Re: [RFC 2/4] jbd2: revert must-not-fail allocation loops back to GFP_NOFAIL

[Michal Hocko <mhocko@suse.cz>]

On Mon 02-03-15 12:33:21, David Rientjes wrote:
> On Mon, 2 Mar 2015, Michal Hocko wrote:
> 
> > This basically reverts 47def82672b3 (jbd2: Remove __GFP_NOFAIL from jbd2
> > layer). The deprecation of __GFP_NOFAIL was a bad choice because it led
> > to open coding the endless loop around the allocator rather than
> > removing the dependency on the non failing allocation. So the
> > deprecation was a clear failure and the reality tells us that
> > __GFP_NOFAIL is not even close to go away.
> > 
> > It is still true that __GFP_NOFAIL allocations are generally discouraged
> > and new uses should be evaluated and an alternative (pre-allocations or
> > reservations) should be considered but it doesn't make any sense to lie
> > the allocator about the requirements. Allocator can take steps to help
> > making a progress if it knows the requirements.
> > 
> 
> The changelog should state that this only changes the source code, there 
> is no functional change since alloc_buffer_head() and 
> kmem_cache_zalloc(transaction_cache) are already implicitly nofail due to 
> the allocation order.  The failure code added by the commit you cite are 
> never executed.

Well, even when those allocation would fail the resulting behavior is
basically the same (modulo congestion_wait which imho doesn't make much
difference). So I would prefer not getting that way and simply stay with
the external loop vs. looping within the allocator.

> I agree that if the implementation of the page allocator were to change 
> with respect to PAGE_ALLOC_COSTLY_ORDER that we'd need __GFP_NOFAIL and 
> that such an allocation is better handled in the page allocator.
> 
> > Signed-off-by: Michal Hocko <mhocko@suse.cz>
> 
> Acked-by: David Rientjes <rientjes@google.com>

Thanks

> GFP_NOFS|__GFP_NOFAIL is scary.

Yes it is but as I've learned nothing unusual in the fs land and the
situation should be improved a lot if we go reservation way suggested by
David. Then __GFP_NOFAIL would consume the pre-reserved memory rather
than trigger OOM killer.
-- 
Michal Hocko
SUSE Labs