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 <stdbool.h>
  28 #include <stdint.h>
  29 #include <sys/uio.h>
  30 #include <unistd.h>
  31 #include <arpa/inet.h>
  32 #include <net/if.h>
  33 #include <linux/if_tun.h>
  34 #include <linux/if_ether.h>
  35 #include <fcntl.h>
  36
  37 #include <rte_eth_tap.h>
  38 #include <tap_flow.h>
  39 #include <tap_netlink.h>
  40 #include <tap_tcmsgs.h>
  41
  42 /* Linux based path to the TUN device */
  43 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  44 #define DEFAULT_TAP_NAME        "dtap"
  45
  46 #define ETH_TAP_IFACE_ARG       "iface"
  47 #define ETH_TAP_SPEED_ARG       "speed"
  48 #define ETH_TAP_REMOTE_ARG      "remote"
  49 #define ETH_TAP_MAC_ARG         "mac"
  50 #define ETH_TAP_MAC_FIXED       "fixed"
  51
  52 static struct rte_vdev_driver pmd_tap_drv;
  53
  54 static const char *valid_arguments[] = {
  55         ETH_TAP_IFACE_ARG,
  56         ETH_TAP_SPEED_ARG,
  57         ETH_TAP_REMOTE_ARG,
  58         ETH_TAP_MAC_ARG,
  59         NULL
  60 };
  61
  62 static int tap_unit;
  63
  64 static volatile uint32_t tap_trigger;   /* Rx trigger */
  65
  66 static struct rte_eth_link pmd_link = {
  67         .link_speed = ETH_SPEED_NUM_10G,
  68         .link_duplex = ETH_LINK_FULL_DUPLEX,
  69         .link_status = ETH_LINK_DOWN,
  70         .link_autoneg = ETH_LINK_AUTONEG
  71 };
  72
  73 static void
  74 tap_trigger_cb(int sig __rte_unused)
  75 {
  76         /* Valid trigger values are nonzero */
  77         tap_trigger = (tap_trigger + 1) | 0x80000000;
  78 }
  79
  80 /* Specifies on what netdevices the ioctl should be applied */
  81 enum ioctl_mode {
  82         LOCAL_AND_REMOTE,
  83         LOCAL_ONLY,
  84         REMOTE_ONLY,
  85 };
  86
  87 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
  88
  89 /* Tun/Tap allocation routine
  90  *
  91  * name is the number of the interface to use, unless NULL to take the host
  92  * supplied name.
  93  */
  94 static int
  95 tun_alloc(struct pmd_internals *pmd)
  96 {
  97         struct ifreq ifr;
  98 #ifdef IFF_MULTI_QUEUE
  99         unsigned int features;
 100 #endif
 101         int fd;
 102
 103         memset(&ifr, 0, sizeof(struct ifreq));
 104
 105         /*
 106          * Do not set IFF_NO_PI as packet information header will be needed
 107          * to check if a received packet has been truncated.
 108          */
 109         ifr.ifr_flags = IFF_TAP;
 110         snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
 111
 112         RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
 113
 114         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 115         if (fd < 0) {
 116                 RTE_LOG(ERR, PMD, "Unable to create TAP interface\n");
 117                 goto error;
 118         }
 119
 120 #ifdef IFF_MULTI_QUEUE
 121         /* Grab the TUN features to verify we can work multi-queue */
 122         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 123                 RTE_LOG(ERR, PMD, "TAP unable to get TUN/TAP features\n");
 124                 goto error;
 125         }
 126         RTE_LOG(DEBUG, PMD, "  TAP Features %08x\n", features);
 127
 128         if (features & IFF_MULTI_QUEUE) {
 129                 RTE_LOG(DEBUG, PMD, "  Multi-queue support for %d queues\n",
 130                         RTE_PMD_TAP_MAX_QUEUES);
 131                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 132         } else
 133 #endif
 134         {
 135                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 136                 RTE_LOG(DEBUG, PMD, "  Single queue only support\n");
 137         }
 138
 139         /* Set the TUN/TAP configuration and set the name if needed */
 140         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 141                 RTE_LOG(WARNING, PMD,
 142                         "Unable to set TUNSETIFF for %s\n",
 143                         ifr.ifr_name);
 144                 perror("TUNSETIFF");
 145                 goto error;
 146         }
 147
 148         /* Always set the file descriptor to non-blocking */
 149         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 150                 RTE_LOG(WARNING, PMD,
 151                         "Unable to set %s to nonblocking\n",
 152                         ifr.ifr_name);
 153                 perror("F_SETFL, NONBLOCK");
 154                 goto error;
 155         }
 156
 157         /* Set up trigger to optimize empty Rx bursts */
 158         errno = 0;
 159         do {
 160                 struct sigaction sa;
 161                 int flags = fcntl(fd, F_GETFL);
 162
 163                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 164                         break;
 165                 if (sa.sa_handler != tap_trigger_cb) {
 166                         /*
 167                          * Make sure SIGIO is not already taken. This is done
 168                          * as late as possible to leave the application a
 169                          * chance to set up its own signal handler first.
 170                          */
 171                         if (sa.sa_handler != SIG_IGN &&
 172                             sa.sa_handler != SIG_DFL) {
 173                                 errno = EBUSY;
 174                                 break;
 175                         }
 176                         sa = (struct sigaction){
 177                                 .sa_flags = SA_RESTART,
 178                                 .sa_handler = tap_trigger_cb,
 179                         };
 180                         if (sigaction(SIGIO, &sa, NULL) == -1)
 181                                 break;
 182                 }
 183                 /* Enable SIGIO on file descriptor */
 184                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 185                 fcntl(fd, F_SETOWN, getpid());
 186         } while (0);
 187         if (errno) {
 188                 /* Disable trigger globally in case of error */
 189                 tap_trigger = 0;
 190                 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
 191                         strerror(errno));
 192         }
 193
 194         return fd;
 195
 196 error:
 197         if (fd > 0)
 198                 close(fd);
 199         return -1;
 200 }
 201
 202 static void
 203 tap_verify_csum(struct rte_mbuf *mbuf)
 204 {
 205         uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
 206         uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
 207         uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
 208         unsigned int l2_len = sizeof(struct ether_hdr);
 209         unsigned int l3_len;
 210         uint16_t cksum = 0;
 211         void *l3_hdr;
 212         void *l4_hdr;
 213
 214         if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
 215                 l2_len += 4;
 216         else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
 217                 l2_len += 8;
 218         /* Don't verify checksum for packets with discontinuous L2 header */
 219         if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
 220                      rte_pktmbuf_data_len(mbuf)))
 221                 return;
 222         l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
 223         if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 224                 struct ipv4_hdr *iph = l3_hdr;
 225
 226                 /* ihl contains the number of 4-byte words in the header */
 227                 l3_len = 4 * (iph->version_ihl & 0xf);
 228                 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
 229                         return;
 230
 231                 cksum = ~rte_raw_cksum(iph, l3_len);
 232                 mbuf->ol_flags |= cksum ?
 233                         PKT_RX_IP_CKSUM_BAD :
 234                         PKT_RX_IP_CKSUM_GOOD;
 235         } else if (l3 == RTE_PTYPE_L3_IPV6) {
 236                 l3_len = sizeof(struct ipv6_hdr);
 237         } else {
 238                 /* IPv6 extensions are not supported */
 239                 return;
 240         }
 241         if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
 242                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 243                 /* Don't verify checksum for multi-segment packets. */
 244                 if (mbuf->nb_segs > 1)
 245                         return;
 246                 if (l3 == RTE_PTYPE_L3_IPV4)
 247                         cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
 248                 else if (l3 == RTE_PTYPE_L3_IPV6)
 249                         cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
 250                 mbuf->ol_flags |= cksum ?
 251                         PKT_RX_L4_CKSUM_BAD :
 252                         PKT_RX_L4_CKSUM_GOOD;
 253         }
 254 }
 255
 256 /* Callback to handle the rx burst of packets to the correct interface and
 257  * file descriptor(s) in a multi-queue setup.
 258  */
 259 static uint16_t
 260 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 261 {
 262         struct rx_queue *rxq = queue;
 263         uint16_t num_rx;
 264         unsigned long num_rx_bytes = 0;
 265         uint32_t trigger = tap_trigger;
 266
 267         if (trigger == rxq->trigger_seen)
 268                 return 0;
 269         if (trigger)
 270                 rxq->trigger_seen = trigger;
 271         rte_compiler_barrier();
 272         for (num_rx = 0; num_rx < nb_pkts; ) {
 273                 struct rte_mbuf *mbuf = rxq->pool;
 274                 struct rte_mbuf *seg = NULL;
 275                 struct rte_mbuf *new_tail = NULL;
 276                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 277                 int len;
 278
 279                 len = readv(rxq->fd, *rxq->iovecs,
 280                             1 + (rxq->rxmode->enable_scatter ?
 281                                  rxq->nb_rx_desc : 1));
 282                 if (len < (int)sizeof(struct tun_pi))
 283                         break;
 284
 285                 /* Packet couldn't fit in the provided mbuf */
 286                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 287                         rxq->stats.ierrors++;
 288                         continue;
 289                 }
 290
 291                 len -= sizeof(struct tun_pi);
 292
 293                 mbuf->pkt_len = len;
 294                 mbuf->port = rxq->in_port;
 295                 while (1) {
 296                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 297
 298                         if (unlikely(!buf)) {
 299                                 rxq->stats.rx_nombuf++;
 300                                 /* No new buf has been allocated: do nothing */
 301                                 if (!new_tail || !seg)
 302                                         goto end;
 303
 304                                 seg->next = NULL;
 305                                 rte_pktmbuf_free(mbuf);
 306
 307                                 goto end;
 308                         }
 309                         seg = seg ? seg->next : mbuf;
 310                         if (rxq->pool == mbuf)
 311                                 rxq->pool = buf;
 312                         if (new_tail)
 313                                 new_tail->next = buf;
 314                         new_tail = buf;
 315                         new_tail->next = seg->next;
 316
 317                         /* iovecs[0] is reserved for packet info (pi) */
 318                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 319                                 buf->buf_len - data_off;
 320                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 321                                 (char *)buf->buf_addr + data_off;
 322
 323                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 324                         seg->data_off = data_off;
 325
 326                         len -= seg->data_len;
 327                         if (len <= 0)
 328                                 break;
 329                         mbuf->nb_segs++;
 330                         /* First segment has headroom, not the others */
 331                         data_off = 0;
 332                 }
 333                 seg->next = NULL;
 334                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 335                                                       RTE_PTYPE_ALL_MASK);
 336                 if (rxq->rxmode->hw_ip_checksum)
 337                         tap_verify_csum(mbuf);
 338
 339                 /* account for the receive frame */
 340                 bufs[num_rx++] = mbuf;
 341                 num_rx_bytes += mbuf->pkt_len;
 342         }
 343 end:
 344         rxq->stats.ipackets += num_rx;
 345         rxq->stats.ibytes += num_rx_bytes;
 346
 347         return num_rx;
 348 }
 349
 350 static uint64_t
 351 tap_tx_offload_get_port_capa(void)
 352 {
 353         /*
 354          * In order to support legacy apps,
 355          * report capabilities also as port capabilities.
 356          */
 357         return DEV_TX_OFFLOAD_IPV4_CKSUM |
 358                DEV_TX_OFFLOAD_UDP_CKSUM |
 359                DEV_TX_OFFLOAD_TCP_CKSUM;
 360 }
 361
 362 static uint64_t
 363 tap_tx_offload_get_queue_capa(void)
 364 {
 365         return DEV_TX_OFFLOAD_IPV4_CKSUM |
 366                DEV_TX_OFFLOAD_UDP_CKSUM |
 367                DEV_TX_OFFLOAD_TCP_CKSUM;
 368 }
 369
 370 static bool
 371 tap_txq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 372 {
 373         uint64_t port_offloads = dev->data->dev_conf.txmode.offloads;
 374         uint64_t queue_supp_offloads = tap_tx_offload_get_queue_capa();
 375         uint64_t port_supp_offloads = tap_tx_offload_get_port_capa();
 376
 377         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 378             offloads)
 379                 return false;
 380         /* Verify we have no conflict with port offloads */
 381         if ((port_offloads ^ offloads) & port_supp_offloads)
 382                 return false;
 383         return true;
 384 }
 385
 386 static void
 387 tap_tx_offload(char *packet, uint64_t ol_flags, unsigned int l2_len,
 388                unsigned int l3_len)
 389 {
 390         void *l3_hdr = packet + l2_len;
 391
 392         if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
 393                 struct ipv4_hdr *iph = l3_hdr;
 394                 uint16_t cksum;
 395
 396                 iph->hdr_checksum = 0;
 397                 cksum = rte_raw_cksum(iph, l3_len);
 398                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 399         }
 400         if (ol_flags & PKT_TX_L4_MASK) {
 401                 uint16_t l4_len;
 402                 uint32_t cksum;
 403                 uint16_t *l4_cksum;
 404                 void *l4_hdr;
 405
 406                 l4_hdr = packet + l2_len + l3_len;
 407                 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
 408                         l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
 409                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 410                         l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
 411                 else
 412                         return;
 413                 *l4_cksum = 0;
 414                 if (ol_flags & PKT_TX_IPV4) {
 415                         struct ipv4_hdr *iph = l3_hdr;
 416
 417                         l4_len = rte_be_to_cpu_16(iph->total_length) - l3_len;
 418                         cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 419                 } else {
 420                         struct ipv6_hdr *ip6h = l3_hdr;
 421
 422                         /* payload_len does not include ext headers */
 423                         l4_len = rte_be_to_cpu_16(ip6h->payload_len) -
 424                                 l3_len + sizeof(struct ipv6_hdr);
 425                         cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 426                 }
 427                 cksum += rte_raw_cksum(l4_hdr, l4_len);
 428                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 429                 cksum = (~cksum) & 0xffff;
 430                 if (cksum == 0)
 431                         cksum = 0xffff;
 432                 *l4_cksum = cksum;
 433         }
 434 }
 435
 436 /* Callback to handle sending packets from the tap interface
 437  */
 438 static uint16_t
 439 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 440 {
 441         struct tx_queue *txq = queue;
 442         uint16_t num_tx = 0;
 443         unsigned long num_tx_bytes = 0;
 444         uint32_t max_size;
 445         int i;
 446
 447         if (unlikely(nb_pkts == 0))
 448                 return 0;
 449
 450         max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
 451         for (i = 0; i < nb_pkts; i++) {
 452                 struct rte_mbuf *mbuf = bufs[num_tx];
 453                 struct iovec iovecs[mbuf->nb_segs + 1];
 454                 struct tun_pi pi = { .flags = 0 };
 455                 struct rte_mbuf *seg = mbuf;
 456                 char m_copy[mbuf->data_len];
 457                 int n;
 458                 int j;
 459
 460                 /* stats.errs will be incremented */
 461                 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
 462                         break;
 463
 464                 iovecs[0].iov_base = &pi;
 465                 iovecs[0].iov_len = sizeof(pi);
 466                 for (j = 1; j <= mbuf->nb_segs; j++) {
 467                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 468                         iovecs[j].iov_base =
 469                                 rte_pktmbuf_mtod(seg, void *);
 470                         seg = seg->next;
 471                 }
 472                 if (txq->csum &&
 473                     ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 474                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 475                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
 476                         /* Support only packets with all data in the same seg */
 477                         if (mbuf->nb_segs > 1)
 478                                 break;
 479                         /* To change checksums, work on a copy of data. */
 480                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 481                                    rte_pktmbuf_data_len(mbuf));
 482                         tap_tx_offload(m_copy, mbuf->ol_flags,
 483                                        mbuf->l2_len, mbuf->l3_len);
 484                         iovecs[1].iov_base = m_copy;
 485                 }
 486                 /* copy the tx frame data */
 487                 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
 488                 if (n <= 0)
 489                         break;
 490
 491                 num_tx++;
 492                 num_tx_bytes += mbuf->pkt_len;
 493                 rte_pktmbuf_free(mbuf);
 494         }
 495
 496         txq->stats.opackets += num_tx;
 497         txq->stats.errs += nb_pkts - num_tx;
 498         txq->stats.obytes += num_tx_bytes;
 499
 500         return num_tx;
 501 }
 502
 503 static const char *
 504 tap_ioctl_req2str(unsigned long request)
 505 {
 506         switch (request) {
 507         case SIOCSIFFLAGS:
 508                 return "SIOCSIFFLAGS";
 509         case SIOCGIFFLAGS:
 510                 return "SIOCGIFFLAGS";
 511         case SIOCGIFHWADDR:
 512                 return "SIOCGIFHWADDR";
 513         case SIOCSIFHWADDR:
 514                 return "SIOCSIFHWADDR";
 515         case SIOCSIFMTU:
 516                 return "SIOCSIFMTU";
 517         }
 518         return "UNKNOWN";
 519 }
 520
 521 static int
 522 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 523           struct ifreq *ifr, int set, enum ioctl_mode mode)
 524 {
 525         short req_flags = ifr->ifr_flags;
 526         int remote = pmd->remote_if_index &&
 527                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 528
 529         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 530                 return 0;
 531         /*
 532          * If there is a remote netdevice, apply ioctl on it, then apply it on
 533          * the tap netdevice.
 534          */
 535 apply:
 536         if (remote)
 537                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
 538         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 539                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 540         switch (request) {
 541         case SIOCSIFFLAGS:
 542                 /* fetch current flags to leave other flags untouched */
 543                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 544                         goto error;
 545                 if (set)
 546                         ifr->ifr_flags |= req_flags;
 547                 else
 548                         ifr->ifr_flags &= ~req_flags;
 549                 break;
 550         case SIOCGIFFLAGS:
 551         case SIOCGIFHWADDR:
 552         case SIOCSIFHWADDR:
 553         case SIOCSIFMTU:
 554                 break;
 555         default:
 556                 RTE_ASSERT(!"unsupported request type: must not happen");
 557         }
 558         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 559                 goto error;
 560         if (remote-- && mode == LOCAL_AND_REMOTE)
 561                 goto apply;
 562         return 0;
 563
 564 error:
 565         RTE_LOG(DEBUG, PMD, "%s: %s(%s) failed: %s(%d)\n", ifr->ifr_name,
 566                 __func__, tap_ioctl_req2str(request), strerror(errno), errno);
 567         return -errno;
 568 }
 569
 570 static int
 571 tap_link_set_down(struct rte_eth_dev *dev)
 572 {
 573         struct pmd_internals *pmd = dev->data->dev_private;
 574         struct ifreq ifr = { .ifr_flags = IFF_UP };
 575
 576         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 577         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 578 }
 579
 580 static int
 581 tap_link_set_up(struct rte_eth_dev *dev)
 582 {
 583         struct pmd_internals *pmd = dev->data->dev_private;
 584         struct ifreq ifr = { .ifr_flags = IFF_UP };
 585
 586         dev->data->dev_link.link_status = ETH_LINK_UP;
 587         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 588 }
 589
 590 static int
 591 tap_dev_start(struct rte_eth_dev *dev)
 592 {
 593         int err;
 594
 595         err = tap_intr_handle_set(dev, 1);
 596         if (err)
 597                 return err;
 598         return tap_link_set_up(dev);
 599 }
 600
 601 /* This function gets called when the current port gets stopped.
 602  */
 603 static void
 604 tap_dev_stop(struct rte_eth_dev *dev)
 605 {
 606         tap_intr_handle_set(dev, 0);
 607         tap_link_set_down(dev);
 608 }
 609
 610 static int
 611 tap_dev_configure(struct rte_eth_dev *dev)
 612 {
 613         uint64_t supp_tx_offloads = tap_tx_offload_get_port_capa();
 614         uint64_t tx_offloads = dev->data->dev_conf.txmode.offloads;
 615
 616         if ((tx_offloads & supp_tx_offloads) != tx_offloads) {
 617                 rte_errno = ENOTSUP;
 618                 RTE_LOG(ERR, PMD,
 619                         "Some Tx offloads are not supported "
 620                         "requested 0x%" PRIx64 " supported 0x%" PRIx64 "\n",
 621                         tx_offloads, supp_tx_offloads);
 622                 return -rte_errno;
 623         }
 624         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 625                 RTE_LOG(ERR, PMD,
 626                         "%s: number of rx queues %d exceeds max num of queues %d\n",
 627                         dev->device->name,
 628                         dev->data->nb_rx_queues,
 629                         RTE_PMD_TAP_MAX_QUEUES);
 630                 return -1;
 631         }
 632         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 633                 RTE_LOG(ERR, PMD,
 634                         "%s: number of tx queues %d exceeds max num of queues %d\n",
 635                         dev->device->name,
 636                         dev->data->nb_tx_queues,
 637                         RTE_PMD_TAP_MAX_QUEUES);
 638                 return -1;
 639         }
 640
 641         RTE_LOG(INFO, PMD, "%s: %p: TX configured queues number: %u\n",
 642              dev->device->name, (void *)dev, dev->data->nb_tx_queues);
 643
 644         RTE_LOG(INFO, PMD, "%s: %p: RX configured queues number: %u\n",
 645              dev->device->name, (void *)dev, dev->data->nb_rx_queues);
 646
 647         return 0;
 648 }
 649
 650 static uint32_t
 651 tap_dev_speed_capa(void)
 652 {
 653         uint32_t speed = pmd_link.link_speed;
 654         uint32_t capa = 0;
 655
 656         if (speed >= ETH_SPEED_NUM_10M)
 657                 capa |= ETH_LINK_SPEED_10M;
 658         if (speed >= ETH_SPEED_NUM_100M)
 659                 capa |= ETH_LINK_SPEED_100M;
 660         if (speed >= ETH_SPEED_NUM_1G)
 661                 capa |= ETH_LINK_SPEED_1G;
 662         if (speed >= ETH_SPEED_NUM_5G)
 663                 capa |= ETH_LINK_SPEED_2_5G;
 664         if (speed >= ETH_SPEED_NUM_5G)
 665                 capa |= ETH_LINK_SPEED_5G;
 666         if (speed >= ETH_SPEED_NUM_10G)
 667                 capa |= ETH_LINK_SPEED_10G;
 668         if (speed >= ETH_SPEED_NUM_20G)
 669                 capa |= ETH_LINK_SPEED_20G;
 670         if (speed >= ETH_SPEED_NUM_25G)
 671                 capa |= ETH_LINK_SPEED_25G;
 672         if (speed >= ETH_SPEED_NUM_40G)
 673                 capa |= ETH_LINK_SPEED_40G;
 674         if (speed >= ETH_SPEED_NUM_50G)
 675                 capa |= ETH_LINK_SPEED_50G;
 676         if (speed >= ETH_SPEED_NUM_56G)
 677                 capa |= ETH_LINK_SPEED_56G;
 678         if (speed >= ETH_SPEED_NUM_100G)
 679                 capa |= ETH_LINK_SPEED_100G;
 680
 681         return capa;
 682 }
 683
 684 static void
 685 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 686 {
 687         struct pmd_internals *internals = dev->data->dev_private;
 688
 689         dev_info->if_index = internals->if_index;
 690         dev_info->max_mac_addrs = 1;
 691         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 692         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
 693         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
 694         dev_info->min_rx_bufsize = 0;
 695         dev_info->pci_dev = NULL;
 696         dev_info->speed_capa = tap_dev_speed_capa();
 697         dev_info->rx_offload_capa = (DEV_RX_OFFLOAD_IPV4_CKSUM |
 698                                      DEV_RX_OFFLOAD_UDP_CKSUM |
 699                                      DEV_RX_OFFLOAD_TCP_CKSUM);
 700         dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
 701         dev_info->tx_offload_capa = dev_info->tx_queue_offload_capa |
 702                                     tap_tx_offload_get_port_capa();
 703 }
 704
 705 static int
 706 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 707 {
 708         unsigned int i, imax;
 709         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 710         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 711         unsigned long rx_nombuf = 0, ierrors = 0;
 712         const struct pmd_internals *pmd = dev->data->dev_private;
 713
 714         /* rx queue statistics */
 715         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 716                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 717         for (i = 0; i < imax; i++) {
 718                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 719                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 720                 rx_total += tap_stats->q_ipackets[i];
 721                 rx_bytes_total += tap_stats->q_ibytes[i];
 722                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
 723                 ierrors += pmd->rxq[i].stats.ierrors;
 724         }
 725
 726         /* tx queue statistics */
 727         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 728                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 729
 730         for (i = 0; i < imax; i++) {
 731                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 732                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 733                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 734                 tx_total += tap_stats->q_opackets[i];
 735                 tx_err_total += tap_stats->q_errors[i];
 736                 tx_bytes_total += tap_stats->q_obytes[i];
 737         }
 738
 739         tap_stats->ipackets = rx_total;
 740         tap_stats->ibytes = rx_bytes_total;
 741         tap_stats->ierrors = ierrors;
 742         tap_stats->rx_nombuf = rx_nombuf;
 743         tap_stats->opackets = tx_total;
 744         tap_stats->oerrors = tx_err_total;
 745         tap_stats->obytes = tx_bytes_total;
 746         return 0;
 747 }
 748
 749 static void
 750 tap_stats_reset(struct rte_eth_dev *dev)
 751 {
 752         int i;
 753         struct pmd_internals *pmd = dev->data->dev_private;
 754
 755         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 756                 pmd->rxq[i].stats.ipackets = 0;
 757                 pmd->rxq[i].stats.ibytes = 0;
 758                 pmd->rxq[i].stats.ierrors = 0;
 759                 pmd->rxq[i].stats.rx_nombuf = 0;
 760
 761                 pmd->txq[i].stats.opackets = 0;
 762                 pmd->txq[i].stats.errs = 0;
 763                 pmd->txq[i].stats.obytes = 0;
 764         }
 765 }
 766
 767 static void
 768 tap_dev_close(struct rte_eth_dev *dev)
 769 {
 770         int i;
 771         struct pmd_internals *internals = dev->data->dev_private;
 772
 773         tap_link_set_down(dev);
 774         tap_flow_flush(dev, NULL);
 775         tap_flow_implicit_flush(internals, NULL);
 776
 777         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 778                 if (internals->rxq[i].fd != -1) {
 779                         close(internals->rxq[i].fd);
 780                         internals->rxq[i].fd = -1;
 781                 }
 782                 if (internals->txq[i].fd != -1) {
 783                         close(internals->txq[i].fd);
 784                         internals->txq[i].fd = -1;
 785                 }
 786         }
 787
 788         if (internals->remote_if_index) {
 789                 /* Restore initial remote state */
 790                 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
 791                                 &internals->remote_initial_flags);
 792         }
 793 }
 794
 795 static void
 796 tap_rx_queue_release(void *queue)
 797 {
 798         struct rx_queue *rxq = queue;
 799
 800         if (rxq && (rxq->fd > 0)) {
 801                 close(rxq->fd);
 802                 rxq->fd = -1;
 803                 rte_pktmbuf_free(rxq->pool);
 804                 rte_free(rxq->iovecs);
 805                 rxq->pool = NULL;
 806                 rxq->iovecs = NULL;
 807         }
 808 }
 809
 810 static void
 811 tap_tx_queue_release(void *queue)
 812 {
 813         struct tx_queue *txq = queue;
 814
 815         if (txq && (txq->fd > 0)) {
 816                 close(txq->fd);
 817                 txq->fd = -1;
 818         }
 819 }
 820
 821 static int
 822 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
 823 {
 824         struct rte_eth_link *dev_link = &dev->data->dev_link;
 825         struct pmd_internals *pmd = dev->data->dev_private;
 826         struct ifreq ifr = { .ifr_flags = 0 };
 827
 828         if (pmd->remote_if_index) {
 829                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
 830                 if (!(ifr.ifr_flags & IFF_UP) ||
 831                     !(ifr.ifr_flags & IFF_RUNNING)) {
 832                         dev_link->link_status = ETH_LINK_DOWN;
 833                         return 0;
 834                 }
 835         }
 836         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
 837         dev_link->link_status =
 838                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
 839                  ETH_LINK_UP :
 840                  ETH_LINK_DOWN);
 841         return 0;
 842 }
 843
 844 static void
 845 tap_promisc_enable(struct rte_eth_dev *dev)
 846 {
 847         struct pmd_internals *pmd = dev->data->dev_private;
 848         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 849
 850         dev->data->promiscuous = 1;
 851         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 852         if (pmd->remote_if_index && !pmd->flow_isolate)
 853                 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
 854 }
 855
 856 static void
 857 tap_promisc_disable(struct rte_eth_dev *dev)
 858 {
 859         struct pmd_internals *pmd = dev->data->dev_private;
 860         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 861
 862         dev->data->promiscuous = 0;
 863         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 864         if (pmd->remote_if_index && !pmd->flow_isolate)
 865                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
 866 }
 867
 868 static void
 869 tap_allmulti_enable(struct rte_eth_dev *dev)
 870 {
 871         struct pmd_internals *pmd = dev->data->dev_private;
 872         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 873
 874         dev->data->all_multicast = 1;
 875         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 876         if (pmd->remote_if_index && !pmd->flow_isolate)
 877                 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
 878 }
 879
 880 static void
 881 tap_allmulti_disable(struct rte_eth_dev *dev)
 882 {
 883         struct pmd_internals *pmd = dev->data->dev_private;
 884         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 885
 886         dev->data->all_multicast = 0;
 887         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 888         if (pmd->remote_if_index && !pmd->flow_isolate)
 889                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
 890 }
 891
 892 static void
 893 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 894 {
 895         struct pmd_internals *pmd = dev->data->dev_private;
 896         enum ioctl_mode mode = LOCAL_ONLY;
 897         struct ifreq ifr;
 898
 899         if (is_zero_ether_addr(mac_addr)) {
 900                 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
 901                         dev->device->name);
 902                 return;
 903         }
 904         /* Check the actual current MAC address on the tap netdevice */
 905         if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
 906                 return;
 907         if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 908                                mac_addr))
 909                 return;
 910         /* Check the current MAC address on the remote */
 911         if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0)
 912                 return;
 913         if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 914                                mac_addr))
 915                 mode = LOCAL_AND_REMOTE;
 916         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
 917         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
 918         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode) < 0)
 919                 return;
 920         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
 921         if (pmd->remote_if_index && !pmd->flow_isolate) {
 922                 /* Replace MAC redirection rule after a MAC change */
 923                 if (tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC) < 0) {
 924                         RTE_LOG(ERR, PMD,
 925                                 "%s: Couldn't delete MAC redirection rule\n",
 926                                 dev->device->name);
 927                         return;
 928                 }
 929                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0)
 930                         RTE_LOG(ERR, PMD,
 931                                 "%s: Couldn't add MAC redirection rule\n",
 932                                 dev->device->name);
 933         }
 934 }
 935
 936 static int
 937 tap_setup_queue(struct rte_eth_dev *dev,
 938                 struct pmd_internals *internals,
 939                 uint16_t qid,
 940                 int is_rx)
 941 {
 942         int *fd;
 943         int *other_fd;
 944         const char *dir;
 945         struct pmd_internals *pmd = dev->data->dev_private;
 946         struct rx_queue *rx = &internals->rxq[qid];
 947         struct tx_queue *tx = &internals->txq[qid];
 948
 949         if (is_rx) {
 950                 fd = &rx->fd;
 951                 other_fd = &tx->fd;
 952                 dir = "rx";
 953         } else {
 954                 fd = &tx->fd;
 955                 other_fd = &rx->fd;
 956                 dir = "tx";
 957         }
 958         if (*fd != -1) {
 959                 /* fd for this queue already exists */
 960                 RTE_LOG(DEBUG, PMD, "%s: fd %d for %s queue qid %d exists\n",
 961                         pmd->name, *fd, dir, qid);
 962         } else if (*other_fd != -1) {
 963                 /* Only other_fd exists. dup it */
 964                 *fd = dup(*other_fd);
 965                 if (*fd < 0) {
 966                         *fd = -1;
 967                         RTE_LOG(ERR, PMD, "%s: dup() failed.\n",
 968                                 pmd->name);
 969                         return -1;
 970                 }
 971                 RTE_LOG(DEBUG, PMD, "%s: dup fd %d for %s queue qid %d (%d)\n",
 972                         pmd->name, *other_fd, dir, qid, *fd);
 973         } else {
 974                 /* Both RX and TX fds do not exist (equal -1). Create fd */
 975                 *fd = tun_alloc(pmd);
 976                 if (*fd < 0) {
 977                         *fd = -1; /* restore original value */
 978                         RTE_LOG(ERR, PMD, "%s: tun_alloc() failed.\n",
 979                                 pmd->name);
 980                         return -1;
 981                 }
 982                 RTE_LOG(DEBUG, PMD, "%s: add %s queue for qid %d fd %d\n",
 983                         pmd->name, dir, qid, *fd);
 984         }
 985
 986         tx->mtu = &dev->data->mtu;
 987         rx->rxmode = &dev->data->dev_conf.rxmode;
 988
 989         return *fd;
 990 }
 991
 992 static int
 993 tap_rx_queue_setup(struct rte_eth_dev *dev,
 994                    uint16_t rx_queue_id,
 995                    uint16_t nb_rx_desc,
 996                    unsigned int socket_id,
 997                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 998                    struct rte_mempool *mp)
 999 {
1000         struct pmd_internals *internals = dev->data->dev_private;
1001         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1002         struct rte_mbuf **tmp = &rxq->pool;
1003         long iov_max = sysconf(_SC_IOV_MAX);
1004         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1005         struct iovec (*iovecs)[nb_desc + 1];
1006         int data_off = RTE_PKTMBUF_HEADROOM;
1007         int ret = 0;
1008         int fd;
1009         int i;
1010
1011         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1012                 RTE_LOG(WARNING, PMD,
1013                         "nb_rx_queues %d too small or mempool NULL\n",
1014                         dev->data->nb_rx_queues);
1015                 return -1;
1016         }
1017
1018         rxq->mp = mp;
1019         rxq->trigger_seen = 1; /* force initial burst */
1020         rxq->in_port = dev->data->port_id;
1021         rxq->nb_rx_desc = nb_desc;
1022         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1023                                     socket_id);
1024         if (!iovecs) {
1025                 RTE_LOG(WARNING, PMD,
1026                         "%s: Couldn't allocate %d RX descriptors\n",
1027                         dev->device->name, nb_desc);
1028                 return -ENOMEM;
1029         }
1030         rxq->iovecs = iovecs;
1031
1032         dev->data->rx_queues[rx_queue_id] = rxq;
1033         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1034         if (fd == -1) {
1035                 ret = fd;
1036                 goto error;
1037         }
1038
1039         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1040         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1041
1042         for (i = 1; i <= nb_desc; i++) {
1043                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1044                 if (!*tmp) {
1045                         RTE_LOG(WARNING, PMD,
1046                                 "%s: couldn't allocate memory for queue %d\n",
1047                                 dev->device->name, rx_queue_id);
1048                         ret = -ENOMEM;
1049                         goto error;
1050                 }
1051                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1052                 (*rxq->iovecs)[i].iov_base =
1053                         (char *)(*tmp)->buf_addr + data_off;
1054                 data_off = 0;
1055                 tmp = &(*tmp)->next;
1056         }
1057
1058         RTE_LOG(DEBUG, PMD, "  RX TAP device name %s, qid %d on fd %d\n",
1059                 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
1060
1061         return 0;
1062
1063 error:
1064         rte_pktmbuf_free(rxq->pool);
1065         rxq->pool = NULL;
1066         rte_free(rxq->iovecs);
1067         rxq->iovecs = NULL;
1068         return ret;
1069 }
1070
1071 static int
1072 tap_tx_queue_setup(struct rte_eth_dev *dev,
1073                    uint16_t tx_queue_id,
1074                    uint16_t nb_tx_desc __rte_unused,
1075                    unsigned int socket_id __rte_unused,
1076                    const struct rte_eth_txconf *tx_conf)
1077 {
1078         struct pmd_internals *internals = dev->data->dev_private;
1079         struct tx_queue *txq;
1080         int ret;
1081
1082         if (tx_queue_id >= dev->data->nb_tx_queues)
1083                 return -1;
1084         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1085         txq = dev->data->tx_queues[tx_queue_id];
1086         /*
1087          * Don't verify port offloads for application which
1088          * use the old API.
1089          */
1090         if (tx_conf != NULL &&
1091             !!(tx_conf->txq_flags & ETH_TXQ_FLAGS_IGNORE)) {
1092                 if (tap_txq_are_offloads_valid(dev, tx_conf->offloads)) {
1093                         txq->csum = !!(tx_conf->offloads &
1094                                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
1095                                          DEV_TX_OFFLOAD_UDP_CKSUM |
1096                                          DEV_TX_OFFLOAD_TCP_CKSUM));
1097                 } else {
1098                         rte_errno = ENOTSUP;
1099                         RTE_LOG(ERR, PMD,
1100                                 "%p: Tx queue offloads 0x%" PRIx64
1101                                 " don't match port offloads 0x%" PRIx64
1102                                 " or supported offloads 0x%" PRIx64,
1103                                 (void *)dev, tx_conf->offloads,
1104                                 dev->data->dev_conf.txmode.offloads,
1105                                 tap_tx_offload_get_port_capa());
1106                         return -rte_errno;
1107                 }
1108         }
1109         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1110         if (ret == -1)
1111                 return -1;
1112         RTE_LOG(DEBUG, PMD,
1113                 "  TX TAP device name %s, qid %d on fd %d csum %s\n",
1114                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd,
1115                 txq->csum ? "on" : "off");
1116
1117         return 0;
1118 }
1119
1120 static int
1121 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1122 {
1123         struct pmd_internals *pmd = dev->data->dev_private;
1124         struct ifreq ifr = { .ifr_mtu = mtu };
1125         int err = 0;
1126
1127         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1128         if (!err)
1129                 dev->data->mtu = mtu;
1130
1131         return err;
1132 }
1133
1134 static int
1135 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1136                      struct ether_addr *mc_addr_set __rte_unused,
1137                      uint32_t nb_mc_addr __rte_unused)
1138 {
1139         /*
1140          * Nothing to do actually: the tap has no filtering whatsoever, every
1141          * packet is received.
1142          */
1143         return 0;
1144 }
1145
1146 static int
1147 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1148 {
1149         struct rte_eth_dev *dev = arg;
1150         struct pmd_internals *pmd = dev->data->dev_private;
1151         struct ifinfomsg *info = NLMSG_DATA(nh);
1152
1153         if (nh->nlmsg_type != RTM_NEWLINK ||
1154             (info->ifi_index != pmd->if_index &&
1155              info->ifi_index != pmd->remote_if_index))
1156                 return 0;
1157         return tap_link_update(dev, 0);
1158 }
1159
1160 static void
1161 tap_dev_intr_handler(void *cb_arg)
1162 {
1163         struct rte_eth_dev *dev = cb_arg;
1164         struct pmd_internals *pmd = dev->data->dev_private;
1165
1166         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1167 }
1168
1169 static int
1170 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1171 {
1172         struct pmd_internals *pmd = dev->data->dev_private;
1173
1174         /* In any case, disable interrupt if the conf is no longer there. */
1175         if (!dev->data->dev_conf.intr_conf.lsc) {
1176                 if (pmd->intr_handle.fd != -1) {
1177                         tap_nl_final(pmd->intr_handle.fd);
1178                         rte_intr_callback_unregister(&pmd->intr_handle,
1179                                 tap_dev_intr_handler, dev);
1180                 }
1181                 return 0;
1182         }
1183         if (set) {
1184                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1185                 if (unlikely(pmd->intr_handle.fd == -1))
1186                         return -EBADF;
1187                 return rte_intr_callback_register(
1188                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1189         }
1190         tap_nl_final(pmd->intr_handle.fd);
1191         return rte_intr_callback_unregister(&pmd->intr_handle,
1192                                             tap_dev_intr_handler, dev);
1193 }
1194
1195 static const uint32_t*
1196 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1197 {
1198         static const uint32_t ptypes[] = {
1199                 RTE_PTYPE_INNER_L2_ETHER,
1200                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1201                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1202                 RTE_PTYPE_INNER_L3_IPV4,
1203                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1204                 RTE_PTYPE_INNER_L3_IPV6,
1205                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1206                 RTE_PTYPE_INNER_L4_FRAG,
1207                 RTE_PTYPE_INNER_L4_UDP,
1208                 RTE_PTYPE_INNER_L4_TCP,
1209                 RTE_PTYPE_INNER_L4_SCTP,
1210                 RTE_PTYPE_L2_ETHER,
1211                 RTE_PTYPE_L2_ETHER_VLAN,
1212                 RTE_PTYPE_L2_ETHER_QINQ,
1213                 RTE_PTYPE_L3_IPV4,
1214                 RTE_PTYPE_L3_IPV4_EXT,
1215                 RTE_PTYPE_L3_IPV6_EXT,
1216                 RTE_PTYPE_L3_IPV6,
1217                 RTE_PTYPE_L4_FRAG,
1218                 RTE_PTYPE_L4_UDP,
1219                 RTE_PTYPE_L4_TCP,
1220                 RTE_PTYPE_L4_SCTP,
1221         };
1222
1223         return ptypes;
1224 }
1225
1226 static int
1227 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1228                   struct rte_eth_fc_conf *fc_conf)
1229 {
1230         fc_conf->mode = RTE_FC_NONE;
1231         return 0;
1232 }
1233
1234 static int
1235 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1236                   struct rte_eth_fc_conf *fc_conf)
1237 {
1238         if (fc_conf->mode != RTE_FC_NONE)
1239                 return -ENOTSUP;
1240         return 0;
1241 }
1242
1243 static const struct eth_dev_ops ops = {
1244         .dev_start              = tap_dev_start,
1245         .dev_stop               = tap_dev_stop,
1246         .dev_close              = tap_dev_close,
1247         .dev_configure          = tap_dev_configure,
1248         .dev_infos_get          = tap_dev_info,
1249         .rx_queue_setup         = tap_rx_queue_setup,
1250         .tx_queue_setup         = tap_tx_queue_setup,
1251         .rx_queue_release       = tap_rx_queue_release,
1252         .tx_queue_release       = tap_tx_queue_release,
1253         .flow_ctrl_get          = tap_flow_ctrl_get,
1254         .flow_ctrl_set          = tap_flow_ctrl_set,
1255         .link_update            = tap_link_update,
1256         .dev_set_link_up        = tap_link_set_up,
1257         .dev_set_link_down      = tap_link_set_down,
1258         .promiscuous_enable     = tap_promisc_enable,
1259         .promiscuous_disable    = tap_promisc_disable,
1260         .allmulticast_enable    = tap_allmulti_enable,
1261         .allmulticast_disable   = tap_allmulti_disable,
1262         .mac_addr_set           = tap_mac_set,
1263         .mtu_set                = tap_mtu_set,
1264         .set_mc_addr_list       = tap_set_mc_addr_list,
1265         .stats_get              = tap_stats_get,
1266         .stats_reset            = tap_stats_reset,
1267         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1268         .filter_ctrl            = tap_dev_filter_ctrl,
1269 };
1270
1271 static int
1272 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1273                    char *remote_iface, int fixed_mac_type)
1274 {
1275         int numa_node = rte_socket_id();
1276         struct rte_eth_dev *dev;
1277         struct pmd_internals *pmd;
1278         struct rte_eth_dev_data *data;
1279         struct ifreq ifr;
1280         int i;
1281
1282         RTE_LOG(DEBUG, PMD, "  TAP device on numa %u\n", rte_socket_id());
1283
1284         data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
1285         if (!data) {
1286                 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
1287                 goto error_exit;
1288         }
1289
1290         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1291         if (!dev) {
1292                 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
1293                 goto error_exit;
1294         }
1295
1296         pmd = dev->data->dev_private;
1297         pmd->dev = dev;
1298         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1299
1300         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1301         if (pmd->ioctl_sock == -1) {
1302                 RTE_LOG(ERR, PMD,
1303                         "TAP Unable to get a socket for management: %s\n",
1304                         strerror(errno));
1305                 goto error_exit;
1306         }
1307
1308         /* Setup some default values */
1309         rte_memcpy(data, dev->data, sizeof(*data));
1310         data->dev_private = pmd;
1311         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1312         data->numa_node = numa_node;
1313
1314         data->dev_link = pmd_link;
1315         data->mac_addrs = &pmd->eth_addr;
1316         /* Set the number of RX and TX queues */
1317         data->nb_rx_queues = 0;
1318         data->nb_tx_queues = 0;
1319
1320         dev->data = data;
1321         dev->dev_ops = &ops;
1322         dev->rx_pkt_burst = pmd_rx_burst;
1323         dev->tx_pkt_burst = pmd_tx_burst;
1324
1325         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1326         pmd->intr_handle.fd = -1;
1327
1328         /* Presetup the fds to -1 as being not valid */
1329         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1330                 pmd->rxq[i].fd = -1;
1331                 pmd->txq[i].fd = -1;
1332         }
1333
1334         if (fixed_mac_type) {
1335                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1336                 static int iface_idx;
1337                 char mac[ETHER_ADDR_LEN] = "\0dtap";
1338
1339                 mac[ETHER_ADDR_LEN - 1] = iface_idx++;
1340                 rte_memcpy(&pmd->eth_addr, mac, ETHER_ADDR_LEN);
1341         } else {
1342                 eth_random_addr((uint8_t *)&pmd->eth_addr);
1343         }
1344
1345         /* Immediately create the netdevice (this will create the 1st queue). */
1346         /* rx queue */
1347         if (tap_setup_queue(dev, pmd, 0, 1) == -1)
1348                 goto error_exit;
1349         /* tx queue */
1350         if (tap_setup_queue(dev, pmd, 0, 0) == -1)
1351                 goto error_exit;
1352
1353         ifr.ifr_mtu = dev->data->mtu;
1354         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1355                 goto error_exit;
1356
1357         memset(&ifr, 0, sizeof(struct ifreq));
1358         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1359         rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr, ETHER_ADDR_LEN);
1360         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1361                 goto error_exit;
1362
1363         /*
1364          * Set up everything related to rte_flow:
1365          * - netlink socket
1366          * - tap / remote if_index
1367          * - mandatory QDISCs
1368          * - rte_flow actual/implicit lists
1369          * - implicit rules
1370          */
1371         pmd->nlsk_fd = tap_nl_init(0);
1372         if (pmd->nlsk_fd == -1) {
1373                 RTE_LOG(WARNING, PMD, "%s: failed to create netlink socket.\n",
1374                         pmd->name);
1375                 goto disable_rte_flow;
1376         }
1377         pmd->if_index = if_nametoindex(pmd->name);
1378         if (!pmd->if_index) {
1379                 RTE_LOG(ERR, PMD, "%s: failed to get if_index.\n", pmd->name);
1380                 goto disable_rte_flow;
1381         }
1382         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1383                 RTE_LOG(ERR, PMD, "%s: failed to create multiq qdisc.\n",
1384                         pmd->name);
1385                 goto disable_rte_flow;
1386         }
1387         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1388                 RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1389                         pmd->name);
1390                 goto disable_rte_flow;
1391         }
1392         LIST_INIT(&pmd->flows);
1393
1394         if (strlen(remote_iface)) {
1395                 pmd->remote_if_index = if_nametoindex(remote_iface);
1396                 if (!pmd->remote_if_index) {
1397                         RTE_LOG(ERR, PMD, "%s: failed to get %s if_index.\n",
1398                                 pmd->name, remote_iface);
1399                         goto error_remote;
1400                 }
1401                 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1402                          "%s", remote_iface);
1403
1404                 /* Save state of remote device */
1405                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1406
1407                 /* Replicate remote MAC address */
1408                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1409                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1410                                 pmd->name, pmd->remote_iface);
1411                         goto error_remote;
1412                 }
1413                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1414                            ETHER_ADDR_LEN);
1415                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1416                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1417                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1418                                 pmd->name, remote_iface);
1419                         goto error_remote;
1420                 }
1421
1422                 /*
1423                  * Flush usually returns negative value because it tries to
1424                  * delete every QDISC (and on a running device, one QDISC at
1425                  * least is needed). Ignore negative return value.
1426                  */
1427                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1428                 if (qdisc_create_ingress(pmd->nlsk_fd,
1429                                          pmd->remote_if_index) < 0) {
1430                         RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1431                                 pmd->remote_iface);
1432                         goto error_remote;
1433                 }
1434                 LIST_INIT(&pmd->implicit_flows);
1435                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1436                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1437                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1438                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1439                         RTE_LOG(ERR, PMD,
1440                                 "%s: failed to create implicit rules.\n",
1441                                 pmd->name);
1442                         goto error_remote;
1443                 }
1444         }
1445
1446         return 0;
1447
1448 disable_rte_flow:
1449         RTE_LOG(ERR, PMD, " Disabling rte flow support: %s(%d)\n",
1450                 strerror(errno), errno);
1451         if (strlen(remote_iface)) {
1452                 RTE_LOG(ERR, PMD, "Remote feature requires flow support.\n");
1453                 goto error_exit;
1454         }
1455         return 0;
1456
1457 error_remote:
1458         RTE_LOG(ERR, PMD, " Can't set up remote feature: %s(%d)\n",
1459                 strerror(errno), errno);
1460         tap_flow_implicit_flush(pmd, NULL);
1461
1462 error_exit:
1463         RTE_LOG(ERR, PMD, "TAP Unable to initialize %s\n",
1464                 rte_vdev_device_name(vdev));
1465
1466         rte_free(data);
1467         return -EINVAL;
1468 }
1469
1470 static int
1471 set_interface_name(const char *key __rte_unused,
1472                    const char *value,
1473                    void *extra_args)
1474 {
1475         char *name = (char *)extra_args;
1476
1477         if (value)
1478                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
1479         else
1480                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1481                          DEFAULT_TAP_NAME, (tap_unit - 1));
1482
1483         return 0;
1484 }
1485
1486 static int
1487 set_interface_speed(const char *key __rte_unused,
1488                     const char *value,
1489                     void *extra_args)
1490 {
1491         *(int *)extra_args = (value) ? atoi(value) : ETH_SPEED_NUM_10G;
1492
1493         return 0;
1494 }
1495
1496 static int
1497 set_remote_iface(const char *key __rte_unused,
1498                  const char *value,
1499                  void *extra_args)
1500 {
1501         char *name = (char *)extra_args;
1502
1503         if (value)
1504                 snprintf(name, RTE_ETH_NAME_MAX_LEN, "%s", value);
1505
1506         return 0;
1507 }
1508
1509 static int
1510 set_mac_type(const char *key __rte_unused,
1511              const char *value,
1512              void *extra_args)
1513 {
1514         if (value &&
1515             !strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED)))
1516                 *(int *)extra_args = 1;
1517         return 0;
1518 }
1519
1520 /* Open a TAP interface device.
1521  */
1522 static int
1523 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1524 {
1525         const char *name, *params;
1526         int ret;
1527         struct rte_kvargs *kvlist = NULL;
1528         int speed;
1529         char tap_name[RTE_ETH_NAME_MAX_LEN];
1530         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1531         int fixed_mac_type = 0;
1532
1533         name = rte_vdev_device_name(dev);
1534         params = rte_vdev_device_args(dev);
1535
1536         speed = ETH_SPEED_NUM_10G;
1537         snprintf(tap_name, sizeof(tap_name), "%s%d",
1538                  DEFAULT_TAP_NAME, tap_unit++);
1539         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1540
1541         if (params && (params[0] != '\0')) {
1542                 RTE_LOG(DEBUG, PMD, "parameters (%s)\n", params);
1543
1544                 kvlist = rte_kvargs_parse(params, valid_arguments);
1545                 if (kvlist) {
1546                         if (rte_kvargs_count(kvlist, ETH_TAP_SPEED_ARG) == 1) {
1547                                 ret = rte_kvargs_process(kvlist,
1548                                                          ETH_TAP_SPEED_ARG,
1549                                                          &set_interface_speed,
1550                                                          &speed);
1551                                 if (ret == -1)
1552                                         goto leave;
1553                         }
1554
1555                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1556                                 ret = rte_kvargs_process(kvlist,
1557                                                          ETH_TAP_IFACE_ARG,
1558                                                          &set_interface_name,
1559                                                          tap_name);
1560                                 if (ret == -1)
1561                                         goto leave;
1562                         }
1563
1564                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1565                                 ret = rte_kvargs_process(kvlist,
1566                                                          ETH_TAP_REMOTE_ARG,
1567                                                          &set_remote_iface,
1568                                                          remote_iface);
1569                                 if (ret == -1)
1570                                         goto leave;
1571                         }
1572
1573                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1574                                 ret = rte_kvargs_process(kvlist,
1575                                                          ETH_TAP_MAC_ARG,
1576                                                          &set_mac_type,
1577                                                          &fixed_mac_type);
1578                                 if (ret == -1)
1579                                         goto leave;
1580                         }
1581                 }
1582         }
1583         pmd_link.link_speed = speed;
1584
1585         RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1586                 name, tap_name);
1587
1588         ret = eth_dev_tap_create(dev, tap_name, remote_iface, fixed_mac_type);
1589
1590 leave:
1591         if (ret == -1) {
1592                 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1593                         name, tap_name);
1594                 tap_unit--;             /* Restore the unit number */
1595         }
1596         rte_kvargs_free(kvlist);
1597
1598         return ret;
1599 }
1600
1601 /* detach a TAP device.
1602  */
1603 static int
1604 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1605 {
1606         struct rte_eth_dev *eth_dev = NULL;
1607         struct pmd_internals *internals;
1608         int i;
1609
1610         RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1611                 rte_socket_id());
1612
1613         /* find the ethdev entry */
1614         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1615         if (!eth_dev)
1616                 return 0;
1617
1618         internals = eth_dev->data->dev_private;
1619         if (internals->nlsk_fd) {
1620                 tap_flow_flush(eth_dev, NULL);
1621                 tap_flow_implicit_flush(internals, NULL);
1622                 tap_nl_final(internals->nlsk_fd);
1623         }
1624         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1625                 if (internals->rxq[i].fd != -1) {
1626                         close(internals->rxq[i].fd);
1627                         internals->rxq[i].fd = -1;
1628                 }
1629                 if (internals->txq[i].fd != -1) {
1630                         close(internals->txq[i].fd);
1631                         internals->txq[i].fd = -1;
1632                 }
1633         }
1634
1635         close(internals->ioctl_sock);
1636         rte_free(eth_dev->data->dev_private);
1637         rte_free(eth_dev->data);
1638
1639         rte_eth_dev_release_port(eth_dev);
1640
1641         return 0;
1642 }
1643
1644 static struct rte_vdev_driver pmd_tap_drv = {
1645         .probe = rte_pmd_tap_probe,
1646         .remove = rte_pmd_tap_remove,
1647 };
1648 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1649 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1650 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1651                               ETH_TAP_IFACE_ARG "=<string> "
1652                               ETH_TAP_SPEED_ARG "=<int> "
1653                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_FIXED " "
1654                               ETH_TAP_REMOTE_ARG "=<string>");