drivers/net/tap/rte_eth_tap.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2016-2017 Intel Corporation
   3  */
   4
   5 #include <rte_atomic.h>
   6 #include <rte_branch_prediction.h>
   7 #include <rte_byteorder.h>
   8 #include <rte_common.h>
   9 #include <rte_mbuf.h>
  10 #include <ethdev_driver.h>
  11 #include <ethdev_vdev.h>
  12 #include <rte_malloc.h>
  13 #include <rte_bus_vdev.h>
  14 #include <rte_kvargs.h>
  15 #include <rte_net.h>
  16 #include <rte_debug.h>
  17 #include <rte_ip.h>
  18 #include <rte_string_fns.h>
  19 #include <rte_ethdev.h>
  20 #include <rte_errno.h>
  21 #include <rte_cycles.h>
  22
  23 #include <sys/types.h>
  24 #include <sys/stat.h>
  25 #include <sys/socket.h>
  26 #include <sys/ioctl.h>
  27 #include <sys/utsname.h>
  28 #include <sys/mman.h>
  29 #include <errno.h>
  30 #include <signal.h>
  31 #include <stdbool.h>
  32 #include <stdint.h>
  33 #include <sys/uio.h>
  34 #include <unistd.h>
  35 #include <arpa/inet.h>
  36 #include <net/if.h>
  37 #include <linux/if_tun.h>
  38 #include <linux/if_ether.h>
  39 #include <fcntl.h>
  40 #include <ctype.h>
  41
  42 #include <tap_rss.h>
  43 #include <rte_eth_tap.h>
  44 #include <tap_flow.h>
  45 #include <tap_netlink.h>
  46 #include <tap_tcmsgs.h>
  47
  48 /* Linux based path to the TUN device */
  49 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  50 #define DEFAULT_TAP_NAME        "dtap"
  51 #define DEFAULT_TUN_NAME        "dtun"
  52
  53 #define ETH_TAP_IFACE_ARG       "iface"
  54 #define ETH_TAP_REMOTE_ARG      "remote"
  55 #define ETH_TAP_MAC_ARG         "mac"
  56 #define ETH_TAP_MAC_FIXED       "fixed"
  57
  58 #define ETH_TAP_USR_MAC_FMT     "xx:xx:xx:xx:xx:xx"
  59 #define ETH_TAP_CMP_MAC_FMT     "0123456789ABCDEFabcdef"
  60 #define ETH_TAP_MAC_ARG_FMT     ETH_TAP_MAC_FIXED "|" ETH_TAP_USR_MAC_FMT
  61
  62 #define TAP_GSO_MBUFS_PER_CORE  128
  63 #define TAP_GSO_MBUF_SEG_SIZE   128
  64 #define TAP_GSO_MBUF_CACHE_SIZE 4
  65 #define TAP_GSO_MBUFS_NUM \
  66         (TAP_GSO_MBUFS_PER_CORE * TAP_GSO_MBUF_CACHE_SIZE)
  67
  68 /* IPC key for queue fds sync */
  69 #define TAP_MP_KEY "tap_mp_sync_queues"
  70 #define TAP_MP_REQ_START_RXTX "tap_mp_req_start_rxtx"
  71
  72 #define TAP_IOV_DEFAULT_MAX 1024
  73
  74 #define TAP_RX_OFFLOAD (RTE_ETH_RX_OFFLOAD_SCATTER |    \
  75                         RTE_ETH_RX_OFFLOAD_IPV4_CKSUM | \
  76                         RTE_ETH_RX_OFFLOAD_UDP_CKSUM |  \
  77                         RTE_ETH_RX_OFFLOAD_TCP_CKSUM)
  78
  79 #define TAP_TX_OFFLOAD (RTE_ETH_TX_OFFLOAD_MULTI_SEGS | \
  80                         RTE_ETH_TX_OFFLOAD_IPV4_CKSUM | \
  81                         RTE_ETH_TX_OFFLOAD_UDP_CKSUM |  \
  82                         RTE_ETH_TX_OFFLOAD_TCP_CKSUM |  \
  83                         RTE_ETH_TX_OFFLOAD_TCP_TSO)
  84
  85 static int tap_devices_count;
  86
  87 static const char *tuntap_types[ETH_TUNTAP_TYPE_MAX] = {
  88         "UNKNOWN", "TUN", "TAP"
  89 };
  90
  91 static const char *valid_arguments[] = {
  92         ETH_TAP_IFACE_ARG,
  93         ETH_TAP_REMOTE_ARG,
  94         ETH_TAP_MAC_ARG,
  95         NULL
  96 };
  97
  98 static volatile uint32_t tap_trigger;   /* Rx trigger */
  99
 100 static struct rte_eth_link pmd_link = {
 101         .link_speed = RTE_ETH_SPEED_NUM_10G,
 102         .link_duplex = RTE_ETH_LINK_FULL_DUPLEX,
 103         .link_status = RTE_ETH_LINK_DOWN,
 104         .link_autoneg = RTE_ETH_LINK_FIXED,
 105 };
 106
 107 static void
 108 tap_trigger_cb(int sig __rte_unused)
 109 {
 110         /* Valid trigger values are nonzero */
 111         tap_trigger = (tap_trigger + 1) | 0x80000000;
 112 }
 113
 114 /* Specifies on what netdevices the ioctl should be applied */
 115 enum ioctl_mode {
 116         LOCAL_AND_REMOTE,
 117         LOCAL_ONLY,
 118         REMOTE_ONLY,
 119 };
 120
 121 /* Message header to synchronize queues via IPC */
 122 struct ipc_queues {
 123         char port_name[RTE_DEV_NAME_MAX_LEN];
 124         int rxq_count;
 125         int txq_count;
 126         /*
 127          * The file descriptors are in the dedicated part
 128          * of the Unix message to be translated by the kernel.
 129          */
 130 };
 131
 132 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
 133
 134 /**
 135  * Tun/Tap allocation routine
 136  *
 137  * @param[in] pmd
 138  *   Pointer to private structure.
 139  *
 140  * @param[in] is_keepalive
 141  *   Keepalive flag
 142  *
 143  * @return
 144  *   -1 on failure, fd on success
 145  */
 146 static int
 147 tun_alloc(struct pmd_internals *pmd, int is_keepalive)
 148 {
 149         struct ifreq ifr;
 150 #ifdef IFF_MULTI_QUEUE
 151         unsigned int features;
 152 #endif
 153         int fd, signo, flags;
 154
 155         memset(&ifr, 0, sizeof(struct ifreq));
 156
 157         /*
 158          * Do not set IFF_NO_PI as packet information header will be needed
 159          * to check if a received packet has been truncated.
 160          */
 161         ifr.ifr_flags = (pmd->type == ETH_TUNTAP_TYPE_TAP) ?
 162                 IFF_TAP : IFF_TUN | IFF_POINTOPOINT;
 163         strlcpy(ifr.ifr_name, pmd->name, IFNAMSIZ);
 164
 165         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 166         if (fd < 0) {
 167                 TAP_LOG(ERR, "Unable to open %s interface", TUN_TAP_DEV_PATH);
 168                 goto error;
 169         }
 170
 171 #ifdef IFF_MULTI_QUEUE
 172         /* Grab the TUN features to verify we can work multi-queue */
 173         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 174                 TAP_LOG(ERR, "unable to get TUN/TAP features");
 175                 goto error;
 176         }
 177         TAP_LOG(DEBUG, "%s Features %08x", TUN_TAP_DEV_PATH, features);
 178
 179         if (features & IFF_MULTI_QUEUE) {
 180                 TAP_LOG(DEBUG, "  Multi-queue support for %d queues",
 181                         RTE_PMD_TAP_MAX_QUEUES);
 182                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 183         } else
 184 #endif
 185         {
 186                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 187                 TAP_LOG(DEBUG, "  Single queue only support");
 188         }
 189
 190         /* Set the TUN/TAP configuration and set the name if needed */
 191         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 192                 TAP_LOG(WARNING, "Unable to set TUNSETIFF for %s: %s",
 193                         ifr.ifr_name, strerror(errno));
 194                 goto error;
 195         }
 196
 197         /*
 198          * Name passed to kernel might be wildcard like dtun%d
 199          * and need to find the resulting device.
 200          */
 201         TAP_LOG(DEBUG, "Device name is '%s'", ifr.ifr_name);
 202         strlcpy(pmd->name, ifr.ifr_name, RTE_ETH_NAME_MAX_LEN);
 203
 204         if (is_keepalive) {
 205                 /*
 206                  * Detach the TUN/TAP keep-alive queue
 207                  * to avoid traffic through it
 208                  */
 209                 ifr.ifr_flags = IFF_DETACH_QUEUE;
 210                 if (ioctl(fd, TUNSETQUEUE, (void *)&ifr) < 0) {
 211                         TAP_LOG(WARNING,
 212                                 "Unable to detach keep-alive queue for %s: %s",
 213                                 ifr.ifr_name, strerror(errno));
 214                         goto error;
 215                 }
 216         }
 217
 218         flags = fcntl(fd, F_GETFL);
 219         if (flags == -1) {
 220                 TAP_LOG(WARNING,
 221                         "Unable to get %s current flags\n",
 222                         ifr.ifr_name);
 223                 goto error;
 224         }
 225
 226         /* Always set the file descriptor to non-blocking */
 227         flags |= O_NONBLOCK;
 228         if (fcntl(fd, F_SETFL, flags) < 0) {
 229                 TAP_LOG(WARNING,
 230                         "Unable to set %s to nonblocking: %s",
 231                         ifr.ifr_name, strerror(errno));
 232                 goto error;
 233         }
 234
 235         /* Find a free realtime signal */
 236         for (signo = SIGRTMIN + 1; signo < SIGRTMAX; signo++) {
 237                 struct sigaction sa;
 238
 239                 if (sigaction(signo, NULL, &sa) == -1) {
 240                         TAP_LOG(WARNING,
 241                                 "Unable to get current rt-signal %d handler",
 242                                 signo);
 243                         goto error;
 244                 }
 245
 246                 /* Already have the handler we want on this signal  */
 247                 if (sa.sa_handler == tap_trigger_cb)
 248                         break;
 249
 250                 /* Is handler in use by application */
 251                 if (sa.sa_handler != SIG_DFL) {
 252                         TAP_LOG(DEBUG,
 253                                 "Skipping used rt-signal %d", signo);
 254                         continue;
 255                 }
 256
 257                 sa = (struct sigaction) {
 258                         .sa_flags = SA_RESTART,
 259                         .sa_handler = tap_trigger_cb,
 260                 };
 261
 262                 if (sigaction(signo, &sa, NULL) == -1) {
 263                         TAP_LOG(WARNING,
 264                                 "Unable to set rt-signal %d handler\n", signo);
 265                         goto error;
 266                 }
 267
 268                 /* Found a good signal to use */
 269                 TAP_LOG(DEBUG,
 270                         "Using rt-signal %d", signo);
 271                 break;
 272         }
 273
 274         if (signo == SIGRTMAX) {
 275                 TAP_LOG(WARNING, "All rt-signals are in use\n");
 276
 277                 /* Disable trigger globally in case of error */
 278                 tap_trigger = 0;
 279                 TAP_LOG(NOTICE, "No Rx trigger signal available\n");
 280         } else {
 281                 /* Enable signal on file descriptor */
 282                 if (fcntl(fd, F_SETSIG, signo) < 0) {
 283                         TAP_LOG(WARNING, "Unable to set signo %d for fd %d: %s",
 284                                 signo, fd, strerror(errno));
 285                         goto error;
 286                 }
 287                 if (fcntl(fd, F_SETFL, flags | O_ASYNC) < 0) {
 288                         TAP_LOG(WARNING, "Unable to set fcntl flags: %s",
 289                                 strerror(errno));
 290                         goto error;
 291                 }
 292
 293                 if (fcntl(fd, F_SETOWN, getpid()) < 0) {
 294                         TAP_LOG(WARNING, "Unable to set fcntl owner: %s",
 295                                 strerror(errno));
 296                         goto error;
 297                 }
 298         }
 299         return fd;
 300
 301 error:
 302         if (fd >= 0)
 303                 close(fd);
 304         return -1;
 305 }
 306
 307 static void
 308 tap_verify_csum(struct rte_mbuf *mbuf)
 309 {
 310         uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
 311         uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
 312         uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
 313         unsigned int l2_len = sizeof(struct rte_ether_hdr);
 314         unsigned int l3_len;
 315         uint16_t cksum = 0;
 316         void *l3_hdr;
 317         void *l4_hdr;
 318         struct rte_udp_hdr *udp_hdr;
 319
 320         if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
 321                 l2_len += 4;
 322         else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
 323                 l2_len += 8;
 324         /* Don't verify checksum for packets with discontinuous L2 header */
 325         if (unlikely(l2_len + sizeof(struct rte_ipv4_hdr) >
 326                      rte_pktmbuf_data_len(mbuf)))
 327                 return;
 328         l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
 329         if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 330                 struct rte_ipv4_hdr *iph = l3_hdr;
 331
 332                 l3_len = rte_ipv4_hdr_len(iph);
 333                 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
 334                         return;
 335                 /* check that the total length reported by header is not
 336                  * greater than the total received size
 337                  */
 338                 if (l2_len + rte_be_to_cpu_16(iph->total_length) >
 339                                 rte_pktmbuf_data_len(mbuf))
 340                         return;
 341
 342                 cksum = ~rte_raw_cksum(iph, l3_len);
 343                 mbuf->ol_flags |= cksum ?
 344                         RTE_MBUF_F_RX_IP_CKSUM_BAD :
 345                         RTE_MBUF_F_RX_IP_CKSUM_GOOD;
 346         } else if (l3 == RTE_PTYPE_L3_IPV6) {
 347                 struct rte_ipv6_hdr *iph = l3_hdr;
 348
 349                 l3_len = sizeof(struct rte_ipv6_hdr);
 350                 /* check that the total length reported by header is not
 351                  * greater than the total received size
 352                  */
 353                 if (l2_len + l3_len + rte_be_to_cpu_16(iph->payload_len) >
 354                                 rte_pktmbuf_data_len(mbuf))
 355                         return;
 356         } else {
 357                 /* - RTE_PTYPE_L3_IPV4_EXT_UNKNOWN cannot happen because
 358                  *   mbuf->packet_type is filled by rte_net_get_ptype() which
 359                  *   never returns this value.
 360                  * - IPv6 extensions are not supported.
 361                  */
 362                 return;
 363         }
 364         if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
 365                 int cksum_ok;
 366
 367                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 368                 /* Don't verify checksum for multi-segment packets. */
 369                 if (mbuf->nb_segs > 1)
 370                         return;
 371                 if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 372                         if (l4 == RTE_PTYPE_L4_UDP) {
 373                                 udp_hdr = (struct rte_udp_hdr *)l4_hdr;
 374                                 if (udp_hdr->dgram_cksum == 0) {
 375                                         /*
 376                                          * For IPv4, a zero UDP checksum
 377                                          * indicates that the sender did not
 378                                          * generate one [RFC 768].
 379                                          */
 380                                         mbuf->ol_flags |= RTE_MBUF_F_RX_L4_CKSUM_NONE;
 381                                         return;
 382                                 }
 383                         }
 384                         cksum_ok = !rte_ipv4_udptcp_cksum_verify(l3_hdr,
 385                                                                  l4_hdr);
 386                 } else { /* l3 == RTE_PTYPE_L3_IPV6, checked above */
 387                         cksum_ok = !rte_ipv6_udptcp_cksum_verify(l3_hdr,
 388                                                                  l4_hdr);
 389                 }
 390                 mbuf->ol_flags |= cksum_ok ?
 391                         RTE_MBUF_F_RX_L4_CKSUM_GOOD : RTE_MBUF_F_RX_L4_CKSUM_BAD;
 392         }
 393 }
 394
 395 static void
 396 tap_rxq_pool_free(struct rte_mbuf *pool)
 397 {
 398         struct rte_mbuf *mbuf = pool;
 399         uint16_t nb_segs = 1;
 400
 401         if (mbuf == NULL)
 402                 return;
 403
 404         while (mbuf->next) {
 405                 mbuf = mbuf->next;
 406                 nb_segs++;
 407         }
 408         pool->nb_segs = nb_segs;
 409         rte_pktmbuf_free(pool);
 410 }
 411
 412 /* Callback to handle the rx burst of packets to the correct interface and
 413  * file descriptor(s) in a multi-queue setup.
 414  */
 415 static uint16_t
 416 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 417 {
 418         struct rx_queue *rxq = queue;
 419         struct pmd_process_private *process_private;
 420         uint16_t num_rx;
 421         unsigned long num_rx_bytes = 0;
 422         uint32_t trigger = tap_trigger;
 423
 424         if (trigger == rxq->trigger_seen)
 425                 return 0;
 426
 427         process_private = rte_eth_devices[rxq->in_port].process_private;
 428         for (num_rx = 0; num_rx < nb_pkts; ) {
 429                 struct rte_mbuf *mbuf = rxq->pool;
 430                 struct rte_mbuf *seg = NULL;
 431                 struct rte_mbuf *new_tail = NULL;
 432                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 433                 int len;
 434
 435                 len = readv(process_private->rxq_fds[rxq->queue_id],
 436                         *rxq->iovecs,
 437                         1 + (rxq->rxmode->offloads & RTE_ETH_RX_OFFLOAD_SCATTER ?
 438                              rxq->nb_rx_desc : 1));
 439                 if (len < (int)sizeof(struct tun_pi))
 440                         break;
 441
 442                 /* Packet couldn't fit in the provided mbuf */
 443                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 444                         rxq->stats.ierrors++;
 445                         continue;
 446                 }
 447
 448                 len -= sizeof(struct tun_pi);
 449
 450                 mbuf->pkt_len = len;
 451                 mbuf->port = rxq->in_port;
 452                 while (1) {
 453                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 454
 455                         if (unlikely(!buf)) {
 456                                 rxq->stats.rx_nombuf++;
 457                                 /* No new buf has been allocated: do nothing */
 458                                 if (!new_tail || !seg)
 459                                         goto end;
 460
 461                                 seg->next = NULL;
 462                                 tap_rxq_pool_free(mbuf);
 463
 464                                 goto end;
 465                         }
 466                         seg = seg ? seg->next : mbuf;
 467                         if (rxq->pool == mbuf)
 468                                 rxq->pool = buf;
 469                         if (new_tail)
 470                                 new_tail->next = buf;
 471                         new_tail = buf;
 472                         new_tail->next = seg->next;
 473
 474                         /* iovecs[0] is reserved for packet info (pi) */
 475                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 476                                 buf->buf_len - data_off;
 477                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 478                                 (char *)buf->buf_addr + data_off;
 479
 480                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 481                         seg->data_off = data_off;
 482
 483                         len -= seg->data_len;
 484                         if (len <= 0)
 485                                 break;
 486                         mbuf->nb_segs++;
 487                         /* First segment has headroom, not the others */
 488                         data_off = 0;
 489                 }
 490                 seg->next = NULL;
 491                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 492                                                       RTE_PTYPE_ALL_MASK);
 493                 if (rxq->rxmode->offloads & RTE_ETH_RX_OFFLOAD_CHECKSUM)
 494                         tap_verify_csum(mbuf);
 495
 496                 /* account for the receive frame */
 497                 bufs[num_rx++] = mbuf;
 498                 num_rx_bytes += mbuf->pkt_len;
 499         }
 500 end:
 501         rxq->stats.ipackets += num_rx;
 502         rxq->stats.ibytes += num_rx_bytes;
 503
 504         if (trigger && num_rx < nb_pkts)
 505                 rxq->trigger_seen = trigger;
 506
 507         return num_rx;
 508 }
 509
 510 /* Finalize l4 checksum calculation */
 511 static void
 512 tap_tx_l4_cksum(uint16_t *l4_cksum, uint16_t l4_phdr_cksum,
 513                 uint32_t l4_raw_cksum)
 514 {
 515         if (l4_cksum) {
 516                 uint32_t cksum;
 517
 518                 cksum = __rte_raw_cksum_reduce(l4_raw_cksum);
 519                 cksum += l4_phdr_cksum;
 520
 521                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 522                 cksum = (~cksum) & 0xffff;
 523                 if (cksum == 0)
 524                         cksum = 0xffff;
 525                 *l4_cksum = cksum;
 526         }
 527 }
 528
 529 /* Accumulate L4 raw checksums */
 530 static void
 531 tap_tx_l4_add_rcksum(char *l4_data, unsigned int l4_len, uint16_t *l4_cksum,
 532                         uint32_t *l4_raw_cksum)
 533 {
 534         if (l4_cksum == NULL)
 535                 return;
 536
 537         *l4_raw_cksum = __rte_raw_cksum(l4_data, l4_len, *l4_raw_cksum);
 538 }
 539
 540 /* L3 and L4 pseudo headers checksum offloads */
 541 static void
 542 tap_tx_l3_cksum(char *packet, uint64_t ol_flags, unsigned int l2_len,
 543                 unsigned int l3_len, unsigned int l4_len, uint16_t **l4_cksum,
 544                 uint16_t *l4_phdr_cksum, uint32_t *l4_raw_cksum)
 545 {
 546         void *l3_hdr = packet + l2_len;
 547
 548         if (ol_flags & (RTE_MBUF_F_TX_IP_CKSUM | RTE_MBUF_F_TX_IPV4)) {
 549                 struct rte_ipv4_hdr *iph = l3_hdr;
 550                 uint16_t cksum;
 551
 552                 iph->hdr_checksum = 0;
 553                 cksum = rte_raw_cksum(iph, l3_len);
 554                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 555         }
 556         if (ol_flags & RTE_MBUF_F_TX_L4_MASK) {
 557                 void *l4_hdr;
 558
 559                 l4_hdr = packet + l2_len + l3_len;
 560                 if ((ol_flags & RTE_MBUF_F_TX_L4_MASK) == RTE_MBUF_F_TX_UDP_CKSUM)
 561                         *l4_cksum = &((struct rte_udp_hdr *)l4_hdr)->dgram_cksum;
 562                 else if ((ol_flags & RTE_MBUF_F_TX_L4_MASK) == RTE_MBUF_F_TX_TCP_CKSUM)
 563                         *l4_cksum = &((struct rte_tcp_hdr *)l4_hdr)->cksum;
 564                 else
 565                         return;
 566                 **l4_cksum = 0;
 567                 if (ol_flags & RTE_MBUF_F_TX_IPV4)
 568                         *l4_phdr_cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 569                 else
 570                         *l4_phdr_cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 571                 *l4_raw_cksum = __rte_raw_cksum(l4_hdr, l4_len, 0);
 572         }
 573 }
 574
 575 static inline int
 576 tap_write_mbufs(struct tx_queue *txq, uint16_t num_mbufs,
 577                         struct rte_mbuf **pmbufs,
 578                         uint16_t *num_packets, unsigned long *num_tx_bytes)
 579 {
 580         int i;
 581         uint16_t l234_hlen;
 582         struct pmd_process_private *process_private;
 583
 584         process_private = rte_eth_devices[txq->out_port].process_private;
 585
 586         for (i = 0; i < num_mbufs; i++) {
 587                 struct rte_mbuf *mbuf = pmbufs[i];
 588                 struct iovec iovecs[mbuf->nb_segs + 2];
 589                 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
 590                 struct rte_mbuf *seg = mbuf;
 591                 char m_copy[mbuf->data_len];
 592                 int proto;
 593                 int n;
 594                 int j;
 595                 int k; /* current index in iovecs for copying segments */
 596                 uint16_t seg_len; /* length of first segment */
 597                 uint16_t nb_segs;
 598                 uint16_t *l4_cksum; /* l4 checksum (pseudo header + payload) */
 599                 uint32_t l4_raw_cksum = 0; /* TCP/UDP payload raw checksum */
 600                 uint16_t l4_phdr_cksum = 0; /* TCP/UDP pseudo header checksum */
 601                 uint16_t is_cksum = 0; /* in case cksum should be offloaded */
 602
 603                 l4_cksum = NULL;
 604                 if (txq->type == ETH_TUNTAP_TYPE_TUN) {
 605                         /*
 606                          * TUN and TAP are created with IFF_NO_PI disabled.
 607                          * For TUN PMD this mandatory as fields are used by
 608                          * Kernel tun.c to determine whether its IP or non IP
 609                          * packets.
 610                          *
 611                          * The logic fetches the first byte of data from mbuf
 612                          * then compares whether its v4 or v6. If first byte
 613                          * is 4 or 6, then protocol field is updated.
 614                          */
 615                         char *buff_data = rte_pktmbuf_mtod(seg, void *);
 616                         proto = (*buff_data & 0xf0);
 617                         pi.proto = (proto == 0x40) ?
 618                                 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4) :
 619                                 ((proto == 0x60) ?
 620                                         rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6) :
 621                                         0x00);
 622                 }
 623
 624                 k = 0;
 625                 iovecs[k].iov_base = &pi;
 626                 iovecs[k].iov_len = sizeof(pi);
 627                 k++;
 628
 629                 nb_segs = mbuf->nb_segs;
 630                 if (txq->csum &&
 631                     ((mbuf->ol_flags & (RTE_MBUF_F_TX_IP_CKSUM | RTE_MBUF_F_TX_IPV4) ||
 632                       (mbuf->ol_flags & RTE_MBUF_F_TX_L4_MASK) == RTE_MBUF_F_TX_UDP_CKSUM ||
 633                       (mbuf->ol_flags & RTE_MBUF_F_TX_L4_MASK) == RTE_MBUF_F_TX_TCP_CKSUM))) {
 634                         is_cksum = 1;
 635
 636                         /* Support only packets with at least layer 4
 637                          * header included in the first segment
 638                          */
 639                         seg_len = rte_pktmbuf_data_len(mbuf);
 640                         l234_hlen = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
 641                         if (seg_len < l234_hlen)
 642                                 return -1;
 643
 644                         /* To change checksums, work on a * copy of l2, l3
 645                          * headers + l4 pseudo header
 646                          */
 647                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 648                                         l234_hlen);
 649                         tap_tx_l3_cksum(m_copy, mbuf->ol_flags,
 650                                        mbuf->l2_len, mbuf->l3_len, mbuf->l4_len,
 651                                        &l4_cksum, &l4_phdr_cksum,
 652                                        &l4_raw_cksum);
 653                         iovecs[k].iov_base = m_copy;
 654                         iovecs[k].iov_len = l234_hlen;
 655                         k++;
 656
 657                         /* Update next iovecs[] beyond l2, l3, l4 headers */
 658                         if (seg_len > l234_hlen) {
 659                                 iovecs[k].iov_len = seg_len - l234_hlen;
 660                                 iovecs[k].iov_base =
 661                                         rte_pktmbuf_mtod(seg, char *) +
 662                                                 l234_hlen;
 663                                 tap_tx_l4_add_rcksum(iovecs[k].iov_base,
 664                                         iovecs[k].iov_len, l4_cksum,
 665                                         &l4_raw_cksum);
 666                                 k++;
 667                                 nb_segs++;
 668                         }
 669                         seg = seg->next;
 670                 }
 671
 672                 for (j = k; j <= nb_segs; j++) {
 673                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 674                         iovecs[j].iov_base = rte_pktmbuf_mtod(seg, void *);
 675                         if (is_cksum)
 676                                 tap_tx_l4_add_rcksum(iovecs[j].iov_base,
 677                                         iovecs[j].iov_len, l4_cksum,
 678                                         &l4_raw_cksum);
 679                         seg = seg->next;
 680                 }
 681
 682                 if (is_cksum)
 683                         tap_tx_l4_cksum(l4_cksum, l4_phdr_cksum, l4_raw_cksum);
 684
 685                 /* copy the tx frame data */
 686                 n = writev(process_private->txq_fds[txq->queue_id], iovecs, j);
 687                 if (n <= 0)
 688                         return -1;
 689
 690                 (*num_packets)++;
 691                 (*num_tx_bytes) += rte_pktmbuf_pkt_len(mbuf);
 692         }
 693         return 0;
 694 }
 695
 696 /* Callback to handle sending packets from the tap interface
 697  */
 698 static uint16_t
 699 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 700 {
 701         struct tx_queue *txq = queue;
 702         uint16_t num_tx = 0;
 703         uint16_t num_packets = 0;
 704         unsigned long num_tx_bytes = 0;
 705         uint32_t max_size;
 706         int i;
 707
 708         if (unlikely(nb_pkts == 0))
 709                 return 0;
 710
 711         struct rte_mbuf *gso_mbufs[MAX_GSO_MBUFS];
 712         max_size = *txq->mtu + (RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN + 4);
 713         for (i = 0; i < nb_pkts; i++) {
 714                 struct rte_mbuf *mbuf_in = bufs[num_tx];
 715                 struct rte_mbuf **mbuf;
 716                 uint16_t num_mbufs = 0;
 717                 uint16_t tso_segsz = 0;
 718                 int ret;
 719                 int num_tso_mbufs;
 720                 uint16_t hdrs_len;
 721                 uint64_t tso;
 722
 723                 tso = mbuf_in->ol_flags & RTE_MBUF_F_TX_TCP_SEG;
 724                 if (tso) {
 725                         struct rte_gso_ctx *gso_ctx = &txq->gso_ctx;
 726
 727                         /* TCP segmentation implies TCP checksum offload */
 728                         mbuf_in->ol_flags |= RTE_MBUF_F_TX_TCP_CKSUM;
 729
 730                         /* gso size is calculated without RTE_ETHER_CRC_LEN */
 731                         hdrs_len = mbuf_in->l2_len + mbuf_in->l3_len +
 732                                         mbuf_in->l4_len;
 733                         tso_segsz = mbuf_in->tso_segsz + hdrs_len;
 734                         if (unlikely(tso_segsz == hdrs_len) ||
 735                                 tso_segsz > *txq->mtu) {
 736                                 txq->stats.errs++;
 737                                 break;
 738                         }
 739                         gso_ctx->gso_size = tso_segsz;
 740                         /* 'mbuf_in' packet to segment */
 741                         num_tso_mbufs = rte_gso_segment(mbuf_in,
 742                                 gso_ctx, /* gso control block */
 743                                 (struct rte_mbuf **)&gso_mbufs, /* out mbufs */
 744                                 RTE_DIM(gso_mbufs)); /* max tso mbufs */
 745
 746                         /* ret contains the number of new created mbufs */
 747                         if (num_tso_mbufs < 0)
 748                                 break;
 749
 750                         if (num_tso_mbufs >= 1) {
 751                                 mbuf = gso_mbufs;
 752                                 num_mbufs = num_tso_mbufs;
 753                         } else {
 754                                 /* 0 means it can be transmitted directly
 755                                  * without gso.
 756                                  */
 757                                 mbuf = &mbuf_in;
 758                                 num_mbufs = 1;
 759                         }
 760                 } else {
 761                         /* stats.errs will be incremented */
 762                         if (rte_pktmbuf_pkt_len(mbuf_in) > max_size)
 763                                 break;
 764
 765                         /* ret 0 indicates no new mbufs were created */
 766                         num_tso_mbufs = 0;
 767                         mbuf = &mbuf_in;
 768                         num_mbufs = 1;
 769                 }
 770
 771                 ret = tap_write_mbufs(txq, num_mbufs, mbuf,
 772                                 &num_packets, &num_tx_bytes);
 773                 if (ret == -1) {
 774                         txq->stats.errs++;
 775                         /* free tso mbufs */
 776                         if (num_tso_mbufs > 0)
 777                                 rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
 778                         break;
 779                 }
 780                 num_tx++;
 781                 /* free original mbuf */
 782                 rte_pktmbuf_free(mbuf_in);
 783                 /* free tso mbufs */
 784                 if (num_tso_mbufs > 0)
 785                         rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
 786         }
 787
 788         txq->stats.opackets += num_packets;
 789         txq->stats.errs += nb_pkts - num_tx;
 790         txq->stats.obytes += num_tx_bytes;
 791
 792         return num_tx;
 793 }
 794
 795 static const char *
 796 tap_ioctl_req2str(unsigned long request)
 797 {
 798         switch (request) {
 799         case SIOCSIFFLAGS:
 800                 return "SIOCSIFFLAGS";
 801         case SIOCGIFFLAGS:
 802                 return "SIOCGIFFLAGS";
 803         case SIOCGIFHWADDR:
 804                 return "SIOCGIFHWADDR";
 805         case SIOCSIFHWADDR:
 806                 return "SIOCSIFHWADDR";
 807         case SIOCSIFMTU:
 808                 return "SIOCSIFMTU";
 809         }
 810         return "UNKNOWN";
 811 }
 812
 813 static int
 814 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 815           struct ifreq *ifr, int set, enum ioctl_mode mode)
 816 {
 817         short req_flags = ifr->ifr_flags;
 818         int remote = pmd->remote_if_index &&
 819                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 820
 821         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 822                 return 0;
 823         /*
 824          * If there is a remote netdevice, apply ioctl on it, then apply it on
 825          * the tap netdevice.
 826          */
 827 apply:
 828         if (remote)
 829                 strlcpy(ifr->ifr_name, pmd->remote_iface, IFNAMSIZ);
 830         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 831                 strlcpy(ifr->ifr_name, pmd->name, IFNAMSIZ);
 832         switch (request) {
 833         case SIOCSIFFLAGS:
 834                 /* fetch current flags to leave other flags untouched */
 835                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 836                         goto error;
 837                 if (set)
 838                         ifr->ifr_flags |= req_flags;
 839                 else
 840                         ifr->ifr_flags &= ~req_flags;
 841                 break;
 842         case SIOCGIFFLAGS:
 843         case SIOCGIFHWADDR:
 844         case SIOCSIFHWADDR:
 845         case SIOCSIFMTU:
 846                 break;
 847         default:
 848                 TAP_LOG(WARNING, "%s: ioctl() called with wrong arg",
 849                         pmd->name);
 850                 return -EINVAL;
 851         }
 852         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 853                 goto error;
 854         if (remote-- && mode == LOCAL_AND_REMOTE)
 855                 goto apply;
 856         return 0;
 857
 858 error:
 859         TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
 860                 tap_ioctl_req2str(request), strerror(errno), errno);
 861         return -errno;
 862 }
 863
 864 static int
 865 tap_link_set_down(struct rte_eth_dev *dev)
 866 {
 867         struct pmd_internals *pmd = dev->data->dev_private;
 868         struct ifreq ifr = { .ifr_flags = IFF_UP };
 869
 870         dev->data->dev_link.link_status = RTE_ETH_LINK_DOWN;
 871         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 872 }
 873
 874 static int
 875 tap_link_set_up(struct rte_eth_dev *dev)
 876 {
 877         struct pmd_internals *pmd = dev->data->dev_private;
 878         struct ifreq ifr = { .ifr_flags = IFF_UP };
 879
 880         dev->data->dev_link.link_status = RTE_ETH_LINK_UP;
 881         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 882 }
 883
 884 static int
 885 tap_mp_req_on_rxtx(struct rte_eth_dev *dev)
 886 {
 887         struct rte_mp_msg msg;
 888         struct ipc_queues *request_param = (struct ipc_queues *)msg.param;
 889         int err;
 890         int fd_iterator = 0;
 891         struct pmd_process_private *process_private = dev->process_private;
 892         int i;
 893
 894         memset(&msg, 0, sizeof(msg));
 895         strlcpy(msg.name, TAP_MP_REQ_START_RXTX, sizeof(msg.name));
 896         strlcpy(request_param->port_name, dev->data->name, sizeof(request_param->port_name));
 897         msg.len_param = sizeof(*request_param);
 898         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 899                 msg.fds[fd_iterator++] = process_private->txq_fds[i];
 900                 msg.num_fds++;
 901                 request_param->txq_count++;
 902         }
 903         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 904                 msg.fds[fd_iterator++] = process_private->rxq_fds[i];
 905                 msg.num_fds++;
 906                 request_param->rxq_count++;
 907         }
 908
 909         err = rte_mp_sendmsg(&msg);
 910         if (err < 0) {
 911                 TAP_LOG(ERR, "Failed to send start req to secondary %d",
 912                         rte_errno);
 913                 return -1;
 914         }
 915
 916         return 0;
 917 }
 918
 919 static int
 920 tap_dev_start(struct rte_eth_dev *dev)
 921 {
 922         int err, i;
 923
 924         if (rte_eal_process_type() == RTE_PROC_PRIMARY)
 925                 tap_mp_req_on_rxtx(dev);
 926
 927         err = tap_intr_handle_set(dev, 1);
 928         if (err)
 929                 return err;
 930
 931         err = tap_link_set_up(dev);
 932         if (err)
 933                 return err;
 934
 935         for (i = 0; i < dev->data->nb_tx_queues; i++)
 936                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 937         for (i = 0; i < dev->data->nb_rx_queues; i++)
 938                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 939
 940         return err;
 941 }
 942
 943 static int
 944 tap_mp_req_start_rxtx(const struct rte_mp_msg *request, __rte_unused const void *peer)
 945 {
 946         struct rte_eth_dev *dev;
 947         const struct ipc_queues *request_param =
 948                 (const struct ipc_queues *)request->param;
 949         int fd_iterator;
 950         int queue;
 951         struct pmd_process_private *process_private;
 952
 953         dev = rte_eth_dev_get_by_name(request_param->port_name);
 954         if (!dev) {
 955                 TAP_LOG(ERR, "Failed to get dev for %s",
 956                         request_param->port_name);
 957                 return -1;
 958         }
 959         process_private = dev->process_private;
 960         fd_iterator = 0;
 961         TAP_LOG(DEBUG, "tap_attach rx_q:%d tx_q:%d\n", request_param->rxq_count,
 962                 request_param->txq_count);
 963         for (queue = 0; queue < request_param->txq_count; queue++)
 964                 process_private->txq_fds[queue] = request->fds[fd_iterator++];
 965         for (queue = 0; queue < request_param->rxq_count; queue++)
 966                 process_private->rxq_fds[queue] = request->fds[fd_iterator++];
 967
 968         return 0;
 969 }
 970
 971 /* This function gets called when the current port gets stopped.
 972  */
 973 static int
 974 tap_dev_stop(struct rte_eth_dev *dev)
 975 {
 976         int i;
 977
 978         for (i = 0; i < dev->data->nb_tx_queues; i++)
 979                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 980         for (i = 0; i < dev->data->nb_rx_queues; i++)
 981                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 982
 983         tap_intr_handle_set(dev, 0);
 984         tap_link_set_down(dev);
 985
 986         return 0;
 987 }
 988
 989 static int
 990 tap_dev_configure(struct rte_eth_dev *dev)
 991 {
 992         struct pmd_internals *pmd = dev->data->dev_private;
 993
 994         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 995                 TAP_LOG(ERR,
 996                         "%s: number of rx queues %d exceeds max num of queues %d",
 997                         dev->device->name,
 998                         dev->data->nb_rx_queues,
 999                         RTE_PMD_TAP_MAX_QUEUES);
1000                 return -1;
1001         }
1002         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
1003                 TAP_LOG(ERR,
1004                         "%s: number of tx queues %d exceeds max num of queues %d",
1005                         dev->device->name,
1006                         dev->data->nb_tx_queues,
1007                         RTE_PMD_TAP_MAX_QUEUES);
1008                 return -1;
1009         }
1010         if (dev->data->nb_rx_queues != dev->data->nb_tx_queues) {
1011                 TAP_LOG(ERR,
1012                         "%s: number of rx queues %d must be equal to number of tx queues %d",
1013                         dev->device->name,
1014                         dev->data->nb_rx_queues,
1015                         dev->data->nb_tx_queues);
1016                 return -1;
1017         }
1018
1019         TAP_LOG(INFO, "%s: %s: TX configured queues number: %u",
1020                 dev->device->name, pmd->name, dev->data->nb_tx_queues);
1021
1022         TAP_LOG(INFO, "%s: %s: RX configured queues number: %u",
1023                 dev->device->name, pmd->name, dev->data->nb_rx_queues);
1024
1025         return 0;
1026 }
1027
1028 static uint32_t
1029 tap_dev_speed_capa(void)
1030 {
1031         uint32_t speed = pmd_link.link_speed;
1032         uint32_t capa = 0;
1033
1034         if (speed >= RTE_ETH_SPEED_NUM_10M)
1035                 capa |= RTE_ETH_LINK_SPEED_10M;
1036         if (speed >= RTE_ETH_SPEED_NUM_100M)
1037                 capa |= RTE_ETH_LINK_SPEED_100M;
1038         if (speed >= RTE_ETH_SPEED_NUM_1G)
1039                 capa |= RTE_ETH_LINK_SPEED_1G;
1040         if (speed >= RTE_ETH_SPEED_NUM_5G)
1041                 capa |= RTE_ETH_LINK_SPEED_2_5G;
1042         if (speed >= RTE_ETH_SPEED_NUM_5G)
1043                 capa |= RTE_ETH_LINK_SPEED_5G;
1044         if (speed >= RTE_ETH_SPEED_NUM_10G)
1045                 capa |= RTE_ETH_LINK_SPEED_10G;
1046         if (speed >= RTE_ETH_SPEED_NUM_20G)
1047                 capa |= RTE_ETH_LINK_SPEED_20G;
1048         if (speed >= RTE_ETH_SPEED_NUM_25G)
1049                 capa |= RTE_ETH_LINK_SPEED_25G;
1050         if (speed >= RTE_ETH_SPEED_NUM_40G)
1051                 capa |= RTE_ETH_LINK_SPEED_40G;
1052         if (speed >= RTE_ETH_SPEED_NUM_50G)
1053                 capa |= RTE_ETH_LINK_SPEED_50G;
1054         if (speed >= RTE_ETH_SPEED_NUM_56G)
1055                 capa |= RTE_ETH_LINK_SPEED_56G;
1056         if (speed >= RTE_ETH_SPEED_NUM_100G)
1057                 capa |= RTE_ETH_LINK_SPEED_100G;
1058
1059         return capa;
1060 }
1061
1062 static int
1063 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1064 {
1065         struct pmd_internals *internals = dev->data->dev_private;
1066
1067         dev_info->if_index = internals->if_index;
1068         dev_info->max_mac_addrs = 1;
1069         dev_info->max_rx_pktlen = (uint32_t)RTE_ETHER_MAX_VLAN_FRAME_LEN;
1070         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
1071         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
1072         dev_info->min_rx_bufsize = 0;
1073         dev_info->speed_capa = tap_dev_speed_capa();
1074         dev_info->rx_queue_offload_capa = TAP_RX_OFFLOAD;
1075         dev_info->rx_offload_capa = dev_info->rx_queue_offload_capa;
1076         dev_info->tx_queue_offload_capa = TAP_TX_OFFLOAD;
1077         dev_info->tx_offload_capa = dev_info->tx_queue_offload_capa;
1078         dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
1079         /*
1080          * limitation: TAP supports all of IP, UDP and TCP hash
1081          * functions together and not in partial combinations
1082          */
1083         dev_info->flow_type_rss_offloads = ~TAP_RSS_HF_MASK;
1084         dev_info->dev_capa &= ~RTE_ETH_DEV_CAPA_FLOW_RULE_KEEP;
1085
1086         return 0;
1087 }
1088
1089 static int
1090 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
1091 {
1092         unsigned int i, imax;
1093         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
1094         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
1095         unsigned long rx_nombuf = 0, ierrors = 0;
1096         const struct pmd_internals *pmd = dev->data->dev_private;
1097
1098         /* rx queue statistics */
1099         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
1100                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
1101         for (i = 0; i < imax; i++) {
1102                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
1103                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
1104                 rx_total += tap_stats->q_ipackets[i];
1105                 rx_bytes_total += tap_stats->q_ibytes[i];
1106                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
1107                 ierrors += pmd->rxq[i].stats.ierrors;
1108         }
1109
1110         /* tx queue statistics */
1111         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
1112                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
1113
1114         for (i = 0; i < imax; i++) {
1115                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
1116                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
1117                 tx_total += tap_stats->q_opackets[i];
1118                 tx_err_total += pmd->txq[i].stats.errs;
1119                 tx_bytes_total += tap_stats->q_obytes[i];
1120         }
1121
1122         tap_stats->ipackets = rx_total;
1123         tap_stats->ibytes = rx_bytes_total;
1124         tap_stats->ierrors = ierrors;
1125         tap_stats->rx_nombuf = rx_nombuf;
1126         tap_stats->opackets = tx_total;
1127         tap_stats->oerrors = tx_err_total;
1128         tap_stats->obytes = tx_bytes_total;
1129         return 0;
1130 }
1131
1132 static int
1133 tap_stats_reset(struct rte_eth_dev *dev)
1134 {
1135         int i;
1136         struct pmd_internals *pmd = dev->data->dev_private;
1137
1138         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1139                 pmd->rxq[i].stats.ipackets = 0;
1140                 pmd->rxq[i].stats.ibytes = 0;
1141                 pmd->rxq[i].stats.ierrors = 0;
1142                 pmd->rxq[i].stats.rx_nombuf = 0;
1143
1144                 pmd->txq[i].stats.opackets = 0;
1145                 pmd->txq[i].stats.errs = 0;
1146                 pmd->txq[i].stats.obytes = 0;
1147         }
1148
1149         return 0;
1150 }
1151
1152 static int
1153 tap_dev_close(struct rte_eth_dev *dev)
1154 {
1155         int i;
1156         struct pmd_internals *internals = dev->data->dev_private;
1157         struct pmd_process_private *process_private = dev->process_private;
1158         struct rx_queue *rxq;
1159
1160         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1161                 rte_free(dev->process_private);
1162                 if (tap_devices_count == 1)
1163                         rte_mp_action_unregister(TAP_MP_REQ_START_RXTX);
1164                 tap_devices_count--;
1165                 return 0;
1166         }
1167
1168         tap_link_set_down(dev);
1169         if (internals->nlsk_fd != -1) {
1170                 tap_flow_flush(dev, NULL);
1171                 tap_flow_implicit_flush(internals, NULL);
1172                 tap_nl_final(internals->nlsk_fd);
1173                 internals->nlsk_fd = -1;
1174         }
1175
1176         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1177                 if (process_private->rxq_fds[i] != -1) {
1178                         rxq = &internals->rxq[i];
1179                         close(process_private->rxq_fds[i]);
1180                         process_private->rxq_fds[i] = -1;
1181                         tap_rxq_pool_free(rxq->pool);
1182                         rte_free(rxq->iovecs);
1183                         rxq->pool = NULL;
1184                         rxq->iovecs = NULL;
1185                 }
1186                 if (process_private->txq_fds[i] != -1) {
1187                         close(process_private->txq_fds[i]);
1188                         process_private->txq_fds[i] = -1;
1189                 }
1190         }
1191
1192         if (internals->remote_if_index) {
1193                 /* Restore initial remote state */
1194                 int ret = ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
1195                                 &internals->remote_initial_flags);
1196                 if (ret)
1197                         TAP_LOG(ERR, "restore remote state failed: %d", ret);
1198
1199         }
1200
1201         rte_mempool_free(internals->gso_ctx_mp);
1202         internals->gso_ctx_mp = NULL;
1203
1204         if (internals->ka_fd != -1) {
1205                 close(internals->ka_fd);
1206                 internals->ka_fd = -1;
1207         }
1208
1209         /* mac_addrs must not be freed alone because part of dev_private */
1210         dev->data->mac_addrs = NULL;
1211
1212         internals = dev->data->dev_private;
1213         TAP_LOG(DEBUG, "Closing %s Ethernet device on numa %u",
1214                 tuntap_types[internals->type], rte_socket_id());
1215
1216         rte_intr_instance_free(internals->intr_handle);
1217
1218         if (internals->ioctl_sock != -1) {
1219                 close(internals->ioctl_sock);
1220                 internals->ioctl_sock = -1;
1221         }
1222         rte_free(dev->process_private);
1223         if (tap_devices_count == 1)
1224                 rte_mp_action_unregister(TAP_MP_KEY);
1225         tap_devices_count--;
1226         /*
1227          * Since TUN device has no more opened file descriptors
1228          * it will be removed from kernel
1229          */
1230
1231         return 0;
1232 }
1233
1234 static void
1235 tap_rx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
1236 {
1237         struct rx_queue *rxq = dev->data->rx_queues[qid];
1238         struct pmd_process_private *process_private;
1239
1240         if (!rxq)
1241                 return;
1242         process_private = rte_eth_devices[rxq->in_port].process_private;
1243         if (process_private->rxq_fds[rxq->queue_id] != -1) {
1244                 close(process_private->rxq_fds[rxq->queue_id]);
1245                 process_private->rxq_fds[rxq->queue_id] = -1;
1246                 tap_rxq_pool_free(rxq->pool);
1247                 rte_free(rxq->iovecs);
1248                 rxq->pool = NULL;
1249                 rxq->iovecs = NULL;
1250         }
1251 }
1252
1253 static void
1254 tap_tx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
1255 {
1256         struct tx_queue *txq = dev->data->tx_queues[qid];
1257         struct pmd_process_private *process_private;
1258
1259         if (!txq)
1260                 return;
1261         process_private = rte_eth_devices[txq->out_port].process_private;
1262
1263         if (process_private->txq_fds[txq->queue_id] != -1) {
1264                 close(process_private->txq_fds[txq->queue_id]);
1265                 process_private->txq_fds[txq->queue_id] = -1;
1266         }
1267 }
1268
1269 static int
1270 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1271 {
1272         struct rte_eth_link *dev_link = &dev->data->dev_link;
1273         struct pmd_internals *pmd = dev->data->dev_private;
1274         struct ifreq ifr = { .ifr_flags = 0 };
1275
1276         if (pmd->remote_if_index) {
1277                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
1278                 if (!(ifr.ifr_flags & IFF_UP) ||
1279                     !(ifr.ifr_flags & IFF_RUNNING)) {
1280                         dev_link->link_status = RTE_ETH_LINK_DOWN;
1281                         return 0;
1282                 }
1283         }
1284         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
1285         dev_link->link_status =
1286                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
1287                  RTE_ETH_LINK_UP :
1288                  RTE_ETH_LINK_DOWN);
1289         return 0;
1290 }
1291
1292 static int
1293 tap_promisc_enable(struct rte_eth_dev *dev)
1294 {
1295         struct pmd_internals *pmd = dev->data->dev_private;
1296         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1297         int ret;
1298
1299         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1300         if (ret != 0)
1301                 return ret;
1302
1303         if (pmd->remote_if_index && !pmd->flow_isolate) {
1304                 dev->data->promiscuous = 1;
1305                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
1306                 if (ret != 0) {
1307                         /* Rollback promisc flag */
1308                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1309                         /*
1310                          * rte_eth_dev_promiscuous_enable() rollback
1311                          * dev->data->promiscuous in the case of failure.
1312                          */
1313                         return ret;
1314                 }
1315         }
1316
1317         return 0;
1318 }
1319
1320 static int
1321 tap_promisc_disable(struct rte_eth_dev *dev)
1322 {
1323         struct pmd_internals *pmd = dev->data->dev_private;
1324         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1325         int ret;
1326
1327         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1328         if (ret != 0)
1329                 return ret;
1330
1331         if (pmd->remote_if_index && !pmd->flow_isolate) {
1332                 dev->data->promiscuous = 0;
1333                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
1334                 if (ret != 0) {
1335                         /* Rollback promisc flag */
1336                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1337                         /*
1338                          * rte_eth_dev_promiscuous_disable() rollback
1339                          * dev->data->promiscuous in the case of failure.
1340                          */
1341                         return ret;
1342                 }
1343         }
1344
1345         return 0;
1346 }
1347
1348 static int
1349 tap_allmulti_enable(struct rte_eth_dev *dev)
1350 {
1351         struct pmd_internals *pmd = dev->data->dev_private;
1352         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1353         int ret;
1354
1355         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1356         if (ret != 0)
1357                 return ret;
1358
1359         if (pmd->remote_if_index && !pmd->flow_isolate) {
1360                 dev->data->all_multicast = 1;
1361                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
1362                 if (ret != 0) {
1363                         /* Rollback allmulti flag */
1364                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1365                         /*
1366                          * rte_eth_dev_allmulticast_enable() rollback
1367                          * dev->data->all_multicast in the case of failure.
1368                          */
1369                         return ret;
1370                 }
1371         }
1372
1373         return 0;
1374 }
1375
1376 static int
1377 tap_allmulti_disable(struct rte_eth_dev *dev)
1378 {
1379         struct pmd_internals *pmd = dev->data->dev_private;
1380         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1381         int ret;
1382
1383         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1384         if (ret != 0)
1385                 return ret;
1386
1387         if (pmd->remote_if_index && !pmd->flow_isolate) {
1388                 dev->data->all_multicast = 0;
1389                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
1390                 if (ret != 0) {
1391                         /* Rollback allmulti flag */
1392                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1393                         /*
1394                          * rte_eth_dev_allmulticast_disable() rollback
1395                          * dev->data->all_multicast in the case of failure.
1396                          */
1397                         return ret;
1398                 }
1399         }
1400
1401         return 0;
1402 }
1403
1404 static int
1405 tap_mac_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1406 {
1407         struct pmd_internals *pmd = dev->data->dev_private;
1408         enum ioctl_mode mode = LOCAL_ONLY;
1409         struct ifreq ifr;
1410         int ret;
1411
1412         if (pmd->type == ETH_TUNTAP_TYPE_TUN) {
1413                 TAP_LOG(ERR, "%s: can't MAC address for TUN",
1414                         dev->device->name);
1415                 return -ENOTSUP;
1416         }
1417
1418         if (rte_is_zero_ether_addr(mac_addr)) {
1419                 TAP_LOG(ERR, "%s: can't set an empty MAC address",
1420                         dev->device->name);
1421                 return -EINVAL;
1422         }
1423         /* Check the actual current MAC address on the tap netdevice */
1424         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
1425         if (ret < 0)
1426                 return ret;
1427         if (rte_is_same_ether_addr(
1428                         (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1429                         mac_addr))
1430                 return 0;
1431         /* Check the current MAC address on the remote */
1432         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
1433         if (ret < 0)
1434                 return ret;
1435         if (!rte_is_same_ether_addr(
1436                         (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1437                         mac_addr))
1438                 mode = LOCAL_AND_REMOTE;
1439         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1440         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, RTE_ETHER_ADDR_LEN);
1441         ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
1442         if (ret < 0)
1443                 return ret;
1444         rte_memcpy(&pmd->eth_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1445         if (pmd->remote_if_index && !pmd->flow_isolate) {
1446                 /* Replace MAC redirection rule after a MAC change */
1447                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
1448                 if (ret < 0) {
1449                         TAP_LOG(ERR,
1450                                 "%s: Couldn't delete MAC redirection rule",
1451                                 dev->device->name);
1452                         return ret;
1453                 }
1454                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1455                 if (ret < 0) {
1456                         TAP_LOG(ERR,
1457                                 "%s: Couldn't add MAC redirection rule",
1458                                 dev->device->name);
1459                         return ret;
1460                 }
1461         }
1462
1463         return 0;
1464 }
1465
1466 static int
1467 tap_gso_ctx_setup(struct rte_gso_ctx *gso_ctx, struct rte_eth_dev *dev)
1468 {
1469         uint32_t gso_types;
1470         char pool_name[64];
1471         struct pmd_internals *pmd = dev->data->dev_private;
1472         int ret;
1473
1474         /* initialize GSO context */
1475         gso_types = RTE_ETH_TX_OFFLOAD_TCP_TSO;
1476         if (!pmd->gso_ctx_mp) {
1477                 /*
1478                  * Create private mbuf pool with TAP_GSO_MBUF_SEG_SIZE
1479                  * bytes size per mbuf use this pool for both direct and
1480                  * indirect mbufs
1481                  */
1482                 ret = snprintf(pool_name, sizeof(pool_name), "mp_%s",
1483                                 dev->device->name);
1484                 if (ret < 0 || ret >= (int)sizeof(pool_name)) {
1485                         TAP_LOG(ERR,
1486                                 "%s: failed to create mbuf pool name for device %s,"
1487                                 "device name too long or output error, ret: %d\n",
1488                                 pmd->name, dev->device->name, ret);
1489                         return -ENAMETOOLONG;
1490                 }
1491                 pmd->gso_ctx_mp = rte_pktmbuf_pool_create(pool_name,
1492                         TAP_GSO_MBUFS_NUM, TAP_GSO_MBUF_CACHE_SIZE, 0,
1493                         RTE_PKTMBUF_HEADROOM + TAP_GSO_MBUF_SEG_SIZE,
1494                         SOCKET_ID_ANY);
1495                 if (!pmd->gso_ctx_mp) {
1496                         TAP_LOG(ERR,
1497                                 "%s: failed to create mbuf pool for device %s\n",
1498                                 pmd->name, dev->device->name);
1499                         return -1;
1500                 }
1501         }
1502
1503         gso_ctx->direct_pool = pmd->gso_ctx_mp;
1504         gso_ctx->indirect_pool = pmd->gso_ctx_mp;
1505         gso_ctx->gso_types = gso_types;
1506         gso_ctx->gso_size = 0; /* gso_size is set in tx_burst() per packet */
1507         gso_ctx->flag = 0;
1508
1509         return 0;
1510 }
1511
1512 static int
1513 tap_setup_queue(struct rte_eth_dev *dev,
1514                 struct pmd_internals *internals,
1515                 uint16_t qid,
1516                 int is_rx)
1517 {
1518         int ret;
1519         int *fd;
1520         int *other_fd;
1521         const char *dir;
1522         struct pmd_internals *pmd = dev->data->dev_private;
1523         struct pmd_process_private *process_private = dev->process_private;
1524         struct rx_queue *rx = &internals->rxq[qid];
1525         struct tx_queue *tx = &internals->txq[qid];
1526         struct rte_gso_ctx *gso_ctx;
1527
1528         if (is_rx) {
1529                 fd = &process_private->rxq_fds[qid];
1530                 other_fd = &process_private->txq_fds[qid];
1531                 dir = "rx";
1532                 gso_ctx = NULL;
1533         } else {
1534                 fd = &process_private->txq_fds[qid];
1535                 other_fd = &process_private->rxq_fds[qid];
1536                 dir = "tx";
1537                 gso_ctx = &tx->gso_ctx;
1538         }
1539         if (*fd != -1) {
1540                 /* fd for this queue already exists */
1541                 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1542                         pmd->name, *fd, dir, qid);
1543                 gso_ctx = NULL;
1544         } else if (*other_fd != -1) {
1545                 /* Only other_fd exists. dup it */
1546                 *fd = dup(*other_fd);
1547                 if (*fd < 0) {
1548                         *fd = -1;
1549                         TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1550                         return -1;
1551                 }
1552                 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1553                         pmd->name, *other_fd, dir, qid, *fd);
1554         } else {
1555                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1556                 *fd = tun_alloc(pmd, 0);
1557                 if (*fd < 0) {
1558                         *fd = -1; /* restore original value */
1559                         TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1560                         return -1;
1561                 }
1562                 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1563                         pmd->name, dir, qid, *fd);
1564         }
1565
1566         tx->mtu = &dev->data->mtu;
1567         rx->rxmode = &dev->data->dev_conf.rxmode;
1568         if (gso_ctx) {
1569                 ret = tap_gso_ctx_setup(gso_ctx, dev);
1570                 if (ret)
1571                         return -1;
1572         }
1573
1574         tx->type = pmd->type;
1575
1576         return *fd;
1577 }
1578
1579 static int
1580 tap_rx_queue_setup(struct rte_eth_dev *dev,
1581                    uint16_t rx_queue_id,
1582                    uint16_t nb_rx_desc,
1583                    unsigned int socket_id,
1584                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1585                    struct rte_mempool *mp)
1586 {
1587         struct pmd_internals *internals = dev->data->dev_private;
1588         struct pmd_process_private *process_private = dev->process_private;
1589         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1590         struct rte_mbuf **tmp = &rxq->pool;
1591         long iov_max = sysconf(_SC_IOV_MAX);
1592
1593         if (iov_max <= 0) {
1594                 TAP_LOG(WARNING,
1595                         "_SC_IOV_MAX is not defined. Using %d as default",
1596                         TAP_IOV_DEFAULT_MAX);
1597                 iov_max = TAP_IOV_DEFAULT_MAX;
1598         }
1599         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1600         struct iovec (*iovecs)[nb_desc + 1];
1601         int data_off = RTE_PKTMBUF_HEADROOM;
1602         int ret = 0;
1603         int fd;
1604         int i;
1605
1606         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1607                 TAP_LOG(WARNING,
1608                         "nb_rx_queues %d too small or mempool NULL",
1609                         dev->data->nb_rx_queues);
1610                 return -1;
1611         }
1612
1613         rxq->mp = mp;
1614         rxq->trigger_seen = 1; /* force initial burst */
1615         rxq->in_port = dev->data->port_id;
1616         rxq->queue_id = rx_queue_id;
1617         rxq->nb_rx_desc = nb_desc;
1618         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1619                                     socket_id);
1620         if (!iovecs) {
1621                 TAP_LOG(WARNING,
1622                         "%s: Couldn't allocate %d RX descriptors",
1623                         dev->device->name, nb_desc);
1624                 return -ENOMEM;
1625         }
1626         rxq->iovecs = iovecs;
1627
1628         dev->data->rx_queues[rx_queue_id] = rxq;
1629         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1630         if (fd == -1) {
1631                 ret = fd;
1632                 goto error;
1633         }
1634
1635         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1636         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1637
1638         for (i = 1; i <= nb_desc; i++) {
1639                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1640                 if (!*tmp) {
1641                         TAP_LOG(WARNING,
1642                                 "%s: couldn't allocate memory for queue %d",
1643                                 dev->device->name, rx_queue_id);
1644                         ret = -ENOMEM;
1645                         goto error;
1646                 }
1647                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1648                 (*rxq->iovecs)[i].iov_base =
1649                         (char *)(*tmp)->buf_addr + data_off;
1650                 data_off = 0;
1651                 tmp = &(*tmp)->next;
1652         }
1653
1654         TAP_LOG(DEBUG, "  RX TUNTAP device name %s, qid %d on fd %d",
1655                 internals->name, rx_queue_id,
1656                 process_private->rxq_fds[rx_queue_id]);
1657
1658         return 0;
1659
1660 error:
1661         tap_rxq_pool_free(rxq->pool);
1662         rxq->pool = NULL;
1663         rte_free(rxq->iovecs);
1664         rxq->iovecs = NULL;
1665         return ret;
1666 }
1667
1668 static int
1669 tap_tx_queue_setup(struct rte_eth_dev *dev,
1670                    uint16_t tx_queue_id,
1671                    uint16_t nb_tx_desc __rte_unused,
1672                    unsigned int socket_id __rte_unused,
1673                    const struct rte_eth_txconf *tx_conf)
1674 {
1675         struct pmd_internals *internals = dev->data->dev_private;
1676         struct pmd_process_private *process_private = dev->process_private;
1677         struct tx_queue *txq;
1678         int ret;
1679         uint64_t offloads;
1680
1681         if (tx_queue_id >= dev->data->nb_tx_queues)
1682                 return -1;
1683         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1684         txq = dev->data->tx_queues[tx_queue_id];
1685         txq->out_port = dev->data->port_id;
1686         txq->queue_id = tx_queue_id;
1687
1688         offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1689         txq->csum = !!(offloads &
1690                         (RTE_ETH_TX_OFFLOAD_IPV4_CKSUM |
1691                          RTE_ETH_TX_OFFLOAD_UDP_CKSUM |
1692                          RTE_ETH_TX_OFFLOAD_TCP_CKSUM));
1693
1694         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1695         if (ret == -1)
1696                 return -1;
1697         TAP_LOG(DEBUG,
1698                 "  TX TUNTAP device name %s, qid %d on fd %d csum %s",
1699                 internals->name, tx_queue_id,
1700                 process_private->txq_fds[tx_queue_id],
1701                 txq->csum ? "on" : "off");
1702
1703         return 0;
1704 }
1705
1706 static int
1707 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1708 {
1709         struct pmd_internals *pmd = dev->data->dev_private;
1710         struct ifreq ifr = { .ifr_mtu = mtu };
1711
1712         return tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1713 }
1714
1715 static int
1716 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1717                      struct rte_ether_addr *mc_addr_set __rte_unused,
1718                      uint32_t nb_mc_addr __rte_unused)
1719 {
1720         /*
1721          * Nothing to do actually: the tap has no filtering whatsoever, every
1722          * packet is received.
1723          */
1724         return 0;
1725 }
1726
1727 static int
1728 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1729 {
1730         struct rte_eth_dev *dev = arg;
1731         struct pmd_internals *pmd = dev->data->dev_private;
1732         struct ifinfomsg *info = NLMSG_DATA(nh);
1733
1734         if (nh->nlmsg_type != RTM_NEWLINK ||
1735             (info->ifi_index != pmd->if_index &&
1736              info->ifi_index != pmd->remote_if_index))
1737                 return 0;
1738         return tap_link_update(dev, 0);
1739 }
1740
1741 static void
1742 tap_dev_intr_handler(void *cb_arg)
1743 {
1744         struct rte_eth_dev *dev = cb_arg;
1745         struct pmd_internals *pmd = dev->data->dev_private;
1746
1747         if (rte_intr_fd_get(pmd->intr_handle) >= 0)
1748                 tap_nl_recv(rte_intr_fd_get(pmd->intr_handle),
1749                             tap_nl_msg_handler, dev);
1750 }
1751
1752 static int
1753 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1754 {
1755         struct pmd_internals *pmd = dev->data->dev_private;
1756         int ret;
1757
1758         /* In any case, disable interrupt if the conf is no longer there. */
1759         if (!dev->data->dev_conf.intr_conf.lsc) {
1760                 if (rte_intr_fd_get(pmd->intr_handle) != -1)
1761                         goto clean;
1762
1763                 return 0;
1764         }
1765         if (set) {
1766                 rte_intr_fd_set(pmd->intr_handle, tap_nl_init(RTMGRP_LINK));
1767                 if (unlikely(rte_intr_fd_get(pmd->intr_handle) == -1))
1768                         return -EBADF;
1769                 return rte_intr_callback_register(
1770                         pmd->intr_handle, tap_dev_intr_handler, dev);
1771         }
1772
1773 clean:
1774         do {
1775                 ret = rte_intr_callback_unregister(pmd->intr_handle,
1776                         tap_dev_intr_handler, dev);
1777                 if (ret >= 0) {
1778                         break;
1779                 } else if (ret == -EAGAIN) {
1780                         rte_delay_ms(100);
1781                 } else {
1782                         TAP_LOG(ERR, "intr callback unregister failed: %d",
1783                                      ret);
1784                         break;
1785                 }
1786         } while (true);
1787
1788         if (rte_intr_fd_get(pmd->intr_handle) >= 0) {
1789                 tap_nl_final(rte_intr_fd_get(pmd->intr_handle));
1790                 rte_intr_fd_set(pmd->intr_handle, -1);
1791         }
1792
1793         return 0;
1794 }
1795
1796 static int
1797 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1798 {
1799         int err;
1800
1801         err = tap_lsc_intr_handle_set(dev, set);
1802         if (err < 0) {
1803                 if (!set)
1804                         tap_rx_intr_vec_set(dev, 0);
1805                 return err;
1806         }
1807         err = tap_rx_intr_vec_set(dev, set);
1808         if (err && set)
1809                 tap_lsc_intr_handle_set(dev, 0);
1810         return err;
1811 }
1812
1813 static const uint32_t*
1814 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1815 {
1816         static const uint32_t ptypes[] = {
1817                 RTE_PTYPE_INNER_L2_ETHER,
1818                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1819                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1820                 RTE_PTYPE_INNER_L3_IPV4,
1821                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1822                 RTE_PTYPE_INNER_L3_IPV6,
1823                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1824                 RTE_PTYPE_INNER_L4_FRAG,
1825                 RTE_PTYPE_INNER_L4_UDP,
1826                 RTE_PTYPE_INNER_L4_TCP,
1827                 RTE_PTYPE_INNER_L4_SCTP,
1828                 RTE_PTYPE_L2_ETHER,
1829                 RTE_PTYPE_L2_ETHER_VLAN,
1830                 RTE_PTYPE_L2_ETHER_QINQ,
1831                 RTE_PTYPE_L3_IPV4,
1832                 RTE_PTYPE_L3_IPV4_EXT,
1833                 RTE_PTYPE_L3_IPV6_EXT,
1834                 RTE_PTYPE_L3_IPV6,
1835                 RTE_PTYPE_L4_FRAG,
1836                 RTE_PTYPE_L4_UDP,
1837                 RTE_PTYPE_L4_TCP,
1838                 RTE_PTYPE_L4_SCTP,
1839         };
1840
1841         return ptypes;
1842 }
1843
1844 static int
1845 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1846                   struct rte_eth_fc_conf *fc_conf)
1847 {
1848         fc_conf->mode = RTE_ETH_FC_NONE;
1849         return 0;
1850 }
1851
1852 static int
1853 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1854                   struct rte_eth_fc_conf *fc_conf)
1855 {
1856         if (fc_conf->mode != RTE_ETH_FC_NONE)
1857                 return -ENOTSUP;
1858         return 0;
1859 }
1860
1861 /**
1862  * DPDK callback to update the RSS hash configuration.
1863  *
1864  * @param dev
1865  *   Pointer to Ethernet device structure.
1866  * @param[in] rss_conf
1867  *   RSS configuration data.
1868  *
1869  * @return
1870  *   0 on success, a negative errno value otherwise and rte_errno is set.
1871  */
1872 static int
1873 tap_rss_hash_update(struct rte_eth_dev *dev,
1874                 struct rte_eth_rss_conf *rss_conf)
1875 {
1876         if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1877                 rte_errno = EINVAL;
1878                 return -rte_errno;
1879         }
1880         if (rss_conf->rss_key && rss_conf->rss_key_len) {
1881                 /*
1882                  * Currently TAP RSS key is hard coded
1883                  * and cannot be updated
1884                  */
1885                 TAP_LOG(ERR,
1886                         "port %u RSS key cannot be updated",
1887                         dev->data->port_id);
1888                 rte_errno = EINVAL;
1889                 return -rte_errno;
1890         }
1891         return 0;
1892 }
1893
1894 static int
1895 tap_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1896 {
1897         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1898
1899         return 0;
1900 }
1901
1902 static int
1903 tap_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1904 {
1905         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1906
1907         return 0;
1908 }
1909
1910 static int
1911 tap_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1912 {
1913         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1914
1915         return 0;
1916 }
1917
1918 static int
1919 tap_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1920 {
1921         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1922
1923         return 0;
1924 }
1925 static const struct eth_dev_ops ops = {
1926         .dev_start              = tap_dev_start,
1927         .dev_stop               = tap_dev_stop,
1928         .dev_close              = tap_dev_close,
1929         .dev_configure          = tap_dev_configure,
1930         .dev_infos_get          = tap_dev_info,
1931         .rx_queue_setup         = tap_rx_queue_setup,
1932         .tx_queue_setup         = tap_tx_queue_setup,
1933         .rx_queue_start         = tap_rx_queue_start,
1934         .tx_queue_start         = tap_tx_queue_start,
1935         .rx_queue_stop          = tap_rx_queue_stop,
1936         .tx_queue_stop          = tap_tx_queue_stop,
1937         .rx_queue_release       = tap_rx_queue_release,
1938         .tx_queue_release       = tap_tx_queue_release,
1939         .flow_ctrl_get          = tap_flow_ctrl_get,
1940         .flow_ctrl_set          = tap_flow_ctrl_set,
1941         .link_update            = tap_link_update,
1942         .dev_set_link_up        = tap_link_set_up,
1943         .dev_set_link_down      = tap_link_set_down,
1944         .promiscuous_enable     = tap_promisc_enable,
1945         .promiscuous_disable    = tap_promisc_disable,
1946         .allmulticast_enable    = tap_allmulti_enable,
1947         .allmulticast_disable   = tap_allmulti_disable,
1948         .mac_addr_set           = tap_mac_set,
1949         .mtu_set                = tap_mtu_set,
1950         .set_mc_addr_list       = tap_set_mc_addr_list,
1951         .stats_get              = tap_stats_get,
1952         .stats_reset            = tap_stats_reset,
1953         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1954         .rss_hash_update        = tap_rss_hash_update,
1955         .flow_ops_get           = tap_dev_flow_ops_get,
1956 };
1957
1958 static int
1959 eth_dev_tap_create(struct rte_vdev_device *vdev, const char *tap_name,
1960                    char *remote_iface, struct rte_ether_addr *mac_addr,
1961                    enum rte_tuntap_type type)
1962 {
1963         int numa_node = rte_socket_id();
1964         struct rte_eth_dev *dev;
1965         struct pmd_internals *pmd;
1966         struct pmd_process_private *process_private;
1967         const char *tuntap_name = tuntap_types[type];
1968         struct rte_eth_dev_data *data;
1969         struct ifreq ifr;
1970         int i;
1971
1972         TAP_LOG(DEBUG, "%s device on numa %u", tuntap_name, rte_socket_id());
1973
1974         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1975         if (!dev) {
1976                 TAP_LOG(ERR, "%s Unable to allocate device struct",
1977                                 tuntap_name);
1978                 goto error_exit_nodev;
1979         }
1980
1981         process_private = (struct pmd_process_private *)
1982                 rte_zmalloc_socket(tap_name, sizeof(struct pmd_process_private),
1983                         RTE_CACHE_LINE_SIZE, dev->device->numa_node);
1984
1985         if (process_private == NULL) {
1986                 TAP_LOG(ERR, "Failed to alloc memory for process private");
1987                 return -1;
1988         }
1989         pmd = dev->data->dev_private;
1990         dev->process_private = process_private;
1991         pmd->dev = dev;
1992         strlcpy(pmd->name, tap_name, sizeof(pmd->name));
1993         pmd->type = type;
1994         pmd->ka_fd = -1;
1995         pmd->nlsk_fd = -1;
1996         pmd->gso_ctx_mp = NULL;
1997
1998         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1999         if (pmd->ioctl_sock == -1) {
2000                 TAP_LOG(ERR,
2001                         "%s Unable to get a socket for management: %s",
2002                         tuntap_name, strerror(errno));
2003                 goto error_exit;
2004         }
2005
2006         /* Allocate interrupt instance */
2007         pmd->intr_handle = rte_intr_instance_alloc(RTE_INTR_INSTANCE_F_SHARED);
2008         if (pmd->intr_handle == NULL) {
2009                 TAP_LOG(ERR, "Failed to allocate intr handle");
2010                 goto error_exit;
2011         }
2012
2013         /* Setup some default values */
2014         data = dev->data;
2015         data->dev_private = pmd;
2016         data->dev_flags = RTE_ETH_DEV_INTR_LSC |
2017                                 RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
2018         data->numa_node = numa_node;
2019
2020         data->dev_link = pmd_link;
2021         data->mac_addrs = &pmd->eth_addr;
2022         /* Set the number of RX and TX queues */
2023         data->nb_rx_queues = 0;
2024         data->nb_tx_queues = 0;
2025
2026         dev->dev_ops = &ops;
2027         dev->rx_pkt_burst = pmd_rx_burst;
2028         dev->tx_pkt_burst = pmd_tx_burst;
2029
2030         rte_intr_type_set(pmd->intr_handle, RTE_INTR_HANDLE_EXT);
2031         rte_intr_fd_set(pmd->intr_handle, -1);
2032         dev->intr_handle = pmd->intr_handle;
2033
2034         /* Presetup the fds to -1 as being not valid */
2035         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
2036                 process_private->rxq_fds[i] = -1;
2037                 process_private->txq_fds[i] = -1;
2038         }
2039
2040         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
2041                 if (rte_is_zero_ether_addr(mac_addr))
2042                         rte_eth_random_addr((uint8_t *)&pmd->eth_addr);
2043                 else
2044                         rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
2045         }
2046
2047         /*
2048          * Allocate a TUN device keep-alive file descriptor that will only be
2049          * closed when the TUN device itself is closed or removed.
2050          * This keep-alive file descriptor will guarantee that the TUN device
2051          * exists even when all of its queues are closed
2052          */
2053         pmd->ka_fd = tun_alloc(pmd, 1);
2054         if (pmd->ka_fd == -1) {
2055                 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
2056                 goto error_exit;
2057         }
2058         TAP_LOG(DEBUG, "allocated %s", pmd->name);
2059
2060         ifr.ifr_mtu = dev->data->mtu;
2061         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
2062                 goto error_exit;
2063
2064         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
2065                 memset(&ifr, 0, sizeof(struct ifreq));
2066                 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
2067                 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
2068                                 RTE_ETHER_ADDR_LEN);
2069                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
2070                         goto error_exit;
2071         }
2072
2073         /*
2074          * Set up everything related to rte_flow:
2075          * - netlink socket
2076          * - tap / remote if_index
2077          * - mandatory QDISCs
2078          * - rte_flow actual/implicit lists
2079          * - implicit rules
2080          */
2081         pmd->nlsk_fd = tap_nl_init(0);
2082         if (pmd->nlsk_fd == -1) {
2083                 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
2084                         pmd->name);
2085                 goto disable_rte_flow;
2086         }
2087         pmd->if_index = if_nametoindex(pmd->name);
2088         if (!pmd->if_index) {
2089                 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
2090                 goto disable_rte_flow;
2091         }
2092         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
2093                 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
2094                         pmd->name);
2095                 goto disable_rte_flow;
2096         }
2097         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
2098                 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
2099                         pmd->name);
2100                 goto disable_rte_flow;
2101         }
2102         LIST_INIT(&pmd->flows);
2103
2104         if (strlen(remote_iface)) {
2105                 pmd->remote_if_index = if_nametoindex(remote_iface);
2106                 if (!pmd->remote_if_index) {
2107                         TAP_LOG(ERR, "%s: failed to get %s if_index.",
2108                                 pmd->name, remote_iface);
2109                         goto error_remote;
2110                 }
2111                 strlcpy(pmd->remote_iface, remote_iface, RTE_ETH_NAME_MAX_LEN);
2112
2113                 /* Save state of remote device */
2114                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
2115
2116                 /* Replicate remote MAC address */
2117                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
2118                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
2119                                 pmd->name, pmd->remote_iface);
2120                         goto error_remote;
2121                 }
2122                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
2123                            RTE_ETHER_ADDR_LEN);
2124                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
2125                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
2126                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
2127                                 pmd->name, remote_iface);
2128                         goto error_remote;
2129                 }
2130
2131                 /*
2132                  * Flush usually returns negative value because it tries to
2133                  * delete every QDISC (and on a running device, one QDISC at
2134                  * least is needed). Ignore negative return value.
2135                  */
2136                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
2137                 if (qdisc_create_ingress(pmd->nlsk_fd,
2138                                          pmd->remote_if_index) < 0) {
2139                         TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
2140                                 pmd->remote_iface);
2141                         goto error_remote;
2142                 }
2143                 LIST_INIT(&pmd->implicit_flows);
2144                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
2145                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
2146                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
2147                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
2148                         TAP_LOG(ERR,
2149                                 "%s: failed to create implicit rules.",
2150                                 pmd->name);
2151                         goto error_remote;
2152                 }
2153         }
2154
2155         rte_eth_dev_probing_finish(dev);
2156         return 0;
2157
2158 disable_rte_flow:
2159         TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
2160                 strerror(errno), errno);
2161         if (strlen(remote_iface)) {
2162                 TAP_LOG(ERR, "Remote feature requires flow support.");
2163                 goto error_exit;
2164         }
2165         rte_eth_dev_probing_finish(dev);
2166         return 0;
2167
2168 error_remote:
2169         TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
2170                 strerror(errno), errno);
2171         tap_flow_implicit_flush(pmd, NULL);
2172
2173 error_exit:
2174         if (pmd->nlsk_fd != -1)
2175                 close(pmd->nlsk_fd);
2176         if (pmd->ka_fd != -1)
2177                 close(pmd->ka_fd);
2178         if (pmd->ioctl_sock != -1)
2179                 close(pmd->ioctl_sock);
2180         /* mac_addrs must not be freed alone because part of dev_private */
2181         dev->data->mac_addrs = NULL;
2182         rte_intr_instance_free(pmd->intr_handle);
2183         rte_eth_dev_release_port(dev);
2184
2185 error_exit_nodev:
2186         TAP_LOG(ERR, "%s Unable to initialize %s",
2187                 tuntap_name, rte_vdev_device_name(vdev));
2188
2189         return -EINVAL;
2190 }
2191
2192 /* make sure name is a possible Linux network device name */
2193 static bool
2194 is_valid_iface(const char *name)
2195 {
2196         if (*name == '\0')
2197                 return false;
2198
2199         if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
2200                 return false;
2201
2202         while (*name) {
2203                 if (*name == '/' || *name == ':' || isspace(*name))
2204                         return false;
2205                 name++;
2206         }
2207         return true;
2208 }
2209
2210 static int
2211 set_interface_name(const char *key __rte_unused,
2212                    const char *value,
2213                    void *extra_args)
2214 {
2215         char *name = (char *)extra_args;
2216
2217         if (value) {
2218                 if (!is_valid_iface(value)) {
2219                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2220                                 value);
2221                         return -1;
2222                 }
2223                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2224         } else {
2225                 /* use tap%d which causes kernel to choose next available */
2226                 strlcpy(name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2227         }
2228         return 0;
2229 }
2230
2231 static int
2232 set_remote_iface(const char *key __rte_unused,
2233                  const char *value,
2234                  void *extra_args)
2235 {
2236         char *name = (char *)extra_args;
2237
2238         if (value) {
2239                 if (!is_valid_iface(value)) {
2240                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2241                                 value);
2242                         return -1;
2243                 }
2244                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2245         }
2246
2247         return 0;
2248 }
2249
2250 static int parse_user_mac(struct rte_ether_addr *user_mac,
2251                 const char *value)
2252 {
2253         unsigned int index = 0;
2254         char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
2255
2256         if (user_mac == NULL || value == NULL)
2257                 return 0;
2258
2259         strlcpy(mac_temp, value, sizeof(mac_temp));
2260         mac_byte = strtok(mac_temp, ":");
2261
2262         while ((mac_byte != NULL) &&
2263                         (strlen(mac_byte) <= 2) &&
2264                         (strlen(mac_byte) == strspn(mac_byte,
2265                                         ETH_TAP_CMP_MAC_FMT))) {
2266                 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
2267                 mac_byte = strtok(NULL, ":");
2268         }
2269
2270         return index;
2271 }
2272
2273 static int
2274 set_mac_type(const char *key __rte_unused,
2275              const char *value,
2276              void *extra_args)
2277 {
2278         struct rte_ether_addr *user_mac = extra_args;
2279
2280         if (!value)
2281                 return 0;
2282
2283         if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
2284                 static int iface_idx;
2285
2286                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
2287                 memcpy((char *)user_mac->addr_bytes, "\0dtap",
2288                         RTE_ETHER_ADDR_LEN);
2289                 user_mac->addr_bytes[RTE_ETHER_ADDR_LEN - 1] =
2290                         iface_idx++ + '0';
2291                 goto success;
2292         }
2293
2294         if (parse_user_mac(user_mac, value) != 6)
2295                 goto error;
2296 success:
2297         TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
2298         return 0;
2299
2300 error:
2301         TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
2302                 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
2303         return -1;
2304 }
2305
2306 /*
2307  * Open a TUN interface device. TUN PMD
2308  * 1) sets tap_type as false
2309  * 2) intakes iface as argument.
2310  * 3) as interface is virtual set speed to 10G
2311  */
2312 static int
2313 rte_pmd_tun_probe(struct rte_vdev_device *dev)
2314 {
2315         const char *name, *params;
2316         int ret;
2317         struct rte_kvargs *kvlist = NULL;
2318         char tun_name[RTE_ETH_NAME_MAX_LEN];
2319         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2320         struct rte_eth_dev *eth_dev;
2321
2322         name = rte_vdev_device_name(dev);
2323         params = rte_vdev_device_args(dev);
2324         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2325
2326         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
2327             strlen(params) == 0) {
2328                 eth_dev = rte_eth_dev_attach_secondary(name);
2329                 if (!eth_dev) {
2330                         TAP_LOG(ERR, "Failed to probe %s", name);
2331                         return -1;
2332                 }
2333                 eth_dev->dev_ops = &ops;
2334                 eth_dev->device = &dev->device;
2335                 rte_eth_dev_probing_finish(eth_dev);
2336                 return 0;
2337         }
2338
2339         /* use tun%d which causes kernel to choose next available */
2340         strlcpy(tun_name, DEFAULT_TUN_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2341
2342         if (params && (params[0] != '\0')) {
2343                 TAP_LOG(DEBUG, "parameters (%s)", params);
2344
2345                 kvlist = rte_kvargs_parse(params, valid_arguments);
2346                 if (kvlist) {
2347                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2348                                 ret = rte_kvargs_process(kvlist,
2349                                         ETH_TAP_IFACE_ARG,
2350                                         &set_interface_name,
2351                                         tun_name);
2352
2353                                 if (ret == -1)
2354                                         goto leave;
2355                         }
2356                 }
2357         }
2358         pmd_link.link_speed = RTE_ETH_SPEED_NUM_10G;
2359
2360         TAP_LOG(DEBUG, "Initializing pmd_tun for %s", name);
2361
2362         ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0,
2363                                  ETH_TUNTAP_TYPE_TUN);
2364
2365 leave:
2366         if (ret == -1) {
2367                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2368                         name, tun_name);
2369         }
2370         rte_kvargs_free(kvlist);
2371
2372         return ret;
2373 }
2374
2375 /* Request queue file descriptors from secondary to primary. */
2376 static int
2377 tap_mp_attach_queues(const char *port_name, struct rte_eth_dev *dev)
2378 {
2379         int ret;
2380         struct timespec timeout = {.tv_sec = 1, .tv_nsec = 0};
2381         struct rte_mp_msg request, *reply;
2382         struct rte_mp_reply replies;
2383         struct ipc_queues *request_param = (struct ipc_queues *)request.param;
2384         struct ipc_queues *reply_param;
2385         struct pmd_process_private *process_private = dev->process_private;
2386         int queue, fd_iterator;
2387
2388         /* Prepare the request */
2389         memset(&request, 0, sizeof(request));
2390         strlcpy(request.name, TAP_MP_KEY, sizeof(request.name));
2391         strlcpy(request_param->port_name, port_name,
2392                 sizeof(request_param->port_name));
2393         request.len_param = sizeof(*request_param);
2394         /* Send request and receive reply */
2395         ret = rte_mp_request_sync(&request, &replies, &timeout);
2396         if (ret < 0 || replies.nb_received != 1) {
2397                 TAP_LOG(ERR, "Failed to request queues from primary: %d",
2398                         rte_errno);
2399                 return -1;
2400         }
2401         reply = &replies.msgs[0];
2402         reply_param = (struct ipc_queues *)reply->param;
2403         TAP_LOG(DEBUG, "Received IPC reply for %s", reply_param->port_name);
2404
2405         /* Attach the queues from received file descriptors */
2406         if (reply_param->rxq_count + reply_param->txq_count != reply->num_fds) {
2407                 TAP_LOG(ERR, "Unexpected number of fds received");
2408                 return -1;
2409         }
2410
2411         dev->data->nb_rx_queues = reply_param->rxq_count;
2412         dev->data->nb_tx_queues = reply_param->txq_count;
2413         fd_iterator = 0;
2414         for (queue = 0; queue < reply_param->rxq_count; queue++)
2415                 process_private->rxq_fds[queue] = reply->fds[fd_iterator++];
2416         for (queue = 0; queue < reply_param->txq_count; queue++)
2417                 process_private->txq_fds[queue] = reply->fds[fd_iterator++];
2418         free(reply);
2419         return 0;
2420 }
2421
2422 /* Send the queue file descriptors from the primary process to secondary. */
2423 static int
2424 tap_mp_sync_queues(const struct rte_mp_msg *request, const void *peer)
2425 {
2426         struct rte_eth_dev *dev;
2427         struct pmd_process_private *process_private;
2428         struct rte_mp_msg reply;
2429         const struct ipc_queues *request_param =
2430                 (const struct ipc_queues *)request->param;
2431         struct ipc_queues *reply_param =
2432                 (struct ipc_queues *)reply.param;
2433         int queue;
2434
2435         /* Get requested port */
2436         TAP_LOG(DEBUG, "Received IPC request for %s", request_param->port_name);
2437         dev = rte_eth_dev_get_by_name(request_param->port_name);
2438         if (!dev) {
2439                 TAP_LOG(ERR, "Failed to get port id for %s",
2440                         request_param->port_name);
2441                 return -1;
2442         }
2443         process_private = dev->process_private;
2444
2445         /* Fill file descriptors for all queues */
2446         reply.num_fds = 0;
2447         reply_param->rxq_count = 0;
2448         if (dev->data->nb_rx_queues + dev->data->nb_tx_queues >
2449                         RTE_MP_MAX_FD_NUM){
2450                 TAP_LOG(ERR, "Number of rx/tx queues exceeds max number of fds");
2451                 return -1;
2452         }
2453
2454         for (queue = 0; queue < dev->data->nb_rx_queues; queue++) {
2455                 reply.fds[reply.num_fds++] = process_private->rxq_fds[queue];
2456                 reply_param->rxq_count++;
2457         }
2458         RTE_ASSERT(reply_param->rxq_count == dev->data->nb_rx_queues);
2459
2460         reply_param->txq_count = 0;
2461         for (queue = 0; queue < dev->data->nb_tx_queues; queue++) {
2462                 reply.fds[reply.num_fds++] = process_private->txq_fds[queue];
2463                 reply_param->txq_count++;
2464         }
2465         RTE_ASSERT(reply_param->txq_count == dev->data->nb_tx_queues);
2466
2467         /* Send reply */
2468         strlcpy(reply.name, request->name, sizeof(reply.name));
2469         strlcpy(reply_param->port_name, request_param->port_name,
2470                 sizeof(reply_param->port_name));
2471         reply.len_param = sizeof(*reply_param);
2472         if (rte_mp_reply(&reply, peer) < 0) {
2473                 TAP_LOG(ERR, "Failed to reply an IPC request to sync queues");
2474                 return -1;
2475         }
2476         return 0;
2477 }
2478
2479 /* Open a TAP interface device.
2480  */
2481 static int
2482 rte_pmd_tap_probe(struct rte_vdev_device *dev)
2483 {
2484         const char *name, *params;
2485         int ret;
2486         struct rte_kvargs *kvlist = NULL;
2487         int speed;
2488         char tap_name[RTE_ETH_NAME_MAX_LEN];
2489         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2490         struct rte_ether_addr user_mac = { .addr_bytes = {0} };
2491         struct rte_eth_dev *eth_dev;
2492         int tap_devices_count_increased = 0;
2493
2494         name = rte_vdev_device_name(dev);
2495         params = rte_vdev_device_args(dev);
2496
2497         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
2498                 eth_dev = rte_eth_dev_attach_secondary(name);
2499                 if (!eth_dev) {
2500                         TAP_LOG(ERR, "Failed to probe %s", name);
2501                         return -1;
2502                 }
2503                 eth_dev->dev_ops = &ops;
2504                 eth_dev->device = &dev->device;
2505                 eth_dev->rx_pkt_burst = pmd_rx_burst;
2506                 eth_dev->tx_pkt_burst = pmd_tx_burst;
2507                 if (!rte_eal_primary_proc_alive(NULL)) {
2508                         TAP_LOG(ERR, "Primary process is missing");
2509                         return -1;
2510                 }
2511                 eth_dev->process_private = (struct pmd_process_private *)
2512                         rte_zmalloc_socket(name,
2513                                 sizeof(struct pmd_process_private),
2514                                 RTE_CACHE_LINE_SIZE,
2515                                 eth_dev->device->numa_node);
2516                 if (eth_dev->process_private == NULL) {
2517                         TAP_LOG(ERR,
2518                                 "Failed to alloc memory for process private");
2519                         return -1;
2520                 }
2521
2522                 ret = tap_mp_attach_queues(name, eth_dev);
2523                 if (ret != 0)
2524                         return -1;
2525
2526                 if (!tap_devices_count) {
2527                         ret = rte_mp_action_register(TAP_MP_REQ_START_RXTX, tap_mp_req_start_rxtx);
2528                         if (ret < 0 && rte_errno != ENOTSUP) {
2529                                 TAP_LOG(ERR, "tap: Failed to register IPC callback: %s",
2530                                         strerror(rte_errno));
2531                                 return -1;
2532                         }
2533                 }
2534                 tap_devices_count++;
2535                 rte_eth_dev_probing_finish(eth_dev);
2536                 return 0;
2537         }
2538
2539         speed = RTE_ETH_SPEED_NUM_10G;
2540
2541         /* use tap%d which causes kernel to choose next available */
2542         strlcpy(tap_name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2543         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2544
2545         if (params && (params[0] != '\0')) {
2546                 TAP_LOG(DEBUG, "parameters (%s)", params);
2547
2548                 kvlist = rte_kvargs_parse(params, valid_arguments);
2549                 if (kvlist) {
2550                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2551                                 ret = rte_kvargs_process(kvlist,
2552                                                          ETH_TAP_IFACE_ARG,
2553                                                          &set_interface_name,
2554                                                          tap_name);
2555                                 if (ret == -1)
2556                                         goto leave;
2557                         }
2558
2559                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
2560                                 ret = rte_kvargs_process(kvlist,
2561                                                          ETH_TAP_REMOTE_ARG,
2562                                                          &set_remote_iface,
2563                                                          remote_iface);
2564                                 if (ret == -1)
2565                                         goto leave;
2566                         }
2567
2568                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
2569                                 ret = rte_kvargs_process(kvlist,
2570                                                          ETH_TAP_MAC_ARG,
2571                                                          &set_mac_type,
2572                                                          &user_mac);
2573                                 if (ret == -1)
2574                                         goto leave;
2575                         }
2576                 }
2577         }
2578         pmd_link.link_speed = speed;
2579
2580         TAP_LOG(DEBUG, "Initializing pmd_tap for %s", name);
2581
2582         /* Register IPC feed callback */
2583         if (!tap_devices_count) {
2584                 ret = rte_mp_action_register(TAP_MP_KEY, tap_mp_sync_queues);
2585                 if (ret < 0 && rte_errno != ENOTSUP) {
2586                         TAP_LOG(ERR, "tap: Failed to register IPC callback: %s",
2587                                 strerror(rte_errno));
2588                         goto leave;
2589                 }
2590         }
2591         tap_devices_count++;
2592         tap_devices_count_increased = 1;
2593         ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
2594                 ETH_TUNTAP_TYPE_TAP);
2595
2596 leave:
2597         if (ret == -1) {
2598                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2599                         name, tap_name);
2600                 if (tap_devices_count_increased == 1) {
2601                         if (tap_devices_count == 1)
2602                                 rte_mp_action_unregister(TAP_MP_KEY);
2603                         tap_devices_count--;
2604                 }
2605         }
2606         rte_kvargs_free(kvlist);
2607
2608         return ret;
2609 }
2610
2611 /* detach a TUNTAP device.
2612  */
2613 static int
2614 rte_pmd_tap_remove(struct rte_vdev_device *dev)
2615 {
2616         struct rte_eth_dev *eth_dev = NULL;
2617
2618         /* find the ethdev entry */
2619         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
2620         if (!eth_dev)
2621                 return 0;
2622
2623         tap_dev_close(eth_dev);
2624         rte_eth_dev_release_port(eth_dev);
2625
2626         return 0;
2627 }
2628
2629 static struct rte_vdev_driver pmd_tun_drv = {
2630         .probe = rte_pmd_tun_probe,
2631         .remove = rte_pmd_tap_remove,
2632 };
2633
2634 static struct rte_vdev_driver pmd_tap_drv = {
2635         .probe = rte_pmd_tap_probe,
2636         .remove = rte_pmd_tap_remove,
2637 };
2638
2639 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
2640 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
2641 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
2642 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
2643                               ETH_TAP_IFACE_ARG "=<string> ");
2644 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
2645                               ETH_TAP_IFACE_ARG "=<string> "
2646                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
2647                               ETH_TAP_REMOTE_ARG "=<string>");
2648 RTE_LOG_REGISTER_DEFAULT(tap_logtype, NOTICE);