drivers/net/tap/rte_eth_tap.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2016-2017 Intel Corporation
   3  */
   4
   5 #include <rte_atomic.h>
   6 #include <rte_branch_prediction.h>
   7 #include <rte_byteorder.h>
   8 #include <rte_common.h>
   9 #include <rte_mbuf.h>
  10 #include <rte_ethdev_driver.h>
  11 #include <rte_ethdev_vdev.h>
  12 #include <rte_malloc.h>
  13 #include <rte_bus_vdev.h>
  14 #include <rte_kvargs.h>
  15 #include <rte_net.h>
  16 #include <rte_debug.h>
  17 #include <rte_ip.h>
  18
  19 #include <sys/types.h>
  20 #include <sys/stat.h>
  21 #include <sys/socket.h>
  22 #include <sys/ioctl.h>
  23 #include <sys/utsname.h>
  24 #include <sys/mman.h>
  25 #include <errno.h>
  26 #include <signal.h>
  27 #include <stdbool.h>
  28 #include <stdint.h>
  29 #include <sys/uio.h>
  30 #include <unistd.h>
  31 #include <arpa/inet.h>
  32 #include <net/if.h>
  33 #include <linux/if_tun.h>
  34 #include <linux/if_ether.h>
  35 #include <fcntl.h>
  36
  37 #include <rte_eth_tap.h>
  38 #include <tap_flow.h>
  39 #include <tap_netlink.h>
  40 #include <tap_tcmsgs.h>
  41
  42 /* Linux based path to the TUN device */
  43 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  44 #define DEFAULT_TAP_NAME        "dtap"
  45
  46 #define ETH_TAP_IFACE_ARG       "iface"
  47 #define ETH_TAP_REMOTE_ARG      "remote"
  48 #define ETH_TAP_MAC_ARG         "mac"
  49 #define ETH_TAP_MAC_FIXED       "fixed"
  50
  51 static struct rte_vdev_driver pmd_tap_drv;
  52
  53 static const char *valid_arguments[] = {
  54         ETH_TAP_IFACE_ARG,
  55         ETH_TAP_REMOTE_ARG,
  56         ETH_TAP_MAC_ARG,
  57         NULL
  58 };
  59
  60 static int tap_unit;
  61
  62 static volatile uint32_t tap_trigger;   /* Rx trigger */
  63
  64 static struct rte_eth_link pmd_link = {
  65         .link_speed = ETH_SPEED_NUM_10G,
  66         .link_duplex = ETH_LINK_FULL_DUPLEX,
  67         .link_status = ETH_LINK_DOWN,
  68         .link_autoneg = ETH_LINK_AUTONEG
  69 };
  70
  71 static void
  72 tap_trigger_cb(int sig __rte_unused)
  73 {
  74         /* Valid trigger values are nonzero */
  75         tap_trigger = (tap_trigger + 1) | 0x80000000;
  76 }
  77
  78 /* Specifies on what netdevices the ioctl should be applied */
  79 enum ioctl_mode {
  80         LOCAL_AND_REMOTE,
  81         LOCAL_ONLY,
  82         REMOTE_ONLY,
  83 };
  84
  85 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
  86
  87 /* Tun/Tap allocation routine
  88  *
  89  * name is the number of the interface to use, unless NULL to take the host
  90  * supplied name.
  91  */
  92 static int
  93 tun_alloc(struct pmd_internals *pmd)
  94 {
  95         struct ifreq ifr;
  96 #ifdef IFF_MULTI_QUEUE
  97         unsigned int features;
  98 #endif
  99         int fd;
 100
 101         memset(&ifr, 0, sizeof(struct ifreq));
 102
 103         /*
 104          * Do not set IFF_NO_PI as packet information header will be needed
 105          * to check if a received packet has been truncated.
 106          */
 107         ifr.ifr_flags = IFF_TAP;
 108         snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
 109
 110         RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
 111
 112         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 113         if (fd < 0) {
 114                 RTE_LOG(ERR, PMD, "Unable to create TAP interface\n");
 115                 goto error;
 116         }
 117
 118 #ifdef IFF_MULTI_QUEUE
 119         /* Grab the TUN features to verify we can work multi-queue */
 120         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 121                 RTE_LOG(ERR, PMD, "TAP unable to get TUN/TAP features\n");
 122                 goto error;
 123         }
 124         RTE_LOG(DEBUG, PMD, "  TAP Features %08x\n", features);
 125
 126         if (features & IFF_MULTI_QUEUE) {
 127                 RTE_LOG(DEBUG, PMD, "  Multi-queue support for %d queues\n",
 128                         RTE_PMD_TAP_MAX_QUEUES);
 129                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 130         } else
 131 #endif
 132         {
 133                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 134                 RTE_LOG(DEBUG, PMD, "  Single queue only support\n");
 135         }
 136
 137         /* Set the TUN/TAP configuration and set the name if needed */
 138         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 139                 RTE_LOG(WARNING, PMD,
 140                         "Unable to set TUNSETIFF for %s\n",
 141                         ifr.ifr_name);
 142                 perror("TUNSETIFF");
 143                 goto error;
 144         }
 145
 146         /* Always set the file descriptor to non-blocking */
 147         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 148                 RTE_LOG(WARNING, PMD,
 149                         "Unable to set %s to nonblocking\n",
 150                         ifr.ifr_name);
 151                 perror("F_SETFL, NONBLOCK");
 152                 goto error;
 153         }
 154
 155         /* Set up trigger to optimize empty Rx bursts */
 156         errno = 0;
 157         do {
 158                 struct sigaction sa;
 159                 int flags = fcntl(fd, F_GETFL);
 160
 161                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 162                         break;
 163                 if (sa.sa_handler != tap_trigger_cb) {
 164                         /*
 165                          * Make sure SIGIO is not already taken. This is done
 166                          * as late as possible to leave the application a
 167                          * chance to set up its own signal handler first.
 168                          */
 169                         if (sa.sa_handler != SIG_IGN &&
 170                             sa.sa_handler != SIG_DFL) {
 171                                 errno = EBUSY;
 172                                 break;
 173                         }
 174                         sa = (struct sigaction){
 175                                 .sa_flags = SA_RESTART,
 176                                 .sa_handler = tap_trigger_cb,
 177                         };
 178                         if (sigaction(SIGIO, &sa, NULL) == -1)
 179                                 break;
 180                 }
 181                 /* Enable SIGIO on file descriptor */
 182                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 183                 fcntl(fd, F_SETOWN, getpid());
 184         } while (0);
 185         if (errno) {
 186                 /* Disable trigger globally in case of error */
 187                 tap_trigger = 0;
 188                 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
 189                         strerror(errno));
 190         }
 191
 192         return fd;
 193
 194 error:
 195         if (fd > 0)
 196                 close(fd);
 197         return -1;
 198 }
 199
 200 static void
 201 tap_verify_csum(struct rte_mbuf *mbuf)
 202 {
 203         uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
 204         uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
 205         uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
 206         unsigned int l2_len = sizeof(struct ether_hdr);
 207         unsigned int l3_len;
 208         uint16_t cksum = 0;
 209         void *l3_hdr;
 210         void *l4_hdr;
 211
 212         if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
 213                 l2_len += 4;
 214         else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
 215                 l2_len += 8;
 216         /* Don't verify checksum for packets with discontinuous L2 header */
 217         if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
 218                      rte_pktmbuf_data_len(mbuf)))
 219                 return;
 220         l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
 221         if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 222                 struct ipv4_hdr *iph = l3_hdr;
 223
 224                 /* ihl contains the number of 4-byte words in the header */
 225                 l3_len = 4 * (iph->version_ihl & 0xf);
 226                 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
 227                         return;
 228
 229                 cksum = ~rte_raw_cksum(iph, l3_len);
 230                 mbuf->ol_flags |= cksum ?
 231                         PKT_RX_IP_CKSUM_BAD :
 232                         PKT_RX_IP_CKSUM_GOOD;
 233         } else if (l3 == RTE_PTYPE_L3_IPV6) {
 234                 l3_len = sizeof(struct ipv6_hdr);
 235         } else {
 236                 /* IPv6 extensions are not supported */
 237                 return;
 238         }
 239         if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
 240                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 241                 /* Don't verify checksum for multi-segment packets. */
 242                 if (mbuf->nb_segs > 1)
 243                         return;
 244                 if (l3 == RTE_PTYPE_L3_IPV4)
 245                         cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
 246                 else if (l3 == RTE_PTYPE_L3_IPV6)
 247                         cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
 248                 mbuf->ol_flags |= cksum ?
 249                         PKT_RX_L4_CKSUM_BAD :
 250                         PKT_RX_L4_CKSUM_GOOD;
 251         }
 252 }
 253
 254 static uint64_t
 255 tap_rx_offload_get_port_capa(void)
 256 {
 257         /*
 258          * In order to support legacy apps,
 259          * report capabilities also as port capabilities.
 260          */
 261         return DEV_RX_OFFLOAD_SCATTER |
 262                DEV_RX_OFFLOAD_IPV4_CKSUM |
 263                DEV_RX_OFFLOAD_UDP_CKSUM |
 264                DEV_RX_OFFLOAD_TCP_CKSUM;
 265 }
 266
 267 static uint64_t
 268 tap_rx_offload_get_queue_capa(void)
 269 {
 270         return DEV_RX_OFFLOAD_SCATTER |
 271                DEV_RX_OFFLOAD_IPV4_CKSUM |
 272                DEV_RX_OFFLOAD_UDP_CKSUM |
 273                DEV_RX_OFFLOAD_TCP_CKSUM;
 274 }
 275
 276 static bool
 277 tap_rxq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 278 {
 279         uint64_t port_offloads = dev->data->dev_conf.rxmode.offloads;
 280         uint64_t queue_supp_offloads = tap_rx_offload_get_queue_capa();
 281         uint64_t port_supp_offloads = tap_rx_offload_get_port_capa();
 282
 283         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 284             offloads)
 285                 return false;
 286         if ((port_offloads ^ offloads) & port_supp_offloads)
 287                 return false;
 288         return true;
 289 }
 290
 291 /* Callback to handle the rx burst of packets to the correct interface and
 292  * file descriptor(s) in a multi-queue setup.
 293  */
 294 static uint16_t
 295 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 296 {
 297         struct rx_queue *rxq = queue;
 298         uint16_t num_rx;
 299         unsigned long num_rx_bytes = 0;
 300         uint32_t trigger = tap_trigger;
 301
 302         if (trigger == rxq->trigger_seen)
 303                 return 0;
 304         if (trigger)
 305                 rxq->trigger_seen = trigger;
 306         rte_compiler_barrier();
 307         for (num_rx = 0; num_rx < nb_pkts; ) {
 308                 struct rte_mbuf *mbuf = rxq->pool;
 309                 struct rte_mbuf *seg = NULL;
 310                 struct rte_mbuf *new_tail = NULL;
 311                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 312                 int len;
 313
 314                 len = readv(rxq->fd, *rxq->iovecs,
 315                             1 +
 316                             (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
 317                              rxq->nb_rx_desc : 1));
 318                 if (len < (int)sizeof(struct tun_pi))
 319                         break;
 320
 321                 /* Packet couldn't fit in the provided mbuf */
 322                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 323                         rxq->stats.ierrors++;
 324                         continue;
 325                 }
 326
 327                 len -= sizeof(struct tun_pi);
 328
 329                 mbuf->pkt_len = len;
 330                 mbuf->port = rxq->in_port;
 331                 while (1) {
 332                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 333
 334                         if (unlikely(!buf)) {
 335                                 rxq->stats.rx_nombuf++;
 336                                 /* No new buf has been allocated: do nothing */
 337                                 if (!new_tail || !seg)
 338                                         goto end;
 339
 340                                 seg->next = NULL;
 341                                 rte_pktmbuf_free(mbuf);
 342
 343                                 goto end;
 344                         }
 345                         seg = seg ? seg->next : mbuf;
 346                         if (rxq->pool == mbuf)
 347                                 rxq->pool = buf;
 348                         if (new_tail)
 349                                 new_tail->next = buf;
 350                         new_tail = buf;
 351                         new_tail->next = seg->next;
 352
 353                         /* iovecs[0] is reserved for packet info (pi) */
 354                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 355                                 buf->buf_len - data_off;
 356                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 357                                 (char *)buf->buf_addr + data_off;
 358
 359                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 360                         seg->data_off = data_off;
 361
 362                         len -= seg->data_len;
 363                         if (len <= 0)
 364                                 break;
 365                         mbuf->nb_segs++;
 366                         /* First segment has headroom, not the others */
 367                         data_off = 0;
 368                 }
 369                 seg->next = NULL;
 370                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 371                                                       RTE_PTYPE_ALL_MASK);
 372                 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
 373                         tap_verify_csum(mbuf);
 374
 375                 /* account for the receive frame */
 376                 bufs[num_rx++] = mbuf;
 377                 num_rx_bytes += mbuf->pkt_len;
 378         }
 379 end:
 380         rxq->stats.ipackets += num_rx;
 381         rxq->stats.ibytes += num_rx_bytes;
 382
 383         return num_rx;
 384 }
 385
 386 static uint64_t
 387 tap_tx_offload_get_port_capa(void)
 388 {
 389         /*
 390          * In order to support legacy apps,
 391          * report capabilities also as port capabilities.
 392          */
 393         return DEV_TX_OFFLOAD_MULTI_SEGS |
 394                DEV_TX_OFFLOAD_IPV4_CKSUM |
 395                DEV_TX_OFFLOAD_UDP_CKSUM |
 396                DEV_TX_OFFLOAD_TCP_CKSUM;
 397 }
 398
 399 static uint64_t
 400 tap_tx_offload_get_queue_capa(void)
 401 {
 402         return DEV_TX_OFFLOAD_MULTI_SEGS |
 403                DEV_TX_OFFLOAD_IPV4_CKSUM |
 404                DEV_TX_OFFLOAD_UDP_CKSUM |
 405                DEV_TX_OFFLOAD_TCP_CKSUM;
 406 }
 407
 408 static bool
 409 tap_txq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 410 {
 411         uint64_t port_offloads = dev->data->dev_conf.txmode.offloads;
 412         uint64_t queue_supp_offloads = tap_tx_offload_get_queue_capa();
 413         uint64_t port_supp_offloads = tap_tx_offload_get_port_capa();
 414
 415         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 416             offloads)
 417                 return false;
 418         /* Verify we have no conflict with port offloads */
 419         if ((port_offloads ^ offloads) & port_supp_offloads)
 420                 return false;
 421         return true;
 422 }
 423
 424 static void
 425 tap_tx_offload(char *packet, uint64_t ol_flags, unsigned int l2_len,
 426                unsigned int l3_len)
 427 {
 428         void *l3_hdr = packet + l2_len;
 429
 430         if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
 431                 struct ipv4_hdr *iph = l3_hdr;
 432                 uint16_t cksum;
 433
 434                 iph->hdr_checksum = 0;
 435                 cksum = rte_raw_cksum(iph, l3_len);
 436                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 437         }
 438         if (ol_flags & PKT_TX_L4_MASK) {
 439                 uint16_t l4_len;
 440                 uint32_t cksum;
 441                 uint16_t *l4_cksum;
 442                 void *l4_hdr;
 443
 444                 l4_hdr = packet + l2_len + l3_len;
 445                 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
 446                         l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
 447                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 448                         l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
 449                 else
 450                         return;
 451                 *l4_cksum = 0;
 452                 if (ol_flags & PKT_TX_IPV4) {
 453                         struct ipv4_hdr *iph = l3_hdr;
 454
 455                         l4_len = rte_be_to_cpu_16(iph->total_length) - l3_len;
 456                         cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 457                 } else {
 458                         struct ipv6_hdr *ip6h = l3_hdr;
 459
 460                         /* payload_len does not include ext headers */
 461                         l4_len = rte_be_to_cpu_16(ip6h->payload_len) -
 462                                 l3_len + sizeof(struct ipv6_hdr);
 463                         cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 464                 }
 465                 cksum += rte_raw_cksum(l4_hdr, l4_len);
 466                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 467                 cksum = (~cksum) & 0xffff;
 468                 if (cksum == 0)
 469                         cksum = 0xffff;
 470                 *l4_cksum = cksum;
 471         }
 472 }
 473
 474 /* Callback to handle sending packets from the tap interface
 475  */
 476 static uint16_t
 477 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 478 {
 479         struct tx_queue *txq = queue;
 480         uint16_t num_tx = 0;
 481         unsigned long num_tx_bytes = 0;
 482         uint32_t max_size;
 483         int i;
 484
 485         if (unlikely(nb_pkts == 0))
 486                 return 0;
 487
 488         max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
 489         for (i = 0; i < nb_pkts; i++) {
 490                 struct rte_mbuf *mbuf = bufs[num_tx];
 491                 struct iovec iovecs[mbuf->nb_segs + 1];
 492                 struct tun_pi pi = { .flags = 0 };
 493                 struct rte_mbuf *seg = mbuf;
 494                 char m_copy[mbuf->data_len];
 495                 int n;
 496                 int j;
 497
 498                 /* stats.errs will be incremented */
 499                 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
 500                         break;
 501
 502                 iovecs[0].iov_base = &pi;
 503                 iovecs[0].iov_len = sizeof(pi);
 504                 for (j = 1; j <= mbuf->nb_segs; j++) {
 505                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 506                         iovecs[j].iov_base =
 507                                 rte_pktmbuf_mtod(seg, void *);
 508                         seg = seg->next;
 509                 }
 510                 if (txq->csum &&
 511                     ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 512                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 513                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
 514                         /* Support only packets with all data in the same seg */
 515                         if (mbuf->nb_segs > 1)
 516                                 break;
 517                         /* To change checksums, work on a copy of data. */
 518                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 519                                    rte_pktmbuf_data_len(mbuf));
 520                         tap_tx_offload(m_copy, mbuf->ol_flags,
 521                                        mbuf->l2_len, mbuf->l3_len);
 522                         iovecs[1].iov_base = m_copy;
 523                 }
 524                 /* copy the tx frame data */
 525                 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
 526                 if (n <= 0)
 527                         break;
 528
 529                 num_tx++;
 530                 num_tx_bytes += mbuf->pkt_len;
 531                 rte_pktmbuf_free(mbuf);
 532         }
 533
 534         txq->stats.opackets += num_tx;
 535         txq->stats.errs += nb_pkts - num_tx;
 536         txq->stats.obytes += num_tx_bytes;
 537
 538         return num_tx;
 539 }
 540
 541 static const char *
 542 tap_ioctl_req2str(unsigned long request)
 543 {
 544         switch (request) {
 545         case SIOCSIFFLAGS:
 546                 return "SIOCSIFFLAGS";
 547         case SIOCGIFFLAGS:
 548                 return "SIOCGIFFLAGS";
 549         case SIOCGIFHWADDR:
 550                 return "SIOCGIFHWADDR";
 551         case SIOCSIFHWADDR:
 552                 return "SIOCSIFHWADDR";
 553         case SIOCSIFMTU:
 554                 return "SIOCSIFMTU";
 555         }
 556         return "UNKNOWN";
 557 }
 558
 559 static int
 560 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 561           struct ifreq *ifr, int set, enum ioctl_mode mode)
 562 {
 563         short req_flags = ifr->ifr_flags;
 564         int remote = pmd->remote_if_index &&
 565                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 566
 567         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 568                 return 0;
 569         /*
 570          * If there is a remote netdevice, apply ioctl on it, then apply it on
 571          * the tap netdevice.
 572          */
 573 apply:
 574         if (remote)
 575                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
 576         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 577                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 578         switch (request) {
 579         case SIOCSIFFLAGS:
 580                 /* fetch current flags to leave other flags untouched */
 581                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 582                         goto error;
 583                 if (set)
 584                         ifr->ifr_flags |= req_flags;
 585                 else
 586                         ifr->ifr_flags &= ~req_flags;
 587                 break;
 588         case SIOCGIFFLAGS:
 589         case SIOCGIFHWADDR:
 590         case SIOCSIFHWADDR:
 591         case SIOCSIFMTU:
 592                 break;
 593         default:
 594                 RTE_ASSERT(!"unsupported request type: must not happen");
 595         }
 596         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 597                 goto error;
 598         if (remote-- && mode == LOCAL_AND_REMOTE)
 599                 goto apply;
 600         return 0;
 601
 602 error:
 603         RTE_LOG(DEBUG, PMD, "%s: %s(%s) failed: %s(%d)\n", ifr->ifr_name,
 604                 __func__, tap_ioctl_req2str(request), strerror(errno), errno);
 605         return -errno;
 606 }
 607
 608 static int
 609 tap_link_set_down(struct rte_eth_dev *dev)
 610 {
 611         struct pmd_internals *pmd = dev->data->dev_private;
 612         struct ifreq ifr = { .ifr_flags = IFF_UP };
 613
 614         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 615         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 616 }
 617
 618 static int
 619 tap_link_set_up(struct rte_eth_dev *dev)
 620 {
 621         struct pmd_internals *pmd = dev->data->dev_private;
 622         struct ifreq ifr = { .ifr_flags = IFF_UP };
 623
 624         dev->data->dev_link.link_status = ETH_LINK_UP;
 625         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 626 }
 627
 628 static int
 629 tap_dev_start(struct rte_eth_dev *dev)
 630 {
 631         int err;
 632
 633         err = tap_intr_handle_set(dev, 1);
 634         if (err)
 635                 return err;
 636         return tap_link_set_up(dev);
 637 }
 638
 639 /* This function gets called when the current port gets stopped.
 640  */
 641 static void
 642 tap_dev_stop(struct rte_eth_dev *dev)
 643 {
 644         tap_intr_handle_set(dev, 0);
 645         tap_link_set_down(dev);
 646 }
 647
 648 static int
 649 tap_dev_configure(struct rte_eth_dev *dev)
 650 {
 651         uint64_t supp_tx_offloads = tap_tx_offload_get_port_capa();
 652         uint64_t tx_offloads = dev->data->dev_conf.txmode.offloads;
 653
 654         if ((tx_offloads & supp_tx_offloads) != tx_offloads) {
 655                 rte_errno = ENOTSUP;
 656                 RTE_LOG(ERR, PMD,
 657                         "Some Tx offloads are not supported "
 658                         "requested 0x%" PRIx64 " supported 0x%" PRIx64 "\n",
 659                         tx_offloads, supp_tx_offloads);
 660                 return -rte_errno;
 661         }
 662         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 663                 RTE_LOG(ERR, PMD,
 664                         "%s: number of rx queues %d exceeds max num of queues %d\n",
 665                         dev->device->name,
 666                         dev->data->nb_rx_queues,
 667                         RTE_PMD_TAP_MAX_QUEUES);
 668                 return -1;
 669         }
 670         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 671                 RTE_LOG(ERR, PMD,
 672                         "%s: number of tx queues %d exceeds max num of queues %d\n",
 673                         dev->device->name,
 674                         dev->data->nb_tx_queues,
 675                         RTE_PMD_TAP_MAX_QUEUES);
 676                 return -1;
 677         }
 678
 679         RTE_LOG(INFO, PMD, "%s: %p: TX configured queues number: %u\n",
 680              dev->device->name, (void *)dev, dev->data->nb_tx_queues);
 681
 682         RTE_LOG(INFO, PMD, "%s: %p: RX configured queues number: %u\n",
 683              dev->device->name, (void *)dev, dev->data->nb_rx_queues);
 684
 685         return 0;
 686 }
 687
 688 static uint32_t
 689 tap_dev_speed_capa(void)
 690 {
 691         uint32_t speed = pmd_link.link_speed;
 692         uint32_t capa = 0;
 693
 694         if (speed >= ETH_SPEED_NUM_10M)
 695                 capa |= ETH_LINK_SPEED_10M;
 696         if (speed >= ETH_SPEED_NUM_100M)
 697                 capa |= ETH_LINK_SPEED_100M;
 698         if (speed >= ETH_SPEED_NUM_1G)
 699                 capa |= ETH_LINK_SPEED_1G;
 700         if (speed >= ETH_SPEED_NUM_5G)
 701                 capa |= ETH_LINK_SPEED_2_5G;
 702         if (speed >= ETH_SPEED_NUM_5G)
 703                 capa |= ETH_LINK_SPEED_5G;
 704         if (speed >= ETH_SPEED_NUM_10G)
 705                 capa |= ETH_LINK_SPEED_10G;
 706         if (speed >= ETH_SPEED_NUM_20G)
 707                 capa |= ETH_LINK_SPEED_20G;
 708         if (speed >= ETH_SPEED_NUM_25G)
 709                 capa |= ETH_LINK_SPEED_25G;
 710         if (speed >= ETH_SPEED_NUM_40G)
 711                 capa |= ETH_LINK_SPEED_40G;
 712         if (speed >= ETH_SPEED_NUM_50G)
 713                 capa |= ETH_LINK_SPEED_50G;
 714         if (speed >= ETH_SPEED_NUM_56G)
 715                 capa |= ETH_LINK_SPEED_56G;
 716         if (speed >= ETH_SPEED_NUM_100G)
 717                 capa |= ETH_LINK_SPEED_100G;
 718
 719         return capa;
 720 }
 721
 722 static void
 723 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 724 {
 725         struct pmd_internals *internals = dev->data->dev_private;
 726
 727         dev_info->if_index = internals->if_index;
 728         dev_info->max_mac_addrs = 1;
 729         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 730         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
 731         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
 732         dev_info->min_rx_bufsize = 0;
 733         dev_info->pci_dev = NULL;
 734         dev_info->speed_capa = tap_dev_speed_capa();
 735         dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
 736         dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
 737                                     dev_info->rx_queue_offload_capa;
 738         dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
 739         dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
 740                                     dev_info->tx_queue_offload_capa;
 741 }
 742
 743 static int
 744 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 745 {
 746         unsigned int i, imax;
 747         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 748         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 749         unsigned long rx_nombuf = 0, ierrors = 0;
 750         const struct pmd_internals *pmd = dev->data->dev_private;
 751
 752         /* rx queue statistics */
 753         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 754                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 755         for (i = 0; i < imax; i++) {
 756                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 757                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 758                 rx_total += tap_stats->q_ipackets[i];
 759                 rx_bytes_total += tap_stats->q_ibytes[i];
 760                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
 761                 ierrors += pmd->rxq[i].stats.ierrors;
 762         }
 763
 764         /* tx queue statistics */
 765         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 766                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 767
 768         for (i = 0; i < imax; i++) {
 769                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 770                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 771                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 772                 tx_total += tap_stats->q_opackets[i];
 773                 tx_err_total += tap_stats->q_errors[i];
 774                 tx_bytes_total += tap_stats->q_obytes[i];
 775         }
 776
 777         tap_stats->ipackets = rx_total;
 778         tap_stats->ibytes = rx_bytes_total;
 779         tap_stats->ierrors = ierrors;
 780         tap_stats->rx_nombuf = rx_nombuf;
 781         tap_stats->opackets = tx_total;
 782         tap_stats->oerrors = tx_err_total;
 783         tap_stats->obytes = tx_bytes_total;
 784         return 0;
 785 }
 786
 787 static void
 788 tap_stats_reset(struct rte_eth_dev *dev)
 789 {
 790         int i;
 791         struct pmd_internals *pmd = dev->data->dev_private;
 792
 793         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 794                 pmd->rxq[i].stats.ipackets = 0;
 795                 pmd->rxq[i].stats.ibytes = 0;
 796                 pmd->rxq[i].stats.ierrors = 0;
 797                 pmd->rxq[i].stats.rx_nombuf = 0;
 798
 799                 pmd->txq[i].stats.opackets = 0;
 800                 pmd->txq[i].stats.errs = 0;
 801                 pmd->txq[i].stats.obytes = 0;
 802         }
 803 }
 804
 805 static void
 806 tap_dev_close(struct rte_eth_dev *dev)
 807 {
 808         int i;
 809         struct pmd_internals *internals = dev->data->dev_private;
 810
 811         tap_link_set_down(dev);
 812         tap_flow_flush(dev, NULL);
 813         tap_flow_implicit_flush(internals, NULL);
 814
 815         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 816                 if (internals->rxq[i].fd != -1) {
 817                         close(internals->rxq[i].fd);
 818                         internals->rxq[i].fd = -1;
 819                 }
 820                 if (internals->txq[i].fd != -1) {
 821                         close(internals->txq[i].fd);
 822                         internals->txq[i].fd = -1;
 823                 }
 824         }
 825
 826         if (internals->remote_if_index) {
 827                 /* Restore initial remote state */
 828                 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
 829                                 &internals->remote_initial_flags);
 830         }
 831 }
 832
 833 static void
 834 tap_rx_queue_release(void *queue)
 835 {
 836         struct rx_queue *rxq = queue;
 837
 838         if (rxq && (rxq->fd > 0)) {
 839                 close(rxq->fd);
 840                 rxq->fd = -1;
 841                 rte_pktmbuf_free(rxq->pool);
 842                 rte_free(rxq->iovecs);
 843                 rxq->pool = NULL;
 844                 rxq->iovecs = NULL;
 845         }
 846 }
 847
 848 static void
 849 tap_tx_queue_release(void *queue)
 850 {
 851         struct tx_queue *txq = queue;
 852
 853         if (txq && (txq->fd > 0)) {
 854                 close(txq->fd);
 855                 txq->fd = -1;
 856         }
 857 }
 858
 859 static int
 860 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
 861 {
 862         struct rte_eth_link *dev_link = &dev->data->dev_link;
 863         struct pmd_internals *pmd = dev->data->dev_private;
 864         struct ifreq ifr = { .ifr_flags = 0 };
 865
 866         if (pmd->remote_if_index) {
 867                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
 868                 if (!(ifr.ifr_flags & IFF_UP) ||
 869                     !(ifr.ifr_flags & IFF_RUNNING)) {
 870                         dev_link->link_status = ETH_LINK_DOWN;
 871                         return 0;
 872                 }
 873         }
 874         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
 875         dev_link->link_status =
 876                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
 877                  ETH_LINK_UP :
 878                  ETH_LINK_DOWN);
 879         return 0;
 880 }
 881
 882 static void
 883 tap_promisc_enable(struct rte_eth_dev *dev)
 884 {
 885         struct pmd_internals *pmd = dev->data->dev_private;
 886         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 887
 888         dev->data->promiscuous = 1;
 889         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 890         if (pmd->remote_if_index && !pmd->flow_isolate)
 891                 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
 892 }
 893
 894 static void
 895 tap_promisc_disable(struct rte_eth_dev *dev)
 896 {
 897         struct pmd_internals *pmd = dev->data->dev_private;
 898         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 899
 900         dev->data->promiscuous = 0;
 901         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 902         if (pmd->remote_if_index && !pmd->flow_isolate)
 903                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
 904 }
 905
 906 static void
 907 tap_allmulti_enable(struct rte_eth_dev *dev)
 908 {
 909         struct pmd_internals *pmd = dev->data->dev_private;
 910         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 911
 912         dev->data->all_multicast = 1;
 913         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 914         if (pmd->remote_if_index && !pmd->flow_isolate)
 915                 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
 916 }
 917
 918 static void
 919 tap_allmulti_disable(struct rte_eth_dev *dev)
 920 {
 921         struct pmd_internals *pmd = dev->data->dev_private;
 922         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 923
 924         dev->data->all_multicast = 0;
 925         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 926         if (pmd->remote_if_index && !pmd->flow_isolate)
 927                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
 928 }
 929
 930 static void
 931 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 932 {
 933         struct pmd_internals *pmd = dev->data->dev_private;
 934         enum ioctl_mode mode = LOCAL_ONLY;
 935         struct ifreq ifr;
 936
 937         if (is_zero_ether_addr(mac_addr)) {
 938                 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
 939                         dev->device->name);
 940                 return;
 941         }
 942         /* Check the actual current MAC address on the tap netdevice */
 943         if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
 944                 return;
 945         if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 946                                mac_addr))
 947                 return;
 948         /* Check the current MAC address on the remote */
 949         if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0)
 950                 return;
 951         if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 952                                mac_addr))
 953                 mode = LOCAL_AND_REMOTE;
 954         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
 955         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
 956         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode) < 0)
 957                 return;
 958         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
 959         if (pmd->remote_if_index && !pmd->flow_isolate) {
 960                 /* Replace MAC redirection rule after a MAC change */
 961                 if (tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC) < 0) {
 962                         RTE_LOG(ERR, PMD,
 963                                 "%s: Couldn't delete MAC redirection rule\n",
 964                                 dev->device->name);
 965                         return;
 966                 }
 967                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0)
 968                         RTE_LOG(ERR, PMD,
 969                                 "%s: Couldn't add MAC redirection rule\n",
 970                                 dev->device->name);
 971         }
 972 }
 973
 974 static int
 975 tap_setup_queue(struct rte_eth_dev *dev,
 976                 struct pmd_internals *internals,
 977                 uint16_t qid,
 978                 int is_rx)
 979 {
 980         int *fd;
 981         int *other_fd;
 982         const char *dir;
 983         struct pmd_internals *pmd = dev->data->dev_private;
 984         struct rx_queue *rx = &internals->rxq[qid];
 985         struct tx_queue *tx = &internals->txq[qid];
 986
 987         if (is_rx) {
 988                 fd = &rx->fd;
 989                 other_fd = &tx->fd;
 990                 dir = "rx";
 991         } else {
 992                 fd = &tx->fd;
 993                 other_fd = &rx->fd;
 994                 dir = "tx";
 995         }
 996         if (*fd != -1) {
 997                 /* fd for this queue already exists */
 998                 RTE_LOG(DEBUG, PMD, "%s: fd %d for %s queue qid %d exists\n",
 999                         pmd->name, *fd, dir, qid);
1000         } else if (*other_fd != -1) {
1001                 /* Only other_fd exists. dup it */
1002                 *fd = dup(*other_fd);
1003                 if (*fd < 0) {
1004                         *fd = -1;
1005                         RTE_LOG(ERR, PMD, "%s: dup() failed.\n",
1006                                 pmd->name);
1007                         return -1;
1008                 }
1009                 RTE_LOG(DEBUG, PMD, "%s: dup fd %d for %s queue qid %d (%d)\n",
1010                         pmd->name, *other_fd, dir, qid, *fd);
1011         } else {
1012                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1013                 *fd = tun_alloc(pmd);
1014                 if (*fd < 0) {
1015                         *fd = -1; /* restore original value */
1016                         RTE_LOG(ERR, PMD, "%s: tun_alloc() failed.\n",
1017                                 pmd->name);
1018                         return -1;
1019                 }
1020                 RTE_LOG(DEBUG, PMD, "%s: add %s queue for qid %d fd %d\n",
1021                         pmd->name, dir, qid, *fd);
1022         }
1023
1024         tx->mtu = &dev->data->mtu;
1025         rx->rxmode = &dev->data->dev_conf.rxmode;
1026
1027         return *fd;
1028 }
1029
1030 static int
1031 tap_rx_queue_setup(struct rte_eth_dev *dev,
1032                    uint16_t rx_queue_id,
1033                    uint16_t nb_rx_desc,
1034                    unsigned int socket_id,
1035                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1036                    struct rte_mempool *mp)
1037 {
1038         struct pmd_internals *internals = dev->data->dev_private;
1039         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1040         struct rte_mbuf **tmp = &rxq->pool;
1041         long iov_max = sysconf(_SC_IOV_MAX);
1042         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1043         struct iovec (*iovecs)[nb_desc + 1];
1044         int data_off = RTE_PKTMBUF_HEADROOM;
1045         int ret = 0;
1046         int fd;
1047         int i;
1048
1049         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1050                 RTE_LOG(WARNING, PMD,
1051                         "nb_rx_queues %d too small or mempool NULL\n",
1052                         dev->data->nb_rx_queues);
1053                 return -1;
1054         }
1055
1056         /* Verify application offloads are valid for our port and queue. */
1057         if (!tap_rxq_are_offloads_valid(dev, rx_conf->offloads)) {
1058                 rte_errno = ENOTSUP;
1059                 RTE_LOG(ERR, PMD,
1060                         "%p: Rx queue offloads 0x%" PRIx64
1061                         " don't match port offloads 0x%" PRIx64
1062                         " or supported offloads 0x%" PRIx64 "\n",
1063                         (void *)dev, rx_conf->offloads,
1064                         dev->data->dev_conf.rxmode.offloads,
1065                         (tap_rx_offload_get_port_capa() |
1066                          tap_rx_offload_get_queue_capa()));
1067                 return -rte_errno;
1068         }
1069         rxq->mp = mp;
1070         rxq->trigger_seen = 1; /* force initial burst */
1071         rxq->in_port = dev->data->port_id;
1072         rxq->nb_rx_desc = nb_desc;
1073         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1074                                     socket_id);
1075         if (!iovecs) {
1076                 RTE_LOG(WARNING, PMD,
1077                         "%s: Couldn't allocate %d RX descriptors\n",
1078                         dev->device->name, nb_desc);
1079                 return -ENOMEM;
1080         }
1081         rxq->iovecs = iovecs;
1082
1083         dev->data->rx_queues[rx_queue_id] = rxq;
1084         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1085         if (fd == -1) {
1086                 ret = fd;
1087                 goto error;
1088         }
1089
1090         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1091         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1092
1093         for (i = 1; i <= nb_desc; i++) {
1094                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1095                 if (!*tmp) {
1096                         RTE_LOG(WARNING, PMD,
1097                                 "%s: couldn't allocate memory for queue %d\n",
1098                                 dev->device->name, rx_queue_id);
1099                         ret = -ENOMEM;
1100                         goto error;
1101                 }
1102                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1103                 (*rxq->iovecs)[i].iov_base =
1104                         (char *)(*tmp)->buf_addr + data_off;
1105                 data_off = 0;
1106                 tmp = &(*tmp)->next;
1107         }
1108
1109         RTE_LOG(DEBUG, PMD, "  RX TAP device name %s, qid %d on fd %d\n",
1110                 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
1111
1112         return 0;
1113
1114 error:
1115         rte_pktmbuf_free(rxq->pool);
1116         rxq->pool = NULL;
1117         rte_free(rxq->iovecs);
1118         rxq->iovecs = NULL;
1119         return ret;
1120 }
1121
1122 static int
1123 tap_tx_queue_setup(struct rte_eth_dev *dev,
1124                    uint16_t tx_queue_id,
1125                    uint16_t nb_tx_desc __rte_unused,
1126                    unsigned int socket_id __rte_unused,
1127                    const struct rte_eth_txconf *tx_conf)
1128 {
1129         struct pmd_internals *internals = dev->data->dev_private;
1130         struct tx_queue *txq;
1131         int ret;
1132
1133         if (tx_queue_id >= dev->data->nb_tx_queues)
1134                 return -1;
1135         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1136         txq = dev->data->tx_queues[tx_queue_id];
1137         /*
1138          * Don't verify port offloads for application which
1139          * use the old API.
1140          */
1141         if (tx_conf != NULL &&
1142             !!(tx_conf->txq_flags & ETH_TXQ_FLAGS_IGNORE)) {
1143                 if (tap_txq_are_offloads_valid(dev, tx_conf->offloads)) {
1144                         txq->csum = !!(tx_conf->offloads &
1145                                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
1146                                          DEV_TX_OFFLOAD_UDP_CKSUM |
1147                                          DEV_TX_OFFLOAD_TCP_CKSUM));
1148                 } else {
1149                         rte_errno = ENOTSUP;
1150                         RTE_LOG(ERR, PMD,
1151                                 "%p: Tx queue offloads 0x%" PRIx64
1152                                 " don't match port offloads 0x%" PRIx64
1153                                 " or supported offloads 0x%" PRIx64,
1154                                 (void *)dev, tx_conf->offloads,
1155                                 dev->data->dev_conf.txmode.offloads,
1156                                 tap_tx_offload_get_port_capa());
1157                         return -rte_errno;
1158                 }
1159         }
1160         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1161         if (ret == -1)
1162                 return -1;
1163         RTE_LOG(DEBUG, PMD,
1164                 "  TX TAP device name %s, qid %d on fd %d csum %s\n",
1165                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd,
1166                 txq->csum ? "on" : "off");
1167
1168         return 0;
1169 }
1170
1171 static int
1172 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1173 {
1174         struct pmd_internals *pmd = dev->data->dev_private;
1175         struct ifreq ifr = { .ifr_mtu = mtu };
1176         int err = 0;
1177
1178         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1179         if (!err)
1180                 dev->data->mtu = mtu;
1181
1182         return err;
1183 }
1184
1185 static int
1186 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1187                      struct ether_addr *mc_addr_set __rte_unused,
1188                      uint32_t nb_mc_addr __rte_unused)
1189 {
1190         /*
1191          * Nothing to do actually: the tap has no filtering whatsoever, every
1192          * packet is received.
1193          */
1194         return 0;
1195 }
1196
1197 static int
1198 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1199 {
1200         struct rte_eth_dev *dev = arg;
1201         struct pmd_internals *pmd = dev->data->dev_private;
1202         struct ifinfomsg *info = NLMSG_DATA(nh);
1203
1204         if (nh->nlmsg_type != RTM_NEWLINK ||
1205             (info->ifi_index != pmd->if_index &&
1206              info->ifi_index != pmd->remote_if_index))
1207                 return 0;
1208         return tap_link_update(dev, 0);
1209 }
1210
1211 static void
1212 tap_dev_intr_handler(void *cb_arg)
1213 {
1214         struct rte_eth_dev *dev = cb_arg;
1215         struct pmd_internals *pmd = dev->data->dev_private;
1216
1217         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1218 }
1219
1220 static int
1221 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1222 {
1223         struct pmd_internals *pmd = dev->data->dev_private;
1224
1225         /* In any case, disable interrupt if the conf is no longer there. */
1226         if (!dev->data->dev_conf.intr_conf.lsc) {
1227                 if (pmd->intr_handle.fd != -1) {
1228                         tap_nl_final(pmd->intr_handle.fd);
1229                         rte_intr_callback_unregister(&pmd->intr_handle,
1230                                 tap_dev_intr_handler, dev);
1231                 }
1232                 return 0;
1233         }
1234         if (set) {
1235                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1236                 if (unlikely(pmd->intr_handle.fd == -1))
1237                         return -EBADF;
1238                 return rte_intr_callback_register(
1239                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1240         }
1241         tap_nl_final(pmd->intr_handle.fd);
1242         return rte_intr_callback_unregister(&pmd->intr_handle,
1243                                             tap_dev_intr_handler, dev);
1244 }
1245
1246 static int
1247 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1248 {
1249         int err;
1250
1251         err = tap_lsc_intr_handle_set(dev, set);
1252         if (err)
1253                 return err;
1254         err = tap_rx_intr_vec_set(dev, set);
1255         if (err && set)
1256                 tap_lsc_intr_handle_set(dev, 0);
1257         return err;
1258 }
1259
1260 static const uint32_t*
1261 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1262 {
1263         static const uint32_t ptypes[] = {
1264                 RTE_PTYPE_INNER_L2_ETHER,
1265                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1266                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1267                 RTE_PTYPE_INNER_L3_IPV4,
1268                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1269                 RTE_PTYPE_INNER_L3_IPV6,
1270                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1271                 RTE_PTYPE_INNER_L4_FRAG,
1272                 RTE_PTYPE_INNER_L4_UDP,
1273                 RTE_PTYPE_INNER_L4_TCP,
1274                 RTE_PTYPE_INNER_L4_SCTP,
1275                 RTE_PTYPE_L2_ETHER,
1276                 RTE_PTYPE_L2_ETHER_VLAN,
1277                 RTE_PTYPE_L2_ETHER_QINQ,
1278                 RTE_PTYPE_L3_IPV4,
1279                 RTE_PTYPE_L3_IPV4_EXT,
1280                 RTE_PTYPE_L3_IPV6_EXT,
1281                 RTE_PTYPE_L3_IPV6,
1282                 RTE_PTYPE_L4_FRAG,
1283                 RTE_PTYPE_L4_UDP,
1284                 RTE_PTYPE_L4_TCP,
1285                 RTE_PTYPE_L4_SCTP,
1286         };
1287
1288         return ptypes;
1289 }
1290
1291 static int
1292 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1293                   struct rte_eth_fc_conf *fc_conf)
1294 {
1295         fc_conf->mode = RTE_FC_NONE;
1296         return 0;
1297 }
1298
1299 static int
1300 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1301                   struct rte_eth_fc_conf *fc_conf)
1302 {
1303         if (fc_conf->mode != RTE_FC_NONE)
1304                 return -ENOTSUP;
1305         return 0;
1306 }
1307
1308 static const struct eth_dev_ops ops = {
1309         .dev_start              = tap_dev_start,
1310         .dev_stop               = tap_dev_stop,
1311         .dev_close              = tap_dev_close,
1312         .dev_configure          = tap_dev_configure,
1313         .dev_infos_get          = tap_dev_info,
1314         .rx_queue_setup         = tap_rx_queue_setup,
1315         .tx_queue_setup         = tap_tx_queue_setup,
1316         .rx_queue_release       = tap_rx_queue_release,
1317         .tx_queue_release       = tap_tx_queue_release,
1318         .flow_ctrl_get          = tap_flow_ctrl_get,
1319         .flow_ctrl_set          = tap_flow_ctrl_set,
1320         .link_update            = tap_link_update,
1321         .dev_set_link_up        = tap_link_set_up,
1322         .dev_set_link_down      = tap_link_set_down,
1323         .promiscuous_enable     = tap_promisc_enable,
1324         .promiscuous_disable    = tap_promisc_disable,
1325         .allmulticast_enable    = tap_allmulti_enable,
1326         .allmulticast_disable   = tap_allmulti_disable,
1327         .mac_addr_set           = tap_mac_set,
1328         .mtu_set                = tap_mtu_set,
1329         .set_mc_addr_list       = tap_set_mc_addr_list,
1330         .stats_get              = tap_stats_get,
1331         .stats_reset            = tap_stats_reset,
1332         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1333         .filter_ctrl            = tap_dev_filter_ctrl,
1334 };
1335
1336 static int
1337 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1338                    char *remote_iface, int fixed_mac_type)
1339 {
1340         int numa_node = rte_socket_id();
1341         struct rte_eth_dev *dev;
1342         struct pmd_internals *pmd;
1343         struct rte_eth_dev_data *data;
1344         struct ifreq ifr;
1345         int i;
1346
1347         RTE_LOG(DEBUG, PMD, "  TAP device on numa %u\n", rte_socket_id());
1348
1349         data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
1350         if (!data) {
1351                 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
1352                 goto error_exit_nodev;
1353         }
1354
1355         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1356         if (!dev) {
1357                 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
1358                 goto error_exit_nodev;
1359         }
1360
1361         pmd = dev->data->dev_private;
1362         pmd->dev = dev;
1363         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1364
1365         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1366         if (pmd->ioctl_sock == -1) {
1367                 RTE_LOG(ERR, PMD,
1368                         "TAP Unable to get a socket for management: %s\n",
1369                         strerror(errno));
1370                 goto error_exit;
1371         }
1372
1373         /* Setup some default values */
1374         rte_memcpy(data, dev->data, sizeof(*data));
1375         data->dev_private = pmd;
1376         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1377         data->numa_node = numa_node;
1378
1379         data->dev_link = pmd_link;
1380         data->mac_addrs = &pmd->eth_addr;
1381         /* Set the number of RX and TX queues */
1382         data->nb_rx_queues = 0;
1383         data->nb_tx_queues = 0;
1384
1385         dev->data = data;
1386         dev->dev_ops = &ops;
1387         dev->rx_pkt_burst = pmd_rx_burst;
1388         dev->tx_pkt_burst = pmd_tx_burst;
1389
1390         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1391         pmd->intr_handle.fd = -1;
1392         dev->intr_handle = &pmd->intr_handle;
1393
1394         /* Presetup the fds to -1 as being not valid */
1395         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1396                 pmd->rxq[i].fd = -1;
1397                 pmd->txq[i].fd = -1;
1398         }
1399
1400         if (fixed_mac_type) {
1401                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1402                 static int iface_idx;
1403                 char mac[ETHER_ADDR_LEN] = "\0dtap";
1404
1405                 mac[ETHER_ADDR_LEN - 1] = iface_idx++;
1406                 rte_memcpy(&pmd->eth_addr, mac, ETHER_ADDR_LEN);
1407         } else {
1408                 eth_random_addr((uint8_t *)&pmd->eth_addr);
1409         }
1410
1411         /* Immediately create the netdevice (this will create the 1st queue). */
1412         /* rx queue */
1413         if (tap_setup_queue(dev, pmd, 0, 1) == -1)
1414                 goto error_exit;
1415         /* tx queue */
1416         if (tap_setup_queue(dev, pmd, 0, 0) == -1)
1417                 goto error_exit;
1418
1419         ifr.ifr_mtu = dev->data->mtu;
1420         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1421                 goto error_exit;
1422
1423         memset(&ifr, 0, sizeof(struct ifreq));
1424         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1425         rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr, ETHER_ADDR_LEN);
1426         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1427                 goto error_exit;
1428
1429         /*
1430          * Set up everything related to rte_flow:
1431          * - netlink socket
1432          * - tap / remote if_index
1433          * - mandatory QDISCs
1434          * - rte_flow actual/implicit lists
1435          * - implicit rules
1436          */
1437         pmd->nlsk_fd = tap_nl_init(0);
1438         if (pmd->nlsk_fd == -1) {
1439                 RTE_LOG(WARNING, PMD, "%s: failed to create netlink socket.\n",
1440                         pmd->name);
1441                 goto disable_rte_flow;
1442         }
1443         pmd->if_index = if_nametoindex(pmd->name);
1444         if (!pmd->if_index) {
1445                 RTE_LOG(ERR, PMD, "%s: failed to get if_index.\n", pmd->name);
1446                 goto disable_rte_flow;
1447         }
1448         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1449                 RTE_LOG(ERR, PMD, "%s: failed to create multiq qdisc.\n",
1450                         pmd->name);
1451                 goto disable_rte_flow;
1452         }
1453         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1454                 RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1455                         pmd->name);
1456                 goto disable_rte_flow;
1457         }
1458         LIST_INIT(&pmd->flows);
1459
1460         if (strlen(remote_iface)) {
1461                 pmd->remote_if_index = if_nametoindex(remote_iface);
1462                 if (!pmd->remote_if_index) {
1463                         RTE_LOG(ERR, PMD, "%s: failed to get %s if_index.\n",
1464                                 pmd->name, remote_iface);
1465                         goto error_remote;
1466                 }
1467                 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1468                          "%s", remote_iface);
1469
1470                 /* Save state of remote device */
1471                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1472
1473                 /* Replicate remote MAC address */
1474                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1475                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1476                                 pmd->name, pmd->remote_iface);
1477                         goto error_remote;
1478                 }
1479                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1480                            ETHER_ADDR_LEN);
1481                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1482                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1483                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1484                                 pmd->name, remote_iface);
1485                         goto error_remote;
1486                 }
1487
1488                 /*
1489                  * Flush usually returns negative value because it tries to
1490                  * delete every QDISC (and on a running device, one QDISC at
1491                  * least is needed). Ignore negative return value.
1492                  */
1493                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1494                 if (qdisc_create_ingress(pmd->nlsk_fd,
1495                                          pmd->remote_if_index) < 0) {
1496                         RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1497                                 pmd->remote_iface);
1498                         goto error_remote;
1499                 }
1500                 LIST_INIT(&pmd->implicit_flows);
1501                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1502                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1503                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1504                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1505                         RTE_LOG(ERR, PMD,
1506                                 "%s: failed to create implicit rules.\n",
1507                                 pmd->name);
1508                         goto error_remote;
1509                 }
1510         }
1511
1512         return 0;
1513
1514 disable_rte_flow:
1515         RTE_LOG(ERR, PMD, " Disabling rte flow support: %s(%d)\n",
1516                 strerror(errno), errno);
1517         if (strlen(remote_iface)) {
1518                 RTE_LOG(ERR, PMD, "Remote feature requires flow support.\n");
1519                 goto error_exit;
1520         }
1521         return 0;
1522
1523 error_remote:
1524         RTE_LOG(ERR, PMD, " Can't set up remote feature: %s(%d)\n",
1525                 strerror(errno), errno);
1526         tap_flow_implicit_flush(pmd, NULL);
1527
1528 error_exit:
1529         if (pmd->ioctl_sock > 0)
1530                 close(pmd->ioctl_sock);
1531         rte_eth_dev_release_port(dev);
1532
1533 error_exit_nodev:
1534         RTE_LOG(ERR, PMD, "TAP Unable to initialize %s\n",
1535                 rte_vdev_device_name(vdev));
1536
1537         rte_free(data);
1538         return -EINVAL;
1539 }
1540
1541 static int
1542 set_interface_name(const char *key __rte_unused,
1543                    const char *value,
1544                    void *extra_args)
1545 {
1546         char *name = (char *)extra_args;
1547
1548         if (value)
1549                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
1550         else
1551                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1552                          DEFAULT_TAP_NAME, (tap_unit - 1));
1553
1554         return 0;
1555 }
1556
1557 static int
1558 set_remote_iface(const char *key __rte_unused,
1559                  const char *value,
1560                  void *extra_args)
1561 {
1562         char *name = (char *)extra_args;
1563
1564         if (value)
1565                 snprintf(name, RTE_ETH_NAME_MAX_LEN, "%s", value);
1566
1567         return 0;
1568 }
1569
1570 static int
1571 set_mac_type(const char *key __rte_unused,
1572              const char *value,
1573              void *extra_args)
1574 {
1575         if (value &&
1576             !strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED)))
1577                 *(int *)extra_args = 1;
1578         return 0;
1579 }
1580
1581 /* Open a TAP interface device.
1582  */
1583 static int
1584 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1585 {
1586         const char *name, *params;
1587         int ret;
1588         struct rte_kvargs *kvlist = NULL;
1589         int speed;
1590         char tap_name[RTE_ETH_NAME_MAX_LEN];
1591         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1592         int fixed_mac_type = 0;
1593
1594         name = rte_vdev_device_name(dev);
1595         params = rte_vdev_device_args(dev);
1596
1597         speed = ETH_SPEED_NUM_10G;
1598         snprintf(tap_name, sizeof(tap_name), "%s%d",
1599                  DEFAULT_TAP_NAME, tap_unit++);
1600         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1601
1602         if (params && (params[0] != '\0')) {
1603                 RTE_LOG(DEBUG, PMD, "parameters (%s)\n", params);
1604
1605                 kvlist = rte_kvargs_parse(params, valid_arguments);
1606                 if (kvlist) {
1607                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1608                                 ret = rte_kvargs_process(kvlist,
1609                                                          ETH_TAP_IFACE_ARG,
1610                                                          &set_interface_name,
1611                                                          tap_name);
1612                                 if (ret == -1)
1613                                         goto leave;
1614                         }
1615
1616                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1617                                 ret = rte_kvargs_process(kvlist,
1618                                                          ETH_TAP_REMOTE_ARG,
1619                                                          &set_remote_iface,
1620                                                          remote_iface);
1621                                 if (ret == -1)
1622                                         goto leave;
1623                         }
1624
1625                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1626                                 ret = rte_kvargs_process(kvlist,
1627                                                          ETH_TAP_MAC_ARG,
1628                                                          &set_mac_type,
1629                                                          &fixed_mac_type);
1630                                 if (ret == -1)
1631                                         goto leave;
1632                         }
1633                 }
1634         }
1635         pmd_link.link_speed = speed;
1636
1637         RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1638                 name, tap_name);
1639
1640         ret = eth_dev_tap_create(dev, tap_name, remote_iface, fixed_mac_type);
1641
1642 leave:
1643         if (ret == -1) {
1644                 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1645                         name, tap_name);
1646                 tap_unit--;             /* Restore the unit number */
1647         }
1648         rte_kvargs_free(kvlist);
1649
1650         return ret;
1651 }
1652
1653 /* detach a TAP device.
1654  */
1655 static int
1656 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1657 {
1658         struct rte_eth_dev *eth_dev = NULL;
1659         struct pmd_internals *internals;
1660         int i;
1661
1662         RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1663                 rte_socket_id());
1664
1665         /* find the ethdev entry */
1666         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1667         if (!eth_dev)
1668                 return 0;
1669
1670         internals = eth_dev->data->dev_private;
1671         if (internals->nlsk_fd) {
1672                 tap_flow_flush(eth_dev, NULL);
1673                 tap_flow_implicit_flush(internals, NULL);
1674                 tap_nl_final(internals->nlsk_fd);
1675         }
1676         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1677                 if (internals->rxq[i].fd != -1) {
1678                         close(internals->rxq[i].fd);
1679                         internals->rxq[i].fd = -1;
1680                 }
1681                 if (internals->txq[i].fd != -1) {
1682                         close(internals->txq[i].fd);
1683                         internals->txq[i].fd = -1;
1684                 }
1685         }
1686
1687         close(internals->ioctl_sock);
1688         rte_free(eth_dev->data->dev_private);
1689         rte_free(eth_dev->data);
1690
1691         rte_eth_dev_release_port(eth_dev);
1692
1693         return 0;
1694 }
1695
1696 static struct rte_vdev_driver pmd_tap_drv = {
1697         .probe = rte_pmd_tap_probe,
1698         .remove = rte_pmd_tap_remove,
1699 };
1700 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1701 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1702 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1703                               ETH_TAP_IFACE_ARG "=<string> "
1704                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_FIXED " "
1705                               ETH_TAP_REMOTE_ARG "=<string>");