- Make sure the server sends a complete response when the client
includes the "Connection: close" header.
- Make sure the server terminates in `m_strand` to avoid
concurrent socket closure and ops processing.
In this repo, `boost::interprocess` was being used soley to make `uint32_t` operations atomic. So I replaced each instance of
`boost::interprocess::ipcdetail::atomic(...)32` with `std::atomic` methods. I replaced member declarations as applicable. For example,
when I needed to change a `volatile uint32_t` into a `std::atomic<uint32_t>`. Sometimes, a member was being used a boolean flag, so
I replaced it with `std::atomic<bool>`.
You may notice that I didn't touch `levin_client_async.h`. That is because this file is entirely unused and will be deleted in PR monero-project#8211.
Additional changes from review:
* Make some local variables const
* Change postfix operators to prefix operators where value was not need
This fixes rapid reconnections failing as the peer hasn't yet
worked out the other side is gone, and will reject "duplicate"
connections until a timeout.
Resetting the timer after shutdown was initiated would keep
a reference to the object inside ASIO, which would keep the
connection alive until the timer timed out
The problem actually exists in two parts:
1. When sending chunks over a connection, if the queue size is
greater than N, the seed is predictable across every monero node.
>"If rand() is used before any calls to srand(), rand() behaves as if
it was seeded with srand(1). Each time rand() is seeded with the same seed, it
must produce the same sequence of values."
2. The CID speaks for itself: "'rand' should not be used for security-related
applications, because linear congruential algorithms are too easy to break."
*But* this is an area of contention.
One could argue that a CSPRNG is warranted in order to fully mitigate any
potential timing attacks based on crafting chunk responses. Others could argue
that the existing LCG, or even an MTG, would suffice (if properly seeded). As a
compromise, I've used an MTG with a full bit space. This should give a healthy
balance of security and speed without relying on the existing crypto library
(which I'm told might break on some systems since epee is not (shouldn't be)
dependent upon the existing crypto library).
new cli options (RPC ones also apply to wallet):
--p2p-bind-ipv6-address (default = "::")
--p2p-bind-port-ipv6 (default same as ipv4 port for given nettype)
--rpc-bind-ipv6-address (default = "::1")
--p2p-use-ipv6 (default false)
--rpc-use-ipv6 (default false)
--p2p-require-ipv4 (default true, if ipv4 bind fails and this is
true, will not continue even if ipv6 bind
successful)
--rpc-require-ipv4 (default true, description as above)
ipv6 addresses are to be specified as "[xx:xx:xx::xx:xx]:port" except
in the cases of the cli args for bind address. For those the square
braces can be omitted.
When closing connections due to exiting, the IO service is
already gone, so the data exchange needed for a gracious SSL
shutdown cannot happen. We just close the socket in that case.
Specifying SSL certificates for peer verification does an exact match,
making it a not-so-obvious alias for the fingerprints option. This
changes the checks to OpenSSL which loads concatenated certificate(s)
from a single file and does a certificate-authority (chain of trust)
check instead. There is no drop in security - a compromised exact match
fingerprint has the same worse case failure. There is increased security
in allowing separate long-term CA key and short-term SSL server keys.
This also removes loading of the system-default CA files if a custom
CA file or certificate fingerprint is specified.
j-berman