adding capa quickstart reference (#1802)

binja: add support for forwarded exports

.devcontainer/Dockerfile

@@ -0,0 +1,21 @@
+# See here for image contents: https://github.com/microsoft/vscode-dev-containers/tree/v0.233.0/containers/python-3/.devcontainer/base.Dockerfile
+
+# [Choice] Python version (use -bullseye variants on local arm64/Apple Silicon): 3, 3.10, 3.9, 3.8, 3.7, 3.6, 3-bullseye, 3.10-bullseye, 3.9-bullseye, 3.8-bullseye, 3.7-bullseye, 3.6-bullseye, 3-buster, 3.10-buster, 3.9-buster, 3.8-buster, 3.7-buster, 3.6-buster
+ARG VARIANT="3.10-bullseye"
+FROM mcr.microsoft.com/vscode/devcontainers/python:0-${VARIANT}
+
+# [Choice] Node.js version: none, lts/*, 16, 14, 12, 10
+ARG NODE_VERSION="none"
+RUN if [ "${NODE_VERSION}" != "none" ]; then su vscode -c "umask 0002 && . /usr/local/share/nvm/nvm.sh && nvm install ${NODE_VERSION} 2>&1"; fi
+
+# [Optional] If your pip requirements rarely change, uncomment this section to add them to the image.
+# COPY requirements.txt /tmp/pip-tmp/
+# RUN pip3 --disable-pip-version-check --no-cache-dir install -r /tmp/pip-tmp/requirements.txt \
+#    && rm -rf /tmp/pip-tmp
+
+# [Optional] Uncomment this section to install additional OS packages.
+# RUN apt-get update && export DEBIAN_FRONTEND=noninteractive \
+#     && apt-get -y install --no-install-recommends <your-package-list-here>
+
+# [Optional] Uncomment this line to install global node packages.
+# RUN su vscode -c "source /usr/local/share/nvm/nvm.sh && npm install -g <your-package-here>" 2>&1

.devcontainer/devcontainer.json

@@ -0,0 +1,51 @@
+// For format details, see https://aka.ms/devcontainer.json. For config options, see the README at:
+// https://github.com/microsoft/vscode-dev-containers/tree/v0.233.0/containers/python-3
+{
+	"name": "Python 3",
+	"build": {
+		"dockerfile": "Dockerfile",
+		"context": "..",
+		"args": { 
+			// Update 'VARIANT' to pick a Python version: 3, 3.10, 3.9, 3.8, 3.7, 3.6
+			// Append -bullseye or -buster to pin to an OS version.
+			// Use -bullseye variants on local on arm64/Apple Silicon.
+			"VARIANT": "3.10",
+			// Options
+			"NODE_VERSION": "none"
+		}
+	},
+
+	// Set *default* container specific settings.json values on container create.
+	"settings": { 
+		"python.defaultInterpreterPath": "/usr/local/bin/python",
+		"python.linting.enabled": true,
+		"python.linting.pylintEnabled": true,
+		"python.formatting.autopep8Path": "/usr/local/py-utils/bin/autopep8",
+		"python.formatting.blackPath": "/usr/local/py-utils/bin/black",
+		"python.formatting.yapfPath": "/usr/local/py-utils/bin/yapf",
+		"python.linting.banditPath": "/usr/local/py-utils/bin/bandit",
+		"python.linting.flake8Path": "/usr/local/py-utils/bin/flake8",
+		"python.linting.mypyPath": "/usr/local/py-utils/bin/mypy",
+		"python.linting.pycodestylePath": "/usr/local/py-utils/bin/pycodestyle",
+		"python.linting.pydocstylePath": "/usr/local/py-utils/bin/pydocstyle",
+		"python.linting.pylintPath": "/usr/local/py-utils/bin/pylint"
+	},
+
+	// Add the IDs of extensions you want installed when the container is created.
+	"extensions": [
+		"ms-python.python",
+		"ms-python.vscode-pylance"
+	],
+
+	// Use 'forwardPorts' to make a list of ports inside the container available locally.
+	// "forwardPorts": [],
+
+	// Use 'postCreateCommand' to run commands after the container is created.
+	"postCreateCommand": "git submodule update --init && pip3 install --user -e .[dev] && pre-commit install",
+
+	// Comment out to connect as root instead. More info: https://aka.ms/vscode-remote/containers/non-root.
+	"remoteUser": "vscode",
+	"features": {
+		"git": "latest"
+	}
+}

.github/CONTRIBUTING.md

@@ -2,7 +2,7 @@
 
 First off, thanks for taking the time to contribute!
 
-The following is a set of guidelines for contributing to capa and its packages, which are hosted in the [FireEye Organization](https://github.com/fireeye) on GitHub. These are mostly guidelines, not rules. Use your best judgment, and feel free to propose changes to this document in a pull request.
+The following is a set of guidelines for contributing to capa and its packages, which are hosted in the [Mandiant Organization](https://github.com/mandiant) on GitHub. These are mostly guidelines, not rules. Use your best judgment, and feel free to propose changes to this document in a pull request.
 
 #### Table Of Contents
 
@@ -31,10 +31,10 @@ This project and everyone participating in it is governed by the [Capa Code of C
 
 ### Capa and its repositories
 
-We host the capa project as three Github repositories:
-  - [capa](https://github.com/fireeye/capa)
-  - [capa-rules](https://github.com/fireeye/capa-rules)
-  - [capa-testfiles](https://github.com/fireeye/capa-testfiles)
+We host the capa project as three GitHub repositories:
+  - [capa](https://github.com/mandiant/capa)
+  - [capa-rules](https://github.com/mandiant/capa-rules)
+  - [capa-testfiles](https://github.com/mandiant/capa-testfiles)
   
 The command line tools, logic engine, and other Python source code are found in the `capa` repository.
 This is the repository to fork when you want to enhance the features, performance, or user interface of capa.
@@ -54,7 +54,7 @@ These are files you'll need in order to run the linter (in `--thorough` mode) an
 ### Design Decisions
 
 When we make a significant decision in how we maintain the project and what we can or cannot support,
- we will document it in the [capa issues tracker](https://github.com/fireeye/capa/issues).
+ we will document it in the [capa issues tracker](https://github.com/mandiant/capa/issues).
 This is the best place review our discussions about what/how/why we do things in the project.
 If you have a question, check to see if it is documented there.
 If it is *not* documented there, or you can't find an answer, please open a issue.
@@ -78,7 +78,7 @@ Fill out [the required template](./ISSUE_TEMPLATE/bug_report.md),
 #### Before Submitting A Bug Report
 
 * **Determine [which repository the problem should be reported in](#capa-and-its-repositories)**.
-* **Perform a [cursory search](https://github.com/fireeye/capa/issues?q=is%3Aissue)** to see if the problem has already been reported. If it has **and the issue is still open**, add a comment to the existing issue instead of opening a new one.
+* **Perform a [cursory search](https://github.com/mandiant/capa/issues?q=is%3Aissue)** to see if the problem has already been reported. If it has **and the issue is still open**, add a comment to the existing issue instead of opening a new one.
 
 #### How Do I Submit A (Good) Bug Report?
 
@@ -101,7 +101,7 @@ Explain the problem and include additional details to help maintainers reproduce
 Provide more context by answering these questions:
 
 * **Did the problem start happening recently** (e.g. after updating to a new version of capa) or was this always a problem?
-* If the problem started happening recently, **can you reproduce the problem in an older version of capa?** What's the most recent version in which the problem doesn't happen? You can download older versions of capa from [the releases page](https://github.com/fireeye/capa/releases).
+* If the problem started happening recently, **can you reproduce the problem in an older version of capa?** What's the most recent version in which the problem doesn't happen? You can download older versions of capa from [the releases page](https://github.com/mandiant/capa/releases).
 * **Can you reliably reproduce the issue?** If not, provide details about how often the problem happens and under which conditions it normally happens.
 * If the problem is related to working with files (e.g. opening and editing files), **does the problem happen for all files and projects or only some?** Does the problem happen only when working with local or remote files (e.g. on network drives), with files of a specific type (e.g. only JavaScript or Python files), with large files or files with very long lines, or with files in a specific encoding? Is there anything else special about the files you are using?
 
@@ -119,7 +119,7 @@ Before creating enhancement suggestions, please check [this list](#before-submit
 #### Before Submitting An Enhancement Suggestion
 
 * **Determine [which repository the enhancement should be suggested in](#capa-and-its-repositories).**
-* **Perform a [cursory search](https://github.com/fireeye/capa/issues?q=is%3Aissue)** to see if the enhancement has already been suggested. If it has, add a comment to the existing issue instead of opening a new one.
+* **Perform a [cursory search](https://github.com/mandiant/capa/issues?q=is%3Aissue)** to see if the enhancement has already been suggested. If it has, add a comment to the existing issue instead of opening a new one.
 
 #### How Do I Submit A (Good) Enhancement Suggestion?
 
@@ -138,15 +138,15 @@ Enhancement suggestions are tracked as [GitHub issues](https://guides.github.com
 
 Unsure where to begin contributing to capa? You can start by looking through these `good-first-issue` and `rule-idea` issues:
 
-* [good-first-issue](https://github.com/fireeye/capa/issues?q=is%3Aissue+is%3Aopen+label%3A%22good+first+issue%22) - issues which should only require a few lines of code, and a test or two.
-* [rule-idea](https://github.com/fireeye/capa-rules/issues?q=is%3Aissue+is%3Aopen+label%3A%22rule+idea%22) - issues that describe potential new rule ideas.
+* [good-first-issue](https://github.com/mandiant/capa/issues?q=is%3Aissue+is%3Aopen+label%3A%22good+first+issue%22) - issues which should only require a few lines of code, and a test or two.
+* [rule-idea](https://github.com/mandiant/capa-rules/issues?q=is%3Aissue+is%3Aopen+label%3A%22rule+idea%22) - issues that describe potential new rule ideas.
 
 Both issue lists are sorted by total number of comments. While not perfect, number of comments is a reasonable proxy for impact a given change will have.
 
 #### Local development
 
 capa and all its resources can be developed locally. 
-For instructions on how to do this, see the "Method 3" section of the [installation guide](https://github.com/fireeye/capa/blob/master/doc/installation.md).
+For instructions on how to do this, see the "Method 3" section of the [installation guide](https://github.com/mandiant/capa/blob/master/doc/installation.md).
 
 ### Pull Requests
 
@@ -159,12 +159,25 @@ The process described here has several goals:
 
 Please follow these steps to have your contribution considered by the maintainers:
 
-1. Follow all instructions in [the template](PULL_REQUEST_TEMPLATE.md)
-2. Follow the [styleguides](#styleguides)
+0. Sign the [Contributor License Agreement](#contributor-license-agreement)
+1. Follow the [styleguides](#styleguides)
+2. Update the CHANGELOG and add tests and documentation. In case they are not needed, indicate it in [the PR template](pull_request_template.md).
 3. After you submit your pull request, verify that all [status checks](https://help.github.com/articles/about-status-checks/) are passing <details><summary>What if the status checks are failing? </summary>If a status check is failing, and you believe that the failure is unrelated to your change, please leave a comment on the pull request explaining why you believe the failure is unrelated. A maintainer will re-run the status check for you. If we conclude that the failure was a false positive, then we will open an issue to track that problem with our status check suite.</details>
 
 While the prerequisites above must be satisfied prior to having your pull request reviewed, the reviewer(s) may ask you to complete additional design work, tests, or other changes before your pull request can be ultimately accepted.
 
+### Contributor License Agreement
+
+Contributions to this project must be accompanied by a Contributor License
+Agreement. You (or your employer) retain the copyright to your contribution,
+this simply gives us permission to use and redistribute your contributions as
+part of the project. Head over to <https://cla.developers.google.com/> to see
+your current agreements on file or to sign a new one.
+
+You generally only need to submit a CLA once, so if you've already submitted one
+(even if it was for a different project), you probably don't need to do it
+again.
+
 ## Styleguides
 
 ### Git Commit Messages
@@ -190,8 +203,8 @@ Our CI pipeline will reformat and enforce the Python styleguide.
 
 All (non-nursery) capa rules must:
 
-  1. pass the [linter](https://github.com/fireeye/capa/blob/master/scripts/lint.py), and
-  2. be formatted with [capafmt](https://github.com/fireeye/capa/blob/master/scripts/capafmt.py)
+  1. pass the [linter](https://github.com/mandiant/capa/blob/master/scripts/lint.py), and
+  2. be formatted with [capafmt](https://github.com/mandiant/capa/blob/master/scripts/capafmt.py)
   
 This ensures that all rules meet the same minimum level of quality and are structured in a consistent way.
 Our CI pipeline will reformat and enforce the capa rules styleguide.

.github/ISSUE_TEMPLATE/bug_report.md

@@ -5,16 +5,16 @@ about: Create a report to help us improve
 ---
 <!--
 # Is your bug report related to capa rules (for example a false positive)?
-We use submodules to separate code, rules and test data. If your issue is related to capa rules, please report it at https://github.com/fireeye/capa-rules/issues.
+We use submodules to separate code, rules and test data. If your issue is related to capa rules, please report it at https://github.com/mandiant/capa-rules/issues.
 
 # Have you checked that your issue isn't already filed?
-Please search if there is a similar issue at https://github.com/fireeye/capa/issues. If there is already a similar issue, please add more details there instead of opening a new one.
+Please search if there is a similar issue at https://github.com/mandiant/capa/issues. If there is already a similar issue, please add more details there instead of opening a new one.
 
 # Have you read capa's Code of Conduct?
-By filing an Issue, you are expected to comply with it, including treating everyone with respect: https://github.com/fireeye/capa/blob/master/.github/CODE_OF_CONDUCT.md
+By filing an Issue, you are expected to comply with it, including treating everyone with respect: https://github.com/mandiant/capa/blob/master/.github/CODE_OF_CONDUCT.md
 
 # Have you read capa's CONTRIBUTING guide?
-It contains helpful information about how to contribute to capa. Check https://github.com/fireeye/capa/blob/master/.github/CONTRIBUTING.md#reporting-bugs
+It contains helpful information about how to contribute to capa. Check https://github.com/mandiant/capa/blob/master/.github/CONTRIBUTING.md#reporting-bugs
 -->
 
 ### Description

.github/ISSUE_TEMPLATE/feature_request.md

@@ -5,16 +5,16 @@ about: Suggest an idea for capa
 ---
 <!--
 # Is your issue related to capa rules (for example an idea for a new rule)?
-We use submodules to separate code, rules and test data. If your issue is related to capa rules, please report it at https://github.com/fireeye/capa-rules/issues.
+We use submodules to separate code, rules and test data. If your issue is related to capa rules, please report it at https://github.com/mandiant/capa-rules/issues.
 
 # Have you checked that your issue isn't already filed?
-Please search if there is a similar issue at https://github.com/fireeye/capa/issues. If there is already a similar issue, please add more details there instead of opening a new one.
+Please search if there is a similar issue at https://github.com/mandiant/capa/issues. If there is already a similar issue, please add more details there instead of opening a new one.
 
 # Have you read capa's Code of Conduct?
-By filing an Issue, you are expected to comply with it, including treating everyone with respect: https://github.com/fireeye/capa/blob/master/.github/CODE_OF_CONDUCT.md
+By filing an Issue, you are expected to comply with it, including treating everyone with respect: https://github.com/mandiant/capa/blob/master/.github/CODE_OF_CONDUCT.md
 
 # Have you read capa's CONTRIBUTING guide?
-It contains helpful information about how to contribute to capa. Check https://github.com/fireeye/capa/blob/master/.github/CONTRIBUTING.md#suggesting-enhancements
+It contains helpful information about how to contribute to capa. Check https://github.com/mandiant/capa/blob/master/.github/CONTRIBUTING.md#suggesting-enhancements
 -->
 
 ### Summary

.github/flake8.ini

@@ -0,0 +1,41 @@
+[flake8]
+max-line-length = 120
+
+extend-ignore =
+    # E203: whitespace before ':'  (black does this)
+    E203,
+    # F401: `foo` imported but unused  (prefer ruff)
+    F401,
+    # F811 Redefinition of unused `foo`  (prefer ruff)
+    F811,
+    # E501 line too long  (prefer black)
+    E501,
+    # B010 Do not call setattr with a constant attribute value
+    B010,
+    # G200 Logging statement uses exception in arguments
+    G200,
+    # SIM102 Use a single if-statement instead of nested if-statements
+    # doesn't provide a space for commenting or logical separation of conditions
+    SIM102,
+    # SIM114 Use logical or and a single body
+    # makes logic trees too complex
+    SIM114,
+    # SIM117 Use 'with Foo, Bar:' instead of multiple with statements
+    # makes lines too long
+    SIM117
+
+per-file-ignores =
+    # T201 print found.
+    #
+    # scripts are meant to print output
+    scripts/*: T201
+    # capa.exe is meant to print output
+    capa/main.py: T201
+    # IDA tests emit results to output window so need to print
+    tests/test_ida_features.py: T201
+    # utility used to find the Binary Ninja API via invoking python.exe
+    capa/features/extractors/binja/find_binja_api.py: T201
+
+copyright-check = True
+copyright-min-file-size = 1 
+copyright-regexp = Copyright \(C\) 2023 Mandiant, Inc. All Rights Reserved.

.github/mypy/mypy.ini

@@ -0,0 +1,91 @@
+[mypy]
+
+[mypy-halo.*]
+ignore_missing_imports = True
+
+[mypy-tqdm.*]
+ignore_missing_imports = True
+
+[mypy-ruamel.*]
+ignore_missing_imports = True
+
+[mypy-networkx.*]
+ignore_missing_imports = True
+
+[mypy-pefile.*]
+ignore_missing_imports = True
+
+[mypy-viv_utils.*]
+ignore_missing_imports = True
+
+[mypy-flirt.*]
+ignore_missing_imports = True
+
+[mypy-lief.*]
+ignore_missing_imports = True
+
+[mypy-idc.*]
+ignore_missing_imports = True
+
+[mypy-vivisect.*]
+ignore_missing_imports = True
+
+[mypy-envi.*]
+ignore_missing_imports = True
+
+[mypy-PE.*]
+ignore_missing_imports = True
+
+[mypy-idaapi.*]
+ignore_missing_imports = True
+
+[mypy-idautils.*]
+ignore_missing_imports = True
+
+[mypy-ida_auto.*]
+ignore_missing_imports = True
+
+[mypy-ida_bytes.*]
+ignore_missing_imports = True
+
+[mypy-ida_nalt.*]
+ignore_missing_imports = True
+
+[mypy-ida_kernwin.*]
+ignore_missing_imports = True
+
+[mypy-ida_settings.*]
+ignore_missing_imports = True
+
+[mypy-ida_funcs.*]
+ignore_missing_imports = True
+
+[mypy-ida_loader.*]
+ignore_missing_imports = True
+
+[mypy-ida_segment.*]
+ignore_missing_imports = True
+
+[mypy-PyQt5.*]
+ignore_missing_imports = True
+
+[mypy-binaryninja.*]
+ignore_missing_imports = True
+
+[mypy-pytest.*]
+ignore_missing_imports = True
+
+[mypy-devtools.*]
+ignore_missing_imports = True
+
+[mypy-elftools.*]
+ignore_missing_imports = True
+
+[mypy-dncil.*]
+ignore_missing_imports = True
+
+[mypy-netnode.*]
+ignore_missing_imports = True
+
+[mypy-ghidra.*]
+ignore_missing_imports = True

.github/pull_request_template.md

@@ -1,32 +1,22 @@
-<!--
-Thank you for contributing to capa! :heart:
 
-IMPORTANT NOTE
-It's most important that you submit your improvements. So even if you don't use this complete template we look forward to collaborating!
+<!--
+Thank you for contributing to capa! <3
 
 Please read capa's CONTRIBUTING guide if you haven't done so already.
-It contains helpful information about how to contribute to capa. Check https://github.com/fireeye/capa/blob/master/.github/CONTRIBUTING.md
+It contains helpful information about how to contribute to capa. Check https://github.com/mandiant/capa/blob/master/.github/CONTRIBUTING.md
 
-PR template based on https://embeddedartistry.com/blog/2017/08/04/a-github-pull-request-template-for-your-projects/
+Please describe the changes in this pull request (PR). Include your motivation and context to help us review.
+
+Please mention the issue your PR addresses (if any):
+closes #issue_number
 -->
 
-### Description
 
-<!-- Please describe the changes in this PR. Including your motivation and context helps us to review. -->
+### Checklist
 
-closes # (issue)
-
-### Type of change
-
-Please update the [CHANGELOG.md](/CHANGELOG.md)
-
-- [ ] Bug fix (non-breaking change which fixes an issue)
-- [ ] New feature (non-breaking change which adds functionality)
-- [ ] Breaking change (fix or feature that would cause existing functionality to not work as expected)
-- [ ] This change requires a documentation update
-  - [ ] I have made the corresponding changes to the documentation
-
-### Tests
-
-- [ ] I have added tests that prove my fix is effective or that my feature works
+<!-- CHANGELOG.md has a `master (unreleased)` section. Please add bug fixes, new features, breaking changes and anything else you think is worthwhile mentioning in the release notes to this file. -->
+- [ ] No CHANGELOG update needed
+<!-- Tests prove that your fix/work as expected and ensure it doesn't break on the feature. -->
 - [ ] No new tests needed
+<!-- Please help us keeping capa documentation up-to-date -->
+- [ ] No documentation update needed

.github/pyinstaller/hooks/hook-smda.py

@@ -1,5 +0,0 @@
-# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
-import PyInstaller.utils.hooks
-
-# ref: https://groups.google.com/g/pyinstaller/c/amWi0-66uZI/m/miPoKfWjBAAJ
-binaries = PyInstaller.utils.hooks.collect_dynamic_libs("capstone")

.github/pyinstaller/hooks/hook-vivisect.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
+# Copyright (C) 2020 Mandiant, Inc. All Rights Reserved.
 
 from PyInstaller.utils.hooks import copy_metadata
 
@@ -38,39 +38,36 @@ hiddenimports = [
     "vivisect",
     "vivisect.analysis",
     "vivisect.analysis.amd64",
-    "vivisect.analysis.amd64",
     "vivisect.analysis.amd64.emulation",
     "vivisect.analysis.amd64.golang",
     "vivisect.analysis.crypto",
-    "vivisect.analysis.crypto",
     "vivisect.analysis.crypto.constants",
     "vivisect.analysis.elf",
-    "vivisect.analysis.elf",
     "vivisect.analysis.elf.elfplt",
+    "vivisect.analysis.elf.elfplt_late",
     "vivisect.analysis.elf.libc_start_main",
     "vivisect.analysis.generic",
-    "vivisect.analysis.generic",
     "vivisect.analysis.generic.codeblocks",
     "vivisect.analysis.generic.emucode",
     "vivisect.analysis.generic.entrypoints",
     "vivisect.analysis.generic.funcentries",
     "vivisect.analysis.generic.impapi",
+    "vivisect.analysis.generic.linker",
     "vivisect.analysis.generic.mkpointers",
+    "vivisect.analysis.generic.noret",
     "vivisect.analysis.generic.pointers",
     "vivisect.analysis.generic.pointertables",
     "vivisect.analysis.generic.relocations",
     "vivisect.analysis.generic.strconst",
     "vivisect.analysis.generic.switchcase",
+    "vivisect.analysis.generic.symswitchcase",
     "vivisect.analysis.generic.thunks",
-    "vivisect.analysis.generic.noret",
-    "vivisect.analysis.i386",
     "vivisect.analysis.i386",
     "vivisect.analysis.i386.calling",
     "vivisect.analysis.i386.golang",
     "vivisect.analysis.i386.importcalls",
     "vivisect.analysis.i386.instrhook",
-    "vivisect.analysis.i386.thunk_bx",
-    "vivisect.analysis.ms",
+    "vivisect.analysis.i386.thunk_reg",
     "vivisect.analysis.ms",
     "vivisect.analysis.ms.hotpatch",
     "vivisect.analysis.ms.localhints",
@@ -81,8 +78,40 @@ hiddenimports = [
     "vivisect.impapi.posix.amd64",
     "vivisect.impapi.posix.i386",
     "vivisect.impapi.windows",
+    "vivisect.impapi.windows.advapi_32",
+    "vivisect.impapi.windows.advapi_64",
     "vivisect.impapi.windows.amd64",
+    "vivisect.impapi.windows.gdi_32",
+    "vivisect.impapi.windows.gdi_64",
     "vivisect.impapi.windows.i386",
+    "vivisect.impapi.windows.kernel_32",
+    "vivisect.impapi.windows.kernel_64",
+    "vivisect.impapi.windows.msvcr100_32",
+    "vivisect.impapi.windows.msvcr100_64",
+    "vivisect.impapi.windows.msvcr110_32",
+    "vivisect.impapi.windows.msvcr110_64",
+    "vivisect.impapi.windows.msvcr120_32",
+    "vivisect.impapi.windows.msvcr120_64",
+    "vivisect.impapi.windows.msvcr71_32",
+    "vivisect.impapi.windows.msvcr80_32",
+    "vivisect.impapi.windows.msvcr80_64",
+    "vivisect.impapi.windows.msvcr90_32",
+    "vivisect.impapi.windows.msvcr90_64",
+    "vivisect.impapi.windows.msvcrt_32",
+    "vivisect.impapi.windows.msvcrt_64",
+    "vivisect.impapi.windows.ntdll_32",
+    "vivisect.impapi.windows.ntdll_64",
+    "vivisect.impapi.windows.ole_32",
+    "vivisect.impapi.windows.ole_64",
+    "vivisect.impapi.windows.rpcrt4_32",
+    "vivisect.impapi.windows.rpcrt4_64",
+    "vivisect.impapi.windows.shell_32",
+    "vivisect.impapi.windows.shell_64",
+    "vivisect.impapi.windows.user_32",
+    "vivisect.impapi.windows.user_64",
+    "vivisect.impapi.windows.ws2plus_32",
+    "vivisect.impapi.windows.ws2plus_64",
+    "vivisect.impapi.winkern",
     "vivisect.impapi.winkern.i386",
     "vivisect.impapi.winkern.amd64",
     "vivisect.parsers.blob",

.github/pyinstaller/pyinstaller.spec

@@ -1,55 +1,40 @@
 # -*- mode: python -*-
-# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
+# Copyright (C) 2020 Mandiant, Inc. All Rights Reserved.
 import os.path
 import subprocess
 
 import wcwidth
 
 
-# when invoking pyinstaller from the project root,
-# this gets run from the project root.
-with open('./capa/version.py', 'wb') as f:
-    # git output will look like:
-    #
-    #     tags/v1.0.0-0-g3af38dc
-    #         ------- tag
-    #                 - commits since
-    #                   g------- git hash fragment
-    version = (subprocess.check_output(["git", "describe", "--always", "--tags", "--long"])
-               .decode("utf-8")
-               .strip()
-               .replace("tags/", ""))
-    f.write(("__version__ = '%s'" % version).encode("utf-8"))
-
 a = Analysis(
     # when invoking pyinstaller from the project root,
     # this gets invoked from the directory of the spec file,
     # i.e. ./.github/pyinstaller
-    ['../../capa/main.py'],
-    pathex=['capa'],
+    ["../../capa/main.py"],
+    pathex=["capa"],
     binaries=None,
     datas=[
         # when invoking pyinstaller from the project root,
         # this gets invoked from the directory of the spec file,
         # i.e. ./.github/pyinstaller
-        ('../../rules', 'rules'),
-
+        ("../../rules", "rules"),
+        ("../../sigs", "sigs"),
+        ("../../cache", "cache"),
         # capa.render.default uses tabulate that depends on wcwidth.
         # it seems wcwidth uses a json file `version.json`
         # and this doesn't get picked up by pyinstaller automatically.
         # so we manually embed the wcwidth resources here.
         #
         # ref: https://stackoverflow.com/a/62278462/87207
-        (os.path.dirname(wcwidth.__file__), 'wcwidth')
+        (os.path.dirname(wcwidth.__file__), "wcwidth"),
     ],
     # when invoking pyinstaller from the project root,
     # this gets run from the project root.
-    hookspath=['.github/pyinstaller/hooks'],
+    hookspath=[".github/pyinstaller/hooks"],
     runtime_hooks=None,
     excludes=[
         # ignore packages that would otherwise be bundled with the .exe.
         # review: build/pyinstaller/xref-pyinstaller.html
-
         # we don't do any GUI stuff, so ignore these modules
         "tkinter",
         "_tkinter",
@@ -59,7 +44,6 @@ a = Analysis(
         # since we don't spawn a notebook, we can safely remove these.
         "IPython",
         "ipywidgets",
-
         # these are pulled in by networkx
         # but we don't need to compute the strongly connected components.
         "numpy",
@@ -67,7 +51,6 @@ a = Analysis(
         "matplotlib",
         "pandas",
         "pytest",
-
         # deps from viv that we don't use.
         # this duplicates the entries in `hook-vivisect`,
         # but works better this way.
@@ -77,36 +60,36 @@ a = Analysis(
         "PyQt5",
         "qt5",
         "pyqtwebengine",
-        "pyasn1"
-    ])
+        "pyasn1",
+        "binaryninja",
+    ],
+)
 
-a.binaries = a.binaries - TOC([
- ('tcl85.dll', None, None),
- ('tk85.dll', None, None),
- ('_tkinter', None, None)])
+a.binaries = a.binaries - TOC([("tcl85.dll", None, None), ("tk85.dll", None, None), ("_tkinter", None, None)])
 
 pyz = PYZ(a.pure, a.zipped_data)
 
-exe = EXE(pyz,
-          a.scripts,
-          a.binaries,
-          a.zipfiles,
-          a.datas,
-          exclude_binaries=False,
-          name='capa',
-          icon='logo.ico',
-          debug=False,
-          strip=None,
-          upx=True,
-          console=True )
+exe = EXE(
+    pyz,
+    a.scripts,
+    a.binaries,
+    a.zipfiles,
+    a.datas,
+    exclude_binaries=False,
+    name="capa",
+    icon="logo.ico",
+    debug=False,
+    strip=None,
+    upx=True,
+    console=True,
+)
 
 # enable the following to debug the contents of the .exe
 #
-#coll = COLLECT(exe,
+# coll = COLLECT(exe,
 #               a.binaries,
 #               a.zipfiles,
 #               a.datas,
 #               strip=None,
 #               upx=True,
 #               name='capa-dat')
-

.github/ruff.toml

@@ -0,0 +1,43 @@
+# Enable the pycodestyle (`E`) and Pyflakes (`F`) rules by default.
+# Unlike Flake8, Ruff doesn't enable pycodestyle warnings (`W`) or
+# McCabe complexity (`C901`) by default.
+select = ["E", "F"]
+
+# Allow autofix for all enabled rules (when `--fix`) is provided.
+fixable = ["ALL"]
+unfixable = []
+
+# E402 module level import not at top of file
+# E722 do not use bare 'except'
+# E501 line too long
+ignore = ["E402", "E722", "E501"]
+
+line-length = 120
+
+exclude = [
+    # Exclude a variety of commonly ignored directories.
+    ".bzr",
+    ".direnv",
+    ".eggs",
+    ".git",
+    ".git-rewrite",
+    ".hg",
+    ".mypy_cache",
+    ".nox",
+    ".pants.d",
+    ".pytype",
+    ".ruff_cache",
+    ".svn",
+    ".tox",
+    ".venv",
+    "__pypackages__",
+    "_build",
+    "buck-out",
+    "build",
+    "dist",
+    "node_modules",
+    "venv",
+    # protobuf generated files
+    "*_pb2.py", 
+    "*_pb2.pyi"
+]

.github/tox.ini

@@ -1,10 +0,0 @@
-[pycodestyle]
-; E402: module level import not at top of file
-; W503: line break before binary operator
-; E231 missing whitespace after ',' (emitted by black)
-; E203 whitespace before ':' (emitted by black)
-ignore = E402,W503,E203,E231
-max-line-length = 160
-statistics = True
-count = True
-exclude = .*

.github/workflows/build.yml

@@ -1,54 +1,96 @@
 name: build
 
 on:
+  pull_request:
+    branches: [ master ]
   release:
     types: [edited, published]
 
+permissions:
+  contents: write
+
 jobs:
   build:
     name: PyInstaller for ${{ matrix.os }}
     runs-on: ${{ matrix.os }}
     strategy:
+      # set to false for debugging
+      fail-fast: true
       matrix:
         include:
-          - os: ubuntu-16.04
+          - os: ubuntu-20.04
             # use old linux so that the shared library versioning is more portable
             artifact_name: capa
             asset_name: linux
           - os: windows-2019
             artifact_name: capa.exe
             asset_name: windows
-          - os: macos-10.15
+          - os: macos-11
+            # use older macOS for assumed better portability
             artifact_name: capa
             asset_name: macos
     steps:
       - name: Checkout capa
-        uses: actions/checkout@v2
+        uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # v3.3.0
         with:
           submodules: true
-      - name: Set up Python 3.9
-        uses: actions/setup-python@v2
+      # using Python 3.8 to support running across multiple operating systems including Windows 7
+      - name: Set up Python 3.8
+        uses: actions/setup-python@d27e3f3d7c64b4bbf8e4abfb9b63b83e846e0435 # v4.5.0
         with:
-          python-version: 3.9
-      - if: matrix.os == 'ubuntu-16.04'
+          python-version: 3.8
+      - if: matrix.os == 'ubuntu-20.04'
         run: sudo apt-get install -y libyaml-dev
-      - name: Install PyInstaller
-        run: pip install 'pyinstaller==4.2'
-      - name: Install capa
-        run: pip install -e .
+      - name: Upgrade pip, setuptools
+        run: python -m pip install --upgrade pip setuptools
+      - name: Install capa with build requirements
+        run: pip install -e .[build]
+      - name: Cache the rule set
+        run: python ./scripts/cache-ruleset.py ./rules/ ./cache/
       - name: Build standalone executable
-        run: pyinstaller .github/pyinstaller/pyinstaller.spec
-      - name: Does it run?
+        run: pyinstaller --log-level DEBUG .github/pyinstaller/pyinstaller.spec
+      - name: Does it run (PE)?
         run: dist/capa "tests/data/Practical Malware Analysis Lab 01-01.dll_"
-      - uses: actions/upload-artifact@v2
+      - name: Does it run (Shellcode)?
+        run: dist/capa "tests/data/499c2a85f6e8142c3f48d4251c9c7cd6.raw32"
+      - name: Does it run (ELF)?
+        run: dist/capa "tests/data/7351f8a40c5450557b24622417fc478d.elf_"
+      - uses: actions/upload-artifact@0b7f8abb1508181956e8e162db84b466c27e18ce # v3.1.2
         with:
           name: ${{ matrix.asset_name }}
           path: dist/${{ matrix.artifact_name }}
 
-  zip:
-    name: zip ${{ matrix.asset_name }}
+  test_run:
+    name: Test run on ${{ matrix.os }}
+    runs-on: ${{ matrix.os }}
+    needs: [build]
+    strategy:
+      matrix:
+        include:
+          # OSs not already tested above
+          - os: ubuntu-22.04
+            artifact_name: capa
+            asset_name: linux
+          - os: windows-2022
+            artifact_name: capa.exe
+            asset_name: windows
+    steps:
+      - name: Download ${{ matrix.asset_name }}
+        uses: actions/download-artifact@9bc31d5ccc31df68ecc42ccf4149144866c47d8a # v3.0.2
+        with:
+          name: ${{ matrix.asset_name }}
+      - name: Set executable flag
+        if: matrix.os != 'windows-2022'
+        run: chmod +x ${{ matrix.artifact_name }}
+      - name: Run capa
+        run: ./${{ matrix.artifact_name }} -h
+
+  zip_and_upload:
+    # upload zipped binaries to Release page
+    if: github.event_name == 'release'
+    name: zip and upload ${{ matrix.asset_name }}
     runs-on: ubuntu-20.04
-    needs: build
+    needs: [build]
     strategy:
       matrix:
         include:
@@ -60,7 +102,7 @@ jobs:
             artifact_name: capa
     steps:
       - name: Download ${{ matrix.asset_name }}
-        uses: actions/download-artifact@v2
+        uses: actions/download-artifact@9bc31d5ccc31df68ecc42ccf4149144866c47d8a # v3.0.2
         with:
           name: ${{ matrix.asset_name }}
       - name: Set executable flag
@@ -70,7 +112,7 @@ jobs:
       - name: Zip ${{ matrix.artifact_name }} into ${{ env.zip_name }}
         run: zip ${{ env.zip_name }} ${{ matrix.artifact_name }}
       - name: Upload ${{ env.zip_name }} to GH Release
-        uses: svenstaro/upload-release-action@v2
+        uses: svenstaro/upload-release-action@2728235f7dc9ff598bd86ce3c274b74f802d2208 # v2
         with:
           repo_token: ${{ secrets.GITHUB_TOKEN}}
           file: ${{ env.zip_name }}

.github/workflows/changelog.yml

@@ -0,0 +1,43 @@
+name: changelog
+
+on:
+  # We need pull_request_target instead of pull_request because a write
+  # repository token is needed to add a review to a PR. DO NOT BUILD
+  # OR RUN UNTRUSTED CODE FROM PRs IN THIS ACTION
+  pull_request_target:
+    types: [opened, edited, synchronize]
+
+permissions: read-all
+
+jobs:
+  check_changelog:
+    # no need to check for dependency updates via dependabot
+    if: github.actor != 'dependabot[bot]' && github.actor != 'dependabot-preview[bot]'
+    runs-on: ubuntu-20.04
+    env:
+      NO_CHANGELOG: '[x] No CHANGELOG update needed'
+    steps:
+    - name: Get changed files
+      id: files
+      uses: Ana06/get-changed-files@e0c398b7065a8d84700c471b6afc4116d1ba4e96 # v2.2.0
+    - name: check changelog updated
+      id: changelog_updated
+      env:
+        PR_BODY: ${{ github.event.pull_request.body }}
+        FILES: ${{ steps.files.outputs.modified }}
+      run: |
+        echo $FILES | grep -qF 'CHANGELOG.md' || echo $PR_BODY | grep -qiF "$NO_CHANGELOG"
+    - name: Reject pull request if no CHANGELOG update
+      if: ${{ always() && steps.changelog_updated.outcome == 'failure' }}
+      uses: Ana06/automatic-pull-request-review@0cf4e8a17ba79344ed3fdd7fed6dd0311d08a9d4 # v0.1.0
+      with:
+        repo-token: ${{ secrets.GITHUB_TOKEN }}
+        event: REQUEST_CHANGES
+        body: "Please add bug fixes, new features, breaking changes and anything else you think is worthwhile mentioning to the `master (unreleased)` section of CHANGELOG.md. If no CHANGELOG update is needed add the following to the PR description: `${{ env.NO_CHANGELOG }}`"
+        allow_duplicate: false
+    - name: Dismiss previous review if CHANGELOG update
+      uses: Ana06/automatic-pull-request-review@0cf4e8a17ba79344ed3fdd7fed6dd0311d08a9d4 # v0.1.0
+      with:
+        repo-token: ${{ secrets.GITHUB_TOKEN }}
+        event: DISMISS
+        body: "CHANGELOG updated or no update needed, thanks! :smile:"

.github/workflows/publish.yml

@@ -1,29 +1,41 @@
-# This workflows will upload a Python Package using Twine when a release is created
-# For more information see: https://help.github.com/en/actions/language-and-framework-guides/using-python-with-github-actions#publishing-to-package-registries
-
+# use PyPI trusted publishing, as described here: 
+# https://blog.trailofbits.com/2023/05/23/trusted-publishing-a-new-benchmark-for-packaging-security/
 name: publish to pypi
 
 on:
   release:
     types: [published]
 
+permissions:
+  contents: write
+
 jobs:
-  deploy:
-    runs-on: ubuntu-20.04
+  pypi-publish:
+    runs-on: ubuntu-latest
+    environment:
+      name: release
+    permissions:
+      id-token: write
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # v3.3.0
       - name: Set up Python
-        uses: actions/setup-python@v2
+        uses: actions/setup-python@d27e3f3d7c64b4bbf8e4abfb9b63b83e846e0435 # v4.5.0
         with:
-          python-version: '2.7'
+          python-version: '3.8'
       - name: Install dependencies
         run: |
           python -m pip install --upgrade pip
-          pip install setuptools wheel twine
-      - name: Build and publish
-        env:
-          TWINE_USERNAME: ${{ secrets.PYPI_USERNAME }}
-          TWINE_PASSWORD: ${{ secrets.PYPI_PASSWORD }}
+          pip install -e .[build]
+      - name: build package
         run: |
-          python setup.py sdist bdist_wheel
-          twine upload --skip-existing dist/*
+          python -m build
+      - name: upload package artifacts
+        uses: actions/upload-artifact@0b7f8abb1508181956e8e162db84b466c27e18ce # v3.1.2
+        with:
+          path: dist/*
+      - name: publish package
+        uses: pypa/gh-action-pypi-publish@f5622bde02b04381239da3573277701ceca8f6a0  # release/v1
+        with:
+          skip-existing: true
+          verbose: true
+          print-hash: true

.github/workflows/scorecard.yml

@@ -0,0 +1,72 @@
+# This workflow uses actions that are not certified by GitHub. They are provided
+# by a third-party and are governed by separate terms of service, privacy
+# policy, and support documentation.
+
+name: Scorecard supply-chain security
+on:
+  # For Branch-Protection check. Only the default branch is supported. See
+  # https://github.com/ossf/scorecard/blob/main/docs/checks.md#branch-protection
+  branch_protection_rule:
+  # To guarantee Maintained check is occasionally updated. See
+  # https://github.com/ossf/scorecard/blob/main/docs/checks.md#maintained
+  schedule:
+    - cron: '43 4 * * 3'
+  push:
+    branches: [ "master" ]
+
+# Declare default permissions as read only.
+permissions: read-all
+
+jobs:
+  analysis:
+    name: Scorecard analysis
+    runs-on: ubuntu-latest
+    permissions:
+      # Needed to upload the results to code-scanning dashboard.
+      security-events: write
+      # Needed to publish results and get a badge (see publish_results below).
+      id-token: write
+      # Uncomment the permissions below if installing in a private repository.
+      # contents: read
+      # actions: read
+
+    steps:
+      - name: "Checkout code"
+        uses: actions/checkout@93ea575cb5d8a053eaa0ac8fa3b40d7e05a33cc8 # v3.1.0
+        with:
+          persist-credentials: false
+
+      - name: "Run analysis"
+        uses: ossf/scorecard-action@99c53751e09b9529366343771cc321ec74e9bd3d # v2.0.6
+        with:
+          results_file: results.sarif
+          results_format: sarif
+          # (Optional) "write" PAT token. Uncomment the `repo_token` line below if:
+          # - you want to enable the Branch-Protection check on a *public* repository, or
+          # - you are installing Scorecard on a *private* repository
+          # To create the PAT, follow the steps in https://github.com/ossf/scorecard-action#authentication-with-pat.
+          # repo_token: ${{ secrets.SCORECARD_TOKEN }}
+
+          # Public repositories:
+          #   - Publish results to OpenSSF REST API for easy access by consumers
+          #   - Allows the repository to include the Scorecard badge.
+          #   - See https://github.com/ossf/scorecard-action#publishing-results.
+          # For private repositories:
+          #   - `publish_results` will always be set to `false`, regardless
+          #     of the value entered here.
+          publish_results: true
+
+      # Upload the results as artifacts (optional). Commenting out will disable uploads of run results in SARIF
+      # format to the repository Actions tab.
+      - name: "Upload artifact"
+        uses: actions/upload-artifact@3cea5372237819ed00197afe530f5a7ea3e805c8 # v3.1.0
+        with:
+          name: SARIF file
+          path: results.sarif
+          retention-days: 5
+
+      # Upload the results to GitHub's code scanning dashboard.
+      - name: "Upload to code-scanning"
+        uses: github/codeql-action/upload-sarif@807578363a7869ca324a79039e6db9c843e0e100 # v2.1.27
+        with:
+          sarif_file: results.sarif

.github/workflows/tag.yml

@@ -4,21 +4,29 @@ on:
   release:
     types: [published]
 
+permissions: read-all
+
 jobs:
   tag:
     name: Tag capa rules
     runs-on: ubuntu-20.04
     steps:
     - name: Checkout capa-rules
-      uses: actions/checkout@v2
+      uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # v3.3.0
       with:
-        repository: fireeye/capa-rules
+        repository: mandiant/capa-rules
         token: ${{ secrets.CAPA_TOKEN }}
     - name: Tag capa-rules
-      run: git tag ${{ github.event.release.tag_name }}
+      run: |
+        # user information is needed to create annotated tags (with a message)
+        git config user.email 'capa-dev@mandiant.com'
+        git config user.name 'Capa Bot'
+        name=${{ github.event.release.tag_name }}
+        git tag $name -m "https://github.com/mandiant/capa/releases/$name"
+        # TODO update branch name-major=${name%%.*}
     - name: Push tag to capa-rules
-      uses: ad-m/github-push-action@master
+      uses: ad-m/github-push-action@0fafdd62b84042d49ec0cb92d9cac7f7ce4ec79e # master
       with:
-        repository: fireeye/capa-rules
+        repository: mandiant/capa-rules
         github_token: ${{ secrets.CAPA_TOKEN }}
         tags: true

.github/workflows/tests.yml

@@ -6,37 +6,60 @@ on:
   pull_request:
     branches: [ master ]
 
+permissions: read-all
+
+# save workspaces to speed up testing
+env:
+  CAPA_SAVE_WORKSPACE: "True"
+
 jobs:
+  changelog_format:
+    runs-on: ubuntu-20.04
+    steps:
+    - name: Checkout capa
+      uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # v3.3.0
+    # The sync GH action in capa-rules relies on a single '- *$' in the CHANGELOG file
+    - name: Ensure CHANGELOG has '- *$'
+      run: |
+        number=$(grep '\- *$' CHANGELOG.md | wc -l)
+        if [ $number != 1 ]; then exit 1; fi
+
   code_style:
     runs-on: ubuntu-20.04
     steps:
     - name: Checkout capa
-      uses: actions/checkout@v2
-    - name: Set up Python 3.8
-      uses: actions/setup-python@v2
+      uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # v3.3.0
+    # use latest available python to take advantage of best performance
+    - name: Set up Python 3.11
+      uses: actions/setup-python@d27e3f3d7c64b4bbf8e4abfb9b63b83e846e0435 # v4.5.0
       with:
-        python-version: 3.8
+        python-version: "3.11"
     - name: Install dependencies
-      run: pip install 'isort==5.*' black
+      run: pip install -e .[dev]
+    - name: Lint with ruff
+      run: pre-commit run ruff
     - name: Lint with isort
-      run: isort --profile black --length-sort --line-width 120 -c .
+      run: pre-commit run isort
     - name: Lint with black
-      run: black -l 120 --check .
+      run: pre-commit run black
+    - name: Lint with flake8
+      run: pre-commit run flake8
+    - name: Check types with mypy
+      run:  pre-commit run mypy
 
   rule_linter:
     runs-on: ubuntu-20.04
     steps:
-    - name: Checkout capa with rules submodule
-      uses: actions/checkout@v2
+    - name: Checkout capa with submodules
+      uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # v3.3.0
       with:
-        submodules: true
-    - name: Set up Python 3.8
-      uses: actions/setup-python@v2
+        submodules: recursive
+    - name: Set up Python 3.11
+      uses: actions/setup-python@d27e3f3d7c64b4bbf8e4abfb9b63b83e846e0435 # v4.5.0
       with:
-        python-version: 3.8
-    # We don't need vivisect, so we can install capa using Python3
+        python-version: "3.11"
     - name: Install capa
-      run: pip install -e .
+      run: pip install -e .[dev]
     - name: Run rule linter
       run: python scripts/lint.py rules/
 
@@ -47,33 +70,131 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        os: [ubuntu-20.04, windows-2019, macos-10.15]
+        os: [ubuntu-20.04, windows-2019, macos-11]
         # across all operating systems
-        python-version: [3.6, 3.9]
+        python-version: ["3.8", "3.11"]
         include:
           # on Ubuntu run these as well
           - os: ubuntu-20.04
-            python-version: 2.7
+            python-version: "3.8"
           - os: ubuntu-20.04
-            python-version: 3.7
+            python-version: "3.9"
           - os: ubuntu-20.04
-            python-version: 3.8
+            python-version: "3.10"
     steps:
     - name: Checkout capa with submodules
-      uses: actions/checkout@v2
+      uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # v3.3.0
       with:
-        submodules: true
+        submodules: recursive
     - name: Set up Python ${{ matrix.python-version }}
-      uses: actions/setup-python@v2
+      uses: actions/setup-python@d27e3f3d7c64b4bbf8e4abfb9b63b83e846e0435 # v4.5.0
       with:
         python-version: ${{ matrix.python-version }}
     - name: Install pyyaml
       if: matrix.os == 'ubuntu-20.04'
       run: sudo apt-get install -y libyaml-dev
-    - name: Install Microsoft Visual C++ 9.0
-      if: matrix.os == 'windows-2019' && matrix.python-version == '2.7'
-      run: choco install vcpython27
     - name: Install capa
       run: pip install -e .[dev]
     - name: Run tests
-      run: pytest tests/
+      run: pytest -v tests/
+
+  binja-tests:
+    name: Binary Ninja tests for ${{ matrix.python-version }}
+    env:
+      BN_SERIAL: ${{ secrets.BN_SERIAL }}
+    runs-on: ubuntu-20.04
+    needs: [code_style, rule_linter]
+    strategy:
+      fail-fast: false
+      matrix:
+        python-version: ["3.8", "3.11"]
+    steps:
+    - name: Checkout capa with submodules
+      # do only run if BN_SERIAL is available, have to do this in every step, see https://github.com/orgs/community/discussions/26726#discussioncomment-3253118
+      if: ${{ env.BN_SERIAL != 0 }}
+      uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # v3.3.0
+      with:
+        submodules: recursive
+    - name: Set up Python ${{ matrix.python-version }}
+      if: ${{ env.BN_SERIAL != 0 }}
+      uses: actions/setup-python@d27e3f3d7c64b4bbf8e4abfb9b63b83e846e0435 # v4.5.0
+      with:
+        python-version: ${{ matrix.python-version }}
+    - name: Install pyyaml
+      if: ${{ env.BN_SERIAL != 0 }}
+      run: sudo apt-get install -y libyaml-dev
+    - name: Install capa
+      if: ${{ env.BN_SERIAL != 0 }}
+      run: pip install -e .[dev]
+    - name: install Binary Ninja
+      if: ${{ env.BN_SERIAL != 0 }}
+      run: |
+        mkdir ./.github/binja
+        curl "https://raw.githubusercontent.com/Vector35/binaryninja-api/6812c97/scripts/download_headless.py" -o ./.github/binja/download_headless.py
+        python ./.github/binja/download_headless.py --serial ${{ env.BN_SERIAL }} --output .github/binja/BinaryNinja-headless.zip
+        unzip .github/binja/BinaryNinja-headless.zip -d .github/binja/
+        python .github/binja/binaryninja/scripts/install_api.py --install-on-root --silent
+    - name: Run tests
+      if: ${{ env.BN_SERIAL != 0 }}
+      env:
+        BN_LICENSE: ${{ secrets.BN_LICENSE }}
+      run: pytest -v tests/test_binja_features.py  # explicitly refer to the binja tests for performance. other tests run above.
+
+  ghidra-tests:
+    name: Ghidra tests for ${{ matrix.python-version }}
+    runs-on: ubuntu-20.04
+    needs: [code_style, rule_linter]
+    strategy:
+      fail-fast: false
+      matrix:
+        python-version: ["3.8", "3.11"]
+        java-version: ["17"]
+        gradle-version: ["7.3"]
+        ghidra-version: ["10.3"]
+        public-version: ["PUBLIC_20230510"] # for ghidra releases
+        jep-version: ["4.1.1"]
+        ghidrathon-version: ["3.0.0"]
+    steps:
+    - name: Checkout capa with submodules
+      uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # v3.3.0
+      with:
+        submodules: true
+    - name: Set up Python ${{ matrix.python-version }}
+      uses: actions/setup-python@d27e3f3d7c64b4bbf8e4abfb9b63b83e846e0435 # v4.5.0
+      with:
+        python-version: ${{ matrix.python-version }}
+    - name: Set up Java ${{ matrix.java-version }}
+      uses: actions/setup-java@5ffc13f4174014e2d4d4572b3d74c3fa61aeb2c2 # v3
+      with:
+        distribution: 'temurin'
+        java-version: ${{ matrix.java-version }}
+    - name: Set up Gradle ${{ matrix.gradle-version }} 
+      uses: gradle/gradle-build-action@40b6781dcdec2762ad36556682ac74e31030cfe2 # v2.5.1
+      with:
+        gradle-version: ${{ matrix.gradle-version }}
+    - name: Install Jep ${{ matrix.jep-version }} 
+      run : pip install jep==${{ matrix.jep-version }}
+    - name: Install Ghidra ${{ matrix.ghidra-version }} 
+      run: |
+        mkdir ./.github/ghidra
+        wget "https://github.com/NationalSecurityAgency/ghidra/releases/download/Ghidra_${{ matrix.ghidra-version }}_build/ghidra_${{ matrix.ghidra-version }}_${{ matrix.public-version }}.zip" -O ./.github/ghidra/ghidra_${{ matrix.ghidra-version }}_PUBLIC.zip
+        unzip .github/ghidra/ghidra_${{ matrix.ghidra-version }}_PUBLIC.zip -d .github/ghidra/
+    - name: Install Ghidrathon
+      run : |
+        mkdir ./.github/ghidrathon
+        curl -o ./.github/ghidrathon/ghidrathon-${{ matrix.ghidrathon-version }}.zip "https://codeload.github.com/mandiant/Ghidrathon/zip/refs/tags/v${{ matrix.ghidrathon-version }}"
+        unzip .github/ghidrathon/ghidrathon-${{ matrix.ghidrathon-version }}.zip -d .github/ghidrathon/
+        gradle -p ./.github/ghidrathon/Ghidrathon-${{ matrix.ghidrathon-version }}/ -PGHIDRA_INSTALL_DIR=$(pwd)/.github/ghidra/ghidra_${{ matrix.ghidra-version }}_PUBLIC
+        unzip .github/ghidrathon/Ghidrathon-${{ matrix.ghidrathon-version }}/dist/*.zip -d .github/ghidra/ghidra_${{ matrix.ghidra-version }}_PUBLIC/Ghidra/Extensions
+    - name: Install pyyaml
+      run: sudo apt-get install -y libyaml-dev
+    - name: Install capa
+      run: pip install -e .[dev] 
+    - name: Run tests
+      run: | 
+        mkdir ./.github/ghidra/project
+        .github/ghidra/ghidra_${{ matrix.ghidra-version }}_PUBLIC/support/analyzeHeadless .github/ghidra/project ghidra_test -Import ./tests/data/mimikatz.exe_ -ScriptPath ./tests/ -PostScript test_ghidra_features.py > ../output.log
+        cat ../output.log
+        exit_code=$(cat ../output.log | grep exit | awk '{print $NF}')
+        exit $exit_code
+ 

.gitignore

@@ -108,9 +108,21 @@ venv.bak/
 *.viv
 *.idb
 *.i64
+.vscode
+
 !rules/lib
 
-# hooks/ci.sh output
-isort-output.log
-black-output.log
-rule-linter-output.log
+scripts/perf/*.txt
+scripts/perf/*.svg
+scripts/perf/*.zip
+
+.direnv
+.envrc
+.DS_Store
+*/.DS_Store
+Pipfile
+Pipfile.lock
+/cache/
+.github/binja/binaryninja
+.github/binja/download_headless.py
+.github/binja/BinaryNinja-headless.zip

.pre-commit-config.yaml

@@ -0,0 +1,111 @@
+# install the pre-commit hooks:
+#
+#    ❯ pre-commit install --hook-type pre-commit
+#    pre-commit installed at .git/hooks/pre-commit
+#
+#    ❯ pre-commit install --hook-type pre-push
+#    pre-commit installed at .git/hooks/pre-push
+#
+# run all linters liks:
+#
+#    ❯ pre-commit run --all-files
+#    isort....................................................................Passed
+#    black....................................................................Passed
+#    ruff.....................................................................Passed
+#    flake8...................................................................Passed
+#    mypy.....................................................................Passed
+#
+# run a single linter like:
+#
+#    ❯ pre-commit run --all-files isort
+#    isort....................................................................Passed
+
+repos:
+-   repo: local
+    hooks:
+    -   id: isort
+        name: isort
+        stages: [commit, push]
+        language: system
+        entry: isort
+        args: 
+        -   "--length-sort"
+        -   "--profile"
+        -   "black"
+        -   "--line-length=120"
+        -   "--skip-glob"
+        -   "*_pb2.py"
+        -   "capa/"
+        -   "scripts/"
+        -   "tests/"
+        always_run: true
+        pass_filenames: false
+
+-   repo: local
+    hooks:
+    -   id: black
+        name: black
+        stages: [commit, push]
+        language: system
+        entry: black
+        args: 
+        -   "--line-length=120"
+        -   "--extend-exclude"
+        -   ".*_pb2.py"
+        -   "capa/"
+        -   "scripts/"
+        -   "tests/"
+        always_run: true
+        pass_filenames: false
+
+-   repo: local
+    hooks:
+    -   id: ruff
+        name: ruff
+        stages: [commit, push]
+        language: system
+        entry: ruff
+        args: 
+        -   "check"
+        -   "--config"
+        -   ".github/ruff.toml"
+        -   "capa/"
+        -   "scripts/"
+        -   "tests/"
+        always_run: true
+        pass_filenames: false
+
+-   repo: local
+    hooks:
+    -   id: flake8
+        name: flake8
+        stages: [commit, push]
+        language: system
+        entry: flake8
+        args: 
+        -   "--config"
+        -   ".github/flake8.ini"
+        -   "--extend-exclude"
+        -   "capa/render/proto/capa_pb2.py"
+        -   "capa/"
+        -   "scripts/"
+        -   "tests/"
+        always_run: true
+        pass_filenames: false
+
+-   repo: local
+    hooks:
+    -   id: mypy
+        name: mypy
+        stages: [commit, push]
+        language: system
+        entry: mypy
+        args: 
+        -   "--check-untyped-defs"
+        -   "--ignore-missing-imports"
+        -   "--config-file=.github/mypy/mypy.ini"
+        -   "capa/"
+        -   "scripts/"
+        -   "tests/"
+        always_run: true
+        pass_filenames: false

LICENSE.txt

@@ -187,7 +187,7 @@
       same "printed page" as the copyright notice for easier
       identification within third-party archives.
 
-   Copyright (C) 2020 FireEye, Inc.
+   Copyright (C) 2023 Mandiant, Inc.
 
    Licensed under the Apache License, Version 2.0 (the "License");
    you may not use this file except in compliance with the License.

README.md

@@ -1,17 +1,21 @@
-![capa](.github/logo.png)
+![capa](https://github.com/mandiant/capa/blob/master/.github/logo.png)
 
 [![PyPI - Python Version](https://img.shields.io/pypi/pyversions/flare-capa)](https://pypi.org/project/flare-capa)
-[![Last release](https://img.shields.io/github/v/release/fireeye/capa)](https://github.com/fireeye/capa/releases)
-[![Number of rules](https://img.shields.io/badge/rules-485-blue.svg)](https://github.com/fireeye/capa-rules)
-[![CI status](https://github.com/fireeye/capa/workflows/CI/badge.svg)](https://github.com/fireeye/capa/actions?query=workflow%3ACI+event%3Apush+branch%3Amaster)
-[![Downloads](https://img.shields.io/github/downloads/fireeye/capa/total)](https://github.com/fireeye/capa/releases)
+[![Last release](https://img.shields.io/github/v/release/mandiant/capa)](https://github.com/mandiant/capa/releases)
+[![Number of rules](https://img.shields.io/badge/rules-832-blue.svg)](https://github.com/mandiant/capa-rules)
+[![CI status](https://github.com/mandiant/capa/workflows/CI/badge.svg)](https://github.com/mandiant/capa/actions?query=workflow%3ACI+event%3Apush+branch%3Amaster)
+[![Downloads](https://img.shields.io/github/downloads/mandiant/capa/total)](https://github.com/mandiant/capa/releases)
 [![License](https://img.shields.io/badge/license-Apache--2.0-green.svg)](LICENSE.txt)
 
 capa detects capabilities in executable files.
-You run it against a PE file or shellcode and it tells you what it thinks the program can do.
+You run it against a PE, ELF, .NET module, or shellcode file and it tells you what it thinks the program can do.
 For example, it might suggest that the file is a backdoor, is capable of installing services, or relies on HTTP to communicate.
 
-Check out the overview in our first [capa blog post](https://www.fireeye.com/blog/threat-research/2020/07/capa-automatically-identify-malware-capabilities.html).
+Check out:
+- the overview in our first [capa blog post](https://www.mandiant.com/resources/capa-automatically-identify-malware-capabilities)
+- the major version 2.0 updates described in our [second blog post](https://www.mandiant.com/resources/capa-2-better-stronger-faster)
+- the major version 3.0 (ELF support) described in the [third blog post](https://www.mandiant.com/resources/elfant-in-the-room-capa-v3)
+- the major version 4.0 (.NET support) described in the [fourth blog post](https://www.mandiant.com/resources/blog/capa-v4-casting-wider-net)
 
 ```
 $ capa.exe suspicious.exe
@@ -63,18 +67,11 @@ $ capa.exe suspicious.exe
 
 # download and usage
 
-Download stable releases of the standalone capa binaries [here](https://github.com/fireeye/capa/releases). You can run the standalone binaries without installation. capa is a command line tool that should be run from the terminal.
+Download stable releases of the standalone capa binaries [here](https://github.com/mandiant/capa/releases). You can run the standalone binaries without installation. capa is a command line tool that should be run from the terminal.
 
-<!--
-Alternatively, you can fetch a nightly build of a standalone binary from one of the following links. These are built using the latest development branch.
-- Windows 64bit: TODO
-- Linux: TODO
-- OSX: TODO
--->
+To use capa as a library or integrate with another tool, see [doc/installation.md](https://github.com/mandiant/capa/blob/master/doc/installation.md) for further setup instructions.
 
-To use capa as a library or integrate with another tool, see [doc/installation.md](doc/installation.md) for further setup instructions.
-
-For more information about how to use capa, see [doc/usage.md](doc/usage.md).
+For more information about how to use capa, see [doc/usage.md](https://github.com/mandiant/capa/blob/master/doc/usage.md).
 
 # example
 
@@ -91,31 +88,40 @@ This is useful for at least two reasons:
   - it shows where within the binary an experienced analyst might study with IDA Pro
 
 ```
-λ capa.exe suspicious.exe -vv
+$ capa.exe suspicious.exe -vv
 ...
 execute shell command and capture output
 namespace   c2/shell
-author      matthew.williams@fireeye.com
+author      matthew.williams@mandiant.com
 scope       function
 att&ck      Execution::Command and Scripting Interpreter::Windows Command Shell [T1059.003]
 references  https://docs.microsoft.com/en-us/windows/win32/api/processthreadsapi/ns-processthreadsapi-startupinfoa
-examples    Practical Malware Analysis Lab 14-02.exe_:0x4011C0
-function @ 0x10003A13
+function @ 0x4011C0
   and:
-    match: create a process with modified I/O handles and window @ 0x10003A13
+    match: create a process with modified I/O handles and window @ 0x4011C0
       and:
+        number: 257 = STARTF_USESTDHANDLES | STARTF_USESHOWWINDOW @ 0x4012B8
         or:
-          api: kernel32.CreateProcess @ 0x10003D6D
-        number: 0x101 @ 0x10003B03
-        or:
-          number: 0x44 @ 0x10003ADC
-        optional:
-          api: kernel32.GetStartupInfo @ 0x10003AE4
-    match: create pipe @ 0x10003A13
+          number: 68 = StartupInfo.cb (size) @ 0x401282
+        or: = API functions that accept a pointer to a STARTUPINFO structure
+          api: kernel32.CreateProcess @ 0x401343
+    match: create pipe @ 0x4011C0
       or:
-        api: kernel32.CreatePipe @ 0x10003ACB
+        api: kernel32.CreatePipe @ 0x40126F, 0x401280
+    optional:
+      match: create thread @ 0x40136A, 0x4013BA
+        or:
+          and:
+            os: windows
+            or:
+              api: kernel32.CreateThread @ 0x4013D7
+        or:
+          and:
+            os: windows
+            or:
+              api: kernel32.CreateThread @ 0x401395
     or:
-      string: cmd.exe /c  @ 0x10003AED
+      string: "cmd.exe" @ 0x4012FD
 ...
 ```
 
@@ -131,36 +137,51 @@ rule:
   meta:
     name: hash data with CRC32
     namespace: data-manipulation/checksum/crc32
-    author: moritz.raabe@fireeye.com
+    authors:
+      - moritz.raabe@mandiant.com
     scope: function
+    mbc:
+      - Data::Checksum::CRC32 [C0032.001]
     examples:
       - 2D3EDC218A90F03089CC01715A9F047F:0x403CBD
       - 7D28CB106CB54876B2A5C111724A07CD:0x402350  # RtlComputeCrc32
+      - 7EFF498DE13CC734262F87E6B3EF38AB:0x100084A6
   features:
     - or:
       - and:
         - mnemonic: shr
-        - number: 0xEDB88320
+        - or:
+          - number: 0xEDB88320
+          - bytes: 00 00 00 00 96 30 07 77 2C 61 0E EE BA 51 09 99 19 C4 6D 07 8F F4 6A 70 35 A5 63 E9 A3 95 64 9E = crc32_tab
         - number: 8
         - characteristic: nzxor
+      - and:
+        - number: 0x8320
+        - number: 0xEDB8
+        - characteristic: nzxor
       - api: RtlComputeCrc32
 ```
 
-The [github.com/fireeye/capa-rules](https://github.com/fireeye/capa-rules) repository contains hundreds of standard library rules that are distributed with capa.
+The [github.com/mandiant/capa-rules](https://github.com/mandiant/capa-rules) repository contains hundreds of standard library rules that are distributed with capa.
 Please learn to write rules and contribute new entries as you find interesting techniques in malware.
 
-If you use IDA Pro, then you can use the [capa explorer](capa/ida/plugin/) plugin.
+If you use IDA Pro, then you can use the [capa explorer](https://github.com/mandiant/capa/tree/master/capa/ida/plugin) plugin.
 capa explorer helps you identify interesting areas of a program and build new capa rules using features extracted directly from your IDA Pro database.
 
-![capa + IDA Pro integration](doc/img/explorer_expanded.png)
+![capa + IDA Pro integration](https://github.com/mandiant/capa/blob/master/doc/img/explorer_expanded.png)
+
+If you use Ghidra, you can use the Python 3 [Ghidra feature extractor](/capa/ghidra/). This integration enables capa to extract features directly from your Ghidra database, which can help you identify capabilities in programs that you analyze using Ghidra.
 
 # further information
 ## capa
-- [doc/installation](doc/installation.md)
-- [doc/usage](doc/usage.md)
-- [doc/limitations](doc/limitations.md)
-- [Contributing Guide](.github/CONTRIBUTING.md)
+- [Installation](https://github.com/mandiant/capa/blob/master/doc/installation.md)
+- [Usage](https://github.com/mandiant/capa/blob/master/doc/usage.md)
+- [Limitations](https://github.com/mandiant/capa/blob/master/doc/limitations.md)
+- [Contributing Guide](https://github.com/mandiant/capa/blob/master/.github/CONTRIBUTING.md)
 
 ## capa rules
-- [capa-rules repository](https://github.com/fireeye/capa-rules)
-- [capa-rules rule format](https://github.com/fireeye/capa-rules/blob/master/doc/format.md)
+- [capa-rules repository](https://github.com/mandiant/capa-rules)
+- [capa-rules rule format](https://github.com/mandiant/capa-rules/blob/master/doc/format.md)
+
+## capa testfiles
+The [capa-testfiles repository](https://github.com/mandiant/capa-testfiles) contains the data we use to test capa's code and rules

capa/engine.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
 # Licensed under the Apache License, Version 2.0 (the "License");
 #  you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at: [package root]/LICENSE.txt
@@ -8,11 +8,29 @@
 
 import copy
 import collections
+from typing import TYPE_CHECKING, Set, Dict, List, Tuple, Union, Mapping, Iterable, Iterator
 
-import capa.features
+import capa.perf
+import capa.features.common
+from capa.features.common import Result, Feature
+from capa.features.address import Address
+
+if TYPE_CHECKING:
+    # circular import, otherwise
+    import capa.rules
 
 
-class Statement(object):
+# a collection of features and the locations at which they are found.
+#
+# used throughout matching as the context in which features are searched:
+# to check if a feature exists, do: `Number(0x10) in features`.
+# to collect the locations of a feature, do: `features[Number(0x10)]`
+#
+# aliased here so that the type can be documented and xref'd.
+FeatureSet = Dict[Feature, Set[Address]]
+
+
+class Statement:
     """
     superclass for structural nodes, such as and/or/not.
     this exists to provide a default impl for `__str__` and `__repr__`,
@@ -20,157 +38,194 @@ class Statement(object):
     """
 
     def __init__(self, description=None):
-        super(Statement, self).__init__()
+        super().__init__()
         self.name = self.__class__.__name__
         self.description = description
 
     def __str__(self):
+        name = self.name.lower()
+        children = ",".join(map(str, self.get_children()))
         if self.description:
-            return "%s(%s = %s)" % (self.name.lower(), ",".join(map(str, self.get_children())), self.description)
+            return f"{name}({children} = {self.description})"
         else:
-            return "%s(%s)" % (self.name.lower(), ",".join(map(str, self.get_children())))
+            return f"{name}({children})"
 
     def __repr__(self):
         return str(self)
 
-    def evaluate(self, ctx):
+    def evaluate(self, features: FeatureSet, short_circuit=True) -> Result:
         """
         classes that inherit `Statement` must implement `evaluate`
 
         args:
-          ctx (defaultdict[Feature, set[VA]])
-
-        returns:
-          Result
+            short_circuit (bool): if true, then statements like and/or/some may short circuit.
         """
         raise NotImplementedError()
 
-    def get_children(self):
+    def get_children(self) -> Iterator[Union["Statement", Feature]]:
         if hasattr(self, "child"):
-            yield self.child
+            # this really confuses mypy because the property may not exist
+            # since its defined in the subclasses.
+            child = self.child  # type: ignore
+            assert isinstance(child, (Statement, Feature))
+            yield child
 
         if hasattr(self, "children"):
             for child in self.children:
+                assert isinstance(child, (Statement, Feature))
                 yield child
 
     def replace_child(self, existing, new):
         if hasattr(self, "child"):
-            if self.child is existing:
+            # this really confuses mypy because the property may not exist
+            # since its defined in the subclasses.
+            if self.child is existing:  # type: ignore
                 self.child = new
 
         if hasattr(self, "children"):
-            for i, child in enumerate(self.children):
+            children = self.children
+            for i, child in enumerate(children):
                 if child is existing:
-                    self.children[i] = new
-
-
-class Result(object):
-    """
-    represents the results of an evaluation of statements against features.
-
-    instances of this class should behave like a bool,
-    e.g. `assert Result(True, ...) == True`
-
-    instances track additional metadata about evaluation results.
-    they contain references to the statement node (e.g. an And statement),
-     as well as the children Result instances.
-
-    we need this so that we can render the tree of expressions and their results.
-    """
-
-    def __init__(self, success, statement, children, locations=None):
-        """
-        args:
-          success (bool)
-          statement (capa.engine.Statement or capa.features.Feature)
-          children (list[Result])
-          locations (iterable[VA])
-        """
-        super(Result, self).__init__()
-        self.success = success
-        self.statement = statement
-        self.children = children
-        self.locations = locations if locations is not None else ()
-
-    def __eq__(self, other):
-        if isinstance(other, bool):
-            return self.success == other
-        return False
-
-    def __bool__(self):
-        return self.success
-
-    def __nonzero__(self):
-        return self.success
+                    children[i] = new
 
 
 class And(Statement):
-    """match if all of the children evaluate to True."""
+    """
+    match if all of the children evaluate to True.
+
+    the order of evaluation is dictated by the property
+    `And.children` (type: List[Statement|Feature]).
+    a query optimizer may safely manipulate the order of these children.
+    """
 
     def __init__(self, children, description=None):
-        super(And, self).__init__(description=description)
+        super().__init__(description=description)
         self.children = children
 
-    def evaluate(self, ctx):
-        results = [child.evaluate(ctx) for child in self.children]
-        success = all(results)
-        return Result(success, self, results)
+    def evaluate(self, ctx, short_circuit=True):
+        capa.perf.counters["evaluate.feature"] += 1
+        capa.perf.counters["evaluate.feature.and"] += 1
+
+        if short_circuit:
+            results = []
+            for child in self.children:
+                result = child.evaluate(ctx, short_circuit=short_circuit)
+                results.append(result)
+                if not result:
+                    # short circuit
+                    return Result(False, self, results)
+
+            return Result(True, self, results)
+        else:
+            results = [child.evaluate(ctx, short_circuit=short_circuit) for child in self.children]
+            success = all(results)
+            return Result(success, self, results)
 
 
 class Or(Statement):
-    """match if any of the children evaluate to True."""
+    """
+    match if any of the children evaluate to True.
+
+    the order of evaluation is dictated by the property
+    `Or.children` (type: List[Statement|Feature]).
+    a query optimizer may safely manipulate the order of these children.
+    """
 
     def __init__(self, children, description=None):
-        super(Or, self).__init__(description=description)
+        super().__init__(description=description)
         self.children = children
 
-    def evaluate(self, ctx):
-        results = [child.evaluate(ctx) for child in self.children]
-        success = any(results)
-        return Result(success, self, results)
+    def evaluate(self, ctx, short_circuit=True):
+        capa.perf.counters["evaluate.feature"] += 1
+        capa.perf.counters["evaluate.feature.or"] += 1
+
+        if short_circuit:
+            results = []
+            for child in self.children:
+                result = child.evaluate(ctx, short_circuit=short_circuit)
+                results.append(result)
+                if result:
+                    # short circuit as soon as we hit one match
+                    return Result(True, self, results)
+
+            return Result(False, self, results)
+        else:
+            results = [child.evaluate(ctx, short_circuit=short_circuit) for child in self.children]
+            success = any(results)
+            return Result(success, self, results)
 
 
 class Not(Statement):
     """match only if the child evaluates to False."""
 
     def __init__(self, child, description=None):
-        super(Not, self).__init__(description=description)
+        super().__init__(description=description)
         self.child = child
 
-    def evaluate(self, ctx):
-        results = [self.child.evaluate(ctx)]
+    def evaluate(self, ctx, short_circuit=True):
+        capa.perf.counters["evaluate.feature"] += 1
+        capa.perf.counters["evaluate.feature.not"] += 1
+
+        results = [self.child.evaluate(ctx, short_circuit=short_circuit)]
         success = not results[0]
         return Result(success, self, results)
 
 
 class Some(Statement):
-    """match if at least N of the children evaluate to True."""
+    """
+    match if at least N of the children evaluate to True.
+
+    the order of evaluation is dictated by the property
+    `Some.children` (type: List[Statement|Feature]).
+    a query optimizer may safely manipulate the order of these children.
+    """
 
     def __init__(self, count, children, description=None):
-        super(Some, self).__init__(description=description)
+        super().__init__(description=description)
         self.count = count
         self.children = children
 
-    def evaluate(self, ctx):
-        results = [child.evaluate(ctx) for child in self.children]
-        # note that here we cast the child result as a bool
-        # because we've overridden `__bool__` above.
-        #
-        # we can't use `if child is True` because the instance is not True.
-        success = sum([1 for child in results if bool(child) is True]) >= self.count
-        return Result(success, self, results)
+    def evaluate(self, ctx, short_circuit=True):
+        capa.perf.counters["evaluate.feature"] += 1
+        capa.perf.counters["evaluate.feature.some"] += 1
+
+        if short_circuit:
+            results = []
+            satisfied_children_count = 0
+            for child in self.children:
+                result = child.evaluate(ctx, short_circuit=short_circuit)
+                results.append(result)
+                if result:
+                    satisfied_children_count += 1
+
+                if satisfied_children_count >= self.count:
+                    # short circuit as soon as we hit the threshold
+                    return Result(True, self, results)
+
+            return Result(False, self, results)
+        else:
+            results = [child.evaluate(ctx, short_circuit=short_circuit) for child in self.children]
+            # note that here we cast the child result as a bool
+            # because we've overridden `__bool__` above.
+            #
+            # we can't use `if child is True` because the instance is not True.
+            success = sum([1 for child in results if bool(child) is True]) >= self.count
+            return Result(success, self, results)
 
 
 class Range(Statement):
     """match if the child is contained in the ctx set with a count in the given range."""
 
     def __init__(self, child, min=None, max=None, description=None):
-        super(Range, self).__init__(description=description)
+        super().__init__(description=description)
         self.child = child
         self.min = min if min is not None else 0
         self.max = max if max is not None else (1 << 64 - 1)
 
-    def evaluate(self, ctx):
+    def evaluate(self, ctx, **kwargs):
+        capa.perf.counters["evaluate.feature"] += 1
+        capa.perf.counters["evaluate.feature.range"] += 1
+
         count = len(ctx.get(self.child, []))
         if self.min == 0 and count == 0:
             return Result(True, self, [])
@@ -179,9 +234,9 @@ class Range(Statement):
 
     def __str__(self):
         if self.max == (1 << 64 - 1):
-            return "range(%s, min=%d, max=infinity)" % (str(self.child), self.min)
+            return f"range({str(self.child)}, min={self.min}, max=infinity)"
         else:
-            return "range(%s, min=%d, max=%d)" % (str(self.child), self.min, self.max)
+            return f"range({str(self.child)}, min={self.min}, max={self.max})"
 
 
 class Subscope(Statement):
@@ -190,59 +245,66 @@ class Subscope(Statement):
     the engine should preprocess rules to extract subscope statements into their own rules.
     """
 
-    def __init__(self, scope, child):
-        super(Subscope, self).__init__()
+    def __init__(self, scope, child, description=None):
+        super().__init__(description=description)
         self.scope = scope
         self.child = child
 
-    def evaluate(self, ctx):
+    def evaluate(self, ctx, **kwargs):
         raise ValueError("cannot evaluate a subscope directly!")
 
 
-def topologically_order_rules(rules):
+# mapping from rule name to list of: (location of match, result object)
+#
+# used throughout matching and rendering to collection the results
+#  of statement evaluation and their locations.
+#
+# to check if a rule matched, do: `"TCP client" in matches`.
+# to find where a rule matched, do: `map(first, matches["TCP client"])`
+# to see how a rule matched, do:
+#
+#     for address, match_details in matches["TCP client"]:
+#         inspect(match_details)
+#
+# aliased here so that the type can be documented and xref'd.
+MatchResults = Mapping[str, List[Tuple[Address, Result]]]
+
+
+def index_rule_matches(features: FeatureSet, rule: "capa.rules.Rule", locations: Iterable[Address]):
     """
-    order the given rules such that dependencies show up before dependents.
-    this means that as we match rules, we can add features for the matches, and these
-     will be matched by subsequent rules if they follow this order.
+    record into the given featureset that the given rule matched at the given locations.
 
-    assumes that the rule dependency graph is a DAG.
+    naively, this is just adding a MatchedRule feature;
+    however, we also want to record matches for the rule's namespaces.
+
+    updates `features` in-place. doesn't modify the remaining arguments.
     """
-    # we evaluate `rules` multiple times, so if its a generator, realize it into a list.
-    rules = list(rules)
-    namespaces = capa.rules.index_rules_by_namespace(rules)
-    rules = {rule.name: rule for rule in rules}
-    seen = set([])
-    ret = []
-
-    def rec(rule):
-        if rule.name in seen:
-            return
-
-        for dep in rule.get_dependencies(namespaces):
-            rec(rules[dep])
-
-        ret.append(rule)
-        seen.add(rule.name)
-
-    for rule in rules.values():
-        rec(rule)
-
-    return ret
+    features[capa.features.common.MatchedRule(rule.name)].update(locations)
+    namespace = rule.meta.get("namespace")
+    if namespace:
+        while namespace:
+            features[capa.features.common.MatchedRule(namespace)].update(locations)
+            namespace, _, _ = namespace.rpartition("/")
 
 
-def match(rules, features, va):
+def match(rules: List["capa.rules.Rule"], features: FeatureSet, addr: Address) -> Tuple[FeatureSet, MatchResults]:
     """
-    Args:
-      rules (List[capa.rules.Rule]): these must already be ordered topologically by dependency.
-      features (Mapping[capa.features.Feature, int]):
-      va (int): location of the features
+    match the given rules against the given features,
+    returning an updated set of features and the matches.
 
-    Returns:
-      Tuple[List[capa.features.Feature], Dict[str, Tuple[int, capa.engine.Result]]]: two-tuple with entries:
-        - list of features used for matching (which may be greater than argument, due to rule match features), and
-        - mapping from rule name to (location of match, result object)
+    the updated features are just like the input,
+    but extended to include the match features (e.g. names of rules that matched).
+    the given feature set is not modified; an updated copy is returned.
+
+    the given list of rules must be ordered topologically by dependency,
+    or else `match` statements will not be handled correctly.
+
+    this routine should be fairly optimized, but is not guaranteed to be the fastest matcher possible.
+    it has a particularly convenient signature: (rules, features) -> matches
+    other strategies can be imagined that match differently; implement these elsewhere.
+    specifically, this routine does "top down" matching of the given rules against the feature set.
     """
-    results = collections.defaultdict(list)
+    results = collections.defaultdict(list)  # type: MatchResults
 
     # copy features so that we can modify it
     # without affecting the caller (keep this function pure)
@@ -251,15 +313,22 @@ def match(rules, features, va):
     features = collections.defaultdict(set, copy.copy(features))
 
     for rule in rules:
-        res = rule.evaluate(features)
+        res = rule.evaluate(features, short_circuit=True)
         if res:
-            results[rule.name].append((va, res))
-            features[capa.features.MatchedRule(rule.name)].add(va)
+            # we first matched the rule with short circuiting enabled.
+            # this is much faster than without short circuiting.
+            # however, we want to collect all results thoroughly,
+            # so once we've found a match quickly,
+            # go back and capture results without short circuiting.
+            res = rule.evaluate(features, short_circuit=False)
 
-            namespace = rule.meta.get("namespace")
-            if namespace:
-                while namespace:
-                    features[capa.features.MatchedRule(namespace)].add(va)
-                    namespace, _, _ = namespace.rpartition("/")
+            # sanity check
+            assert bool(res) is True
+
+            results[rule.name].append((addr, res))
+            # we need to update the current `features`
+            # because subsequent iterations of this loop may use newly added features,
+            # such as rule or namespace matches.
+            index_rule_matches(features, rule, [addr])
 
     return (features, results)

capa/exceptions.py

@@ -0,0 +1,21 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+class UnsupportedRuntimeError(RuntimeError):
+    pass
+
+
+class UnsupportedFormatError(ValueError):
+    pass
+
+
+class UnsupportedArchError(ValueError):
+    pass
+
+
+class UnsupportedOSError(ValueError):
+    pass

capa/features/__init__.py

@@ -1,236 +0,0 @@
-# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
-# Licensed under the Apache License, Version 2.0 (the "License");
-#  you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at: [package root]/LICENSE.txt
-# Unless required by applicable law or agreed to in writing, software distributed under the License
-#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and limitations under the License.
-
-import re
-import sys
-import codecs
-import logging
-
-import capa.engine
-
-logger = logging.getLogger(__name__)
-MAX_BYTES_FEATURE_SIZE = 0x100
-
-# thunks may be chained so we specify a delta to control the depth to which these chains are explored
-THUNK_CHAIN_DEPTH_DELTA = 5
-
-# identifiers for supported architectures names that tweak a feature
-# for example, offset/x32
-ARCH_X32 = "x32"
-ARCH_X64 = "x64"
-VALID_ARCH = (ARCH_X32, ARCH_X64)
-
-
-def bytes_to_str(b):
-    if sys.version_info[0] >= 3:
-        return str(codecs.encode(b, "hex").decode("utf-8"))
-    else:
-        return codecs.encode(b, "hex")
-
-
-def hex_string(h):
-    """ render hex string e.g. "0a40b1" as "0A 40 B1" """
-    return " ".join(h[i : i + 2] for i in range(0, len(h), 2)).upper()
-
-
-def escape_string(s):
-    """escape special characters"""
-    s = repr(s)
-    if not s.startswith(('"', "'")):
-        # u'hello\r\nworld' -> hello\\r\\nworld
-        s = s[2:-1]
-    else:
-        # 'hello\r\nworld' -> hello\\r\\nworld
-        s = s[1:-1]
-    s = s.replace("\\'", "'")  # repr() may escape "'" in some edge cases, remove
-    s = s.replace('"', '\\"')  # repr() does not escape '"', add
-    return s
-
-
-class Feature(object):
-    def __init__(self, value, arch=None, description=None):
-        """
-        Args:
-          value (any): the value of the feature, such as the number or string.
-          arch (str): one of the VALID_ARCH values, or None.
-            When None, then the feature applies to any architecture.
-            Modifies the feature name from `feature` to `feature/arch`, like `offset/x32`.
-          description (str): a human-readable description that explains the feature value.
-        """
-        super(Feature, self).__init__()
-
-        if arch is not None:
-            if arch not in VALID_ARCH:
-                raise ValueError("arch '%s' must be one of %s" % (arch, VALID_ARCH))
-            self.name = self.__class__.__name__.lower() + "/" + arch
-        else:
-            self.name = self.__class__.__name__.lower()
-
-        self.value = value
-        self.arch = arch
-        self.description = description
-
-    def __hash__(self):
-        return hash((self.name, self.value, self.arch))
-
-    def __eq__(self, other):
-        return self.name == other.name and self.value == other.value and self.arch == other.arch
-
-    def get_value_str(self):
-        """
-        render the value of this feature, for use by `__str__` and friends.
-        subclasses should override to customize the rendering.
-
-        Returns: any
-        """
-        return self.value
-
-    def __str__(self):
-        if self.value is not None:
-            if self.description:
-                return "%s(%s = %s)" % (self.name, self.get_value_str(), self.description)
-            else:
-                return "%s(%s)" % (self.name, self.get_value_str())
-        else:
-            return "%s" % self.name
-
-    def __repr__(self):
-        return str(self)
-
-    def evaluate(self, ctx):
-        return capa.engine.Result(self in ctx, self, [], locations=ctx.get(self, []))
-
-    def freeze_serialize(self):
-        if self.arch is not None:
-            return (self.__class__.__name__, [self.value, {"arch": self.arch}])
-        else:
-            return (self.__class__.__name__, [self.value])
-
-    @classmethod
-    def freeze_deserialize(cls, args):
-        # as you can see below in code,
-        # if the last argument is a dictionary,
-        # consider it to be kwargs passed to the feature constructor.
-        if len(args) == 1:
-            return cls(*args)
-        elif isinstance(args[-1], dict):
-            kwargs = args[-1]
-            args = args[:-1]
-            return cls(*args, **kwargs)
-
-
-class MatchedRule(Feature):
-    def __init__(self, value, description=None):
-        super(MatchedRule, self).__init__(value, description=description)
-        self.name = "match"
-
-
-class Characteristic(Feature):
-    def __init__(self, value, description=None):
-        super(Characteristic, self).__init__(value, description=description)
-
-
-class String(Feature):
-    def __init__(self, value, description=None):
-        super(String, self).__init__(value, description=description)
-
-
-class Regex(String):
-    def __init__(self, value, description=None):
-        super(Regex, self).__init__(value, description=description)
-        pat = self.value[len("/") : -len("/")]
-        flags = re.DOTALL
-        if value.endswith("/i"):
-            pat = self.value[len("/") : -len("/i")]
-            flags |= re.IGNORECASE
-        try:
-            self.re = re.compile(pat, flags)
-        except re.error:
-            if value.endswith("/i"):
-                value = value[: -len("i")]
-            raise ValueError(
-                "invalid regular expression: %s it should use Python syntax, try it at https://pythex.org" % value
-            )
-
-    def evaluate(self, ctx):
-        for feature, locations in ctx.items():
-            if not isinstance(feature, (capa.features.String,)):
-                continue
-
-            # `re.search` finds a match anywhere in the given string
-            # which implies leading and/or trailing whitespace.
-            # using this mode cleans is more convenient for rule authors,
-            # so that they don't have to prefix/suffix their terms like: /.*foo.*/.
-            if self.re.search(feature.value):
-                # unlike other features, we cannot return put a reference to `self` directly in a `Result`.
-                # this is because `self` may match on many strings, so we can't stuff the matched value into it.
-                # instead, return a new instance that has a reference to both the regex and the matched value.
-                # see #262.
-                return capa.engine.Result(True, _MatchedRegex(self, feature.value), [], locations=locations)
-
-        return capa.engine.Result(False, _MatchedRegex(self, None), [])
-
-    def __str__(self):
-        return "regex(string =~ %s)" % self.value
-
-
-class _MatchedRegex(Regex):
-    """
-    this represents a specific instance of a regular expression feature match.
-    treat it the same as a `Regex` except it has the `match` field that contains the complete string that matched.
-
-    note: this type should only ever be constructed by `Regex.evaluate()`. it is not part of the public API.
-    """
-
-    def __init__(self, regex, match):
-        """
-        args:
-          regex (Regex): the regex feature that matches
-          match (string|None): the matching string or None if it doesn't match
-        """
-        super(_MatchedRegex, self).__init__(regex.value, description=regex.description)
-        # we want this to collide with the name of `Regex` above,
-        # so that it works nicely with the renderers.
-        self.name = "regex"
-        # this may be None if the regex doesn't match
-        self.match = match
-
-    def __str__(self):
-        return 'regex(string =~ %s, matched = "%s")' % (self.value, self.match)
-
-
-class StringFactory(object):
-    def __new__(self, value, description=None):
-        if value.startswith("/") and (value.endswith("/") or value.endswith("/i")):
-            return Regex(value, description=description)
-        return String(value, description=description)
-
-
-class Bytes(Feature):
-    def __init__(self, value, description=None):
-        super(Bytes, self).__init__(value, description=description)
-
-    def evaluate(self, ctx):
-        for feature, locations in ctx.items():
-            if not isinstance(feature, (capa.features.Bytes,)):
-                continue
-
-            if feature.value.startswith(self.value):
-                return capa.engine.Result(True, self, [], locations=locations)
-
-        return capa.engine.Result(False, self, [])
-
-    def get_value_str(self):
-        return hex_string(bytes_to_str(self.value))
-
-    def freeze_serialize(self):
-        return (self.__class__.__name__, [bytes_to_str(self.value).upper()])
-
-    @classmethod
-    def freeze_deserialize(cls, args):
-        return cls(*[codecs.decode(x, "hex") for x in args])

capa/features/address.py

@@ -0,0 +1,121 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import abc
+
+
+class Address(abc.ABC):
+    @abc.abstractmethod
+    def __eq__(self, other):
+        ...
+
+    @abc.abstractmethod
+    def __lt__(self, other):
+        # implement < so that addresses can be sorted from low to high
+        ...
+
+    @abc.abstractmethod
+    def __hash__(self):
+        # implement hash so that addresses can be used in sets and dicts
+        ...
+
+    @abc.abstractmethod
+    def __repr__(self):
+        # implement repr to help during debugging
+        ...
+
+
+class AbsoluteVirtualAddress(int, Address):
+    """an absolute memory address"""
+
+    def __new__(cls, v):
+        assert v >= 0
+        return int.__new__(cls, v)
+
+    def __repr__(self):
+        return f"absolute(0x{self:x})"
+
+    def __hash__(self):
+        return int.__hash__(self)
+
+
+class RelativeVirtualAddress(int, Address):
+    """a memory address relative to a base address"""
+
+    def __repr__(self):
+        return f"relative(0x{self:x})"
+
+    def __hash__(self):
+        return int.__hash__(self)
+
+
+class FileOffsetAddress(int, Address):
+    """an address relative to the start of a file"""
+
+    def __new__(cls, v):
+        assert v >= 0
+        return int.__new__(cls, v)
+
+    def __repr__(self):
+        return f"file(0x{self:x})"
+
+    def __hash__(self):
+        return int.__hash__(self)
+
+
+class DNTokenAddress(int, Address):
+    """a .NET token"""
+
+    def __new__(cls, token: int):
+        return int.__new__(cls, token)
+
+    def __repr__(self):
+        return f"token(0x{self:x})"
+
+    def __hash__(self):
+        return int.__hash__(self)
+
+
+class DNTokenOffsetAddress(Address):
+    """an offset into an object specified by a .NET token"""
+
+    def __init__(self, token: int, offset: int):
+        assert offset >= 0
+        self.token = token
+        self.offset = offset
+
+    def __eq__(self, other):
+        return (self.token, self.offset) == (other.token, other.offset)
+
+    def __lt__(self, other):
+        return (self.token, self.offset) < (other.token, other.offset)
+
+    def __hash__(self):
+        return hash((self.token, self.offset))
+
+    def __repr__(self):
+        return f"token(0x{self.token:x})+(0x{self.offset:x})"
+
+    def __index__(self):
+        return self.token + self.offset
+
+
+class _NoAddress(Address):
+    def __eq__(self, other):
+        return True
+
+    def __lt__(self, other):
+        return False
+
+    def __hash__(self):
+        return hash(0)
+
+    def __repr__(self):
+        return "no address"
+
+
+NO_ADDRESS = _NoAddress()

capa/features/basicblock.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
 # Licensed under the Apache License, Version 2.0 (the "License");
 #  you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at: [package root]/LICENSE.txt
@@ -6,22 +6,15 @@
 #  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and limitations under the License.
 
-from capa.features import Feature
+from capa.features.common import Feature
 
 
 class BasicBlock(Feature):
-    def __init__(self):
-        super(BasicBlock, self).__init__(None)
+    def __init__(self, description=None):
+        super().__init__(0, description=description)
 
     def __str__(self):
         return "basic block"
 
     def get_value_str(self):
         return ""
-
-    def freeze_serialize(self):
-        return (self.__class__.__name__, [])
-
-    @classmethod
-    def freeze_deserialize(cls, args):
-        return cls()

capa/features/common.py

@@ -0,0 +1,476 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+
+import re
+import abc
+import codecs
+import typing
+import logging
+import collections
+from typing import TYPE_CHECKING, Set, Dict, List, Union, Optional
+
+if TYPE_CHECKING:
+    # circular import, otherwise
+    import capa.engine
+
+import capa.perf
+import capa.features
+import capa.features.extractors.elf
+from capa.features.address import Address
+
+logger = logging.getLogger(__name__)
+MAX_BYTES_FEATURE_SIZE = 0x100
+
+# thunks may be chained so we specify a delta to control the depth to which these chains are explored
+THUNK_CHAIN_DEPTH_DELTA = 5
+
+
+class FeatureAccess:
+    READ = "read"
+    WRITE = "write"
+
+
+VALID_FEATURE_ACCESS = (FeatureAccess.READ, FeatureAccess.WRITE)
+
+
+def bytes_to_str(b: bytes) -> str:
+    return str(codecs.encode(b, "hex").decode("utf-8"))
+
+
+def hex_string(h: str) -> str:
+    """render hex string e.g. "0a40b1" as "0A 40 B1" """
+    return " ".join(h[i : i + 2] for i in range(0, len(h), 2)).upper()
+
+
+def escape_string(s: str) -> str:
+    """escape special characters"""
+    s = repr(s)
+    if not s.startswith(('"', "'")):
+        # u'hello\r\nworld' -> hello\\r\\nworld
+        s = s[2:-1]
+    else:
+        # 'hello\r\nworld' -> hello\\r\\nworld
+        s = s[1:-1]
+    s = s.replace("\\'", "'")  # repr() may escape "'" in some edge cases, remove
+    s = s.replace('"', '\\"')  # repr() does not escape '"', add
+    return s
+
+
+class Result:
+    """
+    represents the results of an evaluation of statements against features.
+
+    instances of this class should behave like a bool,
+    e.g. `assert Result(True, ...) == True`
+
+    instances track additional metadata about evaluation results.
+    they contain references to the statement node (e.g. an And statement),
+     as well as the children Result instances.
+
+    we need this so that we can render the tree of expressions and their results.
+    """
+
+    def __init__(
+        self,
+        success: bool,
+        statement: Union["capa.engine.Statement", "Feature"],
+        children: List["Result"],
+        locations: Optional[Set[Address]] = None,
+    ):
+        super().__init__()
+        self.success = success
+        self.statement = statement
+        self.children = children
+        self.locations = locations if locations is not None else set()
+
+    def __eq__(self, other):
+        if isinstance(other, bool):
+            return self.success == other
+        return False
+
+    def __bool__(self):
+        return self.success
+
+    def __nonzero__(self):
+        return self.success
+
+
+class Feature(abc.ABC):  # noqa: B024
+    # this is an abstract class, since we don't want anyone to instantiate it directly,
+    # but it doesn't have any abstract methods.
+
+    def __init__(
+        self,
+        value: Union[str, int, float, bytes],
+        description: Optional[str] = None,
+    ):
+        """
+        Args:
+          value (any): the value of the feature, such as the number or string.
+          description (str): a human-readable description that explains the feature value.
+        """
+        super().__init__()
+
+        self.name = self.__class__.__name__.lower()
+        self.value = value
+        self.description = description
+
+    def __hash__(self):
+        return hash((self.name, self.value))
+
+    def __eq__(self, other):
+        return self.name == other.name and self.value == other.value
+
+    def __lt__(self, other):
+        # implementing sorting by serializing to JSON is a huge hack.
+        # its slow, inelegant, and probably doesn't work intuitively;
+        # however, we only use it for deterministic output, so it's good enough for now.
+
+        # circular import
+        # we should fix if this wasn't already a huge hack.
+        import capa.features.freeze.features
+
+        return (
+            capa.features.freeze.features.feature_from_capa(self).model_dump_json()
+            < capa.features.freeze.features.feature_from_capa(other).model_dump_json()
+        )
+
+    def get_name_str(self) -> str:
+        """
+        render the name of this feature, for use by `__str__` and friends.
+        subclasses should override to customize the rendering.
+        """
+        return self.name
+
+    def get_value_str(self) -> str:
+        """
+        render the value of this feature, for use by `__str__` and friends.
+        subclasses should override to customize the rendering.
+        """
+        return str(self.value)
+
+    def __str__(self):
+        if self.value is not None:
+            if self.description:
+                return f"{self.get_name_str()}({self.get_value_str()} = {self.description})"
+            else:
+                return f"{self.get_name_str()}({self.get_value_str()})"
+        else:
+            return f"{self.get_name_str()}"
+
+    def __repr__(self):
+        return str(self)
+
+    def evaluate(self, ctx: Dict["Feature", Set[Address]], **kwargs) -> Result:
+        capa.perf.counters["evaluate.feature"] += 1
+        capa.perf.counters["evaluate.feature." + self.name] += 1
+        return Result(self in ctx, self, [], locations=ctx.get(self, set()))
+
+
+class MatchedRule(Feature):
+    def __init__(self, value: str, description=None):
+        super().__init__(value, description=description)
+        self.name = "match"
+
+
+class Characteristic(Feature):
+    def __init__(self, value: str, description=None):
+        super().__init__(value, description=description)
+
+
+class String(Feature):
+    def __init__(self, value: str, description=None):
+        super().__init__(value, description=description)
+
+    def get_value_str(self) -> str:
+        assert isinstance(self.value, str)
+        return escape_string(self.value)
+
+
+class Class(Feature):
+    def __init__(self, value: str, description=None):
+        super().__init__(value, description=description)
+
+
+class Namespace(Feature):
+    def __init__(self, value: str, description=None):
+        super().__init__(value, description=description)
+
+
+class Substring(String):
+    def __init__(self, value: str, description=None):
+        super().__init__(value, description=description)
+        self.value = value
+
+    def evaluate(self, ctx, short_circuit=True):
+        capa.perf.counters["evaluate.feature"] += 1
+        capa.perf.counters["evaluate.feature.substring"] += 1
+
+        # mapping from string value to list of locations.
+        # will unique the locations later on.
+        matches: typing.DefaultDict[str, Set[Address]] = collections.defaultdict(set)
+
+        assert isinstance(self.value, str)
+        for feature, locations in ctx.items():
+            if not isinstance(feature, (String,)):
+                continue
+
+            if not isinstance(feature.value, str):
+                # this is a programming error: String should only contain str
+                raise ValueError("unexpected feature value type")
+
+            if self.value in feature.value:
+                matches[feature.value].update(locations)
+                if short_circuit:
+                    # we found one matching string, thats sufficient to match.
+                    # don't collect other matching strings in this mode.
+                    break
+
+        if matches:
+            # collect all locations
+            locations = set()
+            for locs in matches.values():
+                locations.update(locs)
+
+            # unlike other features, we cannot return put a reference to `self` directly in a `Result`.
+            # this is because `self` may match on many strings, so we can't stuff the matched value into it.
+            # instead, return a new instance that has a reference to both the substring and the matched values.
+            return Result(True, _MatchedSubstring(self, dict(matches)), [], locations=locations)
+        else:
+            return Result(False, _MatchedSubstring(self, {}), [])
+
+    def get_value_str(self) -> str:
+        assert isinstance(self.value, str)
+        return escape_string(self.value)
+
+    def __str__(self):
+        assert isinstance(self.value, str)
+        return f"substring({escape_string(self.value)})"
+
+
+class _MatchedSubstring(Substring):
+    """
+    this represents specific match instances of a substring feature.
+    treat it the same as a `Substring` except it has the `matches` field that contains the complete strings that matched.
+
+    note: this type should only ever be constructed by `Substring.evaluate()`. it is not part of the public API.
+    """
+
+    def __init__(self, substring: Substring, matches: Dict[str, Set[Address]]):
+        """
+        args:
+          substring: the substring feature that matches.
+          match: mapping from matching string to its locations.
+        """
+        super().__init__(str(substring.value), description=substring.description)
+        # we want this to collide with the name of `Substring` above,
+        # so that it works nicely with the renderers.
+        self.name = "substring"
+        # this may be None if the substring doesn't match
+        self.matches = matches
+
+    def __str__(self):
+        matches = ", ".join(f'"{s}"' for s in (self.matches or {}).keys())
+        assert isinstance(self.value, str)
+        return f'substring("{self.value}", matches = {matches})'
+
+
+class Regex(String):
+    def __init__(self, value: str, description=None):
+        super().__init__(value, description=description)
+        self.value = value
+
+        pat = self.value[len("/") : -len("/")]
+        flags = re.DOTALL
+        if value.endswith("/i"):
+            pat = self.value[len("/") : -len("/i")]
+            flags |= re.IGNORECASE
+        try:
+            self.re = re.compile(pat, flags)
+        except re.error as exc:
+            if value.endswith("/i"):
+                value = value[: -len("i")]
+            raise ValueError(
+                f"invalid regular expression: {value} it should use Python syntax, try it at https://pythex.org"
+            ) from exc
+
+    def evaluate(self, ctx, short_circuit=True):
+        capa.perf.counters["evaluate.feature"] += 1
+        capa.perf.counters["evaluate.feature.regex"] += 1
+
+        # mapping from string value to list of locations.
+        # will unique the locations later on.
+        matches: typing.DefaultDict[str, Set[Address]] = collections.defaultdict(set)
+
+        for feature, locations in ctx.items():
+            if not isinstance(feature, (String,)):
+                continue
+
+            if not isinstance(feature.value, str):
+                # this is a programming error: String should only contain str
+                raise ValueError("unexpected feature value type")
+
+            # `re.search` finds a match anywhere in the given string
+            # which implies leading and/or trailing whitespace.
+            # using this mode cleans is more convenient for rule authors,
+            # so that they don't have to prefix/suffix their terms like: /.*foo.*/.
+            if self.re.search(feature.value):
+                matches[feature.value].update(locations)
+                if short_circuit:
+                    # we found one matching string, thats sufficient to match.
+                    # don't collect other matching strings in this mode.
+                    break
+
+        if matches:
+            # collect all locations
+            locations = set()
+            for locs in matches.values():
+                locations.update(locs)
+
+            # unlike other features, we cannot return put a reference to `self` directly in a `Result`.
+            # this is because `self` may match on many strings, so we can't stuff the matched value into it.
+            # instead, return a new instance that has a reference to both the regex and the matched values.
+            # see #262.
+            return Result(True, _MatchedRegex(self, dict(matches)), [], locations=locations)
+        else:
+            return Result(False, _MatchedRegex(self, {}), [])
+
+    def __str__(self):
+        assert isinstance(self.value, str)
+        return f"regex(string =~ {self.value})"
+
+
+class _MatchedRegex(Regex):
+    """
+    this represents specific match instances of a regular expression feature.
+    treat it the same as a `Regex` except it has the `matches` field that contains the complete strings that matched.
+
+    note: this type should only ever be constructed by `Regex.evaluate()`. it is not part of the public API.
+    """
+
+    def __init__(self, regex: Regex, matches: Dict[str, Set[Address]]):
+        """
+        args:
+          regex: the regex feature that matches.
+          matches: mapping from matching string to its locations.
+        """
+        super().__init__(str(regex.value), description=regex.description)
+        # we want this to collide with the name of `Regex` above,
+        # so that it works nicely with the renderers.
+        self.name = "regex"
+        # this may be None if the regex doesn't match
+        self.matches = matches
+
+    def __str__(self):
+        matches = ", ".join(f'"{s}"' for s in (self.matches or {}).keys())
+        assert isinstance(self.value, str)
+        return f"regex(string =~ {self.value}, matches = {matches})"
+
+
+class StringFactory:
+    def __new__(cls, value: str, description=None):
+        if value.startswith("/") and (value.endswith("/") or value.endswith("/i")):
+            return Regex(value, description=description)
+        return String(value, description=description)
+
+
+class Bytes(Feature):
+    def __init__(self, value: bytes, description=None):
+        super().__init__(value, description=description)
+        self.value = value
+
+    def evaluate(self, ctx, **kwargs):
+        capa.perf.counters["evaluate.feature"] += 1
+        capa.perf.counters["evaluate.feature.bytes"] += 1
+
+        assert isinstance(self.value, bytes)
+        for feature, locations in ctx.items():
+            if not isinstance(feature, (Bytes,)):
+                continue
+
+            assert isinstance(feature.value, bytes)
+            if feature.value.startswith(self.value):
+                return Result(True, self, [], locations=locations)
+
+        return Result(False, self, [])
+
+    def get_value_str(self):
+        assert isinstance(self.value, bytes)
+        return hex_string(bytes_to_str(self.value))
+
+
+# other candidates here: https://docs.microsoft.com/en-us/windows/win32/debug/pe-format#machine-types
+ARCH_I386 = "i386"
+ARCH_AMD64 = "amd64"
+# dotnet
+ARCH_ANY = "any"
+VALID_ARCH = (ARCH_I386, ARCH_AMD64, ARCH_ANY)
+
+
+class Arch(Feature):
+    def __init__(self, value: str, description=None):
+        super().__init__(value, description=description)
+        self.name = "arch"
+
+
+OS_WINDOWS = "windows"
+OS_LINUX = "linux"
+OS_MACOS = "macos"
+# dotnet
+OS_ANY = "any"
+VALID_OS = {os.value for os in capa.features.extractors.elf.OS}
+VALID_OS.update({OS_WINDOWS, OS_LINUX, OS_MACOS, OS_ANY})
+# internal only, not to be used in rules
+OS_AUTO = "auto"
+
+
+class OS(Feature):
+    def __init__(self, value: str, description=None):
+        super().__init__(value, description=description)
+        self.name = "os"
+
+    def evaluate(self, ctx, **kwargs):
+        capa.perf.counters["evaluate.feature"] += 1
+        capa.perf.counters["evaluate.feature." + self.name] += 1
+
+        for feature, locations in ctx.items():
+            if not isinstance(feature, (OS,)):
+                continue
+
+            assert isinstance(feature.value, str)
+            if OS_ANY in (self.value, feature.value) or self.value == feature.value:
+                return Result(True, self, [], locations=locations)
+
+        return Result(False, self, [])
+
+
+FORMAT_PE = "pe"
+FORMAT_ELF = "elf"
+FORMAT_DOTNET = "dotnet"
+VALID_FORMAT = (FORMAT_PE, FORMAT_ELF, FORMAT_DOTNET)
+# internal only, not to be used in rules
+FORMAT_AUTO = "auto"
+FORMAT_SC32 = "sc32"
+FORMAT_SC64 = "sc64"
+FORMAT_FREEZE = "freeze"
+FORMAT_RESULT = "result"
+FORMAT_UNKNOWN = "unknown"
+
+
+class Format(Feature):
+    def __init__(self, value: str, description=None):
+        super().__init__(value, description=description)
+        self.name = "format"
+
+
+def is_global_feature(feature):
+    """
+    is this a feature that is extracted at every scope?
+    today, these are OS and arch features.
+    """
+    return isinstance(feature, (OS, Arch))

capa/features/extractors/__init__.py

@@ -1,286 +0,0 @@
-# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
-# Licensed under the Apache License, Version 2.0 (the "License");
-#  you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at: [package root]/LICENSE.txt
-# Unless required by applicable law or agreed to in writing, software distributed under the License
-#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and limitations under the License.
-
-import abc
-
-
-class FeatureExtractor(object):
-    """
-    FeatureExtractor defines the interface for fetching features from a sample.
-
-    There may be multiple backends that support fetching features for capa.
-    For example, we use vivisect by default, but also want to support saving
-     and restoring features from a JSON file.
-    When we restore the features, we'd like to use exactly the same matching logic
-     to find matching rules.
-    Therefore, we can define a FeatureExtractor that provides features from the
-     serialized JSON file and do matching without a binary analysis pass.
-    Also, this provides a way to hook in an IDA backend.
-
-    This class is not instantiated directly; it is the base class for other implementations.
-    """
-
-    __metaclass__ = abc.ABCMeta
-
-    def __init__(self):
-        #
-        # note: a subclass should define ctor parameters for its own use.
-        #  for example, the Vivisect feature extract might require the vw and/or path.
-        # this base class doesn't know what to do with that info, though.
-        #
-        super(FeatureExtractor, self).__init__()
-
-    @abc.abstractmethod
-    def get_base_address(self):
-        """
-        fetch the preferred load address at which the sample was analyzed.
-
-        returns: int
-        """
-        raise NotImplemented
-
-    @abc.abstractmethod
-    def extract_file_features(self):
-        """
-        extract file-scope features.
-
-        example::
-
-            extractor = VivisectFeatureExtractor(vw, path)
-            for feature, va in extractor.get_file_features():
-                print('0x%x: %s', va, feature)
-
-        yields:
-          Tuple[capa.features.Feature, int]: feature and its location
-        """
-        raise NotImplemented
-
-    @abc.abstractmethod
-    def get_functions(self):
-        """
-        enumerate the functions and provide opaque values that will
-         subsequently be provided to `.extract_function_features()`, etc.
-
-        by "opaque value", we mean that this can be any object, as long as it
-         provides enough context to `.extract_function_features()`.
-
-        the opaque value should support casting to int (`__int__`) for the function start address.
-
-        yields:
-          any: the opaque function value.
-        """
-        raise NotImplemented
-
-    @abc.abstractmethod
-    def extract_function_features(self, f):
-        """
-        extract function-scope features.
-        the arguments are opaque values previously provided by `.get_functions()`, etc.
-
-        example::
-
-            extractor = VivisectFeatureExtractor(vw, path)
-            for function in extractor.get_functions():
-                for feature, va in extractor.extract_function_features(function):
-                    print('0x%x: %s', va, feature)
-
-        args:
-          f [any]: an opaque value previously fetched from `.get_functions()`.
-
-        yields:
-          Tuple[capa.features.Feature, int]: feature and its location
-        """
-        raise NotImplemented
-
-    @abc.abstractmethod
-    def get_basic_blocks(self, f):
-        """
-        enumerate the basic blocks in the given function and provide opaque values that will
-         subsequently be provided to `.extract_basic_block_features()`, etc.
-
-        by "opaque value", we mean that this can be any object, as long as it
-         provides enough context to `.extract_basic_block_features()`.
-
-        the opaque value should support casting to int (`__int__`) for the basic block start address.
-
-        yields:
-          any: the opaque basic block value.
-        """
-        raise NotImplemented
-
-    @abc.abstractmethod
-    def extract_basic_block_features(self, f, bb):
-        """
-        extract basic block-scope features.
-        the arguments are opaque values previously provided by `.get_functions()`, etc.
-
-        example::
-
-            extractor = VivisectFeatureExtractor(vw, path)
-            for function in extractor.get_functions():
-                for bb in extractor.get_basic_blocks(function):
-                    for feature, va in extractor.extract_basic_block_features(function, bb):
-                        print('0x%x: %s', va, feature)
-
-        args:
-          f [any]: an opaque value previously fetched from `.get_functions()`.
-          bb [any]: an opaque value previously fetched from `.get_basic_blocks()`.
-
-        yields:
-          Tuple[capa.features.Feature, int]: feature and its location
-        """
-        raise NotImplemented
-
-    @abc.abstractmethod
-    def get_instructions(self, f, bb):
-        """
-        enumerate the instructions in the given basic block and provide opaque values that will
-         subsequently be provided to `.extract_insn_features()`, etc.
-
-        by "opaque value", we mean that this can be any object, as long as it
-         provides enough context to `.extract_insn_features()`.
-
-        the opaque value should support casting to int (`__int__`) for the instruction address.
-
-        yields:
-          any: the opaque function value.
-        """
-        raise NotImplemented
-
-    @abc.abstractmethod
-    def extract_insn_features(self, f, bb, insn):
-        """
-        extract instruction-scope features.
-        the arguments are opaque values previously provided by `.get_functions()`, etc.
-
-        example::
-
-            extractor = VivisectFeatureExtractor(vw, path)
-            for function in extractor.get_functions():
-                for bb in extractor.get_basic_blocks(function):
-                    for insn in extractor.get_instructions(function, bb):
-                        for feature, va in extractor.extract_insn_features(function, bb, insn):
-                            print('0x%x: %s', va, feature)
-
-        args:
-          f [any]: an opaque value previously fetched from `.get_functions()`.
-          bb [any]: an opaque value previously fetched from `.get_basic_blocks()`.
-          insn [any]: an opaque value previously fetched from `.get_instructions()`.
-
-        yields:
-          Tuple[capa.features.Feature, int]: feature and its location
-        """
-        raise NotImplemented
-
-
-class NullFeatureExtractor(FeatureExtractor):
-    """
-    An extractor that extracts some user-provided features.
-    The structure of the single parameter is demonstrated in the example below.
-
-    This is useful for testing, as we can provide expected values and see if matching works.
-    Also, this is how we represent features deserialized from a freeze file.
-
-    example::
-
-        extractor = NullFeatureExtractor({
-            'base address: 0x401000,
-            'file features': [
-                (0x402345, capa.features.Characteristic('embedded pe')),
-            ],
-            'functions': {
-                0x401000: {
-                    'features': [
-                        (0x401000, capa.features.Characteristic('nzxor')),
-                    ],
-                    'basic blocks': {
-                        0x401000: {
-                            'features': [
-                                (0x401000, capa.features.Characteristic('tight-loop')),
-                            ],
-                            'instructions': {
-                                0x401000: {
-                                    'features': [
-                                        (0x401000, capa.features.Characteristic('nzxor')),
-                                    ],
-                                },
-                                0x401002: ...
-                            }
-                        },
-                        0x401005: ...
-                    }
-                },
-                0x40200: ...
-            }
-        )
-    """
-
-    def __init__(self, features):
-        super(NullFeatureExtractor, self).__init__()
-        self.features = features
-
-    def get_base_address(self):
-        return self.features["base address"]
-
-    def extract_file_features(self):
-        for p in self.features.get("file features", []):
-            va, feature = p
-            yield feature, va
-
-    def get_functions(self):
-        for va in sorted(self.features["functions"].keys()):
-            yield va
-
-    def extract_function_features(self, f):
-        for p in self.features.get("functions", {}).get(f, {}).get("features", []):  # noqa: E127 line over-indented
-            va, feature = p
-            yield feature, va
-
-    def get_basic_blocks(self, f):
-        for va in sorted(
-            self.features.get("functions", {})  # noqa: E127 line over-indented
-            .get(f, {})
-            .get("basic blocks", {})
-            .keys()
-        ):
-            yield va
-
-    def extract_basic_block_features(self, f, bb):
-        for p in (
-            self.features.get("functions", {})  # noqa: E127 line over-indented
-            .get(f, {})
-            .get("basic blocks", {})
-            .get(bb, {})
-            .get("features", [])
-        ):
-            va, feature = p
-            yield feature, va
-
-    def get_instructions(self, f, bb):
-        for va in sorted(
-            self.features.get("functions", {})  # noqa: E127 line over-indented
-            .get(f, {})
-            .get("basic blocks", {})
-            .get(bb, {})
-            .get("instructions", {})
-            .keys()
-        ):
-            yield va
-
-    def extract_insn_features(self, f, bb, insn):
-        for p in (
-            self.features.get("functions", {})  # noqa: E127 line over-indented
-            .get(f, {})
-            .get("basic blocks", {})
-            .get(bb, {})
-            .get("instructions", {})
-            .get(insn, {})
-            .get("features", [])
-        ):
-            va, feature = p
-            yield feature, va

capa/features/extractors/base_extractor.py

@@ -0,0 +1,264 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+
+import abc
+import dataclasses
+from typing import Any, Dict, Tuple, Union, Iterator
+from dataclasses import dataclass
+
+import capa.features.address
+from capa.features.common import Feature
+from capa.features.address import Address, AbsoluteVirtualAddress
+
+# feature extractors may reference functions, BBs, insns by opaque handle values.
+# you can use the `.address` property to get and render the address of the feature.
+#
+# these handles are only consumed by routines on
+# the feature extractor from which they were created.
+
+
+@dataclass
+class FunctionHandle:
+    """reference to a function recognized by a feature extractor.
+
+    Attributes:
+        address: the address of the function.
+        inner: extractor-specific data.
+        ctx: a context object for the extractor.
+    """
+
+    address: Address
+    inner: Any
+    ctx: Dict[str, Any] = dataclasses.field(default_factory=dict)
+
+
+@dataclass
+class BBHandle:
+    """reference to a basic block recognized by a feature extractor.
+
+    Attributes:
+        address: the address of the basic block start address.
+        inner: extractor-specific data.
+    """
+
+    address: Address
+    inner: Any
+
+
+@dataclass
+class InsnHandle:
+    """reference to a instruction recognized by a feature extractor.
+
+    Attributes:
+        address: the address of the instruction address.
+        inner: extractor-specific data.
+    """
+
+    address: Address
+    inner: Any
+
+
+class FeatureExtractor:
+    """
+    FeatureExtractor defines the interface for fetching features from a sample.
+
+    There may be multiple backends that support fetching features for capa.
+    For example, we use vivisect by default, but also want to support saving
+     and restoring features from a JSON file.
+    When we restore the features, we'd like to use exactly the same matching logic
+     to find matching rules.
+    Therefore, we can define a FeatureExtractor that provides features from the
+     serialized JSON file and do matching without a binary analysis pass.
+    Also, this provides a way to hook in an IDA backend.
+
+    This class is not instantiated directly; it is the base class for other implementations.
+    """
+
+    __metaclass__ = abc.ABCMeta
+
+    def __init__(self):
+        #
+        # note: a subclass should define ctor parameters for its own use.
+        #  for example, the Vivisect feature extract might require the vw and/or path.
+        # this base class doesn't know what to do with that info, though.
+        #
+        super().__init__()
+
+    @abc.abstractmethod
+    def get_base_address(self) -> Union[AbsoluteVirtualAddress, capa.features.address._NoAddress]:
+        """
+        fetch the preferred load address at which the sample was analyzed.
+
+        when the base address is `NO_ADDRESS`, then the loader has no concept of a preferred load address.
+        such as: shellcode, .NET modules, etc.
+        in these scenarios, RelativeVirtualAddresses aren't used.
+        """
+        raise NotImplementedError()
+
+    @abc.abstractmethod
+    def extract_global_features(self) -> Iterator[Tuple[Feature, Address]]:
+        """
+        extract features found at every scope ("global").
+
+        example::
+
+            extractor = VivisectFeatureExtractor(vw, path)
+            for feature, va in extractor.get_global_features():
+                print('0x%x: %s', va, feature)
+
+        yields:
+          Tuple[Feature, Address]: feature and its location
+        """
+        raise NotImplementedError()
+
+    @abc.abstractmethod
+    def extract_file_features(self) -> Iterator[Tuple[Feature, Address]]:
+        """
+        extract file-scope features.
+
+        example::
+
+            extractor = VivisectFeatureExtractor(vw, path)
+            for feature, va in extractor.get_file_features():
+                print('0x%x: %s', va, feature)
+
+        yields:
+          Tuple[Feature, Address]: feature and its location
+        """
+        raise NotImplementedError()
+
+    @abc.abstractmethod
+    def get_functions(self) -> Iterator[FunctionHandle]:
+        """
+        enumerate the functions and provide opaque values that will
+         subsequently be provided to `.extract_function_features()`, etc.
+        """
+        raise NotImplementedError()
+
+    def is_library_function(self, addr: Address) -> bool:
+        """
+        is the given address a library function?
+        the backend may implement its own function matching algorithm, or none at all.
+        we accept an address here, rather than function object,
+         to handle addresses identified in instructions.
+
+        this information is used to:
+          - filter out matches in library functions (by default), and
+          - recognize when to fetch symbol names for called (non-API) functions
+
+        args:
+          addr (Address): the address of a function.
+
+        returns:
+          bool: True if the given address is the start of a library function.
+        """
+        return False
+
+    def get_function_name(self, addr: Address) -> str:
+        """
+        fetch any recognized name for the given address.
+        this is only guaranteed to return a value when the given function is a recognized library function.
+        we accept a VA here, rather than function object, to handle addresses identified in instructions.
+
+        args:
+          addr (Address): the address of a function.
+
+        returns:
+          str: the function name
+
+        raises:
+          KeyError: when the given function does not have a name.
+        """
+        raise KeyError(addr)
+
+    @abc.abstractmethod
+    def extract_function_features(self, f: FunctionHandle) -> Iterator[Tuple[Feature, Address]]:
+        """
+        extract function-scope features.
+        the arguments are opaque values previously provided by `.get_functions()`, etc.
+
+        example::
+
+            extractor = VivisectFeatureExtractor(vw, path)
+            for function in extractor.get_functions():
+                for feature, address in extractor.extract_function_features(function):
+                    print('0x%x: %s', address, feature)
+
+        args:
+          f [FunctionHandle]: an opaque value previously fetched from `.get_functions()`.
+
+        yields:
+          Tuple[Feature, Address]: feature and its location
+        """
+        raise NotImplementedError()
+
+    @abc.abstractmethod
+    def get_basic_blocks(self, f: FunctionHandle) -> Iterator[BBHandle]:
+        """
+        enumerate the basic blocks in the given function and provide opaque values that will
+         subsequently be provided to `.extract_basic_block_features()`, etc.
+        """
+        raise NotImplementedError()
+
+    @abc.abstractmethod
+    def extract_basic_block_features(self, f: FunctionHandle, bb: BBHandle) -> Iterator[Tuple[Feature, Address]]:
+        """
+        extract basic block-scope features.
+        the arguments are opaque values previously provided by `.get_functions()`, etc.
+
+        example::
+
+            extractor = VivisectFeatureExtractor(vw, path)
+            for function in extractor.get_functions():
+                for bb in extractor.get_basic_blocks(function):
+                    for feature, address in extractor.extract_basic_block_features(function, bb):
+                        print('0x%x: %s', address, feature)
+
+        args:
+          f [FunctionHandle]: an opaque value previously fetched from `.get_functions()`.
+          bb [BBHandle]: an opaque value previously fetched from `.get_basic_blocks()`.
+
+        yields:
+          Tuple[Feature, Address]: feature and its location
+        """
+        raise NotImplementedError()
+
+    @abc.abstractmethod
+    def get_instructions(self, f: FunctionHandle, bb: BBHandle) -> Iterator[InsnHandle]:
+        """
+        enumerate the instructions in the given basic block and provide opaque values that will
+         subsequently be provided to `.extract_insn_features()`, etc.
+        """
+        raise NotImplementedError()
+
+    @abc.abstractmethod
+    def extract_insn_features(
+        self, f: FunctionHandle, bb: BBHandle, insn: InsnHandle
+    ) -> Iterator[Tuple[Feature, Address]]:
+        """
+        extract instruction-scope features.
+        the arguments are opaque values previously provided by `.get_functions()`, etc.
+
+        example::
+
+            extractor = VivisectFeatureExtractor(vw, path)
+            for function in extractor.get_functions():
+                for bb in extractor.get_basic_blocks(function):
+                    for insn in extractor.get_instructions(function, bb):
+                        for feature, address in extractor.extract_insn_features(function, bb, insn):
+                            print('0x%x: %s', address, feature)
+
+        args:
+          f [FunctionHandle]: an opaque value previously fetched from `.get_functions()`.
+          bb [BBHandle]: an opaque value previously fetched from `.get_basic_blocks()`.
+          insn [InsnHandle]: an opaque value previously fetched from `.get_instructions()`.
+
+        yields:
+          Tuple[Feature, Address]: feature and its location
+        """
+        raise NotImplementedError()

capa/features/extractors/binja/basicblock.py

@@ -0,0 +1,184 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+
+import string
+import struct
+from typing import Tuple, Iterator
+
+from binaryninja import Function, Settings
+from binaryninja import BasicBlock as BinjaBasicBlock
+from binaryninja import (
+    BinaryView,
+    SymbolType,
+    RegisterValueType,
+    VariableSourceType,
+    MediumLevelILSetVar,
+    MediumLevelILOperation,
+    MediumLevelILBasicBlock,
+    MediumLevelILInstruction,
+)
+
+from capa.features.common import Feature, Characteristic
+from capa.features.address import Address
+from capa.features.basicblock import BasicBlock
+from capa.features.extractors.helpers import MIN_STACKSTRING_LEN
+from capa.features.extractors.base_extractor import BBHandle, FunctionHandle
+
+use_const_outline: bool = False
+settings: Settings = Settings()
+if settings.contains("analysis.outlining.builtins") and settings.get_bool("analysis.outlining.builtins"):
+    use_const_outline = True
+
+
+def get_printable_len_ascii(s: bytes) -> int:
+    """Return string length if all operand bytes are ascii or utf16-le printable"""
+    count = 0
+    for c in s:
+        if c == 0:
+            return count
+        if c < 127 and chr(c) in string.printable:
+            count += 1
+    return count
+
+
+def get_printable_len_wide(s: bytes) -> int:
+    """Return string length if all operand bytes are ascii or utf16-le printable"""
+    if all(c == 0x00 for c in s[1::2]):
+        return get_printable_len_ascii(s[::2])
+    return 0
+
+
+def get_stack_string_len(f: Function, il: MediumLevelILInstruction) -> int:
+    bv: BinaryView = f.view
+
+    if il.operation != MediumLevelILOperation.MLIL_CALL:
+        return 0
+
+    target = il.dest
+    if target.operation not in [MediumLevelILOperation.MLIL_CONST, MediumLevelILOperation.MLIL_CONST_PTR]:
+        return 0
+
+    addr = target.value.value
+    sym = bv.get_symbol_at(addr)
+    if not sym or sym.type != SymbolType.LibraryFunctionSymbol:
+        return 0
+
+    if sym.name not in ["__builtin_strncpy", "__builtin_strcpy", "__builtin_wcscpy"]:
+        return 0
+
+    if len(il.params) < 2:
+        return 0
+
+    dest = il.params[0]
+    if dest.operation in [MediumLevelILOperation.MLIL_ADDRESS_OF, MediumLevelILOperation.MLIL_VAR]:
+        var = dest.src
+    else:
+        return 0
+
+    if var.source_type != VariableSourceType.StackVariableSourceType:
+        return 0
+
+    src = il.params[1]
+    if src.value.type != RegisterValueType.ConstantDataAggregateValue:
+        return 0
+
+    s = f.get_constant_data(RegisterValueType.ConstantDataAggregateValue, src.value.value)
+    return max(get_printable_len_ascii(bytes(s)), get_printable_len_wide(bytes(s)))
+
+
+def get_printable_len(il: MediumLevelILSetVar) -> int:
+    """Return string length if all operand bytes are ascii or utf16-le printable"""
+    width = il.dest.type.width
+    value = il.src.value.value
+
+    if width == 1:
+        chars = struct.pack("<B", value & 0xFF)
+    elif width == 2:
+        chars = struct.pack("<H", value & 0xFFFF)
+    elif width == 4:
+        chars = struct.pack("<I", value & 0xFFFFFFFF)
+    elif width == 8:
+        chars = struct.pack("<Q", value & 0xFFFFFFFFFFFFFFFF)
+    else:
+        return 0
+
+    def is_printable_ascii(chars_: bytes):
+        return all(c < 127 and chr(c) in string.printable for c in chars_)
+
+    def is_printable_utf16le(chars_: bytes):
+        if all(c == 0x00 for c in chars_[1::2]):
+            return is_printable_ascii(chars_[::2])
+
+    if is_printable_ascii(chars):
+        return width
+
+    if is_printable_utf16le(chars):
+        return width // 2
+
+    return 0
+
+
+def is_mov_imm_to_stack(il: MediumLevelILInstruction) -> bool:
+    """verify instruction moves immediate onto stack"""
+    if il.operation != MediumLevelILOperation.MLIL_SET_VAR:
+        return False
+
+    if il.src.operation != MediumLevelILOperation.MLIL_CONST:
+        return False
+
+    if il.dest.source_type != VariableSourceType.StackVariableSourceType:
+        return False
+
+    return True
+
+
+def bb_contains_stackstring(f: Function, bb: MediumLevelILBasicBlock) -> bool:
+    """check basic block for stackstring indicators
+
+    true if basic block contains enough moves of constant bytes to the stack
+    """
+    count = 0
+    for il in bb:
+        if use_const_outline:
+            count += get_stack_string_len(f, il)
+        else:
+            if is_mov_imm_to_stack(il):
+                count += get_printable_len(il)
+
+    if count > MIN_STACKSTRING_LEN:
+        return True
+    return False
+
+
+def extract_bb_stackstring(fh: FunctionHandle, bbh: BBHandle) -> Iterator[Tuple[Feature, Address]]:
+    """extract stackstring indicators from basic block"""
+    bb: Tuple[BinjaBasicBlock, MediumLevelILBasicBlock] = bbh.inner
+    if bb[1] is not None and bb_contains_stackstring(fh.inner, bb[1]):
+        yield Characteristic("stack string"), bbh.address
+
+
+def extract_bb_tight_loop(fh: FunctionHandle, bbh: BBHandle) -> Iterator[Tuple[Feature, Address]]:
+    """extract tight loop indicators from a basic block"""
+    bb: Tuple[BinjaBasicBlock, MediumLevelILBasicBlock] = bbh.inner
+    for edge in bb[0].outgoing_edges:
+        if edge.target.start == bb[0].start:
+            yield Characteristic("tight loop"), bbh.address
+
+
+def extract_features(fh: FunctionHandle, bbh: BBHandle) -> Iterator[Tuple[Feature, Address]]:
+    """extract basic block features"""
+    for bb_handler in BASIC_BLOCK_HANDLERS:
+        for feature, addr in bb_handler(fh, bbh):
+            yield feature, addr
+    yield BasicBlock(), bbh.address
+
+
+BASIC_BLOCK_HANDLERS = (
+    extract_bb_tight_loop,
+    extract_bb_stackstring,
+)

capa/features/extractors/binja/extractor.py

@@ -0,0 +1,75 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+from typing import List, Tuple, Iterator
+
+import binaryninja as binja
+
+import capa.features.extractors.elf
+import capa.features.extractors.binja.file
+import capa.features.extractors.binja.insn
+import capa.features.extractors.binja.global_
+import capa.features.extractors.binja.function
+import capa.features.extractors.binja.basicblock
+from capa.features.common import Feature
+from capa.features.address import Address, AbsoluteVirtualAddress
+from capa.features.extractors.base_extractor import BBHandle, InsnHandle, FunctionHandle, FeatureExtractor
+
+
+class BinjaFeatureExtractor(FeatureExtractor):
+    def __init__(self, bv: binja.BinaryView):
+        super().__init__()
+        self.bv = bv
+        self.global_features: List[Tuple[Feature, Address]] = []
+        self.global_features.extend(capa.features.extractors.binja.file.extract_file_format(self.bv))
+        self.global_features.extend(capa.features.extractors.binja.global_.extract_os(self.bv))
+        self.global_features.extend(capa.features.extractors.binja.global_.extract_arch(self.bv))
+
+    def get_base_address(self):
+        return AbsoluteVirtualAddress(self.bv.start)
+
+    def extract_global_features(self):
+        yield from self.global_features
+
+    def extract_file_features(self):
+        yield from capa.features.extractors.binja.file.extract_features(self.bv)
+
+    def get_functions(self) -> Iterator[FunctionHandle]:
+        for f in self.bv.functions:
+            yield FunctionHandle(address=AbsoluteVirtualAddress(f.start), inner=f)
+
+    def extract_function_features(self, fh: FunctionHandle) -> Iterator[Tuple[Feature, Address]]:
+        yield from capa.features.extractors.binja.function.extract_features(fh)
+
+    def get_basic_blocks(self, fh: FunctionHandle) -> Iterator[BBHandle]:
+        f: binja.Function = fh.inner
+        # Set up a MLIL basic block dict look up to associate the disassembly basic block with its MLIL basic block
+        mlil_lookup = {}
+        for mlil_bb in f.mlil.basic_blocks:
+            mlil_lookup[mlil_bb.source_block.start] = mlil_bb
+
+        for bb in f.basic_blocks:
+            mlil_bb = mlil_lookup.get(bb.start)
+
+            yield BBHandle(address=AbsoluteVirtualAddress(bb.start), inner=(bb, mlil_bb))
+
+    def extract_basic_block_features(self, fh: FunctionHandle, bbh: BBHandle) -> Iterator[Tuple[Feature, Address]]:
+        yield from capa.features.extractors.binja.basicblock.extract_features(fh, bbh)
+
+    def get_instructions(self, fh: FunctionHandle, bbh: BBHandle) -> Iterator[InsnHandle]:
+        import capa.features.extractors.binja.helpers as binja_helpers
+
+        bb: Tuple[binja.BasicBlock, binja.MediumLevelILBasicBlock] = bbh.inner
+        addr = bb[0].start
+
+        for text, length in bb[0]:
+            insn = binja_helpers.DisassemblyInstruction(addr, length, text)
+            yield InsnHandle(address=AbsoluteVirtualAddress(addr), inner=insn)
+            addr += length
+
+    def extract_insn_features(self, fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle):
+        yield from capa.features.extractors.binja.insn.extract_features(fh, bbh, ih)

capa/features/extractors/binja/file.py

@@ -0,0 +1,187 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+
+import struct
+from typing import Tuple, Iterator
+
+from binaryninja import Segment, BinaryView, SymbolType, SymbolBinding
+
+import capa.features.extractors.common
+import capa.features.extractors.helpers
+import capa.features.extractors.strings
+from capa.features.file import Export, Import, Section, FunctionName
+from capa.features.common import FORMAT_PE, FORMAT_ELF, Format, String, Feature, Characteristic
+from capa.features.address import NO_ADDRESS, Address, FileOffsetAddress, AbsoluteVirtualAddress
+from capa.features.extractors.binja.helpers import read_c_string, unmangle_c_name
+
+
+def check_segment_for_pe(bv: BinaryView, seg: Segment) -> Iterator[Tuple[int, int]]:
+    """check segment for embedded PE
+
+    adapted for binja from:
+    https://github.com/vivisect/vivisect/blob/7be4037b1cecc4551b397f840405a1fc606f9b53/PE/carve.py#L19
+    """
+    mz_xor = [
+        (
+            capa.features.extractors.helpers.xor_static(b"MZ", i),
+            capa.features.extractors.helpers.xor_static(b"PE", i),
+            i,
+        )
+        for i in range(256)
+    ]
+
+    todo = []
+    # If this is the first segment of the binary, skip the first bytes. Otherwise, there will always be a matched
+    # PE at the start of the binaryview.
+    start = seg.start
+    if bv.view_type == "PE" and start == bv.start:
+        start += 1
+
+    for mzx, pex, i in mz_xor:
+        for off, _ in bv.find_all_data(start, seg.end, mzx):
+            todo.append((off, mzx, pex, i))
+
+    while len(todo):
+        off, mzx, pex, i = todo.pop()
+
+        # The MZ header has one field we will check e_lfanew is at 0x3c
+        e_lfanew = off + 0x3C
+
+        if seg.end < (e_lfanew + 4):
+            continue
+
+        newoff = struct.unpack("<I", capa.features.extractors.helpers.xor_static(bv.read(e_lfanew, 4), i))[0]
+
+        peoff = off + newoff
+        if seg.end < (peoff + 2):
+            continue
+
+        if bv.read(peoff, 2) == pex:
+            yield off, i
+
+
+def extract_file_embedded_pe(bv: BinaryView) -> Iterator[Tuple[Feature, Address]]:
+    """extract embedded PE features"""
+    for seg in bv.segments:
+        for ea, _ in check_segment_for_pe(bv, seg):
+            yield Characteristic("embedded pe"), FileOffsetAddress(ea)
+
+
+def extract_file_export_names(bv: BinaryView) -> Iterator[Tuple[Feature, Address]]:
+    """extract function exports"""
+    for sym in bv.get_symbols_of_type(SymbolType.FunctionSymbol):
+        if sym.binding in [SymbolBinding.GlobalBinding, SymbolBinding.WeakBinding]:
+            name = sym.short_name
+            yield Export(name), AbsoluteVirtualAddress(sym.address)
+            unmangled_name = unmangle_c_name(name)
+            if name != unmangled_name:
+                yield Export(unmangled_name), AbsoluteVirtualAddress(sym.address)
+
+    for sym in bv.get_symbols_of_type(SymbolType.DataSymbol):
+        if sym.binding not in [SymbolBinding.GlobalBinding]:
+            continue
+
+        name = sym.short_name
+        if not name.startswith("__forwarder_name"):
+            continue
+
+        # Due to https://github.com/Vector35/binaryninja-api/issues/4641, in binja version 3.5, the symbol's name
+        # does not contain the DLL name. As a workaround, we read the C string at the symbol's address, which contains
+        # both the DLL name and the function name.
+        # Once the above issue is closed in the next binjs stable release, we can update the code here to use the
+        # symbol name directly.
+        name = read_c_string(bv, sym.address, 1024)
+        forwarded_name = capa.features.extractors.helpers.reformat_forwarded_export_name(name)
+        yield Export(forwarded_name), AbsoluteVirtualAddress(sym.address)
+        yield Characteristic("forwarded export"), AbsoluteVirtualAddress(sym.address)
+
+
+def extract_file_import_names(bv: BinaryView) -> Iterator[Tuple[Feature, Address]]:
+    """extract function imports
+
+    1. imports by ordinal:
+     - modulename.#ordinal
+
+    2. imports by name, results in two features to support importname-only
+       matching:
+     - modulename.importname
+     - importname
+    """
+    for sym in bv.get_symbols_of_type(SymbolType.ImportAddressSymbol):
+        lib_name = str(sym.namespace)
+        addr = AbsoluteVirtualAddress(sym.address)
+        for name in capa.features.extractors.helpers.generate_symbols(lib_name, sym.short_name):
+            yield Import(name), addr
+
+        ordinal = sym.ordinal
+        if ordinal != 0 and (lib_name != ""):
+            ordinal_name = f"#{ordinal}"
+            for name in capa.features.extractors.helpers.generate_symbols(lib_name, ordinal_name):
+                yield Import(name), addr
+
+
+def extract_file_section_names(bv: BinaryView) -> Iterator[Tuple[Feature, Address]]:
+    """extract section names"""
+    for name, section in bv.sections.items():
+        yield Section(name), AbsoluteVirtualAddress(section.start)
+
+
+def extract_file_strings(bv: BinaryView) -> Iterator[Tuple[Feature, Address]]:
+    """extract ASCII and UTF-16 LE strings"""
+    for s in bv.strings:
+        yield String(s.value), FileOffsetAddress(s.start)
+
+
+def extract_file_function_names(bv: BinaryView) -> Iterator[Tuple[Feature, Address]]:
+    """
+    extract the names of statically-linked library functions.
+    """
+    for sym_name in bv.symbols:
+        for sym in bv.symbols[sym_name]:
+            if sym.type not in [SymbolType.LibraryFunctionSymbol, SymbolType.FunctionSymbol]:
+                continue
+
+            name = sym.short_name
+            yield FunctionName(name), sym.address
+            if name.startswith("_"):
+                # some linkers may prefix linked routines with a `_` to avoid name collisions.
+                # extract features for both the mangled and un-mangled representations.
+                # e.g. `_fwrite` -> `fwrite`
+                # see: https://stackoverflow.com/a/2628384/87207
+                yield FunctionName(name[1:]), sym.address
+
+
+def extract_file_format(bv: BinaryView) -> Iterator[Tuple[Feature, Address]]:
+    view_type = bv.view_type
+    if view_type in ["PE", "COFF"]:
+        yield Format(FORMAT_PE), NO_ADDRESS
+    elif view_type == "ELF":
+        yield Format(FORMAT_ELF), NO_ADDRESS
+    elif view_type == "Raw":
+        # no file type to return when processing a binary file, but we want to continue processing
+        return
+    else:
+        raise NotImplementedError(f"unexpected file format: {view_type}")
+
+
+def extract_features(bv: BinaryView) -> Iterator[Tuple[Feature, Address]]:
+    """extract file features"""
+    for file_handler in FILE_HANDLERS:
+        for feature, addr in file_handler(bv):
+            yield feature, addr
+
+
+FILE_HANDLERS = (
+    extract_file_export_names,
+    extract_file_import_names,
+    extract_file_strings,
+    extract_file_section_names,
+    extract_file_embedded_pe,
+    extract_file_function_names,
+    extract_file_format,
+)

capa/features/extractors/binja/find_binja_api.py

@@ -0,0 +1,35 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import subprocess
+from pathlib import Path
+
+# When the script gets executed as a standalone executable (via PyInstaller), `import binaryninja` does not work because
+# we have excluded the binaryninja module in `pyinstaller.spec`. The trick here is to call the system Python and try
+# to find out the path of the binaryninja module that has been installed.
+# Note, including the binaryninja module in the `pyintaller.spec` would not work, since the binaryninja module tries to
+# find the binaryninja core e.g., `libbinaryninjacore.dylib`, using a relative path. And this does not work when the
+# binaryninja module is extracted by the PyInstaller.
+code = r"""
+from pathlib import Path
+from importlib import util
+spec = util.find_spec('binaryninja')
+if spec is not None:
+    if len(spec.submodule_search_locations) > 0:
+            path = Path(spec.submodule_search_locations[0])
+            # encode the path with utf8 then convert to hex, make sure it can be read and restored properly
+            print(str(path.parent).encode('utf8').hex())
+"""
+
+
+def find_binja_path() -> Path:
+    raw_output = subprocess.check_output(["python", "-c", code]).decode("ascii").strip()
+    return Path(bytes.fromhex(raw_output).decode("utf8"))
+
+
+if __name__ == "__main__":
+    print(find_binja_path())

capa/features/extractors/binja/function.py

@@ -0,0 +1,104 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+from typing import Tuple, Iterator
+
+from binaryninja import Function, BinaryView, SymbolType, RegisterValueType, LowLevelILOperation
+
+from capa.features.file import FunctionName
+from capa.features.common import Feature, Characteristic
+from capa.features.address import Address, AbsoluteVirtualAddress
+from capa.features.extractors import loops
+from capa.features.extractors.base_extractor import FunctionHandle
+
+
+def extract_function_calls_to(fh: FunctionHandle):
+    """extract callers to a function"""
+    func: Function = fh.inner
+
+    for caller in func.caller_sites:
+        # Everything that is a code reference to the current function is considered a caller, which actually includes
+        # many other references that are NOT a caller. For example, an instruction `push function_start` will also be
+        # considered a caller to the function
+        llil = caller.llil
+        if (llil is None) or llil.operation not in [
+            LowLevelILOperation.LLIL_CALL,
+            LowLevelILOperation.LLIL_CALL_STACK_ADJUST,
+            LowLevelILOperation.LLIL_JUMP,
+            LowLevelILOperation.LLIL_TAILCALL,
+        ]:
+            continue
+
+        if llil.dest.value.type not in [
+            RegisterValueType.ImportedAddressValue,
+            RegisterValueType.ConstantValue,
+            RegisterValueType.ConstantPointerValue,
+        ]:
+            continue
+
+        address = llil.dest.value.value
+        if address != func.start:
+            continue
+
+        yield Characteristic("calls to"), AbsoluteVirtualAddress(caller.address)
+
+
+def extract_function_loop(fh: FunctionHandle):
+    """extract loop indicators from a function"""
+    func: Function = fh.inner
+
+    edges = []
+
+    # construct control flow graph
+    for bb in func.basic_blocks:
+        for edge in bb.outgoing_edges:
+            edges.append((bb.start, edge.target.start))
+
+    if loops.has_loop(edges):
+        yield Characteristic("loop"), fh.address
+
+
+def extract_recursive_call(fh: FunctionHandle):
+    """extract recursive function call"""
+    func: Function = fh.inner
+    bv: BinaryView = func.view
+    if bv is None:
+        return
+
+    for ref in bv.get_code_refs(func.start):
+        if ref.function == func:
+            yield Characteristic("recursive call"), fh.address
+
+
+def extract_function_name(fh: FunctionHandle):
+    """extract function names (e.g., symtab names)"""
+    func: Function = fh.inner
+    bv: BinaryView = func.view
+    if bv is None:
+        return
+
+    for sym in bv.get_symbols(func.start):
+        if sym.type not in [SymbolType.LibraryFunctionSymbol, SymbolType.FunctionSymbol]:
+            continue
+
+        name = sym.short_name
+        yield FunctionName(name), sym.address
+        if name.startswith("_"):
+            # some linkers may prefix linked routines with a `_` to avoid name collisions.
+            # extract features for both the mangled and un-mangled representations.
+            # e.g. `_fwrite` -> `fwrite`
+            # see: https://stackoverflow.com/a/2628384/87207
+            yield FunctionName(name[1:]), sym.address
+
+
+def extract_features(fh: FunctionHandle) -> Iterator[Tuple[Feature, Address]]:
+    for func_handler in FUNCTION_HANDLERS:
+        for feature, addr in func_handler(fh):
+            yield feature, addr
+
+
+FUNCTION_HANDLERS = (extract_function_calls_to, extract_function_loop, extract_recursive_call, extract_function_name)

capa/features/extractors/binja/global_.py

@@ -0,0 +1,60 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import logging
+from typing import Tuple, Iterator
+
+from binaryninja import BinaryView
+
+from capa.features.common import OS, OS_MACOS, ARCH_I386, ARCH_AMD64, OS_WINDOWS, Arch, Feature
+from capa.features.address import NO_ADDRESS, Address
+
+logger = logging.getLogger(__name__)
+
+
+def extract_os(bv: BinaryView) -> Iterator[Tuple[Feature, Address]]:
+    name = bv.platform.name
+    if "-" in name:
+        name = name.split("-")[0]
+
+    if name == "windows":
+        yield OS(OS_WINDOWS), NO_ADDRESS
+
+    elif name == "macos":
+        yield OS(OS_MACOS), NO_ADDRESS
+
+    elif name in ["linux", "freebsd", "decree"]:
+        yield OS(name), NO_ADDRESS
+
+    else:
+        # we likely end up here:
+        #  1. handling shellcode, or
+        #  2. handling a new file format (e.g. macho)
+        #
+        # for (1) we can't do much - its shellcode and all bets are off.
+        # we could maybe accept a further CLI argument to specify the OS,
+        # but i think this would be rarely used.
+        # rules that rely on OS conditions will fail to match on shellcode.
+        #
+        # for (2), this logic will need to be updated as the format is implemented.
+        logger.debug("unsupported file format: %s, will not guess OS", name)
+        return
+
+
+def extract_arch(bv: BinaryView) -> Iterator[Tuple[Feature, Address]]:
+    arch = bv.arch.name
+    if arch == "x86_64":
+        yield Arch(ARCH_AMD64), NO_ADDRESS
+    elif arch == "x86":
+        yield Arch(ARCH_I386), NO_ADDRESS
+    else:
+        # we likely end up here:
+        #  1. handling a new architecture (e.g. aarch64)
+        #
+        # for (1), this logic will need to be updated as the format is implemented.
+        logger.debug("unsupported architecture: %s", arch)
+        return

capa/features/extractors/binja/helpers.py

@@ -0,0 +1,69 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import re
+from typing import List, Callable
+from dataclasses import dataclass
+
+from binaryninja import BinaryView, LowLevelILInstruction
+from binaryninja.architecture import InstructionTextToken
+
+
+@dataclass
+class DisassemblyInstruction:
+    address: int
+    length: int
+    text: List[InstructionTextToken]
+
+
+LLIL_VISITOR = Callable[[LowLevelILInstruction, LowLevelILInstruction, int], bool]
+
+
+def visit_llil_exprs(il: LowLevelILInstruction, func: LLIL_VISITOR):
+    # BN does not really support operand index at the disassembly level, so use the LLIL operand index as a substitute.
+    # Note, this is NOT always guaranteed to be the same as disassembly operand.
+    for i, op in enumerate(il.operands):
+        if isinstance(op, LowLevelILInstruction) and func(op, il, i):
+            visit_llil_exprs(op, func)
+
+
+def unmangle_c_name(name: str) -> str:
+    # https://learn.microsoft.com/en-us/cpp/build/reference/decorated-names?view=msvc-170#FormatC
+    # Possible variations for BaseThreadInitThunk:
+    # @BaseThreadInitThunk@12
+    # _BaseThreadInitThunk
+    # _BaseThreadInitThunk@12
+    # It is also possible for a function to have a `Stub` appended to its name:
+    # _lstrlenWStub@4
+
+    # A small optimization to avoid running the regex too many times
+    # this still increases the unit test execution time from 170s to 200s, should be able to accelerate it
+    #
+    # TODO(xusheng): performance optimizations to improve test execution time
+    # https://github.com/mandiant/capa/issues/1610
+    if name[0] in ["@", "_"]:
+        match = re.match(r"^[@|_](.*?)(Stub)?(@\d+)?$", name)
+        if match:
+            return match.group(1)
+
+    return name
+
+
+def read_c_string(bv: BinaryView, offset: int, max_len: int) -> str:
+    s: List[str] = []
+    while len(s) < max_len:
+        try:
+            c = bv.read(offset + len(s), 1)[0]
+        except Exception:
+            break
+
+        if c == 0:
+            break
+
+        s.append(chr(c))
+
+    return "".join(s)

capa/features/extractors/binja/insn.py

@@ -0,0 +1,586 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+from typing import Any, List, Tuple, Iterator, Optional
+
+from binaryninja import Function
+from binaryninja import BasicBlock as BinjaBasicBlock
+from binaryninja import (
+    BinaryView,
+    ILRegister,
+    SymbolType,
+    BinaryReader,
+    RegisterValueType,
+    LowLevelILOperation,
+    LowLevelILInstruction,
+)
+
+import capa.features.extractors.helpers
+from capa.features.insn import API, MAX_STRUCTURE_SIZE, Number, Offset, Mnemonic, OperandNumber, OperandOffset
+from capa.features.common import MAX_BYTES_FEATURE_SIZE, Bytes, String, Feature, Characteristic
+from capa.features.address import Address, AbsoluteVirtualAddress
+from capa.features.extractors.binja.helpers import DisassemblyInstruction, visit_llil_exprs
+from capa.features.extractors.base_extractor import BBHandle, InsnHandle, FunctionHandle
+
+# security cookie checks may perform non-zeroing XORs, these are expected within a certain
+# byte range within the first and returning basic blocks, this helps to reduce FP features
+SECURITY_COOKIE_BYTES_DELTA = 0x40
+
+
+# check if a function is a stub function to another function/symbol. The criteria is:
+# 1. The function must only have one basic block
+# 2. The function must only make one call/jump to another address
+# If the function being checked is a stub function, returns the target address. Otherwise, return None.
+def is_stub_function(bv: BinaryView, addr: int) -> Optional[int]:
+    funcs = bv.get_functions_at(addr)
+    for func in funcs:
+        if len(func.basic_blocks) != 1:
+            continue
+
+        call_count = 0
+        call_target = None
+        for il in func.llil.instructions:
+            if il.operation in [
+                LowLevelILOperation.LLIL_CALL,
+                LowLevelILOperation.LLIL_CALL_STACK_ADJUST,
+                LowLevelILOperation.LLIL_JUMP,
+                LowLevelILOperation.LLIL_TAILCALL,
+            ]:
+                call_count += 1
+                if il.dest.value.type in [
+                    RegisterValueType.ImportedAddressValue,
+                    RegisterValueType.ConstantValue,
+                    RegisterValueType.ConstantPointerValue,
+                ]:
+                    call_target = il.dest.value.value
+
+        if call_count == 1 and call_target is not None:
+            return call_target
+
+    return None
+
+
+def extract_insn_api_features(fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
+    """
+    parse instruction API features
+
+    example:
+       call dword [0x00473038]
+    """
+    func: Function = fh.inner
+    bv: BinaryView = func.view
+
+    for llil in func.get_llils_at(ih.address):
+        if llil.operation in [
+            LowLevelILOperation.LLIL_CALL,
+            LowLevelILOperation.LLIL_CALL_STACK_ADJUST,
+            LowLevelILOperation.LLIL_JUMP,
+            LowLevelILOperation.LLIL_TAILCALL,
+        ]:
+            if llil.dest.value.type not in [
+                RegisterValueType.ImportedAddressValue,
+                RegisterValueType.ConstantValue,
+                RegisterValueType.ConstantPointerValue,
+            ]:
+                continue
+            address = llil.dest.value.value
+            candidate_addrs = [address]
+            stub_addr = is_stub_function(bv, address)
+            if stub_addr is not None:
+                candidate_addrs.append(stub_addr)
+
+            for address in candidate_addrs:
+                for sym in func.view.get_symbols(address):
+                    if sym is None or sym.type not in [
+                        SymbolType.ImportAddressSymbol,
+                        SymbolType.ImportedFunctionSymbol,
+                        SymbolType.FunctionSymbol,
+                    ]:
+                        continue
+
+                    sym_name = sym.short_name
+
+                    lib_name = ""
+                    import_lib = bv.lookup_imported_object_library(sym.address)
+                    if import_lib is not None:
+                        lib_name = import_lib[0].name
+                        if lib_name.endswith(".dll"):
+                            lib_name = lib_name[:-4]
+                        elif lib_name.endswith(".so"):
+                            lib_name = lib_name[:-3]
+
+                    for name in capa.features.extractors.helpers.generate_symbols(lib_name, sym_name):
+                        yield API(name), ih.address
+
+                    if sym_name.startswith("_"):
+                        for name in capa.features.extractors.helpers.generate_symbols(lib_name, sym_name[1:]):
+                            yield API(name), ih.address
+
+
+def extract_insn_number_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    """
+    parse instruction number features
+    example:
+        push    3136B0h         ; dwControlCode
+    """
+    func: Function = fh.inner
+
+    results: List[Tuple[Any[Number, OperandNumber], Address]] = []
+
+    def llil_checker(il: LowLevelILInstruction, parent: LowLevelILInstruction, index: int) -> bool:
+        if il.operation == LowLevelILOperation.LLIL_LOAD:
+            return False
+
+        if il.operation not in [LowLevelILOperation.LLIL_CONST, LowLevelILOperation.LLIL_CONST_PTR]:
+            return True
+
+        for op in parent.operands:
+            if isinstance(op, ILRegister) and op.name in ["esp", "ebp", "rsp", "rbp", "sp"]:
+                return False
+            elif isinstance(op, LowLevelILInstruction) and op.operation == LowLevelILOperation.LLIL_REG:
+                if op.src.name in ["esp", "ebp", "rsp", "rbp", "sp"]:
+                    return False
+
+        raw_value = il.value.value
+        if parent.operation == LowLevelILOperation.LLIL_SUB:
+            raw_value = -raw_value
+
+        results.append((Number(raw_value), ih.address))
+        results.append((OperandNumber(index, raw_value), ih.address))
+
+        return False
+
+    for llil in func.get_llils_at(ih.address):
+        visit_llil_exprs(llil, llil_checker)
+
+    yield from results
+
+
+def extract_insn_bytes_features(fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
+    """
+    parse referenced byte sequences
+    example:
+        push    offset iid_004118d4_IShellLinkA ; riid
+    """
+    func: Function = fh.inner
+    bv: BinaryView = func.view
+
+    candidate_addrs = set()
+
+    llil = func.get_llil_at(ih.address)
+    if llil is None or llil.operation in [LowLevelILOperation.LLIL_CALL, LowLevelILOperation.LLIL_CALL_STACK_ADJUST]:
+        return
+
+    for ref in bv.get_code_refs_from(ih.address):
+        if ref == ih.address:
+            continue
+
+        if len(bv.get_functions_containing(ref)) > 0:
+            continue
+
+        candidate_addrs.add(ref)
+
+    # collect candidate address by enumerating all integers, https://github.com/Vector35/binaryninja-api/issues/3966
+    def llil_checker(il: LowLevelILInstruction, parent: LowLevelILInstruction, index: int) -> bool:
+        if il.operation in [LowLevelILOperation.LLIL_CONST, LowLevelILOperation.LLIL_CONST_PTR]:
+            value = il.value.value
+            if value > 0:
+                candidate_addrs.add(value)
+            return False
+
+        return True
+
+    for llil in func.get_llils_at(ih.address):
+        visit_llil_exprs(llil, llil_checker)
+
+    for addr in candidate_addrs:
+        extracted_bytes = bv.read(addr, MAX_BYTES_FEATURE_SIZE)
+        if extracted_bytes and not capa.features.extractors.helpers.all_zeros(extracted_bytes):
+            if bv.get_string_at(addr) is None:
+                # don't extract byte features for obvious strings
+                yield Bytes(extracted_bytes), ih.address
+
+
+def extract_insn_string_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    """
+    parse instruction string features
+
+    example:
+        push offset aAcr     ; "ACR  > "
+    """
+    func: Function = fh.inner
+    bv: BinaryView = func.view
+
+    candidate_addrs = set()
+
+    # collect candidate address from code refs directly
+    for ref in bv.get_code_refs_from(ih.address):
+        if ref == ih.address:
+            continue
+
+        if len(bv.get_functions_containing(ref)) > 0:
+            continue
+
+        candidate_addrs.add(ref)
+
+    # collect candidate address by enumerating all integers, https://github.com/Vector35/binaryninja-api/issues/3966
+    def llil_checker(il: LowLevelILInstruction, parent: LowLevelILInstruction, index: int) -> bool:
+        if il.operation in [LowLevelILOperation.LLIL_CONST, LowLevelILOperation.LLIL_CONST_PTR]:
+            value = il.value.value
+            if value > 0:
+                candidate_addrs.add(value)
+            return False
+
+        return True
+
+    for llil in func.get_llils_at(ih.address):
+        visit_llil_exprs(llil, llil_checker)
+
+    # Now we have all the candidate address, check them for string or pointer to string
+    br = BinaryReader(bv)
+    for addr in candidate_addrs:
+        found = bv.get_string_at(addr)
+        if found:
+            yield String(found.value), ih.address
+
+        br.seek(addr)
+        pointer = None
+        if bv.arch.address_size == 4:
+            pointer = br.read32()
+        elif bv.arch.address_size == 8:
+            pointer = br.read64()
+
+        if pointer is not None:
+            found = bv.get_string_at(pointer)
+            if found:
+                yield String(found.value), ih.address
+
+
+def extract_insn_offset_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    """
+    parse instruction structure offset features
+
+    example:
+        .text:0040112F cmp [esi+4], ebx
+    """
+    func: Function = fh.inner
+
+    results: List[Tuple[Any[Offset, OperandOffset], Address]] = []
+    address_size = func.view.arch.address_size * 8
+
+    def llil_checker(il: LowLevelILInstruction, parent: LowLevelILInstruction, index: int) -> bool:
+        #  The most common case, read/write dereference to something like `dword [eax+0x28]`
+        if il.operation in [LowLevelILOperation.LLIL_ADD, LowLevelILOperation.LLIL_SUB]:
+            left = il.left
+            right = il.right
+            # Exclude offsets based on stack/franme pointers
+            if left.operation == LowLevelILOperation.LLIL_REG and left.src.name in ["esp", "ebp", "rsp", "rbp", "sp"]:
+                return True
+
+            if right.operation != LowLevelILOperation.LLIL_CONST:
+                return True
+
+            raw_value = right.value.value
+            # If this is not a dereference, then this must be an add and the offset must be in the range \
+            # [0, MAX_STRUCTURE_SIZE]. For example,
+            # add eax, 0x10,
+            # lea ebx, [eax + 1]
+            if parent.operation not in [LowLevelILOperation.LLIL_LOAD, LowLevelILOperation.LLIL_STORE]:
+                if il.operation != LowLevelILOperation.LLIL_ADD or (not 0 < raw_value < MAX_STRUCTURE_SIZE):
+                    return False
+
+            if address_size > 0:
+                # BN also encodes the constant value as two's complement, we need to restore its original value
+                value = capa.features.extractors.helpers.twos_complement(raw_value, address_size)
+            else:
+                value = raw_value
+
+            results.append((Offset(value), ih.address))
+            results.append((OperandOffset(index, value), ih.address))
+            return False
+
+        # An edge case: for code like `push dword [esi]`, we need to generate a feature for offset 0x0
+        elif il.operation in [LowLevelILOperation.LLIL_LOAD, LowLevelILOperation.LLIL_STORE]:
+            if il.operands[0].operation == LowLevelILOperation.LLIL_REG:
+                results.append((Offset(0), ih.address))
+                results.append((OperandOffset(index, 0), ih.address))
+                return False
+
+        return True
+
+    for llil in func.get_llils_at(ih.address):
+        visit_llil_exprs(llil, llil_checker)
+
+    yield from results
+
+
+def is_nzxor_stack_cookie(f: Function, bb: BinjaBasicBlock, llil: LowLevelILInstruction) -> bool:
+    """check if nzxor exists within stack cookie delta"""
+    # TODO(xusheng): use LLIL SSA to do more accurate analysis
+    # https://github.com/mandiant/capa/issues/1609
+
+    reg_names = []
+    if llil.left.operation == LowLevelILOperation.LLIL_REG:
+        reg_names.append(llil.left.src.name)
+
+    if llil.right.operation == LowLevelILOperation.LLIL_REG:
+        reg_names.append(llil.right.src.name)
+
+    # stack cookie reg should be stack/frame pointer
+    if not any(reg in ["ebp", "esp", "rbp", "rsp", "sp"] for reg in reg_names):
+        return False
+
+    # expect security cookie init in first basic block within first bytes (instructions)
+    if len(bb.incoming_edges) == 0 and llil.address < (bb.start + SECURITY_COOKIE_BYTES_DELTA):
+        return True
+
+    # ... or within last bytes (instructions) before a return
+    if len(bb.outgoing_edges) == 0 and llil.address > (bb.end - SECURITY_COOKIE_BYTES_DELTA):
+        return True
+
+    return False
+
+
+def extract_insn_nzxor_characteristic_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    """
+    parse instruction non-zeroing XOR instruction
+    ignore expected non-zeroing XORs, e.g. security cookies
+    """
+    func: Function = fh.inner
+
+    results = []
+
+    def llil_checker(il: LowLevelILInstruction, parent: LowLevelILInstruction, index: int) -> bool:
+        # If the two operands of the xor instruction are the same, the LLIL will be translated to other instructions,
+        # e.g., <llil: eax = 0>, (LLIL_SET_REG). So we do not need to check whether the two operands are the same.
+        if il.operation == LowLevelILOperation.LLIL_XOR:
+            # Exclude cases related to the stack cookie
+            if is_nzxor_stack_cookie(fh.inner, bbh.inner[0], il):
+                return False
+            results.append((Characteristic("nzxor"), ih.address))
+            return False
+        else:
+            return True
+
+    for llil in func.get_llils_at(ih.address):
+        visit_llil_exprs(llil, llil_checker)
+
+    yield from results
+
+
+def extract_insn_mnemonic_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    """parse instruction mnemonic features"""
+    insn: DisassemblyInstruction = ih.inner
+    yield Mnemonic(insn.text[0].text), ih.address
+
+
+def extract_insn_obfs_call_plus_5_characteristic_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    """
+    parse call $+5 instruction from the given instruction.
+    """
+    insn: DisassemblyInstruction = ih.inner
+    if insn.text[0].text == "call" and insn.text[2].text == "$+5" and insn.length == 5:
+        yield Characteristic("call $+5"), ih.address
+
+
+def extract_insn_peb_access_characteristic_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    """parse instruction peb access
+
+    fs:[0x30] on x86, gs:[0x60] on x64
+    """
+    func: Function = fh.inner
+
+    results = []
+
+    def llil_checker(il: LowLevelILInstruction, parent: LowLevelILOperation, index: int) -> bool:
+        if il.operation != LowLevelILOperation.LLIL_LOAD:
+            return True
+
+        src = il.src
+        if src.operation != LowLevelILOperation.LLIL_ADD:
+            return True
+
+        left = src.left
+        right = src.right
+
+        if left.operation != LowLevelILOperation.LLIL_REG:
+            return True
+
+        reg = left.src.name
+
+        if right.operation != LowLevelILOperation.LLIL_CONST:
+            return True
+
+        value = right.value.value
+        if (reg, value) not in (("fsbase", 0x30), ("gsbase", 0x60)):
+            return True
+
+        results.append((Characteristic("peb access"), ih.address))
+        return False
+
+    for llil in func.get_llils_at(ih.address):
+        visit_llil_exprs(llil, llil_checker)
+
+    yield from results
+
+
+def extract_insn_segment_access_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    """parse instruction fs or gs access"""
+    func: Function = fh.inner
+
+    results = []
+
+    def llil_checker(il: LowLevelILInstruction, parent: LowLevelILInstruction, index: int) -> bool:
+        if il.operation == LowLevelILOperation.LLIL_REG:
+            reg = il.src.name
+            if reg == "fsbase":
+                results.append((Characteristic("fs access"), ih.address))
+                return False
+            elif reg == "gsbase":
+                results.append((Characteristic("gs access"), ih.address))
+                return False
+            return False
+
+        return True
+
+    for llil in func.get_llils_at(ih.address):
+        visit_llil_exprs(llil, llil_checker)
+
+    yield from results
+
+
+def extract_insn_cross_section_cflow(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    """inspect the instruction for a CALL or JMP that crosses section boundaries"""
+    func: Function = fh.inner
+    bv: BinaryView = func.view
+
+    if bv is None:
+        return
+
+    seg1 = bv.get_segment_at(ih.address)
+    sections1 = bv.get_sections_at(ih.address)
+    for ref in bv.get_code_refs_from(ih.address):
+        if len(bv.get_functions_at(ref)) == 0:
+            continue
+
+        seg2 = bv.get_segment_at(ref)
+        sections2 = bv.get_sections_at(ref)
+        if seg1 != seg2 or sections1 != sections2:
+            yield Characteristic("cross section flow"), ih.address
+
+
+def extract_function_calls_from(fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
+    """extract functions calls from features
+
+    most relevant at the function scope, however, its most efficient to extract at the instruction scope
+    """
+    func: Function = fh.inner
+    bv: BinaryView = func.view
+
+    if bv is None:
+        return
+
+    for il in func.get_llils_at(ih.address):
+        if il.operation not in [
+            LowLevelILOperation.LLIL_CALL,
+            LowLevelILOperation.LLIL_CALL_STACK_ADJUST,
+            LowLevelILOperation.LLIL_TAILCALL,
+        ]:
+            continue
+
+        dest = il.dest
+        if dest.operation == LowLevelILOperation.LLIL_CONST_PTR:
+            value = dest.value.value
+            yield Characteristic("calls from"), AbsoluteVirtualAddress(value)
+        elif dest.operation == LowLevelILOperation.LLIL_CONST:
+            yield Characteristic("calls from"), AbsoluteVirtualAddress(dest.value)
+        elif dest.operation == LowLevelILOperation.LLIL_LOAD:
+            indirect_src = dest.src
+            if indirect_src.operation == LowLevelILOperation.LLIL_CONST_PTR:
+                value = indirect_src.value.value
+                yield Characteristic("calls from"), AbsoluteVirtualAddress(value)
+            elif indirect_src.operation == LowLevelILOperation.LLIL_CONST:
+                yield Characteristic("calls from"), AbsoluteVirtualAddress(indirect_src.value)
+        elif dest.operation == LowLevelILOperation.LLIL_REG:
+            if dest.value.type in [
+                RegisterValueType.ImportedAddressValue,
+                RegisterValueType.ConstantValue,
+                RegisterValueType.ConstantPointerValue,
+            ]:
+                yield Characteristic("calls from"), AbsoluteVirtualAddress(dest.value.value)
+
+
+def extract_function_indirect_call_characteristic_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    """extract indirect function calls (e.g., call eax or call dword ptr [edx+4])
+    does not include calls like => call ds:dword_ABD4974
+
+    most relevant at the function or basic block scope;
+    however, its most efficient to extract at the instruction scope
+    """
+    func: Function = fh.inner
+
+    llil = func.get_llil_at(ih.address)
+    if llil is None or llil.operation not in [
+        LowLevelILOperation.LLIL_CALL,
+        LowLevelILOperation.LLIL_CALL_STACK_ADJUST,
+        LowLevelILOperation.LLIL_TAILCALL,
+    ]:
+        return
+
+    if llil.dest.operation in [LowLevelILOperation.LLIL_CONST, LowLevelILOperation.LLIL_CONST_PTR]:
+        return
+
+    if llil.dest.operation == LowLevelILOperation.LLIL_LOAD:
+        src = llil.dest.src
+        if src.operation in [LowLevelILOperation.LLIL_CONST, LowLevelILOperation.LLIL_CONST_PTR]:
+            return
+
+    yield Characteristic("indirect call"), ih.address
+
+
+def extract_features(f: FunctionHandle, bbh: BBHandle, insn: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
+    """extract instruction features"""
+    for inst_handler in INSTRUCTION_HANDLERS:
+        for feature, ea in inst_handler(f, bbh, insn):
+            yield feature, ea
+
+
+INSTRUCTION_HANDLERS = (
+    extract_insn_api_features,
+    extract_insn_number_features,
+    extract_insn_bytes_features,
+    extract_insn_string_features,
+    extract_insn_offset_features,
+    extract_insn_nzxor_characteristic_features,
+    extract_insn_mnemonic_features,
+    extract_insn_obfs_call_plus_5_characteristic_features,
+    extract_insn_peb_access_characteristic_features,
+    extract_insn_cross_section_cflow,
+    extract_insn_segment_access_features,
+    extract_function_calls_from,
+    extract_function_indirect_call_characteristic_features,
+)

capa/features/extractors/common.py

@@ -0,0 +1,135 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import io
+import logging
+import binascii
+import contextlib
+from typing import Tuple, Iterator
+
+import pefile
+
+import capa.features
+import capa.features.extractors.elf
+import capa.features.extractors.pefile
+import capa.features.extractors.strings
+from capa.features.common import (
+    OS,
+    OS_ANY,
+    OS_AUTO,
+    ARCH_ANY,
+    FORMAT_PE,
+    FORMAT_ELF,
+    OS_WINDOWS,
+    FORMAT_FREEZE,
+    FORMAT_RESULT,
+    Arch,
+    Format,
+    String,
+    Feature,
+)
+from capa.features.freeze import is_freeze
+from capa.features.address import NO_ADDRESS, Address, FileOffsetAddress
+
+logger = logging.getLogger(__name__)
+
+# match strings for formats
+MATCH_PE = b"MZ"
+MATCH_ELF = b"\x7fELF"
+MATCH_RESULT = b'{"meta":'
+
+
+def extract_file_strings(buf, **kwargs) -> Iterator[Tuple[String, Address]]:
+    """
+    extract ASCII and UTF-16 LE strings from file
+    """
+    for s in capa.features.extractors.strings.extract_ascii_strings(buf):
+        yield String(s.s), FileOffsetAddress(s.offset)
+
+    for s in capa.features.extractors.strings.extract_unicode_strings(buf):
+        yield String(s.s), FileOffsetAddress(s.offset)
+
+
+def extract_format(buf) -> Iterator[Tuple[Feature, Address]]:
+    if buf.startswith(MATCH_PE):
+        yield Format(FORMAT_PE), NO_ADDRESS
+    elif buf.startswith(MATCH_ELF):
+        yield Format(FORMAT_ELF), NO_ADDRESS
+    elif is_freeze(buf):
+        yield Format(FORMAT_FREEZE), NO_ADDRESS
+    elif buf.startswith(MATCH_RESULT):
+        yield Format(FORMAT_RESULT), NO_ADDRESS
+    else:
+        # we likely end up here:
+        #  1. handling a file format (e.g. macho)
+        #
+        # for (1), this logic will need to be updated as the format is implemented.
+        logger.debug("unsupported file format: %s", binascii.hexlify(buf[:4]).decode("ascii"))
+        return
+
+
+def extract_arch(buf) -> Iterator[Tuple[Feature, Address]]:
+    if buf.startswith(MATCH_PE):
+        yield from capa.features.extractors.pefile.extract_file_arch(pe=pefile.PE(data=buf))
+
+    elif buf.startswith(MATCH_RESULT):
+        yield Arch(ARCH_ANY), NO_ADDRESS
+
+    elif buf.startswith(MATCH_ELF):
+        with contextlib.closing(io.BytesIO(buf)) as f:
+            arch = capa.features.extractors.elf.detect_elf_arch(f)
+
+        if arch not in capa.features.common.VALID_ARCH:
+            logger.debug("unsupported arch: %s", arch)
+            return
+
+        yield Arch(arch), NO_ADDRESS
+
+    else:
+        # we likely end up here:
+        #  1. handling shellcode, or
+        #  2. handling a new file format (e.g. macho)
+        #
+        # for (1) we can't do much - its shellcode and all bets are off.
+        # we could maybe accept a further CLI argument to specify the arch,
+        # but i think this would be rarely used.
+        # rules that rely on arch conditions will fail to match on shellcode.
+        #
+        # for (2), this logic will need to be updated as the format is implemented.
+        logger.debug("unsupported file format: %s, will not guess Arch", binascii.hexlify(buf[:4]).decode("ascii"))
+        return
+
+
+def extract_os(buf, os=OS_AUTO) -> Iterator[Tuple[Feature, Address]]:
+    if os != OS_AUTO:
+        yield OS(os), NO_ADDRESS
+
+    if buf.startswith(MATCH_PE):
+        yield OS(OS_WINDOWS), NO_ADDRESS
+    elif buf.startswith(MATCH_RESULT):
+        yield OS(OS_ANY), NO_ADDRESS
+    elif buf.startswith(MATCH_ELF):
+        with contextlib.closing(io.BytesIO(buf)) as f:
+            os = capa.features.extractors.elf.detect_elf_os(f)
+
+        if os not in capa.features.common.VALID_OS:
+            logger.debug("unsupported os: %s", os)
+            return
+
+        yield OS(os), NO_ADDRESS
+
+    else:
+        # we likely end up here:
+        #  1. handling shellcode, or
+        #  2. handling a new file format (e.g. macho)
+        #
+        # for (1) we can't do much - its shellcode and all bets are off.
+        # rules that rely on OS conditions will fail to match on shellcode.
+        #
+        # for (2), this logic will need to be updated as the format is implemented.
+        logger.debug("unsupported file format: %s, will not guess OS", binascii.hexlify(buf[:4]).decode("ascii"))
+        return

capa/features/extractors/dnfile/extractor.py

@@ -0,0 +1,155 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+
+from __future__ import annotations
+
+from typing import Dict, List, Tuple, Union, Iterator, Optional
+from pathlib import Path
+
+import dnfile
+from dncil.cil.opcode import OpCodes
+
+import capa.features.extractors
+import capa.features.extractors.dotnetfile
+import capa.features.extractors.dnfile.file
+import capa.features.extractors.dnfile.insn
+import capa.features.extractors.dnfile.function
+from capa.features.common import Feature
+from capa.features.address import NO_ADDRESS, Address, DNTokenAddress, DNTokenOffsetAddress
+from capa.features.extractors.dnfile.types import DnType, DnUnmanagedMethod
+from capa.features.extractors.base_extractor import BBHandle, InsnHandle, FunctionHandle, FeatureExtractor
+from capa.features.extractors.dnfile.helpers import (
+    get_dotnet_types,
+    get_dotnet_fields,
+    get_dotnet_managed_imports,
+    get_dotnet_managed_methods,
+    get_dotnet_unmanaged_imports,
+    get_dotnet_managed_method_bodies,
+)
+
+
+class DnFileFeatureExtractorCache:
+    def __init__(self, pe: dnfile.dnPE):
+        self.imports: Dict[int, Union[DnType, DnUnmanagedMethod]] = {}
+        self.native_imports: Dict[int, Union[DnType, DnUnmanagedMethod]] = {}
+        self.methods: Dict[int, Union[DnType, DnUnmanagedMethod]] = {}
+        self.fields: Dict[int, Union[DnType, DnUnmanagedMethod]] = {}
+        self.types: Dict[int, Union[DnType, DnUnmanagedMethod]] = {}
+
+        for import_ in get_dotnet_managed_imports(pe):
+            self.imports[import_.token] = import_
+        for native_import in get_dotnet_unmanaged_imports(pe):
+            self.native_imports[native_import.token] = native_import
+        for method in get_dotnet_managed_methods(pe):
+            self.methods[method.token] = method
+        for field in get_dotnet_fields(pe):
+            self.fields[field.token] = field
+        for type_ in get_dotnet_types(pe):
+            self.types[type_.token] = type_
+
+    def get_import(self, token: int) -> Optional[Union[DnType, DnUnmanagedMethod]]:
+        return self.imports.get(token)
+
+    def get_native_import(self, token: int) -> Optional[Union[DnType, DnUnmanagedMethod]]:
+        return self.native_imports.get(token)
+
+    def get_method(self, token: int) -> Optional[Union[DnType, DnUnmanagedMethod]]:
+        return self.methods.get(token)
+
+    def get_field(self, token: int) -> Optional[Union[DnType, DnUnmanagedMethod]]:
+        return self.fields.get(token)
+
+    def get_type(self, token: int) -> Optional[Union[DnType, DnUnmanagedMethod]]:
+        return self.types.get(token)
+
+
+class DnfileFeatureExtractor(FeatureExtractor):
+    def __init__(self, path: Path):
+        super().__init__()
+        self.pe: dnfile.dnPE = dnfile.dnPE(str(path))
+
+        # pre-compute .NET token lookup tables; each .NET method has access to this cache for feature extraction
+        # most relevant at instruction scope
+        self.token_cache: DnFileFeatureExtractorCache = DnFileFeatureExtractorCache(self.pe)
+
+        # pre-compute these because we'll yield them at *every* scope.
+        self.global_features: List[Tuple[Feature, Address]] = []
+        self.global_features.extend(capa.features.extractors.dotnetfile.extract_file_format())
+        self.global_features.extend(capa.features.extractors.dotnetfile.extract_file_os(pe=self.pe))
+        self.global_features.extend(capa.features.extractors.dotnetfile.extract_file_arch(pe=self.pe))
+
+    def get_base_address(self):
+        return NO_ADDRESS
+
+    def extract_global_features(self):
+        yield from self.global_features
+
+    def extract_file_features(self):
+        yield from capa.features.extractors.dnfile.file.extract_features(self.pe)
+
+    def get_functions(self) -> Iterator[FunctionHandle]:
+        # create a method lookup table
+        methods: Dict[Address, FunctionHandle] = {}
+        for token, method in get_dotnet_managed_method_bodies(self.pe):
+            fh: FunctionHandle = FunctionHandle(
+                address=DNTokenAddress(token),
+                inner=method,
+                ctx={"pe": self.pe, "calls_from": set(), "calls_to": set(), "cache": self.token_cache},
+            )
+
+            # method tokens should be unique
+            assert fh.address not in methods.keys()
+            methods[fh.address] = fh
+
+        # calculate unique calls to/from each method
+        for fh in methods.values():
+            for insn in fh.inner.instructions:
+                if insn.opcode not in (
+                    OpCodes.Call,
+                    OpCodes.Callvirt,
+                    OpCodes.Jmp,
+                    OpCodes.Newobj,
+                ):
+                    continue
+
+                address: DNTokenAddress = DNTokenAddress(insn.operand.value)
+
+                # record call to destination method; note: we only consider MethodDef methods for destinations
+                dest: Optional[FunctionHandle] = methods.get(address)
+                if dest is not None:
+                    dest.ctx["calls_to"].add(fh.address)
+
+                # record call from source method; note: we record all unique calls from a MethodDef method, not just
+                # those calls to other MethodDef methods e.g. calls to imported MemberRef methods
+                fh.ctx["calls_from"].add(address)
+
+        yield from methods.values()
+
+    def extract_function_features(self, fh) -> Iterator[Tuple[Feature, Address]]:
+        yield from capa.features.extractors.dnfile.function.extract_features(fh)
+
+    def get_basic_blocks(self, f) -> Iterator[BBHandle]:
+        # each dotnet method is considered 1 basic block
+        yield BBHandle(
+            address=f.address,
+            inner=f.inner,
+        )
+
+    def extract_basic_block_features(self, fh, bbh):
+        # we don't support basic block features
+        yield from []
+
+    def get_instructions(self, fh, bbh):
+        for insn in bbh.inner.instructions:
+            yield InsnHandle(
+                address=DNTokenOffsetAddress(bbh.address, insn.offset - (fh.inner.offset + fh.inner.header_size)),
+                inner=insn,
+            )
+
+    def extract_insn_features(self, fh, bbh, ih) -> Iterator[Tuple[Feature, Address]]:
+        yield from capa.features.extractors.dnfile.insn.extract_features(fh, bbh, ih)

capa/features/extractors/dnfile/file.py

@@ -0,0 +1,63 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+
+from __future__ import annotations
+
+from typing import Tuple, Iterator
+
+import dnfile
+
+import capa.features.extractors.dotnetfile
+from capa.features.file import Import, FunctionName
+from capa.features.common import Class, Format, String, Feature, Namespace, Characteristic
+from capa.features.address import Address
+
+
+def extract_file_import_names(pe: dnfile.dnPE) -> Iterator[Tuple[Import, Address]]:
+    yield from capa.features.extractors.dotnetfile.extract_file_import_names(pe=pe)
+
+
+def extract_file_format(pe: dnfile.dnPE) -> Iterator[Tuple[Format, Address]]:
+    yield from capa.features.extractors.dotnetfile.extract_file_format(pe=pe)
+
+
+def extract_file_function_names(pe: dnfile.dnPE) -> Iterator[Tuple[FunctionName, Address]]:
+    yield from capa.features.extractors.dotnetfile.extract_file_function_names(pe=pe)
+
+
+def extract_file_strings(pe: dnfile.dnPE) -> Iterator[Tuple[String, Address]]:
+    yield from capa.features.extractors.dotnetfile.extract_file_strings(pe=pe)
+
+
+def extract_file_mixed_mode_characteristic_features(pe: dnfile.dnPE) -> Iterator[Tuple[Characteristic, Address]]:
+    yield from capa.features.extractors.dotnetfile.extract_file_mixed_mode_characteristic_features(pe=pe)
+
+
+def extract_file_namespace_features(pe: dnfile.dnPE) -> Iterator[Tuple[Namespace, Address]]:
+    yield from capa.features.extractors.dotnetfile.extract_file_namespace_features(pe=pe)
+
+
+def extract_file_class_features(pe: dnfile.dnPE) -> Iterator[Tuple[Class, Address]]:
+    yield from capa.features.extractors.dotnetfile.extract_file_class_features(pe=pe)
+
+
+def extract_features(pe: dnfile.dnPE) -> Iterator[Tuple[Feature, Address]]:
+    for file_handler in FILE_HANDLERS:
+        for feature, address in file_handler(pe):
+            yield feature, address
+
+
+FILE_HANDLERS = (
+    extract_file_import_names,
+    extract_file_function_names,
+    extract_file_strings,
+    extract_file_format,
+    extract_file_mixed_mode_characteristic_features,
+    extract_file_namespace_features,
+    extract_file_class_features,
+)

capa/features/extractors/dnfile/function.py

@@ -0,0 +1,50 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+
+from __future__ import annotations
+
+import logging
+from typing import Tuple, Iterator
+
+from capa.features.common import Feature, Characteristic
+from capa.features.address import Address
+from capa.features.extractors.base_extractor import FunctionHandle
+
+logger = logging.getLogger(__name__)
+
+
+def extract_function_calls_to(fh: FunctionHandle) -> Iterator[Tuple[Characteristic, Address]]:
+    """extract callers to a function"""
+    for dest in fh.ctx["calls_to"]:
+        yield Characteristic("calls to"), dest
+
+
+def extract_function_calls_from(fh: FunctionHandle) -> Iterator[Tuple[Characteristic, Address]]:
+    """extract callers from a function"""
+    for src in fh.ctx["calls_from"]:
+        yield Characteristic("calls from"), src
+
+
+def extract_recursive_call(fh: FunctionHandle) -> Iterator[Tuple[Characteristic, Address]]:
+    """extract recursive function call"""
+    if fh.address in fh.ctx["calls_to"]:
+        yield Characteristic("recursive call"), fh.address
+
+
+def extract_function_loop(fh: FunctionHandle) -> Iterator[Tuple[Characteristic, Address]]:
+    """extract loop indicators from a function"""
+    raise NotImplementedError()
+
+
+def extract_features(fh: FunctionHandle) -> Iterator[Tuple[Feature, Address]]:
+    for func_handler in FUNCTION_HANDLERS:
+        for feature, addr in func_handler(fh):
+            yield feature, addr
+
+
+FUNCTION_HANDLERS = (extract_function_calls_to, extract_function_calls_from, extract_recursive_call)

capa/features/extractors/dnfile/helpers.py

@@ -0,0 +1,335 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+
+from __future__ import annotations
+
+import logging
+from typing import Dict, Tuple, Union, Iterator, Optional
+
+import dnfile
+from dncil.cil.body import CilMethodBody
+from dncil.cil.error import MethodBodyFormatError
+from dncil.clr.token import Token, StringToken, InvalidToken
+from dncil.cil.body.reader import CilMethodBodyReaderBase
+
+from capa.features.common import FeatureAccess
+from capa.features.extractors.dnfile.types import DnType, DnUnmanagedMethod
+
+logger = logging.getLogger(__name__)
+
+
+class DnfileMethodBodyReader(CilMethodBodyReaderBase):
+    def __init__(self, pe: dnfile.dnPE, row: dnfile.mdtable.MethodDefRow):
+        self.pe: dnfile.dnPE = pe
+        self.offset: int = self.pe.get_offset_from_rva(row.Rva)
+
+    def read(self, n: int) -> bytes:
+        data: bytes = self.pe.get_data(self.pe.get_rva_from_offset(self.offset), n)
+        self.offset += n
+        return data
+
+    def tell(self) -> int:
+        return self.offset
+
+    def seek(self, offset: int) -> int:
+        self.offset = offset
+        return self.offset
+
+
+def resolve_dotnet_token(pe: dnfile.dnPE, token: Token) -> Union[dnfile.base.MDTableRow, InvalidToken, str]:
+    """map generic token to string or table row"""
+    assert pe.net is not None
+    assert pe.net.mdtables is not None
+
+    if isinstance(token, StringToken):
+        user_string: Optional[str] = read_dotnet_user_string(pe, token)
+        if user_string is None:
+            return InvalidToken(token.value)
+        return user_string
+
+    table: Optional[dnfile.base.ClrMetaDataTable] = pe.net.mdtables.tables.get(token.table)
+    if table is None:
+        # table index is not valid
+        return InvalidToken(token.value)
+
+    try:
+        return table.rows[token.rid - 1]
+    except IndexError:
+        # table index is valid but row index is not valid
+        return InvalidToken(token.value)
+
+
+def read_dotnet_method_body(pe: dnfile.dnPE, row: dnfile.mdtable.MethodDefRow) -> Optional[CilMethodBody]:
+    """read dotnet method body"""
+    try:
+        return CilMethodBody(DnfileMethodBodyReader(pe, row))
+    except MethodBodyFormatError as e:
+        logger.debug("failed to parse managed method body @ 0x%08x (%s)", row.Rva, e)
+        return None
+
+
+def read_dotnet_user_string(pe: dnfile.dnPE, token: StringToken) -> Optional[str]:
+    """read user string from #US stream"""
+    assert pe.net is not None
+
+    if pe.net.user_strings is None:
+        # stream may not exist (seen in obfuscated .NET)
+        logger.debug("#US stream does not exist for stream index 0x%08x", token.rid)
+        return None
+
+    try:
+        user_string: Optional[dnfile.stream.UserString] = pe.net.user_strings.get_us(token.rid)
+    except UnicodeDecodeError as e:
+        logger.debug("failed to decode #US stream index 0x%08x (%s)", token.rid, e)
+        return None
+
+    if user_string is None:
+        return None
+
+    return user_string.value
+
+
+def get_dotnet_managed_imports(pe: dnfile.dnPE) -> Iterator[DnType]:
+    """get managed imports from MemberRef table
+
+    see https://www.ntcore.com/files/dotnetformat.htm
+
+    10 - MemberRef Table
+        Each row represents an imported method
+            Class (index into the TypeRef, ModuleRef, MethodDef, TypeSpec or TypeDef tables)
+            Name (index into String heap)
+    01 - TypeRef Table
+        Each row represents an imported class, its namespace and the assembly which contains it
+            TypeName (index into String heap)
+            TypeNamespace (index into String heap)
+    """
+    for rid, member_ref in iter_dotnet_table(pe, dnfile.mdtable.MemberRef.number):
+        assert isinstance(member_ref, dnfile.mdtable.MemberRefRow)
+
+        if not isinstance(member_ref.Class.row, dnfile.mdtable.TypeRefRow):
+            # only process class imports from TypeRef table
+            continue
+
+        token: int = calculate_dotnet_token_value(dnfile.mdtable.MemberRef.number, rid)
+        access: Optional[str]
+
+        # assume .NET imports starting with get_/set_ are used to access a property
+        if member_ref.Name.startswith("get_"):
+            access = FeatureAccess.READ
+        elif member_ref.Name.startswith("set_"):
+            access = FeatureAccess.WRITE
+        else:
+            access = None
+
+        member_ref_name: str = member_ref.Name
+        if member_ref_name.startswith(("get_", "set_")):
+            # remove get_/set_ from MemberRef name
+            member_ref_name = member_ref_name[4:]
+
+        yield DnType(
+            token,
+            member_ref.Class.row.TypeName,
+            namespace=member_ref.Class.row.TypeNamespace,
+            member=member_ref_name,
+            access=access,
+        )
+
+
+def get_dotnet_methoddef_property_accessors(pe: dnfile.dnPE) -> Iterator[Tuple[int, str]]:
+    """get MethodDef methods used to access properties
+
+    see https://www.ntcore.com/files/dotnetformat.htm
+
+    24 - MethodSemantics Table
+        Links Events and Properties to specific methods. For example one Event can be associated to more methods. A property uses this table to associate get/set methods.
+            Semantics (a 2-byte bitmask of type MethodSemanticsAttributes)
+            Method (index into the MethodDef table)
+            Association (index into the Event or Property table; more precisely, a HasSemantics coded index)
+    """
+    for rid, method_semantics in iter_dotnet_table(pe, dnfile.mdtable.MethodSemantics.number):
+        assert isinstance(method_semantics, dnfile.mdtable.MethodSemanticsRow)
+
+        if method_semantics.Association.row is None:
+            logger.debug("MethodSemantics[0x%X] Association row is None", rid)
+            continue
+
+        if isinstance(method_semantics.Association.row, dnfile.mdtable.EventRow):
+            # ignore events
+            logger.debug("MethodSemantics[0x%X] ignoring Event", rid)
+            continue
+
+        if method_semantics.Method.table is None:
+            logger.debug("MethodSemantics[0x%X] Method table is None", rid)
+            continue
+
+        token: int = calculate_dotnet_token_value(
+            method_semantics.Method.table.number, method_semantics.Method.row_index
+        )
+
+        if method_semantics.Semantics.msSetter:
+            yield token, FeatureAccess.WRITE
+        elif method_semantics.Semantics.msGetter:
+            yield token, FeatureAccess.READ
+
+
+def get_dotnet_managed_methods(pe: dnfile.dnPE) -> Iterator[DnType]:
+    """get managed method names from TypeDef table
+
+    see https://www.ntcore.com/files/dotnetformat.htm
+
+    02 - TypeDef Table
+        Each row represents a class in the current assembly.
+            TypeName (index into String heap)
+            TypeNamespace (index into String heap)
+            MethodList (index into MethodDef table; it marks the first of a contiguous run of Methods owned by this Type)
+    """
+    accessor_map: Dict[int, str] = {}
+    for methoddef, methoddef_access in get_dotnet_methoddef_property_accessors(pe):
+        accessor_map[methoddef] = methoddef_access
+
+    for rid, typedef in iter_dotnet_table(pe, dnfile.mdtable.TypeDef.number):
+        assert isinstance(typedef, dnfile.mdtable.TypeDefRow)
+
+        for idx, method in enumerate(typedef.MethodList):
+            if method.table is None:
+                logger.debug("TypeDef[0x%X] MethodList[0x%X] table is None", rid, idx)
+                continue
+            if method.row is None:
+                logger.debug("TypeDef[0x%X] MethodList[0x%X] row is None", rid, idx)
+                continue
+
+            token: int = calculate_dotnet_token_value(method.table.number, method.row_index)
+            access: Optional[str] = accessor_map.get(token)
+
+            method_name: str = method.row.Name
+            if method_name.startswith(("get_", "set_")):
+                # remove get_/set_
+                method_name = method_name[4:]
+
+            yield DnType(token, typedef.TypeName, namespace=typedef.TypeNamespace, member=method_name, access=access)
+
+
+def get_dotnet_fields(pe: dnfile.dnPE) -> Iterator[DnType]:
+    """get fields from TypeDef table
+
+    see https://www.ntcore.com/files/dotnetformat.htm
+
+    02 - TypeDef Table
+        Each row represents a class in the current assembly.
+            TypeName (index into String heap)
+            TypeNamespace (index into String heap)
+            FieldList (index into Field table; it marks the first of a contiguous run of Fields owned by this Type)
+    """
+    for rid, typedef in iter_dotnet_table(pe, dnfile.mdtable.TypeDef.number):
+        assert isinstance(typedef, dnfile.mdtable.TypeDefRow)
+
+        for idx, field in enumerate(typedef.FieldList):
+            if field.table is None:
+                logger.debug("TypeDef[0x%X] FieldList[0x%X] table is None", rid, idx)
+                continue
+            if field.row is None:
+                logger.debug("TypeDef[0x%X] FieldList[0x%X] row is None", rid, idx)
+                continue
+            token: int = calculate_dotnet_token_value(field.table.number, field.row_index)
+            yield DnType(token, typedef.TypeName, namespace=typedef.TypeNamespace, member=field.row.Name)
+
+
+def get_dotnet_managed_method_bodies(pe: dnfile.dnPE) -> Iterator[Tuple[int, CilMethodBody]]:
+    """get managed methods from MethodDef table"""
+    for rid, method_def in iter_dotnet_table(pe, dnfile.mdtable.MethodDef.number):
+        assert isinstance(method_def, dnfile.mdtable.MethodDefRow)
+
+        if not method_def.ImplFlags.miIL or any((method_def.Flags.mdAbstract, method_def.Flags.mdPinvokeImpl)):
+            # skip methods that do not have a method body
+            continue
+
+        body: Optional[CilMethodBody] = read_dotnet_method_body(pe, method_def)
+        if body is None:
+            logger.debug("MethodDef[0x%X] method body is None", rid)
+            continue
+
+        token: int = calculate_dotnet_token_value(dnfile.mdtable.MethodDef.number, rid)
+        yield token, body
+
+
+def get_dotnet_unmanaged_imports(pe: dnfile.dnPE) -> Iterator[DnUnmanagedMethod]:
+    """get unmanaged imports from ImplMap table
+
+    see https://www.ntcore.com/files/dotnetformat.htm
+
+    28 - ImplMap Table
+        ImplMap table holds information about unmanaged methods that can be reached from managed code, using PInvoke dispatch
+            MemberForwarded (index into the Field or MethodDef table; more precisely, a MemberForwarded coded index)
+            ImportName (index into the String heap)
+            ImportScope (index into the ModuleRef table)
+    """
+    for rid, impl_map in iter_dotnet_table(pe, dnfile.mdtable.ImplMap.number):
+        assert isinstance(impl_map, dnfile.mdtable.ImplMapRow)
+
+        module: str
+        if impl_map.ImportScope.row is None:
+            logger.debug("ImplMap[0x%X] ImportScope row is None", rid)
+            module = ""
+        else:
+            module = impl_map.ImportScope.row.Name
+        method: str = impl_map.ImportName
+
+        member_forward_table: int
+        if impl_map.MemberForwarded.table is None:
+            logger.debug("ImplMap[0x%X] MemberForwarded table is None", rid)
+            continue
+        else:
+            member_forward_table = impl_map.MemberForwarded.table.number
+        member_forward_row: int = impl_map.MemberForwarded.row_index
+
+        # ECMA says "Each row of the ImplMap table associates a row in the MethodDef table (MemberForwarded) with the
+        # name of a routine (ImportName) in some unmanaged DLL (ImportScope)"; so we calculate and map the MemberForwarded
+        # MethodDef table token to help us later record native import method calls made from CIL
+        token: int = calculate_dotnet_token_value(member_forward_table, member_forward_row)
+
+        # like Kernel32.dll
+        if module and "." in module:
+            module = module.split(".")[0]
+
+        # like kernel32.CreateFileA
+        yield DnUnmanagedMethod(token, module, method)
+
+
+def get_dotnet_types(pe: dnfile.dnPE) -> Iterator[DnType]:
+    """get .NET types from TypeDef and TypeRef tables"""
+    for rid, typedef in iter_dotnet_table(pe, dnfile.mdtable.TypeDef.number):
+        assert isinstance(typedef, dnfile.mdtable.TypeDefRow)
+
+        typedef_token: int = calculate_dotnet_token_value(dnfile.mdtable.TypeDef.number, rid)
+        yield DnType(typedef_token, typedef.TypeName, namespace=typedef.TypeNamespace)
+
+    for rid, typeref in iter_dotnet_table(pe, dnfile.mdtable.TypeRef.number):
+        assert isinstance(typeref, dnfile.mdtable.TypeRefRow)
+
+        typeref_token: int = calculate_dotnet_token_value(dnfile.mdtable.TypeRef.number, rid)
+        yield DnType(typeref_token, typeref.TypeName, namespace=typeref.TypeNamespace)
+
+
+def calculate_dotnet_token_value(table: int, rid: int) -> int:
+    return ((table & 0xFF) << Token.TABLE_SHIFT) | (rid & Token.RID_MASK)
+
+
+def is_dotnet_mixed_mode(pe: dnfile.dnPE) -> bool:
+    assert pe.net is not None
+    assert pe.net.Flags is not None
+
+    return not bool(pe.net.Flags.CLR_ILONLY)
+
+
+def iter_dotnet_table(pe: dnfile.dnPE, table_index: int) -> Iterator[Tuple[int, dnfile.base.MDTableRow]]:
+    assert pe.net is not None
+    assert pe.net.mdtables is not None
+
+    for rid, row in enumerate(pe.net.mdtables.tables.get(table_index, [])):
+        # .NET tables are 1-indexed
+        yield rid + 1, row

capa/features/extractors/dnfile/insn.py

@@ -0,0 +1,227 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+
+from __future__ import annotations
+
+import logging
+from typing import TYPE_CHECKING, Tuple, Union, Iterator, Optional
+
+if TYPE_CHECKING:
+    from capa.features.extractors.dnfile.extractor import DnFileFeatureExtractorCache
+
+import dnfile
+from dncil.clr.token import Token, StringToken, InvalidToken
+from dncil.cil.opcode import OpCodes
+
+import capa.features.extractors.helpers
+from capa.features.insn import API, Number, Property
+from capa.features.common import Class, String, Feature, Namespace, FeatureAccess, Characteristic
+from capa.features.address import Address
+from capa.features.extractors.dnfile.types import DnType, DnUnmanagedMethod
+from capa.features.extractors.base_extractor import BBHandle, InsnHandle, FunctionHandle
+from capa.features.extractors.dnfile.helpers import (
+    resolve_dotnet_token,
+    read_dotnet_user_string,
+    calculate_dotnet_token_value,
+)
+
+logger = logging.getLogger(__name__)
+
+
+def get_callee(
+    pe: dnfile.dnPE, cache: DnFileFeatureExtractorCache, token: Token
+) -> Optional[Union[DnType, DnUnmanagedMethod]]:
+    """map .NET token to un/managed (generic) method"""
+    token_: int
+    if token.table == dnfile.mdtable.MethodSpec.number:
+        # map MethodSpec to MethodDef or MemberRef
+        row: Union[dnfile.base.MDTableRow, InvalidToken, str] = resolve_dotnet_token(pe, token)
+        assert isinstance(row, dnfile.mdtable.MethodSpecRow)
+
+        if row.Method.table is None:
+            logger.debug("MethodSpec[0x%X] Method table is None", token.rid)
+            return None
+
+        token_ = calculate_dotnet_token_value(row.Method.table.number, row.Method.row_index)
+    else:
+        token_ = token.value
+
+    callee: Optional[Union[DnType, DnUnmanagedMethod]] = cache.get_import(token_)
+    if callee is None:
+        # we must check unmanaged imports before managed methods because we map forwarded managed methods
+        # to their unmanaged imports; we prefer a forwarded managed method be mapped to its unmanaged import for analysis
+        callee = cache.get_native_import(token_)
+        if callee is None:
+            callee = cache.get_method(token_)
+    return callee
+
+
+def extract_insn_api_features(fh: FunctionHandle, bh, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
+    """parse instruction API features"""
+    if ih.inner.opcode not in (
+        OpCodes.Call,
+        OpCodes.Callvirt,
+        OpCodes.Jmp,
+        OpCodes.Newobj,
+    ):
+        return
+
+    callee: Optional[Union[DnType, DnUnmanagedMethod]] = get_callee(fh.ctx["pe"], fh.ctx["cache"], ih.inner.operand)
+    if isinstance(callee, DnType):
+        # ignore methods used to access properties
+        if callee.access is None:
+            # like System.IO.File::Delete
+            yield API(str(callee)), ih.address
+    elif isinstance(callee, DnUnmanagedMethod):
+        # like kernel32.CreateFileA
+        for name in capa.features.extractors.helpers.generate_symbols(callee.module, callee.method):
+            yield API(name), ih.address
+
+
+def extract_insn_property_features(fh: FunctionHandle, bh, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
+    """parse instruction property features"""
+    name: Optional[str] = None
+    access: Optional[str] = None
+
+    if ih.inner.opcode in (OpCodes.Call, OpCodes.Callvirt, OpCodes.Jmp):
+        # property access via MethodDef or MemberRef
+        callee: Optional[Union[DnType, DnUnmanagedMethod]] = get_callee(fh.ctx["pe"], fh.ctx["cache"], ih.inner.operand)
+        if isinstance(callee, DnType):
+            if callee.access is not None:
+                name = str(callee)
+                access = callee.access
+
+    elif ih.inner.opcode in (OpCodes.Ldfld, OpCodes.Ldflda, OpCodes.Ldsfld, OpCodes.Ldsflda):
+        # property read via Field
+        read_field: Optional[Union[DnType, DnUnmanagedMethod]] = fh.ctx["cache"].get_field(ih.inner.operand.value)
+        if read_field is not None:
+            name = str(read_field)
+            access = FeatureAccess.READ
+
+    elif ih.inner.opcode in (OpCodes.Stfld, OpCodes.Stsfld):
+        # property write via Field
+        write_field: Optional[Union[DnType, DnUnmanagedMethod]] = fh.ctx["cache"].get_field(ih.inner.operand.value)
+        if write_field is not None:
+            name = str(write_field)
+            access = FeatureAccess.WRITE
+
+    if name is not None:
+        if access is not None:
+            yield Property(name, access=access), ih.address
+        yield Property(name), ih.address
+
+
+def extract_insn_namespace_class_features(
+    fh: FunctionHandle, bh, ih: InsnHandle
+) -> Iterator[Tuple[Union[Namespace, Class], Address]]:
+    """parse instruction namespace and class features"""
+    type_: Optional[Union[DnType, DnUnmanagedMethod]] = None
+
+    if ih.inner.opcode in (
+        OpCodes.Call,
+        OpCodes.Callvirt,
+        OpCodes.Jmp,
+        OpCodes.Ldvirtftn,
+        OpCodes.Ldftn,
+        OpCodes.Newobj,
+    ):
+        # method call - includes managed methods (MethodDef, TypeRef) and properties (MethodSemantics, TypeRef)
+        type_ = get_callee(fh.ctx["pe"], fh.ctx["cache"], ih.inner.operand)
+
+    elif ih.inner.opcode in (
+        OpCodes.Ldfld,
+        OpCodes.Ldflda,
+        OpCodes.Ldsfld,
+        OpCodes.Ldsflda,
+        OpCodes.Stfld,
+        OpCodes.Stsfld,
+    ):
+        # field access
+        type_ = fh.ctx["cache"].get_field(ih.inner.operand.value)
+
+    # ECMA 335 VI.C.4.10
+    elif ih.inner.opcode in (
+        OpCodes.Initobj,
+        OpCodes.Box,
+        OpCodes.Castclass,
+        OpCodes.Cpobj,
+        OpCodes.Isinst,
+        OpCodes.Ldelem,
+        OpCodes.Ldelema,
+        OpCodes.Ldobj,
+        OpCodes.Mkrefany,
+        OpCodes.Newarr,
+        OpCodes.Refanyval,
+        OpCodes.Sizeof,
+        OpCodes.Stobj,
+        OpCodes.Unbox,
+        OpCodes.Constrained,
+        OpCodes.Stelem,
+        OpCodes.Unbox_Any,
+    ):
+        # type access
+        type_ = fh.ctx["cache"].get_type(ih.inner.operand.value)
+
+    if isinstance(type_, DnType):
+        yield Class(DnType.format_name(type_.class_, namespace=type_.namespace)), ih.address
+        if type_.namespace:
+            yield Namespace(type_.namespace), ih.address
+
+
+def extract_insn_number_features(fh, bh, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
+    """parse instruction number features"""
+    if ih.inner.is_ldc():
+        yield Number(ih.inner.get_ldc()), ih.address
+
+
+def extract_insn_string_features(fh: FunctionHandle, bh, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
+    """parse instruction string features"""
+    if not ih.inner.is_ldstr():
+        return
+
+    if not isinstance(ih.inner.operand, StringToken):
+        return
+
+    user_string: Optional[str] = read_dotnet_user_string(fh.ctx["pe"], ih.inner.operand)
+    if user_string is None:
+        return
+
+    if len(user_string) >= 4:
+        yield String(user_string), ih.address
+
+
+def extract_unmanaged_call_characteristic_features(
+    fh: FunctionHandle, bb: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Characteristic, Address]]:
+    if ih.inner.opcode not in (OpCodes.Call, OpCodes.Callvirt, OpCodes.Jmp):
+        return
+
+    row: Union[str, InvalidToken, dnfile.base.MDTableRow] = resolve_dotnet_token(fh.ctx["pe"], ih.inner.operand)
+    if not isinstance(row, dnfile.mdtable.MethodDefRow):
+        return
+
+    if any((row.Flags.mdPinvokeImpl, row.ImplFlags.miUnmanaged, row.ImplFlags.miNative)):
+        yield Characteristic("unmanaged call"), ih.address
+
+
+def extract_features(fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
+    """extract instruction features"""
+    for inst_handler in INSTRUCTION_HANDLERS:
+        for feature, addr in inst_handler(fh, bbh, ih):
+            assert isinstance(addr, Address)
+            yield feature, addr
+
+
+INSTRUCTION_HANDLERS = (
+    extract_insn_api_features,
+    extract_insn_property_features,
+    extract_insn_number_features,
+    extract_insn_string_features,
+    extract_insn_namespace_class_features,
+    extract_unmanaged_call_characteristic_features,
+)

capa/features/extractors/dnfile/types.py

@@ -0,0 +1,74 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+
+from typing import Optional
+
+
+class DnType:
+    def __init__(self, token: int, class_: str, namespace: str = "", member: str = "", access: Optional[str] = None):
+        self.token: int = token
+        self.access: Optional[str] = access
+        self.namespace: str = namespace
+        self.class_: str = class_
+
+        if member == ".ctor":
+            member = "ctor"
+        if member == ".cctor":
+            member = "cctor"
+
+        self.member: str = member
+
+    def __hash__(self):
+        return hash((self.token, self.access, self.namespace, self.class_, self.member))
+
+    def __eq__(self, other):
+        return (
+            self.token == other.token
+            and self.access == other.access
+            and self.namespace == other.namespace
+            and self.class_ == other.class_
+            and self.member == other.member
+        )
+
+    def __str__(self):
+        return DnType.format_name(self.class_, namespace=self.namespace, member=self.member)
+
+    def __repr__(self):
+        return str(self)
+
+    @staticmethod
+    def format_name(class_: str, namespace: str = "", member: str = ""):
+        # like File::OpenRead
+        name: str = f"{class_}::{member}" if member else class_
+        if namespace:
+            # like System.IO.File::OpenRead
+            name = f"{namespace}.{name}"
+        return name
+
+
+class DnUnmanagedMethod:
+    def __init__(self, token: int, module: str, method: str):
+        self.token: int = token
+        self.module: str = module
+        self.method: str = method
+
+    def __hash__(self):
+        return hash((self.token, self.module, self.method))
+
+    def __eq__(self, other):
+        return self.token == other.token and self.module == other.module and self.method == other.method
+
+    def __str__(self):
+        return DnUnmanagedMethod.format_name(self.module, self.method)
+
+    def __repr__(self):
+        return str(self)
+
+    @staticmethod
+    def format_name(module, method):
+        return f"{module}.{method}"

capa/features/extractors/dnfile_.py

@@ -0,0 +1,158 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import logging
+from typing import Tuple, Iterator
+from pathlib import Path
+
+import dnfile
+import pefile
+
+from capa.features.common import (
+    OS,
+    OS_ANY,
+    ARCH_ANY,
+    ARCH_I386,
+    FORMAT_PE,
+    ARCH_AMD64,
+    FORMAT_DOTNET,
+    Arch,
+    Format,
+    Feature,
+)
+from capa.features.address import NO_ADDRESS, Address, AbsoluteVirtualAddress
+from capa.features.extractors.base_extractor import FeatureExtractor
+
+logger = logging.getLogger(__name__)
+
+
+def extract_file_format(**kwargs) -> Iterator[Tuple[Feature, Address]]:
+    yield Format(FORMAT_PE), NO_ADDRESS
+    yield Format(FORMAT_DOTNET), NO_ADDRESS
+
+
+def extract_file_os(**kwargs) -> Iterator[Tuple[Feature, Address]]:
+    yield OS(OS_ANY), NO_ADDRESS
+
+
+def extract_file_arch(pe: dnfile.dnPE, **kwargs) -> Iterator[Tuple[Feature, Address]]:
+    # to distinguish in more detail, see https://stackoverflow.com/a/23614024/10548020
+    # .NET 4.5 added option: any CPU, 32-bit preferred
+    assert pe.net is not None
+    assert pe.net.Flags is not None
+
+    if pe.net.Flags.CLR_32BITREQUIRED and pe.PE_TYPE == pefile.OPTIONAL_HEADER_MAGIC_PE:
+        yield Arch(ARCH_I386), NO_ADDRESS
+    elif not pe.net.Flags.CLR_32BITREQUIRED and pe.PE_TYPE == pefile.OPTIONAL_HEADER_MAGIC_PE_PLUS:
+        yield Arch(ARCH_AMD64), NO_ADDRESS
+    else:
+        yield Arch(ARCH_ANY), NO_ADDRESS
+
+
+def extract_file_features(pe: dnfile.dnPE) -> Iterator[Tuple[Feature, Address]]:
+    for file_handler in FILE_HANDLERS:
+        for feature, address in file_handler(pe=pe):  # type: ignore
+            yield feature, address
+
+
+FILE_HANDLERS = (
+    # extract_file_export_names,
+    # extract_file_import_names,
+    # extract_file_section_names,
+    # extract_file_strings,
+    # extract_file_function_names,
+    extract_file_format,
+)
+
+
+def extract_global_features(pe: dnfile.dnPE) -> Iterator[Tuple[Feature, Address]]:
+    for handler in GLOBAL_HANDLERS:
+        for feature, addr in handler(pe=pe):  # type: ignore
+            yield feature, addr
+
+
+GLOBAL_HANDLERS = (
+    extract_file_os,
+    extract_file_arch,
+)
+
+
+class DnfileFeatureExtractor(FeatureExtractor):
+    def __init__(self, path: Path):
+        super().__init__()
+        self.path: Path = path
+        self.pe: dnfile.dnPE = dnfile.dnPE(str(path))
+
+    def get_base_address(self) -> AbsoluteVirtualAddress:
+        return AbsoluteVirtualAddress(0x0)
+
+    def get_entry_point(self) -> int:
+        # self.pe.net.Flags.CLT_NATIVE_ENTRYPOINT
+        #  True: native EP: Token
+        #  False: managed EP: RVA
+        assert self.pe.net is not None
+        assert self.pe.net.struct is not None
+
+        return self.pe.net.struct.EntryPointTokenOrRva
+
+    def extract_global_features(self):
+        yield from extract_global_features(self.pe)
+
+    def extract_file_features(self):
+        yield from extract_file_features(self.pe)
+
+    def is_dotnet_file(self) -> bool:
+        return bool(self.pe.net)
+
+    def is_mixed_mode(self) -> bool:
+        assert self.pe is not None
+        assert self.pe.net is not None
+        assert self.pe.net.Flags is not None
+
+        return not bool(self.pe.net.Flags.CLR_ILONLY)
+
+    def get_runtime_version(self) -> Tuple[int, int]:
+        assert self.pe is not None
+        assert self.pe.net is not None
+        assert self.pe.net.struct is not None
+
+        return self.pe.net.struct.MajorRuntimeVersion, self.pe.net.struct.MinorRuntimeVersion
+
+    def get_meta_version_string(self) -> str:
+        assert self.pe.net is not None
+        assert self.pe.net.metadata is not None
+        assert self.pe.net.metadata.struct is not None
+        assert self.pe.net.metadata.struct.Version is not None
+
+        vbuf = self.pe.net.metadata.struct.Version
+        assert isinstance(vbuf, bytes)
+
+        return vbuf.rstrip(b"\x00").decode("utf-8")
+
+    def get_functions(self):
+        raise NotImplementedError("DnfileFeatureExtractor can only be used to extract file features")
+
+    def extract_function_features(self, f):
+        raise NotImplementedError("DnfileFeatureExtractor can only be used to extract file features")
+
+    def get_basic_blocks(self, f):
+        raise NotImplementedError("DnfileFeatureExtractor can only be used to extract file features")
+
+    def extract_basic_block_features(self, f, bb):
+        raise NotImplementedError("DnfileFeatureExtractor can only be used to extract file features")
+
+    def get_instructions(self, f, bb):
+        raise NotImplementedError("DnfileFeatureExtractor can only be used to extract file features")
+
+    def extract_insn_features(self, f, bb, insn):
+        raise NotImplementedError("DnfileFeatureExtractor can only be used to extract file features")
+
+    def is_library_function(self, va):
+        raise NotImplementedError("DnfileFeatureExtractor can only be used to extract file features")
+
+    def get_function_name(self, va):
+        raise NotImplementedError("DnfileFeatureExtractor can only be used to extract file features")

capa/features/extractors/dotnetfile.py

@@ -0,0 +1,239 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import logging
+from typing import Tuple, Iterator
+from pathlib import Path
+
+import dnfile
+import pefile
+
+import capa.features.extractors.helpers
+from capa.features.file import Import, FunctionName
+from capa.features.common import (
+    OS,
+    OS_ANY,
+    ARCH_ANY,
+    ARCH_I386,
+    FORMAT_PE,
+    ARCH_AMD64,
+    FORMAT_DOTNET,
+    Arch,
+    Class,
+    Format,
+    String,
+    Feature,
+    Namespace,
+    Characteristic,
+)
+from capa.features.address import NO_ADDRESS, Address, DNTokenAddress
+from capa.features.extractors.base_extractor import FeatureExtractor
+from capa.features.extractors.dnfile.helpers import (
+    DnType,
+    iter_dotnet_table,
+    is_dotnet_mixed_mode,
+    get_dotnet_managed_imports,
+    get_dotnet_managed_methods,
+    calculate_dotnet_token_value,
+    get_dotnet_unmanaged_imports,
+)
+
+logger = logging.getLogger(__name__)
+
+
+def extract_file_format(**kwargs) -> Iterator[Tuple[Format, Address]]:
+    yield Format(FORMAT_PE), NO_ADDRESS
+    yield Format(FORMAT_DOTNET), NO_ADDRESS
+
+
+def extract_file_import_names(pe: dnfile.dnPE, **kwargs) -> Iterator[Tuple[Import, Address]]:
+    for method in get_dotnet_managed_imports(pe):
+        # like System.IO.File::OpenRead
+        yield Import(str(method)), DNTokenAddress(method.token)
+
+    for imp in get_dotnet_unmanaged_imports(pe):
+        # like kernel32.CreateFileA
+        for name in capa.features.extractors.helpers.generate_symbols(imp.module, imp.method):
+            yield Import(name), DNTokenAddress(imp.token)
+
+
+def extract_file_function_names(pe: dnfile.dnPE, **kwargs) -> Iterator[Tuple[FunctionName, Address]]:
+    for method in get_dotnet_managed_methods(pe):
+        yield FunctionName(str(method)), DNTokenAddress(method.token)
+
+
+def extract_file_namespace_features(pe: dnfile.dnPE, **kwargs) -> Iterator[Tuple[Namespace, Address]]:
+    """emit namespace features from TypeRef and TypeDef tables"""
+
+    # namespaces may be referenced multiple times, so we need to filter
+    namespaces = set()
+
+    for _, typedef in iter_dotnet_table(pe, dnfile.mdtable.TypeDef.number):
+        # emit internal .NET namespaces
+        assert isinstance(typedef, dnfile.mdtable.TypeDefRow)
+        namespaces.add(typedef.TypeNamespace)
+
+    for _, typeref in iter_dotnet_table(pe, dnfile.mdtable.TypeRef.number):
+        # emit external .NET namespaces
+        assert isinstance(typeref, dnfile.mdtable.TypeRefRow)
+        namespaces.add(typeref.TypeNamespace)
+
+    # namespaces may be empty, discard
+    namespaces.discard("")
+
+    for namespace in namespaces:
+        # namespace do not have an associated token, so we yield 0x0
+        yield Namespace(namespace), NO_ADDRESS
+
+
+def extract_file_class_features(pe: dnfile.dnPE, **kwargs) -> Iterator[Tuple[Class, Address]]:
+    """emit class features from TypeRef and TypeDef tables"""
+    for rid, typedef in iter_dotnet_table(pe, dnfile.mdtable.TypeDef.number):
+        # emit internal .NET classes
+        assert isinstance(typedef, dnfile.mdtable.TypeDefRow)
+
+        token = calculate_dotnet_token_value(dnfile.mdtable.TypeDef.number, rid)
+        yield Class(DnType.format_name(typedef.TypeName, namespace=typedef.TypeNamespace)), DNTokenAddress(token)
+
+    for rid, typeref in iter_dotnet_table(pe, dnfile.mdtable.TypeRef.number):
+        # emit external .NET classes
+        assert isinstance(typeref, dnfile.mdtable.TypeRefRow)
+
+        token = calculate_dotnet_token_value(dnfile.mdtable.TypeRef.number, rid)
+        yield Class(DnType.format_name(typeref.TypeName, namespace=typeref.TypeNamespace)), DNTokenAddress(token)
+
+
+def extract_file_os(**kwargs) -> Iterator[Tuple[OS, Address]]:
+    yield OS(OS_ANY), NO_ADDRESS
+
+
+def extract_file_arch(pe: dnfile.dnPE, **kwargs) -> Iterator[Tuple[Arch, Address]]:
+    # to distinguish in more detail, see https://stackoverflow.com/a/23614024/10548020
+    # .NET 4.5 added option: any CPU, 32-bit preferred
+    assert pe.net is not None
+    assert pe.net.Flags is not None
+
+    if pe.net.Flags.CLR_32BITREQUIRED and pe.PE_TYPE == pefile.OPTIONAL_HEADER_MAGIC_PE:
+        yield Arch(ARCH_I386), NO_ADDRESS
+    elif not pe.net.Flags.CLR_32BITREQUIRED and pe.PE_TYPE == pefile.OPTIONAL_HEADER_MAGIC_PE_PLUS:
+        yield Arch(ARCH_AMD64), NO_ADDRESS
+    else:
+        yield Arch(ARCH_ANY), NO_ADDRESS
+
+
+def extract_file_strings(pe: dnfile.dnPE, **kwargs) -> Iterator[Tuple[String, Address]]:
+    yield from capa.features.extractors.common.extract_file_strings(pe.__data__)
+
+
+def extract_file_mixed_mode_characteristic_features(
+    pe: dnfile.dnPE, **kwargs
+) -> Iterator[Tuple[Characteristic, Address]]:
+    if is_dotnet_mixed_mode(pe):
+        yield Characteristic("mixed mode"), NO_ADDRESS
+
+
+def extract_file_features(pe: dnfile.dnPE) -> Iterator[Tuple[Feature, Address]]:
+    for file_handler in FILE_HANDLERS:
+        for feature, addr in file_handler(pe=pe):  # type: ignore
+            yield feature, addr
+
+
+FILE_HANDLERS = (
+    extract_file_import_names,
+    extract_file_function_names,
+    extract_file_strings,
+    extract_file_format,
+    extract_file_mixed_mode_characteristic_features,
+    extract_file_namespace_features,
+    extract_file_class_features,
+)
+
+
+def extract_global_features(pe: dnfile.dnPE) -> Iterator[Tuple[Feature, Address]]:
+    for handler in GLOBAL_HANDLERS:
+        for feature, va in handler(pe=pe):  # type: ignore
+            yield feature, va
+
+
+GLOBAL_HANDLERS = (
+    extract_file_os,
+    extract_file_arch,
+)
+
+
+class DotnetFileFeatureExtractor(FeatureExtractor):
+    def __init__(self, path: Path):
+        super().__init__()
+        self.path: Path = path
+        self.pe: dnfile.dnPE = dnfile.dnPE(str(path))
+
+    def get_base_address(self):
+        return NO_ADDRESS
+
+    def get_entry_point(self) -> int:
+        # self.pe.net.Flags.CLT_NATIVE_ENTRYPOINT
+        #  True: native EP: Token
+        #  False: managed EP: RVA
+        assert self.pe.net is not None
+        assert self.pe.net.struct is not None
+
+        return self.pe.net.struct.EntryPointTokenOrRva
+
+    def extract_global_features(self):
+        yield from extract_global_features(self.pe)
+
+    def extract_file_features(self):
+        yield from extract_file_features(self.pe)
+
+    def is_dotnet_file(self) -> bool:
+        return bool(self.pe.net)
+
+    def is_mixed_mode(self) -> bool:
+        return is_dotnet_mixed_mode(self.pe)
+
+    def get_runtime_version(self) -> Tuple[int, int]:
+        assert self.pe.net is not None
+        assert self.pe.net.struct is not None
+        assert self.pe.net.struct.MajorRuntimeVersion is not None
+        assert self.pe.net.struct.MinorRuntimeVersion is not None
+
+        return self.pe.net.struct.MajorRuntimeVersion, self.pe.net.struct.MinorRuntimeVersion
+
+    def get_meta_version_string(self) -> str:
+        assert self.pe.net is not None
+        assert self.pe.net.metadata is not None
+        assert self.pe.net.metadata.struct is not None
+        assert self.pe.net.metadata.struct.Version is not None
+
+        vbuf = self.pe.net.metadata.struct.Version
+        assert isinstance(vbuf, bytes)
+
+        return vbuf.rstrip(b"\x00").decode("utf-8")
+
+    def get_functions(self):
+        raise NotImplementedError("DotnetFileFeatureExtractor can only be used to extract file features")
+
+    def extract_function_features(self, f):
+        raise NotImplementedError("DotnetFileFeatureExtractor can only be used to extract file features")
+
+    def get_basic_blocks(self, f):
+        raise NotImplementedError("DotnetFileFeatureExtractor can only be used to extract file features")
+
+    def extract_basic_block_features(self, f, bb):
+        raise NotImplementedError("DotnetFileFeatureExtractor can only be used to extract file features")
+
+    def get_instructions(self, f, bb):
+        raise NotImplementedError("DotnetFileFeatureExtractor can only be used to extract file features")
+
+    def extract_insn_features(self, f, bb, insn):
+        raise NotImplementedError("DotnetFileFeatureExtractor can only be used to extract file features")
+
+    def is_library_function(self, va):
+        raise NotImplementedError("DotnetFileFeatureExtractor can only be used to extract file features")
+
+    def get_function_name(self, va):
+        raise NotImplementedError("DotnetFileFeatureExtractor can only be used to extract file features")

capa/features/extractors/elf.py

@@ -0,0 +1,985 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import struct
+import logging
+import itertools
+import collections
+from enum import Enum
+from typing import Set, Dict, List, Tuple, BinaryIO, Iterator, Optional
+from dataclasses import dataclass
+
+import Elf  # from vivisect
+
+logger = logging.getLogger(__name__)
+
+
+def align(v, alignment):
+    remainder = v % alignment
+    if remainder == 0:
+        return v
+    else:
+        return v + (alignment - remainder)
+
+
+def read_cstr(buf, offset) -> str:
+    s = buf[offset:]
+    s, _, _ = s.partition(b"\x00")
+    return s.decode("utf-8")
+
+
+class CorruptElfFile(ValueError):
+    pass
+
+
+class OS(str, Enum):
+    HPUX = "hpux"
+    NETBSD = "netbsd"
+    LINUX = "linux"
+    HURD = "hurd"
+    _86OPEN = "86open"
+    SOLARIS = "solaris"
+    AIX = "aix"
+    IRIX = "irix"
+    FREEBSD = "freebsd"
+    TRU64 = "tru64"
+    MODESTO = "modesto"
+    OPENBSD = "openbsd"
+    OPENVMS = "openvms"
+    NSK = "nsk"
+    AROS = "aros"
+    FENIXOS = "fenixos"
+    CLOUD = "cloud"
+    SYLLABLE = "syllable"
+    NACL = "nacl"
+    ANDROID = "android"
+
+
+# via readelf: https://github.com/bminor/binutils-gdb/blob/c0e94211e1ac05049a4ce7c192c9d14d1764eb3e/binutils/readelf.c#L19635-L19658
+# and here: https://github.com/bminor/binutils-gdb/blob/34c54daa337da9fadf87d2706d6a590ae1f88f4d/include/elf/common.h#L933-L939
+GNU_ABI_TAG = {
+    0: OS.LINUX,
+    1: OS.HURD,
+    2: OS.SOLARIS,
+    3: OS.FREEBSD,
+    4: OS.NETBSD,
+    5: OS.SYLLABLE,
+    6: OS.NACL,
+}
+
+
+@dataclass
+class Phdr:
+    type: int
+    offset: int
+    vaddr: int
+    paddr: int
+    filesz: int
+    buf: bytes
+
+
+@dataclass
+class Shdr:
+    name: int
+    type: int
+    flags: int
+    addr: int
+    offset: int
+    size: int
+    link: int
+    entsize: int
+    buf: bytes
+
+    @classmethod
+    def from_viv(cls, section, buf: bytes) -> "Shdr":
+        return cls(
+            section.sh_name,
+            section.sh_type,
+            section.sh_flags,
+            section.sh_addr,
+            section.sh_offset,
+            section.sh_size,
+            section.sh_link,
+            section.sh_entsize,
+            buf,
+        )
+
+
+class ELF:
+    def __init__(self, f: BinaryIO):
+        self.f = f
+
+        # these will all be initialized in `_parse()`
+        self.bitness: int
+        self.endian: str
+        self.e_phentsize: int
+        self.e_phnum: int
+        self.e_shentsize: int
+        self.e_shnum: int
+        self.phbuf: bytes
+        self.shbuf: bytes
+
+        self._parse()
+
+    def _parse(self):
+        self.f.seek(0x0)
+        self.file_header = self.f.read(0x40)
+
+        if not self.file_header.startswith(b"\x7fELF"):
+            raise CorruptElfFile("missing magic header")
+
+        ei_class, ei_data = struct.unpack_from("BB", self.file_header, 4)
+        logger.debug("ei_class: 0x%02x ei_data: 0x%02x", ei_class, ei_data)
+        if ei_class == 1:
+            self.bitness = 32
+        elif ei_class == 2:
+            self.bitness = 64
+        else:
+            raise CorruptElfFile(f"invalid ei_class: 0x{ei_class:02x}")
+
+        if ei_data == 1:
+            self.endian = "<"
+        elif ei_data == 2:
+            self.endian = ">"
+        else:
+            raise CorruptElfFile(f"not an ELF file: invalid ei_data: 0x{ei_data:02x}")
+
+        if self.bitness == 32:
+            e_phoff, e_shoff = struct.unpack_from(self.endian + "II", self.file_header, 0x1C)
+            self.e_phentsize, self.e_phnum = struct.unpack_from(self.endian + "HH", self.file_header, 0x2A)
+            self.e_shentsize, self.e_shnum = struct.unpack_from(self.endian + "HH", self.file_header, 0x2E)
+        elif self.bitness == 64:
+            e_phoff, e_shoff = struct.unpack_from(self.endian + "QQ", self.file_header, 0x20)
+            self.e_phentsize, self.e_phnum = struct.unpack_from(self.endian + "HH", self.file_header, 0x36)
+            self.e_shentsize, self.e_shnum = struct.unpack_from(self.endian + "HH", self.file_header, 0x3A)
+        else:
+            raise NotImplementedError()
+
+        logger.debug("e_phoff: 0x%02x e_phentsize: 0x%02x e_phnum: %d", e_phoff, self.e_phentsize, self.e_phnum)
+
+        self.f.seek(e_phoff)
+        program_header_size = self.e_phnum * self.e_phentsize
+        self.phbuf = self.f.read(program_header_size)
+        if len(self.phbuf) != program_header_size:
+            logger.warning("failed to read program headers")
+            self.e_phnum = 0
+
+        self.f.seek(e_shoff)
+        section_header_size = self.e_shnum * self.e_shentsize
+        self.shbuf = self.f.read(section_header_size)
+        if len(self.shbuf) != section_header_size:
+            logger.warning("failed to read section headers")
+            self.e_shnum = 0
+
+    OSABI = {
+        # via pyelftools: https://github.com/eliben/pyelftools/blob/0664de05ed2db3d39041e2d51d19622a8ef4fb0f/elftools/elf/enums.py#L35-L58
+        # some candidates are commented out because the are not useful values,
+        # at least when guessing OSes
+        # 0: "SYSV",  # too often used when OS is not SYSV
+        1: OS.HPUX,
+        2: OS.NETBSD,
+        3: OS.LINUX,
+        4: OS.HURD,
+        5: OS._86OPEN,
+        6: OS.SOLARIS,
+        7: OS.AIX,
+        8: OS.IRIX,
+        9: OS.FREEBSD,
+        10: OS.TRU64,
+        11: OS.MODESTO,
+        12: OS.OPENBSD,
+        13: OS.OPENVMS,
+        14: OS.NSK,
+        15: OS.AROS,
+        16: OS.FENIXOS,
+        17: OS.CLOUD,
+        # 53: "SORTFIX",      # i can't find any reference to this OS, i dont think it exists
+        # 64: "ARM_AEABI",    # not an OS
+        # 97: "ARM",          # not an OS
+        # 255: "STANDALONE",  # not an OS
+    }
+
+    @property
+    def ei_osabi(self) -> Optional[OS]:
+        (ei_osabi,) = struct.unpack_from(self.endian + "B", self.file_header, 7)
+        return ELF.OSABI.get(ei_osabi)
+
+    MACHINE = {
+        # via https://refspecs.linuxfoundation.org/elf/gabi4+/ch4.eheader.html
+        1: "M32",
+        2: "SPARC",
+        3: "i386",
+        4: "68K",
+        5: "88K",
+        6: "486",
+        7: "860",
+        8: "MIPS",
+        9: "S370",
+        10: "MIPS_RS3_LE",
+        11: "RS6000",
+        15: "PA_RISC",
+        16: "nCUBE",
+        17: "VPP500",
+        18: "SPARC32PLUS",
+        19: "960",
+        20: "PPC",
+        21: "PPC64",
+        22: "S390",
+        23: "SPU",
+        36: "V800",
+        37: "FR20",
+        38: "RH32",
+        39: "RCE",
+        40: "ARM",
+        41: "ALPHA",
+        42: "SH",
+        43: "SPARCV9",
+        44: "TRICORE",
+        45: "ARC",
+        46: "H8_300",
+        47: "H8_300H",
+        48: "H8S",
+        49: "H8_500",
+        50: "IA_64",
+        51: "MIPS_X",
+        52: "COLDFIRE",
+        53: "68HC12",
+        54: "MMA",
+        55: "PCP",
+        56: "NCPU",
+        57: "NDR1",
+        58: "STARCORE",
+        59: "ME16",
+        60: "ST100",
+        61: "TINYJ",
+        62: "amd64",
+        63: "PDSP",
+        64: "PDP10",
+        65: "PDP11",
+        66: "FX66",
+        67: "ST9PLUS",
+        68: "ST7",
+        69: "68HC16",
+        70: "68HC11",
+        71: "68HC08",
+        72: "68HC05",
+        73: "SVX",
+        74: "ST19",
+        75: "VAX",
+        76: "CRIS",
+        77: "JAVELIN",
+        78: "FIREPATH",
+        79: "ZSP",
+        80: "MMIX",
+        81: "HUANY",
+        82: "PRISM",
+        83: "AVR",
+        84: "FR30",
+        85: "D10V",
+        86: "D30V",
+        87: "V850",
+        88: "M32R",
+        89: "MN10300",
+        90: "MN10200",
+        91: "PJ",
+        92: "OPENRISC",
+        93: "ARC_A5",
+        94: "XTENSA",
+        95: "VIDEOCORE",
+        96: "TMM_GPP",
+        97: "NS32K",
+        98: "TPC",
+        99: "SNP1K",
+        100: "ST200",
+    }
+
+    @property
+    def e_machine(self) -> Optional[str]:
+        (e_machine,) = struct.unpack_from(self.endian + "H", self.file_header, 0x12)
+        return ELF.MACHINE.get(e_machine)
+
+    def parse_program_header(self, i) -> Phdr:
+        phent_offset = i * self.e_phentsize
+        phent = self.phbuf[phent_offset : phent_offset + self.e_phentsize]
+
+        (p_type,) = struct.unpack_from(self.endian + "I", phent, 0x0)
+        logger.debug("ph:p_type: 0x%04x", p_type)
+
+        if self.bitness == 32:
+            p_offset, p_vaddr, p_paddr, p_filesz = struct.unpack_from(self.endian + "IIII", phent, 0x4)
+        elif self.bitness == 64:
+            p_offset, p_vaddr, p_paddr, p_filesz = struct.unpack_from(self.endian + "QQQQ", phent, 0x8)
+        else:
+            raise NotImplementedError()
+
+        logger.debug("ph:p_offset: 0x%02x p_filesz: 0x%04x", p_offset, p_filesz)
+
+        self.f.seek(p_offset)
+        buf = self.f.read(p_filesz)
+        if len(buf) != p_filesz:
+            raise ValueError("failed to read program header content")
+
+        return Phdr(p_type, p_offset, p_vaddr, p_paddr, p_filesz, buf)
+
+    @property
+    def program_headers(self):
+        for i in range(self.e_phnum):
+            try:
+                yield self.parse_program_header(i)
+            except ValueError:
+                continue
+
+    def parse_section_header(self, i) -> Shdr:
+        shent_offset = i * self.e_shentsize
+        shent = self.shbuf[shent_offset : shent_offset + self.e_shentsize]
+
+        if self.bitness == 32:
+            sh_name, sh_type, sh_flags, sh_addr, sh_offset, sh_size, sh_link, _, _, sh_entsize = struct.unpack_from(
+                self.endian + "IIIIIIIIII", shent, 0x0
+            )
+        elif self.bitness == 64:
+            sh_name, sh_type, sh_flags, sh_addr, sh_offset, sh_size, sh_link, _, _, sh_entsize = struct.unpack_from(
+                self.endian + "IIQQQQIIQQ", shent, 0x0
+            )
+        else:
+            raise NotImplementedError()
+
+        logger.debug("sh:sh_offset: 0x%02x sh_size: 0x%04x", sh_offset, sh_size)
+
+        self.f.seek(sh_offset)
+        buf = self.f.read(sh_size)
+        if len(buf) != sh_size:
+            raise ValueError("failed to read section header content")
+
+        return Shdr(sh_name, sh_type, sh_flags, sh_addr, sh_offset, sh_size, sh_link, sh_entsize, buf)
+
+    @property
+    def section_headers(self):
+        for i in range(self.e_shnum):
+            try:
+                yield self.parse_section_header(i)
+            except ValueError:
+                continue
+
+    @property
+    def linker(self):
+        PT_INTERP = 0x3
+        for phdr in self.program_headers:
+            if phdr.type != PT_INTERP:
+                continue
+
+            return read_cstr(phdr.buf, 0)
+
+    @property
+    def versions_needed(self) -> Dict[str, Set[str]]:
+        # symbol version requirements are stored in the .gnu.version_r section,
+        # which has type SHT_GNU_verneed (0x6ffffffe).
+        #
+        # this contains a linked list of ElfXX_Verneed structs,
+        # each referencing a linked list of ElfXX_Vernaux structs.
+        # strings are stored in the section referenced by the sh_link field of the section header.
+        # each Verneed struct contains a reference to the name of the library,
+        # each Vernaux struct contains a reference to the name of a symbol.
+        SHT_GNU_VERNEED = 0x6FFFFFFE
+        for shdr in self.section_headers:
+            if shdr.type != SHT_GNU_VERNEED:
+                continue
+
+            # the linked section contains strings referenced by the verneed structures.
+            linked_shdr = self.parse_section_header(shdr.link)
+
+            versions_needed = collections.defaultdict(set)
+
+            # read verneed structures from the start of the section
+            # until the vn_next link is 0x0.
+            # each entry describes a shared object that is required by this binary.
+            vn_offset = 0x0
+            while True:
+                # ElfXX_Verneed layout is the same on 32 and 64 bit
+                vn_version, vn_cnt, vn_file, vn_aux, vn_next = struct.unpack_from(
+                    self.endian + "HHIII", shdr.buf, vn_offset
+                )
+                if vn_version != 1:
+                    # unexpected format, don't try to keep parsing
+                    break
+
+                # shared object names, like: "libdl.so.2"
+                so_name = read_cstr(linked_shdr.buf, vn_file)
+
+                # read vernaux structures linked from the verneed structure.
+                # there should be vn_cnt of these.
+                # each entry describes an ABI name required by the shared object.
+                vna_offset = vn_offset + vn_aux
+                for _ in range(vn_cnt):
+                    # ElfXX_Vernaux layout is the same on 32 and 64 bit
+                    _, _, _, vna_name, vna_next = struct.unpack_from(self.endian + "IHHII", shdr.buf, vna_offset)
+
+                    # ABI names, like: "GLIBC_2.2.5"
+                    abi = read_cstr(linked_shdr.buf, vna_name)
+                    versions_needed[so_name].add(abi)
+
+                    vna_offset += vna_next
+
+                vn_offset += vn_next
+                if vn_next == 0:
+                    break
+
+            return dict(versions_needed)
+
+        return {}
+
+    @property
+    def dynamic_entries(self) -> Iterator[Tuple[int, int]]:
+        """
+        read the entries from the dynamic section,
+        yielding the tag and value for each entry.
+        """
+        DT_NULL = 0x0
+        PT_DYNAMIC = 0x2
+        for phdr in self.program_headers:
+            if phdr.type != PT_DYNAMIC:
+                continue
+
+            offset = 0x0
+            while True:
+                if self.bitness == 32:
+                    d_tag, d_val = struct.unpack_from(self.endian + "II", phdr.buf, offset)
+                    offset += 8
+                elif self.bitness == 64:
+                    d_tag, d_val = struct.unpack_from(self.endian + "QQ", phdr.buf, offset)
+                    offset += 16
+                else:
+                    raise NotImplementedError()
+
+                if d_tag == DT_NULL:
+                    break
+
+                yield d_tag, d_val
+
+    @property
+    def strtab(self) -> Optional[bytes]:
+        """
+        fetch the bytes of the string table
+        referenced by the dynamic section.
+        """
+        DT_STRTAB = 0x5
+        DT_STRSZ = 0xA
+
+        strtab_addr = None
+        strtab_size = None
+
+        for d_tag, d_val in self.dynamic_entries:
+            if d_tag == DT_STRTAB:
+                strtab_addr = d_val
+                break
+
+        for d_tag, d_val in self.dynamic_entries:
+            if d_tag == DT_STRSZ:
+                strtab_size = d_val
+                break
+
+        if strtab_addr is None:
+            return None
+
+        if strtab_size is None:
+            return None
+
+        strtab_offset = None
+        for shdr in self.section_headers:
+            # the section header address should be defined
+            if shdr.addr and shdr.addr <= strtab_addr < shdr.addr + shdr.size:
+                strtab_offset = shdr.offset + (strtab_addr - shdr.addr)
+                break
+
+        if strtab_offset is None:
+            return None
+
+        self.f.seek(strtab_offset)
+        strtab_buf = self.f.read(strtab_size)
+
+        if len(strtab_buf) != strtab_size:
+            return None
+
+        return strtab_buf
+
+    @property
+    def needed(self) -> Iterator[str]:
+        """
+        read the names of DT_NEEDED entries from the dynamic section,
+        which correspond to dependencies on other shared objects,
+        like: `libpthread.so.0`
+        """
+        DT_NEEDED = 0x1
+        strtab = self.strtab
+        if not strtab:
+            return
+
+        for d_tag, d_val in self.dynamic_entries:
+            if d_tag != DT_NEEDED:
+                continue
+
+            try:
+                yield read_cstr(strtab, d_val)
+            except UnicodeDecodeError as e:
+                logger.warning("failed to read DT_NEEDED entry: %s", str(e))
+
+    @property
+    def symtab(self) -> Optional[Tuple[Shdr, Shdr]]:
+        """
+        fetch the Shdr for the symtab and the associated strtab.
+        """
+        SHT_SYMTAB = 0x2
+        for shdr in self.section_headers:
+            if shdr.type != SHT_SYMTAB:
+                continue
+
+            # the linked section contains strings referenced by the symtab structures.
+            strtab_shdr = self.parse_section_header(shdr.link)
+
+            return shdr, strtab_shdr
+
+        return None
+
+
+@dataclass
+class ABITag:
+    os: OS
+    kmajor: int
+    kminor: int
+    kpatch: int
+
+
+class PHNote:
+    def __init__(self, endian: str, buf: bytes):
+        self.endian = endian
+        self.buf = buf
+
+        # these will be initialized in `_parse()`
+        self.type_: int
+        self.descsz: int
+        self.name: str
+
+        self._parse()
+
+    def _parse(self):
+        namesz, self.descsz, self.type_ = struct.unpack_from(self.endian + "III", self.buf, 0x0)
+        name_offset = 0xC
+        self.desc_offset = name_offset + align(namesz, 0x4)
+
+        logger.debug("ph:namesz: 0x%02x descsz: 0x%02x type: 0x%04x", namesz, self.descsz, self.type_)
+
+        self.name = self.buf[name_offset : name_offset + namesz].partition(b"\x00")[0].decode("ascii")
+        logger.debug("name: %s", self.name)
+
+    @property
+    def abi_tag(self) -> Optional[ABITag]:
+        if self.type_ != 1:
+            # > The type field shall be 1.
+            # Linux Standard Base Specification 1.2
+            # ref: https://refspecs.linuxfoundation.org/LSB_1.2.0/gLSB/noteabitag.html
+            return None
+
+        if self.name != "GNU":
+            return None
+
+        if self.descsz < 16:
+            return None
+
+        desc = self.buf[self.desc_offset : self.desc_offset + self.descsz]
+        abi_tag, kmajor, kminor, kpatch = struct.unpack_from(self.endian + "IIII", desc, 0x0)
+        logger.debug("GNU_ABI_TAG: 0x%02x", abi_tag)
+
+        os = GNU_ABI_TAG.get(abi_tag)
+        if not os:
+            return None
+
+        logger.debug("abi tag: %s earliest compatible kernel: %d.%d.%d", os, kmajor, kminor, kpatch)
+
+        return ABITag(os, kmajor, kminor, kpatch)
+
+
+class SHNote:
+    def __init__(self, endian: str, buf: bytes):
+        self.endian = endian
+        self.buf = buf
+
+        # these will be initialized in `_parse()`
+        self.type_: int
+        self.descsz: int
+        self.name: str
+
+        self._parse()
+
+    def _parse(self):
+        namesz, self.descsz, self.type_ = struct.unpack_from(self.endian + "III", self.buf, 0x0)
+        name_offset = 0xC
+        self.desc_offset = name_offset + align(namesz, 0x4)
+
+        logger.debug("sh:namesz: 0x%02x descsz: 0x%02x type: 0x%04x", namesz, self.descsz, self.type_)
+
+        name_buf = self.buf[name_offset : name_offset + namesz]
+        self.name = read_cstr(name_buf, 0x0)
+        logger.debug("sh:name: %s", self.name)
+
+    @property
+    def abi_tag(self) -> Optional[ABITag]:
+        if self.name != "GNU":
+            return None
+
+        if self.descsz < 16:
+            return None
+
+        desc = self.buf[self.desc_offset : self.desc_offset + self.descsz]
+        abi_tag, kmajor, kminor, kpatch = struct.unpack_from(self.endian + "IIII", desc, 0x0)
+        logger.debug("GNU_ABI_TAG: 0x%02x", abi_tag)
+
+        os = GNU_ABI_TAG.get(abi_tag)
+        if not os:
+            return None
+
+        logger.debug("abi tag: %s earliest compatible kernel: %d.%d.%d", os, kmajor, kminor, kpatch)
+        return ABITag(os, kmajor, kminor, kpatch)
+
+
+@dataclass
+class Symbol:
+    name_offset: int
+    value: int
+    size: int
+    info: int
+    other: int
+    shndx: int
+
+
+class SymTab:
+    def __init__(
+        self,
+        endian: str,
+        bitness: int,
+        symtab: Shdr,
+        strtab: Shdr,
+    ) -> None:
+        self.symbols: List[Symbol] = []
+
+        self.symtab = symtab
+        self.strtab = strtab
+
+        self._parse(endian, bitness, symtab.buf)
+
+    def _parse(self, endian: str, bitness: int, symtab_buf: bytes) -> None:
+        """
+        return the symbol's information in
+        the order specified by sys/elf32.h
+        """
+        if self.symtab.entsize == 0:
+            return
+
+        for i in range(int(len(self.symtab.buf) / self.symtab.entsize)):
+            if bitness == 32:
+                name_offset, value, size, info, other, shndx = struct.unpack_from(
+                    endian + "IIIBBH", symtab_buf, i * self.symtab.entsize
+                )
+            elif bitness == 64:
+                name_offset, info, other, shndx, value, size = struct.unpack_from(
+                    endian + "IBBHQQ", symtab_buf, i * self.symtab.entsize
+                )
+
+            self.symbols.append(Symbol(name_offset, value, size, info, other, shndx))
+
+    def get_name(self, symbol: Symbol) -> str:
+        """
+        fetch a symbol's name from symtab's
+        associated strings' section (SHT_STRTAB)
+        """
+        if not self.strtab:
+            raise ValueError("no strings found")
+
+        for i in range(symbol.name_offset, self.strtab.size):
+            if self.strtab.buf[i] == 0:
+                return self.strtab.buf[symbol.name_offset : i].decode("utf-8")
+
+        raise ValueError("symbol name not found")
+
+    def get_symbols(self) -> Iterator[Symbol]:
+        """
+        return a tuple: (name, value, size, info, other, shndx)
+        for each symbol contained in the symbol table
+        """
+        yield from self.symbols
+
+    @classmethod
+    def from_viv(cls, elf: Elf.Elf) -> Optional["SymTab"]:
+        endian = "<" if elf.getEndian() == 0 else ">"
+        bitness = elf.bits
+
+        SHT_SYMTAB = 0x2
+        for section in elf.sections:
+            if section.sh_type == SHT_SYMTAB:
+                strtab_section = elf.sections[section.sh_link]
+                sh_symtab = Shdr.from_viv(section, elf.readAtOffset(section.sh_offset, section.sh_size))
+                sh_strtab = Shdr.from_viv(
+                    strtab_section, elf.readAtOffset(strtab_section.sh_offset, strtab_section.sh_size)
+                )
+
+        try:
+            return cls(endian, bitness, sh_symtab, sh_strtab)
+        except NameError:
+            return None
+        except Exception:
+            # all exceptions that could be encountered by
+            # cls._parse() imply a faulty symbol's table.
+            raise CorruptElfFile("malformed symbol's table")
+
+
+def guess_os_from_osabi(elf: ELF) -> Optional[OS]:
+    return elf.ei_osabi
+
+
+def guess_os_from_ph_notes(elf: ELF) -> Optional[OS]:
+    # search for PT_NOTE sections that specify an OS
+    # for example, on Linux there is a GNU section with minimum kernel version
+    PT_NOTE = 0x4
+    for phdr in elf.program_headers:
+        if phdr.type != PT_NOTE:
+            continue
+
+        note = PHNote(elf.endian, phdr.buf)
+
+        if note.type_ != 1:
+            # > The type field shall be 1.
+            # Linux Standard Base Specification 1.2
+            # ref: https://refspecs.linuxfoundation.org/LSB_1.2.0/gLSB/noteabitag.html
+            continue
+
+        if note.name == "Linux":
+            logger.debug("note owner: %s", "LINUX")
+            return OS.LINUX
+        elif note.name == "OpenBSD":
+            logger.debug("note owner: %s", "OPENBSD")
+            return OS.OPENBSD
+        elif note.name == "NetBSD":
+            logger.debug("note owner: %s", "NETBSD")
+            return OS.NETBSD
+        elif note.name == "FreeBSD":
+            logger.debug("note owner: %s", "FREEBSD")
+            return OS.FREEBSD
+        elif note.name == "Android":
+            logger.debug("note owner: %s", "Android")
+            # see the following for parsing the structure:
+            # https://android.googlesource.com/platform/ndk/+/master/parse_elfnote.py
+            return OS.ANDROID
+        elif note.name == "GNU":
+            abi_tag = note.abi_tag
+            if abi_tag:
+                return abi_tag.os
+            else:
+                # cannot make a guess about the OS, but probably linux or hurd
+                pass
+
+    return None
+
+
+def guess_os_from_sh_notes(elf: ELF) -> Optional[OS]:
+    # search for notes stored in sections that aren't visible in program headers.
+    # e.g. .note.Linux in Linux kernel modules.
+    SHT_NOTE = 0x7
+    for shdr in elf.section_headers:
+        if shdr.type != SHT_NOTE:
+            continue
+
+        note = SHNote(elf.endian, shdr.buf)
+
+        if note.name == "Linux":
+            logger.debug("note owner: %s", "LINUX")
+            return OS.LINUX
+        elif note.name == "OpenBSD":
+            logger.debug("note owner: %s", "OPENBSD")
+            return OS.OPENBSD
+        elif note.name == "NetBSD":
+            logger.debug("note owner: %s", "NETBSD")
+            return OS.NETBSD
+        elif note.name == "FreeBSD":
+            logger.debug("note owner: %s", "FREEBSD")
+            return OS.FREEBSD
+        elif note.name == "GNU":
+            abi_tag = note.abi_tag
+            if abi_tag:
+                return abi_tag.os
+            else:
+                # cannot make a guess about the OS, but probably linux or hurd
+                pass
+
+    return None
+
+
+def guess_os_from_linker(elf: ELF) -> Optional[OS]:
+    # search for recognizable dynamic linkers (interpreters)
+    # for example, on linux, we see file paths like: /lib64/ld-linux-x86-64.so.2
+    linker = elf.linker
+    if linker and "ld-linux" in elf.linker:
+        return OS.LINUX
+
+    return None
+
+
+def guess_os_from_abi_versions_needed(elf: ELF) -> Optional[OS]:
+    # then lets look for GLIBC symbol versioning requirements.
+    # this will let us guess about linux/hurd in some cases.
+
+    versions_needed = elf.versions_needed
+    if any(abi.startswith("GLIBC") for abi in itertools.chain(*versions_needed.values())):
+        # there are any GLIBC versions needed
+
+        if elf.e_machine != "i386":
+            # GLIBC runs on Linux and Hurd.
+            # for Hurd, its *only* on i386.
+            # so if we're not on i386, then we're on Linux.
+            return OS.LINUX
+
+        else:
+            # we're on i386, so we could be on either Linux or Hurd.
+            linker = elf.linker
+
+            if linker and "ld-linux" in linker:
+                return OS.LINUX
+
+            elif linker and "/ld.so" in linker:
+                return OS.HURD
+
+            else:
+                # we don't have any good guesses based on versions needed
+                pass
+
+    return None
+
+
+def guess_os_from_needed_dependencies(elf: ELF) -> Optional[OS]:
+    for needed in elf.needed:
+        if needed.startswith("libmachuser.so"):
+            return OS.HURD
+        if needed.startswith("libhurduser.so"):
+            return OS.HURD
+        if needed.startswith("libandroid.so"):
+            return OS.ANDROID
+
+    return None
+
+
+def guess_os_from_symtab(elf: ELF) -> Optional[OS]:
+    shdrs = elf.symtab
+    if not shdrs:
+        # executable does not contain a symbol table
+        # or the symbol's names are stripped
+        return None
+
+    symtab_shdr, strtab_shdr = shdrs
+    symtab = SymTab(elf.endian, elf.bitness, symtab_shdr, strtab_shdr)
+
+    keywords = {
+        OS.LINUX: [
+            "linux",
+            "/linux/",
+        ],
+    }
+
+    for symbol in symtab.get_symbols():
+        sym_name = symtab.get_name(symbol)
+
+        for os, hints in keywords.items():
+            if any(hint in sym_name for hint in hints):
+                return os
+
+    return None
+
+
+def detect_elf_os(f) -> str:
+    """
+    f: type Union[BinaryIO, IDAIO, GHIDRAIO]
+    """
+    try:
+        elf = ELF(f)
+    except Exception as e:
+        logger.warning("Error parsing ELF file: %s", e)
+        return "unknown"
+
+    try:
+        osabi_guess = guess_os_from_osabi(elf)
+        logger.debug("guess: osabi: %s", osabi_guess)
+    except Exception as e:
+        logger.warning("Error guessing OS from OSABI: %s", e)
+        osabi_guess = None
+
+    try:
+        ph_notes_guess = guess_os_from_ph_notes(elf)
+        logger.debug("guess: ph notes: %s", ph_notes_guess)
+    except Exception as e:
+        logger.warning("Error guessing OS from program header notes: %s", e)
+        ph_notes_guess = None
+
+    try:
+        sh_notes_guess = guess_os_from_sh_notes(elf)
+        logger.debug("guess: sh notes: %s", sh_notes_guess)
+    except Exception as e:
+        logger.warning("Error guessing OS from section header notes: %s", e)
+        sh_notes_guess = None
+
+    try:
+        linker_guess = guess_os_from_linker(elf)
+        logger.debug("guess: linker: %s", linker_guess)
+    except Exception as e:
+        logger.warning("Error guessing OS from linker: %s", e)
+        linker_guess = None
+
+    try:
+        abi_versions_needed_guess = guess_os_from_abi_versions_needed(elf)
+        logger.debug("guess: ABI versions needed: %s", abi_versions_needed_guess)
+    except Exception as e:
+        logger.warning("Error guessing OS from ABI versions needed: %s", e)
+        abi_versions_needed_guess = None
+
+    try:
+        needed_dependencies_guess = guess_os_from_needed_dependencies(elf)
+        logger.debug("guess: needed dependencies: %s", needed_dependencies_guess)
+    except Exception as e:
+        logger.warning("Error guessing OS from needed dependencies: %s", e)
+        needed_dependencies_guess = None
+
+    try:
+        symtab_guess = guess_os_from_symtab(elf)
+        logger.debug("guess: pertinent symbol name: %s", symtab_guess)
+    except Exception as e:
+        logger.warning("Error guessing OS from symbol table: %s", e)
+        symtab_guess = None
+
+    ret = None
+
+    if osabi_guess:
+        ret = osabi_guess
+
+    elif ph_notes_guess:
+        ret = ph_notes_guess
+
+    elif sh_notes_guess:
+        ret = sh_notes_guess
+
+    elif linker_guess:
+        ret = linker_guess
+
+    elif abi_versions_needed_guess:
+        ret = abi_versions_needed_guess
+
+    elif needed_dependencies_guess:
+        ret = needed_dependencies_guess
+
+    elif symtab_guess:
+        ret = symtab_guess
+
+    return ret.value if ret is not None else "unknown"
+
+
+def detect_elf_arch(f: BinaryIO) -> str:
+    return ELF(f).e_machine or "unknown"

capa/features/extractors/elffile.py

@@ -0,0 +1,203 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import io
+import logging
+from typing import Tuple, Iterator
+from pathlib import Path
+
+from elftools.elf.elffile import ELFFile, SymbolTableSection
+from elftools.elf.relocation import RelocationSection
+
+import capa.features.extractors.common
+from capa.features.file import Export, Import, Section
+from capa.features.common import OS, FORMAT_ELF, Arch, Format, Feature
+from capa.features.address import NO_ADDRESS, FileOffsetAddress, AbsoluteVirtualAddress
+from capa.features.extractors.base_extractor import FeatureExtractor
+
+logger = logging.getLogger(__name__)
+
+
+def extract_file_export_names(elf: ELFFile, **kwargs):
+    for section in elf.iter_sections():
+        if not isinstance(section, SymbolTableSection):
+            continue
+
+        if section["sh_entsize"] == 0:
+            logger.debug("Symbol table '%s' has a sh_entsize of zero!", section.name)
+            continue
+
+        logger.debug("Symbol table '%s' contains %s entries:", section.name, section.num_symbols())
+
+        for symbol in section.iter_symbols():
+            # The following conditions are based on the following article
+            # http://www.m4b.io/elf/export/binary/analysis/2015/05/25/what-is-an-elf-export.html
+            if not symbol.name:
+                continue
+            if symbol.entry.st_info.type not in ["STT_FUNC", "STT_OBJECT", "STT_IFUNC"]:
+                continue
+            if symbol.entry.st_value == 0:
+                continue
+            if symbol.entry.st_shndx == "SHN_UNDEF":
+                continue
+
+            yield Export(symbol.name), AbsoluteVirtualAddress(symbol.entry.st_value)
+
+
+def extract_file_import_names(elf: ELFFile, **kwargs):
+    # Create a dictionary to store symbol names by their index
+    symbol_names = {}
+
+    # Extract symbol names and store them in the dictionary
+    for section in elf.iter_sections():
+        if not isinstance(section, SymbolTableSection):
+            continue
+
+        for _, symbol in enumerate(section.iter_symbols()):
+            # The following conditions are based on the following article
+            # http://www.m4b.io/elf/export/binary/analysis/2015/05/25/what-is-an-elf-export.html
+            if not symbol.name:
+                continue
+            if symbol.entry.st_info.type not in ["STT_FUNC", "STT_OBJECT", "STT_IFUNC"]:
+                continue
+            if symbol.entry.st_value != 0:
+                continue
+            if symbol.entry.st_shndx != "SHN_UNDEF":
+                continue
+            if symbol.entry.st_name == 0:
+                continue
+
+            symbol_names[_] = symbol.name
+
+    for section in elf.iter_sections():
+        if not isinstance(section, RelocationSection):
+            continue
+
+        if section["sh_entsize"] == 0:
+            logger.debug("Symbol table '%s' has a sh_entsize of zero!", section.name)
+            continue
+
+        logger.debug("Symbol table '%s' contains %s entries:", section.name, section.num_relocations())
+
+        for relocation in section.iter_relocations():
+            # Extract the symbol name from the symbol table using the symbol index in the relocation
+            if relocation["r_info_sym"] not in symbol_names:
+                continue
+            yield Import(symbol_names[relocation["r_info_sym"]]), FileOffsetAddress(relocation["r_offset"])
+
+
+def extract_file_section_names(elf: ELFFile, **kwargs):
+    for section in elf.iter_sections():
+        if section.name:
+            yield Section(section.name), AbsoluteVirtualAddress(section.header.sh_addr)
+        elif section.is_null():
+            yield Section("NULL"), AbsoluteVirtualAddress(section.header.sh_addr)
+
+
+def extract_file_strings(buf, **kwargs):
+    yield from capa.features.extractors.common.extract_file_strings(buf)
+
+
+def extract_file_os(elf: ELFFile, buf, **kwargs):
+    # our current approach does not always get an OS value, e.g. for packed samples
+    # for file limitation purposes, we're more lax here
+    try:
+        os_tuple = next(capa.features.extractors.common.extract_os(buf))
+        yield os_tuple
+    except StopIteration:
+        yield OS("unknown"), NO_ADDRESS
+
+
+def extract_file_format(**kwargs):
+    yield Format(FORMAT_ELF), NO_ADDRESS
+
+
+def extract_file_arch(elf: ELFFile, **kwargs):
+    arch = elf.get_machine_arch()
+    if arch == "x86":
+        yield Arch("i386"), NO_ADDRESS
+    elif arch == "x64":
+        yield Arch("amd64"), NO_ADDRESS
+    else:
+        logger.warning("unsupported architecture: %s", arch)
+
+
+def extract_file_features(elf: ELFFile, buf: bytes) -> Iterator[Tuple[Feature, int]]:
+    for file_handler in FILE_HANDLERS:
+        for feature, addr in file_handler(elf=elf, buf=buf):  # type: ignore
+            yield feature, addr
+
+
+FILE_HANDLERS = (
+    extract_file_export_names,
+    extract_file_import_names,
+    extract_file_section_names,
+    extract_file_strings,
+    # no library matching
+    extract_file_format,
+)
+
+
+def extract_global_features(elf: ELFFile, buf: bytes) -> Iterator[Tuple[Feature, int]]:
+    for global_handler in GLOBAL_HANDLERS:
+        for feature, addr in global_handler(elf=elf, buf=buf):  # type: ignore
+            yield feature, addr
+
+
+GLOBAL_HANDLERS = (
+    extract_file_os,
+    extract_file_arch,
+)
+
+
+class ElfFeatureExtractor(FeatureExtractor):
+    def __init__(self, path: Path):
+        super().__init__()
+        self.path: Path = path
+        self.elf = ELFFile(io.BytesIO(path.read_bytes()))
+
+    def get_base_address(self):
+        # virtual address of the first segment with type LOAD
+        for segment in self.elf.iter_segments():
+            if segment.header.p_type == "PT_LOAD":
+                return AbsoluteVirtualAddress(segment.header.p_vaddr)
+
+    def extract_global_features(self):
+        buf = self.path.read_bytes()
+
+        for feature, addr in extract_global_features(self.elf, buf):
+            yield feature, addr
+
+    def extract_file_features(self):
+        buf = self.path.read_bytes()
+
+        for feature, addr in extract_file_features(self.elf, buf):
+            yield feature, addr
+
+    def get_functions(self):
+        raise NotImplementedError("ElfFeatureExtractor can only be used to extract file features")
+
+    def extract_function_features(self, f):
+        raise NotImplementedError("ElfFeatureExtractor can only be used to extract file features")
+
+    def get_basic_blocks(self, f):
+        raise NotImplementedError("ElfFeatureExtractor can only be used to extract file features")
+
+    def extract_basic_block_features(self, f, bb):
+        raise NotImplementedError("ElfFeatureExtractor can only be used to extract file features")
+
+    def get_instructions(self, f, bb):
+        raise NotImplementedError("ElfFeatureExtractor can only be used to extract file features")
+
+    def extract_insn_features(self, f, bb, insn):
+        raise NotImplementedError("ElfFeatureExtractor can only be used to extract file features")
+
+    def is_library_function(self, addr):
+        raise NotImplementedError("ElfFeatureExtractor can only be used to extract file features")
+
+    def get_function_name(self, addr):
+        raise NotImplementedError("ElfFeatureExtractor can only be used to extract file features")

capa/features/extractors/ghidra/basicblock.py

@@ -0,0 +1,152 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+
+import string
+import struct
+from typing import Tuple, Iterator
+
+import ghidra
+from ghidra.program.model.lang import OperandType
+
+import capa.features.extractors.ghidra.helpers
+from capa.features.common import Feature, Characteristic
+from capa.features.address import Address
+from capa.features.basicblock import BasicBlock
+from capa.features.extractors.helpers import MIN_STACKSTRING_LEN
+from capa.features.extractors.base_extractor import BBHandle, FunctionHandle
+
+
+def get_printable_len(op: ghidra.program.model.scalar.Scalar) -> int:
+    """Return string length if all operand bytes are ascii or utf16-le printable"""
+    op_bit_len = op.bitLength()
+    op_byte_len = op_bit_len // 8
+    op_val = op.getValue()
+
+    if op_bit_len == 8:
+        chars = struct.pack("<B", op_val & 0xFF)
+    elif op_bit_len == 16:
+        chars = struct.pack("<H", op_val & 0xFFFF)
+    elif op_bit_len == 32:
+        chars = struct.pack("<I", op_val & 0xFFFFFFFF)
+    elif op_bit_len == 64:
+        chars = struct.pack("<Q", op_val & 0xFFFFFFFFFFFFFFFF)
+    else:
+        raise ValueError(f"Unhandled operand data type 0x{op_bit_len:x}.")
+
+    def is_printable_ascii(chars_: bytes):
+        return all(c < 127 and chr(c) in string.printable for c in chars_)
+
+    def is_printable_utf16le(chars_: bytes):
+        if all(c == 0x00 for c in chars_[1::2]):
+            return is_printable_ascii(chars_[::2])
+
+    if is_printable_ascii(chars):
+        return op_byte_len
+
+    if is_printable_utf16le(chars):
+        return op_byte_len
+
+    return 0
+
+
+def is_mov_imm_to_stack(insn: ghidra.program.database.code.InstructionDB) -> bool:
+    """verify instruction moves immediate onto stack"""
+
+    # Ghidra will Bitwise OR the OperandTypes to assign multiple
+    # i.e., the first operand is a stackvar (dynamically allocated),
+    # and the second is a scalar value (single int/char/float/etc.)
+    mov_its_ops = [(OperandType.ADDRESS | OperandType.DYNAMIC), OperandType.SCALAR]
+    found = False
+
+    # MOV dword ptr [EBP + local_*], 0x65
+    if insn.getMnemonicString().startswith("MOV"):
+        found = all(insn.getOperandType(i) == mov_its_ops[i] for i in range(2))
+
+    return found
+
+
+def bb_contains_stackstring(bb: ghidra.program.model.block.CodeBlock) -> bool:
+    """check basic block for stackstring indicators
+
+    true if basic block contains enough moves of constant bytes to the stack
+    """
+    count = 0
+    for insn in currentProgram().getListing().getInstructions(bb, True):  # type: ignore [name-defined] # noqa: F821
+        if is_mov_imm_to_stack(insn):
+            count += get_printable_len(insn.getScalar(1))
+        if count > MIN_STACKSTRING_LEN:
+            return True
+    return False
+
+
+def _bb_has_tight_loop(bb: ghidra.program.model.block.CodeBlock):
+    """
+    parse tight loops, true if last instruction in basic block branches to bb start
+    """
+    # Reverse Ordered, first InstructionDB
+    last_insn = currentProgram().getListing().getInstructions(bb, False).next()  # type: ignore [name-defined] # noqa: F821
+
+    if last_insn.getFlowType().isJump():
+        return last_insn.getAddress(0) == bb.getMinAddress()
+
+    return False
+
+
+def extract_bb_stackstring(fh: FunctionHandle, bbh: BBHandle) -> Iterator[Tuple[Feature, Address]]:
+    """extract stackstring indicators from basic block"""
+    bb: ghidra.program.model.block.CodeBlock = bbh.inner
+
+    if bb_contains_stackstring(bb):
+        yield Characteristic("stack string"), bbh.address
+
+
+def extract_bb_tight_loop(fh: FunctionHandle, bbh: BBHandle) -> Iterator[Tuple[Feature, Address]]:
+    """check basic block for tight loop indicators"""
+    bb: ghidra.program.model.block.CodeBlock = bbh.inner
+
+    if _bb_has_tight_loop(bb):
+        yield Characteristic("tight loop"), bbh.address
+
+
+BASIC_BLOCK_HANDLERS = (
+    extract_bb_tight_loop,
+    extract_bb_stackstring,
+)
+
+
+def extract_features(fh: FunctionHandle, bbh: BBHandle) -> Iterator[Tuple[Feature, Address]]:
+    """
+    extract features from the given basic block.
+
+    args:
+        bb: the basic block to process.
+
+    yields:
+      Tuple[Feature, int]: the features and their location found in this basic block.
+    """
+    yield BasicBlock(), bbh.address
+    for bb_handler in BASIC_BLOCK_HANDLERS:
+        for feature, addr in bb_handler(fh, bbh):
+            yield feature, addr
+
+
+def main():
+    features = []
+    from capa.features.extractors.ghidra.extractor import GhidraFeatureExtractor
+
+    for fh in GhidraFeatureExtractor().get_functions():
+        for bbh in capa.features.extractors.ghidra.helpers.get_function_blocks(fh):
+            features.extend(list(extract_features(fh, bbh)))
+
+    import pprint
+
+    pprint.pprint(features)  # noqa: T203
+
+
+if __name__ == "__main__":
+    main()

capa/features/extractors/ghidra/extractor.py

@@ -0,0 +1,75 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+from typing import List, Tuple, Iterator
+
+import capa.features.extractors.ghidra.file
+import capa.features.extractors.ghidra.insn
+import capa.features.extractors.ghidra.global_
+import capa.features.extractors.ghidra.function
+import capa.features.extractors.ghidra.basicblock
+from capa.features.common import Feature
+from capa.features.address import Address, AbsoluteVirtualAddress
+from capa.features.extractors.base_extractor import BBHandle, InsnHandle, FunctionHandle, FeatureExtractor
+
+
+class GhidraFeatureExtractor(FeatureExtractor):
+    def __init__(self):
+        super().__init__()
+        import capa.features.extractors.ghidra.helpers as ghidra_helpers
+
+        self.global_features: List[Tuple[Feature, Address]] = []
+        self.global_features.extend(capa.features.extractors.ghidra.file.extract_file_format())
+        self.global_features.extend(capa.features.extractors.ghidra.global_.extract_os())
+        self.global_features.extend(capa.features.extractors.ghidra.global_.extract_arch())
+        self.imports = ghidra_helpers.get_file_imports()
+        self.externs = ghidra_helpers.get_file_externs()
+        self.fakes = ghidra_helpers.map_fake_import_addrs()
+
+    def get_base_address(self):
+        return AbsoluteVirtualAddress(currentProgram().getImageBase().getOffset())  # type: ignore [name-defined] # noqa: F821
+
+    def extract_global_features(self):
+        yield from self.global_features
+
+    def extract_file_features(self):
+        yield from capa.features.extractors.ghidra.file.extract_features()
+
+    def get_functions(self) -> Iterator[FunctionHandle]:
+        import capa.features.extractors.ghidra.helpers as ghidra_helpers
+
+        for fhandle in ghidra_helpers.get_function_symbols():
+            fh: FunctionHandle = FunctionHandle(
+                address=AbsoluteVirtualAddress(fhandle.getEntryPoint().getOffset()),
+                inner=fhandle,
+                ctx={"imports_cache": self.imports, "externs_cache": self.externs, "fakes_cache": self.fakes},
+            )
+            yield fh
+
+    @staticmethod
+    def get_function(addr: int) -> FunctionHandle:
+        func = getFunctionContaining(toAddr(addr))  # type: ignore [name-defined] # noqa: F821
+        return FunctionHandle(address=AbsoluteVirtualAddress(func.getEntryPoint().getOffset()), inner=func)
+
+    def extract_function_features(self, fh: FunctionHandle) -> Iterator[Tuple[Feature, Address]]:
+        yield from capa.features.extractors.ghidra.function.extract_features(fh)
+
+    def get_basic_blocks(self, fh: FunctionHandle) -> Iterator[BBHandle]:
+        import capa.features.extractors.ghidra.helpers as ghidra_helpers
+
+        yield from ghidra_helpers.get_function_blocks(fh)
+
+    def extract_basic_block_features(self, fh: FunctionHandle, bbh: BBHandle) -> Iterator[Tuple[Feature, Address]]:
+        yield from capa.features.extractors.ghidra.basicblock.extract_features(fh, bbh)
+
+    def get_instructions(self, fh: FunctionHandle, bbh: BBHandle) -> Iterator[InsnHandle]:
+        import capa.features.extractors.ghidra.helpers as ghidra_helpers
+
+        yield from ghidra_helpers.get_insn_in_range(bbh)
+
+    def extract_insn_features(self, fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle):
+        yield from capa.features.extractors.ghidra.insn.extract_features(fh, bbh, ih)

capa/features/extractors/ghidra/file.py

@@ -0,0 +1,202 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import re
+import struct
+from typing import List, Tuple, Iterator
+
+from ghidra.program.model.symbol import SourceType, SymbolType
+
+import capa.features.extractors.common
+import capa.features.extractors.helpers
+import capa.features.extractors.strings
+import capa.features.extractors.ghidra.helpers
+from capa.features.file import Export, Import, Section, FunctionName
+from capa.features.common import FORMAT_PE, FORMAT_ELF, Format, String, Feature, Characteristic
+from capa.features.address import NO_ADDRESS, Address, FileOffsetAddress, AbsoluteVirtualAddress
+
+MAX_OFFSET_PE_AFTER_MZ = 0x200
+
+
+def find_embedded_pe(block_bytez: bytes, mz_xor: List[Tuple[bytes, bytes, int]]) -> Iterator[Tuple[int, int]]:
+    """check segment for embedded PE
+
+    adapted for Ghidra from:
+    https://github.com/vivisect/vivisect/blob/91e8419a861f4977https://github.com/vivisect/vivisect/blob/91e8419a861f49779f18316f155311967e696836/PE/carve.py#L259f18316f155311967e696836/PE/carve.py#L25
+    """
+    todo = []
+
+    for mzx, pex, i in mz_xor:
+        for match in re.finditer(re.escape(mzx), block_bytez):
+            todo.append((match.start(), mzx, pex, i))
+
+    seg_max = len(block_bytez)  # type: ignore [name-defined] # noqa: F821
+    while len(todo):
+        off, mzx, pex, i = todo.pop()
+
+        # MZ header has one field we will check e_lfanew is at 0x3c
+        e_lfanew = off + 0x3C
+
+        if seg_max < e_lfanew + 4:
+            continue
+
+        e_lfanew_bytes = block_bytez[e_lfanew : e_lfanew + 4]
+        newoff = struct.unpack("<I", capa.features.extractors.helpers.xor_static(e_lfanew_bytes, i))[0]
+
+        # assume XOR'd "PE" bytes exist within threshold
+        if newoff > MAX_OFFSET_PE_AFTER_MZ:
+            continue
+
+        peoff = off + newoff
+        if seg_max < peoff + 2:
+            continue
+
+        pe_bytes = block_bytez[peoff : peoff + 2]
+        if pe_bytes == pex:
+            yield off, i
+
+
+def extract_file_embedded_pe() -> Iterator[Tuple[Feature, Address]]:
+    """extract embedded PE features"""
+
+    # pre-compute XOR pairs
+    mz_xor: List[Tuple[bytes, bytes, int]] = [
+        (
+            capa.features.extractors.helpers.xor_static(b"MZ", i),
+            capa.features.extractors.helpers.xor_static(b"PE", i),
+            i,
+        )
+        for i in range(256)
+    ]
+
+    for block in currentProgram().getMemory().getBlocks():  # type: ignore [name-defined] # noqa: F821
+        if not all((block.isLoaded(), block.isInitialized(), "Headers" not in block.getName())):
+            continue
+
+        for off, _ in find_embedded_pe(capa.features.extractors.ghidra.helpers.get_block_bytes(block), mz_xor):
+            # add offset back to block start
+            ea: int = block.getStart().add(off).getOffset()
+
+            yield Characteristic("embedded pe"), FileOffsetAddress(ea)
+
+
+def extract_file_export_names() -> Iterator[Tuple[Feature, Address]]:
+    """extract function exports"""
+    st = currentProgram().getSymbolTable()  # type: ignore [name-defined] # noqa: F821
+    for addr in st.getExternalEntryPointIterator():
+        yield Export(st.getPrimarySymbol(addr).getName()), AbsoluteVirtualAddress(addr.getOffset())
+
+
+def extract_file_import_names() -> Iterator[Tuple[Feature, Address]]:
+    """extract function imports
+
+    1. imports by ordinal:
+     - modulename.#ordinal
+
+    2. imports by name, results in two features to support importname-only
+       matching:
+     - modulename.importname
+     - importname
+    """
+
+    for f in currentProgram().getFunctionManager().getExternalFunctions():  # type: ignore [name-defined] # noqa: F821
+        for r in f.getSymbol().getReferences():
+            if r.getReferenceType().isData():
+                addr = r.getFromAddress().getOffset()  # gets pointer to fake external addr
+
+        fstr = f.toString().split("::")  # format: MODULE.dll::import / MODULE::Ordinal_*
+        if "Ordinal_" in fstr[1]:
+            fstr[1] = f"#{fstr[1].split('_')[1]}"
+
+        for name in capa.features.extractors.helpers.generate_symbols(fstr[0][:-4], fstr[1]):
+            yield Import(name), AbsoluteVirtualAddress(addr)
+
+
+def extract_file_section_names() -> Iterator[Tuple[Feature, Address]]:
+    """extract section names"""
+
+    for block in currentProgram().getMemory().getBlocks():  # type: ignore [name-defined] # noqa: F821
+        yield Section(block.getName()), AbsoluteVirtualAddress(block.getStart().getOffset())
+
+
+def extract_file_strings() -> Iterator[Tuple[Feature, Address]]:
+    """extract ASCII and UTF-16 LE strings"""
+
+    for block in currentProgram().getMemory().getBlocks():  # type: ignore [name-defined] # noqa: F821
+        if block.isInitialized():
+            p_bytes = capa.features.extractors.ghidra.helpers.get_block_bytes(block)
+
+        for s in capa.features.extractors.strings.extract_ascii_strings(p_bytes):
+            offset = block.getStart().getOffset() + s.offset
+            yield String(s.s), FileOffsetAddress(offset)
+
+        for s in capa.features.extractors.strings.extract_unicode_strings(p_bytes):
+            offset = block.getStart().getOffset() + s.offset
+            yield String(s.s), FileOffsetAddress(offset)
+
+
+def extract_file_function_names() -> Iterator[Tuple[Feature, Address]]:
+    """
+    extract the names of statically-linked library functions.
+    """
+
+    for sym in currentProgram().getSymbolTable().getAllSymbols(True):  # type: ignore [name-defined] # noqa: F821
+        # .isExternal() misses more than this config for the function symbols
+        if sym.getSymbolType() == SymbolType.FUNCTION and sym.getSource() == SourceType.ANALYSIS and sym.isGlobal():
+            name = sym.getName()  # starts to resolve names based on Ghidra's FidDB
+            if name.startswith("FID_conflict:"):  # format: FID_conflict:<function-name>
+                name = name[13:]
+            addr = AbsoluteVirtualAddress(sym.getAddress().getOffset())
+            yield FunctionName(name), addr
+            if name.startswith("_"):
+                # some linkers may prefix linked routines with a `_` to avoid name collisions.
+                # extract features for both the mangled and un-mangled representations.
+                # e.g. `_fwrite` -> `fwrite`
+                # see: https://stackoverflow.com/a/2628384/87207
+                yield FunctionName(name[1:]), addr
+
+
+def extract_file_format() -> Iterator[Tuple[Feature, Address]]:
+    ef = currentProgram().getExecutableFormat()  # type: ignore [name-defined] # noqa: F821
+    if "PE" in ef:
+        yield Format(FORMAT_PE), NO_ADDRESS
+    elif "ELF" in ef:
+        yield Format(FORMAT_ELF), NO_ADDRESS
+    elif "Raw" in ef:
+        # no file type to return when processing a binary file, but we want to continue processing
+        return
+    else:
+        raise NotImplementedError(f"unexpected file format: {ef}")
+
+
+def extract_features() -> Iterator[Tuple[Feature, Address]]:
+    """extract file features"""
+    for file_handler in FILE_HANDLERS:
+        for feature, addr in file_handler():
+            yield feature, addr
+
+
+FILE_HANDLERS = (
+    extract_file_embedded_pe,
+    extract_file_export_names,
+    extract_file_import_names,
+    extract_file_section_names,
+    extract_file_strings,
+    extract_file_function_names,
+    extract_file_format,
+)
+
+
+def main():
+    """ """
+    import pprint
+
+    pprint.pprint(list(extract_features()))  # noqa: T203
+
+
+if __name__ == "__main__":
+    main()

capa/features/extractors/ghidra/function.py

@@ -0,0 +1,73 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+from typing import Tuple, Iterator
+
+import ghidra
+from ghidra.program.model.block import BasicBlockModel, SimpleBlockIterator
+
+import capa.features.extractors.ghidra.helpers
+from capa.features.common import Feature, Characteristic
+from capa.features.address import Address, AbsoluteVirtualAddress
+from capa.features.extractors import loops
+from capa.features.extractors.base_extractor import FunctionHandle
+
+
+def extract_function_calls_to(fh: FunctionHandle):
+    """extract callers to a function"""
+    f: ghidra.program.database.function.FunctionDB = fh.inner
+    for ref in f.getSymbol().getReferences():
+        if ref.getReferenceType().isCall():
+            yield Characteristic("calls to"), AbsoluteVirtualAddress(ref.getFromAddress().getOffset())
+
+
+def extract_function_loop(fh: FunctionHandle):
+    f: ghidra.program.database.function.FunctionDB = fh.inner
+
+    edges = []
+    for block in SimpleBlockIterator(BasicBlockModel(currentProgram()), f.getBody(), monitor()):  # type: ignore [name-defined] # noqa: F821
+        dests = block.getDestinations(monitor())  # type: ignore [name-defined] # noqa: F821
+        s_addrs = block.getStartAddresses()
+
+        while dests.hasNext():  # For loop throws Python TypeError
+            for addr in s_addrs:
+                edges.append((addr.getOffset(), dests.next().getDestinationAddress().getOffset()))
+
+    if loops.has_loop(edges):
+        yield Characteristic("loop"), AbsoluteVirtualAddress(f.getEntryPoint().getOffset())
+
+
+def extract_recursive_call(fh: FunctionHandle):
+    f: ghidra.program.database.function.FunctionDB = fh.inner
+
+    for func in f.getCalledFunctions(monitor()):  # type: ignore [name-defined] # noqa: F821
+        if func.getEntryPoint().getOffset() == f.getEntryPoint().getOffset():
+            yield Characteristic("recursive call"), AbsoluteVirtualAddress(f.getEntryPoint().getOffset())
+
+
+def extract_features(fh: FunctionHandle) -> Iterator[Tuple[Feature, Address]]:
+    for func_handler in FUNCTION_HANDLERS:
+        for feature, addr in func_handler(fh):
+            yield feature, addr
+
+
+FUNCTION_HANDLERS = (extract_function_calls_to, extract_function_loop, extract_recursive_call)
+
+
+def main():
+    """ """
+    features = []
+    for fhandle in capa.features.extractors.ghidra.helpers.get_function_symbols():
+        features.extend(list(extract_features(fhandle)))
+
+    import pprint
+
+    pprint.pprint(features)  # noqa: T203
+
+
+if __name__ == "__main__":
+    main()

capa/features/extractors/ghidra/global_.py

@@ -0,0 +1,67 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import logging
+import contextlib
+from typing import Tuple, Iterator
+
+import capa.ghidra.helpers
+import capa.features.extractors.elf
+import capa.features.extractors.ghidra.helpers
+from capa.features.common import OS, ARCH_I386, ARCH_AMD64, OS_WINDOWS, Arch, Feature
+from capa.features.address import NO_ADDRESS, Address
+
+logger = logging.getLogger(__name__)
+
+
+def extract_os() -> Iterator[Tuple[Feature, Address]]:
+    format_name: str = currentProgram().getExecutableFormat()  # type: ignore [name-defined] # noqa: F821
+
+    if "PE" in format_name:
+        yield OS(OS_WINDOWS), NO_ADDRESS
+
+    elif "ELF" in format_name:
+        with contextlib.closing(capa.ghidra.helpers.GHIDRAIO()) as f:
+            os = capa.features.extractors.elf.detect_elf_os(f)
+
+        yield OS(os), NO_ADDRESS
+
+    else:
+        # we likely end up here:
+        #  1. handling shellcode, or
+        #  2. handling a new file format (e.g. macho)
+        #
+        # for (1) we can't do much - its shellcode and all bets are off.
+        # we could maybe accept a further CLI argument to specify the OS,
+        # but i think this would be rarely used.
+        # rules that rely on OS conditions will fail to match on shellcode.
+        #
+        # for (2), this logic will need to be updated as the format is implemented.
+        logger.debug("unsupported file format: %s, will not guess OS", format_name)
+        return
+
+
+def extract_arch() -> Iterator[Tuple[Feature, Address]]:
+    lang_id = currentProgram().getMetadata().get("Language ID")  # type: ignore [name-defined] # noqa: F821
+
+    if "x86" in lang_id and "64" in lang_id:
+        yield Arch(ARCH_AMD64), NO_ADDRESS
+
+    elif "x86" in lang_id and "32" in lang_id:
+        yield Arch(ARCH_I386), NO_ADDRESS
+
+    elif "x86" not in lang_id:
+        logger.debug("unsupported architecture: non-32-bit nor non-64-bit intel")
+        return
+
+    else:
+        # we likely end up here:
+        #  1. handling a new architecture (e.g. aarch64)
+        #
+        # for (1), this logic will need to be updated as the format is implemented.
+        logger.debug("unsupported architecture: %s", lang_id)
+        return

capa/features/extractors/ghidra/helpers.py

@@ -0,0 +1,277 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+from typing import Dict, List, Iterator
+
+import ghidra
+import java.lang
+from ghidra.program.model.lang import OperandType
+from ghidra.program.model.block import BasicBlockModel, SimpleBlockIterator
+from ghidra.program.model.symbol import SourceType, SymbolType
+from ghidra.program.model.address import AddressSpace
+
+import capa.features.extractors.helpers
+from capa.features.common import THUNK_CHAIN_DEPTH_DELTA
+from capa.features.address import AbsoluteVirtualAddress
+from capa.features.extractors.base_extractor import BBHandle, InsnHandle, FunctionHandle
+
+
+def ints_to_bytes(bytez: List[int]) -> bytes:
+    """convert Java signed ints to Python bytes
+
+    args:
+        bytez: list of Java signed ints
+    """
+    return bytes([b & 0xFF for b in bytez])
+
+
+def find_byte_sequence(addr: ghidra.program.model.address.Address, seq: bytes) -> Iterator[int]:
+    """yield all ea of a given byte sequence
+
+    args:
+        addr: start address
+        seq: bytes to search e.g. b"\x01\x03"
+    """
+    seqstr = "".join([f"\\x{b:02x}" for b in seq])
+    eas = findBytes(addr, seqstr, java.lang.Integer.MAX_VALUE, 1)  # type: ignore [name-defined] # noqa: F821
+
+    yield from eas
+
+
+def get_bytes(addr: ghidra.program.model.address.Address, length: int) -> bytes:
+    """yield length bytes at addr
+
+    args:
+        addr: Address to begin pull from
+        length: length of bytes to pull
+    """
+    try:
+        return ints_to_bytes(getBytes(addr, length))  # type: ignore [name-defined] # noqa: F821
+    except RuntimeError:
+        return b""
+
+
+def get_block_bytes(block: ghidra.program.model.mem.MemoryBlock) -> bytes:
+    """yield all bytes in a given block
+
+    args:
+        block: MemoryBlock to pull from
+    """
+    return get_bytes(block.getStart(), block.getSize())
+
+
+def get_function_symbols():
+    """yield all non-external function symbols"""
+    yield from currentProgram().getFunctionManager().getFunctionsNoStubs(True)  # type: ignore [name-defined] # noqa: F821
+
+
+def get_function_blocks(fh: FunctionHandle) -> Iterator[BBHandle]:
+    """yield BBHandle for each bb in a given function"""
+
+    func: ghidra.program.database.function.FunctionDB = fh.inner
+    for bb in SimpleBlockIterator(BasicBlockModel(currentProgram()), func.getBody(), monitor()):  # type: ignore [name-defined] # noqa: F821
+        yield BBHandle(address=AbsoluteVirtualAddress(bb.getMinAddress().getOffset()), inner=bb)
+
+
+def get_insn_in_range(bbh: BBHandle) -> Iterator[InsnHandle]:
+    """yield InshHandle for each insn in a given basicblock"""
+    for insn in currentProgram().getListing().getInstructions(bbh.inner, True):  # type: ignore [name-defined] # noqa: F821
+        yield InsnHandle(address=AbsoluteVirtualAddress(insn.getAddress().getOffset()), inner=insn)
+
+
+def get_file_imports() -> Dict[int, List[str]]:
+    """get all import names & addrs"""
+
+    import_dict: Dict[int, List[str]] = {}
+
+    for f in currentProgram().getFunctionManager().getExternalFunctions():  # type: ignore [name-defined] # noqa: F821
+        for r in f.getSymbol().getReferences():
+            if r.getReferenceType().isData():
+                addr = r.getFromAddress().getOffset()  # gets pointer to fake external addr
+
+        ex_loc = f.getExternalLocation().getAddress()  # map external locations as well (offset into module files)
+
+        fstr = f.toString().split("::")  # format: MODULE.dll::import / MODULE::Ordinal_* / <EXTERNAL>::import
+        if "Ordinal_" in fstr[1]:
+            fstr[1] = f"#{fstr[1].split('_')[1]}"
+
+        # <EXTERNAL> mostly shows up in ELF files, otherwise, strip '.dll' w/ [:-4]
+        fstr[0] = "*" if "<EXTERNAL>" in fstr[0] else fstr[0][:-4]
+
+        for name in capa.features.extractors.helpers.generate_symbols(fstr[0], fstr[1]):
+            import_dict.setdefault(addr, []).append(name)
+            if ex_loc:
+                import_dict.setdefault(ex_loc.getOffset(), []).append(name)
+
+    return import_dict
+
+
+def get_file_externs() -> Dict[int, List[str]]:
+    """
+    Gets function names & addresses of statically-linked library functions
+
+    Ghidra's external namespace is mostly reserved for dynamically-linked
+    imports. Statically-linked functions are part of the global namespace.
+    Filtering on the type, source, and namespace of the symbols yield more
+    statically-linked library functions.
+
+    Example: (PMA Lab 16-01.exe_) 7faafc7e4a5c736ebfee6abbbc812d80:0x407490
+    - __aulldiv
+    - Note: See Symbol Table labels
+    """
+
+    extern_dict: Dict[int, List[str]] = {}
+
+    for sym in currentProgram().getSymbolTable().getAllSymbols(True):  # type: ignore [name-defined] # noqa: F821
+        # .isExternal() misses more than this config for the function symbols
+        if sym.getSymbolType() == SymbolType.FUNCTION and sym.getSource() == SourceType.ANALYSIS and sym.isGlobal():
+            name = sym.getName()  # starts to resolve names based on Ghidra's FidDB
+            if name.startswith("FID_conflict:"):  # format: FID_conflict:<function-name>
+                name = name[13:]
+            extern_dict.setdefault(sym.getAddress().getOffset(), []).append(name)
+            if name.startswith("_"):
+                # some linkers may prefix linked routines with a `_` to avoid name collisions.
+                # extract features for both the mangled and un-mangled representations.
+                # e.g. `_fwrite` -> `fwrite`
+                # see: https://stackoverflow.com/a/2628384/87207
+                extern_dict.setdefault(sym.getAddress().getOffset(), []).append(name[1:])
+
+    return extern_dict
+
+
+def map_fake_import_addrs() -> Dict[int, List[int]]:
+    """
+    Map ghidra's fake import entrypoints to their
+    real addresses
+
+    Helps as many Ghidra Scripting API calls end up returning
+    these external (fake) addresses.
+
+    Undocumented but intended Ghidra behavior:
+     - Import entryPoint fields are stored in the 'EXTERNAL:' AddressSpace.
+       'getEntryPoint()' returns the entryPoint field, which is an offset
+       from the beginning of the assigned AddressSpace. In the case of externals,
+       they start from 1 and increment.
+    https://github.com/NationalSecurityAgency/ghidra/blob/26d4bd9104809747c21f2528cab8aba9aef9acd5/Ghidra/Features/Base/src/test.slow/java/ghidra/program/database/function/ExternalFunctionDBTest.java#L90
+
+    Example: (mimikatz.exe_) 5f66b82558ca92e54e77f216ef4c066c:0x473090
+    - 0x473090 -> PTR_CreateServiceW_00473090
+    - 'EXTERNAL:00000025' -> External Address (ghidra.program.model.address.SpecialAddress)
+    """
+    fake_dict: Dict[int, List[int]] = {}
+
+    for f in currentProgram().getFunctionManager().getExternalFunctions():  # type: ignore [name-defined] # noqa: F821
+        for r in f.getSymbol().getReferences():
+            if r.getReferenceType().isData():
+                fake_dict.setdefault(f.getEntryPoint().getOffset(), []).append(r.getFromAddress().getOffset())
+
+    return fake_dict
+
+
+def check_addr_for_api(
+    addr: ghidra.program.model.address.Address,
+    fakes: Dict[int, List[int]],
+    imports: Dict[int, List[str]],
+    externs: Dict[int, List[str]],
+) -> bool:
+    offset = addr.getOffset()
+
+    fake = fakes.get(offset)
+    if fake:
+        return True
+
+    imp = imports.get(offset)
+    if imp:
+        return True
+
+    extern = externs.get(offset)
+    if extern:
+        return True
+
+    return False
+
+
+def is_call_or_jmp(insn: ghidra.program.database.code.InstructionDB) -> bool:
+    return any(mnem in insn.getMnemonicString() for mnem in ["CALL", "J"])  # JMP, JNE, JNZ, etc
+
+
+def is_sp_modified(insn: ghidra.program.database.code.InstructionDB) -> bool:
+    for i in range(insn.getNumOperands()):
+        if insn.getOperandType(i) == OperandType.REGISTER:
+            return "SP" in insn.getRegister(i).getName() and insn.getOperandRefType(i).isWrite()
+    return False
+
+
+def is_stack_referenced(insn: ghidra.program.database.code.InstructionDB) -> bool:
+    """generic catch-all for stack references"""
+    for i in range(insn.getNumOperands()):
+        if insn.getOperandType(i) == OperandType.REGISTER:
+            if "BP" in insn.getRegister(i).getName():
+                return True
+            else:
+                continue
+
+    return any(ref.isStackReference() for ref in insn.getReferencesFrom())
+
+
+def is_zxor(insn: ghidra.program.database.code.InstructionDB) -> bool:
+    # assume XOR insn
+    # XOR's against the same operand zero out
+    ops = []
+    operands = []
+    for i in range(insn.getNumOperands()):
+        ops.append(insn.getOpObjects(i))
+
+    # Operands stored in a 2D array
+    for j in range(len(ops)):
+        for k in range(len(ops[j])):
+            operands.append(ops[j][k])
+
+    return all(n == operands[0] for n in operands)
+
+
+def handle_thunk(addr: ghidra.program.model.address.Address):
+    """Follow thunk chains down to a reasonable depth"""
+    ref = addr
+    for _ in range(THUNK_CHAIN_DEPTH_DELTA):
+        thunk_jmp = getInstructionAt(ref)  # type: ignore [name-defined] # noqa: F821
+        if thunk_jmp and is_call_or_jmp(thunk_jmp):
+            if OperandType.isAddress(thunk_jmp.getOperandType(0)):
+                ref = thunk_jmp.getAddress(0)
+        else:
+            thunk_dat = getDataContaining(ref)  # type: ignore [name-defined] # noqa: F821
+            if thunk_dat and thunk_dat.isDefined() and thunk_dat.isPointer():
+                ref = thunk_dat.getValue()
+                break  # end of thunk chain reached
+    return ref
+
+
+def dereference_ptr(insn: ghidra.program.database.code.InstructionDB):
+    addr_code = OperandType.ADDRESS | OperandType.CODE
+    to_deref = insn.getAddress(0)
+    dat = getDataContaining(to_deref)  # type: ignore [name-defined] # noqa: F821
+
+    if insn.getOperandType(0) == addr_code:
+        thfunc = getFunctionContaining(to_deref)  # type: ignore [name-defined] # noqa: F821
+        if thfunc and thfunc.isThunk():
+            return handle_thunk(to_deref)
+        else:
+            # if it doesn't poin to a thunk, it's usually a jmp to a label
+            return to_deref
+    if not dat:
+        return to_deref
+    if dat.isDefined() and dat.isPointer():
+        addr = dat.getValue()
+        # now we need to check the addr space to see if it is truly resolvable
+        # ghidra sometimes likes to hand us direct RAM addrs, which typically point
+        # to api calls that we can't actually resolve as such
+        if addr.getAddressSpace().getType() == AddressSpace.TYPE_RAM:
+            return to_deref
+        else:
+            return addr
+    else:
+        return to_deref

capa/features/extractors/ghidra/insn.py

@@ -0,0 +1,521 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+from typing import Any, Dict, Tuple, Iterator
+
+import ghidra
+from ghidra.program.model.lang import OperandType
+from ghidra.program.model.block import SimpleBlockModel
+
+import capa.features.extractors.helpers
+import capa.features.extractors.ghidra.helpers
+from capa.features.insn import API, MAX_STRUCTURE_SIZE, Number, Offset, Mnemonic, OperandNumber, OperandOffset
+from capa.features.common import MAX_BYTES_FEATURE_SIZE, Bytes, String, Feature, Characteristic
+from capa.features.address import Address, AbsoluteVirtualAddress
+from capa.features.extractors.base_extractor import BBHandle, InsnHandle, FunctionHandle
+
+# security cookie checks may perform non-zeroing XORs, these are expected within a certain
+# byte range within the first and returning basic blocks, this helps to reduce FP features
+SECURITY_COOKIE_BYTES_DELTA = 0x40
+
+
+def get_imports(ctx: Dict[str, Any]) -> Dict[int, Any]:
+    """Populate the import cache for this context"""
+    if "imports_cache" not in ctx:
+        ctx["imports_cache"] = capa.features.extractors.ghidra.helpers.get_file_imports()
+    return ctx["imports_cache"]
+
+
+def get_externs(ctx: Dict[str, Any]) -> Dict[int, Any]:
+    """Populate the externs cache for this context"""
+    if "externs_cache" not in ctx:
+        ctx["externs_cache"] = capa.features.extractors.ghidra.helpers.get_file_externs()
+    return ctx["externs_cache"]
+
+
+def get_fakes(ctx: Dict[str, Any]) -> Dict[int, Any]:
+    """Populate the fake import addrs cache for this context"""
+    if "fakes_cache" not in ctx:
+        ctx["fakes_cache"] = capa.features.extractors.ghidra.helpers.map_fake_import_addrs()
+    return ctx["fakes_cache"]
+
+
+def check_for_api_call(
+    insn, externs: Dict[int, Any], fakes: Dict[int, Any], imports: Dict[int, Any], imp_or_ex: bool
+) -> Iterator[Any]:
+    """check instruction for API call
+
+    params:
+        externs - external library functions cache
+        fakes - mapped fake import addresses cache
+        imports - imported functions cache
+        imp_or_ex - flag to check imports or externs
+
+    yields:
+        matched api calls
+    """
+    info = ()
+    funcs = imports if imp_or_ex else externs
+
+    # assume only CALLs or JMPs are passed
+    ref_type = insn.getOperandType(0)
+    addr_data = OperandType.ADDRESS | OperandType.DATA  # needs dereferencing
+    addr_code = OperandType.ADDRESS | OperandType.CODE  # needs dereferencing
+
+    if OperandType.isRegister(ref_type):
+        if OperandType.isAddress(ref_type):
+            # If it's an address in a register, check the mapped fake addrs
+            # since they're dereferenced to their fake addrs
+            op_ref = insn.getAddress(0).getOffset()
+            ref = fakes.get(op_ref)  # obtain the real addr
+            if not ref:
+                return
+        else:
+            return
+    elif ref_type in (addr_data, addr_code) or (OperandType.isIndirect(ref_type) and OperandType.isAddress(ref_type)):
+        # we must dereference and check if the addr is a pointer to an api function
+        addr_ref = capa.features.extractors.ghidra.helpers.dereference_ptr(insn)
+        if not capa.features.extractors.ghidra.helpers.check_addr_for_api(addr_ref, fakes, imports, externs):
+            return
+        ref = addr_ref.getOffset()
+    elif ref_type == OperandType.DYNAMIC | OperandType.ADDRESS or ref_type == OperandType.DYNAMIC:
+        return  # cannot resolve dynamics statically
+    else:
+        # pure address does not need to get dereferenced/ handled
+        addr_ref = insn.getAddress(0)
+        if not addr_ref:
+            # If it returned null, it was an indirect
+            # that had no address reference.
+            # This check is faster than checking for (indirect and not address)
+            return
+        if not capa.features.extractors.ghidra.helpers.check_addr_for_api(addr_ref, fakes, imports, externs):
+            return
+        ref = addr_ref.getOffset()
+
+    if isinstance(ref, list):  # ref from REG | ADDR
+        for r in ref:
+            info = funcs.get(r)  # type: ignore
+            if info:
+                yield info
+    else:
+        info = funcs.get(ref)  # type: ignore
+        if info:
+            yield info
+
+
+def extract_insn_api_features(fh: FunctionHandle, bb: BBHandle, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
+    insn: ghidra.program.database.code.InstructionDB = ih.inner
+
+    if not capa.features.extractors.ghidra.helpers.is_call_or_jmp(insn):
+        return
+
+    externs = get_externs(fh.ctx)
+    fakes = get_fakes(fh.ctx)
+    imports = get_imports(fh.ctx)
+
+    # check calls to imported functions
+    for api in check_for_api_call(insn, externs, fakes, imports, True):
+        for imp in api:
+            yield API(imp), ih.address
+
+    # check calls to extern functions
+    for api in check_for_api_call(insn, externs, fakes, imports, False):
+        for ext in api:
+            yield API(ext), ih.address
+
+
+def extract_insn_number_features(fh: FunctionHandle, bb: BBHandle, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
+    """
+    parse instruction number features
+    example:
+        push    3136B0h         ; dwControlCode
+    """
+    insn: ghidra.program.database.code.InstructionDB = ih.inner
+
+    if insn.getMnemonicString().startswith("RET"):
+        # skip things like:
+        #   .text:0042250E retn 8
+        return
+
+    if capa.features.extractors.ghidra.helpers.is_sp_modified(insn):
+        # skip things like:
+        #   .text:00401145 add esp, 0Ch
+        return
+
+    for i in range(insn.getNumOperands()):
+        # Exceptions for LEA insn:
+        # invalid operand encoding, considered numbers instead of offsets
+        # see: mimikatz.exe_:0x4018C0
+        if insn.getOperandType(i) == OperandType.DYNAMIC and insn.getMnemonicString().startswith("LEA"):
+            # Additional check, avoid yielding "wide" values (ex. mimikatz.exe:0x471EE6 LEA EBX, [ECX + EAX*0x4])
+            op_objs = insn.getOpObjects(i)
+            if len(op_objs) == 3:  # ECX, EAX, 0x4
+                continue
+
+            if isinstance(op_objs[-1], ghidra.program.model.scalar.Scalar):
+                const = op_objs[-1].getUnsignedValue()
+                addr = ih.address
+
+                yield Number(const), addr
+                yield OperandNumber(i, const), addr
+        elif not OperandType.isScalar(insn.getOperandType(i)):
+            # skip things like:
+            #   references, void types
+            continue
+        else:
+            const = insn.getScalar(i).getUnsignedValue()
+            addr = ih.address
+
+            yield Number(const), addr
+            yield OperandNumber(i, const), addr
+
+            if insn.getMnemonicString().startswith("ADD") and 0 < const < MAX_STRUCTURE_SIZE:
+                # for pattern like:
+                #
+                #     add eax, 0x10
+                #
+                # assume 0x10 is also an offset (imagine eax is a pointer).
+                yield Offset(const), addr
+                yield OperandOffset(i, const), addr
+
+
+def extract_insn_offset_features(fh: FunctionHandle, bb: BBHandle, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
+    """
+    parse instruction structure offset features
+
+    example:
+        .text:0040112F cmp [esi+4], ebx
+    """
+    insn: ghidra.program.database.code.InstructionDB = ih.inner
+
+    if insn.getMnemonicString().startswith("LEA"):
+        return
+
+    # ignore any stack references
+    if not capa.features.extractors.ghidra.helpers.is_stack_referenced(insn):
+        # Ghidra stores operands in 2D arrays if they contain offsets
+        for i in range(insn.getNumOperands()):
+            if insn.getOperandType(i) == OperandType.DYNAMIC:  # e.g. [esi + 4]
+                # manual extraction, since the default api calls only work on the 1st dimension of the array
+                op_objs = insn.getOpObjects(i)
+                if isinstance(op_objs[-1], ghidra.program.model.scalar.Scalar):
+                    op_off = op_objs[-1].getValue()
+                    yield Offset(op_off), ih.address
+                    yield OperandOffset(i, op_off), ih.address
+                else:
+                    yield Offset(0), ih.address
+                    yield OperandOffset(i, 0), ih.address
+
+
+def extract_insn_bytes_features(fh: FunctionHandle, bb: BBHandle, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
+    """
+    parse referenced byte sequences
+    example:
+        push    offset iid_004118d4_IShellLinkA ; riid
+    """
+    insn: ghidra.program.database.code.InstructionDB = ih.inner
+
+    if capa.features.extractors.ghidra.helpers.is_call_or_jmp(insn):
+        return
+
+    ref = insn.getAddress()  # init to insn addr
+    for i in range(insn.getNumOperands()):
+        if OperandType.isAddress(insn.getOperandType(i)):
+            ref = insn.getAddress(i)  # pulls pointer if there is one
+
+    if ref != insn.getAddress():  # bail out if there's no pointer
+        ghidra_dat = getDataAt(ref)  # type: ignore [name-defined] # noqa: F821
+        if (
+            ghidra_dat and not ghidra_dat.hasStringValue() and not ghidra_dat.isPointer()
+        ):  # avoid if the data itself is a pointer
+            extracted_bytes = capa.features.extractors.ghidra.helpers.get_bytes(ref, MAX_BYTES_FEATURE_SIZE)
+            if extracted_bytes and not capa.features.extractors.helpers.all_zeros(extracted_bytes):
+                # don't extract byte features for obvious strings
+                yield Bytes(extracted_bytes), ih.address
+
+
+def extract_insn_string_features(fh: FunctionHandle, bb: BBHandle, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
+    """
+    parse instruction string features
+
+    example:
+        push offset aAcr     ; "ACR  > "
+    """
+    insn: ghidra.program.database.code.InstructionDB = ih.inner
+    dyn_addr = OperandType.DYNAMIC | OperandType.ADDRESS
+
+    ref = insn.getAddress()
+    for i in range(insn.getNumOperands()):
+        if OperandType.isScalarAsAddress(insn.getOperandType(i)):
+            ref = insn.getAddress(i)
+        # strings are also referenced dynamically via pointers & arrays, so we need to deref them
+        if insn.getOperandType(i) == dyn_addr:
+            ref = insn.getAddress(i)
+            dat = getDataAt(ref)  # type: ignore [name-defined] # noqa: F821
+            if dat and dat.isPointer():
+                ref = dat.getValue()
+
+    if ref != insn.getAddress():
+        ghidra_dat = getDataAt(ref)  # type: ignore [name-defined] # noqa: F821
+        if ghidra_dat and ghidra_dat.hasStringValue():
+            yield String(ghidra_dat.getValue()), ih.address
+
+
+def extract_insn_mnemonic_features(
+    fh: FunctionHandle, bb: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    """parse instruction mnemonic features"""
+    insn: ghidra.program.database.code.InstructionDB = ih.inner
+
+    yield Mnemonic(insn.getMnemonicString().lower()), ih.address
+
+
+def extract_insn_obfs_call_plus_5_characteristic_features(
+    fh: FunctionHandle, bb: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    """
+    parse call $+5 instruction from the given instruction.
+    """
+    insn: ghidra.program.database.code.InstructionDB = ih.inner
+
+    if not capa.features.extractors.ghidra.helpers.is_call_or_jmp(insn):
+        return
+
+    code_ref = OperandType.ADDRESS | OperandType.CODE
+    ref = insn.getAddress()
+    for i in range(insn.getNumOperands()):
+        if insn.getOperandType(i) == code_ref:
+            ref = insn.getAddress(i)
+
+    if insn.getAddress().add(5) == ref:
+        yield Characteristic("call $+5"), ih.address
+
+
+def extract_insn_segment_access_features(
+    fh: FunctionHandle, bb: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    """parse instruction fs or gs access"""
+    insn: ghidra.program.database.code.InstructionDB = ih.inner
+
+    insn_str = insn.toString()
+
+    if "FS:" in insn_str:
+        yield Characteristic("fs access"), ih.address
+
+    if "GS:" in insn_str:
+        yield Characteristic("gs access"), ih.address
+
+
+def extract_insn_peb_access_characteristic_features(
+    fh: FunctionHandle, bb: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    """parse instruction peb access
+
+    fs:[0x30] on x86, gs:[0x60] on x64
+
+    """
+    insn: ghidra.program.database.code.InstructionDB = ih.inner
+
+    insn_str = insn.toString()
+    if insn_str.startswith(("PUSH", "MOV")):
+        if "FS:[0x30]" in insn_str or "GS:[0x60]" in insn_str:
+            yield Characteristic("peb access"), ih.address
+
+
+def extract_insn_cross_section_cflow(
+    fh: FunctionHandle, bb: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    """inspect the instruction for a CALL or JMP that crosses section boundaries"""
+    insn: ghidra.program.database.code.InstructionDB = ih.inner
+
+    if not capa.features.extractors.ghidra.helpers.is_call_or_jmp(insn):
+        return
+
+    externs = get_externs(fh.ctx)
+    fakes = get_fakes(fh.ctx)
+    imports = get_imports(fh.ctx)
+
+    # OperandType to dereference
+    addr_data = OperandType.ADDRESS | OperandType.DATA
+    addr_code = OperandType.ADDRESS | OperandType.CODE
+
+    ref_type = insn.getOperandType(0)
+
+    # both OperandType flags must be present
+    # bail on REGISTER alone
+    if OperandType.isRegister(ref_type):
+        if OperandType.isAddress(ref_type):
+            ref = insn.getAddress(0)  # Ghidra dereferences REG | ADDR
+            if capa.features.extractors.ghidra.helpers.check_addr_for_api(ref, fakes, imports, externs):
+                return
+        else:
+            return
+    elif ref_type in (addr_data, addr_code) or (OperandType.isIndirect(ref_type) and OperandType.isAddress(ref_type)):
+        # we must dereference and check if the addr is a pointer to an api function
+        ref = capa.features.extractors.ghidra.helpers.dereference_ptr(insn)
+        if capa.features.extractors.ghidra.helpers.check_addr_for_api(ref, fakes, imports, externs):
+            return
+    elif ref_type == OperandType.DYNAMIC | OperandType.ADDRESS or ref_type == OperandType.DYNAMIC:
+        return  # cannot resolve dynamics statically
+    else:
+        # pure address does not need to get dereferenced/ handled
+        ref = insn.getAddress(0)
+        if not ref:
+            # If it returned null, it was an indirect
+            # that had no address reference.
+            # This check is faster than checking for (indirect and not address)
+            return
+        if capa.features.extractors.ghidra.helpers.check_addr_for_api(ref, fakes, imports, externs):
+            return
+
+    this_mem_block = getMemoryBlock(insn.getAddress())  # type: ignore [name-defined] # noqa: F821
+    ref_block = getMemoryBlock(ref)  # type: ignore [name-defined] # noqa: F821
+    if ref_block != this_mem_block:
+        yield Characteristic("cross section flow"), ih.address
+
+
+def extract_function_calls_from(
+    fh: FunctionHandle,
+    bb: BBHandle,
+    ih: InsnHandle,
+) -> Iterator[Tuple[Feature, Address]]:
+    """extract functions calls from features
+
+    most relevant at the function scope, however, its most efficient to extract at the instruction scope
+    """
+    insn: ghidra.program.database.code.InstructionDB = ih.inner
+
+    if insn.getMnemonicString().startswith("CALL"):
+        # This method of "dereferencing" addresses/ pointers
+        # is not as robust as methods in other functions,
+        # but works just fine for this one
+        reference = 0
+        for ref in insn.getReferencesFrom():
+            addr = ref.getToAddress()
+
+            # avoid returning fake addrs
+            if not addr.isExternalAddress():
+                reference = addr.getOffset()
+
+            # if a reference is < 0, then ghidra pulled an offset from a DYNAMIC | ADDR (usually a stackvar)
+            # these cannot be resolved to actual addrs
+            if reference > 0:
+                yield Characteristic("calls from"), AbsoluteVirtualAddress(reference)
+
+
+def extract_function_indirect_call_characteristic_features(
+    fh: FunctionHandle,
+    bb: BBHandle,
+    ih: InsnHandle,
+) -> Iterator[Tuple[Feature, Address]]:
+    """extract indirect function calls (e.g., call eax or call dword ptr [edx+4])
+    does not include calls like => call ds:dword_ABD4974
+
+    most relevant at the function or basic block scope;
+    however, its most efficient to extract at the instruction scope
+    """
+    insn: ghidra.program.database.code.InstructionDB = ih.inner
+
+    if insn.getMnemonicString().startswith("CALL"):
+        if OperandType.isRegister(insn.getOperandType(0)):
+            yield Characteristic("indirect call"), ih.address
+        if OperandType.isIndirect(insn.getOperandType(0)):
+            yield Characteristic("indirect call"), ih.address
+
+
+def check_nzxor_security_cookie_delta(
+    fh: ghidra.program.database.function.FunctionDB, insn: ghidra.program.database.code.InstructionDB
+):
+    """Get the function containing the insn
+    Get the last block of the function that contains the insn
+
+    Check the bb containing the insn
+    Check the last bb of the function containing the insn
+    """
+
+    model = SimpleBlockModel(currentProgram())  # type: ignore [name-defined] # noqa: F821
+    insn_addr = insn.getAddress()
+    func_asv = fh.getBody()
+    first_addr = func_asv.getMinAddress()
+    last_addr = func_asv.getMaxAddress()
+
+    if model.getFirstCodeBlockContaining(
+        first_addr, monitor()  # type: ignore [name-defined] # noqa: F821
+    ) == model.getFirstCodeBlockContaining(
+        last_addr, monitor()  # type: ignore [name-defined] # noqa: F821
+    ):
+        if insn_addr < first_addr.add(SECURITY_COOKIE_BYTES_DELTA):
+            return True
+        else:
+            return insn_addr > last_addr.add(SECURITY_COOKIE_BYTES_DELTA * -1)
+    else:
+        return False
+
+
+def extract_insn_nzxor_characteristic_features(
+    fh: FunctionHandle,
+    bb: BBHandle,
+    ih: InsnHandle,
+) -> Iterator[Tuple[Feature, Address]]:
+    f: ghidra.program.database.function.FunctionDB = fh.inner
+    insn: ghidra.program.database.code.InstructionDB = ih.inner
+
+    if "XOR" not in insn.getMnemonicString():
+        return
+    if capa.features.extractors.ghidra.helpers.is_stack_referenced(insn):
+        return
+    if capa.features.extractors.ghidra.helpers.is_zxor(insn):
+        return
+    if check_nzxor_security_cookie_delta(f, insn):
+        return
+    yield Characteristic("nzxor"), ih.address
+
+
+def extract_features(
+    fh: FunctionHandle,
+    bb: BBHandle,
+    insn: InsnHandle,
+) -> Iterator[Tuple[Feature, Address]]:
+    for insn_handler in INSTRUCTION_HANDLERS:
+        for feature, addr in insn_handler(fh, bb, insn):
+            yield feature, addr
+
+
+INSTRUCTION_HANDLERS = (
+    extract_insn_api_features,
+    extract_insn_number_features,
+    extract_insn_bytes_features,
+    extract_insn_string_features,
+    extract_insn_offset_features,
+    extract_insn_nzxor_characteristic_features,
+    extract_insn_mnemonic_features,
+    extract_insn_obfs_call_plus_5_characteristic_features,
+    extract_insn_peb_access_characteristic_features,
+    extract_insn_cross_section_cflow,
+    extract_insn_segment_access_features,
+    extract_function_calls_from,
+    extract_function_indirect_call_characteristic_features,
+)
+
+
+def main():
+    """ """
+    features = []
+    from capa.features.extractors.ghidra.extractor import GhidraFeatureExtractor
+
+    for fh in GhidraFeatureExtractor().get_functions():
+        for bb in capa.features.extractors.ghidra.helpers.get_function_blocks(fh):
+            for insn in capa.features.extractors.ghidra.helpers.get_insn_in_range(bb):
+                features.extend(list(extract_features(fh, bb, insn)))
+
+    import pprint
+
+    pprint.pprint(features)  # noqa: T203
+
+
+if __name__ == "__main__":
+    main()

capa/features/extractors/helpers.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
 # Licensed under the Apache License, Version 2.0 (the "License");
 #  you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at: [package root]/LICENSE.txt
@@ -6,23 +6,18 @@
 #  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and limitations under the License.
 
-import sys
+import struct
 import builtins
-
-from capa.features.file import Import
-from capa.features.insn import API
+from typing import Tuple, Iterator
 
 MIN_STACKSTRING_LEN = 8
 
 
-def xor_static(data, i):
-    if sys.version_info >= (3, 0):
-        return bytes(c ^ i for c in data)
-    else:
-        return "".join(chr(ord(c) ^ i) for c in data)
+def xor_static(data: bytes, i: int) -> bytes:
+    return bytes(c ^ i for c in data)
 
 
-def is_aw_function(symbol):
+def is_aw_function(symbol: str) -> bool:
     """
     is the given function name an A/W function?
     these are variants of functions that, on Windows, accept either a narrow or wide string.
@@ -34,11 +29,10 @@ def is_aw_function(symbol):
     if symbol[-1] not in ("A", "W"):
         return False
 
-    # second to last character should be lowercase letter
-    return "a" <= symbol[-2] <= "z" or "0" <= symbol[-2] <= "9"
+    return True
 
 
-def is_ordinal(symbol):
+def is_ordinal(symbol: str) -> bool:
     """
     is the given symbol an ordinal that is prefixed by "#"?
     """
@@ -47,7 +41,7 @@ def is_ordinal(symbol):
     return False
 
 
-def generate_symbols(dll, symbol):
+def generate_symbols(dll: str, symbol: str) -> Iterator[str]:
     """
     for a given dll and symbol name, generate variants.
     we over-generate features to make matching easier.
@@ -57,8 +51,11 @@ def generate_symbols(dll, symbol):
       - CreateFileA
       - CreateFile
     """
+    # normalize dll name
+    dll = dll.lower()
+
     # kernel32.CreateFileA
-    yield "%s.%s" % (dll, symbol)
+    yield f"{dll}.{symbol}"
 
     if not is_ordinal(symbol):
         # CreateFileA
@@ -66,18 +63,35 @@ def generate_symbols(dll, symbol):
 
     if is_aw_function(symbol):
         # kernel32.CreateFile
-        yield "%s.%s" % (dll, symbol[:-1])
+        yield f"{dll}.{symbol[:-1]}"
 
         if not is_ordinal(symbol):
             # CreateFile
             yield symbol[:-1]
 
 
-def all_zeros(bytez):
+def reformat_forwarded_export_name(forwarded_name: str) -> str:
+    """
+    a forwarded export has a DLL name/path an symbol name.
+    we want the former to be lowercase, and the latter to be verbatim.
+    """
+
+    # use rpartition so we can split on separator between dll and name.
+    # the dll name can be a full path, like in the case of
+    # ef64d6d7c34250af8e21a10feb931c9b
+    # which i assume means the path can have embedded periods.
+    # so we don't want the first period, we want the last.
+    forwarded_dll, _, forwarded_symbol = forwarded_name.rpartition(".")
+    forwarded_dll = forwarded_dll.lower()
+
+    return f"{forwarded_dll}.{forwarded_symbol}"
+
+
+def all_zeros(bytez: bytes) -> bool:
     return all(b == 0 for b in builtins.bytes(bytez))
 
 
-def twos_complement(val, bits):
+def twos_complement(val: int, bits: int) -> int:
     """
     compute the 2's complement of int value val
 
@@ -90,3 +104,48 @@ def twos_complement(val, bits):
     else:
         # return positive value as is
         return val
+
+
+def carve_pe(pbytes: bytes, offset: int = 0) -> Iterator[Tuple[int, int]]:
+    """
+    Generate (offset, key) tuples of embedded PEs
+
+    Based on the version from vivisect:
+      https://github.com/vivisect/vivisect/blob/7be4037b1cecc4551b397f840405a1fc606f9b53/PE/carve.py#L19
+    And its IDA adaptation:
+      capa/features/extractors/ida/file.py
+    """
+    mz_xor = [
+        (
+            xor_static(b"MZ", key),
+            xor_static(b"PE", key),
+            key,
+        )
+        for key in range(256)
+    ]
+
+    pblen = len(pbytes)
+    todo = [(pbytes.find(mzx, offset), mzx, pex, key) for mzx, pex, key in mz_xor]
+    todo = [(off, mzx, pex, key) for (off, mzx, pex, key) in todo if off != -1]
+
+    while len(todo):
+        off, mzx, pex, key = todo.pop()
+
+        # The MZ header has one field we will check
+        # e_lfanew is at 0x3c
+        e_lfanew = off + 0x3C
+        if pblen < (e_lfanew + 4):
+            continue
+
+        newoff = struct.unpack("<I", xor_static(pbytes[e_lfanew : e_lfanew + 4], key))[0]
+
+        nextres = pbytes.find(mzx, off + 1)
+        if nextres != -1:
+            todo.append((nextres, mzx, pex, key))
+
+        peoff = off + newoff
+        if pblen < (peoff + 2):
+            continue
+
+        if pbytes[peoff : peoff + 2] == pex:
+            yield (off, key)

capa/features/extractors/ida/__init__.py

@@ -1,93 +0,0 @@
-# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
-# Licensed under the Apache License, Version 2.0 (the "License");
-#  you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at: [package root]/LICENSE.txt
-# Unless required by applicable law or agreed to in writing, software distributed under the License
-#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and limitations under the License.
-
-import sys
-import types
-
-import idaapi
-
-import capa.features.extractors.ida.file
-import capa.features.extractors.ida.insn
-import capa.features.extractors.ida.function
-import capa.features.extractors.ida.basicblock
-from capa.features.extractors import FeatureExtractor
-
-
-def get_ea(self):
-    """ """
-    if isinstance(self, (idaapi.BasicBlock, idaapi.func_t)):
-        return self.start_ea
-    if isinstance(self, idaapi.insn_t):
-        return self.ea
-    raise TypeError
-
-
-def add_ea_int_cast(o):
-    """
-    dynamically add a cast-to-int (`__int__`) method to the given object
-    that returns the value of the `.ea` property.
-    this bit of skullduggery lets use cast viv-utils objects as ints.
-    the correct way of doing this is to update viv-utils (or subclass the objects here).
-    """
-    if sys.version_info[0] >= 3:
-        setattr(o, "__int__", types.MethodType(get_ea, o))
-    else:
-        setattr(o, "__int__", types.MethodType(get_ea, o, type(o)))
-    return o
-
-
-class IdaFeatureExtractor(FeatureExtractor):
-    def __init__(self):
-        super(IdaFeatureExtractor, self).__init__()
-
-    def get_base_address(self):
-        return idaapi.get_imagebase()
-
-    def extract_file_features(self):
-        for (feature, ea) in capa.features.extractors.ida.file.extract_features():
-            yield feature, ea
-
-    def get_functions(self):
-        import capa.features.extractors.ida.helpers as ida_helpers
-
-        # data structure shared across functions yielded here.
-        # useful for caching analysis relevant across a single workspace.
-        ctx = {}
-
-        # ignore library functions and thunk functions as identified by IDA
-        for f in ida_helpers.get_functions(skip_thunks=True, skip_libs=True):
-            setattr(f, "ctx", ctx)
-            yield add_ea_int_cast(f)
-
-    @staticmethod
-    def get_function(ea):
-        f = idaapi.get_func(ea)
-        setattr(f, "ctx", {})
-        return add_ea_int_cast(f)
-
-    def extract_function_features(self, f):
-        for (feature, ea) in capa.features.extractors.ida.function.extract_features(f):
-            yield feature, ea
-
-    def get_basic_blocks(self, f):
-        for bb in capa.features.extractors.ida.helpers.get_function_blocks(f):
-            yield add_ea_int_cast(bb)
-
-    def extract_basic_block_features(self, f, bb):
-        for (feature, ea) in capa.features.extractors.ida.basicblock.extract_features(f, bb):
-            yield feature, ea
-
-    def get_instructions(self, f, bb):
-        import capa.features.extractors.ida.helpers as ida_helpers
-
-        for insn in ida_helpers.get_instructions_in_range(bb.start_ea, bb.end_ea):
-            yield add_ea_int_cast(insn)
-
-    def extract_insn_features(self, f, bb, insn):
-        for (feature, ea) in capa.features.extractors.ida.insn.extract_features(f, bb, insn):
-            yield feature, ea

capa/features/extractors/ida/basicblock.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
 # Licensed under the Apache License, Version 2.0 (the "License");
 #  you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at: [package root]/LICENSE.txt
@@ -6,25 +6,23 @@
 #  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and limitations under the License.
 
-import sys
 import string
 import struct
+from typing import Tuple, Iterator
 
 import idaapi
 
 import capa.features.extractors.ida.helpers
-from capa.features import Characteristic
+from capa.features.common import Feature, Characteristic
+from capa.features.address import Address
 from capa.features.basicblock import BasicBlock
 from capa.features.extractors.ida import helpers
 from capa.features.extractors.helpers import MIN_STACKSTRING_LEN
+from capa.features.extractors.base_extractor import BBHandle, FunctionHandle
 
 
-def get_printable_len(op):
-    """Return string length if all operand bytes are ascii or utf16-le printable
-
-    args:
-        op (IDA op_t)
-    """
+def get_printable_len(op: idaapi.op_t) -> int:
+    """Return string length if all operand bytes are ascii or utf16-le printable"""
     op_val = capa.features.extractors.ida.helpers.mask_op_val(op)
 
     if op.dtype == idaapi.dt_byte:
@@ -36,21 +34,14 @@ def get_printable_len(op):
     elif op.dtype == idaapi.dt_qword:
         chars = struct.pack("<Q", op_val)
     else:
-        raise ValueError("Unhandled operand data type 0x%x." % op.dtype)
+        raise ValueError(f"Unhandled operand data type 0x{op.dtype:x}.")
 
-    def is_printable_ascii(chars):
-        if sys.version_info[0] >= 3:
-            return all(c < 127 and chr(c) in string.printable for c in chars)
-        else:
-            return all(ord(c) < 127 and c in string.printable for c in chars)
+    def is_printable_ascii(chars_: bytes):
+        return all(c < 127 and chr(c) in string.printable for c in chars_)
 
-    def is_printable_utf16le(chars):
-        if sys.version_info[0] >= 3:
-            if all(c == 0x00 for c in chars[1::2]):
-                return is_printable_ascii(chars[::2])
-        else:
-            if all(c == "\x00" for c in chars[1::2]):
-                return is_printable_ascii(chars[::2])
+    def is_printable_utf16le(chars_: bytes):
+        if all(c == 0x00 for c in chars_[1::2]):
+            return is_printable_ascii(chars_[::2])
 
     if is_printable_ascii(chars):
         return idaapi.get_dtype_size(op.dtype)
@@ -61,12 +52,8 @@ def get_printable_len(op):
     return 0
 
 
-def is_mov_imm_to_stack(insn):
-    """verify instruction moves immediate onto stack
-
-    args:
-        insn (IDA insn_t)
-    """
+def is_mov_imm_to_stack(insn: idaapi.insn_t) -> bool:
+    """verify instruction moves immediate onto stack"""
     if insn.Op2.type != idaapi.o_imm:
         return False
 
@@ -79,14 +66,10 @@ def is_mov_imm_to_stack(insn):
     return True
 
 
-def bb_contains_stackstring(f, bb):
+def bb_contains_stackstring(f: idaapi.func_t, bb: idaapi.BasicBlock) -> bool:
     """check basic block for stackstring indicators
 
     true if basic block contains enough moves of constant bytes to the stack
-
-    args:
-        f (IDA func_t)
-        bb (IDA BasicBlock)
     """
     count = 0
     for insn in capa.features.extractors.ida.helpers.get_instructions_in_range(bb.start_ea, bb.end_ea):
@@ -97,57 +80,27 @@ def bb_contains_stackstring(f, bb):
     return False
 
 
-def extract_bb_stackstring(f, bb):
-    """extract stackstring indicators from basic block
-
-    args:
-        f (IDA func_t)
-        bb (IDA BasicBlock)
-    """
-    if bb_contains_stackstring(f, bb):
-        yield Characteristic("stack string"), bb.start_ea
+def extract_bb_stackstring(fh: FunctionHandle, bbh: BBHandle) -> Iterator[Tuple[Feature, Address]]:
+    """extract stackstring indicators from basic block"""
+    if bb_contains_stackstring(fh.inner, bbh.inner):
+        yield Characteristic("stack string"), bbh.address
 
 
-def extract_bb_tight_loop(f, bb):
-    """extract tight loop indicators from a basic block
-
-    args:
-        f (IDA func_t)
-        bb (IDA BasicBlock)
-    """
-    if capa.features.extractors.ida.helpers.is_basic_block_tight_loop(bb):
-        yield Characteristic("tight loop"), bb.start_ea
+def extract_bb_tight_loop(fh: FunctionHandle, bbh: BBHandle) -> Iterator[Tuple[Feature, Address]]:
+    """extract tight loop indicators from a basic block"""
+    if capa.features.extractors.ida.helpers.is_basic_block_tight_loop(bbh.inner):
+        yield Characteristic("tight loop"), bbh.address
 
 
-def extract_features(f, bb):
-    """extract basic block features
-
-    args:
-        f (IDA func_t)
-        bb (IDA BasicBlock)
-    """
+def extract_features(fh: FunctionHandle, bbh: BBHandle) -> Iterator[Tuple[Feature, Address]]:
+    """extract basic block features"""
     for bb_handler in BASIC_BLOCK_HANDLERS:
-        for (feature, ea) in bb_handler(f, bb):
-            yield feature, ea
-    yield BasicBlock(), bb.start_ea
+        for feature, addr in bb_handler(fh, bbh):
+            yield feature, addr
+    yield BasicBlock(), bbh.address
 
 
 BASIC_BLOCK_HANDLERS = (
     extract_bb_tight_loop,
     extract_bb_stackstring,
 )
-
-
-def main():
-    features = []
-    for f in helpers.get_functions(skip_thunks=True, skip_libs=True):
-        for bb in idaapi.FlowChart(f, flags=idaapi.FC_PREDS):
-            features.extend(list(extract_features(f, bb)))
-
-    import pprint
-
-    pprint.pprint(features)
-
-
-if __name__ == "__main__":
-    main()

capa/features/extractors/ida/extractor.py

@@ -0,0 +1,71 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+from typing import List, Tuple, Iterator
+
+import idaapi
+
+import capa.ida.helpers
+import capa.features.extractors.elf
+import capa.features.extractors.ida.file
+import capa.features.extractors.ida.insn
+import capa.features.extractors.ida.global_
+import capa.features.extractors.ida.function
+import capa.features.extractors.ida.basicblock
+from capa.features.common import Feature
+from capa.features.address import Address, AbsoluteVirtualAddress
+from capa.features.extractors.base_extractor import BBHandle, InsnHandle, FunctionHandle, FeatureExtractor
+
+
+class IdaFeatureExtractor(FeatureExtractor):
+    def __init__(self):
+        super().__init__()
+        self.global_features: List[Tuple[Feature, Address]] = []
+        self.global_features.extend(capa.features.extractors.ida.file.extract_file_format())
+        self.global_features.extend(capa.features.extractors.ida.global_.extract_os())
+        self.global_features.extend(capa.features.extractors.ida.global_.extract_arch())
+
+    def get_base_address(self):
+        return AbsoluteVirtualAddress(idaapi.get_imagebase())
+
+    def extract_global_features(self):
+        yield from self.global_features
+
+    def extract_file_features(self):
+        yield from capa.features.extractors.ida.file.extract_features()
+
+    def get_functions(self) -> Iterator[FunctionHandle]:
+        import capa.features.extractors.ida.helpers as ida_helpers
+
+        # ignore library functions and thunk functions as identified by IDA
+        yield from ida_helpers.get_functions(skip_thunks=True, skip_libs=True)
+
+    @staticmethod
+    def get_function(ea: int) -> FunctionHandle:
+        f = idaapi.get_func(ea)
+        return FunctionHandle(address=AbsoluteVirtualAddress(f.start_ea), inner=f)
+
+    def extract_function_features(self, fh: FunctionHandle) -> Iterator[Tuple[Feature, Address]]:
+        yield from capa.features.extractors.ida.function.extract_features(fh)
+
+    def get_basic_blocks(self, fh: FunctionHandle) -> Iterator[BBHandle]:
+        import capa.features.extractors.ida.helpers as ida_helpers
+
+        for bb in ida_helpers.get_function_blocks(fh.inner):
+            yield BBHandle(address=AbsoluteVirtualAddress(bb.start_ea), inner=bb)
+
+    def extract_basic_block_features(self, fh: FunctionHandle, bbh: BBHandle) -> Iterator[Tuple[Feature, Address]]:
+        yield from capa.features.extractors.ida.basicblock.extract_features(fh, bbh)
+
+    def get_instructions(self, fh: FunctionHandle, bbh: BBHandle) -> Iterator[InsnHandle]:
+        import capa.features.extractors.ida.helpers as ida_helpers
+
+        for insn in ida_helpers.get_instructions_in_range(bbh.inner.start_ea, bbh.inner.end_ea):
+            yield InsnHandle(address=AbsoluteVirtualAddress(insn.ea), inner=insn)
+
+    def extract_insn_features(self, fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle):
+        yield from capa.features.extractors.ida.insn.extract_features(fh, bbh, ih)

capa/features/extractors/ida/file.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
 # Licensed under the Apache License, Version 2.0 (the "License");
 #  you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at: [package root]/LICENSE.txt
@@ -7,26 +7,29 @@
 # See the License for the specific language governing permissions and limitations under the License.
 
 import struct
+from typing import Tuple, Iterator
 
 import idc
 import idaapi
 import idautils
+import ida_entry
 
+import capa.features.extractors.common
 import capa.features.extractors.helpers
 import capa.features.extractors.strings
 import capa.features.extractors.ida.helpers
-from capa.features import String, Characteristic
-from capa.features.file import Export, Import, Section
+from capa.features.file import Export, Import, Section, FunctionName
+from capa.features.common import FORMAT_PE, FORMAT_ELF, Format, String, Feature, Characteristic
+from capa.features.address import NO_ADDRESS, Address, FileOffsetAddress, AbsoluteVirtualAddress
+
+MAX_OFFSET_PE_AFTER_MZ = 0x200
 
 
-def check_segment_for_pe(seg):
+def check_segment_for_pe(seg: idaapi.segment_t) -> Iterator[Tuple[int, int]]:
     """check segment for embedded PE
 
     adapted for IDA from:
-    https://github.com/vivisect/vivisect/blob/7be4037b1cecc4551b397f840405a1fc606f9b53/PE/carve.py#L19
-
-    args:
-        seg (IDA segment_t)
+    https://github.com/vivisect/vivisect/blob/91e8419a861f49779f18316f155311967e696836/PE/carve.py#L25
     """
     seg_max = seg.end_ea
     mz_xor = [
@@ -39,14 +42,15 @@ def check_segment_for_pe(seg):
     ]
 
     todo = []
-    for (mzx, pex, i) in mz_xor:
+    for mzx, pex, i in mz_xor:
+        # find all segment offsets containing XOR'd "MZ" bytes
         for off in capa.features.extractors.ida.helpers.find_byte_sequence(seg.start_ea, seg.end_ea, mzx):
             todo.append((off, mzx, pex, i))
 
     while len(todo):
         off, mzx, pex, i = todo.pop()
 
-        # The MZ header has one field we will check e_lfanew is at 0x3c
+        # MZ header has one field we will check e_lfanew is at 0x3c
         e_lfanew = off + 0x3C
 
         if seg_max < (e_lfanew + 4):
@@ -54,18 +58,19 @@ def check_segment_for_pe(seg):
 
         newoff = struct.unpack("<I", capa.features.extractors.helpers.xor_static(idc.get_bytes(e_lfanew, 4), i))[0]
 
+        # assume XOR'd "PE" bytes exist within threshold
+        if newoff > MAX_OFFSET_PE_AFTER_MZ:
+            continue
+
         peoff = off + newoff
         if seg_max < (peoff + 2):
             continue
 
         if idc.get_bytes(peoff, 2) == pex:
-            yield (off, i)
-
-        for nextres in capa.features.extractors.ida.helpers.find_byte_sequence(off + 1, seg.end_ea, mzx):
-            todo.append((nextres, mzx, pex, i))
+            yield off, i
 
 
-def extract_file_embedded_pe():
+def extract_file_embedded_pe() -> Iterator[Tuple[Feature, Address]]:
     """extract embedded PE features
 
     IDA must load resource sections for this to be complete
@@ -73,17 +78,23 @@ def extract_file_embedded_pe():
         - Check 'Load resource sections' when opening binary in IDA manually
     """
     for seg in capa.features.extractors.ida.helpers.get_segments(skip_header_segments=True):
-        for (ea, _) in check_segment_for_pe(seg):
-            yield Characteristic("embedded pe"), ea
+        for ea, _ in check_segment_for_pe(seg):
+            yield Characteristic("embedded pe"), FileOffsetAddress(ea)
 
 
-def extract_file_export_names():
-    """ extract function exports """
-    for (_, _, ea, name) in idautils.Entries():
-        yield Export(name), ea
+def extract_file_export_names() -> Iterator[Tuple[Feature, Address]]:
+    """extract function exports"""
+    for _, ordinal, ea, name in idautils.Entries():
+        forwarded_name = ida_entry.get_entry_forwarder(ordinal)
+        if forwarded_name is None:
+            yield Export(name), AbsoluteVirtualAddress(ea)
+        else:
+            forwarded_name = capa.features.extractors.helpers.reformat_forwarded_export_name(forwarded_name)
+            yield Export(forwarded_name), AbsoluteVirtualAddress(ea)
+            yield Characteristic("forwarded export"), AbsoluteVirtualAddress(ea)
 
 
-def extract_file_import_names():
+def extract_file_import_names() -> Iterator[Tuple[Feature, Address]]:
     """extract function imports
 
     1. imports by ordinal:
@@ -94,28 +105,32 @@ def extract_file_import_names():
      - modulename.importname
      - importname
     """
-    for (ea, info) in capa.features.extractors.ida.helpers.get_file_imports().items():
+    for ea, info in capa.features.extractors.ida.helpers.get_file_imports().items():
+        addr = AbsoluteVirtualAddress(ea)
         if info[1] and info[2]:
             # e.g. in mimikatz: ('cabinet', 'FCIAddFile', 11L)
             # extract by name here and by ordinal below
             for name in capa.features.extractors.helpers.generate_symbols(info[0], info[1]):
-                yield Import(name), ea
+                yield Import(name), addr
             dll = info[0]
-            symbol = "#%d" % (info[2])
+            symbol = f"#{info[2]}"
         elif info[1]:
             dll = info[0]
             symbol = info[1]
         elif info[2]:
             dll = info[0]
-            symbol = "#%d" % (info[2])
+            symbol = f"#{info[2]}"
         else:
             continue
 
         for name in capa.features.extractors.helpers.generate_symbols(dll, symbol):
-            yield Import(name), ea
+            yield Import(name), addr
+
+    for ea, info in capa.features.extractors.ida.helpers.get_file_externs().items():
+        yield Import(info[1]), AbsoluteVirtualAddress(ea)
 
 
-def extract_file_section_names():
+def extract_file_section_names() -> Iterator[Tuple[Feature, Address]]:
     """extract section names
 
     IDA must load resource sections for this to be complete
@@ -123,10 +138,10 @@ def extract_file_section_names():
         - Check 'Load resource sections' when opening binary in IDA manually
     """
     for seg in capa.features.extractors.ida.helpers.get_segments(skip_header_segments=True):
-        yield Section(idaapi.get_segm_name(seg)), seg.start_ea
+        yield Section(idaapi.get_segm_name(seg)), AbsoluteVirtualAddress(seg.start_ea)
 
 
-def extract_file_strings():
+def extract_file_strings() -> Iterator[Tuple[Feature, Address]]:
     """extract ASCII and UTF-16 LE strings
 
     IDA must load resource sections for this to be complete
@@ -136,18 +151,50 @@ def extract_file_strings():
     for seg in capa.features.extractors.ida.helpers.get_segments():
         seg_buff = capa.features.extractors.ida.helpers.get_segment_buffer(seg)
 
+        # differing to common string extractor factor in segment offset here
         for s in capa.features.extractors.strings.extract_ascii_strings(seg_buff):
-            yield String(s.s), (seg.start_ea + s.offset)
+            yield String(s.s), FileOffsetAddress(seg.start_ea + s.offset)
 
         for s in capa.features.extractors.strings.extract_unicode_strings(seg_buff):
-            yield String(s.s), (seg.start_ea + s.offset)
+            yield String(s.s), FileOffsetAddress(seg.start_ea + s.offset)
 
 
-def extract_features():
-    """ extract file features """
+def extract_file_function_names() -> Iterator[Tuple[Feature, Address]]:
+    """
+    extract the names of statically-linked library functions.
+    """
+    for ea in idautils.Functions():
+        addr = AbsoluteVirtualAddress(ea)
+        if idaapi.get_func(ea).flags & idaapi.FUNC_LIB:
+            name = idaapi.get_name(ea)
+            yield FunctionName(name), addr
+            if name.startswith("_"):
+                # some linkers may prefix linked routines with a `_` to avoid name collisions.
+                # extract features for both the mangled and un-mangled representations.
+                # e.g. `_fwrite` -> `fwrite`
+                # see: https://stackoverflow.com/a/2628384/87207
+                yield FunctionName(name[1:]), addr
+
+
+def extract_file_format() -> Iterator[Tuple[Feature, Address]]:
+    file_info = idaapi.get_inf_structure()
+
+    if file_info.filetype in (idaapi.f_PE, idaapi.f_COFF):
+        yield Format(FORMAT_PE), NO_ADDRESS
+    elif file_info.filetype == idaapi.f_ELF:
+        yield Format(FORMAT_ELF), NO_ADDRESS
+    elif file_info.filetype == idaapi.f_BIN:
+        # no file type to return when processing a binary file, but we want to continue processing
+        return
+    else:
+        raise NotImplementedError(f"unexpected file format: {file_info.filetype}")
+
+
+def extract_features() -> Iterator[Tuple[Feature, Address]]:
+    """extract file features"""
     for file_handler in FILE_HANDLERS:
-        for feature, va in file_handler():
-            yield feature, va
+        for feature, addr in file_handler():
+            yield feature, addr
 
 
 FILE_HANDLERS = (
@@ -156,15 +203,6 @@ FILE_HANDLERS = (
     extract_file_strings,
     extract_file_section_names,
     extract_file_embedded_pe,
+    extract_file_function_names,
+    extract_file_format,
 )
-
-
-def main():
-    """ """
-    import pprint
-
-    pprint.pprint(list(extract_features()))
-
-
-if __name__ == "__main__":
-    main()

capa/features/extractors/ida/function.py

@@ -1,35 +1,31 @@
-# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
 # Licensed under the Apache License, Version 2.0 (the "License");
 #  you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at: [package root]/LICENSE.txt
 # Unless required by applicable law or agreed to in writing, software distributed under the License
 #  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and limitations under the License.
+from typing import Tuple, Iterator
 
 import idaapi
 import idautils
 
 import capa.features.extractors.ida.helpers
-from capa.features import Characteristic
+from capa.features.common import Feature, Characteristic
+from capa.features.address import Address, AbsoluteVirtualAddress
 from capa.features.extractors import loops
+from capa.features.extractors.base_extractor import FunctionHandle
 
 
-def extract_function_calls_to(f):
-    """extract callers to a function
-
-    args:
-        f (IDA func_t)
-    """
-    for ea in idautils.CodeRefsTo(f.start_ea, True):
-        yield Characteristic("calls to"), ea
+def extract_function_calls_to(fh: FunctionHandle):
+    """extract callers to a function"""
+    for ea in idautils.CodeRefsTo(fh.inner.start_ea, True):
+        yield Characteristic("calls to"), AbsoluteVirtualAddress(ea)
 
 
-def extract_function_loop(f):
-    """extract loop indicators from a function
-
-    args:
-        f (IDA func_t)
-    """
+def extract_function_loop(fh: FunctionHandle):
+    """extract loop indicators from a function"""
+    f: idaapi.func_t = fh.inner
     edges = []
 
     # construct control flow graph
@@ -38,43 +34,19 @@ def extract_function_loop(f):
             edges.append((bb.start_ea, succ.start_ea))
 
     if loops.has_loop(edges):
-        yield Characteristic("loop"), f.start_ea
+        yield Characteristic("loop"), fh.address
 
 
-def extract_recursive_call(f):
-    """extract recursive function call
-
-    args:
-        f (IDA func_t)
-    """
-    if capa.features.extractors.ida.helpers.is_function_recursive(f):
-        yield Characteristic("recursive call"), f.start_ea
+def extract_recursive_call(fh: FunctionHandle):
+    """extract recursive function call"""
+    if capa.features.extractors.ida.helpers.is_function_recursive(fh.inner):
+        yield Characteristic("recursive call"), fh.address
 
 
-def extract_features(f):
-    """extract function features
-
-    arg:
-        f (IDA func_t)
-    """
+def extract_features(fh: FunctionHandle) -> Iterator[Tuple[Feature, Address]]:
     for func_handler in FUNCTION_HANDLERS:
-        for (feature, ea) in func_handler(f):
-            yield feature, ea
+        for feature, addr in func_handler(fh):
+            yield feature, addr
 
 
 FUNCTION_HANDLERS = (extract_function_calls_to, extract_function_loop, extract_recursive_call)
-
-
-def main():
-    """ """
-    features = []
-    for f in capa.features.extractors.ida.get_functions(skip_thunks=True, skip_libs=True):
-        features.extend(list(extract_features(f)))
-
-    import pprint
-
-    pprint.pprint(features)
-
-
-if __name__ == "__main__":
-    main()

capa/features/extractors/ida/global_.py

@@ -0,0 +1,65 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import logging
+import contextlib
+from typing import Tuple, Iterator
+
+import idaapi
+import ida_loader
+
+import capa.ida.helpers
+import capa.features.extractors.elf
+from capa.features.common import OS, ARCH_I386, ARCH_AMD64, OS_WINDOWS, Arch, Feature
+from capa.features.address import NO_ADDRESS, Address
+
+logger = logging.getLogger(__name__)
+
+
+def extract_os() -> Iterator[Tuple[Feature, Address]]:
+    format_name: str = ida_loader.get_file_type_name()
+
+    if "PE" in format_name:
+        yield OS(OS_WINDOWS), NO_ADDRESS
+
+    elif "ELF" in format_name:
+        with contextlib.closing(capa.ida.helpers.IDAIO()) as f:
+            os = capa.features.extractors.elf.detect_elf_os(f)
+
+        yield OS(os), NO_ADDRESS
+
+    else:
+        # we likely end up here:
+        #  1. handling shellcode, or
+        #  2. handling a new file format (e.g. macho)
+        #
+        # for (1) we can't do much - its shellcode and all bets are off.
+        # we could maybe accept a further CLI argument to specify the OS,
+        # but i think this would be rarely used.
+        # rules that rely on OS conditions will fail to match on shellcode.
+        #
+        # for (2), this logic will need to be updated as the format is implemented.
+        logger.debug("unsupported file format: %s, will not guess OS", format_name)
+        return
+
+
+def extract_arch() -> Iterator[Tuple[Feature, Address]]:
+    info: idaapi.idainfo = idaapi.get_inf_structure()
+    if info.procname == "metapc" and info.is_64bit():
+        yield Arch(ARCH_AMD64), NO_ADDRESS
+    elif info.procname == "metapc" and info.is_32bit():
+        yield Arch(ARCH_I386), NO_ADDRESS
+    elif info.procname == "metapc":
+        logger.debug("unsupported architecture: non-32-bit nor non-64-bit intel")
+        return
+    else:
+        # we likely end up here:
+        #  1. handling a new architecture (e.g. aarch64)
+        #
+        # for (1), this logic will need to be updated as the format is implemented.
+        logger.debug("unsupported architecture: %s", info.procname)
+        return

capa/features/extractors/ida/helpers.py

@@ -1,21 +1,24 @@
-# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
 # Licensed under the Apache License, Version 2.0 (the "License");
 #  you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at: [package root]/LICENSE.txt
 # Unless required by applicable law or agreed to in writing, software distributed under the License
 #  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and limitations under the License.
-
-import sys
-import string
+import functools
+from typing import Any, Dict, Tuple, Iterator, Optional
 
 import idc
 import idaapi
 import idautils
 import ida_bytes
+import ida_segment
+
+from capa.features.address import AbsoluteVirtualAddress
+from capa.features.extractors.base_extractor import FunctionHandle
 
 
-def find_byte_sequence(start, end, seq):
+def find_byte_sequence(start: int, end: int, seq: bytes) -> Iterator[int]:
     """yield all ea of a given byte sequence
 
     args:
@@ -23,36 +26,33 @@ def find_byte_sequence(start, end, seq):
         end: max virtual address
         seq: bytes to search e.g. b"\x01\x03"
     """
-    if sys.version_info[0] >= 3:
-        seq = " ".join(["%02x" % b for b in seq])
-    else:
-        seq = " ".join(["%02x" % ord(b) for b in seq])
-
+    seqstr = " ".join([f"{b:02x}" for b in seq])
     while True:
-        ea = idaapi.find_binary(start, end, seq, 0, idaapi.SEARCH_DOWN)
+        # TODO(mike-hunhoff): find_binary is deprecated. Please use ida_bytes.bin_search() instead.
+        # https://github.com/mandiant/capa/issues/1606
+        ea = idaapi.find_binary(start, end, seqstr, 0, idaapi.SEARCH_DOWN)
         if ea == idaapi.BADADDR:
             break
         start = ea + 1
         yield ea
 
 
-def get_functions(start=None, end=None, skip_thunks=False, skip_libs=False):
+def get_functions(
+    start: Optional[int] = None, end: Optional[int] = None, skip_thunks: bool = False, skip_libs: bool = False
+) -> Iterator[FunctionHandle]:
     """get functions, range optional
 
     args:
         start: min virtual address
         end: max virtual address
-
-    ret:
-        yield func_t*
     """
     for ea in idautils.Functions(start=start, end=end):
         f = idaapi.get_func(ea)
         if not (skip_thunks and (f.flags & idaapi.FUNC_THUNK) or skip_libs and (f.flags & idaapi.FUNC_LIB)):
-            yield f
+            yield FunctionHandle(address=AbsoluteVirtualAddress(ea), inner=f)
 
 
-def get_segments(skip_header_segments=False):
+def get_segments(skip_header_segments=False) -> Iterator[idaapi.segment_t]:
     """get list of segments (sections) in the binary image
 
     args:
@@ -64,7 +64,7 @@ def get_segments(skip_header_segments=False):
             yield seg
 
 
-def get_segment_buffer(seg):
+def get_segment_buffer(seg: idaapi.segment_t) -> bytes:
     """return bytes stored in a given segment
 
     decrease buffer size until IDA is able to read bytes from the segment
@@ -82,9 +82,22 @@ def get_segment_buffer(seg):
     return buff if buff else b""
 
 
-def get_file_imports():
-    """ get file imports """
-    imports = {}
+def inspect_import(imports, library, ea, function, ordinal):
+    if function and function.startswith("__imp_"):
+        # handle mangled PE imports
+        function = function[len("__imp_") :]
+
+    if function and "@@" in function:
+        # handle mangled ELF imports, like "fopen@@glibc_2.2.5"
+        function, _, _ = function.partition("@@")
+
+    imports[ea] = (library.lower(), function, ordinal)
+    return True
+
+
+def get_file_imports() -> Dict[int, Tuple[str, str, int]]:
+    """get file imports"""
+    imports: Dict[int, Tuple[str, str, int]] = {}
 
     for idx in range(idaapi.get_import_module_qty()):
         library = idaapi.get_import_module_name(idx)
@@ -92,26 +105,38 @@ def get_file_imports():
         if not library:
             continue
 
-        def inspect_import(ea, function, ordinal):
-            if function and function.startswith("__imp_"):
-                # handle mangled names starting
-                function = function[len("__imp_") :]
-            imports[ea] = (library.lower(), function, ordinal)
-            return True
+        # IDA uses section names for the library of ELF imports, like ".dynsym".
+        # These are not useful to us, we may need to expand this list over time
+        # TODO(williballenthin): find all section names used by IDA
+        # https://github.com/mandiant/capa/issues/1419
+        if library == ".dynsym":
+            library = ""
 
-        idaapi.enum_import_names(idx, inspect_import)
+        cb = functools.partial(inspect_import, imports, library)
+        idaapi.enum_import_names(idx, cb)
 
     return imports
 
 
-def get_instructions_in_range(start, end):
+def get_file_externs() -> Dict[int, Tuple[str, str, int]]:
+    externs = {}
+
+    for seg in get_segments(skip_header_segments=True):
+        if seg.type != ida_segment.SEG_XTRN:
+            continue
+
+        for ea in idautils.Functions(seg.start_ea, seg.end_ea):
+            externs[ea] = ("", idaapi.get_func_name(ea), -1)
+
+    return externs
+
+
+def get_instructions_in_range(start: int, end: int) -> Iterator[idaapi.insn_t]:
     """yield instructions in range
 
     args:
         start: virtual address (inclusive)
         end: virtual address (exclusive)
-    yield:
-        (insn_t*)
     """
     for head in idautils.Heads(start, end):
         insn = idautils.DecodeInstruction(head)
@@ -119,8 +144,8 @@ def get_instructions_in_range(start, end):
             yield insn
 
 
-def is_operand_equal(op1, op2):
-    """ compare two IDA op_t """
+def is_operand_equal(op1: idaapi.op_t, op2: idaapi.op_t) -> bool:
+    """compare two IDA op_t"""
     if op1.flags != op2.flags:
         return False
 
@@ -145,8 +170,8 @@ def is_operand_equal(op1, op2):
     return True
 
 
-def is_basic_block_equal(bb1, bb2):
-    """ compare two IDA BasicBlock """
+def is_basic_block_equal(bb1: idaapi.BasicBlock, bb2: idaapi.BasicBlock) -> bool:
+    """compare two IDA BasicBlock"""
     if bb1.start_ea != bb2.start_ea:
         return False
 
@@ -159,12 +184,12 @@ def is_basic_block_equal(bb1, bb2):
     return True
 
 
-def basic_block_size(bb):
-    """ calculate size of basic block """
+def basic_block_size(bb: idaapi.BasicBlock) -> int:
+    """calculate size of basic block"""
     return bb.end_ea - bb.start_ea
 
 
-def read_bytes_at(ea, count):
+def read_bytes_at(ea: int, count: int) -> bytes:
     """ """
     # check if byte has a value, see get_wide_byte doc
     if not idc.is_loaded(ea):
@@ -177,10 +202,10 @@ def read_bytes_at(ea, count):
         return idc.get_bytes(ea, count)
 
 
-def find_string_at(ea, min=4):
-    """ check if ASCII string exists at a given virtual address """
+def find_string_at(ea: int, min_: int = 4) -> str:
+    """check if ASCII string exists at a given virtual address"""
     found = idaapi.get_strlit_contents(ea, -1, idaapi.STRTYPE_C)
-    if found and len(found) > min:
+    if found and len(found) >= min_:
         try:
             found = found.decode("ascii")
             # hacky check for IDA bug; get_strlit_contents also reads Unicode as
@@ -194,7 +219,7 @@ def find_string_at(ea, min=4):
     return ""
 
 
-def get_op_phrase_info(op):
+def get_op_phrase_info(op: idaapi.op_t) -> Dict:
     """parse phrase features from operand
 
     Pretty much dup of sark's implementation:
@@ -204,7 +229,8 @@ def get_op_phrase_info(op):
         return {}
 
     scale = 1 << ((op.specflag2 & 0xC0) >> 6)
-    offset = op.addr
+    # IDA ea_t may be 32- or 64-bit; we assume displacement can only be 32-bit
+    offset = op.addr & 0xFFFFFFFF
 
     if op.specflag1 == 0:
         index = None
@@ -231,47 +257,45 @@ def get_op_phrase_info(op):
     return {"base": base, "index": index, "scale": scale, "offset": offset}
 
 
-def is_op_write(insn, op):
-    """ Check if an operand is written to (destination operand) """
+def is_op_write(insn: idaapi.insn_t, op: idaapi.op_t) -> bool:
+    """Check if an operand is written to (destination operand)"""
     return idaapi.has_cf_chg(insn.get_canon_feature(), op.n)
 
 
-def is_op_read(insn, op):
-    """ Check if an operand is read from (source operand) """
+def is_op_read(insn: idaapi.insn_t, op: idaapi.op_t) -> bool:
+    """Check if an operand is read from (source operand)"""
     return idaapi.has_cf_use(insn.get_canon_feature(), op.n)
 
 
-def is_op_offset(insn, op):
-    """ Check is an operand has been marked as an offset (by auto-analysis or manually) """
+def is_op_offset(insn: idaapi.insn_t, op: idaapi.op_t) -> bool:
+    """Check is an operand has been marked as an offset (by auto-analysis or manually)"""
     flags = idaapi.get_flags(insn.ea)
     return ida_bytes.is_off(flags, op.n)
 
 
-def is_sp_modified(insn):
-    """ determine if instruction modifies SP, ESP, RSP """
-    for op in get_insn_ops(insn, target_ops=(idaapi.o_reg,)):
-        if op.reg == idautils.procregs.sp.reg and is_op_write(insn, op):
-            # register is stack and written
-            return True
-    return False
+def is_sp_modified(insn: idaapi.insn_t) -> bool:
+    """determine if instruction modifies SP, ESP, RSP"""
+    return any(
+        op.reg == idautils.procregs.sp.reg and is_op_write(insn, op)
+        for op in get_insn_ops(insn, target_ops=(idaapi.o_reg,))
+    )
 
 
-def is_bp_modified(insn):
-    """ check if instruction modifies BP, EBP, RBP """
-    for op in get_insn_ops(insn, target_ops=(idaapi.o_reg,)):
-        if op.reg == idautils.procregs.bp.reg and is_op_write(insn, op):
-            # register is base and written
-            return True
-    return False
+def is_bp_modified(insn: idaapi.insn_t) -> bool:
+    """check if instruction modifies BP, EBP, RBP"""
+    return any(
+        op.reg == idautils.procregs.bp.reg and is_op_write(insn, op)
+        for op in get_insn_ops(insn, target_ops=(idaapi.o_reg,))
+    )
 
 
-def is_frame_register(reg):
-    """ check if register is sp or bp """
+def is_frame_register(reg: int) -> bool:
+    """check if register is sp or bp"""
     return reg in (idautils.procregs.sp.reg, idautils.procregs.bp.reg)
 
 
-def get_insn_ops(insn, target_ops=()):
-    """ yield op_t for instruction, filter on type if specified """
+def get_insn_ops(insn: idaapi.insn_t, target_ops: Optional[Tuple[Any]] = None) -> idaapi.op_t:
+    """yield op_t for instruction, filter on type if specified"""
     for op in insn.ops:
         if op.type == idaapi.o_void:
             # avoid looping all 6 ops if only subset exists
@@ -281,12 +305,12 @@ def get_insn_ops(insn, target_ops=()):
         yield op
 
 
-def is_op_stack_var(ea, index):
-    """ check if operand is a stack variable """
+def is_op_stack_var(ea: int, index: int) -> bool:
+    """check if operand is a stack variable"""
     return idaapi.is_stkvar(idaapi.get_flags(ea), index)
 
 
-def mask_op_val(op):
+def mask_op_val(op: idaapi.op_t) -> int:
     """mask value by data type
 
     necessary due to a bug in AMD64
@@ -306,26 +330,15 @@ def mask_op_val(op):
     return masks.get(op.dtype, op.value) & op.value
 
 
-def is_function_recursive(f):
-    """check if function is recursive
-
-    args:
-        f (IDA func_t)
-    """
-    for ref in idautils.CodeRefsTo(f.start_ea, True):
-        if f.contains(ref):
-            return True
-    return False
+def is_function_recursive(f: idaapi.func_t) -> bool:
+    """check if function is recursive"""
+    return any(f.contains(ref) for ref in idautils.CodeRefsTo(f.start_ea, True))
 
 
-def is_basic_block_tight_loop(bb):
+def is_basic_block_tight_loop(bb: idaapi.BasicBlock) -> bool:
     """check basic block loops to self
 
     true if last instruction in basic block branches to basic block start
-
-    args:
-        f (IDA func_t)
-        bb (IDA BasicBlock)
     """
     bb_end = idc.prev_head(bb.end_ea)
     if bb.start_ea < bb_end:
@@ -335,8 +348,8 @@ def is_basic_block_tight_loop(bb):
     return False
 
 
-def find_data_reference_from_insn(insn, max_depth=10):
-    """ search for data reference from instruction, return address of instruction if no reference exists """
+def find_data_reference_from_insn(insn: idaapi.insn_t, max_depth: int = 10) -> int:
+    """search for data reference from instruction, return address of instruction if no reference exists"""
     depth = 0
     ea = insn.ea
 
@@ -365,19 +378,17 @@ def find_data_reference_from_insn(insn, max_depth=10):
     return ea
 
 
-def get_function_blocks(f):
-    """yield basic blocks contained in specified function
-
-    args:
-        f (IDA func_t)
-    yield:
-        block (IDA BasicBlock)
-    """
+def get_function_blocks(f: idaapi.func_t) -> Iterator[idaapi.BasicBlock]:
+    """yield basic blocks contained in specified function"""
     # leverage idaapi.FC_NOEXT flag to ignore useless external blocks referenced by the function
-    for block in idaapi.FlowChart(f, flags=(idaapi.FC_PREDS | idaapi.FC_NOEXT)):
-        yield block
+    yield from idaapi.FlowChart(f, flags=(idaapi.FC_PREDS | idaapi.FC_NOEXT))
 
 
-def is_basic_block_return(bb):
-    """ check if basic block is return block """
+def is_basic_block_return(bb: idaapi.BasicBlock) -> bool:
+    """check if basic block is return block"""
     return bb.type == idaapi.fcb_ret
+
+
+def has_sib(oper: idaapi.op_t) -> bool:
+    # via: https://reverseengineering.stackexchange.com/a/14300
+    return oper.specflag1 == 1

capa/features/extractors/ida/insn.py

@@ -1,10 +1,11 @@
-# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
 # Licensed under the Apache License, Version 2.0 (the "License");
 #  you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at: [package root]/LICENSE.txt
 # Unless required by applicable law or agreed to in writing, software distributed under the License
 #  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and limitations under the License.
+from typing import Any, Dict, Tuple, Iterator
 
 import idc
 import idaapi
@@ -12,51 +13,30 @@ import idautils
 
 import capa.features.extractors.helpers
 import capa.features.extractors.ida.helpers
-from capa.features import (
-    ARCH_X32,
-    ARCH_X64,
-    MAX_BYTES_FEATURE_SIZE,
-    THUNK_CHAIN_DEPTH_DELTA,
-    Bytes,
-    String,
-    Characteristic,
-)
-from capa.features.insn import API, Number, Offset, Mnemonic
+from capa.features.insn import API, MAX_STRUCTURE_SIZE, Number, Offset, Mnemonic, OperandNumber, OperandOffset
+from capa.features.common import MAX_BYTES_FEATURE_SIZE, THUNK_CHAIN_DEPTH_DELTA, Bytes, String, Feature, Characteristic
+from capa.features.address import Address, AbsoluteVirtualAddress
+from capa.features.extractors.base_extractor import BBHandle, InsnHandle, FunctionHandle
 
 # security cookie checks may perform non-zeroing XORs, these are expected within a certain
 # byte range within the first and returning basic blocks, this helps to reduce FP features
 SECURITY_COOKIE_BYTES_DELTA = 0x40
 
 
-def get_arch(ctx):
-    """
-    fetch the ARCH_* constant for the currently open workspace.
-
-    via Tamir Bahar/@tmr232
-    https://reverseengineering.stackexchange.com/a/11398/17194
-    """
-    if "arch" not in ctx:
-        info = idaapi.get_inf_structure()
-        if info.is_64bit():
-            ctx["arch"] = ARCH_X64
-        elif info.is_32bit():
-            ctx["arch"] = ARCH_X32
-        else:
-            raise ValueError("unexpected architecture")
-    return ctx["arch"]
-
-
-def get_imports(ctx):
+def get_imports(ctx: Dict[str, Any]) -> Dict[int, Any]:
     if "imports_cache" not in ctx:
         ctx["imports_cache"] = capa.features.extractors.ida.helpers.get_file_imports()
     return ctx["imports_cache"]
 
 
-def check_for_api_call(ctx, insn):
-    """ check instruction for API call """
-    if not insn.get_canon_mnem() in ("call", "jmp"):
-        return
+def get_externs(ctx: Dict[str, Any]) -> Dict[int, Any]:
+    if "externs_cache" not in ctx:
+        ctx["externs_cache"] = capa.features.extractors.ida.helpers.get_file_externs()
+    return ctx["externs_cache"]
 
+
+def check_for_api_call(insn: idaapi.insn_t, funcs: Dict[int, Any]) -> Iterator[Any]:
+    """check instruction for API call"""
     info = ()
     ref = insn.ea
 
@@ -72,7 +52,7 @@ def check_for_api_call(ctx, insn):
             except IndexError:
                 break
 
-        info = get_imports(ctx).get(ref, ())
+        info = funcs.get(ref, ())
         if info:
             break
 
@@ -81,37 +61,64 @@ def check_for_api_call(ctx, insn):
             break
 
     if info:
-        yield "%s.%s" % (info[0], info[1])
+        yield info
 
 
-def extract_insn_api_features(f, bb, insn):
-    """parse instruction API features
-
-    args:
-        f (IDA func_t)
-        bb (IDA BasicBlock)
-        insn (IDA insn_t)
+def extract_insn_api_features(fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
+    """
+    parse instruction API features
 
     example:
-        call dword [0x00473038]
+       call dword [0x00473038]
     """
-    for api in check_for_api_call(f.ctx, insn):
-        dll, _, symbol = api.rpartition(".")
-        for name in capa.features.extractors.helpers.generate_symbols(dll, symbol):
-            yield API(name), insn.ea
+    insn: idaapi.insn_t = ih.inner
+
+    if insn.get_canon_mnem() not in ("call", "jmp"):
+        return
+
+    # check calls to imported functions
+    for api in check_for_api_call(insn, get_imports(fh.ctx)):
+        # tuple (<module>, <function>, <ordinal>)
+        for name in capa.features.extractors.helpers.generate_symbols(api[0], api[1]):
+            yield API(name), ih.address
+
+    # check calls to extern functions
+    for api in check_for_api_call(insn, get_externs(fh.ctx)):
+        # tuple (<module>, <function>, <ordinal>)
+        yield API(api[1]), ih.address
+
+    # extract IDA/FLIRT recognized API functions
+    targets = tuple(idautils.CodeRefsFrom(insn.ea, False))
+    if not targets:
+        return
+
+    target = targets[0]
+    target_func = idaapi.get_func(target)
+    if not target_func or target_func.start_ea != target:
+        # not a function (start)
+        return
+
+    if target_func.flags & idaapi.FUNC_LIB:
+        name = idaapi.get_name(target_func.start_ea)
+        yield API(name), ih.address
+        if name.startswith("_"):
+            # some linkers may prefix linked routines with a `_` to avoid name collisions.
+            # extract features for both the mangled and un-mangled representations.
+            # e.g. `_fwrite` -> `fwrite`
+            # see: https://stackoverflow.com/a/2628384/87207
+            yield API(name[1:]), ih.address
 
 
-def extract_insn_number_features(f, bb, insn):
-    """parse instruction number features
-
-    args:
-        f (IDA func_t)
-        bb (IDA BasicBlock)
-        insn (IDA insn_t)
-
+def extract_insn_number_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    """
+    parse instruction number features
     example:
         push    3136B0h         ; dwControlCode
     """
+    insn: idaapi.insn_t = ih.inner
+
     if idaapi.is_ret_insn(insn):
         # skip things like:
         #   .text:0042250E retn 8
@@ -122,7 +129,11 @@ def extract_insn_number_features(f, bb, insn):
         #   .text:00401145 add esp, 0Ch
         return
 
-    for op in capa.features.extractors.ida.helpers.get_insn_ops(insn, target_ops=(idaapi.o_imm, idaapi.o_mem)):
+    for i, op in enumerate(insn.ops):
+        if op.type == idaapi.o_void:
+            break
+        if op.type not in (idaapi.o_imm, idaapi.o_mem):
+            continue
         # skip things like:
         #   .text:00401100 shr eax, offset loc_C
         if capa.features.extractors.ida.helpers.is_op_offset(insn, op):
@@ -133,21 +144,27 @@ def extract_insn_number_features(f, bb, insn):
         else:
             const = op.addr
 
-        yield Number(const), insn.ea
-        yield Number(const, arch=get_arch(f.ctx)), insn.ea
+        yield Number(const), ih.address
+        yield OperandNumber(i, const), ih.address
+
+        if insn.itype == idaapi.NN_add and 0 < const < MAX_STRUCTURE_SIZE and op.type == idaapi.o_imm:
+            # for pattern like:
+            #
+            #     add eax, 0x10
+            #
+            # assume 0x10 is also an offset (imagine eax is a pointer).
+            yield Offset(const), ih.address
+            yield OperandOffset(i, const), ih.address
 
 
-def extract_insn_bytes_features(f, bb, insn):
-    """parse referenced byte sequences
-
-    args:
-        f (IDA func_t)
-        bb (IDA BasicBlock)
-        insn (IDA insn_t)
-
+def extract_insn_bytes_features(fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
+    """
+    parse referenced byte sequences
     example:
         push    offset iid_004118d4_IShellLinkA ; riid
     """
+    insn: idaapi.insn_t = ih.inner
+
     if idaapi.is_call_insn(insn):
         return
 
@@ -155,43 +172,54 @@ def extract_insn_bytes_features(f, bb, insn):
     if ref != insn.ea:
         extracted_bytes = capa.features.extractors.ida.helpers.read_bytes_at(ref, MAX_BYTES_FEATURE_SIZE)
         if extracted_bytes and not capa.features.extractors.helpers.all_zeros(extracted_bytes):
-            yield Bytes(extracted_bytes), insn.ea
+            if not capa.features.extractors.ida.helpers.find_string_at(ref):
+                # don't extract byte features for obvious strings
+                yield Bytes(extracted_bytes), ih.address
 
 
-def extract_insn_string_features(f, bb, insn):
-    """parse instruction string features
-
-    args:
-        f (IDA func_t)
-        bb (IDA BasicBlock)
-        insn (IDA insn_t)
+def extract_insn_string_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    """
+    parse instruction string features
 
     example:
         push offset aAcr     ; "ACR  > "
     """
+    insn: idaapi.insn_t = ih.inner
+
     ref = capa.features.extractors.ida.helpers.find_data_reference_from_insn(insn)
     if ref != insn.ea:
         found = capa.features.extractors.ida.helpers.find_string_at(ref)
         if found:
-            yield String(found), insn.ea
+            yield String(found), ih.address
 
 
-def extract_insn_offset_features(f, bb, insn):
-    """parse instruction structure offset features
-
-    args:
-        f (IDA func_t)
-        bb (IDA BasicBlock)
-        insn (IDA insn_t)
+def extract_insn_offset_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    """
+    parse instruction structure offset features
 
     example:
         .text:0040112F cmp [esi+4], ebx
     """
-    for op in capa.features.extractors.ida.helpers.get_insn_ops(insn, target_ops=(idaapi.o_phrase, idaapi.o_displ)):
+    insn: idaapi.insn_t = ih.inner
+
+    for i, op in enumerate(insn.ops):
+        if op.type == idaapi.o_void:
+            break
+        if op.type not in (idaapi.o_phrase, idaapi.o_displ):
+            continue
         if capa.features.extractors.ida.helpers.is_op_stack_var(insn.ea, op.n):
             continue
+
         p_info = capa.features.extractors.ida.helpers.get_op_phrase_info(op)
-        op_off = p_info.get("offset", 0)
+
+        op_off = p_info.get("offset")
+        if op_off is None:
+            continue
+
         if idaapi.is_mapped(op_off):
             # Ignore:
             #   mov esi, dword_1005B148[esi]
@@ -202,12 +230,32 @@ def extract_insn_offset_features(f, bb, insn):
         # https://stackoverflow.com/questions/31853189/x86-64-assembly-why-displacement-not-64-bits
         op_off = capa.features.extractors.helpers.twos_complement(op_off, 32)
 
-        yield Offset(op_off), insn.ea
-        yield Offset(op_off, arch=get_arch(f.ctx)), insn.ea
+        yield Offset(op_off), ih.address
+        yield OperandOffset(i, op_off), ih.address
+
+        if (
+            insn.itype == idaapi.NN_lea
+            and i == 1
+            # o_displ is used for both:
+            #   [eax+1]
+            #   [eax+ebx+2]
+            and op.type == idaapi.o_displ
+            # but the SIB is only present for [eax+ebx+2]
+            # which we don't want
+            and not capa.features.extractors.ida.helpers.has_sib(op)
+        ):
+            # for pattern like:
+            #
+            #     lea eax, [ebx + 1]
+            #
+            # assume 1 is also an offset (imagine ebx is a zero register).
+            yield Number(op_off), ih.address
+            yield OperandNumber(i, op_off), ih.address
 
 
-def contains_stack_cookie_keywords(s):
-    """check if string contains stack cookie keywords
+def contains_stack_cookie_keywords(s: str) -> bool:
+    """
+    check if string contains stack cookie keywords
 
     Examples:
         xor     ecx, ebp ; StackCookie
@@ -221,7 +269,7 @@ def contains_stack_cookie_keywords(s):
     return any(keyword in s for keyword in ("stack", "security"))
 
 
-def bb_stack_cookie_registers(bb):
+def bb_stack_cookie_registers(bb: idaapi.BasicBlock) -> Iterator[int]:
     """scan basic block for stack cookie operations
 
     yield registers ids that may have been used for stack cookie operations
@@ -255,8 +303,8 @@ def bb_stack_cookie_registers(bb):
                     yield op.reg
 
 
-def is_nzxor_stack_cookie_delta(f, bb, insn):
-    """ check if nzxor exists within stack cookie delta """
+def is_nzxor_stack_cookie_delta(f: idaapi.func_t, bb: idaapi.BasicBlock, insn: idaapi.insn_t) -> bool:
+    """check if nzxor exists within stack cookie delta"""
     # security cookie check should use SP or BP
     if not capa.features.extractors.ida.helpers.is_frame_register(insn.Op2.reg):
         return False
@@ -278,8 +326,8 @@ def is_nzxor_stack_cookie_delta(f, bb, insn):
     return False
 
 
-def is_nzxor_stack_cookie(f, bb, insn):
-    """ check if nzxor is related to stack cookie """
+def is_nzxor_stack_cookie(f: idaapi.func_t, bb: idaapi.BasicBlock, insn: idaapi.insn_t) -> bool:
+    """check if nzxor is related to stack cookie"""
     if contains_stack_cookie_keywords(idaapi.get_cmt(insn.ea, False)):
         # Example:
         #   xor     ecx, ebp        ; StackCookie
@@ -295,37 +343,49 @@ def is_nzxor_stack_cookie(f, bb, insn):
     return False
 
 
-def extract_insn_nzxor_characteristic_features(f, bb, insn):
-    """parse instruction non-zeroing XOR instruction
-
-    ignore expected non-zeroing XORs, e.g. security cookies
-
-    args:
-        f (IDA func_t)
-        bb (IDA BasicBlock)
-        insn (IDA insn_t)
+def extract_insn_nzxor_characteristic_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
     """
+    parse instruction non-zeroing XOR instruction
+    ignore expected non-zeroing XORs, e.g. security cookies
+    """
+    insn: idaapi.insn_t = ih.inner
+
     if insn.itype not in (idaapi.NN_xor, idaapi.NN_xorpd, idaapi.NN_xorps, idaapi.NN_pxor):
         return
     if capa.features.extractors.ida.helpers.is_operand_equal(insn.Op1, insn.Op2):
         return
-    if is_nzxor_stack_cookie(f, bb, insn):
+    if is_nzxor_stack_cookie(fh.inner, bbh.inner, insn):
         return
-    yield Characteristic("nzxor"), insn.ea
+    yield Characteristic("nzxor"), ih.address
 
 
-def extract_insn_mnemonic_features(f, bb, insn):
-    """parse instruction mnemonic features
+def extract_insn_mnemonic_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    """parse instruction mnemonic features"""
+    yield Mnemonic(idc.print_insn_mnem(ih.inner.ea)), ih.address
 
-    args:
-        f (IDA func_t)
-        bb (IDA BasicBlock)
-        insn (IDA insn_t)
+
+def extract_insn_obfs_call_plus_5_characteristic_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
     """
-    yield Mnemonic(insn.get_canon_mnem()), insn.ea
+    parse call $+5 instruction from the given instruction.
+    """
+    insn: idaapi.insn_t = ih.inner
+
+    if not idaapi.is_call_insn(insn):
+        return
+
+    if insn.ea + 5 == idc.get_operand_value(insn.ea, 0):
+        yield Characteristic("call $+5"), ih.address
 
 
-def extract_insn_peb_access_characteristic_features(f, bb, insn):
+def extract_insn_peb_access_characteristic_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
     """parse instruction peb access
 
     fs:[0x30] on x86, gs:[0x60] on x64
@@ -333,51 +393,61 @@ def extract_insn_peb_access_characteristic_features(f, bb, insn):
     TODO:
         IDA should be able to do this..
     """
+    insn: idaapi.insn_t = ih.inner
+
     if insn.itype not in (idaapi.NN_push, idaapi.NN_mov):
         return
 
-    if all(map(lambda op: op.type != idaapi.o_mem, insn.ops)):
+    if all(op.type != idaapi.o_mem for op in insn.ops):
         # try to optimize for only memory references
         return
 
     disasm = idc.GetDisasm(insn.ea)
 
     if " fs:30h" in disasm or " gs:60h" in disasm:
-        # TODO: replace above with proper IDA
-        yield Characteristic("peb access"), insn.ea
+        # TODO(mike-hunhoff): use proper IDA API for fetching segment access
+        # scanning the disassembly text is a hack.
+        # https://github.com/mandiant/capa/issues/1605
+        yield Characteristic("peb access"), ih.address
 
 
-def extract_insn_segment_access_features(f, bb, insn):
+def extract_insn_segment_access_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
     """parse instruction fs or gs access
 
     TODO:
         IDA should be able to do this...
     """
-    if all(map(lambda op: op.type != idaapi.o_mem, insn.ops)):
+    insn: idaapi.insn_t = ih.inner
+
+    if all(op.type != idaapi.o_mem for op in insn.ops):
         # try to optimize for only memory references
         return
 
     disasm = idc.GetDisasm(insn.ea)
 
     if " fs:" in disasm:
-        # TODO: replace above with proper IDA
-        yield Characteristic("fs access"), insn.ea
+        # TODO(mike-hunhoff): use proper IDA API for fetching segment access
+        # scanning the disassembly text is a hack.
+        # https://github.com/mandiant/capa/issues/1605
+        yield Characteristic("fs access"), ih.address
 
     if " gs:" in disasm:
-        # TODO: replace above with proper IDA
-        yield Characteristic("gs access"), insn.ea
+        # TODO(mike-hunhoff): use proper IDA API for fetching segment access
+        # scanning the disassembly text is a hack.
+        # https://github.com/mandiant/capa/issues/1605
+        yield Characteristic("gs access"), ih.address
 
 
-def extract_insn_cross_section_cflow(f, bb, insn):
-    """inspect the instruction for a CALL or JMP that crosses section boundaries
+def extract_insn_cross_section_cflow(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
+    """inspect the instruction for a CALL or JMP that crosses section boundaries"""
+    insn: idaapi.insn_t = ih.inner
 
-    args:
-        f (IDA func_t)
-        bb (IDA BasicBlock)
-        insn (IDA insn_t)
-    """
     for ref in idautils.CodeRefsFrom(insn.ea, False):
-        if ref in get_imports(f.ctx).keys():
+        if ref in get_imports(fh.ctx):
             # ignore API calls
             continue
         if not idaapi.getseg(ref):
@@ -385,50 +455,40 @@ def extract_insn_cross_section_cflow(f, bb, insn):
             continue
         if idaapi.getseg(ref) == idaapi.getseg(insn.ea):
             continue
-        yield Characteristic("cross section flow"), insn.ea
+        yield Characteristic("cross section flow"), ih.address
 
 
-def extract_function_calls_from(f, bb, insn):
+def extract_function_calls_from(fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
     """extract functions calls from features
 
     most relevant at the function scope, however, its most efficient to extract at the instruction scope
-
-    args:
-        f (IDA func_t)
-        bb (IDA BasicBlock)
-        insn (IDA insn_t)
     """
+    insn: idaapi.insn_t = ih.inner
+
     if idaapi.is_call_insn(insn):
         for ref in idautils.CodeRefsFrom(insn.ea, False):
-            yield Characteristic("calls from"), ref
+            yield Characteristic("calls from"), AbsoluteVirtualAddress(ref)
 
 
-def extract_function_indirect_call_characteristic_features(f, bb, insn):
+def extract_function_indirect_call_characteristic_features(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
     """extract indirect function calls (e.g., call eax or call dword ptr [edx+4])
     does not include calls like => call ds:dword_ABD4974
 
     most relevant at the function or basic block scope;
     however, its most efficient to extract at the instruction scope
-
-    args:
-        f (IDA func_t)
-        bb (IDA BasicBlock)
-        insn (IDA insn_t)
     """
+    insn: idaapi.insn_t = ih.inner
+
     if idaapi.is_call_insn(insn) and idc.get_operand_type(insn.ea, 0) in (idc.o_reg, idc.o_phrase, idc.o_displ):
-        yield Characteristic("indirect call"), insn.ea
+        yield Characteristic("indirect call"), ih.address
 
 
-def extract_features(f, bb, insn):
-    """extract instruction features
-
-    args:
-        f (IDA func_t)
-        bb (IDA BasicBlock)
-        insn (IDA insn_t)
-    """
+def extract_features(f: FunctionHandle, bbh: BBHandle, insn: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
+    """extract instruction features"""
     for inst_handler in INSTRUCTION_HANDLERS:
-        for (feature, ea) in inst_handler(f, bb, insn):
+        for feature, ea in inst_handler(f, bbh, insn):
             yield feature, ea
 
 
@@ -440,26 +500,10 @@ INSTRUCTION_HANDLERS = (
     extract_insn_offset_features,
     extract_insn_nzxor_characteristic_features,
     extract_insn_mnemonic_features,
+    extract_insn_obfs_call_plus_5_characteristic_features,
     extract_insn_peb_access_characteristic_features,
     extract_insn_cross_section_cflow,
     extract_insn_segment_access_features,
     extract_function_calls_from,
     extract_function_indirect_call_characteristic_features,
 )
-
-
-def main():
-    """ """
-    features = []
-    for f in capa.features.extractors.ida.helpers.get_functions(skip_thunks=True, skip_libs=True):
-        for bb in idaapi.FlowChart(f, flags=idaapi.FC_PREDS):
-            for insn in capa.features.extractors.ida.helpers.get_instructions_in_range(bb.start_ea, bb.end_ea):
-                features.extend(list(extract_features(f, bb, insn)))
-
-    import pprint
-
-    pprint.pprint(features)
-
-
-if __name__ == "__main__":
-    main()

capa/features/extractors/loops.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
 # Licensed under the Apache License, Version 2.0 (the "License");
 #  you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at: [package root]/LICENSE.txt
@@ -6,7 +6,7 @@
 #  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and limitations under the License.
 
-from networkx import nx
+import networkx
 from networkx.algorithms.components import strongly_connected_components
 
 
@@ -20,6 +20,6 @@ def has_loop(edges, threshold=2):
     returns:
         bool
     """
-    g = nx.DiGraph()
+    g = networkx.DiGraph()
     g.add_edges_from(edges)
     return any(len(comp) >= threshold for comp in strongly_connected_components(g))

capa/features/extractors/null.py

@@ -0,0 +1,79 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+from typing import Dict, List, Tuple
+from dataclasses import dataclass
+
+from capa.features.common import Feature
+from capa.features.address import NO_ADDRESS, Address
+from capa.features.extractors.base_extractor import BBHandle, InsnHandle, FunctionHandle, FeatureExtractor
+
+
+@dataclass
+class InstructionFeatures:
+    features: List[Tuple[Address, Feature]]
+
+
+@dataclass
+class BasicBlockFeatures:
+    features: List[Tuple[Address, Feature]]
+    instructions: Dict[Address, InstructionFeatures]
+
+
+@dataclass
+class FunctionFeatures:
+    features: List[Tuple[Address, Feature]]
+    basic_blocks: Dict[Address, BasicBlockFeatures]
+
+
+@dataclass
+class NullFeatureExtractor(FeatureExtractor):
+    """
+    An extractor that extracts some user-provided features.
+
+    This is useful for testing, as we can provide expected values and see if matching works.
+    """
+
+    base_address: Address
+    global_features: List[Feature]
+    file_features: List[Tuple[Address, Feature]]
+    functions: Dict[Address, FunctionFeatures]
+
+    def get_base_address(self):
+        return self.base_address
+
+    def extract_global_features(self):
+        for feature in self.global_features:
+            yield feature, NO_ADDRESS
+
+    def extract_file_features(self):
+        for address, feature in self.file_features:
+            yield feature, address
+
+    def get_functions(self):
+        for address in sorted(self.functions.keys()):
+            yield FunctionHandle(address, None)
+
+    def extract_function_features(self, f):
+        for address, feature in self.functions[f.address].features:
+            yield feature, address
+
+    def get_basic_blocks(self, f):
+        for address in sorted(self.functions[f.address].basic_blocks.keys()):
+            yield BBHandle(address, None)
+
+    def extract_basic_block_features(self, f, bb):
+        for address, feature in self.functions[f.address].basic_blocks[bb.address].features:
+            yield feature, address
+
+    def get_instructions(self, f, bb):
+        for address in sorted(self.functions[f.address].basic_blocks[bb.address].instructions.keys()):
+            yield InsnHandle(address, None)
+
+    def extract_insn_features(self, f, bb, insn):
+        for address, feature in self.functions[f.address].basic_blocks[bb.address].instructions[insn.address].features:
+            yield feature, address

capa/features/extractors/pefile.py

@@ -0,0 +1,229 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+
+import logging
+from pathlib import Path
+
+import pefile
+
+import capa.features.common
+import capa.features.extractors
+import capa.features.extractors.common
+import capa.features.extractors.helpers
+import capa.features.extractors.strings
+from capa.features.file import Export, Import, Section
+from capa.features.common import OS, ARCH_I386, FORMAT_PE, ARCH_AMD64, OS_WINDOWS, Arch, Format, Characteristic
+from capa.features.address import NO_ADDRESS, FileOffsetAddress, AbsoluteVirtualAddress
+from capa.features.extractors.base_extractor import FeatureExtractor
+
+logger = logging.getLogger(__name__)
+
+
+def extract_file_embedded_pe(buf, **kwargs):
+    for offset, _ in capa.features.extractors.helpers.carve_pe(buf, 1):
+        yield Characteristic("embedded pe"), FileOffsetAddress(offset)
+
+
+def extract_file_export_names(pe, **kwargs):
+    base_address = pe.OPTIONAL_HEADER.ImageBase
+
+    if hasattr(pe, "DIRECTORY_ENTRY_EXPORT"):
+        for export in pe.DIRECTORY_ENTRY_EXPORT.symbols:
+            if not export.name:
+                continue
+            try:
+                name = export.name.partition(b"\x00")[0].decode("ascii")
+            except UnicodeDecodeError:
+                continue
+
+            if export.forwarder is None:
+                va = base_address + export.address
+                yield Export(name), AbsoluteVirtualAddress(va)
+
+            else:
+                try:
+                    forwarded_name = export.forwarder.partition(b"\x00")[0].decode("ascii")
+                except UnicodeDecodeError:
+                    continue
+                forwarded_name = capa.features.extractors.helpers.reformat_forwarded_export_name(forwarded_name)
+                va = base_address + export.address
+                yield Export(forwarded_name), AbsoluteVirtualAddress(va)
+                yield Characteristic("forwarded export"), AbsoluteVirtualAddress(va)
+
+
+def extract_file_import_names(pe, **kwargs):
+    """
+    extract imported function names
+    1. imports by ordinal:
+     - modulename.#ordinal
+    2. imports by name, results in two features to support importname-only matching:
+     - modulename.importname
+     - importname
+    """
+    if hasattr(pe, "DIRECTORY_ENTRY_IMPORT"):
+        for dll in pe.DIRECTORY_ENTRY_IMPORT:
+            try:
+                modname = dll.dll.partition(b"\x00")[0].decode("ascii")
+            except UnicodeDecodeError:
+                continue
+
+            # strip extension
+            modname = modname.rpartition(".")[0].lower()
+
+            for imp in dll.imports:
+                if imp.import_by_ordinal:
+                    impname = f"#{imp.ordinal}"
+                else:
+                    try:
+                        impname = imp.name.partition(b"\x00")[0].decode("ascii")
+                    except UnicodeDecodeError:
+                        continue
+
+                for name in capa.features.extractors.helpers.generate_symbols(modname, impname):
+                    yield Import(name), AbsoluteVirtualAddress(imp.address)
+
+
+def extract_file_section_names(pe, **kwargs):
+    base_address = pe.OPTIONAL_HEADER.ImageBase
+
+    for section in pe.sections:
+        try:
+            name = section.Name.partition(b"\x00")[0].decode("ascii")
+        except UnicodeDecodeError:
+            continue
+
+        yield Section(name), AbsoluteVirtualAddress(base_address + section.VirtualAddress)
+
+
+def extract_file_strings(buf, **kwargs):
+    yield from capa.features.extractors.common.extract_file_strings(buf)
+
+
+def extract_file_function_names(**kwargs):
+    """
+    extract the names of statically-linked library functions.
+    """
+    if False:
+        # using a `yield` here to force this to be a generator, not function.
+        yield NotImplementedError("pefile doesn't have library matching")
+    return
+
+
+def extract_file_os(**kwargs):
+    # assuming PE -> Windows
+    # though i suppose they're also used by UEFI
+    yield OS(OS_WINDOWS), NO_ADDRESS
+
+
+def extract_file_format(**kwargs):
+    yield Format(FORMAT_PE), NO_ADDRESS
+
+
+def extract_file_arch(pe, **kwargs):
+    if pe.FILE_HEADER.Machine == pefile.MACHINE_TYPE["IMAGE_FILE_MACHINE_I386"]:
+        yield Arch(ARCH_I386), NO_ADDRESS
+    elif pe.FILE_HEADER.Machine == pefile.MACHINE_TYPE["IMAGE_FILE_MACHINE_AMD64"]:
+        yield Arch(ARCH_AMD64), NO_ADDRESS
+    else:
+        logger.warning("unsupported architecture: %s", pefile.MACHINE_TYPE[pe.FILE_HEADER.Machine])
+
+
+def extract_file_features(pe, buf):
+    """
+    extract file features from given workspace
+
+    args:
+      pe (pefile.PE): the parsed PE
+      buf: the raw sample bytes
+
+    yields:
+      Tuple[Feature, VA]: a feature and its location.
+    """
+
+    for file_handler in FILE_HANDLERS:
+        # file_handler: type: (pe, bytes) -> Iterable[Tuple[Feature, Address]]
+        for feature, va in file_handler(pe=pe, buf=buf):  # type: ignore
+            yield feature, va
+
+
+FILE_HANDLERS = (
+    extract_file_embedded_pe,
+    extract_file_export_names,
+    extract_file_import_names,
+    extract_file_section_names,
+    extract_file_strings,
+    extract_file_function_names,
+    extract_file_format,
+)
+
+
+def extract_global_features(pe, buf):
+    """
+    extract global features from given workspace
+
+    args:
+      pe (pefile.PE): the parsed PE
+      buf: the raw sample bytes
+
+    yields:
+      Tuple[Feature, VA]: a feature and its location.
+    """
+    for handler in GLOBAL_HANDLERS:
+        # file_handler: type: (pe, bytes) -> Iterable[Tuple[Feature, Address]]
+        for feature, va in handler(pe=pe, buf=buf):  # type: ignore
+            yield feature, va
+
+
+GLOBAL_HANDLERS = (
+    extract_file_os,
+    extract_file_arch,
+)
+
+
+class PefileFeatureExtractor(FeatureExtractor):
+    def __init__(self, path: Path):
+        super().__init__()
+        self.path: Path = path
+        self.pe = pefile.PE(str(path))
+
+    def get_base_address(self):
+        return AbsoluteVirtualAddress(self.pe.OPTIONAL_HEADER.ImageBase)
+
+    def extract_global_features(self):
+        buf = Path(self.path).read_bytes()
+
+        yield from extract_global_features(self.pe, buf)
+
+    def extract_file_features(self):
+        buf = Path(self.path).read_bytes()
+
+        yield from extract_file_features(self.pe, buf)
+
+    def get_functions(self):
+        raise NotImplementedError("PefileFeatureExtract can only be used to extract file features")
+
+    def extract_function_features(self, f):
+        raise NotImplementedError("PefileFeatureExtract can only be used to extract file features")
+
+    def get_basic_blocks(self, f):
+        raise NotImplementedError("PefileFeatureExtract can only be used to extract file features")
+
+    def extract_basic_block_features(self, f, bb):
+        raise NotImplementedError("PefileFeatureExtract can only be used to extract file features")
+
+    def get_instructions(self, f, bb):
+        raise NotImplementedError("PefileFeatureExtract can only be used to extract file features")
+
+    def extract_insn_features(self, f, bb, insn):
+        raise NotImplementedError("PefileFeatureExtract can only be used to extract file features")
+
+    def is_library_function(self, va):
+        raise NotImplementedError("PefileFeatureExtract can only be used to extract file features")
+
+    def get_function_name(self, va):
+        raise NotImplementedError("PefileFeatureExtract can only be used to extract file features")

capa/features/extractors/smda/__init__.py

@@ -1,52 +0,0 @@
-import sys
-import types
-
-from smda.common.SmdaReport import SmdaReport
-from smda.common.SmdaInstruction import SmdaInstruction
-
-import capa.features.extractors.smda.file
-import capa.features.extractors.smda.insn
-import capa.features.extractors.smda.function
-import capa.features.extractors.smda.basicblock
-from capa.main import UnsupportedRuntimeError
-from capa.features.extractors import FeatureExtractor
-
-
-class SmdaFeatureExtractor(FeatureExtractor):
-    def __init__(self, smda_report: SmdaReport, path):
-        super(SmdaFeatureExtractor, self).__init__()
-        if sys.version_info < (3, 0):
-            raise UnsupportedRuntimeError("SMDA should only be used with Python 3.")
-        self.smda_report = smda_report
-        self.path = path
-
-    def get_base_address(self):
-        return self.smda_report.base_addr
-
-    def extract_file_features(self):
-        for feature, va in capa.features.extractors.smda.file.extract_features(self.smda_report, self.path):
-            yield feature, va
-
-    def get_functions(self):
-        for function in self.smda_report.getFunctions():
-            yield function
-
-    def extract_function_features(self, f):
-        for feature, va in capa.features.extractors.smda.function.extract_features(f):
-            yield feature, va
-
-    def get_basic_blocks(self, f):
-        for bb in f.getBlocks():
-            yield bb
-
-    def extract_basic_block_features(self, f, bb):
-        for feature, va in capa.features.extractors.smda.basicblock.extract_features(f, bb):
-            yield feature, va
-
-    def get_instructions(self, f, bb):
-        for smda_ins in bb.getInstructions():
-            yield smda_ins
-
-    def extract_insn_features(self, f, bb, insn):
-        for feature, va in capa.features.extractors.smda.insn.extract_features(f, bb, insn):
-            yield feature, va

capa/features/extractors/smda/basicblock.py

@@ -1,131 +0,0 @@
-import sys
-import string
-import struct
-
-from capa.features import Characteristic
-from capa.features.basicblock import BasicBlock
-from capa.features.extractors.helpers import MIN_STACKSTRING_LEN
-
-
-def _bb_has_tight_loop(f, bb):
-    """
-    parse tight loops, true if last instruction in basic block branches to bb start
-    """
-    return bb.offset in f.blockrefs[bb.offset] if bb.offset in f.blockrefs else False
-
-
-def extract_bb_tight_loop(f, bb):
-    """ check basic block for tight loop indicators """
-    if _bb_has_tight_loop(f, bb):
-        yield Characteristic("tight loop"), bb.offset
-
-
-def _bb_has_stackstring(f, bb):
-    """
-    extract potential stackstring creation, using the following heuristics:
-      - basic block contains enough moves of constant bytes to the stack
-    """
-    count = 0
-    for instr in bb.getInstructions():
-        if is_mov_imm_to_stack(instr):
-            count += get_printable_len(instr.getDetailed())
-        if count > MIN_STACKSTRING_LEN:
-            return True
-    return False
-
-
-def get_operands(smda_ins):
-    return [o.strip() for o in smda_ins.operands.split(",")]
-
-
-def extract_stackstring(f, bb):
-    """ check basic block for stackstring indicators """
-    if _bb_has_stackstring(f, bb):
-        yield Characteristic("stack string"), bb.offset
-
-
-def is_mov_imm_to_stack(smda_ins):
-    """
-    Return if instruction moves immediate onto stack
-    """
-    if not smda_ins.mnemonic.startswith("mov"):
-        return False
-
-    try:
-        dst, src = get_operands(smda_ins)
-    except ValueError:
-        # not two operands
-        return False
-
-    try:
-        int(src, 16)
-    except ValueError:
-        return False
-
-    if not any(regname in dst for regname in ["ebp", "rbp", "esp", "rsp"]):
-        return False
-
-    return True
-
-
-def is_printable_ascii(chars):
-    return all(c < 127 and chr(c) in string.printable for c in chars)
-
-
-def is_printable_utf16le(chars):
-    if all(c == 0x00 for c in chars[1::2]):
-        return is_printable_ascii(chars[::2])
-
-
-def get_printable_len(instr):
-    """
-    Return string length if all operand bytes are ascii or utf16-le printable
-
-    Works on a capstone instruction
-    """
-    # should have exactly two operands for mov immediate
-    if len(instr.operands) != 2:
-        return 0
-
-    op_value = instr.operands[1].value.imm
-
-    if instr.imm_size == 1:
-        chars = struct.pack("<B", op_value & 0xFF)
-    elif instr.imm_size == 2:
-        chars = struct.pack("<H", op_value & 0xFFFF)
-    elif instr.imm_size == 4:
-        chars = struct.pack("<I", op_value & 0xFFFFFFFF)
-    elif instr.imm_size == 8:
-        chars = struct.pack("<Q", op_value & 0xFFFFFFFFFFFFFFFF)
-    else:
-        raise ValueError("Unhandled operand data type 0x%x." % instr.imm_size)
-
-    if is_printable_ascii(chars):
-        return instr.imm_size
-    if is_printable_utf16le(chars):
-        return instr.imm_size // 2
-
-    return 0
-
-
-def extract_features(f, bb):
-    """
-    extract features from the given basic block.
-
-    args:
-      f (smda.common.SmdaFunction): the function from which to extract features
-      bb (smda.common.SmdaBasicBlock): the basic block to process.
-
-    yields:
-      Feature, set[VA]: the features and their location found in this basic block.
-    """
-    yield BasicBlock(), bb.offset
-    for bb_handler in BASIC_BLOCK_HANDLERS:
-        for feature, va in bb_handler(f, bb):
-            yield feature, va
-
-
-BASIC_BLOCK_HANDLERS = (
-    extract_bb_tight_loop,
-    extract_stackstring,
-)

capa/features/extractors/smda/file.py

@@ -1,139 +0,0 @@
-import struct
-
-# if we have SMDA we definitely have lief
-import lief
-
-import capa.features.extractors.helpers
-import capa.features.extractors.strings
-from capa.features import String, Characteristic
-from capa.features.file import Export, Import, Section
-
-
-def carve(pbytes, offset=0):
-    """
-    Return a list of (offset, size, xor) tuples of embedded PEs
-
-    Based on the version from vivisect:
-    https://github.com/vivisect/vivisect/blob/7be4037b1cecc4551b397f840405a1fc606f9b53/PE/carve.py#L19
-    And its IDA adaptation:
-    capa/features/extractors/ida/file.py
-    """
-    mz_xor = [
-        (
-            capa.features.extractors.helpers.xor_static(b"MZ", i),
-            capa.features.extractors.helpers.xor_static(b"PE", i),
-            i,
-        )
-        for i in range(256)
-    ]
-
-    pblen = len(pbytes)
-    todo = [(pbytes.find(mzx, offset), mzx, pex, i) for mzx, pex, i in mz_xor]
-    todo = [(off, mzx, pex, i) for (off, mzx, pex, i) in todo if off != -1]
-
-    while len(todo):
-
-        off, mzx, pex, i = todo.pop()
-
-        # The MZ header has one field we will check
-        # e_lfanew is at 0x3c
-        e_lfanew = off + 0x3C
-        if pblen < (e_lfanew + 4):
-            continue
-
-        newoff = struct.unpack("<I", capa.features.extractors.helpers.xor_static(pbytes[e_lfanew : e_lfanew + 4], i))[0]
-
-        nextres = pbytes.find(mzx, off + 1)
-        if nextres != -1:
-            todo.append((nextres, mzx, pex, i))
-
-        peoff = off + newoff
-        if pblen < (peoff + 2):
-            continue
-
-        if pbytes[peoff : peoff + 2] == pex:
-            yield (off, i)
-
-
-def extract_file_embedded_pe(smda_report, file_path):
-    with open(file_path, "rb") as f:
-        fbytes = f.read()
-
-    for offset, i in carve(fbytes, 1):
-        yield Characteristic("embedded pe"), offset
-
-
-def extract_file_export_names(smda_report, file_path):
-    lief_binary = lief.parse(file_path)
-    if lief_binary is not None:
-        for function in lief_binary.exported_functions:
-            yield Export(function.name), function.address
-
-
-def extract_file_import_names(smda_report, file_path):
-    # extract import table info via LIEF
-    lief_binary = lief.parse(file_path)
-    if not isinstance(lief_binary, lief.PE.Binary):
-        return
-    for imported_library in lief_binary.imports:
-        library_name = imported_library.name.lower()
-        library_name = library_name[:-4] if library_name.endswith(".dll") else library_name
-        for func in imported_library.entries:
-            if func.name:
-                va = func.iat_address + smda_report.base_addr
-                for name in capa.features.extractors.helpers.generate_symbols(library_name, func.name):
-                    yield Import(name), va
-            elif func.is_ordinal:
-                for name in capa.features.extractors.helpers.generate_symbols(library_name, "#%s" % func.ordinal):
-                    yield Import(name), va
-
-
-def extract_file_section_names(smda_report, file_path):
-    lief_binary = lief.parse(file_path)
-    if not isinstance(lief_binary, lief.PE.Binary):
-        return
-    if lief_binary and lief_binary.sections:
-        base_address = lief_binary.optional_header.imagebase
-        for section in lief_binary.sections:
-            yield Section(section.name), base_address + section.virtual_address
-
-
-def extract_file_strings(smda_report, file_path):
-    """
-    extract ASCII and UTF-16 LE strings from file
-    """
-    with open(file_path, "rb") as f:
-        b = f.read()
-
-    for s in capa.features.extractors.strings.extract_ascii_strings(b):
-        yield String(s.s), s.offset
-
-    for s in capa.features.extractors.strings.extract_unicode_strings(b):
-        yield String(s.s), s.offset
-
-
-def extract_features(smda_report, file_path):
-    """
-    extract file features from given workspace
-
-    args:
-      smda_report (smda.common.SmdaReport): a SmdaReport
-      file_path: path to the input file
-
-    yields:
-      Tuple[Feature, VA]: a feature and its location.
-    """
-
-    for file_handler in FILE_HANDLERS:
-        result = file_handler(smda_report, file_path)
-        for feature, va in file_handler(smda_report, file_path):
-            yield feature, va
-
-
-FILE_HANDLERS = (
-    extract_file_embedded_pe,
-    extract_file_export_names,
-    extract_file_import_names,
-    extract_file_section_names,
-    extract_file_strings,
-)

capa/features/extractors/smda/function.py

@@ -1,38 +0,0 @@
-from capa.features import Characteristic
-from capa.features.extractors import loops
-
-
-def extract_function_calls_to(f):
-    for inref in f.inrefs:
-        yield Characteristic("calls to"), inref
-
-
-def extract_function_loop(f):
-    """
-    parse if a function has a loop
-    """
-    edges = []
-    for bb_from, bb_tos in f.blockrefs.items():
-        for bb_to in bb_tos:
-            edges.append((bb_from, bb_to))
-
-    if edges and loops.has_loop(edges):
-        yield Characteristic("loop"), f.offset
-
-
-def extract_features(f):
-    """
-    extract features from the given function.
-
-    args:
-      f (smda.common.SmdaFunction): the function from which to extract features
-
-    yields:
-      Feature, set[VA]: the features and their location found in this function.
-    """
-    for func_handler in FUNCTION_HANDLERS:
-        for feature, va in func_handler(f):
-            yield feature, va
-
-
-FUNCTION_HANDLERS = (extract_function_calls_to, extract_function_loop)

capa/features/extractors/smda/insn.py

@@ -1,393 +0,0 @@
-import re
-import string
-import struct
-
-from smda.common.SmdaReport import SmdaReport
-
-import capa.features.extractors.helpers
-from capa.features import (
-    ARCH_X32,
-    ARCH_X64,
-    MAX_BYTES_FEATURE_SIZE,
-    THUNK_CHAIN_DEPTH_DELTA,
-    Bytes,
-    String,
-    Characteristic,
-)
-from capa.features.insn import API, Number, Offset, Mnemonic
-
-# security cookie checks may perform non-zeroing XORs, these are expected within a certain
-# byte range within the first and returning basic blocks, this helps to reduce FP features
-SECURITY_COOKIE_BYTES_DELTA = 0x40
-PATTERN_HEXNUM = re.compile(r"[+\-] (?P<num>0x[a-fA-F0-9]+)")
-PATTERN_SINGLENUM = re.compile(r"[+\-] (?P<num>[0-9])")
-
-
-def get_arch(smda_report):
-    if smda_report.architecture == "intel":
-        if smda_report.bitness == 32:
-            return ARCH_X32
-        elif smda_report.bitness == 64:
-            return ARCH_X64
-    else:
-        raise NotImplementedError
-
-
-def extract_insn_api_features(f, bb, insn):
-    """parse API features from the given instruction."""
-    if insn.offset in f.apirefs:
-        api_entry = f.apirefs[insn.offset]
-        # reformat
-        dll_name, api_name = api_entry.split("!")
-        dll_name = dll_name.split(".")[0]
-        dll_name = dll_name.lower()
-        for name in capa.features.extractors.helpers.generate_symbols(dll_name, api_name):
-            yield API(name), insn.offset
-    elif insn.offset in f.outrefs:
-        current_function = f
-        current_instruction = insn
-        for index in range(THUNK_CHAIN_DEPTH_DELTA):
-            if current_function and len(current_function.outrefs[current_instruction.offset]) == 1:
-                target = current_function.outrefs[current_instruction.offset][0]
-                referenced_function = current_function.smda_report.getFunction(target)
-                if referenced_function:
-                    # TODO SMDA: implement this function for both jmp and call, checking if function has 1 instruction which refs an API
-                    if referenced_function.isApiThunk():
-                        api_entry = (
-                            referenced_function.apirefs[target] if target in referenced_function.apirefs else None
-                        )
-                        if api_entry:
-                            # reformat
-                            dll_name, api_name = api_entry.split("!")
-                            dll_name = dll_name.split(".")[0]
-                            dll_name = dll_name.lower()
-                            for name in capa.features.extractors.helpers.generate_symbols(dll_name, api_name):
-                                yield API(name), insn.offset
-                    elif referenced_function.num_instructions == 1 and referenced_function.num_outrefs == 1:
-                        current_function = referenced_function
-                        current_instruction = [i for i in referenced_function.getInstructions()][0]
-                else:
-                    return
-
-
-def extract_insn_number_features(f, bb, insn):
-    """parse number features from the given instruction."""
-    # example:
-    #
-    #     push    3136B0h         ; dwControlCode
-    operands = [o.strip() for o in insn.operands.split(",")]
-    if insn.mnemonic == "add" and operands[0] in ["esp", "rsp"]:
-        # skip things like:
-        #
-        #    .text:00401140                 call    sub_407E2B
-        #    .text:00401145                 add     esp, 0Ch
-        return
-    for operand in operands:
-        try:
-            yield Number(int(operand, 16)), insn.offset
-            yield Number(int(operand, 16), arch=get_arch(f.smda_report)), insn.offset
-        except:
-            continue
-
-
-def read_bytes(smda_report, va, num_bytes=None):
-    """
-    read up to MAX_BYTES_FEATURE_SIZE from the given address.
-    """
-
-    rva = va - smda_report.base_addr
-    if smda_report.buffer is None:
-        return
-    buffer_end = len(smda_report.buffer)
-    max_bytes = num_bytes if num_bytes is not None else MAX_BYTES_FEATURE_SIZE
-    if rva + max_bytes > buffer_end:
-        return smda_report.buffer[rva:]
-    else:
-        return smda_report.buffer[rva : rva + max_bytes]
-
-
-def derefs(smda_report, p):
-    """
-    recursively follow the given pointer, yielding the valid memory addresses along the way.
-    useful when you may have a pointer to string, or pointer to pointer to string, etc.
-
-    this is a "do what i mean" type of helper function.
-
-    based on the implementation in viv/insn.py
-    """
-    depth = 0
-    while True:
-        if not smda_report.isAddrWithinMemoryImage(p):
-            return
-        yield p
-
-        bytes_ = read_bytes(smda_report, p, num_bytes=4)
-        val = struct.unpack("I", bytes_)[0]
-
-        # sanity: pointer points to self
-        if val == p:
-            return
-
-        # sanity: avoid chains of pointers that are unreasonably deep
-        depth += 1
-        if depth > 10:
-            return
-
-        p = val
-
-
-def extract_insn_bytes_features(f, bb, insn):
-    """
-    parse byte sequence features from the given instruction.
-    example:
-        #     push    offset iid_004118d4_IShellLinkA ; riid
-    """
-    for data_ref in insn.getDataRefs():
-        for v in derefs(f.smda_report, data_ref):
-            bytes_read = read_bytes(f.smda_report, v)
-            if bytes_read is None:
-                continue
-            if capa.features.extractors.helpers.all_zeros(bytes_read):
-                continue
-
-            yield Bytes(bytes_read), insn.offset
-
-
-def detect_ascii_len(smda_report, offset):
-    if smda_report.buffer is None:
-        return 0
-    ascii_len = 0
-    rva = offset - smda_report.base_addr
-    char = smda_report.buffer[rva]
-    while char < 127 and chr(char) in string.printable:
-        ascii_len += 1
-        rva += 1
-        char = smda_report.buffer[rva]
-    if char == 0:
-        return ascii_len
-    return 0
-
-
-def detect_unicode_len(smda_report, offset):
-    if smda_report.buffer is None:
-        return 0
-    unicode_len = 0
-    rva = offset - smda_report.base_addr
-    char = smda_report.buffer[rva]
-    second_char = smda_report.buffer[rva + 1]
-    while char < 127 and chr(char) in string.printable and second_char == 0:
-        unicode_len += 2
-        rva += 2
-        char = smda_report.buffer[rva]
-        second_char = smda_report.buffer[rva + 1]
-    if char == 0 and second_char == 0:
-        return unicode_len
-    return 0
-
-
-def read_string(smda_report, offset):
-    alen = detect_ascii_len(smda_report, offset)
-    if alen > 1:
-        return read_bytes(smda_report, offset, alen).decode("utf-8")
-    ulen = detect_unicode_len(smda_report, offset)
-    if ulen > 2:
-        return read_bytes(smda_report, offset, ulen).decode("utf-16")
-
-
-def extract_insn_string_features(f, bb, insn):
-    """parse string features from the given instruction."""
-    # example:
-    #
-    #     push    offset aAcr     ; "ACR  > "
-    for data_ref in insn.getDataRefs():
-        for v in derefs(f.smda_report, data_ref):
-            string_read = read_string(f.smda_report, v)
-            if string_read:
-                yield String(string_read.rstrip("\x00")), insn.offset
-
-
-def extract_insn_offset_features(f, bb, insn):
-    """parse structure offset features from the given instruction."""
-    # examples:
-    #
-    #     mov eax, [esi + 4]
-    #     mov eax, [esi + ecx + 16384]
-    operands = [o.strip() for o in insn.operands.split(",")]
-    for operand in operands:
-        if not "ptr" in operand:
-            continue
-        if "esp" in operand or "ebp" in operand or "rbp" in operand:
-            continue
-        number = 0
-        number_hex = re.search(PATTERN_HEXNUM, operand)
-        number_int = re.search(PATTERN_SINGLENUM, operand)
-        if number_hex:
-            number = int(number_hex.group("num"), 16)
-            number = -1 * number if number_hex.group().startswith("-") else number
-        elif number_int:
-            number = int(number_int.group("num"))
-            number = -1 * number if number_int.group().startswith("-") else number
-        yield Offset(number), insn.offset
-        yield Offset(number, arch=get_arch(f.smda_report)), insn.offset
-
-
-def is_security_cookie(f, bb, insn):
-    """
-    check if an instruction is related to security cookie checks
-    """
-    # security cookie check should use SP or BP
-    operands = [o.strip() for o in insn.operands.split(",")]
-    if operands[1] not in ["esp", "ebp", "rsp", "rbp"]:
-        return False
-    for index, block in enumerate(f.getBlocks()):
-        # expect security cookie init in first basic block within first bytes (instructions)
-        block_instructions = [i for i in block.getInstructions()]
-        if index == 0 and insn.offset < (block_instructions[0].offset + SECURITY_COOKIE_BYTES_DELTA):
-            return True
-        # ... or within last bytes (instructions) before a return
-        if block_instructions[-1].mnemonic.startswith("ret") and insn.offset > (
-            block_instructions[-1].offset - SECURITY_COOKIE_BYTES_DELTA
-        ):
-            return True
-    return False
-
-
-def extract_insn_nzxor_characteristic_features(f, bb, insn):
-    """
-    parse non-zeroing XOR instruction from the given instruction.
-    ignore expected non-zeroing XORs, e.g. security cookies.
-    """
-
-    if insn.mnemonic not in ("xor", "xorpd", "xorps", "pxor"):
-        return
-
-    operands = [o.strip() for o in insn.operands.split(",")]
-    if operands[0] == operands[1]:
-        return
-
-    if is_security_cookie(f, bb, insn):
-        return
-
-    yield Characteristic("nzxor"), insn.offset
-
-
-def extract_insn_mnemonic_features(f, bb, insn):
-    """parse mnemonic features from the given instruction."""
-    yield Mnemonic(insn.mnemonic), insn.offset
-
-
-def extract_insn_peb_access_characteristic_features(f, bb, insn):
-    """
-    parse peb access from the given function. fs:[0x30] on x86, gs:[0x60] on x64
-    """
-
-    if insn.mnemonic not in ["push", "mov"]:
-        return
-
-    operands = [o.strip() for o in insn.operands.split(",")]
-    for operand in operands:
-        if "fs:" in operand and "0x30" in operand:
-            yield Characteristic("peb access"), insn.offset
-        elif "gs:" in operand and "0x60" in operand:
-            yield Characteristic("peb access"), insn.offset
-
-
-def extract_insn_segment_access_features(f, bb, insn):
-    """ parse the instruction for access to fs or gs """
-    operands = [o.strip() for o in insn.operands.split(",")]
-    for operand in operands:
-        if "fs:" in operand:
-            yield Characteristic("fs access"), insn.offset
-        elif "gs:" in operand:
-            yield Characteristic("gs access"), insn.offset
-
-
-def extract_insn_cross_section_cflow(f, bb, insn):
-    """
-    inspect the instruction for a CALL or JMP that crosses section boundaries.
-    """
-    if insn.mnemonic in ["call", "jmp"]:
-        if insn.offset in f.apirefs:
-            return
-
-        smda_report = insn.smda_function.smda_report
-        if insn.offset in f.outrefs:
-            for target in f.outrefs[insn.offset]:
-                if smda_report.getSection(insn.offset) != smda_report.getSection(target):
-                    yield Characteristic("cross section flow"), insn.offset
-        elif insn.operands.startswith("0x"):
-            target = int(insn.operands, 16)
-            if smda_report.getSection(insn.offset) != smda_report.getSection(target):
-                yield Characteristic("cross section flow"), insn.offset
-
-
-# this is a feature that's most relevant at the function scope,
-# however, its most efficient to extract at the instruction scope.
-def extract_function_calls_from(f, bb, insn):
-    if insn.mnemonic != "call":
-        return
-
-    if insn.offset in f.outrefs:
-        for outref in f.outrefs[insn.offset]:
-            yield Characteristic("calls from"), outref
-
-            if outref == f.offset:
-                # if we found a jump target and it's the function address
-                # mark as recursive
-                yield Characteristic("recursive call"), outref
-    if insn.offset in f.apirefs:
-        yield Characteristic("calls from"), insn.offset
-
-
-# this is a feature that's most relevant at the function or basic block scope,
-# however, its most efficient to extract at the instruction scope.
-def extract_function_indirect_call_characteristic_features(f, bb, insn):
-    """
-    extract indirect function call characteristic (e.g., call eax or call dword ptr [edx+4])
-    does not include calls like => call ds:dword_ABD4974
-    """
-    if insn.mnemonic != "call":
-        return
-    if insn.operands.startswith("0x"):
-        return False
-    if "qword ptr" in insn.operands and "rip" in insn.operands:
-        return False
-    if insn.operands.startswith("dword ptr [0x"):
-        return False
-    # call edx
-    # call dword ptr [eax+50h]
-    # call qword ptr [rsp+78h]
-    yield Characteristic("indirect call"), insn.offset
-
-
-def extract_features(f, bb, insn):
-    """
-    extract features from the given insn.
-
-    args:
-      f (smda.common.SmdaFunction): the function to process.
-      bb (smda.common.SmdaBasicBlock): the basic block to process.
-      insn (smda.common.SmdaInstruction): the instruction to process.
-
-    yields:
-      Feature, set[VA]: the features and their location found in this insn.
-    """
-    for insn_handler in INSTRUCTION_HANDLERS:
-        for feature, va in insn_handler(f, bb, insn):
-            yield feature, va
-
-
-INSTRUCTION_HANDLERS = (
-    extract_insn_api_features,
-    extract_insn_number_features,
-    extract_insn_string_features,
-    extract_insn_bytes_features,
-    extract_insn_offset_features,
-    extract_insn_nzxor_characteristic_features,
-    extract_insn_mnemonic_features,
-    extract_insn_peb_access_characteristic_features,
-    extract_insn_cross_section_cflow,
-    extract_insn_segment_access_features,
-    extract_function_calls_from,
-    extract_function_indirect_call_characteristic_features,
-)

capa/features/extractors/strings.py

@@ -1,6 +1,6 @@
-# strings code from FLOSS, https://github.com/fireeye/flare-floss
+# strings code from FLOSS, https://github.com/mandiant/flare-floss
 #
-# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
 # Licensed under the Apache License, Version 2.0 (the "License");
 #  you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at: [package root]/LICENSE.txt
@@ -9,6 +9,7 @@
 # See the License for the specific language governing permissions and limitations under the License.
 
 import re
+import contextlib
 from collections import namedtuple
 
 ASCII_BYTE = r" !\"#\$%&\'\(\)\*\+,-\./0123456789:;<=>\?@ABCDEFGHIJKLMNOPQRSTUVWXYZ\[\]\^_`abcdefghijklmnopqrstuvwxyz\{\|\}\\\~\t".encode(
@@ -81,24 +82,5 @@ def extract_unicode_strings(buf, n=4):
         reg = b"((?:[%s]\x00){%d,})" % (ASCII_BYTE, n)
         r = re.compile(reg)
     for match in r.finditer(buf):
-        try:
+        with contextlib.suppress(UnicodeDecodeError):
             yield String(match.group().decode("utf-16"), match.start())
-        except UnicodeDecodeError:
-            pass
-
-
-def main():
-    import sys
-
-    with open(sys.argv[1], "rb") as f:
-        b = f.read()
-
-    for s in extract_ascii_strings(b):
-        print("0x{:x}: {:s}".format(s.offset, s.s))
-
-    for s in extract_unicode_strings(b):
-        print("0x{:x}: {:s}".format(s.offset, s.s))
-
-
-if __name__ == "__main__":
-    main()

capa/features/extractors/viv/__init__.py

@@ -1,81 +0,0 @@
-# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
-# Licensed under the Apache License, Version 2.0 (the "License");
-#  you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at: [package root]/LICENSE.txt
-# Unless required by applicable law or agreed to in writing, software distributed under the License
-#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and limitations under the License.
-
-import types
-
-import viv_utils
-
-import capa.features.extractors
-import capa.features.extractors.viv.file
-import capa.features.extractors.viv.insn
-import capa.features.extractors.viv.function
-import capa.features.extractors.viv.basicblock
-from capa.features.extractors import FeatureExtractor
-
-__all__ = ["file", "function", "basicblock", "insn"]
-
-
-def get_va(self):
-    try:
-        # vivisect type
-        return self.va
-    except AttributeError:
-        pass
-
-    raise TypeError()
-
-
-def add_va_int_cast(o):
-    """
-    dynamically add a cast-to-int (`__int__`) method to the given object
-    that returns the value of the `.va` property.
-
-    this bit of skullduggery lets use cast viv-utils objects as ints.
-    the correct way of doing this is to update viv-utils (or subclass the objects here).
-    """
-    setattr(o, "__int__", types.MethodType(get_va, o))
-    return o
-
-
-class VivisectFeatureExtractor(FeatureExtractor):
-    def __init__(self, vw, path):
-        super(VivisectFeatureExtractor, self).__init__()
-        self.vw = vw
-        self.path = path
-
-    def get_base_address(self):
-        # assume there is only one file loaded into the vw
-        return list(self.vw.filemeta.values())[0]["imagebase"]
-
-    def extract_file_features(self):
-        for feature, va in capa.features.extractors.viv.file.extract_features(self.vw, self.path):
-            yield feature, va
-
-    def get_functions(self):
-        for va in sorted(self.vw.getFunctions()):
-            yield add_va_int_cast(viv_utils.Function(self.vw, va))
-
-    def extract_function_features(self, f):
-        for feature, va in capa.features.extractors.viv.function.extract_features(f):
-            yield feature, va
-
-    def get_basic_blocks(self, f):
-        for bb in f.basic_blocks:
-            yield add_va_int_cast(bb)
-
-    def extract_basic_block_features(self, f, bb):
-        for feature, va in capa.features.extractors.viv.basicblock.extract_features(f, bb):
-            yield feature, va
-
-    def get_instructions(self, f, bb):
-        for insn in bb.instructions:
-            yield add_va_int_cast(insn)
-
-    def extract_insn_features(self, f, bb, insn):
-        for feature, va in capa.features.extractors.viv.insn.extract_features(f, bb, insn):
-            yield feature, va

capa/features/extractors/viv/basicblock.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
 # Licensed under the Apache License, Version 2.0 (the "License");
 #  you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at: [package root]/LICENSE.txt
@@ -8,27 +8,30 @@
 
 import string
 import struct
+from typing import Tuple, Iterator
 
 import envi
-import vivisect.const
+import envi.archs.i386.disasm
 
-from capa.features import Characteristic
+from capa.features.common import Feature, Characteristic
+from capa.features.address import Address, AbsoluteVirtualAddress
 from capa.features.basicblock import BasicBlock
 from capa.features.extractors.helpers import MIN_STACKSTRING_LEN
+from capa.features.extractors.base_extractor import BBHandle, FunctionHandle
 
 
-def interface_extract_basic_block_XXX(f, bb):
+def interface_extract_basic_block_XXX(f: FunctionHandle, bb: BBHandle) -> Iterator[Tuple[Feature, Address]]:
     """
     parse features from the given basic block.
 
     args:
-      f (viv_utils.Function): the function to process.
-      bb (viv_utils.BasicBlock): the basic block to process.
+      f: the function to process.
+      bb: the basic block to process.
 
     yields:
-      (Feature, int): the feature and the address at which its found.
+      (Feature, Address): the feature and the address at which its found.
     """
-    yield NotImplementedError("feature"), NotImplementedError("virtual address")
+    raise NotImplementedError
 
 
 def _bb_has_tight_loop(f, bb):
@@ -37,17 +40,17 @@ def _bb_has_tight_loop(f, bb):
     """
     if len(bb.instructions) > 0:
         for bva, bflags in bb.instructions[-1].getBranches():
-            if bflags & vivisect.envi.BR_COND:
+            if bflags & envi.BR_COND:
                 if bva == bb.va:
                     return True
 
     return False
 
 
-def extract_bb_tight_loop(f, bb):
-    """ check basic block for tight loop indicators """
-    if _bb_has_tight_loop(f, bb):
-        yield Characteristic("tight loop"), bb.va
+def extract_bb_tight_loop(f: FunctionHandle, bb: BBHandle) -> Iterator[Tuple[Feature, Address]]:
+    """check basic block for tight loop indicators"""
+    if _bb_has_tight_loop(f, bb.inner):
+        yield Characteristic("tight loop"), bb.address
 
 
 def _bb_has_stackstring(f, bb):
@@ -67,13 +70,13 @@ def _bb_has_stackstring(f, bb):
     return False
 
 
-def extract_stackstring(f, bb):
-    """ check basic block for stackstring indicators """
-    if _bb_has_stackstring(f, bb):
-        yield Characteristic("stack string"), bb.va
+def extract_stackstring(f: FunctionHandle, bb: BBHandle) -> Iterator[Tuple[Feature, Address]]:
+    """check basic block for stackstring indicators"""
+    if _bb_has_stackstring(f, bb.inner):
+        yield Characteristic("stack string"), bb.address
 
 
-def is_mov_imm_to_stack(instr):
+def is_mov_imm_to_stack(instr: envi.archs.i386.disasm.i386Opcode) -> bool:
     """
     Return if instruction moves immediate onto stack
     """
@@ -89,7 +92,6 @@ def is_mov_imm_to_stack(instr):
     if not src.isImmed():
         return False
 
-    # TODO what about 64-bit operands?
     if not isinstance(dst, envi.archs.i386.disasm.i386SibOper) and not isinstance(
         dst, envi.archs.i386.disasm.i386RegMemOper
     ):
@@ -105,7 +107,7 @@ def is_mov_imm_to_stack(instr):
     return True
 
 
-def get_printable_len(oper):
+def get_printable_len(oper: envi.archs.i386.disasm.i386ImmOper) -> int:
     """
     Return string length if all operand bytes are ascii or utf16-le printable
     """
@@ -117,14 +119,18 @@ def get_printable_len(oper):
         chars = struct.pack("<I", oper.imm)
     elif oper.tsize == 8:
         chars = struct.pack("<Q", oper.imm)
+    else:
+        raise ValueError(f"unexpected oper.tsize: {oper.tsize}")
+
     if is_printable_ascii(chars):
         return oper.tsize
-    if is_printable_utf16le(chars):
+    elif is_printable_utf16le(chars):
         return oper.tsize / 2
-    return 0
+    else:
+        return 0
 
 
-def is_printable_ascii(chars):
+def is_printable_ascii(chars: bytes) -> bool:
     try:
         chars_str = chars.decode("ascii")
     except UnicodeDecodeError:
@@ -133,12 +139,13 @@ def is_printable_ascii(chars):
         return all(c in string.printable for c in chars_str)
 
 
-def is_printable_utf16le(chars):
+def is_printable_utf16le(chars: bytes) -> bool:
     if all(c == b"\x00" for c in chars[1::2]):
         return is_printable_ascii(chars[::2])
+    return False
 
 
-def extract_features(f, bb):
+def extract_features(f: FunctionHandle, bb: BBHandle) -> Iterator[Tuple[Feature, Address]]:
     """
     extract features from the given basic block.
 
@@ -147,12 +154,12 @@ def extract_features(f, bb):
       bb (viv_utils.BasicBlock): the basic block to process.
 
     yields:
-      Feature, set[VA]: the features and their location found in this basic block.
+      Tuple[Feature, int]: the features and their location found in this basic block.
     """
-    yield BasicBlock(), bb.va
+    yield BasicBlock(), AbsoluteVirtualAddress(bb.inner.va)
     for bb_handler in BASIC_BLOCK_HANDLERS:
-        for feature, va in bb_handler(f, bb):
-            yield feature, va
+        for feature, addr in bb_handler(f, bb):
+            yield feature, addr
 
 
 BASIC_BLOCK_HANDLERS = (

capa/features/extractors/viv/extractor.py

@@ -0,0 +1,83 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import logging
+from typing import Any, Dict, List, Tuple, Iterator
+from pathlib import Path
+
+import viv_utils
+import viv_utils.flirt
+
+import capa.features.extractors.common
+import capa.features.extractors.viv.file
+import capa.features.extractors.viv.insn
+import capa.features.extractors.viv.global_
+import capa.features.extractors.viv.function
+import capa.features.extractors.viv.basicblock
+from capa.features.common import Feature
+from capa.features.address import Address, AbsoluteVirtualAddress
+from capa.features.extractors.base_extractor import BBHandle, InsnHandle, FunctionHandle, FeatureExtractor
+
+logger = logging.getLogger(__name__)
+
+
+class VivisectFeatureExtractor(FeatureExtractor):
+    def __init__(self, vw, path: Path, os):
+        super().__init__()
+        self.vw = vw
+        self.path = path
+        self.buf = path.read_bytes()
+
+        # pre-compute these because we'll yield them at *every* scope.
+        self.global_features: List[Tuple[Feature, Address]] = []
+        self.global_features.extend(capa.features.extractors.viv.file.extract_file_format(self.buf))
+        self.global_features.extend(capa.features.extractors.common.extract_os(self.buf, os))
+        self.global_features.extend(capa.features.extractors.viv.global_.extract_arch(self.vw))
+
+    def get_base_address(self):
+        # assume there is only one file loaded into the vw
+        return AbsoluteVirtualAddress(list(self.vw.filemeta.values())[0]["imagebase"])
+
+    def extract_global_features(self):
+        yield from self.global_features
+
+    def extract_file_features(self):
+        yield from capa.features.extractors.viv.file.extract_features(self.vw, self.buf)
+
+    def get_functions(self) -> Iterator[FunctionHandle]:
+        cache: Dict[str, Any] = {}
+        for va in sorted(self.vw.getFunctions()):
+            yield FunctionHandle(
+                address=AbsoluteVirtualAddress(va), inner=viv_utils.Function(self.vw, va), ctx={"cache": cache}
+            )
+
+    def extract_function_features(self, fh: FunctionHandle) -> Iterator[Tuple[Feature, Address]]:
+        yield from capa.features.extractors.viv.function.extract_features(fh)
+
+    def get_basic_blocks(self, fh: FunctionHandle) -> Iterator[BBHandle]:
+        f: viv_utils.Function = fh.inner
+        for bb in f.basic_blocks:
+            yield BBHandle(address=AbsoluteVirtualAddress(bb.va), inner=bb)
+
+    def extract_basic_block_features(self, fh: FunctionHandle, bbh) -> Iterator[Tuple[Feature, Address]]:
+        yield from capa.features.extractors.viv.basicblock.extract_features(fh, bbh)
+
+    def get_instructions(self, fh: FunctionHandle, bbh: BBHandle) -> Iterator[InsnHandle]:
+        bb: viv_utils.BasicBlock = bbh.inner
+        for insn in bb.instructions:
+            yield InsnHandle(address=AbsoluteVirtualAddress(insn.va), inner=insn)
+
+    def extract_insn_features(
+        self, fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+    ) -> Iterator[Tuple[Feature, Address]]:
+        yield from capa.features.extractors.viv.insn.extract_features(fh, bbh, ih)
+
+    def is_library_function(self, addr):
+        return viv_utils.flirt.is_library_function(self.vw, addr)
+
+    def get_function_name(self, addr):
+        return viv_utils.get_function_name(self.vw, addr)

capa/features/extractors/viv/file.py

@@ -1,33 +1,62 @@
-# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
 # Licensed under the Apache License, Version 2.0 (the "License");
 #  you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at: [package root]/LICENSE.txt
 # Unless required by applicable law or agreed to in writing, software distributed under the License
 #  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and limitations under the License.
+from typing import Tuple, Iterator
 
 import PE.carve as pe_carve  # vivisect PE
+import vivisect
+import viv_utils
+import viv_utils.flirt
 
+import capa.features.insn
+import capa.features.extractors.common
 import capa.features.extractors.helpers
 import capa.features.extractors.strings
-from capa.features import String, Characteristic
-from capa.features.file import Export, Import, Section
+from capa.features.file import Export, Import, Section, FunctionName
+from capa.features.common import Feature, Characteristic
+from capa.features.address import Address, FileOffsetAddress, AbsoluteVirtualAddress
 
 
-def extract_file_embedded_pe(vw, file_path):
-    with open(file_path, "rb") as f:
-        fbytes = f.read()
-
-    for offset, i in pe_carve.carve(fbytes, 1):
-        yield Characteristic("embedded pe"), offset
+def extract_file_embedded_pe(buf, **kwargs) -> Iterator[Tuple[Feature, Address]]:
+    for offset, _ in pe_carve.carve(buf, 1):
+        yield Characteristic("embedded pe"), FileOffsetAddress(offset)
 
 
-def extract_file_export_names(vw, file_path):
-    for va, etype, name, _ in vw.getExports():
-        yield Export(name), va
+def get_first_vw_filename(vw: vivisect.VivWorkspace):
+    # vivisect associates metadata with each file that its loaded into the workspace.
+    # capa only loads a single file into each workspace.
+    # so to access the metadata for the file in question, we can just take the first one.
+    # otherwise, we'd have to pass around the module name of the file we're analyzing,
+    # which is a pain.
+    #
+    # so this is a simplifying assumption.
+    return next(iter(vw.filemeta.keys()))
 
 
-def extract_file_import_names(vw, file_path):
+def extract_file_export_names(vw: vivisect.VivWorkspace, **kwargs) -> Iterator[Tuple[Feature, Address]]:
+    for va, _, name, _ in vw.getExports():
+        yield Export(name), AbsoluteVirtualAddress(va)
+
+    if vw.getMeta("Format") == "pe":
+        pe = vw.parsedbin
+        baseaddr = pe.IMAGE_NT_HEADERS.OptionalHeader.ImageBase
+        for rva, _, forwarded_name in vw.getFileMeta(get_first_vw_filename(vw), "forwarders"):
+            try:
+                forwarded_name = forwarded_name.partition(b"\x00")[0].decode("ascii")
+            except UnicodeDecodeError:
+                continue
+
+            forwarded_name = capa.features.extractors.helpers.reformat_forwarded_export_name(forwarded_name)
+            va = baseaddr + rva
+            yield Export(forwarded_name), AbsoluteVirtualAddress(va)
+            yield Characteristic("forwarded export"), AbsoluteVirtualAddress(va)
+
+
+def extract_file_import_names(vw, **kwargs) -> Iterator[Tuple[Feature, Address]]:
     """
     extract imported function names
     1. imports by ordinal:
@@ -38,16 +67,17 @@ def extract_file_import_names(vw, file_path):
     """
     for va, _, _, tinfo in vw.getImports():
         # vivisect source: tinfo = "%s.%s" % (libname, impname)
-        modname, impname = tinfo.split(".")
+        modname, impname = tinfo.split(".", 1)
         if is_viv_ord_impname(impname):
             # replace ord prefix with #
-            impname = "#%s" % impname[len("ord") :]
+            impname = "#" + impname[len("ord") :]
 
+        addr = AbsoluteVirtualAddress(va)
         for name in capa.features.extractors.helpers.generate_symbols(modname, impname):
-            yield Import(name), va
+            yield Import(name), addr
 
 
-def is_viv_ord_impname(impname):
+def is_viv_ord_impname(impname: str) -> bool:
     """
     return if import name matches vivisect's ordinal naming scheme `'ord%d' % ord`
     """
@@ -61,40 +91,51 @@ def is_viv_ord_impname(impname):
         return True
 
 
-def extract_file_section_names(vw, file_path):
+def extract_file_section_names(vw, **kwargs) -> Iterator[Tuple[Feature, Address]]:
     for va, _, segname, _ in vw.getSegments():
-        yield Section(segname), va
+        yield Section(segname), AbsoluteVirtualAddress(va)
 
 
-def extract_file_strings(vw, file_path):
+def extract_file_strings(buf, **kwargs) -> Iterator[Tuple[Feature, Address]]:
+    yield from capa.features.extractors.common.extract_file_strings(buf)
+
+
+def extract_file_function_names(vw, **kwargs) -> Iterator[Tuple[Feature, Address]]:
     """
-    extract ASCII and UTF-16 LE strings from file
+    extract the names of statically-linked library functions.
     """
-    with open(file_path, "rb") as f:
-        b = f.read()
-
-    for s in capa.features.extractors.strings.extract_ascii_strings(b):
-        yield String(s.s), s.offset
-
-    for s in capa.features.extractors.strings.extract_unicode_strings(b):
-        yield String(s.s), s.offset
+    for va in sorted(vw.getFunctions()):
+        addr = AbsoluteVirtualAddress(va)
+        if viv_utils.flirt.is_library_function(vw, va):
+            name = viv_utils.get_function_name(vw, va)
+            yield FunctionName(name), addr
+            if name.startswith("_"):
+                # some linkers may prefix linked routines with a `_` to avoid name collisions.
+                # extract features for both the mangled and un-mangled representations.
+                # e.g. `_fwrite` -> `fwrite`
+                # see: https://stackoverflow.com/a/2628384/87207
+                yield FunctionName(name[1:]), addr
 
 
-def extract_features(vw, file_path):
+def extract_file_format(buf, **kwargs) -> Iterator[Tuple[Feature, Address]]:
+    yield from capa.features.extractors.common.extract_format(buf)
+
+
+def extract_features(vw, buf: bytes) -> Iterator[Tuple[Feature, Address]]:
     """
     extract file features from given workspace
 
     args:
       vw (vivisect.VivWorkspace): the vivisect workspace
-      file_path: path to the input file
+      buf: the raw input file bytes
 
     yields:
-      Tuple[Feature, VA]: a feature and its location.
+      Tuple[Feature, Address]: a feature and its location.
     """
 
     for file_handler in FILE_HANDLERS:
-        for feature, va in file_handler(vw, file_path):
-            yield feature, va
+        for feature, addr in file_handler(vw=vw, buf=buf):  # type: ignore
+            yield feature, addr
 
 
 FILE_HANDLERS = (
@@ -103,4 +144,6 @@ FILE_HANDLERS = (
     extract_file_import_names,
     extract_file_section_names,
     extract_file_strings,
+    extract_file_function_names,
+    extract_file_format,
 )

capa/features/extractors/viv/function.py

@@ -1,70 +1,110 @@
-# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
 # Licensed under the Apache License, Version 2.0 (the "License");
 #  you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at: [package root]/LICENSE.txt
 # Unless required by applicable law or agreed to in writing, software distributed under the License
 #  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and limitations under the License.
+from typing import Tuple, Iterator
 
+import envi
+import viv_utils
 import vivisect.const
 
-from capa.features import Characteristic
+from capa.features.file import FunctionName
+from capa.features.common import Feature, Characteristic
+from capa.features.address import Address, AbsoluteVirtualAddress
 from capa.features.extractors import loops
+from capa.features.extractors.elf import SymTab
+from capa.features.extractors.base_extractor import FunctionHandle
 
 
-def interface_extract_function_XXX(f):
+def interface_extract_function_XXX(fh: FunctionHandle) -> Iterator[Tuple[Feature, Address]]:
     """
     parse features from the given function.
 
     args:
-      f (viv_utils.Function): the function to process.
+      f: the function to process.
 
     yields:
-      (Feature, int): the feature and the address at which its found.
+      (Feature, Address): the feature and the address at which its found.
     """
-    yield NotImplementedError("feature"), NotImplementedError("virtual address")
+    raise NotImplementedError
 
 
-def extract_function_calls_to(f):
+def extract_function_symtab_names(fh: FunctionHandle) -> Iterator[Tuple[Feature, Address]]:
+    if fh.inner.vw.metadata["Format"] == "elf":
+        # the file's symbol table gets added to the metadata of the vivisect workspace.
+        # this is in order to eliminate the computational overhead of refetching symtab each time.
+        if "symtab" not in fh.ctx["cache"]:
+            try:
+                fh.ctx["cache"]["symtab"] = SymTab.from_viv(fh.inner.vw.parsedbin)
+            except Exception:
+                fh.ctx["cache"]["symtab"] = None
+
+        symtab = fh.ctx["cache"]["symtab"]
+        if symtab:
+            for symbol in symtab.get_symbols():
+                sym_name = symtab.get_name(symbol)
+                sym_value = symbol.value
+                sym_info = symbol.info
+
+                STT_FUNC = 0x2
+                if sym_value == fh.address and sym_info & STT_FUNC != 0:
+                    yield FunctionName(sym_name), fh.address
+
+
+def extract_function_calls_to(fhandle: FunctionHandle) -> Iterator[Tuple[Feature, Address]]:
+    f: viv_utils.Function = fhandle.inner
     for src, _, _, _ in f.vw.getXrefsTo(f.va, rtype=vivisect.const.REF_CODE):
-        yield Characteristic("calls to"), src
+        yield Characteristic("calls to"), AbsoluteVirtualAddress(src)
 
 
-def extract_function_loop(f):
+def extract_function_loop(fhandle: FunctionHandle) -> Iterator[Tuple[Feature, Address]]:
     """
     parse if a function has a loop
     """
+    f: viv_utils.Function = fhandle.inner
+
     edges = []
 
     for bb in f.basic_blocks:
         if len(bb.instructions) > 0:
             for bva, bflags in bb.instructions[-1].getBranches():
+                if bva is None:
+                    # vivisect may be unable to recover the call target, e.g. on dynamic calls like `call esi`
+                    # for this bva is None, and we don't want to add it for loop detection, ref: vivisect#574
+                    continue
                 # vivisect does not set branch flags for non-conditional jmp so add explicit check
                 if (
-                    bflags & vivisect.envi.BR_COND
-                    or bflags & vivisect.envi.BR_FALL
-                    or bflags & vivisect.envi.BR_TABLE
+                    bflags & envi.BR_COND
+                    or bflags & envi.BR_FALL
+                    or bflags & envi.BR_TABLE
                     or bb.instructions[-1].mnem == "jmp"
                 ):
                     edges.append((bb.va, bva))
 
     if edges and loops.has_loop(edges):
-        yield Characteristic("loop"), f.va
+        yield Characteristic("loop"), fhandle.address
 
 
-def extract_features(f):
+def extract_features(fh: FunctionHandle) -> Iterator[Tuple[Feature, Address]]:
     """
     extract features from the given function.
 
     args:
-      f (viv_utils.Function): the function from which to extract features
+      fh: the function handle from which to extract features
 
     yields:
-      Feature, set[VA]: the features and their location found in this function.
+      Tuple[Feature, int]: the features and their location found in this function.
     """
     for func_handler in FUNCTION_HANDLERS:
-        for feature, va in func_handler(f):
-            yield feature, va
+        for feature, addr in func_handler(fh):
+            yield feature, addr
 
 
-FUNCTION_HANDLERS = (extract_function_calls_to, extract_function_loop)
+FUNCTION_HANDLERS = (
+    extract_function_symtab_names,
+    extract_function_calls_to,
+    extract_function_loop,
+)

capa/features/extractors/viv/global_.py

@@ -0,0 +1,31 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import logging
+from typing import Tuple, Iterator
+
+from capa.features.common import ARCH_I386, ARCH_AMD64, Arch, Feature
+from capa.features.address import NO_ADDRESS, Address
+
+logger = logging.getLogger(__name__)
+
+
+def extract_arch(vw) -> Iterator[Tuple[Feature, Address]]:
+    arch = vw.getMeta("Architecture")
+    if arch == "amd64":
+        yield Arch(ARCH_AMD64), NO_ADDRESS
+
+    elif arch == "i386":
+        yield Arch(ARCH_I386), NO_ADDRESS
+
+    else:
+        # we likely end up here:
+        #  1. handling a new architecture (e.g. aarch64)
+        #
+        # for (1), this logic will need to be updated as the format is implemented.
+        logger.debug("unsupported architecture: %s", vw.arch.__class__.__name__)
+        return

capa/features/extractors/viv/helpers.py

@@ -1,14 +1,17 @@
-# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
 # Licensed under the Apache License, Version 2.0 (the "License");
 #  you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at: [package root]/LICENSE.txt
 # Unless required by applicable law or agreed to in writing, software distributed under the License
 #  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and limitations under the License.
+from typing import Optional
+
+from vivisect import VivWorkspace
 from vivisect.const import XR_TO, REF_CODE
 
 
-def get_coderef_from(vw, va):
+def get_coderef_from(vw: VivWorkspace, va: int) -> Optional[int]:
     """
     return first code `tova` whose origin is the specified va
     return None if no code reference is found

capa/features/extractors/viv/indirect_calls.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
 # Licensed under the Apache License, Version 2.0 (the "License");
 #  you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at: [package root]/LICENSE.txt
@@ -7,11 +7,13 @@
 # See the License for the specific language governing permissions and limitations under the License.
 
 import collections
+from typing import Set, List, Deque, Tuple, Optional
 
 import envi
 import vivisect.const
 import envi.archs.i386.disasm
 import envi.archs.amd64.disasm
+from vivisect import VivWorkspace
 
 # pull out consts for lookup performance
 i386RegOper = envi.archs.i386.disasm.i386RegOper
@@ -26,7 +28,7 @@ FAR_BRANCH_MASK = envi.BR_PROC | envi.BR_DEREF | envi.BR_ARCH
 DESTRUCTIVE_MNEMONICS = ("mov", "lea", "pop", "xor")
 
 
-def get_previous_instructions(vw, va):
+def get_previous_instructions(vw: VivWorkspace, va: int) -> List[int]:
     """
     collect the instructions that flow to the given address, local to the current function.
 
@@ -40,22 +42,24 @@ def get_previous_instructions(vw, va):
     ret = []
 
     # find the immediate prior instruction.
-    # ensure that it fallsthrough to this one.
+    # ensure that it falls through to this one.
     loc = vw.getPrevLocation(va, adjacent=True)
     if loc is not None:
-        # from vivisect.const:
-        # location: (L_VA, L_SIZE, L_LTYPE, L_TINFO)
-        (pva, _, ptype, pinfo) = vw.getPrevLocation(va, adjacent=True)
+        ploc = vw.getPrevLocation(va, adjacent=True)
+        if ploc is not None:
+            # from vivisect.const:
+            # location: (L_VA, L_SIZE, L_LTYPE, L_TINFO)
+            (pva, _, ptype, pinfo) = ploc
 
-        if ptype == LOC_OP and not (pinfo & IF_NOFALL):
-            ret.append(pva)
+            if ptype == LOC_OP and not (pinfo & IF_NOFALL):
+                ret.append(pva)
 
     # find any code refs, e.g. jmp, to this location.
     # ignore any calls.
     #
     # from vivisect.const:
     # xref: (XR_FROM, XR_TO, XR_RTYPE, XR_RFLAG)
-    for (xfrom, _, _, xflag) in vw.getXrefsTo(va, REF_CODE):
+    for xfrom, _, _, xflag in vw.getXrefsTo(va, REF_CODE):
         if (xflag & FAR_BRANCH_MASK) != 0:
             continue
         ret.append(xfrom)
@@ -67,7 +71,7 @@ class NotFoundError(Exception):
     pass
 
 
-def find_definition(vw, va, reg):
+def find_definition(vw: VivWorkspace, va: int, reg: int) -> Tuple[int, Optional[int]]:
     """
     scan backwards from the given address looking for assignments to the given register.
     if a constant, return that value.
@@ -83,8 +87,8 @@ def find_definition(vw, va, reg):
     raises:
       NotFoundError: when the definition cannot be found.
     """
-    q = collections.deque()
-    seen = set([])
+    q: Deque[int] = collections.deque()
+    seen: Set[int] = set()
 
     q.extend(get_previous_instructions(vw, va))
     while q:
@@ -128,14 +132,14 @@ def find_definition(vw, va, reg):
     raise NotFoundError()
 
 
-def is_indirect_call(vw, va, insn=None):
+def is_indirect_call(vw: VivWorkspace, va: int, insn: envi.Opcode) -> bool:
     if insn is None:
         insn = vw.parseOpcode(va)
 
     return insn.mnem in ("call", "jmp") and isinstance(insn.opers[0], envi.archs.i386.disasm.i386RegOper)
 
 
-def resolve_indirect_call(vw, va, insn=None):
+def resolve_indirect_call(vw: VivWorkspace, va: int, insn: envi.Opcode) -> Tuple[int, Optional[int]]:
     """
     inspect the given indirect call instruction and attempt to resolve the target address.
 

capa/features/extractors/viv/insn.py

@@ -1,26 +1,29 @@
-# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
 # Licensed under the Apache License, Version 2.0 (the "License");
 #  you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at: [package root]/LICENSE.txt
 # Unless required by applicable law or agreed to in writing, software distributed under the License
 #  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and limitations under the License.
+from typing import List, Tuple, Callable, Iterator
 
+import envi
+import envi.exc
+import viv_utils
 import envi.memory
+import viv_utils.flirt
+import envi.archs.i386.regs
+import envi.archs.amd64.regs
 import envi.archs.i386.disasm
+import envi.archs.amd64.disasm
 
 import capa.features.extractors.helpers
 import capa.features.extractors.viv.helpers
-from capa.features import (
-    ARCH_X32,
-    ARCH_X64,
-    MAX_BYTES_FEATURE_SIZE,
-    THUNK_CHAIN_DEPTH_DELTA,
-    Bytes,
-    String,
-    Characteristic,
-)
-from capa.features.insn import API, Number, Offset, Mnemonic
+from capa.features.insn import API, MAX_STRUCTURE_SIZE, Number, Offset, Mnemonic, OperandNumber, OperandOffset
+from capa.features.common import MAX_BYTES_FEATURE_SIZE, THUNK_CHAIN_DEPTH_DELTA, Bytes, String, Feature, Characteristic
+from capa.features.address import Address, AbsoluteVirtualAddress
+from capa.features.extractors.elf import SymTab
+from capa.features.extractors.base_extractor import BBHandle, InsnHandle, FunctionHandle
 from capa.features.extractors.viv.indirect_calls import NotFoundError, resolve_indirect_call
 
 # security cookie checks may perform non-zeroing XORs, these are expected within a certain
@@ -28,27 +31,21 @@ from capa.features.extractors.viv.indirect_calls import NotFoundError, resolve_i
 SECURITY_COOKIE_BYTES_DELTA = 0x40
 
 
-def get_arch(vw):
-    arch = vw.getMeta("Architecture")
-    if arch == "i386":
-        return ARCH_X32
-    elif arch == "amd64":
-        return ARCH_X64
-
-
-def interface_extract_instruction_XXX(f, bb, insn):
+def interface_extract_instruction_XXX(
+    fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
     """
     parse features from the given instruction.
 
     args:
-      f (viv_utils.Function): the function to process.
-      bb (viv_utils.BasicBlock): the basic block to process.
-      insn (vivisect...Instruction): the instruction to process.
+      fh: the function handle to process.
+      bbh: the basic block handle to process.
+      ih: the instruction handle to process.
 
     yields:
-      (Feature, int): the feature and the address at which its found.
+      (Feature, Address): the feature and the address at which its found.
     """
-    yield NotImplementedError("feature"), NotImplementedError("virtual address")
+    raise NotImplementedError
 
 
 def get_imports(vw):
@@ -68,13 +65,15 @@ def get_imports(vw):
         return imports
 
 
-def extract_insn_api_features(f, bb, insn):
-    """parse API features from the given instruction."""
-
-    # example:
-    #
-    #    call dword [0x00473038]
+def extract_insn_api_features(fh: FunctionHandle, bb, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
+    """
+    parse API features from the given instruction.
 
+    example:
+       call dword [0x00473038]
+    """
+    insn: envi.Opcode = ih.inner
+    f: viv_utils.Function = fh.inner
     if insn.mnem not in ("call", "jmp"):
         return
 
@@ -91,12 +90,12 @@ def extract_insn_api_features(f, bb, insn):
         if target in imports:
             dll, symbol = imports[target]
             for name in capa.features.extractors.helpers.generate_symbols(dll, symbol):
-                yield API(name), insn.va
+                yield API(name), ih.address
 
     # call via thunk on x86,
     # see 9324d1a8ae37a36ae560c37448c9705a at 0x407985
     #
-    # this is also how calls to internal functions may be decoded on x64.
+    # this is also how calls to internal functions may be decoded on x32 and x64.
     # see Lab21-01.exe_:0x140001178
     #
     # follow chained thunks, e.g. in 82bf6347acf15e5d883715dc289d8a2b at 0x14005E0FF in
@@ -111,11 +110,46 @@ def extract_insn_api_features(f, bb, insn):
         if not target:
             return
 
+        if f.vw.metadata["Format"] == "elf":
+            if "symtab" not in fh.ctx["cache"]:
+                # the symbol table gets stored as a function's attribute in order to avoid running
+                # this code everytime the call is made, thus preventing the computational overhead.
+                try:
+                    fh.ctx["cache"]["symtab"] = SymTab.from_viv(f.vw.parsedbin)
+                except Exception:
+                    fh.ctx["cache"]["symtab"] = None
+
+            symtab = fh.ctx["cache"]["symtab"]
+            if symtab:
+                for symbol in symtab.get_symbols():
+                    sym_name = symtab.get_name(symbol)
+                    sym_value = symbol.value
+                    sym_info = symbol.info
+
+                    STT_FUNC = 0x2
+                    if sym_value == target and sym_info & STT_FUNC != 0:
+                        yield API(sym_name), ih.address
+
+        if viv_utils.flirt.is_library_function(f.vw, target):
+            name = viv_utils.get_function_name(f.vw, target)
+            yield API(name), ih.address
+            if name.startswith("_"):
+                # some linkers may prefix linked routines with a `_` to avoid name collisions.
+                # extract features for both the mangled and un-mangled representations.
+                # e.g. `_fwrite` -> `fwrite`
+                # see: https://stackoverflow.com/a/2628384/87207
+                yield API(name[1:]), ih.address
+            return
+
         for _ in range(THUNK_CHAIN_DEPTH_DELTA):
             if target in imports:
                 dll, symbol = imports[target]
                 for name in capa.features.extractors.helpers.generate_symbols(dll, symbol):
-                    yield API(name), insn.va
+                    yield API(name), ih.address
+
+            # if jump leads to an ENDBRANCH instruction, skip it
+            if f.vw.getByteDef(target)[1].startswith(b"\xf3\x0f\x1e"):
+                target += 4
 
             target = capa.features.extractors.viv.helpers.get_coderef_from(f.vw, target)
             if not target:
@@ -131,7 +165,7 @@ def extract_insn_api_features(f, bb, insn):
         if target in imports:
             dll, symbol = imports[target]
             for name in capa.features.extractors.helpers.generate_symbols(dll, symbol):
-                yield API(name), insn.va
+                yield API(name), ih.address
 
     elif isinstance(insn.opers[0], envi.archs.i386.disasm.i386RegOper):
         try:
@@ -148,38 +182,7 @@ def extract_insn_api_features(f, bb, insn):
         if target in imports:
             dll, symbol = imports[target]
             for name in capa.features.extractors.helpers.generate_symbols(dll, symbol):
-                yield API(name), insn.va
-
-
-def extract_insn_number_features(f, bb, insn):
-    """parse number features from the given instruction."""
-    # example:
-    #
-    #     push    3136B0h         ; dwControlCode
-    for oper in insn.opers:
-        # this is for both x32 and x64
-        if not isinstance(oper, (envi.archs.i386.disasm.i386ImmOper, envi.archs.i386.disasm.i386ImmMemOper)):
-            continue
-
-        if isinstance(oper, envi.archs.i386.disasm.i386ImmOper):
-            v = oper.getOperValue(oper)
-        else:
-            v = oper.getOperAddr(oper)
-
-        if f.vw.probeMemory(v, 1, envi.memory.MM_READ):
-            # this is a valid address
-            # assume its not also a constant.
-            continue
-
-        if insn.mnem == "add" and insn.opers[0].isReg() and insn.opers[0].reg == envi.archs.i386.disasm.REG_ESP:
-            # skip things like:
-            #
-            #    .text:00401140                 call    sub_407E2B
-            #    .text:00401145                 add     esp, 0Ch
-            return
-
-        yield Number(v), insn.va
-        yield Number(v, arch=get_arch(f.vw)), insn.va
+                yield API(name), ih.address
 
 
 def derefs(vw, p):
@@ -193,8 +196,13 @@ def derefs(vw, p):
     while True:
         if not vw.isValidPointer(p):
             return
+
         yield p
 
+        if vw.isProbablyString(p) or vw.isProbablyUnicode(p):
+            # don't deref strings that coincidentally are pointers
+            return
+
         try:
             next = vw.readMemoryPtr(p)
         except Exception:
@@ -214,7 +222,7 @@ def derefs(vw, p):
         p = next
 
 
-def read_memory(vw, va, size):
+def read_memory(vw, va: int, size: int) -> bytes:
     # as documented in #176, vivisect will not readMemory() when the section is not marked readable.
     #
     # but here, we don't care about permissions.
@@ -227,10 +235,10 @@ def read_memory(vw, va, size):
             mva, msize, mperms, mfname = mmap
             offset = va - mva
             return mbytes[offset : offset + size]
-    raise envi.SegmentationViolation(va)
+    raise envi.exc.SegmentationViolation(va)
 
 
-def read_bytes(vw, va):
+def read_bytes(vw, va: int) -> bytes:
     """
     read up to MAX_BYTES_FEATURE_SIZE from the given address.
 
@@ -239,7 +247,7 @@ def read_bytes(vw, va):
     """
     segm = vw.getSegment(va)
     if not segm:
-        raise envi.SegmentationViolation(va)
+        raise envi.exc.SegmentationViolation(va)
 
     segm_end = segm[0] + segm[1]
     try:
@@ -248,16 +256,19 @@ def read_bytes(vw, va):
             return read_memory(vw, va, segm_end - va)
         else:
             return read_memory(vw, va, MAX_BYTES_FEATURE_SIZE)
-    except envi.SegmentationViolation:
+    except envi.exc.SegmentationViolation:
         raise
 
 
-def extract_insn_bytes_features(f, bb, insn):
+def extract_insn_bytes_features(fh: FunctionHandle, bb, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
     """
     parse byte sequence features from the given instruction.
     example:
         #     push    offset iid_004118d4_IShellLinkA ; riid
     """
+    insn: envi.Opcode = ih.inner
+    f: viv_utils.Function = fh.inner
+
     if insn.mnem == "call":
         return
 
@@ -277,30 +288,39 @@ def extract_insn_bytes_features(f, bb, insn):
         else:
             continue
 
-        for v in derefs(f.vw, v):
+        for vv in derefs(f.vw, v):
             try:
-                buf = read_bytes(f.vw, v)
-            except envi.SegmentationViolation:
+                buf = read_bytes(f.vw, vv)
+            except envi.exc.SegmentationViolation:
                 continue
 
             if capa.features.extractors.helpers.all_zeros(buf):
                 continue
 
-            yield Bytes(buf), insn.va
+            if f.vw.isProbablyString(vv) or f.vw.isProbablyUnicode(vv):
+                # don't extract byte features for obvious strings
+                continue
+
+            yield Bytes(buf), ih.address
 
 
-def read_string(vw, offset):
+def read_string(vw, offset: int) -> str:
     try:
         alen = vw.detectString(offset)
-    except envi.SegmentationViolation:
+    except envi.exc.SegmentationViolation:
         pass
     else:
         if alen > 0:
-            return read_memory(vw, offset, alen).decode("utf-8")
+            buf = read_memory(vw, offset, alen)
+            if b"\x00" in buf:
+                # account for bug #1271.
+                # remove when vivisect is fixed.
+                buf = buf.partition(b"\x00")[0]
+            return buf.decode("utf-8")
 
     try:
         ulen = vw.detectUnicode(offset)
-    except envi.SegmentationViolation:
+    except envi.exc.SegmentationViolation:
         pass
     except IndexError:
         # potential vivisect bug detecting Unicode at segment end
@@ -315,92 +335,24 @@ def read_string(vw, offset):
                 # vivisect seems to mis-detect the end unicode strings
                 # off by two, too short
                 ulen += 2
-            return read_memory(vw, offset, ulen).decode("utf-16")
+            # partition to account for bug #1271.
+            # remove when vivisect is fixed.
+            return read_memory(vw, offset, ulen).decode("utf-16").partition("\x00")[0]
 
     raise ValueError("not a string", offset)
 
 
-def extract_insn_string_features(f, bb, insn):
-    """parse string features from the given instruction."""
-    # example:
-    #
-    #     push    offset aAcr     ; "ACR  > "
-
-    for oper in insn.opers:
-        if isinstance(oper, envi.archs.i386.disasm.i386ImmOper):
-            v = oper.getOperValue(oper)
-        elif isinstance(oper, envi.archs.i386.disasm.i386ImmMemOper):
-            # like 0x10056CB4 in `lea eax, dword [0x10056CB4]`
-            v = oper.imm
-        elif isinstance(oper, envi.archs.i386.disasm.i386SibOper):
-            # like 0x401000 in `mov eax, 0x401000[2 * ebx]`
-            v = oper.imm
-        elif isinstance(oper, envi.archs.amd64.disasm.Amd64RipRelOper):
-            v = oper.getOperAddr(insn)
-        else:
-            continue
-
-        for v in derefs(f.vw, v):
-            try:
-                s = read_string(f.vw, v)
-            except ValueError:
-                continue
-            else:
-                yield String(s.rstrip("\x00")), insn.va
-
-
-def extract_insn_offset_features(f, bb, insn):
-    """parse structure offset features from the given instruction."""
-    # example:
-    #
-    #     .text:0040112F    cmp     [esi+4], ebx
-    for oper in insn.opers:
-
-        # this is for both x32 and x64
-        # like [esi + 4]
-        #       reg   ^
-        #             disp
-        if isinstance(oper, envi.archs.i386.disasm.i386RegMemOper):
-            if oper.reg == envi.archs.i386.disasm.REG_ESP:
-                continue
-
-            if oper.reg == envi.archs.i386.disasm.REG_EBP:
-                continue
-
-            # TODO: do x64 support for real.
-            if oper.reg == envi.archs.amd64.disasm.REG_RBP:
-                continue
-
-            # viv already decodes offsets as signed
-            v = oper.disp
-
-            yield Offset(v), insn.va
-            yield Offset(v, arch=get_arch(f.vw)), insn.va
-
-        # like: [esi + ecx + 16384]
-        #        reg   ^     ^
-        #              index ^
-        #                    disp
-        elif isinstance(oper, envi.archs.i386.disasm.i386SibOper):
-            # viv already decodes offsets as signed
-            v = oper.disp
-
-            yield Offset(v), insn.va
-            yield Offset(v, arch=get_arch(f.vw)), insn.va
-
-
-def is_security_cookie(f, bb, insn):
+def is_security_cookie(f, bb, insn) -> bool:
     """
     check if an instruction is related to security cookie checks
     """
     # security cookie check should use SP or BP
     oper = insn.opers[1]
     if oper.isReg() and oper.reg not in [
-        envi.archs.i386.disasm.REG_ESP,
-        envi.archs.i386.disasm.REG_EBP,
-        # TODO: do x64 support for real.
-        envi.archs.amd64.disasm.REG_RBP,
-        envi.archs.amd64.disasm.REG_RSP,
+        envi.archs.i386.regs.REG_ESP,
+        envi.archs.i386.regs.REG_EBP,
+        envi.archs.amd64.regs.REG_RBP,
+        envi.archs.amd64.regs.REG_RSP,
     ]:
         return False
 
@@ -417,11 +369,17 @@ def is_security_cookie(f, bb, insn):
     return False
 
 
-def extract_insn_nzxor_characteristic_features(f, bb, insn):
+def extract_insn_nzxor_characteristic_features(
+    fh: FunctionHandle, bbhandle: BBHandle, ih: InsnHandle
+) -> Iterator[Tuple[Feature, Address]]:
     """
     parse non-zeroing XOR instruction from the given instruction.
     ignore expected non-zeroing XORs, e.g. security cookies.
     """
+    insn: envi.Opcode = ih.inner
+    bb: viv_utils.BasicBlock = bbhandle.inner
+    f: viv_utils.Function = fh.inner
+
     if insn.mnem not in ("xor", "xorpd", "xorps", "pxor"):
         return
 
@@ -431,19 +389,37 @@ def extract_insn_nzxor_characteristic_features(f, bb, insn):
     if is_security_cookie(f, bb, insn):
         return
 
-    yield Characteristic("nzxor"), insn.va
+    yield Characteristic("nzxor"), ih.address
 
 
-def extract_insn_mnemonic_features(f, bb, insn):
+def extract_insn_mnemonic_features(f, bb, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
     """parse mnemonic features from the given instruction."""
-    yield Mnemonic(insn.mnem), insn.va
+    yield Mnemonic(ih.inner.mnem), ih.address
 
 
-def extract_insn_peb_access_characteristic_features(f, bb, insn):
+def extract_insn_obfs_call_plus_5_characteristic_features(f, bb, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
+    """
+    parse call $+5 instruction from the given instruction.
+    """
+    insn: envi.Opcode = ih.inner
+
+    if insn.mnem != "call":
+        return
+
+    if isinstance(insn.opers[0], envi.archs.i386.disasm.i386PcRelOper):
+        if insn.va + 5 == insn.opers[0].getOperValue(insn):
+            yield Characteristic("call $+5"), ih.address
+
+    if isinstance(insn.opers[0], (envi.archs.i386.disasm.i386ImmMemOper, envi.archs.amd64.disasm.Amd64RipRelOper)):
+        if insn.va + 5 == insn.opers[0].getOperAddr(insn):
+            yield Characteristic("call $+5"), ih.address
+
+
+def extract_insn_peb_access_characteristic_features(f, bb, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
     """
     parse peb access from the given function. fs:[0x30] on x86, gs:[0x60] on x64
     """
-    # TODO handle where fs/gs are loaded into a register or onto the stack and used later
+    insn: envi.Opcode = ih.inner
 
     if insn.mnem not in ["push", "mov"]:
         return
@@ -462,7 +438,7 @@ def extract_insn_peb_access_characteristic_features(f, bb, insn):
             if (isinstance(oper, envi.archs.i386.disasm.i386RegMemOper) and oper.disp == 0x30) or (
                 isinstance(oper, envi.archs.i386.disasm.i386ImmMemOper) and oper.imm == 0x30
             ):
-                yield Characteristic("peb access"), insn.va
+                yield Characteristic("peb access"), ih.address
     elif "gs" in prefix:
         for oper in insn.opers:
             if (
@@ -470,23 +446,25 @@ def extract_insn_peb_access_characteristic_features(f, bb, insn):
                 or (isinstance(oper, envi.archs.amd64.disasm.i386SibOper) and oper.imm == 0x60)
                 or (isinstance(oper, envi.archs.amd64.disasm.i386ImmMemOper) and oper.imm == 0x60)
             ):
-                yield Characteristic("peb access"), insn.va
+                yield Characteristic("peb access"), ih.address
     else:
         pass
 
 
-def extract_insn_segment_access_features(f, bb, insn):
-    """ parse the instruction for access to fs or gs """
+def extract_insn_segment_access_features(f, bb, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
+    """parse the instruction for access to fs or gs"""
+    insn: envi.Opcode = ih.inner
+
     prefix = insn.getPrefixName()
 
     if prefix == "fs":
-        yield Characteristic("fs access"), insn.va
+        yield Characteristic("fs access"), ih.address
 
     if prefix == "gs":
-        yield Characteristic("gs access"), insn.va
+        yield Characteristic("gs access"), ih.address
 
 
-def get_section(vw, va):
+def get_section(vw, va: int):
     for start, length, _, __ in vw.getMemoryMaps():
         if start <= va < start + length:
             return start
@@ -494,10 +472,13 @@ def get_section(vw, va):
     raise KeyError(va)
 
 
-def extract_insn_cross_section_cflow(f, bb, insn):
+def extract_insn_cross_section_cflow(fh: FunctionHandle, bb, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
     """
     inspect the instruction for a CALL or JMP that crosses section boundaries.
     """
+    insn: envi.Opcode = ih.inner
+    f: viv_utils.Function = fh.inner
+
     for va, flags in insn.getBranches():
         if va is None:
             # va may be none for dynamic branches that haven't been resolved, such as `jmp eax`.
@@ -524,7 +505,7 @@ def extract_insn_cross_section_cflow(f, bb, insn):
                     continue
 
             if get_section(f.vw, insn.va) != get_section(f.vw, va):
-                yield Characteristic("cross section flow"), insn.va
+                yield Characteristic("cross section flow"), ih.address
 
         except KeyError:
             continue
@@ -532,7 +513,10 @@ def extract_insn_cross_section_cflow(f, bb, insn):
 
 # this is a feature that's most relevant at the function scope,
 # however, its most efficient to extract at the instruction scope.
-def extract_function_calls_from(f, bb, insn):
+def extract_function_calls_from(fh: FunctionHandle, bb, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
+    insn: envi.Opcode = ih.inner
+    f: viv_utils.Function = fh.inner
+
     if insn.mnem != "call":
         return
 
@@ -542,7 +526,8 @@ def extract_function_calls_from(f, bb, insn):
     if isinstance(insn.opers[0], envi.archs.i386.disasm.i386ImmMemOper):
         oper = insn.opers[0]
         target = oper.getOperAddr(insn)
-        yield Characteristic("calls from"), target
+        if target >= 0:
+            yield Characteristic("calls from"), AbsoluteVirtualAddress(target)
 
     # call via thunk on x86,
     # see 9324d1a8ae37a36ae560c37448c9705a at 0x407985
@@ -551,43 +536,192 @@ def extract_function_calls_from(f, bb, insn):
     # see Lab21-01.exe_:0x140001178
     elif isinstance(insn.opers[0], envi.archs.i386.disasm.i386PcRelOper):
         target = insn.opers[0].getOperValue(insn)
-        yield Characteristic("calls from"), target
+        if target >= 0:
+            yield Characteristic("calls from"), AbsoluteVirtualAddress(target)
 
     # call via IAT, x64
     elif isinstance(insn.opers[0], envi.archs.amd64.disasm.Amd64RipRelOper):
         op = insn.opers[0]
         target = op.getOperAddr(insn)
-        yield Characteristic("calls from"), target
+        if target >= 0:
+            yield Characteristic("calls from"), AbsoluteVirtualAddress(target)
 
     if target and target == f.va:
         # if we found a jump target and it's the function address
         # mark as recursive
-        yield Characteristic("recursive call"), target
+        yield Characteristic("recursive call"), AbsoluteVirtualAddress(target)
 
 
 # this is a feature that's most relevant at the function or basic block scope,
 # however, its most efficient to extract at the instruction scope.
-def extract_function_indirect_call_characteristic_features(f, bb, insn):
+def extract_function_indirect_call_characteristic_features(f, bb, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
     """
     extract indirect function call characteristic (e.g., call eax or call dword ptr [edx+4])
     does not include calls like => call ds:dword_ABD4974
     """
+    insn: envi.Opcode = ih.inner
+
     if insn.mnem != "call":
         return
 
     # Checks below work for x86 and x64
     if isinstance(insn.opers[0], envi.archs.i386.disasm.i386RegOper):
         # call edx
-        yield Characteristic("indirect call"), insn.va
+        yield Characteristic("indirect call"), ih.address
     elif isinstance(insn.opers[0], envi.archs.i386.disasm.i386RegMemOper):
         # call dword ptr [eax+50h]
-        yield Characteristic("indirect call"), insn.va
+        yield Characteristic("indirect call"), ih.address
     elif isinstance(insn.opers[0], envi.archs.i386.disasm.i386SibOper):
         # call qword ptr [rsp+78h]
-        yield Characteristic("indirect call"), insn.va
+        yield Characteristic("indirect call"), ih.address
 
 
-def extract_features(f, bb, insn):
+def extract_op_number_features(
+    fh: FunctionHandle, bb, ih: InsnHandle, i, oper: envi.Operand
+) -> Iterator[Tuple[Feature, Address]]:
+    """parse number features from the given operand.
+
+    example:
+        push    3136B0h         ; dwControlCode
+    """
+    insn: envi.Opcode = ih.inner
+    f: viv_utils.Function = fh.inner
+
+    # this is for both x32 and x64
+    if not isinstance(oper, (envi.archs.i386.disasm.i386ImmOper, envi.archs.i386.disasm.i386ImmMemOper)):
+        return
+
+    if isinstance(oper, envi.archs.i386.disasm.i386ImmOper):
+        v = oper.getOperValue(oper)
+    else:
+        v = oper.getOperAddr(oper)
+
+    if f.vw.probeMemory(v, 1, envi.memory.MM_READ):
+        # this is a valid address
+        # assume its not also a constant.
+        return
+
+    if insn.mnem == "add" and insn.opers[0].isReg() and insn.opers[0].reg == envi.archs.i386.regs.REG_ESP:
+        # skip things like:
+        #
+        #    .text:00401140                 call    sub_407E2B
+        #    .text:00401145                 add     esp, 0Ch
+        return
+
+    yield Number(v), ih.address
+    yield OperandNumber(i, v), ih.address
+
+    if insn.mnem == "add" and 0 < v < MAX_STRUCTURE_SIZE and isinstance(oper, envi.archs.i386.disasm.i386ImmOper):
+        # for pattern like:
+        #
+        #     add eax, 0x10
+        #
+        # assume 0x10 is also an offset (imagine eax is a pointer).
+        yield Offset(v), ih.address
+        yield OperandOffset(i, v), ih.address
+
+
+def extract_op_offset_features(
+    fh: FunctionHandle, bb, ih: InsnHandle, i, oper: envi.Operand
+) -> Iterator[Tuple[Feature, Address]]:
+    """parse structure offset features from the given operand."""
+    # example:
+    #
+    #     .text:0040112F    cmp     [esi+4], ebx
+    insn: envi.Opcode = ih.inner
+    f: viv_utils.Function = fh.inner
+
+    # this is for both x32 and x64
+    # like [esi + 4]
+    #       reg   ^
+    #             disp
+    if isinstance(oper, envi.archs.i386.disasm.i386RegMemOper):
+        if oper.reg == envi.archs.i386.regs.REG_ESP:
+            return
+
+        if oper.reg == envi.archs.i386.regs.REG_EBP:
+            return
+
+        if oper.reg == envi.archs.amd64.regs.REG_RBP:
+            return
+
+        # viv already decodes offsets as signed
+        v = oper.disp
+
+        yield Offset(v), ih.address
+        yield OperandOffset(i, v), ih.address
+
+        if insn.mnem == "lea" and i == 1 and not f.vw.probeMemory(v, 1, envi.memory.MM_READ):
+            # for pattern like:
+            #
+            #     lea eax, [ebx + 1]
+            #
+            # assume 1 is also an offset (imagine ebx is a zero register).
+            yield Number(v), ih.address
+            yield OperandNumber(i, v), ih.address
+
+    # like: [esi + ecx + 16384]
+    #        reg   ^     ^
+    #              index ^
+    #                    disp
+    elif isinstance(oper, envi.archs.i386.disasm.i386SibOper):
+        # viv already decodes offsets as signed
+        v = oper.disp
+
+        yield Offset(v), ih.address
+        yield OperandOffset(i, v), ih.address
+
+
+def extract_op_string_features(
+    fh: FunctionHandle, bb, ih: InsnHandle, i, oper: envi.Operand
+) -> Iterator[Tuple[Feature, Address]]:
+    """parse string features from the given operand."""
+    # example:
+    #
+    #     push    offset aAcr     ; "ACR  > "
+    insn: envi.Opcode = ih.inner
+    f: viv_utils.Function = fh.inner
+
+    if isinstance(oper, envi.archs.i386.disasm.i386ImmOper):
+        v = oper.getOperValue(oper)
+    elif isinstance(oper, envi.archs.i386.disasm.i386ImmMemOper):
+        # like 0x10056CB4 in `lea eax, dword [0x10056CB4]`
+        v = oper.imm
+    elif isinstance(oper, envi.archs.i386.disasm.i386SibOper):
+        # like 0x401000 in `mov eax, 0x401000[2 * ebx]`
+        v = oper.imm
+    elif isinstance(oper, envi.archs.amd64.disasm.Amd64RipRelOper):
+        v = oper.getOperAddr(insn)
+    else:
+        return
+
+    for vv in derefs(f.vw, v):
+        try:
+            s = read_string(f.vw, vv).rstrip("\x00")
+        except ValueError:
+            continue
+        else:
+            if len(s) >= 4:
+                yield String(s), ih.address
+
+
+def extract_operand_features(f: FunctionHandle, bb, insn: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
+    for i, oper in enumerate(insn.inner.opers):
+        for op_handler in OPERAND_HANDLERS:
+            for feature, addr in op_handler(f, bb, insn, i, oper):
+                yield feature, addr
+
+
+OPERAND_HANDLERS: List[
+    Callable[[FunctionHandle, BBHandle, InsnHandle, int, envi.Operand], Iterator[Tuple[Feature, Address]]]
+] = [
+    extract_op_number_features,
+    extract_op_offset_features,
+    extract_op_string_features,
+]
+
+
+def extract_features(f, bb, insn) -> Iterator[Tuple[Feature, Address]]:
     """
     extract features from the given insn.
 
@@ -597,24 +731,23 @@ def extract_features(f, bb, insn):
       insn (vivisect...Instruction): the instruction to process.
 
     yields:
-      Feature, set[VA]: the features and their location found in this insn.
+      Tuple[Feature, Address]: the features and their location found in this insn.
     """
     for insn_handler in INSTRUCTION_HANDLERS:
-        for feature, va in insn_handler(f, bb, insn):
-            yield feature, va
+        for feature, addr in insn_handler(f, bb, insn):
+            yield feature, addr
 
 
-INSTRUCTION_HANDLERS = (
+INSTRUCTION_HANDLERS: List[Callable[[FunctionHandle, BBHandle, InsnHandle], Iterator[Tuple[Feature, Address]]]] = [
     extract_insn_api_features,
-    extract_insn_number_features,
-    extract_insn_string_features,
     extract_insn_bytes_features,
-    extract_insn_offset_features,
     extract_insn_nzxor_characteristic_features,
     extract_insn_mnemonic_features,
+    extract_insn_obfs_call_plus_5_characteristic_features,
     extract_insn_peb_access_characteristic_features,
     extract_insn_cross_section_cflow,
     extract_insn_segment_access_features,
     extract_function_calls_from,
     extract_function_indirect_call_characteristic_features,
-)
+    extract_operand_features,
+]

capa/features/file.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
 # Licensed under the Apache License, Version 2.0 (the "License");
 #  you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at: [package root]/LICENSE.txt
@@ -6,22 +6,33 @@
 #  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and limitations under the License.
 
-from capa.features import Feature
+from capa.features.common import Feature
 
 
 class Export(Feature):
-    def __init__(self, value, description=None):
+    def __init__(self, value: str, description=None):
         # value is export name
-        super(Export, self).__init__(value, description=description)
+        super().__init__(value, description=description)
 
 
 class Import(Feature):
-    def __init__(self, value, description=None):
+    def __init__(self, value: str, description=None):
         # value is import name
-        super(Import, self).__init__(value, description=description)
+        super().__init__(value, description=description)
 
 
 class Section(Feature):
-    def __init__(self, value, description=None):
+    def __init__(self, value: str, description=None):
         # value is section name
-        super(Section, self).__init__(value, description=description)
+        super().__init__(value, description=description)
+
+
+class FunctionName(Feature):
+    """recognized name for statically linked function"""
+
+    def __init__(self, name: str, description=None):
+        # value is function name
+        super().__init__(name, description=description)
+        # override the name property set by `capa.features.Feature`
+        # that would be `functionname` (note missing dash)
+        self.name = "function-name"

capa/features/freeze.py

@@ -1,299 +0,0 @@
-"""
-capa freeze file format: `| capa0000 | + zlib(utf-8(json(...)))`
-
-json format:
-
-    {
-      'version': 1,
-      'base address': int(base address),
-      'functions': {
-        int(function va): {
-          'basic blocks': {
-            int(basic block va): {
-              'instructions': [instruction va, ...]
-            },
-            ...
-          },
-          ...
-        },
-        ...
-      },
-      'scopes': {
-        'file': [
-          (str(name), [any(arg), ...], int(va), ()),
-          ...
-        },
-        'function': [
-          (str(name), [any(arg), ...], int(va), (int(function va), )),
-          ...
-        ],
-        'basic block': [
-          (str(name), [any(arg), ...], int(va), (int(function va),
-                                                 int(basic block va))),
-          ...
-        ],
-        'instruction': [
-          (str(name), [any(arg), ...], int(va), (int(function va),
-                                                 int(basic block va),
-                                                 int(instruction va))),
-          ...
-        ],
-      }
-    }
-
-Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
-Licensed under the Apache License, Version 2.0 (the "License");
- you may not use this file except in compliance with the License.
-You may obtain a copy of the License at: [package root]/LICENSE.txt
-Unless required by applicable law or agreed to in writing, software distributed under the License
- is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-See the License for the specific language governing permissions and limitations under the License.
-"""
-import json
-import zlib
-import logging
-
-import capa.features
-import capa.features.file
-import capa.features.insn
-import capa.features.basicblock
-import capa.features.extractors
-from capa.helpers import hex
-
-logger = logging.getLogger(__name__)
-
-
-def serialize_feature(feature):
-    return feature.freeze_serialize()
-
-
-KNOWN_FEATURES = {F.__name__: F for F in capa.features.Feature.__subclasses__()}
-
-
-def deserialize_feature(doc):
-    F = KNOWN_FEATURES[doc[0]]
-    return F.freeze_deserialize(doc[1])
-
-
-def dumps(extractor):
-    """
-    serialize the given extractor to a string
-
-    args:
-      extractor: capa.features.extractor.FeatureExtractor:
-
-    returns:
-      str: the serialized features.
-    """
-    ret = {
-        "version": 1,
-        "base address": extractor.get_base_address(),
-        "functions": {},
-        "scopes": {
-            "file": [],
-            "function": [],
-            "basic block": [],
-            "instruction": [],
-        },
-    }
-
-    for feature, va in extractor.extract_file_features():
-        ret["scopes"]["file"].append(serialize_feature(feature) + (hex(va), ()))
-
-    for f in extractor.get_functions():
-        ret["functions"][hex(f)] = {}
-
-        for feature, va in extractor.extract_function_features(f):
-            ret["scopes"]["function"].append(serialize_feature(feature) + (hex(va), (hex(f),)))
-
-        for bb in extractor.get_basic_blocks(f):
-            ret["functions"][hex(f)][hex(bb)] = []
-
-            for feature, va in extractor.extract_basic_block_features(f, bb):
-                ret["scopes"]["basic block"].append(
-                    serialize_feature(feature)
-                    + (
-                        hex(va),
-                        (
-                            hex(f),
-                            hex(bb),
-                        ),
-                    )
-                )
-
-            for insnva, insn in sorted(
-                [(insn.__int__(), insn) for insn in extractor.get_instructions(f, bb)], key=lambda p: p[0]
-            ):
-                ret["functions"][hex(f)][hex(bb)].append(hex(insnva))
-
-                for feature, va in extractor.extract_insn_features(f, bb, insn):
-                    ret["scopes"]["instruction"].append(
-                        serialize_feature(feature)
-                        + (
-                            hex(va),
-                            (
-                                hex(f),
-                                hex(bb),
-                                hex(insnva),
-                            ),
-                        )
-                    )
-    return json.dumps(ret)
-
-
-def loads(s):
-    """deserialize a set of features (as a NullFeatureExtractor) from a string."""
-    doc = json.loads(s)
-
-    if doc.get("version") != 1:
-        raise ValueError("unsupported freeze format version: %d" % (doc.get("version")))
-
-    features = {
-        "base address": doc.get("base address"),
-        "file features": [],
-        "functions": {},
-    }
-
-    for fva, function in doc.get("functions", {}).items():
-        fva = int(fva, 0x10)
-        features["functions"][fva] = {
-            "features": [],
-            "basic blocks": {},
-        }
-
-        for bbva, bb in function.items():
-            bbva = int(bbva, 0x10)
-            features["functions"][fva]["basic blocks"][bbva] = {
-                "features": [],
-                "instructions": {},
-            }
-
-            for insnva in bb:
-                insnva = int(insnva, 0x10)
-                features["functions"][fva]["basic blocks"][bbva]["instructions"][insnva] = {
-                    "features": [],
-                }
-
-    # in the following blocks, each entry looks like:
-    #
-    #     ('MatchedRule', ('foo', ), '0x401000', ('0x401000', ))
-    #      ^^^^^^^^^^^^^  ^^^^^^^^^  ^^^^^^^^^^  ^^^^^^^^^^^^^^
-    #      feature name   args       addr         func/bb/insn
-    for feature in doc.get("scopes", {}).get("file", []):
-        va, loc = feature[2:]
-        va = int(va, 0x10)
-        feature = deserialize_feature(feature[:2])
-        features["file features"].append((va, feature))
-
-    for feature in doc.get("scopes", {}).get("function", []):
-        # fetch the pair like:
-        #
-        #     ('0x401000', ('0x401000', ))
-        #      ^^^^^^^^^^  ^^^^^^^^^^^^^^
-        #      addr         func/bb/insn
-        va, loc = feature[2:]
-        va = int(va, 0x10)
-        loc = [int(lo, 0x10) for lo in loc]
-
-        # decode the feature from the pair like:
-        #
-        #     ('MatchedRule', ('foo', ))
-        #      ^^^^^^^^^^^^^  ^^^^^^^^^
-        #      feature name   args
-        feature = deserialize_feature(feature[:2])
-        features["functions"][loc[0]]["features"].append((va, feature))
-
-    for feature in doc.get("scopes", {}).get("basic block", []):
-        va, loc = feature[2:]
-        va = int(va, 0x10)
-        loc = [int(lo, 0x10) for lo in loc]
-        feature = deserialize_feature(feature[:2])
-        features["functions"][loc[0]]["basic blocks"][loc[1]]["features"].append((va, feature))
-
-    for feature in doc.get("scopes", {}).get("instruction", []):
-        va, loc = feature[2:]
-        va = int(va, 0x10)
-        loc = [int(lo, 0x10) for lo in loc]
-        feature = deserialize_feature(feature[:2])
-        features["functions"][loc[0]]["basic blocks"][loc[1]]["instructions"][loc[2]]["features"].append((va, feature))
-
-    return capa.features.extractors.NullFeatureExtractor(features)
-
-
-MAGIC = "capa0000".encode("ascii")
-
-
-def dump(extractor):
-    """serialize the given extractor to a byte array."""
-    return MAGIC + zlib.compress(dumps(extractor).encode("utf-8"))
-
-
-def is_freeze(buf):
-    return buf[: len(MAGIC)] == MAGIC
-
-
-def load(buf):
-    """deserialize a set of features (as a NullFeatureExtractor) from a byte array."""
-    if not is_freeze(buf):
-        raise ValueError("missing magic header")
-    return loads(zlib.decompress(buf[len(MAGIC) :]).decode("utf-8"))
-
-
-def main(argv=None):
-    import sys
-    import argparse
-
-    import capa.main
-
-    if argv is None:
-        argv = sys.argv[1:]
-
-    formats = [
-        ("auto", "(default) detect file type automatically"),
-        ("pe", "Windows PE file"),
-        ("sc32", "32-bit shellcode"),
-        ("sc64", "64-bit shellcode"),
-    ]
-    format_help = ", ".join(["%s: %s" % (f[0], f[1]) for f in formats])
-
-    parser = argparse.ArgumentParser(description="save capa features to a file")
-    parser.add_argument("sample", type=str, help="Path to sample to analyze")
-    parser.add_argument("output", type=str, help="Path to output file")
-    parser.add_argument("-v", "--verbose", action="store_true", help="Enable verbose output")
-    parser.add_argument("-q", "--quiet", action="store_true", help="Disable all output but errors")
-    parser.add_argument(
-        "-f", "--format", choices=[f[0] for f in formats], default="auto", help="Select sample format, %s" % format_help
-    )
-    if sys.version_info >= (3, 0):
-        parser.add_argument(
-            "-b",
-            "--backend",
-            type=str,
-            help="select the backend to use",
-            choices=(capa.main.BACKEND_VIV, capa.main.BACKEND_SMDA),
-            default=capa.main.BACKEND_VIV,
-        )
-    args = parser.parse_args(args=argv)
-
-    if args.quiet:
-        logging.basicConfig(level=logging.ERROR)
-        logging.getLogger().setLevel(logging.ERROR)
-    elif args.verbose:
-        logging.basicConfig(level=logging.DEBUG)
-        logging.getLogger().setLevel(logging.DEBUG)
-    else:
-        logging.basicConfig(level=logging.INFO)
-        logging.getLogger().setLevel(logging.INFO)
-
-    backend = args.backend if sys.version_info > (3, 0) else capa.main.BACKEND_VIV
-    extractor = capa.main.get_extractor(args.sample, args.format, backend)
-    with open(args.output, "wb") as f:
-        f.write(dump(extractor))
-
-    return 0
-
-
-if __name__ == "__main__":
-    import sys
-
-    sys.exit(main())

capa/features/freeze/__init__.py

@@ -0,0 +1,399 @@
+"""
+capa freeze file format: `| capa0000 | + zlib(utf-8(json(...)))`
+
+Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+Licensed under the Apache License, Version 2.0 (the "License");
+ you may not use this file except in compliance with the License.
+You may obtain a copy of the License at: [package root]/LICENSE.txt
+Unless required by applicable law or agreed to in writing, software distributed under the License
+ is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and limitations under the License.
+"""
+import zlib
+import logging
+from enum import Enum
+from typing import List, Tuple, Union
+
+from pydantic import Field, BaseModel, ConfigDict
+
+import capa.helpers
+import capa.version
+import capa.features.file
+import capa.features.insn
+import capa.features.common
+import capa.features.address
+import capa.features.basicblock
+import capa.features.extractors.base_extractor
+from capa.helpers import assert_never
+from capa.features.freeze.features import Feature, feature_from_capa
+
+logger = logging.getLogger(__name__)
+
+
+class HashableModel(BaseModel):
+    model_config = ConfigDict(frozen=True)
+
+
+class AddressType(str, Enum):
+    ABSOLUTE = "absolute"
+    RELATIVE = "relative"
+    FILE = "file"
+    DN_TOKEN = "dn token"
+    DN_TOKEN_OFFSET = "dn token offset"
+    NO_ADDRESS = "no address"
+
+
+class Address(HashableModel):
+    type: AddressType
+    value: Union[int, Tuple[int, int], None] = None  # None default value to support deserialization of NO_ADDRESS
+
+    @classmethod
+    def from_capa(cls, a: capa.features.address.Address) -> "Address":
+        if isinstance(a, capa.features.address.AbsoluteVirtualAddress):
+            return cls(type=AddressType.ABSOLUTE, value=int(a))
+
+        elif isinstance(a, capa.features.address.RelativeVirtualAddress):
+            return cls(type=AddressType.RELATIVE, value=int(a))
+
+        elif isinstance(a, capa.features.address.FileOffsetAddress):
+            return cls(type=AddressType.FILE, value=int(a))
+
+        elif isinstance(a, capa.features.address.DNTokenAddress):
+            return cls(type=AddressType.DN_TOKEN, value=int(a))
+
+        elif isinstance(a, capa.features.address.DNTokenOffsetAddress):
+            return cls(type=AddressType.DN_TOKEN_OFFSET, value=(a.token, a.offset))
+
+        elif a == capa.features.address.NO_ADDRESS or isinstance(a, capa.features.address._NoAddress):
+            return cls(type=AddressType.NO_ADDRESS, value=None)
+
+        elif isinstance(a, capa.features.address.Address) and not issubclass(type(a), capa.features.address.Address):
+            raise ValueError("don't use an Address instance directly")
+
+        elif isinstance(a, capa.features.address.Address):
+            raise ValueError("don't use an Address instance directly")
+
+        else:
+            assert_never(a)
+
+    def to_capa(self) -> capa.features.address.Address:
+        if self.type is AddressType.ABSOLUTE:
+            assert isinstance(self.value, int)
+            return capa.features.address.AbsoluteVirtualAddress(self.value)
+
+        elif self.type is AddressType.RELATIVE:
+            assert isinstance(self.value, int)
+            return capa.features.address.RelativeVirtualAddress(self.value)
+
+        elif self.type is AddressType.FILE:
+            assert isinstance(self.value, int)
+            return capa.features.address.FileOffsetAddress(self.value)
+
+        elif self.type is AddressType.DN_TOKEN:
+            assert isinstance(self.value, int)
+            return capa.features.address.DNTokenAddress(self.value)
+
+        elif self.type is AddressType.DN_TOKEN_OFFSET:
+            assert isinstance(self.value, tuple)
+            token, offset = self.value
+            assert isinstance(token, int)
+            assert isinstance(offset, int)
+            return capa.features.address.DNTokenOffsetAddress(token, offset)
+
+        elif self.type is AddressType.NO_ADDRESS:
+            return capa.features.address.NO_ADDRESS
+
+        else:
+            assert_never(self.type)
+
+    def __lt__(self, other: "Address") -> bool:
+        if self.type != other.type:
+            return self.type < other.type
+
+        if self.type is AddressType.NO_ADDRESS:
+            return True
+
+        else:
+            assert self.type == other.type
+            # mypy doesn't realize we've proven that either
+            # both are ints, or both are tuples of ints.
+            # and both of these are comparable.
+            return self.value < other.value  # type: ignore
+
+
+class GlobalFeature(HashableModel):
+    feature: Feature
+
+
+class FileFeature(HashableModel):
+    address: Address
+    feature: Feature
+
+
+class FunctionFeature(HashableModel):
+    """
+    args:
+        function: the address of the function to which this feature belongs.
+        address: the address at which this feature is found.
+
+    function != address because, e.g., the feature may be found *within* the scope (function).
+    versus right at its starting address.
+    """
+
+    function: Address
+    address: Address
+    feature: Feature
+
+
+class BasicBlockFeature(HashableModel):
+    """
+    args:
+        basic_block: the address of the basic block to which this feature belongs.
+        address: the address at which this feature is found.
+
+    basic_block != address because, e.g., the feature may be found *within* the scope (basic block).
+    versus right at its starting address.
+    """
+
+    basic_block: Address = Field(alias="basic block")
+    address: Address
+    feature: Feature
+    model_config = ConfigDict(populate_by_name=True)
+
+
+class InstructionFeature(HashableModel):
+    """
+    args:
+        instruction: the address of the instruction to which this feature belongs.
+        address: the address at which this feature is found.
+
+    instruction != address because, e.g., the feature may be found *within* the scope (basic block),
+    versus right at its starting address.
+    """
+
+    instruction: Address
+    address: Address
+    feature: Feature
+
+
+class InstructionFeatures(BaseModel):
+    address: Address
+    features: Tuple[InstructionFeature, ...]
+
+
+class BasicBlockFeatures(BaseModel):
+    address: Address
+    features: Tuple[BasicBlockFeature, ...]
+    instructions: Tuple[InstructionFeatures, ...]
+
+
+class FunctionFeatures(BaseModel):
+    address: Address
+    features: Tuple[FunctionFeature, ...]
+    basic_blocks: Tuple[BasicBlockFeatures, ...] = Field(alias="basic blocks")
+    model_config = ConfigDict(populate_by_name=True)
+
+
+class Features(BaseModel):
+    global_: Tuple[GlobalFeature, ...] = Field(alias="global")
+    file: Tuple[FileFeature, ...]
+    functions: Tuple[FunctionFeatures, ...]
+    model_config = ConfigDict(populate_by_name=True)
+
+
+class Extractor(BaseModel):
+    name: str
+    version: str = capa.version.__version__
+    model_config = ConfigDict(populate_by_name=True)
+
+
+class Freeze(BaseModel):
+    version: int = 2
+    base_address: Address = Field(alias="base address")
+    extractor: Extractor
+    features: Features
+    model_config = ConfigDict(populate_by_name=True)
+
+
+def dumps(extractor: capa.features.extractors.base_extractor.FeatureExtractor) -> str:
+    """
+    serialize the given extractor to a string
+    """
+
+    global_features: List[GlobalFeature] = []
+    for feature, _ in extractor.extract_global_features():
+        global_features.append(
+            GlobalFeature(
+                feature=feature_from_capa(feature),
+            )
+        )
+
+    file_features: List[FileFeature] = []
+    for feature, address in extractor.extract_file_features():
+        file_features.append(
+            FileFeature(
+                feature=feature_from_capa(feature),
+                address=Address.from_capa(address),
+            )
+        )
+
+    function_features: List[FunctionFeatures] = []
+    for f in extractor.get_functions():
+        faddr = Address.from_capa(f.address)
+        ffeatures = [
+            FunctionFeature(
+                function=faddr,
+                address=Address.from_capa(addr),
+                feature=feature_from_capa(feature),
+            )
+            for feature, addr in extractor.extract_function_features(f)
+        ]
+
+        basic_blocks = []
+        for bb in extractor.get_basic_blocks(f):
+            bbaddr = Address.from_capa(bb.address)
+            bbfeatures = [
+                BasicBlockFeature(
+                    basic_block=bbaddr,
+                    address=Address.from_capa(addr),
+                    feature=feature_from_capa(feature),
+                )  # type: ignore
+                # Mypy is unable to recognise `basic_block` as a argument due to alias
+                for feature, addr in extractor.extract_basic_block_features(f, bb)
+            ]
+
+            instructions = []
+            for insn in extractor.get_instructions(f, bb):
+                iaddr = Address.from_capa(insn.address)
+                ifeatures = [
+                    InstructionFeature(
+                        instruction=iaddr,
+                        address=Address.from_capa(addr),
+                        feature=feature_from_capa(feature),
+                    )
+                    for feature, addr in extractor.extract_insn_features(f, bb, insn)
+                ]
+
+                instructions.append(
+                    InstructionFeatures(
+                        address=iaddr,
+                        features=tuple(ifeatures),
+                    )
+                )
+
+            basic_blocks.append(
+                BasicBlockFeatures(
+                    address=bbaddr,
+                    features=tuple(bbfeatures),
+                    instructions=tuple(instructions),
+                )
+            )
+
+        function_features.append(
+            FunctionFeatures(
+                address=faddr,
+                features=tuple(ffeatures),
+                basic_blocks=basic_blocks,
+            )  # type: ignore
+            # Mypy is unable to recognise `basic_blocks` as a argument due to alias
+        )
+
+    features = Features(
+        global_=global_features,
+        file=tuple(file_features),
+        functions=tuple(function_features),
+    )  # type: ignore
+    # Mypy is unable to recognise `global_` as a argument due to alias
+
+    freeze = Freeze(
+        version=2,
+        base_address=Address.from_capa(extractor.get_base_address()),
+        extractor=Extractor(name=extractor.__class__.__name__),
+        features=features,
+    )  # type: ignore
+    # Mypy is unable to recognise `base_address` as a argument due to alias
+
+    return freeze.model_dump_json()
+
+
+def loads(s: str) -> capa.features.extractors.base_extractor.FeatureExtractor:
+    """deserialize a set of features (as a NullFeatureExtractor) from a string."""
+    import capa.features.extractors.null as null
+
+    freeze = Freeze.model_validate_json(s)
+    if freeze.version != 2:
+        raise ValueError(f"unsupported freeze format version: {freeze.version}")
+
+    return null.NullFeatureExtractor(
+        base_address=freeze.base_address.to_capa(),
+        global_features=[f.feature.to_capa() for f in freeze.features.global_],
+        file_features=[(f.address.to_capa(), f.feature.to_capa()) for f in freeze.features.file],
+        functions={
+            f.address.to_capa(): null.FunctionFeatures(
+                features=[(fe.address.to_capa(), fe.feature.to_capa()) for fe in f.features],
+                basic_blocks={
+                    bb.address.to_capa(): null.BasicBlockFeatures(
+                        features=[(fe.address.to_capa(), fe.feature.to_capa()) for fe in bb.features],
+                        instructions={
+                            i.address.to_capa(): null.InstructionFeatures(
+                                features=[(fe.address.to_capa(), fe.feature.to_capa()) for fe in i.features]
+                            )
+                            for i in bb.instructions
+                        },
+                    )
+                    for bb in f.basic_blocks
+                },
+            )
+            for f in freeze.features.functions
+        },
+    )
+
+
+MAGIC = "capa0000".encode("ascii")
+
+
+def dump(extractor: capa.features.extractors.base_extractor.FeatureExtractor) -> bytes:
+    """serialize the given extractor to a byte array."""
+    return MAGIC + zlib.compress(dumps(extractor).encode("utf-8"))
+
+
+def is_freeze(buf: bytes) -> bool:
+    return buf[: len(MAGIC)] == MAGIC
+
+
+def load(buf: bytes) -> capa.features.extractors.base_extractor.FeatureExtractor:
+    """deserialize a set of features (as a NullFeatureExtractor) from a byte array."""
+    if not is_freeze(buf):
+        raise ValueError("missing magic header")
+    return loads(zlib.decompress(buf[len(MAGIC) :]).decode("utf-8"))
+
+
+def main(argv=None):
+    import sys
+    import argparse
+    from pathlib import Path
+
+    import capa.main
+
+    if argv is None:
+        argv = sys.argv[1:]
+
+    parser = argparse.ArgumentParser(description="save capa features to a file")
+    capa.main.install_common_args(parser, {"sample", "format", "backend", "os", "signatures"})
+    parser.add_argument("output", type=str, help="Path to output file")
+    args = parser.parse_args(args=argv)
+    capa.main.handle_common_args(args)
+
+    sigpaths = capa.main.get_signatures(args.signatures)
+
+    extractor = capa.main.get_extractor(args.sample, args.format, args.os, args.backend, sigpaths, False)
+
+    Path(args.output).write_bytes(dump(extractor))
+
+    return 0
+
+
+if __name__ == "__main__":
+    import sys
+
+    sys.exit(main())

capa/features/freeze/features.py

@@ -0,0 +1,376 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import binascii
+from typing import Union, Optional
+
+from pydantic import Field, BaseModel, ConfigDict
+
+import capa.features.file
+import capa.features.insn
+import capa.features.common
+import capa.features.basicblock
+
+
+class FeatureModel(BaseModel):
+    model_config = ConfigDict(frozen=True, populate_by_name=True)
+
+    def to_capa(self) -> capa.features.common.Feature:
+        if isinstance(self, OSFeature):
+            return capa.features.common.OS(self.os, description=self.description)
+
+        elif isinstance(self, ArchFeature):
+            return capa.features.common.Arch(self.arch, description=self.description)
+
+        elif isinstance(self, FormatFeature):
+            return capa.features.common.Format(self.format, description=self.description)
+
+        elif isinstance(self, MatchFeature):
+            return capa.features.common.MatchedRule(self.match, description=self.description)
+
+        elif isinstance(
+            self,
+            CharacteristicFeature,
+        ):
+            return capa.features.common.Characteristic(self.characteristic, description=self.description)
+
+        elif isinstance(self, ExportFeature):
+            return capa.features.file.Export(self.export, description=self.description)
+
+        elif isinstance(self, ImportFeature):
+            return capa.features.file.Import(self.import_, description=self.description)
+
+        elif isinstance(self, SectionFeature):
+            return capa.features.file.Section(self.section, description=self.description)
+
+        elif isinstance(self, FunctionNameFeature):
+            return capa.features.file.FunctionName(self.function_name, description=self.description)
+
+        elif isinstance(self, SubstringFeature):
+            return capa.features.common.Substring(self.substring, description=self.description)
+
+        elif isinstance(self, RegexFeature):
+            return capa.features.common.Regex(self.regex, description=self.description)
+
+        elif isinstance(self, StringFeature):
+            return capa.features.common.String(self.string, description=self.description)
+
+        elif isinstance(self, ClassFeature):
+            return capa.features.common.Class(self.class_, description=self.description)
+
+        elif isinstance(self, NamespaceFeature):
+            return capa.features.common.Namespace(self.namespace, description=self.description)
+
+        elif isinstance(self, BasicBlockFeature):
+            return capa.features.basicblock.BasicBlock(description=self.description)
+
+        elif isinstance(self, APIFeature):
+            return capa.features.insn.API(self.api, description=self.description)
+
+        elif isinstance(self, PropertyFeature):
+            return capa.features.insn.Property(self.property, access=self.access, description=self.description)
+
+        elif isinstance(self, NumberFeature):
+            return capa.features.insn.Number(self.number, description=self.description)
+
+        elif isinstance(self, BytesFeature):
+            return capa.features.common.Bytes(binascii.unhexlify(self.bytes), description=self.description)
+
+        elif isinstance(self, OffsetFeature):
+            return capa.features.insn.Offset(self.offset, description=self.description)
+
+        elif isinstance(self, MnemonicFeature):
+            return capa.features.insn.Mnemonic(self.mnemonic, description=self.description)
+
+        elif isinstance(self, OperandNumberFeature):
+            return capa.features.insn.OperandNumber(
+                self.index,
+                self.operand_number,
+                description=self.description,
+            )
+
+        elif isinstance(self, OperandOffsetFeature):
+            return capa.features.insn.OperandOffset(
+                self.index,
+                self.operand_offset,
+                description=self.description,
+            )
+
+        else:
+            raise NotImplementedError(f"Feature.to_capa({type(self)}) not implemented")
+
+
+def feature_from_capa(f: capa.features.common.Feature) -> "Feature":
+    if isinstance(f, capa.features.common.OS):
+        assert isinstance(f.value, str)
+        return OSFeature(os=f.value, description=f.description)
+
+    elif isinstance(f, capa.features.common.Arch):
+        assert isinstance(f.value, str)
+        return ArchFeature(arch=f.value, description=f.description)
+
+    elif isinstance(f, capa.features.common.Format):
+        assert isinstance(f.value, str)
+        return FormatFeature(format=f.value, description=f.description)
+
+    elif isinstance(f, capa.features.common.MatchedRule):
+        assert isinstance(f.value, str)
+        return MatchFeature(match=f.value, description=f.description)
+
+    elif isinstance(f, capa.features.common.Characteristic):
+        assert isinstance(f.value, str)
+        return CharacteristicFeature(characteristic=f.value, description=f.description)
+
+    elif isinstance(f, capa.features.file.Export):
+        assert isinstance(f.value, str)
+        return ExportFeature(export=f.value, description=f.description)
+
+    elif isinstance(f, capa.features.file.Import):
+        assert isinstance(f.value, str)
+        return ImportFeature(import_=f.value, description=f.description)  # type: ignore
+        # Mypy is unable to recognise `import_` as a argument due to alias
+
+    elif isinstance(f, capa.features.file.Section):
+        assert isinstance(f.value, str)
+        return SectionFeature(section=f.value, description=f.description)
+
+    elif isinstance(f, capa.features.file.FunctionName):
+        assert isinstance(f.value, str)
+        return FunctionNameFeature(function_name=f.value, description=f.description)  # type: ignore
+        # Mypy is unable to recognise `function_name` as a argument due to alias
+
+    # must come before check for String due to inheritance
+    elif isinstance(f, capa.features.common.Substring):
+        assert isinstance(f.value, str)
+        return SubstringFeature(substring=f.value, description=f.description)
+
+    # must come before check for String due to inheritance
+    elif isinstance(f, capa.features.common.Regex):
+        assert isinstance(f.value, str)
+        return RegexFeature(regex=f.value, description=f.description)
+
+    elif isinstance(f, capa.features.common.String):
+        assert isinstance(f.value, str)
+        return StringFeature(string=f.value, description=f.description)
+
+    elif isinstance(f, capa.features.common.Class):
+        assert isinstance(f.value, str)
+        return ClassFeature(class_=f.value, description=f.description)  # type: ignore
+        # Mypy is unable to recognise `class_` as a argument due to alias
+
+    elif isinstance(f, capa.features.common.Namespace):
+        assert isinstance(f.value, str)
+        return NamespaceFeature(namespace=f.value, description=f.description)
+
+    elif isinstance(f, capa.features.basicblock.BasicBlock):
+        return BasicBlockFeature(description=f.description)
+
+    elif isinstance(f, capa.features.insn.API):
+        assert isinstance(f.value, str)
+        return APIFeature(api=f.value, description=f.description)
+
+    elif isinstance(f, capa.features.insn.Property):
+        assert isinstance(f.value, str)
+        return PropertyFeature(property=f.value, access=f.access, description=f.description)
+
+    elif isinstance(f, capa.features.insn.Number):
+        assert isinstance(f.value, (int, float))
+        return NumberFeature(number=f.value, description=f.description)
+
+    elif isinstance(f, capa.features.common.Bytes):
+        buf = f.value
+        assert isinstance(buf, bytes)
+        return BytesFeature(bytes=binascii.hexlify(buf).decode("ascii"), description=f.description)
+
+    elif isinstance(f, capa.features.insn.Offset):
+        assert isinstance(f.value, int)
+        return OffsetFeature(offset=f.value, description=f.description)
+
+    elif isinstance(f, capa.features.insn.Mnemonic):
+        assert isinstance(f.value, str)
+        return MnemonicFeature(mnemonic=f.value, description=f.description)
+
+    elif isinstance(f, capa.features.insn.OperandNumber):
+        assert isinstance(f.value, int)
+        return OperandNumberFeature(index=f.index, operand_number=f.value, description=f.description)  # type: ignore
+        # Mypy is unable to recognise `operand_number` as a argument due to alias
+
+    elif isinstance(f, capa.features.insn.OperandOffset):
+        assert isinstance(f.value, int)
+        return OperandOffsetFeature(index=f.index, operand_offset=f.value, description=f.description)  # type: ignore
+        # Mypy is unable to recognise `operand_offset` as a argument due to alias
+
+    else:
+        raise NotImplementedError(f"feature_from_capa({type(f)}) not implemented")
+
+
+class OSFeature(FeatureModel):
+    type: str = "os"
+    os: str
+    description: Optional[str] = None
+
+
+class ArchFeature(FeatureModel):
+    type: str = "arch"
+    arch: str
+    description: Optional[str] = None
+
+
+class FormatFeature(FeatureModel):
+    type: str = "format"
+    format: str
+    description: Optional[str] = None
+
+
+class MatchFeature(FeatureModel):
+    type: str = "match"
+    match: str
+    description: Optional[str] = None
+
+
+class CharacteristicFeature(FeatureModel):
+    type: str = "characteristic"
+    characteristic: str
+    description: Optional[str] = None
+
+
+class ExportFeature(FeatureModel):
+    type: str = "export"
+    export: str
+    description: Optional[str] = None
+
+
+class ImportFeature(FeatureModel):
+    type: str = "import"
+    import_: str = Field(alias="import")
+    description: Optional[str] = None
+
+
+class SectionFeature(FeatureModel):
+    type: str = "section"
+    section: str
+    description: Optional[str] = None
+
+
+class FunctionNameFeature(FeatureModel):
+    type: str = "function name"
+    function_name: str = Field(alias="function name")
+    description: Optional[str] = None
+
+
+class SubstringFeature(FeatureModel):
+    type: str = "substring"
+    substring: str
+    description: Optional[str] = None
+
+
+class RegexFeature(FeatureModel):
+    type: str = "regex"
+    regex: str
+    description: Optional[str] = None
+
+
+class StringFeature(FeatureModel):
+    type: str = "string"
+    string: str
+    description: Optional[str] = None
+
+
+class ClassFeature(FeatureModel):
+    type: str = "class"
+    class_: str = Field(alias="class")
+    description: Optional[str] = None
+
+
+class NamespaceFeature(FeatureModel):
+    type: str = "namespace"
+    namespace: str
+    description: Optional[str] = None
+
+
+class BasicBlockFeature(FeatureModel):
+    type: str = "basic block"
+    description: Optional[str] = None
+
+
+class APIFeature(FeatureModel):
+    type: str = "api"
+    api: str
+    description: Optional[str] = None
+
+
+class PropertyFeature(FeatureModel):
+    type: str = "property"
+    access: Optional[str] = None
+    property: str
+    description: Optional[str] = None
+
+
+class NumberFeature(FeatureModel):
+    type: str = "number"
+    number: Union[int, float]
+    description: Optional[str] = None
+
+
+class BytesFeature(FeatureModel):
+    type: str = "bytes"
+    bytes: str
+    description: Optional[str] = None
+
+
+class OffsetFeature(FeatureModel):
+    type: str = "offset"
+    offset: int
+    description: Optional[str] = None
+
+
+class MnemonicFeature(FeatureModel):
+    type: str = "mnemonic"
+    mnemonic: str
+    description: Optional[str] = None
+
+
+class OperandNumberFeature(FeatureModel):
+    type: str = "operand number"
+    index: int
+    operand_number: int = Field(alias="operand number")
+    description: Optional[str] = None
+
+
+class OperandOffsetFeature(FeatureModel):
+    type: str = "operand offset"
+    index: int
+    operand_offset: int = Field(alias="operand offset")
+    description: Optional[str] = None
+
+
+Feature = Union[
+    OSFeature,
+    ArchFeature,
+    FormatFeature,
+    MatchFeature,
+    CharacteristicFeature,
+    ExportFeature,
+    ImportFeature,
+    SectionFeature,
+    FunctionNameFeature,
+    SubstringFeature,
+    RegexFeature,
+    StringFeature,
+    ClassFeature,
+    NamespaceFeature,
+    APIFeature,
+    PropertyFeature,
+    NumberFeature,
+    BytesFeature,
+    OffsetFeature,
+    MnemonicFeature,
+    OperandNumberFeature,
+    OperandOffsetFeature,
+    # Note! this must be last, see #1161
+    BasicBlockFeature,
+]

capa/features/insn.py

@@ -1,40 +1,169 @@
-# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
 # Licensed under the Apache License, Version 2.0 (the "License");
 #  you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at: [package root]/LICENSE.txt
 # Unless required by applicable law or agreed to in writing, software distributed under the License
 #  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and limitations under the License.
+import abc
+from typing import Union, Optional
 
-from capa.features import Feature
+import capa.helpers
+from capa.features.common import VALID_FEATURE_ACCESS, Feature
+
+
+def hex(n: int) -> str:
+    """render the given number using upper case hex, like: 0x123ABC"""
+    if n < 0:
+        return f"-0x{(-n):X}"
+    else:
+        return f"0x{(n):X}"
 
 
 class API(Feature):
-    def __init__(self, name, description=None):
-        # Downcase library name if given
-        if "." in name:
-            modname, _, impname = name.rpartition(".")
-            name = modname.lower() + "." + impname
+    def __init__(self, name: str, description=None):
+        super().__init__(name, description=description)
 
-        super(API, self).__init__(name, description=description)
+
+class _AccessFeature(Feature, abc.ABC):
+    # superclass: don't use directly
+    def __init__(self, value: str, access: Optional[str] = None, description: Optional[str] = None):
+        super().__init__(value, description=description)
+        if access is not None:
+            if access not in VALID_FEATURE_ACCESS:
+                raise ValueError(f"{self.name} access type {access} not valid")
+        self.access = access
+
+    def __hash__(self):
+        return hash((self.name, self.value, self.access))
+
+    def __eq__(self, other):
+        return super().__eq__(other) and self.access == other.access
+
+    def get_name_str(self) -> str:
+        if self.access is not None:
+            return f"{self.name}/{self.access}"
+        return self.name
+
+
+class Property(_AccessFeature):
+    def __init__(self, value: str, access: Optional[str] = None, description=None):
+        super().__init__(value, access=access, description=description)
 
 
 class Number(Feature):
-    def __init__(self, value, arch=None, description=None):
-        super(Number, self).__init__(value, arch=arch, description=description)
+    def __init__(self, value: Union[int, float], description=None):
+        """
+        args:
+          value (int or float): positive or negative integer, or floating point number.
+
+        the range of the value is:
+          - if positive, the range of u64
+          - if negative, the range of i64
+          - if floating, the range and precision of double
+        """
+        super().__init__(value, description=description)
 
     def get_value_str(self):
-        return "0x%X" % self.value
+        if isinstance(self.value, int):
+            return capa.helpers.hex(self.value)
+        elif isinstance(self.value, float):
+            return str(self.value)
+        else:
+            raise ValueError(f"invalid value type {type(self.value)}")
+
+
+# max recognized structure size (and therefore, offset size)
+MAX_STRUCTURE_SIZE = 0x10000
 
 
 class Offset(Feature):
-    def __init__(self, value, arch=None, description=None):
-        super(Offset, self).__init__(value, arch=arch, description=description)
+    def __init__(self, value: int, description=None):
+        """
+        args:
+          value (int): the offset, which can be positive or negative.
+
+        the range of the value is:
+          - if positive, the range of u64
+          - if negative, the range of i64
+        """
+        super().__init__(value, description=description)
 
     def get_value_str(self):
-        return "0x%X" % self.value
+        assert isinstance(self.value, int)
+        return hex(self.value)
 
 
 class Mnemonic(Feature):
-    def __init__(self, value, description=None):
-        super(Mnemonic, self).__init__(value, description=description)
+    def __init__(self, value: str, description=None):
+        super().__init__(value, description=description)
+
+
+# max number of operands to consider for a given instruction.
+# since we only support Intel and .NET, we can assume this is 3
+# which covers cases up to e.g. "vinserti128 ymm0,ymm0,ymm5,1"
+MAX_OPERAND_COUNT = 4
+MAX_OPERAND_INDEX = MAX_OPERAND_COUNT - 1
+
+
+class _Operand(Feature, abc.ABC):
+    # superclass: don't use directly
+    # subclasses should set self.name and provide the value string formatter
+    def __init__(self, index: int, value: Union[int, float], description=None):
+        super().__init__(value, description=description)
+        self.index = index
+
+    def __hash__(self):
+        return hash((self.name, self.value))
+
+    def __eq__(self, other):
+        return super().__eq__(other) and self.index == other.index
+
+
+class OperandNumber(_Operand):
+    # cached names so we don't do extra string formatting every ctor
+    NAMES = [f"operand[{i}].number" for i in range(MAX_OPERAND_COUNT)]
+
+    # operand[i].number: 0x12
+    def __init__(self, index: int, value: Union[int, float], description=None):
+        """
+        args:
+          value (int or float): positive or negative integer, or floating point number.
+
+        the range of the value is:
+          - if positive, the range of u64
+          - if negative, the range of i64
+          - if floating, the range and precision of double
+        """
+        super().__init__(index, value, description=description)
+        self.name = self.NAMES[index]
+
+    def get_value_str(self) -> str:
+        if isinstance(self.value, int):
+            return capa.helpers.hex(self.value)
+        elif isinstance(self.value, float):
+            return str(self.value)
+        else:
+            raise ValueError("invalid value type")
+
+
+class OperandOffset(_Operand):
+    # cached names so we don't do extra string formatting every ctor
+    NAMES = [f"operand[{i}].offset" for i in range(MAX_OPERAND_COUNT)]
+
+    # operand[i].offset: 0x12
+    def __init__(self, index: int, value: int, description=None):
+        """
+        args:
+          value (int): the offset, which can be positive or negative.
+
+        the range of the value is:
+          - if positive, the range of u64
+          - if negative, the range of i64
+        """
+        super().__init__(index, value, description=description)
+        self.name = self.NAMES[index]
+
+    def get_value_str(self) -> str:
+        assert isinstance(self.value, int)
+        return hex(self.value)

capa/ghidra/README.md

@@ -0,0 +1,172 @@
+<div align="center">
+    <img src="/doc/img/ghidra_backend_logo.png" width=300 height=175>
+</div>
+
+The Ghidra feature extractor is an application of the FLARE team's open-source project, Ghidrathon, to integrate capa with Ghidra using Python 3. capa is a framework that uses a well-defined collection of rules to identify capabilities in a program. You can run capa against a PE file, ELF file, or shellcode and it tells you what it thinks the program can do. For example, it might suggest that the program is a backdoor, can install services, or relies on HTTP to communicate. The Ghidra feature extractor can be used to run capa analysis on your Ghidra databases without needing access to the original binary file.
+
+<img src="/doc/img/ghidra_script_mngr_output.png">
+
+## Getting Started
+
+### Installation
+
+Please ensure that you have the following dependencies installed before continuing:
+
+| Dependency | Version | Source |
+|------------|---------|--------|
+| Ghidrathon | `>= 3.0.0` | https://github.com/mandiant/Ghidrathon |
+| Python | `>= 3.8` | https://www.python.org/downloads |
+| Ghidra | `>= 10.2` | https://ghidra-sre.org |
+
+In order to run capa using using Ghidra, you must install capa as a library, obtain the official capa rules that match the capa version you have installed, and configure the Python 3 script [capa_ghidra.py](/capa/ghidra/capa_ghidra.py). You can do this by completing the following steps using the Python 3 interpreter that you have configured for your Ghidrathon installation:
+
+1. Install capa and its dependencies from PyPI using the following command:
+```bash
+$ pip install flare-capa
+```
+
+2. Download and extract the [official capa rules](https://github.com/mandiant/capa-rules/releases) that match the capa version you have installed. Use the following command to view the version of capa you have installed:
+```bash
+$ pip show flare-capa
+OR
+$ capa --version
+```
+
+3. Copy [capa_ghidra.py](/capa/ghidra/capa_ghidra.py) to your `$USER_HOME/ghidra_scripts` directory or manually add `</path/to/ghidra_capa.py/>` to the Ghidra Script Manager.
+
+## Usage
+
+After completing the installation steps you can execute `capa_ghidra.py` using the Ghidra Script Manager or Headless Analyzer.
+
+### Ghidra Script Manager
+
+To execute `capa_ghidra.py` using the Ghidra Script Manager, first open the Ghidra Script Manager by navigating to `Window > Script Manager` in the Ghidra Code Browser. Next, locate `capa_ghidra.py` by selecting the `Python 3 > capa` category or using the Ghidra Script Manager search funtionality. Finally, double-click `capa_ghidra.py` to execute the script. If you don't see `capa_ghidra.py`, make sure you have copied the script to your `$USER_HOME/ghidra_scripts` directory or manually added `</path/to/ghidra_capa.py/>` to the Ghidra Script Manager
+
+When executed, `capa_ghidra.py` asks you to provide your capa rules directory and preferred output format. `capa_ghidra.py` supports `default`, `verbose`, and `vverbose` output formats when executed from the Ghidra Script Manager. `capa_ghidra.py` writes output to the Ghidra Console Window.
+
+#### Example
+
+The following is an example of running `capa_ghidra.py` using the Ghidra Script Manager:
+
+Selecting capa rules:
+<img src="/doc/img/ghidra_script_mngr_rules.png">
+
+Choosing output format:
+<img src="/doc/img/ghidra_script_mngr_verbosity.png">
+
+Viewing results in Ghidra Console Window:
+<img src="/doc/img/ghidra_script_mngr_output.png">
+
+### Ghidra Headless Analyzer
+
+To execute `capa_ghidra.py` using the Ghidra Headless Analyzer, you can use the Ghidra `analyzeHeadless` script located in your `$GHIDRA_HOME/support` directory. You will need to provide the following arguments to the Ghidra `analyzeHeadless` script:
+
+1. `</path/to/ghidra/project/>`: path to Ghidra project
+2. `<ghidra_project_name>`: name of Ghidra Project
+3. `-process <sample_name>`: name of sample `<sample_name>`
+4. `-ScriptPath </path/to/capa_ghidra/>`: OPTIONAL argument specifying path `</path/to/capa_ghidra/>` to `capa_ghidra.py`
+5. `-PostScript capa_ghidra.py`: executes `capa_ghidra.py` as post-analysis script
+6. `"<capa_args>"`: single, quoted string containing capa arguments that must specify capa rules directory and output format, e.g. `"<path/to/capa/rules> --verbose"`. `capa_ghidra.py` supports `default`, `verbose`, `vverbose` and `json` formats when executed using the Ghidra Headless Analyzer. `capa_ghidra.py` writes output to the console window used to execute the Ghidra `analyzeHeadless` script.
+7. `-processor <languageID>`: required ONLY if sample `<sample_name>` is shellcode. More information on specifying the `<languageID>` can be found in the `$GHIDRA_HOME/support/analyzeHeadlessREADME.html` documentation.
+
+The following is an example of combining these arguments into a single `analyzeHeadless` script command:
+
+```
+$GHIDRA_HOME/support/analyzeHeadless </path/to/ghidra/project/> <ghidra_project_name> -process <sample_name> -PostScript capa_ghidra.py "/path/to/capa/rules/ --verbose"
+```
+
+You may also want to run capa against a sample that you have not yet imported into your Ghidra project. The following is an example of importing a sample and running `capa_ghidra.py` using a single `analyzeHeadless` script command:
+
+```
+$GHIDRA_HOME/support/analyzeHeadless </path/to/ghidra/project/> <ghidra_project_name> -Import </path/to/sample> -PostScript capa_ghidra.py "/path/to/capa/rules/ --verbose"
+```
+
+You can also provide `capa_ghidra.py` the single argument `"help"` to view supported arguments when running the script using the Ghidra Headless Analyzer:
+```
+$GHIDRA_HOME/support/analyzeHeadless </path/to/ghidra/project/> <ghidra_project_name> -process <sample_name> -PostScript capa_ghidra.py "help"
+```
+
+#### Example
+
+The following is an example of running `capa_ghidra.py` against a shellcode sample using the Ghidra `analyzeHeadless` script:
+```
+$ analyzeHeadless /home/wumbo/Desktop/ghidra_projects/ capa_test -process 499c2a85f6e8142c3f48d4251c9c7cd6.raw32 -processor x86:LE:32:default -PostScript capa_ghidra.py "/home/wumbo/capa/rules -vv"
+[...]
+
+INFO  REPORT: Analysis succeeded for file: /499c2a85f6e8142c3f48d4251c9c7cd6.raw32 (HeadlessAnalyzer)  
+INFO  SCRIPT: /home/wumbo/ghidra_scripts/capa_ghidra.py (HeadlessAnalyzer)  
+md5                     499c2a85f6e8142c3f48d4251c9c7cd6                                                                                                                                                                                                    
+sha1
+sha256                  e8e02191c1b38c808d27a899ac164b3675eb5cadd3a8907b0ffa863714000e72
+path                    /home/wumbo/capa/tests/data/499c2a85f6e8142c3f48d4251c9c7cd6.raw32
+timestamp               2023-08-29 17:57:00.946588
+capa version            6.1.0
+os                      unknown os
+format                  Raw Binary
+arch                    x86
+extractor               ghidra
+base address            global
+rules                   /home/wumbo/capa/rules
+function count          42
+library function count  0
+total feature count     1970
+
+contain loop (24 matches, only showing first match of library rule)
+author  moritz.raabe@mandiant.com
+scope   function
+function @ 0x0
+  or:
+    characteristic: loop @ 0x0
+    characteristic: tight loop @ 0x278
+
+contain obfuscated stackstrings
+namespace  anti-analysis/obfuscation/string/stackstring
+author     moritz.raabe@mandiant.com
+scope      basic block
+att&ck     Defense Evasion::Obfuscated Files or Information::Indicator Removal from Tools [T1027.005]
+mbc        Anti-Static Analysis::Executable Code Obfuscation::Argument Obfuscation [B0032.020], Anti-Static Analysis::Executable Code Obfuscation::Stack Strings [B0032.017]
+basic block @ 0x0 in function 0x0
+  characteristic: stack string @ 0x0
+
+encode data using XOR
+namespace  data-manipulation/encoding/xor
+author     moritz.raabe@mandiant.com
+scope      basic block
+att&ck     Defense Evasion::Obfuscated Files or Information [T1027]
+mbc        Defense Evasion::Obfuscated Files or Information::Encoding-Standard Algorithm [E1027.m02], Data::Encode Data::XOR [C0026.002]
+basic block @ 0x8AF in function 0x8A1
+  and:
+    characteristic: tight loop @ 0x8AF
+    characteristic: nzxor @ 0x8C0
+    not: = filter for potential false positives
+      or:
+        or: = unsigned bitwise negation operation (~i)
+          number: 0xFFFFFFFF = bitwise negation for unsigned 32 bits
+          number: 0xFFFFFFFFFFFFFFFF = bitwise negation for unsigned 64 bits
+        or: = signed bitwise negation operation (~i)
+          number: 0xFFFFFFF = bitwise negation for signed 32 bits
+          number: 0xFFFFFFFFFFFFFFF = bitwise negation for signed 64 bits
+        or: = Magic constants used in the implementation of strings functions.
+          number: 0x7EFEFEFF = optimized string constant for 32 bits
+          number: 0x81010101 = -0x81010101 = 0x7EFEFEFF
+          number: 0x81010100 = 0x81010100 = ~0x7EFEFEFF
+          number: 0x7EFEFEFEFEFEFEFF = optimized string constant for 64 bits
+          number: 0x8101010101010101 = -0x8101010101010101 = 0x7EFEFEFEFEFEFEFF
+          number: 0x8101010101010100 = 0x8101010101010100 = ~0x7EFEFEFEFEFEFEFF
+
+get OS information via KUSER_SHARED_DATA
+namespace   host-interaction/os/version
+author      @mr-tz
+scope       function
+att&ck      Discovery::System Information Discovery [T1082]
+references  https://www.geoffchappell.com/studies/windows/km/ntoskrnl/inc/api/ntexapi_x/kuser_shared_data/index.htm
+function @ 0x1CA6
+  or:
+    number: 0x7FFE026C = NtMajorVersion @ 0x1D18
+
+
+
+Script /home/wumbo/ghidra_scripts/capa_ghidra.py called exit with code 0
+
+[...]
+```

capa/ghidra/capa_ghidra.py

@@ -0,0 +1,166 @@
+# Run capa against loaded Ghidra database
+# @author Mike Hunhoff (mehunhoff@google.com)
+# @category Python 3.capa
+
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import sys
+import logging
+import pathlib
+import argparse
+
+import capa
+import capa.main
+import capa.rules
+import capa.ghidra.helpers
+import capa.render.default
+import capa.features.extractors.ghidra.extractor
+
+logger = logging.getLogger("capa_ghidra")
+
+
+def run_headless():
+    parser = argparse.ArgumentParser(description="The FLARE team's open-source tool to integrate capa with Ghidra.")
+
+    parser.add_argument(
+        "rules",
+        type=str,
+        help="path to rule file or directory",
+    )
+    parser.add_argument(
+        "-v", "--verbose", action="store_true", help="enable verbose result document (no effect with --json)"
+    )
+    parser.add_argument(
+        "-vv", "--vverbose", action="store_true", help="enable very verbose result document (no effect with --json)"
+    )
+    parser.add_argument("-d", "--debug", action="store_true", help="enable debugging output on STDERR")
+    parser.add_argument("-q", "--quiet", action="store_true", help="disable all output but errors")
+    parser.add_argument("-j", "--json", action="store_true", help="emit JSON instead of text")
+
+    script_args = list(getScriptArgs())  # type: ignore [name-defined] # noqa: F821
+    if not script_args or len(script_args) > 1:
+        script_args = []
+    else:
+        script_args = script_args[0].split()
+        for idx, arg in enumerate(script_args):
+            if arg.lower() == "help":
+                script_args[idx] = "--help"
+
+    args = parser.parse_args(args=script_args)
+
+    if args.quiet:
+        logging.basicConfig(level=logging.WARNING)
+        logging.getLogger().setLevel(logging.WARNING)
+    elif args.debug:
+        logging.basicConfig(level=logging.DEBUG)
+        logging.getLogger().setLevel(logging.DEBUG)
+    else:
+        logging.basicConfig(level=logging.INFO)
+        logging.getLogger().setLevel(logging.INFO)
+
+    logger.debug("running in Ghidra headless mode")
+
+    rules_path = pathlib.Path(args.rules)
+
+    logger.debug("rule path: %s", rules_path)
+    rules = capa.main.get_rules([rules_path])
+
+    meta = capa.ghidra.helpers.collect_metadata([rules_path])
+    extractor = capa.features.extractors.ghidra.extractor.GhidraFeatureExtractor()
+
+    capabilities, counts = capa.main.find_capabilities(rules, extractor, False)
+
+    meta.analysis.feature_counts = counts["feature_counts"]
+    meta.analysis.library_functions = counts["library_functions"]
+    meta.analysis.layout = capa.main.compute_layout(rules, extractor, capabilities)
+
+    if capa.main.has_file_limitation(rules, capabilities, is_standalone=True):
+        logger.info("capa encountered warnings during analysis")
+
+    if args.json:
+        print(capa.render.json.render(meta, rules, capabilities))  # noqa: T201
+    elif args.vverbose:
+        print(capa.render.vverbose.render(meta, rules, capabilities))  # noqa: T201
+    elif args.verbose:
+        print(capa.render.verbose.render(meta, rules, capabilities))  # noqa: T201
+    else:
+        print(capa.render.default.render(meta, rules, capabilities))  # noqa: T201
+
+    return 0
+
+
+def run_ui():
+    logging.basicConfig(level=logging.INFO)
+    logging.getLogger().setLevel(logging.INFO)
+
+    rules_dir: str = ""
+    try:
+        selected_dir = askDirectory("Choose capa rules directory", "Ok")  # type: ignore [name-defined] # noqa: F821
+        if selected_dir:
+            rules_dir = selected_dir.getPath()
+    except RuntimeError:
+        # RuntimeError thrown when user selects "Cancel"
+        pass
+
+    if not rules_dir:
+        logger.info("You must choose a capa rules directory before running capa.")
+        return capa.main.E_MISSING_RULES
+
+    verbose = askChoice(  # type: ignore [name-defined] # noqa: F821
+        "capa output verbosity", "Choose capa output verbosity", ["default", "verbose", "vverbose"], "default"
+    )
+
+    rules_path: pathlib.Path = pathlib.Path(rules_dir)
+    logger.info("running capa using rules from %s", str(rules_path))
+
+    rules = capa.main.get_rules([rules_path])
+
+    meta = capa.ghidra.helpers.collect_metadata([rules_path])
+    extractor = capa.features.extractors.ghidra.extractor.GhidraFeatureExtractor()
+
+    capabilities, counts = capa.main.find_capabilities(rules, extractor, True)
+
+    meta.analysis.feature_counts = counts["feature_counts"]
+    meta.analysis.library_functions = counts["library_functions"]
+    meta.analysis.layout = capa.main.compute_layout(rules, extractor, capabilities)
+
+    if capa.main.has_file_limitation(rules, capabilities, is_standalone=False):
+        logger.info("capa encountered warnings during analysis")
+
+    if verbose == "vverbose":
+        print(capa.render.vverbose.render(meta, rules, capabilities))  # noqa: T201
+    elif verbose == "verbose":
+        print(capa.render.verbose.render(meta, rules, capabilities))  # noqa: T201
+    else:
+        print(capa.render.default.render(meta, rules, capabilities))  # noqa: T201
+
+    return 0
+
+
+def main():
+    if not capa.ghidra.helpers.is_supported_ghidra_version():
+        return capa.main.E_UNSUPPORTED_GHIDRA_VERSION
+
+    if not capa.ghidra.helpers.is_supported_file_type():
+        return capa.main.E_INVALID_FILE_TYPE
+
+    if not capa.ghidra.helpers.is_supported_arch_type():
+        return capa.main.E_INVALID_FILE_ARCH
+
+    if isRunningHeadless():  # type: ignore [name-defined] # noqa: F821
+        return run_headless()
+    else:
+        return run_ui()
+
+
+if __name__ == "__main__":
+    if sys.version_info < (3, 8):
+        from capa.exceptions import UnsupportedRuntimeError
+
+        raise UnsupportedRuntimeError("This version of capa can only be used with Python 3.8+")
+    sys.exit(main())

capa/ghidra/helpers.py

@@ -0,0 +1,159 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import logging
+import datetime
+import contextlib
+from typing import List
+from pathlib import Path
+
+import capa
+import capa.version
+import capa.features.common
+import capa.features.freeze
+import capa.render.result_document as rdoc
+import capa.features.extractors.ghidra.helpers
+
+logger = logging.getLogger("capa")
+
+# file type as returned by Ghidra
+SUPPORTED_FILE_TYPES = ("Executable and Linking Format (ELF)", "Portable Executable (PE)", "Raw Binary")
+
+
+class GHIDRAIO:
+    """
+    An object that acts as a file-like object,
+    using bytes from the current Ghidra listing.
+    """
+
+    def __init__(self):
+        super().__init__()
+
+        self.offset = 0
+        self.bytes_ = self.get_bytes()
+
+    def seek(self, offset, whence=0):
+        assert whence == 0
+        self.offset = offset
+
+    def read(self, size):
+        logger.debug("reading 0x%x bytes at 0x%x (ea: 0x%x)", size, self.offset, currentProgram().getImageBase().add(self.offset).getOffset())  # type: ignore [name-defined] # noqa: F821
+
+        if size > len(self.bytes_) - self.offset:
+            logger.debug("cannot read 0x%x bytes at 0x%x (ea: BADADDR)", size, self.offset)
+            return b""
+        else:
+            return self.bytes_[self.offset : self.offset + size]
+
+    def close(self):
+        return
+
+    def get_bytes(self):
+        file_bytes = currentProgram().getMemory().getAllFileBytes()[0]  # type: ignore [name-defined] # noqa: F821
+
+        # getOriginalByte() allows for raw file parsing on the Ghidra side
+        # other functions will fail as Ghidra will think that it's reading uninitialized memory
+        bytes_ = [file_bytes.getOriginalByte(i) for i in range(file_bytes.getSize())]
+
+        return capa.features.extractors.ghidra.helpers.ints_to_bytes(bytes_)
+
+
+def is_supported_ghidra_version():
+    version = float(getGhidraVersion()[:4])  # type: ignore [name-defined] # noqa: F821
+    if version < 10.2:
+        warning_msg = "capa does not support this Ghidra version"
+        logger.warning(warning_msg)
+        logger.warning("Your Ghidra version is: %s. Supported versions are: Ghidra >= 10.2", version)
+        return False
+    return True
+
+
+def is_running_headless():
+    return isRunningHeadless()  # type: ignore [name-defined] # noqa: F821
+
+
+def is_supported_file_type():
+    file_info = currentProgram().getExecutableFormat()  # type: ignore [name-defined] # noqa: F821
+    if file_info not in SUPPORTED_FILE_TYPES:
+        logger.error("-" * 80)
+        logger.error(" Input file does not appear to be a supported file type.")
+        logger.error(" ")
+        logger.error(
+            " capa currently only supports analyzing PE, ELF, or binary files containing x86 (32- and 64-bit) shellcode."
+        )
+        logger.error(" If you don't know the input file type, you can try using the `file` utility to guess it.")
+        logger.error("-" * 80)
+        return False
+    return True
+
+
+def is_supported_arch_type():
+    lang_id = str(currentProgram().getLanguageID()).lower()  # type: ignore [name-defined] # noqa: F821
+
+    if not all((lang_id.startswith("x86"), any(arch in lang_id for arch in ("32", "64")))):
+        logger.error("-" * 80)
+        logger.error(" Input file does not appear to target a supported architecture.")
+        logger.error(" ")
+        logger.error(" capa currently only supports analyzing x86 (32- and 64-bit).")
+        logger.error("-" * 80)
+        return False
+    return True
+
+
+def get_file_md5():
+    return currentProgram().getExecutableMD5()  # type: ignore [name-defined] # noqa: F821
+
+
+def get_file_sha256():
+    return currentProgram().getExecutableSHA256()  # type: ignore [name-defined] # noqa: F821
+
+
+def collect_metadata(rules: List[Path]):
+    md5 = get_file_md5()
+    sha256 = get_file_sha256()
+
+    info = currentProgram().getLanguageID().toString()  # type: ignore [name-defined] # noqa: F821
+    if "x86" in info and "64" in info:
+        arch = "x86_64"
+    elif "x86" in info and "32" in info:
+        arch = "x86"
+    else:
+        arch = "unknown arch"
+
+    format_name: str = currentProgram().getExecutableFormat()  # type: ignore [name-defined] # noqa: F821
+    if "PE" in format_name:
+        os = "windows"
+    elif "ELF" in format_name:
+        with contextlib.closing(capa.ghidra.helpers.GHIDRAIO()) as f:
+            os = capa.features.extractors.elf.detect_elf_os(f)
+    else:
+        os = "unknown os"
+
+    return rdoc.Metadata(
+        timestamp=datetime.datetime.now(),
+        version=capa.version.__version__,
+        argv=(),
+        sample=rdoc.Sample(
+            md5=md5,
+            sha1="",
+            sha256=sha256,
+            path=currentProgram().getExecutablePath(),  # type: ignore [name-defined] # noqa: F821
+        ),
+        analysis=rdoc.Analysis(
+            format=currentProgram().getExecutableFormat(),  # type: ignore [name-defined] # noqa: F821
+            arch=arch,
+            os=os,
+            extractor="ghidra",
+            rules=tuple(r.resolve().absolute().as_posix() for r in rules),
+            base_address=capa.features.freeze.Address.from_capa(currentProgram().getImageBase().getOffset()),  # type: ignore [name-defined] # noqa: F821
+            layout=rdoc.Layout(
+                functions=(),
+            ),
+            feature_counts=rdoc.FeatureCounts(file=0, functions=()),
+            library_functions=(),
+        ),
+    )

capa/helpers.py

@@ -1,36 +1,167 @@
-# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
 # Licensed under the Apache License, Version 2.0 (the "License");
 #  you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at: [package root]/LICENSE.txt
 # Unless required by applicable law or agreed to in writing, software distributed under the License
 #  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and limitations under the License.
+import inspect
+import logging
+import contextlib
+import importlib.util
+from typing import NoReturn
+from pathlib import Path
 
-import os
+import tqdm
 
-_hex = hex
+from capa.exceptions import UnsupportedFormatError
+from capa.features.common import FORMAT_PE, FORMAT_SC32, FORMAT_SC64, FORMAT_DOTNET, FORMAT_UNKNOWN, Format
+
+EXTENSIONS_SHELLCODE_32 = ("sc32", "raw32")
+EXTENSIONS_SHELLCODE_64 = ("sc64", "raw64")
+EXTENSIONS_ELF = "elf_"
+
+logger = logging.getLogger("capa")
 
 
-def hex(i):
-    # under py2.7, long integers get formatted with a trailing `L`
-    # and this is not pretty. so strip it out.
-    return _hex(oint(i)).rstrip("L")
+def hex(n: int) -> str:
+    """render the given number using upper case hex, like: 0x123ABC"""
+    if n < 0:
+        return f"-0x{(-n):X}"
+    else:
+        return f"0x{(n):X}"
 
 
-def oint(i):
-    # there seems to be some trouble with using `int(viv_utils.Function)`
-    # with the black magic we do with binding the `__int__()` routine.
-    # i haven't had a chance to debug this yet (and i have no hotel wifi).
-    # so in the meantime, detect this, and call the method directly.
-    try:
-        return int(i)
-    except TypeError:
-        return i.__int__()
-
-
-def get_file_taste(sample_path):
-    if not os.path.exists(sample_path):
-        raise IOError("sample path %s does not exist or cannot be accessed" % sample_path)
-    with open(sample_path, "rb") as f:
-        taste = f.read(8)
+def get_file_taste(sample_path: Path) -> bytes:
+    if not sample_path.exists():
+        raise IOError(f"sample path {sample_path} does not exist or cannot be accessed")
+    taste = sample_path.open("rb").read(8)
     return taste
+
+
+def is_runtime_ida():
+    return importlib.util.find_spec("idc") is not None
+
+
+def is_runtime_ghidra():
+    try:
+        currentProgram  # type: ignore [name-defined] # noqa: F821
+    except NameError:
+        return False
+    return True
+
+
+def assert_never(value) -> NoReturn:
+    # careful: python -O will remove this assertion.
+    # but this is only used for type checking, so it's ok.
+    assert False, f"Unhandled value: {value} ({type(value).__name__})"  # noqa: B011
+
+
+def get_format_from_extension(sample: Path) -> str:
+    if sample.name.endswith(EXTENSIONS_SHELLCODE_32):
+        return FORMAT_SC32
+    elif sample.name.endswith(EXTENSIONS_SHELLCODE_64):
+        return FORMAT_SC64
+    return FORMAT_UNKNOWN
+
+
+def get_auto_format(path: Path) -> str:
+    format_ = get_format(path)
+    if format_ == FORMAT_UNKNOWN:
+        format_ = get_format_from_extension(path)
+    if format_ == FORMAT_UNKNOWN:
+        raise UnsupportedFormatError()
+    return format_
+
+
+def get_format(sample: Path) -> str:
+    # imported locally to avoid import cycle
+    from capa.features.extractors.common import extract_format
+    from capa.features.extractors.dnfile_ import DnfileFeatureExtractor
+
+    buf = sample.read_bytes()
+
+    for feature, _ in extract_format(buf):
+        if feature == Format(FORMAT_PE):
+            dnfile_extractor = DnfileFeatureExtractor(sample)
+            if dnfile_extractor.is_dotnet_file():
+                feature = Format(FORMAT_DOTNET)
+
+        assert isinstance(feature.value, str)
+        return feature.value
+
+    return FORMAT_UNKNOWN
+
+
+@contextlib.contextmanager
+def redirecting_print_to_tqdm(disable_progress):
+    """
+    tqdm (progress bar) expects to have fairly tight control over console output.
+    so calls to `print()` will break the progress bar and make things look bad.
+    so, this context manager temporarily replaces the `print` implementation
+    with one that is compatible with tqdm.
+    via: https://stackoverflow.com/a/42424890/87207
+    """
+    old_print = print  # noqa: T202 [reserved word print used]
+
+    def new_print(*args, **kwargs):
+        # If tqdm.tqdm.write raises error, use builtin print
+        if disable_progress:
+            old_print(*args, **kwargs)
+        else:
+            try:
+                tqdm.tqdm.write(*args, **kwargs)
+            except Exception:
+                old_print(*args, **kwargs)
+
+    try:
+        # Globally replace print with new_print.
+        # Verified this works manually on Python 3.11:
+        #     >>> import inspect
+        #     >>> inspect.builtins
+        #     <module 'builtins' (built-in)>
+        inspect.builtins.print = new_print  # type: ignore
+        yield
+    finally:
+        inspect.builtins.print = old_print  # type: ignore
+
+
+def log_unsupported_format_error():
+    logger.error("-" * 80)
+    logger.error(" Input file does not appear to be a PE or ELF file.")
+    logger.error(" ")
+    logger.error(
+        " capa currently only supports analyzing PE and ELF files (or shellcode, when using --format sc32|sc64)."
+    )
+    logger.error(" If you don't know the input file type, you can try using the `file` utility to guess it.")
+    logger.error("-" * 80)
+
+
+def log_unsupported_os_error():
+    logger.error("-" * 80)
+    logger.error(" Input file does not appear to target a supported OS.")
+    logger.error(" ")
+    logger.error(
+        " capa currently only supports analyzing executables for some operating systems (including Windows and Linux)."
+    )
+    logger.error("-" * 80)
+
+
+def log_unsupported_arch_error():
+    logger.error("-" * 80)
+    logger.error(" Input file does not appear to target a supported architecture.")
+    logger.error(" ")
+    logger.error(" capa currently only supports analyzing x86 (32- and 64-bit).")
+    logger.error("-" * 80)
+
+
+def log_unsupported_runtime_error():
+    logger.error("-" * 80)
+    logger.error(" Unsupported runtime or Python interpreter.")
+    logger.error(" ")
+    logger.error(" capa supports running under Python 3.8 and higher.")
+    logger.error(" ")
+    logger.error(
+        " If you're seeing this message on the command line, please ensure you're running a supported Python version."
+    )
+    logger.error("-" * 80)

capa/ida/helpers.py

@@ -0,0 +1,252 @@
+# Copyright (C) 2023 Mandiant, Inc. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at: [package root]/LICENSE.txt
+# Unless required by applicable law or agreed to in writing, software distributed under the License
+#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and limitations under the License.
+import logging
+import datetime
+import contextlib
+from typing import List, Optional
+from pathlib import Path
+
+import idc
+import idaapi
+import idautils
+import ida_bytes
+import ida_loader
+from netnode import netnode
+
+import capa
+import capa.version
+import capa.render.utils as rutils
+import capa.features.common
+import capa.features.freeze
+import capa.render.result_document as rdoc
+from capa.features.address import AbsoluteVirtualAddress
+
+logger = logging.getLogger("capa")
+
+# file type as returned by idainfo.file_type
+SUPPORTED_FILE_TYPES = (
+    idaapi.f_PE,
+    idaapi.f_ELF,
+    idaapi.f_BIN,
+    idaapi.f_COFF,
+    # idaapi.f_MACHO,
+)
+
+# arch type as returned by idainfo.procname
+SUPPORTED_ARCH_TYPES = ("metapc",)
+
+CAPA_NETNODE = f"$ com.mandiant.capa.v{capa.version.__version__}"
+NETNODE_RESULTS = "results"
+NETNODE_RULES_CACHE_ID = "rules-cache-id"
+
+
+def inform_user_ida_ui(message):
+    # this isn't a logger, this is IDA's logging facility
+    idaapi.info(f"{message}. Please refer to IDA Output window for more information.")  # noqa: G004
+
+
+def is_supported_ida_version():
+    version = float(idaapi.get_kernel_version())
+    if version < 7.4 or version >= 9:
+        warning_msg = "This plugin does not support your IDA Pro version"
+        logger.warning(warning_msg)
+        logger.warning("Your IDA Pro version is: %s. Supported versions are: IDA >= 7.4 and IDA < 9.0.", version)
+        return False
+    return True
+
+
+def is_supported_file_type():
+    file_info = idaapi.get_inf_structure()
+    if file_info.filetype not in SUPPORTED_FILE_TYPES:
+        logger.error("-" * 80)
+        logger.error(" Input file does not appear to be a supported file type.")
+        logger.error(" ")
+        logger.error(
+            " capa currently only supports analyzing PE, ELF, or binary files containing x86 (32- and 64-bit) shellcode."
+        )
+        logger.error(" If you don't know the input file type, you can try using the `file` utility to guess it.")
+        logger.error("-" * 80)
+        return False
+    return True
+
+
+def is_supported_arch_type():
+    file_info = idaapi.get_inf_structure()
+    if file_info.procname not in SUPPORTED_ARCH_TYPES or not any((file_info.is_32bit(), file_info.is_64bit())):
+        logger.error("-" * 80)
+        logger.error(" Input file does not appear to target a supported architecture.")
+        logger.error(" ")
+        logger.error(" capa currently only supports analyzing x86 (32- and 64-bit).")
+        logger.error("-" * 80)
+        return False
+    return True
+
+
+def get_disasm_line(va):
+    """ """
+    return idc.generate_disasm_line(va, idc.GENDSM_FORCE_CODE)
+
+
+def is_func_start(ea):
+    """check if function stat exists at virtual address"""
+    f = idaapi.get_func(ea)
+    return f and f.start_ea == ea
+
+
+def get_func_start_ea(ea):
+    """ """
+    f = idaapi.get_func(ea)
+    return f if f is None else f.start_ea
+
+
+def get_file_md5():
+    """ """
+    md5 = idautils.GetInputFileMD5()
+    if not isinstance(md5, str):
+        md5 = capa.features.common.bytes_to_str(md5)
+    return md5
+
+
+def get_file_sha256():
+    """ """
+    sha256 = idaapi.retrieve_input_file_sha256()
+    if not isinstance(sha256, str):
+        sha256 = capa.features.common.bytes_to_str(sha256)
+    return sha256
+
+
+def collect_metadata(rules: List[Path]):
+    """ """
+    md5 = get_file_md5()
+    sha256 = get_file_sha256()
+
+    info: idaapi.idainfo = idaapi.get_inf_structure()
+    if info.procname == "metapc" and info.is_64bit():
+        arch = "x86_64"
+    elif info.procname == "metapc" and info.is_32bit():
+        arch = "x86"
+    else:
+        arch = "unknown arch"
+
+    format_name: str = ida_loader.get_file_type_name()
+    if "PE" in format_name:
+        os = "windows"
+    elif "ELF" in format_name:
+        with contextlib.closing(capa.ida.helpers.IDAIO()) as f:
+            os = capa.features.extractors.elf.detect_elf_os(f)
+    else:
+        os = "unknown os"
+
+    return rdoc.Metadata(
+        timestamp=datetime.datetime.now(),
+        version=capa.version.__version__,
+        argv=(),
+        sample=rdoc.Sample(
+            md5=md5,
+            sha1="",  # not easily accessible
+            sha256=sha256,
+            path=idaapi.get_input_file_path(),
+        ),
+        analysis=rdoc.Analysis(
+            format=idaapi.get_file_type_name(),
+            arch=arch,
+            os=os,
+            extractor="ida",
+            rules=tuple(r.resolve().absolute().as_posix() for r in rules),
+            base_address=capa.features.freeze.Address.from_capa(idaapi.get_imagebase()),
+            layout=rdoc.Layout(
+                functions=(),
+                # this is updated after capabilities have been collected.
+                # will look like:
+                #
+                # "functions": { 0x401000: { "matched_basic_blocks": [ 0x401000, 0x401005, ... ] }, ... }
+            ),
+            # ignore these for now - not used by IDA plugin.
+            feature_counts=rdoc.FeatureCounts(file=0, functions=()),
+            library_functions=(),
+        ),
+    )
+
+
+class IDAIO:
+    """
+    An object that acts as a file-like object,
+    using bytes from the current IDB workspace.
+    """
+
+    def __init__(self):
+        super().__init__()
+        self.offset = 0
+
+    def seek(self, offset, whence=0):
+        assert whence == 0
+        self.offset = offset
+
+    def read(self, size):
+        ea = ida_loader.get_fileregion_ea(self.offset)
+        if ea == idc.BADADDR:
+            logger.debug("cannot read 0x%x bytes at 0x%x (ea: BADADDR)", size, self.offset)
+            return b""
+
+        logger.debug("reading 0x%x bytes at 0x%x (ea: 0x%x)", size, self.offset, ea)
+
+        # get_bytes returns None on error, for consistency with read always return bytes
+        return ida_bytes.get_bytes(ea, size) or b""
+
+    def close(self):
+        return
+
+
+def save_cached_results(resdoc):
+    logger.debug("saving cached capa results to netnode '%s'", CAPA_NETNODE)
+    n = netnode.Netnode(CAPA_NETNODE)
+    n[NETNODE_RESULTS] = resdoc.json()
+
+
+def idb_contains_cached_results() -> bool:
+    try:
+        n = netnode.Netnode(CAPA_NETNODE)
+        return bool(n.get(NETNODE_RESULTS))
+    except netnode.NetnodeCorruptError as e:
+        logger.exception(str(e))
+        return False
+
+
+def load_and_verify_cached_results() -> Optional[rdoc.ResultDocument]:
+    """verifies that cached results have valid (mapped) addresses for the current database"""
+    logger.debug("loading cached capa results from netnode '%s'", CAPA_NETNODE)
+
+    n = netnode.Netnode(CAPA_NETNODE)
+    doc = rdoc.ResultDocument.model_validate_json(n[NETNODE_RESULTS])
+
+    for rule in rutils.capability_rules(doc):
+        for location_, _ in rule.matches:
+            location = location_.to_capa()
+            if isinstance(location, AbsoluteVirtualAddress):
+                ea = int(location)
+                if not idaapi.is_mapped(ea):
+                    logger.error("cached address %s is not a valid location in this database", hex(ea))
+                    return None
+    return doc
+
+
+def save_rules_cache_id(ruleset_id):
+    logger.debug("saving ruleset ID to netnode '%s'", CAPA_NETNODE)
+    n = netnode.Netnode(CAPA_NETNODE)
+    n[NETNODE_RULES_CACHE_ID] = ruleset_id
+
+
+def load_rules_cache_id():
+    n = netnode.Netnode(CAPA_NETNODE)
+    return n[NETNODE_RULES_CACHE_ID]
+
+
+def delete_cached_results():
+    logger.debug("deleting cached capa data")
+    n = netnode.Netnode(CAPA_NETNODE)
+    del n[NETNODE_RESULTS]

capa/ida/helpers/__init__.py

@@ -1,121 +0,0 @@
-# Copyright (C) 2020 FireEye, Inc. All Rights Reserved.
-# Licensed under the Apache License, Version 2.0 (the "License");
-#  you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at: [package root]/LICENSE.txt
-# Unless required by applicable law or agreed to in writing, software distributed under the License
-#  is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and limitations under the License.
-
-import logging
-import datetime
-
-import idc
-import six
-import idaapi
-import idautils
-
-import capa
-
-logger = logging.getLogger("capa")
-
-SUPPORTED_IDA_VERSIONS = [
-    "7.1",
-    "7.2",
-    "7.3",
-    "7.4",
-    "7.5",
-]
-
-# file type names as returned by idaapi.get_file_type_name()
-SUPPORTED_FILE_TYPES = [
-    "Portable executable for 80386 (PE)",
-    "Portable executable for AMD64 (PE)",
-    "Binary file",  # x86/AMD64 shellcode support
-]
-
-
-def inform_user_ida_ui(message):
-    idaapi.info("%s. Please refer to IDA Output window for more information." % message)
-
-
-def is_supported_ida_version():
-    version = idaapi.get_kernel_version()
-    if version not in SUPPORTED_IDA_VERSIONS:
-        warning_msg = "This plugin does not support your IDA Pro version"
-        logger.warning(warning_msg)
-        logger.warning(
-            "Your IDA Pro version is: %s. Supported versions are: %s." % (version, ", ".join(SUPPORTED_IDA_VERSIONS))
-        )
-        return False
-    return True
-
-
-def is_supported_file_type():
-    file_type = idaapi.get_file_type_name()
-    if file_type not in SUPPORTED_FILE_TYPES:
-        logger.error("-" * 80)
-        logger.error(" Input file does not appear to be a PE file.")
-        logger.error(" ")
-        logger.error(
-            " capa currently only supports analyzing PE files (or binary files containing x86/AMD64 shellcode) with IDA."
-        )
-        logger.error(" If you don't know the input file type, you can try using the `file` utility to guess it.")
-        logger.error("-" * 80)
-        return False
-    return True
-
-
-def get_disasm_line(va):
-    """ """
-    return idc.generate_disasm_line(va, idc.GENDSM_FORCE_CODE)
-
-
-def is_func_start(ea):
-    """ check if function stat exists at virtual address """
-    f = idaapi.get_func(ea)
-    return f and f.start_ea == ea
-
-
-def get_func_start_ea(ea):
-    """ """
-    f = idaapi.get_func(ea)
-    return f if f is None else f.start_ea
-
-
-def get_file_md5():
-    """ """
-    md5 = idautils.GetInputFileMD5()
-    if not isinstance(md5, six.string_types):
-        md5 = capa.features.bytes_to_str(md5)
-    return md5
-
-
-def get_file_sha256():
-    """ """
-    sha256 = idaapi.retrieve_input_file_sha256()
-    if not isinstance(sha256, six.string_types):
-        sha256 = capa.features.bytes_to_str(sha256)
-    return sha256
-
-
-def collect_metadata():
-    """ """
-    md5 = get_file_md5()
-    sha256 = get_file_sha256()
-
-    return {
-        "timestamp": datetime.datetime.now().isoformat(),
-        # "argv" is not relevant here
-        "sample": {
-            "md5": md5,
-            "sha1": "",  # not easily accessible
-            "sha256": sha256,
-            "path": idaapi.get_input_file_path(),
-        },
-        "analysis": {
-            "format": idaapi.get_file_type_name(),
-            "extractor": "ida",
-            "base_address": idaapi.get_imagebase(),
-        },
-        "version": capa.version.__version__,
-    }