> On Aug 21, 2017, at 9:10 AM, Shaohua Li <shli@xxxxxxxxxx> wrote:
>
> On Fri, Aug 18, 2017 at 07:00:20AM +0000, Stephane Thiell wrote:
>>
>>> On Aug 17, 2017, at 10:50 PM, Shaohua Li <shli@xxxxxxxxxx> wrote:
>>>
>>> On Fri, Aug 18, 2017 at 04:13:21AM +0000, Stephane Thiell wrote:
>>>>> On Aug 17, 2017, at 3:58 PM, Shaohua Li <shli@xxxxxxxxxx> wrote:
>>>>>
>>>>> On Tue, Aug 15, 2017 at 07:27:44PM +0000, Stephane Thiell wrote:
>>>>>>
>>>>>>> On Aug 15, 2017, at 11:39 AM, Shaohua Li <shli@xxxxxxxxxx> wrote:
>>>>>>>
>>>>>>> On Tue, Aug 15, 2017 at 05:59:38PM +0000, Stephane Thiell wrote:
>>>>>>>>
>>>>>>>>> On Aug 15, 2017, at 2:24 AM, Guoqing Jiang <gqjiang@xxxxxxxx> wrote:
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> On 08/14/2017 01:29 AM, Stephane Thiell wrote:
>>>>>>>>>> Also known as "kernel BUG at drivers/md/raid5.c:527!"
>>>>>>>>>>
>>>>>>>>>> I’m still struggling with this issue in raid5.c. I currently have one crash per week in average on a system using Lustre and 24 x 10-disk raid6 per server. It did happen with or without check running. I now think there is a race at some point where a released stripe head (batched) is added to conf->inactive_list without sh->batch_head being set back to NULL.
>>>>>>>>>>
>>>>>>>>>> Because sh->batch_head is only being set back to NULL in a single function, break_stripe_batch_list(), my guess is that a call to raid5_release_stripe() on the head stripe is performed elsewhere. I didn’t find where this could come from yet. So perhaps sh->batch_head should be just set to NULL in do_release_stripe(), before being added to the inactive list?
>>>>>>>>>>
>>>>>>>>>> Alternatively, I wonder if it would make more sense to set batch_head to NULL where sh->overwrite_disks is also set back to 0 (like in ops_run_biodrain() today).
>>>>>>>>>>
>>>>>>>>>> Because sh->overwrite_disks is also set to 0 in init_stripe(), my current plan is to try this change, at least as a workaround: remove the BUG_ON(sh->batch_head) from init_stripe() and set sh->batch_head = NULL (I don’t think I need to lock on sh->stripe_lock there?). I will get new hardware soon so I will give it a try on a test system. Any suggestions welcomed…
>>>>>>>>>>
>>>>>>>>>> Quick analysis of the last vmcore:
>>>>>>>>>>
>>>>>>>>>> crash> ps | grep ">"
>>>>>>>>>> ...
>>>>>>>>>>> 16119 16111 22 ffff881f93bace70 RU 0.0 150772 10748 e2label
>>>>>>>>>>> 151448 2 34 ffff881ff753bec0 RU 0.0 0 0 [kiblnd_sd_00_00]
>>>>>>>>>>> 151449 2 2 ffff881ff753edd0 RU 0.0 0 0 [kiblnd_sd_00_01]
>>>>>>>>>>> 154637 2 29 ffff881f776b1f60 RU 0.0 0 0 [md32_raid6]
>>>>>>>>>>> 163643 2 11 ffff8820d200bec0 UN 0.0 0 0 [ll_ost_io01_008]
>>>>>>>>>>> 164083 2 13 ffff8820d2753ec0 UN 0.0 0 0 [ll_ost_io01_034]
>>>>>>>>>>> 171389 2 5 ffff883fd072de20 UN 0.0 0 0 [ll_ost_io01_058]
>>>>>>>>>>> 171391 2 17 ffff883fd072ce70 UN 0.0 0 0 [ll_ost_io01_060]
>>>>>>>>>>> 171399 2 33 ffff883fd9e9ce70 UN 0.0 0 0 [ll_ost_io01_068]
>>>>>>>>>>> 237019 2 15 ffff882715628fb0 UN 0.0 0 0 [ll_ost_io01_084]
>>>>>>>>>>> 237033 2 3 ffff881fb0250fb0 UN 0.0 0 0 [ll_ost_io01_098]
>>>>>>>>>> ...
>>>>>>>>>>
>>>>>>>>>> crash> bt 237033
>>>>>>>>>> PID: 237033 TASK: ffff881fb0250fb0 CPU: 3 COMMAND: "ll_ost_io01_098"
>>>>>>>>>> #0 [ffff8826fa3e7240] machine_kexec at ffffffff81059bdb
>>>>>>>>>> #1 [ffff8826fa3e72a0] __crash_kexec at ffffffff811057c2
>>>>>>>>>> #2 [ffff8826fa3e7370] crash_kexec at ffffffff811058b0
>>>>>>>>>> #3 [ffff8826fa3e7388] oops_end at ffffffff8168fac8
>>>>>>>>>> #4 [ffff8826fa3e73b0] die at ffffffff8102e93b
>>>>>>>>>> #5 [ffff8826fa3e73e0] do_trap at ffffffff8168f180
>>>>>>>>>> #6 [ffff8826fa3e7430] do_invalid_op at ffffffff8102b144
>>>>>>>>>> #7 [ffff8826fa3e74e0] invalid_op at ffffffff81698b5e
>>>>>>>>>> [exception RIP: raid5_get_active_stripe+1809]
>>>>>>>>>> RIP: ffffffffa0fefd71 RSP: ffff8826fa3e7590 RFLAGS: 00010086
>>>>>>>>>> RAX: 0000000000000000 RBX: ffff883df051641c RCX: dead000000000200
>>>>>>>>>> RDX: 0000000000000000 RSI: 0000000000000005 RDI: ffff881e08d54d30
>>>>>>>>>> RBP: ffff8826fa3e7638 R8: ffff881e08d54d40 R9: 0000000000000080
>>>>>>>>>> R10: 0000000000000000 R11: 0000000000000000 R12: ffff883df0516400
>>>>>>>>>> R13: ffff881e08d54d30 R14: 0000000000000000 R15: 0000000000000000
>>>>>>>>>> ORIG_RAX: ffffffffffffffff CS: 0010 SS: 0018
>>>>>>>>>> #8 [ffff8826fa3e7640] raid5_make_request at ffffffffa0feff55 [raid456]
>>>>>>>>>> #9 [ffff8826fa3e7730] md_make_request at ffffffff814f838c
>>>>>>>>>> #10 [ffff8826fa3e7790] generic_make_request at ffffffff812eedc9
>>>>>>>>>> #11 [ffff8826fa3e77d8] submit_bio at ffffffff812eef11
>>>>>>>>>> #12 [ffff8826fa3e7830] osd_submit_bio at ffffffffa10cdbcc [osd_ldiskfs]
>>>>>>>>>> #13 [ffff8826fa3e7840] osd_do_bio at ffffffffa10d03c1 [osd_ldiskfs]
>>>>>>>>>> #14 [ffff8826fa3e7960] osd_write_commit at ffffffffa10d0b94 [osd_ldiskfs]
>>>>>>>>>> #15 [ffff8826fa3e7a00] ofd_commitrw_write at ffffffffa120c734 [ofd]
>>>>>>>>>> #16 [ffff8826fa3e7a78] ofd_commitrw at ffffffffa120ff07 [ofd]
>>>>>>>>>> #17 [ffff8826fa3e7b08] obd_commitrw at ffffffffa0c7dc22 [ptlrpc]
>>>>>>>>>> #18 [ffff8826fa3e7b70] tgt_brw_write at ffffffffa0c55fc1 [ptlrpc]
>>>>>>>>>> #19 [ffff8826fa3e7cd8] tgt_request_handle at ffffffffa0c52275 [ptlrpc]
>>>>>>>>>> #20 [ffff8826fa3e7d20] ptlrpc_server_handle_request at ffffffffa0bfe1fb [ptlrpc]
>>>>>>>>>> #21 [ffff8826fa3e7de8] ptlrpc_main at ffffffffa0c022b0 [ptlrpc]
>>>>>>>>>> #22 [ffff8826fa3e7ec8] kthread at ffffffff810b0a4f
>>>>>>>>>> #23 [ffff8826fa3e7f50] ret_from_fork at ffffffff81697218
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> After some disassembly work, I found out that all active tasks are working on the same md (all are raid6):
>>>>>>>>>>
>>>>>>>>>> md32’s (struct r5conf *) is 0xffff883df0516400
>>>>>>>>>>
>>>>>>>>>> Our crashed task:
>>>>>>>>>>
>>>>>>>>>> crash> printf "%s", ((struct r5conf)*0xffff883df0516400).cache_name[0]
>>>>>>>>>> raid6-md32
>>>>>>>>>>
>>>>>>>>>> Example for md32_raid6 task in release_inactive_stripe_list:
>>>>>>>>>>
>>>>>>>>>> crash> bt 154637
>>>>>>>>>> PID: 154637 TASK: ffff881f776b1f60 CPU: 29 COMMAND: "md32_raid6"
>>>>>>>>>> #0 [ffff883ffc985e48] crash_nmi_callback at ffffffff8104d2e2
>>>>>>>>>> #1 [ffff883ffc985e58] nmi_handle at ffffffff8168fc77
>>>>>>>>>> #2 [ffff883ffc985eb0] do_nmi at ffffffff8168fe83
>>>>>>>>>> #3 [ffff883ffc985ef0] end_repeat_nmi at ffffffff8168f093
>>>>>>>>>> [exception RIP: _raw_spin_lock_irqsave+64]
>>>>>>>>>> RIP: ffffffff8168e5f0 RSP: ffff881e8d0cbc60 RFLAGS: 00000006
>>>>>>>>>> RAX: 00000000000031bb RBX: ffff883df0516740 RCX: 000000000000009a
>>>>>>>>>> RDX: 0000000000000098 RSI: 000000000000009a RDI: ffff883df051641c
>>>>>>>>>> RBP: ffff881e8d0cbc60 R8: 0000000000000206 R9: 0000000000000000
>>>>>>>>>> R10: 0000000000000000 R11: 0000000000000007 R12: ffff883df0516400
>>>>>>>>>> R13: ffff883df051641c R14: 0000000000000006 R15: 0000000000000005
>>>>>>>>>> ORIG_RAX: ffffffffffffffff CS: 0010 SS: 0018
>>>>>>>>>> --- <NMI exception stack> ---
>>>>>>>>>> #4 [ffff881e8d0cbc60] _raw_spin_lock_irqsave at ffffffff8168e5f0
>>>>>>>>>> #5 [ffff881e8d0cbc68] release_inactive_stripe_list at ffffffffa0fe5500 [raid456]
>>>>>>>>>> #6 [ffff881e8d0cbcb0] handle_active_stripes at ffffffffa0ff34c2 [raid456]
>>>>>>>>>> #7 [ffff881e8d0cbd60] raid5d at ffffffffa0ff3c28 [raid456]
>>>>>>>>>> #8 [ffff881e8d0cbe50] md_thread at ffffffff814fb745
>>>>>>>>>> #9 [ffff881e8d0cbec8] kthread at ffffffff810b0a4f
>>>>>>>>>> #10 [ffff881e8d0cbf50] ret_from_fork at ffffffff81697218
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> crash> bt 163643 164083 171389 171391 171399 237019 237033 |grep "exception RIP:"
>>>>>>>>>> [exception RIP: _raw_spin_lock_irq+63]
>>>>>>>>>> [exception RIP: _raw_spin_lock_irq+63]
>>>>>>>>>> [exception RIP: _raw_spin_lock_irq+58]
>>>>>>>>>> [exception RIP: _raw_spin_lock_irq+63]
>>>>>>>>>> [exception RIP: _raw_spin_lock_irq+58]
>>>>>>>>>> [exception RIP: _raw_spin_lock_irq+63]
>>>>>>>>>> [exception RIP: raid5_get_active_stripe+1809]
>>>>>>>>>>
>>>>>>>>>> crash> bt 163643 164083 171389 171391 171399 237019 237033 |grep "raid5_get_active_stripe"
>>>>>>>>>> #5 [ffff883bcee93588] raid5_get_active_stripe at ffffffffa0fef6cb [raid456]
>>>>>>>>>> #5 [ffff883ba4af7588] raid5_get_active_stripe at ffffffffa0fef6cb [raid456]
>>>>>>>>>> #5 [ffff88390877b588] raid5_get_active_stripe at ffffffffa0fef6cb [raid456]
>>>>>>>>>> #5 [ffff88390803b588] raid5_get_active_stripe at ffffffffa0fef6cb [raid456]
>>>>>>>>>> #5 [ffff88390831f588] raid5_get_active_stripe at ffffffffa0fef6cb [raid456]
>>>>>>>>>> #5 [ffff8826fa7f7588] raid5_get_active_stripe at ffffffffa0fef6cb [raid456]
>>>>>>>>>> [exception RIP: raid5_get_active_stripe+1809]
>>>>>>>>>>
>>>>>>>>>> crash> bt 163643 164083 171389 171391 171399 237019 237033 |grep R12
>>>>>>>>>> R10: 0000000000000003 R11: 0000000000000000 R12: ffff883df0516400
>>>>>>>>>> R10: 0000000000000001 R11: 0000000000000000 R12: ffff883df0516400
>>>>>>>>>> R10: 0000000000000005 R11: 0000000000000000 R12: ffff883df0516400
>>>>>>>>>> R10: 0000000000000007 R11: 0000000000000000 R12: ffff883df0516400
>>>>>>>>>> R10: 0000000000000009 R11: 0000000000000000 R12: ffff883df0516400
>>>>>>>>>> R10: 0000000000000009 R11: 0000000000000000 R12: ffff883df0516400
>>>>>>>>>> R10: 0000000000000000 R11: 0000000000000000 R12: ffff883df0516400
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> The bad stripe_head (sh) is available at (usually in %r13):
>>>>>>>>>>
>>>>>>>>>> crash> p ((struct stripe_head)*0xffff881e08d54d30).raid_conf
>>>>>>>>>> $98 = (struct r5conf *) 0xffff883df0516400
>>>>>>>>>>
>>>>>>>>>> and batch_head is not NULL... that's the issue:
>
> Last mail appers not hitting the list, resend.
>
> The new dump info does reveal some infos. Not sure if it's the issue you found,
> but I did find a race condition. Please try below patch and report back:
>
> diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
> index ed5cd705b985..35637fe34820 100644
> --- a/drivers/md/raid5.c
> +++ b/drivers/md/raid5.c
> @@ -806,13 +806,19 @@ static void stripe_add_to_batch_list(struct r5conf *conf, struct stripe_head *sh
> }
>
> /*
> + * We must assign batch_head of this stripe within the
> + * batch_lock, otherwise clear_batch_ready of batch head
> + * stripe could clear BATCH_READY bit of this stripe and this
> + * stripe->batch_head doesn't get assigned, which could
> + * confuse clear_batch_ready for this stripe
> + */
> + sh->batch_head = head->batch_head;
> + /*
> * at this point, head's BATCH_READY could be cleared, but we
> * can still add the stripe to batch list
> */
> list_add(&sh->batch_list, &head->batch_list);
> spin_unlock(&head->batch_head->batch_lock);
> -
> - sh->batch_head = head->batch_head;
> } else {
> head->batch_head = head;
> sh->batch_head = head->batch_head;
Awesome! I will apply your patch today on two of my Lustre servers and report back if I see another occurrence, or after some time if it doesn’t show up. We’ll need to wait for at least a couple weeks to be sure this does actually fix the issue I’m seeing.
Thanks much!
Stephane
��.n��������+%������w��{.n�����{����w��ܨ}���Ơz�j:+v�����w����ޙ��&�)ߡ�a����z�ޗ���ݢj��w�f
