Re: [PATCH] memcg: fix mutex not unlocked on memcg_create_kmem_cache fail path

[Vladimir Davydov <vdavydov@parallels.com>]

Hi, David

Thank you for taking look at this and adding the missing patch
description. WRT your patch, please see the comment inline.

On 01/31/2014 02:39 AM, David Rientjes wrote:
> On Thu, 30 Jan 2014, Andrew Morton wrote:
>
>>> It always was.
>> eh?  kmem_cache_create_memcg()'s kstrdup() will allocate the minimum
>> needed amount of memory.
>>
> Ah, good point.  We could this incrementally on my patch:
>
> diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> --- a/mm/memcontrol.c
> +++ b/mm/memcontrol.c
> @@ -637,6 +637,9 @@ int memcg_limited_groups_array_size;
>   * better kept as an internal representation in cgroup.c. In any case, the
>   * cgrp_id space is not getting any smaller, and we don't have to necessarily
>   * increase ours as well if it increases.
> + *
> + * Updates to MAX_SIZE should update the space for the memcg name in
> + * memcg_create_kmem_cache().
>   */
>  #define MEMCG_CACHES_MIN_SIZE 4
>  #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
> @@ -3400,8 +3403,10 @@ void mem_cgroup_destroy_cache(struct kmem_cache *cachep)
>  static struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg,
>  						  struct kmem_cache *s)
>  {
> -	char *name = NULL;
>  	struct kmem_cache *new;
> +	const char *cgrp_name;
> +	char *name = NULL;
> +	size_t len;
>  
>  	BUG_ON(!memcg_can_account_kmem(memcg));
>  
> @@ -3409,9 +3414,22 @@ static struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg,
>  	if (unlikely(!name))
>  		return NULL;
>  
> +	/*
> +	 * Format of a memcg's kmem cache name:
> +	 * <cache-name>(<memcg-id>:<cgroup-name>)
> +	 */
> +	len = strlen(s->name);
> +	/* Space for parentheses, colon, terminator */
> +	len += 4;
> +	/* MEMCG_CACHES_MAX_SIZE is USHRT_MAX */
> +	len += 5;
> +	BUILD_BUG_ON(MEMCG_CACHES_MAX_SIZE > USHRT_MAX);
> +

This looks cumbersome, IMO. Let's leave it as is for now. AFAIK,
cgroup_name() will be reworked soon so that it won't require RCU-context
(https://lkml.org/lkml/2014/1/28/530). Therefore, it will be possible to
get rid of this pointless tmp_name allocation by making
kmem_cache_create_memcg() take not just name, but printf-like format +
vargs.

Thanks.