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
 107         struct rx_queue rxq[RTE_PMD_TAP_MAX_QUEUES];    /* List of RX queues */
 108         struct tx_queue txq[RTE_PMD_TAP_MAX_QUEUES];    /* List of TX queues */
 109 };
 110
 111 /* Tun/Tap allocation routine
 112  *
 113  * name is the number of the interface to use, unless NULL to take the host
 114  * supplied name.
 115  */
 116 static int
 117 tun_alloc(char *name)
 118 {
 119         struct ifreq ifr;
 120         unsigned int features;
 121         int fd;
 122
 123         memset(&ifr, 0, sizeof(struct ifreq));
 124
 125         ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
 126         if (name && name[0])
 127                 strncpy(ifr.ifr_name, name, IFNAMSIZ);
 128
 129         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 130         if (fd < 0) {
 131                 RTE_LOG(ERR, PMD, "Unable to create TAP interface");
 132                 goto error;
 133         }
 134
 135         /* Grab the TUN features to verify we can work */
 136         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 137                 RTE_LOG(ERR, PMD, "Unable to get TUN/TAP features\n");
 138                 goto error;
 139         }
 140         RTE_LOG(DEBUG, PMD, "TUN/TAP Features %08x\n", features);
 141
 142 #ifdef IFF_MULTI_QUEUE
 143         if (!(features & IFF_MULTI_QUEUE) && (RTE_PMD_TAP_MAX_QUEUES > 1)) {
 144                 RTE_LOG(DEBUG, PMD, "TUN/TAP device only one queue\n");
 145                 goto error;
 146         } else if ((features & IFF_ONE_QUEUE) &&
 147                         (RTE_PMD_TAP_MAX_QUEUES == 1)) {
 148                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 149                 RTE_LOG(DEBUG, PMD, "Single queue only support\n");
 150         } else {
 151                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 152                 RTE_LOG(DEBUG, PMD, "Multi-queue support for %d queues\n",
 153                         RTE_PMD_TAP_MAX_QUEUES);
 154         }
 155 #else
 156         if (RTE_PMD_TAP_MAX_QUEUES > 1) {
 157                 RTE_LOG(DEBUG, PMD, "TUN/TAP device only one queue\n");
 158                 goto error;
 159         } else {
 160                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 161                 RTE_LOG(DEBUG, PMD, "Single queue only support\n");
 162         }
 163 #endif
 164
 165         /* Set the TUN/TAP configuration and get the name if needed */
 166         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 167                 RTE_LOG(ERR, PMD, "Unable to set TUNSETIFF for %s\n",
 168                         ifr.ifr_name);
 169                 perror("TUNSETIFF");
 170                 goto error;
 171         }
 172
 173         /* Always set the file descriptor to non-blocking */
 174         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 175                 RTE_LOG(ERR, PMD, "Unable to set to nonblocking\n");
 176                 perror("F_SETFL, NONBLOCK");
 177                 goto error;
 178         }
 179
 180         /* If the name is different that new name as default */
 181         if (name && strcmp(name, ifr.ifr_name))
 182                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", ifr.ifr_name);
 183
 184         return fd;
 185
 186 error:
 187         if (fd > 0)
 188                 close(fd);
 189         return -1;
 190 }
 191
 192 /* Callback to handle the rx burst of packets to the correct interface and
 193  * file descriptor(s) in a multi-queue setup.
 194  */
 195 static uint16_t
 196 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 197 {
 198         int len;
 199         struct rte_mbuf *mbuf;
 200         struct rx_queue *rxq = queue;
 201         uint16_t num_rx;
 202         unsigned long num_rx_bytes = 0;
 203
 204         for (num_rx = 0; num_rx < nb_pkts; ) {
 205                 /* allocate the next mbuf */
 206                 mbuf = rte_pktmbuf_alloc(rxq->mp);
 207                 if (unlikely(!mbuf)) {
 208                         RTE_LOG(WARNING, PMD, "Unable to allocate mbuf\n");
 209                         break;
 210                 }
 211
 212                 len = read(rxq->fd, rte_pktmbuf_mtod(mbuf, char *),
 213                            rte_pktmbuf_tailroom(mbuf));
 214                 if (len <= 0) {
 215                         rte_pktmbuf_free(mbuf);
 216                         break;
 217                 }
 218
 219                 mbuf->data_len = len;
 220                 mbuf->pkt_len = len;
 221                 mbuf->port = rxq->in_port;
 222
 223                 /* account for the receive frame */
 224                 bufs[num_rx++] = mbuf;
 225                 num_rx_bytes += mbuf->pkt_len;
 226         }
 227         rxq->stats.ipackets += num_rx;
 228         rxq->stats.ibytes += num_rx_bytes;
 229
 230         return num_rx;
 231 }
 232
 233 /* Callback to handle sending packets from the tap interface
 234  */
 235 static uint16_t
 236 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 237 {
 238         struct rte_mbuf *mbuf;
 239         struct tx_queue *txq = queue;
 240         struct pollfd pfd;
 241         uint16_t num_tx = 0;
 242         unsigned long num_tx_bytes = 0;
 243         int i, n;
 244
 245         if (unlikely(nb_pkts == 0))
 246                 return 0;
 247
 248         pfd.events = POLLOUT;
 249         pfd.fd = txq->fd;
 250         for (i = 0; i < nb_pkts; i++) {
 251                 n = poll(&pfd, 1, 0);
 252
 253                 if (n <= 0)
 254                         break;
 255
 256                 if (pfd.revents & POLLOUT) {
 257                         /* copy the tx frame data */
 258                         mbuf = bufs[num_tx];
 259                         n = write(pfd.fd, rte_pktmbuf_mtod(mbuf, void*),
 260                                   rte_pktmbuf_pkt_len(mbuf));
 261                         if (n <= 0)
 262                                 break;
 263
 264                         num_tx++;
 265                         num_tx_bytes += mbuf->pkt_len;
 266                         rte_pktmbuf_free(mbuf);
 267                 }
 268         }
 269
 270         txq->stats.opackets += num_tx;
 271         txq->stats.errs += nb_pkts - num_tx;
 272         txq->stats.obytes += num_tx_bytes;
 273
 274         return num_tx;
 275 }
 276
 277 static int tap_link_set_flags(struct pmd_internals *pmd, short flags, int add)
 278 {
 279         struct ifreq ifr;
 280         int err, s;
 281
 282         /*
 283          * An AF_INET/DGRAM socket is needed for
 284          * SIOCGIFFLAGS/SIOCSIFFLAGS, using fd won't work.
 285          */
 286         s = socket(AF_INET, SOCK_DGRAM, 0);
 287         if (s < 0) {
 288                 RTE_LOG(ERR, PMD,
 289                         "Unable to get a socket to set flags: %s\n",
 290                         strerror(errno));
 291                 return -1;
 292         }
 293         memset(&ifr, 0, sizeof(ifr));
 294         strncpy(ifr.ifr_name, pmd->name, IFNAMSIZ);
 295         err = ioctl(s, SIOCGIFFLAGS, &ifr);
 296         if (err < 0) {
 297                 RTE_LOG(ERR, PMD, "Unable to get tap netdevice flags: %s\n",
 298                         strerror(errno));
 299                 close(s);
 300                 return -1;
 301         }
 302         if (add)
 303                 ifr.ifr_flags |= flags;
 304         else
 305                 ifr.ifr_flags &= ~flags;
 306         err = ioctl(s, SIOCSIFFLAGS, &ifr);
 307         if (err < 0) {
 308                 RTE_LOG(ERR, PMD, "Unable to %s flags 0x%x: %s\n",
 309                         add ? "set" : "unset", flags, strerror(errno));
 310                 close(s);
 311                 return -1;
 312         }
 313         close(s);
 314
 315         return 0;
 316 }
 317
 318 static int
 319 tap_link_set_down(struct rte_eth_dev *dev)
 320 {
 321         struct pmd_internals *pmd = dev->data->dev_private;
 322
 323         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 324         return tap_link_set_flags(pmd, IFF_UP | IFF_NOARP, 0);
 325 }
 326
 327 static int
 328 tap_link_set_up(struct rte_eth_dev *dev)
 329 {
 330         struct pmd_internals *pmd = dev->data->dev_private;
 331
 332         dev->data->dev_link.link_status = ETH_LINK_UP;
 333         return tap_link_set_flags(pmd, IFF_UP | IFF_NOARP, 1);
 334 }
 335
 336 static int
 337 tap_dev_start(struct rte_eth_dev *dev)
 338 {
 339         return tap_link_set_up(dev);
 340 }
 341
 342 /* This function gets called when the current port gets stopped.
 343  */
 344 static void
 345 tap_dev_stop(struct rte_eth_dev *dev)
 346 {
 347         int i;
 348         struct pmd_internals *internals = dev->data->dev_private;
 349
 350         for (i = 0; i < internals->nb_queues; i++)
 351                 if (internals->rxq[i].fd != -1)
 352                         close(internals->rxq[i].fd);
 353         tap_link_set_down(dev);
 354 }
 355
 356 static int
 357 tap_dev_configure(struct rte_eth_dev *dev __rte_unused)
 358 {
 359         return 0;
 360 }
 361
 362 static void
 363 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 364 {
 365         struct pmd_internals *internals = dev->data->dev_private;
 366
 367         dev_info->if_index = internals->if_index;
 368         dev_info->max_mac_addrs = 1;
 369         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 370         dev_info->max_rx_queues = internals->nb_queues;
 371         dev_info->max_tx_queues = internals->nb_queues;
 372         dev_info->min_rx_bufsize = 0;
 373         dev_info->pci_dev = NULL;
 374 }
 375
 376 static void
 377 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 378 {
 379         unsigned int i, imax;
 380         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 381         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 382         const struct pmd_internals *pmd = dev->data->dev_private;
 383
 384         imax = (pmd->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 385                 pmd->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 386
 387         for (i = 0; i < imax; i++) {
 388                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 389                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 390                 rx_total += tap_stats->q_ipackets[i];
 391                 rx_bytes_total += tap_stats->q_ibytes[i];
 392         }
 393
 394         for (i = 0; i < imax; i++) {
 395                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 396                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 397                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 398                 tx_total += tap_stats->q_opackets[i];
 399                 tx_err_total += tap_stats->q_errors[i];
 400                 tx_bytes_total += tap_stats->q_obytes[i];
 401         }
 402
 403         tap_stats->ipackets = rx_total;
 404         tap_stats->ibytes = rx_bytes_total;
 405         tap_stats->opackets = tx_total;
 406         tap_stats->oerrors = tx_err_total;
 407         tap_stats->obytes = tx_bytes_total;
 408 }
 409
 410 static void
 411 tap_stats_reset(struct rte_eth_dev *dev)
 412 {
 413         int i;
 414         struct pmd_internals *pmd = dev->data->dev_private;
 415
 416         for (i = 0; i < pmd->nb_queues; i++) {
 417                 pmd->rxq[i].stats.ipackets = 0;
 418                 pmd->rxq[i].stats.ibytes = 0;
 419         }
 420
 421         for (i = 0; i < pmd->nb_queues; i++) {
 422                 pmd->txq[i].stats.opackets = 0;
 423                 pmd->txq[i].stats.errs = 0;
 424                 pmd->txq[i].stats.obytes = 0;
 425         }
 426 }
 427
 428 static void
 429 tap_dev_close(struct rte_eth_dev *dev __rte_unused)
 430 {
 431 }
 432
 433 static void
 434 tap_rx_queue_release(void *queue)
 435 {
 436         struct rx_queue *rxq = queue;
 437
 438         if (rxq && (rxq->fd > 0)) {
 439                 close(rxq->fd);
 440                 rxq->fd = -1;
 441         }
 442 }
 443
 444 static void
 445 tap_tx_queue_release(void *queue)
 446 {
 447         struct tx_queue *txq = queue;
 448
 449         if (txq && (txq->fd > 0)) {
 450                 close(txq->fd);
 451                 txq->fd = -1;
 452         }
 453 }
 454
 455 static int
 456 tap_link_update(struct rte_eth_dev *dev __rte_unused,
 457                 int wait_to_complete __rte_unused)
 458 {
 459         return 0;
 460 }
 461
 462 static void
 463 tap_promisc_enable(struct rte_eth_dev *dev)
 464 {
 465         struct pmd_internals *pmd = dev->data->dev_private;
 466
 467         dev->data->promiscuous = 1;
 468         tap_link_set_flags(pmd, IFF_PROMISC, 1);
 469 }
 470
 471 static void
 472 tap_promisc_disable(struct rte_eth_dev *dev)
 473 {
 474         struct pmd_internals *pmd = dev->data->dev_private;
 475
 476         dev->data->promiscuous = 0;
 477         tap_link_set_flags(pmd, IFF_PROMISC, 0);
 478 }
 479
 480 static void
 481 tap_allmulti_enable(struct rte_eth_dev *dev)
 482 {
 483         struct pmd_internals *pmd = dev->data->dev_private;
 484
 485         dev->data->all_multicast = 1;
 486         tap_link_set_flags(pmd, IFF_ALLMULTI, 1);
 487 }
 488
 489 static void
 490 tap_allmulti_disable(struct rte_eth_dev *dev)
 491 {
 492         struct pmd_internals *pmd = dev->data->dev_private;
 493
 494         dev->data->all_multicast = 0;
 495         tap_link_set_flags(pmd, IFF_ALLMULTI, 0);
 496 }
 497
 498 static int
 499 tap_setup_queue(struct rte_eth_dev *dev,
 500                 struct pmd_internals *internals,
 501                 uint16_t qid)
 502 {
 503         struct pmd_internals *pmd = dev->data->dev_private;
 504         struct rx_queue *rx = &internals->rxq[qid];
 505         struct tx_queue *tx = &internals->txq[qid];
 506         int fd;
 507
 508         fd = rx->fd;
 509         if (fd < 0) {
 510                 fd = tx->fd;
 511                 if (fd < 0) {
 512                         RTE_LOG(INFO, PMD, "Add queue to TAP %s for qid %d\n",
 513                                 pmd->name, qid);
 514                         fd = tun_alloc(pmd->name);
 515                         if (fd < 0) {
 516                                 RTE_LOG(ERR, PMD, "tun_alloc(%s) failed\n",
 517                                         pmd->name);
 518                                 return -1;
 519                         }
 520                 }
 521         }
 522
 523         rx->fd = fd;
 524         tx->fd = fd;
 525
 526         return fd;
 527 }
 528
 529 static int
 530 rx_setup_queue(struct rte_eth_dev *dev,
 531                 struct pmd_internals *internals,
 532                 uint16_t qid)
 533 {
 534         dev->data->rx_queues[qid] = &internals->rxq[qid];
 535
 536         return tap_setup_queue(dev, internals, qid);
 537 }
 538
 539 static int
 540 tx_setup_queue(struct rte_eth_dev *dev,
 541                 struct pmd_internals *internals,
 542                 uint16_t qid)
 543 {
 544         dev->data->tx_queues[qid] = &internals->txq[qid];
 545
 546         return tap_setup_queue(dev, internals, qid);
 547 }
 548
 549 static int
 550 tap_rx_queue_setup(struct rte_eth_dev *dev,
 551                    uint16_t rx_queue_id,
 552                    uint16_t nb_rx_desc __rte_unused,
 553                    unsigned int socket_id __rte_unused,
 554                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 555                    struct rte_mempool *mp)
 556 {
 557         struct pmd_internals *internals = dev->data->dev_private;
 558         uint16_t buf_size;
 559         int fd;
 560
 561         if ((rx_queue_id >= internals->nb_queues) || !mp) {
 562                 RTE_LOG(ERR, PMD, "nb_queues %d mp %p\n",
 563                         internals->nb_queues, mp);
 564                 return -1;
 565         }
 566
 567         internals->rxq[rx_queue_id].mp = mp;
 568         internals->rxq[rx_queue_id].in_port = dev->data->port_id;
 569
 570         /* Now get the space available for data in the mbuf */
 571         buf_size = (uint16_t)(rte_pktmbuf_data_room_size(mp) -
 572                                 RTE_PKTMBUF_HEADROOM);
 573
 574         if (buf_size < ETH_FRAME_LEN) {
 575                 RTE_LOG(ERR, PMD,
 576                         "%s: %d bytes will not fit in mbuf (%d bytes)\n",
 577                         dev->data->name, ETH_FRAME_LEN, buf_size);
 578                 return -ENOMEM;
 579         }
 580
 581         fd = rx_setup_queue(dev, internals, rx_queue_id);
 582         if (fd == -1)
 583                 return -1;
 584
 585         RTE_LOG(INFO, PMD, "RX TAP device name %s, qid %d on fd %d\n",
 586                 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
 587
 588         return 0;
 589 }
 590
 591 static int
 592 tap_tx_queue_setup(struct rte_eth_dev *dev,
 593                    uint16_t tx_queue_id,
 594                    uint16_t nb_tx_desc __rte_unused,
 595                    unsigned int socket_id __rte_unused,
 596                    const struct rte_eth_txconf *tx_conf __rte_unused)
 597 {
 598         struct pmd_internals *internals = dev->data->dev_private;
 599         int ret;
 600
 601         if (tx_queue_id >= internals->nb_queues)
 602                 return -1;
 603
 604         ret = tx_setup_queue(dev, internals, tx_queue_id);
 605         if (ret == -1)
 606                 return -1;
 607
 608         RTE_LOG(INFO, PMD, "TX TAP device name %s, qid %d on fd %d\n",
 609                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd);
 610
 611         return 0;
 612 }
 613
 614 static const struct eth_dev_ops ops = {
 615         .dev_start              = tap_dev_start,
 616         .dev_stop               = tap_dev_stop,
 617         .dev_close              = tap_dev_close,
 618         .dev_configure          = tap_dev_configure,
 619         .dev_infos_get          = tap_dev_info,
 620         .rx_queue_setup         = tap_rx_queue_setup,
 621         .tx_queue_setup         = tap_tx_queue_setup,
 622         .rx_queue_release       = tap_rx_queue_release,
 623         .tx_queue_release       = tap_tx_queue_release,
 624         .link_update            = tap_link_update,
 625         .dev_set_link_up        = tap_link_set_up,
 626         .dev_set_link_down      = tap_link_set_down,
 627         .promiscuous_enable     = tap_promisc_enable,
 628         .promiscuous_disable    = tap_promisc_disable,
 629         .allmulticast_enable    = tap_allmulti_enable,
 630         .allmulticast_disable   = tap_allmulti_disable,
 631         .stats_get              = tap_stats_get,
 632         .stats_reset            = tap_stats_reset,
 633 };
 634
 635 static int
 636 pmd_mac_address(int fd, struct rte_eth_dev *dev, struct ether_addr *addr)
 637 {
 638         struct ifreq ifr;
 639
 640         if ((fd <= 0) || !dev || !addr)
 641                 return -1;
 642
 643         memset(&ifr, 0, sizeof(ifr));
 644
 645         if (ioctl(fd, SIOCGIFHWADDR, &ifr) == -1) {
 646                 RTE_LOG(ERR, PMD, "ioctl failed (SIOCGIFHWADDR) (%s)\n",
 647                         ifr.ifr_name);
 648                 return -1;
 649         }
 650
 651         rte_memcpy(addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
 652
 653         return 0;
 654 }
 655
 656 static int
 657 eth_dev_tap_create(const char *name, char *tap_name)
 658 {
 659         int numa_node = rte_socket_id();
 660         struct rte_eth_dev *dev = NULL;
 661         struct pmd_internals *pmd = NULL;
 662         struct rte_eth_dev_data *data = NULL;
 663         int i, fd = -1;
 664
 665         RTE_LOG(INFO, PMD,
 666                 "%s: Create TAP Ethernet device with %d queues on numa %u\n",
 667                  name, RTE_PMD_TAP_MAX_QUEUES, rte_socket_id());
 668
 669         data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
 670         if (!data) {
 671                 RTE_LOG(ERR, PMD, "Failed to allocate data\n");
 672                 goto error_exit;
 673         }
 674
 675         pmd = rte_zmalloc_socket(tap_name, sizeof(*pmd), 0, numa_node);
 676         if (!pmd) {
 677                 RTE_LOG(ERR, PMD, "Unable to allocate internal struct\n");
 678                 goto error_exit;
 679         }
 680
 681         /* Use the name and not the tap_name */
 682         dev = rte_eth_dev_allocate(tap_name);
 683         if (!dev) {
 684                 RTE_LOG(ERR, PMD, "Unable to allocate device struct\n");
 685                 goto error_exit;
 686         }
 687
 688         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
 689
 690         pmd->nb_queues = RTE_PMD_TAP_MAX_QUEUES;
 691
 692         /* Setup some default values */
 693         data->dev_private = pmd;
 694         data->port_id = dev->data->port_id;
 695         data->dev_flags = RTE_ETH_DEV_DETACHABLE;
 696         data->kdrv = RTE_KDRV_NONE;
 697         data->drv_name = pmd_tap_drv.driver.name;
 698         data->numa_node = numa_node;
 699
 700         data->dev_link = pmd_link;
 701         data->mac_addrs = &pmd->eth_addr;
 702         data->nb_rx_queues = pmd->nb_queues;
 703         data->nb_tx_queues = pmd->nb_queues;
 704
 705         dev->data = data;
 706         dev->dev_ops = &ops;
 707         dev->driver = NULL;
 708         dev->rx_pkt_burst = pmd_rx_burst;
 709         dev->tx_pkt_burst = pmd_tx_burst;
 710         snprintf(dev->data->name, sizeof(dev->data->name), "%s", name);
 711
 712         /* Create the first Tap device */
 713         fd = tun_alloc(tap_name);
 714         if (fd < 0) {
 715                 RTE_LOG(ERR, PMD, "tun_alloc() failed\n");
 716                 goto error_exit;
 717         }
 718
 719         /* Presetup the fds to -1 as being not working */
 720         for (i = 1; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 721                 pmd->rxq[i].fd = -1;
 722                 pmd->txq[i].fd = -1;
 723         }
 724
 725         /* Take the TUN/TAP fd and place in the first location */
 726         pmd->rxq[0].fd = fd;
 727         pmd->txq[0].fd = fd;
 728
 729         if (pmd_mac_address(fd, dev, &pmd->eth_addr) < 0) {
 730                 RTE_LOG(ERR, PMD, "Unable to get MAC address\n");
 731                 goto error_exit;
 732         }
 733
 734         return 0;
 735
 736 error_exit:
 737         RTE_PMD_DEBUG_TRACE("Unable to initialize %s\n", name);
 738
 739         rte_free(data);
 740         rte_free(pmd);
 741
 742         rte_eth_dev_release_port(dev);
 743
 744         return -EINVAL;
 745 }
 746
 747 static int
 748 set_interface_name(const char *key __rte_unused,
 749                    const char *value,
 750                    void *extra_args)
 751 {
 752         char *name = (char *)extra_args;
 753
 754         if (value)
 755                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
 756         else
 757                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
 758                          DEFAULT_TAP_NAME, (tap_unit - 1));
 759
 760         return 0;
 761 }
 762
 763 static int
 764 set_interface_speed(const char *key __rte_unused,
 765                     const char *value,
 766                     void *extra_args)
 767 {
 768         *(int *)extra_args = (value) ? atoi(value) : ETH_SPEED_NUM_10G;
 769
 770         return 0;
 771 }
 772
 773 /* Open a TAP interface device.
 774  */
 775 static int
 776 rte_pmd_tap_probe(const char *name, const char *params)
 777 {
 778         int ret;
 779         struct rte_kvargs *kvlist = NULL;
 780         int speed;
 781         char tap_name[RTE_ETH_NAME_MAX_LEN];
 782
 783         speed = ETH_SPEED_NUM_10G;
 784         snprintf(tap_name, sizeof(tap_name), "%s%d",
 785                  DEFAULT_TAP_NAME, tap_unit++);
 786
 787         if (params && (params[0] != '\0')) {
 788                 RTE_LOG(INFO, PMD, "paramaters (%s)\n", params);
 789
 790                 kvlist = rte_kvargs_parse(params, valid_arguments);
 791                 if (kvlist) {
 792                         if (rte_kvargs_count(kvlist, ETH_TAP_SPEED_ARG) == 1) {
 793                                 ret = rte_kvargs_process(kvlist,
 794                                                          ETH_TAP_SPEED_ARG,
 795                                                          &set_interface_speed,
 796                                                          &speed);
 797                                 if (ret == -1)
 798                                         goto leave;
 799                         }
 800
 801                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
 802                                 ret = rte_kvargs_process(kvlist,
 803                                                          ETH_TAP_IFACE_ARG,
 804                                                          &set_interface_name,
 805                                                          tap_name);
 806                                 if (ret == -1)
 807                                         goto leave;
 808                         }
 809                 }
 810         }
 811         pmd_link.link_speed = speed;
 812
 813         RTE_LOG(INFO, PMD, "Initializing pmd_tap for %s as %s\n",
 814                 name, tap_name);
 815
 816         ret = eth_dev_tap_create(name, tap_name);
 817
 818 leave:
 819         if (ret == -1) {
 820                 RTE_LOG(INFO, PMD, "Failed to create pmd for %s as %s\n",
 821                         name, tap_name);
 822                 tap_unit--;             /* Restore the unit number */
 823         }
 824         rte_kvargs_free(kvlist);
 825
 826         return ret;
 827 }
 828
 829 /* detach a TAP device.
 830  */
 831 static int
 832 rte_pmd_tap_remove(const char *name)
 833 {
 834         struct rte_eth_dev *eth_dev = NULL;
 835         struct pmd_internals *internals;
 836         int i;
 837
 838         RTE_LOG(INFO, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
 839                 rte_socket_id());
 840
 841         /* find the ethdev entry */
 842         eth_dev = rte_eth_dev_allocated(name);
 843         if (!eth_dev)
 844                 return 0;
 845
 846         internals = eth_dev->data->dev_private;
 847         for (i = 0; i < internals->nb_queues; i++)
 848                 if (internals->rxq[i].fd != -1)
 849                         close(internals->rxq[i].fd);
 850
 851         rte_free(eth_dev->data->dev_private);
 852         rte_free(eth_dev->data);
 853
 854         rte_eth_dev_release_port(eth_dev);
 855
 856         return 0;
 857 }
 858
 859 static struct rte_vdev_driver pmd_tap_drv = {
 860         .probe = rte_pmd_tap_probe,
 861         .remove = rte_pmd_tap_remove,
 862 };
 863 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
 864 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
 865 RTE_PMD_REGISTER_PARAM_STRING(net_tap, "iface=<string>,speed=N");