Re: __blkg_lookup oops with 4.2-rcX

[Tejun Heo <tj@kernel.org>]

Hello,

On Sun, Aug 30, 2015 at 08:30:41AM -0400, Josh Boyer wrote:
> Mike and Jeff suggested you be informed of the oops one of our
> community members is hitting in Fedora with 4.2-rcX.  I thought they
> had already sent this upstream to you, but apparently they didn't.
> 
> The latest oops is below.  That is with 4.2-rc8.  I believe the first
> report was against a merge window 4.2 kernel.  The full bug report is
> here: https://bugzilla.redhat.com/show_bug.cgi?id=1237136
> 
> I believe Mike and Jeff suspected the cgroup writeback patches.
> 
> josh
> 
> lvm vgchange -a n
>   /run/lvm/lvmetad.socket: connect failed: No such file or directory
>   WARNING: Failed to connect to lvmetad. Falling back to internal scanning.
> [   36.157672] BUG: unable to handle kernel NULL pointer dereference
> at 0000000000000558
> [   36.157672] IP: [<ffffffff81389746>] __blkg_lookup+0x26/0x70
...
> [   36.157672]  [<ffffffff8138d14a>] blk_throtl_drain+0x5a/0x110
> [   36.157672]  [<ffffffff8138a108>] blkcg_drain_queue+0x18/0x20
> [   36.157672]  [<ffffffff81369a70>] __blk_drain_queue+0xc0/0x170
> [   36.157672]  [<ffffffff8136a101>] blk_queue_bypass_start+0x61/0x80
> [   36.157672]  [<ffffffff81388c59>] blkcg_deactivate_policy+0x39/0x100
> [   36.157672]  [<ffffffff8138d328>] blk_throtl_exit+0x38/0x50
> [   36.157672]  [<ffffffff8138a14e>] blkcg_exit_queue+0x3e/0x50
> [   36.157672]  [<ffffffff8137016e>] blk_release_queue+0x1e/0xc0
> [   36.157672]  [<ffffffff8139bcba>] kobject_release+0x7a/0x190
> [   36.157672]  [<ffffffff8139bb6f>] kobject_put+0x2f/0x60
> [   36.157672]  [<ffffffff8136a2b1>] blk_cleanup_queue+0x111/0x140
> [   36.157672]  [<ffffffff815f13fc>] cleanup_mapped_device+0xdc/0x100
> [   36.157672]  [<ffffffff815f2311>] __dm_destroy+0x161/0x260
> [   36.157672]  [<ffffffff815f45d3>] dm_destroy+0x13/0x20
> [   36.157672]  [<ffffffff815f9ebd>] dev_remove+0x10d/0x170
> [   36.157672]  [<ffffffff815fa572>] ctl_ioctl+0x232/0x4d0
> [   36.157672]  [<ffffffff815fa823>] dm_ctl_ioctl+0x13/0x20
> [   36.157672]  [<ffffffff8122ebd5>] do_vfs_ioctl+0x295/0x470
> [   36.157672]  [<ffffffff8122ee29>] SyS_ioctl+0x79/0x90
> [   36.157672]  [<ffffffff817750ae>] entry_SYSCALL_64_fastpath+0x12/0x71

I think the offending commit is 776687bce42b ("block, blk-mq: draining
can't be skipped even if bypass_depth was non-zero").  It looks like
the patch makes shutdown path travel data structure which is already
destroyed.  Will post the fix soon.

Thanks.

-- 
tejun