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 #include <rte_net.h>
  42
  43 #include <sys/types.h>
  44 #include <sys/stat.h>
  45 #include <sys/socket.h>
  46 #include <sys/ioctl.h>
  47 #include <sys/utsname.h>
  48 #include <sys/mman.h>
  49 #include <errno.h>
  50 #include <signal.h>
  51 #include <stdint.h>
  52 #include <unistd.h>
  53 #include <arpa/inet.h>
  54 #include <net/if.h>
  55 #include <linux/if_tun.h>
  56 #include <linux/if_ether.h>
  57 #include <linux/version.h>
  58 #include <fcntl.h>
  59
  60 #include <rte_eth_tap.h>
  61 #include <tap_flow.h>
  62 #include <tap_tcmsgs.h>
  63
  64 /* Linux based path to the TUN device */
  65 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  66 #define DEFAULT_TAP_NAME        "dtap"
  67
  68 #define ETH_TAP_IFACE_ARG       "iface"
  69 #define ETH_TAP_SPEED_ARG       "speed"
  70
  71 #ifdef IFF_MULTI_QUEUE
  72 #define RTE_PMD_TAP_MAX_QUEUES  16
  73 #else
  74 #define RTE_PMD_TAP_MAX_QUEUES  1
  75 #endif
  76
  77 #define FLOWER_KERNEL_VERSION KERNEL_VERSION(4, 2, 0)
  78 #define FLOWER_VLAN_KERNEL_VERSION KERNEL_VERSION(4, 9, 0)
  79
  80 static struct rte_vdev_driver pmd_tap_drv;
  81
  82 static const char *valid_arguments[] = {
  83         ETH_TAP_IFACE_ARG,
  84         ETH_TAP_SPEED_ARG,
  85         NULL
  86 };
  87
  88 static int tap_unit;
  89
  90 static volatile uint32_t tap_trigger;   /* Rx trigger */
  91
  92 static struct rte_eth_link pmd_link = {
  93         .link_speed = ETH_SPEED_NUM_10G,
  94         .link_duplex = ETH_LINK_FULL_DUPLEX,
  95         .link_status = ETH_LINK_DOWN,
  96         .link_autoneg = ETH_LINK_SPEED_AUTONEG
  97 };
  98
  99 static void
 100 tap_trigger_cb(int sig __rte_unused)
 101 {
 102         /* Valid trigger values are nonzero */
 103         tap_trigger = (tap_trigger + 1) | 0x80000000;
 104 }
 105
 106 static int
 107 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 108           struct ifreq *ifr, int set);
 109
 110 /* Tun/Tap allocation routine
 111  *
 112  * name is the number of the interface to use, unless NULL to take the host
 113  * supplied name.
 114  */
 115 static int
 116 tun_alloc(struct pmd_internals *pmd, uint16_t qid)
 117 {
 118         struct ifreq ifr;
 119 #ifdef IFF_MULTI_QUEUE
 120         unsigned int features;
 121 #endif
 122         int fd;
 123
 124         memset(&ifr, 0, sizeof(struct ifreq));
 125
 126         ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
 127         snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
 128
 129         RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
 130
 131         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 132         if (fd < 0) {
 133                 RTE_LOG(ERR, PMD, "Unable to create TAP interface");
 134                 goto error;
 135         }
 136
 137 #ifdef IFF_MULTI_QUEUE
 138         /* Grab the TUN features to verify we can work multi-queue */
 139         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 140                 RTE_LOG(ERR, PMD, "TAP unable to get TUN/TAP features\n");
 141                 goto error;
 142         }
 143         RTE_LOG(DEBUG, PMD, "  TAP Features %08x\n", features);
 144
 145         if (features & IFF_MULTI_QUEUE) {
 146                 RTE_LOG(DEBUG, PMD, "  Multi-queue support for %d queues\n",
 147                         RTE_PMD_TAP_MAX_QUEUES);
 148                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 149         } else
 150 #endif
 151         {
 152                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 153                 RTE_LOG(DEBUG, PMD, "  Single queue only support\n");
 154         }
 155
 156         /* Set the TUN/TAP configuration and set the name if needed */
 157         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 158                 RTE_LOG(WARNING, PMD,
 159                         "Unable to set TUNSETIFF for %s\n",
 160                         ifr.ifr_name);
 161                 perror("TUNSETIFF");
 162                 goto error;
 163         }
 164
 165         /* Always set the file descriptor to non-blocking */
 166         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 167                 RTE_LOG(WARNING, PMD,
 168                         "Unable to set %s to nonblocking\n",
 169                         ifr.ifr_name);
 170                 perror("F_SETFL, NONBLOCK");
 171                 goto error;
 172         }
 173
 174         /* Set up trigger to optimize empty Rx bursts */
 175         errno = 0;
 176         do {
 177                 struct sigaction sa;
 178                 int flags = fcntl(fd, F_GETFL);
 179
 180                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 181                         break;
 182                 if (sa.sa_handler != tap_trigger_cb) {
 183                         /*
 184                          * Make sure SIGIO is not already taken. This is done
 185                          * as late as possible to leave the application a
 186                          * chance to set up its own signal handler first.
 187                          */
 188                         if (sa.sa_handler != SIG_IGN &&
 189                             sa.sa_handler != SIG_DFL) {
 190                                 errno = EBUSY;
 191                                 break;
 192                         }
 193                         sa = (struct sigaction){
 194                                 .sa_flags = SA_RESTART,
 195                                 .sa_handler = tap_trigger_cb,
 196                         };
 197                         if (sigaction(SIGIO, &sa, NULL) == -1)
 198                                 break;
 199                 }
 200                 /* Enable SIGIO on file descriptor */
 201                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 202                 fcntl(fd, F_SETOWN, getpid());
 203         } while (0);
 204         if (errno) {
 205                 /* Disable trigger globally in case of error */
 206                 tap_trigger = 0;
 207                 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
 208                         strerror(errno));
 209         }
 210
 211         if (qid == 0) {
 212                 struct ifreq ifr;
 213
 214                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0) < 0)
 215                         goto error;
 216                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
 217                            ETHER_ADDR_LEN);
 218
 219                 pmd->if_index = if_nametoindex(pmd->name);
 220                 if (!pmd->if_index) {
 221                         RTE_LOG(ERR, PMD,
 222                                 "Could not find ifindex for %s: rte_flow won't be usable.\n",
 223                                 pmd->name);
 224                         return fd;
 225                 }
 226                 if (!pmd->flower_support)
 227                         return fd;
 228                 if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
 229                         RTE_LOG(ERR, PMD,
 230                                 "Could not create multiq qdisc for %s: rte_flow won't be usable.\n",
 231                                 pmd->name);
 232                         return fd;
 233                 }
 234                 if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
 235                         RTE_LOG(ERR, PMD,
 236                                 "Could not create multiq qdisc for %s: rte_flow won't be usable.\n",
 237                                 pmd->name);
 238                         return fd;
 239                 }
 240         }
 241
 242         return fd;
 243
 244 error:
 245         if (fd > 0)
 246                 close(fd);
 247         return -1;
 248 }
 249
 250 /* Callback to handle the rx burst of packets to the correct interface and
 251  * file descriptor(s) in a multi-queue setup.
 252  */
 253 static uint16_t
 254 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 255 {
 256         int len;
 257         struct rte_mbuf *mbuf;
 258         struct rx_queue *rxq = queue;
 259         uint16_t num_rx;
 260         unsigned long num_rx_bytes = 0;
 261         uint32_t trigger = tap_trigger;
 262
 263         if (trigger == rxq->trigger_seen)
 264                 return 0;
 265         if (trigger)
 266                 rxq->trigger_seen = trigger;
 267         rte_compiler_barrier();
 268         for (num_rx = 0; num_rx < nb_pkts; ) {
 269                 /* allocate the next mbuf */
 270                 mbuf = rte_pktmbuf_alloc(rxq->mp);
 271                 if (unlikely(!mbuf)) {
 272                         RTE_LOG(WARNING, PMD, "TAP unable to allocate mbuf\n");
 273                         break;
 274                 }
 275
 276                 len = read(rxq->fd, rte_pktmbuf_mtod(mbuf, char *),
 277                            rte_pktmbuf_tailroom(mbuf));
 278                 if (len <= 0) {
 279                         rte_pktmbuf_free(mbuf);
 280                         break;
 281                 }
 282
 283                 mbuf->data_len = len;
 284                 mbuf->pkt_len = len;
 285                 mbuf->port = rxq->in_port;
 286                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 287                                                       RTE_PTYPE_ALL_MASK);
 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         uint32_t max_size;
 309         int i, n;
 310
 311         if (unlikely(nb_pkts == 0))
 312                 return 0;
 313
 314         max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
 315         for (i = 0; i < nb_pkts; i++) {
 316                 /* copy the tx frame data */
 317                 mbuf = bufs[num_tx];
 318                 /* stats.errs will be incremented */
 319                 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
 320                         break;
 321                 n = write(txq->fd,
 322                           rte_pktmbuf_mtod(mbuf, void *),
 323                           rte_pktmbuf_pkt_len(mbuf));
 324                 if (n <= 0)
 325                         break;
 326
 327                 num_tx++;
 328                 num_tx_bytes += mbuf->pkt_len;
 329                 rte_pktmbuf_free(mbuf);
 330         }
 331
 332         txq->stats.opackets += num_tx;
 333         txq->stats.errs += nb_pkts - num_tx;
 334         txq->stats.obytes += num_tx_bytes;
 335
 336         return num_tx;
 337 }
 338
 339 static int
 340 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 341           struct ifreq *ifr, int set)
 342 {
 343         short req_flags = ifr->ifr_flags;
 344
 345         snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 346         switch (request) {
 347         case SIOCSIFFLAGS:
 348                 /* fetch current flags to leave other flags untouched */
 349                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 350                         goto error;
 351                 if (set)
 352                         ifr->ifr_flags |= req_flags;
 353                 else
 354                         ifr->ifr_flags &= ~req_flags;
 355                 break;
 356         case SIOCGIFHWADDR:
 357         case SIOCSIFHWADDR:
 358         case SIOCSIFMTU:
 359                 break;
 360         default:
 361                 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
 362                         pmd->name);
 363                 return -EINVAL;
 364         }
 365         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 366                 goto error;
 367         return 0;
 368
 369 error:
 370         RTE_LOG(ERR, PMD, "%s: ioctl(%lu) failed with error: %s\n",
 371                 ifr->ifr_name, request, strerror(errno));
 372         return -errno;
 373 }
 374
 375 static int
 376 tap_link_set_down(struct rte_eth_dev *dev)
 377 {
 378         struct pmd_internals *pmd = dev->data->dev_private;
 379         struct ifreq ifr = { .ifr_flags = IFF_UP };
 380
 381         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 382         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0);
 383 }
 384
 385 static int
 386 tap_link_set_up(struct rte_eth_dev *dev)
 387 {
 388         struct pmd_internals *pmd = dev->data->dev_private;
 389         struct ifreq ifr = { .ifr_flags = IFF_UP };
 390
 391         dev->data->dev_link.link_status = ETH_LINK_UP;
 392         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1);
 393 }
 394
 395 static int
 396 tap_dev_start(struct rte_eth_dev *dev)
 397 {
 398         return tap_link_set_up(dev);
 399 }
 400
 401 /* This function gets called when the current port gets stopped.
 402  */
 403 static void
 404 tap_dev_stop(struct rte_eth_dev *dev)
 405 {
 406         tap_link_set_down(dev);
 407 }
 408
 409 static int
 410 tap_dev_configure(struct rte_eth_dev *dev __rte_unused)
 411 {
 412         return 0;
 413 }
 414
 415 static uint32_t
 416 tap_dev_speed_capa(void)
 417 {
 418         uint32_t speed = pmd_link.link_speed;
 419         uint32_t capa = 0;
 420
 421         if (speed >= ETH_SPEED_NUM_10M)
 422                 capa |= ETH_LINK_SPEED_10M;
 423         if (speed >= ETH_SPEED_NUM_100M)
 424                 capa |= ETH_LINK_SPEED_100M;
 425         if (speed >= ETH_SPEED_NUM_1G)
 426                 capa |= ETH_LINK_SPEED_1G;
 427         if (speed >= ETH_SPEED_NUM_5G)
 428                 capa |= ETH_LINK_SPEED_2_5G;
 429         if (speed >= ETH_SPEED_NUM_5G)
 430                 capa |= ETH_LINK_SPEED_5G;
 431         if (speed >= ETH_SPEED_NUM_10G)
 432                 capa |= ETH_LINK_SPEED_10G;
 433         if (speed >= ETH_SPEED_NUM_20G)
 434                 capa |= ETH_LINK_SPEED_20G;
 435         if (speed >= ETH_SPEED_NUM_25G)
 436                 capa |= ETH_LINK_SPEED_25G;
 437         if (speed >= ETH_SPEED_NUM_40G)
 438                 capa |= ETH_LINK_SPEED_40G;
 439         if (speed >= ETH_SPEED_NUM_50G)
 440                 capa |= ETH_LINK_SPEED_50G;
 441         if (speed >= ETH_SPEED_NUM_56G)
 442                 capa |= ETH_LINK_SPEED_56G;
 443         if (speed >= ETH_SPEED_NUM_100G)
 444                 capa |= ETH_LINK_SPEED_100G;
 445
 446         return capa;
 447 }
 448
 449 static void
 450 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 451 {
 452         struct pmd_internals *internals = dev->data->dev_private;
 453
 454         dev_info->if_index = internals->if_index;
 455         dev_info->max_mac_addrs = 1;
 456         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 457         dev_info->max_rx_queues = internals->nb_queues;
 458         dev_info->max_tx_queues = internals->nb_queues;
 459         dev_info->min_rx_bufsize = 0;
 460         dev_info->pci_dev = NULL;
 461         dev_info->speed_capa = tap_dev_speed_capa();
 462 }
 463
 464 static void
 465 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 466 {
 467         unsigned int i, imax;
 468         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 469         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 470         const struct pmd_internals *pmd = dev->data->dev_private;
 471
 472         imax = (pmd->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 473                 pmd->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 474
 475         for (i = 0; i < imax; i++) {
 476                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 477                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 478                 rx_total += tap_stats->q_ipackets[i];
 479                 rx_bytes_total += tap_stats->q_ibytes[i];
 480
 481                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 482                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 483                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 484                 tx_total += tap_stats->q_opackets[i];
 485                 tx_err_total += tap_stats->q_errors[i];
 486                 tx_bytes_total += tap_stats->q_obytes[i];
 487         }
 488
 489         tap_stats->ipackets = rx_total;
 490         tap_stats->ibytes = rx_bytes_total;
 491         tap_stats->opackets = tx_total;
 492         tap_stats->oerrors = tx_err_total;
 493         tap_stats->obytes = tx_bytes_total;
 494 }
 495
 496 static void
 497 tap_stats_reset(struct rte_eth_dev *dev)
 498 {
 499         int i;
 500         struct pmd_internals *pmd = dev->data->dev_private;
 501
 502         for (i = 0; i < pmd->nb_queues; i++) {
 503                 pmd->rxq[i].stats.ipackets = 0;
 504                 pmd->rxq[i].stats.ibytes = 0;
 505
 506                 pmd->txq[i].stats.opackets = 0;
 507                 pmd->txq[i].stats.errs = 0;
 508                 pmd->txq[i].stats.obytes = 0;
 509         }
 510 }
 511
 512 static void
 513 tap_dev_close(struct rte_eth_dev *dev __rte_unused)
 514 {
 515         int i;
 516         struct pmd_internals *internals = dev->data->dev_private;
 517
 518         tap_link_set_down(dev);
 519         tap_flow_flush(dev, NULL);
 520
 521         for (i = 0; i < internals->nb_queues; i++) {
 522                 if (internals->rxq[i].fd != -1)
 523                         close(internals->rxq[i].fd);
 524                 internals->rxq[i].fd = -1;
 525                 internals->txq[i].fd = -1;
 526         }
 527 }
 528
 529 static void
 530 tap_rx_queue_release(void *queue)
 531 {
 532         struct rx_queue *rxq = queue;
 533
 534         if (rxq && (rxq->fd > 0)) {
 535                 close(rxq->fd);
 536                 rxq->fd = -1;
 537         }
 538 }
 539
 540 static void
 541 tap_tx_queue_release(void *queue)
 542 {
 543         struct tx_queue *txq = queue;
 544
 545         if (txq && (txq->fd > 0)) {
 546                 close(txq->fd);
 547                 txq->fd = -1;
 548         }
 549 }
 550
 551 static int
 552 tap_link_update(struct rte_eth_dev *dev __rte_unused,
 553                 int wait_to_complete __rte_unused)
 554 {
 555         return 0;
 556 }
 557
 558 static void
 559 tap_promisc_enable(struct rte_eth_dev *dev)
 560 {
 561         struct pmd_internals *pmd = dev->data->dev_private;
 562         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 563
 564         dev->data->promiscuous = 1;
 565         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1);
 566 }
 567
 568 static void
 569 tap_promisc_disable(struct rte_eth_dev *dev)
 570 {
 571         struct pmd_internals *pmd = dev->data->dev_private;
 572         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 573
 574         dev->data->promiscuous = 0;
 575         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0);
 576 }
 577
 578 static void
 579 tap_allmulti_enable(struct rte_eth_dev *dev)
 580 {
 581         struct pmd_internals *pmd = dev->data->dev_private;
 582         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 583
 584         dev->data->all_multicast = 1;
 585         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1);
 586 }
 587
 588 static void
 589 tap_allmulti_disable(struct rte_eth_dev *dev)
 590 {
 591         struct pmd_internals *pmd = dev->data->dev_private;
 592         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 593
 594         dev->data->all_multicast = 0;
 595         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0);
 596 }
 597
 598
 599 static void
 600 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 601 {
 602         struct pmd_internals *pmd = dev->data->dev_private;
 603         struct ifreq ifr;
 604
 605         if (is_zero_ether_addr(mac_addr)) {
 606                 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
 607                         dev->data->name);
 608                 return;
 609         }
 610
 611         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
 612         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
 613         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1) < 0)
 614                 return;
 615         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
 616 }
 617
 618 static int
 619 tap_setup_queue(struct rte_eth_dev *dev,
 620                 struct pmd_internals *internals,
 621                 uint16_t qid)
 622 {
 623         struct pmd_internals *pmd = dev->data->dev_private;
 624         struct rx_queue *rx = &internals->rxq[qid];
 625         struct tx_queue *tx = &internals->txq[qid];
 626         int fd;
 627
 628         fd = rx->fd;
 629         if (fd < 0) {
 630                 fd = tx->fd;
 631                 if (fd < 0) {
 632                         RTE_LOG(INFO, PMD, "Add queue to TAP %s for qid %d\n",
 633                                 pmd->name, qid);
 634                         fd = tun_alloc(pmd, qid);
 635                         if (fd < 0) {
 636                                 RTE_LOG(ERR, PMD, "tun_alloc(%s, %d) failed\n",
 637                                         pmd->name, qid);
 638                                 return -1;
 639                         }
 640                         if (qid == 0) {
 641                                 struct ifreq ifr;
 642
 643                                 ifr.ifr_mtu = dev->data->mtu;
 644                                 if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1) < 0) {
 645                                         close(fd);
 646                                         return -1;
 647                                 }
 648                         }
 649                 }
 650         }
 651
 652         rx->fd = fd;
 653         tx->fd = fd;
 654         tx->mtu = &dev->data->mtu;
 655
 656         return fd;
 657 }
 658
 659 static int
 660 rx_setup_queue(struct rte_eth_dev *dev,
 661                 struct pmd_internals *internals,
 662                 uint16_t qid)
 663 {
 664         dev->data->rx_queues[qid] = &internals->rxq[qid];
 665
 666         return tap_setup_queue(dev, internals, qid);
 667 }
 668
 669 static int
 670 tx_setup_queue(struct rte_eth_dev *dev,
 671                 struct pmd_internals *internals,
 672                 uint16_t qid)
 673 {
 674         dev->data->tx_queues[qid] = &internals->txq[qid];
 675
 676         return tap_setup_queue(dev, internals, qid);
 677 }
 678
 679 static int
 680 tap_rx_queue_setup(struct rte_eth_dev *dev,
 681                    uint16_t rx_queue_id,
 682                    uint16_t nb_rx_desc __rte_unused,
 683                    unsigned int socket_id __rte_unused,
 684                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 685                    struct rte_mempool *mp)
 686 {
 687         struct pmd_internals *internals = dev->data->dev_private;
 688         uint16_t buf_size;
 689         int fd;
 690
 691         if ((rx_queue_id >= internals->nb_queues) || !mp) {
 692                 RTE_LOG(WARNING, PMD,
 693                         "nb_queues %d too small or mempool NULL\n",
 694                         internals->nb_queues);
 695                 return -1;
 696         }
 697
 698         internals->rxq[rx_queue_id].mp = mp;
 699         internals->rxq[rx_queue_id].trigger_seen = 1; /* force initial burst */
 700         internals->rxq[rx_queue_id].in_port = dev->data->port_id;
 701
 702         /* Now get the space available for data in the mbuf */
 703         buf_size = (uint16_t)(rte_pktmbuf_data_room_size(mp) -
 704                                 RTE_PKTMBUF_HEADROOM);
 705
 706         if (buf_size < ETH_FRAME_LEN) {
 707                 RTE_LOG(WARNING, PMD,
 708                         "%s: %d bytes will not fit in mbuf (%d bytes)\n",
 709                         dev->data->name, ETH_FRAME_LEN, buf_size);
 710                 return -ENOMEM;
 711         }
 712
 713         fd = rx_setup_queue(dev, internals, rx_queue_id);
 714         if (fd == -1)
 715                 return -1;
 716
 717         RTE_LOG(DEBUG, PMD, "  RX TAP device name %s, qid %d on fd %d\n",
 718                 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
 719
 720         return 0;
 721 }
 722
 723 static int
 724 tap_tx_queue_setup(struct rte_eth_dev *dev,
 725                    uint16_t tx_queue_id,
 726                    uint16_t nb_tx_desc __rte_unused,
 727                    unsigned int socket_id __rte_unused,
 728                    const struct rte_eth_txconf *tx_conf __rte_unused)
 729 {
 730         struct pmd_internals *internals = dev->data->dev_private;
 731         int ret;
 732
 733         if (tx_queue_id >= internals->nb_queues)
 734                 return -1;
 735
 736         ret = tx_setup_queue(dev, internals, tx_queue_id);
 737         if (ret == -1)
 738                 return -1;
 739
 740         RTE_LOG(DEBUG, PMD, "  TX TAP device name %s, qid %d on fd %d\n",
 741                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd);
 742
 743         return 0;
 744 }
 745
 746 static int
 747 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 748 {
 749         struct pmd_internals *pmd = dev->data->dev_private;
 750         struct ifreq ifr = { .ifr_mtu = mtu };
 751         int err = 0;
 752
 753         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1);
 754         if (!err)
 755                 dev->data->mtu = mtu;
 756
 757         return err;
 758 }
 759
 760 static int
 761 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
 762                      struct ether_addr *mc_addr_set __rte_unused,
 763                      uint32_t nb_mc_addr __rte_unused)
 764 {
 765         /*
 766          * Nothing to do actually: the tap has no filtering whatsoever, every
 767          * packet is received.
 768          */
 769         return 0;
 770 }
 771
 772 static const uint32_t*
 773 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
 774 {
 775         static const uint32_t ptypes[] = {
 776                 RTE_PTYPE_INNER_L2_ETHER,
 777                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
 778                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
 779                 RTE_PTYPE_INNER_L3_IPV4,
 780                 RTE_PTYPE_INNER_L3_IPV4_EXT,
 781                 RTE_PTYPE_INNER_L3_IPV6,
 782                 RTE_PTYPE_INNER_L3_IPV6_EXT,
 783                 RTE_PTYPE_INNER_L4_FRAG,
 784                 RTE_PTYPE_INNER_L4_UDP,
 785                 RTE_PTYPE_INNER_L4_TCP,
 786                 RTE_PTYPE_INNER_L4_SCTP,
 787                 RTE_PTYPE_L2_ETHER,
 788                 RTE_PTYPE_L2_ETHER_VLAN,
 789                 RTE_PTYPE_L2_ETHER_QINQ,
 790                 RTE_PTYPE_L3_IPV4,
 791                 RTE_PTYPE_L3_IPV4_EXT,
 792                 RTE_PTYPE_L3_IPV6_EXT,
 793                 RTE_PTYPE_L3_IPV6,
 794                 RTE_PTYPE_L4_FRAG,
 795                 RTE_PTYPE_L4_UDP,
 796                 RTE_PTYPE_L4_TCP,
 797                 RTE_PTYPE_L4_SCTP,
 798         };
 799
 800         return ptypes;
 801 }
 802
 803 static int
 804 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
 805                   struct rte_eth_fc_conf *fc_conf)
 806 {
 807         fc_conf->mode = RTE_FC_NONE;
 808         return 0;
 809 }
 810
 811 static int
 812 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
 813                   struct rte_eth_fc_conf *fc_conf)
 814 {
 815         if (fc_conf->mode != RTE_FC_NONE)
 816                 return -ENOTSUP;
 817         return 0;
 818 }
 819
 820 static const struct eth_dev_ops ops = {
 821         .dev_start              = tap_dev_start,
 822         .dev_stop               = tap_dev_stop,
 823         .dev_close              = tap_dev_close,
 824         .dev_configure          = tap_dev_configure,
 825         .dev_infos_get          = tap_dev_info,
 826         .rx_queue_setup         = tap_rx_queue_setup,
 827         .tx_queue_setup         = tap_tx_queue_setup,
 828         .rx_queue_release       = tap_rx_queue_release,
 829         .tx_queue_release       = tap_tx_queue_release,
 830         .flow_ctrl_get          = tap_flow_ctrl_get,
 831         .flow_ctrl_set          = tap_flow_ctrl_set,
 832         .link_update            = tap_link_update,
 833         .dev_set_link_up        = tap_link_set_up,
 834         .dev_set_link_down      = tap_link_set_down,
 835         .promiscuous_enable     = tap_promisc_enable,
 836         .promiscuous_disable    = tap_promisc_disable,
 837         .allmulticast_enable    = tap_allmulti_enable,
 838         .allmulticast_disable   = tap_allmulti_disable,
 839         .mac_addr_set           = tap_mac_set,
 840         .mtu_set                = tap_mtu_set,
 841         .set_mc_addr_list       = tap_set_mc_addr_list,
 842         .stats_get              = tap_stats_get,
 843         .stats_reset            = tap_stats_reset,
 844         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
 845         .filter_ctrl            = tap_dev_filter_ctrl,
 846 };
 847
 848 static int
 849 tap_kernel_support(struct pmd_internals *pmd)
 850 {
 851         struct utsname utsname;
 852         int ver[3];
 853
 854         if (uname(&utsname) == -1 ||
 855             sscanf(utsname.release, "%d.%d.%d",
 856                    &ver[0], &ver[1], &ver[2]) != 3)
 857                 return 0;
 858         if (KERNEL_VERSION(ver[0], ver[1], ver[2]) >= FLOWER_KERNEL_VERSION)
 859                 pmd->flower_support = 1;
 860         if (KERNEL_VERSION(ver[0], ver[1], ver[2]) >=
 861             FLOWER_VLAN_KERNEL_VERSION)
 862                 pmd->flower_vlan_support = 1;
 863         return 1;
 864 }
 865
 866 static int
 867 eth_dev_tap_create(const char *name, char *tap_name)
 868 {
 869         int numa_node = rte_socket_id();
 870         struct rte_eth_dev *dev = NULL;
 871         struct pmd_internals *pmd = NULL;
 872         struct rte_eth_dev_data *data = NULL;
 873         int i;
 874
 875         RTE_LOG(DEBUG, PMD, "  TAP device on numa %u\n", rte_socket_id());
 876
 877         data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
 878         if (!data) {
 879                 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
 880                 goto error_exit;
 881         }
 882
 883         pmd = rte_zmalloc_socket(tap_name, sizeof(*pmd), 0, numa_node);
 884         if (!pmd) {
 885                 RTE_LOG(ERR, PMD, "TAP Unable to allocate internal struct\n");
 886                 goto error_exit;
 887         }
 888
 889         /* name in allocation and data->name must be consistent */
 890         snprintf(data->name, sizeof(data->name), "%s", name);
 891         dev = rte_eth_dev_allocate(name);
 892         if (!dev) {
 893                 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
 894                 goto error_exit;
 895         }
 896
 897         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
 898
 899         pmd->nb_queues = RTE_PMD_TAP_MAX_QUEUES;
 900
 901         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
 902         if (pmd->ioctl_sock == -1) {
 903                 RTE_LOG(ERR, PMD,
 904                         "TAP Unable to get a socket for management: %s\n",
 905                         strerror(errno));
 906                 goto error_exit;
 907         }
 908
 909         /* Setup some default values */
 910         data->dev_private = pmd;
 911         data->port_id = dev->data->port_id;
 912         data->mtu = dev->data->mtu;
 913         data->dev_flags = RTE_ETH_DEV_DETACHABLE;
 914         data->kdrv = RTE_KDRV_NONE;
 915         data->drv_name = pmd_tap_drv.driver.name;
 916         data->numa_node = numa_node;
 917
 918         data->dev_link = pmd_link;
 919         data->mac_addrs = &pmd->eth_addr;
 920         data->nb_rx_queues = pmd->nb_queues;
 921         data->nb_tx_queues = pmd->nb_queues;
 922
 923         dev->data = data;
 924         dev->dev_ops = &ops;
 925         dev->driver = NULL;
 926         dev->rx_pkt_burst = pmd_rx_burst;
 927         dev->tx_pkt_burst = pmd_tx_burst;
 928
 929         /* Presetup the fds to -1 as being not valid */
 930         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 931                 pmd->rxq[i].fd = -1;
 932                 pmd->txq[i].fd = -1;
 933         }
 934
 935         tap_kernel_support(pmd);
 936         if (!pmd->flower_support)
 937                 return 0;
 938         LIST_INIT(&pmd->flows);
 939         /*
 940          * If no netlink socket can be created, then it will fail when
 941          * creating/destroying flow rules.
 942          */
 943         pmd->nlsk_fd = nl_init();
 944
 945         return 0;
 946
 947 error_exit:
 948         RTE_LOG(DEBUG, PMD, "TAP Unable to initialize %s\n", name);
 949
 950         rte_free(data);
 951         rte_free(pmd);
 952
 953         rte_eth_dev_release_port(dev);
 954
 955         return -EINVAL;
 956 }
 957
 958 static int
 959 set_interface_name(const char *key __rte_unused,
 960                    const char *value,
 961                    void *extra_args)
 962 {
 963         char *name = (char *)extra_args;
 964
 965         if (value)
 966                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
 967         else
 968                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
 969                          DEFAULT_TAP_NAME, (tap_unit - 1));
 970
 971         return 0;
 972 }
 973
 974 static int
 975 set_interface_speed(const char *key __rte_unused,
 976                     const char *value,
 977                     void *extra_args)
 978 {
 979         *(int *)extra_args = (value) ? atoi(value) : ETH_SPEED_NUM_10G;
 980
 981         return 0;
 982 }
 983
 984 /* Open a TAP interface device.
 985  */
 986 static int
 987 rte_pmd_tap_probe(const char *name, const char *params)
 988 {
 989         int ret;
 990         struct rte_kvargs *kvlist = NULL;
 991         int speed;
 992         char tap_name[RTE_ETH_NAME_MAX_LEN];
 993
 994         speed = ETH_SPEED_NUM_10G;
 995         snprintf(tap_name, sizeof(tap_name), "%s%d",
 996                  DEFAULT_TAP_NAME, tap_unit++);
 997
 998         if (params && (params[0] != '\0')) {
 999                 RTE_LOG(DEBUG, PMD, "paramaters (%s)\n", params);
1000
1001                 kvlist = rte_kvargs_parse(params, valid_arguments);
1002                 if (kvlist) {
1003                         if (rte_kvargs_count(kvlist, ETH_TAP_SPEED_ARG) == 1) {
1004                                 ret = rte_kvargs_process(kvlist,
1005                                                          ETH_TAP_SPEED_ARG,
1006                                                          &set_interface_speed,
1007                                                          &speed);
1008                                 if (ret == -1)
1009                                         goto leave;
1010                         }
1011
1012                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1013                                 ret = rte_kvargs_process(kvlist,
1014                                                          ETH_TAP_IFACE_ARG,
1015                                                          &set_interface_name,
1016                                                          tap_name);
1017                                 if (ret == -1)
1018                                         goto leave;
1019                         }
1020                 }
1021         }
1022         pmd_link.link_speed = speed;
1023
1024         RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1025                 name, tap_name);
1026
1027         ret = eth_dev_tap_create(name, tap_name);
1028
1029 leave:
1030         if (ret == -1) {
1031                 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1032                         name, tap_name);
1033                 tap_unit--;             /* Restore the unit number */
1034         }
1035         rte_kvargs_free(kvlist);
1036
1037         return ret;
1038 }
1039
1040 /* detach a TAP device.
1041  */
1042 static int
1043 rte_pmd_tap_remove(const char *name)
1044 {
1045         struct rte_eth_dev *eth_dev = NULL;
1046         struct pmd_internals *internals;
1047         int i;
1048
1049         RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1050                 rte_socket_id());
1051
1052         /* find the ethdev entry */
1053         eth_dev = rte_eth_dev_allocated(name);
1054         if (!eth_dev)
1055                 return 0;
1056
1057         internals = eth_dev->data->dev_private;
1058         if (internals->flower_support && internals->nlsk_fd) {
1059                 tap_flow_flush(eth_dev, NULL);
1060                 nl_final(internals->nlsk_fd);
1061         }
1062         for (i = 0; i < internals->nb_queues; i++)
1063                 if (internals->rxq[i].fd != -1)
1064                         close(internals->rxq[i].fd);
1065
1066         close(internals->ioctl_sock);
1067         rte_free(eth_dev->data->dev_private);
1068         rte_free(eth_dev->data);
1069
1070         rte_eth_dev_release_port(eth_dev);
1071
1072         return 0;
1073 }
1074
1075 static struct rte_vdev_driver pmd_tap_drv = {
1076         .probe = rte_pmd_tap_probe,
1077         .remove = rte_pmd_tap_remove,
1078 };
1079 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1080 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1081 RTE_PMD_REGISTER_PARAM_STRING(net_tap, "iface=<string>,speed=N");