Btrfs: fix unprotected list operations at btrfs_write_dirty_block_groups

We call btrfs_write_dirty_block_groups() in the critical section of a
transaction's commit, when no other tasks can join the transaction and
add more block groups to the transaction's list of dirty block groups,
so we not taking the dirty block groups spinlock when checking for the
list's emptyness, grabbing its first element or deleting elements from
it.

However there's a special and rare case where we can have a concurrent
task adding elements to this list. We trigger writeback for space
caches before at btrfs_start_dirty_block_groups() and in past iterations
of the loop at btrfs_write_dirty_block_groups(), this means that when
the writeback finishes (which happens asynchronously) it creates a
task for the endio free space work queue that executes
btrfs_finish_ordered_io() - this function is able to join the transaction,
through btrfs_join_transaction_nolock(), and update the free space cache's
inode item in the root tree, which can result in COWing nodes of this tree
and therefore allocation of a new block group can happen, which gets added
to the transaction's list of dirty block groups while the transaction
commit task is operating on it concurrently.

So fix this by taking the dirty block groups spinlock before doing
operations on the dirty block groups list at
btrfs_write_dirty_block_groups().
Signed-off-by: NFilipe Manana <fdmanana@suse.com>

fs/btrfs/extent-tree.c

@@ -3684,11 +3684,21 @@ int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
 		return -ENOMEM;
 
 	/*
-	 * We don't need the lock here since we are protected by the transaction
-	 * commit.  We want to do the cache_save_setup first and then run the
+	 * Even though we are in the critical section of the transaction commit,
+	 * we can still have concurrent tasks adding elements to this
+	 * transaction's list of dirty block groups. These tasks correspond to
+	 * endio free space workers started when writeback finishes for a
+	 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
+	 * allocate new block groups as a result of COWing nodes of the root
+	 * tree when updating the free space inode. The writeback for the space
+	 * caches is triggered by an earlier call to
+	 * btrfs_start_dirty_block_groups() and iterations of the following
+	 * loop.
+	 * Also we want to do the cache_save_setup first and then run the
 	 * delayed refs to make sure we have the best chance at doing this all
 	 * in one shot.
 	 */
+	spin_lock(&cur_trans->dirty_bgs_lock);
 	while (!list_empty(&cur_trans->dirty_bgs)) {
 		cache = list_first_entry(&cur_trans->dirty_bgs,
 					 struct btrfs_block_group_cache,
@@ -3700,11 +3710,13 @@ int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
 		 * finish and then do it all again
 		 */
 		if (!list_empty(&cache->io_list)) {
+			spin_unlock(&cur_trans->dirty_bgs_lock);
 			list_del_init(&cache->io_list);
 			btrfs_wait_cache_io(root, trans, cache,
 					    &cache->io_ctl, path,
 					    cache->key.objectid);
 			btrfs_put_block_group(cache);
+			spin_lock(&cur_trans->dirty_bgs_lock);
 		}
 
 		/*
@@ -3712,6 +3724,7 @@ int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
 		 * on any pending IO
 		 */
 		list_del_init(&cache->dirty_list);
+		spin_unlock(&cur_trans->dirty_bgs_lock);
 		should_put = 1;
 
 		cache_save_setup(cache, trans, path);
@@ -3743,7 +3756,9 @@ int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
 		/* if its not on the io list, we need to put the block group */
 		if (should_put)
 			btrfs_put_block_group(cache);
+		spin_lock(&cur_trans->dirty_bgs_lock);
 	}
+	spin_unlock(&cur_trans->dirty_bgs_lock);
 
 	while (!list_empty(io)) {
 		cache = list_first_entry(io, struct btrfs_block_group_cache,