Btrfs: fix deadlock on tree root leaf when finding free extent

When we are writing out a free space cache, during the transaction commit
phase, we can end up in a deadlock which results in a stack trace like the
following:

 schedule+0x28/0x80
 btrfs_tree_read_lock+0x8e/0x120 [btrfs]
 ? finish_wait+0x80/0x80
 btrfs_read_lock_root_node+0x2f/0x40 [btrfs]
 btrfs_search_slot+0xf6/0x9f0 [btrfs]
 ? evict_refill_and_join+0xd0/0xd0 [btrfs]
 ? inode_insert5+0x119/0x190
 btrfs_lookup_inode+0x3a/0xc0 [btrfs]
 ? kmem_cache_alloc+0x166/0x1d0
 btrfs_iget+0x113/0x690 [btrfs]
 __lookup_free_space_inode+0xd8/0x150 [btrfs]
 lookup_free_space_inode+0x5b/0xb0 [btrfs]
 load_free_space_cache+0x7c/0x170 [btrfs]
 ? cache_block_group+0x72/0x3b0 [btrfs]
 cache_block_group+0x1b3/0x3b0 [btrfs]
 ? finish_wait+0x80/0x80
 find_free_extent+0x799/0x1010 [btrfs]
 btrfs_reserve_extent+0x9b/0x180 [btrfs]
 btrfs_alloc_tree_block+0x1b3/0x4f0 [btrfs]
 __btrfs_cow_block+0x11d/0x500 [btrfs]
 btrfs_cow_block+0xdc/0x180 [btrfs]
 btrfs_search_slot+0x3bd/0x9f0 [btrfs]
 btrfs_lookup_inode+0x3a/0xc0 [btrfs]
 ? kmem_cache_alloc+0x166/0x1d0
 btrfs_update_inode_item+0x46/0x100 [btrfs]
 cache_save_setup+0xe4/0x3a0 [btrfs]
 btrfs_start_dirty_block_groups+0x1be/0x480 [btrfs]
 btrfs_commit_transaction+0xcb/0x8b0 [btrfs]

At cache_save_setup() we need to update the inode item of a block group's
cache which is located in the tree root (fs_info->tree_root), which means
that it may result in COWing a leaf from that tree. If that happens we
need to find a free metadata extent and while looking for one, if we find
a block group which was not cached yet we attempt to load its cache by
calling cache_block_group(). However this function will try to load the
inode of the free space cache, which requires finding the matching inode
item in the tree root - if that inode item is located in the same leaf as
the inode item of the space cache we are updating at cache_save_setup(),
we end up in a deadlock, since we try to obtain a read lock on the same
extent buffer that we previously write locked.

So fix this by using the tree root's commit root when searching for a
block group's free space cache inode item when we are attempting to load
a free space cache. This is safe since block groups once loaded stay in
memory forever, as well as their caches, so after they are first loaded
we will never need to read their inode items again. For new block groups,
once they are created they get their ->cached field set to
BTRFS_CACHE_FINISHED meaning we will not need to read their inode item.

Reported-by: Andrew Nelson <andrew.s.nelson@gmail.com>
Link: https://lore.kernel.org/linux-btrfs/CAPTELenq9x5KOWuQ+fa7h1r3nsJG8vyiTH8+ifjURc_duHh2Wg@mail.gmail.com/
Fixes: 9d66e233c704 ("Btrfs: load free space cache if it exists")
Tested-by: Andrew Nelson <andrew.s.nelson@gmail.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>

1 files changed, 21 insertions, 1 deletions

diff --git a/fs/btrfs/free-space-cache.c b/fs/btrfs/free-space-cache.c
index 4ba0aedc878b..74aa552f4793 100644
--- a/fs/btrfs/free-space-cache.c
+++ b/fs/btrfs/free-space-cache.c
@@ -75,7 +75,8 @@ static struct inode *__lookup_free_space_inode(struct btrfs_root *root,
 	 * sure NOFS is set to keep us from deadlocking.
 	 */
 	nofs_flag = memalloc_nofs_save();
-	inode = btrfs_iget(fs_info->sb, &location, root, NULL);
+	inode = btrfs_iget_path(fs_info->sb, &location, root, NULL, path);
+	btrfs_release_path(path);
 	memalloc_nofs_restore(nofs_flag);
 	if (IS_ERR(inode))
 		return inode;
@@ -838,6 +839,25 @@ int load_free_space_cache(struct btrfs_fs_info *fs_info,
 	path->search_commit_root = 1;
 	path->skip_locking = 1;
 
+	/*
+	 * We must pass a path with search_commit_root set to btrfs_iget in
+	 * order to avoid a deadlock when allocating extents for the tree root.
+	 *
+	 * When we are COWing an extent buffer from the tree root, when looking
+	 * for a free extent, at extent-tree.c:find_free_extent(), we can find
+	 * block group without its free space cache loaded. When we find one
+	 * we must load its space cache which requires reading its free space
+	 * cache's inode item from the root tree. If this inode item is located
+	 * in the same leaf that we started COWing before, then we end up in
+	 * deadlock on the extent buffer (trying to read lock it when we
+	 * previously write locked it).
+	 *
+	 * It's safe to read the inode item using the commit root because
+	 * block groups, once loaded, stay in memory forever (until they are
+	 * removed) as well as their space caches once loaded. New block groups
+	 * once created get their ->cached field set to BTRFS_CACHE_FINISHED so
+	 * we will never try to read their inode item while the fs is mounted.
+	 */
 	inode = lookup_free_space_inode(fs_info, block_group, path);
 	if (IS_ERR(inode)) {
 		btrfs_free_path(path);