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_driver.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 #include <rte_string_fns.h>
  19
  20 #include <sys/types.h>
  21 #include <sys/stat.h>
  22 #include <sys/socket.h>
  23 #include <sys/ioctl.h>
  24 #include <sys/utsname.h>
  25 #include <sys/mman.h>
  26 #include <errno.h>
  27 #include <signal.h>
  28 #include <stdbool.h>
  29 #include <stdint.h>
  30 #include <sys/uio.h>
  31 #include <unistd.h>
  32 #include <arpa/inet.h>
  33 #include <net/if.h>
  34 #include <linux/if_tun.h>
  35 #include <linux/if_ether.h>
  36 #include <fcntl.h>
  37
  38 #include <tap_rss.h>
  39 #include <rte_eth_tap.h>
  40 #include <tap_flow.h>
  41 #include <tap_netlink.h>
  42 #include <tap_tcmsgs.h>
  43
  44 /* Linux based path to the TUN device */
  45 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  46 #define DEFAULT_TAP_NAME        "dtap"
  47 #define DEFAULT_TUN_NAME        "dtun"
  48
  49 #define ETH_TAP_IFACE_ARG       "iface"
  50 #define ETH_TAP_REMOTE_ARG      "remote"
  51 #define ETH_TAP_MAC_ARG         "mac"
  52 #define ETH_TAP_MAC_FIXED       "fixed"
  53
  54 #define ETH_TAP_USR_MAC_FMT     "xx:xx:xx:xx:xx:xx"
  55 #define ETH_TAP_CMP_MAC_FMT     "0123456789ABCDEFabcdef"
  56 #define ETH_TAP_MAC_ARG_FMT     ETH_TAP_MAC_FIXED "|" ETH_TAP_USR_MAC_FMT
  57
  58 static struct rte_vdev_driver pmd_tap_drv;
  59 static struct rte_vdev_driver pmd_tun_drv;
  60
  61 static const char *valid_arguments[] = {
  62         ETH_TAP_IFACE_ARG,
  63         ETH_TAP_REMOTE_ARG,
  64         ETH_TAP_MAC_ARG,
  65         NULL
  66 };
  67
  68 static int tap_unit;
  69 static unsigned int tun_unit;
  70
  71 static int tap_type;
  72 static char tuntap_name[8];
  73
  74 static volatile uint32_t tap_trigger;   /* Rx trigger */
  75
  76 static struct rte_eth_link pmd_link = {
  77         .link_speed = ETH_SPEED_NUM_10G,
  78         .link_duplex = ETH_LINK_FULL_DUPLEX,
  79         .link_status = ETH_LINK_DOWN,
  80         .link_autoneg = ETH_LINK_FIXED,
  81 };
  82
  83 static void
  84 tap_trigger_cb(int sig __rte_unused)
  85 {
  86         /* Valid trigger values are nonzero */
  87         tap_trigger = (tap_trigger + 1) | 0x80000000;
  88 }
  89
  90 /* Specifies on what netdevices the ioctl should be applied */
  91 enum ioctl_mode {
  92         LOCAL_AND_REMOTE,
  93         LOCAL_ONLY,
  94         REMOTE_ONLY,
  95 };
  96
  97 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
  98
  99 /* Tun/Tap allocation routine
 100  *
 101  * name is the number of the interface to use, unless NULL to take the host
 102  * supplied name.
 103  */
 104 static int
 105 tun_alloc(struct pmd_internals *pmd)
 106 {
 107         struct ifreq ifr;
 108 #ifdef IFF_MULTI_QUEUE
 109         unsigned int features;
 110 #endif
 111         int fd;
 112
 113         memset(&ifr, 0, sizeof(struct ifreq));
 114
 115         /*
 116          * Do not set IFF_NO_PI as packet information header will be needed
 117          * to check if a received packet has been truncated.
 118          */
 119         ifr.ifr_flags = (tap_type) ? IFF_TAP : IFF_TUN | IFF_POINTOPOINT;
 120         snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
 121
 122         TAP_LOG(DEBUG, "ifr_name '%s'", ifr.ifr_name);
 123
 124         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 125         if (fd < 0) {
 126                 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
 127                 goto error;
 128         }
 129
 130 #ifdef IFF_MULTI_QUEUE
 131         /* Grab the TUN features to verify we can work multi-queue */
 132         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 133                 TAP_LOG(ERR, "%s unable to get TUN/TAP features",
 134                         tuntap_name);
 135                 goto error;
 136         }
 137         TAP_LOG(DEBUG, "%s Features %08x", tuntap_name, features);
 138
 139         if (features & IFF_MULTI_QUEUE) {
 140                 TAP_LOG(DEBUG, "  Multi-queue support for %d queues",
 141                         RTE_PMD_TAP_MAX_QUEUES);
 142                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 143         } else
 144 #endif
 145         {
 146                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 147                 TAP_LOG(DEBUG, "  Single queue only support");
 148         }
 149
 150         /* Set the TUN/TAP configuration and set the name if needed */
 151         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 152                 TAP_LOG(WARNING, "Unable to set TUNSETIFF for %s: %s",
 153                         ifr.ifr_name, strerror(errno));
 154                 goto error;
 155         }
 156
 157         /* Always set the file descriptor to non-blocking */
 158         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 159                 TAP_LOG(WARNING,
 160                         "Unable to set %s to nonblocking: %s",
 161                         ifr.ifr_name, strerror(errno));
 162                 goto error;
 163         }
 164
 165         /* Set up trigger to optimize empty Rx bursts */
 166         errno = 0;
 167         do {
 168                 struct sigaction sa;
 169                 int flags = fcntl(fd, F_GETFL);
 170
 171                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 172                         break;
 173                 if (sa.sa_handler != tap_trigger_cb) {
 174                         /*
 175                          * Make sure SIGIO is not already taken. This is done
 176                          * as late as possible to leave the application a
 177                          * chance to set up its own signal handler first.
 178                          */
 179                         if (sa.sa_handler != SIG_IGN &&
 180                             sa.sa_handler != SIG_DFL) {
 181                                 errno = EBUSY;
 182                                 break;
 183                         }
 184                         sa = (struct sigaction){
 185                                 .sa_flags = SA_RESTART,
 186                                 .sa_handler = tap_trigger_cb,
 187                         };
 188                         if (sigaction(SIGIO, &sa, NULL) == -1)
 189                                 break;
 190                 }
 191                 /* Enable SIGIO on file descriptor */
 192                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 193                 fcntl(fd, F_SETOWN, getpid());
 194         } while (0);
 195
 196         if (errno) {
 197                 /* Disable trigger globally in case of error */
 198                 tap_trigger = 0;
 199                 TAP_LOG(WARNING, "Rx trigger disabled: %s",
 200                         strerror(errno));
 201         }
 202
 203         return fd;
 204
 205 error:
 206         if (fd > 0)
 207                 close(fd);
 208         return -1;
 209 }
 210
 211 static void
 212 tap_verify_csum(struct rte_mbuf *mbuf)
 213 {
 214         uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
 215         uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
 216         uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
 217         unsigned int l2_len = sizeof(struct ether_hdr);
 218         unsigned int l3_len;
 219         uint16_t cksum = 0;
 220         void *l3_hdr;
 221         void *l4_hdr;
 222
 223         if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
 224                 l2_len += 4;
 225         else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
 226                 l2_len += 8;
 227         /* Don't verify checksum for packets with discontinuous L2 header */
 228         if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
 229                      rte_pktmbuf_data_len(mbuf)))
 230                 return;
 231         l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
 232         if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 233                 struct ipv4_hdr *iph = l3_hdr;
 234
 235                 /* ihl contains the number of 4-byte words in the header */
 236                 l3_len = 4 * (iph->version_ihl & 0xf);
 237                 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
 238                         return;
 239
 240                 cksum = ~rte_raw_cksum(iph, l3_len);
 241                 mbuf->ol_flags |= cksum ?
 242                         PKT_RX_IP_CKSUM_BAD :
 243                         PKT_RX_IP_CKSUM_GOOD;
 244         } else if (l3 == RTE_PTYPE_L3_IPV6) {
 245                 l3_len = sizeof(struct ipv6_hdr);
 246         } else {
 247                 /* IPv6 extensions are not supported */
 248                 return;
 249         }
 250         if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
 251                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 252                 /* Don't verify checksum for multi-segment packets. */
 253                 if (mbuf->nb_segs > 1)
 254                         return;
 255                 if (l3 == RTE_PTYPE_L3_IPV4)
 256                         cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
 257                 else if (l3 == RTE_PTYPE_L3_IPV6)
 258                         cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
 259                 mbuf->ol_flags |= cksum ?
 260                         PKT_RX_L4_CKSUM_BAD :
 261                         PKT_RX_L4_CKSUM_GOOD;
 262         }
 263 }
 264
 265 static uint64_t
 266 tap_rx_offload_get_port_capa(void)
 267 {
 268         /*
 269          * No specific port Rx offload capabilities.
 270          */
 271         return 0;
 272 }
 273
 274 static uint64_t
 275 tap_rx_offload_get_queue_capa(void)
 276 {
 277         return DEV_RX_OFFLOAD_SCATTER |
 278                DEV_RX_OFFLOAD_IPV4_CKSUM |
 279                DEV_RX_OFFLOAD_UDP_CKSUM |
 280                DEV_RX_OFFLOAD_TCP_CKSUM |
 281                DEV_RX_OFFLOAD_CRC_STRIP;
 282 }
 283
 284 /* Callback to handle the rx burst of packets to the correct interface and
 285  * file descriptor(s) in a multi-queue setup.
 286  */
 287 static uint16_t
 288 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 289 {
 290         struct rx_queue *rxq = queue;
 291         uint16_t num_rx;
 292         unsigned long num_rx_bytes = 0;
 293         uint32_t trigger = tap_trigger;
 294
 295         if (trigger == rxq->trigger_seen)
 296                 return 0;
 297         if (trigger)
 298                 rxq->trigger_seen = trigger;
 299         rte_compiler_barrier();
 300         for (num_rx = 0; num_rx < nb_pkts; ) {
 301                 struct rte_mbuf *mbuf = rxq->pool;
 302                 struct rte_mbuf *seg = NULL;
 303                 struct rte_mbuf *new_tail = NULL;
 304                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 305                 int len;
 306
 307                 len = readv(rxq->fd, *rxq->iovecs,
 308                             1 +
 309                             (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
 310                              rxq->nb_rx_desc : 1));
 311                 if (len < (int)sizeof(struct tun_pi))
 312                         break;
 313
 314                 /* Packet couldn't fit in the provided mbuf */
 315                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 316                         rxq->stats.ierrors++;
 317                         continue;
 318                 }
 319
 320                 len -= sizeof(struct tun_pi);
 321
 322                 mbuf->pkt_len = len;
 323                 mbuf->port = rxq->in_port;
 324                 while (1) {
 325                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 326
 327                         if (unlikely(!buf)) {
 328                                 rxq->stats.rx_nombuf++;
 329                                 /* No new buf has been allocated: do nothing */
 330                                 if (!new_tail || !seg)
 331                                         goto end;
 332
 333                                 seg->next = NULL;
 334                                 rte_pktmbuf_free(mbuf);
 335
 336                                 goto end;
 337                         }
 338                         seg = seg ? seg->next : mbuf;
 339                         if (rxq->pool == mbuf)
 340                                 rxq->pool = buf;
 341                         if (new_tail)
 342                                 new_tail->next = buf;
 343                         new_tail = buf;
 344                         new_tail->next = seg->next;
 345
 346                         /* iovecs[0] is reserved for packet info (pi) */
 347                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 348                                 buf->buf_len - data_off;
 349                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 350                                 (char *)buf->buf_addr + data_off;
 351
 352                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 353                         seg->data_off = data_off;
 354
 355                         len -= seg->data_len;
 356                         if (len <= 0)
 357                                 break;
 358                         mbuf->nb_segs++;
 359                         /* First segment has headroom, not the others */
 360                         data_off = 0;
 361                 }
 362                 seg->next = NULL;
 363                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 364                                                       RTE_PTYPE_ALL_MASK);
 365                 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
 366                         tap_verify_csum(mbuf);
 367
 368                 /* account for the receive frame */
 369                 bufs[num_rx++] = mbuf;
 370                 num_rx_bytes += mbuf->pkt_len;
 371         }
 372 end:
 373         rxq->stats.ipackets += num_rx;
 374         rxq->stats.ibytes += num_rx_bytes;
 375
 376         return num_rx;
 377 }
 378
 379 static uint64_t
 380 tap_tx_offload_get_port_capa(void)
 381 {
 382         /*
 383          * No specific port Tx offload capabilities.
 384          */
 385         return 0;
 386 }
 387
 388 static uint64_t
 389 tap_tx_offload_get_queue_capa(void)
 390 {
 391         return DEV_TX_OFFLOAD_MULTI_SEGS |
 392                DEV_TX_OFFLOAD_IPV4_CKSUM |
 393                DEV_TX_OFFLOAD_UDP_CKSUM |
 394                DEV_TX_OFFLOAD_TCP_CKSUM;
 395 }
 396
 397 static void
 398 tap_tx_offload(char *packet, uint64_t ol_flags, unsigned int l2_len,
 399                unsigned int l3_len)
 400 {
 401         void *l3_hdr = packet + l2_len;
 402
 403         if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
 404                 struct ipv4_hdr *iph = l3_hdr;
 405                 uint16_t cksum;
 406
 407                 iph->hdr_checksum = 0;
 408                 cksum = rte_raw_cksum(iph, l3_len);
 409                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 410         }
 411         if (ol_flags & PKT_TX_L4_MASK) {
 412                 uint16_t l4_len;
 413                 uint32_t cksum;
 414                 uint16_t *l4_cksum;
 415                 void *l4_hdr;
 416
 417                 l4_hdr = packet + l2_len + l3_len;
 418                 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
 419                         l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
 420                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 421                         l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
 422                 else
 423                         return;
 424                 *l4_cksum = 0;
 425                 if (ol_flags & PKT_TX_IPV4) {
 426                         struct ipv4_hdr *iph = l3_hdr;
 427
 428                         l4_len = rte_be_to_cpu_16(iph->total_length) - l3_len;
 429                         cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 430                 } else {
 431                         struct ipv6_hdr *ip6h = l3_hdr;
 432
 433                         /* payload_len does not include ext headers */
 434                         l4_len = rte_be_to_cpu_16(ip6h->payload_len) -
 435                                 l3_len + sizeof(struct ipv6_hdr);
 436                         cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 437                 }
 438                 cksum += rte_raw_cksum(l4_hdr, l4_len);
 439                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 440                 cksum = (~cksum) & 0xffff;
 441                 if (cksum == 0)
 442                         cksum = 0xffff;
 443                 *l4_cksum = cksum;
 444         }
 445 }
 446
 447 /* Callback to handle sending packets from the tap interface
 448  */
 449 static uint16_t
 450 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 451 {
 452         struct tx_queue *txq = queue;
 453         uint16_t num_tx = 0;
 454         unsigned long num_tx_bytes = 0;
 455         uint32_t max_size;
 456         int i;
 457
 458         if (unlikely(nb_pkts == 0))
 459                 return 0;
 460
 461         max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
 462         for (i = 0; i < nb_pkts; i++) {
 463                 struct rte_mbuf *mbuf = bufs[num_tx];
 464                 struct iovec iovecs[mbuf->nb_segs + 1];
 465                 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
 466                 struct rte_mbuf *seg = mbuf;
 467                 char m_copy[mbuf->data_len];
 468                 int n;
 469                 int j;
 470
 471                 /* stats.errs will be incremented */
 472                 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
 473                         break;
 474
 475                 /*
 476                  * TUN and TAP are created with IFF_NO_PI disabled.
 477                  * For TUN PMD this mandatory as fields are used by
 478                  * Kernel tun.c to determine whether its IP or non IP
 479                  * packets.
 480                  *
 481                  * The logic fetches the first byte of data from mbuf.
 482                  * compares whether its v4 or v6. If none matches default
 483                  * value 0x00 is taken for protocol field.
 484                  */
 485                 char *buff_data = rte_pktmbuf_mtod(seg, void *);
 486                 j = (*buff_data & 0xf0);
 487                 pi.proto = (j == 0x40) ? 0x0008 :
 488                                 (j == 0x60) ? 0xdd86 : 0x00;
 489
 490                 iovecs[0].iov_base = &pi;
 491                 iovecs[0].iov_len = sizeof(pi);
 492                 for (j = 1; j <= mbuf->nb_segs; j++) {
 493                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 494                         iovecs[j].iov_base =
 495                                 rte_pktmbuf_mtod(seg, void *);
 496                         seg = seg->next;
 497                 }
 498                 if (txq->csum &&
 499                     ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 500                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 501                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
 502                         /* Support only packets with all data in the same seg */
 503                         if (mbuf->nb_segs > 1)
 504                                 break;
 505                         /* To change checksums, work on a copy of data. */
 506                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 507                                    rte_pktmbuf_data_len(mbuf));
 508                         tap_tx_offload(m_copy, mbuf->ol_flags,
 509                                        mbuf->l2_len, mbuf->l3_len);
 510                         iovecs[1].iov_base = m_copy;
 511                 }
 512                 /* copy the tx frame data */
 513                 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
 514                 if (n <= 0)
 515                         break;
 516
 517                 num_tx++;
 518                 num_tx_bytes += mbuf->pkt_len;
 519                 rte_pktmbuf_free(mbuf);
 520         }
 521
 522         txq->stats.opackets += num_tx;
 523         txq->stats.errs += nb_pkts - num_tx;
 524         txq->stats.obytes += num_tx_bytes;
 525
 526         return num_tx;
 527 }
 528
 529 static const char *
 530 tap_ioctl_req2str(unsigned long request)
 531 {
 532         switch (request) {
 533         case SIOCSIFFLAGS:
 534                 return "SIOCSIFFLAGS";
 535         case SIOCGIFFLAGS:
 536                 return "SIOCGIFFLAGS";
 537         case SIOCGIFHWADDR:
 538                 return "SIOCGIFHWADDR";
 539         case SIOCSIFHWADDR:
 540                 return "SIOCSIFHWADDR";
 541         case SIOCSIFMTU:
 542                 return "SIOCSIFMTU";
 543         }
 544         return "UNKNOWN";
 545 }
 546
 547 static int
 548 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 549           struct ifreq *ifr, int set, enum ioctl_mode mode)
 550 {
 551         short req_flags = ifr->ifr_flags;
 552         int remote = pmd->remote_if_index &&
 553                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 554
 555         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 556                 return 0;
 557         /*
 558          * If there is a remote netdevice, apply ioctl on it, then apply it on
 559          * the tap netdevice.
 560          */
 561 apply:
 562         if (remote)
 563                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
 564         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 565                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 566         switch (request) {
 567         case SIOCSIFFLAGS:
 568                 /* fetch current flags to leave other flags untouched */
 569                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 570                         goto error;
 571                 if (set)
 572                         ifr->ifr_flags |= req_flags;
 573                 else
 574                         ifr->ifr_flags &= ~req_flags;
 575                 break;
 576         case SIOCGIFFLAGS:
 577         case SIOCGIFHWADDR:
 578         case SIOCSIFHWADDR:
 579         case SIOCSIFMTU:
 580                 break;
 581         default:
 582                 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
 583                         pmd->name);
 584                 return -EINVAL;
 585         }
 586         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 587                 goto error;
 588         if (remote-- && mode == LOCAL_AND_REMOTE)
 589                 goto apply;
 590         return 0;
 591
 592 error:
 593         TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
 594                 tap_ioctl_req2str(request), strerror(errno), errno);
 595         return -errno;
 596 }
 597
 598 static int
 599 tap_link_set_down(struct rte_eth_dev *dev)
 600 {
 601         struct pmd_internals *pmd = dev->data->dev_private;
 602         struct ifreq ifr = { .ifr_flags = IFF_UP };
 603
 604         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 605         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 606 }
 607
 608 static int
 609 tap_link_set_up(struct rte_eth_dev *dev)
 610 {
 611         struct pmd_internals *pmd = dev->data->dev_private;
 612         struct ifreq ifr = { .ifr_flags = IFF_UP };
 613
 614         dev->data->dev_link.link_status = ETH_LINK_UP;
 615         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 616 }
 617
 618 static int
 619 tap_dev_start(struct rte_eth_dev *dev)
 620 {
 621         int err;
 622
 623         err = tap_intr_handle_set(dev, 1);
 624         if (err)
 625                 return err;
 626         return tap_link_set_up(dev);
 627 }
 628
 629 /* This function gets called when the current port gets stopped.
 630  */
 631 static void
 632 tap_dev_stop(struct rte_eth_dev *dev)
 633 {
 634         tap_intr_handle_set(dev, 0);
 635         tap_link_set_down(dev);
 636 }
 637
 638 static int
 639 tap_dev_configure(struct rte_eth_dev *dev)
 640 {
 641         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 642                 TAP_LOG(ERR,
 643                         "%s: number of rx queues %d exceeds max num of queues %d",
 644                         dev->device->name,
 645                         dev->data->nb_rx_queues,
 646                         RTE_PMD_TAP_MAX_QUEUES);
 647                 return -1;
 648         }
 649         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 650                 TAP_LOG(ERR,
 651                         "%s: number of tx queues %d exceeds max num of queues %d",
 652                         dev->device->name,
 653                         dev->data->nb_tx_queues,
 654                         RTE_PMD_TAP_MAX_QUEUES);
 655                 return -1;
 656         }
 657
 658         TAP_LOG(INFO, "%s: %p: TX configured queues number: %u",
 659                 dev->device->name, (void *)dev, dev->data->nb_tx_queues);
 660
 661         TAP_LOG(INFO, "%s: %p: RX configured queues number: %u",
 662                 dev->device->name, (void *)dev, dev->data->nb_rx_queues);
 663
 664         return 0;
 665 }
 666
 667 static uint32_t
 668 tap_dev_speed_capa(void)
 669 {
 670         uint32_t speed = pmd_link.link_speed;
 671         uint32_t capa = 0;
 672
 673         if (speed >= ETH_SPEED_NUM_10M)
 674                 capa |= ETH_LINK_SPEED_10M;
 675         if (speed >= ETH_SPEED_NUM_100M)
 676                 capa |= ETH_LINK_SPEED_100M;
 677         if (speed >= ETH_SPEED_NUM_1G)
 678                 capa |= ETH_LINK_SPEED_1G;
 679         if (speed >= ETH_SPEED_NUM_5G)
 680                 capa |= ETH_LINK_SPEED_2_5G;
 681         if (speed >= ETH_SPEED_NUM_5G)
 682                 capa |= ETH_LINK_SPEED_5G;
 683         if (speed >= ETH_SPEED_NUM_10G)
 684                 capa |= ETH_LINK_SPEED_10G;
 685         if (speed >= ETH_SPEED_NUM_20G)
 686                 capa |= ETH_LINK_SPEED_20G;
 687         if (speed >= ETH_SPEED_NUM_25G)
 688                 capa |= ETH_LINK_SPEED_25G;
 689         if (speed >= ETH_SPEED_NUM_40G)
 690                 capa |= ETH_LINK_SPEED_40G;
 691         if (speed >= ETH_SPEED_NUM_50G)
 692                 capa |= ETH_LINK_SPEED_50G;
 693         if (speed >= ETH_SPEED_NUM_56G)
 694                 capa |= ETH_LINK_SPEED_56G;
 695         if (speed >= ETH_SPEED_NUM_100G)
 696                 capa |= ETH_LINK_SPEED_100G;
 697
 698         return capa;
 699 }
 700
 701 static void
 702 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 703 {
 704         struct pmd_internals *internals = dev->data->dev_private;
 705
 706         dev_info->if_index = internals->if_index;
 707         dev_info->max_mac_addrs = 1;
 708         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 709         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
 710         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
 711         dev_info->min_rx_bufsize = 0;
 712         dev_info->speed_capa = tap_dev_speed_capa();
 713         dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
 714         dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
 715                                     dev_info->rx_queue_offload_capa;
 716         dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
 717         dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
 718                                     dev_info->tx_queue_offload_capa;
 719         dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
 720         /*
 721          * limitation: TAP supports all of IP, UDP and TCP hash
 722          * functions together and not in partial combinations
 723          */
 724         dev_info->flow_type_rss_offloads = ~TAP_RSS_HF_MASK;
 725 }
 726
 727 static int
 728 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 729 {
 730         unsigned int i, imax;
 731         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 732         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 733         unsigned long rx_nombuf = 0, ierrors = 0;
 734         const struct pmd_internals *pmd = dev->data->dev_private;
 735
 736         /* rx queue statistics */
 737         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 738                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 739         for (i = 0; i < imax; i++) {
 740                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 741                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 742                 rx_total += tap_stats->q_ipackets[i];
 743                 rx_bytes_total += tap_stats->q_ibytes[i];
 744                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
 745                 ierrors += pmd->rxq[i].stats.ierrors;
 746         }
 747
 748         /* tx queue statistics */
 749         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 750                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 751
 752         for (i = 0; i < imax; i++) {
 753                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 754                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 755                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 756                 tx_total += tap_stats->q_opackets[i];
 757                 tx_err_total += tap_stats->q_errors[i];
 758                 tx_bytes_total += tap_stats->q_obytes[i];
 759         }
 760
 761         tap_stats->ipackets = rx_total;
 762         tap_stats->ibytes = rx_bytes_total;
 763         tap_stats->ierrors = ierrors;
 764         tap_stats->rx_nombuf = rx_nombuf;
 765         tap_stats->opackets = tx_total;
 766         tap_stats->oerrors = tx_err_total;
 767         tap_stats->obytes = tx_bytes_total;
 768         return 0;
 769 }
 770
 771 static void
 772 tap_stats_reset(struct rte_eth_dev *dev)
 773 {
 774         int i;
 775         struct pmd_internals *pmd = dev->data->dev_private;
 776
 777         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 778                 pmd->rxq[i].stats.ipackets = 0;
 779                 pmd->rxq[i].stats.ibytes = 0;
 780                 pmd->rxq[i].stats.ierrors = 0;
 781                 pmd->rxq[i].stats.rx_nombuf = 0;
 782
 783                 pmd->txq[i].stats.opackets = 0;
 784                 pmd->txq[i].stats.errs = 0;
 785                 pmd->txq[i].stats.obytes = 0;
 786         }
 787 }
 788
 789 static void
 790 tap_dev_close(struct rte_eth_dev *dev)
 791 {
 792         int i;
 793         struct pmd_internals *internals = dev->data->dev_private;
 794
 795         tap_link_set_down(dev);
 796         tap_flow_flush(dev, NULL);
 797         tap_flow_implicit_flush(internals, NULL);
 798
 799         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 800                 if (internals->rxq[i].fd != -1) {
 801                         close(internals->rxq[i].fd);
 802                         internals->rxq[i].fd = -1;
 803                 }
 804                 if (internals->txq[i].fd != -1) {
 805                         close(internals->txq[i].fd);
 806                         internals->txq[i].fd = -1;
 807                 }
 808         }
 809
 810         if (internals->remote_if_index) {
 811                 /* Restore initial remote state */
 812                 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
 813                                 &internals->remote_initial_flags);
 814         }
 815 }
 816
 817 static void
 818 tap_rx_queue_release(void *queue)
 819 {
 820         struct rx_queue *rxq = queue;
 821
 822         if (rxq && (rxq->fd > 0)) {
 823                 close(rxq->fd);
 824                 rxq->fd = -1;
 825                 rte_pktmbuf_free(rxq->pool);
 826                 rte_free(rxq->iovecs);
 827                 rxq->pool = NULL;
 828                 rxq->iovecs = NULL;
 829         }
 830 }
 831
 832 static void
 833 tap_tx_queue_release(void *queue)
 834 {
 835         struct tx_queue *txq = queue;
 836
 837         if (txq && (txq->fd > 0)) {
 838                 close(txq->fd);
 839                 txq->fd = -1;
 840         }
 841 }
 842
 843 static int
 844 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
 845 {
 846         struct rte_eth_link *dev_link = &dev->data->dev_link;
 847         struct pmd_internals *pmd = dev->data->dev_private;
 848         struct ifreq ifr = { .ifr_flags = 0 };
 849
 850         if (pmd->remote_if_index) {
 851                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
 852                 if (!(ifr.ifr_flags & IFF_UP) ||
 853                     !(ifr.ifr_flags & IFF_RUNNING)) {
 854                         dev_link->link_status = ETH_LINK_DOWN;
 855                         return 0;
 856                 }
 857         }
 858         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
 859         dev_link->link_status =
 860                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
 861                  ETH_LINK_UP :
 862                  ETH_LINK_DOWN);
 863         return 0;
 864 }
 865
 866 static void
 867 tap_promisc_enable(struct rte_eth_dev *dev)
 868 {
 869         struct pmd_internals *pmd = dev->data->dev_private;
 870         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 871
 872         dev->data->promiscuous = 1;
 873         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 874         if (pmd->remote_if_index && !pmd->flow_isolate)
 875                 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
 876 }
 877
 878 static void
 879 tap_promisc_disable(struct rte_eth_dev *dev)
 880 {
 881         struct pmd_internals *pmd = dev->data->dev_private;
 882         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 883
 884         dev->data->promiscuous = 0;
 885         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 886         if (pmd->remote_if_index && !pmd->flow_isolate)
 887                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
 888 }
 889
 890 static void
 891 tap_allmulti_enable(struct rte_eth_dev *dev)
 892 {
 893         struct pmd_internals *pmd = dev->data->dev_private;
 894         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 895
 896         dev->data->all_multicast = 1;
 897         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 898         if (pmd->remote_if_index && !pmd->flow_isolate)
 899                 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
 900 }
 901
 902 static void
 903 tap_allmulti_disable(struct rte_eth_dev *dev)
 904 {
 905         struct pmd_internals *pmd = dev->data->dev_private;
 906         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 907
 908         dev->data->all_multicast = 0;
 909         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 910         if (pmd->remote_if_index && !pmd->flow_isolate)
 911                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
 912 }
 913
 914 static int
 915 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 916 {
 917         struct pmd_internals *pmd = dev->data->dev_private;
 918         enum ioctl_mode mode = LOCAL_ONLY;
 919         struct ifreq ifr;
 920         int ret;
 921
 922         if (is_zero_ether_addr(mac_addr)) {
 923                 TAP_LOG(ERR, "%s: can't set an empty MAC address",
 924                         dev->device->name);
 925                 return -EINVAL;
 926         }
 927         /* Check the actual current MAC address on the tap netdevice */
 928         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
 929         if (ret < 0)
 930                 return ret;
 931         if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 932                                mac_addr))
 933                 return 0;
 934         /* Check the current MAC address on the remote */
 935         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
 936         if (ret < 0)
 937                 return ret;
 938         if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 939                                mac_addr))
 940                 mode = LOCAL_AND_REMOTE;
 941         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
 942         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
 943         ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
 944         if (ret < 0)
 945                 return ret;
 946         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
 947         if (pmd->remote_if_index && !pmd->flow_isolate) {
 948                 /* Replace MAC redirection rule after a MAC change */
 949                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
 950                 if (ret < 0) {
 951                         TAP_LOG(ERR,
 952                                 "%s: Couldn't delete MAC redirection rule",
 953                                 dev->device->name);
 954                         return ret;
 955                 }
 956                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
 957                 if (ret < 0) {
 958                         TAP_LOG(ERR,
 959                                 "%s: Couldn't add MAC redirection rule",
 960                                 dev->device->name);
 961                         return ret;
 962                 }
 963         }
 964
 965         return 0;
 966 }
 967
 968 static int
 969 tap_setup_queue(struct rte_eth_dev *dev,
 970                 struct pmd_internals *internals,
 971                 uint16_t qid,
 972                 int is_rx)
 973 {
 974         int *fd;
 975         int *other_fd;
 976         const char *dir;
 977         struct pmd_internals *pmd = dev->data->dev_private;
 978         struct rx_queue *rx = &internals->rxq[qid];
 979         struct tx_queue *tx = &internals->txq[qid];
 980
 981         if (is_rx) {
 982                 fd = &rx->fd;
 983                 other_fd = &tx->fd;
 984                 dir = "rx";
 985         } else {
 986                 fd = &tx->fd;
 987                 other_fd = &rx->fd;
 988                 dir = "tx";
 989         }
 990         if (*fd != -1) {
 991                 /* fd for this queue already exists */
 992                 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
 993                         pmd->name, *fd, dir, qid);
 994         } else if (*other_fd != -1) {
 995                 /* Only other_fd exists. dup it */
 996                 *fd = dup(*other_fd);
 997                 if (*fd < 0) {
 998                         *fd = -1;
 999                         TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1000                         return -1;
1001                 }
1002                 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1003                         pmd->name, *other_fd, dir, qid, *fd);
1004         } else {
1005                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1006                 *fd = tun_alloc(pmd);
1007                 if (*fd < 0) {
1008                         *fd = -1; /* restore original value */
1009                         TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1010                         return -1;
1011                 }
1012                 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1013                         pmd->name, dir, qid, *fd);
1014         }
1015
1016         tx->mtu = &dev->data->mtu;
1017         rx->rxmode = &dev->data->dev_conf.rxmode;
1018
1019         return *fd;
1020 }
1021
1022 static int
1023 tap_rx_queue_setup(struct rte_eth_dev *dev,
1024                    uint16_t rx_queue_id,
1025                    uint16_t nb_rx_desc,
1026                    unsigned int socket_id,
1027                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1028                    struct rte_mempool *mp)
1029 {
1030         struct pmd_internals *internals = dev->data->dev_private;
1031         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1032         struct rte_mbuf **tmp = &rxq->pool;
1033         long iov_max = sysconf(_SC_IOV_MAX);
1034         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1035         struct iovec (*iovecs)[nb_desc + 1];
1036         int data_off = RTE_PKTMBUF_HEADROOM;
1037         int ret = 0;
1038         int fd;
1039         int i;
1040
1041         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1042                 TAP_LOG(WARNING,
1043                         "nb_rx_queues %d too small or mempool NULL",
1044                         dev->data->nb_rx_queues);
1045                 return -1;
1046         }
1047
1048         rxq->mp = mp;
1049         rxq->trigger_seen = 1; /* force initial burst */
1050         rxq->in_port = dev->data->port_id;
1051         rxq->nb_rx_desc = nb_desc;
1052         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1053                                     socket_id);
1054         if (!iovecs) {
1055                 TAP_LOG(WARNING,
1056                         "%s: Couldn't allocate %d RX descriptors",
1057                         dev->device->name, nb_desc);
1058                 return -ENOMEM;
1059         }
1060         rxq->iovecs = iovecs;
1061
1062         dev->data->rx_queues[rx_queue_id] = rxq;
1063         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1064         if (fd == -1) {
1065                 ret = fd;
1066                 goto error;
1067         }
1068
1069         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1070         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1071
1072         for (i = 1; i <= nb_desc; i++) {
1073                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1074                 if (!*tmp) {
1075                         TAP_LOG(WARNING,
1076                                 "%s: couldn't allocate memory for queue %d",
1077                                 dev->device->name, rx_queue_id);
1078                         ret = -ENOMEM;
1079                         goto error;
1080                 }
1081                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1082                 (*rxq->iovecs)[i].iov_base =
1083                         (char *)(*tmp)->buf_addr + data_off;
1084                 data_off = 0;
1085                 tmp = &(*tmp)->next;
1086         }
1087
1088         TAP_LOG(DEBUG, "  RX TUNTAP device name %s, qid %d on fd %d",
1089                 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
1090
1091         return 0;
1092
1093 error:
1094         rte_pktmbuf_free(rxq->pool);
1095         rxq->pool = NULL;
1096         rte_free(rxq->iovecs);
1097         rxq->iovecs = NULL;
1098         return ret;
1099 }
1100
1101 static int
1102 tap_tx_queue_setup(struct rte_eth_dev *dev,
1103                    uint16_t tx_queue_id,
1104                    uint16_t nb_tx_desc __rte_unused,
1105                    unsigned int socket_id __rte_unused,
1106                    const struct rte_eth_txconf *tx_conf)
1107 {
1108         struct pmd_internals *internals = dev->data->dev_private;
1109         struct tx_queue *txq;
1110         int ret;
1111         uint64_t offloads;
1112
1113         if (tx_queue_id >= dev->data->nb_tx_queues)
1114                 return -1;
1115         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1116         txq = dev->data->tx_queues[tx_queue_id];
1117
1118         offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1119         txq->csum = !!(offloads &
1120                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
1121                          DEV_TX_OFFLOAD_UDP_CKSUM |
1122                          DEV_TX_OFFLOAD_TCP_CKSUM));
1123
1124         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1125         if (ret == -1)
1126                 return -1;
1127         TAP_LOG(DEBUG,
1128                 "  TX TUNTAP device name %s, qid %d on fd %d csum %s",
1129                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd,
1130                 txq->csum ? "on" : "off");
1131
1132         return 0;
1133 }
1134
1135 static int
1136 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1137 {
1138         struct pmd_internals *pmd = dev->data->dev_private;
1139         struct ifreq ifr = { .ifr_mtu = mtu };
1140         int err = 0;
1141
1142         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1143         if (!err)
1144                 dev->data->mtu = mtu;
1145
1146         return err;
1147 }
1148
1149 static int
1150 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1151                      struct ether_addr *mc_addr_set __rte_unused,
1152                      uint32_t nb_mc_addr __rte_unused)
1153 {
1154         /*
1155          * Nothing to do actually: the tap has no filtering whatsoever, every
1156          * packet is received.
1157          */
1158         return 0;
1159 }
1160
1161 static int
1162 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1163 {
1164         struct rte_eth_dev *dev = arg;
1165         struct pmd_internals *pmd = dev->data->dev_private;
1166         struct ifinfomsg *info = NLMSG_DATA(nh);
1167
1168         if (nh->nlmsg_type != RTM_NEWLINK ||
1169             (info->ifi_index != pmd->if_index &&
1170              info->ifi_index != pmd->remote_if_index))
1171                 return 0;
1172         return tap_link_update(dev, 0);
1173 }
1174
1175 static void
1176 tap_dev_intr_handler(void *cb_arg)
1177 {
1178         struct rte_eth_dev *dev = cb_arg;
1179         struct pmd_internals *pmd = dev->data->dev_private;
1180
1181         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1182 }
1183
1184 static int
1185 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1186 {
1187         struct pmd_internals *pmd = dev->data->dev_private;
1188
1189         /* In any case, disable interrupt if the conf is no longer there. */
1190         if (!dev->data->dev_conf.intr_conf.lsc) {
1191                 if (pmd->intr_handle.fd != -1) {
1192                         tap_nl_final(pmd->intr_handle.fd);
1193                         rte_intr_callback_unregister(&pmd->intr_handle,
1194                                 tap_dev_intr_handler, dev);
1195                 }
1196                 return 0;
1197         }
1198         if (set) {
1199                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1200                 if (unlikely(pmd->intr_handle.fd == -1))
1201                         return -EBADF;
1202                 return rte_intr_callback_register(
1203                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1204         }
1205         tap_nl_final(pmd->intr_handle.fd);
1206         return rte_intr_callback_unregister(&pmd->intr_handle,
1207                                             tap_dev_intr_handler, dev);
1208 }
1209
1210 static int
1211 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1212 {
1213         int err;
1214
1215         err = tap_lsc_intr_handle_set(dev, set);
1216         if (err)
1217                 return err;
1218         err = tap_rx_intr_vec_set(dev, set);
1219         if (err && set)
1220                 tap_lsc_intr_handle_set(dev, 0);
1221         return err;
1222 }
1223
1224 static const uint32_t*
1225 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1226 {
1227         static const uint32_t ptypes[] = {
1228                 RTE_PTYPE_INNER_L2_ETHER,
1229                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1230                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1231                 RTE_PTYPE_INNER_L3_IPV4,
1232                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1233                 RTE_PTYPE_INNER_L3_IPV6,
1234                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1235                 RTE_PTYPE_INNER_L4_FRAG,
1236                 RTE_PTYPE_INNER_L4_UDP,
1237                 RTE_PTYPE_INNER_L4_TCP,
1238                 RTE_PTYPE_INNER_L4_SCTP,
1239                 RTE_PTYPE_L2_ETHER,
1240                 RTE_PTYPE_L2_ETHER_VLAN,
1241                 RTE_PTYPE_L2_ETHER_QINQ,
1242                 RTE_PTYPE_L3_IPV4,
1243                 RTE_PTYPE_L3_IPV4_EXT,
1244                 RTE_PTYPE_L3_IPV6_EXT,
1245                 RTE_PTYPE_L3_IPV6,
1246                 RTE_PTYPE_L4_FRAG,
1247                 RTE_PTYPE_L4_UDP,
1248                 RTE_PTYPE_L4_TCP,
1249                 RTE_PTYPE_L4_SCTP,
1250         };
1251
1252         return ptypes;
1253 }
1254
1255 static int
1256 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1257                   struct rte_eth_fc_conf *fc_conf)
1258 {
1259         fc_conf->mode = RTE_FC_NONE;
1260         return 0;
1261 }
1262
1263 static int
1264 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1265                   struct rte_eth_fc_conf *fc_conf)
1266 {
1267         if (fc_conf->mode != RTE_FC_NONE)
1268                 return -ENOTSUP;
1269         return 0;
1270 }
1271
1272 /**
1273  * DPDK callback to update the RSS hash configuration.
1274  *
1275  * @param dev
1276  *   Pointer to Ethernet device structure.
1277  * @param[in] rss_conf
1278  *   RSS configuration data.
1279  *
1280  * @return
1281  *   0 on success, a negative errno value otherwise and rte_errno is set.
1282  */
1283 static int
1284 tap_rss_hash_update(struct rte_eth_dev *dev,
1285                 struct rte_eth_rss_conf *rss_conf)
1286 {
1287         if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1288                 rte_errno = EINVAL;
1289                 return -rte_errno;
1290         }
1291         if (rss_conf->rss_key && rss_conf->rss_key_len) {
1292                 /*
1293                  * Currently TAP RSS key is hard coded
1294                  * and cannot be updated
1295                  */
1296                 TAP_LOG(ERR,
1297                         "port %u RSS key cannot be updated",
1298                         dev->data->port_id);
1299                 rte_errno = EINVAL;
1300                 return -rte_errno;
1301         }
1302         return 0;
1303 }
1304
1305 static const struct eth_dev_ops ops = {
1306         .dev_start              = tap_dev_start,
1307         .dev_stop               = tap_dev_stop,
1308         .dev_close              = tap_dev_close,
1309         .dev_configure          = tap_dev_configure,
1310         .dev_infos_get          = tap_dev_info,
1311         .rx_queue_setup         = tap_rx_queue_setup,
1312         .tx_queue_setup         = tap_tx_queue_setup,
1313         .rx_queue_release       = tap_rx_queue_release,
1314         .tx_queue_release       = tap_tx_queue_release,
1315         .flow_ctrl_get          = tap_flow_ctrl_get,
1316         .flow_ctrl_set          = tap_flow_ctrl_set,
1317         .link_update            = tap_link_update,
1318         .dev_set_link_up        = tap_link_set_up,
1319         .dev_set_link_down      = tap_link_set_down,
1320         .promiscuous_enable     = tap_promisc_enable,
1321         .promiscuous_disable    = tap_promisc_disable,
1322         .allmulticast_enable    = tap_allmulti_enable,
1323         .allmulticast_disable   = tap_allmulti_disable,
1324         .mac_addr_set           = tap_mac_set,
1325         .mtu_set                = tap_mtu_set,
1326         .set_mc_addr_list       = tap_set_mc_addr_list,
1327         .stats_get              = tap_stats_get,
1328         .stats_reset            = tap_stats_reset,
1329         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1330         .rss_hash_update        = tap_rss_hash_update,
1331         .filter_ctrl            = tap_dev_filter_ctrl,
1332 };
1333
1334 static int
1335 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1336                    char *remote_iface, struct ether_addr *mac_addr)
1337 {
1338         int numa_node = rte_socket_id();
1339         struct rte_eth_dev *dev;
1340         struct pmd_internals *pmd;
1341         struct rte_eth_dev_data *data;
1342         struct ifreq ifr;
1343         int i;
1344
1345         TAP_LOG(DEBUG, "%s device on numa %u",
1346                         tuntap_name, rte_socket_id());
1347
1348         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1349         if (!dev) {
1350                 TAP_LOG(ERR, "%s Unable to allocate device struct",
1351                                 tuntap_name);
1352                 goto error_exit_nodev;
1353         }
1354
1355         pmd = dev->data->dev_private;
1356         pmd->dev = dev;
1357         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1358
1359         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1360         if (pmd->ioctl_sock == -1) {
1361                 TAP_LOG(ERR,
1362                         "%s Unable to get a socket for management: %s",
1363                         tuntap_name, strerror(errno));
1364                 goto error_exit;
1365         }
1366
1367         /* Setup some default values */
1368         data = dev->data;
1369         data->dev_private = pmd;
1370         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1371         data->numa_node = numa_node;
1372
1373         data->dev_link = pmd_link;
1374         data->mac_addrs = &pmd->eth_addr;
1375         /* Set the number of RX and TX queues */
1376         data->nb_rx_queues = 0;
1377         data->nb_tx_queues = 0;
1378
1379         dev->dev_ops = &ops;
1380         dev->rx_pkt_burst = pmd_rx_burst;
1381         dev->tx_pkt_burst = pmd_tx_burst;
1382
1383         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1384         pmd->intr_handle.fd = -1;
1385         dev->intr_handle = &pmd->intr_handle;
1386
1387         /* Presetup the fds to -1 as being not valid */
1388         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1389                 pmd->rxq[i].fd = -1;
1390                 pmd->txq[i].fd = -1;
1391         }
1392
1393         if (tap_type) {
1394                 if (is_zero_ether_addr(mac_addr))
1395                         eth_random_addr((uint8_t *)&pmd->eth_addr);
1396                 else
1397                         rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1398         }
1399
1400         /* Immediately create the netdevice (this will create the 1st queue). */
1401         /* rx queue */
1402         if (tap_setup_queue(dev, pmd, 0, 1) == -1)
1403                 goto error_exit;
1404         /* tx queue */
1405         if (tap_setup_queue(dev, pmd, 0, 0) == -1)
1406                 goto error_exit;
1407
1408         ifr.ifr_mtu = dev->data->mtu;
1409         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1410                 goto error_exit;
1411
1412         if (tap_type) {
1413                 memset(&ifr, 0, sizeof(struct ifreq));
1414                 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1415                 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1416                                 ETHER_ADDR_LEN);
1417                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1418                         goto error_exit;
1419         }
1420
1421         /*
1422          * Set up everything related to rte_flow:
1423          * - netlink socket
1424          * - tap / remote if_index
1425          * - mandatory QDISCs
1426          * - rte_flow actual/implicit lists
1427          * - implicit rules
1428          */
1429         pmd->nlsk_fd = tap_nl_init(0);
1430         if (pmd->nlsk_fd == -1) {
1431                 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
1432                         pmd->name);
1433                 goto disable_rte_flow;
1434         }
1435         pmd->if_index = if_nametoindex(pmd->name);
1436         if (!pmd->if_index) {
1437                 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
1438                 goto disable_rte_flow;
1439         }
1440         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1441                 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
1442                         pmd->name);
1443                 goto disable_rte_flow;
1444         }
1445         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1446                 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1447                         pmd->name);
1448                 goto disable_rte_flow;
1449         }
1450         LIST_INIT(&pmd->flows);
1451
1452         if (strlen(remote_iface)) {
1453                 pmd->remote_if_index = if_nametoindex(remote_iface);
1454                 if (!pmd->remote_if_index) {
1455                         TAP_LOG(ERR, "%s: failed to get %s if_index.",
1456                                 pmd->name, remote_iface);
1457                         goto error_remote;
1458                 }
1459                 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1460                          "%s", remote_iface);
1461
1462                 /* Save state of remote device */
1463                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1464
1465                 /* Replicate remote MAC address */
1466                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1467                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1468                                 pmd->name, pmd->remote_iface);
1469                         goto error_remote;
1470                 }
1471                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1472                            ETHER_ADDR_LEN);
1473                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1474                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1475                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1476                                 pmd->name, remote_iface);
1477                         goto error_remote;
1478                 }
1479
1480                 /*
1481                  * Flush usually returns negative value because it tries to
1482                  * delete every QDISC (and on a running device, one QDISC at
1483                  * least is needed). Ignore negative return value.
1484                  */
1485                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1486                 if (qdisc_create_ingress(pmd->nlsk_fd,
1487                                          pmd->remote_if_index) < 0) {
1488                         TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1489                                 pmd->remote_iface);
1490                         goto error_remote;
1491                 }
1492                 LIST_INIT(&pmd->implicit_flows);
1493                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1494                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1495                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1496                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1497                         TAP_LOG(ERR,
1498                                 "%s: failed to create implicit rules.",
1499                                 pmd->name);
1500                         goto error_remote;
1501                 }
1502         }
1503
1504         rte_eth_dev_probing_finish(dev);
1505         return 0;
1506
1507 disable_rte_flow:
1508         TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
1509                 strerror(errno), errno);
1510         if (strlen(remote_iface)) {
1511                 TAP_LOG(ERR, "Remote feature requires flow support.");
1512                 goto error_exit;
1513         }
1514         return 0;
1515
1516 error_remote:
1517         TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
1518                 strerror(errno), errno);
1519         tap_flow_implicit_flush(pmd, NULL);
1520
1521 error_exit:
1522         if (pmd->ioctl_sock > 0)
1523                 close(pmd->ioctl_sock);
1524         rte_eth_dev_release_port(dev);
1525
1526 error_exit_nodev:
1527         TAP_LOG(ERR, "%s Unable to initialize %s",
1528                 tuntap_name, rte_vdev_device_name(vdev));
1529
1530         return -EINVAL;
1531 }
1532
1533 static int
1534 set_interface_name(const char *key __rte_unused,
1535                    const char *value,
1536                    void *extra_args)
1537 {
1538         char *name = (char *)extra_args;
1539
1540         if (value)
1541                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN - 1);
1542         else
1543                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1544                          DEFAULT_TAP_NAME, (tap_unit - 1));
1545
1546         return 0;
1547 }
1548
1549 static int
1550 set_remote_iface(const char *key __rte_unused,
1551                  const char *value,
1552                  void *extra_args)
1553 {
1554         char *name = (char *)extra_args;
1555
1556         if (value)
1557                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
1558
1559         return 0;
1560 }
1561
1562 static int parse_user_mac(struct ether_addr *user_mac,
1563                 const char *value)
1564 {
1565         unsigned int index = 0;
1566         char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
1567
1568         if (user_mac == NULL || value == NULL)
1569                 return 0;
1570
1571         strlcpy(mac_temp, value, sizeof(mac_temp));
1572         mac_byte = strtok(mac_temp, ":");
1573
1574         while ((mac_byte != NULL) &&
1575                         (strlen(mac_byte) <= 2) &&
1576                         (strlen(mac_byte) == strspn(mac_byte,
1577                                         ETH_TAP_CMP_MAC_FMT))) {
1578                 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
1579                 mac_byte = strtok(NULL, ":");
1580         }
1581
1582         return index;
1583 }
1584
1585 static int
1586 set_mac_type(const char *key __rte_unused,
1587              const char *value,
1588              void *extra_args)
1589 {
1590         struct ether_addr *user_mac = extra_args;
1591
1592         if (!value)
1593                 return 0;
1594
1595         if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
1596                 static int iface_idx;
1597
1598                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1599                 memcpy((char *)user_mac->addr_bytes, "\0dtap", ETHER_ADDR_LEN);
1600                 user_mac->addr_bytes[ETHER_ADDR_LEN - 1] = iface_idx++ + '0';
1601                 goto success;
1602         }
1603
1604         if (parse_user_mac(user_mac, value) != 6)
1605                 goto error;
1606 success:
1607         TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
1608         return 0;
1609
1610 error:
1611         TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
1612                 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
1613         return -1;
1614 }
1615
1616 /*
1617  * Open a TUN interface device. TUN PMD
1618  * 1) sets tap_type as false
1619  * 2) intakes iface as argument.
1620  * 3) as interface is virtual set speed to 10G
1621  */
1622 static int
1623 rte_pmd_tun_probe(struct rte_vdev_device *dev)
1624 {
1625         const char *name, *params;
1626         int ret;
1627         struct rte_kvargs *kvlist = NULL;
1628         char tun_name[RTE_ETH_NAME_MAX_LEN];
1629         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1630
1631         tap_type = 0;
1632         strcpy(tuntap_name, "TUN");
1633
1634         name = rte_vdev_device_name(dev);
1635         params = rte_vdev_device_args(dev);
1636         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1637
1638         snprintf(tun_name, sizeof(tun_name), "%s%u",
1639                  DEFAULT_TUN_NAME, tun_unit++);
1640
1641         if (params && (params[0] != '\0')) {
1642                 TAP_LOG(DEBUG, "parameters (%s)", params);
1643
1644                 kvlist = rte_kvargs_parse(params, valid_arguments);
1645                 if (kvlist) {
1646                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1647                                 ret = rte_kvargs_process(kvlist,
1648                                         ETH_TAP_IFACE_ARG,
1649                                         &set_interface_name,
1650                                         tun_name);
1651
1652                                 if (ret == -1)
1653                                         goto leave;
1654                         }
1655                 }
1656         }
1657         pmd_link.link_speed = ETH_SPEED_NUM_10G;
1658
1659         TAP_LOG(NOTICE, "Initializing pmd_tun for %s as %s",
1660                 name, tun_name);
1661
1662         ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0);
1663
1664 leave:
1665         if (ret == -1) {
1666                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
1667                         name, tun_name);
1668                 tun_unit--; /* Restore the unit number */
1669         }
1670         rte_kvargs_free(kvlist);
1671
1672         return ret;
1673 }
1674
1675 /* Open a TAP interface device.
1676  */
1677 static int
1678 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1679 {
1680         const char *name, *params;
1681         int ret;
1682         struct rte_kvargs *kvlist = NULL;
1683         int speed;
1684         char tap_name[RTE_ETH_NAME_MAX_LEN];
1685         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1686         struct ether_addr user_mac = { .addr_bytes = {0} };
1687         struct rte_eth_dev *eth_dev;
1688
1689         tap_type = 1;
1690         strcpy(tuntap_name, "TAP");
1691
1692         name = rte_vdev_device_name(dev);
1693         params = rte_vdev_device_args(dev);
1694
1695         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
1696             strlen(params) == 0) {
1697                 eth_dev = rte_eth_dev_attach_secondary(name);
1698                 if (!eth_dev) {
1699                         TAP_LOG(ERR, "Failed to probe %s", name);
1700                         return -1;
1701                 }
1702                 /* TODO: request info from primary to set up Rx and Tx */
1703                 eth_dev->dev_ops = &ops;
1704                 rte_eth_dev_probing_finish(eth_dev);
1705                 return 0;
1706         }
1707
1708         speed = ETH_SPEED_NUM_10G;
1709         snprintf(tap_name, sizeof(tap_name), "%s%d",
1710                  DEFAULT_TAP_NAME, tap_unit++);
1711         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1712
1713         if (params && (params[0] != '\0')) {
1714                 TAP_LOG(DEBUG, "parameters (%s)", params);
1715
1716                 kvlist = rte_kvargs_parse(params, valid_arguments);
1717                 if (kvlist) {
1718                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1719                                 ret = rte_kvargs_process(kvlist,
1720                                                          ETH_TAP_IFACE_ARG,
1721                                                          &set_interface_name,
1722                                                          tap_name);
1723                                 if (ret == -1)
1724                                         goto leave;
1725                         }
1726
1727                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1728                                 ret = rte_kvargs_process(kvlist,
1729                                                          ETH_TAP_REMOTE_ARG,
1730                                                          &set_remote_iface,
1731                                                          remote_iface);
1732                                 if (ret == -1)
1733                                         goto leave;
1734                         }
1735
1736                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1737                                 ret = rte_kvargs_process(kvlist,
1738                                                          ETH_TAP_MAC_ARG,
1739                                                          &set_mac_type,
1740                                                          &user_mac);
1741                                 if (ret == -1)
1742                                         goto leave;
1743                         }
1744                 }
1745         }
1746         pmd_link.link_speed = speed;
1747
1748         TAP_LOG(NOTICE, "Initializing pmd_tap for %s as %s",
1749                 name, tap_name);
1750
1751         ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac);
1752
1753 leave:
1754         if (ret == -1) {
1755                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
1756                         name, tap_name);
1757                 tap_unit--;             /* Restore the unit number */
1758         }
1759         rte_kvargs_free(kvlist);
1760
1761         return ret;
1762 }
1763
1764 /* detach a TUNTAP device.
1765  */
1766 static int
1767 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1768 {
1769         struct rte_eth_dev *eth_dev = NULL;
1770         struct pmd_internals *internals;
1771         int i;
1772
1773         TAP_LOG(DEBUG, "Closing TUN/TAP Ethernet device on numa %u",
1774                 rte_socket_id());
1775
1776         /* find the ethdev entry */
1777         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1778         if (!eth_dev)
1779                 return 0;
1780
1781         internals = eth_dev->data->dev_private;
1782         if (internals->nlsk_fd) {
1783                 tap_flow_flush(eth_dev, NULL);
1784                 tap_flow_implicit_flush(internals, NULL);
1785                 tap_nl_final(internals->nlsk_fd);
1786         }
1787         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1788                 if (internals->rxq[i].fd != -1) {
1789                         close(internals->rxq[i].fd);
1790                         internals->rxq[i].fd = -1;
1791                 }
1792                 if (internals->txq[i].fd != -1) {
1793                         close(internals->txq[i].fd);
1794                         internals->txq[i].fd = -1;
1795                 }
1796         }
1797
1798         close(internals->ioctl_sock);
1799         rte_free(eth_dev->data->dev_private);
1800
1801         rte_eth_dev_release_port(eth_dev);
1802
1803         return 0;
1804 }
1805
1806 static struct rte_vdev_driver pmd_tun_drv = {
1807         .probe = rte_pmd_tun_probe,
1808         .remove = rte_pmd_tap_remove,
1809 };
1810
1811 static struct rte_vdev_driver pmd_tap_drv = {
1812         .probe = rte_pmd_tap_probe,
1813         .remove = rte_pmd_tap_remove,
1814 };
1815
1816 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1817 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
1818 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1819 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
1820                               ETH_TAP_IFACE_ARG "=<string> ");
1821 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1822                               ETH_TAP_IFACE_ARG "=<string> "
1823                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
1824                               ETH_TAP_REMOTE_ARG "=<string>");
1825 int tap_logtype;
1826
1827 RTE_INIT(tap_init_log);
1828 static void
1829 tap_init_log(void)
1830 {
1831         tap_logtype = rte_log_register("pmd.net.tap");
1832         if (tap_logtype >= 0)
1833                 rte_log_set_level(tap_logtype, RTE_LOG_NOTICE);
1834 }