Re: [dpdk-dev] [dpdk-stable] [PATCH v3 1/5] net/tap: fix mbuf double free when writev fails

[Ferruh Yigit <ferruh.yigit@intel.com>]

On 4/9/2020 1:53 PM, wangyunjian wrote:
> 
> 
>> -----Original Message-----
>> From: Ferruh Yigit [mailto:ferruh.yigit@intel.com]
>> Sent: Thursday, April 9, 2020 5:52 PM
>> To: wangyunjian <wangyunjian@huawei.com>; dev@dpdk.org
>> Cc: keith.wiles@intel.com; Lilijun (Jerry) <jerry.lilijun@huawei.com>; xudingke
>> <xudingke@huawei.com>; stable@dpdk.org
>> Subject: Re: [dpdk-stable] [dpdk-dev] [PATCH v3 1/5] net/tap: fix mbuf double
>> free when writev fails
>>
>> On 4/9/2020 9:03 AM, wangyunjian wrote:
>>>
>>>
>>>> -----Original Message-----
>>>> From: Ferruh Yigit [mailto:ferruh.yigit@intel.com]
>>>> Sent: Tuesday, April 7, 2020 8:35 PM
>>>> To: wangyunjian <wangyunjian@huawei.com>; dev@dpdk.org
>>>> Cc: keith.wiles@intel.com; Lilijun (Jerry)
>>>> <jerry.lilijun@huawei.com>; xudingke <xudingke@huawei.com>;
>>>> stable@dpdk.org
>>>> Subject: Re: [dpdk-stable] [dpdk-dev] [PATCH v3 1/5] net/tap: fix
>>>> mbuf double free when writev fails
>>>>
>>>> On 4/7/2020 5:22 AM, wangyunjian wrote:
>>>>> From: Yunjian Wang <wangyunjian@huawei.com>
>>>>>
>>>>> When the tap_write_mbufs() function return with break, mbuf was
>>>>> freed without incrementing num_packets. This may lead applications
>>>>> also free the mbuf. And the pmd_tx_burst() function should returns
>>>>> the number of original packets it actually sent excluding tso mbufs.
>>>>>
>>>>> Fixes: 9396ad334672 ("net/tap: fix reported number of Tx packets")
>>>>> CC: stable@dpdk.org
>>>>>
>>>>> Signed-off-by: Yunjian Wang <wangyunjian@huawei.com>
>>>>> ---
>>>>>  drivers/net/tap/rte_eth_tap.c | 21 +++++++++++++++------
>>>>>  1 file changed, 15 insertions(+), 6 deletions(-)
>>>>>
>>>>> diff --git a/drivers/net/tap/rte_eth_tap.c
>>>>> b/drivers/net/tap/rte_eth_tap.c index 05470a211..4c4b6b0b2 100644
>>>>> --- a/drivers/net/tap/rte_eth_tap.c
>>>>> +++ b/drivers/net/tap/rte_eth_tap.c
>>>>> @@ -521,7 +521,7 @@ tap_tx_l3_cksum(char *packet, uint64_t ol_flags,
>>>> unsigned int l2_len,
>>>>>  	}
>>>>>  }
>>>>>
>>>>> -static inline void
>>>>> +static inline int
>>>>>  tap_write_mbufs(struct tx_queue *txq, uint16_t num_mbufs,
>>>>>  			struct rte_mbuf **pmbufs,
>>>>>  			uint16_t *num_packets, unsigned long *num_tx_bytes)
>> @@
>>>> -588,7
>>>>> +588,7 @@ tap_write_mbufs(struct tx_queue *txq, uint16_t num_mbufs,
>>>>>  			seg_len = rte_pktmbuf_data_len(mbuf);
>>>>>  			l234_hlen = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
>>>>>  			if (seg_len < l234_hlen)
>>>>> -				break;
>>>>> +				return -1;
>>>>>
>>>>>  			/* To change checksums, work on a * copy of l2, l3
>>>>>  			 * headers + l4 pseudo header
>>>>> @@ -634,10 +634,12 @@ tap_write_mbufs(struct tx_queue *txq,
>> uint16_t
>>>> num_mbufs,
>>>>>  		/* copy the tx frame data */
>>>>>  		n = writev(process_private->txq_fds[txq->queue_id], iovecs, j);
>>>>>  		if (n <= 0)
>>>>> -			break;
>>>>> +			return -1;
>>>>> +
>>>>>  		(*num_packets)++;
>>>>>  		(*num_tx_bytes) += rte_pktmbuf_pkt_len(mbuf);
>>>>>  	}
>>>>> +	return 0;
>>>>>  }
>>>>>
>>>>>  /* Callback to handle sending packets from the tap interface @@
>>>>> -708,8 +710,15 @@ pmd_tx_burst(void *queue, struct rte_mbuf **bufs,
>>>> uint16_t nb_pkts)
>>>>>  			num_mbufs = 1;
>>>>>  		}
>>>>>
>>>>> -		tap_write_mbufs(txq, num_mbufs, mbuf,
>>>>> -				&num_packets, &num_tx_bytes);
>>>>> +		ret = tap_write_mbufs(txq, num_mbufs, mbuf,
>>>>> +				      &num_packets, &num_tx_bytes);
>>>>
>>>> reusing 'ret' here breaks the logic at the end of the loop that free
>>>> tso mbufs, which expects 'ret' is number of mbufs in tso case.
>>>>
>>>>> +		if (ret != 0) {
>>>>> +			txq->stats.errs++;
>>>>> +			/* free tso mbufs */
>>>>> +			for (j = 0; j < ret; j++)
>>>>
>>>> 'ret' only can be '0' or '-1', and we take the branch only when it is
>>>> '-1', so this block is not used at all and it doesn't free any mbuf.
>>>
>>> I'm sorry for my mistakes. I will fix it in next version.
>>> what about following:
>>>
>>> error = tap_write_mbufs(txq, num_mbufs, mbuf,
>>>                      &num_packets, &num_tx_bytes); if (error == -1) {
>>>     txq->stats.errs++;
>>>     /* free tso mbufs */
>>>     for (j = 0; j < ret; j++)
>>>         rte_pktmbuf_free(mbuf[j]);
>>>     break;
>>> }
>>
>> +1, but still needs to free the 'mbuf_in' before break.
> 
> I don't think it needs to free the 'mbuf_in' before break.
> The 'num_tx' does not increase, the caller will free unsent packets.

Yep, you are right.

> 
>>
>> Or maybe it is better to create a new variable like 'num_tso_mbufs' and use it
>> instead of 'ret', which is more readable, and this enables to reuse the 'ret'.
> 
> Thanks for your suggestion, will include it in next version.
> 
> Yunjian
> 
>>
>>>
>>> Thanks
>>> Yunjian
>>>>> +				rte_pktmbuf_free(mbuf[j]);
>>>>
>>>>
>>>> In the no tso case, if the 'tap_write_mbufs()' fails, this doesn't
>>>> free the 'mbuf_in'.
>>>>
>>>>> +			break;
>>>>> +		}
>>>>>  		num_tx++;
>>>>>  		/* free original mbuf */
>>>>>  		rte_pktmbuf_free(mbuf_in);
>>>>> @@ -722,7 +731,7 @@ pmd_tx_burst(void *queue, struct rte_mbuf
>>>>> **bufs,
>>>> uint16_t nb_pkts)
>>>>>  	txq->stats.errs += nb_pkts - num_tx;
>>>>>  	txq->stats.obytes += num_tx_bytes;
>>>>>
>>>>> -	return num_packets;
>>>>> +	return num_tx;
>>>>
>>>> +1 to return number of original packets.
>>>>
>>>>>  }
>>>>>
>>>>>  static const char *
>>>>>
>>>
>