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 #define DEFAULT_TUN_NAME        "dtun"
  47
  48 #define ETH_TAP_IFACE_ARG       "iface"
  49 #define ETH_TAP_REMOTE_ARG      "remote"
  50 #define ETH_TAP_MAC_ARG         "mac"
  51 #define ETH_TAP_MAC_FIXED       "fixed"
  52
  53 #define ETH_TAP_USR_MAC_FMT     "xx:xx:xx:xx:xx:xx"
  54 #define ETH_TAP_CMP_MAC_FMT     "0123456789ABCDEFabcdef"
  55 #define ETH_TAP_MAC_ARG_FMT     ETH_TAP_MAC_FIXED "|" ETH_TAP_USR_MAC_FMT
  56
  57 static struct rte_vdev_driver pmd_tap_drv;
  58 static struct rte_vdev_driver pmd_tun_drv;
  59
  60 static const char *valid_arguments[] = {
  61         ETH_TAP_IFACE_ARG,
  62         ETH_TAP_REMOTE_ARG,
  63         ETH_TAP_MAC_ARG,
  64         NULL
  65 };
  66
  67 static int tap_unit;
  68 static int tun_unit;
  69
  70 static int tap_type;
  71 static char tuntap_name[8];
  72
  73 static volatile uint32_t tap_trigger;   /* Rx trigger */
  74
  75 static struct rte_eth_link pmd_link = {
  76         .link_speed = ETH_SPEED_NUM_10G,
  77         .link_duplex = ETH_LINK_FULL_DUPLEX,
  78         .link_status = ETH_LINK_DOWN,
  79         .link_autoneg = ETH_LINK_AUTONEG
  80 };
  81
  82 static void
  83 tap_trigger_cb(int sig __rte_unused)
  84 {
  85         /* Valid trigger values are nonzero */
  86         tap_trigger = (tap_trigger + 1) | 0x80000000;
  87 }
  88
  89 /* Specifies on what netdevices the ioctl should be applied */
  90 enum ioctl_mode {
  91         LOCAL_AND_REMOTE,
  92         LOCAL_ONLY,
  93         REMOTE_ONLY,
  94 };
  95
  96 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
  97
  98 /* Tun/Tap allocation routine
  99  *
 100  * name is the number of the interface to use, unless NULL to take the host
 101  * supplied name.
 102  */
 103 static int
 104 tun_alloc(struct pmd_internals *pmd)
 105 {
 106         struct ifreq ifr;
 107 #ifdef IFF_MULTI_QUEUE
 108         unsigned int features;
 109 #endif
 110         int fd;
 111
 112         memset(&ifr, 0, sizeof(struct ifreq));
 113
 114         /*
 115          * Do not set IFF_NO_PI as packet information header will be needed
 116          * to check if a received packet has been truncated.
 117          */
 118         ifr.ifr_flags = (tap_type) ? IFF_TAP : IFF_TUN | IFF_POINTOPOINT;
 119         snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
 120
 121         RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
 122
 123         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 124         if (fd < 0) {
 125                 RTE_LOG(ERR, PMD, "Unable to create %s interface\n",
 126                                 tuntap_name);
 127                 goto error;
 128         }
 129
 130 #ifdef IFF_MULTI_QUEUE
 131         /* Grab the TUN features to verify we can work multi-queue */
 132         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 133                 RTE_LOG(ERR, PMD, "%s unable to get TUN/TAP features\n",
 134                                 tuntap_name);
 135                 goto error;
 136         }
 137         RTE_LOG(DEBUG, PMD, "%s Features %08x\n", tuntap_name, features);
 138
 139         if (features & IFF_MULTI_QUEUE) {
 140                 RTE_LOG(DEBUG, PMD, "  Multi-queue support for %d queues\n",
 141                         RTE_PMD_TAP_MAX_QUEUES);
 142                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 143         } else
 144 #endif
 145         {
 146                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 147                 RTE_LOG(DEBUG, PMD, "  Single queue only support\n");
 148         }
 149
 150         /* Set the TUN/TAP configuration and set the name if needed */
 151         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 152                 RTE_LOG(WARNING, PMD,
 153                         "Unable to set TUNSETIFF for %s\n",
 154                         ifr.ifr_name);
 155                 perror("TUNSETIFF");
 156                 goto error;
 157         }
 158
 159         /* Always set the file descriptor to non-blocking */
 160         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 161                 RTE_LOG(WARNING, PMD,
 162                         "Unable to set %s to nonblocking\n",
 163                         ifr.ifr_name);
 164                 perror("F_SETFL, NONBLOCK");
 165                 goto error;
 166         }
 167
 168         /* Set up trigger to optimize empty Rx bursts */
 169         errno = 0;
 170         do {
 171                 struct sigaction sa;
 172                 int flags = fcntl(fd, F_GETFL);
 173
 174                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 175                         break;
 176                 if (sa.sa_handler != tap_trigger_cb) {
 177                         /*
 178                          * Make sure SIGIO is not already taken. This is done
 179                          * as late as possible to leave the application a
 180                          * chance to set up its own signal handler first.
 181                          */
 182                         if (sa.sa_handler != SIG_IGN &&
 183                             sa.sa_handler != SIG_DFL) {
 184                                 errno = EBUSY;
 185                                 break;
 186                         }
 187                         sa = (struct sigaction){
 188                                 .sa_flags = SA_RESTART,
 189                                 .sa_handler = tap_trigger_cb,
 190                         };
 191                         if (sigaction(SIGIO, &sa, NULL) == -1)
 192                                 break;
 193                 }
 194                 /* Enable SIGIO on file descriptor */
 195                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 196                 fcntl(fd, F_SETOWN, getpid());
 197         } while (0);
 198         if (errno) {
 199                 /* Disable trigger globally in case of error */
 200                 tap_trigger = 0;
 201                 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
 202                         strerror(errno));
 203         }
 204
 205         return fd;
 206
 207 error:
 208         if (fd > 0)
 209                 close(fd);
 210         return -1;
 211 }
 212
 213 static void
 214 tap_verify_csum(struct rte_mbuf *mbuf)
 215 {
 216         uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
 217         uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
 218         uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
 219         unsigned int l2_len = sizeof(struct ether_hdr);
 220         unsigned int l3_len;
 221         uint16_t cksum = 0;
 222         void *l3_hdr;
 223         void *l4_hdr;
 224
 225         if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
 226                 l2_len += 4;
 227         else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
 228                 l2_len += 8;
 229         /* Don't verify checksum for packets with discontinuous L2 header */
 230         if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
 231                      rte_pktmbuf_data_len(mbuf)))
 232                 return;
 233         l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
 234         if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 235                 struct ipv4_hdr *iph = l3_hdr;
 236
 237                 /* ihl contains the number of 4-byte words in the header */
 238                 l3_len = 4 * (iph->version_ihl & 0xf);
 239                 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
 240                         return;
 241
 242                 cksum = ~rte_raw_cksum(iph, l3_len);
 243                 mbuf->ol_flags |= cksum ?
 244                         PKT_RX_IP_CKSUM_BAD :
 245                         PKT_RX_IP_CKSUM_GOOD;
 246         } else if (l3 == RTE_PTYPE_L3_IPV6) {
 247                 l3_len = sizeof(struct ipv6_hdr);
 248         } else {
 249                 /* IPv6 extensions are not supported */
 250                 return;
 251         }
 252         if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
 253                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 254                 /* Don't verify checksum for multi-segment packets. */
 255                 if (mbuf->nb_segs > 1)
 256                         return;
 257                 if (l3 == RTE_PTYPE_L3_IPV4)
 258                         cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
 259                 else if (l3 == RTE_PTYPE_L3_IPV6)
 260                         cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
 261                 mbuf->ol_flags |= cksum ?
 262                         PKT_RX_L4_CKSUM_BAD :
 263                         PKT_RX_L4_CKSUM_GOOD;
 264         }
 265 }
 266
 267 static uint64_t
 268 tap_rx_offload_get_port_capa(void)
 269 {
 270         /*
 271          * In order to support legacy apps,
 272          * report capabilities also as port capabilities.
 273          */
 274         return DEV_RX_OFFLOAD_SCATTER |
 275                DEV_RX_OFFLOAD_IPV4_CKSUM |
 276                DEV_RX_OFFLOAD_UDP_CKSUM |
 277                DEV_RX_OFFLOAD_TCP_CKSUM |
 278                DEV_RX_OFFLOAD_CRC_STRIP;
 279 }
 280
 281 static uint64_t
 282 tap_rx_offload_get_queue_capa(void)
 283 {
 284         return DEV_RX_OFFLOAD_SCATTER |
 285                DEV_RX_OFFLOAD_IPV4_CKSUM |
 286                DEV_RX_OFFLOAD_UDP_CKSUM |
 287                DEV_RX_OFFLOAD_TCP_CKSUM |
 288                DEV_RX_OFFLOAD_CRC_STRIP;
 289 }
 290
 291 static bool
 292 tap_rxq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 293 {
 294         uint64_t port_offloads = dev->data->dev_conf.rxmode.offloads;
 295         uint64_t queue_supp_offloads = tap_rx_offload_get_queue_capa();
 296         uint64_t port_supp_offloads = tap_rx_offload_get_port_capa();
 297
 298         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 299             offloads)
 300                 return false;
 301         if ((port_offloads ^ offloads) & port_supp_offloads)
 302                 return false;
 303         return true;
 304 }
 305
 306 /* Callback to handle the rx burst of packets to the correct interface and
 307  * file descriptor(s) in a multi-queue setup.
 308  */
 309 static uint16_t
 310 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 311 {
 312         struct rx_queue *rxq = queue;
 313         uint16_t num_rx;
 314         unsigned long num_rx_bytes = 0;
 315         uint32_t trigger = tap_trigger;
 316
 317         if (trigger == rxq->trigger_seen)
 318                 return 0;
 319         if (trigger)
 320                 rxq->trigger_seen = trigger;
 321         rte_compiler_barrier();
 322         for (num_rx = 0; num_rx < nb_pkts; ) {
 323                 struct rte_mbuf *mbuf = rxq->pool;
 324                 struct rte_mbuf *seg = NULL;
 325                 struct rte_mbuf *new_tail = NULL;
 326                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 327                 int len;
 328
 329                 len = readv(rxq->fd, *rxq->iovecs,
 330                             1 +
 331                             (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
 332                              rxq->nb_rx_desc : 1));
 333                 if (len < (int)sizeof(struct tun_pi))
 334                         break;
 335
 336                 /* Packet couldn't fit in the provided mbuf */
 337                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 338                         rxq->stats.ierrors++;
 339                         continue;
 340                 }
 341
 342                 len -= sizeof(struct tun_pi);
 343
 344                 mbuf->pkt_len = len;
 345                 mbuf->port = rxq->in_port;
 346                 while (1) {
 347                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 348
 349                         if (unlikely(!buf)) {
 350                                 rxq->stats.rx_nombuf++;
 351                                 /* No new buf has been allocated: do nothing */
 352                                 if (!new_tail || !seg)
 353                                         goto end;
 354
 355                                 seg->next = NULL;
 356                                 rte_pktmbuf_free(mbuf);
 357
 358                                 goto end;
 359                         }
 360                         seg = seg ? seg->next : mbuf;
 361                         if (rxq->pool == mbuf)
 362                                 rxq->pool = buf;
 363                         if (new_tail)
 364                                 new_tail->next = buf;
 365                         new_tail = buf;
 366                         new_tail->next = seg->next;
 367
 368                         /* iovecs[0] is reserved for packet info (pi) */
 369                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 370                                 buf->buf_len - data_off;
 371                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 372                                 (char *)buf->buf_addr + data_off;
 373
 374                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 375                         seg->data_off = data_off;
 376
 377                         len -= seg->data_len;
 378                         if (len <= 0)
 379                                 break;
 380                         mbuf->nb_segs++;
 381                         /* First segment has headroom, not the others */
 382                         data_off = 0;
 383                 }
 384                 seg->next = NULL;
 385                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 386                                                       RTE_PTYPE_ALL_MASK);
 387                 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
 388                         tap_verify_csum(mbuf);
 389
 390                 /* account for the receive frame */
 391                 bufs[num_rx++] = mbuf;
 392                 num_rx_bytes += mbuf->pkt_len;
 393         }
 394 end:
 395         rxq->stats.ipackets += num_rx;
 396         rxq->stats.ibytes += num_rx_bytes;
 397
 398         return num_rx;
 399 }
 400
 401 static uint64_t
 402 tap_tx_offload_get_port_capa(void)
 403 {
 404         /*
 405          * In order to support legacy apps,
 406          * report capabilities also as port capabilities.
 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 uint64_t
 415 tap_tx_offload_get_queue_capa(void)
 416 {
 417         return DEV_TX_OFFLOAD_MULTI_SEGS |
 418                DEV_TX_OFFLOAD_IPV4_CKSUM |
 419                DEV_TX_OFFLOAD_UDP_CKSUM |
 420                DEV_TX_OFFLOAD_TCP_CKSUM;
 421 }
 422
 423 static bool
 424 tap_txq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 425 {
 426         uint64_t port_offloads = dev->data->dev_conf.txmode.offloads;
 427         uint64_t queue_supp_offloads = tap_tx_offload_get_queue_capa();
 428         uint64_t port_supp_offloads = tap_tx_offload_get_port_capa();
 429
 430         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 431             offloads)
 432                 return false;
 433         /* Verify we have no conflict with port offloads */
 434         if ((port_offloads ^ offloads) & port_supp_offloads)
 435                 return false;
 436         return true;
 437 }
 438
 439 static void
 440 tap_tx_offload(char *packet, uint64_t ol_flags, unsigned int l2_len,
 441                unsigned int l3_len)
 442 {
 443         void *l3_hdr = packet + l2_len;
 444
 445         if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
 446                 struct ipv4_hdr *iph = l3_hdr;
 447                 uint16_t cksum;
 448
 449                 iph->hdr_checksum = 0;
 450                 cksum = rte_raw_cksum(iph, l3_len);
 451                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 452         }
 453         if (ol_flags & PKT_TX_L4_MASK) {
 454                 uint16_t l4_len;
 455                 uint32_t cksum;
 456                 uint16_t *l4_cksum;
 457                 void *l4_hdr;
 458
 459                 l4_hdr = packet + l2_len + l3_len;
 460                 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
 461                         l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
 462                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 463                         l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
 464                 else
 465                         return;
 466                 *l4_cksum = 0;
 467                 if (ol_flags & PKT_TX_IPV4) {
 468                         struct ipv4_hdr *iph = l3_hdr;
 469
 470                         l4_len = rte_be_to_cpu_16(iph->total_length) - l3_len;
 471                         cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 472                 } else {
 473                         struct ipv6_hdr *ip6h = l3_hdr;
 474
 475                         /* payload_len does not include ext headers */
 476                         l4_len = rte_be_to_cpu_16(ip6h->payload_len) -
 477                                 l3_len + sizeof(struct ipv6_hdr);
 478                         cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 479                 }
 480                 cksum += rte_raw_cksum(l4_hdr, l4_len);
 481                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 482                 cksum = (~cksum) & 0xffff;
 483                 if (cksum == 0)
 484                         cksum = 0xffff;
 485                 *l4_cksum = cksum;
 486         }
 487 }
 488
 489 /* Callback to handle sending packets from the tap interface
 490  */
 491 static uint16_t
 492 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 493 {
 494         struct tx_queue *txq = queue;
 495         uint16_t num_tx = 0;
 496         unsigned long num_tx_bytes = 0;
 497         uint32_t max_size;
 498         int i;
 499
 500         if (unlikely(nb_pkts == 0))
 501                 return 0;
 502
 503         max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
 504         for (i = 0; i < nb_pkts; i++) {
 505                 struct rte_mbuf *mbuf = bufs[num_tx];
 506                 struct iovec iovecs[mbuf->nb_segs + 1];
 507                 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
 508                 struct rte_mbuf *seg = mbuf;
 509                 char m_copy[mbuf->data_len];
 510                 int n;
 511                 int j;
 512
 513                 /* stats.errs will be incremented */
 514                 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
 515                         break;
 516
 517                 /*
 518                  * TUN and TAP are created with IFF_NO_PI disabled.
 519                  * For TUN PMD this mandatory as fields are used by
 520                  * Kernel tun.c to determine whether its IP or non IP
 521                  * packets.
 522                  *
 523                  * The logic fetches the first byte of data from mbuf.
 524                  * compares whether its v4 or v6. If none matches default
 525                  * value 0x00 is taken for protocol field.
 526                  */
 527                 char *buff_data = rte_pktmbuf_mtod(seg, void *);
 528                 j = (*buff_data & 0xf0);
 529                 if (j & (0x40 | 0x60))
 530                         pi.proto = (j == 0x40) ? 0x0008 : 0xdd86;
 531
 532                 iovecs[0].iov_base = &pi;
 533                 iovecs[0].iov_len = sizeof(pi);
 534                 for (j = 1; j <= mbuf->nb_segs; j++) {
 535                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 536                         iovecs[j].iov_base =
 537                                 rte_pktmbuf_mtod(seg, void *);
 538                         seg = seg->next;
 539                 }
 540                 if (txq->csum &&
 541                     ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 542                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 543                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
 544                         /* Support only packets with all data in the same seg */
 545                         if (mbuf->nb_segs > 1)
 546                                 break;
 547                         /* To change checksums, work on a copy of data. */
 548                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 549                                    rte_pktmbuf_data_len(mbuf));
 550                         tap_tx_offload(m_copy, mbuf->ol_flags,
 551                                        mbuf->l2_len, mbuf->l3_len);
 552                         iovecs[1].iov_base = m_copy;
 553                 }
 554                 /* copy the tx frame data */
 555                 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
 556                 if (n <= 0)
 557                         break;
 558
 559                 num_tx++;
 560                 num_tx_bytes += mbuf->pkt_len;
 561                 rte_pktmbuf_free(mbuf);
 562         }
 563
 564         txq->stats.opackets += num_tx;
 565         txq->stats.errs += nb_pkts - num_tx;
 566         txq->stats.obytes += num_tx_bytes;
 567
 568         return num_tx;
 569 }
 570
 571 static const char *
 572 tap_ioctl_req2str(unsigned long request)
 573 {
 574         switch (request) {
 575         case SIOCSIFFLAGS:
 576                 return "SIOCSIFFLAGS";
 577         case SIOCGIFFLAGS:
 578                 return "SIOCGIFFLAGS";
 579         case SIOCGIFHWADDR:
 580                 return "SIOCGIFHWADDR";
 581         case SIOCSIFHWADDR:
 582                 return "SIOCSIFHWADDR";
 583         case SIOCSIFMTU:
 584                 return "SIOCSIFMTU";
 585         }
 586         return "UNKNOWN";
 587 }
 588
 589 static int
 590 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 591           struct ifreq *ifr, int set, enum ioctl_mode mode)
 592 {
 593         short req_flags = ifr->ifr_flags;
 594         int remote = pmd->remote_if_index &&
 595                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 596
 597         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 598                 return 0;
 599         /*
 600          * If there is a remote netdevice, apply ioctl on it, then apply it on
 601          * the tap netdevice.
 602          */
 603 apply:
 604         if (remote)
 605                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
 606         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 607                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 608         switch (request) {
 609         case SIOCSIFFLAGS:
 610                 /* fetch current flags to leave other flags untouched */
 611                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 612                         goto error;
 613                 if (set)
 614                         ifr->ifr_flags |= req_flags;
 615                 else
 616                         ifr->ifr_flags &= ~req_flags;
 617                 break;
 618         case SIOCGIFFLAGS:
 619         case SIOCGIFHWADDR:
 620         case SIOCSIFHWADDR:
 621         case SIOCSIFMTU:
 622                 break;
 623         default:
 624                 RTE_ASSERT(!"unsupported request type: must not happen");
 625         }
 626         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 627                 goto error;
 628         if (remote-- && mode == LOCAL_AND_REMOTE)
 629                 goto apply;
 630         return 0;
 631
 632 error:
 633         RTE_LOG(DEBUG, PMD, "%s: %s(%s) failed: %s(%d)\n", ifr->ifr_name,
 634                 __func__, tap_ioctl_req2str(request), strerror(errno), errno);
 635         return -errno;
 636 }
 637
 638 static int
 639 tap_link_set_down(struct rte_eth_dev *dev)
 640 {
 641         struct pmd_internals *pmd = dev->data->dev_private;
 642         struct ifreq ifr = { .ifr_flags = IFF_UP };
 643
 644         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 645         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 646 }
 647
 648 static int
 649 tap_link_set_up(struct rte_eth_dev *dev)
 650 {
 651         struct pmd_internals *pmd = dev->data->dev_private;
 652         struct ifreq ifr = { .ifr_flags = IFF_UP };
 653
 654         dev->data->dev_link.link_status = ETH_LINK_UP;
 655         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 656 }
 657
 658 static int
 659 tap_dev_start(struct rte_eth_dev *dev)
 660 {
 661         int err;
 662
 663         err = tap_intr_handle_set(dev, 1);
 664         if (err)
 665                 return err;
 666         return tap_link_set_up(dev);
 667 }
 668
 669 /* This function gets called when the current port gets stopped.
 670  */
 671 static void
 672 tap_dev_stop(struct rte_eth_dev *dev)
 673 {
 674         tap_intr_handle_set(dev, 0);
 675         tap_link_set_down(dev);
 676 }
 677
 678 static int
 679 tap_dev_configure(struct rte_eth_dev *dev)
 680 {
 681         uint64_t supp_tx_offloads = tap_tx_offload_get_port_capa();
 682         uint64_t tx_offloads = dev->data->dev_conf.txmode.offloads;
 683
 684         if ((tx_offloads & supp_tx_offloads) != tx_offloads) {
 685                 rte_errno = ENOTSUP;
 686                 RTE_LOG(ERR, PMD,
 687                         "Some Tx offloads are not supported "
 688                         "requested 0x%" PRIx64 " supported 0x%" PRIx64 "\n",
 689                         tx_offloads, supp_tx_offloads);
 690                 return -rte_errno;
 691         }
 692         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 693                 RTE_LOG(ERR, PMD,
 694                         "%s: number of rx queues %d exceeds max num of queues %d\n",
 695                         dev->device->name,
 696                         dev->data->nb_rx_queues,
 697                         RTE_PMD_TAP_MAX_QUEUES);
 698                 return -1;
 699         }
 700         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 701                 RTE_LOG(ERR, PMD,
 702                         "%s: number of tx queues %d exceeds max num of queues %d\n",
 703                         dev->device->name,
 704                         dev->data->nb_tx_queues,
 705                         RTE_PMD_TAP_MAX_QUEUES);
 706                 return -1;
 707         }
 708
 709         RTE_LOG(INFO, PMD, "%s: %p: TX configured queues number: %u\n",
 710              dev->device->name, (void *)dev, dev->data->nb_tx_queues);
 711
 712         RTE_LOG(INFO, PMD, "%s: %p: RX configured queues number: %u\n",
 713              dev->device->name, (void *)dev, dev->data->nb_rx_queues);
 714
 715         return 0;
 716 }
 717
 718 static uint32_t
 719 tap_dev_speed_capa(void)
 720 {
 721         uint32_t speed = pmd_link.link_speed;
 722         uint32_t capa = 0;
 723
 724         if (speed >= ETH_SPEED_NUM_10M)
 725                 capa |= ETH_LINK_SPEED_10M;
 726         if (speed >= ETH_SPEED_NUM_100M)
 727                 capa |= ETH_LINK_SPEED_100M;
 728         if (speed >= ETH_SPEED_NUM_1G)
 729                 capa |= ETH_LINK_SPEED_1G;
 730         if (speed >= ETH_SPEED_NUM_5G)
 731                 capa |= ETH_LINK_SPEED_2_5G;
 732         if (speed >= ETH_SPEED_NUM_5G)
 733                 capa |= ETH_LINK_SPEED_5G;
 734         if (speed >= ETH_SPEED_NUM_10G)
 735                 capa |= ETH_LINK_SPEED_10G;
 736         if (speed >= ETH_SPEED_NUM_20G)
 737                 capa |= ETH_LINK_SPEED_20G;
 738         if (speed >= ETH_SPEED_NUM_25G)
 739                 capa |= ETH_LINK_SPEED_25G;
 740         if (speed >= ETH_SPEED_NUM_40G)
 741                 capa |= ETH_LINK_SPEED_40G;
 742         if (speed >= ETH_SPEED_NUM_50G)
 743                 capa |= ETH_LINK_SPEED_50G;
 744         if (speed >= ETH_SPEED_NUM_56G)
 745                 capa |= ETH_LINK_SPEED_56G;
 746         if (speed >= ETH_SPEED_NUM_100G)
 747                 capa |= ETH_LINK_SPEED_100G;
 748
 749         return capa;
 750 }
 751
 752 static void
 753 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 754 {
 755         struct pmd_internals *internals = dev->data->dev_private;
 756
 757         dev_info->if_index = internals->if_index;
 758         dev_info->max_mac_addrs = 1;
 759         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 760         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
 761         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
 762         dev_info->min_rx_bufsize = 0;
 763         dev_info->speed_capa = tap_dev_speed_capa();
 764         dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
 765         dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
 766                                     dev_info->rx_queue_offload_capa;
 767         dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
 768         dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
 769                                     dev_info->tx_queue_offload_capa;
 770 }
 771
 772 static int
 773 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 774 {
 775         unsigned int i, imax;
 776         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 777         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 778         unsigned long rx_nombuf = 0, ierrors = 0;
 779         const struct pmd_internals *pmd = dev->data->dev_private;
 780
 781         /* rx queue statistics */
 782         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 783                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 784         for (i = 0; i < imax; i++) {
 785                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 786                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 787                 rx_total += tap_stats->q_ipackets[i];
 788                 rx_bytes_total += tap_stats->q_ibytes[i];
 789                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
 790                 ierrors += pmd->rxq[i].stats.ierrors;
 791         }
 792
 793         /* tx queue statistics */
 794         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 795                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 796
 797         for (i = 0; i < imax; i++) {
 798                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 799                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 800                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 801                 tx_total += tap_stats->q_opackets[i];
 802                 tx_err_total += tap_stats->q_errors[i];
 803                 tx_bytes_total += tap_stats->q_obytes[i];
 804         }
 805
 806         tap_stats->ipackets = rx_total;
 807         tap_stats->ibytes = rx_bytes_total;
 808         tap_stats->ierrors = ierrors;
 809         tap_stats->rx_nombuf = rx_nombuf;
 810         tap_stats->opackets = tx_total;
 811         tap_stats->oerrors = tx_err_total;
 812         tap_stats->obytes = tx_bytes_total;
 813         return 0;
 814 }
 815
 816 static void
 817 tap_stats_reset(struct rte_eth_dev *dev)
 818 {
 819         int i;
 820         struct pmd_internals *pmd = dev->data->dev_private;
 821
 822         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 823                 pmd->rxq[i].stats.ipackets = 0;
 824                 pmd->rxq[i].stats.ibytes = 0;
 825                 pmd->rxq[i].stats.ierrors = 0;
 826                 pmd->rxq[i].stats.rx_nombuf = 0;
 827
 828                 pmd->txq[i].stats.opackets = 0;
 829                 pmd->txq[i].stats.errs = 0;
 830                 pmd->txq[i].stats.obytes = 0;
 831         }
 832 }
 833
 834 static void
 835 tap_dev_close(struct rte_eth_dev *dev)
 836 {
 837         int i;
 838         struct pmd_internals *internals = dev->data->dev_private;
 839
 840         tap_link_set_down(dev);
 841         tap_flow_flush(dev, NULL);
 842         tap_flow_implicit_flush(internals, NULL);
 843
 844         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 845                 if (internals->rxq[i].fd != -1) {
 846                         close(internals->rxq[i].fd);
 847                         internals->rxq[i].fd = -1;
 848                 }
 849                 if (internals->txq[i].fd != -1) {
 850                         close(internals->txq[i].fd);
 851                         internals->txq[i].fd = -1;
 852                 }
 853         }
 854
 855         if (internals->remote_if_index) {
 856                 /* Restore initial remote state */
 857                 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
 858                                 &internals->remote_initial_flags);
 859         }
 860 }
 861
 862 static void
 863 tap_rx_queue_release(void *queue)
 864 {
 865         struct rx_queue *rxq = queue;
 866
 867         if (rxq && (rxq->fd > 0)) {
 868                 close(rxq->fd);
 869                 rxq->fd = -1;
 870                 rte_pktmbuf_free(rxq->pool);
 871                 rte_free(rxq->iovecs);
 872                 rxq->pool = NULL;
 873                 rxq->iovecs = NULL;
 874         }
 875 }
 876
 877 static void
 878 tap_tx_queue_release(void *queue)
 879 {
 880         struct tx_queue *txq = queue;
 881
 882         if (txq && (txq->fd > 0)) {
 883                 close(txq->fd);
 884                 txq->fd = -1;
 885         }
 886 }
 887
 888 static int
 889 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
 890 {
 891         struct rte_eth_link *dev_link = &dev->data->dev_link;
 892         struct pmd_internals *pmd = dev->data->dev_private;
 893         struct ifreq ifr = { .ifr_flags = 0 };
 894
 895         if (pmd->remote_if_index) {
 896                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
 897                 if (!(ifr.ifr_flags & IFF_UP) ||
 898                     !(ifr.ifr_flags & IFF_RUNNING)) {
 899                         dev_link->link_status = ETH_LINK_DOWN;
 900                         return 0;
 901                 }
 902         }
 903         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
 904         dev_link->link_status =
 905                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
 906                  ETH_LINK_UP :
 907                  ETH_LINK_DOWN);
 908         return 0;
 909 }
 910
 911 static void
 912 tap_promisc_enable(struct rte_eth_dev *dev)
 913 {
 914         struct pmd_internals *pmd = dev->data->dev_private;
 915         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 916
 917         dev->data->promiscuous = 1;
 918         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 919         if (pmd->remote_if_index && !pmd->flow_isolate)
 920                 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
 921 }
 922
 923 static void
 924 tap_promisc_disable(struct rte_eth_dev *dev)
 925 {
 926         struct pmd_internals *pmd = dev->data->dev_private;
 927         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 928
 929         dev->data->promiscuous = 0;
 930         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 931         if (pmd->remote_if_index && !pmd->flow_isolate)
 932                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
 933 }
 934
 935 static void
 936 tap_allmulti_enable(struct rte_eth_dev *dev)
 937 {
 938         struct pmd_internals *pmd = dev->data->dev_private;
 939         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 940
 941         dev->data->all_multicast = 1;
 942         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 943         if (pmd->remote_if_index && !pmd->flow_isolate)
 944                 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
 945 }
 946
 947 static void
 948 tap_allmulti_disable(struct rte_eth_dev *dev)
 949 {
 950         struct pmd_internals *pmd = dev->data->dev_private;
 951         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 952
 953         dev->data->all_multicast = 0;
 954         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 955         if (pmd->remote_if_index && !pmd->flow_isolate)
 956                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
 957 }
 958
 959 static int
 960 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 961 {
 962         struct pmd_internals *pmd = dev->data->dev_private;
 963         enum ioctl_mode mode = LOCAL_ONLY;
 964         struct ifreq ifr;
 965         int ret;
 966
 967         if (is_zero_ether_addr(mac_addr)) {
 968                 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
 969                         dev->device->name);
 970                 return -EINVAL;
 971         }
 972         /* Check the actual current MAC address on the tap netdevice */
 973         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
 974         if (ret < 0)
 975                 return ret;
 976         if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 977                                mac_addr))
 978                 return 0;
 979         /* Check the current MAC address on the remote */
 980         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
 981         if (ret < 0)
 982                 return ret;
 983         if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 984                                mac_addr))
 985                 mode = LOCAL_AND_REMOTE;
 986         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
 987         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
 988         ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
 989         if (ret < 0)
 990                 return ret;
 991         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
 992         if (pmd->remote_if_index && !pmd->flow_isolate) {
 993                 /* Replace MAC redirection rule after a MAC change */
 994                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
 995                 if (ret < 0) {
 996                         RTE_LOG(ERR, PMD,
 997                                 "%s: Couldn't delete MAC redirection rule\n",
 998                                 dev->device->name);
 999                         return ret;
1000                 }
1001                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1002                 if (ret < 0) {
1003                         RTE_LOG(ERR, PMD,
1004                                 "%s: Couldn't add MAC redirection rule\n",
1005                                 dev->device->name);
1006                         return ret;
1007                 }
1008         }
1009
1010         return 0;
1011 }
1012
1013 static int
1014 tap_setup_queue(struct rte_eth_dev *dev,
1015                 struct pmd_internals *internals,
1016                 uint16_t qid,
1017                 int is_rx)
1018 {
1019         int *fd;
1020         int *other_fd;
1021         const char *dir;
1022         struct pmd_internals *pmd = dev->data->dev_private;
1023         struct rx_queue *rx = &internals->rxq[qid];
1024         struct tx_queue *tx = &internals->txq[qid];
1025
1026         if (is_rx) {
1027                 fd = &rx->fd;
1028                 other_fd = &tx->fd;
1029                 dir = "rx";
1030         } else {
1031                 fd = &tx->fd;
1032                 other_fd = &rx->fd;
1033                 dir = "tx";
1034         }
1035         if (*fd != -1) {
1036                 /* fd for this queue already exists */
1037                 RTE_LOG(DEBUG, PMD, "%s: fd %d for %s queue qid %d exists\n",
1038                         pmd->name, *fd, dir, qid);
1039         } else if (*other_fd != -1) {
1040                 /* Only other_fd exists. dup it */
1041                 *fd = dup(*other_fd);
1042                 if (*fd < 0) {
1043                         *fd = -1;
1044                         RTE_LOG(ERR, PMD, "%s: dup() failed.\n",
1045                                 pmd->name);
1046                         return -1;
1047                 }
1048                 RTE_LOG(DEBUG, PMD, "%s: dup fd %d for %s queue qid %d (%d)\n",
1049                         pmd->name, *other_fd, dir, qid, *fd);
1050         } else {
1051                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1052                 *fd = tun_alloc(pmd);
1053                 if (*fd < 0) {
1054                         *fd = -1; /* restore original value */
1055                         RTE_LOG(ERR, PMD, "%s: tun_alloc() failed.\n",
1056                                 pmd->name);
1057                         return -1;
1058                 }
1059                 RTE_LOG(DEBUG, PMD, "%s: add %s queue for qid %d fd %d\n",
1060                         pmd->name, dir, qid, *fd);
1061         }
1062
1063         tx->mtu = &dev->data->mtu;
1064         rx->rxmode = &dev->data->dev_conf.rxmode;
1065
1066         return *fd;
1067 }
1068
1069 static int
1070 tap_rx_queue_setup(struct rte_eth_dev *dev,
1071                    uint16_t rx_queue_id,
1072                    uint16_t nb_rx_desc,
1073                    unsigned int socket_id,
1074                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1075                    struct rte_mempool *mp)
1076 {
1077         struct pmd_internals *internals = dev->data->dev_private;
1078         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1079         struct rte_mbuf **tmp = &rxq->pool;
1080         long iov_max = sysconf(_SC_IOV_MAX);
1081         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1082         struct iovec (*iovecs)[nb_desc + 1];
1083         int data_off = RTE_PKTMBUF_HEADROOM;
1084         int ret = 0;
1085         int fd;
1086         int i;
1087
1088         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1089                 RTE_LOG(WARNING, PMD,
1090                         "nb_rx_queues %d too small or mempool NULL\n",
1091                         dev->data->nb_rx_queues);
1092                 return -1;
1093         }
1094
1095         /* Verify application offloads are valid for our port and queue. */
1096         if (!tap_rxq_are_offloads_valid(dev, rx_conf->offloads)) {
1097                 rte_errno = ENOTSUP;
1098                 RTE_LOG(ERR, PMD,
1099                         "%p: Rx queue offloads 0x%" PRIx64
1100                         " don't match port offloads 0x%" PRIx64
1101                         " or supported offloads 0x%" PRIx64 "\n",
1102                         (void *)dev, rx_conf->offloads,
1103                         dev->data->dev_conf.rxmode.offloads,
1104                         (tap_rx_offload_get_port_capa() |
1105                          tap_rx_offload_get_queue_capa()));
1106                 return -rte_errno;
1107         }
1108         rxq->mp = mp;
1109         rxq->trigger_seen = 1; /* force initial burst */
1110         rxq->in_port = dev->data->port_id;
1111         rxq->nb_rx_desc = nb_desc;
1112         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1113                                     socket_id);
1114         if (!iovecs) {
1115                 RTE_LOG(WARNING, PMD,
1116                         "%s: Couldn't allocate %d RX descriptors\n",
1117                         dev->device->name, nb_desc);
1118                 return -ENOMEM;
1119         }
1120         rxq->iovecs = iovecs;
1121
1122         dev->data->rx_queues[rx_queue_id] = rxq;
1123         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1124         if (fd == -1) {
1125                 ret = fd;
1126                 goto error;
1127         }
1128
1129         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1130         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1131
1132         for (i = 1; i <= nb_desc; i++) {
1133                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1134                 if (!*tmp) {
1135                         RTE_LOG(WARNING, PMD,
1136                                 "%s: couldn't allocate memory for queue %d\n",
1137                                 dev->device->name, rx_queue_id);
1138                         ret = -ENOMEM;
1139                         goto error;
1140                 }
1141                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1142                 (*rxq->iovecs)[i].iov_base =
1143                         (char *)(*tmp)->buf_addr + data_off;
1144                 data_off = 0;
1145                 tmp = &(*tmp)->next;
1146         }
1147
1148         RTE_LOG(DEBUG, PMD, "  RX TUNTAP device name %s, qid %d on fd %d\n",
1149                 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
1150
1151         return 0;
1152
1153 error:
1154         rte_pktmbuf_free(rxq->pool);
1155         rxq->pool = NULL;
1156         rte_free(rxq->iovecs);
1157         rxq->iovecs = NULL;
1158         return ret;
1159 }
1160
1161 static int
1162 tap_tx_queue_setup(struct rte_eth_dev *dev,
1163                    uint16_t tx_queue_id,
1164                    uint16_t nb_tx_desc __rte_unused,
1165                    unsigned int socket_id __rte_unused,
1166                    const struct rte_eth_txconf *tx_conf)
1167 {
1168         struct pmd_internals *internals = dev->data->dev_private;
1169         struct tx_queue *txq;
1170         int ret;
1171
1172         if (tx_queue_id >= dev->data->nb_tx_queues)
1173                 return -1;
1174         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1175         txq = dev->data->tx_queues[tx_queue_id];
1176         /*
1177          * Don't verify port offloads for application which
1178          * use the old API.
1179          */
1180         if (tx_conf != NULL &&
1181             !!(tx_conf->txq_flags & ETH_TXQ_FLAGS_IGNORE)) {
1182                 if (tap_txq_are_offloads_valid(dev, tx_conf->offloads)) {
1183                         txq->csum = !!(tx_conf->offloads &
1184                                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
1185                                          DEV_TX_OFFLOAD_UDP_CKSUM |
1186                                          DEV_TX_OFFLOAD_TCP_CKSUM));
1187                 } else {
1188                         rte_errno = ENOTSUP;
1189                         RTE_LOG(ERR, PMD,
1190                                 "%p: Tx queue offloads 0x%" PRIx64
1191                                 " don't match port offloads 0x%" PRIx64
1192                                 " or supported offloads 0x%" PRIx64,
1193                                 (void *)dev, tx_conf->offloads,
1194                                 dev->data->dev_conf.txmode.offloads,
1195                                 tap_tx_offload_get_port_capa());
1196                         return -rte_errno;
1197                 }
1198         }
1199         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1200         if (ret == -1)
1201                 return -1;
1202         RTE_LOG(DEBUG, PMD,
1203                 "  TX TUNTAP device name %s, qid %d on fd %d csum %s\n",
1204                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd,
1205                 txq->csum ? "on" : "off");
1206
1207         return 0;
1208 }
1209
1210 static int
1211 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1212 {
1213         struct pmd_internals *pmd = dev->data->dev_private;
1214         struct ifreq ifr = { .ifr_mtu = mtu };
1215         int err = 0;
1216
1217         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1218         if (!err)
1219                 dev->data->mtu = mtu;
1220
1221         return err;
1222 }
1223
1224 static int
1225 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1226                      struct ether_addr *mc_addr_set __rte_unused,
1227                      uint32_t nb_mc_addr __rte_unused)
1228 {
1229         /*
1230          * Nothing to do actually: the tap has no filtering whatsoever, every
1231          * packet is received.
1232          */
1233         return 0;
1234 }
1235
1236 static int
1237 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1238 {
1239         struct rte_eth_dev *dev = arg;
1240         struct pmd_internals *pmd = dev->data->dev_private;
1241         struct ifinfomsg *info = NLMSG_DATA(nh);
1242
1243         if (nh->nlmsg_type != RTM_NEWLINK ||
1244             (info->ifi_index != pmd->if_index &&
1245              info->ifi_index != pmd->remote_if_index))
1246                 return 0;
1247         return tap_link_update(dev, 0);
1248 }
1249
1250 static void
1251 tap_dev_intr_handler(void *cb_arg)
1252 {
1253         struct rte_eth_dev *dev = cb_arg;
1254         struct pmd_internals *pmd = dev->data->dev_private;
1255
1256         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1257 }
1258
1259 static int
1260 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1261 {
1262         struct pmd_internals *pmd = dev->data->dev_private;
1263
1264         /* In any case, disable interrupt if the conf is no longer there. */
1265         if (!dev->data->dev_conf.intr_conf.lsc) {
1266                 if (pmd->intr_handle.fd != -1) {
1267                         tap_nl_final(pmd->intr_handle.fd);
1268                         rte_intr_callback_unregister(&pmd->intr_handle,
1269                                 tap_dev_intr_handler, dev);
1270                 }
1271                 return 0;
1272         }
1273         if (set) {
1274                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1275                 if (unlikely(pmd->intr_handle.fd == -1))
1276                         return -EBADF;
1277                 return rte_intr_callback_register(
1278                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1279         }
1280         tap_nl_final(pmd->intr_handle.fd);
1281         return rte_intr_callback_unregister(&pmd->intr_handle,
1282                                             tap_dev_intr_handler, dev);
1283 }
1284
1285 static int
1286 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1287 {
1288         int err;
1289
1290         err = tap_lsc_intr_handle_set(dev, set);
1291         if (err)
1292                 return err;
1293         err = tap_rx_intr_vec_set(dev, set);
1294         if (err && set)
1295                 tap_lsc_intr_handle_set(dev, 0);
1296         return err;
1297 }
1298
1299 static const uint32_t*
1300 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1301 {
1302         static const uint32_t ptypes[] = {
1303                 RTE_PTYPE_INNER_L2_ETHER,
1304                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1305                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1306                 RTE_PTYPE_INNER_L3_IPV4,
1307                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1308                 RTE_PTYPE_INNER_L3_IPV6,
1309                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1310                 RTE_PTYPE_INNER_L4_FRAG,
1311                 RTE_PTYPE_INNER_L4_UDP,
1312                 RTE_PTYPE_INNER_L4_TCP,
1313                 RTE_PTYPE_INNER_L4_SCTP,
1314                 RTE_PTYPE_L2_ETHER,
1315                 RTE_PTYPE_L2_ETHER_VLAN,
1316                 RTE_PTYPE_L2_ETHER_QINQ,
1317                 RTE_PTYPE_L3_IPV4,
1318                 RTE_PTYPE_L3_IPV4_EXT,
1319                 RTE_PTYPE_L3_IPV6_EXT,
1320                 RTE_PTYPE_L3_IPV6,
1321                 RTE_PTYPE_L4_FRAG,
1322                 RTE_PTYPE_L4_UDP,
1323                 RTE_PTYPE_L4_TCP,
1324                 RTE_PTYPE_L4_SCTP,
1325         };
1326
1327         return ptypes;
1328 }
1329
1330 static int
1331 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1332                   struct rte_eth_fc_conf *fc_conf)
1333 {
1334         fc_conf->mode = RTE_FC_NONE;
1335         return 0;
1336 }
1337
1338 static int
1339 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1340                   struct rte_eth_fc_conf *fc_conf)
1341 {
1342         if (fc_conf->mode != RTE_FC_NONE)
1343                 return -ENOTSUP;
1344         return 0;
1345 }
1346
1347 static const struct eth_dev_ops ops = {
1348         .dev_start              = tap_dev_start,
1349         .dev_stop               = tap_dev_stop,
1350         .dev_close              = tap_dev_close,
1351         .dev_configure          = tap_dev_configure,
1352         .dev_infos_get          = tap_dev_info,
1353         .rx_queue_setup         = tap_rx_queue_setup,
1354         .tx_queue_setup         = tap_tx_queue_setup,
1355         .rx_queue_release       = tap_rx_queue_release,
1356         .tx_queue_release       = tap_tx_queue_release,
1357         .flow_ctrl_get          = tap_flow_ctrl_get,
1358         .flow_ctrl_set          = tap_flow_ctrl_set,
1359         .link_update            = tap_link_update,
1360         .dev_set_link_up        = tap_link_set_up,
1361         .dev_set_link_down      = tap_link_set_down,
1362         .promiscuous_enable     = tap_promisc_enable,
1363         .promiscuous_disable    = tap_promisc_disable,
1364         .allmulticast_enable    = tap_allmulti_enable,
1365         .allmulticast_disable   = tap_allmulti_disable,
1366         .mac_addr_set           = tap_mac_set,
1367         .mtu_set                = tap_mtu_set,
1368         .set_mc_addr_list       = tap_set_mc_addr_list,
1369         .stats_get              = tap_stats_get,
1370         .stats_reset            = tap_stats_reset,
1371         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1372         .filter_ctrl            = tap_dev_filter_ctrl,
1373 };
1374
1375 static int
1376 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1377                    char *remote_iface, struct ether_addr *mac_addr)
1378 {
1379         int numa_node = rte_socket_id();
1380         struct rte_eth_dev *dev;
1381         struct pmd_internals *pmd;
1382         struct rte_eth_dev_data *data;
1383         struct ifreq ifr;
1384         int i;
1385
1386         RTE_LOG(DEBUG, PMD, "%s device on numa %u\n",
1387                         tuntap_name, rte_socket_id());
1388
1389         data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
1390         if (!data) {
1391                 RTE_LOG(ERR, PMD, "%s Failed to allocate data\n", tuntap_name);
1392                 goto error_exit_nodev;
1393         }
1394
1395         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1396         if (!dev) {
1397                 RTE_LOG(ERR, PMD, "%s Unable to allocate device struct\n",
1398                                 tuntap_name);
1399                 goto error_exit_nodev;
1400         }
1401
1402         pmd = dev->data->dev_private;
1403         pmd->dev = dev;
1404         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1405
1406         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1407         if (pmd->ioctl_sock == -1) {
1408                 RTE_LOG(ERR, PMD,
1409                         "%s Unable to get a socket for management: %s\n",
1410                         tuntap_name, strerror(errno));
1411                 goto error_exit;
1412         }
1413
1414         /* Setup some default values */
1415         rte_memcpy(data, dev->data, sizeof(*data));
1416         data->dev_private = pmd;
1417         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1418         data->numa_node = numa_node;
1419
1420         data->dev_link = pmd_link;
1421         data->mac_addrs = &pmd->eth_addr;
1422         /* Set the number of RX and TX queues */
1423         data->nb_rx_queues = 0;
1424         data->nb_tx_queues = 0;
1425
1426         dev->data = data;
1427         dev->dev_ops = &ops;
1428         dev->rx_pkt_burst = pmd_rx_burst;
1429         dev->tx_pkt_burst = pmd_tx_burst;
1430
1431         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1432         pmd->intr_handle.fd = -1;
1433         dev->intr_handle = &pmd->intr_handle;
1434
1435         /* Presetup the fds to -1 as being not valid */
1436         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1437                 pmd->rxq[i].fd = -1;
1438                 pmd->txq[i].fd = -1;
1439         }
1440
1441         if (tap_type) {
1442                 if (is_zero_ether_addr(mac_addr))
1443                         eth_random_addr((uint8_t *)&pmd->eth_addr);
1444                 else
1445                         rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1446         }
1447
1448         /* Immediately create the netdevice (this will create the 1st queue). */
1449         /* rx queue */
1450         if (tap_setup_queue(dev, pmd, 0, 1) == -1)
1451                 goto error_exit;
1452         /* tx queue */
1453         if (tap_setup_queue(dev, pmd, 0, 0) == -1)
1454                 goto error_exit;
1455
1456         ifr.ifr_mtu = dev->data->mtu;
1457         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1458                 goto error_exit;
1459
1460         if (tap_type) {
1461                 memset(&ifr, 0, sizeof(struct ifreq));
1462                 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1463                 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1464                                 ETHER_ADDR_LEN);
1465                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1466                         goto error_exit;
1467         }
1468
1469         /*
1470          * Set up everything related to rte_flow:
1471          * - netlink socket
1472          * - tap / remote if_index
1473          * - mandatory QDISCs
1474          * - rte_flow actual/implicit lists
1475          * - implicit rules
1476          */
1477         pmd->nlsk_fd = tap_nl_init(0);
1478         if (pmd->nlsk_fd == -1) {
1479                 RTE_LOG(WARNING, PMD, "%s: failed to create netlink socket.\n",
1480                         pmd->name);
1481                 goto disable_rte_flow;
1482         }
1483         pmd->if_index = if_nametoindex(pmd->name);
1484         if (!pmd->if_index) {
1485                 RTE_LOG(ERR, PMD, "%s: failed to get if_index.\n", pmd->name);
1486                 goto disable_rte_flow;
1487         }
1488         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1489                 RTE_LOG(ERR, PMD, "%s: failed to create multiq qdisc.\n",
1490                         pmd->name);
1491                 goto disable_rte_flow;
1492         }
1493         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1494                 RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1495                         pmd->name);
1496                 goto disable_rte_flow;
1497         }
1498         LIST_INIT(&pmd->flows);
1499
1500         if (strlen(remote_iface)) {
1501                 pmd->remote_if_index = if_nametoindex(remote_iface);
1502                 if (!pmd->remote_if_index) {
1503                         RTE_LOG(ERR, PMD, "%s: failed to get %s if_index.\n",
1504                                 pmd->name, remote_iface);
1505                         goto error_remote;
1506                 }
1507                 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1508                          "%s", remote_iface);
1509
1510                 /* Save state of remote device */
1511                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1512
1513                 /* Replicate remote MAC address */
1514                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1515                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1516                                 pmd->name, pmd->remote_iface);
1517                         goto error_remote;
1518                 }
1519                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1520                            ETHER_ADDR_LEN);
1521                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1522                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1523                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1524                                 pmd->name, remote_iface);
1525                         goto error_remote;
1526                 }
1527
1528                 /*
1529                  * Flush usually returns negative value because it tries to
1530                  * delete every QDISC (and on a running device, one QDISC at
1531                  * least is needed). Ignore negative return value.
1532                  */
1533                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1534                 if (qdisc_create_ingress(pmd->nlsk_fd,
1535                                          pmd->remote_if_index) < 0) {
1536                         RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1537                                 pmd->remote_iface);
1538                         goto error_remote;
1539                 }
1540                 LIST_INIT(&pmd->implicit_flows);
1541                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1542                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1543                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1544                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1545                         RTE_LOG(ERR, PMD,
1546                                 "%s: failed to create implicit rules.\n",
1547                                 pmd->name);
1548                         goto error_remote;
1549                 }
1550         }
1551
1552         return 0;
1553
1554 disable_rte_flow:
1555         RTE_LOG(ERR, PMD, " Disabling rte flow support: %s(%d)\n",
1556                 strerror(errno), errno);
1557         if (strlen(remote_iface)) {
1558                 RTE_LOG(ERR, PMD, "Remote feature requires flow support.\n");
1559                 goto error_exit;
1560         }
1561         return 0;
1562
1563 error_remote:
1564         RTE_LOG(ERR, PMD, " Can't set up remote feature: %s(%d)\n",
1565                 strerror(errno), errno);
1566         tap_flow_implicit_flush(pmd, NULL);
1567
1568 error_exit:
1569         if (pmd->ioctl_sock > 0)
1570                 close(pmd->ioctl_sock);
1571         rte_eth_dev_release_port(dev);
1572
1573 error_exit_nodev:
1574         RTE_LOG(ERR, PMD, "%s Unable to initialize %s\n",
1575                 tuntap_name, rte_vdev_device_name(vdev));
1576
1577         rte_free(data);
1578         return -EINVAL;
1579 }
1580
1581 static int
1582 set_interface_name(const char *key __rte_unused,
1583                    const char *value,
1584                    void *extra_args)
1585 {
1586         char *name = (char *)extra_args;
1587
1588         if (value)
1589                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN - 1);
1590         else
1591                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1592                          DEFAULT_TAP_NAME, (tap_unit - 1));
1593
1594         return 0;
1595 }
1596
1597 static int
1598 set_remote_iface(const char *key __rte_unused,
1599                  const char *value,
1600                  void *extra_args)
1601 {
1602         char *name = (char *)extra_args;
1603
1604         if (value)
1605                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
1606
1607         return 0;
1608 }
1609
1610 static int parse_user_mac(struct ether_addr *user_mac,
1611                 const char *value)
1612 {
1613         unsigned int index = 0;
1614         char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
1615
1616         if (user_mac == NULL || value == NULL)
1617                 return 0;
1618
1619         strlcpy(mac_temp, value, sizeof(mac_temp));
1620         mac_byte = strtok(mac_temp, ":");
1621
1622         while ((mac_byte != NULL) &&
1623                         (strlen(mac_byte) <= 2) &&
1624                         (strlen(mac_byte) == strspn(mac_byte,
1625                                         ETH_TAP_CMP_MAC_FMT))) {
1626                 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
1627                 mac_byte = strtok(NULL, ":");
1628         }
1629
1630         return index;
1631 }
1632
1633 static int
1634 set_mac_type(const char *key __rte_unused,
1635              const char *value,
1636              void *extra_args)
1637 {
1638         struct ether_addr *user_mac = extra_args;
1639
1640         if (!value)
1641                 return 0;
1642
1643         if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
1644                 static int iface_idx;
1645
1646                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1647                 memcpy((char *)user_mac->addr_bytes, "\0dtap", ETHER_ADDR_LEN);
1648                 user_mac->addr_bytes[ETHER_ADDR_LEN - 1] = iface_idx++ + '0';
1649                 goto success;
1650         }
1651
1652         if (parse_user_mac(user_mac, value) != 6)
1653                 goto error;
1654 success:
1655         RTE_LOG(DEBUG, PMD, "TAP user MAC param (%s)\n", value);
1656         return 0;
1657
1658 error:
1659         RTE_LOG(ERR, PMD, "TAP user MAC (%s) is not in format (%s|%s)\n",
1660                 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
1661         return -1;
1662 }
1663
1664 /*
1665  * Open a TUN interface device. TUN PMD
1666  * 1) sets tap_type as false
1667  * 2) intakes iface as argument.
1668  * 3) as interface is virtual set speed to 10G
1669  */
1670 static int
1671 rte_pmd_tun_probe(struct rte_vdev_device *dev)
1672 {
1673         const char *name, *params;
1674         int ret;
1675         struct rte_kvargs *kvlist = NULL;
1676         char tun_name[RTE_ETH_NAME_MAX_LEN];
1677         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1678
1679         tap_type = 0;
1680         strcpy(tuntap_name, "TUN");
1681
1682         name = rte_vdev_device_name(dev);
1683         params = rte_vdev_device_args(dev);
1684         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1685
1686         if (params && (params[0] != '\0')) {
1687                 RTE_LOG(DEBUG, PMD, "parameters (%s)\n", params);
1688
1689                 kvlist = rte_kvargs_parse(params, valid_arguments);
1690                 if (kvlist) {
1691                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1692                                 ret = rte_kvargs_process(kvlist,
1693                                         ETH_TAP_IFACE_ARG,
1694                                         &set_interface_name,
1695                                         tun_name);
1696
1697                                 if (ret == -1)
1698                                         goto leave;
1699                         }
1700                 }
1701         }
1702         pmd_link.link_speed = ETH_SPEED_NUM_10G;
1703
1704         RTE_LOG(NOTICE, PMD, "Initializing pmd_tun for %s as %s\n",
1705                 name, tun_name);
1706
1707         ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0);
1708
1709 leave:
1710         if (ret == -1) {
1711                 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1712                         name, tun_name);
1713                 tun_unit--; /* Restore the unit number */
1714         }
1715         rte_kvargs_free(kvlist);
1716
1717         return ret;
1718 }
1719
1720 /* Open a TAP interface device.
1721  */
1722 static int
1723 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1724 {
1725         const char *name, *params;
1726         int ret;
1727         struct rte_kvargs *kvlist = NULL;
1728         int speed;
1729         char tap_name[RTE_ETH_NAME_MAX_LEN];
1730         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1731         struct ether_addr user_mac = { .addr_bytes = {0} };
1732
1733         tap_type = 1;
1734         strcpy(tuntap_name, "TAP");
1735
1736         name = rte_vdev_device_name(dev);
1737         params = rte_vdev_device_args(dev);
1738
1739         speed = ETH_SPEED_NUM_10G;
1740         snprintf(tap_name, sizeof(tap_name), "%s%d",
1741                  DEFAULT_TAP_NAME, tap_unit++);
1742         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1743
1744         if (params && (params[0] != '\0')) {
1745                 RTE_LOG(DEBUG, PMD, "parameters (%s)\n", params);
1746
1747                 kvlist = rte_kvargs_parse(params, valid_arguments);
1748                 if (kvlist) {
1749                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1750                                 ret = rte_kvargs_process(kvlist,
1751                                                          ETH_TAP_IFACE_ARG,
1752                                                          &set_interface_name,
1753                                                          tap_name);
1754                                 if (ret == -1)
1755                                         goto leave;
1756                         }
1757
1758                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1759                                 ret = rte_kvargs_process(kvlist,
1760                                                          ETH_TAP_REMOTE_ARG,
1761                                                          &set_remote_iface,
1762                                                          remote_iface);
1763                                 if (ret == -1)
1764                                         goto leave;
1765                         }
1766
1767                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1768                                 ret = rte_kvargs_process(kvlist,
1769                                                          ETH_TAP_MAC_ARG,
1770                                                          &set_mac_type,
1771                                                          &user_mac);
1772                                 if (ret == -1)
1773                                         goto leave;
1774                         }
1775                 }
1776         }
1777         pmd_link.link_speed = speed;
1778
1779         RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1780                 name, tap_name);
1781
1782         ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac);
1783
1784 leave:
1785         if (ret == -1) {
1786                 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1787                         name, tap_name);
1788                 tap_unit--;             /* Restore the unit number */
1789         }
1790         rte_kvargs_free(kvlist);
1791
1792         return ret;
1793 }
1794
1795 /* detach a TUNTAP device.
1796  */
1797 static int
1798 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1799 {
1800         struct rte_eth_dev *eth_dev = NULL;
1801         struct pmd_internals *internals;
1802         int i;
1803
1804         RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1805                 rte_socket_id());
1806
1807         /* find the ethdev entry */
1808         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1809         if (!eth_dev)
1810                 return 0;
1811
1812         internals = eth_dev->data->dev_private;
1813         if (internals->nlsk_fd) {
1814                 tap_flow_flush(eth_dev, NULL);
1815                 tap_flow_implicit_flush(internals, NULL);
1816                 tap_nl_final(internals->nlsk_fd);
1817         }
1818         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1819                 if (internals->rxq[i].fd != -1) {
1820                         close(internals->rxq[i].fd);
1821                         internals->rxq[i].fd = -1;
1822                 }
1823                 if (internals->txq[i].fd != -1) {
1824                         close(internals->txq[i].fd);
1825                         internals->txq[i].fd = -1;
1826                 }
1827         }
1828
1829         close(internals->ioctl_sock);
1830         rte_free(eth_dev->data->dev_private);
1831         rte_free(eth_dev->data);
1832
1833         rte_eth_dev_release_port(eth_dev);
1834
1835         return 0;
1836 }
1837
1838 static struct rte_vdev_driver pmd_tun_drv = {
1839         .probe = rte_pmd_tun_probe,
1840         .remove = rte_pmd_tap_remove,
1841 };
1842
1843 static struct rte_vdev_driver pmd_tap_drv = {
1844         .probe = rte_pmd_tap_probe,
1845         .remove = rte_pmd_tap_remove,
1846 };
1847 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1848 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
1849 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1850 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
1851                               ETH_TAP_IFACE_ARG "=<string> ");
1852 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1853                               ETH_TAP_IFACE_ARG "=<string> "
1854                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
1855                               ETH_TAP_REMOTE_ARG "=<string>");