oomd: fix race with path unavailability when killing cgroups

There can be a situation where systemd-oomd would kill all of the processes
in a cgroup, pid1 would clean up that cgroup, and systemd-oomd would get
ENODEV trying to iterate the cgroup a final time to ensure it was empty.
systemd-oomd sees this as an error and immediately picks a new candidate even
though pressure may have recovered. To counter this, check and handle
path unavailability errnos specially.

Fixes: #22030

diff --git a/src/oom/oomd-util.c b/src/oom/oomd-util.c
index 64ea8cf7e43201c2203273ac321459b39c4774c4..b54bf483d603ea763a9435dc8cd5b850049ce67c 100644 (file)
--- a/src/oom/oomd-util.c
+++ b/src/oom/oomd-util.c
@@ -196,7 +196,14 @@ int oomd_cgroup_kill(const char *path, bool recurse, bool dry_run) {
                 r = cg_kill_recursive(SYSTEMD_CGROUP_CONTROLLER, path, SIGKILL, CGROUP_IGNORE_SELF, pids_killed, log_kill, NULL);
         else
                 r = cg_kill(SYSTEMD_CGROUP_CONTROLLER, path, SIGKILL, CGROUP_IGNORE_SELF, pids_killed, log_kill, NULL);
-        if (r < 0)
+
+        /* The cgroup could have been cleaned up after we have sent SIGKILL to all of the processes, but before
+         * we could do one last iteration of cgroup.procs to check. Or the service unit could have exited and
+         * was removed between picking candidates and coming into this function. In either case, let's log
+         * about it let the caller decide what to do once they know how many PIDs were killed. */
+        if (IN_SET(r, -ENOENT, -ENODEV))
+                log_debug_errno(r, "Error when sending SIGKILL to processes in cgroup path %s, ignoring: %m", path);
+        else if (r < 0)
                 return r;
 
         r = increment_oomd_xattr(path, "user.oomd_kill", set_size(pids_killed));