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 <rte_eth_tap.h>
  39 #include <tap_flow.h>
  40 #include <tap_netlink.h>
  41 #include <tap_tcmsgs.h>
  42
  43 /* Linux based path to the TUN device */
  44 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  45 #define DEFAULT_TAP_NAME        "dtap"
  46
  47 #define ETH_TAP_IFACE_ARG       "iface"
  48 #define ETH_TAP_REMOTE_ARG      "remote"
  49 #define ETH_TAP_MAC_ARG         "mac"
  50 #define ETH_TAP_MAC_FIXED       "fixed"
  51
  52 #define ETH_TAP_USR_MAC_FMT     "xx:xx:xx:xx:xx:xx"
  53 #define ETH_TAP_CMP_MAC_FMT     "0123456789ABCDEFabcdef"
  54 #define ETH_TAP_MAC_ARG_FMT     ETH_TAP_MAC_FIXED "|" ETH_TAP_USR_MAC_FMT
  55
  56 static struct rte_vdev_driver pmd_tap_drv;
  57
  58 static const char *valid_arguments[] = {
  59         ETH_TAP_IFACE_ARG,
  60         ETH_TAP_REMOTE_ARG,
  61         ETH_TAP_MAC_ARG,
  62         NULL
  63 };
  64
  65 static int tap_unit;
  66
  67 static volatile uint32_t tap_trigger;   /* Rx trigger */
  68
  69 static struct rte_eth_link pmd_link = {
  70         .link_speed = ETH_SPEED_NUM_10G,
  71         .link_duplex = ETH_LINK_FULL_DUPLEX,
  72         .link_status = ETH_LINK_DOWN,
  73         .link_autoneg = ETH_LINK_AUTONEG
  74 };
  75
  76 static void
  77 tap_trigger_cb(int sig __rte_unused)
  78 {
  79         /* Valid trigger values are nonzero */
  80         tap_trigger = (tap_trigger + 1) | 0x80000000;
  81 }
  82
  83 /* Specifies on what netdevices the ioctl should be applied */
  84 enum ioctl_mode {
  85         LOCAL_AND_REMOTE,
  86         LOCAL_ONLY,
  87         REMOTE_ONLY,
  88 };
  89
  90 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
  91
  92 /* Tun/Tap allocation routine
  93  *
  94  * name is the number of the interface to use, unless NULL to take the host
  95  * supplied name.
  96  */
  97 static int
  98 tun_alloc(struct pmd_internals *pmd)
  99 {
 100         struct ifreq ifr;
 101 #ifdef IFF_MULTI_QUEUE
 102         unsigned int features;
 103 #endif
 104         int fd;
 105
 106         memset(&ifr, 0, sizeof(struct ifreq));
 107
 108         /*
 109          * Do not set IFF_NO_PI as packet information header will be needed
 110          * to check if a received packet has been truncated.
 111          */
 112         ifr.ifr_flags = IFF_TAP;
 113         snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
 114
 115         RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
 116
 117         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 118         if (fd < 0) {
 119                 RTE_LOG(ERR, PMD, "Unable to create TAP interface\n");
 120                 goto error;
 121         }
 122
 123 #ifdef IFF_MULTI_QUEUE
 124         /* Grab the TUN features to verify we can work multi-queue */
 125         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 126                 RTE_LOG(ERR, PMD, "TAP unable to get TUN/TAP features\n");
 127                 goto error;
 128         }
 129         RTE_LOG(DEBUG, PMD, "  TAP Features %08x\n", features);
 130
 131         if (features & IFF_MULTI_QUEUE) {
 132                 RTE_LOG(DEBUG, PMD, "  Multi-queue support for %d queues\n",
 133                         RTE_PMD_TAP_MAX_QUEUES);
 134                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 135         } else
 136 #endif
 137         {
 138                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 139                 RTE_LOG(DEBUG, PMD, "  Single queue only support\n");
 140         }
 141
 142         /* Set the TUN/TAP configuration and set the name if needed */
 143         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 144                 RTE_LOG(WARNING, PMD,
 145                         "Unable to set TUNSETIFF for %s\n",
 146                         ifr.ifr_name);
 147                 perror("TUNSETIFF");
 148                 goto error;
 149         }
 150
 151         /* Always set the file descriptor to non-blocking */
 152         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 153                 RTE_LOG(WARNING, PMD,
 154                         "Unable to set %s to nonblocking\n",
 155                         ifr.ifr_name);
 156                 perror("F_SETFL, NONBLOCK");
 157                 goto error;
 158         }
 159
 160         /* Set up trigger to optimize empty Rx bursts */
 161         errno = 0;
 162         do {
 163                 struct sigaction sa;
 164                 int flags = fcntl(fd, F_GETFL);
 165
 166                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 167                         break;
 168                 if (sa.sa_handler != tap_trigger_cb) {
 169                         /*
 170                          * Make sure SIGIO is not already taken. This is done
 171                          * as late as possible to leave the application a
 172                          * chance to set up its own signal handler first.
 173                          */
 174                         if (sa.sa_handler != SIG_IGN &&
 175                             sa.sa_handler != SIG_DFL) {
 176                                 errno = EBUSY;
 177                                 break;
 178                         }
 179                         sa = (struct sigaction){
 180                                 .sa_flags = SA_RESTART,
 181                                 .sa_handler = tap_trigger_cb,
 182                         };
 183                         if (sigaction(SIGIO, &sa, NULL) == -1)
 184                                 break;
 185                 }
 186                 /* Enable SIGIO on file descriptor */
 187                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 188                 fcntl(fd, F_SETOWN, getpid());
 189         } while (0);
 190         if (errno) {
 191                 /* Disable trigger globally in case of error */
 192                 tap_trigger = 0;
 193                 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
 194                         strerror(errno));
 195         }
 196
 197         return fd;
 198
 199 error:
 200         if (fd > 0)
 201                 close(fd);
 202         return -1;
 203 }
 204
 205 static void
 206 tap_verify_csum(struct rte_mbuf *mbuf)
 207 {
 208         uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
 209         uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
 210         uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
 211         unsigned int l2_len = sizeof(struct ether_hdr);
 212         unsigned int l3_len;
 213         uint16_t cksum = 0;
 214         void *l3_hdr;
 215         void *l4_hdr;
 216
 217         if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
 218                 l2_len += 4;
 219         else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
 220                 l2_len += 8;
 221         /* Don't verify checksum for packets with discontinuous L2 header */
 222         if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
 223                      rte_pktmbuf_data_len(mbuf)))
 224                 return;
 225         l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
 226         if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 227                 struct ipv4_hdr *iph = l3_hdr;
 228
 229                 /* ihl contains the number of 4-byte words in the header */
 230                 l3_len = 4 * (iph->version_ihl & 0xf);
 231                 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
 232                         return;
 233
 234                 cksum = ~rte_raw_cksum(iph, l3_len);
 235                 mbuf->ol_flags |= cksum ?
 236                         PKT_RX_IP_CKSUM_BAD :
 237                         PKT_RX_IP_CKSUM_GOOD;
 238         } else if (l3 == RTE_PTYPE_L3_IPV6) {
 239                 l3_len = sizeof(struct ipv6_hdr);
 240         } else {
 241                 /* IPv6 extensions are not supported */
 242                 return;
 243         }
 244         if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
 245                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 246                 /* Don't verify checksum for multi-segment packets. */
 247                 if (mbuf->nb_segs > 1)
 248                         return;
 249                 if (l3 == RTE_PTYPE_L3_IPV4)
 250                         cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
 251                 else if (l3 == RTE_PTYPE_L3_IPV6)
 252                         cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
 253                 mbuf->ol_flags |= cksum ?
 254                         PKT_RX_L4_CKSUM_BAD :
 255                         PKT_RX_L4_CKSUM_GOOD;
 256         }
 257 }
 258
 259 static uint64_t
 260 tap_rx_offload_get_port_capa(void)
 261 {
 262         /*
 263          * In order to support legacy apps,
 264          * report capabilities also as port capabilities.
 265          */
 266         return DEV_RX_OFFLOAD_SCATTER |
 267                DEV_RX_OFFLOAD_IPV4_CKSUM |
 268                DEV_RX_OFFLOAD_UDP_CKSUM |
 269                DEV_RX_OFFLOAD_TCP_CKSUM |
 270                DEV_RX_OFFLOAD_CRC_STRIP;
 271 }
 272
 273 static uint64_t
 274 tap_rx_offload_get_queue_capa(void)
 275 {
 276         return DEV_RX_OFFLOAD_SCATTER |
 277                DEV_RX_OFFLOAD_IPV4_CKSUM |
 278                DEV_RX_OFFLOAD_UDP_CKSUM |
 279                DEV_RX_OFFLOAD_TCP_CKSUM |
 280                DEV_RX_OFFLOAD_CRC_STRIP;
 281 }
 282
 283 static bool
 284 tap_rxq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 285 {
 286         uint64_t port_offloads = dev->data->dev_conf.rxmode.offloads;
 287         uint64_t queue_supp_offloads = tap_rx_offload_get_queue_capa();
 288         uint64_t port_supp_offloads = tap_rx_offload_get_port_capa();
 289
 290         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 291             offloads)
 292                 return false;
 293         if ((port_offloads ^ offloads) & port_supp_offloads)
 294                 return false;
 295         return true;
 296 }
 297
 298 /* Callback to handle the rx burst of packets to the correct interface and
 299  * file descriptor(s) in a multi-queue setup.
 300  */
 301 static uint16_t
 302 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 303 {
 304         struct rx_queue *rxq = queue;
 305         uint16_t num_rx;
 306         unsigned long num_rx_bytes = 0;
 307         uint32_t trigger = tap_trigger;
 308
 309         if (trigger == rxq->trigger_seen)
 310                 return 0;
 311         if (trigger)
 312                 rxq->trigger_seen = trigger;
 313         rte_compiler_barrier();
 314         for (num_rx = 0; num_rx < nb_pkts; ) {
 315                 struct rte_mbuf *mbuf = rxq->pool;
 316                 struct rte_mbuf *seg = NULL;
 317                 struct rte_mbuf *new_tail = NULL;
 318                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 319                 int len;
 320
 321                 len = readv(rxq->fd, *rxq->iovecs,
 322                             1 +
 323                             (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
 324                              rxq->nb_rx_desc : 1));
 325                 if (len < (int)sizeof(struct tun_pi))
 326                         break;
 327
 328                 /* Packet couldn't fit in the provided mbuf */
 329                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 330                         rxq->stats.ierrors++;
 331                         continue;
 332                 }
 333
 334                 len -= sizeof(struct tun_pi);
 335
 336                 mbuf->pkt_len = len;
 337                 mbuf->port = rxq->in_port;
 338                 while (1) {
 339                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 340
 341                         if (unlikely(!buf)) {
 342                                 rxq->stats.rx_nombuf++;
 343                                 /* No new buf has been allocated: do nothing */
 344                                 if (!new_tail || !seg)
 345                                         goto end;
 346
 347                                 seg->next = NULL;
 348                                 rte_pktmbuf_free(mbuf);
 349
 350                                 goto end;
 351                         }
 352                         seg = seg ? seg->next : mbuf;
 353                         if (rxq->pool == mbuf)
 354                                 rxq->pool = buf;
 355                         if (new_tail)
 356                                 new_tail->next = buf;
 357                         new_tail = buf;
 358                         new_tail->next = seg->next;
 359
 360                         /* iovecs[0] is reserved for packet info (pi) */
 361                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 362                                 buf->buf_len - data_off;
 363                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 364                                 (char *)buf->buf_addr + data_off;
 365
 366                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 367                         seg->data_off = data_off;
 368
 369                         len -= seg->data_len;
 370                         if (len <= 0)
 371                                 break;
 372                         mbuf->nb_segs++;
 373                         /* First segment has headroom, not the others */
 374                         data_off = 0;
 375                 }
 376                 seg->next = NULL;
 377                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 378                                                       RTE_PTYPE_ALL_MASK);
 379                 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
 380                         tap_verify_csum(mbuf);
 381
 382                 /* account for the receive frame */
 383                 bufs[num_rx++] = mbuf;
 384                 num_rx_bytes += mbuf->pkt_len;
 385         }
 386 end:
 387         rxq->stats.ipackets += num_rx;
 388         rxq->stats.ibytes += num_rx_bytes;
 389
 390         return num_rx;
 391 }
 392
 393 static uint64_t
 394 tap_tx_offload_get_port_capa(void)
 395 {
 396         /*
 397          * In order to support legacy apps,
 398          * report capabilities also as port capabilities.
 399          */
 400         return DEV_TX_OFFLOAD_MULTI_SEGS |
 401                DEV_TX_OFFLOAD_IPV4_CKSUM |
 402                DEV_TX_OFFLOAD_UDP_CKSUM |
 403                DEV_TX_OFFLOAD_TCP_CKSUM;
 404 }
 405
 406 static uint64_t
 407 tap_tx_offload_get_queue_capa(void)
 408 {
 409         return DEV_TX_OFFLOAD_MULTI_SEGS |
 410                DEV_TX_OFFLOAD_IPV4_CKSUM |
 411                DEV_TX_OFFLOAD_UDP_CKSUM |
 412                DEV_TX_OFFLOAD_TCP_CKSUM;
 413 }
 414
 415 static bool
 416 tap_txq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 417 {
 418         uint64_t port_offloads = dev->data->dev_conf.txmode.offloads;
 419         uint64_t queue_supp_offloads = tap_tx_offload_get_queue_capa();
 420         uint64_t port_supp_offloads = tap_tx_offload_get_port_capa();
 421
 422         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 423             offloads)
 424                 return false;
 425         /* Verify we have no conflict with port offloads */
 426         if ((port_offloads ^ offloads) & port_supp_offloads)
 427                 return false;
 428         return true;
 429 }
 430
 431 static void
 432 tap_tx_offload(char *packet, uint64_t ol_flags, unsigned int l2_len,
 433                unsigned int l3_len)
 434 {
 435         void *l3_hdr = packet + l2_len;
 436
 437         if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
 438                 struct ipv4_hdr *iph = l3_hdr;
 439                 uint16_t cksum;
 440
 441                 iph->hdr_checksum = 0;
 442                 cksum = rte_raw_cksum(iph, l3_len);
 443                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 444         }
 445         if (ol_flags & PKT_TX_L4_MASK) {
 446                 uint16_t l4_len;
 447                 uint32_t cksum;
 448                 uint16_t *l4_cksum;
 449                 void *l4_hdr;
 450
 451                 l4_hdr = packet + l2_len + l3_len;
 452                 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
 453                         l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
 454                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 455                         l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
 456                 else
 457                         return;
 458                 *l4_cksum = 0;
 459                 if (ol_flags & PKT_TX_IPV4) {
 460                         struct ipv4_hdr *iph = l3_hdr;
 461
 462                         l4_len = rte_be_to_cpu_16(iph->total_length) - l3_len;
 463                         cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 464                 } else {
 465                         struct ipv6_hdr *ip6h = l3_hdr;
 466
 467                         /* payload_len does not include ext headers */
 468                         l4_len = rte_be_to_cpu_16(ip6h->payload_len) -
 469                                 l3_len + sizeof(struct ipv6_hdr);
 470                         cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 471                 }
 472                 cksum += rte_raw_cksum(l4_hdr, l4_len);
 473                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 474                 cksum = (~cksum) & 0xffff;
 475                 if (cksum == 0)
 476                         cksum = 0xffff;
 477                 *l4_cksum = cksum;
 478         }
 479 }
 480
 481 /* Callback to handle sending packets from the tap interface
 482  */
 483 static uint16_t
 484 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 485 {
 486         struct tx_queue *txq = queue;
 487         uint16_t num_tx = 0;
 488         unsigned long num_tx_bytes = 0;
 489         uint32_t max_size;
 490         int i;
 491
 492         if (unlikely(nb_pkts == 0))
 493                 return 0;
 494
 495         max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
 496         for (i = 0; i < nb_pkts; i++) {
 497                 struct rte_mbuf *mbuf = bufs[num_tx];
 498                 struct iovec iovecs[mbuf->nb_segs + 1];
 499                 struct tun_pi pi = { .flags = 0 };
 500                 struct rte_mbuf *seg = mbuf;
 501                 char m_copy[mbuf->data_len];
 502                 int n;
 503                 int j;
 504
 505                 /* stats.errs will be incremented */
 506                 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
 507                         break;
 508
 509                 iovecs[0].iov_base = &pi;
 510                 iovecs[0].iov_len = sizeof(pi);
 511                 for (j = 1; j <= mbuf->nb_segs; j++) {
 512                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 513                         iovecs[j].iov_base =
 514                                 rte_pktmbuf_mtod(seg, void *);
 515                         seg = seg->next;
 516                 }
 517                 if (txq->csum &&
 518                     ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 519                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 520                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
 521                         /* Support only packets with all data in the same seg */
 522                         if (mbuf->nb_segs > 1)
 523                                 break;
 524                         /* To change checksums, work on a copy of data. */
 525                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 526                                    rte_pktmbuf_data_len(mbuf));
 527                         tap_tx_offload(m_copy, mbuf->ol_flags,
 528                                        mbuf->l2_len, mbuf->l3_len);
 529                         iovecs[1].iov_base = m_copy;
 530                 }
 531                 /* copy the tx frame data */
 532                 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
 533                 if (n <= 0)
 534                         break;
 535
 536                 num_tx++;
 537                 num_tx_bytes += mbuf->pkt_len;
 538                 rte_pktmbuf_free(mbuf);
 539         }
 540
 541         txq->stats.opackets += num_tx;
 542         txq->stats.errs += nb_pkts - num_tx;
 543         txq->stats.obytes += num_tx_bytes;
 544
 545         return num_tx;
 546 }
 547
 548 static const char *
 549 tap_ioctl_req2str(unsigned long request)
 550 {
 551         switch (request) {
 552         case SIOCSIFFLAGS:
 553                 return "SIOCSIFFLAGS";
 554         case SIOCGIFFLAGS:
 555                 return "SIOCGIFFLAGS";
 556         case SIOCGIFHWADDR:
 557                 return "SIOCGIFHWADDR";
 558         case SIOCSIFHWADDR:
 559                 return "SIOCSIFHWADDR";
 560         case SIOCSIFMTU:
 561                 return "SIOCSIFMTU";
 562         }
 563         return "UNKNOWN";
 564 }
 565
 566 static int
 567 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 568           struct ifreq *ifr, int set, enum ioctl_mode mode)
 569 {
 570         short req_flags = ifr->ifr_flags;
 571         int remote = pmd->remote_if_index &&
 572                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 573
 574         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 575                 return 0;
 576         /*
 577          * If there is a remote netdevice, apply ioctl on it, then apply it on
 578          * the tap netdevice.
 579          */
 580 apply:
 581         if (remote)
 582                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
 583         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 584                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 585         switch (request) {
 586         case SIOCSIFFLAGS:
 587                 /* fetch current flags to leave other flags untouched */
 588                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 589                         goto error;
 590                 if (set)
 591                         ifr->ifr_flags |= req_flags;
 592                 else
 593                         ifr->ifr_flags &= ~req_flags;
 594                 break;
 595         case SIOCGIFFLAGS:
 596         case SIOCGIFHWADDR:
 597         case SIOCSIFHWADDR:
 598         case SIOCSIFMTU:
 599                 break;
 600         default:
 601                 RTE_ASSERT(!"unsupported request type: must not happen");
 602         }
 603         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 604                 goto error;
 605         if (remote-- && mode == LOCAL_AND_REMOTE)
 606                 goto apply;
 607         return 0;
 608
 609 error:
 610         RTE_LOG(DEBUG, PMD, "%s: %s(%s) failed: %s(%d)\n", ifr->ifr_name,
 611                 __func__, tap_ioctl_req2str(request), strerror(errno), errno);
 612         return -errno;
 613 }
 614
 615 static int
 616 tap_link_set_down(struct rte_eth_dev *dev)
 617 {
 618         struct pmd_internals *pmd = dev->data->dev_private;
 619         struct ifreq ifr = { .ifr_flags = IFF_UP };
 620
 621         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 622         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 623 }
 624
 625 static int
 626 tap_link_set_up(struct rte_eth_dev *dev)
 627 {
 628         struct pmd_internals *pmd = dev->data->dev_private;
 629         struct ifreq ifr = { .ifr_flags = IFF_UP };
 630
 631         dev->data->dev_link.link_status = ETH_LINK_UP;
 632         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 633 }
 634
 635 static int
 636 tap_dev_start(struct rte_eth_dev *dev)
 637 {
 638         int err;
 639
 640         err = tap_intr_handle_set(dev, 1);
 641         if (err)
 642                 return err;
 643         return tap_link_set_up(dev);
 644 }
 645
 646 /* This function gets called when the current port gets stopped.
 647  */
 648 static void
 649 tap_dev_stop(struct rte_eth_dev *dev)
 650 {
 651         tap_intr_handle_set(dev, 0);
 652         tap_link_set_down(dev);
 653 }
 654
 655 static int
 656 tap_dev_configure(struct rte_eth_dev *dev)
 657 {
 658         uint64_t supp_tx_offloads = tap_tx_offload_get_port_capa();
 659         uint64_t tx_offloads = dev->data->dev_conf.txmode.offloads;
 660
 661         if ((tx_offloads & supp_tx_offloads) != tx_offloads) {
 662                 rte_errno = ENOTSUP;
 663                 RTE_LOG(ERR, PMD,
 664                         "Some Tx offloads are not supported "
 665                         "requested 0x%" PRIx64 " supported 0x%" PRIx64 "\n",
 666                         tx_offloads, supp_tx_offloads);
 667                 return -rte_errno;
 668         }
 669         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 670                 RTE_LOG(ERR, PMD,
 671                         "%s: number of rx queues %d exceeds max num of queues %d\n",
 672                         dev->device->name,
 673                         dev->data->nb_rx_queues,
 674                         RTE_PMD_TAP_MAX_QUEUES);
 675                 return -1;
 676         }
 677         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 678                 RTE_LOG(ERR, PMD,
 679                         "%s: number of tx queues %d exceeds max num of queues %d\n",
 680                         dev->device->name,
 681                         dev->data->nb_tx_queues,
 682                         RTE_PMD_TAP_MAX_QUEUES);
 683                 return -1;
 684         }
 685
 686         RTE_LOG(INFO, PMD, "%s: %p: TX configured queues number: %u\n",
 687              dev->device->name, (void *)dev, dev->data->nb_tx_queues);
 688
 689         RTE_LOG(INFO, PMD, "%s: %p: RX configured queues number: %u\n",
 690              dev->device->name, (void *)dev, dev->data->nb_rx_queues);
 691
 692         return 0;
 693 }
 694
 695 static uint32_t
 696 tap_dev_speed_capa(void)
 697 {
 698         uint32_t speed = pmd_link.link_speed;
 699         uint32_t capa = 0;
 700
 701         if (speed >= ETH_SPEED_NUM_10M)
 702                 capa |= ETH_LINK_SPEED_10M;
 703         if (speed >= ETH_SPEED_NUM_100M)
 704                 capa |= ETH_LINK_SPEED_100M;
 705         if (speed >= ETH_SPEED_NUM_1G)
 706                 capa |= ETH_LINK_SPEED_1G;
 707         if (speed >= ETH_SPEED_NUM_5G)
 708                 capa |= ETH_LINK_SPEED_2_5G;
 709         if (speed >= ETH_SPEED_NUM_5G)
 710                 capa |= ETH_LINK_SPEED_5G;
 711         if (speed >= ETH_SPEED_NUM_10G)
 712                 capa |= ETH_LINK_SPEED_10G;
 713         if (speed >= ETH_SPEED_NUM_20G)
 714                 capa |= ETH_LINK_SPEED_20G;
 715         if (speed >= ETH_SPEED_NUM_25G)
 716                 capa |= ETH_LINK_SPEED_25G;
 717         if (speed >= ETH_SPEED_NUM_40G)
 718                 capa |= ETH_LINK_SPEED_40G;
 719         if (speed >= ETH_SPEED_NUM_50G)
 720                 capa |= ETH_LINK_SPEED_50G;
 721         if (speed >= ETH_SPEED_NUM_56G)
 722                 capa |= ETH_LINK_SPEED_56G;
 723         if (speed >= ETH_SPEED_NUM_100G)
 724                 capa |= ETH_LINK_SPEED_100G;
 725
 726         return capa;
 727 }
 728
 729 static void
 730 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 731 {
 732         struct pmd_internals *internals = dev->data->dev_private;
 733
 734         dev_info->if_index = internals->if_index;
 735         dev_info->max_mac_addrs = 1;
 736         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 737         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
 738         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
 739         dev_info->min_rx_bufsize = 0;
 740         dev_info->pci_dev = NULL;
 741         dev_info->speed_capa = tap_dev_speed_capa();
 742         dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
 743         dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
 744                                     dev_info->rx_queue_offload_capa;
 745         dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
 746         dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
 747                                     dev_info->tx_queue_offload_capa;
 748 }
 749
 750 static int
 751 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 752 {
 753         unsigned int i, imax;
 754         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 755         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 756         unsigned long rx_nombuf = 0, ierrors = 0;
 757         const struct pmd_internals *pmd = dev->data->dev_private;
 758
 759         /* rx queue statistics */
 760         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 761                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 762         for (i = 0; i < imax; i++) {
 763                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 764                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 765                 rx_total += tap_stats->q_ipackets[i];
 766                 rx_bytes_total += tap_stats->q_ibytes[i];
 767                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
 768                 ierrors += pmd->rxq[i].stats.ierrors;
 769         }
 770
 771         /* tx queue statistics */
 772         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 773                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 774
 775         for (i = 0; i < imax; i++) {
 776                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 777                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 778                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 779                 tx_total += tap_stats->q_opackets[i];
 780                 tx_err_total += tap_stats->q_errors[i];
 781                 tx_bytes_total += tap_stats->q_obytes[i];
 782         }
 783
 784         tap_stats->ipackets = rx_total;
 785         tap_stats->ibytes = rx_bytes_total;
 786         tap_stats->ierrors = ierrors;
 787         tap_stats->rx_nombuf = rx_nombuf;
 788         tap_stats->opackets = tx_total;
 789         tap_stats->oerrors = tx_err_total;
 790         tap_stats->obytes = tx_bytes_total;
 791         return 0;
 792 }
 793
 794 static void
 795 tap_stats_reset(struct rte_eth_dev *dev)
 796 {
 797         int i;
 798         struct pmd_internals *pmd = dev->data->dev_private;
 799
 800         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 801                 pmd->rxq[i].stats.ipackets = 0;
 802                 pmd->rxq[i].stats.ibytes = 0;
 803                 pmd->rxq[i].stats.ierrors = 0;
 804                 pmd->rxq[i].stats.rx_nombuf = 0;
 805
 806                 pmd->txq[i].stats.opackets = 0;
 807                 pmd->txq[i].stats.errs = 0;
 808                 pmd->txq[i].stats.obytes = 0;
 809         }
 810 }
 811
 812 static void
 813 tap_dev_close(struct rte_eth_dev *dev)
 814 {
 815         int i;
 816         struct pmd_internals *internals = dev->data->dev_private;
 817
 818         tap_link_set_down(dev);
 819         tap_flow_flush(dev, NULL);
 820         tap_flow_implicit_flush(internals, NULL);
 821
 822         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 823                 if (internals->rxq[i].fd != -1) {
 824                         close(internals->rxq[i].fd);
 825                         internals->rxq[i].fd = -1;
 826                 }
 827                 if (internals->txq[i].fd != -1) {
 828                         close(internals->txq[i].fd);
 829                         internals->txq[i].fd = -1;
 830                 }
 831         }
 832
 833         if (internals->remote_if_index) {
 834                 /* Restore initial remote state */
 835                 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
 836                                 &internals->remote_initial_flags);
 837         }
 838 }
 839
 840 static void
 841 tap_rx_queue_release(void *queue)
 842 {
 843         struct rx_queue *rxq = queue;
 844
 845         if (rxq && (rxq->fd > 0)) {
 846                 close(rxq->fd);
 847                 rxq->fd = -1;
 848                 rte_pktmbuf_free(rxq->pool);
 849                 rte_free(rxq->iovecs);
 850                 rxq->pool = NULL;
 851                 rxq->iovecs = NULL;
 852         }
 853 }
 854
 855 static void
 856 tap_tx_queue_release(void *queue)
 857 {
 858         struct tx_queue *txq = queue;
 859
 860         if (txq && (txq->fd > 0)) {
 861                 close(txq->fd);
 862                 txq->fd = -1;
 863         }
 864 }
 865
 866 static int
 867 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
 868 {
 869         struct rte_eth_link *dev_link = &dev->data->dev_link;
 870         struct pmd_internals *pmd = dev->data->dev_private;
 871         struct ifreq ifr = { .ifr_flags = 0 };
 872
 873         if (pmd->remote_if_index) {
 874                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
 875                 if (!(ifr.ifr_flags & IFF_UP) ||
 876                     !(ifr.ifr_flags & IFF_RUNNING)) {
 877                         dev_link->link_status = ETH_LINK_DOWN;
 878                         return 0;
 879                 }
 880         }
 881         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
 882         dev_link->link_status =
 883                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
 884                  ETH_LINK_UP :
 885                  ETH_LINK_DOWN);
 886         return 0;
 887 }
 888
 889 static void
 890 tap_promisc_enable(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 = 1;
 896         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 897         if (pmd->remote_if_index && !pmd->flow_isolate)
 898                 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
 899 }
 900
 901 static void
 902 tap_promisc_disable(struct rte_eth_dev *dev)
 903 {
 904         struct pmd_internals *pmd = dev->data->dev_private;
 905         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 906
 907         dev->data->promiscuous = 0;
 908         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 909         if (pmd->remote_if_index && !pmd->flow_isolate)
 910                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
 911 }
 912
 913 static void
 914 tap_allmulti_enable(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 = 1;
 920         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 921         if (pmd->remote_if_index && !pmd->flow_isolate)
 922                 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
 923 }
 924
 925 static void
 926 tap_allmulti_disable(struct rte_eth_dev *dev)
 927 {
 928         struct pmd_internals *pmd = dev->data->dev_private;
 929         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 930
 931         dev->data->all_multicast = 0;
 932         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 933         if (pmd->remote_if_index && !pmd->flow_isolate)
 934                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
 935 }
 936
 937 static void
 938 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 939 {
 940         struct pmd_internals *pmd = dev->data->dev_private;
 941         enum ioctl_mode mode = LOCAL_ONLY;
 942         struct ifreq ifr;
 943
 944         if (is_zero_ether_addr(mac_addr)) {
 945                 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
 946                         dev->device->name);
 947                 return;
 948         }
 949         /* Check the actual current MAC address on the tap netdevice */
 950         if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
 951                 return;
 952         if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 953                                mac_addr))
 954                 return;
 955         /* Check the current MAC address on the remote */
 956         if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0)
 957                 return;
 958         if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 959                                mac_addr))
 960                 mode = LOCAL_AND_REMOTE;
 961         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
 962         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
 963         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode) < 0)
 964                 return;
 965         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
 966         if (pmd->remote_if_index && !pmd->flow_isolate) {
 967                 /* Replace MAC redirection rule after a MAC change */
 968                 if (tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC) < 0) {
 969                         RTE_LOG(ERR, PMD,
 970                                 "%s: Couldn't delete MAC redirection rule\n",
 971                                 dev->device->name);
 972                         return;
 973                 }
 974                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0)
 975                         RTE_LOG(ERR, PMD,
 976                                 "%s: Couldn't add MAC redirection rule\n",
 977                                 dev->device->name);
 978         }
 979 }
 980
 981 static int
 982 tap_setup_queue(struct rte_eth_dev *dev,
 983                 struct pmd_internals *internals,
 984                 uint16_t qid,
 985                 int is_rx)
 986 {
 987         int *fd;
 988         int *other_fd;
 989         const char *dir;
 990         struct pmd_internals *pmd = dev->data->dev_private;
 991         struct rx_queue *rx = &internals->rxq[qid];
 992         struct tx_queue *tx = &internals->txq[qid];
 993
 994         if (is_rx) {
 995                 fd = &rx->fd;
 996                 other_fd = &tx->fd;
 997                 dir = "rx";
 998         } else {
 999                 fd = &tx->fd;
1000                 other_fd = &rx->fd;
1001                 dir = "tx";
1002         }
1003         if (*fd != -1) {
1004                 /* fd for this queue already exists */
1005                 RTE_LOG(DEBUG, PMD, "%s: fd %d for %s queue qid %d exists\n",
1006                         pmd->name, *fd, dir, qid);
1007         } else if (*other_fd != -1) {
1008                 /* Only other_fd exists. dup it */
1009                 *fd = dup(*other_fd);
1010                 if (*fd < 0) {
1011                         *fd = -1;
1012                         RTE_LOG(ERR, PMD, "%s: dup() failed.\n",
1013                                 pmd->name);
1014                         return -1;
1015                 }
1016                 RTE_LOG(DEBUG, PMD, "%s: dup fd %d for %s queue qid %d (%d)\n",
1017                         pmd->name, *other_fd, dir, qid, *fd);
1018         } else {
1019                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1020                 *fd = tun_alloc(pmd);
1021                 if (*fd < 0) {
1022                         *fd = -1; /* restore original value */
1023                         RTE_LOG(ERR, PMD, "%s: tun_alloc() failed.\n",
1024                                 pmd->name);
1025                         return -1;
1026                 }
1027                 RTE_LOG(DEBUG, PMD, "%s: add %s queue for qid %d fd %d\n",
1028                         pmd->name, dir, qid, *fd);
1029         }
1030
1031         tx->mtu = &dev->data->mtu;
1032         rx->rxmode = &dev->data->dev_conf.rxmode;
1033
1034         return *fd;
1035 }
1036
1037 static int
1038 tap_rx_queue_setup(struct rte_eth_dev *dev,
1039                    uint16_t rx_queue_id,
1040                    uint16_t nb_rx_desc,
1041                    unsigned int socket_id,
1042                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1043                    struct rte_mempool *mp)
1044 {
1045         struct pmd_internals *internals = dev->data->dev_private;
1046         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1047         struct rte_mbuf **tmp = &rxq->pool;
1048         long iov_max = sysconf(_SC_IOV_MAX);
1049         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1050         struct iovec (*iovecs)[nb_desc + 1];
1051         int data_off = RTE_PKTMBUF_HEADROOM;
1052         int ret = 0;
1053         int fd;
1054         int i;
1055
1056         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1057                 RTE_LOG(WARNING, PMD,
1058                         "nb_rx_queues %d too small or mempool NULL\n",
1059                         dev->data->nb_rx_queues);
1060                 return -1;
1061         }
1062
1063         /* Verify application offloads are valid for our port and queue. */
1064         if (!tap_rxq_are_offloads_valid(dev, rx_conf->offloads)) {
1065                 rte_errno = ENOTSUP;
1066                 RTE_LOG(ERR, PMD,
1067                         "%p: Rx queue offloads 0x%" PRIx64
1068                         " don't match port offloads 0x%" PRIx64
1069                         " or supported offloads 0x%" PRIx64 "\n",
1070                         (void *)dev, rx_conf->offloads,
1071                         dev->data->dev_conf.rxmode.offloads,
1072                         (tap_rx_offload_get_port_capa() |
1073                          tap_rx_offload_get_queue_capa()));
1074                 return -rte_errno;
1075         }
1076         rxq->mp = mp;
1077         rxq->trigger_seen = 1; /* force initial burst */
1078         rxq->in_port = dev->data->port_id;
1079         rxq->nb_rx_desc = nb_desc;
1080         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1081                                     socket_id);
1082         if (!iovecs) {
1083                 RTE_LOG(WARNING, PMD,
1084                         "%s: Couldn't allocate %d RX descriptors\n",
1085                         dev->device->name, nb_desc);
1086                 return -ENOMEM;
1087         }
1088         rxq->iovecs = iovecs;
1089
1090         dev->data->rx_queues[rx_queue_id] = rxq;
1091         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1092         if (fd == -1) {
1093                 ret = fd;
1094                 goto error;
1095         }
1096
1097         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1098         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1099
1100         for (i = 1; i <= nb_desc; i++) {
1101                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1102                 if (!*tmp) {
1103                         RTE_LOG(WARNING, PMD,
1104                                 "%s: couldn't allocate memory for queue %d\n",
1105                                 dev->device->name, rx_queue_id);
1106                         ret = -ENOMEM;
1107                         goto error;
1108                 }
1109                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1110                 (*rxq->iovecs)[i].iov_base =
1111                         (char *)(*tmp)->buf_addr + data_off;
1112                 data_off = 0;
1113                 tmp = &(*tmp)->next;
1114         }
1115
1116         RTE_LOG(DEBUG, PMD, "  RX TAP device name %s, qid %d on fd %d\n",
1117                 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
1118
1119         return 0;
1120
1121 error:
1122         rte_pktmbuf_free(rxq->pool);
1123         rxq->pool = NULL;
1124         rte_free(rxq->iovecs);
1125         rxq->iovecs = NULL;
1126         return ret;
1127 }
1128
1129 static int
1130 tap_tx_queue_setup(struct rte_eth_dev *dev,
1131                    uint16_t tx_queue_id,
1132                    uint16_t nb_tx_desc __rte_unused,
1133                    unsigned int socket_id __rte_unused,
1134                    const struct rte_eth_txconf *tx_conf)
1135 {
1136         struct pmd_internals *internals = dev->data->dev_private;
1137         struct tx_queue *txq;
1138         int ret;
1139
1140         if (tx_queue_id >= dev->data->nb_tx_queues)
1141                 return -1;
1142         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1143         txq = dev->data->tx_queues[tx_queue_id];
1144         /*
1145          * Don't verify port offloads for application which
1146          * use the old API.
1147          */
1148         if (tx_conf != NULL &&
1149             !!(tx_conf->txq_flags & ETH_TXQ_FLAGS_IGNORE)) {
1150                 if (tap_txq_are_offloads_valid(dev, tx_conf->offloads)) {
1151                         txq->csum = !!(tx_conf->offloads &
1152                                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
1153                                          DEV_TX_OFFLOAD_UDP_CKSUM |
1154                                          DEV_TX_OFFLOAD_TCP_CKSUM));
1155                 } else {
1156                         rte_errno = ENOTSUP;
1157                         RTE_LOG(ERR, PMD,
1158                                 "%p: Tx queue offloads 0x%" PRIx64
1159                                 " don't match port offloads 0x%" PRIx64
1160                                 " or supported offloads 0x%" PRIx64,
1161                                 (void *)dev, tx_conf->offloads,
1162                                 dev->data->dev_conf.txmode.offloads,
1163                                 tap_tx_offload_get_port_capa());
1164                         return -rte_errno;
1165                 }
1166         }
1167         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1168         if (ret == -1)
1169                 return -1;
1170         RTE_LOG(DEBUG, PMD,
1171                 "  TX TAP device name %s, qid %d on fd %d csum %s\n",
1172                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd,
1173                 txq->csum ? "on" : "off");
1174
1175         return 0;
1176 }
1177
1178 static int
1179 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1180 {
1181         struct pmd_internals *pmd = dev->data->dev_private;
1182         struct ifreq ifr = { .ifr_mtu = mtu };
1183         int err = 0;
1184
1185         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1186         if (!err)
1187                 dev->data->mtu = mtu;
1188
1189         return err;
1190 }
1191
1192 static int
1193 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1194                      struct ether_addr *mc_addr_set __rte_unused,
1195                      uint32_t nb_mc_addr __rte_unused)
1196 {
1197         /*
1198          * Nothing to do actually: the tap has no filtering whatsoever, every
1199          * packet is received.
1200          */
1201         return 0;
1202 }
1203
1204 static int
1205 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1206 {
1207         struct rte_eth_dev *dev = arg;
1208         struct pmd_internals *pmd = dev->data->dev_private;
1209         struct ifinfomsg *info = NLMSG_DATA(nh);
1210
1211         if (nh->nlmsg_type != RTM_NEWLINK ||
1212             (info->ifi_index != pmd->if_index &&
1213              info->ifi_index != pmd->remote_if_index))
1214                 return 0;
1215         return tap_link_update(dev, 0);
1216 }
1217
1218 static void
1219 tap_dev_intr_handler(void *cb_arg)
1220 {
1221         struct rte_eth_dev *dev = cb_arg;
1222         struct pmd_internals *pmd = dev->data->dev_private;
1223
1224         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1225 }
1226
1227 static int
1228 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1229 {
1230         struct pmd_internals *pmd = dev->data->dev_private;
1231
1232         /* In any case, disable interrupt if the conf is no longer there. */
1233         if (!dev->data->dev_conf.intr_conf.lsc) {
1234                 if (pmd->intr_handle.fd != -1) {
1235                         tap_nl_final(pmd->intr_handle.fd);
1236                         rte_intr_callback_unregister(&pmd->intr_handle,
1237                                 tap_dev_intr_handler, dev);
1238                 }
1239                 return 0;
1240         }
1241         if (set) {
1242                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1243                 if (unlikely(pmd->intr_handle.fd == -1))
1244                         return -EBADF;
1245                 return rte_intr_callback_register(
1246                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1247         }
1248         tap_nl_final(pmd->intr_handle.fd);
1249         return rte_intr_callback_unregister(&pmd->intr_handle,
1250                                             tap_dev_intr_handler, dev);
1251 }
1252
1253 static int
1254 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1255 {
1256         int err;
1257
1258         err = tap_lsc_intr_handle_set(dev, set);
1259         if (err)
1260                 return err;
1261         err = tap_rx_intr_vec_set(dev, set);
1262         if (err && set)
1263                 tap_lsc_intr_handle_set(dev, 0);
1264         return err;
1265 }
1266
1267 static const uint32_t*
1268 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1269 {
1270         static const uint32_t ptypes[] = {
1271                 RTE_PTYPE_INNER_L2_ETHER,
1272                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1273                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1274                 RTE_PTYPE_INNER_L3_IPV4,
1275                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1276                 RTE_PTYPE_INNER_L3_IPV6,
1277                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1278                 RTE_PTYPE_INNER_L4_FRAG,
1279                 RTE_PTYPE_INNER_L4_UDP,
1280                 RTE_PTYPE_INNER_L4_TCP,
1281                 RTE_PTYPE_INNER_L4_SCTP,
1282                 RTE_PTYPE_L2_ETHER,
1283                 RTE_PTYPE_L2_ETHER_VLAN,
1284                 RTE_PTYPE_L2_ETHER_QINQ,
1285                 RTE_PTYPE_L3_IPV4,
1286                 RTE_PTYPE_L3_IPV4_EXT,
1287                 RTE_PTYPE_L3_IPV6_EXT,
1288                 RTE_PTYPE_L3_IPV6,
1289                 RTE_PTYPE_L4_FRAG,
1290                 RTE_PTYPE_L4_UDP,
1291                 RTE_PTYPE_L4_TCP,
1292                 RTE_PTYPE_L4_SCTP,
1293         };
1294
1295         return ptypes;
1296 }
1297
1298 static int
1299 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1300                   struct rte_eth_fc_conf *fc_conf)
1301 {
1302         fc_conf->mode = RTE_FC_NONE;
1303         return 0;
1304 }
1305
1306 static int
1307 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1308                   struct rte_eth_fc_conf *fc_conf)
1309 {
1310         if (fc_conf->mode != RTE_FC_NONE)
1311                 return -ENOTSUP;
1312         return 0;
1313 }
1314
1315 static const struct eth_dev_ops ops = {
1316         .dev_start              = tap_dev_start,
1317         .dev_stop               = tap_dev_stop,
1318         .dev_close              = tap_dev_close,
1319         .dev_configure          = tap_dev_configure,
1320         .dev_infos_get          = tap_dev_info,
1321         .rx_queue_setup         = tap_rx_queue_setup,
1322         .tx_queue_setup         = tap_tx_queue_setup,
1323         .rx_queue_release       = tap_rx_queue_release,
1324         .tx_queue_release       = tap_tx_queue_release,
1325         .flow_ctrl_get          = tap_flow_ctrl_get,
1326         .flow_ctrl_set          = tap_flow_ctrl_set,
1327         .link_update            = tap_link_update,
1328         .dev_set_link_up        = tap_link_set_up,
1329         .dev_set_link_down      = tap_link_set_down,
1330         .promiscuous_enable     = tap_promisc_enable,
1331         .promiscuous_disable    = tap_promisc_disable,
1332         .allmulticast_enable    = tap_allmulti_enable,
1333         .allmulticast_disable   = tap_allmulti_disable,
1334         .mac_addr_set           = tap_mac_set,
1335         .mtu_set                = tap_mtu_set,
1336         .set_mc_addr_list       = tap_set_mc_addr_list,
1337         .stats_get              = tap_stats_get,
1338         .stats_reset            = tap_stats_reset,
1339         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1340         .filter_ctrl            = tap_dev_filter_ctrl,
1341 };
1342
1343 static int
1344 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1345                    char *remote_iface, struct ether_addr *mac_addr)
1346 {
1347         int numa_node = rte_socket_id();
1348         struct rte_eth_dev *dev;
1349         struct pmd_internals *pmd;
1350         struct rte_eth_dev_data *data;
1351         struct ifreq ifr;
1352         int i;
1353
1354         RTE_LOG(DEBUG, PMD, "  TAP device on numa %u\n", rte_socket_id());
1355
1356         data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
1357         if (!data) {
1358                 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
1359                 goto error_exit_nodev;
1360         }
1361
1362         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1363         if (!dev) {
1364                 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
1365                 goto error_exit_nodev;
1366         }
1367
1368         pmd = dev->data->dev_private;
1369         pmd->dev = dev;
1370         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1371
1372         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1373         if (pmd->ioctl_sock == -1) {
1374                 RTE_LOG(ERR, PMD,
1375                         "TAP Unable to get a socket for management: %s\n",
1376                         strerror(errno));
1377                 goto error_exit;
1378         }
1379
1380         /* Setup some default values */
1381         rte_memcpy(data, dev->data, sizeof(*data));
1382         data->dev_private = pmd;
1383         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1384         data->numa_node = numa_node;
1385
1386         data->dev_link = pmd_link;
1387         data->mac_addrs = &pmd->eth_addr;
1388         /* Set the number of RX and TX queues */
1389         data->nb_rx_queues = 0;
1390         data->nb_tx_queues = 0;
1391
1392         dev->data = data;
1393         dev->dev_ops = &ops;
1394         dev->rx_pkt_burst = pmd_rx_burst;
1395         dev->tx_pkt_burst = pmd_tx_burst;
1396
1397         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1398         pmd->intr_handle.fd = -1;
1399         dev->intr_handle = &pmd->intr_handle;
1400
1401         /* Presetup the fds to -1 as being not valid */
1402         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1403                 pmd->rxq[i].fd = -1;
1404                 pmd->txq[i].fd = -1;
1405         }
1406
1407         if (is_zero_ether_addr(mac_addr))
1408                 eth_random_addr((uint8_t *)&pmd->eth_addr);
1409         else
1410                 rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1411
1412         /* Immediately create the netdevice (this will create the 1st queue). */
1413         /* rx queue */
1414         if (tap_setup_queue(dev, pmd, 0, 1) == -1)
1415                 goto error_exit;
1416         /* tx queue */
1417         if (tap_setup_queue(dev, pmd, 0, 0) == -1)
1418                 goto error_exit;
1419
1420         ifr.ifr_mtu = dev->data->mtu;
1421         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1422                 goto error_exit;
1423
1424         memset(&ifr, 0, sizeof(struct ifreq));
1425         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1426         rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr, ETHER_ADDR_LEN);
1427         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1428                 goto error_exit;
1429
1430         /*
1431          * Set up everything related to rte_flow:
1432          * - netlink socket
1433          * - tap / remote if_index
1434          * - mandatory QDISCs
1435          * - rte_flow actual/implicit lists
1436          * - implicit rules
1437          */
1438         pmd->nlsk_fd = tap_nl_init(0);
1439         if (pmd->nlsk_fd == -1) {
1440                 RTE_LOG(WARNING, PMD, "%s: failed to create netlink socket.\n",
1441                         pmd->name);
1442                 goto disable_rte_flow;
1443         }
1444         pmd->if_index = if_nametoindex(pmd->name);
1445         if (!pmd->if_index) {
1446                 RTE_LOG(ERR, PMD, "%s: failed to get if_index.\n", pmd->name);
1447                 goto disable_rte_flow;
1448         }
1449         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1450                 RTE_LOG(ERR, PMD, "%s: failed to create multiq qdisc.\n",
1451                         pmd->name);
1452                 goto disable_rte_flow;
1453         }
1454         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1455                 RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1456                         pmd->name);
1457                 goto disable_rte_flow;
1458         }
1459         LIST_INIT(&pmd->flows);
1460
1461         if (strlen(remote_iface)) {
1462                 pmd->remote_if_index = if_nametoindex(remote_iface);
1463                 if (!pmd->remote_if_index) {
1464                         RTE_LOG(ERR, PMD, "%s: failed to get %s if_index.\n",
1465                                 pmd->name, remote_iface);
1466                         goto error_remote;
1467                 }
1468                 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1469                          "%s", remote_iface);
1470
1471                 /* Save state of remote device */
1472                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1473
1474                 /* Replicate remote MAC address */
1475                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1476                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1477                                 pmd->name, pmd->remote_iface);
1478                         goto error_remote;
1479                 }
1480                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1481                            ETHER_ADDR_LEN);
1482                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1483                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1484                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1485                                 pmd->name, remote_iface);
1486                         goto error_remote;
1487                 }
1488
1489                 /*
1490                  * Flush usually returns negative value because it tries to
1491                  * delete every QDISC (and on a running device, one QDISC at
1492                  * least is needed). Ignore negative return value.
1493                  */
1494                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1495                 if (qdisc_create_ingress(pmd->nlsk_fd,
1496                                          pmd->remote_if_index) < 0) {
1497                         RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1498                                 pmd->remote_iface);
1499                         goto error_remote;
1500                 }
1501                 LIST_INIT(&pmd->implicit_flows);
1502                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1503                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1504                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1505                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1506                         RTE_LOG(ERR, PMD,
1507                                 "%s: failed to create implicit rules.\n",
1508                                 pmd->name);
1509                         goto error_remote;
1510                 }
1511         }
1512
1513         return 0;
1514
1515 disable_rte_flow:
1516         RTE_LOG(ERR, PMD, " Disabling rte flow support: %s(%d)\n",
1517                 strerror(errno), errno);
1518         if (strlen(remote_iface)) {
1519                 RTE_LOG(ERR, PMD, "Remote feature requires flow support.\n");
1520                 goto error_exit;
1521         }
1522         return 0;
1523
1524 error_remote:
1525         RTE_LOG(ERR, PMD, " Can't set up remote feature: %s(%d)\n",
1526                 strerror(errno), errno);
1527         tap_flow_implicit_flush(pmd, NULL);
1528
1529 error_exit:
1530         if (pmd->ioctl_sock > 0)
1531                 close(pmd->ioctl_sock);
1532         rte_eth_dev_release_port(dev);
1533
1534 error_exit_nodev:
1535         RTE_LOG(ERR, PMD, "TAP Unable to initialize %s\n",
1536                 rte_vdev_device_name(vdev));
1537
1538         rte_free(data);
1539         return -EINVAL;
1540 }
1541
1542 static int
1543 set_interface_name(const char *key __rte_unused,
1544                    const char *value,
1545                    void *extra_args)
1546 {
1547         char *name = (char *)extra_args;
1548
1549         if (value)
1550                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN - 1);
1551         else
1552                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1553                          DEFAULT_TAP_NAME, (tap_unit - 1));
1554
1555         return 0;
1556 }
1557
1558 static int
1559 set_remote_iface(const char *key __rte_unused,
1560                  const char *value,
1561                  void *extra_args)
1562 {
1563         char *name = (char *)extra_args;
1564
1565         if (value)
1566                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
1567
1568         return 0;
1569 }
1570
1571 static int parse_user_mac(struct ether_addr *user_mac,
1572                 const char *value)
1573 {
1574         unsigned int index = 0;
1575         char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
1576
1577         if (user_mac == NULL || value == NULL)
1578                 return 0;
1579
1580         strlcpy(mac_temp, value, sizeof(mac_temp));
1581         mac_byte = strtok(mac_temp, ":");
1582
1583         while ((mac_byte != NULL) &&
1584                         (strlen(mac_byte) <= 2) &&
1585                         (strlen(mac_byte) == strspn(mac_byte,
1586                                         ETH_TAP_CMP_MAC_FMT))) {
1587                 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
1588                 mac_byte = strtok(NULL, ":");
1589         }
1590
1591         return index;
1592 }
1593
1594 static int
1595 set_mac_type(const char *key __rte_unused,
1596              const char *value,
1597              void *extra_args)
1598 {
1599         struct ether_addr *user_mac = extra_args;
1600
1601         if (!value)
1602                 return 0;
1603
1604         if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
1605                 static int iface_idx;
1606
1607                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1608                 memcpy((char *)user_mac->addr_bytes, "\0dtap", ETHER_ADDR_LEN);
1609                 user_mac->addr_bytes[ETHER_ADDR_LEN - 1] = iface_idx++ + '0';
1610                 goto success;
1611         }
1612
1613         if (parse_user_mac(user_mac, value) != 6)
1614                 goto error;
1615 success:
1616         RTE_LOG(DEBUG, PMD, "TAP user MAC param (%s)\n", value);
1617         return 0;
1618
1619 error:
1620         RTE_LOG(ERR, PMD, "TAP user MAC (%s) is not in format (%s|%s)\n",
1621                 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
1622         return -1;
1623 }
1624
1625 /* Open a TAP interface device.
1626  */
1627 static int
1628 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1629 {
1630         const char *name, *params;
1631         int ret;
1632         struct rte_kvargs *kvlist = NULL;
1633         int speed;
1634         char tap_name[RTE_ETH_NAME_MAX_LEN];
1635         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1636         struct ether_addr user_mac = { .addr_bytes = {0} };
1637
1638         name = rte_vdev_device_name(dev);
1639         params = rte_vdev_device_args(dev);
1640
1641         speed = ETH_SPEED_NUM_10G;
1642         snprintf(tap_name, sizeof(tap_name), "%s%d",
1643                  DEFAULT_TAP_NAME, tap_unit++);
1644         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1645
1646         if (params && (params[0] != '\0')) {
1647                 RTE_LOG(DEBUG, PMD, "parameters (%s)\n", params);
1648
1649                 kvlist = rte_kvargs_parse(params, valid_arguments);
1650                 if (kvlist) {
1651                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1652                                 ret = rte_kvargs_process(kvlist,
1653                                                          ETH_TAP_IFACE_ARG,
1654                                                          &set_interface_name,
1655                                                          tap_name);
1656                                 if (ret == -1)
1657                                         goto leave;
1658                         }
1659
1660                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1661                                 ret = rte_kvargs_process(kvlist,
1662                                                          ETH_TAP_REMOTE_ARG,
1663                                                          &set_remote_iface,
1664                                                          remote_iface);
1665                                 if (ret == -1)
1666                                         goto leave;
1667                         }
1668
1669                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1670                                 ret = rte_kvargs_process(kvlist,
1671                                                          ETH_TAP_MAC_ARG,
1672                                                          &set_mac_type,
1673                                                          &user_mac);
1674                                 if (ret == -1)
1675                                         goto leave;
1676                         }
1677                 }
1678         }
1679         pmd_link.link_speed = speed;
1680
1681         RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1682                 name, tap_name);
1683
1684         ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac);
1685
1686 leave:
1687         if (ret == -1) {
1688                 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1689                         name, tap_name);
1690                 tap_unit--;             /* Restore the unit number */
1691         }
1692         rte_kvargs_free(kvlist);
1693
1694         return ret;
1695 }
1696
1697 /* detach a TAP device.
1698  */
1699 static int
1700 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1701 {
1702         struct rte_eth_dev *eth_dev = NULL;
1703         struct pmd_internals *internals;
1704         int i;
1705
1706         RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1707                 rte_socket_id());
1708
1709         /* find the ethdev entry */
1710         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1711         if (!eth_dev)
1712                 return 0;
1713
1714         internals = eth_dev->data->dev_private;
1715         if (internals->nlsk_fd) {
1716                 tap_flow_flush(eth_dev, NULL);
1717                 tap_flow_implicit_flush(internals, NULL);
1718                 tap_nl_final(internals->nlsk_fd);
1719         }
1720         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1721                 if (internals->rxq[i].fd != -1) {
1722                         close(internals->rxq[i].fd);
1723                         internals->rxq[i].fd = -1;
1724                 }
1725                 if (internals->txq[i].fd != -1) {
1726                         close(internals->txq[i].fd);
1727                         internals->txq[i].fd = -1;
1728                 }
1729         }
1730
1731         close(internals->ioctl_sock);
1732         rte_free(eth_dev->data->dev_private);
1733         rte_free(eth_dev->data);
1734
1735         rte_eth_dev_release_port(eth_dev);
1736
1737         return 0;
1738 }
1739
1740 static struct rte_vdev_driver pmd_tap_drv = {
1741         .probe = rte_pmd_tap_probe,
1742         .remove = rte_pmd_tap_remove,
1743 };
1744 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1745 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1746 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1747                               ETH_TAP_IFACE_ARG "=<string> "
1748                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
1749                               ETH_TAP_REMOTE_ARG "=<string>");