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
  19 #include <sys/types.h>
  20 #include <sys/stat.h>
  21 #include <sys/socket.h>
  22 #include <sys/ioctl.h>
  23 #include <sys/utsname.h>
  24 #include <sys/mman.h>
  25 #include <errno.h>
  26 #include <signal.h>
  27 #include <stdbool.h>
  28 #include <stdint.h>
  29 #include <sys/uio.h>
  30 #include <unistd.h>
  31 #include <arpa/inet.h>
  32 #include <net/if.h>
  33 #include <linux/if_tun.h>
  34 #include <linux/if_ether.h>
  35 #include <fcntl.h>
  36
  37 #include <rte_eth_tap.h>
  38 #include <tap_flow.h>
  39 #include <tap_netlink.h>
  40 #include <tap_tcmsgs.h>
  41
  42 /* Linux based path to the TUN device */
  43 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  44 #define DEFAULT_TAP_NAME        "dtap"
  45
  46 #define ETH_TAP_IFACE_ARG       "iface"
  47 #define ETH_TAP_REMOTE_ARG      "remote"
  48 #define ETH_TAP_MAC_ARG         "mac"
  49 #define ETH_TAP_MAC_FIXED       "fixed"
  50
  51 #define ETH_TAP_USR_MAC_FMT     "xx:xx:xx:xx:xx:xx"
  52 #define ETH_TAP_CMP_MAC_FMT     "0123456789ABCDEFabcdef"
  53 #define ETH_TAP_MAC_ARG_FMT     ETH_TAP_MAC_FIXED "|" ETH_TAP_USR_MAC_FMT
  54
  55 static struct rte_vdev_driver pmd_tap_drv;
  56
  57 static const char *valid_arguments[] = {
  58         ETH_TAP_IFACE_ARG,
  59         ETH_TAP_REMOTE_ARG,
  60         ETH_TAP_MAC_ARG,
  61         NULL
  62 };
  63
  64 static int tap_unit;
  65
  66 static volatile uint32_t tap_trigger;   /* Rx trigger */
  67
  68 static struct rte_eth_link pmd_link = {
  69         .link_speed = ETH_SPEED_NUM_10G,
  70         .link_duplex = ETH_LINK_FULL_DUPLEX,
  71         .link_status = ETH_LINK_DOWN,
  72         .link_autoneg = ETH_LINK_AUTONEG
  73 };
  74
  75 static void
  76 tap_trigger_cb(int sig __rte_unused)
  77 {
  78         /* Valid trigger values are nonzero */
  79         tap_trigger = (tap_trigger + 1) | 0x80000000;
  80 }
  81
  82 /* Specifies on what netdevices the ioctl should be applied */
  83 enum ioctl_mode {
  84         LOCAL_AND_REMOTE,
  85         LOCAL_ONLY,
  86         REMOTE_ONLY,
  87 };
  88
  89 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
  90
  91 /* Tun/Tap allocation routine
  92  *
  93  * name is the number of the interface to use, unless NULL to take the host
  94  * supplied name.
  95  */
  96 static int
  97 tun_alloc(struct pmd_internals *pmd)
  98 {
  99         struct ifreq ifr;
 100 #ifdef IFF_MULTI_QUEUE
 101         unsigned int features;
 102 #endif
 103         int fd;
 104
 105         memset(&ifr, 0, sizeof(struct ifreq));
 106
 107         /*
 108          * Do not set IFF_NO_PI as packet information header will be needed
 109          * to check if a received packet has been truncated.
 110          */
 111         ifr.ifr_flags = IFF_TAP;
 112         snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
 113
 114         RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
 115
 116         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 117         if (fd < 0) {
 118                 RTE_LOG(ERR, PMD, "Unable to create TAP interface\n");
 119                 goto error;
 120         }
 121
 122 #ifdef IFF_MULTI_QUEUE
 123         /* Grab the TUN features to verify we can work multi-queue */
 124         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 125                 RTE_LOG(ERR, PMD, "TAP unable to get TUN/TAP features\n");
 126                 goto error;
 127         }
 128         RTE_LOG(DEBUG, PMD, "  TAP Features %08x\n", features);
 129
 130         if (features & IFF_MULTI_QUEUE) {
 131                 RTE_LOG(DEBUG, PMD, "  Multi-queue support for %d queues\n",
 132                         RTE_PMD_TAP_MAX_QUEUES);
 133                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 134         } else
 135 #endif
 136         {
 137                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 138                 RTE_LOG(DEBUG, PMD, "  Single queue only support\n");
 139         }
 140
 141         /* Set the TUN/TAP configuration and set the name if needed */
 142         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 143                 RTE_LOG(WARNING, PMD,
 144                         "Unable to set TUNSETIFF for %s\n",
 145                         ifr.ifr_name);
 146                 perror("TUNSETIFF");
 147                 goto error;
 148         }
 149
 150         /* Always set the file descriptor to non-blocking */
 151         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 152                 RTE_LOG(WARNING, PMD,
 153                         "Unable to set %s to nonblocking\n",
 154                         ifr.ifr_name);
 155                 perror("F_SETFL, NONBLOCK");
 156                 goto error;
 157         }
 158
 159         /* Set up trigger to optimize empty Rx bursts */
 160         errno = 0;
 161         do {
 162                 struct sigaction sa;
 163                 int flags = fcntl(fd, F_GETFL);
 164
 165                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 166                         break;
 167                 if (sa.sa_handler != tap_trigger_cb) {
 168                         /*
 169                          * Make sure SIGIO is not already taken. This is done
 170                          * as late as possible to leave the application a
 171                          * chance to set up its own signal handler first.
 172                          */
 173                         if (sa.sa_handler != SIG_IGN &&
 174                             sa.sa_handler != SIG_DFL) {
 175                                 errno = EBUSY;
 176                                 break;
 177                         }
 178                         sa = (struct sigaction){
 179                                 .sa_flags = SA_RESTART,
 180                                 .sa_handler = tap_trigger_cb,
 181                         };
 182                         if (sigaction(SIGIO, &sa, NULL) == -1)
 183                                 break;
 184                 }
 185                 /* Enable SIGIO on file descriptor */
 186                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 187                 fcntl(fd, F_SETOWN, getpid());
 188         } while (0);
 189         if (errno) {
 190                 /* Disable trigger globally in case of error */
 191                 tap_trigger = 0;
 192                 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
 193                         strerror(errno));
 194         }
 195
 196         return fd;
 197
 198 error:
 199         if (fd > 0)
 200                 close(fd);
 201         return -1;
 202 }
 203
 204 static void
 205 tap_verify_csum(struct rte_mbuf *mbuf)
 206 {
 207         uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
 208         uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
 209         uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
 210         unsigned int l2_len = sizeof(struct ether_hdr);
 211         unsigned int l3_len;
 212         uint16_t cksum = 0;
 213         void *l3_hdr;
 214         void *l4_hdr;
 215
 216         if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
 217                 l2_len += 4;
 218         else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
 219                 l2_len += 8;
 220         /* Don't verify checksum for packets with discontinuous L2 header */
 221         if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
 222                      rte_pktmbuf_data_len(mbuf)))
 223                 return;
 224         l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
 225         if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 226                 struct ipv4_hdr *iph = l3_hdr;
 227
 228                 /* ihl contains the number of 4-byte words in the header */
 229                 l3_len = 4 * (iph->version_ihl & 0xf);
 230                 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
 231                         return;
 232
 233                 cksum = ~rte_raw_cksum(iph, l3_len);
 234                 mbuf->ol_flags |= cksum ?
 235                         PKT_RX_IP_CKSUM_BAD :
 236                         PKT_RX_IP_CKSUM_GOOD;
 237         } else if (l3 == RTE_PTYPE_L3_IPV6) {
 238                 l3_len = sizeof(struct ipv6_hdr);
 239         } else {
 240                 /* IPv6 extensions are not supported */
 241                 return;
 242         }
 243         if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
 244                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 245                 /* Don't verify checksum for multi-segment packets. */
 246                 if (mbuf->nb_segs > 1)
 247                         return;
 248                 if (l3 == RTE_PTYPE_L3_IPV4)
 249                         cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
 250                 else if (l3 == RTE_PTYPE_L3_IPV6)
 251                         cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
 252                 mbuf->ol_flags |= cksum ?
 253                         PKT_RX_L4_CKSUM_BAD :
 254                         PKT_RX_L4_CKSUM_GOOD;
 255         }
 256 }
 257
 258 static uint64_t
 259 tap_rx_offload_get_port_capa(void)
 260 {
 261         /*
 262          * In order to support legacy apps,
 263          * report capabilities also as port capabilities.
 264          */
 265         return DEV_RX_OFFLOAD_SCATTER |
 266                DEV_RX_OFFLOAD_IPV4_CKSUM |
 267                DEV_RX_OFFLOAD_UDP_CKSUM |
 268                DEV_RX_OFFLOAD_TCP_CKSUM |
 269                DEV_RX_OFFLOAD_CRC_STRIP;
 270 }
 271
 272 static uint64_t
 273 tap_rx_offload_get_queue_capa(void)
 274 {
 275         return DEV_RX_OFFLOAD_SCATTER |
 276                DEV_RX_OFFLOAD_IPV4_CKSUM |
 277                DEV_RX_OFFLOAD_UDP_CKSUM |
 278                DEV_RX_OFFLOAD_TCP_CKSUM |
 279                DEV_RX_OFFLOAD_CRC_STRIP;
 280 }
 281
 282 static bool
 283 tap_rxq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 284 {
 285         uint64_t port_offloads = dev->data->dev_conf.rxmode.offloads;
 286         uint64_t queue_supp_offloads = tap_rx_offload_get_queue_capa();
 287         uint64_t port_supp_offloads = tap_rx_offload_get_port_capa();
 288
 289         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 290             offloads)
 291                 return false;
 292         if ((port_offloads ^ offloads) & port_supp_offloads)
 293                 return false;
 294         return true;
 295 }
 296
 297 /* Callback to handle the rx burst of packets to the correct interface and
 298  * file descriptor(s) in a multi-queue setup.
 299  */
 300 static uint16_t
 301 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 302 {
 303         struct rx_queue *rxq = queue;
 304         uint16_t num_rx;
 305         unsigned long num_rx_bytes = 0;
 306         uint32_t trigger = tap_trigger;
 307
 308         if (trigger == rxq->trigger_seen)
 309                 return 0;
 310         if (trigger)
 311                 rxq->trigger_seen = trigger;
 312         rte_compiler_barrier();
 313         for (num_rx = 0; num_rx < nb_pkts; ) {
 314                 struct rte_mbuf *mbuf = rxq->pool;
 315                 struct rte_mbuf *seg = NULL;
 316                 struct rte_mbuf *new_tail = NULL;
 317                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 318                 int len;
 319
 320                 len = readv(rxq->fd, *rxq->iovecs,
 321                             1 +
 322                             (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
 323                              rxq->nb_rx_desc : 1));
 324                 if (len < (int)sizeof(struct tun_pi))
 325                         break;
 326
 327                 /* Packet couldn't fit in the provided mbuf */
 328                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 329                         rxq->stats.ierrors++;
 330                         continue;
 331                 }
 332
 333                 len -= sizeof(struct tun_pi);
 334
 335                 mbuf->pkt_len = len;
 336                 mbuf->port = rxq->in_port;
 337                 while (1) {
 338                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 339
 340                         if (unlikely(!buf)) {
 341                                 rxq->stats.rx_nombuf++;
 342                                 /* No new buf has been allocated: do nothing */
 343                                 if (!new_tail || !seg)
 344                                         goto end;
 345
 346                                 seg->next = NULL;
 347                                 rte_pktmbuf_free(mbuf);
 348
 349                                 goto end;
 350                         }
 351                         seg = seg ? seg->next : mbuf;
 352                         if (rxq->pool == mbuf)
 353                                 rxq->pool = buf;
 354                         if (new_tail)
 355                                 new_tail->next = buf;
 356                         new_tail = buf;
 357                         new_tail->next = seg->next;
 358
 359                         /* iovecs[0] is reserved for packet info (pi) */
 360                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 361                                 buf->buf_len - data_off;
 362                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 363                                 (char *)buf->buf_addr + data_off;
 364
 365                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 366                         seg->data_off = data_off;
 367
 368                         len -= seg->data_len;
 369                         if (len <= 0)
 370                                 break;
 371                         mbuf->nb_segs++;
 372                         /* First segment has headroom, not the others */
 373                         data_off = 0;
 374                 }
 375                 seg->next = NULL;
 376                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 377                                                       RTE_PTYPE_ALL_MASK);
 378                 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
 379                         tap_verify_csum(mbuf);
 380
 381                 /* account for the receive frame */
 382                 bufs[num_rx++] = mbuf;
 383                 num_rx_bytes += mbuf->pkt_len;
 384         }
 385 end:
 386         rxq->stats.ipackets += num_rx;
 387         rxq->stats.ibytes += num_rx_bytes;
 388
 389         return num_rx;
 390 }
 391
 392 static uint64_t
 393 tap_tx_offload_get_port_capa(void)
 394 {
 395         /*
 396          * In order to support legacy apps,
 397          * report capabilities also as port capabilities.
 398          */
 399         return DEV_TX_OFFLOAD_MULTI_SEGS |
 400                DEV_TX_OFFLOAD_IPV4_CKSUM |
 401                DEV_TX_OFFLOAD_UDP_CKSUM |
 402                DEV_TX_OFFLOAD_TCP_CKSUM;
 403 }
 404
 405 static uint64_t
 406 tap_tx_offload_get_queue_capa(void)
 407 {
 408         return DEV_TX_OFFLOAD_MULTI_SEGS |
 409                DEV_TX_OFFLOAD_IPV4_CKSUM |
 410                DEV_TX_OFFLOAD_UDP_CKSUM |
 411                DEV_TX_OFFLOAD_TCP_CKSUM;
 412 }
 413
 414 static bool
 415 tap_txq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 416 {
 417         uint64_t port_offloads = dev->data->dev_conf.txmode.offloads;
 418         uint64_t queue_supp_offloads = tap_tx_offload_get_queue_capa();
 419         uint64_t port_supp_offloads = tap_tx_offload_get_port_capa();
 420
 421         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 422             offloads)
 423                 return false;
 424         /* Verify we have no conflict with port offloads */
 425         if ((port_offloads ^ offloads) & port_supp_offloads)
 426                 return false;
 427         return true;
 428 }
 429
 430 static void
 431 tap_tx_offload(char *packet, uint64_t ol_flags, unsigned int l2_len,
 432                unsigned int l3_len)
 433 {
 434         void *l3_hdr = packet + l2_len;
 435
 436         if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
 437                 struct ipv4_hdr *iph = l3_hdr;
 438                 uint16_t cksum;
 439
 440                 iph->hdr_checksum = 0;
 441                 cksum = rte_raw_cksum(iph, l3_len);
 442                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 443         }
 444         if (ol_flags & PKT_TX_L4_MASK) {
 445                 uint16_t l4_len;
 446                 uint32_t cksum;
 447                 uint16_t *l4_cksum;
 448                 void *l4_hdr;
 449
 450                 l4_hdr = packet + l2_len + l3_len;
 451                 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
 452                         l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
 453                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 454                         l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
 455                 else
 456                         return;
 457                 *l4_cksum = 0;
 458                 if (ol_flags & PKT_TX_IPV4) {
 459                         struct ipv4_hdr *iph = l3_hdr;
 460
 461                         l4_len = rte_be_to_cpu_16(iph->total_length) - l3_len;
 462                         cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 463                 } else {
 464                         struct ipv6_hdr *ip6h = l3_hdr;
 465
 466                         /* payload_len does not include ext headers */
 467                         l4_len = rte_be_to_cpu_16(ip6h->payload_len) -
 468                                 l3_len + sizeof(struct ipv6_hdr);
 469                         cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 470                 }
 471                 cksum += rte_raw_cksum(l4_hdr, l4_len);
 472                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 473                 cksum = (~cksum) & 0xffff;
 474                 if (cksum == 0)
 475                         cksum = 0xffff;
 476                 *l4_cksum = cksum;
 477         }
 478 }
 479
 480 /* Callback to handle sending packets from the tap interface
 481  */
 482 static uint16_t
 483 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 484 {
 485         struct tx_queue *txq = queue;
 486         uint16_t num_tx = 0;
 487         unsigned long num_tx_bytes = 0;
 488         uint32_t max_size;
 489         int i;
 490
 491         if (unlikely(nb_pkts == 0))
 492                 return 0;
 493
 494         max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
 495         for (i = 0; i < nb_pkts; i++) {
 496                 struct rte_mbuf *mbuf = bufs[num_tx];
 497                 struct iovec iovecs[mbuf->nb_segs + 1];
 498                 struct tun_pi pi = { .flags = 0 };
 499                 struct rte_mbuf *seg = mbuf;
 500                 char m_copy[mbuf->data_len];
 501                 int n;
 502                 int j;
 503
 504                 /* stats.errs will be incremented */
 505                 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
 506                         break;
 507
 508                 iovecs[0].iov_base = &pi;
 509                 iovecs[0].iov_len = sizeof(pi);
 510                 for (j = 1; j <= mbuf->nb_segs; j++) {
 511                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 512                         iovecs[j].iov_base =
 513                                 rte_pktmbuf_mtod(seg, void *);
 514                         seg = seg->next;
 515                 }
 516                 if (txq->csum &&
 517                     ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 518                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 519                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
 520                         /* Support only packets with all data in the same seg */
 521                         if (mbuf->nb_segs > 1)
 522                                 break;
 523                         /* To change checksums, work on a copy of data. */
 524                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 525                                    rte_pktmbuf_data_len(mbuf));
 526                         tap_tx_offload(m_copy, mbuf->ol_flags,
 527                                        mbuf->l2_len, mbuf->l3_len);
 528                         iovecs[1].iov_base = m_copy;
 529                 }
 530                 /* copy the tx frame data */
 531                 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
 532                 if (n <= 0)
 533                         break;
 534
 535                 num_tx++;
 536                 num_tx_bytes += mbuf->pkt_len;
 537                 rte_pktmbuf_free(mbuf);
 538         }
 539
 540         txq->stats.opackets += num_tx;
 541         txq->stats.errs += nb_pkts - num_tx;
 542         txq->stats.obytes += num_tx_bytes;
 543
 544         return num_tx;
 545 }
 546
 547 static const char *
 548 tap_ioctl_req2str(unsigned long request)
 549 {
 550         switch (request) {
 551         case SIOCSIFFLAGS:
 552                 return "SIOCSIFFLAGS";
 553         case SIOCGIFFLAGS:
 554                 return "SIOCGIFFLAGS";
 555         case SIOCGIFHWADDR:
 556                 return "SIOCGIFHWADDR";
 557         case SIOCSIFHWADDR:
 558                 return "SIOCSIFHWADDR";
 559         case SIOCSIFMTU:
 560                 return "SIOCSIFMTU";
 561         }
 562         return "UNKNOWN";
 563 }
 564
 565 static int
 566 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 567           struct ifreq *ifr, int set, enum ioctl_mode mode)
 568 {
 569         short req_flags = ifr->ifr_flags;
 570         int remote = pmd->remote_if_index &&
 571                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 572
 573         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 574                 return 0;
 575         /*
 576          * If there is a remote netdevice, apply ioctl on it, then apply it on
 577          * the tap netdevice.
 578          */
 579 apply:
 580         if (remote)
 581                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
 582         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 583                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 584         switch (request) {
 585         case SIOCSIFFLAGS:
 586                 /* fetch current flags to leave other flags untouched */
 587                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 588                         goto error;
 589                 if (set)
 590                         ifr->ifr_flags |= req_flags;
 591                 else
 592                         ifr->ifr_flags &= ~req_flags;
 593                 break;
 594         case SIOCGIFFLAGS:
 595         case SIOCGIFHWADDR:
 596         case SIOCSIFHWADDR:
 597         case SIOCSIFMTU:
 598                 break;
 599         default:
 600                 RTE_ASSERT(!"unsupported request type: must not happen");
 601         }
 602         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 603                 goto error;
 604         if (remote-- && mode == LOCAL_AND_REMOTE)
 605                 goto apply;
 606         return 0;
 607
 608 error:
 609         RTE_LOG(DEBUG, PMD, "%s: %s(%s) failed: %s(%d)\n", ifr->ifr_name,
 610                 __func__, tap_ioctl_req2str(request), strerror(errno), errno);
 611         return -errno;
 612 }
 613
 614 static int
 615 tap_link_set_down(struct rte_eth_dev *dev)
 616 {
 617         struct pmd_internals *pmd = dev->data->dev_private;
 618         struct ifreq ifr = { .ifr_flags = IFF_UP };
 619
 620         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 621         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 622 }
 623
 624 static int
 625 tap_link_set_up(struct rte_eth_dev *dev)
 626 {
 627         struct pmd_internals *pmd = dev->data->dev_private;
 628         struct ifreq ifr = { .ifr_flags = IFF_UP };
 629
 630         dev->data->dev_link.link_status = ETH_LINK_UP;
 631         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 632 }
 633
 634 static int
 635 tap_dev_start(struct rte_eth_dev *dev)
 636 {
 637         int err;
 638
 639         err = tap_intr_handle_set(dev, 1);
 640         if (err)
 641                 return err;
 642         return tap_link_set_up(dev);
 643 }
 644
 645 /* This function gets called when the current port gets stopped.
 646  */
 647 static void
 648 tap_dev_stop(struct rte_eth_dev *dev)
 649 {
 650         tap_intr_handle_set(dev, 0);
 651         tap_link_set_down(dev);
 652 }
 653
 654 static int
 655 tap_dev_configure(struct rte_eth_dev *dev)
 656 {
 657         uint64_t supp_tx_offloads = tap_tx_offload_get_port_capa();
 658         uint64_t tx_offloads = dev->data->dev_conf.txmode.offloads;
 659
 660         if ((tx_offloads & supp_tx_offloads) != tx_offloads) {
 661                 rte_errno = ENOTSUP;
 662                 RTE_LOG(ERR, PMD,
 663                         "Some Tx offloads are not supported "
 664                         "requested 0x%" PRIx64 " supported 0x%" PRIx64 "\n",
 665                         tx_offloads, supp_tx_offloads);
 666                 return -rte_errno;
 667         }
 668         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 669                 RTE_LOG(ERR, PMD,
 670                         "%s: number of rx queues %d exceeds max num of queues %d\n",
 671                         dev->device->name,
 672                         dev->data->nb_rx_queues,
 673                         RTE_PMD_TAP_MAX_QUEUES);
 674                 return -1;
 675         }
 676         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 677                 RTE_LOG(ERR, PMD,
 678                         "%s: number of tx queues %d exceeds max num of queues %d\n",
 679                         dev->device->name,
 680                         dev->data->nb_tx_queues,
 681                         RTE_PMD_TAP_MAX_QUEUES);
 682                 return -1;
 683         }
 684
 685         RTE_LOG(INFO, PMD, "%s: %p: TX configured queues number: %u\n",
 686              dev->device->name, (void *)dev, dev->data->nb_tx_queues);
 687
 688         RTE_LOG(INFO, PMD, "%s: %p: RX configured queues number: %u\n",
 689              dev->device->name, (void *)dev, dev->data->nb_rx_queues);
 690
 691         return 0;
 692 }
 693
 694 static uint32_t
 695 tap_dev_speed_capa(void)
 696 {
 697         uint32_t speed = pmd_link.link_speed;
 698         uint32_t capa = 0;
 699
 700         if (speed >= ETH_SPEED_NUM_10M)
 701                 capa |= ETH_LINK_SPEED_10M;
 702         if (speed >= ETH_SPEED_NUM_100M)
 703                 capa |= ETH_LINK_SPEED_100M;
 704         if (speed >= ETH_SPEED_NUM_1G)
 705                 capa |= ETH_LINK_SPEED_1G;
 706         if (speed >= ETH_SPEED_NUM_5G)
 707                 capa |= ETH_LINK_SPEED_2_5G;
 708         if (speed >= ETH_SPEED_NUM_5G)
 709                 capa |= ETH_LINK_SPEED_5G;
 710         if (speed >= ETH_SPEED_NUM_10G)
 711                 capa |= ETH_LINK_SPEED_10G;
 712         if (speed >= ETH_SPEED_NUM_20G)
 713                 capa |= ETH_LINK_SPEED_20G;
 714         if (speed >= ETH_SPEED_NUM_25G)
 715                 capa |= ETH_LINK_SPEED_25G;
 716         if (speed >= ETH_SPEED_NUM_40G)
 717                 capa |= ETH_LINK_SPEED_40G;
 718         if (speed >= ETH_SPEED_NUM_50G)
 719                 capa |= ETH_LINK_SPEED_50G;
 720         if (speed >= ETH_SPEED_NUM_56G)
 721                 capa |= ETH_LINK_SPEED_56G;
 722         if (speed >= ETH_SPEED_NUM_100G)
 723                 capa |= ETH_LINK_SPEED_100G;
 724
 725         return capa;
 726 }
 727
 728 static void
 729 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 730 {
 731         struct pmd_internals *internals = dev->data->dev_private;
 732
 733         dev_info->if_index = internals->if_index;
 734         dev_info->max_mac_addrs = 1;
 735         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 736         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
 737         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
 738         dev_info->min_rx_bufsize = 0;
 739         dev_info->pci_dev = NULL;
 740         dev_info->speed_capa = tap_dev_speed_capa();
 741         dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
 742         dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
 743                                     dev_info->rx_queue_offload_capa;
 744         dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
 745         dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
 746                                     dev_info->tx_queue_offload_capa;
 747 }
 748
 749 static int
 750 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 751 {
 752         unsigned int i, imax;
 753         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 754         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 755         unsigned long rx_nombuf = 0, ierrors = 0;
 756         const struct pmd_internals *pmd = dev->data->dev_private;
 757
 758         /* rx queue statistics */
 759         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 760                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 761         for (i = 0; i < imax; i++) {
 762                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 763                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 764                 rx_total += tap_stats->q_ipackets[i];
 765                 rx_bytes_total += tap_stats->q_ibytes[i];
 766                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
 767                 ierrors += pmd->rxq[i].stats.ierrors;
 768         }
 769
 770         /* tx queue statistics */
 771         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 772                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 773
 774         for (i = 0; i < imax; i++) {
 775                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 776                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 777                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 778                 tx_total += tap_stats->q_opackets[i];
 779                 tx_err_total += tap_stats->q_errors[i];
 780                 tx_bytes_total += tap_stats->q_obytes[i];
 781         }
 782
 783         tap_stats->ipackets = rx_total;
 784         tap_stats->ibytes = rx_bytes_total;
 785         tap_stats->ierrors = ierrors;
 786         tap_stats->rx_nombuf = rx_nombuf;
 787         tap_stats->opackets = tx_total;
 788         tap_stats->oerrors = tx_err_total;
 789         tap_stats->obytes = tx_bytes_total;
 790         return 0;
 791 }
 792
 793 static void
 794 tap_stats_reset(struct rte_eth_dev *dev)
 795 {
 796         int i;
 797         struct pmd_internals *pmd = dev->data->dev_private;
 798
 799         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 800                 pmd->rxq[i].stats.ipackets = 0;
 801                 pmd->rxq[i].stats.ibytes = 0;
 802                 pmd->rxq[i].stats.ierrors = 0;
 803                 pmd->rxq[i].stats.rx_nombuf = 0;
 804
 805                 pmd->txq[i].stats.opackets = 0;
 806                 pmd->txq[i].stats.errs = 0;
 807                 pmd->txq[i].stats.obytes = 0;
 808         }
 809 }
 810
 811 static void
 812 tap_dev_close(struct rte_eth_dev *dev)
 813 {
 814         int i;
 815         struct pmd_internals *internals = dev->data->dev_private;
 816
 817         tap_link_set_down(dev);
 818         tap_flow_flush(dev, NULL);
 819         tap_flow_implicit_flush(internals, NULL);
 820
 821         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 822                 if (internals->rxq[i].fd != -1) {
 823                         close(internals->rxq[i].fd);
 824                         internals->rxq[i].fd = -1;
 825                 }
 826                 if (internals->txq[i].fd != -1) {
 827                         close(internals->txq[i].fd);
 828                         internals->txq[i].fd = -1;
 829                 }
 830         }
 831
 832         if (internals->remote_if_index) {
 833                 /* Restore initial remote state */
 834                 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
 835                                 &internals->remote_initial_flags);
 836         }
 837 }
 838
 839 static void
 840 tap_rx_queue_release(void *queue)
 841 {
 842         struct rx_queue *rxq = queue;
 843
 844         if (rxq && (rxq->fd > 0)) {
 845                 close(rxq->fd);
 846                 rxq->fd = -1;
 847                 rte_pktmbuf_free(rxq->pool);
 848                 rte_free(rxq->iovecs);
 849                 rxq->pool = NULL;
 850                 rxq->iovecs = NULL;
 851         }
 852 }
 853
 854 static void
 855 tap_tx_queue_release(void *queue)
 856 {
 857         struct tx_queue *txq = queue;
 858
 859         if (txq && (txq->fd > 0)) {
 860                 close(txq->fd);
 861                 txq->fd = -1;
 862         }
 863 }
 864
 865 static int
 866 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
 867 {
 868         struct rte_eth_link *dev_link = &dev->data->dev_link;
 869         struct pmd_internals *pmd = dev->data->dev_private;
 870         struct ifreq ifr = { .ifr_flags = 0 };
 871
 872         if (pmd->remote_if_index) {
 873                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
 874                 if (!(ifr.ifr_flags & IFF_UP) ||
 875                     !(ifr.ifr_flags & IFF_RUNNING)) {
 876                         dev_link->link_status = ETH_LINK_DOWN;
 877                         return 0;
 878                 }
 879         }
 880         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
 881         dev_link->link_status =
 882                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
 883                  ETH_LINK_UP :
 884                  ETH_LINK_DOWN);
 885         return 0;
 886 }
 887
 888 static void
 889 tap_promisc_enable(struct rte_eth_dev *dev)
 890 {
 891         struct pmd_internals *pmd = dev->data->dev_private;
 892         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 893
 894         dev->data->promiscuous = 1;
 895         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 896         if (pmd->remote_if_index && !pmd->flow_isolate)
 897                 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
 898 }
 899
 900 static void
 901 tap_promisc_disable(struct rte_eth_dev *dev)
 902 {
 903         struct pmd_internals *pmd = dev->data->dev_private;
 904         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 905
 906         dev->data->promiscuous = 0;
 907         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 908         if (pmd->remote_if_index && !pmd->flow_isolate)
 909                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
 910 }
 911
 912 static void
 913 tap_allmulti_enable(struct rte_eth_dev *dev)
 914 {
 915         struct pmd_internals *pmd = dev->data->dev_private;
 916         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 917
 918         dev->data->all_multicast = 1;
 919         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 920         if (pmd->remote_if_index && !pmd->flow_isolate)
 921                 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
 922 }
 923
 924 static void
 925 tap_allmulti_disable(struct rte_eth_dev *dev)
 926 {
 927         struct pmd_internals *pmd = dev->data->dev_private;
 928         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 929
 930         dev->data->all_multicast = 0;
 931         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 932         if (pmd->remote_if_index && !pmd->flow_isolate)
 933                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
 934 }
 935
 936 static void
 937 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 938 {
 939         struct pmd_internals *pmd = dev->data->dev_private;
 940         enum ioctl_mode mode = LOCAL_ONLY;
 941         struct ifreq ifr;
 942
 943         if (is_zero_ether_addr(mac_addr)) {
 944                 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
 945                         dev->device->name);
 946                 return;
 947         }
 948         /* Check the actual current MAC address on the tap netdevice */
 949         if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
 950                 return;
 951         if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 952                                mac_addr))
 953                 return;
 954         /* Check the current MAC address on the remote */
 955         if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0)
 956                 return;
 957         if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 958                                mac_addr))
 959                 mode = LOCAL_AND_REMOTE;
 960         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
 961         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
 962         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode) < 0)
 963                 return;
 964         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
 965         if (pmd->remote_if_index && !pmd->flow_isolate) {
 966                 /* Replace MAC redirection rule after a MAC change */
 967                 if (tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC) < 0) {
 968                         RTE_LOG(ERR, PMD,
 969                                 "%s: Couldn't delete MAC redirection rule\n",
 970                                 dev->device->name);
 971                         return;
 972                 }
 973                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0)
 974                         RTE_LOG(ERR, PMD,
 975                                 "%s: Couldn't add MAC redirection rule\n",
 976                                 dev->device->name);
 977         }
 978 }
 979
 980 static int
 981 tap_setup_queue(struct rte_eth_dev *dev,
 982                 struct pmd_internals *internals,
 983                 uint16_t qid,
 984                 int is_rx)
 985 {
 986         int *fd;
 987         int *other_fd;
 988         const char *dir;
 989         struct pmd_internals *pmd = dev->data->dev_private;
 990         struct rx_queue *rx = &internals->rxq[qid];
 991         struct tx_queue *tx = &internals->txq[qid];
 992
 993         if (is_rx) {
 994                 fd = &rx->fd;
 995                 other_fd = &tx->fd;
 996                 dir = "rx";
 997         } else {
 998                 fd = &tx->fd;
 999                 other_fd = &rx->fd;
1000                 dir = "tx";
1001         }
1002         if (*fd != -1) {
1003                 /* fd for this queue already exists */
1004                 RTE_LOG(DEBUG, PMD, "%s: fd %d for %s queue qid %d exists\n",
1005                         pmd->name, *fd, dir, qid);
1006         } else if (*other_fd != -1) {
1007                 /* Only other_fd exists. dup it */
1008                 *fd = dup(*other_fd);
1009                 if (*fd < 0) {
1010                         *fd = -1;
1011                         RTE_LOG(ERR, PMD, "%s: dup() failed.\n",
1012                                 pmd->name);
1013                         return -1;
1014                 }
1015                 RTE_LOG(DEBUG, PMD, "%s: dup fd %d for %s queue qid %d (%d)\n",
1016                         pmd->name, *other_fd, dir, qid, *fd);
1017         } else {
1018                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1019                 *fd = tun_alloc(pmd);
1020                 if (*fd < 0) {
1021                         *fd = -1; /* restore original value */
1022                         RTE_LOG(ERR, PMD, "%s: tun_alloc() failed.\n",
1023                                 pmd->name);
1024                         return -1;
1025                 }
1026                 RTE_LOG(DEBUG, PMD, "%s: add %s queue for qid %d fd %d\n",
1027                         pmd->name, dir, qid, *fd);
1028         }
1029
1030         tx->mtu = &dev->data->mtu;
1031         rx->rxmode = &dev->data->dev_conf.rxmode;
1032
1033         return *fd;
1034 }
1035
1036 static int
1037 tap_rx_queue_setup(struct rte_eth_dev *dev,
1038                    uint16_t rx_queue_id,
1039                    uint16_t nb_rx_desc,
1040                    unsigned int socket_id,
1041                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1042                    struct rte_mempool *mp)
1043 {
1044         struct pmd_internals *internals = dev->data->dev_private;
1045         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1046         struct rte_mbuf **tmp = &rxq->pool;
1047         long iov_max = sysconf(_SC_IOV_MAX);
1048         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1049         struct iovec (*iovecs)[nb_desc + 1];
1050         int data_off = RTE_PKTMBUF_HEADROOM;
1051         int ret = 0;
1052         int fd;
1053         int i;
1054
1055         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1056                 RTE_LOG(WARNING, PMD,
1057                         "nb_rx_queues %d too small or mempool NULL\n",
1058                         dev->data->nb_rx_queues);
1059                 return -1;
1060         }
1061
1062         /* Verify application offloads are valid for our port and queue. */
1063         if (!tap_rxq_are_offloads_valid(dev, rx_conf->offloads)) {
1064                 rte_errno = ENOTSUP;
1065                 RTE_LOG(ERR, PMD,
1066                         "%p: Rx queue offloads 0x%" PRIx64
1067                         " don't match port offloads 0x%" PRIx64
1068                         " or supported offloads 0x%" PRIx64 "\n",
1069                         (void *)dev, rx_conf->offloads,
1070                         dev->data->dev_conf.rxmode.offloads,
1071                         (tap_rx_offload_get_port_capa() |
1072                          tap_rx_offload_get_queue_capa()));
1073                 return -rte_errno;
1074         }
1075         rxq->mp = mp;
1076         rxq->trigger_seen = 1; /* force initial burst */
1077         rxq->in_port = dev->data->port_id;
1078         rxq->nb_rx_desc = nb_desc;
1079         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1080                                     socket_id);
1081         if (!iovecs) {
1082                 RTE_LOG(WARNING, PMD,
1083                         "%s: Couldn't allocate %d RX descriptors\n",
1084                         dev->device->name, nb_desc);
1085                 return -ENOMEM;
1086         }
1087         rxq->iovecs = iovecs;
1088
1089         dev->data->rx_queues[rx_queue_id] = rxq;
1090         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1091         if (fd == -1) {
1092                 ret = fd;
1093                 goto error;
1094         }
1095
1096         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1097         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1098
1099         for (i = 1; i <= nb_desc; i++) {
1100                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1101                 if (!*tmp) {
1102                         RTE_LOG(WARNING, PMD,
1103                                 "%s: couldn't allocate memory for queue %d\n",
1104                                 dev->device->name, rx_queue_id);
1105                         ret = -ENOMEM;
1106                         goto error;
1107                 }
1108                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1109                 (*rxq->iovecs)[i].iov_base =
1110                         (char *)(*tmp)->buf_addr + data_off;
1111                 data_off = 0;
1112                 tmp = &(*tmp)->next;
1113         }
1114
1115         RTE_LOG(DEBUG, PMD, "  RX TAP device name %s, qid %d on fd %d\n",
1116                 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
1117
1118         return 0;
1119
1120 error:
1121         rte_pktmbuf_free(rxq->pool);
1122         rxq->pool = NULL;
1123         rte_free(rxq->iovecs);
1124         rxq->iovecs = NULL;
1125         return ret;
1126 }
1127
1128 static int
1129 tap_tx_queue_setup(struct rte_eth_dev *dev,
1130                    uint16_t tx_queue_id,
1131                    uint16_t nb_tx_desc __rte_unused,
1132                    unsigned int socket_id __rte_unused,
1133                    const struct rte_eth_txconf *tx_conf)
1134 {
1135         struct pmd_internals *internals = dev->data->dev_private;
1136         struct tx_queue *txq;
1137         int ret;
1138
1139         if (tx_queue_id >= dev->data->nb_tx_queues)
1140                 return -1;
1141         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1142         txq = dev->data->tx_queues[tx_queue_id];
1143         /*
1144          * Don't verify port offloads for application which
1145          * use the old API.
1146          */
1147         if (tx_conf != NULL &&
1148             !!(tx_conf->txq_flags & ETH_TXQ_FLAGS_IGNORE)) {
1149                 if (tap_txq_are_offloads_valid(dev, tx_conf->offloads)) {
1150                         txq->csum = !!(tx_conf->offloads &
1151                                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
1152                                          DEV_TX_OFFLOAD_UDP_CKSUM |
1153                                          DEV_TX_OFFLOAD_TCP_CKSUM));
1154                 } else {
1155                         rte_errno = ENOTSUP;
1156                         RTE_LOG(ERR, PMD,
1157                                 "%p: Tx queue offloads 0x%" PRIx64
1158                                 " don't match port offloads 0x%" PRIx64
1159                                 " or supported offloads 0x%" PRIx64,
1160                                 (void *)dev, tx_conf->offloads,
1161                                 dev->data->dev_conf.txmode.offloads,
1162                                 tap_tx_offload_get_port_capa());
1163                         return -rte_errno;
1164                 }
1165         }
1166         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1167         if (ret == -1)
1168                 return -1;
1169         RTE_LOG(DEBUG, PMD,
1170                 "  TX TAP device name %s, qid %d on fd %d csum %s\n",
1171                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd,
1172                 txq->csum ? "on" : "off");
1173
1174         return 0;
1175 }
1176
1177 static int
1178 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1179 {
1180         struct pmd_internals *pmd = dev->data->dev_private;
1181         struct ifreq ifr = { .ifr_mtu = mtu };
1182         int err = 0;
1183
1184         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1185         if (!err)
1186                 dev->data->mtu = mtu;
1187
1188         return err;
1189 }
1190
1191 static int
1192 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1193                      struct ether_addr *mc_addr_set __rte_unused,
1194                      uint32_t nb_mc_addr __rte_unused)
1195 {
1196         /*
1197          * Nothing to do actually: the tap has no filtering whatsoever, every
1198          * packet is received.
1199          */
1200         return 0;
1201 }
1202
1203 static int
1204 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1205 {
1206         struct rte_eth_dev *dev = arg;
1207         struct pmd_internals *pmd = dev->data->dev_private;
1208         struct ifinfomsg *info = NLMSG_DATA(nh);
1209
1210         if (nh->nlmsg_type != RTM_NEWLINK ||
1211             (info->ifi_index != pmd->if_index &&
1212              info->ifi_index != pmd->remote_if_index))
1213                 return 0;
1214         return tap_link_update(dev, 0);
1215 }
1216
1217 static void
1218 tap_dev_intr_handler(void *cb_arg)
1219 {
1220         struct rte_eth_dev *dev = cb_arg;
1221         struct pmd_internals *pmd = dev->data->dev_private;
1222
1223         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1224 }
1225
1226 static int
1227 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1228 {
1229         struct pmd_internals *pmd = dev->data->dev_private;
1230
1231         /* In any case, disable interrupt if the conf is no longer there. */
1232         if (!dev->data->dev_conf.intr_conf.lsc) {
1233                 if (pmd->intr_handle.fd != -1) {
1234                         tap_nl_final(pmd->intr_handle.fd);
1235                         rte_intr_callback_unregister(&pmd->intr_handle,
1236                                 tap_dev_intr_handler, dev);
1237                 }
1238                 return 0;
1239         }
1240         if (set) {
1241                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1242                 if (unlikely(pmd->intr_handle.fd == -1))
1243                         return -EBADF;
1244                 return rte_intr_callback_register(
1245                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1246         }
1247         tap_nl_final(pmd->intr_handle.fd);
1248         return rte_intr_callback_unregister(&pmd->intr_handle,
1249                                             tap_dev_intr_handler, dev);
1250 }
1251
1252 static int
1253 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1254 {
1255         int err;
1256
1257         err = tap_lsc_intr_handle_set(dev, set);
1258         if (err)
1259                 return err;
1260         err = tap_rx_intr_vec_set(dev, set);
1261         if (err && set)
1262                 tap_lsc_intr_handle_set(dev, 0);
1263         return err;
1264 }
1265
1266 static const uint32_t*
1267 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1268 {
1269         static const uint32_t ptypes[] = {
1270                 RTE_PTYPE_INNER_L2_ETHER,
1271                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1272                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1273                 RTE_PTYPE_INNER_L3_IPV4,
1274                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1275                 RTE_PTYPE_INNER_L3_IPV6,
1276                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1277                 RTE_PTYPE_INNER_L4_FRAG,
1278                 RTE_PTYPE_INNER_L4_UDP,
1279                 RTE_PTYPE_INNER_L4_TCP,
1280                 RTE_PTYPE_INNER_L4_SCTP,
1281                 RTE_PTYPE_L2_ETHER,
1282                 RTE_PTYPE_L2_ETHER_VLAN,
1283                 RTE_PTYPE_L2_ETHER_QINQ,
1284                 RTE_PTYPE_L3_IPV4,
1285                 RTE_PTYPE_L3_IPV4_EXT,
1286                 RTE_PTYPE_L3_IPV6_EXT,
1287                 RTE_PTYPE_L3_IPV6,
1288                 RTE_PTYPE_L4_FRAG,
1289                 RTE_PTYPE_L4_UDP,
1290                 RTE_PTYPE_L4_TCP,
1291                 RTE_PTYPE_L4_SCTP,
1292         };
1293
1294         return ptypes;
1295 }
1296
1297 static int
1298 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1299                   struct rte_eth_fc_conf *fc_conf)
1300 {
1301         fc_conf->mode = RTE_FC_NONE;
1302         return 0;
1303 }
1304
1305 static int
1306 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1307                   struct rte_eth_fc_conf *fc_conf)
1308 {
1309         if (fc_conf->mode != RTE_FC_NONE)
1310                 return -ENOTSUP;
1311         return 0;
1312 }
1313
1314 static const struct eth_dev_ops ops = {
1315         .dev_start              = tap_dev_start,
1316         .dev_stop               = tap_dev_stop,
1317         .dev_close              = tap_dev_close,
1318         .dev_configure          = tap_dev_configure,
1319         .dev_infos_get          = tap_dev_info,
1320         .rx_queue_setup         = tap_rx_queue_setup,
1321         .tx_queue_setup         = tap_tx_queue_setup,
1322         .rx_queue_release       = tap_rx_queue_release,
1323         .tx_queue_release       = tap_tx_queue_release,
1324         .flow_ctrl_get          = tap_flow_ctrl_get,
1325         .flow_ctrl_set          = tap_flow_ctrl_set,
1326         .link_update            = tap_link_update,
1327         .dev_set_link_up        = tap_link_set_up,
1328         .dev_set_link_down      = tap_link_set_down,
1329         .promiscuous_enable     = tap_promisc_enable,
1330         .promiscuous_disable    = tap_promisc_disable,
1331         .allmulticast_enable    = tap_allmulti_enable,
1332         .allmulticast_disable   = tap_allmulti_disable,
1333         .mac_addr_set           = tap_mac_set,
1334         .mtu_set                = tap_mtu_set,
1335         .set_mc_addr_list       = tap_set_mc_addr_list,
1336         .stats_get              = tap_stats_get,
1337         .stats_reset            = tap_stats_reset,
1338         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1339         .filter_ctrl            = tap_dev_filter_ctrl,
1340 };
1341
1342 static int
1343 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1344                    char *remote_iface, struct ether_addr *mac_addr)
1345 {
1346         int numa_node = rte_socket_id();
1347         struct rte_eth_dev *dev;
1348         struct pmd_internals *pmd;
1349         struct rte_eth_dev_data *data;
1350         struct ifreq ifr;
1351         int i;
1352
1353         RTE_LOG(DEBUG, PMD, "  TAP device on numa %u\n", rte_socket_id());
1354
1355         data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
1356         if (!data) {
1357                 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
1358                 goto error_exit_nodev;
1359         }
1360
1361         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1362         if (!dev) {
1363                 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
1364                 goto error_exit_nodev;
1365         }
1366
1367         pmd = dev->data->dev_private;
1368         pmd->dev = dev;
1369         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1370
1371         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1372         if (pmd->ioctl_sock == -1) {
1373                 RTE_LOG(ERR, PMD,
1374                         "TAP Unable to get a socket for management: %s\n",
1375                         strerror(errno));
1376                 goto error_exit;
1377         }
1378
1379         /* Setup some default values */
1380         rte_memcpy(data, dev->data, sizeof(*data));
1381         data->dev_private = pmd;
1382         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1383         data->numa_node = numa_node;
1384
1385         data->dev_link = pmd_link;
1386         data->mac_addrs = &pmd->eth_addr;
1387         /* Set the number of RX and TX queues */
1388         data->nb_rx_queues = 0;
1389         data->nb_tx_queues = 0;
1390
1391         dev->data = data;
1392         dev->dev_ops = &ops;
1393         dev->rx_pkt_burst = pmd_rx_burst;
1394         dev->tx_pkt_burst = pmd_tx_burst;
1395
1396         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1397         pmd->intr_handle.fd = -1;
1398         dev->intr_handle = &pmd->intr_handle;
1399
1400         /* Presetup the fds to -1 as being not valid */
1401         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1402                 pmd->rxq[i].fd = -1;
1403                 pmd->txq[i].fd = -1;
1404         }
1405
1406         if (is_zero_ether_addr(mac_addr))
1407                 eth_random_addr((uint8_t *)&pmd->eth_addr);
1408         else
1409                 rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(mac_addr));
1410
1411         /* Immediately create the netdevice (this will create the 1st queue). */
1412         /* rx queue */
1413         if (tap_setup_queue(dev, pmd, 0, 1) == -1)
1414                 goto error_exit;
1415         /* tx queue */
1416         if (tap_setup_queue(dev, pmd, 0, 0) == -1)
1417                 goto error_exit;
1418
1419         ifr.ifr_mtu = dev->data->mtu;
1420         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1421                 goto error_exit;
1422
1423         memset(&ifr, 0, sizeof(struct ifreq));
1424         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1425         rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr, ETHER_ADDR_LEN);
1426         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1427                 goto error_exit;
1428
1429         /*
1430          * Set up everything related to rte_flow:
1431          * - netlink socket
1432          * - tap / remote if_index
1433          * - mandatory QDISCs
1434          * - rte_flow actual/implicit lists
1435          * - implicit rules
1436          */
1437         pmd->nlsk_fd = tap_nl_init(0);
1438         if (pmd->nlsk_fd == -1) {
1439                 RTE_LOG(WARNING, PMD, "%s: failed to create netlink socket.\n",
1440                         pmd->name);
1441                 goto disable_rte_flow;
1442         }
1443         pmd->if_index = if_nametoindex(pmd->name);
1444         if (!pmd->if_index) {
1445                 RTE_LOG(ERR, PMD, "%s: failed to get if_index.\n", pmd->name);
1446                 goto disable_rte_flow;
1447         }
1448         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1449                 RTE_LOG(ERR, PMD, "%s: failed to create multiq qdisc.\n",
1450                         pmd->name);
1451                 goto disable_rte_flow;
1452         }
1453         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1454                 RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1455                         pmd->name);
1456                 goto disable_rte_flow;
1457         }
1458         LIST_INIT(&pmd->flows);
1459
1460         if (strlen(remote_iface)) {
1461                 pmd->remote_if_index = if_nametoindex(remote_iface);
1462                 if (!pmd->remote_if_index) {
1463                         RTE_LOG(ERR, PMD, "%s: failed to get %s if_index.\n",
1464                                 pmd->name, remote_iface);
1465                         goto error_remote;
1466                 }
1467                 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1468                          "%s", remote_iface);
1469
1470                 /* Save state of remote device */
1471                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1472
1473                 /* Replicate remote MAC address */
1474                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1475                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1476                                 pmd->name, pmd->remote_iface);
1477                         goto error_remote;
1478                 }
1479                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1480                            ETHER_ADDR_LEN);
1481                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1482                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1483                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1484                                 pmd->name, remote_iface);
1485                         goto error_remote;
1486                 }
1487
1488                 /*
1489                  * Flush usually returns negative value because it tries to
1490                  * delete every QDISC (and on a running device, one QDISC at
1491                  * least is needed). Ignore negative return value.
1492                  */
1493                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1494                 if (qdisc_create_ingress(pmd->nlsk_fd,
1495                                          pmd->remote_if_index) < 0) {
1496                         RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1497                                 pmd->remote_iface);
1498                         goto error_remote;
1499                 }
1500                 LIST_INIT(&pmd->implicit_flows);
1501                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1502                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1503                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1504                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1505                         RTE_LOG(ERR, PMD,
1506                                 "%s: failed to create implicit rules.\n",
1507                                 pmd->name);
1508                         goto error_remote;
1509                 }
1510         }
1511
1512         return 0;
1513
1514 disable_rte_flow:
1515         RTE_LOG(ERR, PMD, " Disabling rte flow support: %s(%d)\n",
1516                 strerror(errno), errno);
1517         if (strlen(remote_iface)) {
1518                 RTE_LOG(ERR, PMD, "Remote feature requires flow support.\n");
1519                 goto error_exit;
1520         }
1521         return 0;
1522
1523 error_remote:
1524         RTE_LOG(ERR, PMD, " Can't set up remote feature: %s(%d)\n",
1525                 strerror(errno), errno);
1526         tap_flow_implicit_flush(pmd, NULL);
1527
1528 error_exit:
1529         if (pmd->ioctl_sock > 0)
1530                 close(pmd->ioctl_sock);
1531         rte_eth_dev_release_port(dev);
1532
1533 error_exit_nodev:
1534         RTE_LOG(ERR, PMD, "TAP Unable to initialize %s\n",
1535                 rte_vdev_device_name(vdev));
1536
1537         rte_free(data);
1538         return -EINVAL;
1539 }
1540
1541 static int
1542 set_interface_name(const char *key __rte_unused,
1543                    const char *value,
1544                    void *extra_args)
1545 {
1546         char *name = (char *)extra_args;
1547
1548         if (value)
1549                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
1550         else
1551                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1552                          DEFAULT_TAP_NAME, (tap_unit - 1));
1553
1554         return 0;
1555 }
1556
1557 static int
1558 set_remote_iface(const char *key __rte_unused,
1559                  const char *value,
1560                  void *extra_args)
1561 {
1562         char *name = (char *)extra_args;
1563
1564         if (value)
1565                 snprintf(name, RTE_ETH_NAME_MAX_LEN, "%s", value);
1566
1567         return 0;
1568 }
1569
1570 static int parse_user_mac(struct ether_addr *user_mac,
1571                 const char *value)
1572 {
1573         unsigned int index = 0;
1574         char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
1575
1576         if (user_mac == NULL || value == NULL)
1577                 return 0;
1578
1579         snprintf(mac_temp, sizeof(mac_temp), "%s", value);
1580         mac_byte = strtok(mac_temp, ":");
1581
1582         while ((mac_byte != NULL) &&
1583                         (strlen(mac_byte) <= 2) &&
1584                         (strlen(mac_byte) == strspn(mac_byte,
1585                                         ETH_TAP_CMP_MAC_FMT))) {
1586                 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
1587                 mac_byte = strtok(NULL, ":");
1588         }
1589
1590         return index;
1591 }
1592
1593 static int
1594 set_mac_type(const char *key __rte_unused,
1595              const char *value,
1596              void *extra_args)
1597 {
1598         struct ether_addr *user_mac = extra_args;
1599
1600         if (!value)
1601                 return 0;
1602
1603         if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
1604                 static int iface_idx;
1605
1606                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1607                 memcpy((char *)user_mac->addr_bytes, "\0dtap", ETHER_ADDR_LEN);
1608                 user_mac->addr_bytes[ETHER_ADDR_LEN - 1] = iface_idx++ + '0';
1609                 goto success;
1610         }
1611
1612         if (parse_user_mac(user_mac, value) != 6)
1613                 goto error;
1614 success:
1615         RTE_LOG(DEBUG, PMD, "TAP user MAC param (%s)\n", value);
1616         return 0;
1617
1618 error:
1619         RTE_LOG(ERR, PMD, "TAP user MAC (%s) is not in format (%s|%s)\n",
1620                 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
1621         return -1;
1622 }
1623
1624 /* Open a TAP interface device.
1625  */
1626 static int
1627 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1628 {
1629         const char *name, *params;
1630         int ret;
1631         struct rte_kvargs *kvlist = NULL;
1632         int speed;
1633         char tap_name[RTE_ETH_NAME_MAX_LEN];
1634         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1635         struct ether_addr user_mac = { .addr_bytes = {0} };
1636
1637         name = rte_vdev_device_name(dev);
1638         params = rte_vdev_device_args(dev);
1639
1640         speed = ETH_SPEED_NUM_10G;
1641         snprintf(tap_name, sizeof(tap_name), "%s%d",
1642                  DEFAULT_TAP_NAME, tap_unit++);
1643         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1644
1645         if (params && (params[0] != '\0')) {
1646                 RTE_LOG(DEBUG, PMD, "parameters (%s)\n", params);
1647
1648                 kvlist = rte_kvargs_parse(params, valid_arguments);
1649                 if (kvlist) {
1650                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1651                                 ret = rte_kvargs_process(kvlist,
1652                                                          ETH_TAP_IFACE_ARG,
1653                                                          &set_interface_name,
1654                                                          tap_name);
1655                                 if (ret == -1)
1656                                         goto leave;
1657                         }
1658
1659                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1660                                 ret = rte_kvargs_process(kvlist,
1661                                                          ETH_TAP_REMOTE_ARG,
1662                                                          &set_remote_iface,
1663                                                          remote_iface);
1664                                 if (ret == -1)
1665                                         goto leave;
1666                         }
1667
1668                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1669                                 ret = rte_kvargs_process(kvlist,
1670                                                          ETH_TAP_MAC_ARG,
1671                                                          &set_mac_type,
1672                                                          &user_mac);
1673                                 if (ret == -1)
1674                                         goto leave;
1675                         }
1676                 }
1677         }
1678         pmd_link.link_speed = speed;
1679
1680         RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1681                 name, tap_name);
1682
1683         ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac);
1684
1685 leave:
1686         if (ret == -1) {
1687                 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1688                         name, tap_name);
1689                 tap_unit--;             /* Restore the unit number */
1690         }
1691         rte_kvargs_free(kvlist);
1692
1693         return ret;
1694 }
1695
1696 /* detach a TAP device.
1697  */
1698 static int
1699 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1700 {
1701         struct rte_eth_dev *eth_dev = NULL;
1702         struct pmd_internals *internals;
1703         int i;
1704
1705         RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1706                 rte_socket_id());
1707
1708         /* find the ethdev entry */
1709         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1710         if (!eth_dev)
1711                 return 0;
1712
1713         internals = eth_dev->data->dev_private;
1714         if (internals->nlsk_fd) {
1715                 tap_flow_flush(eth_dev, NULL);
1716                 tap_flow_implicit_flush(internals, NULL);
1717                 tap_nl_final(internals->nlsk_fd);
1718         }
1719         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1720                 if (internals->rxq[i].fd != -1) {
1721                         close(internals->rxq[i].fd);
1722                         internals->rxq[i].fd = -1;
1723                 }
1724                 if (internals->txq[i].fd != -1) {
1725                         close(internals->txq[i].fd);
1726                         internals->txq[i].fd = -1;
1727                 }
1728         }
1729
1730         close(internals->ioctl_sock);
1731         rte_free(eth_dev->data->dev_private);
1732         rte_free(eth_dev->data);
1733
1734         rte_eth_dev_release_port(eth_dev);
1735
1736         return 0;
1737 }
1738
1739 static struct rte_vdev_driver pmd_tap_drv = {
1740         .probe = rte_pmd_tap_probe,
1741         .remove = rte_pmd_tap_remove,
1742 };
1743 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1744 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1745 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1746                               ETH_TAP_IFACE_ARG "=<string> "
1747                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
1748                               ETH_TAP_REMOTE_ARG "=<string>");