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                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 354                 /* Don't verify checksum for multi-segment packets. */
 355                 if (mbuf->nb_segs > 1)
 356                         return;
 357                 if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 358                         if (l4 == RTE_PTYPE_L4_UDP) {
 359                                 udp_hdr = (struct rte_udp_hdr *)l4_hdr;
 360                                 if (udp_hdr->dgram_cksum == 0) {
 361                                         /*
 362                                          * For IPv4, a zero UDP checksum
 363                                          * indicates that the sender did not
 364                                          * generate one [RFC 768].
 365                                          */
 366                                         mbuf->ol_flags |= PKT_RX_L4_CKSUM_NONE;
 367                                         return;
 368                                 }
 369                         }
 370                         cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
 371                 } else { /* l3 == RTE_PTYPE_L3_IPV6, checked above */
 372                         cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
 373                 }
 374                 mbuf->ol_flags |= cksum ?
 375                         PKT_RX_L4_CKSUM_BAD :
 376                         PKT_RX_L4_CKSUM_GOOD;
 377         }
 378 }
 379
 380 static uint64_t
 381 tap_rx_offload_get_port_capa(void)
 382 {
 383         /*
 384          * No specific port Rx offload capabilities.
 385          */
 386         return 0;
 387 }
 388
 389 static uint64_t
 390 tap_rx_offload_get_queue_capa(void)
 391 {
 392         return DEV_RX_OFFLOAD_SCATTER |
 393                DEV_RX_OFFLOAD_IPV4_CKSUM |
 394                DEV_RX_OFFLOAD_UDP_CKSUM |
 395                DEV_RX_OFFLOAD_TCP_CKSUM;
 396 }
 397
 398 static void
 399 tap_rxq_pool_free(struct rte_mbuf *pool)
 400 {
 401         struct rte_mbuf *mbuf = pool;
 402         uint16_t nb_segs = 1;
 403
 404         if (mbuf == NULL)
 405                 return;
 406
 407         while (mbuf->next) {
 408                 mbuf = mbuf->next;
 409                 nb_segs++;
 410         }
 411         pool->nb_segs = nb_segs;
 412         rte_pktmbuf_free(pool);
 413 }
 414
 415 /* Callback to handle the rx burst of packets to the correct interface and
 416  * file descriptor(s) in a multi-queue setup.
 417  */
 418 static uint16_t
 419 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 420 {
 421         struct rx_queue *rxq = queue;
 422         struct pmd_process_private *process_private;
 423         uint16_t num_rx;
 424         unsigned long num_rx_bytes = 0;
 425         uint32_t trigger = tap_trigger;
 426
 427         if (trigger == rxq->trigger_seen)
 428                 return 0;
 429
 430         process_private = rte_eth_devices[rxq->in_port].process_private;
 431         for (num_rx = 0; num_rx < nb_pkts; ) {
 432                 struct rte_mbuf *mbuf = rxq->pool;
 433                 struct rte_mbuf *seg = NULL;
 434                 struct rte_mbuf *new_tail = NULL;
 435                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 436                 int len;
 437
 438                 len = readv(process_private->rxq_fds[rxq->queue_id],
 439                         *rxq->iovecs,
 440                         1 + (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
 441                              rxq->nb_rx_desc : 1));
 442                 if (len < (int)sizeof(struct tun_pi))
 443                         break;
 444
 445                 /* Packet couldn't fit in the provided mbuf */
 446                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 447                         rxq->stats.ierrors++;
 448                         continue;
 449                 }
 450
 451                 len -= sizeof(struct tun_pi);
 452
 453                 mbuf->pkt_len = len;
 454                 mbuf->port = rxq->in_port;
 455                 while (1) {
 456                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 457
 458                         if (unlikely(!buf)) {
 459                                 rxq->stats.rx_nombuf++;
 460                                 /* No new buf has been allocated: do nothing */
 461                                 if (!new_tail || !seg)
 462                                         goto end;
 463
 464                                 seg->next = NULL;
 465                                 tap_rxq_pool_free(mbuf);
 466
 467                                 goto end;
 468                         }
 469                         seg = seg ? seg->next : mbuf;
 470                         if (rxq->pool == mbuf)
 471                                 rxq->pool = buf;
 472                         if (new_tail)
 473                                 new_tail->next = buf;
 474                         new_tail = buf;
 475                         new_tail->next = seg->next;
 476
 477                         /* iovecs[0] is reserved for packet info (pi) */
 478                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 479                                 buf->buf_len - data_off;
 480                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 481                                 (char *)buf->buf_addr + data_off;
 482
 483                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 484                         seg->data_off = data_off;
 485
 486                         len -= seg->data_len;
 487                         if (len <= 0)
 488                                 break;
 489                         mbuf->nb_segs++;
 490                         /* First segment has headroom, not the others */
 491                         data_off = 0;
 492                 }
 493                 seg->next = NULL;
 494                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 495                                                       RTE_PTYPE_ALL_MASK);
 496                 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
 497                         tap_verify_csum(mbuf);
 498
 499                 /* account for the receive frame */
 500                 bufs[num_rx++] = mbuf;
 501                 num_rx_bytes += mbuf->pkt_len;
 502         }
 503 end:
 504         rxq->stats.ipackets += num_rx;
 505         rxq->stats.ibytes += num_rx_bytes;
 506
 507         if (trigger && num_rx < nb_pkts)
 508                 rxq->trigger_seen = trigger;
 509
 510         return num_rx;
 511 }
 512
 513 static uint64_t
 514 tap_tx_offload_get_port_capa(void)
 515 {
 516         /*
 517          * No specific port Tx offload capabilities.
 518          */
 519         return 0;
 520 }
 521
 522 static uint64_t
 523 tap_tx_offload_get_queue_capa(void)
 524 {
 525         return DEV_TX_OFFLOAD_MULTI_SEGS |
 526                DEV_TX_OFFLOAD_IPV4_CKSUM |
 527                DEV_TX_OFFLOAD_UDP_CKSUM |
 528                DEV_TX_OFFLOAD_TCP_CKSUM |
 529                DEV_TX_OFFLOAD_TCP_TSO;
 530 }
 531
 532 /* Finalize l4 checksum calculation */
 533 static void
 534 tap_tx_l4_cksum(uint16_t *l4_cksum, uint16_t l4_phdr_cksum,
 535                 uint32_t l4_raw_cksum)
 536 {
 537         if (l4_cksum) {
 538                 uint32_t cksum;
 539
 540                 cksum = __rte_raw_cksum_reduce(l4_raw_cksum);
 541                 cksum += l4_phdr_cksum;
 542
 543                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 544                 cksum = (~cksum) & 0xffff;
 545                 if (cksum == 0)
 546                         cksum = 0xffff;
 547                 *l4_cksum = cksum;
 548         }
 549 }
 550
 551 /* Accumaulate L4 raw checksums */
 552 static void
 553 tap_tx_l4_add_rcksum(char *l4_data, unsigned int l4_len, uint16_t *l4_cksum,
 554                         uint32_t *l4_raw_cksum)
 555 {
 556         if (l4_cksum == NULL)
 557                 return;
 558
 559         *l4_raw_cksum = __rte_raw_cksum(l4_data, l4_len, *l4_raw_cksum);
 560 }
 561
 562 /* L3 and L4 pseudo headers checksum offloads */
 563 static void
 564 tap_tx_l3_cksum(char *packet, uint64_t ol_flags, unsigned int l2_len,
 565                 unsigned int l3_len, unsigned int l4_len, uint16_t **l4_cksum,
 566                 uint16_t *l4_phdr_cksum, uint32_t *l4_raw_cksum)
 567 {
 568         void *l3_hdr = packet + l2_len;
 569
 570         if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
 571                 struct rte_ipv4_hdr *iph = l3_hdr;
 572                 uint16_t cksum;
 573
 574                 iph->hdr_checksum = 0;
 575                 cksum = rte_raw_cksum(iph, l3_len);
 576                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 577         }
 578         if (ol_flags & PKT_TX_L4_MASK) {
 579                 void *l4_hdr;
 580
 581                 l4_hdr = packet + l2_len + l3_len;
 582                 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
 583                         *l4_cksum = &((struct rte_udp_hdr *)l4_hdr)->dgram_cksum;
 584                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 585                         *l4_cksum = &((struct rte_tcp_hdr *)l4_hdr)->cksum;
 586                 else
 587                         return;
 588                 **l4_cksum = 0;
 589                 if (ol_flags & PKT_TX_IPV4)
 590                         *l4_phdr_cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 591                 else
 592                         *l4_phdr_cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 593                 *l4_raw_cksum = __rte_raw_cksum(l4_hdr, l4_len, 0);
 594         }
 595 }
 596
 597 static inline int
 598 tap_write_mbufs(struct tx_queue *txq, uint16_t num_mbufs,
 599                         struct rte_mbuf **pmbufs,
 600                         uint16_t *num_packets, unsigned long *num_tx_bytes)
 601 {
 602         int i;
 603         uint16_t l234_hlen;
 604         struct pmd_process_private *process_private;
 605
 606         process_private = rte_eth_devices[txq->out_port].process_private;
 607
 608         for (i = 0; i < num_mbufs; i++) {
 609                 struct rte_mbuf *mbuf = pmbufs[i];
 610                 struct iovec iovecs[mbuf->nb_segs + 2];
 611                 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
 612                 struct rte_mbuf *seg = mbuf;
 613                 char m_copy[mbuf->data_len];
 614                 int proto;
 615                 int n;
 616                 int j;
 617                 int k; /* current index in iovecs for copying segments */
 618                 uint16_t seg_len; /* length of first segment */
 619                 uint16_t nb_segs;
 620                 uint16_t *l4_cksum; /* l4 checksum (pseudo header + payload) */
 621                 uint32_t l4_raw_cksum = 0; /* TCP/UDP payload raw checksum */
 622                 uint16_t l4_phdr_cksum = 0; /* TCP/UDP pseudo header checksum */
 623                 uint16_t is_cksum = 0; /* in case cksum should be offloaded */
 624
 625                 l4_cksum = NULL;
 626                 if (txq->type == ETH_TUNTAP_TYPE_TUN) {
 627                         /*
 628                          * TUN and TAP are created with IFF_NO_PI disabled.
 629                          * For TUN PMD this mandatory as fields are used by
 630                          * Kernel tun.c to determine whether its IP or non IP
 631                          * packets.
 632                          *
 633                          * The logic fetches the first byte of data from mbuf
 634                          * then compares whether its v4 or v6. If first byte
 635                          * is 4 or 6, then protocol field is updated.
 636                          */
 637                         char *buff_data = rte_pktmbuf_mtod(seg, void *);
 638                         proto = (*buff_data & 0xf0);
 639                         pi.proto = (proto == 0x40) ?
 640                                 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4) :
 641                                 ((proto == 0x60) ?
 642                                         rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6) :
 643                                         0x00);
 644                 }
 645
 646                 k = 0;
 647                 iovecs[k].iov_base = &pi;
 648                 iovecs[k].iov_len = sizeof(pi);
 649                 k++;
 650
 651                 nb_segs = mbuf->nb_segs;
 652                 if (txq->csum &&
 653                     ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 654                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 655                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
 656                         is_cksum = 1;
 657
 658                         /* Support only packets with at least layer 4
 659                          * header included in the first segment
 660                          */
 661                         seg_len = rte_pktmbuf_data_len(mbuf);
 662                         l234_hlen = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
 663                         if (seg_len < l234_hlen)
 664                                 return -1;
 665
 666                         /* To change checksums, work on a * copy of l2, l3
 667                          * headers + l4 pseudo header
 668                          */
 669                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 670                                         l234_hlen);
 671                         tap_tx_l3_cksum(m_copy, mbuf->ol_flags,
 672                                        mbuf->l2_len, mbuf->l3_len, mbuf->l4_len,
 673                                        &l4_cksum, &l4_phdr_cksum,
 674                                        &l4_raw_cksum);
 675                         iovecs[k].iov_base = m_copy;
 676                         iovecs[k].iov_len = l234_hlen;
 677                         k++;
 678
 679                         /* Update next iovecs[] beyond l2, l3, l4 headers */
 680                         if (seg_len > l234_hlen) {
 681                                 iovecs[k].iov_len = seg_len - l234_hlen;
 682                                 iovecs[k].iov_base =
 683                                         rte_pktmbuf_mtod(seg, char *) +
 684                                                 l234_hlen;
 685                                 tap_tx_l4_add_rcksum(iovecs[k].iov_base,
 686                                         iovecs[k].iov_len, l4_cksum,
 687                                         &l4_raw_cksum);
 688                                 k++;
 689                                 nb_segs++;
 690                         }
 691                         seg = seg->next;
 692                 }
 693
 694                 for (j = k; j <= nb_segs; j++) {
 695                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 696                         iovecs[j].iov_base = rte_pktmbuf_mtod(seg, void *);
 697                         if (is_cksum)
 698                                 tap_tx_l4_add_rcksum(iovecs[j].iov_base,
 699                                         iovecs[j].iov_len, l4_cksum,
 700                                         &l4_raw_cksum);
 701                         seg = seg->next;
 702                 }
 703
 704                 if (is_cksum)
 705                         tap_tx_l4_cksum(l4_cksum, l4_phdr_cksum, l4_raw_cksum);
 706
 707                 /* copy the tx frame data */
 708                 n = writev(process_private->txq_fds[txq->queue_id], iovecs, j);
 709                 if (n <= 0)
 710                         return -1;
 711
 712                 (*num_packets)++;
 713                 (*num_tx_bytes) += rte_pktmbuf_pkt_len(mbuf);
 714         }
 715         return 0;
 716 }
 717
 718 /* Callback to handle sending packets from the tap interface
 719  */
 720 static uint16_t
 721 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 722 {
 723         struct tx_queue *txq = queue;
 724         uint16_t num_tx = 0;
 725         uint16_t num_packets = 0;
 726         unsigned long num_tx_bytes = 0;
 727         uint32_t max_size;
 728         int i;
 729
 730         if (unlikely(nb_pkts == 0))
 731                 return 0;
 732
 733         struct rte_mbuf *gso_mbufs[MAX_GSO_MBUFS];
 734         max_size = *txq->mtu + (RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN + 4);
 735         for (i = 0; i < nb_pkts; i++) {
 736                 struct rte_mbuf *mbuf_in = bufs[num_tx];
 737                 struct rte_mbuf **mbuf;
 738                 uint16_t num_mbufs = 0;
 739                 uint16_t tso_segsz = 0;
 740                 int ret;
 741                 int num_tso_mbufs;
 742                 uint16_t hdrs_len;
 743                 uint64_t tso;
 744
 745                 tso = mbuf_in->ol_flags & PKT_TX_TCP_SEG;
 746                 if (tso) {
 747                         struct rte_gso_ctx *gso_ctx = &txq->gso_ctx;
 748
 749                         /* TCP segmentation implies TCP checksum offload */
 750                         mbuf_in->ol_flags |= PKT_TX_TCP_CKSUM;
 751
 752                         /* gso size is calculated without RTE_ETHER_CRC_LEN */
 753                         hdrs_len = mbuf_in->l2_len + mbuf_in->l3_len +
 754                                         mbuf_in->l4_len;
 755                         tso_segsz = mbuf_in->tso_segsz + hdrs_len;
 756                         if (unlikely(tso_segsz == hdrs_len) ||
 757                                 tso_segsz > *txq->mtu) {
 758                                 txq->stats.errs++;
 759                                 break;
 760                         }
 761                         gso_ctx->gso_size = tso_segsz;
 762                         /* 'mbuf_in' packet to segment */
 763                         num_tso_mbufs = rte_gso_segment(mbuf_in,
 764                                 gso_ctx, /* gso control block */
 765                                 (struct rte_mbuf **)&gso_mbufs, /* out mbufs */
 766                                 RTE_DIM(gso_mbufs)); /* max tso mbufs */
 767
 768                         /* ret contains the number of new created mbufs */
 769                         if (num_tso_mbufs < 0)
 770                                 break;
 771
 772                         if (num_tso_mbufs >= 1) {
 773                                 mbuf = gso_mbufs;
 774                                 num_mbufs = num_tso_mbufs;
 775                         } else {
 776                                 /* 0 means it can be transmitted directly
 777                                  * without gso.
 778                                  */
 779                                 mbuf = &mbuf_in;
 780                                 num_mbufs = 1;
 781                         }
 782                 } else {
 783                         /* stats.errs will be incremented */
 784                         if (rte_pktmbuf_pkt_len(mbuf_in) > max_size)
 785                                 break;
 786
 787                         /* ret 0 indicates no new mbufs were created */
 788                         num_tso_mbufs = 0;
 789                         mbuf = &mbuf_in;
 790                         num_mbufs = 1;
 791                 }
 792
 793                 ret = tap_write_mbufs(txq, num_mbufs, mbuf,
 794                                 &num_packets, &num_tx_bytes);
 795                 if (ret == -1) {
 796                         txq->stats.errs++;
 797                         /* free tso mbufs */
 798                         if (num_tso_mbufs > 0)
 799                                 rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
 800                         break;
 801                 }
 802                 num_tx++;
 803                 /* free original mbuf */
 804                 rte_pktmbuf_free(mbuf_in);
 805                 /* free tso mbufs */
 806                 if (num_tso_mbufs > 0)
 807                         rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
 808         }
 809
 810         txq->stats.opackets += num_packets;
 811         txq->stats.errs += nb_pkts - num_tx;
 812         txq->stats.obytes += num_tx_bytes;
 813
 814         return num_tx;
 815 }
 816
 817 static const char *
 818 tap_ioctl_req2str(unsigned long request)
 819 {
 820         switch (request) {
 821         case SIOCSIFFLAGS:
 822                 return "SIOCSIFFLAGS";
 823         case SIOCGIFFLAGS:
 824                 return "SIOCGIFFLAGS";
 825         case SIOCGIFHWADDR:
 826                 return "SIOCGIFHWADDR";
 827         case SIOCSIFHWADDR:
 828                 return "SIOCSIFHWADDR";
 829         case SIOCSIFMTU:
 830                 return "SIOCSIFMTU";
 831         }
 832         return "UNKNOWN";
 833 }
 834
 835 static int
 836 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 837           struct ifreq *ifr, int set, enum ioctl_mode mode)
 838 {
 839         short req_flags = ifr->ifr_flags;
 840         int remote = pmd->remote_if_index &&
 841                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 842
 843         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 844                 return 0;
 845         /*
 846          * If there is a remote netdevice, apply ioctl on it, then apply it on
 847          * the tap netdevice.
 848          */
 849 apply:
 850         if (remote)
 851                 strlcpy(ifr->ifr_name, pmd->remote_iface, IFNAMSIZ);
 852         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 853                 strlcpy(ifr->ifr_name, pmd->name, IFNAMSIZ);
 854         switch (request) {
 855         case SIOCSIFFLAGS:
 856                 /* fetch current flags to leave other flags untouched */
 857                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 858                         goto error;
 859                 if (set)
 860                         ifr->ifr_flags |= req_flags;
 861                 else
 862                         ifr->ifr_flags &= ~req_flags;
 863                 break;
 864         case SIOCGIFFLAGS:
 865         case SIOCGIFHWADDR:
 866         case SIOCSIFHWADDR:
 867         case SIOCSIFMTU:
 868                 break;
 869         default:
 870                 TAP_LOG(WARNING, "%s: ioctl() called with wrong arg",
 871                         pmd->name);
 872                 return -EINVAL;
 873         }
 874         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 875                 goto error;
 876         if (remote-- && mode == LOCAL_AND_REMOTE)
 877                 goto apply;
 878         return 0;
 879
 880 error:
 881         TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
 882                 tap_ioctl_req2str(request), strerror(errno), errno);
 883         return -errno;
 884 }
 885
 886 static int
 887 tap_link_set_down(struct rte_eth_dev *dev)
 888 {
 889         struct pmd_internals *pmd = dev->data->dev_private;
 890         struct ifreq ifr = { .ifr_flags = IFF_UP };
 891
 892         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 893         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 894 }
 895
 896 static int
 897 tap_link_set_up(struct rte_eth_dev *dev)
 898 {
 899         struct pmd_internals *pmd = dev->data->dev_private;
 900         struct ifreq ifr = { .ifr_flags = IFF_UP };
 901
 902         dev->data->dev_link.link_status = ETH_LINK_UP;
 903         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 904 }
 905
 906 static int
 907 tap_dev_start(struct rte_eth_dev *dev)
 908 {
 909         int err, i;
 910
 911         err = tap_intr_handle_set(dev, 1);
 912         if (err)
 913                 return err;
 914
 915         err = tap_link_set_up(dev);
 916         if (err)
 917                 return err;
 918
 919         for (i = 0; i < dev->data->nb_tx_queues; i++)
 920                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 921         for (i = 0; i < dev->data->nb_rx_queues; i++)
 922                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 923
 924         return err;
 925 }
 926
 927 /* This function gets called when the current port gets stopped.
 928  */
 929 static int
 930 tap_dev_stop(struct rte_eth_dev *dev)
 931 {
 932         int i;
 933
 934         for (i = 0; i < dev->data->nb_tx_queues; i++)
 935                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 936         for (i = 0; i < dev->data->nb_rx_queues; i++)
 937                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 938
 939         tap_intr_handle_set(dev, 0);
 940         tap_link_set_down(dev);
 941
 942         return 0;
 943 }
 944
 945 static int
 946 tap_dev_configure(struct rte_eth_dev *dev)
 947 {
 948         struct pmd_internals *pmd = dev->data->dev_private;
 949
 950         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 951                 TAP_LOG(ERR,
 952                         "%s: number of rx queues %d exceeds max num of queues %d",
 953                         dev->device->name,
 954                         dev->data->nb_rx_queues,
 955                         RTE_PMD_TAP_MAX_QUEUES);
 956                 return -1;
 957         }
 958         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 959                 TAP_LOG(ERR,
 960                         "%s: number of tx queues %d exceeds max num of queues %d",
 961                         dev->device->name,
 962                         dev->data->nb_tx_queues,
 963                         RTE_PMD_TAP_MAX_QUEUES);
 964                 return -1;
 965         }
 966
 967         TAP_LOG(INFO, "%s: %s: TX configured queues number: %u",
 968                 dev->device->name, pmd->name, dev->data->nb_tx_queues);
 969
 970         TAP_LOG(INFO, "%s: %s: RX configured queues number: %u",
 971                 dev->device->name, pmd->name, dev->data->nb_rx_queues);
 972
 973         return 0;
 974 }
 975
 976 static uint32_t
 977 tap_dev_speed_capa(void)
 978 {
 979         uint32_t speed = pmd_link.link_speed;
 980         uint32_t capa = 0;
 981
 982         if (speed >= ETH_SPEED_NUM_10M)
 983                 capa |= ETH_LINK_SPEED_10M;
 984         if (speed >= ETH_SPEED_NUM_100M)
 985                 capa |= ETH_LINK_SPEED_100M;
 986         if (speed >= ETH_SPEED_NUM_1G)
 987                 capa |= ETH_LINK_SPEED_1G;
 988         if (speed >= ETH_SPEED_NUM_5G)
 989                 capa |= ETH_LINK_SPEED_2_5G;
 990         if (speed >= ETH_SPEED_NUM_5G)
 991                 capa |= ETH_LINK_SPEED_5G;
 992         if (speed >= ETH_SPEED_NUM_10G)
 993                 capa |= ETH_LINK_SPEED_10G;
 994         if (speed >= ETH_SPEED_NUM_20G)
 995                 capa |= ETH_LINK_SPEED_20G;
 996         if (speed >= ETH_SPEED_NUM_25G)
 997                 capa |= ETH_LINK_SPEED_25G;
 998         if (speed >= ETH_SPEED_NUM_40G)
 999                 capa |= ETH_LINK_SPEED_40G;
1000         if (speed >= ETH_SPEED_NUM_50G)
1001                 capa |= ETH_LINK_SPEED_50G;
1002         if (speed >= ETH_SPEED_NUM_56G)
1003                 capa |= ETH_LINK_SPEED_56G;
1004         if (speed >= ETH_SPEED_NUM_100G)
1005                 capa |= ETH_LINK_SPEED_100G;
1006
1007         return capa;
1008 }
1009
1010 static int
1011 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1012 {
1013         struct pmd_internals *internals = dev->data->dev_private;
1014
1015         dev_info->if_index = internals->if_index;
1016         dev_info->max_mac_addrs = 1;
1017         dev_info->max_rx_pktlen = (uint32_t)RTE_ETHER_MAX_VLAN_FRAME_LEN;
1018         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
1019         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
1020         dev_info->min_rx_bufsize = 0;
1021         dev_info->speed_capa = tap_dev_speed_capa();
1022         dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
1023         dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
1024                                     dev_info->rx_queue_offload_capa;
1025         dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
1026         dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
1027                                     dev_info->tx_queue_offload_capa;
1028         dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
1029         /*
1030          * limitation: TAP supports all of IP, UDP and TCP hash
1031          * functions together and not in partial combinations
1032          */
1033         dev_info->flow_type_rss_offloads = ~TAP_RSS_HF_MASK;
1034
1035         return 0;
1036 }
1037
1038 static int
1039 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
1040 {
1041         unsigned int i, imax;
1042         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
1043         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
1044         unsigned long rx_nombuf = 0, ierrors = 0;
1045         const struct pmd_internals *pmd = dev->data->dev_private;
1046
1047         /* rx queue statistics */
1048         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
1049                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
1050         for (i = 0; i < imax; i++) {
1051                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
1052                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
1053                 rx_total += tap_stats->q_ipackets[i];
1054                 rx_bytes_total += tap_stats->q_ibytes[i];
1055                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
1056                 ierrors += pmd->rxq[i].stats.ierrors;
1057         }
1058
1059         /* tx queue statistics */
1060         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
1061                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
1062
1063         for (i = 0; i < imax; i++) {
1064                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
1065                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
1066                 tx_total += tap_stats->q_opackets[i];
1067                 tx_err_total += pmd->txq[i].stats.errs;
1068                 tx_bytes_total += tap_stats->q_obytes[i];
1069         }
1070
1071         tap_stats->ipackets = rx_total;
1072         tap_stats->ibytes = rx_bytes_total;
1073         tap_stats->ierrors = ierrors;
1074         tap_stats->rx_nombuf = rx_nombuf;
1075         tap_stats->opackets = tx_total;
1076         tap_stats->oerrors = tx_err_total;
1077         tap_stats->obytes = tx_bytes_total;
1078         return 0;
1079 }
1080
1081 static int
1082 tap_stats_reset(struct rte_eth_dev *dev)
1083 {
1084         int i;
1085         struct pmd_internals *pmd = dev->data->dev_private;
1086
1087         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1088                 pmd->rxq[i].stats.ipackets = 0;
1089                 pmd->rxq[i].stats.ibytes = 0;
1090                 pmd->rxq[i].stats.ierrors = 0;
1091                 pmd->rxq[i].stats.rx_nombuf = 0;
1092
1093                 pmd->txq[i].stats.opackets = 0;
1094                 pmd->txq[i].stats.errs = 0;
1095                 pmd->txq[i].stats.obytes = 0;
1096         }
1097
1098         return 0;
1099 }
1100
1101 static int
1102 tap_dev_close(struct rte_eth_dev *dev)
1103 {
1104         int i;
1105         struct pmd_internals *internals = dev->data->dev_private;
1106         struct pmd_process_private *process_private = dev->process_private;
1107         struct rx_queue *rxq;
1108
1109         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1110                 rte_free(dev->process_private);
1111                 return 0;
1112         }
1113
1114         tap_link_set_down(dev);
1115         if (internals->nlsk_fd != -1) {
1116                 tap_flow_flush(dev, NULL);
1117                 tap_flow_implicit_flush(internals, NULL);
1118                 tap_nl_final(internals->nlsk_fd);
1119                 internals->nlsk_fd = -1;
1120         }
1121
1122         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1123                 if (process_private->rxq_fds[i] != -1) {
1124                         rxq = &internals->rxq[i];
1125                         close(process_private->rxq_fds[i]);
1126                         process_private->rxq_fds[i] = -1;
1127                         tap_rxq_pool_free(rxq->pool);
1128                         rte_free(rxq->iovecs);
1129                         rxq->pool = NULL;
1130                         rxq->iovecs = NULL;
1131                 }
1132                 if (process_private->txq_fds[i] != -1) {
1133                         close(process_private->txq_fds[i]);
1134                         process_private->txq_fds[i] = -1;
1135                 }
1136         }
1137
1138         if (internals->remote_if_index) {
1139                 /* Restore initial remote state */
1140                 int ret = ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
1141                                 &internals->remote_initial_flags);
1142                 if (ret)
1143                         TAP_LOG(ERR, "restore remote state failed: %d", ret);
1144
1145         }
1146
1147         rte_mempool_free(internals->gso_ctx_mp);
1148         internals->gso_ctx_mp = NULL;
1149
1150         if (internals->ka_fd != -1) {
1151                 close(internals->ka_fd);
1152                 internals->ka_fd = -1;
1153         }
1154
1155         /* mac_addrs must not be freed alone because part of dev_private */
1156         dev->data->mac_addrs = NULL;
1157
1158         internals = dev->data->dev_private;
1159         TAP_LOG(DEBUG, "Closing %s Ethernet device on numa %u",
1160                 tuntap_types[internals->type], rte_socket_id());
1161
1162         if (internals->ioctl_sock != -1) {
1163                 close(internals->ioctl_sock);
1164                 internals->ioctl_sock = -1;
1165         }
1166         rte_free(dev->process_private);
1167         if (tap_devices_count == 1)
1168                 rte_mp_action_unregister(TAP_MP_KEY);
1169         tap_devices_count--;
1170         /*
1171          * Since TUN device has no more opened file descriptors
1172          * it will be removed from kernel
1173          */
1174
1175         return 0;
1176 }
1177
1178 static void
1179 tap_rx_queue_release(void *queue)
1180 {
1181         struct rx_queue *rxq = queue;
1182         struct pmd_process_private *process_private;
1183
1184         if (!rxq)
1185                 return;
1186         process_private = rte_eth_devices[rxq->in_port].process_private;
1187         if (process_private->rxq_fds[rxq->queue_id] != -1) {
1188                 close(process_private->rxq_fds[rxq->queue_id]);
1189                 process_private->rxq_fds[rxq->queue_id] = -1;
1190                 tap_rxq_pool_free(rxq->pool);
1191                 rte_free(rxq->iovecs);
1192                 rxq->pool = NULL;
1193                 rxq->iovecs = NULL;
1194         }
1195 }
1196
1197 static void
1198 tap_tx_queue_release(void *queue)
1199 {
1200         struct tx_queue *txq = queue;
1201         struct pmd_process_private *process_private;
1202
1203         if (!txq)
1204                 return;
1205         process_private = rte_eth_devices[txq->out_port].process_private;
1206
1207         if (process_private->txq_fds[txq->queue_id] != -1) {
1208                 close(process_private->txq_fds[txq->queue_id]);
1209                 process_private->txq_fds[txq->queue_id] = -1;
1210         }
1211 }
1212
1213 static int
1214 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1215 {
1216         struct rte_eth_link *dev_link = &dev->data->dev_link;
1217         struct pmd_internals *pmd = dev->data->dev_private;
1218         struct ifreq ifr = { .ifr_flags = 0 };
1219
1220         if (pmd->remote_if_index) {
1221                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
1222                 if (!(ifr.ifr_flags & IFF_UP) ||
1223                     !(ifr.ifr_flags & IFF_RUNNING)) {
1224                         dev_link->link_status = ETH_LINK_DOWN;
1225                         return 0;
1226                 }
1227         }
1228         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
1229         dev_link->link_status =
1230                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
1231                  ETH_LINK_UP :
1232                  ETH_LINK_DOWN);
1233         return 0;
1234 }
1235
1236 static int
1237 tap_promisc_enable(struct rte_eth_dev *dev)
1238 {
1239         struct pmd_internals *pmd = dev->data->dev_private;
1240         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1241         int ret;
1242
1243         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1244         if (ret != 0)
1245                 return ret;
1246
1247         if (pmd->remote_if_index && !pmd->flow_isolate) {
1248                 dev->data->promiscuous = 1;
1249                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
1250                 if (ret != 0) {
1251                         /* Rollback promisc flag */
1252                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1253                         /*
1254                          * rte_eth_dev_promiscuous_enable() rollback
1255                          * dev->data->promiscuous in the case of failure.
1256                          */
1257                         return ret;
1258                 }
1259         }
1260
1261         return 0;
1262 }
1263
1264 static int
1265 tap_promisc_disable(struct rte_eth_dev *dev)
1266 {
1267         struct pmd_internals *pmd = dev->data->dev_private;
1268         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1269         int ret;
1270
1271         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1272         if (ret != 0)
1273                 return ret;
1274
1275         if (pmd->remote_if_index && !pmd->flow_isolate) {
1276                 dev->data->promiscuous = 0;
1277                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
1278                 if (ret != 0) {
1279                         /* Rollback promisc flag */
1280                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1281                         /*
1282                          * rte_eth_dev_promiscuous_disable() rollback
1283                          * dev->data->promiscuous in the case of failure.
1284                          */
1285                         return ret;
1286                 }
1287         }
1288
1289         return 0;
1290 }
1291
1292 static int
1293 tap_allmulti_enable(struct rte_eth_dev *dev)
1294 {
1295         struct pmd_internals *pmd = dev->data->dev_private;
1296         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1297         int ret;
1298
1299         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1300         if (ret != 0)
1301                 return ret;
1302
1303         if (pmd->remote_if_index && !pmd->flow_isolate) {
1304                 dev->data->all_multicast = 1;
1305                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
1306                 if (ret != 0) {
1307                         /* Rollback allmulti flag */
1308                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1309                         /*
1310                          * rte_eth_dev_allmulticast_enable() rollback
1311                          * dev->data->all_multicast in the case of failure.
1312                          */
1313                         return ret;
1314                 }
1315         }
1316
1317         return 0;
1318 }
1319
1320 static int
1321 tap_allmulti_disable(struct rte_eth_dev *dev)
1322 {
1323         struct pmd_internals *pmd = dev->data->dev_private;
1324         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1325         int ret;
1326
1327         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1328         if (ret != 0)
1329                 return ret;
1330
1331         if (pmd->remote_if_index && !pmd->flow_isolate) {
1332                 dev->data->all_multicast = 0;
1333                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
1334                 if (ret != 0) {
1335                         /* Rollback allmulti flag */
1336                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1337                         /*
1338                          * rte_eth_dev_allmulticast_disable() rollback
1339                          * dev->data->all_multicast in the case of failure.
1340                          */
1341                         return ret;
1342                 }
1343         }
1344
1345         return 0;
1346 }
1347
1348 static int
1349 tap_mac_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1350 {
1351         struct pmd_internals *pmd = dev->data->dev_private;
1352         enum ioctl_mode mode = LOCAL_ONLY;
1353         struct ifreq ifr;
1354         int ret;
1355
1356         if (pmd->type == ETH_TUNTAP_TYPE_TUN) {
1357                 TAP_LOG(ERR, "%s: can't MAC address for TUN",
1358                         dev->device->name);
1359                 return -ENOTSUP;
1360         }
1361
1362         if (rte_is_zero_ether_addr(mac_addr)) {
1363                 TAP_LOG(ERR, "%s: can't set an empty MAC address",
1364                         dev->device->name);
1365                 return -EINVAL;
1366         }
1367         /* Check the actual current MAC address on the tap netdevice */
1368         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
1369         if (ret < 0)
1370                 return ret;
1371         if (rte_is_same_ether_addr(
1372                         (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1373                         mac_addr))
1374                 return 0;
1375         /* Check the current MAC address on the remote */
1376         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
1377         if (ret < 0)
1378                 return ret;
1379         if (!rte_is_same_ether_addr(
1380                         (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1381                         mac_addr))
1382                 mode = LOCAL_AND_REMOTE;
1383         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1384         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, RTE_ETHER_ADDR_LEN);
1385         ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
1386         if (ret < 0)
1387                 return ret;
1388         rte_memcpy(&pmd->eth_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1389         if (pmd->remote_if_index && !pmd->flow_isolate) {
1390                 /* Replace MAC redirection rule after a MAC change */
1391                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
1392                 if (ret < 0) {
1393                         TAP_LOG(ERR,
1394                                 "%s: Couldn't delete MAC redirection rule",
1395                                 dev->device->name);
1396                         return ret;
1397                 }
1398                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1399                 if (ret < 0) {
1400                         TAP_LOG(ERR,
1401                                 "%s: Couldn't add MAC redirection rule",
1402                                 dev->device->name);
1403                         return ret;
1404                 }
1405         }
1406
1407         return 0;
1408 }
1409
1410 static int
1411 tap_gso_ctx_setup(struct rte_gso_ctx *gso_ctx, struct rte_eth_dev *dev)
1412 {
1413         uint32_t gso_types;
1414         char pool_name[64];
1415         struct pmd_internals *pmd = dev->data->dev_private;
1416         int ret;
1417
1418         /* initialize GSO context */
1419         gso_types = DEV_TX_OFFLOAD_TCP_TSO;
1420         if (!pmd->gso_ctx_mp) {
1421                 /*
1422                  * Create private mbuf pool with TAP_GSO_MBUF_SEG_SIZE
1423                  * bytes size per mbuf use this pool for both direct and
1424                  * indirect mbufs
1425                  */
1426                 ret = snprintf(pool_name, sizeof(pool_name), "mp_%s",
1427                                 dev->device->name);
1428                 if (ret < 0 || ret >= (int)sizeof(pool_name)) {
1429                         TAP_LOG(ERR,
1430                                 "%s: failed to create mbuf pool name for device %s,"
1431                                 "device name too long or output error, ret: %d\n",
1432                                 pmd->name, dev->device->name, ret);
1433                         return -ENAMETOOLONG;
1434                 }
1435                 pmd->gso_ctx_mp = rte_pktmbuf_pool_create(pool_name,
1436                         TAP_GSO_MBUFS_NUM, TAP_GSO_MBUF_CACHE_SIZE, 0,
1437                         RTE_PKTMBUF_HEADROOM + TAP_GSO_MBUF_SEG_SIZE,
1438                         SOCKET_ID_ANY);
1439                 if (!pmd->gso_ctx_mp) {
1440                         TAP_LOG(ERR,
1441                                 "%s: failed to create mbuf pool for device %s\n",
1442                                 pmd->name, dev->device->name);
1443                         return -1;
1444                 }
1445         }
1446
1447         gso_ctx->direct_pool = pmd->gso_ctx_mp;
1448         gso_ctx->indirect_pool = pmd->gso_ctx_mp;
1449         gso_ctx->gso_types = gso_types;
1450         gso_ctx->gso_size = 0; /* gso_size is set in tx_burst() per packet */
1451         gso_ctx->flag = 0;
1452
1453         return 0;
1454 }
1455
1456 static int
1457 tap_setup_queue(struct rte_eth_dev *dev,
1458                 struct pmd_internals *internals,
1459                 uint16_t qid,
1460                 int is_rx)
1461 {
1462         int ret;
1463         int *fd;
1464         int *other_fd;
1465         const char *dir;
1466         struct pmd_internals *pmd = dev->data->dev_private;
1467         struct pmd_process_private *process_private = dev->process_private;
1468         struct rx_queue *rx = &internals->rxq[qid];
1469         struct tx_queue *tx = &internals->txq[qid];
1470         struct rte_gso_ctx *gso_ctx;
1471
1472         if (is_rx) {
1473                 fd = &process_private->rxq_fds[qid];
1474                 other_fd = &process_private->txq_fds[qid];
1475                 dir = "rx";
1476                 gso_ctx = NULL;
1477         } else {
1478                 fd = &process_private->txq_fds[qid];
1479                 other_fd = &process_private->rxq_fds[qid];
1480                 dir = "tx";
1481                 gso_ctx = &tx->gso_ctx;
1482         }
1483         if (*fd != -1) {
1484                 /* fd for this queue already exists */
1485                 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1486                         pmd->name, *fd, dir, qid);
1487                 gso_ctx = NULL;
1488         } else if (*other_fd != -1) {
1489                 /* Only other_fd exists. dup it */
1490                 *fd = dup(*other_fd);
1491                 if (*fd < 0) {
1492                         *fd = -1;
1493                         TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1494                         return -1;
1495                 }
1496                 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1497                         pmd->name, *other_fd, dir, qid, *fd);
1498         } else {
1499                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1500                 *fd = tun_alloc(pmd, 0);
1501                 if (*fd < 0) {
1502                         *fd = -1; /* restore original value */
1503                         TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1504                         return -1;
1505                 }
1506                 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1507                         pmd->name, dir, qid, *fd);
1508         }
1509
1510         tx->mtu = &dev->data->mtu;
1511         rx->rxmode = &dev->data->dev_conf.rxmode;
1512         if (gso_ctx) {
1513                 ret = tap_gso_ctx_setup(gso_ctx, dev);
1514                 if (ret)
1515                         return -1;
1516         }
1517
1518         tx->type = pmd->type;
1519
1520         return *fd;
1521 }
1522
1523 static int
1524 tap_rx_queue_setup(struct rte_eth_dev *dev,
1525                    uint16_t rx_queue_id,
1526                    uint16_t nb_rx_desc,
1527                    unsigned int socket_id,
1528                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1529                    struct rte_mempool *mp)
1530 {
1531         struct pmd_internals *internals = dev->data->dev_private;
1532         struct pmd_process_private *process_private = dev->process_private;
1533         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1534         struct rte_mbuf **tmp = &rxq->pool;
1535         long iov_max = sysconf(_SC_IOV_MAX);
1536
1537         if (iov_max <= 0) {
1538                 TAP_LOG(WARNING,
1539                         "_SC_IOV_MAX is not defined. Using %d as default",
1540                         TAP_IOV_DEFAULT_MAX);
1541                 iov_max = TAP_IOV_DEFAULT_MAX;
1542         }
1543         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1544         struct iovec (*iovecs)[nb_desc + 1];
1545         int data_off = RTE_PKTMBUF_HEADROOM;
1546         int ret = 0;
1547         int fd;
1548         int i;
1549
1550         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1551                 TAP_LOG(WARNING,
1552                         "nb_rx_queues %d too small or mempool NULL",
1553                         dev->data->nb_rx_queues);
1554                 return -1;
1555         }
1556
1557         rxq->mp = mp;
1558         rxq->trigger_seen = 1; /* force initial burst */
1559         rxq->in_port = dev->data->port_id;
1560         rxq->queue_id = rx_queue_id;
1561         rxq->nb_rx_desc = nb_desc;
1562         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1563                                     socket_id);
1564         if (!iovecs) {
1565                 TAP_LOG(WARNING,
1566                         "%s: Couldn't allocate %d RX descriptors",
1567                         dev->device->name, nb_desc);
1568                 return -ENOMEM;
1569         }
1570         rxq->iovecs = iovecs;
1571
1572         dev->data->rx_queues[rx_queue_id] = rxq;
1573         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1574         if (fd == -1) {
1575                 ret = fd;
1576                 goto error;
1577         }
1578
1579         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1580         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1581
1582         for (i = 1; i <= nb_desc; i++) {
1583                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1584                 if (!*tmp) {
1585                         TAP_LOG(WARNING,
1586                                 "%s: couldn't allocate memory for queue %d",
1587                                 dev->device->name, rx_queue_id);
1588                         ret = -ENOMEM;
1589                         goto error;
1590                 }
1591                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1592                 (*rxq->iovecs)[i].iov_base =
1593                         (char *)(*tmp)->buf_addr + data_off;
1594                 data_off = 0;
1595                 tmp = &(*tmp)->next;
1596         }
1597
1598         TAP_LOG(DEBUG, "  RX TUNTAP device name %s, qid %d on fd %d",
1599                 internals->name, rx_queue_id,
1600                 process_private->rxq_fds[rx_queue_id]);
1601
1602         return 0;
1603
1604 error:
1605         tap_rxq_pool_free(rxq->pool);
1606         rxq->pool = NULL;
1607         rte_free(rxq->iovecs);
1608         rxq->iovecs = NULL;
1609         return ret;
1610 }
1611
1612 static int
1613 tap_tx_queue_setup(struct rte_eth_dev *dev,
1614                    uint16_t tx_queue_id,
1615                    uint16_t nb_tx_desc __rte_unused,
1616                    unsigned int socket_id __rte_unused,
1617                    const struct rte_eth_txconf *tx_conf)
1618 {
1619         struct pmd_internals *internals = dev->data->dev_private;
1620         struct pmd_process_private *process_private = dev->process_private;
1621         struct tx_queue *txq;
1622         int ret;
1623         uint64_t offloads;
1624
1625         if (tx_queue_id >= dev->data->nb_tx_queues)
1626                 return -1;
1627         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1628         txq = dev->data->tx_queues[tx_queue_id];
1629         txq->out_port = dev->data->port_id;
1630         txq->queue_id = tx_queue_id;
1631
1632         offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1633         txq->csum = !!(offloads &
1634                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
1635                          DEV_TX_OFFLOAD_UDP_CKSUM |
1636                          DEV_TX_OFFLOAD_TCP_CKSUM));
1637
1638         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1639         if (ret == -1)
1640                 return -1;
1641         TAP_LOG(DEBUG,
1642                 "  TX TUNTAP device name %s, qid %d on fd %d csum %s",
1643                 internals->name, tx_queue_id,
1644                 process_private->txq_fds[tx_queue_id],
1645                 txq->csum ? "on" : "off");
1646
1647         return 0;
1648 }
1649
1650 static int
1651 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1652 {
1653         struct pmd_internals *pmd = dev->data->dev_private;
1654         struct ifreq ifr = { .ifr_mtu = mtu };
1655         int err = 0;
1656
1657         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1658         if (!err)
1659                 dev->data->mtu = mtu;
1660
1661         return err;
1662 }
1663
1664 static int
1665 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1666                      struct rte_ether_addr *mc_addr_set __rte_unused,
1667                      uint32_t nb_mc_addr __rte_unused)
1668 {
1669         /*
1670          * Nothing to do actually: the tap has no filtering whatsoever, every
1671          * packet is received.
1672          */
1673         return 0;
1674 }
1675
1676 static int
1677 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1678 {
1679         struct rte_eth_dev *dev = arg;
1680         struct pmd_internals *pmd = dev->data->dev_private;
1681         struct ifinfomsg *info = NLMSG_DATA(nh);
1682
1683         if (nh->nlmsg_type != RTM_NEWLINK ||
1684             (info->ifi_index != pmd->if_index &&
1685              info->ifi_index != pmd->remote_if_index))
1686                 return 0;
1687         return tap_link_update(dev, 0);
1688 }
1689
1690 static void
1691 tap_dev_intr_handler(void *cb_arg)
1692 {
1693         struct rte_eth_dev *dev = cb_arg;
1694         struct pmd_internals *pmd = dev->data->dev_private;
1695
1696         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1697 }
1698
1699 static int
1700 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1701 {
1702         struct pmd_internals *pmd = dev->data->dev_private;
1703         int ret;
1704
1705         /* In any case, disable interrupt if the conf is no longer there. */
1706         if (!dev->data->dev_conf.intr_conf.lsc) {
1707                 if (pmd->intr_handle.fd != -1) {
1708                         goto clean;
1709                 }
1710                 return 0;
1711         }
1712         if (set) {
1713                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1714                 if (unlikely(pmd->intr_handle.fd == -1))
1715                         return -EBADF;
1716                 return rte_intr_callback_register(
1717                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1718         }
1719
1720 clean:
1721         do {
1722                 ret = rte_intr_callback_unregister(&pmd->intr_handle,
1723                         tap_dev_intr_handler, dev);
1724                 if (ret >= 0) {
1725                         break;
1726                 } else if (ret == -EAGAIN) {
1727                         rte_delay_ms(100);
1728                 } else {
1729                         TAP_LOG(ERR, "intr callback unregister failed: %d",
1730                                      ret);
1731                         break;
1732                 }
1733         } while (true);
1734
1735         tap_nl_final(pmd->intr_handle.fd);
1736         pmd->intr_handle.fd = -1;
1737
1738         return 0;
1739 }
1740
1741 static int
1742 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1743 {
1744         int err;
1745
1746         err = tap_lsc_intr_handle_set(dev, set);
1747         if (err < 0) {
1748                 if (!set)
1749                         tap_rx_intr_vec_set(dev, 0);
1750                 return err;
1751         }
1752         err = tap_rx_intr_vec_set(dev, set);
1753         if (err && set)
1754                 tap_lsc_intr_handle_set(dev, 0);
1755         return err;
1756 }
1757
1758 static const uint32_t*
1759 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1760 {
1761         static const uint32_t ptypes[] = {
1762                 RTE_PTYPE_INNER_L2_ETHER,
1763                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1764                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1765                 RTE_PTYPE_INNER_L3_IPV4,
1766                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1767                 RTE_PTYPE_INNER_L3_IPV6,
1768                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1769                 RTE_PTYPE_INNER_L4_FRAG,
1770                 RTE_PTYPE_INNER_L4_UDP,
1771                 RTE_PTYPE_INNER_L4_TCP,
1772                 RTE_PTYPE_INNER_L4_SCTP,
1773                 RTE_PTYPE_L2_ETHER,
1774                 RTE_PTYPE_L2_ETHER_VLAN,
1775                 RTE_PTYPE_L2_ETHER_QINQ,
1776                 RTE_PTYPE_L3_IPV4,
1777                 RTE_PTYPE_L3_IPV4_EXT,
1778                 RTE_PTYPE_L3_IPV6_EXT,
1779                 RTE_PTYPE_L3_IPV6,
1780                 RTE_PTYPE_L4_FRAG,
1781                 RTE_PTYPE_L4_UDP,
1782                 RTE_PTYPE_L4_TCP,
1783                 RTE_PTYPE_L4_SCTP,
1784         };
1785
1786         return ptypes;
1787 }
1788
1789 static int
1790 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1791                   struct rte_eth_fc_conf *fc_conf)
1792 {
1793         fc_conf->mode = RTE_FC_NONE;
1794         return 0;
1795 }
1796
1797 static int
1798 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1799                   struct rte_eth_fc_conf *fc_conf)
1800 {
1801         if (fc_conf->mode != RTE_FC_NONE)
1802                 return -ENOTSUP;
1803         return 0;
1804 }
1805
1806 /**
1807  * DPDK callback to update the RSS hash configuration.
1808  *
1809  * @param dev
1810  *   Pointer to Ethernet device structure.
1811  * @param[in] rss_conf
1812  *   RSS configuration data.
1813  *
1814  * @return
1815  *   0 on success, a negative errno value otherwise and rte_errno is set.
1816  */
1817 static int
1818 tap_rss_hash_update(struct rte_eth_dev *dev,
1819                 struct rte_eth_rss_conf *rss_conf)
1820 {
1821         if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1822                 rte_errno = EINVAL;
1823                 return -rte_errno;
1824         }
1825         if (rss_conf->rss_key && rss_conf->rss_key_len) {
1826                 /*
1827                  * Currently TAP RSS key is hard coded
1828                  * and cannot be updated
1829                  */
1830                 TAP_LOG(ERR,
1831                         "port %u RSS key cannot be updated",
1832                         dev->data->port_id);
1833                 rte_errno = EINVAL;
1834                 return -rte_errno;
1835         }
1836         return 0;
1837 }
1838
1839 static int
1840 tap_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1841 {
1842         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1843
1844         return 0;
1845 }
1846
1847 static int
1848 tap_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1849 {
1850         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1851
1852         return 0;
1853 }
1854
1855 static int
1856 tap_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1857 {
1858         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1859
1860         return 0;
1861 }
1862
1863 static int
1864 tap_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1865 {
1866         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1867
1868         return 0;
1869 }
1870 static const struct eth_dev_ops ops = {
1871         .dev_start              = tap_dev_start,
1872         .dev_stop               = tap_dev_stop,
1873         .dev_close              = tap_dev_close,
1874         .dev_configure          = tap_dev_configure,
1875         .dev_infos_get          = tap_dev_info,
1876         .rx_queue_setup         = tap_rx_queue_setup,
1877         .tx_queue_setup         = tap_tx_queue_setup,
1878         .rx_queue_start         = tap_rx_queue_start,
1879         .tx_queue_start         = tap_tx_queue_start,
1880         .rx_queue_stop          = tap_rx_queue_stop,
1881         .tx_queue_stop          = tap_tx_queue_stop,
1882         .rx_queue_release       = tap_rx_queue_release,
1883         .tx_queue_release       = tap_tx_queue_release,
1884         .flow_ctrl_get          = tap_flow_ctrl_get,
1885         .flow_ctrl_set          = tap_flow_ctrl_set,
1886         .link_update            = tap_link_update,
1887         .dev_set_link_up        = tap_link_set_up,
1888         .dev_set_link_down      = tap_link_set_down,
1889         .promiscuous_enable     = tap_promisc_enable,
1890         .promiscuous_disable    = tap_promisc_disable,
1891         .allmulticast_enable    = tap_allmulti_enable,
1892         .allmulticast_disable   = tap_allmulti_disable,
1893         .mac_addr_set           = tap_mac_set,
1894         .mtu_set                = tap_mtu_set,
1895         .set_mc_addr_list       = tap_set_mc_addr_list,
1896         .stats_get              = tap_stats_get,
1897         .stats_reset            = tap_stats_reset,
1898         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1899         .rss_hash_update        = tap_rss_hash_update,
1900         .flow_ops_get           = tap_dev_flow_ops_get,
1901 };
1902
1903 static int
1904 eth_dev_tap_create(struct rte_vdev_device *vdev, const char *tap_name,
1905                    char *remote_iface, struct rte_ether_addr *mac_addr,
1906                    enum rte_tuntap_type type)
1907 {
1908         int numa_node = rte_socket_id();
1909         struct rte_eth_dev *dev;
1910         struct pmd_internals *pmd;
1911         struct pmd_process_private *process_private;
1912         const char *tuntap_name = tuntap_types[type];
1913         struct rte_eth_dev_data *data;
1914         struct ifreq ifr;
1915         int i;
1916
1917         TAP_LOG(DEBUG, "%s device on numa %u", tuntap_name, rte_socket_id());
1918
1919         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1920         if (!dev) {
1921                 TAP_LOG(ERR, "%s Unable to allocate device struct",
1922                                 tuntap_name);
1923                 goto error_exit_nodev;
1924         }
1925
1926         process_private = (struct pmd_process_private *)
1927                 rte_zmalloc_socket(tap_name, sizeof(struct pmd_process_private),
1928                         RTE_CACHE_LINE_SIZE, dev->device->numa_node);
1929
1930         if (process_private == NULL) {
1931                 TAP_LOG(ERR, "Failed to alloc memory for process private");
1932                 return -1;
1933         }
1934         pmd = dev->data->dev_private;
1935         dev->process_private = process_private;
1936         pmd->dev = dev;
1937         strlcpy(pmd->name, tap_name, sizeof(pmd->name));
1938         pmd->type = type;
1939         pmd->ka_fd = -1;
1940         pmd->nlsk_fd = -1;
1941         pmd->gso_ctx_mp = NULL;
1942
1943         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1944         if (pmd->ioctl_sock == -1) {
1945                 TAP_LOG(ERR,
1946                         "%s Unable to get a socket for management: %s",
1947                         tuntap_name, strerror(errno));
1948                 goto error_exit;
1949         }
1950
1951         /* Setup some default values */
1952         data = dev->data;
1953         data->dev_private = pmd;
1954         data->dev_flags = RTE_ETH_DEV_INTR_LSC |
1955                                 RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1956         data->numa_node = numa_node;
1957
1958         data->dev_link = pmd_link;
1959         data->mac_addrs = &pmd->eth_addr;
1960         /* Set the number of RX and TX queues */
1961         data->nb_rx_queues = 0;
1962         data->nb_tx_queues = 0;
1963
1964         dev->dev_ops = &ops;
1965         dev->rx_pkt_burst = pmd_rx_burst;
1966         dev->tx_pkt_burst = pmd_tx_burst;
1967
1968         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1969         pmd->intr_handle.fd = -1;
1970         dev->intr_handle = &pmd->intr_handle;
1971
1972         /* Presetup the fds to -1 as being not valid */
1973         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1974                 process_private->rxq_fds[i] = -1;
1975                 process_private->txq_fds[i] = -1;
1976         }
1977
1978         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1979                 if (rte_is_zero_ether_addr(mac_addr))
1980                         rte_eth_random_addr((uint8_t *)&pmd->eth_addr);
1981                 else
1982                         rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1983         }
1984
1985         /*
1986          * Allocate a TUN device keep-alive file descriptor that will only be
1987          * closed when the TUN device itself is closed or removed.
1988          * This keep-alive file descriptor will guarantee that the TUN device
1989          * exists even when all of its queues are closed
1990          */
1991         pmd->ka_fd = tun_alloc(pmd, 1);
1992         if (pmd->ka_fd == -1) {
1993                 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
1994                 goto error_exit;
1995         }
1996         TAP_LOG(DEBUG, "allocated %s", pmd->name);
1997
1998         ifr.ifr_mtu = dev->data->mtu;
1999         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
2000                 goto error_exit;
2001
2002         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
2003                 memset(&ifr, 0, sizeof(struct ifreq));
2004                 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
2005                 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
2006                                 RTE_ETHER_ADDR_LEN);
2007                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
2008                         goto error_exit;
2009         }
2010
2011         /*
2012          * Set up everything related to rte_flow:
2013          * - netlink socket
2014          * - tap / remote if_index
2015          * - mandatory QDISCs
2016          * - rte_flow actual/implicit lists
2017          * - implicit rules
2018          */
2019         pmd->nlsk_fd = tap_nl_init(0);
2020         if (pmd->nlsk_fd == -1) {
2021                 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
2022                         pmd->name);
2023                 goto disable_rte_flow;
2024         }
2025         pmd->if_index = if_nametoindex(pmd->name);
2026         if (!pmd->if_index) {
2027                 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
2028                 goto disable_rte_flow;
2029         }
2030         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
2031                 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
2032                         pmd->name);
2033                 goto disable_rte_flow;
2034         }
2035         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
2036                 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
2037                         pmd->name);
2038                 goto disable_rte_flow;
2039         }
2040         LIST_INIT(&pmd->flows);
2041
2042         if (strlen(remote_iface)) {
2043                 pmd->remote_if_index = if_nametoindex(remote_iface);
2044                 if (!pmd->remote_if_index) {
2045                         TAP_LOG(ERR, "%s: failed to get %s if_index.",
2046                                 pmd->name, remote_iface);
2047                         goto error_remote;
2048                 }
2049                 strlcpy(pmd->remote_iface, remote_iface, RTE_ETH_NAME_MAX_LEN);
2050
2051                 /* Save state of remote device */
2052                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
2053
2054                 /* Replicate remote MAC address */
2055                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
2056                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
2057                                 pmd->name, pmd->remote_iface);
2058                         goto error_remote;
2059                 }
2060                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
2061                            RTE_ETHER_ADDR_LEN);
2062                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
2063                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
2064                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
2065                                 pmd->name, remote_iface);
2066                         goto error_remote;
2067                 }
2068
2069                 /*
2070                  * Flush usually returns negative value because it tries to
2071                  * delete every QDISC (and on a running device, one QDISC at
2072                  * least is needed). Ignore negative return value.
2073                  */
2074                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
2075                 if (qdisc_create_ingress(pmd->nlsk_fd,
2076                                          pmd->remote_if_index) < 0) {
2077                         TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
2078                                 pmd->remote_iface);
2079                         goto error_remote;
2080                 }
2081                 LIST_INIT(&pmd->implicit_flows);
2082                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
2083                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
2084                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
2085                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
2086                         TAP_LOG(ERR,
2087                                 "%s: failed to create implicit rules.",
2088                                 pmd->name);
2089                         goto error_remote;
2090                 }
2091         }
2092
2093         rte_eth_dev_probing_finish(dev);
2094         return 0;
2095
2096 disable_rte_flow:
2097         TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
2098                 strerror(errno), errno);
2099         if (strlen(remote_iface)) {
2100                 TAP_LOG(ERR, "Remote feature requires flow support.");
2101                 goto error_exit;
2102         }
2103         rte_eth_dev_probing_finish(dev);
2104         return 0;
2105
2106 error_remote:
2107         TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
2108                 strerror(errno), errno);
2109         tap_flow_implicit_flush(pmd, NULL);
2110
2111 error_exit:
2112         if (pmd->nlsk_fd != -1)
2113                 close(pmd->nlsk_fd);
2114         if (pmd->ka_fd != -1)
2115                 close(pmd->ka_fd);
2116         if (pmd->ioctl_sock != -1)
2117                 close(pmd->ioctl_sock);
2118         /* mac_addrs must not be freed alone because part of dev_private */
2119         dev->data->mac_addrs = NULL;
2120         rte_eth_dev_release_port(dev);
2121
2122 error_exit_nodev:
2123         TAP_LOG(ERR, "%s Unable to initialize %s",
2124                 tuntap_name, rte_vdev_device_name(vdev));
2125
2126         return -EINVAL;
2127 }
2128
2129 /* make sure name is a possible Linux network device name */
2130 static bool
2131 is_valid_iface(const char *name)
2132 {
2133         if (*name == '\0')
2134                 return false;
2135
2136         if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
2137                 return false;
2138
2139         while (*name) {
2140                 if (*name == '/' || *name == ':' || isspace(*name))
2141                         return false;
2142                 name++;
2143         }
2144         return true;
2145 }
2146
2147 static int
2148 set_interface_name(const char *key __rte_unused,
2149                    const char *value,
2150                    void *extra_args)
2151 {
2152         char *name = (char *)extra_args;
2153
2154         if (value) {
2155                 if (!is_valid_iface(value)) {
2156                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2157                                 value);
2158                         return -1;
2159                 }
2160                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2161         } else {
2162                 /* use tap%d which causes kernel to choose next available */
2163                 strlcpy(name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2164         }
2165         return 0;
2166 }
2167
2168 static int
2169 set_remote_iface(const char *key __rte_unused,
2170                  const char *value,
2171                  void *extra_args)
2172 {
2173         char *name = (char *)extra_args;
2174
2175         if (value) {
2176                 if (!is_valid_iface(value)) {
2177                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2178                                 value);
2179                         return -1;
2180                 }
2181                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2182         }
2183
2184         return 0;
2185 }
2186
2187 static int parse_user_mac(struct rte_ether_addr *user_mac,
2188                 const char *value)
2189 {
2190         unsigned int index = 0;
2191         char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
2192
2193         if (user_mac == NULL || value == NULL)
2194                 return 0;
2195
2196         strlcpy(mac_temp, value, sizeof(mac_temp));
2197         mac_byte = strtok(mac_temp, ":");
2198
2199         while ((mac_byte != NULL) &&
2200                         (strlen(mac_byte) <= 2) &&
2201                         (strlen(mac_byte) == strspn(mac_byte,
2202                                         ETH_TAP_CMP_MAC_FMT))) {
2203                 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
2204                 mac_byte = strtok(NULL, ":");
2205         }
2206
2207         return index;
2208 }
2209
2210 static int
2211 set_mac_type(const char *key __rte_unused,
2212              const char *value,
2213              void *extra_args)
2214 {
2215         struct rte_ether_addr *user_mac = extra_args;
2216
2217         if (!value)
2218                 return 0;
2219
2220         if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
2221                 static int iface_idx;
2222
2223                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
2224                 memcpy((char *)user_mac->addr_bytes, "\0dtap",
2225                         RTE_ETHER_ADDR_LEN);
2226                 user_mac->addr_bytes[RTE_ETHER_ADDR_LEN - 1] =
2227                         iface_idx++ + '0';
2228                 goto success;
2229         }
2230
2231         if (parse_user_mac(user_mac, value) != 6)
2232                 goto error;
2233 success:
2234         TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
2235         return 0;
2236
2237 error:
2238         TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
2239                 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
2240         return -1;
2241 }
2242
2243 /*
2244  * Open a TUN interface device. TUN PMD
2245  * 1) sets tap_type as false
2246  * 2) intakes iface as argument.
2247  * 3) as interface is virtual set speed to 10G
2248  */
2249 static int
2250 rte_pmd_tun_probe(struct rte_vdev_device *dev)
2251 {
2252         const char *name, *params;
2253         int ret;
2254         struct rte_kvargs *kvlist = NULL;
2255         char tun_name[RTE_ETH_NAME_MAX_LEN];
2256         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2257         struct rte_eth_dev *eth_dev;
2258
2259         name = rte_vdev_device_name(dev);
2260         params = rte_vdev_device_args(dev);
2261         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2262
2263         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
2264             strlen(params) == 0) {
2265                 eth_dev = rte_eth_dev_attach_secondary(name);
2266                 if (!eth_dev) {
2267                         TAP_LOG(ERR, "Failed to probe %s", name);
2268                         return -1;
2269                 }
2270                 eth_dev->dev_ops = &ops;
2271                 eth_dev->device = &dev->device;
2272                 rte_eth_dev_probing_finish(eth_dev);
2273                 return 0;
2274         }
2275
2276         /* use tun%d which causes kernel to choose next available */
2277         strlcpy(tun_name, DEFAULT_TUN_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2278
2279         if (params && (params[0] != '\0')) {
2280                 TAP_LOG(DEBUG, "parameters (%s)", params);
2281
2282                 kvlist = rte_kvargs_parse(params, valid_arguments);
2283                 if (kvlist) {
2284                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2285                                 ret = rte_kvargs_process(kvlist,
2286                                         ETH_TAP_IFACE_ARG,
2287                                         &set_interface_name,
2288                                         tun_name);
2289
2290                                 if (ret == -1)
2291                                         goto leave;
2292                         }
2293                 }
2294         }
2295         pmd_link.link_speed = ETH_SPEED_NUM_10G;
2296
2297         TAP_LOG(DEBUG, "Initializing pmd_tun for %s", name);
2298
2299         ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0,
2300                                  ETH_TUNTAP_TYPE_TUN);
2301
2302 leave:
2303         if (ret == -1) {
2304                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2305                         name, tun_name);
2306         }
2307         rte_kvargs_free(kvlist);
2308
2309         return ret;
2310 }
2311
2312 /* Request queue file descriptors from secondary to primary. */
2313 static int
2314 tap_mp_attach_queues(const char *port_name, struct rte_eth_dev *dev)
2315 {
2316         int ret;
2317         struct timespec timeout = {.tv_sec = 1, .tv_nsec = 0};
2318         struct rte_mp_msg request, *reply;
2319         struct rte_mp_reply replies;
2320         struct ipc_queues *request_param = (struct ipc_queues *)request.param;
2321         struct ipc_queues *reply_param;
2322         struct pmd_process_private *process_private = dev->process_private;
2323         int queue, fd_iterator;
2324
2325         /* Prepare the request */
2326         memset(&request, 0, sizeof(request));
2327         strlcpy(request.name, TAP_MP_KEY, sizeof(request.name));
2328         strlcpy(request_param->port_name, port_name,
2329                 sizeof(request_param->port_name));
2330         request.len_param = sizeof(*request_param);
2331         /* Send request and receive reply */
2332         ret = rte_mp_request_sync(&request, &replies, &timeout);
2333         if (ret < 0 || replies.nb_received != 1) {
2334                 TAP_LOG(ERR, "Failed to request queues from primary: %d",
2335                         rte_errno);
2336                 return -1;
2337         }
2338         reply = &replies.msgs[0];
2339         reply_param = (struct ipc_queues *)reply->param;
2340         TAP_LOG(DEBUG, "Received IPC reply for %s", reply_param->port_name);
2341
2342         /* Attach the queues from received file descriptors */
2343         if (reply_param->rxq_count + reply_param->txq_count != reply->num_fds) {
2344                 TAP_LOG(ERR, "Unexpected number of fds received");
2345                 return -1;
2346         }
2347
2348         dev->data->nb_rx_queues = reply_param->rxq_count;
2349         dev->data->nb_tx_queues = reply_param->txq_count;
2350         fd_iterator = 0;
2351         for (queue = 0; queue < reply_param->rxq_count; queue++)
2352                 process_private->rxq_fds[queue] = reply->fds[fd_iterator++];
2353         for (queue = 0; queue < reply_param->txq_count; queue++)
2354                 process_private->txq_fds[queue] = reply->fds[fd_iterator++];
2355         free(reply);
2356         return 0;
2357 }
2358
2359 /* Send the queue file descriptors from the primary process to secondary. */
2360 static int
2361 tap_mp_sync_queues(const struct rte_mp_msg *request, const void *peer)
2362 {
2363         struct rte_eth_dev *dev;
2364         struct pmd_process_private *process_private;
2365         struct rte_mp_msg reply;
2366         const struct ipc_queues *request_param =
2367                 (const struct ipc_queues *)request->param;
2368         struct ipc_queues *reply_param =
2369                 (struct ipc_queues *)reply.param;
2370         uint16_t port_id;
2371         int queue;
2372         int ret;
2373
2374         /* Get requested port */
2375         TAP_LOG(DEBUG, "Received IPC request for %s", request_param->port_name);
2376         ret = rte_eth_dev_get_port_by_name(request_param->port_name, &port_id);
2377         if (ret) {
2378                 TAP_LOG(ERR, "Failed to get port id for %s",
2379                         request_param->port_name);
2380                 return -1;
2381         }
2382         dev = &rte_eth_devices[port_id];
2383         process_private = dev->process_private;
2384
2385         /* Fill file descriptors for all queues */
2386         reply.num_fds = 0;
2387         reply_param->rxq_count = 0;
2388         if (dev->data->nb_rx_queues + dev->data->nb_tx_queues >
2389                         RTE_MP_MAX_FD_NUM){
2390                 TAP_LOG(ERR, "Number of rx/tx queues exceeds max number of fds");
2391                 return -1;
2392         }
2393
2394         for (queue = 0; queue < dev->data->nb_rx_queues; queue++) {
2395                 reply.fds[reply.num_fds++] = process_private->rxq_fds[queue];
2396                 reply_param->rxq_count++;
2397         }
2398         RTE_ASSERT(reply_param->rxq_count == dev->data->nb_rx_queues);
2399
2400         reply_param->txq_count = 0;
2401         for (queue = 0; queue < dev->data->nb_tx_queues; queue++) {
2402                 reply.fds[reply.num_fds++] = process_private->txq_fds[queue];
2403                 reply_param->txq_count++;
2404         }
2405         RTE_ASSERT(reply_param->txq_count == dev->data->nb_tx_queues);
2406
2407         /* Send reply */
2408         strlcpy(reply.name, request->name, sizeof(reply.name));
2409         strlcpy(reply_param->port_name, request_param->port_name,
2410                 sizeof(reply_param->port_name));
2411         reply.len_param = sizeof(*reply_param);
2412         if (rte_mp_reply(&reply, peer) < 0) {
2413                 TAP_LOG(ERR, "Failed to reply an IPC request to sync queues");
2414                 return -1;
2415         }
2416         return 0;
2417 }
2418
2419 /* Open a TAP interface device.
2420  */
2421 static int
2422 rte_pmd_tap_probe(struct rte_vdev_device *dev)
2423 {
2424         const char *name, *params;
2425         int ret;
2426         struct rte_kvargs *kvlist = NULL;
2427         int speed;
2428         char tap_name[RTE_ETH_NAME_MAX_LEN];
2429         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2430         struct rte_ether_addr user_mac = { .addr_bytes = {0} };
2431         struct rte_eth_dev *eth_dev;
2432         int tap_devices_count_increased = 0;
2433
2434         name = rte_vdev_device_name(dev);
2435         params = rte_vdev_device_args(dev);
2436
2437         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
2438                 eth_dev = rte_eth_dev_attach_secondary(name);
2439                 if (!eth_dev) {
2440                         TAP_LOG(ERR, "Failed to probe %s", name);
2441                         return -1;
2442                 }
2443                 eth_dev->dev_ops = &ops;
2444                 eth_dev->device = &dev->device;
2445                 eth_dev->rx_pkt_burst = pmd_rx_burst;
2446                 eth_dev->tx_pkt_burst = pmd_tx_burst;
2447                 if (!rte_eal_primary_proc_alive(NULL)) {
2448                         TAP_LOG(ERR, "Primary process is missing");
2449                         return -1;
2450                 }
2451                 eth_dev->process_private = (struct pmd_process_private *)
2452                         rte_zmalloc_socket(name,
2453                                 sizeof(struct pmd_process_private),
2454                                 RTE_CACHE_LINE_SIZE,
2455                                 eth_dev->device->numa_node);
2456                 if (eth_dev->process_private == NULL) {
2457                         TAP_LOG(ERR,
2458                                 "Failed to alloc memory for process private");
2459                         return -1;
2460                 }
2461
2462                 ret = tap_mp_attach_queues(name, eth_dev);
2463                 if (ret != 0)
2464                         return -1;
2465                 rte_eth_dev_probing_finish(eth_dev);
2466                 return 0;
2467         }
2468
2469         speed = ETH_SPEED_NUM_10G;
2470
2471         /* use tap%d which causes kernel to choose next available */
2472         strlcpy(tap_name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2473         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2474
2475         if (params && (params[0] != '\0')) {
2476                 TAP_LOG(DEBUG, "parameters (%s)", params);
2477
2478                 kvlist = rte_kvargs_parse(params, valid_arguments);
2479                 if (kvlist) {
2480                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2481                                 ret = rte_kvargs_process(kvlist,
2482                                                          ETH_TAP_IFACE_ARG,
2483                                                          &set_interface_name,
2484                                                          tap_name);
2485                                 if (ret == -1)
2486                                         goto leave;
2487                         }
2488
2489                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
2490                                 ret = rte_kvargs_process(kvlist,
2491                                                          ETH_TAP_REMOTE_ARG,
2492                                                          &set_remote_iface,
2493                                                          remote_iface);
2494                                 if (ret == -1)
2495                                         goto leave;
2496                         }
2497
2498                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
2499                                 ret = rte_kvargs_process(kvlist,
2500                                                          ETH_TAP_MAC_ARG,
2501                                                          &set_mac_type,
2502                                                          &user_mac);
2503                                 if (ret == -1)
2504                                         goto leave;
2505                         }
2506                 }
2507         }
2508         pmd_link.link_speed = speed;
2509
2510         TAP_LOG(DEBUG, "Initializing pmd_tap for %s", name);
2511
2512         /* Register IPC feed callback */
2513         if (!tap_devices_count) {
2514                 ret = rte_mp_action_register(TAP_MP_KEY, tap_mp_sync_queues);
2515                 if (ret < 0 && rte_errno != ENOTSUP) {
2516                         TAP_LOG(ERR, "tap: Failed to register IPC callback: %s",
2517                                 strerror(rte_errno));
2518                         goto leave;
2519                 }
2520         }
2521         tap_devices_count++;
2522         tap_devices_count_increased = 1;
2523         ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
2524                 ETH_TUNTAP_TYPE_TAP);
2525
2526 leave:
2527         if (ret == -1) {
2528                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2529                         name, tap_name);
2530                 if (tap_devices_count_increased == 1) {
2531                         if (tap_devices_count == 1)
2532                                 rte_mp_action_unregister(TAP_MP_KEY);
2533                         tap_devices_count--;
2534                 }
2535         }
2536         rte_kvargs_free(kvlist);
2537
2538         return ret;
2539 }
2540
2541 /* detach a TUNTAP device.
2542  */
2543 static int
2544 rte_pmd_tap_remove(struct rte_vdev_device *dev)
2545 {
2546         struct rte_eth_dev *eth_dev = NULL;
2547
2548         /* find the ethdev entry */
2549         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
2550         if (!eth_dev)
2551                 return 0;
2552
2553         tap_dev_close(eth_dev);
2554         rte_eth_dev_release_port(eth_dev);
2555
2556         return 0;
2557 }
2558
2559 static struct rte_vdev_driver pmd_tun_drv = {
2560         .probe = rte_pmd_tun_probe,
2561         .remove = rte_pmd_tap_remove,
2562 };
2563
2564 static struct rte_vdev_driver pmd_tap_drv = {
2565         .probe = rte_pmd_tap_probe,
2566         .remove = rte_pmd_tap_remove,
2567 };
2568
2569 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
2570 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
2571 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
2572 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
2573                               ETH_TAP_IFACE_ARG "=<string> ");
2574 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
2575                               ETH_TAP_IFACE_ARG "=<string> "
2576                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
2577                               ETH_TAP_REMOTE_ARG "=<string>");
2578 RTE_LOG_REGISTER_DEFAULT(tap_logtype, NOTICE);