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
  42 #include <sys/types.h>
  43 #include <sys/stat.h>
  44 #include <sys/socket.h>
  45 #include <sys/ioctl.h>
  46 #include <sys/mman.h>
  47 #include <errno.h>
  48 #include <signal.h>
  49 #include <stdint.h>
  50 #include <unistd.h>
  51 #include <arpa/inet.h>
  52 #include <linux/if.h>
  53 #include <linux/if_tun.h>
  54 #include <linux/if_ether.h>
  55 #include <fcntl.h>
  56
  57 /* Linux based path to the TUN device */
  58 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  59 #define DEFAULT_TAP_NAME        "dtap"
  60
  61 #define ETH_TAP_IFACE_ARG       "iface"
  62 #define ETH_TAP_SPEED_ARG       "speed"
  63
  64 #ifdef IFF_MULTI_QUEUE
  65 #define RTE_PMD_TAP_MAX_QUEUES  16
  66 #else
  67 #define RTE_PMD_TAP_MAX_QUEUES  1
  68 #endif
  69
  70 static struct rte_vdev_driver pmd_tap_drv;
  71
  72 static const char *valid_arguments[] = {
  73         ETH_TAP_IFACE_ARG,
  74         ETH_TAP_SPEED_ARG,
  75         NULL
  76 };
  77
  78 static int tap_unit;
  79
  80 static volatile uint32_t tap_trigger;   /* Rx trigger */
  81
  82 static struct rte_eth_link pmd_link = {
  83         .link_speed = ETH_SPEED_NUM_10G,
  84         .link_duplex = ETH_LINK_FULL_DUPLEX,
  85         .link_status = ETH_LINK_DOWN,
  86         .link_autoneg = ETH_LINK_SPEED_AUTONEG
  87 };
  88
  89 struct pkt_stats {
  90         uint64_t opackets;              /* Number of output packets */
  91         uint64_t ipackets;              /* Number of input packets */
  92         uint64_t obytes;                /* Number of bytes on output */
  93         uint64_t ibytes;                /* Number of bytes on input */
  94         uint64_t errs;                  /* Number of error packets */
  95 };
  96
  97 struct rx_queue {
  98         struct rte_mempool *mp;         /* Mempool for RX packets */
  99         uint32_t trigger_seen;          /* Last seen Rx trigger value */
 100         uint16_t in_port;               /* Port ID */
 101         int fd;
 102
 103         struct pkt_stats stats;         /* Stats for this RX queue */
 104 };
 105
 106 struct tx_queue {
 107         int fd;
 108         struct pkt_stats stats;         /* Stats for this TX queue */
 109 };
 110
 111 struct pmd_internals {
 112         char name[RTE_ETH_NAME_MAX_LEN];        /* Internal Tap device name */
 113         uint16_t nb_queues;             /* Number of queues supported */
 114         struct ether_addr eth_addr;     /* Mac address of the device port */
 115
 116         int if_index;                   /* IF_INDEX for the port */
 117         int ioctl_sock;                 /* socket for ioctl calls */
 118
 119         struct rx_queue rxq[RTE_PMD_TAP_MAX_QUEUES];    /* List of RX queues */
 120         struct tx_queue txq[RTE_PMD_TAP_MAX_QUEUES];    /* List of TX queues */
 121 };
 122
 123 static void
 124 tap_trigger_cb(int sig __rte_unused)
 125 {
 126         /* Valid trigger values are nonzero */
 127         tap_trigger = (tap_trigger + 1) | 0x80000000;
 128 }
 129
 130 /* Tun/Tap allocation routine
 131  *
 132  * name is the number of the interface to use, unless NULL to take the host
 133  * supplied name.
 134  */
 135 static int
 136 tun_alloc(struct pmd_internals *pmd, uint16_t qid)
 137 {
 138         struct ifreq ifr;
 139 #ifdef IFF_MULTI_QUEUE
 140         unsigned int features;
 141 #endif
 142         int fd;
 143
 144         memset(&ifr, 0, sizeof(struct ifreq));
 145
 146         ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
 147         snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
 148
 149         RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
 150
 151         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 152         if (fd < 0) {
 153                 RTE_LOG(ERR, PMD, "Unable to create TAP interface");
 154                 goto error;
 155         }
 156
 157 #ifdef IFF_MULTI_QUEUE
 158         /* Grab the TUN features to verify we can work multi-queue */
 159         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 160                 RTE_LOG(ERR, PMD, "TAP unable to get TUN/TAP features\n");
 161                 goto error;
 162         }
 163         RTE_LOG(DEBUG, PMD, "  TAP Features %08x\n", features);
 164
 165         if (features & IFF_MULTI_QUEUE) {
 166                 RTE_LOG(DEBUG, PMD, "  Multi-queue support for %d queues\n",
 167                         RTE_PMD_TAP_MAX_QUEUES);
 168                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 169         } else
 170 #endif
 171         {
 172                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 173                 RTE_LOG(DEBUG, PMD, "  Single queue only support\n");
 174         }
 175
 176         /* Set the TUN/TAP configuration and set the name if needed */
 177         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 178                 RTE_LOG(WARNING, PMD,
 179                         "Unable to set TUNSETIFF for %s\n",
 180                         ifr.ifr_name);
 181                 perror("TUNSETIFF");
 182                 goto error;
 183         }
 184
 185         /* Always set the file descriptor to non-blocking */
 186         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 187                 RTE_LOG(WARNING, PMD,
 188                         "Unable to set %s to nonblocking\n",
 189                         ifr.ifr_name);
 190                 perror("F_SETFL, NONBLOCK");
 191                 goto error;
 192         }
 193
 194         /* Set up trigger to optimize empty Rx bursts */
 195         errno = 0;
 196         do {
 197                 struct sigaction sa;
 198                 int flags = fcntl(fd, F_GETFL);
 199
 200                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 201                         break;
 202                 if (sa.sa_handler != tap_trigger_cb) {
 203                         /*
 204                          * Make sure SIGIO is not already taken. This is done
 205                          * as late as possible to leave the application a
 206                          * chance to set up its own signal handler first.
 207                          */
 208                         if (sa.sa_handler != SIG_IGN &&
 209                             sa.sa_handler != SIG_DFL) {
 210                                 errno = EBUSY;
 211                                 break;
 212                         }
 213                         sa = (struct sigaction){
 214                                 .sa_flags = SA_RESTART,
 215                                 .sa_handler = tap_trigger_cb,
 216                         };
 217                         if (sigaction(SIGIO, &sa, NULL) == -1)
 218                                 break;
 219                 }
 220                 /* Enable SIGIO on file descriptor */
 221                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 222                 fcntl(fd, F_SETOWN, getpid());
 223         } while (0);
 224         if (errno) {
 225                 /* Disable trigger globally in case of error */
 226                 tap_trigger = 0;
 227                 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
 228                         strerror(errno));
 229         }
 230
 231         if (qid == 0) {
 232                 if (ioctl(fd, SIOCGIFHWADDR, &ifr) == -1) {
 233                         RTE_LOG(ERR, PMD, "ioctl failed (SIOCGIFHWADDR) (%s)\n",
 234                                 ifr.ifr_name);
 235                         goto error;
 236                 }
 237
 238                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data, ETH_ALEN);
 239         }
 240
 241         return fd;
 242
 243 error:
 244         if (fd > 0)
 245                 close(fd);
 246         return -1;
 247 }
 248
 249 /* Callback to handle the rx burst of packets to the correct interface and
 250  * file descriptor(s) in a multi-queue setup.
 251  */
 252 static uint16_t
 253 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 254 {
 255         int len;
 256         struct rte_mbuf *mbuf;
 257         struct rx_queue *rxq = queue;
 258         uint16_t num_rx;
 259         unsigned long num_rx_bytes = 0;
 260         uint32_t trigger = tap_trigger;
 261
 262         if (trigger == rxq->trigger_seen)
 263                 return 0;
 264         if (trigger)
 265                 rxq->trigger_seen = trigger;
 266         rte_compiler_barrier();
 267         for (num_rx = 0; num_rx < nb_pkts; ) {
 268                 /* allocate the next mbuf */
 269                 mbuf = rte_pktmbuf_alloc(rxq->mp);
 270                 if (unlikely(!mbuf)) {
 271                         RTE_LOG(WARNING, PMD, "TAP unable to allocate mbuf\n");
 272                         break;
 273                 }
 274
 275                 len = read(rxq->fd, rte_pktmbuf_mtod(mbuf, char *),
 276                            rte_pktmbuf_tailroom(mbuf));
 277                 if (len <= 0) {
 278                         rte_pktmbuf_free(mbuf);
 279                         break;
 280                 }
 281
 282                 mbuf->data_len = len;
 283                 mbuf->pkt_len = len;
 284                 mbuf->port = rxq->in_port;
 285
 286                 /* account for the receive frame */
 287                 bufs[num_rx++] = mbuf;
 288                 num_rx_bytes += mbuf->pkt_len;
 289         }
 290         rxq->stats.ipackets += num_rx;
 291         rxq->stats.ibytes += num_rx_bytes;
 292
 293         return num_rx;
 294 }
 295
 296 /* Callback to handle sending packets from the tap interface
 297  */
 298 static uint16_t
 299 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 300 {
 301         struct rte_mbuf *mbuf;
 302         struct tx_queue *txq = queue;
 303         uint16_t num_tx = 0;
 304         unsigned long num_tx_bytes = 0;
 305         int i, n;
 306
 307         if (unlikely(nb_pkts == 0))
 308                 return 0;
 309
 310         for (i = 0; i < nb_pkts; i++) {
 311                 /* copy the tx frame data */
 312                 mbuf = bufs[num_tx];
 313                 n = write(txq->fd,
 314                           rte_pktmbuf_mtod(mbuf, void *),
 315                           rte_pktmbuf_pkt_len(mbuf));
 316                 if (n <= 0)
 317                         break;
 318
 319                 num_tx++;
 320                 num_tx_bytes += mbuf->pkt_len;
 321                 rte_pktmbuf_free(mbuf);
 322         }
 323
 324         txq->stats.opackets += num_tx;
 325         txq->stats.errs += nb_pkts - num_tx;
 326         txq->stats.obytes += num_tx_bytes;
 327
 328         return num_tx;
 329 }
 330
 331 static int
 332 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 333           struct ifreq *ifr, int set)
 334 {
 335         short req_flags = ifr->ifr_flags;
 336
 337         snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 338         switch (request) {
 339         case SIOCSIFFLAGS:
 340                 /* fetch current flags to leave other flags untouched */
 341                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 342                         goto error;
 343                 if (set)
 344                         ifr->ifr_flags |= req_flags;
 345                 else
 346                         ifr->ifr_flags &= ~req_flags;
 347                 break;
 348         default:
 349                 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
 350                         pmd->name);
 351                 return -EINVAL;
 352         }
 353         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 354                 goto error;
 355         return 0;
 356
 357 error:
 358         RTE_LOG(ERR, PMD, "%s: ioctl(%lu) failed with error: %s\n",
 359                 ifr->ifr_name, request, strerror(errno));
 360         return -errno;
 361 }
 362
 363 static int
 364 tap_link_set_down(struct rte_eth_dev *dev)
 365 {
 366         struct pmd_internals *pmd = dev->data->dev_private;
 367         struct ifreq ifr = { .ifr_flags = IFF_UP };
 368
 369         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 370         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0);
 371 }
 372
 373 static int
 374 tap_link_set_up(struct rte_eth_dev *dev)
 375 {
 376         struct pmd_internals *pmd = dev->data->dev_private;
 377         struct ifreq ifr = { .ifr_flags = IFF_UP };
 378
 379         dev->data->dev_link.link_status = ETH_LINK_UP;
 380         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1);
 381 }
 382
 383 static int
 384 tap_dev_start(struct rte_eth_dev *dev)
 385 {
 386         return tap_link_set_up(dev);
 387 }
 388
 389 /* This function gets called when the current port gets stopped.
 390  */
 391 static void
 392 tap_dev_stop(struct rte_eth_dev *dev)
 393 {
 394         tap_link_set_down(dev);
 395 }
 396
 397 static int
 398 tap_dev_configure(struct rte_eth_dev *dev __rte_unused)
 399 {
 400         return 0;
 401 }
 402
 403 static void
 404 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 405 {
 406         struct pmd_internals *internals = dev->data->dev_private;
 407
 408         dev_info->if_index = internals->if_index;
 409         dev_info->max_mac_addrs = 1;
 410         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 411         dev_info->max_rx_queues = internals->nb_queues;
 412         dev_info->max_tx_queues = internals->nb_queues;
 413         dev_info->min_rx_bufsize = 0;
 414         dev_info->pci_dev = NULL;
 415 }
 416
 417 static void
 418 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 419 {
 420         unsigned int i, imax;
 421         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 422         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 423         const struct pmd_internals *pmd = dev->data->dev_private;
 424
 425         imax = (pmd->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 426                 pmd->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 427
 428         for (i = 0; i < imax; i++) {
 429                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 430                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 431                 rx_total += tap_stats->q_ipackets[i];
 432                 rx_bytes_total += tap_stats->q_ibytes[i];
 433
 434                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 435                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 436                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 437                 tx_total += tap_stats->q_opackets[i];
 438                 tx_err_total += tap_stats->q_errors[i];
 439                 tx_bytes_total += tap_stats->q_obytes[i];
 440         }
 441
 442         tap_stats->ipackets = rx_total;
 443         tap_stats->ibytes = rx_bytes_total;
 444         tap_stats->opackets = tx_total;
 445         tap_stats->oerrors = tx_err_total;
 446         tap_stats->obytes = tx_bytes_total;
 447 }
 448
 449 static void
 450 tap_stats_reset(struct rte_eth_dev *dev)
 451 {
 452         int i;
 453         struct pmd_internals *pmd = dev->data->dev_private;
 454
 455         for (i = 0; i < pmd->nb_queues; i++) {
 456                 pmd->rxq[i].stats.ipackets = 0;
 457                 pmd->rxq[i].stats.ibytes = 0;
 458
 459                 pmd->txq[i].stats.opackets = 0;
 460                 pmd->txq[i].stats.errs = 0;
 461                 pmd->txq[i].stats.obytes = 0;
 462         }
 463 }
 464
 465 static void
 466 tap_dev_close(struct rte_eth_dev *dev __rte_unused)
 467 {
 468         int i;
 469         struct pmd_internals *internals = dev->data->dev_private;
 470
 471         tap_link_set_down(dev);
 472
 473         for (i = 0; i < internals->nb_queues; i++) {
 474                 if (internals->rxq[i].fd != -1)
 475                         close(internals->rxq[i].fd);
 476                 internals->rxq[i].fd = -1;
 477                 internals->txq[i].fd = -1;
 478         }
 479 }
 480
 481 static void
 482 tap_rx_queue_release(void *queue)
 483 {
 484         struct rx_queue *rxq = queue;
 485
 486         if (rxq && (rxq->fd > 0)) {
 487                 close(rxq->fd);
 488                 rxq->fd = -1;
 489         }
 490 }
 491
 492 static void
 493 tap_tx_queue_release(void *queue)
 494 {
 495         struct tx_queue *txq = queue;
 496
 497         if (txq && (txq->fd > 0)) {
 498                 close(txq->fd);
 499                 txq->fd = -1;
 500         }
 501 }
 502
 503 static int
 504 tap_link_update(struct rte_eth_dev *dev __rte_unused,
 505                 int wait_to_complete __rte_unused)
 506 {
 507         return 0;
 508 }
 509
 510 static void
 511 tap_promisc_enable(struct rte_eth_dev *dev)
 512 {
 513         struct pmd_internals *pmd = dev->data->dev_private;
 514         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 515
 516         dev->data->promiscuous = 1;
 517         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1);
 518 }
 519
 520 static void
 521 tap_promisc_disable(struct rte_eth_dev *dev)
 522 {
 523         struct pmd_internals *pmd = dev->data->dev_private;
 524         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 525
 526         dev->data->promiscuous = 0;
 527         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0);
 528 }
 529
 530 static void
 531 tap_allmulti_enable(struct rte_eth_dev *dev)
 532 {
 533         struct pmd_internals *pmd = dev->data->dev_private;
 534         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 535
 536         dev->data->all_multicast = 1;
 537         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1);
 538 }
 539
 540 static void
 541 tap_allmulti_disable(struct rte_eth_dev *dev)
 542 {
 543         struct pmd_internals *pmd = dev->data->dev_private;
 544         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 545
 546         dev->data->all_multicast = 0;
 547         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0);
 548 }
 549
 550 static int
 551 tap_setup_queue(struct rte_eth_dev *dev,
 552                 struct pmd_internals *internals,
 553                 uint16_t qid)
 554 {
 555         struct pmd_internals *pmd = dev->data->dev_private;
 556         struct rx_queue *rx = &internals->rxq[qid];
 557         struct tx_queue *tx = &internals->txq[qid];
 558         int fd;
 559
 560         fd = rx->fd;
 561         if (fd < 0) {
 562                 fd = tx->fd;
 563                 if (fd < 0) {
 564                         RTE_LOG(INFO, PMD, "Add queue to TAP %s for qid %d\n",
 565                                 pmd->name, qid);
 566                         fd = tun_alloc(pmd, qid);
 567                         if (fd < 0) {
 568                                 RTE_LOG(ERR, PMD, "tun_alloc(%s, %d) failed\n",
 569                                         pmd->name, qid);
 570                                 return -1;
 571                         }
 572                 }
 573         }
 574
 575         rx->fd = fd;
 576         tx->fd = fd;
 577
 578         return fd;
 579 }
 580
 581 static int
 582 rx_setup_queue(struct rte_eth_dev *dev,
 583                 struct pmd_internals *internals,
 584                 uint16_t qid)
 585 {
 586         dev->data->rx_queues[qid] = &internals->rxq[qid];
 587
 588         return tap_setup_queue(dev, internals, qid);
 589 }
 590
 591 static int
 592 tx_setup_queue(struct rte_eth_dev *dev,
 593                 struct pmd_internals *internals,
 594                 uint16_t qid)
 595 {
 596         dev->data->tx_queues[qid] = &internals->txq[qid];
 597
 598         return tap_setup_queue(dev, internals, qid);
 599 }
 600
 601 static int
 602 tap_rx_queue_setup(struct rte_eth_dev *dev,
 603                    uint16_t rx_queue_id,
 604                    uint16_t nb_rx_desc __rte_unused,
 605                    unsigned int socket_id __rte_unused,
 606                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 607                    struct rte_mempool *mp)
 608 {
 609         struct pmd_internals *internals = dev->data->dev_private;
 610         uint16_t buf_size;
 611         int fd;
 612
 613         if ((rx_queue_id >= internals->nb_queues) || !mp) {
 614                 RTE_LOG(WARNING, PMD,
 615                         "nb_queues %d too small or mempool NULL\n",
 616                         internals->nb_queues);
 617                 return -1;
 618         }
 619
 620         internals->rxq[rx_queue_id].mp = mp;
 621         internals->rxq[rx_queue_id].trigger_seen = 1; /* force initial burst */
 622         internals->rxq[rx_queue_id].in_port = dev->data->port_id;
 623
 624         /* Now get the space available for data in the mbuf */
 625         buf_size = (uint16_t)(rte_pktmbuf_data_room_size(mp) -
 626                                 RTE_PKTMBUF_HEADROOM);
 627
 628         if (buf_size < ETH_FRAME_LEN) {
 629                 RTE_LOG(WARNING, PMD,
 630                         "%s: %d bytes will not fit in mbuf (%d bytes)\n",
 631                         dev->data->name, ETH_FRAME_LEN, buf_size);
 632                 return -ENOMEM;
 633         }
 634
 635         fd = rx_setup_queue(dev, internals, rx_queue_id);
 636         if (fd == -1)
 637                 return -1;
 638
 639         RTE_LOG(DEBUG, PMD, "  RX TAP device name %s, qid %d on fd %d\n",
 640                 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
 641
 642         return 0;
 643 }
 644
 645 static int
 646 tap_tx_queue_setup(struct rte_eth_dev *dev,
 647                    uint16_t tx_queue_id,
 648                    uint16_t nb_tx_desc __rte_unused,
 649                    unsigned int socket_id __rte_unused,
 650                    const struct rte_eth_txconf *tx_conf __rte_unused)
 651 {
 652         struct pmd_internals *internals = dev->data->dev_private;
 653         int ret;
 654
 655         if (tx_queue_id >= internals->nb_queues)
 656                 return -1;
 657
 658         ret = tx_setup_queue(dev, internals, tx_queue_id);
 659         if (ret == -1)
 660                 return -1;
 661
 662         RTE_LOG(DEBUG, PMD, "  TX TAP device name %s, qid %d on fd %d\n",
 663                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd);
 664
 665         return 0;
 666 }
 667
 668 static const struct eth_dev_ops ops = {
 669         .dev_start              = tap_dev_start,
 670         .dev_stop               = tap_dev_stop,
 671         .dev_close              = tap_dev_close,
 672         .dev_configure          = tap_dev_configure,
 673         .dev_infos_get          = tap_dev_info,
 674         .rx_queue_setup         = tap_rx_queue_setup,
 675         .tx_queue_setup         = tap_tx_queue_setup,
 676         .rx_queue_release       = tap_rx_queue_release,
 677         .tx_queue_release       = tap_tx_queue_release,
 678         .link_update            = tap_link_update,
 679         .dev_set_link_up        = tap_link_set_up,
 680         .dev_set_link_down      = tap_link_set_down,
 681         .promiscuous_enable     = tap_promisc_enable,
 682         .promiscuous_disable    = tap_promisc_disable,
 683         .allmulticast_enable    = tap_allmulti_enable,
 684         .allmulticast_disable   = tap_allmulti_disable,
 685         .stats_get              = tap_stats_get,
 686         .stats_reset            = tap_stats_reset,
 687 };
 688
 689 static int
 690 eth_dev_tap_create(const char *name, char *tap_name)
 691 {
 692         int numa_node = rte_socket_id();
 693         struct rte_eth_dev *dev = NULL;
 694         struct pmd_internals *pmd = NULL;
 695         struct rte_eth_dev_data *data = NULL;
 696         int i;
 697
 698         RTE_LOG(DEBUG, PMD, "  TAP device on numa %u\n", rte_socket_id());
 699
 700         data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
 701         if (!data) {
 702                 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
 703                 goto error_exit;
 704         }
 705
 706         pmd = rte_zmalloc_socket(tap_name, sizeof(*pmd), 0, numa_node);
 707         if (!pmd) {
 708                 RTE_LOG(ERR, PMD, "TAP Unable to allocate internal struct\n");
 709                 goto error_exit;
 710         }
 711
 712         /* name in allocation and data->name must be consistent */
 713         snprintf(data->name, sizeof(data->name), "%s", name);
 714         dev = rte_eth_dev_allocate(name);
 715         if (!dev) {
 716                 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
 717                 goto error_exit;
 718         }
 719
 720         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
 721
 722         pmd->nb_queues = RTE_PMD_TAP_MAX_QUEUES;
 723
 724         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
 725         if (pmd->ioctl_sock == -1) {
 726                 RTE_LOG(ERR, PMD,
 727                         "TAP Unable to get a socket for management: %s\n",
 728                         strerror(errno));
 729                 goto error_exit;
 730         }
 731
 732         /* Setup some default values */
 733         data->dev_private = pmd;
 734         data->port_id = dev->data->port_id;
 735         data->dev_flags = RTE_ETH_DEV_DETACHABLE;
 736         data->kdrv = RTE_KDRV_NONE;
 737         data->drv_name = pmd_tap_drv.driver.name;
 738         data->numa_node = numa_node;
 739
 740         data->dev_link = pmd_link;
 741         data->mac_addrs = &pmd->eth_addr;
 742         data->nb_rx_queues = pmd->nb_queues;
 743         data->nb_tx_queues = pmd->nb_queues;
 744
 745         dev->data = data;
 746         dev->dev_ops = &ops;
 747         dev->driver = NULL;
 748         dev->rx_pkt_burst = pmd_rx_burst;
 749         dev->tx_pkt_burst = pmd_tx_burst;
 750
 751         /* Presetup the fds to -1 as being not valid */
 752         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 753                 pmd->rxq[i].fd = -1;
 754                 pmd->txq[i].fd = -1;
 755         }
 756
 757         return 0;
 758
 759 error_exit:
 760         RTE_LOG(DEBUG, PMD, "TAP Unable to initialize %s\n", name);
 761
 762         rte_free(data);
 763         rte_free(pmd);
 764
 765         rte_eth_dev_release_port(dev);
 766
 767         return -EINVAL;
 768 }
 769
 770 static int
 771 set_interface_name(const char *key __rte_unused,
 772                    const char *value,
 773                    void *extra_args)
 774 {
 775         char *name = (char *)extra_args;
 776
 777         if (value)
 778                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
 779         else
 780                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
 781                          DEFAULT_TAP_NAME, (tap_unit - 1));
 782
 783         return 0;
 784 }
 785
 786 static int
 787 set_interface_speed(const char *key __rte_unused,
 788                     const char *value,
 789                     void *extra_args)
 790 {
 791         *(int *)extra_args = (value) ? atoi(value) : ETH_SPEED_NUM_10G;
 792
 793         return 0;
 794 }
 795
 796 /* Open a TAP interface device.
 797  */
 798 static int
 799 rte_pmd_tap_probe(const char *name, const char *params)
 800 {
 801         int ret;
 802         struct rte_kvargs *kvlist = NULL;
 803         int speed;
 804         char tap_name[RTE_ETH_NAME_MAX_LEN];
 805
 806         speed = ETH_SPEED_NUM_10G;
 807         snprintf(tap_name, sizeof(tap_name), "%s%d",
 808                  DEFAULT_TAP_NAME, tap_unit++);
 809
 810         if (params && (params[0] != '\0')) {
 811                 RTE_LOG(DEBUG, PMD, "paramaters (%s)\n", params);
 812
 813                 kvlist = rte_kvargs_parse(params, valid_arguments);
 814                 if (kvlist) {
 815                         if (rte_kvargs_count(kvlist, ETH_TAP_SPEED_ARG) == 1) {
 816                                 ret = rte_kvargs_process(kvlist,
 817                                                          ETH_TAP_SPEED_ARG,
 818                                                          &set_interface_speed,
 819                                                          &speed);
 820                                 if (ret == -1)
 821                                         goto leave;
 822                         }
 823
 824                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
 825                                 ret = rte_kvargs_process(kvlist,
 826                                                          ETH_TAP_IFACE_ARG,
 827                                                          &set_interface_name,
 828                                                          tap_name);
 829                                 if (ret == -1)
 830                                         goto leave;
 831                         }
 832                 }
 833         }
 834         pmd_link.link_speed = speed;
 835
 836         RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
 837                 name, tap_name);
 838
 839         ret = eth_dev_tap_create(name, tap_name);
 840
 841 leave:
 842         if (ret == -1) {
 843                 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
 844                         name, tap_name);
 845                 tap_unit--;             /* Restore the unit number */
 846         }
 847         rte_kvargs_free(kvlist);
 848
 849         return ret;
 850 }
 851
 852 /* detach a TAP device.
 853  */
 854 static int
 855 rte_pmd_tap_remove(const char *name)
 856 {
 857         struct rte_eth_dev *eth_dev = NULL;
 858         struct pmd_internals *internals;
 859         int i;
 860
 861         RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
 862                 rte_socket_id());
 863
 864         /* find the ethdev entry */
 865         eth_dev = rte_eth_dev_allocated(name);
 866         if (!eth_dev)
 867                 return 0;
 868
 869         internals = eth_dev->data->dev_private;
 870         for (i = 0; i < internals->nb_queues; i++)
 871                 if (internals->rxq[i].fd != -1)
 872                         close(internals->rxq[i].fd);
 873
 874         close(internals->ioctl_sock);
 875         rte_free(eth_dev->data->dev_private);
 876         rte_free(eth_dev->data);
 877
 878         rte_eth_dev_release_port(eth_dev);
 879
 880         return 0;
 881 }
 882
 883 static struct rte_vdev_driver pmd_tap_drv = {
 884         .probe = rte_pmd_tap_probe,
 885         .remove = rte_pmd_tap_remove,
 886 };
 887 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
 888 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
 889 RTE_PMD_REGISTER_PARAM_STRING(net_tap, "iface=<string>,speed=N");