Support migration membarrier() PRIVATE_EXPEDITED registrations

[osctobe]

Probe registration of membarrier() for MEMBARRIER_CMD_PRIVATE_EXPEDITED* calls on dump and restore the registrations.

This is a best-effort implementation that ignores unknown modes. The proper migration of membarrier() needs kernel support for querying the state (currently e.g. registration for MEMBARRIER_CMD_GLOBAL_EXPEDITED is not possible to detect).

[adrianreber]

Please open the PR against the criu-dev branch.

[mihalicyn]

Hi @osctobe!

please, rebase your branch against criu-dev.

[codecov-commenter]

Codecov Report

Patch coverage has no change and project coverage change: +0.02% 🎉

Comparison is base (e1cda9f) 70.57% compared to head (4948d09) 70.60%.

❗ Current head 4948d09 differs from pull request most recent head e6738b6. Consider uploading reports for the commit e6738b6 to get more accurate results

Additional details and impacted files

@@             Coverage Diff              @@
##           criu-dev    #2025      +/-   ##
============================================
+ Coverage     70.57%   70.60%   +0.02%     
============================================
  Files           133      133              
  Lines         33314    33319       +5     
============================================
+ Hits          23513    23524      +11     
+ Misses         9801     9795       -6     

Files Changed	Coverage Δ
criu/cr-dump.c	75.85% <ø> (ø)
criu/cr-restore.c	67.38% <ø> (ø)
criu/include/parasite.h	100.00% <ø> (ø)

... and 3 files with indirect coverage changes

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[mclapinski]

I think this could use a test.

[Snorch]

Thanks for a good PR! Please see some comments. And yes you need a zdtm test for it.

[github-actions]

A friendly reminder that this PR had no activity for 30 days.

[osctobe]

Rebased and fixed. A test is attached.

[mclapinski]

1. Check if you actually fixed Snorch's comments. If yes, you can close conversation. I'm pretty sure you didn't fix the comment about the comments' format.
2. Split the test into a separate commit.
3. Fix compile issues reported by the CI.

[avagin]

This is a best-effort implementation that ignores unknown modes. The proper migration of membarrier() needs kernel support for querying the state (currently e.g. registration for MEMBARRIER_CMD_GLOBAL_EXPEDITED is not possible to detect).

Now we have MEMBARRIER_CMD_GET_REGISTRATIONS
torvalds/linux@544a4f2

Should we start using it to solve this problem?

[mclapinski]

MEMBARRIER_CMD_GET_REGISTRATIONS exists only on fairly new kernels. IMO it's better to merge this and then write a new PR to use MEMBARRIER_CMD_GET_REGISTRATIONS with this implementation as fallback.

[osctobe]

@avagin ping? BTW, I think even with the no-auto-close tag this PR seems to be falling through the cracks.

[mclapinski]

Still fails CI. I don't know about the Docker test but the other 2 are real failures.

[avagin]

Merged with a few minor fixes.

@osctobe I don't merge CL-s that have CI failures without explanations why they are unrelated to new changes.