Btrfs: do not take cleanup_work_sem in btrfs_run_delayed_iputs()
authorSage Weil <sage@inktank.com>
Thu, 30 Aug 2012 22:26:17 +0000 (16:26 -0600)
committerChris Mason <chris.mason@fusionio.com>
Mon, 1 Oct 2012 19:19:08 +0000 (15:19 -0400)
Josef has suggested that this is not necessary.  Removing it also avoids
this lockdep splat (after the new sb_internal locking stuff was added):

[  604.090449] ======================================================
[  604.114819] [ INFO: possible circular locking dependency detected ]
[  604.139262] 3.6.0-rc2-ceph-00144-g463b030 #1 Not tainted
[  604.162193] -------------------------------------------------------
[  604.186139] btrfs-cleaner/6669 is trying to acquire lock:
[  604.209555]  (sb_internal#2){.+.+..}, at: [<ffffffffa0042b84>] start_transaction+0x124/0x430 [btrfs]
[  604.257100]
[  604.257100] but task is already holding lock:
[  604.300366]  (&fs_info->cleanup_work_sem){.+.+..}, at: [<ffffffffa0048002>] btrfs_run_delayed_iputs+0x72/0x130 [btrfs]
[  604.352989]
[  604.352989] which lock already depends on the new lock.
[  604.352989]
[  604.427104]
[  604.427104] the existing dependency chain (in reverse order) is:
[  604.478493]
[  604.478493] -> #1 (&fs_info->cleanup_work_sem){.+.+..}:
[  604.529313]        [<ffffffff810b2c82>] lock_acquire+0xa2/0x140
[  604.559621]        [<ffffffff81632b69>] down_read+0x39/0x4e
[  604.589382]        [<ffffffffa004db98>] btrfs_lookup_dentry+0x218/0x550 [btrfs]
[  604.596161] btrfs: unlinked 1 orphans
[  604.675002]        [<ffffffffa006aadd>] create_subvol+0x62d/0x690 [btrfs]
[  604.708859]        [<ffffffffa006d666>] btrfs_mksubvol.isra.52+0x346/0x3a0 [btrfs]
[  604.772466]        [<ffffffffa006d7f2>] btrfs_ioctl_snap_create_transid+0x132/0x190 [btrfs]
[  604.842245]        [<ffffffffa006d8ae>] btrfs_ioctl_snap_create+0x5e/0x80 [btrfs]
[  604.912852]        [<ffffffffa00708ae>] btrfs_ioctl+0x138e/0x1990 [btrfs]
[  604.951888]        [<ffffffff8118e9b8>] do_vfs_ioctl+0x98/0x560
[  604.989961]        [<ffffffff8118ef11>] sys_ioctl+0x91/0xa0
[  605.026628]        [<ffffffff8163d569>] system_call_fastpath+0x16/0x1b
[  605.064404]
[  605.064404] -> #0 (sb_internal#2){.+.+..}:
[  605.126832]        [<ffffffff810b25e8>] __lock_acquire+0x1ac8/0x1b90
[  605.163671]        [<ffffffff810b2c82>] lock_acquire+0xa2/0x140
[  605.200228]        [<ffffffff8117dac6>] __sb_start_write+0xc6/0x1b0
[  605.236818]        [<ffffffffa0042b84>] start_transaction+0x124/0x430 [btrfs]
[  605.274029]        [<ffffffffa00431a3>] btrfs_start_transaction+0x13/0x20 [btrfs]
[  605.340520]        [<ffffffffa004ccfa>] btrfs_evict_inode+0x19a/0x330 [btrfs]
[  605.378720]        [<ffffffff811972c8>] evict+0xb8/0x1c0
[  605.416057]        [<ffffffff811974d5>] iput+0x105/0x210
[  605.452373]        [<ffffffffa0048082>] btrfs_run_delayed_iputs+0xf2/0x130 [btrfs]
[  605.521627]        [<ffffffffa003b5e1>] cleaner_kthread+0xa1/0x120 [btrfs]
[  605.560520]        [<ffffffff810791ee>] kthread+0xae/0xc0
[  605.598094]        [<ffffffff8163e744>] kernel_thread_helper+0x4/0x10
[  605.636499]
[  605.636499] other info that might help us debug this:
[  605.636499]
[  605.736504]  Possible unsafe locking scenario:
[  605.736504]
[  605.801931]        CPU0                    CPU1
[  605.835126]        ----                    ----
[  605.867093]   lock(&fs_info->cleanup_work_sem);
[  605.898594]                                lock(sb_internal#2);
[  605.931954]                                lock(&fs_info->cleanup_work_sem);
[  605.965359]   lock(sb_internal#2);
[  605.994758]
[  605.994758]  *** DEADLOCK ***
[  605.994758]
[  606.075281] 2 locks held by btrfs-cleaner/6669:
[  606.104528]  #0:  (&fs_info->cleaner_mutex){+.+...}, at: [<ffffffffa003b5d5>] cleaner_kthread+0x95/0x120 [btrfs]
[  606.165626]  #1:  (&fs_info->cleanup_work_sem){.+.+..}, at: [<ffffffffa0048002>] btrfs_run_delayed_iputs+0x72/0x130 [btrfs]
[  606.231297]
[  606.231297] stack backtrace:
[  606.287723] Pid: 6669, comm: btrfs-cleaner Not tainted 3.6.0-rc2-ceph-00144-g463b030 #1
[  606.347823] Call Trace:
[  606.376184]  [<ffffffff8162a77c>] print_circular_bug+0x1fb/0x20c
[  606.409243]  [<ffffffff810b25e8>] __lock_acquire+0x1ac8/0x1b90
[  606.441343]  [<ffffffffa0042b84>] ? start_transaction+0x124/0x430 [btrfs]
[  606.474583]  [<ffffffff810b2c82>] lock_acquire+0xa2/0x140
[  606.505934]  [<ffffffffa0042b84>] ? start_transaction+0x124/0x430 [btrfs]
[  606.539429]  [<ffffffff8132babd>] ? do_raw_spin_unlock+0x5d/0xb0
[  606.571719]  [<ffffffff8117dac6>] __sb_start_write+0xc6/0x1b0
[  606.603498]  [<ffffffffa0042b84>] ? start_transaction+0x124/0x430 [btrfs]
[  606.637405]  [<ffffffffa0042b84>] ? start_transaction+0x124/0x430 [btrfs]
[  606.670165]  [<ffffffff81172e75>] ? kmem_cache_alloc+0xb5/0x160
[  606.702144]  [<ffffffffa0042b84>] start_transaction+0x124/0x430 [btrfs]
[  606.735562]  [<ffffffffa00256a6>] ? block_rsv_add_bytes+0x56/0x80 [btrfs]
[  606.769861]  [<ffffffffa00431a3>] btrfs_start_transaction+0x13/0x20 [btrfs]
[  606.804575]  [<ffffffffa004ccfa>] btrfs_evict_inode+0x19a/0x330 [btrfs]
[  606.838756]  [<ffffffff81634c6b>] ? _raw_spin_unlock+0x2b/0x40
[  606.872010]  [<ffffffff811972c8>] evict+0xb8/0x1c0
[  606.903800]  [<ffffffff811974d5>] iput+0x105/0x210
[  606.935416]  [<ffffffffa0048082>] btrfs_run_delayed_iputs+0xf2/0x130 [btrfs]
[  606.970510]  [<ffffffffa003b5d5>] ? cleaner_kthread+0x95/0x120 [btrfs]
[  607.005648]  [<ffffffffa003b5e1>] cleaner_kthread+0xa1/0x120 [btrfs]
[  607.040724]  [<ffffffffa003b540>] ? btrfs_destroy_delayed_refs.isra.102+0x220/0x220 [btrfs]
[  607.104740]  [<ffffffff810791ee>] kthread+0xae/0xc0
[  607.137119]  [<ffffffff810b379d>] ? trace_hardirqs_on+0xd/0x10
[  607.169797]  [<ffffffff8163e744>] kernel_thread_helper+0x4/0x10
[  607.202472]  [<ffffffff81635430>] ? retint_restore_args+0x13/0x13
[  607.235884]  [<ffffffff81079140>] ? flush_kthread_work+0x1a0/0x1a0
[  607.268731]  [<ffffffff8163e740>] ? gs_change+0x13/0x13

Signed-off-by: Sage Weil <sage@inktank.com>
fs/btrfs/inode.c

index 8cb272c84089c670ddea7bbd85995cbcff357add..073af0724bc024dfc3fb9ebe2ccc9d1cfa567673 100644 (file)
@@ -2118,7 +2118,6 @@ void btrfs_run_delayed_iputs(struct btrfs_root *root)
        if (empty)
                return;
 
-       down_read(&root->fs_info->cleanup_work_sem);
        spin_lock(&fs_info->delayed_iput_lock);
        list_splice_init(&fs_info->delayed_iputs, &list);
        spin_unlock(&fs_info->delayed_iput_lock);
@@ -2129,7 +2128,6 @@ void btrfs_run_delayed_iputs(struct btrfs_root *root)
                iput(delayed->inode);
                kfree(delayed);
        }
-       up_read(&root->fs_info->cleanup_work_sem);
 }
 
 enum btrfs_orphan_cleanup_state {