Btrfs: fix hang when loading existing inode cache off disk
authorFilipe Manana <fdmanana@suse.com>
Thu, 4 Jul 2019 15:24:09 +0000 (16:24 +0100)
committerDavid Sterba <dsterba@suse.com>
Mon, 9 Sep 2019 12:59:01 +0000 (14:59 +0200)
If we are able to load an existing inode cache off disk, we set the state
of the cache to BTRFS_CACHE_FINISHED, but we don't wake up any one waiting
for the cache to be available. This means that anyone waiting for the
cache to be available, waiting on the condition that either its state is
BTRFS_CACHE_FINISHED or its available free space is greather than zero,
can hang forever.

This could be observed running fstests with MOUNT_OPTIONS="-o inode_cache",
in particular test case generic/161 triggered it very frequently for me,
producing a trace like the following:

  [63795.739712] BTRFS info (device sdc): enabling inode map caching
  [63795.739714] BTRFS info (device sdc): disk space caching is enabled
  [63795.739716] BTRFS info (device sdc): has skinny extents
  [64036.653886] INFO: task btrfs-transacti:3917 blocked for more than 120 seconds.
  [64036.654079]       Not tainted 5.2.0-rc4-btrfs-next-50 #1
  [64036.654143] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
  [64036.654232] btrfs-transacti D    0  3917      2 0x80004000
  [64036.654239] Call Trace:
  [64036.654258]  ? __schedule+0x3ae/0x7b0
  [64036.654271]  schedule+0x3a/0xb0
  [64036.654325]  btrfs_commit_transaction+0x978/0xae0 [btrfs]
  [64036.654339]  ? remove_wait_queue+0x60/0x60
  [64036.654395]  transaction_kthread+0x146/0x180 [btrfs]
  [64036.654450]  ? btrfs_cleanup_transaction+0x620/0x620 [btrfs]
  [64036.654456]  kthread+0x103/0x140
  [64036.654464]  ? kthread_create_worker_on_cpu+0x70/0x70
  [64036.654476]  ret_from_fork+0x3a/0x50
  [64036.654504] INFO: task xfs_io:3919 blocked for more than 120 seconds.
  [64036.654568]       Not tainted 5.2.0-rc4-btrfs-next-50 #1
  [64036.654617] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
  [64036.654685] xfs_io          D    0  3919   3633 0x00000000
  [64036.654691] Call Trace:
  [64036.654703]  ? __schedule+0x3ae/0x7b0
  [64036.654716]  schedule+0x3a/0xb0
  [64036.654756]  btrfs_find_free_ino+0xa9/0x120 [btrfs]
  [64036.654764]  ? remove_wait_queue+0x60/0x60
  [64036.654809]  btrfs_create+0x72/0x1f0 [btrfs]
  [64036.654822]  lookup_open+0x6bc/0x790
  [64036.654849]  path_openat+0x3bc/0xc00
  [64036.654854]  ? __lock_acquire+0x331/0x1cb0
  [64036.654869]  do_filp_open+0x99/0x110
  [64036.654884]  ? __alloc_fd+0xee/0x200
  [64036.654895]  ? do_raw_spin_unlock+0x49/0xc0
  [64036.654909]  ? do_sys_open+0x132/0x220
  [64036.654913]  do_sys_open+0x132/0x220
  [64036.654926]  do_syscall_64+0x60/0x1d0
  [64036.654933]  entry_SYSCALL_64_after_hwframe+0x49/0xbe

Fix this by adding a wake_up() call right after setting the cache state to
BTRFS_CACHE_FINISHED, at start_caching(), when we are able to load the
cache from disk.

Fixes: 82d5902d9c681b ("Btrfs: Support reading/writing on disk free ino cache")
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
fs/btrfs/inode-map.c

index 2e8bb402050b9593a4f88b7224e2b73618479b8b..84b2c9ee52a74641b6629674761dddd5b6a24d9c 100644 (file)
@@ -146,6 +146,7 @@ static void start_caching(struct btrfs_root *root)
                spin_lock(&root->ino_cache_lock);
                root->ino_cache_state = BTRFS_CACHE_FINISHED;
                spin_unlock(&root->ino_cache_lock);
+               wake_up(&root->ino_cache_wait);
                return;
        }