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_DEFTTL  64   /* from RFC 1340. */
  32
  33 #define IP_DN_FRAGMENT_FLAG 0x0040
  34
  35 /* Used to compare MAC addresses. */
  36 #define MAC_ADDR_CMP 0xFFFFFFFFFFFFULL
  37
  38 /* Configurable number of RX/TX ring descriptors */
  39 #define RTE_TEST_RX_DESC_DEFAULT 1024
  40 #define RTE_TEST_TX_DESC_DEFAULT 512
  41
  42 /* Default inner VLAN ID */
  43 #define INNER_VLAN_ID 100
  44
  45 /* VXLAN device */
  46 struct vxlan_conf vxdev;
  47
  48 struct rte_ipv4_hdr app_ip_hdr[VXLAN_N_PORTS];
  49 struct rte_ether_hdr app_l2_hdr[VXLAN_N_PORTS];
  50
  51 /* local VTEP IP address */
  52 uint8_t vxlan_multicast_ips[2][4] = { {239, 1, 1, 1 }, {239, 1, 2, 1 } };
  53
  54 /* Remote VTEP IP address */
  55 uint8_t vxlan_overlay_ips[2][4] = { {192, 168, 10, 1}, {192, 168, 30, 1} };
  56
  57 /* Remote VTEP MAC address */
  58 uint8_t peer_mac[6] = {0x00, 0x11, 0x01, 0x00, 0x00, 0x01};
  59
  60 /* VXLAN RX filter type */
  61 uint8_t tep_filter_type[] = {RTE_TUNNEL_FILTER_IMAC_TENID,
  62                         RTE_TUNNEL_FILTER_IMAC_IVLAN_TENID,
  63                         RTE_TUNNEL_FILTER_OMAC_TENID_IMAC,};
  64
  65 /* Options for configuring ethernet port */
  66 static struct rte_eth_conf port_conf = {
  67         .rxmode = {
  68                 .split_hdr_size = 0,
  69         },
  70         .txmode = {
  71                 .mq_mode = ETH_MQ_TX_NONE,
  72                 .offloads = (DEV_TX_OFFLOAD_IPV4_CKSUM |
  73                              DEV_TX_OFFLOAD_UDP_CKSUM |
  74                              DEV_TX_OFFLOAD_TCP_CKSUM |
  75                              DEV_TX_OFFLOAD_SCTP_CKSUM |
  76                              DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM |
  77                              DEV_TX_OFFLOAD_TCP_TSO |
  78                              DEV_TX_OFFLOAD_MULTI_SEGS |
  79                              DEV_TX_OFFLOAD_VXLAN_TNL_TSO),
  80         },
  81 };
  82
  83 /**
  84  * The one or two device(s) that belongs to the same tenant ID can
  85  * be assigned in a VM.
  86  */
  87 const uint16_t tenant_id_conf[] = {
  88         1000, 1000, 1001, 1001, 1002, 1002, 1003, 1003,
  89         1004, 1004, 1005, 1005, 1006, 1006, 1007, 1007,
  90         1008, 1008, 1009, 1009, 1010, 1010, 1011, 1011,
  91         1012, 1012, 1013, 1013, 1014, 1014, 1015, 1015,
  92         1016, 1016, 1017, 1017, 1018, 1018, 1019, 1019,
  93         1020, 1020, 1021, 1021, 1022, 1022, 1023, 1023,
  94         1024, 1024, 1025, 1025, 1026, 1026, 1027, 1027,
  95         1028, 1028, 1029, 1029, 1030, 1030, 1031, 1031,
  96 };
  97
  98 /**
  99  * Initialises a given port using global settings and with the rx buffers
 100  * coming from the mbuf_pool passed as parameter
 101  */
 102 int
 103 vxlan_port_init(uint16_t port, struct rte_mempool *mbuf_pool)
 104 {
 105         int retval;
 106         uint16_t q;
 107         struct rte_eth_dev_info dev_info;
 108         uint16_t rx_rings, tx_rings = (uint16_t)rte_lcore_count();
 109         uint16_t rx_ring_size = RTE_TEST_RX_DESC_DEFAULT;
 110         uint16_t tx_ring_size = RTE_TEST_TX_DESC_DEFAULT;
 111         struct rte_eth_udp_tunnel tunnel_udp;
 112         struct rte_eth_rxconf *rxconf;
 113         struct rte_eth_txconf *txconf;
 114         struct vxlan_conf *pconf = &vxdev;
 115         struct rte_eth_conf local_port_conf = port_conf;
 116
 117         pconf->dst_port = udp_port;
 118
 119         retval = rte_eth_dev_info_get(port, &dev_info);
 120         if (retval != 0)
 121                 rte_exit(EXIT_FAILURE,
 122                         "Error during getting device (port %u) info: %s\n",
 123                         port, strerror(-retval));
 124
 125         if (dev_info.max_rx_queues > MAX_QUEUES) {
 126                 rte_exit(EXIT_FAILURE,
 127                         "please define MAX_QUEUES no less than %u in %s\n",
 128                         dev_info.max_rx_queues, __FILE__);
 129         }
 130
 131         rxconf = &dev_info.default_rxconf;
 132         txconf = &dev_info.default_txconf;
 133
 134         if (!rte_eth_dev_is_valid_port(port))
 135                 return -1;
 136
 137         rx_rings = nb_devices;
 138         if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
 139                 local_port_conf.txmode.offloads |=
 140                         DEV_TX_OFFLOAD_MBUF_FAST_FREE;
 141         /* Configure ethernet device. */
 142         retval = rte_eth_dev_configure(port, rx_rings, tx_rings,
 143                                        &local_port_conf);
 144         if (retval != 0)
 145                 return retval;
 146
 147         retval = rte_eth_dev_adjust_nb_rx_tx_desc(port, &rx_ring_size,
 148                         &tx_ring_size);
 149         if (retval != 0)
 150                 return retval;
 151
 152         /* Setup the queues. */
 153         rxconf->offloads = local_port_conf.rxmode.offloads;
 154         for (q = 0; q < rx_rings; q++) {
 155                 retval = rte_eth_rx_queue_setup(port, q, rx_ring_size,
 156                                                 rte_eth_dev_socket_id(port),
 157                                                 rxconf,
 158                                                 mbuf_pool);
 159                 if (retval < 0)
 160                         return retval;
 161         }
 162         txconf->offloads = local_port_conf.txmode.offloads;
 163         for (q = 0; q < tx_rings; q++) {
 164                 retval = rte_eth_tx_queue_setup(port, q, tx_ring_size,
 165                                                 rte_eth_dev_socket_id(port),
 166                                                 txconf);
 167                 if (retval < 0)
 168                         return retval;
 169         }
 170
 171         /* Start the device. */
 172         retval  = rte_eth_dev_start(port);
 173         if (retval < 0)
 174                 return retval;
 175
 176         /* Configure UDP port for UDP tunneling */
 177         tunnel_udp.udp_port = udp_port;
 178         tunnel_udp.prot_type = RTE_TUNNEL_TYPE_VXLAN;
 179         retval = rte_eth_dev_udp_tunnel_port_add(port, &tunnel_udp);
 180         if (retval < 0)
 181                 return retval;
 182         retval = rte_eth_macaddr_get(port, &ports_eth_addr[port]);
 183         if (retval < 0)
 184                 return retval;
 185
 186         RTE_LOG(INFO, PORT, "Port %u MAC: %02"PRIx8" %02"PRIx8" %02"PRIx8
 187                         " %02"PRIx8" %02"PRIx8" %02"PRIx8"\n",
 188                         port,
 189                         ports_eth_addr[port].addr_bytes[0],
 190                         ports_eth_addr[port].addr_bytes[1],
 191                         ports_eth_addr[port].addr_bytes[2],
 192                         ports_eth_addr[port].addr_bytes[3],
 193                         ports_eth_addr[port].addr_bytes[4],
 194                         ports_eth_addr[port].addr_bytes[5]);
 195
 196         if (tso_segsz != 0) {
 197                 if ((dev_info.tx_offload_capa & DEV_TX_OFFLOAD_TCP_TSO) == 0)
 198                         RTE_LOG(WARNING, PORT,
 199                                 "hardware TSO offload is not supported\n");
 200         }
 201         return 0;
 202 }
 203
 204 static int
 205 vxlan_rx_process(struct rte_mbuf *pkt)
 206 {
 207         int ret = 0;
 208
 209         if (rx_decap)
 210                 ret = decapsulation(pkt);
 211
 212         return ret;
 213 }
 214
 215 static void
 216 vxlan_tx_process(uint8_t queue_id, struct rte_mbuf *pkt)
 217 {
 218         if (tx_encap)
 219                 encapsulation(pkt, queue_id);
 220
 221         return;
 222 }
 223
 224 /*
 225  * This function learns the MAC address of the device and set init
 226  * L2 header and L3 header info.
 227  */
 228 int
 229 vxlan_link(struct vhost_dev *vdev, struct rte_mbuf *m)
 230 {
 231         int i, ret;
 232         struct rte_ether_hdr *pkt_hdr;
 233         uint64_t portid = vdev->vid;
 234         struct rte_ipv4_hdr *ip;
 235
 236         struct rte_eth_tunnel_filter_conf tunnel_filter_conf;
 237
 238         if (unlikely(portid >= VXLAN_N_PORTS)) {
 239                 RTE_LOG(INFO, VHOST_DATA,
 240                         "(%d) WARNING: Not configuring device,"
 241                         "as already have %d ports for VXLAN.",
 242                         vdev->vid, VXLAN_N_PORTS);
 243                 return -1;
 244         }
 245
 246         /* Learn MAC address of guest device from packet */
 247         pkt_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 248         if (rte_is_same_ether_addr(&(pkt_hdr->s_addr), &vdev->mac_address)) {
 249                 RTE_LOG(INFO, VHOST_DATA,
 250                         "(%d) WARNING: This device is using an existing"
 251                         " MAC address and has not been registered.\n",
 252                         vdev->vid);
 253                 return -1;
 254         }
 255
 256         for (i = 0; i < RTE_ETHER_ADDR_LEN; i++) {
 257                 vdev->mac_address.addr_bytes[i] =
 258                         vxdev.port[portid].vport_mac.addr_bytes[i] =
 259                         pkt_hdr->s_addr.addr_bytes[i];
 260                 vxdev.port[portid].peer_mac.addr_bytes[i] = peer_mac[i];
 261         }
 262
 263         memset(&tunnel_filter_conf, 0,
 264                 sizeof(struct rte_eth_tunnel_filter_conf));
 265
 266         rte_ether_addr_copy(&ports_eth_addr[0], &tunnel_filter_conf.outer_mac);
 267         tunnel_filter_conf.filter_type = tep_filter_type[filter_idx];
 268
 269         /* inner MAC */
 270         rte_ether_addr_copy(&vdev->mac_address, &tunnel_filter_conf.inner_mac);
 271
 272         tunnel_filter_conf.queue_id = vdev->rx_q;
 273         tunnel_filter_conf.tenant_id = tenant_id_conf[vdev->rx_q];
 274
 275         if (tep_filter_type[filter_idx] == RTE_TUNNEL_FILTER_IMAC_IVLAN_TENID)
 276                 tunnel_filter_conf.inner_vlan = INNER_VLAN_ID;
 277
 278         tunnel_filter_conf.tunnel_type = RTE_TUNNEL_TYPE_VXLAN;
 279
 280         ret = rte_eth_dev_filter_ctrl(ports[0],
 281                 RTE_ETH_FILTER_TUNNEL,
 282                 RTE_ETH_FILTER_ADD,
 283                 &tunnel_filter_conf);
 284         if (ret) {
 285                 RTE_LOG(ERR, VHOST_DATA,
 286                         "%d Failed to add device MAC address to cloud filter\n",
 287                 vdev->rx_q);
 288                 return -1;
 289         }
 290
 291         /* Print out inner MAC and VNI info. */
 292         RTE_LOG(INFO, VHOST_DATA,
 293                 "(%d) MAC_ADDRESS %02x:%02x:%02x:%02x:%02x:%02x and VNI %d registered\n",
 294                 vdev->rx_q,
 295                 vdev->mac_address.addr_bytes[0],
 296                 vdev->mac_address.addr_bytes[1],
 297                 vdev->mac_address.addr_bytes[2],
 298                 vdev->mac_address.addr_bytes[3],
 299                 vdev->mac_address.addr_bytes[4],
 300                 vdev->mac_address.addr_bytes[5],
 301                 tenant_id_conf[vdev->rx_q]);
 302
 303         vxdev.port[portid].vport_id = portid;
 304
 305         for (i = 0; i < 4; i++) {
 306                 /* Local VTEP IP */
 307                 vxdev.port_ip |= vxlan_multicast_ips[portid][i] << (8 * i);
 308                 /* Remote VTEP IP */
 309                 vxdev.port[portid].peer_ip |=
 310                         vxlan_overlay_ips[portid][i] << (8 * i);
 311         }
 312
 313         vxdev.out_key = tenant_id_conf[vdev->rx_q];
 314         rte_ether_addr_copy(&vxdev.port[portid].peer_mac,
 315                         &app_l2_hdr[portid].d_addr);
 316         rte_ether_addr_copy(&ports_eth_addr[0],
 317                         &app_l2_hdr[portid].s_addr);
 318         app_l2_hdr[portid].ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4);
 319
 320         ip = &app_ip_hdr[portid];
 321         ip->version_ihl = RTE_IPV4_VHL_DEF;
 322         ip->type_of_service = 0;
 323         ip->total_length = 0;
 324         ip->packet_id = 0;
 325         ip->fragment_offset = IP_DN_FRAGMENT_FLAG;
 326         ip->time_to_live = IP_DEFTTL;
 327         ip->next_proto_id = IPPROTO_UDP;
 328         ip->hdr_checksum = 0;
 329         ip->src_addr = vxdev.port_ip;
 330         ip->dst_addr = vxdev.port[portid].peer_ip;
 331
 332         /* Set device as ready for RX. */
 333         vdev->ready = DEVICE_RX;
 334
 335         return 0;
 336 }
 337
 338 /**
 339  * Removes cloud filter. Ensures that nothing is adding buffers to the RX
 340  * queue before disabling RX on the device.
 341  */
 342 void
 343 vxlan_unlink(struct vhost_dev *vdev)
 344 {
 345         unsigned i = 0, rx_count;
 346         int ret;
 347         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 348         struct rte_eth_tunnel_filter_conf tunnel_filter_conf;
 349
 350         if (vdev->ready == DEVICE_RX) {
 351                 memset(&tunnel_filter_conf, 0,
 352                         sizeof(struct rte_eth_tunnel_filter_conf));
 353
 354                 rte_ether_addr_copy(&ports_eth_addr[0],
 355                                 &tunnel_filter_conf.outer_mac);
 356                 rte_ether_addr_copy(&vdev->mac_address,
 357                                 &tunnel_filter_conf.inner_mac);
 358                 tunnel_filter_conf.tenant_id = tenant_id_conf[vdev->rx_q];
 359                 tunnel_filter_conf.filter_type = tep_filter_type[filter_idx];
 360
 361                 if (tep_filter_type[filter_idx] ==
 362                         RTE_TUNNEL_FILTER_IMAC_IVLAN_TENID)
 363                         tunnel_filter_conf.inner_vlan = INNER_VLAN_ID;
 364
 365                 tunnel_filter_conf.queue_id = vdev->rx_q;
 366                 tunnel_filter_conf.tunnel_type = RTE_TUNNEL_TYPE_VXLAN;
 367
 368                 ret = rte_eth_dev_filter_ctrl(ports[0],
 369                                 RTE_ETH_FILTER_TUNNEL,
 370                                 RTE_ETH_FILTER_DELETE,
 371                                 &tunnel_filter_conf);
 372                 if (ret) {
 373                         RTE_LOG(ERR, VHOST_DATA,
 374                                 "%d Failed to add device MAC address to cloud filter\n",
 375                                 vdev->rx_q);
 376                         return;
 377                 }
 378                 for (i = 0; i < RTE_ETHER_ADDR_LEN; i++)
 379                         vdev->mac_address.addr_bytes[i] = 0;
 380
 381                 /* Clear out the receive buffers */
 382                 rx_count = rte_eth_rx_burst(ports[0],
 383                                 (uint16_t)vdev->rx_q,
 384                                 pkts_burst, MAX_PKT_BURST);
 385
 386                 while (rx_count) {
 387                         for (i = 0; i < rx_count; i++)
 388                                 rte_pktmbuf_free(pkts_burst[i]);
 389
 390                         rx_count = rte_eth_rx_burst(ports[0],
 391                                         (uint16_t)vdev->rx_q,
 392                                         pkts_burst, MAX_PKT_BURST);
 393                 }
 394                 vdev->ready = DEVICE_MAC_LEARNING;
 395         }
 396 }
 397
 398 /* Transmit packets after encapsulating */
 399 int
 400 vxlan_tx_pkts(uint16_t port_id, uint16_t queue_id,
 401                 struct rte_mbuf **tx_pkts, uint16_t nb_pkts) {
 402         int ret = 0;
 403         uint16_t i;
 404
 405         for (i = 0; i < nb_pkts; i++)
 406                 vxlan_tx_process(queue_id, tx_pkts[i]);
 407
 408         ret = rte_eth_tx_burst(port_id, queue_id, tx_pkts, nb_pkts);
 409
 410         return ret;
 411 }
 412
 413 /* Check for decapsulation and pass packets directly to VIRTIO device */
 414 int
 415 vxlan_rx_pkts(int vid, struct rte_mbuf **pkts_burst, uint32_t rx_count)
 416 {
 417         uint32_t i = 0;
 418         uint32_t count = 0;
 419         int ret;
 420         struct rte_mbuf *pkts_valid[rx_count];
 421
 422         for (i = 0; i < rx_count; i++) {
 423                 if (enable_stats) {
 424                         rte_atomic64_add(
 425                                 &dev_statistics[vid].rx_bad_ip_csum,
 426                                 (pkts_burst[i]->ol_flags & PKT_RX_IP_CKSUM_BAD)
 427                                 != 0);
 428                         rte_atomic64_add(
 429                                 &dev_statistics[vid].rx_bad_ip_csum,
 430                                 (pkts_burst[i]->ol_flags & PKT_RX_L4_CKSUM_BAD)
 431                                 != 0);
 432                 }
 433                 ret = vxlan_rx_process(pkts_burst[i]);
 434                 if (unlikely(ret < 0))
 435                         continue;
 436
 437                 pkts_valid[count] = pkts_burst[i];
 438                         count++;
 439         }
 440
 441         ret = rte_vhost_enqueue_burst(vid, VIRTIO_RXQ, pkts_valid, count);
 442         return ret;
 443 }