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         if (internals->ioctl_sock != -1) {
1217                 close(internals->ioctl_sock);
1218                 internals->ioctl_sock = -1;
1219         }
1220         rte_free(dev->process_private);
1221         if (tap_devices_count == 1)
1222                 rte_mp_action_unregister(TAP_MP_KEY);
1223         tap_devices_count--;
1224         /*
1225          * Since TUN device has no more opened file descriptors
1226          * it will be removed from kernel
1227          */
1228
1229         return 0;
1230 }
1231
1232 static void
1233 tap_rx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
1234 {
1235         struct rx_queue *rxq = dev->data->rx_queues[qid];
1236         struct pmd_process_private *process_private;
1237
1238         if (!rxq)
1239                 return;
1240         process_private = rte_eth_devices[rxq->in_port].process_private;
1241         if (process_private->rxq_fds[rxq->queue_id] != -1) {
1242                 close(process_private->rxq_fds[rxq->queue_id]);
1243                 process_private->rxq_fds[rxq->queue_id] = -1;
1244                 tap_rxq_pool_free(rxq->pool);
1245                 rte_free(rxq->iovecs);
1246                 rxq->pool = NULL;
1247                 rxq->iovecs = NULL;
1248         }
1249 }
1250
1251 static void
1252 tap_tx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
1253 {
1254         struct tx_queue *txq = dev->data->tx_queues[qid];
1255         struct pmd_process_private *process_private;
1256
1257         if (!txq)
1258                 return;
1259         process_private = rte_eth_devices[txq->out_port].process_private;
1260
1261         if (process_private->txq_fds[txq->queue_id] != -1) {
1262                 close(process_private->txq_fds[txq->queue_id]);
1263                 process_private->txq_fds[txq->queue_id] = -1;
1264         }
1265 }
1266
1267 static int
1268 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1269 {
1270         struct rte_eth_link *dev_link = &dev->data->dev_link;
1271         struct pmd_internals *pmd = dev->data->dev_private;
1272         struct ifreq ifr = { .ifr_flags = 0 };
1273
1274         if (pmd->remote_if_index) {
1275                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
1276                 if (!(ifr.ifr_flags & IFF_UP) ||
1277                     !(ifr.ifr_flags & IFF_RUNNING)) {
1278                         dev_link->link_status = RTE_ETH_LINK_DOWN;
1279                         return 0;
1280                 }
1281         }
1282         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
1283         dev_link->link_status =
1284                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
1285                  RTE_ETH_LINK_UP :
1286                  RTE_ETH_LINK_DOWN);
1287         return 0;
1288 }
1289
1290 static int
1291 tap_promisc_enable(struct rte_eth_dev *dev)
1292 {
1293         struct pmd_internals *pmd = dev->data->dev_private;
1294         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1295         int ret;
1296
1297         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1298         if (ret != 0)
1299                 return ret;
1300
1301         if (pmd->remote_if_index && !pmd->flow_isolate) {
1302                 dev->data->promiscuous = 1;
1303                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
1304                 if (ret != 0) {
1305                         /* Rollback promisc flag */
1306                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1307                         /*
1308                          * rte_eth_dev_promiscuous_enable() rollback
1309                          * dev->data->promiscuous in the case of failure.
1310                          */
1311                         return ret;
1312                 }
1313         }
1314
1315         return 0;
1316 }
1317
1318 static int
1319 tap_promisc_disable(struct rte_eth_dev *dev)
1320 {
1321         struct pmd_internals *pmd = dev->data->dev_private;
1322         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1323         int ret;
1324
1325         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1326         if (ret != 0)
1327                 return ret;
1328
1329         if (pmd->remote_if_index && !pmd->flow_isolate) {
1330                 dev->data->promiscuous = 0;
1331                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
1332                 if (ret != 0) {
1333                         /* Rollback promisc flag */
1334                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1335                         /*
1336                          * rte_eth_dev_promiscuous_disable() rollback
1337                          * dev->data->promiscuous in the case of failure.
1338                          */
1339                         return ret;
1340                 }
1341         }
1342
1343         return 0;
1344 }
1345
1346 static int
1347 tap_allmulti_enable(struct rte_eth_dev *dev)
1348 {
1349         struct pmd_internals *pmd = dev->data->dev_private;
1350         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1351         int ret;
1352
1353         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1354         if (ret != 0)
1355                 return ret;
1356
1357         if (pmd->remote_if_index && !pmd->flow_isolate) {
1358                 dev->data->all_multicast = 1;
1359                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
1360                 if (ret != 0) {
1361                         /* Rollback allmulti flag */
1362                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1363                         /*
1364                          * rte_eth_dev_allmulticast_enable() rollback
1365                          * dev->data->all_multicast in the case of failure.
1366                          */
1367                         return ret;
1368                 }
1369         }
1370
1371         return 0;
1372 }
1373
1374 static int
1375 tap_allmulti_disable(struct rte_eth_dev *dev)
1376 {
1377         struct pmd_internals *pmd = dev->data->dev_private;
1378         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1379         int ret;
1380
1381         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1382         if (ret != 0)
1383                 return ret;
1384
1385         if (pmd->remote_if_index && !pmd->flow_isolate) {
1386                 dev->data->all_multicast = 0;
1387                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
1388                 if (ret != 0) {
1389                         /* Rollback allmulti flag */
1390                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1391                         /*
1392                          * rte_eth_dev_allmulticast_disable() rollback
1393                          * dev->data->all_multicast in the case of failure.
1394                          */
1395                         return ret;
1396                 }
1397         }
1398
1399         return 0;
1400 }
1401
1402 static int
1403 tap_mac_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1404 {
1405         struct pmd_internals *pmd = dev->data->dev_private;
1406         enum ioctl_mode mode = LOCAL_ONLY;
1407         struct ifreq ifr;
1408         int ret;
1409
1410         if (pmd->type == ETH_TUNTAP_TYPE_TUN) {
1411                 TAP_LOG(ERR, "%s: can't MAC address for TUN",
1412                         dev->device->name);
1413                 return -ENOTSUP;
1414         }
1415
1416         if (rte_is_zero_ether_addr(mac_addr)) {
1417                 TAP_LOG(ERR, "%s: can't set an empty MAC address",
1418                         dev->device->name);
1419                 return -EINVAL;
1420         }
1421         /* Check the actual current MAC address on the tap netdevice */
1422         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
1423         if (ret < 0)
1424                 return ret;
1425         if (rte_is_same_ether_addr(
1426                         (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1427                         mac_addr))
1428                 return 0;
1429         /* Check the current MAC address on the remote */
1430         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
1431         if (ret < 0)
1432                 return ret;
1433         if (!rte_is_same_ether_addr(
1434                         (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1435                         mac_addr))
1436                 mode = LOCAL_AND_REMOTE;
1437         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1438         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, RTE_ETHER_ADDR_LEN);
1439         ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
1440         if (ret < 0)
1441                 return ret;
1442         rte_memcpy(&pmd->eth_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1443         if (pmd->remote_if_index && !pmd->flow_isolate) {
1444                 /* Replace MAC redirection rule after a MAC change */
1445                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
1446                 if (ret < 0) {
1447                         TAP_LOG(ERR,
1448                                 "%s: Couldn't delete MAC redirection rule",
1449                                 dev->device->name);
1450                         return ret;
1451                 }
1452                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1453                 if (ret < 0) {
1454                         TAP_LOG(ERR,
1455                                 "%s: Couldn't add MAC redirection rule",
1456                                 dev->device->name);
1457                         return ret;
1458                 }
1459         }
1460
1461         return 0;
1462 }
1463
1464 static int
1465 tap_gso_ctx_setup(struct rte_gso_ctx *gso_ctx, struct rte_eth_dev *dev)
1466 {
1467         uint32_t gso_types;
1468         char pool_name[64];
1469         struct pmd_internals *pmd = dev->data->dev_private;
1470         int ret;
1471
1472         /* initialize GSO context */
1473         gso_types = RTE_ETH_TX_OFFLOAD_TCP_TSO;
1474         if (!pmd->gso_ctx_mp) {
1475                 /*
1476                  * Create private mbuf pool with TAP_GSO_MBUF_SEG_SIZE
1477                  * bytes size per mbuf use this pool for both direct and
1478                  * indirect mbufs
1479                  */
1480                 ret = snprintf(pool_name, sizeof(pool_name), "mp_%s",
1481                                 dev->device->name);
1482                 if (ret < 0 || ret >= (int)sizeof(pool_name)) {
1483                         TAP_LOG(ERR,
1484                                 "%s: failed to create mbuf pool name for device %s,"
1485                                 "device name too long or output error, ret: %d\n",
1486                                 pmd->name, dev->device->name, ret);
1487                         return -ENAMETOOLONG;
1488                 }
1489                 pmd->gso_ctx_mp = rte_pktmbuf_pool_create(pool_name,
1490                         TAP_GSO_MBUFS_NUM, TAP_GSO_MBUF_CACHE_SIZE, 0,
1491                         RTE_PKTMBUF_HEADROOM + TAP_GSO_MBUF_SEG_SIZE,
1492                         SOCKET_ID_ANY);
1493                 if (!pmd->gso_ctx_mp) {
1494                         TAP_LOG(ERR,
1495                                 "%s: failed to create mbuf pool for device %s\n",
1496                                 pmd->name, dev->device->name);
1497                         return -1;
1498                 }
1499         }
1500
1501         gso_ctx->direct_pool = pmd->gso_ctx_mp;
1502         gso_ctx->indirect_pool = pmd->gso_ctx_mp;
1503         gso_ctx->gso_types = gso_types;
1504         gso_ctx->gso_size = 0; /* gso_size is set in tx_burst() per packet */
1505         gso_ctx->flag = 0;
1506
1507         return 0;
1508 }
1509
1510 static int
1511 tap_setup_queue(struct rte_eth_dev *dev,
1512                 struct pmd_internals *internals,
1513                 uint16_t qid,
1514                 int is_rx)
1515 {
1516         int ret;
1517         int *fd;
1518         int *other_fd;
1519         const char *dir;
1520         struct pmd_internals *pmd = dev->data->dev_private;
1521         struct pmd_process_private *process_private = dev->process_private;
1522         struct rx_queue *rx = &internals->rxq[qid];
1523         struct tx_queue *tx = &internals->txq[qid];
1524         struct rte_gso_ctx *gso_ctx;
1525
1526         if (is_rx) {
1527                 fd = &process_private->rxq_fds[qid];
1528                 other_fd = &process_private->txq_fds[qid];
1529                 dir = "rx";
1530                 gso_ctx = NULL;
1531         } else {
1532                 fd = &process_private->txq_fds[qid];
1533                 other_fd = &process_private->rxq_fds[qid];
1534                 dir = "tx";
1535                 gso_ctx = &tx->gso_ctx;
1536         }
1537         if (*fd != -1) {
1538                 /* fd for this queue already exists */
1539                 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1540                         pmd->name, *fd, dir, qid);
1541                 gso_ctx = NULL;
1542         } else if (*other_fd != -1) {
1543                 /* Only other_fd exists. dup it */
1544                 *fd = dup(*other_fd);
1545                 if (*fd < 0) {
1546                         *fd = -1;
1547                         TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1548                         return -1;
1549                 }
1550                 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1551                         pmd->name, *other_fd, dir, qid, *fd);
1552         } else {
1553                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1554                 *fd = tun_alloc(pmd, 0);
1555                 if (*fd < 0) {
1556                         *fd = -1; /* restore original value */
1557                         TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1558                         return -1;
1559                 }
1560                 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1561                         pmd->name, dir, qid, *fd);
1562         }
1563
1564         tx->mtu = &dev->data->mtu;
1565         rx->rxmode = &dev->data->dev_conf.rxmode;
1566         if (gso_ctx) {
1567                 ret = tap_gso_ctx_setup(gso_ctx, dev);
1568                 if (ret)
1569                         return -1;
1570         }
1571
1572         tx->type = pmd->type;
1573
1574         return *fd;
1575 }
1576
1577 static int
1578 tap_rx_queue_setup(struct rte_eth_dev *dev,
1579                    uint16_t rx_queue_id,
1580                    uint16_t nb_rx_desc,
1581                    unsigned int socket_id,
1582                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1583                    struct rte_mempool *mp)
1584 {
1585         struct pmd_internals *internals = dev->data->dev_private;
1586         struct pmd_process_private *process_private = dev->process_private;
1587         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1588         struct rte_mbuf **tmp = &rxq->pool;
1589         long iov_max = sysconf(_SC_IOV_MAX);
1590
1591         if (iov_max <= 0) {
1592                 TAP_LOG(WARNING,
1593                         "_SC_IOV_MAX is not defined. Using %d as default",
1594                         TAP_IOV_DEFAULT_MAX);
1595                 iov_max = TAP_IOV_DEFAULT_MAX;
1596         }
1597         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1598         struct iovec (*iovecs)[nb_desc + 1];
1599         int data_off = RTE_PKTMBUF_HEADROOM;
1600         int ret = 0;
1601         int fd;
1602         int i;
1603
1604         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1605                 TAP_LOG(WARNING,
1606                         "nb_rx_queues %d too small or mempool NULL",
1607                         dev->data->nb_rx_queues);
1608                 return -1;
1609         }
1610
1611         rxq->mp = mp;
1612         rxq->trigger_seen = 1; /* force initial burst */
1613         rxq->in_port = dev->data->port_id;
1614         rxq->queue_id = rx_queue_id;
1615         rxq->nb_rx_desc = nb_desc;
1616         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1617                                     socket_id);
1618         if (!iovecs) {
1619                 TAP_LOG(WARNING,
1620                         "%s: Couldn't allocate %d RX descriptors",
1621                         dev->device->name, nb_desc);
1622                 return -ENOMEM;
1623         }
1624         rxq->iovecs = iovecs;
1625
1626         dev->data->rx_queues[rx_queue_id] = rxq;
1627         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1628         if (fd == -1) {
1629                 ret = fd;
1630                 goto error;
1631         }
1632
1633         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1634         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1635
1636         for (i = 1; i <= nb_desc; i++) {
1637                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1638                 if (!*tmp) {
1639                         TAP_LOG(WARNING,
1640                                 "%s: couldn't allocate memory for queue %d",
1641                                 dev->device->name, rx_queue_id);
1642                         ret = -ENOMEM;
1643                         goto error;
1644                 }
1645                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1646                 (*rxq->iovecs)[i].iov_base =
1647                         (char *)(*tmp)->buf_addr + data_off;
1648                 data_off = 0;
1649                 tmp = &(*tmp)->next;
1650         }
1651
1652         TAP_LOG(DEBUG, "  RX TUNTAP device name %s, qid %d on fd %d",
1653                 internals->name, rx_queue_id,
1654                 process_private->rxq_fds[rx_queue_id]);
1655
1656         return 0;
1657
1658 error:
1659         tap_rxq_pool_free(rxq->pool);
1660         rxq->pool = NULL;
1661         rte_free(rxq->iovecs);
1662         rxq->iovecs = NULL;
1663         return ret;
1664 }
1665
1666 static int
1667 tap_tx_queue_setup(struct rte_eth_dev *dev,
1668                    uint16_t tx_queue_id,
1669                    uint16_t nb_tx_desc __rte_unused,
1670                    unsigned int socket_id __rte_unused,
1671                    const struct rte_eth_txconf *tx_conf)
1672 {
1673         struct pmd_internals *internals = dev->data->dev_private;
1674         struct pmd_process_private *process_private = dev->process_private;
1675         struct tx_queue *txq;
1676         int ret;
1677         uint64_t offloads;
1678
1679         if (tx_queue_id >= dev->data->nb_tx_queues)
1680                 return -1;
1681         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1682         txq = dev->data->tx_queues[tx_queue_id];
1683         txq->out_port = dev->data->port_id;
1684         txq->queue_id = tx_queue_id;
1685
1686         offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1687         txq->csum = !!(offloads &
1688                         (RTE_ETH_TX_OFFLOAD_IPV4_CKSUM |
1689                          RTE_ETH_TX_OFFLOAD_UDP_CKSUM |
1690                          RTE_ETH_TX_OFFLOAD_TCP_CKSUM));
1691
1692         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1693         if (ret == -1)
1694                 return -1;
1695         TAP_LOG(DEBUG,
1696                 "  TX TUNTAP device name %s, qid %d on fd %d csum %s",
1697                 internals->name, tx_queue_id,
1698                 process_private->txq_fds[tx_queue_id],
1699                 txq->csum ? "on" : "off");
1700
1701         return 0;
1702 }
1703
1704 static int
1705 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1706 {
1707         struct pmd_internals *pmd = dev->data->dev_private;
1708         struct ifreq ifr = { .ifr_mtu = mtu };
1709
1710         return tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1711 }
1712
1713 static int
1714 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1715                      struct rte_ether_addr *mc_addr_set __rte_unused,
1716                      uint32_t nb_mc_addr __rte_unused)
1717 {
1718         /*
1719          * Nothing to do actually: the tap has no filtering whatsoever, every
1720          * packet is received.
1721          */
1722         return 0;
1723 }
1724
1725 static int
1726 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1727 {
1728         struct rte_eth_dev *dev = arg;
1729         struct pmd_internals *pmd = dev->data->dev_private;
1730         struct ifinfomsg *info = NLMSG_DATA(nh);
1731
1732         if (nh->nlmsg_type != RTM_NEWLINK ||
1733             (info->ifi_index != pmd->if_index &&
1734              info->ifi_index != pmd->remote_if_index))
1735                 return 0;
1736         return tap_link_update(dev, 0);
1737 }
1738
1739 static void
1740 tap_dev_intr_handler(void *cb_arg)
1741 {
1742         struct rte_eth_dev *dev = cb_arg;
1743         struct pmd_internals *pmd = dev->data->dev_private;
1744
1745         if (rte_intr_fd_get(pmd->intr_handle) >= 0)
1746                 tap_nl_recv(rte_intr_fd_get(pmd->intr_handle),
1747                             tap_nl_msg_handler, dev);
1748 }
1749
1750 static int
1751 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1752 {
1753         struct pmd_internals *pmd = dev->data->dev_private;
1754         int ret;
1755
1756         /* In any case, disable interrupt if the conf is no longer there. */
1757         if (!dev->data->dev_conf.intr_conf.lsc) {
1758                 if (rte_intr_fd_get(pmd->intr_handle) != -1)
1759                         goto clean;
1760
1761                 return 0;
1762         }
1763         if (set) {
1764                 rte_intr_fd_set(pmd->intr_handle, tap_nl_init(RTMGRP_LINK));
1765                 if (unlikely(rte_intr_fd_get(pmd->intr_handle) == -1))
1766                         return -EBADF;
1767                 return rte_intr_callback_register(
1768                         pmd->intr_handle, tap_dev_intr_handler, dev);
1769         }
1770
1771 clean:
1772         do {
1773                 ret = rte_intr_callback_unregister(pmd->intr_handle,
1774                         tap_dev_intr_handler, dev);
1775                 if (ret >= 0) {
1776                         break;
1777                 } else if (ret == -EAGAIN) {
1778                         rte_delay_ms(100);
1779                 } else {
1780                         TAP_LOG(ERR, "intr callback unregister failed: %d",
1781                                      ret);
1782                         break;
1783                 }
1784         } while (true);
1785
1786         if (rte_intr_fd_get(pmd->intr_handle) >= 0) {
1787                 tap_nl_final(rte_intr_fd_get(pmd->intr_handle));
1788                 rte_intr_fd_set(pmd->intr_handle, -1);
1789         }
1790
1791         return 0;
1792 }
1793
1794 static int
1795 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1796 {
1797         int err;
1798
1799         err = tap_lsc_intr_handle_set(dev, set);
1800         if (err < 0) {
1801                 if (!set)
1802                         tap_rx_intr_vec_set(dev, 0);
1803                 return err;
1804         }
1805         err = tap_rx_intr_vec_set(dev, set);
1806         if (err && set)
1807                 tap_lsc_intr_handle_set(dev, 0);
1808         return err;
1809 }
1810
1811 static const uint32_t*
1812 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1813 {
1814         static const uint32_t ptypes[] = {
1815                 RTE_PTYPE_INNER_L2_ETHER,
1816                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1817                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1818                 RTE_PTYPE_INNER_L3_IPV4,
1819                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1820                 RTE_PTYPE_INNER_L3_IPV6,
1821                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1822                 RTE_PTYPE_INNER_L4_FRAG,
1823                 RTE_PTYPE_INNER_L4_UDP,
1824                 RTE_PTYPE_INNER_L4_TCP,
1825                 RTE_PTYPE_INNER_L4_SCTP,
1826                 RTE_PTYPE_L2_ETHER,
1827                 RTE_PTYPE_L2_ETHER_VLAN,
1828                 RTE_PTYPE_L2_ETHER_QINQ,
1829                 RTE_PTYPE_L3_IPV4,
1830                 RTE_PTYPE_L3_IPV4_EXT,
1831                 RTE_PTYPE_L3_IPV6_EXT,
1832                 RTE_PTYPE_L3_IPV6,
1833                 RTE_PTYPE_L4_FRAG,
1834                 RTE_PTYPE_L4_UDP,
1835                 RTE_PTYPE_L4_TCP,
1836                 RTE_PTYPE_L4_SCTP,
1837         };
1838
1839         return ptypes;
1840 }
1841
1842 static int
1843 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1844                   struct rte_eth_fc_conf *fc_conf)
1845 {
1846         fc_conf->mode = RTE_ETH_FC_NONE;
1847         return 0;
1848 }
1849
1850 static int
1851 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1852                   struct rte_eth_fc_conf *fc_conf)
1853 {
1854         if (fc_conf->mode != RTE_ETH_FC_NONE)
1855                 return -ENOTSUP;
1856         return 0;
1857 }
1858
1859 /**
1860  * DPDK callback to update the RSS hash configuration.
1861  *
1862  * @param dev
1863  *   Pointer to Ethernet device structure.
1864  * @param[in] rss_conf
1865  *   RSS configuration data.
1866  *
1867  * @return
1868  *   0 on success, a negative errno value otherwise and rte_errno is set.
1869  */
1870 static int
1871 tap_rss_hash_update(struct rte_eth_dev *dev,
1872                 struct rte_eth_rss_conf *rss_conf)
1873 {
1874         if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1875                 rte_errno = EINVAL;
1876                 return -rte_errno;
1877         }
1878         if (rss_conf->rss_key && rss_conf->rss_key_len) {
1879                 /*
1880                  * Currently TAP RSS key is hard coded
1881                  * and cannot be updated
1882                  */
1883                 TAP_LOG(ERR,
1884                         "port %u RSS key cannot be updated",
1885                         dev->data->port_id);
1886                 rte_errno = EINVAL;
1887                 return -rte_errno;
1888         }
1889         return 0;
1890 }
1891
1892 static int
1893 tap_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1894 {
1895         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1896
1897         return 0;
1898 }
1899
1900 static int
1901 tap_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1902 {
1903         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1904
1905         return 0;
1906 }
1907
1908 static int
1909 tap_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1910 {
1911         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1912
1913         return 0;
1914 }
1915
1916 static int
1917 tap_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1918 {
1919         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1920
1921         return 0;
1922 }
1923 static const struct eth_dev_ops ops = {
1924         .dev_start              = tap_dev_start,
1925         .dev_stop               = tap_dev_stop,
1926         .dev_close              = tap_dev_close,
1927         .dev_configure          = tap_dev_configure,
1928         .dev_infos_get          = tap_dev_info,
1929         .rx_queue_setup         = tap_rx_queue_setup,
1930         .tx_queue_setup         = tap_tx_queue_setup,
1931         .rx_queue_start         = tap_rx_queue_start,
1932         .tx_queue_start         = tap_tx_queue_start,
1933         .rx_queue_stop          = tap_rx_queue_stop,
1934         .tx_queue_stop          = tap_tx_queue_stop,
1935         .rx_queue_release       = tap_rx_queue_release,
1936         .tx_queue_release       = tap_tx_queue_release,
1937         .flow_ctrl_get          = tap_flow_ctrl_get,
1938         .flow_ctrl_set          = tap_flow_ctrl_set,
1939         .link_update            = tap_link_update,
1940         .dev_set_link_up        = tap_link_set_up,
1941         .dev_set_link_down      = tap_link_set_down,
1942         .promiscuous_enable     = tap_promisc_enable,
1943         .promiscuous_disable    = tap_promisc_disable,
1944         .allmulticast_enable    = tap_allmulti_enable,
1945         .allmulticast_disable   = tap_allmulti_disable,
1946         .mac_addr_set           = tap_mac_set,
1947         .mtu_set                = tap_mtu_set,
1948         .set_mc_addr_list       = tap_set_mc_addr_list,
1949         .stats_get              = tap_stats_get,
1950         .stats_reset            = tap_stats_reset,
1951         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1952         .rss_hash_update        = tap_rss_hash_update,
1953         .flow_ops_get           = tap_dev_flow_ops_get,
1954 };
1955
1956 static int
1957 eth_dev_tap_create(struct rte_vdev_device *vdev, const char *tap_name,
1958                    char *remote_iface, struct rte_ether_addr *mac_addr,
1959                    enum rte_tuntap_type type)
1960 {
1961         int numa_node = rte_socket_id();
1962         struct rte_eth_dev *dev;
1963         struct pmd_internals *pmd;
1964         struct pmd_process_private *process_private;
1965         const char *tuntap_name = tuntap_types[type];
1966         struct rte_eth_dev_data *data;
1967         struct ifreq ifr;
1968         int i;
1969
1970         TAP_LOG(DEBUG, "%s device on numa %u", tuntap_name, rte_socket_id());
1971
1972         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1973         if (!dev) {
1974                 TAP_LOG(ERR, "%s Unable to allocate device struct",
1975                                 tuntap_name);
1976                 goto error_exit_nodev;
1977         }
1978
1979         process_private = (struct pmd_process_private *)
1980                 rte_zmalloc_socket(tap_name, sizeof(struct pmd_process_private),
1981                         RTE_CACHE_LINE_SIZE, dev->device->numa_node);
1982
1983         if (process_private == NULL) {
1984                 TAP_LOG(ERR, "Failed to alloc memory for process private");
1985                 return -1;
1986         }
1987         pmd = dev->data->dev_private;
1988         dev->process_private = process_private;
1989         pmd->dev = dev;
1990         strlcpy(pmd->name, tap_name, sizeof(pmd->name));
1991         pmd->type = type;
1992         pmd->ka_fd = -1;
1993         pmd->nlsk_fd = -1;
1994         pmd->gso_ctx_mp = NULL;
1995
1996         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1997         if (pmd->ioctl_sock == -1) {
1998                 TAP_LOG(ERR,
1999                         "%s Unable to get a socket for management: %s",
2000                         tuntap_name, strerror(errno));
2001                 goto error_exit;
2002         }
2003
2004         /* Allocate interrupt instance */
2005         pmd->intr_handle = rte_intr_instance_alloc(RTE_INTR_INSTANCE_F_SHARED);
2006         if (pmd->intr_handle == NULL) {
2007                 TAP_LOG(ERR, "Failed to allocate intr handle");
2008                 goto error_exit;
2009         }
2010
2011         /* Setup some default values */
2012         data = dev->data;
2013         data->dev_private = pmd;
2014         data->dev_flags = RTE_ETH_DEV_INTR_LSC |
2015                                 RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
2016         data->numa_node = numa_node;
2017
2018         data->dev_link = pmd_link;
2019         data->mac_addrs = &pmd->eth_addr;
2020         /* Set the number of RX and TX queues */
2021         data->nb_rx_queues = 0;
2022         data->nb_tx_queues = 0;
2023
2024         dev->dev_ops = &ops;
2025         dev->rx_pkt_burst = pmd_rx_burst;
2026         dev->tx_pkt_burst = pmd_tx_burst;
2027
2028         rte_intr_type_set(pmd->intr_handle, RTE_INTR_HANDLE_EXT);
2029         rte_intr_fd_set(pmd->intr_handle, -1);
2030         dev->intr_handle = pmd->intr_handle;
2031
2032         /* Presetup the fds to -1 as being not valid */
2033         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
2034                 process_private->rxq_fds[i] = -1;
2035                 process_private->txq_fds[i] = -1;
2036         }
2037
2038         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
2039                 if (rte_is_zero_ether_addr(mac_addr))
2040                         rte_eth_random_addr((uint8_t *)&pmd->eth_addr);
2041                 else
2042                         rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
2043         }
2044
2045         /*
2046          * Allocate a TUN device keep-alive file descriptor that will only be
2047          * closed when the TUN device itself is closed or removed.
2048          * This keep-alive file descriptor will guarantee that the TUN device
2049          * exists even when all of its queues are closed
2050          */
2051         pmd->ka_fd = tun_alloc(pmd, 1);
2052         if (pmd->ka_fd == -1) {
2053                 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
2054                 goto error_exit;
2055         }
2056         TAP_LOG(DEBUG, "allocated %s", pmd->name);
2057
2058         ifr.ifr_mtu = dev->data->mtu;
2059         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
2060                 goto error_exit;
2061
2062         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
2063                 memset(&ifr, 0, sizeof(struct ifreq));
2064                 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
2065                 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
2066                                 RTE_ETHER_ADDR_LEN);
2067                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
2068                         goto error_exit;
2069         }
2070
2071         /*
2072          * Set up everything related to rte_flow:
2073          * - netlink socket
2074          * - tap / remote if_index
2075          * - mandatory QDISCs
2076          * - rte_flow actual/implicit lists
2077          * - implicit rules
2078          */
2079         pmd->nlsk_fd = tap_nl_init(0);
2080         if (pmd->nlsk_fd == -1) {
2081                 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
2082                         pmd->name);
2083                 goto disable_rte_flow;
2084         }
2085         pmd->if_index = if_nametoindex(pmd->name);
2086         if (!pmd->if_index) {
2087                 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
2088                 goto disable_rte_flow;
2089         }
2090         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
2091                 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
2092                         pmd->name);
2093                 goto disable_rte_flow;
2094         }
2095         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
2096                 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
2097                         pmd->name);
2098                 goto disable_rte_flow;
2099         }
2100         LIST_INIT(&pmd->flows);
2101
2102         if (strlen(remote_iface)) {
2103                 pmd->remote_if_index = if_nametoindex(remote_iface);
2104                 if (!pmd->remote_if_index) {
2105                         TAP_LOG(ERR, "%s: failed to get %s if_index.",
2106                                 pmd->name, remote_iface);
2107                         goto error_remote;
2108                 }
2109                 strlcpy(pmd->remote_iface, remote_iface, RTE_ETH_NAME_MAX_LEN);
2110
2111                 /* Save state of remote device */
2112                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
2113
2114                 /* Replicate remote MAC address */
2115                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
2116                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
2117                                 pmd->name, pmd->remote_iface);
2118                         goto error_remote;
2119                 }
2120                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
2121                            RTE_ETHER_ADDR_LEN);
2122                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
2123                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
2124                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
2125                                 pmd->name, remote_iface);
2126                         goto error_remote;
2127                 }
2128
2129                 /*
2130                  * Flush usually returns negative value because it tries to
2131                  * delete every QDISC (and on a running device, one QDISC at
2132                  * least is needed). Ignore negative return value.
2133                  */
2134                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
2135                 if (qdisc_create_ingress(pmd->nlsk_fd,
2136                                          pmd->remote_if_index) < 0) {
2137                         TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
2138                                 pmd->remote_iface);
2139                         goto error_remote;
2140                 }
2141                 LIST_INIT(&pmd->implicit_flows);
2142                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
2143                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
2144                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
2145                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
2146                         TAP_LOG(ERR,
2147                                 "%s: failed to create implicit rules.",
2148                                 pmd->name);
2149                         goto error_remote;
2150                 }
2151         }
2152
2153         rte_eth_dev_probing_finish(dev);
2154         return 0;
2155
2156 disable_rte_flow:
2157         TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
2158                 strerror(errno), errno);
2159         if (strlen(remote_iface)) {
2160                 TAP_LOG(ERR, "Remote feature requires flow support.");
2161                 goto error_exit;
2162         }
2163         rte_eth_dev_probing_finish(dev);
2164         return 0;
2165
2166 error_remote:
2167         TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
2168                 strerror(errno), errno);
2169         tap_flow_implicit_flush(pmd, NULL);
2170
2171 error_exit:
2172         if (pmd->nlsk_fd != -1)
2173                 close(pmd->nlsk_fd);
2174         if (pmd->ka_fd != -1)
2175                 close(pmd->ka_fd);
2176         if (pmd->ioctl_sock != -1)
2177                 close(pmd->ioctl_sock);
2178         /* mac_addrs must not be freed alone because part of dev_private */
2179         dev->data->mac_addrs = NULL;
2180         rte_eth_dev_release_port(dev);
2181         rte_intr_instance_free(pmd->intr_handle);
2182
2183 error_exit_nodev:
2184         TAP_LOG(ERR, "%s Unable to initialize %s",
2185                 tuntap_name, rte_vdev_device_name(vdev));
2186
2187         return -EINVAL;
2188 }
2189
2190 /* make sure name is a possible Linux network device name */
2191 static bool
2192 is_valid_iface(const char *name)
2193 {
2194         if (*name == '\0')
2195                 return false;
2196
2197         if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
2198                 return false;
2199
2200         while (*name) {
2201                 if (*name == '/' || *name == ':' || isspace(*name))
2202                         return false;
2203                 name++;
2204         }
2205         return true;
2206 }
2207
2208 static int
2209 set_interface_name(const char *key __rte_unused,
2210                    const char *value,
2211                    void *extra_args)
2212 {
2213         char *name = (char *)extra_args;
2214
2215         if (value) {
2216                 if (!is_valid_iface(value)) {
2217                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2218                                 value);
2219                         return -1;
2220                 }
2221                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2222         } else {
2223                 /* use tap%d which causes kernel to choose next available */
2224                 strlcpy(name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2225         }
2226         return 0;
2227 }
2228
2229 static int
2230 set_remote_iface(const char *key __rte_unused,
2231                  const char *value,
2232                  void *extra_args)
2233 {
2234         char *name = (char *)extra_args;
2235
2236         if (value) {
2237                 if (!is_valid_iface(value)) {
2238                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2239                                 value);
2240                         return -1;
2241                 }
2242                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2243         }
2244
2245         return 0;
2246 }
2247
2248 static int parse_user_mac(struct rte_ether_addr *user_mac,
2249                 const char *value)
2250 {
2251         unsigned int index = 0;
2252         char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
2253
2254         if (user_mac == NULL || value == NULL)
2255                 return 0;
2256
2257         strlcpy(mac_temp, value, sizeof(mac_temp));
2258         mac_byte = strtok(mac_temp, ":");
2259
2260         while ((mac_byte != NULL) &&
2261                         (strlen(mac_byte) <= 2) &&
2262                         (strlen(mac_byte) == strspn(mac_byte,
2263                                         ETH_TAP_CMP_MAC_FMT))) {
2264                 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
2265                 mac_byte = strtok(NULL, ":");
2266         }
2267
2268         return index;
2269 }
2270
2271 static int
2272 set_mac_type(const char *key __rte_unused,
2273              const char *value,
2274              void *extra_args)
2275 {
2276         struct rte_ether_addr *user_mac = extra_args;
2277
2278         if (!value)
2279                 return 0;
2280
2281         if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
2282                 static int iface_idx;
2283
2284                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
2285                 memcpy((char *)user_mac->addr_bytes, "\0dtap",
2286                         RTE_ETHER_ADDR_LEN);
2287                 user_mac->addr_bytes[RTE_ETHER_ADDR_LEN - 1] =
2288                         iface_idx++ + '0';
2289                 goto success;
2290         }
2291
2292         if (parse_user_mac(user_mac, value) != 6)
2293                 goto error;
2294 success:
2295         TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
2296         return 0;
2297
2298 error:
2299         TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
2300                 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
2301         return -1;
2302 }
2303
2304 /*
2305  * Open a TUN interface device. TUN PMD
2306  * 1) sets tap_type as false
2307  * 2) intakes iface as argument.
2308  * 3) as interface is virtual set speed to 10G
2309  */
2310 static int
2311 rte_pmd_tun_probe(struct rte_vdev_device *dev)
2312 {
2313         const char *name, *params;
2314         int ret;
2315         struct rte_kvargs *kvlist = NULL;
2316         char tun_name[RTE_ETH_NAME_MAX_LEN];
2317         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2318         struct rte_eth_dev *eth_dev;
2319
2320         name = rte_vdev_device_name(dev);
2321         params = rte_vdev_device_args(dev);
2322         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2323
2324         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
2325             strlen(params) == 0) {
2326                 eth_dev = rte_eth_dev_attach_secondary(name);
2327                 if (!eth_dev) {
2328                         TAP_LOG(ERR, "Failed to probe %s", name);
2329                         return -1;
2330                 }
2331                 eth_dev->dev_ops = &ops;
2332                 eth_dev->device = &dev->device;
2333                 rte_eth_dev_probing_finish(eth_dev);
2334                 return 0;
2335         }
2336
2337         /* use tun%d which causes kernel to choose next available */
2338         strlcpy(tun_name, DEFAULT_TUN_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2339
2340         if (params && (params[0] != '\0')) {
2341                 TAP_LOG(DEBUG, "parameters (%s)", params);
2342
2343                 kvlist = rte_kvargs_parse(params, valid_arguments);
2344                 if (kvlist) {
2345                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2346                                 ret = rte_kvargs_process(kvlist,
2347                                         ETH_TAP_IFACE_ARG,
2348                                         &set_interface_name,
2349                                         tun_name);
2350
2351                                 if (ret == -1)
2352                                         goto leave;
2353                         }
2354                 }
2355         }
2356         pmd_link.link_speed = RTE_ETH_SPEED_NUM_10G;
2357
2358         TAP_LOG(DEBUG, "Initializing pmd_tun for %s", name);
2359
2360         ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0,
2361                                  ETH_TUNTAP_TYPE_TUN);
2362
2363 leave:
2364         if (ret == -1) {
2365                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2366                         name, tun_name);
2367         }
2368         rte_kvargs_free(kvlist);
2369
2370         return ret;
2371 }
2372
2373 /* Request queue file descriptors from secondary to primary. */
2374 static int
2375 tap_mp_attach_queues(const char *port_name, struct rte_eth_dev *dev)
2376 {
2377         int ret;
2378         struct timespec timeout = {.tv_sec = 1, .tv_nsec = 0};
2379         struct rte_mp_msg request, *reply;
2380         struct rte_mp_reply replies;
2381         struct ipc_queues *request_param = (struct ipc_queues *)request.param;
2382         struct ipc_queues *reply_param;
2383         struct pmd_process_private *process_private = dev->process_private;
2384         int queue, fd_iterator;
2385
2386         /* Prepare the request */
2387         memset(&request, 0, sizeof(request));
2388         strlcpy(request.name, TAP_MP_KEY, sizeof(request.name));
2389         strlcpy(request_param->port_name, port_name,
2390                 sizeof(request_param->port_name));
2391         request.len_param = sizeof(*request_param);
2392         /* Send request and receive reply */
2393         ret = rte_mp_request_sync(&request, &replies, &timeout);
2394         if (ret < 0 || replies.nb_received != 1) {
2395                 TAP_LOG(ERR, "Failed to request queues from primary: %d",
2396                         rte_errno);
2397                 return -1;
2398         }
2399         reply = &replies.msgs[0];
2400         reply_param = (struct ipc_queues *)reply->param;
2401         TAP_LOG(DEBUG, "Received IPC reply for %s", reply_param->port_name);
2402
2403         /* Attach the queues from received file descriptors */
2404         if (reply_param->rxq_count + reply_param->txq_count != reply->num_fds) {
2405                 TAP_LOG(ERR, "Unexpected number of fds received");
2406                 return -1;
2407         }
2408
2409         dev->data->nb_rx_queues = reply_param->rxq_count;
2410         dev->data->nb_tx_queues = reply_param->txq_count;
2411         fd_iterator = 0;
2412         for (queue = 0; queue < reply_param->rxq_count; queue++)
2413                 process_private->rxq_fds[queue] = reply->fds[fd_iterator++];
2414         for (queue = 0; queue < reply_param->txq_count; queue++)
2415                 process_private->txq_fds[queue] = reply->fds[fd_iterator++];
2416         free(reply);
2417         return 0;
2418 }
2419
2420 /* Send the queue file descriptors from the primary process to secondary. */
2421 static int
2422 tap_mp_sync_queues(const struct rte_mp_msg *request, const void *peer)
2423 {
2424         struct rte_eth_dev *dev;
2425         struct pmd_process_private *process_private;
2426         struct rte_mp_msg reply;
2427         const struct ipc_queues *request_param =
2428                 (const struct ipc_queues *)request->param;
2429         struct ipc_queues *reply_param =
2430                 (struct ipc_queues *)reply.param;
2431         int queue;
2432
2433         /* Get requested port */
2434         TAP_LOG(DEBUG, "Received IPC request for %s", request_param->port_name);
2435         dev = rte_eth_dev_get_by_name(request_param->port_name);
2436         if (!dev) {
2437                 TAP_LOG(ERR, "Failed to get port id for %s",
2438                         request_param->port_name);
2439                 return -1;
2440         }
2441         process_private = dev->process_private;
2442
2443         /* Fill file descriptors for all queues */
2444         reply.num_fds = 0;
2445         reply_param->rxq_count = 0;
2446         if (dev->data->nb_rx_queues + dev->data->nb_tx_queues >
2447                         RTE_MP_MAX_FD_NUM){
2448                 TAP_LOG(ERR, "Number of rx/tx queues exceeds max number of fds");
2449                 return -1;
2450         }
2451
2452         for (queue = 0; queue < dev->data->nb_rx_queues; queue++) {
2453                 reply.fds[reply.num_fds++] = process_private->rxq_fds[queue];
2454                 reply_param->rxq_count++;
2455         }
2456         RTE_ASSERT(reply_param->rxq_count == dev->data->nb_rx_queues);
2457
2458         reply_param->txq_count = 0;
2459         for (queue = 0; queue < dev->data->nb_tx_queues; queue++) {
2460                 reply.fds[reply.num_fds++] = process_private->txq_fds[queue];
2461                 reply_param->txq_count++;
2462         }
2463         RTE_ASSERT(reply_param->txq_count == dev->data->nb_tx_queues);
2464
2465         /* Send reply */
2466         strlcpy(reply.name, request->name, sizeof(reply.name));
2467         strlcpy(reply_param->port_name, request_param->port_name,
2468                 sizeof(reply_param->port_name));
2469         reply.len_param = sizeof(*reply_param);
2470         if (rte_mp_reply(&reply, peer) < 0) {
2471                 TAP_LOG(ERR, "Failed to reply an IPC request to sync queues");
2472                 return -1;
2473         }
2474         return 0;
2475 }
2476
2477 /* Open a TAP interface device.
2478  */
2479 static int
2480 rte_pmd_tap_probe(struct rte_vdev_device *dev)
2481 {
2482         const char *name, *params;
2483         int ret;
2484         struct rte_kvargs *kvlist = NULL;
2485         int speed;
2486         char tap_name[RTE_ETH_NAME_MAX_LEN];
2487         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2488         struct rte_ether_addr user_mac = { .addr_bytes = {0} };
2489         struct rte_eth_dev *eth_dev;
2490         int tap_devices_count_increased = 0;
2491
2492         name = rte_vdev_device_name(dev);
2493         params = rte_vdev_device_args(dev);
2494
2495         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
2496                 eth_dev = rte_eth_dev_attach_secondary(name);
2497                 if (!eth_dev) {
2498                         TAP_LOG(ERR, "Failed to probe %s", name);
2499                         return -1;
2500                 }
2501                 eth_dev->dev_ops = &ops;
2502                 eth_dev->device = &dev->device;
2503                 eth_dev->rx_pkt_burst = pmd_rx_burst;
2504                 eth_dev->tx_pkt_burst = pmd_tx_burst;
2505                 if (!rte_eal_primary_proc_alive(NULL)) {
2506                         TAP_LOG(ERR, "Primary process is missing");
2507                         return -1;
2508                 }
2509                 eth_dev->process_private = (struct pmd_process_private *)
2510                         rte_zmalloc_socket(name,
2511                                 sizeof(struct pmd_process_private),
2512                                 RTE_CACHE_LINE_SIZE,
2513                                 eth_dev->device->numa_node);
2514                 if (eth_dev->process_private == NULL) {
2515                         TAP_LOG(ERR,
2516                                 "Failed to alloc memory for process private");
2517                         return -1;
2518                 }
2519
2520                 ret = tap_mp_attach_queues(name, eth_dev);
2521                 if (ret != 0)
2522                         return -1;
2523
2524                 if (!tap_devices_count) {
2525                         ret = rte_mp_action_register(TAP_MP_REQ_START_RXTX, tap_mp_req_start_rxtx);
2526                         if (ret < 0 && rte_errno != ENOTSUP) {
2527                                 TAP_LOG(ERR, "tap: Failed to register IPC callback: %s",
2528                                         strerror(rte_errno));
2529                                 return -1;
2530                         }
2531                 }
2532                 tap_devices_count++;
2533                 rte_eth_dev_probing_finish(eth_dev);
2534                 return 0;
2535         }
2536
2537         speed = RTE_ETH_SPEED_NUM_10G;
2538
2539         /* use tap%d which causes kernel to choose next available */
2540         strlcpy(tap_name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2541         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2542
2543         if (params && (params[0] != '\0')) {
2544                 TAP_LOG(DEBUG, "parameters (%s)", params);
2545
2546                 kvlist = rte_kvargs_parse(params, valid_arguments);
2547                 if (kvlist) {
2548                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2549                                 ret = rte_kvargs_process(kvlist,
2550                                                          ETH_TAP_IFACE_ARG,
2551                                                          &set_interface_name,
2552                                                          tap_name);
2553                                 if (ret == -1)
2554                                         goto leave;
2555                         }
2556
2557                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
2558                                 ret = rte_kvargs_process(kvlist,
2559                                                          ETH_TAP_REMOTE_ARG,
2560                                                          &set_remote_iface,
2561                                                          remote_iface);
2562                                 if (ret == -1)
2563                                         goto leave;
2564                         }
2565
2566                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
2567                                 ret = rte_kvargs_process(kvlist,
2568                                                          ETH_TAP_MAC_ARG,
2569                                                          &set_mac_type,
2570                                                          &user_mac);
2571                                 if (ret == -1)
2572                                         goto leave;
2573                         }
2574                 }
2575         }
2576         pmd_link.link_speed = speed;
2577
2578         TAP_LOG(DEBUG, "Initializing pmd_tap for %s", name);
2579
2580         /* Register IPC feed callback */
2581         if (!tap_devices_count) {
2582                 ret = rte_mp_action_register(TAP_MP_KEY, tap_mp_sync_queues);
2583                 if (ret < 0 && rte_errno != ENOTSUP) {
2584                         TAP_LOG(ERR, "tap: Failed to register IPC callback: %s",
2585                                 strerror(rte_errno));
2586                         goto leave;
2587                 }
2588         }
2589         tap_devices_count++;
2590         tap_devices_count_increased = 1;
2591         ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
2592                 ETH_TUNTAP_TYPE_TAP);
2593
2594 leave:
2595         if (ret == -1) {
2596                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2597                         name, tap_name);
2598                 if (tap_devices_count_increased == 1) {
2599                         if (tap_devices_count == 1)
2600                                 rte_mp_action_unregister(TAP_MP_KEY);
2601                         tap_devices_count--;
2602                 }
2603         }
2604         rte_kvargs_free(kvlist);
2605
2606         return ret;
2607 }
2608
2609 /* detach a TUNTAP device.
2610  */
2611 static int
2612 rte_pmd_tap_remove(struct rte_vdev_device *dev)
2613 {
2614         struct rte_eth_dev *eth_dev = NULL;
2615
2616         /* find the ethdev entry */
2617         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
2618         if (!eth_dev)
2619                 return 0;
2620
2621         tap_dev_close(eth_dev);
2622         rte_eth_dev_release_port(eth_dev);
2623
2624         return 0;
2625 }
2626
2627 static struct rte_vdev_driver pmd_tun_drv = {
2628         .probe = rte_pmd_tun_probe,
2629         .remove = rte_pmd_tap_remove,
2630 };
2631
2632 static struct rte_vdev_driver pmd_tap_drv = {
2633         .probe = rte_pmd_tap_probe,
2634         .remove = rte_pmd_tap_remove,
2635 };
2636
2637 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
2638 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
2639 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
2640 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
2641                               ETH_TAP_IFACE_ARG "=<string> ");
2642 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
2643                               ETH_TAP_IFACE_ARG "=<string> "
2644                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
2645                               ETH_TAP_REMOTE_ARG "=<string>");
2646 RTE_LOG_REGISTER_DEFAULT(tap_logtype, NOTICE);