Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Subject: Re: [syzbot] KASAN: use-after-free Read in disk_part_iter_next (2)
To:     Christoph Hellwig <hch@lst.de>,
        syzbot <syzbot+8fede7e30c7cee0de139@syzkaller.appspotmail.com>
Cc:     axboe@kernel.dk, linux-block@vger.kernel.org,
        linux-kernel@vger.kernel.org, syzkaller-bugs@googlegroups.com,
        tj@kernel.org
References: <00000000000053da9405bd7d2644@google.com>
 <000000000000a17fb305be0b3f38@google.com> <20210322071801.GE3440@lst.de>
From:   Bart Van Assche <bvanassche@acm.org>
Message-ID: <94abea02-0c6f-cd0b-5aa4-fef08f9b2ab8@acm.org>
Date:   Mon, 22 Mar 2021 15:21:19 -0700
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:78.0) Gecko/20100101
 Thunderbird/78.8.0
MIME-Version: 1.0
In-Reply-To: <20210322071801.GE3440@lst.de>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 3/22/21 12:18 AM, Christoph Hellwig wrote:
> I've been running the reproducer on a KASAN enable VM for about
> 15 minutes now, but haven't been able to reproduce it.
> 
> Is there a way to inject this proposed fix into the syzbot queue?
> 
> diff --git a/block/partitions/core.c b/block/partitions/core.c
> index 1a7558917c47d6..f5d5872b89d57e 100644
> --- a/block/partitions/core.c
> +++ b/block/partitions/core.c
> @@ -288,15 +288,12 @@ struct device_type part_type = {
>   void delete_partition(struct block_device *part)
>   {
>   	xa_erase(&part->bd_disk->part_tbl, part->bd_partno);
> -	kobject_put(part->bd_holder_dir);
> -	device_del(&part->bd_device);
> -
> -	/*
> -	 * Remove the block device from the inode hash, so that it cannot be
> -	 * looked up any more even when openers still hold references.
> -	 */
>   	remove_inode_hash(part->bd_inode);
>   
> +	synchronize_rcu();
> +
> +	kobject_put(part->bd_holder_dir);
> +	device_del(&part->bd_device);
>   	put_device(&part->bd_device);
>   }

Hi Christoph,

disk_part_iter_next() calls bdgrab() before returning a pointer to a 
certain partition. 'part' is only freed if its reference count drops to 
zero. The function that frees the partition information, 
bdev_free_inode(), is invoked via call_rcu(). bdgrab() fails if the 
refcount of a partition is zero. Does that mean that it is not necessary 
to call synchronize_rcu() between xa_erase() and put_device()?

Thanks,

Bart.