- Use `-b` most everywhere, it will save time.
- Start some intel tests upon the input file being fully read instead of
at an arbitrary time.
- Improve termination condition for some sumstats/cluster tests.
- Filter uninteresting output from some supervisor tests.
- Test for `notice_policy.log` is no longer needed.
Previously, a single `icmp_conn` record was built per ICMP "connection"
and re-used for all events generated from it. This may have been a
historical attempt at performance optimization, but:
* By default, Zeek does not load any scripts that handle ICMP events.
* The one script Zeek ships with that does handle ICMP events,
"detect-traceroute", is already noted as being disabled due to
potential performance problems of doing that kind of analysis.
* Re-use of the original `icmp_conn` record tends to misreport
TTL and length values since they come from original packet instead
of the current one.
* Even if we chose to still re-use `icmp_conn` records and just fill
in a new TTL and length value each packet, a user script could have
stored a reference to the record and not be expecting those values
to be changed out from underneath them.
Now, a new `icmp_info` record is created/populated in all ICMP events
and should be used instead of `icmp_conn`. It also removes the
orig_h/resp_h fields as those are redundant with what's already
available in the connection record.
Duplicate script `@load` directives are now detected by comparing
against canonical paths formed by realpath(). This fixes the previous,
unexpected behavior of treating scripts that hardlink to same
inode as duplicates: such links will now be loaded as distinct scripts
since their canonical path differs.
* Variables of `string` type can now be set to an empty string
* Trying to set a variable with non-`string` type to an empty value
now emits an error instead of silently doing nothing
* Providing an invalid identifier now emits an "unknown identifier"
error instead of silently doing nothing
This introduces a new sampling state-map for expired connections to fix
segfaults that previously occured when passing in a `connection` record
to `Reporter::conn_weird()` for which the internal `Connection` object
had already been expired and deleted. This also introduces a new event
called `expired_conn_weird`, which is similar to `conn_weird`, except
the full `connection` record is no longer available, just the `conn_id`
and UID string.
- Updated the logic significantly: still filters out ICMP from being
considered an active service (like before) and adds a new
"Known::service_udp_requires_response" option (defaults to true) for
whether to require UDP server response before being considered an
active service.
* 'topic/dopheide/known-services' of https://github.com/dopheide-esnet/zeek:
Log services with unknown protocols
- Minor whitespace/format adjustments
* 'network-time-init' of https://github.com/J-Gras/zeek:
Fix documentation for network_time_init.
Improve network_time_init() test and add baseline.
Add network_time_init() event.
Add test for network_time behavior.
This controls whether ports given by "udp_content_delivery_ports_orig" and
"udp_content_delivery_ports_orig" are in terms of the UDP packet's
destination port or by the Connection's "responder" port (the former is
the unchanged default behavior).
Connection durations were being compared "less than" instead of "greater
than" the desired duration threshold as provided to
set_current_conn_bytes_threshold() or
ConnThreshold::set_duration_threshold()
This changes the decapsulation logic for GRE/ERSPAN payloads to re-use
existing Layer 2 parsing logic that already handles things like 802.1Q
tags correctly before going on to process the inner IPv4/IPv6 payload.
These were previously reporting leaks due to various allocations not
getting cleaned up during the stack unwind, but at the current state of
the transition toward IntrusivePtr usage, theses tests no longer leak.
* Generally increase timeouts for tests that have recent transient
failures
* Change any test that relied on `btest-bg-wait -k` since that's never
going to play with with CI systems. Instead, we always need to have
a well-defined termination condition in the test itself (and most
already did, so didn't really need the `-k` flag anyway).
Or otherwise convert into a regular btest if it didn't already seem to
be covered.
There's no need for a separate memory leak test group since compiling
with LeakSanitizer now covers leak checking for the full btest suite.
E.g. ones that throw interpreter exceptions, as those are currently
known to potentially cause leaks. Fixing the underlying leaks involves
the larger task of more IntrusivePtr usage.
Reference cycles may also cause leaks.
Adjustments during merge:
- kept the UNKNOWN Log::ID as placeholder value
- changed the coverage.find-bro-logs test to check for arbitrary $path
field values instead of just string literals
- don't force EnumVal to unsigned integer since the relevant union member
is the signed integer and added the relevant enum values/types to
.bif files for easier access
- compare FILE* versus file name to check for stdout equality (don't
think it matters much, just a bit more efficient)
- minor whitespace/style tweaks
* origin/topic/dev/print-to-log:
Added a non boolean configuration and other changes as suggested by Jon
Allow Print Statements to be redirected to a Log# This is a combination of 3 commits.
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.
Now, in addition to setting thresholds for bytes and packet, one can set
a threshold for connection duration. Note that the threshold event is
only raised once the next packet in the connection is seen.
This also fixes a small pre-existing bug, in which a bunch of warnings
were raised if someone just used the lower-level functions without going
through the higher-level scripting API.
For broker.log and cluster.log: there was a race condition. A worker's
first IOSource that it processes is potentially Broker if there were
no packets available yet and thread scheduling happens to work out
such that network connections (inside CAF threads) become established
before we enter the main I/O loop. Such peering establishments would
generate logs with timestamp 0 as there was not yet any code path
taken that would update network_time.
For reporter.log: any non-worker (packet-processing) node would just
unnecessarily use a timestamp of 0 for their reporter messages.
