md/raid5: drain released_stripes before destroying the cache

grow_one_stripe() calls raid5_release_stripe() for each new stripe.
When mddev->thread is already set, that queues the stripe on
conf->released_stripes and wakes the thread. Only raid5d moves that
list onto inactive_list.

Takeover calls setup_conf() while mddev->thread still belongs to the
previous personality, so those stripes stay on released_stripes. If
setup_conf() then fails, shrink_stripes() only frees inactive_list and
kmem_cache_destroy() warns that objects remain.

Move stripes still queued on released_stripes onto inactive_list
before shrink_stripes() frees them.

Fixes: 773ca82fa1 ("raid5: make release_stripe lockless")
Reported-by: syzbot+6a132745caefcf42cb0d@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/linux-raid/6ab6b743.7eec07f4.c25c.0006.GAE@google.com/T/#t
Signed-off-by: Li Youhong <liyouhong@kylinos.cn>
Link: https://patch.msgid.link/20260929083916.50620-1-dayou5941@163.com
Signed-off-by: Yu Kuai <yukuai@fygo.io>
This commit is contained in:
Li Youhong 2026-09-29 16:39:16 +08:00 committed by Yu Kuai
parent 900bf42e0f
commit af560e357a

View File

@ -2713,6 +2713,37 @@ static int drop_one_stripe(struct r5conf *conf)
static void shrink_stripes(struct r5conf *conf)
{
struct list_head temp[NR_STRIPE_HASH_LOCKS];
int i;
/*
* grow_one_stripe() parks new stripes on released_stripes when
* mddev->thread is already set. Takeover leaves the previous
* personality's thread there, so raid5d never moves them to
* inactive_list. Drain them before the cache is destroyed.
* slab_cache is assigned only after device_lock and the hash
* lists exist; earlier abort paths have nothing to drain.
*/
if (conf->slab_cache) {
for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++)
INIT_LIST_HEAD(&temp[i]);
spin_lock_irq(&conf->device_lock);
release_stripe_list(conf, temp);
spin_unlock_irq(&conf->device_lock);
for (i = 0; i < NR_STRIPE_HASH_LOCKS; i++) {
if (list_empty(&temp[i]))
continue;
spin_lock_irq(conf->hash_locks + i);
if (list_empty(conf->inactive_list + i))
atomic_dec(&conf->empty_inactive_list_nr);
list_splice_tail_init(&temp[i],
conf->inactive_list + i);
spin_unlock_irq(conf->hash_locks + i);
}
}
while (conf->max_nr_stripes &&
drop_one_stripe(conf))
;