Btrfs: fix deadlock when using free space tree due to block group creation

When modifying the free space tree we can end up COWing one of its extent
buffers which in turn might result in allocating a new chunk, which in
turn can result in flushing (finish creation) of pending block groups. If
that happens we can deadlock because creating a pending block group needs
to update the free space tree, and if any of the updates tries to modify
the same extent buffer that we are COWing, we end up in a deadlock since
we try to write lock twice the same extent buffer.

So fix this by skipping pending block group creation if we are COWing an
extent buffer from the free space tree. This is a case missed by commit
5ce555578e ("Btrfs: fix deadlock when writing out free space caches").

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=202173
Fixes: 5ce555578e ("Btrfs: fix deadlock when writing out free space caches")
CC: stable@vger.kernel.org # 4.18+
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
This commit is contained in:
Filipe Manana 2019-01-08 11:44:41 +00:00 committed by David Sterba
parent d8b5524242
commit a6d8654d88
1 changed files with 9 additions and 7 deletions

View File

@ -1016,19 +1016,21 @@ static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
parent_start = parent->start;
/*
* If we are COWing a node/leaf from the extent, chunk or device trees,
* make sure that we do not finish block group creation of pending block
* groups. We do this to avoid a deadlock.
* If we are COWing a node/leaf from the extent, chunk, device or free
* space trees, make sure that we do not finish block group creation of
* pending block groups. We do this to avoid a deadlock.
* COWing can result in allocation of a new chunk, and flushing pending
* block groups (btrfs_create_pending_block_groups()) can be triggered
* when finishing allocation of a new chunk. Creation of a pending block
* group modifies the extent, chunk and device trees, therefore we could
* deadlock with ourselves since we are holding a lock on an extent
* buffer that btrfs_create_pending_block_groups() may try to COW later.
* group modifies the extent, chunk, device and free space trees,
* therefore we could deadlock with ourselves since we are holding a
* lock on an extent buffer that btrfs_create_pending_block_groups() may
* try to COW later.
*/
if (root == fs_info->extent_root ||
root == fs_info->chunk_root ||
root == fs_info->dev_root)
root == fs_info->dev_root ||
root == fs_info->free_space_root)
trans->can_flush_pending_bgs = false;
cow = btrfs_alloc_tree_block(trans, root, parent_start,