[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <ee1b4215-2846-841f-9f05-6a66e3e7665a@kernel.dk>
Date: Tue, 3 Oct 2017 08:36:16 -0600
From: Jens Axboe <axboe@...nel.dk>
To: Jan Kara <jack@...e.cz>
Cc: linux-kernel@...r.kernel.org, linux-fsdevel@...r.kernel.org,
hannes@...xchg.org, torvalds@...ux-foundation.org
Subject: Re: [PATCH 02/12] buffer: grow_dev_page() should use __GFP_NOFAIL for
all cases
On 10/03/2017 06:25 AM, Jan Kara wrote:
> On Tue 03-10-17 14:10:49, Jan Kara wrote:
>> On Wed 27-09-17 14:13:49, Jens Axboe wrote:
>>> We currently it it for find_or_create_page(), which means that it
>>> cannot fail. Ensure we also pass in 'retry == true' to
>>> alloc_page_buffers(), which also ensure that it cannot fail.
>>>
>>> After this, there are no failure cases in grow_dev_page() that
>>> occur because of a failed memory allocation.
>>>
>>> Signed-off-by: Jens Axboe <axboe@...nel.dk>
>>
>> Makes sense. You can add:
>>
>> Reviewed-by: Jan Kara <jack@...e.cz>
>
> I forgot one question though:
>
>>> page = find_or_create_page(inode->i_mapping, index, gfp_mask);
>>> - if (!page)
>>> - return ret;
>
> Are we sure find_or_create_page() cannot fail for other reasons than
> ENOMEM? Currently it seems to be the case AFAICT but it isn't obvious to me
> that is guaranteed in future as well...
It looks up a page, or allocates and adds one. If we tell the allocator
that we cannot tolerate failure, then I don't see what else would be
able to make it fail. That function is such an integral part of
lookup/create for the page cache.
--
Jens Axboe
Powered by blists - more mailing lists