Re: [RFC PATCH 1/2] ext4: Fix possible deadlock with local interrupts disabled and page-draining IPI

[Nikolay Borisov <kernel@kyup.com>]

On 10/13/2015 04:14 PM, Jan Kara wrote:
> On Tue 13-10-15 13:37:16, Nikolay Borisov wrote:
>>
>>
>> On 10/13/2015 11:15 AM, Jan Kara wrote:
>>> On Mon 12-10-15 17:51:07, Nikolay Borisov wrote:
>>>> Hello and thanks for the reply,
>>>>
>>>> On 10/12/2015 04:40 PM, Jan Kara wrote:
>>>>> On Fri 09-10-15 11:03:30, Nikolay Borisov wrote:
>>>>>> On 10/09/2015 10:37 AM, Hillf Danton wrote:
>>>>>>>>>> @@ -109,8 +109,8 @@ static void ext4_finish_bio(struct bio *bio)
>>>>>>>>>>  			if (bio->bi_error)
>>>>>>>>>>  				buffer_io_error(bh);
>>>>>>>>>>  		} while ((bh = bh->b_this_page) != head);
>>>>>>>>>> -		bit_spin_unlock(BH_Uptodate_Lock, &head->b_state);
>>>>>>>>>>  		local_irq_restore(flags);
>>>>>>>>>
>>>>>>>>> What if it takes 100ms to unlock after IRQ restored?
>>>>>>>>
>>>>>>>> I'm not sure I understand in what direction you are going? Care to
>>>>>>>> elaborate?
>>>>>>>>
>>>>>>> Your change introduces extra time cost the lock waiter has to pay in
>>>>>>> the case that irq happens before the lock is released.
>>>>>>
>>>>>> [CC filesystem and mm people. For reference the thread starts here:
>>>>>>  http://thread.gmane.org/gmane.linux.kernel/2056996 ]
>>>>>>
>>>>>> Right, I see what you mean and it's a good point but when doing the
>>>>>> patches I was striving for correctness and starting a discussion, hence
>>>>>> the RFC. In any case I'd personally choose correctness over performance
>>>>>> always ;).
>>>>>>
>>>>>> As I'm not an fs/ext4 expert and have added the relevant parties (please
>>>>>> use reply-all from now on so that the thread is not being cut in the
>>>>>> middle) who will be able to say whether it impact is going to be that
>>>>>> big. I guess in this particular code path worrying about this is prudent
>>>>>> as writeback sounds like a heavily used path.
>>>>>>
>>>>>> Maybe the problem should be approached from a different angle e.g.
>>>>>> drain_all_pages and its reliance on the fact that the IPI will always be
>>>>>> delivered in some finite amount of time? But what if a cpu with disabled
>>>>>> interrupts is waiting on the task issuing the IPI?
>>>>>
>>>>> So I have looked through your patch and also original report (thread starts
>>>>> here: https://lkml.org/lkml/2015/10/8/341) and IMHO one question hasn't
>>>>> been properly answered yet: Who is holding BH_Uptodate_Lock we are spinning
>>>>> on? You have suggested in https://lkml.org/lkml/2015/10/8/464 that it was
>>>>> __block_write_full_page_endio() call but that cannot really be the case.
>>>>> BH_Uptodate_Lock is used only in IO completion handlers -
>>>>> end_buffer_async_read, end_buffer_async_write, ext4_finish_bio. So there
>>>>> really should be some end_io function running on some other CPU which holds
>>>>> BH_Uptodate_Lock for that buffer.
>>>>
>>>> I did check all the call traces of the current processes on the machine
>>>> at the time of the hard lockup and none of the 3 functions you mentioned
>>>> were in any of the call chains. But while I was looking the code of
>>>> end_buffer_async_write and in the comments I saw it was mentioned that
>>>> those completion handler were called from __block_write_full_page_endio
>>>> so that's what pointed my attention to that function. But you are right
>>>> that it doesn't take the BH lock.
>>>>
>>>> Furthermore the fact that the BH_Async_Write flag is set points me in
>>>> the direction that end_buffer_async_write should have been executing but
>>>> as I said issuing "bt" for all the tasks didn't show this function.
>>>
>>> Actually ext4_bio_write_page() also sets BH_Async_Write so that seems like
>>> a more likely place where that flag got set since ext4_finish_bio() was
>>> then handling IO completion.
>>>
>>>> I'm beginning to wonder if it's possible that a single bit memory error
>>>> has crept up, but this still seems like a long shot...
>>>
>>> Yup. Possible but a long shot. Is the problem reproducible in any way?
>>
>> Okay, I rule out hardware issue since a different server today 
>> experienced the same hard lockup. One thing which looks 
>> suspicious to me are the repetitions of bio_endio/clone_endio: 
>>
>> Oct 13 03:16:54 10.80.5.48 Call Trace:
>> Oct 13 03:16:54 10.80.5.48 <NMI>
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff81651631>] dump_stack+0x58/0x7f
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff81089a6c>] warn_slowpath_common+0x8c/0xc0
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff81089b56>] warn_slowpath_fmt+0x46/0x50
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff811015f8>] watchdog_overflow_callback+0x98/0xc0
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff81132d0c>] __perf_event_overflow+0x9c/0x250
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff81133664>] perf_event_overflow+0x14/0x20
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff81061796>] intel_pmu_handle_irq+0x1d6/0x3e0
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff8105b4c4>] perf_event_nmi_handler+0x34/0x60
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff8104c152>] nmi_handle+0xa2/0x1a0
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff8104c3b4>] do_nmi+0x164/0x430
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff81656e2e>] end_repeat_nmi+0x1a/0x1e
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff8125be19>] ? ext4_finish_bio+0x279/0x2a0
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff8125be19>] ? ext4_finish_bio+0x279/0x2a0
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff8125be19>] ? ext4_finish_bio+0x279/0x2a0
>> Oct 13 03:16:54 10.80.5.48 <<EOE>> 
>> Oct 13 03:16:54 10.80.5.48 <IRQ> 
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff8125c2c8>] ext4_end_bio+0xc8/0x120
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff811dbf1d>] bio_endio+0x1d/0x40
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff81546781>] dec_pending+0x1c1/0x360
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff81546996>] clone_endio+0x76/0xa0
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff811dbf1d>] bio_endio+0x1d/0x40
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff81546781>] dec_pending+0x1c1/0x360
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff81546996>] clone_endio+0x76/0xa0
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff811dbf1d>] bio_endio+0x1d/0x40
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff81546781>] dec_pending+0x1c1/0x360
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff81546996>] clone_endio+0x76/0xa0
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff811dbf1d>] bio_endio+0x1d/0x40
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff812fad2b>] blk_update_request+0x21b/0x450
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff810e7797>] ? generic_exec_single+0xa7/0xb0
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff812faf87>] blk_update_bidi_request+0x27/0xb0
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff810e7817>] ? __smp_call_function_single+0x77/0x120
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff812fcc7f>] blk_end_bidi_request+0x2f/0x80
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff812fcd20>] blk_end_request+0x10/0x20
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff813fdc1c>] scsi_io_completion+0xbc/0x620
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff813f57f9>] scsi_finish_command+0xc9/0x130
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff813fe2e7>] scsi_softirq_done+0x147/0x170
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff813035ad>] blk_done_softirq+0x7d/0x90
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff8108ed87>] __do_softirq+0x137/0x2e0
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff81658a0c>] call_softirq+0x1c/0x30
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff8104a35d>] do_softirq+0x8d/0xc0
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff8108e925>] irq_exit+0x95/0xa0
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff81658f76>] do_IRQ+0x66/0xe0
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff816567ef>] common_interrupt+0x6f/0x6f
>> Oct 13 03:16:54 10.80.5.48 <EOI> 
>> Oct 13 03:16:54 10.80.5.48 [<ffffffff81656836>] ? retint_swapgs+0xe/0x13
>> Oct 13 03:16:54 10.80.5.48 ---[ end trace 4a0584a583c66b92 ]---
>>
>> Doing addr2line on ffffffff8125c2c8 shows:
>> /home/projects/linux-stable/fs/ext4/page-io.c:335 which for me is the
>> last bio_put in ext4_end_bio. However, the ? addresses, right at the
>> beginning of the NMI stack (ffffffff8125be19) map to inner loop in
>> bit_spin_lock:
>>
>> } while (test_bit(bitnum, addr));
>>
>> and this is in line with my initial bug report. 
> 
> OK.
> 
>> Unfortunately I wasn't able to acquire a crashdump since the machine
>> hard-locked way too fast.
>>
>> On a slightly different note is it possible to
>> panic the machine via NMIs? Since if all the CPUs are hard lockedup they
>> cannot process sysrq interrupts?
> 
> Certainly it's possible to do that - the easiest way is actually to use
> 
> nmi_watchdog=panic
> 
> Then panic will automatically trigger when watchdog fires.
> 
>>>> Btw I think in any case the spin_lock patch is wrong as this code can be
>>>> called from within softirq context and we do want to be interrupt safe
>>>> at that point.
>>>
>>> Agreed, that patch is definitely wrong.
>>>
>>>>> BTW: I suppose the filesystem uses 4k blocksize, doesn't it?
>>>>
>>>> Unfortunately I cannot tell you with 100% certainty, since on this
>>>> server there are multiple block devices with blocksize either 1k or 4k.
>>>> So it is one of these. If you know a way to extract this information
>>>> from a vmcore file I'd be happy to do it.
>>>
>>> Well, if you have a crashdump, then bh->b_size is the block size. So just
>>> check that for the bh we are spinning on.
>>
>> Turns out in my original email the bh->b_size was shown : 
>> b_size = 0x400 == 1k. So the filesystem is not 4k but 1k. 
> 
> OK, then I have a theory. We can manipulate bh->b_state in a non-atomic
> manner in _ext4_get_block(). If we happen to do that on the first buffer in
> a page while IO completes on another buffer in the same page, we could in
> theory mess up and miss clearing of BH_Uptodate_Lock flag. Can you try
> whether the attached patch fixes your problem?

I just want to make sure I have fully understood the problem. So the way
I understand what you have said is that since the blocksize is 1k this
potentially means we might need up to 4 buffer heads to map everything
in a page. But as you mentioned the locking in ext4_finish_bio is done
only on the first buffer_head in this set of bh. At the same time in
_ext4_get_block bh->b_state is modified non-atomically as you said, and
this can happen in one of those 4bh used to map the whole page? Correct?

If my reasoning is correct then I see no reason why this corruption
couldn't happen even with blocksize == pagesize since ext4_get_block's
non-atomic modification of the bh->b_state could still race with
ext4_finish_bio's atomic modification (even though ext4_finish_bio would
just work on a single buffer head)?

Sorry if the question seems stupid but I just want to fully wrap my head
around the internals of it.

Thanks

> 
> 								Honza
>