Re: XFS lock warning, 2.6.29-rc8

Re: XFS lock warning, 2.6.29-rc8

[Dave Chinner]

[ changed cc to xfs@oss.sgi.com. linux-xfs@vger.kernel.org does not
exist. ]

On Sun, Mar 15, 2009 at 02:59:01PM +0200, Denys Fedoryschenko wrote:
> hi
> 
> Here is what i got suddently. But seems everything is working.
> If any other information required, please let me know.

Known issue.

> [39993.560234] the existing dependency chain (in reverse order) is:
> [39993.560236]
> [39993.560236] -> #1 (&mm->mmap_sem){----}:
> [39993.560240]        [<ffffffff8105b9fc>] __lock_acquire+0x12b0/0x161a
> [39993.560244]        [<ffffffff8105bdbb>] lock_acquire+0x55/0x71
> [39993.560247]        [<ffffffff813fc2a5>] down_read+0x34/0x41
> [39993.560252]        [<ffffffff813ffbfe>] do_page_fault+0x3fa/0x848
> [39993.560256]        [<ffffffff813fda2f>] page_fault+0x1f/0x30
> [39993.560258]        [<ffffffff8107e13d>] generic_file_aio_read+0x35f/0x5da
> [39993.560268]        [<ffffffff811650d8>] xfs_read+0x17d/0x1f5
> [39993.560272]        [<ffffffff81161646>] xfs_file_aio_read+0x5f/0x61
> [39993.560275]        [<ffffffff810a88ff>] do_sync_read+0xe7/0x12d
> [39993.560284]        [<ffffffff810a928c>] vfs_read+0xab/0x134
> [39993.560287]        [<ffffffff810a93d9>] sys_read+0x47/0x6f
> [39993.560289]        [<ffffffff8100c0db>] system_call_fastpath+0x16/0x1b
> [39993.560293]        [<ffffffffffffffff>] 0xffffffffffffffff
> [39993.560302]
> [39993.560302] -> #0 (&(&ip->i_iolock)->mr_lock){----}:
> [39993.560306]        [<ffffffff8105b6cb>] __lock_acquire+0xf7f/0x161a
> [39993.560309]        [<ffffffff8105bdbb>] lock_acquire+0x55/0x71
> [39993.560312]        [<ffffffff8104ebde>] down_write_nested+0x37/0x46
> [39993.560316]        [<ffffffff811426aa>] xfs_ilock+0x27/0x79
> [39993.560318]        [<ffffffff8115b40e>] xfs_inactive+0x14d/0x411
> [39993.560322]        [<ffffffff8116552f>] xfs_fs_clear_inode+0x83/0x85
> [39993.560325]        [<ffffffff810ba8f7>] clear_inode+0x8a/0xe3
> [39993.560329]        [<ffffffff810bb044>] generic_delete_inode+0xf8/0x173
> [39993.560332]        [<ffffffff810bb0d6>] generic_drop_inode+0x17/0x1d5
> [39993.560335]        [<ffffffff810ba1c7>] iput+0x61/0x65
> [39993.560338]        [<ffffffff810b7554>] dentry_iput+0xa9/0xc1
> [39993.560341]        [<ffffffff810b7651>] d_kill+0x40/0x60
> [39993.560344]        [<ffffffff810b942d>] dput+0x141/0x14e
> [39993.560347]        [<ffffffff810a9a85>] __fput+0x173/0x198
> [39993.560350]        [<ffffffff810a9ac2>] fput+0x18/0x1a
> [39993.560353]        [<ffffffff8109411e>] remove_vma+0x3e/0x63
> [39993.560355]        [<ffffffff8109503c>] do_munmap+0x2e9/0x30b
> [39993.560358]        [<ffffffff8109509e>] sys_munmap+0x40/0x59
> [39993.560361]        [<ffffffff8100c0db>] system_call_fastpath+0x16/0x1b
> [39993.560363]        [<ffffffffffffffff>] 0xffffffffffffffff

This is a VM problem where it calls fput() with the mmap_sem() held
in remove_vma(). It makes the incorrect assumption that filesystems
will never use the same lock in the IO path and the inode release path.

This can deadlock if you are really unlucky.

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

fput under mmap_sem

[Christoph Hellwig]

On Mon, Mar 16, 2009 at 09:19:21AM +1100, Dave Chinner wrote:
> This is a VM problem where it calls fput() with the mmap_sem() held
> in remove_vma(). It makes the incorrect assumption that filesystems
> will never use the same lock in the IO path and the inode release path.
> 
> This can deadlock if you are really unlucky.

I really wonder why other filesystems haven't hit this yet.  Any chance
we can get the fput moved out of mmap_sem to get rid of this class of
problems?

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: fput under mmap_sem

[Nick Piggin]

On Wednesday 18 March 2009 18:13:13 Christoph Hellwig wrote:
> On Mon, Mar 16, 2009 at 09:19:21AM +1100, Dave Chinner wrote:
> > This is a VM problem where it calls fput() with the mmap_sem() held
> > in remove_vma(). It makes the incorrect assumption that filesystems
> > will never use the same lock in the IO path and the inode release path.
> >
> > This can deadlock if you are really unlucky.
>
> I really wonder why other filesystems haven't hit this yet.  Any chance
> we can get the fput moved out of mmap_sem to get rid of this class of
> problems?

Yes I don't think there is any reason against holding the file refcount
a little longer, but it can be pretty nasty to do in practice because
of deep call chains and sometimes multiple fputs within a given lock
section.

Possibly the easiest and quickest way will be to move aio's deferred
__fput into a usable API and try that. If there are any performance
problems, then we can try to move those fputs out from the mmap_sem
one at a time (eg. I don't expect fput for vma replacement/merging to
often cause the refcount to reach 0, and this is one of the harder
places; wheras fput for a simple munmap might be more often causing
refcount to reach 0 but it should be simpler to move out of mmap_sem
if it is a problem).

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs