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 tap_link_set_flags(struct pmd_internals *pmd, short flags, int add)
 279 {
 280         struct ifreq ifr;
 281         int err, s;
 282
 283         /*
 284          * An AF_INET/DGRAM socket is needed for
 285          * SIOCGIFFLAGS/SIOCSIFFLAGS, using fd won't work.
 286          */
 287         s = socket(AF_INET, SOCK_DGRAM, 0);
 288         if (s < 0) {
 289                 RTE_LOG(ERR, PMD,
 290                         "Unable to get a socket to set flags: %s\n",
 291                         strerror(errno));
 292                 return -1;
 293         }
 294         memset(&ifr, 0, sizeof(ifr));
 295         strncpy(ifr.ifr_name, pmd->name, IFNAMSIZ);
 296         err = ioctl(s, SIOCGIFFLAGS, &ifr);
 297         if (err < 0) {
 298                 RTE_LOG(ERR, PMD, "Unable to get tap netdevice flags: %s\n",
 299                         strerror(errno));
 300                 close(s);
 301                 return -1;
 302         }
 303         if (add)
 304                 ifr.ifr_flags |= flags;
 305         else
 306                 ifr.ifr_flags &= ~flags;
 307         err = ioctl(s, SIOCSIFFLAGS, &ifr);
 308         if (err < 0) {
 309                 RTE_LOG(ERR, PMD, "Unable to %s flags 0x%x: %s\n",
 310                         add ? "set" : "unset", flags, strerror(errno));
 311                 close(s);
 312                 return -1;
 313         }
 314         close(s);
 315
 316         return 0;
 317 }
 318
 319 static int
 320 tap_link_set_down(struct rte_eth_dev *dev)
 321 {
 322         struct pmd_internals *pmd = dev->data->dev_private;
 323
 324         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 325         return tap_link_set_flags(pmd, IFF_UP | IFF_NOARP, 0);
 326 }
 327
 328 static int
 329 tap_link_set_up(struct rte_eth_dev *dev)
 330 {
 331         struct pmd_internals *pmd = dev->data->dev_private;
 332
 333         dev->data->dev_link.link_status = ETH_LINK_UP;
 334         return tap_link_set_flags(pmd, IFF_UP | IFF_NOARP, 1);
 335 }
 336
 337 static int
 338 tap_dev_start(struct rte_eth_dev *dev)
 339 {
 340         return tap_link_set_up(dev);
 341 }
 342
 343 /* This function gets called when the current port gets stopped.
 344  */
 345 static void
 346 tap_dev_stop(struct rte_eth_dev *dev)
 347 {
 348         int i;
 349         struct pmd_internals *internals = dev->data->dev_private;
 350
 351         for (i = 0; i < internals->nb_queues; i++)
 352                 if (internals->fds[i] != -1)
 353                         close(internals->fds[i]);
 354         tap_link_set_down(dev);
 355 }
 356
 357 static int
 358 tap_dev_configure(struct rte_eth_dev *dev __rte_unused)
 359 {
 360         return 0;
 361 }
 362
 363 static void
 364 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 365 {
 366         struct pmd_internals *internals = dev->data->dev_private;
 367
 368         dev_info->if_index = internals->if_index;
 369         dev_info->max_mac_addrs = 1;
 370         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 371         dev_info->max_rx_queues = internals->nb_queues;
 372         dev_info->max_tx_queues = internals->nb_queues;
 373         dev_info->min_rx_bufsize = 0;
 374         dev_info->pci_dev = NULL;
 375 }
 376
 377 static void
 378 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 379 {
 380         unsigned int i, imax;
 381         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 382         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 383         const struct pmd_internals *pmd = dev->data->dev_private;
 384
 385         imax = (pmd->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 386                 pmd->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 387
 388         for (i = 0; i < imax; i++) {
 389                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 390                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 391                 rx_total += tap_stats->q_ipackets[i];
 392                 rx_bytes_total += tap_stats->q_ibytes[i];
 393         }
 394
 395         for (i = 0; i < imax; i++) {
 396                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 397                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 398                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 399                 tx_total += tap_stats->q_opackets[i];
 400                 tx_err_total += tap_stats->q_errors[i];
 401                 tx_bytes_total += tap_stats->q_obytes[i];
 402         }
 403
 404         tap_stats->ipackets = rx_total;
 405         tap_stats->ibytes = rx_bytes_total;
 406         tap_stats->opackets = tx_total;
 407         tap_stats->oerrors = tx_err_total;
 408         tap_stats->obytes = tx_bytes_total;
 409 }
 410
 411 static void
 412 tap_stats_reset(struct rte_eth_dev *dev)
 413 {
 414         int i;
 415         struct pmd_internals *pmd = dev->data->dev_private;
 416
 417         for (i = 0; i < pmd->nb_queues; i++) {
 418                 pmd->rxq[i].stats.ipackets = 0;
 419                 pmd->rxq[i].stats.ibytes = 0;
 420         }
 421
 422         for (i = 0; i < pmd->nb_queues; i++) {
 423                 pmd->txq[i].stats.opackets = 0;
 424                 pmd->txq[i].stats.errs = 0;
 425                 pmd->txq[i].stats.obytes = 0;
 426         }
 427 }
 428
 429 static void
 430 tap_dev_close(struct rte_eth_dev *dev __rte_unused)
 431 {
 432 }
 433
 434 static void
 435 tap_rx_queue_release(void *queue)
 436 {
 437         struct rx_queue *rxq = queue;
 438
 439         if (rxq && (rxq->fd > 0)) {
 440                 close(rxq->fd);
 441                 rxq->fd = -1;
 442         }
 443 }
 444
 445 static void
 446 tap_tx_queue_release(void *queue)
 447 {
 448         struct tx_queue *txq = queue;
 449
 450         if (txq && (txq->fd > 0)) {
 451                 close(txq->fd);
 452                 txq->fd = -1;
 453         }
 454 }
 455
 456 static int
 457 tap_link_update(struct rte_eth_dev *dev __rte_unused,
 458                 int wait_to_complete __rte_unused)
 459 {
 460         return 0;
 461 }
 462
 463 static void
 464 tap_promisc_enable(struct rte_eth_dev *dev)
 465 {
 466         struct pmd_internals *pmd = dev->data->dev_private;
 467
 468         dev->data->promiscuous = 1;
 469         tap_link_set_flags(pmd, IFF_PROMISC, 1);
 470 }
 471
 472 static void
 473 tap_promisc_disable(struct rte_eth_dev *dev)
 474 {
 475         struct pmd_internals *pmd = dev->data->dev_private;
 476
 477         dev->data->promiscuous = 0;
 478         tap_link_set_flags(pmd, IFF_PROMISC, 0);
 479 }
 480
 481 static void
 482 tap_allmulti_enable(struct rte_eth_dev *dev)
 483 {
 484         struct pmd_internals *pmd = dev->data->dev_private;
 485
 486         dev->data->all_multicast = 1;
 487         tap_link_set_flags(pmd, IFF_ALLMULTI, 1);
 488 }
 489
 490 static void
 491 tap_allmulti_disable(struct rte_eth_dev *dev)
 492 {
 493         struct pmd_internals *pmd = dev->data->dev_private;
 494
 495         dev->data->all_multicast = 0;
 496         tap_link_set_flags(pmd, IFF_ALLMULTI, 0);
 497 }
 498
 499 static int
 500 tap_setup_queue(struct rte_eth_dev *dev,
 501                 struct pmd_internals *internals,
 502                 uint16_t qid)
 503 {
 504         struct pmd_internals *pmd = dev->data->dev_private;
 505         struct rx_queue *rx = &internals->rxq[qid];
 506         struct tx_queue *tx = &internals->txq[qid];
 507         int fd;
 508
 509         fd = rx->fd;
 510         if (fd < 0) {
 511                 fd = tx->fd;
 512                 if (fd < 0) {
 513                         RTE_LOG(INFO, PMD, "Add queue to TAP %s for qid %d\n",
 514                                 pmd->name, qid);
 515                         fd = tun_alloc(pmd->name);
 516                         if (fd < 0) {
 517                                 RTE_LOG(ERR, PMD, "tun_alloc(%s) failed\n",
 518                                         pmd->name);
 519                                 return -1;
 520                         }
 521                 }
 522         }
 523
 524         rx->fd = fd;
 525         tx->fd = fd;
 526
 527         return fd;
 528 }
 529
 530 static int
 531 rx_setup_queue(struct rte_eth_dev *dev,
 532                 struct pmd_internals *internals,
 533                 uint16_t qid)
 534 {
 535         dev->data->rx_queues[qid] = &internals->rxq[qid];
 536
 537         return tap_setup_queue(dev, internals, qid);
 538 }
 539
 540 static int
 541 tx_setup_queue(struct rte_eth_dev *dev,
 542                 struct pmd_internals *internals,
 543                 uint16_t qid)
 544 {
 545         dev->data->tx_queues[qid] = &internals->txq[qid];
 546
 547         return tap_setup_queue(dev, internals, qid);
 548 }
 549
 550 static int
 551 tap_rx_queue_setup(struct rte_eth_dev *dev,
 552                    uint16_t rx_queue_id,
 553                    uint16_t nb_rx_desc __rte_unused,
 554                    unsigned int socket_id __rte_unused,
 555                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 556                    struct rte_mempool *mp)
 557 {
 558         struct pmd_internals *internals = dev->data->dev_private;
 559         uint16_t buf_size;
 560         int fd;
 561
 562         if ((rx_queue_id >= internals->nb_queues) || !mp) {
 563                 RTE_LOG(ERR, PMD, "nb_queues %d mp %p\n",
 564                         internals->nb_queues, mp);
 565                 return -1;
 566         }
 567
 568         internals->rxq[rx_queue_id].mp = mp;
 569         internals->rxq[rx_queue_id].in_port = dev->data->port_id;
 570
 571         /* Now get the space available for data in the mbuf */
 572         buf_size = (uint16_t)(rte_pktmbuf_data_room_size(mp) -
 573                                 RTE_PKTMBUF_HEADROOM);
 574
 575         if (buf_size < ETH_FRAME_LEN) {
 576                 RTE_LOG(ERR, PMD,
 577                         "%s: %d bytes will not fit in mbuf (%d bytes)\n",
 578                         dev->data->name, ETH_FRAME_LEN, buf_size);
 579                 return -ENOMEM;
 580         }
 581
 582         fd = rx_setup_queue(dev, internals, rx_queue_id);
 583         if (fd == -1)
 584                 return -1;
 585
 586         internals->fds[rx_queue_id] = fd;
 587         RTE_LOG(INFO, PMD, "RX TAP device name %s, qid %d on fd %d\n",
 588                 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
 589
 590         return 0;
 591 }
 592
 593 static int
 594 tap_tx_queue_setup(struct rte_eth_dev *dev,
 595                    uint16_t tx_queue_id,
 596                    uint16_t nb_tx_desc __rte_unused,
 597                    unsigned int socket_id __rte_unused,
 598                    const struct rte_eth_txconf *tx_conf __rte_unused)
 599 {
 600         struct pmd_internals *internals = dev->data->dev_private;
 601         int ret;
 602
 603         if (tx_queue_id >= internals->nb_queues)
 604                 return -1;
 605
 606         ret = tx_setup_queue(dev, internals, tx_queue_id);
 607         if (ret == -1)
 608                 return -1;
 609
 610         RTE_LOG(INFO, PMD, "TX TAP device name %s, qid %d on fd %d\n",
 611                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd);
 612
 613         return 0;
 614 }
 615
 616 static const struct eth_dev_ops ops = {
 617         .dev_start              = tap_dev_start,
 618         .dev_stop               = tap_dev_stop,
 619         .dev_close              = tap_dev_close,
 620         .dev_configure          = tap_dev_configure,
 621         .dev_infos_get          = tap_dev_info,
 622         .rx_queue_setup         = tap_rx_queue_setup,
 623         .tx_queue_setup         = tap_tx_queue_setup,
 624         .rx_queue_release       = tap_rx_queue_release,
 625         .tx_queue_release       = tap_tx_queue_release,
 626         .link_update            = tap_link_update,
 627         .dev_set_link_up        = tap_link_set_up,
 628         .dev_set_link_down      = tap_link_set_down,
 629         .promiscuous_enable     = tap_promisc_enable,
 630         .promiscuous_disable    = tap_promisc_disable,
 631         .allmulticast_enable    = tap_allmulti_enable,
 632         .allmulticast_disable   = tap_allmulti_disable,
 633         .stats_get              = tap_stats_get,
 634         .stats_reset            = tap_stats_reset,
 635 };
 636
 637 static int
 638 pmd_mac_address(int fd, struct rte_eth_dev *dev, struct ether_addr *addr)
 639 {
 640         struct ifreq ifr;
 641
 642         if ((fd <= 0) || !dev || !addr)
 643                 return -1;
 644
 645         memset(&ifr, 0, sizeof(ifr));
 646
 647         if (ioctl(fd, SIOCGIFHWADDR, &ifr) == -1) {
 648                 RTE_LOG(ERR, PMD, "ioctl failed (SIOCGIFHWADDR) (%s)\n",
 649                         ifr.ifr_name);
 650                 return -1;
 651         }
 652
 653         rte_memcpy(addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
 654
 655         return 0;
 656 }
 657
 658 static int
 659 eth_dev_tap_create(const char *name, char *tap_name)
 660 {
 661         int numa_node = rte_socket_id();
 662         struct rte_eth_dev *dev = NULL;
 663         struct pmd_internals *pmd = NULL;
 664         struct rte_eth_dev_data *data = NULL;
 665         int i, fd = -1;
 666
 667         RTE_LOG(INFO, PMD,
 668                 "%s: Create TAP Ethernet device with %d queues on numa %u\n",
 669                  name, RTE_PMD_TAP_MAX_QUEUES, rte_socket_id());
 670
 671         data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
 672         if (!data) {
 673                 RTE_LOG(ERR, PMD, "Failed to allocate data\n");
 674                 goto error_exit;
 675         }
 676
 677         pmd = rte_zmalloc_socket(tap_name, sizeof(*pmd), 0, numa_node);
 678         if (!pmd) {
 679                 RTE_LOG(ERR, PMD, "Unable to allocate internal struct\n");
 680                 goto error_exit;
 681         }
 682
 683         /* Use the name and not the tap_name */
 684         dev = rte_eth_dev_allocate(tap_name);
 685         if (!dev) {
 686                 RTE_LOG(ERR, PMD, "Unable to allocate device struct\n");
 687                 goto error_exit;
 688         }
 689
 690         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
 691
 692         pmd->nb_queues = RTE_PMD_TAP_MAX_QUEUES;
 693
 694         /* Setup some default values */
 695         data->dev_private = pmd;
 696         data->port_id = dev->data->port_id;
 697         data->dev_flags = RTE_ETH_DEV_DETACHABLE;
 698         data->kdrv = RTE_KDRV_NONE;
 699         data->drv_name = pmd_tap_drv.driver.name;
 700         data->numa_node = numa_node;
 701
 702         data->dev_link = pmd_link;
 703         data->mac_addrs = &pmd->eth_addr;
 704         data->nb_rx_queues = pmd->nb_queues;
 705         data->nb_tx_queues = pmd->nb_queues;
 706
 707         dev->data = data;
 708         dev->dev_ops = &ops;
 709         dev->driver = NULL;
 710         dev->rx_pkt_burst = pmd_rx_burst;
 711         dev->tx_pkt_burst = pmd_tx_burst;
 712         snprintf(dev->data->name, sizeof(dev->data->name), "%s", name);
 713
 714         /* Create the first Tap device */
 715         fd = tun_alloc(tap_name);
 716         if (fd < 0) {
 717                 RTE_LOG(ERR, PMD, "tun_alloc() failed\n");
 718                 goto error_exit;
 719         }
 720
 721         /* Presetup the fds to -1 as being not working */
 722         for (i = 1; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 723                 pmd->fds[i] = -1;
 724                 pmd->rxq[i].fd = -1;
 725                 pmd->txq[i].fd = -1;
 726         }
 727
 728         /* Take the TUN/TAP fd and place in the first location */
 729         pmd->rxq[0].fd = fd;
 730         pmd->txq[0].fd = fd;
 731         pmd->fds[0] = fd;
 732
 733         if (pmd_mac_address(fd, dev, &pmd->eth_addr) < 0) {
 734                 RTE_LOG(ERR, PMD, "Unable to get MAC address\n");
 735                 goto error_exit;
 736         }
 737
 738         return 0;
 739
 740 error_exit:
 741         RTE_PMD_DEBUG_TRACE("Unable to initialize %s\n", name);
 742
 743         rte_free(data);
 744         rte_free(pmd);
 745
 746         rte_eth_dev_release_port(dev);
 747
 748         return -EINVAL;
 749 }
 750
 751 static int
 752 set_interface_name(const char *key __rte_unused,
 753                    const char *value,
 754                    void *extra_args)
 755 {
 756         char *name = (char *)extra_args;
 757
 758         if (value)
 759                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
 760         else
 761                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
 762                          DEFAULT_TAP_NAME, (tap_unit - 1));
 763
 764         return 0;
 765 }
 766
 767 static int
 768 set_interface_speed(const char *key __rte_unused,
 769                     const char *value,
 770                     void *extra_args)
 771 {
 772         *(int *)extra_args = (value) ? atoi(value) : ETH_SPEED_NUM_10G;
 773
 774         return 0;
 775 }
 776
 777 /* Open a TAP interface device.
 778  */
 779 static int
 780 rte_pmd_tap_probe(const char *name, const char *params)
 781 {
 782         int ret;
 783         struct rte_kvargs *kvlist = NULL;
 784         int speed;
 785         char tap_name[RTE_ETH_NAME_MAX_LEN];
 786
 787         speed = ETH_SPEED_NUM_10G;
 788         snprintf(tap_name, sizeof(tap_name), "%s%d",
 789                  DEFAULT_TAP_NAME, tap_unit++);
 790
 791         if (params && (params[0] != '\0')) {
 792                 RTE_LOG(INFO, PMD, "paramaters (%s)\n", params);
 793
 794                 kvlist = rte_kvargs_parse(params, valid_arguments);
 795                 if (kvlist) {
 796                         if (rte_kvargs_count(kvlist, ETH_TAP_SPEED_ARG) == 1) {
 797                                 ret = rte_kvargs_process(kvlist,
 798                                                          ETH_TAP_SPEED_ARG,
 799                                                          &set_interface_speed,
 800                                                          &speed);
 801                                 if (ret == -1)
 802                                         goto leave;
 803                         }
 804
 805                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
 806                                 ret = rte_kvargs_process(kvlist,
 807                                                          ETH_TAP_IFACE_ARG,
 808                                                          &set_interface_name,
 809                                                          tap_name);
 810                                 if (ret == -1)
 811                                         goto leave;
 812                         }
 813                 }
 814         }
 815         pmd_link.link_speed = speed;
 816
 817         RTE_LOG(INFO, PMD, "Initializing pmd_tap for %s as %s\n",
 818                 name, tap_name);
 819
 820         ret = eth_dev_tap_create(name, tap_name);
 821
 822 leave:
 823         if (ret == -1) {
 824                 RTE_LOG(INFO, PMD, "Failed to create pmd for %s as %s\n",
 825                         name, tap_name);
 826                 tap_unit--;             /* Restore the unit number */
 827         }
 828         rte_kvargs_free(kvlist);
 829
 830         return ret;
 831 }
 832
 833 /* detach a TAP device.
 834  */
 835 static int
 836 rte_pmd_tap_remove(const char *name)
 837 {
 838         struct rte_eth_dev *eth_dev = NULL;
 839         struct pmd_internals *internals;
 840         int i;
 841
 842         RTE_LOG(INFO, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
 843                 rte_socket_id());
 844
 845         /* find the ethdev entry */
 846         eth_dev = rte_eth_dev_allocated(name);
 847         if (!eth_dev)
 848                 return 0;
 849
 850         internals = eth_dev->data->dev_private;
 851         for (i = 0; i < internals->nb_queues; i++)
 852                 if (internals->fds[i] != -1)
 853                         close(internals->fds[i]);
 854
 855         rte_free(eth_dev->data->dev_private);
 856         rte_free(eth_dev->data);
 857
 858         rte_eth_dev_release_port(eth_dev);
 859
 860         return 0;
 861 }
 862
 863 static struct rte_vdev_driver pmd_tap_drv = {
 864         .probe = rte_pmd_tap_probe,
 865         .remove = rte_pmd_tap_remove,
 866 };
 867 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
 868 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
 869 RTE_PMD_REGISTER_PARAM_STRING(net_tap, "iface=<string>,speed=N");