drivers/net/tap/rte_eth_tap.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2016-2017 Intel Corporation
   3  */
   4
   5 #include <rte_atomic.h>
   6 #include <rte_branch_prediction.h>
   7 #include <rte_byteorder.h>
   8 #include <rte_common.h>
   9 #include <rte_mbuf.h>
  10 #include <rte_ethdev.h>
  11 #include <rte_ethdev_vdev.h>
  12 #include <rte_malloc.h>
  13 #include <rte_bus_vdev.h>
  14 #include <rte_kvargs.h>
  15 #include <rte_net.h>
  16 #include <rte_debug.h>
  17 #include <rte_ip.h>
  18
  19 #include <sys/types.h>
  20 #include <sys/stat.h>
  21 #include <sys/socket.h>
  22 #include <sys/ioctl.h>
  23 #include <sys/utsname.h>
  24 #include <sys/mman.h>
  25 #include <errno.h>
  26 #include <signal.h>
  27 #include <stdint.h>
  28 #include <sys/uio.h>
  29 #include <unistd.h>
  30 #include <arpa/inet.h>
  31 #include <net/if.h>
  32 #include <linux/if_tun.h>
  33 #include <linux/if_ether.h>
  34 #include <fcntl.h>
  35
  36 #include <rte_eth_tap.h>
  37 #include <tap_flow.h>
  38 #include <tap_netlink.h>
  39 #include <tap_tcmsgs.h>
  40
  41 /* Linux based path to the TUN device */
  42 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  43 #define DEFAULT_TAP_NAME        "dtap"
  44
  45 #define ETH_TAP_IFACE_ARG       "iface"
  46 #define ETH_TAP_SPEED_ARG       "speed"
  47 #define ETH_TAP_REMOTE_ARG      "remote"
  48 #define ETH_TAP_MAC_ARG         "mac"
  49 #define ETH_TAP_MAC_FIXED       "fixed"
  50
  51 static struct rte_vdev_driver pmd_tap_drv;
  52
  53 static const char *valid_arguments[] = {
  54         ETH_TAP_IFACE_ARG,
  55         ETH_TAP_SPEED_ARG,
  56         ETH_TAP_REMOTE_ARG,
  57         ETH_TAP_MAC_ARG,
  58         NULL
  59 };
  60
  61 static int tap_unit;
  62
  63 static volatile uint32_t tap_trigger;   /* Rx trigger */
  64
  65 static struct rte_eth_link pmd_link = {
  66         .link_speed = ETH_SPEED_NUM_10G,
  67         .link_duplex = ETH_LINK_FULL_DUPLEX,
  68         .link_status = ETH_LINK_DOWN,
  69         .link_autoneg = ETH_LINK_AUTONEG
  70 };
  71
  72 static void
  73 tap_trigger_cb(int sig __rte_unused)
  74 {
  75         /* Valid trigger values are nonzero */
  76         tap_trigger = (tap_trigger + 1) | 0x80000000;
  77 }
  78
  79 /* Specifies on what netdevices the ioctl should be applied */
  80 enum ioctl_mode {
  81         LOCAL_AND_REMOTE,
  82         LOCAL_ONLY,
  83         REMOTE_ONLY,
  84 };
  85
  86 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
  87
  88 /* Tun/Tap allocation routine
  89  *
  90  * name is the number of the interface to use, unless NULL to take the host
  91  * supplied name.
  92  */
  93 static int
  94 tun_alloc(struct pmd_internals *pmd)
  95 {
  96         struct ifreq ifr;
  97 #ifdef IFF_MULTI_QUEUE
  98         unsigned int features;
  99 #endif
 100         int fd;
 101
 102         memset(&ifr, 0, sizeof(struct ifreq));
 103
 104         /*
 105          * Do not set IFF_NO_PI as packet information header will be needed
 106          * to check if a received packet has been truncated.
 107          */
 108         ifr.ifr_flags = IFF_TAP;
 109         snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
 110
 111         RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
 112
 113         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 114         if (fd < 0) {
 115                 RTE_LOG(ERR, PMD, "Unable to create TAP interface\n");
 116                 goto error;
 117         }
 118
 119 #ifdef IFF_MULTI_QUEUE
 120         /* Grab the TUN features to verify we can work multi-queue */
 121         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 122                 RTE_LOG(ERR, PMD, "TAP unable to get TUN/TAP features\n");
 123                 goto error;
 124         }
 125         RTE_LOG(DEBUG, PMD, "  TAP Features %08x\n", features);
 126
 127         if (features & IFF_MULTI_QUEUE) {
 128                 RTE_LOG(DEBUG, PMD, "  Multi-queue support for %d queues\n",
 129                         RTE_PMD_TAP_MAX_QUEUES);
 130                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 131         } else
 132 #endif
 133         {
 134                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 135                 RTE_LOG(DEBUG, PMD, "  Single queue only support\n");
 136         }
 137
 138         /* Set the TUN/TAP configuration and set the name if needed */
 139         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 140                 RTE_LOG(WARNING, PMD,
 141                         "Unable to set TUNSETIFF for %s\n",
 142                         ifr.ifr_name);
 143                 perror("TUNSETIFF");
 144                 goto error;
 145         }
 146
 147         /* Always set the file descriptor to non-blocking */
 148         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 149                 RTE_LOG(WARNING, PMD,
 150                         "Unable to set %s to nonblocking\n",
 151                         ifr.ifr_name);
 152                 perror("F_SETFL, NONBLOCK");
 153                 goto error;
 154         }
 155
 156         /* Set up trigger to optimize empty Rx bursts */
 157         errno = 0;
 158         do {
 159                 struct sigaction sa;
 160                 int flags = fcntl(fd, F_GETFL);
 161
 162                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 163                         break;
 164                 if (sa.sa_handler != tap_trigger_cb) {
 165                         /*
 166                          * Make sure SIGIO is not already taken. This is done
 167                          * as late as possible to leave the application a
 168                          * chance to set up its own signal handler first.
 169                          */
 170                         if (sa.sa_handler != SIG_IGN &&
 171                             sa.sa_handler != SIG_DFL) {
 172                                 errno = EBUSY;
 173                                 break;
 174                         }
 175                         sa = (struct sigaction){
 176                                 .sa_flags = SA_RESTART,
 177                                 .sa_handler = tap_trigger_cb,
 178                         };
 179                         if (sigaction(SIGIO, &sa, NULL) == -1)
 180                                 break;
 181                 }
 182                 /* Enable SIGIO on file descriptor */
 183                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 184                 fcntl(fd, F_SETOWN, getpid());
 185         } while (0);
 186         if (errno) {
 187                 /* Disable trigger globally in case of error */
 188                 tap_trigger = 0;
 189                 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
 190                         strerror(errno));
 191         }
 192
 193         return fd;
 194
 195 error:
 196         if (fd > 0)
 197                 close(fd);
 198         return -1;
 199 }
 200
 201 static void
 202 tap_verify_csum(struct rte_mbuf *mbuf)
 203 {
 204         uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
 205         uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
 206         uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
 207         unsigned int l2_len = sizeof(struct ether_hdr);
 208         unsigned int l3_len;
 209         uint16_t cksum = 0;
 210         void *l3_hdr;
 211         void *l4_hdr;
 212
 213         if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
 214                 l2_len += 4;
 215         else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
 216                 l2_len += 8;
 217         /* Don't verify checksum for packets with discontinuous L2 header */
 218         if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
 219                      rte_pktmbuf_data_len(mbuf)))
 220                 return;
 221         l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
 222         if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 223                 struct ipv4_hdr *iph = l3_hdr;
 224
 225                 /* ihl contains the number of 4-byte words in the header */
 226                 l3_len = 4 * (iph->version_ihl & 0xf);
 227                 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
 228                         return;
 229
 230                 cksum = ~rte_raw_cksum(iph, l3_len);
 231                 mbuf->ol_flags |= cksum ?
 232                         PKT_RX_IP_CKSUM_BAD :
 233                         PKT_RX_IP_CKSUM_GOOD;
 234         } else if (l3 == RTE_PTYPE_L3_IPV6) {
 235                 l3_len = sizeof(struct ipv6_hdr);
 236         } else {
 237                 /* IPv6 extensions are not supported */
 238                 return;
 239         }
 240         if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
 241                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 242                 /* Don't verify checksum for multi-segment packets. */
 243                 if (mbuf->nb_segs > 1)
 244                         return;
 245                 if (l3 == RTE_PTYPE_L3_IPV4)
 246                         cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
 247                 else if (l3 == RTE_PTYPE_L3_IPV6)
 248                         cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
 249                 mbuf->ol_flags |= cksum ?
 250                         PKT_RX_L4_CKSUM_BAD :
 251                         PKT_RX_L4_CKSUM_GOOD;
 252         }
 253 }
 254
 255 /* Callback to handle the rx burst of packets to the correct interface and
 256  * file descriptor(s) in a multi-queue setup.
 257  */
 258 static uint16_t
 259 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 260 {
 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                 struct rte_mbuf *mbuf = rxq->pool;
 273                 struct rte_mbuf *seg = NULL;
 274                 struct rte_mbuf *new_tail = NULL;
 275                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 276                 int len;
 277
 278                 len = readv(rxq->fd, *rxq->iovecs,
 279                             1 + (rxq->rxmode->enable_scatter ?
 280                                  rxq->nb_rx_desc : 1));
 281                 if (len < (int)sizeof(struct tun_pi))
 282                         break;
 283
 284                 /* Packet couldn't fit in the provided mbuf */
 285                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 286                         rxq->stats.ierrors++;
 287                         continue;
 288                 }
 289
 290                 len -= sizeof(struct tun_pi);
 291
 292                 mbuf->pkt_len = len;
 293                 mbuf->port = rxq->in_port;
 294                 while (1) {
 295                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 296
 297                         if (unlikely(!buf)) {
 298                                 rxq->stats.rx_nombuf++;
 299                                 /* No new buf has been allocated: do nothing */
 300                                 if (!new_tail || !seg)
 301                                         goto end;
 302
 303                                 seg->next = NULL;
 304                                 rte_pktmbuf_free(mbuf);
 305
 306                                 goto end;
 307                         }
 308                         seg = seg ? seg->next : mbuf;
 309                         if (rxq->pool == mbuf)
 310                                 rxq->pool = buf;
 311                         if (new_tail)
 312                                 new_tail->next = buf;
 313                         new_tail = buf;
 314                         new_tail->next = seg->next;
 315
 316                         /* iovecs[0] is reserved for packet info (pi) */
 317                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 318                                 buf->buf_len - data_off;
 319                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 320                                 (char *)buf->buf_addr + data_off;
 321
 322                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 323                         seg->data_off = data_off;
 324
 325                         len -= seg->data_len;
 326                         if (len <= 0)
 327                                 break;
 328                         mbuf->nb_segs++;
 329                         /* First segment has headroom, not the others */
 330                         data_off = 0;
 331                 }
 332                 seg->next = NULL;
 333                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 334                                                       RTE_PTYPE_ALL_MASK);
 335                 if (rxq->rxmode->hw_ip_checksum)
 336                         tap_verify_csum(mbuf);
 337
 338                 /* account for the receive frame */
 339                 bufs[num_rx++] = mbuf;
 340                 num_rx_bytes += mbuf->pkt_len;
 341         }
 342 end:
 343         rxq->stats.ipackets += num_rx;
 344         rxq->stats.ibytes += num_rx_bytes;
 345
 346         return num_rx;
 347 }
 348
 349 static void
 350 tap_tx_offload(char *packet, uint64_t ol_flags, unsigned int l2_len,
 351                unsigned int l3_len)
 352 {
 353         void *l3_hdr = packet + l2_len;
 354
 355         if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
 356                 struct ipv4_hdr *iph = l3_hdr;
 357                 uint16_t cksum;
 358
 359                 iph->hdr_checksum = 0;
 360                 cksum = rte_raw_cksum(iph, l3_len);
 361                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 362         }
 363         if (ol_flags & PKT_TX_L4_MASK) {
 364                 uint16_t l4_len;
 365                 uint32_t cksum;
 366                 uint16_t *l4_cksum;
 367                 void *l4_hdr;
 368
 369                 l4_hdr = packet + l2_len + l3_len;
 370                 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
 371                         l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
 372                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 373                         l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
 374                 else
 375                         return;
 376                 *l4_cksum = 0;
 377                 if (ol_flags & PKT_TX_IPV4) {
 378                         struct ipv4_hdr *iph = l3_hdr;
 379
 380                         l4_len = rte_be_to_cpu_16(iph->total_length) - l3_len;
 381                         cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 382                 } else {
 383                         struct ipv6_hdr *ip6h = l3_hdr;
 384
 385                         /* payload_len does not include ext headers */
 386                         l4_len = rte_be_to_cpu_16(ip6h->payload_len) -
 387                                 l3_len + sizeof(struct ipv6_hdr);
 388                         cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 389                 }
 390                 cksum += rte_raw_cksum(l4_hdr, l4_len);
 391                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 392                 cksum = (~cksum) & 0xffff;
 393                 if (cksum == 0)
 394                         cksum = 0xffff;
 395                 *l4_cksum = cksum;
 396         }
 397 }
 398
 399 /* Callback to handle sending packets from the tap interface
 400  */
 401 static uint16_t
 402 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 403 {
 404         struct tx_queue *txq = queue;
 405         uint16_t num_tx = 0;
 406         unsigned long num_tx_bytes = 0;
 407         uint32_t max_size;
 408         int i;
 409
 410         if (unlikely(nb_pkts == 0))
 411                 return 0;
 412
 413         max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
 414         for (i = 0; i < nb_pkts; i++) {
 415                 struct rte_mbuf *mbuf = bufs[num_tx];
 416                 struct iovec iovecs[mbuf->nb_segs + 1];
 417                 struct tun_pi pi = { .flags = 0 };
 418                 struct rte_mbuf *seg = mbuf;
 419                 char m_copy[mbuf->data_len];
 420                 int n;
 421                 int j;
 422
 423                 /* stats.errs will be incremented */
 424                 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
 425                         break;
 426
 427                 iovecs[0].iov_base = &pi;
 428                 iovecs[0].iov_len = sizeof(pi);
 429                 for (j = 1; j <= mbuf->nb_segs; j++) {
 430                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 431                         iovecs[j].iov_base =
 432                                 rte_pktmbuf_mtod(seg, void *);
 433                         seg = seg->next;
 434                 }
 435                 if (mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 436                     (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 437                     (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM) {
 438                         /* Support only packets with all data in the same seg */
 439                         if (mbuf->nb_segs > 1)
 440                                 break;
 441                         /* To change checksums, work on a copy of data. */
 442                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 443                                    rte_pktmbuf_data_len(mbuf));
 444                         tap_tx_offload(m_copy, mbuf->ol_flags,
 445                                        mbuf->l2_len, mbuf->l3_len);
 446                         iovecs[1].iov_base = m_copy;
 447                 }
 448                 /* copy the tx frame data */
 449                 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
 450                 if (n <= 0)
 451                         break;
 452
 453                 num_tx++;
 454                 num_tx_bytes += mbuf->pkt_len;
 455                 rte_pktmbuf_free(mbuf);
 456         }
 457
 458         txq->stats.opackets += num_tx;
 459         txq->stats.errs += nb_pkts - num_tx;
 460         txq->stats.obytes += num_tx_bytes;
 461
 462         return num_tx;
 463 }
 464
 465 static const char *
 466 tap_ioctl_req2str(unsigned long request)
 467 {
 468         switch (request) {
 469         case SIOCSIFFLAGS:
 470                 return "SIOCSIFFLAGS";
 471         case SIOCGIFFLAGS:
 472                 return "SIOCGIFFLAGS";
 473         case SIOCGIFHWADDR:
 474                 return "SIOCGIFHWADDR";
 475         case SIOCSIFHWADDR:
 476                 return "SIOCSIFHWADDR";
 477         case SIOCSIFMTU:
 478                 return "SIOCSIFMTU";
 479         }
 480         return "UNKNOWN";
 481 }
 482
 483 static int
 484 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 485           struct ifreq *ifr, int set, enum ioctl_mode mode)
 486 {
 487         short req_flags = ifr->ifr_flags;
 488         int remote = pmd->remote_if_index &&
 489                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 490
 491         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 492                 return 0;
 493         /*
 494          * If there is a remote netdevice, apply ioctl on it, then apply it on
 495          * the tap netdevice.
 496          */
 497 apply:
 498         if (remote)
 499                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
 500         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 501                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 502         switch (request) {
 503         case SIOCSIFFLAGS:
 504                 /* fetch current flags to leave other flags untouched */
 505                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 506                         goto error;
 507                 if (set)
 508                         ifr->ifr_flags |= req_flags;
 509                 else
 510                         ifr->ifr_flags &= ~req_flags;
 511                 break;
 512         case SIOCGIFFLAGS:
 513         case SIOCGIFHWADDR:
 514         case SIOCSIFHWADDR:
 515         case SIOCSIFMTU:
 516                 break;
 517         default:
 518                 RTE_ASSERT(!"unsupported request type: must not happen");
 519         }
 520         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 521                 goto error;
 522         if (remote-- && mode == LOCAL_AND_REMOTE)
 523                 goto apply;
 524         return 0;
 525
 526 error:
 527         RTE_LOG(DEBUG, PMD, "%s: %s(%s) failed: %s(%d)\n", ifr->ifr_name,
 528                 __func__, tap_ioctl_req2str(request), strerror(errno), errno);
 529         return -errno;
 530 }
 531
 532 static int
 533 tap_link_set_down(struct rte_eth_dev *dev)
 534 {
 535         struct pmd_internals *pmd = dev->data->dev_private;
 536         struct ifreq ifr = { .ifr_flags = IFF_UP };
 537
 538         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 539         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 540 }
 541
 542 static int
 543 tap_link_set_up(struct rte_eth_dev *dev)
 544 {
 545         struct pmd_internals *pmd = dev->data->dev_private;
 546         struct ifreq ifr = { .ifr_flags = IFF_UP };
 547
 548         dev->data->dev_link.link_status = ETH_LINK_UP;
 549         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 550 }
 551
 552 static int
 553 tap_dev_start(struct rte_eth_dev *dev)
 554 {
 555         int err;
 556
 557         err = tap_intr_handle_set(dev, 1);
 558         if (err)
 559                 return err;
 560         return tap_link_set_up(dev);
 561 }
 562
 563 /* This function gets called when the current port gets stopped.
 564  */
 565 static void
 566 tap_dev_stop(struct rte_eth_dev *dev)
 567 {
 568         tap_intr_handle_set(dev, 0);
 569         tap_link_set_down(dev);
 570 }
 571
 572 static int
 573 tap_dev_configure(struct rte_eth_dev *dev)
 574 {
 575         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 576                 RTE_LOG(ERR, PMD,
 577                         "%s: number of rx queues %d exceeds max num of queues %d\n",
 578                         dev->device->name,
 579                         dev->data->nb_rx_queues,
 580                         RTE_PMD_TAP_MAX_QUEUES);
 581                 return -1;
 582         }
 583         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 584                 RTE_LOG(ERR, PMD,
 585                         "%s: number of tx queues %d exceeds max num of queues %d\n",
 586                         dev->device->name,
 587                         dev->data->nb_tx_queues,
 588                         RTE_PMD_TAP_MAX_QUEUES);
 589                 return -1;
 590         }
 591
 592         RTE_LOG(INFO, PMD, "%s: %p: TX configured queues number: %u\n",
 593              dev->device->name, (void *)dev, dev->data->nb_tx_queues);
 594
 595         RTE_LOG(INFO, PMD, "%s: %p: RX configured queues number: %u\n",
 596              dev->device->name, (void *)dev, dev->data->nb_rx_queues);
 597
 598         return 0;
 599 }
 600
 601 static uint32_t
 602 tap_dev_speed_capa(void)
 603 {
 604         uint32_t speed = pmd_link.link_speed;
 605         uint32_t capa = 0;
 606
 607         if (speed >= ETH_SPEED_NUM_10M)
 608                 capa |= ETH_LINK_SPEED_10M;
 609         if (speed >= ETH_SPEED_NUM_100M)
 610                 capa |= ETH_LINK_SPEED_100M;
 611         if (speed >= ETH_SPEED_NUM_1G)
 612                 capa |= ETH_LINK_SPEED_1G;
 613         if (speed >= ETH_SPEED_NUM_5G)
 614                 capa |= ETH_LINK_SPEED_2_5G;
 615         if (speed >= ETH_SPEED_NUM_5G)
 616                 capa |= ETH_LINK_SPEED_5G;
 617         if (speed >= ETH_SPEED_NUM_10G)
 618                 capa |= ETH_LINK_SPEED_10G;
 619         if (speed >= ETH_SPEED_NUM_20G)
 620                 capa |= ETH_LINK_SPEED_20G;
 621         if (speed >= ETH_SPEED_NUM_25G)
 622                 capa |= ETH_LINK_SPEED_25G;
 623         if (speed >= ETH_SPEED_NUM_40G)
 624                 capa |= ETH_LINK_SPEED_40G;
 625         if (speed >= ETH_SPEED_NUM_50G)
 626                 capa |= ETH_LINK_SPEED_50G;
 627         if (speed >= ETH_SPEED_NUM_56G)
 628                 capa |= ETH_LINK_SPEED_56G;
 629         if (speed >= ETH_SPEED_NUM_100G)
 630                 capa |= ETH_LINK_SPEED_100G;
 631
 632         return capa;
 633 }
 634
 635 static void
 636 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 637 {
 638         struct pmd_internals *internals = dev->data->dev_private;
 639
 640         dev_info->if_index = internals->if_index;
 641         dev_info->max_mac_addrs = 1;
 642         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 643         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
 644         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
 645         dev_info->min_rx_bufsize = 0;
 646         dev_info->pci_dev = NULL;
 647         dev_info->speed_capa = tap_dev_speed_capa();
 648         dev_info->rx_offload_capa = (DEV_RX_OFFLOAD_IPV4_CKSUM |
 649                                      DEV_RX_OFFLOAD_UDP_CKSUM |
 650                                      DEV_RX_OFFLOAD_TCP_CKSUM);
 651         dev_info->tx_offload_capa =
 652                 (DEV_TX_OFFLOAD_IPV4_CKSUM |
 653                  DEV_TX_OFFLOAD_UDP_CKSUM |
 654                  DEV_TX_OFFLOAD_TCP_CKSUM);
 655 }
 656
 657 static int
 658 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 659 {
 660         unsigned int i, imax;
 661         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 662         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 663         unsigned long rx_nombuf = 0, ierrors = 0;
 664         const struct pmd_internals *pmd = dev->data->dev_private;
 665
 666         /* rx queue statistics */
 667         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 668                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 669         for (i = 0; i < imax; i++) {
 670                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 671                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 672                 rx_total += tap_stats->q_ipackets[i];
 673                 rx_bytes_total += tap_stats->q_ibytes[i];
 674                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
 675                 ierrors += pmd->rxq[i].stats.ierrors;
 676         }
 677
 678         /* tx queue statistics */
 679         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 680                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 681
 682         for (i = 0; i < imax; i++) {
 683                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 684                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 685                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 686                 tx_total += tap_stats->q_opackets[i];
 687                 tx_err_total += tap_stats->q_errors[i];
 688                 tx_bytes_total += tap_stats->q_obytes[i];
 689         }
 690
 691         tap_stats->ipackets = rx_total;
 692         tap_stats->ibytes = rx_bytes_total;
 693         tap_stats->ierrors = ierrors;
 694         tap_stats->rx_nombuf = rx_nombuf;
 695         tap_stats->opackets = tx_total;
 696         tap_stats->oerrors = tx_err_total;
 697         tap_stats->obytes = tx_bytes_total;
 698         return 0;
 699 }
 700
 701 static void
 702 tap_stats_reset(struct rte_eth_dev *dev)
 703 {
 704         int i;
 705         struct pmd_internals *pmd = dev->data->dev_private;
 706
 707         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 708                 pmd->rxq[i].stats.ipackets = 0;
 709                 pmd->rxq[i].stats.ibytes = 0;
 710                 pmd->rxq[i].stats.ierrors = 0;
 711                 pmd->rxq[i].stats.rx_nombuf = 0;
 712
 713                 pmd->txq[i].stats.opackets = 0;
 714                 pmd->txq[i].stats.errs = 0;
 715                 pmd->txq[i].stats.obytes = 0;
 716         }
 717 }
 718
 719 static void
 720 tap_dev_close(struct rte_eth_dev *dev)
 721 {
 722         int i;
 723         struct pmd_internals *internals = dev->data->dev_private;
 724
 725         tap_link_set_down(dev);
 726         tap_flow_flush(dev, NULL);
 727         tap_flow_implicit_flush(internals, NULL);
 728
 729         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 730                 if (internals->rxq[i].fd != -1) {
 731                         close(internals->rxq[i].fd);
 732                         internals->rxq[i].fd = -1;
 733                 }
 734                 if (internals->txq[i].fd != -1) {
 735                         close(internals->txq[i].fd);
 736                         internals->txq[i].fd = -1;
 737                 }
 738         }
 739
 740         if (internals->remote_if_index) {
 741                 /* Restore initial remote state */
 742                 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
 743                                 &internals->remote_initial_flags);
 744         }
 745 }
 746
 747 static void
 748 tap_rx_queue_release(void *queue)
 749 {
 750         struct rx_queue *rxq = queue;
 751
 752         if (rxq && (rxq->fd > 0)) {
 753                 close(rxq->fd);
 754                 rxq->fd = -1;
 755                 rte_pktmbuf_free(rxq->pool);
 756                 rte_free(rxq->iovecs);
 757                 rxq->pool = NULL;
 758                 rxq->iovecs = NULL;
 759         }
 760 }
 761
 762 static void
 763 tap_tx_queue_release(void *queue)
 764 {
 765         struct tx_queue *txq = queue;
 766
 767         if (txq && (txq->fd > 0)) {
 768                 close(txq->fd);
 769                 txq->fd = -1;
 770         }
 771 }
 772
 773 static int
 774 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
 775 {
 776         struct rte_eth_link *dev_link = &dev->data->dev_link;
 777         struct pmd_internals *pmd = dev->data->dev_private;
 778         struct ifreq ifr = { .ifr_flags = 0 };
 779
 780         if (pmd->remote_if_index) {
 781                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
 782                 if (!(ifr.ifr_flags & IFF_UP) ||
 783                     !(ifr.ifr_flags & IFF_RUNNING)) {
 784                         dev_link->link_status = ETH_LINK_DOWN;
 785                         return 0;
 786                 }
 787         }
 788         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
 789         dev_link->link_status =
 790                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
 791                  ETH_LINK_UP :
 792                  ETH_LINK_DOWN);
 793         return 0;
 794 }
 795
 796 static void
 797 tap_promisc_enable(struct rte_eth_dev *dev)
 798 {
 799         struct pmd_internals *pmd = dev->data->dev_private;
 800         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 801
 802         dev->data->promiscuous = 1;
 803         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 804         if (pmd->remote_if_index && !pmd->flow_isolate)
 805                 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
 806 }
 807
 808 static void
 809 tap_promisc_disable(struct rte_eth_dev *dev)
 810 {
 811         struct pmd_internals *pmd = dev->data->dev_private;
 812         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 813
 814         dev->data->promiscuous = 0;
 815         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 816         if (pmd->remote_if_index && !pmd->flow_isolate)
 817                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
 818 }
 819
 820 static void
 821 tap_allmulti_enable(struct rte_eth_dev *dev)
 822 {
 823         struct pmd_internals *pmd = dev->data->dev_private;
 824         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 825
 826         dev->data->all_multicast = 1;
 827         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 828         if (pmd->remote_if_index && !pmd->flow_isolate)
 829                 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
 830 }
 831
 832 static void
 833 tap_allmulti_disable(struct rte_eth_dev *dev)
 834 {
 835         struct pmd_internals *pmd = dev->data->dev_private;
 836         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 837
 838         dev->data->all_multicast = 0;
 839         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 840         if (pmd->remote_if_index && !pmd->flow_isolate)
 841                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
 842 }
 843
 844 static void
 845 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 846 {
 847         struct pmd_internals *pmd = dev->data->dev_private;
 848         enum ioctl_mode mode = LOCAL_ONLY;
 849         struct ifreq ifr;
 850
 851         if (is_zero_ether_addr(mac_addr)) {
 852                 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
 853                         dev->device->name);
 854                 return;
 855         }
 856         /* Check the actual current MAC address on the tap netdevice */
 857         if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
 858                 return;
 859         if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 860                                mac_addr))
 861                 return;
 862         /* Check the current MAC address on the remote */
 863         if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0)
 864                 return;
 865         if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 866                                mac_addr))
 867                 mode = LOCAL_AND_REMOTE;
 868         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
 869         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
 870         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode) < 0)
 871                 return;
 872         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
 873         if (pmd->remote_if_index && !pmd->flow_isolate) {
 874                 /* Replace MAC redirection rule after a MAC change */
 875                 if (tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC) < 0) {
 876                         RTE_LOG(ERR, PMD,
 877                                 "%s: Couldn't delete MAC redirection rule\n",
 878                                 dev->device->name);
 879                         return;
 880                 }
 881                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0)
 882                         RTE_LOG(ERR, PMD,
 883                                 "%s: Couldn't add MAC redirection rule\n",
 884                                 dev->device->name);
 885         }
 886 }
 887
 888 static int
 889 tap_setup_queue(struct rte_eth_dev *dev,
 890                 struct pmd_internals *internals,
 891                 uint16_t qid,
 892                 int is_rx)
 893 {
 894         int *fd;
 895         int *other_fd;
 896         const char *dir;
 897         struct pmd_internals *pmd = dev->data->dev_private;
 898         struct rx_queue *rx = &internals->rxq[qid];
 899         struct tx_queue *tx = &internals->txq[qid];
 900
 901         if (is_rx) {
 902                 fd = &rx->fd;
 903                 other_fd = &tx->fd;
 904                 dir = "rx";
 905         } else {
 906                 fd = &tx->fd;
 907                 other_fd = &rx->fd;
 908                 dir = "tx";
 909         }
 910         if (*fd != -1) {
 911                 /* fd for this queue already exists */
 912                 RTE_LOG(DEBUG, PMD, "%s: fd %d for %s queue qid %d exists\n",
 913                         pmd->name, *fd, dir, qid);
 914         } else if (*other_fd != -1) {
 915                 /* Only other_fd exists. dup it */
 916                 *fd = dup(*other_fd);
 917                 if (*fd < 0) {
 918                         *fd = -1;
 919                         RTE_LOG(ERR, PMD, "%s: dup() failed.\n",
 920                                 pmd->name);
 921                         return -1;
 922                 }
 923                 RTE_LOG(DEBUG, PMD, "%s: dup fd %d for %s queue qid %d (%d)\n",
 924                         pmd->name, *other_fd, dir, qid, *fd);
 925         } else {
 926                 /* Both RX and TX fds do not exist (equal -1). Create fd */
 927                 *fd = tun_alloc(pmd);
 928                 if (*fd < 0) {
 929                         *fd = -1; /* restore original value */
 930                         RTE_LOG(ERR, PMD, "%s: tun_alloc() failed.\n",
 931                                 pmd->name);
 932                         return -1;
 933                 }
 934                 RTE_LOG(DEBUG, PMD, "%s: add %s queue for qid %d fd %d\n",
 935                         pmd->name, dir, qid, *fd);
 936         }
 937
 938         tx->mtu = &dev->data->mtu;
 939         rx->rxmode = &dev->data->dev_conf.rxmode;
 940
 941         return *fd;
 942 }
 943
 944 static int
 945 tap_rx_queue_setup(struct rte_eth_dev *dev,
 946                    uint16_t rx_queue_id,
 947                    uint16_t nb_rx_desc,
 948                    unsigned int socket_id,
 949                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 950                    struct rte_mempool *mp)
 951 {
 952         struct pmd_internals *internals = dev->data->dev_private;
 953         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
 954         struct rte_mbuf **tmp = &rxq->pool;
 955         long iov_max = sysconf(_SC_IOV_MAX);
 956         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
 957         struct iovec (*iovecs)[nb_desc + 1];
 958         int data_off = RTE_PKTMBUF_HEADROOM;
 959         int ret = 0;
 960         int fd;
 961         int i;
 962
 963         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
 964                 RTE_LOG(WARNING, PMD,
 965                         "nb_rx_queues %d too small or mempool NULL\n",
 966                         dev->data->nb_rx_queues);
 967                 return -1;
 968         }
 969
 970         rxq->mp = mp;
 971         rxq->trigger_seen = 1; /* force initial burst */
 972         rxq->in_port = dev->data->port_id;
 973         rxq->nb_rx_desc = nb_desc;
 974         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
 975                                     socket_id);
 976         if (!iovecs) {
 977                 RTE_LOG(WARNING, PMD,
 978                         "%s: Couldn't allocate %d RX descriptors\n",
 979                         dev->device->name, nb_desc);
 980                 return -ENOMEM;
 981         }
 982         rxq->iovecs = iovecs;
 983
 984         dev->data->rx_queues[rx_queue_id] = rxq;
 985         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
 986         if (fd == -1) {
 987                 ret = fd;
 988                 goto error;
 989         }
 990
 991         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
 992         (*rxq->iovecs)[0].iov_base = &rxq->pi;
 993
 994         for (i = 1; i <= nb_desc; i++) {
 995                 *tmp = rte_pktmbuf_alloc(rxq->mp);
 996                 if (!*tmp) {
 997                         RTE_LOG(WARNING, PMD,
 998                                 "%s: couldn't allocate memory for queue %d\n",
 999                                 dev->device->name, rx_queue_id);
1000                         ret = -ENOMEM;
1001                         goto error;
1002                 }
1003                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1004                 (*rxq->iovecs)[i].iov_base =
1005                         (char *)(*tmp)->buf_addr + data_off;
1006                 data_off = 0;
1007                 tmp = &(*tmp)->next;
1008         }
1009
1010         RTE_LOG(DEBUG, PMD, "  RX TAP device name %s, qid %d on fd %d\n",
1011                 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
1012
1013         return 0;
1014
1015 error:
1016         rte_pktmbuf_free(rxq->pool);
1017         rxq->pool = NULL;
1018         rte_free(rxq->iovecs);
1019         rxq->iovecs = NULL;
1020         return ret;
1021 }
1022
1023 static int
1024 tap_tx_queue_setup(struct rte_eth_dev *dev,
1025                    uint16_t tx_queue_id,
1026                    uint16_t nb_tx_desc __rte_unused,
1027                    unsigned int socket_id __rte_unused,
1028                    const struct rte_eth_txconf *tx_conf __rte_unused)
1029 {
1030         struct pmd_internals *internals = dev->data->dev_private;
1031         int ret;
1032
1033         if (tx_queue_id >= dev->data->nb_tx_queues)
1034                 return -1;
1035
1036         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1037         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1038         if (ret == -1)
1039                 return -1;
1040
1041         RTE_LOG(DEBUG, PMD, "  TX TAP device name %s, qid %d on fd %d\n",
1042                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd);
1043
1044         return 0;
1045 }
1046
1047 static int
1048 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1049 {
1050         struct pmd_internals *pmd = dev->data->dev_private;
1051         struct ifreq ifr = { .ifr_mtu = mtu };
1052         int err = 0;
1053
1054         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1055         if (!err)
1056                 dev->data->mtu = mtu;
1057
1058         return err;
1059 }
1060
1061 static int
1062 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1063                      struct ether_addr *mc_addr_set __rte_unused,
1064                      uint32_t nb_mc_addr __rte_unused)
1065 {
1066         /*
1067          * Nothing to do actually: the tap has no filtering whatsoever, every
1068          * packet is received.
1069          */
1070         return 0;
1071 }
1072
1073 static int
1074 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1075 {
1076         struct rte_eth_dev *dev = arg;
1077         struct pmd_internals *pmd = dev->data->dev_private;
1078         struct ifinfomsg *info = NLMSG_DATA(nh);
1079
1080         if (nh->nlmsg_type != RTM_NEWLINK ||
1081             (info->ifi_index != pmd->if_index &&
1082              info->ifi_index != pmd->remote_if_index))
1083                 return 0;
1084         return tap_link_update(dev, 0);
1085 }
1086
1087 static void
1088 tap_dev_intr_handler(void *cb_arg)
1089 {
1090         struct rte_eth_dev *dev = cb_arg;
1091         struct pmd_internals *pmd = dev->data->dev_private;
1092
1093         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1094 }
1095
1096 static int
1097 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1098 {
1099         struct pmd_internals *pmd = dev->data->dev_private;
1100
1101         /* In any case, disable interrupt if the conf is no longer there. */
1102         if (!dev->data->dev_conf.intr_conf.lsc) {
1103                 if (pmd->intr_handle.fd != -1) {
1104                         tap_nl_final(pmd->intr_handle.fd);
1105                         rte_intr_callback_unregister(&pmd->intr_handle,
1106                                 tap_dev_intr_handler, dev);
1107                 }
1108                 return 0;
1109         }
1110         if (set) {
1111                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1112                 if (unlikely(pmd->intr_handle.fd == -1))
1113                         return -EBADF;
1114                 return rte_intr_callback_register(
1115                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1116         }
1117         tap_nl_final(pmd->intr_handle.fd);
1118         return rte_intr_callback_unregister(&pmd->intr_handle,
1119                                             tap_dev_intr_handler, dev);
1120 }
1121
1122 static const uint32_t*
1123 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1124 {
1125         static const uint32_t ptypes[] = {
1126                 RTE_PTYPE_INNER_L2_ETHER,
1127                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1128                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1129                 RTE_PTYPE_INNER_L3_IPV4,
1130                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1131                 RTE_PTYPE_INNER_L3_IPV6,
1132                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1133                 RTE_PTYPE_INNER_L4_FRAG,
1134                 RTE_PTYPE_INNER_L4_UDP,
1135                 RTE_PTYPE_INNER_L4_TCP,
1136                 RTE_PTYPE_INNER_L4_SCTP,
1137                 RTE_PTYPE_L2_ETHER,
1138                 RTE_PTYPE_L2_ETHER_VLAN,
1139                 RTE_PTYPE_L2_ETHER_QINQ,
1140                 RTE_PTYPE_L3_IPV4,
1141                 RTE_PTYPE_L3_IPV4_EXT,
1142                 RTE_PTYPE_L3_IPV6_EXT,
1143                 RTE_PTYPE_L3_IPV6,
1144                 RTE_PTYPE_L4_FRAG,
1145                 RTE_PTYPE_L4_UDP,
1146                 RTE_PTYPE_L4_TCP,
1147                 RTE_PTYPE_L4_SCTP,
1148         };
1149
1150         return ptypes;
1151 }
1152
1153 static int
1154 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1155                   struct rte_eth_fc_conf *fc_conf)
1156 {
1157         fc_conf->mode = RTE_FC_NONE;
1158         return 0;
1159 }
1160
1161 static int
1162 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1163                   struct rte_eth_fc_conf *fc_conf)
1164 {
1165         if (fc_conf->mode != RTE_FC_NONE)
1166                 return -ENOTSUP;
1167         return 0;
1168 }
1169
1170 static const struct eth_dev_ops ops = {
1171         .dev_start              = tap_dev_start,
1172         .dev_stop               = tap_dev_stop,
1173         .dev_close              = tap_dev_close,
1174         .dev_configure          = tap_dev_configure,
1175         .dev_infos_get          = tap_dev_info,
1176         .rx_queue_setup         = tap_rx_queue_setup,
1177         .tx_queue_setup         = tap_tx_queue_setup,
1178         .rx_queue_release       = tap_rx_queue_release,
1179         .tx_queue_release       = tap_tx_queue_release,
1180         .flow_ctrl_get          = tap_flow_ctrl_get,
1181         .flow_ctrl_set          = tap_flow_ctrl_set,
1182         .link_update            = tap_link_update,
1183         .dev_set_link_up        = tap_link_set_up,
1184         .dev_set_link_down      = tap_link_set_down,
1185         .promiscuous_enable     = tap_promisc_enable,
1186         .promiscuous_disable    = tap_promisc_disable,
1187         .allmulticast_enable    = tap_allmulti_enable,
1188         .allmulticast_disable   = tap_allmulti_disable,
1189         .mac_addr_set           = tap_mac_set,
1190         .mtu_set                = tap_mtu_set,
1191         .set_mc_addr_list       = tap_set_mc_addr_list,
1192         .stats_get              = tap_stats_get,
1193         .stats_reset            = tap_stats_reset,
1194         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1195         .filter_ctrl            = tap_dev_filter_ctrl,
1196 };
1197
1198 static int
1199 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1200                    char *remote_iface, int fixed_mac_type)
1201 {
1202         int numa_node = rte_socket_id();
1203         struct rte_eth_dev *dev;
1204         struct pmd_internals *pmd;
1205         struct rte_eth_dev_data *data;
1206         struct ifreq ifr;
1207         int i;
1208
1209         RTE_LOG(DEBUG, PMD, "  TAP device on numa %u\n", rte_socket_id());
1210
1211         data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
1212         if (!data) {
1213                 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
1214                 goto error_exit;
1215         }
1216
1217         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1218         if (!dev) {
1219                 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
1220                 goto error_exit;
1221         }
1222
1223         pmd = dev->data->dev_private;
1224         pmd->dev = dev;
1225         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1226
1227         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1228         if (pmd->ioctl_sock == -1) {
1229                 RTE_LOG(ERR, PMD,
1230                         "TAP Unable to get a socket for management: %s\n",
1231                         strerror(errno));
1232                 goto error_exit;
1233         }
1234
1235         /* Setup some default values */
1236         rte_memcpy(data, dev->data, sizeof(*data));
1237         data->dev_private = pmd;
1238         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1239         data->numa_node = numa_node;
1240
1241         data->dev_link = pmd_link;
1242         data->mac_addrs = &pmd->eth_addr;
1243         /* Set the number of RX and TX queues */
1244         data->nb_rx_queues = 0;
1245         data->nb_tx_queues = 0;
1246
1247         dev->data = data;
1248         dev->dev_ops = &ops;
1249         dev->rx_pkt_burst = pmd_rx_burst;
1250         dev->tx_pkt_burst = pmd_tx_burst;
1251
1252         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1253         pmd->intr_handle.fd = -1;
1254
1255         /* Presetup the fds to -1 as being not valid */
1256         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1257                 pmd->rxq[i].fd = -1;
1258                 pmd->txq[i].fd = -1;
1259         }
1260
1261         if (fixed_mac_type) {
1262                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1263                 static int iface_idx;
1264                 char mac[ETHER_ADDR_LEN] = "\0dtap";
1265
1266                 mac[ETHER_ADDR_LEN - 1] = iface_idx++;
1267                 rte_memcpy(&pmd->eth_addr, mac, ETHER_ADDR_LEN);
1268         } else {
1269                 eth_random_addr((uint8_t *)&pmd->eth_addr);
1270         }
1271
1272         /* Immediately create the netdevice (this will create the 1st queue). */
1273         /* rx queue */
1274         if (tap_setup_queue(dev, pmd, 0, 1) == -1)
1275                 goto error_exit;
1276         /* tx queue */
1277         if (tap_setup_queue(dev, pmd, 0, 0) == -1)
1278                 goto error_exit;
1279
1280         ifr.ifr_mtu = dev->data->mtu;
1281         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1282                 goto error_exit;
1283
1284         memset(&ifr, 0, sizeof(struct ifreq));
1285         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1286         rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr, ETHER_ADDR_LEN);
1287         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1288                 goto error_exit;
1289
1290         /*
1291          * Set up everything related to rte_flow:
1292          * - netlink socket
1293          * - tap / remote if_index
1294          * - mandatory QDISCs
1295          * - rte_flow actual/implicit lists
1296          * - implicit rules
1297          */
1298         pmd->nlsk_fd = tap_nl_init(0);
1299         if (pmd->nlsk_fd == -1) {
1300                 RTE_LOG(WARNING, PMD, "%s: failed to create netlink socket.\n",
1301                         pmd->name);
1302                 goto disable_rte_flow;
1303         }
1304         pmd->if_index = if_nametoindex(pmd->name);
1305         if (!pmd->if_index) {
1306                 RTE_LOG(ERR, PMD, "%s: failed to get if_index.\n", pmd->name);
1307                 goto disable_rte_flow;
1308         }
1309         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1310                 RTE_LOG(ERR, PMD, "%s: failed to create multiq qdisc.\n",
1311                         pmd->name);
1312                 goto disable_rte_flow;
1313         }
1314         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1315                 RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1316                         pmd->name);
1317                 goto disable_rte_flow;
1318         }
1319         LIST_INIT(&pmd->flows);
1320
1321         if (strlen(remote_iface)) {
1322                 pmd->remote_if_index = if_nametoindex(remote_iface);
1323                 if (!pmd->remote_if_index) {
1324                         RTE_LOG(ERR, PMD, "%s: failed to get %s if_index.\n",
1325                                 pmd->name, remote_iface);
1326                         goto error_remote;
1327                 }
1328                 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1329                          "%s", remote_iface);
1330
1331                 /* Save state of remote device */
1332                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1333
1334                 /* Replicate remote MAC address */
1335                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1336                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1337                                 pmd->name, pmd->remote_iface);
1338                         goto error_remote;
1339                 }
1340                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1341                            ETHER_ADDR_LEN);
1342                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1343                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1344                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1345                                 pmd->name, remote_iface);
1346                         goto error_remote;
1347                 }
1348
1349                 /*
1350                  * Flush usually returns negative value because it tries to
1351                  * delete every QDISC (and on a running device, one QDISC at
1352                  * least is needed). Ignore negative return value.
1353                  */
1354                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1355                 if (qdisc_create_ingress(pmd->nlsk_fd,
1356                                          pmd->remote_if_index) < 0) {
1357                         RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1358                                 pmd->remote_iface);
1359                         goto error_remote;
1360                 }
1361                 LIST_INIT(&pmd->implicit_flows);
1362                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1363                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1364                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1365                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1366                         RTE_LOG(ERR, PMD,
1367                                 "%s: failed to create implicit rules.\n",
1368                                 pmd->name);
1369                         goto error_remote;
1370                 }
1371         }
1372
1373         return 0;
1374
1375 disable_rte_flow:
1376         RTE_LOG(ERR, PMD, " Disabling rte flow support: %s(%d)\n",
1377                 strerror(errno), errno);
1378         if (strlen(remote_iface)) {
1379                 RTE_LOG(ERR, PMD, "Remote feature requires flow support.\n");
1380                 goto error_exit;
1381         }
1382         return 0;
1383
1384 error_remote:
1385         RTE_LOG(ERR, PMD, " Can't set up remote feature: %s(%d)\n",
1386                 strerror(errno), errno);
1387         tap_flow_implicit_flush(pmd, NULL);
1388
1389 error_exit:
1390         RTE_LOG(ERR, PMD, "TAP Unable to initialize %s\n",
1391                 rte_vdev_device_name(vdev));
1392
1393         rte_free(data);
1394         return -EINVAL;
1395 }
1396
1397 static int
1398 set_interface_name(const char *key __rte_unused,
1399                    const char *value,
1400                    void *extra_args)
1401 {
1402         char *name = (char *)extra_args;
1403
1404         if (value)
1405                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
1406         else
1407                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1408                          DEFAULT_TAP_NAME, (tap_unit - 1));
1409
1410         return 0;
1411 }
1412
1413 static int
1414 set_interface_speed(const char *key __rte_unused,
1415                     const char *value,
1416                     void *extra_args)
1417 {
1418         *(int *)extra_args = (value) ? atoi(value) : ETH_SPEED_NUM_10G;
1419
1420         return 0;
1421 }
1422
1423 static int
1424 set_remote_iface(const char *key __rte_unused,
1425                  const char *value,
1426                  void *extra_args)
1427 {
1428         char *name = (char *)extra_args;
1429
1430         if (value)
1431                 snprintf(name, RTE_ETH_NAME_MAX_LEN, "%s", value);
1432
1433         return 0;
1434 }
1435
1436 static int
1437 set_mac_type(const char *key __rte_unused,
1438              const char *value,
1439              void *extra_args)
1440 {
1441         if (value &&
1442             !strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED)))
1443                 *(int *)extra_args = 1;
1444         return 0;
1445 }
1446
1447 /* Open a TAP interface device.
1448  */
1449 static int
1450 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1451 {
1452         const char *name, *params;
1453         int ret;
1454         struct rte_kvargs *kvlist = NULL;
1455         int speed;
1456         char tap_name[RTE_ETH_NAME_MAX_LEN];
1457         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1458         int fixed_mac_type = 0;
1459
1460         name = rte_vdev_device_name(dev);
1461         params = rte_vdev_device_args(dev);
1462
1463         speed = ETH_SPEED_NUM_10G;
1464         snprintf(tap_name, sizeof(tap_name), "%s%d",
1465                  DEFAULT_TAP_NAME, tap_unit++);
1466         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1467
1468         if (params && (params[0] != '\0')) {
1469                 RTE_LOG(DEBUG, PMD, "parameters (%s)\n", params);
1470
1471                 kvlist = rte_kvargs_parse(params, valid_arguments);
1472                 if (kvlist) {
1473                         if (rte_kvargs_count(kvlist, ETH_TAP_SPEED_ARG) == 1) {
1474                                 ret = rte_kvargs_process(kvlist,
1475                                                          ETH_TAP_SPEED_ARG,
1476                                                          &set_interface_speed,
1477                                                          &speed);
1478                                 if (ret == -1)
1479                                         goto leave;
1480                         }
1481
1482                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1483                                 ret = rte_kvargs_process(kvlist,
1484                                                          ETH_TAP_IFACE_ARG,
1485                                                          &set_interface_name,
1486                                                          tap_name);
1487                                 if (ret == -1)
1488                                         goto leave;
1489                         }
1490
1491                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1492                                 ret = rte_kvargs_process(kvlist,
1493                                                          ETH_TAP_REMOTE_ARG,
1494                                                          &set_remote_iface,
1495                                                          remote_iface);
1496                                 if (ret == -1)
1497                                         goto leave;
1498                         }
1499
1500                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1501                                 ret = rte_kvargs_process(kvlist,
1502                                                          ETH_TAP_MAC_ARG,
1503                                                          &set_mac_type,
1504                                                          &fixed_mac_type);
1505                                 if (ret == -1)
1506                                         goto leave;
1507                         }
1508                 }
1509         }
1510         pmd_link.link_speed = speed;
1511
1512         RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1513                 name, tap_name);
1514
1515         ret = eth_dev_tap_create(dev, tap_name, remote_iface, fixed_mac_type);
1516
1517 leave:
1518         if (ret == -1) {
1519                 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1520                         name, tap_name);
1521                 tap_unit--;             /* Restore the unit number */
1522         }
1523         rte_kvargs_free(kvlist);
1524
1525         return ret;
1526 }
1527
1528 /* detach a TAP device.
1529  */
1530 static int
1531 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1532 {
1533         struct rte_eth_dev *eth_dev = NULL;
1534         struct pmd_internals *internals;
1535         int i;
1536
1537         RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1538                 rte_socket_id());
1539
1540         /* find the ethdev entry */
1541         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1542         if (!eth_dev)
1543                 return 0;
1544
1545         internals = eth_dev->data->dev_private;
1546         if (internals->nlsk_fd) {
1547                 tap_flow_flush(eth_dev, NULL);
1548                 tap_flow_implicit_flush(internals, NULL);
1549                 tap_nl_final(internals->nlsk_fd);
1550         }
1551         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1552                 if (internals->rxq[i].fd != -1) {
1553                         close(internals->rxq[i].fd);
1554                         internals->rxq[i].fd = -1;
1555                 }
1556                 if (internals->txq[i].fd != -1) {
1557                         close(internals->txq[i].fd);
1558                         internals->txq[i].fd = -1;
1559                 }
1560         }
1561
1562         close(internals->ioctl_sock);
1563         rte_free(eth_dev->data->dev_private);
1564         rte_free(eth_dev->data);
1565
1566         rte_eth_dev_release_port(eth_dev);
1567
1568         return 0;
1569 }
1570
1571 static struct rte_vdev_driver pmd_tap_drv = {
1572         .probe = rte_pmd_tap_probe,
1573         .remove = rte_pmd_tap_remove,
1574 };
1575 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1576 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1577 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1578                               ETH_TAP_IFACE_ARG "=<string> "
1579                               ETH_TAP_SPEED_ARG "=<int> "
1580                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_FIXED " "
1581                               ETH_TAP_REMOTE_ARG "=<string>");