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

[David Rientjes]

On Thu, 30 Jan 2014, Andrew Morton wrote:

> Well gee, how did that one get through?
> 
> What was the point in permanently allocating tmp_name, btw?  "This
> static temporary buffer is used to prevent from pointless shortliving
> allocation"?  That's daft - memcg_create_kmem_cache() is not a fastpath
> and there are a million places in the kernel where we could permanently
> leak memory because it is "pointless" to allocate on demand.
> 
> The allocation of PATH_MAX bytes is unfortunate - kasprintf() wouild
> work well here, but cgroup_name()'s need for rcu_read_lock() screws us
> up.
> 

What's funnier is that tmp_name isn't required at all since 
kmem_cache_create_memcg() is just going to do a kstrdup() on it anyway, so 
you could easily just pass in the pointer to memory that has been 
allocated for s->name rather than allocating memory twice.

> So how about doing this?
> 
> From: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> Subject: mm/memcontrol.c: memcg_create_kmem_cache() tweaks
> 
> Allocate tmp_name on demand rather than permanently consuming PATH_MAX
> bytes of memory.  This permits a small reduction in the mutex hold time as
> well.
> 
> Cc: Glauber Costa <glommer@xxxxxxxxxxxxx>
> Cc: Johannes Weiner <hannes@xxxxxxxxxxx>
> Cc: Michal Hocko <mhocko@xxxxxxx>
> Cc: Vladimir Davydov <vdavydov@xxxxxxxxxxxxx>
> Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> ---
> 
>  mm/memcontrol.c |   11 +++++------
>  1 file changed, 5 insertions(+), 6 deletions(-)
> 
> diff -puN mm/memcontrol.c~mm-memcontrolc-memcg_create_kmem_cache-tweaks mm/memcontrol.c
> --- a/mm/memcontrol.c~mm-memcontrolc-memcg_create_kmem_cache-tweaks
> +++ a/mm/memcontrol.c
> @@ -3401,17 +3401,14 @@ static struct kmem_cache *memcg_create_k
>  						  struct kmem_cache *s)
>  {
>  	struct kmem_cache *new = NULL;
> -	static char *tmp_name = NULL;
> +	static char *tmp_name;

You're keeping it static and the mutex so you're still keeping it global, 
ok...

>  	static DEFINE_MUTEX(mutex);	/* protects tmp_name */
>  
>  	BUG_ON(!memcg_can_account_kmem(memcg));
>  
> -	mutex_lock(&mutex);
>  	/*
> -	 * kmem_cache_create_memcg duplicates the given name and
> -	 * cgroup_name for this name requires RCU context.
> -	 * This static temporary buffer is used to prevent from
> -	 * pointless shortliving allocation.
> +	 * kmem_cache_create_memcg duplicates the given name and cgroup_name()
> +	 * for this name requires rcu_read_lock().
>  	 */
>  	if (!tmp_name) {
>  		tmp_name = kmalloc(PATH_MAX, GFP_KERNEL);

Eek, memory leak.  Two concurrent calls to memcg_create_kmem_cache() find 
!tmp_name and do the kmalloc() concurrently.

> @@ -3419,6 +3416,7 @@ static struct kmem_cache *memcg_create_k
>  			goto out;
>  	}
>  
> +	mutex_lock(&mutex);
>  	rcu_read_lock();
>  	snprintf(tmp_name, PATH_MAX, "%s(%d:%s)", s->name,
>  			 memcg_cache_id(memcg), cgroup_name(memcg->css.cgroup));
> @@ -3432,6 +3430,7 @@ static struct kmem_cache *memcg_create_k
>  		new = s;
>  out:
>  	mutex_unlock(&mutex);
> +	kfree(tmp_name);

Why would we free the global buffer?

>  	return new;
>  }
>  
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/