zeeke / sriov-cni / 31509142621

63%

master: 46%

push

github

CmdDel: handle NSPathNotNSErr to prevent VF leak on corrupted netns

When a netns bind mount gets corrupted — e.g. a non-recursive bind mount
of /run/netns strips the nsfs submounts due to shared mount propagation —
statfs() returns TMPFS_MAGIC (0x01021994) instead of NSFS_MAGIC. In this
case ns.GetNS() returns NSPathNotNSErr, but CmdDel only handles
NSPathNotExistErr. The unhandled error causes CmdDel to fail, which makes
CRI-O retry teardown in a loop. The pod stays stuck and the
sriov-device-plugin keeps allocating the same VF to new sandbox attempts
that all fail with "LoadConf(): VF pci addr is required".

Treat NSPathNotNSErr the same as NSPathNotExistErr: the netns is not
usable, so return success and let CRI-O proceed with cleanup. The VF
cannot be moved back via ReleaseVF (the netns is inaccessible), but it
will be recovered by the kernel once all references to the dead netns are released.

Signed-off-by: Andrea Panattoni <apanatto@redhat.com>

6 of 6 new or added lines in 1 file covered. (100.0%)

972 of 1546 relevant lines covered (62.87%)

8.66 hits per line