drivers/net/tap/rte_eth_tap.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2016-2017 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <rte_atomic.h>
  35 #include <rte_branch_prediction.h>
  36 #include <rte_common.h>
  37 #include <rte_mbuf.h>
  38 #include <rte_ethdev.h>
  39 #include <rte_ethdev_vdev.h>
  40 #include <rte_malloc.h>
  41 #include <rte_vdev.h>
  42 #include <rte_kvargs.h>
  43 #include <rte_net.h>
  44 #include <rte_debug.h>
  45
  46 #include <sys/types.h>
  47 #include <sys/stat.h>
  48 #include <sys/socket.h>
  49 #include <sys/ioctl.h>
  50 #include <sys/utsname.h>
  51 #include <sys/mman.h>
  52 #include <errno.h>
  53 #include <signal.h>
  54 #include <stdint.h>
  55 #include <sys/uio.h>
  56 #include <unistd.h>
  57 #include <arpa/inet.h>
  58 #include <net/if.h>
  59 #include <linux/if_tun.h>
  60 #include <linux/if_ether.h>
  61 #include <linux/version.h>
  62 #include <fcntl.h>
  63
  64 #include <rte_eth_tap.h>
  65 #include <tap_flow.h>
  66 #include <tap_netlink.h>
  67 #include <tap_tcmsgs.h>
  68
  69 /* Linux based path to the TUN device */
  70 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  71 #define DEFAULT_TAP_NAME        "dtap"
  72
  73 #define ETH_TAP_IFACE_ARG       "iface"
  74 #define ETH_TAP_SPEED_ARG       "speed"
  75 #define ETH_TAP_REMOTE_ARG      "remote"
  76 #define ETH_TAP_MAC_ARG         "mac"
  77 #define ETH_TAP_MAC_FIXED       "fixed"
  78
  79 #define FLOWER_KERNEL_VERSION KERNEL_VERSION(4, 2, 0)
  80 #define FLOWER_VLAN_KERNEL_VERSION KERNEL_VERSION(4, 9, 0)
  81
  82 static struct rte_vdev_driver pmd_tap_drv;
  83
  84 static const char *valid_arguments[] = {
  85         ETH_TAP_IFACE_ARG,
  86         ETH_TAP_SPEED_ARG,
  87         ETH_TAP_REMOTE_ARG,
  88         ETH_TAP_MAC_ARG,
  89         NULL
  90 };
  91
  92 static int tap_unit;
  93
  94 static volatile uint32_t tap_trigger;   /* Rx trigger */
  95
  96 static struct rte_eth_link pmd_link = {
  97         .link_speed = ETH_SPEED_NUM_10G,
  98         .link_duplex = ETH_LINK_FULL_DUPLEX,
  99         .link_status = ETH_LINK_DOWN,
 100         .link_autoneg = ETH_LINK_SPEED_AUTONEG
 101 };
 102
 103 static void
 104 tap_trigger_cb(int sig __rte_unused)
 105 {
 106         /* Valid trigger values are nonzero */
 107         tap_trigger = (tap_trigger + 1) | 0x80000000;
 108 }
 109
 110 /* Specifies on what netdevices the ioctl should be applied */
 111 enum ioctl_mode {
 112         LOCAL_AND_REMOTE,
 113         LOCAL_ONLY,
 114         REMOTE_ONLY,
 115 };
 116
 117 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
 118
 119 /* Tun/Tap allocation routine
 120  *
 121  * name is the number of the interface to use, unless NULL to take the host
 122  * supplied name.
 123  */
 124 static int
 125 tun_alloc(struct pmd_internals *pmd)
 126 {
 127         struct ifreq ifr;
 128 #ifdef IFF_MULTI_QUEUE
 129         unsigned int features;
 130 #endif
 131         int fd;
 132
 133         memset(&ifr, 0, sizeof(struct ifreq));
 134
 135         /*
 136          * Do not set IFF_NO_PI as packet information header will be needed
 137          * to check if a received packet has been truncated.
 138          */
 139         ifr.ifr_flags = IFF_TAP;
 140         snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
 141
 142         RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
 143
 144         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 145         if (fd < 0) {
 146                 RTE_LOG(ERR, PMD, "Unable to create TAP interface");
 147                 goto error;
 148         }
 149
 150 #ifdef IFF_MULTI_QUEUE
 151         /* Grab the TUN features to verify we can work multi-queue */
 152         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 153                 RTE_LOG(ERR, PMD, "TAP unable to get TUN/TAP features\n");
 154                 goto error;
 155         }
 156         RTE_LOG(DEBUG, PMD, "  TAP Features %08x\n", features);
 157
 158         if (features & IFF_MULTI_QUEUE) {
 159                 RTE_LOG(DEBUG, PMD, "  Multi-queue support for %d queues\n",
 160                         RTE_PMD_TAP_MAX_QUEUES);
 161                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 162         } else
 163 #endif
 164         {
 165                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 166                 RTE_LOG(DEBUG, PMD, "  Single queue only support\n");
 167         }
 168
 169         /* Set the TUN/TAP configuration and set the name if needed */
 170         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 171                 RTE_LOG(WARNING, PMD,
 172                         "Unable to set TUNSETIFF for %s\n",
 173                         ifr.ifr_name);
 174                 perror("TUNSETIFF");
 175                 goto error;
 176         }
 177
 178         /* Always set the file descriptor to non-blocking */
 179         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 180                 RTE_LOG(WARNING, PMD,
 181                         "Unable to set %s to nonblocking\n",
 182                         ifr.ifr_name);
 183                 perror("F_SETFL, NONBLOCK");
 184                 goto error;
 185         }
 186
 187         /* Set up trigger to optimize empty Rx bursts */
 188         errno = 0;
 189         do {
 190                 struct sigaction sa;
 191                 int flags = fcntl(fd, F_GETFL);
 192
 193                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 194                         break;
 195                 if (sa.sa_handler != tap_trigger_cb) {
 196                         /*
 197                          * Make sure SIGIO is not already taken. This is done
 198                          * as late as possible to leave the application a
 199                          * chance to set up its own signal handler first.
 200                          */
 201                         if (sa.sa_handler != SIG_IGN &&
 202                             sa.sa_handler != SIG_DFL) {
 203                                 errno = EBUSY;
 204                                 break;
 205                         }
 206                         sa = (struct sigaction){
 207                                 .sa_flags = SA_RESTART,
 208                                 .sa_handler = tap_trigger_cb,
 209                         };
 210                         if (sigaction(SIGIO, &sa, NULL) == -1)
 211                                 break;
 212                 }
 213                 /* Enable SIGIO on file descriptor */
 214                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 215                 fcntl(fd, F_SETOWN, getpid());
 216         } while (0);
 217         if (errno) {
 218                 /* Disable trigger globally in case of error */
 219                 tap_trigger = 0;
 220                 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
 221                         strerror(errno));
 222         }
 223
 224         return fd;
 225
 226 error:
 227         if (fd > 0)
 228                 close(fd);
 229         return -1;
 230 }
 231
 232 /* Callback to handle the rx burst of packets to the correct interface and
 233  * file descriptor(s) in a multi-queue setup.
 234  */
 235 static uint16_t
 236 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 237 {
 238         struct rx_queue *rxq = queue;
 239         uint16_t num_rx;
 240         unsigned long num_rx_bytes = 0;
 241         uint32_t trigger = tap_trigger;
 242
 243         if (trigger == rxq->trigger_seen)
 244                 return 0;
 245         if (trigger)
 246                 rxq->trigger_seen = trigger;
 247         rte_compiler_barrier();
 248         for (num_rx = 0; num_rx < nb_pkts; ) {
 249                 struct rte_mbuf *mbuf = rxq->pool;
 250                 struct rte_mbuf *seg = NULL;
 251                 struct rte_mbuf *new_tail = NULL;
 252                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 253                 int len;
 254
 255                 len = readv(rxq->fd, *rxq->iovecs,
 256                             1 + (rxq->rxmode->enable_scatter ?
 257                                  rxq->nb_rx_desc : 1));
 258                 if (len < (int)sizeof(struct tun_pi))
 259                         break;
 260
 261                 /* Packet couldn't fit in the provided mbuf */
 262                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 263                         rxq->stats.ierrors++;
 264                         continue;
 265                 }
 266
 267                 len -= sizeof(struct tun_pi);
 268
 269                 mbuf->pkt_len = len;
 270                 mbuf->port = rxq->in_port;
 271                 while (1) {
 272                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 273
 274                         if (unlikely(!buf)) {
 275                                 rxq->stats.rx_nombuf++;
 276                                 /* No new buf has been allocated: do nothing */
 277                                 if (!new_tail || !seg)
 278                                         goto end;
 279
 280                                 seg->next = NULL;
 281                                 rte_pktmbuf_free(mbuf);
 282
 283                                 goto end;
 284                         }
 285                         seg = seg ? seg->next : mbuf;
 286                         if (rxq->pool == mbuf)
 287                                 rxq->pool = buf;
 288                         if (new_tail)
 289                                 new_tail->next = buf;
 290                         new_tail = buf;
 291                         new_tail->next = seg->next;
 292
 293                         /* iovecs[0] is reserved for packet info (pi) */
 294                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 295                                 buf->buf_len - data_off;
 296                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 297                                 (char *)buf->buf_addr + data_off;
 298
 299                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 300                         seg->data_off = data_off;
 301
 302                         len -= seg->data_len;
 303                         if (len <= 0)
 304                                 break;
 305                         mbuf->nb_segs++;
 306                         /* First segment has headroom, not the others */
 307                         data_off = 0;
 308                 }
 309                 seg->next = NULL;
 310                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 311                                                       RTE_PTYPE_ALL_MASK);
 312
 313                 /* account for the receive frame */
 314                 bufs[num_rx++] = mbuf;
 315                 num_rx_bytes += mbuf->pkt_len;
 316         }
 317 end:
 318         rxq->stats.ipackets += num_rx;
 319         rxq->stats.ibytes += num_rx_bytes;
 320
 321         return num_rx;
 322 }
 323
 324 /* Callback to handle sending packets from the tap interface
 325  */
 326 static uint16_t
 327 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 328 {
 329         struct tx_queue *txq = queue;
 330         uint16_t num_tx = 0;
 331         unsigned long num_tx_bytes = 0;
 332         uint32_t max_size;
 333         int i;
 334
 335         if (unlikely(nb_pkts == 0))
 336                 return 0;
 337
 338         max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
 339         for (i = 0; i < nb_pkts; i++) {
 340                 struct rte_mbuf *mbuf = bufs[num_tx];
 341                 struct iovec iovecs[mbuf->nb_segs + 1];
 342                 struct tun_pi pi = { .flags = 0 };
 343                 struct rte_mbuf *seg = mbuf;
 344                 int n;
 345                 int j;
 346
 347                 /* stats.errs will be incremented */
 348                 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
 349                         break;
 350
 351                 iovecs[0].iov_base = &pi;
 352                 iovecs[0].iov_len = sizeof(pi);
 353                 for (j = 1; j <= mbuf->nb_segs; j++) {
 354                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 355                         iovecs[j].iov_base =
 356                                 rte_pktmbuf_mtod(seg, void *);
 357                         seg = seg->next;
 358                 }
 359                 /* copy the tx frame data */
 360                 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
 361                 if (n <= 0)
 362                         break;
 363
 364                 num_tx++;
 365                 num_tx_bytes += mbuf->pkt_len;
 366                 rte_pktmbuf_free(mbuf);
 367         }
 368
 369         txq->stats.opackets += num_tx;
 370         txq->stats.errs += nb_pkts - num_tx;
 371         txq->stats.obytes += num_tx_bytes;
 372
 373         return num_tx;
 374 }
 375
 376 static const char *
 377 tap_ioctl_req2str(unsigned long request)
 378 {
 379         switch (request) {
 380         case SIOCSIFFLAGS:
 381                 return "SIOCSIFFLAGS";
 382         case SIOCGIFFLAGS:
 383                 return "SIOCGIFFLAGS";
 384         case SIOCGIFHWADDR:
 385                 return "SIOCGIFHWADDR";
 386         case SIOCSIFHWADDR:
 387                 return "SIOCSIFHWADDR";
 388         case SIOCSIFMTU:
 389                 return "SIOCSIFMTU";
 390         }
 391         return "UNKNOWN";
 392 }
 393
 394 static int
 395 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 396           struct ifreq *ifr, int set, enum ioctl_mode mode)
 397 {
 398         short req_flags = ifr->ifr_flags;
 399         int remote = pmd->remote_if_index &&
 400                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 401
 402         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 403                 return 0;
 404         /*
 405          * If there is a remote netdevice, apply ioctl on it, then apply it on
 406          * the tap netdevice.
 407          */
 408 apply:
 409         if (remote)
 410                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
 411         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 412                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 413         switch (request) {
 414         case SIOCSIFFLAGS:
 415                 /* fetch current flags to leave other flags untouched */
 416                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 417                         goto error;
 418                 if (set)
 419                         ifr->ifr_flags |= req_flags;
 420                 else
 421                         ifr->ifr_flags &= ~req_flags;
 422                 break;
 423         case SIOCGIFFLAGS:
 424         case SIOCGIFHWADDR:
 425         case SIOCSIFHWADDR:
 426         case SIOCSIFMTU:
 427                 break;
 428         default:
 429                 RTE_ASSERT(!"unsupported request type: must not happen");
 430         }
 431         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 432                 goto error;
 433         if (remote-- && mode == LOCAL_AND_REMOTE)
 434                 goto apply;
 435         return 0;
 436
 437 error:
 438         RTE_LOG(DEBUG, PMD, "%s: %s(%s) failed: %s(%d)\n", ifr->ifr_name,
 439                 __func__, tap_ioctl_req2str(request), strerror(errno), errno);
 440         return -errno;
 441 }
 442
 443 static int
 444 tap_link_set_down(struct rte_eth_dev *dev)
 445 {
 446         struct pmd_internals *pmd = dev->data->dev_private;
 447         struct ifreq ifr = { .ifr_flags = IFF_UP };
 448
 449         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 450         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 451 }
 452
 453 static int
 454 tap_link_set_up(struct rte_eth_dev *dev)
 455 {
 456         struct pmd_internals *pmd = dev->data->dev_private;
 457         struct ifreq ifr = { .ifr_flags = IFF_UP };
 458
 459         dev->data->dev_link.link_status = ETH_LINK_UP;
 460         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 461 }
 462
 463 static int
 464 tap_dev_start(struct rte_eth_dev *dev)
 465 {
 466         int err;
 467
 468         err = tap_intr_handle_set(dev, 1);
 469         if (err)
 470                 return err;
 471         return tap_link_set_up(dev);
 472 }
 473
 474 /* This function gets called when the current port gets stopped.
 475  */
 476 static void
 477 tap_dev_stop(struct rte_eth_dev *dev)
 478 {
 479         tap_intr_handle_set(dev, 0);
 480         tap_link_set_down(dev);
 481 }
 482
 483 static int
 484 tap_dev_configure(struct rte_eth_dev *dev __rte_unused)
 485 {
 486         return 0;
 487 }
 488
 489 static uint32_t
 490 tap_dev_speed_capa(void)
 491 {
 492         uint32_t speed = pmd_link.link_speed;
 493         uint32_t capa = 0;
 494
 495         if (speed >= ETH_SPEED_NUM_10M)
 496                 capa |= ETH_LINK_SPEED_10M;
 497         if (speed >= ETH_SPEED_NUM_100M)
 498                 capa |= ETH_LINK_SPEED_100M;
 499         if (speed >= ETH_SPEED_NUM_1G)
 500                 capa |= ETH_LINK_SPEED_1G;
 501         if (speed >= ETH_SPEED_NUM_5G)
 502                 capa |= ETH_LINK_SPEED_2_5G;
 503         if (speed >= ETH_SPEED_NUM_5G)
 504                 capa |= ETH_LINK_SPEED_5G;
 505         if (speed >= ETH_SPEED_NUM_10G)
 506                 capa |= ETH_LINK_SPEED_10G;
 507         if (speed >= ETH_SPEED_NUM_20G)
 508                 capa |= ETH_LINK_SPEED_20G;
 509         if (speed >= ETH_SPEED_NUM_25G)
 510                 capa |= ETH_LINK_SPEED_25G;
 511         if (speed >= ETH_SPEED_NUM_40G)
 512                 capa |= ETH_LINK_SPEED_40G;
 513         if (speed >= ETH_SPEED_NUM_50G)
 514                 capa |= ETH_LINK_SPEED_50G;
 515         if (speed >= ETH_SPEED_NUM_56G)
 516                 capa |= ETH_LINK_SPEED_56G;
 517         if (speed >= ETH_SPEED_NUM_100G)
 518                 capa |= ETH_LINK_SPEED_100G;
 519
 520         return capa;
 521 }
 522
 523 static void
 524 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 525 {
 526         struct pmd_internals *internals = dev->data->dev_private;
 527
 528         dev_info->if_index = internals->if_index;
 529         dev_info->max_mac_addrs = 1;
 530         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 531         dev_info->max_rx_queues = internals->nb_queues;
 532         dev_info->max_tx_queues = internals->nb_queues;
 533         dev_info->min_rx_bufsize = 0;
 534         dev_info->pci_dev = NULL;
 535         dev_info->speed_capa = tap_dev_speed_capa();
 536 }
 537
 538 static void
 539 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 540 {
 541         unsigned int i, imax;
 542         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 543         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 544         unsigned long rx_nombuf = 0, ierrors = 0;
 545         const struct pmd_internals *pmd = dev->data->dev_private;
 546
 547         imax = (pmd->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 548                 pmd->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 549
 550         for (i = 0; i < imax; i++) {
 551                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 552                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 553                 rx_total += tap_stats->q_ipackets[i];
 554                 rx_bytes_total += tap_stats->q_ibytes[i];
 555                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
 556                 ierrors += pmd->rxq[i].stats.ierrors;
 557
 558                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 559                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 560                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 561                 tx_total += tap_stats->q_opackets[i];
 562                 tx_err_total += tap_stats->q_errors[i];
 563                 tx_bytes_total += tap_stats->q_obytes[i];
 564         }
 565
 566         tap_stats->ipackets = rx_total;
 567         tap_stats->ibytes = rx_bytes_total;
 568         tap_stats->ierrors = ierrors;
 569         tap_stats->rx_nombuf = rx_nombuf;
 570         tap_stats->opackets = tx_total;
 571         tap_stats->oerrors = tx_err_total;
 572         tap_stats->obytes = tx_bytes_total;
 573 }
 574
 575 static void
 576 tap_stats_reset(struct rte_eth_dev *dev)
 577 {
 578         int i;
 579         struct pmd_internals *pmd = dev->data->dev_private;
 580
 581         for (i = 0; i < pmd->nb_queues; i++) {
 582                 pmd->rxq[i].stats.ipackets = 0;
 583                 pmd->rxq[i].stats.ibytes = 0;
 584                 pmd->rxq[i].stats.ierrors = 0;
 585                 pmd->rxq[i].stats.rx_nombuf = 0;
 586
 587                 pmd->txq[i].stats.opackets = 0;
 588                 pmd->txq[i].stats.errs = 0;
 589                 pmd->txq[i].stats.obytes = 0;
 590         }
 591 }
 592
 593 static void
 594 tap_dev_close(struct rte_eth_dev *dev)
 595 {
 596         int i;
 597         struct pmd_internals *internals = dev->data->dev_private;
 598
 599         tap_link_set_down(dev);
 600         tap_flow_flush(dev, NULL);
 601         tap_flow_implicit_flush(internals, NULL);
 602
 603         for (i = 0; i < internals->nb_queues; i++) {
 604                 if (internals->rxq[i].fd != -1)
 605                         close(internals->rxq[i].fd);
 606                 internals->rxq[i].fd = -1;
 607                 internals->txq[i].fd = -1;
 608         }
 609 }
 610
 611 static void
 612 tap_rx_queue_release(void *queue)
 613 {
 614         struct rx_queue *rxq = queue;
 615
 616         if (rxq && (rxq->fd > 0)) {
 617                 close(rxq->fd);
 618                 rxq->fd = -1;
 619                 rte_pktmbuf_free(rxq->pool);
 620                 rte_free(rxq->iovecs);
 621                 rxq->pool = NULL;
 622                 rxq->iovecs = NULL;
 623         }
 624 }
 625
 626 static void
 627 tap_tx_queue_release(void *queue)
 628 {
 629         struct tx_queue *txq = queue;
 630
 631         if (txq && (txq->fd > 0)) {
 632                 close(txq->fd);
 633                 txq->fd = -1;
 634         }
 635 }
 636
 637 static int
 638 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
 639 {
 640         struct rte_eth_link *dev_link = &dev->data->dev_link;
 641         struct pmd_internals *pmd = dev->data->dev_private;
 642         struct ifreq ifr = { .ifr_flags = 0 };
 643
 644         if (pmd->remote_if_index) {
 645                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
 646                 if (!(ifr.ifr_flags & IFF_UP) ||
 647                     !(ifr.ifr_flags & IFF_RUNNING)) {
 648                         dev_link->link_status = ETH_LINK_DOWN;
 649                         return 0;
 650                 }
 651         }
 652         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
 653         dev_link->link_status =
 654                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
 655                  ETH_LINK_UP :
 656                  ETH_LINK_DOWN);
 657         return 0;
 658 }
 659
 660 static void
 661 tap_promisc_enable(struct rte_eth_dev *dev)
 662 {
 663         struct pmd_internals *pmd = dev->data->dev_private;
 664         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 665
 666         dev->data->promiscuous = 1;
 667         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 668         if (pmd->remote_if_index)
 669                 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
 670 }
 671
 672 static void
 673 tap_promisc_disable(struct rte_eth_dev *dev)
 674 {
 675         struct pmd_internals *pmd = dev->data->dev_private;
 676         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 677
 678         dev->data->promiscuous = 0;
 679         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 680         if (pmd->remote_if_index)
 681                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
 682 }
 683
 684 static void
 685 tap_allmulti_enable(struct rte_eth_dev *dev)
 686 {
 687         struct pmd_internals *pmd = dev->data->dev_private;
 688         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 689
 690         dev->data->all_multicast = 1;
 691         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 692         if (pmd->remote_if_index)
 693                 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
 694 }
 695
 696 static void
 697 tap_allmulti_disable(struct rte_eth_dev *dev)
 698 {
 699         struct pmd_internals *pmd = dev->data->dev_private;
 700         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 701
 702         dev->data->all_multicast = 0;
 703         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 704         if (pmd->remote_if_index)
 705                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
 706 }
 707
 708 static void
 709 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 710 {
 711         struct pmd_internals *pmd = dev->data->dev_private;
 712         enum ioctl_mode mode = LOCAL_ONLY;
 713         struct ifreq ifr;
 714
 715         if (is_zero_ether_addr(mac_addr)) {
 716                 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
 717                         dev->data->name);
 718                 return;
 719         }
 720         /* Check the actual current MAC address on the tap netdevice */
 721         if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
 722                 return;
 723         if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 724                                mac_addr))
 725                 return;
 726         /* Check the current MAC address on the remote */
 727         if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0)
 728                 return;
 729         if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 730                                mac_addr))
 731                 mode = LOCAL_AND_REMOTE;
 732         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
 733         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
 734         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode) < 0)
 735                 return;
 736         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
 737         if (pmd->remote_if_index) {
 738                 /* Replace MAC redirection rule after a MAC change */
 739                 if (tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC) < 0) {
 740                         RTE_LOG(ERR, PMD,
 741                                 "%s: Couldn't delete MAC redirection rule\n",
 742                                 dev->data->name);
 743                         return;
 744                 }
 745                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0)
 746                         RTE_LOG(ERR, PMD,
 747                                 "%s: Couldn't add MAC redirection rule\n",
 748                                 dev->data->name);
 749         }
 750 }
 751
 752 static int
 753 tap_setup_queue(struct rte_eth_dev *dev,
 754                 struct pmd_internals *internals,
 755                 uint16_t qid)
 756 {
 757         struct pmd_internals *pmd = dev->data->dev_private;
 758         struct rx_queue *rx = &internals->rxq[qid];
 759         struct tx_queue *tx = &internals->txq[qid];
 760         int fd = rx->fd == -1 ? tx->fd : rx->fd;
 761
 762         if (fd == -1) {
 763                 RTE_LOG(INFO, PMD, "Add queue to TAP %s for qid %d\n",
 764                         pmd->name, qid);
 765                 fd = tun_alloc(pmd);
 766                 if (fd < 0) {
 767                         RTE_LOG(ERR, PMD, "%s: tun_alloc() failed.\n",
 768                                 pmd->name);
 769                         return -1;
 770                 }
 771         }
 772
 773         rx->fd = fd;
 774         tx->fd = fd;
 775         tx->mtu = &dev->data->mtu;
 776         rx->rxmode = &dev->data->dev_conf.rxmode;
 777
 778         return fd;
 779 }
 780
 781 static int
 782 tap_rx_queue_setup(struct rte_eth_dev *dev,
 783                    uint16_t rx_queue_id,
 784                    uint16_t nb_rx_desc,
 785                    unsigned int socket_id,
 786                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 787                    struct rte_mempool *mp)
 788 {
 789         struct pmd_internals *internals = dev->data->dev_private;
 790         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
 791         struct rte_mbuf **tmp = &rxq->pool;
 792         long iov_max = sysconf(_SC_IOV_MAX);
 793         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
 794         struct iovec (*iovecs)[nb_desc + 1];
 795         int data_off = RTE_PKTMBUF_HEADROOM;
 796         int ret = 0;
 797         int fd;
 798         int i;
 799
 800         if ((rx_queue_id >= internals->nb_queues) || !mp) {
 801                 RTE_LOG(WARNING, PMD,
 802                         "nb_queues %d too small or mempool NULL\n",
 803                         internals->nb_queues);
 804                 return -1;
 805         }
 806
 807         rxq->mp = mp;
 808         rxq->trigger_seen = 1; /* force initial burst */
 809         rxq->in_port = dev->data->port_id;
 810         rxq->nb_rx_desc = nb_desc;
 811         iovecs = rte_zmalloc_socket(dev->data->name, sizeof(*iovecs), 0,
 812                                     socket_id);
 813         if (!iovecs) {
 814                 RTE_LOG(WARNING, PMD,
 815                         "%s: Couldn't allocate %d RX descriptors\n",
 816                         dev->data->name, nb_desc);
 817                 return -ENOMEM;
 818         }
 819         rxq->iovecs = iovecs;
 820
 821         dev->data->rx_queues[rx_queue_id] = rxq;
 822         fd = tap_setup_queue(dev, internals, rx_queue_id);
 823         if (fd == -1) {
 824                 ret = fd;
 825                 goto error;
 826         }
 827
 828         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
 829         (*rxq->iovecs)[0].iov_base = &rxq->pi;
 830
 831         for (i = 1; i <= nb_desc; i++) {
 832                 *tmp = rte_pktmbuf_alloc(rxq->mp);
 833                 if (!*tmp) {
 834                         RTE_LOG(WARNING, PMD,
 835                                 "%s: couldn't allocate memory for queue %d\n",
 836                                 dev->data->name, rx_queue_id);
 837                         ret = -ENOMEM;
 838                         goto error;
 839                 }
 840                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
 841                 (*rxq->iovecs)[i].iov_base =
 842                         (char *)(*tmp)->buf_addr + data_off;
 843                 data_off = 0;
 844                 tmp = &(*tmp)->next;
 845         }
 846
 847         RTE_LOG(DEBUG, PMD, "  RX TAP device name %s, qid %d on fd %d\n",
 848                 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
 849
 850         return 0;
 851
 852 error:
 853         rte_pktmbuf_free(rxq->pool);
 854         rxq->pool = NULL;
 855         rte_free(rxq->iovecs);
 856         rxq->iovecs = NULL;
 857         return ret;
 858 }
 859
 860 static int
 861 tap_tx_queue_setup(struct rte_eth_dev *dev,
 862                    uint16_t tx_queue_id,
 863                    uint16_t nb_tx_desc __rte_unused,
 864                    unsigned int socket_id __rte_unused,
 865                    const struct rte_eth_txconf *tx_conf __rte_unused)
 866 {
 867         struct pmd_internals *internals = dev->data->dev_private;
 868         int ret;
 869
 870         if (tx_queue_id >= internals->nb_queues)
 871                 return -1;
 872
 873         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
 874         ret = tap_setup_queue(dev, internals, tx_queue_id);
 875         if (ret == -1)
 876                 return -1;
 877
 878         RTE_LOG(DEBUG, PMD, "  TX TAP device name %s, qid %d on fd %d\n",
 879                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd);
 880
 881         return 0;
 882 }
 883
 884 static int
 885 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 886 {
 887         struct pmd_internals *pmd = dev->data->dev_private;
 888         struct ifreq ifr = { .ifr_mtu = mtu };
 889         int err = 0;
 890
 891         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
 892         if (!err)
 893                 dev->data->mtu = mtu;
 894
 895         return err;
 896 }
 897
 898 static int
 899 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
 900                      struct ether_addr *mc_addr_set __rte_unused,
 901                      uint32_t nb_mc_addr __rte_unused)
 902 {
 903         /*
 904          * Nothing to do actually: the tap has no filtering whatsoever, every
 905          * packet is received.
 906          */
 907         return 0;
 908 }
 909
 910 static int
 911 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
 912 {
 913         struct rte_eth_dev *dev = arg;
 914         struct pmd_internals *pmd = dev->data->dev_private;
 915         struct ifinfomsg *info = NLMSG_DATA(nh);
 916
 917         if (nh->nlmsg_type != RTM_NEWLINK ||
 918             (info->ifi_index != pmd->if_index &&
 919              info->ifi_index != pmd->remote_if_index))
 920                 return 0;
 921         return tap_link_update(dev, 0);
 922 }
 923
 924 static void
 925 tap_dev_intr_handler(void *cb_arg)
 926 {
 927         struct rte_eth_dev *dev = cb_arg;
 928         struct pmd_internals *pmd = dev->data->dev_private;
 929
 930         nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
 931 }
 932
 933 static int
 934 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
 935 {
 936         struct pmd_internals *pmd = dev->data->dev_private;
 937
 938         /* In any case, disable interrupt if the conf is no longer there. */
 939         if (!dev->data->dev_conf.intr_conf.lsc) {
 940                 if (pmd->intr_handle.fd != -1)
 941                         nl_final(pmd->intr_handle.fd);
 942                 rte_intr_callback_unregister(
 943                         &pmd->intr_handle, tap_dev_intr_handler, dev);
 944                 return 0;
 945         }
 946         if (set) {
 947                 pmd->intr_handle.fd = nl_init(RTMGRP_LINK);
 948                 if (unlikely(pmd->intr_handle.fd == -1))
 949                         return -EBADF;
 950                 return rte_intr_callback_register(
 951                         &pmd->intr_handle, tap_dev_intr_handler, dev);
 952         }
 953         nl_final(pmd->intr_handle.fd);
 954         return rte_intr_callback_unregister(&pmd->intr_handle,
 955                                             tap_dev_intr_handler, dev);
 956 }
 957
 958 static const uint32_t*
 959 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
 960 {
 961         static const uint32_t ptypes[] = {
 962                 RTE_PTYPE_INNER_L2_ETHER,
 963                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
 964                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
 965                 RTE_PTYPE_INNER_L3_IPV4,
 966                 RTE_PTYPE_INNER_L3_IPV4_EXT,
 967                 RTE_PTYPE_INNER_L3_IPV6,
 968                 RTE_PTYPE_INNER_L3_IPV6_EXT,
 969                 RTE_PTYPE_INNER_L4_FRAG,
 970                 RTE_PTYPE_INNER_L4_UDP,
 971                 RTE_PTYPE_INNER_L4_TCP,
 972                 RTE_PTYPE_INNER_L4_SCTP,
 973                 RTE_PTYPE_L2_ETHER,
 974                 RTE_PTYPE_L2_ETHER_VLAN,
 975                 RTE_PTYPE_L2_ETHER_QINQ,
 976                 RTE_PTYPE_L3_IPV4,
 977                 RTE_PTYPE_L3_IPV4_EXT,
 978                 RTE_PTYPE_L3_IPV6_EXT,
 979                 RTE_PTYPE_L3_IPV6,
 980                 RTE_PTYPE_L4_FRAG,
 981                 RTE_PTYPE_L4_UDP,
 982                 RTE_PTYPE_L4_TCP,
 983                 RTE_PTYPE_L4_SCTP,
 984         };
 985
 986         return ptypes;
 987 }
 988
 989 static int
 990 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
 991                   struct rte_eth_fc_conf *fc_conf)
 992 {
 993         fc_conf->mode = RTE_FC_NONE;
 994         return 0;
 995 }
 996
 997 static int
 998 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
 999                   struct rte_eth_fc_conf *fc_conf)
1000 {
1001         if (fc_conf->mode != RTE_FC_NONE)
1002                 return -ENOTSUP;
1003         return 0;
1004 }
1005
1006 static const struct eth_dev_ops ops = {
1007         .dev_start              = tap_dev_start,
1008         .dev_stop               = tap_dev_stop,
1009         .dev_close              = tap_dev_close,
1010         .dev_configure          = tap_dev_configure,
1011         .dev_infos_get          = tap_dev_info,
1012         .rx_queue_setup         = tap_rx_queue_setup,
1013         .tx_queue_setup         = tap_tx_queue_setup,
1014         .rx_queue_release       = tap_rx_queue_release,
1015         .tx_queue_release       = tap_tx_queue_release,
1016         .flow_ctrl_get          = tap_flow_ctrl_get,
1017         .flow_ctrl_set          = tap_flow_ctrl_set,
1018         .link_update            = tap_link_update,
1019         .dev_set_link_up        = tap_link_set_up,
1020         .dev_set_link_down      = tap_link_set_down,
1021         .promiscuous_enable     = tap_promisc_enable,
1022         .promiscuous_disable    = tap_promisc_disable,
1023         .allmulticast_enable    = tap_allmulti_enable,
1024         .allmulticast_disable   = tap_allmulti_disable,
1025         .mac_addr_set           = tap_mac_set,
1026         .mtu_set                = tap_mtu_set,
1027         .set_mc_addr_list       = tap_set_mc_addr_list,
1028         .stats_get              = tap_stats_get,
1029         .stats_reset            = tap_stats_reset,
1030         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1031         .filter_ctrl            = tap_dev_filter_ctrl,
1032 };
1033
1034 static int
1035 tap_kernel_support(struct pmd_internals *pmd)
1036 {
1037         struct utsname utsname;
1038         int ver[3];
1039
1040         if (uname(&utsname) == -1 ||
1041             sscanf(utsname.release, "%d.%d.%d",
1042                    &ver[0], &ver[1], &ver[2]) != 3)
1043                 return 0;
1044         if (KERNEL_VERSION(ver[0], ver[1], ver[2]) >= FLOWER_KERNEL_VERSION)
1045                 pmd->flower_support = 1;
1046         if (KERNEL_VERSION(ver[0], ver[1], ver[2]) >=
1047             FLOWER_VLAN_KERNEL_VERSION)
1048                 pmd->flower_vlan_support = 1;
1049         return 1;
1050 }
1051
1052 static int
1053 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1054                    char *remote_iface, int fixed_mac_type)
1055 {
1056         int numa_node = rte_socket_id();
1057         struct rte_eth_dev *dev;
1058         struct pmd_internals *pmd;
1059         struct rte_eth_dev_data *data;
1060         struct ifreq ifr;
1061         int i;
1062
1063         RTE_LOG(DEBUG, PMD, "  TAP device on numa %u\n", rte_socket_id());
1064
1065         data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
1066         if (!data) {
1067                 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
1068                 goto error_exit;
1069         }
1070
1071         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1072         if (!dev) {
1073                 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
1074                 goto error_exit;
1075         }
1076
1077         pmd = dev->data->dev_private;
1078         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1079         pmd->nb_queues = RTE_PMD_TAP_MAX_QUEUES;
1080
1081         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1082         if (pmd->ioctl_sock == -1) {
1083                 RTE_LOG(ERR, PMD,
1084                         "TAP Unable to get a socket for management: %s\n",
1085                         strerror(errno));
1086                 goto error_exit;
1087         }
1088
1089         /* Setup some default values */
1090         rte_memcpy(data, dev->data, sizeof(*data));
1091         data->dev_private = pmd;
1092         data->dev_flags = RTE_ETH_DEV_DETACHABLE | RTE_ETH_DEV_INTR_LSC;
1093         data->numa_node = numa_node;
1094         data->drv_name = pmd_tap_drv.driver.name;
1095
1096         data->dev_link = pmd_link;
1097         data->mac_addrs = &pmd->eth_addr;
1098         data->nb_rx_queues = pmd->nb_queues;
1099         data->nb_tx_queues = pmd->nb_queues;
1100
1101         dev->data = data;
1102         dev->dev_ops = &ops;
1103         dev->rx_pkt_burst = pmd_rx_burst;
1104         dev->tx_pkt_burst = pmd_tx_burst;
1105
1106         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1107         pmd->intr_handle.fd = -1;
1108
1109         /* Presetup the fds to -1 as being not valid */
1110         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1111                 pmd->rxq[i].fd = -1;
1112                 pmd->txq[i].fd = -1;
1113         }
1114
1115         if (fixed_mac_type) {
1116                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1117                 static int iface_idx;
1118                 char mac[ETHER_ADDR_LEN] = "\0dtap";
1119
1120                 mac[ETHER_ADDR_LEN - 1] = iface_idx++;
1121                 rte_memcpy(&pmd->eth_addr, mac, ETHER_ADDR_LEN);
1122         } else {
1123                 eth_random_addr((uint8_t *)&pmd->eth_addr);
1124         }
1125
1126         /* Immediately create the netdevice (this will create the 1st queue). */
1127         if (tap_setup_queue(dev, pmd, 0) == -1)
1128                 goto error_exit;
1129
1130         ifr.ifr_mtu = dev->data->mtu;
1131         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1132                 goto error_exit;
1133
1134         memset(&ifr, 0, sizeof(struct ifreq));
1135         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1136         rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr, ETHER_ADDR_LEN);
1137         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1138                 goto error_exit;
1139
1140         tap_kernel_support(pmd);
1141         if (!pmd->flower_support) {
1142                 if (remote_iface[0]) {
1143                         RTE_LOG(ERR, PMD,
1144                                 "%s: kernel does not support TC rules, required for remote feature.",
1145                                 pmd->name);
1146                         goto error_exit;
1147                 } else {
1148                         RTE_LOG(INFO, PMD,
1149                                 "%s: kernel too old for Flow API support.\n",
1150                                 pmd->name);
1151                         return 0;
1152                 }
1153         }
1154
1155         /*
1156          * Set up everything related to rte_flow:
1157          * - netlink socket
1158          * - tap / remote if_index
1159          * - mandatory QDISCs
1160          * - rte_flow actual/implicit lists
1161          * - implicit rules
1162          */
1163         pmd->nlsk_fd = nl_init(0);
1164         if (pmd->nlsk_fd == -1) {
1165                 RTE_LOG(WARNING, PMD, "%s: failed to create netlink socket.",
1166                         pmd->name);
1167                 goto disable_rte_flow;
1168         }
1169         pmd->if_index = if_nametoindex(pmd->name);
1170         if (!pmd->if_index) {
1171                 RTE_LOG(ERR, PMD, "%s: failed to get if_index.", pmd->name);
1172                 goto disable_rte_flow;
1173         }
1174         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1175                 RTE_LOG(ERR, PMD, "%s: failed to create multiq qdisc.",
1176                         pmd->name);
1177                 goto disable_rte_flow;
1178         }
1179         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1180                 RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.",
1181                         pmd->name);
1182                 goto disable_rte_flow;
1183         }
1184         LIST_INIT(&pmd->flows);
1185
1186         if (strlen(remote_iface)) {
1187                 pmd->remote_if_index = if_nametoindex(remote_iface);
1188                 if (!pmd->remote_if_index) {
1189                         RTE_LOG(ERR, PMD, "%s: failed to get %s if_index.",
1190                                 pmd->name, remote_iface);
1191                         goto error_remote;
1192                 }
1193                 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1194                          "%s", remote_iface);
1195                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1196                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.",
1197                                 pmd->name, pmd->remote_iface);
1198                         goto error_remote;
1199                 }
1200                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1201                            ETHER_ADDR_LEN);
1202                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1203                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1204                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.",
1205                                 pmd->name, remote_iface);
1206                         goto error_remote;
1207                 }
1208
1209                 /*
1210                  * Flush usually returns negative value because it tries to
1211                  * delete every QDISC (and on a running device, one QDISC at
1212                  * least is needed). Ignore negative return value.
1213                  */
1214                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1215                 if (qdisc_create_ingress(pmd->nlsk_fd,
1216                                          pmd->remote_if_index) < 0) {
1217                         RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.",
1218                                 pmd->remote_iface);
1219                         goto error_remote;
1220                 }
1221                 LIST_INIT(&pmd->implicit_flows);
1222                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1223                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1224                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1225                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1226                         RTE_LOG(ERR, PMD,
1227                                 "%s: failed to create implicit rules.",
1228                                 pmd->name);
1229                         goto error_remote;
1230                 }
1231         }
1232
1233         return 0;
1234
1235 disable_rte_flow:
1236         RTE_LOG(ERR, PMD, " Disabling rte flow support: %s(%d)\n",
1237                 strerror(errno), errno);
1238         if (strlen(remote_iface)) {
1239                 RTE_LOG(ERR, PMD, "Remote feature requires flow support.\n");
1240                 goto error_exit;
1241         }
1242         pmd->flower_support = 0;
1243         return 0;
1244
1245 error_remote:
1246         RTE_LOG(ERR, PMD, " Can't set up remote feature: %s(%d)\n",
1247                 strerror(errno), errno);
1248         tap_flow_implicit_flush(pmd, NULL);
1249
1250 error_exit:
1251         RTE_LOG(ERR, PMD, "TAP Unable to initialize %s\n",
1252                 rte_vdev_device_name(vdev));
1253
1254         rte_free(data);
1255         return -EINVAL;
1256 }
1257
1258 static int
1259 set_interface_name(const char *key __rte_unused,
1260                    const char *value,
1261                    void *extra_args)
1262 {
1263         char *name = (char *)extra_args;
1264
1265         if (value)
1266                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
1267         else
1268                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1269                          DEFAULT_TAP_NAME, (tap_unit - 1));
1270
1271         return 0;
1272 }
1273
1274 static int
1275 set_interface_speed(const char *key __rte_unused,
1276                     const char *value,
1277                     void *extra_args)
1278 {
1279         *(int *)extra_args = (value) ? atoi(value) : ETH_SPEED_NUM_10G;
1280
1281         return 0;
1282 }
1283
1284 static int
1285 set_remote_iface(const char *key __rte_unused,
1286                  const char *value,
1287                  void *extra_args)
1288 {
1289         char *name = (char *)extra_args;
1290
1291         if (value)
1292                 snprintf(name, RTE_ETH_NAME_MAX_LEN, "%s", value);
1293
1294         return 0;
1295 }
1296
1297 static int
1298 set_mac_type(const char *key __rte_unused,
1299              const char *value,
1300              void *extra_args)
1301 {
1302         if (value &&
1303             !strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED)))
1304                 *(int *)extra_args = 1;
1305         return 0;
1306 }
1307
1308 /* Open a TAP interface device.
1309  */
1310 static int
1311 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1312 {
1313         const char *name, *params;
1314         int ret;
1315         struct rte_kvargs *kvlist = NULL;
1316         int speed;
1317         char tap_name[RTE_ETH_NAME_MAX_LEN];
1318         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1319         int fixed_mac_type = 0;
1320
1321         name = rte_vdev_device_name(dev);
1322         params = rte_vdev_device_args(dev);
1323
1324         speed = ETH_SPEED_NUM_10G;
1325         snprintf(tap_name, sizeof(tap_name), "%s%d",
1326                  DEFAULT_TAP_NAME, tap_unit++);
1327         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1328
1329         if (params && (params[0] != '\0')) {
1330                 RTE_LOG(DEBUG, PMD, "paramaters (%s)\n", params);
1331
1332                 kvlist = rte_kvargs_parse(params, valid_arguments);
1333                 if (kvlist) {
1334                         if (rte_kvargs_count(kvlist, ETH_TAP_SPEED_ARG) == 1) {
1335                                 ret = rte_kvargs_process(kvlist,
1336                                                          ETH_TAP_SPEED_ARG,
1337                                                          &set_interface_speed,
1338                                                          &speed);
1339                                 if (ret == -1)
1340                                         goto leave;
1341                         }
1342
1343                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1344                                 ret = rte_kvargs_process(kvlist,
1345                                                          ETH_TAP_IFACE_ARG,
1346                                                          &set_interface_name,
1347                                                          tap_name);
1348                                 if (ret == -1)
1349                                         goto leave;
1350                         }
1351
1352                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1353                                 ret = rte_kvargs_process(kvlist,
1354                                                          ETH_TAP_REMOTE_ARG,
1355                                                          &set_remote_iface,
1356                                                          remote_iface);
1357                                 if (ret == -1)
1358                                         goto leave;
1359                         }
1360
1361                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1362                                 ret = rte_kvargs_process(kvlist,
1363                                                          ETH_TAP_MAC_ARG,
1364                                                          &set_mac_type,
1365                                                          &fixed_mac_type);
1366                                 if (ret == -1)
1367                                         goto leave;
1368                         }
1369                 }
1370         }
1371         pmd_link.link_speed = speed;
1372
1373         RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1374                 name, tap_name);
1375
1376         ret = eth_dev_tap_create(dev, tap_name, remote_iface, fixed_mac_type);
1377
1378 leave:
1379         if (ret == -1) {
1380                 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1381                         name, tap_name);
1382                 tap_unit--;             /* Restore the unit number */
1383         }
1384         rte_kvargs_free(kvlist);
1385
1386         return ret;
1387 }
1388
1389 /* detach a TAP device.
1390  */
1391 static int
1392 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1393 {
1394         struct rte_eth_dev *eth_dev = NULL;
1395         struct pmd_internals *internals;
1396         int i;
1397
1398         RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1399                 rte_socket_id());
1400
1401         /* find the ethdev entry */
1402         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1403         if (!eth_dev)
1404                 return 0;
1405
1406         internals = eth_dev->data->dev_private;
1407         if (internals->flower_support && internals->nlsk_fd) {
1408                 tap_flow_flush(eth_dev, NULL);
1409                 tap_flow_implicit_flush(internals, NULL);
1410                 nl_final(internals->nlsk_fd);
1411         }
1412         for (i = 0; i < internals->nb_queues; i++)
1413                 if (internals->rxq[i].fd != -1)
1414                         close(internals->rxq[i].fd);
1415
1416         close(internals->ioctl_sock);
1417         rte_free(eth_dev->data->dev_private);
1418         rte_free(eth_dev->data);
1419
1420         rte_eth_dev_release_port(eth_dev);
1421
1422         return 0;
1423 }
1424
1425 static struct rte_vdev_driver pmd_tap_drv = {
1426         .probe = rte_pmd_tap_probe,
1427         .remove = rte_pmd_tap_remove,
1428 };
1429 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1430 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1431 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1432                               ETH_TAP_IFACE_ARG "=<string> "
1433                               ETH_TAP_SPEED_ARG "=<int> "
1434                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_FIXED " "
1435                               ETH_TAP_REMOTE_ARG "=<string>");