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