linux-kernel - Re: [PATCH 4/6] xfs: use memalloc_nofs_{save,restore} instead of memalloc

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: Windows password security audit tool. GUI, reports in PDF.

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

Date:   Tue, 7 Feb 2017 09:51:50 +1100
From:   Dave Chinner <david@...morbit.com>
To:     Michal Hocko <mhocko@...nel.org>
Cc:     "Darrick J. Wong" <darrick.wong@...cle.com>,
        Matthew Wilcox <willy@...radead.org>,
        Andrew Morton <akpm@...ux-foundation.org>,
        linux-fsdevel@...r.kernel.org, linux-mm@...ck.org,
        djwong@...nel.org, Theodore Ts'o <tytso@....edu>,
        Chris Mason <clm@...com>, David Sterba <dsterba@...e.cz>,
        Jan Kara <jack@...e.cz>, ceph-devel@...r.kernel.org,
        cluster-devel@...hat.com, linux-nfs@...r.kernel.org,
        logfs@...fs.org, linux-xfs@...r.kernel.org,
        linux-ext4@...r.kernel.org, linux-btrfs@...r.kernel.org,
        linux-mtd@...ts.infradead.org, reiserfs-devel@...r.kernel.org,
        linux-ntfs-dev@...ts.sourceforge.net,
        linux-f2fs-devel@...ts.sourceforge.net,
        linux-afs@...ts.infradead.org, LKML <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH 4/6] xfs: use memalloc_nofs_{save,restore} instead of
 memalloc_noio*

On Mon, Feb 06, 2017 at 07:47:43PM +0100, Michal Hocko wrote:
> On Mon 06-02-17 10:32:37, Darrick J. Wong wrote:
> > On Mon, Feb 06, 2017 at 06:44:15PM +0100, Michal Hocko wrote:
> > > On Mon 06-02-17 07:39:23, Matthew Wilcox wrote:
> > > > On Mon, Feb 06, 2017 at 03:07:16PM +0100, Michal Hocko wrote:
> > > > > +++ b/fs/xfs/xfs_buf.c
> > > > > @@ -442,17 +442,17 @@ _xfs_buf_map_pages(
> > > > >  		bp->b_addr = NULL;
> > > > >  	} else {
> > > > >  		int retried = 0;
> > > > > -		unsigned noio_flag;
> > > > > +		unsigned nofs_flag;
> > > > >  
> > > > >  		/*
> > > > >  		 * vm_map_ram() will allocate auxillary structures (e.g.
> > > > >  		 * pagetables) with GFP_KERNEL, yet we are likely to be under
> > > > >  		 * GFP_NOFS context here. Hence we need to tell memory reclaim
> > > > > -		 * that we are in such a context via PF_MEMALLOC_NOIO to prevent
> > > > > +		 * that we are in such a context via PF_MEMALLOC_NOFS to prevent
> > > > >  		 * memory reclaim re-entering the filesystem here and
> > > > >  		 * potentially deadlocking.
> > > > >  		 */
> > > > 
> > > > This comment feels out of date ... how about:
> > > 
> > > which part is out of date?
> > > 
> > > > 
> > > > 		/*
> > > > 		 * vm_map_ram will allocate auxiliary structures (eg page
> > > > 		 * tables) with GFP_KERNEL.  If that tries to reclaim memory
> > > > 		 * by calling back into this filesystem, we may deadlock.
> > > > 		 * Prevent that by setting the NOFS flag.
> > > > 		 */
> > > 
> > > dunno, the previous wording seems clear enough to me. Maybe little bit
> > > more chatty than yours but I am not sure this is worth changing.
> > 
> > I prefer to keep the "...yet we are likely to be under GFP_NOFS..."
> > wording of the old comment because it captures the uncertainty of
> > whether or not we actually are already under NOFS.  If someone actually
> > has audited this code well enough to know for sure then yes let's change
> > the comment, but I haven't gone that far.
> 
> I believe we can drop the memalloc_nofs_save then as well because either
> we are called from a potentially dangerous context and thus we are in
> the nofs scope we we do not need the protection at all.

No, absolutely not. "Belief" is not a sufficient justification for
removing low level deadlock avoidance infrastructure. This code
needs to remain in _xfs_buf_map_pages() until a full audit of the
caller paths is done and we're 100% certain that there are no
lurking deadlocks.

For example, I'm pretty sure we can call into _xfs_buf_map_pages()
outside of a transaction context but with an inode ILOCK held
exclusively. If we then recurse into memory reclaim and try to run a
transaction during reclaim, we have an inverted ILOCK vs transaction
locking order. i.e. we are not allowed to call xfs_trans_reserve()
with an ILOCK held as that can deadlock the log:  log full, locked
inode pins tail of log, inode cannot be flushed because ILOCK is
held by caller waiting for log space to become available....

i.e. there are certain situations where holding a ILOCK is a
deadlock vector. See xfs_lock_inodes() for an example of the lengths
we go to avoid ILOCK based log deadlocks like this...

Cheers,

Dave.
-- 
Dave Chinner
david@...morbit.com