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 #include <rte_ethdev.h>
  20 #include <rte_errno.h>
  21
  22 #include <assert.h>
  23 #include <sys/types.h>
  24 #include <sys/stat.h>
  25 #include <sys/socket.h>
  26 #include <sys/ioctl.h>
  27 #include <sys/utsname.h>
  28 #include <sys/mman.h>
  29 #include <errno.h>
  30 #include <signal.h>
  31 #include <stdbool.h>
  32 #include <stdint.h>
  33 #include <sys/uio.h>
  34 #include <unistd.h>
  35 #include <arpa/inet.h>
  36 #include <net/if.h>
  37 #include <linux/if_tun.h>
  38 #include <linux/if_ether.h>
  39 #include <fcntl.h>
  40
  41 #include <tap_rss.h>
  42 #include <rte_eth_tap.h>
  43 #include <tap_flow.h>
  44 #include <tap_netlink.h>
  45 #include <tap_tcmsgs.h>
  46
  47 /* Linux based path to the TUN device */
  48 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  49 #define DEFAULT_TAP_NAME        "dtap"
  50 #define DEFAULT_TUN_NAME        "dtun"
  51
  52 #define ETH_TAP_IFACE_ARG       "iface"
  53 #define ETH_TAP_REMOTE_ARG      "remote"
  54 #define ETH_TAP_MAC_ARG         "mac"
  55 #define ETH_TAP_MAC_FIXED       "fixed"
  56
  57 #define ETH_TAP_USR_MAC_FMT     "xx:xx:xx:xx:xx:xx"
  58 #define ETH_TAP_CMP_MAC_FMT     "0123456789ABCDEFabcdef"
  59 #define ETH_TAP_MAC_ARG_FMT     ETH_TAP_MAC_FIXED "|" ETH_TAP_USR_MAC_FMT
  60
  61 #define TAP_GSO_MBUFS_PER_CORE  128
  62 #define TAP_GSO_MBUF_SEG_SIZE   128
  63 #define TAP_GSO_MBUF_CACHE_SIZE 4
  64 #define TAP_GSO_MBUFS_NUM \
  65         (TAP_GSO_MBUFS_PER_CORE * TAP_GSO_MBUF_CACHE_SIZE)
  66
  67 /* IPC key for queue fds sync */
  68 #define TAP_MP_KEY "tap_mp_sync_queues"
  69
  70 static int tap_devices_count;
  71 static struct rte_vdev_driver pmd_tap_drv;
  72 static struct rte_vdev_driver pmd_tun_drv;
  73
  74 static const char *valid_arguments[] = {
  75         ETH_TAP_IFACE_ARG,
  76         ETH_TAP_REMOTE_ARG,
  77         ETH_TAP_MAC_ARG,
  78         NULL
  79 };
  80
  81 static unsigned int tap_unit;
  82 static unsigned int tun_unit;
  83
  84 static char tuntap_name[8];
  85
  86 static volatile uint32_t tap_trigger;   /* Rx trigger */
  87
  88 static struct rte_eth_link pmd_link = {
  89         .link_speed = ETH_SPEED_NUM_10G,
  90         .link_duplex = ETH_LINK_FULL_DUPLEX,
  91         .link_status = ETH_LINK_DOWN,
  92         .link_autoneg = ETH_LINK_FIXED,
  93 };
  94
  95 static void
  96 tap_trigger_cb(int sig __rte_unused)
  97 {
  98         /* Valid trigger values are nonzero */
  99         tap_trigger = (tap_trigger + 1) | 0x80000000;
 100 }
 101
 102 /* Specifies on what netdevices the ioctl should be applied */
 103 enum ioctl_mode {
 104         LOCAL_AND_REMOTE,
 105         LOCAL_ONLY,
 106         REMOTE_ONLY,
 107 };
 108
 109 /* Message header to synchronize queues via IPC */
 110 struct ipc_queues {
 111         char port_name[RTE_DEV_NAME_MAX_LEN];
 112         int rxq_count;
 113         int txq_count;
 114         /*
 115          * The file descriptors are in the dedicated part
 116          * of the Unix message to be translated by the kernel.
 117          */
 118 };
 119
 120 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
 121
 122 /**
 123  * Tun/Tap allocation routine
 124  *
 125  * @param[in] pmd
 126  *   Pointer to private structure.
 127  *
 128  * @param[in] is_keepalive
 129  *   Keepalive flag
 130  *
 131  * @return
 132  *   -1 on failure, fd on success
 133  */
 134 static int
 135 tun_alloc(struct pmd_internals *pmd, int is_keepalive)
 136 {
 137         struct ifreq ifr;
 138 #ifdef IFF_MULTI_QUEUE
 139         unsigned int features;
 140 #endif
 141         int fd;
 142
 143         memset(&ifr, 0, sizeof(struct ifreq));
 144
 145         /*
 146          * Do not set IFF_NO_PI as packet information header will be needed
 147          * to check if a received packet has been truncated.
 148          */
 149         ifr.ifr_flags = (pmd->type == ETH_TUNTAP_TYPE_TAP) ?
 150                 IFF_TAP : IFF_TUN | IFF_POINTOPOINT;
 151         strlcpy(ifr.ifr_name, pmd->name, IFNAMSIZ);
 152
 153         TAP_LOG(DEBUG, "ifr_name '%s'", ifr.ifr_name);
 154
 155         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 156         if (fd < 0) {
 157                 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
 158                 goto error;
 159         }
 160
 161 #ifdef IFF_MULTI_QUEUE
 162         /* Grab the TUN features to verify we can work multi-queue */
 163         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 164                 TAP_LOG(ERR, "%s unable to get TUN/TAP features",
 165                         tuntap_name);
 166                 goto error;
 167         }
 168         TAP_LOG(DEBUG, "%s Features %08x", tuntap_name, features);
 169
 170         if (features & IFF_MULTI_QUEUE) {
 171                 TAP_LOG(DEBUG, "  Multi-queue support for %d queues",
 172                         RTE_PMD_TAP_MAX_QUEUES);
 173                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 174         } else
 175 #endif
 176         {
 177                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 178                 TAP_LOG(DEBUG, "  Single queue only support");
 179         }
 180
 181         /* Set the TUN/TAP configuration and set the name if needed */
 182         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 183                 TAP_LOG(WARNING, "Unable to set TUNSETIFF for %s: %s",
 184                         ifr.ifr_name, strerror(errno));
 185                 goto error;
 186         }
 187
 188         if (is_keepalive) {
 189                 /*
 190                  * Detach the TUN/TAP keep-alive queue
 191                  * to avoid traffic through it
 192                  */
 193                 ifr.ifr_flags = IFF_DETACH_QUEUE;
 194                 if (ioctl(fd, TUNSETQUEUE, (void *)&ifr) < 0) {
 195                         TAP_LOG(WARNING,
 196                                 "Unable to detach keep-alive queue for %s: %s",
 197                                 ifr.ifr_name, strerror(errno));
 198                         goto error;
 199                 }
 200         }
 201
 202         /* Always set the file descriptor to non-blocking */
 203         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 204                 TAP_LOG(WARNING,
 205                         "Unable to set %s to nonblocking: %s",
 206                         ifr.ifr_name, strerror(errno));
 207                 goto error;
 208         }
 209
 210         /* Set up trigger to optimize empty Rx bursts */
 211         errno = 0;
 212         do {
 213                 struct sigaction sa;
 214                 int flags = fcntl(fd, F_GETFL);
 215
 216                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 217                         break;
 218                 if (sa.sa_handler != tap_trigger_cb) {
 219                         /*
 220                          * Make sure SIGIO is not already taken. This is done
 221                          * as late as possible to leave the application a
 222                          * chance to set up its own signal handler first.
 223                          */
 224                         if (sa.sa_handler != SIG_IGN &&
 225                             sa.sa_handler != SIG_DFL) {
 226                                 errno = EBUSY;
 227                                 break;
 228                         }
 229                         sa = (struct sigaction){
 230                                 .sa_flags = SA_RESTART,
 231                                 .sa_handler = tap_trigger_cb,
 232                         };
 233                         if (sigaction(SIGIO, &sa, NULL) == -1)
 234                                 break;
 235                 }
 236                 /* Enable SIGIO on file descriptor */
 237                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 238                 fcntl(fd, F_SETOWN, getpid());
 239         } while (0);
 240
 241         if (errno) {
 242                 /* Disable trigger globally in case of error */
 243                 tap_trigger = 0;
 244                 TAP_LOG(WARNING, "Rx trigger disabled: %s",
 245                         strerror(errno));
 246         }
 247
 248         return fd;
 249
 250 error:
 251         if (fd >= 0)
 252                 close(fd);
 253         return -1;
 254 }
 255
 256 static void
 257 tap_verify_csum(struct rte_mbuf *mbuf)
 258 {
 259         uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
 260         uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
 261         uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
 262         unsigned int l2_len = sizeof(struct ether_hdr);
 263         unsigned int l3_len;
 264         uint16_t cksum = 0;
 265         void *l3_hdr;
 266         void *l4_hdr;
 267
 268         if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
 269                 l2_len += 4;
 270         else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
 271                 l2_len += 8;
 272         /* Don't verify checksum for packets with discontinuous L2 header */
 273         if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
 274                      rte_pktmbuf_data_len(mbuf)))
 275                 return;
 276         l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
 277         if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 278                 struct ipv4_hdr *iph = l3_hdr;
 279
 280                 /* ihl contains the number of 4-byte words in the header */
 281                 l3_len = 4 * (iph->version_ihl & 0xf);
 282                 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
 283                         return;
 284                 /* check that the total length reported by header is not
 285                  * greater than the total received size
 286                  */
 287                 if (l2_len + rte_be_to_cpu_16(iph->total_length) >
 288                                 rte_pktmbuf_data_len(mbuf))
 289                         return;
 290
 291                 cksum = ~rte_raw_cksum(iph, l3_len);
 292                 mbuf->ol_flags |= cksum ?
 293                         PKT_RX_IP_CKSUM_BAD :
 294                         PKT_RX_IP_CKSUM_GOOD;
 295         } else if (l3 == RTE_PTYPE_L3_IPV6) {
 296                 struct ipv6_hdr *iph = l3_hdr;
 297
 298                 l3_len = sizeof(struct ipv6_hdr);
 299                 /* check that the total length reported by header is not
 300                  * greater than the total received size
 301                  */
 302                 if (l2_len + l3_len + rte_be_to_cpu_16(iph->payload_len) >
 303                                 rte_pktmbuf_data_len(mbuf))
 304                         return;
 305         } else {
 306                 /* IPv6 extensions are not supported */
 307                 return;
 308         }
 309         if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
 310                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 311                 /* Don't verify checksum for multi-segment packets. */
 312                 if (mbuf->nb_segs > 1)
 313                         return;
 314                 if (l3 == RTE_PTYPE_L3_IPV4)
 315                         cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
 316                 else if (l3 == RTE_PTYPE_L3_IPV6)
 317                         cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
 318                 mbuf->ol_flags |= cksum ?
 319                         PKT_RX_L4_CKSUM_BAD :
 320                         PKT_RX_L4_CKSUM_GOOD;
 321         }
 322 }
 323
 324 static uint64_t
 325 tap_rx_offload_get_port_capa(void)
 326 {
 327         /*
 328          * No specific port Rx offload capabilities.
 329          */
 330         return 0;
 331 }
 332
 333 static uint64_t
 334 tap_rx_offload_get_queue_capa(void)
 335 {
 336         return DEV_RX_OFFLOAD_SCATTER |
 337                DEV_RX_OFFLOAD_IPV4_CKSUM |
 338                DEV_RX_OFFLOAD_UDP_CKSUM |
 339                DEV_RX_OFFLOAD_TCP_CKSUM;
 340 }
 341
 342 /* Callback to handle the rx burst of packets to the correct interface and
 343  * file descriptor(s) in a multi-queue setup.
 344  */
 345 static uint16_t
 346 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 347 {
 348         struct rx_queue *rxq = queue;
 349         struct pmd_process_private *process_private;
 350         uint16_t num_rx;
 351         unsigned long num_rx_bytes = 0;
 352         uint32_t trigger = tap_trigger;
 353
 354         if (trigger == rxq->trigger_seen)
 355                 return 0;
 356         if (trigger)
 357                 rxq->trigger_seen = trigger;
 358         process_private = rte_eth_devices[rxq->in_port].process_private;
 359         rte_compiler_barrier();
 360         for (num_rx = 0; num_rx < nb_pkts; ) {
 361                 struct rte_mbuf *mbuf = rxq->pool;
 362                 struct rte_mbuf *seg = NULL;
 363                 struct rte_mbuf *new_tail = NULL;
 364                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 365                 int len;
 366
 367                 len = readv(process_private->rxq_fds[rxq->queue_id],
 368                         *rxq->iovecs,
 369                         1 + (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
 370                              rxq->nb_rx_desc : 1));
 371                 if (len < (int)sizeof(struct tun_pi))
 372                         break;
 373
 374                 /* Packet couldn't fit in the provided mbuf */
 375                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 376                         rxq->stats.ierrors++;
 377                         continue;
 378                 }
 379
 380                 len -= sizeof(struct tun_pi);
 381
 382                 mbuf->pkt_len = len;
 383                 mbuf->port = rxq->in_port;
 384                 while (1) {
 385                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 386
 387                         if (unlikely(!buf)) {
 388                                 rxq->stats.rx_nombuf++;
 389                                 /* No new buf has been allocated: do nothing */
 390                                 if (!new_tail || !seg)
 391                                         goto end;
 392
 393                                 seg->next = NULL;
 394                                 rte_pktmbuf_free(mbuf);
 395
 396                                 goto end;
 397                         }
 398                         seg = seg ? seg->next : mbuf;
 399                         if (rxq->pool == mbuf)
 400                                 rxq->pool = buf;
 401                         if (new_tail)
 402                                 new_tail->next = buf;
 403                         new_tail = buf;
 404                         new_tail->next = seg->next;
 405
 406                         /* iovecs[0] is reserved for packet info (pi) */
 407                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 408                                 buf->buf_len - data_off;
 409                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 410                                 (char *)buf->buf_addr + data_off;
 411
 412                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 413                         seg->data_off = data_off;
 414
 415                         len -= seg->data_len;
 416                         if (len <= 0)
 417                                 break;
 418                         mbuf->nb_segs++;
 419                         /* First segment has headroom, not the others */
 420                         data_off = 0;
 421                 }
 422                 seg->next = NULL;
 423                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 424                                                       RTE_PTYPE_ALL_MASK);
 425                 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
 426                         tap_verify_csum(mbuf);
 427
 428                 /* account for the receive frame */
 429                 bufs[num_rx++] = mbuf;
 430                 num_rx_bytes += mbuf->pkt_len;
 431         }
 432 end:
 433         rxq->stats.ipackets += num_rx;
 434         rxq->stats.ibytes += num_rx_bytes;
 435
 436         return num_rx;
 437 }
 438
 439 static uint64_t
 440 tap_tx_offload_get_port_capa(void)
 441 {
 442         /*
 443          * No specific port Tx offload capabilities.
 444          */
 445         return 0;
 446 }
 447
 448 static uint64_t
 449 tap_tx_offload_get_queue_capa(void)
 450 {
 451         return DEV_TX_OFFLOAD_MULTI_SEGS |
 452                DEV_TX_OFFLOAD_IPV4_CKSUM |
 453                DEV_TX_OFFLOAD_UDP_CKSUM |
 454                DEV_TX_OFFLOAD_TCP_CKSUM |
 455                DEV_TX_OFFLOAD_TCP_TSO;
 456 }
 457
 458 /* Finalize l4 checksum calculation */
 459 static void
 460 tap_tx_l4_cksum(uint16_t *l4_cksum, uint16_t l4_phdr_cksum,
 461                 uint32_t l4_raw_cksum)
 462 {
 463         if (l4_cksum) {
 464                 uint32_t cksum;
 465
 466                 cksum = __rte_raw_cksum_reduce(l4_raw_cksum);
 467                 cksum += l4_phdr_cksum;
 468
 469                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 470                 cksum = (~cksum) & 0xffff;
 471                 if (cksum == 0)
 472                         cksum = 0xffff;
 473                 *l4_cksum = cksum;
 474         }
 475 }
 476
 477 /* Accumaulate L4 raw checksums */
 478 static void
 479 tap_tx_l4_add_rcksum(char *l4_data, unsigned int l4_len, uint16_t *l4_cksum,
 480                         uint32_t *l4_raw_cksum)
 481 {
 482         if (l4_cksum == NULL)
 483                 return;
 484
 485         *l4_raw_cksum = __rte_raw_cksum(l4_data, l4_len, *l4_raw_cksum);
 486 }
 487
 488 /* L3 and L4 pseudo headers checksum offloads */
 489 static void
 490 tap_tx_l3_cksum(char *packet, uint64_t ol_flags, unsigned int l2_len,
 491                 unsigned int l3_len, unsigned int l4_len, uint16_t **l4_cksum,
 492                 uint16_t *l4_phdr_cksum, uint32_t *l4_raw_cksum)
 493 {
 494         void *l3_hdr = packet + l2_len;
 495
 496         if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
 497                 struct ipv4_hdr *iph = l3_hdr;
 498                 uint16_t cksum;
 499
 500                 iph->hdr_checksum = 0;
 501                 cksum = rte_raw_cksum(iph, l3_len);
 502                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 503         }
 504         if (ol_flags & PKT_TX_L4_MASK) {
 505                 void *l4_hdr;
 506
 507                 l4_hdr = packet + l2_len + l3_len;
 508                 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
 509                         *l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
 510                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 511                         *l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
 512                 else
 513                         return;
 514                 **l4_cksum = 0;
 515                 if (ol_flags & PKT_TX_IPV4)
 516                         *l4_phdr_cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 517                 else
 518                         *l4_phdr_cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 519                 *l4_raw_cksum = __rte_raw_cksum(l4_hdr, l4_len, 0);
 520         }
 521 }
 522
 523 static inline void
 524 tap_write_mbufs(struct tx_queue *txq, uint16_t num_mbufs,
 525                         struct rte_mbuf **pmbufs,
 526                         uint16_t *num_packets, unsigned long *num_tx_bytes)
 527 {
 528         int i;
 529         uint16_t l234_hlen;
 530         struct pmd_process_private *process_private;
 531
 532         process_private = rte_eth_devices[txq->out_port].process_private;
 533
 534         for (i = 0; i < num_mbufs; i++) {
 535                 struct rte_mbuf *mbuf = pmbufs[i];
 536                 struct iovec iovecs[mbuf->nb_segs + 2];
 537                 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
 538                 struct rte_mbuf *seg = mbuf;
 539                 char m_copy[mbuf->data_len];
 540                 int proto;
 541                 int n;
 542                 int j;
 543                 int k; /* current index in iovecs for copying segments */
 544                 uint16_t seg_len; /* length of first segment */
 545                 uint16_t nb_segs;
 546                 uint16_t *l4_cksum; /* l4 checksum (pseudo header + payload) */
 547                 uint32_t l4_raw_cksum = 0; /* TCP/UDP payload raw checksum */
 548                 uint16_t l4_phdr_cksum = 0; /* TCP/UDP pseudo header checksum */
 549                 uint16_t is_cksum = 0; /* in case cksum should be offloaded */
 550
 551                 l4_cksum = NULL;
 552                 if (txq->type == ETH_TUNTAP_TYPE_TUN) {
 553                         /*
 554                          * TUN and TAP are created with IFF_NO_PI disabled.
 555                          * For TUN PMD this mandatory as fields are used by
 556                          * Kernel tun.c to determine whether its IP or non IP
 557                          * packets.
 558                          *
 559                          * The logic fetches the first byte of data from mbuf
 560                          * then compares whether its v4 or v6. If first byte
 561                          * is 4 or 6, then protocol field is updated.
 562                          */
 563                         char *buff_data = rte_pktmbuf_mtod(seg, void *);
 564                         proto = (*buff_data & 0xf0);
 565                         pi.proto = (proto == 0x40) ?
 566                                 rte_cpu_to_be_16(ETHER_TYPE_IPv4) :
 567                                 ((proto == 0x60) ?
 568                                         rte_cpu_to_be_16(ETHER_TYPE_IPv6) :
 569                                         0x00);
 570                 }
 571
 572                 k = 0;
 573                 iovecs[k].iov_base = &pi;
 574                 iovecs[k].iov_len = sizeof(pi);
 575                 k++;
 576
 577                 nb_segs = mbuf->nb_segs;
 578                 if (txq->csum &&
 579                     ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 580                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 581                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
 582                         is_cksum = 1;
 583
 584                         /* Support only packets with at least layer 4
 585                          * header included in the first segment
 586                          */
 587                         seg_len = rte_pktmbuf_data_len(mbuf);
 588                         l234_hlen = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
 589                         if (seg_len < l234_hlen)
 590                                 break;
 591
 592                         /* To change checksums, work on a * copy of l2, l3
 593                          * headers + l4 pseudo header
 594                          */
 595                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 596                                         l234_hlen);
 597                         tap_tx_l3_cksum(m_copy, mbuf->ol_flags,
 598                                        mbuf->l2_len, mbuf->l3_len, mbuf->l4_len,
 599                                        &l4_cksum, &l4_phdr_cksum,
 600                                        &l4_raw_cksum);
 601                         iovecs[k].iov_base = m_copy;
 602                         iovecs[k].iov_len = l234_hlen;
 603                         k++;
 604
 605                         /* Update next iovecs[] beyond l2, l3, l4 headers */
 606                         if (seg_len > l234_hlen) {
 607                                 iovecs[k].iov_len = seg_len - l234_hlen;
 608                                 iovecs[k].iov_base =
 609                                         rte_pktmbuf_mtod(seg, char *) +
 610                                                 l234_hlen;
 611                                 tap_tx_l4_add_rcksum(iovecs[k].iov_base,
 612                                         iovecs[k].iov_len, l4_cksum,
 613                                         &l4_raw_cksum);
 614                                 k++;
 615                                 nb_segs++;
 616                         }
 617                         seg = seg->next;
 618                 }
 619
 620                 for (j = k; j <= nb_segs; j++) {
 621                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 622                         iovecs[j].iov_base = rte_pktmbuf_mtod(seg, void *);
 623                         if (is_cksum)
 624                                 tap_tx_l4_add_rcksum(iovecs[j].iov_base,
 625                                         iovecs[j].iov_len, l4_cksum,
 626                                         &l4_raw_cksum);
 627                         seg = seg->next;
 628                 }
 629
 630                 if (is_cksum)
 631                         tap_tx_l4_cksum(l4_cksum, l4_phdr_cksum, l4_raw_cksum);
 632
 633                 /* copy the tx frame data */
 634                 n = writev(process_private->txq_fds[txq->queue_id], iovecs, j);
 635                 if (n <= 0)
 636                         break;
 637                 (*num_packets)++;
 638                 (*num_tx_bytes) += rte_pktmbuf_pkt_len(mbuf);
 639         }
 640 }
 641
 642 /* Callback to handle sending packets from the tap interface
 643  */
 644 static uint16_t
 645 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 646 {
 647         struct tx_queue *txq = queue;
 648         uint16_t num_tx = 0;
 649         uint16_t num_packets = 0;
 650         unsigned long num_tx_bytes = 0;
 651         uint32_t max_size;
 652         int i;
 653
 654         if (unlikely(nb_pkts == 0))
 655                 return 0;
 656
 657         struct rte_mbuf *gso_mbufs[MAX_GSO_MBUFS];
 658         max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
 659         for (i = 0; i < nb_pkts; i++) {
 660                 struct rte_mbuf *mbuf_in = bufs[num_tx];
 661                 struct rte_mbuf **mbuf;
 662                 uint16_t num_mbufs = 0;
 663                 uint16_t tso_segsz = 0;
 664                 int ret;
 665                 uint16_t hdrs_len;
 666                 int j;
 667                 uint64_t tso;
 668
 669                 tso = mbuf_in->ol_flags & PKT_TX_TCP_SEG;
 670                 if (tso) {
 671                         struct rte_gso_ctx *gso_ctx = &txq->gso_ctx;
 672
 673                         assert(gso_ctx != NULL);
 674
 675                         /* TCP segmentation implies TCP checksum offload */
 676                         mbuf_in->ol_flags |= PKT_TX_TCP_CKSUM;
 677
 678                         /* gso size is calculated without ETHER_CRC_LEN */
 679                         hdrs_len = mbuf_in->l2_len + mbuf_in->l3_len +
 680                                         mbuf_in->l4_len;
 681                         tso_segsz = mbuf_in->tso_segsz + hdrs_len;
 682                         if (unlikely(tso_segsz == hdrs_len) ||
 683                                 tso_segsz > *txq->mtu) {
 684                                 txq->stats.errs++;
 685                                 break;
 686                         }
 687                         gso_ctx->gso_size = tso_segsz;
 688                         ret = rte_gso_segment(mbuf_in, /* packet to segment */
 689                                 gso_ctx, /* gso control block */
 690                                 (struct rte_mbuf **)&gso_mbufs, /* out mbufs */
 691                                 RTE_DIM(gso_mbufs)); /* max tso mbufs */
 692
 693                         /* ret contains the number of new created mbufs */
 694                         if (ret < 0)
 695                                 break;
 696
 697                         mbuf = gso_mbufs;
 698                         num_mbufs = ret;
 699                 } else {
 700                         /* stats.errs will be incremented */
 701                         if (rte_pktmbuf_pkt_len(mbuf_in) > max_size)
 702                                 break;
 703
 704                         /* ret 0 indicates no new mbufs were created */
 705                         ret = 0;
 706                         mbuf = &mbuf_in;
 707                         num_mbufs = 1;
 708                 }
 709
 710                 tap_write_mbufs(txq, num_mbufs, mbuf,
 711                                 &num_packets, &num_tx_bytes);
 712                 num_tx++;
 713                 /* free original mbuf */
 714                 rte_pktmbuf_free(mbuf_in);
 715                 /* free tso mbufs */
 716                 for (j = 0; j < ret; j++)
 717                         rte_pktmbuf_free(mbuf[j]);
 718         }
 719
 720         txq->stats.opackets += num_packets;
 721         txq->stats.errs += nb_pkts - num_tx;
 722         txq->stats.obytes += num_tx_bytes;
 723
 724         return num_packets;
 725 }
 726
 727 static const char *
 728 tap_ioctl_req2str(unsigned long request)
 729 {
 730         switch (request) {
 731         case SIOCSIFFLAGS:
 732                 return "SIOCSIFFLAGS";
 733         case SIOCGIFFLAGS:
 734                 return "SIOCGIFFLAGS";
 735         case SIOCGIFHWADDR:
 736                 return "SIOCGIFHWADDR";
 737         case SIOCSIFHWADDR:
 738                 return "SIOCSIFHWADDR";
 739         case SIOCSIFMTU:
 740                 return "SIOCSIFMTU";
 741         }
 742         return "UNKNOWN";
 743 }
 744
 745 static int
 746 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 747           struct ifreq *ifr, int set, enum ioctl_mode mode)
 748 {
 749         short req_flags = ifr->ifr_flags;
 750         int remote = pmd->remote_if_index &&
 751                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 752
 753         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 754                 return 0;
 755         /*
 756          * If there is a remote netdevice, apply ioctl on it, then apply it on
 757          * the tap netdevice.
 758          */
 759 apply:
 760         if (remote)
 761                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
 762         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 763                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 764         switch (request) {
 765         case SIOCSIFFLAGS:
 766                 /* fetch current flags to leave other flags untouched */
 767                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 768                         goto error;
 769                 if (set)
 770                         ifr->ifr_flags |= req_flags;
 771                 else
 772                         ifr->ifr_flags &= ~req_flags;
 773                 break;
 774         case SIOCGIFFLAGS:
 775         case SIOCGIFHWADDR:
 776         case SIOCSIFHWADDR:
 777         case SIOCSIFMTU:
 778                 break;
 779         default:
 780                 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
 781                         pmd->name);
 782                 return -EINVAL;
 783         }
 784         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 785                 goto error;
 786         if (remote-- && mode == LOCAL_AND_REMOTE)
 787                 goto apply;
 788         return 0;
 789
 790 error:
 791         TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
 792                 tap_ioctl_req2str(request), strerror(errno), errno);
 793         return -errno;
 794 }
 795
 796 static int
 797 tap_link_set_down(struct rte_eth_dev *dev)
 798 {
 799         struct pmd_internals *pmd = dev->data->dev_private;
 800         struct ifreq ifr = { .ifr_flags = IFF_UP };
 801
 802         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 803         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 804 }
 805
 806 static int
 807 tap_link_set_up(struct rte_eth_dev *dev)
 808 {
 809         struct pmd_internals *pmd = dev->data->dev_private;
 810         struct ifreq ifr = { .ifr_flags = IFF_UP };
 811
 812         dev->data->dev_link.link_status = ETH_LINK_UP;
 813         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 814 }
 815
 816 static int
 817 tap_dev_start(struct rte_eth_dev *dev)
 818 {
 819         int err, i;
 820
 821         err = tap_intr_handle_set(dev, 1);
 822         if (err)
 823                 return err;
 824
 825         err = tap_link_set_up(dev);
 826         if (err)
 827                 return err;
 828
 829         for (i = 0; i < dev->data->nb_tx_queues; i++)
 830                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 831         for (i = 0; i < dev->data->nb_rx_queues; i++)
 832                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 833
 834         return err;
 835 }
 836
 837 /* This function gets called when the current port gets stopped.
 838  */
 839 static void
 840 tap_dev_stop(struct rte_eth_dev *dev)
 841 {
 842         int i;
 843
 844         for (i = 0; i < dev->data->nb_tx_queues; i++)
 845                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 846         for (i = 0; i < dev->data->nb_rx_queues; i++)
 847                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 848
 849         tap_intr_handle_set(dev, 0);
 850         tap_link_set_down(dev);
 851 }
 852
 853 static int
 854 tap_dev_configure(struct rte_eth_dev *dev)
 855 {
 856         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 857                 TAP_LOG(ERR,
 858                         "%s: number of rx queues %d exceeds max num of queues %d",
 859                         dev->device->name,
 860                         dev->data->nb_rx_queues,
 861                         RTE_PMD_TAP_MAX_QUEUES);
 862                 return -1;
 863         }
 864         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 865                 TAP_LOG(ERR,
 866                         "%s: number of tx queues %d exceeds max num of queues %d",
 867                         dev->device->name,
 868                         dev->data->nb_tx_queues,
 869                         RTE_PMD_TAP_MAX_QUEUES);
 870                 return -1;
 871         }
 872
 873         TAP_LOG(INFO, "%s: %p: TX configured queues number: %u",
 874                 dev->device->name, (void *)dev, dev->data->nb_tx_queues);
 875
 876         TAP_LOG(INFO, "%s: %p: RX configured queues number: %u",
 877                 dev->device->name, (void *)dev, dev->data->nb_rx_queues);
 878
 879         return 0;
 880 }
 881
 882 static uint32_t
 883 tap_dev_speed_capa(void)
 884 {
 885         uint32_t speed = pmd_link.link_speed;
 886         uint32_t capa = 0;
 887
 888         if (speed >= ETH_SPEED_NUM_10M)
 889                 capa |= ETH_LINK_SPEED_10M;
 890         if (speed >= ETH_SPEED_NUM_100M)
 891                 capa |= ETH_LINK_SPEED_100M;
 892         if (speed >= ETH_SPEED_NUM_1G)
 893                 capa |= ETH_LINK_SPEED_1G;
 894         if (speed >= ETH_SPEED_NUM_5G)
 895                 capa |= ETH_LINK_SPEED_2_5G;
 896         if (speed >= ETH_SPEED_NUM_5G)
 897                 capa |= ETH_LINK_SPEED_5G;
 898         if (speed >= ETH_SPEED_NUM_10G)
 899                 capa |= ETH_LINK_SPEED_10G;
 900         if (speed >= ETH_SPEED_NUM_20G)
 901                 capa |= ETH_LINK_SPEED_20G;
 902         if (speed >= ETH_SPEED_NUM_25G)
 903                 capa |= ETH_LINK_SPEED_25G;
 904         if (speed >= ETH_SPEED_NUM_40G)
 905                 capa |= ETH_LINK_SPEED_40G;
 906         if (speed >= ETH_SPEED_NUM_50G)
 907                 capa |= ETH_LINK_SPEED_50G;
 908         if (speed >= ETH_SPEED_NUM_56G)
 909                 capa |= ETH_LINK_SPEED_56G;
 910         if (speed >= ETH_SPEED_NUM_100G)
 911                 capa |= ETH_LINK_SPEED_100G;
 912
 913         return capa;
 914 }
 915
 916 static void
 917 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 918 {
 919         struct pmd_internals *internals = dev->data->dev_private;
 920
 921         dev_info->if_index = internals->if_index;
 922         dev_info->max_mac_addrs = 1;
 923         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 924         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
 925         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
 926         dev_info->min_rx_bufsize = 0;
 927         dev_info->speed_capa = tap_dev_speed_capa();
 928         dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
 929         dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
 930                                     dev_info->rx_queue_offload_capa;
 931         dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
 932         dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
 933                                     dev_info->tx_queue_offload_capa;
 934         dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
 935         /*
 936          * limitation: TAP supports all of IP, UDP and TCP hash
 937          * functions together and not in partial combinations
 938          */
 939         dev_info->flow_type_rss_offloads = ~TAP_RSS_HF_MASK;
 940 }
 941
 942 static int
 943 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 944 {
 945         unsigned int i, imax;
 946         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 947         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 948         unsigned long rx_nombuf = 0, ierrors = 0;
 949         const struct pmd_internals *pmd = dev->data->dev_private;
 950
 951         /* rx queue statistics */
 952         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 953                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 954         for (i = 0; i < imax; i++) {
 955                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 956                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 957                 rx_total += tap_stats->q_ipackets[i];
 958                 rx_bytes_total += tap_stats->q_ibytes[i];
 959                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
 960                 ierrors += pmd->rxq[i].stats.ierrors;
 961         }
 962
 963         /* tx queue statistics */
 964         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 965                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 966
 967         for (i = 0; i < imax; i++) {
 968                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 969                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 970                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 971                 tx_total += tap_stats->q_opackets[i];
 972                 tx_err_total += tap_stats->q_errors[i];
 973                 tx_bytes_total += tap_stats->q_obytes[i];
 974         }
 975
 976         tap_stats->ipackets = rx_total;
 977         tap_stats->ibytes = rx_bytes_total;
 978         tap_stats->ierrors = ierrors;
 979         tap_stats->rx_nombuf = rx_nombuf;
 980         tap_stats->opackets = tx_total;
 981         tap_stats->oerrors = tx_err_total;
 982         tap_stats->obytes = tx_bytes_total;
 983         return 0;
 984 }
 985
 986 static void
 987 tap_stats_reset(struct rte_eth_dev *dev)
 988 {
 989         int i;
 990         struct pmd_internals *pmd = dev->data->dev_private;
 991
 992         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 993                 pmd->rxq[i].stats.ipackets = 0;
 994                 pmd->rxq[i].stats.ibytes = 0;
 995                 pmd->rxq[i].stats.ierrors = 0;
 996                 pmd->rxq[i].stats.rx_nombuf = 0;
 997
 998                 pmd->txq[i].stats.opackets = 0;
 999                 pmd->txq[i].stats.errs = 0;
1000                 pmd->txq[i].stats.obytes = 0;
1001         }
1002 }
1003
1004 static void
1005 tap_dev_close(struct rte_eth_dev *dev)
1006 {
1007         int i;
1008         struct pmd_internals *internals = dev->data->dev_private;
1009         struct pmd_process_private *process_private = dev->process_private;
1010
1011         tap_link_set_down(dev);
1012         tap_flow_flush(dev, NULL);
1013         tap_flow_implicit_flush(internals, NULL);
1014
1015         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1016                 if (process_private->rxq_fds[i] != -1) {
1017                         close(process_private->rxq_fds[i]);
1018                         process_private->rxq_fds[i] = -1;
1019                 }
1020                 if (process_private->txq_fds[i] != -1) {
1021                         close(process_private->txq_fds[i]);
1022                         process_private->txq_fds[i] = -1;
1023                 }
1024         }
1025
1026         if (internals->remote_if_index) {
1027                 /* Restore initial remote state */
1028                 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
1029                                 &internals->remote_initial_flags);
1030         }
1031
1032         if (internals->ka_fd != -1) {
1033                 close(internals->ka_fd);
1034                 internals->ka_fd = -1;
1035         }
1036         /*
1037          * Since TUN device has no more opened file descriptors
1038          * it will be removed from kernel
1039          */
1040 }
1041
1042 static void
1043 tap_rx_queue_release(void *queue)
1044 {
1045         struct rx_queue *rxq = queue;
1046         struct pmd_process_private *process_private;
1047
1048         if (!rxq)
1049                 return;
1050         process_private = rte_eth_devices[rxq->in_port].process_private;
1051         if (process_private->rxq_fds[rxq->queue_id] > 0) {
1052                 close(process_private->rxq_fds[rxq->queue_id]);
1053                 process_private->rxq_fds[rxq->queue_id] = -1;
1054                 rte_pktmbuf_free(rxq->pool);
1055                 rte_free(rxq->iovecs);
1056                 rxq->pool = NULL;
1057                 rxq->iovecs = NULL;
1058         }
1059 }
1060
1061 static void
1062 tap_tx_queue_release(void *queue)
1063 {
1064         struct tx_queue *txq = queue;
1065         struct pmd_process_private *process_private;
1066
1067         if (!txq)
1068                 return;
1069         process_private = rte_eth_devices[txq->out_port].process_private;
1070
1071         if (process_private->txq_fds[txq->queue_id] > 0) {
1072                 close(process_private->txq_fds[txq->queue_id]);
1073                 process_private->txq_fds[txq->queue_id] = -1;
1074         }
1075 }
1076
1077 static int
1078 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1079 {
1080         struct rte_eth_link *dev_link = &dev->data->dev_link;
1081         struct pmd_internals *pmd = dev->data->dev_private;
1082         struct ifreq ifr = { .ifr_flags = 0 };
1083
1084         if (pmd->remote_if_index) {
1085                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
1086                 if (!(ifr.ifr_flags & IFF_UP) ||
1087                     !(ifr.ifr_flags & IFF_RUNNING)) {
1088                         dev_link->link_status = ETH_LINK_DOWN;
1089                         return 0;
1090                 }
1091         }
1092         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
1093         dev_link->link_status =
1094                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
1095                  ETH_LINK_UP :
1096                  ETH_LINK_DOWN);
1097         return 0;
1098 }
1099
1100 static void
1101 tap_promisc_enable(struct rte_eth_dev *dev)
1102 {
1103         struct pmd_internals *pmd = dev->data->dev_private;
1104         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1105
1106         dev->data->promiscuous = 1;
1107         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1108         if (pmd->remote_if_index && !pmd->flow_isolate)
1109                 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
1110 }
1111
1112 static void
1113 tap_promisc_disable(struct rte_eth_dev *dev)
1114 {
1115         struct pmd_internals *pmd = dev->data->dev_private;
1116         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1117
1118         dev->data->promiscuous = 0;
1119         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1120         if (pmd->remote_if_index && !pmd->flow_isolate)
1121                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
1122 }
1123
1124 static void
1125 tap_allmulti_enable(struct rte_eth_dev *dev)
1126 {
1127         struct pmd_internals *pmd = dev->data->dev_private;
1128         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1129
1130         dev->data->all_multicast = 1;
1131         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1132         if (pmd->remote_if_index && !pmd->flow_isolate)
1133                 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
1134 }
1135
1136 static void
1137 tap_allmulti_disable(struct rte_eth_dev *dev)
1138 {
1139         struct pmd_internals *pmd = dev->data->dev_private;
1140         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1141
1142         dev->data->all_multicast = 0;
1143         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1144         if (pmd->remote_if_index && !pmd->flow_isolate)
1145                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
1146 }
1147
1148 static int
1149 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1150 {
1151         struct pmd_internals *pmd = dev->data->dev_private;
1152         enum ioctl_mode mode = LOCAL_ONLY;
1153         struct ifreq ifr;
1154         int ret;
1155
1156         if (pmd->type == ETH_TUNTAP_TYPE_TUN) {
1157                 TAP_LOG(ERR, "%s: can't MAC address for TUN",
1158                         dev->device->name);
1159                 return -ENOTSUP;
1160         }
1161
1162         if (is_zero_ether_addr(mac_addr)) {
1163                 TAP_LOG(ERR, "%s: can't set an empty MAC address",
1164                         dev->device->name);
1165                 return -EINVAL;
1166         }
1167         /* Check the actual current MAC address on the tap netdevice */
1168         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
1169         if (ret < 0)
1170                 return ret;
1171         if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
1172                                mac_addr))
1173                 return 0;
1174         /* Check the current MAC address on the remote */
1175         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
1176         if (ret < 0)
1177                 return ret;
1178         if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
1179                                mac_addr))
1180                 mode = LOCAL_AND_REMOTE;
1181         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1182         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
1183         ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
1184         if (ret < 0)
1185                 return ret;
1186         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
1187         if (pmd->remote_if_index && !pmd->flow_isolate) {
1188                 /* Replace MAC redirection rule after a MAC change */
1189                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
1190                 if (ret < 0) {
1191                         TAP_LOG(ERR,
1192                                 "%s: Couldn't delete MAC redirection rule",
1193                                 dev->device->name);
1194                         return ret;
1195                 }
1196                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1197                 if (ret < 0) {
1198                         TAP_LOG(ERR,
1199                                 "%s: Couldn't add MAC redirection rule",
1200                                 dev->device->name);
1201                         return ret;
1202                 }
1203         }
1204
1205         return 0;
1206 }
1207
1208 static int
1209 tap_gso_ctx_setup(struct rte_gso_ctx *gso_ctx, struct rte_eth_dev *dev)
1210 {
1211         uint32_t gso_types;
1212         char pool_name[64];
1213
1214         /*
1215          * Create private mbuf pool with TAP_GSO_MBUF_SEG_SIZE bytes
1216          * size per mbuf use this pool for both direct and indirect mbufs
1217          */
1218
1219         struct rte_mempool *mp;      /* Mempool for GSO packets */
1220
1221         /* initialize GSO context */
1222         gso_types = DEV_TX_OFFLOAD_TCP_TSO;
1223         snprintf(pool_name, sizeof(pool_name), "mp_%s", dev->device->name);
1224         mp = rte_mempool_lookup((const char *)pool_name);
1225         if (!mp) {
1226                 mp = rte_pktmbuf_pool_create(pool_name, TAP_GSO_MBUFS_NUM,
1227                         TAP_GSO_MBUF_CACHE_SIZE, 0,
1228                         RTE_PKTMBUF_HEADROOM + TAP_GSO_MBUF_SEG_SIZE,
1229                         SOCKET_ID_ANY);
1230                 if (!mp) {
1231                         struct pmd_internals *pmd = dev->data->dev_private;
1232                         RTE_LOG(DEBUG, PMD, "%s: failed to create mbuf pool for device %s\n",
1233                                 pmd->name, dev->device->name);
1234                         return -1;
1235                 }
1236         }
1237
1238         gso_ctx->direct_pool = mp;
1239         gso_ctx->indirect_pool = mp;
1240         gso_ctx->gso_types = gso_types;
1241         gso_ctx->gso_size = 0; /* gso_size is set in tx_burst() per packet */
1242         gso_ctx->flag = 0;
1243
1244         return 0;
1245 }
1246
1247 static int
1248 tap_setup_queue(struct rte_eth_dev *dev,
1249                 struct pmd_internals *internals,
1250                 uint16_t qid,
1251                 int is_rx)
1252 {
1253         int ret;
1254         int *fd;
1255         int *other_fd;
1256         const char *dir;
1257         struct pmd_internals *pmd = dev->data->dev_private;
1258         struct pmd_process_private *process_private = dev->process_private;
1259         struct rx_queue *rx = &internals->rxq[qid];
1260         struct tx_queue *tx = &internals->txq[qid];
1261         struct rte_gso_ctx *gso_ctx;
1262
1263         if (is_rx) {
1264                 fd = &process_private->rxq_fds[qid];
1265                 other_fd = &process_private->txq_fds[qid];
1266                 dir = "rx";
1267                 gso_ctx = NULL;
1268         } else {
1269                 fd = &process_private->txq_fds[qid];
1270                 other_fd = &process_private->rxq_fds[qid];
1271                 dir = "tx";
1272                 gso_ctx = &tx->gso_ctx;
1273         }
1274         if (*fd != -1) {
1275                 /* fd for this queue already exists */
1276                 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1277                         pmd->name, *fd, dir, qid);
1278                 gso_ctx = NULL;
1279         } else if (*other_fd != -1) {
1280                 /* Only other_fd exists. dup it */
1281                 *fd = dup(*other_fd);
1282                 if (*fd < 0) {
1283                         *fd = -1;
1284                         TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1285                         return -1;
1286                 }
1287                 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1288                         pmd->name, *other_fd, dir, qid, *fd);
1289         } else {
1290                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1291                 *fd = tun_alloc(pmd, 0);
1292                 if (*fd < 0) {
1293                         *fd = -1; /* restore original value */
1294                         TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1295                         return -1;
1296                 }
1297                 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1298                         pmd->name, dir, qid, *fd);
1299         }
1300
1301         tx->mtu = &dev->data->mtu;
1302         rx->rxmode = &dev->data->dev_conf.rxmode;
1303         if (gso_ctx) {
1304                 ret = tap_gso_ctx_setup(gso_ctx, dev);
1305                 if (ret)
1306                         return -1;
1307         }
1308
1309         tx->type = pmd->type;
1310
1311         return *fd;
1312 }
1313
1314 static int
1315 tap_rx_queue_setup(struct rte_eth_dev *dev,
1316                    uint16_t rx_queue_id,
1317                    uint16_t nb_rx_desc,
1318                    unsigned int socket_id,
1319                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1320                    struct rte_mempool *mp)
1321 {
1322         struct pmd_internals *internals = dev->data->dev_private;
1323         struct pmd_process_private *process_private = dev->process_private;
1324         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1325         struct rte_mbuf **tmp = &rxq->pool;
1326         long iov_max = sysconf(_SC_IOV_MAX);
1327         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1328         struct iovec (*iovecs)[nb_desc + 1];
1329         int data_off = RTE_PKTMBUF_HEADROOM;
1330         int ret = 0;
1331         int fd;
1332         int i;
1333
1334         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1335                 TAP_LOG(WARNING,
1336                         "nb_rx_queues %d too small or mempool NULL",
1337                         dev->data->nb_rx_queues);
1338                 return -1;
1339         }
1340
1341         rxq->mp = mp;
1342         rxq->trigger_seen = 1; /* force initial burst */
1343         rxq->in_port = dev->data->port_id;
1344         rxq->queue_id = rx_queue_id;
1345         rxq->nb_rx_desc = nb_desc;
1346         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1347                                     socket_id);
1348         if (!iovecs) {
1349                 TAP_LOG(WARNING,
1350                         "%s: Couldn't allocate %d RX descriptors",
1351                         dev->device->name, nb_desc);
1352                 return -ENOMEM;
1353         }
1354         rxq->iovecs = iovecs;
1355
1356         dev->data->rx_queues[rx_queue_id] = rxq;
1357         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1358         if (fd == -1) {
1359                 ret = fd;
1360                 goto error;
1361         }
1362
1363         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1364         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1365
1366         for (i = 1; i <= nb_desc; i++) {
1367                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1368                 if (!*tmp) {
1369                         TAP_LOG(WARNING,
1370                                 "%s: couldn't allocate memory for queue %d",
1371                                 dev->device->name, rx_queue_id);
1372                         ret = -ENOMEM;
1373                         goto error;
1374                 }
1375                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1376                 (*rxq->iovecs)[i].iov_base =
1377                         (char *)(*tmp)->buf_addr + data_off;
1378                 data_off = 0;
1379                 tmp = &(*tmp)->next;
1380         }
1381
1382         TAP_LOG(DEBUG, "  RX TUNTAP device name %s, qid %d on fd %d",
1383                 internals->name, rx_queue_id,
1384                 process_private->rxq_fds[rx_queue_id]);
1385
1386         return 0;
1387
1388 error:
1389         rte_pktmbuf_free(rxq->pool);
1390         rxq->pool = NULL;
1391         rte_free(rxq->iovecs);
1392         rxq->iovecs = NULL;
1393         return ret;
1394 }
1395
1396 static int
1397 tap_tx_queue_setup(struct rte_eth_dev *dev,
1398                    uint16_t tx_queue_id,
1399                    uint16_t nb_tx_desc __rte_unused,
1400                    unsigned int socket_id __rte_unused,
1401                    const struct rte_eth_txconf *tx_conf)
1402 {
1403         struct pmd_internals *internals = dev->data->dev_private;
1404         struct pmd_process_private *process_private = dev->process_private;
1405         struct tx_queue *txq;
1406         int ret;
1407         uint64_t offloads;
1408
1409         if (tx_queue_id >= dev->data->nb_tx_queues)
1410                 return -1;
1411         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1412         txq = dev->data->tx_queues[tx_queue_id];
1413         txq->out_port = dev->data->port_id;
1414         txq->queue_id = tx_queue_id;
1415
1416         offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1417         txq->csum = !!(offloads &
1418                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
1419                          DEV_TX_OFFLOAD_UDP_CKSUM |
1420                          DEV_TX_OFFLOAD_TCP_CKSUM));
1421
1422         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1423         if (ret == -1)
1424                 return -1;
1425         TAP_LOG(DEBUG,
1426                 "  TX TUNTAP device name %s, qid %d on fd %d csum %s",
1427                 internals->name, tx_queue_id,
1428                 process_private->txq_fds[tx_queue_id],
1429                 txq->csum ? "on" : "off");
1430
1431         return 0;
1432 }
1433
1434 static int
1435 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1436 {
1437         struct pmd_internals *pmd = dev->data->dev_private;
1438         struct ifreq ifr = { .ifr_mtu = mtu };
1439         int err = 0;
1440
1441         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1442         if (!err)
1443                 dev->data->mtu = mtu;
1444
1445         return err;
1446 }
1447
1448 static int
1449 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1450                      struct ether_addr *mc_addr_set __rte_unused,
1451                      uint32_t nb_mc_addr __rte_unused)
1452 {
1453         /*
1454          * Nothing to do actually: the tap has no filtering whatsoever, every
1455          * packet is received.
1456          */
1457         return 0;
1458 }
1459
1460 static int
1461 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1462 {
1463         struct rte_eth_dev *dev = arg;
1464         struct pmd_internals *pmd = dev->data->dev_private;
1465         struct ifinfomsg *info = NLMSG_DATA(nh);
1466
1467         if (nh->nlmsg_type != RTM_NEWLINK ||
1468             (info->ifi_index != pmd->if_index &&
1469              info->ifi_index != pmd->remote_if_index))
1470                 return 0;
1471         return tap_link_update(dev, 0);
1472 }
1473
1474 static void
1475 tap_dev_intr_handler(void *cb_arg)
1476 {
1477         struct rte_eth_dev *dev = cb_arg;
1478         struct pmd_internals *pmd = dev->data->dev_private;
1479
1480         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1481 }
1482
1483 static int
1484 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1485 {
1486         struct pmd_internals *pmd = dev->data->dev_private;
1487
1488         /* In any case, disable interrupt if the conf is no longer there. */
1489         if (!dev->data->dev_conf.intr_conf.lsc) {
1490                 if (pmd->intr_handle.fd != -1) {
1491                         tap_nl_final(pmd->intr_handle.fd);
1492                         rte_intr_callback_unregister(&pmd->intr_handle,
1493                                 tap_dev_intr_handler, dev);
1494                 }
1495                 return 0;
1496         }
1497         if (set) {
1498                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1499                 if (unlikely(pmd->intr_handle.fd == -1))
1500                         return -EBADF;
1501                 return rte_intr_callback_register(
1502                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1503         }
1504         tap_nl_final(pmd->intr_handle.fd);
1505         return rte_intr_callback_unregister(&pmd->intr_handle,
1506                                             tap_dev_intr_handler, dev);
1507 }
1508
1509 static int
1510 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1511 {
1512         int err;
1513
1514         err = tap_lsc_intr_handle_set(dev, set);
1515         if (err)
1516                 return err;
1517         err = tap_rx_intr_vec_set(dev, set);
1518         if (err && set)
1519                 tap_lsc_intr_handle_set(dev, 0);
1520         return err;
1521 }
1522
1523 static const uint32_t*
1524 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1525 {
1526         static const uint32_t ptypes[] = {
1527                 RTE_PTYPE_INNER_L2_ETHER,
1528                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1529                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1530                 RTE_PTYPE_INNER_L3_IPV4,
1531                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1532                 RTE_PTYPE_INNER_L3_IPV6,
1533                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1534                 RTE_PTYPE_INNER_L4_FRAG,
1535                 RTE_PTYPE_INNER_L4_UDP,
1536                 RTE_PTYPE_INNER_L4_TCP,
1537                 RTE_PTYPE_INNER_L4_SCTP,
1538                 RTE_PTYPE_L2_ETHER,
1539                 RTE_PTYPE_L2_ETHER_VLAN,
1540                 RTE_PTYPE_L2_ETHER_QINQ,
1541                 RTE_PTYPE_L3_IPV4,
1542                 RTE_PTYPE_L3_IPV4_EXT,
1543                 RTE_PTYPE_L3_IPV6_EXT,
1544                 RTE_PTYPE_L3_IPV6,
1545                 RTE_PTYPE_L4_FRAG,
1546                 RTE_PTYPE_L4_UDP,
1547                 RTE_PTYPE_L4_TCP,
1548                 RTE_PTYPE_L4_SCTP,
1549         };
1550
1551         return ptypes;
1552 }
1553
1554 static int
1555 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1556                   struct rte_eth_fc_conf *fc_conf)
1557 {
1558         fc_conf->mode = RTE_FC_NONE;
1559         return 0;
1560 }
1561
1562 static int
1563 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1564                   struct rte_eth_fc_conf *fc_conf)
1565 {
1566         if (fc_conf->mode != RTE_FC_NONE)
1567                 return -ENOTSUP;
1568         return 0;
1569 }
1570
1571 /**
1572  * DPDK callback to update the RSS hash configuration.
1573  *
1574  * @param dev
1575  *   Pointer to Ethernet device structure.
1576  * @param[in] rss_conf
1577  *   RSS configuration data.
1578  *
1579  * @return
1580  *   0 on success, a negative errno value otherwise and rte_errno is set.
1581  */
1582 static int
1583 tap_rss_hash_update(struct rte_eth_dev *dev,
1584                 struct rte_eth_rss_conf *rss_conf)
1585 {
1586         if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1587                 rte_errno = EINVAL;
1588                 return -rte_errno;
1589         }
1590         if (rss_conf->rss_key && rss_conf->rss_key_len) {
1591                 /*
1592                  * Currently TAP RSS key is hard coded
1593                  * and cannot be updated
1594                  */
1595                 TAP_LOG(ERR,
1596                         "port %u RSS key cannot be updated",
1597                         dev->data->port_id);
1598                 rte_errno = EINVAL;
1599                 return -rte_errno;
1600         }
1601         return 0;
1602 }
1603
1604 static int
1605 tap_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1606 {
1607         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1608
1609         return 0;
1610 }
1611
1612 static int
1613 tap_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1614 {
1615         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1616
1617         return 0;
1618 }
1619
1620 static int
1621 tap_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1622 {
1623         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1624
1625         return 0;
1626 }
1627
1628 static int
1629 tap_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1630 {
1631         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1632
1633         return 0;
1634 }
1635 static const struct eth_dev_ops ops = {
1636         .dev_start              = tap_dev_start,
1637         .dev_stop               = tap_dev_stop,
1638         .dev_close              = tap_dev_close,
1639         .dev_configure          = tap_dev_configure,
1640         .dev_infos_get          = tap_dev_info,
1641         .rx_queue_setup         = tap_rx_queue_setup,
1642         .tx_queue_setup         = tap_tx_queue_setup,
1643         .rx_queue_start         = tap_rx_queue_start,
1644         .tx_queue_start         = tap_tx_queue_start,
1645         .rx_queue_stop          = tap_rx_queue_stop,
1646         .tx_queue_stop          = tap_tx_queue_stop,
1647         .rx_queue_release       = tap_rx_queue_release,
1648         .tx_queue_release       = tap_tx_queue_release,
1649         .flow_ctrl_get          = tap_flow_ctrl_get,
1650         .flow_ctrl_set          = tap_flow_ctrl_set,
1651         .link_update            = tap_link_update,
1652         .dev_set_link_up        = tap_link_set_up,
1653         .dev_set_link_down      = tap_link_set_down,
1654         .promiscuous_enable     = tap_promisc_enable,
1655         .promiscuous_disable    = tap_promisc_disable,
1656         .allmulticast_enable    = tap_allmulti_enable,
1657         .allmulticast_disable   = tap_allmulti_disable,
1658         .mac_addr_set           = tap_mac_set,
1659         .mtu_set                = tap_mtu_set,
1660         .set_mc_addr_list       = tap_set_mc_addr_list,
1661         .stats_get              = tap_stats_get,
1662         .stats_reset            = tap_stats_reset,
1663         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1664         .rss_hash_update        = tap_rss_hash_update,
1665         .filter_ctrl            = tap_dev_filter_ctrl,
1666 };
1667
1668 static int
1669 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1670                    char *remote_iface, struct ether_addr *mac_addr,
1671                    enum rte_tuntap_type type)
1672 {
1673         int numa_node = rte_socket_id();
1674         struct rte_eth_dev *dev;
1675         struct pmd_internals *pmd;
1676         struct pmd_process_private *process_private;
1677         struct rte_eth_dev_data *data;
1678         struct ifreq ifr;
1679         int i;
1680
1681         TAP_LOG(DEBUG, "%s device on numa %u",
1682                         tuntap_name, rte_socket_id());
1683
1684         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1685         if (!dev) {
1686                 TAP_LOG(ERR, "%s Unable to allocate device struct",
1687                                 tuntap_name);
1688                 goto error_exit_nodev;
1689         }
1690
1691         process_private = (struct pmd_process_private *)
1692                 rte_zmalloc_socket(tap_name, sizeof(struct pmd_process_private),
1693                         RTE_CACHE_LINE_SIZE, dev->device->numa_node);
1694
1695         if (process_private == NULL) {
1696                 TAP_LOG(ERR, "Failed to alloc memory for process private");
1697                 return -1;
1698         }
1699         pmd = dev->data->dev_private;
1700         dev->process_private = process_private;
1701         pmd->dev = dev;
1702         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1703         pmd->type = type;
1704
1705         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1706         if (pmd->ioctl_sock == -1) {
1707                 TAP_LOG(ERR,
1708                         "%s Unable to get a socket for management: %s",
1709                         tuntap_name, strerror(errno));
1710                 goto error_exit;
1711         }
1712
1713         /* Setup some default values */
1714         data = dev->data;
1715         data->dev_private = pmd;
1716         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1717         data->numa_node = numa_node;
1718
1719         data->dev_link = pmd_link;
1720         data->mac_addrs = &pmd->eth_addr;
1721         /* Set the number of RX and TX queues */
1722         data->nb_rx_queues = 0;
1723         data->nb_tx_queues = 0;
1724
1725         dev->dev_ops = &ops;
1726         dev->rx_pkt_burst = pmd_rx_burst;
1727         dev->tx_pkt_burst = pmd_tx_burst;
1728
1729         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1730         pmd->intr_handle.fd = -1;
1731         dev->intr_handle = &pmd->intr_handle;
1732
1733         /* Presetup the fds to -1 as being not valid */
1734         pmd->ka_fd = -1;
1735         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1736                 process_private->rxq_fds[i] = -1;
1737                 process_private->txq_fds[i] = -1;
1738         }
1739
1740         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1741                 if (is_zero_ether_addr(mac_addr))
1742                         eth_random_addr((uint8_t *)&pmd->eth_addr);
1743                 else
1744                         rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1745         }
1746
1747         /*
1748          * Allocate a TUN device keep-alive file descriptor that will only be
1749          * closed when the TUN device itself is closed or removed.
1750          * This keep-alive file descriptor will guarantee that the TUN device
1751          * exists even when all of its queues are closed
1752          */
1753         pmd->ka_fd = tun_alloc(pmd, 1);
1754         if (pmd->ka_fd == -1) {
1755                 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
1756                 goto error_exit;
1757         }
1758
1759         ifr.ifr_mtu = dev->data->mtu;
1760         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1761                 goto error_exit;
1762
1763         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1764                 memset(&ifr, 0, sizeof(struct ifreq));
1765                 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1766                 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1767                                 ETHER_ADDR_LEN);
1768                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1769                         goto error_exit;
1770         }
1771
1772         /*
1773          * Set up everything related to rte_flow:
1774          * - netlink socket
1775          * - tap / remote if_index
1776          * - mandatory QDISCs
1777          * - rte_flow actual/implicit lists
1778          * - implicit rules
1779          */
1780         pmd->nlsk_fd = tap_nl_init(0);
1781         if (pmd->nlsk_fd == -1) {
1782                 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
1783                         pmd->name);
1784                 goto disable_rte_flow;
1785         }
1786         pmd->if_index = if_nametoindex(pmd->name);
1787         if (!pmd->if_index) {
1788                 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
1789                 goto disable_rte_flow;
1790         }
1791         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1792                 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
1793                         pmd->name);
1794                 goto disable_rte_flow;
1795         }
1796         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1797                 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1798                         pmd->name);
1799                 goto disable_rte_flow;
1800         }
1801         LIST_INIT(&pmd->flows);
1802
1803         if (strlen(remote_iface)) {
1804                 pmd->remote_if_index = if_nametoindex(remote_iface);
1805                 if (!pmd->remote_if_index) {
1806                         TAP_LOG(ERR, "%s: failed to get %s if_index.",
1807                                 pmd->name, remote_iface);
1808                         goto error_remote;
1809                 }
1810                 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1811                          "%s", remote_iface);
1812
1813                 /* Save state of remote device */
1814                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1815
1816                 /* Replicate remote MAC address */
1817                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1818                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1819                                 pmd->name, pmd->remote_iface);
1820                         goto error_remote;
1821                 }
1822                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1823                            ETHER_ADDR_LEN);
1824                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1825                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1826                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1827                                 pmd->name, remote_iface);
1828                         goto error_remote;
1829                 }
1830
1831                 /*
1832                  * Flush usually returns negative value because it tries to
1833                  * delete every QDISC (and on a running device, one QDISC at
1834                  * least is needed). Ignore negative return value.
1835                  */
1836                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1837                 if (qdisc_create_ingress(pmd->nlsk_fd,
1838                                          pmd->remote_if_index) < 0) {
1839                         TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1840                                 pmd->remote_iface);
1841                         goto error_remote;
1842                 }
1843                 LIST_INIT(&pmd->implicit_flows);
1844                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1845                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1846                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1847                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1848                         TAP_LOG(ERR,
1849                                 "%s: failed to create implicit rules.",
1850                                 pmd->name);
1851                         goto error_remote;
1852                 }
1853         }
1854
1855         rte_eth_dev_probing_finish(dev);
1856         return 0;
1857
1858 disable_rte_flow:
1859         TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
1860                 strerror(errno), errno);
1861         if (strlen(remote_iface)) {
1862                 TAP_LOG(ERR, "Remote feature requires flow support.");
1863                 goto error_exit;
1864         }
1865         rte_eth_dev_probing_finish(dev);
1866         return 0;
1867
1868 error_remote:
1869         TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
1870                 strerror(errno), errno);
1871         tap_flow_implicit_flush(pmd, NULL);
1872
1873 error_exit:
1874         if (pmd->ioctl_sock > 0)
1875                 close(pmd->ioctl_sock);
1876         /* mac_addrs must not be freed alone because part of dev_private */
1877         dev->data->mac_addrs = NULL;
1878         rte_eth_dev_release_port(dev);
1879
1880 error_exit_nodev:
1881         TAP_LOG(ERR, "%s Unable to initialize %s",
1882                 tuntap_name, rte_vdev_device_name(vdev));
1883
1884         return -EINVAL;
1885 }
1886
1887 static int
1888 set_interface_name(const char *key __rte_unused,
1889                    const char *value,
1890                    void *extra_args)
1891 {
1892         char *name = (char *)extra_args;
1893
1894         if (value)
1895                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN - 1);
1896         else
1897                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1898                          DEFAULT_TAP_NAME, (tap_unit - 1));
1899
1900         return 0;
1901 }
1902
1903 static int
1904 set_remote_iface(const char *key __rte_unused,
1905                  const char *value,
1906                  void *extra_args)
1907 {
1908         char *name = (char *)extra_args;
1909
1910         if (value)
1911                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
1912
1913         return 0;
1914 }
1915
1916 static int parse_user_mac(struct ether_addr *user_mac,
1917                 const char *value)
1918 {
1919         unsigned int index = 0;
1920         char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
1921
1922         if (user_mac == NULL || value == NULL)
1923                 return 0;
1924
1925         strlcpy(mac_temp, value, sizeof(mac_temp));
1926         mac_byte = strtok(mac_temp, ":");
1927
1928         while ((mac_byte != NULL) &&
1929                         (strlen(mac_byte) <= 2) &&
1930                         (strlen(mac_byte) == strspn(mac_byte,
1931                                         ETH_TAP_CMP_MAC_FMT))) {
1932                 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
1933                 mac_byte = strtok(NULL, ":");
1934         }
1935
1936         return index;
1937 }
1938
1939 static int
1940 set_mac_type(const char *key __rte_unused,
1941              const char *value,
1942              void *extra_args)
1943 {
1944         struct ether_addr *user_mac = extra_args;
1945
1946         if (!value)
1947                 return 0;
1948
1949         if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
1950                 static int iface_idx;
1951
1952                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1953                 memcpy((char *)user_mac->addr_bytes, "\0dtap", ETHER_ADDR_LEN);
1954                 user_mac->addr_bytes[ETHER_ADDR_LEN - 1] = iface_idx++ + '0';
1955                 goto success;
1956         }
1957
1958         if (parse_user_mac(user_mac, value) != 6)
1959                 goto error;
1960 success:
1961         TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
1962         return 0;
1963
1964 error:
1965         TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
1966                 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
1967         return -1;
1968 }
1969
1970 /*
1971  * Open a TUN interface device. TUN PMD
1972  * 1) sets tap_type as false
1973  * 2) intakes iface as argument.
1974  * 3) as interface is virtual set speed to 10G
1975  */
1976 static int
1977 rte_pmd_tun_probe(struct rte_vdev_device *dev)
1978 {
1979         const char *name, *params;
1980         int ret;
1981         struct rte_kvargs *kvlist = NULL;
1982         char tun_name[RTE_ETH_NAME_MAX_LEN];
1983         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1984         struct rte_eth_dev *eth_dev;
1985
1986         strcpy(tuntap_name, "TUN");
1987
1988         name = rte_vdev_device_name(dev);
1989         params = rte_vdev_device_args(dev);
1990         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1991
1992         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
1993             strlen(params) == 0) {
1994                 eth_dev = rte_eth_dev_attach_secondary(name);
1995                 if (!eth_dev) {
1996                         TAP_LOG(ERR, "Failed to probe %s", name);
1997                         return -1;
1998                 }
1999                 eth_dev->dev_ops = &ops;
2000                 eth_dev->device = &dev->device;
2001                 rte_eth_dev_probing_finish(eth_dev);
2002                 return 0;
2003         }
2004
2005         snprintf(tun_name, sizeof(tun_name), "%s%u",
2006                  DEFAULT_TUN_NAME, tun_unit++);
2007
2008         if (params && (params[0] != '\0')) {
2009                 TAP_LOG(DEBUG, "parameters (%s)", params);
2010
2011                 kvlist = rte_kvargs_parse(params, valid_arguments);
2012                 if (kvlist) {
2013                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2014                                 ret = rte_kvargs_process(kvlist,
2015                                         ETH_TAP_IFACE_ARG,
2016                                         &set_interface_name,
2017                                         tun_name);
2018
2019                                 if (ret == -1)
2020                                         goto leave;
2021                         }
2022                 }
2023         }
2024         pmd_link.link_speed = ETH_SPEED_NUM_10G;
2025
2026         TAP_LOG(NOTICE, "Initializing pmd_tun for %s as %s",
2027                 name, tun_name);
2028
2029         ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0,
2030                 ETH_TUNTAP_TYPE_TUN);
2031
2032 leave:
2033         if (ret == -1) {
2034                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2035                         name, tun_name);
2036                 tun_unit--; /* Restore the unit number */
2037         }
2038         rte_kvargs_free(kvlist);
2039
2040         return ret;
2041 }
2042
2043 /* Request queue file descriptors from secondary to primary. */
2044 static int
2045 tap_mp_attach_queues(const char *port_name, struct rte_eth_dev *dev)
2046 {
2047         int ret;
2048         struct timespec timeout = {.tv_sec = 1, .tv_nsec = 0};
2049         struct rte_mp_msg request, *reply;
2050         struct rte_mp_reply replies;
2051         struct ipc_queues *request_param = (struct ipc_queues *)request.param;
2052         struct ipc_queues *reply_param;
2053         struct pmd_process_private *process_private = dev->process_private;
2054         int queue, fd_iterator;
2055
2056         /* Prepare the request */
2057         strlcpy(request.name, TAP_MP_KEY, sizeof(request.name));
2058         strlcpy(request_param->port_name, port_name,
2059                 sizeof(request_param->port_name));
2060         request.len_param = sizeof(*request_param);
2061         /* Send request and receive reply */
2062         ret = rte_mp_request_sync(&request, &replies, &timeout);
2063         if (ret < 0) {
2064                 TAP_LOG(ERR, "Failed to request queues from primary: %d",
2065                         rte_errno);
2066                 return -1;
2067         }
2068         reply = &replies.msgs[0];
2069         reply_param = (struct ipc_queues *)reply->param;
2070         TAP_LOG(DEBUG, "Received IPC reply for %s", reply_param->port_name);
2071
2072         /* Attach the queues from received file descriptors */
2073         dev->data->nb_rx_queues = reply_param->rxq_count;
2074         dev->data->nb_tx_queues = reply_param->txq_count;
2075         fd_iterator = 0;
2076         for (queue = 0; queue < reply_param->rxq_count; queue++)
2077                 process_private->rxq_fds[queue] = reply->fds[fd_iterator++];
2078         for (queue = 0; queue < reply_param->txq_count; queue++)
2079                 process_private->txq_fds[queue] = reply->fds[fd_iterator++];
2080
2081         return 0;
2082 }
2083
2084 /* Send the queue file descriptors from the primary process to secondary. */
2085 static int
2086 tap_mp_sync_queues(const struct rte_mp_msg *request, const void *peer)
2087 {
2088         struct rte_eth_dev *dev;
2089         struct pmd_process_private *process_private;
2090         struct rte_mp_msg reply;
2091         const struct ipc_queues *request_param =
2092                 (const struct ipc_queues *)request->param;
2093         struct ipc_queues *reply_param =
2094                 (struct ipc_queues *)reply.param;
2095         uint16_t port_id;
2096         int queue;
2097         int ret;
2098
2099         /* Get requested port */
2100         TAP_LOG(DEBUG, "Received IPC request for %s", request_param->port_name);
2101         ret = rte_eth_dev_get_port_by_name(request_param->port_name, &port_id);
2102         if (ret) {
2103                 TAP_LOG(ERR, "Failed to get port id for %s",
2104                         request_param->port_name);
2105                 return -1;
2106         }
2107         dev = &rte_eth_devices[port_id];
2108         process_private = dev->process_private;
2109
2110         /* Fill file descriptors for all queues */
2111         reply.num_fds = 0;
2112         reply_param->rxq_count = 0;
2113         for (queue = 0; queue < dev->data->nb_rx_queues; queue++) {
2114                 reply.fds[reply.num_fds++] = process_private->rxq_fds[queue];
2115                 reply_param->rxq_count++;
2116         }
2117         RTE_ASSERT(reply_param->rxq_count == dev->data->nb_rx_queues);
2118         RTE_ASSERT(reply_param->txq_count == dev->data->nb_tx_queues);
2119         RTE_ASSERT(reply.num_fds <= RTE_MP_MAX_FD_NUM);
2120
2121         reply_param->txq_count = 0;
2122         for (queue = 0; queue < dev->data->nb_tx_queues; queue++) {
2123                 reply.fds[reply.num_fds++] = process_private->txq_fds[queue];
2124                 reply_param->txq_count++;
2125         }
2126
2127         /* Send reply */
2128         strlcpy(reply.name, request->name, sizeof(reply.name));
2129         strlcpy(reply_param->port_name, request_param->port_name,
2130                 sizeof(reply_param->port_name));
2131         reply.len_param = sizeof(*reply_param);
2132         if (rte_mp_reply(&reply, peer) < 0) {
2133                 TAP_LOG(ERR, "Failed to reply an IPC request to sync queues");
2134                 return -1;
2135         }
2136         return 0;
2137 }
2138
2139 /* Open a TAP interface device.
2140  */
2141 static int
2142 rte_pmd_tap_probe(struct rte_vdev_device *dev)
2143 {
2144         const char *name, *params;
2145         int ret;
2146         struct rte_kvargs *kvlist = NULL;
2147         int speed;
2148         char tap_name[RTE_ETH_NAME_MAX_LEN];
2149         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2150         struct ether_addr user_mac = { .addr_bytes = {0} };
2151         struct rte_eth_dev *eth_dev;
2152         int tap_devices_count_increased = 0;
2153
2154         strcpy(tuntap_name, "TAP");
2155
2156         name = rte_vdev_device_name(dev);
2157         params = rte_vdev_device_args(dev);
2158
2159         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
2160                 eth_dev = rte_eth_dev_attach_secondary(name);
2161                 if (!eth_dev) {
2162                         TAP_LOG(ERR, "Failed to probe %s", name);
2163                         return -1;
2164                 }
2165                 eth_dev->dev_ops = &ops;
2166                 eth_dev->device = &dev->device;
2167                 eth_dev->rx_pkt_burst = pmd_rx_burst;
2168                 eth_dev->tx_pkt_burst = pmd_tx_burst;
2169                 if (!rte_eal_primary_proc_alive(NULL)) {
2170                         TAP_LOG(ERR, "Primary process is missing");
2171                         return -1;
2172                 }
2173                 eth_dev->process_private = (struct pmd_process_private *)
2174                         rte_zmalloc_socket(name,
2175                                 sizeof(struct pmd_process_private),
2176                                 RTE_CACHE_LINE_SIZE,
2177                                 eth_dev->device->numa_node);
2178                 if (eth_dev->process_private == NULL) {
2179                         TAP_LOG(ERR,
2180                                 "Failed to alloc memory for process private");
2181                         return -1;
2182                 }
2183
2184                 ret = tap_mp_attach_queues(name, eth_dev);
2185                 if (ret != 0)
2186                         return -1;
2187                 rte_eth_dev_probing_finish(eth_dev);
2188                 return 0;
2189         }
2190
2191         speed = ETH_SPEED_NUM_10G;
2192         snprintf(tap_name, sizeof(tap_name), "%s%u",
2193                  DEFAULT_TAP_NAME, tap_unit++);
2194         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2195
2196         if (params && (params[0] != '\0')) {
2197                 TAP_LOG(DEBUG, "parameters (%s)", params);
2198
2199                 kvlist = rte_kvargs_parse(params, valid_arguments);
2200                 if (kvlist) {
2201                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2202                                 ret = rte_kvargs_process(kvlist,
2203                                                          ETH_TAP_IFACE_ARG,
2204                                                          &set_interface_name,
2205                                                          tap_name);
2206                                 if (ret == -1)
2207                                         goto leave;
2208                         }
2209
2210                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
2211                                 ret = rte_kvargs_process(kvlist,
2212                                                          ETH_TAP_REMOTE_ARG,
2213                                                          &set_remote_iface,
2214                                                          remote_iface);
2215                                 if (ret == -1)
2216                                         goto leave;
2217                         }
2218
2219                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
2220                                 ret = rte_kvargs_process(kvlist,
2221                                                          ETH_TAP_MAC_ARG,
2222                                                          &set_mac_type,
2223                                                          &user_mac);
2224                                 if (ret == -1)
2225                                         goto leave;
2226                         }
2227                 }
2228         }
2229         pmd_link.link_speed = speed;
2230
2231         TAP_LOG(NOTICE, "Initializing pmd_tap for %s as %s",
2232                 name, tap_name);
2233
2234         /* Register IPC feed callback */
2235         if (!tap_devices_count) {
2236                 ret = rte_mp_action_register(TAP_MP_KEY, tap_mp_sync_queues);
2237                 if (ret < 0) {
2238                         TAP_LOG(ERR, "%s: Failed to register IPC callback: %s",
2239                                 tuntap_name, strerror(rte_errno));
2240                         goto leave;
2241                 }
2242         }
2243         tap_devices_count++;
2244         tap_devices_count_increased = 1;
2245         ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
2246                 ETH_TUNTAP_TYPE_TAP);
2247
2248 leave:
2249         if (ret == -1) {
2250                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2251                         name, tap_name);
2252                 if (tap_devices_count_increased == 1) {
2253                         if (tap_devices_count == 1)
2254                                 rte_mp_action_unregister(TAP_MP_KEY);
2255                         tap_devices_count--;
2256                 }
2257                 tap_unit--;             /* Restore the unit number */
2258         }
2259         rte_kvargs_free(kvlist);
2260
2261         return ret;
2262 }
2263
2264 /* detach a TUNTAP device.
2265  */
2266 static int
2267 rte_pmd_tap_remove(struct rte_vdev_device *dev)
2268 {
2269         struct rte_eth_dev *eth_dev = NULL;
2270         struct pmd_internals *internals;
2271         struct pmd_process_private *process_private;
2272         int i;
2273
2274         /* find the ethdev entry */
2275         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
2276         if (!eth_dev)
2277                 return -ENODEV;
2278
2279         /* mac_addrs must not be freed alone because part of dev_private */
2280         eth_dev->data->mac_addrs = NULL;
2281
2282         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2283                 return rte_eth_dev_release_port(eth_dev);
2284
2285         internals = eth_dev->data->dev_private;
2286         process_private = eth_dev->process_private;
2287
2288         TAP_LOG(DEBUG, "Closing %s Ethernet device on numa %u",
2289                 (internals->type == ETH_TUNTAP_TYPE_TAP) ? "TAP" : "TUN",
2290                 rte_socket_id());
2291
2292         if (internals->nlsk_fd) {
2293                 tap_flow_flush(eth_dev, NULL);
2294                 tap_flow_implicit_flush(internals, NULL);
2295                 tap_nl_final(internals->nlsk_fd);
2296         }
2297         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
2298                 if (process_private->rxq_fds[i] != -1) {
2299                         close(process_private->rxq_fds[i]);
2300                         process_private->rxq_fds[i] = -1;
2301                 }
2302                 if (process_private->txq_fds[i] != -1) {
2303                         close(process_private->txq_fds[i]);
2304                         process_private->txq_fds[i] = -1;
2305                 }
2306         }
2307
2308         close(internals->ioctl_sock);
2309         rte_free(eth_dev->process_private);
2310         if (tap_devices_count == 1)
2311                 rte_mp_action_unregister(TAP_MP_KEY);
2312         tap_devices_count--;
2313         rte_eth_dev_release_port(eth_dev);
2314
2315         if (internals->ka_fd != -1) {
2316                 close(internals->ka_fd);
2317                 internals->ka_fd = -1;
2318         }
2319         return 0;
2320 }
2321
2322 static struct rte_vdev_driver pmd_tun_drv = {
2323         .probe = rte_pmd_tun_probe,
2324         .remove = rte_pmd_tap_remove,
2325 };
2326
2327 static struct rte_vdev_driver pmd_tap_drv = {
2328         .probe = rte_pmd_tap_probe,
2329         .remove = rte_pmd_tap_remove,
2330 };
2331
2332 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
2333 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
2334 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
2335 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
2336                               ETH_TAP_IFACE_ARG "=<string> ");
2337 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
2338                               ETH_TAP_IFACE_ARG "=<string> "
2339                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
2340                               ETH_TAP_REMOTE_ARG "=<string>");
2341 int tap_logtype;
2342
2343 RTE_INIT(tap_init_log)
2344 {
2345         tap_logtype = rte_log_register("pmd.net.tap");
2346         if (tap_logtype >= 0)
2347                 rte_log_set_level(tap_logtype, RTE_LOG_NOTICE);
2348 }