drivers/net/tap/rte_eth_tap.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2016-2017 Intel Corporation
   3  */
   4
   5 #include <rte_atomic.h>
   6 #include <rte_branch_prediction.h>
   7 #include <rte_byteorder.h>
   8 #include <rte_common.h>
   9 #include <rte_mbuf.h>
  10 #include <rte_ethdev_driver.h>
  11 #include <rte_ethdev_vdev.h>
  12 #include <rte_malloc.h>
  13 #include <rte_bus_vdev.h>
  14 #include <rte_kvargs.h>
  15 #include <rte_net.h>
  16 #include <rte_debug.h>
  17 #include <rte_ip.h>
  18 #include <rte_string_fns.h>
  19
  20 #include <sys/types.h>
  21 #include <sys/stat.h>
  22 #include <sys/socket.h>
  23 #include <sys/ioctl.h>
  24 #include <sys/utsname.h>
  25 #include <sys/mman.h>
  26 #include <errno.h>
  27 #include <signal.h>
  28 #include <stdbool.h>
  29 #include <stdint.h>
  30 #include <sys/uio.h>
  31 #include <unistd.h>
  32 #include <arpa/inet.h>
  33 #include <net/if.h>
  34 #include <linux/if_tun.h>
  35 #include <linux/if_ether.h>
  36 #include <fcntl.h>
  37
  38 #include <rte_eth_tap.h>
  39 #include <tap_flow.h>
  40 #include <tap_netlink.h>
  41 #include <tap_tcmsgs.h>
  42
  43 /* Linux based path to the TUN device */
  44 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  45 #define DEFAULT_TAP_NAME        "dtap"
  46 #define DEFAULT_TUN_NAME        "dtun"
  47
  48 #define ETH_TAP_IFACE_ARG       "iface"
  49 #define ETH_TAP_REMOTE_ARG      "remote"
  50 #define ETH_TAP_MAC_ARG         "mac"
  51 #define ETH_TAP_MAC_FIXED       "fixed"
  52
  53 #define ETH_TAP_USR_MAC_FMT     "xx:xx:xx:xx:xx:xx"
  54 #define ETH_TAP_CMP_MAC_FMT     "0123456789ABCDEFabcdef"
  55 #define ETH_TAP_MAC_ARG_FMT     ETH_TAP_MAC_FIXED "|" ETH_TAP_USR_MAC_FMT
  56
  57 static struct rte_vdev_driver pmd_tap_drv;
  58 static struct rte_vdev_driver pmd_tun_drv;
  59
  60 static const char *valid_arguments[] = {
  61         ETH_TAP_IFACE_ARG,
  62         ETH_TAP_REMOTE_ARG,
  63         ETH_TAP_MAC_ARG,
  64         NULL
  65 };
  66
  67 static int tap_unit;
  68 static int tun_unit;
  69
  70 static int tap_type;
  71 static char tuntap_name[8];
  72
  73 static volatile uint32_t tap_trigger;   /* Rx trigger */
  74
  75 static struct rte_eth_link pmd_link = {
  76         .link_speed = ETH_SPEED_NUM_10G,
  77         .link_duplex = ETH_LINK_FULL_DUPLEX,
  78         .link_status = ETH_LINK_DOWN,
  79         .link_autoneg = ETH_LINK_AUTONEG
  80 };
  81
  82 static void
  83 tap_trigger_cb(int sig __rte_unused)
  84 {
  85         /* Valid trigger values are nonzero */
  86         tap_trigger = (tap_trigger + 1) | 0x80000000;
  87 }
  88
  89 /* Specifies on what netdevices the ioctl should be applied */
  90 enum ioctl_mode {
  91         LOCAL_AND_REMOTE,
  92         LOCAL_ONLY,
  93         REMOTE_ONLY,
  94 };
  95
  96 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
  97
  98 /* Tun/Tap allocation routine
  99  *
 100  * name is the number of the interface to use, unless NULL to take the host
 101  * supplied name.
 102  */
 103 static int
 104 tun_alloc(struct pmd_internals *pmd)
 105 {
 106         struct ifreq ifr;
 107 #ifdef IFF_MULTI_QUEUE
 108         unsigned int features;
 109 #endif
 110         int fd;
 111
 112         memset(&ifr, 0, sizeof(struct ifreq));
 113
 114         /*
 115          * Do not set IFF_NO_PI as packet information header will be needed
 116          * to check if a received packet has been truncated.
 117          */
 118         ifr.ifr_flags = (tap_type) ? IFF_TAP : IFF_TUN | IFF_POINTOPOINT;
 119         snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
 120
 121         RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
 122
 123         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 124         if (fd < 0) {
 125                 RTE_LOG(ERR, PMD, "Unable to create %s interface\n",
 126                                 tuntap_name);
 127                 goto error;
 128         }
 129
 130 #ifdef IFF_MULTI_QUEUE
 131         /* Grab the TUN features to verify we can work multi-queue */
 132         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 133                 RTE_LOG(ERR, PMD, "%s unable to get TUN/TAP features\n",
 134                                 tuntap_name);
 135                 goto error;
 136         }
 137         RTE_LOG(DEBUG, PMD, "%s Features %08x\n", tuntap_name, features);
 138
 139         if (features & IFF_MULTI_QUEUE) {
 140                 RTE_LOG(DEBUG, PMD, "  Multi-queue support for %d queues\n",
 141                         RTE_PMD_TAP_MAX_QUEUES);
 142                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 143         } else
 144 #endif
 145         {
 146                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 147                 RTE_LOG(DEBUG, PMD, "  Single queue only support\n");
 148         }
 149
 150         /* Set the TUN/TAP configuration and set the name if needed */
 151         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 152                 RTE_LOG(WARNING, PMD,
 153                         "Unable to set TUNSETIFF for %s\n",
 154                         ifr.ifr_name);
 155                 perror("TUNSETIFF");
 156                 goto error;
 157         }
 158
 159         /* Always set the file descriptor to non-blocking */
 160         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 161                 RTE_LOG(WARNING, PMD,
 162                         "Unable to set %s to nonblocking\n",
 163                         ifr.ifr_name);
 164                 perror("F_SETFL, NONBLOCK");
 165                 goto error;
 166         }
 167
 168         /* Set up trigger to optimize empty Rx bursts */
 169         errno = 0;
 170         do {
 171                 struct sigaction sa;
 172                 int flags = fcntl(fd, F_GETFL);
 173
 174                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 175                         break;
 176                 if (sa.sa_handler != tap_trigger_cb) {
 177                         /*
 178                          * Make sure SIGIO is not already taken. This is done
 179                          * as late as possible to leave the application a
 180                          * chance to set up its own signal handler first.
 181                          */
 182                         if (sa.sa_handler != SIG_IGN &&
 183                             sa.sa_handler != SIG_DFL) {
 184                                 errno = EBUSY;
 185                                 break;
 186                         }
 187                         sa = (struct sigaction){
 188                                 .sa_flags = SA_RESTART,
 189                                 .sa_handler = tap_trigger_cb,
 190                         };
 191                         if (sigaction(SIGIO, &sa, NULL) == -1)
 192                                 break;
 193                 }
 194                 /* Enable SIGIO on file descriptor */
 195                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 196                 fcntl(fd, F_SETOWN, getpid());
 197         } while (0);
 198         if (errno) {
 199                 /* Disable trigger globally in case of error */
 200                 tap_trigger = 0;
 201                 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
 202                         strerror(errno));
 203         }
 204
 205         return fd;
 206
 207 error:
 208         if (fd > 0)
 209                 close(fd);
 210         return -1;
 211 }
 212
 213 static void
 214 tap_verify_csum(struct rte_mbuf *mbuf)
 215 {
 216         uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
 217         uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
 218         uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
 219         unsigned int l2_len = sizeof(struct ether_hdr);
 220         unsigned int l3_len;
 221         uint16_t cksum = 0;
 222         void *l3_hdr;
 223         void *l4_hdr;
 224
 225         if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
 226                 l2_len += 4;
 227         else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
 228                 l2_len += 8;
 229         /* Don't verify checksum for packets with discontinuous L2 header */
 230         if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
 231                      rte_pktmbuf_data_len(mbuf)))
 232                 return;
 233         l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
 234         if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 235                 struct ipv4_hdr *iph = l3_hdr;
 236
 237                 /* ihl contains the number of 4-byte words in the header */
 238                 l3_len = 4 * (iph->version_ihl & 0xf);
 239                 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
 240                         return;
 241
 242                 cksum = ~rte_raw_cksum(iph, l3_len);
 243                 mbuf->ol_flags |= cksum ?
 244                         PKT_RX_IP_CKSUM_BAD :
 245                         PKT_RX_IP_CKSUM_GOOD;
 246         } else if (l3 == RTE_PTYPE_L3_IPV6) {
 247                 l3_len = sizeof(struct ipv6_hdr);
 248         } else {
 249                 /* IPv6 extensions are not supported */
 250                 return;
 251         }
 252         if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
 253                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 254                 /* Don't verify checksum for multi-segment packets. */
 255                 if (mbuf->nb_segs > 1)
 256                         return;
 257                 if (l3 == RTE_PTYPE_L3_IPV4)
 258                         cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
 259                 else if (l3 == RTE_PTYPE_L3_IPV6)
 260                         cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
 261                 mbuf->ol_flags |= cksum ?
 262                         PKT_RX_L4_CKSUM_BAD :
 263                         PKT_RX_L4_CKSUM_GOOD;
 264         }
 265 }
 266
 267 static uint64_t
 268 tap_rx_offload_get_port_capa(void)
 269 {
 270         /*
 271          * In order to support legacy apps,
 272          * report capabilities also as port capabilities.
 273          */
 274         return DEV_RX_OFFLOAD_SCATTER |
 275                DEV_RX_OFFLOAD_IPV4_CKSUM |
 276                DEV_RX_OFFLOAD_UDP_CKSUM |
 277                DEV_RX_OFFLOAD_TCP_CKSUM |
 278                DEV_RX_OFFLOAD_CRC_STRIP;
 279 }
 280
 281 static uint64_t
 282 tap_rx_offload_get_queue_capa(void)
 283 {
 284         return DEV_RX_OFFLOAD_SCATTER |
 285                DEV_RX_OFFLOAD_IPV4_CKSUM |
 286                DEV_RX_OFFLOAD_UDP_CKSUM |
 287                DEV_RX_OFFLOAD_TCP_CKSUM |
 288                DEV_RX_OFFLOAD_CRC_STRIP;
 289 }
 290
 291 static bool
 292 tap_rxq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 293 {
 294         uint64_t port_offloads = dev->data->dev_conf.rxmode.offloads;
 295         uint64_t queue_supp_offloads = tap_rx_offload_get_queue_capa();
 296         uint64_t port_supp_offloads = tap_rx_offload_get_port_capa();
 297
 298         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 299             offloads)
 300                 return false;
 301         if ((port_offloads ^ offloads) & port_supp_offloads)
 302                 return false;
 303         return true;
 304 }
 305
 306 /* Callback to handle the rx burst of packets to the correct interface and
 307  * file descriptor(s) in a multi-queue setup.
 308  */
 309 static uint16_t
 310 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 311 {
 312         struct rx_queue *rxq = queue;
 313         uint16_t num_rx;
 314         unsigned long num_rx_bytes = 0;
 315         uint32_t trigger = tap_trigger;
 316
 317         if (trigger == rxq->trigger_seen)
 318                 return 0;
 319         if (trigger)
 320                 rxq->trigger_seen = trigger;
 321         rte_compiler_barrier();
 322         for (num_rx = 0; num_rx < nb_pkts; ) {
 323                 struct rte_mbuf *mbuf = rxq->pool;
 324                 struct rte_mbuf *seg = NULL;
 325                 struct rte_mbuf *new_tail = NULL;
 326                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 327                 int len;
 328
 329                 len = readv(rxq->fd, *rxq->iovecs,
 330                             1 +
 331                             (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
 332                              rxq->nb_rx_desc : 1));
 333                 if (len < (int)sizeof(struct tun_pi))
 334                         break;
 335
 336                 /* Packet couldn't fit in the provided mbuf */
 337                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 338                         rxq->stats.ierrors++;
 339                         continue;
 340                 }
 341
 342                 len -= sizeof(struct tun_pi);
 343
 344                 mbuf->pkt_len = len;
 345                 mbuf->port = rxq->in_port;
 346                 while (1) {
 347                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 348
 349                         if (unlikely(!buf)) {
 350                                 rxq->stats.rx_nombuf++;
 351                                 /* No new buf has been allocated: do nothing */
 352                                 if (!new_tail || !seg)
 353                                         goto end;
 354
 355                                 seg->next = NULL;
 356                                 rte_pktmbuf_free(mbuf);
 357
 358                                 goto end;
 359                         }
 360                         seg = seg ? seg->next : mbuf;
 361                         if (rxq->pool == mbuf)
 362                                 rxq->pool = buf;
 363                         if (new_tail)
 364                                 new_tail->next = buf;
 365                         new_tail = buf;
 366                         new_tail->next = seg->next;
 367
 368                         /* iovecs[0] is reserved for packet info (pi) */
 369                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 370                                 buf->buf_len - data_off;
 371                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 372                                 (char *)buf->buf_addr + data_off;
 373
 374                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 375                         seg->data_off = data_off;
 376
 377                         len -= seg->data_len;
 378                         if (len <= 0)
 379                                 break;
 380                         mbuf->nb_segs++;
 381                         /* First segment has headroom, not the others */
 382                         data_off = 0;
 383                 }
 384                 seg->next = NULL;
 385                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 386                                                       RTE_PTYPE_ALL_MASK);
 387                 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
 388                         tap_verify_csum(mbuf);
 389
 390                 /* account for the receive frame */
 391                 bufs[num_rx++] = mbuf;
 392                 num_rx_bytes += mbuf->pkt_len;
 393         }
 394 end:
 395         rxq->stats.ipackets += num_rx;
 396         rxq->stats.ibytes += num_rx_bytes;
 397
 398         return num_rx;
 399 }
 400
 401 static uint64_t
 402 tap_tx_offload_get_port_capa(void)
 403 {
 404         /*
 405          * In order to support legacy apps,
 406          * report capabilities also as port capabilities.
 407          */
 408         return DEV_TX_OFFLOAD_MULTI_SEGS |
 409                DEV_TX_OFFLOAD_IPV4_CKSUM |
 410                DEV_TX_OFFLOAD_UDP_CKSUM |
 411                DEV_TX_OFFLOAD_TCP_CKSUM;
 412 }
 413
 414 static uint64_t
 415 tap_tx_offload_get_queue_capa(void)
 416 {
 417         return DEV_TX_OFFLOAD_MULTI_SEGS |
 418                DEV_TX_OFFLOAD_IPV4_CKSUM |
 419                DEV_TX_OFFLOAD_UDP_CKSUM |
 420                DEV_TX_OFFLOAD_TCP_CKSUM;
 421 }
 422
 423 static bool
 424 tap_txq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 425 {
 426         uint64_t port_offloads = dev->data->dev_conf.txmode.offloads;
 427         uint64_t queue_supp_offloads = tap_tx_offload_get_queue_capa();
 428         uint64_t port_supp_offloads = tap_tx_offload_get_port_capa();
 429
 430         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 431             offloads)
 432                 return false;
 433         /* Verify we have no conflict with port offloads */
 434         if ((port_offloads ^ offloads) & port_supp_offloads)
 435                 return false;
 436         return true;
 437 }
 438
 439 static void
 440 tap_tx_offload(char *packet, uint64_t ol_flags, unsigned int l2_len,
 441                unsigned int l3_len)
 442 {
 443         void *l3_hdr = packet + l2_len;
 444
 445         if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
 446                 struct ipv4_hdr *iph = l3_hdr;
 447                 uint16_t cksum;
 448
 449                 iph->hdr_checksum = 0;
 450                 cksum = rte_raw_cksum(iph, l3_len);
 451                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 452         }
 453         if (ol_flags & PKT_TX_L4_MASK) {
 454                 uint16_t l4_len;
 455                 uint32_t cksum;
 456                 uint16_t *l4_cksum;
 457                 void *l4_hdr;
 458
 459                 l4_hdr = packet + l2_len + l3_len;
 460                 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
 461                         l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
 462                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 463                         l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
 464                 else
 465                         return;
 466                 *l4_cksum = 0;
 467                 if (ol_flags & PKT_TX_IPV4) {
 468                         struct ipv4_hdr *iph = l3_hdr;
 469
 470                         l4_len = rte_be_to_cpu_16(iph->total_length) - l3_len;
 471                         cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 472                 } else {
 473                         struct ipv6_hdr *ip6h = l3_hdr;
 474
 475                         /* payload_len does not include ext headers */
 476                         l4_len = rte_be_to_cpu_16(ip6h->payload_len) -
 477                                 l3_len + sizeof(struct ipv6_hdr);
 478                         cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 479                 }
 480                 cksum += rte_raw_cksum(l4_hdr, l4_len);
 481                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 482                 cksum = (~cksum) & 0xffff;
 483                 if (cksum == 0)
 484                         cksum = 0xffff;
 485                 *l4_cksum = cksum;
 486         }
 487 }
 488
 489 /* Callback to handle sending packets from the tap interface
 490  */
 491 static uint16_t
 492 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 493 {
 494         struct tx_queue *txq = queue;
 495         uint16_t num_tx = 0;
 496         unsigned long num_tx_bytes = 0;
 497         uint32_t max_size;
 498         int i;
 499
 500         if (unlikely(nb_pkts == 0))
 501                 return 0;
 502
 503         max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
 504         for (i = 0; i < nb_pkts; i++) {
 505                 struct rte_mbuf *mbuf = bufs[num_tx];
 506                 struct iovec iovecs[mbuf->nb_segs + 1];
 507                 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
 508                 struct rte_mbuf *seg = mbuf;
 509                 char m_copy[mbuf->data_len];
 510                 int n;
 511                 int j;
 512
 513                 /* stats.errs will be incremented */
 514                 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
 515                         break;
 516
 517                 /*
 518                  * TUN and TAP are created with IFF_NO_PI disabled.
 519                  * For TUN PMD this mandatory as fields are used by
 520                  * Kernel tun.c to determine whether its IP or non IP
 521                  * packets.
 522                  *
 523                  * The logic fetches the first byte of data from mbuf.
 524                  * compares whether its v4 or v6. If none matches default
 525                  * value 0x00 is taken for protocol field.
 526                  */
 527                 char *buff_data = rte_pktmbuf_mtod(seg, void *);
 528                 j = (*buff_data & 0xf0);
 529                 if (j & (0x40 | 0x60))
 530                         pi.proto = (j == 0x40) ? 0x0008 : 0xdd86;
 531
 532                 iovecs[0].iov_base = &pi;
 533                 iovecs[0].iov_len = sizeof(pi);
 534                 for (j = 1; j <= mbuf->nb_segs; j++) {
 535                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 536                         iovecs[j].iov_base =
 537                                 rte_pktmbuf_mtod(seg, void *);
 538                         seg = seg->next;
 539                 }
 540                 if (txq->csum &&
 541                     ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 542                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 543                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
 544                         /* Support only packets with all data in the same seg */
 545                         if (mbuf->nb_segs > 1)
 546                                 break;
 547                         /* To change checksums, work on a copy of data. */
 548                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 549                                    rte_pktmbuf_data_len(mbuf));
 550                         tap_tx_offload(m_copy, mbuf->ol_flags,
 551                                        mbuf->l2_len, mbuf->l3_len);
 552                         iovecs[1].iov_base = m_copy;
 553                 }
 554                 /* copy the tx frame data */
 555                 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
 556                 if (n <= 0)
 557                         break;
 558
 559                 num_tx++;
 560                 num_tx_bytes += mbuf->pkt_len;
 561                 rte_pktmbuf_free(mbuf);
 562         }
 563
 564         txq->stats.opackets += num_tx;
 565         txq->stats.errs += nb_pkts - num_tx;
 566         txq->stats.obytes += num_tx_bytes;
 567
 568         return num_tx;
 569 }
 570
 571 static const char *
 572 tap_ioctl_req2str(unsigned long request)
 573 {
 574         switch (request) {
 575         case SIOCSIFFLAGS:
 576                 return "SIOCSIFFLAGS";
 577         case SIOCGIFFLAGS:
 578                 return "SIOCGIFFLAGS";
 579         case SIOCGIFHWADDR:
 580                 return "SIOCGIFHWADDR";
 581         case SIOCSIFHWADDR:
 582                 return "SIOCSIFHWADDR";
 583         case SIOCSIFMTU:
 584                 return "SIOCSIFMTU";
 585         }
 586         return "UNKNOWN";
 587 }
 588
 589 static int
 590 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 591           struct ifreq *ifr, int set, enum ioctl_mode mode)
 592 {
 593         short req_flags = ifr->ifr_flags;
 594         int remote = pmd->remote_if_index &&
 595                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 596
 597         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 598                 return 0;
 599         /*
 600          * If there is a remote netdevice, apply ioctl on it, then apply it on
 601          * the tap netdevice.
 602          */
 603 apply:
 604         if (remote)
 605                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
 606         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 607                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 608         switch (request) {
 609         case SIOCSIFFLAGS:
 610                 /* fetch current flags to leave other flags untouched */
 611                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 612                         goto error;
 613                 if (set)
 614                         ifr->ifr_flags |= req_flags;
 615                 else
 616                         ifr->ifr_flags &= ~req_flags;
 617                 break;
 618         case SIOCGIFFLAGS:
 619         case SIOCGIFHWADDR:
 620         case SIOCSIFHWADDR:
 621         case SIOCSIFMTU:
 622                 break;
 623         default:
 624                 RTE_ASSERT(!"unsupported request type: must not happen");
 625         }
 626         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 627                 goto error;
 628         if (remote-- && mode == LOCAL_AND_REMOTE)
 629                 goto apply;
 630         return 0;
 631
 632 error:
 633         RTE_LOG(DEBUG, PMD, "%s: %s(%s) failed: %s(%d)\n", ifr->ifr_name,
 634                 __func__, tap_ioctl_req2str(request), strerror(errno), errno);
 635         return -errno;
 636 }
 637
 638 static int
 639 tap_link_set_down(struct rte_eth_dev *dev)
 640 {
 641         struct pmd_internals *pmd = dev->data->dev_private;
 642         struct ifreq ifr = { .ifr_flags = IFF_UP };
 643
 644         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 645         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 646 }
 647
 648 static int
 649 tap_link_set_up(struct rte_eth_dev *dev)
 650 {
 651         struct pmd_internals *pmd = dev->data->dev_private;
 652         struct ifreq ifr = { .ifr_flags = IFF_UP };
 653
 654         dev->data->dev_link.link_status = ETH_LINK_UP;
 655         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 656 }
 657
 658 static int
 659 tap_dev_start(struct rte_eth_dev *dev)
 660 {
 661         int err;
 662
 663         err = tap_intr_handle_set(dev, 1);
 664         if (err)
 665                 return err;
 666         return tap_link_set_up(dev);
 667 }
 668
 669 /* This function gets called when the current port gets stopped.
 670  */
 671 static void
 672 tap_dev_stop(struct rte_eth_dev *dev)
 673 {
 674         tap_intr_handle_set(dev, 0);
 675         tap_link_set_down(dev);
 676 }
 677
 678 static int
 679 tap_dev_configure(struct rte_eth_dev *dev)
 680 {
 681         uint64_t supp_tx_offloads = tap_tx_offload_get_port_capa();
 682         uint64_t tx_offloads = dev->data->dev_conf.txmode.offloads;
 683
 684         if ((tx_offloads & supp_tx_offloads) != tx_offloads) {
 685                 rte_errno = ENOTSUP;
 686                 RTE_LOG(ERR, PMD,
 687                         "Some Tx offloads are not supported "
 688                         "requested 0x%" PRIx64 " supported 0x%" PRIx64 "\n",
 689                         tx_offloads, supp_tx_offloads);
 690                 return -rte_errno;
 691         }
 692         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 693                 RTE_LOG(ERR, PMD,
 694                         "%s: number of rx queues %d exceeds max num of queues %d\n",
 695                         dev->device->name,
 696                         dev->data->nb_rx_queues,
 697                         RTE_PMD_TAP_MAX_QUEUES);
 698                 return -1;
 699         }
 700         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 701                 RTE_LOG(ERR, PMD,
 702                         "%s: number of tx queues %d exceeds max num of queues %d\n",
 703                         dev->device->name,
 704                         dev->data->nb_tx_queues,
 705                         RTE_PMD_TAP_MAX_QUEUES);
 706                 return -1;
 707         }
 708
 709         RTE_LOG(INFO, PMD, "%s: %p: TX configured queues number: %u\n",
 710              dev->device->name, (void *)dev, dev->data->nb_tx_queues);
 711
 712         RTE_LOG(INFO, PMD, "%s: %p: RX configured queues number: %u\n",
 713              dev->device->name, (void *)dev, dev->data->nb_rx_queues);
 714
 715         return 0;
 716 }
 717
 718 static uint32_t
 719 tap_dev_speed_capa(void)
 720 {
 721         uint32_t speed = pmd_link.link_speed;
 722         uint32_t capa = 0;
 723
 724         if (speed >= ETH_SPEED_NUM_10M)
 725                 capa |= ETH_LINK_SPEED_10M;
 726         if (speed >= ETH_SPEED_NUM_100M)
 727                 capa |= ETH_LINK_SPEED_100M;
 728         if (speed >= ETH_SPEED_NUM_1G)
 729                 capa |= ETH_LINK_SPEED_1G;
 730         if (speed >= ETH_SPEED_NUM_5G)
 731                 capa |= ETH_LINK_SPEED_2_5G;
 732         if (speed >= ETH_SPEED_NUM_5G)
 733                 capa |= ETH_LINK_SPEED_5G;
 734         if (speed >= ETH_SPEED_NUM_10G)
 735                 capa |= ETH_LINK_SPEED_10G;
 736         if (speed >= ETH_SPEED_NUM_20G)
 737                 capa |= ETH_LINK_SPEED_20G;
 738         if (speed >= ETH_SPEED_NUM_25G)
 739                 capa |= ETH_LINK_SPEED_25G;
 740         if (speed >= ETH_SPEED_NUM_40G)
 741                 capa |= ETH_LINK_SPEED_40G;
 742         if (speed >= ETH_SPEED_NUM_50G)
 743                 capa |= ETH_LINK_SPEED_50G;
 744         if (speed >= ETH_SPEED_NUM_56G)
 745                 capa |= ETH_LINK_SPEED_56G;
 746         if (speed >= ETH_SPEED_NUM_100G)
 747                 capa |= ETH_LINK_SPEED_100G;
 748
 749         return capa;
 750 }
 751
 752 static void
 753 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 754 {
 755         struct pmd_internals *internals = dev->data->dev_private;
 756
 757         dev_info->if_index = internals->if_index;
 758         dev_info->max_mac_addrs = 1;
 759         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 760         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
 761         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
 762         dev_info->min_rx_bufsize = 0;
 763         dev_info->pci_dev = NULL;
 764         dev_info->speed_capa = tap_dev_speed_capa();
 765         dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
 766         dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
 767                                     dev_info->rx_queue_offload_capa;
 768         dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
 769         dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
 770                                     dev_info->tx_queue_offload_capa;
 771 }
 772
 773 static int
 774 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 775 {
 776         unsigned int i, imax;
 777         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 778         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 779         unsigned long rx_nombuf = 0, ierrors = 0;
 780         const struct pmd_internals *pmd = dev->data->dev_private;
 781
 782         /* rx queue statistics */
 783         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 784                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 785         for (i = 0; i < imax; i++) {
 786                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 787                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 788                 rx_total += tap_stats->q_ipackets[i];
 789                 rx_bytes_total += tap_stats->q_ibytes[i];
 790                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
 791                 ierrors += pmd->rxq[i].stats.ierrors;
 792         }
 793
 794         /* tx queue statistics */
 795         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 796                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 797
 798         for (i = 0; i < imax; i++) {
 799                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 800                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 801                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 802                 tx_total += tap_stats->q_opackets[i];
 803                 tx_err_total += tap_stats->q_errors[i];
 804                 tx_bytes_total += tap_stats->q_obytes[i];
 805         }
 806
 807         tap_stats->ipackets = rx_total;
 808         tap_stats->ibytes = rx_bytes_total;
 809         tap_stats->ierrors = ierrors;
 810         tap_stats->rx_nombuf = rx_nombuf;
 811         tap_stats->opackets = tx_total;
 812         tap_stats->oerrors = tx_err_total;
 813         tap_stats->obytes = tx_bytes_total;
 814         return 0;
 815 }
 816
 817 static void
 818 tap_stats_reset(struct rte_eth_dev *dev)
 819 {
 820         int i;
 821         struct pmd_internals *pmd = dev->data->dev_private;
 822
 823         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 824                 pmd->rxq[i].stats.ipackets = 0;
 825                 pmd->rxq[i].stats.ibytes = 0;
 826                 pmd->rxq[i].stats.ierrors = 0;
 827                 pmd->rxq[i].stats.rx_nombuf = 0;
 828
 829                 pmd->txq[i].stats.opackets = 0;
 830                 pmd->txq[i].stats.errs = 0;
 831                 pmd->txq[i].stats.obytes = 0;
 832         }
 833 }
 834
 835 static void
 836 tap_dev_close(struct rte_eth_dev *dev)
 837 {
 838         int i;
 839         struct pmd_internals *internals = dev->data->dev_private;
 840
 841         tap_link_set_down(dev);
 842         tap_flow_flush(dev, NULL);
 843         tap_flow_implicit_flush(internals, NULL);
 844
 845         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 846                 if (internals->rxq[i].fd != -1) {
 847                         close(internals->rxq[i].fd);
 848                         internals->rxq[i].fd = -1;
 849                 }
 850                 if (internals->txq[i].fd != -1) {
 851                         close(internals->txq[i].fd);
 852                         internals->txq[i].fd = -1;
 853                 }
 854         }
 855
 856         if (internals->remote_if_index) {
 857                 /* Restore initial remote state */
 858                 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
 859                                 &internals->remote_initial_flags);
 860         }
 861 }
 862
 863 static void
 864 tap_rx_queue_release(void *queue)
 865 {
 866         struct rx_queue *rxq = queue;
 867
 868         if (rxq && (rxq->fd > 0)) {
 869                 close(rxq->fd);
 870                 rxq->fd = -1;
 871                 rte_pktmbuf_free(rxq->pool);
 872                 rte_free(rxq->iovecs);
 873                 rxq->pool = NULL;
 874                 rxq->iovecs = NULL;
 875         }
 876 }
 877
 878 static void
 879 tap_tx_queue_release(void *queue)
 880 {
 881         struct tx_queue *txq = queue;
 882
 883         if (txq && (txq->fd > 0)) {
 884                 close(txq->fd);
 885                 txq->fd = -1;
 886         }
 887 }
 888
 889 static int
 890 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
 891 {
 892         struct rte_eth_link *dev_link = &dev->data->dev_link;
 893         struct pmd_internals *pmd = dev->data->dev_private;
 894         struct ifreq ifr = { .ifr_flags = 0 };
 895
 896         if (pmd->remote_if_index) {
 897                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
 898                 if (!(ifr.ifr_flags & IFF_UP) ||
 899                     !(ifr.ifr_flags & IFF_RUNNING)) {
 900                         dev_link->link_status = ETH_LINK_DOWN;
 901                         return 0;
 902                 }
 903         }
 904         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
 905         dev_link->link_status =
 906                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
 907                  ETH_LINK_UP :
 908                  ETH_LINK_DOWN);
 909         return 0;
 910 }
 911
 912 static void
 913 tap_promisc_enable(struct rte_eth_dev *dev)
 914 {
 915         struct pmd_internals *pmd = dev->data->dev_private;
 916         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 917
 918         dev->data->promiscuous = 1;
 919         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 920         if (pmd->remote_if_index && !pmd->flow_isolate)
 921                 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
 922 }
 923
 924 static void
 925 tap_promisc_disable(struct rte_eth_dev *dev)
 926 {
 927         struct pmd_internals *pmd = dev->data->dev_private;
 928         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 929
 930         dev->data->promiscuous = 0;
 931         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 932         if (pmd->remote_if_index && !pmd->flow_isolate)
 933                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
 934 }
 935
 936 static void
 937 tap_allmulti_enable(struct rte_eth_dev *dev)
 938 {
 939         struct pmd_internals *pmd = dev->data->dev_private;
 940         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 941
 942         dev->data->all_multicast = 1;
 943         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 944         if (pmd->remote_if_index && !pmd->flow_isolate)
 945                 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
 946 }
 947
 948 static void
 949 tap_allmulti_disable(struct rte_eth_dev *dev)
 950 {
 951         struct pmd_internals *pmd = dev->data->dev_private;
 952         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 953
 954         dev->data->all_multicast = 0;
 955         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 956         if (pmd->remote_if_index && !pmd->flow_isolate)
 957                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
 958 }
 959
 960 static void
 961 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 962 {
 963         struct pmd_internals *pmd = dev->data->dev_private;
 964         enum ioctl_mode mode = LOCAL_ONLY;
 965         struct ifreq ifr;
 966
 967         if (is_zero_ether_addr(mac_addr)) {
 968                 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
 969                         dev->device->name);
 970                 return;
 971         }
 972         /* Check the actual current MAC address on the tap netdevice */
 973         if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
 974                 return;
 975         if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 976                                mac_addr))
 977                 return;
 978         /* Check the current MAC address on the remote */
 979         if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0)
 980                 return;
 981         if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 982                                mac_addr))
 983                 mode = LOCAL_AND_REMOTE;
 984         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
 985         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
 986         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode) < 0)
 987                 return;
 988         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
 989         if (pmd->remote_if_index && !pmd->flow_isolate) {
 990                 /* Replace MAC redirection rule after a MAC change */
 991                 if (tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC) < 0) {
 992                         RTE_LOG(ERR, PMD,
 993                                 "%s: Couldn't delete MAC redirection rule\n",
 994                                 dev->device->name);
 995                         return;
 996                 }
 997                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0)
 998                         RTE_LOG(ERR, PMD,
 999                                 "%s: Couldn't add MAC redirection rule\n",
1000                                 dev->device->name);
1001         }
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                 RTE_LOG(DEBUG, PMD, "%s: fd %d for %s queue qid %d exists\n",
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                         RTE_LOG(ERR, PMD, "%s: dup() failed.\n",
1036                                 pmd->name);
1037                         return -1;
1038                 }
1039                 RTE_LOG(DEBUG, PMD, "%s: dup fd %d for %s queue qid %d (%d)\n",
1040                         pmd->name, *other_fd, dir, qid, *fd);
1041         } else {
1042                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1043                 *fd = tun_alloc(pmd);
1044                 if (*fd < 0) {
1045                         *fd = -1; /* restore original value */
1046                         RTE_LOG(ERR, PMD, "%s: tun_alloc() failed.\n",
1047                                 pmd->name);
1048                         return -1;
1049                 }
1050                 RTE_LOG(DEBUG, PMD, "%s: add %s queue for qid %d fd %d\n",
1051                         pmd->name, dir, qid, *fd);
1052         }
1053
1054         tx->mtu = &dev->data->mtu;
1055         rx->rxmode = &dev->data->dev_conf.rxmode;
1056
1057         return *fd;
1058 }
1059
1060 static int
1061 tap_rx_queue_setup(struct rte_eth_dev *dev,
1062                    uint16_t rx_queue_id,
1063                    uint16_t nb_rx_desc,
1064                    unsigned int socket_id,
1065                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1066                    struct rte_mempool *mp)
1067 {
1068         struct pmd_internals *internals = dev->data->dev_private;
1069         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1070         struct rte_mbuf **tmp = &rxq->pool;
1071         long iov_max = sysconf(_SC_IOV_MAX);
1072         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1073         struct iovec (*iovecs)[nb_desc + 1];
1074         int data_off = RTE_PKTMBUF_HEADROOM;
1075         int ret = 0;
1076         int fd;
1077         int i;
1078
1079         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1080                 RTE_LOG(WARNING, PMD,
1081                         "nb_rx_queues %d too small or mempool NULL\n",
1082                         dev->data->nb_rx_queues);
1083                 return -1;
1084         }
1085
1086         /* Verify application offloads are valid for our port and queue. */
1087         if (!tap_rxq_are_offloads_valid(dev, rx_conf->offloads)) {
1088                 rte_errno = ENOTSUP;
1089                 RTE_LOG(ERR, PMD,
1090                         "%p: Rx queue offloads 0x%" PRIx64
1091                         " don't match port offloads 0x%" PRIx64
1092                         " or supported offloads 0x%" PRIx64 "\n",
1093                         (void *)dev, rx_conf->offloads,
1094                         dev->data->dev_conf.rxmode.offloads,
1095                         (tap_rx_offload_get_port_capa() |
1096                          tap_rx_offload_get_queue_capa()));
1097                 return -rte_errno;
1098         }
1099         rxq->mp = mp;
1100         rxq->trigger_seen = 1; /* force initial burst */
1101         rxq->in_port = dev->data->port_id;
1102         rxq->nb_rx_desc = nb_desc;
1103         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1104                                     socket_id);
1105         if (!iovecs) {
1106                 RTE_LOG(WARNING, PMD,
1107                         "%s: Couldn't allocate %d RX descriptors\n",
1108                         dev->device->name, nb_desc);
1109                 return -ENOMEM;
1110         }
1111         rxq->iovecs = iovecs;
1112
1113         dev->data->rx_queues[rx_queue_id] = rxq;
1114         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1115         if (fd == -1) {
1116                 ret = fd;
1117                 goto error;
1118         }
1119
1120         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1121         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1122
1123         for (i = 1; i <= nb_desc; i++) {
1124                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1125                 if (!*tmp) {
1126                         RTE_LOG(WARNING, PMD,
1127                                 "%s: couldn't allocate memory for queue %d\n",
1128                                 dev->device->name, rx_queue_id);
1129                         ret = -ENOMEM;
1130                         goto error;
1131                 }
1132                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1133                 (*rxq->iovecs)[i].iov_base =
1134                         (char *)(*tmp)->buf_addr + data_off;
1135                 data_off = 0;
1136                 tmp = &(*tmp)->next;
1137         }
1138
1139         RTE_LOG(DEBUG, PMD, "  RX TUNTAP device name %s, qid %d on fd %d\n",
1140                 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
1141
1142         return 0;
1143
1144 error:
1145         rte_pktmbuf_free(rxq->pool);
1146         rxq->pool = NULL;
1147         rte_free(rxq->iovecs);
1148         rxq->iovecs = NULL;
1149         return ret;
1150 }
1151
1152 static int
1153 tap_tx_queue_setup(struct rte_eth_dev *dev,
1154                    uint16_t tx_queue_id,
1155                    uint16_t nb_tx_desc __rte_unused,
1156                    unsigned int socket_id __rte_unused,
1157                    const struct rte_eth_txconf *tx_conf)
1158 {
1159         struct pmd_internals *internals = dev->data->dev_private;
1160         struct tx_queue *txq;
1161         int ret;
1162
1163         if (tx_queue_id >= dev->data->nb_tx_queues)
1164                 return -1;
1165         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1166         txq = dev->data->tx_queues[tx_queue_id];
1167         /*
1168          * Don't verify port offloads for application which
1169          * use the old API.
1170          */
1171         if (tx_conf != NULL &&
1172             !!(tx_conf->txq_flags & ETH_TXQ_FLAGS_IGNORE)) {
1173                 if (tap_txq_are_offloads_valid(dev, tx_conf->offloads)) {
1174                         txq->csum = !!(tx_conf->offloads &
1175                                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
1176                                          DEV_TX_OFFLOAD_UDP_CKSUM |
1177                                          DEV_TX_OFFLOAD_TCP_CKSUM));
1178                 } else {
1179                         rte_errno = ENOTSUP;
1180                         RTE_LOG(ERR, PMD,
1181                                 "%p: Tx queue offloads 0x%" PRIx64
1182                                 " don't match port offloads 0x%" PRIx64
1183                                 " or supported offloads 0x%" PRIx64,
1184                                 (void *)dev, tx_conf->offloads,
1185                                 dev->data->dev_conf.txmode.offloads,
1186                                 tap_tx_offload_get_port_capa());
1187                         return -rte_errno;
1188                 }
1189         }
1190         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1191         if (ret == -1)
1192                 return -1;
1193         RTE_LOG(DEBUG, PMD,
1194                 "  TX TUNTAP device name %s, qid %d on fd %d csum %s\n",
1195                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd,
1196                 txq->csum ? "on" : "off");
1197
1198         return 0;
1199 }
1200
1201 static int
1202 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1203 {
1204         struct pmd_internals *pmd = dev->data->dev_private;
1205         struct ifreq ifr = { .ifr_mtu = mtu };
1206         int err = 0;
1207
1208         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1209         if (!err)
1210                 dev->data->mtu = mtu;
1211
1212         return err;
1213 }
1214
1215 static int
1216 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1217                      struct ether_addr *mc_addr_set __rte_unused,
1218                      uint32_t nb_mc_addr __rte_unused)
1219 {
1220         /*
1221          * Nothing to do actually: the tap has no filtering whatsoever, every
1222          * packet is received.
1223          */
1224         return 0;
1225 }
1226
1227 static int
1228 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1229 {
1230         struct rte_eth_dev *dev = arg;
1231         struct pmd_internals *pmd = dev->data->dev_private;
1232         struct ifinfomsg *info = NLMSG_DATA(nh);
1233
1234         if (nh->nlmsg_type != RTM_NEWLINK ||
1235             (info->ifi_index != pmd->if_index &&
1236              info->ifi_index != pmd->remote_if_index))
1237                 return 0;
1238         return tap_link_update(dev, 0);
1239 }
1240
1241 static void
1242 tap_dev_intr_handler(void *cb_arg)
1243 {
1244         struct rte_eth_dev *dev = cb_arg;
1245         struct pmd_internals *pmd = dev->data->dev_private;
1246
1247         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1248 }
1249
1250 static int
1251 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1252 {
1253         struct pmd_internals *pmd = dev->data->dev_private;
1254
1255         /* In any case, disable interrupt if the conf is no longer there. */
1256         if (!dev->data->dev_conf.intr_conf.lsc) {
1257                 if (pmd->intr_handle.fd != -1) {
1258                         tap_nl_final(pmd->intr_handle.fd);
1259                         rte_intr_callback_unregister(&pmd->intr_handle,
1260                                 tap_dev_intr_handler, dev);
1261                 }
1262                 return 0;
1263         }
1264         if (set) {
1265                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1266                 if (unlikely(pmd->intr_handle.fd == -1))
1267                         return -EBADF;
1268                 return rte_intr_callback_register(
1269                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1270         }
1271         tap_nl_final(pmd->intr_handle.fd);
1272         return rte_intr_callback_unregister(&pmd->intr_handle,
1273                                             tap_dev_intr_handler, dev);
1274 }
1275
1276 static int
1277 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1278 {
1279         int err;
1280
1281         err = tap_lsc_intr_handle_set(dev, set);
1282         if (err)
1283                 return err;
1284         err = tap_rx_intr_vec_set(dev, set);
1285         if (err && set)
1286                 tap_lsc_intr_handle_set(dev, 0);
1287         return err;
1288 }
1289
1290 static const uint32_t*
1291 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1292 {
1293         static const uint32_t ptypes[] = {
1294                 RTE_PTYPE_INNER_L2_ETHER,
1295                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1296                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1297                 RTE_PTYPE_INNER_L3_IPV4,
1298                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1299                 RTE_PTYPE_INNER_L3_IPV6,
1300                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1301                 RTE_PTYPE_INNER_L4_FRAG,
1302                 RTE_PTYPE_INNER_L4_UDP,
1303                 RTE_PTYPE_INNER_L4_TCP,
1304                 RTE_PTYPE_INNER_L4_SCTP,
1305                 RTE_PTYPE_L2_ETHER,
1306                 RTE_PTYPE_L2_ETHER_VLAN,
1307                 RTE_PTYPE_L2_ETHER_QINQ,
1308                 RTE_PTYPE_L3_IPV4,
1309                 RTE_PTYPE_L3_IPV4_EXT,
1310                 RTE_PTYPE_L3_IPV6_EXT,
1311                 RTE_PTYPE_L3_IPV6,
1312                 RTE_PTYPE_L4_FRAG,
1313                 RTE_PTYPE_L4_UDP,
1314                 RTE_PTYPE_L4_TCP,
1315                 RTE_PTYPE_L4_SCTP,
1316         };
1317
1318         return ptypes;
1319 }
1320
1321 static int
1322 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1323                   struct rte_eth_fc_conf *fc_conf)
1324 {
1325         fc_conf->mode = RTE_FC_NONE;
1326         return 0;
1327 }
1328
1329 static int
1330 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1331                   struct rte_eth_fc_conf *fc_conf)
1332 {
1333         if (fc_conf->mode != RTE_FC_NONE)
1334                 return -ENOTSUP;
1335         return 0;
1336 }
1337
1338 static const struct eth_dev_ops ops = {
1339         .dev_start              = tap_dev_start,
1340         .dev_stop               = tap_dev_stop,
1341         .dev_close              = tap_dev_close,
1342         .dev_configure          = tap_dev_configure,
1343         .dev_infos_get          = tap_dev_info,
1344         .rx_queue_setup         = tap_rx_queue_setup,
1345         .tx_queue_setup         = tap_tx_queue_setup,
1346         .rx_queue_release       = tap_rx_queue_release,
1347         .tx_queue_release       = tap_tx_queue_release,
1348         .flow_ctrl_get          = tap_flow_ctrl_get,
1349         .flow_ctrl_set          = tap_flow_ctrl_set,
1350         .link_update            = tap_link_update,
1351         .dev_set_link_up        = tap_link_set_up,
1352         .dev_set_link_down      = tap_link_set_down,
1353         .promiscuous_enable     = tap_promisc_enable,
1354         .promiscuous_disable    = tap_promisc_disable,
1355         .allmulticast_enable    = tap_allmulti_enable,
1356         .allmulticast_disable   = tap_allmulti_disable,
1357         .mac_addr_set           = tap_mac_set,
1358         .mtu_set                = tap_mtu_set,
1359         .set_mc_addr_list       = tap_set_mc_addr_list,
1360         .stats_get              = tap_stats_get,
1361         .stats_reset            = tap_stats_reset,
1362         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1363         .filter_ctrl            = tap_dev_filter_ctrl,
1364 };
1365
1366 static int
1367 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1368                    char *remote_iface, struct ether_addr *mac_addr)
1369 {
1370         int numa_node = rte_socket_id();
1371         struct rte_eth_dev *dev;
1372         struct pmd_internals *pmd;
1373         struct rte_eth_dev_data *data;
1374         struct ifreq ifr;
1375         int i;
1376
1377         RTE_LOG(DEBUG, PMD, "%s device on numa %u\n",
1378                         tuntap_name, rte_socket_id());
1379
1380         data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
1381         if (!data) {
1382                 RTE_LOG(ERR, PMD, "%s Failed to allocate data\n", tuntap_name);
1383                 goto error_exit_nodev;
1384         }
1385
1386         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1387         if (!dev) {
1388                 RTE_LOG(ERR, PMD, "%s Unable to allocate device struct\n",
1389                                 tuntap_name);
1390                 goto error_exit_nodev;
1391         }
1392
1393         pmd = dev->data->dev_private;
1394         pmd->dev = dev;
1395         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1396
1397         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1398         if (pmd->ioctl_sock == -1) {
1399                 RTE_LOG(ERR, PMD,
1400                         "%s Unable to get a socket for management: %s\n",
1401                         tuntap_name, strerror(errno));
1402                 goto error_exit;
1403         }
1404
1405         /* Setup some default values */
1406         rte_memcpy(data, dev->data, sizeof(*data));
1407         data->dev_private = pmd;
1408         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1409         data->numa_node = numa_node;
1410
1411         data->dev_link = pmd_link;
1412         data->mac_addrs = &pmd->eth_addr;
1413         /* Set the number of RX and TX queues */
1414         data->nb_rx_queues = 0;
1415         data->nb_tx_queues = 0;
1416
1417         dev->data = data;
1418         dev->dev_ops = &ops;
1419         dev->rx_pkt_burst = pmd_rx_burst;
1420         dev->tx_pkt_burst = pmd_tx_burst;
1421
1422         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1423         pmd->intr_handle.fd = -1;
1424         dev->intr_handle = &pmd->intr_handle;
1425
1426         /* Presetup the fds to -1 as being not valid */
1427         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1428                 pmd->rxq[i].fd = -1;
1429                 pmd->txq[i].fd = -1;
1430         }
1431
1432         if (tap_type) {
1433                 if (is_zero_ether_addr(mac_addr))
1434                         eth_random_addr((uint8_t *)&pmd->eth_addr);
1435                 else
1436                         rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1437         }
1438
1439         /* Immediately create the netdevice (this will create the 1st queue). */
1440         /* rx queue */
1441         if (tap_setup_queue(dev, pmd, 0, 1) == -1)
1442                 goto error_exit;
1443         /* tx queue */
1444         if (tap_setup_queue(dev, pmd, 0, 0) == -1)
1445                 goto error_exit;
1446
1447         ifr.ifr_mtu = dev->data->mtu;
1448         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1449                 goto error_exit;
1450
1451         if (tap_type) {
1452                 memset(&ifr, 0, sizeof(struct ifreq));
1453                 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1454                 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1455                                 ETHER_ADDR_LEN);
1456                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1457                         goto error_exit;
1458         }
1459
1460         /*
1461          * Set up everything related to rte_flow:
1462          * - netlink socket
1463          * - tap / remote if_index
1464          * - mandatory QDISCs
1465          * - rte_flow actual/implicit lists
1466          * - implicit rules
1467          */
1468         pmd->nlsk_fd = tap_nl_init(0);
1469         if (pmd->nlsk_fd == -1) {
1470                 RTE_LOG(WARNING, PMD, "%s: failed to create netlink socket.\n",
1471                         pmd->name);
1472                 goto disable_rte_flow;
1473         }
1474         pmd->if_index = if_nametoindex(pmd->name);
1475         if (!pmd->if_index) {
1476                 RTE_LOG(ERR, PMD, "%s: failed to get if_index.\n", pmd->name);
1477                 goto disable_rte_flow;
1478         }
1479         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1480                 RTE_LOG(ERR, PMD, "%s: failed to create multiq qdisc.\n",
1481                         pmd->name);
1482                 goto disable_rte_flow;
1483         }
1484         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1485                 RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1486                         pmd->name);
1487                 goto disable_rte_flow;
1488         }
1489         LIST_INIT(&pmd->flows);
1490
1491         if (strlen(remote_iface)) {
1492                 pmd->remote_if_index = if_nametoindex(remote_iface);
1493                 if (!pmd->remote_if_index) {
1494                         RTE_LOG(ERR, PMD, "%s: failed to get %s if_index.\n",
1495                                 pmd->name, remote_iface);
1496                         goto error_remote;
1497                 }
1498                 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1499                          "%s", remote_iface);
1500
1501                 /* Save state of remote device */
1502                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1503
1504                 /* Replicate remote MAC address */
1505                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1506                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1507                                 pmd->name, pmd->remote_iface);
1508                         goto error_remote;
1509                 }
1510                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1511                            ETHER_ADDR_LEN);
1512                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1513                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1514                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1515                                 pmd->name, remote_iface);
1516                         goto error_remote;
1517                 }
1518
1519                 /*
1520                  * Flush usually returns negative value because it tries to
1521                  * delete every QDISC (and on a running device, one QDISC at
1522                  * least is needed). Ignore negative return value.
1523                  */
1524                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1525                 if (qdisc_create_ingress(pmd->nlsk_fd,
1526                                          pmd->remote_if_index) < 0) {
1527                         RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1528                                 pmd->remote_iface);
1529                         goto error_remote;
1530                 }
1531                 LIST_INIT(&pmd->implicit_flows);
1532                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1533                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1534                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1535                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1536                         RTE_LOG(ERR, PMD,
1537                                 "%s: failed to create implicit rules.\n",
1538                                 pmd->name);
1539                         goto error_remote;
1540                 }
1541         }
1542
1543         return 0;
1544
1545 disable_rte_flow:
1546         RTE_LOG(ERR, PMD, " Disabling rte flow support: %s(%d)\n",
1547                 strerror(errno), errno);
1548         if (strlen(remote_iface)) {
1549                 RTE_LOG(ERR, PMD, "Remote feature requires flow support.\n");
1550                 goto error_exit;
1551         }
1552         return 0;
1553
1554 error_remote:
1555         RTE_LOG(ERR, PMD, " Can't set up remote feature: %s(%d)\n",
1556                 strerror(errno), errno);
1557         tap_flow_implicit_flush(pmd, NULL);
1558
1559 error_exit:
1560         if (pmd->ioctl_sock > 0)
1561                 close(pmd->ioctl_sock);
1562         rte_eth_dev_release_port(dev);
1563
1564 error_exit_nodev:
1565         RTE_LOG(ERR, PMD, "%s Unable to initialize %s\n",
1566                 tuntap_name, rte_vdev_device_name(vdev));
1567
1568         rte_free(data);
1569         return -EINVAL;
1570 }
1571
1572 static int
1573 set_interface_name(const char *key __rte_unused,
1574                    const char *value,
1575                    void *extra_args)
1576 {
1577         char *name = (char *)extra_args;
1578
1579         if (value)
1580                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN - 1);
1581         else
1582                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1583                          DEFAULT_TAP_NAME, (tap_unit - 1));
1584
1585         return 0;
1586 }
1587
1588 static int
1589 set_remote_iface(const char *key __rte_unused,
1590                  const char *value,
1591                  void *extra_args)
1592 {
1593         char *name = (char *)extra_args;
1594
1595         if (value)
1596                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
1597
1598         return 0;
1599 }
1600
1601 static int parse_user_mac(struct ether_addr *user_mac,
1602                 const char *value)
1603 {
1604         unsigned int index = 0;
1605         char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
1606
1607         if (user_mac == NULL || value == NULL)
1608                 return 0;
1609
1610         strlcpy(mac_temp, value, sizeof(mac_temp));
1611         mac_byte = strtok(mac_temp, ":");
1612
1613         while ((mac_byte != NULL) &&
1614                         (strlen(mac_byte) <= 2) &&
1615                         (strlen(mac_byte) == strspn(mac_byte,
1616                                         ETH_TAP_CMP_MAC_FMT))) {
1617                 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
1618                 mac_byte = strtok(NULL, ":");
1619         }
1620
1621         return index;
1622 }
1623
1624 static int
1625 set_mac_type(const char *key __rte_unused,
1626              const char *value,
1627              void *extra_args)
1628 {
1629         struct ether_addr *user_mac = extra_args;
1630
1631         if (!value)
1632                 return 0;
1633
1634         if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
1635                 static int iface_idx;
1636
1637                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1638                 memcpy((char *)user_mac->addr_bytes, "\0dtap", ETHER_ADDR_LEN);
1639                 user_mac->addr_bytes[ETHER_ADDR_LEN - 1] = iface_idx++ + '0';
1640                 goto success;
1641         }
1642
1643         if (parse_user_mac(user_mac, value) != 6)
1644                 goto error;
1645 success:
1646         RTE_LOG(DEBUG, PMD, "TAP user MAC param (%s)\n", value);
1647         return 0;
1648
1649 error:
1650         RTE_LOG(ERR, PMD, "TAP user MAC (%s) is not in format (%s|%s)\n",
1651                 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
1652         return -1;
1653 }
1654
1655 /*
1656  * Open a TUN interface device. TUN PMD
1657  * 1) sets tap_type as false
1658  * 2) intakes iface as argument.
1659  * 3) as interface is virtual set speed to 10G
1660  */
1661 static int
1662 rte_pmd_tun_probe(struct rte_vdev_device *dev)
1663 {
1664         const char *name, *params;
1665         int ret;
1666         struct rte_kvargs *kvlist = NULL;
1667         char tun_name[RTE_ETH_NAME_MAX_LEN];
1668         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1669
1670         tap_type = 0;
1671         strcpy(tuntap_name, "TUN");
1672
1673         name = rte_vdev_device_name(dev);
1674         params = rte_vdev_device_args(dev);
1675         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1676
1677         if (params && (params[0] != '\0')) {
1678                 RTE_LOG(DEBUG, PMD, "parameters (%s)\n", params);
1679
1680                 kvlist = rte_kvargs_parse(params, valid_arguments);
1681                 if (kvlist) {
1682                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1683                                 ret = rte_kvargs_process(kvlist,
1684                                         ETH_TAP_IFACE_ARG,
1685                                         &set_interface_name,
1686                                         tun_name);
1687
1688                                 if (ret == -1)
1689                                         goto leave;
1690                         }
1691                 }
1692         }
1693         pmd_link.link_speed = ETH_SPEED_NUM_10G;
1694
1695         RTE_LOG(NOTICE, PMD, "Initializing pmd_tun for %s as %s\n",
1696                 name, tun_name);
1697
1698         ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0);
1699
1700 leave:
1701         if (ret == -1) {
1702                 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1703                         name, tun_name);
1704                 tun_unit--; /* Restore the unit number */
1705         }
1706         rte_kvargs_free(kvlist);
1707
1708         return ret;
1709 }
1710
1711 /* Open a TAP interface device.
1712  */
1713 static int
1714 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1715 {
1716         const char *name, *params;
1717         int ret;
1718         struct rte_kvargs *kvlist = NULL;
1719         int speed;
1720         char tap_name[RTE_ETH_NAME_MAX_LEN];
1721         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1722         struct ether_addr user_mac = { .addr_bytes = {0} };
1723
1724         tap_type = 1;
1725         strcpy(tuntap_name, "TAP");
1726
1727         name = rte_vdev_device_name(dev);
1728         params = rte_vdev_device_args(dev);
1729
1730         speed = ETH_SPEED_NUM_10G;
1731         snprintf(tap_name, sizeof(tap_name), "%s%d",
1732                  DEFAULT_TAP_NAME, tap_unit++);
1733         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1734
1735         if (params && (params[0] != '\0')) {
1736                 RTE_LOG(DEBUG, PMD, "parameters (%s)\n", params);
1737
1738                 kvlist = rte_kvargs_parse(params, valid_arguments);
1739                 if (kvlist) {
1740                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1741                                 ret = rte_kvargs_process(kvlist,
1742                                                          ETH_TAP_IFACE_ARG,
1743                                                          &set_interface_name,
1744                                                          tap_name);
1745                                 if (ret == -1)
1746                                         goto leave;
1747                         }
1748
1749                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1750                                 ret = rte_kvargs_process(kvlist,
1751                                                          ETH_TAP_REMOTE_ARG,
1752                                                          &set_remote_iface,
1753                                                          remote_iface);
1754                                 if (ret == -1)
1755                                         goto leave;
1756                         }
1757
1758                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1759                                 ret = rte_kvargs_process(kvlist,
1760                                                          ETH_TAP_MAC_ARG,
1761                                                          &set_mac_type,
1762                                                          &user_mac);
1763                                 if (ret == -1)
1764                                         goto leave;
1765                         }
1766                 }
1767         }
1768         pmd_link.link_speed = speed;
1769
1770         RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1771                 name, tap_name);
1772
1773         ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac);
1774
1775 leave:
1776         if (ret == -1) {
1777                 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1778                         name, tap_name);
1779                 tap_unit--;             /* Restore the unit number */
1780         }
1781         rte_kvargs_free(kvlist);
1782
1783         return ret;
1784 }
1785
1786 /* detach a TUNTAP device.
1787  */
1788 static int
1789 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1790 {
1791         struct rte_eth_dev *eth_dev = NULL;
1792         struct pmd_internals *internals;
1793         int i;
1794
1795         RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1796                 rte_socket_id());
1797
1798         /* find the ethdev entry */
1799         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1800         if (!eth_dev)
1801                 return 0;
1802
1803         internals = eth_dev->data->dev_private;
1804         if (internals->nlsk_fd) {
1805                 tap_flow_flush(eth_dev, NULL);
1806                 tap_flow_implicit_flush(internals, NULL);
1807                 tap_nl_final(internals->nlsk_fd);
1808         }
1809         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1810                 if (internals->rxq[i].fd != -1) {
1811                         close(internals->rxq[i].fd);
1812                         internals->rxq[i].fd = -1;
1813                 }
1814                 if (internals->txq[i].fd != -1) {
1815                         close(internals->txq[i].fd);
1816                         internals->txq[i].fd = -1;
1817                 }
1818         }
1819
1820         close(internals->ioctl_sock);
1821         rte_free(eth_dev->data->dev_private);
1822         rte_free(eth_dev->data);
1823
1824         rte_eth_dev_release_port(eth_dev);
1825
1826         return 0;
1827 }
1828
1829 static struct rte_vdev_driver pmd_tun_drv = {
1830         .probe = rte_pmd_tun_probe,
1831         .remove = rte_pmd_tap_remove,
1832 };
1833
1834 static struct rte_vdev_driver pmd_tap_drv = {
1835         .probe = rte_pmd_tap_probe,
1836         .remove = rte_pmd_tap_remove,
1837 };
1838 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1839 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
1840 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1841 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
1842                               ETH_TAP_IFACE_ARG "=<string> ");
1843 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1844                               ETH_TAP_IFACE_ARG "=<string> "
1845                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
1846                               ETH_TAP_REMOTE_ARG "=<string>");