cgroup: Add annotation for holding namespace_sem in current_cgns_cgroup_from_root()
authorYafang Shao <laoar.shao@gmail.com>
Sun, 29 Oct 2023 06:14:31 +0000 (06:14 +0000)
committerTejun Heo <tj@kernel.org>
Thu, 9 Nov 2023 23:25:47 +0000 (13:25 -1000)
When I initially examined the function current_cgns_cgroup_from_root(), I
was perplexed by its lack of holding cgroup_mutex. However, after Michal
explained the reason[0] to me, I realized that it already holds the
namespace_sem. I believe this intricacy could also confuse others, so it
would be advisable to include an annotation for clarification.

After we replace the cgroup_mutex with RCU read lock, if current doesn't
hold the namespace_sem, the root cgroup will be NULL. So let's add a
WARN_ON_ONCE() for it.

[0]. https://lore.kernel.org/bpf/afdnpo3jz2ic2ampud7swd6so5carkilts2mkygcaw67vbw6yh@5b5mncf7qyet

Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
Cc: Michal Koutny <mkoutny@suse.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
kernel/cgroup/cgroup.c

index 9bb255e41cf2a25862e24618c106912c6e1a38f5..4e610863cc3771ccae619239b5df6ae6c7ddc93e 100644 (file)
@@ -1420,6 +1420,11 @@ current_cgns_cgroup_from_root(struct cgroup_root *root)
 
        rcu_read_unlock();
 
+       /*
+        * The namespace_sem is held by current, so the root cgroup can't
+        * be umounted. Therefore, we can ensure that the res is non-NULL.
+        */
+       WARN_ON_ONCE(!res);
        return res;
 }