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 static int
 131 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 132           struct ifreq *ifr, int set);
 133
 134 /* Tun/Tap allocation routine
 135  *
 136  * name is the number of the interface to use, unless NULL to take the host
 137  * supplied name.
 138  */
 139 static int
 140 tun_alloc(struct pmd_internals *pmd, uint16_t qid)
 141 {
 142         struct ifreq ifr;
 143 #ifdef IFF_MULTI_QUEUE
 144         unsigned int features;
 145 #endif
 146         int fd;
 147
 148         memset(&ifr, 0, sizeof(struct ifreq));
 149
 150         ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
 151         snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
 152
 153         RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
 154
 155         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 156         if (fd < 0) {
 157                 RTE_LOG(ERR, PMD, "Unable to create TAP interface");
 158                 goto error;
 159         }
 160
 161 #ifdef IFF_MULTI_QUEUE
 162         /* Grab the TUN features to verify we can work multi-queue */
 163         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 164                 RTE_LOG(ERR, PMD, "TAP unable to get TUN/TAP features\n");
 165                 goto error;
 166         }
 167         RTE_LOG(DEBUG, PMD, "  TAP Features %08x\n", features);
 168
 169         if (features & IFF_MULTI_QUEUE) {
 170                 RTE_LOG(DEBUG, PMD, "  Multi-queue support for %d queues\n",
 171                         RTE_PMD_TAP_MAX_QUEUES);
 172                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 173         } else
 174 #endif
 175         {
 176                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 177                 RTE_LOG(DEBUG, PMD, "  Single queue only support\n");
 178         }
 179
 180         /* Set the TUN/TAP configuration and set the name if needed */
 181         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 182                 RTE_LOG(WARNING, PMD,
 183                         "Unable to set TUNSETIFF for %s\n",
 184                         ifr.ifr_name);
 185                 perror("TUNSETIFF");
 186                 goto error;
 187         }
 188
 189         /* Always set the file descriptor to non-blocking */
 190         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 191                 RTE_LOG(WARNING, PMD,
 192                         "Unable to set %s to nonblocking\n",
 193                         ifr.ifr_name);
 194                 perror("F_SETFL, NONBLOCK");
 195                 goto error;
 196         }
 197
 198         /* Set up trigger to optimize empty Rx bursts */
 199         errno = 0;
 200         do {
 201                 struct sigaction sa;
 202                 int flags = fcntl(fd, F_GETFL);
 203
 204                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 205                         break;
 206                 if (sa.sa_handler != tap_trigger_cb) {
 207                         /*
 208                          * Make sure SIGIO is not already taken. This is done
 209                          * as late as possible to leave the application a
 210                          * chance to set up its own signal handler first.
 211                          */
 212                         if (sa.sa_handler != SIG_IGN &&
 213                             sa.sa_handler != SIG_DFL) {
 214                                 errno = EBUSY;
 215                                 break;
 216                         }
 217                         sa = (struct sigaction){
 218                                 .sa_flags = SA_RESTART,
 219                                 .sa_handler = tap_trigger_cb,
 220                         };
 221                         if (sigaction(SIGIO, &sa, NULL) == -1)
 222                                 break;
 223                 }
 224                 /* Enable SIGIO on file descriptor */
 225                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 226                 fcntl(fd, F_SETOWN, getpid());
 227         } while (0);
 228         if (errno) {
 229                 /* Disable trigger globally in case of error */
 230                 tap_trigger = 0;
 231                 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
 232                         strerror(errno));
 233         }
 234
 235         if (qid == 0) {
 236                 struct ifreq ifr;
 237
 238                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0) < 0)
 239                         goto error;
 240                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
 241                            ETHER_ADDR_LEN);
 242         }
 243
 244         return fd;
 245
 246 error:
 247         if (fd > 0)
 248                 close(fd);
 249         return -1;
 250 }
 251
 252 /* Callback to handle the rx burst of packets to the correct interface and
 253  * file descriptor(s) in a multi-queue setup.
 254  */
 255 static uint16_t
 256 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 257 {
 258         int len;
 259         struct rte_mbuf *mbuf;
 260         struct rx_queue *rxq = queue;
 261         uint16_t num_rx;
 262         unsigned long num_rx_bytes = 0;
 263         uint32_t trigger = tap_trigger;
 264
 265         if (trigger == rxq->trigger_seen)
 266                 return 0;
 267         if (trigger)
 268                 rxq->trigger_seen = trigger;
 269         rte_compiler_barrier();
 270         for (num_rx = 0; num_rx < nb_pkts; ) {
 271                 /* allocate the next mbuf */
 272                 mbuf = rte_pktmbuf_alloc(rxq->mp);
 273                 if (unlikely(!mbuf)) {
 274                         RTE_LOG(WARNING, PMD, "TAP unable to allocate mbuf\n");
 275                         break;
 276                 }
 277
 278                 len = read(rxq->fd, rte_pktmbuf_mtod(mbuf, char *),
 279                            rte_pktmbuf_tailroom(mbuf));
 280                 if (len <= 0) {
 281                         rte_pktmbuf_free(mbuf);
 282                         break;
 283                 }
 284
 285                 mbuf->data_len = len;
 286                 mbuf->pkt_len = len;
 287                 mbuf->port = rxq->in_port;
 288
 289                 /* account for the receive frame */
 290                 bufs[num_rx++] = mbuf;
 291                 num_rx_bytes += mbuf->pkt_len;
 292         }
 293         rxq->stats.ipackets += num_rx;
 294         rxq->stats.ibytes += num_rx_bytes;
 295
 296         return num_rx;
 297 }
 298
 299 /* Callback to handle sending packets from the tap interface
 300  */
 301 static uint16_t
 302 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 303 {
 304         struct rte_mbuf *mbuf;
 305         struct tx_queue *txq = queue;
 306         uint16_t num_tx = 0;
 307         unsigned long num_tx_bytes = 0;
 308         int i, n;
 309
 310         if (unlikely(nb_pkts == 0))
 311                 return 0;
 312
 313         for (i = 0; i < nb_pkts; i++) {
 314                 /* copy the tx frame data */
 315                 mbuf = bufs[num_tx];
 316                 n = write(txq->fd,
 317                           rte_pktmbuf_mtod(mbuf, void *),
 318                           rte_pktmbuf_pkt_len(mbuf));
 319                 if (n <= 0)
 320                         break;
 321
 322                 num_tx++;
 323                 num_tx_bytes += mbuf->pkt_len;
 324                 rte_pktmbuf_free(mbuf);
 325         }
 326
 327         txq->stats.opackets += num_tx;
 328         txq->stats.errs += nb_pkts - num_tx;
 329         txq->stats.obytes += num_tx_bytes;
 330
 331         return num_tx;
 332 }
 333
 334 static int
 335 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 336           struct ifreq *ifr, int set)
 337 {
 338         short req_flags = ifr->ifr_flags;
 339
 340         snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 341         switch (request) {
 342         case SIOCSIFFLAGS:
 343                 /* fetch current flags to leave other flags untouched */
 344                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 345                         goto error;
 346                 if (set)
 347                         ifr->ifr_flags |= req_flags;
 348                 else
 349                         ifr->ifr_flags &= ~req_flags;
 350                 break;
 351         case SIOCGIFHWADDR:
 352         case SIOCSIFHWADDR:
 353                 break;
 354         default:
 355                 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
 356                         pmd->name);
 357                 return -EINVAL;
 358         }
 359         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 360                 goto error;
 361         return 0;
 362
 363 error:
 364         RTE_LOG(ERR, PMD, "%s: ioctl(%lu) failed with error: %s\n",
 365                 ifr->ifr_name, request, strerror(errno));
 366         return -errno;
 367 }
 368
 369 static int
 370 tap_link_set_down(struct rte_eth_dev *dev)
 371 {
 372         struct pmd_internals *pmd = dev->data->dev_private;
 373         struct ifreq ifr = { .ifr_flags = IFF_UP };
 374
 375         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 376         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0);
 377 }
 378
 379 static int
 380 tap_link_set_up(struct rte_eth_dev *dev)
 381 {
 382         struct pmd_internals *pmd = dev->data->dev_private;
 383         struct ifreq ifr = { .ifr_flags = IFF_UP };
 384
 385         dev->data->dev_link.link_status = ETH_LINK_UP;
 386         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1);
 387 }
 388
 389 static int
 390 tap_dev_start(struct rte_eth_dev *dev)
 391 {
 392         return tap_link_set_up(dev);
 393 }
 394
 395 /* This function gets called when the current port gets stopped.
 396  */
 397 static void
 398 tap_dev_stop(struct rte_eth_dev *dev)
 399 {
 400         tap_link_set_down(dev);
 401 }
 402
 403 static int
 404 tap_dev_configure(struct rte_eth_dev *dev __rte_unused)
 405 {
 406         return 0;
 407 }
 408
 409 static void
 410 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 411 {
 412         struct pmd_internals *internals = dev->data->dev_private;
 413
 414         dev_info->if_index = internals->if_index;
 415         dev_info->max_mac_addrs = 1;
 416         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 417         dev_info->max_rx_queues = internals->nb_queues;
 418         dev_info->max_tx_queues = internals->nb_queues;
 419         dev_info->min_rx_bufsize = 0;
 420         dev_info->pci_dev = NULL;
 421 }
 422
 423 static void
 424 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 425 {
 426         unsigned int i, imax;
 427         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 428         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 429         const struct pmd_internals *pmd = dev->data->dev_private;
 430
 431         imax = (pmd->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 432                 pmd->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 433
 434         for (i = 0; i < imax; i++) {
 435                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 436                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 437                 rx_total += tap_stats->q_ipackets[i];
 438                 rx_bytes_total += tap_stats->q_ibytes[i];
 439
 440                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 441                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 442                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 443                 tx_total += tap_stats->q_opackets[i];
 444                 tx_err_total += tap_stats->q_errors[i];
 445                 tx_bytes_total += tap_stats->q_obytes[i];
 446         }
 447
 448         tap_stats->ipackets = rx_total;
 449         tap_stats->ibytes = rx_bytes_total;
 450         tap_stats->opackets = tx_total;
 451         tap_stats->oerrors = tx_err_total;
 452         tap_stats->obytes = tx_bytes_total;
 453 }
 454
 455 static void
 456 tap_stats_reset(struct rte_eth_dev *dev)
 457 {
 458         int i;
 459         struct pmd_internals *pmd = dev->data->dev_private;
 460
 461         for (i = 0; i < pmd->nb_queues; i++) {
 462                 pmd->rxq[i].stats.ipackets = 0;
 463                 pmd->rxq[i].stats.ibytes = 0;
 464
 465                 pmd->txq[i].stats.opackets = 0;
 466                 pmd->txq[i].stats.errs = 0;
 467                 pmd->txq[i].stats.obytes = 0;
 468         }
 469 }
 470
 471 static void
 472 tap_dev_close(struct rte_eth_dev *dev __rte_unused)
 473 {
 474         int i;
 475         struct pmd_internals *internals = dev->data->dev_private;
 476
 477         tap_link_set_down(dev);
 478
 479         for (i = 0; i < internals->nb_queues; i++) {
 480                 if (internals->rxq[i].fd != -1)
 481                         close(internals->rxq[i].fd);
 482                 internals->rxq[i].fd = -1;
 483                 internals->txq[i].fd = -1;
 484         }
 485 }
 486
 487 static void
 488 tap_rx_queue_release(void *queue)
 489 {
 490         struct rx_queue *rxq = queue;
 491
 492         if (rxq && (rxq->fd > 0)) {
 493                 close(rxq->fd);
 494                 rxq->fd = -1;
 495         }
 496 }
 497
 498 static void
 499 tap_tx_queue_release(void *queue)
 500 {
 501         struct tx_queue *txq = queue;
 502
 503         if (txq && (txq->fd > 0)) {
 504                 close(txq->fd);
 505                 txq->fd = -1;
 506         }
 507 }
 508
 509 static int
 510 tap_link_update(struct rte_eth_dev *dev __rte_unused,
 511                 int wait_to_complete __rte_unused)
 512 {
 513         return 0;
 514 }
 515
 516 static void
 517 tap_promisc_enable(struct rte_eth_dev *dev)
 518 {
 519         struct pmd_internals *pmd = dev->data->dev_private;
 520         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 521
 522         dev->data->promiscuous = 1;
 523         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 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         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 531
 532         dev->data->promiscuous = 0;
 533         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0);
 534 }
 535
 536 static void
 537 tap_allmulti_enable(struct rte_eth_dev *dev)
 538 {
 539         struct pmd_internals *pmd = dev->data->dev_private;
 540         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 541
 542         dev->data->all_multicast = 1;
 543         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1);
 544 }
 545
 546 static void
 547 tap_allmulti_disable(struct rte_eth_dev *dev)
 548 {
 549         struct pmd_internals *pmd = dev->data->dev_private;
 550         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 551
 552         dev->data->all_multicast = 0;
 553         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0);
 554 }
 555
 556
 557 static void
 558 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 559 {
 560         struct pmd_internals *pmd = dev->data->dev_private;
 561         struct ifreq ifr;
 562
 563         if (is_zero_ether_addr(mac_addr)) {
 564                 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
 565                         dev->data->name);
 566                 return;
 567         }
 568
 569         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
 570         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
 571         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1) < 0)
 572                 return;
 573         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
 574 }
 575
 576 static int
 577 tap_setup_queue(struct rte_eth_dev *dev,
 578                 struct pmd_internals *internals,
 579                 uint16_t qid)
 580 {
 581         struct pmd_internals *pmd = dev->data->dev_private;
 582         struct rx_queue *rx = &internals->rxq[qid];
 583         struct tx_queue *tx = &internals->txq[qid];
 584         int fd;
 585
 586         fd = rx->fd;
 587         if (fd < 0) {
 588                 fd = tx->fd;
 589                 if (fd < 0) {
 590                         RTE_LOG(INFO, PMD, "Add queue to TAP %s for qid %d\n",
 591                                 pmd->name, qid);
 592                         fd = tun_alloc(pmd, qid);
 593                         if (fd < 0) {
 594                                 RTE_LOG(ERR, PMD, "tun_alloc(%s, %d) failed\n",
 595                                         pmd->name, qid);
 596                                 return -1;
 597                         }
 598                 }
 599         }
 600
 601         rx->fd = fd;
 602         tx->fd = fd;
 603
 604         return fd;
 605 }
 606
 607 static int
 608 rx_setup_queue(struct rte_eth_dev *dev,
 609                 struct pmd_internals *internals,
 610                 uint16_t qid)
 611 {
 612         dev->data->rx_queues[qid] = &internals->rxq[qid];
 613
 614         return tap_setup_queue(dev, internals, qid);
 615 }
 616
 617 static int
 618 tx_setup_queue(struct rte_eth_dev *dev,
 619                 struct pmd_internals *internals,
 620                 uint16_t qid)
 621 {
 622         dev->data->tx_queues[qid] = &internals->txq[qid];
 623
 624         return tap_setup_queue(dev, internals, qid);
 625 }
 626
 627 static int
 628 tap_rx_queue_setup(struct rte_eth_dev *dev,
 629                    uint16_t rx_queue_id,
 630                    uint16_t nb_rx_desc __rte_unused,
 631                    unsigned int socket_id __rte_unused,
 632                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 633                    struct rte_mempool *mp)
 634 {
 635         struct pmd_internals *internals = dev->data->dev_private;
 636         uint16_t buf_size;
 637         int fd;
 638
 639         if ((rx_queue_id >= internals->nb_queues) || !mp) {
 640                 RTE_LOG(WARNING, PMD,
 641                         "nb_queues %d too small or mempool NULL\n",
 642                         internals->nb_queues);
 643                 return -1;
 644         }
 645
 646         internals->rxq[rx_queue_id].mp = mp;
 647         internals->rxq[rx_queue_id].trigger_seen = 1; /* force initial burst */
 648         internals->rxq[rx_queue_id].in_port = dev->data->port_id;
 649
 650         /* Now get the space available for data in the mbuf */
 651         buf_size = (uint16_t)(rte_pktmbuf_data_room_size(mp) -
 652                                 RTE_PKTMBUF_HEADROOM);
 653
 654         if (buf_size < ETH_FRAME_LEN) {
 655                 RTE_LOG(WARNING, PMD,
 656                         "%s: %d bytes will not fit in mbuf (%d bytes)\n",
 657                         dev->data->name, ETH_FRAME_LEN, buf_size);
 658                 return -ENOMEM;
 659         }
 660
 661         fd = rx_setup_queue(dev, internals, rx_queue_id);
 662         if (fd == -1)
 663                 return -1;
 664
 665         RTE_LOG(DEBUG, PMD, "  RX TAP device name %s, qid %d on fd %d\n",
 666                 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
 667
 668         return 0;
 669 }
 670
 671 static int
 672 tap_tx_queue_setup(struct rte_eth_dev *dev,
 673                    uint16_t tx_queue_id,
 674                    uint16_t nb_tx_desc __rte_unused,
 675                    unsigned int socket_id __rte_unused,
 676                    const struct rte_eth_txconf *tx_conf __rte_unused)
 677 {
 678         struct pmd_internals *internals = dev->data->dev_private;
 679         int ret;
 680
 681         if (tx_queue_id >= internals->nb_queues)
 682                 return -1;
 683
 684         ret = tx_setup_queue(dev, internals, tx_queue_id);
 685         if (ret == -1)
 686                 return -1;
 687
 688         RTE_LOG(DEBUG, PMD, "  TX TAP device name %s, qid %d on fd %d\n",
 689                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd);
 690
 691         return 0;
 692 }
 693
 694 static const struct eth_dev_ops ops = {
 695         .dev_start              = tap_dev_start,
 696         .dev_stop               = tap_dev_stop,
 697         .dev_close              = tap_dev_close,
 698         .dev_configure          = tap_dev_configure,
 699         .dev_infos_get          = tap_dev_info,
 700         .rx_queue_setup         = tap_rx_queue_setup,
 701         .tx_queue_setup         = tap_tx_queue_setup,
 702         .rx_queue_release       = tap_rx_queue_release,
 703         .tx_queue_release       = tap_tx_queue_release,
 704         .link_update            = tap_link_update,
 705         .dev_set_link_up        = tap_link_set_up,
 706         .dev_set_link_down      = tap_link_set_down,
 707         .promiscuous_enable     = tap_promisc_enable,
 708         .promiscuous_disable    = tap_promisc_disable,
 709         .allmulticast_enable    = tap_allmulti_enable,
 710         .allmulticast_disable   = tap_allmulti_disable,
 711         .mac_addr_set           = tap_mac_set,
 712         .stats_get              = tap_stats_get,
 713         .stats_reset            = tap_stats_reset,
 714 };
 715
 716 static int
 717 eth_dev_tap_create(const char *name, char *tap_name)
 718 {
 719         int numa_node = rte_socket_id();
 720         struct rte_eth_dev *dev = NULL;
 721         struct pmd_internals *pmd = NULL;
 722         struct rte_eth_dev_data *data = NULL;
 723         int i;
 724
 725         RTE_LOG(DEBUG, PMD, "  TAP device on numa %u\n", rte_socket_id());
 726
 727         data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
 728         if (!data) {
 729                 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
 730                 goto error_exit;
 731         }
 732
 733         pmd = rte_zmalloc_socket(tap_name, sizeof(*pmd), 0, numa_node);
 734         if (!pmd) {
 735                 RTE_LOG(ERR, PMD, "TAP Unable to allocate internal struct\n");
 736                 goto error_exit;
 737         }
 738
 739         /* name in allocation and data->name must be consistent */
 740         snprintf(data->name, sizeof(data->name), "%s", name);
 741         dev = rte_eth_dev_allocate(name);
 742         if (!dev) {
 743                 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
 744                 goto error_exit;
 745         }
 746
 747         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
 748
 749         pmd->nb_queues = RTE_PMD_TAP_MAX_QUEUES;
 750
 751         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
 752         if (pmd->ioctl_sock == -1) {
 753                 RTE_LOG(ERR, PMD,
 754                         "TAP Unable to get a socket for management: %s\n",
 755                         strerror(errno));
 756                 goto error_exit;
 757         }
 758
 759         /* Setup some default values */
 760         data->dev_private = pmd;
 761         data->port_id = dev->data->port_id;
 762         data->dev_flags = RTE_ETH_DEV_DETACHABLE;
 763         data->kdrv = RTE_KDRV_NONE;
 764         data->drv_name = pmd_tap_drv.driver.name;
 765         data->numa_node = numa_node;
 766
 767         data->dev_link = pmd_link;
 768         data->mac_addrs = &pmd->eth_addr;
 769         data->nb_rx_queues = pmd->nb_queues;
 770         data->nb_tx_queues = pmd->nb_queues;
 771
 772         dev->data = data;
 773         dev->dev_ops = &ops;
 774         dev->driver = NULL;
 775         dev->rx_pkt_burst = pmd_rx_burst;
 776         dev->tx_pkt_burst = pmd_tx_burst;
 777
 778         /* Presetup the fds to -1 as being not valid */
 779         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 780                 pmd->rxq[i].fd = -1;
 781                 pmd->txq[i].fd = -1;
 782         }
 783
 784         return 0;
 785
 786 error_exit:
 787         RTE_LOG(DEBUG, PMD, "TAP Unable to initialize %s\n", name);
 788
 789         rte_free(data);
 790         rte_free(pmd);
 791
 792         rte_eth_dev_release_port(dev);
 793
 794         return -EINVAL;
 795 }
 796
 797 static int
 798 set_interface_name(const char *key __rte_unused,
 799                    const char *value,
 800                    void *extra_args)
 801 {
 802         char *name = (char *)extra_args;
 803
 804         if (value)
 805                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
 806         else
 807                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
 808                          DEFAULT_TAP_NAME, (tap_unit - 1));
 809
 810         return 0;
 811 }
 812
 813 static int
 814 set_interface_speed(const char *key __rte_unused,
 815                     const char *value,
 816                     void *extra_args)
 817 {
 818         *(int *)extra_args = (value) ? atoi(value) : ETH_SPEED_NUM_10G;
 819
 820         return 0;
 821 }
 822
 823 /* Open a TAP interface device.
 824  */
 825 static int
 826 rte_pmd_tap_probe(const char *name, const char *params)
 827 {
 828         int ret;
 829         struct rte_kvargs *kvlist = NULL;
 830         int speed;
 831         char tap_name[RTE_ETH_NAME_MAX_LEN];
 832
 833         speed = ETH_SPEED_NUM_10G;
 834         snprintf(tap_name, sizeof(tap_name), "%s%d",
 835                  DEFAULT_TAP_NAME, tap_unit++);
 836
 837         if (params && (params[0] != '\0')) {
 838                 RTE_LOG(DEBUG, PMD, "paramaters (%s)\n", params);
 839
 840                 kvlist = rte_kvargs_parse(params, valid_arguments);
 841                 if (kvlist) {
 842                         if (rte_kvargs_count(kvlist, ETH_TAP_SPEED_ARG) == 1) {
 843                                 ret = rte_kvargs_process(kvlist,
 844                                                          ETH_TAP_SPEED_ARG,
 845                                                          &set_interface_speed,
 846                                                          &speed);
 847                                 if (ret == -1)
 848                                         goto leave;
 849                         }
 850
 851                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
 852                                 ret = rte_kvargs_process(kvlist,
 853                                                          ETH_TAP_IFACE_ARG,
 854                                                          &set_interface_name,
 855                                                          tap_name);
 856                                 if (ret == -1)
 857                                         goto leave;
 858                         }
 859                 }
 860         }
 861         pmd_link.link_speed = speed;
 862
 863         RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
 864                 name, tap_name);
 865
 866         ret = eth_dev_tap_create(name, tap_name);
 867
 868 leave:
 869         if (ret == -1) {
 870                 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
 871                         name, tap_name);
 872                 tap_unit--;             /* Restore the unit number */
 873         }
 874         rte_kvargs_free(kvlist);
 875
 876         return ret;
 877 }
 878
 879 /* detach a TAP device.
 880  */
 881 static int
 882 rte_pmd_tap_remove(const char *name)
 883 {
 884         struct rte_eth_dev *eth_dev = NULL;
 885         struct pmd_internals *internals;
 886         int i;
 887
 888         RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
 889                 rte_socket_id());
 890
 891         /* find the ethdev entry */
 892         eth_dev = rte_eth_dev_allocated(name);
 893         if (!eth_dev)
 894                 return 0;
 895
 896         internals = eth_dev->data->dev_private;
 897         for (i = 0; i < internals->nb_queues; i++)
 898                 if (internals->rxq[i].fd != -1)
 899                         close(internals->rxq[i].fd);
 900
 901         close(internals->ioctl_sock);
 902         rte_free(eth_dev->data->dev_private);
 903         rte_free(eth_dev->data);
 904
 905         rte_eth_dev_release_port(eth_dev);
 906
 907         return 0;
 908 }
 909
 910 static struct rte_vdev_driver pmd_tap_drv = {
 911         .probe = rte_pmd_tap_probe,
 912         .remove = rte_pmd_tap_remove,
 913 };
 914 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
 915 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
 916 RTE_PMD_REGISTER_PARAM_STRING(net_tap, "iface=<string>,speed=N");