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 <ethdev_driver.h>
  11 #include <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         struct rte_udp_hdr *udp_hdr;
 307
 308         if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
 309                 l2_len += 4;
 310         else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
 311                 l2_len += 8;
 312         /* Don't verify checksum for packets with discontinuous L2 header */
 313         if (unlikely(l2_len + sizeof(struct rte_ipv4_hdr) >
 314                      rte_pktmbuf_data_len(mbuf)))
 315                 return;
 316         l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
 317         if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 318                 struct rte_ipv4_hdr *iph = l3_hdr;
 319
 320                 l3_len = rte_ipv4_hdr_len(iph);
 321                 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
 322                         return;
 323                 /* check that the total length reported by header is not
 324                  * greater than the total received size
 325                  */
 326                 if (l2_len + rte_be_to_cpu_16(iph->total_length) >
 327                                 rte_pktmbuf_data_len(mbuf))
 328                         return;
 329
 330                 cksum = ~rte_raw_cksum(iph, l3_len);
 331                 mbuf->ol_flags |= cksum ?
 332                         PKT_RX_IP_CKSUM_BAD :
 333                         PKT_RX_IP_CKSUM_GOOD;
 334         } else if (l3 == RTE_PTYPE_L3_IPV6) {
 335                 struct rte_ipv6_hdr *iph = l3_hdr;
 336
 337                 l3_len = sizeof(struct rte_ipv6_hdr);
 338                 /* check that the total length reported by header is not
 339                  * greater than the total received size
 340                  */
 341                 if (l2_len + l3_len + rte_be_to_cpu_16(iph->payload_len) >
 342                                 rte_pktmbuf_data_len(mbuf))
 343                         return;
 344         } else {
 345                 /* - RTE_PTYPE_L3_IPV4_EXT_UNKNOWN cannot happen because
 346                  *   mbuf->packet_type is filled by rte_net_get_ptype() which
 347                  *   never returns this value.
 348                  * - IPv6 extensions are not supported.
 349                  */
 350                 return;
 351         }
 352         if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
 353                 int cksum_ok;
 354
 355                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 356                 /* Don't verify checksum for multi-segment packets. */
 357                 if (mbuf->nb_segs > 1)
 358                         return;
 359                 if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 360                         if (l4 == RTE_PTYPE_L4_UDP) {
 361                                 udp_hdr = (struct rte_udp_hdr *)l4_hdr;
 362                                 if (udp_hdr->dgram_cksum == 0) {
 363                                         /*
 364                                          * For IPv4, a zero UDP checksum
 365                                          * indicates that the sender did not
 366                                          * generate one [RFC 768].
 367                                          */
 368                                         mbuf->ol_flags |= PKT_RX_L4_CKSUM_NONE;
 369                                         return;
 370                                 }
 371                         }
 372                         cksum_ok = !rte_ipv4_udptcp_cksum_verify(l3_hdr,
 373                                                                  l4_hdr);
 374                 } else { /* l3 == RTE_PTYPE_L3_IPV6, checked above */
 375                         cksum_ok = !rte_ipv6_udptcp_cksum_verify(l3_hdr,
 376                                                                  l4_hdr);
 377                 }
 378                 mbuf->ol_flags |= cksum_ok ?
 379                         PKT_RX_L4_CKSUM_GOOD : PKT_RX_L4_CKSUM_BAD;
 380         }
 381 }
 382
 383 static uint64_t
 384 tap_rx_offload_get_queue_capa(void)
 385 {
 386         return DEV_RX_OFFLOAD_SCATTER |
 387                DEV_RX_OFFLOAD_IPV4_CKSUM |
 388                DEV_RX_OFFLOAD_UDP_CKSUM |
 389                DEV_RX_OFFLOAD_TCP_CKSUM;
 390 }
 391
 392 static void
 393 tap_rxq_pool_free(struct rte_mbuf *pool)
 394 {
 395         struct rte_mbuf *mbuf = pool;
 396         uint16_t nb_segs = 1;
 397
 398         if (mbuf == NULL)
 399                 return;
 400
 401         while (mbuf->next) {
 402                 mbuf = mbuf->next;
 403                 nb_segs++;
 404         }
 405         pool->nb_segs = nb_segs;
 406         rte_pktmbuf_free(pool);
 407 }
 408
 409 /* Callback to handle the rx burst of packets to the correct interface and
 410  * file descriptor(s) in a multi-queue setup.
 411  */
 412 static uint16_t
 413 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 414 {
 415         struct rx_queue *rxq = queue;
 416         struct pmd_process_private *process_private;
 417         uint16_t num_rx;
 418         unsigned long num_rx_bytes = 0;
 419         uint32_t trigger = tap_trigger;
 420
 421         if (trigger == rxq->trigger_seen)
 422                 return 0;
 423
 424         process_private = rte_eth_devices[rxq->in_port].process_private;
 425         for (num_rx = 0; num_rx < nb_pkts; ) {
 426                 struct rte_mbuf *mbuf = rxq->pool;
 427                 struct rte_mbuf *seg = NULL;
 428                 struct rte_mbuf *new_tail = NULL;
 429                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 430                 int len;
 431
 432                 len = readv(process_private->rxq_fds[rxq->queue_id],
 433                         *rxq->iovecs,
 434                         1 + (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
 435                              rxq->nb_rx_desc : 1));
 436                 if (len < (int)sizeof(struct tun_pi))
 437                         break;
 438
 439                 /* Packet couldn't fit in the provided mbuf */
 440                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 441                         rxq->stats.ierrors++;
 442                         continue;
 443                 }
 444
 445                 len -= sizeof(struct tun_pi);
 446
 447                 mbuf->pkt_len = len;
 448                 mbuf->port = rxq->in_port;
 449                 while (1) {
 450                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 451
 452                         if (unlikely(!buf)) {
 453                                 rxq->stats.rx_nombuf++;
 454                                 /* No new buf has been allocated: do nothing */
 455                                 if (!new_tail || !seg)
 456                                         goto end;
 457
 458                                 seg->next = NULL;
 459                                 tap_rxq_pool_free(mbuf);
 460
 461                                 goto end;
 462                         }
 463                         seg = seg ? seg->next : mbuf;
 464                         if (rxq->pool == mbuf)
 465                                 rxq->pool = buf;
 466                         if (new_tail)
 467                                 new_tail->next = buf;
 468                         new_tail = buf;
 469                         new_tail->next = seg->next;
 470
 471                         /* iovecs[0] is reserved for packet info (pi) */
 472                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 473                                 buf->buf_len - data_off;
 474                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 475                                 (char *)buf->buf_addr + data_off;
 476
 477                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 478                         seg->data_off = data_off;
 479
 480                         len -= seg->data_len;
 481                         if (len <= 0)
 482                                 break;
 483                         mbuf->nb_segs++;
 484                         /* First segment has headroom, not the others */
 485                         data_off = 0;
 486                 }
 487                 seg->next = NULL;
 488                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 489                                                       RTE_PTYPE_ALL_MASK);
 490                 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
 491                         tap_verify_csum(mbuf);
 492
 493                 /* account for the receive frame */
 494                 bufs[num_rx++] = mbuf;
 495                 num_rx_bytes += mbuf->pkt_len;
 496         }
 497 end:
 498         rxq->stats.ipackets += num_rx;
 499         rxq->stats.ibytes += num_rx_bytes;
 500
 501         if (trigger && num_rx < nb_pkts)
 502                 rxq->trigger_seen = trigger;
 503
 504         return num_rx;
 505 }
 506
 507 static uint64_t
 508 tap_tx_offload_get_queue_capa(void)
 509 {
 510         return DEV_TX_OFFLOAD_MULTI_SEGS |
 511                DEV_TX_OFFLOAD_IPV4_CKSUM |
 512                DEV_TX_OFFLOAD_UDP_CKSUM |
 513                DEV_TX_OFFLOAD_TCP_CKSUM |
 514                DEV_TX_OFFLOAD_TCP_TSO;
 515 }
 516
 517 /* Finalize l4 checksum calculation */
 518 static void
 519 tap_tx_l4_cksum(uint16_t *l4_cksum, uint16_t l4_phdr_cksum,
 520                 uint32_t l4_raw_cksum)
 521 {
 522         if (l4_cksum) {
 523                 uint32_t cksum;
 524
 525                 cksum = __rte_raw_cksum_reduce(l4_raw_cksum);
 526                 cksum += l4_phdr_cksum;
 527
 528                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 529                 cksum = (~cksum) & 0xffff;
 530                 if (cksum == 0)
 531                         cksum = 0xffff;
 532                 *l4_cksum = cksum;
 533         }
 534 }
 535
 536 /* Accumaulate L4 raw checksums */
 537 static void
 538 tap_tx_l4_add_rcksum(char *l4_data, unsigned int l4_len, uint16_t *l4_cksum,
 539                         uint32_t *l4_raw_cksum)
 540 {
 541         if (l4_cksum == NULL)
 542                 return;
 543
 544         *l4_raw_cksum = __rte_raw_cksum(l4_data, l4_len, *l4_raw_cksum);
 545 }
 546
 547 /* L3 and L4 pseudo headers checksum offloads */
 548 static void
 549 tap_tx_l3_cksum(char *packet, uint64_t ol_flags, unsigned int l2_len,
 550                 unsigned int l3_len, unsigned int l4_len, uint16_t **l4_cksum,
 551                 uint16_t *l4_phdr_cksum, uint32_t *l4_raw_cksum)
 552 {
 553         void *l3_hdr = packet + l2_len;
 554
 555         if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
 556                 struct rte_ipv4_hdr *iph = l3_hdr;
 557                 uint16_t cksum;
 558
 559                 iph->hdr_checksum = 0;
 560                 cksum = rte_raw_cksum(iph, l3_len);
 561                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 562         }
 563         if (ol_flags & PKT_TX_L4_MASK) {
 564                 void *l4_hdr;
 565
 566                 l4_hdr = packet + l2_len + l3_len;
 567                 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
 568                         *l4_cksum = &((struct rte_udp_hdr *)l4_hdr)->dgram_cksum;
 569                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 570                         *l4_cksum = &((struct rte_tcp_hdr *)l4_hdr)->cksum;
 571                 else
 572                         return;
 573                 **l4_cksum = 0;
 574                 if (ol_flags & PKT_TX_IPV4)
 575                         *l4_phdr_cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 576                 else
 577                         *l4_phdr_cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 578                 *l4_raw_cksum = __rte_raw_cksum(l4_hdr, l4_len, 0);
 579         }
 580 }
 581
 582 static inline int
 583 tap_write_mbufs(struct tx_queue *txq, uint16_t num_mbufs,
 584                         struct rte_mbuf **pmbufs,
 585                         uint16_t *num_packets, unsigned long *num_tx_bytes)
 586 {
 587         int i;
 588         uint16_t l234_hlen;
 589         struct pmd_process_private *process_private;
 590
 591         process_private = rte_eth_devices[txq->out_port].process_private;
 592
 593         for (i = 0; i < num_mbufs; i++) {
 594                 struct rte_mbuf *mbuf = pmbufs[i];
 595                 struct iovec iovecs[mbuf->nb_segs + 2];
 596                 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
 597                 struct rte_mbuf *seg = mbuf;
 598                 char m_copy[mbuf->data_len];
 599                 int proto;
 600                 int n;
 601                 int j;
 602                 int k; /* current index in iovecs for copying segments */
 603                 uint16_t seg_len; /* length of first segment */
 604                 uint16_t nb_segs;
 605                 uint16_t *l4_cksum; /* l4 checksum (pseudo header + payload) */
 606                 uint32_t l4_raw_cksum = 0; /* TCP/UDP payload raw checksum */
 607                 uint16_t l4_phdr_cksum = 0; /* TCP/UDP pseudo header checksum */
 608                 uint16_t is_cksum = 0; /* in case cksum should be offloaded */
 609
 610                 l4_cksum = NULL;
 611                 if (txq->type == ETH_TUNTAP_TYPE_TUN) {
 612                         /*
 613                          * TUN and TAP are created with IFF_NO_PI disabled.
 614                          * For TUN PMD this mandatory as fields are used by
 615                          * Kernel tun.c to determine whether its IP or non IP
 616                          * packets.
 617                          *
 618                          * The logic fetches the first byte of data from mbuf
 619                          * then compares whether its v4 or v6. If first byte
 620                          * is 4 or 6, then protocol field is updated.
 621                          */
 622                         char *buff_data = rte_pktmbuf_mtod(seg, void *);
 623                         proto = (*buff_data & 0xf0);
 624                         pi.proto = (proto == 0x40) ?
 625                                 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4) :
 626                                 ((proto == 0x60) ?
 627                                         rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6) :
 628                                         0x00);
 629                 }
 630
 631                 k = 0;
 632                 iovecs[k].iov_base = &pi;
 633                 iovecs[k].iov_len = sizeof(pi);
 634                 k++;
 635
 636                 nb_segs = mbuf->nb_segs;
 637                 if (txq->csum &&
 638                     ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 639                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 640                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
 641                         is_cksum = 1;
 642
 643                         /* Support only packets with at least layer 4
 644                          * header included in the first segment
 645                          */
 646                         seg_len = rte_pktmbuf_data_len(mbuf);
 647                         l234_hlen = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
 648                         if (seg_len < l234_hlen)
 649                                 return -1;
 650
 651                         /* To change checksums, work on a * copy of l2, l3
 652                          * headers + l4 pseudo header
 653                          */
 654                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 655                                         l234_hlen);
 656                         tap_tx_l3_cksum(m_copy, mbuf->ol_flags,
 657                                        mbuf->l2_len, mbuf->l3_len, mbuf->l4_len,
 658                                        &l4_cksum, &l4_phdr_cksum,
 659                                        &l4_raw_cksum);
 660                         iovecs[k].iov_base = m_copy;
 661                         iovecs[k].iov_len = l234_hlen;
 662                         k++;
 663
 664                         /* Update next iovecs[] beyond l2, l3, l4 headers */
 665                         if (seg_len > l234_hlen) {
 666                                 iovecs[k].iov_len = seg_len - l234_hlen;
 667                                 iovecs[k].iov_base =
 668                                         rte_pktmbuf_mtod(seg, char *) +
 669                                                 l234_hlen;
 670                                 tap_tx_l4_add_rcksum(iovecs[k].iov_base,
 671                                         iovecs[k].iov_len, l4_cksum,
 672                                         &l4_raw_cksum);
 673                                 k++;
 674                                 nb_segs++;
 675                         }
 676                         seg = seg->next;
 677                 }
 678
 679                 for (j = k; j <= nb_segs; j++) {
 680                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 681                         iovecs[j].iov_base = rte_pktmbuf_mtod(seg, void *);
 682                         if (is_cksum)
 683                                 tap_tx_l4_add_rcksum(iovecs[j].iov_base,
 684                                         iovecs[j].iov_len, l4_cksum,
 685                                         &l4_raw_cksum);
 686                         seg = seg->next;
 687                 }
 688
 689                 if (is_cksum)
 690                         tap_tx_l4_cksum(l4_cksum, l4_phdr_cksum, l4_raw_cksum);
 691
 692                 /* copy the tx frame data */
 693                 n = writev(process_private->txq_fds[txq->queue_id], iovecs, j);
 694                 if (n <= 0)
 695                         return -1;
 696
 697                 (*num_packets)++;
 698                 (*num_tx_bytes) += rte_pktmbuf_pkt_len(mbuf);
 699         }
 700         return 0;
 701 }
 702
 703 /* Callback to handle sending packets from the tap interface
 704  */
 705 static uint16_t
 706 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 707 {
 708         struct tx_queue *txq = queue;
 709         uint16_t num_tx = 0;
 710         uint16_t num_packets = 0;
 711         unsigned long num_tx_bytes = 0;
 712         uint32_t max_size;
 713         int i;
 714
 715         if (unlikely(nb_pkts == 0))
 716                 return 0;
 717
 718         struct rte_mbuf *gso_mbufs[MAX_GSO_MBUFS];
 719         max_size = *txq->mtu + (RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN + 4);
 720         for (i = 0; i < nb_pkts; i++) {
 721                 struct rte_mbuf *mbuf_in = bufs[num_tx];
 722                 struct rte_mbuf **mbuf;
 723                 uint16_t num_mbufs = 0;
 724                 uint16_t tso_segsz = 0;
 725                 int ret;
 726                 int num_tso_mbufs;
 727                 uint16_t hdrs_len;
 728                 uint64_t tso;
 729
 730                 tso = mbuf_in->ol_flags & PKT_TX_TCP_SEG;
 731                 if (tso) {
 732                         struct rte_gso_ctx *gso_ctx = &txq->gso_ctx;
 733
 734                         /* TCP segmentation implies TCP checksum offload */
 735                         mbuf_in->ol_flags |= PKT_TX_TCP_CKSUM;
 736
 737                         /* gso size is calculated without RTE_ETHER_CRC_LEN */
 738                         hdrs_len = mbuf_in->l2_len + mbuf_in->l3_len +
 739                                         mbuf_in->l4_len;
 740                         tso_segsz = mbuf_in->tso_segsz + hdrs_len;
 741                         if (unlikely(tso_segsz == hdrs_len) ||
 742                                 tso_segsz > *txq->mtu) {
 743                                 txq->stats.errs++;
 744                                 break;
 745                         }
 746                         gso_ctx->gso_size = tso_segsz;
 747                         /* 'mbuf_in' packet to segment */
 748                         num_tso_mbufs = rte_gso_segment(mbuf_in,
 749                                 gso_ctx, /* gso control block */
 750                                 (struct rte_mbuf **)&gso_mbufs, /* out mbufs */
 751                                 RTE_DIM(gso_mbufs)); /* max tso mbufs */
 752
 753                         /* ret contains the number of new created mbufs */
 754                         if (num_tso_mbufs < 0)
 755                                 break;
 756
 757                         if (num_tso_mbufs >= 1) {
 758                                 mbuf = gso_mbufs;
 759                                 num_mbufs = num_tso_mbufs;
 760                         } else {
 761                                 /* 0 means it can be transmitted directly
 762                                  * without gso.
 763                                  */
 764                                 mbuf = &mbuf_in;
 765                                 num_mbufs = 1;
 766                         }
 767                 } else {
 768                         /* stats.errs will be incremented */
 769                         if (rte_pktmbuf_pkt_len(mbuf_in) > max_size)
 770                                 break;
 771
 772                         /* ret 0 indicates no new mbufs were created */
 773                         num_tso_mbufs = 0;
 774                         mbuf = &mbuf_in;
 775                         num_mbufs = 1;
 776                 }
 777
 778                 ret = tap_write_mbufs(txq, num_mbufs, mbuf,
 779                                 &num_packets, &num_tx_bytes);
 780                 if (ret == -1) {
 781                         txq->stats.errs++;
 782                         /* free tso mbufs */
 783                         if (num_tso_mbufs > 0)
 784                                 rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
 785                         break;
 786                 }
 787                 num_tx++;
 788                 /* free original mbuf */
 789                 rte_pktmbuf_free(mbuf_in);
 790                 /* free tso mbufs */
 791                 if (num_tso_mbufs > 0)
 792                         rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
 793         }
 794
 795         txq->stats.opackets += num_packets;
 796         txq->stats.errs += nb_pkts - num_tx;
 797         txq->stats.obytes += num_tx_bytes;
 798
 799         return num_tx;
 800 }
 801
 802 static const char *
 803 tap_ioctl_req2str(unsigned long request)
 804 {
 805         switch (request) {
 806         case SIOCSIFFLAGS:
 807                 return "SIOCSIFFLAGS";
 808         case SIOCGIFFLAGS:
 809                 return "SIOCGIFFLAGS";
 810         case SIOCGIFHWADDR:
 811                 return "SIOCGIFHWADDR";
 812         case SIOCSIFHWADDR:
 813                 return "SIOCSIFHWADDR";
 814         case SIOCSIFMTU:
 815                 return "SIOCSIFMTU";
 816         }
 817         return "UNKNOWN";
 818 }
 819
 820 static int
 821 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 822           struct ifreq *ifr, int set, enum ioctl_mode mode)
 823 {
 824         short req_flags = ifr->ifr_flags;
 825         int remote = pmd->remote_if_index &&
 826                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 827
 828         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 829                 return 0;
 830         /*
 831          * If there is a remote netdevice, apply ioctl on it, then apply it on
 832          * the tap netdevice.
 833          */
 834 apply:
 835         if (remote)
 836                 strlcpy(ifr->ifr_name, pmd->remote_iface, IFNAMSIZ);
 837         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 838                 strlcpy(ifr->ifr_name, pmd->name, IFNAMSIZ);
 839         switch (request) {
 840         case SIOCSIFFLAGS:
 841                 /* fetch current flags to leave other flags untouched */
 842                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 843                         goto error;
 844                 if (set)
 845                         ifr->ifr_flags |= req_flags;
 846                 else
 847                         ifr->ifr_flags &= ~req_flags;
 848                 break;
 849         case SIOCGIFFLAGS:
 850         case SIOCGIFHWADDR:
 851         case SIOCSIFHWADDR:
 852         case SIOCSIFMTU:
 853                 break;
 854         default:
 855                 TAP_LOG(WARNING, "%s: ioctl() called with wrong arg",
 856                         pmd->name);
 857                 return -EINVAL;
 858         }
 859         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 860                 goto error;
 861         if (remote-- && mode == LOCAL_AND_REMOTE)
 862                 goto apply;
 863         return 0;
 864
 865 error:
 866         TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
 867                 tap_ioctl_req2str(request), strerror(errno), errno);
 868         return -errno;
 869 }
 870
 871 static int
 872 tap_link_set_down(struct rte_eth_dev *dev)
 873 {
 874         struct pmd_internals *pmd = dev->data->dev_private;
 875         struct ifreq ifr = { .ifr_flags = IFF_UP };
 876
 877         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 878         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 879 }
 880
 881 static int
 882 tap_link_set_up(struct rte_eth_dev *dev)
 883 {
 884         struct pmd_internals *pmd = dev->data->dev_private;
 885         struct ifreq ifr = { .ifr_flags = IFF_UP };
 886
 887         dev->data->dev_link.link_status = ETH_LINK_UP;
 888         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 889 }
 890
 891 static int
 892 tap_dev_start(struct rte_eth_dev *dev)
 893 {
 894         int err, i;
 895
 896         err = tap_intr_handle_set(dev, 1);
 897         if (err)
 898                 return err;
 899
 900         err = tap_link_set_up(dev);
 901         if (err)
 902                 return err;
 903
 904         for (i = 0; i < dev->data->nb_tx_queues; i++)
 905                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 906         for (i = 0; i < dev->data->nb_rx_queues; i++)
 907                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 908
 909         return err;
 910 }
 911
 912 /* This function gets called when the current port gets stopped.
 913  */
 914 static int
 915 tap_dev_stop(struct rte_eth_dev *dev)
 916 {
 917         int i;
 918
 919         for (i = 0; i < dev->data->nb_tx_queues; i++)
 920                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 921         for (i = 0; i < dev->data->nb_rx_queues; i++)
 922                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 923
 924         tap_intr_handle_set(dev, 0);
 925         tap_link_set_down(dev);
 926
 927         return 0;
 928 }
 929
 930 static int
 931 tap_dev_configure(struct rte_eth_dev *dev)
 932 {
 933         struct pmd_internals *pmd = dev->data->dev_private;
 934
 935         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 936                 TAP_LOG(ERR,
 937                         "%s: number of rx queues %d exceeds max num of queues %d",
 938                         dev->device->name,
 939                         dev->data->nb_rx_queues,
 940                         RTE_PMD_TAP_MAX_QUEUES);
 941                 return -1;
 942         }
 943         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 944                 TAP_LOG(ERR,
 945                         "%s: number of tx queues %d exceeds max num of queues %d",
 946                         dev->device->name,
 947                         dev->data->nb_tx_queues,
 948                         RTE_PMD_TAP_MAX_QUEUES);
 949                 return -1;
 950         }
 951
 952         TAP_LOG(INFO, "%s: %s: TX configured queues number: %u",
 953                 dev->device->name, pmd->name, dev->data->nb_tx_queues);
 954
 955         TAP_LOG(INFO, "%s: %s: RX configured queues number: %u",
 956                 dev->device->name, pmd->name, dev->data->nb_rx_queues);
 957
 958         return 0;
 959 }
 960
 961 static uint32_t
 962 tap_dev_speed_capa(void)
 963 {
 964         uint32_t speed = pmd_link.link_speed;
 965         uint32_t capa = 0;
 966
 967         if (speed >= ETH_SPEED_NUM_10M)
 968                 capa |= ETH_LINK_SPEED_10M;
 969         if (speed >= ETH_SPEED_NUM_100M)
 970                 capa |= ETH_LINK_SPEED_100M;
 971         if (speed >= ETH_SPEED_NUM_1G)
 972                 capa |= ETH_LINK_SPEED_1G;
 973         if (speed >= ETH_SPEED_NUM_5G)
 974                 capa |= ETH_LINK_SPEED_2_5G;
 975         if (speed >= ETH_SPEED_NUM_5G)
 976                 capa |= ETH_LINK_SPEED_5G;
 977         if (speed >= ETH_SPEED_NUM_10G)
 978                 capa |= ETH_LINK_SPEED_10G;
 979         if (speed >= ETH_SPEED_NUM_20G)
 980                 capa |= ETH_LINK_SPEED_20G;
 981         if (speed >= ETH_SPEED_NUM_25G)
 982                 capa |= ETH_LINK_SPEED_25G;
 983         if (speed >= ETH_SPEED_NUM_40G)
 984                 capa |= ETH_LINK_SPEED_40G;
 985         if (speed >= ETH_SPEED_NUM_50G)
 986                 capa |= ETH_LINK_SPEED_50G;
 987         if (speed >= ETH_SPEED_NUM_56G)
 988                 capa |= ETH_LINK_SPEED_56G;
 989         if (speed >= ETH_SPEED_NUM_100G)
 990                 capa |= ETH_LINK_SPEED_100G;
 991
 992         return capa;
 993 }
 994
 995 static int
 996 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 997 {
 998         struct pmd_internals *internals = dev->data->dev_private;
 999
1000         dev_info->if_index = internals->if_index;
1001         dev_info->max_mac_addrs = 1;
1002         dev_info->max_rx_pktlen = (uint32_t)RTE_ETHER_MAX_VLAN_FRAME_LEN;
1003         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
1004         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
1005         dev_info->min_rx_bufsize = 0;
1006         dev_info->speed_capa = tap_dev_speed_capa();
1007         dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
1008         dev_info->rx_offload_capa = dev_info->rx_queue_offload_capa;
1009         dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
1010         dev_info->tx_offload_capa = dev_info->tx_queue_offload_capa;
1011         dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
1012         /*
1013          * limitation: TAP supports all of IP, UDP and TCP hash
1014          * functions together and not in partial combinations
1015          */
1016         dev_info->flow_type_rss_offloads = ~TAP_RSS_HF_MASK;
1017
1018         return 0;
1019 }
1020
1021 static int
1022 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
1023 {
1024         unsigned int i, imax;
1025         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
1026         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
1027         unsigned long rx_nombuf = 0, ierrors = 0;
1028         const struct pmd_internals *pmd = dev->data->dev_private;
1029
1030         /* rx queue statistics */
1031         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
1032                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
1033         for (i = 0; i < imax; i++) {
1034                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
1035                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
1036                 rx_total += tap_stats->q_ipackets[i];
1037                 rx_bytes_total += tap_stats->q_ibytes[i];
1038                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
1039                 ierrors += pmd->rxq[i].stats.ierrors;
1040         }
1041
1042         /* tx queue statistics */
1043         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
1044                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
1045
1046         for (i = 0; i < imax; i++) {
1047                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
1048                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
1049                 tx_total += tap_stats->q_opackets[i];
1050                 tx_err_total += pmd->txq[i].stats.errs;
1051                 tx_bytes_total += tap_stats->q_obytes[i];
1052         }
1053
1054         tap_stats->ipackets = rx_total;
1055         tap_stats->ibytes = rx_bytes_total;
1056         tap_stats->ierrors = ierrors;
1057         tap_stats->rx_nombuf = rx_nombuf;
1058         tap_stats->opackets = tx_total;
1059         tap_stats->oerrors = tx_err_total;
1060         tap_stats->obytes = tx_bytes_total;
1061         return 0;
1062 }
1063
1064 static int
1065 tap_stats_reset(struct rte_eth_dev *dev)
1066 {
1067         int i;
1068         struct pmd_internals *pmd = dev->data->dev_private;
1069
1070         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1071                 pmd->rxq[i].stats.ipackets = 0;
1072                 pmd->rxq[i].stats.ibytes = 0;
1073                 pmd->rxq[i].stats.ierrors = 0;
1074                 pmd->rxq[i].stats.rx_nombuf = 0;
1075
1076                 pmd->txq[i].stats.opackets = 0;
1077                 pmd->txq[i].stats.errs = 0;
1078                 pmd->txq[i].stats.obytes = 0;
1079         }
1080
1081         return 0;
1082 }
1083
1084 static int
1085 tap_dev_close(struct rte_eth_dev *dev)
1086 {
1087         int i;
1088         struct pmd_internals *internals = dev->data->dev_private;
1089         struct pmd_process_private *process_private = dev->process_private;
1090         struct rx_queue *rxq;
1091
1092         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1093                 rte_free(dev->process_private);
1094                 return 0;
1095         }
1096
1097         tap_link_set_down(dev);
1098         if (internals->nlsk_fd != -1) {
1099                 tap_flow_flush(dev, NULL);
1100                 tap_flow_implicit_flush(internals, NULL);
1101                 tap_nl_final(internals->nlsk_fd);
1102                 internals->nlsk_fd = -1;
1103         }
1104
1105         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1106                 if (process_private->rxq_fds[i] != -1) {
1107                         rxq = &internals->rxq[i];
1108                         close(process_private->rxq_fds[i]);
1109                         process_private->rxq_fds[i] = -1;
1110                         tap_rxq_pool_free(rxq->pool);
1111                         rte_free(rxq->iovecs);
1112                         rxq->pool = NULL;
1113                         rxq->iovecs = NULL;
1114                 }
1115                 if (process_private->txq_fds[i] != -1) {
1116                         close(process_private->txq_fds[i]);
1117                         process_private->txq_fds[i] = -1;
1118                 }
1119         }
1120
1121         if (internals->remote_if_index) {
1122                 /* Restore initial remote state */
1123                 int ret = ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
1124                                 &internals->remote_initial_flags);
1125                 if (ret)
1126                         TAP_LOG(ERR, "restore remote state failed: %d", ret);
1127
1128         }
1129
1130         rte_mempool_free(internals->gso_ctx_mp);
1131         internals->gso_ctx_mp = NULL;
1132
1133         if (internals->ka_fd != -1) {
1134                 close(internals->ka_fd);
1135                 internals->ka_fd = -1;
1136         }
1137
1138         /* mac_addrs must not be freed alone because part of dev_private */
1139         dev->data->mac_addrs = NULL;
1140
1141         internals = dev->data->dev_private;
1142         TAP_LOG(DEBUG, "Closing %s Ethernet device on numa %u",
1143                 tuntap_types[internals->type], rte_socket_id());
1144
1145         if (internals->ioctl_sock != -1) {
1146                 close(internals->ioctl_sock);
1147                 internals->ioctl_sock = -1;
1148         }
1149         rte_free(dev->process_private);
1150         if (tap_devices_count == 1)
1151                 rte_mp_action_unregister(TAP_MP_KEY);
1152         tap_devices_count--;
1153         /*
1154          * Since TUN device has no more opened file descriptors
1155          * it will be removed from kernel
1156          */
1157
1158         return 0;
1159 }
1160
1161 static void
1162 tap_rx_queue_release(void *queue)
1163 {
1164         struct rx_queue *rxq = queue;
1165         struct pmd_process_private *process_private;
1166
1167         if (!rxq)
1168                 return;
1169         process_private = rte_eth_devices[rxq->in_port].process_private;
1170         if (process_private->rxq_fds[rxq->queue_id] != -1) {
1171                 close(process_private->rxq_fds[rxq->queue_id]);
1172                 process_private->rxq_fds[rxq->queue_id] = -1;
1173                 tap_rxq_pool_free(rxq->pool);
1174                 rte_free(rxq->iovecs);
1175                 rxq->pool = NULL;
1176                 rxq->iovecs = NULL;
1177         }
1178 }
1179
1180 static void
1181 tap_tx_queue_release(void *queue)
1182 {
1183         struct tx_queue *txq = queue;
1184         struct pmd_process_private *process_private;
1185
1186         if (!txq)
1187                 return;
1188         process_private = rte_eth_devices[txq->out_port].process_private;
1189
1190         if (process_private->txq_fds[txq->queue_id] != -1) {
1191                 close(process_private->txq_fds[txq->queue_id]);
1192                 process_private->txq_fds[txq->queue_id] = -1;
1193         }
1194 }
1195
1196 static int
1197 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1198 {
1199         struct rte_eth_link *dev_link = &dev->data->dev_link;
1200         struct pmd_internals *pmd = dev->data->dev_private;
1201         struct ifreq ifr = { .ifr_flags = 0 };
1202
1203         if (pmd->remote_if_index) {
1204                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
1205                 if (!(ifr.ifr_flags & IFF_UP) ||
1206                     !(ifr.ifr_flags & IFF_RUNNING)) {
1207                         dev_link->link_status = ETH_LINK_DOWN;
1208                         return 0;
1209                 }
1210         }
1211         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
1212         dev_link->link_status =
1213                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
1214                  ETH_LINK_UP :
1215                  ETH_LINK_DOWN);
1216         return 0;
1217 }
1218
1219 static int
1220 tap_promisc_enable(struct rte_eth_dev *dev)
1221 {
1222         struct pmd_internals *pmd = dev->data->dev_private;
1223         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1224         int ret;
1225
1226         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1227         if (ret != 0)
1228                 return ret;
1229
1230         if (pmd->remote_if_index && !pmd->flow_isolate) {
1231                 dev->data->promiscuous = 1;
1232                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
1233                 if (ret != 0) {
1234                         /* Rollback promisc flag */
1235                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1236                         /*
1237                          * rte_eth_dev_promiscuous_enable() rollback
1238                          * dev->data->promiscuous in the case of failure.
1239                          */
1240                         return ret;
1241                 }
1242         }
1243
1244         return 0;
1245 }
1246
1247 static int
1248 tap_promisc_disable(struct rte_eth_dev *dev)
1249 {
1250         struct pmd_internals *pmd = dev->data->dev_private;
1251         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1252         int ret;
1253
1254         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1255         if (ret != 0)
1256                 return ret;
1257
1258         if (pmd->remote_if_index && !pmd->flow_isolate) {
1259                 dev->data->promiscuous = 0;
1260                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
1261                 if (ret != 0) {
1262                         /* Rollback promisc flag */
1263                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1264                         /*
1265                          * rte_eth_dev_promiscuous_disable() rollback
1266                          * dev->data->promiscuous in the case of failure.
1267                          */
1268                         return ret;
1269                 }
1270         }
1271
1272         return 0;
1273 }
1274
1275 static int
1276 tap_allmulti_enable(struct rte_eth_dev *dev)
1277 {
1278         struct pmd_internals *pmd = dev->data->dev_private;
1279         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1280         int ret;
1281
1282         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1283         if (ret != 0)
1284                 return ret;
1285
1286         if (pmd->remote_if_index && !pmd->flow_isolate) {
1287                 dev->data->all_multicast = 1;
1288                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
1289                 if (ret != 0) {
1290                         /* Rollback allmulti flag */
1291                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1292                         /*
1293                          * rte_eth_dev_allmulticast_enable() rollback
1294                          * dev->data->all_multicast in the case of failure.
1295                          */
1296                         return ret;
1297                 }
1298         }
1299
1300         return 0;
1301 }
1302
1303 static int
1304 tap_allmulti_disable(struct rte_eth_dev *dev)
1305 {
1306         struct pmd_internals *pmd = dev->data->dev_private;
1307         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1308         int ret;
1309
1310         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1311         if (ret != 0)
1312                 return ret;
1313
1314         if (pmd->remote_if_index && !pmd->flow_isolate) {
1315                 dev->data->all_multicast = 0;
1316                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
1317                 if (ret != 0) {
1318                         /* Rollback allmulti flag */
1319                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1320                         /*
1321                          * rte_eth_dev_allmulticast_disable() rollback
1322                          * dev->data->all_multicast in the case of failure.
1323                          */
1324                         return ret;
1325                 }
1326         }
1327
1328         return 0;
1329 }
1330
1331 static int
1332 tap_mac_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1333 {
1334         struct pmd_internals *pmd = dev->data->dev_private;
1335         enum ioctl_mode mode = LOCAL_ONLY;
1336         struct ifreq ifr;
1337         int ret;
1338
1339         if (pmd->type == ETH_TUNTAP_TYPE_TUN) {
1340                 TAP_LOG(ERR, "%s: can't MAC address for TUN",
1341                         dev->device->name);
1342                 return -ENOTSUP;
1343         }
1344
1345         if (rte_is_zero_ether_addr(mac_addr)) {
1346                 TAP_LOG(ERR, "%s: can't set an empty MAC address",
1347                         dev->device->name);
1348                 return -EINVAL;
1349         }
1350         /* Check the actual current MAC address on the tap netdevice */
1351         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
1352         if (ret < 0)
1353                 return ret;
1354         if (rte_is_same_ether_addr(
1355                         (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1356                         mac_addr))
1357                 return 0;
1358         /* Check the current MAC address on the remote */
1359         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
1360         if (ret < 0)
1361                 return ret;
1362         if (!rte_is_same_ether_addr(
1363                         (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1364                         mac_addr))
1365                 mode = LOCAL_AND_REMOTE;
1366         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1367         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, RTE_ETHER_ADDR_LEN);
1368         ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
1369         if (ret < 0)
1370                 return ret;
1371         rte_memcpy(&pmd->eth_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1372         if (pmd->remote_if_index && !pmd->flow_isolate) {
1373                 /* Replace MAC redirection rule after a MAC change */
1374                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
1375                 if (ret < 0) {
1376                         TAP_LOG(ERR,
1377                                 "%s: Couldn't delete MAC redirection rule",
1378                                 dev->device->name);
1379                         return ret;
1380                 }
1381                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1382                 if (ret < 0) {
1383                         TAP_LOG(ERR,
1384                                 "%s: Couldn't add MAC redirection rule",
1385                                 dev->device->name);
1386                         return ret;
1387                 }
1388         }
1389
1390         return 0;
1391 }
1392
1393 static int
1394 tap_gso_ctx_setup(struct rte_gso_ctx *gso_ctx, struct rte_eth_dev *dev)
1395 {
1396         uint32_t gso_types;
1397         char pool_name[64];
1398         struct pmd_internals *pmd = dev->data->dev_private;
1399         int ret;
1400
1401         /* initialize GSO context */
1402         gso_types = DEV_TX_OFFLOAD_TCP_TSO;
1403         if (!pmd->gso_ctx_mp) {
1404                 /*
1405                  * Create private mbuf pool with TAP_GSO_MBUF_SEG_SIZE
1406                  * bytes size per mbuf use this pool for both direct and
1407                  * indirect mbufs
1408                  */
1409                 ret = snprintf(pool_name, sizeof(pool_name), "mp_%s",
1410                                 dev->device->name);
1411                 if (ret < 0 || ret >= (int)sizeof(pool_name)) {
1412                         TAP_LOG(ERR,
1413                                 "%s: failed to create mbuf pool name for device %s,"
1414                                 "device name too long or output error, ret: %d\n",
1415                                 pmd->name, dev->device->name, ret);
1416                         return -ENAMETOOLONG;
1417                 }
1418                 pmd->gso_ctx_mp = rte_pktmbuf_pool_create(pool_name,
1419                         TAP_GSO_MBUFS_NUM, TAP_GSO_MBUF_CACHE_SIZE, 0,
1420                         RTE_PKTMBUF_HEADROOM + TAP_GSO_MBUF_SEG_SIZE,
1421                         SOCKET_ID_ANY);
1422                 if (!pmd->gso_ctx_mp) {
1423                         TAP_LOG(ERR,
1424                                 "%s: failed to create mbuf pool for device %s\n",
1425                                 pmd->name, dev->device->name);
1426                         return -1;
1427                 }
1428         }
1429
1430         gso_ctx->direct_pool = pmd->gso_ctx_mp;
1431         gso_ctx->indirect_pool = pmd->gso_ctx_mp;
1432         gso_ctx->gso_types = gso_types;
1433         gso_ctx->gso_size = 0; /* gso_size is set in tx_burst() per packet */
1434         gso_ctx->flag = 0;
1435
1436         return 0;
1437 }
1438
1439 static int
1440 tap_setup_queue(struct rte_eth_dev *dev,
1441                 struct pmd_internals *internals,
1442                 uint16_t qid,
1443                 int is_rx)
1444 {
1445         int ret;
1446         int *fd;
1447         int *other_fd;
1448         const char *dir;
1449         struct pmd_internals *pmd = dev->data->dev_private;
1450         struct pmd_process_private *process_private = dev->process_private;
1451         struct rx_queue *rx = &internals->rxq[qid];
1452         struct tx_queue *tx = &internals->txq[qid];
1453         struct rte_gso_ctx *gso_ctx;
1454
1455         if (is_rx) {
1456                 fd = &process_private->rxq_fds[qid];
1457                 other_fd = &process_private->txq_fds[qid];
1458                 dir = "rx";
1459                 gso_ctx = NULL;
1460         } else {
1461                 fd = &process_private->txq_fds[qid];
1462                 other_fd = &process_private->rxq_fds[qid];
1463                 dir = "tx";
1464                 gso_ctx = &tx->gso_ctx;
1465         }
1466         if (*fd != -1) {
1467                 /* fd for this queue already exists */
1468                 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1469                         pmd->name, *fd, dir, qid);
1470                 gso_ctx = NULL;
1471         } else if (*other_fd != -1) {
1472                 /* Only other_fd exists. dup it */
1473                 *fd = dup(*other_fd);
1474                 if (*fd < 0) {
1475                         *fd = -1;
1476                         TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1477                         return -1;
1478                 }
1479                 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1480                         pmd->name, *other_fd, dir, qid, *fd);
1481         } else {
1482                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1483                 *fd = tun_alloc(pmd, 0);
1484                 if (*fd < 0) {
1485                         *fd = -1; /* restore original value */
1486                         TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1487                         return -1;
1488                 }
1489                 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1490                         pmd->name, dir, qid, *fd);
1491         }
1492
1493         tx->mtu = &dev->data->mtu;
1494         rx->rxmode = &dev->data->dev_conf.rxmode;
1495         if (gso_ctx) {
1496                 ret = tap_gso_ctx_setup(gso_ctx, dev);
1497                 if (ret)
1498                         return -1;
1499         }
1500
1501         tx->type = pmd->type;
1502
1503         return *fd;
1504 }
1505
1506 static int
1507 tap_rx_queue_setup(struct rte_eth_dev *dev,
1508                    uint16_t rx_queue_id,
1509                    uint16_t nb_rx_desc,
1510                    unsigned int socket_id,
1511                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1512                    struct rte_mempool *mp)
1513 {
1514         struct pmd_internals *internals = dev->data->dev_private;
1515         struct pmd_process_private *process_private = dev->process_private;
1516         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1517         struct rte_mbuf **tmp = &rxq->pool;
1518         long iov_max = sysconf(_SC_IOV_MAX);
1519
1520         if (iov_max <= 0) {
1521                 TAP_LOG(WARNING,
1522                         "_SC_IOV_MAX is not defined. Using %d as default",
1523                         TAP_IOV_DEFAULT_MAX);
1524                 iov_max = TAP_IOV_DEFAULT_MAX;
1525         }
1526         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1527         struct iovec (*iovecs)[nb_desc + 1];
1528         int data_off = RTE_PKTMBUF_HEADROOM;
1529         int ret = 0;
1530         int fd;
1531         int i;
1532
1533         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1534                 TAP_LOG(WARNING,
1535                         "nb_rx_queues %d too small or mempool NULL",
1536                         dev->data->nb_rx_queues);
1537                 return -1;
1538         }
1539
1540         rxq->mp = mp;
1541         rxq->trigger_seen = 1; /* force initial burst */
1542         rxq->in_port = dev->data->port_id;
1543         rxq->queue_id = rx_queue_id;
1544         rxq->nb_rx_desc = nb_desc;
1545         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1546                                     socket_id);
1547         if (!iovecs) {
1548                 TAP_LOG(WARNING,
1549                         "%s: Couldn't allocate %d RX descriptors",
1550                         dev->device->name, nb_desc);
1551                 return -ENOMEM;
1552         }
1553         rxq->iovecs = iovecs;
1554
1555         dev->data->rx_queues[rx_queue_id] = rxq;
1556         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1557         if (fd == -1) {
1558                 ret = fd;
1559                 goto error;
1560         }
1561
1562         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1563         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1564
1565         for (i = 1; i <= nb_desc; i++) {
1566                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1567                 if (!*tmp) {
1568                         TAP_LOG(WARNING,
1569                                 "%s: couldn't allocate memory for queue %d",
1570                                 dev->device->name, rx_queue_id);
1571                         ret = -ENOMEM;
1572                         goto error;
1573                 }
1574                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1575                 (*rxq->iovecs)[i].iov_base =
1576                         (char *)(*tmp)->buf_addr + data_off;
1577                 data_off = 0;
1578                 tmp = &(*tmp)->next;
1579         }
1580
1581         TAP_LOG(DEBUG, "  RX TUNTAP device name %s, qid %d on fd %d",
1582                 internals->name, rx_queue_id,
1583                 process_private->rxq_fds[rx_queue_id]);
1584
1585         return 0;
1586
1587 error:
1588         tap_rxq_pool_free(rxq->pool);
1589         rxq->pool = NULL;
1590         rte_free(rxq->iovecs);
1591         rxq->iovecs = NULL;
1592         return ret;
1593 }
1594
1595 static int
1596 tap_tx_queue_setup(struct rte_eth_dev *dev,
1597                    uint16_t tx_queue_id,
1598                    uint16_t nb_tx_desc __rte_unused,
1599                    unsigned int socket_id __rte_unused,
1600                    const struct rte_eth_txconf *tx_conf)
1601 {
1602         struct pmd_internals *internals = dev->data->dev_private;
1603         struct pmd_process_private *process_private = dev->process_private;
1604         struct tx_queue *txq;
1605         int ret;
1606         uint64_t offloads;
1607
1608         if (tx_queue_id >= dev->data->nb_tx_queues)
1609                 return -1;
1610         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1611         txq = dev->data->tx_queues[tx_queue_id];
1612         txq->out_port = dev->data->port_id;
1613         txq->queue_id = tx_queue_id;
1614
1615         offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1616         txq->csum = !!(offloads &
1617                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
1618                          DEV_TX_OFFLOAD_UDP_CKSUM |
1619                          DEV_TX_OFFLOAD_TCP_CKSUM));
1620
1621         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1622         if (ret == -1)
1623                 return -1;
1624         TAP_LOG(DEBUG,
1625                 "  TX TUNTAP device name %s, qid %d on fd %d csum %s",
1626                 internals->name, tx_queue_id,
1627                 process_private->txq_fds[tx_queue_id],
1628                 txq->csum ? "on" : "off");
1629
1630         return 0;
1631 }
1632
1633 static int
1634 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1635 {
1636         struct pmd_internals *pmd = dev->data->dev_private;
1637         struct ifreq ifr = { .ifr_mtu = mtu };
1638         int err = 0;
1639
1640         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1641         if (!err)
1642                 dev->data->mtu = mtu;
1643
1644         return err;
1645 }
1646
1647 static int
1648 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1649                      struct rte_ether_addr *mc_addr_set __rte_unused,
1650                      uint32_t nb_mc_addr __rte_unused)
1651 {
1652         /*
1653          * Nothing to do actually: the tap has no filtering whatsoever, every
1654          * packet is received.
1655          */
1656         return 0;
1657 }
1658
1659 static int
1660 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1661 {
1662         struct rte_eth_dev *dev = arg;
1663         struct pmd_internals *pmd = dev->data->dev_private;
1664         struct ifinfomsg *info = NLMSG_DATA(nh);
1665
1666         if (nh->nlmsg_type != RTM_NEWLINK ||
1667             (info->ifi_index != pmd->if_index &&
1668              info->ifi_index != pmd->remote_if_index))
1669                 return 0;
1670         return tap_link_update(dev, 0);
1671 }
1672
1673 static void
1674 tap_dev_intr_handler(void *cb_arg)
1675 {
1676         struct rte_eth_dev *dev = cb_arg;
1677         struct pmd_internals *pmd = dev->data->dev_private;
1678
1679         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1680 }
1681
1682 static int
1683 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1684 {
1685         struct pmd_internals *pmd = dev->data->dev_private;
1686         int ret;
1687
1688         /* In any case, disable interrupt if the conf is no longer there. */
1689         if (!dev->data->dev_conf.intr_conf.lsc) {
1690                 if (pmd->intr_handle.fd != -1) {
1691                         goto clean;
1692                 }
1693                 return 0;
1694         }
1695         if (set) {
1696                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1697                 if (unlikely(pmd->intr_handle.fd == -1))
1698                         return -EBADF;
1699                 return rte_intr_callback_register(
1700                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1701         }
1702
1703 clean:
1704         do {
1705                 ret = rte_intr_callback_unregister(&pmd->intr_handle,
1706                         tap_dev_intr_handler, dev);
1707                 if (ret >= 0) {
1708                         break;
1709                 } else if (ret == -EAGAIN) {
1710                         rte_delay_ms(100);
1711                 } else {
1712                         TAP_LOG(ERR, "intr callback unregister failed: %d",
1713                                      ret);
1714                         break;
1715                 }
1716         } while (true);
1717
1718         tap_nl_final(pmd->intr_handle.fd);
1719         pmd->intr_handle.fd = -1;
1720
1721         return 0;
1722 }
1723
1724 static int
1725 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1726 {
1727         int err;
1728
1729         err = tap_lsc_intr_handle_set(dev, set);
1730         if (err < 0) {
1731                 if (!set)
1732                         tap_rx_intr_vec_set(dev, 0);
1733                 return err;
1734         }
1735         err = tap_rx_intr_vec_set(dev, set);
1736         if (err && set)
1737                 tap_lsc_intr_handle_set(dev, 0);
1738         return err;
1739 }
1740
1741 static const uint32_t*
1742 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1743 {
1744         static const uint32_t ptypes[] = {
1745                 RTE_PTYPE_INNER_L2_ETHER,
1746                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1747                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1748                 RTE_PTYPE_INNER_L3_IPV4,
1749                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1750                 RTE_PTYPE_INNER_L3_IPV6,
1751                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1752                 RTE_PTYPE_INNER_L4_FRAG,
1753                 RTE_PTYPE_INNER_L4_UDP,
1754                 RTE_PTYPE_INNER_L4_TCP,
1755                 RTE_PTYPE_INNER_L4_SCTP,
1756                 RTE_PTYPE_L2_ETHER,
1757                 RTE_PTYPE_L2_ETHER_VLAN,
1758                 RTE_PTYPE_L2_ETHER_QINQ,
1759                 RTE_PTYPE_L3_IPV4,
1760                 RTE_PTYPE_L3_IPV4_EXT,
1761                 RTE_PTYPE_L3_IPV6_EXT,
1762                 RTE_PTYPE_L3_IPV6,
1763                 RTE_PTYPE_L4_FRAG,
1764                 RTE_PTYPE_L4_UDP,
1765                 RTE_PTYPE_L4_TCP,
1766                 RTE_PTYPE_L4_SCTP,
1767         };
1768
1769         return ptypes;
1770 }
1771
1772 static int
1773 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1774                   struct rte_eth_fc_conf *fc_conf)
1775 {
1776         fc_conf->mode = RTE_FC_NONE;
1777         return 0;
1778 }
1779
1780 static int
1781 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1782                   struct rte_eth_fc_conf *fc_conf)
1783 {
1784         if (fc_conf->mode != RTE_FC_NONE)
1785                 return -ENOTSUP;
1786         return 0;
1787 }
1788
1789 /**
1790  * DPDK callback to update the RSS hash configuration.
1791  *
1792  * @param dev
1793  *   Pointer to Ethernet device structure.
1794  * @param[in] rss_conf
1795  *   RSS configuration data.
1796  *
1797  * @return
1798  *   0 on success, a negative errno value otherwise and rte_errno is set.
1799  */
1800 static int
1801 tap_rss_hash_update(struct rte_eth_dev *dev,
1802                 struct rte_eth_rss_conf *rss_conf)
1803 {
1804         if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1805                 rte_errno = EINVAL;
1806                 return -rte_errno;
1807         }
1808         if (rss_conf->rss_key && rss_conf->rss_key_len) {
1809                 /*
1810                  * Currently TAP RSS key is hard coded
1811                  * and cannot be updated
1812                  */
1813                 TAP_LOG(ERR,
1814                         "port %u RSS key cannot be updated",
1815                         dev->data->port_id);
1816                 rte_errno = EINVAL;
1817                 return -rte_errno;
1818         }
1819         return 0;
1820 }
1821
1822 static int
1823 tap_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1824 {
1825         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1826
1827         return 0;
1828 }
1829
1830 static int
1831 tap_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1832 {
1833         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1834
1835         return 0;
1836 }
1837
1838 static int
1839 tap_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1840 {
1841         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1842
1843         return 0;
1844 }
1845
1846 static int
1847 tap_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1848 {
1849         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1850
1851         return 0;
1852 }
1853 static const struct eth_dev_ops ops = {
1854         .dev_start              = tap_dev_start,
1855         .dev_stop               = tap_dev_stop,
1856         .dev_close              = tap_dev_close,
1857         .dev_configure          = tap_dev_configure,
1858         .dev_infos_get          = tap_dev_info,
1859         .rx_queue_setup         = tap_rx_queue_setup,
1860         .tx_queue_setup         = tap_tx_queue_setup,
1861         .rx_queue_start         = tap_rx_queue_start,
1862         .tx_queue_start         = tap_tx_queue_start,
1863         .rx_queue_stop          = tap_rx_queue_stop,
1864         .tx_queue_stop          = tap_tx_queue_stop,
1865         .rx_queue_release       = tap_rx_queue_release,
1866         .tx_queue_release       = tap_tx_queue_release,
1867         .flow_ctrl_get          = tap_flow_ctrl_get,
1868         .flow_ctrl_set          = tap_flow_ctrl_set,
1869         .link_update            = tap_link_update,
1870         .dev_set_link_up        = tap_link_set_up,
1871         .dev_set_link_down      = tap_link_set_down,
1872         .promiscuous_enable     = tap_promisc_enable,
1873         .promiscuous_disable    = tap_promisc_disable,
1874         .allmulticast_enable    = tap_allmulti_enable,
1875         .allmulticast_disable   = tap_allmulti_disable,
1876         .mac_addr_set           = tap_mac_set,
1877         .mtu_set                = tap_mtu_set,
1878         .set_mc_addr_list       = tap_set_mc_addr_list,
1879         .stats_get              = tap_stats_get,
1880         .stats_reset            = tap_stats_reset,
1881         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1882         .rss_hash_update        = tap_rss_hash_update,
1883         .flow_ops_get           = tap_dev_flow_ops_get,
1884 };
1885
1886 static int
1887 eth_dev_tap_create(struct rte_vdev_device *vdev, const char *tap_name,
1888                    char *remote_iface, struct rte_ether_addr *mac_addr,
1889                    enum rte_tuntap_type type)
1890 {
1891         int numa_node = rte_socket_id();
1892         struct rte_eth_dev *dev;
1893         struct pmd_internals *pmd;
1894         struct pmd_process_private *process_private;
1895         const char *tuntap_name = tuntap_types[type];
1896         struct rte_eth_dev_data *data;
1897         struct ifreq ifr;
1898         int i;
1899
1900         TAP_LOG(DEBUG, "%s device on numa %u", tuntap_name, rte_socket_id());
1901
1902         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1903         if (!dev) {
1904                 TAP_LOG(ERR, "%s Unable to allocate device struct",
1905                                 tuntap_name);
1906                 goto error_exit_nodev;
1907         }
1908
1909         process_private = (struct pmd_process_private *)
1910                 rte_zmalloc_socket(tap_name, sizeof(struct pmd_process_private),
1911                         RTE_CACHE_LINE_SIZE, dev->device->numa_node);
1912
1913         if (process_private == NULL) {
1914                 TAP_LOG(ERR, "Failed to alloc memory for process private");
1915                 return -1;
1916         }
1917         pmd = dev->data->dev_private;
1918         dev->process_private = process_private;
1919         pmd->dev = dev;
1920         strlcpy(pmd->name, tap_name, sizeof(pmd->name));
1921         pmd->type = type;
1922         pmd->ka_fd = -1;
1923         pmd->nlsk_fd = -1;
1924         pmd->gso_ctx_mp = NULL;
1925
1926         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1927         if (pmd->ioctl_sock == -1) {
1928                 TAP_LOG(ERR,
1929                         "%s Unable to get a socket for management: %s",
1930                         tuntap_name, strerror(errno));
1931                 goto error_exit;
1932         }
1933
1934         /* Setup some default values */
1935         data = dev->data;
1936         data->dev_private = pmd;
1937         data->dev_flags = RTE_ETH_DEV_INTR_LSC |
1938                                 RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1939         data->numa_node = numa_node;
1940
1941         data->dev_link = pmd_link;
1942         data->mac_addrs = &pmd->eth_addr;
1943         /* Set the number of RX and TX queues */
1944         data->nb_rx_queues = 0;
1945         data->nb_tx_queues = 0;
1946
1947         dev->dev_ops = &ops;
1948         dev->rx_pkt_burst = pmd_rx_burst;
1949         dev->tx_pkt_burst = pmd_tx_burst;
1950
1951         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1952         pmd->intr_handle.fd = -1;
1953         dev->intr_handle = &pmd->intr_handle;
1954
1955         /* Presetup the fds to -1 as being not valid */
1956         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1957                 process_private->rxq_fds[i] = -1;
1958                 process_private->txq_fds[i] = -1;
1959         }
1960
1961         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1962                 if (rte_is_zero_ether_addr(mac_addr))
1963                         rte_eth_random_addr((uint8_t *)&pmd->eth_addr);
1964                 else
1965                         rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1966         }
1967
1968         /*
1969          * Allocate a TUN device keep-alive file descriptor that will only be
1970          * closed when the TUN device itself is closed or removed.
1971          * This keep-alive file descriptor will guarantee that the TUN device
1972          * exists even when all of its queues are closed
1973          */
1974         pmd->ka_fd = tun_alloc(pmd, 1);
1975         if (pmd->ka_fd == -1) {
1976                 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
1977                 goto error_exit;
1978         }
1979         TAP_LOG(DEBUG, "allocated %s", pmd->name);
1980
1981         ifr.ifr_mtu = dev->data->mtu;
1982         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1983                 goto error_exit;
1984
1985         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1986                 memset(&ifr, 0, sizeof(struct ifreq));
1987                 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1988                 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1989                                 RTE_ETHER_ADDR_LEN);
1990                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1991                         goto error_exit;
1992         }
1993
1994         /*
1995          * Set up everything related to rte_flow:
1996          * - netlink socket
1997          * - tap / remote if_index
1998          * - mandatory QDISCs
1999          * - rte_flow actual/implicit lists
2000          * - implicit rules
2001          */
2002         pmd->nlsk_fd = tap_nl_init(0);
2003         if (pmd->nlsk_fd == -1) {
2004                 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
2005                         pmd->name);
2006                 goto disable_rte_flow;
2007         }
2008         pmd->if_index = if_nametoindex(pmd->name);
2009         if (!pmd->if_index) {
2010                 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
2011                 goto disable_rte_flow;
2012         }
2013         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
2014                 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
2015                         pmd->name);
2016                 goto disable_rte_flow;
2017         }
2018         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
2019                 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
2020                         pmd->name);
2021                 goto disable_rte_flow;
2022         }
2023         LIST_INIT(&pmd->flows);
2024
2025         if (strlen(remote_iface)) {
2026                 pmd->remote_if_index = if_nametoindex(remote_iface);
2027                 if (!pmd->remote_if_index) {
2028                         TAP_LOG(ERR, "%s: failed to get %s if_index.",
2029                                 pmd->name, remote_iface);
2030                         goto error_remote;
2031                 }
2032                 strlcpy(pmd->remote_iface, remote_iface, RTE_ETH_NAME_MAX_LEN);
2033
2034                 /* Save state of remote device */
2035                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
2036
2037                 /* Replicate remote MAC address */
2038                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
2039                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
2040                                 pmd->name, pmd->remote_iface);
2041                         goto error_remote;
2042                 }
2043                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
2044                            RTE_ETHER_ADDR_LEN);
2045                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
2046                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
2047                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
2048                                 pmd->name, remote_iface);
2049                         goto error_remote;
2050                 }
2051
2052                 /*
2053                  * Flush usually returns negative value because it tries to
2054                  * delete every QDISC (and on a running device, one QDISC at
2055                  * least is needed). Ignore negative return value.
2056                  */
2057                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
2058                 if (qdisc_create_ingress(pmd->nlsk_fd,
2059                                          pmd->remote_if_index) < 0) {
2060                         TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
2061                                 pmd->remote_iface);
2062                         goto error_remote;
2063                 }
2064                 LIST_INIT(&pmd->implicit_flows);
2065                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
2066                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
2067                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
2068                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
2069                         TAP_LOG(ERR,
2070                                 "%s: failed to create implicit rules.",
2071                                 pmd->name);
2072                         goto error_remote;
2073                 }
2074         }
2075
2076         rte_eth_dev_probing_finish(dev);
2077         return 0;
2078
2079 disable_rte_flow:
2080         TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
2081                 strerror(errno), errno);
2082         if (strlen(remote_iface)) {
2083                 TAP_LOG(ERR, "Remote feature requires flow support.");
2084                 goto error_exit;
2085         }
2086         rte_eth_dev_probing_finish(dev);
2087         return 0;
2088
2089 error_remote:
2090         TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
2091                 strerror(errno), errno);
2092         tap_flow_implicit_flush(pmd, NULL);
2093
2094 error_exit:
2095         if (pmd->nlsk_fd != -1)
2096                 close(pmd->nlsk_fd);
2097         if (pmd->ka_fd != -1)
2098                 close(pmd->ka_fd);
2099         if (pmd->ioctl_sock != -1)
2100                 close(pmd->ioctl_sock);
2101         /* mac_addrs must not be freed alone because part of dev_private */
2102         dev->data->mac_addrs = NULL;
2103         rte_eth_dev_release_port(dev);
2104
2105 error_exit_nodev:
2106         TAP_LOG(ERR, "%s Unable to initialize %s",
2107                 tuntap_name, rte_vdev_device_name(vdev));
2108
2109         return -EINVAL;
2110 }
2111
2112 /* make sure name is a possible Linux network device name */
2113 static bool
2114 is_valid_iface(const char *name)
2115 {
2116         if (*name == '\0')
2117                 return false;
2118
2119         if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
2120                 return false;
2121
2122         while (*name) {
2123                 if (*name == '/' || *name == ':' || isspace(*name))
2124                         return false;
2125                 name++;
2126         }
2127         return true;
2128 }
2129
2130 static int
2131 set_interface_name(const char *key __rte_unused,
2132                    const char *value,
2133                    void *extra_args)
2134 {
2135         char *name = (char *)extra_args;
2136
2137         if (value) {
2138                 if (!is_valid_iface(value)) {
2139                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2140                                 value);
2141                         return -1;
2142                 }
2143                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2144         } else {
2145                 /* use tap%d which causes kernel to choose next available */
2146                 strlcpy(name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2147         }
2148         return 0;
2149 }
2150
2151 static int
2152 set_remote_iface(const char *key __rte_unused,
2153                  const char *value,
2154                  void *extra_args)
2155 {
2156         char *name = (char *)extra_args;
2157
2158         if (value) {
2159                 if (!is_valid_iface(value)) {
2160                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2161                                 value);
2162                         return -1;
2163                 }
2164                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2165         }
2166
2167         return 0;
2168 }
2169
2170 static int parse_user_mac(struct rte_ether_addr *user_mac,
2171                 const char *value)
2172 {
2173         unsigned int index = 0;
2174         char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
2175
2176         if (user_mac == NULL || value == NULL)
2177                 return 0;
2178
2179         strlcpy(mac_temp, value, sizeof(mac_temp));
2180         mac_byte = strtok(mac_temp, ":");
2181
2182         while ((mac_byte != NULL) &&
2183                         (strlen(mac_byte) <= 2) &&
2184                         (strlen(mac_byte) == strspn(mac_byte,
2185                                         ETH_TAP_CMP_MAC_FMT))) {
2186                 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
2187                 mac_byte = strtok(NULL, ":");
2188         }
2189
2190         return index;
2191 }
2192
2193 static int
2194 set_mac_type(const char *key __rte_unused,
2195              const char *value,
2196              void *extra_args)
2197 {
2198         struct rte_ether_addr *user_mac = extra_args;
2199
2200         if (!value)
2201                 return 0;
2202
2203         if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
2204                 static int iface_idx;
2205
2206                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
2207                 memcpy((char *)user_mac->addr_bytes, "\0dtap",
2208                         RTE_ETHER_ADDR_LEN);
2209                 user_mac->addr_bytes[RTE_ETHER_ADDR_LEN - 1] =
2210                         iface_idx++ + '0';
2211                 goto success;
2212         }
2213
2214         if (parse_user_mac(user_mac, value) != 6)
2215                 goto error;
2216 success:
2217         TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
2218         return 0;
2219
2220 error:
2221         TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
2222                 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
2223         return -1;
2224 }
2225
2226 /*
2227  * Open a TUN interface device. TUN PMD
2228  * 1) sets tap_type as false
2229  * 2) intakes iface as argument.
2230  * 3) as interface is virtual set speed to 10G
2231  */
2232 static int
2233 rte_pmd_tun_probe(struct rte_vdev_device *dev)
2234 {
2235         const char *name, *params;
2236         int ret;
2237         struct rte_kvargs *kvlist = NULL;
2238         char tun_name[RTE_ETH_NAME_MAX_LEN];
2239         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2240         struct rte_eth_dev *eth_dev;
2241
2242         name = rte_vdev_device_name(dev);
2243         params = rte_vdev_device_args(dev);
2244         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2245
2246         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
2247             strlen(params) == 0) {
2248                 eth_dev = rte_eth_dev_attach_secondary(name);
2249                 if (!eth_dev) {
2250                         TAP_LOG(ERR, "Failed to probe %s", name);
2251                         return -1;
2252                 }
2253                 eth_dev->dev_ops = &ops;
2254                 eth_dev->device = &dev->device;
2255                 rte_eth_dev_probing_finish(eth_dev);
2256                 return 0;
2257         }
2258
2259         /* use tun%d which causes kernel to choose next available */
2260         strlcpy(tun_name, DEFAULT_TUN_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2261
2262         if (params && (params[0] != '\0')) {
2263                 TAP_LOG(DEBUG, "parameters (%s)", params);
2264
2265                 kvlist = rte_kvargs_parse(params, valid_arguments);
2266                 if (kvlist) {
2267                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2268                                 ret = rte_kvargs_process(kvlist,
2269                                         ETH_TAP_IFACE_ARG,
2270                                         &set_interface_name,
2271                                         tun_name);
2272
2273                                 if (ret == -1)
2274                                         goto leave;
2275                         }
2276                 }
2277         }
2278         pmd_link.link_speed = ETH_SPEED_NUM_10G;
2279
2280         TAP_LOG(DEBUG, "Initializing pmd_tun for %s", name);
2281
2282         ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0,
2283                                  ETH_TUNTAP_TYPE_TUN);
2284
2285 leave:
2286         if (ret == -1) {
2287                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2288                         name, tun_name);
2289         }
2290         rte_kvargs_free(kvlist);
2291
2292         return ret;
2293 }
2294
2295 /* Request queue file descriptors from secondary to primary. */
2296 static int
2297 tap_mp_attach_queues(const char *port_name, struct rte_eth_dev *dev)
2298 {
2299         int ret;
2300         struct timespec timeout = {.tv_sec = 1, .tv_nsec = 0};
2301         struct rte_mp_msg request, *reply;
2302         struct rte_mp_reply replies;
2303         struct ipc_queues *request_param = (struct ipc_queues *)request.param;
2304         struct ipc_queues *reply_param;
2305         struct pmd_process_private *process_private = dev->process_private;
2306         int queue, fd_iterator;
2307
2308         /* Prepare the request */
2309         memset(&request, 0, sizeof(request));
2310         strlcpy(request.name, TAP_MP_KEY, sizeof(request.name));
2311         strlcpy(request_param->port_name, port_name,
2312                 sizeof(request_param->port_name));
2313         request.len_param = sizeof(*request_param);
2314         /* Send request and receive reply */
2315         ret = rte_mp_request_sync(&request, &replies, &timeout);
2316         if (ret < 0 || replies.nb_received != 1) {
2317                 TAP_LOG(ERR, "Failed to request queues from primary: %d",
2318                         rte_errno);
2319                 return -1;
2320         }
2321         reply = &replies.msgs[0];
2322         reply_param = (struct ipc_queues *)reply->param;
2323         TAP_LOG(DEBUG, "Received IPC reply for %s", reply_param->port_name);
2324
2325         /* Attach the queues from received file descriptors */
2326         if (reply_param->rxq_count + reply_param->txq_count != reply->num_fds) {
2327                 TAP_LOG(ERR, "Unexpected number of fds received");
2328                 return -1;
2329         }
2330
2331         dev->data->nb_rx_queues = reply_param->rxq_count;
2332         dev->data->nb_tx_queues = reply_param->txq_count;
2333         fd_iterator = 0;
2334         for (queue = 0; queue < reply_param->rxq_count; queue++)
2335                 process_private->rxq_fds[queue] = reply->fds[fd_iterator++];
2336         for (queue = 0; queue < reply_param->txq_count; queue++)
2337                 process_private->txq_fds[queue] = reply->fds[fd_iterator++];
2338         free(reply);
2339         return 0;
2340 }
2341
2342 /* Send the queue file descriptors from the primary process to secondary. */
2343 static int
2344 tap_mp_sync_queues(const struct rte_mp_msg *request, const void *peer)
2345 {
2346         struct rte_eth_dev *dev;
2347         struct pmd_process_private *process_private;
2348         struct rte_mp_msg reply;
2349         const struct ipc_queues *request_param =
2350                 (const struct ipc_queues *)request->param;
2351         struct ipc_queues *reply_param =
2352                 (struct ipc_queues *)reply.param;
2353         uint16_t port_id;
2354         int queue;
2355         int ret;
2356
2357         /* Get requested port */
2358         TAP_LOG(DEBUG, "Received IPC request for %s", request_param->port_name);
2359         ret = rte_eth_dev_get_port_by_name(request_param->port_name, &port_id);
2360         if (ret) {
2361                 TAP_LOG(ERR, "Failed to get port id for %s",
2362                         request_param->port_name);
2363                 return -1;
2364         }
2365         dev = &rte_eth_devices[port_id];
2366         process_private = dev->process_private;
2367
2368         /* Fill file descriptors for all queues */
2369         reply.num_fds = 0;
2370         reply_param->rxq_count = 0;
2371         if (dev->data->nb_rx_queues + dev->data->nb_tx_queues >
2372                         RTE_MP_MAX_FD_NUM){
2373                 TAP_LOG(ERR, "Number of rx/tx queues exceeds max number of fds");
2374                 return -1;
2375         }
2376
2377         for (queue = 0; queue < dev->data->nb_rx_queues; queue++) {
2378                 reply.fds[reply.num_fds++] = process_private->rxq_fds[queue];
2379                 reply_param->rxq_count++;
2380         }
2381         RTE_ASSERT(reply_param->rxq_count == dev->data->nb_rx_queues);
2382
2383         reply_param->txq_count = 0;
2384         for (queue = 0; queue < dev->data->nb_tx_queues; queue++) {
2385                 reply.fds[reply.num_fds++] = process_private->txq_fds[queue];
2386                 reply_param->txq_count++;
2387         }
2388         RTE_ASSERT(reply_param->txq_count == dev->data->nb_tx_queues);
2389
2390         /* Send reply */
2391         strlcpy(reply.name, request->name, sizeof(reply.name));
2392         strlcpy(reply_param->port_name, request_param->port_name,
2393                 sizeof(reply_param->port_name));
2394         reply.len_param = sizeof(*reply_param);
2395         if (rte_mp_reply(&reply, peer) < 0) {
2396                 TAP_LOG(ERR, "Failed to reply an IPC request to sync queues");
2397                 return -1;
2398         }
2399         return 0;
2400 }
2401
2402 /* Open a TAP interface device.
2403  */
2404 static int
2405 rte_pmd_tap_probe(struct rte_vdev_device *dev)
2406 {
2407         const char *name, *params;
2408         int ret;
2409         struct rte_kvargs *kvlist = NULL;
2410         int speed;
2411         char tap_name[RTE_ETH_NAME_MAX_LEN];
2412         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2413         struct rte_ether_addr user_mac = { .addr_bytes = {0} };
2414         struct rte_eth_dev *eth_dev;
2415         int tap_devices_count_increased = 0;
2416
2417         name = rte_vdev_device_name(dev);
2418         params = rte_vdev_device_args(dev);
2419
2420         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
2421                 eth_dev = rte_eth_dev_attach_secondary(name);
2422                 if (!eth_dev) {
2423                         TAP_LOG(ERR, "Failed to probe %s", name);
2424                         return -1;
2425                 }
2426                 eth_dev->dev_ops = &ops;
2427                 eth_dev->device = &dev->device;
2428                 eth_dev->rx_pkt_burst = pmd_rx_burst;
2429                 eth_dev->tx_pkt_burst = pmd_tx_burst;
2430                 if (!rte_eal_primary_proc_alive(NULL)) {
2431                         TAP_LOG(ERR, "Primary process is missing");
2432                         return -1;
2433                 }
2434                 eth_dev->process_private = (struct pmd_process_private *)
2435                         rte_zmalloc_socket(name,
2436                                 sizeof(struct pmd_process_private),
2437                                 RTE_CACHE_LINE_SIZE,
2438                                 eth_dev->device->numa_node);
2439                 if (eth_dev->process_private == NULL) {
2440                         TAP_LOG(ERR,
2441                                 "Failed to alloc memory for process private");
2442                         return -1;
2443                 }
2444
2445                 ret = tap_mp_attach_queues(name, eth_dev);
2446                 if (ret != 0)
2447                         return -1;
2448                 rte_eth_dev_probing_finish(eth_dev);
2449                 return 0;
2450         }
2451
2452         speed = ETH_SPEED_NUM_10G;
2453
2454         /* use tap%d which causes kernel to choose next available */
2455         strlcpy(tap_name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2456         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2457
2458         if (params && (params[0] != '\0')) {
2459                 TAP_LOG(DEBUG, "parameters (%s)", params);
2460
2461                 kvlist = rte_kvargs_parse(params, valid_arguments);
2462                 if (kvlist) {
2463                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2464                                 ret = rte_kvargs_process(kvlist,
2465                                                          ETH_TAP_IFACE_ARG,
2466                                                          &set_interface_name,
2467                                                          tap_name);
2468                                 if (ret == -1)
2469                                         goto leave;
2470                         }
2471
2472                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
2473                                 ret = rte_kvargs_process(kvlist,
2474                                                          ETH_TAP_REMOTE_ARG,
2475                                                          &set_remote_iface,
2476                                                          remote_iface);
2477                                 if (ret == -1)
2478                                         goto leave;
2479                         }
2480
2481                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
2482                                 ret = rte_kvargs_process(kvlist,
2483                                                          ETH_TAP_MAC_ARG,
2484                                                          &set_mac_type,
2485                                                          &user_mac);
2486                                 if (ret == -1)
2487                                         goto leave;
2488                         }
2489                 }
2490         }
2491         pmd_link.link_speed = speed;
2492
2493         TAP_LOG(DEBUG, "Initializing pmd_tap for %s", name);
2494
2495         /* Register IPC feed callback */
2496         if (!tap_devices_count) {
2497                 ret = rte_mp_action_register(TAP_MP_KEY, tap_mp_sync_queues);
2498                 if (ret < 0 && rte_errno != ENOTSUP) {
2499                         TAP_LOG(ERR, "tap: Failed to register IPC callback: %s",
2500                                 strerror(rte_errno));
2501                         goto leave;
2502                 }
2503         }
2504         tap_devices_count++;
2505         tap_devices_count_increased = 1;
2506         ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
2507                 ETH_TUNTAP_TYPE_TAP);
2508
2509 leave:
2510         if (ret == -1) {
2511                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2512                         name, tap_name);
2513                 if (tap_devices_count_increased == 1) {
2514                         if (tap_devices_count == 1)
2515                                 rte_mp_action_unregister(TAP_MP_KEY);
2516                         tap_devices_count--;
2517                 }
2518         }
2519         rte_kvargs_free(kvlist);
2520
2521         return ret;
2522 }
2523
2524 /* detach a TUNTAP device.
2525  */
2526 static int
2527 rte_pmd_tap_remove(struct rte_vdev_device *dev)
2528 {
2529         struct rte_eth_dev *eth_dev = NULL;
2530
2531         /* find the ethdev entry */
2532         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
2533         if (!eth_dev)
2534                 return 0;
2535
2536         tap_dev_close(eth_dev);
2537         rte_eth_dev_release_port(eth_dev);
2538
2539         return 0;
2540 }
2541
2542 static struct rte_vdev_driver pmd_tun_drv = {
2543         .probe = rte_pmd_tun_probe,
2544         .remove = rte_pmd_tap_remove,
2545 };
2546
2547 static struct rte_vdev_driver pmd_tap_drv = {
2548         .probe = rte_pmd_tap_probe,
2549         .remove = rte_pmd_tap_remove,
2550 };
2551
2552 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
2553 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
2554 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
2555 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
2556                               ETH_TAP_IFACE_ARG "=<string> ");
2557 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
2558                               ETH_TAP_IFACE_ARG "=<string> "
2559                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
2560                               ETH_TAP_REMOTE_ARG "=<string>");
2561 RTE_LOG_REGISTER_DEFAULT(tap_logtype, NOTICE);