drivers/net/tap/rte_eth_tap.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2016-2017 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_ethdev_vdev.h>
  40 #include <rte_malloc.h>
  41 #include <rte_vdev.h>
  42 #include <rte_kvargs.h>
  43 #include <rte_net.h>
  44 #include <rte_debug.h>
  45
  46 #include <sys/types.h>
  47 #include <sys/stat.h>
  48 #include <sys/socket.h>
  49 #include <sys/ioctl.h>
  50 #include <sys/utsname.h>
  51 #include <sys/mman.h>
  52 #include <errno.h>
  53 #include <signal.h>
  54 #include <stdint.h>
  55 #include <sys/uio.h>
  56 #include <unistd.h>
  57 #include <arpa/inet.h>
  58 #include <net/if.h>
  59 #include <linux/if_tun.h>
  60 #include <linux/if_ether.h>
  61 #include <linux/version.h>
  62 #include <fcntl.h>
  63
  64 #include <rte_eth_tap.h>
  65 #include <tap_flow.h>
  66 #include <tap_netlink.h>
  67 #include <tap_tcmsgs.h>
  68
  69 /* Linux based path to the TUN device */
  70 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  71 #define DEFAULT_TAP_NAME        "dtap"
  72
  73 #define ETH_TAP_IFACE_ARG       "iface"
  74 #define ETH_TAP_SPEED_ARG       "speed"
  75 #define ETH_TAP_REMOTE_ARG      "remote"
  76 #define ETH_TAP_MAC_ARG         "mac"
  77 #define ETH_TAP_MAC_FIXED       "fixed"
  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         ETH_TAP_REMOTE_ARG,
  88         ETH_TAP_MAC_ARG,
  89         NULL
  90 };
  91
  92 static int tap_unit;
  93
  94 static volatile uint32_t tap_trigger;   /* Rx trigger */
  95
  96 static struct rte_eth_link pmd_link = {
  97         .link_speed = ETH_SPEED_NUM_10G,
  98         .link_duplex = ETH_LINK_FULL_DUPLEX,
  99         .link_status = ETH_LINK_DOWN,
 100         .link_autoneg = ETH_LINK_SPEED_AUTONEG
 101 };
 102
 103 static void
 104 tap_trigger_cb(int sig __rte_unused)
 105 {
 106         /* Valid trigger values are nonzero */
 107         tap_trigger = (tap_trigger + 1) | 0x80000000;
 108 }
 109
 110 /* Specifies on what netdevices the ioctl should be applied */
 111 enum ioctl_mode {
 112         LOCAL_AND_REMOTE,
 113         LOCAL_ONLY,
 114         REMOTE_ONLY,
 115 };
 116
 117 static int
 118 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 119           struct ifreq *ifr, int set, enum ioctl_mode mode);
 120
 121 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
 122
 123 /* Tun/Tap allocation routine
 124  *
 125  * name is the number of the interface to use, unless NULL to take the host
 126  * supplied name.
 127  */
 128 static int
 129 tun_alloc(struct pmd_internals *pmd, uint16_t qid)
 130 {
 131         struct ifreq ifr;
 132 #ifdef IFF_MULTI_QUEUE
 133         unsigned int features;
 134 #endif
 135         int fd;
 136
 137         memset(&ifr, 0, sizeof(struct ifreq));
 138
 139         /*
 140          * Do not set IFF_NO_PI as packet information header will be needed
 141          * to check if a received packet has been truncated.
 142          */
 143         ifr.ifr_flags = IFF_TAP;
 144         snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
 145
 146         RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
 147
 148         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 149         if (fd < 0) {
 150                 RTE_LOG(ERR, PMD, "Unable to create TAP interface");
 151                 goto error;
 152         }
 153
 154 #ifdef IFF_MULTI_QUEUE
 155         /* Grab the TUN features to verify we can work multi-queue */
 156         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 157                 RTE_LOG(ERR, PMD, "TAP unable to get TUN/TAP features\n");
 158                 goto error;
 159         }
 160         RTE_LOG(DEBUG, PMD, "  TAP Features %08x\n", features);
 161
 162         if (features & IFF_MULTI_QUEUE) {
 163                 RTE_LOG(DEBUG, PMD, "  Multi-queue support for %d queues\n",
 164                         RTE_PMD_TAP_MAX_QUEUES);
 165                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 166         } else
 167 #endif
 168         {
 169                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 170                 RTE_LOG(DEBUG, PMD, "  Single queue only support\n");
 171         }
 172
 173         /* Set the TUN/TAP configuration and set the name if needed */
 174         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 175                 RTE_LOG(WARNING, PMD,
 176                         "Unable to set TUNSETIFF for %s\n",
 177                         ifr.ifr_name);
 178                 perror("TUNSETIFF");
 179                 goto error;
 180         }
 181
 182         /* Always set the file descriptor to non-blocking */
 183         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 184                 RTE_LOG(WARNING, PMD,
 185                         "Unable to set %s to nonblocking\n",
 186                         ifr.ifr_name);
 187                 perror("F_SETFL, NONBLOCK");
 188                 goto error;
 189         }
 190
 191         /* Set up trigger to optimize empty Rx bursts */
 192         errno = 0;
 193         do {
 194                 struct sigaction sa;
 195                 int flags = fcntl(fd, F_GETFL);
 196
 197                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 198                         break;
 199                 if (sa.sa_handler != tap_trigger_cb) {
 200                         /*
 201                          * Make sure SIGIO is not already taken. This is done
 202                          * as late as possible to leave the application a
 203                          * chance to set up its own signal handler first.
 204                          */
 205                         if (sa.sa_handler != SIG_IGN &&
 206                             sa.sa_handler != SIG_DFL) {
 207                                 errno = EBUSY;
 208                                 break;
 209                         }
 210                         sa = (struct sigaction){
 211                                 .sa_flags = SA_RESTART,
 212                                 .sa_handler = tap_trigger_cb,
 213                         };
 214                         if (sigaction(SIGIO, &sa, NULL) == -1)
 215                                 break;
 216                 }
 217                 /* Enable SIGIO on file descriptor */
 218                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 219                 fcntl(fd, F_SETOWN, getpid());
 220         } while (0);
 221         if (errno) {
 222                 /* Disable trigger globally in case of error */
 223                 tap_trigger = 0;
 224                 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
 225                         strerror(errno));
 226         }
 227
 228         if (qid == 0) {
 229                 struct ifreq ifr;
 230
 231                 /*
 232                  * pmd->eth_addr contains the desired MAC, either from remote
 233                  * or from a random assignment. Sync it with the tap netdevice.
 234                  */
 235                 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
 236                 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
 237                            ETHER_ADDR_LEN);
 238                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
 239                         goto error;
 240
 241                 pmd->if_index = if_nametoindex(pmd->name);
 242                 if (!pmd->if_index) {
 243                         RTE_LOG(ERR, PMD,
 244                                 "Could not find ifindex for %s: rte_flow won't be usable.\n",
 245                                 pmd->name);
 246                         return fd;
 247                 }
 248                 if (!pmd->flower_support)
 249                         return fd;
 250                 if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
 251                         RTE_LOG(ERR, PMD,
 252                                 "Could not create multiq qdisc for %s: rte_flow won't be usable.\n",
 253                                 pmd->name);
 254                         return fd;
 255                 }
 256                 if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
 257                         RTE_LOG(ERR, PMD,
 258                                 "Could not create multiq qdisc for %s: rte_flow won't be usable.\n",
 259                                 pmd->name);
 260                         return fd;
 261                 }
 262                 if (pmd->remote_if_index) {
 263                         /*
 264                          * Flush usually returns negative value because it tries
 265                          * to delete every QDISC (and on a running device, one
 266                          * QDISC at least is needed). Ignore negative return
 267                          * value.
 268                          */
 269                         qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
 270                         if (qdisc_create_ingress(pmd->nlsk_fd,
 271                                                  pmd->remote_if_index) < 0)
 272                                 goto remote_fail;
 273                         LIST_INIT(&pmd->implicit_flows);
 274                         if (tap_flow_implicit_create(
 275                                     pmd, TAP_REMOTE_LOCAL_MAC) < 0)
 276                                 goto remote_fail;
 277                         if (tap_flow_implicit_create(
 278                                     pmd, TAP_REMOTE_BROADCAST) < 0)
 279                                 goto remote_fail;
 280                         if (tap_flow_implicit_create(
 281                                     pmd, TAP_REMOTE_BROADCASTV6) < 0)
 282                                 goto remote_fail;
 283                         if (tap_flow_implicit_create(
 284                                     pmd, TAP_REMOTE_TX) < 0)
 285                                 goto remote_fail;
 286                 }
 287         }
 288
 289         return fd;
 290
 291 remote_fail:
 292         RTE_LOG(ERR, PMD,
 293                 "Could not set up remote flow rules for %s: remote disabled.\n",
 294                 pmd->name);
 295         pmd->remote_if_index = 0;
 296         tap_flow_implicit_flush(pmd, NULL);
 297         return fd;
 298
 299 error:
 300         if (fd > 0)
 301                 close(fd);
 302         return -1;
 303 }
 304
 305 /* Callback to handle the rx burst of packets to the correct interface and
 306  * file descriptor(s) in a multi-queue setup.
 307  */
 308 static uint16_t
 309 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 310 {
 311         struct rx_queue *rxq = queue;
 312         uint16_t num_rx;
 313         unsigned long num_rx_bytes = 0;
 314         uint32_t trigger = tap_trigger;
 315
 316         if (trigger == rxq->trigger_seen)
 317                 return 0;
 318         if (trigger)
 319                 rxq->trigger_seen = trigger;
 320         rte_compiler_barrier();
 321         for (num_rx = 0; num_rx < nb_pkts; ) {
 322                 struct rte_mbuf *mbuf = rxq->pool;
 323                 struct rte_mbuf *seg = NULL;
 324                 struct rte_mbuf *new_tail = NULL;
 325                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 326                 int len;
 327
 328                 len = readv(rxq->fd, *rxq->iovecs,
 329                             1 + (rxq->rxmode->enable_scatter ?
 330                                  rxq->nb_rx_desc : 1));
 331                 if (len < (int)sizeof(struct tun_pi))
 332                         break;
 333
 334                 /* Packet couldn't fit in the provided mbuf */
 335                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 336                         rxq->stats.ierrors++;
 337                         continue;
 338                 }
 339
 340                 len -= sizeof(struct tun_pi);
 341
 342                 mbuf->pkt_len = len;
 343                 mbuf->port = rxq->in_port;
 344                 while (1) {
 345                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 346
 347                         if (unlikely(!buf)) {
 348                                 rxq->stats.rx_nombuf++;
 349                                 /* No new buf has been allocated: do nothing */
 350                                 if (!new_tail || !seg)
 351                                         goto end;
 352
 353                                 seg->next = NULL;
 354                                 rte_pktmbuf_free(mbuf);
 355
 356                                 goto end;
 357                         }
 358                         seg = seg ? seg->next : mbuf;
 359                         if (rxq->pool == mbuf)
 360                                 rxq->pool = buf;
 361                         if (new_tail)
 362                                 new_tail->next = buf;
 363                         new_tail = buf;
 364                         new_tail->next = seg->next;
 365
 366                         /* iovecs[0] is reserved for packet info (pi) */
 367                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 368                                 buf->buf_len - data_off;
 369                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 370                                 (char *)buf->buf_addr + data_off;
 371
 372                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 373                         seg->data_off = data_off;
 374
 375                         len -= seg->data_len;
 376                         if (len <= 0)
 377                                 break;
 378                         mbuf->nb_segs++;
 379                         /* First segment has headroom, not the others */
 380                         data_off = 0;
 381                 }
 382                 seg->next = NULL;
 383                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 384                                                       RTE_PTYPE_ALL_MASK);
 385
 386                 /* account for the receive frame */
 387                 bufs[num_rx++] = mbuf;
 388                 num_rx_bytes += mbuf->pkt_len;
 389         }
 390 end:
 391         rxq->stats.ipackets += num_rx;
 392         rxq->stats.ibytes += num_rx_bytes;
 393
 394         return num_rx;
 395 }
 396
 397 /* Callback to handle sending packets from the tap interface
 398  */
 399 static uint16_t
 400 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 401 {
 402         struct tx_queue *txq = queue;
 403         uint16_t num_tx = 0;
 404         unsigned long num_tx_bytes = 0;
 405         uint32_t max_size;
 406         int i;
 407
 408         if (unlikely(nb_pkts == 0))
 409                 return 0;
 410
 411         max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
 412         for (i = 0; i < nb_pkts; i++) {
 413                 struct rte_mbuf *mbuf = bufs[num_tx];
 414                 struct iovec iovecs[mbuf->nb_segs + 1];
 415                 struct tun_pi pi = { .flags = 0 };
 416                 struct rte_mbuf *seg = mbuf;
 417                 int n;
 418                 int j;
 419
 420                 /* stats.errs will be incremented */
 421                 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
 422                         break;
 423
 424                 iovecs[0].iov_base = &pi;
 425                 iovecs[0].iov_len = sizeof(pi);
 426                 for (j = 1; j <= mbuf->nb_segs; j++) {
 427                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 428                         iovecs[j].iov_base =
 429                                 rte_pktmbuf_mtod(seg, void *);
 430                         seg = seg->next;
 431                 }
 432                 /* copy the tx frame data */
 433                 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
 434                 if (n <= 0)
 435                         break;
 436
 437                 num_tx++;
 438                 num_tx_bytes += mbuf->pkt_len;
 439                 rte_pktmbuf_free(mbuf);
 440         }
 441
 442         txq->stats.opackets += num_tx;
 443         txq->stats.errs += nb_pkts - num_tx;
 444         txq->stats.obytes += num_tx_bytes;
 445
 446         return num_tx;
 447 }
 448
 449 static const char *
 450 tap_ioctl_req2str(unsigned long request)
 451 {
 452         switch (request) {
 453         case SIOCSIFFLAGS:
 454                 return "SIOCSIFFLAGS";
 455         case SIOCGIFFLAGS:
 456                 return "SIOCGIFFLAGS";
 457         case SIOCGIFHWADDR:
 458                 return "SIOCGIFHWADDR";
 459         case SIOCSIFHWADDR:
 460                 return "SIOCSIFHWADDR";
 461         case SIOCSIFMTU:
 462                 return "SIOCSIFMTU";
 463         }
 464         return "UNKNOWN";
 465 }
 466
 467 static int
 468 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 469           struct ifreq *ifr, int set, enum ioctl_mode mode)
 470 {
 471         short req_flags = ifr->ifr_flags;
 472         int remote = pmd->remote_if_index &&
 473                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 474
 475         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 476                 return 0;
 477         /*
 478          * If there is a remote netdevice, apply ioctl on it, then apply it on
 479          * the tap netdevice.
 480          */
 481 apply:
 482         if (remote)
 483                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
 484         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 485                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 486         switch (request) {
 487         case SIOCSIFFLAGS:
 488                 /* fetch current flags to leave other flags untouched */
 489                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 490                         goto error;
 491                 if (set)
 492                         ifr->ifr_flags |= req_flags;
 493                 else
 494                         ifr->ifr_flags &= ~req_flags;
 495                 break;
 496         case SIOCGIFFLAGS:
 497         case SIOCGIFHWADDR:
 498         case SIOCSIFHWADDR:
 499         case SIOCSIFMTU:
 500                 break;
 501         default:
 502                 RTE_ASSERT(!"unsupported request type: must not happen");
 503         }
 504         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 505                 goto error;
 506         if (remote-- && mode == LOCAL_AND_REMOTE)
 507                 goto apply;
 508         return 0;
 509
 510 error:
 511         RTE_LOG(DEBUG, PMD, "%s: %s(%s) failed: %s(%d)\n", ifr->ifr_name,
 512                 __func__, tap_ioctl_req2str(request), strerror(errno), errno);
 513         return -errno;
 514 }
 515
 516 static int
 517 tap_link_set_down(struct rte_eth_dev *dev)
 518 {
 519         struct pmd_internals *pmd = dev->data->dev_private;
 520         struct ifreq ifr = { .ifr_flags = IFF_UP };
 521
 522         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 523         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 524 }
 525
 526 static int
 527 tap_link_set_up(struct rte_eth_dev *dev)
 528 {
 529         struct pmd_internals *pmd = dev->data->dev_private;
 530         struct ifreq ifr = { .ifr_flags = IFF_UP };
 531
 532         dev->data->dev_link.link_status = ETH_LINK_UP;
 533         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 534 }
 535
 536 static int
 537 tap_dev_start(struct rte_eth_dev *dev)
 538 {
 539         int err;
 540
 541         err = tap_intr_handle_set(dev, 1);
 542         if (err)
 543                 return err;
 544         return tap_link_set_up(dev);
 545 }
 546
 547 /* This function gets called when the current port gets stopped.
 548  */
 549 static void
 550 tap_dev_stop(struct rte_eth_dev *dev)
 551 {
 552         tap_intr_handle_set(dev, 0);
 553         tap_link_set_down(dev);
 554 }
 555
 556 static int
 557 tap_dev_configure(struct rte_eth_dev *dev __rte_unused)
 558 {
 559         return 0;
 560 }
 561
 562 static uint32_t
 563 tap_dev_speed_capa(void)
 564 {
 565         uint32_t speed = pmd_link.link_speed;
 566         uint32_t capa = 0;
 567
 568         if (speed >= ETH_SPEED_NUM_10M)
 569                 capa |= ETH_LINK_SPEED_10M;
 570         if (speed >= ETH_SPEED_NUM_100M)
 571                 capa |= ETH_LINK_SPEED_100M;
 572         if (speed >= ETH_SPEED_NUM_1G)
 573                 capa |= ETH_LINK_SPEED_1G;
 574         if (speed >= ETH_SPEED_NUM_5G)
 575                 capa |= ETH_LINK_SPEED_2_5G;
 576         if (speed >= ETH_SPEED_NUM_5G)
 577                 capa |= ETH_LINK_SPEED_5G;
 578         if (speed >= ETH_SPEED_NUM_10G)
 579                 capa |= ETH_LINK_SPEED_10G;
 580         if (speed >= ETH_SPEED_NUM_20G)
 581                 capa |= ETH_LINK_SPEED_20G;
 582         if (speed >= ETH_SPEED_NUM_25G)
 583                 capa |= ETH_LINK_SPEED_25G;
 584         if (speed >= ETH_SPEED_NUM_40G)
 585                 capa |= ETH_LINK_SPEED_40G;
 586         if (speed >= ETH_SPEED_NUM_50G)
 587                 capa |= ETH_LINK_SPEED_50G;
 588         if (speed >= ETH_SPEED_NUM_56G)
 589                 capa |= ETH_LINK_SPEED_56G;
 590         if (speed >= ETH_SPEED_NUM_100G)
 591                 capa |= ETH_LINK_SPEED_100G;
 592
 593         return capa;
 594 }
 595
 596 static void
 597 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 598 {
 599         struct pmd_internals *internals = dev->data->dev_private;
 600
 601         dev_info->if_index = internals->if_index;
 602         dev_info->max_mac_addrs = 1;
 603         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 604         dev_info->max_rx_queues = internals->nb_queues;
 605         dev_info->max_tx_queues = internals->nb_queues;
 606         dev_info->min_rx_bufsize = 0;
 607         dev_info->pci_dev = NULL;
 608         dev_info->speed_capa = tap_dev_speed_capa();
 609 }
 610
 611 static void
 612 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 613 {
 614         unsigned int i, imax;
 615         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 616         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 617         unsigned long rx_nombuf = 0, ierrors = 0;
 618         const struct pmd_internals *pmd = dev->data->dev_private;
 619
 620         imax = (pmd->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 621                 pmd->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 622
 623         for (i = 0; i < imax; i++) {
 624                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 625                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 626                 rx_total += tap_stats->q_ipackets[i];
 627                 rx_bytes_total += tap_stats->q_ibytes[i];
 628                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
 629                 ierrors += pmd->rxq[i].stats.ierrors;
 630
 631                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 632                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 633                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 634                 tx_total += tap_stats->q_opackets[i];
 635                 tx_err_total += tap_stats->q_errors[i];
 636                 tx_bytes_total += tap_stats->q_obytes[i];
 637         }
 638
 639         tap_stats->ipackets = rx_total;
 640         tap_stats->ibytes = rx_bytes_total;
 641         tap_stats->ierrors = ierrors;
 642         tap_stats->rx_nombuf = rx_nombuf;
 643         tap_stats->opackets = tx_total;
 644         tap_stats->oerrors = tx_err_total;
 645         tap_stats->obytes = tx_bytes_total;
 646 }
 647
 648 static void
 649 tap_stats_reset(struct rte_eth_dev *dev)
 650 {
 651         int i;
 652         struct pmd_internals *pmd = dev->data->dev_private;
 653
 654         for (i = 0; i < pmd->nb_queues; i++) {
 655                 pmd->rxq[i].stats.ipackets = 0;
 656                 pmd->rxq[i].stats.ibytes = 0;
 657                 pmd->rxq[i].stats.ierrors = 0;
 658                 pmd->rxq[i].stats.rx_nombuf = 0;
 659
 660                 pmd->txq[i].stats.opackets = 0;
 661                 pmd->txq[i].stats.errs = 0;
 662                 pmd->txq[i].stats.obytes = 0;
 663         }
 664 }
 665
 666 static void
 667 tap_dev_close(struct rte_eth_dev *dev __rte_unused)
 668 {
 669         int i;
 670         struct pmd_internals *internals = dev->data->dev_private;
 671
 672         tap_link_set_down(dev);
 673         tap_flow_flush(dev, NULL);
 674         tap_flow_implicit_flush(internals, NULL);
 675
 676         for (i = 0; i < internals->nb_queues; i++) {
 677                 if (internals->rxq[i].fd != -1)
 678                         close(internals->rxq[i].fd);
 679                 internals->rxq[i].fd = -1;
 680                 internals->txq[i].fd = -1;
 681         }
 682 }
 683
 684 static void
 685 tap_rx_queue_release(void *queue)
 686 {
 687         struct rx_queue *rxq = queue;
 688
 689         if (rxq && (rxq->fd > 0)) {
 690                 close(rxq->fd);
 691                 rxq->fd = -1;
 692                 rte_pktmbuf_free(rxq->pool);
 693                 rte_free(rxq->iovecs);
 694                 rxq->pool = NULL;
 695                 rxq->iovecs = NULL;
 696         }
 697 }
 698
 699 static void
 700 tap_tx_queue_release(void *queue)
 701 {
 702         struct tx_queue *txq = queue;
 703
 704         if (txq && (txq->fd > 0)) {
 705                 close(txq->fd);
 706                 txq->fd = -1;
 707         }
 708 }
 709
 710 static int
 711 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
 712 {
 713         struct rte_eth_link *dev_link = &dev->data->dev_link;
 714         struct pmd_internals *pmd = dev->data->dev_private;
 715         struct ifreq ifr = { .ifr_flags = 0 };
 716
 717         if (pmd->remote_if_index) {
 718                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
 719                 if (!(ifr.ifr_flags & IFF_UP) ||
 720                     !(ifr.ifr_flags & IFF_RUNNING)) {
 721                         dev_link->link_status = ETH_LINK_DOWN;
 722                         return 0;
 723                 }
 724         }
 725         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
 726         dev_link->link_status =
 727                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
 728                  ETH_LINK_UP :
 729                  ETH_LINK_DOWN);
 730         return 0;
 731 }
 732
 733 static void
 734 tap_promisc_enable(struct rte_eth_dev *dev)
 735 {
 736         struct pmd_internals *pmd = dev->data->dev_private;
 737         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 738
 739         dev->data->promiscuous = 1;
 740         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 741         if (pmd->remote_if_index)
 742                 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
 743 }
 744
 745 static void
 746 tap_promisc_disable(struct rte_eth_dev *dev)
 747 {
 748         struct pmd_internals *pmd = dev->data->dev_private;
 749         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 750
 751         dev->data->promiscuous = 0;
 752         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 753         if (pmd->remote_if_index)
 754                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
 755 }
 756
 757 static void
 758 tap_allmulti_enable(struct rte_eth_dev *dev)
 759 {
 760         struct pmd_internals *pmd = dev->data->dev_private;
 761         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 762
 763         dev->data->all_multicast = 1;
 764         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 765         if (pmd->remote_if_index)
 766                 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
 767 }
 768
 769 static void
 770 tap_allmulti_disable(struct rte_eth_dev *dev)
 771 {
 772         struct pmd_internals *pmd = dev->data->dev_private;
 773         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 774
 775         dev->data->all_multicast = 0;
 776         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 777         if (pmd->remote_if_index)
 778                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
 779 }
 780
 781
 782 static void
 783 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 784 {
 785         struct pmd_internals *pmd = dev->data->dev_private;
 786         struct ifreq ifr;
 787
 788         if (is_zero_ether_addr(mac_addr)) {
 789                 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
 790                         dev->data->name);
 791                 return;
 792         }
 793         /* Check the actual current MAC address on the tap netdevice */
 794         if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY) != 0)
 795                 return;
 796         if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 797                                mac_addr))
 798                 return;
 799
 800         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
 801         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
 802         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, LOCAL_AND_REMOTE) < 0)
 803                 return;
 804         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
 805         if (pmd->remote_if_index) {
 806                 /* Replace MAC redirection rule after a MAC change */
 807                 if (tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC) < 0) {
 808                         RTE_LOG(ERR, PMD,
 809                                 "%s: Couldn't delete MAC redirection rule\n",
 810                                 dev->data->name);
 811                         return;
 812                 }
 813                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0)
 814                         RTE_LOG(ERR, PMD,
 815                                 "%s: Couldn't add MAC redirection rule\n",
 816                                 dev->data->name);
 817         }
 818 }
 819
 820 static int
 821 tap_setup_queue(struct rte_eth_dev *dev,
 822                 struct pmd_internals *internals,
 823                 uint16_t qid)
 824 {
 825         struct pmd_internals *pmd = dev->data->dev_private;
 826         struct rx_queue *rx = &internals->rxq[qid];
 827         struct tx_queue *tx = &internals->txq[qid];
 828         int fd;
 829
 830         fd = rx->fd;
 831         if (fd < 0) {
 832                 fd = tx->fd;
 833                 if (fd < 0) {
 834                         RTE_LOG(INFO, PMD, "Add queue to TAP %s for qid %d\n",
 835                                 pmd->name, qid);
 836                         fd = tun_alloc(pmd, qid);
 837                         if (fd < 0) {
 838                                 RTE_LOG(ERR, PMD, "tun_alloc(%s, %d) failed\n",
 839                                         pmd->name, qid);
 840                                 return -1;
 841                         }
 842                         if (qid == 0) {
 843                                 struct ifreq ifr;
 844
 845                                 ifr.ifr_mtu = dev->data->mtu;
 846                                 if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1,
 847                                               LOCAL_AND_REMOTE) < 0) {
 848                                         close(fd);
 849                                         return -1;
 850                                 }
 851                         }
 852                 }
 853         }
 854
 855         rx->fd = fd;
 856         tx->fd = fd;
 857         tx->mtu = &dev->data->mtu;
 858         rx->rxmode = &dev->data->dev_conf.rxmode;
 859
 860         return fd;
 861 }
 862
 863 static int
 864 rx_setup_queue(struct rte_eth_dev *dev,
 865                 struct pmd_internals *internals,
 866                 uint16_t qid)
 867 {
 868         dev->data->rx_queues[qid] = &internals->rxq[qid];
 869
 870         return tap_setup_queue(dev, internals, qid);
 871 }
 872
 873 static int
 874 tx_setup_queue(struct rte_eth_dev *dev,
 875                 struct pmd_internals *internals,
 876                 uint16_t qid)
 877 {
 878         dev->data->tx_queues[qid] = &internals->txq[qid];
 879
 880         return tap_setup_queue(dev, internals, qid);
 881 }
 882
 883 static int
 884 tap_rx_queue_setup(struct rte_eth_dev *dev,
 885                    uint16_t rx_queue_id,
 886                    uint16_t nb_rx_desc,
 887                    unsigned int socket_id,
 888                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 889                    struct rte_mempool *mp)
 890 {
 891         struct pmd_internals *internals = dev->data->dev_private;
 892         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
 893         struct rte_mbuf **tmp = &rxq->pool;
 894         long iov_max = sysconf(_SC_IOV_MAX);
 895         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
 896         struct iovec (*iovecs)[nb_desc + 1];
 897         int data_off = RTE_PKTMBUF_HEADROOM;
 898         int ret = 0;
 899         int fd;
 900         int i;
 901
 902         if ((rx_queue_id >= internals->nb_queues) || !mp) {
 903                 RTE_LOG(WARNING, PMD,
 904                         "nb_queues %d too small or mempool NULL\n",
 905                         internals->nb_queues);
 906                 return -1;
 907         }
 908
 909         rxq->mp = mp;
 910         rxq->trigger_seen = 1; /* force initial burst */
 911         rxq->in_port = dev->data->port_id;
 912         rxq->nb_rx_desc = nb_desc;
 913         iovecs = rte_zmalloc_socket(dev->data->name, sizeof(*iovecs), 0,
 914                                     socket_id);
 915         if (!iovecs) {
 916                 RTE_LOG(WARNING, PMD,
 917                         "%s: Couldn't allocate %d RX descriptors\n",
 918                         dev->data->name, nb_desc);
 919                 return -ENOMEM;
 920         }
 921         rxq->iovecs = iovecs;
 922
 923         fd = rx_setup_queue(dev, internals, rx_queue_id);
 924         if (fd == -1) {
 925                 ret = fd;
 926                 goto error;
 927         }
 928
 929         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
 930         (*rxq->iovecs)[0].iov_base = &rxq->pi;
 931
 932         for (i = 1; i <= nb_desc; i++) {
 933                 *tmp = rte_pktmbuf_alloc(rxq->mp);
 934                 if (!*tmp) {
 935                         RTE_LOG(WARNING, PMD,
 936                                 "%s: couldn't allocate memory for queue %d\n",
 937                                 dev->data->name, rx_queue_id);
 938                         ret = -ENOMEM;
 939                         goto error;
 940                 }
 941                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
 942                 (*rxq->iovecs)[i].iov_base =
 943                         (char *)(*tmp)->buf_addr + data_off;
 944                 data_off = 0;
 945                 tmp = &(*tmp)->next;
 946         }
 947
 948         RTE_LOG(DEBUG, PMD, "  RX TAP device name %s, qid %d on fd %d\n",
 949                 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
 950
 951         return 0;
 952
 953 error:
 954         rte_pktmbuf_free(rxq->pool);
 955         rxq->pool = NULL;
 956         rte_free(rxq->iovecs);
 957         rxq->iovecs = NULL;
 958         return ret;
 959 }
 960
 961 static int
 962 tap_tx_queue_setup(struct rte_eth_dev *dev,
 963                    uint16_t tx_queue_id,
 964                    uint16_t nb_tx_desc __rte_unused,
 965                    unsigned int socket_id __rte_unused,
 966                    const struct rte_eth_txconf *tx_conf __rte_unused)
 967 {
 968         struct pmd_internals *internals = dev->data->dev_private;
 969         int ret;
 970
 971         if (tx_queue_id >= internals->nb_queues)
 972                 return -1;
 973
 974         ret = tx_setup_queue(dev, internals, tx_queue_id);
 975         if (ret == -1)
 976                 return -1;
 977
 978         RTE_LOG(DEBUG, PMD, "  TX TAP device name %s, qid %d on fd %d\n",
 979                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd);
 980
 981         return 0;
 982 }
 983
 984 static int
 985 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
 986 {
 987         struct pmd_internals *pmd = dev->data->dev_private;
 988         struct ifreq ifr = { .ifr_mtu = mtu };
 989         int err = 0;
 990
 991         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
 992         if (!err)
 993                 dev->data->mtu = mtu;
 994
 995         return err;
 996 }
 997
 998 static int
 999 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1000                      struct ether_addr *mc_addr_set __rte_unused,
1001                      uint32_t nb_mc_addr __rte_unused)
1002 {
1003         /*
1004          * Nothing to do actually: the tap has no filtering whatsoever, every
1005          * packet is received.
1006          */
1007         return 0;
1008 }
1009
1010 static int
1011 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1012 {
1013         struct rte_eth_dev *dev = arg;
1014         struct pmd_internals *pmd = dev->data->dev_private;
1015         struct ifinfomsg *info = NLMSG_DATA(nh);
1016
1017         if (nh->nlmsg_type != RTM_NEWLINK ||
1018             (info->ifi_index != pmd->if_index &&
1019              info->ifi_index != pmd->remote_if_index))
1020                 return 0;
1021         return tap_link_update(dev, 0);
1022 }
1023
1024 static void
1025 tap_dev_intr_handler(void *cb_arg)
1026 {
1027         struct rte_eth_dev *dev = cb_arg;
1028         struct pmd_internals *pmd = dev->data->dev_private;
1029
1030         nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1031 }
1032
1033 static int
1034 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1035 {
1036         struct pmd_internals *pmd = dev->data->dev_private;
1037
1038         /* In any case, disable interrupt if the conf is no longer there. */
1039         if (!dev->data->dev_conf.intr_conf.lsc) {
1040                 if (pmd->intr_handle.fd != -1)
1041                         nl_final(pmd->intr_handle.fd);
1042                 rte_intr_callback_unregister(
1043                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1044                 return 0;
1045         }
1046         if (set) {
1047                 pmd->intr_handle.fd = nl_init(RTMGRP_LINK);
1048                 if (unlikely(pmd->intr_handle.fd == -1))
1049                         return -EBADF;
1050                 return rte_intr_callback_register(
1051                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1052         }
1053         nl_final(pmd->intr_handle.fd);
1054         return rte_intr_callback_unregister(&pmd->intr_handle,
1055                                             tap_dev_intr_handler, dev);
1056 }
1057
1058 static const uint32_t*
1059 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1060 {
1061         static const uint32_t ptypes[] = {
1062                 RTE_PTYPE_INNER_L2_ETHER,
1063                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1064                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1065                 RTE_PTYPE_INNER_L3_IPV4,
1066                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1067                 RTE_PTYPE_INNER_L3_IPV6,
1068                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1069                 RTE_PTYPE_INNER_L4_FRAG,
1070                 RTE_PTYPE_INNER_L4_UDP,
1071                 RTE_PTYPE_INNER_L4_TCP,
1072                 RTE_PTYPE_INNER_L4_SCTP,
1073                 RTE_PTYPE_L2_ETHER,
1074                 RTE_PTYPE_L2_ETHER_VLAN,
1075                 RTE_PTYPE_L2_ETHER_QINQ,
1076                 RTE_PTYPE_L3_IPV4,
1077                 RTE_PTYPE_L3_IPV4_EXT,
1078                 RTE_PTYPE_L3_IPV6_EXT,
1079                 RTE_PTYPE_L3_IPV6,
1080                 RTE_PTYPE_L4_FRAG,
1081                 RTE_PTYPE_L4_UDP,
1082                 RTE_PTYPE_L4_TCP,
1083                 RTE_PTYPE_L4_SCTP,
1084         };
1085
1086         return ptypes;
1087 }
1088
1089 static int
1090 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1091                   struct rte_eth_fc_conf *fc_conf)
1092 {
1093         fc_conf->mode = RTE_FC_NONE;
1094         return 0;
1095 }
1096
1097 static int
1098 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1099                   struct rte_eth_fc_conf *fc_conf)
1100 {
1101         if (fc_conf->mode != RTE_FC_NONE)
1102                 return -ENOTSUP;
1103         return 0;
1104 }
1105
1106 static const struct eth_dev_ops ops = {
1107         .dev_start              = tap_dev_start,
1108         .dev_stop               = tap_dev_stop,
1109         .dev_close              = tap_dev_close,
1110         .dev_configure          = tap_dev_configure,
1111         .dev_infos_get          = tap_dev_info,
1112         .rx_queue_setup         = tap_rx_queue_setup,
1113         .tx_queue_setup         = tap_tx_queue_setup,
1114         .rx_queue_release       = tap_rx_queue_release,
1115         .tx_queue_release       = tap_tx_queue_release,
1116         .flow_ctrl_get          = tap_flow_ctrl_get,
1117         .flow_ctrl_set          = tap_flow_ctrl_set,
1118         .link_update            = tap_link_update,
1119         .dev_set_link_up        = tap_link_set_up,
1120         .dev_set_link_down      = tap_link_set_down,
1121         .promiscuous_enable     = tap_promisc_enable,
1122         .promiscuous_disable    = tap_promisc_disable,
1123         .allmulticast_enable    = tap_allmulti_enable,
1124         .allmulticast_disable   = tap_allmulti_disable,
1125         .mac_addr_set           = tap_mac_set,
1126         .mtu_set                = tap_mtu_set,
1127         .set_mc_addr_list       = tap_set_mc_addr_list,
1128         .stats_get              = tap_stats_get,
1129         .stats_reset            = tap_stats_reset,
1130         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1131         .filter_ctrl            = tap_dev_filter_ctrl,
1132 };
1133
1134 static int
1135 tap_kernel_support(struct pmd_internals *pmd)
1136 {
1137         struct utsname utsname;
1138         int ver[3];
1139
1140         if (uname(&utsname) == -1 ||
1141             sscanf(utsname.release, "%d.%d.%d",
1142                    &ver[0], &ver[1], &ver[2]) != 3)
1143                 return 0;
1144         if (KERNEL_VERSION(ver[0], ver[1], ver[2]) >= FLOWER_KERNEL_VERSION)
1145                 pmd->flower_support = 1;
1146         if (KERNEL_VERSION(ver[0], ver[1], ver[2]) >=
1147             FLOWER_VLAN_KERNEL_VERSION)
1148                 pmd->flower_vlan_support = 1;
1149         return 1;
1150 }
1151
1152 static int
1153 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1154                    char *remote_iface, int fixed_mac_type)
1155 {
1156         int numa_node = rte_socket_id();
1157         struct rte_eth_dev *dev;
1158         struct pmd_internals *pmd;
1159         struct rte_eth_dev_data *data;
1160         int i;
1161
1162         RTE_LOG(DEBUG, PMD, "  TAP device on numa %u\n", rte_socket_id());
1163
1164         data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
1165         if (!data) {
1166                 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
1167                 goto error_exit;
1168         }
1169
1170         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1171         if (!dev) {
1172                 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
1173                 goto error_exit;
1174         }
1175
1176         pmd = dev->data->dev_private;
1177         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1178         pmd->nb_queues = RTE_PMD_TAP_MAX_QUEUES;
1179
1180         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1181         if (pmd->ioctl_sock == -1) {
1182                 RTE_LOG(ERR, PMD,
1183                         "TAP Unable to get a socket for management: %s\n",
1184                         strerror(errno));
1185                 goto error_exit;
1186         }
1187
1188         /* Setup some default values */
1189         rte_memcpy(data, dev->data, sizeof(*data));
1190         data->dev_private = pmd;
1191         data->dev_flags = RTE_ETH_DEV_DETACHABLE | RTE_ETH_DEV_INTR_LSC;
1192         data->numa_node = numa_node;
1193         data->drv_name = pmd_tap_drv.driver.name;
1194
1195         data->dev_link = pmd_link;
1196         data->mac_addrs = &pmd->eth_addr;
1197         data->nb_rx_queues = pmd->nb_queues;
1198         data->nb_tx_queues = pmd->nb_queues;
1199
1200         dev->data = data;
1201         dev->dev_ops = &ops;
1202         dev->rx_pkt_burst = pmd_rx_burst;
1203         dev->tx_pkt_burst = pmd_tx_burst;
1204
1205         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1206         pmd->intr_handle.fd = -1;
1207
1208         /* Presetup the fds to -1 as being not valid */
1209         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1210                 pmd->rxq[i].fd = -1;
1211                 pmd->txq[i].fd = -1;
1212         }
1213
1214         if (fixed_mac_type) {
1215                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1216                 static int iface_idx;
1217                 char mac[ETHER_ADDR_LEN] = "\0dtap";
1218
1219                 mac[ETHER_ADDR_LEN - 1] = iface_idx++;
1220                 rte_memcpy(&pmd->eth_addr, mac, ETHER_ADDR_LEN);
1221         } else {
1222                 eth_random_addr((uint8_t *)&pmd->eth_addr);
1223         }
1224
1225         tap_kernel_support(pmd);
1226         if (!pmd->flower_support)
1227                 return 0;
1228         LIST_INIT(&pmd->flows);
1229         /*
1230          * If no netlink socket can be created, then it will fail when
1231          * creating/destroying flow rules.
1232          */
1233         pmd->nlsk_fd = nl_init(0);
1234         if (strlen(remote_iface)) {
1235                 struct ifreq ifr;
1236
1237                 pmd->remote_if_index = if_nametoindex(remote_iface);
1238                 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1239                          "%s", remote_iface);
1240                 if (!pmd->remote_if_index) {
1241                         RTE_LOG(ERR, PMD, "Could not find %s ifindex: "
1242                                 "remote interface will remain unconfigured\n",
1243                                 remote_iface);
1244                         return 0;
1245                 }
1246                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0)
1247                         goto error_exit;
1248                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1249                            ETHER_ADDR_LEN);
1250         }
1251
1252         return 0;
1253
1254 error_exit:
1255         RTE_LOG(DEBUG, PMD, "TAP Unable to initialize %s\n",
1256                 rte_vdev_device_name(vdev));
1257
1258         rte_free(data);
1259         return -EINVAL;
1260 }
1261
1262 static int
1263 set_interface_name(const char *key __rte_unused,
1264                    const char *value,
1265                    void *extra_args)
1266 {
1267         char *name = (char *)extra_args;
1268
1269         if (value)
1270                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
1271         else
1272                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1273                          DEFAULT_TAP_NAME, (tap_unit - 1));
1274
1275         return 0;
1276 }
1277
1278 static int
1279 set_interface_speed(const char *key __rte_unused,
1280                     const char *value,
1281                     void *extra_args)
1282 {
1283         *(int *)extra_args = (value) ? atoi(value) : ETH_SPEED_NUM_10G;
1284
1285         return 0;
1286 }
1287
1288 static int
1289 set_remote_iface(const char *key __rte_unused,
1290                  const char *value,
1291                  void *extra_args)
1292 {
1293         char *name = (char *)extra_args;
1294
1295         if (value)
1296                 snprintf(name, RTE_ETH_NAME_MAX_LEN, "%s", value);
1297
1298         return 0;
1299 }
1300
1301 static int
1302 set_mac_type(const char *key __rte_unused,
1303              const char *value,
1304              void *extra_args)
1305 {
1306         if (value &&
1307             !strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED)))
1308                 *(int *)extra_args = 1;
1309         return 0;
1310 }
1311
1312 /* Open a TAP interface device.
1313  */
1314 static int
1315 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1316 {
1317         const char *name, *params;
1318         int ret;
1319         struct rte_kvargs *kvlist = NULL;
1320         int speed;
1321         char tap_name[RTE_ETH_NAME_MAX_LEN];
1322         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1323         int fixed_mac_type = 0;
1324
1325         name = rte_vdev_device_name(dev);
1326         params = rte_vdev_device_args(dev);
1327
1328         speed = ETH_SPEED_NUM_10G;
1329         snprintf(tap_name, sizeof(tap_name), "%s%d",
1330                  DEFAULT_TAP_NAME, tap_unit++);
1331         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1332
1333         if (params && (params[0] != '\0')) {
1334                 RTE_LOG(DEBUG, PMD, "paramaters (%s)\n", params);
1335
1336                 kvlist = rte_kvargs_parse(params, valid_arguments);
1337                 if (kvlist) {
1338                         if (rte_kvargs_count(kvlist, ETH_TAP_SPEED_ARG) == 1) {
1339                                 ret = rte_kvargs_process(kvlist,
1340                                                          ETH_TAP_SPEED_ARG,
1341                                                          &set_interface_speed,
1342                                                          &speed);
1343                                 if (ret == -1)
1344                                         goto leave;
1345                         }
1346
1347                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1348                                 ret = rte_kvargs_process(kvlist,
1349                                                          ETH_TAP_IFACE_ARG,
1350                                                          &set_interface_name,
1351                                                          tap_name);
1352                                 if (ret == -1)
1353                                         goto leave;
1354                         }
1355
1356                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1357                                 ret = rte_kvargs_process(kvlist,
1358                                                          ETH_TAP_REMOTE_ARG,
1359                                                          &set_remote_iface,
1360                                                          remote_iface);
1361                                 if (ret == -1)
1362                                         goto leave;
1363                         }
1364
1365                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1366                                 ret = rte_kvargs_process(kvlist,
1367                                                          ETH_TAP_MAC_ARG,
1368                                                          &set_mac_type,
1369                                                          &fixed_mac_type);
1370                                 if (ret == -1)
1371                                         goto leave;
1372                         }
1373                 }
1374         }
1375         pmd_link.link_speed = speed;
1376
1377         RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1378                 name, tap_name);
1379
1380         ret = eth_dev_tap_create(dev, tap_name, remote_iface, fixed_mac_type);
1381
1382 leave:
1383         if (ret == -1) {
1384                 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1385                         name, tap_name);
1386                 tap_unit--;             /* Restore the unit number */
1387         }
1388         rte_kvargs_free(kvlist);
1389
1390         return ret;
1391 }
1392
1393 /* detach a TAP device.
1394  */
1395 static int
1396 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1397 {
1398         struct rte_eth_dev *eth_dev = NULL;
1399         struct pmd_internals *internals;
1400         int i;
1401
1402         RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1403                 rte_socket_id());
1404
1405         /* find the ethdev entry */
1406         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1407         if (!eth_dev)
1408                 return 0;
1409
1410         internals = eth_dev->data->dev_private;
1411         if (internals->flower_support && internals->nlsk_fd) {
1412                 tap_flow_flush(eth_dev, NULL);
1413                 tap_flow_implicit_flush(internals, NULL);
1414                 nl_final(internals->nlsk_fd);
1415         }
1416         for (i = 0; i < internals->nb_queues; i++)
1417                 if (internals->rxq[i].fd != -1)
1418                         close(internals->rxq[i].fd);
1419
1420         close(internals->ioctl_sock);
1421         rte_free(eth_dev->data->dev_private);
1422         rte_free(eth_dev->data);
1423
1424         rte_eth_dev_release_port(eth_dev);
1425
1426         return 0;
1427 }
1428
1429 static struct rte_vdev_driver pmd_tap_drv = {
1430         .probe = rte_pmd_tap_probe,
1431         .remove = rte_pmd_tap_remove,
1432 };
1433 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1434 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1435 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1436                               ETH_TAP_IFACE_ARG "=<string> "
1437                               ETH_TAP_SPEED_ARG "=<int> "
1438                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_FIXED " "
1439                               ETH_TAP_REMOTE_ARG "=<string>");