Received-SPF: pass (google.com: domain of linux-ext4-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Thu, 29 Apr 2021 23:09:56 +0300
From:   Pavel Skripkin <paskripkin@gmail.com>
To:     "Theodore Ts'o" <tytso@mit.edu>
Cc:     Vegard Nossum <vegard.nossum@oracle.com>,
        akpm@linux-foundation.org, peterz@infradead.org, axboe@kernel.dk,
        pmladek@suse.com, adilger.kernel@dilger.ca,
        linux-ext4@vger.kernel.org, linux-kernel@vger.kernel.org,
        syzbot+d9e482e303930fa4f6ff@syzkaller.appspotmail.com
Subject: Re: [PATCH] ext4: fix memory leak in ext4_fill_super
Message-ID: <20210429230956.6ad23897@gmail.com>
In-Reply-To: <YIrnPXJo/n68NrQs@mit.edu>
References: <20210428172828.12589-1-paskripkin@gmail.com>
        <3c3877a4-fef2-9e24-f99f-2ecc46deb7e4@oracle.com>
        <20210429143354.418248a7@gmail.com>
        <YIrnPXJo/n68NrQs@mit.edu>
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Precedence: bulk

On Thu, 29 Apr 2021 13:05:01 -0400
"Theodore Ts'o" <tytso@mit.edu> wrote:

> On Thu, Apr 29, 2021 at 02:33:54PM +0300, Pavel Skripkin wrote:
> > 
> > There is a chance, that kthread_stop() call will happen before
> > threadfn call. It means, that kthread_stop() return value must be
> > checked everywhere, isn't it? Otherwise, there are a lot of
> > potential memory leaks, because some developers rely on the fact,
> > that data allocated for the thread will be freed _inside_ thread
> > function.
> 
> That's not the only potential way that we could leak memory.  Earlier
> in kthread(), if this memory allocation fails,
> 
> 	self = kzalloc(sizeof(*self), GFP_KERNEL);
> 
> we will exit with -ENOMEM.  So at the very least all callers of
> kthread_stop() also need to check for -ENOMEM as well as -EINTR ---
> or, be somehow sure that the thread function was successfully called
> and started.  In this particular case, the ext4 mount code had just
> started the kmmpd thread, and then detected that something else had
> gone wrong, and failed the mount before the kmmpd thread ever had a
> chance to run.

There is a small problem about -ENOMEM:

static int kmmpd(void *data)
{
...
			retval = read_mmp_block(sb, &bh_check, mmp_block);
			if (retval) {
				ext4_error_err(sb, -retval,
					       "error reading MMP data: %d",
					       retval);
				goto exit_thread;
			}
...

exit_thread:
	EXT4_SB(sb)->s_mmp_tsk = NULL;
	kfree(data);
	brelse(bh);
	return retval;
}

read_mmp_block can return -ENOMEM. In this case double free will happen.
I believe, we can change `return retval;` to `return 0;`, because
kthread return value isn't checked anywhere.

What do You think about it? 


With regards,
Pavel Skripkin