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         RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
 122
 123         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 124         if (fd < 0) {
 125                 RTE_LOG(ERR, PMD, "Unable to create %s interface\n",
 126                                 tuntap_name);
 127                 goto error;
 128         }
 129
 130 #ifdef IFF_MULTI_QUEUE
 131         /* Grab the TUN features to verify we can work multi-queue */
 132         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 133                 RTE_LOG(ERR, PMD, "%s unable to get TUN/TAP features\n",
 134                                 tuntap_name);
 135                 goto error;
 136         }
 137         RTE_LOG(DEBUG, PMD, "%s Features %08x\n", tuntap_name, features);
 138
 139         if (features & IFF_MULTI_QUEUE) {
 140                 RTE_LOG(DEBUG, PMD, "  Multi-queue support for %d queues\n",
 141                         RTE_PMD_TAP_MAX_QUEUES);
 142                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 143         } else
 144 #endif
 145         {
 146                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 147                 RTE_LOG(DEBUG, PMD, "  Single queue only support\n");
 148         }
 149
 150         /* Set the TUN/TAP configuration and set the name if needed */
 151         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 152                 RTE_LOG(WARNING, PMD,
 153                         "Unable to set TUNSETIFF for %s\n",
 154                         ifr.ifr_name);
 155                 perror("TUNSETIFF");
 156                 goto error;
 157         }
 158
 159         /* Always set the file descriptor to non-blocking */
 160         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 161                 RTE_LOG(WARNING, PMD,
 162                         "Unable to set %s to nonblocking\n",
 163                         ifr.ifr_name);
 164                 perror("F_SETFL, NONBLOCK");
 165                 goto error;
 166         }
 167
 168         /* Set up trigger to optimize empty Rx bursts */
 169         errno = 0;
 170         do {
 171                 struct sigaction sa;
 172                 int flags = fcntl(fd, F_GETFL);
 173
 174                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 175                         break;
 176                 if (sa.sa_handler != tap_trigger_cb) {
 177                         /*
 178                          * Make sure SIGIO is not already taken. This is done
 179                          * as late as possible to leave the application a
 180                          * chance to set up its own signal handler first.
 181                          */
 182                         if (sa.sa_handler != SIG_IGN &&
 183                             sa.sa_handler != SIG_DFL) {
 184                                 errno = EBUSY;
 185                                 break;
 186                         }
 187                         sa = (struct sigaction){
 188                                 .sa_flags = SA_RESTART,
 189                                 .sa_handler = tap_trigger_cb,
 190                         };
 191                         if (sigaction(SIGIO, &sa, NULL) == -1)
 192                                 break;
 193                 }
 194                 /* Enable SIGIO on file descriptor */
 195                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 196                 fcntl(fd, F_SETOWN, getpid());
 197         } while (0);
 198         if (errno) {
 199                 /* Disable trigger globally in case of error */
 200                 tap_trigger = 0;
 201                 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
 202                         strerror(errno));
 203         }
 204
 205         return fd;
 206
 207 error:
 208         if (fd > 0)
 209                 close(fd);
 210         return -1;
 211 }
 212
 213 static void
 214 tap_verify_csum(struct rte_mbuf *mbuf)
 215 {
 216         uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
 217         uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
 218         uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
 219         unsigned int l2_len = sizeof(struct ether_hdr);
 220         unsigned int l3_len;
 221         uint16_t cksum = 0;
 222         void *l3_hdr;
 223         void *l4_hdr;
 224
 225         if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
 226                 l2_len += 4;
 227         else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
 228                 l2_len += 8;
 229         /* Don't verify checksum for packets with discontinuous L2 header */
 230         if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
 231                      rte_pktmbuf_data_len(mbuf)))
 232                 return;
 233         l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
 234         if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 235                 struct ipv4_hdr *iph = l3_hdr;
 236
 237                 /* ihl contains the number of 4-byte words in the header */
 238                 l3_len = 4 * (iph->version_ihl & 0xf);
 239                 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
 240                         return;
 241
 242                 cksum = ~rte_raw_cksum(iph, l3_len);
 243                 mbuf->ol_flags |= cksum ?
 244                         PKT_RX_IP_CKSUM_BAD :
 245                         PKT_RX_IP_CKSUM_GOOD;
 246         } else if (l3 == RTE_PTYPE_L3_IPV6) {
 247                 l3_len = sizeof(struct ipv6_hdr);
 248         } else {
 249                 /* IPv6 extensions are not supported */
 250                 return;
 251         }
 252         if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
 253                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 254                 /* Don't verify checksum for multi-segment packets. */
 255                 if (mbuf->nb_segs > 1)
 256                         return;
 257                 if (l3 == RTE_PTYPE_L3_IPV4)
 258                         cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
 259                 else if (l3 == RTE_PTYPE_L3_IPV6)
 260                         cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
 261                 mbuf->ol_flags |= cksum ?
 262                         PKT_RX_L4_CKSUM_BAD :
 263                         PKT_RX_L4_CKSUM_GOOD;
 264         }
 265 }
 266
 267 static uint64_t
 268 tap_rx_offload_get_port_capa(void)
 269 {
 270         /*
 271          * No specific port Rx offload capabilities.
 272          */
 273         return 0;
 274 }
 275
 276 static uint64_t
 277 tap_rx_offload_get_queue_capa(void)
 278 {
 279         return DEV_RX_OFFLOAD_SCATTER |
 280                DEV_RX_OFFLOAD_IPV4_CKSUM |
 281                DEV_RX_OFFLOAD_UDP_CKSUM |
 282                DEV_RX_OFFLOAD_TCP_CKSUM |
 283                DEV_RX_OFFLOAD_CRC_STRIP;
 284 }
 285
 286 static bool
 287 tap_rxq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 288 {
 289         uint64_t port_offloads = dev->data->dev_conf.rxmode.offloads;
 290         uint64_t queue_supp_offloads = tap_rx_offload_get_queue_capa();
 291         uint64_t port_supp_offloads = tap_rx_offload_get_port_capa();
 292
 293         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 294             offloads)
 295                 return false;
 296         if ((port_offloads ^ offloads) & port_supp_offloads)
 297                 return false;
 298         return true;
 299 }
 300
 301 /* Callback to handle the rx burst of packets to the correct interface and
 302  * file descriptor(s) in a multi-queue setup.
 303  */
 304 static uint16_t
 305 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 306 {
 307         struct rx_queue *rxq = queue;
 308         uint16_t num_rx;
 309         unsigned long num_rx_bytes = 0;
 310         uint32_t trigger = tap_trigger;
 311
 312         if (trigger == rxq->trigger_seen)
 313                 return 0;
 314         if (trigger)
 315                 rxq->trigger_seen = trigger;
 316         rte_compiler_barrier();
 317         for (num_rx = 0; num_rx < nb_pkts; ) {
 318                 struct rte_mbuf *mbuf = rxq->pool;
 319                 struct rte_mbuf *seg = NULL;
 320                 struct rte_mbuf *new_tail = NULL;
 321                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 322                 int len;
 323
 324                 len = readv(rxq->fd, *rxq->iovecs,
 325                             1 +
 326                             (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
 327                              rxq->nb_rx_desc : 1));
 328                 if (len < (int)sizeof(struct tun_pi))
 329                         break;
 330
 331                 /* Packet couldn't fit in the provided mbuf */
 332                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 333                         rxq->stats.ierrors++;
 334                         continue;
 335                 }
 336
 337                 len -= sizeof(struct tun_pi);
 338
 339                 mbuf->pkt_len = len;
 340                 mbuf->port = rxq->in_port;
 341                 while (1) {
 342                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 343
 344                         if (unlikely(!buf)) {
 345                                 rxq->stats.rx_nombuf++;
 346                                 /* No new buf has been allocated: do nothing */
 347                                 if (!new_tail || !seg)
 348                                         goto end;
 349
 350                                 seg->next = NULL;
 351                                 rte_pktmbuf_free(mbuf);
 352
 353                                 goto end;
 354                         }
 355                         seg = seg ? seg->next : mbuf;
 356                         if (rxq->pool == mbuf)
 357                                 rxq->pool = buf;
 358                         if (new_tail)
 359                                 new_tail->next = buf;
 360                         new_tail = buf;
 361                         new_tail->next = seg->next;
 362
 363                         /* iovecs[0] is reserved for packet info (pi) */
 364                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 365                                 buf->buf_len - data_off;
 366                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 367                                 (char *)buf->buf_addr + data_off;
 368
 369                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 370                         seg->data_off = data_off;
 371
 372                         len -= seg->data_len;
 373                         if (len <= 0)
 374                                 break;
 375                         mbuf->nb_segs++;
 376                         /* First segment has headroom, not the others */
 377                         data_off = 0;
 378                 }
 379                 seg->next = NULL;
 380                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 381                                                       RTE_PTYPE_ALL_MASK);
 382                 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
 383                         tap_verify_csum(mbuf);
 384
 385                 /* account for the receive frame */
 386                 bufs[num_rx++] = mbuf;
 387                 num_rx_bytes += mbuf->pkt_len;
 388         }
 389 end:
 390         rxq->stats.ipackets += num_rx;
 391         rxq->stats.ibytes += num_rx_bytes;
 392
 393         return num_rx;
 394 }
 395
 396 static uint64_t
 397 tap_tx_offload_get_port_capa(void)
 398 {
 399         /*
 400          * No specific port Tx offload capabilities.
 401          */
 402         return 0;
 403 }
 404
 405 static uint64_t
 406 tap_tx_offload_get_queue_capa(void)
 407 {
 408         return DEV_TX_OFFLOAD_MULTI_SEGS |
 409                DEV_TX_OFFLOAD_IPV4_CKSUM |
 410                DEV_TX_OFFLOAD_UDP_CKSUM |
 411                DEV_TX_OFFLOAD_TCP_CKSUM;
 412 }
 413
 414 static bool
 415 tap_txq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 416 {
 417         uint64_t port_offloads = dev->data->dev_conf.txmode.offloads;
 418         uint64_t queue_supp_offloads = tap_tx_offload_get_queue_capa();
 419         uint64_t port_supp_offloads = tap_tx_offload_get_port_capa();
 420
 421         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 422             offloads)
 423                 return false;
 424         /* Verify we have no conflict with port offloads */
 425         if ((port_offloads ^ offloads) & port_supp_offloads)
 426                 return false;
 427         return true;
 428 }
 429
 430 static void
 431 tap_tx_offload(char *packet, uint64_t ol_flags, unsigned int l2_len,
 432                unsigned int l3_len)
 433 {
 434         void *l3_hdr = packet + l2_len;
 435
 436         if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
 437                 struct ipv4_hdr *iph = l3_hdr;
 438                 uint16_t cksum;
 439
 440                 iph->hdr_checksum = 0;
 441                 cksum = rte_raw_cksum(iph, l3_len);
 442                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 443         }
 444         if (ol_flags & PKT_TX_L4_MASK) {
 445                 uint16_t l4_len;
 446                 uint32_t cksum;
 447                 uint16_t *l4_cksum;
 448                 void *l4_hdr;
 449
 450                 l4_hdr = packet + l2_len + l3_len;
 451                 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
 452                         l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
 453                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 454                         l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
 455                 else
 456                         return;
 457                 *l4_cksum = 0;
 458                 if (ol_flags & PKT_TX_IPV4) {
 459                         struct ipv4_hdr *iph = l3_hdr;
 460
 461                         l4_len = rte_be_to_cpu_16(iph->total_length) - l3_len;
 462                         cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 463                 } else {
 464                         struct ipv6_hdr *ip6h = l3_hdr;
 465
 466                         /* payload_len does not include ext headers */
 467                         l4_len = rte_be_to_cpu_16(ip6h->payload_len) -
 468                                 l3_len + sizeof(struct ipv6_hdr);
 469                         cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 470                 }
 471                 cksum += rte_raw_cksum(l4_hdr, l4_len);
 472                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 473                 cksum = (~cksum) & 0xffff;
 474                 if (cksum == 0)
 475                         cksum = 0xffff;
 476                 *l4_cksum = cksum;
 477         }
 478 }
 479
 480 /* Callback to handle sending packets from the tap interface
 481  */
 482 static uint16_t
 483 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 484 {
 485         struct tx_queue *txq = queue;
 486         uint16_t num_tx = 0;
 487         unsigned long num_tx_bytes = 0;
 488         uint32_t max_size;
 489         int i;
 490
 491         if (unlikely(nb_pkts == 0))
 492                 return 0;
 493
 494         max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
 495         for (i = 0; i < nb_pkts; i++) {
 496                 struct rte_mbuf *mbuf = bufs[num_tx];
 497                 struct iovec iovecs[mbuf->nb_segs + 1];
 498                 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
 499                 struct rte_mbuf *seg = mbuf;
 500                 char m_copy[mbuf->data_len];
 501                 int n;
 502                 int j;
 503
 504                 /* stats.errs will be incremented */
 505                 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
 506                         break;
 507
 508                 /*
 509                  * TUN and TAP are created with IFF_NO_PI disabled.
 510                  * For TUN PMD this mandatory as fields are used by
 511                  * Kernel tun.c to determine whether its IP or non IP
 512                  * packets.
 513                  *
 514                  * The logic fetches the first byte of data from mbuf.
 515                  * compares whether its v4 or v6. If none matches default
 516                  * value 0x00 is taken for protocol field.
 517                  */
 518                 char *buff_data = rte_pktmbuf_mtod(seg, void *);
 519                 j = (*buff_data & 0xf0);
 520                 pi.proto = (j == 0x40) ? 0x0008 :
 521                                 (j == 0x60) ? 0xdd86 : 0x00;
 522
 523                 iovecs[0].iov_base = &pi;
 524                 iovecs[0].iov_len = sizeof(pi);
 525                 for (j = 1; j <= mbuf->nb_segs; j++) {
 526                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 527                         iovecs[j].iov_base =
 528                                 rte_pktmbuf_mtod(seg, void *);
 529                         seg = seg->next;
 530                 }
 531                 if (txq->csum &&
 532                     ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 533                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 534                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
 535                         /* Support only packets with all data in the same seg */
 536                         if (mbuf->nb_segs > 1)
 537                                 break;
 538                         /* To change checksums, work on a copy of data. */
 539                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 540                                    rte_pktmbuf_data_len(mbuf));
 541                         tap_tx_offload(m_copy, mbuf->ol_flags,
 542                                        mbuf->l2_len, mbuf->l3_len);
 543                         iovecs[1].iov_base = m_copy;
 544                 }
 545                 /* copy the tx frame data */
 546                 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
 547                 if (n <= 0)
 548                         break;
 549
 550                 num_tx++;
 551                 num_tx_bytes += mbuf->pkt_len;
 552                 rte_pktmbuf_free(mbuf);
 553         }
 554
 555         txq->stats.opackets += num_tx;
 556         txq->stats.errs += nb_pkts - num_tx;
 557         txq->stats.obytes += num_tx_bytes;
 558
 559         return num_tx;
 560 }
 561
 562 static const char *
 563 tap_ioctl_req2str(unsigned long request)
 564 {
 565         switch (request) {
 566         case SIOCSIFFLAGS:
 567                 return "SIOCSIFFLAGS";
 568         case SIOCGIFFLAGS:
 569                 return "SIOCGIFFLAGS";
 570         case SIOCGIFHWADDR:
 571                 return "SIOCGIFHWADDR";
 572         case SIOCSIFHWADDR:
 573                 return "SIOCSIFHWADDR";
 574         case SIOCSIFMTU:
 575                 return "SIOCSIFMTU";
 576         }
 577         return "UNKNOWN";
 578 }
 579
 580 static int
 581 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 582           struct ifreq *ifr, int set, enum ioctl_mode mode)
 583 {
 584         short req_flags = ifr->ifr_flags;
 585         int remote = pmd->remote_if_index &&
 586                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 587
 588         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 589                 return 0;
 590         /*
 591          * If there is a remote netdevice, apply ioctl on it, then apply it on
 592          * the tap netdevice.
 593          */
 594 apply:
 595         if (remote)
 596                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
 597         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 598                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 599         switch (request) {
 600         case SIOCSIFFLAGS:
 601                 /* fetch current flags to leave other flags untouched */
 602                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 603                         goto error;
 604                 if (set)
 605                         ifr->ifr_flags |= req_flags;
 606                 else
 607                         ifr->ifr_flags &= ~req_flags;
 608                 break;
 609         case SIOCGIFFLAGS:
 610         case SIOCGIFHWADDR:
 611         case SIOCSIFHWADDR:
 612         case SIOCSIFMTU:
 613                 break;
 614         default:
 615                 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
 616                         pmd->name);
 617                 return -EINVAL;
 618         }
 619         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 620                 goto error;
 621         if (remote-- && mode == LOCAL_AND_REMOTE)
 622                 goto apply;
 623         return 0;
 624
 625 error:
 626         RTE_LOG(DEBUG, PMD, "%s: %s(%s) failed: %s(%d)\n", ifr->ifr_name,
 627                 __func__, tap_ioctl_req2str(request), strerror(errno), errno);
 628         return -errno;
 629 }
 630
 631 static int
 632 tap_link_set_down(struct rte_eth_dev *dev)
 633 {
 634         struct pmd_internals *pmd = dev->data->dev_private;
 635         struct ifreq ifr = { .ifr_flags = IFF_UP };
 636
 637         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 638         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 639 }
 640
 641 static int
 642 tap_link_set_up(struct rte_eth_dev *dev)
 643 {
 644         struct pmd_internals *pmd = dev->data->dev_private;
 645         struct ifreq ifr = { .ifr_flags = IFF_UP };
 646
 647         dev->data->dev_link.link_status = ETH_LINK_UP;
 648         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 649 }
 650
 651 static int
 652 tap_dev_start(struct rte_eth_dev *dev)
 653 {
 654         int err;
 655
 656         err = tap_intr_handle_set(dev, 1);
 657         if (err)
 658                 return err;
 659         return tap_link_set_up(dev);
 660 }
 661
 662 /* This function gets called when the current port gets stopped.
 663  */
 664 static void
 665 tap_dev_stop(struct rte_eth_dev *dev)
 666 {
 667         tap_intr_handle_set(dev, 0);
 668         tap_link_set_down(dev);
 669 }
 670
 671 static int
 672 tap_dev_configure(struct rte_eth_dev *dev)
 673 {
 674         uint64_t supp_tx_offloads = tap_tx_offload_get_port_capa() |
 675                                 tap_tx_offload_get_queue_capa();
 676         uint64_t tx_offloads = dev->data->dev_conf.txmode.offloads;
 677
 678         if ((tx_offloads & supp_tx_offloads) != tx_offloads) {
 679                 rte_errno = ENOTSUP;
 680                 RTE_LOG(ERR, PMD,
 681                         "Some Tx offloads are not supported "
 682                         "requested 0x%" PRIx64 " supported 0x%" PRIx64 "\n",
 683                         tx_offloads, supp_tx_offloads);
 684                 return -rte_errno;
 685         }
 686         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 687                 RTE_LOG(ERR, PMD,
 688                         "%s: number of rx queues %d exceeds max num of queues %d\n",
 689                         dev->device->name,
 690                         dev->data->nb_rx_queues,
 691                         RTE_PMD_TAP_MAX_QUEUES);
 692                 return -1;
 693         }
 694         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 695                 RTE_LOG(ERR, PMD,
 696                         "%s: number of tx queues %d exceeds max num of queues %d\n",
 697                         dev->device->name,
 698                         dev->data->nb_tx_queues,
 699                         RTE_PMD_TAP_MAX_QUEUES);
 700                 return -1;
 701         }
 702
 703         RTE_LOG(INFO, PMD, "%s: %p: TX configured queues number: %u\n",
 704              dev->device->name, (void *)dev, dev->data->nb_tx_queues);
 705
 706         RTE_LOG(INFO, PMD, "%s: %p: RX configured queues number: %u\n",
 707              dev->device->name, (void *)dev, dev->data->nb_rx_queues);
 708
 709         return 0;
 710 }
 711
 712 static uint32_t
 713 tap_dev_speed_capa(void)
 714 {
 715         uint32_t speed = pmd_link.link_speed;
 716         uint32_t capa = 0;
 717
 718         if (speed >= ETH_SPEED_NUM_10M)
 719                 capa |= ETH_LINK_SPEED_10M;
 720         if (speed >= ETH_SPEED_NUM_100M)
 721                 capa |= ETH_LINK_SPEED_100M;
 722         if (speed >= ETH_SPEED_NUM_1G)
 723                 capa |= ETH_LINK_SPEED_1G;
 724         if (speed >= ETH_SPEED_NUM_5G)
 725                 capa |= ETH_LINK_SPEED_2_5G;
 726         if (speed >= ETH_SPEED_NUM_5G)
 727                 capa |= ETH_LINK_SPEED_5G;
 728         if (speed >= ETH_SPEED_NUM_10G)
 729                 capa |= ETH_LINK_SPEED_10G;
 730         if (speed >= ETH_SPEED_NUM_20G)
 731                 capa |= ETH_LINK_SPEED_20G;
 732         if (speed >= ETH_SPEED_NUM_25G)
 733                 capa |= ETH_LINK_SPEED_25G;
 734         if (speed >= ETH_SPEED_NUM_40G)
 735                 capa |= ETH_LINK_SPEED_40G;
 736         if (speed >= ETH_SPEED_NUM_50G)
 737                 capa |= ETH_LINK_SPEED_50G;
 738         if (speed >= ETH_SPEED_NUM_56G)
 739                 capa |= ETH_LINK_SPEED_56G;
 740         if (speed >= ETH_SPEED_NUM_100G)
 741                 capa |= ETH_LINK_SPEED_100G;
 742
 743         return capa;
 744 }
 745
 746 static void
 747 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 748 {
 749         struct pmd_internals *internals = dev->data->dev_private;
 750
 751         dev_info->if_index = internals->if_index;
 752         dev_info->max_mac_addrs = 1;
 753         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 754         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
 755         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
 756         dev_info->min_rx_bufsize = 0;
 757         dev_info->speed_capa = tap_dev_speed_capa();
 758         dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
 759         dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
 760                                     dev_info->rx_queue_offload_capa;
 761         dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
 762         dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
 763                                     dev_info->tx_queue_offload_capa;
 764 }
 765
 766 static int
 767 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 768 {
 769         unsigned int i, imax;
 770         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 771         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 772         unsigned long rx_nombuf = 0, ierrors = 0;
 773         const struct pmd_internals *pmd = dev->data->dev_private;
 774
 775         /* rx queue statistics */
 776         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 777                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 778         for (i = 0; i < imax; i++) {
 779                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 780                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 781                 rx_total += tap_stats->q_ipackets[i];
 782                 rx_bytes_total += tap_stats->q_ibytes[i];
 783                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
 784                 ierrors += pmd->rxq[i].stats.ierrors;
 785         }
 786
 787         /* tx queue statistics */
 788         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 789                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 790
 791         for (i = 0; i < imax; i++) {
 792                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 793                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 794                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 795                 tx_total += tap_stats->q_opackets[i];
 796                 tx_err_total += tap_stats->q_errors[i];
 797                 tx_bytes_total += tap_stats->q_obytes[i];
 798         }
 799
 800         tap_stats->ipackets = rx_total;
 801         tap_stats->ibytes = rx_bytes_total;
 802         tap_stats->ierrors = ierrors;
 803         tap_stats->rx_nombuf = rx_nombuf;
 804         tap_stats->opackets = tx_total;
 805         tap_stats->oerrors = tx_err_total;
 806         tap_stats->obytes = tx_bytes_total;
 807         return 0;
 808 }
 809
 810 static void
 811 tap_stats_reset(struct rte_eth_dev *dev)
 812 {
 813         int i;
 814         struct pmd_internals *pmd = dev->data->dev_private;
 815
 816         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 817                 pmd->rxq[i].stats.ipackets = 0;
 818                 pmd->rxq[i].stats.ibytes = 0;
 819                 pmd->rxq[i].stats.ierrors = 0;
 820                 pmd->rxq[i].stats.rx_nombuf = 0;
 821
 822                 pmd->txq[i].stats.opackets = 0;
 823                 pmd->txq[i].stats.errs = 0;
 824                 pmd->txq[i].stats.obytes = 0;
 825         }
 826 }
 827
 828 static void
 829 tap_dev_close(struct rte_eth_dev *dev)
 830 {
 831         int i;
 832         struct pmd_internals *internals = dev->data->dev_private;
 833
 834         tap_link_set_down(dev);
 835         tap_flow_flush(dev, NULL);
 836         tap_flow_implicit_flush(internals, NULL);
 837
 838         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 839                 if (internals->rxq[i].fd != -1) {
 840                         close(internals->rxq[i].fd);
 841                         internals->rxq[i].fd = -1;
 842                 }
 843                 if (internals->txq[i].fd != -1) {
 844                         close(internals->txq[i].fd);
 845                         internals->txq[i].fd = -1;
 846                 }
 847         }
 848
 849         if (internals->remote_if_index) {
 850                 /* Restore initial remote state */
 851                 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
 852                                 &internals->remote_initial_flags);
 853         }
 854 }
 855
 856 static void
 857 tap_rx_queue_release(void *queue)
 858 {
 859         struct rx_queue *rxq = queue;
 860
 861         if (rxq && (rxq->fd > 0)) {
 862                 close(rxq->fd);
 863                 rxq->fd = -1;
 864                 rte_pktmbuf_free(rxq->pool);
 865                 rte_free(rxq->iovecs);
 866                 rxq->pool = NULL;
 867                 rxq->iovecs = NULL;
 868         }
 869 }
 870
 871 static void
 872 tap_tx_queue_release(void *queue)
 873 {
 874         struct tx_queue *txq = queue;
 875
 876         if (txq && (txq->fd > 0)) {
 877                 close(txq->fd);
 878                 txq->fd = -1;
 879         }
 880 }
 881
 882 static int
 883 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
 884 {
 885         struct rte_eth_link *dev_link = &dev->data->dev_link;
 886         struct pmd_internals *pmd = dev->data->dev_private;
 887         struct ifreq ifr = { .ifr_flags = 0 };
 888
 889         if (pmd->remote_if_index) {
 890                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
 891                 if (!(ifr.ifr_flags & IFF_UP) ||
 892                     !(ifr.ifr_flags & IFF_RUNNING)) {
 893                         dev_link->link_status = ETH_LINK_DOWN;
 894                         return 0;
 895                 }
 896         }
 897         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
 898         dev_link->link_status =
 899                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
 900                  ETH_LINK_UP :
 901                  ETH_LINK_DOWN);
 902         return 0;
 903 }
 904
 905 static void
 906 tap_promisc_enable(struct rte_eth_dev *dev)
 907 {
 908         struct pmd_internals *pmd = dev->data->dev_private;
 909         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 910
 911         dev->data->promiscuous = 1;
 912         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 913         if (pmd->remote_if_index && !pmd->flow_isolate)
 914                 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
 915 }
 916
 917 static void
 918 tap_promisc_disable(struct rte_eth_dev *dev)
 919 {
 920         struct pmd_internals *pmd = dev->data->dev_private;
 921         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 922
 923         dev->data->promiscuous = 0;
 924         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 925         if (pmd->remote_if_index && !pmd->flow_isolate)
 926                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
 927 }
 928
 929 static void
 930 tap_allmulti_enable(struct rte_eth_dev *dev)
 931 {
 932         struct pmd_internals *pmd = dev->data->dev_private;
 933         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 934
 935         dev->data->all_multicast = 1;
 936         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 937         if (pmd->remote_if_index && !pmd->flow_isolate)
 938                 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
 939 }
 940
 941 static void
 942 tap_allmulti_disable(struct rte_eth_dev *dev)
 943 {
 944         struct pmd_internals *pmd = dev->data->dev_private;
 945         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 946
 947         dev->data->all_multicast = 0;
 948         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 949         if (pmd->remote_if_index && !pmd->flow_isolate)
 950                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
 951 }
 952
 953 static int
 954 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 955 {
 956         struct pmd_internals *pmd = dev->data->dev_private;
 957         enum ioctl_mode mode = LOCAL_ONLY;
 958         struct ifreq ifr;
 959         int ret;
 960
 961         if (is_zero_ether_addr(mac_addr)) {
 962                 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
 963                         dev->device->name);
 964                 return -EINVAL;
 965         }
 966         /* Check the actual current MAC address on the tap netdevice */
 967         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
 968         if (ret < 0)
 969                 return ret;
 970         if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 971                                mac_addr))
 972                 return 0;
 973         /* Check the current MAC address on the remote */
 974         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
 975         if (ret < 0)
 976                 return ret;
 977         if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 978                                mac_addr))
 979                 mode = LOCAL_AND_REMOTE;
 980         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
 981         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
 982         ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
 983         if (ret < 0)
 984                 return ret;
 985         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
 986         if (pmd->remote_if_index && !pmd->flow_isolate) {
 987                 /* Replace MAC redirection rule after a MAC change */
 988                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
 989                 if (ret < 0) {
 990                         RTE_LOG(ERR, PMD,
 991                                 "%s: Couldn't delete MAC redirection rule\n",
 992                                 dev->device->name);
 993                         return ret;
 994                 }
 995                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
 996                 if (ret < 0) {
 997                         RTE_LOG(ERR, PMD,
 998                                 "%s: Couldn't add MAC redirection rule\n",
 999                                 dev->device->name);
1000                         return ret;
1001                 }
1002         }
1003
1004         return 0;
1005 }
1006
1007 static int
1008 tap_setup_queue(struct rte_eth_dev *dev,
1009                 struct pmd_internals *internals,
1010                 uint16_t qid,
1011                 int is_rx)
1012 {
1013         int *fd;
1014         int *other_fd;
1015         const char *dir;
1016         struct pmd_internals *pmd = dev->data->dev_private;
1017         struct rx_queue *rx = &internals->rxq[qid];
1018         struct tx_queue *tx = &internals->txq[qid];
1019
1020         if (is_rx) {
1021                 fd = &rx->fd;
1022                 other_fd = &tx->fd;
1023                 dir = "rx";
1024         } else {
1025                 fd = &tx->fd;
1026                 other_fd = &rx->fd;
1027                 dir = "tx";
1028         }
1029         if (*fd != -1) {
1030                 /* fd for this queue already exists */
1031                 RTE_LOG(DEBUG, PMD, "%s: fd %d for %s queue qid %d exists\n",
1032                         pmd->name, *fd, dir, qid);
1033         } else if (*other_fd != -1) {
1034                 /* Only other_fd exists. dup it */
1035                 *fd = dup(*other_fd);
1036                 if (*fd < 0) {
1037                         *fd = -1;
1038                         RTE_LOG(ERR, PMD, "%s: dup() failed.\n",
1039                                 pmd->name);
1040                         return -1;
1041                 }
1042                 RTE_LOG(DEBUG, PMD, "%s: dup fd %d for %s queue qid %d (%d)\n",
1043                         pmd->name, *other_fd, dir, qid, *fd);
1044         } else {
1045                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1046                 *fd = tun_alloc(pmd);
1047                 if (*fd < 0) {
1048                         *fd = -1; /* restore original value */
1049                         RTE_LOG(ERR, PMD, "%s: tun_alloc() failed.\n",
1050                                 pmd->name);
1051                         return -1;
1052                 }
1053                 RTE_LOG(DEBUG, PMD, "%s: add %s queue for qid %d fd %d\n",
1054                         pmd->name, dir, qid, *fd);
1055         }
1056
1057         tx->mtu = &dev->data->mtu;
1058         rx->rxmode = &dev->data->dev_conf.rxmode;
1059
1060         return *fd;
1061 }
1062
1063 static int
1064 tap_rx_queue_setup(struct rte_eth_dev *dev,
1065                    uint16_t rx_queue_id,
1066                    uint16_t nb_rx_desc,
1067                    unsigned int socket_id,
1068                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1069                    struct rte_mempool *mp)
1070 {
1071         struct pmd_internals *internals = dev->data->dev_private;
1072         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1073         struct rte_mbuf **tmp = &rxq->pool;
1074         long iov_max = sysconf(_SC_IOV_MAX);
1075         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1076         struct iovec (*iovecs)[nb_desc + 1];
1077         int data_off = RTE_PKTMBUF_HEADROOM;
1078         int ret = 0;
1079         int fd;
1080         int i;
1081
1082         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1083                 RTE_LOG(WARNING, PMD,
1084                         "nb_rx_queues %d too small or mempool NULL\n",
1085                         dev->data->nb_rx_queues);
1086                 return -1;
1087         }
1088
1089         /* Verify application offloads are valid for our port and queue. */
1090         if (!tap_rxq_are_offloads_valid(dev, rx_conf->offloads)) {
1091                 rte_errno = ENOTSUP;
1092                 RTE_LOG(ERR, PMD,
1093                         "%p: Rx queue offloads 0x%" PRIx64
1094                         " don't match port offloads 0x%" PRIx64
1095                         " or supported offloads 0x%" PRIx64 "\n",
1096                         (void *)dev, rx_conf->offloads,
1097                         dev->data->dev_conf.rxmode.offloads,
1098                         (tap_rx_offload_get_port_capa() |
1099                          tap_rx_offload_get_queue_capa()));
1100                 return -rte_errno;
1101         }
1102         rxq->mp = mp;
1103         rxq->trigger_seen = 1; /* force initial burst */
1104         rxq->in_port = dev->data->port_id;
1105         rxq->nb_rx_desc = nb_desc;
1106         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1107                                     socket_id);
1108         if (!iovecs) {
1109                 RTE_LOG(WARNING, PMD,
1110                         "%s: Couldn't allocate %d RX descriptors\n",
1111                         dev->device->name, nb_desc);
1112                 return -ENOMEM;
1113         }
1114         rxq->iovecs = iovecs;
1115
1116         dev->data->rx_queues[rx_queue_id] = rxq;
1117         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1118         if (fd == -1) {
1119                 ret = fd;
1120                 goto error;
1121         }
1122
1123         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1124         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1125
1126         for (i = 1; i <= nb_desc; i++) {
1127                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1128                 if (!*tmp) {
1129                         RTE_LOG(WARNING, PMD,
1130                                 "%s: couldn't allocate memory for queue %d\n",
1131                                 dev->device->name, rx_queue_id);
1132                         ret = -ENOMEM;
1133                         goto error;
1134                 }
1135                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1136                 (*rxq->iovecs)[i].iov_base =
1137                         (char *)(*tmp)->buf_addr + data_off;
1138                 data_off = 0;
1139                 tmp = &(*tmp)->next;
1140         }
1141
1142         RTE_LOG(DEBUG, PMD, "  RX TUNTAP device name %s, qid %d on fd %d\n",
1143                 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
1144
1145         return 0;
1146
1147 error:
1148         rte_pktmbuf_free(rxq->pool);
1149         rxq->pool = NULL;
1150         rte_free(rxq->iovecs);
1151         rxq->iovecs = NULL;
1152         return ret;
1153 }
1154
1155 static int
1156 tap_tx_queue_setup(struct rte_eth_dev *dev,
1157                    uint16_t tx_queue_id,
1158                    uint16_t nb_tx_desc __rte_unused,
1159                    unsigned int socket_id __rte_unused,
1160                    const struct rte_eth_txconf *tx_conf)
1161 {
1162         struct pmd_internals *internals = dev->data->dev_private;
1163         struct tx_queue *txq;
1164         int ret;
1165
1166         if (tx_queue_id >= dev->data->nb_tx_queues)
1167                 return -1;
1168         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1169         txq = dev->data->tx_queues[tx_queue_id];
1170         /*
1171          * Don't verify port offloads for application which
1172          * use the old API.
1173          */
1174         if (tx_conf != NULL &&
1175             !!(tx_conf->txq_flags & ETH_TXQ_FLAGS_IGNORE)) {
1176                 if (tap_txq_are_offloads_valid(dev, tx_conf->offloads)) {
1177                         txq->csum = !!(tx_conf->offloads &
1178                                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
1179                                          DEV_TX_OFFLOAD_UDP_CKSUM |
1180                                          DEV_TX_OFFLOAD_TCP_CKSUM));
1181                 } else {
1182                         rte_errno = ENOTSUP;
1183                         RTE_LOG(ERR, PMD,
1184                                 "%p: Tx queue offloads 0x%" PRIx64
1185                                 " don't match port offloads 0x%" PRIx64
1186                                 " or supported offloads 0x%" PRIx64,
1187                                 (void *)dev, tx_conf->offloads,
1188                                 dev->data->dev_conf.txmode.offloads,
1189                                 (tap_tx_offload_get_port_capa() |
1190                                 tap_tx_offload_get_queue_capa()));
1191                         return -rte_errno;
1192                 }
1193         }
1194         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1195         if (ret == -1)
1196                 return -1;
1197         RTE_LOG(DEBUG, PMD,
1198                 "  TX TUNTAP device name %s, qid %d on fd %d csum %s\n",
1199                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd,
1200                 txq->csum ? "on" : "off");
1201
1202         return 0;
1203 }
1204
1205 static int
1206 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1207 {
1208         struct pmd_internals *pmd = dev->data->dev_private;
1209         struct ifreq ifr = { .ifr_mtu = mtu };
1210         int err = 0;
1211
1212         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1213         if (!err)
1214                 dev->data->mtu = mtu;
1215
1216         return err;
1217 }
1218
1219 static int
1220 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1221                      struct ether_addr *mc_addr_set __rte_unused,
1222                      uint32_t nb_mc_addr __rte_unused)
1223 {
1224         /*
1225          * Nothing to do actually: the tap has no filtering whatsoever, every
1226          * packet is received.
1227          */
1228         return 0;
1229 }
1230
1231 static int
1232 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1233 {
1234         struct rte_eth_dev *dev = arg;
1235         struct pmd_internals *pmd = dev->data->dev_private;
1236         struct ifinfomsg *info = NLMSG_DATA(nh);
1237
1238         if (nh->nlmsg_type != RTM_NEWLINK ||
1239             (info->ifi_index != pmd->if_index &&
1240              info->ifi_index != pmd->remote_if_index))
1241                 return 0;
1242         return tap_link_update(dev, 0);
1243 }
1244
1245 static void
1246 tap_dev_intr_handler(void *cb_arg)
1247 {
1248         struct rte_eth_dev *dev = cb_arg;
1249         struct pmd_internals *pmd = dev->data->dev_private;
1250
1251         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1252 }
1253
1254 static int
1255 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1256 {
1257         struct pmd_internals *pmd = dev->data->dev_private;
1258
1259         /* In any case, disable interrupt if the conf is no longer there. */
1260         if (!dev->data->dev_conf.intr_conf.lsc) {
1261                 if (pmd->intr_handle.fd != -1) {
1262                         tap_nl_final(pmd->intr_handle.fd);
1263                         rte_intr_callback_unregister(&pmd->intr_handle,
1264                                 tap_dev_intr_handler, dev);
1265                 }
1266                 return 0;
1267         }
1268         if (set) {
1269                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1270                 if (unlikely(pmd->intr_handle.fd == -1))
1271                         return -EBADF;
1272                 return rte_intr_callback_register(
1273                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1274         }
1275         tap_nl_final(pmd->intr_handle.fd);
1276         return rte_intr_callback_unregister(&pmd->intr_handle,
1277                                             tap_dev_intr_handler, dev);
1278 }
1279
1280 static int
1281 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1282 {
1283         int err;
1284
1285         err = tap_lsc_intr_handle_set(dev, set);
1286         if (err)
1287                 return err;
1288         err = tap_rx_intr_vec_set(dev, set);
1289         if (err && set)
1290                 tap_lsc_intr_handle_set(dev, 0);
1291         return err;
1292 }
1293
1294 static const uint32_t*
1295 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1296 {
1297         static const uint32_t ptypes[] = {
1298                 RTE_PTYPE_INNER_L2_ETHER,
1299                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1300                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1301                 RTE_PTYPE_INNER_L3_IPV4,
1302                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1303                 RTE_PTYPE_INNER_L3_IPV6,
1304                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1305                 RTE_PTYPE_INNER_L4_FRAG,
1306                 RTE_PTYPE_INNER_L4_UDP,
1307                 RTE_PTYPE_INNER_L4_TCP,
1308                 RTE_PTYPE_INNER_L4_SCTP,
1309                 RTE_PTYPE_L2_ETHER,
1310                 RTE_PTYPE_L2_ETHER_VLAN,
1311                 RTE_PTYPE_L2_ETHER_QINQ,
1312                 RTE_PTYPE_L3_IPV4,
1313                 RTE_PTYPE_L3_IPV4_EXT,
1314                 RTE_PTYPE_L3_IPV6_EXT,
1315                 RTE_PTYPE_L3_IPV6,
1316                 RTE_PTYPE_L4_FRAG,
1317                 RTE_PTYPE_L4_UDP,
1318                 RTE_PTYPE_L4_TCP,
1319                 RTE_PTYPE_L4_SCTP,
1320         };
1321
1322         return ptypes;
1323 }
1324
1325 static int
1326 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1327                   struct rte_eth_fc_conf *fc_conf)
1328 {
1329         fc_conf->mode = RTE_FC_NONE;
1330         return 0;
1331 }
1332
1333 static int
1334 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1335                   struct rte_eth_fc_conf *fc_conf)
1336 {
1337         if (fc_conf->mode != RTE_FC_NONE)
1338                 return -ENOTSUP;
1339         return 0;
1340 }
1341
1342 static const struct eth_dev_ops ops = {
1343         .dev_start              = tap_dev_start,
1344         .dev_stop               = tap_dev_stop,
1345         .dev_close              = tap_dev_close,
1346         .dev_configure          = tap_dev_configure,
1347         .dev_infos_get          = tap_dev_info,
1348         .rx_queue_setup         = tap_rx_queue_setup,
1349         .tx_queue_setup         = tap_tx_queue_setup,
1350         .rx_queue_release       = tap_rx_queue_release,
1351         .tx_queue_release       = tap_tx_queue_release,
1352         .flow_ctrl_get          = tap_flow_ctrl_get,
1353         .flow_ctrl_set          = tap_flow_ctrl_set,
1354         .link_update            = tap_link_update,
1355         .dev_set_link_up        = tap_link_set_up,
1356         .dev_set_link_down      = tap_link_set_down,
1357         .promiscuous_enable     = tap_promisc_enable,
1358         .promiscuous_disable    = tap_promisc_disable,
1359         .allmulticast_enable    = tap_allmulti_enable,
1360         .allmulticast_disable   = tap_allmulti_disable,
1361         .mac_addr_set           = tap_mac_set,
1362         .mtu_set                = tap_mtu_set,
1363         .set_mc_addr_list       = tap_set_mc_addr_list,
1364         .stats_get              = tap_stats_get,
1365         .stats_reset            = tap_stats_reset,
1366         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1367         .filter_ctrl            = tap_dev_filter_ctrl,
1368 };
1369
1370 static int
1371 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1372                    char *remote_iface, struct ether_addr *mac_addr)
1373 {
1374         int numa_node = rte_socket_id();
1375         struct rte_eth_dev *dev;
1376         struct pmd_internals *pmd;
1377         struct rte_eth_dev_data *data;
1378         struct ifreq ifr;
1379         int i;
1380
1381         RTE_LOG(DEBUG, PMD, "%s device on numa %u\n",
1382                         tuntap_name, rte_socket_id());
1383
1384         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1385         if (!dev) {
1386                 RTE_LOG(ERR, PMD, "%s Unable to allocate device struct\n",
1387                                 tuntap_name);
1388                 goto error_exit_nodev;
1389         }
1390
1391         pmd = dev->data->dev_private;
1392         pmd->dev = dev;
1393         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1394
1395         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1396         if (pmd->ioctl_sock == -1) {
1397                 RTE_LOG(ERR, PMD,
1398                         "%s Unable to get a socket for management: %s\n",
1399                         tuntap_name, strerror(errno));
1400                 goto error_exit;
1401         }
1402
1403         /* Setup some default values */
1404         data = dev->data;
1405         data->dev_private = pmd;
1406         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1407         data->numa_node = numa_node;
1408
1409         data->dev_link = pmd_link;
1410         data->mac_addrs = &pmd->eth_addr;
1411         /* Set the number of RX and TX queues */
1412         data->nb_rx_queues = 0;
1413         data->nb_tx_queues = 0;
1414
1415         dev->dev_ops = &ops;
1416         dev->rx_pkt_burst = pmd_rx_burst;
1417         dev->tx_pkt_burst = pmd_tx_burst;
1418
1419         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1420         pmd->intr_handle.fd = -1;
1421         dev->intr_handle = &pmd->intr_handle;
1422
1423         /* Presetup the fds to -1 as being not valid */
1424         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1425                 pmd->rxq[i].fd = -1;
1426                 pmd->txq[i].fd = -1;
1427         }
1428
1429         if (tap_type) {
1430                 if (is_zero_ether_addr(mac_addr))
1431                         eth_random_addr((uint8_t *)&pmd->eth_addr);
1432                 else
1433                         rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1434         }
1435
1436         /* Immediately create the netdevice (this will create the 1st queue). */
1437         /* rx queue */
1438         if (tap_setup_queue(dev, pmd, 0, 1) == -1)
1439                 goto error_exit;
1440         /* tx queue */
1441         if (tap_setup_queue(dev, pmd, 0, 0) == -1)
1442                 goto error_exit;
1443
1444         ifr.ifr_mtu = dev->data->mtu;
1445         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1446                 goto error_exit;
1447
1448         if (tap_type) {
1449                 memset(&ifr, 0, sizeof(struct ifreq));
1450                 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1451                 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1452                                 ETHER_ADDR_LEN);
1453                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1454                         goto error_exit;
1455         }
1456
1457         /*
1458          * Set up everything related to rte_flow:
1459          * - netlink socket
1460          * - tap / remote if_index
1461          * - mandatory QDISCs
1462          * - rte_flow actual/implicit lists
1463          * - implicit rules
1464          */
1465         pmd->nlsk_fd = tap_nl_init(0);
1466         if (pmd->nlsk_fd == -1) {
1467                 RTE_LOG(WARNING, PMD, "%s: failed to create netlink socket.\n",
1468                         pmd->name);
1469                 goto disable_rte_flow;
1470         }
1471         pmd->if_index = if_nametoindex(pmd->name);
1472         if (!pmd->if_index) {
1473                 RTE_LOG(ERR, PMD, "%s: failed to get if_index.\n", pmd->name);
1474                 goto disable_rte_flow;
1475         }
1476         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1477                 RTE_LOG(ERR, PMD, "%s: failed to create multiq qdisc.\n",
1478                         pmd->name);
1479                 goto disable_rte_flow;
1480         }
1481         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1482                 RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1483                         pmd->name);
1484                 goto disable_rte_flow;
1485         }
1486         LIST_INIT(&pmd->flows);
1487
1488         if (strlen(remote_iface)) {
1489                 pmd->remote_if_index = if_nametoindex(remote_iface);
1490                 if (!pmd->remote_if_index) {
1491                         RTE_LOG(ERR, PMD, "%s: failed to get %s if_index.\n",
1492                                 pmd->name, remote_iface);
1493                         goto error_remote;
1494                 }
1495                 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1496                          "%s", remote_iface);
1497
1498                 /* Save state of remote device */
1499                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1500
1501                 /* Replicate remote MAC address */
1502                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1503                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1504                                 pmd->name, pmd->remote_iface);
1505                         goto error_remote;
1506                 }
1507                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1508                            ETHER_ADDR_LEN);
1509                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1510                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1511                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1512                                 pmd->name, remote_iface);
1513                         goto error_remote;
1514                 }
1515
1516                 /*
1517                  * Flush usually returns negative value because it tries to
1518                  * delete every QDISC (and on a running device, one QDISC at
1519                  * least is needed). Ignore negative return value.
1520                  */
1521                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1522                 if (qdisc_create_ingress(pmd->nlsk_fd,
1523                                          pmd->remote_if_index) < 0) {
1524                         RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1525                                 pmd->remote_iface);
1526                         goto error_remote;
1527                 }
1528                 LIST_INIT(&pmd->implicit_flows);
1529                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1530                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1531                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1532                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1533                         RTE_LOG(ERR, PMD,
1534                                 "%s: failed to create implicit rules.\n",
1535                                 pmd->name);
1536                         goto error_remote;
1537                 }
1538         }
1539
1540         return 0;
1541
1542 disable_rte_flow:
1543         RTE_LOG(ERR, PMD, " Disabling rte flow support: %s(%d)\n",
1544                 strerror(errno), errno);
1545         if (strlen(remote_iface)) {
1546                 RTE_LOG(ERR, PMD, "Remote feature requires flow support.\n");
1547                 goto error_exit;
1548         }
1549         return 0;
1550
1551 error_remote:
1552         RTE_LOG(ERR, PMD, " Can't set up remote feature: %s(%d)\n",
1553                 strerror(errno), errno);
1554         tap_flow_implicit_flush(pmd, NULL);
1555
1556 error_exit:
1557         if (pmd->ioctl_sock > 0)
1558                 close(pmd->ioctl_sock);
1559         rte_eth_dev_release_port(dev);
1560
1561 error_exit_nodev:
1562         RTE_LOG(ERR, PMD, "%s Unable to initialize %s\n",
1563                 tuntap_name, rte_vdev_device_name(vdev));
1564
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         struct rte_eth_dev *eth_dev;
1720
1721         tap_type = 1;
1722         strcpy(tuntap_name, "TAP");
1723
1724         name = rte_vdev_device_name(dev);
1725         params = rte_vdev_device_args(dev);
1726
1727         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
1728             strlen(params) == 0) {
1729                 eth_dev = rte_eth_dev_attach_secondary(name);
1730                 if (!eth_dev) {
1731                         RTE_LOG(ERR, PMD, "Failed to probe %s\n", name);
1732                         return -1;
1733                 }
1734                 /* TODO: request info from primary to set up Rx and Tx */
1735                 eth_dev->dev_ops = &ops;
1736                 return 0;
1737         }
1738
1739         speed = ETH_SPEED_NUM_10G;
1740         snprintf(tap_name, sizeof(tap_name), "%s%d",
1741                  DEFAULT_TAP_NAME, tap_unit++);
1742         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1743
1744         if (params && (params[0] != '\0')) {
1745                 RTE_LOG(DEBUG, PMD, "parameters (%s)\n", params);
1746
1747                 kvlist = rte_kvargs_parse(params, valid_arguments);
1748                 if (kvlist) {
1749                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1750                                 ret = rte_kvargs_process(kvlist,
1751                                                          ETH_TAP_IFACE_ARG,
1752                                                          &set_interface_name,
1753                                                          tap_name);
1754                                 if (ret == -1)
1755                                         goto leave;
1756                         }
1757
1758                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1759                                 ret = rte_kvargs_process(kvlist,
1760                                                          ETH_TAP_REMOTE_ARG,
1761                                                          &set_remote_iface,
1762                                                          remote_iface);
1763                                 if (ret == -1)
1764                                         goto leave;
1765                         }
1766
1767                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1768                                 ret = rte_kvargs_process(kvlist,
1769                                                          ETH_TAP_MAC_ARG,
1770                                                          &set_mac_type,
1771                                                          &user_mac);
1772                                 if (ret == -1)
1773                                         goto leave;
1774                         }
1775                 }
1776         }
1777         pmd_link.link_speed = speed;
1778
1779         RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1780                 name, tap_name);
1781
1782         ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac);
1783
1784 leave:
1785         if (ret == -1) {
1786                 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1787                         name, tap_name);
1788                 tap_unit--;             /* Restore the unit number */
1789         }
1790         rte_kvargs_free(kvlist);
1791
1792         return ret;
1793 }
1794
1795 /* detach a TUNTAP device.
1796  */
1797 static int
1798 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1799 {
1800         struct rte_eth_dev *eth_dev = NULL;
1801         struct pmd_internals *internals;
1802         int i;
1803
1804         RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1805                 rte_socket_id());
1806
1807         /* find the ethdev entry */
1808         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1809         if (!eth_dev)
1810                 return 0;
1811
1812         internals = eth_dev->data->dev_private;
1813         if (internals->nlsk_fd) {
1814                 tap_flow_flush(eth_dev, NULL);
1815                 tap_flow_implicit_flush(internals, NULL);
1816                 tap_nl_final(internals->nlsk_fd);
1817         }
1818         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1819                 if (internals->rxq[i].fd != -1) {
1820                         close(internals->rxq[i].fd);
1821                         internals->rxq[i].fd = -1;
1822                 }
1823                 if (internals->txq[i].fd != -1) {
1824                         close(internals->txq[i].fd);
1825                         internals->txq[i].fd = -1;
1826                 }
1827         }
1828
1829         close(internals->ioctl_sock);
1830         rte_free(eth_dev->data->dev_private);
1831
1832         rte_eth_dev_release_port(eth_dev);
1833
1834         return 0;
1835 }
1836
1837 static struct rte_vdev_driver pmd_tun_drv = {
1838         .probe = rte_pmd_tun_probe,
1839         .remove = rte_pmd_tap_remove,
1840 };
1841
1842 static struct rte_vdev_driver pmd_tap_drv = {
1843         .probe = rte_pmd_tap_probe,
1844         .remove = rte_pmd_tap_remove,
1845 };
1846 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1847 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
1848 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1849 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
1850                               ETH_TAP_IFACE_ARG "=<string> ");
1851 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1852                               ETH_TAP_IFACE_ARG "=<string> "
1853                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
1854                               ETH_TAP_REMOTE_ARG "=<string>");