Cgroup: align with upstream 6.11 CONFIG_MEMCG_V1 CONFIG_CPUSETS_V1

[tixxdz]

What happened?

Tests on bpf-next are failing https://github.com/cilium/tetragon/actions/runs/11500178416/job/32010347721?pr=2615 due to memory controller being split, where v1 is now under its own:

• CONFIG_MEMCG_V1 https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=e93d4166b40a84df83c4d5cb4c709d022808ef9b
• Do not report disabled v1 controllers in /proc/cgroups https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=af000ce85293b8e608f696f0c6c280bc3a75887f

Our cgroup logic works by trying to detect available controllers from /proc/cgroups and uses the memory then pids where they should always be available, however due this change memory v1 controller is not available there, so we switch to use pids, but I suspect that test is failing due to some race conditions as we run the test quickly, however in container load tests are passing which suggest the race.

Anyway since now the memory v1 logic was split, we have to check:

• If memory v1 is not mounted, adapt the userspace detection logic and ensure that we continue to read cgroup names correctly so we work transparently with both cgroupv1 or cgroupv2
  • Added more checks to ensure that cgroupv1 is properly set, and ensured that cgroupv2 also works.
• Fix the failing test and maybe cover new cases
  • We have tests to cover all cases, some already detected this change, and we added new ones.

Tetragon Version

head

Kernel Version

6.11

Kubernetes Version

No response

Bugtool

No response

Relevant log output

No response

Anything else?

No response

[tixxdz]

Fix is here: #3053

[tixxdz]

Merged fix and added more tests to ensure we handle new cgroupv1 configs