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/mman.h>
  48 #include <errno.h>
  49 #include <signal.h>
  50 #include <stdint.h>
  51 #include <unistd.h>
  52 #include <arpa/inet.h>
  53 #include <linux/if.h>
  54 #include <linux/if_tun.h>
  55 #include <linux/if_ether.h>
  56 #include <fcntl.h>
  57
  58 /* Linux based path to the TUN device */
  59 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  60 #define DEFAULT_TAP_NAME        "dtap"
  61
  62 #define ETH_TAP_IFACE_ARG       "iface"
  63 #define ETH_TAP_SPEED_ARG       "speed"
  64
  65 #ifdef IFF_MULTI_QUEUE
  66 #define RTE_PMD_TAP_MAX_QUEUES  16
  67 #else
  68 #define RTE_PMD_TAP_MAX_QUEUES  1
  69 #endif
  70
  71 static struct rte_vdev_driver pmd_tap_drv;
  72
  73 static const char *valid_arguments[] = {
  74         ETH_TAP_IFACE_ARG,
  75         ETH_TAP_SPEED_ARG,
  76         NULL
  77 };
  78
  79 static int tap_unit;
  80
  81 static volatile uint32_t tap_trigger;   /* Rx trigger */
  82
  83 static struct rte_eth_link pmd_link = {
  84         .link_speed = ETH_SPEED_NUM_10G,
  85         .link_duplex = ETH_LINK_FULL_DUPLEX,
  86         .link_status = ETH_LINK_DOWN,
  87         .link_autoneg = ETH_LINK_SPEED_AUTONEG
  88 };
  89
  90 struct pkt_stats {
  91         uint64_t opackets;              /* Number of output packets */
  92         uint64_t ipackets;              /* Number of input packets */
  93         uint64_t obytes;                /* Number of bytes on output */
  94         uint64_t ibytes;                /* Number of bytes on input */
  95         uint64_t errs;                  /* Number of error packets */
  96 };
  97
  98 struct rx_queue {
  99         struct rte_mempool *mp;         /* Mempool for RX packets */
 100         uint32_t trigger_seen;          /* Last seen Rx trigger value */
 101         uint16_t in_port;               /* Port ID */
 102         int fd;
 103
 104         struct pkt_stats stats;         /* Stats for this RX queue */
 105 };
 106
 107 struct tx_queue {
 108         int fd;
 109         struct pkt_stats stats;         /* Stats for this TX queue */
 110 };
 111
 112 struct pmd_internals {
 113         char name[RTE_ETH_NAME_MAX_LEN];        /* Internal Tap device name */
 114         uint16_t nb_queues;             /* Number of queues supported */
 115         struct ether_addr eth_addr;     /* Mac address of the device port */
 116
 117         int if_index;                   /* IF_INDEX for the port */
 118         int ioctl_sock;                 /* socket for ioctl calls */
 119
 120         struct rx_queue rxq[RTE_PMD_TAP_MAX_QUEUES];    /* List of RX queues */
 121         struct tx_queue txq[RTE_PMD_TAP_MAX_QUEUES];    /* List of TX queues */
 122 };
 123
 124 static void
 125 tap_trigger_cb(int sig __rte_unused)
 126 {
 127         /* Valid trigger values are nonzero */
 128         tap_trigger = (tap_trigger + 1) | 0x80000000;
 129 }
 130
 131 static int
 132 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 133           struct ifreq *ifr, int set);
 134
 135 /* Tun/Tap allocation routine
 136  *
 137  * name is the number of the interface to use, unless NULL to take the host
 138  * supplied name.
 139  */
 140 static int
 141 tun_alloc(struct pmd_internals *pmd, uint16_t qid)
 142 {
 143         struct ifreq ifr;
 144 #ifdef IFF_MULTI_QUEUE
 145         unsigned int features;
 146 #endif
 147         int fd;
 148
 149         memset(&ifr, 0, sizeof(struct ifreq));
 150
 151         ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
 152         snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
 153
 154         RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
 155
 156         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 157         if (fd < 0) {
 158                 RTE_LOG(ERR, PMD, "Unable to create TAP interface");
 159                 goto error;
 160         }
 161
 162 #ifdef IFF_MULTI_QUEUE
 163         /* Grab the TUN features to verify we can work multi-queue */
 164         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 165                 RTE_LOG(ERR, PMD, "TAP unable to get TUN/TAP features\n");
 166                 goto error;
 167         }
 168         RTE_LOG(DEBUG, PMD, "  TAP Features %08x\n", features);
 169
 170         if (features & IFF_MULTI_QUEUE) {
 171                 RTE_LOG(DEBUG, PMD, "  Multi-queue support for %d queues\n",
 172                         RTE_PMD_TAP_MAX_QUEUES);
 173                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 174         } else
 175 #endif
 176         {
 177                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 178                 RTE_LOG(DEBUG, PMD, "  Single queue only support\n");
 179         }
 180
 181         /* Set the TUN/TAP configuration and set the name if needed */
 182         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 183                 RTE_LOG(WARNING, PMD,
 184                         "Unable to set TUNSETIFF for %s\n",
 185                         ifr.ifr_name);
 186                 perror("TUNSETIFF");
 187                 goto error;
 188         }
 189
 190         /* Always set the file descriptor to non-blocking */
 191         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 192                 RTE_LOG(WARNING, PMD,
 193                         "Unable to set %s to nonblocking\n",
 194                         ifr.ifr_name);
 195                 perror("F_SETFL, NONBLOCK");
 196                 goto error;
 197         }
 198
 199         /* Set up trigger to optimize empty Rx bursts */
 200         errno = 0;
 201         do {
 202                 struct sigaction sa;
 203                 int flags = fcntl(fd, F_GETFL);
 204
 205                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 206                         break;
 207                 if (sa.sa_handler != tap_trigger_cb) {
 208                         /*
 209                          * Make sure SIGIO is not already taken. This is done
 210                          * as late as possible to leave the application a
 211                          * chance to set up its own signal handler first.
 212                          */
 213                         if (sa.sa_handler != SIG_IGN &&
 214                             sa.sa_handler != SIG_DFL) {
 215                                 errno = EBUSY;
 216                                 break;
 217                         }
 218                         sa = (struct sigaction){
 219                                 .sa_flags = SA_RESTART,
 220                                 .sa_handler = tap_trigger_cb,
 221                         };
 222                         if (sigaction(SIGIO, &sa, NULL) == -1)
 223                                 break;
 224                 }
 225                 /* Enable SIGIO on file descriptor */
 226                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 227                 fcntl(fd, F_SETOWN, getpid());
 228         } while (0);
 229         if (errno) {
 230                 /* Disable trigger globally in case of error */
 231                 tap_trigger = 0;
 232                 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
 233                         strerror(errno));
 234         }
 235
 236         if (qid == 0) {
 237                 struct ifreq ifr;
 238
 239                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0) < 0)
 240                         goto error;
 241                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
 242                            ETHER_ADDR_LEN);
 243         }
 244
 245         return fd;
 246
 247 error:
 248         if (fd > 0)
 249                 close(fd);
 250         return -1;
 251 }
 252
 253 /* Callback to handle the rx burst of packets to the correct interface and
 254  * file descriptor(s) in a multi-queue setup.
 255  */
 256 static uint16_t
 257 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 258 {
 259         int len;
 260         struct rte_mbuf *mbuf;
 261         struct rx_queue *rxq = queue;
 262         uint16_t num_rx;
 263         unsigned long num_rx_bytes = 0;
 264         uint32_t trigger = tap_trigger;
 265
 266         if (trigger == rxq->trigger_seen)
 267                 return 0;
 268         if (trigger)
 269                 rxq->trigger_seen = trigger;
 270         rte_compiler_barrier();
 271         for (num_rx = 0; num_rx < nb_pkts; ) {
 272                 /* allocate the next mbuf */
 273                 mbuf = rte_pktmbuf_alloc(rxq->mp);
 274                 if (unlikely(!mbuf)) {
 275                         RTE_LOG(WARNING, PMD, "TAP unable to allocate mbuf\n");
 276                         break;
 277                 }
 278
 279                 len = read(rxq->fd, rte_pktmbuf_mtod(mbuf, char *),
 280                            rte_pktmbuf_tailroom(mbuf));
 281                 if (len <= 0) {
 282                         rte_pktmbuf_free(mbuf);
 283                         break;
 284                 }
 285
 286                 mbuf->data_len = len;
 287                 mbuf->pkt_len = len;
 288                 mbuf->port = rxq->in_port;
 289                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 290                                                       RTE_PTYPE_ALL_MASK);
 291
 292                 /* account for the receive frame */
 293                 bufs[num_rx++] = mbuf;
 294                 num_rx_bytes += mbuf->pkt_len;
 295         }
 296         rxq->stats.ipackets += num_rx;
 297         rxq->stats.ibytes += num_rx_bytes;
 298
 299         return num_rx;
 300 }
 301
 302 /* Callback to handle sending packets from the tap interface
 303  */
 304 static uint16_t
 305 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 306 {
 307         struct rte_mbuf *mbuf;
 308         struct tx_queue *txq = queue;
 309         uint16_t num_tx = 0;
 310         unsigned long num_tx_bytes = 0;
 311         int i, n;
 312
 313         if (unlikely(nb_pkts == 0))
 314                 return 0;
 315
 316         for (i = 0; i < nb_pkts; i++) {
 317                 /* copy the tx frame data */
 318                 mbuf = bufs[num_tx];
 319                 n = write(txq->fd,
 320                           rte_pktmbuf_mtod(mbuf, void *),
 321                           rte_pktmbuf_pkt_len(mbuf));
 322                 if (n <= 0)
 323                         break;
 324
 325                 num_tx++;
 326                 num_tx_bytes += mbuf->pkt_len;
 327                 rte_pktmbuf_free(mbuf);
 328         }
 329
 330         txq->stats.opackets += num_tx;
 331         txq->stats.errs += nb_pkts - num_tx;
 332         txq->stats.obytes += num_tx_bytes;
 333
 334         return num_tx;
 335 }
 336
 337 static int
 338 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 339           struct ifreq *ifr, int set)
 340 {
 341         short req_flags = ifr->ifr_flags;
 342
 343         snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 344         switch (request) {
 345         case SIOCSIFFLAGS:
 346                 /* fetch current flags to leave other flags untouched */
 347                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 348                         goto error;
 349                 if (set)
 350                         ifr->ifr_flags |= req_flags;
 351                 else
 352                         ifr->ifr_flags &= ~req_flags;
 353                 break;
 354         case SIOCGIFHWADDR:
 355         case SIOCSIFHWADDR:
 356         case SIOCSIFMTU:
 357                 break;
 358         default:
 359                 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
 360                         pmd->name);
 361                 return -EINVAL;
 362         }
 363         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 364                 goto error;
 365         return 0;
 366
 367 error:
 368         RTE_LOG(ERR, PMD, "%s: ioctl(%lu) failed with error: %s\n",
 369                 ifr->ifr_name, request, strerror(errno));
 370         return -errno;
 371 }
 372
 373 static int
 374 tap_link_set_down(struct rte_eth_dev *dev)
 375 {
 376         struct pmd_internals *pmd = dev->data->dev_private;
 377         struct ifreq ifr = { .ifr_flags = IFF_UP };
 378
 379         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 380         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0);
 381 }
 382
 383 static int
 384 tap_link_set_up(struct rte_eth_dev *dev)
 385 {
 386         struct pmd_internals *pmd = dev->data->dev_private;
 387         struct ifreq ifr = { .ifr_flags = IFF_UP };
 388
 389         dev->data->dev_link.link_status = ETH_LINK_UP;
 390         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1);
 391 }
 392
 393 static int
 394 tap_dev_start(struct rte_eth_dev *dev)
 395 {
 396         return tap_link_set_up(dev);
 397 }
 398
 399 /* This function gets called when the current port gets stopped.
 400  */
 401 static void
 402 tap_dev_stop(struct rte_eth_dev *dev)
 403 {
 404         tap_link_set_down(dev);
 405 }
 406
 407 static int
 408 tap_dev_configure(struct rte_eth_dev *dev __rte_unused)
 409 {
 410         return 0;
 411 }
 412
 413 static uint32_t
 414 tap_dev_speed_capa(void)
 415 {
 416         uint32_t speed = pmd_link.link_speed;
 417         uint32_t capa = 0;
 418
 419         if (speed >= ETH_SPEED_NUM_10M)
 420                 capa |= ETH_LINK_SPEED_10M;
 421         if (speed >= ETH_SPEED_NUM_100M)
 422                 capa |= ETH_LINK_SPEED_100M;
 423         if (speed >= ETH_SPEED_NUM_1G)
 424                 capa |= ETH_LINK_SPEED_1G;
 425         if (speed >= ETH_SPEED_NUM_5G)
 426                 capa |= ETH_LINK_SPEED_2_5G;
 427         if (speed >= ETH_SPEED_NUM_5G)
 428                 capa |= ETH_LINK_SPEED_5G;
 429         if (speed >= ETH_SPEED_NUM_10G)
 430                 capa |= ETH_LINK_SPEED_10G;
 431         if (speed >= ETH_SPEED_NUM_20G)
 432                 capa |= ETH_LINK_SPEED_20G;
 433         if (speed >= ETH_SPEED_NUM_25G)
 434                 capa |= ETH_LINK_SPEED_25G;
 435         if (speed >= ETH_SPEED_NUM_40G)
 436                 capa |= ETH_LINK_SPEED_40G;
 437         if (speed >= ETH_SPEED_NUM_50G)
 438                 capa |= ETH_LINK_SPEED_50G;
 439         if (speed >= ETH_SPEED_NUM_56G)
 440                 capa |= ETH_LINK_SPEED_56G;
 441         if (speed >= ETH_SPEED_NUM_100G)
 442                 capa |= ETH_LINK_SPEED_100G;
 443
 444         return capa;
 445 }
 446
 447 static void
 448 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 449 {
 450         struct pmd_internals *internals = dev->data->dev_private;
 451
 452         dev_info->if_index = internals->if_index;
 453         dev_info->max_mac_addrs = 1;
 454         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 455         dev_info->max_rx_queues = internals->nb_queues;
 456         dev_info->max_tx_queues = internals->nb_queues;
 457         dev_info->min_rx_bufsize = 0;
 458         dev_info->pci_dev = NULL;
 459         dev_info->speed_capa = tap_dev_speed_capa();
 460 }
 461
 462 static void
 463 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 464 {
 465         unsigned int i, imax;
 466         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 467         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 468         const struct pmd_internals *pmd = dev->data->dev_private;
 469
 470         imax = (pmd->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 471                 pmd->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 472
 473         for (i = 0; i < imax; i++) {
 474                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 475                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 476                 rx_total += tap_stats->q_ipackets[i];
 477                 rx_bytes_total += tap_stats->q_ibytes[i];
 478
 479                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 480                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 481                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 482                 tx_total += tap_stats->q_opackets[i];
 483                 tx_err_total += tap_stats->q_errors[i];
 484                 tx_bytes_total += tap_stats->q_obytes[i];
 485         }
 486
 487         tap_stats->ipackets = rx_total;
 488         tap_stats->ibytes = rx_bytes_total;
 489         tap_stats->opackets = tx_total;
 490         tap_stats->oerrors = tx_err_total;
 491         tap_stats->obytes = tx_bytes_total;
 492 }
 493
 494 static void
 495 tap_stats_reset(struct rte_eth_dev *dev)
 496 {
 497         int i;
 498         struct pmd_internals *pmd = dev->data->dev_private;
 499
 500         for (i = 0; i < pmd->nb_queues; i++) {
 501                 pmd->rxq[i].stats.ipackets = 0;
 502                 pmd->rxq[i].stats.ibytes = 0;
 503
 504                 pmd->txq[i].stats.opackets = 0;
 505                 pmd->txq[i].stats.errs = 0;
 506                 pmd->txq[i].stats.obytes = 0;
 507         }
 508 }
 509
 510 static void
 511 tap_dev_close(struct rte_eth_dev *dev __rte_unused)
 512 {
 513         int i;
 514         struct pmd_internals *internals = dev->data->dev_private;
 515
 516         tap_link_set_down(dev);
 517
 518         for (i = 0; i < internals->nb_queues; i++) {
 519                 if (internals->rxq[i].fd != -1)
 520                         close(internals->rxq[i].fd);
 521                 internals->rxq[i].fd = -1;
 522                 internals->txq[i].fd = -1;
 523         }
 524 }
 525
 526 static void
 527 tap_rx_queue_release(void *queue)
 528 {
 529         struct rx_queue *rxq = queue;
 530
 531         if (rxq && (rxq->fd > 0)) {
 532                 close(rxq->fd);
 533                 rxq->fd = -1;
 534         }
 535 }
 536
 537 static void
 538 tap_tx_queue_release(void *queue)
 539 {
 540         struct tx_queue *txq = queue;
 541
 542         if (txq && (txq->fd > 0)) {
 543                 close(txq->fd);
 544                 txq->fd = -1;
 545         }
 546 }
 547
 548 static int
 549 tap_link_update(struct rte_eth_dev *dev __rte_unused,
 550                 int wait_to_complete __rte_unused)
 551 {
 552         return 0;
 553 }
 554
 555 static void
 556 tap_promisc_enable(struct rte_eth_dev *dev)
 557 {
 558         struct pmd_internals *pmd = dev->data->dev_private;
 559         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 560
 561         dev->data->promiscuous = 1;
 562         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1);
 563 }
 564
 565 static void
 566 tap_promisc_disable(struct rte_eth_dev *dev)
 567 {
 568         struct pmd_internals *pmd = dev->data->dev_private;
 569         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 570
 571         dev->data->promiscuous = 0;
 572         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0);
 573 }
 574
 575 static void
 576 tap_allmulti_enable(struct rte_eth_dev *dev)
 577 {
 578         struct pmd_internals *pmd = dev->data->dev_private;
 579         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 580
 581         dev->data->all_multicast = 1;
 582         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1);
 583 }
 584
 585 static void
 586 tap_allmulti_disable(struct rte_eth_dev *dev)
 587 {
 588         struct pmd_internals *pmd = dev->data->dev_private;
 589         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 590
 591         dev->data->all_multicast = 0;
 592         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0);
 593 }
 594
 595
 596 static void
 597 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 598 {
 599         struct pmd_internals *pmd = dev->data->dev_private;
 600         struct ifreq ifr;
 601
 602         if (is_zero_ether_addr(mac_addr)) {
 603                 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
 604                         dev->data->name);
 605                 return;
 606         }
 607
 608         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
 609         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
 610         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1) < 0)
 611                 return;
 612         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
 613 }
 614
 615 static int
 616 tap_setup_queue(struct rte_eth_dev *dev,
 617                 struct pmd_internals *internals,
 618                 uint16_t qid)
 619 {
 620         struct pmd_internals *pmd = dev->data->dev_private;
 621         struct rx_queue *rx = &internals->rxq[qid];
 622         struct tx_queue *tx = &internals->txq[qid];
 623         int fd;
 624
 625         fd = rx->fd;
 626         if (fd < 0) {
 627                 fd = tx->fd;
 628                 if (fd < 0) {
 629                         RTE_LOG(INFO, PMD, "Add queue to TAP %s for qid %d\n",
 630                                 pmd->name, qid);
 631                         fd = tun_alloc(pmd, qid);
 632                         if (fd < 0) {
 633                                 RTE_LOG(ERR, PMD, "tun_alloc(%s, %d) failed\n",
 634                                         pmd->name, qid);
 635                                 return -1;
 636                         }
 637                         if (qid == 0) {
 638                                 struct ifreq ifr;
 639
 640                                 ifr.ifr_mtu = dev->data->mtu;
 641                                 if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1) < 0) {
 642                                         close(fd);
 643                                         return -1;
 644                                 }
 645                         }
 646                 }
 647         }
 648
 649         rx->fd = fd;
 650         tx->fd = fd;
 651
 652         return fd;
 653 }
 654
 655 static int
 656 rx_setup_queue(struct rte_eth_dev *dev,
 657                 struct pmd_internals *internals,
 658                 uint16_t qid)
 659 {
 660         dev->data->rx_queues[qid] = &internals->rxq[qid];
 661
 662         return tap_setup_queue(dev, internals, qid);
 663 }
 664
 665 static int
 666 tx_setup_queue(struct rte_eth_dev *dev,
 667                 struct pmd_internals *internals,
 668                 uint16_t qid)
 669 {
 670         dev->data->tx_queues[qid] = &internals->txq[qid];
 671
 672         return tap_setup_queue(dev, internals, qid);
 673 }
 674
 675 static int
 676 tap_rx_queue_setup(struct rte_eth_dev *dev,
 677                    uint16_t rx_queue_id,
 678                    uint16_t nb_rx_desc __rte_unused,
 679                    unsigned int socket_id __rte_unused,
 680                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 681                    struct rte_mempool *mp)
 682 {
 683         struct pmd_internals *internals = dev->data->dev_private;
 684         uint16_t buf_size;
 685         int fd;
 686
 687         if ((rx_queue_id >= internals->nb_queues) || !mp) {
 688                 RTE_LOG(WARNING, PMD,
 689                         "nb_queues %d too small or mempool NULL\n",
 690                         internals->nb_queues);
 691                 return -1;
 692         }
 693
 694         internals->rxq[rx_queue_id].mp = mp;
 695         internals->rxq[rx_queue_id].trigger_seen = 1; /* force initial burst */
 696         internals->rxq[rx_queue_id].in_port = dev->data->port_id;
 697
 698         /* Now get the space available for data in the mbuf */
 699         buf_size = (uint16_t)(rte_pktmbuf_data_room_size(mp) -
 700                                 RTE_PKTMBUF_HEADROOM);
 701
 702         if (buf_size < ETH_FRAME_LEN) {
 703                 RTE_LOG(WARNING, PMD,
 704                         "%s: %d bytes will not fit in mbuf (%d bytes)\n",
 705                         dev->data->name, ETH_FRAME_LEN, buf_size);
 706                 return -ENOMEM;
 707         }
 708
 709         fd = rx_setup_queue(dev, internals, rx_queue_id);
 710         if (fd == -1)
 711                 return -1;
 712
 713         RTE_LOG(DEBUG, PMD, "  RX TAP device name %s, qid %d on fd %d\n",
 714                 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
 715
 716         return 0;
 717 }
 718
 719 static int
 720 tap_tx_queue_setup(struct rte_eth_dev *dev,
 721                    uint16_t tx_queue_id,
 722                    uint16_t nb_tx_desc __rte_unused,
 723                    unsigned int socket_id __rte_unused,
 724                    const struct rte_eth_txconf *tx_conf __rte_unused)
 725 {
 726         struct pmd_internals *internals = dev->data->dev_private;
 727         int ret;
 728
 729         if (tx_queue_id >= internals->nb_queues)
 730                 return -1;
 731
 732         ret = tx_setup_queue(dev, internals, tx_queue_id);
 733         if (ret == -1)
 734                 return -1;
 735
 736         RTE_LOG(DEBUG, PMD, "  TX TAP device name %s, qid %d on fd %d\n",
 737                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd);
 738
 739         return 0;
 740 }
 741
 742 static int
 743 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 744 {
 745         struct pmd_internals *pmd = dev->data->dev_private;
 746         struct ifreq ifr = { .ifr_mtu = mtu };
 747         int err = 0;
 748
 749         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1);
 750         if (!err)
 751                 dev->data->mtu = mtu;
 752
 753         return err;
 754 }
 755
 756 static int
 757 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
 758                      struct ether_addr *mc_addr_set __rte_unused,
 759                      uint32_t nb_mc_addr __rte_unused)
 760 {
 761         /*
 762          * Nothing to do actually: the tap has no filtering whatsoever, every
 763          * packet is received.
 764          */
 765         return 0;
 766 }
 767
 768 static const uint32_t*
 769 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
 770 {
 771         static const uint32_t ptypes[] = {
 772                 RTE_PTYPE_INNER_L2_ETHER,
 773                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
 774                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
 775                 RTE_PTYPE_INNER_L3_IPV4,
 776                 RTE_PTYPE_INNER_L3_IPV4_EXT,
 777                 RTE_PTYPE_INNER_L3_IPV6,
 778                 RTE_PTYPE_INNER_L3_IPV6_EXT,
 779                 RTE_PTYPE_INNER_L4_FRAG,
 780                 RTE_PTYPE_INNER_L4_UDP,
 781                 RTE_PTYPE_INNER_L4_TCP,
 782                 RTE_PTYPE_INNER_L4_SCTP,
 783                 RTE_PTYPE_L2_ETHER,
 784                 RTE_PTYPE_L2_ETHER_VLAN,
 785                 RTE_PTYPE_L2_ETHER_QINQ,
 786                 RTE_PTYPE_L3_IPV4,
 787                 RTE_PTYPE_L3_IPV4_EXT,
 788                 RTE_PTYPE_L3_IPV6_EXT,
 789                 RTE_PTYPE_L3_IPV6,
 790                 RTE_PTYPE_L4_FRAG,
 791                 RTE_PTYPE_L4_UDP,
 792                 RTE_PTYPE_L4_TCP,
 793                 RTE_PTYPE_L4_SCTP,
 794         };
 795
 796         return ptypes;
 797 }
 798
 799 static const struct eth_dev_ops ops = {
 800         .dev_start              = tap_dev_start,
 801         .dev_stop               = tap_dev_stop,
 802         .dev_close              = tap_dev_close,
 803         .dev_configure          = tap_dev_configure,
 804         .dev_infos_get          = tap_dev_info,
 805         .rx_queue_setup         = tap_rx_queue_setup,
 806         .tx_queue_setup         = tap_tx_queue_setup,
 807         .rx_queue_release       = tap_rx_queue_release,
 808         .tx_queue_release       = tap_tx_queue_release,
 809         .link_update            = tap_link_update,
 810         .dev_set_link_up        = tap_link_set_up,
 811         .dev_set_link_down      = tap_link_set_down,
 812         .promiscuous_enable     = tap_promisc_enable,
 813         .promiscuous_disable    = tap_promisc_disable,
 814         .allmulticast_enable    = tap_allmulti_enable,
 815         .allmulticast_disable   = tap_allmulti_disable,
 816         .mac_addr_set           = tap_mac_set,
 817         .mtu_set                = tap_mtu_set,
 818         .set_mc_addr_list       = tap_set_mc_addr_list,
 819         .stats_get              = tap_stats_get,
 820         .stats_reset            = tap_stats_reset,
 821         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
 822 };
 823
 824 static int
 825 eth_dev_tap_create(const char *name, char *tap_name)
 826 {
 827         int numa_node = rte_socket_id();
 828         struct rte_eth_dev *dev = NULL;
 829         struct pmd_internals *pmd = NULL;
 830         struct rte_eth_dev_data *data = NULL;
 831         int i;
 832
 833         RTE_LOG(DEBUG, PMD, "  TAP device on numa %u\n", rte_socket_id());
 834
 835         data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
 836         if (!data) {
 837                 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
 838                 goto error_exit;
 839         }
 840
 841         pmd = rte_zmalloc_socket(tap_name, sizeof(*pmd), 0, numa_node);
 842         if (!pmd) {
 843                 RTE_LOG(ERR, PMD, "TAP Unable to allocate internal struct\n");
 844                 goto error_exit;
 845         }
 846
 847         /* name in allocation and data->name must be consistent */
 848         snprintf(data->name, sizeof(data->name), "%s", name);
 849         dev = rte_eth_dev_allocate(name);
 850         if (!dev) {
 851                 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
 852                 goto error_exit;
 853         }
 854
 855         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
 856
 857         pmd->nb_queues = RTE_PMD_TAP_MAX_QUEUES;
 858
 859         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
 860         if (pmd->ioctl_sock == -1) {
 861                 RTE_LOG(ERR, PMD,
 862                         "TAP Unable to get a socket for management: %s\n",
 863                         strerror(errno));
 864                 goto error_exit;
 865         }
 866
 867         /* Setup some default values */
 868         data->dev_private = pmd;
 869         data->port_id = dev->data->port_id;
 870         data->mtu = dev->data->mtu;
 871         data->dev_flags = RTE_ETH_DEV_DETACHABLE;
 872         data->kdrv = RTE_KDRV_NONE;
 873         data->drv_name = pmd_tap_drv.driver.name;
 874         data->numa_node = numa_node;
 875
 876         data->dev_link = pmd_link;
 877         data->mac_addrs = &pmd->eth_addr;
 878         data->nb_rx_queues = pmd->nb_queues;
 879         data->nb_tx_queues = pmd->nb_queues;
 880
 881         dev->data = data;
 882         dev->dev_ops = &ops;
 883         dev->driver = NULL;
 884         dev->rx_pkt_burst = pmd_rx_burst;
 885         dev->tx_pkt_burst = pmd_tx_burst;
 886
 887         /* Presetup the fds to -1 as being not valid */
 888         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 889                 pmd->rxq[i].fd = -1;
 890                 pmd->txq[i].fd = -1;
 891         }
 892
 893         return 0;
 894
 895 error_exit:
 896         RTE_LOG(DEBUG, PMD, "TAP Unable to initialize %s\n", name);
 897
 898         rte_free(data);
 899         rte_free(pmd);
 900
 901         rte_eth_dev_release_port(dev);
 902
 903         return -EINVAL;
 904 }
 905
 906 static int
 907 set_interface_name(const char *key __rte_unused,
 908                    const char *value,
 909                    void *extra_args)
 910 {
 911         char *name = (char *)extra_args;
 912
 913         if (value)
 914                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
 915         else
 916                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
 917                          DEFAULT_TAP_NAME, (tap_unit - 1));
 918
 919         return 0;
 920 }
 921
 922 static int
 923 set_interface_speed(const char *key __rte_unused,
 924                     const char *value,
 925                     void *extra_args)
 926 {
 927         *(int *)extra_args = (value) ? atoi(value) : ETH_SPEED_NUM_10G;
 928
 929         return 0;
 930 }
 931
 932 /* Open a TAP interface device.
 933  */
 934 static int
 935 rte_pmd_tap_probe(const char *name, const char *params)
 936 {
 937         int ret;
 938         struct rte_kvargs *kvlist = NULL;
 939         int speed;
 940         char tap_name[RTE_ETH_NAME_MAX_LEN];
 941
 942         speed = ETH_SPEED_NUM_10G;
 943         snprintf(tap_name, sizeof(tap_name), "%s%d",
 944                  DEFAULT_TAP_NAME, tap_unit++);
 945
 946         if (params && (params[0] != '\0')) {
 947                 RTE_LOG(DEBUG, PMD, "paramaters (%s)\n", params);
 948
 949                 kvlist = rte_kvargs_parse(params, valid_arguments);
 950                 if (kvlist) {
 951                         if (rte_kvargs_count(kvlist, ETH_TAP_SPEED_ARG) == 1) {
 952                                 ret = rte_kvargs_process(kvlist,
 953                                                          ETH_TAP_SPEED_ARG,
 954                                                          &set_interface_speed,
 955                                                          &speed);
 956                                 if (ret == -1)
 957                                         goto leave;
 958                         }
 959
 960                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
 961                                 ret = rte_kvargs_process(kvlist,
 962                                                          ETH_TAP_IFACE_ARG,
 963                                                          &set_interface_name,
 964                                                          tap_name);
 965                                 if (ret == -1)
 966                                         goto leave;
 967                         }
 968                 }
 969         }
 970         pmd_link.link_speed = speed;
 971
 972         RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
 973                 name, tap_name);
 974
 975         ret = eth_dev_tap_create(name, tap_name);
 976
 977 leave:
 978         if (ret == -1) {
 979                 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
 980                         name, tap_name);
 981                 tap_unit--;             /* Restore the unit number */
 982         }
 983         rte_kvargs_free(kvlist);
 984
 985         return ret;
 986 }
 987
 988 /* detach a TAP device.
 989  */
 990 static int
 991 rte_pmd_tap_remove(const char *name)
 992 {
 993         struct rte_eth_dev *eth_dev = NULL;
 994         struct pmd_internals *internals;
 995         int i;
 996
 997         RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
 998                 rte_socket_id());
 999
1000         /* find the ethdev entry */
1001         eth_dev = rte_eth_dev_allocated(name);
1002         if (!eth_dev)
1003                 return 0;
1004
1005         internals = eth_dev->data->dev_private;
1006         for (i = 0; i < internals->nb_queues; i++)
1007                 if (internals->rxq[i].fd != -1)
1008                         close(internals->rxq[i].fd);
1009
1010         close(internals->ioctl_sock);
1011         rte_free(eth_dev->data->dev_private);
1012         rte_free(eth_dev->data);
1013
1014         rte_eth_dev_release_port(eth_dev);
1015
1016         return 0;
1017 }
1018
1019 static struct rte_vdev_driver pmd_tap_drv = {
1020         .probe = rte_pmd_tap_probe,
1021         .remove = rte_pmd_tap_remove,
1022 };
1023 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1024 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1025 RTE_PMD_REGISTER_PARAM_STRING(net_tap, "iface=<string>,speed=N");