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_branch_prediction.h>
  36 #include <rte_common.h>
  37 #include <rte_mbuf.h>
  38 #include <rte_ethdev.h>
  39 #include <rte_malloc.h>
  40 #include <rte_vdev.h>
  41 #include <rte_kvargs.h>
  42 #include <rte_net.h>
  43
  44 #include <sys/types.h>
  45 #include <sys/stat.h>
  46 #include <sys/socket.h>
  47 #include <sys/ioctl.h>
  48 #include <sys/utsname.h>
  49 #include <sys/mman.h>
  50 #include <errno.h>
  51 #include <signal.h>
  52 #include <stdint.h>
  53 #include <sys/uio.h>
  54 #include <unistd.h>
  55 #include <arpa/inet.h>
  56 #include <net/if.h>
  57 #include <linux/if_tun.h>
  58 #include <linux/if_ether.h>
  59 #include <linux/version.h>
  60 #include <fcntl.h>
  61
  62 #include <rte_eth_tap.h>
  63 #include <tap_flow.h>
  64 #include <tap_tcmsgs.h>
  65
  66 /* Linux based path to the TUN device */
  67 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  68 #define DEFAULT_TAP_NAME        "dtap"
  69
  70 #define ETH_TAP_IFACE_ARG       "iface"
  71 #define ETH_TAP_SPEED_ARG       "speed"
  72
  73 #ifdef IFF_MULTI_QUEUE
  74 #define RTE_PMD_TAP_MAX_QUEUES  16
  75 #else
  76 #define RTE_PMD_TAP_MAX_QUEUES  1
  77 #endif
  78
  79 #define FLOWER_KERNEL_VERSION KERNEL_VERSION(4, 2, 0)
  80 #define FLOWER_VLAN_KERNEL_VERSION KERNEL_VERSION(4, 9, 0)
  81
  82 static struct rte_vdev_driver pmd_tap_drv;
  83
  84 static const char *valid_arguments[] = {
  85         ETH_TAP_IFACE_ARG,
  86         ETH_TAP_SPEED_ARG,
  87         NULL
  88 };
  89
  90 static int tap_unit;
  91
  92 static volatile uint32_t tap_trigger;   /* Rx trigger */
  93
  94 static struct rte_eth_link pmd_link = {
  95         .link_speed = ETH_SPEED_NUM_10G,
  96         .link_duplex = ETH_LINK_FULL_DUPLEX,
  97         .link_status = ETH_LINK_DOWN,
  98         .link_autoneg = ETH_LINK_SPEED_AUTONEG
  99 };
 100
 101 static void
 102 tap_trigger_cb(int sig __rte_unused)
 103 {
 104         /* Valid trigger values are nonzero */
 105         tap_trigger = (tap_trigger + 1) | 0x80000000;
 106 }
 107
 108 static int
 109 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 110           struct ifreq *ifr, int set);
 111
 112 /* Tun/Tap allocation routine
 113  *
 114  * name is the number of the interface to use, unless NULL to take the host
 115  * supplied name.
 116  */
 117 static int
 118 tun_alloc(struct pmd_internals *pmd, uint16_t qid)
 119 {
 120         struct ifreq ifr;
 121 #ifdef IFF_MULTI_QUEUE
 122         unsigned int features;
 123 #endif
 124         int fd;
 125
 126         memset(&ifr, 0, sizeof(struct ifreq));
 127
 128         /*
 129          * Do not set IFF_NO_PI as packet information header will be needed
 130          * to check if a received packet has been truncated.
 131          */
 132         ifr.ifr_flags = IFF_TAP;
 133         snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
 134
 135         RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
 136
 137         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 138         if (fd < 0) {
 139                 RTE_LOG(ERR, PMD, "Unable to create TAP interface");
 140                 goto error;
 141         }
 142
 143 #ifdef IFF_MULTI_QUEUE
 144         /* Grab the TUN features to verify we can work multi-queue */
 145         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 146                 RTE_LOG(ERR, PMD, "TAP unable to get TUN/TAP features\n");
 147                 goto error;
 148         }
 149         RTE_LOG(DEBUG, PMD, "  TAP Features %08x\n", features);
 150
 151         if (features & IFF_MULTI_QUEUE) {
 152                 RTE_LOG(DEBUG, PMD, "  Multi-queue support for %d queues\n",
 153                         RTE_PMD_TAP_MAX_QUEUES);
 154                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 155         } else
 156 #endif
 157         {
 158                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 159                 RTE_LOG(DEBUG, PMD, "  Single queue only support\n");
 160         }
 161
 162         /* Set the TUN/TAP configuration and set the name if needed */
 163         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 164                 RTE_LOG(WARNING, PMD,
 165                         "Unable to set TUNSETIFF for %s\n",
 166                         ifr.ifr_name);
 167                 perror("TUNSETIFF");
 168                 goto error;
 169         }
 170
 171         /* Always set the file descriptor to non-blocking */
 172         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 173                 RTE_LOG(WARNING, PMD,
 174                         "Unable to set %s to nonblocking\n",
 175                         ifr.ifr_name);
 176                 perror("F_SETFL, NONBLOCK");
 177                 goto error;
 178         }
 179
 180         /* Set up trigger to optimize empty Rx bursts */
 181         errno = 0;
 182         do {
 183                 struct sigaction sa;
 184                 int flags = fcntl(fd, F_GETFL);
 185
 186                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 187                         break;
 188                 if (sa.sa_handler != tap_trigger_cb) {
 189                         /*
 190                          * Make sure SIGIO is not already taken. This is done
 191                          * as late as possible to leave the application a
 192                          * chance to set up its own signal handler first.
 193                          */
 194                         if (sa.sa_handler != SIG_IGN &&
 195                             sa.sa_handler != SIG_DFL) {
 196                                 errno = EBUSY;
 197                                 break;
 198                         }
 199                         sa = (struct sigaction){
 200                                 .sa_flags = SA_RESTART,
 201                                 .sa_handler = tap_trigger_cb,
 202                         };
 203                         if (sigaction(SIGIO, &sa, NULL) == -1)
 204                                 break;
 205                 }
 206                 /* Enable SIGIO on file descriptor */
 207                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 208                 fcntl(fd, F_SETOWN, getpid());
 209         } while (0);
 210         if (errno) {
 211                 /* Disable trigger globally in case of error */
 212                 tap_trigger = 0;
 213                 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
 214                         strerror(errno));
 215         }
 216
 217         if (qid == 0) {
 218                 struct ifreq ifr;
 219
 220                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0) < 0)
 221                         goto error;
 222                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
 223                            ETHER_ADDR_LEN);
 224
 225                 pmd->if_index = if_nametoindex(pmd->name);
 226                 if (!pmd->if_index) {
 227                         RTE_LOG(ERR, PMD,
 228                                 "Could not find ifindex for %s: rte_flow won't be usable.\n",
 229                                 pmd->name);
 230                         return fd;
 231                 }
 232                 if (!pmd->flower_support)
 233                         return fd;
 234                 if (qdisc_create_multiq(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                 if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
 241                         RTE_LOG(ERR, PMD,
 242                                 "Could not create multiq qdisc for %s: rte_flow won't be usable.\n",
 243                                 pmd->name);
 244                         return fd;
 245                 }
 246         }
 247
 248         return fd;
 249
 250 error:
 251         if (fd > 0)
 252                 close(fd);
 253         return -1;
 254 }
 255
 256 /* Callback to handle the rx burst of packets to the correct interface and
 257  * file descriptor(s) in a multi-queue setup.
 258  */
 259 static uint16_t
 260 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 261 {
 262         struct rx_queue *rxq = queue;
 263         uint16_t num_rx;
 264         unsigned long num_rx_bytes = 0;
 265         uint32_t trigger = tap_trigger;
 266
 267         if (trigger == rxq->trigger_seen)
 268                 return 0;
 269         if (trigger)
 270                 rxq->trigger_seen = trigger;
 271         rte_compiler_barrier();
 272         for (num_rx = 0; num_rx < nb_pkts; ) {
 273                 struct rte_mbuf *mbuf = rxq->pool;
 274                 struct rte_mbuf *seg = NULL;
 275                 struct rte_mbuf *new_tail = NULL;
 276                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 277                 int len;
 278
 279                 len = readv(rxq->fd, *rxq->iovecs,
 280                             1 + (rxq->rxmode->enable_scatter ?
 281                                  rxq->nb_rx_desc : 1));
 282                 if (len < (int)sizeof(struct tun_pi))
 283                         break;
 284
 285                 /* Packet couldn't fit in the provided mbuf */
 286                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 287                         rxq->stats.ierrors++;
 288                         continue;
 289                 }
 290
 291                 len -= sizeof(struct tun_pi);
 292
 293                 mbuf->pkt_len = len;
 294                 mbuf->port = rxq->in_port;
 295                 while (1) {
 296                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 297
 298                         if (unlikely(!buf)) {
 299                                 rxq->stats.rx_nombuf++;
 300                                 /* No new buf has been allocated: do nothing */
 301                                 if (!new_tail || !seg)
 302                                         goto end;
 303
 304                                 seg->next = NULL;
 305                                 rte_pktmbuf_free(mbuf);
 306
 307                                 goto end;
 308                         }
 309                         seg = seg ? seg->next : mbuf;
 310                         if (rxq->pool == mbuf)
 311                                 rxq->pool = buf;
 312                         if (new_tail)
 313                                 new_tail->next = buf;
 314                         new_tail = buf;
 315                         new_tail->next = seg->next;
 316
 317                         /* iovecs[0] is reserved for packet info (pi) */
 318                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 319                                 buf->buf_len - data_off;
 320                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 321                                 (char *)buf->buf_addr + data_off;
 322
 323                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 324                         seg->data_off = data_off;
 325
 326                         len -= seg->data_len;
 327                         if (len <= 0)
 328                                 break;
 329                         mbuf->nb_segs++;
 330                         /* First segment has headroom, not the others */
 331                         data_off = 0;
 332                 }
 333                 seg->next = NULL;
 334                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 335                                                       RTE_PTYPE_ALL_MASK);
 336
 337                 /* account for the receive frame */
 338                 bufs[num_rx++] = mbuf;
 339                 num_rx_bytes += mbuf->pkt_len;
 340         }
 341 end:
 342         rxq->stats.ipackets += num_rx;
 343         rxq->stats.ibytes += num_rx_bytes;
 344
 345         return num_rx;
 346 }
 347
 348 /* Callback to handle sending packets from the tap interface
 349  */
 350 static uint16_t
 351 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 352 {
 353         struct tx_queue *txq = queue;
 354         uint16_t num_tx = 0;
 355         unsigned long num_tx_bytes = 0;
 356         uint32_t max_size;
 357         int i;
 358
 359         if (unlikely(nb_pkts == 0))
 360                 return 0;
 361
 362         max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
 363         for (i = 0; i < nb_pkts; i++) {
 364                 struct rte_mbuf *mbuf = bufs[num_tx];
 365                 struct iovec iovecs[mbuf->nb_segs + 1];
 366                 struct tun_pi pi = { .flags = 0 };
 367                 struct rte_mbuf *seg = mbuf;
 368                 int n;
 369                 int j;
 370
 371                 /* stats.errs will be incremented */
 372                 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
 373                         break;
 374
 375                 iovecs[0].iov_base = &pi;
 376                 iovecs[0].iov_len = sizeof(pi);
 377                 for (j = 1; j <= mbuf->nb_segs; j++) {
 378                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 379                         iovecs[j].iov_base =
 380                                 rte_pktmbuf_mtod(seg, void *);
 381                         seg = seg->next;
 382                 }
 383                 /* copy the tx frame data */
 384                 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
 385                 if (n <= 0)
 386                         break;
 387
 388                 num_tx++;
 389                 num_tx_bytes += mbuf->pkt_len;
 390                 rte_pktmbuf_free(mbuf);
 391         }
 392
 393         txq->stats.opackets += num_tx;
 394         txq->stats.errs += nb_pkts - num_tx;
 395         txq->stats.obytes += num_tx_bytes;
 396
 397         return num_tx;
 398 }
 399
 400 static int
 401 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 402           struct ifreq *ifr, int set)
 403 {
 404         short req_flags = ifr->ifr_flags;
 405
 406         snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 407         switch (request) {
 408         case SIOCSIFFLAGS:
 409                 /* fetch current flags to leave other flags untouched */
 410                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 411                         goto error;
 412                 if (set)
 413                         ifr->ifr_flags |= req_flags;
 414                 else
 415                         ifr->ifr_flags &= ~req_flags;
 416                 break;
 417         case SIOCGIFHWADDR:
 418         case SIOCSIFHWADDR:
 419         case SIOCSIFMTU:
 420                 break;
 421         default:
 422                 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
 423                         pmd->name);
 424                 return -EINVAL;
 425         }
 426         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 427                 goto error;
 428         return 0;
 429
 430 error:
 431         RTE_LOG(ERR, PMD, "%s: ioctl(%lu) failed with error: %s\n",
 432                 ifr->ifr_name, request, strerror(errno));
 433         return -errno;
 434 }
 435
 436 static int
 437 tap_link_set_down(struct rte_eth_dev *dev)
 438 {
 439         struct pmd_internals *pmd = dev->data->dev_private;
 440         struct ifreq ifr = { .ifr_flags = IFF_UP };
 441
 442         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 443         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0);
 444 }
 445
 446 static int
 447 tap_link_set_up(struct rte_eth_dev *dev)
 448 {
 449         struct pmd_internals *pmd = dev->data->dev_private;
 450         struct ifreq ifr = { .ifr_flags = IFF_UP };
 451
 452         dev->data->dev_link.link_status = ETH_LINK_UP;
 453         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1);
 454 }
 455
 456 static int
 457 tap_dev_start(struct rte_eth_dev *dev)
 458 {
 459         return tap_link_set_up(dev);
 460 }
 461
 462 /* This function gets called when the current port gets stopped.
 463  */
 464 static void
 465 tap_dev_stop(struct rte_eth_dev *dev)
 466 {
 467         tap_link_set_down(dev);
 468 }
 469
 470 static int
 471 tap_dev_configure(struct rte_eth_dev *dev __rte_unused)
 472 {
 473         return 0;
 474 }
 475
 476 static uint32_t
 477 tap_dev_speed_capa(void)
 478 {
 479         uint32_t speed = pmd_link.link_speed;
 480         uint32_t capa = 0;
 481
 482         if (speed >= ETH_SPEED_NUM_10M)
 483                 capa |= ETH_LINK_SPEED_10M;
 484         if (speed >= ETH_SPEED_NUM_100M)
 485                 capa |= ETH_LINK_SPEED_100M;
 486         if (speed >= ETH_SPEED_NUM_1G)
 487                 capa |= ETH_LINK_SPEED_1G;
 488         if (speed >= ETH_SPEED_NUM_5G)
 489                 capa |= ETH_LINK_SPEED_2_5G;
 490         if (speed >= ETH_SPEED_NUM_5G)
 491                 capa |= ETH_LINK_SPEED_5G;
 492         if (speed >= ETH_SPEED_NUM_10G)
 493                 capa |= ETH_LINK_SPEED_10G;
 494         if (speed >= ETH_SPEED_NUM_20G)
 495                 capa |= ETH_LINK_SPEED_20G;
 496         if (speed >= ETH_SPEED_NUM_25G)
 497                 capa |= ETH_LINK_SPEED_25G;
 498         if (speed >= ETH_SPEED_NUM_40G)
 499                 capa |= ETH_LINK_SPEED_40G;
 500         if (speed >= ETH_SPEED_NUM_50G)
 501                 capa |= ETH_LINK_SPEED_50G;
 502         if (speed >= ETH_SPEED_NUM_56G)
 503                 capa |= ETH_LINK_SPEED_56G;
 504         if (speed >= ETH_SPEED_NUM_100G)
 505                 capa |= ETH_LINK_SPEED_100G;
 506
 507         return capa;
 508 }
 509
 510 static void
 511 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 512 {
 513         struct pmd_internals *internals = dev->data->dev_private;
 514
 515         dev_info->if_index = internals->if_index;
 516         dev_info->max_mac_addrs = 1;
 517         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 518         dev_info->max_rx_queues = internals->nb_queues;
 519         dev_info->max_tx_queues = internals->nb_queues;
 520         dev_info->min_rx_bufsize = 0;
 521         dev_info->pci_dev = NULL;
 522         dev_info->speed_capa = tap_dev_speed_capa();
 523 }
 524
 525 static void
 526 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 527 {
 528         unsigned int i, imax;
 529         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 530         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 531         unsigned long rx_nombuf = 0, ierrors = 0;
 532         const struct pmd_internals *pmd = dev->data->dev_private;
 533
 534         imax = (pmd->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 535                 pmd->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 536
 537         for (i = 0; i < imax; i++) {
 538                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 539                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 540                 rx_total += tap_stats->q_ipackets[i];
 541                 rx_bytes_total += tap_stats->q_ibytes[i];
 542                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
 543                 ierrors += pmd->rxq[i].stats.ierrors;
 544
 545                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 546                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 547                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 548                 tx_total += tap_stats->q_opackets[i];
 549                 tx_err_total += tap_stats->q_errors[i];
 550                 tx_bytes_total += tap_stats->q_obytes[i];
 551         }
 552
 553         tap_stats->ipackets = rx_total;
 554         tap_stats->ibytes = rx_bytes_total;
 555         tap_stats->ierrors = ierrors;
 556         tap_stats->rx_nombuf = rx_nombuf;
 557         tap_stats->opackets = tx_total;
 558         tap_stats->oerrors = tx_err_total;
 559         tap_stats->obytes = tx_bytes_total;
 560 }
 561
 562 static void
 563 tap_stats_reset(struct rte_eth_dev *dev)
 564 {
 565         int i;
 566         struct pmd_internals *pmd = dev->data->dev_private;
 567
 568         for (i = 0; i < pmd->nb_queues; i++) {
 569                 pmd->rxq[i].stats.ipackets = 0;
 570                 pmd->rxq[i].stats.ibytes = 0;
 571                 pmd->rxq[i].stats.ierrors = 0;
 572                 pmd->rxq[i].stats.rx_nombuf = 0;
 573
 574                 pmd->txq[i].stats.opackets = 0;
 575                 pmd->txq[i].stats.errs = 0;
 576                 pmd->txq[i].stats.obytes = 0;
 577         }
 578 }
 579
 580 static void
 581 tap_dev_close(struct rte_eth_dev *dev __rte_unused)
 582 {
 583         int i;
 584         struct pmd_internals *internals = dev->data->dev_private;
 585
 586         tap_link_set_down(dev);
 587         tap_flow_flush(dev, NULL);
 588
 589         for (i = 0; i < internals->nb_queues; i++) {
 590                 if (internals->rxq[i].fd != -1)
 591                         close(internals->rxq[i].fd);
 592                 internals->rxq[i].fd = -1;
 593                 internals->txq[i].fd = -1;
 594         }
 595 }
 596
 597 static void
 598 tap_rx_queue_release(void *queue)
 599 {
 600         struct rx_queue *rxq = queue;
 601
 602         if (rxq && (rxq->fd > 0)) {
 603                 close(rxq->fd);
 604                 rxq->fd = -1;
 605                 rte_pktmbuf_free(rxq->pool);
 606                 rte_free(rxq->iovecs);
 607                 rxq->pool = NULL;
 608                 rxq->iovecs = NULL;
 609         }
 610 }
 611
 612 static void
 613 tap_tx_queue_release(void *queue)
 614 {
 615         struct tx_queue *txq = queue;
 616
 617         if (txq && (txq->fd > 0)) {
 618                 close(txq->fd);
 619                 txq->fd = -1;
 620         }
 621 }
 622
 623 static int
 624 tap_link_update(struct rte_eth_dev *dev __rte_unused,
 625                 int wait_to_complete __rte_unused)
 626 {
 627         return 0;
 628 }
 629
 630 static void
 631 tap_promisc_enable(struct rte_eth_dev *dev)
 632 {
 633         struct pmd_internals *pmd = dev->data->dev_private;
 634         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 635
 636         dev->data->promiscuous = 1;
 637         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1);
 638 }
 639
 640 static void
 641 tap_promisc_disable(struct rte_eth_dev *dev)
 642 {
 643         struct pmd_internals *pmd = dev->data->dev_private;
 644         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 645
 646         dev->data->promiscuous = 0;
 647         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0);
 648 }
 649
 650 static void
 651 tap_allmulti_enable(struct rte_eth_dev *dev)
 652 {
 653         struct pmd_internals *pmd = dev->data->dev_private;
 654         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 655
 656         dev->data->all_multicast = 1;
 657         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1);
 658 }
 659
 660 static void
 661 tap_allmulti_disable(struct rte_eth_dev *dev)
 662 {
 663         struct pmd_internals *pmd = dev->data->dev_private;
 664         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 665
 666         dev->data->all_multicast = 0;
 667         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0);
 668 }
 669
 670
 671 static void
 672 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 673 {
 674         struct pmd_internals *pmd = dev->data->dev_private;
 675         struct ifreq ifr;
 676
 677         if (is_zero_ether_addr(mac_addr)) {
 678                 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
 679                         dev->data->name);
 680                 return;
 681         }
 682
 683         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
 684         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
 685         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1) < 0)
 686                 return;
 687         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
 688 }
 689
 690 static int
 691 tap_setup_queue(struct rte_eth_dev *dev,
 692                 struct pmd_internals *internals,
 693                 uint16_t qid)
 694 {
 695         struct pmd_internals *pmd = dev->data->dev_private;
 696         struct rx_queue *rx = &internals->rxq[qid];
 697         struct tx_queue *tx = &internals->txq[qid];
 698         int fd;
 699
 700         fd = rx->fd;
 701         if (fd < 0) {
 702                 fd = tx->fd;
 703                 if (fd < 0) {
 704                         RTE_LOG(INFO, PMD, "Add queue to TAP %s for qid %d\n",
 705                                 pmd->name, qid);
 706                         fd = tun_alloc(pmd, qid);
 707                         if (fd < 0) {
 708                                 RTE_LOG(ERR, PMD, "tun_alloc(%s, %d) failed\n",
 709                                         pmd->name, qid);
 710                                 return -1;
 711                         }
 712                         if (qid == 0) {
 713                                 struct ifreq ifr;
 714
 715                                 ifr.ifr_mtu = dev->data->mtu;
 716                                 if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1) < 0) {
 717                                         close(fd);
 718                                         return -1;
 719                                 }
 720                         }
 721                 }
 722         }
 723
 724         rx->fd = fd;
 725         tx->fd = fd;
 726         tx->mtu = &dev->data->mtu;
 727         rx->rxmode = &dev->data->dev_conf.rxmode;
 728
 729         return fd;
 730 }
 731
 732 static int
 733 rx_setup_queue(struct rte_eth_dev *dev,
 734                 struct pmd_internals *internals,
 735                 uint16_t qid)
 736 {
 737         dev->data->rx_queues[qid] = &internals->rxq[qid];
 738
 739         return tap_setup_queue(dev, internals, qid);
 740 }
 741
 742 static int
 743 tx_setup_queue(struct rte_eth_dev *dev,
 744                 struct pmd_internals *internals,
 745                 uint16_t qid)
 746 {
 747         dev->data->tx_queues[qid] = &internals->txq[qid];
 748
 749         return tap_setup_queue(dev, internals, qid);
 750 }
 751
 752 static int
 753 tap_rx_queue_setup(struct rte_eth_dev *dev,
 754                    uint16_t rx_queue_id,
 755                    uint16_t nb_rx_desc,
 756                    unsigned int socket_id,
 757                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 758                    struct rte_mempool *mp)
 759 {
 760         struct pmd_internals *internals = dev->data->dev_private;
 761         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
 762         struct rte_mbuf **tmp = &rxq->pool;
 763         struct iovec (*iovecs)[nb_rx_desc + 1];
 764         int data_off = RTE_PKTMBUF_HEADROOM;
 765         uint16_t buf_size;
 766         int ret = 0;
 767         int fd;
 768         int i;
 769
 770         if ((rx_queue_id >= internals->nb_queues) || !mp) {
 771                 RTE_LOG(WARNING, PMD,
 772                         "nb_queues %d too small or mempool NULL\n",
 773                         internals->nb_queues);
 774                 return -1;
 775         }
 776
 777         rxq->mp = mp;
 778         rxq->trigger_seen = 1; /* force initial burst */
 779         rxq->in_port = dev->data->port_id;
 780         rxq->nb_rx_desc = nb_rx_desc;
 781         iovecs = rte_zmalloc_socket(dev->data->name, sizeof(*iovecs), 0,
 782                                     socket_id);
 783         if (!iovecs) {
 784                 RTE_LOG(WARNING, PMD,
 785                         "%s: Couldn't allocate %d RX descriptors\n",
 786                         dev->data->name, nb_rx_desc);
 787                 return -ENOMEM;
 788         }
 789         rxq->iovecs = iovecs;
 790
 791         /* Now get the space available for data in the mbuf */
 792         buf_size = (uint16_t)(rte_pktmbuf_data_room_size(mp) -
 793                                 RTE_PKTMBUF_HEADROOM);
 794
 795         if (buf_size < ETH_FRAME_LEN) {
 796                 RTE_LOG(WARNING, PMD,
 797                         "%s: %d bytes will not fit in mbuf (%d bytes)\n",
 798                         dev->data->name, ETH_FRAME_LEN, buf_size);
 799                 ret = -ENOMEM;
 800                 goto error;
 801         }
 802
 803         fd = rx_setup_queue(dev, internals, rx_queue_id);
 804         if (fd == -1) {
 805                 ret = fd;
 806                 goto error;
 807         }
 808
 809         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
 810         (*rxq->iovecs)[0].iov_base = &rxq->pi;
 811
 812         for (i = 1; i <= nb_rx_desc; i++) {
 813                 *tmp = rte_pktmbuf_alloc(rxq->mp);
 814                 if (!*tmp) {
 815                         RTE_LOG(WARNING, PMD,
 816                                 "%s: couldn't allocate memory for queue %d\n",
 817                                 dev->data->name, rx_queue_id);
 818                         ret = -ENOMEM;
 819                         goto error;
 820                 }
 821                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
 822                 (*rxq->iovecs)[i].iov_base =
 823                         (char *)(*tmp)->buf_addr + data_off;
 824                 data_off = 0;
 825                 tmp = &(*tmp)->next;
 826         }
 827
 828         RTE_LOG(DEBUG, PMD, "  RX TAP device name %s, qid %d on fd %d\n",
 829                 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
 830
 831         return 0;
 832
 833 error:
 834         rte_pktmbuf_free(rxq->pool);
 835         rxq->pool = NULL;
 836         rte_free(rxq->iovecs);
 837         rxq->iovecs = NULL;
 838         return ret;
 839 }
 840
 841 static int
 842 tap_tx_queue_setup(struct rte_eth_dev *dev,
 843                    uint16_t tx_queue_id,
 844                    uint16_t nb_tx_desc __rte_unused,
 845                    unsigned int socket_id __rte_unused,
 846                    const struct rte_eth_txconf *tx_conf __rte_unused)
 847 {
 848         struct pmd_internals *internals = dev->data->dev_private;
 849         int ret;
 850
 851         if (tx_queue_id >= internals->nb_queues)
 852                 return -1;
 853
 854         ret = tx_setup_queue(dev, internals, tx_queue_id);
 855         if (ret == -1)
 856                 return -1;
 857
 858         RTE_LOG(DEBUG, PMD, "  TX TAP device name %s, qid %d on fd %d\n",
 859                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd);
 860
 861         return 0;
 862 }
 863
 864 static int
 865 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 866 {
 867         struct pmd_internals *pmd = dev->data->dev_private;
 868         struct ifreq ifr = { .ifr_mtu = mtu };
 869         int err = 0;
 870
 871         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1);
 872         if (!err)
 873                 dev->data->mtu = mtu;
 874
 875         return err;
 876 }
 877
 878 static int
 879 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
 880                      struct ether_addr *mc_addr_set __rte_unused,
 881                      uint32_t nb_mc_addr __rte_unused)
 882 {
 883         /*
 884          * Nothing to do actually: the tap has no filtering whatsoever, every
 885          * packet is received.
 886          */
 887         return 0;
 888 }
 889
 890 static const uint32_t*
 891 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
 892 {
 893         static const uint32_t ptypes[] = {
 894                 RTE_PTYPE_INNER_L2_ETHER,
 895                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
 896                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
 897                 RTE_PTYPE_INNER_L3_IPV4,
 898                 RTE_PTYPE_INNER_L3_IPV4_EXT,
 899                 RTE_PTYPE_INNER_L3_IPV6,
 900                 RTE_PTYPE_INNER_L3_IPV6_EXT,
 901                 RTE_PTYPE_INNER_L4_FRAG,
 902                 RTE_PTYPE_INNER_L4_UDP,
 903                 RTE_PTYPE_INNER_L4_TCP,
 904                 RTE_PTYPE_INNER_L4_SCTP,
 905                 RTE_PTYPE_L2_ETHER,
 906                 RTE_PTYPE_L2_ETHER_VLAN,
 907                 RTE_PTYPE_L2_ETHER_QINQ,
 908                 RTE_PTYPE_L3_IPV4,
 909                 RTE_PTYPE_L3_IPV4_EXT,
 910                 RTE_PTYPE_L3_IPV6_EXT,
 911                 RTE_PTYPE_L3_IPV6,
 912                 RTE_PTYPE_L4_FRAG,
 913                 RTE_PTYPE_L4_UDP,
 914                 RTE_PTYPE_L4_TCP,
 915                 RTE_PTYPE_L4_SCTP,
 916         };
 917
 918         return ptypes;
 919 }
 920
 921 static int
 922 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
 923                   struct rte_eth_fc_conf *fc_conf)
 924 {
 925         fc_conf->mode = RTE_FC_NONE;
 926         return 0;
 927 }
 928
 929 static int
 930 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
 931                   struct rte_eth_fc_conf *fc_conf)
 932 {
 933         if (fc_conf->mode != RTE_FC_NONE)
 934                 return -ENOTSUP;
 935         return 0;
 936 }
 937
 938 static const struct eth_dev_ops ops = {
 939         .dev_start              = tap_dev_start,
 940         .dev_stop               = tap_dev_stop,
 941         .dev_close              = tap_dev_close,
 942         .dev_configure          = tap_dev_configure,
 943         .dev_infos_get          = tap_dev_info,
 944         .rx_queue_setup         = tap_rx_queue_setup,
 945         .tx_queue_setup         = tap_tx_queue_setup,
 946         .rx_queue_release       = tap_rx_queue_release,
 947         .tx_queue_release       = tap_tx_queue_release,
 948         .flow_ctrl_get          = tap_flow_ctrl_get,
 949         .flow_ctrl_set          = tap_flow_ctrl_set,
 950         .link_update            = tap_link_update,
 951         .dev_set_link_up        = tap_link_set_up,
 952         .dev_set_link_down      = tap_link_set_down,
 953         .promiscuous_enable     = tap_promisc_enable,
 954         .promiscuous_disable    = tap_promisc_disable,
 955         .allmulticast_enable    = tap_allmulti_enable,
 956         .allmulticast_disable   = tap_allmulti_disable,
 957         .mac_addr_set           = tap_mac_set,
 958         .mtu_set                = tap_mtu_set,
 959         .set_mc_addr_list       = tap_set_mc_addr_list,
 960         .stats_get              = tap_stats_get,
 961         .stats_reset            = tap_stats_reset,
 962         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
 963         .filter_ctrl            = tap_dev_filter_ctrl,
 964 };
 965
 966 static int
 967 tap_kernel_support(struct pmd_internals *pmd)
 968 {
 969         struct utsname utsname;
 970         int ver[3];
 971
 972         if (uname(&utsname) == -1 ||
 973             sscanf(utsname.release, "%d.%d.%d",
 974                    &ver[0], &ver[1], &ver[2]) != 3)
 975                 return 0;
 976         if (KERNEL_VERSION(ver[0], ver[1], ver[2]) >= FLOWER_KERNEL_VERSION)
 977                 pmd->flower_support = 1;
 978         if (KERNEL_VERSION(ver[0], ver[1], ver[2]) >=
 979             FLOWER_VLAN_KERNEL_VERSION)
 980                 pmd->flower_vlan_support = 1;
 981         return 1;
 982 }
 983
 984 static int
 985 eth_dev_tap_create(const char *name, char *tap_name)
 986 {
 987         int numa_node = rte_socket_id();
 988         struct rte_eth_dev *dev = NULL;
 989         struct pmd_internals *pmd = NULL;
 990         struct rte_eth_dev_data *data = NULL;
 991         int i;
 992
 993         RTE_LOG(DEBUG, PMD, "  TAP device on numa %u\n", rte_socket_id());
 994
 995         data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
 996         if (!data) {
 997                 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
 998                 goto error_exit;
 999         }
1000
1001         pmd = rte_zmalloc_socket(tap_name, sizeof(*pmd), 0, numa_node);
1002         if (!pmd) {
1003                 RTE_LOG(ERR, PMD, "TAP Unable to allocate internal struct\n");
1004                 goto error_exit;
1005         }
1006
1007         /* name in allocation and data->name must be consistent */
1008         snprintf(data->name, sizeof(data->name), "%s", name);
1009         dev = rte_eth_dev_allocate(name);
1010         if (!dev) {
1011                 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
1012                 goto error_exit;
1013         }
1014
1015         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1016
1017         pmd->nb_queues = RTE_PMD_TAP_MAX_QUEUES;
1018
1019         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1020         if (pmd->ioctl_sock == -1) {
1021                 RTE_LOG(ERR, PMD,
1022                         "TAP Unable to get a socket for management: %s\n",
1023                         strerror(errno));
1024                 goto error_exit;
1025         }
1026
1027         /* Setup some default values */
1028         data->dev_private = pmd;
1029         data->port_id = dev->data->port_id;
1030         data->mtu = dev->data->mtu;
1031         data->dev_flags = RTE_ETH_DEV_DETACHABLE;
1032         data->kdrv = RTE_KDRV_NONE;
1033         data->drv_name = pmd_tap_drv.driver.name;
1034         data->numa_node = numa_node;
1035
1036         data->dev_link = pmd_link;
1037         data->mac_addrs = &pmd->eth_addr;
1038         data->nb_rx_queues = pmd->nb_queues;
1039         data->nb_tx_queues = pmd->nb_queues;
1040
1041         dev->data = data;
1042         dev->dev_ops = &ops;
1043         dev->driver = NULL;
1044         dev->rx_pkt_burst = pmd_rx_burst;
1045         dev->tx_pkt_burst = pmd_tx_burst;
1046
1047         /* Presetup the fds to -1 as being not valid */
1048         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1049                 pmd->rxq[i].fd = -1;
1050                 pmd->txq[i].fd = -1;
1051         }
1052
1053         tap_kernel_support(pmd);
1054         if (!pmd->flower_support)
1055                 return 0;
1056         LIST_INIT(&pmd->flows);
1057         /*
1058          * If no netlink socket can be created, then it will fail when
1059          * creating/destroying flow rules.
1060          */
1061         pmd->nlsk_fd = nl_init();
1062
1063         return 0;
1064
1065 error_exit:
1066         RTE_LOG(DEBUG, PMD, "TAP Unable to initialize %s\n", name);
1067
1068         rte_free(data);
1069         rte_free(pmd);
1070
1071         rte_eth_dev_release_port(dev);
1072
1073         return -EINVAL;
1074 }
1075
1076 static int
1077 set_interface_name(const char *key __rte_unused,
1078                    const char *value,
1079                    void *extra_args)
1080 {
1081         char *name = (char *)extra_args;
1082
1083         if (value)
1084                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
1085         else
1086                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1087                          DEFAULT_TAP_NAME, (tap_unit - 1));
1088
1089         return 0;
1090 }
1091
1092 static int
1093 set_interface_speed(const char *key __rte_unused,
1094                     const char *value,
1095                     void *extra_args)
1096 {
1097         *(int *)extra_args = (value) ? atoi(value) : ETH_SPEED_NUM_10G;
1098
1099         return 0;
1100 }
1101
1102 /* Open a TAP interface device.
1103  */
1104 static int
1105 rte_pmd_tap_probe(const char *name, const char *params)
1106 {
1107         int ret;
1108         struct rte_kvargs *kvlist = NULL;
1109         int speed;
1110         char tap_name[RTE_ETH_NAME_MAX_LEN];
1111
1112         speed = ETH_SPEED_NUM_10G;
1113         snprintf(tap_name, sizeof(tap_name), "%s%d",
1114                  DEFAULT_TAP_NAME, tap_unit++);
1115
1116         if (params && (params[0] != '\0')) {
1117                 RTE_LOG(DEBUG, PMD, "paramaters (%s)\n", params);
1118
1119                 kvlist = rte_kvargs_parse(params, valid_arguments);
1120                 if (kvlist) {
1121                         if (rte_kvargs_count(kvlist, ETH_TAP_SPEED_ARG) == 1) {
1122                                 ret = rte_kvargs_process(kvlist,
1123                                                          ETH_TAP_SPEED_ARG,
1124                                                          &set_interface_speed,
1125                                                          &speed);
1126                                 if (ret == -1)
1127                                         goto leave;
1128                         }
1129
1130                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1131                                 ret = rte_kvargs_process(kvlist,
1132                                                          ETH_TAP_IFACE_ARG,
1133                                                          &set_interface_name,
1134                                                          tap_name);
1135                                 if (ret == -1)
1136                                         goto leave;
1137                         }
1138                 }
1139         }
1140         pmd_link.link_speed = speed;
1141
1142         RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1143                 name, tap_name);
1144
1145         ret = eth_dev_tap_create(name, tap_name);
1146
1147 leave:
1148         if (ret == -1) {
1149                 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1150                         name, tap_name);
1151                 tap_unit--;             /* Restore the unit number */
1152         }
1153         rte_kvargs_free(kvlist);
1154
1155         return ret;
1156 }
1157
1158 /* detach a TAP device.
1159  */
1160 static int
1161 rte_pmd_tap_remove(const char *name)
1162 {
1163         struct rte_eth_dev *eth_dev = NULL;
1164         struct pmd_internals *internals;
1165         int i;
1166
1167         RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1168                 rte_socket_id());
1169
1170         /* find the ethdev entry */
1171         eth_dev = rte_eth_dev_allocated(name);
1172         if (!eth_dev)
1173                 return 0;
1174
1175         internals = eth_dev->data->dev_private;
1176         if (internals->flower_support && internals->nlsk_fd) {
1177                 tap_flow_flush(eth_dev, NULL);
1178                 nl_final(internals->nlsk_fd);
1179         }
1180         for (i = 0; i < internals->nb_queues; i++)
1181                 if (internals->rxq[i].fd != -1)
1182                         close(internals->rxq[i].fd);
1183
1184         close(internals->ioctl_sock);
1185         rte_free(eth_dev->data->dev_private);
1186         rte_free(eth_dev->data);
1187
1188         rte_eth_dev_release_port(eth_dev);
1189
1190         return 0;
1191 }
1192
1193 static struct rte_vdev_driver pmd_tap_drv = {
1194         .probe = rte_pmd_tap_probe,
1195         .remove = rte_pmd_tap_remove,
1196 };
1197 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1198 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1199 RTE_PMD_REGISTER_PARAM_STRING(net_tap, "iface=<string>,speed=N");