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_FIXED,
  80 };
  81
  82 static void
  83 tap_trigger_cb(int sig __rte_unused)
  84 {
  85         /* Valid trigger values are nonzero */
  86         tap_trigger = (tap_trigger + 1) | 0x80000000;
  87 }
  88
  89 /* Specifies on what netdevices the ioctl should be applied */
  90 enum ioctl_mode {
  91         LOCAL_AND_REMOTE,
  92         LOCAL_ONLY,
  93         REMOTE_ONLY,
  94 };
  95
  96 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
  97
  98 /* Tun/Tap allocation routine
  99  *
 100  * name is the number of the interface to use, unless NULL to take the host
 101  * supplied name.
 102  */
 103 static int
 104 tun_alloc(struct pmd_internals *pmd)
 105 {
 106         struct ifreq ifr;
 107 #ifdef IFF_MULTI_QUEUE
 108         unsigned int features;
 109 #endif
 110         int fd;
 111
 112         memset(&ifr, 0, sizeof(struct ifreq));
 113
 114         /*
 115          * Do not set IFF_NO_PI as packet information header will be needed
 116          * to check if a received packet has been truncated.
 117          */
 118         ifr.ifr_flags = (tap_type) ? IFF_TAP : IFF_TUN | IFF_POINTOPOINT;
 119         snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
 120
 121         TAP_LOG(DEBUG, "ifr_name '%s'", ifr.ifr_name);
 122
 123         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 124         if (fd < 0) {
 125                 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
 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                 TAP_LOG(ERR, "%s unable to get TUN/TAP features",
 133                         tuntap_name);
 134                 goto error;
 135         }
 136         TAP_LOG(DEBUG, "%s Features %08x", tuntap_name, features);
 137
 138         if (features & IFF_MULTI_QUEUE) {
 139                 TAP_LOG(DEBUG, "  Multi-queue support for %d queues",
 140                         RTE_PMD_TAP_MAX_QUEUES);
 141                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 142         } else
 143 #endif
 144         {
 145                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 146                 TAP_LOG(DEBUG, "  Single queue only support");
 147         }
 148
 149         /* Set the TUN/TAP configuration and set the name if needed */
 150         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 151                 TAP_LOG(WARNING, "Unable to set TUNSETIFF for %s: %s",
 152                         ifr.ifr_name, strerror(errno));
 153                 goto error;
 154         }
 155
 156         /* Always set the file descriptor to non-blocking */
 157         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 158                 TAP_LOG(WARNING,
 159                         "Unable to set %s to nonblocking: %s",
 160                         ifr.ifr_name, strerror(errno));
 161                 goto error;
 162         }
 163
 164         /* Set up trigger to optimize empty Rx bursts */
 165         errno = 0;
 166         do {
 167                 struct sigaction sa;
 168                 int flags = fcntl(fd, F_GETFL);
 169
 170                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 171                         break;
 172                 if (sa.sa_handler != tap_trigger_cb) {
 173                         /*
 174                          * Make sure SIGIO is not already taken. This is done
 175                          * as late as possible to leave the application a
 176                          * chance to set up its own signal handler first.
 177                          */
 178                         if (sa.sa_handler != SIG_IGN &&
 179                             sa.sa_handler != SIG_DFL) {
 180                                 errno = EBUSY;
 181                                 break;
 182                         }
 183                         sa = (struct sigaction){
 184                                 .sa_flags = SA_RESTART,
 185                                 .sa_handler = tap_trigger_cb,
 186                         };
 187                         if (sigaction(SIGIO, &sa, NULL) == -1)
 188                                 break;
 189                 }
 190                 /* Enable SIGIO on file descriptor */
 191                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 192                 fcntl(fd, F_SETOWN, getpid());
 193         } while (0);
 194
 195         if (errno) {
 196                 /* Disable trigger globally in case of error */
 197                 tap_trigger = 0;
 198                 TAP_LOG(WARNING, "Rx trigger disabled: %s",
 199                         strerror(errno));
 200         }
 201
 202         return fd;
 203
 204 error:
 205         if (fd > 0)
 206                 close(fd);
 207         return -1;
 208 }
 209
 210 static void
 211 tap_verify_csum(struct rte_mbuf *mbuf)
 212 {
 213         uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
 214         uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
 215         uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
 216         unsigned int l2_len = sizeof(struct ether_hdr);
 217         unsigned int l3_len;
 218         uint16_t cksum = 0;
 219         void *l3_hdr;
 220         void *l4_hdr;
 221
 222         if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
 223                 l2_len += 4;
 224         else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
 225                 l2_len += 8;
 226         /* Don't verify checksum for packets with discontinuous L2 header */
 227         if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
 228                      rte_pktmbuf_data_len(mbuf)))
 229                 return;
 230         l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
 231         if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 232                 struct ipv4_hdr *iph = l3_hdr;
 233
 234                 /* ihl contains the number of 4-byte words in the header */
 235                 l3_len = 4 * (iph->version_ihl & 0xf);
 236                 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
 237                         return;
 238
 239                 cksum = ~rte_raw_cksum(iph, l3_len);
 240                 mbuf->ol_flags |= cksum ?
 241                         PKT_RX_IP_CKSUM_BAD :
 242                         PKT_RX_IP_CKSUM_GOOD;
 243         } else if (l3 == RTE_PTYPE_L3_IPV6) {
 244                 l3_len = sizeof(struct ipv6_hdr);
 245         } else {
 246                 /* IPv6 extensions are not supported */
 247                 return;
 248         }
 249         if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
 250                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 251                 /* Don't verify checksum for multi-segment packets. */
 252                 if (mbuf->nb_segs > 1)
 253                         return;
 254                 if (l3 == RTE_PTYPE_L3_IPV4)
 255                         cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
 256                 else if (l3 == RTE_PTYPE_L3_IPV6)
 257                         cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
 258                 mbuf->ol_flags |= cksum ?
 259                         PKT_RX_L4_CKSUM_BAD :
 260                         PKT_RX_L4_CKSUM_GOOD;
 261         }
 262 }
 263
 264 static uint64_t
 265 tap_rx_offload_get_port_capa(void)
 266 {
 267         /*
 268          * No specific port Rx offload capabilities.
 269          */
 270         return 0;
 271 }
 272
 273 static uint64_t
 274 tap_rx_offload_get_queue_capa(void)
 275 {
 276         return DEV_RX_OFFLOAD_SCATTER |
 277                DEV_RX_OFFLOAD_IPV4_CKSUM |
 278                DEV_RX_OFFLOAD_UDP_CKSUM |
 279                DEV_RX_OFFLOAD_TCP_CKSUM |
 280                DEV_RX_OFFLOAD_CRC_STRIP;
 281 }
 282
 283 static bool
 284 tap_rxq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 285 {
 286         uint64_t port_offloads = dev->data->dev_conf.rxmode.offloads;
 287         uint64_t queue_supp_offloads = tap_rx_offload_get_queue_capa();
 288         uint64_t port_supp_offloads = tap_rx_offload_get_port_capa();
 289
 290         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 291             offloads)
 292                 return false;
 293         if ((port_offloads ^ offloads) & port_supp_offloads)
 294                 return false;
 295         return true;
 296 }
 297
 298 /* Callback to handle the rx burst of packets to the correct interface and
 299  * file descriptor(s) in a multi-queue setup.
 300  */
 301 static uint16_t
 302 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 303 {
 304         struct rx_queue *rxq = queue;
 305         uint16_t num_rx;
 306         unsigned long num_rx_bytes = 0;
 307         uint32_t trigger = tap_trigger;
 308
 309         if (trigger == rxq->trigger_seen)
 310                 return 0;
 311         if (trigger)
 312                 rxq->trigger_seen = trigger;
 313         rte_compiler_barrier();
 314         for (num_rx = 0; num_rx < nb_pkts; ) {
 315                 struct rte_mbuf *mbuf = rxq->pool;
 316                 struct rte_mbuf *seg = NULL;
 317                 struct rte_mbuf *new_tail = NULL;
 318                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 319                 int len;
 320
 321                 len = readv(rxq->fd, *rxq->iovecs,
 322                             1 +
 323                             (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
 324                              rxq->nb_rx_desc : 1));
 325                 if (len < (int)sizeof(struct tun_pi))
 326                         break;
 327
 328                 /* Packet couldn't fit in the provided mbuf */
 329                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 330                         rxq->stats.ierrors++;
 331                         continue;
 332                 }
 333
 334                 len -= sizeof(struct tun_pi);
 335
 336                 mbuf->pkt_len = len;
 337                 mbuf->port = rxq->in_port;
 338                 while (1) {
 339                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 340
 341                         if (unlikely(!buf)) {
 342                                 rxq->stats.rx_nombuf++;
 343                                 /* No new buf has been allocated: do nothing */
 344                                 if (!new_tail || !seg)
 345                                         goto end;
 346
 347                                 seg->next = NULL;
 348                                 rte_pktmbuf_free(mbuf);
 349
 350                                 goto end;
 351                         }
 352                         seg = seg ? seg->next : mbuf;
 353                         if (rxq->pool == mbuf)
 354                                 rxq->pool = buf;
 355                         if (new_tail)
 356                                 new_tail->next = buf;
 357                         new_tail = buf;
 358                         new_tail->next = seg->next;
 359
 360                         /* iovecs[0] is reserved for packet info (pi) */
 361                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 362                                 buf->buf_len - data_off;
 363                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 364                                 (char *)buf->buf_addr + data_off;
 365
 366                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 367                         seg->data_off = data_off;
 368
 369                         len -= seg->data_len;
 370                         if (len <= 0)
 371                                 break;
 372                         mbuf->nb_segs++;
 373                         /* First segment has headroom, not the others */
 374                         data_off = 0;
 375                 }
 376                 seg->next = NULL;
 377                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 378                                                       RTE_PTYPE_ALL_MASK);
 379                 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
 380                         tap_verify_csum(mbuf);
 381
 382                 /* account for the receive frame */
 383                 bufs[num_rx++] = mbuf;
 384                 num_rx_bytes += mbuf->pkt_len;
 385         }
 386 end:
 387         rxq->stats.ipackets += num_rx;
 388         rxq->stats.ibytes += num_rx_bytes;
 389
 390         return num_rx;
 391 }
 392
 393 static uint64_t
 394 tap_tx_offload_get_port_capa(void)
 395 {
 396         /*
 397          * No specific port Tx offload capabilities.
 398          */
 399         return 0;
 400 }
 401
 402 static uint64_t
 403 tap_tx_offload_get_queue_capa(void)
 404 {
 405         return DEV_TX_OFFLOAD_MULTI_SEGS |
 406                DEV_TX_OFFLOAD_IPV4_CKSUM |
 407                DEV_TX_OFFLOAD_UDP_CKSUM |
 408                DEV_TX_OFFLOAD_TCP_CKSUM;
 409 }
 410
 411 static bool
 412 tap_txq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 413 {
 414         uint64_t port_offloads = dev->data->dev_conf.txmode.offloads;
 415         uint64_t queue_supp_offloads = tap_tx_offload_get_queue_capa();
 416         uint64_t port_supp_offloads = tap_tx_offload_get_port_capa();
 417
 418         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 419             offloads)
 420                 return false;
 421         /* Verify we have no conflict with port offloads */
 422         if ((port_offloads ^ offloads) & port_supp_offloads)
 423                 return false;
 424         return true;
 425 }
 426
 427 static void
 428 tap_tx_offload(char *packet, uint64_t ol_flags, unsigned int l2_len,
 429                unsigned int l3_len)
 430 {
 431         void *l3_hdr = packet + l2_len;
 432
 433         if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
 434                 struct ipv4_hdr *iph = l3_hdr;
 435                 uint16_t cksum;
 436
 437                 iph->hdr_checksum = 0;
 438                 cksum = rte_raw_cksum(iph, l3_len);
 439                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 440         }
 441         if (ol_flags & PKT_TX_L4_MASK) {
 442                 uint16_t l4_len;
 443                 uint32_t cksum;
 444                 uint16_t *l4_cksum;
 445                 void *l4_hdr;
 446
 447                 l4_hdr = packet + l2_len + l3_len;
 448                 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
 449                         l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
 450                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 451                         l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
 452                 else
 453                         return;
 454                 *l4_cksum = 0;
 455                 if (ol_flags & PKT_TX_IPV4) {
 456                         struct ipv4_hdr *iph = l3_hdr;
 457
 458                         l4_len = rte_be_to_cpu_16(iph->total_length) - l3_len;
 459                         cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 460                 } else {
 461                         struct ipv6_hdr *ip6h = l3_hdr;
 462
 463                         /* payload_len does not include ext headers */
 464                         l4_len = rte_be_to_cpu_16(ip6h->payload_len) -
 465                                 l3_len + sizeof(struct ipv6_hdr);
 466                         cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 467                 }
 468                 cksum += rte_raw_cksum(l4_hdr, l4_len);
 469                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 470                 cksum = (~cksum) & 0xffff;
 471                 if (cksum == 0)
 472                         cksum = 0xffff;
 473                 *l4_cksum = cksum;
 474         }
 475 }
 476
 477 /* Callback to handle sending packets from the tap interface
 478  */
 479 static uint16_t
 480 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 481 {
 482         struct tx_queue *txq = queue;
 483         uint16_t num_tx = 0;
 484         unsigned long num_tx_bytes = 0;
 485         uint32_t max_size;
 486         int i;
 487
 488         if (unlikely(nb_pkts == 0))
 489                 return 0;
 490
 491         max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
 492         for (i = 0; i < nb_pkts; i++) {
 493                 struct rte_mbuf *mbuf = bufs[num_tx];
 494                 struct iovec iovecs[mbuf->nb_segs + 1];
 495                 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
 496                 struct rte_mbuf *seg = mbuf;
 497                 char m_copy[mbuf->data_len];
 498                 int n;
 499                 int j;
 500
 501                 /* stats.errs will be incremented */
 502                 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
 503                         break;
 504
 505                 /*
 506                  * TUN and TAP are created with IFF_NO_PI disabled.
 507                  * For TUN PMD this mandatory as fields are used by
 508                  * Kernel tun.c to determine whether its IP or non IP
 509                  * packets.
 510                  *
 511                  * The logic fetches the first byte of data from mbuf.
 512                  * compares whether its v4 or v6. If none matches default
 513                  * value 0x00 is taken for protocol field.
 514                  */
 515                 char *buff_data = rte_pktmbuf_mtod(seg, void *);
 516                 j = (*buff_data & 0xf0);
 517                 pi.proto = (j == 0x40) ? 0x0008 :
 518                                 (j == 0x60) ? 0xdd86 : 0x00;
 519
 520                 iovecs[0].iov_base = &pi;
 521                 iovecs[0].iov_len = sizeof(pi);
 522                 for (j = 1; j <= mbuf->nb_segs; j++) {
 523                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 524                         iovecs[j].iov_base =
 525                                 rte_pktmbuf_mtod(seg, void *);
 526                         seg = seg->next;
 527                 }
 528                 if (txq->csum &&
 529                     ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 530                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 531                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
 532                         /* Support only packets with all data in the same seg */
 533                         if (mbuf->nb_segs > 1)
 534                                 break;
 535                         /* To change checksums, work on a copy of data. */
 536                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 537                                    rte_pktmbuf_data_len(mbuf));
 538                         tap_tx_offload(m_copy, mbuf->ol_flags,
 539                                        mbuf->l2_len, mbuf->l3_len);
 540                         iovecs[1].iov_base = m_copy;
 541                 }
 542                 /* copy the tx frame data */
 543                 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
 544                 if (n <= 0)
 545                         break;
 546
 547                 num_tx++;
 548                 num_tx_bytes += mbuf->pkt_len;
 549                 rte_pktmbuf_free(mbuf);
 550         }
 551
 552         txq->stats.opackets += num_tx;
 553         txq->stats.errs += nb_pkts - num_tx;
 554         txq->stats.obytes += num_tx_bytes;
 555
 556         return num_tx;
 557 }
 558
 559 static const char *
 560 tap_ioctl_req2str(unsigned long request)
 561 {
 562         switch (request) {
 563         case SIOCSIFFLAGS:
 564                 return "SIOCSIFFLAGS";
 565         case SIOCGIFFLAGS:
 566                 return "SIOCGIFFLAGS";
 567         case SIOCGIFHWADDR:
 568                 return "SIOCGIFHWADDR";
 569         case SIOCSIFHWADDR:
 570                 return "SIOCSIFHWADDR";
 571         case SIOCSIFMTU:
 572                 return "SIOCSIFMTU";
 573         }
 574         return "UNKNOWN";
 575 }
 576
 577 static int
 578 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 579           struct ifreq *ifr, int set, enum ioctl_mode mode)
 580 {
 581         short req_flags = ifr->ifr_flags;
 582         int remote = pmd->remote_if_index &&
 583                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 584
 585         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 586                 return 0;
 587         /*
 588          * If there is a remote netdevice, apply ioctl on it, then apply it on
 589          * the tap netdevice.
 590          */
 591 apply:
 592         if (remote)
 593                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
 594         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 595                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 596         switch (request) {
 597         case SIOCSIFFLAGS:
 598                 /* fetch current flags to leave other flags untouched */
 599                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 600                         goto error;
 601                 if (set)
 602                         ifr->ifr_flags |= req_flags;
 603                 else
 604                         ifr->ifr_flags &= ~req_flags;
 605                 break;
 606         case SIOCGIFFLAGS:
 607         case SIOCGIFHWADDR:
 608         case SIOCSIFHWADDR:
 609         case SIOCSIFMTU:
 610                 break;
 611         default:
 612                 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
 613                         pmd->name);
 614                 return -EINVAL;
 615         }
 616         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 617                 goto error;
 618         if (remote-- && mode == LOCAL_AND_REMOTE)
 619                 goto apply;
 620         return 0;
 621
 622 error:
 623         TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
 624                 tap_ioctl_req2str(request), strerror(errno), errno);
 625         return -errno;
 626 }
 627
 628 static int
 629 tap_link_set_down(struct rte_eth_dev *dev)
 630 {
 631         struct pmd_internals *pmd = dev->data->dev_private;
 632         struct ifreq ifr = { .ifr_flags = IFF_UP };
 633
 634         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 635         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 636 }
 637
 638 static int
 639 tap_link_set_up(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_UP;
 645         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 646 }
 647
 648 static int
 649 tap_dev_start(struct rte_eth_dev *dev)
 650 {
 651         int err;
 652
 653         err = tap_intr_handle_set(dev, 1);
 654         if (err)
 655                 return err;
 656         return tap_link_set_up(dev);
 657 }
 658
 659 /* This function gets called when the current port gets stopped.
 660  */
 661 static void
 662 tap_dev_stop(struct rte_eth_dev *dev)
 663 {
 664         tap_intr_handle_set(dev, 0);
 665         tap_link_set_down(dev);
 666 }
 667
 668 static int
 669 tap_dev_configure(struct rte_eth_dev *dev)
 670 {
 671         uint64_t supp_tx_offloads = tap_tx_offload_get_port_capa() |
 672                                 tap_tx_offload_get_queue_capa();
 673         uint64_t tx_offloads = dev->data->dev_conf.txmode.offloads;
 674
 675         if ((tx_offloads & supp_tx_offloads) != tx_offloads) {
 676                 rte_errno = ENOTSUP;
 677                 TAP_LOG(ERR,
 678                         "Some Tx offloads are not supported "
 679                         "requested 0x%" PRIx64 " supported 0x%" PRIx64,
 680                         tx_offloads, supp_tx_offloads);
 681                 return -rte_errno;
 682         }
 683         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 684                 TAP_LOG(ERR,
 685                         "%s: number of rx queues %d exceeds max num of queues %d",
 686                         dev->device->name,
 687                         dev->data->nb_rx_queues,
 688                         RTE_PMD_TAP_MAX_QUEUES);
 689                 return -1;
 690         }
 691         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 692                 TAP_LOG(ERR,
 693                         "%s: number of tx queues %d exceeds max num of queues %d",
 694                         dev->device->name,
 695                         dev->data->nb_tx_queues,
 696                         RTE_PMD_TAP_MAX_QUEUES);
 697                 return -1;
 698         }
 699
 700         TAP_LOG(INFO, "%s: %p: TX configured queues number: %u",
 701                 dev->device->name, (void *)dev, dev->data->nb_tx_queues);
 702
 703         TAP_LOG(INFO, "%s: %p: RX configured queues number: %u",
 704                 dev->device->name, (void *)dev, dev->data->nb_rx_queues);
 705
 706         return 0;
 707 }
 708
 709 static uint32_t
 710 tap_dev_speed_capa(void)
 711 {
 712         uint32_t speed = pmd_link.link_speed;
 713         uint32_t capa = 0;
 714
 715         if (speed >= ETH_SPEED_NUM_10M)
 716                 capa |= ETH_LINK_SPEED_10M;
 717         if (speed >= ETH_SPEED_NUM_100M)
 718                 capa |= ETH_LINK_SPEED_100M;
 719         if (speed >= ETH_SPEED_NUM_1G)
 720                 capa |= ETH_LINK_SPEED_1G;
 721         if (speed >= ETH_SPEED_NUM_5G)
 722                 capa |= ETH_LINK_SPEED_2_5G;
 723         if (speed >= ETH_SPEED_NUM_5G)
 724                 capa |= ETH_LINK_SPEED_5G;
 725         if (speed >= ETH_SPEED_NUM_10G)
 726                 capa |= ETH_LINK_SPEED_10G;
 727         if (speed >= ETH_SPEED_NUM_20G)
 728                 capa |= ETH_LINK_SPEED_20G;
 729         if (speed >= ETH_SPEED_NUM_25G)
 730                 capa |= ETH_LINK_SPEED_25G;
 731         if (speed >= ETH_SPEED_NUM_40G)
 732                 capa |= ETH_LINK_SPEED_40G;
 733         if (speed >= ETH_SPEED_NUM_50G)
 734                 capa |= ETH_LINK_SPEED_50G;
 735         if (speed >= ETH_SPEED_NUM_56G)
 736                 capa |= ETH_LINK_SPEED_56G;
 737         if (speed >= ETH_SPEED_NUM_100G)
 738                 capa |= ETH_LINK_SPEED_100G;
 739
 740         return capa;
 741 }
 742
 743 static void
 744 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 745 {
 746         struct pmd_internals *internals = dev->data->dev_private;
 747
 748         dev_info->if_index = internals->if_index;
 749         dev_info->max_mac_addrs = 1;
 750         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 751         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
 752         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
 753         dev_info->min_rx_bufsize = 0;
 754         dev_info->speed_capa = tap_dev_speed_capa();
 755         dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
 756         dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
 757                                     dev_info->rx_queue_offload_capa;
 758         dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
 759         dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
 760                                     dev_info->tx_queue_offload_capa;
 761 }
 762
 763 static int
 764 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 765 {
 766         unsigned int i, imax;
 767         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 768         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 769         unsigned long rx_nombuf = 0, ierrors = 0;
 770         const struct pmd_internals *pmd = dev->data->dev_private;
 771
 772         /* rx queue statistics */
 773         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 774                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 775         for (i = 0; i < imax; i++) {
 776                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 777                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 778                 rx_total += tap_stats->q_ipackets[i];
 779                 rx_bytes_total += tap_stats->q_ibytes[i];
 780                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
 781                 ierrors += pmd->rxq[i].stats.ierrors;
 782         }
 783
 784         /* tx queue statistics */
 785         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 786                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 787
 788         for (i = 0; i < imax; i++) {
 789                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 790                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 791                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 792                 tx_total += tap_stats->q_opackets[i];
 793                 tx_err_total += tap_stats->q_errors[i];
 794                 tx_bytes_total += tap_stats->q_obytes[i];
 795         }
 796
 797         tap_stats->ipackets = rx_total;
 798         tap_stats->ibytes = rx_bytes_total;
 799         tap_stats->ierrors = ierrors;
 800         tap_stats->rx_nombuf = rx_nombuf;
 801         tap_stats->opackets = tx_total;
 802         tap_stats->oerrors = tx_err_total;
 803         tap_stats->obytes = tx_bytes_total;
 804         return 0;
 805 }
 806
 807 static void
 808 tap_stats_reset(struct rte_eth_dev *dev)
 809 {
 810         int i;
 811         struct pmd_internals *pmd = dev->data->dev_private;
 812
 813         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 814                 pmd->rxq[i].stats.ipackets = 0;
 815                 pmd->rxq[i].stats.ibytes = 0;
 816                 pmd->rxq[i].stats.ierrors = 0;
 817                 pmd->rxq[i].stats.rx_nombuf = 0;
 818
 819                 pmd->txq[i].stats.opackets = 0;
 820                 pmd->txq[i].stats.errs = 0;
 821                 pmd->txq[i].stats.obytes = 0;
 822         }
 823 }
 824
 825 static void
 826 tap_dev_close(struct rte_eth_dev *dev)
 827 {
 828         int i;
 829         struct pmd_internals *internals = dev->data->dev_private;
 830
 831         tap_link_set_down(dev);
 832         tap_flow_flush(dev, NULL);
 833         tap_flow_implicit_flush(internals, NULL);
 834
 835         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 836                 if (internals->rxq[i].fd != -1) {
 837                         close(internals->rxq[i].fd);
 838                         internals->rxq[i].fd = -1;
 839                 }
 840                 if (internals->txq[i].fd != -1) {
 841                         close(internals->txq[i].fd);
 842                         internals->txq[i].fd = -1;
 843                 }
 844         }
 845
 846         if (internals->remote_if_index) {
 847                 /* Restore initial remote state */
 848                 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
 849                                 &internals->remote_initial_flags);
 850         }
 851 }
 852
 853 static void
 854 tap_rx_queue_release(void *queue)
 855 {
 856         struct rx_queue *rxq = queue;
 857
 858         if (rxq && (rxq->fd > 0)) {
 859                 close(rxq->fd);
 860                 rxq->fd = -1;
 861                 rte_pktmbuf_free(rxq->pool);
 862                 rte_free(rxq->iovecs);
 863                 rxq->pool = NULL;
 864                 rxq->iovecs = NULL;
 865         }
 866 }
 867
 868 static void
 869 tap_tx_queue_release(void *queue)
 870 {
 871         struct tx_queue *txq = queue;
 872
 873         if (txq && (txq->fd > 0)) {
 874                 close(txq->fd);
 875                 txq->fd = -1;
 876         }
 877 }
 878
 879 static int
 880 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
 881 {
 882         struct rte_eth_link *dev_link = &dev->data->dev_link;
 883         struct pmd_internals *pmd = dev->data->dev_private;
 884         struct ifreq ifr = { .ifr_flags = 0 };
 885
 886         if (pmd->remote_if_index) {
 887                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
 888                 if (!(ifr.ifr_flags & IFF_UP) ||
 889                     !(ifr.ifr_flags & IFF_RUNNING)) {
 890                         dev_link->link_status = ETH_LINK_DOWN;
 891                         return 0;
 892                 }
 893         }
 894         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
 895         dev_link->link_status =
 896                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
 897                  ETH_LINK_UP :
 898                  ETH_LINK_DOWN);
 899         return 0;
 900 }
 901
 902 static void
 903 tap_promisc_enable(struct rte_eth_dev *dev)
 904 {
 905         struct pmd_internals *pmd = dev->data->dev_private;
 906         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 907
 908         dev->data->promiscuous = 1;
 909         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 910         if (pmd->remote_if_index && !pmd->flow_isolate)
 911                 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
 912 }
 913
 914 static void
 915 tap_promisc_disable(struct rte_eth_dev *dev)
 916 {
 917         struct pmd_internals *pmd = dev->data->dev_private;
 918         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 919
 920         dev->data->promiscuous = 0;
 921         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 922         if (pmd->remote_if_index && !pmd->flow_isolate)
 923                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
 924 }
 925
 926 static void
 927 tap_allmulti_enable(struct rte_eth_dev *dev)
 928 {
 929         struct pmd_internals *pmd = dev->data->dev_private;
 930         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 931
 932         dev->data->all_multicast = 1;
 933         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 934         if (pmd->remote_if_index && !pmd->flow_isolate)
 935                 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
 936 }
 937
 938 static void
 939 tap_allmulti_disable(struct rte_eth_dev *dev)
 940 {
 941         struct pmd_internals *pmd = dev->data->dev_private;
 942         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 943
 944         dev->data->all_multicast = 0;
 945         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 946         if (pmd->remote_if_index && !pmd->flow_isolate)
 947                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
 948 }
 949
 950 static int
 951 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 952 {
 953         struct pmd_internals *pmd = dev->data->dev_private;
 954         enum ioctl_mode mode = LOCAL_ONLY;
 955         struct ifreq ifr;
 956         int ret;
 957
 958         if (is_zero_ether_addr(mac_addr)) {
 959                 TAP_LOG(ERR, "%s: can't set an empty MAC address",
 960                         dev->device->name);
 961                 return -EINVAL;
 962         }
 963         /* Check the actual current MAC address on the tap netdevice */
 964         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
 965         if (ret < 0)
 966                 return ret;
 967         if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 968                                mac_addr))
 969                 return 0;
 970         /* Check the current MAC address on the remote */
 971         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
 972         if (ret < 0)
 973                 return ret;
 974         if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 975                                mac_addr))
 976                 mode = LOCAL_AND_REMOTE;
 977         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
 978         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
 979         ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
 980         if (ret < 0)
 981                 return ret;
 982         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
 983         if (pmd->remote_if_index && !pmd->flow_isolate) {
 984                 /* Replace MAC redirection rule after a MAC change */
 985                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
 986                 if (ret < 0) {
 987                         TAP_LOG(ERR,
 988                                 "%s: Couldn't delete MAC redirection rule",
 989                                 dev->device->name);
 990                         return ret;
 991                 }
 992                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
 993                 if (ret < 0) {
 994                         TAP_LOG(ERR,
 995                                 "%s: Couldn't add MAC redirection rule",
 996                                 dev->device->name);
 997                         return ret;
 998                 }
 999         }
1000
1001         return 0;
1002 }
1003
1004 static int
1005 tap_setup_queue(struct rte_eth_dev *dev,
1006                 struct pmd_internals *internals,
1007                 uint16_t qid,
1008                 int is_rx)
1009 {
1010         int *fd;
1011         int *other_fd;
1012         const char *dir;
1013         struct pmd_internals *pmd = dev->data->dev_private;
1014         struct rx_queue *rx = &internals->rxq[qid];
1015         struct tx_queue *tx = &internals->txq[qid];
1016
1017         if (is_rx) {
1018                 fd = &rx->fd;
1019                 other_fd = &tx->fd;
1020                 dir = "rx";
1021         } else {
1022                 fd = &tx->fd;
1023                 other_fd = &rx->fd;
1024                 dir = "tx";
1025         }
1026         if (*fd != -1) {
1027                 /* fd for this queue already exists */
1028                 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1029                         pmd->name, *fd, dir, qid);
1030         } else if (*other_fd != -1) {
1031                 /* Only other_fd exists. dup it */
1032                 *fd = dup(*other_fd);
1033                 if (*fd < 0) {
1034                         *fd = -1;
1035                         TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1036                         return -1;
1037                 }
1038                 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1039                         pmd->name, *other_fd, dir, qid, *fd);
1040         } else {
1041                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1042                 *fd = tun_alloc(pmd);
1043                 if (*fd < 0) {
1044                         *fd = -1; /* restore original value */
1045                         TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1046                         return -1;
1047                 }
1048                 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
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                 TAP_LOG(WARNING,
1079                         "nb_rx_queues %d too small or mempool NULL",
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                 TAP_LOG(ERR,
1088                         "%p: Rx queue offloads 0x%" PRIx64
1089                         " don't match port offloads 0x%" PRIx64
1090                         " or supported offloads 0x%" PRIx64,
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                 TAP_LOG(WARNING,
1105                         "%s: Couldn't allocate %d RX descriptors",
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                         TAP_LOG(WARNING,
1125                                 "%s: couldn't allocate memory for queue %d",
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         TAP_LOG(DEBUG, "  RX TUNTAP device name %s, qid %d on fd %d",
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                         TAP_LOG(ERR,
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                                 tap_tx_offload_get_queue_capa()));
1186                         return -rte_errno;
1187                 }
1188         }
1189         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1190         if (ret == -1)
1191                 return -1;
1192         TAP_LOG(DEBUG,
1193                 "  TX TUNTAP device name %s, qid %d on fd %d csum %s",
1194                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd,
1195                 txq->csum ? "on" : "off");
1196
1197         return 0;
1198 }
1199
1200 static int
1201 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1202 {
1203         struct pmd_internals *pmd = dev->data->dev_private;
1204         struct ifreq ifr = { .ifr_mtu = mtu };
1205         int err = 0;
1206
1207         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1208         if (!err)
1209                 dev->data->mtu = mtu;
1210
1211         return err;
1212 }
1213
1214 static int
1215 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1216                      struct ether_addr *mc_addr_set __rte_unused,
1217                      uint32_t nb_mc_addr __rte_unused)
1218 {
1219         /*
1220          * Nothing to do actually: the tap has no filtering whatsoever, every
1221          * packet is received.
1222          */
1223         return 0;
1224 }
1225
1226 static int
1227 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1228 {
1229         struct rte_eth_dev *dev = arg;
1230         struct pmd_internals *pmd = dev->data->dev_private;
1231         struct ifinfomsg *info = NLMSG_DATA(nh);
1232
1233         if (nh->nlmsg_type != RTM_NEWLINK ||
1234             (info->ifi_index != pmd->if_index &&
1235              info->ifi_index != pmd->remote_if_index))
1236                 return 0;
1237         return tap_link_update(dev, 0);
1238 }
1239
1240 static void
1241 tap_dev_intr_handler(void *cb_arg)
1242 {
1243         struct rte_eth_dev *dev = cb_arg;
1244         struct pmd_internals *pmd = dev->data->dev_private;
1245
1246         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1247 }
1248
1249 static int
1250 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1251 {
1252         struct pmd_internals *pmd = dev->data->dev_private;
1253
1254         /* In any case, disable interrupt if the conf is no longer there. */
1255         if (!dev->data->dev_conf.intr_conf.lsc) {
1256                 if (pmd->intr_handle.fd != -1) {
1257                         tap_nl_final(pmd->intr_handle.fd);
1258                         rte_intr_callback_unregister(&pmd->intr_handle,
1259                                 tap_dev_intr_handler, dev);
1260                 }
1261                 return 0;
1262         }
1263         if (set) {
1264                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1265                 if (unlikely(pmd->intr_handle.fd == -1))
1266                         return -EBADF;
1267                 return rte_intr_callback_register(
1268                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1269         }
1270         tap_nl_final(pmd->intr_handle.fd);
1271         return rte_intr_callback_unregister(&pmd->intr_handle,
1272                                             tap_dev_intr_handler, dev);
1273 }
1274
1275 static int
1276 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1277 {
1278         int err;
1279
1280         err = tap_lsc_intr_handle_set(dev, set);
1281         if (err)
1282                 return err;
1283         err = tap_rx_intr_vec_set(dev, set);
1284         if (err && set)
1285                 tap_lsc_intr_handle_set(dev, 0);
1286         return err;
1287 }
1288
1289 static const uint32_t*
1290 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1291 {
1292         static const uint32_t ptypes[] = {
1293                 RTE_PTYPE_INNER_L2_ETHER,
1294                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1295                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1296                 RTE_PTYPE_INNER_L3_IPV4,
1297                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1298                 RTE_PTYPE_INNER_L3_IPV6,
1299                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1300                 RTE_PTYPE_INNER_L4_FRAG,
1301                 RTE_PTYPE_INNER_L4_UDP,
1302                 RTE_PTYPE_INNER_L4_TCP,
1303                 RTE_PTYPE_INNER_L4_SCTP,
1304                 RTE_PTYPE_L2_ETHER,
1305                 RTE_PTYPE_L2_ETHER_VLAN,
1306                 RTE_PTYPE_L2_ETHER_QINQ,
1307                 RTE_PTYPE_L3_IPV4,
1308                 RTE_PTYPE_L3_IPV4_EXT,
1309                 RTE_PTYPE_L3_IPV6_EXT,
1310                 RTE_PTYPE_L3_IPV6,
1311                 RTE_PTYPE_L4_FRAG,
1312                 RTE_PTYPE_L4_UDP,
1313                 RTE_PTYPE_L4_TCP,
1314                 RTE_PTYPE_L4_SCTP,
1315         };
1316
1317         return ptypes;
1318 }
1319
1320 static int
1321 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1322                   struct rte_eth_fc_conf *fc_conf)
1323 {
1324         fc_conf->mode = RTE_FC_NONE;
1325         return 0;
1326 }
1327
1328 static int
1329 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1330                   struct rte_eth_fc_conf *fc_conf)
1331 {
1332         if (fc_conf->mode != RTE_FC_NONE)
1333                 return -ENOTSUP;
1334         return 0;
1335 }
1336
1337 static const struct eth_dev_ops ops = {
1338         .dev_start              = tap_dev_start,
1339         .dev_stop               = tap_dev_stop,
1340         .dev_close              = tap_dev_close,
1341         .dev_configure          = tap_dev_configure,
1342         .dev_infos_get          = tap_dev_info,
1343         .rx_queue_setup         = tap_rx_queue_setup,
1344         .tx_queue_setup         = tap_tx_queue_setup,
1345         .rx_queue_release       = tap_rx_queue_release,
1346         .tx_queue_release       = tap_tx_queue_release,
1347         .flow_ctrl_get          = tap_flow_ctrl_get,
1348         .flow_ctrl_set          = tap_flow_ctrl_set,
1349         .link_update            = tap_link_update,
1350         .dev_set_link_up        = tap_link_set_up,
1351         .dev_set_link_down      = tap_link_set_down,
1352         .promiscuous_enable     = tap_promisc_enable,
1353         .promiscuous_disable    = tap_promisc_disable,
1354         .allmulticast_enable    = tap_allmulti_enable,
1355         .allmulticast_disable   = tap_allmulti_disable,
1356         .mac_addr_set           = tap_mac_set,
1357         .mtu_set                = tap_mtu_set,
1358         .set_mc_addr_list       = tap_set_mc_addr_list,
1359         .stats_get              = tap_stats_get,
1360         .stats_reset            = tap_stats_reset,
1361         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1362         .filter_ctrl            = tap_dev_filter_ctrl,
1363 };
1364
1365 static int
1366 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1367                    char *remote_iface, struct ether_addr *mac_addr)
1368 {
1369         int numa_node = rte_socket_id();
1370         struct rte_eth_dev *dev;
1371         struct pmd_internals *pmd;
1372         struct rte_eth_dev_data *data;
1373         struct ifreq ifr;
1374         int i;
1375
1376         TAP_LOG(DEBUG, "%s device on numa %u",
1377                         tuntap_name, rte_socket_id());
1378
1379         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1380         if (!dev) {
1381                 TAP_LOG(ERR, "%s Unable to allocate device struct",
1382                                 tuntap_name);
1383                 goto error_exit_nodev;
1384         }
1385
1386         pmd = dev->data->dev_private;
1387         pmd->dev = dev;
1388         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1389
1390         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1391         if (pmd->ioctl_sock == -1) {
1392                 TAP_LOG(ERR,
1393                         "%s Unable to get a socket for management: %s",
1394                         tuntap_name, strerror(errno));
1395                 goto error_exit;
1396         }
1397
1398         /* Setup some default values */
1399         data = dev->data;
1400         data->dev_private = pmd;
1401         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1402         data->numa_node = numa_node;
1403
1404         data->dev_link = pmd_link;
1405         data->mac_addrs = &pmd->eth_addr;
1406         /* Set the number of RX and TX queues */
1407         data->nb_rx_queues = 0;
1408         data->nb_tx_queues = 0;
1409
1410         dev->dev_ops = &ops;
1411         dev->rx_pkt_burst = pmd_rx_burst;
1412         dev->tx_pkt_burst = pmd_tx_burst;
1413
1414         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1415         pmd->intr_handle.fd = -1;
1416         dev->intr_handle = &pmd->intr_handle;
1417
1418         /* Presetup the fds to -1 as being not valid */
1419         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1420                 pmd->rxq[i].fd = -1;
1421                 pmd->txq[i].fd = -1;
1422         }
1423
1424         if (tap_type) {
1425                 if (is_zero_ether_addr(mac_addr))
1426                         eth_random_addr((uint8_t *)&pmd->eth_addr);
1427                 else
1428                         rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1429         }
1430
1431         /* Immediately create the netdevice (this will create the 1st queue). */
1432         /* rx queue */
1433         if (tap_setup_queue(dev, pmd, 0, 1) == -1)
1434                 goto error_exit;
1435         /* tx queue */
1436         if (tap_setup_queue(dev, pmd, 0, 0) == -1)
1437                 goto error_exit;
1438
1439         ifr.ifr_mtu = dev->data->mtu;
1440         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1441                 goto error_exit;
1442
1443         if (tap_type) {
1444                 memset(&ifr, 0, sizeof(struct ifreq));
1445                 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1446                 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1447                                 ETHER_ADDR_LEN);
1448                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1449                         goto error_exit;
1450         }
1451
1452         /*
1453          * Set up everything related to rte_flow:
1454          * - netlink socket
1455          * - tap / remote if_index
1456          * - mandatory QDISCs
1457          * - rte_flow actual/implicit lists
1458          * - implicit rules
1459          */
1460         pmd->nlsk_fd = tap_nl_init(0);
1461         if (pmd->nlsk_fd == -1) {
1462                 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
1463                         pmd->name);
1464                 goto disable_rte_flow;
1465         }
1466         pmd->if_index = if_nametoindex(pmd->name);
1467         if (!pmd->if_index) {
1468                 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
1469                 goto disable_rte_flow;
1470         }
1471         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1472                 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
1473                         pmd->name);
1474                 goto disable_rte_flow;
1475         }
1476         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1477                 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1478                         pmd->name);
1479                 goto disable_rte_flow;
1480         }
1481         LIST_INIT(&pmd->flows);
1482
1483         if (strlen(remote_iface)) {
1484                 pmd->remote_if_index = if_nametoindex(remote_iface);
1485                 if (!pmd->remote_if_index) {
1486                         TAP_LOG(ERR, "%s: failed to get %s if_index.",
1487                                 pmd->name, remote_iface);
1488                         goto error_remote;
1489                 }
1490                 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1491                          "%s", remote_iface);
1492
1493                 /* Save state of remote device */
1494                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1495
1496                 /* Replicate remote MAC address */
1497                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1498                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1499                                 pmd->name, pmd->remote_iface);
1500                         goto error_remote;
1501                 }
1502                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1503                            ETHER_ADDR_LEN);
1504                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1505                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1506                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1507                                 pmd->name, remote_iface);
1508                         goto error_remote;
1509                 }
1510
1511                 /*
1512                  * Flush usually returns negative value because it tries to
1513                  * delete every QDISC (and on a running device, one QDISC at
1514                  * least is needed). Ignore negative return value.
1515                  */
1516                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1517                 if (qdisc_create_ingress(pmd->nlsk_fd,
1518                                          pmd->remote_if_index) < 0) {
1519                         TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1520                                 pmd->remote_iface);
1521                         goto error_remote;
1522                 }
1523                 LIST_INIT(&pmd->implicit_flows);
1524                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1525                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1526                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1527                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1528                         TAP_LOG(ERR,
1529                                 "%s: failed to create implicit rules.",
1530                                 pmd->name);
1531                         goto error_remote;
1532                 }
1533         }
1534
1535         return 0;
1536
1537 disable_rte_flow:
1538         TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
1539                 strerror(errno), errno);
1540         if (strlen(remote_iface)) {
1541                 TAP_LOG(ERR, "Remote feature requires flow support.");
1542                 goto error_exit;
1543         }
1544         return 0;
1545
1546 error_remote:
1547         TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
1548                 strerror(errno), errno);
1549         tap_flow_implicit_flush(pmd, NULL);
1550
1551 error_exit:
1552         if (pmd->ioctl_sock > 0)
1553                 close(pmd->ioctl_sock);
1554         rte_eth_dev_release_port(dev);
1555
1556 error_exit_nodev:
1557         TAP_LOG(ERR, "%s Unable to initialize %s",
1558                 tuntap_name, rte_vdev_device_name(vdev));
1559
1560         return -EINVAL;
1561 }
1562
1563 static int
1564 set_interface_name(const char *key __rte_unused,
1565                    const char *value,
1566                    void *extra_args)
1567 {
1568         char *name = (char *)extra_args;
1569
1570         if (value)
1571                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN - 1);
1572         else
1573                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1574                          DEFAULT_TAP_NAME, (tap_unit - 1));
1575
1576         return 0;
1577 }
1578
1579 static int
1580 set_remote_iface(const char *key __rte_unused,
1581                  const char *value,
1582                  void *extra_args)
1583 {
1584         char *name = (char *)extra_args;
1585
1586         if (value)
1587                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
1588
1589         return 0;
1590 }
1591
1592 static int parse_user_mac(struct ether_addr *user_mac,
1593                 const char *value)
1594 {
1595         unsigned int index = 0;
1596         char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
1597
1598         if (user_mac == NULL || value == NULL)
1599                 return 0;
1600
1601         strlcpy(mac_temp, value, sizeof(mac_temp));
1602         mac_byte = strtok(mac_temp, ":");
1603
1604         while ((mac_byte != NULL) &&
1605                         (strlen(mac_byte) <= 2) &&
1606                         (strlen(mac_byte) == strspn(mac_byte,
1607                                         ETH_TAP_CMP_MAC_FMT))) {
1608                 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
1609                 mac_byte = strtok(NULL, ":");
1610         }
1611
1612         return index;
1613 }
1614
1615 static int
1616 set_mac_type(const char *key __rte_unused,
1617              const char *value,
1618              void *extra_args)
1619 {
1620         struct ether_addr *user_mac = extra_args;
1621
1622         if (!value)
1623                 return 0;
1624
1625         if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
1626                 static int iface_idx;
1627
1628                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1629                 memcpy((char *)user_mac->addr_bytes, "\0dtap", ETHER_ADDR_LEN);
1630                 user_mac->addr_bytes[ETHER_ADDR_LEN - 1] = iface_idx++ + '0';
1631                 goto success;
1632         }
1633
1634         if (parse_user_mac(user_mac, value) != 6)
1635                 goto error;
1636 success:
1637         TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
1638         return 0;
1639
1640 error:
1641         TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
1642                 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
1643         return -1;
1644 }
1645
1646 /*
1647  * Open a TUN interface device. TUN PMD
1648  * 1) sets tap_type as false
1649  * 2) intakes iface as argument.
1650  * 3) as interface is virtual set speed to 10G
1651  */
1652 static int
1653 rte_pmd_tun_probe(struct rte_vdev_device *dev)
1654 {
1655         const char *name, *params;
1656         int ret;
1657         struct rte_kvargs *kvlist = NULL;
1658         char tun_name[RTE_ETH_NAME_MAX_LEN];
1659         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1660
1661         tap_type = 0;
1662         strcpy(tuntap_name, "TUN");
1663
1664         name = rte_vdev_device_name(dev);
1665         params = rte_vdev_device_args(dev);
1666         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1667
1668         if (params && (params[0] != '\0')) {
1669                 TAP_LOG(DEBUG, "parameters (%s)", params);
1670
1671                 kvlist = rte_kvargs_parse(params, valid_arguments);
1672                 if (kvlist) {
1673                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1674                                 ret = rte_kvargs_process(kvlist,
1675                                         ETH_TAP_IFACE_ARG,
1676                                         &set_interface_name,
1677                                         tun_name);
1678
1679                                 if (ret == -1)
1680                                         goto leave;
1681                         }
1682                 }
1683         }
1684         pmd_link.link_speed = ETH_SPEED_NUM_10G;
1685
1686         TAP_LOG(NOTICE, "Initializing pmd_tun for %s as %s",
1687                 name, tun_name);
1688
1689         ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0);
1690
1691 leave:
1692         if (ret == -1) {
1693                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
1694                         name, tun_name);
1695                 tun_unit--; /* Restore the unit number */
1696         }
1697         rte_kvargs_free(kvlist);
1698
1699         return ret;
1700 }
1701
1702 /* Open a TAP interface device.
1703  */
1704 static int
1705 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1706 {
1707         const char *name, *params;
1708         int ret;
1709         struct rte_kvargs *kvlist = NULL;
1710         int speed;
1711         char tap_name[RTE_ETH_NAME_MAX_LEN];
1712         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1713         struct ether_addr user_mac = { .addr_bytes = {0} };
1714         struct rte_eth_dev *eth_dev;
1715
1716         tap_type = 1;
1717         strcpy(tuntap_name, "TAP");
1718
1719         name = rte_vdev_device_name(dev);
1720         params = rte_vdev_device_args(dev);
1721
1722         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
1723             strlen(params) == 0) {
1724                 eth_dev = rte_eth_dev_attach_secondary(name);
1725                 if (!eth_dev) {
1726                         TAP_LOG(ERR, "Failed to probe %s", name);
1727                         return -1;
1728                 }
1729                 /* TODO: request info from primary to set up Rx and Tx */
1730                 eth_dev->dev_ops = &ops;
1731                 return 0;
1732         }
1733
1734         speed = ETH_SPEED_NUM_10G;
1735         snprintf(tap_name, sizeof(tap_name), "%s%d",
1736                  DEFAULT_TAP_NAME, tap_unit++);
1737         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1738
1739         if (params && (params[0] != '\0')) {
1740                 TAP_LOG(DEBUG, "parameters (%s)", params);
1741
1742                 kvlist = rte_kvargs_parse(params, valid_arguments);
1743                 if (kvlist) {
1744                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1745                                 ret = rte_kvargs_process(kvlist,
1746                                                          ETH_TAP_IFACE_ARG,
1747                                                          &set_interface_name,
1748                                                          tap_name);
1749                                 if (ret == -1)
1750                                         goto leave;
1751                         }
1752
1753                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1754                                 ret = rte_kvargs_process(kvlist,
1755                                                          ETH_TAP_REMOTE_ARG,
1756                                                          &set_remote_iface,
1757                                                          remote_iface);
1758                                 if (ret == -1)
1759                                         goto leave;
1760                         }
1761
1762                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1763                                 ret = rte_kvargs_process(kvlist,
1764                                                          ETH_TAP_MAC_ARG,
1765                                                          &set_mac_type,
1766                                                          &user_mac);
1767                                 if (ret == -1)
1768                                         goto leave;
1769                         }
1770                 }
1771         }
1772         pmd_link.link_speed = speed;
1773
1774         TAP_LOG(NOTICE, "Initializing pmd_tap for %s as %s",
1775                 name, tap_name);
1776
1777         ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac);
1778
1779 leave:
1780         if (ret == -1) {
1781                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
1782                         name, tap_name);
1783                 tap_unit--;             /* Restore the unit number */
1784         }
1785         rte_kvargs_free(kvlist);
1786
1787         return ret;
1788 }
1789
1790 /* detach a TUNTAP device.
1791  */
1792 static int
1793 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1794 {
1795         struct rte_eth_dev *eth_dev = NULL;
1796         struct pmd_internals *internals;
1797         int i;
1798
1799         TAP_LOG(DEBUG, "Closing TUN/TAP Ethernet device on numa %u",
1800                 rte_socket_id());
1801
1802         /* find the ethdev entry */
1803         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1804         if (!eth_dev)
1805                 return 0;
1806
1807         internals = eth_dev->data->dev_private;
1808         if (internals->nlsk_fd) {
1809                 tap_flow_flush(eth_dev, NULL);
1810                 tap_flow_implicit_flush(internals, NULL);
1811                 tap_nl_final(internals->nlsk_fd);
1812         }
1813         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1814                 if (internals->rxq[i].fd != -1) {
1815                         close(internals->rxq[i].fd);
1816                         internals->rxq[i].fd = -1;
1817                 }
1818                 if (internals->txq[i].fd != -1) {
1819                         close(internals->txq[i].fd);
1820                         internals->txq[i].fd = -1;
1821                 }
1822         }
1823
1824         close(internals->ioctl_sock);
1825         rte_free(eth_dev->data->dev_private);
1826
1827         rte_eth_dev_release_port(eth_dev);
1828
1829         return 0;
1830 }
1831
1832 static struct rte_vdev_driver pmd_tun_drv = {
1833         .probe = rte_pmd_tun_probe,
1834         .remove = rte_pmd_tap_remove,
1835 };
1836
1837 static struct rte_vdev_driver pmd_tap_drv = {
1838         .probe = rte_pmd_tap_probe,
1839         .remove = rte_pmd_tap_remove,
1840 };
1841
1842 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1843 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
1844 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1845 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
1846                               ETH_TAP_IFACE_ARG "=<string> ");
1847 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1848                               ETH_TAP_IFACE_ARG "=<string> "
1849                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
1850                               ETH_TAP_REMOTE_ARG "=<string>");
1851 int tap_logtype;
1852
1853 RTE_INIT(tap_init_log);
1854 static void
1855 tap_init_log(void)
1856 {
1857         tap_logtype = rte_log_register("pmd.net.tap");
1858         if (tap_logtype >= 0)
1859                 rte_log_set_level(tap_logtype, RTE_LOG_NOTICE);
1860 }