]> git.ipfire.org Git - thirdparty/kernel/stable.git/commit
release_task: kill unnecessary rcu_read_lock() around dec_rlimit_ucounts()
authorOleg Nesterov <oleg@redhat.com>
Sun, 26 Oct 2025 14:31:40 +0000 (15:31 +0100)
committerAndrew Morton <akpm@linux-foundation.org>
Wed, 12 Nov 2025 18:00:17 +0000 (10:00 -0800)
commitc25d24d0f4c6c0954f2b0eb1fc69d293e88192bf
tree8e8b2bf1259cbf56feee5795f84c2c0a9d191c0c
parentf9925019f49e75f425623f46f379182dc3a1e8cd
release_task: kill unnecessary rcu_read_lock() around dec_rlimit_ucounts()

rcu_read_lock() was added to shut RCU-lockdep up when this code used
__task_cred()->rcu_dereference(), but after the commit 21d1c5e386bc
("Reimplement RLIMIT_NPROC on top of ucounts") it is no longer needed:
task_ucounts()->task_cred_xxx() takes rcu_read_lock() itself.

NOTE: task_ucounts() returns the pointer to another rcu-protected data,
struct ucounts.  So it should either be used when task->real_cred and thus
task->real_cred->ucounts is stable (release_task, copy_process,
copy_creds), or it should be called under rcu_read_lock().  In both cases
it is pointless to take rcu_read_lock() to read the cred->ucounts pointer.

Link: https://lkml.kernel.org/r/20251026143140.GA22463@redhat.com
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Alexey Gladkov <legion@kernel.org>
Cc: David Howells <dhowells@redhat.com>
Cc: Mateusz Guzik <mjguzik@gmail.com>
Cc: "Paul E . McKenney" <paulmck@kernel.org>
Cc: Kees Cook <kees@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
kernel/exit.c