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 TAP interface\n");
 126                 goto error;
 127         }
 128
 129 #ifdef IFF_MULTI_QUEUE
 130         /* Grab the TUN features to verify we can work multi-queue */
 131         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 132                 RTE_LOG(ERR, PMD, "TAP unable to get TUN/TAP features\n");
 133                 goto error;
 134         }
 135         RTE_LOG(DEBUG, PMD, "  TAP Features %08x\n", features);
 136
 137         if (features & IFF_MULTI_QUEUE) {
 138                 RTE_LOG(DEBUG, PMD, "  Multi-queue support for %d queues\n",
 139                         RTE_PMD_TAP_MAX_QUEUES);
 140                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 141         } else
 142 #endif
 143         {
 144                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 145                 RTE_LOG(DEBUG, PMD, "  Single queue only support\n");
 146         }
 147
 148         /* Set the TUN/TAP configuration and set the name if needed */
 149         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 150                 RTE_LOG(WARNING, PMD,
 151                         "Unable to set TUNSETIFF for %s\n",
 152                         ifr.ifr_name);
 153                 perror("TUNSETIFF");
 154                 goto error;
 155         }
 156
 157         /* Always set the file descriptor to non-blocking */
 158         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 159                 RTE_LOG(WARNING, PMD,
 160                         "Unable to set %s to nonblocking\n",
 161                         ifr.ifr_name);
 162                 perror("F_SETFL, NONBLOCK");
 163                 goto error;
 164         }
 165
 166         /* Set up trigger to optimize empty Rx bursts */
 167         errno = 0;
 168         do {
 169                 struct sigaction sa;
 170                 int flags = fcntl(fd, F_GETFL);
 171
 172                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 173                         break;
 174                 if (sa.sa_handler != tap_trigger_cb) {
 175                         /*
 176                          * Make sure SIGIO is not already taken. This is done
 177                          * as late as possible to leave the application a
 178                          * chance to set up its own signal handler first.
 179                          */
 180                         if (sa.sa_handler != SIG_IGN &&
 181                             sa.sa_handler != SIG_DFL) {
 182                                 errno = EBUSY;
 183                                 break;
 184                         }
 185                         sa = (struct sigaction){
 186                                 .sa_flags = SA_RESTART,
 187                                 .sa_handler = tap_trigger_cb,
 188                         };
 189                         if (sigaction(SIGIO, &sa, NULL) == -1)
 190                                 break;
 191                 }
 192                 /* Enable SIGIO on file descriptor */
 193                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 194                 fcntl(fd, F_SETOWN, getpid());
 195         } while (0);
 196         if (errno) {
 197                 /* Disable trigger globally in case of error */
 198                 tap_trigger = 0;
 199                 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
 200                         strerror(errno));
 201         }
 202
 203         return fd;
 204
 205 error:
 206         if (fd > 0)
 207                 close(fd);
 208         return -1;
 209 }
 210
 211 static void
 212 tap_verify_csum(struct rte_mbuf *mbuf)
 213 {
 214         uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
 215         uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
 216         uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
 217         unsigned int l2_len = sizeof(struct ether_hdr);
 218         unsigned int l3_len;
 219         uint16_t cksum = 0;
 220         void *l3_hdr;
 221         void *l4_hdr;
 222
 223         if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
 224                 l2_len += 4;
 225         else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
 226                 l2_len += 8;
 227         /* Don't verify checksum for packets with discontinuous L2 header */
 228         if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
 229                      rte_pktmbuf_data_len(mbuf)))
 230                 return;
 231         l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
 232         if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 233                 struct ipv4_hdr *iph = l3_hdr;
 234
 235                 /* ihl contains the number of 4-byte words in the header */
 236                 l3_len = 4 * (iph->version_ihl & 0xf);
 237                 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
 238                         return;
 239
 240                 cksum = ~rte_raw_cksum(iph, l3_len);
 241                 mbuf->ol_flags |= cksum ?
 242                         PKT_RX_IP_CKSUM_BAD :
 243                         PKT_RX_IP_CKSUM_GOOD;
 244         } else if (l3 == RTE_PTYPE_L3_IPV6) {
 245                 l3_len = sizeof(struct ipv6_hdr);
 246         } else {
 247                 /* IPv6 extensions are not supported */
 248                 return;
 249         }
 250         if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
 251                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 252                 /* Don't verify checksum for multi-segment packets. */
 253                 if (mbuf->nb_segs > 1)
 254                         return;
 255                 if (l3 == RTE_PTYPE_L3_IPV4)
 256                         cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
 257                 else if (l3 == RTE_PTYPE_L3_IPV6)
 258                         cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
 259                 mbuf->ol_flags |= cksum ?
 260                         PKT_RX_L4_CKSUM_BAD :
 261                         PKT_RX_L4_CKSUM_GOOD;
 262         }
 263 }
 264
 265 static uint64_t
 266 tap_rx_offload_get_port_capa(void)
 267 {
 268         /*
 269          * In order to support legacy apps,
 270          * report capabilities also as port capabilities.
 271          */
 272         return DEV_RX_OFFLOAD_SCATTER |
 273                DEV_RX_OFFLOAD_IPV4_CKSUM |
 274                DEV_RX_OFFLOAD_UDP_CKSUM |
 275                DEV_RX_OFFLOAD_TCP_CKSUM |
 276                DEV_RX_OFFLOAD_CRC_STRIP;
 277 }
 278
 279 static uint64_t
 280 tap_rx_offload_get_queue_capa(void)
 281 {
 282         return DEV_RX_OFFLOAD_SCATTER |
 283                DEV_RX_OFFLOAD_IPV4_CKSUM |
 284                DEV_RX_OFFLOAD_UDP_CKSUM |
 285                DEV_RX_OFFLOAD_TCP_CKSUM |
 286                DEV_RX_OFFLOAD_CRC_STRIP;
 287 }
 288
 289 static bool
 290 tap_rxq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 291 {
 292         uint64_t port_offloads = dev->data->dev_conf.rxmode.offloads;
 293         uint64_t queue_supp_offloads = tap_rx_offload_get_queue_capa();
 294         uint64_t port_supp_offloads = tap_rx_offload_get_port_capa();
 295
 296         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 297             offloads)
 298                 return false;
 299         if ((port_offloads ^ offloads) & port_supp_offloads)
 300                 return false;
 301         return true;
 302 }
 303
 304 /* Callback to handle the rx burst of packets to the correct interface and
 305  * file descriptor(s) in a multi-queue setup.
 306  */
 307 static uint16_t
 308 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 309 {
 310         struct rx_queue *rxq = queue;
 311         uint16_t num_rx;
 312         unsigned long num_rx_bytes = 0;
 313         uint32_t trigger = tap_trigger;
 314
 315         if (trigger == rxq->trigger_seen)
 316                 return 0;
 317         if (trigger)
 318                 rxq->trigger_seen = trigger;
 319         rte_compiler_barrier();
 320         for (num_rx = 0; num_rx < nb_pkts; ) {
 321                 struct rte_mbuf *mbuf = rxq->pool;
 322                 struct rte_mbuf *seg = NULL;
 323                 struct rte_mbuf *new_tail = NULL;
 324                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 325                 int len;
 326
 327                 len = readv(rxq->fd, *rxq->iovecs,
 328                             1 +
 329                             (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
 330                              rxq->nb_rx_desc : 1));
 331                 if (len < (int)sizeof(struct tun_pi))
 332                         break;
 333
 334                 /* Packet couldn't fit in the provided mbuf */
 335                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 336                         rxq->stats.ierrors++;
 337                         continue;
 338                 }
 339
 340                 len -= sizeof(struct tun_pi);
 341
 342                 mbuf->pkt_len = len;
 343                 mbuf->port = rxq->in_port;
 344                 while (1) {
 345                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 346
 347                         if (unlikely(!buf)) {
 348                                 rxq->stats.rx_nombuf++;
 349                                 /* No new buf has been allocated: do nothing */
 350                                 if (!new_tail || !seg)
 351                                         goto end;
 352
 353                                 seg->next = NULL;
 354                                 rte_pktmbuf_free(mbuf);
 355
 356                                 goto end;
 357                         }
 358                         seg = seg ? seg->next : mbuf;
 359                         if (rxq->pool == mbuf)
 360                                 rxq->pool = buf;
 361                         if (new_tail)
 362                                 new_tail->next = buf;
 363                         new_tail = buf;
 364                         new_tail->next = seg->next;
 365
 366                         /* iovecs[0] is reserved for packet info (pi) */
 367                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 368                                 buf->buf_len - data_off;
 369                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 370                                 (char *)buf->buf_addr + data_off;
 371
 372                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 373                         seg->data_off = data_off;
 374
 375                         len -= seg->data_len;
 376                         if (len <= 0)
 377                                 break;
 378                         mbuf->nb_segs++;
 379                         /* First segment has headroom, not the others */
 380                         data_off = 0;
 381                 }
 382                 seg->next = NULL;
 383                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 384                                                       RTE_PTYPE_ALL_MASK);
 385                 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
 386                         tap_verify_csum(mbuf);
 387
 388                 /* account for the receive frame */
 389                 bufs[num_rx++] = mbuf;
 390                 num_rx_bytes += mbuf->pkt_len;
 391         }
 392 end:
 393         rxq->stats.ipackets += num_rx;
 394         rxq->stats.ibytes += num_rx_bytes;
 395
 396         return num_rx;
 397 }
 398
 399 static uint64_t
 400 tap_tx_offload_get_port_capa(void)
 401 {
 402         /*
 403          * In order to support legacy apps,
 404          * report capabilities also as port capabilities.
 405          */
 406         return DEV_TX_OFFLOAD_MULTI_SEGS |
 407                DEV_TX_OFFLOAD_IPV4_CKSUM |
 408                DEV_TX_OFFLOAD_UDP_CKSUM |
 409                DEV_TX_OFFLOAD_TCP_CKSUM;
 410 }
 411
 412 static uint64_t
 413 tap_tx_offload_get_queue_capa(void)
 414 {
 415         return DEV_TX_OFFLOAD_MULTI_SEGS |
 416                DEV_TX_OFFLOAD_IPV4_CKSUM |
 417                DEV_TX_OFFLOAD_UDP_CKSUM |
 418                DEV_TX_OFFLOAD_TCP_CKSUM;
 419 }
 420
 421 static bool
 422 tap_txq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 423 {
 424         uint64_t port_offloads = dev->data->dev_conf.txmode.offloads;
 425         uint64_t queue_supp_offloads = tap_tx_offload_get_queue_capa();
 426         uint64_t port_supp_offloads = tap_tx_offload_get_port_capa();
 427
 428         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 429             offloads)
 430                 return false;
 431         /* Verify we have no conflict with port offloads */
 432         if ((port_offloads ^ offloads) & port_supp_offloads)
 433                 return false;
 434         return true;
 435 }
 436
 437 static void
 438 tap_tx_offload(char *packet, uint64_t ol_flags, unsigned int l2_len,
 439                unsigned int l3_len)
 440 {
 441         void *l3_hdr = packet + l2_len;
 442
 443         if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
 444                 struct ipv4_hdr *iph = l3_hdr;
 445                 uint16_t cksum;
 446
 447                 iph->hdr_checksum = 0;
 448                 cksum = rte_raw_cksum(iph, l3_len);
 449                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 450         }
 451         if (ol_flags & PKT_TX_L4_MASK) {
 452                 uint16_t l4_len;
 453                 uint32_t cksum;
 454                 uint16_t *l4_cksum;
 455                 void *l4_hdr;
 456
 457                 l4_hdr = packet + l2_len + l3_len;
 458                 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
 459                         l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
 460                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 461                         l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
 462                 else
 463                         return;
 464                 *l4_cksum = 0;
 465                 if (ol_flags & PKT_TX_IPV4) {
 466                         struct ipv4_hdr *iph = l3_hdr;
 467
 468                         l4_len = rte_be_to_cpu_16(iph->total_length) - l3_len;
 469                         cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 470                 } else {
 471                         struct ipv6_hdr *ip6h = l3_hdr;
 472
 473                         /* payload_len does not include ext headers */
 474                         l4_len = rte_be_to_cpu_16(ip6h->payload_len) -
 475                                 l3_len + sizeof(struct ipv6_hdr);
 476                         cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 477                 }
 478                 cksum += rte_raw_cksum(l4_hdr, l4_len);
 479                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 480                 cksum = (~cksum) & 0xffff;
 481                 if (cksum == 0)
 482                         cksum = 0xffff;
 483                 *l4_cksum = cksum;
 484         }
 485 }
 486
 487 /* Callback to handle sending packets from the tap interface
 488  */
 489 static uint16_t
 490 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 491 {
 492         struct tx_queue *txq = queue;
 493         uint16_t num_tx = 0;
 494         unsigned long num_tx_bytes = 0;
 495         uint32_t max_size;
 496         int i;
 497
 498         if (unlikely(nb_pkts == 0))
 499                 return 0;
 500
 501         max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
 502         for (i = 0; i < nb_pkts; i++) {
 503                 struct rte_mbuf *mbuf = bufs[num_tx];
 504                 struct iovec iovecs[mbuf->nb_segs + 1];
 505                 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
 506                 struct rte_mbuf *seg = mbuf;
 507                 char m_copy[mbuf->data_len];
 508                 int n;
 509                 int j;
 510
 511                 /* stats.errs will be incremented */
 512                 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
 513                         break;
 514
 515                 /*
 516                  * TUN and TAP are created with IFF_NO_PI disabled.
 517                  * For TUN PMD this mandatory as fields are used by
 518                  * Kernel tun.c to determine whether its IP or non IP
 519                  * packets.
 520                  *
 521                  * The logic fetches the first byte of data from mbuf.
 522                  * compares whether its v4 or v6. If none matches default
 523                  * value 0x00 is taken for protocol field.
 524                  */
 525                 char *buff_data = rte_pktmbuf_mtod(seg, void *);
 526                 j = (*buff_data & 0xf0);
 527                 if (j & (0x40 | 0x60))
 528                         pi.proto = (j == 0x40) ? 0x0008 : 0xdd86;
 529
 530                 iovecs[0].iov_base = &pi;
 531                 iovecs[0].iov_len = sizeof(pi);
 532                 for (j = 1; j <= mbuf->nb_segs; j++) {
 533                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 534                         iovecs[j].iov_base =
 535                                 rte_pktmbuf_mtod(seg, void *);
 536                         seg = seg->next;
 537                 }
 538                 if (txq->csum &&
 539                     ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 540                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 541                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
 542                         /* Support only packets with all data in the same seg */
 543                         if (mbuf->nb_segs > 1)
 544                                 break;
 545                         /* To change checksums, work on a copy of data. */
 546                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 547                                    rte_pktmbuf_data_len(mbuf));
 548                         tap_tx_offload(m_copy, mbuf->ol_flags,
 549                                        mbuf->l2_len, mbuf->l3_len);
 550                         iovecs[1].iov_base = m_copy;
 551                 }
 552                 /* copy the tx frame data */
 553                 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
 554                 if (n <= 0)
 555                         break;
 556
 557                 num_tx++;
 558                 num_tx_bytes += mbuf->pkt_len;
 559                 rte_pktmbuf_free(mbuf);
 560         }
 561
 562         txq->stats.opackets += num_tx;
 563         txq->stats.errs += nb_pkts - num_tx;
 564         txq->stats.obytes += num_tx_bytes;
 565
 566         return num_tx;
 567 }
 568
 569 static const char *
 570 tap_ioctl_req2str(unsigned long request)
 571 {
 572         switch (request) {
 573         case SIOCSIFFLAGS:
 574                 return "SIOCSIFFLAGS";
 575         case SIOCGIFFLAGS:
 576                 return "SIOCGIFFLAGS";
 577         case SIOCGIFHWADDR:
 578                 return "SIOCGIFHWADDR";
 579         case SIOCSIFHWADDR:
 580                 return "SIOCSIFHWADDR";
 581         case SIOCSIFMTU:
 582                 return "SIOCSIFMTU";
 583         }
 584         return "UNKNOWN";
 585 }
 586
 587 static int
 588 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 589           struct ifreq *ifr, int set, enum ioctl_mode mode)
 590 {
 591         short req_flags = ifr->ifr_flags;
 592         int remote = pmd->remote_if_index &&
 593                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 594
 595         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 596                 return 0;
 597         /*
 598          * If there is a remote netdevice, apply ioctl on it, then apply it on
 599          * the tap netdevice.
 600          */
 601 apply:
 602         if (remote)
 603                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
 604         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 605                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 606         switch (request) {
 607         case SIOCSIFFLAGS:
 608                 /* fetch current flags to leave other flags untouched */
 609                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 610                         goto error;
 611                 if (set)
 612                         ifr->ifr_flags |= req_flags;
 613                 else
 614                         ifr->ifr_flags &= ~req_flags;
 615                 break;
 616         case SIOCGIFFLAGS:
 617         case SIOCGIFHWADDR:
 618         case SIOCSIFHWADDR:
 619         case SIOCSIFMTU:
 620                 break;
 621         default:
 622                 RTE_ASSERT(!"unsupported request type: must not happen");
 623         }
 624         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 625                 goto error;
 626         if (remote-- && mode == LOCAL_AND_REMOTE)
 627                 goto apply;
 628         return 0;
 629
 630 error:
 631         RTE_LOG(DEBUG, PMD, "%s: %s(%s) failed: %s(%d)\n", ifr->ifr_name,
 632                 __func__, tap_ioctl_req2str(request), strerror(errno), errno);
 633         return -errno;
 634 }
 635
 636 static int
 637 tap_link_set_down(struct rte_eth_dev *dev)
 638 {
 639         struct pmd_internals *pmd = dev->data->dev_private;
 640         struct ifreq ifr = { .ifr_flags = IFF_UP };
 641
 642         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 643         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 644 }
 645
 646 static int
 647 tap_link_set_up(struct rte_eth_dev *dev)
 648 {
 649         struct pmd_internals *pmd = dev->data->dev_private;
 650         struct ifreq ifr = { .ifr_flags = IFF_UP };
 651
 652         dev->data->dev_link.link_status = ETH_LINK_UP;
 653         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 654 }
 655
 656 static int
 657 tap_dev_start(struct rte_eth_dev *dev)
 658 {
 659         int err;
 660
 661         err = tap_intr_handle_set(dev, 1);
 662         if (err)
 663                 return err;
 664         return tap_link_set_up(dev);
 665 }
 666
 667 /* This function gets called when the current port gets stopped.
 668  */
 669 static void
 670 tap_dev_stop(struct rte_eth_dev *dev)
 671 {
 672         tap_intr_handle_set(dev, 0);
 673         tap_link_set_down(dev);
 674 }
 675
 676 static int
 677 tap_dev_configure(struct rte_eth_dev *dev)
 678 {
 679         uint64_t supp_tx_offloads = tap_tx_offload_get_port_capa();
 680         uint64_t tx_offloads = dev->data->dev_conf.txmode.offloads;
 681
 682         if ((tx_offloads & supp_tx_offloads) != tx_offloads) {
 683                 rte_errno = ENOTSUP;
 684                 RTE_LOG(ERR, PMD,
 685                         "Some Tx offloads are not supported "
 686                         "requested 0x%" PRIx64 " supported 0x%" PRIx64 "\n",
 687                         tx_offloads, supp_tx_offloads);
 688                 return -rte_errno;
 689         }
 690         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 691                 RTE_LOG(ERR, PMD,
 692                         "%s: number of rx queues %d exceeds max num of queues %d\n",
 693                         dev->device->name,
 694                         dev->data->nb_rx_queues,
 695                         RTE_PMD_TAP_MAX_QUEUES);
 696                 return -1;
 697         }
 698         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 699                 RTE_LOG(ERR, PMD,
 700                         "%s: number of tx queues %d exceeds max num of queues %d\n",
 701                         dev->device->name,
 702                         dev->data->nb_tx_queues,
 703                         RTE_PMD_TAP_MAX_QUEUES);
 704                 return -1;
 705         }
 706
 707         RTE_LOG(INFO, PMD, "%s: %p: TX configured queues number: %u\n",
 708              dev->device->name, (void *)dev, dev->data->nb_tx_queues);
 709
 710         RTE_LOG(INFO, PMD, "%s: %p: RX configured queues number: %u\n",
 711              dev->device->name, (void *)dev, dev->data->nb_rx_queues);
 712
 713         return 0;
 714 }
 715
 716 static uint32_t
 717 tap_dev_speed_capa(void)
 718 {
 719         uint32_t speed = pmd_link.link_speed;
 720         uint32_t capa = 0;
 721
 722         if (speed >= ETH_SPEED_NUM_10M)
 723                 capa |= ETH_LINK_SPEED_10M;
 724         if (speed >= ETH_SPEED_NUM_100M)
 725                 capa |= ETH_LINK_SPEED_100M;
 726         if (speed >= ETH_SPEED_NUM_1G)
 727                 capa |= ETH_LINK_SPEED_1G;
 728         if (speed >= ETH_SPEED_NUM_5G)
 729                 capa |= ETH_LINK_SPEED_2_5G;
 730         if (speed >= ETH_SPEED_NUM_5G)
 731                 capa |= ETH_LINK_SPEED_5G;
 732         if (speed >= ETH_SPEED_NUM_10G)
 733                 capa |= ETH_LINK_SPEED_10G;
 734         if (speed >= ETH_SPEED_NUM_20G)
 735                 capa |= ETH_LINK_SPEED_20G;
 736         if (speed >= ETH_SPEED_NUM_25G)
 737                 capa |= ETH_LINK_SPEED_25G;
 738         if (speed >= ETH_SPEED_NUM_40G)
 739                 capa |= ETH_LINK_SPEED_40G;
 740         if (speed >= ETH_SPEED_NUM_50G)
 741                 capa |= ETH_LINK_SPEED_50G;
 742         if (speed >= ETH_SPEED_NUM_56G)
 743                 capa |= ETH_LINK_SPEED_56G;
 744         if (speed >= ETH_SPEED_NUM_100G)
 745                 capa |= ETH_LINK_SPEED_100G;
 746
 747         return capa;
 748 }
 749
 750 static void
 751 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 752 {
 753         struct pmd_internals *internals = dev->data->dev_private;
 754
 755         dev_info->if_index = internals->if_index;
 756         dev_info->max_mac_addrs = 1;
 757         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 758         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
 759         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
 760         dev_info->min_rx_bufsize = 0;
 761         dev_info->pci_dev = NULL;
 762         dev_info->speed_capa = tap_dev_speed_capa();
 763         dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
 764         dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
 765                                     dev_info->rx_queue_offload_capa;
 766         dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
 767         dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
 768                                     dev_info->tx_queue_offload_capa;
 769 }
 770
 771 static int
 772 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 773 {
 774         unsigned int i, imax;
 775         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 776         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 777         unsigned long rx_nombuf = 0, ierrors = 0;
 778         const struct pmd_internals *pmd = dev->data->dev_private;
 779
 780         /* rx queue statistics */
 781         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 782                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 783         for (i = 0; i < imax; i++) {
 784                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 785                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 786                 rx_total += tap_stats->q_ipackets[i];
 787                 rx_bytes_total += tap_stats->q_ibytes[i];
 788                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
 789                 ierrors += pmd->rxq[i].stats.ierrors;
 790         }
 791
 792         /* tx queue statistics */
 793         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 794                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 795
 796         for (i = 0; i < imax; i++) {
 797                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 798                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 799                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 800                 tx_total += tap_stats->q_opackets[i];
 801                 tx_err_total += tap_stats->q_errors[i];
 802                 tx_bytes_total += tap_stats->q_obytes[i];
 803         }
 804
 805         tap_stats->ipackets = rx_total;
 806         tap_stats->ibytes = rx_bytes_total;
 807         tap_stats->ierrors = ierrors;
 808         tap_stats->rx_nombuf = rx_nombuf;
 809         tap_stats->opackets = tx_total;
 810         tap_stats->oerrors = tx_err_total;
 811         tap_stats->obytes = tx_bytes_total;
 812         return 0;
 813 }
 814
 815 static void
 816 tap_stats_reset(struct rte_eth_dev *dev)
 817 {
 818         int i;
 819         struct pmd_internals *pmd = dev->data->dev_private;
 820
 821         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 822                 pmd->rxq[i].stats.ipackets = 0;
 823                 pmd->rxq[i].stats.ibytes = 0;
 824                 pmd->rxq[i].stats.ierrors = 0;
 825                 pmd->rxq[i].stats.rx_nombuf = 0;
 826
 827                 pmd->txq[i].stats.opackets = 0;
 828                 pmd->txq[i].stats.errs = 0;
 829                 pmd->txq[i].stats.obytes = 0;
 830         }
 831 }
 832
 833 static void
 834 tap_dev_close(struct rte_eth_dev *dev)
 835 {
 836         int i;
 837         struct pmd_internals *internals = dev->data->dev_private;
 838
 839         tap_link_set_down(dev);
 840         tap_flow_flush(dev, NULL);
 841         tap_flow_implicit_flush(internals, NULL);
 842
 843         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 844                 if (internals->rxq[i].fd != -1) {
 845                         close(internals->rxq[i].fd);
 846                         internals->rxq[i].fd = -1;
 847                 }
 848                 if (internals->txq[i].fd != -1) {
 849                         close(internals->txq[i].fd);
 850                         internals->txq[i].fd = -1;
 851                 }
 852         }
 853
 854         if (internals->remote_if_index) {
 855                 /* Restore initial remote state */
 856                 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
 857                                 &internals->remote_initial_flags);
 858         }
 859 }
 860
 861 static void
 862 tap_rx_queue_release(void *queue)
 863 {
 864         struct rx_queue *rxq = queue;
 865
 866         if (rxq && (rxq->fd > 0)) {
 867                 close(rxq->fd);
 868                 rxq->fd = -1;
 869                 rte_pktmbuf_free(rxq->pool);
 870                 rte_free(rxq->iovecs);
 871                 rxq->pool = NULL;
 872                 rxq->iovecs = NULL;
 873         }
 874 }
 875
 876 static void
 877 tap_tx_queue_release(void *queue)
 878 {
 879         struct tx_queue *txq = queue;
 880
 881         if (txq && (txq->fd > 0)) {
 882                 close(txq->fd);
 883                 txq->fd = -1;
 884         }
 885 }
 886
 887 static int
 888 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
 889 {
 890         struct rte_eth_link *dev_link = &dev->data->dev_link;
 891         struct pmd_internals *pmd = dev->data->dev_private;
 892         struct ifreq ifr = { .ifr_flags = 0 };
 893
 894         if (pmd->remote_if_index) {
 895                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
 896                 if (!(ifr.ifr_flags & IFF_UP) ||
 897                     !(ifr.ifr_flags & IFF_RUNNING)) {
 898                         dev_link->link_status = ETH_LINK_DOWN;
 899                         return 0;
 900                 }
 901         }
 902         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
 903         dev_link->link_status =
 904                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
 905                  ETH_LINK_UP :
 906                  ETH_LINK_DOWN);
 907         return 0;
 908 }
 909
 910 static void
 911 tap_promisc_enable(struct rte_eth_dev *dev)
 912 {
 913         struct pmd_internals *pmd = dev->data->dev_private;
 914         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 915
 916         dev->data->promiscuous = 1;
 917         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 918         if (pmd->remote_if_index && !pmd->flow_isolate)
 919                 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
 920 }
 921
 922 static void
 923 tap_promisc_disable(struct rte_eth_dev *dev)
 924 {
 925         struct pmd_internals *pmd = dev->data->dev_private;
 926         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 927
 928         dev->data->promiscuous = 0;
 929         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 930         if (pmd->remote_if_index && !pmd->flow_isolate)
 931                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
 932 }
 933
 934 static void
 935 tap_allmulti_enable(struct rte_eth_dev *dev)
 936 {
 937         struct pmd_internals *pmd = dev->data->dev_private;
 938         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 939
 940         dev->data->all_multicast = 1;
 941         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 942         if (pmd->remote_if_index && !pmd->flow_isolate)
 943                 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
 944 }
 945
 946 static void
 947 tap_allmulti_disable(struct rte_eth_dev *dev)
 948 {
 949         struct pmd_internals *pmd = dev->data->dev_private;
 950         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 951
 952         dev->data->all_multicast = 0;
 953         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 954         if (pmd->remote_if_index && !pmd->flow_isolate)
 955                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
 956 }
 957
 958 static void
 959 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 960 {
 961         struct pmd_internals *pmd = dev->data->dev_private;
 962         enum ioctl_mode mode = LOCAL_ONLY;
 963         struct ifreq ifr;
 964
 965         if (is_zero_ether_addr(mac_addr)) {
 966                 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
 967                         dev->device->name);
 968                 return;
 969         }
 970         /* Check the actual current MAC address on the tap netdevice */
 971         if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
 972                 return;
 973         if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 974                                mac_addr))
 975                 return;
 976         /* Check the current MAC address on the remote */
 977         if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0)
 978                 return;
 979         if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 980                                mac_addr))
 981                 mode = LOCAL_AND_REMOTE;
 982         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
 983         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
 984         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode) < 0)
 985                 return;
 986         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
 987         if (pmd->remote_if_index && !pmd->flow_isolate) {
 988                 /* Replace MAC redirection rule after a MAC change */
 989                 if (tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC) < 0) {
 990                         RTE_LOG(ERR, PMD,
 991                                 "%s: Couldn't delete MAC redirection rule\n",
 992                                 dev->device->name);
 993                         return;
 994                 }
 995                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0)
 996                         RTE_LOG(ERR, PMD,
 997                                 "%s: Couldn't add MAC redirection rule\n",
 998                                 dev->device->name);
 999         }
1000 }
1001
1002 static int
1003 tap_setup_queue(struct rte_eth_dev *dev,
1004                 struct pmd_internals *internals,
1005                 uint16_t qid,
1006                 int is_rx)
1007 {
1008         int *fd;
1009         int *other_fd;
1010         const char *dir;
1011         struct pmd_internals *pmd = dev->data->dev_private;
1012         struct rx_queue *rx = &internals->rxq[qid];
1013         struct tx_queue *tx = &internals->txq[qid];
1014
1015         if (is_rx) {
1016                 fd = &rx->fd;
1017                 other_fd = &tx->fd;
1018                 dir = "rx";
1019         } else {
1020                 fd = &tx->fd;
1021                 other_fd = &rx->fd;
1022                 dir = "tx";
1023         }
1024         if (*fd != -1) {
1025                 /* fd for this queue already exists */
1026                 RTE_LOG(DEBUG, PMD, "%s: fd %d for %s queue qid %d exists\n",
1027                         pmd->name, *fd, dir, qid);
1028         } else if (*other_fd != -1) {
1029                 /* Only other_fd exists. dup it */
1030                 *fd = dup(*other_fd);
1031                 if (*fd < 0) {
1032                         *fd = -1;
1033                         RTE_LOG(ERR, PMD, "%s: dup() failed.\n",
1034                                 pmd->name);
1035                         return -1;
1036                 }
1037                 RTE_LOG(DEBUG, PMD, "%s: dup fd %d for %s queue qid %d (%d)\n",
1038                         pmd->name, *other_fd, dir, qid, *fd);
1039         } else {
1040                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1041                 *fd = tun_alloc(pmd);
1042                 if (*fd < 0) {
1043                         *fd = -1; /* restore original value */
1044                         RTE_LOG(ERR, PMD, "%s: tun_alloc() failed.\n",
1045                                 pmd->name);
1046                         return -1;
1047                 }
1048                 RTE_LOG(DEBUG, PMD, "%s: add %s queue for qid %d fd %d\n",
1049                         pmd->name, dir, qid, *fd);
1050         }
1051
1052         tx->mtu = &dev->data->mtu;
1053         rx->rxmode = &dev->data->dev_conf.rxmode;
1054
1055         return *fd;
1056 }
1057
1058 static int
1059 tap_rx_queue_setup(struct rte_eth_dev *dev,
1060                    uint16_t rx_queue_id,
1061                    uint16_t nb_rx_desc,
1062                    unsigned int socket_id,
1063                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1064                    struct rte_mempool *mp)
1065 {
1066         struct pmd_internals *internals = dev->data->dev_private;
1067         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1068         struct rte_mbuf **tmp = &rxq->pool;
1069         long iov_max = sysconf(_SC_IOV_MAX);
1070         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1071         struct iovec (*iovecs)[nb_desc + 1];
1072         int data_off = RTE_PKTMBUF_HEADROOM;
1073         int ret = 0;
1074         int fd;
1075         int i;
1076
1077         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1078                 RTE_LOG(WARNING, PMD,
1079                         "nb_rx_queues %d too small or mempool NULL\n",
1080                         dev->data->nb_rx_queues);
1081                 return -1;
1082         }
1083
1084         /* Verify application offloads are valid for our port and queue. */
1085         if (!tap_rxq_are_offloads_valid(dev, rx_conf->offloads)) {
1086                 rte_errno = ENOTSUP;
1087                 RTE_LOG(ERR, PMD,
1088                         "%p: Rx queue offloads 0x%" PRIx64
1089                         " don't match port offloads 0x%" PRIx64
1090                         " or supported offloads 0x%" PRIx64 "\n",
1091                         (void *)dev, rx_conf->offloads,
1092                         dev->data->dev_conf.rxmode.offloads,
1093                         (tap_rx_offload_get_port_capa() |
1094                          tap_rx_offload_get_queue_capa()));
1095                 return -rte_errno;
1096         }
1097         rxq->mp = mp;
1098         rxq->trigger_seen = 1; /* force initial burst */
1099         rxq->in_port = dev->data->port_id;
1100         rxq->nb_rx_desc = nb_desc;
1101         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1102                                     socket_id);
1103         if (!iovecs) {
1104                 RTE_LOG(WARNING, PMD,
1105                         "%s: Couldn't allocate %d RX descriptors\n",
1106                         dev->device->name, nb_desc);
1107                 return -ENOMEM;
1108         }
1109         rxq->iovecs = iovecs;
1110
1111         dev->data->rx_queues[rx_queue_id] = rxq;
1112         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1113         if (fd == -1) {
1114                 ret = fd;
1115                 goto error;
1116         }
1117
1118         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1119         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1120
1121         for (i = 1; i <= nb_desc; i++) {
1122                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1123                 if (!*tmp) {
1124                         RTE_LOG(WARNING, PMD,
1125                                 "%s: couldn't allocate memory for queue %d\n",
1126                                 dev->device->name, rx_queue_id);
1127                         ret = -ENOMEM;
1128                         goto error;
1129                 }
1130                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1131                 (*rxq->iovecs)[i].iov_base =
1132                         (char *)(*tmp)->buf_addr + data_off;
1133                 data_off = 0;
1134                 tmp = &(*tmp)->next;
1135         }
1136
1137         RTE_LOG(DEBUG, PMD, "  RX TAP device name %s, qid %d on fd %d\n",
1138                 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
1139
1140         return 0;
1141
1142 error:
1143         rte_pktmbuf_free(rxq->pool);
1144         rxq->pool = NULL;
1145         rte_free(rxq->iovecs);
1146         rxq->iovecs = NULL;
1147         return ret;
1148 }
1149
1150 static int
1151 tap_tx_queue_setup(struct rte_eth_dev *dev,
1152                    uint16_t tx_queue_id,
1153                    uint16_t nb_tx_desc __rte_unused,
1154                    unsigned int socket_id __rte_unused,
1155                    const struct rte_eth_txconf *tx_conf)
1156 {
1157         struct pmd_internals *internals = dev->data->dev_private;
1158         struct tx_queue *txq;
1159         int ret;
1160
1161         if (tx_queue_id >= dev->data->nb_tx_queues)
1162                 return -1;
1163         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1164         txq = dev->data->tx_queues[tx_queue_id];
1165         /*
1166          * Don't verify port offloads for application which
1167          * use the old API.
1168          */
1169         if (tx_conf != NULL &&
1170             !!(tx_conf->txq_flags & ETH_TXQ_FLAGS_IGNORE)) {
1171                 if (tap_txq_are_offloads_valid(dev, tx_conf->offloads)) {
1172                         txq->csum = !!(tx_conf->offloads &
1173                                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
1174                                          DEV_TX_OFFLOAD_UDP_CKSUM |
1175                                          DEV_TX_OFFLOAD_TCP_CKSUM));
1176                 } else {
1177                         rte_errno = ENOTSUP;
1178                         RTE_LOG(ERR, PMD,
1179                                 "%p: Tx queue offloads 0x%" PRIx64
1180                                 " don't match port offloads 0x%" PRIx64
1181                                 " or supported offloads 0x%" PRIx64,
1182                                 (void *)dev, tx_conf->offloads,
1183                                 dev->data->dev_conf.txmode.offloads,
1184                                 tap_tx_offload_get_port_capa());
1185                         return -rte_errno;
1186                 }
1187         }
1188         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1189         if (ret == -1)
1190                 return -1;
1191         RTE_LOG(DEBUG, PMD,
1192                 "  TX TAP device name %s, qid %d on fd %d csum %s\n",
1193                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd,
1194                 txq->csum ? "on" : "off");
1195
1196         return 0;
1197 }
1198
1199 static int
1200 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1201 {
1202         struct pmd_internals *pmd = dev->data->dev_private;
1203         struct ifreq ifr = { .ifr_mtu = mtu };
1204         int err = 0;
1205
1206         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1207         if (!err)
1208                 dev->data->mtu = mtu;
1209
1210         return err;
1211 }
1212
1213 static int
1214 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1215                      struct ether_addr *mc_addr_set __rte_unused,
1216                      uint32_t nb_mc_addr __rte_unused)
1217 {
1218         /*
1219          * Nothing to do actually: the tap has no filtering whatsoever, every
1220          * packet is received.
1221          */
1222         return 0;
1223 }
1224
1225 static int
1226 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1227 {
1228         struct rte_eth_dev *dev = arg;
1229         struct pmd_internals *pmd = dev->data->dev_private;
1230         struct ifinfomsg *info = NLMSG_DATA(nh);
1231
1232         if (nh->nlmsg_type != RTM_NEWLINK ||
1233             (info->ifi_index != pmd->if_index &&
1234              info->ifi_index != pmd->remote_if_index))
1235                 return 0;
1236         return tap_link_update(dev, 0);
1237 }
1238
1239 static void
1240 tap_dev_intr_handler(void *cb_arg)
1241 {
1242         struct rte_eth_dev *dev = cb_arg;
1243         struct pmd_internals *pmd = dev->data->dev_private;
1244
1245         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1246 }
1247
1248 static int
1249 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1250 {
1251         struct pmd_internals *pmd = dev->data->dev_private;
1252
1253         /* In any case, disable interrupt if the conf is no longer there. */
1254         if (!dev->data->dev_conf.intr_conf.lsc) {
1255                 if (pmd->intr_handle.fd != -1) {
1256                         tap_nl_final(pmd->intr_handle.fd);
1257                         rte_intr_callback_unregister(&pmd->intr_handle,
1258                                 tap_dev_intr_handler, dev);
1259                 }
1260                 return 0;
1261         }
1262         if (set) {
1263                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1264                 if (unlikely(pmd->intr_handle.fd == -1))
1265                         return -EBADF;
1266                 return rte_intr_callback_register(
1267                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1268         }
1269         tap_nl_final(pmd->intr_handle.fd);
1270         return rte_intr_callback_unregister(&pmd->intr_handle,
1271                                             tap_dev_intr_handler, dev);
1272 }
1273
1274 static int
1275 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1276 {
1277         int err;
1278
1279         err = tap_lsc_intr_handle_set(dev, set);
1280         if (err)
1281                 return err;
1282         err = tap_rx_intr_vec_set(dev, set);
1283         if (err && set)
1284                 tap_lsc_intr_handle_set(dev, 0);
1285         return err;
1286 }
1287
1288 static const uint32_t*
1289 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1290 {
1291         static const uint32_t ptypes[] = {
1292                 RTE_PTYPE_INNER_L2_ETHER,
1293                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1294                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1295                 RTE_PTYPE_INNER_L3_IPV4,
1296                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1297                 RTE_PTYPE_INNER_L3_IPV6,
1298                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1299                 RTE_PTYPE_INNER_L4_FRAG,
1300                 RTE_PTYPE_INNER_L4_UDP,
1301                 RTE_PTYPE_INNER_L4_TCP,
1302                 RTE_PTYPE_INNER_L4_SCTP,
1303                 RTE_PTYPE_L2_ETHER,
1304                 RTE_PTYPE_L2_ETHER_VLAN,
1305                 RTE_PTYPE_L2_ETHER_QINQ,
1306                 RTE_PTYPE_L3_IPV4,
1307                 RTE_PTYPE_L3_IPV4_EXT,
1308                 RTE_PTYPE_L3_IPV6_EXT,
1309                 RTE_PTYPE_L3_IPV6,
1310                 RTE_PTYPE_L4_FRAG,
1311                 RTE_PTYPE_L4_UDP,
1312                 RTE_PTYPE_L4_TCP,
1313                 RTE_PTYPE_L4_SCTP,
1314         };
1315
1316         return ptypes;
1317 }
1318
1319 static int
1320 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1321                   struct rte_eth_fc_conf *fc_conf)
1322 {
1323         fc_conf->mode = RTE_FC_NONE;
1324         return 0;
1325 }
1326
1327 static int
1328 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1329                   struct rte_eth_fc_conf *fc_conf)
1330 {
1331         if (fc_conf->mode != RTE_FC_NONE)
1332                 return -ENOTSUP;
1333         return 0;
1334 }
1335
1336 static const struct eth_dev_ops ops = {
1337         .dev_start              = tap_dev_start,
1338         .dev_stop               = tap_dev_stop,
1339         .dev_close              = tap_dev_close,
1340         .dev_configure          = tap_dev_configure,
1341         .dev_infos_get          = tap_dev_info,
1342         .rx_queue_setup         = tap_rx_queue_setup,
1343         .tx_queue_setup         = tap_tx_queue_setup,
1344         .rx_queue_release       = tap_rx_queue_release,
1345         .tx_queue_release       = tap_tx_queue_release,
1346         .flow_ctrl_get          = tap_flow_ctrl_get,
1347         .flow_ctrl_set          = tap_flow_ctrl_set,
1348         .link_update            = tap_link_update,
1349         .dev_set_link_up        = tap_link_set_up,
1350         .dev_set_link_down      = tap_link_set_down,
1351         .promiscuous_enable     = tap_promisc_enable,
1352         .promiscuous_disable    = tap_promisc_disable,
1353         .allmulticast_enable    = tap_allmulti_enable,
1354         .allmulticast_disable   = tap_allmulti_disable,
1355         .mac_addr_set           = tap_mac_set,
1356         .mtu_set                = tap_mtu_set,
1357         .set_mc_addr_list       = tap_set_mc_addr_list,
1358         .stats_get              = tap_stats_get,
1359         .stats_reset            = tap_stats_reset,
1360         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1361         .filter_ctrl            = tap_dev_filter_ctrl,
1362 };
1363
1364 static int
1365 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1366                    char *remote_iface, struct ether_addr *mac_addr)
1367 {
1368         int numa_node = rte_socket_id();
1369         struct rte_eth_dev *dev;
1370         struct pmd_internals *pmd;
1371         struct rte_eth_dev_data *data;
1372         struct ifreq ifr;
1373         int i;
1374
1375         RTE_LOG(DEBUG, PMD, "  TAP device on numa %u\n", rte_socket_id());
1376
1377         data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
1378         if (!data) {
1379                 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
1380                 goto error_exit_nodev;
1381         }
1382
1383         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1384         if (!dev) {
1385                 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
1386                 goto error_exit_nodev;
1387         }
1388
1389         pmd = dev->data->dev_private;
1390         pmd->dev = dev;
1391         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1392
1393         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1394         if (pmd->ioctl_sock == -1) {
1395                 RTE_LOG(ERR, PMD,
1396                         "TAP Unable to get a socket for management: %s\n",
1397                         strerror(errno));
1398                 goto error_exit;
1399         }
1400
1401         /* Setup some default values */
1402         rte_memcpy(data, dev->data, sizeof(*data));
1403         data->dev_private = pmd;
1404         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1405         data->numa_node = numa_node;
1406
1407         data->dev_link = pmd_link;
1408         data->mac_addrs = &pmd->eth_addr;
1409         /* Set the number of RX and TX queues */
1410         data->nb_rx_queues = 0;
1411         data->nb_tx_queues = 0;
1412
1413         dev->data = data;
1414         dev->dev_ops = &ops;
1415         dev->rx_pkt_burst = pmd_rx_burst;
1416         dev->tx_pkt_burst = pmd_tx_burst;
1417
1418         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1419         pmd->intr_handle.fd = -1;
1420         dev->intr_handle = &pmd->intr_handle;
1421
1422         /* Presetup the fds to -1 as being not valid */
1423         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1424                 pmd->rxq[i].fd = -1;
1425                 pmd->txq[i].fd = -1;
1426         }
1427
1428         if (tap_type) {
1429                 if (is_zero_ether_addr(mac_addr))
1430                         eth_random_addr((uint8_t *)&pmd->eth_addr);
1431                 else
1432                         rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1433         }
1434
1435         /* Immediately create the netdevice (this will create the 1st queue). */
1436         /* rx queue */
1437         if (tap_setup_queue(dev, pmd, 0, 1) == -1)
1438                 goto error_exit;
1439         /* tx queue */
1440         if (tap_setup_queue(dev, pmd, 0, 0) == -1)
1441                 goto error_exit;
1442
1443         ifr.ifr_mtu = dev->data->mtu;
1444         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1445                 goto error_exit;
1446
1447         if (tap_type) {
1448                 memset(&ifr, 0, sizeof(struct ifreq));
1449                 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1450                 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1451                                 ETHER_ADDR_LEN);
1452                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1453                         goto error_exit;
1454         }
1455
1456         /*
1457          * Set up everything related to rte_flow:
1458          * - netlink socket
1459          * - tap / remote if_index
1460          * - mandatory QDISCs
1461          * - rte_flow actual/implicit lists
1462          * - implicit rules
1463          */
1464         pmd->nlsk_fd = tap_nl_init(0);
1465         if (pmd->nlsk_fd == -1) {
1466                 RTE_LOG(WARNING, PMD, "%s: failed to create netlink socket.\n",
1467                         pmd->name);
1468                 goto disable_rte_flow;
1469         }
1470         pmd->if_index = if_nametoindex(pmd->name);
1471         if (!pmd->if_index) {
1472                 RTE_LOG(ERR, PMD, "%s: failed to get if_index.\n", pmd->name);
1473                 goto disable_rte_flow;
1474         }
1475         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1476                 RTE_LOG(ERR, PMD, "%s: failed to create multiq qdisc.\n",
1477                         pmd->name);
1478                 goto disable_rte_flow;
1479         }
1480         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1481                 RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1482                         pmd->name);
1483                 goto disable_rte_flow;
1484         }
1485         LIST_INIT(&pmd->flows);
1486
1487         if (strlen(remote_iface)) {
1488                 pmd->remote_if_index = if_nametoindex(remote_iface);
1489                 if (!pmd->remote_if_index) {
1490                         RTE_LOG(ERR, PMD, "%s: failed to get %s if_index.\n",
1491                                 pmd->name, remote_iface);
1492                         goto error_remote;
1493                 }
1494                 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1495                          "%s", remote_iface);
1496
1497                 /* Save state of remote device */
1498                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1499
1500                 /* Replicate remote MAC address */
1501                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1502                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1503                                 pmd->name, pmd->remote_iface);
1504                         goto error_remote;
1505                 }
1506                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1507                            ETHER_ADDR_LEN);
1508                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1509                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1510                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1511                                 pmd->name, remote_iface);
1512                         goto error_remote;
1513                 }
1514
1515                 /*
1516                  * Flush usually returns negative value because it tries to
1517                  * delete every QDISC (and on a running device, one QDISC at
1518                  * least is needed). Ignore negative return value.
1519                  */
1520                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1521                 if (qdisc_create_ingress(pmd->nlsk_fd,
1522                                          pmd->remote_if_index) < 0) {
1523                         RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1524                                 pmd->remote_iface);
1525                         goto error_remote;
1526                 }
1527                 LIST_INIT(&pmd->implicit_flows);
1528                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1529                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1530                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1531                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1532                         RTE_LOG(ERR, PMD,
1533                                 "%s: failed to create implicit rules.\n",
1534                                 pmd->name);
1535                         goto error_remote;
1536                 }
1537         }
1538
1539         return 0;
1540
1541 disable_rte_flow:
1542         RTE_LOG(ERR, PMD, " Disabling rte flow support: %s(%d)\n",
1543                 strerror(errno), errno);
1544         if (strlen(remote_iface)) {
1545                 RTE_LOG(ERR, PMD, "Remote feature requires flow support.\n");
1546                 goto error_exit;
1547         }
1548         return 0;
1549
1550 error_remote:
1551         RTE_LOG(ERR, PMD, " Can't set up remote feature: %s(%d)\n",
1552                 strerror(errno), errno);
1553         tap_flow_implicit_flush(pmd, NULL);
1554
1555 error_exit:
1556         if (pmd->ioctl_sock > 0)
1557                 close(pmd->ioctl_sock);
1558         rte_eth_dev_release_port(dev);
1559
1560 error_exit_nodev:
1561         RTE_LOG(ERR, PMD, "TAP Unable to initialize %s\n",
1562                 rte_vdev_device_name(vdev));
1563
1564         rte_free(data);
1565         return -EINVAL;
1566 }
1567
1568 static int
1569 set_interface_name(const char *key __rte_unused,
1570                    const char *value,
1571                    void *extra_args)
1572 {
1573         char *name = (char *)extra_args;
1574
1575         if (value)
1576                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN - 1);
1577         else
1578                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1579                          DEFAULT_TAP_NAME, (tap_unit - 1));
1580
1581         return 0;
1582 }
1583
1584 static int
1585 set_remote_iface(const char *key __rte_unused,
1586                  const char *value,
1587                  void *extra_args)
1588 {
1589         char *name = (char *)extra_args;
1590
1591         if (value)
1592                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
1593
1594         return 0;
1595 }
1596
1597 static int parse_user_mac(struct ether_addr *user_mac,
1598                 const char *value)
1599 {
1600         unsigned int index = 0;
1601         char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
1602
1603         if (user_mac == NULL || value == NULL)
1604                 return 0;
1605
1606         strlcpy(mac_temp, value, sizeof(mac_temp));
1607         mac_byte = strtok(mac_temp, ":");
1608
1609         while ((mac_byte != NULL) &&
1610                         (strlen(mac_byte) <= 2) &&
1611                         (strlen(mac_byte) == strspn(mac_byte,
1612                                         ETH_TAP_CMP_MAC_FMT))) {
1613                 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
1614                 mac_byte = strtok(NULL, ":");
1615         }
1616
1617         return index;
1618 }
1619
1620 static int
1621 set_mac_type(const char *key __rte_unused,
1622              const char *value,
1623              void *extra_args)
1624 {
1625         struct ether_addr *user_mac = extra_args;
1626
1627         if (!value)
1628                 return 0;
1629
1630         if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
1631                 static int iface_idx;
1632
1633                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1634                 memcpy((char *)user_mac->addr_bytes, "\0dtap", ETHER_ADDR_LEN);
1635                 user_mac->addr_bytes[ETHER_ADDR_LEN - 1] = iface_idx++ + '0';
1636                 goto success;
1637         }
1638
1639         if (parse_user_mac(user_mac, value) != 6)
1640                 goto error;
1641 success:
1642         RTE_LOG(DEBUG, PMD, "TAP user MAC param (%s)\n", value);
1643         return 0;
1644
1645 error:
1646         RTE_LOG(ERR, PMD, "TAP user MAC (%s) is not in format (%s|%s)\n",
1647                 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
1648         return -1;
1649 }
1650
1651 /*
1652  * Open a TUN interface device. TUN PMD
1653  * 1) sets tap_type as false
1654  * 2) intakes iface as argument.
1655  * 3) as interface is virtual set speed to 10G
1656  */
1657 static int
1658 rte_pmd_tun_probe(struct rte_vdev_device *dev)
1659 {
1660         const char *name, *params;
1661         int ret;
1662         struct rte_kvargs *kvlist = NULL;
1663         char tun_name[RTE_ETH_NAME_MAX_LEN];
1664         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1665
1666         tap_type = 0;
1667         strcpy(tuntap_name, "TUN");
1668
1669         name = rte_vdev_device_name(dev);
1670         params = rte_vdev_device_args(dev);
1671         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1672
1673         if (params && (params[0] != '\0')) {
1674                 RTE_LOG(DEBUG, PMD, "parameters (%s)\n", params);
1675
1676                 kvlist = rte_kvargs_parse(params, valid_arguments);
1677                 if (kvlist) {
1678                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1679                                 ret = rte_kvargs_process(kvlist,
1680                                         ETH_TAP_IFACE_ARG,
1681                                         &set_interface_name,
1682                                         tun_name);
1683
1684                                 if (ret == -1)
1685                                         goto leave;
1686                         }
1687                 }
1688         }
1689         pmd_link.link_speed = ETH_SPEED_NUM_10G;
1690
1691         RTE_LOG(NOTICE, PMD, "Initializing pmd_tun for %s as %s\n",
1692                 name, tun_name);
1693
1694         ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0);
1695
1696 leave:
1697         if (ret == -1) {
1698                 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1699                         name, tun_name);
1700                 tun_unit--; /* Restore the unit number */
1701         }
1702         rte_kvargs_free(kvlist);
1703
1704         return ret;
1705 }
1706
1707 /* Open a TAP interface device.
1708  */
1709 static int
1710 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1711 {
1712         const char *name, *params;
1713         int ret;
1714         struct rte_kvargs *kvlist = NULL;
1715         int speed;
1716         char tap_name[RTE_ETH_NAME_MAX_LEN];
1717         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1718         struct ether_addr user_mac = { .addr_bytes = {0} };
1719
1720         tap_type = 1;
1721         strcpy(tuntap_name, "TAP");
1722
1723         name = rte_vdev_device_name(dev);
1724         params = rte_vdev_device_args(dev);
1725
1726         speed = ETH_SPEED_NUM_10G;
1727         snprintf(tap_name, sizeof(tap_name), "%s%d",
1728                  DEFAULT_TAP_NAME, tap_unit++);
1729         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1730
1731         if (params && (params[0] != '\0')) {
1732                 RTE_LOG(DEBUG, PMD, "parameters (%s)\n", params);
1733
1734                 kvlist = rte_kvargs_parse(params, valid_arguments);
1735                 if (kvlist) {
1736                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1737                                 ret = rte_kvargs_process(kvlist,
1738                                                          ETH_TAP_IFACE_ARG,
1739                                                          &set_interface_name,
1740                                                          tap_name);
1741                                 if (ret == -1)
1742                                         goto leave;
1743                         }
1744
1745                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1746                                 ret = rte_kvargs_process(kvlist,
1747                                                          ETH_TAP_REMOTE_ARG,
1748                                                          &set_remote_iface,
1749                                                          remote_iface);
1750                                 if (ret == -1)
1751                                         goto leave;
1752                         }
1753
1754                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1755                                 ret = rte_kvargs_process(kvlist,
1756                                                          ETH_TAP_MAC_ARG,
1757                                                          &set_mac_type,
1758                                                          &user_mac);
1759                                 if (ret == -1)
1760                                         goto leave;
1761                         }
1762                 }
1763         }
1764         pmd_link.link_speed = speed;
1765
1766         RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1767                 name, tap_name);
1768
1769         ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac);
1770
1771 leave:
1772         if (ret == -1) {
1773                 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1774                         name, tap_name);
1775                 tap_unit--;             /* Restore the unit number */
1776         }
1777         rte_kvargs_free(kvlist);
1778
1779         return ret;
1780 }
1781
1782 /* detach a TUNTAP device.
1783  */
1784 static int
1785 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1786 {
1787         struct rte_eth_dev *eth_dev = NULL;
1788         struct pmd_internals *internals;
1789         int i;
1790
1791         RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1792                 rte_socket_id());
1793
1794         /* find the ethdev entry */
1795         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1796         if (!eth_dev)
1797                 return 0;
1798
1799         internals = eth_dev->data->dev_private;
1800         if (internals->nlsk_fd) {
1801                 tap_flow_flush(eth_dev, NULL);
1802                 tap_flow_implicit_flush(internals, NULL);
1803                 tap_nl_final(internals->nlsk_fd);
1804         }
1805         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1806                 if (internals->rxq[i].fd != -1) {
1807                         close(internals->rxq[i].fd);
1808                         internals->rxq[i].fd = -1;
1809                 }
1810                 if (internals->txq[i].fd != -1) {
1811                         close(internals->txq[i].fd);
1812                         internals->txq[i].fd = -1;
1813                 }
1814         }
1815
1816         close(internals->ioctl_sock);
1817         rte_free(eth_dev->data->dev_private);
1818         rte_free(eth_dev->data);
1819
1820         rte_eth_dev_release_port(eth_dev);
1821
1822         return 0;
1823 }
1824
1825 static struct rte_vdev_driver pmd_tun_drv = {
1826         .probe = rte_pmd_tun_probe,
1827         .remove = rte_pmd_tap_remove,
1828 };
1829
1830 static struct rte_vdev_driver pmd_tap_drv = {
1831         .probe = rte_pmd_tap_probe,
1832         .remove = rte_pmd_tap_remove,
1833 };
1834 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1835 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
1836 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1837 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
1838                               ETH_TAP_IFACE_ARG "=<string> ");
1839 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1840                               ETH_TAP_IFACE_ARG "=<string> "
1841                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
1842                               ETH_TAP_REMOTE_ARG "=<string>");