Re: [PATCH] slub: fix failure when we delete and create a slab cache

From: Andrew Morton
Date: Fri Jun 15 2018 - 18:58:24 EST


On Fri, 15 Jun 2018 18:25:29 -0400 (EDT) Mikulas Patocka <mpatocka@xxxxxxxxxx> wrote:

> In the kernel 4.17 I removed some code from dm-bufio that did slab cache
> merging (21bb13276768) - both slab and slub support merging caches with
> identical attributes, so dm-bufio now just calls kmem_cache_create and
> relies on implicit merging.
>
> This uncovered a bug in the slub subsystem - if we delete a cache and
> immediatelly create another cache with the same attributes, it fails
> because of duplicate filename in /sys/kernel/slab/. The slub subsystem
> offloads freeing the cache to a workqueue - and if we create the new cache
> before the workqueue runs, it complains because of duplicate filename in
> sysfs.

Huh. Surprised that such an obvious blooper survived this long. I
guess a rapid del+add is uncommon.

> This patch fixes the bug by moving the call of kobject_del from
> sysfs_slab_remove_workfn to shutdown_cache. kobject_del must be called
> while we hold slab_mutex - so that the sysfs entry is deleted before a
> cache with the same attributes could be created.
>
>
> Running device-mapper-test-suite with:

Nice changelog, btw.

> --- linux-2.6.orig/include/linux/slub_def.h
> +++ linux-2.6/include/linux/slub_def.h
> @@ -156,8 +156,12 @@ struct kmem_cache {
>
> #ifdef CONFIG_SYSFS
> #define SLAB_SUPPORTS_SYSFS
> +void sysfs_slab_unlink(struct kmem_cache *);
> void sysfs_slab_release(struct kmem_cache *);
> #else
> +static inline void sysfs_slab_unlink(struct kmem_cache *s)
> +{
> +}
> static inline void sysfs_slab_release(struct kmem_cache *s)
> {
> }

hm, that's pretty old-school. We could replace SLAB_SUPPORTS_SYSFS
with CONFIG_SLAB_SUPPORTS_SYSFS, move the above logic into slab.h and..

> --- linux-2.6.orig/mm/slab_common.c
> +++ linux-2.6/mm/slab_common.c
> @@ -566,10 +566,14 @@ static int shutdown_cache(struct kmem_ca
> list_del(&s->list);
>
> if (s->flags & SLAB_TYPESAFE_BY_RCU) {
> +#ifdef SLAB_SUPPORTS_SYSFS
> + sysfs_slab_unlink(s);
> +#endif
> list_add_tail(&s->list, &slab_caches_to_rcu_destroy);
> schedule_work(&slab_caches_to_rcu_destroy_work);
> } else {
> #ifdef SLAB_SUPPORTS_SYSFS
> + sysfs_slab_unlink(s);
> sysfs_slab_release(s);
> #else
> slab_kmem_cache_release(s);

remove a bunch of ifdefs. But that would be a separate thing.