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 #include <ctype.h>
  41
  42 #include <tap_rss.h>
  43 #include <rte_eth_tap.h>
  44 #include <tap_flow.h>
  45 #include <tap_netlink.h>
  46 #include <tap_tcmsgs.h>
  47
  48 /* Linux based path to the TUN device */
  49 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  50 #define DEFAULT_TAP_NAME        "dtap"
  51 #define DEFAULT_TUN_NAME        "dtun"
  52
  53 #define ETH_TAP_IFACE_ARG       "iface"
  54 #define ETH_TAP_REMOTE_ARG      "remote"
  55 #define ETH_TAP_MAC_ARG         "mac"
  56 #define ETH_TAP_MAC_FIXED       "fixed"
  57
  58 #define ETH_TAP_USR_MAC_FMT     "xx:xx:xx:xx:xx:xx"
  59 #define ETH_TAP_CMP_MAC_FMT     "0123456789ABCDEFabcdef"
  60 #define ETH_TAP_MAC_ARG_FMT     ETH_TAP_MAC_FIXED "|" ETH_TAP_USR_MAC_FMT
  61
  62 #define TAP_GSO_MBUFS_PER_CORE  128
  63 #define TAP_GSO_MBUF_SEG_SIZE   128
  64 #define TAP_GSO_MBUF_CACHE_SIZE 4
  65 #define TAP_GSO_MBUFS_NUM \
  66         (TAP_GSO_MBUFS_PER_CORE * TAP_GSO_MBUF_CACHE_SIZE)
  67
  68 /* IPC key for queue fds sync */
  69 #define TAP_MP_KEY "tap_mp_sync_queues"
  70
  71 static int tap_devices_count;
  72 static struct rte_vdev_driver pmd_tap_drv;
  73 static struct rte_vdev_driver pmd_tun_drv;
  74
  75 static const char *valid_arguments[] = {
  76         ETH_TAP_IFACE_ARG,
  77         ETH_TAP_REMOTE_ARG,
  78         ETH_TAP_MAC_ARG,
  79         NULL
  80 };
  81
  82 static unsigned int tap_unit;
  83 static unsigned int tun_unit;
  84
  85 static char tuntap_name[8];
  86
  87 static volatile uint32_t tap_trigger;   /* Rx trigger */
  88
  89 static struct rte_eth_link pmd_link = {
  90         .link_speed = ETH_SPEED_NUM_10G,
  91         .link_duplex = ETH_LINK_FULL_DUPLEX,
  92         .link_status = ETH_LINK_DOWN,
  93         .link_autoneg = ETH_LINK_FIXED,
  94 };
  95
  96 static void
  97 tap_trigger_cb(int sig __rte_unused)
  98 {
  99         /* Valid trigger values are nonzero */
 100         tap_trigger = (tap_trigger + 1) | 0x80000000;
 101 }
 102
 103 /* Specifies on what netdevices the ioctl should be applied */
 104 enum ioctl_mode {
 105         LOCAL_AND_REMOTE,
 106         LOCAL_ONLY,
 107         REMOTE_ONLY,
 108 };
 109
 110 /* Message header to synchronize queues via IPC */
 111 struct ipc_queues {
 112         char port_name[RTE_DEV_NAME_MAX_LEN];
 113         int rxq_count;
 114         int txq_count;
 115         /*
 116          * The file descriptors are in the dedicated part
 117          * of the Unix message to be translated by the kernel.
 118          */
 119 };
 120
 121 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
 122
 123 /**
 124  * Tun/Tap allocation routine
 125  *
 126  * @param[in] pmd
 127  *   Pointer to private structure.
 128  *
 129  * @param[in] is_keepalive
 130  *   Keepalive flag
 131  *
 132  * @return
 133  *   -1 on failure, fd on success
 134  */
 135 static int
 136 tun_alloc(struct pmd_internals *pmd, int is_keepalive)
 137 {
 138         struct ifreq ifr;
 139 #ifdef IFF_MULTI_QUEUE
 140         unsigned int features;
 141 #endif
 142         int fd;
 143
 144         memset(&ifr, 0, sizeof(struct ifreq));
 145
 146         /*
 147          * Do not set IFF_NO_PI as packet information header will be needed
 148          * to check if a received packet has been truncated.
 149          */
 150         ifr.ifr_flags = (pmd->type == ETH_TUNTAP_TYPE_TAP) ?
 151                 IFF_TAP : IFF_TUN | IFF_POINTOPOINT;
 152         strlcpy(ifr.ifr_name, pmd->name, IFNAMSIZ);
 153
 154         TAP_LOG(DEBUG, "ifr_name '%s'", ifr.ifr_name);
 155
 156         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 157         if (fd < 0) {
 158                 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
 159                 goto error;
 160         }
 161
 162 #ifdef IFF_MULTI_QUEUE
 163         /* Grab the TUN features to verify we can work multi-queue */
 164         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 165                 TAP_LOG(ERR, "%s unable to get TUN/TAP features",
 166                         tuntap_name);
 167                 goto error;
 168         }
 169         TAP_LOG(DEBUG, "%s Features %08x", tuntap_name, features);
 170
 171         if (features & IFF_MULTI_QUEUE) {
 172                 TAP_LOG(DEBUG, "  Multi-queue support for %d queues",
 173                         RTE_PMD_TAP_MAX_QUEUES);
 174                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 175         } else
 176 #endif
 177         {
 178                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 179                 TAP_LOG(DEBUG, "  Single queue only support");
 180         }
 181
 182         /* Set the TUN/TAP configuration and set the name if needed */
 183         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 184                 TAP_LOG(WARNING, "Unable to set TUNSETIFF for %s: %s",
 185                         ifr.ifr_name, strerror(errno));
 186                 goto error;
 187         }
 188
 189         if (is_keepalive) {
 190                 /*
 191                  * Detach the TUN/TAP keep-alive queue
 192                  * to avoid traffic through it
 193                  */
 194                 ifr.ifr_flags = IFF_DETACH_QUEUE;
 195                 if (ioctl(fd, TUNSETQUEUE, (void *)&ifr) < 0) {
 196                         TAP_LOG(WARNING,
 197                                 "Unable to detach keep-alive queue for %s: %s",
 198                                 ifr.ifr_name, strerror(errno));
 199                         goto error;
 200                 }
 201         }
 202
 203         /* Always set the file descriptor to non-blocking */
 204         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 205                 TAP_LOG(WARNING,
 206                         "Unable to set %s to nonblocking: %s",
 207                         ifr.ifr_name, strerror(errno));
 208                 goto error;
 209         }
 210
 211         /* Set up trigger to optimize empty Rx bursts */
 212         errno = 0;
 213         do {
 214                 struct sigaction sa;
 215                 int flags = fcntl(fd, F_GETFL);
 216
 217                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 218                         break;
 219                 if (sa.sa_handler != tap_trigger_cb) {
 220                         /*
 221                          * Make sure SIGIO is not already taken. This is done
 222                          * as late as possible to leave the application a
 223                          * chance to set up its own signal handler first.
 224                          */
 225                         if (sa.sa_handler != SIG_IGN &&
 226                             sa.sa_handler != SIG_DFL) {
 227                                 errno = EBUSY;
 228                                 break;
 229                         }
 230                         sa = (struct sigaction){
 231                                 .sa_flags = SA_RESTART,
 232                                 .sa_handler = tap_trigger_cb,
 233                         };
 234                         if (sigaction(SIGIO, &sa, NULL) == -1)
 235                                 break;
 236                 }
 237                 /* Enable SIGIO on file descriptor */
 238                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 239                 fcntl(fd, F_SETOWN, getpid());
 240         } while (0);
 241
 242         if (errno) {
 243                 /* Disable trigger globally in case of error */
 244                 tap_trigger = 0;
 245                 TAP_LOG(WARNING, "Rx trigger disabled: %s",
 246                         strerror(errno));
 247         }
 248
 249         return fd;
 250
 251 error:
 252         if (fd >= 0)
 253                 close(fd);
 254         return -1;
 255 }
 256
 257 static void
 258 tap_verify_csum(struct rte_mbuf *mbuf)
 259 {
 260         uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
 261         uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
 262         uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
 263         unsigned int l2_len = sizeof(struct ether_hdr);
 264         unsigned int l3_len;
 265         uint16_t cksum = 0;
 266         void *l3_hdr;
 267         void *l4_hdr;
 268
 269         if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
 270                 l2_len += 4;
 271         else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
 272                 l2_len += 8;
 273         /* Don't verify checksum for packets with discontinuous L2 header */
 274         if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
 275                      rte_pktmbuf_data_len(mbuf)))
 276                 return;
 277         l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
 278         if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 279                 struct ipv4_hdr *iph = l3_hdr;
 280
 281                 /* ihl contains the number of 4-byte words in the header */
 282                 l3_len = 4 * (iph->version_ihl & 0xf);
 283                 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
 284                         return;
 285                 /* check that the total length reported by header is not
 286                  * greater than the total received size
 287                  */
 288                 if (l2_len + rte_be_to_cpu_16(iph->total_length) >
 289                                 rte_pktmbuf_data_len(mbuf))
 290                         return;
 291
 292                 cksum = ~rte_raw_cksum(iph, l3_len);
 293                 mbuf->ol_flags |= cksum ?
 294                         PKT_RX_IP_CKSUM_BAD :
 295                         PKT_RX_IP_CKSUM_GOOD;
 296         } else if (l3 == RTE_PTYPE_L3_IPV6) {
 297                 struct ipv6_hdr *iph = l3_hdr;
 298
 299                 l3_len = sizeof(struct ipv6_hdr);
 300                 /* check that the total length reported by header is not
 301                  * greater than the total received size
 302                  */
 303                 if (l2_len + l3_len + rte_be_to_cpu_16(iph->payload_len) >
 304                                 rte_pktmbuf_data_len(mbuf))
 305                         return;
 306         } else {
 307                 /* IPv6 extensions are not supported */
 308                 return;
 309         }
 310         if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
 311                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 312                 /* Don't verify checksum for multi-segment packets. */
 313                 if (mbuf->nb_segs > 1)
 314                         return;
 315                 if (l3 == RTE_PTYPE_L3_IPV4)
 316                         cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
 317                 else if (l3 == RTE_PTYPE_L3_IPV6)
 318                         cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
 319                 mbuf->ol_flags |= cksum ?
 320                         PKT_RX_L4_CKSUM_BAD :
 321                         PKT_RX_L4_CKSUM_GOOD;
 322         }
 323 }
 324
 325 static uint64_t
 326 tap_rx_offload_get_port_capa(void)
 327 {
 328         /*
 329          * No specific port Rx offload capabilities.
 330          */
 331         return 0;
 332 }
 333
 334 static uint64_t
 335 tap_rx_offload_get_queue_capa(void)
 336 {
 337         return DEV_RX_OFFLOAD_SCATTER |
 338                DEV_RX_OFFLOAD_IPV4_CKSUM |
 339                DEV_RX_OFFLOAD_UDP_CKSUM |
 340                DEV_RX_OFFLOAD_TCP_CKSUM;
 341 }
 342
 343 /* Callback to handle the rx burst of packets to the correct interface and
 344  * file descriptor(s) in a multi-queue setup.
 345  */
 346 static uint16_t
 347 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 348 {
 349         struct rx_queue *rxq = queue;
 350         struct pmd_process_private *process_private;
 351         uint16_t num_rx;
 352         unsigned long num_rx_bytes = 0;
 353         uint32_t trigger = tap_trigger;
 354
 355         if (trigger == rxq->trigger_seen)
 356                 return 0;
 357         if (trigger)
 358                 rxq->trigger_seen = trigger;
 359         process_private = rte_eth_devices[rxq->in_port].process_private;
 360         rte_compiler_barrier();
 361         for (num_rx = 0; num_rx < nb_pkts; ) {
 362                 struct rte_mbuf *mbuf = rxq->pool;
 363                 struct rte_mbuf *seg = NULL;
 364                 struct rte_mbuf *new_tail = NULL;
 365                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 366                 int len;
 367
 368                 len = readv(process_private->rxq_fds[rxq->queue_id],
 369                         *rxq->iovecs,
 370                         1 + (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
 371                              rxq->nb_rx_desc : 1));
 372                 if (len < (int)sizeof(struct tun_pi))
 373                         break;
 374
 375                 /* Packet couldn't fit in the provided mbuf */
 376                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 377                         rxq->stats.ierrors++;
 378                         continue;
 379                 }
 380
 381                 len -= sizeof(struct tun_pi);
 382
 383                 mbuf->pkt_len = len;
 384                 mbuf->port = rxq->in_port;
 385                 while (1) {
 386                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 387
 388                         if (unlikely(!buf)) {
 389                                 rxq->stats.rx_nombuf++;
 390                                 /* No new buf has been allocated: do nothing */
 391                                 if (!new_tail || !seg)
 392                                         goto end;
 393
 394                                 seg->next = NULL;
 395                                 rte_pktmbuf_free(mbuf);
 396
 397                                 goto end;
 398                         }
 399                         seg = seg ? seg->next : mbuf;
 400                         if (rxq->pool == mbuf)
 401                                 rxq->pool = buf;
 402                         if (new_tail)
 403                                 new_tail->next = buf;
 404                         new_tail = buf;
 405                         new_tail->next = seg->next;
 406
 407                         /* iovecs[0] is reserved for packet info (pi) */
 408                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 409                                 buf->buf_len - data_off;
 410                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 411                                 (char *)buf->buf_addr + data_off;
 412
 413                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 414                         seg->data_off = data_off;
 415
 416                         len -= seg->data_len;
 417                         if (len <= 0)
 418                                 break;
 419                         mbuf->nb_segs++;
 420                         /* First segment has headroom, not the others */
 421                         data_off = 0;
 422                 }
 423                 seg->next = NULL;
 424                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 425                                                       RTE_PTYPE_ALL_MASK);
 426                 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
 427                         tap_verify_csum(mbuf);
 428
 429                 /* account for the receive frame */
 430                 bufs[num_rx++] = mbuf;
 431                 num_rx_bytes += mbuf->pkt_len;
 432         }
 433 end:
 434         rxq->stats.ipackets += num_rx;
 435         rxq->stats.ibytes += num_rx_bytes;
 436
 437         return num_rx;
 438 }
 439
 440 static uint64_t
 441 tap_tx_offload_get_port_capa(void)
 442 {
 443         /*
 444          * No specific port Tx offload capabilities.
 445          */
 446         return 0;
 447 }
 448
 449 static uint64_t
 450 tap_tx_offload_get_queue_capa(void)
 451 {
 452         return DEV_TX_OFFLOAD_MULTI_SEGS |
 453                DEV_TX_OFFLOAD_IPV4_CKSUM |
 454                DEV_TX_OFFLOAD_UDP_CKSUM |
 455                DEV_TX_OFFLOAD_TCP_CKSUM |
 456                DEV_TX_OFFLOAD_TCP_TSO;
 457 }
 458
 459 /* Finalize l4 checksum calculation */
 460 static void
 461 tap_tx_l4_cksum(uint16_t *l4_cksum, uint16_t l4_phdr_cksum,
 462                 uint32_t l4_raw_cksum)
 463 {
 464         if (l4_cksum) {
 465                 uint32_t cksum;
 466
 467                 cksum = __rte_raw_cksum_reduce(l4_raw_cksum);
 468                 cksum += l4_phdr_cksum;
 469
 470                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 471                 cksum = (~cksum) & 0xffff;
 472                 if (cksum == 0)
 473                         cksum = 0xffff;
 474                 *l4_cksum = cksum;
 475         }
 476 }
 477
 478 /* Accumaulate L4 raw checksums */
 479 static void
 480 tap_tx_l4_add_rcksum(char *l4_data, unsigned int l4_len, uint16_t *l4_cksum,
 481                         uint32_t *l4_raw_cksum)
 482 {
 483         if (l4_cksum == NULL)
 484                 return;
 485
 486         *l4_raw_cksum = __rte_raw_cksum(l4_data, l4_len, *l4_raw_cksum);
 487 }
 488
 489 /* L3 and L4 pseudo headers checksum offloads */
 490 static void
 491 tap_tx_l3_cksum(char *packet, uint64_t ol_flags, unsigned int l2_len,
 492                 unsigned int l3_len, unsigned int l4_len, uint16_t **l4_cksum,
 493                 uint16_t *l4_phdr_cksum, uint32_t *l4_raw_cksum)
 494 {
 495         void *l3_hdr = packet + l2_len;
 496
 497         if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
 498                 struct ipv4_hdr *iph = l3_hdr;
 499                 uint16_t cksum;
 500
 501                 iph->hdr_checksum = 0;
 502                 cksum = rte_raw_cksum(iph, l3_len);
 503                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 504         }
 505         if (ol_flags & PKT_TX_L4_MASK) {
 506                 void *l4_hdr;
 507
 508                 l4_hdr = packet + l2_len + l3_len;
 509                 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
 510                         *l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
 511                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 512                         *l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
 513                 else
 514                         return;
 515                 **l4_cksum = 0;
 516                 if (ol_flags & PKT_TX_IPV4)
 517                         *l4_phdr_cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 518                 else
 519                         *l4_phdr_cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 520                 *l4_raw_cksum = __rte_raw_cksum(l4_hdr, l4_len, 0);
 521         }
 522 }
 523
 524 static inline void
 525 tap_write_mbufs(struct tx_queue *txq, uint16_t num_mbufs,
 526                         struct rte_mbuf **pmbufs,
 527                         uint16_t *num_packets, unsigned long *num_tx_bytes)
 528 {
 529         int i;
 530         uint16_t l234_hlen;
 531         struct pmd_process_private *process_private;
 532
 533         process_private = rte_eth_devices[txq->out_port].process_private;
 534
 535         for (i = 0; i < num_mbufs; i++) {
 536                 struct rte_mbuf *mbuf = pmbufs[i];
 537                 struct iovec iovecs[mbuf->nb_segs + 2];
 538                 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
 539                 struct rte_mbuf *seg = mbuf;
 540                 char m_copy[mbuf->data_len];
 541                 int proto;
 542                 int n;
 543                 int j;
 544                 int k; /* current index in iovecs for copying segments */
 545                 uint16_t seg_len; /* length of first segment */
 546                 uint16_t nb_segs;
 547                 uint16_t *l4_cksum; /* l4 checksum (pseudo header + payload) */
 548                 uint32_t l4_raw_cksum = 0; /* TCP/UDP payload raw checksum */
 549                 uint16_t l4_phdr_cksum = 0; /* TCP/UDP pseudo header checksum */
 550                 uint16_t is_cksum = 0; /* in case cksum should be offloaded */
 551
 552                 l4_cksum = NULL;
 553                 if (txq->type == ETH_TUNTAP_TYPE_TUN) {
 554                         /*
 555                          * TUN and TAP are created with IFF_NO_PI disabled.
 556                          * For TUN PMD this mandatory as fields are used by
 557                          * Kernel tun.c to determine whether its IP or non IP
 558                          * packets.
 559                          *
 560                          * The logic fetches the first byte of data from mbuf
 561                          * then compares whether its v4 or v6. If first byte
 562                          * is 4 or 6, then protocol field is updated.
 563                          */
 564                         char *buff_data = rte_pktmbuf_mtod(seg, void *);
 565                         proto = (*buff_data & 0xf0);
 566                         pi.proto = (proto == 0x40) ?
 567                                 rte_cpu_to_be_16(ETHER_TYPE_IPv4) :
 568                                 ((proto == 0x60) ?
 569                                         rte_cpu_to_be_16(ETHER_TYPE_IPv6) :
 570                                         0x00);
 571                 }
 572
 573                 k = 0;
 574                 iovecs[k].iov_base = &pi;
 575                 iovecs[k].iov_len = sizeof(pi);
 576                 k++;
 577
 578                 nb_segs = mbuf->nb_segs;
 579                 if (txq->csum &&
 580                     ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 581                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 582                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
 583                         is_cksum = 1;
 584
 585                         /* Support only packets with at least layer 4
 586                          * header included in the first segment
 587                          */
 588                         seg_len = rte_pktmbuf_data_len(mbuf);
 589                         l234_hlen = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
 590                         if (seg_len < l234_hlen)
 591                                 break;
 592
 593                         /* To change checksums, work on a * copy of l2, l3
 594                          * headers + l4 pseudo header
 595                          */
 596                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 597                                         l234_hlen);
 598                         tap_tx_l3_cksum(m_copy, mbuf->ol_flags,
 599                                        mbuf->l2_len, mbuf->l3_len, mbuf->l4_len,
 600                                        &l4_cksum, &l4_phdr_cksum,
 601                                        &l4_raw_cksum);
 602                         iovecs[k].iov_base = m_copy;
 603                         iovecs[k].iov_len = l234_hlen;
 604                         k++;
 605
 606                         /* Update next iovecs[] beyond l2, l3, l4 headers */
 607                         if (seg_len > l234_hlen) {
 608                                 iovecs[k].iov_len = seg_len - l234_hlen;
 609                                 iovecs[k].iov_base =
 610                                         rte_pktmbuf_mtod(seg, char *) +
 611                                                 l234_hlen;
 612                                 tap_tx_l4_add_rcksum(iovecs[k].iov_base,
 613                                         iovecs[k].iov_len, l4_cksum,
 614                                         &l4_raw_cksum);
 615                                 k++;
 616                                 nb_segs++;
 617                         }
 618                         seg = seg->next;
 619                 }
 620
 621                 for (j = k; j <= nb_segs; j++) {
 622                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 623                         iovecs[j].iov_base = rte_pktmbuf_mtod(seg, void *);
 624                         if (is_cksum)
 625                                 tap_tx_l4_add_rcksum(iovecs[j].iov_base,
 626                                         iovecs[j].iov_len, l4_cksum,
 627                                         &l4_raw_cksum);
 628                         seg = seg->next;
 629                 }
 630
 631                 if (is_cksum)
 632                         tap_tx_l4_cksum(l4_cksum, l4_phdr_cksum, l4_raw_cksum);
 633
 634                 /* copy the tx frame data */
 635                 n = writev(process_private->txq_fds[txq->queue_id], iovecs, j);
 636                 if (n <= 0)
 637                         break;
 638                 (*num_packets)++;
 639                 (*num_tx_bytes) += rte_pktmbuf_pkt_len(mbuf);
 640         }
 641 }
 642
 643 /* Callback to handle sending packets from the tap interface
 644  */
 645 static uint16_t
 646 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 647 {
 648         struct tx_queue *txq = queue;
 649         uint16_t num_tx = 0;
 650         uint16_t num_packets = 0;
 651         unsigned long num_tx_bytes = 0;
 652         uint32_t max_size;
 653         int i;
 654
 655         if (unlikely(nb_pkts == 0))
 656                 return 0;
 657
 658         struct rte_mbuf *gso_mbufs[MAX_GSO_MBUFS];
 659         max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
 660         for (i = 0; i < nb_pkts; i++) {
 661                 struct rte_mbuf *mbuf_in = bufs[num_tx];
 662                 struct rte_mbuf **mbuf;
 663                 uint16_t num_mbufs = 0;
 664                 uint16_t tso_segsz = 0;
 665                 int ret;
 666                 uint16_t hdrs_len;
 667                 int j;
 668                 uint64_t tso;
 669
 670                 tso = mbuf_in->ol_flags & PKT_TX_TCP_SEG;
 671                 if (tso) {
 672                         struct rte_gso_ctx *gso_ctx = &txq->gso_ctx;
 673
 674                         assert(gso_ctx != NULL);
 675
 676                         /* TCP segmentation implies TCP checksum offload */
 677                         mbuf_in->ol_flags |= PKT_TX_TCP_CKSUM;
 678
 679                         /* gso size is calculated without ETHER_CRC_LEN */
 680                         hdrs_len = mbuf_in->l2_len + mbuf_in->l3_len +
 681                                         mbuf_in->l4_len;
 682                         tso_segsz = mbuf_in->tso_segsz + hdrs_len;
 683                         if (unlikely(tso_segsz == hdrs_len) ||
 684                                 tso_segsz > *txq->mtu) {
 685                                 txq->stats.errs++;
 686                                 break;
 687                         }
 688                         gso_ctx->gso_size = tso_segsz;
 689                         ret = rte_gso_segment(mbuf_in, /* packet to segment */
 690                                 gso_ctx, /* gso control block */
 691                                 (struct rte_mbuf **)&gso_mbufs, /* out mbufs */
 692                                 RTE_DIM(gso_mbufs)); /* max tso mbufs */
 693
 694                         /* ret contains the number of new created mbufs */
 695                         if (ret < 0)
 696                                 break;
 697
 698                         mbuf = gso_mbufs;
 699                         num_mbufs = ret;
 700                 } else {
 701                         /* stats.errs will be incremented */
 702                         if (rte_pktmbuf_pkt_len(mbuf_in) > max_size)
 703                                 break;
 704
 705                         /* ret 0 indicates no new mbufs were created */
 706                         ret = 0;
 707                         mbuf = &mbuf_in;
 708                         num_mbufs = 1;
 709                 }
 710
 711                 tap_write_mbufs(txq, num_mbufs, mbuf,
 712                                 &num_packets, &num_tx_bytes);
 713                 num_tx++;
 714                 /* free original mbuf */
 715                 rte_pktmbuf_free(mbuf_in);
 716                 /* free tso mbufs */
 717                 for (j = 0; j < ret; j++)
 718                         rte_pktmbuf_free(mbuf[j]);
 719         }
 720
 721         txq->stats.opackets += num_packets;
 722         txq->stats.errs += nb_pkts - num_tx;
 723         txq->stats.obytes += num_tx_bytes;
 724
 725         return num_packets;
 726 }
 727
 728 static const char *
 729 tap_ioctl_req2str(unsigned long request)
 730 {
 731         switch (request) {
 732         case SIOCSIFFLAGS:
 733                 return "SIOCSIFFLAGS";
 734         case SIOCGIFFLAGS:
 735                 return "SIOCGIFFLAGS";
 736         case SIOCGIFHWADDR:
 737                 return "SIOCGIFHWADDR";
 738         case SIOCSIFHWADDR:
 739                 return "SIOCSIFHWADDR";
 740         case SIOCSIFMTU:
 741                 return "SIOCSIFMTU";
 742         }
 743         return "UNKNOWN";
 744 }
 745
 746 static int
 747 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 748           struct ifreq *ifr, int set, enum ioctl_mode mode)
 749 {
 750         short req_flags = ifr->ifr_flags;
 751         int remote = pmd->remote_if_index &&
 752                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 753
 754         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 755                 return 0;
 756         /*
 757          * If there is a remote netdevice, apply ioctl on it, then apply it on
 758          * the tap netdevice.
 759          */
 760 apply:
 761         if (remote)
 762                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
 763         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 764                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 765         switch (request) {
 766         case SIOCSIFFLAGS:
 767                 /* fetch current flags to leave other flags untouched */
 768                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 769                         goto error;
 770                 if (set)
 771                         ifr->ifr_flags |= req_flags;
 772                 else
 773                         ifr->ifr_flags &= ~req_flags;
 774                 break;
 775         case SIOCGIFFLAGS:
 776         case SIOCGIFHWADDR:
 777         case SIOCSIFHWADDR:
 778         case SIOCSIFMTU:
 779                 break;
 780         default:
 781                 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
 782                         pmd->name);
 783                 return -EINVAL;
 784         }
 785         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 786                 goto error;
 787         if (remote-- && mode == LOCAL_AND_REMOTE)
 788                 goto apply;
 789         return 0;
 790
 791 error:
 792         TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
 793                 tap_ioctl_req2str(request), strerror(errno), errno);
 794         return -errno;
 795 }
 796
 797 static int
 798 tap_link_set_down(struct rte_eth_dev *dev)
 799 {
 800         struct pmd_internals *pmd = dev->data->dev_private;
 801         struct ifreq ifr = { .ifr_flags = IFF_UP };
 802
 803         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 804         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 805 }
 806
 807 static int
 808 tap_link_set_up(struct rte_eth_dev *dev)
 809 {
 810         struct pmd_internals *pmd = dev->data->dev_private;
 811         struct ifreq ifr = { .ifr_flags = IFF_UP };
 812
 813         dev->data->dev_link.link_status = ETH_LINK_UP;
 814         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 815 }
 816
 817 static int
 818 tap_dev_start(struct rte_eth_dev *dev)
 819 {
 820         int err, i;
 821
 822         err = tap_intr_handle_set(dev, 1);
 823         if (err)
 824                 return err;
 825
 826         err = tap_link_set_up(dev);
 827         if (err)
 828                 return err;
 829
 830         for (i = 0; i < dev->data->nb_tx_queues; i++)
 831                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 832         for (i = 0; i < dev->data->nb_rx_queues; i++)
 833                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 834
 835         return err;
 836 }
 837
 838 /* This function gets called when the current port gets stopped.
 839  */
 840 static void
 841 tap_dev_stop(struct rte_eth_dev *dev)
 842 {
 843         int i;
 844
 845         for (i = 0; i < dev->data->nb_tx_queues; i++)
 846                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 847         for (i = 0; i < dev->data->nb_rx_queues; i++)
 848                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 849
 850         tap_intr_handle_set(dev, 0);
 851         tap_link_set_down(dev);
 852 }
 853
 854 static int
 855 tap_dev_configure(struct rte_eth_dev *dev)
 856 {
 857         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 858                 TAP_LOG(ERR,
 859                         "%s: number of rx queues %d exceeds max num of queues %d",
 860                         dev->device->name,
 861                         dev->data->nb_rx_queues,
 862                         RTE_PMD_TAP_MAX_QUEUES);
 863                 return -1;
 864         }
 865         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 866                 TAP_LOG(ERR,
 867                         "%s: number of tx queues %d exceeds max num of queues %d",
 868                         dev->device->name,
 869                         dev->data->nb_tx_queues,
 870                         RTE_PMD_TAP_MAX_QUEUES);
 871                 return -1;
 872         }
 873
 874         TAP_LOG(INFO, "%s: %p: TX configured queues number: %u",
 875                 dev->device->name, (void *)dev, dev->data->nb_tx_queues);
 876
 877         TAP_LOG(INFO, "%s: %p: RX configured queues number: %u",
 878                 dev->device->name, (void *)dev, dev->data->nb_rx_queues);
 879
 880         return 0;
 881 }
 882
 883 static uint32_t
 884 tap_dev_speed_capa(void)
 885 {
 886         uint32_t speed = pmd_link.link_speed;
 887         uint32_t capa = 0;
 888
 889         if (speed >= ETH_SPEED_NUM_10M)
 890                 capa |= ETH_LINK_SPEED_10M;
 891         if (speed >= ETH_SPEED_NUM_100M)
 892                 capa |= ETH_LINK_SPEED_100M;
 893         if (speed >= ETH_SPEED_NUM_1G)
 894                 capa |= ETH_LINK_SPEED_1G;
 895         if (speed >= ETH_SPEED_NUM_5G)
 896                 capa |= ETH_LINK_SPEED_2_5G;
 897         if (speed >= ETH_SPEED_NUM_5G)
 898                 capa |= ETH_LINK_SPEED_5G;
 899         if (speed >= ETH_SPEED_NUM_10G)
 900                 capa |= ETH_LINK_SPEED_10G;
 901         if (speed >= ETH_SPEED_NUM_20G)
 902                 capa |= ETH_LINK_SPEED_20G;
 903         if (speed >= ETH_SPEED_NUM_25G)
 904                 capa |= ETH_LINK_SPEED_25G;
 905         if (speed >= ETH_SPEED_NUM_40G)
 906                 capa |= ETH_LINK_SPEED_40G;
 907         if (speed >= ETH_SPEED_NUM_50G)
 908                 capa |= ETH_LINK_SPEED_50G;
 909         if (speed >= ETH_SPEED_NUM_56G)
 910                 capa |= ETH_LINK_SPEED_56G;
 911         if (speed >= ETH_SPEED_NUM_100G)
 912                 capa |= ETH_LINK_SPEED_100G;
 913
 914         return capa;
 915 }
 916
 917 static void
 918 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 919 {
 920         struct pmd_internals *internals = dev->data->dev_private;
 921
 922         dev_info->if_index = internals->if_index;
 923         dev_info->max_mac_addrs = 1;
 924         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 925         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
 926         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
 927         dev_info->min_rx_bufsize = 0;
 928         dev_info->speed_capa = tap_dev_speed_capa();
 929         dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
 930         dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
 931                                     dev_info->rx_queue_offload_capa;
 932         dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
 933         dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
 934                                     dev_info->tx_queue_offload_capa;
 935         dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
 936         /*
 937          * limitation: TAP supports all of IP, UDP and TCP hash
 938          * functions together and not in partial combinations
 939          */
 940         dev_info->flow_type_rss_offloads = ~TAP_RSS_HF_MASK;
 941 }
 942
 943 static int
 944 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 945 {
 946         unsigned int i, imax;
 947         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 948         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 949         unsigned long rx_nombuf = 0, ierrors = 0;
 950         const struct pmd_internals *pmd = dev->data->dev_private;
 951
 952         /* rx queue statistics */
 953         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 954                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 955         for (i = 0; i < imax; i++) {
 956                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 957                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 958                 rx_total += tap_stats->q_ipackets[i];
 959                 rx_bytes_total += tap_stats->q_ibytes[i];
 960                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
 961                 ierrors += pmd->rxq[i].stats.ierrors;
 962         }
 963
 964         /* tx queue statistics */
 965         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 966                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 967
 968         for (i = 0; i < imax; i++) {
 969                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 970                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 971                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 972                 tx_total += tap_stats->q_opackets[i];
 973                 tx_err_total += tap_stats->q_errors[i];
 974                 tx_bytes_total += tap_stats->q_obytes[i];
 975         }
 976
 977         tap_stats->ipackets = rx_total;
 978         tap_stats->ibytes = rx_bytes_total;
 979         tap_stats->ierrors = ierrors;
 980         tap_stats->rx_nombuf = rx_nombuf;
 981         tap_stats->opackets = tx_total;
 982         tap_stats->oerrors = tx_err_total;
 983         tap_stats->obytes = tx_bytes_total;
 984         return 0;
 985 }
 986
 987 static void
 988 tap_stats_reset(struct rte_eth_dev *dev)
 989 {
 990         int i;
 991         struct pmd_internals *pmd = dev->data->dev_private;
 992
 993         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 994                 pmd->rxq[i].stats.ipackets = 0;
 995                 pmd->rxq[i].stats.ibytes = 0;
 996                 pmd->rxq[i].stats.ierrors = 0;
 997                 pmd->rxq[i].stats.rx_nombuf = 0;
 998
 999                 pmd->txq[i].stats.opackets = 0;
1000                 pmd->txq[i].stats.errs = 0;
1001                 pmd->txq[i].stats.obytes = 0;
1002         }
1003 }
1004
1005 static void
1006 tap_dev_close(struct rte_eth_dev *dev)
1007 {
1008         int i;
1009         struct pmd_internals *internals = dev->data->dev_private;
1010         struct pmd_process_private *process_private = dev->process_private;
1011
1012         tap_link_set_down(dev);
1013         tap_flow_flush(dev, NULL);
1014         tap_flow_implicit_flush(internals, NULL);
1015
1016         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1017                 if (process_private->rxq_fds[i] != -1) {
1018                         close(process_private->rxq_fds[i]);
1019                         process_private->rxq_fds[i] = -1;
1020                 }
1021                 if (process_private->txq_fds[i] != -1) {
1022                         close(process_private->txq_fds[i]);
1023                         process_private->txq_fds[i] = -1;
1024                 }
1025         }
1026
1027         if (internals->remote_if_index) {
1028                 /* Restore initial remote state */
1029                 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
1030                                 &internals->remote_initial_flags);
1031         }
1032
1033         if (internals->ka_fd != -1) {
1034                 close(internals->ka_fd);
1035                 internals->ka_fd = -1;
1036         }
1037         /*
1038          * Since TUN device has no more opened file descriptors
1039          * it will be removed from kernel
1040          */
1041 }
1042
1043 static void
1044 tap_rx_queue_release(void *queue)
1045 {
1046         struct rx_queue *rxq = queue;
1047         struct pmd_process_private *process_private;
1048
1049         if (!rxq)
1050                 return;
1051         process_private = rte_eth_devices[rxq->in_port].process_private;
1052         if (process_private->rxq_fds[rxq->queue_id] > 0) {
1053                 close(process_private->rxq_fds[rxq->queue_id]);
1054                 process_private->rxq_fds[rxq->queue_id] = -1;
1055                 rte_pktmbuf_free(rxq->pool);
1056                 rte_free(rxq->iovecs);
1057                 rxq->pool = NULL;
1058                 rxq->iovecs = NULL;
1059         }
1060 }
1061
1062 static void
1063 tap_tx_queue_release(void *queue)
1064 {
1065         struct tx_queue *txq = queue;
1066         struct pmd_process_private *process_private;
1067
1068         if (!txq)
1069                 return;
1070         process_private = rte_eth_devices[txq->out_port].process_private;
1071
1072         if (process_private->txq_fds[txq->queue_id] > 0) {
1073                 close(process_private->txq_fds[txq->queue_id]);
1074                 process_private->txq_fds[txq->queue_id] = -1;
1075         }
1076 }
1077
1078 static int
1079 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1080 {
1081         struct rte_eth_link *dev_link = &dev->data->dev_link;
1082         struct pmd_internals *pmd = dev->data->dev_private;
1083         struct ifreq ifr = { .ifr_flags = 0 };
1084
1085         if (pmd->remote_if_index) {
1086                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
1087                 if (!(ifr.ifr_flags & IFF_UP) ||
1088                     !(ifr.ifr_flags & IFF_RUNNING)) {
1089                         dev_link->link_status = ETH_LINK_DOWN;
1090                         return 0;
1091                 }
1092         }
1093         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
1094         dev_link->link_status =
1095                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
1096                  ETH_LINK_UP :
1097                  ETH_LINK_DOWN);
1098         return 0;
1099 }
1100
1101 static void
1102 tap_promisc_enable(struct rte_eth_dev *dev)
1103 {
1104         struct pmd_internals *pmd = dev->data->dev_private;
1105         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1106
1107         dev->data->promiscuous = 1;
1108         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1109         if (pmd->remote_if_index && !pmd->flow_isolate)
1110                 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
1111 }
1112
1113 static void
1114 tap_promisc_disable(struct rte_eth_dev *dev)
1115 {
1116         struct pmd_internals *pmd = dev->data->dev_private;
1117         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1118
1119         dev->data->promiscuous = 0;
1120         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1121         if (pmd->remote_if_index && !pmd->flow_isolate)
1122                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
1123 }
1124
1125 static void
1126 tap_allmulti_enable(struct rte_eth_dev *dev)
1127 {
1128         struct pmd_internals *pmd = dev->data->dev_private;
1129         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1130
1131         dev->data->all_multicast = 1;
1132         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1133         if (pmd->remote_if_index && !pmd->flow_isolate)
1134                 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
1135 }
1136
1137 static void
1138 tap_allmulti_disable(struct rte_eth_dev *dev)
1139 {
1140         struct pmd_internals *pmd = dev->data->dev_private;
1141         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1142
1143         dev->data->all_multicast = 0;
1144         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1145         if (pmd->remote_if_index && !pmd->flow_isolate)
1146                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
1147 }
1148
1149 static int
1150 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1151 {
1152         struct pmd_internals *pmd = dev->data->dev_private;
1153         enum ioctl_mode mode = LOCAL_ONLY;
1154         struct ifreq ifr;
1155         int ret;
1156
1157         if (pmd->type == ETH_TUNTAP_TYPE_TUN) {
1158                 TAP_LOG(ERR, "%s: can't MAC address for TUN",
1159                         dev->device->name);
1160                 return -ENOTSUP;
1161         }
1162
1163         if (is_zero_ether_addr(mac_addr)) {
1164                 TAP_LOG(ERR, "%s: can't set an empty MAC address",
1165                         dev->device->name);
1166                 return -EINVAL;
1167         }
1168         /* Check the actual current MAC address on the tap netdevice */
1169         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
1170         if (ret < 0)
1171                 return ret;
1172         if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
1173                                mac_addr))
1174                 return 0;
1175         /* Check the current MAC address on the remote */
1176         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
1177         if (ret < 0)
1178                 return ret;
1179         if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
1180                                mac_addr))
1181                 mode = LOCAL_AND_REMOTE;
1182         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1183         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
1184         ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
1185         if (ret < 0)
1186                 return ret;
1187         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
1188         if (pmd->remote_if_index && !pmd->flow_isolate) {
1189                 /* Replace MAC redirection rule after a MAC change */
1190                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
1191                 if (ret < 0) {
1192                         TAP_LOG(ERR,
1193                                 "%s: Couldn't delete MAC redirection rule",
1194                                 dev->device->name);
1195                         return ret;
1196                 }
1197                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1198                 if (ret < 0) {
1199                         TAP_LOG(ERR,
1200                                 "%s: Couldn't add MAC redirection rule",
1201                                 dev->device->name);
1202                         return ret;
1203                 }
1204         }
1205
1206         return 0;
1207 }
1208
1209 static int
1210 tap_gso_ctx_setup(struct rte_gso_ctx *gso_ctx, struct rte_eth_dev *dev)
1211 {
1212         uint32_t gso_types;
1213         char pool_name[64];
1214
1215         /*
1216          * Create private mbuf pool with TAP_GSO_MBUF_SEG_SIZE bytes
1217          * size per mbuf use this pool for both direct and indirect mbufs
1218          */
1219
1220         struct rte_mempool *mp;      /* Mempool for GSO packets */
1221
1222         /* initialize GSO context */
1223         gso_types = DEV_TX_OFFLOAD_TCP_TSO;
1224         snprintf(pool_name, sizeof(pool_name), "mp_%s", dev->device->name);
1225         mp = rte_mempool_lookup((const char *)pool_name);
1226         if (!mp) {
1227                 mp = rte_pktmbuf_pool_create(pool_name, TAP_GSO_MBUFS_NUM,
1228                         TAP_GSO_MBUF_CACHE_SIZE, 0,
1229                         RTE_PKTMBUF_HEADROOM + TAP_GSO_MBUF_SEG_SIZE,
1230                         SOCKET_ID_ANY);
1231                 if (!mp) {
1232                         struct pmd_internals *pmd = dev->data->dev_private;
1233                         RTE_LOG(DEBUG, PMD, "%s: failed to create mbuf pool for device %s\n",
1234                                 pmd->name, dev->device->name);
1235                         return -1;
1236                 }
1237         }
1238
1239         gso_ctx->direct_pool = mp;
1240         gso_ctx->indirect_pool = mp;
1241         gso_ctx->gso_types = gso_types;
1242         gso_ctx->gso_size = 0; /* gso_size is set in tx_burst() per packet */
1243         gso_ctx->flag = 0;
1244
1245         return 0;
1246 }
1247
1248 static int
1249 tap_setup_queue(struct rte_eth_dev *dev,
1250                 struct pmd_internals *internals,
1251                 uint16_t qid,
1252                 int is_rx)
1253 {
1254         int ret;
1255         int *fd;
1256         int *other_fd;
1257         const char *dir;
1258         struct pmd_internals *pmd = dev->data->dev_private;
1259         struct pmd_process_private *process_private = dev->process_private;
1260         struct rx_queue *rx = &internals->rxq[qid];
1261         struct tx_queue *tx = &internals->txq[qid];
1262         struct rte_gso_ctx *gso_ctx;
1263
1264         if (is_rx) {
1265                 fd = &process_private->rxq_fds[qid];
1266                 other_fd = &process_private->txq_fds[qid];
1267                 dir = "rx";
1268                 gso_ctx = NULL;
1269         } else {
1270                 fd = &process_private->txq_fds[qid];
1271                 other_fd = &process_private->rxq_fds[qid];
1272                 dir = "tx";
1273                 gso_ctx = &tx->gso_ctx;
1274         }
1275         if (*fd != -1) {
1276                 /* fd for this queue already exists */
1277                 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1278                         pmd->name, *fd, dir, qid);
1279                 gso_ctx = NULL;
1280         } else if (*other_fd != -1) {
1281                 /* Only other_fd exists. dup it */
1282                 *fd = dup(*other_fd);
1283                 if (*fd < 0) {
1284                         *fd = -1;
1285                         TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1286                         return -1;
1287                 }
1288                 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1289                         pmd->name, *other_fd, dir, qid, *fd);
1290         } else {
1291                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1292                 *fd = tun_alloc(pmd, 0);
1293                 if (*fd < 0) {
1294                         *fd = -1; /* restore original value */
1295                         TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1296                         return -1;
1297                 }
1298                 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1299                         pmd->name, dir, qid, *fd);
1300         }
1301
1302         tx->mtu = &dev->data->mtu;
1303         rx->rxmode = &dev->data->dev_conf.rxmode;
1304         if (gso_ctx) {
1305                 ret = tap_gso_ctx_setup(gso_ctx, dev);
1306                 if (ret)
1307                         return -1;
1308         }
1309
1310         tx->type = pmd->type;
1311
1312         return *fd;
1313 }
1314
1315 static int
1316 tap_rx_queue_setup(struct rte_eth_dev *dev,
1317                    uint16_t rx_queue_id,
1318                    uint16_t nb_rx_desc,
1319                    unsigned int socket_id,
1320                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1321                    struct rte_mempool *mp)
1322 {
1323         struct pmd_internals *internals = dev->data->dev_private;
1324         struct pmd_process_private *process_private = dev->process_private;
1325         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1326         struct rte_mbuf **tmp = &rxq->pool;
1327         long iov_max = sysconf(_SC_IOV_MAX);
1328         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1329         struct iovec (*iovecs)[nb_desc + 1];
1330         int data_off = RTE_PKTMBUF_HEADROOM;
1331         int ret = 0;
1332         int fd;
1333         int i;
1334
1335         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1336                 TAP_LOG(WARNING,
1337                         "nb_rx_queues %d too small or mempool NULL",
1338                         dev->data->nb_rx_queues);
1339                 return -1;
1340         }
1341
1342         rxq->mp = mp;
1343         rxq->trigger_seen = 1; /* force initial burst */
1344         rxq->in_port = dev->data->port_id;
1345         rxq->queue_id = rx_queue_id;
1346         rxq->nb_rx_desc = nb_desc;
1347         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1348                                     socket_id);
1349         if (!iovecs) {
1350                 TAP_LOG(WARNING,
1351                         "%s: Couldn't allocate %d RX descriptors",
1352                         dev->device->name, nb_desc);
1353                 return -ENOMEM;
1354         }
1355         rxq->iovecs = iovecs;
1356
1357         dev->data->rx_queues[rx_queue_id] = rxq;
1358         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1359         if (fd == -1) {
1360                 ret = fd;
1361                 goto error;
1362         }
1363
1364         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1365         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1366
1367         for (i = 1; i <= nb_desc; i++) {
1368                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1369                 if (!*tmp) {
1370                         TAP_LOG(WARNING,
1371                                 "%s: couldn't allocate memory for queue %d",
1372                                 dev->device->name, rx_queue_id);
1373                         ret = -ENOMEM;
1374                         goto error;
1375                 }
1376                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1377                 (*rxq->iovecs)[i].iov_base =
1378                         (char *)(*tmp)->buf_addr + data_off;
1379                 data_off = 0;
1380                 tmp = &(*tmp)->next;
1381         }
1382
1383         TAP_LOG(DEBUG, "  RX TUNTAP device name %s, qid %d on fd %d",
1384                 internals->name, rx_queue_id,
1385                 process_private->rxq_fds[rx_queue_id]);
1386
1387         return 0;
1388
1389 error:
1390         rte_pktmbuf_free(rxq->pool);
1391         rxq->pool = NULL;
1392         rte_free(rxq->iovecs);
1393         rxq->iovecs = NULL;
1394         return ret;
1395 }
1396
1397 static int
1398 tap_tx_queue_setup(struct rte_eth_dev *dev,
1399                    uint16_t tx_queue_id,
1400                    uint16_t nb_tx_desc __rte_unused,
1401                    unsigned int socket_id __rte_unused,
1402                    const struct rte_eth_txconf *tx_conf)
1403 {
1404         struct pmd_internals *internals = dev->data->dev_private;
1405         struct pmd_process_private *process_private = dev->process_private;
1406         struct tx_queue *txq;
1407         int ret;
1408         uint64_t offloads;
1409
1410         if (tx_queue_id >= dev->data->nb_tx_queues)
1411                 return -1;
1412         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1413         txq = dev->data->tx_queues[tx_queue_id];
1414         txq->out_port = dev->data->port_id;
1415         txq->queue_id = tx_queue_id;
1416
1417         offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1418         txq->csum = !!(offloads &
1419                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
1420                          DEV_TX_OFFLOAD_UDP_CKSUM |
1421                          DEV_TX_OFFLOAD_TCP_CKSUM));
1422
1423         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1424         if (ret == -1)
1425                 return -1;
1426         TAP_LOG(DEBUG,
1427                 "  TX TUNTAP device name %s, qid %d on fd %d csum %s",
1428                 internals->name, tx_queue_id,
1429                 process_private->txq_fds[tx_queue_id],
1430                 txq->csum ? "on" : "off");
1431
1432         return 0;
1433 }
1434
1435 static int
1436 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1437 {
1438         struct pmd_internals *pmd = dev->data->dev_private;
1439         struct ifreq ifr = { .ifr_mtu = mtu };
1440         int err = 0;
1441
1442         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1443         if (!err)
1444                 dev->data->mtu = mtu;
1445
1446         return err;
1447 }
1448
1449 static int
1450 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1451                      struct ether_addr *mc_addr_set __rte_unused,
1452                      uint32_t nb_mc_addr __rte_unused)
1453 {
1454         /*
1455          * Nothing to do actually: the tap has no filtering whatsoever, every
1456          * packet is received.
1457          */
1458         return 0;
1459 }
1460
1461 static int
1462 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1463 {
1464         struct rte_eth_dev *dev = arg;
1465         struct pmd_internals *pmd = dev->data->dev_private;
1466         struct ifinfomsg *info = NLMSG_DATA(nh);
1467
1468         if (nh->nlmsg_type != RTM_NEWLINK ||
1469             (info->ifi_index != pmd->if_index &&
1470              info->ifi_index != pmd->remote_if_index))
1471                 return 0;
1472         return tap_link_update(dev, 0);
1473 }
1474
1475 static void
1476 tap_dev_intr_handler(void *cb_arg)
1477 {
1478         struct rte_eth_dev *dev = cb_arg;
1479         struct pmd_internals *pmd = dev->data->dev_private;
1480
1481         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1482 }
1483
1484 static int
1485 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1486 {
1487         struct pmd_internals *pmd = dev->data->dev_private;
1488
1489         /* In any case, disable interrupt if the conf is no longer there. */
1490         if (!dev->data->dev_conf.intr_conf.lsc) {
1491                 if (pmd->intr_handle.fd != -1) {
1492                         tap_nl_final(pmd->intr_handle.fd);
1493                         rte_intr_callback_unregister(&pmd->intr_handle,
1494                                 tap_dev_intr_handler, dev);
1495                 }
1496                 return 0;
1497         }
1498         if (set) {
1499                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1500                 if (unlikely(pmd->intr_handle.fd == -1))
1501                         return -EBADF;
1502                 return rte_intr_callback_register(
1503                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1504         }
1505         tap_nl_final(pmd->intr_handle.fd);
1506         return rte_intr_callback_unregister(&pmd->intr_handle,
1507                                             tap_dev_intr_handler, dev);
1508 }
1509
1510 static int
1511 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1512 {
1513         int err;
1514
1515         err = tap_lsc_intr_handle_set(dev, set);
1516         if (err)
1517                 return err;
1518         err = tap_rx_intr_vec_set(dev, set);
1519         if (err && set)
1520                 tap_lsc_intr_handle_set(dev, 0);
1521         return err;
1522 }
1523
1524 static const uint32_t*
1525 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1526 {
1527         static const uint32_t ptypes[] = {
1528                 RTE_PTYPE_INNER_L2_ETHER,
1529                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1530                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1531                 RTE_PTYPE_INNER_L3_IPV4,
1532                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1533                 RTE_PTYPE_INNER_L3_IPV6,
1534                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1535                 RTE_PTYPE_INNER_L4_FRAG,
1536                 RTE_PTYPE_INNER_L4_UDP,
1537                 RTE_PTYPE_INNER_L4_TCP,
1538                 RTE_PTYPE_INNER_L4_SCTP,
1539                 RTE_PTYPE_L2_ETHER,
1540                 RTE_PTYPE_L2_ETHER_VLAN,
1541                 RTE_PTYPE_L2_ETHER_QINQ,
1542                 RTE_PTYPE_L3_IPV4,
1543                 RTE_PTYPE_L3_IPV4_EXT,
1544                 RTE_PTYPE_L3_IPV6_EXT,
1545                 RTE_PTYPE_L3_IPV6,
1546                 RTE_PTYPE_L4_FRAG,
1547                 RTE_PTYPE_L4_UDP,
1548                 RTE_PTYPE_L4_TCP,
1549                 RTE_PTYPE_L4_SCTP,
1550         };
1551
1552         return ptypes;
1553 }
1554
1555 static int
1556 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1557                   struct rte_eth_fc_conf *fc_conf)
1558 {
1559         fc_conf->mode = RTE_FC_NONE;
1560         return 0;
1561 }
1562
1563 static int
1564 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1565                   struct rte_eth_fc_conf *fc_conf)
1566 {
1567         if (fc_conf->mode != RTE_FC_NONE)
1568                 return -ENOTSUP;
1569         return 0;
1570 }
1571
1572 /**
1573  * DPDK callback to update the RSS hash configuration.
1574  *
1575  * @param dev
1576  *   Pointer to Ethernet device structure.
1577  * @param[in] rss_conf
1578  *   RSS configuration data.
1579  *
1580  * @return
1581  *   0 on success, a negative errno value otherwise and rte_errno is set.
1582  */
1583 static int
1584 tap_rss_hash_update(struct rte_eth_dev *dev,
1585                 struct rte_eth_rss_conf *rss_conf)
1586 {
1587         if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1588                 rte_errno = EINVAL;
1589                 return -rte_errno;
1590         }
1591         if (rss_conf->rss_key && rss_conf->rss_key_len) {
1592                 /*
1593                  * Currently TAP RSS key is hard coded
1594                  * and cannot be updated
1595                  */
1596                 TAP_LOG(ERR,
1597                         "port %u RSS key cannot be updated",
1598                         dev->data->port_id);
1599                 rte_errno = EINVAL;
1600                 return -rte_errno;
1601         }
1602         return 0;
1603 }
1604
1605 static int
1606 tap_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1607 {
1608         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1609
1610         return 0;
1611 }
1612
1613 static int
1614 tap_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1615 {
1616         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1617
1618         return 0;
1619 }
1620
1621 static int
1622 tap_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1623 {
1624         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1625
1626         return 0;
1627 }
1628
1629 static int
1630 tap_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1631 {
1632         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1633
1634         return 0;
1635 }
1636 static const struct eth_dev_ops ops = {
1637         .dev_start              = tap_dev_start,
1638         .dev_stop               = tap_dev_stop,
1639         .dev_close              = tap_dev_close,
1640         .dev_configure          = tap_dev_configure,
1641         .dev_infos_get          = tap_dev_info,
1642         .rx_queue_setup         = tap_rx_queue_setup,
1643         .tx_queue_setup         = tap_tx_queue_setup,
1644         .rx_queue_start         = tap_rx_queue_start,
1645         .tx_queue_start         = tap_tx_queue_start,
1646         .rx_queue_stop          = tap_rx_queue_stop,
1647         .tx_queue_stop          = tap_tx_queue_stop,
1648         .rx_queue_release       = tap_rx_queue_release,
1649         .tx_queue_release       = tap_tx_queue_release,
1650         .flow_ctrl_get          = tap_flow_ctrl_get,
1651         .flow_ctrl_set          = tap_flow_ctrl_set,
1652         .link_update            = tap_link_update,
1653         .dev_set_link_up        = tap_link_set_up,
1654         .dev_set_link_down      = tap_link_set_down,
1655         .promiscuous_enable     = tap_promisc_enable,
1656         .promiscuous_disable    = tap_promisc_disable,
1657         .allmulticast_enable    = tap_allmulti_enable,
1658         .allmulticast_disable   = tap_allmulti_disable,
1659         .mac_addr_set           = tap_mac_set,
1660         .mtu_set                = tap_mtu_set,
1661         .set_mc_addr_list       = tap_set_mc_addr_list,
1662         .stats_get              = tap_stats_get,
1663         .stats_reset            = tap_stats_reset,
1664         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1665         .rss_hash_update        = tap_rss_hash_update,
1666         .filter_ctrl            = tap_dev_filter_ctrl,
1667 };
1668
1669 static int
1670 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1671                    char *remote_iface, struct ether_addr *mac_addr,
1672                    enum rte_tuntap_type type)
1673 {
1674         int numa_node = rte_socket_id();
1675         struct rte_eth_dev *dev;
1676         struct pmd_internals *pmd;
1677         struct pmd_process_private *process_private;
1678         struct rte_eth_dev_data *data;
1679         struct ifreq ifr;
1680         int i;
1681
1682         TAP_LOG(DEBUG, "%s device on numa %u",
1683                         tuntap_name, rte_socket_id());
1684
1685         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1686         if (!dev) {
1687                 TAP_LOG(ERR, "%s Unable to allocate device struct",
1688                                 tuntap_name);
1689                 goto error_exit_nodev;
1690         }
1691
1692         process_private = (struct pmd_process_private *)
1693                 rte_zmalloc_socket(tap_name, sizeof(struct pmd_process_private),
1694                         RTE_CACHE_LINE_SIZE, dev->device->numa_node);
1695
1696         if (process_private == NULL) {
1697                 TAP_LOG(ERR, "Failed to alloc memory for process private");
1698                 return -1;
1699         }
1700         pmd = dev->data->dev_private;
1701         dev->process_private = process_private;
1702         pmd->dev = dev;
1703         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1704         pmd->type = type;
1705
1706         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1707         if (pmd->ioctl_sock == -1) {
1708                 TAP_LOG(ERR,
1709                         "%s Unable to get a socket for management: %s",
1710                         tuntap_name, strerror(errno));
1711                 goto error_exit;
1712         }
1713
1714         /* Setup some default values */
1715         data = dev->data;
1716         data->dev_private = pmd;
1717         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1718         data->numa_node = numa_node;
1719
1720         data->dev_link = pmd_link;
1721         data->mac_addrs = &pmd->eth_addr;
1722         /* Set the number of RX and TX queues */
1723         data->nb_rx_queues = 0;
1724         data->nb_tx_queues = 0;
1725
1726         dev->dev_ops = &ops;
1727         dev->rx_pkt_burst = pmd_rx_burst;
1728         dev->tx_pkt_burst = pmd_tx_burst;
1729
1730         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1731         pmd->intr_handle.fd = -1;
1732         dev->intr_handle = &pmd->intr_handle;
1733
1734         /* Presetup the fds to -1 as being not valid */
1735         pmd->ka_fd = -1;
1736         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1737                 process_private->rxq_fds[i] = -1;
1738                 process_private->txq_fds[i] = -1;
1739         }
1740
1741         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1742                 if (is_zero_ether_addr(mac_addr))
1743                         eth_random_addr((uint8_t *)&pmd->eth_addr);
1744                 else
1745                         rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1746         }
1747
1748         /*
1749          * Allocate a TUN device keep-alive file descriptor that will only be
1750          * closed when the TUN device itself is closed or removed.
1751          * This keep-alive file descriptor will guarantee that the TUN device
1752          * exists even when all of its queues are closed
1753          */
1754         pmd->ka_fd = tun_alloc(pmd, 1);
1755         if (pmd->ka_fd == -1) {
1756                 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
1757                 goto error_exit;
1758         }
1759
1760         ifr.ifr_mtu = dev->data->mtu;
1761         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1762                 goto error_exit;
1763
1764         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1765                 memset(&ifr, 0, sizeof(struct ifreq));
1766                 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1767                 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1768                                 ETHER_ADDR_LEN);
1769                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1770                         goto error_exit;
1771         }
1772
1773         /*
1774          * Set up everything related to rte_flow:
1775          * - netlink socket
1776          * - tap / remote if_index
1777          * - mandatory QDISCs
1778          * - rte_flow actual/implicit lists
1779          * - implicit rules
1780          */
1781         pmd->nlsk_fd = tap_nl_init(0);
1782         if (pmd->nlsk_fd == -1) {
1783                 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
1784                         pmd->name);
1785                 goto disable_rte_flow;
1786         }
1787         pmd->if_index = if_nametoindex(pmd->name);
1788         if (!pmd->if_index) {
1789                 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
1790                 goto disable_rte_flow;
1791         }
1792         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1793                 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
1794                         pmd->name);
1795                 goto disable_rte_flow;
1796         }
1797         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1798                 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1799                         pmd->name);
1800                 goto disable_rte_flow;
1801         }
1802         LIST_INIT(&pmd->flows);
1803
1804         if (strlen(remote_iface)) {
1805                 pmd->remote_if_index = if_nametoindex(remote_iface);
1806                 if (!pmd->remote_if_index) {
1807                         TAP_LOG(ERR, "%s: failed to get %s if_index.",
1808                                 pmd->name, remote_iface);
1809                         goto error_remote;
1810                 }
1811                 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1812                          "%s", remote_iface);
1813
1814                 /* Save state of remote device */
1815                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1816
1817                 /* Replicate remote MAC address */
1818                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1819                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1820                                 pmd->name, pmd->remote_iface);
1821                         goto error_remote;
1822                 }
1823                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1824                            ETHER_ADDR_LEN);
1825                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1826                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1827                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1828                                 pmd->name, remote_iface);
1829                         goto error_remote;
1830                 }
1831
1832                 /*
1833                  * Flush usually returns negative value because it tries to
1834                  * delete every QDISC (and on a running device, one QDISC at
1835                  * least is needed). Ignore negative return value.
1836                  */
1837                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1838                 if (qdisc_create_ingress(pmd->nlsk_fd,
1839                                          pmd->remote_if_index) < 0) {
1840                         TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1841                                 pmd->remote_iface);
1842                         goto error_remote;
1843                 }
1844                 LIST_INIT(&pmd->implicit_flows);
1845                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1846                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1847                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1848                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1849                         TAP_LOG(ERR,
1850                                 "%s: failed to create implicit rules.",
1851                                 pmd->name);
1852                         goto error_remote;
1853                 }
1854         }
1855
1856         rte_eth_dev_probing_finish(dev);
1857         return 0;
1858
1859 disable_rte_flow:
1860         TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
1861                 strerror(errno), errno);
1862         if (strlen(remote_iface)) {
1863                 TAP_LOG(ERR, "Remote feature requires flow support.");
1864                 goto error_exit;
1865         }
1866         rte_eth_dev_probing_finish(dev);
1867         return 0;
1868
1869 error_remote:
1870         TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
1871                 strerror(errno), errno);
1872         tap_flow_implicit_flush(pmd, NULL);
1873
1874 error_exit:
1875         if (pmd->ioctl_sock > 0)
1876                 close(pmd->ioctl_sock);
1877         /* mac_addrs must not be freed alone because part of dev_private */
1878         dev->data->mac_addrs = NULL;
1879         rte_eth_dev_release_port(dev);
1880
1881 error_exit_nodev:
1882         TAP_LOG(ERR, "%s Unable to initialize %s",
1883                 tuntap_name, rte_vdev_device_name(vdev));
1884
1885         return -EINVAL;
1886 }
1887
1888 /* make sure name is a possible Linux network device name */
1889 static bool
1890 is_valid_iface(const char *name)
1891 {
1892         if (*name == '\0')
1893                 return false;
1894
1895         if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1896                 return false;
1897
1898         while (*name) {
1899                 if (*name == '/' || *name == ':' || isspace(*name))
1900                         return false;
1901                 name++;
1902         }
1903         return true;
1904 }
1905
1906 static int
1907 set_interface_name(const char *key __rte_unused,
1908                    const char *value,
1909                    void *extra_args)
1910 {
1911         char *name = (char *)extra_args;
1912
1913         if (value) {
1914                 if (!is_valid_iface(value)) {
1915                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
1916                                 value);
1917                         return -1;
1918                 }
1919                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
1920         } else {
1921                 snprintf(name, RTE_ETH_NAME_MAX_LEN, "%s%d",
1922                          DEFAULT_TAP_NAME, tap_unit - 1);
1923         }
1924         return 0;
1925 }
1926
1927 static int
1928 set_remote_iface(const char *key __rte_unused,
1929                  const char *value,
1930                  void *extra_args)
1931 {
1932         char *name = (char *)extra_args;
1933
1934         if (value) {
1935                 if (!is_valid_iface(value)) {
1936                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
1937                                 value);
1938                         return -1;
1939                 }
1940                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
1941         }
1942
1943         return 0;
1944 }
1945
1946 static int parse_user_mac(struct ether_addr *user_mac,
1947                 const char *value)
1948 {
1949         unsigned int index = 0;
1950         char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
1951
1952         if (user_mac == NULL || value == NULL)
1953                 return 0;
1954
1955         strlcpy(mac_temp, value, sizeof(mac_temp));
1956         mac_byte = strtok(mac_temp, ":");
1957
1958         while ((mac_byte != NULL) &&
1959                         (strlen(mac_byte) <= 2) &&
1960                         (strlen(mac_byte) == strspn(mac_byte,
1961                                         ETH_TAP_CMP_MAC_FMT))) {
1962                 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
1963                 mac_byte = strtok(NULL, ":");
1964         }
1965
1966         return index;
1967 }
1968
1969 static int
1970 set_mac_type(const char *key __rte_unused,
1971              const char *value,
1972              void *extra_args)
1973 {
1974         struct ether_addr *user_mac = extra_args;
1975
1976         if (!value)
1977                 return 0;
1978
1979         if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
1980                 static int iface_idx;
1981
1982                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1983                 memcpy((char *)user_mac->addr_bytes, "\0dtap", ETHER_ADDR_LEN);
1984                 user_mac->addr_bytes[ETHER_ADDR_LEN - 1] = iface_idx++ + '0';
1985                 goto success;
1986         }
1987
1988         if (parse_user_mac(user_mac, value) != 6)
1989                 goto error;
1990 success:
1991         TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
1992         return 0;
1993
1994 error:
1995         TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
1996                 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
1997         return -1;
1998 }
1999
2000 /*
2001  * Open a TUN interface device. TUN PMD
2002  * 1) sets tap_type as false
2003  * 2) intakes iface as argument.
2004  * 3) as interface is virtual set speed to 10G
2005  */
2006 static int
2007 rte_pmd_tun_probe(struct rte_vdev_device *dev)
2008 {
2009         const char *name, *params;
2010         int ret;
2011         struct rte_kvargs *kvlist = NULL;
2012         char tun_name[RTE_ETH_NAME_MAX_LEN];
2013         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2014         struct rte_eth_dev *eth_dev;
2015
2016         strcpy(tuntap_name, "TUN");
2017
2018         name = rte_vdev_device_name(dev);
2019         params = rte_vdev_device_args(dev);
2020         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2021
2022         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
2023             strlen(params) == 0) {
2024                 eth_dev = rte_eth_dev_attach_secondary(name);
2025                 if (!eth_dev) {
2026                         TAP_LOG(ERR, "Failed to probe %s", name);
2027                         return -1;
2028                 }
2029                 eth_dev->dev_ops = &ops;
2030                 eth_dev->device = &dev->device;
2031                 rte_eth_dev_probing_finish(eth_dev);
2032                 return 0;
2033         }
2034
2035         snprintf(tun_name, sizeof(tun_name), "%s%u",
2036                  DEFAULT_TUN_NAME, tun_unit++);
2037
2038         if (params && (params[0] != '\0')) {
2039                 TAP_LOG(DEBUG, "parameters (%s)", params);
2040
2041                 kvlist = rte_kvargs_parse(params, valid_arguments);
2042                 if (kvlist) {
2043                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2044                                 ret = rte_kvargs_process(kvlist,
2045                                         ETH_TAP_IFACE_ARG,
2046                                         &set_interface_name,
2047                                         tun_name);
2048
2049                                 if (ret == -1)
2050                                         goto leave;
2051                         }
2052                 }
2053         }
2054         pmd_link.link_speed = ETH_SPEED_NUM_10G;
2055
2056         TAP_LOG(NOTICE, "Initializing pmd_tun for %s as %s",
2057                 name, tun_name);
2058
2059         ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0,
2060                 ETH_TUNTAP_TYPE_TUN);
2061
2062 leave:
2063         if (ret == -1) {
2064                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2065                         name, tun_name);
2066                 tun_unit--; /* Restore the unit number */
2067         }
2068         rte_kvargs_free(kvlist);
2069
2070         return ret;
2071 }
2072
2073 /* Request queue file descriptors from secondary to primary. */
2074 static int
2075 tap_mp_attach_queues(const char *port_name, struct rte_eth_dev *dev)
2076 {
2077         int ret;
2078         struct timespec timeout = {.tv_sec = 1, .tv_nsec = 0};
2079         struct rte_mp_msg request, *reply;
2080         struct rte_mp_reply replies;
2081         struct ipc_queues *request_param = (struct ipc_queues *)request.param;
2082         struct ipc_queues *reply_param;
2083         struct pmd_process_private *process_private = dev->process_private;
2084         int queue, fd_iterator;
2085
2086         /* Prepare the request */
2087         strlcpy(request.name, TAP_MP_KEY, sizeof(request.name));
2088         strlcpy(request_param->port_name, port_name,
2089                 sizeof(request_param->port_name));
2090         request.len_param = sizeof(*request_param);
2091         /* Send request and receive reply */
2092         ret = rte_mp_request_sync(&request, &replies, &timeout);
2093         if (ret < 0) {
2094                 TAP_LOG(ERR, "Failed to request queues from primary: %d",
2095                         rte_errno);
2096                 return -1;
2097         }
2098         reply = &replies.msgs[0];
2099         reply_param = (struct ipc_queues *)reply->param;
2100         TAP_LOG(DEBUG, "Received IPC reply for %s", reply_param->port_name);
2101
2102         /* Attach the queues from received file descriptors */
2103         dev->data->nb_rx_queues = reply_param->rxq_count;
2104         dev->data->nb_tx_queues = reply_param->txq_count;
2105         fd_iterator = 0;
2106         for (queue = 0; queue < reply_param->rxq_count; queue++)
2107                 process_private->rxq_fds[queue] = reply->fds[fd_iterator++];
2108         for (queue = 0; queue < reply_param->txq_count; queue++)
2109                 process_private->txq_fds[queue] = reply->fds[fd_iterator++];
2110
2111         return 0;
2112 }
2113
2114 /* Send the queue file descriptors from the primary process to secondary. */
2115 static int
2116 tap_mp_sync_queues(const struct rte_mp_msg *request, const void *peer)
2117 {
2118         struct rte_eth_dev *dev;
2119         struct pmd_process_private *process_private;
2120         struct rte_mp_msg reply;
2121         const struct ipc_queues *request_param =
2122                 (const struct ipc_queues *)request->param;
2123         struct ipc_queues *reply_param =
2124                 (struct ipc_queues *)reply.param;
2125         uint16_t port_id;
2126         int queue;
2127         int ret;
2128
2129         /* Get requested port */
2130         TAP_LOG(DEBUG, "Received IPC request for %s", request_param->port_name);
2131         ret = rte_eth_dev_get_port_by_name(request_param->port_name, &port_id);
2132         if (ret) {
2133                 TAP_LOG(ERR, "Failed to get port id for %s",
2134                         request_param->port_name);
2135                 return -1;
2136         }
2137         dev = &rte_eth_devices[port_id];
2138         process_private = dev->process_private;
2139
2140         /* Fill file descriptors for all queues */
2141         reply.num_fds = 0;
2142         reply_param->rxq_count = 0;
2143         for (queue = 0; queue < dev->data->nb_rx_queues; queue++) {
2144                 reply.fds[reply.num_fds++] = process_private->rxq_fds[queue];
2145                 reply_param->rxq_count++;
2146         }
2147         RTE_ASSERT(reply_param->rxq_count == dev->data->nb_rx_queues);
2148         RTE_ASSERT(reply_param->txq_count == dev->data->nb_tx_queues);
2149         RTE_ASSERT(reply.num_fds <= RTE_MP_MAX_FD_NUM);
2150
2151         reply_param->txq_count = 0;
2152         for (queue = 0; queue < dev->data->nb_tx_queues; queue++) {
2153                 reply.fds[reply.num_fds++] = process_private->txq_fds[queue];
2154                 reply_param->txq_count++;
2155         }
2156
2157         /* Send reply */
2158         strlcpy(reply.name, request->name, sizeof(reply.name));
2159         strlcpy(reply_param->port_name, request_param->port_name,
2160                 sizeof(reply_param->port_name));
2161         reply.len_param = sizeof(*reply_param);
2162         if (rte_mp_reply(&reply, peer) < 0) {
2163                 TAP_LOG(ERR, "Failed to reply an IPC request to sync queues");
2164                 return -1;
2165         }
2166         return 0;
2167 }
2168
2169 /* Open a TAP interface device.
2170  */
2171 static int
2172 rte_pmd_tap_probe(struct rte_vdev_device *dev)
2173 {
2174         const char *name, *params;
2175         int ret;
2176         struct rte_kvargs *kvlist = NULL;
2177         int speed;
2178         char tap_name[RTE_ETH_NAME_MAX_LEN];
2179         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2180         struct ether_addr user_mac = { .addr_bytes = {0} };
2181         struct rte_eth_dev *eth_dev;
2182         int tap_devices_count_increased = 0;
2183
2184         strcpy(tuntap_name, "TAP");
2185
2186         name = rte_vdev_device_name(dev);
2187         params = rte_vdev_device_args(dev);
2188
2189         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
2190                 eth_dev = rte_eth_dev_attach_secondary(name);
2191                 if (!eth_dev) {
2192                         TAP_LOG(ERR, "Failed to probe %s", name);
2193                         return -1;
2194                 }
2195                 eth_dev->dev_ops = &ops;
2196                 eth_dev->device = &dev->device;
2197                 eth_dev->rx_pkt_burst = pmd_rx_burst;
2198                 eth_dev->tx_pkt_burst = pmd_tx_burst;
2199                 if (!rte_eal_primary_proc_alive(NULL)) {
2200                         TAP_LOG(ERR, "Primary process is missing");
2201                         return -1;
2202                 }
2203                 eth_dev->process_private = (struct pmd_process_private *)
2204                         rte_zmalloc_socket(name,
2205                                 sizeof(struct pmd_process_private),
2206                                 RTE_CACHE_LINE_SIZE,
2207                                 eth_dev->device->numa_node);
2208                 if (eth_dev->process_private == NULL) {
2209                         TAP_LOG(ERR,
2210                                 "Failed to alloc memory for process private");
2211                         return -1;
2212                 }
2213
2214                 ret = tap_mp_attach_queues(name, eth_dev);
2215                 if (ret != 0)
2216                         return -1;
2217                 rte_eth_dev_probing_finish(eth_dev);
2218                 return 0;
2219         }
2220
2221         speed = ETH_SPEED_NUM_10G;
2222         snprintf(tap_name, sizeof(tap_name), "%s%u",
2223                  DEFAULT_TAP_NAME, tap_unit++);
2224         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2225
2226         if (params && (params[0] != '\0')) {
2227                 TAP_LOG(DEBUG, "parameters (%s)", params);
2228
2229                 kvlist = rte_kvargs_parse(params, valid_arguments);
2230                 if (kvlist) {
2231                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2232                                 ret = rte_kvargs_process(kvlist,
2233                                                          ETH_TAP_IFACE_ARG,
2234                                                          &set_interface_name,
2235                                                          tap_name);
2236                                 if (ret == -1)
2237                                         goto leave;
2238                         }
2239
2240                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
2241                                 ret = rte_kvargs_process(kvlist,
2242                                                          ETH_TAP_REMOTE_ARG,
2243                                                          &set_remote_iface,
2244                                                          remote_iface);
2245                                 if (ret == -1)
2246                                         goto leave;
2247                         }
2248
2249                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
2250                                 ret = rte_kvargs_process(kvlist,
2251                                                          ETH_TAP_MAC_ARG,
2252                                                          &set_mac_type,
2253                                                          &user_mac);
2254                                 if (ret == -1)
2255                                         goto leave;
2256                         }
2257                 }
2258         }
2259         pmd_link.link_speed = speed;
2260
2261         TAP_LOG(NOTICE, "Initializing pmd_tap for %s as %s",
2262                 name, tap_name);
2263
2264         /* Register IPC feed callback */
2265         if (!tap_devices_count) {
2266                 ret = rte_mp_action_register(TAP_MP_KEY, tap_mp_sync_queues);
2267                 if (ret < 0) {
2268                         TAP_LOG(ERR, "%s: Failed to register IPC callback: %s",
2269                                 tuntap_name, strerror(rte_errno));
2270                         goto leave;
2271                 }
2272         }
2273         tap_devices_count++;
2274         tap_devices_count_increased = 1;
2275         ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
2276                 ETH_TUNTAP_TYPE_TAP);
2277
2278 leave:
2279         if (ret == -1) {
2280                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2281                         name, tap_name);
2282                 if (tap_devices_count_increased == 1) {
2283                         if (tap_devices_count == 1)
2284                                 rte_mp_action_unregister(TAP_MP_KEY);
2285                         tap_devices_count--;
2286                 }
2287                 tap_unit--;             /* Restore the unit number */
2288         }
2289         rte_kvargs_free(kvlist);
2290
2291         return ret;
2292 }
2293
2294 /* detach a TUNTAP device.
2295  */
2296 static int
2297 rte_pmd_tap_remove(struct rte_vdev_device *dev)
2298 {
2299         struct rte_eth_dev *eth_dev = NULL;
2300         struct pmd_internals *internals;
2301         struct pmd_process_private *process_private;
2302         int i;
2303
2304         /* find the ethdev entry */
2305         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
2306         if (!eth_dev)
2307                 return -ENODEV;
2308
2309         /* mac_addrs must not be freed alone because part of dev_private */
2310         eth_dev->data->mac_addrs = NULL;
2311
2312         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2313                 return rte_eth_dev_release_port(eth_dev);
2314
2315         internals = eth_dev->data->dev_private;
2316         process_private = eth_dev->process_private;
2317
2318         TAP_LOG(DEBUG, "Closing %s Ethernet device on numa %u",
2319                 (internals->type == ETH_TUNTAP_TYPE_TAP) ? "TAP" : "TUN",
2320                 rte_socket_id());
2321
2322         if (internals->nlsk_fd) {
2323                 tap_flow_flush(eth_dev, NULL);
2324                 tap_flow_implicit_flush(internals, NULL);
2325                 tap_nl_final(internals->nlsk_fd);
2326         }
2327         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
2328                 if (process_private->rxq_fds[i] != -1) {
2329                         close(process_private->rxq_fds[i]);
2330                         process_private->rxq_fds[i] = -1;
2331                 }
2332                 if (process_private->txq_fds[i] != -1) {
2333                         close(process_private->txq_fds[i]);
2334                         process_private->txq_fds[i] = -1;
2335                 }
2336         }
2337
2338         close(internals->ioctl_sock);
2339         rte_free(eth_dev->process_private);
2340         if (tap_devices_count == 1)
2341                 rte_mp_action_unregister(TAP_MP_KEY);
2342         tap_devices_count--;
2343         rte_eth_dev_release_port(eth_dev);
2344
2345         if (internals->ka_fd != -1) {
2346                 close(internals->ka_fd);
2347                 internals->ka_fd = -1;
2348         }
2349         return 0;
2350 }
2351
2352 static struct rte_vdev_driver pmd_tun_drv = {
2353         .probe = rte_pmd_tun_probe,
2354         .remove = rte_pmd_tap_remove,
2355 };
2356
2357 static struct rte_vdev_driver pmd_tap_drv = {
2358         .probe = rte_pmd_tap_probe,
2359         .remove = rte_pmd_tap_remove,
2360 };
2361
2362 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
2363 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
2364 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
2365 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
2366                               ETH_TAP_IFACE_ARG "=<string> ");
2367 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
2368                               ETH_TAP_IFACE_ARG "=<string> "
2369                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
2370                               ETH_TAP_REMOTE_ARG "=<string>");
2371 int tap_logtype;
2372
2373 RTE_INIT(tap_init_log)
2374 {
2375         tap_logtype = rte_log_register("pmd.net.tap");
2376         if (tap_logtype >= 0)
2377                 rte_log_set_level(tap_logtype, RTE_LOG_NOTICE);
2378 }