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 int
 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         return 0;
 917 }
 918
 919 static int
 920 tap_dev_configure(struct rte_eth_dev *dev)
 921 {
 922         struct pmd_internals *pmd = dev->data->dev_private;
 923
 924         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 925                 TAP_LOG(ERR,
 926                         "%s: number of rx queues %d exceeds max num of queues %d",
 927                         dev->device->name,
 928                         dev->data->nb_rx_queues,
 929                         RTE_PMD_TAP_MAX_QUEUES);
 930                 return -1;
 931         }
 932         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 933                 TAP_LOG(ERR,
 934                         "%s: number of tx queues %d exceeds max num of queues %d",
 935                         dev->device->name,
 936                         dev->data->nb_tx_queues,
 937                         RTE_PMD_TAP_MAX_QUEUES);
 938                 return -1;
 939         }
 940
 941         TAP_LOG(INFO, "%s: %s: TX configured queues number: %u",
 942                 dev->device->name, pmd->name, dev->data->nb_tx_queues);
 943
 944         TAP_LOG(INFO, "%s: %s: RX configured queues number: %u",
 945                 dev->device->name, pmd->name, dev->data->nb_rx_queues);
 946
 947         return 0;
 948 }
 949
 950 static uint32_t
 951 tap_dev_speed_capa(void)
 952 {
 953         uint32_t speed = pmd_link.link_speed;
 954         uint32_t capa = 0;
 955
 956         if (speed >= ETH_SPEED_NUM_10M)
 957                 capa |= ETH_LINK_SPEED_10M;
 958         if (speed >= ETH_SPEED_NUM_100M)
 959                 capa |= ETH_LINK_SPEED_100M;
 960         if (speed >= ETH_SPEED_NUM_1G)
 961                 capa |= ETH_LINK_SPEED_1G;
 962         if (speed >= ETH_SPEED_NUM_5G)
 963                 capa |= ETH_LINK_SPEED_2_5G;
 964         if (speed >= ETH_SPEED_NUM_5G)
 965                 capa |= ETH_LINK_SPEED_5G;
 966         if (speed >= ETH_SPEED_NUM_10G)
 967                 capa |= ETH_LINK_SPEED_10G;
 968         if (speed >= ETH_SPEED_NUM_20G)
 969                 capa |= ETH_LINK_SPEED_20G;
 970         if (speed >= ETH_SPEED_NUM_25G)
 971                 capa |= ETH_LINK_SPEED_25G;
 972         if (speed >= ETH_SPEED_NUM_40G)
 973                 capa |= ETH_LINK_SPEED_40G;
 974         if (speed >= ETH_SPEED_NUM_50G)
 975                 capa |= ETH_LINK_SPEED_50G;
 976         if (speed >= ETH_SPEED_NUM_56G)
 977                 capa |= ETH_LINK_SPEED_56G;
 978         if (speed >= ETH_SPEED_NUM_100G)
 979                 capa |= ETH_LINK_SPEED_100G;
 980
 981         return capa;
 982 }
 983
 984 static int
 985 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 986 {
 987         struct pmd_internals *internals = dev->data->dev_private;
 988
 989         dev_info->if_index = internals->if_index;
 990         dev_info->max_mac_addrs = 1;
 991         dev_info->max_rx_pktlen = (uint32_t)RTE_ETHER_MAX_VLAN_FRAME_LEN;
 992         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
 993         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
 994         dev_info->min_rx_bufsize = 0;
 995         dev_info->speed_capa = tap_dev_speed_capa();
 996         dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
 997         dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
 998                                     dev_info->rx_queue_offload_capa;
 999         dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
1000         dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
1001                                     dev_info->tx_queue_offload_capa;
1002         dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
1003         /*
1004          * limitation: TAP supports all of IP, UDP and TCP hash
1005          * functions together and not in partial combinations
1006          */
1007         dev_info->flow_type_rss_offloads = ~TAP_RSS_HF_MASK;
1008
1009         return 0;
1010 }
1011
1012 static int
1013 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
1014 {
1015         unsigned int i, imax;
1016         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
1017         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
1018         unsigned long rx_nombuf = 0, ierrors = 0;
1019         const struct pmd_internals *pmd = dev->data->dev_private;
1020
1021         /* rx queue statistics */
1022         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
1023                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
1024         for (i = 0; i < imax; i++) {
1025                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
1026                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
1027                 rx_total += tap_stats->q_ipackets[i];
1028                 rx_bytes_total += tap_stats->q_ibytes[i];
1029                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
1030                 ierrors += pmd->rxq[i].stats.ierrors;
1031         }
1032
1033         /* tx queue statistics */
1034         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
1035                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
1036
1037         for (i = 0; i < imax; i++) {
1038                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
1039                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
1040                 tx_total += tap_stats->q_opackets[i];
1041                 tx_err_total += pmd->txq[i].stats.errs;
1042                 tx_bytes_total += tap_stats->q_obytes[i];
1043         }
1044
1045         tap_stats->ipackets = rx_total;
1046         tap_stats->ibytes = rx_bytes_total;
1047         tap_stats->ierrors = ierrors;
1048         tap_stats->rx_nombuf = rx_nombuf;
1049         tap_stats->opackets = tx_total;
1050         tap_stats->oerrors = tx_err_total;
1051         tap_stats->obytes = tx_bytes_total;
1052         return 0;
1053 }
1054
1055 static int
1056 tap_stats_reset(struct rte_eth_dev *dev)
1057 {
1058         int i;
1059         struct pmd_internals *pmd = dev->data->dev_private;
1060
1061         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1062                 pmd->rxq[i].stats.ipackets = 0;
1063                 pmd->rxq[i].stats.ibytes = 0;
1064                 pmd->rxq[i].stats.ierrors = 0;
1065                 pmd->rxq[i].stats.rx_nombuf = 0;
1066
1067                 pmd->txq[i].stats.opackets = 0;
1068                 pmd->txq[i].stats.errs = 0;
1069                 pmd->txq[i].stats.obytes = 0;
1070         }
1071
1072         return 0;
1073 }
1074
1075 static int
1076 tap_dev_close(struct rte_eth_dev *dev)
1077 {
1078         int i;
1079         struct pmd_internals *internals = dev->data->dev_private;
1080         struct pmd_process_private *process_private = dev->process_private;
1081         struct rx_queue *rxq;
1082
1083         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1084                 rte_free(dev->process_private);
1085                 return 0;
1086         }
1087
1088         tap_link_set_down(dev);
1089         if (internals->nlsk_fd != -1) {
1090                 tap_flow_flush(dev, NULL);
1091                 tap_flow_implicit_flush(internals, NULL);
1092                 tap_nl_final(internals->nlsk_fd);
1093                 internals->nlsk_fd = -1;
1094         }
1095
1096         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1097                 if (process_private->rxq_fds[i] != -1) {
1098                         rxq = &internals->rxq[i];
1099                         close(process_private->rxq_fds[i]);
1100                         process_private->rxq_fds[i] = -1;
1101                         tap_rxq_pool_free(rxq->pool);
1102                         rte_free(rxq->iovecs);
1103                         rxq->pool = NULL;
1104                         rxq->iovecs = NULL;
1105                 }
1106                 if (process_private->txq_fds[i] != -1) {
1107                         close(process_private->txq_fds[i]);
1108                         process_private->txq_fds[i] = -1;
1109                 }
1110         }
1111
1112         if (internals->remote_if_index) {
1113                 /* Restore initial remote state */
1114                 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
1115                                 &internals->remote_initial_flags);
1116         }
1117
1118         rte_mempool_free(internals->gso_ctx_mp);
1119         internals->gso_ctx_mp = NULL;
1120
1121         if (internals->ka_fd != -1) {
1122                 close(internals->ka_fd);
1123                 internals->ka_fd = -1;
1124         }
1125
1126         /* mac_addrs must not be freed alone because part of dev_private */
1127         dev->data->mac_addrs = NULL;
1128
1129         internals = dev->data->dev_private;
1130         TAP_LOG(DEBUG, "Closing %s Ethernet device on numa %u",
1131                 tuntap_types[internals->type], rte_socket_id());
1132
1133         if (internals->ioctl_sock != -1) {
1134                 close(internals->ioctl_sock);
1135                 internals->ioctl_sock = -1;
1136         }
1137         rte_free(dev->process_private);
1138         if (tap_devices_count == 1)
1139                 rte_mp_action_unregister(TAP_MP_KEY);
1140         tap_devices_count--;
1141         /*
1142          * Since TUN device has no more opened file descriptors
1143          * it will be removed from kernel
1144          */
1145
1146         return 0;
1147 }
1148
1149 static void
1150 tap_rx_queue_release(void *queue)
1151 {
1152         struct rx_queue *rxq = queue;
1153         struct pmd_process_private *process_private;
1154
1155         if (!rxq)
1156                 return;
1157         process_private = rte_eth_devices[rxq->in_port].process_private;
1158         if (process_private->rxq_fds[rxq->queue_id] != -1) {
1159                 close(process_private->rxq_fds[rxq->queue_id]);
1160                 process_private->rxq_fds[rxq->queue_id] = -1;
1161                 tap_rxq_pool_free(rxq->pool);
1162                 rte_free(rxq->iovecs);
1163                 rxq->pool = NULL;
1164                 rxq->iovecs = NULL;
1165         }
1166 }
1167
1168 static void
1169 tap_tx_queue_release(void *queue)
1170 {
1171         struct tx_queue *txq = queue;
1172         struct pmd_process_private *process_private;
1173
1174         if (!txq)
1175                 return;
1176         process_private = rte_eth_devices[txq->out_port].process_private;
1177
1178         if (process_private->txq_fds[txq->queue_id] != -1) {
1179                 close(process_private->txq_fds[txq->queue_id]);
1180                 process_private->txq_fds[txq->queue_id] = -1;
1181         }
1182 }
1183
1184 static int
1185 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1186 {
1187         struct rte_eth_link *dev_link = &dev->data->dev_link;
1188         struct pmd_internals *pmd = dev->data->dev_private;
1189         struct ifreq ifr = { .ifr_flags = 0 };
1190
1191         if (pmd->remote_if_index) {
1192                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
1193                 if (!(ifr.ifr_flags & IFF_UP) ||
1194                     !(ifr.ifr_flags & IFF_RUNNING)) {
1195                         dev_link->link_status = ETH_LINK_DOWN;
1196                         return 0;
1197                 }
1198         }
1199         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
1200         dev_link->link_status =
1201                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
1202                  ETH_LINK_UP :
1203                  ETH_LINK_DOWN);
1204         return 0;
1205 }
1206
1207 static int
1208 tap_promisc_enable(struct rte_eth_dev *dev)
1209 {
1210         struct pmd_internals *pmd = dev->data->dev_private;
1211         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1212         int ret;
1213
1214         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1215         if (ret != 0)
1216                 return ret;
1217
1218         if (pmd->remote_if_index && !pmd->flow_isolate) {
1219                 dev->data->promiscuous = 1;
1220                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
1221                 if (ret != 0) {
1222                         /* Rollback promisc flag */
1223                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1224                         /*
1225                          * rte_eth_dev_promiscuous_enable() rollback
1226                          * dev->data->promiscuous in the case of failure.
1227                          */
1228                         return ret;
1229                 }
1230         }
1231
1232         return 0;
1233 }
1234
1235 static int
1236 tap_promisc_disable(struct rte_eth_dev *dev)
1237 {
1238         struct pmd_internals *pmd = dev->data->dev_private;
1239         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1240         int ret;
1241
1242         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1243         if (ret != 0)
1244                 return ret;
1245
1246         if (pmd->remote_if_index && !pmd->flow_isolate) {
1247                 dev->data->promiscuous = 0;
1248                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
1249                 if (ret != 0) {
1250                         /* Rollback promisc flag */
1251                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1252                         /*
1253                          * rte_eth_dev_promiscuous_disable() rollback
1254                          * dev->data->promiscuous in the case of failure.
1255                          */
1256                         return ret;
1257                 }
1258         }
1259
1260         return 0;
1261 }
1262
1263 static int
1264 tap_allmulti_enable(struct rte_eth_dev *dev)
1265 {
1266         struct pmd_internals *pmd = dev->data->dev_private;
1267         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1268         int ret;
1269
1270         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1271         if (ret != 0)
1272                 return ret;
1273
1274         if (pmd->remote_if_index && !pmd->flow_isolate) {
1275                 dev->data->all_multicast = 1;
1276                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
1277                 if (ret != 0) {
1278                         /* Rollback allmulti flag */
1279                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1280                         /*
1281                          * rte_eth_dev_allmulticast_enable() rollback
1282                          * dev->data->all_multicast in the case of failure.
1283                          */
1284                         return ret;
1285                 }
1286         }
1287
1288         return 0;
1289 }
1290
1291 static int
1292 tap_allmulti_disable(struct rte_eth_dev *dev)
1293 {
1294         struct pmd_internals *pmd = dev->data->dev_private;
1295         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1296         int ret;
1297
1298         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1299         if (ret != 0)
1300                 return ret;
1301
1302         if (pmd->remote_if_index && !pmd->flow_isolate) {
1303                 dev->data->all_multicast = 0;
1304                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
1305                 if (ret != 0) {
1306                         /* Rollback allmulti flag */
1307                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1308                         /*
1309                          * rte_eth_dev_allmulticast_disable() rollback
1310                          * dev->data->all_multicast in the case of failure.
1311                          */
1312                         return ret;
1313                 }
1314         }
1315
1316         return 0;
1317 }
1318
1319 static int
1320 tap_mac_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1321 {
1322         struct pmd_internals *pmd = dev->data->dev_private;
1323         enum ioctl_mode mode = LOCAL_ONLY;
1324         struct ifreq ifr;
1325         int ret;
1326
1327         if (pmd->type == ETH_TUNTAP_TYPE_TUN) {
1328                 TAP_LOG(ERR, "%s: can't MAC address for TUN",
1329                         dev->device->name);
1330                 return -ENOTSUP;
1331         }
1332
1333         if (rte_is_zero_ether_addr(mac_addr)) {
1334                 TAP_LOG(ERR, "%s: can't set an empty MAC address",
1335                         dev->device->name);
1336                 return -EINVAL;
1337         }
1338         /* Check the actual current MAC address on the tap netdevice */
1339         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
1340         if (ret < 0)
1341                 return ret;
1342         if (rte_is_same_ether_addr(
1343                         (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1344                         mac_addr))
1345                 return 0;
1346         /* Check the current MAC address on the remote */
1347         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
1348         if (ret < 0)
1349                 return ret;
1350         if (!rte_is_same_ether_addr(
1351                         (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1352                         mac_addr))
1353                 mode = LOCAL_AND_REMOTE;
1354         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1355         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, RTE_ETHER_ADDR_LEN);
1356         ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
1357         if (ret < 0)
1358                 return ret;
1359         rte_memcpy(&pmd->eth_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1360         if (pmd->remote_if_index && !pmd->flow_isolate) {
1361                 /* Replace MAC redirection rule after a MAC change */
1362                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
1363                 if (ret < 0) {
1364                         TAP_LOG(ERR,
1365                                 "%s: Couldn't delete MAC redirection rule",
1366                                 dev->device->name);
1367                         return ret;
1368                 }
1369                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1370                 if (ret < 0) {
1371                         TAP_LOG(ERR,
1372                                 "%s: Couldn't add MAC redirection rule",
1373                                 dev->device->name);
1374                         return ret;
1375                 }
1376         }
1377
1378         return 0;
1379 }
1380
1381 static int
1382 tap_gso_ctx_setup(struct rte_gso_ctx *gso_ctx, struct rte_eth_dev *dev)
1383 {
1384         uint32_t gso_types;
1385         char pool_name[64];
1386         struct pmd_internals *pmd = dev->data->dev_private;
1387         int ret;
1388
1389         /* initialize GSO context */
1390         gso_types = DEV_TX_OFFLOAD_TCP_TSO;
1391         if (!pmd->gso_ctx_mp) {
1392                 /*
1393                  * Create private mbuf pool with TAP_GSO_MBUF_SEG_SIZE
1394                  * bytes size per mbuf use this pool for both direct and
1395                  * indirect mbufs
1396                  */
1397                 ret = snprintf(pool_name, sizeof(pool_name), "mp_%s",
1398                                 dev->device->name);
1399                 if (ret < 0 || ret >= (int)sizeof(pool_name)) {
1400                         TAP_LOG(ERR,
1401                                 "%s: failed to create mbuf pool name for device %s,"
1402                                 "device name too long or output error, ret: %d\n",
1403                                 pmd->name, dev->device->name, ret);
1404                         return -ENAMETOOLONG;
1405                 }
1406                 pmd->gso_ctx_mp = rte_pktmbuf_pool_create(pool_name,
1407                         TAP_GSO_MBUFS_NUM, TAP_GSO_MBUF_CACHE_SIZE, 0,
1408                         RTE_PKTMBUF_HEADROOM + TAP_GSO_MBUF_SEG_SIZE,
1409                         SOCKET_ID_ANY);
1410                 if (!pmd->gso_ctx_mp) {
1411                         TAP_LOG(ERR,
1412                                 "%s: failed to create mbuf pool for device %s\n",
1413                                 pmd->name, dev->device->name);
1414                         return -1;
1415                 }
1416         }
1417
1418         gso_ctx->direct_pool = pmd->gso_ctx_mp;
1419         gso_ctx->indirect_pool = pmd->gso_ctx_mp;
1420         gso_ctx->gso_types = gso_types;
1421         gso_ctx->gso_size = 0; /* gso_size is set in tx_burst() per packet */
1422         gso_ctx->flag = 0;
1423
1424         return 0;
1425 }
1426
1427 static int
1428 tap_setup_queue(struct rte_eth_dev *dev,
1429                 struct pmd_internals *internals,
1430                 uint16_t qid,
1431                 int is_rx)
1432 {
1433         int ret;
1434         int *fd;
1435         int *other_fd;
1436         const char *dir;
1437         struct pmd_internals *pmd = dev->data->dev_private;
1438         struct pmd_process_private *process_private = dev->process_private;
1439         struct rx_queue *rx = &internals->rxq[qid];
1440         struct tx_queue *tx = &internals->txq[qid];
1441         struct rte_gso_ctx *gso_ctx;
1442
1443         if (is_rx) {
1444                 fd = &process_private->rxq_fds[qid];
1445                 other_fd = &process_private->txq_fds[qid];
1446                 dir = "rx";
1447                 gso_ctx = NULL;
1448         } else {
1449                 fd = &process_private->txq_fds[qid];
1450                 other_fd = &process_private->rxq_fds[qid];
1451                 dir = "tx";
1452                 gso_ctx = &tx->gso_ctx;
1453         }
1454         if (*fd != -1) {
1455                 /* fd for this queue already exists */
1456                 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1457                         pmd->name, *fd, dir, qid);
1458                 gso_ctx = NULL;
1459         } else if (*other_fd != -1) {
1460                 /* Only other_fd exists. dup it */
1461                 *fd = dup(*other_fd);
1462                 if (*fd < 0) {
1463                         *fd = -1;
1464                         TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1465                         return -1;
1466                 }
1467                 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1468                         pmd->name, *other_fd, dir, qid, *fd);
1469         } else {
1470                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1471                 *fd = tun_alloc(pmd, 0);
1472                 if (*fd < 0) {
1473                         *fd = -1; /* restore original value */
1474                         TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1475                         return -1;
1476                 }
1477                 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1478                         pmd->name, dir, qid, *fd);
1479         }
1480
1481         tx->mtu = &dev->data->mtu;
1482         rx->rxmode = &dev->data->dev_conf.rxmode;
1483         if (gso_ctx) {
1484                 ret = tap_gso_ctx_setup(gso_ctx, dev);
1485                 if (ret)
1486                         return -1;
1487         }
1488
1489         tx->type = pmd->type;
1490
1491         return *fd;
1492 }
1493
1494 static int
1495 tap_rx_queue_setup(struct rte_eth_dev *dev,
1496                    uint16_t rx_queue_id,
1497                    uint16_t nb_rx_desc,
1498                    unsigned int socket_id,
1499                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1500                    struct rte_mempool *mp)
1501 {
1502         struct pmd_internals *internals = dev->data->dev_private;
1503         struct pmd_process_private *process_private = dev->process_private;
1504         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1505         struct rte_mbuf **tmp = &rxq->pool;
1506         long iov_max = sysconf(_SC_IOV_MAX);
1507
1508         if (iov_max <= 0) {
1509                 TAP_LOG(WARNING,
1510                         "_SC_IOV_MAX is not defined. Using %d as default",
1511                         TAP_IOV_DEFAULT_MAX);
1512                 iov_max = TAP_IOV_DEFAULT_MAX;
1513         }
1514         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1515         struct iovec (*iovecs)[nb_desc + 1];
1516         int data_off = RTE_PKTMBUF_HEADROOM;
1517         int ret = 0;
1518         int fd;
1519         int i;
1520
1521         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1522                 TAP_LOG(WARNING,
1523                         "nb_rx_queues %d too small or mempool NULL",
1524                         dev->data->nb_rx_queues);
1525                 return -1;
1526         }
1527
1528         rxq->mp = mp;
1529         rxq->trigger_seen = 1; /* force initial burst */
1530         rxq->in_port = dev->data->port_id;
1531         rxq->queue_id = rx_queue_id;
1532         rxq->nb_rx_desc = nb_desc;
1533         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1534                                     socket_id);
1535         if (!iovecs) {
1536                 TAP_LOG(WARNING,
1537                         "%s: Couldn't allocate %d RX descriptors",
1538                         dev->device->name, nb_desc);
1539                 return -ENOMEM;
1540         }
1541         rxq->iovecs = iovecs;
1542
1543         dev->data->rx_queues[rx_queue_id] = rxq;
1544         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1545         if (fd == -1) {
1546                 ret = fd;
1547                 goto error;
1548         }
1549
1550         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1551         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1552
1553         for (i = 1; i <= nb_desc; i++) {
1554                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1555                 if (!*tmp) {
1556                         TAP_LOG(WARNING,
1557                                 "%s: couldn't allocate memory for queue %d",
1558                                 dev->device->name, rx_queue_id);
1559                         ret = -ENOMEM;
1560                         goto error;
1561                 }
1562                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1563                 (*rxq->iovecs)[i].iov_base =
1564                         (char *)(*tmp)->buf_addr + data_off;
1565                 data_off = 0;
1566                 tmp = &(*tmp)->next;
1567         }
1568
1569         TAP_LOG(DEBUG, "  RX TUNTAP device name %s, qid %d on fd %d",
1570                 internals->name, rx_queue_id,
1571                 process_private->rxq_fds[rx_queue_id]);
1572
1573         return 0;
1574
1575 error:
1576         tap_rxq_pool_free(rxq->pool);
1577         rxq->pool = NULL;
1578         rte_free(rxq->iovecs);
1579         rxq->iovecs = NULL;
1580         return ret;
1581 }
1582
1583 static int
1584 tap_tx_queue_setup(struct rte_eth_dev *dev,
1585                    uint16_t tx_queue_id,
1586                    uint16_t nb_tx_desc __rte_unused,
1587                    unsigned int socket_id __rte_unused,
1588                    const struct rte_eth_txconf *tx_conf)
1589 {
1590         struct pmd_internals *internals = dev->data->dev_private;
1591         struct pmd_process_private *process_private = dev->process_private;
1592         struct tx_queue *txq;
1593         int ret;
1594         uint64_t offloads;
1595
1596         if (tx_queue_id >= dev->data->nb_tx_queues)
1597                 return -1;
1598         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1599         txq = dev->data->tx_queues[tx_queue_id];
1600         txq->out_port = dev->data->port_id;
1601         txq->queue_id = tx_queue_id;
1602
1603         offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1604         txq->csum = !!(offloads &
1605                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
1606                          DEV_TX_OFFLOAD_UDP_CKSUM |
1607                          DEV_TX_OFFLOAD_TCP_CKSUM));
1608
1609         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1610         if (ret == -1)
1611                 return -1;
1612         TAP_LOG(DEBUG,
1613                 "  TX TUNTAP device name %s, qid %d on fd %d csum %s",
1614                 internals->name, tx_queue_id,
1615                 process_private->txq_fds[tx_queue_id],
1616                 txq->csum ? "on" : "off");
1617
1618         return 0;
1619 }
1620
1621 static int
1622 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1623 {
1624         struct pmd_internals *pmd = dev->data->dev_private;
1625         struct ifreq ifr = { .ifr_mtu = mtu };
1626         int err = 0;
1627
1628         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1629         if (!err)
1630                 dev->data->mtu = mtu;
1631
1632         return err;
1633 }
1634
1635 static int
1636 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1637                      struct rte_ether_addr *mc_addr_set __rte_unused,
1638                      uint32_t nb_mc_addr __rte_unused)
1639 {
1640         /*
1641          * Nothing to do actually: the tap has no filtering whatsoever, every
1642          * packet is received.
1643          */
1644         return 0;
1645 }
1646
1647 static int
1648 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1649 {
1650         struct rte_eth_dev *dev = arg;
1651         struct pmd_internals *pmd = dev->data->dev_private;
1652         struct ifinfomsg *info = NLMSG_DATA(nh);
1653
1654         if (nh->nlmsg_type != RTM_NEWLINK ||
1655             (info->ifi_index != pmd->if_index &&
1656              info->ifi_index != pmd->remote_if_index))
1657                 return 0;
1658         return tap_link_update(dev, 0);
1659 }
1660
1661 static void
1662 tap_dev_intr_handler(void *cb_arg)
1663 {
1664         struct rte_eth_dev *dev = cb_arg;
1665         struct pmd_internals *pmd = dev->data->dev_private;
1666
1667         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1668 }
1669
1670 static int
1671 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1672 {
1673         struct pmd_internals *pmd = dev->data->dev_private;
1674         int ret;
1675
1676         /* In any case, disable interrupt if the conf is no longer there. */
1677         if (!dev->data->dev_conf.intr_conf.lsc) {
1678                 if (pmd->intr_handle.fd != -1) {
1679                         goto clean;
1680                 }
1681                 return 0;
1682         }
1683         if (set) {
1684                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1685                 if (unlikely(pmd->intr_handle.fd == -1))
1686                         return -EBADF;
1687                 return rte_intr_callback_register(
1688                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1689         }
1690
1691 clean:
1692         do {
1693                 ret = rte_intr_callback_unregister(&pmd->intr_handle,
1694                         tap_dev_intr_handler, dev);
1695                 if (ret >= 0) {
1696                         break;
1697                 } else if (ret == -EAGAIN) {
1698                         rte_delay_ms(100);
1699                 } else {
1700                         TAP_LOG(ERR, "intr callback unregister failed: %d",
1701                                      ret);
1702                         break;
1703                 }
1704         } while (true);
1705
1706         tap_nl_final(pmd->intr_handle.fd);
1707         pmd->intr_handle.fd = -1;
1708
1709         return 0;
1710 }
1711
1712 static int
1713 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1714 {
1715         int err;
1716
1717         err = tap_lsc_intr_handle_set(dev, set);
1718         if (err < 0) {
1719                 if (!set)
1720                         tap_rx_intr_vec_set(dev, 0);
1721                 return err;
1722         }
1723         err = tap_rx_intr_vec_set(dev, set);
1724         if (err && set)
1725                 tap_lsc_intr_handle_set(dev, 0);
1726         return err;
1727 }
1728
1729 static const uint32_t*
1730 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1731 {
1732         static const uint32_t ptypes[] = {
1733                 RTE_PTYPE_INNER_L2_ETHER,
1734                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1735                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1736                 RTE_PTYPE_INNER_L3_IPV4,
1737                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1738                 RTE_PTYPE_INNER_L3_IPV6,
1739                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1740                 RTE_PTYPE_INNER_L4_FRAG,
1741                 RTE_PTYPE_INNER_L4_UDP,
1742                 RTE_PTYPE_INNER_L4_TCP,
1743                 RTE_PTYPE_INNER_L4_SCTP,
1744                 RTE_PTYPE_L2_ETHER,
1745                 RTE_PTYPE_L2_ETHER_VLAN,
1746                 RTE_PTYPE_L2_ETHER_QINQ,
1747                 RTE_PTYPE_L3_IPV4,
1748                 RTE_PTYPE_L3_IPV4_EXT,
1749                 RTE_PTYPE_L3_IPV6_EXT,
1750                 RTE_PTYPE_L3_IPV6,
1751                 RTE_PTYPE_L4_FRAG,
1752                 RTE_PTYPE_L4_UDP,
1753                 RTE_PTYPE_L4_TCP,
1754                 RTE_PTYPE_L4_SCTP,
1755         };
1756
1757         return ptypes;
1758 }
1759
1760 static int
1761 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1762                   struct rte_eth_fc_conf *fc_conf)
1763 {
1764         fc_conf->mode = RTE_FC_NONE;
1765         return 0;
1766 }
1767
1768 static int
1769 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1770                   struct rte_eth_fc_conf *fc_conf)
1771 {
1772         if (fc_conf->mode != RTE_FC_NONE)
1773                 return -ENOTSUP;
1774         return 0;
1775 }
1776
1777 /**
1778  * DPDK callback to update the RSS hash configuration.
1779  *
1780  * @param dev
1781  *   Pointer to Ethernet device structure.
1782  * @param[in] rss_conf
1783  *   RSS configuration data.
1784  *
1785  * @return
1786  *   0 on success, a negative errno value otherwise and rte_errno is set.
1787  */
1788 static int
1789 tap_rss_hash_update(struct rte_eth_dev *dev,
1790                 struct rte_eth_rss_conf *rss_conf)
1791 {
1792         if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1793                 rte_errno = EINVAL;
1794                 return -rte_errno;
1795         }
1796         if (rss_conf->rss_key && rss_conf->rss_key_len) {
1797                 /*
1798                  * Currently TAP RSS key is hard coded
1799                  * and cannot be updated
1800                  */
1801                 TAP_LOG(ERR,
1802                         "port %u RSS key cannot be updated",
1803                         dev->data->port_id);
1804                 rte_errno = EINVAL;
1805                 return -rte_errno;
1806         }
1807         return 0;
1808 }
1809
1810 static int
1811 tap_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1812 {
1813         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1814
1815         return 0;
1816 }
1817
1818 static int
1819 tap_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1820 {
1821         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1822
1823         return 0;
1824 }
1825
1826 static int
1827 tap_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1828 {
1829         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1830
1831         return 0;
1832 }
1833
1834 static int
1835 tap_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1836 {
1837         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1838
1839         return 0;
1840 }
1841 static const struct eth_dev_ops ops = {
1842         .dev_start              = tap_dev_start,
1843         .dev_stop               = tap_dev_stop,
1844         .dev_close              = tap_dev_close,
1845         .dev_configure          = tap_dev_configure,
1846         .dev_infos_get          = tap_dev_info,
1847         .rx_queue_setup         = tap_rx_queue_setup,
1848         .tx_queue_setup         = tap_tx_queue_setup,
1849         .rx_queue_start         = tap_rx_queue_start,
1850         .tx_queue_start         = tap_tx_queue_start,
1851         .rx_queue_stop          = tap_rx_queue_stop,
1852         .tx_queue_stop          = tap_tx_queue_stop,
1853         .rx_queue_release       = tap_rx_queue_release,
1854         .tx_queue_release       = tap_tx_queue_release,
1855         .flow_ctrl_get          = tap_flow_ctrl_get,
1856         .flow_ctrl_set          = tap_flow_ctrl_set,
1857         .link_update            = tap_link_update,
1858         .dev_set_link_up        = tap_link_set_up,
1859         .dev_set_link_down      = tap_link_set_down,
1860         .promiscuous_enable     = tap_promisc_enable,
1861         .promiscuous_disable    = tap_promisc_disable,
1862         .allmulticast_enable    = tap_allmulti_enable,
1863         .allmulticast_disable   = tap_allmulti_disable,
1864         .mac_addr_set           = tap_mac_set,
1865         .mtu_set                = tap_mtu_set,
1866         .set_mc_addr_list       = tap_set_mc_addr_list,
1867         .stats_get              = tap_stats_get,
1868         .stats_reset            = tap_stats_reset,
1869         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1870         .rss_hash_update        = tap_rss_hash_update,
1871         .filter_ctrl            = tap_dev_filter_ctrl,
1872 };
1873
1874 static int
1875 eth_dev_tap_create(struct rte_vdev_device *vdev, const char *tap_name,
1876                    char *remote_iface, struct rte_ether_addr *mac_addr,
1877                    enum rte_tuntap_type type)
1878 {
1879         int numa_node = rte_socket_id();
1880         struct rte_eth_dev *dev;
1881         struct pmd_internals *pmd;
1882         struct pmd_process_private *process_private;
1883         const char *tuntap_name = tuntap_types[type];
1884         struct rte_eth_dev_data *data;
1885         struct ifreq ifr;
1886         int i;
1887
1888         TAP_LOG(DEBUG, "%s device on numa %u", tuntap_name, rte_socket_id());
1889
1890         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1891         if (!dev) {
1892                 TAP_LOG(ERR, "%s Unable to allocate device struct",
1893                                 tuntap_name);
1894                 goto error_exit_nodev;
1895         }
1896
1897         process_private = (struct pmd_process_private *)
1898                 rte_zmalloc_socket(tap_name, sizeof(struct pmd_process_private),
1899                         RTE_CACHE_LINE_SIZE, dev->device->numa_node);
1900
1901         if (process_private == NULL) {
1902                 TAP_LOG(ERR, "Failed to alloc memory for process private");
1903                 return -1;
1904         }
1905         pmd = dev->data->dev_private;
1906         dev->process_private = process_private;
1907         pmd->dev = dev;
1908         strlcpy(pmd->name, tap_name, sizeof(pmd->name));
1909         pmd->type = type;
1910         pmd->ka_fd = -1;
1911         pmd->nlsk_fd = -1;
1912         pmd->gso_ctx_mp = NULL;
1913
1914         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1915         if (pmd->ioctl_sock == -1) {
1916                 TAP_LOG(ERR,
1917                         "%s Unable to get a socket for management: %s",
1918                         tuntap_name, strerror(errno));
1919                 goto error_exit;
1920         }
1921
1922         /* Setup some default values */
1923         data = dev->data;
1924         data->dev_private = pmd;
1925         data->dev_flags = RTE_ETH_DEV_INTR_LSC |
1926                                 RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1927         data->numa_node = numa_node;
1928
1929         data->dev_link = pmd_link;
1930         data->mac_addrs = &pmd->eth_addr;
1931         /* Set the number of RX and TX queues */
1932         data->nb_rx_queues = 0;
1933         data->nb_tx_queues = 0;
1934
1935         dev->dev_ops = &ops;
1936         dev->rx_pkt_burst = pmd_rx_burst;
1937         dev->tx_pkt_burst = pmd_tx_burst;
1938
1939         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1940         pmd->intr_handle.fd = -1;
1941         dev->intr_handle = &pmd->intr_handle;
1942
1943         /* Presetup the fds to -1 as being not valid */
1944         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1945                 process_private->rxq_fds[i] = -1;
1946                 process_private->txq_fds[i] = -1;
1947         }
1948
1949         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1950                 if (rte_is_zero_ether_addr(mac_addr))
1951                         rte_eth_random_addr((uint8_t *)&pmd->eth_addr);
1952                 else
1953                         rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1954         }
1955
1956         /*
1957          * Allocate a TUN device keep-alive file descriptor that will only be
1958          * closed when the TUN device itself is closed or removed.
1959          * This keep-alive file descriptor will guarantee that the TUN device
1960          * exists even when all of its queues are closed
1961          */
1962         pmd->ka_fd = tun_alloc(pmd, 1);
1963         if (pmd->ka_fd == -1) {
1964                 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
1965                 goto error_exit;
1966         }
1967         TAP_LOG(DEBUG, "allocated %s", pmd->name);
1968
1969         ifr.ifr_mtu = dev->data->mtu;
1970         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1971                 goto error_exit;
1972
1973         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1974                 memset(&ifr, 0, sizeof(struct ifreq));
1975                 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1976                 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1977                                 RTE_ETHER_ADDR_LEN);
1978                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1979                         goto error_exit;
1980         }
1981
1982         /*
1983          * Set up everything related to rte_flow:
1984          * - netlink socket
1985          * - tap / remote if_index
1986          * - mandatory QDISCs
1987          * - rte_flow actual/implicit lists
1988          * - implicit rules
1989          */
1990         pmd->nlsk_fd = tap_nl_init(0);
1991         if (pmd->nlsk_fd == -1) {
1992                 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
1993                         pmd->name);
1994                 goto disable_rte_flow;
1995         }
1996         pmd->if_index = if_nametoindex(pmd->name);
1997         if (!pmd->if_index) {
1998                 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
1999                 goto disable_rte_flow;
2000         }
2001         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
2002                 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
2003                         pmd->name);
2004                 goto disable_rte_flow;
2005         }
2006         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
2007                 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
2008                         pmd->name);
2009                 goto disable_rte_flow;
2010         }
2011         LIST_INIT(&pmd->flows);
2012
2013         if (strlen(remote_iface)) {
2014                 pmd->remote_if_index = if_nametoindex(remote_iface);
2015                 if (!pmd->remote_if_index) {
2016                         TAP_LOG(ERR, "%s: failed to get %s if_index.",
2017                                 pmd->name, remote_iface);
2018                         goto error_remote;
2019                 }
2020                 strlcpy(pmd->remote_iface, remote_iface, RTE_ETH_NAME_MAX_LEN);
2021
2022                 /* Save state of remote device */
2023                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
2024
2025                 /* Replicate remote MAC address */
2026                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
2027                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
2028                                 pmd->name, pmd->remote_iface);
2029                         goto error_remote;
2030                 }
2031                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
2032                            RTE_ETHER_ADDR_LEN);
2033                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
2034                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
2035                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
2036                                 pmd->name, remote_iface);
2037                         goto error_remote;
2038                 }
2039
2040                 /*
2041                  * Flush usually returns negative value because it tries to
2042                  * delete every QDISC (and on a running device, one QDISC at
2043                  * least is needed). Ignore negative return value.
2044                  */
2045                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
2046                 if (qdisc_create_ingress(pmd->nlsk_fd,
2047                                          pmd->remote_if_index) < 0) {
2048                         TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
2049                                 pmd->remote_iface);
2050                         goto error_remote;
2051                 }
2052                 LIST_INIT(&pmd->implicit_flows);
2053                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
2054                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
2055                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
2056                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
2057                         TAP_LOG(ERR,
2058                                 "%s: failed to create implicit rules.",
2059                                 pmd->name);
2060                         goto error_remote;
2061                 }
2062         }
2063
2064         rte_eth_dev_probing_finish(dev);
2065         return 0;
2066
2067 disable_rte_flow:
2068         TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
2069                 strerror(errno), errno);
2070         if (strlen(remote_iface)) {
2071                 TAP_LOG(ERR, "Remote feature requires flow support.");
2072                 goto error_exit;
2073         }
2074         rte_eth_dev_probing_finish(dev);
2075         return 0;
2076
2077 error_remote:
2078         TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
2079                 strerror(errno), errno);
2080         tap_flow_implicit_flush(pmd, NULL);
2081
2082 error_exit:
2083         if (pmd->nlsk_fd != -1)
2084                 close(pmd->nlsk_fd);
2085         if (pmd->ka_fd != -1)
2086                 close(pmd->ka_fd);
2087         if (pmd->ioctl_sock != -1)
2088                 close(pmd->ioctl_sock);
2089         /* mac_addrs must not be freed alone because part of dev_private */
2090         dev->data->mac_addrs = NULL;
2091         rte_eth_dev_release_port(dev);
2092
2093 error_exit_nodev:
2094         TAP_LOG(ERR, "%s Unable to initialize %s",
2095                 tuntap_name, rte_vdev_device_name(vdev));
2096
2097         return -EINVAL;
2098 }
2099
2100 /* make sure name is a possible Linux network device name */
2101 static bool
2102 is_valid_iface(const char *name)
2103 {
2104         if (*name == '\0')
2105                 return false;
2106
2107         if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
2108                 return false;
2109
2110         while (*name) {
2111                 if (*name == '/' || *name == ':' || isspace(*name))
2112                         return false;
2113                 name++;
2114         }
2115         return true;
2116 }
2117
2118 static int
2119 set_interface_name(const char *key __rte_unused,
2120                    const char *value,
2121                    void *extra_args)
2122 {
2123         char *name = (char *)extra_args;
2124
2125         if (value) {
2126                 if (!is_valid_iface(value)) {
2127                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2128                                 value);
2129                         return -1;
2130                 }
2131                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2132         } else {
2133                 /* use tap%d which causes kernel to choose next available */
2134                 strlcpy(name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2135         }
2136         return 0;
2137 }
2138
2139 static int
2140 set_remote_iface(const char *key __rte_unused,
2141                  const char *value,
2142                  void *extra_args)
2143 {
2144         char *name = (char *)extra_args;
2145
2146         if (value) {
2147                 if (!is_valid_iface(value)) {
2148                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2149                                 value);
2150                         return -1;
2151                 }
2152                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2153         }
2154
2155         return 0;
2156 }
2157
2158 static int parse_user_mac(struct rte_ether_addr *user_mac,
2159                 const char *value)
2160 {
2161         unsigned int index = 0;
2162         char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
2163
2164         if (user_mac == NULL || value == NULL)
2165                 return 0;
2166
2167         strlcpy(mac_temp, value, sizeof(mac_temp));
2168         mac_byte = strtok(mac_temp, ":");
2169
2170         while ((mac_byte != NULL) &&
2171                         (strlen(mac_byte) <= 2) &&
2172                         (strlen(mac_byte) == strspn(mac_byte,
2173                                         ETH_TAP_CMP_MAC_FMT))) {
2174                 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
2175                 mac_byte = strtok(NULL, ":");
2176         }
2177
2178         return index;
2179 }
2180
2181 static int
2182 set_mac_type(const char *key __rte_unused,
2183              const char *value,
2184              void *extra_args)
2185 {
2186         struct rte_ether_addr *user_mac = extra_args;
2187
2188         if (!value)
2189                 return 0;
2190
2191         if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
2192                 static int iface_idx;
2193
2194                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
2195                 memcpy((char *)user_mac->addr_bytes, "\0dtap",
2196                         RTE_ETHER_ADDR_LEN);
2197                 user_mac->addr_bytes[RTE_ETHER_ADDR_LEN - 1] =
2198                         iface_idx++ + '0';
2199                 goto success;
2200         }
2201
2202         if (parse_user_mac(user_mac, value) != 6)
2203                 goto error;
2204 success:
2205         TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
2206         return 0;
2207
2208 error:
2209         TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
2210                 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
2211         return -1;
2212 }
2213
2214 /*
2215  * Open a TUN interface device. TUN PMD
2216  * 1) sets tap_type as false
2217  * 2) intakes iface as argument.
2218  * 3) as interface is virtual set speed to 10G
2219  */
2220 static int
2221 rte_pmd_tun_probe(struct rte_vdev_device *dev)
2222 {
2223         const char *name, *params;
2224         int ret;
2225         struct rte_kvargs *kvlist = NULL;
2226         char tun_name[RTE_ETH_NAME_MAX_LEN];
2227         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2228         struct rte_eth_dev *eth_dev;
2229
2230         name = rte_vdev_device_name(dev);
2231         params = rte_vdev_device_args(dev);
2232         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2233
2234         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
2235             strlen(params) == 0) {
2236                 eth_dev = rte_eth_dev_attach_secondary(name);
2237                 if (!eth_dev) {
2238                         TAP_LOG(ERR, "Failed to probe %s", name);
2239                         return -1;
2240                 }
2241                 eth_dev->dev_ops = &ops;
2242                 eth_dev->device = &dev->device;
2243                 rte_eth_dev_probing_finish(eth_dev);
2244                 return 0;
2245         }
2246
2247         /* use tun%d which causes kernel to choose next available */
2248         strlcpy(tun_name, DEFAULT_TUN_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2249
2250         if (params && (params[0] != '\0')) {
2251                 TAP_LOG(DEBUG, "parameters (%s)", params);
2252
2253                 kvlist = rte_kvargs_parse(params, valid_arguments);
2254                 if (kvlist) {
2255                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2256                                 ret = rte_kvargs_process(kvlist,
2257                                         ETH_TAP_IFACE_ARG,
2258                                         &set_interface_name,
2259                                         tun_name);
2260
2261                                 if (ret == -1)
2262                                         goto leave;
2263                         }
2264                 }
2265         }
2266         pmd_link.link_speed = ETH_SPEED_NUM_10G;
2267
2268         TAP_LOG(DEBUG, "Initializing pmd_tun for %s", name);
2269
2270         ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0,
2271                                  ETH_TUNTAP_TYPE_TUN);
2272
2273 leave:
2274         if (ret == -1) {
2275                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2276                         name, tun_name);
2277         }
2278         rte_kvargs_free(kvlist);
2279
2280         return ret;
2281 }
2282
2283 /* Request queue file descriptors from secondary to primary. */
2284 static int
2285 tap_mp_attach_queues(const char *port_name, struct rte_eth_dev *dev)
2286 {
2287         int ret;
2288         struct timespec timeout = {.tv_sec = 1, .tv_nsec = 0};
2289         struct rte_mp_msg request, *reply;
2290         struct rte_mp_reply replies;
2291         struct ipc_queues *request_param = (struct ipc_queues *)request.param;
2292         struct ipc_queues *reply_param;
2293         struct pmd_process_private *process_private = dev->process_private;
2294         int queue, fd_iterator;
2295
2296         /* Prepare the request */
2297         memset(&request, 0, sizeof(request));
2298         strlcpy(request.name, TAP_MP_KEY, sizeof(request.name));
2299         strlcpy(request_param->port_name, port_name,
2300                 sizeof(request_param->port_name));
2301         request.len_param = sizeof(*request_param);
2302         /* Send request and receive reply */
2303         ret = rte_mp_request_sync(&request, &replies, &timeout);
2304         if (ret < 0 || replies.nb_received != 1) {
2305                 TAP_LOG(ERR, "Failed to request queues from primary: %d",
2306                         rte_errno);
2307                 return -1;
2308         }
2309         reply = &replies.msgs[0];
2310         reply_param = (struct ipc_queues *)reply->param;
2311         TAP_LOG(DEBUG, "Received IPC reply for %s", reply_param->port_name);
2312
2313         /* Attach the queues from received file descriptors */
2314         if (reply_param->rxq_count + reply_param->txq_count != reply->num_fds) {
2315                 TAP_LOG(ERR, "Unexpected number of fds received");
2316                 return -1;
2317         }
2318
2319         dev->data->nb_rx_queues = reply_param->rxq_count;
2320         dev->data->nb_tx_queues = reply_param->txq_count;
2321         fd_iterator = 0;
2322         for (queue = 0; queue < reply_param->rxq_count; queue++)
2323                 process_private->rxq_fds[queue] = reply->fds[fd_iterator++];
2324         for (queue = 0; queue < reply_param->txq_count; queue++)
2325                 process_private->txq_fds[queue] = reply->fds[fd_iterator++];
2326         free(reply);
2327         return 0;
2328 }
2329
2330 /* Send the queue file descriptors from the primary process to secondary. */
2331 static int
2332 tap_mp_sync_queues(const struct rte_mp_msg *request, const void *peer)
2333 {
2334         struct rte_eth_dev *dev;
2335         struct pmd_process_private *process_private;
2336         struct rte_mp_msg reply;
2337         const struct ipc_queues *request_param =
2338                 (const struct ipc_queues *)request->param;
2339         struct ipc_queues *reply_param =
2340                 (struct ipc_queues *)reply.param;
2341         uint16_t port_id;
2342         int queue;
2343         int ret;
2344
2345         /* Get requested port */
2346         TAP_LOG(DEBUG, "Received IPC request for %s", request_param->port_name);
2347         ret = rte_eth_dev_get_port_by_name(request_param->port_name, &port_id);
2348         if (ret) {
2349                 TAP_LOG(ERR, "Failed to get port id for %s",
2350                         request_param->port_name);
2351                 return -1;
2352         }
2353         dev = &rte_eth_devices[port_id];
2354         process_private = dev->process_private;
2355
2356         /* Fill file descriptors for all queues */
2357         reply.num_fds = 0;
2358         reply_param->rxq_count = 0;
2359         if (dev->data->nb_rx_queues + dev->data->nb_tx_queues >
2360                         RTE_MP_MAX_FD_NUM){
2361                 TAP_LOG(ERR, "Number of rx/tx queues exceeds max number of fds");
2362                 return -1;
2363         }
2364
2365         for (queue = 0; queue < dev->data->nb_rx_queues; queue++) {
2366                 reply.fds[reply.num_fds++] = process_private->rxq_fds[queue];
2367                 reply_param->rxq_count++;
2368         }
2369         RTE_ASSERT(reply_param->rxq_count == dev->data->nb_rx_queues);
2370
2371         reply_param->txq_count = 0;
2372         for (queue = 0; queue < dev->data->nb_tx_queues; queue++) {
2373                 reply.fds[reply.num_fds++] = process_private->txq_fds[queue];
2374                 reply_param->txq_count++;
2375         }
2376         RTE_ASSERT(reply_param->txq_count == dev->data->nb_tx_queues);
2377
2378         /* Send reply */
2379         strlcpy(reply.name, request->name, sizeof(reply.name));
2380         strlcpy(reply_param->port_name, request_param->port_name,
2381                 sizeof(reply_param->port_name));
2382         reply.len_param = sizeof(*reply_param);
2383         if (rte_mp_reply(&reply, peer) < 0) {
2384                 TAP_LOG(ERR, "Failed to reply an IPC request to sync queues");
2385                 return -1;
2386         }
2387         return 0;
2388 }
2389
2390 /* Open a TAP interface device.
2391  */
2392 static int
2393 rte_pmd_tap_probe(struct rte_vdev_device *dev)
2394 {
2395         const char *name, *params;
2396         int ret;
2397         struct rte_kvargs *kvlist = NULL;
2398         int speed;
2399         char tap_name[RTE_ETH_NAME_MAX_LEN];
2400         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2401         struct rte_ether_addr user_mac = { .addr_bytes = {0} };
2402         struct rte_eth_dev *eth_dev;
2403         int tap_devices_count_increased = 0;
2404
2405         name = rte_vdev_device_name(dev);
2406         params = rte_vdev_device_args(dev);
2407
2408         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
2409                 eth_dev = rte_eth_dev_attach_secondary(name);
2410                 if (!eth_dev) {
2411                         TAP_LOG(ERR, "Failed to probe %s", name);
2412                         return -1;
2413                 }
2414                 eth_dev->dev_ops = &ops;
2415                 eth_dev->device = &dev->device;
2416                 eth_dev->rx_pkt_burst = pmd_rx_burst;
2417                 eth_dev->tx_pkt_burst = pmd_tx_burst;
2418                 if (!rte_eal_primary_proc_alive(NULL)) {
2419                         TAP_LOG(ERR, "Primary process is missing");
2420                         return -1;
2421                 }
2422                 eth_dev->process_private = (struct pmd_process_private *)
2423                         rte_zmalloc_socket(name,
2424                                 sizeof(struct pmd_process_private),
2425                                 RTE_CACHE_LINE_SIZE,
2426                                 eth_dev->device->numa_node);
2427                 if (eth_dev->process_private == NULL) {
2428                         TAP_LOG(ERR,
2429                                 "Failed to alloc memory for process private");
2430                         return -1;
2431                 }
2432
2433                 ret = tap_mp_attach_queues(name, eth_dev);
2434                 if (ret != 0)
2435                         return -1;
2436                 rte_eth_dev_probing_finish(eth_dev);
2437                 return 0;
2438         }
2439
2440         speed = ETH_SPEED_NUM_10G;
2441
2442         /* use tap%d which causes kernel to choose next available */
2443         strlcpy(tap_name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2444         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2445
2446         if (params && (params[0] != '\0')) {
2447                 TAP_LOG(DEBUG, "parameters (%s)", params);
2448
2449                 kvlist = rte_kvargs_parse(params, valid_arguments);
2450                 if (kvlist) {
2451                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2452                                 ret = rte_kvargs_process(kvlist,
2453                                                          ETH_TAP_IFACE_ARG,
2454                                                          &set_interface_name,
2455                                                          tap_name);
2456                                 if (ret == -1)
2457                                         goto leave;
2458                         }
2459
2460                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
2461                                 ret = rte_kvargs_process(kvlist,
2462                                                          ETH_TAP_REMOTE_ARG,
2463                                                          &set_remote_iface,
2464                                                          remote_iface);
2465                                 if (ret == -1)
2466                                         goto leave;
2467                         }
2468
2469                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
2470                                 ret = rte_kvargs_process(kvlist,
2471                                                          ETH_TAP_MAC_ARG,
2472                                                          &set_mac_type,
2473                                                          &user_mac);
2474                                 if (ret == -1)
2475                                         goto leave;
2476                         }
2477                 }
2478         }
2479         pmd_link.link_speed = speed;
2480
2481         TAP_LOG(DEBUG, "Initializing pmd_tap for %s", name);
2482
2483         /* Register IPC feed callback */
2484         if (!tap_devices_count) {
2485                 ret = rte_mp_action_register(TAP_MP_KEY, tap_mp_sync_queues);
2486                 if (ret < 0 && rte_errno != ENOTSUP) {
2487                         TAP_LOG(ERR, "tap: Failed to register IPC callback: %s",
2488                                 strerror(rte_errno));
2489                         goto leave;
2490                 }
2491         }
2492         tap_devices_count++;
2493         tap_devices_count_increased = 1;
2494         ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
2495                 ETH_TUNTAP_TYPE_TAP);
2496
2497 leave:
2498         if (ret == -1) {
2499                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2500                         name, tap_name);
2501                 if (tap_devices_count_increased == 1) {
2502                         if (tap_devices_count == 1)
2503                                 rte_mp_action_unregister(TAP_MP_KEY);
2504                         tap_devices_count--;
2505                 }
2506         }
2507         rte_kvargs_free(kvlist);
2508
2509         return ret;
2510 }
2511
2512 /* detach a TUNTAP device.
2513  */
2514 static int
2515 rte_pmd_tap_remove(struct rte_vdev_device *dev)
2516 {
2517         struct rte_eth_dev *eth_dev = NULL;
2518
2519         /* find the ethdev entry */
2520         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
2521         if (!eth_dev)
2522                 return 0;
2523
2524         tap_dev_close(eth_dev);
2525         rte_eth_dev_release_port(eth_dev);
2526
2527         return 0;
2528 }
2529
2530 static struct rte_vdev_driver pmd_tun_drv = {
2531         .probe = rte_pmd_tun_probe,
2532         .remove = rte_pmd_tap_remove,
2533 };
2534
2535 static struct rte_vdev_driver pmd_tap_drv = {
2536         .probe = rte_pmd_tap_probe,
2537         .remove = rte_pmd_tap_remove,
2538 };
2539
2540 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
2541 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
2542 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
2543 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
2544                               ETH_TAP_IFACE_ARG "=<string> ");
2545 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
2546                               ETH_TAP_IFACE_ARG "=<string> "
2547                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
2548                               ETH_TAP_REMOTE_ARG "=<string>");
2549 RTE_LOG_REGISTER(tap_logtype, pmd.net.tap, NOTICE);