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 __rte_unused)
 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
 709 static void
 710 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 711 {
 712         struct pmd_internals *pmd = dev->data->dev_private;
 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
 727         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
 728         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
 729         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, LOCAL_AND_REMOTE) < 0)
 730                 return;
 731         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
 732         if (pmd->remote_if_index) {
 733                 /* Replace MAC redirection rule after a MAC change */
 734                 if (tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC) < 0) {
 735                         RTE_LOG(ERR, PMD,
 736                                 "%s: Couldn't delete MAC redirection rule\n",
 737                                 dev->data->name);
 738                         return;
 739                 }
 740                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0)
 741                         RTE_LOG(ERR, PMD,
 742                                 "%s: Couldn't add MAC redirection rule\n",
 743                                 dev->data->name);
 744         }
 745 }
 746
 747 static int
 748 tap_setup_queue(struct rte_eth_dev *dev,
 749                 struct pmd_internals *internals,
 750                 uint16_t qid)
 751 {
 752         struct pmd_internals *pmd = dev->data->dev_private;
 753         struct rx_queue *rx = &internals->rxq[qid];
 754         struct tx_queue *tx = &internals->txq[qid];
 755         int fd = rx->fd == -1 ? tx->fd : rx->fd;
 756
 757         if (fd == -1) {
 758                 RTE_LOG(INFO, PMD, "Add queue to TAP %s for qid %d\n",
 759                         pmd->name, qid);
 760                 fd = tun_alloc(pmd);
 761                 if (fd < 0) {
 762                         RTE_LOG(ERR, PMD, "%s: tun_alloc() failed.\n",
 763                                 pmd->name);
 764                         return -1;
 765                 }
 766         }
 767
 768         rx->fd = fd;
 769         tx->fd = fd;
 770         tx->mtu = &dev->data->mtu;
 771         rx->rxmode = &dev->data->dev_conf.rxmode;
 772
 773         return fd;
 774 }
 775
 776 static int
 777 tap_rx_queue_setup(struct rte_eth_dev *dev,
 778                    uint16_t rx_queue_id,
 779                    uint16_t nb_rx_desc,
 780                    unsigned int socket_id,
 781                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 782                    struct rte_mempool *mp)
 783 {
 784         struct pmd_internals *internals = dev->data->dev_private;
 785         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
 786         struct rte_mbuf **tmp = &rxq->pool;
 787         long iov_max = sysconf(_SC_IOV_MAX);
 788         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
 789         struct iovec (*iovecs)[nb_desc + 1];
 790         int data_off = RTE_PKTMBUF_HEADROOM;
 791         int ret = 0;
 792         int fd;
 793         int i;
 794
 795         if ((rx_queue_id >= internals->nb_queues) || !mp) {
 796                 RTE_LOG(WARNING, PMD,
 797                         "nb_queues %d too small or mempool NULL\n",
 798                         internals->nb_queues);
 799                 return -1;
 800         }
 801
 802         rxq->mp = mp;
 803         rxq->trigger_seen = 1; /* force initial burst */
 804         rxq->in_port = dev->data->port_id;
 805         rxq->nb_rx_desc = nb_desc;
 806         iovecs = rte_zmalloc_socket(dev->data->name, sizeof(*iovecs), 0,
 807                                     socket_id);
 808         if (!iovecs) {
 809                 RTE_LOG(WARNING, PMD,
 810                         "%s: Couldn't allocate %d RX descriptors\n",
 811                         dev->data->name, nb_desc);
 812                 return -ENOMEM;
 813         }
 814         rxq->iovecs = iovecs;
 815
 816         dev->data->rx_queues[rx_queue_id] = rxq;
 817         fd = tap_setup_queue(dev, internals, rx_queue_id);
 818         if (fd == -1) {
 819                 ret = fd;
 820                 goto error;
 821         }
 822
 823         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
 824         (*rxq->iovecs)[0].iov_base = &rxq->pi;
 825
 826         for (i = 1; i <= nb_desc; i++) {
 827                 *tmp = rte_pktmbuf_alloc(rxq->mp);
 828                 if (!*tmp) {
 829                         RTE_LOG(WARNING, PMD,
 830                                 "%s: couldn't allocate memory for queue %d\n",
 831                                 dev->data->name, rx_queue_id);
 832                         ret = -ENOMEM;
 833                         goto error;
 834                 }
 835                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
 836                 (*rxq->iovecs)[i].iov_base =
 837                         (char *)(*tmp)->buf_addr + data_off;
 838                 data_off = 0;
 839                 tmp = &(*tmp)->next;
 840         }
 841
 842         RTE_LOG(DEBUG, PMD, "  RX TAP device name %s, qid %d on fd %d\n",
 843                 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
 844
 845         return 0;
 846
 847 error:
 848         rte_pktmbuf_free(rxq->pool);
 849         rxq->pool = NULL;
 850         rte_free(rxq->iovecs);
 851         rxq->iovecs = NULL;
 852         return ret;
 853 }
 854
 855 static int
 856 tap_tx_queue_setup(struct rte_eth_dev *dev,
 857                    uint16_t tx_queue_id,
 858                    uint16_t nb_tx_desc __rte_unused,
 859                    unsigned int socket_id __rte_unused,
 860                    const struct rte_eth_txconf *tx_conf __rte_unused)
 861 {
 862         struct pmd_internals *internals = dev->data->dev_private;
 863         int ret;
 864
 865         if (tx_queue_id >= internals->nb_queues)
 866                 return -1;
 867
 868         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
 869         ret = tap_setup_queue(dev, internals, tx_queue_id);
 870         if (ret == -1)
 871                 return -1;
 872
 873         RTE_LOG(DEBUG, PMD, "  TX TAP device name %s, qid %d on fd %d\n",
 874                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd);
 875
 876         return 0;
 877 }
 878
 879 static int
 880 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 881 {
 882         struct pmd_internals *pmd = dev->data->dev_private;
 883         struct ifreq ifr = { .ifr_mtu = mtu };
 884         int err = 0;
 885
 886         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
 887         if (!err)
 888                 dev->data->mtu = mtu;
 889
 890         return err;
 891 }
 892
 893 static int
 894 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
 895                      struct ether_addr *mc_addr_set __rte_unused,
 896                      uint32_t nb_mc_addr __rte_unused)
 897 {
 898         /*
 899          * Nothing to do actually: the tap has no filtering whatsoever, every
 900          * packet is received.
 901          */
 902         return 0;
 903 }
 904
 905 static int
 906 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
 907 {
 908         struct rte_eth_dev *dev = arg;
 909         struct pmd_internals *pmd = dev->data->dev_private;
 910         struct ifinfomsg *info = NLMSG_DATA(nh);
 911
 912         if (nh->nlmsg_type != RTM_NEWLINK ||
 913             (info->ifi_index != pmd->if_index &&
 914              info->ifi_index != pmd->remote_if_index))
 915                 return 0;
 916         return tap_link_update(dev, 0);
 917 }
 918
 919 static void
 920 tap_dev_intr_handler(void *cb_arg)
 921 {
 922         struct rte_eth_dev *dev = cb_arg;
 923         struct pmd_internals *pmd = dev->data->dev_private;
 924
 925         nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
 926 }
 927
 928 static int
 929 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
 930 {
 931         struct pmd_internals *pmd = dev->data->dev_private;
 932
 933         /* In any case, disable interrupt if the conf is no longer there. */
 934         if (!dev->data->dev_conf.intr_conf.lsc) {
 935                 if (pmd->intr_handle.fd != -1)
 936                         nl_final(pmd->intr_handle.fd);
 937                 rte_intr_callback_unregister(
 938                         &pmd->intr_handle, tap_dev_intr_handler, dev);
 939                 return 0;
 940         }
 941         if (set) {
 942                 pmd->intr_handle.fd = nl_init(RTMGRP_LINK);
 943                 if (unlikely(pmd->intr_handle.fd == -1))
 944                         return -EBADF;
 945                 return rte_intr_callback_register(
 946                         &pmd->intr_handle, tap_dev_intr_handler, dev);
 947         }
 948         nl_final(pmd->intr_handle.fd);
 949         return rte_intr_callback_unregister(&pmd->intr_handle,
 950                                             tap_dev_intr_handler, dev);
 951 }
 952
 953 static const uint32_t*
 954 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
 955 {
 956         static const uint32_t ptypes[] = {
 957                 RTE_PTYPE_INNER_L2_ETHER,
 958                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
 959                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
 960                 RTE_PTYPE_INNER_L3_IPV4,
 961                 RTE_PTYPE_INNER_L3_IPV4_EXT,
 962                 RTE_PTYPE_INNER_L3_IPV6,
 963                 RTE_PTYPE_INNER_L3_IPV6_EXT,
 964                 RTE_PTYPE_INNER_L4_FRAG,
 965                 RTE_PTYPE_INNER_L4_UDP,
 966                 RTE_PTYPE_INNER_L4_TCP,
 967                 RTE_PTYPE_INNER_L4_SCTP,
 968                 RTE_PTYPE_L2_ETHER,
 969                 RTE_PTYPE_L2_ETHER_VLAN,
 970                 RTE_PTYPE_L2_ETHER_QINQ,
 971                 RTE_PTYPE_L3_IPV4,
 972                 RTE_PTYPE_L3_IPV4_EXT,
 973                 RTE_PTYPE_L3_IPV6_EXT,
 974                 RTE_PTYPE_L3_IPV6,
 975                 RTE_PTYPE_L4_FRAG,
 976                 RTE_PTYPE_L4_UDP,
 977                 RTE_PTYPE_L4_TCP,
 978                 RTE_PTYPE_L4_SCTP,
 979         };
 980
 981         return ptypes;
 982 }
 983
 984 static int
 985 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
 986                   struct rte_eth_fc_conf *fc_conf)
 987 {
 988         fc_conf->mode = RTE_FC_NONE;
 989         return 0;
 990 }
 991
 992 static int
 993 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
 994                   struct rte_eth_fc_conf *fc_conf)
 995 {
 996         if (fc_conf->mode != RTE_FC_NONE)
 997                 return -ENOTSUP;
 998         return 0;
 999 }
1000
1001 static const struct eth_dev_ops ops = {
1002         .dev_start              = tap_dev_start,
1003         .dev_stop               = tap_dev_stop,
1004         .dev_close              = tap_dev_close,
1005         .dev_configure          = tap_dev_configure,
1006         .dev_infos_get          = tap_dev_info,
1007         .rx_queue_setup         = tap_rx_queue_setup,
1008         .tx_queue_setup         = tap_tx_queue_setup,
1009         .rx_queue_release       = tap_rx_queue_release,
1010         .tx_queue_release       = tap_tx_queue_release,
1011         .flow_ctrl_get          = tap_flow_ctrl_get,
1012         .flow_ctrl_set          = tap_flow_ctrl_set,
1013         .link_update            = tap_link_update,
1014         .dev_set_link_up        = tap_link_set_up,
1015         .dev_set_link_down      = tap_link_set_down,
1016         .promiscuous_enable     = tap_promisc_enable,
1017         .promiscuous_disable    = tap_promisc_disable,
1018         .allmulticast_enable    = tap_allmulti_enable,
1019         .allmulticast_disable   = tap_allmulti_disable,
1020         .mac_addr_set           = tap_mac_set,
1021         .mtu_set                = tap_mtu_set,
1022         .set_mc_addr_list       = tap_set_mc_addr_list,
1023         .stats_get              = tap_stats_get,
1024         .stats_reset            = tap_stats_reset,
1025         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1026         .filter_ctrl            = tap_dev_filter_ctrl,
1027 };
1028
1029 static int
1030 tap_kernel_support(struct pmd_internals *pmd)
1031 {
1032         struct utsname utsname;
1033         int ver[3];
1034
1035         if (uname(&utsname) == -1 ||
1036             sscanf(utsname.release, "%d.%d.%d",
1037                    &ver[0], &ver[1], &ver[2]) != 3)
1038                 return 0;
1039         if (KERNEL_VERSION(ver[0], ver[1], ver[2]) >= FLOWER_KERNEL_VERSION)
1040                 pmd->flower_support = 1;
1041         if (KERNEL_VERSION(ver[0], ver[1], ver[2]) >=
1042             FLOWER_VLAN_KERNEL_VERSION)
1043                 pmd->flower_vlan_support = 1;
1044         return 1;
1045 }
1046
1047 static int
1048 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1049                    char *remote_iface, int fixed_mac_type)
1050 {
1051         int numa_node = rte_socket_id();
1052         struct rte_eth_dev *dev;
1053         struct pmd_internals *pmd;
1054         struct rte_eth_dev_data *data;
1055         struct ifreq ifr;
1056         int i;
1057
1058         RTE_LOG(DEBUG, PMD, "  TAP device on numa %u\n", rte_socket_id());
1059
1060         data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
1061         if (!data) {
1062                 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
1063                 goto error_exit;
1064         }
1065
1066         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1067         if (!dev) {
1068                 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
1069                 goto error_exit;
1070         }
1071
1072         pmd = dev->data->dev_private;
1073         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1074         pmd->nb_queues = RTE_PMD_TAP_MAX_QUEUES;
1075
1076         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1077         if (pmd->ioctl_sock == -1) {
1078                 RTE_LOG(ERR, PMD,
1079                         "TAP Unable to get a socket for management: %s\n",
1080                         strerror(errno));
1081                 goto error_exit;
1082         }
1083
1084         /* Setup some default values */
1085         rte_memcpy(data, dev->data, sizeof(*data));
1086         data->dev_private = pmd;
1087         data->dev_flags = RTE_ETH_DEV_DETACHABLE | RTE_ETH_DEV_INTR_LSC;
1088         data->numa_node = numa_node;
1089         data->drv_name = pmd_tap_drv.driver.name;
1090
1091         data->dev_link = pmd_link;
1092         data->mac_addrs = &pmd->eth_addr;
1093         data->nb_rx_queues = pmd->nb_queues;
1094         data->nb_tx_queues = pmd->nb_queues;
1095
1096         dev->data = data;
1097         dev->dev_ops = &ops;
1098         dev->rx_pkt_burst = pmd_rx_burst;
1099         dev->tx_pkt_burst = pmd_tx_burst;
1100
1101         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1102         pmd->intr_handle.fd = -1;
1103
1104         /* Presetup the fds to -1 as being not valid */
1105         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1106                 pmd->rxq[i].fd = -1;
1107                 pmd->txq[i].fd = -1;
1108         }
1109
1110         if (fixed_mac_type) {
1111                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1112                 static int iface_idx;
1113                 char mac[ETHER_ADDR_LEN] = "\0dtap";
1114
1115                 mac[ETHER_ADDR_LEN - 1] = iface_idx++;
1116                 rte_memcpy(&pmd->eth_addr, mac, ETHER_ADDR_LEN);
1117         } else {
1118                 eth_random_addr((uint8_t *)&pmd->eth_addr);
1119         }
1120
1121         /* Immediately create the netdevice (this will create the 1st queue). */
1122         if (tap_setup_queue(dev, pmd, 0) == -1)
1123                 goto error_exit;
1124
1125         ifr.ifr_mtu = dev->data->mtu;
1126         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1127                 goto error_exit;
1128
1129         memset(&ifr, 0, sizeof(struct ifreq));
1130         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1131         rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr, ETHER_ADDR_LEN);
1132         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1133                 goto error_exit;
1134
1135         tap_kernel_support(pmd);
1136         if (!pmd->flower_support) {
1137                 if (remote_iface[0]) {
1138                         RTE_LOG(ERR, PMD,
1139                                 "%s: kernel does not support TC rules, required for remote feature.",
1140                                 pmd->name);
1141                         goto error_exit;
1142                 } else {
1143                         RTE_LOG(INFO, PMD,
1144                                 "%s: kernel too old for Flow API support.\n",
1145                                 pmd->name);
1146                         return 0;
1147                 }
1148         }
1149
1150         /*
1151          * Set up everything related to rte_flow:
1152          * - netlink socket
1153          * - tap / remote if_index
1154          * - mandatory QDISCs
1155          * - rte_flow actual/implicit lists
1156          * - implicit rules
1157          */
1158         pmd->nlsk_fd = nl_init(0);
1159         if (pmd->nlsk_fd == -1) {
1160                 RTE_LOG(WARNING, PMD, "%s: failed to create netlink socket.",
1161                         pmd->name);
1162                 goto disable_rte_flow;
1163         }
1164         pmd->if_index = if_nametoindex(pmd->name);
1165         if (!pmd->if_index) {
1166                 RTE_LOG(ERR, PMD, "%s: failed to get if_index.", pmd->name);
1167                 goto disable_rte_flow;
1168         }
1169         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1170                 RTE_LOG(ERR, PMD, "%s: failed to create multiq qdisc.",
1171                         pmd->name);
1172                 goto disable_rte_flow;
1173         }
1174         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1175                 RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.",
1176                         pmd->name);
1177                 goto disable_rte_flow;
1178         }
1179         LIST_INIT(&pmd->flows);
1180
1181         if (strlen(remote_iface)) {
1182                 pmd->remote_if_index = if_nametoindex(remote_iface);
1183                 if (!pmd->remote_if_index) {
1184                         RTE_LOG(ERR, PMD, "%s: failed to get %s if_index.",
1185                                 pmd->name, remote_iface);
1186                         goto error_remote;
1187                 }
1188                 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1189                          "%s", remote_iface);
1190                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1191                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.",
1192                                 pmd->name, pmd->remote_iface);
1193                         goto error_remote;
1194                 }
1195                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1196                            ETHER_ADDR_LEN);
1197                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1198                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1199                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.",
1200                                 pmd->name, remote_iface);
1201                         goto error_remote;
1202                 }
1203
1204                 /*
1205                  * Flush usually returns negative value because it tries to
1206                  * delete every QDISC (and on a running device, one QDISC at
1207                  * least is needed). Ignore negative return value.
1208                  */
1209                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1210                 if (qdisc_create_ingress(pmd->nlsk_fd,
1211                                          pmd->remote_if_index) < 0) {
1212                         RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.",
1213                                 pmd->remote_iface);
1214                         goto error_remote;
1215                 }
1216                 LIST_INIT(&pmd->implicit_flows);
1217                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1218                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1219                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1220                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1221                         RTE_LOG(ERR, PMD,
1222                                 "%s: failed to create implicit rules.",
1223                                 pmd->name);
1224                         goto error_remote;
1225                 }
1226         }
1227
1228         return 0;
1229
1230 disable_rte_flow:
1231         RTE_LOG(ERR, PMD, " Disabling rte flow support: %s(%d)\n",
1232                 strerror(errno), errno);
1233         if (strlen(remote_iface)) {
1234                 RTE_LOG(ERR, PMD, "Remote feature requires flow support.\n");
1235                 goto error_exit;
1236         }
1237         pmd->flower_support = 0;
1238         return 0;
1239
1240 error_remote:
1241         RTE_LOG(ERR, PMD, " Can't set up remote feature: %s(%d)\n",
1242                 strerror(errno), errno);
1243         tap_flow_implicit_flush(pmd, NULL);
1244
1245 error_exit:
1246         RTE_LOG(ERR, PMD, "TAP Unable to initialize %s\n",
1247                 rte_vdev_device_name(vdev));
1248
1249         rte_free(data);
1250         return -EINVAL;
1251 }
1252
1253 static int
1254 set_interface_name(const char *key __rte_unused,
1255                    const char *value,
1256                    void *extra_args)
1257 {
1258         char *name = (char *)extra_args;
1259
1260         if (value)
1261                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
1262         else
1263                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1264                          DEFAULT_TAP_NAME, (tap_unit - 1));
1265
1266         return 0;
1267 }
1268
1269 static int
1270 set_interface_speed(const char *key __rte_unused,
1271                     const char *value,
1272                     void *extra_args)
1273 {
1274         *(int *)extra_args = (value) ? atoi(value) : ETH_SPEED_NUM_10G;
1275
1276         return 0;
1277 }
1278
1279 static int
1280 set_remote_iface(const char *key __rte_unused,
1281                  const char *value,
1282                  void *extra_args)
1283 {
1284         char *name = (char *)extra_args;
1285
1286         if (value)
1287                 snprintf(name, RTE_ETH_NAME_MAX_LEN, "%s", value);
1288
1289         return 0;
1290 }
1291
1292 static int
1293 set_mac_type(const char *key __rte_unused,
1294              const char *value,
1295              void *extra_args)
1296 {
1297         if (value &&
1298             !strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED)))
1299                 *(int *)extra_args = 1;
1300         return 0;
1301 }
1302
1303 /* Open a TAP interface device.
1304  */
1305 static int
1306 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1307 {
1308         const char *name, *params;
1309         int ret;
1310         struct rte_kvargs *kvlist = NULL;
1311         int speed;
1312         char tap_name[RTE_ETH_NAME_MAX_LEN];
1313         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1314         int fixed_mac_type = 0;
1315
1316         name = rte_vdev_device_name(dev);
1317         params = rte_vdev_device_args(dev);
1318
1319         speed = ETH_SPEED_NUM_10G;
1320         snprintf(tap_name, sizeof(tap_name), "%s%d",
1321                  DEFAULT_TAP_NAME, tap_unit++);
1322         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1323
1324         if (params && (params[0] != '\0')) {
1325                 RTE_LOG(DEBUG, PMD, "paramaters (%s)\n", params);
1326
1327                 kvlist = rte_kvargs_parse(params, valid_arguments);
1328                 if (kvlist) {
1329                         if (rte_kvargs_count(kvlist, ETH_TAP_SPEED_ARG) == 1) {
1330                                 ret = rte_kvargs_process(kvlist,
1331                                                          ETH_TAP_SPEED_ARG,
1332                                                          &set_interface_speed,
1333                                                          &speed);
1334                                 if (ret == -1)
1335                                         goto leave;
1336                         }
1337
1338                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1339                                 ret = rte_kvargs_process(kvlist,
1340                                                          ETH_TAP_IFACE_ARG,
1341                                                          &set_interface_name,
1342                                                          tap_name);
1343                                 if (ret == -1)
1344                                         goto leave;
1345                         }
1346
1347                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1348                                 ret = rte_kvargs_process(kvlist,
1349                                                          ETH_TAP_REMOTE_ARG,
1350                                                          &set_remote_iface,
1351                                                          remote_iface);
1352                                 if (ret == -1)
1353                                         goto leave;
1354                         }
1355
1356                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1357                                 ret = rte_kvargs_process(kvlist,
1358                                                          ETH_TAP_MAC_ARG,
1359                                                          &set_mac_type,
1360                                                          &fixed_mac_type);
1361                                 if (ret == -1)
1362                                         goto leave;
1363                         }
1364                 }
1365         }
1366         pmd_link.link_speed = speed;
1367
1368         RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1369                 name, tap_name);
1370
1371         ret = eth_dev_tap_create(dev, tap_name, remote_iface, fixed_mac_type);
1372
1373 leave:
1374         if (ret == -1) {
1375                 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1376                         name, tap_name);
1377                 tap_unit--;             /* Restore the unit number */
1378         }
1379         rte_kvargs_free(kvlist);
1380
1381         return ret;
1382 }
1383
1384 /* detach a TAP device.
1385  */
1386 static int
1387 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1388 {
1389         struct rte_eth_dev *eth_dev = NULL;
1390         struct pmd_internals *internals;
1391         int i;
1392
1393         RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1394                 rte_socket_id());
1395
1396         /* find the ethdev entry */
1397         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1398         if (!eth_dev)
1399                 return 0;
1400
1401         internals = eth_dev->data->dev_private;
1402         if (internals->flower_support && internals->nlsk_fd) {
1403                 tap_flow_flush(eth_dev, NULL);
1404                 tap_flow_implicit_flush(internals, NULL);
1405                 nl_final(internals->nlsk_fd);
1406         }
1407         for (i = 0; i < internals->nb_queues; i++)
1408                 if (internals->rxq[i].fd != -1)
1409                         close(internals->rxq[i].fd);
1410
1411         close(internals->ioctl_sock);
1412         rte_free(eth_dev->data->dev_private);
1413         rte_free(eth_dev->data);
1414
1415         rte_eth_dev_release_port(eth_dev);
1416
1417         return 0;
1418 }
1419
1420 static struct rte_vdev_driver pmd_tap_drv = {
1421         .probe = rte_pmd_tap_probe,
1422         .remove = rte_pmd_tap_remove,
1423 };
1424 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1425 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1426 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1427                               ETH_TAP_IFACE_ARG "=<string> "
1428                               ETH_TAP_SPEED_ARG "=<int> "
1429                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_FIXED " "
1430                               ETH_TAP_REMOTE_ARG "=<string>");