Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:5 as permitted sender) client-ip=2620:137:e000::3:5;
Subject: Re: [PATCH -next] nbd: get config_lock before sock_shutdown
To:     Jens Axboe <axboe@kernel.dk>,
        Zhong Jinghua <zhongjinghua@huaweicloud.com>,
        josef@toxicpanda.com
Cc:     linux-block@vger.kernel.org, nbd@other.debian.org,
        linux-kernel@vger.kernel.org, yi.zhang@huawei.com,
        "yukuai (C)" <yukuai3@huawei.com>
References: <20230707062256.1271948-1-zhongjinghua@huaweicloud.com>
 <1b67a9dd-c28a-661a-3a46-dab509d4c34e@kernel.dk>
From:   Yu Kuai <yukuai1@huaweicloud.com>
Message-ID: <ae49487b-9aa8-fe37-792b-676f7e70b23b@huaweicloud.com>
Date:   Thu, 28 Sep 2023 14:04:36 +0800
User-Agent: Mozilla/5.0 (Windows NT 10.0; WOW64; rv:60.0) Gecko/20100101
 Thunderbird/60.8.0
MIME-Version: 1.0
In-Reply-To: <1b67a9dd-c28a-661a-3a46-dab509d4c34e@kernel.dk>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk

Hi,

在 2023/08/01 8:27, Jens Axboe 写道:
> On 7/7/23 12:22?AM, Zhong Jinghua wrote:
>> Config->socks in sock_shutdown may trigger a UAF problem.
>> The reason is that sock_shutdown does not hold the config_lock,
>> so that nbd_ioctl can release config->socks at this time.
>>
>> T0: NBD_SET_SOCK
>> T1: NBD_DO_IT
>>
>> T0						T1
>>
>> nbd_ioctl
>>    mutex_lock(&nbd->config_lock)
>>    // get lock
>>    __nbd_ioctl
>>      nbd_start_device_ioctl
>>        nbd_start_device
>>         mutex_unlock(&nbd->config_lock)
>>           // relase lock
>>           wait_event_interruptible
>>           (kill, enter sock_shutdown)
>>           sock_shutdown
>> 					nbd_ioctl
>> 					  mutex_lock(&nbd->config_lock)
>> 					  // get lock
>> 					  __nbd_ioctl
>> 					    nbd_add_socket
>> 					      krealloc
>> 						kfree(p)
>> 					        //config->socks is NULL
>>             nbd_sock *nsock = config->socks // error
>>
>> Fix it by moving config_lock up before sock_shutdown.
>>
>> Signed-off-by: Zhong Jinghua <zhongjinghua@huaweicloud.com>
>> ---
>>   drivers/block/nbd.c | 7 ++++++-
>>   1 file changed, 6 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/block/nbd.c b/drivers/block/nbd.c
>> index c410cf29fb0c..accbe99ebb7e 100644
>> --- a/drivers/block/nbd.c
>> +++ b/drivers/block/nbd.c
>> @@ -1428,13 +1428,18 @@ static int nbd_start_device_ioctl(struct nbd_device *nbd)
>>   	mutex_unlock(&nbd->config_lock);
>>   	ret = wait_event_interruptible(config->recv_wq,
>>   					 atomic_read(&config->recv_threads) == 0);
>> +
>> +	/*
>> +	 * recv_work in flush_workqueue will not get this lock, because nbd_open
>> +	 * will hold nbd->config_refs
>> +	 */
>> +	mutex_lock(&nbd->config_lock);
>>   	if (ret) {
>>   		sock_shutdown(nbd);
>>   		nbd_clear_que(nbd);
>>   	}
>>   
>>   	flush_workqueue(nbd->recv_workq);
>> -	mutex_lock(&nbd->config_lock);
> 
> Feels pretty iffy to hold config_lock over the flush. If anything off
> recv_work() ever grabs it, we'd be stuck. Your comment assumes that the
> only case this will currently happen is if we drop the last ref, or at
> least that's the case that'd do it even if you don't mention it
> explicitly.
> 
> Maybe this is all fine, but recv_work() should have a comment matching
> this one, and this comment should be more descriptive as well.

Jinghua,

Please add comment as Jens suggested, and resend this patch.

Thanks,
Kuai

>