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 #define TAP_IOV_DEFAULT_MAX 1024
  72
  73 static int tap_devices_count;
  74 static struct rte_vdev_driver pmd_tap_drv;
  75 static struct rte_vdev_driver pmd_tun_drv;
  76
  77 static const char *valid_arguments[] = {
  78         ETH_TAP_IFACE_ARG,
  79         ETH_TAP_REMOTE_ARG,
  80         ETH_TAP_MAC_ARG,
  81         NULL
  82 };
  83
  84 static volatile uint32_t tap_trigger;   /* Rx trigger */
  85
  86 static struct rte_eth_link pmd_link = {
  87         .link_speed = ETH_SPEED_NUM_10G,
  88         .link_duplex = ETH_LINK_FULL_DUPLEX,
  89         .link_status = ETH_LINK_DOWN,
  90         .link_autoneg = ETH_LINK_FIXED,
  91 };
  92
  93 static void
  94 tap_trigger_cb(int sig __rte_unused)
  95 {
  96         /* Valid trigger values are nonzero */
  97         tap_trigger = (tap_trigger + 1) | 0x80000000;
  98 }
  99
 100 /* Specifies on what netdevices the ioctl should be applied */
 101 enum ioctl_mode {
 102         LOCAL_AND_REMOTE,
 103         LOCAL_ONLY,
 104         REMOTE_ONLY,
 105 };
 106
 107 /* Message header to synchronize queues via IPC */
 108 struct ipc_queues {
 109         char port_name[RTE_DEV_NAME_MAX_LEN];
 110         int rxq_count;
 111         int txq_count;
 112         /*
 113          * The file descriptors are in the dedicated part
 114          * of the Unix message to be translated by the kernel.
 115          */
 116 };
 117
 118 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
 119
 120 /**
 121  * Tun/Tap allocation routine
 122  *
 123  * @param[in] pmd
 124  *   Pointer to private structure.
 125  *
 126  * @param[in] is_keepalive
 127  *   Keepalive flag
 128  *
 129  * @return
 130  *   -1 on failure, fd on success
 131  */
 132 static int
 133 tun_alloc(struct pmd_internals *pmd, int is_keepalive)
 134 {
 135         struct ifreq ifr;
 136 #ifdef IFF_MULTI_QUEUE
 137         unsigned int features;
 138 #endif
 139         int fd;
 140
 141         memset(&ifr, 0, sizeof(struct ifreq));
 142
 143         /*
 144          * Do not set IFF_NO_PI as packet information header will be needed
 145          * to check if a received packet has been truncated.
 146          */
 147         ifr.ifr_flags = (pmd->type == ETH_TUNTAP_TYPE_TAP) ?
 148                 IFF_TAP : IFF_TUN | IFF_POINTOPOINT;
 149         strlcpy(ifr.ifr_name, pmd->name, IFNAMSIZ);
 150
 151         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 152         if (fd < 0) {
 153                 TAP_LOG(ERR, "Unable to open %s interface", TUN_TAP_DEV_PATH);
 154                 goto error;
 155         }
 156
 157 #ifdef IFF_MULTI_QUEUE
 158         /* Grab the TUN features to verify we can work multi-queue */
 159         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 160                 TAP_LOG(ERR, "unable to get TUN/TAP features");
 161                 goto error;
 162         }
 163         TAP_LOG(DEBUG, "%s Features %08x", TUN_TAP_DEV_PATH, features);
 164
 165         if (features & IFF_MULTI_QUEUE) {
 166                 TAP_LOG(DEBUG, "  Multi-queue support for %d queues",
 167                         RTE_PMD_TAP_MAX_QUEUES);
 168                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 169         } else
 170 #endif
 171         {
 172                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 173                 TAP_LOG(DEBUG, "  Single queue only support");
 174         }
 175
 176         /* Set the TUN/TAP configuration and set the name if needed */
 177         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 178                 TAP_LOG(WARNING, "Unable to set TUNSETIFF for %s: %s",
 179                         ifr.ifr_name, strerror(errno));
 180                 goto error;
 181         }
 182
 183         /*
 184          * Name passed to kernel might be wildcard like dtun%d
 185          * and need to find the resulting device.
 186          */
 187         TAP_LOG(DEBUG, "Device name is '%s'", ifr.ifr_name);
 188         strlcpy(pmd->name, ifr.ifr_name, RTE_ETH_NAME_MAX_LEN);
 189
 190         if (is_keepalive) {
 191                 /*
 192                  * Detach the TUN/TAP keep-alive queue
 193                  * to avoid traffic through it
 194                  */
 195                 ifr.ifr_flags = IFF_DETACH_QUEUE;
 196                 if (ioctl(fd, TUNSETQUEUE, (void *)&ifr) < 0) {
 197                         TAP_LOG(WARNING,
 198                                 "Unable to detach keep-alive queue for %s: %s",
 199                                 ifr.ifr_name, strerror(errno));
 200                         goto error;
 201                 }
 202         }
 203
 204         /* Always set the file descriptor to non-blocking */
 205         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 206                 TAP_LOG(WARNING,
 207                         "Unable to set %s to nonblocking: %s",
 208                         ifr.ifr_name, strerror(errno));
 209                 goto error;
 210         }
 211
 212         /* Set up trigger to optimize empty Rx bursts */
 213         errno = 0;
 214         do {
 215                 struct sigaction sa;
 216                 int flags = fcntl(fd, F_GETFL);
 217
 218                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 219                         break;
 220                 if (sa.sa_handler != tap_trigger_cb) {
 221                         /*
 222                          * Make sure SIGIO is not already taken. This is done
 223                          * as late as possible to leave the application a
 224                          * chance to set up its own signal handler first.
 225                          */
 226                         if (sa.sa_handler != SIG_IGN &&
 227                             sa.sa_handler != SIG_DFL) {
 228                                 errno = EBUSY;
 229                                 break;
 230                         }
 231                         sa = (struct sigaction){
 232                                 .sa_flags = SA_RESTART,
 233                                 .sa_handler = tap_trigger_cb,
 234                         };
 235                         if (sigaction(SIGIO, &sa, NULL) == -1)
 236                                 break;
 237                 }
 238                 /* Enable SIGIO on file descriptor */
 239                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 240                 fcntl(fd, F_SETOWN, getpid());
 241         } while (0);
 242
 243         if (errno) {
 244                 /* Disable trigger globally in case of error */
 245                 tap_trigger = 0;
 246                 TAP_LOG(WARNING, "Rx trigger disabled: %s",
 247                         strerror(errno));
 248         }
 249
 250         return fd;
 251
 252 error:
 253         if (fd >= 0)
 254                 close(fd);
 255         return -1;
 256 }
 257
 258 static void
 259 tap_verify_csum(struct rte_mbuf *mbuf)
 260 {
 261         uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
 262         uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
 263         uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
 264         unsigned int l2_len = sizeof(struct rte_ether_hdr);
 265         unsigned int l3_len;
 266         uint16_t cksum = 0;
 267         void *l3_hdr;
 268         void *l4_hdr;
 269
 270         if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
 271                 l2_len += 4;
 272         else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
 273                 l2_len += 8;
 274         /* Don't verify checksum for packets with discontinuous L2 header */
 275         if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
 276                      rte_pktmbuf_data_len(mbuf)))
 277                 return;
 278         l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
 279         if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 280                 struct ipv4_hdr *iph = l3_hdr;
 281
 282                 /* ihl contains the number of 4-byte words in the header */
 283                 l3_len = 4 * (iph->version_ihl & 0xf);
 284                 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
 285                         return;
 286                 /* check that the total length reported by header is not
 287                  * greater than the total received size
 288                  */
 289                 if (l2_len + rte_be_to_cpu_16(iph->total_length) >
 290                                 rte_pktmbuf_data_len(mbuf))
 291                         return;
 292
 293                 cksum = ~rte_raw_cksum(iph, l3_len);
 294                 mbuf->ol_flags |= cksum ?
 295                         PKT_RX_IP_CKSUM_BAD :
 296                         PKT_RX_IP_CKSUM_GOOD;
 297         } else if (l3 == RTE_PTYPE_L3_IPV6) {
 298                 struct ipv6_hdr *iph = l3_hdr;
 299
 300                 l3_len = sizeof(struct ipv6_hdr);
 301                 /* check that the total length reported by header is not
 302                  * greater than the total received size
 303                  */
 304                 if (l2_len + l3_len + rte_be_to_cpu_16(iph->payload_len) >
 305                                 rte_pktmbuf_data_len(mbuf))
 306                         return;
 307         } else {
 308                 /* IPv6 extensions are not supported */
 309                 return;
 310         }
 311         if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
 312                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 313                 /* Don't verify checksum for multi-segment packets. */
 314                 if (mbuf->nb_segs > 1)
 315                         return;
 316                 if (l3 == RTE_PTYPE_L3_IPV4)
 317                         cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
 318                 else if (l3 == RTE_PTYPE_L3_IPV6)
 319                         cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
 320                 mbuf->ol_flags |= cksum ?
 321                         PKT_RX_L4_CKSUM_BAD :
 322                         PKT_RX_L4_CKSUM_GOOD;
 323         }
 324 }
 325
 326 static uint64_t
 327 tap_rx_offload_get_port_capa(void)
 328 {
 329         /*
 330          * No specific port Rx offload capabilities.
 331          */
 332         return 0;
 333 }
 334
 335 static uint64_t
 336 tap_rx_offload_get_queue_capa(void)
 337 {
 338         return DEV_RX_OFFLOAD_SCATTER |
 339                DEV_RX_OFFLOAD_IPV4_CKSUM |
 340                DEV_RX_OFFLOAD_UDP_CKSUM |
 341                DEV_RX_OFFLOAD_TCP_CKSUM;
 342 }
 343
 344 /* Callback to handle the rx burst of packets to the correct interface and
 345  * file descriptor(s) in a multi-queue setup.
 346  */
 347 static uint16_t
 348 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 349 {
 350         struct rx_queue *rxq = queue;
 351         struct pmd_process_private *process_private;
 352         uint16_t num_rx;
 353         unsigned long num_rx_bytes = 0;
 354         uint32_t trigger = tap_trigger;
 355
 356         if (trigger == rxq->trigger_seen)
 357                 return 0;
 358         if (trigger)
 359                 rxq->trigger_seen = trigger;
 360         process_private = rte_eth_devices[rxq->in_port].process_private;
 361         rte_compiler_barrier();
 362         for (num_rx = 0; num_rx < nb_pkts; ) {
 363                 struct rte_mbuf *mbuf = rxq->pool;
 364                 struct rte_mbuf *seg = NULL;
 365                 struct rte_mbuf *new_tail = NULL;
 366                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 367                 int len;
 368
 369                 len = readv(process_private->rxq_fds[rxq->queue_id],
 370                         *rxq->iovecs,
 371                         1 + (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
 372                              rxq->nb_rx_desc : 1));
 373                 if (len < (int)sizeof(struct tun_pi))
 374                         break;
 375
 376                 /* Packet couldn't fit in the provided mbuf */
 377                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 378                         rxq->stats.ierrors++;
 379                         continue;
 380                 }
 381
 382                 len -= sizeof(struct tun_pi);
 383
 384                 mbuf->pkt_len = len;
 385                 mbuf->port = rxq->in_port;
 386                 while (1) {
 387                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 388
 389                         if (unlikely(!buf)) {
 390                                 rxq->stats.rx_nombuf++;
 391                                 /* No new buf has been allocated: do nothing */
 392                                 if (!new_tail || !seg)
 393                                         goto end;
 394
 395                                 seg->next = NULL;
 396                                 rte_pktmbuf_free(mbuf);
 397
 398                                 goto end;
 399                         }
 400                         seg = seg ? seg->next : mbuf;
 401                         if (rxq->pool == mbuf)
 402                                 rxq->pool = buf;
 403                         if (new_tail)
 404                                 new_tail->next = buf;
 405                         new_tail = buf;
 406                         new_tail->next = seg->next;
 407
 408                         /* iovecs[0] is reserved for packet info (pi) */
 409                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 410                                 buf->buf_len - data_off;
 411                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 412                                 (char *)buf->buf_addr + data_off;
 413
 414                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 415                         seg->data_off = data_off;
 416
 417                         len -= seg->data_len;
 418                         if (len <= 0)
 419                                 break;
 420                         mbuf->nb_segs++;
 421                         /* First segment has headroom, not the others */
 422                         data_off = 0;
 423                 }
 424                 seg->next = NULL;
 425                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 426                                                       RTE_PTYPE_ALL_MASK);
 427                 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
 428                         tap_verify_csum(mbuf);
 429
 430                 /* account for the receive frame */
 431                 bufs[num_rx++] = mbuf;
 432                 num_rx_bytes += mbuf->pkt_len;
 433         }
 434 end:
 435         rxq->stats.ipackets += num_rx;
 436         rxq->stats.ibytes += num_rx_bytes;
 437
 438         return num_rx;
 439 }
 440
 441 static uint64_t
 442 tap_tx_offload_get_port_capa(void)
 443 {
 444         /*
 445          * No specific port Tx offload capabilities.
 446          */
 447         return 0;
 448 }
 449
 450 static uint64_t
 451 tap_tx_offload_get_queue_capa(void)
 452 {
 453         return DEV_TX_OFFLOAD_MULTI_SEGS |
 454                DEV_TX_OFFLOAD_IPV4_CKSUM |
 455                DEV_TX_OFFLOAD_UDP_CKSUM |
 456                DEV_TX_OFFLOAD_TCP_CKSUM |
 457                DEV_TX_OFFLOAD_TCP_TSO;
 458 }
 459
 460 /* Finalize l4 checksum calculation */
 461 static void
 462 tap_tx_l4_cksum(uint16_t *l4_cksum, uint16_t l4_phdr_cksum,
 463                 uint32_t l4_raw_cksum)
 464 {
 465         if (l4_cksum) {
 466                 uint32_t cksum;
 467
 468                 cksum = __rte_raw_cksum_reduce(l4_raw_cksum);
 469                 cksum += l4_phdr_cksum;
 470
 471                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 472                 cksum = (~cksum) & 0xffff;
 473                 if (cksum == 0)
 474                         cksum = 0xffff;
 475                 *l4_cksum = cksum;
 476         }
 477 }
 478
 479 /* Accumaulate L4 raw checksums */
 480 static void
 481 tap_tx_l4_add_rcksum(char *l4_data, unsigned int l4_len, uint16_t *l4_cksum,
 482                         uint32_t *l4_raw_cksum)
 483 {
 484         if (l4_cksum == NULL)
 485                 return;
 486
 487         *l4_raw_cksum = __rte_raw_cksum(l4_data, l4_len, *l4_raw_cksum);
 488 }
 489
 490 /* L3 and L4 pseudo headers checksum offloads */
 491 static void
 492 tap_tx_l3_cksum(char *packet, uint64_t ol_flags, unsigned int l2_len,
 493                 unsigned int l3_len, unsigned int l4_len, uint16_t **l4_cksum,
 494                 uint16_t *l4_phdr_cksum, uint32_t *l4_raw_cksum)
 495 {
 496         void *l3_hdr = packet + l2_len;
 497
 498         if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
 499                 struct ipv4_hdr *iph = l3_hdr;
 500                 uint16_t cksum;
 501
 502                 iph->hdr_checksum = 0;
 503                 cksum = rte_raw_cksum(iph, l3_len);
 504                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 505         }
 506         if (ol_flags & PKT_TX_L4_MASK) {
 507                 void *l4_hdr;
 508
 509                 l4_hdr = packet + l2_len + l3_len;
 510                 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
 511                         *l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
 512                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 513                         *l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
 514                 else
 515                         return;
 516                 **l4_cksum = 0;
 517                 if (ol_flags & PKT_TX_IPV4)
 518                         *l4_phdr_cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 519                 else
 520                         *l4_phdr_cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 521                 *l4_raw_cksum = __rte_raw_cksum(l4_hdr, l4_len, 0);
 522         }
 523 }
 524
 525 static inline void
 526 tap_write_mbufs(struct tx_queue *txq, uint16_t num_mbufs,
 527                         struct rte_mbuf **pmbufs,
 528                         uint16_t *num_packets, unsigned long *num_tx_bytes)
 529 {
 530         int i;
 531         uint16_t l234_hlen;
 532         struct pmd_process_private *process_private;
 533
 534         process_private = rte_eth_devices[txq->out_port].process_private;
 535
 536         for (i = 0; i < num_mbufs; i++) {
 537                 struct rte_mbuf *mbuf = pmbufs[i];
 538                 struct iovec iovecs[mbuf->nb_segs + 2];
 539                 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
 540                 struct rte_mbuf *seg = mbuf;
 541                 char m_copy[mbuf->data_len];
 542                 int proto;
 543                 int n;
 544                 int j;
 545                 int k; /* current index in iovecs for copying segments */
 546                 uint16_t seg_len; /* length of first segment */
 547                 uint16_t nb_segs;
 548                 uint16_t *l4_cksum; /* l4 checksum (pseudo header + payload) */
 549                 uint32_t l4_raw_cksum = 0; /* TCP/UDP payload raw checksum */
 550                 uint16_t l4_phdr_cksum = 0; /* TCP/UDP pseudo header checksum */
 551                 uint16_t is_cksum = 0; /* in case cksum should be offloaded */
 552
 553                 l4_cksum = NULL;
 554                 if (txq->type == ETH_TUNTAP_TYPE_TUN) {
 555                         /*
 556                          * TUN and TAP are created with IFF_NO_PI disabled.
 557                          * For TUN PMD this mandatory as fields are used by
 558                          * Kernel tun.c to determine whether its IP or non IP
 559                          * packets.
 560                          *
 561                          * The logic fetches the first byte of data from mbuf
 562                          * then compares whether its v4 or v6. If first byte
 563                          * is 4 or 6, then protocol field is updated.
 564                          */
 565                         char *buff_data = rte_pktmbuf_mtod(seg, void *);
 566                         proto = (*buff_data & 0xf0);
 567                         pi.proto = (proto == 0x40) ?
 568                                 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPv4) :
 569                                 ((proto == 0x60) ?
 570                                         rte_cpu_to_be_16(RTE_ETHER_TYPE_IPv6) :
 571                                         0x00);
 572                 }
 573
 574                 k = 0;
 575                 iovecs[k].iov_base = &pi;
 576                 iovecs[k].iov_len = sizeof(pi);
 577                 k++;
 578
 579                 nb_segs = mbuf->nb_segs;
 580                 if (txq->csum &&
 581                     ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 582                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 583                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
 584                         is_cksum = 1;
 585
 586                         /* Support only packets with at least layer 4
 587                          * header included in the first segment
 588                          */
 589                         seg_len = rte_pktmbuf_data_len(mbuf);
 590                         l234_hlen = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
 591                         if (seg_len < l234_hlen)
 592                                 break;
 593
 594                         /* To change checksums, work on a * copy of l2, l3
 595                          * headers + l4 pseudo header
 596                          */
 597                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 598                                         l234_hlen);
 599                         tap_tx_l3_cksum(m_copy, mbuf->ol_flags,
 600                                        mbuf->l2_len, mbuf->l3_len, mbuf->l4_len,
 601                                        &l4_cksum, &l4_phdr_cksum,
 602                                        &l4_raw_cksum);
 603                         iovecs[k].iov_base = m_copy;
 604                         iovecs[k].iov_len = l234_hlen;
 605                         k++;
 606
 607                         /* Update next iovecs[] beyond l2, l3, l4 headers */
 608                         if (seg_len > l234_hlen) {
 609                                 iovecs[k].iov_len = seg_len - l234_hlen;
 610                                 iovecs[k].iov_base =
 611                                         rte_pktmbuf_mtod(seg, char *) +
 612                                                 l234_hlen;
 613                                 tap_tx_l4_add_rcksum(iovecs[k].iov_base,
 614                                         iovecs[k].iov_len, l4_cksum,
 615                                         &l4_raw_cksum);
 616                                 k++;
 617                                 nb_segs++;
 618                         }
 619                         seg = seg->next;
 620                 }
 621
 622                 for (j = k; j <= nb_segs; j++) {
 623                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 624                         iovecs[j].iov_base = rte_pktmbuf_mtod(seg, void *);
 625                         if (is_cksum)
 626                                 tap_tx_l4_add_rcksum(iovecs[j].iov_base,
 627                                         iovecs[j].iov_len, l4_cksum,
 628                                         &l4_raw_cksum);
 629                         seg = seg->next;
 630                 }
 631
 632                 if (is_cksum)
 633                         tap_tx_l4_cksum(l4_cksum, l4_phdr_cksum, l4_raw_cksum);
 634
 635                 /* copy the tx frame data */
 636                 n = writev(process_private->txq_fds[txq->queue_id], iovecs, j);
 637                 if (n <= 0)
 638                         break;
 639                 (*num_packets)++;
 640                 (*num_tx_bytes) += rte_pktmbuf_pkt_len(mbuf);
 641         }
 642 }
 643
 644 /* Callback to handle sending packets from the tap interface
 645  */
 646 static uint16_t
 647 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 648 {
 649         struct tx_queue *txq = queue;
 650         uint16_t num_tx = 0;
 651         uint16_t num_packets = 0;
 652         unsigned long num_tx_bytes = 0;
 653         uint32_t max_size;
 654         int i;
 655
 656         if (unlikely(nb_pkts == 0))
 657                 return 0;
 658
 659         struct rte_mbuf *gso_mbufs[MAX_GSO_MBUFS];
 660         max_size = *txq->mtu + (RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN + 4);
 661         for (i = 0; i < nb_pkts; i++) {
 662                 struct rte_mbuf *mbuf_in = bufs[num_tx];
 663                 struct rte_mbuf **mbuf;
 664                 uint16_t num_mbufs = 0;
 665                 uint16_t tso_segsz = 0;
 666                 int ret;
 667                 uint16_t hdrs_len;
 668                 int j;
 669                 uint64_t tso;
 670
 671                 tso = mbuf_in->ol_flags & PKT_TX_TCP_SEG;
 672                 if (tso) {
 673                         struct rte_gso_ctx *gso_ctx = &txq->gso_ctx;
 674
 675                         assert(gso_ctx != NULL);
 676
 677                         /* TCP segmentation implies TCP checksum offload */
 678                         mbuf_in->ol_flags |= PKT_TX_TCP_CKSUM;
 679
 680                         /* gso size is calculated without RTE_ETHER_CRC_LEN */
 681                         hdrs_len = mbuf_in->l2_len + mbuf_in->l3_len +
 682                                         mbuf_in->l4_len;
 683                         tso_segsz = mbuf_in->tso_segsz + hdrs_len;
 684                         if (unlikely(tso_segsz == hdrs_len) ||
 685                                 tso_segsz > *txq->mtu) {
 686                                 txq->stats.errs++;
 687                                 break;
 688                         }
 689                         gso_ctx->gso_size = tso_segsz;
 690                         ret = rte_gso_segment(mbuf_in, /* packet to segment */
 691                                 gso_ctx, /* gso control block */
 692                                 (struct rte_mbuf **)&gso_mbufs, /* out mbufs */
 693                                 RTE_DIM(gso_mbufs)); /* max tso mbufs */
 694
 695                         /* ret contains the number of new created mbufs */
 696                         if (ret < 0)
 697                                 break;
 698
 699                         mbuf = gso_mbufs;
 700                         num_mbufs = ret;
 701                 } else {
 702                         /* stats.errs will be incremented */
 703                         if (rte_pktmbuf_pkt_len(mbuf_in) > max_size)
 704                                 break;
 705
 706                         /* ret 0 indicates no new mbufs were created */
 707                         ret = 0;
 708                         mbuf = &mbuf_in;
 709                         num_mbufs = 1;
 710                 }
 711
 712                 tap_write_mbufs(txq, num_mbufs, mbuf,
 713                                 &num_packets, &num_tx_bytes);
 714                 num_tx++;
 715                 /* free original mbuf */
 716                 rte_pktmbuf_free(mbuf_in);
 717                 /* free tso mbufs */
 718                 for (j = 0; j < ret; j++)
 719                         rte_pktmbuf_free(mbuf[j]);
 720         }
 721
 722         txq->stats.opackets += num_packets;
 723         txq->stats.errs += nb_pkts - num_tx;
 724         txq->stats.obytes += num_tx_bytes;
 725
 726         return num_packets;
 727 }
 728
 729 static const char *
 730 tap_ioctl_req2str(unsigned long request)
 731 {
 732         switch (request) {
 733         case SIOCSIFFLAGS:
 734                 return "SIOCSIFFLAGS";
 735         case SIOCGIFFLAGS:
 736                 return "SIOCGIFFLAGS";
 737         case SIOCGIFHWADDR:
 738                 return "SIOCGIFHWADDR";
 739         case SIOCSIFHWADDR:
 740                 return "SIOCSIFHWADDR";
 741         case SIOCSIFMTU:
 742                 return "SIOCSIFMTU";
 743         }
 744         return "UNKNOWN";
 745 }
 746
 747 static int
 748 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 749           struct ifreq *ifr, int set, enum ioctl_mode mode)
 750 {
 751         short req_flags = ifr->ifr_flags;
 752         int remote = pmd->remote_if_index &&
 753                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 754
 755         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 756                 return 0;
 757         /*
 758          * If there is a remote netdevice, apply ioctl on it, then apply it on
 759          * the tap netdevice.
 760          */
 761 apply:
 762         if (remote)
 763                 strlcpy(ifr->ifr_name, pmd->remote_iface, IFNAMSIZ);
 764         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 765                 strlcpy(ifr->ifr_name, pmd->name, IFNAMSIZ);
 766         switch (request) {
 767         case SIOCSIFFLAGS:
 768                 /* fetch current flags to leave other flags untouched */
 769                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 770                         goto error;
 771                 if (set)
 772                         ifr->ifr_flags |= req_flags;
 773                 else
 774                         ifr->ifr_flags &= ~req_flags;
 775                 break;
 776         case SIOCGIFFLAGS:
 777         case SIOCGIFHWADDR:
 778         case SIOCSIFHWADDR:
 779         case SIOCSIFMTU:
 780                 break;
 781         default:
 782                 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
 783                         pmd->name);
 784                 return -EINVAL;
 785         }
 786         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 787                 goto error;
 788         if (remote-- && mode == LOCAL_AND_REMOTE)
 789                 goto apply;
 790         return 0;
 791
 792 error:
 793         TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
 794                 tap_ioctl_req2str(request), strerror(errno), errno);
 795         return -errno;
 796 }
 797
 798 static int
 799 tap_link_set_down(struct rte_eth_dev *dev)
 800 {
 801         struct pmd_internals *pmd = dev->data->dev_private;
 802         struct ifreq ifr = { .ifr_flags = IFF_UP };
 803
 804         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 805         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 806 }
 807
 808 static int
 809 tap_link_set_up(struct rte_eth_dev *dev)
 810 {
 811         struct pmd_internals *pmd = dev->data->dev_private;
 812         struct ifreq ifr = { .ifr_flags = IFF_UP };
 813
 814         dev->data->dev_link.link_status = ETH_LINK_UP;
 815         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 816 }
 817
 818 static int
 819 tap_dev_start(struct rte_eth_dev *dev)
 820 {
 821         int err, i;
 822
 823         err = tap_intr_handle_set(dev, 1);
 824         if (err)
 825                 return err;
 826
 827         err = tap_link_set_up(dev);
 828         if (err)
 829                 return err;
 830
 831         for (i = 0; i < dev->data->nb_tx_queues; i++)
 832                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 833         for (i = 0; i < dev->data->nb_rx_queues; i++)
 834                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 835
 836         return err;
 837 }
 838
 839 /* This function gets called when the current port gets stopped.
 840  */
 841 static void
 842 tap_dev_stop(struct rte_eth_dev *dev)
 843 {
 844         int i;
 845
 846         for (i = 0; i < dev->data->nb_tx_queues; i++)
 847                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 848         for (i = 0; i < dev->data->nb_rx_queues; i++)
 849                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 850
 851         tap_intr_handle_set(dev, 0);
 852         tap_link_set_down(dev);
 853 }
 854
 855 static int
 856 tap_dev_configure(struct rte_eth_dev *dev)
 857 {
 858         struct pmd_internals *pmd = dev->data->dev_private;
 859
 860         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 861                 TAP_LOG(ERR,
 862                         "%s: number of rx queues %d exceeds max num of queues %d",
 863                         dev->device->name,
 864                         dev->data->nb_rx_queues,
 865                         RTE_PMD_TAP_MAX_QUEUES);
 866                 return -1;
 867         }
 868         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 869                 TAP_LOG(ERR,
 870                         "%s: number of tx queues %d exceeds max num of queues %d",
 871                         dev->device->name,
 872                         dev->data->nb_tx_queues,
 873                         RTE_PMD_TAP_MAX_QUEUES);
 874                 return -1;
 875         }
 876
 877         TAP_LOG(INFO, "%s: %s: TX configured queues number: %u",
 878                 dev->device->name, pmd->name, dev->data->nb_tx_queues);
 879
 880         TAP_LOG(INFO, "%s: %s: RX configured queues number: %u",
 881                 dev->device->name, pmd->name, dev->data->nb_rx_queues);
 882
 883         return 0;
 884 }
 885
 886 static uint32_t
 887 tap_dev_speed_capa(void)
 888 {
 889         uint32_t speed = pmd_link.link_speed;
 890         uint32_t capa = 0;
 891
 892         if (speed >= ETH_SPEED_NUM_10M)
 893                 capa |= ETH_LINK_SPEED_10M;
 894         if (speed >= ETH_SPEED_NUM_100M)
 895                 capa |= ETH_LINK_SPEED_100M;
 896         if (speed >= ETH_SPEED_NUM_1G)
 897                 capa |= ETH_LINK_SPEED_1G;
 898         if (speed >= ETH_SPEED_NUM_5G)
 899                 capa |= ETH_LINK_SPEED_2_5G;
 900         if (speed >= ETH_SPEED_NUM_5G)
 901                 capa |= ETH_LINK_SPEED_5G;
 902         if (speed >= ETH_SPEED_NUM_10G)
 903                 capa |= ETH_LINK_SPEED_10G;
 904         if (speed >= ETH_SPEED_NUM_20G)
 905                 capa |= ETH_LINK_SPEED_20G;
 906         if (speed >= ETH_SPEED_NUM_25G)
 907                 capa |= ETH_LINK_SPEED_25G;
 908         if (speed >= ETH_SPEED_NUM_40G)
 909                 capa |= ETH_LINK_SPEED_40G;
 910         if (speed >= ETH_SPEED_NUM_50G)
 911                 capa |= ETH_LINK_SPEED_50G;
 912         if (speed >= ETH_SPEED_NUM_56G)
 913                 capa |= ETH_LINK_SPEED_56G;
 914         if (speed >= ETH_SPEED_NUM_100G)
 915                 capa |= ETH_LINK_SPEED_100G;
 916
 917         return capa;
 918 }
 919
 920 static void
 921 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 922 {
 923         struct pmd_internals *internals = dev->data->dev_private;
 924
 925         dev_info->if_index = internals->if_index;
 926         dev_info->max_mac_addrs = 1;
 927         dev_info->max_rx_pktlen = (uint32_t)RTE_ETHER_MAX_VLAN_FRAME_LEN;
 928         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
 929         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
 930         dev_info->min_rx_bufsize = 0;
 931         dev_info->speed_capa = tap_dev_speed_capa();
 932         dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
 933         dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
 934                                     dev_info->rx_queue_offload_capa;
 935         dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
 936         dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
 937                                     dev_info->tx_queue_offload_capa;
 938         dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
 939         /*
 940          * limitation: TAP supports all of IP, UDP and TCP hash
 941          * functions together and not in partial combinations
 942          */
 943         dev_info->flow_type_rss_offloads = ~TAP_RSS_HF_MASK;
 944 }
 945
 946 static int
 947 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 948 {
 949         unsigned int i, imax;
 950         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 951         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 952         unsigned long rx_nombuf = 0, ierrors = 0;
 953         const struct pmd_internals *pmd = dev->data->dev_private;
 954
 955         /* rx queue statistics */
 956         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 957                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 958         for (i = 0; i < imax; i++) {
 959                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 960                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 961                 rx_total += tap_stats->q_ipackets[i];
 962                 rx_bytes_total += tap_stats->q_ibytes[i];
 963                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
 964                 ierrors += pmd->rxq[i].stats.ierrors;
 965         }
 966
 967         /* tx queue statistics */
 968         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 969                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 970
 971         for (i = 0; i < imax; i++) {
 972                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 973                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 974                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 975                 tx_total += tap_stats->q_opackets[i];
 976                 tx_err_total += tap_stats->q_errors[i];
 977                 tx_bytes_total += tap_stats->q_obytes[i];
 978         }
 979
 980         tap_stats->ipackets = rx_total;
 981         tap_stats->ibytes = rx_bytes_total;
 982         tap_stats->ierrors = ierrors;
 983         tap_stats->rx_nombuf = rx_nombuf;
 984         tap_stats->opackets = tx_total;
 985         tap_stats->oerrors = tx_err_total;
 986         tap_stats->obytes = tx_bytes_total;
 987         return 0;
 988 }
 989
 990 static void
 991 tap_stats_reset(struct rte_eth_dev *dev)
 992 {
 993         int i;
 994         struct pmd_internals *pmd = dev->data->dev_private;
 995
 996         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 997                 pmd->rxq[i].stats.ipackets = 0;
 998                 pmd->rxq[i].stats.ibytes = 0;
 999                 pmd->rxq[i].stats.ierrors = 0;
1000                 pmd->rxq[i].stats.rx_nombuf = 0;
1001
1002                 pmd->txq[i].stats.opackets = 0;
1003                 pmd->txq[i].stats.errs = 0;
1004                 pmd->txq[i].stats.obytes = 0;
1005         }
1006 }
1007
1008 static void
1009 tap_dev_close(struct rte_eth_dev *dev)
1010 {
1011         int i;
1012         struct pmd_internals *internals = dev->data->dev_private;
1013         struct pmd_process_private *process_private = dev->process_private;
1014
1015         tap_link_set_down(dev);
1016         tap_flow_flush(dev, NULL);
1017         tap_flow_implicit_flush(internals, NULL);
1018
1019         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1020                 if (process_private->rxq_fds[i] != -1) {
1021                         close(process_private->rxq_fds[i]);
1022                         process_private->rxq_fds[i] = -1;
1023                 }
1024                 if (process_private->txq_fds[i] != -1) {
1025                         close(process_private->txq_fds[i]);
1026                         process_private->txq_fds[i] = -1;
1027                 }
1028         }
1029
1030         if (internals->remote_if_index) {
1031                 /* Restore initial remote state */
1032                 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
1033                                 &internals->remote_initial_flags);
1034         }
1035
1036         if (internals->ka_fd != -1) {
1037                 close(internals->ka_fd);
1038                 internals->ka_fd = -1;
1039         }
1040         /*
1041          * Since TUN device has no more opened file descriptors
1042          * it will be removed from kernel
1043          */
1044 }
1045
1046 static void
1047 tap_rx_queue_release(void *queue)
1048 {
1049         struct rx_queue *rxq = queue;
1050         struct pmd_process_private *process_private;
1051
1052         if (!rxq)
1053                 return;
1054         process_private = rte_eth_devices[rxq->in_port].process_private;
1055         if (process_private->rxq_fds[rxq->queue_id] > 0) {
1056                 close(process_private->rxq_fds[rxq->queue_id]);
1057                 process_private->rxq_fds[rxq->queue_id] = -1;
1058                 rte_pktmbuf_free(rxq->pool);
1059                 rte_free(rxq->iovecs);
1060                 rxq->pool = NULL;
1061                 rxq->iovecs = NULL;
1062         }
1063 }
1064
1065 static void
1066 tap_tx_queue_release(void *queue)
1067 {
1068         struct tx_queue *txq = queue;
1069         struct pmd_process_private *process_private;
1070
1071         if (!txq)
1072                 return;
1073         process_private = rte_eth_devices[txq->out_port].process_private;
1074
1075         if (process_private->txq_fds[txq->queue_id] > 0) {
1076                 close(process_private->txq_fds[txq->queue_id]);
1077                 process_private->txq_fds[txq->queue_id] = -1;
1078         }
1079 }
1080
1081 static int
1082 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1083 {
1084         struct rte_eth_link *dev_link = &dev->data->dev_link;
1085         struct pmd_internals *pmd = dev->data->dev_private;
1086         struct ifreq ifr = { .ifr_flags = 0 };
1087
1088         if (pmd->remote_if_index) {
1089                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
1090                 if (!(ifr.ifr_flags & IFF_UP) ||
1091                     !(ifr.ifr_flags & IFF_RUNNING)) {
1092                         dev_link->link_status = ETH_LINK_DOWN;
1093                         return 0;
1094                 }
1095         }
1096         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
1097         dev_link->link_status =
1098                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
1099                  ETH_LINK_UP :
1100                  ETH_LINK_DOWN);
1101         return 0;
1102 }
1103
1104 static void
1105 tap_promisc_enable(struct rte_eth_dev *dev)
1106 {
1107         struct pmd_internals *pmd = dev->data->dev_private;
1108         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1109
1110         dev->data->promiscuous = 1;
1111         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1112         if (pmd->remote_if_index && !pmd->flow_isolate)
1113                 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
1114 }
1115
1116 static void
1117 tap_promisc_disable(struct rte_eth_dev *dev)
1118 {
1119         struct pmd_internals *pmd = dev->data->dev_private;
1120         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1121
1122         dev->data->promiscuous = 0;
1123         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1124         if (pmd->remote_if_index && !pmd->flow_isolate)
1125                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
1126 }
1127
1128 static void
1129 tap_allmulti_enable(struct rte_eth_dev *dev)
1130 {
1131         struct pmd_internals *pmd = dev->data->dev_private;
1132         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1133
1134         dev->data->all_multicast = 1;
1135         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1136         if (pmd->remote_if_index && !pmd->flow_isolate)
1137                 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
1138 }
1139
1140 static void
1141 tap_allmulti_disable(struct rte_eth_dev *dev)
1142 {
1143         struct pmd_internals *pmd = dev->data->dev_private;
1144         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1145
1146         dev->data->all_multicast = 0;
1147         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1148         if (pmd->remote_if_index && !pmd->flow_isolate)
1149                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
1150 }
1151
1152 static int
1153 tap_mac_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1154 {
1155         struct pmd_internals *pmd = dev->data->dev_private;
1156         enum ioctl_mode mode = LOCAL_ONLY;
1157         struct ifreq ifr;
1158         int ret;
1159
1160         if (pmd->type == ETH_TUNTAP_TYPE_TUN) {
1161                 TAP_LOG(ERR, "%s: can't MAC address for TUN",
1162                         dev->device->name);
1163                 return -ENOTSUP;
1164         }
1165
1166         if (rte_is_zero_ether_addr(mac_addr)) {
1167                 TAP_LOG(ERR, "%s: can't set an empty MAC address",
1168                         dev->device->name);
1169                 return -EINVAL;
1170         }
1171         /* Check the actual current MAC address on the tap netdevice */
1172         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
1173         if (ret < 0)
1174                 return ret;
1175         if (rte_is_same_ether_addr(
1176                         (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1177                         mac_addr))
1178                 return 0;
1179         /* Check the current MAC address on the remote */
1180         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
1181         if (ret < 0)
1182                 return ret;
1183         if (!rte_is_same_ether_addr(
1184                         (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1185                         mac_addr))
1186                 mode = LOCAL_AND_REMOTE;
1187         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1188         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, RTE_ETHER_ADDR_LEN);
1189         ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
1190         if (ret < 0)
1191                 return ret;
1192         rte_memcpy(&pmd->eth_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1193         if (pmd->remote_if_index && !pmd->flow_isolate) {
1194                 /* Replace MAC redirection rule after a MAC change */
1195                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
1196                 if (ret < 0) {
1197                         TAP_LOG(ERR,
1198                                 "%s: Couldn't delete MAC redirection rule",
1199                                 dev->device->name);
1200                         return ret;
1201                 }
1202                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1203                 if (ret < 0) {
1204                         TAP_LOG(ERR,
1205                                 "%s: Couldn't add MAC redirection rule",
1206                                 dev->device->name);
1207                         return ret;
1208                 }
1209         }
1210
1211         return 0;
1212 }
1213
1214 static int
1215 tap_gso_ctx_setup(struct rte_gso_ctx *gso_ctx, struct rte_eth_dev *dev)
1216 {
1217         uint32_t gso_types;
1218         char pool_name[64];
1219
1220         /*
1221          * Create private mbuf pool with TAP_GSO_MBUF_SEG_SIZE bytes
1222          * size per mbuf use this pool for both direct and indirect mbufs
1223          */
1224
1225         struct rte_mempool *mp;      /* Mempool for GSO packets */
1226
1227         /* initialize GSO context */
1228         gso_types = DEV_TX_OFFLOAD_TCP_TSO;
1229         snprintf(pool_name, sizeof(pool_name), "mp_%s", dev->device->name);
1230         mp = rte_mempool_lookup((const char *)pool_name);
1231         if (!mp) {
1232                 mp = rte_pktmbuf_pool_create(pool_name, TAP_GSO_MBUFS_NUM,
1233                         TAP_GSO_MBUF_CACHE_SIZE, 0,
1234                         RTE_PKTMBUF_HEADROOM + TAP_GSO_MBUF_SEG_SIZE,
1235                         SOCKET_ID_ANY);
1236                 if (!mp) {
1237                         struct pmd_internals *pmd = dev->data->dev_private;
1238                         RTE_LOG(DEBUG, PMD, "%s: failed to create mbuf pool for device %s\n",
1239                                 pmd->name, dev->device->name);
1240                         return -1;
1241                 }
1242         }
1243
1244         gso_ctx->direct_pool = mp;
1245         gso_ctx->indirect_pool = mp;
1246         gso_ctx->gso_types = gso_types;
1247         gso_ctx->gso_size = 0; /* gso_size is set in tx_burst() per packet */
1248         gso_ctx->flag = 0;
1249
1250         return 0;
1251 }
1252
1253 static int
1254 tap_setup_queue(struct rte_eth_dev *dev,
1255                 struct pmd_internals *internals,
1256                 uint16_t qid,
1257                 int is_rx)
1258 {
1259         int ret;
1260         int *fd;
1261         int *other_fd;
1262         const char *dir;
1263         struct pmd_internals *pmd = dev->data->dev_private;
1264         struct pmd_process_private *process_private = dev->process_private;
1265         struct rx_queue *rx = &internals->rxq[qid];
1266         struct tx_queue *tx = &internals->txq[qid];
1267         struct rte_gso_ctx *gso_ctx;
1268
1269         if (is_rx) {
1270                 fd = &process_private->rxq_fds[qid];
1271                 other_fd = &process_private->txq_fds[qid];
1272                 dir = "rx";
1273                 gso_ctx = NULL;
1274         } else {
1275                 fd = &process_private->txq_fds[qid];
1276                 other_fd = &process_private->rxq_fds[qid];
1277                 dir = "tx";
1278                 gso_ctx = &tx->gso_ctx;
1279         }
1280         if (*fd != -1) {
1281                 /* fd for this queue already exists */
1282                 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1283                         pmd->name, *fd, dir, qid);
1284                 gso_ctx = NULL;
1285         } else if (*other_fd != -1) {
1286                 /* Only other_fd exists. dup it */
1287                 *fd = dup(*other_fd);
1288                 if (*fd < 0) {
1289                         *fd = -1;
1290                         TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1291                         return -1;
1292                 }
1293                 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1294                         pmd->name, *other_fd, dir, qid, *fd);
1295         } else {
1296                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1297                 *fd = tun_alloc(pmd, 0);
1298                 if (*fd < 0) {
1299                         *fd = -1; /* restore original value */
1300                         TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1301                         return -1;
1302                 }
1303                 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1304                         pmd->name, dir, qid, *fd);
1305         }
1306
1307         tx->mtu = &dev->data->mtu;
1308         rx->rxmode = &dev->data->dev_conf.rxmode;
1309         if (gso_ctx) {
1310                 ret = tap_gso_ctx_setup(gso_ctx, dev);
1311                 if (ret)
1312                         return -1;
1313         }
1314
1315         tx->type = pmd->type;
1316
1317         return *fd;
1318 }
1319
1320 static int
1321 tap_rx_queue_setup(struct rte_eth_dev *dev,
1322                    uint16_t rx_queue_id,
1323                    uint16_t nb_rx_desc,
1324                    unsigned int socket_id,
1325                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1326                    struct rte_mempool *mp)
1327 {
1328         struct pmd_internals *internals = dev->data->dev_private;
1329         struct pmd_process_private *process_private = dev->process_private;
1330         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1331         struct rte_mbuf **tmp = &rxq->pool;
1332         long iov_max = sysconf(_SC_IOV_MAX);
1333
1334         if (iov_max <= 0) {
1335                 TAP_LOG(WARNING,
1336                         "_SC_IOV_MAX is not defined. Using %d as default",
1337                         TAP_IOV_DEFAULT_MAX);
1338                 iov_max = TAP_IOV_DEFAULT_MAX;
1339         }
1340         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1341         struct iovec (*iovecs)[nb_desc + 1];
1342         int data_off = RTE_PKTMBUF_HEADROOM;
1343         int ret = 0;
1344         int fd;
1345         int i;
1346
1347         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1348                 TAP_LOG(WARNING,
1349                         "nb_rx_queues %d too small or mempool NULL",
1350                         dev->data->nb_rx_queues);
1351                 return -1;
1352         }
1353
1354         rxq->mp = mp;
1355         rxq->trigger_seen = 1; /* force initial burst */
1356         rxq->in_port = dev->data->port_id;
1357         rxq->queue_id = rx_queue_id;
1358         rxq->nb_rx_desc = nb_desc;
1359         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1360                                     socket_id);
1361         if (!iovecs) {
1362                 TAP_LOG(WARNING,
1363                         "%s: Couldn't allocate %d RX descriptors",
1364                         dev->device->name, nb_desc);
1365                 return -ENOMEM;
1366         }
1367         rxq->iovecs = iovecs;
1368
1369         dev->data->rx_queues[rx_queue_id] = rxq;
1370         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1371         if (fd == -1) {
1372                 ret = fd;
1373                 goto error;
1374         }
1375
1376         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1377         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1378
1379         for (i = 1; i <= nb_desc; i++) {
1380                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1381                 if (!*tmp) {
1382                         TAP_LOG(WARNING,
1383                                 "%s: couldn't allocate memory for queue %d",
1384                                 dev->device->name, rx_queue_id);
1385                         ret = -ENOMEM;
1386                         goto error;
1387                 }
1388                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1389                 (*rxq->iovecs)[i].iov_base =
1390                         (char *)(*tmp)->buf_addr + data_off;
1391                 data_off = 0;
1392                 tmp = &(*tmp)->next;
1393         }
1394
1395         TAP_LOG(DEBUG, "  RX TUNTAP device name %s, qid %d on fd %d",
1396                 internals->name, rx_queue_id,
1397                 process_private->rxq_fds[rx_queue_id]);
1398
1399         return 0;
1400
1401 error:
1402         rte_pktmbuf_free(rxq->pool);
1403         rxq->pool = NULL;
1404         rte_free(rxq->iovecs);
1405         rxq->iovecs = NULL;
1406         return ret;
1407 }
1408
1409 static int
1410 tap_tx_queue_setup(struct rte_eth_dev *dev,
1411                    uint16_t tx_queue_id,
1412                    uint16_t nb_tx_desc __rte_unused,
1413                    unsigned int socket_id __rte_unused,
1414                    const struct rte_eth_txconf *tx_conf)
1415 {
1416         struct pmd_internals *internals = dev->data->dev_private;
1417         struct pmd_process_private *process_private = dev->process_private;
1418         struct tx_queue *txq;
1419         int ret;
1420         uint64_t offloads;
1421
1422         if (tx_queue_id >= dev->data->nb_tx_queues)
1423                 return -1;
1424         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1425         txq = dev->data->tx_queues[tx_queue_id];
1426         txq->out_port = dev->data->port_id;
1427         txq->queue_id = tx_queue_id;
1428
1429         offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1430         txq->csum = !!(offloads &
1431                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
1432                          DEV_TX_OFFLOAD_UDP_CKSUM |
1433                          DEV_TX_OFFLOAD_TCP_CKSUM));
1434
1435         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1436         if (ret == -1)
1437                 return -1;
1438         TAP_LOG(DEBUG,
1439                 "  TX TUNTAP device name %s, qid %d on fd %d csum %s",
1440                 internals->name, tx_queue_id,
1441                 process_private->txq_fds[tx_queue_id],
1442                 txq->csum ? "on" : "off");
1443
1444         return 0;
1445 }
1446
1447 static int
1448 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1449 {
1450         struct pmd_internals *pmd = dev->data->dev_private;
1451         struct ifreq ifr = { .ifr_mtu = mtu };
1452         int err = 0;
1453
1454         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1455         if (!err)
1456                 dev->data->mtu = mtu;
1457
1458         return err;
1459 }
1460
1461 static int
1462 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1463                      struct rte_ether_addr *mc_addr_set __rte_unused,
1464                      uint32_t nb_mc_addr __rte_unused)
1465 {
1466         /*
1467          * Nothing to do actually: the tap has no filtering whatsoever, every
1468          * packet is received.
1469          */
1470         return 0;
1471 }
1472
1473 static int
1474 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1475 {
1476         struct rte_eth_dev *dev = arg;
1477         struct pmd_internals *pmd = dev->data->dev_private;
1478         struct ifinfomsg *info = NLMSG_DATA(nh);
1479
1480         if (nh->nlmsg_type != RTM_NEWLINK ||
1481             (info->ifi_index != pmd->if_index &&
1482              info->ifi_index != pmd->remote_if_index))
1483                 return 0;
1484         return tap_link_update(dev, 0);
1485 }
1486
1487 static void
1488 tap_dev_intr_handler(void *cb_arg)
1489 {
1490         struct rte_eth_dev *dev = cb_arg;
1491         struct pmd_internals *pmd = dev->data->dev_private;
1492
1493         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1494 }
1495
1496 static int
1497 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1498 {
1499         struct pmd_internals *pmd = dev->data->dev_private;
1500
1501         /* In any case, disable interrupt if the conf is no longer there. */
1502         if (!dev->data->dev_conf.intr_conf.lsc) {
1503                 if (pmd->intr_handle.fd != -1) {
1504                         tap_nl_final(pmd->intr_handle.fd);
1505                         rte_intr_callback_unregister(&pmd->intr_handle,
1506                                 tap_dev_intr_handler, dev);
1507                 }
1508                 return 0;
1509         }
1510         if (set) {
1511                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1512                 if (unlikely(pmd->intr_handle.fd == -1))
1513                         return -EBADF;
1514                 return rte_intr_callback_register(
1515                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1516         }
1517         tap_nl_final(pmd->intr_handle.fd);
1518         return rte_intr_callback_unregister(&pmd->intr_handle,
1519                                             tap_dev_intr_handler, dev);
1520 }
1521
1522 static int
1523 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1524 {
1525         int err;
1526
1527         err = tap_lsc_intr_handle_set(dev, set);
1528         if (err)
1529                 return err;
1530         err = tap_rx_intr_vec_set(dev, set);
1531         if (err && set)
1532                 tap_lsc_intr_handle_set(dev, 0);
1533         return err;
1534 }
1535
1536 static const uint32_t*
1537 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1538 {
1539         static const uint32_t ptypes[] = {
1540                 RTE_PTYPE_INNER_L2_ETHER,
1541                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1542                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1543                 RTE_PTYPE_INNER_L3_IPV4,
1544                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1545                 RTE_PTYPE_INNER_L3_IPV6,
1546                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1547                 RTE_PTYPE_INNER_L4_FRAG,
1548                 RTE_PTYPE_INNER_L4_UDP,
1549                 RTE_PTYPE_INNER_L4_TCP,
1550                 RTE_PTYPE_INNER_L4_SCTP,
1551                 RTE_PTYPE_L2_ETHER,
1552                 RTE_PTYPE_L2_ETHER_VLAN,
1553                 RTE_PTYPE_L2_ETHER_QINQ,
1554                 RTE_PTYPE_L3_IPV4,
1555                 RTE_PTYPE_L3_IPV4_EXT,
1556                 RTE_PTYPE_L3_IPV6_EXT,
1557                 RTE_PTYPE_L3_IPV6,
1558                 RTE_PTYPE_L4_FRAG,
1559                 RTE_PTYPE_L4_UDP,
1560                 RTE_PTYPE_L4_TCP,
1561                 RTE_PTYPE_L4_SCTP,
1562         };
1563
1564         return ptypes;
1565 }
1566
1567 static int
1568 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1569                   struct rte_eth_fc_conf *fc_conf)
1570 {
1571         fc_conf->mode = RTE_FC_NONE;
1572         return 0;
1573 }
1574
1575 static int
1576 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1577                   struct rte_eth_fc_conf *fc_conf)
1578 {
1579         if (fc_conf->mode != RTE_FC_NONE)
1580                 return -ENOTSUP;
1581         return 0;
1582 }
1583
1584 /**
1585  * DPDK callback to update the RSS hash configuration.
1586  *
1587  * @param dev
1588  *   Pointer to Ethernet device structure.
1589  * @param[in] rss_conf
1590  *   RSS configuration data.
1591  *
1592  * @return
1593  *   0 on success, a negative errno value otherwise and rte_errno is set.
1594  */
1595 static int
1596 tap_rss_hash_update(struct rte_eth_dev *dev,
1597                 struct rte_eth_rss_conf *rss_conf)
1598 {
1599         if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1600                 rte_errno = EINVAL;
1601                 return -rte_errno;
1602         }
1603         if (rss_conf->rss_key && rss_conf->rss_key_len) {
1604                 /*
1605                  * Currently TAP RSS key is hard coded
1606                  * and cannot be updated
1607                  */
1608                 TAP_LOG(ERR,
1609                         "port %u RSS key cannot be updated",
1610                         dev->data->port_id);
1611                 rte_errno = EINVAL;
1612                 return -rte_errno;
1613         }
1614         return 0;
1615 }
1616
1617 static int
1618 tap_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1619 {
1620         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1621
1622         return 0;
1623 }
1624
1625 static int
1626 tap_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1627 {
1628         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1629
1630         return 0;
1631 }
1632
1633 static int
1634 tap_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1635 {
1636         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1637
1638         return 0;
1639 }
1640
1641 static int
1642 tap_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1643 {
1644         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1645
1646         return 0;
1647 }
1648 static const struct eth_dev_ops ops = {
1649         .dev_start              = tap_dev_start,
1650         .dev_stop               = tap_dev_stop,
1651         .dev_close              = tap_dev_close,
1652         .dev_configure          = tap_dev_configure,
1653         .dev_infos_get          = tap_dev_info,
1654         .rx_queue_setup         = tap_rx_queue_setup,
1655         .tx_queue_setup         = tap_tx_queue_setup,
1656         .rx_queue_start         = tap_rx_queue_start,
1657         .tx_queue_start         = tap_tx_queue_start,
1658         .rx_queue_stop          = tap_rx_queue_stop,
1659         .tx_queue_stop          = tap_tx_queue_stop,
1660         .rx_queue_release       = tap_rx_queue_release,
1661         .tx_queue_release       = tap_tx_queue_release,
1662         .flow_ctrl_get          = tap_flow_ctrl_get,
1663         .flow_ctrl_set          = tap_flow_ctrl_set,
1664         .link_update            = tap_link_update,
1665         .dev_set_link_up        = tap_link_set_up,
1666         .dev_set_link_down      = tap_link_set_down,
1667         .promiscuous_enable     = tap_promisc_enable,
1668         .promiscuous_disable    = tap_promisc_disable,
1669         .allmulticast_enable    = tap_allmulti_enable,
1670         .allmulticast_disable   = tap_allmulti_disable,
1671         .mac_addr_set           = tap_mac_set,
1672         .mtu_set                = tap_mtu_set,
1673         .set_mc_addr_list       = tap_set_mc_addr_list,
1674         .stats_get              = tap_stats_get,
1675         .stats_reset            = tap_stats_reset,
1676         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1677         .rss_hash_update        = tap_rss_hash_update,
1678         .filter_ctrl            = tap_dev_filter_ctrl,
1679 };
1680
1681 static const char *tuntap_types[ETH_TUNTAP_TYPE_MAX] = {
1682         "UNKNOWN", "TUN", "TAP"
1683 };
1684
1685 static int
1686 eth_dev_tap_create(struct rte_vdev_device *vdev, const char *tap_name,
1687                    char *remote_iface, struct rte_ether_addr *mac_addr,
1688                    enum rte_tuntap_type type)
1689 {
1690         int numa_node = rte_socket_id();
1691         struct rte_eth_dev *dev;
1692         struct pmd_internals *pmd;
1693         struct pmd_process_private *process_private;
1694         const char *tuntap_name = tuntap_types[type];
1695         struct rte_eth_dev_data *data;
1696         struct ifreq ifr;
1697         int i;
1698
1699         TAP_LOG(DEBUG, "%s device on numa %u", tuntap_name, rte_socket_id());
1700
1701         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1702         if (!dev) {
1703                 TAP_LOG(ERR, "%s Unable to allocate device struct",
1704                                 tuntap_name);
1705                 goto error_exit_nodev;
1706         }
1707
1708         process_private = (struct pmd_process_private *)
1709                 rte_zmalloc_socket(tap_name, sizeof(struct pmd_process_private),
1710                         RTE_CACHE_LINE_SIZE, dev->device->numa_node);
1711
1712         if (process_private == NULL) {
1713                 TAP_LOG(ERR, "Failed to alloc memory for process private");
1714                 return -1;
1715         }
1716         pmd = dev->data->dev_private;
1717         dev->process_private = process_private;
1718         pmd->dev = dev;
1719         strlcpy(pmd->name, tap_name, sizeof(pmd->name));
1720         pmd->type = type;
1721
1722         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1723         if (pmd->ioctl_sock == -1) {
1724                 TAP_LOG(ERR,
1725                         "%s Unable to get a socket for management: %s",
1726                         tuntap_name, strerror(errno));
1727                 goto error_exit;
1728         }
1729
1730         /* Setup some default values */
1731         data = dev->data;
1732         data->dev_private = pmd;
1733         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1734         data->numa_node = numa_node;
1735
1736         data->dev_link = pmd_link;
1737         data->mac_addrs = &pmd->eth_addr;
1738         /* Set the number of RX and TX queues */
1739         data->nb_rx_queues = 0;
1740         data->nb_tx_queues = 0;
1741
1742         dev->dev_ops = &ops;
1743         dev->rx_pkt_burst = pmd_rx_burst;
1744         dev->tx_pkt_burst = pmd_tx_burst;
1745
1746         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1747         pmd->intr_handle.fd = -1;
1748         dev->intr_handle = &pmd->intr_handle;
1749
1750         /* Presetup the fds to -1 as being not valid */
1751         pmd->ka_fd = -1;
1752         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1753                 process_private->rxq_fds[i] = -1;
1754                 process_private->txq_fds[i] = -1;
1755         }
1756
1757         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1758                 if (rte_is_zero_ether_addr(mac_addr))
1759                         rte_eth_random_addr((uint8_t *)&pmd->eth_addr);
1760                 else
1761                         rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1762         }
1763
1764         /*
1765          * Allocate a TUN device keep-alive file descriptor that will only be
1766          * closed when the TUN device itself is closed or removed.
1767          * This keep-alive file descriptor will guarantee that the TUN device
1768          * exists even when all of its queues are closed
1769          */
1770         pmd->ka_fd = tun_alloc(pmd, 1);
1771         if (pmd->ka_fd == -1) {
1772                 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
1773                 goto error_exit;
1774         }
1775         TAP_LOG(DEBUG, "allocated %s", pmd->name);
1776
1777         ifr.ifr_mtu = dev->data->mtu;
1778         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1779                 goto error_exit;
1780
1781         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1782                 memset(&ifr, 0, sizeof(struct ifreq));
1783                 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1784                 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1785                                 RTE_ETHER_ADDR_LEN);
1786                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1787                         goto error_exit;
1788         }
1789
1790         /*
1791          * Set up everything related to rte_flow:
1792          * - netlink socket
1793          * - tap / remote if_index
1794          * - mandatory QDISCs
1795          * - rte_flow actual/implicit lists
1796          * - implicit rules
1797          */
1798         pmd->nlsk_fd = tap_nl_init(0);
1799         if (pmd->nlsk_fd == -1) {
1800                 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
1801                         pmd->name);
1802                 goto disable_rte_flow;
1803         }
1804         pmd->if_index = if_nametoindex(pmd->name);
1805         if (!pmd->if_index) {
1806                 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
1807                 goto disable_rte_flow;
1808         }
1809         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1810                 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
1811                         pmd->name);
1812                 goto disable_rte_flow;
1813         }
1814         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1815                 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1816                         pmd->name);
1817                 goto disable_rte_flow;
1818         }
1819         LIST_INIT(&pmd->flows);
1820
1821         if (strlen(remote_iface)) {
1822                 pmd->remote_if_index = if_nametoindex(remote_iface);
1823                 if (!pmd->remote_if_index) {
1824                         TAP_LOG(ERR, "%s: failed to get %s if_index.",
1825                                 pmd->name, remote_iface);
1826                         goto error_remote;
1827                 }
1828                 strlcpy(pmd->remote_iface, remote_iface, RTE_ETH_NAME_MAX_LEN);
1829
1830                 /* Save state of remote device */
1831                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1832
1833                 /* Replicate remote MAC address */
1834                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1835                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1836                                 pmd->name, pmd->remote_iface);
1837                         goto error_remote;
1838                 }
1839                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1840                            RTE_ETHER_ADDR_LEN);
1841                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1842                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1843                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1844                                 pmd->name, remote_iface);
1845                         goto error_remote;
1846                 }
1847
1848                 /*
1849                  * Flush usually returns negative value because it tries to
1850                  * delete every QDISC (and on a running device, one QDISC at
1851                  * least is needed). Ignore negative return value.
1852                  */
1853                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1854                 if (qdisc_create_ingress(pmd->nlsk_fd,
1855                                          pmd->remote_if_index) < 0) {
1856                         TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1857                                 pmd->remote_iface);
1858                         goto error_remote;
1859                 }
1860                 LIST_INIT(&pmd->implicit_flows);
1861                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1862                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1863                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1864                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1865                         TAP_LOG(ERR,
1866                                 "%s: failed to create implicit rules.",
1867                                 pmd->name);
1868                         goto error_remote;
1869                 }
1870         }
1871
1872         rte_eth_dev_probing_finish(dev);
1873         return 0;
1874
1875 disable_rte_flow:
1876         TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
1877                 strerror(errno), errno);
1878         if (strlen(remote_iface)) {
1879                 TAP_LOG(ERR, "Remote feature requires flow support.");
1880                 goto error_exit;
1881         }
1882         rte_eth_dev_probing_finish(dev);
1883         return 0;
1884
1885 error_remote:
1886         TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
1887                 strerror(errno), errno);
1888         tap_flow_implicit_flush(pmd, NULL);
1889
1890 error_exit:
1891         if (pmd->ioctl_sock > 0)
1892                 close(pmd->ioctl_sock);
1893         /* mac_addrs must not be freed alone because part of dev_private */
1894         dev->data->mac_addrs = NULL;
1895         rte_eth_dev_release_port(dev);
1896
1897 error_exit_nodev:
1898         TAP_LOG(ERR, "%s Unable to initialize %s",
1899                 tuntap_name, rte_vdev_device_name(vdev));
1900
1901         return -EINVAL;
1902 }
1903
1904 /* make sure name is a possible Linux network device name */
1905 static bool
1906 is_valid_iface(const char *name)
1907 {
1908         if (*name == '\0')
1909                 return false;
1910
1911         if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1912                 return false;
1913
1914         while (*name) {
1915                 if (*name == '/' || *name == ':' || isspace(*name))
1916                         return false;
1917                 name++;
1918         }
1919         return true;
1920 }
1921
1922 static int
1923 set_interface_name(const char *key __rte_unused,
1924                    const char *value,
1925                    void *extra_args)
1926 {
1927         char *name = (char *)extra_args;
1928
1929         if (value) {
1930                 if (!is_valid_iface(value)) {
1931                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
1932                                 value);
1933                         return -1;
1934                 }
1935                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
1936         } else {
1937                 /* use tap%d which causes kernel to choose next available */
1938                 strlcpy(name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
1939         }
1940         return 0;
1941 }
1942
1943 static int
1944 set_remote_iface(const char *key __rte_unused,
1945                  const char *value,
1946                  void *extra_args)
1947 {
1948         char *name = (char *)extra_args;
1949
1950         if (value) {
1951                 if (!is_valid_iface(value)) {
1952                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
1953                                 value);
1954                         return -1;
1955                 }
1956                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
1957         }
1958
1959         return 0;
1960 }
1961
1962 static int parse_user_mac(struct rte_ether_addr *user_mac,
1963                 const char *value)
1964 {
1965         unsigned int index = 0;
1966         char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
1967
1968         if (user_mac == NULL || value == NULL)
1969                 return 0;
1970
1971         strlcpy(mac_temp, value, sizeof(mac_temp));
1972         mac_byte = strtok(mac_temp, ":");
1973
1974         while ((mac_byte != NULL) &&
1975                         (strlen(mac_byte) <= 2) &&
1976                         (strlen(mac_byte) == strspn(mac_byte,
1977                                         ETH_TAP_CMP_MAC_FMT))) {
1978                 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
1979                 mac_byte = strtok(NULL, ":");
1980         }
1981
1982         return index;
1983 }
1984
1985 static int
1986 set_mac_type(const char *key __rte_unused,
1987              const char *value,
1988              void *extra_args)
1989 {
1990         struct rte_ether_addr *user_mac = extra_args;
1991
1992         if (!value)
1993                 return 0;
1994
1995         if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
1996                 static int iface_idx;
1997
1998                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1999                 memcpy((char *)user_mac->addr_bytes, "\0dtap",
2000                         RTE_ETHER_ADDR_LEN);
2001                 user_mac->addr_bytes[RTE_ETHER_ADDR_LEN - 1] =
2002                         iface_idx++ + '0';
2003                 goto success;
2004         }
2005
2006         if (parse_user_mac(user_mac, value) != 6)
2007                 goto error;
2008 success:
2009         TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
2010         return 0;
2011
2012 error:
2013         TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
2014                 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
2015         return -1;
2016 }
2017
2018 /*
2019  * Open a TUN interface device. TUN PMD
2020  * 1) sets tap_type as false
2021  * 2) intakes iface as argument.
2022  * 3) as interface is virtual set speed to 10G
2023  */
2024 static int
2025 rte_pmd_tun_probe(struct rte_vdev_device *dev)
2026 {
2027         const char *name, *params;
2028         int ret;
2029         struct rte_kvargs *kvlist = NULL;
2030         char tun_name[RTE_ETH_NAME_MAX_LEN];
2031         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2032         struct rte_eth_dev *eth_dev;
2033
2034         name = rte_vdev_device_name(dev);
2035         params = rte_vdev_device_args(dev);
2036         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2037
2038         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
2039             strlen(params) == 0) {
2040                 eth_dev = rte_eth_dev_attach_secondary(name);
2041                 if (!eth_dev) {
2042                         TAP_LOG(ERR, "Failed to probe %s", name);
2043                         return -1;
2044                 }
2045                 eth_dev->dev_ops = &ops;
2046                 eth_dev->device = &dev->device;
2047                 rte_eth_dev_probing_finish(eth_dev);
2048                 return 0;
2049         }
2050
2051         /* use tun%d which causes kernel to choose next available */
2052         strlcpy(tun_name, DEFAULT_TUN_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2053
2054         if (params && (params[0] != '\0')) {
2055                 TAP_LOG(DEBUG, "parameters (%s)", params);
2056
2057                 kvlist = rte_kvargs_parse(params, valid_arguments);
2058                 if (kvlist) {
2059                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2060                                 ret = rte_kvargs_process(kvlist,
2061                                         ETH_TAP_IFACE_ARG,
2062                                         &set_interface_name,
2063                                         tun_name);
2064
2065                                 if (ret == -1)
2066                                         goto leave;
2067                         }
2068                 }
2069         }
2070         pmd_link.link_speed = ETH_SPEED_NUM_10G;
2071
2072         TAP_LOG(DEBUG, "Initializing pmd_tun for %s", name);
2073
2074         ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0,
2075                                  ETH_TUNTAP_TYPE_TUN);
2076
2077 leave:
2078         if (ret == -1) {
2079                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2080                         name, tun_name);
2081         }
2082         rte_kvargs_free(kvlist);
2083
2084         return ret;
2085 }
2086
2087 /* Request queue file descriptors from secondary to primary. */
2088 static int
2089 tap_mp_attach_queues(const char *port_name, struct rte_eth_dev *dev)
2090 {
2091         int ret;
2092         struct timespec timeout = {.tv_sec = 1, .tv_nsec = 0};
2093         struct rte_mp_msg request, *reply;
2094         struct rte_mp_reply replies;
2095         struct ipc_queues *request_param = (struct ipc_queues *)request.param;
2096         struct ipc_queues *reply_param;
2097         struct pmd_process_private *process_private = dev->process_private;
2098         int queue, fd_iterator;
2099
2100         /* Prepare the request */
2101         memset(&request, 0, sizeof(request));
2102         strlcpy(request.name, TAP_MP_KEY, sizeof(request.name));
2103         strlcpy(request_param->port_name, port_name,
2104                 sizeof(request_param->port_name));
2105         request.len_param = sizeof(*request_param);
2106         /* Send request and receive reply */
2107         ret = rte_mp_request_sync(&request, &replies, &timeout);
2108         if (ret < 0 || replies.nb_received != 1) {
2109                 TAP_LOG(ERR, "Failed to request queues from primary: %d",
2110                         rte_errno);
2111                 return -1;
2112         }
2113         reply = &replies.msgs[0];
2114         reply_param = (struct ipc_queues *)reply->param;
2115         TAP_LOG(DEBUG, "Received IPC reply for %s", reply_param->port_name);
2116
2117         /* Attach the queues from received file descriptors */
2118         if (reply_param->rxq_count + reply_param->txq_count != reply->num_fds) {
2119                 TAP_LOG(ERR, "Unexpected number of fds received");
2120                 return -1;
2121         }
2122
2123         dev->data->nb_rx_queues = reply_param->rxq_count;
2124         dev->data->nb_tx_queues = reply_param->txq_count;
2125         fd_iterator = 0;
2126         for (queue = 0; queue < reply_param->rxq_count; queue++)
2127                 process_private->rxq_fds[queue] = reply->fds[fd_iterator++];
2128         for (queue = 0; queue < reply_param->txq_count; queue++)
2129                 process_private->txq_fds[queue] = reply->fds[fd_iterator++];
2130         free(reply);
2131         return 0;
2132 }
2133
2134 /* Send the queue file descriptors from the primary process to secondary. */
2135 static int
2136 tap_mp_sync_queues(const struct rte_mp_msg *request, const void *peer)
2137 {
2138         struct rte_eth_dev *dev;
2139         struct pmd_process_private *process_private;
2140         struct rte_mp_msg reply;
2141         const struct ipc_queues *request_param =
2142                 (const struct ipc_queues *)request->param;
2143         struct ipc_queues *reply_param =
2144                 (struct ipc_queues *)reply.param;
2145         uint16_t port_id;
2146         int queue;
2147         int ret;
2148
2149         /* Get requested port */
2150         TAP_LOG(DEBUG, "Received IPC request for %s", request_param->port_name);
2151         ret = rte_eth_dev_get_port_by_name(request_param->port_name, &port_id);
2152         if (ret) {
2153                 TAP_LOG(ERR, "Failed to get port id for %s",
2154                         request_param->port_name);
2155                 return -1;
2156         }
2157         dev = &rte_eth_devices[port_id];
2158         process_private = dev->process_private;
2159
2160         /* Fill file descriptors for all queues */
2161         reply.num_fds = 0;
2162         reply_param->rxq_count = 0;
2163         if (dev->data->nb_rx_queues + dev->data->nb_tx_queues >
2164                         RTE_MP_MAX_FD_NUM){
2165                 TAP_LOG(ERR, "Number of rx/tx queues exceeds max number of fds");
2166                 return -1;
2167         }
2168
2169         for (queue = 0; queue < dev->data->nb_rx_queues; queue++) {
2170                 reply.fds[reply.num_fds++] = process_private->rxq_fds[queue];
2171                 reply_param->rxq_count++;
2172         }
2173         RTE_ASSERT(reply_param->rxq_count == dev->data->nb_rx_queues);
2174
2175         reply_param->txq_count = 0;
2176         for (queue = 0; queue < dev->data->nb_tx_queues; queue++) {
2177                 reply.fds[reply.num_fds++] = process_private->txq_fds[queue];
2178                 reply_param->txq_count++;
2179         }
2180         RTE_ASSERT(reply_param->txq_count == dev->data->nb_tx_queues);
2181
2182         /* Send reply */
2183         strlcpy(reply.name, request->name, sizeof(reply.name));
2184         strlcpy(reply_param->port_name, request_param->port_name,
2185                 sizeof(reply_param->port_name));
2186         reply.len_param = sizeof(*reply_param);
2187         if (rte_mp_reply(&reply, peer) < 0) {
2188                 TAP_LOG(ERR, "Failed to reply an IPC request to sync queues");
2189                 return -1;
2190         }
2191         return 0;
2192 }
2193
2194 /* Open a TAP interface device.
2195  */
2196 static int
2197 rte_pmd_tap_probe(struct rte_vdev_device *dev)
2198 {
2199         const char *name, *params;
2200         int ret;
2201         struct rte_kvargs *kvlist = NULL;
2202         int speed;
2203         char tap_name[RTE_ETH_NAME_MAX_LEN];
2204         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2205         struct rte_ether_addr user_mac = { .addr_bytes = {0} };
2206         struct rte_eth_dev *eth_dev;
2207         int tap_devices_count_increased = 0;
2208
2209         name = rte_vdev_device_name(dev);
2210         params = rte_vdev_device_args(dev);
2211
2212         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
2213                 eth_dev = rte_eth_dev_attach_secondary(name);
2214                 if (!eth_dev) {
2215                         TAP_LOG(ERR, "Failed to probe %s", name);
2216                         return -1;
2217                 }
2218                 eth_dev->dev_ops = &ops;
2219                 eth_dev->device = &dev->device;
2220                 eth_dev->rx_pkt_burst = pmd_rx_burst;
2221                 eth_dev->tx_pkt_burst = pmd_tx_burst;
2222                 if (!rte_eal_primary_proc_alive(NULL)) {
2223                         TAP_LOG(ERR, "Primary process is missing");
2224                         return -1;
2225                 }
2226                 eth_dev->process_private = (struct pmd_process_private *)
2227                         rte_zmalloc_socket(name,
2228                                 sizeof(struct pmd_process_private),
2229                                 RTE_CACHE_LINE_SIZE,
2230                                 eth_dev->device->numa_node);
2231                 if (eth_dev->process_private == NULL) {
2232                         TAP_LOG(ERR,
2233                                 "Failed to alloc memory for process private");
2234                         return -1;
2235                 }
2236
2237                 ret = tap_mp_attach_queues(name, eth_dev);
2238                 if (ret != 0)
2239                         return -1;
2240                 rte_eth_dev_probing_finish(eth_dev);
2241                 return 0;
2242         }
2243
2244         speed = ETH_SPEED_NUM_10G;
2245
2246         /* use tap%d which causes kernel to choose next available */
2247         strlcpy(tap_name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2248         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2249
2250         if (params && (params[0] != '\0')) {
2251                 TAP_LOG(DEBUG, "parameters (%s)", params);
2252
2253                 kvlist = rte_kvargs_parse(params, valid_arguments);
2254                 if (kvlist) {
2255                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2256                                 ret = rte_kvargs_process(kvlist,
2257                                                          ETH_TAP_IFACE_ARG,
2258                                                          &set_interface_name,
2259                                                          tap_name);
2260                                 if (ret == -1)
2261                                         goto leave;
2262                         }
2263
2264                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
2265                                 ret = rte_kvargs_process(kvlist,
2266                                                          ETH_TAP_REMOTE_ARG,
2267                                                          &set_remote_iface,
2268                                                          remote_iface);
2269                                 if (ret == -1)
2270                                         goto leave;
2271                         }
2272
2273                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
2274                                 ret = rte_kvargs_process(kvlist,
2275                                                          ETH_TAP_MAC_ARG,
2276                                                          &set_mac_type,
2277                                                          &user_mac);
2278                                 if (ret == -1)
2279                                         goto leave;
2280                         }
2281                 }
2282         }
2283         pmd_link.link_speed = speed;
2284
2285         TAP_LOG(DEBUG, "Initializing pmd_tap for %s", name);
2286
2287         /* Register IPC feed callback */
2288         if (!tap_devices_count) {
2289                 ret = rte_mp_action_register(TAP_MP_KEY, tap_mp_sync_queues);
2290                 if (ret < 0) {
2291                         TAP_LOG(ERR, "tap: Failed to register IPC callback: %s",
2292                                 strerror(rte_errno));
2293                         goto leave;
2294                 }
2295         }
2296         tap_devices_count++;
2297         tap_devices_count_increased = 1;
2298         ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
2299                 ETH_TUNTAP_TYPE_TAP);
2300
2301 leave:
2302         if (ret == -1) {
2303                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2304                         name, tap_name);
2305                 if (tap_devices_count_increased == 1) {
2306                         if (tap_devices_count == 1)
2307                                 rte_mp_action_unregister(TAP_MP_KEY);
2308                         tap_devices_count--;
2309                 }
2310         }
2311         rte_kvargs_free(kvlist);
2312
2313         return ret;
2314 }
2315
2316 /* detach a TUNTAP device.
2317  */
2318 static int
2319 rte_pmd_tap_remove(struct rte_vdev_device *dev)
2320 {
2321         struct rte_eth_dev *eth_dev = NULL;
2322         struct pmd_internals *internals;
2323         struct pmd_process_private *process_private;
2324         int i;
2325
2326         /* find the ethdev entry */
2327         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
2328         if (!eth_dev)
2329                 return -ENODEV;
2330
2331         /* mac_addrs must not be freed alone because part of dev_private */
2332         eth_dev->data->mac_addrs = NULL;
2333
2334         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2335                 return rte_eth_dev_release_port(eth_dev);
2336
2337         internals = eth_dev->data->dev_private;
2338         process_private = eth_dev->process_private;
2339
2340         TAP_LOG(DEBUG, "Closing %s Ethernet device on numa %u",
2341                 tuntap_types[internals->type], rte_socket_id());
2342
2343         if (internals->nlsk_fd) {
2344                 tap_flow_flush(eth_dev, NULL);
2345                 tap_flow_implicit_flush(internals, NULL);
2346                 tap_nl_final(internals->nlsk_fd);
2347         }
2348         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
2349                 if (process_private->rxq_fds[i] != -1) {
2350                         close(process_private->rxq_fds[i]);
2351                         process_private->rxq_fds[i] = -1;
2352                 }
2353                 if (process_private->txq_fds[i] != -1) {
2354                         close(process_private->txq_fds[i]);
2355                         process_private->txq_fds[i] = -1;
2356                 }
2357         }
2358
2359         close(internals->ioctl_sock);
2360         rte_free(eth_dev->process_private);
2361         if (tap_devices_count == 1)
2362                 rte_mp_action_unregister(TAP_MP_KEY);
2363         tap_devices_count--;
2364         rte_eth_dev_release_port(eth_dev);
2365
2366         if (internals->ka_fd != -1) {
2367                 close(internals->ka_fd);
2368                 internals->ka_fd = -1;
2369         }
2370         return 0;
2371 }
2372
2373 static struct rte_vdev_driver pmd_tun_drv = {
2374         .probe = rte_pmd_tun_probe,
2375         .remove = rte_pmd_tap_remove,
2376 };
2377
2378 static struct rte_vdev_driver pmd_tap_drv = {
2379         .probe = rte_pmd_tap_probe,
2380         .remove = rte_pmd_tap_remove,
2381 };
2382
2383 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
2384 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
2385 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
2386 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
2387                               ETH_TAP_IFACE_ARG "=<string> ");
2388 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
2389                               ETH_TAP_IFACE_ARG "=<string> "
2390                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
2391                               ETH_TAP_REMOTE_ARG "=<string>");
2392 int tap_logtype;
2393
2394 RTE_INIT(tap_init_log)
2395 {
2396         tap_logtype = rte_log_register("pmd.net.tap");
2397         if (tap_logtype >= 0)
2398                 rte_log_set_level(tap_logtype, RTE_LOG_NOTICE);
2399 }