Re: [patch] fix up lock order reversal in writeback

[Eric Sandeen <sandeen@redhat.com>]

On 11/18/10 12:28 AM, Andrew Morton wrote:
> On Thu, 18 Nov 2010 17:00:00 +1100 Nick Piggin <npiggin@kernel.dk> wrote:
> 
>> On Wed, Nov 17, 2010 at 07:29:00PM -0800, Andrew Morton wrote:
>>> On Wed, 17 Nov 2010 22:06:13 -0500 "Ted Ts'o" <tytso@mit.edu> wrote:
>>>
>>>> On Wed, Nov 17, 2010 at 05:10:57PM +1100, Nick Piggin wrote:
>>>>> On Tue, Nov 16, 2010 at 11:05:52PM -0600, Eric Sandeen wrote:
>>>>>> On 11/16/10 10:38 PM, Nick Piggin wrote:
>>>>>>>> as for the locking problems ... sorry about that!
>>>>>>>
>>>>>>> That's no problem. So is that an ack? :)
>>>>>>>
>>>>>>
>>>>>> I'd like to test it with the original case it was supposed to solve; will
>>>>>> do that tomorrow.
>>>>>
>>>>> OK, but it shouldn't make much difference, unless there is a lot of
>>>>> strange activity happening on the sb (like mount / umount / remount /
>>>>> freeze / etc).
>>>>
>>>> This makes sense to me as well.
>>>>
>>>> Acked-by: "Theodore Ts'o" <tytso@mit.edu>
>>>>
>>>> So how do we want to send this patch to Linus?  It's a writeback
>>>> change, so through some mm tree?
>>>
>>> It's in my todo pile.  Even though the patch sucks, but not as much as
>>> its changelog does.  Am not particularly happy merging an alleged
>>> bugfix where the bug is, and I quote, "I saw a lock order warning on
>>> ext4 trigger".  I mean, wtf?  How is anyone supposed to review the code
>>> based on that??  Or to understand it a year from now?
>>
>> Sorry bout the confusion, it was supposed to be "i_mutex", and then it
>> would have been a bit more obvious.
>>
>>
>>> When I get to it I'll troll this email thread and might be able to
>>> kludge together a description which might be able to fool people into
>>> thinking it makes sense.
>>
>> "Lock order reversal between s_umount and i_mutex".
>>
>> i_mutex nests inside s_umount in some writeback paths (it was the end
>> io handler to convert unwritten extents IIRC). But hmm, wouldn't that
>> be a bug? We aren't allowed to take i_mutex inside writeback, are we?
> 
> I'm not sure that s_umount versus i_mutex has come up before.
> 
> Logically I'd expect i_mutex to nest inside s_umount.  Because s_umount
> is a per-superblock thing, and i_mutex is a per-file thing, and files
> live under superblocks.  Nesting s_umount outside i_mutex creates
> complex deadlock graphs between the various i_mutexes, I think.
> 
> Someone tell me if btrfs has the same bug, via its call to
> writeback_inodes_sb_nr_if_idle()?
> 
> I don't see why these functions need s_umount at all, if they're called
> from within ->write_begin against an inode on that superblock.  If the
> superblock can get itself disappeared while we're running ->write_begin
> on it, we have problems, no?
> 
> In which case I'd suggest just removing the down_read(s_umount) and
> specifying that the caller must pin the superblock via some means.
> 
> Only we can't do that because we need to hold s_umount until the
> bdi_queue_work() worker has done its work.
> 
> The fact that a call to ->write_begin can randomly return with s_umount
> held, to be randomly released at some random time in the future is a
> bit ugly, isn't it?  write_begin is a pretty low-level, per-inode
> thing.
> 
> It'd be better if we pinned these superblocks via refcounting, not via
> holding s_umount but even then, having ->write_begin randomly bump sb
> refcounts for random periods of time is still pretty ugly.
> 
> What a pickle.
> 
> Can we just delete writeback_inodes_sb_nr_if_idle() and
> writeback_inodes_sb_if_idle()?  The changelog for 17bd55d037a02 is
> pretty handwavy - do we know that deleting these things would make a
> jot of difference?

Really?  I thought it was pretty decent ;)

Anyway, xfstests 204, "Test out ENOSPC flushing on small filesystems."
shows the problem clearly, IIRC.  I should have included that in the
changelog, I suppose, sorry.

-Eric
 
> And why _do_ we need to hold s_umount during the bdi_queue_work()
> handover?  Would simply bumping s_count suffice?
>