btrfs: fix lockdep splat in btrfs_recover_relocation
authorJosef Bacik <josef@toxicpanda.com>
Wed, 16 Dec 2020 16:22:14 +0000 (11:22 -0500)
committerDavid Sterba <dsterba@suse.com>
Mon, 18 Jan 2021 14:44:59 +0000 (15:44 +0100)
While testing the error paths of relocation I hit the following lockdep
splat:

  ======================================================
  WARNING: possible circular locking dependency detected
  5.10.0-rc6+ #217 Not tainted
  ------------------------------------------------------
  mount/779 is trying to acquire lock:
  ffffa0e676945418 (&fs_info->balance_mutex){+.+.}-{3:3}, at: btrfs_recover_balance+0x2f0/0x340

  but task is already holding lock:
  ffffa0e60ee31da8 (btrfs-root-00){++++}-{3:3}, at: __btrfs_tree_read_lock+0x27/0x100

  which lock already depends on the new lock.

  the existing dependency chain (in reverse order) is:

  -> #2 (btrfs-root-00){++++}-{3:3}:
 down_read_nested+0x43/0x130
 __btrfs_tree_read_lock+0x27/0x100
 btrfs_read_lock_root_node+0x31/0x40
 btrfs_search_slot+0x462/0x8f0
 btrfs_update_root+0x55/0x2b0
 btrfs_drop_snapshot+0x398/0x750
 clean_dirty_subvols+0xdf/0x120
 btrfs_recover_relocation+0x534/0x5a0
 btrfs_start_pre_rw_mount+0xcb/0x170
 open_ctree+0x151f/0x1726
 btrfs_mount_root.cold+0x12/0xea
 legacy_get_tree+0x30/0x50
 vfs_get_tree+0x28/0xc0
 vfs_kern_mount.part.0+0x71/0xb0
 btrfs_mount+0x10d/0x380
 legacy_get_tree+0x30/0x50
 vfs_get_tree+0x28/0xc0
 path_mount+0x433/0xc10
 __x64_sys_mount+0xe3/0x120
 do_syscall_64+0x33/0x40
 entry_SYSCALL_64_after_hwframe+0x44/0xa9

  -> #1 (sb_internal#2){.+.+}-{0:0}:
 start_transaction+0x444/0x700
 insert_balance_item.isra.0+0x37/0x320
 btrfs_balance+0x354/0xf40
 btrfs_ioctl_balance+0x2cf/0x380
 __x64_sys_ioctl+0x83/0xb0
 do_syscall_64+0x33/0x40
 entry_SYSCALL_64_after_hwframe+0x44/0xa9

  -> #0 (&fs_info->balance_mutex){+.+.}-{3:3}:
 __lock_acquire+0x1120/0x1e10
 lock_acquire+0x116/0x370
 __mutex_lock+0x7e/0x7b0
 btrfs_recover_balance+0x2f0/0x340
 open_ctree+0x1095/0x1726
 btrfs_mount_root.cold+0x12/0xea
 legacy_get_tree+0x30/0x50
 vfs_get_tree+0x28/0xc0
 vfs_kern_mount.part.0+0x71/0xb0
 btrfs_mount+0x10d/0x380
 legacy_get_tree+0x30/0x50
 vfs_get_tree+0x28/0xc0
 path_mount+0x433/0xc10
 __x64_sys_mount+0xe3/0x120
 do_syscall_64+0x33/0x40
 entry_SYSCALL_64_after_hwframe+0x44/0xa9

  other info that might help us debug this:

  Chain exists of:
    &fs_info->balance_mutex --> sb_internal#2 --> btrfs-root-00

   Possible unsafe locking scenario:

 CPU0                    CPU1
 ----                    ----
    lock(btrfs-root-00);
 lock(sb_internal#2);
 lock(btrfs-root-00);
    lock(&fs_info->balance_mutex);

   *** DEADLOCK ***

  2 locks held by mount/779:
   #0: ffffa0e60dc040e0 (&type->s_umount_key#47/1){+.+.}-{3:3}, at: alloc_super+0xb5/0x380
   #1: ffffa0e60ee31da8 (btrfs-root-00){++++}-{3:3}, at: __btrfs_tree_read_lock+0x27/0x100

  stack backtrace:
  CPU: 0 PID: 779 Comm: mount Not tainted 5.10.0-rc6+ #217
  Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.13.0-2.fc32 04/01/2014
  Call Trace:
   dump_stack+0x8b/0xb0
   check_noncircular+0xcf/0xf0
   ? trace_call_bpf+0x139/0x260
   __lock_acquire+0x1120/0x1e10
   lock_acquire+0x116/0x370
   ? btrfs_recover_balance+0x2f0/0x340
   __mutex_lock+0x7e/0x7b0
   ? btrfs_recover_balance+0x2f0/0x340
   ? btrfs_recover_balance+0x2f0/0x340
   ? rcu_read_lock_sched_held+0x3f/0x80
   ? kmem_cache_alloc_trace+0x2c4/0x2f0
   ? btrfs_get_64+0x5e/0x100
   btrfs_recover_balance+0x2f0/0x340
   open_ctree+0x1095/0x1726
   btrfs_mount_root.cold+0x12/0xea
   ? rcu_read_lock_sched_held+0x3f/0x80
   legacy_get_tree+0x30/0x50
   vfs_get_tree+0x28/0xc0
   vfs_kern_mount.part.0+0x71/0xb0
   btrfs_mount+0x10d/0x380
   ? __kmalloc_track_caller+0x2f2/0x320
   legacy_get_tree+0x30/0x50
   vfs_get_tree+0x28/0xc0
   ? capable+0x3a/0x60
   path_mount+0x433/0xc10
   __x64_sys_mount+0xe3/0x120
   do_syscall_64+0x33/0x40
   entry_SYSCALL_64_after_hwframe+0x44/0xa9

This is straightforward to fix, simply release the path before we setup
the balance_ctl.

CC: stable@vger.kernel.org # 4.4+
Reviewed-by: Qu Wenruo <wqu@suse.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
fs/btrfs/volumes.c

index 2c0aa03..0c7f4f6 100644 (file)
@@ -4318,6 +4318,8 @@ int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
                btrfs_warn(fs_info,
        "balance: cannot set exclusive op status, resume manually");
 
+       btrfs_release_path(path);
+
        mutex_lock(&fs_info->balance_mutex);
        BUG_ON(fs_info->balance_ctl);
        spin_lock(&fs_info->balance_lock);