drivers/net/tap/rte_eth_tap.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2016-2017 Intel Corporation
   3  */
   4
   5 #include <rte_atomic.h>
   6 #include <rte_branch_prediction.h>
   7 #include <rte_byteorder.h>
   8 #include <rte_common.h>
   9 #include <rte_mbuf.h>
  10 #include <rte_ethdev_driver.h>
  11 #include <rte_ethdev_vdev.h>
  12 #include <rte_malloc.h>
  13 #include <rte_bus_vdev.h>
  14 #include <rte_kvargs.h>
  15 #include <rte_net.h>
  16 #include <rte_debug.h>
  17 #include <rte_ip.h>
  18 #include <rte_string_fns.h>
  19 #include <rte_ethdev.h>
  20 #include <rte_errno.h>
  21 #include <rte_cycles.h>
  22
  23 #include <sys/types.h>
  24 #include <sys/stat.h>
  25 #include <sys/socket.h>
  26 #include <sys/ioctl.h>
  27 #include <sys/utsname.h>
  28 #include <sys/mman.h>
  29 #include <errno.h>
  30 #include <signal.h>
  31 #include <stdbool.h>
  32 #include <stdint.h>
  33 #include <sys/uio.h>
  34 #include <unistd.h>
  35 #include <arpa/inet.h>
  36 #include <net/if.h>
  37 #include <linux/if_tun.h>
  38 #include <linux/if_ether.h>
  39 #include <fcntl.h>
  40 #include <ctype.h>
  41
  42 #include <tap_rss.h>
  43 #include <rte_eth_tap.h>
  44 #include <tap_flow.h>
  45 #include <tap_netlink.h>
  46 #include <tap_tcmsgs.h>
  47
  48 /* Linux based path to the TUN device */
  49 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  50 #define DEFAULT_TAP_NAME        "dtap"
  51 #define DEFAULT_TUN_NAME        "dtun"
  52
  53 #define ETH_TAP_IFACE_ARG       "iface"
  54 #define ETH_TAP_REMOTE_ARG      "remote"
  55 #define ETH_TAP_MAC_ARG         "mac"
  56 #define ETH_TAP_MAC_FIXED       "fixed"
  57
  58 #define ETH_TAP_USR_MAC_FMT     "xx:xx:xx:xx:xx:xx"
  59 #define ETH_TAP_CMP_MAC_FMT     "0123456789ABCDEFabcdef"
  60 #define ETH_TAP_MAC_ARG_FMT     ETH_TAP_MAC_FIXED "|" ETH_TAP_USR_MAC_FMT
  61
  62 #define TAP_GSO_MBUFS_PER_CORE  128
  63 #define TAP_GSO_MBUF_SEG_SIZE   128
  64 #define TAP_GSO_MBUF_CACHE_SIZE 4
  65 #define TAP_GSO_MBUFS_NUM \
  66         (TAP_GSO_MBUFS_PER_CORE * TAP_GSO_MBUF_CACHE_SIZE)
  67
  68 /* IPC key for queue fds sync */
  69 #define TAP_MP_KEY "tap_mp_sync_queues"
  70
  71 #define TAP_IOV_DEFAULT_MAX 1024
  72
  73 static int tap_devices_count;
  74
  75 static const char *valid_arguments[] = {
  76         ETH_TAP_IFACE_ARG,
  77         ETH_TAP_REMOTE_ARG,
  78         ETH_TAP_MAC_ARG,
  79         NULL
  80 };
  81
  82 static volatile uint32_t tap_trigger;   /* Rx trigger */
  83
  84 static struct rte_eth_link pmd_link = {
  85         .link_speed = ETH_SPEED_NUM_10G,
  86         .link_duplex = ETH_LINK_FULL_DUPLEX,
  87         .link_status = ETH_LINK_DOWN,
  88         .link_autoneg = ETH_LINK_FIXED,
  89 };
  90
  91 static void
  92 tap_trigger_cb(int sig __rte_unused)
  93 {
  94         /* Valid trigger values are nonzero */
  95         tap_trigger = (tap_trigger + 1) | 0x80000000;
  96 }
  97
  98 /* Specifies on what netdevices the ioctl should be applied */
  99 enum ioctl_mode {
 100         LOCAL_AND_REMOTE,
 101         LOCAL_ONLY,
 102         REMOTE_ONLY,
 103 };
 104
 105 /* Message header to synchronize queues via IPC */
 106 struct ipc_queues {
 107         char port_name[RTE_DEV_NAME_MAX_LEN];
 108         int rxq_count;
 109         int txq_count;
 110         /*
 111          * The file descriptors are in the dedicated part
 112          * of the Unix message to be translated by the kernel.
 113          */
 114 };
 115
 116 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
 117
 118 /**
 119  * Tun/Tap allocation routine
 120  *
 121  * @param[in] pmd
 122  *   Pointer to private structure.
 123  *
 124  * @param[in] is_keepalive
 125  *   Keepalive flag
 126  *
 127  * @return
 128  *   -1 on failure, fd on success
 129  */
 130 static int
 131 tun_alloc(struct pmd_internals *pmd, int is_keepalive)
 132 {
 133         struct ifreq ifr;
 134 #ifdef IFF_MULTI_QUEUE
 135         unsigned int features;
 136 #endif
 137         int fd, signo, flags;
 138
 139         memset(&ifr, 0, sizeof(struct ifreq));
 140
 141         /*
 142          * Do not set IFF_NO_PI as packet information header will be needed
 143          * to check if a received packet has been truncated.
 144          */
 145         ifr.ifr_flags = (pmd->type == ETH_TUNTAP_TYPE_TAP) ?
 146                 IFF_TAP : IFF_TUN | IFF_POINTOPOINT;
 147         strlcpy(ifr.ifr_name, pmd->name, IFNAMSIZ);
 148
 149         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 150         if (fd < 0) {
 151                 TAP_LOG(ERR, "Unable to open %s interface", TUN_TAP_DEV_PATH);
 152                 goto error;
 153         }
 154
 155 #ifdef IFF_MULTI_QUEUE
 156         /* Grab the TUN features to verify we can work multi-queue */
 157         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 158                 TAP_LOG(ERR, "unable to get TUN/TAP features");
 159                 goto error;
 160         }
 161         TAP_LOG(DEBUG, "%s Features %08x", TUN_TAP_DEV_PATH, features);
 162
 163         if (features & IFF_MULTI_QUEUE) {
 164                 TAP_LOG(DEBUG, "  Multi-queue support for %d queues",
 165                         RTE_PMD_TAP_MAX_QUEUES);
 166                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 167         } else
 168 #endif
 169         {
 170                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 171                 TAP_LOG(DEBUG, "  Single queue only support");
 172         }
 173
 174         /* Set the TUN/TAP configuration and set the name if needed */
 175         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 176                 TAP_LOG(WARNING, "Unable to set TUNSETIFF for %s: %s",
 177                         ifr.ifr_name, strerror(errno));
 178                 goto error;
 179         }
 180
 181         /*
 182          * Name passed to kernel might be wildcard like dtun%d
 183          * and need to find the resulting device.
 184          */
 185         TAP_LOG(DEBUG, "Device name is '%s'", ifr.ifr_name);
 186         strlcpy(pmd->name, ifr.ifr_name, RTE_ETH_NAME_MAX_LEN);
 187
 188         if (is_keepalive) {
 189                 /*
 190                  * Detach the TUN/TAP keep-alive queue
 191                  * to avoid traffic through it
 192                  */
 193                 ifr.ifr_flags = IFF_DETACH_QUEUE;
 194                 if (ioctl(fd, TUNSETQUEUE, (void *)&ifr) < 0) {
 195                         TAP_LOG(WARNING,
 196                                 "Unable to detach keep-alive queue for %s: %s",
 197                                 ifr.ifr_name, strerror(errno));
 198                         goto error;
 199                 }
 200         }
 201
 202         flags = fcntl(fd, F_GETFL);
 203         if (flags == -1) {
 204                 TAP_LOG(WARNING,
 205                         "Unable to get %s current flags\n",
 206                         ifr.ifr_name);
 207                 goto error;
 208         }
 209
 210         /* Always set the file descriptor to non-blocking */
 211         flags |= O_NONBLOCK;
 212         if (fcntl(fd, F_SETFL, flags) < 0) {
 213                 TAP_LOG(WARNING,
 214                         "Unable to set %s to nonblocking: %s",
 215                         ifr.ifr_name, strerror(errno));
 216                 goto error;
 217         }
 218
 219         /* Find a free realtime signal */
 220         for (signo = SIGRTMIN + 1; signo < SIGRTMAX; signo++) {
 221                 struct sigaction sa;
 222
 223                 if (sigaction(signo, NULL, &sa) == -1) {
 224                         TAP_LOG(WARNING,
 225                                 "Unable to get current rt-signal %d handler",
 226                                 signo);
 227                         goto error;
 228                 }
 229
 230                 /* Already have the handler we want on this signal  */
 231                 if (sa.sa_handler == tap_trigger_cb)
 232                         break;
 233
 234                 /* Is handler in use by application */
 235                 if (sa.sa_handler != SIG_DFL) {
 236                         TAP_LOG(DEBUG,
 237                                 "Skipping used rt-signal %d", signo);
 238                         continue;
 239                 }
 240
 241                 sa = (struct sigaction) {
 242                         .sa_flags = SA_RESTART,
 243                         .sa_handler = tap_trigger_cb,
 244                 };
 245
 246                 if (sigaction(signo, &sa, NULL) == -1) {
 247                         TAP_LOG(WARNING,
 248                                 "Unable to set rt-signal %d handler\n", signo);
 249                         goto error;
 250                 }
 251
 252                 /* Found a good signal to use */
 253                 TAP_LOG(DEBUG,
 254                         "Using rt-signal %d", signo);
 255                 break;
 256         }
 257
 258         if (signo == SIGRTMAX) {
 259                 TAP_LOG(WARNING, "All rt-signals are in use\n");
 260
 261                 /* Disable trigger globally in case of error */
 262                 tap_trigger = 0;
 263                 TAP_LOG(NOTICE, "No Rx trigger signal available\n");
 264         } else {
 265                 /* Enable signal on file descriptor */
 266                 if (fcntl(fd, F_SETSIG, signo) < 0) {
 267                         TAP_LOG(WARNING, "Unable to set signo %d for fd %d: %s",
 268                                 signo, fd, strerror(errno));
 269                         goto error;
 270                 }
 271                 if (fcntl(fd, F_SETFL, flags | O_ASYNC) < 0) {
 272                         TAP_LOG(WARNING, "Unable to set fcntl flags: %s",
 273                                 strerror(errno));
 274                         goto error;
 275                 }
 276
 277                 if (fcntl(fd, F_SETOWN, getpid()) < 0) {
 278                         TAP_LOG(WARNING, "Unable to set fcntl owner: %s",
 279                                 strerror(errno));
 280                         goto error;
 281                 }
 282         }
 283         return fd;
 284
 285 error:
 286         if (fd >= 0)
 287                 close(fd);
 288         return -1;
 289 }
 290
 291 static void
 292 tap_verify_csum(struct rte_mbuf *mbuf)
 293 {
 294         uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
 295         uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
 296         uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
 297         unsigned int l2_len = sizeof(struct rte_ether_hdr);
 298         unsigned int l3_len;
 299         uint16_t cksum = 0;
 300         void *l3_hdr;
 301         void *l4_hdr;
 302
 303         if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
 304                 l2_len += 4;
 305         else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
 306                 l2_len += 8;
 307         /* Don't verify checksum for packets with discontinuous L2 header */
 308         if (unlikely(l2_len + sizeof(struct rte_ipv4_hdr) >
 309                      rte_pktmbuf_data_len(mbuf)))
 310                 return;
 311         l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
 312         if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 313                 struct rte_ipv4_hdr *iph = l3_hdr;
 314
 315                 /* ihl contains the number of 4-byte words in the header */
 316                 l3_len = 4 * (iph->version_ihl & 0xf);
 317                 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
 318                         return;
 319                 /* check that the total length reported by header is not
 320                  * greater than the total received size
 321                  */
 322                 if (l2_len + rte_be_to_cpu_16(iph->total_length) >
 323                                 rte_pktmbuf_data_len(mbuf))
 324                         return;
 325
 326                 cksum = ~rte_raw_cksum(iph, l3_len);
 327                 mbuf->ol_flags |= cksum ?
 328                         PKT_RX_IP_CKSUM_BAD :
 329                         PKT_RX_IP_CKSUM_GOOD;
 330         } else if (l3 == RTE_PTYPE_L3_IPV6) {
 331                 struct rte_ipv6_hdr *iph = l3_hdr;
 332
 333                 l3_len = sizeof(struct rte_ipv6_hdr);
 334                 /* check that the total length reported by header is not
 335                  * greater than the total received size
 336                  */
 337                 if (l2_len + l3_len + rte_be_to_cpu_16(iph->payload_len) >
 338                                 rte_pktmbuf_data_len(mbuf))
 339                         return;
 340         } else {
 341                 /* IPv6 extensions are not supported */
 342                 return;
 343         }
 344         if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
 345                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 346                 /* Don't verify checksum for multi-segment packets. */
 347                 if (mbuf->nb_segs > 1)
 348                         return;
 349                 if (l3 == RTE_PTYPE_L3_IPV4)
 350                         cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
 351                 else if (l3 == RTE_PTYPE_L3_IPV6)
 352                         cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
 353                 mbuf->ol_flags |= cksum ?
 354                         PKT_RX_L4_CKSUM_BAD :
 355                         PKT_RX_L4_CKSUM_GOOD;
 356         }
 357 }
 358
 359 static uint64_t
 360 tap_rx_offload_get_port_capa(void)
 361 {
 362         /*
 363          * No specific port Rx offload capabilities.
 364          */
 365         return 0;
 366 }
 367
 368 static uint64_t
 369 tap_rx_offload_get_queue_capa(void)
 370 {
 371         return DEV_RX_OFFLOAD_SCATTER |
 372                DEV_RX_OFFLOAD_IPV4_CKSUM |
 373                DEV_RX_OFFLOAD_UDP_CKSUM |
 374                DEV_RX_OFFLOAD_TCP_CKSUM;
 375 }
 376
 377 static void
 378 tap_rxq_pool_free(struct rte_mbuf *pool)
 379 {
 380         struct rte_mbuf *mbuf = pool;
 381         uint16_t nb_segs = 1;
 382
 383         if (mbuf == NULL)
 384                 return;
 385
 386         while (mbuf->next) {
 387                 mbuf = mbuf->next;
 388                 nb_segs++;
 389         }
 390         pool->nb_segs = nb_segs;
 391         rte_pktmbuf_free(pool);
 392 }
 393
 394 /* Callback to handle the rx burst of packets to the correct interface and
 395  * file descriptor(s) in a multi-queue setup.
 396  */
 397 static uint16_t
 398 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 399 {
 400         struct rx_queue *rxq = queue;
 401         struct pmd_process_private *process_private;
 402         uint16_t num_rx;
 403         unsigned long num_rx_bytes = 0;
 404         uint32_t trigger = tap_trigger;
 405
 406         if (trigger == rxq->trigger_seen)
 407                 return 0;
 408
 409         process_private = rte_eth_devices[rxq->in_port].process_private;
 410         for (num_rx = 0; num_rx < nb_pkts; ) {
 411                 struct rte_mbuf *mbuf = rxq->pool;
 412                 struct rte_mbuf *seg = NULL;
 413                 struct rte_mbuf *new_tail = NULL;
 414                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 415                 int len;
 416
 417                 len = readv(process_private->rxq_fds[rxq->queue_id],
 418                         *rxq->iovecs,
 419                         1 + (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
 420                              rxq->nb_rx_desc : 1));
 421                 if (len < (int)sizeof(struct tun_pi))
 422                         break;
 423
 424                 /* Packet couldn't fit in the provided mbuf */
 425                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 426                         rxq->stats.ierrors++;
 427                         continue;
 428                 }
 429
 430                 len -= sizeof(struct tun_pi);
 431
 432                 mbuf->pkt_len = len;
 433                 mbuf->port = rxq->in_port;
 434                 while (1) {
 435                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 436
 437                         if (unlikely(!buf)) {
 438                                 rxq->stats.rx_nombuf++;
 439                                 /* No new buf has been allocated: do nothing */
 440                                 if (!new_tail || !seg)
 441                                         goto end;
 442
 443                                 seg->next = NULL;
 444                                 tap_rxq_pool_free(mbuf);
 445
 446                                 goto end;
 447                         }
 448                         seg = seg ? seg->next : mbuf;
 449                         if (rxq->pool == mbuf)
 450                                 rxq->pool = buf;
 451                         if (new_tail)
 452                                 new_tail->next = buf;
 453                         new_tail = buf;
 454                         new_tail->next = seg->next;
 455
 456                         /* iovecs[0] is reserved for packet info (pi) */
 457                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 458                                 buf->buf_len - data_off;
 459                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 460                                 (char *)buf->buf_addr + data_off;
 461
 462                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 463                         seg->data_off = data_off;
 464
 465                         len -= seg->data_len;
 466                         if (len <= 0)
 467                                 break;
 468                         mbuf->nb_segs++;
 469                         /* First segment has headroom, not the others */
 470                         data_off = 0;
 471                 }
 472                 seg->next = NULL;
 473                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 474                                                       RTE_PTYPE_ALL_MASK);
 475                 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
 476                         tap_verify_csum(mbuf);
 477
 478                 /* account for the receive frame */
 479                 bufs[num_rx++] = mbuf;
 480                 num_rx_bytes += mbuf->pkt_len;
 481         }
 482 end:
 483         rxq->stats.ipackets += num_rx;
 484         rxq->stats.ibytes += num_rx_bytes;
 485
 486         if (trigger && num_rx < nb_pkts)
 487                 rxq->trigger_seen = trigger;
 488
 489         return num_rx;
 490 }
 491
 492 static uint64_t
 493 tap_tx_offload_get_port_capa(void)
 494 {
 495         /*
 496          * No specific port Tx offload capabilities.
 497          */
 498         return 0;
 499 }
 500
 501 static uint64_t
 502 tap_tx_offload_get_queue_capa(void)
 503 {
 504         return DEV_TX_OFFLOAD_MULTI_SEGS |
 505                DEV_TX_OFFLOAD_IPV4_CKSUM |
 506                DEV_TX_OFFLOAD_UDP_CKSUM |
 507                DEV_TX_OFFLOAD_TCP_CKSUM |
 508                DEV_TX_OFFLOAD_TCP_TSO;
 509 }
 510
 511 /* Finalize l4 checksum calculation */
 512 static void
 513 tap_tx_l4_cksum(uint16_t *l4_cksum, uint16_t l4_phdr_cksum,
 514                 uint32_t l4_raw_cksum)
 515 {
 516         if (l4_cksum) {
 517                 uint32_t cksum;
 518
 519                 cksum = __rte_raw_cksum_reduce(l4_raw_cksum);
 520                 cksum += l4_phdr_cksum;
 521
 522                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 523                 cksum = (~cksum) & 0xffff;
 524                 if (cksum == 0)
 525                         cksum = 0xffff;
 526                 *l4_cksum = cksum;
 527         }
 528 }
 529
 530 /* Accumaulate L4 raw checksums */
 531 static void
 532 tap_tx_l4_add_rcksum(char *l4_data, unsigned int l4_len, uint16_t *l4_cksum,
 533                         uint32_t *l4_raw_cksum)
 534 {
 535         if (l4_cksum == NULL)
 536                 return;
 537
 538         *l4_raw_cksum = __rte_raw_cksum(l4_data, l4_len, *l4_raw_cksum);
 539 }
 540
 541 /* L3 and L4 pseudo headers checksum offloads */
 542 static void
 543 tap_tx_l3_cksum(char *packet, uint64_t ol_flags, unsigned int l2_len,
 544                 unsigned int l3_len, unsigned int l4_len, uint16_t **l4_cksum,
 545                 uint16_t *l4_phdr_cksum, uint32_t *l4_raw_cksum)
 546 {
 547         void *l3_hdr = packet + l2_len;
 548
 549         if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
 550                 struct rte_ipv4_hdr *iph = l3_hdr;
 551                 uint16_t cksum;
 552
 553                 iph->hdr_checksum = 0;
 554                 cksum = rte_raw_cksum(iph, l3_len);
 555                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 556         }
 557         if (ol_flags & PKT_TX_L4_MASK) {
 558                 void *l4_hdr;
 559
 560                 l4_hdr = packet + l2_len + l3_len;
 561                 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
 562                         *l4_cksum = &((struct rte_udp_hdr *)l4_hdr)->dgram_cksum;
 563                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 564                         *l4_cksum = &((struct rte_tcp_hdr *)l4_hdr)->cksum;
 565                 else
 566                         return;
 567                 **l4_cksum = 0;
 568                 if (ol_flags & PKT_TX_IPV4)
 569                         *l4_phdr_cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 570                 else
 571                         *l4_phdr_cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 572                 *l4_raw_cksum = __rte_raw_cksum(l4_hdr, l4_len, 0);
 573         }
 574 }
 575
 576 static inline int
 577 tap_write_mbufs(struct tx_queue *txq, uint16_t num_mbufs,
 578                         struct rte_mbuf **pmbufs,
 579                         uint16_t *num_packets, unsigned long *num_tx_bytes)
 580 {
 581         int i;
 582         uint16_t l234_hlen;
 583         struct pmd_process_private *process_private;
 584
 585         process_private = rte_eth_devices[txq->out_port].process_private;
 586
 587         for (i = 0; i < num_mbufs; i++) {
 588                 struct rte_mbuf *mbuf = pmbufs[i];
 589                 struct iovec iovecs[mbuf->nb_segs + 2];
 590                 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
 591                 struct rte_mbuf *seg = mbuf;
 592                 char m_copy[mbuf->data_len];
 593                 int proto;
 594                 int n;
 595                 int j;
 596                 int k; /* current index in iovecs for copying segments */
 597                 uint16_t seg_len; /* length of first segment */
 598                 uint16_t nb_segs;
 599                 uint16_t *l4_cksum; /* l4 checksum (pseudo header + payload) */
 600                 uint32_t l4_raw_cksum = 0; /* TCP/UDP payload raw checksum */
 601                 uint16_t l4_phdr_cksum = 0; /* TCP/UDP pseudo header checksum */
 602                 uint16_t is_cksum = 0; /* in case cksum should be offloaded */
 603
 604                 l4_cksum = NULL;
 605                 if (txq->type == ETH_TUNTAP_TYPE_TUN) {
 606                         /*
 607                          * TUN and TAP are created with IFF_NO_PI disabled.
 608                          * For TUN PMD this mandatory as fields are used by
 609                          * Kernel tun.c to determine whether its IP or non IP
 610                          * packets.
 611                          *
 612                          * The logic fetches the first byte of data from mbuf
 613                          * then compares whether its v4 or v6. If first byte
 614                          * is 4 or 6, then protocol field is updated.
 615                          */
 616                         char *buff_data = rte_pktmbuf_mtod(seg, void *);
 617                         proto = (*buff_data & 0xf0);
 618                         pi.proto = (proto == 0x40) ?
 619                                 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4) :
 620                                 ((proto == 0x60) ?
 621                                         rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6) :
 622                                         0x00);
 623                 }
 624
 625                 k = 0;
 626                 iovecs[k].iov_base = &pi;
 627                 iovecs[k].iov_len = sizeof(pi);
 628                 k++;
 629
 630                 nb_segs = mbuf->nb_segs;
 631                 if (txq->csum &&
 632                     ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 633                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 634                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
 635                         is_cksum = 1;
 636
 637                         /* Support only packets with at least layer 4
 638                          * header included in the first segment
 639                          */
 640                         seg_len = rte_pktmbuf_data_len(mbuf);
 641                         l234_hlen = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
 642                         if (seg_len < l234_hlen)
 643                                 return -1;
 644
 645                         /* To change checksums, work on a * copy of l2, l3
 646                          * headers + l4 pseudo header
 647                          */
 648                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 649                                         l234_hlen);
 650                         tap_tx_l3_cksum(m_copy, mbuf->ol_flags,
 651                                        mbuf->l2_len, mbuf->l3_len, mbuf->l4_len,
 652                                        &l4_cksum, &l4_phdr_cksum,
 653                                        &l4_raw_cksum);
 654                         iovecs[k].iov_base = m_copy;
 655                         iovecs[k].iov_len = l234_hlen;
 656                         k++;
 657
 658                         /* Update next iovecs[] beyond l2, l3, l4 headers */
 659                         if (seg_len > l234_hlen) {
 660                                 iovecs[k].iov_len = seg_len - l234_hlen;
 661                                 iovecs[k].iov_base =
 662                                         rte_pktmbuf_mtod(seg, char *) +
 663                                                 l234_hlen;
 664                                 tap_tx_l4_add_rcksum(iovecs[k].iov_base,
 665                                         iovecs[k].iov_len, l4_cksum,
 666                                         &l4_raw_cksum);
 667                                 k++;
 668                                 nb_segs++;
 669                         }
 670                         seg = seg->next;
 671                 }
 672
 673                 for (j = k; j <= nb_segs; j++) {
 674                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 675                         iovecs[j].iov_base = rte_pktmbuf_mtod(seg, void *);
 676                         if (is_cksum)
 677                                 tap_tx_l4_add_rcksum(iovecs[j].iov_base,
 678                                         iovecs[j].iov_len, l4_cksum,
 679                                         &l4_raw_cksum);
 680                         seg = seg->next;
 681                 }
 682
 683                 if (is_cksum)
 684                         tap_tx_l4_cksum(l4_cksum, l4_phdr_cksum, l4_raw_cksum);
 685
 686                 /* copy the tx frame data */
 687                 n = writev(process_private->txq_fds[txq->queue_id], iovecs, j);
 688                 if (n <= 0)
 689                         return -1;
 690
 691                 (*num_packets)++;
 692                 (*num_tx_bytes) += rte_pktmbuf_pkt_len(mbuf);
 693         }
 694         return 0;
 695 }
 696
 697 /* Callback to handle sending packets from the tap interface
 698  */
 699 static uint16_t
 700 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 701 {
 702         struct tx_queue *txq = queue;
 703         uint16_t num_tx = 0;
 704         uint16_t num_packets = 0;
 705         unsigned long num_tx_bytes = 0;
 706         uint32_t max_size;
 707         int i;
 708
 709         if (unlikely(nb_pkts == 0))
 710                 return 0;
 711
 712         struct rte_mbuf *gso_mbufs[MAX_GSO_MBUFS];
 713         max_size = *txq->mtu + (RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN + 4);
 714         for (i = 0; i < nb_pkts; i++) {
 715                 struct rte_mbuf *mbuf_in = bufs[num_tx];
 716                 struct rte_mbuf **mbuf;
 717                 uint16_t num_mbufs = 0;
 718                 uint16_t tso_segsz = 0;
 719                 int ret;
 720                 int num_tso_mbufs;
 721                 uint16_t hdrs_len;
 722                 uint64_t tso;
 723
 724                 tso = mbuf_in->ol_flags & PKT_TX_TCP_SEG;
 725                 if (tso) {
 726                         struct rte_gso_ctx *gso_ctx = &txq->gso_ctx;
 727
 728                         /* TCP segmentation implies TCP checksum offload */
 729                         mbuf_in->ol_flags |= PKT_TX_TCP_CKSUM;
 730
 731                         /* gso size is calculated without RTE_ETHER_CRC_LEN */
 732                         hdrs_len = mbuf_in->l2_len + mbuf_in->l3_len +
 733                                         mbuf_in->l4_len;
 734                         tso_segsz = mbuf_in->tso_segsz + hdrs_len;
 735                         if (unlikely(tso_segsz == hdrs_len) ||
 736                                 tso_segsz > *txq->mtu) {
 737                                 txq->stats.errs++;
 738                                 break;
 739                         }
 740                         gso_ctx->gso_size = tso_segsz;
 741                         /* 'mbuf_in' packet to segment */
 742                         num_tso_mbufs = rte_gso_segment(mbuf_in,
 743                                 gso_ctx, /* gso control block */
 744                                 (struct rte_mbuf **)&gso_mbufs, /* out mbufs */
 745                                 RTE_DIM(gso_mbufs)); /* max tso mbufs */
 746
 747                         /* ret contains the number of new created mbufs */
 748                         if (num_tso_mbufs < 0)
 749                                 break;
 750
 751                         mbuf = gso_mbufs;
 752                         num_mbufs = num_tso_mbufs;
 753                 } else {
 754                         /* stats.errs will be incremented */
 755                         if (rte_pktmbuf_pkt_len(mbuf_in) > max_size)
 756                                 break;
 757
 758                         /* ret 0 indicates no new mbufs were created */
 759                         num_tso_mbufs = 0;
 760                         mbuf = &mbuf_in;
 761                         num_mbufs = 1;
 762                 }
 763
 764                 ret = tap_write_mbufs(txq, num_mbufs, mbuf,
 765                                 &num_packets, &num_tx_bytes);
 766                 if (ret == -1) {
 767                         txq->stats.errs++;
 768                         /* free tso mbufs */
 769                         if (num_tso_mbufs > 0)
 770                                 rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
 771                         break;
 772                 }
 773                 num_tx++;
 774                 /* free original mbuf */
 775                 rte_pktmbuf_free(mbuf_in);
 776                 /* free tso mbufs */
 777                 if (num_tso_mbufs > 0)
 778                         rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
 779         }
 780
 781         txq->stats.opackets += num_packets;
 782         txq->stats.errs += nb_pkts - num_tx;
 783         txq->stats.obytes += num_tx_bytes;
 784
 785         return num_tx;
 786 }
 787
 788 static const char *
 789 tap_ioctl_req2str(unsigned long request)
 790 {
 791         switch (request) {
 792         case SIOCSIFFLAGS:
 793                 return "SIOCSIFFLAGS";
 794         case SIOCGIFFLAGS:
 795                 return "SIOCGIFFLAGS";
 796         case SIOCGIFHWADDR:
 797                 return "SIOCGIFHWADDR";
 798         case SIOCSIFHWADDR:
 799                 return "SIOCSIFHWADDR";
 800         case SIOCSIFMTU:
 801                 return "SIOCSIFMTU";
 802         }
 803         return "UNKNOWN";
 804 }
 805
 806 static int
 807 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 808           struct ifreq *ifr, int set, enum ioctl_mode mode)
 809 {
 810         short req_flags = ifr->ifr_flags;
 811         int remote = pmd->remote_if_index &&
 812                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 813
 814         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 815                 return 0;
 816         /*
 817          * If there is a remote netdevice, apply ioctl on it, then apply it on
 818          * the tap netdevice.
 819          */
 820 apply:
 821         if (remote)
 822                 strlcpy(ifr->ifr_name, pmd->remote_iface, IFNAMSIZ);
 823         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 824                 strlcpy(ifr->ifr_name, pmd->name, IFNAMSIZ);
 825         switch (request) {
 826         case SIOCSIFFLAGS:
 827                 /* fetch current flags to leave other flags untouched */
 828                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 829                         goto error;
 830                 if (set)
 831                         ifr->ifr_flags |= req_flags;
 832                 else
 833                         ifr->ifr_flags &= ~req_flags;
 834                 break;
 835         case SIOCGIFFLAGS:
 836         case SIOCGIFHWADDR:
 837         case SIOCSIFHWADDR:
 838         case SIOCSIFMTU:
 839                 break;
 840         default:
 841                 TAP_LOG(WARNING, "%s: ioctl() called with wrong arg",
 842                         pmd->name);
 843                 return -EINVAL;
 844         }
 845         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 846                 goto error;
 847         if (remote-- && mode == LOCAL_AND_REMOTE)
 848                 goto apply;
 849         return 0;
 850
 851 error:
 852         TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
 853                 tap_ioctl_req2str(request), strerror(errno), errno);
 854         return -errno;
 855 }
 856
 857 static int
 858 tap_link_set_down(struct rte_eth_dev *dev)
 859 {
 860         struct pmd_internals *pmd = dev->data->dev_private;
 861         struct ifreq ifr = { .ifr_flags = IFF_UP };
 862
 863         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 864         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 865 }
 866
 867 static int
 868 tap_link_set_up(struct rte_eth_dev *dev)
 869 {
 870         struct pmd_internals *pmd = dev->data->dev_private;
 871         struct ifreq ifr = { .ifr_flags = IFF_UP };
 872
 873         dev->data->dev_link.link_status = ETH_LINK_UP;
 874         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 875 }
 876
 877 static int
 878 tap_dev_start(struct rte_eth_dev *dev)
 879 {
 880         int err, i;
 881
 882         err = tap_intr_handle_set(dev, 1);
 883         if (err)
 884                 return err;
 885
 886         err = tap_link_set_up(dev);
 887         if (err)
 888                 return err;
 889
 890         for (i = 0; i < dev->data->nb_tx_queues; i++)
 891                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 892         for (i = 0; i < dev->data->nb_rx_queues; i++)
 893                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 894
 895         return err;
 896 }
 897
 898 /* This function gets called when the current port gets stopped.
 899  */
 900 static void
 901 tap_dev_stop(struct rte_eth_dev *dev)
 902 {
 903         int i;
 904
 905         for (i = 0; i < dev->data->nb_tx_queues; i++)
 906                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 907         for (i = 0; i < dev->data->nb_rx_queues; i++)
 908                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 909
 910         tap_intr_handle_set(dev, 0);
 911         tap_link_set_down(dev);
 912 }
 913
 914 static int
 915 tap_dev_configure(struct rte_eth_dev *dev)
 916 {
 917         struct pmd_internals *pmd = dev->data->dev_private;
 918
 919         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 920                 TAP_LOG(ERR,
 921                         "%s: number of rx queues %d exceeds max num of queues %d",
 922                         dev->device->name,
 923                         dev->data->nb_rx_queues,
 924                         RTE_PMD_TAP_MAX_QUEUES);
 925                 return -1;
 926         }
 927         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 928                 TAP_LOG(ERR,
 929                         "%s: number of tx queues %d exceeds max num of queues %d",
 930                         dev->device->name,
 931                         dev->data->nb_tx_queues,
 932                         RTE_PMD_TAP_MAX_QUEUES);
 933                 return -1;
 934         }
 935
 936         TAP_LOG(INFO, "%s: %s: TX configured queues number: %u",
 937                 dev->device->name, pmd->name, dev->data->nb_tx_queues);
 938
 939         TAP_LOG(INFO, "%s: %s: RX configured queues number: %u",
 940                 dev->device->name, pmd->name, dev->data->nb_rx_queues);
 941
 942         return 0;
 943 }
 944
 945 static uint32_t
 946 tap_dev_speed_capa(void)
 947 {
 948         uint32_t speed = pmd_link.link_speed;
 949         uint32_t capa = 0;
 950
 951         if (speed >= ETH_SPEED_NUM_10M)
 952                 capa |= ETH_LINK_SPEED_10M;
 953         if (speed >= ETH_SPEED_NUM_100M)
 954                 capa |= ETH_LINK_SPEED_100M;
 955         if (speed >= ETH_SPEED_NUM_1G)
 956                 capa |= ETH_LINK_SPEED_1G;
 957         if (speed >= ETH_SPEED_NUM_5G)
 958                 capa |= ETH_LINK_SPEED_2_5G;
 959         if (speed >= ETH_SPEED_NUM_5G)
 960                 capa |= ETH_LINK_SPEED_5G;
 961         if (speed >= ETH_SPEED_NUM_10G)
 962                 capa |= ETH_LINK_SPEED_10G;
 963         if (speed >= ETH_SPEED_NUM_20G)
 964                 capa |= ETH_LINK_SPEED_20G;
 965         if (speed >= ETH_SPEED_NUM_25G)
 966                 capa |= ETH_LINK_SPEED_25G;
 967         if (speed >= ETH_SPEED_NUM_40G)
 968                 capa |= ETH_LINK_SPEED_40G;
 969         if (speed >= ETH_SPEED_NUM_50G)
 970                 capa |= ETH_LINK_SPEED_50G;
 971         if (speed >= ETH_SPEED_NUM_56G)
 972                 capa |= ETH_LINK_SPEED_56G;
 973         if (speed >= ETH_SPEED_NUM_100G)
 974                 capa |= ETH_LINK_SPEED_100G;
 975
 976         return capa;
 977 }
 978
 979 static int
 980 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 981 {
 982         struct pmd_internals *internals = dev->data->dev_private;
 983
 984         dev_info->if_index = internals->if_index;
 985         dev_info->max_mac_addrs = 1;
 986         dev_info->max_rx_pktlen = (uint32_t)RTE_ETHER_MAX_VLAN_FRAME_LEN;
 987         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
 988         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
 989         dev_info->min_rx_bufsize = 0;
 990         dev_info->speed_capa = tap_dev_speed_capa();
 991         dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
 992         dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
 993                                     dev_info->rx_queue_offload_capa;
 994         dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
 995         dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
 996                                     dev_info->tx_queue_offload_capa;
 997         dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
 998         /*
 999          * limitation: TAP supports all of IP, UDP and TCP hash
1000          * functions together and not in partial combinations
1001          */
1002         dev_info->flow_type_rss_offloads = ~TAP_RSS_HF_MASK;
1003
1004         return 0;
1005 }
1006
1007 static int
1008 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
1009 {
1010         unsigned int i, imax;
1011         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
1012         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
1013         unsigned long rx_nombuf = 0, ierrors = 0;
1014         const struct pmd_internals *pmd = dev->data->dev_private;
1015
1016         /* rx queue statistics */
1017         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
1018                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
1019         for (i = 0; i < imax; i++) {
1020                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
1021                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
1022                 rx_total += tap_stats->q_ipackets[i];
1023                 rx_bytes_total += tap_stats->q_ibytes[i];
1024                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
1025                 ierrors += pmd->rxq[i].stats.ierrors;
1026         }
1027
1028         /* tx queue statistics */
1029         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
1030                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
1031
1032         for (i = 0; i < imax; i++) {
1033                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
1034                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
1035                 tx_total += tap_stats->q_opackets[i];
1036                 tx_err_total += pmd->txq[i].stats.errs;
1037                 tx_bytes_total += tap_stats->q_obytes[i];
1038         }
1039
1040         tap_stats->ipackets = rx_total;
1041         tap_stats->ibytes = rx_bytes_total;
1042         tap_stats->ierrors = ierrors;
1043         tap_stats->rx_nombuf = rx_nombuf;
1044         tap_stats->opackets = tx_total;
1045         tap_stats->oerrors = tx_err_total;
1046         tap_stats->obytes = tx_bytes_total;
1047         return 0;
1048 }
1049
1050 static int
1051 tap_stats_reset(struct rte_eth_dev *dev)
1052 {
1053         int i;
1054         struct pmd_internals *pmd = dev->data->dev_private;
1055
1056         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1057                 pmd->rxq[i].stats.ipackets = 0;
1058                 pmd->rxq[i].stats.ibytes = 0;
1059                 pmd->rxq[i].stats.ierrors = 0;
1060                 pmd->rxq[i].stats.rx_nombuf = 0;
1061
1062                 pmd->txq[i].stats.opackets = 0;
1063                 pmd->txq[i].stats.errs = 0;
1064                 pmd->txq[i].stats.obytes = 0;
1065         }
1066
1067         return 0;
1068 }
1069
1070 static void
1071 tap_dev_close(struct rte_eth_dev *dev)
1072 {
1073         int i;
1074         struct pmd_internals *internals = dev->data->dev_private;
1075         struct pmd_process_private *process_private = dev->process_private;
1076         struct rx_queue *rxq;
1077
1078         tap_link_set_down(dev);
1079         if (internals->nlsk_fd != -1) {
1080                 tap_flow_flush(dev, NULL);
1081                 tap_flow_implicit_flush(internals, NULL);
1082                 tap_nl_final(internals->nlsk_fd);
1083                 internals->nlsk_fd = -1;
1084         }
1085
1086         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1087                 if (process_private->rxq_fds[i] != -1) {
1088                         rxq = &internals->rxq[i];
1089                         close(process_private->rxq_fds[i]);
1090                         process_private->rxq_fds[i] = -1;
1091                         tap_rxq_pool_free(rxq->pool);
1092                         rte_free(rxq->iovecs);
1093                         rxq->pool = NULL;
1094                         rxq->iovecs = NULL;
1095                 }
1096                 if (process_private->txq_fds[i] != -1) {
1097                         close(process_private->txq_fds[i]);
1098                         process_private->txq_fds[i] = -1;
1099                 }
1100         }
1101
1102         if (internals->remote_if_index) {
1103                 /* Restore initial remote state */
1104                 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
1105                                 &internals->remote_initial_flags);
1106         }
1107
1108         if (internals->ka_fd != -1) {
1109                 close(internals->ka_fd);
1110                 internals->ka_fd = -1;
1111         }
1112         /*
1113          * Since TUN device has no more opened file descriptors
1114          * it will be removed from kernel
1115          */
1116 }
1117
1118 static void
1119 tap_rx_queue_release(void *queue)
1120 {
1121         struct rx_queue *rxq = queue;
1122         struct pmd_process_private *process_private;
1123
1124         if (!rxq)
1125                 return;
1126         process_private = rte_eth_devices[rxq->in_port].process_private;
1127         if (process_private->rxq_fds[rxq->queue_id] != -1) {
1128                 close(process_private->rxq_fds[rxq->queue_id]);
1129                 process_private->rxq_fds[rxq->queue_id] = -1;
1130                 tap_rxq_pool_free(rxq->pool);
1131                 rte_free(rxq->iovecs);
1132                 rxq->pool = NULL;
1133                 rxq->iovecs = NULL;
1134         }
1135 }
1136
1137 static void
1138 tap_tx_queue_release(void *queue)
1139 {
1140         struct tx_queue *txq = queue;
1141         struct pmd_process_private *process_private;
1142
1143         if (!txq)
1144                 return;
1145         process_private = rte_eth_devices[txq->out_port].process_private;
1146
1147         if (process_private->txq_fds[txq->queue_id] != -1) {
1148                 close(process_private->txq_fds[txq->queue_id]);
1149                 process_private->txq_fds[txq->queue_id] = -1;
1150         }
1151 }
1152
1153 static int
1154 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1155 {
1156         struct rte_eth_link *dev_link = &dev->data->dev_link;
1157         struct pmd_internals *pmd = dev->data->dev_private;
1158         struct ifreq ifr = { .ifr_flags = 0 };
1159
1160         if (pmd->remote_if_index) {
1161                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
1162                 if (!(ifr.ifr_flags & IFF_UP) ||
1163                     !(ifr.ifr_flags & IFF_RUNNING)) {
1164                         dev_link->link_status = ETH_LINK_DOWN;
1165                         return 0;
1166                 }
1167         }
1168         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
1169         dev_link->link_status =
1170                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
1171                  ETH_LINK_UP :
1172                  ETH_LINK_DOWN);
1173         return 0;
1174 }
1175
1176 static int
1177 tap_promisc_enable(struct rte_eth_dev *dev)
1178 {
1179         struct pmd_internals *pmd = dev->data->dev_private;
1180         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1181         int ret;
1182
1183         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1184         if (ret != 0)
1185                 return ret;
1186
1187         if (pmd->remote_if_index && !pmd->flow_isolate) {
1188                 dev->data->promiscuous = 1;
1189                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
1190                 if (ret != 0) {
1191                         /* Rollback promisc flag */
1192                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1193                         /*
1194                          * rte_eth_dev_promiscuous_enable() rollback
1195                          * dev->data->promiscuous in the case of failure.
1196                          */
1197                         return ret;
1198                 }
1199         }
1200
1201         return 0;
1202 }
1203
1204 static int
1205 tap_promisc_disable(struct rte_eth_dev *dev)
1206 {
1207         struct pmd_internals *pmd = dev->data->dev_private;
1208         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1209         int ret;
1210
1211         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1212         if (ret != 0)
1213                 return ret;
1214
1215         if (pmd->remote_if_index && !pmd->flow_isolate) {
1216                 dev->data->promiscuous = 0;
1217                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
1218                 if (ret != 0) {
1219                         /* Rollback promisc flag */
1220                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1221                         /*
1222                          * rte_eth_dev_promiscuous_disable() rollback
1223                          * dev->data->promiscuous in the case of failure.
1224                          */
1225                         return ret;
1226                 }
1227         }
1228
1229         return 0;
1230 }
1231
1232 static int
1233 tap_allmulti_enable(struct rte_eth_dev *dev)
1234 {
1235         struct pmd_internals *pmd = dev->data->dev_private;
1236         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1237         int ret;
1238
1239         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1240         if (ret != 0)
1241                 return ret;
1242
1243         if (pmd->remote_if_index && !pmd->flow_isolate) {
1244                 dev->data->all_multicast = 1;
1245                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
1246                 if (ret != 0) {
1247                         /* Rollback allmulti flag */
1248                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1249                         /*
1250                          * rte_eth_dev_allmulticast_enable() rollback
1251                          * dev->data->all_multicast in the case of failure.
1252                          */
1253                         return ret;
1254                 }
1255         }
1256
1257         return 0;
1258 }
1259
1260 static int
1261 tap_allmulti_disable(struct rte_eth_dev *dev)
1262 {
1263         struct pmd_internals *pmd = dev->data->dev_private;
1264         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1265         int ret;
1266
1267         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1268         if (ret != 0)
1269                 return ret;
1270
1271         if (pmd->remote_if_index && !pmd->flow_isolate) {
1272                 dev->data->all_multicast = 0;
1273                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
1274                 if (ret != 0) {
1275                         /* Rollback allmulti flag */
1276                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1277                         /*
1278                          * rte_eth_dev_allmulticast_disable() rollback
1279                          * dev->data->all_multicast in the case of failure.
1280                          */
1281                         return ret;
1282                 }
1283         }
1284
1285         return 0;
1286 }
1287
1288 static int
1289 tap_mac_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1290 {
1291         struct pmd_internals *pmd = dev->data->dev_private;
1292         enum ioctl_mode mode = LOCAL_ONLY;
1293         struct ifreq ifr;
1294         int ret;
1295
1296         if (pmd->type == ETH_TUNTAP_TYPE_TUN) {
1297                 TAP_LOG(ERR, "%s: can't MAC address for TUN",
1298                         dev->device->name);
1299                 return -ENOTSUP;
1300         }
1301
1302         if (rte_is_zero_ether_addr(mac_addr)) {
1303                 TAP_LOG(ERR, "%s: can't set an empty MAC address",
1304                         dev->device->name);
1305                 return -EINVAL;
1306         }
1307         /* Check the actual current MAC address on the tap netdevice */
1308         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
1309         if (ret < 0)
1310                 return ret;
1311         if (rte_is_same_ether_addr(
1312                         (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1313                         mac_addr))
1314                 return 0;
1315         /* Check the current MAC address on the remote */
1316         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
1317         if (ret < 0)
1318                 return ret;
1319         if (!rte_is_same_ether_addr(
1320                         (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1321                         mac_addr))
1322                 mode = LOCAL_AND_REMOTE;
1323         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1324         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, RTE_ETHER_ADDR_LEN);
1325         ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
1326         if (ret < 0)
1327                 return ret;
1328         rte_memcpy(&pmd->eth_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1329         if (pmd->remote_if_index && !pmd->flow_isolate) {
1330                 /* Replace MAC redirection rule after a MAC change */
1331                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
1332                 if (ret < 0) {
1333                         TAP_LOG(ERR,
1334                                 "%s: Couldn't delete MAC redirection rule",
1335                                 dev->device->name);
1336                         return ret;
1337                 }
1338                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1339                 if (ret < 0) {
1340                         TAP_LOG(ERR,
1341                                 "%s: Couldn't add MAC redirection rule",
1342                                 dev->device->name);
1343                         return ret;
1344                 }
1345         }
1346
1347         return 0;
1348 }
1349
1350 static int
1351 tap_gso_ctx_setup(struct rte_gso_ctx *gso_ctx, struct rte_eth_dev *dev)
1352 {
1353         uint32_t gso_types;
1354         char pool_name[64];
1355
1356         /*
1357          * Create private mbuf pool with TAP_GSO_MBUF_SEG_SIZE bytes
1358          * size per mbuf use this pool for both direct and indirect mbufs
1359          */
1360
1361         struct rte_mempool *mp;      /* Mempool for GSO packets */
1362
1363         /* initialize GSO context */
1364         gso_types = DEV_TX_OFFLOAD_TCP_TSO;
1365         snprintf(pool_name, sizeof(pool_name), "mp_%s", dev->device->name);
1366         mp = rte_mempool_lookup((const char *)pool_name);
1367         if (!mp) {
1368                 mp = rte_pktmbuf_pool_create(pool_name, TAP_GSO_MBUFS_NUM,
1369                         TAP_GSO_MBUF_CACHE_SIZE, 0,
1370                         RTE_PKTMBUF_HEADROOM + TAP_GSO_MBUF_SEG_SIZE,
1371                         SOCKET_ID_ANY);
1372                 if (!mp) {
1373                         struct pmd_internals *pmd = dev->data->dev_private;
1374
1375                         TAP_LOG(ERR,
1376                                 "%s: failed to create mbuf pool for device %s\n",
1377                                 pmd->name, dev->device->name);
1378                         return -1;
1379                 }
1380         }
1381
1382         gso_ctx->direct_pool = mp;
1383         gso_ctx->indirect_pool = mp;
1384         gso_ctx->gso_types = gso_types;
1385         gso_ctx->gso_size = 0; /* gso_size is set in tx_burst() per packet */
1386         gso_ctx->flag = 0;
1387
1388         return 0;
1389 }
1390
1391 static int
1392 tap_setup_queue(struct rte_eth_dev *dev,
1393                 struct pmd_internals *internals,
1394                 uint16_t qid,
1395                 int is_rx)
1396 {
1397         int ret;
1398         int *fd;
1399         int *other_fd;
1400         const char *dir;
1401         struct pmd_internals *pmd = dev->data->dev_private;
1402         struct pmd_process_private *process_private = dev->process_private;
1403         struct rx_queue *rx = &internals->rxq[qid];
1404         struct tx_queue *tx = &internals->txq[qid];
1405         struct rte_gso_ctx *gso_ctx;
1406
1407         if (is_rx) {
1408                 fd = &process_private->rxq_fds[qid];
1409                 other_fd = &process_private->txq_fds[qid];
1410                 dir = "rx";
1411                 gso_ctx = NULL;
1412         } else {
1413                 fd = &process_private->txq_fds[qid];
1414                 other_fd = &process_private->rxq_fds[qid];
1415                 dir = "tx";
1416                 gso_ctx = &tx->gso_ctx;
1417         }
1418         if (*fd != -1) {
1419                 /* fd for this queue already exists */
1420                 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1421                         pmd->name, *fd, dir, qid);
1422                 gso_ctx = NULL;
1423         } else if (*other_fd != -1) {
1424                 /* Only other_fd exists. dup it */
1425                 *fd = dup(*other_fd);
1426                 if (*fd < 0) {
1427                         *fd = -1;
1428                         TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1429                         return -1;
1430                 }
1431                 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1432                         pmd->name, *other_fd, dir, qid, *fd);
1433         } else {
1434                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1435                 *fd = tun_alloc(pmd, 0);
1436                 if (*fd < 0) {
1437                         *fd = -1; /* restore original value */
1438                         TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1439                         return -1;
1440                 }
1441                 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1442                         pmd->name, dir, qid, *fd);
1443         }
1444
1445         tx->mtu = &dev->data->mtu;
1446         rx->rxmode = &dev->data->dev_conf.rxmode;
1447         if (gso_ctx) {
1448                 ret = tap_gso_ctx_setup(gso_ctx, dev);
1449                 if (ret)
1450                         return -1;
1451         }
1452
1453         tx->type = pmd->type;
1454
1455         return *fd;
1456 }
1457
1458 static int
1459 tap_rx_queue_setup(struct rte_eth_dev *dev,
1460                    uint16_t rx_queue_id,
1461                    uint16_t nb_rx_desc,
1462                    unsigned int socket_id,
1463                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1464                    struct rte_mempool *mp)
1465 {
1466         struct pmd_internals *internals = dev->data->dev_private;
1467         struct pmd_process_private *process_private = dev->process_private;
1468         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1469         struct rte_mbuf **tmp = &rxq->pool;
1470         long iov_max = sysconf(_SC_IOV_MAX);
1471
1472         if (iov_max <= 0) {
1473                 TAP_LOG(WARNING,
1474                         "_SC_IOV_MAX is not defined. Using %d as default",
1475                         TAP_IOV_DEFAULT_MAX);
1476                 iov_max = TAP_IOV_DEFAULT_MAX;
1477         }
1478         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1479         struct iovec (*iovecs)[nb_desc + 1];
1480         int data_off = RTE_PKTMBUF_HEADROOM;
1481         int ret = 0;
1482         int fd;
1483         int i;
1484
1485         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1486                 TAP_LOG(WARNING,
1487                         "nb_rx_queues %d too small or mempool NULL",
1488                         dev->data->nb_rx_queues);
1489                 return -1;
1490         }
1491
1492         rxq->mp = mp;
1493         rxq->trigger_seen = 1; /* force initial burst */
1494         rxq->in_port = dev->data->port_id;
1495         rxq->queue_id = rx_queue_id;
1496         rxq->nb_rx_desc = nb_desc;
1497         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1498                                     socket_id);
1499         if (!iovecs) {
1500                 TAP_LOG(WARNING,
1501                         "%s: Couldn't allocate %d RX descriptors",
1502                         dev->device->name, nb_desc);
1503                 return -ENOMEM;
1504         }
1505         rxq->iovecs = iovecs;
1506
1507         dev->data->rx_queues[rx_queue_id] = rxq;
1508         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1509         if (fd == -1) {
1510                 ret = fd;
1511                 goto error;
1512         }
1513
1514         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1515         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1516
1517         for (i = 1; i <= nb_desc; i++) {
1518                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1519                 if (!*tmp) {
1520                         TAP_LOG(WARNING,
1521                                 "%s: couldn't allocate memory for queue %d",
1522                                 dev->device->name, rx_queue_id);
1523                         ret = -ENOMEM;
1524                         goto error;
1525                 }
1526                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1527                 (*rxq->iovecs)[i].iov_base =
1528                         (char *)(*tmp)->buf_addr + data_off;
1529                 data_off = 0;
1530                 tmp = &(*tmp)->next;
1531         }
1532
1533         TAP_LOG(DEBUG, "  RX TUNTAP device name %s, qid %d on fd %d",
1534                 internals->name, rx_queue_id,
1535                 process_private->rxq_fds[rx_queue_id]);
1536
1537         return 0;
1538
1539 error:
1540         tap_rxq_pool_free(rxq->pool);
1541         rxq->pool = NULL;
1542         rte_free(rxq->iovecs);
1543         rxq->iovecs = NULL;
1544         return ret;
1545 }
1546
1547 static int
1548 tap_tx_queue_setup(struct rte_eth_dev *dev,
1549                    uint16_t tx_queue_id,
1550                    uint16_t nb_tx_desc __rte_unused,
1551                    unsigned int socket_id __rte_unused,
1552                    const struct rte_eth_txconf *tx_conf)
1553 {
1554         struct pmd_internals *internals = dev->data->dev_private;
1555         struct pmd_process_private *process_private = dev->process_private;
1556         struct tx_queue *txq;
1557         int ret;
1558         uint64_t offloads;
1559
1560         if (tx_queue_id >= dev->data->nb_tx_queues)
1561                 return -1;
1562         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1563         txq = dev->data->tx_queues[tx_queue_id];
1564         txq->out_port = dev->data->port_id;
1565         txq->queue_id = tx_queue_id;
1566
1567         offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1568         txq->csum = !!(offloads &
1569                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
1570                          DEV_TX_OFFLOAD_UDP_CKSUM |
1571                          DEV_TX_OFFLOAD_TCP_CKSUM));
1572
1573         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1574         if (ret == -1)
1575                 return -1;
1576         TAP_LOG(DEBUG,
1577                 "  TX TUNTAP device name %s, qid %d on fd %d csum %s",
1578                 internals->name, tx_queue_id,
1579                 process_private->txq_fds[tx_queue_id],
1580                 txq->csum ? "on" : "off");
1581
1582         return 0;
1583 }
1584
1585 static int
1586 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1587 {
1588         struct pmd_internals *pmd = dev->data->dev_private;
1589         struct ifreq ifr = { .ifr_mtu = mtu };
1590         int err = 0;
1591
1592         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1593         if (!err)
1594                 dev->data->mtu = mtu;
1595
1596         return err;
1597 }
1598
1599 static int
1600 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1601                      struct rte_ether_addr *mc_addr_set __rte_unused,
1602                      uint32_t nb_mc_addr __rte_unused)
1603 {
1604         /*
1605          * Nothing to do actually: the tap has no filtering whatsoever, every
1606          * packet is received.
1607          */
1608         return 0;
1609 }
1610
1611 static int
1612 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1613 {
1614         struct rte_eth_dev *dev = arg;
1615         struct pmd_internals *pmd = dev->data->dev_private;
1616         struct ifinfomsg *info = NLMSG_DATA(nh);
1617
1618         if (nh->nlmsg_type != RTM_NEWLINK ||
1619             (info->ifi_index != pmd->if_index &&
1620              info->ifi_index != pmd->remote_if_index))
1621                 return 0;
1622         return tap_link_update(dev, 0);
1623 }
1624
1625 static void
1626 tap_dev_intr_handler(void *cb_arg)
1627 {
1628         struct rte_eth_dev *dev = cb_arg;
1629         struct pmd_internals *pmd = dev->data->dev_private;
1630
1631         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1632 }
1633
1634 static int
1635 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1636 {
1637         struct pmd_internals *pmd = dev->data->dev_private;
1638         int ret;
1639
1640         /* In any case, disable interrupt if the conf is no longer there. */
1641         if (!dev->data->dev_conf.intr_conf.lsc) {
1642                 if (pmd->intr_handle.fd != -1) {
1643                         goto clean;
1644                 }
1645                 return 0;
1646         }
1647         if (set) {
1648                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1649                 if (unlikely(pmd->intr_handle.fd == -1))
1650                         return -EBADF;
1651                 return rte_intr_callback_register(
1652                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1653         }
1654
1655 clean:
1656         do {
1657                 ret = rte_intr_callback_unregister(&pmd->intr_handle,
1658                         tap_dev_intr_handler, dev);
1659                 if (ret >= 0) {
1660                         break;
1661                 } else if (ret == -EAGAIN) {
1662                         rte_delay_ms(100);
1663                 } else {
1664                         TAP_LOG(ERR, "intr callback unregister failed: %d",
1665                                      ret);
1666                         break;
1667                 }
1668         } while (true);
1669
1670         tap_nl_final(pmd->intr_handle.fd);
1671         pmd->intr_handle.fd = -1;
1672
1673         return 0;
1674 }
1675
1676 static int
1677 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1678 {
1679         int err;
1680
1681         err = tap_lsc_intr_handle_set(dev, set);
1682         if (err < 0) {
1683                 if (!set)
1684                         tap_rx_intr_vec_set(dev, 0);
1685                 return err;
1686         }
1687         err = tap_rx_intr_vec_set(dev, set);
1688         if (err && set)
1689                 tap_lsc_intr_handle_set(dev, 0);
1690         return err;
1691 }
1692
1693 static const uint32_t*
1694 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1695 {
1696         static const uint32_t ptypes[] = {
1697                 RTE_PTYPE_INNER_L2_ETHER,
1698                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1699                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1700                 RTE_PTYPE_INNER_L3_IPV4,
1701                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1702                 RTE_PTYPE_INNER_L3_IPV6,
1703                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1704                 RTE_PTYPE_INNER_L4_FRAG,
1705                 RTE_PTYPE_INNER_L4_UDP,
1706                 RTE_PTYPE_INNER_L4_TCP,
1707                 RTE_PTYPE_INNER_L4_SCTP,
1708                 RTE_PTYPE_L2_ETHER,
1709                 RTE_PTYPE_L2_ETHER_VLAN,
1710                 RTE_PTYPE_L2_ETHER_QINQ,
1711                 RTE_PTYPE_L3_IPV4,
1712                 RTE_PTYPE_L3_IPV4_EXT,
1713                 RTE_PTYPE_L3_IPV6_EXT,
1714                 RTE_PTYPE_L3_IPV6,
1715                 RTE_PTYPE_L4_FRAG,
1716                 RTE_PTYPE_L4_UDP,
1717                 RTE_PTYPE_L4_TCP,
1718                 RTE_PTYPE_L4_SCTP,
1719         };
1720
1721         return ptypes;
1722 }
1723
1724 static int
1725 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1726                   struct rte_eth_fc_conf *fc_conf)
1727 {
1728         fc_conf->mode = RTE_FC_NONE;
1729         return 0;
1730 }
1731
1732 static int
1733 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1734                   struct rte_eth_fc_conf *fc_conf)
1735 {
1736         if (fc_conf->mode != RTE_FC_NONE)
1737                 return -ENOTSUP;
1738         return 0;
1739 }
1740
1741 /**
1742  * DPDK callback to update the RSS hash configuration.
1743  *
1744  * @param dev
1745  *   Pointer to Ethernet device structure.
1746  * @param[in] rss_conf
1747  *   RSS configuration data.
1748  *
1749  * @return
1750  *   0 on success, a negative errno value otherwise and rte_errno is set.
1751  */
1752 static int
1753 tap_rss_hash_update(struct rte_eth_dev *dev,
1754                 struct rte_eth_rss_conf *rss_conf)
1755 {
1756         if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1757                 rte_errno = EINVAL;
1758                 return -rte_errno;
1759         }
1760         if (rss_conf->rss_key && rss_conf->rss_key_len) {
1761                 /*
1762                  * Currently TAP RSS key is hard coded
1763                  * and cannot be updated
1764                  */
1765                 TAP_LOG(ERR,
1766                         "port %u RSS key cannot be updated",
1767                         dev->data->port_id);
1768                 rte_errno = EINVAL;
1769                 return -rte_errno;
1770         }
1771         return 0;
1772 }
1773
1774 static int
1775 tap_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1776 {
1777         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1778
1779         return 0;
1780 }
1781
1782 static int
1783 tap_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1784 {
1785         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1786
1787         return 0;
1788 }
1789
1790 static int
1791 tap_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1792 {
1793         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1794
1795         return 0;
1796 }
1797
1798 static int
1799 tap_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1800 {
1801         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1802
1803         return 0;
1804 }
1805 static const struct eth_dev_ops ops = {
1806         .dev_start              = tap_dev_start,
1807         .dev_stop               = tap_dev_stop,
1808         .dev_close              = tap_dev_close,
1809         .dev_configure          = tap_dev_configure,
1810         .dev_infos_get          = tap_dev_info,
1811         .rx_queue_setup         = tap_rx_queue_setup,
1812         .tx_queue_setup         = tap_tx_queue_setup,
1813         .rx_queue_start         = tap_rx_queue_start,
1814         .tx_queue_start         = tap_tx_queue_start,
1815         .rx_queue_stop          = tap_rx_queue_stop,
1816         .tx_queue_stop          = tap_tx_queue_stop,
1817         .rx_queue_release       = tap_rx_queue_release,
1818         .tx_queue_release       = tap_tx_queue_release,
1819         .flow_ctrl_get          = tap_flow_ctrl_get,
1820         .flow_ctrl_set          = tap_flow_ctrl_set,
1821         .link_update            = tap_link_update,
1822         .dev_set_link_up        = tap_link_set_up,
1823         .dev_set_link_down      = tap_link_set_down,
1824         .promiscuous_enable     = tap_promisc_enable,
1825         .promiscuous_disable    = tap_promisc_disable,
1826         .allmulticast_enable    = tap_allmulti_enable,
1827         .allmulticast_disable   = tap_allmulti_disable,
1828         .mac_addr_set           = tap_mac_set,
1829         .mtu_set                = tap_mtu_set,
1830         .set_mc_addr_list       = tap_set_mc_addr_list,
1831         .stats_get              = tap_stats_get,
1832         .stats_reset            = tap_stats_reset,
1833         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1834         .rss_hash_update        = tap_rss_hash_update,
1835         .filter_ctrl            = tap_dev_filter_ctrl,
1836 };
1837
1838 static const char *tuntap_types[ETH_TUNTAP_TYPE_MAX] = {
1839         "UNKNOWN", "TUN", "TAP"
1840 };
1841
1842 static int
1843 eth_dev_tap_create(struct rte_vdev_device *vdev, const char *tap_name,
1844                    char *remote_iface, struct rte_ether_addr *mac_addr,
1845                    enum rte_tuntap_type type)
1846 {
1847         int numa_node = rte_socket_id();
1848         struct rte_eth_dev *dev;
1849         struct pmd_internals *pmd;
1850         struct pmd_process_private *process_private;
1851         const char *tuntap_name = tuntap_types[type];
1852         struct rte_eth_dev_data *data;
1853         struct ifreq ifr;
1854         int i;
1855
1856         TAP_LOG(DEBUG, "%s device on numa %u", tuntap_name, rte_socket_id());
1857
1858         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1859         if (!dev) {
1860                 TAP_LOG(ERR, "%s Unable to allocate device struct",
1861                                 tuntap_name);
1862                 goto error_exit_nodev;
1863         }
1864
1865         process_private = (struct pmd_process_private *)
1866                 rte_zmalloc_socket(tap_name, sizeof(struct pmd_process_private),
1867                         RTE_CACHE_LINE_SIZE, dev->device->numa_node);
1868
1869         if (process_private == NULL) {
1870                 TAP_LOG(ERR, "Failed to alloc memory for process private");
1871                 return -1;
1872         }
1873         pmd = dev->data->dev_private;
1874         dev->process_private = process_private;
1875         pmd->dev = dev;
1876         strlcpy(pmd->name, tap_name, sizeof(pmd->name));
1877         pmd->type = type;
1878         pmd->ka_fd = -1;
1879         pmd->nlsk_fd = -1;
1880
1881         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1882         if (pmd->ioctl_sock == -1) {
1883                 TAP_LOG(ERR,
1884                         "%s Unable to get a socket for management: %s",
1885                         tuntap_name, strerror(errno));
1886                 goto error_exit;
1887         }
1888
1889         /* Setup some default values */
1890         data = dev->data;
1891         data->dev_private = pmd;
1892         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1893         data->numa_node = numa_node;
1894
1895         data->dev_link = pmd_link;
1896         data->mac_addrs = &pmd->eth_addr;
1897         /* Set the number of RX and TX queues */
1898         data->nb_rx_queues = 0;
1899         data->nb_tx_queues = 0;
1900
1901         dev->dev_ops = &ops;
1902         dev->rx_pkt_burst = pmd_rx_burst;
1903         dev->tx_pkt_burst = pmd_tx_burst;
1904
1905         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1906         pmd->intr_handle.fd = -1;
1907         dev->intr_handle = &pmd->intr_handle;
1908
1909         /* Presetup the fds to -1 as being not valid */
1910         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1911                 process_private->rxq_fds[i] = -1;
1912                 process_private->txq_fds[i] = -1;
1913         }
1914
1915         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1916                 if (rte_is_zero_ether_addr(mac_addr))
1917                         rte_eth_random_addr((uint8_t *)&pmd->eth_addr);
1918                 else
1919                         rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1920         }
1921
1922         /*
1923          * Allocate a TUN device keep-alive file descriptor that will only be
1924          * closed when the TUN device itself is closed or removed.
1925          * This keep-alive file descriptor will guarantee that the TUN device
1926          * exists even when all of its queues are closed
1927          */
1928         pmd->ka_fd = tun_alloc(pmd, 1);
1929         if (pmd->ka_fd == -1) {
1930                 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
1931                 goto error_exit;
1932         }
1933         TAP_LOG(DEBUG, "allocated %s", pmd->name);
1934
1935         ifr.ifr_mtu = dev->data->mtu;
1936         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1937                 goto error_exit;
1938
1939         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1940                 memset(&ifr, 0, sizeof(struct ifreq));
1941                 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1942                 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1943                                 RTE_ETHER_ADDR_LEN);
1944                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1945                         goto error_exit;
1946         }
1947
1948         /*
1949          * Set up everything related to rte_flow:
1950          * - netlink socket
1951          * - tap / remote if_index
1952          * - mandatory QDISCs
1953          * - rte_flow actual/implicit lists
1954          * - implicit rules
1955          */
1956         pmd->nlsk_fd = tap_nl_init(0);
1957         if (pmd->nlsk_fd == -1) {
1958                 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
1959                         pmd->name);
1960                 goto disable_rte_flow;
1961         }
1962         pmd->if_index = if_nametoindex(pmd->name);
1963         if (!pmd->if_index) {
1964                 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
1965                 goto disable_rte_flow;
1966         }
1967         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1968                 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
1969                         pmd->name);
1970                 goto disable_rte_flow;
1971         }
1972         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1973                 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1974                         pmd->name);
1975                 goto disable_rte_flow;
1976         }
1977         LIST_INIT(&pmd->flows);
1978
1979         if (strlen(remote_iface)) {
1980                 pmd->remote_if_index = if_nametoindex(remote_iface);
1981                 if (!pmd->remote_if_index) {
1982                         TAP_LOG(ERR, "%s: failed to get %s if_index.",
1983                                 pmd->name, remote_iface);
1984                         goto error_remote;
1985                 }
1986                 strlcpy(pmd->remote_iface, remote_iface, RTE_ETH_NAME_MAX_LEN);
1987
1988                 /* Save state of remote device */
1989                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1990
1991                 /* Replicate remote MAC address */
1992                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1993                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1994                                 pmd->name, pmd->remote_iface);
1995                         goto error_remote;
1996                 }
1997                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1998                            RTE_ETHER_ADDR_LEN);
1999                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
2000                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
2001                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
2002                                 pmd->name, remote_iface);
2003                         goto error_remote;
2004                 }
2005
2006                 /*
2007                  * Flush usually returns negative value because it tries to
2008                  * delete every QDISC (and on a running device, one QDISC at
2009                  * least is needed). Ignore negative return value.
2010                  */
2011                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
2012                 if (qdisc_create_ingress(pmd->nlsk_fd,
2013                                          pmd->remote_if_index) < 0) {
2014                         TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
2015                                 pmd->remote_iface);
2016                         goto error_remote;
2017                 }
2018                 LIST_INIT(&pmd->implicit_flows);
2019                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
2020                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
2021                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
2022                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
2023                         TAP_LOG(ERR,
2024                                 "%s: failed to create implicit rules.",
2025                                 pmd->name);
2026                         goto error_remote;
2027                 }
2028         }
2029
2030         rte_eth_dev_probing_finish(dev);
2031         return 0;
2032
2033 disable_rte_flow:
2034         TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
2035                 strerror(errno), errno);
2036         if (strlen(remote_iface)) {
2037                 TAP_LOG(ERR, "Remote feature requires flow support.");
2038                 goto error_exit;
2039         }
2040         rte_eth_dev_probing_finish(dev);
2041         return 0;
2042
2043 error_remote:
2044         TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
2045                 strerror(errno), errno);
2046         tap_flow_implicit_flush(pmd, NULL);
2047
2048 error_exit:
2049         if (pmd->nlsk_fd != -1)
2050                 close(pmd->nlsk_fd);
2051         if (pmd->ka_fd != -1)
2052                 close(pmd->ka_fd);
2053         if (pmd->ioctl_sock != -1)
2054                 close(pmd->ioctl_sock);
2055         /* mac_addrs must not be freed alone because part of dev_private */
2056         dev->data->mac_addrs = NULL;
2057         rte_eth_dev_release_port(dev);
2058
2059 error_exit_nodev:
2060         TAP_LOG(ERR, "%s Unable to initialize %s",
2061                 tuntap_name, rte_vdev_device_name(vdev));
2062
2063         return -EINVAL;
2064 }
2065
2066 /* make sure name is a possible Linux network device name */
2067 static bool
2068 is_valid_iface(const char *name)
2069 {
2070         if (*name == '\0')
2071                 return false;
2072
2073         if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
2074                 return false;
2075
2076         while (*name) {
2077                 if (*name == '/' || *name == ':' || isspace(*name))
2078                         return false;
2079                 name++;
2080         }
2081         return true;
2082 }
2083
2084 static int
2085 set_interface_name(const char *key __rte_unused,
2086                    const char *value,
2087                    void *extra_args)
2088 {
2089         char *name = (char *)extra_args;
2090
2091         if (value) {
2092                 if (!is_valid_iface(value)) {
2093                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2094                                 value);
2095                         return -1;
2096                 }
2097                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2098         } else {
2099                 /* use tap%d which causes kernel to choose next available */
2100                 strlcpy(name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2101         }
2102         return 0;
2103 }
2104
2105 static int
2106 set_remote_iface(const char *key __rte_unused,
2107                  const char *value,
2108                  void *extra_args)
2109 {
2110         char *name = (char *)extra_args;
2111
2112         if (value) {
2113                 if (!is_valid_iface(value)) {
2114                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2115                                 value);
2116                         return -1;
2117                 }
2118                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2119         }
2120
2121         return 0;
2122 }
2123
2124 static int parse_user_mac(struct rte_ether_addr *user_mac,
2125                 const char *value)
2126 {
2127         unsigned int index = 0;
2128         char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
2129
2130         if (user_mac == NULL || value == NULL)
2131                 return 0;
2132
2133         strlcpy(mac_temp, value, sizeof(mac_temp));
2134         mac_byte = strtok(mac_temp, ":");
2135
2136         while ((mac_byte != NULL) &&
2137                         (strlen(mac_byte) <= 2) &&
2138                         (strlen(mac_byte) == strspn(mac_byte,
2139                                         ETH_TAP_CMP_MAC_FMT))) {
2140                 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
2141                 mac_byte = strtok(NULL, ":");
2142         }
2143
2144         return index;
2145 }
2146
2147 static int
2148 set_mac_type(const char *key __rte_unused,
2149              const char *value,
2150              void *extra_args)
2151 {
2152         struct rte_ether_addr *user_mac = extra_args;
2153
2154         if (!value)
2155                 return 0;
2156
2157         if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
2158                 static int iface_idx;
2159
2160                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
2161                 memcpy((char *)user_mac->addr_bytes, "\0dtap",
2162                         RTE_ETHER_ADDR_LEN);
2163                 user_mac->addr_bytes[RTE_ETHER_ADDR_LEN - 1] =
2164                         iface_idx++ + '0';
2165                 goto success;
2166         }
2167
2168         if (parse_user_mac(user_mac, value) != 6)
2169                 goto error;
2170 success:
2171         TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
2172         return 0;
2173
2174 error:
2175         TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
2176                 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
2177         return -1;
2178 }
2179
2180 /*
2181  * Open a TUN interface device. TUN PMD
2182  * 1) sets tap_type as false
2183  * 2) intakes iface as argument.
2184  * 3) as interface is virtual set speed to 10G
2185  */
2186 static int
2187 rte_pmd_tun_probe(struct rte_vdev_device *dev)
2188 {
2189         const char *name, *params;
2190         int ret;
2191         struct rte_kvargs *kvlist = NULL;
2192         char tun_name[RTE_ETH_NAME_MAX_LEN];
2193         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2194         struct rte_eth_dev *eth_dev;
2195
2196         name = rte_vdev_device_name(dev);
2197         params = rte_vdev_device_args(dev);
2198         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2199
2200         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
2201             strlen(params) == 0) {
2202                 eth_dev = rte_eth_dev_attach_secondary(name);
2203                 if (!eth_dev) {
2204                         TAP_LOG(ERR, "Failed to probe %s", name);
2205                         return -1;
2206                 }
2207                 eth_dev->dev_ops = &ops;
2208                 eth_dev->device = &dev->device;
2209                 rte_eth_dev_probing_finish(eth_dev);
2210                 return 0;
2211         }
2212
2213         /* use tun%d which causes kernel to choose next available */
2214         strlcpy(tun_name, DEFAULT_TUN_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2215
2216         if (params && (params[0] != '\0')) {
2217                 TAP_LOG(DEBUG, "parameters (%s)", params);
2218
2219                 kvlist = rte_kvargs_parse(params, valid_arguments);
2220                 if (kvlist) {
2221                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2222                                 ret = rte_kvargs_process(kvlist,
2223                                         ETH_TAP_IFACE_ARG,
2224                                         &set_interface_name,
2225                                         tun_name);
2226
2227                                 if (ret == -1)
2228                                         goto leave;
2229                         }
2230                 }
2231         }
2232         pmd_link.link_speed = ETH_SPEED_NUM_10G;
2233
2234         TAP_LOG(DEBUG, "Initializing pmd_tun for %s", name);
2235
2236         ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0,
2237                                  ETH_TUNTAP_TYPE_TUN);
2238
2239 leave:
2240         if (ret == -1) {
2241                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2242                         name, tun_name);
2243         }
2244         rte_kvargs_free(kvlist);
2245
2246         return ret;
2247 }
2248
2249 /* Request queue file descriptors from secondary to primary. */
2250 static int
2251 tap_mp_attach_queues(const char *port_name, struct rte_eth_dev *dev)
2252 {
2253         int ret;
2254         struct timespec timeout = {.tv_sec = 1, .tv_nsec = 0};
2255         struct rte_mp_msg request, *reply;
2256         struct rte_mp_reply replies;
2257         struct ipc_queues *request_param = (struct ipc_queues *)request.param;
2258         struct ipc_queues *reply_param;
2259         struct pmd_process_private *process_private = dev->process_private;
2260         int queue, fd_iterator;
2261
2262         /* Prepare the request */
2263         memset(&request, 0, sizeof(request));
2264         strlcpy(request.name, TAP_MP_KEY, sizeof(request.name));
2265         strlcpy(request_param->port_name, port_name,
2266                 sizeof(request_param->port_name));
2267         request.len_param = sizeof(*request_param);
2268         /* Send request and receive reply */
2269         ret = rte_mp_request_sync(&request, &replies, &timeout);
2270         if (ret < 0 || replies.nb_received != 1) {
2271                 TAP_LOG(ERR, "Failed to request queues from primary: %d",
2272                         rte_errno);
2273                 return -1;
2274         }
2275         reply = &replies.msgs[0];
2276         reply_param = (struct ipc_queues *)reply->param;
2277         TAP_LOG(DEBUG, "Received IPC reply for %s", reply_param->port_name);
2278
2279         /* Attach the queues from received file descriptors */
2280         if (reply_param->rxq_count + reply_param->txq_count != reply->num_fds) {
2281                 TAP_LOG(ERR, "Unexpected number of fds received");
2282                 return -1;
2283         }
2284
2285         dev->data->nb_rx_queues = reply_param->rxq_count;
2286         dev->data->nb_tx_queues = reply_param->txq_count;
2287         fd_iterator = 0;
2288         for (queue = 0; queue < reply_param->rxq_count; queue++)
2289                 process_private->rxq_fds[queue] = reply->fds[fd_iterator++];
2290         for (queue = 0; queue < reply_param->txq_count; queue++)
2291                 process_private->txq_fds[queue] = reply->fds[fd_iterator++];
2292         free(reply);
2293         return 0;
2294 }
2295
2296 /* Send the queue file descriptors from the primary process to secondary. */
2297 static int
2298 tap_mp_sync_queues(const struct rte_mp_msg *request, const void *peer)
2299 {
2300         struct rte_eth_dev *dev;
2301         struct pmd_process_private *process_private;
2302         struct rte_mp_msg reply;
2303         const struct ipc_queues *request_param =
2304                 (const struct ipc_queues *)request->param;
2305         struct ipc_queues *reply_param =
2306                 (struct ipc_queues *)reply.param;
2307         uint16_t port_id;
2308         int queue;
2309         int ret;
2310
2311         /* Get requested port */
2312         TAP_LOG(DEBUG, "Received IPC request for %s", request_param->port_name);
2313         ret = rte_eth_dev_get_port_by_name(request_param->port_name, &port_id);
2314         if (ret) {
2315                 TAP_LOG(ERR, "Failed to get port id for %s",
2316                         request_param->port_name);
2317                 return -1;
2318         }
2319         dev = &rte_eth_devices[port_id];
2320         process_private = dev->process_private;
2321
2322         /* Fill file descriptors for all queues */
2323         reply.num_fds = 0;
2324         reply_param->rxq_count = 0;
2325         if (dev->data->nb_rx_queues + dev->data->nb_tx_queues >
2326                         RTE_MP_MAX_FD_NUM){
2327                 TAP_LOG(ERR, "Number of rx/tx queues exceeds max number of fds");
2328                 return -1;
2329         }
2330
2331         for (queue = 0; queue < dev->data->nb_rx_queues; queue++) {
2332                 reply.fds[reply.num_fds++] = process_private->rxq_fds[queue];
2333                 reply_param->rxq_count++;
2334         }
2335         RTE_ASSERT(reply_param->rxq_count == dev->data->nb_rx_queues);
2336
2337         reply_param->txq_count = 0;
2338         for (queue = 0; queue < dev->data->nb_tx_queues; queue++) {
2339                 reply.fds[reply.num_fds++] = process_private->txq_fds[queue];
2340                 reply_param->txq_count++;
2341         }
2342         RTE_ASSERT(reply_param->txq_count == dev->data->nb_tx_queues);
2343
2344         /* Send reply */
2345         strlcpy(reply.name, request->name, sizeof(reply.name));
2346         strlcpy(reply_param->port_name, request_param->port_name,
2347                 sizeof(reply_param->port_name));
2348         reply.len_param = sizeof(*reply_param);
2349         if (rte_mp_reply(&reply, peer) < 0) {
2350                 TAP_LOG(ERR, "Failed to reply an IPC request to sync queues");
2351                 return -1;
2352         }
2353         return 0;
2354 }
2355
2356 /* Open a TAP interface device.
2357  */
2358 static int
2359 rte_pmd_tap_probe(struct rte_vdev_device *dev)
2360 {
2361         const char *name, *params;
2362         int ret;
2363         struct rte_kvargs *kvlist = NULL;
2364         int speed;
2365         char tap_name[RTE_ETH_NAME_MAX_LEN];
2366         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2367         struct rte_ether_addr user_mac = { .addr_bytes = {0} };
2368         struct rte_eth_dev *eth_dev;
2369         int tap_devices_count_increased = 0;
2370
2371         name = rte_vdev_device_name(dev);
2372         params = rte_vdev_device_args(dev);
2373
2374         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
2375                 eth_dev = rte_eth_dev_attach_secondary(name);
2376                 if (!eth_dev) {
2377                         TAP_LOG(ERR, "Failed to probe %s", name);
2378                         return -1;
2379                 }
2380                 eth_dev->dev_ops = &ops;
2381                 eth_dev->device = &dev->device;
2382                 eth_dev->rx_pkt_burst = pmd_rx_burst;
2383                 eth_dev->tx_pkt_burst = pmd_tx_burst;
2384                 if (!rte_eal_primary_proc_alive(NULL)) {
2385                         TAP_LOG(ERR, "Primary process is missing");
2386                         return -1;
2387                 }
2388                 eth_dev->process_private = (struct pmd_process_private *)
2389                         rte_zmalloc_socket(name,
2390                                 sizeof(struct pmd_process_private),
2391                                 RTE_CACHE_LINE_SIZE,
2392                                 eth_dev->device->numa_node);
2393                 if (eth_dev->process_private == NULL) {
2394                         TAP_LOG(ERR,
2395                                 "Failed to alloc memory for process private");
2396                         return -1;
2397                 }
2398
2399                 ret = tap_mp_attach_queues(name, eth_dev);
2400                 if (ret != 0)
2401                         return -1;
2402                 rte_eth_dev_probing_finish(eth_dev);
2403                 return 0;
2404         }
2405
2406         speed = ETH_SPEED_NUM_10G;
2407
2408         /* use tap%d which causes kernel to choose next available */
2409         strlcpy(tap_name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2410         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2411
2412         if (params && (params[0] != '\0')) {
2413                 TAP_LOG(DEBUG, "parameters (%s)", params);
2414
2415                 kvlist = rte_kvargs_parse(params, valid_arguments);
2416                 if (kvlist) {
2417                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2418                                 ret = rte_kvargs_process(kvlist,
2419                                                          ETH_TAP_IFACE_ARG,
2420                                                          &set_interface_name,
2421                                                          tap_name);
2422                                 if (ret == -1)
2423                                         goto leave;
2424                         }
2425
2426                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
2427                                 ret = rte_kvargs_process(kvlist,
2428                                                          ETH_TAP_REMOTE_ARG,
2429                                                          &set_remote_iface,
2430                                                          remote_iface);
2431                                 if (ret == -1)
2432                                         goto leave;
2433                         }
2434
2435                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
2436                                 ret = rte_kvargs_process(kvlist,
2437                                                          ETH_TAP_MAC_ARG,
2438                                                          &set_mac_type,
2439                                                          &user_mac);
2440                                 if (ret == -1)
2441                                         goto leave;
2442                         }
2443                 }
2444         }
2445         pmd_link.link_speed = speed;
2446
2447         TAP_LOG(DEBUG, "Initializing pmd_tap for %s", name);
2448
2449         /* Register IPC feed callback */
2450         if (!tap_devices_count) {
2451                 ret = rte_mp_action_register(TAP_MP_KEY, tap_mp_sync_queues);
2452                 if (ret < 0 && rte_errno != ENOTSUP) {
2453                         TAP_LOG(ERR, "tap: Failed to register IPC callback: %s",
2454                                 strerror(rte_errno));
2455                         goto leave;
2456                 }
2457         }
2458         tap_devices_count++;
2459         tap_devices_count_increased = 1;
2460         ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
2461                 ETH_TUNTAP_TYPE_TAP);
2462
2463 leave:
2464         if (ret == -1) {
2465                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2466                         name, tap_name);
2467                 if (tap_devices_count_increased == 1) {
2468                         if (tap_devices_count == 1)
2469                                 rte_mp_action_unregister(TAP_MP_KEY);
2470                         tap_devices_count--;
2471                 }
2472         }
2473         rte_kvargs_free(kvlist);
2474
2475         return ret;
2476 }
2477
2478 /* detach a TUNTAP device.
2479  */
2480 static int
2481 rte_pmd_tap_remove(struct rte_vdev_device *dev)
2482 {
2483         struct rte_eth_dev *eth_dev = NULL;
2484         struct pmd_internals *internals;
2485
2486         /* find the ethdev entry */
2487         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
2488         if (!eth_dev)
2489                 return -ENODEV;
2490
2491         /* mac_addrs must not be freed alone because part of dev_private */
2492         eth_dev->data->mac_addrs = NULL;
2493
2494         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2495                 return rte_eth_dev_release_port(eth_dev);
2496
2497         tap_dev_close(eth_dev);
2498
2499         internals = eth_dev->data->dev_private;
2500         TAP_LOG(DEBUG, "Closing %s Ethernet device on numa %u",
2501                 tuntap_types[internals->type], rte_socket_id());
2502
2503         close(internals->ioctl_sock);
2504         rte_free(eth_dev->process_private);
2505         if (tap_devices_count == 1)
2506                 rte_mp_action_unregister(TAP_MP_KEY);
2507         tap_devices_count--;
2508         rte_eth_dev_release_port(eth_dev);
2509
2510         return 0;
2511 }
2512
2513 static struct rte_vdev_driver pmd_tun_drv = {
2514         .probe = rte_pmd_tun_probe,
2515         .remove = rte_pmd_tap_remove,
2516 };
2517
2518 static struct rte_vdev_driver pmd_tap_drv = {
2519         .probe = rte_pmd_tap_probe,
2520         .remove = rte_pmd_tap_remove,
2521 };
2522
2523 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
2524 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
2525 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
2526 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
2527                               ETH_TAP_IFACE_ARG "=<string> ");
2528 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
2529                               ETH_TAP_IFACE_ARG "=<string> "
2530                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
2531                               ETH_TAP_REMOTE_ARG "=<string>");
2532 RTE_LOG_REGISTER(tap_logtype, pmd.net.tap, NOTICE);