Reduce UDS/segment name clashes across same-service instances

[ankor2023]

Add a PID file name hash to the names of the shared memory segments and
Unix Domain Sockets. Since all instances running on the same host are
supposed to have unique PID files, this addition significantly reduces
the probability of name clashes when running multiple Squid instances
with the same service name (i.e. the same squid -n parameter value
that defaults to "squid").

A clash may still happen if two different PID file names have the same
hash or if multiple instances disable PID file management with
pid_filename none. Clashes may also happen in environments where Squid
does not even use service name for naming shared memory segments.

Examples of UDS and shared memory segment names (while using default
service name):

/var/run/squid/squid-SLWQ-kid-1.ipc
/var/run/squid/squid-SLWQ-coordinator.ipc
/dev/shm/squid-SLWQ-tls_session_cache.shm
/dev/shm/squid-SLWQ-transients_map_slices.shm

This change is a reference point for automated CONTRIBUTORS updates.

[yadij]

Build issues are because the src/Instance.cc is not linked against the unit tests for Rock and UFS cache types. See src/Makefile.am "tests for fs/*". Please run "make check" locally to check the fix for these and find any other tests impacted by this new code.

Also, why the murmur hash? The original proposal was just to add the PID number (from the "master" or "coordinator" process). If we do need to hash for any reason Squid uses libnettle, for a selection of speed and security options.

[rousskov]

Amos: The original proposal was just to add the PID number

It was not. Proposal A was to add a hash of the full PID file name.

Adding a PID number instead of a hash would solve the problem with segment clashes as well but would create a worse problem: Segments left by crashed Squid instances will accumulate instead of being removed (because each instance is likely to have a unique master PID value, even instances that have the same PID file name).

Amos: If we do need to hash for any reason Squid uses libnettle, for a selection of speed and security options.

I agree that this PR should not introduce a new hash function. However, please do not rely on libnettle for this PR either; essentially, ignore that libnettle exists. See one of the change requests for a specific hash function recommendation.

[rousskov]

@kinkie, I have removed backport-to-v7 label just in case you do not want to backport this PR that might trigger more PSHMNAMLEN errors in some cases (or want to wait for a followup PR as discussed earlier). Please restore that label if needed!

[rousskov]

Newer FreeBSD version for package ztrack:
To ignore this error set IGNORE_OSVERSION=yes
  - package: 1305000
  - running userland: 1304000
repository FreeBSD contains packages for wrong OS version

@kinkie, if restarting that failed FreeBSD v13.4 staging test is not the answer, then please help getting past that package versioning error.

Edit: @yadij has already flagged the same problem at #2110 (comment). I will just follow the discussion in that PR to reduce the noise here.

[kinkie]

CI unblocked by PR #2112

[rousskov]

• .... PR that automatically replaces long segment names with short hashes
• .... PR that manually renames existing segments

To add one more data point for the future followup PR(s) to consider, here is a segment name I just saw in the logs produced by an older (v5- or v6-based?) Squid:

kid3| 54,3| mem/Segment.cc(148) open: opened /squid-usr.local.squid.var.cache.rocks.1_map_filenos.shm segment: 260

That name is 55-characters long and cannot be shortened enough by simply shortening map_filenos or other hard-coded labels because its longest component is derived from the admin-configured 33-character cache_dir location: /usr/local/squid/var/cache/rocks/1. This is why I crossed out the "manually renames existing segments" PR idea above.

If we want to fully support SMP Squid on MacOS, then we need a different way to produce short names: Either a hash or some kind of enumeration. To avoid completely mishandling, for example, instances where different kids have a different set of cache_dir directives or a different cache_dir directives order, both options would have to use a central hashing/enumeration authority (e.g., a new special/dedicated shared memory segment mapping full names to hashes or enumeration IDs).

[yadij]

backport to v7 completed.