Re: [PATCH 1/2] fs: fsnotify: account fsnotify metadata to kmemcg

On Tue 26-06-18 12:07:57, Shakeel Butt wrote:

On Tue, Jun 26, 2018 at 11:55 AM Johannes Weiner [off-list ref] wrote:

quoted

On Tue, Jun 26, 2018 at 11:00:53AM -0700, Shakeel Butt wrote:

quoted

On Mon, Jun 25, 2018 at 10:49 PM Amir Goldstein [off-list ref] wrote:

quoted

...

quoted

The verb 'unuse' takes an argument memcg and 'uses' it - too weird.
You can use 'override'/'revert' verbs like override_creds or just call
memalloc_use_memcg(old_memcg) since there is no reference taken
anyway in use_memcg and no reference released in unuse_memcg.

Otherwise looks good to me.

Thanks for your feedback. Just using memalloc_use_memcg(old_memcg) and
ignoring the return seems more simple. I will wait for feedback from
other before changing anything.

We're not nesting calls to memalloc_use_memcg(), right? So we don't
have to return old_memcg and don't have to pass anything to unuse, it
can always set current->active_memcg to NULL.

For buffer_head, the allocation is done with GFP_NOFS. So, I think
there is no chance of nesting. The fsnotify uses GFP_KERNEL but based
on my limited understanding of fsnotify, there should not be any
nesting i.e. the allocation triggering reclaim which trigger fsnotify
events. Though I would like Amir or Jan to confirm there is no nesting
possible.

You are correct. Fsnotify events are generated only as a result of some
syscall, not due to reclaim or stuff like that.

								Honza
-- 
Jan Kara [off-list ref]
SUSE Labs, CR