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_IPV4_CKSUM |
 394                DEV_TX_OFFLOAD_UDP_CKSUM |
 395                DEV_TX_OFFLOAD_TCP_CKSUM;
 396 }
 397
 398 static uint64_t
 399 tap_tx_offload_get_queue_capa(void)
 400 {
 401         return DEV_TX_OFFLOAD_IPV4_CKSUM |
 402                DEV_TX_OFFLOAD_UDP_CKSUM |
 403                DEV_TX_OFFLOAD_TCP_CKSUM;
 404 }
 405
 406 static bool
 407 tap_txq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 408 {
 409         uint64_t port_offloads = dev->data->dev_conf.txmode.offloads;
 410         uint64_t queue_supp_offloads = tap_tx_offload_get_queue_capa();
 411         uint64_t port_supp_offloads = tap_tx_offload_get_port_capa();
 412
 413         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 414             offloads)
 415                 return false;
 416         /* Verify we have no conflict with port offloads */
 417         if ((port_offloads ^ offloads) & port_supp_offloads)
 418                 return false;
 419         return true;
 420 }
 421
 422 static void
 423 tap_tx_offload(char *packet, uint64_t ol_flags, unsigned int l2_len,
 424                unsigned int l3_len)
 425 {
 426         void *l3_hdr = packet + l2_len;
 427
 428         if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
 429                 struct ipv4_hdr *iph = l3_hdr;
 430                 uint16_t cksum;
 431
 432                 iph->hdr_checksum = 0;
 433                 cksum = rte_raw_cksum(iph, l3_len);
 434                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 435         }
 436         if (ol_flags & PKT_TX_L4_MASK) {
 437                 uint16_t l4_len;
 438                 uint32_t cksum;
 439                 uint16_t *l4_cksum;
 440                 void *l4_hdr;
 441
 442                 l4_hdr = packet + l2_len + l3_len;
 443                 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
 444                         l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
 445                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 446                         l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
 447                 else
 448                         return;
 449                 *l4_cksum = 0;
 450                 if (ol_flags & PKT_TX_IPV4) {
 451                         struct ipv4_hdr *iph = l3_hdr;
 452
 453                         l4_len = rte_be_to_cpu_16(iph->total_length) - l3_len;
 454                         cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 455                 } else {
 456                         struct ipv6_hdr *ip6h = l3_hdr;
 457
 458                         /* payload_len does not include ext headers */
 459                         l4_len = rte_be_to_cpu_16(ip6h->payload_len) -
 460                                 l3_len + sizeof(struct ipv6_hdr);
 461                         cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 462                 }
 463                 cksum += rte_raw_cksum(l4_hdr, l4_len);
 464                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 465                 cksum = (~cksum) & 0xffff;
 466                 if (cksum == 0)
 467                         cksum = 0xffff;
 468                 *l4_cksum = cksum;
 469         }
 470 }
 471
 472 /* Callback to handle sending packets from the tap interface
 473  */
 474 static uint16_t
 475 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 476 {
 477         struct tx_queue *txq = queue;
 478         uint16_t num_tx = 0;
 479         unsigned long num_tx_bytes = 0;
 480         uint32_t max_size;
 481         int i;
 482
 483         if (unlikely(nb_pkts == 0))
 484                 return 0;
 485
 486         max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
 487         for (i = 0; i < nb_pkts; i++) {
 488                 struct rte_mbuf *mbuf = bufs[num_tx];
 489                 struct iovec iovecs[mbuf->nb_segs + 1];
 490                 struct tun_pi pi = { .flags = 0 };
 491                 struct rte_mbuf *seg = mbuf;
 492                 char m_copy[mbuf->data_len];
 493                 int n;
 494                 int j;
 495
 496                 /* stats.errs will be incremented */
 497                 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
 498                         break;
 499
 500                 iovecs[0].iov_base = &pi;
 501                 iovecs[0].iov_len = sizeof(pi);
 502                 for (j = 1; j <= mbuf->nb_segs; j++) {
 503                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 504                         iovecs[j].iov_base =
 505                                 rte_pktmbuf_mtod(seg, void *);
 506                         seg = seg->next;
 507                 }
 508                 if (txq->csum &&
 509                     ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 510                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 511                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
 512                         /* Support only packets with all data in the same seg */
 513                         if (mbuf->nb_segs > 1)
 514                                 break;
 515                         /* To change checksums, work on a copy of data. */
 516                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 517                                    rte_pktmbuf_data_len(mbuf));
 518                         tap_tx_offload(m_copy, mbuf->ol_flags,
 519                                        mbuf->l2_len, mbuf->l3_len);
 520                         iovecs[1].iov_base = m_copy;
 521                 }
 522                 /* copy the tx frame data */
 523                 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
 524                 if (n <= 0)
 525                         break;
 526
 527                 num_tx++;
 528                 num_tx_bytes += mbuf->pkt_len;
 529                 rte_pktmbuf_free(mbuf);
 530         }
 531
 532         txq->stats.opackets += num_tx;
 533         txq->stats.errs += nb_pkts - num_tx;
 534         txq->stats.obytes += num_tx_bytes;
 535
 536         return num_tx;
 537 }
 538
 539 static const char *
 540 tap_ioctl_req2str(unsigned long request)
 541 {
 542         switch (request) {
 543         case SIOCSIFFLAGS:
 544                 return "SIOCSIFFLAGS";
 545         case SIOCGIFFLAGS:
 546                 return "SIOCGIFFLAGS";
 547         case SIOCGIFHWADDR:
 548                 return "SIOCGIFHWADDR";
 549         case SIOCSIFHWADDR:
 550                 return "SIOCSIFHWADDR";
 551         case SIOCSIFMTU:
 552                 return "SIOCSIFMTU";
 553         }
 554         return "UNKNOWN";
 555 }
 556
 557 static int
 558 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 559           struct ifreq *ifr, int set, enum ioctl_mode mode)
 560 {
 561         short req_flags = ifr->ifr_flags;
 562         int remote = pmd->remote_if_index &&
 563                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 564
 565         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 566                 return 0;
 567         /*
 568          * If there is a remote netdevice, apply ioctl on it, then apply it on
 569          * the tap netdevice.
 570          */
 571 apply:
 572         if (remote)
 573                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
 574         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 575                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 576         switch (request) {
 577         case SIOCSIFFLAGS:
 578                 /* fetch current flags to leave other flags untouched */
 579                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 580                         goto error;
 581                 if (set)
 582                         ifr->ifr_flags |= req_flags;
 583                 else
 584                         ifr->ifr_flags &= ~req_flags;
 585                 break;
 586         case SIOCGIFFLAGS:
 587         case SIOCGIFHWADDR:
 588         case SIOCSIFHWADDR:
 589         case SIOCSIFMTU:
 590                 break;
 591         default:
 592                 RTE_ASSERT(!"unsupported request type: must not happen");
 593         }
 594         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 595                 goto error;
 596         if (remote-- && mode == LOCAL_AND_REMOTE)
 597                 goto apply;
 598         return 0;
 599
 600 error:
 601         RTE_LOG(DEBUG, PMD, "%s: %s(%s) failed: %s(%d)\n", ifr->ifr_name,
 602                 __func__, tap_ioctl_req2str(request), strerror(errno), errno);
 603         return -errno;
 604 }
 605
 606 static int
 607 tap_link_set_down(struct rte_eth_dev *dev)
 608 {
 609         struct pmd_internals *pmd = dev->data->dev_private;
 610         struct ifreq ifr = { .ifr_flags = IFF_UP };
 611
 612         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 613         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 614 }
 615
 616 static int
 617 tap_link_set_up(struct rte_eth_dev *dev)
 618 {
 619         struct pmd_internals *pmd = dev->data->dev_private;
 620         struct ifreq ifr = { .ifr_flags = IFF_UP };
 621
 622         dev->data->dev_link.link_status = ETH_LINK_UP;
 623         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 624 }
 625
 626 static int
 627 tap_dev_start(struct rte_eth_dev *dev)
 628 {
 629         int err;
 630
 631         err = tap_intr_handle_set(dev, 1);
 632         if (err)
 633                 return err;
 634         return tap_link_set_up(dev);
 635 }
 636
 637 /* This function gets called when the current port gets stopped.
 638  */
 639 static void
 640 tap_dev_stop(struct rte_eth_dev *dev)
 641 {
 642         tap_intr_handle_set(dev, 0);
 643         tap_link_set_down(dev);
 644 }
 645
 646 static int
 647 tap_dev_configure(struct rte_eth_dev *dev)
 648 {
 649         uint64_t supp_tx_offloads = tap_tx_offload_get_port_capa();
 650         uint64_t tx_offloads = dev->data->dev_conf.txmode.offloads;
 651
 652         if ((tx_offloads & supp_tx_offloads) != tx_offloads) {
 653                 rte_errno = ENOTSUP;
 654                 RTE_LOG(ERR, PMD,
 655                         "Some Tx offloads are not supported "
 656                         "requested 0x%" PRIx64 " supported 0x%" PRIx64 "\n",
 657                         tx_offloads, supp_tx_offloads);
 658                 return -rte_errno;
 659         }
 660         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 661                 RTE_LOG(ERR, PMD,
 662                         "%s: number of rx queues %d exceeds max num of queues %d\n",
 663                         dev->device->name,
 664                         dev->data->nb_rx_queues,
 665                         RTE_PMD_TAP_MAX_QUEUES);
 666                 return -1;
 667         }
 668         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 669                 RTE_LOG(ERR, PMD,
 670                         "%s: number of tx queues %d exceeds max num of queues %d\n",
 671                         dev->device->name,
 672                         dev->data->nb_tx_queues,
 673                         RTE_PMD_TAP_MAX_QUEUES);
 674                 return -1;
 675         }
 676
 677         RTE_LOG(INFO, PMD, "%s: %p: TX configured queues number: %u\n",
 678              dev->device->name, (void *)dev, dev->data->nb_tx_queues);
 679
 680         RTE_LOG(INFO, PMD, "%s: %p: RX configured queues number: %u\n",
 681              dev->device->name, (void *)dev, dev->data->nb_rx_queues);
 682
 683         return 0;
 684 }
 685
 686 static uint32_t
 687 tap_dev_speed_capa(void)
 688 {
 689         uint32_t speed = pmd_link.link_speed;
 690         uint32_t capa = 0;
 691
 692         if (speed >= ETH_SPEED_NUM_10M)
 693                 capa |= ETH_LINK_SPEED_10M;
 694         if (speed >= ETH_SPEED_NUM_100M)
 695                 capa |= ETH_LINK_SPEED_100M;
 696         if (speed >= ETH_SPEED_NUM_1G)
 697                 capa |= ETH_LINK_SPEED_1G;
 698         if (speed >= ETH_SPEED_NUM_5G)
 699                 capa |= ETH_LINK_SPEED_2_5G;
 700         if (speed >= ETH_SPEED_NUM_5G)
 701                 capa |= ETH_LINK_SPEED_5G;
 702         if (speed >= ETH_SPEED_NUM_10G)
 703                 capa |= ETH_LINK_SPEED_10G;
 704         if (speed >= ETH_SPEED_NUM_20G)
 705                 capa |= ETH_LINK_SPEED_20G;
 706         if (speed >= ETH_SPEED_NUM_25G)
 707                 capa |= ETH_LINK_SPEED_25G;
 708         if (speed >= ETH_SPEED_NUM_40G)
 709                 capa |= ETH_LINK_SPEED_40G;
 710         if (speed >= ETH_SPEED_NUM_50G)
 711                 capa |= ETH_LINK_SPEED_50G;
 712         if (speed >= ETH_SPEED_NUM_56G)
 713                 capa |= ETH_LINK_SPEED_56G;
 714         if (speed >= ETH_SPEED_NUM_100G)
 715                 capa |= ETH_LINK_SPEED_100G;
 716
 717         return capa;
 718 }
 719
 720 static void
 721 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 722 {
 723         struct pmd_internals *internals = dev->data->dev_private;
 724
 725         dev_info->if_index = internals->if_index;
 726         dev_info->max_mac_addrs = 1;
 727         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 728         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
 729         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
 730         dev_info->min_rx_bufsize = 0;
 731         dev_info->pci_dev = NULL;
 732         dev_info->speed_capa = tap_dev_speed_capa();
 733         dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
 734         dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
 735                                     dev_info->rx_queue_offload_capa;
 736         dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
 737         dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
 738                                     dev_info->tx_queue_offload_capa;
 739 }
 740
 741 static int
 742 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 743 {
 744         unsigned int i, imax;
 745         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 746         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 747         unsigned long rx_nombuf = 0, ierrors = 0;
 748         const struct pmd_internals *pmd = dev->data->dev_private;
 749
 750         /* rx queue statistics */
 751         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 752                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 753         for (i = 0; i < imax; i++) {
 754                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 755                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 756                 rx_total += tap_stats->q_ipackets[i];
 757                 rx_bytes_total += tap_stats->q_ibytes[i];
 758                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
 759                 ierrors += pmd->rxq[i].stats.ierrors;
 760         }
 761
 762         /* tx queue statistics */
 763         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 764                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 765
 766         for (i = 0; i < imax; i++) {
 767                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 768                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 769                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 770                 tx_total += tap_stats->q_opackets[i];
 771                 tx_err_total += tap_stats->q_errors[i];
 772                 tx_bytes_total += tap_stats->q_obytes[i];
 773         }
 774
 775         tap_stats->ipackets = rx_total;
 776         tap_stats->ibytes = rx_bytes_total;
 777         tap_stats->ierrors = ierrors;
 778         tap_stats->rx_nombuf = rx_nombuf;
 779         tap_stats->opackets = tx_total;
 780         tap_stats->oerrors = tx_err_total;
 781         tap_stats->obytes = tx_bytes_total;
 782         return 0;
 783 }
 784
 785 static void
 786 tap_stats_reset(struct rte_eth_dev *dev)
 787 {
 788         int i;
 789         struct pmd_internals *pmd = dev->data->dev_private;
 790
 791         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 792                 pmd->rxq[i].stats.ipackets = 0;
 793                 pmd->rxq[i].stats.ibytes = 0;
 794                 pmd->rxq[i].stats.ierrors = 0;
 795                 pmd->rxq[i].stats.rx_nombuf = 0;
 796
 797                 pmd->txq[i].stats.opackets = 0;
 798                 pmd->txq[i].stats.errs = 0;
 799                 pmd->txq[i].stats.obytes = 0;
 800         }
 801 }
 802
 803 static void
 804 tap_dev_close(struct rte_eth_dev *dev)
 805 {
 806         int i;
 807         struct pmd_internals *internals = dev->data->dev_private;
 808
 809         tap_link_set_down(dev);
 810         tap_flow_flush(dev, NULL);
 811         tap_flow_implicit_flush(internals, NULL);
 812
 813         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 814                 if (internals->rxq[i].fd != -1) {
 815                         close(internals->rxq[i].fd);
 816                         internals->rxq[i].fd = -1;
 817                 }
 818                 if (internals->txq[i].fd != -1) {
 819                         close(internals->txq[i].fd);
 820                         internals->txq[i].fd = -1;
 821                 }
 822         }
 823
 824         if (internals->remote_if_index) {
 825                 /* Restore initial remote state */
 826                 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
 827                                 &internals->remote_initial_flags);
 828         }
 829 }
 830
 831 static void
 832 tap_rx_queue_release(void *queue)
 833 {
 834         struct rx_queue *rxq = queue;
 835
 836         if (rxq && (rxq->fd > 0)) {
 837                 close(rxq->fd);
 838                 rxq->fd = -1;
 839                 rte_pktmbuf_free(rxq->pool);
 840                 rte_free(rxq->iovecs);
 841                 rxq->pool = NULL;
 842                 rxq->iovecs = NULL;
 843         }
 844 }
 845
 846 static void
 847 tap_tx_queue_release(void *queue)
 848 {
 849         struct tx_queue *txq = queue;
 850
 851         if (txq && (txq->fd > 0)) {
 852                 close(txq->fd);
 853                 txq->fd = -1;
 854         }
 855 }
 856
 857 static int
 858 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
 859 {
 860         struct rte_eth_link *dev_link = &dev->data->dev_link;
 861         struct pmd_internals *pmd = dev->data->dev_private;
 862         struct ifreq ifr = { .ifr_flags = 0 };
 863
 864         if (pmd->remote_if_index) {
 865                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
 866                 if (!(ifr.ifr_flags & IFF_UP) ||
 867                     !(ifr.ifr_flags & IFF_RUNNING)) {
 868                         dev_link->link_status = ETH_LINK_DOWN;
 869                         return 0;
 870                 }
 871         }
 872         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
 873         dev_link->link_status =
 874                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
 875                  ETH_LINK_UP :
 876                  ETH_LINK_DOWN);
 877         return 0;
 878 }
 879
 880 static void
 881 tap_promisc_enable(struct rte_eth_dev *dev)
 882 {
 883         struct pmd_internals *pmd = dev->data->dev_private;
 884         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 885
 886         dev->data->promiscuous = 1;
 887         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 888         if (pmd->remote_if_index && !pmd->flow_isolate)
 889                 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
 890 }
 891
 892 static void
 893 tap_promisc_disable(struct rte_eth_dev *dev)
 894 {
 895         struct pmd_internals *pmd = dev->data->dev_private;
 896         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 897
 898         dev->data->promiscuous = 0;
 899         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 900         if (pmd->remote_if_index && !pmd->flow_isolate)
 901                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
 902 }
 903
 904 static void
 905 tap_allmulti_enable(struct rte_eth_dev *dev)
 906 {
 907         struct pmd_internals *pmd = dev->data->dev_private;
 908         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 909
 910         dev->data->all_multicast = 1;
 911         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 912         if (pmd->remote_if_index && !pmd->flow_isolate)
 913                 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
 914 }
 915
 916 static void
 917 tap_allmulti_disable(struct rte_eth_dev *dev)
 918 {
 919         struct pmd_internals *pmd = dev->data->dev_private;
 920         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 921
 922         dev->data->all_multicast = 0;
 923         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 924         if (pmd->remote_if_index && !pmd->flow_isolate)
 925                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
 926 }
 927
 928 static void
 929 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 930 {
 931         struct pmd_internals *pmd = dev->data->dev_private;
 932         enum ioctl_mode mode = LOCAL_ONLY;
 933         struct ifreq ifr;
 934
 935         if (is_zero_ether_addr(mac_addr)) {
 936                 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
 937                         dev->device->name);
 938                 return;
 939         }
 940         /* Check the actual current MAC address on the tap netdevice */
 941         if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
 942                 return;
 943         if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 944                                mac_addr))
 945                 return;
 946         /* Check the current MAC address on the remote */
 947         if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0)
 948                 return;
 949         if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 950                                mac_addr))
 951                 mode = LOCAL_AND_REMOTE;
 952         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
 953         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
 954         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode) < 0)
 955                 return;
 956         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
 957         if (pmd->remote_if_index && !pmd->flow_isolate) {
 958                 /* Replace MAC redirection rule after a MAC change */
 959                 if (tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC) < 0) {
 960                         RTE_LOG(ERR, PMD,
 961                                 "%s: Couldn't delete MAC redirection rule\n",
 962                                 dev->device->name);
 963                         return;
 964                 }
 965                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0)
 966                         RTE_LOG(ERR, PMD,
 967                                 "%s: Couldn't add MAC redirection rule\n",
 968                                 dev->device->name);
 969         }
 970 }
 971
 972 static int
 973 tap_setup_queue(struct rte_eth_dev *dev,
 974                 struct pmd_internals *internals,
 975                 uint16_t qid,
 976                 int is_rx)
 977 {
 978         int *fd;
 979         int *other_fd;
 980         const char *dir;
 981         struct pmd_internals *pmd = dev->data->dev_private;
 982         struct rx_queue *rx = &internals->rxq[qid];
 983         struct tx_queue *tx = &internals->txq[qid];
 984
 985         if (is_rx) {
 986                 fd = &rx->fd;
 987                 other_fd = &tx->fd;
 988                 dir = "rx";
 989         } else {
 990                 fd = &tx->fd;
 991                 other_fd = &rx->fd;
 992                 dir = "tx";
 993         }
 994         if (*fd != -1) {
 995                 /* fd for this queue already exists */
 996                 RTE_LOG(DEBUG, PMD, "%s: fd %d for %s queue qid %d exists\n",
 997                         pmd->name, *fd, dir, qid);
 998         } else if (*other_fd != -1) {
 999                 /* Only other_fd exists. dup it */
1000                 *fd = dup(*other_fd);
1001                 if (*fd < 0) {
1002                         *fd = -1;
1003                         RTE_LOG(ERR, PMD, "%s: dup() failed.\n",
1004                                 pmd->name);
1005                         return -1;
1006                 }
1007                 RTE_LOG(DEBUG, PMD, "%s: dup fd %d for %s queue qid %d (%d)\n",
1008                         pmd->name, *other_fd, dir, qid, *fd);
1009         } else {
1010                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1011                 *fd = tun_alloc(pmd);
1012                 if (*fd < 0) {
1013                         *fd = -1; /* restore original value */
1014                         RTE_LOG(ERR, PMD, "%s: tun_alloc() failed.\n",
1015                                 pmd->name);
1016                         return -1;
1017                 }
1018                 RTE_LOG(DEBUG, PMD, "%s: add %s queue for qid %d fd %d\n",
1019                         pmd->name, dir, qid, *fd);
1020         }
1021
1022         tx->mtu = &dev->data->mtu;
1023         rx->rxmode = &dev->data->dev_conf.rxmode;
1024
1025         return *fd;
1026 }
1027
1028 static int
1029 tap_rx_queue_setup(struct rte_eth_dev *dev,
1030                    uint16_t rx_queue_id,
1031                    uint16_t nb_rx_desc,
1032                    unsigned int socket_id,
1033                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1034                    struct rte_mempool *mp)
1035 {
1036         struct pmd_internals *internals = dev->data->dev_private;
1037         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1038         struct rte_mbuf **tmp = &rxq->pool;
1039         long iov_max = sysconf(_SC_IOV_MAX);
1040         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1041         struct iovec (*iovecs)[nb_desc + 1];
1042         int data_off = RTE_PKTMBUF_HEADROOM;
1043         int ret = 0;
1044         int fd;
1045         int i;
1046
1047         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1048                 RTE_LOG(WARNING, PMD,
1049                         "nb_rx_queues %d too small or mempool NULL\n",
1050                         dev->data->nb_rx_queues);
1051                 return -1;
1052         }
1053
1054         /* Verify application offloads are valid for our port and queue. */
1055         if (!tap_rxq_are_offloads_valid(dev, rx_conf->offloads)) {
1056                 rte_errno = ENOTSUP;
1057                 RTE_LOG(ERR, PMD,
1058                         "%p: Rx queue offloads 0x%" PRIx64
1059                         " don't match port offloads 0x%" PRIx64
1060                         " or supported offloads 0x%" PRIx64 "\n",
1061                         (void *)dev, rx_conf->offloads,
1062                         dev->data->dev_conf.rxmode.offloads,
1063                         (tap_rx_offload_get_port_capa() |
1064                          tap_rx_offload_get_queue_capa()));
1065                 return -rte_errno;
1066         }
1067         rxq->mp = mp;
1068         rxq->trigger_seen = 1; /* force initial burst */
1069         rxq->in_port = dev->data->port_id;
1070         rxq->nb_rx_desc = nb_desc;
1071         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1072                                     socket_id);
1073         if (!iovecs) {
1074                 RTE_LOG(WARNING, PMD,
1075                         "%s: Couldn't allocate %d RX descriptors\n",
1076                         dev->device->name, nb_desc);
1077                 return -ENOMEM;
1078         }
1079         rxq->iovecs = iovecs;
1080
1081         dev->data->rx_queues[rx_queue_id] = rxq;
1082         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1083         if (fd == -1) {
1084                 ret = fd;
1085                 goto error;
1086         }
1087
1088         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1089         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1090
1091         for (i = 1; i <= nb_desc; i++) {
1092                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1093                 if (!*tmp) {
1094                         RTE_LOG(WARNING, PMD,
1095                                 "%s: couldn't allocate memory for queue %d\n",
1096                                 dev->device->name, rx_queue_id);
1097                         ret = -ENOMEM;
1098                         goto error;
1099                 }
1100                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1101                 (*rxq->iovecs)[i].iov_base =
1102                         (char *)(*tmp)->buf_addr + data_off;
1103                 data_off = 0;
1104                 tmp = &(*tmp)->next;
1105         }
1106
1107         RTE_LOG(DEBUG, PMD, "  RX TAP device name %s, qid %d on fd %d\n",
1108                 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
1109
1110         return 0;
1111
1112 error:
1113         rte_pktmbuf_free(rxq->pool);
1114         rxq->pool = NULL;
1115         rte_free(rxq->iovecs);
1116         rxq->iovecs = NULL;
1117         return ret;
1118 }
1119
1120 static int
1121 tap_tx_queue_setup(struct rte_eth_dev *dev,
1122                    uint16_t tx_queue_id,
1123                    uint16_t nb_tx_desc __rte_unused,
1124                    unsigned int socket_id __rte_unused,
1125                    const struct rte_eth_txconf *tx_conf)
1126 {
1127         struct pmd_internals *internals = dev->data->dev_private;
1128         struct tx_queue *txq;
1129         int ret;
1130
1131         if (tx_queue_id >= dev->data->nb_tx_queues)
1132                 return -1;
1133         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1134         txq = dev->data->tx_queues[tx_queue_id];
1135         /*
1136          * Don't verify port offloads for application which
1137          * use the old API.
1138          */
1139         if (tx_conf != NULL &&
1140             !!(tx_conf->txq_flags & ETH_TXQ_FLAGS_IGNORE)) {
1141                 if (tap_txq_are_offloads_valid(dev, tx_conf->offloads)) {
1142                         txq->csum = !!(tx_conf->offloads &
1143                                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
1144                                          DEV_TX_OFFLOAD_UDP_CKSUM |
1145                                          DEV_TX_OFFLOAD_TCP_CKSUM));
1146                 } else {
1147                         rte_errno = ENOTSUP;
1148                         RTE_LOG(ERR, PMD,
1149                                 "%p: Tx queue offloads 0x%" PRIx64
1150                                 " don't match port offloads 0x%" PRIx64
1151                                 " or supported offloads 0x%" PRIx64,
1152                                 (void *)dev, tx_conf->offloads,
1153                                 dev->data->dev_conf.txmode.offloads,
1154                                 tap_tx_offload_get_port_capa());
1155                         return -rte_errno;
1156                 }
1157         }
1158         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1159         if (ret == -1)
1160                 return -1;
1161         RTE_LOG(DEBUG, PMD,
1162                 "  TX TAP device name %s, qid %d on fd %d csum %s\n",
1163                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd,
1164                 txq->csum ? "on" : "off");
1165
1166         return 0;
1167 }
1168
1169 static int
1170 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1171 {
1172         struct pmd_internals *pmd = dev->data->dev_private;
1173         struct ifreq ifr = { .ifr_mtu = mtu };
1174         int err = 0;
1175
1176         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1177         if (!err)
1178                 dev->data->mtu = mtu;
1179
1180         return err;
1181 }
1182
1183 static int
1184 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1185                      struct ether_addr *mc_addr_set __rte_unused,
1186                      uint32_t nb_mc_addr __rte_unused)
1187 {
1188         /*
1189          * Nothing to do actually: the tap has no filtering whatsoever, every
1190          * packet is received.
1191          */
1192         return 0;
1193 }
1194
1195 static int
1196 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1197 {
1198         struct rte_eth_dev *dev = arg;
1199         struct pmd_internals *pmd = dev->data->dev_private;
1200         struct ifinfomsg *info = NLMSG_DATA(nh);
1201
1202         if (nh->nlmsg_type != RTM_NEWLINK ||
1203             (info->ifi_index != pmd->if_index &&
1204              info->ifi_index != pmd->remote_if_index))
1205                 return 0;
1206         return tap_link_update(dev, 0);
1207 }
1208
1209 static void
1210 tap_dev_intr_handler(void *cb_arg)
1211 {
1212         struct rte_eth_dev *dev = cb_arg;
1213         struct pmd_internals *pmd = dev->data->dev_private;
1214
1215         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1216 }
1217
1218 static int
1219 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1220 {
1221         struct pmd_internals *pmd = dev->data->dev_private;
1222
1223         /* In any case, disable interrupt if the conf is no longer there. */
1224         if (!dev->data->dev_conf.intr_conf.lsc) {
1225                 if (pmd->intr_handle.fd != -1) {
1226                         tap_nl_final(pmd->intr_handle.fd);
1227                         rte_intr_callback_unregister(&pmd->intr_handle,
1228                                 tap_dev_intr_handler, dev);
1229                 }
1230                 return 0;
1231         }
1232         if (set) {
1233                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1234                 if (unlikely(pmd->intr_handle.fd == -1))
1235                         return -EBADF;
1236                 return rte_intr_callback_register(
1237                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1238         }
1239         tap_nl_final(pmd->intr_handle.fd);
1240         return rte_intr_callback_unregister(&pmd->intr_handle,
1241                                             tap_dev_intr_handler, dev);
1242 }
1243
1244 static int
1245 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1246 {
1247         int err;
1248
1249         err = tap_lsc_intr_handle_set(dev, set);
1250         if (err)
1251                 return err;
1252         err = tap_rx_intr_vec_set(dev, set);
1253         if (err && set)
1254                 tap_lsc_intr_handle_set(dev, 0);
1255         return err;
1256 }
1257
1258 static const uint32_t*
1259 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1260 {
1261         static const uint32_t ptypes[] = {
1262                 RTE_PTYPE_INNER_L2_ETHER,
1263                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1264                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1265                 RTE_PTYPE_INNER_L3_IPV4,
1266                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1267                 RTE_PTYPE_INNER_L3_IPV6,
1268                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1269                 RTE_PTYPE_INNER_L4_FRAG,
1270                 RTE_PTYPE_INNER_L4_UDP,
1271                 RTE_PTYPE_INNER_L4_TCP,
1272                 RTE_PTYPE_INNER_L4_SCTP,
1273                 RTE_PTYPE_L2_ETHER,
1274                 RTE_PTYPE_L2_ETHER_VLAN,
1275                 RTE_PTYPE_L2_ETHER_QINQ,
1276                 RTE_PTYPE_L3_IPV4,
1277                 RTE_PTYPE_L3_IPV4_EXT,
1278                 RTE_PTYPE_L3_IPV6_EXT,
1279                 RTE_PTYPE_L3_IPV6,
1280                 RTE_PTYPE_L4_FRAG,
1281                 RTE_PTYPE_L4_UDP,
1282                 RTE_PTYPE_L4_TCP,
1283                 RTE_PTYPE_L4_SCTP,
1284         };
1285
1286         return ptypes;
1287 }
1288
1289 static int
1290 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1291                   struct rte_eth_fc_conf *fc_conf)
1292 {
1293         fc_conf->mode = RTE_FC_NONE;
1294         return 0;
1295 }
1296
1297 static int
1298 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1299                   struct rte_eth_fc_conf *fc_conf)
1300 {
1301         if (fc_conf->mode != RTE_FC_NONE)
1302                 return -ENOTSUP;
1303         return 0;
1304 }
1305
1306 static const struct eth_dev_ops ops = {
1307         .dev_start              = tap_dev_start,
1308         .dev_stop               = tap_dev_stop,
1309         .dev_close              = tap_dev_close,
1310         .dev_configure          = tap_dev_configure,
1311         .dev_infos_get          = tap_dev_info,
1312         .rx_queue_setup         = tap_rx_queue_setup,
1313         .tx_queue_setup         = tap_tx_queue_setup,
1314         .rx_queue_release       = tap_rx_queue_release,
1315         .tx_queue_release       = tap_tx_queue_release,
1316         .flow_ctrl_get          = tap_flow_ctrl_get,
1317         .flow_ctrl_set          = tap_flow_ctrl_set,
1318         .link_update            = tap_link_update,
1319         .dev_set_link_up        = tap_link_set_up,
1320         .dev_set_link_down      = tap_link_set_down,
1321         .promiscuous_enable     = tap_promisc_enable,
1322         .promiscuous_disable    = tap_promisc_disable,
1323         .allmulticast_enable    = tap_allmulti_enable,
1324         .allmulticast_disable   = tap_allmulti_disable,
1325         .mac_addr_set           = tap_mac_set,
1326         .mtu_set                = tap_mtu_set,
1327         .set_mc_addr_list       = tap_set_mc_addr_list,
1328         .stats_get              = tap_stats_get,
1329         .stats_reset            = tap_stats_reset,
1330         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1331         .filter_ctrl            = tap_dev_filter_ctrl,
1332 };
1333
1334 static int
1335 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1336                    char *remote_iface, int fixed_mac_type)
1337 {
1338         int numa_node = rte_socket_id();
1339         struct rte_eth_dev *dev;
1340         struct pmd_internals *pmd;
1341         struct rte_eth_dev_data *data;
1342         struct ifreq ifr;
1343         int i;
1344
1345         RTE_LOG(DEBUG, PMD, "  TAP device on numa %u\n", rte_socket_id());
1346
1347         data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
1348         if (!data) {
1349                 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
1350                 goto error_exit;
1351         }
1352
1353         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1354         if (!dev) {
1355                 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
1356                 goto error_exit;
1357         }
1358
1359         pmd = dev->data->dev_private;
1360         pmd->dev = dev;
1361         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1362
1363         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1364         if (pmd->ioctl_sock == -1) {
1365                 RTE_LOG(ERR, PMD,
1366                         "TAP Unable to get a socket for management: %s\n",
1367                         strerror(errno));
1368                 goto error_exit;
1369         }
1370
1371         /* Setup some default values */
1372         rte_memcpy(data, dev->data, sizeof(*data));
1373         data->dev_private = pmd;
1374         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1375         data->numa_node = numa_node;
1376
1377         data->dev_link = pmd_link;
1378         data->mac_addrs = &pmd->eth_addr;
1379         /* Set the number of RX and TX queues */
1380         data->nb_rx_queues = 0;
1381         data->nb_tx_queues = 0;
1382
1383         dev->data = data;
1384         dev->dev_ops = &ops;
1385         dev->rx_pkt_burst = pmd_rx_burst;
1386         dev->tx_pkt_burst = pmd_tx_burst;
1387
1388         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1389         pmd->intr_handle.fd = -1;
1390         dev->intr_handle = &pmd->intr_handle;
1391
1392         /* Presetup the fds to -1 as being not valid */
1393         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1394                 pmd->rxq[i].fd = -1;
1395                 pmd->txq[i].fd = -1;
1396         }
1397
1398         if (fixed_mac_type) {
1399                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1400                 static int iface_idx;
1401                 char mac[ETHER_ADDR_LEN] = "\0dtap";
1402
1403                 mac[ETHER_ADDR_LEN - 1] = iface_idx++;
1404                 rte_memcpy(&pmd->eth_addr, mac, ETHER_ADDR_LEN);
1405         } else {
1406                 eth_random_addr((uint8_t *)&pmd->eth_addr);
1407         }
1408
1409         /* Immediately create the netdevice (this will create the 1st queue). */
1410         /* rx queue */
1411         if (tap_setup_queue(dev, pmd, 0, 1) == -1)
1412                 goto error_exit;
1413         /* tx queue */
1414         if (tap_setup_queue(dev, pmd, 0, 0) == -1)
1415                 goto error_exit;
1416
1417         ifr.ifr_mtu = dev->data->mtu;
1418         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1419                 goto error_exit;
1420
1421         memset(&ifr, 0, sizeof(struct ifreq));
1422         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1423         rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr, ETHER_ADDR_LEN);
1424         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1425                 goto error_exit;
1426
1427         /*
1428          * Set up everything related to rte_flow:
1429          * - netlink socket
1430          * - tap / remote if_index
1431          * - mandatory QDISCs
1432          * - rte_flow actual/implicit lists
1433          * - implicit rules
1434          */
1435         pmd->nlsk_fd = tap_nl_init(0);
1436         if (pmd->nlsk_fd == -1) {
1437                 RTE_LOG(WARNING, PMD, "%s: failed to create netlink socket.\n",
1438                         pmd->name);
1439                 goto disable_rte_flow;
1440         }
1441         pmd->if_index = if_nametoindex(pmd->name);
1442         if (!pmd->if_index) {
1443                 RTE_LOG(ERR, PMD, "%s: failed to get if_index.\n", pmd->name);
1444                 goto disable_rte_flow;
1445         }
1446         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1447                 RTE_LOG(ERR, PMD, "%s: failed to create multiq qdisc.\n",
1448                         pmd->name);
1449                 goto disable_rte_flow;
1450         }
1451         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1452                 RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1453                         pmd->name);
1454                 goto disable_rte_flow;
1455         }
1456         LIST_INIT(&pmd->flows);
1457
1458         if (strlen(remote_iface)) {
1459                 pmd->remote_if_index = if_nametoindex(remote_iface);
1460                 if (!pmd->remote_if_index) {
1461                         RTE_LOG(ERR, PMD, "%s: failed to get %s if_index.\n",
1462                                 pmd->name, remote_iface);
1463                         goto error_remote;
1464                 }
1465                 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1466                          "%s", remote_iface);
1467
1468                 /* Save state of remote device */
1469                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1470
1471                 /* Replicate remote MAC address */
1472                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1473                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1474                                 pmd->name, pmd->remote_iface);
1475                         goto error_remote;
1476                 }
1477                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1478                            ETHER_ADDR_LEN);
1479                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1480                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1481                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1482                                 pmd->name, remote_iface);
1483                         goto error_remote;
1484                 }
1485
1486                 /*
1487                  * Flush usually returns negative value because it tries to
1488                  * delete every QDISC (and on a running device, one QDISC at
1489                  * least is needed). Ignore negative return value.
1490                  */
1491                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1492                 if (qdisc_create_ingress(pmd->nlsk_fd,
1493                                          pmd->remote_if_index) < 0) {
1494                         RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1495                                 pmd->remote_iface);
1496                         goto error_remote;
1497                 }
1498                 LIST_INIT(&pmd->implicit_flows);
1499                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1500                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1501                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1502                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1503                         RTE_LOG(ERR, PMD,
1504                                 "%s: failed to create implicit rules.\n",
1505                                 pmd->name);
1506                         goto error_remote;
1507                 }
1508         }
1509
1510         return 0;
1511
1512 disable_rte_flow:
1513         RTE_LOG(ERR, PMD, " Disabling rte flow support: %s(%d)\n",
1514                 strerror(errno), errno);
1515         if (strlen(remote_iface)) {
1516                 RTE_LOG(ERR, PMD, "Remote feature requires flow support.\n");
1517                 goto error_exit;
1518         }
1519         return 0;
1520
1521 error_remote:
1522         RTE_LOG(ERR, PMD, " Can't set up remote feature: %s(%d)\n",
1523                 strerror(errno), errno);
1524         tap_flow_implicit_flush(pmd, NULL);
1525
1526 error_exit:
1527         RTE_LOG(ERR, PMD, "TAP Unable to initialize %s\n",
1528                 rte_vdev_device_name(vdev));
1529
1530         rte_free(data);
1531         return -EINVAL;
1532 }
1533
1534 static int
1535 set_interface_name(const char *key __rte_unused,
1536                    const char *value,
1537                    void *extra_args)
1538 {
1539         char *name = (char *)extra_args;
1540
1541         if (value)
1542                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
1543         else
1544                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1545                          DEFAULT_TAP_NAME, (tap_unit - 1));
1546
1547         return 0;
1548 }
1549
1550 static int
1551 set_remote_iface(const char *key __rte_unused,
1552                  const char *value,
1553                  void *extra_args)
1554 {
1555         char *name = (char *)extra_args;
1556
1557         if (value)
1558                 snprintf(name, RTE_ETH_NAME_MAX_LEN, "%s", value);
1559
1560         return 0;
1561 }
1562
1563 static int
1564 set_mac_type(const char *key __rte_unused,
1565              const char *value,
1566              void *extra_args)
1567 {
1568         if (value &&
1569             !strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED)))
1570                 *(int *)extra_args = 1;
1571         return 0;
1572 }
1573
1574 /* Open a TAP interface device.
1575  */
1576 static int
1577 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1578 {
1579         const char *name, *params;
1580         int ret;
1581         struct rte_kvargs *kvlist = NULL;
1582         int speed;
1583         char tap_name[RTE_ETH_NAME_MAX_LEN];
1584         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1585         int fixed_mac_type = 0;
1586
1587         name = rte_vdev_device_name(dev);
1588         params = rte_vdev_device_args(dev);
1589
1590         speed = ETH_SPEED_NUM_10G;
1591         snprintf(tap_name, sizeof(tap_name), "%s%d",
1592                  DEFAULT_TAP_NAME, tap_unit++);
1593         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1594
1595         if (params && (params[0] != '\0')) {
1596                 RTE_LOG(DEBUG, PMD, "parameters (%s)\n", params);
1597
1598                 kvlist = rte_kvargs_parse(params, valid_arguments);
1599                 if (kvlist) {
1600                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1601                                 ret = rte_kvargs_process(kvlist,
1602                                                          ETH_TAP_IFACE_ARG,
1603                                                          &set_interface_name,
1604                                                          tap_name);
1605                                 if (ret == -1)
1606                                         goto leave;
1607                         }
1608
1609                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1610                                 ret = rte_kvargs_process(kvlist,
1611                                                          ETH_TAP_REMOTE_ARG,
1612                                                          &set_remote_iface,
1613                                                          remote_iface);
1614                                 if (ret == -1)
1615                                         goto leave;
1616                         }
1617
1618                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1619                                 ret = rte_kvargs_process(kvlist,
1620                                                          ETH_TAP_MAC_ARG,
1621                                                          &set_mac_type,
1622                                                          &fixed_mac_type);
1623                                 if (ret == -1)
1624                                         goto leave;
1625                         }
1626                 }
1627         }
1628         pmd_link.link_speed = speed;
1629
1630         RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1631                 name, tap_name);
1632
1633         ret = eth_dev_tap_create(dev, tap_name, remote_iface, fixed_mac_type);
1634
1635 leave:
1636         if (ret == -1) {
1637                 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1638                         name, tap_name);
1639                 tap_unit--;             /* Restore the unit number */
1640         }
1641         rte_kvargs_free(kvlist);
1642
1643         return ret;
1644 }
1645
1646 /* detach a TAP device.
1647  */
1648 static int
1649 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1650 {
1651         struct rte_eth_dev *eth_dev = NULL;
1652         struct pmd_internals *internals;
1653         int i;
1654
1655         RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1656                 rte_socket_id());
1657
1658         /* find the ethdev entry */
1659         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1660         if (!eth_dev)
1661                 return 0;
1662
1663         internals = eth_dev->data->dev_private;
1664         if (internals->nlsk_fd) {
1665                 tap_flow_flush(eth_dev, NULL);
1666                 tap_flow_implicit_flush(internals, NULL);
1667                 tap_nl_final(internals->nlsk_fd);
1668         }
1669         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1670                 if (internals->rxq[i].fd != -1) {
1671                         close(internals->rxq[i].fd);
1672                         internals->rxq[i].fd = -1;
1673                 }
1674                 if (internals->txq[i].fd != -1) {
1675                         close(internals->txq[i].fd);
1676                         internals->txq[i].fd = -1;
1677                 }
1678         }
1679
1680         close(internals->ioctl_sock);
1681         rte_free(eth_dev->data->dev_private);
1682         rte_free(eth_dev->data);
1683
1684         rte_eth_dev_release_port(eth_dev);
1685
1686         return 0;
1687 }
1688
1689 static struct rte_vdev_driver pmd_tap_drv = {
1690         .probe = rte_pmd_tap_probe,
1691         .remove = rte_pmd_tap_remove,
1692 };
1693 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1694 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1695 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1696                               ETH_TAP_IFACE_ARG "=<string> "
1697                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_FIXED " "
1698                               ETH_TAP_REMOTE_ARG "=<string>");