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_mbuf.h>
  35 #include <rte_ethdev.h>
  36 #include <rte_malloc.h>
  37 #include <rte_vdev.h>
  38 #include <rte_kvargs.h>
  39
  40 #include <sys/types.h>
  41 #include <sys/stat.h>
  42 #include <sys/socket.h>
  43 #include <sys/ioctl.h>
  44 #include <sys/mman.h>
  45 #include <unistd.h>
  46 #include <poll.h>
  47 #include <arpa/inet.h>
  48 #include <linux/if.h>
  49 #include <linux/if_tun.h>
  50 #include <linux/if_ether.h>
  51 #include <fcntl.h>
  52
  53 /* Linux based path to the TUN device */
  54 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  55 #define DEFAULT_TAP_NAME        "dtap"
  56
  57 #define ETH_TAP_IFACE_ARG       "iface"
  58 #define ETH_TAP_SPEED_ARG       "speed"
  59
  60 #define RTE_PMD_TAP_MAX_QUEUES  16
  61
  62 static struct rte_vdev_driver pmd_tap_drv;
  63
  64 static const char *valid_arguments[] = {
  65         ETH_TAP_IFACE_ARG,
  66         ETH_TAP_SPEED_ARG,
  67         NULL
  68 };
  69
  70 static int tap_unit;
  71
  72 static struct rte_eth_link pmd_link = {
  73         .link_speed = ETH_SPEED_NUM_10G,
  74         .link_duplex = ETH_LINK_FULL_DUPLEX,
  75         .link_status = ETH_LINK_DOWN,
  76         .link_autoneg = ETH_LINK_SPEED_AUTONEG
  77 };
  78
  79 struct pkt_stats {
  80         uint64_t opackets;              /* Number of output packets */
  81         uint64_t ipackets;              /* Number of input packets */
  82         uint64_t obytes;                /* Number of bytes on output */
  83         uint64_t ibytes;                /* Number of bytes on input */
  84         uint64_t errs;                  /* Number of error packets */
  85 };
  86
  87 struct rx_queue {
  88         struct rte_mempool *mp;         /* Mempool for RX packets */
  89         uint16_t in_port;               /* Port ID */
  90         int fd;
  91
  92         struct pkt_stats stats;         /* Stats for this RX queue */
  93 };
  94
  95 struct tx_queue {
  96         int fd;
  97         struct pkt_stats stats;         /* Stats for this TX queue */
  98 };
  99
 100 struct pmd_internals {
 101         char name[RTE_ETH_NAME_MAX_LEN];        /* Internal Tap device name */
 102         uint16_t nb_queues;             /* Number of queues supported */
 103         struct ether_addr eth_addr;     /* Mac address of the device port */
 104
 105         int if_index;                   /* IF_INDEX for the port */
 106         int fds[RTE_PMD_TAP_MAX_QUEUES]; /* List of all file descriptors */
 107
 108         struct rx_queue rxq[RTE_PMD_TAP_MAX_QUEUES];    /* List of RX queues */
 109         struct tx_queue txq[RTE_PMD_TAP_MAX_QUEUES];    /* List of TX queues */
 110 };
 111
 112 /* Tun/Tap allocation routine
 113  *
 114  * name is the number of the interface to use, unless NULL to take the host
 115  * supplied name.
 116  */
 117 static int
 118 tun_alloc(char *name)
 119 {
 120         struct ifreq ifr;
 121         unsigned int features;
 122         int fd;
 123
 124         memset(&ifr, 0, sizeof(struct ifreq));
 125
 126         ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
 127         if (name && name[0])
 128                 strncpy(ifr.ifr_name, name, IFNAMSIZ);
 129
 130         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 131         if (fd < 0) {
 132                 RTE_LOG(ERR, PMD, "Unable to create TAP interface");
 133                 goto error;
 134         }
 135
 136         /* Grab the TUN features to verify we can work */
 137         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 138                 RTE_LOG(ERR, PMD, "Unable to get TUN/TAP features\n");
 139                 goto error;
 140         }
 141         RTE_LOG(DEBUG, PMD, "TUN/TAP Features %08x\n", features);
 142
 143 #ifdef IFF_MULTI_QUEUE
 144         if (!(features & IFF_MULTI_QUEUE) && (RTE_PMD_TAP_MAX_QUEUES > 1)) {
 145                 RTE_LOG(DEBUG, PMD, "TUN/TAP device only one queue\n");
 146                 goto error;
 147         } else if ((features & IFF_ONE_QUEUE) &&
 148                         (RTE_PMD_TAP_MAX_QUEUES == 1)) {
 149                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 150                 RTE_LOG(DEBUG, PMD, "Single queue only support\n");
 151         } else {
 152                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 153                 RTE_LOG(DEBUG, PMD, "Multi-queue support for %d queues\n",
 154                         RTE_PMD_TAP_MAX_QUEUES);
 155         }
 156 #else
 157         if (RTE_PMD_TAP_MAX_QUEUES > 1) {
 158                 RTE_LOG(DEBUG, PMD, "TUN/TAP device only one queue\n");
 159                 goto error;
 160         } else {
 161                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 162                 RTE_LOG(DEBUG, PMD, "Single queue only support\n");
 163         }
 164 #endif
 165
 166         /* Set the TUN/TAP configuration and get the name if needed */
 167         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 168                 RTE_LOG(ERR, PMD, "Unable to set TUNSETIFF for %s\n",
 169                         ifr.ifr_name);
 170                 perror("TUNSETIFF");
 171                 goto error;
 172         }
 173
 174         /* Always set the file descriptor to non-blocking */
 175         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 176                 RTE_LOG(ERR, PMD, "Unable to set to nonblocking\n");
 177                 perror("F_SETFL, NONBLOCK");
 178                 goto error;
 179         }
 180
 181         /* If the name is different that new name as default */
 182         if (name && strcmp(name, ifr.ifr_name))
 183                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", ifr.ifr_name);
 184
 185         return fd;
 186
 187 error:
 188         if (fd > 0)
 189                 close(fd);
 190         return -1;
 191 }
 192
 193 /* Callback to handle the rx burst of packets to the correct interface and
 194  * file descriptor(s) in a multi-queue setup.
 195  */
 196 static uint16_t
 197 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 198 {
 199         int len;
 200         struct rte_mbuf *mbuf;
 201         struct rx_queue *rxq = queue;
 202         uint16_t num_rx;
 203         unsigned long num_rx_bytes = 0;
 204
 205         for (num_rx = 0; num_rx < nb_pkts; ) {
 206                 /* allocate the next mbuf */
 207                 mbuf = rte_pktmbuf_alloc(rxq->mp);
 208                 if (unlikely(!mbuf)) {
 209                         RTE_LOG(WARNING, PMD, "Unable to allocate mbuf\n");
 210                         break;
 211                 }
 212
 213                 len = read(rxq->fd, rte_pktmbuf_mtod(mbuf, char *),
 214                            rte_pktmbuf_tailroom(mbuf));
 215                 if (len <= 0) {
 216                         rte_pktmbuf_free(mbuf);
 217                         break;
 218                 }
 219
 220                 mbuf->data_len = len;
 221                 mbuf->pkt_len = len;
 222                 mbuf->port = rxq->in_port;
 223
 224                 /* account for the receive frame */
 225                 bufs[num_rx++] = mbuf;
 226                 num_rx_bytes += mbuf->pkt_len;
 227         }
 228         rxq->stats.ipackets += num_rx;
 229         rxq->stats.ibytes += num_rx_bytes;
 230
 231         return num_rx;
 232 }
 233
 234 /* Callback to handle sending packets from the tap interface
 235  */
 236 static uint16_t
 237 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 238 {
 239         struct rte_mbuf *mbuf;
 240         struct tx_queue *txq = queue;
 241         struct pollfd pfd;
 242         uint16_t num_tx = 0;
 243         unsigned long num_tx_bytes = 0;
 244         int i, n;
 245
 246         if (unlikely(nb_pkts == 0))
 247                 return 0;
 248
 249         pfd.events = POLLOUT;
 250         pfd.fd = txq->fd;
 251         for (i = 0; i < nb_pkts; i++) {
 252                 n = poll(&pfd, 1, 0);
 253
 254                 if (n <= 0)
 255                         break;
 256
 257                 if (pfd.revents & POLLOUT) {
 258                         /* copy the tx frame data */
 259                         mbuf = bufs[num_tx];
 260                         n = write(pfd.fd, rte_pktmbuf_mtod(mbuf, void*),
 261                                   rte_pktmbuf_pkt_len(mbuf));
 262                         if (n <= 0)
 263                                 break;
 264
 265                         num_tx++;
 266                         num_tx_bytes += mbuf->pkt_len;
 267                         rte_pktmbuf_free(mbuf);
 268                 }
 269         }
 270
 271         txq->stats.opackets += num_tx;
 272         txq->stats.errs += nb_pkts - num_tx;
 273         txq->stats.obytes += num_tx_bytes;
 274
 275         return num_tx;
 276 }
 277
 278 static int
 279 tap_dev_start(struct rte_eth_dev *dev)
 280 {
 281         /* Force the Link up */
 282         dev->data->dev_link.link_status = ETH_LINK_UP;
 283
 284         return 0;
 285 }
 286
 287 /* This function gets called when the current port gets stopped.
 288  */
 289 static void
 290 tap_dev_stop(struct rte_eth_dev *dev)
 291 {
 292         int i;
 293         struct pmd_internals *internals = dev->data->dev_private;
 294
 295         for (i = 0; i < internals->nb_queues; i++)
 296                 if (internals->fds[i] != -1)
 297                         close(internals->fds[i]);
 298
 299         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 300 }
 301
 302 static int
 303 tap_dev_configure(struct rte_eth_dev *dev __rte_unused)
 304 {
 305         return 0;
 306 }
 307
 308 static void
 309 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 310 {
 311         struct pmd_internals *internals = dev->data->dev_private;
 312
 313         dev_info->if_index = internals->if_index;
 314         dev_info->max_mac_addrs = 1;
 315         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 316         dev_info->max_rx_queues = internals->nb_queues;
 317         dev_info->max_tx_queues = internals->nb_queues;
 318         dev_info->min_rx_bufsize = 0;
 319         dev_info->pci_dev = NULL;
 320 }
 321
 322 static void
 323 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 324 {
 325         unsigned int i, imax;
 326         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 327         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 328         const struct pmd_internals *pmd = dev->data->dev_private;
 329
 330         imax = (pmd->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 331                 pmd->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 332
 333         for (i = 0; i < imax; i++) {
 334                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 335                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 336                 rx_total += tap_stats->q_ipackets[i];
 337                 rx_bytes_total += tap_stats->q_ibytes[i];
 338         }
 339
 340         for (i = 0; i < imax; i++) {
 341                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 342                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 343                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 344                 tx_total += tap_stats->q_opackets[i];
 345                 tx_err_total += tap_stats->q_errors[i];
 346                 tx_bytes_total += tap_stats->q_obytes[i];
 347         }
 348
 349         tap_stats->ipackets = rx_total;
 350         tap_stats->ibytes = rx_bytes_total;
 351         tap_stats->opackets = tx_total;
 352         tap_stats->oerrors = tx_err_total;
 353         tap_stats->obytes = tx_bytes_total;
 354 }
 355
 356 static void
 357 tap_stats_reset(struct rte_eth_dev *dev)
 358 {
 359         int i;
 360         struct pmd_internals *pmd = dev->data->dev_private;
 361
 362         for (i = 0; i < pmd->nb_queues; i++) {
 363                 pmd->rxq[i].stats.ipackets = 0;
 364                 pmd->rxq[i].stats.ibytes = 0;
 365         }
 366
 367         for (i = 0; i < pmd->nb_queues; i++) {
 368                 pmd->txq[i].stats.opackets = 0;
 369                 pmd->txq[i].stats.errs = 0;
 370                 pmd->txq[i].stats.obytes = 0;
 371         }
 372 }
 373
 374 static void
 375 tap_dev_close(struct rte_eth_dev *dev __rte_unused)
 376 {
 377 }
 378
 379 static void
 380 tap_rx_queue_release(void *queue)
 381 {
 382         struct rx_queue *rxq = queue;
 383
 384         if (rxq && (rxq->fd > 0)) {
 385                 close(rxq->fd);
 386                 rxq->fd = -1;
 387         }
 388 }
 389
 390 static void
 391 tap_tx_queue_release(void *queue)
 392 {
 393         struct tx_queue *txq = queue;
 394
 395         if (txq && (txq->fd > 0)) {
 396                 close(txq->fd);
 397                 txq->fd = -1;
 398         }
 399 }
 400
 401 static int
 402 tap_link_update(struct rte_eth_dev *dev __rte_unused,
 403                 int wait_to_complete __rte_unused)
 404 {
 405         return 0;
 406 }
 407
 408 static int
 409 tap_setup_queue(struct rte_eth_dev *dev,
 410                 struct pmd_internals *internals,
 411                 uint16_t qid)
 412 {
 413         struct rx_queue *rx = &internals->rxq[qid];
 414         struct tx_queue *tx = &internals->txq[qid];
 415         int fd;
 416
 417         fd = rx->fd;
 418         if (fd < 0) {
 419                 fd = tx->fd;
 420                 if (fd < 0) {
 421                         RTE_LOG(INFO, PMD, "Add queue to TAP %s for qid %d\n",
 422                                 dev->data->name, qid);
 423                         fd = tun_alloc(dev->data->name);
 424                         if (fd < 0) {
 425                                 RTE_LOG(ERR, PMD, "tun_alloc(%s) failed\n",
 426                                         dev->data->name);
 427                                 return -1;
 428                         }
 429                 }
 430         }
 431
 432         rx->fd = fd;
 433         tx->fd = fd;
 434
 435         return fd;
 436 }
 437
 438 static int
 439 rx_setup_queue(struct rte_eth_dev *dev,
 440                 struct pmd_internals *internals,
 441                 uint16_t qid)
 442 {
 443         dev->data->rx_queues[qid] = &internals->rxq[qid];
 444
 445         return tap_setup_queue(dev, internals, qid);
 446 }
 447
 448 static int
 449 tx_setup_queue(struct rte_eth_dev *dev,
 450                 struct pmd_internals *internals,
 451                 uint16_t qid)
 452 {
 453         dev->data->tx_queues[qid] = &internals->txq[qid];
 454
 455         return tap_setup_queue(dev, internals, qid);
 456 }
 457
 458 static int
 459 tap_rx_queue_setup(struct rte_eth_dev *dev,
 460                    uint16_t rx_queue_id,
 461                    uint16_t nb_rx_desc __rte_unused,
 462                    unsigned int socket_id __rte_unused,
 463                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 464                    struct rte_mempool *mp)
 465 {
 466         struct pmd_internals *internals = dev->data->dev_private;
 467         uint16_t buf_size;
 468         int fd;
 469
 470         if ((rx_queue_id >= internals->nb_queues) || !mp) {
 471                 RTE_LOG(ERR, PMD, "nb_queues %d mp %p\n",
 472                         internals->nb_queues, mp);
 473                 return -1;
 474         }
 475
 476         internals->rxq[rx_queue_id].mp = mp;
 477         internals->rxq[rx_queue_id].in_port = dev->data->port_id;
 478
 479         /* Now get the space available for data in the mbuf */
 480         buf_size = (uint16_t)(rte_pktmbuf_data_room_size(mp) -
 481                                 RTE_PKTMBUF_HEADROOM);
 482
 483         if (buf_size < ETH_FRAME_LEN) {
 484                 RTE_LOG(ERR, PMD,
 485                         "%s: %d bytes will not fit in mbuf (%d bytes)\n",
 486                         dev->data->name, ETH_FRAME_LEN, buf_size);
 487                 return -ENOMEM;
 488         }
 489
 490         fd = rx_setup_queue(dev, internals, rx_queue_id);
 491         if (fd == -1)
 492                 return -1;
 493
 494         internals->fds[rx_queue_id] = fd;
 495         RTE_LOG(INFO, PMD, "RX TAP device name %s, qid %d on fd %d\n",
 496                 dev->data->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
 497
 498         return 0;
 499 }
 500
 501 static int
 502 tap_tx_queue_setup(struct rte_eth_dev *dev,
 503                    uint16_t tx_queue_id,
 504                    uint16_t nb_tx_desc __rte_unused,
 505                    unsigned int socket_id __rte_unused,
 506                    const struct rte_eth_txconf *tx_conf __rte_unused)
 507 {
 508         struct pmd_internals *internals = dev->data->dev_private;
 509         int ret;
 510
 511         if (tx_queue_id >= internals->nb_queues)
 512                 return -1;
 513
 514         ret = tx_setup_queue(dev, internals, tx_queue_id);
 515         if (ret == -1)
 516                 return -1;
 517
 518         RTE_LOG(INFO, PMD, "TX TAP device name %s, qid %d on fd %d\n",
 519                 dev->data->name, tx_queue_id, internals->txq[tx_queue_id].fd);
 520
 521         return 0;
 522 }
 523
 524 static const struct eth_dev_ops ops = {
 525         .dev_start              = tap_dev_start,
 526         .dev_stop               = tap_dev_stop,
 527         .dev_close              = tap_dev_close,
 528         .dev_configure          = tap_dev_configure,
 529         .dev_infos_get          = tap_dev_info,
 530         .rx_queue_setup         = tap_rx_queue_setup,
 531         .tx_queue_setup         = tap_tx_queue_setup,
 532         .rx_queue_release       = tap_rx_queue_release,
 533         .tx_queue_release       = tap_tx_queue_release,
 534         .link_update            = tap_link_update,
 535         .stats_get              = tap_stats_get,
 536         .stats_reset            = tap_stats_reset,
 537 };
 538
 539 static int
 540 pmd_mac_address(int fd, struct rte_eth_dev *dev, struct ether_addr *addr)
 541 {
 542         struct ifreq ifr;
 543
 544         if ((fd <= 0) || !dev || !addr)
 545                 return -1;
 546
 547         memset(&ifr, 0, sizeof(ifr));
 548
 549         if (ioctl(fd, SIOCGIFHWADDR, &ifr) == -1) {
 550                 RTE_LOG(ERR, PMD, "ioctl failed (SIOCGIFHWADDR) (%s)\n",
 551                         ifr.ifr_name);
 552                 return -1;
 553         }
 554
 555         /* Set the host based MAC address to this special MAC format */
 556         ifr.ifr_hwaddr.sa_data[0] = 'T';
 557         ifr.ifr_hwaddr.sa_data[1] = 'a';
 558         ifr.ifr_hwaddr.sa_data[2] = 'p';
 559         ifr.ifr_hwaddr.sa_data[3] = '-';
 560         ifr.ifr_hwaddr.sa_data[4] = dev->data->port_id;
 561         ifr.ifr_hwaddr.sa_data[5] = dev->data->numa_node;
 562         if (ioctl(fd, SIOCSIFHWADDR, &ifr) == -1) {
 563                 RTE_LOG(ERR, PMD, "%s: ioctl failed (SIOCSIFHWADDR) (%s)\n",
 564                         dev->data->name, ifr.ifr_name);
 565                 return -1;
 566         }
 567
 568         /* Set the local application MAC address, needs to be different then
 569          * the host based MAC address.
 570          */
 571         ifr.ifr_hwaddr.sa_data[0] = 'd';
 572         ifr.ifr_hwaddr.sa_data[1] = 'n';
 573         ifr.ifr_hwaddr.sa_data[2] = 'e';
 574         ifr.ifr_hwaddr.sa_data[3] = 't';
 575         ifr.ifr_hwaddr.sa_data[4] = dev->data->port_id;
 576         ifr.ifr_hwaddr.sa_data[5] = dev->data->numa_node;
 577         rte_memcpy(addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
 578
 579         return 0;
 580 }
 581
 582 static int
 583 eth_dev_tap_create(const char *name, char *tap_name)
 584 {
 585         int numa_node = rte_socket_id();
 586         struct rte_eth_dev *dev = NULL;
 587         struct pmd_internals *pmd = NULL;
 588         struct rte_eth_dev_data *data = NULL;
 589         int i, fd = -1;
 590
 591         RTE_LOG(INFO, PMD,
 592                 "%s: Create TAP Ethernet device with %d queues on numa %u\n",
 593                  name, RTE_PMD_TAP_MAX_QUEUES, rte_socket_id());
 594
 595         data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
 596         if (!data) {
 597                 RTE_LOG(INFO, PMD, "Failed to allocate data\n");
 598                 goto error_exit;
 599         }
 600
 601         pmd = rte_zmalloc_socket(tap_name, sizeof(*pmd), 0, numa_node);
 602         if (!pmd) {
 603                 RTE_LOG(INFO, PMD, "Unable to allocate internal struct\n");
 604                 goto error_exit;
 605         }
 606
 607         /* Use the name and not the tap_name */
 608         dev = rte_eth_dev_allocate(tap_name);
 609         if (!dev) {
 610                 RTE_LOG(INFO, PMD, "Unable to allocate device struct\n");
 611                 goto error_exit;
 612         }
 613
 614         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
 615
 616         pmd->nb_queues = RTE_PMD_TAP_MAX_QUEUES;
 617
 618         /* Setup some default values */
 619         data->dev_private = pmd;
 620         data->port_id = dev->data->port_id;
 621         data->dev_flags = RTE_ETH_DEV_DETACHABLE;
 622         data->kdrv = RTE_KDRV_NONE;
 623         data->drv_name = pmd_tap_drv.driver.name;
 624         data->numa_node = numa_node;
 625
 626         data->dev_link = pmd_link;
 627         data->mac_addrs = &pmd->eth_addr;
 628         data->nb_rx_queues = pmd->nb_queues;
 629         data->nb_tx_queues = pmd->nb_queues;
 630
 631         dev->data = data;
 632         dev->dev_ops = &ops;
 633         dev->driver = NULL;
 634         dev->rx_pkt_burst = pmd_rx_burst;
 635         dev->tx_pkt_burst = pmd_tx_burst;
 636         snprintf(dev->data->name, sizeof(dev->data->name), "%s", name);
 637
 638         /* Create the first Tap device */
 639         fd = tun_alloc(tap_name);
 640         if (fd < 0) {
 641                 RTE_LOG(INFO, PMD, "tun_alloc() failed\n");
 642                 goto error_exit;
 643         }
 644
 645         /* Presetup the fds to -1 as being not working */
 646         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 647                 pmd->fds[i] = -1;
 648                 pmd->rxq[i].fd = -1;
 649                 pmd->txq[i].fd = -1;
 650         }
 651
 652         /* Take the TUN/TAP fd and place in the first location */
 653         pmd->rxq[0].fd = fd;
 654         pmd->txq[0].fd = fd;
 655         pmd->fds[0] = fd;
 656
 657         if (pmd_mac_address(fd, dev, &pmd->eth_addr) < 0) {
 658                 RTE_LOG(INFO, PMD, "Unable to get MAC address\n");
 659                 goto error_exit;
 660         }
 661
 662         return 0;
 663
 664 error_exit:
 665         RTE_PMD_DEBUG_TRACE("Unable to initialize %s\n", name);
 666
 667         rte_free(data);
 668         rte_free(pmd);
 669
 670         rte_eth_dev_release_port(dev);
 671
 672         return -EINVAL;
 673 }
 674
 675 static int
 676 set_interface_name(const char *key __rte_unused,
 677                    const char *value,
 678                    void *extra_args)
 679 {
 680         char *name = (char *)extra_args;
 681
 682         if (value)
 683                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
 684         else
 685                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
 686                          DEFAULT_TAP_NAME, (tap_unit - 1));
 687
 688         return 0;
 689 }
 690
 691 static int
 692 set_interface_speed(const char *key __rte_unused,
 693                     const char *value,
 694                     void *extra_args)
 695 {
 696         *(int *)extra_args = (value) ? atoi(value) : ETH_SPEED_NUM_10G;
 697
 698         return 0;
 699 }
 700
 701 /* Open a TAP interface device.
 702  */
 703 static int
 704 rte_pmd_tap_probe(const char *name, const char *params)
 705 {
 706         int ret;
 707         struct rte_kvargs *kvlist = NULL;
 708         int speed;
 709         char tap_name[RTE_ETH_NAME_MAX_LEN];
 710
 711         speed = ETH_SPEED_NUM_10G;
 712         snprintf(tap_name, sizeof(tap_name), "%s%d",
 713                  DEFAULT_TAP_NAME, tap_unit++);
 714
 715         RTE_LOG(INFO, PMD, "Initializing pmd_tap for %s as %s\n",
 716                 name, tap_name);
 717
 718         if (params && (params[0] != '\0')) {
 719                 RTE_LOG(INFO, PMD, "paramaters (%s)\n", params);
 720
 721                 kvlist = rte_kvargs_parse(params, valid_arguments);
 722                 if (kvlist) {
 723                         if (rte_kvargs_count(kvlist, ETH_TAP_SPEED_ARG) == 1) {
 724                                 ret = rte_kvargs_process(kvlist,
 725                                                          ETH_TAP_SPEED_ARG,
 726                                                          &set_interface_speed,
 727                                                          &speed);
 728                                 if (ret == -1)
 729                                         goto leave;
 730                         }
 731
 732                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
 733                                 ret = rte_kvargs_process(kvlist,
 734                                                          ETH_TAP_IFACE_ARG,
 735                                                          &set_interface_name,
 736                                                          tap_name);
 737                                 if (ret == -1)
 738                                         goto leave;
 739                         }
 740                 }
 741         }
 742         pmd_link.link_speed = speed;
 743
 744         ret = eth_dev_tap_create(name, tap_name);
 745
 746 leave:
 747         if (ret == -1) {
 748                 RTE_LOG(INFO, PMD, "Failed to create pmd for %s as %s\n",
 749                         name, tap_name);
 750                 tap_unit--;             /* Restore the unit number */
 751         }
 752         rte_kvargs_free(kvlist);
 753
 754         return ret;
 755 }
 756
 757 /* detach a TAP device.
 758  */
 759 static int
 760 rte_pmd_tap_remove(const char *name)
 761 {
 762         struct rte_eth_dev *eth_dev = NULL;
 763         struct pmd_internals *internals;
 764         int i;
 765
 766         RTE_LOG(INFO, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
 767                 rte_socket_id());
 768
 769         /* find the ethdev entry */
 770         eth_dev = rte_eth_dev_allocated(name);
 771         if (!eth_dev)
 772                 return 0;
 773
 774         internals = eth_dev->data->dev_private;
 775         for (i = 0; i < internals->nb_queues; i++)
 776                 if (internals->fds[i] != -1)
 777                         close(internals->fds[i]);
 778
 779         rte_free(eth_dev->data->dev_private);
 780         rte_free(eth_dev->data);
 781
 782         rte_eth_dev_release_port(eth_dev);
 783
 784         return 0;
 785 }
 786
 787 static struct rte_vdev_driver pmd_tap_drv = {
 788         .probe = rte_pmd_tap_probe,
 789         .remove = rte_pmd_tap_remove,
 790 };
 791 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
 792 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
 793 RTE_PMD_REGISTER_PARAM_STRING(net_tap, "iface=<string>,speed=N");