ceph/src/spdk/dpdk/examples/tep_termination/vxlan.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2015 Intel Corporation
   3  */
   4
   5 #include <stdint.h>
   6 #include <rte_mbuf.h>
   7 #include <rte_hash_crc.h>
   8 #include <rte_byteorder.h>
   9 #include <rte_udp.h>
  10 #include <rte_tcp.h>
  11 #include <rte_sctp.h>
  12
  13 #include "main.h"
  14 #include "vxlan.h"
  15
  16 static uint16_t
  17 get_psd_sum(void *l3_hdr, uint16_t ethertype, uint64_t ol_flags)
  18 {
  19         if (ethertype == ETHER_TYPE_IPv4)
  20                 return rte_ipv4_phdr_cksum(l3_hdr, ol_flags);
  21         else /* assume ethertype == ETHER_TYPE_IPv6 */
  22                 return rte_ipv6_phdr_cksum(l3_hdr, ol_flags);
  23 }
  24
  25 /**
  26  * Parse an ethernet header to fill the ethertype, outer_l2_len, outer_l3_len and
  27  * ipproto. This function is able to recognize IPv4/IPv6 with one optional vlan
  28  * header.
  29  */
  30 static void
  31 parse_ethernet(struct ether_hdr *eth_hdr, union tunnel_offload_info *info,
  32                 uint8_t *l4_proto)
  33 {
  34         struct ipv4_hdr *ipv4_hdr;
  35         struct ipv6_hdr *ipv6_hdr;
  36         uint16_t ethertype;
  37
  38         info->outer_l2_len = sizeof(struct ether_hdr);
  39         ethertype = rte_be_to_cpu_16(eth_hdr->ether_type);
  40
  41         if (ethertype == ETHER_TYPE_VLAN) {
  42                 struct vlan_hdr *vlan_hdr = (struct vlan_hdr *)(eth_hdr + 1);
  43                 info->outer_l2_len  += sizeof(struct vlan_hdr);
  44                 ethertype = rte_be_to_cpu_16(vlan_hdr->eth_proto);
  45         }
  46
  47         switch (ethertype) {
  48         case ETHER_TYPE_IPv4:
  49                 ipv4_hdr = (struct ipv4_hdr *)
  50                         ((char *)eth_hdr + info->outer_l2_len);
  51                 info->outer_l3_len = sizeof(struct ipv4_hdr);
  52                 *l4_proto = ipv4_hdr->next_proto_id;
  53                 break;
  54         case ETHER_TYPE_IPv6:
  55                 ipv6_hdr = (struct ipv6_hdr *)
  56                         ((char *)eth_hdr + info->outer_l2_len);
  57                 info->outer_l3_len = sizeof(struct ipv6_hdr);
  58                 *l4_proto = ipv6_hdr->proto;
  59                 break;
  60         default:
  61                 info->outer_l3_len = 0;
  62                 *l4_proto = 0;
  63                 break;
  64         }
  65 }
  66
  67 /**
  68  * Calculate the checksum of a packet in hardware
  69  */
  70 static uint64_t
  71 process_inner_cksums(struct ether_hdr *eth_hdr, union tunnel_offload_info *info)
  72 {
  73         void *l3_hdr = NULL;
  74         uint8_t l4_proto;
  75         uint16_t ethertype;
  76         struct ipv4_hdr *ipv4_hdr;
  77         struct ipv6_hdr *ipv6_hdr;
  78         struct udp_hdr *udp_hdr;
  79         struct tcp_hdr *tcp_hdr;
  80         struct sctp_hdr *sctp_hdr;
  81         uint64_t ol_flags = 0;
  82
  83         info->l2_len = sizeof(struct ether_hdr);
  84         ethertype = rte_be_to_cpu_16(eth_hdr->ether_type);
  85
  86         if (ethertype == ETHER_TYPE_VLAN) {
  87                 struct vlan_hdr *vlan_hdr = (struct vlan_hdr *)(eth_hdr + 1);
  88                 info->l2_len  += sizeof(struct vlan_hdr);
  89                 ethertype = rte_be_to_cpu_16(vlan_hdr->eth_proto);
  90         }
  91
  92         l3_hdr = (char *)eth_hdr + info->l2_len;
  93
  94         if (ethertype == ETHER_TYPE_IPv4) {
  95                 ipv4_hdr = (struct ipv4_hdr *)l3_hdr;
  96                 ipv4_hdr->hdr_checksum = 0;
  97                 ol_flags |= PKT_TX_IPV4;
  98                 ol_flags |= PKT_TX_IP_CKSUM;
  99                 info->l3_len = sizeof(struct ipv4_hdr);
 100                 l4_proto = ipv4_hdr->next_proto_id;
 101         } else if (ethertype == ETHER_TYPE_IPv6) {
 102                 ipv6_hdr = (struct ipv6_hdr *)l3_hdr;
 103                 info->l3_len = sizeof(struct ipv6_hdr);
 104                 l4_proto = ipv6_hdr->proto;
 105                 ol_flags |= PKT_TX_IPV6;
 106         } else
 107                 return 0; /* packet type not supported, nothing to do */
 108
 109         if (l4_proto == IPPROTO_UDP) {
 110                 udp_hdr = (struct udp_hdr *)((char *)l3_hdr + info->l3_len);
 111                 ol_flags |= PKT_TX_UDP_CKSUM;
 112                 udp_hdr->dgram_cksum = get_psd_sum(l3_hdr,
 113                                 ethertype, ol_flags);
 114         } else if (l4_proto == IPPROTO_TCP) {
 115                 tcp_hdr = (struct tcp_hdr *)((char *)l3_hdr + info->l3_len);
 116                 /* Put PKT_TX_TCP_SEG bit setting before get_psd_sum(), because
 117                  * it depends on PKT_TX_TCP_SEG to calculate pseudo-header
 118                  * checksum.
 119                  */
 120                 if (tso_segsz != 0) {
 121                         ol_flags |= PKT_TX_TCP_SEG;
 122                         info->tso_segsz = tso_segsz;
 123                         info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
 124                 }
 125                 ol_flags |= PKT_TX_TCP_CKSUM;
 126                 tcp_hdr->cksum = get_psd_sum(l3_hdr, ethertype, ol_flags);
 127
 128         } else if (l4_proto == IPPROTO_SCTP) {
 129                 sctp_hdr = (struct sctp_hdr *)((char *)l3_hdr + info->l3_len);
 130                 sctp_hdr->cksum = 0;
 131                 ol_flags |= PKT_TX_SCTP_CKSUM;
 132         }
 133
 134         return ol_flags;
 135 }
 136
 137 int
 138 decapsulation(struct rte_mbuf *pkt)
 139 {
 140         uint8_t l4_proto = 0;
 141         uint16_t outer_header_len;
 142         struct udp_hdr *udp_hdr;
 143         union tunnel_offload_info info = { .data = 0 };
 144         struct ether_hdr *phdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
 145
 146         parse_ethernet(phdr, &info, &l4_proto);
 147
 148         if (l4_proto != IPPROTO_UDP)
 149                 return -1;
 150
 151         udp_hdr = (struct udp_hdr *)((char *)phdr +
 152                 info.outer_l2_len + info.outer_l3_len);
 153
 154         /** check udp destination port, 4789 is the default vxlan port
 155          * (rfc7348) or that the rx offload flag is set (i40e only
 156          * currently)*/
 157         if (udp_hdr->dst_port != rte_cpu_to_be_16(DEFAULT_VXLAN_PORT) &&
 158                 (pkt->packet_type & RTE_PTYPE_TUNNEL_MASK) == 0)
 159                 return -1;
 160         outer_header_len = info.outer_l2_len + info.outer_l3_len
 161                 + sizeof(struct udp_hdr) + sizeof(struct vxlan_hdr);
 162
 163         rte_pktmbuf_adj(pkt, outer_header_len);
 164
 165         return 0;
 166 }
 167
 168 void
 169 encapsulation(struct rte_mbuf *m, uint8_t queue_id)
 170 {
 171         uint vport_id;
 172         uint64_t ol_flags = 0;
 173         uint32_t old_len = m->pkt_len, hash;
 174         union tunnel_offload_info tx_offload = { .data = 0 };
 175         struct ether_hdr *phdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 176
 177         /*Allocate space for new ethernet, IPv4, UDP and VXLAN headers*/
 178         struct ether_hdr *pneth = (struct ether_hdr *) rte_pktmbuf_prepend(m,
 179                 sizeof(struct ether_hdr) + sizeof(struct ipv4_hdr)
 180                 + sizeof(struct udp_hdr) + sizeof(struct vxlan_hdr));
 181
 182         struct ipv4_hdr *ip = (struct ipv4_hdr *) &pneth[1];
 183         struct udp_hdr *udp = (struct udp_hdr *) &ip[1];
 184         struct vxlan_hdr *vxlan = (struct vxlan_hdr *) &udp[1];
 185
 186         /* convert TX queue ID to vport ID */
 187         vport_id = queue_id - 1;
 188
 189         /* replace original Ethernet header with ours */
 190         pneth = rte_memcpy(pneth, &app_l2_hdr[vport_id],
 191                 sizeof(struct ether_hdr));
 192
 193         /* copy in IP header */
 194         ip = rte_memcpy(ip, &app_ip_hdr[vport_id],
 195                 sizeof(struct ipv4_hdr));
 196         ip->total_length = rte_cpu_to_be_16(m->pkt_len
 197                                 - sizeof(struct ether_hdr));
 198
 199         /* outer IP checksum */
 200         ol_flags |= PKT_TX_OUTER_IP_CKSUM;
 201         ip->hdr_checksum = 0;
 202
 203         /* inner IP checksum offload */
 204         if (tx_checksum) {
 205                 ol_flags |= process_inner_cksums(phdr, &tx_offload);
 206                 m->l2_len = tx_offload.l2_len;
 207                 m->l3_len = tx_offload.l3_len;
 208                 m->l4_len = tx_offload.l4_len;
 209                 m->l2_len += ETHER_VXLAN_HLEN;
 210         }
 211
 212         m->outer_l2_len = sizeof(struct ether_hdr);
 213         m->outer_l3_len = sizeof(struct ipv4_hdr);
 214
 215         ol_flags |= PKT_TX_TUNNEL_VXLAN;
 216
 217         m->ol_flags |= ol_flags;
 218         m->tso_segsz = tx_offload.tso_segsz;
 219
 220         /*VXLAN HEADER*/
 221         vxlan->vx_flags = rte_cpu_to_be_32(VXLAN_HF_VNI);
 222         vxlan->vx_vni = rte_cpu_to_be_32(vxdev.out_key << 8);
 223
 224         /*UDP HEADER*/
 225         udp->dgram_cksum = 0;
 226         udp->dgram_len = rte_cpu_to_be_16(old_len
 227                                 + sizeof(struct udp_hdr)
 228                                 + sizeof(struct vxlan_hdr));
 229
 230         udp->dst_port = rte_cpu_to_be_16(vxdev.dst_port);
 231         hash = rte_hash_crc(phdr, 2 * ETHER_ADDR_LEN, phdr->ether_type);
 232         udp->src_port = rte_cpu_to_be_16((((uint64_t) hash * PORT_RANGE) >> 32)
 233                                         + PORT_MIN);
 234
 235         return;
 236 }