btrfs: Fix lockdep warning of wr_ctx->wr_lock in scrub_free_wr_ctx()
authorZhao Lei <zhaolei@cn.fujitsu.com>
Thu, 4 Jun 2015 12:09:15 +0000 (20:09 +0800)
committerChris Mason <clm@fb.com>
Wed, 10 Jun 2015 14:04:52 +0000 (07:04 -0700)
commit20b2e3029eef277cd93a46a991004260057e1a9e
tree043bc01801f7fc64f9910fd19254bb1d7b95e7be
parente1d227a42ea2b4664f94212bd1106b9a3413ffb8
btrfs: Fix lockdep warning of wr_ctx->wr_lock in scrub_free_wr_ctx()

lockdep report following warning in test:
 [25176.843958] =================================
 [25176.844519] [ INFO: inconsistent lock state ]
 [25176.845047] 4.1.0-rc3 #22 Tainted: G        W
 [25176.845591] ---------------------------------
 [25176.846153] inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage.
 [25176.846713] fsstress/26661 [HC0[0]:SC1[1]:HE1:SE0] takes:
 [25176.847246]  (&wr_ctx->wr_lock){+.?...}, at: [<ffffffffa04cdc6d>] scrub_free_ctx+0x2d/0xf0 [btrfs]
 [25176.847838] {SOFTIRQ-ON-W} state was registered at:
 [25176.848396]   [<ffffffff810bf460>] __lock_acquire+0x6a0/0xe10
 [25176.848955]   [<ffffffff810bfd1e>] lock_acquire+0xce/0x2c0
 [25176.849491]   [<ffffffff816489af>] mutex_lock_nested+0x7f/0x410
 [25176.850029]   [<ffffffffa04d04ff>] scrub_stripe+0x4df/0x1080 [btrfs]
 [25176.850575]   [<ffffffffa04d11b1>] scrub_chunk.isra.19+0x111/0x130 [btrfs]
 [25176.851110]   [<ffffffffa04d144c>] scrub_enumerate_chunks+0x27c/0x510 [btrfs]
 [25176.851660]   [<ffffffffa04d3b87>] btrfs_scrub_dev+0x1c7/0x6c0 [btrfs]
 [25176.852189]   [<ffffffffa04e918e>] btrfs_dev_replace_start+0x36e/0x450 [btrfs]
 [25176.852771]   [<ffffffffa04a98e0>] btrfs_ioctl+0x1e10/0x2d20 [btrfs]
 [25176.853315]   [<ffffffff8121c5b8>] do_vfs_ioctl+0x318/0x570
 [25176.853868]   [<ffffffff8121c851>] SyS_ioctl+0x41/0x80
 [25176.854406]   [<ffffffff8164da17>] system_call_fastpath+0x12/0x6f
 [25176.854935] irq event stamp: 51506
 [25176.855511] hardirqs last  enabled at (51506): [<ffffffff810d4ce5>] vprintk_emit+0x225/0x5e0
 [25176.856059] hardirqs last disabled at (51505): [<ffffffff810d4b77>] vprintk_emit+0xb7/0x5e0
 [25176.856642] softirqs last  enabled at (50886): [<ffffffff81067a23>] __do_softirq+0x363/0x640
 [25176.857184] softirqs last disabled at (50949): [<ffffffff8106804d>] irq_exit+0x10d/0x120
 [25176.857746]
 other info that might help us debug this:
 [25176.858845]  Possible unsafe locking scenario:
 [25176.859981]        CPU0
 [25176.860537]        ----
 [25176.861059]   lock(&wr_ctx->wr_lock);
 [25176.861705]   <Interrupt>
 [25176.862272]     lock(&wr_ctx->wr_lock);
 [25176.862881]
  *** DEADLOCK ***

Reason:
 Above warning is caused by:
 Interrupt
 -> bio_endio()
 -> ...
 -> scrub_put_ctx()
 -> scrub_free_ctx() *1
 -> ...
 -> mutex_lock(&wr_ctx->wr_lock);

 scrub_put_ctx() is allowed to be called in end_bio interrupt, but
 in code design, it will never call scrub_free_ctx(sctx) in interrupe
 context(above *1), because btrfs_scrub_dev() get one additional
 reference of sctx->refs, which makes scrub_free_ctx() only called
 withine btrfs_scrub_dev().

 Now the code runs out of our wish, because free sequence in
 scrub_pending_bio_dec() have a gap.

 Current code:
 -----------------------------------+-----------------------------------
 scrub_pending_bio_dec()            |  btrfs_scrub_dev
 -----------------------------------+-----------------------------------
 atomic_dec(&sctx->bios_in_flight); |
 wake_up(&sctx->list_wait);         |
                                    | scrub_put_ctx()
                                    | -> atomic_dec_and_test(&sctx->refs)
 scrub_put_ctx(sctx);               |
 -> atomic_dec_and_test(&sctx->refs)|
 -> scrub_free_ctx()                |
 -----------------------------------+-----------------------------------

 We expected:
 -----------------------------------+-----------------------------------
 scrub_pending_bio_dec()            |  btrfs_scrub_dev
 -----------------------------------+-----------------------------------
 atomic_dec(&sctx->bios_in_flight); |
 wake_up(&sctx->list_wait);         |
 scrub_put_ctx(sctx);               |
 -> atomic_dec_and_test(&sctx->refs)|
                                    | scrub_put_ctx()
                                    | -> atomic_dec_and_test(&sctx->refs)
                                    | -> scrub_free_ctx()
 -----------------------------------+-----------------------------------

Fix:
 Move scrub_pending_bio_dec() to a workqueue, to avoid this function run
 in interrupt context.
 Tested by check tracelog in debug.

Changelog v1->v2:
 Use workqueue instead of adjust function call sequence in v1,
 because v1 will introduce a bug pointed out by:
 Filipe David Manana <fdmanana@gmail.com>

Reported-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
fs/btrfs/async-thread.c
fs/btrfs/async-thread.h
fs/btrfs/ctree.h
fs/btrfs/scrub.c