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 char tuntap_name[8];
  72
  73 static volatile uint32_t tap_trigger;   /* Rx trigger */
  74
  75 static struct rte_eth_link pmd_link = {
  76         .link_speed = ETH_SPEED_NUM_10G,
  77         .link_duplex = ETH_LINK_FULL_DUPLEX,
  78         .link_status = ETH_LINK_DOWN,
  79         .link_autoneg = ETH_LINK_FIXED,
  80 };
  81
  82 static void
  83 tap_trigger_cb(int sig __rte_unused)
  84 {
  85         /* Valid trigger values are nonzero */
  86         tap_trigger = (tap_trigger + 1) | 0x80000000;
  87 }
  88
  89 /* Specifies on what netdevices the ioctl should be applied */
  90 enum ioctl_mode {
  91         LOCAL_AND_REMOTE,
  92         LOCAL_ONLY,
  93         REMOTE_ONLY,
  94 };
  95
  96 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
  97
  98 /* Tun/Tap allocation routine
  99  *
 100  * name is the number of the interface to use, unless NULL to take the host
 101  * supplied name.
 102  */
 103 static int
 104 tun_alloc(struct pmd_internals *pmd)
 105 {
 106         struct ifreq ifr;
 107 #ifdef IFF_MULTI_QUEUE
 108         unsigned int features;
 109 #endif
 110         int fd;
 111
 112         memset(&ifr, 0, sizeof(struct ifreq));
 113
 114         /*
 115          * Do not set IFF_NO_PI as packet information header will be needed
 116          * to check if a received packet has been truncated.
 117          */
 118         ifr.ifr_flags = (pmd->type == ETH_TUNTAP_TYPE_TAP) ?
 119                 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                 if (txq->type == ETH_TUNTAP_TYPE_TUN) {
 476                         /*
 477                          * TUN and TAP are created with IFF_NO_PI disabled.
 478                          * For TUN PMD this mandatory as fields are used by
 479                          * Kernel tun.c to determine whether its IP or non IP
 480                          * packets.
 481                          *
 482                          * The logic fetches the first byte of data from mbuf
 483                          * then compares whether its v4 or v6. If first byte
 484                          * is 4 or 6, then protocol field is updated.
 485                          */
 486                         char *buff_data = rte_pktmbuf_mtod(seg, void *);
 487                         j = (*buff_data & 0xf0);
 488                         pi.proto = (j == 0x40) ? rte_cpu_to_be_16(ETHER_TYPE_IPv4) :
 489                                 (j == 0x60) ? rte_cpu_to_be_16(ETHER_TYPE_IPv6) : 0x00;
 490                 }
 491
 492                 iovecs[0].iov_base = &pi;
 493                 iovecs[0].iov_len = sizeof(pi);
 494                 for (j = 1; j <= mbuf->nb_segs; j++) {
 495                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 496                         iovecs[j].iov_base =
 497                                 rte_pktmbuf_mtod(seg, void *);
 498                         seg = seg->next;
 499                 }
 500                 if (txq->csum &&
 501                     ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 502                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 503                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
 504                         /* Support only packets with all data in the same seg */
 505                         if (mbuf->nb_segs > 1)
 506                                 break;
 507                         /* To change checksums, work on a copy of data. */
 508                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 509                                    rte_pktmbuf_data_len(mbuf));
 510                         tap_tx_offload(m_copy, mbuf->ol_flags,
 511                                        mbuf->l2_len, mbuf->l3_len);
 512                         iovecs[1].iov_base = m_copy;
 513                 }
 514                 /* copy the tx frame data */
 515                 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
 516                 if (n <= 0)
 517                         break;
 518
 519                 num_tx++;
 520                 num_tx_bytes += mbuf->pkt_len;
 521                 rte_pktmbuf_free(mbuf);
 522         }
 523
 524         txq->stats.opackets += num_tx;
 525         txq->stats.errs += nb_pkts - num_tx;
 526         txq->stats.obytes += num_tx_bytes;
 527
 528         return num_tx;
 529 }
 530
 531 static const char *
 532 tap_ioctl_req2str(unsigned long request)
 533 {
 534         switch (request) {
 535         case SIOCSIFFLAGS:
 536                 return "SIOCSIFFLAGS";
 537         case SIOCGIFFLAGS:
 538                 return "SIOCGIFFLAGS";
 539         case SIOCGIFHWADDR:
 540                 return "SIOCGIFHWADDR";
 541         case SIOCSIFHWADDR:
 542                 return "SIOCSIFHWADDR";
 543         case SIOCSIFMTU:
 544                 return "SIOCSIFMTU";
 545         }
 546         return "UNKNOWN";
 547 }
 548
 549 static int
 550 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 551           struct ifreq *ifr, int set, enum ioctl_mode mode)
 552 {
 553         short req_flags = ifr->ifr_flags;
 554         int remote = pmd->remote_if_index &&
 555                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 556
 557         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 558                 return 0;
 559         /*
 560          * If there is a remote netdevice, apply ioctl on it, then apply it on
 561          * the tap netdevice.
 562          */
 563 apply:
 564         if (remote)
 565                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
 566         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 567                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 568         switch (request) {
 569         case SIOCSIFFLAGS:
 570                 /* fetch current flags to leave other flags untouched */
 571                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 572                         goto error;
 573                 if (set)
 574                         ifr->ifr_flags |= req_flags;
 575                 else
 576                         ifr->ifr_flags &= ~req_flags;
 577                 break;
 578         case SIOCGIFFLAGS:
 579         case SIOCGIFHWADDR:
 580         case SIOCSIFHWADDR:
 581         case SIOCSIFMTU:
 582                 break;
 583         default:
 584                 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
 585                         pmd->name);
 586                 return -EINVAL;
 587         }
 588         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 589                 goto error;
 590         if (remote-- && mode == LOCAL_AND_REMOTE)
 591                 goto apply;
 592         return 0;
 593
 594 error:
 595         TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
 596                 tap_ioctl_req2str(request), strerror(errno), errno);
 597         return -errno;
 598 }
 599
 600 static int
 601 tap_link_set_down(struct rte_eth_dev *dev)
 602 {
 603         struct pmd_internals *pmd = dev->data->dev_private;
 604         struct ifreq ifr = { .ifr_flags = IFF_UP };
 605
 606         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 607         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 608 }
 609
 610 static int
 611 tap_link_set_up(struct rte_eth_dev *dev)
 612 {
 613         struct pmd_internals *pmd = dev->data->dev_private;
 614         struct ifreq ifr = { .ifr_flags = IFF_UP };
 615
 616         dev->data->dev_link.link_status = ETH_LINK_UP;
 617         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 618 }
 619
 620 static int
 621 tap_dev_start(struct rte_eth_dev *dev)
 622 {
 623         int err;
 624
 625         err = tap_intr_handle_set(dev, 1);
 626         if (err)
 627                 return err;
 628         return tap_link_set_up(dev);
 629 }
 630
 631 /* This function gets called when the current port gets stopped.
 632  */
 633 static void
 634 tap_dev_stop(struct rte_eth_dev *dev)
 635 {
 636         tap_intr_handle_set(dev, 0);
 637         tap_link_set_down(dev);
 638 }
 639
 640 static int
 641 tap_dev_configure(struct rte_eth_dev *dev)
 642 {
 643         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 644                 TAP_LOG(ERR,
 645                         "%s: number of rx queues %d exceeds max num of queues %d",
 646                         dev->device->name,
 647                         dev->data->nb_rx_queues,
 648                         RTE_PMD_TAP_MAX_QUEUES);
 649                 return -1;
 650         }
 651         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 652                 TAP_LOG(ERR,
 653                         "%s: number of tx queues %d exceeds max num of queues %d",
 654                         dev->device->name,
 655                         dev->data->nb_tx_queues,
 656                         RTE_PMD_TAP_MAX_QUEUES);
 657                 return -1;
 658         }
 659
 660         TAP_LOG(INFO, "%s: %p: TX configured queues number: %u",
 661                 dev->device->name, (void *)dev, dev->data->nb_tx_queues);
 662
 663         TAP_LOG(INFO, "%s: %p: RX configured queues number: %u",
 664                 dev->device->name, (void *)dev, dev->data->nb_rx_queues);
 665
 666         return 0;
 667 }
 668
 669 static uint32_t
 670 tap_dev_speed_capa(void)
 671 {
 672         uint32_t speed = pmd_link.link_speed;
 673         uint32_t capa = 0;
 674
 675         if (speed >= ETH_SPEED_NUM_10M)
 676                 capa |= ETH_LINK_SPEED_10M;
 677         if (speed >= ETH_SPEED_NUM_100M)
 678                 capa |= ETH_LINK_SPEED_100M;
 679         if (speed >= ETH_SPEED_NUM_1G)
 680                 capa |= ETH_LINK_SPEED_1G;
 681         if (speed >= ETH_SPEED_NUM_5G)
 682                 capa |= ETH_LINK_SPEED_2_5G;
 683         if (speed >= ETH_SPEED_NUM_5G)
 684                 capa |= ETH_LINK_SPEED_5G;
 685         if (speed >= ETH_SPEED_NUM_10G)
 686                 capa |= ETH_LINK_SPEED_10G;
 687         if (speed >= ETH_SPEED_NUM_20G)
 688                 capa |= ETH_LINK_SPEED_20G;
 689         if (speed >= ETH_SPEED_NUM_25G)
 690                 capa |= ETH_LINK_SPEED_25G;
 691         if (speed >= ETH_SPEED_NUM_40G)
 692                 capa |= ETH_LINK_SPEED_40G;
 693         if (speed >= ETH_SPEED_NUM_50G)
 694                 capa |= ETH_LINK_SPEED_50G;
 695         if (speed >= ETH_SPEED_NUM_56G)
 696                 capa |= ETH_LINK_SPEED_56G;
 697         if (speed >= ETH_SPEED_NUM_100G)
 698                 capa |= ETH_LINK_SPEED_100G;
 699
 700         return capa;
 701 }
 702
 703 static void
 704 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 705 {
 706         struct pmd_internals *internals = dev->data->dev_private;
 707
 708         dev_info->if_index = internals->if_index;
 709         dev_info->max_mac_addrs = 1;
 710         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 711         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
 712         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
 713         dev_info->min_rx_bufsize = 0;
 714         dev_info->speed_capa = tap_dev_speed_capa();
 715         dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
 716         dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
 717                                     dev_info->rx_queue_offload_capa;
 718         dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
 719         dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
 720                                     dev_info->tx_queue_offload_capa;
 721         dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
 722         /*
 723          * limitation: TAP supports all of IP, UDP and TCP hash
 724          * functions together and not in partial combinations
 725          */
 726         dev_info->flow_type_rss_offloads = ~TAP_RSS_HF_MASK;
 727 }
 728
 729 static int
 730 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 731 {
 732         unsigned int i, imax;
 733         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 734         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 735         unsigned long rx_nombuf = 0, ierrors = 0;
 736         const struct pmd_internals *pmd = dev->data->dev_private;
 737
 738         /* rx queue statistics */
 739         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 740                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 741         for (i = 0; i < imax; i++) {
 742                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 743                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 744                 rx_total += tap_stats->q_ipackets[i];
 745                 rx_bytes_total += tap_stats->q_ibytes[i];
 746                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
 747                 ierrors += pmd->rxq[i].stats.ierrors;
 748         }
 749
 750         /* tx queue statistics */
 751         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 752                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 753
 754         for (i = 0; i < imax; i++) {
 755                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 756                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 757                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 758                 tx_total += tap_stats->q_opackets[i];
 759                 tx_err_total += tap_stats->q_errors[i];
 760                 tx_bytes_total += tap_stats->q_obytes[i];
 761         }
 762
 763         tap_stats->ipackets = rx_total;
 764         tap_stats->ibytes = rx_bytes_total;
 765         tap_stats->ierrors = ierrors;
 766         tap_stats->rx_nombuf = rx_nombuf;
 767         tap_stats->opackets = tx_total;
 768         tap_stats->oerrors = tx_err_total;
 769         tap_stats->obytes = tx_bytes_total;
 770         return 0;
 771 }
 772
 773 static void
 774 tap_stats_reset(struct rte_eth_dev *dev)
 775 {
 776         int i;
 777         struct pmd_internals *pmd = dev->data->dev_private;
 778
 779         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 780                 pmd->rxq[i].stats.ipackets = 0;
 781                 pmd->rxq[i].stats.ibytes = 0;
 782                 pmd->rxq[i].stats.ierrors = 0;
 783                 pmd->rxq[i].stats.rx_nombuf = 0;
 784
 785                 pmd->txq[i].stats.opackets = 0;
 786                 pmd->txq[i].stats.errs = 0;
 787                 pmd->txq[i].stats.obytes = 0;
 788         }
 789 }
 790
 791 static void
 792 tap_dev_close(struct rte_eth_dev *dev)
 793 {
 794         int i;
 795         struct pmd_internals *internals = dev->data->dev_private;
 796
 797         tap_link_set_down(dev);
 798         tap_flow_flush(dev, NULL);
 799         tap_flow_implicit_flush(internals, NULL);
 800
 801         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 802                 if (internals->rxq[i].fd != -1) {
 803                         close(internals->rxq[i].fd);
 804                         internals->rxq[i].fd = -1;
 805                 }
 806                 if (internals->txq[i].fd != -1) {
 807                         close(internals->txq[i].fd);
 808                         internals->txq[i].fd = -1;
 809                 }
 810         }
 811
 812         if (internals->remote_if_index) {
 813                 /* Restore initial remote state */
 814                 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
 815                                 &internals->remote_initial_flags);
 816         }
 817
 818         if (internals->ka_fd != -1) {
 819                 close(internals->ka_fd);
 820                 internals->ka_fd = -1;
 821         }
 822         /*
 823          * Since TUN device has no more opened file descriptors
 824          * it will be removed from kernel
 825          */
 826 }
 827
 828 static void
 829 tap_rx_queue_release(void *queue)
 830 {
 831         struct rx_queue *rxq = queue;
 832
 833         if (rxq && (rxq->fd > 0)) {
 834                 close(rxq->fd);
 835                 rxq->fd = -1;
 836                 rte_pktmbuf_free(rxq->pool);
 837                 rte_free(rxq->iovecs);
 838                 rxq->pool = NULL;
 839                 rxq->iovecs = NULL;
 840         }
 841 }
 842
 843 static void
 844 tap_tx_queue_release(void *queue)
 845 {
 846         struct tx_queue *txq = queue;
 847
 848         if (txq && (txq->fd > 0)) {
 849                 close(txq->fd);
 850                 txq->fd = -1;
 851         }
 852 }
 853
 854 static int
 855 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
 856 {
 857         struct rte_eth_link *dev_link = &dev->data->dev_link;
 858         struct pmd_internals *pmd = dev->data->dev_private;
 859         struct ifreq ifr = { .ifr_flags = 0 };
 860
 861         if (pmd->remote_if_index) {
 862                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
 863                 if (!(ifr.ifr_flags & IFF_UP) ||
 864                     !(ifr.ifr_flags & IFF_RUNNING)) {
 865                         dev_link->link_status = ETH_LINK_DOWN;
 866                         return 0;
 867                 }
 868         }
 869         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
 870         dev_link->link_status =
 871                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
 872                  ETH_LINK_UP :
 873                  ETH_LINK_DOWN);
 874         return 0;
 875 }
 876
 877 static void
 878 tap_promisc_enable(struct rte_eth_dev *dev)
 879 {
 880         struct pmd_internals *pmd = dev->data->dev_private;
 881         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 882
 883         dev->data->promiscuous = 1;
 884         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 885         if (pmd->remote_if_index && !pmd->flow_isolate)
 886                 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
 887 }
 888
 889 static void
 890 tap_promisc_disable(struct rte_eth_dev *dev)
 891 {
 892         struct pmd_internals *pmd = dev->data->dev_private;
 893         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 894
 895         dev->data->promiscuous = 0;
 896         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 897         if (pmd->remote_if_index && !pmd->flow_isolate)
 898                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
 899 }
 900
 901 static void
 902 tap_allmulti_enable(struct rte_eth_dev *dev)
 903 {
 904         struct pmd_internals *pmd = dev->data->dev_private;
 905         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 906
 907         dev->data->all_multicast = 1;
 908         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 909         if (pmd->remote_if_index && !pmd->flow_isolate)
 910                 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
 911 }
 912
 913 static void
 914 tap_allmulti_disable(struct rte_eth_dev *dev)
 915 {
 916         struct pmd_internals *pmd = dev->data->dev_private;
 917         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 918
 919         dev->data->all_multicast = 0;
 920         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 921         if (pmd->remote_if_index && !pmd->flow_isolate)
 922                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
 923 }
 924
 925 static int
 926 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 927 {
 928         struct pmd_internals *pmd = dev->data->dev_private;
 929         enum ioctl_mode mode = LOCAL_ONLY;
 930         struct ifreq ifr;
 931         int ret;
 932
 933         if (pmd->type == ETH_TUNTAP_TYPE_TUN) {
 934                 TAP_LOG(ERR, "%s: can't MAC address for TUN",
 935                         dev->device->name);
 936                 return -ENOTSUP;
 937         }
 938
 939         if (is_zero_ether_addr(mac_addr)) {
 940                 TAP_LOG(ERR, "%s: can't set an empty MAC address",
 941                         dev->device->name);
 942                 return -EINVAL;
 943         }
 944         /* Check the actual current MAC address on the tap netdevice */
 945         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
 946         if (ret < 0)
 947                 return ret;
 948         if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 949                                mac_addr))
 950                 return 0;
 951         /* Check the current MAC address on the remote */
 952         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
 953         if (ret < 0)
 954                 return ret;
 955         if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 956                                mac_addr))
 957                 mode = LOCAL_AND_REMOTE;
 958         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
 959         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
 960         ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
 961         if (ret < 0)
 962                 return ret;
 963         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
 964         if (pmd->remote_if_index && !pmd->flow_isolate) {
 965                 /* Replace MAC redirection rule after a MAC change */
 966                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
 967                 if (ret < 0) {
 968                         TAP_LOG(ERR,
 969                                 "%s: Couldn't delete MAC redirection rule",
 970                                 dev->device->name);
 971                         return ret;
 972                 }
 973                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
 974                 if (ret < 0) {
 975                         TAP_LOG(ERR,
 976                                 "%s: Couldn't add MAC redirection rule",
 977                                 dev->device->name);
 978                         return ret;
 979                 }
 980         }
 981
 982         return 0;
 983 }
 984
 985 static int
 986 tap_setup_queue(struct rte_eth_dev *dev,
 987                 struct pmd_internals *internals,
 988                 uint16_t qid,
 989                 int is_rx)
 990 {
 991         int *fd;
 992         int *other_fd;
 993         const char *dir;
 994         struct pmd_internals *pmd = dev->data->dev_private;
 995         struct rx_queue *rx = &internals->rxq[qid];
 996         struct tx_queue *tx = &internals->txq[qid];
 997
 998         if (is_rx) {
 999                 fd = &rx->fd;
1000                 other_fd = &tx->fd;
1001                 dir = "rx";
1002         } else {
1003                 fd = &tx->fd;
1004                 other_fd = &rx->fd;
1005                 dir = "tx";
1006         }
1007         if (*fd != -1) {
1008                 /* fd for this queue already exists */
1009                 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1010                         pmd->name, *fd, dir, qid);
1011         } else if (*other_fd != -1) {
1012                 /* Only other_fd exists. dup it */
1013                 *fd = dup(*other_fd);
1014                 if (*fd < 0) {
1015                         *fd = -1;
1016                         TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1017                         return -1;
1018                 }
1019                 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1020                         pmd->name, *other_fd, dir, qid, *fd);
1021         } else {
1022                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1023                 *fd = tun_alloc(pmd);
1024                 if (*fd < 0) {
1025                         *fd = -1; /* restore original value */
1026                         TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1027                         return -1;
1028                 }
1029                 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1030                         pmd->name, dir, qid, *fd);
1031         }
1032
1033         tx->mtu = &dev->data->mtu;
1034         rx->rxmode = &dev->data->dev_conf.rxmode;
1035
1036         tx->type = pmd->type;
1037
1038         return *fd;
1039 }
1040
1041 static int
1042 tap_rx_queue_setup(struct rte_eth_dev *dev,
1043                    uint16_t rx_queue_id,
1044                    uint16_t nb_rx_desc,
1045                    unsigned int socket_id,
1046                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1047                    struct rte_mempool *mp)
1048 {
1049         struct pmd_internals *internals = dev->data->dev_private;
1050         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1051         struct rte_mbuf **tmp = &rxq->pool;
1052         long iov_max = sysconf(_SC_IOV_MAX);
1053         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1054         struct iovec (*iovecs)[nb_desc + 1];
1055         int data_off = RTE_PKTMBUF_HEADROOM;
1056         int ret = 0;
1057         int fd;
1058         int i;
1059
1060         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1061                 TAP_LOG(WARNING,
1062                         "nb_rx_queues %d too small or mempool NULL",
1063                         dev->data->nb_rx_queues);
1064                 return -1;
1065         }
1066
1067         rxq->mp = mp;
1068         rxq->trigger_seen = 1; /* force initial burst */
1069         rxq->in_port = dev->data->port_id;
1070         rxq->nb_rx_desc = nb_desc;
1071         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1072                                     socket_id);
1073         if (!iovecs) {
1074                 TAP_LOG(WARNING,
1075                         "%s: Couldn't allocate %d RX descriptors",
1076                         dev->device->name, nb_desc);
1077                 return -ENOMEM;
1078         }
1079         rxq->iovecs = iovecs;
1080
1081         dev->data->rx_queues[rx_queue_id] = rxq;
1082         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1083         if (fd == -1) {
1084                 ret = fd;
1085                 goto error;
1086         }
1087
1088         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1089         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1090
1091         for (i = 1; i <= nb_desc; i++) {
1092                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1093                 if (!*tmp) {
1094                         TAP_LOG(WARNING,
1095                                 "%s: couldn't allocate memory for queue %d",
1096                                 dev->device->name, rx_queue_id);
1097                         ret = -ENOMEM;
1098                         goto error;
1099                 }
1100                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1101                 (*rxq->iovecs)[i].iov_base =
1102                         (char *)(*tmp)->buf_addr + data_off;
1103                 data_off = 0;
1104                 tmp = &(*tmp)->next;
1105         }
1106
1107         TAP_LOG(DEBUG, "  RX TUNTAP device name %s, qid %d on fd %d",
1108                 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
1109
1110         return 0;
1111
1112 error:
1113         rte_pktmbuf_free(rxq->pool);
1114         rxq->pool = NULL;
1115         rte_free(rxq->iovecs);
1116         rxq->iovecs = NULL;
1117         return ret;
1118 }
1119
1120 static int
1121 tap_tx_queue_setup(struct rte_eth_dev *dev,
1122                    uint16_t tx_queue_id,
1123                    uint16_t nb_tx_desc __rte_unused,
1124                    unsigned int socket_id __rte_unused,
1125                    const struct rte_eth_txconf *tx_conf)
1126 {
1127         struct pmd_internals *internals = dev->data->dev_private;
1128         struct tx_queue *txq;
1129         int ret;
1130         uint64_t offloads;
1131
1132         if (tx_queue_id >= dev->data->nb_tx_queues)
1133                 return -1;
1134         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1135         txq = dev->data->tx_queues[tx_queue_id];
1136
1137         offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1138         txq->csum = !!(offloads &
1139                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
1140                          DEV_TX_OFFLOAD_UDP_CKSUM |
1141                          DEV_TX_OFFLOAD_TCP_CKSUM));
1142
1143         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1144         if (ret == -1)
1145                 return -1;
1146         TAP_LOG(DEBUG,
1147                 "  TX TUNTAP device name %s, qid %d on fd %d csum %s",
1148                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd,
1149                 txq->csum ? "on" : "off");
1150
1151         return 0;
1152 }
1153
1154 static int
1155 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1156 {
1157         struct pmd_internals *pmd = dev->data->dev_private;
1158         struct ifreq ifr = { .ifr_mtu = mtu };
1159         int err = 0;
1160
1161         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1162         if (!err)
1163                 dev->data->mtu = mtu;
1164
1165         return err;
1166 }
1167
1168 static int
1169 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1170                      struct ether_addr *mc_addr_set __rte_unused,
1171                      uint32_t nb_mc_addr __rte_unused)
1172 {
1173         /*
1174          * Nothing to do actually: the tap has no filtering whatsoever, every
1175          * packet is received.
1176          */
1177         return 0;
1178 }
1179
1180 static int
1181 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1182 {
1183         struct rte_eth_dev *dev = arg;
1184         struct pmd_internals *pmd = dev->data->dev_private;
1185         struct ifinfomsg *info = NLMSG_DATA(nh);
1186
1187         if (nh->nlmsg_type != RTM_NEWLINK ||
1188             (info->ifi_index != pmd->if_index &&
1189              info->ifi_index != pmd->remote_if_index))
1190                 return 0;
1191         return tap_link_update(dev, 0);
1192 }
1193
1194 static void
1195 tap_dev_intr_handler(void *cb_arg)
1196 {
1197         struct rte_eth_dev *dev = cb_arg;
1198         struct pmd_internals *pmd = dev->data->dev_private;
1199
1200         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1201 }
1202
1203 static int
1204 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1205 {
1206         struct pmd_internals *pmd = dev->data->dev_private;
1207
1208         /* In any case, disable interrupt if the conf is no longer there. */
1209         if (!dev->data->dev_conf.intr_conf.lsc) {
1210                 if (pmd->intr_handle.fd != -1) {
1211                         tap_nl_final(pmd->intr_handle.fd);
1212                         rte_intr_callback_unregister(&pmd->intr_handle,
1213                                 tap_dev_intr_handler, dev);
1214                 }
1215                 return 0;
1216         }
1217         if (set) {
1218                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1219                 if (unlikely(pmd->intr_handle.fd == -1))
1220                         return -EBADF;
1221                 return rte_intr_callback_register(
1222                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1223         }
1224         tap_nl_final(pmd->intr_handle.fd);
1225         return rte_intr_callback_unregister(&pmd->intr_handle,
1226                                             tap_dev_intr_handler, dev);
1227 }
1228
1229 static int
1230 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1231 {
1232         int err;
1233
1234         err = tap_lsc_intr_handle_set(dev, set);
1235         if (err)
1236                 return err;
1237         err = tap_rx_intr_vec_set(dev, set);
1238         if (err && set)
1239                 tap_lsc_intr_handle_set(dev, 0);
1240         return err;
1241 }
1242
1243 static const uint32_t*
1244 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1245 {
1246         static const uint32_t ptypes[] = {
1247                 RTE_PTYPE_INNER_L2_ETHER,
1248                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1249                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1250                 RTE_PTYPE_INNER_L3_IPV4,
1251                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1252                 RTE_PTYPE_INNER_L3_IPV6,
1253                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1254                 RTE_PTYPE_INNER_L4_FRAG,
1255                 RTE_PTYPE_INNER_L4_UDP,
1256                 RTE_PTYPE_INNER_L4_TCP,
1257                 RTE_PTYPE_INNER_L4_SCTP,
1258                 RTE_PTYPE_L2_ETHER,
1259                 RTE_PTYPE_L2_ETHER_VLAN,
1260                 RTE_PTYPE_L2_ETHER_QINQ,
1261                 RTE_PTYPE_L3_IPV4,
1262                 RTE_PTYPE_L3_IPV4_EXT,
1263                 RTE_PTYPE_L3_IPV6_EXT,
1264                 RTE_PTYPE_L3_IPV6,
1265                 RTE_PTYPE_L4_FRAG,
1266                 RTE_PTYPE_L4_UDP,
1267                 RTE_PTYPE_L4_TCP,
1268                 RTE_PTYPE_L4_SCTP,
1269         };
1270
1271         return ptypes;
1272 }
1273
1274 static int
1275 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1276                   struct rte_eth_fc_conf *fc_conf)
1277 {
1278         fc_conf->mode = RTE_FC_NONE;
1279         return 0;
1280 }
1281
1282 static int
1283 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1284                   struct rte_eth_fc_conf *fc_conf)
1285 {
1286         if (fc_conf->mode != RTE_FC_NONE)
1287                 return -ENOTSUP;
1288         return 0;
1289 }
1290
1291 /**
1292  * DPDK callback to update the RSS hash configuration.
1293  *
1294  * @param dev
1295  *   Pointer to Ethernet device structure.
1296  * @param[in] rss_conf
1297  *   RSS configuration data.
1298  *
1299  * @return
1300  *   0 on success, a negative errno value otherwise and rte_errno is set.
1301  */
1302 static int
1303 tap_rss_hash_update(struct rte_eth_dev *dev,
1304                 struct rte_eth_rss_conf *rss_conf)
1305 {
1306         if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1307                 rte_errno = EINVAL;
1308                 return -rte_errno;
1309         }
1310         if (rss_conf->rss_key && rss_conf->rss_key_len) {
1311                 /*
1312                  * Currently TAP RSS key is hard coded
1313                  * and cannot be updated
1314                  */
1315                 TAP_LOG(ERR,
1316                         "port %u RSS key cannot be updated",
1317                         dev->data->port_id);
1318                 rte_errno = EINVAL;
1319                 return -rte_errno;
1320         }
1321         return 0;
1322 }
1323
1324 static const struct eth_dev_ops ops = {
1325         .dev_start              = tap_dev_start,
1326         .dev_stop               = tap_dev_stop,
1327         .dev_close              = tap_dev_close,
1328         .dev_configure          = tap_dev_configure,
1329         .dev_infos_get          = tap_dev_info,
1330         .rx_queue_setup         = tap_rx_queue_setup,
1331         .tx_queue_setup         = tap_tx_queue_setup,
1332         .rx_queue_release       = tap_rx_queue_release,
1333         .tx_queue_release       = tap_tx_queue_release,
1334         .flow_ctrl_get          = tap_flow_ctrl_get,
1335         .flow_ctrl_set          = tap_flow_ctrl_set,
1336         .link_update            = tap_link_update,
1337         .dev_set_link_up        = tap_link_set_up,
1338         .dev_set_link_down      = tap_link_set_down,
1339         .promiscuous_enable     = tap_promisc_enable,
1340         .promiscuous_disable    = tap_promisc_disable,
1341         .allmulticast_enable    = tap_allmulti_enable,
1342         .allmulticast_disable   = tap_allmulti_disable,
1343         .mac_addr_set           = tap_mac_set,
1344         .mtu_set                = tap_mtu_set,
1345         .set_mc_addr_list       = tap_set_mc_addr_list,
1346         .stats_get              = tap_stats_get,
1347         .stats_reset            = tap_stats_reset,
1348         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1349         .rss_hash_update        = tap_rss_hash_update,
1350         .filter_ctrl            = tap_dev_filter_ctrl,
1351 };
1352
1353 static int
1354 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1355                    char *remote_iface, struct ether_addr *mac_addr,
1356                    enum rte_tuntap_type type)
1357 {
1358         int numa_node = rte_socket_id();
1359         struct rte_eth_dev *dev;
1360         struct pmd_internals *pmd;
1361         struct rte_eth_dev_data *data;
1362         struct ifreq ifr;
1363         int i;
1364
1365         TAP_LOG(DEBUG, "%s device on numa %u",
1366                         tuntap_name, rte_socket_id());
1367
1368         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1369         if (!dev) {
1370                 TAP_LOG(ERR, "%s Unable to allocate device struct",
1371                                 tuntap_name);
1372                 goto error_exit_nodev;
1373         }
1374
1375         pmd = dev->data->dev_private;
1376         pmd->dev = dev;
1377         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1378         pmd->type = type;
1379
1380         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1381         if (pmd->ioctl_sock == -1) {
1382                 TAP_LOG(ERR,
1383                         "%s Unable to get a socket for management: %s",
1384                         tuntap_name, strerror(errno));
1385                 goto error_exit;
1386         }
1387
1388         /* Setup some default values */
1389         data = dev->data;
1390         data->dev_private = pmd;
1391         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1392         data->numa_node = numa_node;
1393
1394         data->dev_link = pmd_link;
1395         data->mac_addrs = &pmd->eth_addr;
1396         /* Set the number of RX and TX queues */
1397         data->nb_rx_queues = 0;
1398         data->nb_tx_queues = 0;
1399
1400         dev->dev_ops = &ops;
1401         dev->rx_pkt_burst = pmd_rx_burst;
1402         dev->tx_pkt_burst = pmd_tx_burst;
1403
1404         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1405         pmd->intr_handle.fd = -1;
1406         dev->intr_handle = &pmd->intr_handle;
1407
1408         /* Presetup the fds to -1 as being not valid */
1409         pmd->ka_fd = -1;
1410         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1411                 pmd->rxq[i].fd = -1;
1412                 pmd->txq[i].fd = -1;
1413         }
1414
1415         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1416                 if (is_zero_ether_addr(mac_addr))
1417                         eth_random_addr((uint8_t *)&pmd->eth_addr);
1418                 else
1419                         rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1420         }
1421
1422         /*
1423          * Allocate a TUN device keep-alive file descriptor that will only be
1424          * closed when the TUN device itself is closed or removed.
1425          * This keep-alive file descriptor will guarantee that the TUN device
1426          * exists even when all of its queues are closed
1427          */
1428         pmd->ka_fd = tun_alloc(pmd);
1429         if (pmd->ka_fd == -1) {
1430                 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
1431                 goto error_exit;
1432         }
1433
1434         ifr.ifr_mtu = dev->data->mtu;
1435         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1436                 goto error_exit;
1437
1438         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1439                 memset(&ifr, 0, sizeof(struct ifreq));
1440                 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1441                 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1442                                 ETHER_ADDR_LEN);
1443                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1444                         goto error_exit;
1445         }
1446
1447         /*
1448          * Set up everything related to rte_flow:
1449          * - netlink socket
1450          * - tap / remote if_index
1451          * - mandatory QDISCs
1452          * - rte_flow actual/implicit lists
1453          * - implicit rules
1454          */
1455         pmd->nlsk_fd = tap_nl_init(0);
1456         if (pmd->nlsk_fd == -1) {
1457                 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
1458                         pmd->name);
1459                 goto disable_rte_flow;
1460         }
1461         pmd->if_index = if_nametoindex(pmd->name);
1462         if (!pmd->if_index) {
1463                 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
1464                 goto disable_rte_flow;
1465         }
1466         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1467                 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
1468                         pmd->name);
1469                 goto disable_rte_flow;
1470         }
1471         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1472                 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1473                         pmd->name);
1474                 goto disable_rte_flow;
1475         }
1476         LIST_INIT(&pmd->flows);
1477
1478         if (strlen(remote_iface)) {
1479                 pmd->remote_if_index = if_nametoindex(remote_iface);
1480                 if (!pmd->remote_if_index) {
1481                         TAP_LOG(ERR, "%s: failed to get %s if_index.",
1482                                 pmd->name, remote_iface);
1483                         goto error_remote;
1484                 }
1485                 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1486                          "%s", remote_iface);
1487
1488                 /* Save state of remote device */
1489                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1490
1491                 /* Replicate remote MAC address */
1492                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1493                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1494                                 pmd->name, pmd->remote_iface);
1495                         goto error_remote;
1496                 }
1497                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1498                            ETHER_ADDR_LEN);
1499                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1500                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1501                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1502                                 pmd->name, remote_iface);
1503                         goto error_remote;
1504                 }
1505
1506                 /*
1507                  * Flush usually returns negative value because it tries to
1508                  * delete every QDISC (and on a running device, one QDISC at
1509                  * least is needed). Ignore negative return value.
1510                  */
1511                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1512                 if (qdisc_create_ingress(pmd->nlsk_fd,
1513                                          pmd->remote_if_index) < 0) {
1514                         TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1515                                 pmd->remote_iface);
1516                         goto error_remote;
1517                 }
1518                 LIST_INIT(&pmd->implicit_flows);
1519                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1520                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1521                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1522                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1523                         TAP_LOG(ERR,
1524                                 "%s: failed to create implicit rules.",
1525                                 pmd->name);
1526                         goto error_remote;
1527                 }
1528         }
1529
1530         rte_eth_dev_probing_finish(dev);
1531         return 0;
1532
1533 disable_rte_flow:
1534         TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
1535                 strerror(errno), errno);
1536         if (strlen(remote_iface)) {
1537                 TAP_LOG(ERR, "Remote feature requires flow support.");
1538                 goto error_exit;
1539         }
1540         return 0;
1541
1542 error_remote:
1543         TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
1544                 strerror(errno), errno);
1545         tap_flow_implicit_flush(pmd, NULL);
1546
1547 error_exit:
1548         if (pmd->ioctl_sock > 0)
1549                 close(pmd->ioctl_sock);
1550         rte_eth_dev_release_port(dev);
1551
1552 error_exit_nodev:
1553         TAP_LOG(ERR, "%s Unable to initialize %s",
1554                 tuntap_name, rte_vdev_device_name(vdev));
1555
1556         return -EINVAL;
1557 }
1558
1559 static int
1560 set_interface_name(const char *key __rte_unused,
1561                    const char *value,
1562                    void *extra_args)
1563 {
1564         char *name = (char *)extra_args;
1565
1566         if (value)
1567                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN - 1);
1568         else
1569                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1570                          DEFAULT_TAP_NAME, (tap_unit - 1));
1571
1572         return 0;
1573 }
1574
1575 static int
1576 set_remote_iface(const char *key __rte_unused,
1577                  const char *value,
1578                  void *extra_args)
1579 {
1580         char *name = (char *)extra_args;
1581
1582         if (value)
1583                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
1584
1585         return 0;
1586 }
1587
1588 static int parse_user_mac(struct ether_addr *user_mac,
1589                 const char *value)
1590 {
1591         unsigned int index = 0;
1592         char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
1593
1594         if (user_mac == NULL || value == NULL)
1595                 return 0;
1596
1597         strlcpy(mac_temp, value, sizeof(mac_temp));
1598         mac_byte = strtok(mac_temp, ":");
1599
1600         while ((mac_byte != NULL) &&
1601                         (strlen(mac_byte) <= 2) &&
1602                         (strlen(mac_byte) == strspn(mac_byte,
1603                                         ETH_TAP_CMP_MAC_FMT))) {
1604                 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
1605                 mac_byte = strtok(NULL, ":");
1606         }
1607
1608         return index;
1609 }
1610
1611 static int
1612 set_mac_type(const char *key __rte_unused,
1613              const char *value,
1614              void *extra_args)
1615 {
1616         struct ether_addr *user_mac = extra_args;
1617
1618         if (!value)
1619                 return 0;
1620
1621         if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
1622                 static int iface_idx;
1623
1624                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1625                 memcpy((char *)user_mac->addr_bytes, "\0dtap", ETHER_ADDR_LEN);
1626                 user_mac->addr_bytes[ETHER_ADDR_LEN - 1] = iface_idx++ + '0';
1627                 goto success;
1628         }
1629
1630         if (parse_user_mac(user_mac, value) != 6)
1631                 goto error;
1632 success:
1633         TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
1634         return 0;
1635
1636 error:
1637         TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
1638                 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
1639         return -1;
1640 }
1641
1642 /*
1643  * Open a TUN interface device. TUN PMD
1644  * 1) sets tap_type as false
1645  * 2) intakes iface as argument.
1646  * 3) as interface is virtual set speed to 10G
1647  */
1648 static int
1649 rte_pmd_tun_probe(struct rte_vdev_device *dev)
1650 {
1651         const char *name, *params;
1652         int ret;
1653         struct rte_kvargs *kvlist = NULL;
1654         char tun_name[RTE_ETH_NAME_MAX_LEN];
1655         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1656         struct rte_eth_dev *eth_dev;
1657
1658         strcpy(tuntap_name, "TUN");
1659
1660         name = rte_vdev_device_name(dev);
1661         params = rte_vdev_device_args(dev);
1662         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1663
1664         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
1665             strlen(params) == 0) {
1666                 eth_dev = rte_eth_dev_attach_secondary(name);
1667                 if (!eth_dev) {
1668                         TAP_LOG(ERR, "Failed to probe %s", name);
1669                         return -1;
1670                 }
1671                 eth_dev->dev_ops = &ops;
1672                 return 0;
1673         }
1674
1675         snprintf(tun_name, sizeof(tun_name), "%s%u",
1676                  DEFAULT_TUN_NAME, tun_unit++);
1677
1678         if (params && (params[0] != '\0')) {
1679                 TAP_LOG(DEBUG, "parameters (%s)", params);
1680
1681                 kvlist = rte_kvargs_parse(params, valid_arguments);
1682                 if (kvlist) {
1683                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1684                                 ret = rte_kvargs_process(kvlist,
1685                                         ETH_TAP_IFACE_ARG,
1686                                         &set_interface_name,
1687                                         tun_name);
1688
1689                                 if (ret == -1)
1690                                         goto leave;
1691                         }
1692                 }
1693         }
1694         pmd_link.link_speed = ETH_SPEED_NUM_10G;
1695
1696         TAP_LOG(NOTICE, "Initializing pmd_tun for %s as %s",
1697                 name, tun_name);
1698
1699         ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0,
1700                 ETH_TUNTAP_TYPE_TUN);
1701
1702 leave:
1703         if (ret == -1) {
1704                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
1705                         name, tun_name);
1706                 tun_unit--; /* Restore the unit number */
1707         }
1708         rte_kvargs_free(kvlist);
1709
1710         return ret;
1711 }
1712
1713 /* Open a TAP interface device.
1714  */
1715 static int
1716 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1717 {
1718         const char *name, *params;
1719         int ret;
1720         struct rte_kvargs *kvlist = NULL;
1721         int speed;
1722         char tap_name[RTE_ETH_NAME_MAX_LEN];
1723         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1724         struct ether_addr user_mac = { .addr_bytes = {0} };
1725         struct rte_eth_dev *eth_dev;
1726
1727         strcpy(tuntap_name, "TAP");
1728
1729         name = rte_vdev_device_name(dev);
1730         params = rte_vdev_device_args(dev);
1731
1732         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
1733             strlen(params) == 0) {
1734                 eth_dev = rte_eth_dev_attach_secondary(name);
1735                 if (!eth_dev) {
1736                         TAP_LOG(ERR, "Failed to probe %s", name);
1737                         return -1;
1738                 }
1739                 /* TODO: request info from primary to set up Rx and Tx */
1740                 eth_dev->dev_ops = &ops;
1741                 rte_eth_dev_probing_finish(eth_dev);
1742                 return 0;
1743         }
1744
1745         speed = ETH_SPEED_NUM_10G;
1746         snprintf(tap_name, sizeof(tap_name), "%s%d",
1747                  DEFAULT_TAP_NAME, tap_unit++);
1748         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1749
1750         if (params && (params[0] != '\0')) {
1751                 TAP_LOG(DEBUG, "parameters (%s)", params);
1752
1753                 kvlist = rte_kvargs_parse(params, valid_arguments);
1754                 if (kvlist) {
1755                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1756                                 ret = rte_kvargs_process(kvlist,
1757                                                          ETH_TAP_IFACE_ARG,
1758                                                          &set_interface_name,
1759                                                          tap_name);
1760                                 if (ret == -1)
1761                                         goto leave;
1762                         }
1763
1764                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1765                                 ret = rte_kvargs_process(kvlist,
1766                                                          ETH_TAP_REMOTE_ARG,
1767                                                          &set_remote_iface,
1768                                                          remote_iface);
1769                                 if (ret == -1)
1770                                         goto leave;
1771                         }
1772
1773                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1774                                 ret = rte_kvargs_process(kvlist,
1775                                                          ETH_TAP_MAC_ARG,
1776                                                          &set_mac_type,
1777                                                          &user_mac);
1778                                 if (ret == -1)
1779                                         goto leave;
1780                         }
1781                 }
1782         }
1783         pmd_link.link_speed = speed;
1784
1785         TAP_LOG(NOTICE, "Initializing pmd_tap for %s as %s",
1786                 name, tap_name);
1787
1788         ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
1789                 ETH_TUNTAP_TYPE_TAP);
1790
1791 leave:
1792         if (ret == -1) {
1793                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
1794                         name, tap_name);
1795                 tap_unit--;             /* Restore the unit number */
1796         }
1797         rte_kvargs_free(kvlist);
1798
1799         return ret;
1800 }
1801
1802 /* detach a TUNTAP device.
1803  */
1804 static int
1805 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1806 {
1807         struct rte_eth_dev *eth_dev = NULL;
1808         struct pmd_internals *internals;
1809         int i;
1810
1811         /* find the ethdev entry */
1812         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1813         if (!eth_dev)
1814                 return 0;
1815
1816         internals = eth_dev->data->dev_private;
1817
1818         TAP_LOG(DEBUG, "Closing %s Ethernet device on numa %u",
1819                 (internals->type == ETH_TUNTAP_TYPE_TAP) ? "TAP" : "TUN",
1820                 rte_socket_id());
1821
1822         if (internals->nlsk_fd) {
1823                 tap_flow_flush(eth_dev, NULL);
1824                 tap_flow_implicit_flush(internals, NULL);
1825                 tap_nl_final(internals->nlsk_fd);
1826         }
1827         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1828                 if (internals->rxq[i].fd != -1) {
1829                         close(internals->rxq[i].fd);
1830                         internals->rxq[i].fd = -1;
1831                 }
1832                 if (internals->txq[i].fd != -1) {
1833                         close(internals->txq[i].fd);
1834                         internals->txq[i].fd = -1;
1835                 }
1836         }
1837
1838         close(internals->ioctl_sock);
1839         rte_free(eth_dev->data->dev_private);
1840         rte_eth_dev_release_port(eth_dev);
1841
1842         if (internals->ka_fd != -1) {
1843                 close(internals->ka_fd);
1844                 internals->ka_fd = -1;
1845         }
1846         return 0;
1847 }
1848
1849 static struct rte_vdev_driver pmd_tun_drv = {
1850         .probe = rte_pmd_tun_probe,
1851         .remove = rte_pmd_tap_remove,
1852 };
1853
1854 static struct rte_vdev_driver pmd_tap_drv = {
1855         .probe = rte_pmd_tap_probe,
1856         .remove = rte_pmd_tap_remove,
1857 };
1858
1859 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1860 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
1861 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1862 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
1863                               ETH_TAP_IFACE_ARG "=<string> ");
1864 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1865                               ETH_TAP_IFACE_ARG "=<string> "
1866                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
1867                               ETH_TAP_REMOTE_ARG "=<string>");
1868 int tap_logtype;
1869
1870 RTE_INIT(tap_init_log);
1871 static void
1872 tap_init_log(void)
1873 {
1874         tap_logtype = rte_log_register("pmd.net.tap");
1875         if (tap_logtype >= 0)
1876                 rte_log_set_level(tap_logtype, RTE_LOG_NOTICE);
1877 }