linux-kernel - Re: [patch 1/6] md: remove dependency on __GFP

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Mon, 23 Aug 2010 12:51:00 -0700
From:	Andrew Morton <akpm@...ux-foundation.org>
To:	David Rientjes <rientjes@...gle.com>
Cc:	Neil Brown <neilb@...e.de>, Alasdair G Kergon <agk@...hat.com>,
	linux-raid@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [patch 1/6] md: remove dependency on __GFP_NOFAIL

On Mon, 23 Aug 2010 12:35:22 -0700 (PDT)
David Rientjes <rientjes@...gle.com> wrote:

> On Mon, 23 Aug 2010, Andrew Morton wrote:
> 
> > > diff --git a/drivers/md/dm-region-hash.c b/drivers/md/dm-region-hash.c
> > > --- a/drivers/md/dm-region-hash.c
> > > +++ b/drivers/md/dm-region-hash.c
> > > @@ -289,8 +289,12 @@ static struct dm_region *__rh_alloc(struct dm_region_hash *rh, region_t region)
> > >  	struct dm_region *reg, *nreg;
> > >  
> > >  	nreg = mempool_alloc(rh->region_pool, GFP_ATOMIC);
> > > -	if (unlikely(!nreg))
> > > -		nreg = kmalloc(sizeof(*nreg), GFP_NOIO | __GFP_NOFAIL);
> > > +	if (unlikely(!nreg)) {
> > > +		/* FIXME: this may potentially loop forever */
> > > +		do {
> > > +			nreg = kmalloc(sizeof(*nreg), GFP_NOIO);
> > > +		} while (!nreg);
> > > +	}
> > >  
> > >  	nreg->state = rh->log->type->in_sync(rh->log, region, 1) ?
> > >  		      DM_RH_CLEAN : DM_RH_NOSYNC;
> > 
> > erm.
> > 
> > The reason for adding GFP_NOFAIL in the first place was my observation
> > that the kernel contained lots of open-coded retry-for-ever loops.
> > 
> > All of these are wrong, bad, buggy and mustfix.  So we consolidated the
> > wrongbadbuggymustfix concept into the core MM so that miscreants could
> > be easily identified and hopefully fixed.
> > 
> 
> That consolidation would have been unnecessary, then, since all 
> allocations with order < PAGE_ALLOC_COSTLY_ORDER automatically loop 
> indefinitely in the page allocator.

The difference is that an order-0 !__GFP_NOFAIL allocation attempt can
fail due to oom-killing.  Unless someone broke that.

>  struct dm_region allocations would 
> already do that.
> 
> So this retry loop doesn't actually do anything that the page allocator 
> already doesn't, with or without __GFP_NOFAIL.  The difference here is 
> that
> 
>  - it doesn't depend on the page allocator's implementation, which may
>    change over time, and
> 
>  - it adds documentation so that the subsystems doing these loops can
>    (hopefully) fix these problems later, although their appear to be
>    geniune cases where little other options are available.
> 
> > I think that simply undoing that change is a bad idea - it allows the
> > wrongbadbuggymustfix code to hide from view.
> > 
> 
> It removes several branches from the page allocator.

None on the fast path.

> > The correct way to remove __GFP_NOFAIL is to fix the
> > wrongbadbuggymustfix code properly.
> > 
> 
> If the prerequisite for removing __GFP_NOFAIL is that nobody must ever 
> loop indefinitely looking for memory or smaller order allocations don't 
> implicitly retry, then there's little chance it'll ever get removed since 
> they've existed for years without anybody cleaning them up.

The JBD one is hard - I haven't looked at the others.

We should fix them.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/