cgroup: preserve cgroup2 superblock options on dump

[rocker-zhang]

criu reconfigures the host cgroup2 mount during checkpoint, dropping its superblock options.

When collecting cgroups for dump, __new_open_cgroupfs() opens the unified cgroup2 hierarchy with fsopen("cgroup2") and then cr_fsconfig(FSCONFIG_CMD_CREATE) (or, without fsopen support, mount("none", ..., "cgroup2", 0, NULL)), neither of which carries any superblock option. cgroup2 has a single superblock shared by every mount, so this reconfigures the live host mount and drops options criu did not set, such as nsdelegate, memory_recursiveprot and memory_hugetlb_accounting. The detached-mount teardown only removes criu's own mount instance, so the host cgroup2 is left altered after a checkpoint (#3029).

The fix reads the options of the existing cgroup2 mount from /proc/self/mountinfo and replays the known superblock flags before FSCONFIG_CMD_CREATE (and passes them as mount data in the fsopen-less path), leaving the host mount intact.

I reproduced this in a throwaway VM (its own kernel, so the mutated superblock is the VM's): the current sequence strips nsdelegate,memory_recursiveprot from an existing cgroup2 mount, and the replay preserves them. Builds on x86_64 and aarch64.

Two things I would like direction on. The fix replays an allowlist of known superblock flags; an alternative is to clone the existing mount with open_tree(OPEN_TREE_CLONE) and avoid reconfiguring the superblock at all, which is a larger change. Separately, preserving these options across checkpoint/restore (rather than only avoiding the host mutation) would need a new field in cg_controller_entry, which is an image-format change I have left out of this PR.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 82.35294% with 6 lines in your changes missing coverage. Please review.
✅ Project coverage is 57.10%. Comparing base (fc29bfe) to head (ba6eead).

Files with missing lines	Patch %	Lines
criu/cgroup.c	82.35%	6 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff              @@
##           criu-dev    #3036      +/-   ##
============================================
+ Coverage     57.04%   57.10%   +0.06%     
============================================
  Files           154      154              
  Lines         40534    40566      +32     
  Branches       8882     8892      +10     
============================================
+ Hits          23123    23167      +44     
+ Misses        17057    17045      -12     
  Partials        354      354              

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[avagin]

@Snorch could you please review this pr?

[Snorch]

The problem reproduces:

cat /proc/self/mountinfo  | grep cgroup2
48 46 0:30 / /sys/fs/cgroup rw,nosuid,nodev,noexec,relatime shared:7 - cgroup2 cgroup2 rw,seclabel,nsdelegate,memory_recursiveprot,memory_hugetlb_accounting

python3 test/zdtm.py run -t zdtm/static/cgroup00 --ignore-taint

cat /proc/self/mountinfo  | grep cgroup2
48 46 0:30 / /sys/fs/cgroup rw,nosuid,nodev,noexec,relatime shared:7 - cgroup2 cgroup2 rw,seclabel

[Snorch]

Parsing mountinfo is generally slow (especially host's mountinfo). Preferably we want to avoid doing this one more time (in ideal world we should be switching to listmount+statmount everywhere).

What about alternative approach:

int fd = open("/sys/fs/cgroup", O_PATH);
struct mnt_id_req req = { .size = sizeof(req), .mnt_fd = fd,
                          .param = STATMOUNT_MNT_OPTS };
statmount(&req, sm, bufsz, STATMOUNT_BY_FD);

We will have sb opts in sm->mnt_opts.

Note: Need to verify that it's cgroup2 mount and not something else from other stamount fields.

Yes it only works when cgroup-v2 is mounted into /sys/fs/cgroup on host (which is pretty much always), overwise we have no other option except searching for it which should probably be done via listmount/statmount.