[RFC][PATCH 0/3] fix memcg to do swap account in right way (avoid swap account leak)

[KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>]

Thank you all for comments to previous patches. This is new one.
(Please see when you have free time, I don't want to annoy anyone ;)

Major difference of this version to old ones is
 - old ones tries to fix swap handling itself...
 - this one tries to fix memcg's swap accounting.

I like ideas in this patch set. (But I may need more tests.)

Major concept of this patch set for fixing mis-accounting of memcg is
"ignore a ref from swapcache when we uncharge swap."

Consists of following 3 patches. Maybe patch 1/3 can be a concern for people who
don't use memcg.

 [1/3] Adding SWAP_HAS_CACHE flag to swap_map[] array.
 Add an flag to indicate "there is swap cache" instead of "refcnt from swapcache"
 By this, we'll be able to know refcnt to swap without find_get_page(swapper_space).

 [2/3] fix memcg to handle refcnt to swap.
 There is an issue that "all swap references gone but it can't be freed/uncharged
 because its swapcache is not on memcg's LRU".
 To fix this, this patch tries to unaccount swap even if there is swap-cache.
 Need careful tests (and some fix) but I think this is a good way to go.

 This patch uncharge swap account but swp_entry is still used by swap-cache.
 So, some more work to reclaim unnecesary swap-cache will be required (at vm_swap_full())

 [3/3] count # of swap caches with "unused swp_entries".
 This patch just counts # of swap caches whose swp_entry has no reference.
 This counter + vm_swap_full() will allow us to write a function to reclaim
 swp_entry which is unused.

Any comments are welcome.

Thanks,
-Kame

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>