drivers/net/tap/rte_eth_tap.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2016 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_common.h>
  36 #include <rte_mbuf.h>
  37 #include <rte_ethdev.h>
  38 #include <rte_malloc.h>
  39 #include <rte_vdev.h>
  40 #include <rte_kvargs.h>
  41 #include <rte_net.h>
  42
  43 #include <sys/types.h>
  44 #include <sys/stat.h>
  45 #include <sys/socket.h>
  46 #include <sys/ioctl.h>
  47 #include <sys/mman.h>
  48 #include <errno.h>
  49 #include <signal.h>
  50 #include <stdint.h>
  51 #include <unistd.h>
  52 #include <arpa/inet.h>
  53 #include <linux/if.h>
  54 #include <linux/if_tun.h>
  55 #include <linux/if_ether.h>
  56 #include <fcntl.h>
  57
  58 #include <rte_eth_tap.h>
  59
  60 /* Linux based path to the TUN device */
  61 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  62 #define DEFAULT_TAP_NAME        "dtap"
  63
  64 #define ETH_TAP_IFACE_ARG       "iface"
  65 #define ETH_TAP_SPEED_ARG       "speed"
  66
  67 #ifdef IFF_MULTI_QUEUE
  68 #define RTE_PMD_TAP_MAX_QUEUES  16
  69 #else
  70 #define RTE_PMD_TAP_MAX_QUEUES  1
  71 #endif
  72
  73 static struct rte_vdev_driver pmd_tap_drv;
  74
  75 static const char *valid_arguments[] = {
  76         ETH_TAP_IFACE_ARG,
  77         ETH_TAP_SPEED_ARG,
  78         NULL
  79 };
  80
  81 static int tap_unit;
  82
  83 static volatile uint32_t tap_trigger;   /* Rx trigger */
  84
  85 static struct rte_eth_link pmd_link = {
  86         .link_speed = ETH_SPEED_NUM_10G,
  87         .link_duplex = ETH_LINK_FULL_DUPLEX,
  88         .link_status = ETH_LINK_DOWN,
  89         .link_autoneg = ETH_LINK_SPEED_AUTONEG
  90 };
  91
  92 static void
  93 tap_trigger_cb(int sig __rte_unused)
  94 {
  95         /* Valid trigger values are nonzero */
  96         tap_trigger = (tap_trigger + 1) | 0x80000000;
  97 }
  98
  99 static int
 100 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 101           struct ifreq *ifr, int set);
 102
 103 /* Tun/Tap allocation routine
 104  *
 105  * name is the number of the interface to use, unless NULL to take the host
 106  * supplied name.
 107  */
 108 static int
 109 tun_alloc(struct pmd_internals *pmd, uint16_t qid)
 110 {
 111         struct ifreq ifr;
 112 #ifdef IFF_MULTI_QUEUE
 113         unsigned int features;
 114 #endif
 115         int fd;
 116
 117         memset(&ifr, 0, sizeof(struct ifreq));
 118
 119         ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
 120         snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
 121
 122         RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
 123
 124         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 125         if (fd < 0) {
 126                 RTE_LOG(ERR, PMD, "Unable to create TAP interface");
 127                 goto error;
 128         }
 129
 130 #ifdef IFF_MULTI_QUEUE
 131         /* Grab the TUN features to verify we can work multi-queue */
 132         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 133                 RTE_LOG(ERR, PMD, "TAP unable to get TUN/TAP features\n");
 134                 goto error;
 135         }
 136         RTE_LOG(DEBUG, PMD, "  TAP Features %08x\n", features);
 137
 138         if (features & IFF_MULTI_QUEUE) {
 139                 RTE_LOG(DEBUG, PMD, "  Multi-queue support for %d queues\n",
 140                         RTE_PMD_TAP_MAX_QUEUES);
 141                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 142         } else
 143 #endif
 144         {
 145                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 146                 RTE_LOG(DEBUG, PMD, "  Single queue only support\n");
 147         }
 148
 149         /* Set the TUN/TAP configuration and set the name if needed */
 150         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 151                 RTE_LOG(WARNING, PMD,
 152                         "Unable to set TUNSETIFF for %s\n",
 153                         ifr.ifr_name);
 154                 perror("TUNSETIFF");
 155                 goto error;
 156         }
 157
 158         /* Always set the file descriptor to non-blocking */
 159         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 160                 RTE_LOG(WARNING, PMD,
 161                         "Unable to set %s to nonblocking\n",
 162                         ifr.ifr_name);
 163                 perror("F_SETFL, NONBLOCK");
 164                 goto error;
 165         }
 166
 167         /* Set up trigger to optimize empty Rx bursts */
 168         errno = 0;
 169         do {
 170                 struct sigaction sa;
 171                 int flags = fcntl(fd, F_GETFL);
 172
 173                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 174                         break;
 175                 if (sa.sa_handler != tap_trigger_cb) {
 176                         /*
 177                          * Make sure SIGIO is not already taken. This is done
 178                          * as late as possible to leave the application a
 179                          * chance to set up its own signal handler first.
 180                          */
 181                         if (sa.sa_handler != SIG_IGN &&
 182                             sa.sa_handler != SIG_DFL) {
 183                                 errno = EBUSY;
 184                                 break;
 185                         }
 186                         sa = (struct sigaction){
 187                                 .sa_flags = SA_RESTART,
 188                                 .sa_handler = tap_trigger_cb,
 189                         };
 190                         if (sigaction(SIGIO, &sa, NULL) == -1)
 191                                 break;
 192                 }
 193                 /* Enable SIGIO on file descriptor */
 194                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 195                 fcntl(fd, F_SETOWN, getpid());
 196         } while (0);
 197         if (errno) {
 198                 /* Disable trigger globally in case of error */
 199                 tap_trigger = 0;
 200                 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
 201                         strerror(errno));
 202         }
 203
 204         if (qid == 0) {
 205                 struct ifreq ifr;
 206
 207                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0) < 0)
 208                         goto error;
 209                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
 210                            ETHER_ADDR_LEN);
 211         }
 212
 213         return fd;
 214
 215 error:
 216         if (fd > 0)
 217                 close(fd);
 218         return -1;
 219 }
 220
 221 /* Callback to handle the rx burst of packets to the correct interface and
 222  * file descriptor(s) in a multi-queue setup.
 223  */
 224 static uint16_t
 225 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 226 {
 227         int len;
 228         struct rte_mbuf *mbuf;
 229         struct rx_queue *rxq = queue;
 230         uint16_t num_rx;
 231         unsigned long num_rx_bytes = 0;
 232         uint32_t trigger = tap_trigger;
 233
 234         if (trigger == rxq->trigger_seen)
 235                 return 0;
 236         if (trigger)
 237                 rxq->trigger_seen = trigger;
 238         rte_compiler_barrier();
 239         for (num_rx = 0; num_rx < nb_pkts; ) {
 240                 /* allocate the next mbuf */
 241                 mbuf = rte_pktmbuf_alloc(rxq->mp);
 242                 if (unlikely(!mbuf)) {
 243                         RTE_LOG(WARNING, PMD, "TAP unable to allocate mbuf\n");
 244                         break;
 245                 }
 246
 247                 len = read(rxq->fd, rte_pktmbuf_mtod(mbuf, char *),
 248                            rte_pktmbuf_tailroom(mbuf));
 249                 if (len <= 0) {
 250                         rte_pktmbuf_free(mbuf);
 251                         break;
 252                 }
 253
 254                 mbuf->data_len = len;
 255                 mbuf->pkt_len = len;
 256                 mbuf->port = rxq->in_port;
 257                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 258                                                       RTE_PTYPE_ALL_MASK);
 259
 260                 /* account for the receive frame */
 261                 bufs[num_rx++] = mbuf;
 262                 num_rx_bytes += mbuf->pkt_len;
 263         }
 264         rxq->stats.ipackets += num_rx;
 265         rxq->stats.ibytes += num_rx_bytes;
 266
 267         return num_rx;
 268 }
 269
 270 /* Callback to handle sending packets from the tap interface
 271  */
 272 static uint16_t
 273 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 274 {
 275         struct rte_mbuf *mbuf;
 276         struct tx_queue *txq = queue;
 277         uint16_t num_tx = 0;
 278         unsigned long num_tx_bytes = 0;
 279         int i, n;
 280
 281         if (unlikely(nb_pkts == 0))
 282                 return 0;
 283
 284         for (i = 0; i < nb_pkts; i++) {
 285                 /* copy the tx frame data */
 286                 mbuf = bufs[num_tx];
 287                 n = write(txq->fd,
 288                           rte_pktmbuf_mtod(mbuf, void *),
 289                           rte_pktmbuf_pkt_len(mbuf));
 290                 if (n <= 0)
 291                         break;
 292
 293                 num_tx++;
 294                 num_tx_bytes += mbuf->pkt_len;
 295                 rte_pktmbuf_free(mbuf);
 296         }
 297
 298         txq->stats.opackets += num_tx;
 299         txq->stats.errs += nb_pkts - num_tx;
 300         txq->stats.obytes += num_tx_bytes;
 301
 302         return num_tx;
 303 }
 304
 305 static int
 306 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 307           struct ifreq *ifr, int set)
 308 {
 309         short req_flags = ifr->ifr_flags;
 310
 311         snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 312         switch (request) {
 313         case SIOCSIFFLAGS:
 314                 /* fetch current flags to leave other flags untouched */
 315                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 316                         goto error;
 317                 if (set)
 318                         ifr->ifr_flags |= req_flags;
 319                 else
 320                         ifr->ifr_flags &= ~req_flags;
 321                 break;
 322         case SIOCGIFHWADDR:
 323         case SIOCSIFHWADDR:
 324         case SIOCSIFMTU:
 325                 break;
 326         default:
 327                 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
 328                         pmd->name);
 329                 return -EINVAL;
 330         }
 331         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 332                 goto error;
 333         return 0;
 334
 335 error:
 336         RTE_LOG(ERR, PMD, "%s: ioctl(%lu) failed with error: %s\n",
 337                 ifr->ifr_name, request, strerror(errno));
 338         return -errno;
 339 }
 340
 341 static int
 342 tap_link_set_down(struct rte_eth_dev *dev)
 343 {
 344         struct pmd_internals *pmd = dev->data->dev_private;
 345         struct ifreq ifr = { .ifr_flags = IFF_UP };
 346
 347         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 348         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0);
 349 }
 350
 351 static int
 352 tap_link_set_up(struct rte_eth_dev *dev)
 353 {
 354         struct pmd_internals *pmd = dev->data->dev_private;
 355         struct ifreq ifr = { .ifr_flags = IFF_UP };
 356
 357         dev->data->dev_link.link_status = ETH_LINK_UP;
 358         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1);
 359 }
 360
 361 static int
 362 tap_dev_start(struct rte_eth_dev *dev)
 363 {
 364         return tap_link_set_up(dev);
 365 }
 366
 367 /* This function gets called when the current port gets stopped.
 368  */
 369 static void
 370 tap_dev_stop(struct rte_eth_dev *dev)
 371 {
 372         tap_link_set_down(dev);
 373 }
 374
 375 static int
 376 tap_dev_configure(struct rte_eth_dev *dev __rte_unused)
 377 {
 378         return 0;
 379 }
 380
 381 static uint32_t
 382 tap_dev_speed_capa(void)
 383 {
 384         uint32_t speed = pmd_link.link_speed;
 385         uint32_t capa = 0;
 386
 387         if (speed >= ETH_SPEED_NUM_10M)
 388                 capa |= ETH_LINK_SPEED_10M;
 389         if (speed >= ETH_SPEED_NUM_100M)
 390                 capa |= ETH_LINK_SPEED_100M;
 391         if (speed >= ETH_SPEED_NUM_1G)
 392                 capa |= ETH_LINK_SPEED_1G;
 393         if (speed >= ETH_SPEED_NUM_5G)
 394                 capa |= ETH_LINK_SPEED_2_5G;
 395         if (speed >= ETH_SPEED_NUM_5G)
 396                 capa |= ETH_LINK_SPEED_5G;
 397         if (speed >= ETH_SPEED_NUM_10G)
 398                 capa |= ETH_LINK_SPEED_10G;
 399         if (speed >= ETH_SPEED_NUM_20G)
 400                 capa |= ETH_LINK_SPEED_20G;
 401         if (speed >= ETH_SPEED_NUM_25G)
 402                 capa |= ETH_LINK_SPEED_25G;
 403         if (speed >= ETH_SPEED_NUM_40G)
 404                 capa |= ETH_LINK_SPEED_40G;
 405         if (speed >= ETH_SPEED_NUM_50G)
 406                 capa |= ETH_LINK_SPEED_50G;
 407         if (speed >= ETH_SPEED_NUM_56G)
 408                 capa |= ETH_LINK_SPEED_56G;
 409         if (speed >= ETH_SPEED_NUM_100G)
 410                 capa |= ETH_LINK_SPEED_100G;
 411
 412         return capa;
 413 }
 414
 415 static void
 416 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 417 {
 418         struct pmd_internals *internals = dev->data->dev_private;
 419
 420         dev_info->if_index = internals->if_index;
 421         dev_info->max_mac_addrs = 1;
 422         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 423         dev_info->max_rx_queues = internals->nb_queues;
 424         dev_info->max_tx_queues = internals->nb_queues;
 425         dev_info->min_rx_bufsize = 0;
 426         dev_info->pci_dev = NULL;
 427         dev_info->speed_capa = tap_dev_speed_capa();
 428 }
 429
 430 static void
 431 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 432 {
 433         unsigned int i, imax;
 434         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 435         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 436         const struct pmd_internals *pmd = dev->data->dev_private;
 437
 438         imax = (pmd->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 439                 pmd->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 440
 441         for (i = 0; i < imax; i++) {
 442                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 443                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 444                 rx_total += tap_stats->q_ipackets[i];
 445                 rx_bytes_total += tap_stats->q_ibytes[i];
 446
 447                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 448                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 449                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 450                 tx_total += tap_stats->q_opackets[i];
 451                 tx_err_total += tap_stats->q_errors[i];
 452                 tx_bytes_total += tap_stats->q_obytes[i];
 453         }
 454
 455         tap_stats->ipackets = rx_total;
 456         tap_stats->ibytes = rx_bytes_total;
 457         tap_stats->opackets = tx_total;
 458         tap_stats->oerrors = tx_err_total;
 459         tap_stats->obytes = tx_bytes_total;
 460 }
 461
 462 static void
 463 tap_stats_reset(struct rte_eth_dev *dev)
 464 {
 465         int i;
 466         struct pmd_internals *pmd = dev->data->dev_private;
 467
 468         for (i = 0; i < pmd->nb_queues; i++) {
 469                 pmd->rxq[i].stats.ipackets = 0;
 470                 pmd->rxq[i].stats.ibytes = 0;
 471
 472                 pmd->txq[i].stats.opackets = 0;
 473                 pmd->txq[i].stats.errs = 0;
 474                 pmd->txq[i].stats.obytes = 0;
 475         }
 476 }
 477
 478 static void
 479 tap_dev_close(struct rte_eth_dev *dev __rte_unused)
 480 {
 481         int i;
 482         struct pmd_internals *internals = dev->data->dev_private;
 483
 484         tap_link_set_down(dev);
 485
 486         for (i = 0; i < internals->nb_queues; i++) {
 487                 if (internals->rxq[i].fd != -1)
 488                         close(internals->rxq[i].fd);
 489                 internals->rxq[i].fd = -1;
 490                 internals->txq[i].fd = -1;
 491         }
 492 }
 493
 494 static void
 495 tap_rx_queue_release(void *queue)
 496 {
 497         struct rx_queue *rxq = queue;
 498
 499         if (rxq && (rxq->fd > 0)) {
 500                 close(rxq->fd);
 501                 rxq->fd = -1;
 502         }
 503 }
 504
 505 static void
 506 tap_tx_queue_release(void *queue)
 507 {
 508         struct tx_queue *txq = queue;
 509
 510         if (txq && (txq->fd > 0)) {
 511                 close(txq->fd);
 512                 txq->fd = -1;
 513         }
 514 }
 515
 516 static int
 517 tap_link_update(struct rte_eth_dev *dev __rte_unused,
 518                 int wait_to_complete __rte_unused)
 519 {
 520         return 0;
 521 }
 522
 523 static void
 524 tap_promisc_enable(struct rte_eth_dev *dev)
 525 {
 526         struct pmd_internals *pmd = dev->data->dev_private;
 527         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 528
 529         dev->data->promiscuous = 1;
 530         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1);
 531 }
 532
 533 static void
 534 tap_promisc_disable(struct rte_eth_dev *dev)
 535 {
 536         struct pmd_internals *pmd = dev->data->dev_private;
 537         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 538
 539         dev->data->promiscuous = 0;
 540         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0);
 541 }
 542
 543 static void
 544 tap_allmulti_enable(struct rte_eth_dev *dev)
 545 {
 546         struct pmd_internals *pmd = dev->data->dev_private;
 547         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 548
 549         dev->data->all_multicast = 1;
 550         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1);
 551 }
 552
 553 static void
 554 tap_allmulti_disable(struct rte_eth_dev *dev)
 555 {
 556         struct pmd_internals *pmd = dev->data->dev_private;
 557         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 558
 559         dev->data->all_multicast = 0;
 560         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0);
 561 }
 562
 563
 564 static void
 565 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 566 {
 567         struct pmd_internals *pmd = dev->data->dev_private;
 568         struct ifreq ifr;
 569
 570         if (is_zero_ether_addr(mac_addr)) {
 571                 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
 572                         dev->data->name);
 573                 return;
 574         }
 575
 576         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
 577         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
 578         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1) < 0)
 579                 return;
 580         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
 581 }
 582
 583 static int
 584 tap_setup_queue(struct rte_eth_dev *dev,
 585                 struct pmd_internals *internals,
 586                 uint16_t qid)
 587 {
 588         struct pmd_internals *pmd = dev->data->dev_private;
 589         struct rx_queue *rx = &internals->rxq[qid];
 590         struct tx_queue *tx = &internals->txq[qid];
 591         int fd;
 592
 593         fd = rx->fd;
 594         if (fd < 0) {
 595                 fd = tx->fd;
 596                 if (fd < 0) {
 597                         RTE_LOG(INFO, PMD, "Add queue to TAP %s for qid %d\n",
 598                                 pmd->name, qid);
 599                         fd = tun_alloc(pmd, qid);
 600                         if (fd < 0) {
 601                                 RTE_LOG(ERR, PMD, "tun_alloc(%s, %d) failed\n",
 602                                         pmd->name, qid);
 603                                 return -1;
 604                         }
 605                         if (qid == 0) {
 606                                 struct ifreq ifr;
 607
 608                                 ifr.ifr_mtu = dev->data->mtu;
 609                                 if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1) < 0) {
 610                                         close(fd);
 611                                         return -1;
 612                                 }
 613                         }
 614                 }
 615         }
 616
 617         rx->fd = fd;
 618         tx->fd = fd;
 619
 620         return fd;
 621 }
 622
 623 static int
 624 rx_setup_queue(struct rte_eth_dev *dev,
 625                 struct pmd_internals *internals,
 626                 uint16_t qid)
 627 {
 628         dev->data->rx_queues[qid] = &internals->rxq[qid];
 629
 630         return tap_setup_queue(dev, internals, qid);
 631 }
 632
 633 static int
 634 tx_setup_queue(struct rte_eth_dev *dev,
 635                 struct pmd_internals *internals,
 636                 uint16_t qid)
 637 {
 638         dev->data->tx_queues[qid] = &internals->txq[qid];
 639
 640         return tap_setup_queue(dev, internals, qid);
 641 }
 642
 643 static int
 644 tap_rx_queue_setup(struct rte_eth_dev *dev,
 645                    uint16_t rx_queue_id,
 646                    uint16_t nb_rx_desc __rte_unused,
 647                    unsigned int socket_id __rte_unused,
 648                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 649                    struct rte_mempool *mp)
 650 {
 651         struct pmd_internals *internals = dev->data->dev_private;
 652         uint16_t buf_size;
 653         int fd;
 654
 655         if ((rx_queue_id >= internals->nb_queues) || !mp) {
 656                 RTE_LOG(WARNING, PMD,
 657                         "nb_queues %d too small or mempool NULL\n",
 658                         internals->nb_queues);
 659                 return -1;
 660         }
 661
 662         internals->rxq[rx_queue_id].mp = mp;
 663         internals->rxq[rx_queue_id].trigger_seen = 1; /* force initial burst */
 664         internals->rxq[rx_queue_id].in_port = dev->data->port_id;
 665
 666         /* Now get the space available for data in the mbuf */
 667         buf_size = (uint16_t)(rte_pktmbuf_data_room_size(mp) -
 668                                 RTE_PKTMBUF_HEADROOM);
 669
 670         if (buf_size < ETH_FRAME_LEN) {
 671                 RTE_LOG(WARNING, PMD,
 672                         "%s: %d bytes will not fit in mbuf (%d bytes)\n",
 673                         dev->data->name, ETH_FRAME_LEN, buf_size);
 674                 return -ENOMEM;
 675         }
 676
 677         fd = rx_setup_queue(dev, internals, rx_queue_id);
 678         if (fd == -1)
 679                 return -1;
 680
 681         RTE_LOG(DEBUG, PMD, "  RX TAP device name %s, qid %d on fd %d\n",
 682                 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
 683
 684         return 0;
 685 }
 686
 687 static int
 688 tap_tx_queue_setup(struct rte_eth_dev *dev,
 689                    uint16_t tx_queue_id,
 690                    uint16_t nb_tx_desc __rte_unused,
 691                    unsigned int socket_id __rte_unused,
 692                    const struct rte_eth_txconf *tx_conf __rte_unused)
 693 {
 694         struct pmd_internals *internals = dev->data->dev_private;
 695         int ret;
 696
 697         if (tx_queue_id >= internals->nb_queues)
 698                 return -1;
 699
 700         ret = tx_setup_queue(dev, internals, tx_queue_id);
 701         if (ret == -1)
 702                 return -1;
 703
 704         RTE_LOG(DEBUG, PMD, "  TX TAP device name %s, qid %d on fd %d\n",
 705                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd);
 706
 707         return 0;
 708 }
 709
 710 static int
 711 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 712 {
 713         struct pmd_internals *pmd = dev->data->dev_private;
 714         struct ifreq ifr = { .ifr_mtu = mtu };
 715         int err = 0;
 716
 717         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1);
 718         if (!err)
 719                 dev->data->mtu = mtu;
 720
 721         return err;
 722 }
 723
 724 static int
 725 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
 726                      struct ether_addr *mc_addr_set __rte_unused,
 727                      uint32_t nb_mc_addr __rte_unused)
 728 {
 729         /*
 730          * Nothing to do actually: the tap has no filtering whatsoever, every
 731          * packet is received.
 732          */
 733         return 0;
 734 }
 735
 736 static const uint32_t*
 737 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
 738 {
 739         static const uint32_t ptypes[] = {
 740                 RTE_PTYPE_INNER_L2_ETHER,
 741                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
 742                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
 743                 RTE_PTYPE_INNER_L3_IPV4,
 744                 RTE_PTYPE_INNER_L3_IPV4_EXT,
 745                 RTE_PTYPE_INNER_L3_IPV6,
 746                 RTE_PTYPE_INNER_L3_IPV6_EXT,
 747                 RTE_PTYPE_INNER_L4_FRAG,
 748                 RTE_PTYPE_INNER_L4_UDP,
 749                 RTE_PTYPE_INNER_L4_TCP,
 750                 RTE_PTYPE_INNER_L4_SCTP,
 751                 RTE_PTYPE_L2_ETHER,
 752                 RTE_PTYPE_L2_ETHER_VLAN,
 753                 RTE_PTYPE_L2_ETHER_QINQ,
 754                 RTE_PTYPE_L3_IPV4,
 755                 RTE_PTYPE_L3_IPV4_EXT,
 756                 RTE_PTYPE_L3_IPV6_EXT,
 757                 RTE_PTYPE_L3_IPV6,
 758                 RTE_PTYPE_L4_FRAG,
 759                 RTE_PTYPE_L4_UDP,
 760                 RTE_PTYPE_L4_TCP,
 761                 RTE_PTYPE_L4_SCTP,
 762         };
 763
 764         return ptypes;
 765 }
 766
 767 static int
 768 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
 769                   struct rte_eth_fc_conf *fc_conf)
 770 {
 771         fc_conf->mode = RTE_FC_NONE;
 772         return 0;
 773 }
 774
 775 static int
 776 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
 777                   struct rte_eth_fc_conf *fc_conf)
 778 {
 779         if (fc_conf->mode != RTE_FC_NONE)
 780                 return -ENOTSUP;
 781         return 0;
 782 }
 783
 784 static const struct eth_dev_ops ops = {
 785         .dev_start              = tap_dev_start,
 786         .dev_stop               = tap_dev_stop,
 787         .dev_close              = tap_dev_close,
 788         .dev_configure          = tap_dev_configure,
 789         .dev_infos_get          = tap_dev_info,
 790         .rx_queue_setup         = tap_rx_queue_setup,
 791         .tx_queue_setup         = tap_tx_queue_setup,
 792         .rx_queue_release       = tap_rx_queue_release,
 793         .tx_queue_release       = tap_tx_queue_release,
 794         .flow_ctrl_get          = tap_flow_ctrl_get,
 795         .flow_ctrl_set          = tap_flow_ctrl_set,
 796         .link_update            = tap_link_update,
 797         .dev_set_link_up        = tap_link_set_up,
 798         .dev_set_link_down      = tap_link_set_down,
 799         .promiscuous_enable     = tap_promisc_enable,
 800         .promiscuous_disable    = tap_promisc_disable,
 801         .allmulticast_enable    = tap_allmulti_enable,
 802         .allmulticast_disable   = tap_allmulti_disable,
 803         .mac_addr_set           = tap_mac_set,
 804         .mtu_set                = tap_mtu_set,
 805         .set_mc_addr_list       = tap_set_mc_addr_list,
 806         .stats_get              = tap_stats_get,
 807         .stats_reset            = tap_stats_reset,
 808         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
 809 };
 810
 811 static int
 812 eth_dev_tap_create(const char *name, char *tap_name)
 813 {
 814         int numa_node = rte_socket_id();
 815         struct rte_eth_dev *dev = NULL;
 816         struct pmd_internals *pmd = NULL;
 817         struct rte_eth_dev_data *data = NULL;
 818         int i;
 819
 820         RTE_LOG(DEBUG, PMD, "  TAP device on numa %u\n", rte_socket_id());
 821
 822         data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
 823         if (!data) {
 824                 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
 825                 goto error_exit;
 826         }
 827
 828         pmd = rte_zmalloc_socket(tap_name, sizeof(*pmd), 0, numa_node);
 829         if (!pmd) {
 830                 RTE_LOG(ERR, PMD, "TAP Unable to allocate internal struct\n");
 831                 goto error_exit;
 832         }
 833
 834         /* name in allocation and data->name must be consistent */
 835         snprintf(data->name, sizeof(data->name), "%s", name);
 836         dev = rte_eth_dev_allocate(name);
 837         if (!dev) {
 838                 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
 839                 goto error_exit;
 840         }
 841
 842         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
 843
 844         pmd->nb_queues = RTE_PMD_TAP_MAX_QUEUES;
 845
 846         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
 847         if (pmd->ioctl_sock == -1) {
 848                 RTE_LOG(ERR, PMD,
 849                         "TAP Unable to get a socket for management: %s\n",
 850                         strerror(errno));
 851                 goto error_exit;
 852         }
 853
 854         /* Setup some default values */
 855         data->dev_private = pmd;
 856         data->port_id = dev->data->port_id;
 857         data->mtu = dev->data->mtu;
 858         data->dev_flags = RTE_ETH_DEV_DETACHABLE;
 859         data->kdrv = RTE_KDRV_NONE;
 860         data->drv_name = pmd_tap_drv.driver.name;
 861         data->numa_node = numa_node;
 862
 863         data->dev_link = pmd_link;
 864         data->mac_addrs = &pmd->eth_addr;
 865         data->nb_rx_queues = pmd->nb_queues;
 866         data->nb_tx_queues = pmd->nb_queues;
 867
 868         dev->data = data;
 869         dev->dev_ops = &ops;
 870         dev->driver = NULL;
 871         dev->rx_pkt_burst = pmd_rx_burst;
 872         dev->tx_pkt_burst = pmd_tx_burst;
 873
 874         /* Presetup the fds to -1 as being not valid */
 875         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 876                 pmd->rxq[i].fd = -1;
 877                 pmd->txq[i].fd = -1;
 878         }
 879
 880         return 0;
 881
 882 error_exit:
 883         RTE_LOG(DEBUG, PMD, "TAP Unable to initialize %s\n", name);
 884
 885         rte_free(data);
 886         rte_free(pmd);
 887
 888         rte_eth_dev_release_port(dev);
 889
 890         return -EINVAL;
 891 }
 892
 893 static int
 894 set_interface_name(const char *key __rte_unused,
 895                    const char *value,
 896                    void *extra_args)
 897 {
 898         char *name = (char *)extra_args;
 899
 900         if (value)
 901                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
 902         else
 903                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
 904                          DEFAULT_TAP_NAME, (tap_unit - 1));
 905
 906         return 0;
 907 }
 908
 909 static int
 910 set_interface_speed(const char *key __rte_unused,
 911                     const char *value,
 912                     void *extra_args)
 913 {
 914         *(int *)extra_args = (value) ? atoi(value) : ETH_SPEED_NUM_10G;
 915
 916         return 0;
 917 }
 918
 919 /* Open a TAP interface device.
 920  */
 921 static int
 922 rte_pmd_tap_probe(const char *name, const char *params)
 923 {
 924         int ret;
 925         struct rte_kvargs *kvlist = NULL;
 926         int speed;
 927         char tap_name[RTE_ETH_NAME_MAX_LEN];
 928
 929         speed = ETH_SPEED_NUM_10G;
 930         snprintf(tap_name, sizeof(tap_name), "%s%d",
 931                  DEFAULT_TAP_NAME, tap_unit++);
 932
 933         if (params && (params[0] != '\0')) {
 934                 RTE_LOG(DEBUG, PMD, "paramaters (%s)\n", params);
 935
 936                 kvlist = rte_kvargs_parse(params, valid_arguments);
 937                 if (kvlist) {
 938                         if (rte_kvargs_count(kvlist, ETH_TAP_SPEED_ARG) == 1) {
 939                                 ret = rte_kvargs_process(kvlist,
 940                                                          ETH_TAP_SPEED_ARG,
 941                                                          &set_interface_speed,
 942                                                          &speed);
 943                                 if (ret == -1)
 944                                         goto leave;
 945                         }
 946
 947                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
 948                                 ret = rte_kvargs_process(kvlist,
 949                                                          ETH_TAP_IFACE_ARG,
 950                                                          &set_interface_name,
 951                                                          tap_name);
 952                                 if (ret == -1)
 953                                         goto leave;
 954                         }
 955                 }
 956         }
 957         pmd_link.link_speed = speed;
 958
 959         RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
 960                 name, tap_name);
 961
 962         ret = eth_dev_tap_create(name, tap_name);
 963
 964 leave:
 965         if (ret == -1) {
 966                 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
 967                         name, tap_name);
 968                 tap_unit--;             /* Restore the unit number */
 969         }
 970         rte_kvargs_free(kvlist);
 971
 972         return ret;
 973 }
 974
 975 /* detach a TAP device.
 976  */
 977 static int
 978 rte_pmd_tap_remove(const char *name)
 979 {
 980         struct rte_eth_dev *eth_dev = NULL;
 981         struct pmd_internals *internals;
 982         int i;
 983
 984         RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
 985                 rte_socket_id());
 986
 987         /* find the ethdev entry */
 988         eth_dev = rte_eth_dev_allocated(name);
 989         if (!eth_dev)
 990                 return 0;
 991
 992         internals = eth_dev->data->dev_private;
 993         for (i = 0; i < internals->nb_queues; i++)
 994                 if (internals->rxq[i].fd != -1)
 995                         close(internals->rxq[i].fd);
 996
 997         close(internals->ioctl_sock);
 998         rte_free(eth_dev->data->dev_private);
 999         rte_free(eth_dev->data);
1000
1001         rte_eth_dev_release_port(eth_dev);
1002
1003         return 0;
1004 }
1005
1006 static struct rte_vdev_driver pmd_tap_drv = {
1007         .probe = rte_pmd_tap_probe,
1008         .remove = rte_pmd_tap_remove,
1009 };
1010 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1011 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1012 RTE_PMD_REGISTER_PARAM_STRING(net_tap, "iface=<string>,speed=N");