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 <sys/types.h>
  23 #include <sys/stat.h>
  24 #include <sys/socket.h>
  25 #include <sys/ioctl.h>
  26 #include <sys/utsname.h>
  27 #include <sys/mman.h>
  28 #include <errno.h>
  29 #include <signal.h>
  30 #include <stdbool.h>
  31 #include <stdint.h>
  32 #include <sys/uio.h>
  33 #include <unistd.h>
  34 #include <arpa/inet.h>
  35 #include <net/if.h>
  36 #include <linux/if_tun.h>
  37 #include <linux/if_ether.h>
  38 #include <fcntl.h>
  39 #include <ctype.h>
  40
  41 #include <tap_rss.h>
  42 #include <rte_eth_tap.h>
  43 #include <tap_flow.h>
  44 #include <tap_netlink.h>
  45 #include <tap_tcmsgs.h>
  46
  47 /* Linux based path to the TUN device */
  48 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  49 #define DEFAULT_TAP_NAME        "dtap"
  50 #define DEFAULT_TUN_NAME        "dtun"
  51
  52 #define ETH_TAP_IFACE_ARG       "iface"
  53 #define ETH_TAP_REMOTE_ARG      "remote"
  54 #define ETH_TAP_MAC_ARG         "mac"
  55 #define ETH_TAP_MAC_FIXED       "fixed"
  56
  57 #define ETH_TAP_USR_MAC_FMT     "xx:xx:xx:xx:xx:xx"
  58 #define ETH_TAP_CMP_MAC_FMT     "0123456789ABCDEFabcdef"
  59 #define ETH_TAP_MAC_ARG_FMT     ETH_TAP_MAC_FIXED "|" ETH_TAP_USR_MAC_FMT
  60
  61 #define TAP_GSO_MBUFS_PER_CORE  128
  62 #define TAP_GSO_MBUF_SEG_SIZE   128
  63 #define TAP_GSO_MBUF_CACHE_SIZE 4
  64 #define TAP_GSO_MBUFS_NUM \
  65         (TAP_GSO_MBUFS_PER_CORE * TAP_GSO_MBUF_CACHE_SIZE)
  66
  67 /* IPC key for queue fds sync */
  68 #define TAP_MP_KEY "tap_mp_sync_queues"
  69
  70 #define TAP_IOV_DEFAULT_MAX 1024
  71
  72 static int tap_devices_count;
  73
  74 static const char *valid_arguments[] = {
  75         ETH_TAP_IFACE_ARG,
  76         ETH_TAP_REMOTE_ARG,
  77         ETH_TAP_MAC_ARG,
  78         NULL
  79 };
  80
  81 static volatile uint32_t tap_trigger;   /* Rx trigger */
  82
  83 static struct rte_eth_link pmd_link = {
  84         .link_speed = ETH_SPEED_NUM_10G,
  85         .link_duplex = ETH_LINK_FULL_DUPLEX,
  86         .link_status = ETH_LINK_DOWN,
  87         .link_autoneg = ETH_LINK_FIXED,
  88 };
  89
  90 static void
  91 tap_trigger_cb(int sig __rte_unused)
  92 {
  93         /* Valid trigger values are nonzero */
  94         tap_trigger = (tap_trigger + 1) | 0x80000000;
  95 }
  96
  97 /* Specifies on what netdevices the ioctl should be applied */
  98 enum ioctl_mode {
  99         LOCAL_AND_REMOTE,
 100         LOCAL_ONLY,
 101         REMOTE_ONLY,
 102 };
 103
 104 /* Message header to synchronize queues via IPC */
 105 struct ipc_queues {
 106         char port_name[RTE_DEV_NAME_MAX_LEN];
 107         int rxq_count;
 108         int txq_count;
 109         /*
 110          * The file descriptors are in the dedicated part
 111          * of the Unix message to be translated by the kernel.
 112          */
 113 };
 114
 115 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
 116
 117 /**
 118  * Tun/Tap allocation routine
 119  *
 120  * @param[in] pmd
 121  *   Pointer to private structure.
 122  *
 123  * @param[in] is_keepalive
 124  *   Keepalive flag
 125  *
 126  * @return
 127  *   -1 on failure, fd on success
 128  */
 129 static int
 130 tun_alloc(struct pmd_internals *pmd, int is_keepalive)
 131 {
 132         struct ifreq ifr;
 133 #ifdef IFF_MULTI_QUEUE
 134         unsigned int features;
 135 #endif
 136         int fd;
 137
 138         memset(&ifr, 0, sizeof(struct ifreq));
 139
 140         /*
 141          * Do not set IFF_NO_PI as packet information header will be needed
 142          * to check if a received packet has been truncated.
 143          */
 144         ifr.ifr_flags = (pmd->type == ETH_TUNTAP_TYPE_TAP) ?
 145                 IFF_TAP : IFF_TUN | IFF_POINTOPOINT;
 146         strlcpy(ifr.ifr_name, pmd->name, IFNAMSIZ);
 147
 148         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 149         if (fd < 0) {
 150                 TAP_LOG(ERR, "Unable to open %s interface", TUN_TAP_DEV_PATH);
 151                 goto error;
 152         }
 153
 154 #ifdef IFF_MULTI_QUEUE
 155         /* Grab the TUN features to verify we can work multi-queue */
 156         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 157                 TAP_LOG(ERR, "unable to get TUN/TAP features");
 158                 goto error;
 159         }
 160         TAP_LOG(DEBUG, "%s Features %08x", TUN_TAP_DEV_PATH, features);
 161
 162         if (features & IFF_MULTI_QUEUE) {
 163                 TAP_LOG(DEBUG, "  Multi-queue support for %d queues",
 164                         RTE_PMD_TAP_MAX_QUEUES);
 165                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 166         } else
 167 #endif
 168         {
 169                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 170                 TAP_LOG(DEBUG, "  Single queue only support");
 171         }
 172
 173         /* Set the TUN/TAP configuration and set the name if needed */
 174         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 175                 TAP_LOG(WARNING, "Unable to set TUNSETIFF for %s: %s",
 176                         ifr.ifr_name, strerror(errno));
 177                 goto error;
 178         }
 179
 180         /*
 181          * Name passed to kernel might be wildcard like dtun%d
 182          * and need to find the resulting device.
 183          */
 184         TAP_LOG(DEBUG, "Device name is '%s'", ifr.ifr_name);
 185         strlcpy(pmd->name, ifr.ifr_name, RTE_ETH_NAME_MAX_LEN);
 186
 187         if (is_keepalive) {
 188                 /*
 189                  * Detach the TUN/TAP keep-alive queue
 190                  * to avoid traffic through it
 191                  */
 192                 ifr.ifr_flags = IFF_DETACH_QUEUE;
 193                 if (ioctl(fd, TUNSETQUEUE, (void *)&ifr) < 0) {
 194                         TAP_LOG(WARNING,
 195                                 "Unable to detach keep-alive queue for %s: %s",
 196                                 ifr.ifr_name, strerror(errno));
 197                         goto error;
 198                 }
 199         }
 200
 201         /* Always set the file descriptor to non-blocking */
 202         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 203                 TAP_LOG(WARNING,
 204                         "Unable to set %s to nonblocking: %s",
 205                         ifr.ifr_name, strerror(errno));
 206                 goto error;
 207         }
 208
 209         /* Set up trigger to optimize empty Rx bursts */
 210         errno = 0;
 211         do {
 212                 struct sigaction sa;
 213                 int flags = fcntl(fd, F_GETFL);
 214
 215                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 216                         break;
 217                 if (sa.sa_handler != tap_trigger_cb) {
 218                         /*
 219                          * Make sure SIGIO is not already taken. This is done
 220                          * as late as possible to leave the application a
 221                          * chance to set up its own signal handler first.
 222                          */
 223                         if (sa.sa_handler != SIG_IGN &&
 224                             sa.sa_handler != SIG_DFL) {
 225                                 errno = EBUSY;
 226                                 break;
 227                         }
 228                         sa = (struct sigaction){
 229                                 .sa_flags = SA_RESTART,
 230                                 .sa_handler = tap_trigger_cb,
 231                         };
 232                         if (sigaction(SIGIO, &sa, NULL) == -1)
 233                                 break;
 234                 }
 235                 /* Enable SIGIO on file descriptor */
 236                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 237                 fcntl(fd, F_SETOWN, getpid());
 238         } while (0);
 239
 240         if (errno) {
 241                 /* Disable trigger globally in case of error */
 242                 tap_trigger = 0;
 243                 TAP_LOG(WARNING, "Rx trigger disabled: %s",
 244                         strerror(errno));
 245         }
 246
 247         return fd;
 248
 249 error:
 250         if (fd >= 0)
 251                 close(fd);
 252         return -1;
 253 }
 254
 255 static void
 256 tap_verify_csum(struct rte_mbuf *mbuf)
 257 {
 258         uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
 259         uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
 260         uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
 261         unsigned int l2_len = sizeof(struct rte_ether_hdr);
 262         unsigned int l3_len;
 263         uint16_t cksum = 0;
 264         void *l3_hdr;
 265         void *l4_hdr;
 266
 267         if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
 268                 l2_len += 4;
 269         else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
 270                 l2_len += 8;
 271         /* Don't verify checksum for packets with discontinuous L2 header */
 272         if (unlikely(l2_len + sizeof(struct rte_ipv4_hdr) >
 273                      rte_pktmbuf_data_len(mbuf)))
 274                 return;
 275         l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
 276         if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 277                 struct rte_ipv4_hdr *iph = l3_hdr;
 278
 279                 /* ihl contains the number of 4-byte words in the header */
 280                 l3_len = 4 * (iph->version_ihl & 0xf);
 281                 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
 282                         return;
 283                 /* check that the total length reported by header is not
 284                  * greater than the total received size
 285                  */
 286                 if (l2_len + rte_be_to_cpu_16(iph->total_length) >
 287                                 rte_pktmbuf_data_len(mbuf))
 288                         return;
 289
 290                 cksum = ~rte_raw_cksum(iph, l3_len);
 291                 mbuf->ol_flags |= cksum ?
 292                         PKT_RX_IP_CKSUM_BAD :
 293                         PKT_RX_IP_CKSUM_GOOD;
 294         } else if (l3 == RTE_PTYPE_L3_IPV6) {
 295                 struct rte_ipv6_hdr *iph = l3_hdr;
 296
 297                 l3_len = sizeof(struct rte_ipv6_hdr);
 298                 /* check that the total length reported by header is not
 299                  * greater than the total received size
 300                  */
 301                 if (l2_len + l3_len + rte_be_to_cpu_16(iph->payload_len) >
 302                                 rte_pktmbuf_data_len(mbuf))
 303                         return;
 304         } else {
 305                 /* IPv6 extensions are not supported */
 306                 return;
 307         }
 308         if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
 309                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 310                 /* Don't verify checksum for multi-segment packets. */
 311                 if (mbuf->nb_segs > 1)
 312                         return;
 313                 if (l3 == RTE_PTYPE_L3_IPV4)
 314                         cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
 315                 else if (l3 == RTE_PTYPE_L3_IPV6)
 316                         cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
 317                 mbuf->ol_flags |= cksum ?
 318                         PKT_RX_L4_CKSUM_BAD :
 319                         PKT_RX_L4_CKSUM_GOOD;
 320         }
 321 }
 322
 323 static uint64_t
 324 tap_rx_offload_get_port_capa(void)
 325 {
 326         /*
 327          * No specific port Rx offload capabilities.
 328          */
 329         return 0;
 330 }
 331
 332 static uint64_t
 333 tap_rx_offload_get_queue_capa(void)
 334 {
 335         return DEV_RX_OFFLOAD_SCATTER |
 336                DEV_RX_OFFLOAD_IPV4_CKSUM |
 337                DEV_RX_OFFLOAD_UDP_CKSUM |
 338                DEV_RX_OFFLOAD_TCP_CKSUM;
 339 }
 340
 341 /* Callback to handle the rx burst of packets to the correct interface and
 342  * file descriptor(s) in a multi-queue setup.
 343  */
 344 static uint16_t
 345 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 346 {
 347         struct rx_queue *rxq = queue;
 348         struct pmd_process_private *process_private;
 349         uint16_t num_rx;
 350         unsigned long num_rx_bytes = 0;
 351         uint32_t trigger = tap_trigger;
 352
 353         if (trigger == rxq->trigger_seen)
 354                 return 0;
 355
 356         process_private = rte_eth_devices[rxq->in_port].process_private;
 357         for (num_rx = 0; num_rx < nb_pkts; ) {
 358                 struct rte_mbuf *mbuf = rxq->pool;
 359                 struct rte_mbuf *seg = NULL;
 360                 struct rte_mbuf *new_tail = NULL;
 361                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 362                 int len;
 363
 364                 len = readv(process_private->rxq_fds[rxq->queue_id],
 365                         *rxq->iovecs,
 366                         1 + (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
 367                              rxq->nb_rx_desc : 1));
 368                 if (len < (int)sizeof(struct tun_pi))
 369                         break;
 370
 371                 /* Packet couldn't fit in the provided mbuf */
 372                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 373                         rxq->stats.ierrors++;
 374                         continue;
 375                 }
 376
 377                 len -= sizeof(struct tun_pi);
 378
 379                 mbuf->pkt_len = len;
 380                 mbuf->port = rxq->in_port;
 381                 while (1) {
 382                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 383
 384                         if (unlikely(!buf)) {
 385                                 rxq->stats.rx_nombuf++;
 386                                 /* No new buf has been allocated: do nothing */
 387                                 if (!new_tail || !seg)
 388                                         goto end;
 389
 390                                 seg->next = NULL;
 391                                 rte_pktmbuf_free(mbuf);
 392
 393                                 goto end;
 394                         }
 395                         seg = seg ? seg->next : mbuf;
 396                         if (rxq->pool == mbuf)
 397                                 rxq->pool = buf;
 398                         if (new_tail)
 399                                 new_tail->next = buf;
 400                         new_tail = buf;
 401                         new_tail->next = seg->next;
 402
 403                         /* iovecs[0] is reserved for packet info (pi) */
 404                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 405                                 buf->buf_len - data_off;
 406                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 407                                 (char *)buf->buf_addr + data_off;
 408
 409                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 410                         seg->data_off = data_off;
 411
 412                         len -= seg->data_len;
 413                         if (len <= 0)
 414                                 break;
 415                         mbuf->nb_segs++;
 416                         /* First segment has headroom, not the others */
 417                         data_off = 0;
 418                 }
 419                 seg->next = NULL;
 420                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 421                                                       RTE_PTYPE_ALL_MASK);
 422                 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
 423                         tap_verify_csum(mbuf);
 424
 425                 /* account for the receive frame */
 426                 bufs[num_rx++] = mbuf;
 427                 num_rx_bytes += mbuf->pkt_len;
 428         }
 429 end:
 430         rxq->stats.ipackets += num_rx;
 431         rxq->stats.ibytes += num_rx_bytes;
 432
 433         if (trigger && num_rx < nb_pkts)
 434                 rxq->trigger_seen = trigger;
 435
 436         return num_rx;
 437 }
 438
 439 static uint64_t
 440 tap_tx_offload_get_port_capa(void)
 441 {
 442         /*
 443          * No specific port Tx offload capabilities.
 444          */
 445         return 0;
 446 }
 447
 448 static uint64_t
 449 tap_tx_offload_get_queue_capa(void)
 450 {
 451         return DEV_TX_OFFLOAD_MULTI_SEGS |
 452                DEV_TX_OFFLOAD_IPV4_CKSUM |
 453                DEV_TX_OFFLOAD_UDP_CKSUM |
 454                DEV_TX_OFFLOAD_TCP_CKSUM |
 455                DEV_TX_OFFLOAD_TCP_TSO;
 456 }
 457
 458 /* Finalize l4 checksum calculation */
 459 static void
 460 tap_tx_l4_cksum(uint16_t *l4_cksum, uint16_t l4_phdr_cksum,
 461                 uint32_t l4_raw_cksum)
 462 {
 463         if (l4_cksum) {
 464                 uint32_t cksum;
 465
 466                 cksum = __rte_raw_cksum_reduce(l4_raw_cksum);
 467                 cksum += l4_phdr_cksum;
 468
 469                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 470                 cksum = (~cksum) & 0xffff;
 471                 if (cksum == 0)
 472                         cksum = 0xffff;
 473                 *l4_cksum = cksum;
 474         }
 475 }
 476
 477 /* Accumaulate L4 raw checksums */
 478 static void
 479 tap_tx_l4_add_rcksum(char *l4_data, unsigned int l4_len, uint16_t *l4_cksum,
 480                         uint32_t *l4_raw_cksum)
 481 {
 482         if (l4_cksum == NULL)
 483                 return;
 484
 485         *l4_raw_cksum = __rte_raw_cksum(l4_data, l4_len, *l4_raw_cksum);
 486 }
 487
 488 /* L3 and L4 pseudo headers checksum offloads */
 489 static void
 490 tap_tx_l3_cksum(char *packet, uint64_t ol_flags, unsigned int l2_len,
 491                 unsigned int l3_len, unsigned int l4_len, uint16_t **l4_cksum,
 492                 uint16_t *l4_phdr_cksum, uint32_t *l4_raw_cksum)
 493 {
 494         void *l3_hdr = packet + l2_len;
 495
 496         if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
 497                 struct rte_ipv4_hdr *iph = l3_hdr;
 498                 uint16_t cksum;
 499
 500                 iph->hdr_checksum = 0;
 501                 cksum = rte_raw_cksum(iph, l3_len);
 502                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 503         }
 504         if (ol_flags & PKT_TX_L4_MASK) {
 505                 void *l4_hdr;
 506
 507                 l4_hdr = packet + l2_len + l3_len;
 508                 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
 509                         *l4_cksum = &((struct rte_udp_hdr *)l4_hdr)->dgram_cksum;
 510                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 511                         *l4_cksum = &((struct rte_tcp_hdr *)l4_hdr)->cksum;
 512                 else
 513                         return;
 514                 **l4_cksum = 0;
 515                 if (ol_flags & PKT_TX_IPV4)
 516                         *l4_phdr_cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 517                 else
 518                         *l4_phdr_cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 519                 *l4_raw_cksum = __rte_raw_cksum(l4_hdr, l4_len, 0);
 520         }
 521 }
 522
 523 static inline int
 524 tap_write_mbufs(struct tx_queue *txq, uint16_t num_mbufs,
 525                         struct rte_mbuf **pmbufs,
 526                         uint16_t *num_packets, unsigned long *num_tx_bytes)
 527 {
 528         int i;
 529         uint16_t l234_hlen;
 530         struct pmd_process_private *process_private;
 531
 532         process_private = rte_eth_devices[txq->out_port].process_private;
 533
 534         for (i = 0; i < num_mbufs; i++) {
 535                 struct rte_mbuf *mbuf = pmbufs[i];
 536                 struct iovec iovecs[mbuf->nb_segs + 2];
 537                 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
 538                 struct rte_mbuf *seg = mbuf;
 539                 char m_copy[mbuf->data_len];
 540                 int proto;
 541                 int n;
 542                 int j;
 543                 int k; /* current index in iovecs for copying segments */
 544                 uint16_t seg_len; /* length of first segment */
 545                 uint16_t nb_segs;
 546                 uint16_t *l4_cksum; /* l4 checksum (pseudo header + payload) */
 547                 uint32_t l4_raw_cksum = 0; /* TCP/UDP payload raw checksum */
 548                 uint16_t l4_phdr_cksum = 0; /* TCP/UDP pseudo header checksum */
 549                 uint16_t is_cksum = 0; /* in case cksum should be offloaded */
 550
 551                 l4_cksum = NULL;
 552                 if (txq->type == ETH_TUNTAP_TYPE_TUN) {
 553                         /*
 554                          * TUN and TAP are created with IFF_NO_PI disabled.
 555                          * For TUN PMD this mandatory as fields are used by
 556                          * Kernel tun.c to determine whether its IP or non IP
 557                          * packets.
 558                          *
 559                          * The logic fetches the first byte of data from mbuf
 560                          * then compares whether its v4 or v6. If first byte
 561                          * is 4 or 6, then protocol field is updated.
 562                          */
 563                         char *buff_data = rte_pktmbuf_mtod(seg, void *);
 564                         proto = (*buff_data & 0xf0);
 565                         pi.proto = (proto == 0x40) ?
 566                                 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4) :
 567                                 ((proto == 0x60) ?
 568                                         rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6) :
 569                                         0x00);
 570                 }
 571
 572                 k = 0;
 573                 iovecs[k].iov_base = &pi;
 574                 iovecs[k].iov_len = sizeof(pi);
 575                 k++;
 576
 577                 nb_segs = mbuf->nb_segs;
 578                 if (txq->csum &&
 579                     ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 580                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 581                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
 582                         is_cksum = 1;
 583
 584                         /* Support only packets with at least layer 4
 585                          * header included in the first segment
 586                          */
 587                         seg_len = rte_pktmbuf_data_len(mbuf);
 588                         l234_hlen = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
 589                         if (seg_len < l234_hlen)
 590                                 return -1;
 591
 592                         /* To change checksums, work on a * copy of l2, l3
 593                          * headers + l4 pseudo header
 594                          */
 595                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 596                                         l234_hlen);
 597                         tap_tx_l3_cksum(m_copy, mbuf->ol_flags,
 598                                        mbuf->l2_len, mbuf->l3_len, mbuf->l4_len,
 599                                        &l4_cksum, &l4_phdr_cksum,
 600                                        &l4_raw_cksum);
 601                         iovecs[k].iov_base = m_copy;
 602                         iovecs[k].iov_len = l234_hlen;
 603                         k++;
 604
 605                         /* Update next iovecs[] beyond l2, l3, l4 headers */
 606                         if (seg_len > l234_hlen) {
 607                                 iovecs[k].iov_len = seg_len - l234_hlen;
 608                                 iovecs[k].iov_base =
 609                                         rte_pktmbuf_mtod(seg, char *) +
 610                                                 l234_hlen;
 611                                 tap_tx_l4_add_rcksum(iovecs[k].iov_base,
 612                                         iovecs[k].iov_len, l4_cksum,
 613                                         &l4_raw_cksum);
 614                                 k++;
 615                                 nb_segs++;
 616                         }
 617                         seg = seg->next;
 618                 }
 619
 620                 for (j = k; j <= nb_segs; j++) {
 621                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 622                         iovecs[j].iov_base = rte_pktmbuf_mtod(seg, void *);
 623                         if (is_cksum)
 624                                 tap_tx_l4_add_rcksum(iovecs[j].iov_base,
 625                                         iovecs[j].iov_len, l4_cksum,
 626                                         &l4_raw_cksum);
 627                         seg = seg->next;
 628                 }
 629
 630                 if (is_cksum)
 631                         tap_tx_l4_cksum(l4_cksum, l4_phdr_cksum, l4_raw_cksum);
 632
 633                 /* copy the tx frame data */
 634                 n = writev(process_private->txq_fds[txq->queue_id], iovecs, j);
 635                 if (n <= 0)
 636                         return -1;
 637
 638                 (*num_packets)++;
 639                 (*num_tx_bytes) += rte_pktmbuf_pkt_len(mbuf);
 640         }
 641         return 0;
 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                 int num_tso_mbufs;
 668                 uint16_t hdrs_len;
 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                         /* TCP segmentation implies TCP checksum offload */
 676                         mbuf_in->ol_flags |= PKT_TX_TCP_CKSUM;
 677
 678                         /* gso size is calculated without RTE_ETHER_CRC_LEN */
 679                         hdrs_len = mbuf_in->l2_len + mbuf_in->l3_len +
 680                                         mbuf_in->l4_len;
 681                         tso_segsz = mbuf_in->tso_segsz + hdrs_len;
 682                         if (unlikely(tso_segsz == hdrs_len) ||
 683                                 tso_segsz > *txq->mtu) {
 684                                 txq->stats.errs++;
 685                                 break;
 686                         }
 687                         gso_ctx->gso_size = tso_segsz;
 688                         /* 'mbuf_in' packet to segment */
 689                         num_tso_mbufs = rte_gso_segment(mbuf_in,
 690                                 gso_ctx, /* gso control block */
 691                                 (struct rte_mbuf **)&gso_mbufs, /* out mbufs */
 692                                 RTE_DIM(gso_mbufs)); /* max tso mbufs */
 693
 694                         /* ret contains the number of new created mbufs */
 695                         if (num_tso_mbufs < 0)
 696                                 break;
 697
 698                         mbuf = gso_mbufs;
 699                         num_mbufs = num_tso_mbufs;
 700                 } else {
 701                         /* stats.errs will be incremented */
 702                         if (rte_pktmbuf_pkt_len(mbuf_in) > max_size)
 703                                 break;
 704
 705                         /* ret 0 indicates no new mbufs were created */
 706                         num_tso_mbufs = 0;
 707                         mbuf = &mbuf_in;
 708                         num_mbufs = 1;
 709                 }
 710
 711                 ret = tap_write_mbufs(txq, num_mbufs, mbuf,
 712                                 &num_packets, &num_tx_bytes);
 713                 if (ret == -1) {
 714                         txq->stats.errs++;
 715                         /* free tso mbufs */
 716                         if (num_tso_mbufs > 0)
 717                                 rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
 718                         break;
 719                 }
 720                 num_tx++;
 721                 /* free original mbuf */
 722                 rte_pktmbuf_free(mbuf_in);
 723                 /* free tso mbufs */
 724                 if (num_tso_mbufs > 0)
 725                         rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
 726         }
 727
 728         txq->stats.opackets += num_packets;
 729         txq->stats.errs += nb_pkts - num_tx;
 730         txq->stats.obytes += num_tx_bytes;
 731
 732         return num_tx;
 733 }
 734
 735 static const char *
 736 tap_ioctl_req2str(unsigned long request)
 737 {
 738         switch (request) {
 739         case SIOCSIFFLAGS:
 740                 return "SIOCSIFFLAGS";
 741         case SIOCGIFFLAGS:
 742                 return "SIOCGIFFLAGS";
 743         case SIOCGIFHWADDR:
 744                 return "SIOCGIFHWADDR";
 745         case SIOCSIFHWADDR:
 746                 return "SIOCSIFHWADDR";
 747         case SIOCSIFMTU:
 748                 return "SIOCSIFMTU";
 749         }
 750         return "UNKNOWN";
 751 }
 752
 753 static int
 754 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 755           struct ifreq *ifr, int set, enum ioctl_mode mode)
 756 {
 757         short req_flags = ifr->ifr_flags;
 758         int remote = pmd->remote_if_index &&
 759                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 760
 761         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 762                 return 0;
 763         /*
 764          * If there is a remote netdevice, apply ioctl on it, then apply it on
 765          * the tap netdevice.
 766          */
 767 apply:
 768         if (remote)
 769                 strlcpy(ifr->ifr_name, pmd->remote_iface, IFNAMSIZ);
 770         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 771                 strlcpy(ifr->ifr_name, pmd->name, IFNAMSIZ);
 772         switch (request) {
 773         case SIOCSIFFLAGS:
 774                 /* fetch current flags to leave other flags untouched */
 775                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 776                         goto error;
 777                 if (set)
 778                         ifr->ifr_flags |= req_flags;
 779                 else
 780                         ifr->ifr_flags &= ~req_flags;
 781                 break;
 782         case SIOCGIFFLAGS:
 783         case SIOCGIFHWADDR:
 784         case SIOCSIFHWADDR:
 785         case SIOCSIFMTU:
 786                 break;
 787         default:
 788                 TAP_LOG(WARNING, "%s: ioctl() called with wrong arg",
 789                         pmd->name);
 790                 return -EINVAL;
 791         }
 792         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 793                 goto error;
 794         if (remote-- && mode == LOCAL_AND_REMOTE)
 795                 goto apply;
 796         return 0;
 797
 798 error:
 799         TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
 800                 tap_ioctl_req2str(request), strerror(errno), errno);
 801         return -errno;
 802 }
 803
 804 static int
 805 tap_link_set_down(struct rte_eth_dev *dev)
 806 {
 807         struct pmd_internals *pmd = dev->data->dev_private;
 808         struct ifreq ifr = { .ifr_flags = IFF_UP };
 809
 810         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 811         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 812 }
 813
 814 static int
 815 tap_link_set_up(struct rte_eth_dev *dev)
 816 {
 817         struct pmd_internals *pmd = dev->data->dev_private;
 818         struct ifreq ifr = { .ifr_flags = IFF_UP };
 819
 820         dev->data->dev_link.link_status = ETH_LINK_UP;
 821         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 822 }
 823
 824 static int
 825 tap_dev_start(struct rte_eth_dev *dev)
 826 {
 827         int err, i;
 828
 829         err = tap_intr_handle_set(dev, 1);
 830         if (err)
 831                 return err;
 832
 833         err = tap_link_set_up(dev);
 834         if (err)
 835                 return err;
 836
 837         for (i = 0; i < dev->data->nb_tx_queues; i++)
 838                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 839         for (i = 0; i < dev->data->nb_rx_queues; i++)
 840                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 841
 842         return err;
 843 }
 844
 845 /* This function gets called when the current port gets stopped.
 846  */
 847 static void
 848 tap_dev_stop(struct rte_eth_dev *dev)
 849 {
 850         int i;
 851
 852         for (i = 0; i < dev->data->nb_tx_queues; i++)
 853                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 854         for (i = 0; i < dev->data->nb_rx_queues; i++)
 855                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 856
 857         tap_intr_handle_set(dev, 0);
 858         tap_link_set_down(dev);
 859 }
 860
 861 static int
 862 tap_dev_configure(struct rte_eth_dev *dev)
 863 {
 864         struct pmd_internals *pmd = dev->data->dev_private;
 865
 866         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 867                 TAP_LOG(ERR,
 868                         "%s: number of rx queues %d exceeds max num of queues %d",
 869                         dev->device->name,
 870                         dev->data->nb_rx_queues,
 871                         RTE_PMD_TAP_MAX_QUEUES);
 872                 return -1;
 873         }
 874         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 875                 TAP_LOG(ERR,
 876                         "%s: number of tx queues %d exceeds max num of queues %d",
 877                         dev->device->name,
 878                         dev->data->nb_tx_queues,
 879                         RTE_PMD_TAP_MAX_QUEUES);
 880                 return -1;
 881         }
 882
 883         TAP_LOG(INFO, "%s: %s: TX configured queues number: %u",
 884                 dev->device->name, pmd->name, dev->data->nb_tx_queues);
 885
 886         TAP_LOG(INFO, "%s: %s: RX configured queues number: %u",
 887                 dev->device->name, pmd->name, dev->data->nb_rx_queues);
 888
 889         return 0;
 890 }
 891
 892 static uint32_t
 893 tap_dev_speed_capa(void)
 894 {
 895         uint32_t speed = pmd_link.link_speed;
 896         uint32_t capa = 0;
 897
 898         if (speed >= ETH_SPEED_NUM_10M)
 899                 capa |= ETH_LINK_SPEED_10M;
 900         if (speed >= ETH_SPEED_NUM_100M)
 901                 capa |= ETH_LINK_SPEED_100M;
 902         if (speed >= ETH_SPEED_NUM_1G)
 903                 capa |= ETH_LINK_SPEED_1G;
 904         if (speed >= ETH_SPEED_NUM_5G)
 905                 capa |= ETH_LINK_SPEED_2_5G;
 906         if (speed >= ETH_SPEED_NUM_5G)
 907                 capa |= ETH_LINK_SPEED_5G;
 908         if (speed >= ETH_SPEED_NUM_10G)
 909                 capa |= ETH_LINK_SPEED_10G;
 910         if (speed >= ETH_SPEED_NUM_20G)
 911                 capa |= ETH_LINK_SPEED_20G;
 912         if (speed >= ETH_SPEED_NUM_25G)
 913                 capa |= ETH_LINK_SPEED_25G;
 914         if (speed >= ETH_SPEED_NUM_40G)
 915                 capa |= ETH_LINK_SPEED_40G;
 916         if (speed >= ETH_SPEED_NUM_50G)
 917                 capa |= ETH_LINK_SPEED_50G;
 918         if (speed >= ETH_SPEED_NUM_56G)
 919                 capa |= ETH_LINK_SPEED_56G;
 920         if (speed >= ETH_SPEED_NUM_100G)
 921                 capa |= ETH_LINK_SPEED_100G;
 922
 923         return capa;
 924 }
 925
 926 static int
 927 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 928 {
 929         struct pmd_internals *internals = dev->data->dev_private;
 930
 931         dev_info->if_index = internals->if_index;
 932         dev_info->max_mac_addrs = 1;
 933         dev_info->max_rx_pktlen = (uint32_t)RTE_ETHER_MAX_VLAN_FRAME_LEN;
 934         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
 935         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
 936         dev_info->min_rx_bufsize = 0;
 937         dev_info->speed_capa = tap_dev_speed_capa();
 938         dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
 939         dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
 940                                     dev_info->rx_queue_offload_capa;
 941         dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
 942         dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
 943                                     dev_info->tx_queue_offload_capa;
 944         dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
 945         /*
 946          * limitation: TAP supports all of IP, UDP and TCP hash
 947          * functions together and not in partial combinations
 948          */
 949         dev_info->flow_type_rss_offloads = ~TAP_RSS_HF_MASK;
 950
 951         return 0;
 952 }
 953
 954 static int
 955 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 956 {
 957         unsigned int i, imax;
 958         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 959         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 960         unsigned long rx_nombuf = 0, ierrors = 0;
 961         const struct pmd_internals *pmd = dev->data->dev_private;
 962
 963         /* rx queue statistics */
 964         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 965                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 966         for (i = 0; i < imax; i++) {
 967                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 968                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 969                 rx_total += tap_stats->q_ipackets[i];
 970                 rx_bytes_total += tap_stats->q_ibytes[i];
 971                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
 972                 ierrors += pmd->rxq[i].stats.ierrors;
 973         }
 974
 975         /* tx queue statistics */
 976         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 977                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 978
 979         for (i = 0; i < imax; i++) {
 980                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 981                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 982                 tx_total += tap_stats->q_opackets[i];
 983                 tx_err_total += pmd->txq[i].stats.errs;
 984                 tx_bytes_total += tap_stats->q_obytes[i];
 985         }
 986
 987         tap_stats->ipackets = rx_total;
 988         tap_stats->ibytes = rx_bytes_total;
 989         tap_stats->ierrors = ierrors;
 990         tap_stats->rx_nombuf = rx_nombuf;
 991         tap_stats->opackets = tx_total;
 992         tap_stats->oerrors = tx_err_total;
 993         tap_stats->obytes = tx_bytes_total;
 994         return 0;
 995 }
 996
 997 static int
 998 tap_stats_reset(struct rte_eth_dev *dev)
 999 {
1000         int i;
1001         struct pmd_internals *pmd = dev->data->dev_private;
1002
1003         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1004                 pmd->rxq[i].stats.ipackets = 0;
1005                 pmd->rxq[i].stats.ibytes = 0;
1006                 pmd->rxq[i].stats.ierrors = 0;
1007                 pmd->rxq[i].stats.rx_nombuf = 0;
1008
1009                 pmd->txq[i].stats.opackets = 0;
1010                 pmd->txq[i].stats.errs = 0;
1011                 pmd->txq[i].stats.obytes = 0;
1012         }
1013
1014         return 0;
1015 }
1016
1017 static void
1018 tap_dev_close(struct rte_eth_dev *dev)
1019 {
1020         int i;
1021         struct pmd_internals *internals = dev->data->dev_private;
1022         struct pmd_process_private *process_private = dev->process_private;
1023         struct rx_queue *rxq;
1024
1025         tap_link_set_down(dev);
1026         if (internals->nlsk_fd != -1) {
1027                 tap_flow_flush(dev, NULL);
1028                 tap_flow_implicit_flush(internals, NULL);
1029                 tap_nl_final(internals->nlsk_fd);
1030                 internals->nlsk_fd = -1;
1031         }
1032
1033         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1034                 if (process_private->rxq_fds[i] != -1) {
1035                         rxq = &internals->rxq[i];
1036                         close(process_private->rxq_fds[i]);
1037                         process_private->rxq_fds[i] = -1;
1038                         rte_pktmbuf_free(rxq->pool);
1039                         rte_free(rxq->iovecs);
1040                         rxq->pool = NULL;
1041                         rxq->iovecs = NULL;
1042                 }
1043                 if (process_private->txq_fds[i] != -1) {
1044                         close(process_private->txq_fds[i]);
1045                         process_private->txq_fds[i] = -1;
1046                 }
1047         }
1048
1049         if (internals->remote_if_index) {
1050                 /* Restore initial remote state */
1051                 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
1052                                 &internals->remote_initial_flags);
1053         }
1054
1055         if (internals->ka_fd != -1) {
1056                 close(internals->ka_fd);
1057                 internals->ka_fd = -1;
1058         }
1059         /*
1060          * Since TUN device has no more opened file descriptors
1061          * it will be removed from kernel
1062          */
1063 }
1064
1065 static void
1066 tap_rx_queue_release(void *queue)
1067 {
1068         struct rx_queue *rxq = queue;
1069         struct pmd_process_private *process_private;
1070
1071         if (!rxq)
1072                 return;
1073         process_private = rte_eth_devices[rxq->in_port].process_private;
1074         if (process_private->rxq_fds[rxq->queue_id] > 0) {
1075                 close(process_private->rxq_fds[rxq->queue_id]);
1076                 process_private->rxq_fds[rxq->queue_id] = -1;
1077                 rte_pktmbuf_free(rxq->pool);
1078                 rte_free(rxq->iovecs);
1079                 rxq->pool = NULL;
1080                 rxq->iovecs = NULL;
1081         }
1082 }
1083
1084 static void
1085 tap_tx_queue_release(void *queue)
1086 {
1087         struct tx_queue *txq = queue;
1088         struct pmd_process_private *process_private;
1089
1090         if (!txq)
1091                 return;
1092         process_private = rte_eth_devices[txq->out_port].process_private;
1093
1094         if (process_private->txq_fds[txq->queue_id] > 0) {
1095                 close(process_private->txq_fds[txq->queue_id]);
1096                 process_private->txq_fds[txq->queue_id] = -1;
1097         }
1098 }
1099
1100 static int
1101 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1102 {
1103         struct rte_eth_link *dev_link = &dev->data->dev_link;
1104         struct pmd_internals *pmd = dev->data->dev_private;
1105         struct ifreq ifr = { .ifr_flags = 0 };
1106
1107         if (pmd->remote_if_index) {
1108                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
1109                 if (!(ifr.ifr_flags & IFF_UP) ||
1110                     !(ifr.ifr_flags & IFF_RUNNING)) {
1111                         dev_link->link_status = ETH_LINK_DOWN;
1112                         return 0;
1113                 }
1114         }
1115         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
1116         dev_link->link_status =
1117                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
1118                  ETH_LINK_UP :
1119                  ETH_LINK_DOWN);
1120         return 0;
1121 }
1122
1123 static int
1124 tap_promisc_enable(struct rte_eth_dev *dev)
1125 {
1126         struct pmd_internals *pmd = dev->data->dev_private;
1127         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1128         int ret;
1129
1130         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1131         if (ret != 0)
1132                 return ret;
1133
1134         if (pmd->remote_if_index && !pmd->flow_isolate) {
1135                 dev->data->promiscuous = 1;
1136                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
1137                 if (ret != 0) {
1138                         /* Rollback promisc flag */
1139                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1140                         /*
1141                          * rte_eth_dev_promiscuous_enable() rollback
1142                          * dev->data->promiscuous in the case of failure.
1143                          */
1144                         return ret;
1145                 }
1146         }
1147
1148         return 0;
1149 }
1150
1151 static int
1152 tap_promisc_disable(struct rte_eth_dev *dev)
1153 {
1154         struct pmd_internals *pmd = dev->data->dev_private;
1155         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1156         int ret;
1157
1158         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1159         if (ret != 0)
1160                 return ret;
1161
1162         if (pmd->remote_if_index && !pmd->flow_isolate) {
1163                 dev->data->promiscuous = 0;
1164                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
1165                 if (ret != 0) {
1166                         /* Rollback promisc flag */
1167                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1168                         /*
1169                          * rte_eth_dev_promiscuous_disable() rollback
1170                          * dev->data->promiscuous in the case of failure.
1171                          */
1172                         return ret;
1173                 }
1174         }
1175
1176         return 0;
1177 }
1178
1179 static int
1180 tap_allmulti_enable(struct rte_eth_dev *dev)
1181 {
1182         struct pmd_internals *pmd = dev->data->dev_private;
1183         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1184         int ret;
1185
1186         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1187         if (ret != 0)
1188                 return ret;
1189
1190         if (pmd->remote_if_index && !pmd->flow_isolate) {
1191                 dev->data->all_multicast = 1;
1192                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
1193                 if (ret != 0) {
1194                         /* Rollback allmulti flag */
1195                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1196                         /*
1197                          * rte_eth_dev_allmulticast_enable() rollback
1198                          * dev->data->all_multicast in the case of failure.
1199                          */
1200                         return ret;
1201                 }
1202         }
1203
1204         return 0;
1205 }
1206
1207 static int
1208 tap_allmulti_disable(struct rte_eth_dev *dev)
1209 {
1210         struct pmd_internals *pmd = dev->data->dev_private;
1211         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1212         int ret;
1213
1214         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1215         if (ret != 0)
1216                 return ret;
1217
1218         if (pmd->remote_if_index && !pmd->flow_isolate) {
1219                 dev->data->all_multicast = 0;
1220                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
1221                 if (ret != 0) {
1222                         /* Rollback allmulti flag */
1223                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1224                         /*
1225                          * rte_eth_dev_allmulticast_disable() rollback
1226                          * dev->data->all_multicast in the case of failure.
1227                          */
1228                         return ret;
1229                 }
1230         }
1231
1232         return 0;
1233 }
1234
1235 static int
1236 tap_mac_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1237 {
1238         struct pmd_internals *pmd = dev->data->dev_private;
1239         enum ioctl_mode mode = LOCAL_ONLY;
1240         struct ifreq ifr;
1241         int ret;
1242
1243         if (pmd->type == ETH_TUNTAP_TYPE_TUN) {
1244                 TAP_LOG(ERR, "%s: can't MAC address for TUN",
1245                         dev->device->name);
1246                 return -ENOTSUP;
1247         }
1248
1249         if (rte_is_zero_ether_addr(mac_addr)) {
1250                 TAP_LOG(ERR, "%s: can't set an empty MAC address",
1251                         dev->device->name);
1252                 return -EINVAL;
1253         }
1254         /* Check the actual current MAC address on the tap netdevice */
1255         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
1256         if (ret < 0)
1257                 return ret;
1258         if (rte_is_same_ether_addr(
1259                         (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1260                         mac_addr))
1261                 return 0;
1262         /* Check the current MAC address on the remote */
1263         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
1264         if (ret < 0)
1265                 return ret;
1266         if (!rte_is_same_ether_addr(
1267                         (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1268                         mac_addr))
1269                 mode = LOCAL_AND_REMOTE;
1270         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1271         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, RTE_ETHER_ADDR_LEN);
1272         ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
1273         if (ret < 0)
1274                 return ret;
1275         rte_memcpy(&pmd->eth_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1276         if (pmd->remote_if_index && !pmd->flow_isolate) {
1277                 /* Replace MAC redirection rule after a MAC change */
1278                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
1279                 if (ret < 0) {
1280                         TAP_LOG(ERR,
1281                                 "%s: Couldn't delete MAC redirection rule",
1282                                 dev->device->name);
1283                         return ret;
1284                 }
1285                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1286                 if (ret < 0) {
1287                         TAP_LOG(ERR,
1288                                 "%s: Couldn't add MAC redirection rule",
1289                                 dev->device->name);
1290                         return ret;
1291                 }
1292         }
1293
1294         return 0;
1295 }
1296
1297 static int
1298 tap_gso_ctx_setup(struct rte_gso_ctx *gso_ctx, struct rte_eth_dev *dev)
1299 {
1300         uint32_t gso_types;
1301         char pool_name[64];
1302
1303         /*
1304          * Create private mbuf pool with TAP_GSO_MBUF_SEG_SIZE bytes
1305          * size per mbuf use this pool for both direct and indirect mbufs
1306          */
1307
1308         struct rte_mempool *mp;      /* Mempool for GSO packets */
1309
1310         /* initialize GSO context */
1311         gso_types = DEV_TX_OFFLOAD_TCP_TSO;
1312         snprintf(pool_name, sizeof(pool_name), "mp_%s", dev->device->name);
1313         mp = rte_mempool_lookup((const char *)pool_name);
1314         if (!mp) {
1315                 mp = rte_pktmbuf_pool_create(pool_name, TAP_GSO_MBUFS_NUM,
1316                         TAP_GSO_MBUF_CACHE_SIZE, 0,
1317                         RTE_PKTMBUF_HEADROOM + TAP_GSO_MBUF_SEG_SIZE,
1318                         SOCKET_ID_ANY);
1319                 if (!mp) {
1320                         struct pmd_internals *pmd = dev->data->dev_private;
1321
1322                         TAP_LOG(ERR,
1323                                 "%s: failed to create mbuf pool for device %s\n",
1324                                 pmd->name, dev->device->name);
1325                         return -1;
1326                 }
1327         }
1328
1329         gso_ctx->direct_pool = mp;
1330         gso_ctx->indirect_pool = mp;
1331         gso_ctx->gso_types = gso_types;
1332         gso_ctx->gso_size = 0; /* gso_size is set in tx_burst() per packet */
1333         gso_ctx->flag = 0;
1334
1335         return 0;
1336 }
1337
1338 static int
1339 tap_setup_queue(struct rte_eth_dev *dev,
1340                 struct pmd_internals *internals,
1341                 uint16_t qid,
1342                 int is_rx)
1343 {
1344         int ret;
1345         int *fd;
1346         int *other_fd;
1347         const char *dir;
1348         struct pmd_internals *pmd = dev->data->dev_private;
1349         struct pmd_process_private *process_private = dev->process_private;
1350         struct rx_queue *rx = &internals->rxq[qid];
1351         struct tx_queue *tx = &internals->txq[qid];
1352         struct rte_gso_ctx *gso_ctx;
1353
1354         if (is_rx) {
1355                 fd = &process_private->rxq_fds[qid];
1356                 other_fd = &process_private->txq_fds[qid];
1357                 dir = "rx";
1358                 gso_ctx = NULL;
1359         } else {
1360                 fd = &process_private->txq_fds[qid];
1361                 other_fd = &process_private->rxq_fds[qid];
1362                 dir = "tx";
1363                 gso_ctx = &tx->gso_ctx;
1364         }
1365         if (*fd != -1) {
1366                 /* fd for this queue already exists */
1367                 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1368                         pmd->name, *fd, dir, qid);
1369                 gso_ctx = NULL;
1370         } else if (*other_fd != -1) {
1371                 /* Only other_fd exists. dup it */
1372                 *fd = dup(*other_fd);
1373                 if (*fd < 0) {
1374                         *fd = -1;
1375                         TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1376                         return -1;
1377                 }
1378                 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1379                         pmd->name, *other_fd, dir, qid, *fd);
1380         } else {
1381                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1382                 *fd = tun_alloc(pmd, 0);
1383                 if (*fd < 0) {
1384                         *fd = -1; /* restore original value */
1385                         TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1386                         return -1;
1387                 }
1388                 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1389                         pmd->name, dir, qid, *fd);
1390         }
1391
1392         tx->mtu = &dev->data->mtu;
1393         rx->rxmode = &dev->data->dev_conf.rxmode;
1394         if (gso_ctx) {
1395                 ret = tap_gso_ctx_setup(gso_ctx, dev);
1396                 if (ret)
1397                         return -1;
1398         }
1399
1400         tx->type = pmd->type;
1401
1402         return *fd;
1403 }
1404
1405 static int
1406 tap_rx_queue_setup(struct rte_eth_dev *dev,
1407                    uint16_t rx_queue_id,
1408                    uint16_t nb_rx_desc,
1409                    unsigned int socket_id,
1410                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1411                    struct rte_mempool *mp)
1412 {
1413         struct pmd_internals *internals = dev->data->dev_private;
1414         struct pmd_process_private *process_private = dev->process_private;
1415         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1416         struct rte_mbuf **tmp = &rxq->pool;
1417         long iov_max = sysconf(_SC_IOV_MAX);
1418
1419         if (iov_max <= 0) {
1420                 TAP_LOG(WARNING,
1421                         "_SC_IOV_MAX is not defined. Using %d as default",
1422                         TAP_IOV_DEFAULT_MAX);
1423                 iov_max = TAP_IOV_DEFAULT_MAX;
1424         }
1425         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1426         struct iovec (*iovecs)[nb_desc + 1];
1427         int data_off = RTE_PKTMBUF_HEADROOM;
1428         int ret = 0;
1429         int fd;
1430         int i;
1431
1432         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1433                 TAP_LOG(WARNING,
1434                         "nb_rx_queues %d too small or mempool NULL",
1435                         dev->data->nb_rx_queues);
1436                 return -1;
1437         }
1438
1439         rxq->mp = mp;
1440         rxq->trigger_seen = 1; /* force initial burst */
1441         rxq->in_port = dev->data->port_id;
1442         rxq->queue_id = rx_queue_id;
1443         rxq->nb_rx_desc = nb_desc;
1444         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1445                                     socket_id);
1446         if (!iovecs) {
1447                 TAP_LOG(WARNING,
1448                         "%s: Couldn't allocate %d RX descriptors",
1449                         dev->device->name, nb_desc);
1450                 return -ENOMEM;
1451         }
1452         rxq->iovecs = iovecs;
1453
1454         dev->data->rx_queues[rx_queue_id] = rxq;
1455         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1456         if (fd == -1) {
1457                 ret = fd;
1458                 goto error;
1459         }
1460
1461         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1462         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1463
1464         for (i = 1; i <= nb_desc; i++) {
1465                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1466                 if (!*tmp) {
1467                         TAP_LOG(WARNING,
1468                                 "%s: couldn't allocate memory for queue %d",
1469                                 dev->device->name, rx_queue_id);
1470                         ret = -ENOMEM;
1471                         goto error;
1472                 }
1473                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1474                 (*rxq->iovecs)[i].iov_base =
1475                         (char *)(*tmp)->buf_addr + data_off;
1476                 data_off = 0;
1477                 tmp = &(*tmp)->next;
1478         }
1479
1480         TAP_LOG(DEBUG, "  RX TUNTAP device name %s, qid %d on fd %d",
1481                 internals->name, rx_queue_id,
1482                 process_private->rxq_fds[rx_queue_id]);
1483
1484         return 0;
1485
1486 error:
1487         rte_pktmbuf_free(rxq->pool);
1488         rxq->pool = NULL;
1489         rte_free(rxq->iovecs);
1490         rxq->iovecs = NULL;
1491         return ret;
1492 }
1493
1494 static int
1495 tap_tx_queue_setup(struct rte_eth_dev *dev,
1496                    uint16_t tx_queue_id,
1497                    uint16_t nb_tx_desc __rte_unused,
1498                    unsigned int socket_id __rte_unused,
1499                    const struct rte_eth_txconf *tx_conf)
1500 {
1501         struct pmd_internals *internals = dev->data->dev_private;
1502         struct pmd_process_private *process_private = dev->process_private;
1503         struct tx_queue *txq;
1504         int ret;
1505         uint64_t offloads;
1506
1507         if (tx_queue_id >= dev->data->nb_tx_queues)
1508                 return -1;
1509         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1510         txq = dev->data->tx_queues[tx_queue_id];
1511         txq->out_port = dev->data->port_id;
1512         txq->queue_id = tx_queue_id;
1513
1514         offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1515         txq->csum = !!(offloads &
1516                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
1517                          DEV_TX_OFFLOAD_UDP_CKSUM |
1518                          DEV_TX_OFFLOAD_TCP_CKSUM));
1519
1520         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1521         if (ret == -1)
1522                 return -1;
1523         TAP_LOG(DEBUG,
1524                 "  TX TUNTAP device name %s, qid %d on fd %d csum %s",
1525                 internals->name, tx_queue_id,
1526                 process_private->txq_fds[tx_queue_id],
1527                 txq->csum ? "on" : "off");
1528
1529         return 0;
1530 }
1531
1532 static int
1533 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1534 {
1535         struct pmd_internals *pmd = dev->data->dev_private;
1536         struct ifreq ifr = { .ifr_mtu = mtu };
1537         int err = 0;
1538
1539         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1540         if (!err)
1541                 dev->data->mtu = mtu;
1542
1543         return err;
1544 }
1545
1546 static int
1547 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1548                      struct rte_ether_addr *mc_addr_set __rte_unused,
1549                      uint32_t nb_mc_addr __rte_unused)
1550 {
1551         /*
1552          * Nothing to do actually: the tap has no filtering whatsoever, every
1553          * packet is received.
1554          */
1555         return 0;
1556 }
1557
1558 static int
1559 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1560 {
1561         struct rte_eth_dev *dev = arg;
1562         struct pmd_internals *pmd = dev->data->dev_private;
1563         struct ifinfomsg *info = NLMSG_DATA(nh);
1564
1565         if (nh->nlmsg_type != RTM_NEWLINK ||
1566             (info->ifi_index != pmd->if_index &&
1567              info->ifi_index != pmd->remote_if_index))
1568                 return 0;
1569         return tap_link_update(dev, 0);
1570 }
1571
1572 static void
1573 tap_dev_intr_handler(void *cb_arg)
1574 {
1575         struct rte_eth_dev *dev = cb_arg;
1576         struct pmd_internals *pmd = dev->data->dev_private;
1577
1578         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1579 }
1580
1581 static int
1582 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1583 {
1584         struct pmd_internals *pmd = dev->data->dev_private;
1585
1586         /* In any case, disable interrupt if the conf is no longer there. */
1587         if (!dev->data->dev_conf.intr_conf.lsc) {
1588                 if (pmd->intr_handle.fd != -1) {
1589                         tap_nl_final(pmd->intr_handle.fd);
1590                         rte_intr_callback_unregister(&pmd->intr_handle,
1591                                 tap_dev_intr_handler, dev);
1592                 }
1593                 return 0;
1594         }
1595         if (set) {
1596                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1597                 if (unlikely(pmd->intr_handle.fd == -1))
1598                         return -EBADF;
1599                 return rte_intr_callback_register(
1600                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1601         }
1602         tap_nl_final(pmd->intr_handle.fd);
1603         return rte_intr_callback_unregister(&pmd->intr_handle,
1604                                             tap_dev_intr_handler, dev);
1605 }
1606
1607 static int
1608 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1609 {
1610         int err;
1611
1612         err = tap_lsc_intr_handle_set(dev, set);
1613         if (err < 0) {
1614                 if (!set)
1615                         tap_rx_intr_vec_set(dev, 0);
1616                 return err;
1617         }
1618         err = tap_rx_intr_vec_set(dev, set);
1619         if (err && set)
1620                 tap_lsc_intr_handle_set(dev, 0);
1621         return err;
1622 }
1623
1624 static const uint32_t*
1625 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1626 {
1627         static const uint32_t ptypes[] = {
1628                 RTE_PTYPE_INNER_L2_ETHER,
1629                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1630                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1631                 RTE_PTYPE_INNER_L3_IPV4,
1632                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1633                 RTE_PTYPE_INNER_L3_IPV6,
1634                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1635                 RTE_PTYPE_INNER_L4_FRAG,
1636                 RTE_PTYPE_INNER_L4_UDP,
1637                 RTE_PTYPE_INNER_L4_TCP,
1638                 RTE_PTYPE_INNER_L4_SCTP,
1639                 RTE_PTYPE_L2_ETHER,
1640                 RTE_PTYPE_L2_ETHER_VLAN,
1641                 RTE_PTYPE_L2_ETHER_QINQ,
1642                 RTE_PTYPE_L3_IPV4,
1643                 RTE_PTYPE_L3_IPV4_EXT,
1644                 RTE_PTYPE_L3_IPV6_EXT,
1645                 RTE_PTYPE_L3_IPV6,
1646                 RTE_PTYPE_L4_FRAG,
1647                 RTE_PTYPE_L4_UDP,
1648                 RTE_PTYPE_L4_TCP,
1649                 RTE_PTYPE_L4_SCTP,
1650         };
1651
1652         return ptypes;
1653 }
1654
1655 static int
1656 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1657                   struct rte_eth_fc_conf *fc_conf)
1658 {
1659         fc_conf->mode = RTE_FC_NONE;
1660         return 0;
1661 }
1662
1663 static int
1664 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1665                   struct rte_eth_fc_conf *fc_conf)
1666 {
1667         if (fc_conf->mode != RTE_FC_NONE)
1668                 return -ENOTSUP;
1669         return 0;
1670 }
1671
1672 /**
1673  * DPDK callback to update the RSS hash configuration.
1674  *
1675  * @param dev
1676  *   Pointer to Ethernet device structure.
1677  * @param[in] rss_conf
1678  *   RSS configuration data.
1679  *
1680  * @return
1681  *   0 on success, a negative errno value otherwise and rte_errno is set.
1682  */
1683 static int
1684 tap_rss_hash_update(struct rte_eth_dev *dev,
1685                 struct rte_eth_rss_conf *rss_conf)
1686 {
1687         if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1688                 rte_errno = EINVAL;
1689                 return -rte_errno;
1690         }
1691         if (rss_conf->rss_key && rss_conf->rss_key_len) {
1692                 /*
1693                  * Currently TAP RSS key is hard coded
1694                  * and cannot be updated
1695                  */
1696                 TAP_LOG(ERR,
1697                         "port %u RSS key cannot be updated",
1698                         dev->data->port_id);
1699                 rte_errno = EINVAL;
1700                 return -rte_errno;
1701         }
1702         return 0;
1703 }
1704
1705 static int
1706 tap_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1707 {
1708         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1709
1710         return 0;
1711 }
1712
1713 static int
1714 tap_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1715 {
1716         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1717
1718         return 0;
1719 }
1720
1721 static int
1722 tap_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1723 {
1724         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1725
1726         return 0;
1727 }
1728
1729 static int
1730 tap_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1731 {
1732         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1733
1734         return 0;
1735 }
1736 static const struct eth_dev_ops ops = {
1737         .dev_start              = tap_dev_start,
1738         .dev_stop               = tap_dev_stop,
1739         .dev_close              = tap_dev_close,
1740         .dev_configure          = tap_dev_configure,
1741         .dev_infos_get          = tap_dev_info,
1742         .rx_queue_setup         = tap_rx_queue_setup,
1743         .tx_queue_setup         = tap_tx_queue_setup,
1744         .rx_queue_start         = tap_rx_queue_start,
1745         .tx_queue_start         = tap_tx_queue_start,
1746         .rx_queue_stop          = tap_rx_queue_stop,
1747         .tx_queue_stop          = tap_tx_queue_stop,
1748         .rx_queue_release       = tap_rx_queue_release,
1749         .tx_queue_release       = tap_tx_queue_release,
1750         .flow_ctrl_get          = tap_flow_ctrl_get,
1751         .flow_ctrl_set          = tap_flow_ctrl_set,
1752         .link_update            = tap_link_update,
1753         .dev_set_link_up        = tap_link_set_up,
1754         .dev_set_link_down      = tap_link_set_down,
1755         .promiscuous_enable     = tap_promisc_enable,
1756         .promiscuous_disable    = tap_promisc_disable,
1757         .allmulticast_enable    = tap_allmulti_enable,
1758         .allmulticast_disable   = tap_allmulti_disable,
1759         .mac_addr_set           = tap_mac_set,
1760         .mtu_set                = tap_mtu_set,
1761         .set_mc_addr_list       = tap_set_mc_addr_list,
1762         .stats_get              = tap_stats_get,
1763         .stats_reset            = tap_stats_reset,
1764         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1765         .rss_hash_update        = tap_rss_hash_update,
1766         .filter_ctrl            = tap_dev_filter_ctrl,
1767 };
1768
1769 static const char *tuntap_types[ETH_TUNTAP_TYPE_MAX] = {
1770         "UNKNOWN", "TUN", "TAP"
1771 };
1772
1773 static int
1774 eth_dev_tap_create(struct rte_vdev_device *vdev, const char *tap_name,
1775                    char *remote_iface, struct rte_ether_addr *mac_addr,
1776                    enum rte_tuntap_type type)
1777 {
1778         int numa_node = rte_socket_id();
1779         struct rte_eth_dev *dev;
1780         struct pmd_internals *pmd;
1781         struct pmd_process_private *process_private;
1782         const char *tuntap_name = tuntap_types[type];
1783         struct rte_eth_dev_data *data;
1784         struct ifreq ifr;
1785         int i;
1786
1787         TAP_LOG(DEBUG, "%s device on numa %u", tuntap_name, rte_socket_id());
1788
1789         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1790         if (!dev) {
1791                 TAP_LOG(ERR, "%s Unable to allocate device struct",
1792                                 tuntap_name);
1793                 goto error_exit_nodev;
1794         }
1795
1796         process_private = (struct pmd_process_private *)
1797                 rte_zmalloc_socket(tap_name, sizeof(struct pmd_process_private),
1798                         RTE_CACHE_LINE_SIZE, dev->device->numa_node);
1799
1800         if (process_private == NULL) {
1801                 TAP_LOG(ERR, "Failed to alloc memory for process private");
1802                 return -1;
1803         }
1804         pmd = dev->data->dev_private;
1805         dev->process_private = process_private;
1806         pmd->dev = dev;
1807         strlcpy(pmd->name, tap_name, sizeof(pmd->name));
1808         pmd->type = type;
1809
1810         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1811         if (pmd->ioctl_sock == -1) {
1812                 TAP_LOG(ERR,
1813                         "%s Unable to get a socket for management: %s",
1814                         tuntap_name, strerror(errno));
1815                 goto error_exit;
1816         }
1817
1818         /* Setup some default values */
1819         data = dev->data;
1820         data->dev_private = pmd;
1821         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1822         data->numa_node = numa_node;
1823
1824         data->dev_link = pmd_link;
1825         data->mac_addrs = &pmd->eth_addr;
1826         /* Set the number of RX and TX queues */
1827         data->nb_rx_queues = 0;
1828         data->nb_tx_queues = 0;
1829
1830         dev->dev_ops = &ops;
1831         dev->rx_pkt_burst = pmd_rx_burst;
1832         dev->tx_pkt_burst = pmd_tx_burst;
1833
1834         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1835         pmd->intr_handle.fd = -1;
1836         dev->intr_handle = &pmd->intr_handle;
1837
1838         /* Presetup the fds to -1 as being not valid */
1839         pmd->ka_fd = -1;
1840         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1841                 process_private->rxq_fds[i] = -1;
1842                 process_private->txq_fds[i] = -1;
1843         }
1844
1845         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1846                 if (rte_is_zero_ether_addr(mac_addr))
1847                         rte_eth_random_addr((uint8_t *)&pmd->eth_addr);
1848                 else
1849                         rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1850         }
1851
1852         /*
1853          * Allocate a TUN device keep-alive file descriptor that will only be
1854          * closed when the TUN device itself is closed or removed.
1855          * This keep-alive file descriptor will guarantee that the TUN device
1856          * exists even when all of its queues are closed
1857          */
1858         pmd->ka_fd = tun_alloc(pmd, 1);
1859         if (pmd->ka_fd == -1) {
1860                 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
1861                 goto error_exit;
1862         }
1863         TAP_LOG(DEBUG, "allocated %s", pmd->name);
1864
1865         ifr.ifr_mtu = dev->data->mtu;
1866         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1867                 goto error_exit;
1868
1869         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1870                 memset(&ifr, 0, sizeof(struct ifreq));
1871                 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1872                 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1873                                 RTE_ETHER_ADDR_LEN);
1874                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1875                         goto error_exit;
1876         }
1877
1878         /*
1879          * Set up everything related to rte_flow:
1880          * - netlink socket
1881          * - tap / remote if_index
1882          * - mandatory QDISCs
1883          * - rte_flow actual/implicit lists
1884          * - implicit rules
1885          */
1886         pmd->nlsk_fd = tap_nl_init(0);
1887         if (pmd->nlsk_fd == -1) {
1888                 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
1889                         pmd->name);
1890                 goto disable_rte_flow;
1891         }
1892         pmd->if_index = if_nametoindex(pmd->name);
1893         if (!pmd->if_index) {
1894                 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
1895                 goto disable_rte_flow;
1896         }
1897         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1898                 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
1899                         pmd->name);
1900                 goto disable_rte_flow;
1901         }
1902         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1903                 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1904                         pmd->name);
1905                 goto disable_rte_flow;
1906         }
1907         LIST_INIT(&pmd->flows);
1908
1909         if (strlen(remote_iface)) {
1910                 pmd->remote_if_index = if_nametoindex(remote_iface);
1911                 if (!pmd->remote_if_index) {
1912                         TAP_LOG(ERR, "%s: failed to get %s if_index.",
1913                                 pmd->name, remote_iface);
1914                         goto error_remote;
1915                 }
1916                 strlcpy(pmd->remote_iface, remote_iface, RTE_ETH_NAME_MAX_LEN);
1917
1918                 /* Save state of remote device */
1919                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1920
1921                 /* Replicate remote MAC address */
1922                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1923                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1924                                 pmd->name, pmd->remote_iface);
1925                         goto error_remote;
1926                 }
1927                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1928                            RTE_ETHER_ADDR_LEN);
1929                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1930                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1931                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1932                                 pmd->name, remote_iface);
1933                         goto error_remote;
1934                 }
1935
1936                 /*
1937                  * Flush usually returns negative value because it tries to
1938                  * delete every QDISC (and on a running device, one QDISC at
1939                  * least is needed). Ignore negative return value.
1940                  */
1941                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1942                 if (qdisc_create_ingress(pmd->nlsk_fd,
1943                                          pmd->remote_if_index) < 0) {
1944                         TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1945                                 pmd->remote_iface);
1946                         goto error_remote;
1947                 }
1948                 LIST_INIT(&pmd->implicit_flows);
1949                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1950                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1951                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1952                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1953                         TAP_LOG(ERR,
1954                                 "%s: failed to create implicit rules.",
1955                                 pmd->name);
1956                         goto error_remote;
1957                 }
1958         }
1959
1960         rte_eth_dev_probing_finish(dev);
1961         return 0;
1962
1963 disable_rte_flow:
1964         TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
1965                 strerror(errno), errno);
1966         if (strlen(remote_iface)) {
1967                 TAP_LOG(ERR, "Remote feature requires flow support.");
1968                 goto error_exit;
1969         }
1970         rte_eth_dev_probing_finish(dev);
1971         return 0;
1972
1973 error_remote:
1974         TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
1975                 strerror(errno), errno);
1976         tap_flow_implicit_flush(pmd, NULL);
1977
1978 error_exit:
1979         if (pmd->ioctl_sock > 0)
1980                 close(pmd->ioctl_sock);
1981         /* mac_addrs must not be freed alone because part of dev_private */
1982         dev->data->mac_addrs = NULL;
1983         rte_eth_dev_release_port(dev);
1984
1985 error_exit_nodev:
1986         TAP_LOG(ERR, "%s Unable to initialize %s",
1987                 tuntap_name, rte_vdev_device_name(vdev));
1988
1989         return -EINVAL;
1990 }
1991
1992 /* make sure name is a possible Linux network device name */
1993 static bool
1994 is_valid_iface(const char *name)
1995 {
1996         if (*name == '\0')
1997                 return false;
1998
1999         if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
2000                 return false;
2001
2002         while (*name) {
2003                 if (*name == '/' || *name == ':' || isspace(*name))
2004                         return false;
2005                 name++;
2006         }
2007         return true;
2008 }
2009
2010 static int
2011 set_interface_name(const char *key __rte_unused,
2012                    const char *value,
2013                    void *extra_args)
2014 {
2015         char *name = (char *)extra_args;
2016
2017         if (value) {
2018                 if (!is_valid_iface(value)) {
2019                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2020                                 value);
2021                         return -1;
2022                 }
2023                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2024         } else {
2025                 /* use tap%d which causes kernel to choose next available */
2026                 strlcpy(name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2027         }
2028         return 0;
2029 }
2030
2031 static int
2032 set_remote_iface(const char *key __rte_unused,
2033                  const char *value,
2034                  void *extra_args)
2035 {
2036         char *name = (char *)extra_args;
2037
2038         if (value) {
2039                 if (!is_valid_iface(value)) {
2040                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2041                                 value);
2042                         return -1;
2043                 }
2044                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2045         }
2046
2047         return 0;
2048 }
2049
2050 static int parse_user_mac(struct rte_ether_addr *user_mac,
2051                 const char *value)
2052 {
2053         unsigned int index = 0;
2054         char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
2055
2056         if (user_mac == NULL || value == NULL)
2057                 return 0;
2058
2059         strlcpy(mac_temp, value, sizeof(mac_temp));
2060         mac_byte = strtok(mac_temp, ":");
2061
2062         while ((mac_byte != NULL) &&
2063                         (strlen(mac_byte) <= 2) &&
2064                         (strlen(mac_byte) == strspn(mac_byte,
2065                                         ETH_TAP_CMP_MAC_FMT))) {
2066                 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
2067                 mac_byte = strtok(NULL, ":");
2068         }
2069
2070         return index;
2071 }
2072
2073 static int
2074 set_mac_type(const char *key __rte_unused,
2075              const char *value,
2076              void *extra_args)
2077 {
2078         struct rte_ether_addr *user_mac = extra_args;
2079
2080         if (!value)
2081                 return 0;
2082
2083         if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
2084                 static int iface_idx;
2085
2086                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
2087                 memcpy((char *)user_mac->addr_bytes, "\0dtap",
2088                         RTE_ETHER_ADDR_LEN);
2089                 user_mac->addr_bytes[RTE_ETHER_ADDR_LEN - 1] =
2090                         iface_idx++ + '0';
2091                 goto success;
2092         }
2093
2094         if (parse_user_mac(user_mac, value) != 6)
2095                 goto error;
2096 success:
2097         TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
2098         return 0;
2099
2100 error:
2101         TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
2102                 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
2103         return -1;
2104 }
2105
2106 /*
2107  * Open a TUN interface device. TUN PMD
2108  * 1) sets tap_type as false
2109  * 2) intakes iface as argument.
2110  * 3) as interface is virtual set speed to 10G
2111  */
2112 static int
2113 rte_pmd_tun_probe(struct rte_vdev_device *dev)
2114 {
2115         const char *name, *params;
2116         int ret;
2117         struct rte_kvargs *kvlist = NULL;
2118         char tun_name[RTE_ETH_NAME_MAX_LEN];
2119         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2120         struct rte_eth_dev *eth_dev;
2121
2122         name = rte_vdev_device_name(dev);
2123         params = rte_vdev_device_args(dev);
2124         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2125
2126         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
2127             strlen(params) == 0) {
2128                 eth_dev = rte_eth_dev_attach_secondary(name);
2129                 if (!eth_dev) {
2130                         TAP_LOG(ERR, "Failed to probe %s", name);
2131                         return -1;
2132                 }
2133                 eth_dev->dev_ops = &ops;
2134                 eth_dev->device = &dev->device;
2135                 rte_eth_dev_probing_finish(eth_dev);
2136                 return 0;
2137         }
2138
2139         /* use tun%d which causes kernel to choose next available */
2140         strlcpy(tun_name, DEFAULT_TUN_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2141
2142         if (params && (params[0] != '\0')) {
2143                 TAP_LOG(DEBUG, "parameters (%s)", params);
2144
2145                 kvlist = rte_kvargs_parse(params, valid_arguments);
2146                 if (kvlist) {
2147                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2148                                 ret = rte_kvargs_process(kvlist,
2149                                         ETH_TAP_IFACE_ARG,
2150                                         &set_interface_name,
2151                                         tun_name);
2152
2153                                 if (ret == -1)
2154                                         goto leave;
2155                         }
2156                 }
2157         }
2158         pmd_link.link_speed = ETH_SPEED_NUM_10G;
2159
2160         TAP_LOG(DEBUG, "Initializing pmd_tun for %s", name);
2161
2162         ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0,
2163                                  ETH_TUNTAP_TYPE_TUN);
2164
2165 leave:
2166         if (ret == -1) {
2167                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2168                         name, tun_name);
2169         }
2170         rte_kvargs_free(kvlist);
2171
2172         return ret;
2173 }
2174
2175 /* Request queue file descriptors from secondary to primary. */
2176 static int
2177 tap_mp_attach_queues(const char *port_name, struct rte_eth_dev *dev)
2178 {
2179         int ret;
2180         struct timespec timeout = {.tv_sec = 1, .tv_nsec = 0};
2181         struct rte_mp_msg request, *reply;
2182         struct rte_mp_reply replies;
2183         struct ipc_queues *request_param = (struct ipc_queues *)request.param;
2184         struct ipc_queues *reply_param;
2185         struct pmd_process_private *process_private = dev->process_private;
2186         int queue, fd_iterator;
2187
2188         /* Prepare the request */
2189         memset(&request, 0, sizeof(request));
2190         strlcpy(request.name, TAP_MP_KEY, sizeof(request.name));
2191         strlcpy(request_param->port_name, port_name,
2192                 sizeof(request_param->port_name));
2193         request.len_param = sizeof(*request_param);
2194         /* Send request and receive reply */
2195         ret = rte_mp_request_sync(&request, &replies, &timeout);
2196         if (ret < 0 || replies.nb_received != 1) {
2197                 TAP_LOG(ERR, "Failed to request queues from primary: %d",
2198                         rte_errno);
2199                 return -1;
2200         }
2201         reply = &replies.msgs[0];
2202         reply_param = (struct ipc_queues *)reply->param;
2203         TAP_LOG(DEBUG, "Received IPC reply for %s", reply_param->port_name);
2204
2205         /* Attach the queues from received file descriptors */
2206         if (reply_param->rxq_count + reply_param->txq_count != reply->num_fds) {
2207                 TAP_LOG(ERR, "Unexpected number of fds received");
2208                 return -1;
2209         }
2210
2211         dev->data->nb_rx_queues = reply_param->rxq_count;
2212         dev->data->nb_tx_queues = reply_param->txq_count;
2213         fd_iterator = 0;
2214         for (queue = 0; queue < reply_param->rxq_count; queue++)
2215                 process_private->rxq_fds[queue] = reply->fds[fd_iterator++];
2216         for (queue = 0; queue < reply_param->txq_count; queue++)
2217                 process_private->txq_fds[queue] = reply->fds[fd_iterator++];
2218         free(reply);
2219         return 0;
2220 }
2221
2222 /* Send the queue file descriptors from the primary process to secondary. */
2223 static int
2224 tap_mp_sync_queues(const struct rte_mp_msg *request, const void *peer)
2225 {
2226         struct rte_eth_dev *dev;
2227         struct pmd_process_private *process_private;
2228         struct rte_mp_msg reply;
2229         const struct ipc_queues *request_param =
2230                 (const struct ipc_queues *)request->param;
2231         struct ipc_queues *reply_param =
2232                 (struct ipc_queues *)reply.param;
2233         uint16_t port_id;
2234         int queue;
2235         int ret;
2236
2237         /* Get requested port */
2238         TAP_LOG(DEBUG, "Received IPC request for %s", request_param->port_name);
2239         ret = rte_eth_dev_get_port_by_name(request_param->port_name, &port_id);
2240         if (ret) {
2241                 TAP_LOG(ERR, "Failed to get port id for %s",
2242                         request_param->port_name);
2243                 return -1;
2244         }
2245         dev = &rte_eth_devices[port_id];
2246         process_private = dev->process_private;
2247
2248         /* Fill file descriptors for all queues */
2249         reply.num_fds = 0;
2250         reply_param->rxq_count = 0;
2251         if (dev->data->nb_rx_queues + dev->data->nb_tx_queues >
2252                         RTE_MP_MAX_FD_NUM){
2253                 TAP_LOG(ERR, "Number of rx/tx queues exceeds max number of fds");
2254                 return -1;
2255         }
2256
2257         for (queue = 0; queue < dev->data->nb_rx_queues; queue++) {
2258                 reply.fds[reply.num_fds++] = process_private->rxq_fds[queue];
2259                 reply_param->rxq_count++;
2260         }
2261         RTE_ASSERT(reply_param->rxq_count == dev->data->nb_rx_queues);
2262
2263         reply_param->txq_count = 0;
2264         for (queue = 0; queue < dev->data->nb_tx_queues; queue++) {
2265                 reply.fds[reply.num_fds++] = process_private->txq_fds[queue];
2266                 reply_param->txq_count++;
2267         }
2268         RTE_ASSERT(reply_param->txq_count == dev->data->nb_tx_queues);
2269
2270         /* Send reply */
2271         strlcpy(reply.name, request->name, sizeof(reply.name));
2272         strlcpy(reply_param->port_name, request_param->port_name,
2273                 sizeof(reply_param->port_name));
2274         reply.len_param = sizeof(*reply_param);
2275         if (rte_mp_reply(&reply, peer) < 0) {
2276                 TAP_LOG(ERR, "Failed to reply an IPC request to sync queues");
2277                 return -1;
2278         }
2279         return 0;
2280 }
2281
2282 /* Open a TAP interface device.
2283  */
2284 static int
2285 rte_pmd_tap_probe(struct rte_vdev_device *dev)
2286 {
2287         const char *name, *params;
2288         int ret;
2289         struct rte_kvargs *kvlist = NULL;
2290         int speed;
2291         char tap_name[RTE_ETH_NAME_MAX_LEN];
2292         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2293         struct rte_ether_addr user_mac = { .addr_bytes = {0} };
2294         struct rte_eth_dev *eth_dev;
2295         int tap_devices_count_increased = 0;
2296
2297         name = rte_vdev_device_name(dev);
2298         params = rte_vdev_device_args(dev);
2299
2300         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
2301                 eth_dev = rte_eth_dev_attach_secondary(name);
2302                 if (!eth_dev) {
2303                         TAP_LOG(ERR, "Failed to probe %s", name);
2304                         return -1;
2305                 }
2306                 eth_dev->dev_ops = &ops;
2307                 eth_dev->device = &dev->device;
2308                 eth_dev->rx_pkt_burst = pmd_rx_burst;
2309                 eth_dev->tx_pkt_burst = pmd_tx_burst;
2310                 if (!rte_eal_primary_proc_alive(NULL)) {
2311                         TAP_LOG(ERR, "Primary process is missing");
2312                         return -1;
2313                 }
2314                 eth_dev->process_private = (struct pmd_process_private *)
2315                         rte_zmalloc_socket(name,
2316                                 sizeof(struct pmd_process_private),
2317                                 RTE_CACHE_LINE_SIZE,
2318                                 eth_dev->device->numa_node);
2319                 if (eth_dev->process_private == NULL) {
2320                         TAP_LOG(ERR,
2321                                 "Failed to alloc memory for process private");
2322                         return -1;
2323                 }
2324
2325                 ret = tap_mp_attach_queues(name, eth_dev);
2326                 if (ret != 0)
2327                         return -1;
2328                 rte_eth_dev_probing_finish(eth_dev);
2329                 return 0;
2330         }
2331
2332         speed = ETH_SPEED_NUM_10G;
2333
2334         /* use tap%d which causes kernel to choose next available */
2335         strlcpy(tap_name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2336         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2337
2338         if (params && (params[0] != '\0')) {
2339                 TAP_LOG(DEBUG, "parameters (%s)", params);
2340
2341                 kvlist = rte_kvargs_parse(params, valid_arguments);
2342                 if (kvlist) {
2343                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2344                                 ret = rte_kvargs_process(kvlist,
2345                                                          ETH_TAP_IFACE_ARG,
2346                                                          &set_interface_name,
2347                                                          tap_name);
2348                                 if (ret == -1)
2349                                         goto leave;
2350                         }
2351
2352                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
2353                                 ret = rte_kvargs_process(kvlist,
2354                                                          ETH_TAP_REMOTE_ARG,
2355                                                          &set_remote_iface,
2356                                                          remote_iface);
2357                                 if (ret == -1)
2358                                         goto leave;
2359                         }
2360
2361                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
2362                                 ret = rte_kvargs_process(kvlist,
2363                                                          ETH_TAP_MAC_ARG,
2364                                                          &set_mac_type,
2365                                                          &user_mac);
2366                                 if (ret == -1)
2367                                         goto leave;
2368                         }
2369                 }
2370         }
2371         pmd_link.link_speed = speed;
2372
2373         TAP_LOG(DEBUG, "Initializing pmd_tap for %s", name);
2374
2375         /* Register IPC feed callback */
2376         if (!tap_devices_count) {
2377                 ret = rte_mp_action_register(TAP_MP_KEY, tap_mp_sync_queues);
2378                 if (ret < 0 && rte_errno != ENOTSUP) {
2379                         TAP_LOG(ERR, "tap: Failed to register IPC callback: %s",
2380                                 strerror(rte_errno));
2381                         goto leave;
2382                 }
2383         }
2384         tap_devices_count++;
2385         tap_devices_count_increased = 1;
2386         ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
2387                 ETH_TUNTAP_TYPE_TAP);
2388
2389 leave:
2390         if (ret == -1) {
2391                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2392                         name, tap_name);
2393                 if (tap_devices_count_increased == 1) {
2394                         if (tap_devices_count == 1)
2395                                 rte_mp_action_unregister(TAP_MP_KEY);
2396                         tap_devices_count--;
2397                 }
2398         }
2399         rte_kvargs_free(kvlist);
2400
2401         return ret;
2402 }
2403
2404 /* detach a TUNTAP device.
2405  */
2406 static int
2407 rte_pmd_tap_remove(struct rte_vdev_device *dev)
2408 {
2409         struct rte_eth_dev *eth_dev = NULL;
2410         struct pmd_internals *internals;
2411
2412         /* find the ethdev entry */
2413         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
2414         if (!eth_dev)
2415                 return -ENODEV;
2416
2417         /* mac_addrs must not be freed alone because part of dev_private */
2418         eth_dev->data->mac_addrs = NULL;
2419
2420         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2421                 return rte_eth_dev_release_port(eth_dev);
2422
2423         tap_dev_close(eth_dev);
2424
2425         internals = eth_dev->data->dev_private;
2426         TAP_LOG(DEBUG, "Closing %s Ethernet device on numa %u",
2427                 tuntap_types[internals->type], rte_socket_id());
2428
2429         close(internals->ioctl_sock);
2430         rte_free(eth_dev->process_private);
2431         if (tap_devices_count == 1)
2432                 rte_mp_action_unregister(TAP_MP_KEY);
2433         tap_devices_count--;
2434         rte_eth_dev_release_port(eth_dev);
2435
2436         if (internals->ka_fd != -1) {
2437                 close(internals->ka_fd);
2438                 internals->ka_fd = -1;
2439         }
2440         return 0;
2441 }
2442
2443 static struct rte_vdev_driver pmd_tun_drv = {
2444         .probe = rte_pmd_tun_probe,
2445         .remove = rte_pmd_tap_remove,
2446 };
2447
2448 static struct rte_vdev_driver pmd_tap_drv = {
2449         .probe = rte_pmd_tap_probe,
2450         .remove = rte_pmd_tap_remove,
2451 };
2452
2453 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
2454 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
2455 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
2456 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
2457                               ETH_TAP_IFACE_ARG "=<string> ");
2458 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
2459                               ETH_TAP_IFACE_ARG "=<string> "
2460                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
2461                               ETH_TAP_REMOTE_ARG "=<string>");
2462 int tap_logtype;
2463
2464 RTE_INIT(tap_init_log)
2465 {
2466         tap_logtype = rte_log_register("pmd.net.tap");
2467         if (tap_logtype >= 0)
2468                 rte_log_set_level(tap_logtype, RTE_LOG_NOTICE);
2469 }