Date:   Tue, 7 Mar 2023 19:18:47 +0000
From:   Eric Biggers <ebiggers@kernel.org>
To:     Dan Carpenter <error27@gmail.com>
Cc:     jack@suse.cz, linux-ext4@vger.kernel.org
Subject: Re: [bug report] ext4: Fix possible corruption when moving a
 directory
Message-ID: <ZAeOFzbhCNvskQ6b@gmail.com>
References: <5efbe1b9-ad8b-4a4f-b422-24824d2b775c@kili.mountain>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <5efbe1b9-ad8b-4a4f-b422-24824d2b775c@kili.mountain>
Precedence: bulk

On Tue, Mar 07, 2023 at 06:56:28PM +0300, Dan Carpenter wrote:
> Hello Jan Kara,
> 
> The patch 0813299c586b: "ext4: Fix possible corruption when moving a
> directory" from Jan 26, 2023, leads to the following Smatch static
> checker warning:
> 
> 	fs/ext4/namei.c:4017 ext4_rename()
> 	error: double unlocked '&old.inode->i_rwsem' (orig line 3882)
> 
[...]
>     3875                 /*
>     3876                  * We need to protect against old.inode directory getting
>     3877                  * converted from inline directory format into a normal one.
>     3878                  */
>     3879                 inode_lock_nested(old.inode, I_MUTEX_NONDIR2);
>     3880                 retval = ext4_rename_dir_prepare(handle, &old);
>     3881                 if (retval) {
>     3882                         inode_unlock(old.inode);
> 
> The issue here is that ext4_rename_dir_prepare() sets old.dir_bh and
> then returns -EFSCORRUPTED.  It results in an unlock here and then again
> after the goto.

That analysis looks correct.  FYI, I think this is the same as the syzbot report
"[ext4?] WARNING: bad unlock balance in ext4_rename2"
(https://lore.kernel.org/linux-ext4/000000000000435c6905f639ae8e@google.com).

- Eric