net: new checksum functions
[dpdk.git] / app / test-pmd / csumonly.c
index abc7a9a..2fe3727 100644 (file)
@@ -1,13 +1,14 @@
 /*-
  *   BSD LICENSE
- * 
- *   Copyright(c) 2010-2013 Intel Corporation. All rights reserved.
+ *
+ *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
+ *   Copyright 2014 6WIND S.A.
  *   All rights reserved.
- * 
+ *
  *   Redistribution and use in source and binary forms, with or without
  *   modification, are permitted provided that the following conditions
  *   are met:
- * 
+ *
  *     * Redistributions of source code must retain the above copyright
  *       notice, this list of conditions and the following disclaimer.
  *     * Redistributions in binary form must reproduce the above copyright
@@ -17,7 +18,7 @@
  *     * Neither the name of Intel Corporation nor the names of its
  *       contributors may be used to endorse or promote products derived
  *       from this software without specific prior written permission.
- * 
+ *
  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
 #include <rte_string_fns.h>
 #include "testpmd.h"
 
-
-
 #define IP_DEFTTL  64   /* from RFC 1340. */
 #define IP_VERSION 0x40
 #define IP_HDRLEN  0x05 /* default IP header length == five 32-bits words. */
 #define IP_VHL_DEF (IP_VERSION | IP_HDRLEN)
 
-static inline uint16_t
-get_16b_sum(uint16_t *ptr16, uint32_t nr)
-{
-       uint32_t sum = 0;
-       while (nr > 1)
-       {
-               sum +=*ptr16;
-               nr -= sizeof(uint16_t);
-               ptr16++;
-               if (sum > UINT16_MAX)
-                       sum -= UINT16_MAX;
-       }
-
-       /* If length is in odd bytes */
-       if (nr)
-               sum += *((uint8_t*)ptr16);
+/* we cannot use htons() from arpa/inet.h due to name conflicts, and we
+ * cannot use rte_cpu_to_be_16() on a constant in a switch/case */
+#if __BYTE_ORDER == __LITTLE_ENDIAN
+#define _htons(x) ((uint16_t)((((x) & 0x00ffU) << 8) | (((x) & 0xff00U) >> 8)))
+#else
+#define _htons(x) (x)
+#endif
 
-       sum = ((sum & 0xffff0000) >> 16) + (sum & 0xffff);
-       sum &= 0x0ffff;
-       return (uint16_t)sum;
+static uint16_t
+get_psd_sum(void *l3_hdr, uint16_t ethertype)
+{
+       if (ethertype == _htons(ETHER_TYPE_IPv4))
+               return rte_ipv4_phdr_cksum(l3_hdr);
+       else /* assume ethertype == ETHER_TYPE_IPv6 */
+               return rte_ipv6_phdr_cksum(l3_hdr);
 }
 
-static inline uint16_t
-get_ipv4_cksum(struct ipv4_hdr *ipv4_hdr)
+static uint16_t
+get_udptcp_checksum(void *l3_hdr, void *l4_hdr, uint16_t ethertype)
 {
-       uint16_t cksum;
-       cksum = get_16b_sum((uint16_t*)ipv4_hdr, sizeof(struct ipv4_hdr));
-       return (uint16_t)((cksum == 0xffff)?cksum:~cksum);
+       if (ethertype == _htons(ETHER_TYPE_IPv4))
+               return rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
+       else /* assume ethertype == ETHER_TYPE_IPv6 */
+               return rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
 }
 
-
-static inline uint16_t
-get_ipv4_psd_sum (struct ipv4_hdr * ip_hdr)
+/*
+ * Parse an ethernet header to fill the ethertype, l2_len, l3_len and
+ * ipproto. This function is able to recognize IPv4/IPv6 with one optional vlan
+ * header.
+ */
+static void
+parse_ethernet(struct ether_hdr *eth_hdr, uint16_t *ethertype, uint16_t *l2_len,
+       uint16_t *l3_len, uint8_t *l4_proto)
 {
-       /* Pseudo Header for IPv4/UDP/TCP checksum */
-       union ipv4_psd_header {
-               struct {
-                       uint32_t src_addr; /* IP address of source host. */
-                       uint32_t dst_addr; /* IP address of destination host(s). */
-                       uint8_t  zero;     /* zero. */
-                       uint8_t  proto;    /* L4 protocol type. */
-                       uint16_t len;      /* L4 length. */
-               } __attribute__((__packed__));
-               uint16_t u16_arr[0];
-       } psd_hdr;
-
-       psd_hdr.src_addr = ip_hdr->src_addr;
-       psd_hdr.dst_addr = ip_hdr->dst_addr;
-       psd_hdr.zero     = 0;
-       psd_hdr.proto    = ip_hdr->next_proto_id;
-       psd_hdr.len      = rte_cpu_to_be_16((uint16_t)(rte_be_to_cpu_16(ip_hdr->total_length)
-                               - sizeof(struct ipv4_hdr)));
-       return get_16b_sum(psd_hdr.u16_arr, sizeof(psd_hdr));
+       struct ipv4_hdr *ipv4_hdr;
+       struct ipv6_hdr *ipv6_hdr;
+
+       *l2_len = sizeof(struct ether_hdr);
+       *ethertype = eth_hdr->ether_type;
+
+       if (*ethertype == _htons(ETHER_TYPE_VLAN)) {
+               struct vlan_hdr *vlan_hdr = (struct vlan_hdr *)(eth_hdr + 1);
+
+               *l2_len  += sizeof(struct vlan_hdr);
+               *ethertype = vlan_hdr->eth_proto;
+       }
+
+       switch (*ethertype) {
+       case _htons(ETHER_TYPE_IPv4):
+               ipv4_hdr = (struct ipv4_hdr *) ((char *)eth_hdr + *l2_len);
+               *l3_len = (ipv4_hdr->version_ihl & 0x0f) * 4;
+               *l4_proto = ipv4_hdr->next_proto_id;
+               break;
+       case _htons(ETHER_TYPE_IPv6):
+               ipv6_hdr = (struct ipv6_hdr *) ((char *)eth_hdr + *l2_len);
+               *l3_len = sizeof(struct ipv6_hdr);
+               *l4_proto = ipv6_hdr->proto;
+               break;
+       default:
+               *l3_len = 0;
+               *l4_proto = 0;
+               break;
+       }
 }
 
-static inline uint16_t
-get_ipv6_psd_sum (struct ipv6_hdr * ip_hdr)
+/* modify the IPv4 or IPv4 source address of a packet */
+static void
+change_ip_addresses(void *l3_hdr, uint16_t ethertype)
 {
-       /* Pseudo Header for IPv6/UDP/TCP checksum */
-       union ipv6_psd_header {
-               struct {
-                       uint8_t src_addr[16]; /* IP address of source host. */
-                       uint8_t dst_addr[16]; /* IP address of destination host(s). */
-                       uint32_t len;         /* L4 length. */
-                       uint32_t proto;       /* L4 protocol - top 3 bytes must be zero */
-               } __attribute__((__packed__));
-
-               uint16_t u16_arr[0]; /* allow use as 16-bit values with safe aliasing */
-       } psd_hdr;
-
-       rte_memcpy(&psd_hdr.src_addr, ip_hdr->src_addr,
-                       sizeof(ip_hdr->src_addr) + sizeof(ip_hdr->dst_addr));
-       psd_hdr.len       = ip_hdr->payload_len;
-       psd_hdr.proto     = (ip_hdr->proto << 24);
-
-       return get_16b_sum(psd_hdr.u16_arr, sizeof(psd_hdr));
+       struct ipv4_hdr *ipv4_hdr = l3_hdr;
+       struct ipv6_hdr *ipv6_hdr = l3_hdr;
+
+       if (ethertype == _htons(ETHER_TYPE_IPv4)) {
+               ipv4_hdr->src_addr =
+                       rte_cpu_to_be_32(rte_be_to_cpu_32(ipv4_hdr->src_addr) + 1);
+       } else if (ethertype == _htons(ETHER_TYPE_IPv6)) {
+               ipv6_hdr->src_addr[15] = ipv6_hdr->src_addr[15] + 1;
+       }
 }
 
-static inline uint16_t
-get_ipv4_udptcp_checksum(struct ipv4_hdr *ipv4_hdr, uint16_t *l4_hdr)
+/* if possible, calculate the checksum of a packet in hw or sw,
+ * depending on the testpmd command line configuration */
+static uint64_t
+process_inner_cksums(void *l3_hdr, uint16_t ethertype, uint16_t l3_len,
+       uint8_t l4_proto, uint16_t testpmd_ol_flags)
 {
-       uint32_t cksum;
-       uint32_t l4_len;
-
-       l4_len = rte_be_to_cpu_16(ipv4_hdr->total_length) - sizeof(struct ipv4_hdr);
-
-       cksum = get_16b_sum(l4_hdr, l4_len);
-       cksum += get_ipv4_psd_sum(ipv4_hdr);
-
-       cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
-       cksum = (~cksum) & 0xffff;
-       if (cksum == 0)
-               cksum = 0xffff;
-       return (uint16_t)cksum;
+       struct ipv4_hdr *ipv4_hdr = l3_hdr;
+       struct udp_hdr *udp_hdr;
+       struct tcp_hdr *tcp_hdr;
+       struct sctp_hdr *sctp_hdr;
+       uint64_t ol_flags = 0;
+
+       if (ethertype == _htons(ETHER_TYPE_IPv4)) {
+               ipv4_hdr = l3_hdr;
+               ipv4_hdr->hdr_checksum = 0;
+
+               if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_IP_CKSUM)
+                       ol_flags |= PKT_TX_IP_CKSUM;
+               else
+                       ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
+
+               ol_flags |= PKT_TX_IPV4;
+       } else if (ethertype == _htons(ETHER_TYPE_IPv6))
+               ol_flags |= PKT_TX_IPV6;
+       else
+               return 0; /* packet type not supported, nothing to do */
+
+       if (l4_proto == IPPROTO_UDP) {
+               udp_hdr = (struct udp_hdr *)((char *)l3_hdr + l3_len);
+               /* do not recalculate udp cksum if it was 0 */
+               if (udp_hdr->dgram_cksum != 0) {
+                       udp_hdr->dgram_cksum = 0;
+                       if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_UDP_CKSUM) {
+                               ol_flags |= PKT_TX_UDP_CKSUM;
+                               udp_hdr->dgram_cksum = get_psd_sum(l3_hdr,
+                                       ethertype);
+                       } else {
+                               udp_hdr->dgram_cksum =
+                                       get_udptcp_checksum(l3_hdr, udp_hdr,
+                                               ethertype);
+                       }
+               }
+       } else if (l4_proto == IPPROTO_TCP) {
+               tcp_hdr = (struct tcp_hdr *)((char *)l3_hdr + l3_len);
+               tcp_hdr->cksum = 0;
+               if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_TCP_CKSUM) {
+                       ol_flags |= PKT_TX_TCP_CKSUM;
+                       tcp_hdr->cksum = get_psd_sum(l3_hdr, ethertype);
+               } else {
+                       tcp_hdr->cksum =
+                               get_udptcp_checksum(l3_hdr, tcp_hdr, ethertype);
+               }
+       } else if (l4_proto == IPPROTO_SCTP) {
+               sctp_hdr = (struct sctp_hdr *)((char *)l3_hdr + l3_len);
+               sctp_hdr->cksum = 0;
+               /* sctp payload must be a multiple of 4 to be
+                * offloaded */
+               if ((testpmd_ol_flags & TESTPMD_TX_OFFLOAD_SCTP_CKSUM) &&
+                       ((ipv4_hdr->total_length & 0x3) == 0)) {
+                       ol_flags |= PKT_TX_SCTP_CKSUM;
+               } else {
+                       /* XXX implement CRC32c, example available in
+                        * RFC3309 */
+               }
+       }
 
+       return ol_flags;
 }
 
-static inline uint16_t
-get_ipv6_udptcp_checksum(struct ipv6_hdr *ipv6_hdr, uint16_t *l4_hdr)
+/* Calculate the checksum of outer header (only vxlan is supported,
+ * meaning IP + UDP). The caller already checked that it's a vxlan
+ * packet */
+static uint64_t
+process_outer_cksums(void *outer_l3_hdr, uint16_t outer_ethertype,
+       uint16_t outer_l3_len, uint16_t testpmd_ol_flags)
 {
-       uint32_t cksum;
-       uint32_t l4_len;
+       struct ipv4_hdr *ipv4_hdr = outer_l3_hdr;
+       struct ipv6_hdr *ipv6_hdr = outer_l3_hdr;
+       struct udp_hdr *udp_hdr;
+       uint64_t ol_flags = 0;
 
-       l4_len = rte_be_to_cpu_16(ipv6_hdr->payload_len);
+       if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_VXLAN_CKSUM)
+               ol_flags |= PKT_TX_VXLAN_CKSUM;
 
-       cksum = get_16b_sum(l4_hdr, l4_len);
-       cksum += get_ipv6_psd_sum(ipv6_hdr);
+       if (outer_ethertype == _htons(ETHER_TYPE_IPv4)) {
+               ipv4_hdr->hdr_checksum = 0;
 
-       cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
-       cksum = (~cksum) & 0xffff;
-       if (cksum == 0)
-               cksum = 0xffff;
+               if ((testpmd_ol_flags & TESTPMD_TX_OFFLOAD_VXLAN_CKSUM) == 0)
+                       ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
+       }
 
-       return (uint16_t)cksum;
-}
+       udp_hdr = (struct udp_hdr *)((char *)outer_l3_hdr + outer_l3_len);
+       /* do not recalculate udp cksum if it was 0 */
+       if (udp_hdr->dgram_cksum != 0) {
+               udp_hdr->dgram_cksum = 0;
+               if ((testpmd_ol_flags & TESTPMD_TX_OFFLOAD_VXLAN_CKSUM) == 0) {
+                       if (outer_ethertype == _htons(ETHER_TYPE_IPv4))
+                               udp_hdr->dgram_cksum =
+                                       rte_ipv4_udptcp_cksum(ipv4_hdr, udp_hdr);
+                       else
+                               udp_hdr->dgram_cksum =
+                                       rte_ipv6_udptcp_cksum(ipv6_hdr, udp_hdr);
+               }
+       }
 
+       return ol_flags;
+}
 
 /*
- * Forwarding of packets. Change the checksum field with HW or SW methods
- * The HW/SW method selection depends on the ol_flags on every packet
+ * Receive a burst of packets, and for each packet:
+ *  - parse packet, and try to recognize a supported packet type (1)
+ *  - if it's not a supported packet type, don't touch the packet, else:
+ *  - modify the IPs in inner headers and in outer headers if any
+ *  - reprocess the checksum of all supported layers. This is done in SW
+ *    or HW, depending on testpmd command line configuration
+ * Then transmit packets on the output port.
+ *
+ * (1) Supported packets are:
+ *   Ether / (vlan) / IP|IP6 / UDP|TCP|SCTP .
+ *   Ether / (vlan) / outer IP|IP6 / outer UDP / VxLAN / Ether / IP|IP6 /
+ *           UDP|TCP|SCTP
+ *
+ * The testpmd command line for this forward engine sets the flags
+ * TESTPMD_TX_OFFLOAD_* in ports[tx_port].tx_ol_flags. They control
+ * wether a checksum must be calculated in software or in hardware. The
+ * IP, UDP, TCP and SCTP flags always concern the inner layer.  The
+ * VxLAN flag concerns the outer IP and UDP layer (if packet is
+ * recognized as a vxlan packet).
  */
 static void
 pkt_burst_checksum_forward(struct fwd_stream *fs)
 {
-       struct rte_mbuf  *pkts_burst[MAX_PKT_BURST];
-       struct rte_port  *txp;
-       struct rte_mbuf  *mb;
+       struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
+       struct rte_port *txp;
+       struct rte_mbuf *m;
        struct ether_hdr *eth_hdr;
-       struct ipv4_hdr  *ipv4_hdr;
-       struct ipv6_hdr  *ipv6_hdr;
-       struct udp_hdr   *udp_hdr;
-       struct tcp_hdr   *tcp_hdr;
-       struct sctp_hdr  *sctp_hdr;
-
+       void *l3_hdr = NULL, *outer_l3_hdr = NULL; /* can be IPv4 or IPv6 */
+       struct udp_hdr *udp_hdr;
        uint16_t nb_rx;
        uint16_t nb_tx;
        uint16_t i;
-       uint16_t ol_flags;
-       uint16_t pkt_ol_flags;
-       uint16_t tx_ol_flags;
-       uint16_t l4_proto;
-       uint16_t eth_type;
-       uint8_t  l2_len;
-       uint8_t  l3_len;
-
+       uint64_t ol_flags;
+       uint16_t testpmd_ol_flags;
+       uint8_t l4_proto;
+       uint16_t ethertype = 0, outer_ethertype = 0;
+       uint16_t  l2_len = 0, l3_len = 0, outer_l2_len = 0, outer_l3_len = 0;
+       int tunnel = 0;
        uint32_t rx_bad_ip_csum;
        uint32_t rx_bad_l4_csum;
 
@@ -238,9 +320,7 @@ pkt_burst_checksum_forward(struct fwd_stream *fs)
        start_tsc = rte_rdtsc();
 #endif
 
-       /*
-        * Receive a burst of packets and forward them.
-        */
+       /* receive a burst of packet */
        nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
                                 nb_pkt_per_burst);
        if (unlikely(nb_rx == 0))
@@ -254,185 +334,107 @@ pkt_burst_checksum_forward(struct fwd_stream *fs)
        rx_bad_l4_csum = 0;
 
        txp = &ports[fs->tx_port];
-       tx_ol_flags = txp->tx_ol_flags;
+       testpmd_ol_flags = txp->tx_ol_flags;
 
        for (i = 0; i < nb_rx; i++) {
 
-               mb = pkts_burst[i];
-               l2_len  = sizeof(struct ether_hdr);
-               pkt_ol_flags = mb->ol_flags;
-               ol_flags = (uint16_t) (pkt_ol_flags & (~PKT_TX_L4_MASK));
-
-               eth_hdr = (struct ether_hdr *) mb->pkt.data;
-               eth_type = rte_be_to_cpu_16(eth_hdr->ether_type);
-               if (eth_type == ETHER_TYPE_VLAN) {
-                       /* Only allow single VLAN label here */
-                       l2_len  += sizeof(struct vlan_hdr);
-                        eth_type = rte_be_to_cpu_16(*(uint16_t *)
-                               ((uintptr_t)&eth_hdr->ether_type +
-                               sizeof(struct vlan_hdr)));
+               ol_flags = 0;
+               tunnel = 0;
+               m = pkts_burst[i];
+
+               /* Update the L3/L4 checksum error packet statistics */
+               rx_bad_ip_csum += ((m->ol_flags & PKT_RX_IP_CKSUM_BAD) != 0);
+               rx_bad_l4_csum += ((m->ol_flags & PKT_RX_L4_CKSUM_BAD) != 0);
+
+               /* step 1: dissect packet, parsing optional vlan, ip4/ip6, vxlan
+                * and inner headers */
+
+               eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
+               parse_ethernet(eth_hdr, &ethertype, &l2_len, &l3_len, &l4_proto);
+               l3_hdr = (char *)eth_hdr + l2_len;
+
+               /* check if it's a supported tunnel (only vxlan for now) */
+               if (l4_proto == IPPROTO_UDP) {
+                       udp_hdr = (struct udp_hdr *)((char *)l3_hdr + l3_len);
+
+                       /* currently, this flag is set by i40e only if the
+                        * packet is vxlan */
+                       if (((m->ol_flags & PKT_RX_TUNNEL_IPV4_HDR) ||
+                                       (m->ol_flags & PKT_RX_TUNNEL_IPV6_HDR)))
+                               tunnel = 1;
+                       /* else check udp destination port, 4789 is the default
+                        * vxlan port (rfc7348) */
+                       else if (udp_hdr->dst_port == _htons(4789))
+                               tunnel = 1;
+
+                       if (tunnel == 1) {
+                               outer_ethertype = ethertype;
+                               outer_l2_len = l2_len;
+                               outer_l3_len = l3_len;
+                               outer_l3_hdr = l3_hdr;
+
+                               eth_hdr = (struct ether_hdr *)((char *)udp_hdr +
+                                       sizeof(struct udp_hdr) +
+                                       sizeof(struct vxlan_hdr));
+
+                               parse_ethernet(eth_hdr, &ethertype, &l2_len,
+                                       &l3_len, &l4_proto);
+                               l3_hdr = (char *)eth_hdr + l2_len;
+                       }
                }
 
-               /* Update the L3/L4 checksum error packet count  */
-               rx_bad_ip_csum += (uint16_t) ((pkt_ol_flags & PKT_RX_IP_CKSUM_BAD) != 0);
-               rx_bad_l4_csum += (uint16_t) ((pkt_ol_flags & PKT_RX_L4_CKSUM_BAD) != 0);
-
-               /*
-                * Try to figure out L3 packet type by SW.
-                */
-               if ((pkt_ol_flags & (PKT_RX_IPV4_HDR | PKT_RX_IPV4_HDR_EXT |
-                               PKT_RX_IPV6_HDR | PKT_RX_IPV6_HDR_EXT)) == 0) {
-                       if (eth_type == ETHER_TYPE_IPv4)
-                               pkt_ol_flags |= PKT_RX_IPV4_HDR;
-                       else if (eth_type == ETHER_TYPE_IPv6)
-                               pkt_ol_flags |= PKT_RX_IPV6_HDR;
-               }
+               /* step 2: change all source IPs (v4 or v6) so we need
+                * to recompute the chksums even if they were correct */
 
-               /*
-                * Simplify the protocol parsing
-                * Assuming the incoming packets format as
-                *      Ethernet2 + optional single VLAN
-                *      + ipv4 or ipv6
-                *      + udp or tcp or sctp or others
-                */
-               if (pkt_ol_flags & PKT_RX_IPV4_HDR) {
+               change_ip_addresses(l3_hdr, ethertype);
+               if (tunnel == 1)
+                       change_ip_addresses(outer_l3_hdr, outer_ethertype);
 
-                       /* Do not support ipv4 option field */
-                       l3_len = sizeof(struct ipv4_hdr) ;
+               /* step 3: depending on user command line configuration,
+                * recompute checksum either in software or flag the
+                * mbuf to offload the calculation to the NIC */
 
-                       ipv4_hdr = (struct ipv4_hdr *) (rte_pktmbuf_mtod(mb,
-                                       unsigned char *) + l2_len);
+               /* process checksums of inner headers first */
+               ol_flags |= process_inner_cksums(l3_hdr, ethertype,
+                       l3_len, l4_proto, testpmd_ol_flags);
 
-                       l4_proto = ipv4_hdr->next_proto_id;
+               /* Then process outer headers if any. Note that the software
+                * checksum will be wrong if one of the inner checksums is
+                * processed in hardware. */
+               if (tunnel == 1) {
+                       ol_flags |= process_outer_cksums(outer_l3_hdr,
+                               outer_ethertype, outer_l3_len, testpmd_ol_flags);
+               }
 
-                       /* Do not delete, this is required by HW*/
-                       ipv4_hdr->hdr_checksum = 0;
+               /* step 4: fill the mbuf meta data (flags and header lengths) */
 
-                       if (tx_ol_flags & 0x1) {
-                               /* HW checksum */
-                               ol_flags |= PKT_TX_IP_CKSUM;
+               if (tunnel == 1) {
+                       if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_VXLAN_CKSUM) {
+                               m->l2_len = outer_l2_len;
+                               m->l3_len = outer_l3_len;
+                               m->inner_l2_len = l2_len;
+                               m->inner_l3_len = l3_len;
                        }
                        else {
-                               /* SW checksum calculation */
-                               ipv4_hdr->src_addr++;
-                               ipv4_hdr->hdr_checksum = get_ipv4_cksum(ipv4_hdr);
-                       }
-
-                       if (l4_proto == IPPROTO_UDP) {
-                               udp_hdr = (struct udp_hdr*) (rte_pktmbuf_mtod(mb,
-                                               unsigned char *) + l2_len + l3_len);
-                               if (tx_ol_flags & 0x2) {
-                                       /* HW Offload */
-                                       ol_flags |= PKT_TX_UDP_CKSUM;
-                                       /* Pseudo header sum need be set properly */
-                                       udp_hdr->dgram_cksum = get_ipv4_psd_sum(ipv4_hdr);
-                               }
-                               else {
-                                       /* SW Implementation, clear checksum field first */
-                                       udp_hdr->dgram_cksum = 0;
-                                       udp_hdr->dgram_cksum = get_ipv4_udptcp_checksum(ipv4_hdr,
-                                                       (uint16_t*)udp_hdr);
-                               }
-                       }
-                       else if (l4_proto == IPPROTO_TCP){
-                               tcp_hdr = (struct tcp_hdr*) (rte_pktmbuf_mtod(mb,
-                                               unsigned char *) + l2_len + l3_len);
-                               if (tx_ol_flags & 0x4) {
-                                       ol_flags |= PKT_TX_TCP_CKSUM;
-                                       tcp_hdr->cksum = get_ipv4_psd_sum(ipv4_hdr);
-                               }
-                               else {
-                                       tcp_hdr->cksum = 0;
-                                       tcp_hdr->cksum = get_ipv4_udptcp_checksum(ipv4_hdr,
-                                                       (uint16_t*)tcp_hdr);
-                               }
-                       }
-                       else if (l4_proto == IPPROTO_SCTP) {
-                               sctp_hdr = (struct sctp_hdr*) (rte_pktmbuf_mtod(mb,
-                                               unsigned char *) + l2_len + l3_len);
-
-                               if (tx_ol_flags & 0x8) {
-                                       ol_flags |= PKT_TX_SCTP_CKSUM;
-                                       sctp_hdr->cksum = 0;
-
-                                       /* Sanity check, only number of 4 bytes supported */
-                                       if ((rte_be_to_cpu_16(ipv4_hdr->total_length) % 4) != 0)
-                                               printf("sctp payload must be a multiple "
-                                                       "of 4 bytes for checksum offload");
-                               }
-                               else {
-                                       sctp_hdr->cksum = 0;
-                                       /* CRC32c sample code available in RFC3309 */
-                               }
-                       }
-                       /* End of L4 Handling*/
-               }
-               else if (pkt_ol_flags & PKT_RX_IPV6_HDR) {
-
-                       ipv6_hdr = (struct ipv6_hdr *) (rte_pktmbuf_mtod(mb,
-                                       unsigned char *) + l2_len);
-                       l3_len = sizeof(struct ipv6_hdr) ;
-                       l4_proto = ipv6_hdr->proto;
-
-                       if (l4_proto == IPPROTO_UDP) {
-                               udp_hdr = (struct udp_hdr*) (rte_pktmbuf_mtod(mb,
-                                               unsigned char *) + l2_len + l3_len);
-                               if (tx_ol_flags & 0x2) {
-                                       /* HW Offload */
-                                       ol_flags |= PKT_TX_UDP_CKSUM;
-                                       udp_hdr->dgram_cksum = get_ipv6_psd_sum(ipv6_hdr);
-                               }
-                               else {
-                                       /* SW Implementation */
-                                       /* checksum field need be clear first */
-                                       udp_hdr->dgram_cksum = 0;
-                                       udp_hdr->dgram_cksum = get_ipv6_udptcp_checksum(ipv6_hdr,
-                                                       (uint16_t*)udp_hdr);
-                               }
-                       }
-                       else if (l4_proto == IPPROTO_TCP) {
-                               tcp_hdr = (struct tcp_hdr*) (rte_pktmbuf_mtod(mb,
-                                               unsigned char *) + l2_len + l3_len);
-                               if (tx_ol_flags & 0x4) {
-                                       ol_flags |= PKT_TX_TCP_CKSUM;
-                                       tcp_hdr->cksum = get_ipv6_psd_sum(ipv6_hdr);
-                               }
-                               else {
-                                       tcp_hdr->cksum = 0;
-                                       tcp_hdr->cksum = get_ipv6_udptcp_checksum(ipv6_hdr,
-                                                       (uint16_t*)tcp_hdr);
-                               }
-                       }
-                       else if (l4_proto == IPPROTO_SCTP) {
-                               sctp_hdr = (struct sctp_hdr*) (rte_pktmbuf_mtod(mb,
-                                               unsigned char *) + l2_len + l3_len);
-
-                               if (tx_ol_flags & 0x8) {
-                                       ol_flags |= PKT_TX_SCTP_CKSUM;
-                                       sctp_hdr->cksum = 0;
-                                       /* Sanity check, only number of 4 bytes supported by HW */
-                                       if ((rte_be_to_cpu_16(ipv6_hdr->payload_len) % 4) != 0)
-                                               printf("sctp payload must be a multiple "
-                                                       "of 4 bytes for checksum offload");
-                               }
-                               else {
-                                       /* CRC32c sample code available in RFC3309 */
-                                       sctp_hdr->cksum = 0;
-                               }
-                       } else {
-                               printf("Test flow control for 1G PMD \n");
+                               /* if we don't do vxlan cksum in hw,
+                                  outer checksum will be wrong because
+                                  we changed the ip, but it shows that
+                                  we can process the inner header cksum
+                                  in the nic */
+                               m->l2_len = outer_l2_len + outer_l3_len +
+                                       sizeof(struct udp_hdr) +
+                                       sizeof(struct vxlan_hdr) + l2_len;
+                               m->l3_len = l3_len;
                        }
-                       /* End of L6 Handling*/
-               }
-               else {
-                       l3_len = 0;
-                       printf("Unhandled packet type: %#hx\n", eth_type);
+               } else {
+                       /* this is only useful if an offload flag is
+                        * set, but it does not hurt to fill it in any
+                        * case */
+                       m->l2_len = l2_len;
+                       m->l3_len = l3_len;
                }
+               m->ol_flags = ol_flags;
 
-               /* Combine the packet header write. VLAN is not consider here */
-               mb->pkt.vlan_macip.f.l2_len = l2_len;
-               mb->pkt.vlan_macip.f.l3_len = l3_len;
-               mb->ol_flags = ol_flags;
        }
        nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, pkts_burst, nb_rx);
        fs->tx_packets += nb_tx;
@@ -455,7 +457,6 @@ pkt_burst_checksum_forward(struct fwd_stream *fs)
 #endif
 }
 
-
 struct fwd_engine csum_fwd_engine = {
        .fwd_mode_name  = "csum",
        .port_fwd_begin = NULL,