drivers/net/tap/rte_eth_tap.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2016-2017 Intel Corporation
   3  */
   4
   5 #include <rte_atomic.h>
   6 #include <rte_branch_prediction.h>
   7 #include <rte_byteorder.h>
   8 #include <rte_common.h>
   9 #include <rte_mbuf.h>
  10 #include <rte_ethdev_driver.h>
  11 #include <rte_ethdev_vdev.h>
  12 #include <rte_malloc.h>
  13 #include <rte_bus_vdev.h>
  14 #include <rte_kvargs.h>
  15 #include <rte_net.h>
  16 #include <rte_debug.h>
  17 #include <rte_ip.h>
  18 #include <rte_string_fns.h>
  19 #include <rte_ethdev.h>
  20 #include <rte_errno.h>
  21 #include <rte_cycles.h>
  22
  23 #include <sys/types.h>
  24 #include <sys/stat.h>
  25 #include <sys/socket.h>
  26 #include <sys/ioctl.h>
  27 #include <sys/utsname.h>
  28 #include <sys/mman.h>
  29 #include <errno.h>
  30 #include <signal.h>
  31 #include <stdbool.h>
  32 #include <stdint.h>
  33 #include <sys/uio.h>
  34 #include <unistd.h>
  35 #include <arpa/inet.h>
  36 #include <net/if.h>
  37 #include <linux/if_tun.h>
  38 #include <linux/if_ether.h>
  39 #include <fcntl.h>
  40 #include <ctype.h>
  41
  42 #include <tap_rss.h>
  43 #include <rte_eth_tap.h>
  44 #include <tap_flow.h>
  45 #include <tap_netlink.h>
  46 #include <tap_tcmsgs.h>
  47
  48 /* Linux based path to the TUN device */
  49 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  50 #define DEFAULT_TAP_NAME        "dtap"
  51 #define DEFAULT_TUN_NAME        "dtun"
  52
  53 #define ETH_TAP_IFACE_ARG       "iface"
  54 #define ETH_TAP_REMOTE_ARG      "remote"
  55 #define ETH_TAP_MAC_ARG         "mac"
  56 #define ETH_TAP_MAC_FIXED       "fixed"
  57
  58 #define ETH_TAP_USR_MAC_FMT     "xx:xx:xx:xx:xx:xx"
  59 #define ETH_TAP_CMP_MAC_FMT     "0123456789ABCDEFabcdef"
  60 #define ETH_TAP_MAC_ARG_FMT     ETH_TAP_MAC_FIXED "|" ETH_TAP_USR_MAC_FMT
  61
  62 #define TAP_GSO_MBUFS_PER_CORE  128
  63 #define TAP_GSO_MBUF_SEG_SIZE   128
  64 #define TAP_GSO_MBUF_CACHE_SIZE 4
  65 #define TAP_GSO_MBUFS_NUM \
  66         (TAP_GSO_MBUFS_PER_CORE * TAP_GSO_MBUF_CACHE_SIZE)
  67
  68 /* IPC key for queue fds sync */
  69 #define TAP_MP_KEY "tap_mp_sync_queues"
  70
  71 #define TAP_IOV_DEFAULT_MAX 1024
  72
  73 static int tap_devices_count;
  74
  75 static const char *tuntap_types[ETH_TUNTAP_TYPE_MAX] = {
  76         "UNKNOWN", "TUN", "TAP"
  77 };
  78
  79 static const char *valid_arguments[] = {
  80         ETH_TAP_IFACE_ARG,
  81         ETH_TAP_REMOTE_ARG,
  82         ETH_TAP_MAC_ARG,
  83         NULL
  84 };
  85
  86 static volatile uint32_t tap_trigger;   /* Rx trigger */
  87
  88 static struct rte_eth_link pmd_link = {
  89         .link_speed = ETH_SPEED_NUM_10G,
  90         .link_duplex = ETH_LINK_FULL_DUPLEX,
  91         .link_status = ETH_LINK_DOWN,
  92         .link_autoneg = ETH_LINK_FIXED,
  93 };
  94
  95 static void
  96 tap_trigger_cb(int sig __rte_unused)
  97 {
  98         /* Valid trigger values are nonzero */
  99         tap_trigger = (tap_trigger + 1) | 0x80000000;
 100 }
 101
 102 /* Specifies on what netdevices the ioctl should be applied */
 103 enum ioctl_mode {
 104         LOCAL_AND_REMOTE,
 105         LOCAL_ONLY,
 106         REMOTE_ONLY,
 107 };
 108
 109 /* Message header to synchronize queues via IPC */
 110 struct ipc_queues {
 111         char port_name[RTE_DEV_NAME_MAX_LEN];
 112         int rxq_count;
 113         int txq_count;
 114         /*
 115          * The file descriptors are in the dedicated part
 116          * of the Unix message to be translated by the kernel.
 117          */
 118 };
 119
 120 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
 121
 122 /**
 123  * Tun/Tap allocation routine
 124  *
 125  * @param[in] pmd
 126  *   Pointer to private structure.
 127  *
 128  * @param[in] is_keepalive
 129  *   Keepalive flag
 130  *
 131  * @return
 132  *   -1 on failure, fd on success
 133  */
 134 static int
 135 tun_alloc(struct pmd_internals *pmd, int is_keepalive)
 136 {
 137         struct ifreq ifr;
 138 #ifdef IFF_MULTI_QUEUE
 139         unsigned int features;
 140 #endif
 141         int fd, signo, flags;
 142
 143         memset(&ifr, 0, sizeof(struct ifreq));
 144
 145         /*
 146          * Do not set IFF_NO_PI as packet information header will be needed
 147          * to check if a received packet has been truncated.
 148          */
 149         ifr.ifr_flags = (pmd->type == ETH_TUNTAP_TYPE_TAP) ?
 150                 IFF_TAP : IFF_TUN | IFF_POINTOPOINT;
 151         strlcpy(ifr.ifr_name, pmd->name, IFNAMSIZ);
 152
 153         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 154         if (fd < 0) {
 155                 TAP_LOG(ERR, "Unable to open %s interface", TUN_TAP_DEV_PATH);
 156                 goto error;
 157         }
 158
 159 #ifdef IFF_MULTI_QUEUE
 160         /* Grab the TUN features to verify we can work multi-queue */
 161         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 162                 TAP_LOG(ERR, "unable to get TUN/TAP features");
 163                 goto error;
 164         }
 165         TAP_LOG(DEBUG, "%s Features %08x", TUN_TAP_DEV_PATH, features);
 166
 167         if (features & IFF_MULTI_QUEUE) {
 168                 TAP_LOG(DEBUG, "  Multi-queue support for %d queues",
 169                         RTE_PMD_TAP_MAX_QUEUES);
 170                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 171         } else
 172 #endif
 173         {
 174                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 175                 TAP_LOG(DEBUG, "  Single queue only support");
 176         }
 177
 178         /* Set the TUN/TAP configuration and set the name if needed */
 179         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 180                 TAP_LOG(WARNING, "Unable to set TUNSETIFF for %s: %s",
 181                         ifr.ifr_name, strerror(errno));
 182                 goto error;
 183         }
 184
 185         /*
 186          * Name passed to kernel might be wildcard like dtun%d
 187          * and need to find the resulting device.
 188          */
 189         TAP_LOG(DEBUG, "Device name is '%s'", ifr.ifr_name);
 190         strlcpy(pmd->name, ifr.ifr_name, RTE_ETH_NAME_MAX_LEN);
 191
 192         if (is_keepalive) {
 193                 /*
 194                  * Detach the TUN/TAP keep-alive queue
 195                  * to avoid traffic through it
 196                  */
 197                 ifr.ifr_flags = IFF_DETACH_QUEUE;
 198                 if (ioctl(fd, TUNSETQUEUE, (void *)&ifr) < 0) {
 199                         TAP_LOG(WARNING,
 200                                 "Unable to detach keep-alive queue for %s: %s",
 201                                 ifr.ifr_name, strerror(errno));
 202                         goto error;
 203                 }
 204         }
 205
 206         flags = fcntl(fd, F_GETFL);
 207         if (flags == -1) {
 208                 TAP_LOG(WARNING,
 209                         "Unable to get %s current flags\n",
 210                         ifr.ifr_name);
 211                 goto error;
 212         }
 213
 214         /* Always set the file descriptor to non-blocking */
 215         flags |= O_NONBLOCK;
 216         if (fcntl(fd, F_SETFL, flags) < 0) {
 217                 TAP_LOG(WARNING,
 218                         "Unable to set %s to nonblocking: %s",
 219                         ifr.ifr_name, strerror(errno));
 220                 goto error;
 221         }
 222
 223         /* Find a free realtime signal */
 224         for (signo = SIGRTMIN + 1; signo < SIGRTMAX; signo++) {
 225                 struct sigaction sa;
 226
 227                 if (sigaction(signo, NULL, &sa) == -1) {
 228                         TAP_LOG(WARNING,
 229                                 "Unable to get current rt-signal %d handler",
 230                                 signo);
 231                         goto error;
 232                 }
 233
 234                 /* Already have the handler we want on this signal  */
 235                 if (sa.sa_handler == tap_trigger_cb)
 236                         break;
 237
 238                 /* Is handler in use by application */
 239                 if (sa.sa_handler != SIG_DFL) {
 240                         TAP_LOG(DEBUG,
 241                                 "Skipping used rt-signal %d", signo);
 242                         continue;
 243                 }
 244
 245                 sa = (struct sigaction) {
 246                         .sa_flags = SA_RESTART,
 247                         .sa_handler = tap_trigger_cb,
 248                 };
 249
 250                 if (sigaction(signo, &sa, NULL) == -1) {
 251                         TAP_LOG(WARNING,
 252                                 "Unable to set rt-signal %d handler\n", signo);
 253                         goto error;
 254                 }
 255
 256                 /* Found a good signal to use */
 257                 TAP_LOG(DEBUG,
 258                         "Using rt-signal %d", signo);
 259                 break;
 260         }
 261
 262         if (signo == SIGRTMAX) {
 263                 TAP_LOG(WARNING, "All rt-signals are in use\n");
 264
 265                 /* Disable trigger globally in case of error */
 266                 tap_trigger = 0;
 267                 TAP_LOG(NOTICE, "No Rx trigger signal available\n");
 268         } else {
 269                 /* Enable signal on file descriptor */
 270                 if (fcntl(fd, F_SETSIG, signo) < 0) {
 271                         TAP_LOG(WARNING, "Unable to set signo %d for fd %d: %s",
 272                                 signo, fd, strerror(errno));
 273                         goto error;
 274                 }
 275                 if (fcntl(fd, F_SETFL, flags | O_ASYNC) < 0) {
 276                         TAP_LOG(WARNING, "Unable to set fcntl flags: %s",
 277                                 strerror(errno));
 278                         goto error;
 279                 }
 280
 281                 if (fcntl(fd, F_SETOWN, getpid()) < 0) {
 282                         TAP_LOG(WARNING, "Unable to set fcntl owner: %s",
 283                                 strerror(errno));
 284                         goto error;
 285                 }
 286         }
 287         return fd;
 288
 289 error:
 290         if (fd >= 0)
 291                 close(fd);
 292         return -1;
 293 }
 294
 295 static void
 296 tap_verify_csum(struct rte_mbuf *mbuf)
 297 {
 298         uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
 299         uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
 300         uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
 301         unsigned int l2_len = sizeof(struct rte_ether_hdr);
 302         unsigned int l3_len;
 303         uint16_t cksum = 0;
 304         void *l3_hdr;
 305         void *l4_hdr;
 306
 307         if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
 308                 l2_len += 4;
 309         else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
 310                 l2_len += 8;
 311         /* Don't verify checksum for packets with discontinuous L2 header */
 312         if (unlikely(l2_len + sizeof(struct rte_ipv4_hdr) >
 313                      rte_pktmbuf_data_len(mbuf)))
 314                 return;
 315         l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
 316         if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 317                 struct rte_ipv4_hdr *iph = l3_hdr;
 318
 319                 l3_len = rte_ipv4_hdr_len(iph);
 320                 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
 321                         return;
 322                 /* check that the total length reported by header is not
 323                  * greater than the total received size
 324                  */
 325                 if (l2_len + rte_be_to_cpu_16(iph->total_length) >
 326                                 rte_pktmbuf_data_len(mbuf))
 327                         return;
 328
 329                 cksum = ~rte_raw_cksum(iph, l3_len);
 330                 mbuf->ol_flags |= cksum ?
 331                         PKT_RX_IP_CKSUM_BAD :
 332                         PKT_RX_IP_CKSUM_GOOD;
 333         } else if (l3 == RTE_PTYPE_L3_IPV6) {
 334                 struct rte_ipv6_hdr *iph = l3_hdr;
 335
 336                 l3_len = sizeof(struct rte_ipv6_hdr);
 337                 /* check that the total length reported by header is not
 338                  * greater than the total received size
 339                  */
 340                 if (l2_len + l3_len + rte_be_to_cpu_16(iph->payload_len) >
 341                                 rte_pktmbuf_data_len(mbuf))
 342                         return;
 343         } else {
 344                 /* IPv6 extensions are not supported */
 345                 return;
 346         }
 347         if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
 348                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 349                 /* Don't verify checksum for multi-segment packets. */
 350                 if (mbuf->nb_segs > 1)
 351                         return;
 352                 if (l3 == RTE_PTYPE_L3_IPV4)
 353                         cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
 354                 else if (l3 == RTE_PTYPE_L3_IPV6)
 355                         cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
 356                 mbuf->ol_flags |= cksum ?
 357                         PKT_RX_L4_CKSUM_BAD :
 358                         PKT_RX_L4_CKSUM_GOOD;
 359         }
 360 }
 361
 362 static uint64_t
 363 tap_rx_offload_get_port_capa(void)
 364 {
 365         /*
 366          * No specific port Rx offload capabilities.
 367          */
 368         return 0;
 369 }
 370
 371 static uint64_t
 372 tap_rx_offload_get_queue_capa(void)
 373 {
 374         return DEV_RX_OFFLOAD_SCATTER |
 375                DEV_RX_OFFLOAD_IPV4_CKSUM |
 376                DEV_RX_OFFLOAD_UDP_CKSUM |
 377                DEV_RX_OFFLOAD_TCP_CKSUM;
 378 }
 379
 380 static void
 381 tap_rxq_pool_free(struct rte_mbuf *pool)
 382 {
 383         struct rte_mbuf *mbuf = pool;
 384         uint16_t nb_segs = 1;
 385
 386         if (mbuf == NULL)
 387                 return;
 388
 389         while (mbuf->next) {
 390                 mbuf = mbuf->next;
 391                 nb_segs++;
 392         }
 393         pool->nb_segs = nb_segs;
 394         rte_pktmbuf_free(pool);
 395 }
 396
 397 /* Callback to handle the rx burst of packets to the correct interface and
 398  * file descriptor(s) in a multi-queue setup.
 399  */
 400 static uint16_t
 401 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 402 {
 403         struct rx_queue *rxq = queue;
 404         struct pmd_process_private *process_private;
 405         uint16_t num_rx;
 406         unsigned long num_rx_bytes = 0;
 407         uint32_t trigger = tap_trigger;
 408
 409         if (trigger == rxq->trigger_seen)
 410                 return 0;
 411
 412         process_private = rte_eth_devices[rxq->in_port].process_private;
 413         for (num_rx = 0; num_rx < nb_pkts; ) {
 414                 struct rte_mbuf *mbuf = rxq->pool;
 415                 struct rte_mbuf *seg = NULL;
 416                 struct rte_mbuf *new_tail = NULL;
 417                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 418                 int len;
 419
 420                 len = readv(process_private->rxq_fds[rxq->queue_id],
 421                         *rxq->iovecs,
 422                         1 + (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
 423                              rxq->nb_rx_desc : 1));
 424                 if (len < (int)sizeof(struct tun_pi))
 425                         break;
 426
 427                 /* Packet couldn't fit in the provided mbuf */
 428                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 429                         rxq->stats.ierrors++;
 430                         continue;
 431                 }
 432
 433                 len -= sizeof(struct tun_pi);
 434
 435                 mbuf->pkt_len = len;
 436                 mbuf->port = rxq->in_port;
 437                 while (1) {
 438                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 439
 440                         if (unlikely(!buf)) {
 441                                 rxq->stats.rx_nombuf++;
 442                                 /* No new buf has been allocated: do nothing */
 443                                 if (!new_tail || !seg)
 444                                         goto end;
 445
 446                                 seg->next = NULL;
 447                                 tap_rxq_pool_free(mbuf);
 448
 449                                 goto end;
 450                         }
 451                         seg = seg ? seg->next : mbuf;
 452                         if (rxq->pool == mbuf)
 453                                 rxq->pool = buf;
 454                         if (new_tail)
 455                                 new_tail->next = buf;
 456                         new_tail = buf;
 457                         new_tail->next = seg->next;
 458
 459                         /* iovecs[0] is reserved for packet info (pi) */
 460                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 461                                 buf->buf_len - data_off;
 462                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 463                                 (char *)buf->buf_addr + data_off;
 464
 465                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 466                         seg->data_off = data_off;
 467
 468                         len -= seg->data_len;
 469                         if (len <= 0)
 470                                 break;
 471                         mbuf->nb_segs++;
 472                         /* First segment has headroom, not the others */
 473                         data_off = 0;
 474                 }
 475                 seg->next = NULL;
 476                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 477                                                       RTE_PTYPE_ALL_MASK);
 478                 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
 479                         tap_verify_csum(mbuf);
 480
 481                 /* account for the receive frame */
 482                 bufs[num_rx++] = mbuf;
 483                 num_rx_bytes += mbuf->pkt_len;
 484         }
 485 end:
 486         rxq->stats.ipackets += num_rx;
 487         rxq->stats.ibytes += num_rx_bytes;
 488
 489         if (trigger && num_rx < nb_pkts)
 490                 rxq->trigger_seen = trigger;
 491
 492         return num_rx;
 493 }
 494
 495 static uint64_t
 496 tap_tx_offload_get_port_capa(void)
 497 {
 498         /*
 499          * No specific port Tx offload capabilities.
 500          */
 501         return 0;
 502 }
 503
 504 static uint64_t
 505 tap_tx_offload_get_queue_capa(void)
 506 {
 507         return DEV_TX_OFFLOAD_MULTI_SEGS |
 508                DEV_TX_OFFLOAD_IPV4_CKSUM |
 509                DEV_TX_OFFLOAD_UDP_CKSUM |
 510                DEV_TX_OFFLOAD_TCP_CKSUM |
 511                DEV_TX_OFFLOAD_TCP_TSO;
 512 }
 513
 514 /* Finalize l4 checksum calculation */
 515 static void
 516 tap_tx_l4_cksum(uint16_t *l4_cksum, uint16_t l4_phdr_cksum,
 517                 uint32_t l4_raw_cksum)
 518 {
 519         if (l4_cksum) {
 520                 uint32_t cksum;
 521
 522                 cksum = __rte_raw_cksum_reduce(l4_raw_cksum);
 523                 cksum += l4_phdr_cksum;
 524
 525                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 526                 cksum = (~cksum) & 0xffff;
 527                 if (cksum == 0)
 528                         cksum = 0xffff;
 529                 *l4_cksum = cksum;
 530         }
 531 }
 532
 533 /* Accumaulate L4 raw checksums */
 534 static void
 535 tap_tx_l4_add_rcksum(char *l4_data, unsigned int l4_len, uint16_t *l4_cksum,
 536                         uint32_t *l4_raw_cksum)
 537 {
 538         if (l4_cksum == NULL)
 539                 return;
 540
 541         *l4_raw_cksum = __rte_raw_cksum(l4_data, l4_len, *l4_raw_cksum);
 542 }
 543
 544 /* L3 and L4 pseudo headers checksum offloads */
 545 static void
 546 tap_tx_l3_cksum(char *packet, uint64_t ol_flags, unsigned int l2_len,
 547                 unsigned int l3_len, unsigned int l4_len, uint16_t **l4_cksum,
 548                 uint16_t *l4_phdr_cksum, uint32_t *l4_raw_cksum)
 549 {
 550         void *l3_hdr = packet + l2_len;
 551
 552         if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
 553                 struct rte_ipv4_hdr *iph = l3_hdr;
 554                 uint16_t cksum;
 555
 556                 iph->hdr_checksum = 0;
 557                 cksum = rte_raw_cksum(iph, l3_len);
 558                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 559         }
 560         if (ol_flags & PKT_TX_L4_MASK) {
 561                 void *l4_hdr;
 562
 563                 l4_hdr = packet + l2_len + l3_len;
 564                 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
 565                         *l4_cksum = &((struct rte_udp_hdr *)l4_hdr)->dgram_cksum;
 566                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 567                         *l4_cksum = &((struct rte_tcp_hdr *)l4_hdr)->cksum;
 568                 else
 569                         return;
 570                 **l4_cksum = 0;
 571                 if (ol_flags & PKT_TX_IPV4)
 572                         *l4_phdr_cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 573                 else
 574                         *l4_phdr_cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 575                 *l4_raw_cksum = __rte_raw_cksum(l4_hdr, l4_len, 0);
 576         }
 577 }
 578
 579 static inline int
 580 tap_write_mbufs(struct tx_queue *txq, uint16_t num_mbufs,
 581                         struct rte_mbuf **pmbufs,
 582                         uint16_t *num_packets, unsigned long *num_tx_bytes)
 583 {
 584         int i;
 585         uint16_t l234_hlen;
 586         struct pmd_process_private *process_private;
 587
 588         process_private = rte_eth_devices[txq->out_port].process_private;
 589
 590         for (i = 0; i < num_mbufs; i++) {
 591                 struct rte_mbuf *mbuf = pmbufs[i];
 592                 struct iovec iovecs[mbuf->nb_segs + 2];
 593                 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
 594                 struct rte_mbuf *seg = mbuf;
 595                 char m_copy[mbuf->data_len];
 596                 int proto;
 597                 int n;
 598                 int j;
 599                 int k; /* current index in iovecs for copying segments */
 600                 uint16_t seg_len; /* length of first segment */
 601                 uint16_t nb_segs;
 602                 uint16_t *l4_cksum; /* l4 checksum (pseudo header + payload) */
 603                 uint32_t l4_raw_cksum = 0; /* TCP/UDP payload raw checksum */
 604                 uint16_t l4_phdr_cksum = 0; /* TCP/UDP pseudo header checksum */
 605                 uint16_t is_cksum = 0; /* in case cksum should be offloaded */
 606
 607                 l4_cksum = NULL;
 608                 if (txq->type == ETH_TUNTAP_TYPE_TUN) {
 609                         /*
 610                          * TUN and TAP are created with IFF_NO_PI disabled.
 611                          * For TUN PMD this mandatory as fields are used by
 612                          * Kernel tun.c to determine whether its IP or non IP
 613                          * packets.
 614                          *
 615                          * The logic fetches the first byte of data from mbuf
 616                          * then compares whether its v4 or v6. If first byte
 617                          * is 4 or 6, then protocol field is updated.
 618                          */
 619                         char *buff_data = rte_pktmbuf_mtod(seg, void *);
 620                         proto = (*buff_data & 0xf0);
 621                         pi.proto = (proto == 0x40) ?
 622                                 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4) :
 623                                 ((proto == 0x60) ?
 624                                         rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6) :
 625                                         0x00);
 626                 }
 627
 628                 k = 0;
 629                 iovecs[k].iov_base = &pi;
 630                 iovecs[k].iov_len = sizeof(pi);
 631                 k++;
 632
 633                 nb_segs = mbuf->nb_segs;
 634                 if (txq->csum &&
 635                     ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 636                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 637                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
 638                         is_cksum = 1;
 639
 640                         /* Support only packets with at least layer 4
 641                          * header included in the first segment
 642                          */
 643                         seg_len = rte_pktmbuf_data_len(mbuf);
 644                         l234_hlen = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
 645                         if (seg_len < l234_hlen)
 646                                 return -1;
 647
 648                         /* To change checksums, work on a * copy of l2, l3
 649                          * headers + l4 pseudo header
 650                          */
 651                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 652                                         l234_hlen);
 653                         tap_tx_l3_cksum(m_copy, mbuf->ol_flags,
 654                                        mbuf->l2_len, mbuf->l3_len, mbuf->l4_len,
 655                                        &l4_cksum, &l4_phdr_cksum,
 656                                        &l4_raw_cksum);
 657                         iovecs[k].iov_base = m_copy;
 658                         iovecs[k].iov_len = l234_hlen;
 659                         k++;
 660
 661                         /* Update next iovecs[] beyond l2, l3, l4 headers */
 662                         if (seg_len > l234_hlen) {
 663                                 iovecs[k].iov_len = seg_len - l234_hlen;
 664                                 iovecs[k].iov_base =
 665                                         rte_pktmbuf_mtod(seg, char *) +
 666                                                 l234_hlen;
 667                                 tap_tx_l4_add_rcksum(iovecs[k].iov_base,
 668                                         iovecs[k].iov_len, l4_cksum,
 669                                         &l4_raw_cksum);
 670                                 k++;
 671                                 nb_segs++;
 672                         }
 673                         seg = seg->next;
 674                 }
 675
 676                 for (j = k; j <= nb_segs; j++) {
 677                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 678                         iovecs[j].iov_base = rte_pktmbuf_mtod(seg, void *);
 679                         if (is_cksum)
 680                                 tap_tx_l4_add_rcksum(iovecs[j].iov_base,
 681                                         iovecs[j].iov_len, l4_cksum,
 682                                         &l4_raw_cksum);
 683                         seg = seg->next;
 684                 }
 685
 686                 if (is_cksum)
 687                         tap_tx_l4_cksum(l4_cksum, l4_phdr_cksum, l4_raw_cksum);
 688
 689                 /* copy the tx frame data */
 690                 n = writev(process_private->txq_fds[txq->queue_id], iovecs, j);
 691                 if (n <= 0)
 692                         return -1;
 693
 694                 (*num_packets)++;
 695                 (*num_tx_bytes) += rte_pktmbuf_pkt_len(mbuf);
 696         }
 697         return 0;
 698 }
 699
 700 /* Callback to handle sending packets from the tap interface
 701  */
 702 static uint16_t
 703 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 704 {
 705         struct tx_queue *txq = queue;
 706         uint16_t num_tx = 0;
 707         uint16_t num_packets = 0;
 708         unsigned long num_tx_bytes = 0;
 709         uint32_t max_size;
 710         int i;
 711
 712         if (unlikely(nb_pkts == 0))
 713                 return 0;
 714
 715         struct rte_mbuf *gso_mbufs[MAX_GSO_MBUFS];
 716         max_size = *txq->mtu + (RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN + 4);
 717         for (i = 0; i < nb_pkts; i++) {
 718                 struct rte_mbuf *mbuf_in = bufs[num_tx];
 719                 struct rte_mbuf **mbuf;
 720                 uint16_t num_mbufs = 0;
 721                 uint16_t tso_segsz = 0;
 722                 int ret;
 723                 int num_tso_mbufs;
 724                 uint16_t hdrs_len;
 725                 uint64_t tso;
 726
 727                 tso = mbuf_in->ol_flags & PKT_TX_TCP_SEG;
 728                 if (tso) {
 729                         struct rte_gso_ctx *gso_ctx = &txq->gso_ctx;
 730
 731                         /* TCP segmentation implies TCP checksum offload */
 732                         mbuf_in->ol_flags |= PKT_TX_TCP_CKSUM;
 733
 734                         /* gso size is calculated without RTE_ETHER_CRC_LEN */
 735                         hdrs_len = mbuf_in->l2_len + mbuf_in->l3_len +
 736                                         mbuf_in->l4_len;
 737                         tso_segsz = mbuf_in->tso_segsz + hdrs_len;
 738                         if (unlikely(tso_segsz == hdrs_len) ||
 739                                 tso_segsz > *txq->mtu) {
 740                                 txq->stats.errs++;
 741                                 break;
 742                         }
 743                         gso_ctx->gso_size = tso_segsz;
 744                         /* 'mbuf_in' packet to segment */
 745                         num_tso_mbufs = rte_gso_segment(mbuf_in,
 746                                 gso_ctx, /* gso control block */
 747                                 (struct rte_mbuf **)&gso_mbufs, /* out mbufs */
 748                                 RTE_DIM(gso_mbufs)); /* max tso mbufs */
 749
 750                         /* ret contains the number of new created mbufs */
 751                         if (num_tso_mbufs < 0)
 752                                 break;
 753
 754                         if (num_tso_mbufs >= 1) {
 755                                 mbuf = gso_mbufs;
 756                                 num_mbufs = num_tso_mbufs;
 757                         } else {
 758                                 /* 0 means it can be transmitted directly
 759                                  * without gso.
 760                                  */
 761                                 mbuf = &mbuf_in;
 762                                 num_mbufs = 1;
 763                         }
 764                 } else {
 765                         /* stats.errs will be incremented */
 766                         if (rte_pktmbuf_pkt_len(mbuf_in) > max_size)
 767                                 break;
 768
 769                         /* ret 0 indicates no new mbufs were created */
 770                         num_tso_mbufs = 0;
 771                         mbuf = &mbuf_in;
 772                         num_mbufs = 1;
 773                 }
 774
 775                 ret = tap_write_mbufs(txq, num_mbufs, mbuf,
 776                                 &num_packets, &num_tx_bytes);
 777                 if (ret == -1) {
 778                         txq->stats.errs++;
 779                         /* free tso mbufs */
 780                         if (num_tso_mbufs > 0)
 781                                 rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
 782                         break;
 783                 }
 784                 num_tx++;
 785                 /* free original mbuf */
 786                 rte_pktmbuf_free(mbuf_in);
 787                 /* free tso mbufs */
 788                 if (num_tso_mbufs > 0)
 789                         rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
 790         }
 791
 792         txq->stats.opackets += num_packets;
 793         txq->stats.errs += nb_pkts - num_tx;
 794         txq->stats.obytes += num_tx_bytes;
 795
 796         return num_tx;
 797 }
 798
 799 static const char *
 800 tap_ioctl_req2str(unsigned long request)
 801 {
 802         switch (request) {
 803         case SIOCSIFFLAGS:
 804                 return "SIOCSIFFLAGS";
 805         case SIOCGIFFLAGS:
 806                 return "SIOCGIFFLAGS";
 807         case SIOCGIFHWADDR:
 808                 return "SIOCGIFHWADDR";
 809         case SIOCSIFHWADDR:
 810                 return "SIOCSIFHWADDR";
 811         case SIOCSIFMTU:
 812                 return "SIOCSIFMTU";
 813         }
 814         return "UNKNOWN";
 815 }
 816
 817 static int
 818 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 819           struct ifreq *ifr, int set, enum ioctl_mode mode)
 820 {
 821         short req_flags = ifr->ifr_flags;
 822         int remote = pmd->remote_if_index &&
 823                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 824
 825         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 826                 return 0;
 827         /*
 828          * If there is a remote netdevice, apply ioctl on it, then apply it on
 829          * the tap netdevice.
 830          */
 831 apply:
 832         if (remote)
 833                 strlcpy(ifr->ifr_name, pmd->remote_iface, IFNAMSIZ);
 834         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 835                 strlcpy(ifr->ifr_name, pmd->name, IFNAMSIZ);
 836         switch (request) {
 837         case SIOCSIFFLAGS:
 838                 /* fetch current flags to leave other flags untouched */
 839                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 840                         goto error;
 841                 if (set)
 842                         ifr->ifr_flags |= req_flags;
 843                 else
 844                         ifr->ifr_flags &= ~req_flags;
 845                 break;
 846         case SIOCGIFFLAGS:
 847         case SIOCGIFHWADDR:
 848         case SIOCSIFHWADDR:
 849         case SIOCSIFMTU:
 850                 break;
 851         default:
 852                 TAP_LOG(WARNING, "%s: ioctl() called with wrong arg",
 853                         pmd->name);
 854                 return -EINVAL;
 855         }
 856         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 857                 goto error;
 858         if (remote-- && mode == LOCAL_AND_REMOTE)
 859                 goto apply;
 860         return 0;
 861
 862 error:
 863         TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
 864                 tap_ioctl_req2str(request), strerror(errno), errno);
 865         return -errno;
 866 }
 867
 868 static int
 869 tap_link_set_down(struct rte_eth_dev *dev)
 870 {
 871         struct pmd_internals *pmd = dev->data->dev_private;
 872         struct ifreq ifr = { .ifr_flags = IFF_UP };
 873
 874         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 875         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 876 }
 877
 878 static int
 879 tap_link_set_up(struct rte_eth_dev *dev)
 880 {
 881         struct pmd_internals *pmd = dev->data->dev_private;
 882         struct ifreq ifr = { .ifr_flags = IFF_UP };
 883
 884         dev->data->dev_link.link_status = ETH_LINK_UP;
 885         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 886 }
 887
 888 static int
 889 tap_dev_start(struct rte_eth_dev *dev)
 890 {
 891         int err, i;
 892
 893         err = tap_intr_handle_set(dev, 1);
 894         if (err)
 895                 return err;
 896
 897         err = tap_link_set_up(dev);
 898         if (err)
 899                 return err;
 900
 901         for (i = 0; i < dev->data->nb_tx_queues; i++)
 902                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 903         for (i = 0; i < dev->data->nb_rx_queues; i++)
 904                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 905
 906         return err;
 907 }
 908
 909 /* This function gets called when the current port gets stopped.
 910  */
 911 static int
 912 tap_dev_stop(struct rte_eth_dev *dev)
 913 {
 914         int i;
 915
 916         for (i = 0; i < dev->data->nb_tx_queues; i++)
 917                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 918         for (i = 0; i < dev->data->nb_rx_queues; i++)
 919                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 920
 921         tap_intr_handle_set(dev, 0);
 922         tap_link_set_down(dev);
 923
 924         return 0;
 925 }
 926
 927 static int
 928 tap_dev_configure(struct rte_eth_dev *dev)
 929 {
 930         struct pmd_internals *pmd = dev->data->dev_private;
 931
 932         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 933                 TAP_LOG(ERR,
 934                         "%s: number of rx queues %d exceeds max num of queues %d",
 935                         dev->device->name,
 936                         dev->data->nb_rx_queues,
 937                         RTE_PMD_TAP_MAX_QUEUES);
 938                 return -1;
 939         }
 940         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 941                 TAP_LOG(ERR,
 942                         "%s: number of tx queues %d exceeds max num of queues %d",
 943                         dev->device->name,
 944                         dev->data->nb_tx_queues,
 945                         RTE_PMD_TAP_MAX_QUEUES);
 946                 return -1;
 947         }
 948
 949         TAP_LOG(INFO, "%s: %s: TX configured queues number: %u",
 950                 dev->device->name, pmd->name, dev->data->nb_tx_queues);
 951
 952         TAP_LOG(INFO, "%s: %s: RX configured queues number: %u",
 953                 dev->device->name, pmd->name, dev->data->nb_rx_queues);
 954
 955         return 0;
 956 }
 957
 958 static uint32_t
 959 tap_dev_speed_capa(void)
 960 {
 961         uint32_t speed = pmd_link.link_speed;
 962         uint32_t capa = 0;
 963
 964         if (speed >= ETH_SPEED_NUM_10M)
 965                 capa |= ETH_LINK_SPEED_10M;
 966         if (speed >= ETH_SPEED_NUM_100M)
 967                 capa |= ETH_LINK_SPEED_100M;
 968         if (speed >= ETH_SPEED_NUM_1G)
 969                 capa |= ETH_LINK_SPEED_1G;
 970         if (speed >= ETH_SPEED_NUM_5G)
 971                 capa |= ETH_LINK_SPEED_2_5G;
 972         if (speed >= ETH_SPEED_NUM_5G)
 973                 capa |= ETH_LINK_SPEED_5G;
 974         if (speed >= ETH_SPEED_NUM_10G)
 975                 capa |= ETH_LINK_SPEED_10G;
 976         if (speed >= ETH_SPEED_NUM_20G)
 977                 capa |= ETH_LINK_SPEED_20G;
 978         if (speed >= ETH_SPEED_NUM_25G)
 979                 capa |= ETH_LINK_SPEED_25G;
 980         if (speed >= ETH_SPEED_NUM_40G)
 981                 capa |= ETH_LINK_SPEED_40G;
 982         if (speed >= ETH_SPEED_NUM_50G)
 983                 capa |= ETH_LINK_SPEED_50G;
 984         if (speed >= ETH_SPEED_NUM_56G)
 985                 capa |= ETH_LINK_SPEED_56G;
 986         if (speed >= ETH_SPEED_NUM_100G)
 987                 capa |= ETH_LINK_SPEED_100G;
 988
 989         return capa;
 990 }
 991
 992 static int
 993 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 994 {
 995         struct pmd_internals *internals = dev->data->dev_private;
 996
 997         dev_info->if_index = internals->if_index;
 998         dev_info->max_mac_addrs = 1;
 999         dev_info->max_rx_pktlen = (uint32_t)RTE_ETHER_MAX_VLAN_FRAME_LEN;
1000         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
1001         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
1002         dev_info->min_rx_bufsize = 0;
1003         dev_info->speed_capa = tap_dev_speed_capa();
1004         dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
1005         dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
1006                                     dev_info->rx_queue_offload_capa;
1007         dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
1008         dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
1009                                     dev_info->tx_queue_offload_capa;
1010         dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
1011         /*
1012          * limitation: TAP supports all of IP, UDP and TCP hash
1013          * functions together and not in partial combinations
1014          */
1015         dev_info->flow_type_rss_offloads = ~TAP_RSS_HF_MASK;
1016
1017         return 0;
1018 }
1019
1020 static int
1021 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
1022 {
1023         unsigned int i, imax;
1024         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
1025         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
1026         unsigned long rx_nombuf = 0, ierrors = 0;
1027         const struct pmd_internals *pmd = dev->data->dev_private;
1028
1029         /* rx queue statistics */
1030         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
1031                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
1032         for (i = 0; i < imax; i++) {
1033                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
1034                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
1035                 rx_total += tap_stats->q_ipackets[i];
1036                 rx_bytes_total += tap_stats->q_ibytes[i];
1037                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
1038                 ierrors += pmd->rxq[i].stats.ierrors;
1039         }
1040
1041         /* tx queue statistics */
1042         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
1043                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
1044
1045         for (i = 0; i < imax; i++) {
1046                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
1047                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
1048                 tx_total += tap_stats->q_opackets[i];
1049                 tx_err_total += pmd->txq[i].stats.errs;
1050                 tx_bytes_total += tap_stats->q_obytes[i];
1051         }
1052
1053         tap_stats->ipackets = rx_total;
1054         tap_stats->ibytes = rx_bytes_total;
1055         tap_stats->ierrors = ierrors;
1056         tap_stats->rx_nombuf = rx_nombuf;
1057         tap_stats->opackets = tx_total;
1058         tap_stats->oerrors = tx_err_total;
1059         tap_stats->obytes = tx_bytes_total;
1060         return 0;
1061 }
1062
1063 static int
1064 tap_stats_reset(struct rte_eth_dev *dev)
1065 {
1066         int i;
1067         struct pmd_internals *pmd = dev->data->dev_private;
1068
1069         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1070                 pmd->rxq[i].stats.ipackets = 0;
1071                 pmd->rxq[i].stats.ibytes = 0;
1072                 pmd->rxq[i].stats.ierrors = 0;
1073                 pmd->rxq[i].stats.rx_nombuf = 0;
1074
1075                 pmd->txq[i].stats.opackets = 0;
1076                 pmd->txq[i].stats.errs = 0;
1077                 pmd->txq[i].stats.obytes = 0;
1078         }
1079
1080         return 0;
1081 }
1082
1083 static int
1084 tap_dev_close(struct rte_eth_dev *dev)
1085 {
1086         int i;
1087         struct pmd_internals *internals = dev->data->dev_private;
1088         struct pmd_process_private *process_private = dev->process_private;
1089         struct rx_queue *rxq;
1090
1091         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1092                 rte_free(dev->process_private);
1093                 return 0;
1094         }
1095
1096         tap_link_set_down(dev);
1097         if (internals->nlsk_fd != -1) {
1098                 tap_flow_flush(dev, NULL);
1099                 tap_flow_implicit_flush(internals, NULL);
1100                 tap_nl_final(internals->nlsk_fd);
1101                 internals->nlsk_fd = -1;
1102         }
1103
1104         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1105                 if (process_private->rxq_fds[i] != -1) {
1106                         rxq = &internals->rxq[i];
1107                         close(process_private->rxq_fds[i]);
1108                         process_private->rxq_fds[i] = -1;
1109                         tap_rxq_pool_free(rxq->pool);
1110                         rte_free(rxq->iovecs);
1111                         rxq->pool = NULL;
1112                         rxq->iovecs = NULL;
1113                 }
1114                 if (process_private->txq_fds[i] != -1) {
1115                         close(process_private->txq_fds[i]);
1116                         process_private->txq_fds[i] = -1;
1117                 }
1118         }
1119
1120         if (internals->remote_if_index) {
1121                 /* Restore initial remote state */
1122                 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
1123                                 &internals->remote_initial_flags);
1124         }
1125
1126         rte_mempool_free(internals->gso_ctx_mp);
1127         internals->gso_ctx_mp = NULL;
1128
1129         if (internals->ka_fd != -1) {
1130                 close(internals->ka_fd);
1131                 internals->ka_fd = -1;
1132         }
1133
1134         /* mac_addrs must not be freed alone because part of dev_private */
1135         dev->data->mac_addrs = NULL;
1136
1137         internals = dev->data->dev_private;
1138         TAP_LOG(DEBUG, "Closing %s Ethernet device on numa %u",
1139                 tuntap_types[internals->type], rte_socket_id());
1140
1141         if (internals->ioctl_sock != -1) {
1142                 close(internals->ioctl_sock);
1143                 internals->ioctl_sock = -1;
1144         }
1145         rte_free(dev->process_private);
1146         if (tap_devices_count == 1)
1147                 rte_mp_action_unregister(TAP_MP_KEY);
1148         tap_devices_count--;
1149         /*
1150          * Since TUN device has no more opened file descriptors
1151          * it will be removed from kernel
1152          */
1153
1154         return 0;
1155 }
1156
1157 static void
1158 tap_rx_queue_release(void *queue)
1159 {
1160         struct rx_queue *rxq = queue;
1161         struct pmd_process_private *process_private;
1162
1163         if (!rxq)
1164                 return;
1165         process_private = rte_eth_devices[rxq->in_port].process_private;
1166         if (process_private->rxq_fds[rxq->queue_id] != -1) {
1167                 close(process_private->rxq_fds[rxq->queue_id]);
1168                 process_private->rxq_fds[rxq->queue_id] = -1;
1169                 tap_rxq_pool_free(rxq->pool);
1170                 rte_free(rxq->iovecs);
1171                 rxq->pool = NULL;
1172                 rxq->iovecs = NULL;
1173         }
1174 }
1175
1176 static void
1177 tap_tx_queue_release(void *queue)
1178 {
1179         struct tx_queue *txq = queue;
1180         struct pmd_process_private *process_private;
1181
1182         if (!txq)
1183                 return;
1184         process_private = rte_eth_devices[txq->out_port].process_private;
1185
1186         if (process_private->txq_fds[txq->queue_id] != -1) {
1187                 close(process_private->txq_fds[txq->queue_id]);
1188                 process_private->txq_fds[txq->queue_id] = -1;
1189         }
1190 }
1191
1192 static int
1193 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1194 {
1195         struct rte_eth_link *dev_link = &dev->data->dev_link;
1196         struct pmd_internals *pmd = dev->data->dev_private;
1197         struct ifreq ifr = { .ifr_flags = 0 };
1198
1199         if (pmd->remote_if_index) {
1200                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
1201                 if (!(ifr.ifr_flags & IFF_UP) ||
1202                     !(ifr.ifr_flags & IFF_RUNNING)) {
1203                         dev_link->link_status = ETH_LINK_DOWN;
1204                         return 0;
1205                 }
1206         }
1207         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
1208         dev_link->link_status =
1209                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
1210                  ETH_LINK_UP :
1211                  ETH_LINK_DOWN);
1212         return 0;
1213 }
1214
1215 static int
1216 tap_promisc_enable(struct rte_eth_dev *dev)
1217 {
1218         struct pmd_internals *pmd = dev->data->dev_private;
1219         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1220         int ret;
1221
1222         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1223         if (ret != 0)
1224                 return ret;
1225
1226         if (pmd->remote_if_index && !pmd->flow_isolate) {
1227                 dev->data->promiscuous = 1;
1228                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
1229                 if (ret != 0) {
1230                         /* Rollback promisc flag */
1231                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1232                         /*
1233                          * rte_eth_dev_promiscuous_enable() rollback
1234                          * dev->data->promiscuous in the case of failure.
1235                          */
1236                         return ret;
1237                 }
1238         }
1239
1240         return 0;
1241 }
1242
1243 static int
1244 tap_promisc_disable(struct rte_eth_dev *dev)
1245 {
1246         struct pmd_internals *pmd = dev->data->dev_private;
1247         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1248         int ret;
1249
1250         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1251         if (ret != 0)
1252                 return ret;
1253
1254         if (pmd->remote_if_index && !pmd->flow_isolate) {
1255                 dev->data->promiscuous = 0;
1256                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
1257                 if (ret != 0) {
1258                         /* Rollback promisc flag */
1259                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1260                         /*
1261                          * rte_eth_dev_promiscuous_disable() rollback
1262                          * dev->data->promiscuous in the case of failure.
1263                          */
1264                         return ret;
1265                 }
1266         }
1267
1268         return 0;
1269 }
1270
1271 static int
1272 tap_allmulti_enable(struct rte_eth_dev *dev)
1273 {
1274         struct pmd_internals *pmd = dev->data->dev_private;
1275         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1276         int ret;
1277
1278         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1279         if (ret != 0)
1280                 return ret;
1281
1282         if (pmd->remote_if_index && !pmd->flow_isolate) {
1283                 dev->data->all_multicast = 1;
1284                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
1285                 if (ret != 0) {
1286                         /* Rollback allmulti flag */
1287                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1288                         /*
1289                          * rte_eth_dev_allmulticast_enable() rollback
1290                          * dev->data->all_multicast in the case of failure.
1291                          */
1292                         return ret;
1293                 }
1294         }
1295
1296         return 0;
1297 }
1298
1299 static int
1300 tap_allmulti_disable(struct rte_eth_dev *dev)
1301 {
1302         struct pmd_internals *pmd = dev->data->dev_private;
1303         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1304         int ret;
1305
1306         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1307         if (ret != 0)
1308                 return ret;
1309
1310         if (pmd->remote_if_index && !pmd->flow_isolate) {
1311                 dev->data->all_multicast = 0;
1312                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
1313                 if (ret != 0) {
1314                         /* Rollback allmulti flag */
1315                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1316                         /*
1317                          * rte_eth_dev_allmulticast_disable() rollback
1318                          * dev->data->all_multicast in the case of failure.
1319                          */
1320                         return ret;
1321                 }
1322         }
1323
1324         return 0;
1325 }
1326
1327 static int
1328 tap_mac_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1329 {
1330         struct pmd_internals *pmd = dev->data->dev_private;
1331         enum ioctl_mode mode = LOCAL_ONLY;
1332         struct ifreq ifr;
1333         int ret;
1334
1335         if (pmd->type == ETH_TUNTAP_TYPE_TUN) {
1336                 TAP_LOG(ERR, "%s: can't MAC address for TUN",
1337                         dev->device->name);
1338                 return -ENOTSUP;
1339         }
1340
1341         if (rte_is_zero_ether_addr(mac_addr)) {
1342                 TAP_LOG(ERR, "%s: can't set an empty MAC address",
1343                         dev->device->name);
1344                 return -EINVAL;
1345         }
1346         /* Check the actual current MAC address on the tap netdevice */
1347         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
1348         if (ret < 0)
1349                 return ret;
1350         if (rte_is_same_ether_addr(
1351                         (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1352                         mac_addr))
1353                 return 0;
1354         /* Check the current MAC address on the remote */
1355         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
1356         if (ret < 0)
1357                 return ret;
1358         if (!rte_is_same_ether_addr(
1359                         (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1360                         mac_addr))
1361                 mode = LOCAL_AND_REMOTE;
1362         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1363         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, RTE_ETHER_ADDR_LEN);
1364         ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
1365         if (ret < 0)
1366                 return ret;
1367         rte_memcpy(&pmd->eth_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1368         if (pmd->remote_if_index && !pmd->flow_isolate) {
1369                 /* Replace MAC redirection rule after a MAC change */
1370                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
1371                 if (ret < 0) {
1372                         TAP_LOG(ERR,
1373                                 "%s: Couldn't delete MAC redirection rule",
1374                                 dev->device->name);
1375                         return ret;
1376                 }
1377                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1378                 if (ret < 0) {
1379                         TAP_LOG(ERR,
1380                                 "%s: Couldn't add MAC redirection rule",
1381                                 dev->device->name);
1382                         return ret;
1383                 }
1384         }
1385
1386         return 0;
1387 }
1388
1389 static int
1390 tap_gso_ctx_setup(struct rte_gso_ctx *gso_ctx, struct rte_eth_dev *dev)
1391 {
1392         uint32_t gso_types;
1393         char pool_name[64];
1394         struct pmd_internals *pmd = dev->data->dev_private;
1395         int ret;
1396
1397         /* initialize GSO context */
1398         gso_types = DEV_TX_OFFLOAD_TCP_TSO;
1399         if (!pmd->gso_ctx_mp) {
1400                 /*
1401                  * Create private mbuf pool with TAP_GSO_MBUF_SEG_SIZE
1402                  * bytes size per mbuf use this pool for both direct and
1403                  * indirect mbufs
1404                  */
1405                 ret = snprintf(pool_name, sizeof(pool_name), "mp_%s",
1406                                 dev->device->name);
1407                 if (ret < 0 || ret >= (int)sizeof(pool_name)) {
1408                         TAP_LOG(ERR,
1409                                 "%s: failed to create mbuf pool name for device %s,"
1410                                 "device name too long or output error, ret: %d\n",
1411                                 pmd->name, dev->device->name, ret);
1412                         return -ENAMETOOLONG;
1413                 }
1414                 pmd->gso_ctx_mp = rte_pktmbuf_pool_create(pool_name,
1415                         TAP_GSO_MBUFS_NUM, TAP_GSO_MBUF_CACHE_SIZE, 0,
1416                         RTE_PKTMBUF_HEADROOM + TAP_GSO_MBUF_SEG_SIZE,
1417                         SOCKET_ID_ANY);
1418                 if (!pmd->gso_ctx_mp) {
1419                         TAP_LOG(ERR,
1420                                 "%s: failed to create mbuf pool for device %s\n",
1421                                 pmd->name, dev->device->name);
1422                         return -1;
1423                 }
1424         }
1425
1426         gso_ctx->direct_pool = pmd->gso_ctx_mp;
1427         gso_ctx->indirect_pool = pmd->gso_ctx_mp;
1428         gso_ctx->gso_types = gso_types;
1429         gso_ctx->gso_size = 0; /* gso_size is set in tx_burst() per packet */
1430         gso_ctx->flag = 0;
1431
1432         return 0;
1433 }
1434
1435 static int
1436 tap_setup_queue(struct rte_eth_dev *dev,
1437                 struct pmd_internals *internals,
1438                 uint16_t qid,
1439                 int is_rx)
1440 {
1441         int ret;
1442         int *fd;
1443         int *other_fd;
1444         const char *dir;
1445         struct pmd_internals *pmd = dev->data->dev_private;
1446         struct pmd_process_private *process_private = dev->process_private;
1447         struct rx_queue *rx = &internals->rxq[qid];
1448         struct tx_queue *tx = &internals->txq[qid];
1449         struct rte_gso_ctx *gso_ctx;
1450
1451         if (is_rx) {
1452                 fd = &process_private->rxq_fds[qid];
1453                 other_fd = &process_private->txq_fds[qid];
1454                 dir = "rx";
1455                 gso_ctx = NULL;
1456         } else {
1457                 fd = &process_private->txq_fds[qid];
1458                 other_fd = &process_private->rxq_fds[qid];
1459                 dir = "tx";
1460                 gso_ctx = &tx->gso_ctx;
1461         }
1462         if (*fd != -1) {
1463                 /* fd for this queue already exists */
1464                 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1465                         pmd->name, *fd, dir, qid);
1466                 gso_ctx = NULL;
1467         } else if (*other_fd != -1) {
1468                 /* Only other_fd exists. dup it */
1469                 *fd = dup(*other_fd);
1470                 if (*fd < 0) {
1471                         *fd = -1;
1472                         TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1473                         return -1;
1474                 }
1475                 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1476                         pmd->name, *other_fd, dir, qid, *fd);
1477         } else {
1478                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1479                 *fd = tun_alloc(pmd, 0);
1480                 if (*fd < 0) {
1481                         *fd = -1; /* restore original value */
1482                         TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1483                         return -1;
1484                 }
1485                 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1486                         pmd->name, dir, qid, *fd);
1487         }
1488
1489         tx->mtu = &dev->data->mtu;
1490         rx->rxmode = &dev->data->dev_conf.rxmode;
1491         if (gso_ctx) {
1492                 ret = tap_gso_ctx_setup(gso_ctx, dev);
1493                 if (ret)
1494                         return -1;
1495         }
1496
1497         tx->type = pmd->type;
1498
1499         return *fd;
1500 }
1501
1502 static int
1503 tap_rx_queue_setup(struct rte_eth_dev *dev,
1504                    uint16_t rx_queue_id,
1505                    uint16_t nb_rx_desc,
1506                    unsigned int socket_id,
1507                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1508                    struct rte_mempool *mp)
1509 {
1510         struct pmd_internals *internals = dev->data->dev_private;
1511         struct pmd_process_private *process_private = dev->process_private;
1512         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1513         struct rte_mbuf **tmp = &rxq->pool;
1514         long iov_max = sysconf(_SC_IOV_MAX);
1515
1516         if (iov_max <= 0) {
1517                 TAP_LOG(WARNING,
1518                         "_SC_IOV_MAX is not defined. Using %d as default",
1519                         TAP_IOV_DEFAULT_MAX);
1520                 iov_max = TAP_IOV_DEFAULT_MAX;
1521         }
1522         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1523         struct iovec (*iovecs)[nb_desc + 1];
1524         int data_off = RTE_PKTMBUF_HEADROOM;
1525         int ret = 0;
1526         int fd;
1527         int i;
1528
1529         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1530                 TAP_LOG(WARNING,
1531                         "nb_rx_queues %d too small or mempool NULL",
1532                         dev->data->nb_rx_queues);
1533                 return -1;
1534         }
1535
1536         rxq->mp = mp;
1537         rxq->trigger_seen = 1; /* force initial burst */
1538         rxq->in_port = dev->data->port_id;
1539         rxq->queue_id = rx_queue_id;
1540         rxq->nb_rx_desc = nb_desc;
1541         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1542                                     socket_id);
1543         if (!iovecs) {
1544                 TAP_LOG(WARNING,
1545                         "%s: Couldn't allocate %d RX descriptors",
1546                         dev->device->name, nb_desc);
1547                 return -ENOMEM;
1548         }
1549         rxq->iovecs = iovecs;
1550
1551         dev->data->rx_queues[rx_queue_id] = rxq;
1552         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1553         if (fd == -1) {
1554                 ret = fd;
1555                 goto error;
1556         }
1557
1558         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1559         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1560
1561         for (i = 1; i <= nb_desc; i++) {
1562                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1563                 if (!*tmp) {
1564                         TAP_LOG(WARNING,
1565                                 "%s: couldn't allocate memory for queue %d",
1566                                 dev->device->name, rx_queue_id);
1567                         ret = -ENOMEM;
1568                         goto error;
1569                 }
1570                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1571                 (*rxq->iovecs)[i].iov_base =
1572                         (char *)(*tmp)->buf_addr + data_off;
1573                 data_off = 0;
1574                 tmp = &(*tmp)->next;
1575         }
1576
1577         TAP_LOG(DEBUG, "  RX TUNTAP device name %s, qid %d on fd %d",
1578                 internals->name, rx_queue_id,
1579                 process_private->rxq_fds[rx_queue_id]);
1580
1581         return 0;
1582
1583 error:
1584         tap_rxq_pool_free(rxq->pool);
1585         rxq->pool = NULL;
1586         rte_free(rxq->iovecs);
1587         rxq->iovecs = NULL;
1588         return ret;
1589 }
1590
1591 static int
1592 tap_tx_queue_setup(struct rte_eth_dev *dev,
1593                    uint16_t tx_queue_id,
1594                    uint16_t nb_tx_desc __rte_unused,
1595                    unsigned int socket_id __rte_unused,
1596                    const struct rte_eth_txconf *tx_conf)
1597 {
1598         struct pmd_internals *internals = dev->data->dev_private;
1599         struct pmd_process_private *process_private = dev->process_private;
1600         struct tx_queue *txq;
1601         int ret;
1602         uint64_t offloads;
1603
1604         if (tx_queue_id >= dev->data->nb_tx_queues)
1605                 return -1;
1606         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1607         txq = dev->data->tx_queues[tx_queue_id];
1608         txq->out_port = dev->data->port_id;
1609         txq->queue_id = tx_queue_id;
1610
1611         offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1612         txq->csum = !!(offloads &
1613                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
1614                          DEV_TX_OFFLOAD_UDP_CKSUM |
1615                          DEV_TX_OFFLOAD_TCP_CKSUM));
1616
1617         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1618         if (ret == -1)
1619                 return -1;
1620         TAP_LOG(DEBUG,
1621                 "  TX TUNTAP device name %s, qid %d on fd %d csum %s",
1622                 internals->name, tx_queue_id,
1623                 process_private->txq_fds[tx_queue_id],
1624                 txq->csum ? "on" : "off");
1625
1626         return 0;
1627 }
1628
1629 static int
1630 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1631 {
1632         struct pmd_internals *pmd = dev->data->dev_private;
1633         struct ifreq ifr = { .ifr_mtu = mtu };
1634         int err = 0;
1635
1636         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1637         if (!err)
1638                 dev->data->mtu = mtu;
1639
1640         return err;
1641 }
1642
1643 static int
1644 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1645                      struct rte_ether_addr *mc_addr_set __rte_unused,
1646                      uint32_t nb_mc_addr __rte_unused)
1647 {
1648         /*
1649          * Nothing to do actually: the tap has no filtering whatsoever, every
1650          * packet is received.
1651          */
1652         return 0;
1653 }
1654
1655 static int
1656 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1657 {
1658         struct rte_eth_dev *dev = arg;
1659         struct pmd_internals *pmd = dev->data->dev_private;
1660         struct ifinfomsg *info = NLMSG_DATA(nh);
1661
1662         if (nh->nlmsg_type != RTM_NEWLINK ||
1663             (info->ifi_index != pmd->if_index &&
1664              info->ifi_index != pmd->remote_if_index))
1665                 return 0;
1666         return tap_link_update(dev, 0);
1667 }
1668
1669 static void
1670 tap_dev_intr_handler(void *cb_arg)
1671 {
1672         struct rte_eth_dev *dev = cb_arg;
1673         struct pmd_internals *pmd = dev->data->dev_private;
1674
1675         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1676 }
1677
1678 static int
1679 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1680 {
1681         struct pmd_internals *pmd = dev->data->dev_private;
1682         int ret;
1683
1684         /* In any case, disable interrupt if the conf is no longer there. */
1685         if (!dev->data->dev_conf.intr_conf.lsc) {
1686                 if (pmd->intr_handle.fd != -1) {
1687                         goto clean;
1688                 }
1689                 return 0;
1690         }
1691         if (set) {
1692                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1693                 if (unlikely(pmd->intr_handle.fd == -1))
1694                         return -EBADF;
1695                 return rte_intr_callback_register(
1696                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1697         }
1698
1699 clean:
1700         do {
1701                 ret = rte_intr_callback_unregister(&pmd->intr_handle,
1702                         tap_dev_intr_handler, dev);
1703                 if (ret >= 0) {
1704                         break;
1705                 } else if (ret == -EAGAIN) {
1706                         rte_delay_ms(100);
1707                 } else {
1708                         TAP_LOG(ERR, "intr callback unregister failed: %d",
1709                                      ret);
1710                         break;
1711                 }
1712         } while (true);
1713
1714         tap_nl_final(pmd->intr_handle.fd);
1715         pmd->intr_handle.fd = -1;
1716
1717         return 0;
1718 }
1719
1720 static int
1721 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1722 {
1723         int err;
1724
1725         err = tap_lsc_intr_handle_set(dev, set);
1726         if (err < 0) {
1727                 if (!set)
1728                         tap_rx_intr_vec_set(dev, 0);
1729                 return err;
1730         }
1731         err = tap_rx_intr_vec_set(dev, set);
1732         if (err && set)
1733                 tap_lsc_intr_handle_set(dev, 0);
1734         return err;
1735 }
1736
1737 static const uint32_t*
1738 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1739 {
1740         static const uint32_t ptypes[] = {
1741                 RTE_PTYPE_INNER_L2_ETHER,
1742                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1743                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1744                 RTE_PTYPE_INNER_L3_IPV4,
1745                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1746                 RTE_PTYPE_INNER_L3_IPV6,
1747                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1748                 RTE_PTYPE_INNER_L4_FRAG,
1749                 RTE_PTYPE_INNER_L4_UDP,
1750                 RTE_PTYPE_INNER_L4_TCP,
1751                 RTE_PTYPE_INNER_L4_SCTP,
1752                 RTE_PTYPE_L2_ETHER,
1753                 RTE_PTYPE_L2_ETHER_VLAN,
1754                 RTE_PTYPE_L2_ETHER_QINQ,
1755                 RTE_PTYPE_L3_IPV4,
1756                 RTE_PTYPE_L3_IPV4_EXT,
1757                 RTE_PTYPE_L3_IPV6_EXT,
1758                 RTE_PTYPE_L3_IPV6,
1759                 RTE_PTYPE_L4_FRAG,
1760                 RTE_PTYPE_L4_UDP,
1761                 RTE_PTYPE_L4_TCP,
1762                 RTE_PTYPE_L4_SCTP,
1763         };
1764
1765         return ptypes;
1766 }
1767
1768 static int
1769 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1770                   struct rte_eth_fc_conf *fc_conf)
1771 {
1772         fc_conf->mode = RTE_FC_NONE;
1773         return 0;
1774 }
1775
1776 static int
1777 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1778                   struct rte_eth_fc_conf *fc_conf)
1779 {
1780         if (fc_conf->mode != RTE_FC_NONE)
1781                 return -ENOTSUP;
1782         return 0;
1783 }
1784
1785 /**
1786  * DPDK callback to update the RSS hash configuration.
1787  *
1788  * @param dev
1789  *   Pointer to Ethernet device structure.
1790  * @param[in] rss_conf
1791  *   RSS configuration data.
1792  *
1793  * @return
1794  *   0 on success, a negative errno value otherwise and rte_errno is set.
1795  */
1796 static int
1797 tap_rss_hash_update(struct rte_eth_dev *dev,
1798                 struct rte_eth_rss_conf *rss_conf)
1799 {
1800         if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1801                 rte_errno = EINVAL;
1802                 return -rte_errno;
1803         }
1804         if (rss_conf->rss_key && rss_conf->rss_key_len) {
1805                 /*
1806                  * Currently TAP RSS key is hard coded
1807                  * and cannot be updated
1808                  */
1809                 TAP_LOG(ERR,
1810                         "port %u RSS key cannot be updated",
1811                         dev->data->port_id);
1812                 rte_errno = EINVAL;
1813                 return -rte_errno;
1814         }
1815         return 0;
1816 }
1817
1818 static int
1819 tap_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1820 {
1821         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1822
1823         return 0;
1824 }
1825
1826 static int
1827 tap_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1828 {
1829         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1830
1831         return 0;
1832 }
1833
1834 static int
1835 tap_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1836 {
1837         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1838
1839         return 0;
1840 }
1841
1842 static int
1843 tap_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1844 {
1845         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1846
1847         return 0;
1848 }
1849 static const struct eth_dev_ops ops = {
1850         .dev_start              = tap_dev_start,
1851         .dev_stop               = tap_dev_stop,
1852         .dev_close              = tap_dev_close,
1853         .dev_configure          = tap_dev_configure,
1854         .dev_infos_get          = tap_dev_info,
1855         .rx_queue_setup         = tap_rx_queue_setup,
1856         .tx_queue_setup         = tap_tx_queue_setup,
1857         .rx_queue_start         = tap_rx_queue_start,
1858         .tx_queue_start         = tap_tx_queue_start,
1859         .rx_queue_stop          = tap_rx_queue_stop,
1860         .tx_queue_stop          = tap_tx_queue_stop,
1861         .rx_queue_release       = tap_rx_queue_release,
1862         .tx_queue_release       = tap_tx_queue_release,
1863         .flow_ctrl_get          = tap_flow_ctrl_get,
1864         .flow_ctrl_set          = tap_flow_ctrl_set,
1865         .link_update            = tap_link_update,
1866         .dev_set_link_up        = tap_link_set_up,
1867         .dev_set_link_down      = tap_link_set_down,
1868         .promiscuous_enable     = tap_promisc_enable,
1869         .promiscuous_disable    = tap_promisc_disable,
1870         .allmulticast_enable    = tap_allmulti_enable,
1871         .allmulticast_disable   = tap_allmulti_disable,
1872         .mac_addr_set           = tap_mac_set,
1873         .mtu_set                = tap_mtu_set,
1874         .set_mc_addr_list       = tap_set_mc_addr_list,
1875         .stats_get              = tap_stats_get,
1876         .stats_reset            = tap_stats_reset,
1877         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1878         .rss_hash_update        = tap_rss_hash_update,
1879         .filter_ctrl            = tap_dev_filter_ctrl,
1880 };
1881
1882 static int
1883 eth_dev_tap_create(struct rte_vdev_device *vdev, const char *tap_name,
1884                    char *remote_iface, struct rte_ether_addr *mac_addr,
1885                    enum rte_tuntap_type type)
1886 {
1887         int numa_node = rte_socket_id();
1888         struct rte_eth_dev *dev;
1889         struct pmd_internals *pmd;
1890         struct pmd_process_private *process_private;
1891         const char *tuntap_name = tuntap_types[type];
1892         struct rte_eth_dev_data *data;
1893         struct ifreq ifr;
1894         int i;
1895
1896         TAP_LOG(DEBUG, "%s device on numa %u", tuntap_name, rte_socket_id());
1897
1898         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1899         if (!dev) {
1900                 TAP_LOG(ERR, "%s Unable to allocate device struct",
1901                                 tuntap_name);
1902                 goto error_exit_nodev;
1903         }
1904
1905         process_private = (struct pmd_process_private *)
1906                 rte_zmalloc_socket(tap_name, sizeof(struct pmd_process_private),
1907                         RTE_CACHE_LINE_SIZE, dev->device->numa_node);
1908
1909         if (process_private == NULL) {
1910                 TAP_LOG(ERR, "Failed to alloc memory for process private");
1911                 return -1;
1912         }
1913         pmd = dev->data->dev_private;
1914         dev->process_private = process_private;
1915         pmd->dev = dev;
1916         strlcpy(pmd->name, tap_name, sizeof(pmd->name));
1917         pmd->type = type;
1918         pmd->ka_fd = -1;
1919         pmd->nlsk_fd = -1;
1920         pmd->gso_ctx_mp = NULL;
1921
1922         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1923         if (pmd->ioctl_sock == -1) {
1924                 TAP_LOG(ERR,
1925                         "%s Unable to get a socket for management: %s",
1926                         tuntap_name, strerror(errno));
1927                 goto error_exit;
1928         }
1929
1930         /* Setup some default values */
1931         data = dev->data;
1932         data->dev_private = pmd;
1933         data->dev_flags = RTE_ETH_DEV_INTR_LSC |
1934                                 RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1935         data->numa_node = numa_node;
1936
1937         data->dev_link = pmd_link;
1938         data->mac_addrs = &pmd->eth_addr;
1939         /* Set the number of RX and TX queues */
1940         data->nb_rx_queues = 0;
1941         data->nb_tx_queues = 0;
1942
1943         dev->dev_ops = &ops;
1944         dev->rx_pkt_burst = pmd_rx_burst;
1945         dev->tx_pkt_burst = pmd_tx_burst;
1946
1947         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1948         pmd->intr_handle.fd = -1;
1949         dev->intr_handle = &pmd->intr_handle;
1950
1951         /* Presetup the fds to -1 as being not valid */
1952         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1953                 process_private->rxq_fds[i] = -1;
1954                 process_private->txq_fds[i] = -1;
1955         }
1956
1957         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1958                 if (rte_is_zero_ether_addr(mac_addr))
1959                         rte_eth_random_addr((uint8_t *)&pmd->eth_addr);
1960                 else
1961                         rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1962         }
1963
1964         /*
1965          * Allocate a TUN device keep-alive file descriptor that will only be
1966          * closed when the TUN device itself is closed or removed.
1967          * This keep-alive file descriptor will guarantee that the TUN device
1968          * exists even when all of its queues are closed
1969          */
1970         pmd->ka_fd = tun_alloc(pmd, 1);
1971         if (pmd->ka_fd == -1) {
1972                 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
1973                 goto error_exit;
1974         }
1975         TAP_LOG(DEBUG, "allocated %s", pmd->name);
1976
1977         ifr.ifr_mtu = dev->data->mtu;
1978         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1979                 goto error_exit;
1980
1981         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1982                 memset(&ifr, 0, sizeof(struct ifreq));
1983                 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1984                 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1985                                 RTE_ETHER_ADDR_LEN);
1986                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1987                         goto error_exit;
1988         }
1989
1990         /*
1991          * Set up everything related to rte_flow:
1992          * - netlink socket
1993          * - tap / remote if_index
1994          * - mandatory QDISCs
1995          * - rte_flow actual/implicit lists
1996          * - implicit rules
1997          */
1998         pmd->nlsk_fd = tap_nl_init(0);
1999         if (pmd->nlsk_fd == -1) {
2000                 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
2001                         pmd->name);
2002                 goto disable_rte_flow;
2003         }
2004         pmd->if_index = if_nametoindex(pmd->name);
2005         if (!pmd->if_index) {
2006                 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
2007                 goto disable_rte_flow;
2008         }
2009         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
2010                 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
2011                         pmd->name);
2012                 goto disable_rte_flow;
2013         }
2014         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
2015                 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
2016                         pmd->name);
2017                 goto disable_rte_flow;
2018         }
2019         LIST_INIT(&pmd->flows);
2020
2021         if (strlen(remote_iface)) {
2022                 pmd->remote_if_index = if_nametoindex(remote_iface);
2023                 if (!pmd->remote_if_index) {
2024                         TAP_LOG(ERR, "%s: failed to get %s if_index.",
2025                                 pmd->name, remote_iface);
2026                         goto error_remote;
2027                 }
2028                 strlcpy(pmd->remote_iface, remote_iface, RTE_ETH_NAME_MAX_LEN);
2029
2030                 /* Save state of remote device */
2031                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
2032
2033                 /* Replicate remote MAC address */
2034                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
2035                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
2036                                 pmd->name, pmd->remote_iface);
2037                         goto error_remote;
2038                 }
2039                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
2040                            RTE_ETHER_ADDR_LEN);
2041                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
2042                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
2043                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
2044                                 pmd->name, remote_iface);
2045                         goto error_remote;
2046                 }
2047
2048                 /*
2049                  * Flush usually returns negative value because it tries to
2050                  * delete every QDISC (and on a running device, one QDISC at
2051                  * least is needed). Ignore negative return value.
2052                  */
2053                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
2054                 if (qdisc_create_ingress(pmd->nlsk_fd,
2055                                          pmd->remote_if_index) < 0) {
2056                         TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
2057                                 pmd->remote_iface);
2058                         goto error_remote;
2059                 }
2060                 LIST_INIT(&pmd->implicit_flows);
2061                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
2062                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
2063                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
2064                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
2065                         TAP_LOG(ERR,
2066                                 "%s: failed to create implicit rules.",
2067                                 pmd->name);
2068                         goto error_remote;
2069                 }
2070         }
2071
2072         rte_eth_dev_probing_finish(dev);
2073         return 0;
2074
2075 disable_rte_flow:
2076         TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
2077                 strerror(errno), errno);
2078         if (strlen(remote_iface)) {
2079                 TAP_LOG(ERR, "Remote feature requires flow support.");
2080                 goto error_exit;
2081         }
2082         rte_eth_dev_probing_finish(dev);
2083         return 0;
2084
2085 error_remote:
2086         TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
2087                 strerror(errno), errno);
2088         tap_flow_implicit_flush(pmd, NULL);
2089
2090 error_exit:
2091         if (pmd->nlsk_fd != -1)
2092                 close(pmd->nlsk_fd);
2093         if (pmd->ka_fd != -1)
2094                 close(pmd->ka_fd);
2095         if (pmd->ioctl_sock != -1)
2096                 close(pmd->ioctl_sock);
2097         /* mac_addrs must not be freed alone because part of dev_private */
2098         dev->data->mac_addrs = NULL;
2099         rte_eth_dev_release_port(dev);
2100
2101 error_exit_nodev:
2102         TAP_LOG(ERR, "%s Unable to initialize %s",
2103                 tuntap_name, rte_vdev_device_name(vdev));
2104
2105         return -EINVAL;
2106 }
2107
2108 /* make sure name is a possible Linux network device name */
2109 static bool
2110 is_valid_iface(const char *name)
2111 {
2112         if (*name == '\0')
2113                 return false;
2114
2115         if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
2116                 return false;
2117
2118         while (*name) {
2119                 if (*name == '/' || *name == ':' || isspace(*name))
2120                         return false;
2121                 name++;
2122         }
2123         return true;
2124 }
2125
2126 static int
2127 set_interface_name(const char *key __rte_unused,
2128                    const char *value,
2129                    void *extra_args)
2130 {
2131         char *name = (char *)extra_args;
2132
2133         if (value) {
2134                 if (!is_valid_iface(value)) {
2135                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2136                                 value);
2137                         return -1;
2138                 }
2139                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2140         } else {
2141                 /* use tap%d which causes kernel to choose next available */
2142                 strlcpy(name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2143         }
2144         return 0;
2145 }
2146
2147 static int
2148 set_remote_iface(const char *key __rte_unused,
2149                  const char *value,
2150                  void *extra_args)
2151 {
2152         char *name = (char *)extra_args;
2153
2154         if (value) {
2155                 if (!is_valid_iface(value)) {
2156                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2157                                 value);
2158                         return -1;
2159                 }
2160                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2161         }
2162
2163         return 0;
2164 }
2165
2166 static int parse_user_mac(struct rte_ether_addr *user_mac,
2167                 const char *value)
2168 {
2169         unsigned int index = 0;
2170         char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
2171
2172         if (user_mac == NULL || value == NULL)
2173                 return 0;
2174
2175         strlcpy(mac_temp, value, sizeof(mac_temp));
2176         mac_byte = strtok(mac_temp, ":");
2177
2178         while ((mac_byte != NULL) &&
2179                         (strlen(mac_byte) <= 2) &&
2180                         (strlen(mac_byte) == strspn(mac_byte,
2181                                         ETH_TAP_CMP_MAC_FMT))) {
2182                 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
2183                 mac_byte = strtok(NULL, ":");
2184         }
2185
2186         return index;
2187 }
2188
2189 static int
2190 set_mac_type(const char *key __rte_unused,
2191              const char *value,
2192              void *extra_args)
2193 {
2194         struct rte_ether_addr *user_mac = extra_args;
2195
2196         if (!value)
2197                 return 0;
2198
2199         if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
2200                 static int iface_idx;
2201
2202                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
2203                 memcpy((char *)user_mac->addr_bytes, "\0dtap",
2204                         RTE_ETHER_ADDR_LEN);
2205                 user_mac->addr_bytes[RTE_ETHER_ADDR_LEN - 1] =
2206                         iface_idx++ + '0';
2207                 goto success;
2208         }
2209
2210         if (parse_user_mac(user_mac, value) != 6)
2211                 goto error;
2212 success:
2213         TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
2214         return 0;
2215
2216 error:
2217         TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
2218                 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
2219         return -1;
2220 }
2221
2222 /*
2223  * Open a TUN interface device. TUN PMD
2224  * 1) sets tap_type as false
2225  * 2) intakes iface as argument.
2226  * 3) as interface is virtual set speed to 10G
2227  */
2228 static int
2229 rte_pmd_tun_probe(struct rte_vdev_device *dev)
2230 {
2231         const char *name, *params;
2232         int ret;
2233         struct rte_kvargs *kvlist = NULL;
2234         char tun_name[RTE_ETH_NAME_MAX_LEN];
2235         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2236         struct rte_eth_dev *eth_dev;
2237
2238         name = rte_vdev_device_name(dev);
2239         params = rte_vdev_device_args(dev);
2240         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2241
2242         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
2243             strlen(params) == 0) {
2244                 eth_dev = rte_eth_dev_attach_secondary(name);
2245                 if (!eth_dev) {
2246                         TAP_LOG(ERR, "Failed to probe %s", name);
2247                         return -1;
2248                 }
2249                 eth_dev->dev_ops = &ops;
2250                 eth_dev->device = &dev->device;
2251                 rte_eth_dev_probing_finish(eth_dev);
2252                 return 0;
2253         }
2254
2255         /* use tun%d which causes kernel to choose next available */
2256         strlcpy(tun_name, DEFAULT_TUN_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2257
2258         if (params && (params[0] != '\0')) {
2259                 TAP_LOG(DEBUG, "parameters (%s)", params);
2260
2261                 kvlist = rte_kvargs_parse(params, valid_arguments);
2262                 if (kvlist) {
2263                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2264                                 ret = rte_kvargs_process(kvlist,
2265                                         ETH_TAP_IFACE_ARG,
2266                                         &set_interface_name,
2267                                         tun_name);
2268
2269                                 if (ret == -1)
2270                                         goto leave;
2271                         }
2272                 }
2273         }
2274         pmd_link.link_speed = ETH_SPEED_NUM_10G;
2275
2276         TAP_LOG(DEBUG, "Initializing pmd_tun for %s", name);
2277
2278         ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0,
2279                                  ETH_TUNTAP_TYPE_TUN);
2280
2281 leave:
2282         if (ret == -1) {
2283                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2284                         name, tun_name);
2285         }
2286         rte_kvargs_free(kvlist);
2287
2288         return ret;
2289 }
2290
2291 /* Request queue file descriptors from secondary to primary. */
2292 static int
2293 tap_mp_attach_queues(const char *port_name, struct rte_eth_dev *dev)
2294 {
2295         int ret;
2296         struct timespec timeout = {.tv_sec = 1, .tv_nsec = 0};
2297         struct rte_mp_msg request, *reply;
2298         struct rte_mp_reply replies;
2299         struct ipc_queues *request_param = (struct ipc_queues *)request.param;
2300         struct ipc_queues *reply_param;
2301         struct pmd_process_private *process_private = dev->process_private;
2302         int queue, fd_iterator;
2303
2304         /* Prepare the request */
2305         memset(&request, 0, sizeof(request));
2306         strlcpy(request.name, TAP_MP_KEY, sizeof(request.name));
2307         strlcpy(request_param->port_name, port_name,
2308                 sizeof(request_param->port_name));
2309         request.len_param = sizeof(*request_param);
2310         /* Send request and receive reply */
2311         ret = rte_mp_request_sync(&request, &replies, &timeout);
2312         if (ret < 0 || replies.nb_received != 1) {
2313                 TAP_LOG(ERR, "Failed to request queues from primary: %d",
2314                         rte_errno);
2315                 return -1;
2316         }
2317         reply = &replies.msgs[0];
2318         reply_param = (struct ipc_queues *)reply->param;
2319         TAP_LOG(DEBUG, "Received IPC reply for %s", reply_param->port_name);
2320
2321         /* Attach the queues from received file descriptors */
2322         if (reply_param->rxq_count + reply_param->txq_count != reply->num_fds) {
2323                 TAP_LOG(ERR, "Unexpected number of fds received");
2324                 return -1;
2325         }
2326
2327         dev->data->nb_rx_queues = reply_param->rxq_count;
2328         dev->data->nb_tx_queues = reply_param->txq_count;
2329         fd_iterator = 0;
2330         for (queue = 0; queue < reply_param->rxq_count; queue++)
2331                 process_private->rxq_fds[queue] = reply->fds[fd_iterator++];
2332         for (queue = 0; queue < reply_param->txq_count; queue++)
2333                 process_private->txq_fds[queue] = reply->fds[fd_iterator++];
2334         free(reply);
2335         return 0;
2336 }
2337
2338 /* Send the queue file descriptors from the primary process to secondary. */
2339 static int
2340 tap_mp_sync_queues(const struct rte_mp_msg *request, const void *peer)
2341 {
2342         struct rte_eth_dev *dev;
2343         struct pmd_process_private *process_private;
2344         struct rte_mp_msg reply;
2345         const struct ipc_queues *request_param =
2346                 (const struct ipc_queues *)request->param;
2347         struct ipc_queues *reply_param =
2348                 (struct ipc_queues *)reply.param;
2349         uint16_t port_id;
2350         int queue;
2351         int ret;
2352
2353         /* Get requested port */
2354         TAP_LOG(DEBUG, "Received IPC request for %s", request_param->port_name);
2355         ret = rte_eth_dev_get_port_by_name(request_param->port_name, &port_id);
2356         if (ret) {
2357                 TAP_LOG(ERR, "Failed to get port id for %s",
2358                         request_param->port_name);
2359                 return -1;
2360         }
2361         dev = &rte_eth_devices[port_id];
2362         process_private = dev->process_private;
2363
2364         /* Fill file descriptors for all queues */
2365         reply.num_fds = 0;
2366         reply_param->rxq_count = 0;
2367         if (dev->data->nb_rx_queues + dev->data->nb_tx_queues >
2368                         RTE_MP_MAX_FD_NUM){
2369                 TAP_LOG(ERR, "Number of rx/tx queues exceeds max number of fds");
2370                 return -1;
2371         }
2372
2373         for (queue = 0; queue < dev->data->nb_rx_queues; queue++) {
2374                 reply.fds[reply.num_fds++] = process_private->rxq_fds[queue];
2375                 reply_param->rxq_count++;
2376         }
2377         RTE_ASSERT(reply_param->rxq_count == dev->data->nb_rx_queues);
2378
2379         reply_param->txq_count = 0;
2380         for (queue = 0; queue < dev->data->nb_tx_queues; queue++) {
2381                 reply.fds[reply.num_fds++] = process_private->txq_fds[queue];
2382                 reply_param->txq_count++;
2383         }
2384         RTE_ASSERT(reply_param->txq_count == dev->data->nb_tx_queues);
2385
2386         /* Send reply */
2387         strlcpy(reply.name, request->name, sizeof(reply.name));
2388         strlcpy(reply_param->port_name, request_param->port_name,
2389                 sizeof(reply_param->port_name));
2390         reply.len_param = sizeof(*reply_param);
2391         if (rte_mp_reply(&reply, peer) < 0) {
2392                 TAP_LOG(ERR, "Failed to reply an IPC request to sync queues");
2393                 return -1;
2394         }
2395         return 0;
2396 }
2397
2398 /* Open a TAP interface device.
2399  */
2400 static int
2401 rte_pmd_tap_probe(struct rte_vdev_device *dev)
2402 {
2403         const char *name, *params;
2404         int ret;
2405         struct rte_kvargs *kvlist = NULL;
2406         int speed;
2407         char tap_name[RTE_ETH_NAME_MAX_LEN];
2408         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2409         struct rte_ether_addr user_mac = { .addr_bytes = {0} };
2410         struct rte_eth_dev *eth_dev;
2411         int tap_devices_count_increased = 0;
2412
2413         name = rte_vdev_device_name(dev);
2414         params = rte_vdev_device_args(dev);
2415
2416         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
2417                 eth_dev = rte_eth_dev_attach_secondary(name);
2418                 if (!eth_dev) {
2419                         TAP_LOG(ERR, "Failed to probe %s", name);
2420                         return -1;
2421                 }
2422                 eth_dev->dev_ops = &ops;
2423                 eth_dev->device = &dev->device;
2424                 eth_dev->rx_pkt_burst = pmd_rx_burst;
2425                 eth_dev->tx_pkt_burst = pmd_tx_burst;
2426                 if (!rte_eal_primary_proc_alive(NULL)) {
2427                         TAP_LOG(ERR, "Primary process is missing");
2428                         return -1;
2429                 }
2430                 eth_dev->process_private = (struct pmd_process_private *)
2431                         rte_zmalloc_socket(name,
2432                                 sizeof(struct pmd_process_private),
2433                                 RTE_CACHE_LINE_SIZE,
2434                                 eth_dev->device->numa_node);
2435                 if (eth_dev->process_private == NULL) {
2436                         TAP_LOG(ERR,
2437                                 "Failed to alloc memory for process private");
2438                         return -1;
2439                 }
2440
2441                 ret = tap_mp_attach_queues(name, eth_dev);
2442                 if (ret != 0)
2443                         return -1;
2444                 rte_eth_dev_probing_finish(eth_dev);
2445                 return 0;
2446         }
2447
2448         speed = ETH_SPEED_NUM_10G;
2449
2450         /* use tap%d which causes kernel to choose next available */
2451         strlcpy(tap_name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2452         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2453
2454         if (params && (params[0] != '\0')) {
2455                 TAP_LOG(DEBUG, "parameters (%s)", params);
2456
2457                 kvlist = rte_kvargs_parse(params, valid_arguments);
2458                 if (kvlist) {
2459                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2460                                 ret = rte_kvargs_process(kvlist,
2461                                                          ETH_TAP_IFACE_ARG,
2462                                                          &set_interface_name,
2463                                                          tap_name);
2464                                 if (ret == -1)
2465                                         goto leave;
2466                         }
2467
2468                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
2469                                 ret = rte_kvargs_process(kvlist,
2470                                                          ETH_TAP_REMOTE_ARG,
2471                                                          &set_remote_iface,
2472                                                          remote_iface);
2473                                 if (ret == -1)
2474                                         goto leave;
2475                         }
2476
2477                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
2478                                 ret = rte_kvargs_process(kvlist,
2479                                                          ETH_TAP_MAC_ARG,
2480                                                          &set_mac_type,
2481                                                          &user_mac);
2482                                 if (ret == -1)
2483                                         goto leave;
2484                         }
2485                 }
2486         }
2487         pmd_link.link_speed = speed;
2488
2489         TAP_LOG(DEBUG, "Initializing pmd_tap for %s", name);
2490
2491         /* Register IPC feed callback */
2492         if (!tap_devices_count) {
2493                 ret = rte_mp_action_register(TAP_MP_KEY, tap_mp_sync_queues);
2494                 if (ret < 0 && rte_errno != ENOTSUP) {
2495                         TAP_LOG(ERR, "tap: Failed to register IPC callback: %s",
2496                                 strerror(rte_errno));
2497                         goto leave;
2498                 }
2499         }
2500         tap_devices_count++;
2501         tap_devices_count_increased = 1;
2502         ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
2503                 ETH_TUNTAP_TYPE_TAP);
2504
2505 leave:
2506         if (ret == -1) {
2507                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2508                         name, tap_name);
2509                 if (tap_devices_count_increased == 1) {
2510                         if (tap_devices_count == 1)
2511                                 rte_mp_action_unregister(TAP_MP_KEY);
2512                         tap_devices_count--;
2513                 }
2514         }
2515         rte_kvargs_free(kvlist);
2516
2517         return ret;
2518 }
2519
2520 /* detach a TUNTAP device.
2521  */
2522 static int
2523 rte_pmd_tap_remove(struct rte_vdev_device *dev)
2524 {
2525         struct rte_eth_dev *eth_dev = NULL;
2526
2527         /* find the ethdev entry */
2528         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
2529         if (!eth_dev)
2530                 return 0;
2531
2532         tap_dev_close(eth_dev);
2533         rte_eth_dev_release_port(eth_dev);
2534
2535         return 0;
2536 }
2537
2538 static struct rte_vdev_driver pmd_tun_drv = {
2539         .probe = rte_pmd_tun_probe,
2540         .remove = rte_pmd_tap_remove,
2541 };
2542
2543 static struct rte_vdev_driver pmd_tap_drv = {
2544         .probe = rte_pmd_tap_probe,
2545         .remove = rte_pmd_tap_remove,
2546 };
2547
2548 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
2549 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
2550 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
2551 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
2552                               ETH_TAP_IFACE_ARG "=<string> ");
2553 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
2554                               ETH_TAP_IFACE_ARG "=<string> "
2555                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
2556                               ETH_TAP_REMOTE_ARG "=<string>");
2557 RTE_LOG_REGISTER(tap_logtype, pmd.net.tap, NOTICE);