In https://github.com/zeek/zeek/pull/2191, we added endpoint flipping
for cases where a connection starts with a SYN/ACK followed by ACK or
data. The goal was to treat the connection as productive and go ahead
and parse it. But the TCP analyzer could continue to consider it
partial after flipping, meaning that app layers would bail out. #2426
shows such a case: HTTP gets correctly activated after flipping
through content inspection, but it won't process anything because
`IsPartial()` returns true. As the is-partial state reflects
whether we saw the first packets each in direction, this patch now
overrides that state for the originally missing SYN after flipping.
We actually had the same problem at a couple of other locations already
as well. One of that only happened to work because of the originally
inconsistent state flipping that was fixed in the previous commit. The
corresponding unit test now broke after that change. This commit
updates that logic as well to override the state.
This fix is a bit of a hack, but the best solution I could think of
without introducing larger changes.
Closes#2426.
Changes during merge:
- Add dedicated test (w/ trace "client_timestamp_enabled.pcapng" from Cloudshark)
- Change types from signed to unsigned.
- Add cast for bit-shifting operand.
- clang-format run
The RSTOS0 `conn_state` label is documented as "Originator sent a SYN
followed by a RST, never saw SYN-ACK from responder", but was previously
applied to cases where no originator SYN exists, like a single RST-only
packet.
And switch Zeek's base scripts over to using it in place of
"connection_state_remove". The difference between the two is
that "connection_state_remove" is raised for all events while
"successful_connection_remove" excludes TCP connections that were never
established (just SYN packets). There can be performance benefits
to this change for some use-cases.
There's also a new event called ``connection_successful`` and a new
``connection`` record field named "successful" to help indicate this new
property of connections.
Or more generally, signatures would not work correctly for any case
where the first TCP packet seen contained payload data, regardless of
its TCP flags.
- Since it's just the handshake packets out of order, they're no
longer treated as partial connections, which some protocol analyzers
immediately refuse to look at.
- The TCP_Reassembler "is_orig" state failed to change, which led to
protocol analyzers sometimes using the wrong value for that.
- Add a unit test which exercises the Connection::FlipRoles() code
path (i.e. the SYN/SYN-ACK reversal situation).
Addresses BIT-1148.
The previous behavior was to accomodate SYN/FIN/RST-filtered traces by
not reporting missing data (via the content_gap event) for such
connections. The new behavior always reports gaps for connections that
are established and terminate normally, but sequence numbers indicate
that all data packets of the connection were missed. The behavior can
be reverted by redef'ing "detect_filtered_trace".
In the case multiple FIN packets are seen from a TCP endpoint (e.g.
when one is retransmitted), only the first counted towards a byte in the
sequence space. This could cause a subsequent FIN packet to induce an
incorrect wrap around in the sequence numbers (e.g. the retransmitted
FIN packet now is one sequence number behind the the first) and
misleadingly large connection sizes. The change is to always treat a
FIN packet as counting one byte in to the sequence space.
That field is based on TCP sequence numbers and on seeing a SYN followed
by a failed RST injection response, the initial sequence number tracked
the value in the injection (most likely zero) instead of value in
subsequent SYN response. This could make c$resp$size be set to large
values when it's not really.
Also removed some dead code paths.
