ext4: Fix deadlock during directory rename

As lockdep properly warns, we should not be locking i_rwsem while having
transactions started as the proper lock ordering used by all directory
handling operations is i_rwsem -> transaction start. Fix the lock
ordering by moving the locking of the directory earlier in
ext4_rename().

Reported-by: syzbot+9d16c39efb5fade84574@syzkaller.appspotmail.com
Fixes: 0813299 ("ext4: Fix possible corruption when moving a directory")
Link: https://syzkaller.appspot.com/bug?extid=9d16c39efb5fade84574
Signed-off-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20230301141004.15087-1-jack@suse.cz
Signed-off-by: Theodore Ts'o <tytso@mit.edu>

fs/ext4/namei.c

@@ -3813,10 +3813,20 @@ static int ext4_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
 			return retval;
 	}
 
+	/*
+	 * We need to protect against old.inode directory getting converted
+	 * from inline directory format into a normal one.
+	 */
+	if (S_ISDIR(old.inode->i_mode))
+		inode_lock_nested(old.inode, I_MUTEX_NONDIR2);
+
 	old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de,
 				 &old.inlined);
-	if (IS_ERR(old.bh))
-		return PTR_ERR(old.bh);
+	if (IS_ERR(old.bh)) {
+		retval = PTR_ERR(old.bh);
+		goto unlock_moved_dir;
+	}
+
 	/*
 	 *  Check for inode number is _not_ due to possible IO errors.
 	 *  We might rmdir the source, keep it as pwd of some process
@@ -3873,11 +3883,6 @@ static int ext4_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
 			if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
 				goto end_rename;
 		}
-		/*
-		 * We need to protect against old.inode directory getting
-		 * converted from inline directory format into a normal one.
-		 */
-		inode_lock_nested(old.inode, I_MUTEX_NONDIR2);
 		retval = ext4_rename_dir_prepare(handle, &old);
 		if (retval) {
 			inode_unlock(old.inode);
@@ -4014,12 +4019,15 @@ static int ext4_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
 	} else {
 		ext4_journal_stop(handle);
 	}
-	if (old.dir_bh)
-		inode_unlock(old.inode);
 release_bh:
 	brelse(old.dir_bh);
 	brelse(old.bh);
 	brelse(new.bh);
+
+unlock_moved_dir:
+	if (S_ISDIR(old.inode->i_mode))
+		inode_unlock(old.inode);
+
 	return retval;
 }