sync-related lockdep warnings

sync-related lockdep warnings

[Sage Weil]

Hi-

I've been regularly getting a lockdep warning on inode_lock vs tree->lock.  
It is quickly triggered by my code, which calls ioctl(fd, BTRFS_IOC_SYNC) 
(which just does a btrfs_sync_fs) at regular intervals.

	http://ceph.newdream.net/dump/btrfs-lockdep-sync-ioctl.txt

The second warning is similar, but looks to be a bit more revealing.  It 
is easily triggered by 'while [ 1 ] ; do sync ; done' and then something 
like 'echo a > a' a few times.

	http://ceph.newdream.net/dump/btrfs-lockdep-sync.txt

Let me know if there's any other info on my end that would help sort this 
out...

sage

Re: sync-related lockdep warnings

[Chris Mason]

On Mon, 2008-12-15 at 11:01 -0800, Sage Weil wrote:
> Hi-
> 
> I've been regularly getting a lockdep warning on inode_lock vs tree->lock.  
> It is quickly triggered by my code, which calls ioctl(fd, BTRFS_IOC_SYNC) 
> (which just does a btrfs_sync_fs) at regular intervals.
> 
> 	http://ceph.newdream.net/dump/btrfs-lockdep-sync-ioctl.txt
> 
> The second warning is similar, but looks to be a bit more revealing.  It 
> is easily triggered by 'while [ 1 ] ; do sync ; done' and then something 
> like 'echo a > a' a few times.
> 
> 	http://ceph.newdream.net/dump/btrfs-lockdep-sync.txt
> 
> Let me know if there's any other info on my end that would help sort this 
> out...

Thanks.  I think this is harmless, but I don't actually need do disable
irqs on the delalloc lock.  So, I'm doing some testing here with a
non-irq variant.

-chris

Re: sync-related lockdep warnings

[Chris Mason]

On Mon, 2008-12-15 at 11:01 -0800, Sage Weil wrote:
> Hi-
> 
> I've been regularly getting a lockdep warning on inode_lock vs tree->lock.  
> It is quickly triggered by my code, which calls ioctl(fd, BTRFS_IOC_SYNC) 
> (which just does a btrfs_sync_fs) at regular intervals.
> 
> 	http://ceph.newdream.net/dump/btrfs-lockdep-sync-ioctl.txt
> 
> The second warning is similar, but looks to be a bit more revealing.  It 
> is easily triggered by 'while [ 1 ] ; do sync ; done' and then something 
> like 'echo a > a' a few times.
> 
> 	http://ceph.newdream.net/dump/btrfs-lockdep-sync.txt
> 
> Let me know if there's any other info on my end that would help sort this 
> out...

If you pull from btrfs-unstable, this should be fixed.

Thanks,
Chris

Re: sync-related lockdep warnings

[Chris Mason]

On Mon, 2008-12-15 at 16:01 -0500, Chris Mason wrote:
> On Mon, 2008-12-15 at 11:01 -0800, Sage Weil wrote:
> > Hi-
> > 
> > I've been regularly getting a lockdep warning on inode_lock vs tree->lock.  
> > It is quickly triggered by my code, which calls ioctl(fd, BTRFS_IOC_SYNC) 
> > (which just does a btrfs_sync_fs) at regular intervals.
> > 
> > 	http://ceph.newdream.net/dump/btrfs-lockdep-sync-ioctl.txt
> > 
> > The second warning is similar, but looks to be a bit more revealing.  It 
> > is easily triggered by 'while [ 1 ] ; do sync ; done' and then something 
> > like 'echo a > a' a few times.
> > 
> > 	http://ceph.newdream.net/dump/btrfs-lockdep-sync.txt
> > 
> > Let me know if there's any other info on my end that would help sort this 
> > out...
> 
> If you pull from btrfs-unstable, this should be fixed.

Well, not so much fixed as traded for a different lockdep warning of the
same time.  I've got a new patch in testing here ;)

-chris