btrfs: zoned: traverse devices under chunk_mutex in btrfs_can_activate_zone
authorJohannes Thumshirn <johannes.thumshirn@wdc.com>
Mon, 7 Mar 2022 10:47:17 +0000 (02:47 -0800)
committerDavid Sterba <dsterba@suse.com>
Wed, 23 Mar 2022 22:34:09 +0000 (23:34 +0100)
commit0b9e66762aa0cda2a9c2d5542d64e04dac528fa6
tree175649b3cf0dc35fde44aaa5f0185dd39290e92f
parentd3e29967079c522ce1c5cab0e9fab2c280b977eb
btrfs: zoned: traverse devices under chunk_mutex in btrfs_can_activate_zone

btrfs_can_activate_zone() can be called with the device_list_mutex already
held, which will lead to a deadlock:

insert_dev_extents() // Takes device_list_mutex
`-> insert_dev_extent()
 `-> btrfs_insert_empty_item()
  `-> btrfs_insert_empty_items()
   `-> btrfs_search_slot()
    `-> btrfs_cow_block()
     `-> __btrfs_cow_block()
      `-> btrfs_alloc_tree_block()
       `-> btrfs_reserve_extent()
        `-> find_free_extent()
         `-> find_free_extent_update_loop()
          `-> can_allocate_chunk()
           `-> btrfs_can_activate_zone() // Takes device_list_mutex again

Instead of using the RCU on fs_devices->device_list we
can use fs_devices->alloc_list, protected by the chunk_mutex to traverse
the list of active devices.

We are in the chunk allocation thread. The newer chunk allocation
happens from the devices in the fs_device->alloc_list protected by the
chunk_mutex.

  btrfs_create_chunk()
    lockdep_assert_held(&info->chunk_mutex);
    gather_device_info
      list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list)

Also, a device that reappears after the mount won't join the alloc_list
yet and, it will be in the dev_list, which we don't want to consider in
the context of the chunk alloc.

  [15.166572] WARNING: possible recursive locking detected
  [15.167117] 5.17.0-rc6-dennis #79 Not tainted
  [15.167487] --------------------------------------------
  [15.167733] kworker/u8:3/146 is trying to acquire lock:
  [15.167733] ffff888102962ee0 (&fs_devs->device_list_mutex){+.+.}-{3:3}, at: find_free_extent+0x15a/0x14f0 [btrfs]
  [15.167733]
  [15.167733] but task is already holding lock:
  [15.167733] ffff888102962ee0 (&fs_devs->device_list_mutex){+.+.}-{3:3}, at: btrfs_create_pending_block_groups+0x20a/0x560 [btrfs]
  [15.167733]
  [15.167733] other info that might help us debug this:
  [15.167733]  Possible unsafe locking scenario:
  [15.167733]
  [15.171834]        CPU0
  [15.171834]        ----
  [15.171834]   lock(&fs_devs->device_list_mutex);
  [15.171834]   lock(&fs_devs->device_list_mutex);
  [15.171834]
  [15.171834]  *** DEADLOCK ***
  [15.171834]
  [15.171834]  May be due to missing lock nesting notation
  [15.171834]
  [15.171834] 5 locks held by kworker/u8:3/146:
  [15.171834]  #0: ffff888100050938 ((wq_completion)events_unbound){+.+.}-{0:0}, at: process_one_work+0x1c3/0x5a0
  [15.171834]  #1: ffffc9000067be80 ((work_completion)(&fs_info->async_data_reclaim_work)){+.+.}-{0:0}, at: process_one_work+0x1c3/0x5a0
  [15.176244]  #2: ffff88810521e620 (sb_internal){.+.+}-{0:0}, at: flush_space+0x335/0x600 [btrfs]
  [15.176244]  #3: ffff888102962ee0 (&fs_devs->device_list_mutex){+.+.}-{3:3}, at: btrfs_create_pending_block_groups+0x20a/0x560 [btrfs]
  [15.176244]  #4: ffff8881152e4b78 (btrfs-dev-00){++++}-{3:3}, at: __btrfs_tree_lock+0x27/0x130 [btrfs]
  [15.179641]
  [15.179641] stack backtrace:
  [15.179641] CPU: 1 PID: 146 Comm: kworker/u8:3 Not tainted 5.17.0-rc6-dennis #79
  [15.179641] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.15.0-1.fc35 04/01/2014
  [15.179641] Workqueue: events_unbound btrfs_async_reclaim_data_space [btrfs]
  [15.179641] Call Trace:
  [15.179641]  <TASK>
  [15.179641]  dump_stack_lvl+0x45/0x59
  [15.179641]  __lock_acquire.cold+0x217/0x2b2
  [15.179641]  lock_acquire+0xbf/0x2b0
  [15.183838]  ? find_free_extent+0x15a/0x14f0 [btrfs]
  [15.183838]  __mutex_lock+0x8e/0x970
  [15.183838]  ? find_free_extent+0x15a/0x14f0 [btrfs]
  [15.183838]  ? find_free_extent+0x15a/0x14f0 [btrfs]
  [15.183838]  ? lock_is_held_type+0xd7/0x130
  [15.183838]  ? find_free_extent+0x15a/0x14f0 [btrfs]
  [15.183838]  find_free_extent+0x15a/0x14f0 [btrfs]
  [15.183838]  ? _raw_spin_unlock+0x24/0x40
  [15.183838]  ? btrfs_get_alloc_profile+0x106/0x230 [btrfs]
  [15.187601]  btrfs_reserve_extent+0x131/0x260 [btrfs]
  [15.187601]  btrfs_alloc_tree_block+0xb5/0x3b0 [btrfs]
  [15.187601]  __btrfs_cow_block+0x138/0x600 [btrfs]
  [15.187601]  btrfs_cow_block+0x10f/0x230 [btrfs]
  [15.187601]  btrfs_search_slot+0x55f/0xbc0 [btrfs]
  [15.187601]  ? lock_is_held_type+0xd7/0x130
  [15.187601]  btrfs_insert_empty_items+0x2d/0x60 [btrfs]
  [15.187601]  btrfs_create_pending_block_groups+0x2b3/0x560 [btrfs]
  [15.187601]  __btrfs_end_transaction+0x36/0x2a0 [btrfs]
  [15.192037]  flush_space+0x374/0x600 [btrfs]
  [15.192037]  ? find_held_lock+0x2b/0x80
  [15.192037]  ? btrfs_async_reclaim_data_space+0x49/0x180 [btrfs]
  [15.192037]  ? lock_release+0x131/0x2b0
  [15.192037]  btrfs_async_reclaim_data_space+0x70/0x180 [btrfs]
  [15.192037]  process_one_work+0x24c/0x5a0
  [15.192037]  worker_thread+0x4a/0x3d0

Fixes: a85f05e59bc1 ("btrfs: zoned: avoid chunk allocation if active block group has enough space")
CC: stable@vger.kernel.org # 5.16+
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
fs/btrfs/zoned.c