[dm-devel] [PATCH] dm-bufio: avoid a possible __vmalloc deadlock

[Mike]

Hi

Sorry for my late catch up. Yes, there is a usage on dm_bufio_client_create()

struct dm_buffer *b = alloc_buffer(c, GFP_KERNEL);

Since there is no __GFP_NORETRY, alloc_buffer will fallback to
vmalloc. Also, I suspect here we'd better use GFP_NOIO &
__GFP_NORETRY?

Thanks
Mike


On Mon, Apr 8, 2013 at 6:00 AM, Joe Thornber <thornber at redhat.com> wrote:
> On Fri, Apr 05, 2013 at 03:57:50PM -0400, Mikulas Patocka wrote:
>> Hi
>>
>> This patch avoids a possible deadlock in dm-bufio. (it could happen only
>> with large block size, at most PAGE_SIZE << MAX_ORDER (typically 8MiB).
>
> Does anything currently use such large block sizes? (dm-thin and
> dm-cache stick to 4k blocks).
>
> - Joe