lkml.org 
[lkml]   [2018]   [Dec]   [16]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH] squashfs: enable __GFP_FS in ->readpage to prevent hang in mem alloc
On Sun, Dec 16, 2018 at 05:38:13PM +0800, Hou Tao wrote:
> Hi,
>
> On 2018/12/15 22:38, Matthew Wilcox wrote:
> > On Tue, Dec 04, 2018 at 10:08:40AM +0800, Hou Tao wrote:
> >> There is no need to disable __GFP_FS in ->readpage:
> >> * It's a read-only fs, so there will be no dirty/writeback page and
> >> there will be no deadlock against the caller's locked page
> >> * It just allocates one page, so compaction will not be invoked
> >> * It doesn't take any inode lock, so the reclamation of inode will be fine
> >>
> >> And no __GFP_FS may lead to hang in __alloc_pages_slowpath() if a
> >> squashfs page fault occurs in the context of a memory hogger, because
> >> the hogger will not be killed due to the logic in __alloc_pages_may_oom().
> >
> > I don't understand your argument here. There's a comment in
> > __alloc_pages_may_oom() saying that we _should_ treat GFP_NOFS
> > specially, but we currently don't.
> I am trying to say that if __GFP_FS is used in pagecache_get_page() when it tries
> to allocate a new page for squashfs, that will be no possibility of dead-lock for
> squashfs.
>
> We do treat GFP_NOFS specially in out_of_memory():
>
> /*
> * The OOM killer does not compensate for IO-less reclaim.
> * pagefault_out_of_memory lost its gfp context so we have to
> * make sure exclude 0 mask - all other users should have at least
> * ___GFP_DIRECT_RECLAIM to get here.
> */
> if (oc->gfp_mask && !(oc->gfp_mask & __GFP_FS))
> return true;
>
> So if GFP_FS is used, no task will be killed because we will return from
> out_of_memory() prematurely. And that will lead to an infinite loop in
> __alloc_pages_slowpath() as we have observed:
>
> * a squashfs page fault occurred in the context of a memory hogger
> * the page used for page fault allocated successfully
> * in squashfs_readpage() squashfs will try to allocate other pages
> in the same 128KB block, and __GFP_NOFS is used (actually GFP_HIGHUSER_MOVABLE & ~__GFP_FS)
> * in __alloc_pages_slowpath() we can not get any pages through reclamation
> (because most of memory is used by the current task) and we also can not kill
> the current task (due to __GFP_NOFS), and it will loop forever until it's killed.

Ah, yes, that makes perfect sense. Thank you for the explanation.

I wonder if the correct fix, however, is not to move the check for
GFP_NOFS in out_of_memory() down to below the check whether to kill
the current task. That would solve your problem, and I don't _think_
it would cause any new ones. Michal, you touched this code last, what
do you think?

\
 
 \ /
  Last update: 2018-12-17 04:52    [W:0.066 / U:0.408 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site