move sigs to capa directory

build(deps-dev): bump pytest from 7.4.3 to 7.4.4

.devcontainer/devcontainer.json

@@ -41,7 +41,7 @@
 	// "forwardPorts": [],
 
 	// Use 'postCreateCommand' to run commands after the container is created.
-	"postCreateCommand": "git submodule update --init && pip3 install --user -e .[dev]",
+	"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",

.github/CONTRIBUTING.md

@@ -31,7 +31,7 @@ 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:
+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)
@@ -159,12 +159,25 @@ The process described here has several goals:
 
 Please follow these steps to have your contribution considered by the maintainers:
 
+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

.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

@@ -21,9 +21,6 @@ ignore_missing_imports = True
 [mypy-flirt.*]
 ignore_missing_imports = True
 
-[mypy-smda.*]
-ignore_missing_imports = True
-
 [mypy-lief.*]
 ignore_missing_imports = True
 
@@ -45,9 +42,15 @@ 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
 
@@ -60,6 +63,9 @@ ignore_missing_imports = True
 [mypy-ida_loader.*]
 ignore_missing_imports = True
 
+[mypy-ida_segment.*]
+ignore_missing_imports = True
+
 [mypy-PyQt5.*]
 ignore_missing_imports = True
 
@@ -76,4 +82,10 @@ ignore_missing_imports = True
 ignore_missing_imports = True
 
 [mypy-dncil.*]
-ignore_missing_imports = True
+ignore_missing_imports = True
+
+[mypy-netnode.*]
+ignore_missing_imports = True
+
+[mypy-ghidra.*]
+ignore_missing_imports = True

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

@@ -1,5 +0,0 @@
-# Copyright (C) 2020 Mandiant, 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

@@ -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.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

@@ -6,31 +6,6 @@ import subprocess
 import wcwidth
 
 
-# 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/", "")
-)
-# when invoking pyinstaller from the project root, this gets run from the project root.
-with open("./capa/version.py", "r", encoding="utf-8") as f:
-    lines = f.read()
-# version.py contains the version string and other helper functions
-# here we manually replace the version value substring with the result of the above git output
-VERSION_DEF = "__version__ = "
-s = lines.index(VERSION_DEF)
-e = s + len(VERSION_DEF)
-off_rest_file = e + lines[e:].index("\n")
-lines = lines[s:e] + f'"{version}"' + lines[off_rest_file:]
-with open("./capa/version.py", "w", encoding="utf-8") as f:
-    f.write(lines)
-
 a = Analysis(
     # when invoking pyinstaller from the project root,
     # this gets invoked from the directory of the spec file,
@@ -43,7 +18,8 @@ a = Analysis(
         # this gets invoked from the directory of the spec file,
         # i.e. ./.github/pyinstaller
         ("../../rules", "rules"),
-        ("../../sigs", "sigs"),
+        ("../../capa/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.
@@ -85,6 +61,7 @@ a = Analysis(
         "qt5",
         "pyqtwebengine",
         "pyasn1",
+        "binaryninja",
     ],
 )
 
@@ -102,7 +79,7 @@ exe = EXE(
     name="capa",
     icon="logo.ico",
     debug=False,
-    strip=None,
+    strip=False,
     upx=True,
     console=True,
 )

.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

@@ -6,42 +6,54 @@ on:
   release:
     types: [edited, published]
 
+permissions:
+  contents: write
+
 jobs:
   build:
-    name: PyInstaller for ${{ matrix.os }}
+    name: PyInstaller for ${{ matrix.os }} / Py ${{ matrix.python_version }}
     runs-on: ${{ matrix.os }}
     strategy:
       # set to false for debugging
       fail-fast: true
       matrix:
+        # using Python 3.8 to support running across multiple operating systems including Windows 7
         include:
-          - os: ubuntu-18.04
+          - os: ubuntu-20.04
             # use old linux so that the shared library versioning is more portable
             artifact_name: capa
             asset_name: linux
-          - os: windows-2022
+            python_version: 3.8
+          - os: ubuntu-20.04
+            artifact_name: capa
+            asset_name: linux-py311
+            python_version: 3.11
+          - os: windows-2019
             artifact_name: capa.exe
             asset_name: windows
-          - os: macos-10.15
+            python_version: 3.8
+          - os: macos-11
             # use older macOS for assumed better portability
             artifact_name: capa
             asset_name: macos
+            python_version: 3.8
     steps:
       - name: Checkout capa
-        uses: actions/checkout@v2
+        uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # v3.3.0
         with:
           submodules: true
-      # using Python 3.8 to support running across multiple operating systems including Windows 7
-      - name: Set up Python 3.8
-        uses: actions/setup-python@v2
+      - name: Set up Python ${{ matrix.python_version }}
+        uses: actions/setup-python@d27e3f3d7c64b4bbf8e4abfb9b63b83e846e0435 # v4.5.0
         with:
-          python-version: 3.8
-      - if: matrix.os == 'ubuntu-18.04'
+          python-version: ${{ matrix.python_version }}
+      - if: matrix.os == 'ubuntu-20.04'
         run: sudo apt-get install -y libyaml-dev
       - 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 --log-level DEBUG .github/pyinstaller/pyinstaller.spec
       - name: Does it run (PE)?
@@ -50,35 +62,39 @@ jobs:
         run: dist/capa "tests/data/499c2a85f6e8142c3f48d4251c9c7cd6.raw32"
       - name: Does it run (ELF)?
         run: dist/capa "tests/data/7351f8a40c5450557b24622417fc478d.elf_"
-      - uses: actions/upload-artifact@v2
+      - name: Does it run (CAPE)?
+        run: |
+          7z e "tests/data/dynamic/cape/v2.2/d46900384c78863420fb3e297d0a2f743cd2b6b3f7f82bf64059a168e07aceb7.json.gz"
+          dist/capa "d46900384c78863420fb3e297d0a2f743cd2b6b3f7f82bf64059a168e07aceb7.json"
+      - uses: actions/upload-artifact@0b7f8abb1508181956e8e162db84b466c27e18ce # v3.1.2
         with:
           name: ${{ matrix.asset_name }}
           path: dist/${{ matrix.artifact_name }}
 
   test_run:
-    name: Test run on ${{ matrix.os }}
+    name: Test run on ${{ matrix.os }} / ${{ matrix.asset_name }}
     runs-on: ${{ matrix.os }}
     needs: [build]
     strategy:
       matrix:
         include:
           # OSs not already tested above
-          - os: ubuntu-18.04
+          - os: ubuntu-22.04
             artifact_name: capa
             asset_name: linux
-          - os: ubuntu-20.04
+          - os: ubuntu-22.04
             artifact_name: capa
-            asset_name: linux
+            asset_name: linux-py311
           - os: windows-2022
             artifact_name: capa.exe
             asset_name: windows
     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
-        if: matrix.os != 'windows-2022'      
+        if: matrix.os != 'windows-2022'
         run: chmod +x ${{ matrix.artifact_name }}
       - name: Run capa
         run: ./${{ matrix.artifact_name }} -h
@@ -94,13 +110,15 @@ jobs:
         include:
           - asset_name: linux
             artifact_name: capa
+          - asset_name: linux-py311
+            artifact_name: capa
           - asset_name: windows
             artifact_name: capa.exe
           - asset_name: macos
             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
@@ -110,7 +128,7 @@ 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

@@ -7,6 +7,8 @@ on:
   pull_request_target:
     types: [opened, edited, synchronize]
 
+permissions: read-all
+
 jobs:
   check_changelog:
     # no need to check for dependency updates via dependabot
@@ -17,7 +19,7 @@ jobs:
     steps:
     - name: Get changed files
       id: files
-      uses: Ana06/get-changed-files@v1.2
+      uses: Ana06/get-changed-files@e0c398b7065a8d84700c471b6afc4116d1ba4e96 # v2.2.0
     - name: check changelog updated
       id: changelog_updated
       env:
@@ -27,14 +29,14 @@ jobs:
         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@v0.1.0
+      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@v0.1.0
+      uses: Ana06/automatic-pull-request-review@0cf4e8a17ba79344ed3fdd7fed6dd0311d08a9d4 # v0.1.0
       with:
         repo-token: ${{ secrets.GITHUB_TOKEN }}
         event: DISMISS

.github/workflows/pip-audit.yml

@@ -0,0 +1,21 @@
+name: PIP audit
+
+on:
+  schedule:
+  - cron: '0 8 * * 1'
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    timeout-minutes: 20
+    strategy:
+      matrix:
+        python-version: ["3.11"]
+
+    steps:
+      - name: Check out repository code
+        uses: actions/checkout@v4
+
+      - uses: pypa/gh-action-pip-audit@v1.0.8
+        with:
+          inputs: .

.github/workflows/publish.yml

@@ -1,30 +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: '3.7'
+          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

@@ -4,13 +4,15 @@ 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: mandiant/capa-rules
         token: ${{ secrets.CAPA_TOKEN }}
@@ -23,7 +25,7 @@ jobs:
         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: mandiant/capa-rules
         github_token: ${{ secrets.CAPA_TOKEN }}

.github/workflows/tests.yml

@@ -6,6 +6,8 @@ on:
   pull_request:
     branches: [ master ]
 
+permissions: read-all
+
 # save workspaces to speed up testing
 env:
   CAPA_SAVE_WORKSPACE: "True"
@@ -15,7 +17,7 @@ jobs:
     runs-on: ubuntu-20.04
     steps:
     - name: Checkout capa
-      uses: actions/checkout@v2
+      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: |
@@ -26,35 +28,38 @@ jobs:
     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 -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 --show-diff-on-failure
     - name: Lint with black
-      run: black -l 120 --check .
-    - name: Lint with pycodestyle
-      run: pycodestyle --show-source capa/ scripts/ tests/
+      run: pre-commit run black --show-diff-on-failure
+    - name: Lint with flake8
+      run: pre-commit run flake8 --hook-stage manual
     - name: Check types with mypy
-      run: mypy --config-file .github/mypy/mypy.ini capa/ scripts/ tests/
+      run:  pre-commit run mypy --hook-stage manual
 
   rule_linter:
     runs-on: ubuntu-20.04
     steps:
     - name: Checkout capa with submodules
-      uses: actions/checkout@v2
+      uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # v3.3.0
       with:
         submodules: recursive
-    - name: Set up Python 3.8
-      uses: actions/setup-python@v2
+    - 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 capa
-      run: pip install -e .
+      run: pip install -e .[dev]
     - name: Run rule linter
       run: python scripts/lint.py rules/
 
@@ -67,20 +72,22 @@ jobs:
       matrix:
         os: [ubuntu-20.04, windows-2019, macos-11]
         # across all operating systems
-        python-version: ["3.7", "3.10"]
+        python-version: ["3.8", "3.11"]
         include:
           # on Ubuntu run these as well
           - os: ubuntu-20.04
             python-version: "3.8"
           - os: ubuntu-20.04
             python-version: "3.9"
+          - os: ubuntu-20.04
+            python-version: "3.10"
     steps:
     - name: Checkout capa with submodules
-      uses: actions/checkout@v2
+      uses: actions/checkout@ac593985615ec2ede58e132d2e21d2b1cbd6127c # v3.3.0
       with:
         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
@@ -88,5 +95,110 @@ jobs:
       run: sudo apt-get install -y libyaml-dev
     - name: Install capa
       run: pip install -e .[dev]
+    - name: Run tests (fast)
+      # this set of tests runs about 80% of the cases in 20% of the time,
+      # and should catch most errors quickly.
+      run:  pre-commit run pytest-fast --all-files --hook-stage manual
     - name: Run 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: [tests]
+    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: [tests]
+    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

@@ -108,17 +108,21 @@ venv.bak/
 *.viv
 *.idb
 *.i64
+.vscode
+
 !rules/lib
 
-# hooks/ci.sh output
-isort-output.log
-black-output.log
-rule-linter-output.log
-.vscode
 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

.gitmodules

@@ -1,6 +1,8 @@
 [submodule "rules"]
 	path = rules
 	url = ../capa-rules.git
+	branch = dynamic-syntax
 [submodule "tests/data"]
 	path = tests/data
 	url = ../capa-testfiles.git
+	branch = dynamic-feature-extractor

.pre-commit-config.yaml

@@ -0,0 +1,129 @@
+# 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, manual]
+        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, manual]
+        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, manual]
+        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: [push, manual]
+        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: [push, manual]
+        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
+
+-   repo: local
+    hooks:
+    -   id: pytest-fast
+        name: pytest (fast)
+        stages: [manual]
+        language: system
+        entry: pytest
+        args: 
+        -   "tests/"
+        -   "--ignore=tests/test_binja_features.py"
+        -   "--ignore=tests/test_ghidra_features.py"
+        -   "--ignore=tests/test_ida_features.py"
+        -   "--ignore=tests/test_viv_features.py"
+        -   "--ignore=tests/test_main.py"
+        -   "--ignore=tests/test_scripts.py"
+        always_run: true
+        pass_filenames: false

CHANGELOG.md

@@ -3,23 +3,404 @@
 ## master (unreleased)
 
 ### New Features
+- add Ghidra backend #1770 #1767 @colton-gabertan @mike-hunhoff
+- add dynamic analysis via CAPE sandbox reports #48 #1535 @yelhamer
+  - add call scope #771 @yelhamer
+  - add thread scope #1517 @yelhamer
+  - add process scope #1517 @yelhamer
+  - rules: change `meta.scope` to `meta.scopes` @yelhamer
+  - protobuf: add `Metadata.flavor` @williballenthin
+- binja: add support for forwarded exports #1646 @xusheng6
+- binja: add support for symtab names #1504 @xusheng6
+- add com class/interface features #322 @Aayush-goel-04
+- dotnet: emit enclosing class information for nested classes #1780 #1913 @bkojusner @mike-hunhoff
 
 ### Breaking Changes
 
-### New Rules (0)
+- remove the `SCOPE_*` constants in favor of the `Scope` enum #1764 @williballenthin
+- protobuf: deprecate `RuleMetadata.scope` in favor of `RuleMetadata.scopes` @williballenthin
+- protobuf: deprecate `Metadata.analysis` in favor of `Metadata.analysis2` that is dynamic analysis aware @williballenthin
+- update freeze format to v3, adding support for dynamic analysis @williballenthin
+- extractor: ignore DLL name for api features #1815 @mr-tz
 
-- 
+### New Rules (39)
+
+- nursery/get-ntoskrnl-base-address @mr-tz
+- host-interaction/network/connectivity/set-tcp-connection-state @johnk3r
+- nursery/capture-process-snapshot-data @mr-tz
+- collection/network/capture-packets-using-sharppcap jakub.jozwiak@mandiant.com
+- nursery/communicate-with-kernel-module-via-netlink-socket-on-linux michael.hunhoff@mandiant.com
+- nursery/get-current-pid-on-linux michael.hunhoff@mandiant.com
+- nursery/get-file-system-information-on-linux michael.hunhoff@mandiant.com
+- nursery/get-password-database-entry-on-linux michael.hunhoff@mandiant.com
+- nursery/mark-thread-detached-on-linux michael.hunhoff@mandiant.com
+- nursery/persist-via-gnome-autostart-on-linux michael.hunhoff@mandiant.com
+- nursery/set-thread-name-on-linux michael.hunhoff@mandiant.com
+- load-code/dotnet/load-windows-common-language-runtime michael.hunhoff@mandiant.com blas.kojusner@mandiant.com jakub.jozwiak@mandiant.com
+- nursery/log-keystrokes-via-input-method-manager @mr-tz
+- nursery/encrypt-data-using-rc4-via-systemfunction032 richard.weiss@mandiant.com
+- nursery/add-value-to-global-atom-table @mr-tz
+- nursery/enumerate-processes-that-use-resource @Ana06
+- host-interaction/process/inject/allocate-or-change-rwx-memory @mr-tz
+- lib/allocate-or-change-rw-memory 0x534a@mailbox.org @mr-tz
+- lib/change-memory-protection @mr-tz
+- anti-analysis/anti-av/patch-antimalware-scan-interface-function jakub.jozwiak@mandiant.com
+- executable/dotnet-singlefile/bundled-with-dotnet-single-file-deployment sara.rincon@mandiant.com
+- internal/limitation/file/internal-dotnet-single-file-deployment-limitation sara.rincon@mandiant.com
+- data-manipulation/encoding/encode-data-using-add-xor-sub-operations jakub.jozwiak@mandiant.com
+- nursery/access-camera-in-dotnet-on-android michael.hunhoff@mandiant.com
+- nursery/capture-microphone-audio-in-dotnet-on-android michael.hunhoff@mandiant.com
+- nursery/capture-screenshot-in-dotnet-on-android michael.hunhoff@mandiant.com
+- nursery/check-for-incoming-call-in-dotnet-on-android michael.hunhoff@mandiant.com
+- nursery/check-for-outgoing-call-in-dotnet-on-android michael.hunhoff@mandiant.com
+- nursery/compiled-with-xamarin michael.hunhoff@mandiant.com
+- nursery/get-os-version-in-dotnet-on-android michael.hunhoff@mandiant.com
+- data-manipulation/compression/create-cabinet-on-windows michael.hunhoff@mandiant.com jakub.jozwiak@mandiant.com
+- data-manipulation/compression/extract-cabinet-on-windows jakub.jozwiak@mandiant.com
+- lib/create-file-decompression-interface-context-on-windows jakub.jozwiak@mandiant.com
+- nursery/enumerate-files-in-dotnet moritz.raabe@mandiant.com anushka.virgaonkar@mandiant.com
+- nursery/get-mac-address-in-dotnet moritz.raabe@mandiant.com michael.hunhoff@mandiant.com echernofsky@google.com
+- nursery/get-current-process-command-line william.ballenthin@mandiant.com
+- nursery/get-current-process-file-path william.ballenthin@mandiant.com
+- nursery/hook-routines-via-dlsym-rtld_next william.ballenthin@mandiant.com
+-
 
 ### Bug Fixes
-
+- ghidra: fix `ints_to_bytes` performance #1761 @mike-hunhoff
+- binja: improve function call site detection @xusheng6
+- binja: use `binaryninja.load` to open files @xusheng6
+- binja: bump binja version to 3.5 #1789 @xusheng6
+- elf: better detect ELF OS via GCC .ident directives #1928 @williballenthin
 
 ### capa explorer IDA Pro plugin
 
 ### Development
+- update ATT&CK/MBC data for linting #1932 @mr-tz
 
 ### Raw diffs
-- [capa v4.0.1...master](https://github.com/mandiant/capa/compare/v4.0.1...master)
-- [capa-rules v4.0.1...master](https://github.com/mandiant/capa-rules/compare/v4.0.1...master)
+- [capa v6.1.0...master](https://github.com/mandiant/capa/compare/v6.1.0...master)
+- [capa-rules v6.1.0...master](https://github.com/mandiant/capa-rules/compare/v6.1.0...master)
+
+## v6.1.0
+
+capa v6.1.0 is a bug fix release, most notably fixing unhandled exceptions in the capa explorer IDA Pro plugin.
+@Aayush-Goel-04 put a lot of effort into improving code quality and adding a script for rule authors.
+The script shows which features are present in a sample but not referenced by any existing rule.
+You could use this script to find opportunities for new rules.
+
+Speaking of new rules, we have eight additions, coming from Ronnie, Jakub, Moritz, Ervin, and still@teamt5.org!
+
+### New Features
+- ELF: implement import and export name extractor #1607 #1608 @Aayush-Goel-04
+- bump pydantic from 1.10.9 to 2.1.1 #1582 @Aayush-Goel-04
+- develop script to highlight features not used during matching #331 @Aayush-Goel-04
+
+### New Rules (8)
+
+- executable/pe/export/forwarded-export ronnie.salomonsen@mandiant.com
+- host-interaction/bootloader/get-uefi-variable jakub.jozwiak@mandiant.com
+- host-interaction/bootloader/set-uefi-variable jakub.jozwiak@mandiant.com
+- nursery/enumerate-device-drivers-on-linux @mr-tz
+- anti-analysis/anti-vm/vm-detection/check-for-foreground-window-switch ervin.ocampo@mandiant.com
+- linking/static/sqlite3/linked-against-cppsqlite3 still@teamt5.org
+- linking/static/sqlite3/linked-against-sqlite3 still@teamt5.org
+
+### Bug Fixes
+
+- rules: fix forwarded export characteristic #1656 @RonnieSalomonsen
+- Binary Ninja: Fix stack string detection #1473 @xusheng6
+- linter: skip native API check for NtProtectVirtualMemory #1675 @williballenthin 
+- OS: detect Android ELF files #1705 @williballenthin
+- ELF: fix parsing of symtab #1704 @williballenthin
+- result document: don't use deprecated pydantic functions #1718 @williballenthin
+- pytest: don't mark IDA tests as pytest tests #1719 @williballenthin
+
+### capa explorer IDA Pro plugin
+- fix unhandled exception when resolving rule path #1693 @mike-hunhoff
+
+### Raw diffs
+- [capa v6.0.0...v6.1.0](https://github.com/mandiant/capa/compare/v6.0.0...v6.1.0)
+- [capa-rules v6.0.0...v6.1.0](https://github.com/mandiant/capa-rules/compare/v6.0.0...v6.1.0)
+
+## v6.0.0
+
+capa v6.0 brings many bug fixes and quality improvements, including 64 rule updates and 26 new rules. We're now publishing to PyPI via [Trusted Publishing](https://blog.pypi.org/posts/2023-04-20-introducing-trusted-publishers/) and have migrated to using a `pyproject.toml` file. @Aayush-Goel-04 contributed a lot of new code across many files, so please welcome them to the project, along with @anders-v @crowface28 @dkelly2e @RonnieSalomonsen and @ejfocampo as first-time rule contributors!
+
+For those that use capa as a library, we've introduced some limited breaking changes that better represent data types (versus less-structured data like dictionaries and strings). With the recent deprecation, we've also dropped support for Python 3.7.
+
+### New Features
+- add script to detect feature overlap between new and existing capa rules [#1451](https://github.com/mandiant/capa/issues/1451) [@Aayush-Goel-04](https://github.com/aayush-goel-04)
+- extract forwarded exports from PE files #1624 @williballenthin
+- extract function and API names from ELF symtab entries @yelhamer https://github.com/mandiant/capa-rules/issues/736
+- use fancy box drawing characters for default output #1586 @williballenthin
+
+### Breaking Changes
+- use a class to represent Metadata (not dict) #1411 @Aayush-Goel-04 @manasghandat
+- use pathlib.Path to represent file paths #1534 @Aayush-Goel-04
+- Python 3.8 is now the minimum supported Python version #1578 @williballenthin
+- Require a Contributor License Agreement (CLA) for PRs going forward #1642 @williballenthin
+
+### New Rules (26)
+
+- load-code/shellcode/execute-shellcode-via-windows-callback-function ervin.ocampo@mandiant.com jakub.jozwiak@mandiant.com
+- nursery/execute-shellcode-via-indirect-call ronnie.salomonsen@mandiant.com
+- data-manipulation/encryption/aes/encrypt-data-using-aes-mixcolumns-step @mr-tz
+- linking/static/aplib/linked-against-aplib still@teamt5.org
+- communication/mailslot/read-from-mailslot nick.simonian@mandiant.com
+- nursery/hash-data-using-sha512managed-in-dotnet jonathanlepore@google.com
+- nursery/compiled-with-exescript jonathanlepore@google.com
+- nursery/check-for-sandbox-via-mac-address-ouis-in-dotnet jonathanlepore@google.com
+- host-interaction/hardware/enumerate-devices-by-category @mr-tz
+- host-interaction/service/continue-service @mr-tz
+- host-interaction/service/pause-service @mr-tz
+- persistence/exchange/act-as-exchange-transport-agent jakub.jozwiak@mandiant.com
+- host-interaction/file-system/create-virtual-file-system-in-dotnet jakub.jozwiak@mandiant.com
+- compiler/cx_freeze/compiled-with-cx_freeze @mr-tz jakub.jozwiak@mandiant.com
+- communication/socket/create-vmci-socket jakub.jozwiak@mandiant.com
+- persistence/office/act-as-excel-xll-add-in jakub.jozwiak@mandiant.com
+- persistence/office/act-as-office-com-add-in jakub.jozwiak@mandiant.com
+- persistence/office/act-as-word-wll-add-in jakub.jozwiak@mandiant.com
+- anti-analysis/anti-debugging/debugger-evasion/hide-thread-from-debugger michael.hunhoff@mandiant.com jakub.jozwiak@mandiant.com
+- host-interaction/memory/create-new-application-domain-in-dotnet jakub.jozwiak@mandiant.com
+- host-interaction/gui/switch-active-desktop jakub.jozwiak@mandiant.com
+- host-interaction/service/query-service-configuration @mr-tz
+- anti-analysis/anti-av/patch-event-tracing-for-windows-function jakub.jozwiak@mandiant.com
+- data-manipulation/encoding/xor/covertly-decode-and-write-data-to-windows-directory-using-indirect-calls dan.kelly@mandiant.com
+- linking/runtime-linking/resolve-function-by-brute-ratel-badger-hash jakub.jozwiak@mandiant.com
+
+### Bug Fixes
+- extractor: add a Binary Ninja test that asserts its version #1487 @xusheng6
+- extractor: update Binary Ninja stack string detection after the new constant outlining feature #1473 @xusheng6
+- extractor: update vivisect Arch extraction #1334 @mr-tz
+- extractor: avoid Binary Ninja exception when analyzing certain files #1441 @xusheng6 
+- symtab: fix struct.unpack() format for 64-bit ELF files @yelhamer
+- symtab: safeguard against ZeroDivisionError for files containing a symtab with a null entry size @yelhamer
+- improve ELF strtab and needed parsing @mr-tz
+- better handle exceptional cases when parsing ELF files #1458 @Aayush-Goel-04
+- improved testing coverage for Binary Ninja backend #1446 @Aayush-Goel-04
+- add logging and print redirect to tqdm for capa main #749 @Aayush-Goel-04
+- extractor: fix binja installation path detection does not work with Python 3.11
+- tests: refine the IDA test runner script #1513 @williballenthin
+- output: don't leave behind traces of progress bar @williballenthin
+- import-to-ida: fix bug introduced with JSON report changes in v5 #1584 @williballenthin
+- main: don't show spinner when emitting debug messages #1636 @williballenthin
+- rules: add forwarded export characteristics to rule syntax file scope #1653 @RonnieSalomonsen
+
+### capa explorer IDA Pro plugin
+
+### Development
+- update ATT&CK/MBC data for linting #1568 @mr-tz
+- log time taken to analyze each function #1290 @williballenthin
+- tests: make fixture available via conftest.py #1592 @williballenthin
+- publish via PyPI trusted publishing #1491 @williballenthin
+- migrate to pyproject.toml #1301 @williballenthin
+- use [pre-commit](https://pre-commit.com/) to invoke linters #1579 @williballenthin
+
+
+### Raw diffs
+- [capa v5.1.0...v6.0.0](https://github.com/mandiant/capa/compare/v5.1.0...v6.0.0)
+- [capa-rules v5.1.0...v6.0.0](https://github.com/mandiant/capa-rules/compare/v5.1.0...v6.0.0)
+
+## v5.1.0
+capa version 5.1.0 adds a Protocol Buffers (protobuf) format for result documents. Additionally, the [Vector35](https://vector35.com/) team contributed a new feature extractor using Binary Ninja. Other new features are a new CLI flag to override the detected operating system, functionality to read and render existing result documents, and a output color format that's easier to read.
+
+Over 25 capa rules have been added and improved.
+
+Thanks for all the support, especially to @xusheng6, @captainGeech42, @ggold7046, @manasghandat, @ooprathamm, @linpeiyu164, @yelhamer, @HongThatCong, @naikordian, @stevemk14ebr, @emtuls, @raymondlleong, @bkojusner, @joren485, and everyone else who submitted bugs and provided feedback!
+
+### New Features
+- add protobuf format for result documents #1219 @williballenthin @mr-tz 
+- extractor: add Binary Ninja feature extractor @xusheng6
+- new cli flag `--os` to override auto-detected operating system for a sample @captainGeech42
+- change colour/highlight to "cyan" instead of "blue" for better readability #1384 @ggold7046
+- add new format to parse output json back to capa #1396 @ooprathamm
+- parse ELF symbols' names to guess OS #1403 @yelhamer
+
+### New Rules (26)
+
+- persistence/scheduled-tasks/schedule-task-via-at joren485
+- data-manipulation/prng/generate-random-numbers-via-rtlgenrandom william.ballenthin@mandiant.com
+- communication/ip/convert-ip-address-from-string @mr-tz
+- data-manipulation/compression/compress-data-via-zlib-inflate-or-deflate blas.kojusner@mandiant.com
+- executable/installer/dotnet/packaged-as-single-file-dotnet-application michael.hunhoff@mandiant.com
+- communication/socket/create-raw-socket blas.kojusner@mandiant.com
+- communication/http/reference-http-user-agent-string @mr-tz
+- communication/http/get-http-content-length william.ballenthin@mandiant.com
+- nursery/move-directory michael.hunhoff@mandiant.com
+- nursery/get-http-request-uri william.ballenthin@mandiant.com
+- nursery/create-zip-archive-in-dotnet michael.hunhoff@mandiant.com
+- nursery/extract-zip-archive-in-dotnet anushka.virgaonkar@mandiant.com michael.hunhoff@mandiant.com
+- data-manipulation/encryption/tea/decrypt-data-using-tea william.ballenthin@mandiant.com raymond.leong@mandiant.com
+- data-manipulation/encryption/tea/encrypt-data-using-tea william.ballenthin@mandiant.com raymond.leong@mandiant.com
+- data-manipulation/encryption/xtea/encrypt-data-using-xtea raymond.leong@mandiant.com
+- data-manipulation/encryption/xxtea/encrypt-data-using-xxtea raymond.leong@mandiant.com
+- nursery/hash-data-using-ripemd128 raymond.leong@mandiant.com
+- nursery/hash-data-using-ripemd256 raymond.leong@mandiant.com
+- nursery/hash-data-using-ripemd320 raymond.leong@mandiant.com
+- nursery/set-web-proxy-in-dotnet michael.hunhoff@mandiant.com
+- nursery/check-for-windows-sandbox-via-subdirectory echernofsky@google.com
+- nursery/enumerate-pe-sections-in-dotnet @mr-tz
+- nursery/destroy-software-breakpoint-capability echernofsky@google.com
+- nursery/send-data-to-internet michael.hunhoff@mandiant.com
+- nursery/compiled-with-cx_freeze @mr-tz
+- nursery/contain-a-thread-local-storage-tls-section-in-dotnet michael.hunhoff@mandiant.com
+
+### Bug Fixes
+- extractor: interface of cache modified to prevent extracting file and global features multiple times @stevemk14ebr
+- extractor: removed '.dynsym' as the library name for ELF imports #1318 @stevemk14ebr 
+- extractor: fix vivisect loop detection corner case #1310 @mr-tz
+- match: extend OS characteristic to match OS_ANY to all supported OSes #1324 @mike-hunhoff
+- extractor: fix IDA and vivisect string and bytes features overlap and tests #1327 #1336 @xusheng6
+
+### capa explorer IDA Pro plugin
+- rule generator plugin now loads faster when jumping between functions @stevemk14ebr
+- fix exception when plugin loaded in IDA hosted under idat #1341 @mike-hunhoff
+- improve embedded PE detection performance and reduce FP potential #1344 @mike-hunhoff
+
+### Raw diffs
+- [capa v5.0.0...v5.1.0](https://github.com/mandiant/capa/compare/v5.0.0...v5.1.0)
+- [capa-rules v5.0.0...v5.1.0](https://github.com/mandiant/capa-rules/compare/v5.0.0...v5.1.0)
+
+
+## v5.0.0 (2023-02-08)
+This capa version comes with major improvements and additions to better handle .NET binaries. To showcase this we've updated and added over 30 .NET rules.  
+
+Additionally, capa now caches its rule set for better performance. The capa explorer also caches its analysis results, so that multiple IDA Pro or plugin invocations don't need to repeat the same analysis.
+
+We have removed the SMDA backend and changed the program return codes to be positive numbers.
+
+Other improvements to highlight include better ELF OS detection, various rendering bug fixes, and enhancements to the feature extraction. We've also added support for Python 3.11.
+
+Thanks for all the support, especially to @jsoref, @bkojusner, @edeca, @richardweiss80, @joren485, @ryantxu1, @mwilliams31, @anushkavirgaonkar, @MalwareMechanic, @Still34, @dzbeck, @johnk3r, and everyone else who submitted bugs and provided feedback!
+
+### New Features
+- verify rule metadata format on load #1160 @mr-tz
+- dotnet: emit property features #1168 @anushkavirgaonkar
+- dotnet: emit API features for objects created via the newobj instruction #1186 @mike-hunhoff
+- dotnet: emit API features for generic methods #1231 @mike-hunhoff
+- Python 3.11 support #1192 @williballenthin
+- dotnet: emit calls to/from MethodDef methods #1236 @mike-hunhoff
+- dotnet: emit namespace/class features for ldvirtftn/ldftn instructions #1241 @mike-hunhoff
+- dotnet: emit namespace/class features for type references #1242 @mike-hunhoff
+- dotnet: extract dotnet and pe format #1187 @mr-tz
+- don't render all library rule matches in vverbose output #1174 @mr-tz
+- cache the rule set across invocations for better performance #1212 @williballenthin
+- update ATT&CK/MBC data for linting #1297 @mr-tz
+
+### Breaking Changes
+- remove SMDA backend #1062 @williballenthin
+- error return codes are now positive numbers #1269 @mr-tz
+
+### New Rules (77)
+
+- collection/use-dotnet-library-sharpclipboard @johnk3r
+- data-manipulation/encryption/aes/use-dotnet-library-encryptdecryptutils @johnk3r
+- data-manipulation/json/use-dotnet-library-newtonsoftjson @johnk3r
+- data-manipulation/svg/use-dotnet-library-sharpvectors @johnk3r
+- executable/resource/embed-dependencies-as-resources-using-fodycostura @johnk3r @mr-tz
+- communication/ftp/send/send-file-using-ftp michael.hunhof@mandiant.com anushka.virgaonkar@mandiant.com
+- nursery/extract-zip-archive anushka.virgaonkar@mandiant.com
+- nursery/allocate-unmanaged-memory-in-dotnet michael.hunhoff@mandiant.com
+- nursery/check-file-extension-in-dotnet michael.hunhoff@mandiant.com
+- nursery/decode-data-using-base64-in-dotnet michael.hunhoff@mandiant.com
+- nursery/deserialize-json-in-dotnet michael.hunhoff@mandiant.com
+- nursery/find-data-using-regex-in-dotnet michael.hunhoff@mandiant.com
+- nursery/generate-random-filename-in-dotnet michael.hunhoff@mandiant.com
+- nursery/get-os-version-in-dotnet michael.hunhoff@mandiant.com
+- nursery/load-xml-in-dotnet michael.hunhoff@mandiant.com
+- nursery/manipulate-unmanaged-memory-in-dotnet michael.hunhoff@mandiant.com
+- nursery/save-image-in-dotnet michael.hunhoff@mandiant.com
+- nursery/send-email-in-dotnet michael.hunhoff@mandiant.com
+- nursery/serialize-json-in-dotnet michael.hunhoff@mandiant.com
+- nursery/set-http-user-agent-in-dotnet michael.hunhoff@mandiant.com
+- nursery/compile-csharp-in-dotnet michael.hunhoff@mandiant.com
+- nursery/compile-visual-basic-in-dotnet michael.hunhoff@mandiant.com
+- nursery/compress-data-using-gzip-in-dotnet michael.hunhoff@mandiant.com
+- nursery/execute-sqlite-statement-in-dotnet michael.hunhoff@mandiant.com
+- nursery/execute-via-asynchronous-task-in-dotnet michael.hunhoff@mandiant.com
+- nursery/execute-via-timer-in-dotnet michael.hunhoff@mandiant.com
+- nursery/execute-wmi-query-in-dotnet michael.hunhoff@mandiant.com
+- nursery/manipulate-network-credentials-in-dotnet michael.hunhoff@mandiant.com
+- nursery/encrypt-data-using-aes william.ballenthin@mandiant.com Ivan Kwiatkowski (@JusticeRage)
+- host-interaction/uac/bypass/bypass-uac-via-rpc david.cannings@pwc.com david@edeca.net
+- nursery/check-for-vm-using-instruction-vpcext richard.weiss@mandiant.com
+- nursery/get-windows-directory-from-kuser_shared_data david.cannings@pwc.com
+- nursery/encrypt-data-using-openssl-dsa Ana06
+- nursery/encrypt-data-using-openssl-ecdsa Ana06
+- nursery/encrypt-data-using-openssl-rsa Ana06
+- runtime/dotnet/execute-via-dotnet-startup-hook william.ballenthin@mandiant.com
+- host-interaction/console/manipulate-console-buffer william.ballenthin@mandiant.com michael.hunhoff@mandiant.com
+- nursery/access-wmi-data-in-dotnet michael.hunhoff@mandiant.com
+- nursery/allocate-unmanaged-memory-via-dotnet michael.hunhoff@mandiant.com
+- nursery/generate-random-bytes-in-dotnet michael.hunhoff@mandiant.com
+- nursery/manipulate-console-window michael.hunhoff@mandiant.com
+- nursery/obfuscated-with-koivm michael.hunhoff@mandiant.com
+- nursery/implement-com-dll moritz.raabe@mandiant.com
+- nursery/linked-against-libsodium @mr-tz
+- compiler/nuitka/compiled-with-nuitka @williballenthin
+- nursery/authenticate-data-with-md5-mac william.ballenthin@mandiant.com
+- nursery/resolve-function-by-djb2-hash still@teamt5.org
+- host-interaction/mutex/create-semaphore-on-linux @ramen0x3f
+- host-interaction/mutex/lock-semaphore-on-linux @ramen0x3f
+- host-interaction/mutex/unlock-semaphore-on-linux @ramen0x3f
+- data-manipulation/hashing/sha384/hash-data-using-sha384 william.ballenthin@mandiant.com
+- data-manipulation/hashing/sha512/hash-data-using-sha512 william.ballenthin@mandiant.com
+- nursery/decode-data-using-url-encoding michael.hunhoff@mandiant.com
+- nursery/manipulate-user-privileges michael.hunhoff@mandiant.com
+- lib/get-os-version @mr-tz
+- nursery/decrypt-data-using-tea william.ballenthin@mandiant.com
+- nursery/encrypt-data-using-tea william.ballenthin@mandiant.com
+- nursery/hash-data-using-whirlpool william.ballenthin@mandiant.com
+- nursery/reference-base58-string william.ballenthin@mandiant.com
+- communication/mailslot/create-mailslot william.ballenthin@mandiant.com
+- executable/resource/access-dotnet-resource @mr-tz
+- linking/static/linked-against-cpp-standard-library @mr-tz
+- data-manipulation/compression/compress-data-using-lzo david@edeca.net david.cannings@pwc.com
+- data-manipulation/compression/decompress-data-using-lzo david@edeca.net david.cannings@pwc.com
+- communication/socket/tcp/create-tcp-socket-via-raw-afd-driver william.ballenthin@mandiant.com
+- host-interaction/process/map-section-object william.ballenthin@mandiant.com
+- lib/create-or-open-section-object william.ballenthin@mandiant.com
+- load-code/dotnet/execute-dotnet-assembly-via-clr-host blas.kojusner@mandiant.com
+- load-code/execute-vbscript-javascript-or-jscript-in-memory blas.kojusner@mandiant.com
+- host-interaction/file-system/reference-absolute-stream-path-on-windows blas.kojusner@mandiant.com
+- nursery/generate-method-via-reflection-in-dotnet michael.hunhoff@mandiant.com
+- nursery/unmanaged-call-via-dynamic-pinvoke-in-dotnet michael.hunhoff@mandiant.com
+
+### Bug Fixes
+- render: convert feature attributes to aliased dictionary for vverbose #1152 @mike-hunhoff
+- decouple Token dependency / extractor and features #1139 @mr-tz
+- update pydantic model to guarantee type coercion #1176 @mike-hunhoff
+- do not overwrite version in version.py during PyInstaller build #1169 @mr-tz
+- render: fix vverbose rendering of offsets #1215 @williballenthin
+- elf: better detect OS via GLIBC ABI version needed and dependencies #1221 @williballenthin
+- dotnet: address unhandled exceptions with improved type checking #1230 @mike-hunhoff
+- fix import-to-ida script formatting #1208 @williballenthin
+- render: fix verbose rendering of scopes #1263 @williballenthin
+- rules: better detect invalid rules #1282 @williballenthin
+- show-features: better render strings with embedded whitespace #1267 @williballenthin
+- handle vivisect bug around strings at instruction level, use min length 4 #1271 @williballenthin @mr-tz
+- extractor: guard against invalid "calls from" features #1177 @mr-tz
+- extractor: add format to global features #1258 @mr-tz
+- extractor: discover all strings with length >= 4 #1280 @mr-tz
+- extractor: don't extract byte features for strings #1293 @mr-tz
+
+### capa explorer IDA Pro plugin
+- fix: display instruction items #1154 @mr-tz
+- fix: accept only plaintext pasted content #1194 @williballenthin
+- fix: UnboundLocalError #1217 @williballenthin
+- extractor: add support for COFF files and extern functions #1223 @mike-hunhoff
+- doc: improve error messaging and documentation related to capa rule set #1249 @mike-hunhoff
+- fix: assume 32-bit displacement for offsets #1250 @mike-hunhoff
+- generator: refactor caching and matching #1251 @mike-hunhoff
+- fix: improve exception handling to prevent IDA from locking up when errors occur #1262 @mike-hunhoff
+- verify rule metadata using Pydantic #1167 @mr-tz
+- extractor: make read consistent with file object behavior #1254 @mr-tz
+- fix: UnboundLocalError x2 #1302 @mike-hunhoff
+- cache capa results across IDA sessions #1279 @mr-tz
+
+### Raw diffs
+- [capa v4.0.1...v5.0.0](https://github.com/mandiant/capa/compare/v4.0.1...v5.0.0)
+- [capa-rules v4.0.1...v5.0.0](https://github.com/mandiant/capa-rules/compare/v4.0.1...v5.0.0)
 
 
 ## v4.0.1 (2022-08-15)
@@ -1245,7 +1626,7 @@ Download a standalone binary below and checkout the readme [here on GitHub](http
   - setup: pin vivisect version @williballenthin
   - setup: bump vivisect dependency version @williballenthin
   - setup: set Python project name to `flare-capa` @williballenthin
-  - ci: run tests and linter via Github Actions @Ana06
+  - ci: run tests and linter via GitHub Actions @Ana06
   - hooks: run style checkers and hide stashed output @Ana06
   - linter: ignore period in rule filename @williballenthin
   - linter: warn on nursery rule with no changes needed @williballenthin

LICENSE.txt

@@ -187,7 +187,7 @@
       same "printed page" as the copyright notice for easier
       identification within third-party archives.
 
-   Copyright (C) 2020 Mandiant, 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

@@ -2,13 +2,13 @@
 
 [![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/mandiant/capa)](https://github.com/mandiant/capa/releases)
-[![Number of rules](https://img.shields.io/badge/rules-703-blue.svg)](https://github.com/mandiant/capa-rules)
+[![Number of rules](https://img.shields.io/badge/rules-864-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, ELF, .NET module, or shellcode file and it tells you what it thinks the program can do.
+You run it against a PE, ELF, .NET module, shellcode file, or a sandbox report 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:
@@ -125,6 +125,96 @@ function @ 0x4011C0
 ...
 ```
 
+Additionally, capa also supports analyzing [CAPE](https://github.com/kevoreilly/CAPEv2) sandbox reports for dynamic capabilty extraction.
+In order to use this, you first submit your sample to CAPE for analysis, and then run capa against the generated report (JSON).
+
+Here's an example of running capa against a packed binary, and then running capa against the CAPE report of that binary:
+
+```yaml
+$ capa 05be49819139a3fdcdbddbdefd298398779521f3d68daa25275cc77508e42310.exe
+WARNING:capa.capabilities.common:--------------------------------------------------------------------------------
+WARNING:capa.capabilities.common: This sample appears to be packed.
+WARNING:capa.capabilities.common: 
+WARNING:capa.capabilities.common: Packed samples have often been obfuscated to hide their logic.
+WARNING:capa.capabilities.common: capa cannot handle obfuscation well using static analysis. This means the results may be misleading or incomplete.
+WARNING:capa.capabilities.common: If possible, you should try to unpack this input file before analyzing it with capa.
+WARNING:capa.capabilities.common: Alternatively, run the sample in a supported sandbox and invoke capa against the report to obtain dynamic analysis results.
+WARNING:capa.capabilities.common: 
+WARNING:capa.capabilities.common: Identified via rule: (internal) packer file limitation
+WARNING:capa.capabilities.common: 
+WARNING:capa.capabilities.common: Use -v or -vv if you really want to see the capabilities identified by capa.
+WARNING:capa.capabilities.common:--------------------------------------------------------------------------------
+
+$ capa 05be49819139a3fdcdbddbdefd298398779521f3d68daa25275cc77508e42310.json
+
+┍━━━━━━━━━━━━━━━━━━━━━━━━┯━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┑
+│ ATT&CK Tactic          │ ATT&CK Technique                                                                   │
+┝━━━━━━━━━━━━━━━━━━━━━━━━┿━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┥
+│ CREDENTIAL ACCESS      │ Credentials from Password Stores T1555                                             │
+├────────────────────────┼────────────────────────────────────────────────────────────────────────────────────┤
+│ DEFENSE EVASION        │ File and Directory Permissions Modification T1222                                  │
+│                        │ Modify Registry T1112                                                              │
+│                        │ Obfuscated Files or Information T1027                                              │
+│                        │ Virtualization/Sandbox Evasion::User Activity Based Checks T1497.002               │
+├────────────────────────┼────────────────────────────────────────────────────────────────────────────────────┤
+│ DISCOVERY              │ Account Discovery T1087                                                            │
+│                        │ Application Window Discovery T1010                                                 │
+│                        │ File and Directory Discovery T1083                                                 │
+│                        │ Query Registry T1012                                                               │
+│                        │ System Information Discovery T1082                                                 │
+│                        │ System Location Discovery::System Language Discovery T1614.001                     │
+│                        │ System Owner/User Discovery T1033                                                  │
+├────────────────────────┼────────────────────────────────────────────────────────────────────────────────────┤
+│ EXECUTION              │ System Services::Service Execution T1569.002                                       │
+├────────────────────────┼────────────────────────────────────────────────────────────────────────────────────┤
+│ PERSISTENCE            │ Boot or Logon Autostart Execution::Registry Run Keys / Startup Folder T1547.001    │
+│                        │ Boot or Logon Autostart Execution::Winlogon Helper DLL T1547.004                   │
+│                        │ Create or Modify System Process::Windows Service T1543.003                         │
+┕━━━━━━━━━━━━━━━━━━━━━━━━┷━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┙
+
+┍━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┯━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┑
+│ Capability                                           │ Namespace                                            │
+┝━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┿━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┥
+│ check for unmoving mouse cursor (3 matches)          │ anti-analysis/anti-vm/vm-detection                   │
+│ gather bitkinex information                          │ collection/file-managers                             │
+│ gather classicftp information                        │ collection/file-managers                             │
+│ gather filezilla information                         │ collection/file-managers                             │
+│ gather total-commander information                   │ collection/file-managers                             │
+│ gather ultrafxp information                          │ collection/file-managers                             │
+│ resolve DNS (23 matches)                             │ communication/dns                                    │
+│ initialize Winsock library (7 matches)               │ communication/socket                                 │
+│ act as TCP client (3 matches)                        │ communication/tcp/client                             │
+│ create new key via CryptAcquireContext               │ data-manipulation/encryption                         │
+│ encrypt or decrypt via WinCrypt                      │ data-manipulation/encryption                         │
+│ hash data via WinCrypt                               │ data-manipulation/hashing                            │
+│ initialize hashing via WinCrypt                      │ data-manipulation/hashing                            │
+│ hash data with MD5                                   │ data-manipulation/hashing/md5                        │
+│ generate random numbers via WinAPI                   │ data-manipulation/prng                               │
+│ extract resource via kernel32 functions (2 matches)  │ executable/resource                                  │
+│ interact with driver via control codes (2 matches)   │ host-interaction/driver                              │
+│ get Program Files directory (18 matches)             │ host-interaction/file-system                         │
+│ get common file path (575 matches)                   │ host-interaction/file-system                         │
+│ create directory (2 matches)                         │ host-interaction/file-system/create                  │
+│ delete file                                          │ host-interaction/file-system/delete                  │
+│ get file attributes (122 matches)                    │ host-interaction/file-system/meta                    │
+│ set file attributes (8 matches)                      │ host-interaction/file-system/meta                    │
+│ move file                                            │ host-interaction/file-system/move                    │
+│ find taskbar (3 matches)                             │ host-interaction/gui/taskbar/find                    │
+│ get keyboard layout (12 matches)                     │ host-interaction/hardware/keyboard                   │
+│ get disk size                                        │ host-interaction/hardware/storage                    │
+│ get hostname (4 matches)                             │ host-interaction/os/hostname                         │
+│ allocate or change RWX memory (3 matches)            │ host-interaction/process/inject                      │
+│ query or enumerate registry key (3 matches)          │ host-interaction/registry                            │
+│ query or enumerate registry value (8 matches)        │ host-interaction/registry                            │
+│ delete registry key                                  │ host-interaction/registry/delete                     │
+│ start service                                        │ host-interaction/service/start                       │
+│ get session user name                                │ host-interaction/session                             │
+│ persist via Run registry key                         │ persistence/registry/run                             │
+│ persist via Winlogon Helper DLL registry key         │ persistence/registry/winlogon-helper                 │
+│ persist via Windows service (2 matches)              │ persistence/service                                  │
+┕━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┷━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┙
+```
+
 capa uses a collection of rules to identify capabilities within a program.
 These rules are easy to write, even for those new to reverse engineering.
 By authoring rules, you can extend the capabilities that capa recognizes.
@@ -135,31 +225,30 @@ Here's an example rule used by capa:
 ```yaml
 rule:
   meta:
-    name: hash data with CRC32
-    namespace: data-manipulation/checksum/crc32
+    name: create TCP socket
+    namespace: communication/socket/tcp
     authors:
-      - moritz.raabe@mandiant.com
-    scope: function
+      - william.ballenthin@mandiant.com
+      - joakim@intezer.com
+      - anushka.virgaonkar@mandiant.com
+    scopes:
+      static: basic block
+      dynamic: call
     mbc:
-      - Data::Checksum::CRC32 [C0032.001]
+      - Communication::Socket Communication::Create TCP Socket [C0001.011]
     examples:
-      - 2D3EDC218A90F03089CC01715A9F047F:0x403CBD
-      - 7D28CB106CB54876B2A5C111724A07CD:0x402350  # RtlComputeCrc32
-      - 7EFF498DE13CC734262F87E6B3EF38AB:0x100084A6
+      - Practical Malware Analysis Lab 01-01.dll_:0x10001010
   features:
     - or:
       - and:
-        - mnemonic: shr
+        - number: 6 = IPPROTO_TCP
+        - number: 1 = SOCK_STREAM
+        - number: 2 = AF_INET
         - 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
+          - api: ws2_32.socket
+          - api: ws2_32.WSASocket
+          - api: socket
+      - property/read: System.Net.Sockets.TcpClient::Client
 ```
 
 The [github.com/mandiant/capa-rules](https://github.com/mandiant/capa-rules) repository contains hundreds of standard library rules that are distributed with capa.
@@ -170,6 +259,8 @@ capa explorer helps you identify interesting areas of a program and build new ca
 
 ![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
 - [Installation](https://github.com/mandiant/capa/blob/master/doc/installation.md)

capa/capabilities/common.py

@@ -0,0 +1,79 @@
+# -*- coding: utf-8 -*-
+# 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 itertools
+import collections
+from typing import Any, Tuple
+
+from capa.rules import Scope, RuleSet
+from capa.engine import FeatureSet, MatchResults
+from capa.features.address import NO_ADDRESS
+from capa.features.extractors.base_extractor import FeatureExtractor, StaticFeatureExtractor, DynamicFeatureExtractor
+
+logger = logging.getLogger(__name__)
+
+
+def find_file_capabilities(ruleset: RuleSet, extractor: FeatureExtractor, function_features: FeatureSet):
+    file_features: FeatureSet = collections.defaultdict(set)
+
+    for feature, va in itertools.chain(extractor.extract_file_features(), extractor.extract_global_features()):
+        # not all file features may have virtual addresses.
+        # if not, then at least ensure the feature shows up in the index.
+        # the set of addresses will still be empty.
+        if va:
+            file_features[feature].add(va)
+        else:
+            if feature not in file_features:
+                file_features[feature] = set()
+
+    logger.debug("analyzed file and extracted %d features", len(file_features))
+
+    file_features.update(function_features)
+
+    _, matches = ruleset.match(Scope.FILE, file_features, NO_ADDRESS)
+    return matches, len(file_features)
+
+
+def has_file_limitation(rules: RuleSet, capabilities: MatchResults, is_standalone=True) -> bool:
+    file_limitation_rules = list(filter(lambda r: r.is_file_limitation_rule(), rules.rules.values()))
+
+    for file_limitation_rule in file_limitation_rules:
+        if file_limitation_rule.name not in capabilities:
+            continue
+
+        logger.warning("-" * 80)
+        for line in file_limitation_rule.meta.get("description", "").split("\n"):
+            logger.warning(" %s", line)
+        logger.warning(" Identified via rule: %s", file_limitation_rule.name)
+        if is_standalone:
+            logger.warning(" ")
+            logger.warning(" Use -v or -vv if you really want to see the capabilities identified by capa.")
+        logger.warning("-" * 80)
+
+        # bail on first file limitation
+        return True
+
+    return False
+
+
+def find_capabilities(
+    ruleset: RuleSet, extractor: FeatureExtractor, disable_progress=None, **kwargs
+) -> Tuple[MatchResults, Any]:
+    from capa.capabilities.static import find_static_capabilities
+    from capa.capabilities.dynamic import find_dynamic_capabilities
+
+    if isinstance(extractor, StaticFeatureExtractor):
+        # for the time being, extractors are either static or dynamic.
+        # Remove this assertion once that has changed
+        assert not isinstance(extractor, DynamicFeatureExtractor)
+        return find_static_capabilities(ruleset, extractor, disable_progress=disable_progress, **kwargs)
+    if isinstance(extractor, DynamicFeatureExtractor):
+        return find_dynamic_capabilities(ruleset, extractor, disable_progress=disable_progress, **kwargs)
+
+    raise ValueError(f"unexpected extractor type: {extractor.__class__.__name__}")

capa/capabilities/dynamic.py

@@ -0,0 +1,198 @@
+# -*- coding: utf-8 -*-
+# 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 itertools
+import collections
+from typing import Any, Tuple
+
+import tqdm
+
+import capa.perf
+import capa.features.freeze as frz
+import capa.render.result_document as rdoc
+from capa.rules import Scope, RuleSet
+from capa.engine import FeatureSet, MatchResults
+from capa.helpers import redirecting_print_to_tqdm
+from capa.capabilities.common import find_file_capabilities
+from capa.features.extractors.base_extractor import CallHandle, ThreadHandle, ProcessHandle, DynamicFeatureExtractor
+
+logger = logging.getLogger(__name__)
+
+
+def find_call_capabilities(
+    ruleset: RuleSet, extractor: DynamicFeatureExtractor, ph: ProcessHandle, th: ThreadHandle, ch: CallHandle
+) -> Tuple[FeatureSet, MatchResults]:
+    """
+    find matches for the given rules for the given call.
+
+    returns: tuple containing (features for call, match results for call)
+    """
+    # all features found for the call.
+    features: FeatureSet = collections.defaultdict(set)
+
+    for feature, addr in itertools.chain(
+        extractor.extract_call_features(ph, th, ch), extractor.extract_global_features()
+    ):
+        features[feature].add(addr)
+
+    # matches found at this thread.
+    _, matches = ruleset.match(Scope.CALL, features, ch.address)
+
+    for rule_name, res in matches.items():
+        rule = ruleset[rule_name]
+        for addr, _ in res:
+            capa.engine.index_rule_matches(features, rule, [addr])
+
+    return features, matches
+
+
+def find_thread_capabilities(
+    ruleset: RuleSet, extractor: DynamicFeatureExtractor, ph: ProcessHandle, th: ThreadHandle
+) -> Tuple[FeatureSet, MatchResults, MatchResults]:
+    """
+    find matches for the given rules within the given thread.
+
+    returns: tuple containing (features for thread, match results for thread, match results for calls)
+    """
+    # all features found within this thread,
+    # includes features found within calls.
+    features: FeatureSet = collections.defaultdict(set)
+
+    # matches found at the call scope.
+    # might be found at different calls, thats ok.
+    call_matches: MatchResults = collections.defaultdict(list)
+
+    for ch in extractor.get_calls(ph, th):
+        ifeatures, imatches = find_call_capabilities(ruleset, extractor, ph, th, ch)
+        for feature, vas in ifeatures.items():
+            features[feature].update(vas)
+
+        for rule_name, res in imatches.items():
+            call_matches[rule_name].extend(res)
+
+    for feature, va in itertools.chain(extractor.extract_thread_features(ph, th), extractor.extract_global_features()):
+        features[feature].add(va)
+
+    # matches found within this thread.
+    _, matches = ruleset.match(Scope.THREAD, features, th.address)
+
+    for rule_name, res in matches.items():
+        rule = ruleset[rule_name]
+        for va, _ in res:
+            capa.engine.index_rule_matches(features, rule, [va])
+
+    return features, matches, call_matches
+
+
+def find_process_capabilities(
+    ruleset: RuleSet, extractor: DynamicFeatureExtractor, ph: ProcessHandle
+) -> Tuple[MatchResults, MatchResults, MatchResults, int]:
+    """
+    find matches for the given rules within the given process.
+
+    returns: tuple containing (match results for process, match results for threads, match results for calls, number of features)
+    """
+    # all features found within this process,
+    # includes features found within threads (and calls).
+    process_features: FeatureSet = collections.defaultdict(set)
+
+    # matches found at the basic threads.
+    # might be found at different threads, thats ok.
+    thread_matches: MatchResults = collections.defaultdict(list)
+
+    # matches found at the call scope.
+    # might be found at different calls, thats ok.
+    call_matches: MatchResults = collections.defaultdict(list)
+
+    for th in extractor.get_threads(ph):
+        features, tmatches, cmatches = find_thread_capabilities(ruleset, extractor, ph, th)
+        for feature, vas in features.items():
+            process_features[feature].update(vas)
+
+        for rule_name, res in tmatches.items():
+            thread_matches[rule_name].extend(res)
+
+        for rule_name, res in cmatches.items():
+            call_matches[rule_name].extend(res)
+
+    for feature, va in itertools.chain(extractor.extract_process_features(ph), extractor.extract_global_features()):
+        process_features[feature].add(va)
+
+    _, process_matches = ruleset.match(Scope.PROCESS, process_features, ph.address)
+    return process_matches, thread_matches, call_matches, len(process_features)
+
+
+def find_dynamic_capabilities(
+    ruleset: RuleSet, extractor: DynamicFeatureExtractor, disable_progress=None
+) -> Tuple[MatchResults, Any]:
+    all_process_matches: MatchResults = collections.defaultdict(list)
+    all_thread_matches: MatchResults = collections.defaultdict(list)
+    all_call_matches: MatchResults = collections.defaultdict(list)
+
+    feature_counts = rdoc.DynamicFeatureCounts(file=0, processes=())
+
+    assert isinstance(extractor, DynamicFeatureExtractor)
+    with redirecting_print_to_tqdm(disable_progress):
+        with tqdm.contrib.logging.logging_redirect_tqdm():
+            pbar = tqdm.tqdm
+            if disable_progress:
+                # do not use tqdm to avoid unnecessary side effects when caller intends
+                # to disable progress completely
+                def pbar(s, *args, **kwargs):
+                    return s
+
+            processes = list(extractor.get_processes())
+
+            pb = pbar(processes, desc="matching", unit=" processes", leave=False)
+            for p in pb:
+                process_matches, thread_matches, call_matches, feature_count = find_process_capabilities(
+                    ruleset, extractor, p
+                )
+                feature_counts.processes += (
+                    rdoc.ProcessFeatureCount(address=frz.Address.from_capa(p.address), count=feature_count),
+                )
+                logger.debug("analyzed %s and extracted %d features", p.address, feature_count)
+
+                for rule_name, res in process_matches.items():
+                    all_process_matches[rule_name].extend(res)
+                for rule_name, res in thread_matches.items():
+                    all_thread_matches[rule_name].extend(res)
+                for rule_name, res in call_matches.items():
+                    all_call_matches[rule_name].extend(res)
+
+    # collection of features that captures the rule matches within process and thread scopes.
+    # mapping from feature (matched rule) to set of addresses at which it matched.
+    process_and_lower_features: FeatureSet = collections.defaultdict(set)
+    for rule_name, results in itertools.chain(
+        all_process_matches.items(), all_thread_matches.items(), all_call_matches.items()
+    ):
+        locations = {p[0] for p in results}
+        rule = ruleset[rule_name]
+        capa.engine.index_rule_matches(process_and_lower_features, rule, locations)
+
+    all_file_matches, feature_count = find_file_capabilities(ruleset, extractor, process_and_lower_features)
+    feature_counts.file = feature_count
+
+    matches = dict(
+        itertools.chain(
+            # each rule exists in exactly one scope,
+            # so there won't be any overlap among these following MatchResults,
+            # and we can merge the dictionaries naively.
+            all_thread_matches.items(),
+            all_process_matches.items(),
+            all_call_matches.items(),
+            all_file_matches.items(),
+        )
+    )
+
+    meta = {
+        "feature_counts": feature_counts,
+    }
+
+    return matches, meta

capa/capabilities/static.py

@@ -0,0 +1,233 @@
+# -*- coding: utf-8 -*-
+# 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 time
+import logging
+import itertools
+import collections
+from typing import Any, Tuple
+
+import tqdm.contrib.logging
+
+import capa.perf
+import capa.features.freeze as frz
+import capa.render.result_document as rdoc
+from capa.rules import Scope, RuleSet
+from capa.engine import FeatureSet, MatchResults
+from capa.helpers import redirecting_print_to_tqdm
+from capa.capabilities.common import find_file_capabilities
+from capa.features.extractors.base_extractor import BBHandle, InsnHandle, FunctionHandle, StaticFeatureExtractor
+
+logger = logging.getLogger(__name__)
+
+
+def find_instruction_capabilities(
+    ruleset: RuleSet, extractor: StaticFeatureExtractor, f: FunctionHandle, bb: BBHandle, insn: InsnHandle
+) -> Tuple[FeatureSet, MatchResults]:
+    """
+    find matches for the given rules for the given instruction.
+
+    returns: tuple containing (features for instruction, match results for instruction)
+    """
+    # all features found for the instruction.
+    features: FeatureSet = collections.defaultdict(set)
+
+    for feature, addr in itertools.chain(
+        extractor.extract_insn_features(f, bb, insn), extractor.extract_global_features()
+    ):
+        features[feature].add(addr)
+
+    # matches found at this instruction.
+    _, matches = ruleset.match(Scope.INSTRUCTION, features, insn.address)
+
+    for rule_name, res in matches.items():
+        rule = ruleset[rule_name]
+        for addr, _ in res:
+            capa.engine.index_rule_matches(features, rule, [addr])
+
+    return features, matches
+
+
+def find_basic_block_capabilities(
+    ruleset: RuleSet, extractor: StaticFeatureExtractor, f: FunctionHandle, bb: BBHandle
+) -> Tuple[FeatureSet, MatchResults, MatchResults]:
+    """
+    find matches for the given rules within the given basic block.
+
+    returns: tuple containing (features for basic block, match results for basic block, match results for instructions)
+    """
+    # all features found within this basic block,
+    # includes features found within instructions.
+    features: FeatureSet = collections.defaultdict(set)
+
+    # matches found at the instruction scope.
+    # might be found at different instructions, thats ok.
+    insn_matches: MatchResults = collections.defaultdict(list)
+
+    for insn in extractor.get_instructions(f, bb):
+        ifeatures, imatches = find_instruction_capabilities(ruleset, extractor, f, bb, insn)
+        for feature, vas in ifeatures.items():
+            features[feature].update(vas)
+
+        for rule_name, res in imatches.items():
+            insn_matches[rule_name].extend(res)
+
+    for feature, va in itertools.chain(
+        extractor.extract_basic_block_features(f, bb), extractor.extract_global_features()
+    ):
+        features[feature].add(va)
+
+    # matches found within this basic block.
+    _, matches = ruleset.match(Scope.BASIC_BLOCK, features, bb.address)
+
+    for rule_name, res in matches.items():
+        rule = ruleset[rule_name]
+        for va, _ in res:
+            capa.engine.index_rule_matches(features, rule, [va])
+
+    return features, matches, insn_matches
+
+
+def find_code_capabilities(
+    ruleset: RuleSet, extractor: StaticFeatureExtractor, fh: FunctionHandle
+) -> Tuple[MatchResults, MatchResults, MatchResults, int]:
+    """
+    find matches for the given rules within the given function.
+
+    returns: tuple containing (match results for function, match results for basic blocks, match results for instructions, number of features)
+    """
+    # all features found within this function,
+    # includes features found within basic blocks (and instructions).
+    function_features: FeatureSet = collections.defaultdict(set)
+
+    # matches found at the basic block scope.
+    # might be found at different basic blocks, thats ok.
+    bb_matches: MatchResults = collections.defaultdict(list)
+
+    # matches found at the instruction scope.
+    # might be found at different instructions, thats ok.
+    insn_matches: MatchResults = collections.defaultdict(list)
+
+    for bb in extractor.get_basic_blocks(fh):
+        features, bmatches, imatches = find_basic_block_capabilities(ruleset, extractor, fh, bb)
+        for feature, vas in features.items():
+            function_features[feature].update(vas)
+
+        for rule_name, res in bmatches.items():
+            bb_matches[rule_name].extend(res)
+
+        for rule_name, res in imatches.items():
+            insn_matches[rule_name].extend(res)
+
+    for feature, va in itertools.chain(extractor.extract_function_features(fh), extractor.extract_global_features()):
+        function_features[feature].add(va)
+
+    _, function_matches = ruleset.match(Scope.FUNCTION, function_features, fh.address)
+    return function_matches, bb_matches, insn_matches, len(function_features)
+
+
+def find_static_capabilities(
+    ruleset: RuleSet, extractor: StaticFeatureExtractor, disable_progress=None
+) -> Tuple[MatchResults, Any]:
+    all_function_matches: MatchResults = collections.defaultdict(list)
+    all_bb_matches: MatchResults = collections.defaultdict(list)
+    all_insn_matches: MatchResults = collections.defaultdict(list)
+
+    feature_counts = rdoc.StaticFeatureCounts(file=0, functions=())
+    library_functions: Tuple[rdoc.LibraryFunction, ...] = ()
+
+    assert isinstance(extractor, StaticFeatureExtractor)
+    with redirecting_print_to_tqdm(disable_progress):
+        with tqdm.contrib.logging.logging_redirect_tqdm():
+            pbar = tqdm.tqdm
+            if capa.helpers.is_runtime_ghidra():
+                # Ghidrathon interpreter cannot properly handle
+                # the TMonitor thread that is created via a monitor_interval
+                # > 0
+                pbar.monitor_interval = 0
+            if disable_progress:
+                # do not use tqdm to avoid unnecessary side effects when caller intends
+                # to disable progress completely
+                def pbar(s, *args, **kwargs):
+                    return s
+
+            functions = list(extractor.get_functions())
+            n_funcs = len(functions)
+
+            pb = pbar(functions, desc="matching", unit=" functions", postfix="skipped 0 library functions", leave=False)
+            for f in pb:
+                t0 = time.time()
+                if extractor.is_library_function(f.address):
+                    function_name = extractor.get_function_name(f.address)
+                    logger.debug("skipping library function 0x%x (%s)", f.address, function_name)
+                    library_functions += (
+                        rdoc.LibraryFunction(address=frz.Address.from_capa(f.address), name=function_name),
+                    )
+                    n_libs = len(library_functions)
+                    percentage = round(100 * (n_libs / n_funcs))
+                    if isinstance(pb, tqdm.tqdm):
+                        pb.set_postfix_str(f"skipped {n_libs} library functions ({percentage}%)")
+                    continue
+
+                function_matches, bb_matches, insn_matches, feature_count = find_code_capabilities(
+                    ruleset, extractor, f
+                )
+                feature_counts.functions += (
+                    rdoc.FunctionFeatureCount(address=frz.Address.from_capa(f.address), count=feature_count),
+                )
+                t1 = time.time()
+
+                match_count = sum(len(res) for res in function_matches.values())
+                match_count += sum(len(res) for res in bb_matches.values())
+                match_count += sum(len(res) for res in insn_matches.values())
+                logger.debug(
+                    "analyzed function 0x%x and extracted %d features, %d matches in %0.02fs",
+                    f.address,
+                    feature_count,
+                    match_count,
+                    t1 - t0,
+                )
+
+                for rule_name, res in function_matches.items():
+                    all_function_matches[rule_name].extend(res)
+                for rule_name, res in bb_matches.items():
+                    all_bb_matches[rule_name].extend(res)
+                for rule_name, res in insn_matches.items():
+                    all_insn_matches[rule_name].extend(res)
+
+    # collection of features that captures the rule matches within function, BB, and instruction scopes.
+    # mapping from feature (matched rule) to set of addresses at which it matched.
+    function_and_lower_features: FeatureSet = collections.defaultdict(set)
+    for rule_name, results in itertools.chain(
+        all_function_matches.items(), all_bb_matches.items(), all_insn_matches.items()
+    ):
+        locations = {p[0] for p in results}
+        rule = ruleset[rule_name]
+        capa.engine.index_rule_matches(function_and_lower_features, rule, locations)
+
+    all_file_matches, feature_count = find_file_capabilities(ruleset, extractor, function_and_lower_features)
+    feature_counts.file = feature_count
+
+    matches = dict(
+        itertools.chain(
+            # each rule exists in exactly one scope,
+            # so there won't be any overlap among these following MatchResults,
+            # and we can merge the dictionaries naively.
+            all_insn_matches.items(),
+            all_bb_matches.items(),
+            all_function_matches.items(),
+            all_file_matches.items(),
+        )
+    )
+
+    meta = {
+        "feature_counts": feature_counts,
+        "library_functions": library_functions,
+    }
+
+    return matches, meta

capa/engine.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 Mandiant, 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
@@ -8,7 +8,7 @@
 
 import copy
 import collections
-from typing import TYPE_CHECKING, Set, Dict, List, Tuple, Mapping, Iterable
+from typing import TYPE_CHECKING, Set, Dict, List, Tuple, Union, Mapping, Iterable, Iterator
 
 import capa.perf
 import capa.features.common
@@ -38,15 +38,17 @@ class Statement:
     """
 
     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)
@@ -60,21 +62,28 @@ class Statement:
         """
         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 getattr(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"):
-            children = getattr(self, "children")
+            children = self.children
             for i, child in enumerate(children):
                 if child is existing:
                     children[i] = new
@@ -90,7 +99,7 @@ class And(Statement):
     """
 
     def __init__(self, children, description=None):
-        super(And, self).__init__(description=description)
+        super().__init__(description=description)
         self.children = children
 
     def evaluate(self, ctx, short_circuit=True):
@@ -123,7 +132,7 @@ class Or(Statement):
     """
 
     def __init__(self, children, description=None):
-        super(Or, self).__init__(description=description)
+        super().__init__(description=description)
         self.children = children
 
     def evaluate(self, ctx, short_circuit=True):
@@ -150,7 +159,7 @@ 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, short_circuit=True):
@@ -172,7 +181,7 @@ class Some(Statement):
     """
 
     def __init__(self, count, children, description=None):
-        super(Some, self).__init__(description=description)
+        super().__init__(description=description)
         self.count = count
         self.children = children
 
@@ -208,7 +217,7 @@ 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)
@@ -225,9 +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):
@@ -237,7 +246,7 @@ class Subscope(Statement):
     """
 
     def __init__(self, scope, child, description=None):
-        super(Subscope, self).__init__(description=description)
+        super().__init__(description=description)
         self.scope = scope
         self.child = child
 
@@ -295,7 +304,7 @@ def match(rules: List["capa.rules.Rule"], features: FeatureSet, addr: Address) -
     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)  # type: MatchResults
+    results: MatchResults = collections.defaultdict(list)
 
     # copy features so that we can modify it
     # without affecting the caller (keep this function pure)

capa/exceptions.py

@@ -1,3 +1,10 @@
+# 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
 
@@ -12,3 +19,7 @@ class UnsupportedArchError(ValueError):
 
 class UnsupportedOSError(ValueError):
     pass
+
+
+class EmptyReportError(ValueError):
+    pass

capa/features/address.py

@@ -1,7 +1,12 @@
+# 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 dncil.clr.token import Token
-
 
 class Address(abc.ABC):
     @abc.abstractmethod
@@ -34,6 +39,82 @@ class AbsoluteVirtualAddress(int, Address):
     def __repr__(self):
         return f"absolute(0x{self:x})"
 
+    def __hash__(self):
+        return int.__hash__(self)
+
+
+class ProcessAddress(Address):
+    """an address of a process in a dynamic execution trace"""
+
+    def __init__(self, pid: int, ppid: int = 0):
+        assert ppid >= 0
+        assert pid > 0
+        self.ppid = ppid
+        self.pid = pid
+
+    def __repr__(self):
+        return "process(%s%s)" % (
+            f"ppid: {self.ppid}, " if self.ppid > 0 else "",
+            f"pid: {self.pid}",
+        )
+
+    def __hash__(self):
+        return hash((self.ppid, self.pid))
+
+    def __eq__(self, other):
+        assert isinstance(other, ProcessAddress)
+        return (self.ppid, self.pid) == (other.ppid, other.pid)
+
+    def __lt__(self, other):
+        assert isinstance(other, ProcessAddress)
+        return (self.ppid, self.pid) < (other.ppid, other.pid)
+
+
+class ThreadAddress(Address):
+    """addresses a thread in a dynamic execution trace"""
+
+    def __init__(self, process: ProcessAddress, tid: int):
+        assert tid >= 0
+        self.process = process
+        self.tid = tid
+
+    def __repr__(self):
+        return f"{self.process}, thread(tid: {self.tid})"
+
+    def __hash__(self):
+        return hash((self.process, self.tid))
+
+    def __eq__(self, other):
+        assert isinstance(other, ThreadAddress)
+        return (self.process, self.tid) == (other.process, other.tid)
+
+    def __lt__(self, other):
+        assert isinstance(other, ThreadAddress)
+        return (self.process, self.tid) < (other.process, other.tid)
+
+
+class DynamicCallAddress(Address):
+    """addesses a call in a dynamic execution trace"""
+
+    def __init__(self, thread: ThreadAddress, id: int):
+        assert id >= 0
+        self.thread = thread
+        self.id = id
+
+    def __repr__(self):
+        return f"{self.thread}, call(id: {self.id})"
+
+    def __hash__(self):
+        return hash((self.thread, self.id))
+
+    def __eq__(self, other):
+        assert isinstance(other, DynamicCallAddress)
+        return (self.thread, self.id) == (other.thread, other.id)
+
+    def __lt__(self, other):
+        assert isinstance(other, DynamicCallAddress)
+        return (self.thread, self.id) < (other.thread, other.id)
+
 
 class RelativeVirtualAddress(int, Address):
     """a memory address relative to a base address"""
@@ -41,6 +122,9 @@ class RelativeVirtualAddress(int, 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"""
@@ -52,52 +136,45 @@ class FileOffsetAddress(int, Address):
     def __repr__(self):
         return f"file(0x{self:x})"
 
+    def __hash__(self):
+        return int.__hash__(self)
 
-class DNTokenAddress(Address):
+
+class DNTokenAddress(int, Address):
     """a .NET token"""
 
-    def __init__(self, token: Token):
-        self.token = token
-
-    def __eq__(self, other):
-        return self.token.value == other.token.value
-
-    def __lt__(self, other):
-        return self.token.value < other.token.value
-
-    def __hash__(self):
-        return hash(self.token.value)
+    def __new__(cls, token: int):
+        return int.__new__(cls, token)
 
     def __repr__(self):
-        return f"token(0x{self.token.value:x})"
+        return f"token(0x{self:x})"
 
-    def __index__(self):
-        # returns the object converted to an integer
-        return self.token.value
+    def __hash__(self):
+        return int.__hash__(self)
 
 
 class DNTokenOffsetAddress(Address):
     """an offset into an object specified by a .NET token"""
 
-    def __init__(self, token: Token, offset: int):
+    def __init__(self, token: int, offset: int):
         assert offset >= 0
         self.token = token
         self.offset = offset
 
     def __eq__(self, other):
-        return (self.token.value, self.offset) == (other.token.value, other.offset)
+        return (self.token, self.offset) == (other.token, other.offset)
 
     def __lt__(self, other):
-        return (self.token.value, self.offset) < (other.token.value, other.offset)
+        return (self.token, self.offset) < (other.token, other.offset)
 
     def __hash__(self):
-        return hash((self.token.value, self.offset))
+        return hash((self.token, self.offset))
 
     def __repr__(self):
-        return f"token(0x{self.token.value:x})+(0x{self.offset:x})"
+        return f"token(0x{self.token:x})+(0x{self.offset:x})"
 
     def __index__(self):
-        return self.token.value + self.offset
+        return self.token + self.offset
 
 
 class _NoAddress(Address):

capa/features/basicblock.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 Mandiant, 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
@@ -11,7 +11,7 @@ from capa.features.common import Feature
 
 class BasicBlock(Feature):
     def __init__(self, description=None):
-        super(BasicBlock, self).__init__(None, description=description)
+        super().__init__(0, description=description)
 
     def __str__(self):
         return "basic block"

capa/features/com/__init__.py

@@ -0,0 +1,36 @@
+# 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 enum import Enum
+from typing import Dict, List
+
+from capa.helpers import assert_never
+
+
+class ComType(Enum):
+    CLASS = "class"
+    INTERFACE = "interface"
+
+
+COM_PREFIXES = {
+    ComType.CLASS: "CLSID_",
+    ComType.INTERFACE: "IID_",
+}
+
+
+def load_com_database(com_type: ComType) -> Dict[str, List[str]]:
+    # lazy load these python files since they are so large.
+    # that is, don't load them unless a COM feature is being handled.
+    import capa.features.com.classes
+    import capa.features.com.interfaces
+
+    if com_type == ComType.CLASS:
+        return capa.features.com.classes.COM_CLASSES
+    elif com_type == ComType.INTERFACE:
+        return capa.features.com.interfaces.COM_INTERFACES
+    else:
+        assert_never(com_type)

capa/features/common.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 Mandiant, 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
@@ -9,9 +9,10 @@
 import re
 import abc
 import codecs
+import typing
 import logging
 import collections
-from typing import TYPE_CHECKING, Set, Dict, List, Union, Optional, Sequence
+from typing import TYPE_CHECKING, Set, Dict, List, Union, Optional
 
 if TYPE_CHECKING:
     # circular import, otherwise
@@ -29,6 +30,14 @@ MAX_BYTES_FEATURE_SIZE = 0x100
 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"))
 
@@ -73,7 +82,7 @@ class Result:
         children: List["Result"],
         locations: Optional[Set[Address]] = None,
     ):
-        super(Result, self).__init__()
+        super().__init__()
         self.success = success
         self.statement = statement
         self.children = children
@@ -91,16 +100,23 @@ class Result:
         return self.success
 
 
-class Feature(abc.ABC):
-    def __init__(self, value: Union[str, int, float, bytes], description=None):
+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(Feature, self).__init__()
-        self.name = self.__class__.__name__.lower()
+        super().__init__()
 
+        self.name = self.__class__.__name__.lower()
         self.value = value
         self.description = description
 
@@ -111,31 +127,41 @@ class Feature(abc.ABC):
         return self.name == other.name and self.value == other.value
 
     def __lt__(self, other):
-        # TODO: this is a huge hack!
+        # 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).json()
-            < capa.features.freeze.features.feature_from_capa(other).json()
+            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.
-
-        Returns: any
         """
         return str(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)
+                return f"{self.get_name_str()}({self.get_value_str()} = {self.description})"
             else:
-                return "%s(%s)" % (self.name, self.get_value_str())
+                return f"{self.get_name_str()}({self.get_value_str()})"
         else:
-            return "%s" % self.name
+            return f"{self.get_name_str()}"
 
     def __repr__(self):
         return str(self)
@@ -148,33 +174,37 @@ class Feature(abc.ABC):
 
 class MatchedRule(Feature):
     def __init__(self, value: str, description=None):
-        super(MatchedRule, self).__init__(value, description=description)
+        super().__init__(value, description=description)
         self.name = "match"
 
 
 class Characteristic(Feature):
     def __init__(self, value: str, description=None):
-        super(Characteristic, self).__init__(value, description=description)
+        super().__init__(value, description=description)
 
 
 class String(Feature):
     def __init__(self, value: str, description=None):
-        super(String, self).__init__(value, description=description)
+        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(Class, self).__init__(value, description=description)
+        super().__init__(value, description=description)
 
 
 class Namespace(Feature):
     def __init__(self, value: str, description=None):
-        super(Namespace, self).__init__(value, description=description)
+        super().__init__(value, description=description)
 
 
 class Substring(String):
     def __init__(self, value: str, description=None):
-        super(Substring, self).__init__(value, description=description)
+        super().__init__(value, description=description)
         self.value = value
 
     def evaluate(self, ctx, short_circuit=True):
@@ -183,8 +213,9 @@ class Substring(String):
 
         # mapping from string value to list of locations.
         # will unique the locations later on.
-        matches = collections.defaultdict(list)
+        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
@@ -194,32 +225,32 @@ class Substring(String):
                 raise ValueError("unexpected feature value type")
 
             if self.value in feature.value:
-                matches[feature.value].extend(locations)
+                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:
-            # finalize: defaultdict -> dict
-            # which makes json serialization easier
-            matches = dict(matches)
-
             # collect all locations
             locations = set()
-            for s in matches.keys():
-                matches[s] = list(set(matches[s]))
-                locations.update(matches[s])
+            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, matches), [], locations=locations)
+            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):
-        return "substring(%s)" % self.value
+        assert isinstance(self.value, str)
+        return f"substring({escape_string(self.value)})"
 
 
 class _MatchedSubstring(Substring):
@@ -236,7 +267,7 @@ class _MatchedSubstring(Substring):
           substring: the substring feature that matches.
           match: mapping from matching string to its locations.
         """
-        super(_MatchedSubstring, self).__init__(str(substring.value), description=substring.description)
+        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"
@@ -244,15 +275,14 @@ class _MatchedSubstring(Substring):
         self.matches = matches
 
     def __str__(self):
-        return 'substring("%s", matches = %s)' % (
-            self.value,
-            ", ".join(map(lambda s: '"' + s + '"', (self.matches or {}).keys())),
-        )
+        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(Regex, self).__init__(value, description=description)
+        super().__init__(value, description=description)
         self.value = value
 
         pat = self.value[len("/") : -len("/")]
@@ -262,12 +292,12 @@ class Regex(String):
             flags |= re.IGNORECASE
         try:
             self.re = re.compile(pat, flags)
-        except re.error:
+        except re.error as exc:
             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
-            )
+                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
@@ -275,7 +305,7 @@ class Regex(String):
 
         # mapping from string value to list of locations.
         # will unique the locations later on.
-        matches = collections.defaultdict(list)
+        matches: typing.DefaultDict[str, Set[Address]] = collections.defaultdict(set)
 
         for feature, locations in ctx.items():
             if not isinstance(feature, (String,)):
@@ -290,33 +320,29 @@ class Regex(String):
             # 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].extend(locations)
+                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:
-            # finalize: defaultdict -> dict
-            # which makes json serialization easier
-            matches = dict(matches)
-
             # collect all locations
             locations = set()
-            for s in matches.keys():
-                matches[s] = list(set(matches[s]))
-                locations.update(matches[s])
+            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, matches), [], locations=locations)
+            return Result(True, _MatchedRegex(self, dict(matches)), [], locations=locations)
         else:
             return Result(False, _MatchedRegex(self, {}), [])
 
     def __str__(self):
-        return "regex(string =~ %s)" % self.value
+        assert isinstance(self.value, str)
+        return f"regex(string =~ {self.value})"
 
 
 class _MatchedRegex(Regex):
@@ -333,7 +359,7 @@ class _MatchedRegex(Regex):
           regex: the regex feature that matches.
           matches: mapping from matching string to its locations.
         """
-        super(_MatchedRegex, self).__init__(str(regex.value), description=regex.description)
+        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"
@@ -341,10 +367,9 @@ class _MatchedRegex(Regex):
         self.matches = matches
 
     def __str__(self):
-        return "regex(string =~ %s, matches = %s)" % (
-            self.value,
-            ", ".join(map(lambda s: '"' + s + '"', (self.matches or {}).keys())),
-        )
+        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:
@@ -356,23 +381,26 @@ class StringFactory:
 
 class Bytes(Feature):
     def __init__(self, value: bytes, description=None):
-        super(Bytes, self).__init__(value, description=description)
+        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))
 
 
@@ -386,7 +414,7 @@ VALID_ARCH = (ARCH_I386, ARCH_AMD64, ARCH_ANY)
 
 class Arch(Feature):
     def __init__(self, value: str, description=None):
-        super(Arch, self).__init__(value, description=description)
+        super().__init__(value, description=description)
         self.name = "arch"
 
 
@@ -397,13 +425,29 @@ OS_MACOS = "macos"
 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(OS, self).__init__(value, description=description)
+        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"
@@ -413,13 +457,25 @@ VALID_FORMAT = (FORMAT_PE, FORMAT_ELF, FORMAT_DOTNET)
 FORMAT_AUTO = "auto"
 FORMAT_SC32 = "sc32"
 FORMAT_SC64 = "sc64"
+FORMAT_CAPE = "cape"
+STATIC_FORMATS = {
+    FORMAT_SC32,
+    FORMAT_SC64,
+    FORMAT_PE,
+    FORMAT_ELF,
+    FORMAT_DOTNET,
+}
+DYNAMIC_FORMATS = {
+    FORMAT_CAPE,
+}
 FORMAT_FREEZE = "freeze"
+FORMAT_RESULT = "result"
 FORMAT_UNKNOWN = "unknown"
 
 
 class Format(Feature):
     def __init__(self, value: str, description=None):
-        super(Format, self).__init__(value, description=description)
+        super().__init__(value, description=description)
         self.name = "format"
 
 

capa/features/extractors/base_extractor.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 Mandiant, 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
@@ -7,13 +7,18 @@
 # See the License for the specific language governing permissions and limitations under the License.
 
 import abc
+import hashlib
 import dataclasses
 from typing import Any, Dict, Tuple, Union, Iterator
 from dataclasses import dataclass
 
+# TODO(williballenthin): use typing.TypeAlias directly when Python 3.9 is deprecated
+# https://github.com/mandiant/capa/issues/1699
+from typing_extensions import TypeAlias
+
 import capa.features.address
 from capa.features.common import Feature
-from capa.features.address import Address, AbsoluteVirtualAddress
+from capa.features.address import Address, ThreadAddress, ProcessAddress, DynamicCallAddress, 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.
@@ -22,6 +27,24 @@ from capa.features.address import Address, AbsoluteVirtualAddress
 # the feature extractor from which they were created.
 
 
+@dataclass
+class SampleHashes:
+    md5: str
+    sha1: str
+    sha256: str
+
+    @classmethod
+    def from_bytes(cls, buf: bytes) -> "SampleHashes":
+        md5 = hashlib.md5()
+        sha1 = hashlib.sha1()
+        sha256 = hashlib.sha256()
+        md5.update(buf)
+        sha1.update(buf)
+        sha256.update(buf)
+
+        return cls(md5=md5.hexdigest(), sha1=sha1.hexdigest(), sha256=sha256.hexdigest())
+
+
 @dataclass
 class FunctionHandle:
     """reference to a function recognized by a feature extractor.
@@ -63,16 +86,18 @@ class InsnHandle:
     inner: Any
 
 
-class FeatureExtractor:
+class StaticFeatureExtractor:
     """
-    FeatureExtractor defines the interface for fetching features from a sample.
+    StaticFeatureExtractor defines the interface for fetching features from a
+    sample without running it; extractors that rely on the execution trace of
+    a sample must implement the other sibling class, DynamicFeatureExtracor.
 
     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
+    Therefore, we can define a StaticFeatureExtractor 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.
 
@@ -81,13 +106,14 @@ class FeatureExtractor:
 
     __metaclass__ = abc.ABCMeta
 
-    def __init__(self):
+    def __init__(self, hashes: SampleHashes):
         #
         # 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__()
+        super().__init__()
+        self._sample_hashes = hashes
 
     @abc.abstractmethod
     def get_base_address(self) -> Union[AbsoluteVirtualAddress, capa.features.address._NoAddress]:
@@ -100,6 +126,12 @@ class FeatureExtractor:
         """
         raise NotImplementedError()
 
+    def get_sample_hashes(self) -> SampleHashes:
+        """
+        fetch the hashes for the sample contained within the extractor.
+        """
+        return self._sample_hashes
+
     @abc.abstractmethod
     def extract_global_features(self) -> Iterator[Tuple[Feature, Address]]:
         """
@@ -262,3 +294,177 @@ class FeatureExtractor:
           Tuple[Feature, Address]: feature and its location
         """
         raise NotImplementedError()
+
+
+@dataclass
+class ProcessHandle:
+    """
+    reference to a process extracted by the sandbox.
+
+    Attributes:
+        address: process's address (pid)
+        inner: sandbox-specific data
+    """
+
+    address: ProcessAddress
+    inner: Any
+
+
+@dataclass
+class ThreadHandle:
+    """
+    reference to a thread extracted by the sandbox.
+
+    Attributes:
+        address: thread's address (tid)
+        inner: sandbox-specific data
+    """
+
+    address: ThreadAddress
+    inner: Any
+
+
+@dataclass
+class CallHandle:
+    """
+    reference to an api call extracted by the sandbox.
+
+    Attributes:
+        address: call's address, such as event index or id
+        inner: sandbox-specific data
+    """
+
+    address: DynamicCallAddress
+    inner: Any
+
+
+class DynamicFeatureExtractor:
+    """
+    DynamicFeatureExtractor defines the interface for fetching features from a
+    sandbox' analysis of a sample; extractors that rely on statically analyzing
+    a sample must implement the sibling extractor, StaticFeatureExtractor.
+
+    Features are grouped mainly into threads that alongside their meta-features are also grouped into
+    processes (that also have their own features). Other scopes (such as function and file) may also apply
+    for a specific sandbox.
+
+    This class is not instantiated directly; it is the base class for other implementations.
+    """
+
+    __metaclass__ = abc.ABCMeta
+
+    def __init__(self, hashes: SampleHashes):
+        #
+        # 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__()
+        self._sample_hashes = hashes
+
+    def get_sample_hashes(self) -> SampleHashes:
+        """
+        fetch the hashes for the sample contained within the extractor.
+        """
+        return self._sample_hashes
+
+    @abc.abstractmethod
+    def extract_global_features(self) -> Iterator[Tuple[Feature, Address]]:
+        """
+        extract features found at every scope ("global").
+
+        example::
+
+            extractor = CapeFeatureExtractor.from_report(json.loads(buf))
+            for feature, addr in extractor.get_global_features():
+                print(addr, 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 = CapeFeatureExtractor.from_report(json.loads(buf))
+            for feature, addr in extractor.get_file_features():
+                print(addr, feature)
+
+        yields:
+          Tuple[Feature, Address]: feature and its location
+        """
+        raise NotImplementedError()
+
+    @abc.abstractmethod
+    def get_processes(self) -> Iterator[ProcessHandle]:
+        """
+        Enumerate processes in the trace.
+        """
+        raise NotImplementedError()
+
+    @abc.abstractmethod
+    def extract_process_features(self, ph: ProcessHandle) -> Iterator[Tuple[Feature, Address]]:
+        """
+        Yields all the features of a process. These include:
+        - file features of the process' image
+        """
+        raise NotImplementedError()
+
+    @abc.abstractmethod
+    def get_process_name(self, ph: ProcessHandle) -> str:
+        """
+        Returns the human-readable name for the given process,
+        such as the filename.
+        """
+        raise NotImplementedError()
+
+    @abc.abstractmethod
+    def get_threads(self, ph: ProcessHandle) -> Iterator[ThreadHandle]:
+        """
+        Enumerate threads in the given process.
+        """
+        raise NotImplementedError()
+
+    @abc.abstractmethod
+    def extract_thread_features(self, ph: ProcessHandle, th: ThreadHandle) -> Iterator[Tuple[Feature, Address]]:
+        """
+        Yields all the features of a thread. These include:
+        - sequenced api traces
+        """
+        raise NotImplementedError()
+
+    @abc.abstractmethod
+    def get_calls(self, ph: ProcessHandle, th: ThreadHandle) -> Iterator[CallHandle]:
+        """
+        Enumerate calls in the given thread
+        """
+        raise NotImplementedError()
+
+    @abc.abstractmethod
+    def extract_call_features(
+        self, ph: ProcessHandle, th: ThreadHandle, ch: CallHandle
+    ) -> Iterator[Tuple[Feature, Address]]:
+        """
+        Yields all features of a call. These include:
+        - api name
+        - bytes/strings/numbers extracted from arguments
+        """
+        raise NotImplementedError()
+
+    @abc.abstractmethod
+    def get_call_name(self, ph: ProcessHandle, th: ThreadHandle, ch: CallHandle) -> str:
+        """
+        Returns the human-readable name for the given call,
+        such as as rendered API log entry, like:
+
+            Foo(1, "two", b"\x00\x11") -> -1
+        """
+        raise NotImplementedError()
+
+
+FeatureExtractor: TypeAlias = Union[StaticFeatureExtractor, DynamicFeatureExtractor]

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 +1,81 @@
+# 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,
+    SampleHashes,
+    FunctionHandle,
+    StaticFeatureExtractor,
+)
+
+
+class BinjaFeatureExtractor(StaticFeatureExtractor):
+    def __init__(self, bv: binja.BinaryView):
+        super().__init__(hashes=SampleHashes.from_bytes(bv.file.raw.read(0, len(bv.file.raw))))
+        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, include_dll=True):
+            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, include_dll=True):
+                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/cape/call.py

@@ -0,0 +1,62 @@
+# 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.helpers import assert_never
+from capa.features.insn import API, Number
+from capa.features.common import String, Feature
+from capa.features.address import Address
+from capa.features.extractors.cape.models import Call
+from capa.features.extractors.base_extractor import CallHandle, ThreadHandle, ProcessHandle
+
+logger = logging.getLogger(__name__)
+
+
+def extract_call_features(ph: ProcessHandle, th: ThreadHandle, ch: CallHandle) -> Iterator[Tuple[Feature, Address]]:
+    """
+    this method extracts the given call's features (such as API name and arguments),
+    and returns them as API, Number, and String features.
+
+    args:
+      ph: process handle (for defining the extraction scope)
+      th: thread handle (for defining the extraction scope)
+      ch: call handle (for defining the extraction scope)
+
+    yields:
+      Feature, address; where Feature is either: API, Number, or String.
+    """
+    call: Call = ch.inner
+
+    # list similar to disassembly: arguments right-to-left, call
+    for arg in reversed(call.arguments):
+        value = arg.value
+        if isinstance(value, list) and len(value) == 0:
+            # unsure why CAPE captures arguments as empty lists?
+            continue
+
+        elif isinstance(value, str):
+            yield String(value), ch.address
+
+        elif isinstance(value, int):
+            yield Number(value), ch.address
+
+        else:
+            assert_never(value)
+
+    yield API(call.api), ch.address
+
+
+def extract_features(ph: ProcessHandle, th: ThreadHandle, ch: CallHandle) -> Iterator[Tuple[Feature, Address]]:
+    for handler in CALL_HANDLERS:
+        for feature, addr in handler(ph, th, ch):
+            yield feature, addr
+
+
+CALL_HANDLERS = (extract_call_features,)

capa/features/extractors/cape/extractor.py

@@ -0,0 +1,145 @@
+# 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 Dict, Tuple, Union, Iterator
+
+import capa.features.extractors.cape.call
+import capa.features.extractors.cape.file
+import capa.features.extractors.cape.thread
+import capa.features.extractors.cape.global_
+import capa.features.extractors.cape.process
+from capa.exceptions import EmptyReportError, UnsupportedFormatError
+from capa.features.common import Feature, Characteristic
+from capa.features.address import NO_ADDRESS, Address, AbsoluteVirtualAddress, _NoAddress
+from capa.features.extractors.cape.models import Call, Static, Process, CapeReport
+from capa.features.extractors.base_extractor import (
+    CallHandle,
+    SampleHashes,
+    ThreadHandle,
+    ProcessHandle,
+    DynamicFeatureExtractor,
+)
+
+logger = logging.getLogger(__name__)
+
+TESTED_VERSIONS = {"2.2-CAPE", "2.4-CAPE"}
+
+
+class CapeExtractor(DynamicFeatureExtractor):
+    def __init__(self, report: CapeReport):
+        super().__init__(
+            hashes=SampleHashes(
+                md5=report.target.file.md5.lower(),
+                sha1=report.target.file.sha1.lower(),
+                sha256=report.target.file.sha256.lower(),
+            )
+        )
+        self.report: CapeReport = report
+
+        # pre-compute these because we'll yield them at *every* scope.
+        self.global_features = list(capa.features.extractors.cape.global_.extract_features(self.report))
+
+    def get_base_address(self) -> Union[AbsoluteVirtualAddress, _NoAddress, None]:
+        # value according to the PE header, the actual trace may use a different imagebase
+        assert self.report.static is not None and self.report.static.pe is not None
+        return AbsoluteVirtualAddress(self.report.static.pe.imagebase)
+
+    def extract_global_features(self) -> Iterator[Tuple[Feature, Address]]:
+        yield from self.global_features
+
+    def extract_file_features(self) -> Iterator[Tuple[Feature, Address]]:
+        yield from capa.features.extractors.cape.file.extract_features(self.report)
+
+    def get_processes(self) -> Iterator[ProcessHandle]:
+        yield from capa.features.extractors.cape.file.get_processes(self.report)
+
+    def extract_process_features(self, ph: ProcessHandle) -> Iterator[Tuple[Feature, Address]]:
+        yield from capa.features.extractors.cape.process.extract_features(ph)
+
+    def get_process_name(self, ph) -> str:
+        process: Process = ph.inner
+        return process.process_name
+
+    def get_threads(self, ph: ProcessHandle) -> Iterator[ThreadHandle]:
+        yield from capa.features.extractors.cape.process.get_threads(ph)
+
+    def extract_thread_features(self, ph: ProcessHandle, th: ThreadHandle) -> Iterator[Tuple[Feature, Address]]:
+        if False:
+            # force this routine to be a generator,
+            # but we don't actually have any elements to generate.
+            yield Characteristic("never"), NO_ADDRESS
+        return
+
+    def get_calls(self, ph: ProcessHandle, th: ThreadHandle) -> Iterator[CallHandle]:
+        yield from capa.features.extractors.cape.thread.get_calls(ph, th)
+
+    def extract_call_features(
+        self, ph: ProcessHandle, th: ThreadHandle, ch: CallHandle
+    ) -> Iterator[Tuple[Feature, Address]]:
+        yield from capa.features.extractors.cape.call.extract_features(ph, th, ch)
+
+    def get_call_name(self, ph, th, ch) -> str:
+        call: Call = ch.inner
+
+        parts = []
+        parts.append(call.api)
+        parts.append("(")
+        for argument in call.arguments:
+            parts.append(argument.name)
+            parts.append("=")
+
+            if argument.pretty_value:
+                parts.append(argument.pretty_value)
+            else:
+                if isinstance(argument.value, int):
+                    parts.append(hex(argument.value))
+                elif isinstance(argument.value, str):
+                    parts.append('"')
+                    parts.append(argument.value)
+                    parts.append('"')
+                elif isinstance(argument.value, list):
+                    pass
+                else:
+                    capa.helpers.assert_never(argument.value)
+
+            parts.append(", ")
+        if call.arguments:
+            # remove the trailing comma
+            parts.pop()
+        parts.append(")")
+        parts.append(" -> ")
+        if call.pretty_return:
+            parts.append(call.pretty_return)
+        else:
+            parts.append(hex(call.return_))
+
+        return "".join(parts)
+
+    @classmethod
+    def from_report(cls, report: Dict) -> "CapeExtractor":
+        cr = CapeReport.model_validate(report)
+
+        if cr.info.version not in TESTED_VERSIONS:
+            logger.warning("CAPE version '%s' not tested/supported yet", cr.info.version)
+
+        # observed in 2.4-CAPE reports from capesandbox.com
+        if cr.static is None and cr.target.file.pe is not None:
+            cr.static = Static()
+            cr.static.pe = cr.target.file.pe
+
+        if cr.static is None:
+            raise UnsupportedFormatError("CAPE report missing static analysis")
+
+        if cr.static.pe is None:
+            raise UnsupportedFormatError("CAPE report missing PE analysis")
+
+        if len(cr.behavior.processes) == 0:
+            raise EmptyReportError("CAPE did not capture any processes")
+
+        return cls(cr)

capa/features/extractors/cape/file.py

@@ -0,0 +1,132 @@
+# 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.file import Export, Import, Section
+from capa.features.common import String, Feature
+from capa.features.address import NO_ADDRESS, Address, ProcessAddress, AbsoluteVirtualAddress
+from capa.features.extractors.helpers import generate_symbols
+from capa.features.extractors.cape.models import CapeReport
+from capa.features.extractors.base_extractor import ProcessHandle
+
+logger = logging.getLogger(__name__)
+
+
+def get_processes(report: CapeReport) -> Iterator[ProcessHandle]:
+    """
+    get all the created processes for a sample
+    """
+    seen_processes = {}
+    for process in report.behavior.processes:
+        addr = ProcessAddress(pid=process.process_id, ppid=process.parent_id)
+        yield ProcessHandle(address=addr, inner=process)
+
+        # check for pid and ppid reuse
+        if addr not in seen_processes:
+            seen_processes[addr] = [process]
+        else:
+            logger.warning(
+                "pid and ppid reuse detected between process %s and process%s: %s",
+                process,
+                "es" if len(seen_processes[addr]) > 1 else "",
+                seen_processes[addr],
+            )
+            seen_processes[addr].append(process)
+
+
+def extract_import_names(report: CapeReport) -> Iterator[Tuple[Feature, Address]]:
+    """
+    extract imported function names
+    """
+    assert report.static is not None and report.static.pe is not None
+    imports = report.static.pe.imports
+
+    if isinstance(imports, dict):
+        imports = list(imports.values())
+
+    assert isinstance(imports, list)
+
+    for library in imports:
+        for function in library.imports:
+            if not function.name:
+                continue
+
+            for name in generate_symbols(library.dll, function.name, include_dll=True):
+                yield Import(name), AbsoluteVirtualAddress(function.address)
+
+
+def extract_export_names(report: CapeReport) -> Iterator[Tuple[Feature, Address]]:
+    assert report.static is not None and report.static.pe is not None
+    for function in report.static.pe.exports:
+        yield Export(function.name), AbsoluteVirtualAddress(function.address)
+
+
+def extract_section_names(report: CapeReport) -> Iterator[Tuple[Feature, Address]]:
+    assert report.static is not None and report.static.pe is not None
+    for section in report.static.pe.sections:
+        yield Section(section.name), AbsoluteVirtualAddress(section.virtual_address)
+
+
+def extract_file_strings(report: CapeReport) -> Iterator[Tuple[Feature, Address]]:
+    if report.strings is not None:
+        for string in report.strings:
+            yield String(string), NO_ADDRESS
+
+
+def extract_used_regkeys(report: CapeReport) -> Iterator[Tuple[Feature, Address]]:
+    for regkey in report.behavior.summary.keys:
+        yield String(regkey), NO_ADDRESS
+
+
+def extract_used_files(report: CapeReport) -> Iterator[Tuple[Feature, Address]]:
+    for file in report.behavior.summary.files:
+        yield String(file), NO_ADDRESS
+
+
+def extract_used_mutexes(report: CapeReport) -> Iterator[Tuple[Feature, Address]]:
+    for mutex in report.behavior.summary.mutexes:
+        yield String(mutex), NO_ADDRESS
+
+
+def extract_used_commands(report: CapeReport) -> Iterator[Tuple[Feature, Address]]:
+    for cmd in report.behavior.summary.executed_commands:
+        yield String(cmd), NO_ADDRESS
+
+
+def extract_used_apis(report: CapeReport) -> Iterator[Tuple[Feature, Address]]:
+    for symbol in report.behavior.summary.resolved_apis:
+        yield String(symbol), NO_ADDRESS
+
+
+def extract_used_services(report: CapeReport) -> Iterator[Tuple[Feature, Address]]:
+    for svc in report.behavior.summary.created_services:
+        yield String(svc), NO_ADDRESS
+    for svc in report.behavior.summary.started_services:
+        yield String(svc), NO_ADDRESS
+
+
+def extract_features(report: CapeReport) -> Iterator[Tuple[Feature, Address]]:
+    for handler in FILE_HANDLERS:
+        for feature, addr in handler(report):
+            yield feature, addr
+
+
+FILE_HANDLERS = (
+    extract_import_names,
+    extract_export_names,
+    extract_section_names,
+    extract_file_strings,
+    extract_used_regkeys,
+    extract_used_files,
+    extract_used_mutexes,
+    extract_used_commands,
+    extract_used_apis,
+    extract_used_services,
+)

capa/features/extractors/cape/global_.py

@@ -0,0 +1,93 @@
+# 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 (
+    OS,
+    OS_ANY,
+    OS_LINUX,
+    ARCH_I386,
+    FORMAT_PE,
+    ARCH_AMD64,
+    FORMAT_ELF,
+    OS_WINDOWS,
+    Arch,
+    Format,
+    Feature,
+)
+from capa.features.address import NO_ADDRESS, Address
+from capa.features.extractors.cape.models import CapeReport
+
+logger = logging.getLogger(__name__)
+
+
+def extract_arch(report: CapeReport) -> Iterator[Tuple[Feature, Address]]:
+    if "Intel 80386" in report.target.file.type:
+        yield Arch(ARCH_I386), NO_ADDRESS
+    elif "x86-64" in report.target.file.type:
+        yield Arch(ARCH_AMD64), NO_ADDRESS
+    else:
+        logger.warning("unrecognized Architecture: %s", report.target.file.type)
+        raise ValueError(
+            f"unrecognized Architecture from the CAPE report; output of file command: {report.target.file.type}"
+        )
+
+
+def extract_format(report: CapeReport) -> Iterator[Tuple[Feature, Address]]:
+    if "PE" in report.target.file.type:
+        yield Format(FORMAT_PE), NO_ADDRESS
+    elif "ELF" in report.target.file.type:
+        yield Format(FORMAT_ELF), NO_ADDRESS
+    else:
+        logger.warning("unknown file format, file command output: %s", report.target.file.type)
+        raise ValueError(
+            "unrecognized file format from the CAPE report; output of file command: {report.target.file.type}"
+        )
+
+
+def extract_os(report: CapeReport) -> Iterator[Tuple[Feature, Address]]:
+    # this variable contains the output of the file command
+    file_output = report.target.file.type
+
+    if "windows" in file_output.lower():
+        yield OS(OS_WINDOWS), NO_ADDRESS
+    elif "elf" in file_output.lower():
+        # operating systems recognized by the file command: https://github.com/file/file/blob/master/src/readelf.c#L609
+        if "Linux" in file_output:
+            yield OS(OS_LINUX), NO_ADDRESS
+        elif "Hurd" in file_output:
+            yield OS("hurd"), NO_ADDRESS
+        elif "Solaris" in file_output:
+            yield OS("solaris"), NO_ADDRESS
+        elif "kFreeBSD" in file_output:
+            yield OS("freebsd"), NO_ADDRESS
+        elif "kNetBSD" in file_output:
+            yield OS("netbsd"), NO_ADDRESS
+        else:
+            # if the operating system information is missing from the cape report, it's likely a bug
+            logger.warning("unrecognized OS: %s", file_output)
+            raise ValueError("unrecognized OS from the CAPE report; output of file command: {file_output}")
+    else:
+        # the sample is shellcode
+        logger.debug("unsupported file format, file command output: %s", file_output)
+        yield OS(OS_ANY), NO_ADDRESS
+
+
+def extract_features(report: CapeReport) -> Iterator[Tuple[Feature, Address]]:
+    for global_handler in GLOBAL_HANDLER:
+        for feature, addr in global_handler(report):
+            yield feature, addr
+
+
+GLOBAL_HANDLER = (
+    extract_format,
+    extract_os,
+    extract_arch,
+)

capa/features/extractors/cape/helpers.py

@@ -0,0 +1,29 @@
+# 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, List
+
+from capa.features.extractors.base_extractor import ProcessHandle
+
+
+def find_process(processes: List[Dict[str, Any]], ph: ProcessHandle) -> Dict[str, Any]:
+    """
+    find a specific process identified by a process handler.
+
+    args:
+      processes: a list of processes extracted by CAPE
+      ph: handle of the sought process
+
+    return:
+      a CAPE-defined dictionary for the sought process' information
+    """
+
+    for process in processes:
+        if ph.address.ppid == process["parent_id"] and ph.address.pid == process["process_id"]:
+            return process
+    return {}

capa/features/extractors/cape/models.py

@@ -0,0 +1,446 @@
+# 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 Any, Dict, List, Union, Literal, Optional
+
+from pydantic import Field, BaseModel, ConfigDict
+from typing_extensions import Annotated, TypeAlias
+from pydantic.functional_validators import BeforeValidator
+
+
+def validate_hex_int(value):
+    if isinstance(value, str):
+        return int(value, 16) if value.startswith("0x") else int(value, 10)
+    else:
+        return value
+
+
+def validate_hex_bytes(value):
+    return binascii.unhexlify(value) if isinstance(value, str) else value
+
+
+HexInt = Annotated[int, BeforeValidator(validate_hex_int)]
+HexBytes = Annotated[bytes, BeforeValidator(validate_hex_bytes)]
+
+
+# a model that *cannot* have extra fields
+# if they do, pydantic raises an exception.
+# use this for models we rely upon and cannot change.
+#
+# for things that may be extended and we don't care,
+# use FlexibleModel.
+class ExactModel(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+
+
+# a model that can have extra fields that we ignore.
+# use this if we don't want to raise an exception for extra
+# data fields that we didn't expect.
+class FlexibleModel(BaseModel):
+    pass
+
+
+# use this type to indicate that we won't model this data.
+# because its not relevant to our use in capa.
+#
+# while its nice to have full coverage of the data shape,
+# it can easily change and break our parsing.
+# so we really only want to describe what we'll use.
+Skip: TypeAlias = Optional[Any]
+
+
+# mark fields that we haven't seen yet and need to model.
+# pydantic should raise an error when encountering data
+# in a field with this type.
+# then we can update the model with the discovered shape.
+TODO: TypeAlias = None
+ListTODO: TypeAlias = List[None]
+DictTODO: TypeAlias = ExactModel
+
+EmptyDict: TypeAlias = BaseModel
+EmptyList: TypeAlias = List[Any]
+
+
+class Info(FlexibleModel):
+    version: str
+
+
+class ImportedSymbol(ExactModel):
+    address: HexInt
+    name: Optional[str] = None
+
+
+class ImportedDll(ExactModel):
+    dll: str
+    imports: List[ImportedSymbol]
+
+
+class DirectoryEntry(ExactModel):
+    name: str
+    virtual_address: HexInt
+    size: HexInt
+
+
+class Section(ExactModel):
+    name: str
+    raw_address: HexInt
+    virtual_address: HexInt
+    virtual_size: HexInt
+    size_of_data: HexInt
+    characteristics: str
+    characteristics_raw: HexInt
+    entropy: float
+
+
+class Resource(ExactModel):
+    name: str
+    language: Optional[str] = None
+    sublanguage: str
+    filetype: Optional[str]
+    offset: HexInt
+    size: HexInt
+    entropy: float
+
+
+class DigitalSigner(FlexibleModel):
+    md5_fingerprint: str
+    not_after: str
+    not_before: str
+    serial_number: str
+    sha1_fingerprint: str
+    sha256_fingerprint: str
+
+    issuer_commonName: Optional[str] = None
+    issuer_countryName: Optional[str] = None
+    issuer_localityName: Optional[str] = None
+    issuer_organizationName: Optional[str] = None
+    issuer_stateOrProvinceName: Optional[str] = None
+
+    subject_commonName: Optional[str] = None
+    subject_countryName: Optional[str] = None
+    subject_localityName: Optional[str] = None
+    subject_organizationName: Optional[str] = None
+    subject_stateOrProvinceName: Optional[str] = None
+
+    extensions_authorityInfoAccess_caIssuers: Optional[str] = None
+    extensions_authorityKeyIdentifier: Optional[str] = None
+    extensions_cRLDistributionPoints_0: Optional[str] = None
+    extensions_certificatePolicies_0: Optional[str] = None
+    extensions_subjectAltName_0: Optional[str] = None
+    extensions_subjectKeyIdentifier: Optional[str] = None
+
+
+class AuxSigner(ExactModel):
+    name: str
+    issued_to: str = Field(alias="Issued to")
+    issued_by: str = Field(alias="Issued by")
+    expires: str = Field(alias="Expires")
+    sha1_hash: str = Field(alias="SHA1 hash")
+
+
+class Signer(ExactModel):
+    aux_sha1: Optional[str] = None
+    aux_timestamp: Optional[str] = None
+    aux_valid: Optional[bool] = None
+    aux_error: Optional[bool] = None
+    aux_error_desc: Optional[str] = None
+    aux_signers: Optional[List[AuxSigner]] = None
+
+
+class Overlay(ExactModel):
+    offset: HexInt
+    size: HexInt
+
+
+class KV(ExactModel):
+    name: str
+    value: str
+
+
+class ExportedSymbol(ExactModel):
+    address: HexInt
+    name: str
+    ordinal: int
+
+
+class PE(ExactModel):
+    peid_signatures: TODO
+    imagebase: HexInt
+    entrypoint: HexInt
+    reported_checksum: HexInt
+    actual_checksum: HexInt
+    osversion: str
+    pdbpath: Optional[str] = None
+    timestamp: str
+
+    # List[ImportedDll], or Dict[basename(dll), ImportedDll]
+    imports: Union[List[ImportedDll], Dict[str, ImportedDll]]
+    imported_dll_count: Optional[int] = None
+    imphash: str
+
+    exported_dll_name: Optional[str] = None
+    exports: List[ExportedSymbol]
+
+    dirents: List[DirectoryEntry]
+    sections: List[Section]
+
+    ep_bytes: Optional[HexBytes] = None
+
+    overlay: Optional[Overlay] = None
+    resources: List[Resource]
+    versioninfo: List[KV]
+
+    # base64 encoded data
+    icon: Optional[str] = None
+    # MD5-like hash
+    icon_hash: Optional[str] = None
+    # MD5-like hash
+    icon_fuzzy: Optional[str] = None
+    # short hex string
+    icon_dhash: Optional[str] = None
+
+    digital_signers: List[DigitalSigner]
+    guest_signers: Signer
+
+
+# TODO(mr-tz): target.file.dotnet, target.file.extracted_files, target.file.extracted_files_tool,
+#  target.file.extracted_files_time
+# https://github.com/mandiant/capa/issues/1814
+class File(FlexibleModel):
+    type: str
+    cape_type_code: Optional[int] = None
+    cape_type: Optional[str] = None
+
+    pid: Optional[Union[int, Literal[""]]] = None
+    name: Union[List[str], str]
+    path: str
+    guest_paths: Union[List[str], str, None]
+    timestamp: Optional[str] = None
+
+    #
+    # hashes
+    #
+    crc32: str
+    md5: str
+    sha1: str
+    sha256: str
+    sha512: str
+    sha3_384: str
+    ssdeep: str
+    # unsure why this would ever be "False"
+    tlsh: Optional[Union[str, bool]] = None
+    rh_hash: Optional[str] = None
+
+    #
+    # other metadata, static analysis
+    #
+    size: int
+    pe: Optional[PE] = None
+    ep_bytes: Optional[HexBytes] = None
+    entrypoint: Optional[int] = None
+    data: Optional[str] = None
+    strings: Optional[List[str]] = None
+
+    #
+    # detections (skip)
+    #
+    yara: Skip = None
+    cape_yara: Skip = None
+    clamav: Skip = None
+    virustotal: Skip = None
+
+
+class ProcessFile(File):
+    #
+    # like a File, but also has dynamic analysis results
+    #
+    pid: Optional[int] = None
+    process_path: Optional[str] = None
+    process_name: Optional[str] = None
+    module_path: Optional[str] = None
+    virtual_address: Optional[HexInt] = None
+    target_pid: Optional[Union[int, str]] = None
+    target_path: Optional[str] = None
+    target_process: Optional[str] = None
+
+
+class Argument(ExactModel):
+    name: str
+    # unsure why empty list is provided here
+    value: Union[HexInt, int, str, EmptyList]
+    pretty_value: Optional[str] = None
+
+
+class Call(ExactModel):
+    timestamp: str
+    thread_id: int
+    category: str
+
+    api: str
+
+    arguments: List[Argument]
+    status: bool
+    return_: HexInt = Field(alias="return")
+    pretty_return: Optional[str] = None
+
+    repeated: int
+
+    # virtual addresses
+    caller: HexInt
+    parentcaller: HexInt
+
+    # index into calls array
+    id: int
+
+
+class Process(ExactModel):
+    process_id: int
+    process_name: str
+    parent_id: int
+    module_path: str
+    first_seen: str
+    calls: List[Call]
+    threads: List[int]
+    environ: Dict[str, str]
+
+
+class ProcessTree(ExactModel):
+    name: str
+    pid: int
+    parent_id: int
+    module_path: str
+    threads: List[int]
+    environ: Dict[str, str]
+    children: List["ProcessTree"]
+
+
+class Summary(ExactModel):
+    files: List[str]
+    read_files: List[str]
+    write_files: List[str]
+    delete_files: List[str]
+    keys: List[str]
+    read_keys: List[str]
+    write_keys: List[str]
+    delete_keys: List[str]
+    executed_commands: List[str]
+    resolved_apis: List[str]
+    mutexes: List[str]
+    created_services: List[str]
+    started_services: List[str]
+
+
+class EncryptedBuffer(ExactModel):
+    process_name: str
+    pid: int
+
+    api_call: str
+    buffer: str
+    buffer_size: Optional[int] = None
+    crypt_key: Optional[Union[HexInt, str]] = None
+
+
+class Behavior(ExactModel):
+    summary: Summary
+
+    # list of processes, of threads, of calls
+    processes: List[Process]
+    # tree of processes
+    processtree: List[ProcessTree]
+
+    anomaly: List[str]
+    encryptedbuffers: List[EncryptedBuffer]
+    # these are small objects that describe atomic events,
+    # like file move, registery access.
+    # we'll detect the same with our API call analyis.
+    enhanced: Skip = None
+
+
+class Target(ExactModel):
+    category: str
+    file: File
+    pe: Optional[PE] = None
+
+
+class Static(ExactModel):
+    pe: Optional[PE] = None
+    flare_capa: Skip = None
+
+
+class Cape(ExactModel):
+    payloads: List[ProcessFile]
+    configs: Skip = None
+
+
+# flexible because there may be more sorts of analysis
+# but we only care about the ones described here.
+class CapeReport(FlexibleModel):
+    # the input file, I think
+    target: Target
+    # info about the processing job, like machine and distributed metadata.
+    info: Info
+
+    #
+    # static analysis results
+    #
+    static: Optional[Static] = None
+    strings: Optional[List[str]] = None
+
+    #
+    # dynamic analysis results
+    #
+    # post-processed results: process tree, anomalies, etc
+    behavior: Behavior
+
+    # post-processed results: payloads and extracted configs
+    CAPE: Optional[Cape] = None
+    dropped: Optional[List[File]] = None
+    procdump: Optional[List[ProcessFile]] = None
+    procmemory: ListTODO
+
+    # =========================================================================
+    # information we won't use in capa
+    #
+
+    #
+    # NBIs and HBIs
+    # these are super interesting, but they don't enable use to detect behaviors.
+    # they take a lot of code to model and details to maintain.
+    #
+    # if we come up with a future use for this, go ahead and re-enable!
+    #
+    network: Skip = None
+    suricata: Skip = None
+    curtain: Skip = None
+    sysmon: Skip = None
+    url_analysis: Skip = None
+
+    # screenshot hash values
+    deduplicated_shots: Skip = None
+    # k-v pairs describing the time it took to run each stage.
+    statistics: Skip = None
+    # k-v pairs of ATT&CK ID to signature name or similar.
+    ttps: Skip = None
+    # debug log messages
+    debug: Skip = None
+
+    # various signature matches
+    # we could potentially extend capa to use this info one day,
+    # though it would be quite sandbox-specific,
+    # and more detection-oriented than capability detection.
+    signatures: Skip = None
+    malfamily_tag: Optional[str] = None
+    malscore: float
+    detections: Skip = None
+    detections2pid: Optional[Dict[int, List[str]]] = None
+    # AV detections for the sample.
+    virustotal: Skip = None
+
+    @classmethod
+    def from_buf(cls, buf: bytes) -> "CapeReport":
+        return cls.model_validate_json(buf)

capa/features/extractors/cape/process.py

@@ -0,0 +1,48 @@
+# 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 List, Tuple, Iterator
+
+from capa.features.common import String, Feature
+from capa.features.address import Address, ThreadAddress
+from capa.features.extractors.cape.models import Process
+from capa.features.extractors.base_extractor import ThreadHandle, ProcessHandle
+
+logger = logging.getLogger(__name__)
+
+
+def get_threads(ph: ProcessHandle) -> Iterator[ThreadHandle]:
+    """
+    get the threads associated with a given process
+    """
+    process: Process = ph.inner
+    threads: List[int] = process.threads
+
+    for thread in threads:
+        address: ThreadAddress = ThreadAddress(process=ph.address, tid=thread)
+        yield ThreadHandle(address=address, inner={})
+
+
+def extract_environ_strings(ph: ProcessHandle) -> Iterator[Tuple[Feature, Address]]:
+    """
+    extract strings from a process' provided environment variables.
+    """
+    process: Process = ph.inner
+
+    for value in (value for value in process.environ.values() if value):
+        yield String(value), ph.address
+
+
+def extract_features(ph: ProcessHandle) -> Iterator[Tuple[Feature, Address]]:
+    for handler in PROCESS_HANDLERS:
+        for feature, addr in handler(ph):
+            yield feature, addr
+
+
+PROCESS_HANDLERS = (extract_environ_strings,)

capa/features/extractors/cape/thread.py

@@ -0,0 +1,32 @@
+# 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 Iterator
+
+from capa.features.address import DynamicCallAddress
+from capa.features.extractors.helpers import generate_symbols
+from capa.features.extractors.cape.models import Process
+from capa.features.extractors.base_extractor import CallHandle, ThreadHandle, ProcessHandle
+
+logger = logging.getLogger(__name__)
+
+
+def get_calls(ph: ProcessHandle, th: ThreadHandle) -> Iterator[CallHandle]:
+    process: Process = ph.inner
+
+    tid = th.address.tid
+    for call_index, call in enumerate(process.calls):
+        if call.thread_id != tid:
+            continue
+
+        for symbol in generate_symbols("", call.api):
+            call.api = symbol
+
+            addr = DynamicCallAddress(thread=th.address, id=call_index)
+            yield CallHandle(address=addr, inner=call)

capa/features/extractors/common.py

@@ -1,4 +1,12 @@
+# 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 re
 import logging
 import binascii
 import contextlib
@@ -9,12 +17,33 @@ import pefile
 import capa.features
 import capa.features.extractors.elf
 import capa.features.extractors.pefile
-from capa.features.common import OS, FORMAT_PE, FORMAT_ELF, OS_WINDOWS, FORMAT_FREEZE, Arch, Format, String, Feature
+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":'
+MATCH_JSON_OBJECT = b'{"'
+
 
 def extract_file_strings(buf, **kwargs) -> Iterator[Tuple[String, Address]]:
     """
@@ -28,12 +57,19 @@ def extract_file_strings(buf, **kwargs) -> Iterator[Tuple[String, Address]]:
 
 
 def extract_format(buf) -> Iterator[Tuple[Feature, Address]]:
-    if buf.startswith(b"MZ"):
+    if buf.startswith(MATCH_PE):
         yield Format(FORMAT_PE), NO_ADDRESS
-    elif buf.startswith(b"\x7fELF"):
+    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
+    elif re.sub(rb"\s", b"", buf[:20]).startswith(MATCH_JSON_OBJECT):
+        # potential start of JSON object data without whitespace
+        # we don't know what it is exactly, but may support it (e.g. a dynamic CAPE sandbox report)
+        # skip verdict here and let subsequent code analyze this further
+        return
     else:
         # we likely end up here:
         #  1. handling a file format (e.g. macho)
@@ -44,10 +80,13 @@ def extract_format(buf) -> Iterator[Tuple[Feature, Address]]:
 
 
 def extract_arch(buf) -> Iterator[Tuple[Feature, Address]]:
-    if buf.startswith(b"MZ"):
+    if buf.startswith(MATCH_PE):
         yield from capa.features.extractors.pefile.extract_file_arch(pe=pefile.PE(data=buf))
 
-    elif buf.startswith(b"\x7fELF"):
+    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)
 
@@ -63,7 +102,7 @@ def extract_arch(buf) -> Iterator[Tuple[Feature, Address]]:
         #  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 futher CLI argument to specify the arch,
+        # 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.
         #
@@ -72,10 +111,15 @@ def extract_arch(buf) -> Iterator[Tuple[Feature, Address]]:
         return
 
 
-def extract_os(buf) -> Iterator[Tuple[Feature, Address]]:
-    if buf.startswith(b"MZ"):
+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(b"\x7fELF"):
+    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)
 
@@ -91,8 +135,6 @@ def extract_os(buf) -> Iterator[Tuple[Feature, Address]]:
         #  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 futher 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.

capa/features/extractors/dnfile/extractor.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 Mandiant, 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 +8,84 @@
 
 from __future__ import annotations
 
-from typing import List, Tuple, Iterator
+from typing import Dict, List, Tuple, Union, Iterator, Optional
+from pathlib import Path
 
 import dnfile
-from dncil.clr.token import Token
+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, AbsoluteVirtualAddress
-from capa.features.extractors.base_extractor import BBHandle, InsnHandle, FunctionHandle, FeatureExtractor
-from capa.features.extractors.dnfile.helpers import get_dotnet_managed_method_bodies
+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,
+    SampleHashes,
+    FunctionHandle,
+    StaticFeatureExtractor,
+)
+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 DnfileFeatureExtractor(FeatureExtractor):
-    def __init__(self, path: str):
-        super(DnfileFeatureExtractor, self).__init__()
-        self.pe: dnfile.dnPE = dnfile.dnPE(path)
+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(StaticFeatureExtractor):
+    def __init__(self, path: Path):
+        self.pe: dnfile.dnPE = dnfile.dnPE(str(path))
+        super().__init__(hashes=SampleHashes.from_bytes(path.read_bytes()))
+
+        # 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))
 
@@ -42,12 +99,45 @@ class DnfileFeatureExtractor(FeatureExtractor):
         yield from capa.features.extractors.dnfile.file.extract_features(self.pe)
 
     def get_functions(self) -> Iterator[FunctionHandle]:
-        for token, f in get_dotnet_managed_method_bodies(self.pe):
-            yield FunctionHandle(address=DNTokenAddress(Token(token)), inner=f, ctx={"pe": self.pe})
+        # 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},
+            )
 
-    def extract_function_features(self, f):
-        # TODO
-        yield from []
+            # 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
@@ -63,7 +153,7 @@ class DnfileFeatureExtractor(FeatureExtractor):
     def get_instructions(self, fh, bbh):
         for insn in bbh.inner.instructions:
             yield InsnHandle(
-                address=DNTokenOffsetAddress(bbh.address.token, insn.offset - (fh.inner.offset + fh.inner.header_size)),
+                address=DNTokenOffsetAddress(bbh.address, insn.offset - (fh.inner.offset + fh.inner.header_size)),
                 inner=insn,
             )
 

capa/features/extractors/dnfile/file.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 Mandiant, 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
@@ -48,7 +48,7 @@ def extract_file_class_features(pe: dnfile.dnPE) -> Iterator[Tuple[Class, Addres
 
 def extract_features(pe: dnfile.dnPE) -> Iterator[Tuple[Feature, Address]]:
     for file_handler in FILE_HANDLERS:
-        for (feature, address) in file_handler(pe):
+        for feature, address in file_handler(pe):
             yield feature, address
 
 

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

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 Mandiant, 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
@@ -9,7 +9,7 @@
 from __future__ import annotations
 
 import logging
-from typing import Any, Tuple, Iterator, Optional
+from typing import Dict, Tuple, Union, Iterator, Optional
 
 import dnfile
 from dncil.cil.body import CilMethodBody
@@ -17,10 +17,10 @@ from dncil.cil.error import MethodBodyFormatError
 from dncil.clr.token import Token, StringToken, InvalidToken
 from dncil.cil.body.reader import CilMethodBodyReaderBase
 
-logger = logging.getLogger(__name__)
+from capa.features.common import FeatureAccess
+from capa.features.extractors.dnfile.types import DnType, DnUnmanagedMethod
 
-# key indexes to dotnet metadata tables
-DOTNET_META_TABLES_BY_INDEX = {table.value: table.name for table in dnfile.enums.MetadataTables}
+logger = logging.getLogger(__name__)
 
 
 class DnfileMethodBodyReader(CilMethodBodyReaderBase):
@@ -41,90 +41,20 @@ class DnfileMethodBodyReader(CilMethodBodyReaderBase):
         return self.offset
 
 
-class DnClass(object):
-    def __init__(self, token: int, namespace: str, classname: str):
-        self.token: int = token
-        self.namespace: str = namespace
-        self.classname: str = classname
-
-    def __hash__(self):
-        return hash((self.token,))
-
-    def __eq__(self, other):
-        return self.token == other.token
-
-    def __str__(self):
-        return DnClass.format_name(self.namespace, self.classname)
-
-    def __repr__(self):
-        return str(self)
-
-    @staticmethod
-    def format_name(namespace: str, classname: str):
-        name: str = classname
-        if namespace:
-            # like System.IO.File::OpenRead
-            name = f"{namespace}.{name}"
-        return name
-
-
-class DnMethod(DnClass):
-    def __init__(self, token: int, namespace: str, classname: str, methodname: str):
-        super(DnMethod, self).__init__(token, namespace, classname)
-        self.methodname: str = methodname
-
-    def __str__(self):
-        return DnMethod.format_name(self.namespace, self.classname, self.methodname)
-
-    @staticmethod
-    def format_name(namespace: str, classname: str, methodname: str):  # type: ignore
-        # like File::OpenRead
-        name: str = f"{classname}::{methodname}"
-        if namespace:
-            # like System.IO.File::OpenRead
-            name = f"{namespace}.{name}"
-        return name
-
-
-class DnUnmanagedMethod:
-    def __init__(self, token: int, modulename: str, methodname: str):
-        self.token: int = token
-        self.modulename: str = modulename
-        self.methodname: str = methodname
-
-    def __hash__(self):
-        return hash((self.token,))
-
-    def __eq__(self, other):
-        return self.token == other.token
-
-    def __str__(self):
-        return DnUnmanagedMethod.format_name(self.modulename, self.methodname)
-
-    def __repr__(self):
-        return str(self)
-
-    @staticmethod
-    def format_name(modulename, methodname):
-        return f"{modulename}.{methodname}"
-
-
-def resolve_dotnet_token(pe: dnfile.dnPE, token: Token) -> Any:
+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_name: str = DOTNET_META_TABLES_BY_INDEX.get(token.table, "")
-    if not table_name:
-        # table_index is not valid
-        return InvalidToken(token.value)
-
-    table: Any = getattr(pe.net.mdtables, table_name, None)
+    table: Optional[dnfile.base.ClrMetaDataTable] = pe.net.mdtables.tables.get(token.table)
     if table is None:
-        # table index is valid but table is not present
+        # table index is not valid
         return InvalidToken(token.value)
 
     try:
@@ -139,16 +69,23 @@ def read_dotnet_method_body(pe: dnfile.dnPE, row: dnfile.mdtable.MethodDefRow) -
     try:
         return CilMethodBody(DnfileMethodBodyReader(pe, row))
     except MethodBodyFormatError as e:
-        logger.warn("failed to parse managed method body @ 0x%08x (%s)" % (row.Rva, 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.warn("failed to decode #US stream index 0x%08x (%s)" % (token.rid, e))
+        logger.debug("failed to decode #US stream index 0x%08x (%s)", token.rid, e)
         return None
 
     if user_string is None:
@@ -157,7 +94,7 @@ def read_dotnet_user_string(pe: dnfile.dnPE, token: StringToken) -> Optional[str
     return user_string.value
 
 
-def get_dotnet_managed_imports(pe: dnfile.dnPE) -> Iterator[DnMethod]:
+def get_dotnet_managed_imports(pe: dnfile.dnPE) -> Iterator[DnType]:
     """get managed imports from MemberRef table
 
     see https://www.ntcore.com/files/dotnetformat.htm
@@ -171,15 +108,80 @@ def get_dotnet_managed_imports(pe: dnfile.dnPE) -> Iterator[DnMethod]:
             TypeName (index into String heap)
             TypeNamespace (index into String heap)
     """
-    for (rid, row) in enumerate(iter_dotnet_table(pe, "MemberRef")):
-        if not isinstance(row.Class.row, dnfile.mdtable.TypeRefRow):
+    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(pe.net.mdtables.MemberRef.number, rid + 1)
-        yield DnMethod(token, row.Class.row.TypeNamespace, row.Class.row.TypeName, row.Name)
+        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:]
+
+        typerefnamespace, typerefname = resolve_nested_typeref_name(
+            member_ref.Class.row_index, member_ref.Class.row, pe
+        )
+
+        yield DnType(
+            token,
+            typerefname,
+            namespace=typerefnamespace,
+            member=member_ref_name,
+            access=access,
+        )
 
 
-def get_dotnet_managed_methods(pe: dnfile.dnPE) -> Iterator[DnMethod]:
+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
@@ -188,29 +190,83 @@ def get_dotnet_managed_methods(pe: dnfile.dnPE) -> Iterator[DnMethod]:
         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 continguous run of Methods owned by this Type)
+            MethodList (index into MethodDef table; it marks the first of a contiguous run of Methods owned by this Type)
     """
-    for row in iter_dotnet_table(pe, "TypeDef"):
-        for index in row.MethodList:
-            token = calculate_dotnet_token_value(index.table.number, index.row_index)
-            yield DnMethod(token, row.TypeNamespace, row.TypeName, index.row.Name)
+    nested_class_table = get_dotnet_nested_class_table_index(pe)
+
+    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:]
+
+            typedefnamespace, typedefname = resolve_nested_typedef_name(nested_class_table, rid, typedef, pe)
+
+            yield DnType(token, typedefname, namespace=typedefnamespace, 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)
+    """
+    nested_class_table = get_dotnet_nested_class_table_index(pe)
+
+    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
+
+            typedefnamespace, typedefname = resolve_nested_typedef_name(nested_class_table, rid, typedef, pe)
+
+            token: int = calculate_dotnet_token_value(field.table.number, field.row_index)
+            yield DnType(token, typedefname, namespace=typedefnamespace, member=field.row.Name)
 
 
 def get_dotnet_managed_method_bodies(pe: dnfile.dnPE) -> Iterator[Tuple[int, CilMethodBody]]:
     """get managed methods from MethodDef table"""
-    if not hasattr(pe.net.mdtables, "MethodDef"):
-        return
+    for rid, method_def in iter_dotnet_table(pe, dnfile.mdtable.MethodDef.number):
+        assert isinstance(method_def, dnfile.mdtable.MethodDefRow)
 
-    for (rid, row) in enumerate(pe.net.mdtables.MethodDef):
-        if not row.ImplFlags.miIL or any((row.Flags.mdAbstract, row.Flags.mdPinvokeImpl)):
+        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, row)
+        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.enums.MetadataTables.MethodDef.value, rid + 1)
+        token: int = calculate_dotnet_token_value(dnfile.mdtable.MethodDef.number, rid)
         yield token, body
 
 
@@ -225,37 +281,168 @@ def get_dotnet_unmanaged_imports(pe: dnfile.dnPE) -> Iterator[DnUnmanagedMethod]
             ImportName (index into the String heap)
             ImportScope (index into the ModuleRef table)
     """
-    for row in iter_dotnet_table(pe, "ImplMap"):
-        modulename: str = row.ImportScope.row.Name
-        methodname: str = row.ImportName
+    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(row.MemberForwarded.table.number, row.MemberForwarded.row_index)
+        token: int = calculate_dotnet_token_value(member_forward_table, member_forward_row)
 
         # like Kernel32.dll
-        if modulename and "." in modulename:
-            modulename = modulename.split(".")[0]
+        if module and "." in module:
+            module = module.split(".")[0]
 
         # like kernel32.CreateFileA
-        yield DnUnmanagedMethod(token, modulename, methodname)
+        yield DnUnmanagedMethod(token, module, method)
+
+
+def get_dotnet_table_row(pe: dnfile.dnPE, table_index: int, row_index: int) -> Optional[dnfile.base.MDTableRow]:
+    assert pe.net is not None
+    assert pe.net.mdtables is not None
+
+    if row_index - 1 <= 0:
+        return None
+
+    try:
+        table = pe.net.mdtables.tables.get(table_index, [])
+        return table[row_index - 1]
+    except IndexError:
+        return None
+
+
+def resolve_nested_typedef_name(
+    nested_class_table: dict, index: int, typedef: dnfile.mdtable.TypeDefRow, pe: dnfile.dnPE
+) -> Tuple[str, Tuple[str, ...]]:
+    """Resolves all nested TypeDef class names. Returns the namespace as a str and the nested TypeRef name as a tuple"""
+
+    if index in nested_class_table:
+        typedef_name = []
+        name = typedef.TypeName
+
+        # Append the current typedef name
+        typedef_name.append(name)
+
+        while nested_class_table[index] in nested_class_table:
+            # Iterate through the typedef table to resolve the nested name
+            table_row = get_dotnet_table_row(pe, dnfile.mdtable.TypeDef.number, nested_class_table[index])
+            if table_row is None:
+                return typedef.TypeNamespace, tuple(typedef_name[::-1])
+
+            name = table_row.TypeName
+            typedef_name.append(name)
+            index = nested_class_table[index]
+
+        # Document the root enclosing details
+        table_row = get_dotnet_table_row(pe, dnfile.mdtable.TypeDef.number, nested_class_table[index])
+        if table_row is None:
+            return typedef.TypeNamespace, tuple(typedef_name[::-1])
+
+        enclosing_name = table_row.TypeName
+        typedef_name.append(enclosing_name)
+
+        return table_row.TypeNamespace, tuple(typedef_name[::-1])
+
+    else:
+        return typedef.TypeNamespace, (typedef.TypeName,)
+
+
+def resolve_nested_typeref_name(
+    index: int, typeref: dnfile.mdtable.TypeRefRow, pe: dnfile.dnPE
+) -> Tuple[str, Tuple[str, ...]]:
+    """Resolves all nested TypeRef class names. Returns the namespace as a str and the nested TypeRef name as a tuple"""
+    # If the ResolutionScope decodes to a typeRef type then it is nested
+    if isinstance(typeref.ResolutionScope.table, dnfile.mdtable.TypeRef):
+        typeref_name = []
+        name = typeref.TypeName
+        # Not appending the current typeref name to avoid potential duplicate
+
+        # Validate index
+        table_row = get_dotnet_table_row(pe, dnfile.mdtable.TypeRef.number, index)
+        if table_row is None:
+            return typeref.TypeNamespace, (typeref.TypeName,)
+
+        while isinstance(table_row.ResolutionScope.table, dnfile.mdtable.TypeRef):
+            # Iterate through the typeref table to resolve the nested name
+            typeref_name.append(name)
+            name = table_row.TypeName
+            table_row = get_dotnet_table_row(pe, dnfile.mdtable.TypeRef.number, table_row.ResolutionScope.row_index)
+            if table_row is None:
+                return typeref.TypeNamespace, tuple(typeref_name[::-1])
+
+        # Document the root enclosing details
+        typeref_name.append(table_row.TypeName)
+
+        return table_row.TypeNamespace, tuple(typeref_name[::-1])
+
+    else:
+        return typeref.TypeNamespace, (typeref.TypeName,)
+
+
+def get_dotnet_nested_class_table_index(pe: dnfile.dnPE) -> Dict[int, int]:
+    """Build index for EnclosingClass based off the NestedClass row index in the nestedclass table"""
+    nested_class_table = {}
+
+    # Used to find nested classes in typedef
+    for _, nestedclass in iter_dotnet_table(pe, dnfile.mdtable.NestedClass.number):
+        assert isinstance(nestedclass, dnfile.mdtable.NestedClassRow)
+        nested_class_table[nestedclass.NestedClass.row_index] = nestedclass.EnclosingClass.row_index
+
+    return nested_class_table
+
+
+def get_dotnet_types(pe: dnfile.dnPE) -> Iterator[DnType]:
+    """get .NET types from TypeDef and TypeRef tables"""
+    nested_class_table = get_dotnet_nested_class_table_index(pe)
+
+    for rid, typedef in iter_dotnet_table(pe, dnfile.mdtable.TypeDef.number):
+        assert isinstance(typedef, dnfile.mdtable.TypeDefRow)
+
+        typedefnamespace, typedefname = resolve_nested_typedef_name(nested_class_table, rid, typedef, pe)
+
+        typedef_token: int = calculate_dotnet_token_value(dnfile.mdtable.TypeDef.number, rid)
+        yield DnType(typedef_token, typedefname, namespace=typedefnamespace)
+
+    for rid, typeref in iter_dotnet_table(pe, dnfile.mdtable.TypeRef.number):
+        assert isinstance(typeref, dnfile.mdtable.TypeRefRow)
+
+        typerefnamespace, typerefname = resolve_nested_typeref_name(typeref.ResolutionScope.row_index, typeref, pe)
+
+        typeref_token: int = calculate_dotnet_token_value(dnfile.mdtable.TypeRef.number, rid)
+        yield DnType(typeref_token, typerefname, namespace=typerefnamespace)
 
 
 def calculate_dotnet_token_value(table: int, rid: int) -> int:
     return ((table & 0xFF) << Token.TABLE_SHIFT) | (rid & Token.RID_MASK)
 
 
-def is_dotnet_table_valid(pe: dnfile.dnPE, table_name: str) -> bool:
-    return bool(getattr(pe.net.mdtables, table_name, None))
-
-
 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, name: str) -> Iterator[Any]:
-    if not is_dotnet_table_valid(pe, name):
-        return
-    for row in getattr(pe.net.mdtables, name):
-        yield row
+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

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 Mandiant, 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
@@ -8,175 +8,220 @@
 
 from __future__ import annotations
 
-from typing import Any, Dict, Tuple, Union, Iterator, Optional
+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.cil.body import CilMethodBody
 from dncil.clr.token import Token, StringToken, InvalidToken
 from dncil.cil.opcode import OpCodes
-from dncil.cil.instruction import Instruction
 
 import capa.features.extractors.helpers
-from capa.features.insn import API, Number
-from capa.features.common import Class, String, Feature, Namespace, Characteristic
+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 (
-    DnClass,
-    DnMethod,
-    DnUnmanagedMethod,
     resolve_dotnet_token,
     read_dotnet_user_string,
-    get_dotnet_managed_imports,
-    get_dotnet_managed_methods,
-    get_dotnet_unmanaged_imports,
+    calculate_dotnet_token_value,
 )
 
-
-def get_managed_imports(ctx: Dict) -> Dict:
-    if "managed_imports_cache" not in ctx:
-        ctx["managed_imports_cache"] = {}
-        for method in get_dotnet_managed_imports(ctx["pe"]):
-            ctx["managed_imports_cache"][method.token] = method
-    return ctx["managed_imports_cache"]
+logger = logging.getLogger(__name__)
 
 
-def get_unmanaged_imports(ctx: Dict) -> Dict:
-    if "unmanaged_imports_cache" not in ctx:
-        ctx["unmanaged_imports_cache"] = {}
-        for imp in get_dotnet_unmanaged_imports(ctx["pe"]):
-            ctx["unmanaged_imports_cache"][imp.token] = imp
-    return ctx["unmanaged_imports_cache"]
+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
 
-def get_methods(ctx: Dict) -> Dict:
-    if "methods_cache" not in ctx:
-        ctx["methods_cache"] = {}
-        for method in get_dotnet_managed_methods(ctx["pe"]):
-            ctx["methods_cache"][method.token] = method
-    return ctx["methods_cache"]
+        token_ = calculate_dotnet_token_value(row.Method.table.number, row.Method.row_index)
+    else:
+        token_ = token.value
 
-
-def get_callee(ctx: Dict, token: int) -> Union[DnMethod, DnUnmanagedMethod, None]:
-    """map dotnet token to un/managed method"""
-    callee: Union[DnMethod, DnUnmanagedMethod, None] = get_managed_imports(ctx).get(token, None)
-    if not callee:
+    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 = get_unmanaged_imports(ctx).get(token, None)
-        if not callee:
-            callee = get_methods(ctx).get(token, None)
+        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"""
-    insn: Instruction = ih.inner
-
-    if insn.opcode not in (OpCodes.Call, OpCodes.Callvirt, OpCodes.Jmp, OpCodes.Calli):
+    if ih.inner.opcode not in (
+        OpCodes.Call,
+        OpCodes.Callvirt,
+        OpCodes.Jmp,
+        OpCodes.Newobj,
+    ):
         return
 
-    callee: Union[DnMethod, DnUnmanagedMethod, None] = get_callee(fh.ctx, insn.operand.value)
-    if callee is None:
-        return
-
-    if isinstance(callee, DnUnmanagedMethod):
+    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.modulename, callee.methodname):
+        for name in capa.features.extractors.helpers.generate_symbols(callee.module, callee.method):
             yield API(name), ih.address
-    else:
-        # like System.IO.File::Delete
-        yield API(str(callee)), ih.address
 
 
-def extract_insn_class_features(fh: FunctionHandle, bh, ih: InsnHandle) -> Iterator[Tuple[Class, Address]]:
-    """parse instruction class features"""
-    if ih.inner.opcode not in (OpCodes.Call, OpCodes.Callvirt, OpCodes.Jmp, OpCodes.Calli):
-        return
+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
 
-    row: Any = resolve_dotnet_token(fh.ctx["pe"], Token(ih.inner.operand.value))
+    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
 
-    if not isinstance(row, dnfile.mdtable.MemberRefRow):
-        return
-    if not isinstance(row.Class.row, (dnfile.mdtable.TypeRefRow, dnfile.mdtable.TypeDefRow)):
-        return
+    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
 
-    yield Class(DnClass.format_name(row.Class.row.TypeNamespace, row.Class.row.TypeName)), ih.address
+    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_features(fh: FunctionHandle, bh, ih: InsnHandle) -> Iterator[Tuple[Namespace, Address]]:
-    """parse instruction namespace features"""
-    if ih.inner.opcode not in (OpCodes.Call, OpCodes.Callvirt, OpCodes.Jmp, OpCodes.Calli):
-        return
+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
 
-    row: Any = resolve_dotnet_token(fh.ctx["pe"], Token(ih.inner.operand.value))
+    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)
 
-    if not isinstance(row, dnfile.mdtable.MemberRefRow):
-        return
-    if not isinstance(row.Class.row, (dnfile.mdtable.TypeRefRow, dnfile.mdtable.TypeDefRow)):
-        return
-    if not row.Class.row.TypeNamespace:
-        return
+    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)
 
-    yield Namespace(row.Class.row.TypeNamespace), ih.address
+    # 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"""
-    insn: Instruction = ih.inner
-
-    if insn.is_ldc():
-        yield Number(insn.get_ldc()), ih.address
+    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"""
-    f: CilMethodBody = fh.inner
-    insn: Instruction = ih.inner
-
-    if not insn.is_ldstr():
+    if not ih.inner.is_ldstr():
         return
 
-    if not isinstance(insn.operand, StringToken):
+    if not isinstance(ih.inner.operand, StringToken):
         return
 
-    user_string: Optional[str] = read_dotnet_user_string(fh.ctx["pe"], insn.operand)
+    user_string: Optional[str] = read_dotnet_user_string(fh.ctx["pe"], ih.inner.operand)
     if user_string is None:
         return
 
-    yield String(user_string), ih.address
+    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]]:
-    insn: Instruction = ih.inner
-    if insn.opcode not in (OpCodes.Call, OpCodes.Callvirt, OpCodes.Jmp, OpCodes.Calli):
+    if ih.inner.opcode not in (OpCodes.Call, OpCodes.Callvirt, OpCodes.Jmp):
         return
 
-    token: Any = resolve_dotnet_token(fh.ctx["pe"], insn.operand)
-    if isinstance(token, InvalidToken):
-        return
-    if not isinstance(token, dnfile.mdtable.MethodDefRow):
+    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((token.Flags.mdPinvokeImpl, token.ImplFlags.miUnmanaged, token.ImplFlags.miNative)):
+    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):
+        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_features,
-    extract_insn_class_features,
+    extract_insn_namespace_class_features,
     extract_unmanaged_call_characteristic_features,
 )

capa/features/extractors/dnfile/types.py

@@ -0,0 +1,80 @@
+# 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, Optional
+
+
+class DnType:
+    def __init__(
+        self, token: int, class_: Tuple[str, ...], namespace: str = "", member: str = "", access: Optional[str] = None
+    ):
+        self.token: int = token
+        self.access: Optional[str] = access
+        self.namespace: str = namespace
+        self.class_: Tuple[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_: Tuple[str, ...], namespace: str = "", member: str = ""):
+        if len(class_) > 1:
+            class_str = "/".join(class_)  # Concat items in tuple, separated by a "/"
+        else:
+            class_str = "".join(class_)  # Convert tuple to str
+        # like File::OpenRead
+        name: str = f"{class_str}::{member}" if member else class_str
+        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

@@ -1,116 +0,0 @@
-import logging
-from typing import Tuple, Iterator
-
-import dnfile
-import pefile
-
-from capa.features.common import OS, OS_ANY, ARCH_ANY, ARCH_I386, 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_DOTNET), NO_ADDRESS
-
-
-def extract_file_os(**kwargs) -> Iterator[Tuple[Feature, Address]]:
-    yield OS(OS_ANY), NO_ADDRESS
-
-
-def extract_file_arch(pe, **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
-    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: str):
-        super(DnfileFeatureExtractor, self).__init__()
-        self.path: str = path
-        self.pe: dnfile.dnPE = dnfile.dnPE(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
-        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 not bool(self.pe.net.Flags.CLR_ILONLY)
-
-    def get_runtime_version(self) -> Tuple[int, int]:
-        return self.pe.net.struct.MajorRuntimeVersion, self.pe.net.struct.MinorRuntimeVersion
-
-    def get_meta_version_string(self) -> str:
-        return self.pe.net.metadata.struct.Version.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

@@ -1,10 +1,16 @@
+# 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 itertools
 from typing import Tuple, Iterator
+from pathlib import Path
 
 import dnfile
 import pefile
-from dncil.clr.token import Token
 
 import capa.features.extractors.helpers
 from capa.features.file import Import, FunctionName
@@ -13,6 +19,7 @@ from capa.features.common import (
     OS_ANY,
     ARCH_ANY,
     ARCH_I386,
+    FORMAT_PE,
     ARCH_AMD64,
     FORMAT_DOTNET,
     Arch,
@@ -23,40 +30,43 @@ from capa.features.common import (
     Namespace,
     Characteristic,
 )
-from capa.features.address import NO_ADDRESS, Address, DNTokenAddress, DNTokenOffsetAddress, AbsoluteVirtualAddress
-from capa.features.extractors.base_extractor import FeatureExtractor
+from capa.features.address import NO_ADDRESS, Address, DNTokenAddress
+from capa.features.extractors.dnfile.types import DnType
+from capa.features.extractors.base_extractor import SampleHashes, StaticFeatureExtractor
 from capa.features.extractors.dnfile.helpers import (
-    DnClass,
-    DnMethod,
     iter_dotnet_table,
     is_dotnet_mixed_mode,
     get_dotnet_managed_imports,
     get_dotnet_managed_methods,
+    resolve_nested_typedef_name,
+    resolve_nested_typeref_name,
     calculate_dotnet_token_value,
     get_dotnet_unmanaged_imports,
+    get_dotnet_nested_class_table_index,
 )
 
 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(Token(method.token))
+        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.modulename, imp.methodname):
-            yield Import(name), DNTokenAddress(Token(imp.token))
+        for name in capa.features.extractors.helpers.generate_symbols(imp.module, imp.method, include_dll=True):
+            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(Token(method.token))
+        yield FunctionName(str(method)), DNTokenAddress(method.token)
 
 
 def extract_file_namespace_features(pe: dnfile.dnPE, **kwargs) -> Iterator[Tuple[Namespace, Address]]:
@@ -65,11 +75,15 @@ def extract_file_namespace_features(pe: dnfile.dnPE, **kwargs) -> Iterator[Tuple
     # namespaces may be referenced multiple times, so we need to filter
     namespaces = set()
 
-    for row in iter_dotnet_table(pe, "TypeDef"):
-        namespaces.add(row.TypeNamespace)
+    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 row in iter_dotnet_table(pe, "TypeRef"):
-        namespaces.add(row.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("")
@@ -81,13 +95,25 @@ def extract_file_namespace_features(pe: dnfile.dnPE, **kwargs) -> Iterator[Tuple
 
 def extract_file_class_features(pe: dnfile.dnPE, **kwargs) -> Iterator[Tuple[Class, Address]]:
     """emit class features from TypeRef and TypeDef tables"""
-    for (rid, row) in enumerate(iter_dotnet_table(pe, "TypeDef")):
-        token = calculate_dotnet_token_value(pe.net.mdtables.TypeDef.number, rid + 1)
-        yield Class(DnClass.format_name(row.TypeNamespace, row.TypeName)), DNTokenAddress(Token(token))
+    nested_class_table = get_dotnet_nested_class_table_index(pe)
 
-    for (rid, row) in enumerate(iter_dotnet_table(pe, "TypeRef")):
-        token = calculate_dotnet_token_value(pe.net.mdtables.TypeRef.number, rid + 1)
-        yield Class(DnClass.format_name(row.TypeNamespace, row.TypeName)), DNTokenAddress(Token(token))
+    for rid, typedef in iter_dotnet_table(pe, dnfile.mdtable.TypeDef.number):
+        # emit internal .NET classes
+        assert isinstance(typedef, dnfile.mdtable.TypeDefRow)
+
+        typedefnamespace, typedefname = resolve_nested_typedef_name(nested_class_table, rid, typedef, pe)
+
+        token = calculate_dotnet_token_value(dnfile.mdtable.TypeDef.number, rid)
+        yield Class(DnType.format_name(typedefname, namespace=typedefnamespace)), DNTokenAddress(token)
+
+    for rid, typeref in iter_dotnet_table(pe, dnfile.mdtable.TypeRef.number):
+        # emit external .NET classes
+        assert isinstance(typeref, dnfile.mdtable.TypeRefRow)
+
+        typerefnamespace, typerefname = resolve_nested_typeref_name(typeref.ResolutionScope.row_index, typeref, pe)
+
+        token = calculate_dotnet_token_value(dnfile.mdtable.TypeRef.number, rid)
+        yield Class(DnType.format_name(typerefname, namespace=typerefnamespace)), DNTokenAddress(token)
 
 
 def extract_file_os(**kwargs) -> Iterator[Tuple[OS, Address]]:
@@ -97,6 +123,9 @@ def extract_file_os(**kwargs) -> Iterator[Tuple[OS, 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:
@@ -145,11 +174,11 @@ GLOBAL_HANDLERS = (
 )
 
 
-class DotnetFileFeatureExtractor(FeatureExtractor):
-    def __init__(self, path: str):
-        super(DotnetFileFeatureExtractor, self).__init__()
-        self.path: str = path
-        self.pe: dnfile.dnPE = dnfile.dnPE(path)
+class DotnetFileFeatureExtractor(StaticFeatureExtractor):
+    def __init__(self, path: Path):
+        super().__init__(hashes=SampleHashes.from_bytes(path.read_bytes()))
+        self.path: Path = path
+        self.pe: dnfile.dnPE = dnfile.dnPE(str(path))
 
     def get_base_address(self):
         return NO_ADDRESS
@@ -158,6 +187,9 @@ class DotnetFileFeatureExtractor(FeatureExtractor):
         # 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):
@@ -173,10 +205,23 @@ class DotnetFileFeatureExtractor(FeatureExtractor):
         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:
-        return self.pe.net.metadata.struct.Version.rstrip(b"\x00").decode("utf-8")
+        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")

capa/features/extractors/elffile.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 Mandiant, 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
@@ -7,43 +7,90 @@
 # See the License for the specific language governing permissions and limitations under the License.
 import io
 import logging
-import contextlib
 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 Import, Section
+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.elf import Arch as ElfArch
-from capa.features.extractors.base_extractor import FeatureExtractor
+from capa.features.extractors.base_extractor import SampleHashes, StaticFeatureExtractor
 
 logger = logging.getLogger(__name__)
 
 
-def extract_file_import_names(elf, **kwargs):
-    # see https://github.com/eliben/pyelftools/blob/0664de05ed2db3d39041e2d51d19622a8ef4fb0f/scripts/readelf.py#L372
-    symbol_tables = [(idx, s) for idx, s in enumerate(elf.iter_sections()) if isinstance(s, SymbolTableSection)]
-
-    for section_index, section in symbol_tables:
+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))
+            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()))
+        logger.debug("Symbol table '%s' contains %s entries:", section.name, section.num_symbols())
 
-        for nsym, symbol in enumerate(section.iter_symbols()):
-            if symbol.name and symbol.entry.st_info.type == "STT_FUNC":
-                # TODO symbol address
-                # TODO symbol version info?
-                yield Import(symbol.name), FileOffsetAddress(0x0)
+        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_section_names(elf, **kwargs):
+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)
@@ -55,7 +102,7 @@ def extract_file_strings(buf, **kwargs):
     yield from capa.features.extractors.common.extract_file_strings(buf)
 
 
-def extract_file_os(elf, buf, **kwargs):
+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:
@@ -69,13 +116,12 @@ def extract_file_format(**kwargs):
     yield Format(FORMAT_ELF), NO_ADDRESS
 
 
-def extract_file_arch(elf, **kwargs):
-    # TODO merge with capa.features.extractors.elf.detect_elf_arch()
+def extract_file_arch(elf: ELFFile, **kwargs):
     arch = elf.get_machine_arch()
     if arch == "x86":
-        yield Arch(ElfArch.I386), NO_ADDRESS
+        yield Arch("i386"), NO_ADDRESS
     elif arch == "x64":
-        yield Arch(ElfArch.AMD64), NO_ADDRESS
+        yield Arch("amd64"), NO_ADDRESS
     else:
         logger.warning("unsupported architecture: %s", arch)
 
@@ -87,7 +133,7 @@ def extract_file_features(elf: ELFFile, buf: bytes) -> Iterator[Tuple[Feature, i
 
 
 FILE_HANDLERS = (
-    # TODO extract_file_export_names,
+    extract_file_export_names,
     extract_file_import_names,
     extract_file_section_names,
     extract_file_strings,
@@ -108,12 +154,11 @@ GLOBAL_HANDLERS = (
 )
 
 
-class ElfFeatureExtractor(FeatureExtractor):
-    def __init__(self, path: str):
-        super(ElfFeatureExtractor, self).__init__()
-        self.path = path
-        with open(self.path, "rb") as f:
-            self.elf = ELFFile(io.BytesIO(f.read()))
+class ElfFeatureExtractor(StaticFeatureExtractor):
+    def __init__(self, path: Path):
+        super().__init__(SampleHashes.from_bytes(path.read_bytes()))
+        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
@@ -122,15 +167,13 @@ class ElfFeatureExtractor(FeatureExtractor):
                 return AbsoluteVirtualAddress(segment.header.p_vaddr)
 
     def extract_global_features(self):
-        with open(self.path, "rb") as f:
-            buf = f.read()
+        buf = self.path.read_bytes()
 
         for feature, addr in extract_global_features(self.elf, buf):
             yield feature, addr
 
     def extract_file_features(self):
-        with open(self.path, "rb") as f:
-            buf = f.read()
+        buf = self.path.read_bytes()
 
         for feature, addr in extract_file_features(self.elf, buf):
             yield feature, addr
@@ -153,8 +196,8 @@ class ElfFeatureExtractor(FeatureExtractor):
     def extract_insn_features(self, f, bb, insn):
         raise NotImplementedError("ElfFeatureExtractor can only be used to extract file features")
 
-    def is_library_function(self, va):
+    def is_library_function(self, addr):
         raise NotImplementedError("ElfFeatureExtractor can only be used to extract file features")
 
-    def get_function_name(self, va):
+    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 +1,93 @@
+# 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,
+    SampleHashes,
+    FunctionHandle,
+    StaticFeatureExtractor,
+)
+
+
+class GhidraFeatureExtractor(StaticFeatureExtractor):
+    def __init__(self):
+        import capa.features.extractors.ghidra.helpers as ghidra_helpers
+
+        super().__init__(
+            SampleHashes(
+                md5=capa.ghidra.helpers.get_file_md5(),
+                # ghidra doesn't expose this hash.
+                # https://ghidra.re/ghidra_docs/api/ghidra/program/model/listing/Program.html
+                #
+                # the hashes are stored in the database, not computed on the fly,
+                # so its probably not trivial to add SHA1.
+                sha1="",
+                sha256=capa.ghidra.helpers.get_file_sha256(),
+            )
+        )
+
+        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 +1,204 @@
+# 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)  # 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], include_dll=True):
+            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 not block.isInitialized():
+            continue
+
+        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 +1,301 @@
+# 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
+
+
+def find_data_references_from_insn(insn, max_depth: int = 10):
+    """yield data references from given instruction"""
+    for reference in insn.getReferencesFrom():
+        if not reference.getReferenceType().isData():
+            # only care about data references
+            continue
+
+        to_addr = reference.getToAddress()
+
+        for _ in range(max_depth - 1):
+            data = getDataAt(to_addr)  # type: ignore [name-defined] # noqa: F821
+            if data and data.isPointer():
+                ptr_value = data.getValue()
+
+                if ptr_value is None:
+                    break
+
+                to_addr = ptr_value
+            else:
+                break
+
+        yield to_addr

capa/features/extractors/ghidra/insn.py

@@ -0,0 +1,503 @@
+# 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
+
+
+OPERAND_TYPE_DYNAMIC_ADDRESS = OperandType.DYNAMIC | OperandType.ADDRESS
+
+
+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 == OPERAND_TYPE_DYNAMIC_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
+
+    if capa.features.extractors.ghidra.helpers.is_stack_referenced(insn):
+        # ignore stack references
+        return
+
+    # 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 not op_objs:
+                continue
+
+            if isinstance(op_objs[-1], ghidra.program.model.scalar.Scalar):
+                op_off = op_objs[-1].getValue()
+            else:
+                op_off = 0
+
+            yield Offset(op_off), ih.address
+            yield OperandOffset(i, op_off), 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
+    """
+    for addr in capa.features.extractors.ghidra.helpers.find_data_references_from_insn(ih.inner):
+        data = getDataAt(addr)  # type: ignore [name-defined] # noqa: F821
+        if data and not data.hasStringValue():
+            extracted_bytes = capa.features.extractors.ghidra.helpers.get_bytes(addr, MAX_BYTES_FEATURE_SIZE)
+            if extracted_bytes and not capa.features.extractors.helpers.all_zeros(extracted_bytes):
+                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  > "
+    """
+    for addr in capa.features.extractors.ghidra.helpers.find_data_references_from_insn(ih.inner):
+        data = getDataAt(addr)  # type: ignore [name-defined] # noqa: F821
+        if data and data.hasStringValue():
+            yield String(data.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 == OPERAND_TYPE_DYNAMIC_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 Mandiant, 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
@@ -41,35 +41,64 @@ def is_ordinal(symbol: str) -> bool:
     return False
 
 
-def generate_symbols(dll: str, symbol: str) -> Iterator[str]:
+def generate_symbols(dll: str, symbol: str, include_dll=False) -> Iterator[str]:
     """
     for a given dll and symbol name, generate variants.
     we over-generate features to make matching easier.
     these include:
-      - kernel32.CreateFileA
-      - kernel32.CreateFile
       - CreateFileA
       - CreateFile
+      - ws2_32.#1
+
+    note that since capa v7 only `import` features and APIs called via ordinal include DLL names:
+      - kernel32.CreateFileA
+      - kernel32.CreateFile
+      - ws2_32.#1
+
+    for `api` features dll names are good for documentation but not used during matching
     """
     # normalize dll name
     dll = dll.lower()
 
-    # kernel32.CreateFileA
-    yield "%s.%s" % (dll, symbol)
+    # trim extensions observed in dynamic traces
+    dll = dll[0:-4] if dll.endswith(".dll") else dll
+    dll = dll[0:-4] if dll.endswith(".drv") else dll
+
+    if include_dll or is_ordinal(symbol):
+        # ws2_32.#1
+        # kernel32.CreateFileA
+        yield f"{dll}.{symbol}"
 
     if not is_ordinal(symbol):
         # CreateFileA
         yield symbol
 
-    if is_aw_function(symbol):
-        # kernel32.CreateFile
-        yield "%s.%s" % (dll, symbol[:-1])
+        if is_aw_function(symbol):
+            if include_dll:
+                # kernel32.CreateFile
+                yield f"{dll}.{symbol[:-1]}"
 
-        if not is_ordinal(symbol):
             # CreateFile
             yield symbol[:-1]
 
 
+def reformat_forwarded_export_name(forwarded_name: str) -> str:
+    """
+    a forwarded export has a DLL name/path and 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))
 
@@ -112,7 +141,6 @@ def carve_pe(pbytes: bytes, offset: int = 0) -> Iterator[Tuple[int, int]]:
     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

capa/features/extractors/ida/basicblock.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 Mandiant, 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
@@ -34,7 +34,7 @@ def get_printable_len(op: idaapi.op_t) -> int:
     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_: bytes):
         return all(c < 127 and chr(c) in string.printable for c in chars_)
@@ -95,7 +95,7 @@ def extract_bb_tight_loop(fh: FunctionHandle, bbh: BBHandle) -> Iterator[Tuple[F
 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):
+        for feature, addr in bb_handler(fh, bbh):
             yield feature, addr
     yield BasicBlock(), bbh.address
 
@@ -104,19 +104,3 @@ BASIC_BLOCK_HANDLERS = (
     extract_bb_tight_loop,
     extract_bb_stackstring,
 )
-
-
-def main():
-    features = []
-    for fhandle in helpers.get_functions(skip_thunks=True, skip_libs=True):
-        f: idaapi.func_t = fhandle.inner
-        for bb in idaapi.FlowChart(f, flags=idaapi.FC_PREDS):
-            features.extend(list(extract_features(fhandle, bb)))
-
-    import pprint
-
-    pprint.pprint(features)
-
-
-if __name__ == "__main__":
-    main()

capa/features/extractors/ida/extractor.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 Mandiant, 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
@@ -8,6 +8,7 @@
 from typing import List, Tuple, Iterator
 
 import idaapi
+import ida_nalt
 
 import capa.ida.helpers
 import capa.features.extractors.elf
@@ -18,13 +19,24 @@ 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
+from capa.features.extractors.base_extractor import (
+    BBHandle,
+    InsnHandle,
+    SampleHashes,
+    FunctionHandle,
+    StaticFeatureExtractor,
+)
 
 
-class IdaFeatureExtractor(FeatureExtractor):
+class IdaFeatureExtractor(StaticFeatureExtractor):
     def __init__(self):
-        super(IdaFeatureExtractor, self).__init__()
+        super().__init__(
+            hashes=SampleHashes(
+                md5=ida_nalt.retrieve_input_file_md5(), sha1="(unknown)", sha256=ida_nalt.retrieve_input_file_sha256()
+            )
+        )
         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())
 

capa/features/extractors/ida/file.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 Mandiant, 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
@@ -12,6 +12,7 @@ from typing import Tuple, Iterator
 import idc
 import idaapi
 import idautils
+import ida_entry
 
 import capa.features.extractors.common
 import capa.features.extractors.helpers
@@ -21,12 +22,14 @@ 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: 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
+    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: idaapi.segment_t) -> Iterator[Tuple[int, int]]:
     ]
 
     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):
@@ -54,6 +58,10 @@ def check_segment_for_pe(seg: idaapi.segment_t) -> Iterator[Tuple[int, int]]:
 
         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
@@ -61,9 +69,6 @@ def check_segment_for_pe(seg: idaapi.segment_t) -> Iterator[Tuple[int, int]]:
         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))
-
 
 def extract_file_embedded_pe() -> Iterator[Tuple[Feature, Address]]:
     """extract embedded PE features
@@ -73,14 +78,20 @@ def extract_file_embedded_pe() -> Iterator[Tuple[Feature, Address]]:
         - 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):
+        for ea, _ in check_segment_for_pe(seg):
             yield Characteristic("embedded pe"), FileOffsetAddress(ea)
 
 
 def extract_file_export_names() -> Iterator[Tuple[Feature, Address]]:
     """extract function exports"""
-    for (_, _, ea, name) in idautils.Entries():
-        yield Export(name), AbsoluteVirtualAddress(ea)
+    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() -> Iterator[Tuple[Feature, Address]]:
@@ -94,27 +105,30 @@ def extract_file_import_names() -> Iterator[Tuple[Feature, Address]]:
      - 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]):
+            for name in capa.features.extractors.helpers.generate_symbols(info[0], info[1], include_dll=True):
                 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):
+        for name in capa.features.extractors.helpers.generate_symbols(dll, symbol, include_dll=True):
             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() -> Iterator[Tuple[Feature, Address]]:
     """extract section names
@@ -165,7 +179,7 @@ def extract_file_function_names() -> Iterator[Tuple[Feature, Address]]:
 def extract_file_format() -> Iterator[Tuple[Feature, Address]]:
     file_info = idaapi.get_inf_structure()
 
-    if file_info.filetype == idaapi.f_PE:
+    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
@@ -173,7 +187,7 @@ def extract_file_format() -> Iterator[Tuple[Feature, Address]]:
         # no file type to return when processing a binary file, but we want to continue processing
         return
     else:
-        raise NotImplementedError("file format: %d" % file_info.filetype)
+        raise NotImplementedError(f"unexpected file format: {file_info.filetype}")
 
 
 def extract_features() -> Iterator[Tuple[Feature, Address]]:
@@ -192,14 +206,3 @@ FILE_HANDLERS = (
     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

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 Mandiant, 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
@@ -45,23 +45,8 @@ def extract_recursive_call(fh: FunctionHandle):
 
 def extract_features(fh: FunctionHandle) -> Iterator[Tuple[Feature, Address]]:
     for func_handler in FUNCTION_HANDLERS:
-        for (feature, addr) in func_handler(fh):
+        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.ida.helpers.get_functions(skip_thunks=True, skip_libs=True):
-        features.extend(list(extract_features(fhandle)))
-
-    import pprint
-
-    pprint.pprint(features)
-
-
-if __name__ == "__main__":
-    main()

capa/features/extractors/ida/global_.py

@@ -1,3 +1,10 @@
+# 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
@@ -31,7 +38,7 @@ def extract_os() -> Iterator[Tuple[Feature, Address]]:
         #  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 futher CLI argument to specify the OS,
+        # 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.
         #

capa/features/extractors/ida/helpers.py

@@ -1,16 +1,18 @@
-# Copyright (C) 2020 Mandiant, 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 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
@@ -24,9 +26,10 @@ def find_byte_sequence(start: int, end: int, seq: bytes) -> Iterator[int]:
         end: max virtual address
         seq: bytes to search e.g. b"\x01\x03"
     """
-    seqstr = " ".join(["%02x" % b for b in seq])
+    seqstr = " ".join([f"{b:02x}" for b in seq])
     while True:
-        # TODO find_binary: Deprecated. Please use ida_bytes.bin_search() instead.
+        # 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
@@ -35,7 +38,7 @@ def find_byte_sequence(start: int, end: int, seq: bytes) -> Iterator[int]:
 
 
 def get_functions(
-    start: int = None, end: int = None, skip_thunks: bool = False, skip_libs: bool = False
+    start: Optional[int] = None, end: Optional[int] = None, skip_thunks: bool = False, skip_libs: bool = False
 ) -> Iterator[FunctionHandle]:
     """get functions, range optional
 
@@ -79,9 +82,22 @@ def get_segment_buffer(seg: idaapi.segment_t) -> bytes:
     return buff if buff else b""
 
 
+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 = {}
+    imports: Dict[int, Tuple[str, str, int]] = {}
 
     for idx in range(idaapi.get_import_module_qty()):
         library = idaapi.get_import_module_name(idx)
@@ -89,26 +105,32 @@ def get_file_imports() -> Dict[int, Tuple[str, str, int]]:
         if not library:
             continue
 
-        # IDA uses section names for the library of ELF imports, like ".dynsym"
-        library = library.lstrip(".")
+        # 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 = ""
 
-        def inspect_import(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
-
-        idaapi.enum_import_names(idx, inspect_import)
+        cb = functools.partial(inspect_import, imports, library)
+        idaapi.enum_import_names(idx, cb)
 
     return imports
 
 
+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
 
@@ -183,7 +205,7 @@ def read_bytes_at(ea: int, count: int) -> bytes:
 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
@@ -207,7 +229,8 @@ def get_op_phrase_info(op: idaapi.op_t) -> Dict:
         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
@@ -252,20 +275,18 @@ def is_op_offset(insn: idaapi.insn_t, op: idaapi.op_t) -> bool:
 
 def is_sp_modified(insn: idaapi.insn_t) -> bool:
     """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
+    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: idaapi.insn_t) -> bool:
     """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
+    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: int) -> bool:
@@ -273,7 +294,7 @@ def is_frame_register(reg: int) -> bool:
     return reg in (idautils.procregs.sp.reg, idautils.procregs.bp.reg)
 
 
-def get_insn_ops(insn: idaapi.insn_t, target_ops: Tuple[Any] = None) -> idaapi.op_t:
+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:
@@ -311,10 +332,7 @@ def mask_op_val(op: idaapi.op_t) -> int:
 
 def is_function_recursive(f: idaapi.func_t) -> bool:
     """check if function is recursive"""
-    for ref in idautils.CodeRefsTo(f.start_ea, True):
-        if f.contains(ref):
-            return True
-    return False
+    return any(f.contains(ref) for ref in idautils.CodeRefsTo(f.start_ea, True))
 
 
 def is_basic_block_tight_loop(bb: idaapi.BasicBlock) -> bool:
@@ -363,8 +381,7 @@ def find_data_reference_from_insn(insn: idaapi.insn_t, max_depth: int = 10) -> i
 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: idaapi.BasicBlock) -> bool:

capa/features/extractors/ida/insn.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 Mandiant, 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
@@ -23,13 +23,19 @@ from capa.features.extractors.base_extractor import BBHandle, InsnHandle, Functi
 SECURITY_COOKIE_BYTES_DELTA = 0x40
 
 
-def get_imports(ctx: Dict[str, Any]) -> Dict[str, Any]:
+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: Dict[str, Any], insn: idaapi.insn_t) -> Iterator[str]:
+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
@@ -46,7 +52,7 @@ def check_for_api_call(ctx: Dict[str, Any], insn: idaapi.insn_t) -> Iterator[str
             except IndexError:
                 break
 
-        info = get_imports(ctx).get(ref, ())
+        info = funcs.get(ref, ())
         if info:
             break
 
@@ -55,7 +61,7 @@ def check_for_api_call(ctx: Dict[str, Any], insn: idaapi.insn_t) -> Iterator[str
             break
 
     if info:
-        yield "%s.%s" % (info[0], info[1])
+        yield info
 
 
 def extract_insn_api_features(fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
@@ -67,14 +73,20 @@ def extract_insn_api_features(fh: FunctionHandle, bbh: BBHandle, ih: InsnHandle)
     """
     insn: idaapi.insn_t = ih.inner
 
-    if not insn.get_canon_mnem() in ("call", "jmp"):
+    if insn.get_canon_mnem() not in ("call", "jmp"):
         return
 
-    for api in check_for_api_call(fh.ctx, insn):
-        dll, _, symbol = api.rpartition(".")
-        for name in capa.features.extractors.helpers.generate_symbols(dll, symbol):
+    # 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:
@@ -160,7 +172,9 @@ def extract_insn_bytes_features(fh: FunctionHandle, bbh: BBHandle, ih: InsnHandl
     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), ih.address
+            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(
@@ -201,7 +215,11 @@ def extract_insn_offset_features(
             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]
@@ -380,14 +398,16 @@ def extract_insn_peb_access_characteristic_features(
     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
+        # 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
 
 
@@ -401,18 +421,22 @@ def extract_insn_segment_access_features(
     """
     insn: idaapi.insn_t = ih.inner
 
-    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:" in disasm:
-        # TODO: replace above with proper IDA
+        # 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
+        # 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
 
 
@@ -423,7 +447,7 @@ def extract_insn_cross_section_cflow(
     insn: idaapi.insn_t = ih.inner
 
     for ref in idautils.CodeRefsFrom(insn.ea, False):
-        if ref in get_imports(fh.ctx).keys():
+        if ref in get_imports(fh.ctx):
             # ignore API calls
             continue
         if not idaapi.getseg(ref):
@@ -464,7 +488,7 @@ def extract_function_indirect_call_characteristic_features(
 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):
+        for feature, ea in inst_handler(f, bbh, insn):
             yield feature, ea
 
 
@@ -483,20 +507,3 @@ INSTRUCTION_HANDLERS = (
     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 Mandiant, 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

capa/features/extractors/null.py

@@ -1,9 +1,28 @@
-from typing import Dict, List, Tuple
+# 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, Union
 from dataclasses import dataclass
 
+from typing_extensions import TypeAlias
+
 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
+from capa.features.address import NO_ADDRESS, Address, ThreadAddress, ProcessAddress, DynamicCallAddress
+from capa.features.extractors.base_extractor import (
+    BBHandle,
+    CallHandle,
+    InsnHandle,
+    SampleHashes,
+    ThreadHandle,
+    ProcessHandle,
+    FunctionHandle,
+    StaticFeatureExtractor,
+    DynamicFeatureExtractor,
+)
 
 
 @dataclass
@@ -24,7 +43,7 @@ class FunctionFeatures:
 
 
 @dataclass
-class NullFeatureExtractor(FeatureExtractor):
+class NullStaticFeatureExtractor(StaticFeatureExtractor):
     """
     An extractor that extracts some user-provided features.
 
@@ -32,6 +51,7 @@ class NullFeatureExtractor(FeatureExtractor):
     """
 
     base_address: Address
+    sample_hashes: SampleHashes
     global_features: List[Feature]
     file_features: List[Tuple[Address, Feature]]
     functions: Dict[Address, FunctionFeatures]
@@ -39,6 +59,9 @@ class NullFeatureExtractor(FeatureExtractor):
     def get_base_address(self):
         return self.base_address
 
+    def get_sample_hashes(self) -> SampleHashes:
+        return self.sample_hashes
+
     def extract_global_features(self):
         for feature in self.global_features:
             yield feature, NO_ADDRESS
@@ -52,26 +75,96 @@ class NullFeatureExtractor(FeatureExtractor):
             yield FunctionHandle(address, None)
 
     def extract_function_features(self, f):
-        for address, feature in self.functions.get(f.address, {}).features:
+        for address, feature in self.functions[f.address].features:
             yield feature, address
 
     def get_basic_blocks(self, f):
-        for address in sorted(self.functions.get(f.address, {}).basic_blocks.keys()):
+        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.get(f.address, {}).basic_blocks.get(bb.address, {}).features:
+        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.get(f.address, {}).basic_blocks.get(bb.address, {}).instructions.keys()):
+        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.get(f.address, {})
-            .basic_blocks.get(bb.address, {})
-            .instructions.get(insn.address, {})
-            .features
-        ):
+        for address, feature in self.functions[f.address].basic_blocks[bb.address].instructions[insn.address].features:
             yield feature, address
+
+
+@dataclass
+class CallFeatures:
+    name: str
+    features: List[Tuple[Address, Feature]]
+
+
+@dataclass
+class ThreadFeatures:
+    features: List[Tuple[Address, Feature]]
+    calls: Dict[Address, CallFeatures]
+
+
+@dataclass
+class ProcessFeatures:
+    features: List[Tuple[Address, Feature]]
+    threads: Dict[Address, ThreadFeatures]
+    name: str
+
+
+@dataclass
+class NullDynamicFeatureExtractor(DynamicFeatureExtractor):
+    base_address: Address
+    sample_hashes: SampleHashes
+    global_features: List[Feature]
+    file_features: List[Tuple[Address, Feature]]
+    processes: Dict[Address, ProcessFeatures]
+
+    def extract_global_features(self):
+        for feature in self.global_features:
+            yield feature, NO_ADDRESS
+
+    def get_sample_hashes(self) -> SampleHashes:
+        return self.sample_hashes
+
+    def extract_file_features(self):
+        for address, feature in self.file_features:
+            yield feature, address
+
+    def get_processes(self):
+        for address in sorted(self.processes.keys()):
+            assert isinstance(address, ProcessAddress)
+            yield ProcessHandle(address=address, inner={})
+
+    def extract_process_features(self, ph):
+        for addr, feature in self.processes[ph.address].features:
+            yield feature, addr
+
+    def get_process_name(self, ph) -> str:
+        return self.processes[ph.address].name
+
+    def get_threads(self, ph):
+        for address in sorted(self.processes[ph.address].threads.keys()):
+            assert isinstance(address, ThreadAddress)
+            yield ThreadHandle(address=address, inner={})
+
+    def extract_thread_features(self, ph, th):
+        for addr, feature in self.processes[ph.address].threads[th.address].features:
+            yield feature, addr
+
+    def get_calls(self, ph, th):
+        for address in sorted(self.processes[ph.address].threads[th.address].calls.keys()):
+            assert isinstance(address, DynamicCallAddress)
+            yield CallHandle(address=address, inner={})
+
+    def extract_call_features(self, ph, th, ch):
+        for address, feature in self.processes[ph.address].threads[th.address].calls[ch.address].features:
+            yield feature, address
+
+    def get_call_name(self, ph, th, ch) -> str:
+        return self.processes[ph.address].threads[th.address].calls[ch.address].name
+
+
+NullFeatureExtractor: TypeAlias = Union[NullStaticFeatureExtractor, NullDynamicFeatureExtractor]

capa/features/extractors/pefile.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 Mandiant, 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
@@ -7,6 +7,7 @@
 # See the License for the specific language governing permissions and limitations under the License.
 
 import logging
+from pathlib import Path
 
 import pefile
 
@@ -18,7 +19,7 @@ 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
+from capa.features.extractors.base_extractor import SampleHashes, StaticFeatureExtractor
 
 logger = logging.getLogger(__name__)
 
@@ -39,8 +40,20 @@ def extract_file_export_names(pe, **kwargs):
                 name = export.name.partition(b"\x00")[0].decode("ascii")
             except UnicodeDecodeError:
                 continue
-            va = base_address + export.address
-            yield Export(name), AbsoluteVirtualAddress(va)
+
+            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):
@@ -64,14 +77,14 @@ def extract_file_import_names(pe, **kwargs):
 
             for imp in dll.imports:
                 if imp.import_by_ordinal:
-                    impname = "#%s" % imp.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):
+                for name in capa.features.extractors.helpers.generate_symbols(modname, impname, include_dll=True):
                     yield Import(name), AbsoluteVirtualAddress(imp.address)
 
 
@@ -133,7 +146,8 @@ def extract_file_features(pe, buf):
     """
 
     for file_handler in FILE_HANDLERS:
-        for feature, va in file_handler(pe=pe, buf=buf):
+        # file_handler: type: (pe, bytes) -> Iterable[Tuple[Feature, Address]]
+        for feature, va in file_handler(pe=pe, buf=buf):  # type: ignore
             yield feature, va
 
 
@@ -160,7 +174,8 @@ def extract_global_features(pe, buf):
       Tuple[Feature, VA]: a feature and its location.
     """
     for handler in GLOBAL_HANDLERS:
-        for feature, va in handler(pe=pe, buf=buf):
+        # file_handler: type: (pe, bytes) -> Iterable[Tuple[Feature, Address]]
+        for feature, va in handler(pe=pe, buf=buf):  # type: ignore
             yield feature, va
 
 
@@ -170,24 +185,22 @@ GLOBAL_HANDLERS = (
 )
 
 
-class PefileFeatureExtractor(FeatureExtractor):
-    def __init__(self, path: str):
-        super(PefileFeatureExtractor, self).__init__()
-        self.path = path
-        self.pe = pefile.PE(path)
+class PefileFeatureExtractor(StaticFeatureExtractor):
+    def __init__(self, path: Path):
+        super().__init__(hashes=SampleHashes.from_bytes(path.read_bytes()))
+        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):
-        with open(self.path, "rb") as f:
-            buf = f.read()
+        buf = Path(self.path).read_bytes()
 
         yield from extract_global_features(self.pe, buf)
 
     def extract_file_features(self):
-        with open(self.path, "rb") as f:
-            buf = f.read()
+        buf = Path(self.path).read_bytes()
 
         yield from extract_file_features(self.pe, buf)
 

capa/features/extractors/smda/basicblock.py

@@ -1,133 +0,0 @@
-import string
-import struct
-from typing import Tuple, Iterator
-
-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 _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: FunctionHandle, bb: BBHandle) -> Iterator[Tuple[Feature, Address]]:
-    """check basic block for tight loop indicators"""
-    if _bb_has_tight_loop(f.inner, bb.inner):
-        yield Characteristic("tight loop"), bb.address
-
-
-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: FunctionHandle, bb: BBHandle) -> Iterator[Tuple[Feature, Address]]:
-    """check basic block for stackstring indicators"""
-    if _bb_has_stackstring(f.inner, bb.inner):
-        yield Characteristic("stack string"), bb.address
-
-
-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: FunctionHandle, bb: BBHandle) -> Iterator[Tuple[Feature, Address]]:
-    """
-    extract features from the given basic block.
-
-    args:
-      f: the function from which to extract features
-      bb: the basic block to process.
-
-    yields:
-      Tuple[Feature, Address]: the features and their location found in this basic block.
-    """
-    yield BasicBlock(), bb.address
-    for bb_handler in BASIC_BLOCK_HANDLERS:
-        for feature, addr in bb_handler(f, bb):
-            yield feature, addr
-
-
-BASIC_BLOCK_HANDLERS = (
-    extract_bb_tight_loop,
-    extract_stackstring,
-)

capa/features/extractors/smda/extractor.py

@@ -1,57 +0,0 @@
-from typing import List, Tuple
-
-from smda.common.SmdaReport import SmdaReport
-
-import capa.features.extractors.common
-import capa.features.extractors.smda.file
-import capa.features.extractors.smda.insn
-import capa.features.extractors.smda.global_
-import capa.features.extractors.smda.function
-import capa.features.extractors.smda.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 SmdaFeatureExtractor(FeatureExtractor):
-    def __init__(self, smda_report: SmdaReport, path):
-        super(SmdaFeatureExtractor, self).__init__()
-        self.smda_report = smda_report
-        self.path = path
-        with open(self.path, "rb") as f:
-            self.buf = f.read()
-
-        # 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.common.extract_os(self.buf))
-        self.global_features.extend(capa.features.extractors.smda.global_.extract_arch(self.smda_report))
-
-    def get_base_address(self):
-        return AbsoluteVirtualAddress(self.smda_report.base_addr)
-
-    def extract_global_features(self):
-        yield from self.global_features
-
-    def extract_file_features(self):
-        yield from capa.features.extractors.smda.file.extract_features(self.smda_report, self.buf)
-
-    def get_functions(self):
-        for function in self.smda_report.getFunctions():
-            yield FunctionHandle(address=AbsoluteVirtualAddress(function.offset), inner=function)
-
-    def extract_function_features(self, fh):
-        yield from capa.features.extractors.smda.function.extract_features(fh)
-
-    def get_basic_blocks(self, fh):
-        for bb in fh.inner.getBlocks():
-            yield BBHandle(address=AbsoluteVirtualAddress(bb.offset), inner=bb)
-
-    def extract_basic_block_features(self, fh, bbh):
-        yield from capa.features.extractors.smda.basicblock.extract_features(fh, bbh)
-
-    def get_instructions(self, fh, bbh):
-        for smda_ins in bbh.inner.getInstructions():
-            yield InsnHandle(address=AbsoluteVirtualAddress(smda_ins.offset), inner=smda_ins)
-
-    def extract_insn_features(self, fh, bbh, ih):
-        yield from capa.features.extractors.smda.insn.extract_features(fh, bbh, ih)

capa/features/extractors/smda/file.py

@@ -1,103 +0,0 @@
-# if we have SMDA we definitely have lief
-import lief
-
-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 String, Characteristic
-from capa.features.address import FileOffsetAddress, AbsoluteVirtualAddress
-
-
-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(buf, **kwargs):
-    lief_binary = lief.parse(buf)
-
-    if lief_binary is not None:
-        for function in lief_binary.exported_functions:
-            yield Export(function.name), AbsoluteVirtualAddress(function.address)
-
-
-def extract_file_import_names(smda_report, buf):
-    # extract import table info via LIEF
-    lief_binary = lief.parse(buf)
-    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:
-            va = func.iat_address + smda_report.base_addr
-            if func.name:
-                for name in capa.features.extractors.helpers.generate_symbols(library_name, func.name):
-                    yield Import(name), AbsoluteVirtualAddress(va)
-            elif func.is_ordinal:
-                for name in capa.features.extractors.helpers.generate_symbols(library_name, "#%s" % func.ordinal):
-                    yield Import(name), AbsoluteVirtualAddress(va)
-
-
-def extract_file_section_names(buf, **kwargs):
-    lief_binary = lief.parse(buf)
-    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), AbsoluteVirtualAddress(base_address + section.virtual_address)
-
-
-def extract_file_strings(buf, **kwargs):
-    """
-    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_file_function_names(smda_report, **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("SMDA doesn't have library matching")
-    return
-
-
-def extract_file_format(buf, **kwargs):
-    yield from capa.features.extractors.common.extract_format(buf)
-
-
-def extract_features(smda_report, buf):
-    """
-    extract file features from given workspace
-
-    args:
-      smda_report (smda.common.SmdaReport): a SmdaReport
-      buf: the raw bytes of the sample
-
-    yields:
-      Tuple[Feature, VA]: a feature and its location.
-    """
-
-    for file_handler in FILE_HANDLERS:
-        for feature, addr in file_handler(smda_report=smda_report, buf=buf):
-            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,
-)

capa/features/extractors/smda/function.py

@@ -1,42 +0,0 @@
-from typing import Tuple, Iterator
-
-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: FunctionHandle) -> Iterator[Tuple[Feature, Address]]:
-    for inref in f.inner.inrefs:
-        yield Characteristic("calls to"), AbsoluteVirtualAddress(inref)
-
-
-def extract_function_loop(f: FunctionHandle) -> Iterator[Tuple[Feature, Address]]:
-    """
-    parse if a function has a loop
-    """
-    edges = []
-    for bb_from, bb_tos in f.inner.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.address
-
-
-def extract_features(f: FunctionHandle):
-    """
-    extract features from the given function.
-
-    args:
-      f: the function from which to extract features
-
-    yields:
-      Tuple[Feature, Address]: the features and their location found in this function.
-    """
-    for func_handler in FUNCTION_HANDLERS:
-        for feature, addr in func_handler(f):
-            yield feature, addr
-
-
-FUNCTION_HANDLERS = (extract_function_calls_to, extract_function_loop)

capa/features/extractors/smda/global_.py

@@ -1,21 +0,0 @@
-import logging
-
-from capa.features.common import ARCH_I386, ARCH_AMD64, Arch
-from capa.features.address import NO_ADDRESS
-
-logger = logging.getLogger(__name__)
-
-
-def extract_arch(smda_report):
-    if smda_report.architecture == "intel":
-        if smda_report.bitness == 32:
-            yield Arch(ARCH_I386), NO_ADDRESS
-        elif smda_report.bitness == 64:
-            yield Arch(ARCH_AMD64), 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", smda_report.architecture)
-        return

capa/features/extractors/smda/insn.py

@@ -1,455 +0,0 @@
-import re
-import string
-import struct
-from typing import Tuple, Iterator
-
-import smda
-
-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, 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
-PATTERN_HEXNUM = re.compile(r"[+\-] (?P<num>0x[a-fA-F0-9]+)")
-PATTERN_SINGLENUM = re.compile(r"[+\-] (?P<num>[0-9])")
-
-
-def extract_insn_api_features(fh: FunctionHandle, bb, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
-    """parse API features from the given instruction."""
-    f: smda.Function = fh.inner
-    insn: smda.Insn = ih.inner
-
-    if ih.address in f.apirefs:
-        api_entry = f.apirefs[ih.address]
-        # 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), ih.address
-    elif ih.address 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), ih.address
-                    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(fh: FunctionHandle, bb, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
-    """parse number features from the given instruction."""
-    # example:
-    #
-    #     push    3136B0h         ; dwControlCode
-    f: smda.Function = fh.inner
-    insn: smda.Insn = ih.inner
-
-    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 i, operand in enumerate(operands):
-        try:
-            # The result of bitwise operations is calculated as though carried out
-            # in two’s complement with an infinite number of sign bits
-            value = int(operand, 16) & ((1 << f.smda_report.bitness) - 1)
-        except ValueError:
-            continue
-        else:
-            yield Number(value), ih.address
-            yield OperandNumber(i, value), ih.address
-
-            if insn.mnemonic == "add" and 0 < value < MAX_STRUCTURE_SIZE:
-                # for pattern like:
-                #
-                #     add eax, 0x10
-                #
-                # assume 0x10 is also an offset (imagine eax is a pointer).
-                yield Offset(value), ih.address
-                yield OperandOffset(i, value), ih.address
-
-
-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:
-        raise ValueError("buffer is empty")
-    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(fh: FunctionHandle, bb, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
-    """
-    parse byte sequence features from the given instruction.
-    example:
-        #     push    offset iid_004118d4_IShellLinkA ; riid
-    """
-    f: smda.Function = fh.inner
-    insn: smda.Insn = ih.inner
-
-    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), ih.address
-
-
-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(fh: FunctionHandle, bb, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
-    """parse string features from the given instruction."""
-    # example:
-    #
-    #     push    offset aAcr     ; "ACR  > "
-    f: smda.Function = fh.inner
-    insn: smda.Insn = ih.inner
-
-    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")), ih.address
-
-
-def extract_insn_offset_features(f, bb, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
-    """parse structure offset features from the given instruction."""
-    # examples:
-    #
-    #     mov eax, [esi + 4]
-    #     mov eax, [esi + ecx + 16384]
-    insn: smda.Insn = ih.inner
-
-    operands = [o.strip() for o in insn.operands.split(",")]
-    for i, operand in enumerate(operands):
-        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
-
-        if "ptr" not in operand:
-            if (
-                insn.mnemonic == "lea"
-                and i == 1
-                and (operand.count("+") + operand.count("-")) == 1
-                and operand.count("*") == 0
-            ):
-                # for pattern like:
-                #
-                #     lea eax, [ebx + 1]
-                #
-                # assume 1 is also an offset (imagine ebx is a zero register).
-                yield Number(number), ih.address
-                yield OperandNumber(i, number), ih.address
-
-            continue
-
-        yield Offset(number), ih.address
-        yield OperandOffset(i, number), ih.address
-
-
-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(
-    fh: FunctionHandle, bh: 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.
-    """
-    f: smda.Function = fh.inner
-    bb: smda.BasicBlock = bh.inner
-    insn: smda.Insn = ih.inner
-
-    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"), ih.address
-
-
-def extract_insn_mnemonic_features(f, bb, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
-    """parse mnemonic features from the given instruction."""
-    yield Mnemonic(ih.inner.mnemonic), ih.address
-
-
-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: smda.Insn = ih.inner
-
-    if insn.mnemonic != "call":
-        return
-
-    if not insn.operands.startswith("0x"):
-        return
-
-    if int(insn.operands, 16) == insn.offset + 5:
-        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
-    """
-    insn: smda.Insn = ih.inner
-
-    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"), ih.address
-        elif "gs:" in operand and "0x60" in operand:
-            yield Characteristic("peb access"), ih.address
-
-
-def extract_insn_segment_access_features(f, bb, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
-    """parse the instruction for access to fs or gs"""
-    insn: smda.Insn = ih.inner
-
-    operands = [o.strip() for o in insn.operands.split(",")]
-    for operand in operands:
-        if "fs:" in operand:
-            yield Characteristic("fs access"), ih.address
-        elif "gs:" in operand:
-            yield Characteristic("gs access"), ih.address
-
-
-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.
-    """
-    f: smda.Function = fh.inner
-    insn: smda.Insn = ih.inner
-
-    if insn.mnemonic in ["call", "jmp"]:
-        if ih.address in f.apirefs:
-            return
-
-        smda_report = insn.smda_function.smda_report
-        if ih.address in f.outrefs:
-            for target in f.outrefs[ih.address]:
-                if smda_report.getSection(ih.address) != smda_report.getSection(target):
-                    yield Characteristic("cross section flow"), ih.address
-        elif insn.operands.startswith("0x"):
-            target = int(insn.operands, 16)
-            if smda_report.getSection(ih.address) != smda_report.getSection(target):
-                yield Characteristic("cross section flow"), ih.address
-
-
-# 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(fh: FunctionHandle, bb, ih: InsnHandle) -> Iterator[Tuple[Feature, Address]]:
-    f: smda.Function = fh.inner
-    insn: smda.Insn = ih.inner
-
-    if insn.mnemonic != "call":
-        return
-
-    if ih.address in f.outrefs:
-        for outref in f.outrefs[ih.address]:
-            yield Characteristic("calls from"), AbsoluteVirtualAddress(outref)
-
-            if outref == f.offset:
-                # if we found a jump target and it's the function address
-                # mark as recursive
-                yield Characteristic("recursive call"), AbsoluteVirtualAddress(outref)
-    if ih.address in f.apirefs:
-        yield Characteristic("calls from"), ih.address
-
-
-# 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, 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: smda.Insn = ih.inner
-
-    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"), ih.address
-
-
-def extract_features(f, bb, insn):
-    """
-    extract features from the given insn.
-
-    args:
-      f: the function to process.
-      bb: the basic block to process.
-      insn: the instruction to process.
-
-    yields:
-      Tuple[Feature, Address]: the features and their location found in this insn.
-    """
-    for insn_handler in INSTRUCTION_HANDLERS:
-        for feature, addr in insn_handler(f, bb, insn):
-            yield feature, addr
-
-
-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_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/strings.py

@@ -1,6 +1,6 @@
 # strings code from FLOSS, https://github.com/mandiant/flare-floss
 #
-# Copyright (C) 2020 Mandiant, 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/basicblock.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 Mandiant, 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
@@ -31,7 +31,7 @@ def interface_extract_basic_block_XXX(f: FunctionHandle, bb: BBHandle) -> Iterat
     yields:
       (Feature, Address): the feature and the address at which its found.
     """
-    ...
+    raise NotImplementedError
 
 
 def _bb_has_tight_loop(f, bb):
@@ -92,7 +92,6 @@ def is_mov_imm_to_stack(instr: envi.archs.i386.disasm.i386Opcode) -> bool:
     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
     ):
@@ -121,7 +120,7 @@ def get_printable_len(oper: envi.archs.i386.disasm.i386ImmOper) -> int:
     elif oper.tsize == 8:
         chars = struct.pack("<Q", oper.imm)
     else:
-        raise ValueError("unexpected oper.tsize: %d" % (oper.tsize))
+        raise ValueError(f"unexpected oper.tsize: {oper.tsize}")
 
     if is_printable_ascii(chars):
         return oper.tsize
@@ -141,7 +140,7 @@ def is_printable_ascii(chars: bytes) -> bool:
 
 
 def is_printable_utf16le(chars: bytes) -> bool:
-    if all(c == b"\x00" for c in chars[1::2]):
+    if all(c == 0x0 for c in chars[1::2]):
         return is_printable_ascii(chars[::2])
     return False
 

capa/features/extractors/viv/extractor.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 Mandiant, 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 +6,8 @@
 #  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 List, Tuple, Iterator
+from typing import Any, Dict, List, Tuple, Iterator
+from pathlib import Path
 
 import viv_utils
 import viv_utils.flirt
@@ -19,22 +20,28 @@ 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
+from capa.features.extractors.base_extractor import (
+    BBHandle,
+    InsnHandle,
+    SampleHashes,
+    FunctionHandle,
+    StaticFeatureExtractor,
+)
 
 logger = logging.getLogger(__name__)
 
 
-class VivisectFeatureExtractor(FeatureExtractor):
-    def __init__(self, vw, path):
-        super(VivisectFeatureExtractor, self).__init__()
+class VivisectFeatureExtractor(StaticFeatureExtractor):
+    def __init__(self, vw, path: Path, os):
         self.vw = vw
         self.path = path
-        with open(self.path, "rb") as f:
-            self.buf = f.read()
+        self.buf = path.read_bytes()
+        super().__init__(hashes=SampleHashes.from_bytes(self.buf))
 
         # 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.common.extract_os(self.buf))
+        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):
@@ -48,8 +55,11 @@ class VivisectFeatureExtractor(FeatureExtractor):
         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))
+            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)

capa/features/extractors/viv/file.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 Mandiant, 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
@@ -8,6 +8,7 @@
 from typing import Tuple, Iterator
 
 import PE.carve as pe_carve  # vivisect PE
+import vivisect
 import viv_utils
 import viv_utils.flirt
 
@@ -16,7 +17,7 @@ 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 String, Feature, Characteristic
+from capa.features.common import Feature, Characteristic
 from capa.features.address import Address, FileOffsetAddress, AbsoluteVirtualAddress
 
 
@@ -25,10 +26,35 @@ def extract_file_embedded_pe(buf, **kwargs) -> Iterator[Tuple[Feature, Address]]
         yield Characteristic("embedded pe"), FileOffsetAddress(offset)
 
 
-def extract_file_export_names(vw, **kwargs) -> Iterator[Tuple[Feature, Address]]:
+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_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]]:
     """
@@ -44,10 +70,10 @@ def extract_file_import_names(vw, **kwargs) -> Iterator[Tuple[Feature, Address]]
         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):
+        for name in capa.features.extractors.helpers.generate_symbols(modname, impname, include_dll=True):
             yield Import(name), addr
 
 

capa/features/extractors/viv/function.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 Mandiant, 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
@@ -11,9 +11,11 @@ import envi
 import viv_utils
 import vivisect.const
 
+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
 
 
@@ -27,7 +29,29 @@ def interface_extract_function_XXX(fh: FunctionHandle) -> Iterator[Tuple[Feature
     yields:
       (Feature, Address): the feature and the address at which its found.
     """
-    ...
+    raise NotImplementedError
+
+
+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]]:
@@ -47,6 +71,10 @@ def extract_function_loop(fhandle: FunctionHandle) -> Iterator[Tuple[Feature, Ad
     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 & envi.BR_COND
@@ -75,4 +103,8 @@ def extract_features(fh: FunctionHandle) -> Iterator[Tuple[Feature, Address]]:
             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

@@ -1,9 +1,13 @@
+# 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
 
-import envi.archs.i386
-import envi.archs.amd64
-
 from capa.features.common import ARCH_I386, ARCH_AMD64, Arch, Feature
 from capa.features.address import NO_ADDRESS, Address
 
@@ -11,10 +15,11 @@ logger = logging.getLogger(__name__)
 
 
 def extract_arch(vw) -> Iterator[Tuple[Feature, Address]]:
-    if isinstance(vw.arch, envi.archs.amd64.Amd64Module):
+    arch = vw.getMeta("Architecture")
+    if arch == "amd64":
         yield Arch(ARCH_AMD64), NO_ADDRESS
 
-    elif isinstance(vw.arch, envi.archs.i386.i386Module):
+    elif arch == "i386":
         yield Arch(ARCH_I386), NO_ADDRESS
 
     else:

capa/features/extractors/viv/helpers.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 Mandiant, 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

capa/features/extractors/viv/indirect_calls.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 Mandiant, 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
@@ -7,7 +7,7 @@
 # See the License for the specific language governing permissions and limitations under the License.
 
 import collections
-from typing import Set, List, Deque, Tuple, Union, Optional
+from typing import Set, List, Deque, Tuple, Optional
 
 import envi
 import vivisect.const
@@ -42,7 +42,7 @@ def get_previous_instructions(vw: VivWorkspace, va: int) -> List[int]:
     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:
         ploc = vw.getPrevLocation(va, adjacent=True)
@@ -59,7 +59,7 @@ def get_previous_instructions(vw: VivWorkspace, va: int) -> List[int]:
     #
     # 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)
@@ -71,7 +71,7 @@ class NotFoundError(Exception):
     pass
 
 
-def find_definition(vw: VivWorkspace, va: int, reg: int) -> Tuple[int, Union[int, None]]:
+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.
@@ -87,8 +87,8 @@ def find_definition(vw: VivWorkspace, va: int, reg: int) -> Tuple[int, Union[int
     raises:
       NotFoundError: when the definition cannot be found.
     """
-    q = collections.deque()  # type: Deque[int]
-    seen = set([])  # type: Set[int]
+    q: Deque[int] = collections.deque()
+    seen: Set[int] = set()
 
     q.extend(get_previous_instructions(vw, va))
     while q:

capa/features/extractors/viv/insn.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 Mandiant, 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
@@ -22,6 +22,7 @@ import capa.features.extractors.viv.helpers
 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
 
@@ -44,7 +45,7 @@ def interface_extract_instruction_XXX(
     yields:
       (Feature, Address): the feature and the address at which its found.
     """
-    ...
+    raise NotImplementedError
 
 
 def get_imports(vw):
@@ -109,6 +110,26 @@ def extract_insn_api_features(fh: FunctionHandle, bb, ih: InsnHandle) -> Iterato
         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
@@ -175,8 +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:
@@ -262,15 +288,19 @@ def extract_insn_bytes_features(fh: FunctionHandle, bb, ih: InsnHandle) -> Itera
         else:
             continue
 
-        for v in derefs(f.vw, v):
+        for vv in derefs(f.vw, v):
             try:
-                buf = read_bytes(f.vw, v)
+                buf = read_bytes(f.vw, vv)
             except envi.exc.SegmentationViolation:
                 continue
 
             if capa.features.extractors.helpers.all_zeros(buf):
                 continue
 
+            if f.vw.isProbablyString(vv) or f.vw.isProbablyUnicode(vv):
+                # don't extract byte features for obvious strings
+                continue
+
             yield Bytes(buf), ih.address
 
 
@@ -281,7 +311,12 @@ def read_string(vw, offset: int) -> str:
         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)
@@ -300,7 +335,9 @@ def read_string(vw, offset: int) -> str:
                 # 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)
 
@@ -314,7 +351,6 @@ def is_security_cookie(f, bb, insn) -> bool:
     if oper.isReg() and oper.reg not in [
         envi.archs.i386.regs.REG_ESP,
         envi.archs.i386.regs.REG_EBP,
-        # TODO: do x64 support for real.
         envi.archs.amd64.regs.REG_RBP,
         envi.archs.amd64.regs.REG_RSP,
     ]:
@@ -374,9 +410,7 @@ def extract_insn_obfs_call_plus_5_characteristic_features(f, bb, ih: InsnHandle)
         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) or isinstance(
-        insn.opers[0], envi.archs.amd64.disasm.Amd64RipRelOper
-    ):
+    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
 
@@ -385,7 +419,6 @@ def extract_insn_peb_access_characteristic_features(f, bb, ih: InsnHandle) -> It
     """
     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"]:
@@ -493,7 +526,8 @@ def extract_function_calls_from(fh: FunctionHandle, bb, ih: InsnHandle) -> Itera
     if isinstance(insn.opers[0], envi.archs.i386.disasm.i386ImmMemOper):
         oper = insn.opers[0]
         target = oper.getOperAddr(insn)
-        yield Characteristic("calls from"), AbsoluteVirtualAddress(target)
+        if target >= 0:
+            yield Characteristic("calls from"), AbsoluteVirtualAddress(target)
 
     # call via thunk on x86,
     # see 9324d1a8ae37a36ae560c37448c9705a at 0x407985
@@ -509,7 +543,8 @@ def extract_function_calls_from(fh: FunctionHandle, bb, ih: InsnHandle) -> Itera
     elif isinstance(insn.opers[0], envi.archs.amd64.disasm.Amd64RipRelOper):
         op = insn.opers[0]
         target = op.getOperAddr(insn)
-        yield Characteristic("calls from"), AbsoluteVirtualAddress(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
@@ -607,7 +642,6 @@ def extract_op_offset_features(
         if oper.reg == envi.archs.i386.regs.REG_EBP:
             return
 
-        # TODO: do x64 support for real.
         if oper.reg == envi.archs.amd64.regs.REG_RBP:
             return
 
@@ -661,13 +695,14 @@ def extract_op_string_features(
     else:
         return
 
-    for v in derefs(f.vw, v):
+    for vv in derefs(f.vw, v):
         try:
-            s = read_string(f.vw, v)
+            s = read_string(f.vw, vv).rstrip("\x00")
         except ValueError:
             continue
         else:
-            yield String(s.rstrip("\x00")), ih.address
+            if len(s) >= 4:
+                yield String(s), ih.address
 
 
 def extract_operand_features(f: FunctionHandle, bb, insn: InsnHandle) -> Iterator[Tuple[Feature, Address]]:

capa/features/file.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2020 Mandiant, 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
@@ -12,19 +12,19 @@ from capa.features.common import Feature
 class Export(Feature):
     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: 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: str, description=None):
         # value is section name
-        super(Section, self).__init__(value, description=description)
+        super().__init__(value, description=description)
 
 
 class FunctionName(Feature):
@@ -32,7 +32,7 @@ class FunctionName(Feature):
 
     def __init__(self, name: str, description=None):
         # value is function name
-        super(FunctionName, self).__init__(name, description=description)
+        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/__init__.py

@@ -1,7 +1,7 @@
 """
 capa freeze file format: `| capa0000 | + zlib(utf-8(json(...)))`
 
-Copyright (C) 2020 Mandiant, 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
@@ -9,13 +9,17 @@ Unless required by applicable law or agreed to in writing, software distributed
  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
 from enum import Enum
-from typing import Any, List, Tuple
+from typing import List, Tuple, Union, Literal
 
-import dncil.clr.token
-from pydantic import Field, BaseModel
+from pydantic import Field, BaseModel, ConfigDict
+
+# TODO(williballenthin): use typing.TypeAlias directly in Python 3.10+
+# https://github.com/mandiant/capa/issues/1699
+from typing_extensions import TypeAlias
 
 import capa.helpers
 import capa.version
@@ -24,16 +28,23 @@ import capa.features.insn
 import capa.features.common
 import capa.features.address
 import capa.features.basicblock
-import capa.features.extractors.base_extractor
+import capa.features.extractors.null as null
 from capa.helpers import assert_never
 from capa.features.freeze.features import Feature, feature_from_capa
+from capa.features.extractors.base_extractor import (
+    SampleHashes,
+    FeatureExtractor,
+    StaticFeatureExtractor,
+    DynamicFeatureExtractor,
+)
 
 logger = logging.getLogger(__name__)
 
+CURRENT_VERSION = 3
+
 
 class HashableModel(BaseModel):
-    class Config:
-        frozen = True
+    model_config = ConfigDict(frozen=True)
 
 
 class AddressType(str, Enum):
@@ -42,12 +53,15 @@ class AddressType(str, Enum):
     FILE = "file"
     DN_TOKEN = "dn token"
     DN_TOKEN_OFFSET = "dn token offset"
+    PROCESS = "process"
+    THREAD = "thread"
+    CALL = "call"
     NO_ADDRESS = "no address"
 
 
 class Address(HashableModel):
     type: AddressType
-    value: Any
+    value: Union[int, Tuple[int, ...], None] = None  # None default value to support deserialization of NO_ADDRESS
 
     @classmethod
     def from_capa(cls, a: capa.features.address.Address) -> "Address":
@@ -61,10 +75,19 @@ class Address(HashableModel):
             return cls(type=AddressType.FILE, value=int(a))
 
         elif isinstance(a, capa.features.address.DNTokenAddress):
-            return cls(type=AddressType.DN_TOKEN, value=a.token.value)
+            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.value, a.offset))
+            return cls(type=AddressType.DN_TOKEN_OFFSET, value=(a.token, a.offset))
+
+        elif isinstance(a, capa.features.address.ProcessAddress):
+            return cls(type=AddressType.PROCESS, value=(a.ppid, a.pid))
+
+        elif isinstance(a, capa.features.address.ThreadAddress):
+            return cls(type=AddressType.THREAD, value=(a.process.ppid, a.process.pid, a.tid))
+
+        elif isinstance(a, capa.features.address.DynamicCallAddress):
+            return cls(type=AddressType.CALL, value=(a.thread.process.ppid, a.thread.process.pid, a.thread.tid, a.id))
 
         elif a == capa.features.address.NO_ADDRESS or isinstance(a, capa.features.address._NoAddress):
             return cls(type=AddressType.NO_ADDRESS, value=None)
@@ -80,20 +103,54 @@ class Address(HashableModel):
 
     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:
-            return capa.features.address.DNTokenAddress(dncil.clr.token.Token(self.value))
+            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
-            return capa.features.address.DNTokenOffsetAddress(dncil.clr.token.Token(token), offset)
+            assert isinstance(token, int)
+            assert isinstance(offset, int)
+            return capa.features.address.DNTokenOffsetAddress(token, offset)
+
+        elif self.type is AddressType.PROCESS:
+            assert isinstance(self.value, tuple)
+            ppid, pid = self.value
+            assert isinstance(ppid, int)
+            assert isinstance(pid, int)
+            return capa.features.address.ProcessAddress(ppid=ppid, pid=pid)
+
+        elif self.type is AddressType.THREAD:
+            assert isinstance(self.value, tuple)
+            ppid, pid, tid = self.value
+            assert isinstance(ppid, int)
+            assert isinstance(pid, int)
+            assert isinstance(tid, int)
+            return capa.features.address.ThreadAddress(
+                process=capa.features.address.ProcessAddress(ppid=ppid, pid=pid), tid=tid
+            )
+
+        elif self.type is AddressType.CALL:
+            assert isinstance(self.value, tuple)
+            ppid, pid, tid, id_ = self.value
+            return capa.features.address.DynamicCallAddress(
+                thread=capa.features.address.ThreadAddress(
+                    process=capa.features.address.ProcessAddress(ppid=ppid, pid=pid), tid=tid
+                ),
+                id=id_,
+            )
 
         elif self.type is AddressType.NO_ADDRESS:
             return capa.features.address.NO_ADDRESS
@@ -109,7 +166,11 @@ class Address(HashableModel):
             return True
 
         else:
-            return self.value < other.value
+            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):
@@ -121,6 +182,48 @@ class FileFeature(HashableModel):
     feature: Feature
 
 
+class ProcessFeature(HashableModel):
+    """
+    args:
+        process: the address of the process to which this feature belongs.
+        address: the address at which this feature is found.
+
+    process != address because, e.g., the feature may be found *within* the scope (process).
+    """
+
+    process: Address
+    address: Address
+    feature: Feature
+
+
+class ThreadFeature(HashableModel):
+    """
+    args:
+        thread: the address of the thread to which this feature belongs.
+        address: the address at which this feature is found.
+
+    thread != address because, e.g., the feature may be found *within* the scope (thread).
+    """
+
+    thread: Address
+    address: Address
+    feature: Feature
+
+
+class CallFeature(HashableModel):
+    """
+    args:
+        call: the address of the call to which this feature belongs.
+        address: the address at which this feature is found.
+
+    call != address for consistency with Process and Thread.
+    """
+
+    call: Address
+    address: Address
+    feature: Feature
+
+
 class FunctionFeature(HashableModel):
     """
     args:
@@ -146,9 +249,10 @@ class BasicBlockFeature(HashableModel):
     versus right at its starting address.
     """
 
-    basic_block: Address
+    basic_block: Address = Field(alias="basic block")
     address: Address
     feature: Feature
+    model_config = ConfigDict(populate_by_name=True)
 
 
 class InstructionFeature(HashableModel):
@@ -157,8 +261,7 @@ class InstructionFeature(HashableModel):
         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 because, for consistency with Function and BasicBlock.
     """
 
     instruction: Address
@@ -180,44 +283,66 @@ class BasicBlockFeatures(BaseModel):
 class FunctionFeatures(BaseModel):
     address: Address
     features: Tuple[FunctionFeature, ...]
-    basic_blocks: Tuple[BasicBlockFeatures, ...] = Field(alias="basic block")
-
-    class Config:
-        allow_population_by_field_name = True
+    basic_blocks: Tuple[BasicBlockFeatures, ...] = Field(alias="basic blocks")
+    model_config = ConfigDict(populate_by_name=True)
 
 
-class Features(BaseModel):
+class CallFeatures(BaseModel):
+    address: Address
+    name: str
+    features: Tuple[CallFeature, ...]
+
+
+class ThreadFeatures(BaseModel):
+    address: Address
+    features: Tuple[ThreadFeature, ...]
+    calls: Tuple[CallFeatures, ...]
+
+
+class ProcessFeatures(BaseModel):
+    address: Address
+    name: str
+    features: Tuple[ProcessFeature, ...]
+    threads: Tuple[ThreadFeatures, ...]
+
+
+class StaticFeatures(BaseModel):
     global_: Tuple[GlobalFeature, ...] = Field(alias="global")
     file: Tuple[FileFeature, ...]
     functions: Tuple[FunctionFeatures, ...]
+    model_config = ConfigDict(populate_by_name=True)
 
-    class Config:
-        allow_population_by_field_name = True
+
+class DynamicFeatures(BaseModel):
+    global_: Tuple[GlobalFeature, ...] = Field(alias="global")
+    file: Tuple[FileFeature, ...]
+    processes: Tuple[ProcessFeatures, ...]
+    model_config = ConfigDict(populate_by_name=True)
+
+
+Features: TypeAlias = Union[StaticFeatures, DynamicFeatures]
 
 
 class Extractor(BaseModel):
     name: str
     version: str = capa.version.__version__
-
-    class Config:
-        allow_population_by_field_name = True
+    model_config = ConfigDict(populate_by_name=True)
 
 
 class Freeze(BaseModel):
-    version: int = 2
+    version: int = CURRENT_VERSION
     base_address: Address = Field(alias="base address")
+    sample_hashes: SampleHashes
+    flavor: Literal["static", "dynamic"]
     extractor: Extractor
     features: Features
-
-    class Config:
-        allow_population_by_field_name = True
+    model_config = ConfigDict(populate_by_name=True)
 
 
-def dumps(extractor: capa.features.extractors.base_extractor.FeatureExtractor) -> str:
+def dumps_static(extractor: StaticFeatureExtractor) -> str:
     """
     serialize the given extractor to a string
     """
-
     global_features: List[GlobalFeature] = []
     for feature, _ in extractor.extract_global_features():
         global_features.append(
@@ -255,7 +380,8 @@ def dumps(extractor: capa.features.extractors.base_extractor.FeatureExtractor) -
                     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)
             ]
 
@@ -274,52 +400,168 @@ def dumps(extractor: capa.features.extractors.base_extractor.FeatureExtractor) -
                 instructions.append(
                     InstructionFeatures(
                         address=iaddr,
-                        features=ifeatures,
+                        features=tuple(ifeatures),
                     )
                 )
 
             basic_blocks.append(
                 BasicBlockFeatures(
                     address=bbaddr,
-                    features=bbfeatures,
-                    instructions=instructions,
+                    features=tuple(bbfeatures),
+                    instructions=tuple(instructions),
                 )
             )
 
         function_features.append(
             FunctionFeatures(
                 address=faddr,
-                features=ffeatures,
+                features=tuple(ffeatures),
                 basic_blocks=basic_blocks,
+            )  # type: ignore
+            # Mypy is unable to recognise `basic_blocks` as a argument due to alias
+        )
+
+    features = StaticFeatures(
+        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=CURRENT_VERSION,
+        base_address=Address.from_capa(extractor.get_base_address()),
+        sample_hashes=extractor.get_sample_hashes(),
+        flavor="static",
+        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 dumps_dynamic(extractor: DynamicFeatureExtractor) -> 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),
             )
         )
 
-    features = Features(
+    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),
+            )
+        )
+
+    process_features: List[ProcessFeatures] = []
+    for p in extractor.get_processes():
+        paddr = Address.from_capa(p.address)
+        pname = extractor.get_process_name(p)
+        pfeatures = [
+            ProcessFeature(
+                process=paddr,
+                address=Address.from_capa(addr),
+                feature=feature_from_capa(feature),
+            )
+            for feature, addr in extractor.extract_process_features(p)
+        ]
+
+        threads = []
+        for t in extractor.get_threads(p):
+            taddr = Address.from_capa(t.address)
+            tfeatures = [
+                ThreadFeature(
+                    basic_block=taddr,
+                    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_thread_features(p, t)
+            ]
+
+            calls = []
+            for call in extractor.get_calls(p, t):
+                caddr = Address.from_capa(call.address)
+                cname = extractor.get_call_name(p, t, call)
+                cfeatures = [
+                    CallFeature(
+                        call=caddr,
+                        address=Address.from_capa(addr),
+                        feature=feature_from_capa(feature),
+                    )
+                    for feature, addr in extractor.extract_call_features(p, t, call)
+                ]
+
+                calls.append(
+                    CallFeatures(
+                        address=caddr,
+                        name=cname,
+                        features=tuple(cfeatures),
+                    )
+                )
+
+            threads.append(
+                ThreadFeatures(
+                    address=taddr,
+                    features=tuple(tfeatures),
+                    calls=tuple(calls),
+                )
+            )
+
+        process_features.append(
+            ProcessFeatures(
+                address=paddr,
+                name=pname,
+                features=tuple(pfeatures),
+                threads=tuple(threads),
+            )
+        )
+
+    features = DynamicFeatures(
         global_=global_features,
-        file=file_features,
-        functions=function_features,
-    )
+        file=tuple(file_features),
+        processes=tuple(process_features),
+    )  # type: ignore
+    # Mypy is unable to recognise `global_` as a argument due to alias
+
+    # workaround around mypy issue: https://github.com/python/mypy/issues/1424
+    get_base_addr = getattr(extractor, "get_base_addr", None)
+    base_addr = get_base_addr() if get_base_addr else capa.features.address.NO_ADDRESS
 
     freeze = Freeze(
-        version=2,
-        base_address=Address.from_capa(extractor.get_base_address()),
+        version=CURRENT_VERSION,
+        base_address=Address.from_capa(base_addr),
+        sample_hashes=extractor.get_sample_hashes(),
+        flavor="dynamic",
         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.json()
+    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
+def loads_static(s: str) -> StaticFeatureExtractor:
+    """deserialize a set of features (as a NullStaticFeatureExtractor) from a string."""
+    freeze = Freeze.model_validate_json(s)
+    if freeze.version != CURRENT_VERSION:
+        raise ValueError(f"unsupported freeze format version: {freeze.version}")
 
-    freeze = Freeze.parse_raw(s)
-    if freeze.version != 2:
-        raise ValueError("unsupported freeze format version: %d", freeze.version)
+    assert freeze.flavor == "static"
+    assert isinstance(freeze.features, StaticFeatures)
 
-    return null.NullFeatureExtractor(
+    return null.NullStaticFeatureExtractor(
         base_address=freeze.base_address.to_capa(),
+        sample_hashes=freeze.sample_hashes,
         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={
@@ -343,10 +585,59 @@ def loads(s: str) -> capa.features.extractors.base_extractor.FeatureExtractor:
     )
 
 
+def loads_dynamic(s: str) -> DynamicFeatureExtractor:
+    """deserialize a set of features (as a NullDynamicFeatureExtractor) from a string."""
+    freeze = Freeze.model_validate_json(s)
+    if freeze.version != CURRENT_VERSION:
+        raise ValueError(f"unsupported freeze format version: {freeze.version}")
+
+    assert freeze.flavor == "dynamic"
+    assert isinstance(freeze.features, DynamicFeatures)
+
+    return null.NullDynamicFeatureExtractor(
+        base_address=freeze.base_address.to_capa(),
+        sample_hashes=freeze.sample_hashes,
+        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],
+        processes={
+            p.address.to_capa(): null.ProcessFeatures(
+                name=p.name,
+                features=[(fe.address.to_capa(), fe.feature.to_capa()) for fe in p.features],
+                threads={
+                    t.address.to_capa(): null.ThreadFeatures(
+                        features=[(fe.address.to_capa(), fe.feature.to_capa()) for fe in t.features],
+                        calls={
+                            c.address.to_capa(): null.CallFeatures(
+                                name=c.name,
+                                features=[(fe.address.to_capa(), fe.feature.to_capa()) for fe in c.features],
+                            )
+                            for c in t.calls
+                        },
+                    )
+                    for t in p.threads
+                },
+            )
+            for p in freeze.features.processes
+        },
+    )
+
+
 MAGIC = "capa0000".encode("ascii")
 
 
-def dump(extractor: capa.features.extractors.base_extractor.FeatureExtractor) -> bytes:
+def dumps(extractor: FeatureExtractor) -> str:
+    """serialize the given extractor to a string."""
+    if isinstance(extractor, StaticFeatureExtractor):
+        doc = dumps_static(extractor)
+    elif isinstance(extractor, DynamicFeatureExtractor):
+        doc = dumps_dynamic(extractor)
+    else:
+        raise ValueError("Invalid feature extractor")
+
+    return doc
+
+
+def dump(extractor: FeatureExtractor) -> bytes:
     """serialize the given extractor to a byte array."""
     return MAGIC + zlib.compress(dumps(extractor).encode("utf-8"))
 
@@ -355,16 +646,34 @@ def is_freeze(buf: bytes) -> bool:
     return buf[: len(MAGIC)] == MAGIC
 
 
-def load(buf: bytes) -> capa.features.extractors.base_extractor.FeatureExtractor:
+def loads(s: str):
+    doc = json.loads(s)
+
+    if doc["version"] != CURRENT_VERSION:
+        raise ValueError(f"unsupported freeze format version: {doc['version']}")
+
+    if doc["flavor"] == "static":
+        return loads_static(s)
+    elif doc["flavor"] == "dynamic":
+        return loads_dynamic(s)
+    else:
+        raise ValueError(f"unsupported freeze format flavor: {doc['flavor']}")
+
+
+def load(buf: bytes):
     """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"))
+
+    s = zlib.decompress(buf[len(MAGIC) :]).decode("utf-8")
+
+    return loads(s)
 
 
 def main(argv=None):
     import sys
     import argparse
+    from pathlib import Path
 
     import capa.main
 
@@ -372,17 +681,16 @@ def main(argv=None):
         argv = sys.argv[1:]
 
     parser = argparse.ArgumentParser(description="save capa features to a file")
-    capa.main.install_common_args(parser, {"sample", "format", "backend", "signatures"})
+    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.backend, sigpaths, False)
+    extractor = capa.main.get_extractor(args.sample, args.format, args.os, args.backend, sigpaths, False)
 
-    with open(args.output, "wb") as f:
-        f.write(dump(extractor))
+    Path(args.output).write_bytes(dump(extractor))
 
     return 0