dm multipath: use updated MPATHF_QUEUE_IO on mapping for bio-based mpath
authorGabriel Krisman Bertazi <krisman@collabora.com>
Tue, 28 Apr 2020 00:39:11 +0000 (20:39 -0400)
committerMike Snitzer <snitzer@redhat.com>
Tue, 28 Apr 2020 23:51:46 +0000 (19:51 -0400)
commit5686dee34dbfe0238c0274e0454fa0174ac0a57a
tree4da147978f63e361e39e169d6233ab714bb30a30
parentad4e80a639fc61d5ecebb03caa5cdbfb91fcebfc
dm multipath: use updated MPATHF_QUEUE_IO on mapping for bio-based mpath

When adding devices that don't have a scsi_dh on a BIO based multipath,
I was able to consistently hit the warning below and lock-up the system.

The problem is that __map_bio reads the flag before it potentially being
modified by choose_pgpath, and ends up using the older value.

The WARN_ON below is not trivially linked to the issue. It goes like
this: The activate_path delayed_work is not initialized for non-scsi_dh
devices, but we always set MPATHF_QUEUE_IO, asking for initialization.
That is fine, since MPATHF_QUEUE_IO would be cleared in choose_pgpath.
Nevertheless, only for BIO-based mpath, we cache the flag before calling
choose_pgpath, and use the older version when deciding if we should
initialize the path.  Therefore, we end up trying to initialize the
paths, and calling the non-initialized activate_path work.

[   82.437100] ------------[ cut here ]------------
[   82.437659] WARNING: CPU: 3 PID: 602 at kernel/workqueue.c:1624
  __queue_delayed_work+0x71/0x90
[   82.438436] Modules linked in:
[   82.438911] CPU: 3 PID: 602 Comm: systemd-udevd Not tainted 5.6.0-rc6+ #339
[   82.439680] RIP: 0010:__queue_delayed_work+0x71/0x90
[   82.440287] Code: c1 48 89 4a 50 81 ff 00 02 00 00 75 2a 4c 89 cf e9
94 d6 07 00 e9 7f e9 ff ff 0f 0b eb c7 0f 0b 48 81 7a 58 40 74 a8 94 74
a7 <0f> 0b 48 83 7a 48 00 74 a5 0f 0b eb a1 89 fe 4c 89 cf e9 c8 c4 07
[   82.441719] RSP: 0018:ffffb738803977c0 EFLAGS: 00010007
[   82.442121] RAX: ffffa086389f9740 RBX: 0000000000000002 RCX: 0000000000000000
[   82.442718] RDX: ffffa086350dd930 RSI: ffffa0863d76f600 RDI: 0000000000000200
[   82.443484] RBP: 0000000000000200 R08: 0000000000000000 R09: ffffa086350dd970
[   82.444128] R10: 0000000000000000 R11: 0000000000000000 R12: ffffa086350dd930
[   82.444773] R13: ffffa0863d76f600 R14: 0000000000000000 R15: ffffa08636738008
[   82.445427] FS:  00007f6abfe9dd40(0000) GS:ffffa0863dd80000(0000) knlGS:00000
[   82.446040] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   82.446478] CR2: 0000557d288db4e8 CR3: 0000000078b36000 CR4: 00000000000006e0
[   82.447104] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[   82.447561] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[   82.448012] Call Trace:
[   82.448164]  queue_delayed_work_on+0x6d/0x80
[   82.448472]  __pg_init_all_paths+0x7b/0xf0
[   82.448714]  pg_init_all_paths+0x26/0x40
[   82.448980]  __multipath_map_bio.isra.0+0x84/0x210
[   82.449267]  __map_bio+0x3c/0x1f0
[   82.449468]  __split_and_process_non_flush+0x14a/0x1b0
[   82.449775]  __split_and_process_bio+0xde/0x340
[   82.450045]  ? dm_get_live_table+0x5/0xb0
[   82.450278]  dm_process_bio+0x98/0x290
[   82.450518]  dm_make_request+0x54/0x120
[   82.450778]  generic_make_request+0xd2/0x3e0
[   82.451038]  ? submit_bio+0x3c/0x150
[   82.451278]  submit_bio+0x3c/0x150
[   82.451492]  mpage_readpages+0x129/0x160
[   82.451756]  ? bdev_evict_inode+0x1d0/0x1d0
[   82.452033]  read_pages+0x72/0x170
[   82.452260]  __do_page_cache_readahead+0x1ba/0x1d0
[   82.452624]  force_page_cache_readahead+0x96/0x110
[   82.452903]  generic_file_read_iter+0x84f/0xae0
[   82.453192]  ? __seccomp_filter+0x7c/0x670
[   82.453547]  new_sync_read+0x10e/0x190
[   82.453883]  vfs_read+0x9d/0x150
[   82.454172]  ksys_read+0x65/0xe0
[   82.454466]  do_syscall_64+0x4e/0x210
[   82.454828]  entry_SYSCALL_64_after_hwframe+0x49/0xbe
[...]
[   82.462501] ---[ end trace bb39975e9cf45daa ]---

Cc: stable@vger.kernel.org
Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
drivers/md/dm-mpath.c