Btrfs: fix racy system chunk allocation when setting block group ro

If while setting a block group read-only we end up allocating a system
chunk, through check_system_chunk(), we were not doing it while holding
the chunk mutex which is a problem if a concurrent chunk allocation is
happening, through do_chunk_alloc(), as it means both block groups can
end up using the same logical addresses and physical regions in the
device(s). So make sure we hold the chunk mutex.

Cc: stable@vger.kernel.org  # 4.0+
Fixes: 2f08108 ("btrfs: delete chunk allocation attemp when
                      setting block group ro")

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>

fs/btrfs/extent-tree.c

@@ -8842,7 +8842,9 @@ int btrfs_set_block_group_ro(struct btrfs_root *root,
 out:
 	if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
 		alloc_flags = update_block_group_flags(root, cache->flags);
+		lock_chunks(root->fs_info->chunk_root);
 		check_system_chunk(trans, root, alloc_flags);
+		unlock_chunks(root->fs_info->chunk_root);
 	}
 	mutex_unlock(&root->fs_info->ro_block_group_mutex);
 

fs/btrfs/volumes.c

@@ -4625,6 +4625,7 @@ int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
 {
 	u64 chunk_offset;
 
+	ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
 	chunk_offset = find_next_chunk(extent_root->fs_info);
 	return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
 }