Re: [dpdk-dev] [PATCH] examples/ip_fragmentation: fix check of packet type

["Ananyev, Konstantin" <konstantin.ananyev@intel.com>]

Hi Wei,

> -----Original Message-----
> From: Dai, Wei
> Sent: Friday, March 10, 2017 3:28 AM
> To: dev@dpdk.org; Ananyev, Konstantin <konstantin.ananyev@intel.com>
> Cc: Dai, Wei <wei.dai@intel.com>; stable@dpdk.org
> Subject: [PATCH] examples/ip_fragmentation: fix check of packet type
> 
> The packet_type in mbuf is not correctly filled by ixgbe 82599 NIC.
> To use the ether_type in ethernet header to check packet type is
> more reliaber.
> 
> Fixes: 3c0184cc0c60 ("examples: replace some offload flags with packet type")
> Fixes: ab351fe1c95c ("mbuf: remove packet type from offload flags")
> 
> Cc: stable@dpdk.org
> 
> Reported-by: Fangfang Wei <fangfangx.wei@intel.com>
> Signed-off-by: Wei Dai <wei.dai@intel.com>
> Tested-by: Fagnfang Wei <fangfangx.wei@intel.com>
> ---
>  examples/ip_fragmentation/main.c | 10 ++++++----
>  1 file changed, 6 insertions(+), 4 deletions(-)
> 
> diff --git a/examples/ip_fragmentation/main.c b/examples/ip_fragmentation/main.c
> index e1e32c6..8612984 100644
> --- a/examples/ip_fragmentation/main.c
> +++ b/examples/ip_fragmentation/main.c
> @@ -268,6 +268,7 @@ l3fwd_simple_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf,
>  	uint32_t i, len, next_hop_ipv4;
>  	uint8_t next_hop_ipv6, port_out, ipv6;
>  	int32_t len2;
> +	struct ether_hdr *eth_hdr;
> 
>  	ipv6 = 0;
>  	rxq = &qconf->rx_queue_list[queueid];
> @@ -276,13 +277,14 @@ l3fwd_simple_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf,
>  	port_out = port_in;
> 
>  	/* Remove the Ethernet header and trailer from the input packet */
> +	eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
>  	rte_pktmbuf_adj(m, (uint16_t)sizeof(struct ether_hdr));
> 
>  	/* Build transmission burst */
>  	len = qconf->tx_mbufs[port_out].len;
> 
>  	/* if this is an IPv4 packet */
> -	if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
> +	if (eth_hdr->ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4)) {
>  		struct ipv4_hdr *ip_hdr;
>  		uint32_t ip_dst;
>  		/* Read the lookup key (i.e. ip_dst) from the input packet */
> @@ -316,7 +318,7 @@ l3fwd_simple_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf,
>  			if (unlikely (len2 < 0))
>  				return;
>  		}
> -	} else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
> +	} else if (eth_hdr->ether_type == rte_be_to_cpu_16(ETHER_TYPE_IPv6)) {
>  		/* if this is an IPv6 packet */
>  		struct ipv6_hdr *ip_hdr;
> 
> @@ -363,8 +365,8 @@ l3fwd_simple_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf,
>  		void *d_addr_bytes;
> 
>  		m = qconf->tx_mbufs[port_out].m_table[i];
> -		struct ether_hdr *eth_hdr = (struct ether_hdr *)
> -			rte_pktmbuf_prepend(m, (uint16_t)sizeof(struct ether_hdr));
> +		eth_hdr = (struct ether_hdr *)rte_pktmbuf_prepend(m,
> +			(uint16_t)sizeof(struct ether_hdr));
>  		if (eth_hdr == NULL) {
>  			rte_panic("No headroom in mbuf.\n");
>  		}

Thanks for the fix.
Would it be more convenient to do what l3fwd does:
Check what ptype capabilities are provided by HW, if no ptype support detected,
then install an RX callback?
Konstantin

> --
> 2.7.4