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