[PATCH] cgroup: replace BUG_ON()s in compare_css_sets() with WARN_ON_ONCE()

[PATCH] cgroup: replace BUG_ON()s in compare_css_sets() with WARN_ON_ONCE()

[Shaojie Sun]

compare_css_sets() walks the cgrp_links lists of two css_sets in
lockstep.  Both lists carry exactly one link per hierarchy root, so a
length or ordering mismatch just means the two css_sets are not the same
one, which is what the function already reports with "return false".
The caller keeps walking the hash chain and allocates a new css_set if
nothing matches.

None of these checks is a user-reachable error path; they are invariant
checks on kernel-internal bookkeeping that are expected never to fire,
and each one has a way out.

No functional change for a consistent state.

Signed-off-by: Shaojie Sun <sunshaojie@kylinos.cn>
---
 kernel/cgroup/cgroup.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/kernel/cgroup/cgroup.c b/kernel/cgroup/cgroup.c
index a284225f8dc9..81c82d137fa2 100644
--- a/kernel/cgroup/cgroup.c
+++ b/kernel/cgroup/cgroup.c
@@ -1024,10 +1024,12 @@ static bool compare_css_sets(struct css_set *cset,
 		l2 = l2->next;
 		/* See if we reached the end - both lists are equal length. */
 		if (l1 == &cset->cgrp_links) {
-			BUG_ON(l2 != &old_cset->cgrp_links);
+			if (WARN_ON_ONCE(l2 != &old_cset->cgrp_links))
+				return false;
 			break;
 		} else {
-			BUG_ON(l2 == &old_cset->cgrp_links);
+			if (WARN_ON_ONCE(l2 == &old_cset->cgrp_links))
+				return false;
 		}
 		/* Locate the cgroups associated with these links. */
 		link1 = list_entry(l1, struct cgrp_cset_link, cgrp_link);
@@ -1035,7 +1037,8 @@ static bool compare_css_sets(struct css_set *cset,
 		cgrp1 = link1->cgrp;
 		cgrp2 = link2->cgrp;
 		/* Hierarchies should be linked in the same order. */
-		BUG_ON(cgrp1->root != cgrp2->root);
+		if (WARN_ON_ONCE(cgrp1->root != cgrp2->root))
+			return false;
 
 		/*
 		 * If this hierarchy is the hierarchy of the cgroup
-- 
2.50.1

Re: [PATCH] cgroup: replace BUG_ON()s in compare_css_sets() with WARN_ON_ONCE()

[Tejun Heo]

On Thu, Nov 05, 2026 at 08:48:13PM +0800, Shaojie Sun wrote:
> compare_css_sets() walks the cgrp_links lists of two css_sets in
> lockstep.  Both lists carry exactly one link per hierarchy root, so a
> length or ordering mismatch just means the two css_sets are not the same
> one, which is what the function already reports with "return false".
> The caller keeps walking the hash chain and allocates a new css_set if
> nothing matches.
> 
> None of these checks is a user-reachable error path; they are invariant
> checks on kernel-internal bookkeeping that are expected never to fire,
> and each one has a way out.
> 
> No functional change for a consistent state.

Why are we changing this now? These never fired, and if they ever fire,
we're in a very deep doodoo.

Thanks.

-- 
tejun