examples/tep_termination/vxlan_setup.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2015 Intel Corporation
   3  */
   4
   5 #include <getopt.h>
   6 #include <linux/if_ether.h>
   7 #include <linux/if_vlan.h>
   8 #include <linux/virtio_net.h>
   9 #include <linux/virtio_ring.h>
  10 #include <sys/param.h>
  11 #include <unistd.h>
  12
  13 #include <rte_ethdev.h>
  14 #include <rte_log.h>
  15 #include <rte_string_fns.h>
  16 #include <rte_mbuf.h>
  17 #include <rte_malloc.h>
  18 #include <rte_ip.h>
  19 #include <rte_udp.h>
  20 #include <rte_tcp.h>
  21
  22 #include "main.h"
  23 #include "rte_vhost.h"
  24 #include "vxlan.h"
  25 #include "vxlan_setup.h"
  26
  27 #define IPV4_HEADER_LEN 20
  28 #define UDP_HEADER_LEN  8
  29 #define VXLAN_HEADER_LEN 8
  30
  31 #define IP_VERSION 0x40
  32 #define IP_HDRLEN  0x05 /* default IP header length == five 32-bits words. */
  33 #define IP_DEFTTL  64   /* from RFC 1340. */
  34 #define IP_VHL_DEF (IP_VERSION | IP_HDRLEN)
  35
  36 #define IP_DN_FRAGMENT_FLAG 0x0040
  37
  38 /* Used to compare MAC addresses. */
  39 #define MAC_ADDR_CMP 0xFFFFFFFFFFFFULL
  40
  41 /* Configurable number of RX/TX ring descriptors */
  42 #define RTE_TEST_RX_DESC_DEFAULT 1024
  43 #define RTE_TEST_TX_DESC_DEFAULT 512
  44
  45 /* Default inner VLAN ID */
  46 #define INNER_VLAN_ID 100
  47
  48 /* VXLAN device */
  49 struct vxlan_conf vxdev;
  50
  51 struct ipv4_hdr app_ip_hdr[VXLAN_N_PORTS];
  52 struct ether_hdr app_l2_hdr[VXLAN_N_PORTS];
  53
  54 /* local VTEP IP address */
  55 uint8_t vxlan_multicast_ips[2][4] = { {239, 1, 1, 1 }, {239, 1, 2, 1 } };
  56
  57 /* Remote VTEP IP address */
  58 uint8_t vxlan_overlay_ips[2][4] = { {192, 168, 10, 1}, {192, 168, 30, 1} };
  59
  60 /* Remote VTEP MAC address */
  61 uint8_t peer_mac[6] = {0x00, 0x11, 0x01, 0x00, 0x00, 0x01};
  62
  63 /* VXLAN RX filter type */
  64 uint8_t tep_filter_type[] = {RTE_TUNNEL_FILTER_IMAC_TENID,
  65                         RTE_TUNNEL_FILTER_IMAC_IVLAN_TENID,
  66                         RTE_TUNNEL_FILTER_OMAC_TENID_IMAC,};
  67
  68 /* Options for configuring ethernet port */
  69 static struct rte_eth_conf port_conf = {
  70         .rxmode = {
  71                 .split_hdr_size = 0,
  72                 .ignore_offload_bitfield = 1,
  73                 .offloads = DEV_RX_OFFLOAD_CRC_STRIP,
  74         },
  75         .txmode = {
  76                 .mq_mode = ETH_MQ_TX_NONE,
  77                 .offloads = (DEV_TX_OFFLOAD_IPV4_CKSUM |
  78                              DEV_TX_OFFLOAD_UDP_CKSUM |
  79                              DEV_TX_OFFLOAD_TCP_CKSUM |
  80                              DEV_TX_OFFLOAD_SCTP_CKSUM |
  81                              DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM |
  82                              DEV_TX_OFFLOAD_TCP_TSO |
  83                              DEV_TX_OFFLOAD_MULTI_SEGS |
  84                              DEV_TX_OFFLOAD_VXLAN_TNL_TSO),
  85         },
  86 };
  87
  88 /**
  89  * The one or two device(s) that belongs to the same tenant ID can
  90  * be assigned in a VM.
  91  */
  92 const uint16_t tenant_id_conf[] = {
  93         1000, 1000, 1001, 1001, 1002, 1002, 1003, 1003,
  94         1004, 1004, 1005, 1005, 1006, 1006, 1007, 1007,
  95         1008, 1008, 1009, 1009, 1010, 1010, 1011, 1011,
  96         1012, 1012, 1013, 1013, 1014, 1014, 1015, 1015,
  97         1016, 1016, 1017, 1017, 1018, 1018, 1019, 1019,
  98         1020, 1020, 1021, 1021, 1022, 1022, 1023, 1023,
  99         1024, 1024, 1025, 1025, 1026, 1026, 1027, 1027,
 100         1028, 1028, 1029, 1029, 1030, 1030, 1031, 1031,
 101 };
 102
 103 /**
 104  * Initialises a given port using global settings and with the rx buffers
 105  * coming from the mbuf_pool passed as parameter
 106  */
 107 int
 108 vxlan_port_init(uint16_t port, struct rte_mempool *mbuf_pool)
 109 {
 110         int retval;
 111         uint16_t q;
 112         struct rte_eth_dev_info dev_info;
 113         uint16_t rx_rings, tx_rings = (uint16_t)rte_lcore_count();
 114         uint16_t rx_ring_size = RTE_TEST_RX_DESC_DEFAULT;
 115         uint16_t tx_ring_size = RTE_TEST_TX_DESC_DEFAULT;
 116         struct rte_eth_udp_tunnel tunnel_udp;
 117         struct rte_eth_rxconf *rxconf;
 118         struct rte_eth_txconf *txconf;
 119         struct vxlan_conf *pconf = &vxdev;
 120         struct rte_eth_conf local_port_conf = port_conf;
 121
 122         pconf->dst_port = udp_port;
 123
 124         rte_eth_dev_info_get(port, &dev_info);
 125
 126         if (dev_info.max_rx_queues > MAX_QUEUES) {
 127                 rte_exit(EXIT_FAILURE,
 128                         "please define MAX_QUEUES no less than %u in %s\n",
 129                         dev_info.max_rx_queues, __FILE__);
 130         }
 131
 132         rxconf = &dev_info.default_rxconf;
 133         txconf = &dev_info.default_txconf;
 134         txconf->txq_flags = ETH_TXQ_FLAGS_IGNORE;
 135
 136         if (!rte_eth_dev_is_valid_port(port))
 137                 return -1;
 138
 139         rx_rings = nb_devices;
 140         if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
 141                 local_port_conf.txmode.offloads |=
 142                         DEV_TX_OFFLOAD_MBUF_FAST_FREE;
 143         /* Configure ethernet device. */
 144         retval = rte_eth_dev_configure(port, rx_rings, tx_rings,
 145                                        &local_port_conf);
 146         if (retval != 0)
 147                 return retval;
 148
 149         retval = rte_eth_dev_adjust_nb_rx_tx_desc(port, &rx_ring_size,
 150                         &tx_ring_size);
 151         if (retval != 0)
 152                 return retval;
 153
 154         /* Setup the queues. */
 155         rxconf->offloads = local_port_conf.rxmode.offloads;
 156         for (q = 0; q < rx_rings; q++) {
 157                 retval = rte_eth_rx_queue_setup(port, q, rx_ring_size,
 158                                                 rte_eth_dev_socket_id(port),
 159                                                 rxconf,
 160                                                 mbuf_pool);
 161                 if (retval < 0)
 162                         return retval;
 163         }
 164         txconf->offloads = local_port_conf.txmode.offloads;
 165         for (q = 0; q < tx_rings; q++) {
 166                 retval = rte_eth_tx_queue_setup(port, q, tx_ring_size,
 167                                                 rte_eth_dev_socket_id(port),
 168                                                 txconf);
 169                 if (retval < 0)
 170                         return retval;
 171         }
 172
 173         /* Start the device. */
 174         retval  = rte_eth_dev_start(port);
 175         if (retval < 0)
 176                 return retval;
 177
 178         /* Configure UDP port for UDP tunneling */
 179         tunnel_udp.udp_port = udp_port;
 180         tunnel_udp.prot_type = RTE_TUNNEL_TYPE_VXLAN;
 181         retval = rte_eth_dev_udp_tunnel_port_add(port, &tunnel_udp);
 182         if (retval < 0)
 183                 return retval;
 184         rte_eth_macaddr_get(port, &ports_eth_addr[port]);
 185         RTE_LOG(INFO, PORT, "Port %u MAC: %02"PRIx8" %02"PRIx8" %02"PRIx8
 186                         " %02"PRIx8" %02"PRIx8" %02"PRIx8"\n",
 187                         port,
 188                         ports_eth_addr[port].addr_bytes[0],
 189                         ports_eth_addr[port].addr_bytes[1],
 190                         ports_eth_addr[port].addr_bytes[2],
 191                         ports_eth_addr[port].addr_bytes[3],
 192                         ports_eth_addr[port].addr_bytes[4],
 193                         ports_eth_addr[port].addr_bytes[5]);
 194
 195         if (tso_segsz != 0) {
 196                 struct rte_eth_dev_info dev_info;
 197                 rte_eth_dev_info_get(port, &dev_info);
 198                 if ((dev_info.tx_offload_capa & DEV_TX_OFFLOAD_TCP_TSO) == 0)
 199                         RTE_LOG(WARNING, PORT,
 200                                 "hardware TSO offload is not supported\n");
 201         }
 202         return 0;
 203 }
 204
 205 static int
 206 vxlan_rx_process(struct rte_mbuf *pkt)
 207 {
 208         int ret = 0;
 209
 210         if (rx_decap)
 211                 ret = decapsulation(pkt);
 212
 213         return ret;
 214 }
 215
 216 static void
 217 vxlan_tx_process(uint8_t queue_id, struct rte_mbuf *pkt)
 218 {
 219         if (tx_encap)
 220                 encapsulation(pkt, queue_id);
 221
 222         return;
 223 }
 224
 225 /*
 226  * This function learns the MAC address of the device and set init
 227  * L2 header and L3 header info.
 228  */
 229 int
 230 vxlan_link(struct vhost_dev *vdev, struct rte_mbuf *m)
 231 {
 232         int i, ret;
 233         struct ether_hdr *pkt_hdr;
 234         uint64_t portid = vdev->vid;
 235         struct ipv4_hdr *ip;
 236
 237         struct rte_eth_tunnel_filter_conf tunnel_filter_conf;
 238
 239         if (unlikely(portid >= VXLAN_N_PORTS)) {
 240                 RTE_LOG(INFO, VHOST_DATA,
 241                         "(%d) WARNING: Not configuring device,"
 242                         "as already have %d ports for VXLAN.",
 243                         vdev->vid, VXLAN_N_PORTS);
 244                 return -1;
 245         }
 246
 247         /* Learn MAC address of guest device from packet */
 248         pkt_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 249         if (is_same_ether_addr(&(pkt_hdr->s_addr), &vdev->mac_address)) {
 250                 RTE_LOG(INFO, VHOST_DATA,
 251                         "(%d) WARNING: This device is using an existing"
 252                         " MAC address and has not been registered.\n",
 253                         vdev->vid);
 254                 return -1;
 255         }
 256
 257         for (i = 0; i < ETHER_ADDR_LEN; i++) {
 258                 vdev->mac_address.addr_bytes[i] =
 259                         vxdev.port[portid].vport_mac.addr_bytes[i] =
 260                         pkt_hdr->s_addr.addr_bytes[i];
 261                 vxdev.port[portid].peer_mac.addr_bytes[i] = peer_mac[i];
 262         }
 263
 264         memset(&tunnel_filter_conf, 0,
 265                 sizeof(struct rte_eth_tunnel_filter_conf));
 266
 267         ether_addr_copy(&ports_eth_addr[0], &tunnel_filter_conf.outer_mac);
 268         tunnel_filter_conf.filter_type = tep_filter_type[filter_idx];
 269
 270         /* inner MAC */
 271         ether_addr_copy(&vdev->mac_address, &tunnel_filter_conf.inner_mac);
 272
 273         tunnel_filter_conf.queue_id = vdev->rx_q;
 274         tunnel_filter_conf.tenant_id = tenant_id_conf[vdev->rx_q];
 275
 276         if (tep_filter_type[filter_idx] == RTE_TUNNEL_FILTER_IMAC_IVLAN_TENID)
 277                 tunnel_filter_conf.inner_vlan = INNER_VLAN_ID;
 278
 279         tunnel_filter_conf.tunnel_type = RTE_TUNNEL_TYPE_VXLAN;
 280
 281         ret = rte_eth_dev_filter_ctrl(ports[0],
 282                 RTE_ETH_FILTER_TUNNEL,
 283                 RTE_ETH_FILTER_ADD,
 284                 &tunnel_filter_conf);
 285         if (ret) {
 286                 RTE_LOG(ERR, VHOST_DATA,
 287                         "%d Failed to add device MAC address to cloud filter\n",
 288                 vdev->rx_q);
 289                 return -1;
 290         }
 291
 292         /* Print out inner MAC and VNI info. */
 293         RTE_LOG(INFO, VHOST_DATA,
 294                 "(%d) MAC_ADDRESS %02x:%02x:%02x:%02x:%02x:%02x and VNI %d registered\n",
 295                 vdev->rx_q,
 296                 vdev->mac_address.addr_bytes[0],
 297                 vdev->mac_address.addr_bytes[1],
 298                 vdev->mac_address.addr_bytes[2],
 299                 vdev->mac_address.addr_bytes[3],
 300                 vdev->mac_address.addr_bytes[4],
 301                 vdev->mac_address.addr_bytes[5],
 302                 tenant_id_conf[vdev->rx_q]);
 303
 304         vxdev.port[portid].vport_id = portid;
 305
 306         for (i = 0; i < 4; i++) {
 307                 /* Local VTEP IP */
 308                 vxdev.port_ip |= vxlan_multicast_ips[portid][i] << (8 * i);
 309                 /* Remote VTEP IP */
 310                 vxdev.port[portid].peer_ip |=
 311                         vxlan_overlay_ips[portid][i] << (8 * i);
 312         }
 313
 314         vxdev.out_key = tenant_id_conf[vdev->rx_q];
 315         ether_addr_copy(&vxdev.port[portid].peer_mac,
 316                         &app_l2_hdr[portid].d_addr);
 317         ether_addr_copy(&ports_eth_addr[0],
 318                         &app_l2_hdr[portid].s_addr);
 319         app_l2_hdr[portid].ether_type = rte_cpu_to_be_16(ETHER_TYPE_IPv4);
 320
 321         ip = &app_ip_hdr[portid];
 322         ip->version_ihl = IP_VHL_DEF;
 323         ip->type_of_service = 0;
 324         ip->total_length = 0;
 325         ip->packet_id = 0;
 326         ip->fragment_offset = IP_DN_FRAGMENT_FLAG;
 327         ip->time_to_live = IP_DEFTTL;
 328         ip->next_proto_id = IPPROTO_UDP;
 329         ip->hdr_checksum = 0;
 330         ip->src_addr = vxdev.port_ip;
 331         ip->dst_addr = vxdev.port[portid].peer_ip;
 332
 333         /* Set device as ready for RX. */
 334         vdev->ready = DEVICE_RX;
 335
 336         return 0;
 337 }
 338
 339 /**
 340  * Removes cloud filter. Ensures that nothing is adding buffers to the RX
 341  * queue before disabling RX on the device.
 342  */
 343 void
 344 vxlan_unlink(struct vhost_dev *vdev)
 345 {
 346         unsigned i = 0, rx_count;
 347         int ret;
 348         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 349         struct rte_eth_tunnel_filter_conf tunnel_filter_conf;
 350
 351         if (vdev->ready == DEVICE_RX) {
 352                 memset(&tunnel_filter_conf, 0,
 353                         sizeof(struct rte_eth_tunnel_filter_conf));
 354
 355                 ether_addr_copy(&ports_eth_addr[0], &tunnel_filter_conf.outer_mac);
 356                 ether_addr_copy(&vdev->mac_address, &tunnel_filter_conf.inner_mac);
 357                 tunnel_filter_conf.tenant_id = tenant_id_conf[vdev->rx_q];
 358                 tunnel_filter_conf.filter_type = tep_filter_type[filter_idx];
 359
 360                 if (tep_filter_type[filter_idx] ==
 361                         RTE_TUNNEL_FILTER_IMAC_IVLAN_TENID)
 362                         tunnel_filter_conf.inner_vlan = INNER_VLAN_ID;
 363
 364                 tunnel_filter_conf.queue_id = vdev->rx_q;
 365                 tunnel_filter_conf.tunnel_type = RTE_TUNNEL_TYPE_VXLAN;
 366
 367                 ret = rte_eth_dev_filter_ctrl(ports[0],
 368                                 RTE_ETH_FILTER_TUNNEL,
 369                                 RTE_ETH_FILTER_DELETE,
 370                                 &tunnel_filter_conf);
 371                 if (ret) {
 372                         RTE_LOG(ERR, VHOST_DATA,
 373                                 "%d Failed to add device MAC address to cloud filter\n",
 374                                 vdev->rx_q);
 375                         return;
 376                 }
 377                 for (i = 0; i < ETHER_ADDR_LEN; i++)
 378                         vdev->mac_address.addr_bytes[i] = 0;
 379
 380                 /* Clear out the receive buffers */
 381                 rx_count = rte_eth_rx_burst(ports[0],
 382                                 (uint16_t)vdev->rx_q,
 383                                 pkts_burst, MAX_PKT_BURST);
 384
 385                 while (rx_count) {
 386                         for (i = 0; i < rx_count; i++)
 387                                 rte_pktmbuf_free(pkts_burst[i]);
 388
 389                         rx_count = rte_eth_rx_burst(ports[0],
 390                                         (uint16_t)vdev->rx_q,
 391                                         pkts_burst, MAX_PKT_BURST);
 392                 }
 393                 vdev->ready = DEVICE_MAC_LEARNING;
 394         }
 395 }
 396
 397 /* Transmit packets after encapsulating */
 398 int
 399 vxlan_tx_pkts(uint16_t port_id, uint16_t queue_id,
 400                 struct rte_mbuf **tx_pkts, uint16_t nb_pkts) {
 401         int ret = 0;
 402         uint16_t i;
 403
 404         for (i = 0; i < nb_pkts; i++)
 405                 vxlan_tx_process(queue_id, tx_pkts[i]);
 406
 407         ret = rte_eth_tx_burst(port_id, queue_id, tx_pkts, nb_pkts);
 408
 409         return ret;
 410 }
 411
 412 /* Check for decapsulation and pass packets directly to VIRTIO device */
 413 int
 414 vxlan_rx_pkts(int vid, struct rte_mbuf **pkts_burst, uint32_t rx_count)
 415 {
 416         uint32_t i = 0;
 417         uint32_t count = 0;
 418         int ret;
 419         struct rte_mbuf *pkts_valid[rx_count];
 420
 421         for (i = 0; i < rx_count; i++) {
 422                 if (enable_stats) {
 423                         rte_atomic64_add(
 424                                 &dev_statistics[vid].rx_bad_ip_csum,
 425                                 (pkts_burst[i]->ol_flags & PKT_RX_IP_CKSUM_BAD)
 426                                 != 0);
 427                         rte_atomic64_add(
 428                                 &dev_statistics[vid].rx_bad_ip_csum,
 429                                 (pkts_burst[i]->ol_flags & PKT_RX_L4_CKSUM_BAD)
 430                                 != 0);
 431                 }
 432                 ret = vxlan_rx_process(pkts_burst[i]);
 433                 if (unlikely(ret < 0))
 434                         continue;
 435
 436                 pkts_valid[count] = pkts_burst[i];
 437                         count++;
 438         }
 439
 440         ret = rte_vhost_enqueue_burst(vid, VIRTIO_RXQ, pkts_valid, count);
 441         return ret;
 442 }