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 #include <rte_cycles.h>
  22
  23 #include <sys/types.h>
  24 #include <sys/stat.h>
  25 #include <sys/socket.h>
  26 #include <sys/ioctl.h>
  27 #include <sys/utsname.h>
  28 #include <sys/mman.h>
  29 #include <errno.h>
  30 #include <signal.h>
  31 #include <stdbool.h>
  32 #include <stdint.h>
  33 #include <sys/uio.h>
  34 #include <unistd.h>
  35 #include <arpa/inet.h>
  36 #include <net/if.h>
  37 #include <linux/if_tun.h>
  38 #include <linux/if_ether.h>
  39 #include <fcntl.h>
  40 #include <ctype.h>
  41
  42 #include <tap_rss.h>
  43 #include <rte_eth_tap.h>
  44 #include <tap_flow.h>
  45 #include <tap_netlink.h>
  46 #include <tap_tcmsgs.h>
  47
  48 /* Linux based path to the TUN device */
  49 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  50 #define DEFAULT_TAP_NAME        "dtap"
  51 #define DEFAULT_TUN_NAME        "dtun"
  52
  53 #define ETH_TAP_IFACE_ARG       "iface"
  54 #define ETH_TAP_REMOTE_ARG      "remote"
  55 #define ETH_TAP_MAC_ARG         "mac"
  56 #define ETH_TAP_MAC_FIXED       "fixed"
  57
  58 #define ETH_TAP_USR_MAC_FMT     "xx:xx:xx:xx:xx:xx"
  59 #define ETH_TAP_CMP_MAC_FMT     "0123456789ABCDEFabcdef"
  60 #define ETH_TAP_MAC_ARG_FMT     ETH_TAP_MAC_FIXED "|" ETH_TAP_USR_MAC_FMT
  61
  62 #define TAP_GSO_MBUFS_PER_CORE  128
  63 #define TAP_GSO_MBUF_SEG_SIZE   128
  64 #define TAP_GSO_MBUF_CACHE_SIZE 4
  65 #define TAP_GSO_MBUFS_NUM \
  66         (TAP_GSO_MBUFS_PER_CORE * TAP_GSO_MBUF_CACHE_SIZE)
  67
  68 /* IPC key for queue fds sync */
  69 #define TAP_MP_KEY "tap_mp_sync_queues"
  70
  71 #define TAP_IOV_DEFAULT_MAX 1024
  72
  73 static int tap_devices_count;
  74
  75 static const char *valid_arguments[] = {
  76         ETH_TAP_IFACE_ARG,
  77         ETH_TAP_REMOTE_ARG,
  78         ETH_TAP_MAC_ARG,
  79         NULL
  80 };
  81
  82 static volatile uint32_t tap_trigger;   /* Rx trigger */
  83
  84 static struct rte_eth_link pmd_link = {
  85         .link_speed = ETH_SPEED_NUM_10G,
  86         .link_duplex = ETH_LINK_FULL_DUPLEX,
  87         .link_status = ETH_LINK_DOWN,
  88         .link_autoneg = ETH_LINK_FIXED,
  89 };
  90
  91 static void
  92 tap_trigger_cb(int sig __rte_unused)
  93 {
  94         /* Valid trigger values are nonzero */
  95         tap_trigger = (tap_trigger + 1) | 0x80000000;
  96 }
  97
  98 /* Specifies on what netdevices the ioctl should be applied */
  99 enum ioctl_mode {
 100         LOCAL_AND_REMOTE,
 101         LOCAL_ONLY,
 102         REMOTE_ONLY,
 103 };
 104
 105 /* Message header to synchronize queues via IPC */
 106 struct ipc_queues {
 107         char port_name[RTE_DEV_NAME_MAX_LEN];
 108         int rxq_count;
 109         int txq_count;
 110         /*
 111          * The file descriptors are in the dedicated part
 112          * of the Unix message to be translated by the kernel.
 113          */
 114 };
 115
 116 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
 117
 118 /**
 119  * Tun/Tap allocation routine
 120  *
 121  * @param[in] pmd
 122  *   Pointer to private structure.
 123  *
 124  * @param[in] is_keepalive
 125  *   Keepalive flag
 126  *
 127  * @return
 128  *   -1 on failure, fd on success
 129  */
 130 static int
 131 tun_alloc(struct pmd_internals *pmd, int is_keepalive)
 132 {
 133         struct ifreq ifr;
 134 #ifdef IFF_MULTI_QUEUE
 135         unsigned int features;
 136 #endif
 137         int fd;
 138
 139         memset(&ifr, 0, sizeof(struct ifreq));
 140
 141         /*
 142          * Do not set IFF_NO_PI as packet information header will be needed
 143          * to check if a received packet has been truncated.
 144          */
 145         ifr.ifr_flags = (pmd->type == ETH_TUNTAP_TYPE_TAP) ?
 146                 IFF_TAP : IFF_TUN | IFF_POINTOPOINT;
 147         strlcpy(ifr.ifr_name, pmd->name, IFNAMSIZ);
 148
 149         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 150         if (fd < 0) {
 151                 TAP_LOG(ERR, "Unable to open %s interface", TUN_TAP_DEV_PATH);
 152                 goto error;
 153         }
 154
 155 #ifdef IFF_MULTI_QUEUE
 156         /* Grab the TUN features to verify we can work multi-queue */
 157         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 158                 TAP_LOG(ERR, "unable to get TUN/TAP features");
 159                 goto error;
 160         }
 161         TAP_LOG(DEBUG, "%s Features %08x", TUN_TAP_DEV_PATH, features);
 162
 163         if (features & IFF_MULTI_QUEUE) {
 164                 TAP_LOG(DEBUG, "  Multi-queue support for %d queues",
 165                         RTE_PMD_TAP_MAX_QUEUES);
 166                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 167         } else
 168 #endif
 169         {
 170                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 171                 TAP_LOG(DEBUG, "  Single queue only support");
 172         }
 173
 174         /* Set the TUN/TAP configuration and set the name if needed */
 175         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 176                 TAP_LOG(WARNING, "Unable to set TUNSETIFF for %s: %s",
 177                         ifr.ifr_name, strerror(errno));
 178                 goto error;
 179         }
 180
 181         /*
 182          * Name passed to kernel might be wildcard like dtun%d
 183          * and need to find the resulting device.
 184          */
 185         TAP_LOG(DEBUG, "Device name is '%s'", ifr.ifr_name);
 186         strlcpy(pmd->name, ifr.ifr_name, RTE_ETH_NAME_MAX_LEN);
 187
 188         if (is_keepalive) {
 189                 /*
 190                  * Detach the TUN/TAP keep-alive queue
 191                  * to avoid traffic through it
 192                  */
 193                 ifr.ifr_flags = IFF_DETACH_QUEUE;
 194                 if (ioctl(fd, TUNSETQUEUE, (void *)&ifr) < 0) {
 195                         TAP_LOG(WARNING,
 196                                 "Unable to detach keep-alive queue for %s: %s",
 197                                 ifr.ifr_name, strerror(errno));
 198                         goto error;
 199                 }
 200         }
 201
 202         /* Always set the file descriptor to non-blocking */
 203         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 204                 TAP_LOG(WARNING,
 205                         "Unable to set %s to nonblocking: %s",
 206                         ifr.ifr_name, strerror(errno));
 207                 goto error;
 208         }
 209
 210         /* Set up trigger to optimize empty Rx bursts */
 211         errno = 0;
 212         do {
 213                 struct sigaction sa;
 214                 int flags = fcntl(fd, F_GETFL);
 215
 216                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 217                         break;
 218                 if (sa.sa_handler != tap_trigger_cb) {
 219                         /*
 220                          * Make sure SIGIO is not already taken. This is done
 221                          * as late as possible to leave the application a
 222                          * chance to set up its own signal handler first.
 223                          */
 224                         if (sa.sa_handler != SIG_IGN &&
 225                             sa.sa_handler != SIG_DFL) {
 226                                 errno = EBUSY;
 227                                 break;
 228                         }
 229                         sa = (struct sigaction){
 230                                 .sa_flags = SA_RESTART,
 231                                 .sa_handler = tap_trigger_cb,
 232                         };
 233                         if (sigaction(SIGIO, &sa, NULL) == -1)
 234                                 break;
 235                 }
 236                 /* Enable SIGIO on file descriptor */
 237                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 238                 fcntl(fd, F_SETOWN, getpid());
 239         } while (0);
 240
 241         if (errno) {
 242                 /* Disable trigger globally in case of error */
 243                 tap_trigger = 0;
 244                 TAP_LOG(WARNING, "Rx trigger disabled: %s",
 245                         strerror(errno));
 246         }
 247
 248         return fd;
 249
 250 error:
 251         if (fd >= 0)
 252                 close(fd);
 253         return -1;
 254 }
 255
 256 static void
 257 tap_verify_csum(struct rte_mbuf *mbuf)
 258 {
 259         uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
 260         uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
 261         uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
 262         unsigned int l2_len = sizeof(struct rte_ether_hdr);
 263         unsigned int l3_len;
 264         uint16_t cksum = 0;
 265         void *l3_hdr;
 266         void *l4_hdr;
 267
 268         if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
 269                 l2_len += 4;
 270         else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
 271                 l2_len += 8;
 272         /* Don't verify checksum for packets with discontinuous L2 header */
 273         if (unlikely(l2_len + sizeof(struct rte_ipv4_hdr) >
 274                      rte_pktmbuf_data_len(mbuf)))
 275                 return;
 276         l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
 277         if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 278                 struct rte_ipv4_hdr *iph = l3_hdr;
 279
 280                 /* ihl contains the number of 4-byte words in the header */
 281                 l3_len = 4 * (iph->version_ihl & 0xf);
 282                 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
 283                         return;
 284                 /* check that the total length reported by header is not
 285                  * greater than the total received size
 286                  */
 287                 if (l2_len + rte_be_to_cpu_16(iph->total_length) >
 288                                 rte_pktmbuf_data_len(mbuf))
 289                         return;
 290
 291                 cksum = ~rte_raw_cksum(iph, l3_len);
 292                 mbuf->ol_flags |= cksum ?
 293                         PKT_RX_IP_CKSUM_BAD :
 294                         PKT_RX_IP_CKSUM_GOOD;
 295         } else if (l3 == RTE_PTYPE_L3_IPV6) {
 296                 struct rte_ipv6_hdr *iph = l3_hdr;
 297
 298                 l3_len = sizeof(struct rte_ipv6_hdr);
 299                 /* check that the total length reported by header is not
 300                  * greater than the total received size
 301                  */
 302                 if (l2_len + l3_len + rte_be_to_cpu_16(iph->payload_len) >
 303                                 rte_pktmbuf_data_len(mbuf))
 304                         return;
 305         } else {
 306                 /* IPv6 extensions are not supported */
 307                 return;
 308         }
 309         if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
 310                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 311                 /* Don't verify checksum for multi-segment packets. */
 312                 if (mbuf->nb_segs > 1)
 313                         return;
 314                 if (l3 == RTE_PTYPE_L3_IPV4)
 315                         cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
 316                 else if (l3 == RTE_PTYPE_L3_IPV6)
 317                         cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
 318                 mbuf->ol_flags |= cksum ?
 319                         PKT_RX_L4_CKSUM_BAD :
 320                         PKT_RX_L4_CKSUM_GOOD;
 321         }
 322 }
 323
 324 static uint64_t
 325 tap_rx_offload_get_port_capa(void)
 326 {
 327         /*
 328          * No specific port Rx offload capabilities.
 329          */
 330         return 0;
 331 }
 332
 333 static uint64_t
 334 tap_rx_offload_get_queue_capa(void)
 335 {
 336         return DEV_RX_OFFLOAD_SCATTER |
 337                DEV_RX_OFFLOAD_IPV4_CKSUM |
 338                DEV_RX_OFFLOAD_UDP_CKSUM |
 339                DEV_RX_OFFLOAD_TCP_CKSUM;
 340 }
 341
 342 static void
 343 tap_rxq_pool_free(struct rte_mbuf *pool)
 344 {
 345         struct rte_mbuf *mbuf = pool;
 346         uint16_t nb_segs = 1;
 347
 348         if (mbuf == NULL)
 349                 return;
 350
 351         while (mbuf->next) {
 352                 mbuf = mbuf->next;
 353                 nb_segs++;
 354         }
 355         pool->nb_segs = nb_segs;
 356         rte_pktmbuf_free(pool);
 357 }
 358
 359 /* Callback to handle the rx burst of packets to the correct interface and
 360  * file descriptor(s) in a multi-queue setup.
 361  */
 362 static uint16_t
 363 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 364 {
 365         struct rx_queue *rxq = queue;
 366         struct pmd_process_private *process_private;
 367         uint16_t num_rx;
 368         unsigned long num_rx_bytes = 0;
 369         uint32_t trigger = tap_trigger;
 370
 371         if (trigger == rxq->trigger_seen)
 372                 return 0;
 373
 374         process_private = rte_eth_devices[rxq->in_port].process_private;
 375         for (num_rx = 0; num_rx < nb_pkts; ) {
 376                 struct rte_mbuf *mbuf = rxq->pool;
 377                 struct rte_mbuf *seg = NULL;
 378                 struct rte_mbuf *new_tail = NULL;
 379                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 380                 int len;
 381
 382                 len = readv(process_private->rxq_fds[rxq->queue_id],
 383                         *rxq->iovecs,
 384                         1 + (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
 385                              rxq->nb_rx_desc : 1));
 386                 if (len < (int)sizeof(struct tun_pi))
 387                         break;
 388
 389                 /* Packet couldn't fit in the provided mbuf */
 390                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 391                         rxq->stats.ierrors++;
 392                         continue;
 393                 }
 394
 395                 len -= sizeof(struct tun_pi);
 396
 397                 mbuf->pkt_len = len;
 398                 mbuf->port = rxq->in_port;
 399                 while (1) {
 400                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 401
 402                         if (unlikely(!buf)) {
 403                                 rxq->stats.rx_nombuf++;
 404                                 /* No new buf has been allocated: do nothing */
 405                                 if (!new_tail || !seg)
 406                                         goto end;
 407
 408                                 seg->next = NULL;
 409                                 tap_rxq_pool_free(mbuf);
 410
 411                                 goto end;
 412                         }
 413                         seg = seg ? seg->next : mbuf;
 414                         if (rxq->pool == mbuf)
 415                                 rxq->pool = buf;
 416                         if (new_tail)
 417                                 new_tail->next = buf;
 418                         new_tail = buf;
 419                         new_tail->next = seg->next;
 420
 421                         /* iovecs[0] is reserved for packet info (pi) */
 422                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 423                                 buf->buf_len - data_off;
 424                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 425                                 (char *)buf->buf_addr + data_off;
 426
 427                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 428                         seg->data_off = data_off;
 429
 430                         len -= seg->data_len;
 431                         if (len <= 0)
 432                                 break;
 433                         mbuf->nb_segs++;
 434                         /* First segment has headroom, not the others */
 435                         data_off = 0;
 436                 }
 437                 seg->next = NULL;
 438                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 439                                                       RTE_PTYPE_ALL_MASK);
 440                 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
 441                         tap_verify_csum(mbuf);
 442
 443                 /* account for the receive frame */
 444                 bufs[num_rx++] = mbuf;
 445                 num_rx_bytes += mbuf->pkt_len;
 446         }
 447 end:
 448         rxq->stats.ipackets += num_rx;
 449         rxq->stats.ibytes += num_rx_bytes;
 450
 451         if (trigger && num_rx < nb_pkts)
 452                 rxq->trigger_seen = trigger;
 453
 454         return num_rx;
 455 }
 456
 457 static uint64_t
 458 tap_tx_offload_get_port_capa(void)
 459 {
 460         /*
 461          * No specific port Tx offload capabilities.
 462          */
 463         return 0;
 464 }
 465
 466 static uint64_t
 467 tap_tx_offload_get_queue_capa(void)
 468 {
 469         return DEV_TX_OFFLOAD_MULTI_SEGS |
 470                DEV_TX_OFFLOAD_IPV4_CKSUM |
 471                DEV_TX_OFFLOAD_UDP_CKSUM |
 472                DEV_TX_OFFLOAD_TCP_CKSUM |
 473                DEV_TX_OFFLOAD_TCP_TSO;
 474 }
 475
 476 /* Finalize l4 checksum calculation */
 477 static void
 478 tap_tx_l4_cksum(uint16_t *l4_cksum, uint16_t l4_phdr_cksum,
 479                 uint32_t l4_raw_cksum)
 480 {
 481         if (l4_cksum) {
 482                 uint32_t cksum;
 483
 484                 cksum = __rte_raw_cksum_reduce(l4_raw_cksum);
 485                 cksum += l4_phdr_cksum;
 486
 487                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 488                 cksum = (~cksum) & 0xffff;
 489                 if (cksum == 0)
 490                         cksum = 0xffff;
 491                 *l4_cksum = cksum;
 492         }
 493 }
 494
 495 /* Accumaulate L4 raw checksums */
 496 static void
 497 tap_tx_l4_add_rcksum(char *l4_data, unsigned int l4_len, uint16_t *l4_cksum,
 498                         uint32_t *l4_raw_cksum)
 499 {
 500         if (l4_cksum == NULL)
 501                 return;
 502
 503         *l4_raw_cksum = __rte_raw_cksum(l4_data, l4_len, *l4_raw_cksum);
 504 }
 505
 506 /* L3 and L4 pseudo headers checksum offloads */
 507 static void
 508 tap_tx_l3_cksum(char *packet, uint64_t ol_flags, unsigned int l2_len,
 509                 unsigned int l3_len, unsigned int l4_len, uint16_t **l4_cksum,
 510                 uint16_t *l4_phdr_cksum, uint32_t *l4_raw_cksum)
 511 {
 512         void *l3_hdr = packet + l2_len;
 513
 514         if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
 515                 struct rte_ipv4_hdr *iph = l3_hdr;
 516                 uint16_t cksum;
 517
 518                 iph->hdr_checksum = 0;
 519                 cksum = rte_raw_cksum(iph, l3_len);
 520                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 521         }
 522         if (ol_flags & PKT_TX_L4_MASK) {
 523                 void *l4_hdr;
 524
 525                 l4_hdr = packet + l2_len + l3_len;
 526                 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
 527                         *l4_cksum = &((struct rte_udp_hdr *)l4_hdr)->dgram_cksum;
 528                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 529                         *l4_cksum = &((struct rte_tcp_hdr *)l4_hdr)->cksum;
 530                 else
 531                         return;
 532                 **l4_cksum = 0;
 533                 if (ol_flags & PKT_TX_IPV4)
 534                         *l4_phdr_cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 535                 else
 536                         *l4_phdr_cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 537                 *l4_raw_cksum = __rte_raw_cksum(l4_hdr, l4_len, 0);
 538         }
 539 }
 540
 541 static inline int
 542 tap_write_mbufs(struct tx_queue *txq, uint16_t num_mbufs,
 543                         struct rte_mbuf **pmbufs,
 544                         uint16_t *num_packets, unsigned long *num_tx_bytes)
 545 {
 546         int i;
 547         uint16_t l234_hlen;
 548         struct pmd_process_private *process_private;
 549
 550         process_private = rte_eth_devices[txq->out_port].process_private;
 551
 552         for (i = 0; i < num_mbufs; i++) {
 553                 struct rte_mbuf *mbuf = pmbufs[i];
 554                 struct iovec iovecs[mbuf->nb_segs + 2];
 555                 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
 556                 struct rte_mbuf *seg = mbuf;
 557                 char m_copy[mbuf->data_len];
 558                 int proto;
 559                 int n;
 560                 int j;
 561                 int k; /* current index in iovecs for copying segments */
 562                 uint16_t seg_len; /* length of first segment */
 563                 uint16_t nb_segs;
 564                 uint16_t *l4_cksum; /* l4 checksum (pseudo header + payload) */
 565                 uint32_t l4_raw_cksum = 0; /* TCP/UDP payload raw checksum */
 566                 uint16_t l4_phdr_cksum = 0; /* TCP/UDP pseudo header checksum */
 567                 uint16_t is_cksum = 0; /* in case cksum should be offloaded */
 568
 569                 l4_cksum = NULL;
 570                 if (txq->type == ETH_TUNTAP_TYPE_TUN) {
 571                         /*
 572                          * TUN and TAP are created with IFF_NO_PI disabled.
 573                          * For TUN PMD this mandatory as fields are used by
 574                          * Kernel tun.c to determine whether its IP or non IP
 575                          * packets.
 576                          *
 577                          * The logic fetches the first byte of data from mbuf
 578                          * then compares whether its v4 or v6. If first byte
 579                          * is 4 or 6, then protocol field is updated.
 580                          */
 581                         char *buff_data = rte_pktmbuf_mtod(seg, void *);
 582                         proto = (*buff_data & 0xf0);
 583                         pi.proto = (proto == 0x40) ?
 584                                 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4) :
 585                                 ((proto == 0x60) ?
 586                                         rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6) :
 587                                         0x00);
 588                 }
 589
 590                 k = 0;
 591                 iovecs[k].iov_base = &pi;
 592                 iovecs[k].iov_len = sizeof(pi);
 593                 k++;
 594
 595                 nb_segs = mbuf->nb_segs;
 596                 if (txq->csum &&
 597                     ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 598                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 599                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
 600                         is_cksum = 1;
 601
 602                         /* Support only packets with at least layer 4
 603                          * header included in the first segment
 604                          */
 605                         seg_len = rte_pktmbuf_data_len(mbuf);
 606                         l234_hlen = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
 607                         if (seg_len < l234_hlen)
 608                                 return -1;
 609
 610                         /* To change checksums, work on a * copy of l2, l3
 611                          * headers + l4 pseudo header
 612                          */
 613                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 614                                         l234_hlen);
 615                         tap_tx_l3_cksum(m_copy, mbuf->ol_flags,
 616                                        mbuf->l2_len, mbuf->l3_len, mbuf->l4_len,
 617                                        &l4_cksum, &l4_phdr_cksum,
 618                                        &l4_raw_cksum);
 619                         iovecs[k].iov_base = m_copy;
 620                         iovecs[k].iov_len = l234_hlen;
 621                         k++;
 622
 623                         /* Update next iovecs[] beyond l2, l3, l4 headers */
 624                         if (seg_len > l234_hlen) {
 625                                 iovecs[k].iov_len = seg_len - l234_hlen;
 626                                 iovecs[k].iov_base =
 627                                         rte_pktmbuf_mtod(seg, char *) +
 628                                                 l234_hlen;
 629                                 tap_tx_l4_add_rcksum(iovecs[k].iov_base,
 630                                         iovecs[k].iov_len, l4_cksum,
 631                                         &l4_raw_cksum);
 632                                 k++;
 633                                 nb_segs++;
 634                         }
 635                         seg = seg->next;
 636                 }
 637
 638                 for (j = k; j <= nb_segs; j++) {
 639                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 640                         iovecs[j].iov_base = rte_pktmbuf_mtod(seg, void *);
 641                         if (is_cksum)
 642                                 tap_tx_l4_add_rcksum(iovecs[j].iov_base,
 643                                         iovecs[j].iov_len, l4_cksum,
 644                                         &l4_raw_cksum);
 645                         seg = seg->next;
 646                 }
 647
 648                 if (is_cksum)
 649                         tap_tx_l4_cksum(l4_cksum, l4_phdr_cksum, l4_raw_cksum);
 650
 651                 /* copy the tx frame data */
 652                 n = writev(process_private->txq_fds[txq->queue_id], iovecs, j);
 653                 if (n <= 0)
 654                         return -1;
 655
 656                 (*num_packets)++;
 657                 (*num_tx_bytes) += rte_pktmbuf_pkt_len(mbuf);
 658         }
 659         return 0;
 660 }
 661
 662 /* Callback to handle sending packets from the tap interface
 663  */
 664 static uint16_t
 665 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 666 {
 667         struct tx_queue *txq = queue;
 668         uint16_t num_tx = 0;
 669         uint16_t num_packets = 0;
 670         unsigned long num_tx_bytes = 0;
 671         uint32_t max_size;
 672         int i;
 673
 674         if (unlikely(nb_pkts == 0))
 675                 return 0;
 676
 677         struct rte_mbuf *gso_mbufs[MAX_GSO_MBUFS];
 678         max_size = *txq->mtu + (RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN + 4);
 679         for (i = 0; i < nb_pkts; i++) {
 680                 struct rte_mbuf *mbuf_in = bufs[num_tx];
 681                 struct rte_mbuf **mbuf;
 682                 uint16_t num_mbufs = 0;
 683                 uint16_t tso_segsz = 0;
 684                 int ret;
 685                 int num_tso_mbufs;
 686                 uint16_t hdrs_len;
 687                 uint64_t tso;
 688
 689                 tso = mbuf_in->ol_flags & PKT_TX_TCP_SEG;
 690                 if (tso) {
 691                         struct rte_gso_ctx *gso_ctx = &txq->gso_ctx;
 692
 693                         /* TCP segmentation implies TCP checksum offload */
 694                         mbuf_in->ol_flags |= PKT_TX_TCP_CKSUM;
 695
 696                         /* gso size is calculated without RTE_ETHER_CRC_LEN */
 697                         hdrs_len = mbuf_in->l2_len + mbuf_in->l3_len +
 698                                         mbuf_in->l4_len;
 699                         tso_segsz = mbuf_in->tso_segsz + hdrs_len;
 700                         if (unlikely(tso_segsz == hdrs_len) ||
 701                                 tso_segsz > *txq->mtu) {
 702                                 txq->stats.errs++;
 703                                 break;
 704                         }
 705                         gso_ctx->gso_size = tso_segsz;
 706                         /* 'mbuf_in' packet to segment */
 707                         num_tso_mbufs = rte_gso_segment(mbuf_in,
 708                                 gso_ctx, /* gso control block */
 709                                 (struct rte_mbuf **)&gso_mbufs, /* out mbufs */
 710                                 RTE_DIM(gso_mbufs)); /* max tso mbufs */
 711
 712                         /* ret contains the number of new created mbufs */
 713                         if (num_tso_mbufs < 0)
 714                                 break;
 715
 716                         mbuf = gso_mbufs;
 717                         num_mbufs = num_tso_mbufs;
 718                 } else {
 719                         /* stats.errs will be incremented */
 720                         if (rte_pktmbuf_pkt_len(mbuf_in) > max_size)
 721                                 break;
 722
 723                         /* ret 0 indicates no new mbufs were created */
 724                         num_tso_mbufs = 0;
 725                         mbuf = &mbuf_in;
 726                         num_mbufs = 1;
 727                 }
 728
 729                 ret = tap_write_mbufs(txq, num_mbufs, mbuf,
 730                                 &num_packets, &num_tx_bytes);
 731                 if (ret == -1) {
 732                         txq->stats.errs++;
 733                         /* free tso mbufs */
 734                         if (num_tso_mbufs > 0)
 735                                 rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
 736                         break;
 737                 }
 738                 num_tx++;
 739                 /* free original mbuf */
 740                 rte_pktmbuf_free(mbuf_in);
 741                 /* free tso mbufs */
 742                 if (num_tso_mbufs > 0)
 743                         rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
 744         }
 745
 746         txq->stats.opackets += num_packets;
 747         txq->stats.errs += nb_pkts - num_tx;
 748         txq->stats.obytes += num_tx_bytes;
 749
 750         return num_tx;
 751 }
 752
 753 static const char *
 754 tap_ioctl_req2str(unsigned long request)
 755 {
 756         switch (request) {
 757         case SIOCSIFFLAGS:
 758                 return "SIOCSIFFLAGS";
 759         case SIOCGIFFLAGS:
 760                 return "SIOCGIFFLAGS";
 761         case SIOCGIFHWADDR:
 762                 return "SIOCGIFHWADDR";
 763         case SIOCSIFHWADDR:
 764                 return "SIOCSIFHWADDR";
 765         case SIOCSIFMTU:
 766                 return "SIOCSIFMTU";
 767         }
 768         return "UNKNOWN";
 769 }
 770
 771 static int
 772 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 773           struct ifreq *ifr, int set, enum ioctl_mode mode)
 774 {
 775         short req_flags = ifr->ifr_flags;
 776         int remote = pmd->remote_if_index &&
 777                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 778
 779         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 780                 return 0;
 781         /*
 782          * If there is a remote netdevice, apply ioctl on it, then apply it on
 783          * the tap netdevice.
 784          */
 785 apply:
 786         if (remote)
 787                 strlcpy(ifr->ifr_name, pmd->remote_iface, IFNAMSIZ);
 788         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 789                 strlcpy(ifr->ifr_name, pmd->name, IFNAMSIZ);
 790         switch (request) {
 791         case SIOCSIFFLAGS:
 792                 /* fetch current flags to leave other flags untouched */
 793                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 794                         goto error;
 795                 if (set)
 796                         ifr->ifr_flags |= req_flags;
 797                 else
 798                         ifr->ifr_flags &= ~req_flags;
 799                 break;
 800         case SIOCGIFFLAGS:
 801         case SIOCGIFHWADDR:
 802         case SIOCSIFHWADDR:
 803         case SIOCSIFMTU:
 804                 break;
 805         default:
 806                 TAP_LOG(WARNING, "%s: ioctl() called with wrong arg",
 807                         pmd->name);
 808                 return -EINVAL;
 809         }
 810         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 811                 goto error;
 812         if (remote-- && mode == LOCAL_AND_REMOTE)
 813                 goto apply;
 814         return 0;
 815
 816 error:
 817         TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
 818                 tap_ioctl_req2str(request), strerror(errno), errno);
 819         return -errno;
 820 }
 821
 822 static int
 823 tap_link_set_down(struct rte_eth_dev *dev)
 824 {
 825         struct pmd_internals *pmd = dev->data->dev_private;
 826         struct ifreq ifr = { .ifr_flags = IFF_UP };
 827
 828         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 829         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 830 }
 831
 832 static int
 833 tap_link_set_up(struct rte_eth_dev *dev)
 834 {
 835         struct pmd_internals *pmd = dev->data->dev_private;
 836         struct ifreq ifr = { .ifr_flags = IFF_UP };
 837
 838         dev->data->dev_link.link_status = ETH_LINK_UP;
 839         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 840 }
 841
 842 static int
 843 tap_dev_start(struct rte_eth_dev *dev)
 844 {
 845         int err, i;
 846
 847         err = tap_intr_handle_set(dev, 1);
 848         if (err)
 849                 return err;
 850
 851         err = tap_link_set_up(dev);
 852         if (err)
 853                 return err;
 854
 855         for (i = 0; i < dev->data->nb_tx_queues; i++)
 856                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 857         for (i = 0; i < dev->data->nb_rx_queues; i++)
 858                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 859
 860         return err;
 861 }
 862
 863 /* This function gets called when the current port gets stopped.
 864  */
 865 static void
 866 tap_dev_stop(struct rte_eth_dev *dev)
 867 {
 868         int i;
 869
 870         for (i = 0; i < dev->data->nb_tx_queues; i++)
 871                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 872         for (i = 0; i < dev->data->nb_rx_queues; i++)
 873                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 874
 875         tap_intr_handle_set(dev, 0);
 876         tap_link_set_down(dev);
 877 }
 878
 879 static int
 880 tap_dev_configure(struct rte_eth_dev *dev)
 881 {
 882         struct pmd_internals *pmd = dev->data->dev_private;
 883
 884         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 885                 TAP_LOG(ERR,
 886                         "%s: number of rx queues %d exceeds max num of queues %d",
 887                         dev->device->name,
 888                         dev->data->nb_rx_queues,
 889                         RTE_PMD_TAP_MAX_QUEUES);
 890                 return -1;
 891         }
 892         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 893                 TAP_LOG(ERR,
 894                         "%s: number of tx queues %d exceeds max num of queues %d",
 895                         dev->device->name,
 896                         dev->data->nb_tx_queues,
 897                         RTE_PMD_TAP_MAX_QUEUES);
 898                 return -1;
 899         }
 900
 901         TAP_LOG(INFO, "%s: %s: TX configured queues number: %u",
 902                 dev->device->name, pmd->name, dev->data->nb_tx_queues);
 903
 904         TAP_LOG(INFO, "%s: %s: RX configured queues number: %u",
 905                 dev->device->name, pmd->name, dev->data->nb_rx_queues);
 906
 907         return 0;
 908 }
 909
 910 static uint32_t
 911 tap_dev_speed_capa(void)
 912 {
 913         uint32_t speed = pmd_link.link_speed;
 914         uint32_t capa = 0;
 915
 916         if (speed >= ETH_SPEED_NUM_10M)
 917                 capa |= ETH_LINK_SPEED_10M;
 918         if (speed >= ETH_SPEED_NUM_100M)
 919                 capa |= ETH_LINK_SPEED_100M;
 920         if (speed >= ETH_SPEED_NUM_1G)
 921                 capa |= ETH_LINK_SPEED_1G;
 922         if (speed >= ETH_SPEED_NUM_5G)
 923                 capa |= ETH_LINK_SPEED_2_5G;
 924         if (speed >= ETH_SPEED_NUM_5G)
 925                 capa |= ETH_LINK_SPEED_5G;
 926         if (speed >= ETH_SPEED_NUM_10G)
 927                 capa |= ETH_LINK_SPEED_10G;
 928         if (speed >= ETH_SPEED_NUM_20G)
 929                 capa |= ETH_LINK_SPEED_20G;
 930         if (speed >= ETH_SPEED_NUM_25G)
 931                 capa |= ETH_LINK_SPEED_25G;
 932         if (speed >= ETH_SPEED_NUM_40G)
 933                 capa |= ETH_LINK_SPEED_40G;
 934         if (speed >= ETH_SPEED_NUM_50G)
 935                 capa |= ETH_LINK_SPEED_50G;
 936         if (speed >= ETH_SPEED_NUM_56G)
 937                 capa |= ETH_LINK_SPEED_56G;
 938         if (speed >= ETH_SPEED_NUM_100G)
 939                 capa |= ETH_LINK_SPEED_100G;
 940
 941         return capa;
 942 }
 943
 944 static int
 945 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 946 {
 947         struct pmd_internals *internals = dev->data->dev_private;
 948
 949         dev_info->if_index = internals->if_index;
 950         dev_info->max_mac_addrs = 1;
 951         dev_info->max_rx_pktlen = (uint32_t)RTE_ETHER_MAX_VLAN_FRAME_LEN;
 952         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
 953         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
 954         dev_info->min_rx_bufsize = 0;
 955         dev_info->speed_capa = tap_dev_speed_capa();
 956         dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
 957         dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
 958                                     dev_info->rx_queue_offload_capa;
 959         dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
 960         dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
 961                                     dev_info->tx_queue_offload_capa;
 962         dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
 963         /*
 964          * limitation: TAP supports all of IP, UDP and TCP hash
 965          * functions together and not in partial combinations
 966          */
 967         dev_info->flow_type_rss_offloads = ~TAP_RSS_HF_MASK;
 968
 969         return 0;
 970 }
 971
 972 static int
 973 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 974 {
 975         unsigned int i, imax;
 976         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 977         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 978         unsigned long rx_nombuf = 0, ierrors = 0;
 979         const struct pmd_internals *pmd = dev->data->dev_private;
 980
 981         /* rx queue statistics */
 982         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 983                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 984         for (i = 0; i < imax; i++) {
 985                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 986                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 987                 rx_total += tap_stats->q_ipackets[i];
 988                 rx_bytes_total += tap_stats->q_ibytes[i];
 989                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
 990                 ierrors += pmd->rxq[i].stats.ierrors;
 991         }
 992
 993         /* tx queue statistics */
 994         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 995                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 996
 997         for (i = 0; i < imax; i++) {
 998                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 999                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
1000                 tx_total += tap_stats->q_opackets[i];
1001                 tx_err_total += pmd->txq[i].stats.errs;
1002                 tx_bytes_total += tap_stats->q_obytes[i];
1003         }
1004
1005         tap_stats->ipackets = rx_total;
1006         tap_stats->ibytes = rx_bytes_total;
1007         tap_stats->ierrors = ierrors;
1008         tap_stats->rx_nombuf = rx_nombuf;
1009         tap_stats->opackets = tx_total;
1010         tap_stats->oerrors = tx_err_total;
1011         tap_stats->obytes = tx_bytes_total;
1012         return 0;
1013 }
1014
1015 static int
1016 tap_stats_reset(struct rte_eth_dev *dev)
1017 {
1018         int i;
1019         struct pmd_internals *pmd = dev->data->dev_private;
1020
1021         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1022                 pmd->rxq[i].stats.ipackets = 0;
1023                 pmd->rxq[i].stats.ibytes = 0;
1024                 pmd->rxq[i].stats.ierrors = 0;
1025                 pmd->rxq[i].stats.rx_nombuf = 0;
1026
1027                 pmd->txq[i].stats.opackets = 0;
1028                 pmd->txq[i].stats.errs = 0;
1029                 pmd->txq[i].stats.obytes = 0;
1030         }
1031
1032         return 0;
1033 }
1034
1035 static void
1036 tap_dev_close(struct rte_eth_dev *dev)
1037 {
1038         int i;
1039         struct pmd_internals *internals = dev->data->dev_private;
1040         struct pmd_process_private *process_private = dev->process_private;
1041         struct rx_queue *rxq;
1042
1043         tap_link_set_down(dev);
1044         if (internals->nlsk_fd != -1) {
1045                 tap_flow_flush(dev, NULL);
1046                 tap_flow_implicit_flush(internals, NULL);
1047                 tap_nl_final(internals->nlsk_fd);
1048                 internals->nlsk_fd = -1;
1049         }
1050
1051         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1052                 if (process_private->rxq_fds[i] != -1) {
1053                         rxq = &internals->rxq[i];
1054                         close(process_private->rxq_fds[i]);
1055                         process_private->rxq_fds[i] = -1;
1056                         tap_rxq_pool_free(rxq->pool);
1057                         rte_free(rxq->iovecs);
1058                         rxq->pool = NULL;
1059                         rxq->iovecs = NULL;
1060                 }
1061                 if (process_private->txq_fds[i] != -1) {
1062                         close(process_private->txq_fds[i]);
1063                         process_private->txq_fds[i] = -1;
1064                 }
1065         }
1066
1067         if (internals->remote_if_index) {
1068                 /* Restore initial remote state */
1069                 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
1070                                 &internals->remote_initial_flags);
1071         }
1072
1073         if (internals->ka_fd != -1) {
1074                 close(internals->ka_fd);
1075                 internals->ka_fd = -1;
1076         }
1077         /*
1078          * Since TUN device has no more opened file descriptors
1079          * it will be removed from kernel
1080          */
1081 }
1082
1083 static void
1084 tap_rx_queue_release(void *queue)
1085 {
1086         struct rx_queue *rxq = queue;
1087         struct pmd_process_private *process_private;
1088
1089         if (!rxq)
1090                 return;
1091         process_private = rte_eth_devices[rxq->in_port].process_private;
1092         if (process_private->rxq_fds[rxq->queue_id] > 0) {
1093                 close(process_private->rxq_fds[rxq->queue_id]);
1094                 process_private->rxq_fds[rxq->queue_id] = -1;
1095                 tap_rxq_pool_free(rxq->pool);
1096                 rte_free(rxq->iovecs);
1097                 rxq->pool = NULL;
1098                 rxq->iovecs = NULL;
1099         }
1100 }
1101
1102 static void
1103 tap_tx_queue_release(void *queue)
1104 {
1105         struct tx_queue *txq = queue;
1106         struct pmd_process_private *process_private;
1107
1108         if (!txq)
1109                 return;
1110         process_private = rte_eth_devices[txq->out_port].process_private;
1111
1112         if (process_private->txq_fds[txq->queue_id] > 0) {
1113                 close(process_private->txq_fds[txq->queue_id]);
1114                 process_private->txq_fds[txq->queue_id] = -1;
1115         }
1116 }
1117
1118 static int
1119 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1120 {
1121         struct rte_eth_link *dev_link = &dev->data->dev_link;
1122         struct pmd_internals *pmd = dev->data->dev_private;
1123         struct ifreq ifr = { .ifr_flags = 0 };
1124
1125         if (pmd->remote_if_index) {
1126                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
1127                 if (!(ifr.ifr_flags & IFF_UP) ||
1128                     !(ifr.ifr_flags & IFF_RUNNING)) {
1129                         dev_link->link_status = ETH_LINK_DOWN;
1130                         return 0;
1131                 }
1132         }
1133         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
1134         dev_link->link_status =
1135                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
1136                  ETH_LINK_UP :
1137                  ETH_LINK_DOWN);
1138         return 0;
1139 }
1140
1141 static int
1142 tap_promisc_enable(struct rte_eth_dev *dev)
1143 {
1144         struct pmd_internals *pmd = dev->data->dev_private;
1145         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1146         int ret;
1147
1148         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1149         if (ret != 0)
1150                 return ret;
1151
1152         if (pmd->remote_if_index && !pmd->flow_isolate) {
1153                 dev->data->promiscuous = 1;
1154                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
1155                 if (ret != 0) {
1156                         /* Rollback promisc flag */
1157                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1158                         /*
1159                          * rte_eth_dev_promiscuous_enable() rollback
1160                          * dev->data->promiscuous in the case of failure.
1161                          */
1162                         return ret;
1163                 }
1164         }
1165
1166         return 0;
1167 }
1168
1169 static int
1170 tap_promisc_disable(struct rte_eth_dev *dev)
1171 {
1172         struct pmd_internals *pmd = dev->data->dev_private;
1173         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1174         int ret;
1175
1176         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1177         if (ret != 0)
1178                 return ret;
1179
1180         if (pmd->remote_if_index && !pmd->flow_isolate) {
1181                 dev->data->promiscuous = 0;
1182                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
1183                 if (ret != 0) {
1184                         /* Rollback promisc flag */
1185                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1186                         /*
1187                          * rte_eth_dev_promiscuous_disable() rollback
1188                          * dev->data->promiscuous in the case of failure.
1189                          */
1190                         return ret;
1191                 }
1192         }
1193
1194         return 0;
1195 }
1196
1197 static int
1198 tap_allmulti_enable(struct rte_eth_dev *dev)
1199 {
1200         struct pmd_internals *pmd = dev->data->dev_private;
1201         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1202         int ret;
1203
1204         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1205         if (ret != 0)
1206                 return ret;
1207
1208         if (pmd->remote_if_index && !pmd->flow_isolate) {
1209                 dev->data->all_multicast = 1;
1210                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
1211                 if (ret != 0) {
1212                         /* Rollback allmulti flag */
1213                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1214                         /*
1215                          * rte_eth_dev_allmulticast_enable() rollback
1216                          * dev->data->all_multicast in the case of failure.
1217                          */
1218                         return ret;
1219                 }
1220         }
1221
1222         return 0;
1223 }
1224
1225 static int
1226 tap_allmulti_disable(struct rte_eth_dev *dev)
1227 {
1228         struct pmd_internals *pmd = dev->data->dev_private;
1229         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1230         int ret;
1231
1232         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1233         if (ret != 0)
1234                 return ret;
1235
1236         if (pmd->remote_if_index && !pmd->flow_isolate) {
1237                 dev->data->all_multicast = 0;
1238                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
1239                 if (ret != 0) {
1240                         /* Rollback allmulti flag */
1241                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1242                         /*
1243                          * rte_eth_dev_allmulticast_disable() rollback
1244                          * dev->data->all_multicast in the case of failure.
1245                          */
1246                         return ret;
1247                 }
1248         }
1249
1250         return 0;
1251 }
1252
1253 static int
1254 tap_mac_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1255 {
1256         struct pmd_internals *pmd = dev->data->dev_private;
1257         enum ioctl_mode mode = LOCAL_ONLY;
1258         struct ifreq ifr;
1259         int ret;
1260
1261         if (pmd->type == ETH_TUNTAP_TYPE_TUN) {
1262                 TAP_LOG(ERR, "%s: can't MAC address for TUN",
1263                         dev->device->name);
1264                 return -ENOTSUP;
1265         }
1266
1267         if (rte_is_zero_ether_addr(mac_addr)) {
1268                 TAP_LOG(ERR, "%s: can't set an empty MAC address",
1269                         dev->device->name);
1270                 return -EINVAL;
1271         }
1272         /* Check the actual current MAC address on the tap netdevice */
1273         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
1274         if (ret < 0)
1275                 return ret;
1276         if (rte_is_same_ether_addr(
1277                         (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1278                         mac_addr))
1279                 return 0;
1280         /* Check the current MAC address on the remote */
1281         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
1282         if (ret < 0)
1283                 return ret;
1284         if (!rte_is_same_ether_addr(
1285                         (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1286                         mac_addr))
1287                 mode = LOCAL_AND_REMOTE;
1288         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1289         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, RTE_ETHER_ADDR_LEN);
1290         ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
1291         if (ret < 0)
1292                 return ret;
1293         rte_memcpy(&pmd->eth_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1294         if (pmd->remote_if_index && !pmd->flow_isolate) {
1295                 /* Replace MAC redirection rule after a MAC change */
1296                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
1297                 if (ret < 0) {
1298                         TAP_LOG(ERR,
1299                                 "%s: Couldn't delete MAC redirection rule",
1300                                 dev->device->name);
1301                         return ret;
1302                 }
1303                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1304                 if (ret < 0) {
1305                         TAP_LOG(ERR,
1306                                 "%s: Couldn't add MAC redirection rule",
1307                                 dev->device->name);
1308                         return ret;
1309                 }
1310         }
1311
1312         return 0;
1313 }
1314
1315 static int
1316 tap_gso_ctx_setup(struct rte_gso_ctx *gso_ctx, struct rte_eth_dev *dev)
1317 {
1318         uint32_t gso_types;
1319         char pool_name[64];
1320
1321         /*
1322          * Create private mbuf pool with TAP_GSO_MBUF_SEG_SIZE bytes
1323          * size per mbuf use this pool for both direct and indirect mbufs
1324          */
1325
1326         struct rte_mempool *mp;      /* Mempool for GSO packets */
1327
1328         /* initialize GSO context */
1329         gso_types = DEV_TX_OFFLOAD_TCP_TSO;
1330         snprintf(pool_name, sizeof(pool_name), "mp_%s", dev->device->name);
1331         mp = rte_mempool_lookup((const char *)pool_name);
1332         if (!mp) {
1333                 mp = rte_pktmbuf_pool_create(pool_name, TAP_GSO_MBUFS_NUM,
1334                         TAP_GSO_MBUF_CACHE_SIZE, 0,
1335                         RTE_PKTMBUF_HEADROOM + TAP_GSO_MBUF_SEG_SIZE,
1336                         SOCKET_ID_ANY);
1337                 if (!mp) {
1338                         struct pmd_internals *pmd = dev->data->dev_private;
1339
1340                         TAP_LOG(ERR,
1341                                 "%s: failed to create mbuf pool for device %s\n",
1342                                 pmd->name, dev->device->name);
1343                         return -1;
1344                 }
1345         }
1346
1347         gso_ctx->direct_pool = mp;
1348         gso_ctx->indirect_pool = mp;
1349         gso_ctx->gso_types = gso_types;
1350         gso_ctx->gso_size = 0; /* gso_size is set in tx_burst() per packet */
1351         gso_ctx->flag = 0;
1352
1353         return 0;
1354 }
1355
1356 static int
1357 tap_setup_queue(struct rte_eth_dev *dev,
1358                 struct pmd_internals *internals,
1359                 uint16_t qid,
1360                 int is_rx)
1361 {
1362         int ret;
1363         int *fd;
1364         int *other_fd;
1365         const char *dir;
1366         struct pmd_internals *pmd = dev->data->dev_private;
1367         struct pmd_process_private *process_private = dev->process_private;
1368         struct rx_queue *rx = &internals->rxq[qid];
1369         struct tx_queue *tx = &internals->txq[qid];
1370         struct rte_gso_ctx *gso_ctx;
1371
1372         if (is_rx) {
1373                 fd = &process_private->rxq_fds[qid];
1374                 other_fd = &process_private->txq_fds[qid];
1375                 dir = "rx";
1376                 gso_ctx = NULL;
1377         } else {
1378                 fd = &process_private->txq_fds[qid];
1379                 other_fd = &process_private->rxq_fds[qid];
1380                 dir = "tx";
1381                 gso_ctx = &tx->gso_ctx;
1382         }
1383         if (*fd != -1) {
1384                 /* fd for this queue already exists */
1385                 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1386                         pmd->name, *fd, dir, qid);
1387                 gso_ctx = NULL;
1388         } else if (*other_fd != -1) {
1389                 /* Only other_fd exists. dup it */
1390                 *fd = dup(*other_fd);
1391                 if (*fd < 0) {
1392                         *fd = -1;
1393                         TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1394                         return -1;
1395                 }
1396                 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1397                         pmd->name, *other_fd, dir, qid, *fd);
1398         } else {
1399                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1400                 *fd = tun_alloc(pmd, 0);
1401                 if (*fd < 0) {
1402                         *fd = -1; /* restore original value */
1403                         TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1404                         return -1;
1405                 }
1406                 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1407                         pmd->name, dir, qid, *fd);
1408         }
1409
1410         tx->mtu = &dev->data->mtu;
1411         rx->rxmode = &dev->data->dev_conf.rxmode;
1412         if (gso_ctx) {
1413                 ret = tap_gso_ctx_setup(gso_ctx, dev);
1414                 if (ret)
1415                         return -1;
1416         }
1417
1418         tx->type = pmd->type;
1419
1420         return *fd;
1421 }
1422
1423 static int
1424 tap_rx_queue_setup(struct rte_eth_dev *dev,
1425                    uint16_t rx_queue_id,
1426                    uint16_t nb_rx_desc,
1427                    unsigned int socket_id,
1428                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1429                    struct rte_mempool *mp)
1430 {
1431         struct pmd_internals *internals = dev->data->dev_private;
1432         struct pmd_process_private *process_private = dev->process_private;
1433         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1434         struct rte_mbuf **tmp = &rxq->pool;
1435         long iov_max = sysconf(_SC_IOV_MAX);
1436
1437         if (iov_max <= 0) {
1438                 TAP_LOG(WARNING,
1439                         "_SC_IOV_MAX is not defined. Using %d as default",
1440                         TAP_IOV_DEFAULT_MAX);
1441                 iov_max = TAP_IOV_DEFAULT_MAX;
1442         }
1443         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1444         struct iovec (*iovecs)[nb_desc + 1];
1445         int data_off = RTE_PKTMBUF_HEADROOM;
1446         int ret = 0;
1447         int fd;
1448         int i;
1449
1450         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1451                 TAP_LOG(WARNING,
1452                         "nb_rx_queues %d too small or mempool NULL",
1453                         dev->data->nb_rx_queues);
1454                 return -1;
1455         }
1456
1457         rxq->mp = mp;
1458         rxq->trigger_seen = 1; /* force initial burst */
1459         rxq->in_port = dev->data->port_id;
1460         rxq->queue_id = rx_queue_id;
1461         rxq->nb_rx_desc = nb_desc;
1462         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1463                                     socket_id);
1464         if (!iovecs) {
1465                 TAP_LOG(WARNING,
1466                         "%s: Couldn't allocate %d RX descriptors",
1467                         dev->device->name, nb_desc);
1468                 return -ENOMEM;
1469         }
1470         rxq->iovecs = iovecs;
1471
1472         dev->data->rx_queues[rx_queue_id] = rxq;
1473         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1474         if (fd == -1) {
1475                 ret = fd;
1476                 goto error;
1477         }
1478
1479         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1480         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1481
1482         for (i = 1; i <= nb_desc; i++) {
1483                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1484                 if (!*tmp) {
1485                         TAP_LOG(WARNING,
1486                                 "%s: couldn't allocate memory for queue %d",
1487                                 dev->device->name, rx_queue_id);
1488                         ret = -ENOMEM;
1489                         goto error;
1490                 }
1491                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1492                 (*rxq->iovecs)[i].iov_base =
1493                         (char *)(*tmp)->buf_addr + data_off;
1494                 data_off = 0;
1495                 tmp = &(*tmp)->next;
1496         }
1497
1498         TAP_LOG(DEBUG, "  RX TUNTAP device name %s, qid %d on fd %d",
1499                 internals->name, rx_queue_id,
1500                 process_private->rxq_fds[rx_queue_id]);
1501
1502         return 0;
1503
1504 error:
1505         tap_rxq_pool_free(rxq->pool);
1506         rxq->pool = NULL;
1507         rte_free(rxq->iovecs);
1508         rxq->iovecs = NULL;
1509         return ret;
1510 }
1511
1512 static int
1513 tap_tx_queue_setup(struct rte_eth_dev *dev,
1514                    uint16_t tx_queue_id,
1515                    uint16_t nb_tx_desc __rte_unused,
1516                    unsigned int socket_id __rte_unused,
1517                    const struct rte_eth_txconf *tx_conf)
1518 {
1519         struct pmd_internals *internals = dev->data->dev_private;
1520         struct pmd_process_private *process_private = dev->process_private;
1521         struct tx_queue *txq;
1522         int ret;
1523         uint64_t offloads;
1524
1525         if (tx_queue_id >= dev->data->nb_tx_queues)
1526                 return -1;
1527         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1528         txq = dev->data->tx_queues[tx_queue_id];
1529         txq->out_port = dev->data->port_id;
1530         txq->queue_id = tx_queue_id;
1531
1532         offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1533         txq->csum = !!(offloads &
1534                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
1535                          DEV_TX_OFFLOAD_UDP_CKSUM |
1536                          DEV_TX_OFFLOAD_TCP_CKSUM));
1537
1538         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1539         if (ret == -1)
1540                 return -1;
1541         TAP_LOG(DEBUG,
1542                 "  TX TUNTAP device name %s, qid %d on fd %d csum %s",
1543                 internals->name, tx_queue_id,
1544                 process_private->txq_fds[tx_queue_id],
1545                 txq->csum ? "on" : "off");
1546
1547         return 0;
1548 }
1549
1550 static int
1551 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1552 {
1553         struct pmd_internals *pmd = dev->data->dev_private;
1554         struct ifreq ifr = { .ifr_mtu = mtu };
1555         int err = 0;
1556
1557         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1558         if (!err)
1559                 dev->data->mtu = mtu;
1560
1561         return err;
1562 }
1563
1564 static int
1565 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1566                      struct rte_ether_addr *mc_addr_set __rte_unused,
1567                      uint32_t nb_mc_addr __rte_unused)
1568 {
1569         /*
1570          * Nothing to do actually: the tap has no filtering whatsoever, every
1571          * packet is received.
1572          */
1573         return 0;
1574 }
1575
1576 static int
1577 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1578 {
1579         struct rte_eth_dev *dev = arg;
1580         struct pmd_internals *pmd = dev->data->dev_private;
1581         struct ifinfomsg *info = NLMSG_DATA(nh);
1582
1583         if (nh->nlmsg_type != RTM_NEWLINK ||
1584             (info->ifi_index != pmd->if_index &&
1585              info->ifi_index != pmd->remote_if_index))
1586                 return 0;
1587         return tap_link_update(dev, 0);
1588 }
1589
1590 static void
1591 tap_dev_intr_handler(void *cb_arg)
1592 {
1593         struct rte_eth_dev *dev = cb_arg;
1594         struct pmd_internals *pmd = dev->data->dev_private;
1595
1596         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1597 }
1598
1599 static int
1600 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1601 {
1602         struct pmd_internals *pmd = dev->data->dev_private;
1603         int ret;
1604
1605         /* In any case, disable interrupt if the conf is no longer there. */
1606         if (!dev->data->dev_conf.intr_conf.lsc) {
1607                 if (pmd->intr_handle.fd != -1) {
1608                         goto clean;
1609                 }
1610                 return 0;
1611         }
1612         if (set) {
1613                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1614                 if (unlikely(pmd->intr_handle.fd == -1))
1615                         return -EBADF;
1616                 return rte_intr_callback_register(
1617                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1618         }
1619
1620 clean:
1621         do {
1622                 ret = rte_intr_callback_unregister(&pmd->intr_handle,
1623                         tap_dev_intr_handler, dev);
1624                 if (ret >= 0) {
1625                         break;
1626                 } else if (ret == -EAGAIN) {
1627                         rte_delay_ms(100);
1628                 } else {
1629                         TAP_LOG(ERR, "intr callback unregister failed: %d",
1630                                      ret);
1631                         break;
1632                 }
1633         } while (true);
1634
1635         tap_nl_final(pmd->intr_handle.fd);
1636         pmd->intr_handle.fd = -1;
1637
1638         return 0;
1639 }
1640
1641 static int
1642 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1643 {
1644         int err;
1645
1646         err = tap_lsc_intr_handle_set(dev, set);
1647         if (err < 0) {
1648                 if (!set)
1649                         tap_rx_intr_vec_set(dev, 0);
1650                 return err;
1651         }
1652         err = tap_rx_intr_vec_set(dev, set);
1653         if (err && set)
1654                 tap_lsc_intr_handle_set(dev, 0);
1655         return err;
1656 }
1657
1658 static const uint32_t*
1659 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1660 {
1661         static const uint32_t ptypes[] = {
1662                 RTE_PTYPE_INNER_L2_ETHER,
1663                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1664                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1665                 RTE_PTYPE_INNER_L3_IPV4,
1666                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1667                 RTE_PTYPE_INNER_L3_IPV6,
1668                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1669                 RTE_PTYPE_INNER_L4_FRAG,
1670                 RTE_PTYPE_INNER_L4_UDP,
1671                 RTE_PTYPE_INNER_L4_TCP,
1672                 RTE_PTYPE_INNER_L4_SCTP,
1673                 RTE_PTYPE_L2_ETHER,
1674                 RTE_PTYPE_L2_ETHER_VLAN,
1675                 RTE_PTYPE_L2_ETHER_QINQ,
1676                 RTE_PTYPE_L3_IPV4,
1677                 RTE_PTYPE_L3_IPV4_EXT,
1678                 RTE_PTYPE_L3_IPV6_EXT,
1679                 RTE_PTYPE_L3_IPV6,
1680                 RTE_PTYPE_L4_FRAG,
1681                 RTE_PTYPE_L4_UDP,
1682                 RTE_PTYPE_L4_TCP,
1683                 RTE_PTYPE_L4_SCTP,
1684         };
1685
1686         return ptypes;
1687 }
1688
1689 static int
1690 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1691                   struct rte_eth_fc_conf *fc_conf)
1692 {
1693         fc_conf->mode = RTE_FC_NONE;
1694         return 0;
1695 }
1696
1697 static int
1698 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1699                   struct rte_eth_fc_conf *fc_conf)
1700 {
1701         if (fc_conf->mode != RTE_FC_NONE)
1702                 return -ENOTSUP;
1703         return 0;
1704 }
1705
1706 /**
1707  * DPDK callback to update the RSS hash configuration.
1708  *
1709  * @param dev
1710  *   Pointer to Ethernet device structure.
1711  * @param[in] rss_conf
1712  *   RSS configuration data.
1713  *
1714  * @return
1715  *   0 on success, a negative errno value otherwise and rte_errno is set.
1716  */
1717 static int
1718 tap_rss_hash_update(struct rte_eth_dev *dev,
1719                 struct rte_eth_rss_conf *rss_conf)
1720 {
1721         if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1722                 rte_errno = EINVAL;
1723                 return -rte_errno;
1724         }
1725         if (rss_conf->rss_key && rss_conf->rss_key_len) {
1726                 /*
1727                  * Currently TAP RSS key is hard coded
1728                  * and cannot be updated
1729                  */
1730                 TAP_LOG(ERR,
1731                         "port %u RSS key cannot be updated",
1732                         dev->data->port_id);
1733                 rte_errno = EINVAL;
1734                 return -rte_errno;
1735         }
1736         return 0;
1737 }
1738
1739 static int
1740 tap_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1741 {
1742         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1743
1744         return 0;
1745 }
1746
1747 static int
1748 tap_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1749 {
1750         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1751
1752         return 0;
1753 }
1754
1755 static int
1756 tap_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1757 {
1758         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1759
1760         return 0;
1761 }
1762
1763 static int
1764 tap_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1765 {
1766         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1767
1768         return 0;
1769 }
1770 static const struct eth_dev_ops ops = {
1771         .dev_start              = tap_dev_start,
1772         .dev_stop               = tap_dev_stop,
1773         .dev_close              = tap_dev_close,
1774         .dev_configure          = tap_dev_configure,
1775         .dev_infos_get          = tap_dev_info,
1776         .rx_queue_setup         = tap_rx_queue_setup,
1777         .tx_queue_setup         = tap_tx_queue_setup,
1778         .rx_queue_start         = tap_rx_queue_start,
1779         .tx_queue_start         = tap_tx_queue_start,
1780         .rx_queue_stop          = tap_rx_queue_stop,
1781         .tx_queue_stop          = tap_tx_queue_stop,
1782         .rx_queue_release       = tap_rx_queue_release,
1783         .tx_queue_release       = tap_tx_queue_release,
1784         .flow_ctrl_get          = tap_flow_ctrl_get,
1785         .flow_ctrl_set          = tap_flow_ctrl_set,
1786         .link_update            = tap_link_update,
1787         .dev_set_link_up        = tap_link_set_up,
1788         .dev_set_link_down      = tap_link_set_down,
1789         .promiscuous_enable     = tap_promisc_enable,
1790         .promiscuous_disable    = tap_promisc_disable,
1791         .allmulticast_enable    = tap_allmulti_enable,
1792         .allmulticast_disable   = tap_allmulti_disable,
1793         .mac_addr_set           = tap_mac_set,
1794         .mtu_set                = tap_mtu_set,
1795         .set_mc_addr_list       = tap_set_mc_addr_list,
1796         .stats_get              = tap_stats_get,
1797         .stats_reset            = tap_stats_reset,
1798         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1799         .rss_hash_update        = tap_rss_hash_update,
1800         .filter_ctrl            = tap_dev_filter_ctrl,
1801 };
1802
1803 static const char *tuntap_types[ETH_TUNTAP_TYPE_MAX] = {
1804         "UNKNOWN", "TUN", "TAP"
1805 };
1806
1807 static int
1808 eth_dev_tap_create(struct rte_vdev_device *vdev, const char *tap_name,
1809                    char *remote_iface, struct rte_ether_addr *mac_addr,
1810                    enum rte_tuntap_type type)
1811 {
1812         int numa_node = rte_socket_id();
1813         struct rte_eth_dev *dev;
1814         struct pmd_internals *pmd;
1815         struct pmd_process_private *process_private;
1816         const char *tuntap_name = tuntap_types[type];
1817         struct rte_eth_dev_data *data;
1818         struct ifreq ifr;
1819         int i;
1820
1821         TAP_LOG(DEBUG, "%s device on numa %u", tuntap_name, rte_socket_id());
1822
1823         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1824         if (!dev) {
1825                 TAP_LOG(ERR, "%s Unable to allocate device struct",
1826                                 tuntap_name);
1827                 goto error_exit_nodev;
1828         }
1829
1830         process_private = (struct pmd_process_private *)
1831                 rte_zmalloc_socket(tap_name, sizeof(struct pmd_process_private),
1832                         RTE_CACHE_LINE_SIZE, dev->device->numa_node);
1833
1834         if (process_private == NULL) {
1835                 TAP_LOG(ERR, "Failed to alloc memory for process private");
1836                 return -1;
1837         }
1838         pmd = dev->data->dev_private;
1839         dev->process_private = process_private;
1840         pmd->dev = dev;
1841         strlcpy(pmd->name, tap_name, sizeof(pmd->name));
1842         pmd->type = type;
1843         pmd->ka_fd = -1;
1844         pmd->nlsk_fd = -1;
1845
1846         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1847         if (pmd->ioctl_sock == -1) {
1848                 TAP_LOG(ERR,
1849                         "%s Unable to get a socket for management: %s",
1850                         tuntap_name, strerror(errno));
1851                 goto error_exit;
1852         }
1853
1854         /* Setup some default values */
1855         data = dev->data;
1856         data->dev_private = pmd;
1857         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1858         data->numa_node = numa_node;
1859
1860         data->dev_link = pmd_link;
1861         data->mac_addrs = &pmd->eth_addr;
1862         /* Set the number of RX and TX queues */
1863         data->nb_rx_queues = 0;
1864         data->nb_tx_queues = 0;
1865
1866         dev->dev_ops = &ops;
1867         dev->rx_pkt_burst = pmd_rx_burst;
1868         dev->tx_pkt_burst = pmd_tx_burst;
1869
1870         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1871         pmd->intr_handle.fd = -1;
1872         dev->intr_handle = &pmd->intr_handle;
1873
1874         /* Presetup the fds to -1 as being not valid */
1875         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1876                 process_private->rxq_fds[i] = -1;
1877                 process_private->txq_fds[i] = -1;
1878         }
1879
1880         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1881                 if (rte_is_zero_ether_addr(mac_addr))
1882                         rte_eth_random_addr((uint8_t *)&pmd->eth_addr);
1883                 else
1884                         rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1885         }
1886
1887         /*
1888          * Allocate a TUN device keep-alive file descriptor that will only be
1889          * closed when the TUN device itself is closed or removed.
1890          * This keep-alive file descriptor will guarantee that the TUN device
1891          * exists even when all of its queues are closed
1892          */
1893         pmd->ka_fd = tun_alloc(pmd, 1);
1894         if (pmd->ka_fd == -1) {
1895                 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
1896                 goto error_exit;
1897         }
1898         TAP_LOG(DEBUG, "allocated %s", pmd->name);
1899
1900         ifr.ifr_mtu = dev->data->mtu;
1901         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1902                 goto error_exit;
1903
1904         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1905                 memset(&ifr, 0, sizeof(struct ifreq));
1906                 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1907                 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1908                                 RTE_ETHER_ADDR_LEN);
1909                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1910                         goto error_exit;
1911         }
1912
1913         /*
1914          * Set up everything related to rte_flow:
1915          * - netlink socket
1916          * - tap / remote if_index
1917          * - mandatory QDISCs
1918          * - rte_flow actual/implicit lists
1919          * - implicit rules
1920          */
1921         pmd->nlsk_fd = tap_nl_init(0);
1922         if (pmd->nlsk_fd == -1) {
1923                 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
1924                         pmd->name);
1925                 goto disable_rte_flow;
1926         }
1927         pmd->if_index = if_nametoindex(pmd->name);
1928         if (!pmd->if_index) {
1929                 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
1930                 goto disable_rte_flow;
1931         }
1932         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1933                 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
1934                         pmd->name);
1935                 goto disable_rte_flow;
1936         }
1937         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1938                 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1939                         pmd->name);
1940                 goto disable_rte_flow;
1941         }
1942         LIST_INIT(&pmd->flows);
1943
1944         if (strlen(remote_iface)) {
1945                 pmd->remote_if_index = if_nametoindex(remote_iface);
1946                 if (!pmd->remote_if_index) {
1947                         TAP_LOG(ERR, "%s: failed to get %s if_index.",
1948                                 pmd->name, remote_iface);
1949                         goto error_remote;
1950                 }
1951                 strlcpy(pmd->remote_iface, remote_iface, RTE_ETH_NAME_MAX_LEN);
1952
1953                 /* Save state of remote device */
1954                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1955
1956                 /* Replicate remote MAC address */
1957                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1958                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1959                                 pmd->name, pmd->remote_iface);
1960                         goto error_remote;
1961                 }
1962                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1963                            RTE_ETHER_ADDR_LEN);
1964                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1965                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1966                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1967                                 pmd->name, remote_iface);
1968                         goto error_remote;
1969                 }
1970
1971                 /*
1972                  * Flush usually returns negative value because it tries to
1973                  * delete every QDISC (and on a running device, one QDISC at
1974                  * least is needed). Ignore negative return value.
1975                  */
1976                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1977                 if (qdisc_create_ingress(pmd->nlsk_fd,
1978                                          pmd->remote_if_index) < 0) {
1979                         TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1980                                 pmd->remote_iface);
1981                         goto error_remote;
1982                 }
1983                 LIST_INIT(&pmd->implicit_flows);
1984                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1985                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1986                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1987                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1988                         TAP_LOG(ERR,
1989                                 "%s: failed to create implicit rules.",
1990                                 pmd->name);
1991                         goto error_remote;
1992                 }
1993         }
1994
1995         rte_eth_dev_probing_finish(dev);
1996         return 0;
1997
1998 disable_rte_flow:
1999         TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
2000                 strerror(errno), errno);
2001         if (strlen(remote_iface)) {
2002                 TAP_LOG(ERR, "Remote feature requires flow support.");
2003                 goto error_exit;
2004         }
2005         rte_eth_dev_probing_finish(dev);
2006         return 0;
2007
2008 error_remote:
2009         TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
2010                 strerror(errno), errno);
2011         tap_flow_implicit_flush(pmd, NULL);
2012
2013 error_exit:
2014         if (pmd->nlsk_fd != -1)
2015                 close(pmd->nlsk_fd);
2016         if (pmd->ka_fd != -1)
2017                 close(pmd->ka_fd);
2018         if (pmd->ioctl_sock != -1)
2019                 close(pmd->ioctl_sock);
2020         /* mac_addrs must not be freed alone because part of dev_private */
2021         dev->data->mac_addrs = NULL;
2022         rte_eth_dev_release_port(dev);
2023
2024 error_exit_nodev:
2025         TAP_LOG(ERR, "%s Unable to initialize %s",
2026                 tuntap_name, rte_vdev_device_name(vdev));
2027
2028         return -EINVAL;
2029 }
2030
2031 /* make sure name is a possible Linux network device name */
2032 static bool
2033 is_valid_iface(const char *name)
2034 {
2035         if (*name == '\0')
2036                 return false;
2037
2038         if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
2039                 return false;
2040
2041         while (*name) {
2042                 if (*name == '/' || *name == ':' || isspace(*name))
2043                         return false;
2044                 name++;
2045         }
2046         return true;
2047 }
2048
2049 static int
2050 set_interface_name(const char *key __rte_unused,
2051                    const char *value,
2052                    void *extra_args)
2053 {
2054         char *name = (char *)extra_args;
2055
2056         if (value) {
2057                 if (!is_valid_iface(value)) {
2058                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2059                                 value);
2060                         return -1;
2061                 }
2062                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2063         } else {
2064                 /* use tap%d which causes kernel to choose next available */
2065                 strlcpy(name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2066         }
2067         return 0;
2068 }
2069
2070 static int
2071 set_remote_iface(const char *key __rte_unused,
2072                  const char *value,
2073                  void *extra_args)
2074 {
2075         char *name = (char *)extra_args;
2076
2077         if (value) {
2078                 if (!is_valid_iface(value)) {
2079                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2080                                 value);
2081                         return -1;
2082                 }
2083                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2084         }
2085
2086         return 0;
2087 }
2088
2089 static int parse_user_mac(struct rte_ether_addr *user_mac,
2090                 const char *value)
2091 {
2092         unsigned int index = 0;
2093         char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
2094
2095         if (user_mac == NULL || value == NULL)
2096                 return 0;
2097
2098         strlcpy(mac_temp, value, sizeof(mac_temp));
2099         mac_byte = strtok(mac_temp, ":");
2100
2101         while ((mac_byte != NULL) &&
2102                         (strlen(mac_byte) <= 2) &&
2103                         (strlen(mac_byte) == strspn(mac_byte,
2104                                         ETH_TAP_CMP_MAC_FMT))) {
2105                 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
2106                 mac_byte = strtok(NULL, ":");
2107         }
2108
2109         return index;
2110 }
2111
2112 static int
2113 set_mac_type(const char *key __rte_unused,
2114              const char *value,
2115              void *extra_args)
2116 {
2117         struct rte_ether_addr *user_mac = extra_args;
2118
2119         if (!value)
2120                 return 0;
2121
2122         if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
2123                 static int iface_idx;
2124
2125                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
2126                 memcpy((char *)user_mac->addr_bytes, "\0dtap",
2127                         RTE_ETHER_ADDR_LEN);
2128                 user_mac->addr_bytes[RTE_ETHER_ADDR_LEN - 1] =
2129                         iface_idx++ + '0';
2130                 goto success;
2131         }
2132
2133         if (parse_user_mac(user_mac, value) != 6)
2134                 goto error;
2135 success:
2136         TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
2137         return 0;
2138
2139 error:
2140         TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
2141                 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
2142         return -1;
2143 }
2144
2145 /*
2146  * Open a TUN interface device. TUN PMD
2147  * 1) sets tap_type as false
2148  * 2) intakes iface as argument.
2149  * 3) as interface is virtual set speed to 10G
2150  */
2151 static int
2152 rte_pmd_tun_probe(struct rte_vdev_device *dev)
2153 {
2154         const char *name, *params;
2155         int ret;
2156         struct rte_kvargs *kvlist = NULL;
2157         char tun_name[RTE_ETH_NAME_MAX_LEN];
2158         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2159         struct rte_eth_dev *eth_dev;
2160
2161         name = rte_vdev_device_name(dev);
2162         params = rte_vdev_device_args(dev);
2163         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2164
2165         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
2166             strlen(params) == 0) {
2167                 eth_dev = rte_eth_dev_attach_secondary(name);
2168                 if (!eth_dev) {
2169                         TAP_LOG(ERR, "Failed to probe %s", name);
2170                         return -1;
2171                 }
2172                 eth_dev->dev_ops = &ops;
2173                 eth_dev->device = &dev->device;
2174                 rte_eth_dev_probing_finish(eth_dev);
2175                 return 0;
2176         }
2177
2178         /* use tun%d which causes kernel to choose next available */
2179         strlcpy(tun_name, DEFAULT_TUN_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2180
2181         if (params && (params[0] != '\0')) {
2182                 TAP_LOG(DEBUG, "parameters (%s)", params);
2183
2184                 kvlist = rte_kvargs_parse(params, valid_arguments);
2185                 if (kvlist) {
2186                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2187                                 ret = rte_kvargs_process(kvlist,
2188                                         ETH_TAP_IFACE_ARG,
2189                                         &set_interface_name,
2190                                         tun_name);
2191
2192                                 if (ret == -1)
2193                                         goto leave;
2194                         }
2195                 }
2196         }
2197         pmd_link.link_speed = ETH_SPEED_NUM_10G;
2198
2199         TAP_LOG(DEBUG, "Initializing pmd_tun for %s", name);
2200
2201         ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0,
2202                                  ETH_TUNTAP_TYPE_TUN);
2203
2204 leave:
2205         if (ret == -1) {
2206                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2207                         name, tun_name);
2208         }
2209         rte_kvargs_free(kvlist);
2210
2211         return ret;
2212 }
2213
2214 /* Request queue file descriptors from secondary to primary. */
2215 static int
2216 tap_mp_attach_queues(const char *port_name, struct rte_eth_dev *dev)
2217 {
2218         int ret;
2219         struct timespec timeout = {.tv_sec = 1, .tv_nsec = 0};
2220         struct rte_mp_msg request, *reply;
2221         struct rte_mp_reply replies;
2222         struct ipc_queues *request_param = (struct ipc_queues *)request.param;
2223         struct ipc_queues *reply_param;
2224         struct pmd_process_private *process_private = dev->process_private;
2225         int queue, fd_iterator;
2226
2227         /* Prepare the request */
2228         memset(&request, 0, sizeof(request));
2229         strlcpy(request.name, TAP_MP_KEY, sizeof(request.name));
2230         strlcpy(request_param->port_name, port_name,
2231                 sizeof(request_param->port_name));
2232         request.len_param = sizeof(*request_param);
2233         /* Send request and receive reply */
2234         ret = rte_mp_request_sync(&request, &replies, &timeout);
2235         if (ret < 0 || replies.nb_received != 1) {
2236                 TAP_LOG(ERR, "Failed to request queues from primary: %d",
2237                         rte_errno);
2238                 return -1;
2239         }
2240         reply = &replies.msgs[0];
2241         reply_param = (struct ipc_queues *)reply->param;
2242         TAP_LOG(DEBUG, "Received IPC reply for %s", reply_param->port_name);
2243
2244         /* Attach the queues from received file descriptors */
2245         if (reply_param->rxq_count + reply_param->txq_count != reply->num_fds) {
2246                 TAP_LOG(ERR, "Unexpected number of fds received");
2247                 return -1;
2248         }
2249
2250         dev->data->nb_rx_queues = reply_param->rxq_count;
2251         dev->data->nb_tx_queues = reply_param->txq_count;
2252         fd_iterator = 0;
2253         for (queue = 0; queue < reply_param->rxq_count; queue++)
2254                 process_private->rxq_fds[queue] = reply->fds[fd_iterator++];
2255         for (queue = 0; queue < reply_param->txq_count; queue++)
2256                 process_private->txq_fds[queue] = reply->fds[fd_iterator++];
2257         free(reply);
2258         return 0;
2259 }
2260
2261 /* Send the queue file descriptors from the primary process to secondary. */
2262 static int
2263 tap_mp_sync_queues(const struct rte_mp_msg *request, const void *peer)
2264 {
2265         struct rte_eth_dev *dev;
2266         struct pmd_process_private *process_private;
2267         struct rte_mp_msg reply;
2268         const struct ipc_queues *request_param =
2269                 (const struct ipc_queues *)request->param;
2270         struct ipc_queues *reply_param =
2271                 (struct ipc_queues *)reply.param;
2272         uint16_t port_id;
2273         int queue;
2274         int ret;
2275
2276         /* Get requested port */
2277         TAP_LOG(DEBUG, "Received IPC request for %s", request_param->port_name);
2278         ret = rte_eth_dev_get_port_by_name(request_param->port_name, &port_id);
2279         if (ret) {
2280                 TAP_LOG(ERR, "Failed to get port id for %s",
2281                         request_param->port_name);
2282                 return -1;
2283         }
2284         dev = &rte_eth_devices[port_id];
2285         process_private = dev->process_private;
2286
2287         /* Fill file descriptors for all queues */
2288         reply.num_fds = 0;
2289         reply_param->rxq_count = 0;
2290         if (dev->data->nb_rx_queues + dev->data->nb_tx_queues >
2291                         RTE_MP_MAX_FD_NUM){
2292                 TAP_LOG(ERR, "Number of rx/tx queues exceeds max number of fds");
2293                 return -1;
2294         }
2295
2296         for (queue = 0; queue < dev->data->nb_rx_queues; queue++) {
2297                 reply.fds[reply.num_fds++] = process_private->rxq_fds[queue];
2298                 reply_param->rxq_count++;
2299         }
2300         RTE_ASSERT(reply_param->rxq_count == dev->data->nb_rx_queues);
2301
2302         reply_param->txq_count = 0;
2303         for (queue = 0; queue < dev->data->nb_tx_queues; queue++) {
2304                 reply.fds[reply.num_fds++] = process_private->txq_fds[queue];
2305                 reply_param->txq_count++;
2306         }
2307         RTE_ASSERT(reply_param->txq_count == dev->data->nb_tx_queues);
2308
2309         /* Send reply */
2310         strlcpy(reply.name, request->name, sizeof(reply.name));
2311         strlcpy(reply_param->port_name, request_param->port_name,
2312                 sizeof(reply_param->port_name));
2313         reply.len_param = sizeof(*reply_param);
2314         if (rte_mp_reply(&reply, peer) < 0) {
2315                 TAP_LOG(ERR, "Failed to reply an IPC request to sync queues");
2316                 return -1;
2317         }
2318         return 0;
2319 }
2320
2321 /* Open a TAP interface device.
2322  */
2323 static int
2324 rte_pmd_tap_probe(struct rte_vdev_device *dev)
2325 {
2326         const char *name, *params;
2327         int ret;
2328         struct rte_kvargs *kvlist = NULL;
2329         int speed;
2330         char tap_name[RTE_ETH_NAME_MAX_LEN];
2331         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2332         struct rte_ether_addr user_mac = { .addr_bytes = {0} };
2333         struct rte_eth_dev *eth_dev;
2334         int tap_devices_count_increased = 0;
2335
2336         name = rte_vdev_device_name(dev);
2337         params = rte_vdev_device_args(dev);
2338
2339         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
2340                 eth_dev = rte_eth_dev_attach_secondary(name);
2341                 if (!eth_dev) {
2342                         TAP_LOG(ERR, "Failed to probe %s", name);
2343                         return -1;
2344                 }
2345                 eth_dev->dev_ops = &ops;
2346                 eth_dev->device = &dev->device;
2347                 eth_dev->rx_pkt_burst = pmd_rx_burst;
2348                 eth_dev->tx_pkt_burst = pmd_tx_burst;
2349                 if (!rte_eal_primary_proc_alive(NULL)) {
2350                         TAP_LOG(ERR, "Primary process is missing");
2351                         return -1;
2352                 }
2353                 eth_dev->process_private = (struct pmd_process_private *)
2354                         rte_zmalloc_socket(name,
2355                                 sizeof(struct pmd_process_private),
2356                                 RTE_CACHE_LINE_SIZE,
2357                                 eth_dev->device->numa_node);
2358                 if (eth_dev->process_private == NULL) {
2359                         TAP_LOG(ERR,
2360                                 "Failed to alloc memory for process private");
2361                         return -1;
2362                 }
2363
2364                 ret = tap_mp_attach_queues(name, eth_dev);
2365                 if (ret != 0)
2366                         return -1;
2367                 rte_eth_dev_probing_finish(eth_dev);
2368                 return 0;
2369         }
2370
2371         speed = ETH_SPEED_NUM_10G;
2372
2373         /* use tap%d which causes kernel to choose next available */
2374         strlcpy(tap_name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2375         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2376
2377         if (params && (params[0] != '\0')) {
2378                 TAP_LOG(DEBUG, "parameters (%s)", params);
2379
2380                 kvlist = rte_kvargs_parse(params, valid_arguments);
2381                 if (kvlist) {
2382                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2383                                 ret = rte_kvargs_process(kvlist,
2384                                                          ETH_TAP_IFACE_ARG,
2385                                                          &set_interface_name,
2386                                                          tap_name);
2387                                 if (ret == -1)
2388                                         goto leave;
2389                         }
2390
2391                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
2392                                 ret = rte_kvargs_process(kvlist,
2393                                                          ETH_TAP_REMOTE_ARG,
2394                                                          &set_remote_iface,
2395                                                          remote_iface);
2396                                 if (ret == -1)
2397                                         goto leave;
2398                         }
2399
2400                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
2401                                 ret = rte_kvargs_process(kvlist,
2402                                                          ETH_TAP_MAC_ARG,
2403                                                          &set_mac_type,
2404                                                          &user_mac);
2405                                 if (ret == -1)
2406                                         goto leave;
2407                         }
2408                 }
2409         }
2410         pmd_link.link_speed = speed;
2411
2412         TAP_LOG(DEBUG, "Initializing pmd_tap for %s", name);
2413
2414         /* Register IPC feed callback */
2415         if (!tap_devices_count) {
2416                 ret = rte_mp_action_register(TAP_MP_KEY, tap_mp_sync_queues);
2417                 if (ret < 0 && rte_errno != ENOTSUP) {
2418                         TAP_LOG(ERR, "tap: Failed to register IPC callback: %s",
2419                                 strerror(rte_errno));
2420                         goto leave;
2421                 }
2422         }
2423         tap_devices_count++;
2424         tap_devices_count_increased = 1;
2425         ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
2426                 ETH_TUNTAP_TYPE_TAP);
2427
2428 leave:
2429         if (ret == -1) {
2430                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2431                         name, tap_name);
2432                 if (tap_devices_count_increased == 1) {
2433                         if (tap_devices_count == 1)
2434                                 rte_mp_action_unregister(TAP_MP_KEY);
2435                         tap_devices_count--;
2436                 }
2437         }
2438         rte_kvargs_free(kvlist);
2439
2440         return ret;
2441 }
2442
2443 /* detach a TUNTAP device.
2444  */
2445 static int
2446 rte_pmd_tap_remove(struct rte_vdev_device *dev)
2447 {
2448         struct rte_eth_dev *eth_dev = NULL;
2449         struct pmd_internals *internals;
2450
2451         /* find the ethdev entry */
2452         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
2453         if (!eth_dev)
2454                 return -ENODEV;
2455
2456         /* mac_addrs must not be freed alone because part of dev_private */
2457         eth_dev->data->mac_addrs = NULL;
2458
2459         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2460                 return rte_eth_dev_release_port(eth_dev);
2461
2462         tap_dev_close(eth_dev);
2463
2464         internals = eth_dev->data->dev_private;
2465         TAP_LOG(DEBUG, "Closing %s Ethernet device on numa %u",
2466                 tuntap_types[internals->type], rte_socket_id());
2467
2468         close(internals->ioctl_sock);
2469         rte_free(eth_dev->process_private);
2470         if (tap_devices_count == 1)
2471                 rte_mp_action_unregister(TAP_MP_KEY);
2472         tap_devices_count--;
2473         rte_eth_dev_release_port(eth_dev);
2474
2475         return 0;
2476 }
2477
2478 static struct rte_vdev_driver pmd_tun_drv = {
2479         .probe = rte_pmd_tun_probe,
2480         .remove = rte_pmd_tap_remove,
2481 };
2482
2483 static struct rte_vdev_driver pmd_tap_drv = {
2484         .probe = rte_pmd_tap_probe,
2485         .remove = rte_pmd_tap_remove,
2486 };
2487
2488 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
2489 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
2490 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
2491 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
2492                               ETH_TAP_IFACE_ARG "=<string> ");
2493 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
2494                               ETH_TAP_IFACE_ARG "=<string> "
2495                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
2496                               ETH_TAP_REMOTE_ARG "=<string>");
2497 int tap_logtype;
2498
2499 RTE_INIT(tap_init_log)
2500 {
2501         tap_logtype = rte_log_register("pmd.net.tap");
2502         if (tap_logtype >= 0)
2503                 rte_log_set_level(tap_logtype, RTE_LOG_NOTICE);
2504 }