Re: [PATCH v3] memcg: fix leak on wrong LRU with FUSE

[Johannes Weiner <hannes@cmpxchg.org>]

On Wed, Mar 09, 2011 at 04:48:01PM +0900, KAMEZAWA Hiroyuki wrote:
> On Wed, 9 Mar 2011 15:07:50 +0900
> KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> wrote:
> 
> > > } else {
> > > 	/* shmem */
> > > 	if (PageSwapCache(page)) {
> > > 		..
> > > 	} else {
> > > 		..
> > > 	}
> > > }
> > > 
> > > Otherwise, the page cache will be charged twice.
> > > 
> > 
> > Ahh, thanks. I'll send v3.
> > 
> 
> Okay, this is a fixed one.
> ==
> 
> fs/fuse/dev.c::fuse_try_move_page() does
> 
>    (1) remove a page by ->steal()
>    (2) re-add the page to page cache 
>    (3) link the page to LRU if it was not on LRU at (1)
> 
> This implies the page is _on_ LRU when it's added to radix-tree.
> So, the page is added to  memory cgroup while it's on LRU.
> because LRU is lazy and no one flushs it.
> 
> This is the same behavior as SwapCache and needs special care as
>  - remove page from LRU before overwrite pc->mem_cgroup.
>  - add page to LRU after overwrite pc->mem_cgroup.
> 
> And we need to taking care of pagevec.
> 
> If PageLRU(page) is set before we add PCG_USED bit, the page
> will not be added to memcg's LRU (in short period).
> So, regardlress of PageLRU(page) value before commit_charge(),
> we need to check PageLRU(page) after commit_charge().
> 
> Changelog v2=>v3:
>   - fixed double accounting.
> 
> Changelog v1=>v2:
>   - clean up.
>   - cover !PageLRU() by pagevec case.
> 
> 
> Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>

Thanks for the fix.  I have a few comments below.  Only nitpicks
though, the patch looks correct to me.

Reviewed-by: Johannes Weiner <hannes@cmpxchg.org>

> @@ -2431,9 +2430,28 @@ static void
>  __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
>  					enum charge_type ctype);
>  
> +static void
> +__mem_cgroup_commit_charge_lrucare(struct page *page, struct mem_cgroup *mem,
> +					enum charge_type ctype)
> +{
> +	struct page_cgroup *pc = lookup_page_cgroup(page);
> +	/*
> +	 * In some case, SwapCache, FUSE(splice_buf->radixtree), the page
> +	 * is already on LRU. It means the page may on some other page_cgroup's
> +	 * LRU. Take care of it.
> +	 */
> +	if (unlikely(PageLRU(page)))
> +		mem_cgroup_lru_del_before_commit(page);

Do we need the extra check?  mem_cgroup_lru_del_before_commit() will
do the right thing if the page is not on the list.

> +	__mem_cgroup_commit_charge(mem, page, 1, pc, ctype);
> +	if (unlikely(PageLRU(page)))
> +		mem_cgroup_lru_add_after_commit(page);

Same here, mem_cgroup_lru_add_after_commit() has its own check for
PG_lru.

> @@ -2468,14 +2486,16 @@ int mem_cgroup_cache_charge(struct page 
>  	if (unlikely(!mm))
>  		mm = &init_mm;
>  
> -	if (page_is_file_cache(page))
> -		return mem_cgroup_charge_common(page, mm, gfp_mask,
> -				MEM_CGROUP_CHARGE_TYPE_CACHE);
> -
> +	if (page_is_file_cache(page)) {
> +		ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, &mem, true);
> +		if (ret || !mem)
> +			return ret;
> +		__mem_cgroup_commit_charge_lrucare(page, mem,
> +					MEM_CGROUP_CHARGE_TYPE_CACHE);

I think the comment about why we need to take care of the LRU status
would make more sense here (rather than in the _lrucare function),
because it is here where you make handling the lru a consequence of
the page being a file page.

How about this?

		/*
		 * FUSE reuses pages without going through the final
		 * put that would remove them from the LRU list, make
		 * sure that they get relinked properly.
		 */