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 *tuntap_types[ETH_TUNTAP_TYPE_MAX] = {
  76         "UNKNOWN", "TUN", "TAP"
  77 };
  78
  79 static const char *valid_arguments[] = {
  80         ETH_TAP_IFACE_ARG,
  81         ETH_TAP_REMOTE_ARG,
  82         ETH_TAP_MAC_ARG,
  83         NULL
  84 };
  85
  86 static volatile uint32_t tap_trigger;   /* Rx trigger */
  87
  88 static struct rte_eth_link pmd_link = {
  89         .link_speed = ETH_SPEED_NUM_10G,
  90         .link_duplex = ETH_LINK_FULL_DUPLEX,
  91         .link_status = ETH_LINK_DOWN,
  92         .link_autoneg = ETH_LINK_FIXED,
  93 };
  94
  95 static void
  96 tap_trigger_cb(int sig __rte_unused)
  97 {
  98         /* Valid trigger values are nonzero */
  99         tap_trigger = (tap_trigger + 1) | 0x80000000;
 100 }
 101
 102 /* Specifies on what netdevices the ioctl should be applied */
 103 enum ioctl_mode {
 104         LOCAL_AND_REMOTE,
 105         LOCAL_ONLY,
 106         REMOTE_ONLY,
 107 };
 108
 109 /* Message header to synchronize queues via IPC */
 110 struct ipc_queues {
 111         char port_name[RTE_DEV_NAME_MAX_LEN];
 112         int rxq_count;
 113         int txq_count;
 114         /*
 115          * The file descriptors are in the dedicated part
 116          * of the Unix message to be translated by the kernel.
 117          */
 118 };
 119
 120 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
 121
 122 /**
 123  * Tun/Tap allocation routine
 124  *
 125  * @param[in] pmd
 126  *   Pointer to private structure.
 127  *
 128  * @param[in] is_keepalive
 129  *   Keepalive flag
 130  *
 131  * @return
 132  *   -1 on failure, fd on success
 133  */
 134 static int
 135 tun_alloc(struct pmd_internals *pmd, int is_keepalive)
 136 {
 137         struct ifreq ifr;
 138 #ifdef IFF_MULTI_QUEUE
 139         unsigned int features;
 140 #endif
 141         int fd, signo, flags;
 142
 143         memset(&ifr, 0, sizeof(struct ifreq));
 144
 145         /*
 146          * Do not set IFF_NO_PI as packet information header will be needed
 147          * to check if a received packet has been truncated.
 148          */
 149         ifr.ifr_flags = (pmd->type == ETH_TUNTAP_TYPE_TAP) ?
 150                 IFF_TAP : IFF_TUN | IFF_POINTOPOINT;
 151         strlcpy(ifr.ifr_name, pmd->name, IFNAMSIZ);
 152
 153         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 154         if (fd < 0) {
 155                 TAP_LOG(ERR, "Unable to open %s interface", TUN_TAP_DEV_PATH);
 156                 goto error;
 157         }
 158
 159 #ifdef IFF_MULTI_QUEUE
 160         /* Grab the TUN features to verify we can work multi-queue */
 161         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 162                 TAP_LOG(ERR, "unable to get TUN/TAP features");
 163                 goto error;
 164         }
 165         TAP_LOG(DEBUG, "%s Features %08x", TUN_TAP_DEV_PATH, features);
 166
 167         if (features & IFF_MULTI_QUEUE) {
 168                 TAP_LOG(DEBUG, "  Multi-queue support for %d queues",
 169                         RTE_PMD_TAP_MAX_QUEUES);
 170                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 171         } else
 172 #endif
 173         {
 174                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 175                 TAP_LOG(DEBUG, "  Single queue only support");
 176         }
 177
 178         /* Set the TUN/TAP configuration and set the name if needed */
 179         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 180                 TAP_LOG(WARNING, "Unable to set TUNSETIFF for %s: %s",
 181                         ifr.ifr_name, strerror(errno));
 182                 goto error;
 183         }
 184
 185         /*
 186          * Name passed to kernel might be wildcard like dtun%d
 187          * and need to find the resulting device.
 188          */
 189         TAP_LOG(DEBUG, "Device name is '%s'", ifr.ifr_name);
 190         strlcpy(pmd->name, ifr.ifr_name, RTE_ETH_NAME_MAX_LEN);
 191
 192         if (is_keepalive) {
 193                 /*
 194                  * Detach the TUN/TAP keep-alive queue
 195                  * to avoid traffic through it
 196                  */
 197                 ifr.ifr_flags = IFF_DETACH_QUEUE;
 198                 if (ioctl(fd, TUNSETQUEUE, (void *)&ifr) < 0) {
 199                         TAP_LOG(WARNING,
 200                                 "Unable to detach keep-alive queue for %s: %s",
 201                                 ifr.ifr_name, strerror(errno));
 202                         goto error;
 203                 }
 204         }
 205
 206         flags = fcntl(fd, F_GETFL);
 207         if (flags == -1) {
 208                 TAP_LOG(WARNING,
 209                         "Unable to get %s current flags\n",
 210                         ifr.ifr_name);
 211                 goto error;
 212         }
 213
 214         /* Always set the file descriptor to non-blocking */
 215         flags |= O_NONBLOCK;
 216         if (fcntl(fd, F_SETFL, flags) < 0) {
 217                 TAP_LOG(WARNING,
 218                         "Unable to set %s to nonblocking: %s",
 219                         ifr.ifr_name, strerror(errno));
 220                 goto error;
 221         }
 222
 223         /* Find a free realtime signal */
 224         for (signo = SIGRTMIN + 1; signo < SIGRTMAX; signo++) {
 225                 struct sigaction sa;
 226
 227                 if (sigaction(signo, NULL, &sa) == -1) {
 228                         TAP_LOG(WARNING,
 229                                 "Unable to get current rt-signal %d handler",
 230                                 signo);
 231                         goto error;
 232                 }
 233
 234                 /* Already have the handler we want on this signal  */
 235                 if (sa.sa_handler == tap_trigger_cb)
 236                         break;
 237
 238                 /* Is handler in use by application */
 239                 if (sa.sa_handler != SIG_DFL) {
 240                         TAP_LOG(DEBUG,
 241                                 "Skipping used rt-signal %d", signo);
 242                         continue;
 243                 }
 244
 245                 sa = (struct sigaction) {
 246                         .sa_flags = SA_RESTART,
 247                         .sa_handler = tap_trigger_cb,
 248                 };
 249
 250                 if (sigaction(signo, &sa, NULL) == -1) {
 251                         TAP_LOG(WARNING,
 252                                 "Unable to set rt-signal %d handler\n", signo);
 253                         goto error;
 254                 }
 255
 256                 /* Found a good signal to use */
 257                 TAP_LOG(DEBUG,
 258                         "Using rt-signal %d", signo);
 259                 break;
 260         }
 261
 262         if (signo == SIGRTMAX) {
 263                 TAP_LOG(WARNING, "All rt-signals are in use\n");
 264
 265                 /* Disable trigger globally in case of error */
 266                 tap_trigger = 0;
 267                 TAP_LOG(NOTICE, "No Rx trigger signal available\n");
 268         } else {
 269                 /* Enable signal on file descriptor */
 270                 if (fcntl(fd, F_SETSIG, signo) < 0) {
 271                         TAP_LOG(WARNING, "Unable to set signo %d for fd %d: %s",
 272                                 signo, fd, strerror(errno));
 273                         goto error;
 274                 }
 275                 if (fcntl(fd, F_SETFL, flags | O_ASYNC) < 0) {
 276                         TAP_LOG(WARNING, "Unable to set fcntl flags: %s",
 277                                 strerror(errno));
 278                         goto error;
 279                 }
 280
 281                 if (fcntl(fd, F_SETOWN, getpid()) < 0) {
 282                         TAP_LOG(WARNING, "Unable to set fcntl owner: %s",
 283                                 strerror(errno));
 284                         goto error;
 285                 }
 286         }
 287         return fd;
 288
 289 error:
 290         if (fd >= 0)
 291                 close(fd);
 292         return -1;
 293 }
 294
 295 static void
 296 tap_verify_csum(struct rte_mbuf *mbuf)
 297 {
 298         uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
 299         uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
 300         uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
 301         unsigned int l2_len = sizeof(struct rte_ether_hdr);
 302         unsigned int l3_len;
 303         uint16_t cksum = 0;
 304         void *l3_hdr;
 305         void *l4_hdr;
 306
 307         if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
 308                 l2_len += 4;
 309         else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
 310                 l2_len += 8;
 311         /* Don't verify checksum for packets with discontinuous L2 header */
 312         if (unlikely(l2_len + sizeof(struct rte_ipv4_hdr) >
 313                      rte_pktmbuf_data_len(mbuf)))
 314                 return;
 315         l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
 316         if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 317                 struct rte_ipv4_hdr *iph = l3_hdr;
 318
 319                 l3_len = rte_ipv4_hdr_len(iph);
 320                 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
 321                         return;
 322                 /* check that the total length reported by header is not
 323                  * greater than the total received size
 324                  */
 325                 if (l2_len + rte_be_to_cpu_16(iph->total_length) >
 326                                 rte_pktmbuf_data_len(mbuf))
 327                         return;
 328
 329                 cksum = ~rte_raw_cksum(iph, l3_len);
 330                 mbuf->ol_flags |= cksum ?
 331                         PKT_RX_IP_CKSUM_BAD :
 332                         PKT_RX_IP_CKSUM_GOOD;
 333         } else if (l3 == RTE_PTYPE_L3_IPV6) {
 334                 struct rte_ipv6_hdr *iph = l3_hdr;
 335
 336                 l3_len = sizeof(struct rte_ipv6_hdr);
 337                 /* check that the total length reported by header is not
 338                  * greater than the total received size
 339                  */
 340                 if (l2_len + l3_len + rte_be_to_cpu_16(iph->payload_len) >
 341                                 rte_pktmbuf_data_len(mbuf))
 342                         return;
 343         } else {
 344                 /* IPv6 extensions are not supported */
 345                 return;
 346         }
 347         if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
 348                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 349                 /* Don't verify checksum for multi-segment packets. */
 350                 if (mbuf->nb_segs > 1)
 351                         return;
 352                 if (l3 == RTE_PTYPE_L3_IPV4)
 353                         cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
 354                 else if (l3 == RTE_PTYPE_L3_IPV6)
 355                         cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
 356                 mbuf->ol_flags |= cksum ?
 357                         PKT_RX_L4_CKSUM_BAD :
 358                         PKT_RX_L4_CKSUM_GOOD;
 359         }
 360 }
 361
 362 static uint64_t
 363 tap_rx_offload_get_port_capa(void)
 364 {
 365         /*
 366          * No specific port Rx offload capabilities.
 367          */
 368         return 0;
 369 }
 370
 371 static uint64_t
 372 tap_rx_offload_get_queue_capa(void)
 373 {
 374         return DEV_RX_OFFLOAD_SCATTER |
 375                DEV_RX_OFFLOAD_IPV4_CKSUM |
 376                DEV_RX_OFFLOAD_UDP_CKSUM |
 377                DEV_RX_OFFLOAD_TCP_CKSUM;
 378 }
 379
 380 static void
 381 tap_rxq_pool_free(struct rte_mbuf *pool)
 382 {
 383         struct rte_mbuf *mbuf = pool;
 384         uint16_t nb_segs = 1;
 385
 386         if (mbuf == NULL)
 387                 return;
 388
 389         while (mbuf->next) {
 390                 mbuf = mbuf->next;
 391                 nb_segs++;
 392         }
 393         pool->nb_segs = nb_segs;
 394         rte_pktmbuf_free(pool);
 395 }
 396
 397 /* Callback to handle the rx burst of packets to the correct interface and
 398  * file descriptor(s) in a multi-queue setup.
 399  */
 400 static uint16_t
 401 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 402 {
 403         struct rx_queue *rxq = queue;
 404         struct pmd_process_private *process_private;
 405         uint16_t num_rx;
 406         unsigned long num_rx_bytes = 0;
 407         uint32_t trigger = tap_trigger;
 408
 409         if (trigger == rxq->trigger_seen)
 410                 return 0;
 411
 412         process_private = rte_eth_devices[rxq->in_port].process_private;
 413         for (num_rx = 0; num_rx < nb_pkts; ) {
 414                 struct rte_mbuf *mbuf = rxq->pool;
 415                 struct rte_mbuf *seg = NULL;
 416                 struct rte_mbuf *new_tail = NULL;
 417                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 418                 int len;
 419
 420                 len = readv(process_private->rxq_fds[rxq->queue_id],
 421                         *rxq->iovecs,
 422                         1 + (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
 423                              rxq->nb_rx_desc : 1));
 424                 if (len < (int)sizeof(struct tun_pi))
 425                         break;
 426
 427                 /* Packet couldn't fit in the provided mbuf */
 428                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 429                         rxq->stats.ierrors++;
 430                         continue;
 431                 }
 432
 433                 len -= sizeof(struct tun_pi);
 434
 435                 mbuf->pkt_len = len;
 436                 mbuf->port = rxq->in_port;
 437                 while (1) {
 438                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 439
 440                         if (unlikely(!buf)) {
 441                                 rxq->stats.rx_nombuf++;
 442                                 /* No new buf has been allocated: do nothing */
 443                                 if (!new_tail || !seg)
 444                                         goto end;
 445
 446                                 seg->next = NULL;
 447                                 tap_rxq_pool_free(mbuf);
 448
 449                                 goto end;
 450                         }
 451                         seg = seg ? seg->next : mbuf;
 452                         if (rxq->pool == mbuf)
 453                                 rxq->pool = buf;
 454                         if (new_tail)
 455                                 new_tail->next = buf;
 456                         new_tail = buf;
 457                         new_tail->next = seg->next;
 458
 459                         /* iovecs[0] is reserved for packet info (pi) */
 460                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 461                                 buf->buf_len - data_off;
 462                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 463                                 (char *)buf->buf_addr + data_off;
 464
 465                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 466                         seg->data_off = data_off;
 467
 468                         len -= seg->data_len;
 469                         if (len <= 0)
 470                                 break;
 471                         mbuf->nb_segs++;
 472                         /* First segment has headroom, not the others */
 473                         data_off = 0;
 474                 }
 475                 seg->next = NULL;
 476                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 477                                                       RTE_PTYPE_ALL_MASK);
 478                 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
 479                         tap_verify_csum(mbuf);
 480
 481                 /* account for the receive frame */
 482                 bufs[num_rx++] = mbuf;
 483                 num_rx_bytes += mbuf->pkt_len;
 484         }
 485 end:
 486         rxq->stats.ipackets += num_rx;
 487         rxq->stats.ibytes += num_rx_bytes;
 488
 489         if (trigger && num_rx < nb_pkts)
 490                 rxq->trigger_seen = trigger;
 491
 492         return num_rx;
 493 }
 494
 495 static uint64_t
 496 tap_tx_offload_get_port_capa(void)
 497 {
 498         /*
 499          * No specific port Tx offload capabilities.
 500          */
 501         return 0;
 502 }
 503
 504 static uint64_t
 505 tap_tx_offload_get_queue_capa(void)
 506 {
 507         return DEV_TX_OFFLOAD_MULTI_SEGS |
 508                DEV_TX_OFFLOAD_IPV4_CKSUM |
 509                DEV_TX_OFFLOAD_UDP_CKSUM |
 510                DEV_TX_OFFLOAD_TCP_CKSUM |
 511                DEV_TX_OFFLOAD_TCP_TSO;
 512 }
 513
 514 /* Finalize l4 checksum calculation */
 515 static void
 516 tap_tx_l4_cksum(uint16_t *l4_cksum, uint16_t l4_phdr_cksum,
 517                 uint32_t l4_raw_cksum)
 518 {
 519         if (l4_cksum) {
 520                 uint32_t cksum;
 521
 522                 cksum = __rte_raw_cksum_reduce(l4_raw_cksum);
 523                 cksum += l4_phdr_cksum;
 524
 525                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 526                 cksum = (~cksum) & 0xffff;
 527                 if (cksum == 0)
 528                         cksum = 0xffff;
 529                 *l4_cksum = cksum;
 530         }
 531 }
 532
 533 /* Accumaulate L4 raw checksums */
 534 static void
 535 tap_tx_l4_add_rcksum(char *l4_data, unsigned int l4_len, uint16_t *l4_cksum,
 536                         uint32_t *l4_raw_cksum)
 537 {
 538         if (l4_cksum == NULL)
 539                 return;
 540
 541         *l4_raw_cksum = __rte_raw_cksum(l4_data, l4_len, *l4_raw_cksum);
 542 }
 543
 544 /* L3 and L4 pseudo headers checksum offloads */
 545 static void
 546 tap_tx_l3_cksum(char *packet, uint64_t ol_flags, unsigned int l2_len,
 547                 unsigned int l3_len, unsigned int l4_len, uint16_t **l4_cksum,
 548                 uint16_t *l4_phdr_cksum, uint32_t *l4_raw_cksum)
 549 {
 550         void *l3_hdr = packet + l2_len;
 551
 552         if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
 553                 struct rte_ipv4_hdr *iph = l3_hdr;
 554                 uint16_t cksum;
 555
 556                 iph->hdr_checksum = 0;
 557                 cksum = rte_raw_cksum(iph, l3_len);
 558                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 559         }
 560         if (ol_flags & PKT_TX_L4_MASK) {
 561                 void *l4_hdr;
 562
 563                 l4_hdr = packet + l2_len + l3_len;
 564                 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
 565                         *l4_cksum = &((struct rte_udp_hdr *)l4_hdr)->dgram_cksum;
 566                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 567                         *l4_cksum = &((struct rte_tcp_hdr *)l4_hdr)->cksum;
 568                 else
 569                         return;
 570                 **l4_cksum = 0;
 571                 if (ol_flags & PKT_TX_IPV4)
 572                         *l4_phdr_cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 573                 else
 574                         *l4_phdr_cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 575                 *l4_raw_cksum = __rte_raw_cksum(l4_hdr, l4_len, 0);
 576         }
 577 }
 578
 579 static inline int
 580 tap_write_mbufs(struct tx_queue *txq, uint16_t num_mbufs,
 581                         struct rte_mbuf **pmbufs,
 582                         uint16_t *num_packets, unsigned long *num_tx_bytes)
 583 {
 584         int i;
 585         uint16_t l234_hlen;
 586         struct pmd_process_private *process_private;
 587
 588         process_private = rte_eth_devices[txq->out_port].process_private;
 589
 590         for (i = 0; i < num_mbufs; i++) {
 591                 struct rte_mbuf *mbuf = pmbufs[i];
 592                 struct iovec iovecs[mbuf->nb_segs + 2];
 593                 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
 594                 struct rte_mbuf *seg = mbuf;
 595                 char m_copy[mbuf->data_len];
 596                 int proto;
 597                 int n;
 598                 int j;
 599                 int k; /* current index in iovecs for copying segments */
 600                 uint16_t seg_len; /* length of first segment */
 601                 uint16_t nb_segs;
 602                 uint16_t *l4_cksum; /* l4 checksum (pseudo header + payload) */
 603                 uint32_t l4_raw_cksum = 0; /* TCP/UDP payload raw checksum */
 604                 uint16_t l4_phdr_cksum = 0; /* TCP/UDP pseudo header checksum */
 605                 uint16_t is_cksum = 0; /* in case cksum should be offloaded */
 606
 607                 l4_cksum = NULL;
 608                 if (txq->type == ETH_TUNTAP_TYPE_TUN) {
 609                         /*
 610                          * TUN and TAP are created with IFF_NO_PI disabled.
 611                          * For TUN PMD this mandatory as fields are used by
 612                          * Kernel tun.c to determine whether its IP or non IP
 613                          * packets.
 614                          *
 615                          * The logic fetches the first byte of data from mbuf
 616                          * then compares whether its v4 or v6. If first byte
 617                          * is 4 or 6, then protocol field is updated.
 618                          */
 619                         char *buff_data = rte_pktmbuf_mtod(seg, void *);
 620                         proto = (*buff_data & 0xf0);
 621                         pi.proto = (proto == 0x40) ?
 622                                 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4) :
 623                                 ((proto == 0x60) ?
 624                                         rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6) :
 625                                         0x00);
 626                 }
 627
 628                 k = 0;
 629                 iovecs[k].iov_base = &pi;
 630                 iovecs[k].iov_len = sizeof(pi);
 631                 k++;
 632
 633                 nb_segs = mbuf->nb_segs;
 634                 if (txq->csum &&
 635                     ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 636                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 637                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
 638                         is_cksum = 1;
 639
 640                         /* Support only packets with at least layer 4
 641                          * header included in the first segment
 642                          */
 643                         seg_len = rte_pktmbuf_data_len(mbuf);
 644                         l234_hlen = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
 645                         if (seg_len < l234_hlen)
 646                                 return -1;
 647
 648                         /* To change checksums, work on a * copy of l2, l3
 649                          * headers + l4 pseudo header
 650                          */
 651                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 652                                         l234_hlen);
 653                         tap_tx_l3_cksum(m_copy, mbuf->ol_flags,
 654                                        mbuf->l2_len, mbuf->l3_len, mbuf->l4_len,
 655                                        &l4_cksum, &l4_phdr_cksum,
 656                                        &l4_raw_cksum);
 657                         iovecs[k].iov_base = m_copy;
 658                         iovecs[k].iov_len = l234_hlen;
 659                         k++;
 660
 661                         /* Update next iovecs[] beyond l2, l3, l4 headers */
 662                         if (seg_len > l234_hlen) {
 663                                 iovecs[k].iov_len = seg_len - l234_hlen;
 664                                 iovecs[k].iov_base =
 665                                         rte_pktmbuf_mtod(seg, char *) +
 666                                                 l234_hlen;
 667                                 tap_tx_l4_add_rcksum(iovecs[k].iov_base,
 668                                         iovecs[k].iov_len, l4_cksum,
 669                                         &l4_raw_cksum);
 670                                 k++;
 671                                 nb_segs++;
 672                         }
 673                         seg = seg->next;
 674                 }
 675
 676                 for (j = k; j <= nb_segs; j++) {
 677                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 678                         iovecs[j].iov_base = rte_pktmbuf_mtod(seg, void *);
 679                         if (is_cksum)
 680                                 tap_tx_l4_add_rcksum(iovecs[j].iov_base,
 681                                         iovecs[j].iov_len, l4_cksum,
 682                                         &l4_raw_cksum);
 683                         seg = seg->next;
 684                 }
 685
 686                 if (is_cksum)
 687                         tap_tx_l4_cksum(l4_cksum, l4_phdr_cksum, l4_raw_cksum);
 688
 689                 /* copy the tx frame data */
 690                 n = writev(process_private->txq_fds[txq->queue_id], iovecs, j);
 691                 if (n <= 0)
 692                         return -1;
 693
 694                 (*num_packets)++;
 695                 (*num_tx_bytes) += rte_pktmbuf_pkt_len(mbuf);
 696         }
 697         return 0;
 698 }
 699
 700 /* Callback to handle sending packets from the tap interface
 701  */
 702 static uint16_t
 703 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 704 {
 705         struct tx_queue *txq = queue;
 706         uint16_t num_tx = 0;
 707         uint16_t num_packets = 0;
 708         unsigned long num_tx_bytes = 0;
 709         uint32_t max_size;
 710         int i;
 711
 712         if (unlikely(nb_pkts == 0))
 713                 return 0;
 714
 715         struct rte_mbuf *gso_mbufs[MAX_GSO_MBUFS];
 716         max_size = *txq->mtu + (RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN + 4);
 717         for (i = 0; i < nb_pkts; i++) {
 718                 struct rte_mbuf *mbuf_in = bufs[num_tx];
 719                 struct rte_mbuf **mbuf;
 720                 uint16_t num_mbufs = 0;
 721                 uint16_t tso_segsz = 0;
 722                 int ret;
 723                 int num_tso_mbufs;
 724                 uint16_t hdrs_len;
 725                 uint64_t tso;
 726
 727                 tso = mbuf_in->ol_flags & PKT_TX_TCP_SEG;
 728                 if (tso) {
 729                         struct rte_gso_ctx *gso_ctx = &txq->gso_ctx;
 730
 731                         /* TCP segmentation implies TCP checksum offload */
 732                         mbuf_in->ol_flags |= PKT_TX_TCP_CKSUM;
 733
 734                         /* gso size is calculated without RTE_ETHER_CRC_LEN */
 735                         hdrs_len = mbuf_in->l2_len + mbuf_in->l3_len +
 736                                         mbuf_in->l4_len;
 737                         tso_segsz = mbuf_in->tso_segsz + hdrs_len;
 738                         if (unlikely(tso_segsz == hdrs_len) ||
 739                                 tso_segsz > *txq->mtu) {
 740                                 txq->stats.errs++;
 741                                 break;
 742                         }
 743                         gso_ctx->gso_size = tso_segsz;
 744                         /* 'mbuf_in' packet to segment */
 745                         num_tso_mbufs = rte_gso_segment(mbuf_in,
 746                                 gso_ctx, /* gso control block */
 747                                 (struct rte_mbuf **)&gso_mbufs, /* out mbufs */
 748                                 RTE_DIM(gso_mbufs)); /* max tso mbufs */
 749
 750                         /* ret contains the number of new created mbufs */
 751                         if (num_tso_mbufs < 0)
 752                                 break;
 753
 754                         mbuf = gso_mbufs;
 755                         num_mbufs = num_tso_mbufs;
 756                 } else {
 757                         /* stats.errs will be incremented */
 758                         if (rte_pktmbuf_pkt_len(mbuf_in) > max_size)
 759                                 break;
 760
 761                         /* ret 0 indicates no new mbufs were created */
 762                         num_tso_mbufs = 0;
 763                         mbuf = &mbuf_in;
 764                         num_mbufs = 1;
 765                 }
 766
 767                 ret = tap_write_mbufs(txq, num_mbufs, mbuf,
 768                                 &num_packets, &num_tx_bytes);
 769                 if (ret == -1) {
 770                         txq->stats.errs++;
 771                         /* free tso mbufs */
 772                         if (num_tso_mbufs > 0)
 773                                 rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
 774                         break;
 775                 }
 776                 num_tx++;
 777                 /* free original mbuf */
 778                 rte_pktmbuf_free(mbuf_in);
 779                 /* free tso mbufs */
 780                 if (num_tso_mbufs > 0)
 781                         rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
 782         }
 783
 784         txq->stats.opackets += num_packets;
 785         txq->stats.errs += nb_pkts - num_tx;
 786         txq->stats.obytes += num_tx_bytes;
 787
 788         return num_tx;
 789 }
 790
 791 static const char *
 792 tap_ioctl_req2str(unsigned long request)
 793 {
 794         switch (request) {
 795         case SIOCSIFFLAGS:
 796                 return "SIOCSIFFLAGS";
 797         case SIOCGIFFLAGS:
 798                 return "SIOCGIFFLAGS";
 799         case SIOCGIFHWADDR:
 800                 return "SIOCGIFHWADDR";
 801         case SIOCSIFHWADDR:
 802                 return "SIOCSIFHWADDR";
 803         case SIOCSIFMTU:
 804                 return "SIOCSIFMTU";
 805         }
 806         return "UNKNOWN";
 807 }
 808
 809 static int
 810 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 811           struct ifreq *ifr, int set, enum ioctl_mode mode)
 812 {
 813         short req_flags = ifr->ifr_flags;
 814         int remote = pmd->remote_if_index &&
 815                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 816
 817         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 818                 return 0;
 819         /*
 820          * If there is a remote netdevice, apply ioctl on it, then apply it on
 821          * the tap netdevice.
 822          */
 823 apply:
 824         if (remote)
 825                 strlcpy(ifr->ifr_name, pmd->remote_iface, IFNAMSIZ);
 826         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 827                 strlcpy(ifr->ifr_name, pmd->name, IFNAMSIZ);
 828         switch (request) {
 829         case SIOCSIFFLAGS:
 830                 /* fetch current flags to leave other flags untouched */
 831                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 832                         goto error;
 833                 if (set)
 834                         ifr->ifr_flags |= req_flags;
 835                 else
 836                         ifr->ifr_flags &= ~req_flags;
 837                 break;
 838         case SIOCGIFFLAGS:
 839         case SIOCGIFHWADDR:
 840         case SIOCSIFHWADDR:
 841         case SIOCSIFMTU:
 842                 break;
 843         default:
 844                 TAP_LOG(WARNING, "%s: ioctl() called with wrong arg",
 845                         pmd->name);
 846                 return -EINVAL;
 847         }
 848         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 849                 goto error;
 850         if (remote-- && mode == LOCAL_AND_REMOTE)
 851                 goto apply;
 852         return 0;
 853
 854 error:
 855         TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
 856                 tap_ioctl_req2str(request), strerror(errno), errno);
 857         return -errno;
 858 }
 859
 860 static int
 861 tap_link_set_down(struct rte_eth_dev *dev)
 862 {
 863         struct pmd_internals *pmd = dev->data->dev_private;
 864         struct ifreq ifr = { .ifr_flags = IFF_UP };
 865
 866         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 867         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 868 }
 869
 870 static int
 871 tap_link_set_up(struct rte_eth_dev *dev)
 872 {
 873         struct pmd_internals *pmd = dev->data->dev_private;
 874         struct ifreq ifr = { .ifr_flags = IFF_UP };
 875
 876         dev->data->dev_link.link_status = ETH_LINK_UP;
 877         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 878 }
 879
 880 static int
 881 tap_dev_start(struct rte_eth_dev *dev)
 882 {
 883         int err, i;
 884
 885         err = tap_intr_handle_set(dev, 1);
 886         if (err)
 887                 return err;
 888
 889         err = tap_link_set_up(dev);
 890         if (err)
 891                 return err;
 892
 893         for (i = 0; i < dev->data->nb_tx_queues; i++)
 894                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 895         for (i = 0; i < dev->data->nb_rx_queues; i++)
 896                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 897
 898         return err;
 899 }
 900
 901 /* This function gets called when the current port gets stopped.
 902  */
 903 static void
 904 tap_dev_stop(struct rte_eth_dev *dev)
 905 {
 906         int i;
 907
 908         for (i = 0; i < dev->data->nb_tx_queues; i++)
 909                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 910         for (i = 0; i < dev->data->nb_rx_queues; i++)
 911                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 912
 913         tap_intr_handle_set(dev, 0);
 914         tap_link_set_down(dev);
 915 }
 916
 917 static int
 918 tap_dev_configure(struct rte_eth_dev *dev)
 919 {
 920         struct pmd_internals *pmd = dev->data->dev_private;
 921
 922         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 923                 TAP_LOG(ERR,
 924                         "%s: number of rx queues %d exceeds max num of queues %d",
 925                         dev->device->name,
 926                         dev->data->nb_rx_queues,
 927                         RTE_PMD_TAP_MAX_QUEUES);
 928                 return -1;
 929         }
 930         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 931                 TAP_LOG(ERR,
 932                         "%s: number of tx queues %d exceeds max num of queues %d",
 933                         dev->device->name,
 934                         dev->data->nb_tx_queues,
 935                         RTE_PMD_TAP_MAX_QUEUES);
 936                 return -1;
 937         }
 938
 939         TAP_LOG(INFO, "%s: %s: TX configured queues number: %u",
 940                 dev->device->name, pmd->name, dev->data->nb_tx_queues);
 941
 942         TAP_LOG(INFO, "%s: %s: RX configured queues number: %u",
 943                 dev->device->name, pmd->name, dev->data->nb_rx_queues);
 944
 945         return 0;
 946 }
 947
 948 static uint32_t
 949 tap_dev_speed_capa(void)
 950 {
 951         uint32_t speed = pmd_link.link_speed;
 952         uint32_t capa = 0;
 953
 954         if (speed >= ETH_SPEED_NUM_10M)
 955                 capa |= ETH_LINK_SPEED_10M;
 956         if (speed >= ETH_SPEED_NUM_100M)
 957                 capa |= ETH_LINK_SPEED_100M;
 958         if (speed >= ETH_SPEED_NUM_1G)
 959                 capa |= ETH_LINK_SPEED_1G;
 960         if (speed >= ETH_SPEED_NUM_5G)
 961                 capa |= ETH_LINK_SPEED_2_5G;
 962         if (speed >= ETH_SPEED_NUM_5G)
 963                 capa |= ETH_LINK_SPEED_5G;
 964         if (speed >= ETH_SPEED_NUM_10G)
 965                 capa |= ETH_LINK_SPEED_10G;
 966         if (speed >= ETH_SPEED_NUM_20G)
 967                 capa |= ETH_LINK_SPEED_20G;
 968         if (speed >= ETH_SPEED_NUM_25G)
 969                 capa |= ETH_LINK_SPEED_25G;
 970         if (speed >= ETH_SPEED_NUM_40G)
 971                 capa |= ETH_LINK_SPEED_40G;
 972         if (speed >= ETH_SPEED_NUM_50G)
 973                 capa |= ETH_LINK_SPEED_50G;
 974         if (speed >= ETH_SPEED_NUM_56G)
 975                 capa |= ETH_LINK_SPEED_56G;
 976         if (speed >= ETH_SPEED_NUM_100G)
 977                 capa |= ETH_LINK_SPEED_100G;
 978
 979         return capa;
 980 }
 981
 982 static int
 983 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 984 {
 985         struct pmd_internals *internals = dev->data->dev_private;
 986
 987         dev_info->if_index = internals->if_index;
 988         dev_info->max_mac_addrs = 1;
 989         dev_info->max_rx_pktlen = (uint32_t)RTE_ETHER_MAX_VLAN_FRAME_LEN;
 990         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
 991         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
 992         dev_info->min_rx_bufsize = 0;
 993         dev_info->speed_capa = tap_dev_speed_capa();
 994         dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
 995         dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
 996                                     dev_info->rx_queue_offload_capa;
 997         dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
 998         dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
 999                                     dev_info->tx_queue_offload_capa;
1000         dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
1001         /*
1002          * limitation: TAP supports all of IP, UDP and TCP hash
1003          * functions together and not in partial combinations
1004          */
1005         dev_info->flow_type_rss_offloads = ~TAP_RSS_HF_MASK;
1006
1007         return 0;
1008 }
1009
1010 static int
1011 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
1012 {
1013         unsigned int i, imax;
1014         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
1015         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
1016         unsigned long rx_nombuf = 0, ierrors = 0;
1017         const struct pmd_internals *pmd = dev->data->dev_private;
1018
1019         /* rx queue statistics */
1020         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
1021                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
1022         for (i = 0; i < imax; i++) {
1023                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
1024                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
1025                 rx_total += tap_stats->q_ipackets[i];
1026                 rx_bytes_total += tap_stats->q_ibytes[i];
1027                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
1028                 ierrors += pmd->rxq[i].stats.ierrors;
1029         }
1030
1031         /* tx queue statistics */
1032         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
1033                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
1034
1035         for (i = 0; i < imax; i++) {
1036                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
1037                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
1038                 tx_total += tap_stats->q_opackets[i];
1039                 tx_err_total += pmd->txq[i].stats.errs;
1040                 tx_bytes_total += tap_stats->q_obytes[i];
1041         }
1042
1043         tap_stats->ipackets = rx_total;
1044         tap_stats->ibytes = rx_bytes_total;
1045         tap_stats->ierrors = ierrors;
1046         tap_stats->rx_nombuf = rx_nombuf;
1047         tap_stats->opackets = tx_total;
1048         tap_stats->oerrors = tx_err_total;
1049         tap_stats->obytes = tx_bytes_total;
1050         return 0;
1051 }
1052
1053 static int
1054 tap_stats_reset(struct rte_eth_dev *dev)
1055 {
1056         int i;
1057         struct pmd_internals *pmd = dev->data->dev_private;
1058
1059         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1060                 pmd->rxq[i].stats.ipackets = 0;
1061                 pmd->rxq[i].stats.ibytes = 0;
1062                 pmd->rxq[i].stats.ierrors = 0;
1063                 pmd->rxq[i].stats.rx_nombuf = 0;
1064
1065                 pmd->txq[i].stats.opackets = 0;
1066                 pmd->txq[i].stats.errs = 0;
1067                 pmd->txq[i].stats.obytes = 0;
1068         }
1069
1070         return 0;
1071 }
1072
1073 static int
1074 tap_dev_close(struct rte_eth_dev *dev)
1075 {
1076         int i;
1077         struct pmd_internals *internals = dev->data->dev_private;
1078         struct pmd_process_private *process_private = dev->process_private;
1079         struct rx_queue *rxq;
1080
1081         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1082                 rte_free(dev->process_private);
1083                 return 0;
1084         }
1085
1086         tap_link_set_down(dev);
1087         if (internals->nlsk_fd != -1) {
1088                 tap_flow_flush(dev, NULL);
1089                 tap_flow_implicit_flush(internals, NULL);
1090                 tap_nl_final(internals->nlsk_fd);
1091                 internals->nlsk_fd = -1;
1092         }
1093
1094         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1095                 if (process_private->rxq_fds[i] != -1) {
1096                         rxq = &internals->rxq[i];
1097                         close(process_private->rxq_fds[i]);
1098                         process_private->rxq_fds[i] = -1;
1099                         tap_rxq_pool_free(rxq->pool);
1100                         rte_free(rxq->iovecs);
1101                         rxq->pool = NULL;
1102                         rxq->iovecs = NULL;
1103                 }
1104                 if (process_private->txq_fds[i] != -1) {
1105                         close(process_private->txq_fds[i]);
1106                         process_private->txq_fds[i] = -1;
1107                 }
1108         }
1109
1110         if (internals->remote_if_index) {
1111                 /* Restore initial remote state */
1112                 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
1113                                 &internals->remote_initial_flags);
1114         }
1115
1116         rte_mempool_free(internals->gso_ctx_mp);
1117         internals->gso_ctx_mp = NULL;
1118
1119         if (internals->ka_fd != -1) {
1120                 close(internals->ka_fd);
1121                 internals->ka_fd = -1;
1122         }
1123
1124         /* mac_addrs must not be freed alone because part of dev_private */
1125         dev->data->mac_addrs = NULL;
1126
1127         internals = dev->data->dev_private;
1128         TAP_LOG(DEBUG, "Closing %s Ethernet device on numa %u",
1129                 tuntap_types[internals->type], rte_socket_id());
1130
1131         if (internals->ioctl_sock != -1) {
1132                 close(internals->ioctl_sock);
1133                 internals->ioctl_sock = -1;
1134         }
1135         rte_free(dev->process_private);
1136         dev->process_private = NULL;
1137         if (tap_devices_count == 1)
1138                 rte_mp_action_unregister(TAP_MP_KEY);
1139         tap_devices_count--;
1140         /*
1141          * Since TUN device has no more opened file descriptors
1142          * it will be removed from kernel
1143          */
1144
1145         return 0;
1146 }
1147
1148 static void
1149 tap_rx_queue_release(void *queue)
1150 {
1151         struct rx_queue *rxq = queue;
1152         struct pmd_process_private *process_private;
1153
1154         if (!rxq)
1155                 return;
1156         process_private = rte_eth_devices[rxq->in_port].process_private;
1157         if (process_private->rxq_fds[rxq->queue_id] != -1) {
1158                 close(process_private->rxq_fds[rxq->queue_id]);
1159                 process_private->rxq_fds[rxq->queue_id] = -1;
1160                 tap_rxq_pool_free(rxq->pool);
1161                 rte_free(rxq->iovecs);
1162                 rxq->pool = NULL;
1163                 rxq->iovecs = NULL;
1164         }
1165 }
1166
1167 static void
1168 tap_tx_queue_release(void *queue)
1169 {
1170         struct tx_queue *txq = queue;
1171         struct pmd_process_private *process_private;
1172
1173         if (!txq)
1174                 return;
1175         process_private = rte_eth_devices[txq->out_port].process_private;
1176
1177         if (process_private->txq_fds[txq->queue_id] != -1) {
1178                 close(process_private->txq_fds[txq->queue_id]);
1179                 process_private->txq_fds[txq->queue_id] = -1;
1180         }
1181 }
1182
1183 static int
1184 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1185 {
1186         struct rte_eth_link *dev_link = &dev->data->dev_link;
1187         struct pmd_internals *pmd = dev->data->dev_private;
1188         struct ifreq ifr = { .ifr_flags = 0 };
1189
1190         if (pmd->remote_if_index) {
1191                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
1192                 if (!(ifr.ifr_flags & IFF_UP) ||
1193                     !(ifr.ifr_flags & IFF_RUNNING)) {
1194                         dev_link->link_status = ETH_LINK_DOWN;
1195                         return 0;
1196                 }
1197         }
1198         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
1199         dev_link->link_status =
1200                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
1201                  ETH_LINK_UP :
1202                  ETH_LINK_DOWN);
1203         return 0;
1204 }
1205
1206 static int
1207 tap_promisc_enable(struct rte_eth_dev *dev)
1208 {
1209         struct pmd_internals *pmd = dev->data->dev_private;
1210         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1211         int ret;
1212
1213         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1214         if (ret != 0)
1215                 return ret;
1216
1217         if (pmd->remote_if_index && !pmd->flow_isolate) {
1218                 dev->data->promiscuous = 1;
1219                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
1220                 if (ret != 0) {
1221                         /* Rollback promisc flag */
1222                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1223                         /*
1224                          * rte_eth_dev_promiscuous_enable() rollback
1225                          * dev->data->promiscuous in the case of failure.
1226                          */
1227                         return ret;
1228                 }
1229         }
1230
1231         return 0;
1232 }
1233
1234 static int
1235 tap_promisc_disable(struct rte_eth_dev *dev)
1236 {
1237         struct pmd_internals *pmd = dev->data->dev_private;
1238         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1239         int ret;
1240
1241         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1242         if (ret != 0)
1243                 return ret;
1244
1245         if (pmd->remote_if_index && !pmd->flow_isolate) {
1246                 dev->data->promiscuous = 0;
1247                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
1248                 if (ret != 0) {
1249                         /* Rollback promisc flag */
1250                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1251                         /*
1252                          * rte_eth_dev_promiscuous_disable() rollback
1253                          * dev->data->promiscuous in the case of failure.
1254                          */
1255                         return ret;
1256                 }
1257         }
1258
1259         return 0;
1260 }
1261
1262 static int
1263 tap_allmulti_enable(struct rte_eth_dev *dev)
1264 {
1265         struct pmd_internals *pmd = dev->data->dev_private;
1266         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1267         int ret;
1268
1269         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1270         if (ret != 0)
1271                 return ret;
1272
1273         if (pmd->remote_if_index && !pmd->flow_isolate) {
1274                 dev->data->all_multicast = 1;
1275                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
1276                 if (ret != 0) {
1277                         /* Rollback allmulti flag */
1278                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1279                         /*
1280                          * rte_eth_dev_allmulticast_enable() rollback
1281                          * dev->data->all_multicast in the case of failure.
1282                          */
1283                         return ret;
1284                 }
1285         }
1286
1287         return 0;
1288 }
1289
1290 static int
1291 tap_allmulti_disable(struct rte_eth_dev *dev)
1292 {
1293         struct pmd_internals *pmd = dev->data->dev_private;
1294         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1295         int ret;
1296
1297         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1298         if (ret != 0)
1299                 return ret;
1300
1301         if (pmd->remote_if_index && !pmd->flow_isolate) {
1302                 dev->data->all_multicast = 0;
1303                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
1304                 if (ret != 0) {
1305                         /* Rollback allmulti flag */
1306                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1307                         /*
1308                          * rte_eth_dev_allmulticast_disable() rollback
1309                          * dev->data->all_multicast in the case of failure.
1310                          */
1311                         return ret;
1312                 }
1313         }
1314
1315         return 0;
1316 }
1317
1318 static int
1319 tap_mac_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1320 {
1321         struct pmd_internals *pmd = dev->data->dev_private;
1322         enum ioctl_mode mode = LOCAL_ONLY;
1323         struct ifreq ifr;
1324         int ret;
1325
1326         if (pmd->type == ETH_TUNTAP_TYPE_TUN) {
1327                 TAP_LOG(ERR, "%s: can't MAC address for TUN",
1328                         dev->device->name);
1329                 return -ENOTSUP;
1330         }
1331
1332         if (rte_is_zero_ether_addr(mac_addr)) {
1333                 TAP_LOG(ERR, "%s: can't set an empty MAC address",
1334                         dev->device->name);
1335                 return -EINVAL;
1336         }
1337         /* Check the actual current MAC address on the tap netdevice */
1338         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
1339         if (ret < 0)
1340                 return ret;
1341         if (rte_is_same_ether_addr(
1342                         (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1343                         mac_addr))
1344                 return 0;
1345         /* Check the current MAC address on the remote */
1346         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
1347         if (ret < 0)
1348                 return ret;
1349         if (!rte_is_same_ether_addr(
1350                         (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1351                         mac_addr))
1352                 mode = LOCAL_AND_REMOTE;
1353         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1354         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, RTE_ETHER_ADDR_LEN);
1355         ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
1356         if (ret < 0)
1357                 return ret;
1358         rte_memcpy(&pmd->eth_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1359         if (pmd->remote_if_index && !pmd->flow_isolate) {
1360                 /* Replace MAC redirection rule after a MAC change */
1361                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
1362                 if (ret < 0) {
1363                         TAP_LOG(ERR,
1364                                 "%s: Couldn't delete MAC redirection rule",
1365                                 dev->device->name);
1366                         return ret;
1367                 }
1368                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1369                 if (ret < 0) {
1370                         TAP_LOG(ERR,
1371                                 "%s: Couldn't add MAC redirection rule",
1372                                 dev->device->name);
1373                         return ret;
1374                 }
1375         }
1376
1377         return 0;
1378 }
1379
1380 static int
1381 tap_gso_ctx_setup(struct rte_gso_ctx *gso_ctx, struct rte_eth_dev *dev)
1382 {
1383         uint32_t gso_types;
1384         char pool_name[64];
1385         struct pmd_internals *pmd = dev->data->dev_private;
1386         int ret;
1387
1388         /* initialize GSO context */
1389         gso_types = DEV_TX_OFFLOAD_TCP_TSO;
1390         if (!pmd->gso_ctx_mp) {
1391                 /*
1392                  * Create private mbuf pool with TAP_GSO_MBUF_SEG_SIZE
1393                  * bytes size per mbuf use this pool for both direct and
1394                  * indirect mbufs
1395                  */
1396                 ret = snprintf(pool_name, sizeof(pool_name), "mp_%s",
1397                                 dev->device->name);
1398                 if (ret < 0 || ret >= (int)sizeof(pool_name)) {
1399                         TAP_LOG(ERR,
1400                                 "%s: failed to create mbuf pool name for device %s,"
1401                                 "device name too long or output error, ret: %d\n",
1402                                 pmd->name, dev->device->name, ret);
1403                         return -ENAMETOOLONG;
1404                 }
1405                 pmd->gso_ctx_mp = rte_pktmbuf_pool_create(pool_name,
1406                         TAP_GSO_MBUFS_NUM, TAP_GSO_MBUF_CACHE_SIZE, 0,
1407                         RTE_PKTMBUF_HEADROOM + TAP_GSO_MBUF_SEG_SIZE,
1408                         SOCKET_ID_ANY);
1409                 if (!pmd->gso_ctx_mp) {
1410                         TAP_LOG(ERR,
1411                                 "%s: failed to create mbuf pool for device %s\n",
1412                                 pmd->name, dev->device->name);
1413                         return -1;
1414                 }
1415         }
1416
1417         gso_ctx->direct_pool = pmd->gso_ctx_mp;
1418         gso_ctx->indirect_pool = pmd->gso_ctx_mp;
1419         gso_ctx->gso_types = gso_types;
1420         gso_ctx->gso_size = 0; /* gso_size is set in tx_burst() per packet */
1421         gso_ctx->flag = 0;
1422
1423         return 0;
1424 }
1425
1426 static int
1427 tap_setup_queue(struct rte_eth_dev *dev,
1428                 struct pmd_internals *internals,
1429                 uint16_t qid,
1430                 int is_rx)
1431 {
1432         int ret;
1433         int *fd;
1434         int *other_fd;
1435         const char *dir;
1436         struct pmd_internals *pmd = dev->data->dev_private;
1437         struct pmd_process_private *process_private = dev->process_private;
1438         struct rx_queue *rx = &internals->rxq[qid];
1439         struct tx_queue *tx = &internals->txq[qid];
1440         struct rte_gso_ctx *gso_ctx;
1441
1442         if (is_rx) {
1443                 fd = &process_private->rxq_fds[qid];
1444                 other_fd = &process_private->txq_fds[qid];
1445                 dir = "rx";
1446                 gso_ctx = NULL;
1447         } else {
1448                 fd = &process_private->txq_fds[qid];
1449                 other_fd = &process_private->rxq_fds[qid];
1450                 dir = "tx";
1451                 gso_ctx = &tx->gso_ctx;
1452         }
1453         if (*fd != -1) {
1454                 /* fd for this queue already exists */
1455                 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1456                         pmd->name, *fd, dir, qid);
1457                 gso_ctx = NULL;
1458         } else if (*other_fd != -1) {
1459                 /* Only other_fd exists. dup it */
1460                 *fd = dup(*other_fd);
1461                 if (*fd < 0) {
1462                         *fd = -1;
1463                         TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1464                         return -1;
1465                 }
1466                 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1467                         pmd->name, *other_fd, dir, qid, *fd);
1468         } else {
1469                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1470                 *fd = tun_alloc(pmd, 0);
1471                 if (*fd < 0) {
1472                         *fd = -1; /* restore original value */
1473                         TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1474                         return -1;
1475                 }
1476                 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1477                         pmd->name, dir, qid, *fd);
1478         }
1479
1480         tx->mtu = &dev->data->mtu;
1481         rx->rxmode = &dev->data->dev_conf.rxmode;
1482         if (gso_ctx) {
1483                 ret = tap_gso_ctx_setup(gso_ctx, dev);
1484                 if (ret)
1485                         return -1;
1486         }
1487
1488         tx->type = pmd->type;
1489
1490         return *fd;
1491 }
1492
1493 static int
1494 tap_rx_queue_setup(struct rte_eth_dev *dev,
1495                    uint16_t rx_queue_id,
1496                    uint16_t nb_rx_desc,
1497                    unsigned int socket_id,
1498                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1499                    struct rte_mempool *mp)
1500 {
1501         struct pmd_internals *internals = dev->data->dev_private;
1502         struct pmd_process_private *process_private = dev->process_private;
1503         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1504         struct rte_mbuf **tmp = &rxq->pool;
1505         long iov_max = sysconf(_SC_IOV_MAX);
1506
1507         if (iov_max <= 0) {
1508                 TAP_LOG(WARNING,
1509                         "_SC_IOV_MAX is not defined. Using %d as default",
1510                         TAP_IOV_DEFAULT_MAX);
1511                 iov_max = TAP_IOV_DEFAULT_MAX;
1512         }
1513         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1514         struct iovec (*iovecs)[nb_desc + 1];
1515         int data_off = RTE_PKTMBUF_HEADROOM;
1516         int ret = 0;
1517         int fd;
1518         int i;
1519
1520         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1521                 TAP_LOG(WARNING,
1522                         "nb_rx_queues %d too small or mempool NULL",
1523                         dev->data->nb_rx_queues);
1524                 return -1;
1525         }
1526
1527         rxq->mp = mp;
1528         rxq->trigger_seen = 1; /* force initial burst */
1529         rxq->in_port = dev->data->port_id;
1530         rxq->queue_id = rx_queue_id;
1531         rxq->nb_rx_desc = nb_desc;
1532         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1533                                     socket_id);
1534         if (!iovecs) {
1535                 TAP_LOG(WARNING,
1536                         "%s: Couldn't allocate %d RX descriptors",
1537                         dev->device->name, nb_desc);
1538                 return -ENOMEM;
1539         }
1540         rxq->iovecs = iovecs;
1541
1542         dev->data->rx_queues[rx_queue_id] = rxq;
1543         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1544         if (fd == -1) {
1545                 ret = fd;
1546                 goto error;
1547         }
1548
1549         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1550         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1551
1552         for (i = 1; i <= nb_desc; i++) {
1553                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1554                 if (!*tmp) {
1555                         TAP_LOG(WARNING,
1556                                 "%s: couldn't allocate memory for queue %d",
1557                                 dev->device->name, rx_queue_id);
1558                         ret = -ENOMEM;
1559                         goto error;
1560                 }
1561                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1562                 (*rxq->iovecs)[i].iov_base =
1563                         (char *)(*tmp)->buf_addr + data_off;
1564                 data_off = 0;
1565                 tmp = &(*tmp)->next;
1566         }
1567
1568         TAP_LOG(DEBUG, "  RX TUNTAP device name %s, qid %d on fd %d",
1569                 internals->name, rx_queue_id,
1570                 process_private->rxq_fds[rx_queue_id]);
1571
1572         return 0;
1573
1574 error:
1575         tap_rxq_pool_free(rxq->pool);
1576         rxq->pool = NULL;
1577         rte_free(rxq->iovecs);
1578         rxq->iovecs = NULL;
1579         return ret;
1580 }
1581
1582 static int
1583 tap_tx_queue_setup(struct rte_eth_dev *dev,
1584                    uint16_t tx_queue_id,
1585                    uint16_t nb_tx_desc __rte_unused,
1586                    unsigned int socket_id __rte_unused,
1587                    const struct rte_eth_txconf *tx_conf)
1588 {
1589         struct pmd_internals *internals = dev->data->dev_private;
1590         struct pmd_process_private *process_private = dev->process_private;
1591         struct tx_queue *txq;
1592         int ret;
1593         uint64_t offloads;
1594
1595         if (tx_queue_id >= dev->data->nb_tx_queues)
1596                 return -1;
1597         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1598         txq = dev->data->tx_queues[tx_queue_id];
1599         txq->out_port = dev->data->port_id;
1600         txq->queue_id = tx_queue_id;
1601
1602         offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1603         txq->csum = !!(offloads &
1604                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
1605                          DEV_TX_OFFLOAD_UDP_CKSUM |
1606                          DEV_TX_OFFLOAD_TCP_CKSUM));
1607
1608         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1609         if (ret == -1)
1610                 return -1;
1611         TAP_LOG(DEBUG,
1612                 "  TX TUNTAP device name %s, qid %d on fd %d csum %s",
1613                 internals->name, tx_queue_id,
1614                 process_private->txq_fds[tx_queue_id],
1615                 txq->csum ? "on" : "off");
1616
1617         return 0;
1618 }
1619
1620 static int
1621 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1622 {
1623         struct pmd_internals *pmd = dev->data->dev_private;
1624         struct ifreq ifr = { .ifr_mtu = mtu };
1625         int err = 0;
1626
1627         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1628         if (!err)
1629                 dev->data->mtu = mtu;
1630
1631         return err;
1632 }
1633
1634 static int
1635 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1636                      struct rte_ether_addr *mc_addr_set __rte_unused,
1637                      uint32_t nb_mc_addr __rte_unused)
1638 {
1639         /*
1640          * Nothing to do actually: the tap has no filtering whatsoever, every
1641          * packet is received.
1642          */
1643         return 0;
1644 }
1645
1646 static int
1647 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1648 {
1649         struct rte_eth_dev *dev = arg;
1650         struct pmd_internals *pmd = dev->data->dev_private;
1651         struct ifinfomsg *info = NLMSG_DATA(nh);
1652
1653         if (nh->nlmsg_type != RTM_NEWLINK ||
1654             (info->ifi_index != pmd->if_index &&
1655              info->ifi_index != pmd->remote_if_index))
1656                 return 0;
1657         return tap_link_update(dev, 0);
1658 }
1659
1660 static void
1661 tap_dev_intr_handler(void *cb_arg)
1662 {
1663         struct rte_eth_dev *dev = cb_arg;
1664         struct pmd_internals *pmd = dev->data->dev_private;
1665
1666         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1667 }
1668
1669 static int
1670 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1671 {
1672         struct pmd_internals *pmd = dev->data->dev_private;
1673         int ret;
1674
1675         /* In any case, disable interrupt if the conf is no longer there. */
1676         if (!dev->data->dev_conf.intr_conf.lsc) {
1677                 if (pmd->intr_handle.fd != -1) {
1678                         goto clean;
1679                 }
1680                 return 0;
1681         }
1682         if (set) {
1683                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1684                 if (unlikely(pmd->intr_handle.fd == -1))
1685                         return -EBADF;
1686                 return rte_intr_callback_register(
1687                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1688         }
1689
1690 clean:
1691         do {
1692                 ret = rte_intr_callback_unregister(&pmd->intr_handle,
1693                         tap_dev_intr_handler, dev);
1694                 if (ret >= 0) {
1695                         break;
1696                 } else if (ret == -EAGAIN) {
1697                         rte_delay_ms(100);
1698                 } else {
1699                         TAP_LOG(ERR, "intr callback unregister failed: %d",
1700                                      ret);
1701                         break;
1702                 }
1703         } while (true);
1704
1705         tap_nl_final(pmd->intr_handle.fd);
1706         pmd->intr_handle.fd = -1;
1707
1708         return 0;
1709 }
1710
1711 static int
1712 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1713 {
1714         int err;
1715
1716         err = tap_lsc_intr_handle_set(dev, set);
1717         if (err < 0) {
1718                 if (!set)
1719                         tap_rx_intr_vec_set(dev, 0);
1720                 return err;
1721         }
1722         err = tap_rx_intr_vec_set(dev, set);
1723         if (err && set)
1724                 tap_lsc_intr_handle_set(dev, 0);
1725         return err;
1726 }
1727
1728 static const uint32_t*
1729 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1730 {
1731         static const uint32_t ptypes[] = {
1732                 RTE_PTYPE_INNER_L2_ETHER,
1733                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1734                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1735                 RTE_PTYPE_INNER_L3_IPV4,
1736                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1737                 RTE_PTYPE_INNER_L3_IPV6,
1738                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1739                 RTE_PTYPE_INNER_L4_FRAG,
1740                 RTE_PTYPE_INNER_L4_UDP,
1741                 RTE_PTYPE_INNER_L4_TCP,
1742                 RTE_PTYPE_INNER_L4_SCTP,
1743                 RTE_PTYPE_L2_ETHER,
1744                 RTE_PTYPE_L2_ETHER_VLAN,
1745                 RTE_PTYPE_L2_ETHER_QINQ,
1746                 RTE_PTYPE_L3_IPV4,
1747                 RTE_PTYPE_L3_IPV4_EXT,
1748                 RTE_PTYPE_L3_IPV6_EXT,
1749                 RTE_PTYPE_L3_IPV6,
1750                 RTE_PTYPE_L4_FRAG,
1751                 RTE_PTYPE_L4_UDP,
1752                 RTE_PTYPE_L4_TCP,
1753                 RTE_PTYPE_L4_SCTP,
1754         };
1755
1756         return ptypes;
1757 }
1758
1759 static int
1760 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1761                   struct rte_eth_fc_conf *fc_conf)
1762 {
1763         fc_conf->mode = RTE_FC_NONE;
1764         return 0;
1765 }
1766
1767 static int
1768 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1769                   struct rte_eth_fc_conf *fc_conf)
1770 {
1771         if (fc_conf->mode != RTE_FC_NONE)
1772                 return -ENOTSUP;
1773         return 0;
1774 }
1775
1776 /**
1777  * DPDK callback to update the RSS hash configuration.
1778  *
1779  * @param dev
1780  *   Pointer to Ethernet device structure.
1781  * @param[in] rss_conf
1782  *   RSS configuration data.
1783  *
1784  * @return
1785  *   0 on success, a negative errno value otherwise and rte_errno is set.
1786  */
1787 static int
1788 tap_rss_hash_update(struct rte_eth_dev *dev,
1789                 struct rte_eth_rss_conf *rss_conf)
1790 {
1791         if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1792                 rte_errno = EINVAL;
1793                 return -rte_errno;
1794         }
1795         if (rss_conf->rss_key && rss_conf->rss_key_len) {
1796                 /*
1797                  * Currently TAP RSS key is hard coded
1798                  * and cannot be updated
1799                  */
1800                 TAP_LOG(ERR,
1801                         "port %u RSS key cannot be updated",
1802                         dev->data->port_id);
1803                 rte_errno = EINVAL;
1804                 return -rte_errno;
1805         }
1806         return 0;
1807 }
1808
1809 static int
1810 tap_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1811 {
1812         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1813
1814         return 0;
1815 }
1816
1817 static int
1818 tap_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1819 {
1820         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1821
1822         return 0;
1823 }
1824
1825 static int
1826 tap_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1827 {
1828         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1829
1830         return 0;
1831 }
1832
1833 static int
1834 tap_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1835 {
1836         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1837
1838         return 0;
1839 }
1840 static const struct eth_dev_ops ops = {
1841         .dev_start              = tap_dev_start,
1842         .dev_stop               = tap_dev_stop,
1843         .dev_close              = tap_dev_close,
1844         .dev_configure          = tap_dev_configure,
1845         .dev_infos_get          = tap_dev_info,
1846         .rx_queue_setup         = tap_rx_queue_setup,
1847         .tx_queue_setup         = tap_tx_queue_setup,
1848         .rx_queue_start         = tap_rx_queue_start,
1849         .tx_queue_start         = tap_tx_queue_start,
1850         .rx_queue_stop          = tap_rx_queue_stop,
1851         .tx_queue_stop          = tap_tx_queue_stop,
1852         .rx_queue_release       = tap_rx_queue_release,
1853         .tx_queue_release       = tap_tx_queue_release,
1854         .flow_ctrl_get          = tap_flow_ctrl_get,
1855         .flow_ctrl_set          = tap_flow_ctrl_set,
1856         .link_update            = tap_link_update,
1857         .dev_set_link_up        = tap_link_set_up,
1858         .dev_set_link_down      = tap_link_set_down,
1859         .promiscuous_enable     = tap_promisc_enable,
1860         .promiscuous_disable    = tap_promisc_disable,
1861         .allmulticast_enable    = tap_allmulti_enable,
1862         .allmulticast_disable   = tap_allmulti_disable,
1863         .mac_addr_set           = tap_mac_set,
1864         .mtu_set                = tap_mtu_set,
1865         .set_mc_addr_list       = tap_set_mc_addr_list,
1866         .stats_get              = tap_stats_get,
1867         .stats_reset            = tap_stats_reset,
1868         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1869         .rss_hash_update        = tap_rss_hash_update,
1870         .filter_ctrl            = tap_dev_filter_ctrl,
1871 };
1872
1873 static int
1874 eth_dev_tap_create(struct rte_vdev_device *vdev, const char *tap_name,
1875                    char *remote_iface, struct rte_ether_addr *mac_addr,
1876                    enum rte_tuntap_type type)
1877 {
1878         int numa_node = rte_socket_id();
1879         struct rte_eth_dev *dev;
1880         struct pmd_internals *pmd;
1881         struct pmd_process_private *process_private;
1882         const char *tuntap_name = tuntap_types[type];
1883         struct rte_eth_dev_data *data;
1884         struct ifreq ifr;
1885         int i;
1886
1887         TAP_LOG(DEBUG, "%s device on numa %u", tuntap_name, rte_socket_id());
1888
1889         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1890         if (!dev) {
1891                 TAP_LOG(ERR, "%s Unable to allocate device struct",
1892                                 tuntap_name);
1893                 goto error_exit_nodev;
1894         }
1895
1896         process_private = (struct pmd_process_private *)
1897                 rte_zmalloc_socket(tap_name, sizeof(struct pmd_process_private),
1898                         RTE_CACHE_LINE_SIZE, dev->device->numa_node);
1899
1900         if (process_private == NULL) {
1901                 TAP_LOG(ERR, "Failed to alloc memory for process private");
1902                 return -1;
1903         }
1904         pmd = dev->data->dev_private;
1905         dev->process_private = process_private;
1906         pmd->dev = dev;
1907         strlcpy(pmd->name, tap_name, sizeof(pmd->name));
1908         pmd->type = type;
1909         pmd->ka_fd = -1;
1910         pmd->nlsk_fd = -1;
1911         pmd->gso_ctx_mp = NULL;
1912
1913         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1914         if (pmd->ioctl_sock == -1) {
1915                 TAP_LOG(ERR,
1916                         "%s Unable to get a socket for management: %s",
1917                         tuntap_name, strerror(errno));
1918                 goto error_exit;
1919         }
1920
1921         /* Setup some default values */
1922         data = dev->data;
1923         data->dev_private = pmd;
1924         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1925         data->numa_node = numa_node;
1926
1927         data->dev_link = pmd_link;
1928         data->mac_addrs = &pmd->eth_addr;
1929         /* Set the number of RX and TX queues */
1930         data->nb_rx_queues = 0;
1931         data->nb_tx_queues = 0;
1932
1933         dev->dev_ops = &ops;
1934         dev->rx_pkt_burst = pmd_rx_burst;
1935         dev->tx_pkt_burst = pmd_tx_burst;
1936
1937         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1938         pmd->intr_handle.fd = -1;
1939         dev->intr_handle = &pmd->intr_handle;
1940
1941         /* Presetup the fds to -1 as being not valid */
1942         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1943                 process_private->rxq_fds[i] = -1;
1944                 process_private->txq_fds[i] = -1;
1945         }
1946
1947         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1948                 if (rte_is_zero_ether_addr(mac_addr))
1949                         rte_eth_random_addr((uint8_t *)&pmd->eth_addr);
1950                 else
1951                         rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1952         }
1953
1954         /*
1955          * Allocate a TUN device keep-alive file descriptor that will only be
1956          * closed when the TUN device itself is closed or removed.
1957          * This keep-alive file descriptor will guarantee that the TUN device
1958          * exists even when all of its queues are closed
1959          */
1960         pmd->ka_fd = tun_alloc(pmd, 1);
1961         if (pmd->ka_fd == -1) {
1962                 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
1963                 goto error_exit;
1964         }
1965         TAP_LOG(DEBUG, "allocated %s", pmd->name);
1966
1967         ifr.ifr_mtu = dev->data->mtu;
1968         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1969                 goto error_exit;
1970
1971         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1972                 memset(&ifr, 0, sizeof(struct ifreq));
1973                 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1974                 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1975                                 RTE_ETHER_ADDR_LEN);
1976                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1977                         goto error_exit;
1978         }
1979
1980         /*
1981          * Set up everything related to rte_flow:
1982          * - netlink socket
1983          * - tap / remote if_index
1984          * - mandatory QDISCs
1985          * - rte_flow actual/implicit lists
1986          * - implicit rules
1987          */
1988         pmd->nlsk_fd = tap_nl_init(0);
1989         if (pmd->nlsk_fd == -1) {
1990                 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
1991                         pmd->name);
1992                 goto disable_rte_flow;
1993         }
1994         pmd->if_index = if_nametoindex(pmd->name);
1995         if (!pmd->if_index) {
1996                 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
1997                 goto disable_rte_flow;
1998         }
1999         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
2000                 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
2001                         pmd->name);
2002                 goto disable_rte_flow;
2003         }
2004         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
2005                 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
2006                         pmd->name);
2007                 goto disable_rte_flow;
2008         }
2009         LIST_INIT(&pmd->flows);
2010
2011         if (strlen(remote_iface)) {
2012                 pmd->remote_if_index = if_nametoindex(remote_iface);
2013                 if (!pmd->remote_if_index) {
2014                         TAP_LOG(ERR, "%s: failed to get %s if_index.",
2015                                 pmd->name, remote_iface);
2016                         goto error_remote;
2017                 }
2018                 strlcpy(pmd->remote_iface, remote_iface, RTE_ETH_NAME_MAX_LEN);
2019
2020                 /* Save state of remote device */
2021                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
2022
2023                 /* Replicate remote MAC address */
2024                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
2025                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
2026                                 pmd->name, pmd->remote_iface);
2027                         goto error_remote;
2028                 }
2029                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
2030                            RTE_ETHER_ADDR_LEN);
2031                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
2032                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
2033                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
2034                                 pmd->name, remote_iface);
2035                         goto error_remote;
2036                 }
2037
2038                 /*
2039                  * Flush usually returns negative value because it tries to
2040                  * delete every QDISC (and on a running device, one QDISC at
2041                  * least is needed). Ignore negative return value.
2042                  */
2043                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
2044                 if (qdisc_create_ingress(pmd->nlsk_fd,
2045                                          pmd->remote_if_index) < 0) {
2046                         TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
2047                                 pmd->remote_iface);
2048                         goto error_remote;
2049                 }
2050                 LIST_INIT(&pmd->implicit_flows);
2051                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
2052                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
2053                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
2054                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
2055                         TAP_LOG(ERR,
2056                                 "%s: failed to create implicit rules.",
2057                                 pmd->name);
2058                         goto error_remote;
2059                 }
2060         }
2061
2062         rte_eth_dev_probing_finish(dev);
2063         return 0;
2064
2065 disable_rte_flow:
2066         TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
2067                 strerror(errno), errno);
2068         if (strlen(remote_iface)) {
2069                 TAP_LOG(ERR, "Remote feature requires flow support.");
2070                 goto error_exit;
2071         }
2072         rte_eth_dev_probing_finish(dev);
2073         return 0;
2074
2075 error_remote:
2076         TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
2077                 strerror(errno), errno);
2078         tap_flow_implicit_flush(pmd, NULL);
2079
2080 error_exit:
2081         if (pmd->nlsk_fd != -1)
2082                 close(pmd->nlsk_fd);
2083         if (pmd->ka_fd != -1)
2084                 close(pmd->ka_fd);
2085         if (pmd->ioctl_sock != -1)
2086                 close(pmd->ioctl_sock);
2087         /* mac_addrs must not be freed alone because part of dev_private */
2088         dev->data->mac_addrs = NULL;
2089         rte_eth_dev_release_port(dev);
2090
2091 error_exit_nodev:
2092         TAP_LOG(ERR, "%s Unable to initialize %s",
2093                 tuntap_name, rte_vdev_device_name(vdev));
2094
2095         return -EINVAL;
2096 }
2097
2098 /* make sure name is a possible Linux network device name */
2099 static bool
2100 is_valid_iface(const char *name)
2101 {
2102         if (*name == '\0')
2103                 return false;
2104
2105         if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
2106                 return false;
2107
2108         while (*name) {
2109                 if (*name == '/' || *name == ':' || isspace(*name))
2110                         return false;
2111                 name++;
2112         }
2113         return true;
2114 }
2115
2116 static int
2117 set_interface_name(const char *key __rte_unused,
2118                    const char *value,
2119                    void *extra_args)
2120 {
2121         char *name = (char *)extra_args;
2122
2123         if (value) {
2124                 if (!is_valid_iface(value)) {
2125                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2126                                 value);
2127                         return -1;
2128                 }
2129                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2130         } else {
2131                 /* use tap%d which causes kernel to choose next available */
2132                 strlcpy(name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2133         }
2134         return 0;
2135 }
2136
2137 static int
2138 set_remote_iface(const char *key __rte_unused,
2139                  const char *value,
2140                  void *extra_args)
2141 {
2142         char *name = (char *)extra_args;
2143
2144         if (value) {
2145                 if (!is_valid_iface(value)) {
2146                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2147                                 value);
2148                         return -1;
2149                 }
2150                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2151         }
2152
2153         return 0;
2154 }
2155
2156 static int parse_user_mac(struct rte_ether_addr *user_mac,
2157                 const char *value)
2158 {
2159         unsigned int index = 0;
2160         char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
2161
2162         if (user_mac == NULL || value == NULL)
2163                 return 0;
2164
2165         strlcpy(mac_temp, value, sizeof(mac_temp));
2166         mac_byte = strtok(mac_temp, ":");
2167
2168         while ((mac_byte != NULL) &&
2169                         (strlen(mac_byte) <= 2) &&
2170                         (strlen(mac_byte) == strspn(mac_byte,
2171                                         ETH_TAP_CMP_MAC_FMT))) {
2172                 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
2173                 mac_byte = strtok(NULL, ":");
2174         }
2175
2176         return index;
2177 }
2178
2179 static int
2180 set_mac_type(const char *key __rte_unused,
2181              const char *value,
2182              void *extra_args)
2183 {
2184         struct rte_ether_addr *user_mac = extra_args;
2185
2186         if (!value)
2187                 return 0;
2188
2189         if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
2190                 static int iface_idx;
2191
2192                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
2193                 memcpy((char *)user_mac->addr_bytes, "\0dtap",
2194                         RTE_ETHER_ADDR_LEN);
2195                 user_mac->addr_bytes[RTE_ETHER_ADDR_LEN - 1] =
2196                         iface_idx++ + '0';
2197                 goto success;
2198         }
2199
2200         if (parse_user_mac(user_mac, value) != 6)
2201                 goto error;
2202 success:
2203         TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
2204         return 0;
2205
2206 error:
2207         TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
2208                 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
2209         return -1;
2210 }
2211
2212 /*
2213  * Open a TUN interface device. TUN PMD
2214  * 1) sets tap_type as false
2215  * 2) intakes iface as argument.
2216  * 3) as interface is virtual set speed to 10G
2217  */
2218 static int
2219 rte_pmd_tun_probe(struct rte_vdev_device *dev)
2220 {
2221         const char *name, *params;
2222         int ret;
2223         struct rte_kvargs *kvlist = NULL;
2224         char tun_name[RTE_ETH_NAME_MAX_LEN];
2225         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2226         struct rte_eth_dev *eth_dev;
2227
2228         name = rte_vdev_device_name(dev);
2229         params = rte_vdev_device_args(dev);
2230         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2231
2232         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
2233             strlen(params) == 0) {
2234                 eth_dev = rte_eth_dev_attach_secondary(name);
2235                 if (!eth_dev) {
2236                         TAP_LOG(ERR, "Failed to probe %s", name);
2237                         return -1;
2238                 }
2239                 eth_dev->dev_ops = &ops;
2240                 eth_dev->device = &dev->device;
2241                 rte_eth_dev_probing_finish(eth_dev);
2242                 return 0;
2243         }
2244
2245         /* use tun%d which causes kernel to choose next available */
2246         strlcpy(tun_name, DEFAULT_TUN_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2247
2248         if (params && (params[0] != '\0')) {
2249                 TAP_LOG(DEBUG, "parameters (%s)", params);
2250
2251                 kvlist = rte_kvargs_parse(params, valid_arguments);
2252                 if (kvlist) {
2253                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2254                                 ret = rte_kvargs_process(kvlist,
2255                                         ETH_TAP_IFACE_ARG,
2256                                         &set_interface_name,
2257                                         tun_name);
2258
2259                                 if (ret == -1)
2260                                         goto leave;
2261                         }
2262                 }
2263         }
2264         pmd_link.link_speed = ETH_SPEED_NUM_10G;
2265
2266         TAP_LOG(DEBUG, "Initializing pmd_tun for %s", name);
2267
2268         ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0,
2269                                  ETH_TUNTAP_TYPE_TUN);
2270
2271 leave:
2272         if (ret == -1) {
2273                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2274                         name, tun_name);
2275         }
2276         rte_kvargs_free(kvlist);
2277
2278         return ret;
2279 }
2280
2281 /* Request queue file descriptors from secondary to primary. */
2282 static int
2283 tap_mp_attach_queues(const char *port_name, struct rte_eth_dev *dev)
2284 {
2285         int ret;
2286         struct timespec timeout = {.tv_sec = 1, .tv_nsec = 0};
2287         struct rte_mp_msg request, *reply;
2288         struct rte_mp_reply replies;
2289         struct ipc_queues *request_param = (struct ipc_queues *)request.param;
2290         struct ipc_queues *reply_param;
2291         struct pmd_process_private *process_private = dev->process_private;
2292         int queue, fd_iterator;
2293
2294         /* Prepare the request */
2295         memset(&request, 0, sizeof(request));
2296         strlcpy(request.name, TAP_MP_KEY, sizeof(request.name));
2297         strlcpy(request_param->port_name, port_name,
2298                 sizeof(request_param->port_name));
2299         request.len_param = sizeof(*request_param);
2300         /* Send request and receive reply */
2301         ret = rte_mp_request_sync(&request, &replies, &timeout);
2302         if (ret < 0 || replies.nb_received != 1) {
2303                 TAP_LOG(ERR, "Failed to request queues from primary: %d",
2304                         rte_errno);
2305                 return -1;
2306         }
2307         reply = &replies.msgs[0];
2308         reply_param = (struct ipc_queues *)reply->param;
2309         TAP_LOG(DEBUG, "Received IPC reply for %s", reply_param->port_name);
2310
2311         /* Attach the queues from received file descriptors */
2312         if (reply_param->rxq_count + reply_param->txq_count != reply->num_fds) {
2313                 TAP_LOG(ERR, "Unexpected number of fds received");
2314                 return -1;
2315         }
2316
2317         dev->data->nb_rx_queues = reply_param->rxq_count;
2318         dev->data->nb_tx_queues = reply_param->txq_count;
2319         fd_iterator = 0;
2320         for (queue = 0; queue < reply_param->rxq_count; queue++)
2321                 process_private->rxq_fds[queue] = reply->fds[fd_iterator++];
2322         for (queue = 0; queue < reply_param->txq_count; queue++)
2323                 process_private->txq_fds[queue] = reply->fds[fd_iterator++];
2324         free(reply);
2325         return 0;
2326 }
2327
2328 /* Send the queue file descriptors from the primary process to secondary. */
2329 static int
2330 tap_mp_sync_queues(const struct rte_mp_msg *request, const void *peer)
2331 {
2332         struct rte_eth_dev *dev;
2333         struct pmd_process_private *process_private;
2334         struct rte_mp_msg reply;
2335         const struct ipc_queues *request_param =
2336                 (const struct ipc_queues *)request->param;
2337         struct ipc_queues *reply_param =
2338                 (struct ipc_queues *)reply.param;
2339         uint16_t port_id;
2340         int queue;
2341         int ret;
2342
2343         /* Get requested port */
2344         TAP_LOG(DEBUG, "Received IPC request for %s", request_param->port_name);
2345         ret = rte_eth_dev_get_port_by_name(request_param->port_name, &port_id);
2346         if (ret) {
2347                 TAP_LOG(ERR, "Failed to get port id for %s",
2348                         request_param->port_name);
2349                 return -1;
2350         }
2351         dev = &rte_eth_devices[port_id];
2352         process_private = dev->process_private;
2353
2354         /* Fill file descriptors for all queues */
2355         reply.num_fds = 0;
2356         reply_param->rxq_count = 0;
2357         if (dev->data->nb_rx_queues + dev->data->nb_tx_queues >
2358                         RTE_MP_MAX_FD_NUM){
2359                 TAP_LOG(ERR, "Number of rx/tx queues exceeds max number of fds");
2360                 return -1;
2361         }
2362
2363         for (queue = 0; queue < dev->data->nb_rx_queues; queue++) {
2364                 reply.fds[reply.num_fds++] = process_private->rxq_fds[queue];
2365                 reply_param->rxq_count++;
2366         }
2367         RTE_ASSERT(reply_param->rxq_count == dev->data->nb_rx_queues);
2368
2369         reply_param->txq_count = 0;
2370         for (queue = 0; queue < dev->data->nb_tx_queues; queue++) {
2371                 reply.fds[reply.num_fds++] = process_private->txq_fds[queue];
2372                 reply_param->txq_count++;
2373         }
2374         RTE_ASSERT(reply_param->txq_count == dev->data->nb_tx_queues);
2375
2376         /* Send reply */
2377         strlcpy(reply.name, request->name, sizeof(reply.name));
2378         strlcpy(reply_param->port_name, request_param->port_name,
2379                 sizeof(reply_param->port_name));
2380         reply.len_param = sizeof(*reply_param);
2381         if (rte_mp_reply(&reply, peer) < 0) {
2382                 TAP_LOG(ERR, "Failed to reply an IPC request to sync queues");
2383                 return -1;
2384         }
2385         return 0;
2386 }
2387
2388 /* Open a TAP interface device.
2389  */
2390 static int
2391 rte_pmd_tap_probe(struct rte_vdev_device *dev)
2392 {
2393         const char *name, *params;
2394         int ret;
2395         struct rte_kvargs *kvlist = NULL;
2396         int speed;
2397         char tap_name[RTE_ETH_NAME_MAX_LEN];
2398         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2399         struct rte_ether_addr user_mac = { .addr_bytes = {0} };
2400         struct rte_eth_dev *eth_dev;
2401         int tap_devices_count_increased = 0;
2402
2403         name = rte_vdev_device_name(dev);
2404         params = rte_vdev_device_args(dev);
2405
2406         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
2407                 eth_dev = rte_eth_dev_attach_secondary(name);
2408                 if (!eth_dev) {
2409                         TAP_LOG(ERR, "Failed to probe %s", name);
2410                         return -1;
2411                 }
2412                 eth_dev->dev_ops = &ops;
2413                 eth_dev->device = &dev->device;
2414                 eth_dev->rx_pkt_burst = pmd_rx_burst;
2415                 eth_dev->tx_pkt_burst = pmd_tx_burst;
2416                 if (!rte_eal_primary_proc_alive(NULL)) {
2417                         TAP_LOG(ERR, "Primary process is missing");
2418                         return -1;
2419                 }
2420                 eth_dev->process_private = (struct pmd_process_private *)
2421                         rte_zmalloc_socket(name,
2422                                 sizeof(struct pmd_process_private),
2423                                 RTE_CACHE_LINE_SIZE,
2424                                 eth_dev->device->numa_node);
2425                 if (eth_dev->process_private == NULL) {
2426                         TAP_LOG(ERR,
2427                                 "Failed to alloc memory for process private");
2428                         return -1;
2429                 }
2430
2431                 ret = tap_mp_attach_queues(name, eth_dev);
2432                 if (ret != 0)
2433                         return -1;
2434                 rte_eth_dev_probing_finish(eth_dev);
2435                 return 0;
2436         }
2437
2438         speed = ETH_SPEED_NUM_10G;
2439
2440         /* use tap%d which causes kernel to choose next available */
2441         strlcpy(tap_name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2442         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2443
2444         if (params && (params[0] != '\0')) {
2445                 TAP_LOG(DEBUG, "parameters (%s)", params);
2446
2447                 kvlist = rte_kvargs_parse(params, valid_arguments);
2448                 if (kvlist) {
2449                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2450                                 ret = rte_kvargs_process(kvlist,
2451                                                          ETH_TAP_IFACE_ARG,
2452                                                          &set_interface_name,
2453                                                          tap_name);
2454                                 if (ret == -1)
2455                                         goto leave;
2456                         }
2457
2458                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
2459                                 ret = rte_kvargs_process(kvlist,
2460                                                          ETH_TAP_REMOTE_ARG,
2461                                                          &set_remote_iface,
2462                                                          remote_iface);
2463                                 if (ret == -1)
2464                                         goto leave;
2465                         }
2466
2467                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
2468                                 ret = rte_kvargs_process(kvlist,
2469                                                          ETH_TAP_MAC_ARG,
2470                                                          &set_mac_type,
2471                                                          &user_mac);
2472                                 if (ret == -1)
2473                                         goto leave;
2474                         }
2475                 }
2476         }
2477         pmd_link.link_speed = speed;
2478
2479         TAP_LOG(DEBUG, "Initializing pmd_tap for %s", name);
2480
2481         /* Register IPC feed callback */
2482         if (!tap_devices_count) {
2483                 ret = rte_mp_action_register(TAP_MP_KEY, tap_mp_sync_queues);
2484                 if (ret < 0 && rte_errno != ENOTSUP) {
2485                         TAP_LOG(ERR, "tap: Failed to register IPC callback: %s",
2486                                 strerror(rte_errno));
2487                         goto leave;
2488                 }
2489         }
2490         tap_devices_count++;
2491         tap_devices_count_increased = 1;
2492         ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
2493                 ETH_TUNTAP_TYPE_TAP);
2494
2495 leave:
2496         if (ret == -1) {
2497                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2498                         name, tap_name);
2499                 if (tap_devices_count_increased == 1) {
2500                         if (tap_devices_count == 1)
2501                                 rte_mp_action_unregister(TAP_MP_KEY);
2502                         tap_devices_count--;
2503                 }
2504         }
2505         rte_kvargs_free(kvlist);
2506
2507         return ret;
2508 }
2509
2510 /* detach a TUNTAP device.
2511  */
2512 static int
2513 rte_pmd_tap_remove(struct rte_vdev_device *dev)
2514 {
2515         struct rte_eth_dev *eth_dev = NULL;
2516
2517         /* find the ethdev entry */
2518         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
2519         if (!eth_dev)
2520                 return 0;
2521
2522         tap_dev_close(eth_dev);
2523         rte_eth_dev_release_port(eth_dev);
2524
2525         return 0;
2526 }
2527
2528 static struct rte_vdev_driver pmd_tun_drv = {
2529         .probe = rte_pmd_tun_probe,
2530         .remove = rte_pmd_tap_remove,
2531 };
2532
2533 static struct rte_vdev_driver pmd_tap_drv = {
2534         .probe = rte_pmd_tap_probe,
2535         .remove = rte_pmd_tap_remove,
2536 };
2537
2538 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
2539 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
2540 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
2541 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
2542                               ETH_TAP_IFACE_ARG "=<string> ");
2543 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
2544                               ETH_TAP_IFACE_ARG "=<string> "
2545                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
2546                               ETH_TAP_REMOTE_ARG "=<string>");
2547 RTE_LOG_REGISTER(tap_logtype, pmd.net.tap, NOTICE);