ceph/src/spdk/dpdk/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                 .offloads = DEV_RX_OFFLOAD_CRC_STRIP,
  73         },
  74         .txmode = {
  75                 .mq_mode = ETH_MQ_TX_NONE,
  76                 .offloads = (DEV_TX_OFFLOAD_IPV4_CKSUM |
  77                              DEV_TX_OFFLOAD_UDP_CKSUM |
  78                              DEV_TX_OFFLOAD_TCP_CKSUM |
  79                              DEV_TX_OFFLOAD_SCTP_CKSUM |
  80                              DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM |
  81                              DEV_TX_OFFLOAD_TCP_TSO |
  82                              DEV_TX_OFFLOAD_MULTI_SEGS |
  83                              DEV_TX_OFFLOAD_VXLAN_TNL_TSO),
  84         },
  85 };
  86
  87 /**
  88  * The one or two device(s) that belongs to the same tenant ID can
  89  * be assigned in a VM.
  90  */
  91 const uint16_t tenant_id_conf[] = {
  92         1000, 1000, 1001, 1001, 1002, 1002, 1003, 1003,
  93         1004, 1004, 1005, 1005, 1006, 1006, 1007, 1007,
  94         1008, 1008, 1009, 1009, 1010, 1010, 1011, 1011,
  95         1012, 1012, 1013, 1013, 1014, 1014, 1015, 1015,
  96         1016, 1016, 1017, 1017, 1018, 1018, 1019, 1019,
  97         1020, 1020, 1021, 1021, 1022, 1022, 1023, 1023,
  98         1024, 1024, 1025, 1025, 1026, 1026, 1027, 1027,
  99         1028, 1028, 1029, 1029, 1030, 1030, 1031, 1031,
 100 };
 101
 102 /**
 103  * Initialises a given port using global settings and with the rx buffers
 104  * coming from the mbuf_pool passed as parameter
 105  */
 106 int
 107 vxlan_port_init(uint16_t port, struct rte_mempool *mbuf_pool)
 108 {
 109         int retval;
 110         uint16_t q;
 111         struct rte_eth_dev_info dev_info;
 112         uint16_t rx_rings, tx_rings = (uint16_t)rte_lcore_count();
 113         uint16_t rx_ring_size = RTE_TEST_RX_DESC_DEFAULT;
 114         uint16_t tx_ring_size = RTE_TEST_TX_DESC_DEFAULT;
 115         struct rte_eth_udp_tunnel tunnel_udp;
 116         struct rte_eth_rxconf *rxconf;
 117         struct rte_eth_txconf *txconf;
 118         struct vxlan_conf *pconf = &vxdev;
 119         struct rte_eth_conf local_port_conf = port_conf;
 120
 121         pconf->dst_port = udp_port;
 122
 123         rte_eth_dev_info_get(port, &dev_info);
 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         rte_eth_macaddr_get(port, &ports_eth_addr[port]);
 183         RTE_LOG(INFO, PORT, "Port %u MAC: %02"PRIx8" %02"PRIx8" %02"PRIx8
 184                         " %02"PRIx8" %02"PRIx8" %02"PRIx8"\n",
 185                         port,
 186                         ports_eth_addr[port].addr_bytes[0],
 187                         ports_eth_addr[port].addr_bytes[1],
 188                         ports_eth_addr[port].addr_bytes[2],
 189                         ports_eth_addr[port].addr_bytes[3],
 190                         ports_eth_addr[port].addr_bytes[4],
 191                         ports_eth_addr[port].addr_bytes[5]);
 192
 193         if (tso_segsz != 0) {
 194                 struct rte_eth_dev_info dev_info;
 195                 rte_eth_dev_info_get(port, &dev_info);
 196                 if ((dev_info.tx_offload_capa & DEV_TX_OFFLOAD_TCP_TSO) == 0)
 197                         RTE_LOG(WARNING, PORT,
 198                                 "hardware TSO offload is not supported\n");
 199         }
 200         return 0;
 201 }
 202
 203 static int
 204 vxlan_rx_process(struct rte_mbuf *pkt)
 205 {
 206         int ret = 0;
 207
 208         if (rx_decap)
 209                 ret = decapsulation(pkt);
 210
 211         return ret;
 212 }
 213
 214 static void
 215 vxlan_tx_process(uint8_t queue_id, struct rte_mbuf *pkt)
 216 {
 217         if (tx_encap)
 218                 encapsulation(pkt, queue_id);
 219
 220         return;
 221 }
 222
 223 /*
 224  * This function learns the MAC address of the device and set init
 225  * L2 header and L3 header info.
 226  */
 227 int
 228 vxlan_link(struct vhost_dev *vdev, struct rte_mbuf *m)
 229 {
 230         int i, ret;
 231         struct ether_hdr *pkt_hdr;
 232         uint64_t portid = vdev->vid;
 233         struct ipv4_hdr *ip;
 234
 235         struct rte_eth_tunnel_filter_conf tunnel_filter_conf;
 236
 237         if (unlikely(portid >= VXLAN_N_PORTS)) {
 238                 RTE_LOG(INFO, VHOST_DATA,
 239                         "(%d) WARNING: Not configuring device,"
 240                         "as already have %d ports for VXLAN.",
 241                         vdev->vid, VXLAN_N_PORTS);
 242                 return -1;
 243         }
 244
 245         /* Learn MAC address of guest device from packet */
 246         pkt_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 247         if (is_same_ether_addr(&(pkt_hdr->s_addr), &vdev->mac_address)) {
 248                 RTE_LOG(INFO, VHOST_DATA,
 249                         "(%d) WARNING: This device is using an existing"
 250                         " MAC address and has not been registered.\n",
 251                         vdev->vid);
 252                 return -1;
 253         }
 254
 255         for (i = 0; i < ETHER_ADDR_LEN; i++) {
 256                 vdev->mac_address.addr_bytes[i] =
 257                         vxdev.port[portid].vport_mac.addr_bytes[i] =
 258                         pkt_hdr->s_addr.addr_bytes[i];
 259                 vxdev.port[portid].peer_mac.addr_bytes[i] = peer_mac[i];
 260         }
 261
 262         memset(&tunnel_filter_conf, 0,
 263                 sizeof(struct rte_eth_tunnel_filter_conf));
 264
 265         ether_addr_copy(&ports_eth_addr[0], &tunnel_filter_conf.outer_mac);
 266         tunnel_filter_conf.filter_type = tep_filter_type[filter_idx];
 267
 268         /* inner MAC */
 269         ether_addr_copy(&vdev->mac_address, &tunnel_filter_conf.inner_mac);
 270
 271         tunnel_filter_conf.queue_id = vdev->rx_q;
 272         tunnel_filter_conf.tenant_id = tenant_id_conf[vdev->rx_q];
 273
 274         if (tep_filter_type[filter_idx] == RTE_TUNNEL_FILTER_IMAC_IVLAN_TENID)
 275                 tunnel_filter_conf.inner_vlan = INNER_VLAN_ID;
 276
 277         tunnel_filter_conf.tunnel_type = RTE_TUNNEL_TYPE_VXLAN;
 278
 279         ret = rte_eth_dev_filter_ctrl(ports[0],
 280                 RTE_ETH_FILTER_TUNNEL,
 281                 RTE_ETH_FILTER_ADD,
 282                 &tunnel_filter_conf);
 283         if (ret) {
 284                 RTE_LOG(ERR, VHOST_DATA,
 285                         "%d Failed to add device MAC address to cloud filter\n",
 286                 vdev->rx_q);
 287                 return -1;
 288         }
 289
 290         /* Print out inner MAC and VNI info. */
 291         RTE_LOG(INFO, VHOST_DATA,
 292                 "(%d) MAC_ADDRESS %02x:%02x:%02x:%02x:%02x:%02x and VNI %d registered\n",
 293                 vdev->rx_q,
 294                 vdev->mac_address.addr_bytes[0],
 295                 vdev->mac_address.addr_bytes[1],
 296                 vdev->mac_address.addr_bytes[2],
 297                 vdev->mac_address.addr_bytes[3],
 298                 vdev->mac_address.addr_bytes[4],
 299                 vdev->mac_address.addr_bytes[5],
 300                 tenant_id_conf[vdev->rx_q]);
 301
 302         vxdev.port[portid].vport_id = portid;
 303
 304         for (i = 0; i < 4; i++) {
 305                 /* Local VTEP IP */
 306                 vxdev.port_ip |= vxlan_multicast_ips[portid][i] << (8 * i);
 307                 /* Remote VTEP IP */
 308                 vxdev.port[portid].peer_ip |=
 309                         vxlan_overlay_ips[portid][i] << (8 * i);
 310         }
 311
 312         vxdev.out_key = tenant_id_conf[vdev->rx_q];
 313         ether_addr_copy(&vxdev.port[portid].peer_mac,
 314                         &app_l2_hdr[portid].d_addr);
 315         ether_addr_copy(&ports_eth_addr[0],
 316                         &app_l2_hdr[portid].s_addr);
 317         app_l2_hdr[portid].ether_type = rte_cpu_to_be_16(ETHER_TYPE_IPv4);
 318
 319         ip = &app_ip_hdr[portid];
 320         ip->version_ihl = IP_VHL_DEF;
 321         ip->type_of_service = 0;
 322         ip->total_length = 0;
 323         ip->packet_id = 0;
 324         ip->fragment_offset = IP_DN_FRAGMENT_FLAG;
 325         ip->time_to_live = IP_DEFTTL;
 326         ip->next_proto_id = IPPROTO_UDP;
 327         ip->hdr_checksum = 0;
 328         ip->src_addr = vxdev.port_ip;
 329         ip->dst_addr = vxdev.port[portid].peer_ip;
 330
 331         /* Set device as ready for RX. */
 332         vdev->ready = DEVICE_RX;
 333
 334         return 0;
 335 }
 336
 337 /**
 338  * Removes cloud filter. Ensures that nothing is adding buffers to the RX
 339  * queue before disabling RX on the device.
 340  */
 341 void
 342 vxlan_unlink(struct vhost_dev *vdev)
 343 {
 344         unsigned i = 0, rx_count;
 345         int ret;
 346         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 347         struct rte_eth_tunnel_filter_conf tunnel_filter_conf;
 348
 349         if (vdev->ready == DEVICE_RX) {
 350                 memset(&tunnel_filter_conf, 0,
 351                         sizeof(struct rte_eth_tunnel_filter_conf));
 352
 353                 ether_addr_copy(&ports_eth_addr[0], &tunnel_filter_conf.outer_mac);
 354                 ether_addr_copy(&vdev->mac_address, &tunnel_filter_conf.inner_mac);
 355                 tunnel_filter_conf.tenant_id = tenant_id_conf[vdev->rx_q];
 356                 tunnel_filter_conf.filter_type = tep_filter_type[filter_idx];
 357
 358                 if (tep_filter_type[filter_idx] ==
 359                         RTE_TUNNEL_FILTER_IMAC_IVLAN_TENID)
 360                         tunnel_filter_conf.inner_vlan = INNER_VLAN_ID;
 361
 362                 tunnel_filter_conf.queue_id = vdev->rx_q;
 363                 tunnel_filter_conf.tunnel_type = RTE_TUNNEL_TYPE_VXLAN;
 364
 365                 ret = rte_eth_dev_filter_ctrl(ports[0],
 366                                 RTE_ETH_FILTER_TUNNEL,
 367                                 RTE_ETH_FILTER_DELETE,
 368                                 &tunnel_filter_conf);
 369                 if (ret) {
 370                         RTE_LOG(ERR, VHOST_DATA,
 371                                 "%d Failed to add device MAC address to cloud filter\n",
 372                                 vdev->rx_q);
 373                         return;
 374                 }
 375                 for (i = 0; i < ETHER_ADDR_LEN; i++)
 376                         vdev->mac_address.addr_bytes[i] = 0;
 377
 378                 /* Clear out the receive buffers */
 379                 rx_count = rte_eth_rx_burst(ports[0],
 380                                 (uint16_t)vdev->rx_q,
 381                                 pkts_burst, MAX_PKT_BURST);
 382
 383                 while (rx_count) {
 384                         for (i = 0; i < rx_count; i++)
 385                                 rte_pktmbuf_free(pkts_burst[i]);
 386
 387                         rx_count = rte_eth_rx_burst(ports[0],
 388                                         (uint16_t)vdev->rx_q,
 389                                         pkts_burst, MAX_PKT_BURST);
 390                 }
 391                 vdev->ready = DEVICE_MAC_LEARNING;
 392         }
 393 }
 394
 395 /* Transmit packets after encapsulating */
 396 int
 397 vxlan_tx_pkts(uint16_t port_id, uint16_t queue_id,
 398                 struct rte_mbuf **tx_pkts, uint16_t nb_pkts) {
 399         int ret = 0;
 400         uint16_t i;
 401
 402         for (i = 0; i < nb_pkts; i++)
 403                 vxlan_tx_process(queue_id, tx_pkts[i]);
 404
 405         ret = rte_eth_tx_burst(port_id, queue_id, tx_pkts, nb_pkts);
 406
 407         return ret;
 408 }
 409
 410 /* Check for decapsulation and pass packets directly to VIRTIO device */
 411 int
 412 vxlan_rx_pkts(int vid, struct rte_mbuf **pkts_burst, uint32_t rx_count)
 413 {
 414         uint32_t i = 0;
 415         uint32_t count = 0;
 416         int ret;
 417         struct rte_mbuf *pkts_valid[rx_count];
 418
 419         for (i = 0; i < rx_count; i++) {
 420                 if (enable_stats) {
 421                         rte_atomic64_add(
 422                                 &dev_statistics[vid].rx_bad_ip_csum,
 423                                 (pkts_burst[i]->ol_flags & PKT_RX_IP_CKSUM_BAD)
 424                                 != 0);
 425                         rte_atomic64_add(
 426                                 &dev_statistics[vid].rx_bad_ip_csum,
 427                                 (pkts_burst[i]->ol_flags & PKT_RX_L4_CKSUM_BAD)
 428                                 != 0);
 429                 }
 430                 ret = vxlan_rx_process(pkts_burst[i]);
 431                 if (unlikely(ret < 0))
 432                         continue;
 433
 434                 pkts_valid[count] = pkts_burst[i];
 435                         count++;
 436         }
 437
 438         ret = rte_vhost_enqueue_burst(vid, VIRTIO_RXQ, pkts_valid, count);
 439         return ret;
 440 }