Re: [dpdk-dev] [PATCH] ether: check the first segment length on SW VLAN insertion

[Andrew Rybchenko <arybchenko@solarflare.com>]

On 5/29/20 8:43 AM, Stephen Hemminger wrote:
> On Wed, 27 May 2020 15:31:41 +0100
> Andrew Rybchenko <arybchenko@solarflare.com> wrote:
> 
>> SW VLAN insertion relies on Ethernet addresses location in contigous
>> memory (do not split across mbuf segments). There is no any formal
>> requirements on data location and mbuf structure which guarantee it.
>> So, check it explicitly to avoid corrupted packets if the condition
>> is violated. Typically software VLAN insertion is done on Tx prepare
>> stage and application will get indication that the packet is invalid
>> and cannot be transmitted.
>>
>> Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
>> ---
>>  lib/librte_net/rte_ether.h | 4 ++++
>>  1 file changed, 4 insertions(+)
>>
>> diff --git a/lib/librte_net/rte_ether.h b/lib/librte_net/rte_ether.h
>> index 0ae4e75b6c..d7c076bba8 100644
>> --- a/lib/librte_net/rte_ether.h
>> +++ b/lib/librte_net/rte_ether.h
>> @@ -357,6 +357,10 @@ static inline int rte_vlan_insert(struct rte_mbuf **m)
>>  	if (!RTE_MBUF_DIRECT(*m) || rte_mbuf_refcnt_read(*m) > 1)
>>  		return -EINVAL;
>>  
>> +	/* Can't insert header if the first segment is too short */
>> +	if (rte_pktmbuf_data_len(*m) < 2 * RTE_ETHER_ADDR_LEN)
>> +		return -EINVAL;
> 
> Looks good, but you could also make it handle the fragment case with:
> 
> diff --git a/lib/librte_net/rte_ether.h b/lib/librte_net/rte_ether.h
> index 0ae4e75b6c58..4d0e310a4fac 100644
> --- a/lib/librte_net/rte_ether.h
> +++ b/lib/librte_net/rte_ether.h
> @@ -350,14 +350,18 @@ static inline int rte_vlan_strip(struct rte_mbuf *m)
>   */
>  static inline int rte_vlan_insert(struct rte_mbuf **m)
>  {
> -	struct rte_ether_hdr *oh, *nh;
> +	struct rte_ether_hdr *nh, tmp;
> +	const struct rte_ether_hdr *oh;
>  	struct rte_vlan_hdr *vh;
>  
>  	/* Can't insert header if mbuf is shared */
>  	if (!RTE_MBUF_DIRECT(*m) || rte_mbuf_refcnt_read(*m) > 1)
>  		return -EINVAL;
>  
> -	oh = rte_pktmbuf_mtod(*m, struct rte_ether_hdr *);
> +	oh = rte_pktmbuf_read(*m, 0, sizeof(*oh), &tmp);
> +	if (unlikely(oh == NULL))
> +		return -EINVAL;
> +
>  	nh = (struct rte_ether_hdr *)
>  		rte_pktmbuf_prepend(*m, sizeof(struct rte_vlan_hdr));
>  	if (nh == NULL)
> 

It is more complicated since memmove() below should be
rewritten  in a similar way as rte_pktmbuf_read(), but
write. I'm not sure it worse the effort.