drivers/net/tap/rte_eth_tap.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2016-2017 Intel Corporation
   3  */
   4
   5 #include <rte_atomic.h>
   6 #include <rte_branch_prediction.h>
   7 #include <rte_byteorder.h>
   8 #include <rte_common.h>
   9 #include <rte_mbuf.h>
  10 #include <rte_ethdev.h>
  11 #include <rte_ethdev_vdev.h>
  12 #include <rte_malloc.h>
  13 #include <rte_bus_vdev.h>
  14 #include <rte_kvargs.h>
  15 #include <rte_net.h>
  16 #include <rte_debug.h>
  17 #include <rte_ip.h>
  18
  19 #include <sys/types.h>
  20 #include <sys/stat.h>
  21 #include <sys/socket.h>
  22 #include <sys/ioctl.h>
  23 #include <sys/utsname.h>
  24 #include <sys/mman.h>
  25 #include <errno.h>
  26 #include <signal.h>
  27 #include <stdint.h>
  28 #include <sys/uio.h>
  29 #include <unistd.h>
  30 #include <arpa/inet.h>
  31 #include <net/if.h>
  32 #include <linux/if_tun.h>
  33 #include <linux/if_ether.h>
  34 #include <linux/version.h>
  35 #include <fcntl.h>
  36
  37 #include <rte_eth_tap.h>
  38 #include <tap_flow.h>
  39 #include <tap_netlink.h>
  40 #include <tap_tcmsgs.h>
  41
  42 /* Linux based path to the TUN device */
  43 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  44 #define DEFAULT_TAP_NAME        "dtap"
  45
  46 #define ETH_TAP_IFACE_ARG       "iface"
  47 #define ETH_TAP_SPEED_ARG       "speed"
  48 #define ETH_TAP_REMOTE_ARG      "remote"
  49 #define ETH_TAP_MAC_ARG         "mac"
  50 #define ETH_TAP_MAC_FIXED       "fixed"
  51
  52 #define FLOWER_KERNEL_VERSION KERNEL_VERSION(4, 2, 0)
  53 #define FLOWER_VLAN_KERNEL_VERSION KERNEL_VERSION(4, 9, 0)
  54
  55 static struct rte_vdev_driver pmd_tap_drv;
  56
  57 static const char *valid_arguments[] = {
  58         ETH_TAP_IFACE_ARG,
  59         ETH_TAP_SPEED_ARG,
  60         ETH_TAP_REMOTE_ARG,
  61         ETH_TAP_MAC_ARG,
  62         NULL
  63 };
  64
  65 static int tap_unit;
  66
  67 static volatile uint32_t tap_trigger;   /* Rx trigger */
  68
  69 static struct rte_eth_link pmd_link = {
  70         .link_speed = ETH_SPEED_NUM_10G,
  71         .link_duplex = ETH_LINK_FULL_DUPLEX,
  72         .link_status = ETH_LINK_DOWN,
  73         .link_autoneg = ETH_LINK_AUTONEG
  74 };
  75
  76 static void
  77 tap_trigger_cb(int sig __rte_unused)
  78 {
  79         /* Valid trigger values are nonzero */
  80         tap_trigger = (tap_trigger + 1) | 0x80000000;
  81 }
  82
  83 /* Specifies on what netdevices the ioctl should be applied */
  84 enum ioctl_mode {
  85         LOCAL_AND_REMOTE,
  86         LOCAL_ONLY,
  87         REMOTE_ONLY,
  88 };
  89
  90 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
  91
  92 /* Tun/Tap allocation routine
  93  *
  94  * name is the number of the interface to use, unless NULL to take the host
  95  * supplied name.
  96  */
  97 static int
  98 tun_alloc(struct pmd_internals *pmd)
  99 {
 100         struct ifreq ifr;
 101 #ifdef IFF_MULTI_QUEUE
 102         unsigned int features;
 103 #endif
 104         int fd;
 105
 106         memset(&ifr, 0, sizeof(struct ifreq));
 107
 108         /*
 109          * Do not set IFF_NO_PI as packet information header will be needed
 110          * to check if a received packet has been truncated.
 111          */
 112         ifr.ifr_flags = IFF_TAP;
 113         snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
 114
 115         RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
 116
 117         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 118         if (fd < 0) {
 119                 RTE_LOG(ERR, PMD, "Unable to create TAP interface\n");
 120                 goto error;
 121         }
 122
 123 #ifdef IFF_MULTI_QUEUE
 124         /* Grab the TUN features to verify we can work multi-queue */
 125         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 126                 RTE_LOG(ERR, PMD, "TAP unable to get TUN/TAP features\n");
 127                 goto error;
 128         }
 129         RTE_LOG(DEBUG, PMD, "  TAP Features %08x\n", features);
 130
 131         if (features & IFF_MULTI_QUEUE) {
 132                 RTE_LOG(DEBUG, PMD, "  Multi-queue support for %d queues\n",
 133                         RTE_PMD_TAP_MAX_QUEUES);
 134                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 135         } else
 136 #endif
 137         {
 138                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 139                 RTE_LOG(DEBUG, PMD, "  Single queue only support\n");
 140         }
 141
 142         /* Set the TUN/TAP configuration and set the name if needed */
 143         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 144                 RTE_LOG(WARNING, PMD,
 145                         "Unable to set TUNSETIFF for %s\n",
 146                         ifr.ifr_name);
 147                 perror("TUNSETIFF");
 148                 goto error;
 149         }
 150
 151         /* Always set the file descriptor to non-blocking */
 152         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 153                 RTE_LOG(WARNING, PMD,
 154                         "Unable to set %s to nonblocking\n",
 155                         ifr.ifr_name);
 156                 perror("F_SETFL, NONBLOCK");
 157                 goto error;
 158         }
 159
 160         /* Set up trigger to optimize empty Rx bursts */
 161         errno = 0;
 162         do {
 163                 struct sigaction sa;
 164                 int flags = fcntl(fd, F_GETFL);
 165
 166                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 167                         break;
 168                 if (sa.sa_handler != tap_trigger_cb) {
 169                         /*
 170                          * Make sure SIGIO is not already taken. This is done
 171                          * as late as possible to leave the application a
 172                          * chance to set up its own signal handler first.
 173                          */
 174                         if (sa.sa_handler != SIG_IGN &&
 175                             sa.sa_handler != SIG_DFL) {
 176                                 errno = EBUSY;
 177                                 break;
 178                         }
 179                         sa = (struct sigaction){
 180                                 .sa_flags = SA_RESTART,
 181                                 .sa_handler = tap_trigger_cb,
 182                         };
 183                         if (sigaction(SIGIO, &sa, NULL) == -1)
 184                                 break;
 185                 }
 186                 /* Enable SIGIO on file descriptor */
 187                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 188                 fcntl(fd, F_SETOWN, getpid());
 189         } while (0);
 190         if (errno) {
 191                 /* Disable trigger globally in case of error */
 192                 tap_trigger = 0;
 193                 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
 194                         strerror(errno));
 195         }
 196
 197         return fd;
 198
 199 error:
 200         if (fd > 0)
 201                 close(fd);
 202         return -1;
 203 }
 204
 205 static void
 206 tap_verify_csum(struct rte_mbuf *mbuf)
 207 {
 208         uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
 209         uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
 210         uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
 211         unsigned int l2_len = sizeof(struct ether_hdr);
 212         unsigned int l3_len;
 213         uint16_t cksum = 0;
 214         void *l3_hdr;
 215         void *l4_hdr;
 216
 217         if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
 218                 l2_len += 4;
 219         else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
 220                 l2_len += 8;
 221         /* Don't verify checksum for packets with discontinuous L2 header */
 222         if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
 223                      rte_pktmbuf_data_len(mbuf)))
 224                 return;
 225         l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
 226         if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 227                 struct ipv4_hdr *iph = l3_hdr;
 228
 229                 /* ihl contains the number of 4-byte words in the header */
 230                 l3_len = 4 * (iph->version_ihl & 0xf);
 231                 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
 232                         return;
 233
 234                 cksum = ~rte_raw_cksum(iph, l3_len);
 235                 mbuf->ol_flags |= cksum ?
 236                         PKT_RX_IP_CKSUM_BAD :
 237                         PKT_RX_IP_CKSUM_GOOD;
 238         } else if (l3 == RTE_PTYPE_L3_IPV6) {
 239                 l3_len = sizeof(struct ipv6_hdr);
 240         } else {
 241                 /* IPv6 extensions are not supported */
 242                 return;
 243         }
 244         if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
 245                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 246                 /* Don't verify checksum for multi-segment packets. */
 247                 if (mbuf->nb_segs > 1)
 248                         return;
 249                 if (l3 == RTE_PTYPE_L3_IPV4)
 250                         cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
 251                 else if (l3 == RTE_PTYPE_L3_IPV6)
 252                         cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
 253                 mbuf->ol_flags |= cksum ?
 254                         PKT_RX_L4_CKSUM_BAD :
 255                         PKT_RX_L4_CKSUM_GOOD;
 256         }
 257 }
 258
 259 /* Callback to handle the rx burst of packets to the correct interface and
 260  * file descriptor(s) in a multi-queue setup.
 261  */
 262 static uint16_t
 263 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 264 {
 265         struct rx_queue *rxq = queue;
 266         uint16_t num_rx;
 267         unsigned long num_rx_bytes = 0;
 268         uint32_t trigger = tap_trigger;
 269
 270         if (trigger == rxq->trigger_seen)
 271                 return 0;
 272         if (trigger)
 273                 rxq->trigger_seen = trigger;
 274         rte_compiler_barrier();
 275         for (num_rx = 0; num_rx < nb_pkts; ) {
 276                 struct rte_mbuf *mbuf = rxq->pool;
 277                 struct rte_mbuf *seg = NULL;
 278                 struct rte_mbuf *new_tail = NULL;
 279                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 280                 int len;
 281
 282                 len = readv(rxq->fd, *rxq->iovecs,
 283                             1 + (rxq->rxmode->enable_scatter ?
 284                                  rxq->nb_rx_desc : 1));
 285                 if (len < (int)sizeof(struct tun_pi))
 286                         break;
 287
 288                 /* Packet couldn't fit in the provided mbuf */
 289                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 290                         rxq->stats.ierrors++;
 291                         continue;
 292                 }
 293
 294                 len -= sizeof(struct tun_pi);
 295
 296                 mbuf->pkt_len = len;
 297                 mbuf->port = rxq->in_port;
 298                 while (1) {
 299                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 300
 301                         if (unlikely(!buf)) {
 302                                 rxq->stats.rx_nombuf++;
 303                                 /* No new buf has been allocated: do nothing */
 304                                 if (!new_tail || !seg)
 305                                         goto end;
 306
 307                                 seg->next = NULL;
 308                                 rte_pktmbuf_free(mbuf);
 309
 310                                 goto end;
 311                         }
 312                         seg = seg ? seg->next : mbuf;
 313                         if (rxq->pool == mbuf)
 314                                 rxq->pool = buf;
 315                         if (new_tail)
 316                                 new_tail->next = buf;
 317                         new_tail = buf;
 318                         new_tail->next = seg->next;
 319
 320                         /* iovecs[0] is reserved for packet info (pi) */
 321                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 322                                 buf->buf_len - data_off;
 323                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 324                                 (char *)buf->buf_addr + data_off;
 325
 326                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 327                         seg->data_off = data_off;
 328
 329                         len -= seg->data_len;
 330                         if (len <= 0)
 331                                 break;
 332                         mbuf->nb_segs++;
 333                         /* First segment has headroom, not the others */
 334                         data_off = 0;
 335                 }
 336                 seg->next = NULL;
 337                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 338                                                       RTE_PTYPE_ALL_MASK);
 339                 if (rxq->rxmode->hw_ip_checksum)
 340                         tap_verify_csum(mbuf);
 341
 342                 /* account for the receive frame */
 343                 bufs[num_rx++] = mbuf;
 344                 num_rx_bytes += mbuf->pkt_len;
 345         }
 346 end:
 347         rxq->stats.ipackets += num_rx;
 348         rxq->stats.ibytes += num_rx_bytes;
 349
 350         return num_rx;
 351 }
 352
 353 static void
 354 tap_tx_offload(char *packet, uint64_t ol_flags, unsigned int l2_len,
 355                unsigned int l3_len)
 356 {
 357         void *l3_hdr = packet + l2_len;
 358
 359         if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
 360                 struct ipv4_hdr *iph = l3_hdr;
 361                 uint16_t cksum;
 362
 363                 iph->hdr_checksum = 0;
 364                 cksum = rte_raw_cksum(iph, l3_len);
 365                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 366         }
 367         if (ol_flags & PKT_TX_L4_MASK) {
 368                 uint16_t l4_len;
 369                 uint32_t cksum;
 370                 uint16_t *l4_cksum;
 371                 void *l4_hdr;
 372
 373                 l4_hdr = packet + l2_len + l3_len;
 374                 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
 375                         l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
 376                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 377                         l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
 378                 else
 379                         return;
 380                 *l4_cksum = 0;
 381                 if (ol_flags & PKT_TX_IPV4) {
 382                         struct ipv4_hdr *iph = l3_hdr;
 383
 384                         l4_len = rte_be_to_cpu_16(iph->total_length) - l3_len;
 385                         cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 386                 } else {
 387                         struct ipv6_hdr *ip6h = l3_hdr;
 388
 389                         /* payload_len does not include ext headers */
 390                         l4_len = rte_be_to_cpu_16(ip6h->payload_len) -
 391                                 l3_len + sizeof(struct ipv6_hdr);
 392                         cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 393                 }
 394                 cksum += rte_raw_cksum(l4_hdr, l4_len);
 395                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 396                 cksum = (~cksum) & 0xffff;
 397                 if (cksum == 0)
 398                         cksum = 0xffff;
 399                 *l4_cksum = cksum;
 400         }
 401 }
 402
 403 /* Callback to handle sending packets from the tap interface
 404  */
 405 static uint16_t
 406 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 407 {
 408         struct tx_queue *txq = queue;
 409         uint16_t num_tx = 0;
 410         unsigned long num_tx_bytes = 0;
 411         uint32_t max_size;
 412         int i;
 413
 414         if (unlikely(nb_pkts == 0))
 415                 return 0;
 416
 417         max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
 418         for (i = 0; i < nb_pkts; i++) {
 419                 struct rte_mbuf *mbuf = bufs[num_tx];
 420                 struct iovec iovecs[mbuf->nb_segs + 1];
 421                 struct tun_pi pi = { .flags = 0 };
 422                 struct rte_mbuf *seg = mbuf;
 423                 char m_copy[mbuf->data_len];
 424                 int n;
 425                 int j;
 426
 427                 /* stats.errs will be incremented */
 428                 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
 429                         break;
 430
 431                 iovecs[0].iov_base = &pi;
 432                 iovecs[0].iov_len = sizeof(pi);
 433                 for (j = 1; j <= mbuf->nb_segs; j++) {
 434                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 435                         iovecs[j].iov_base =
 436                                 rte_pktmbuf_mtod(seg, void *);
 437                         seg = seg->next;
 438                 }
 439                 if (mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 440                     (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 441                     (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM) {
 442                         /* Support only packets with all data in the same seg */
 443                         if (mbuf->nb_segs > 1)
 444                                 break;
 445                         /* To change checksums, work on a copy of data. */
 446                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 447                                    rte_pktmbuf_data_len(mbuf));
 448                         tap_tx_offload(m_copy, mbuf->ol_flags,
 449                                        mbuf->l2_len, mbuf->l3_len);
 450                         iovecs[1].iov_base = m_copy;
 451                 }
 452                 /* copy the tx frame data */
 453                 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
 454                 if (n <= 0)
 455                         break;
 456
 457                 num_tx++;
 458                 num_tx_bytes += mbuf->pkt_len;
 459                 rte_pktmbuf_free(mbuf);
 460         }
 461
 462         txq->stats.opackets += num_tx;
 463         txq->stats.errs += nb_pkts - num_tx;
 464         txq->stats.obytes += num_tx_bytes;
 465
 466         return num_tx;
 467 }
 468
 469 static const char *
 470 tap_ioctl_req2str(unsigned long request)
 471 {
 472         switch (request) {
 473         case SIOCSIFFLAGS:
 474                 return "SIOCSIFFLAGS";
 475         case SIOCGIFFLAGS:
 476                 return "SIOCGIFFLAGS";
 477         case SIOCGIFHWADDR:
 478                 return "SIOCGIFHWADDR";
 479         case SIOCSIFHWADDR:
 480                 return "SIOCSIFHWADDR";
 481         case SIOCSIFMTU:
 482                 return "SIOCSIFMTU";
 483         }
 484         return "UNKNOWN";
 485 }
 486
 487 static int
 488 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 489           struct ifreq *ifr, int set, enum ioctl_mode mode)
 490 {
 491         short req_flags = ifr->ifr_flags;
 492         int remote = pmd->remote_if_index &&
 493                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 494
 495         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 496                 return 0;
 497         /*
 498          * If there is a remote netdevice, apply ioctl on it, then apply it on
 499          * the tap netdevice.
 500          */
 501 apply:
 502         if (remote)
 503                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
 504         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 505                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 506         switch (request) {
 507         case SIOCSIFFLAGS:
 508                 /* fetch current flags to leave other flags untouched */
 509                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 510                         goto error;
 511                 if (set)
 512                         ifr->ifr_flags |= req_flags;
 513                 else
 514                         ifr->ifr_flags &= ~req_flags;
 515                 break;
 516         case SIOCGIFFLAGS:
 517         case SIOCGIFHWADDR:
 518         case SIOCSIFHWADDR:
 519         case SIOCSIFMTU:
 520                 break;
 521         default:
 522                 RTE_ASSERT(!"unsupported request type: must not happen");
 523         }
 524         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 525                 goto error;
 526         if (remote-- && mode == LOCAL_AND_REMOTE)
 527                 goto apply;
 528         return 0;
 529
 530 error:
 531         RTE_LOG(DEBUG, PMD, "%s: %s(%s) failed: %s(%d)\n", ifr->ifr_name,
 532                 __func__, tap_ioctl_req2str(request), strerror(errno), errno);
 533         return -errno;
 534 }
 535
 536 static int
 537 tap_link_set_down(struct rte_eth_dev *dev)
 538 {
 539         struct pmd_internals *pmd = dev->data->dev_private;
 540         struct ifreq ifr = { .ifr_flags = IFF_UP };
 541
 542         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 543         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 544 }
 545
 546 static int
 547 tap_link_set_up(struct rte_eth_dev *dev)
 548 {
 549         struct pmd_internals *pmd = dev->data->dev_private;
 550         struct ifreq ifr = { .ifr_flags = IFF_UP };
 551
 552         dev->data->dev_link.link_status = ETH_LINK_UP;
 553         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 554 }
 555
 556 static int
 557 tap_dev_start(struct rte_eth_dev *dev)
 558 {
 559         int err;
 560
 561         err = tap_intr_handle_set(dev, 1);
 562         if (err)
 563                 return err;
 564         return tap_link_set_up(dev);
 565 }
 566
 567 /* This function gets called when the current port gets stopped.
 568  */
 569 static void
 570 tap_dev_stop(struct rte_eth_dev *dev)
 571 {
 572         tap_intr_handle_set(dev, 0);
 573         tap_link_set_down(dev);
 574 }
 575
 576 static int
 577 tap_dev_configure(struct rte_eth_dev *dev)
 578 {
 579         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 580                 RTE_LOG(ERR, PMD,
 581                         "%s: number of rx queues %d exceeds max num of queues %d\n",
 582                         dev->device->name,
 583                         dev->data->nb_rx_queues,
 584                         RTE_PMD_TAP_MAX_QUEUES);
 585                 return -1;
 586         }
 587         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 588                 RTE_LOG(ERR, PMD,
 589                         "%s: number of tx queues %d exceeds max num of queues %d\n",
 590                         dev->device->name,
 591                         dev->data->nb_tx_queues,
 592                         RTE_PMD_TAP_MAX_QUEUES);
 593                 return -1;
 594         }
 595
 596         RTE_LOG(INFO, PMD, "%s: %p: TX configured queues number: %u\n",
 597              dev->device->name, (void *)dev, dev->data->nb_tx_queues);
 598
 599         RTE_LOG(INFO, PMD, "%s: %p: RX configured queues number: %u\n",
 600              dev->device->name, (void *)dev, dev->data->nb_rx_queues);
 601
 602         return 0;
 603 }
 604
 605 static uint32_t
 606 tap_dev_speed_capa(void)
 607 {
 608         uint32_t speed = pmd_link.link_speed;
 609         uint32_t capa = 0;
 610
 611         if (speed >= ETH_SPEED_NUM_10M)
 612                 capa |= ETH_LINK_SPEED_10M;
 613         if (speed >= ETH_SPEED_NUM_100M)
 614                 capa |= ETH_LINK_SPEED_100M;
 615         if (speed >= ETH_SPEED_NUM_1G)
 616                 capa |= ETH_LINK_SPEED_1G;
 617         if (speed >= ETH_SPEED_NUM_5G)
 618                 capa |= ETH_LINK_SPEED_2_5G;
 619         if (speed >= ETH_SPEED_NUM_5G)
 620                 capa |= ETH_LINK_SPEED_5G;
 621         if (speed >= ETH_SPEED_NUM_10G)
 622                 capa |= ETH_LINK_SPEED_10G;
 623         if (speed >= ETH_SPEED_NUM_20G)
 624                 capa |= ETH_LINK_SPEED_20G;
 625         if (speed >= ETH_SPEED_NUM_25G)
 626                 capa |= ETH_LINK_SPEED_25G;
 627         if (speed >= ETH_SPEED_NUM_40G)
 628                 capa |= ETH_LINK_SPEED_40G;
 629         if (speed >= ETH_SPEED_NUM_50G)
 630                 capa |= ETH_LINK_SPEED_50G;
 631         if (speed >= ETH_SPEED_NUM_56G)
 632                 capa |= ETH_LINK_SPEED_56G;
 633         if (speed >= ETH_SPEED_NUM_100G)
 634                 capa |= ETH_LINK_SPEED_100G;
 635
 636         return capa;
 637 }
 638
 639 static void
 640 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 641 {
 642         struct pmd_internals *internals = dev->data->dev_private;
 643
 644         dev_info->if_index = internals->if_index;
 645         dev_info->max_mac_addrs = 1;
 646         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 647         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
 648         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
 649         dev_info->min_rx_bufsize = 0;
 650         dev_info->pci_dev = NULL;
 651         dev_info->speed_capa = tap_dev_speed_capa();
 652         dev_info->rx_offload_capa = (DEV_RX_OFFLOAD_IPV4_CKSUM |
 653                                      DEV_RX_OFFLOAD_UDP_CKSUM |
 654                                      DEV_RX_OFFLOAD_TCP_CKSUM);
 655         dev_info->tx_offload_capa =
 656                 (DEV_TX_OFFLOAD_IPV4_CKSUM |
 657                  DEV_TX_OFFLOAD_UDP_CKSUM |
 658                  DEV_TX_OFFLOAD_TCP_CKSUM);
 659 }
 660
 661 static int
 662 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 663 {
 664         unsigned int i, imax;
 665         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 666         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 667         unsigned long rx_nombuf = 0, ierrors = 0;
 668         const struct pmd_internals *pmd = dev->data->dev_private;
 669
 670         /* rx queue statistics */
 671         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 672                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 673         for (i = 0; i < imax; i++) {
 674                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 675                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 676                 rx_total += tap_stats->q_ipackets[i];
 677                 rx_bytes_total += tap_stats->q_ibytes[i];
 678                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
 679                 ierrors += pmd->rxq[i].stats.ierrors;
 680         }
 681
 682         /* tx queue statistics */
 683         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 684                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 685
 686         for (i = 0; i < imax; i++) {
 687                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 688                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 689                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 690                 tx_total += tap_stats->q_opackets[i];
 691                 tx_err_total += tap_stats->q_errors[i];
 692                 tx_bytes_total += tap_stats->q_obytes[i];
 693         }
 694
 695         tap_stats->ipackets = rx_total;
 696         tap_stats->ibytes = rx_bytes_total;
 697         tap_stats->ierrors = ierrors;
 698         tap_stats->rx_nombuf = rx_nombuf;
 699         tap_stats->opackets = tx_total;
 700         tap_stats->oerrors = tx_err_total;
 701         tap_stats->obytes = tx_bytes_total;
 702         return 0;
 703 }
 704
 705 static void
 706 tap_stats_reset(struct rte_eth_dev *dev)
 707 {
 708         int i;
 709         struct pmd_internals *pmd = dev->data->dev_private;
 710
 711         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 712                 pmd->rxq[i].stats.ipackets = 0;
 713                 pmd->rxq[i].stats.ibytes = 0;
 714                 pmd->rxq[i].stats.ierrors = 0;
 715                 pmd->rxq[i].stats.rx_nombuf = 0;
 716
 717                 pmd->txq[i].stats.opackets = 0;
 718                 pmd->txq[i].stats.errs = 0;
 719                 pmd->txq[i].stats.obytes = 0;
 720         }
 721 }
 722
 723 static void
 724 tap_dev_close(struct rte_eth_dev *dev)
 725 {
 726         int i;
 727         struct pmd_internals *internals = dev->data->dev_private;
 728
 729         tap_link_set_down(dev);
 730         tap_flow_flush(dev, NULL);
 731         tap_flow_implicit_flush(internals, NULL);
 732
 733         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 734                 if (internals->rxq[i].fd != -1) {
 735                         close(internals->rxq[i].fd);
 736                         internals->rxq[i].fd = -1;
 737                 }
 738                 if (internals->txq[i].fd != -1) {
 739                         close(internals->txq[i].fd);
 740                         internals->txq[i].fd = -1;
 741                 }
 742         }
 743
 744         if (internals->remote_if_index) {
 745                 /* Restore initial remote state */
 746                 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
 747                                 &internals->remote_initial_flags);
 748         }
 749 }
 750
 751 static void
 752 tap_rx_queue_release(void *queue)
 753 {
 754         struct rx_queue *rxq = queue;
 755
 756         if (rxq && (rxq->fd > 0)) {
 757                 close(rxq->fd);
 758                 rxq->fd = -1;
 759                 rte_pktmbuf_free(rxq->pool);
 760                 rte_free(rxq->iovecs);
 761                 rxq->pool = NULL;
 762                 rxq->iovecs = NULL;
 763         }
 764 }
 765
 766 static void
 767 tap_tx_queue_release(void *queue)
 768 {
 769         struct tx_queue *txq = queue;
 770
 771         if (txq && (txq->fd > 0)) {
 772                 close(txq->fd);
 773                 txq->fd = -1;
 774         }
 775 }
 776
 777 static int
 778 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
 779 {
 780         struct rte_eth_link *dev_link = &dev->data->dev_link;
 781         struct pmd_internals *pmd = dev->data->dev_private;
 782         struct ifreq ifr = { .ifr_flags = 0 };
 783
 784         if (pmd->remote_if_index) {
 785                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
 786                 if (!(ifr.ifr_flags & IFF_UP) ||
 787                     !(ifr.ifr_flags & IFF_RUNNING)) {
 788                         dev_link->link_status = ETH_LINK_DOWN;
 789                         return 0;
 790                 }
 791         }
 792         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
 793         dev_link->link_status =
 794                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
 795                  ETH_LINK_UP :
 796                  ETH_LINK_DOWN);
 797         return 0;
 798 }
 799
 800 static void
 801 tap_promisc_enable(struct rte_eth_dev *dev)
 802 {
 803         struct pmd_internals *pmd = dev->data->dev_private;
 804         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 805
 806         dev->data->promiscuous = 1;
 807         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 808         if (pmd->remote_if_index && !pmd->flow_isolate)
 809                 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
 810 }
 811
 812 static void
 813 tap_promisc_disable(struct rte_eth_dev *dev)
 814 {
 815         struct pmd_internals *pmd = dev->data->dev_private;
 816         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 817
 818         dev->data->promiscuous = 0;
 819         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 820         if (pmd->remote_if_index && !pmd->flow_isolate)
 821                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
 822 }
 823
 824 static void
 825 tap_allmulti_enable(struct rte_eth_dev *dev)
 826 {
 827         struct pmd_internals *pmd = dev->data->dev_private;
 828         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 829
 830         dev->data->all_multicast = 1;
 831         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 832         if (pmd->remote_if_index && !pmd->flow_isolate)
 833                 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
 834 }
 835
 836 static void
 837 tap_allmulti_disable(struct rte_eth_dev *dev)
 838 {
 839         struct pmd_internals *pmd = dev->data->dev_private;
 840         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 841
 842         dev->data->all_multicast = 0;
 843         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 844         if (pmd->remote_if_index && !pmd->flow_isolate)
 845                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
 846 }
 847
 848 static void
 849 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 850 {
 851         struct pmd_internals *pmd = dev->data->dev_private;
 852         enum ioctl_mode mode = LOCAL_ONLY;
 853         struct ifreq ifr;
 854
 855         if (is_zero_ether_addr(mac_addr)) {
 856                 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
 857                         dev->device->name);
 858                 return;
 859         }
 860         /* Check the actual current MAC address on the tap netdevice */
 861         if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
 862                 return;
 863         if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 864                                mac_addr))
 865                 return;
 866         /* Check the current MAC address on the remote */
 867         if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0)
 868                 return;
 869         if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 870                                mac_addr))
 871                 mode = LOCAL_AND_REMOTE;
 872         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
 873         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
 874         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode) < 0)
 875                 return;
 876         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
 877         if (pmd->remote_if_index && !pmd->flow_isolate) {
 878                 /* Replace MAC redirection rule after a MAC change */
 879                 if (tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC) < 0) {
 880                         RTE_LOG(ERR, PMD,
 881                                 "%s: Couldn't delete MAC redirection rule\n",
 882                                 dev->device->name);
 883                         return;
 884                 }
 885                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0)
 886                         RTE_LOG(ERR, PMD,
 887                                 "%s: Couldn't add MAC redirection rule\n",
 888                                 dev->device->name);
 889         }
 890 }
 891
 892 static int
 893 tap_setup_queue(struct rte_eth_dev *dev,
 894                 struct pmd_internals *internals,
 895                 uint16_t qid,
 896                 int is_rx)
 897 {
 898         int *fd;
 899         int *other_fd;
 900         const char *dir;
 901         struct pmd_internals *pmd = dev->data->dev_private;
 902         struct rx_queue *rx = &internals->rxq[qid];
 903         struct tx_queue *tx = &internals->txq[qid];
 904
 905         if (is_rx) {
 906                 fd = &rx->fd;
 907                 other_fd = &tx->fd;
 908                 dir = "rx";
 909         } else {
 910                 fd = &tx->fd;
 911                 other_fd = &rx->fd;
 912                 dir = "tx";
 913         }
 914         if (*fd != -1) {
 915                 /* fd for this queue already exists */
 916                 RTE_LOG(DEBUG, PMD, "%s: fd %d for %s queue qid %d exists\n",
 917                         pmd->name, *fd, dir, qid);
 918         } else if (*other_fd != -1) {
 919                 /* Only other_fd exists. dup it */
 920                 *fd = dup(*other_fd);
 921                 if (*fd < 0) {
 922                         *fd = -1;
 923                         RTE_LOG(ERR, PMD, "%s: dup() failed.\n",
 924                                 pmd->name);
 925                         return -1;
 926                 }
 927                 RTE_LOG(DEBUG, PMD, "%s: dup fd %d for %s queue qid %d (%d)\n",
 928                         pmd->name, *other_fd, dir, qid, *fd);
 929         } else {
 930                 /* Both RX and TX fds do not exist (equal -1). Create fd */
 931                 *fd = tun_alloc(pmd);
 932                 if (*fd < 0) {
 933                         *fd = -1; /* restore original value */
 934                         RTE_LOG(ERR, PMD, "%s: tun_alloc() failed.\n",
 935                                 pmd->name);
 936                         return -1;
 937                 }
 938                 RTE_LOG(DEBUG, PMD, "%s: add %s queue for qid %d fd %d\n",
 939                         pmd->name, dir, qid, *fd);
 940         }
 941
 942         tx->mtu = &dev->data->mtu;
 943         rx->rxmode = &dev->data->dev_conf.rxmode;
 944
 945         return *fd;
 946 }
 947
 948 static int
 949 tap_rx_queue_setup(struct rte_eth_dev *dev,
 950                    uint16_t rx_queue_id,
 951                    uint16_t nb_rx_desc,
 952                    unsigned int socket_id,
 953                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 954                    struct rte_mempool *mp)
 955 {
 956         struct pmd_internals *internals = dev->data->dev_private;
 957         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
 958         struct rte_mbuf **tmp = &rxq->pool;
 959         long iov_max = sysconf(_SC_IOV_MAX);
 960         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
 961         struct iovec (*iovecs)[nb_desc + 1];
 962         int data_off = RTE_PKTMBUF_HEADROOM;
 963         int ret = 0;
 964         int fd;
 965         int i;
 966
 967         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
 968                 RTE_LOG(WARNING, PMD,
 969                         "nb_rx_queues %d too small or mempool NULL\n",
 970                         dev->data->nb_rx_queues);
 971                 return -1;
 972         }
 973
 974         rxq->mp = mp;
 975         rxq->trigger_seen = 1; /* force initial burst */
 976         rxq->in_port = dev->data->port_id;
 977         rxq->nb_rx_desc = nb_desc;
 978         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
 979                                     socket_id);
 980         if (!iovecs) {
 981                 RTE_LOG(WARNING, PMD,
 982                         "%s: Couldn't allocate %d RX descriptors\n",
 983                         dev->device->name, nb_desc);
 984                 return -ENOMEM;
 985         }
 986         rxq->iovecs = iovecs;
 987
 988         dev->data->rx_queues[rx_queue_id] = rxq;
 989         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
 990         if (fd == -1) {
 991                 ret = fd;
 992                 goto error;
 993         }
 994
 995         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
 996         (*rxq->iovecs)[0].iov_base = &rxq->pi;
 997
 998         for (i = 1; i <= nb_desc; i++) {
 999                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1000                 if (!*tmp) {
1001                         RTE_LOG(WARNING, PMD,
1002                                 "%s: couldn't allocate memory for queue %d\n",
1003                                 dev->device->name, rx_queue_id);
1004                         ret = -ENOMEM;
1005                         goto error;
1006                 }
1007                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1008                 (*rxq->iovecs)[i].iov_base =
1009                         (char *)(*tmp)->buf_addr + data_off;
1010                 data_off = 0;
1011                 tmp = &(*tmp)->next;
1012         }
1013
1014         RTE_LOG(DEBUG, PMD, "  RX TAP device name %s, qid %d on fd %d\n",
1015                 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
1016
1017         return 0;
1018
1019 error:
1020         rte_pktmbuf_free(rxq->pool);
1021         rxq->pool = NULL;
1022         rte_free(rxq->iovecs);
1023         rxq->iovecs = NULL;
1024         return ret;
1025 }
1026
1027 static int
1028 tap_tx_queue_setup(struct rte_eth_dev *dev,
1029                    uint16_t tx_queue_id,
1030                    uint16_t nb_tx_desc __rte_unused,
1031                    unsigned int socket_id __rte_unused,
1032                    const struct rte_eth_txconf *tx_conf __rte_unused)
1033 {
1034         struct pmd_internals *internals = dev->data->dev_private;
1035         int ret;
1036
1037         if (tx_queue_id >= dev->data->nb_tx_queues)
1038                 return -1;
1039
1040         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1041         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1042         if (ret == -1)
1043                 return -1;
1044
1045         RTE_LOG(DEBUG, PMD, "  TX TAP device name %s, qid %d on fd %d\n",
1046                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd);
1047
1048         return 0;
1049 }
1050
1051 static int
1052 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1053 {
1054         struct pmd_internals *pmd = dev->data->dev_private;
1055         struct ifreq ifr = { .ifr_mtu = mtu };
1056         int err = 0;
1057
1058         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1059         if (!err)
1060                 dev->data->mtu = mtu;
1061
1062         return err;
1063 }
1064
1065 static int
1066 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1067                      struct ether_addr *mc_addr_set __rte_unused,
1068                      uint32_t nb_mc_addr __rte_unused)
1069 {
1070         /*
1071          * Nothing to do actually: the tap has no filtering whatsoever, every
1072          * packet is received.
1073          */
1074         return 0;
1075 }
1076
1077 static int
1078 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1079 {
1080         struct rte_eth_dev *dev = arg;
1081         struct pmd_internals *pmd = dev->data->dev_private;
1082         struct ifinfomsg *info = NLMSG_DATA(nh);
1083
1084         if (nh->nlmsg_type != RTM_NEWLINK ||
1085             (info->ifi_index != pmd->if_index &&
1086              info->ifi_index != pmd->remote_if_index))
1087                 return 0;
1088         return tap_link_update(dev, 0);
1089 }
1090
1091 static void
1092 tap_dev_intr_handler(void *cb_arg)
1093 {
1094         struct rte_eth_dev *dev = cb_arg;
1095         struct pmd_internals *pmd = dev->data->dev_private;
1096
1097         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1098 }
1099
1100 static int
1101 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1102 {
1103         struct pmd_internals *pmd = dev->data->dev_private;
1104
1105         /* In any case, disable interrupt if the conf is no longer there. */
1106         if (!dev->data->dev_conf.intr_conf.lsc) {
1107                 if (pmd->intr_handle.fd != -1) {
1108                         tap_nl_final(pmd->intr_handle.fd);
1109                         rte_intr_callback_unregister(&pmd->intr_handle,
1110                                 tap_dev_intr_handler, dev);
1111                 }
1112                 return 0;
1113         }
1114         if (set) {
1115                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1116                 if (unlikely(pmd->intr_handle.fd == -1))
1117                         return -EBADF;
1118                 return rte_intr_callback_register(
1119                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1120         }
1121         tap_nl_final(pmd->intr_handle.fd);
1122         return rte_intr_callback_unregister(&pmd->intr_handle,
1123                                             tap_dev_intr_handler, dev);
1124 }
1125
1126 static const uint32_t*
1127 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1128 {
1129         static const uint32_t ptypes[] = {
1130                 RTE_PTYPE_INNER_L2_ETHER,
1131                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1132                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1133                 RTE_PTYPE_INNER_L3_IPV4,
1134                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1135                 RTE_PTYPE_INNER_L3_IPV6,
1136                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1137                 RTE_PTYPE_INNER_L4_FRAG,
1138                 RTE_PTYPE_INNER_L4_UDP,
1139                 RTE_PTYPE_INNER_L4_TCP,
1140                 RTE_PTYPE_INNER_L4_SCTP,
1141                 RTE_PTYPE_L2_ETHER,
1142                 RTE_PTYPE_L2_ETHER_VLAN,
1143                 RTE_PTYPE_L2_ETHER_QINQ,
1144                 RTE_PTYPE_L3_IPV4,
1145                 RTE_PTYPE_L3_IPV4_EXT,
1146                 RTE_PTYPE_L3_IPV6_EXT,
1147                 RTE_PTYPE_L3_IPV6,
1148                 RTE_PTYPE_L4_FRAG,
1149                 RTE_PTYPE_L4_UDP,
1150                 RTE_PTYPE_L4_TCP,
1151                 RTE_PTYPE_L4_SCTP,
1152         };
1153
1154         return ptypes;
1155 }
1156
1157 static int
1158 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1159                   struct rte_eth_fc_conf *fc_conf)
1160 {
1161         fc_conf->mode = RTE_FC_NONE;
1162         return 0;
1163 }
1164
1165 static int
1166 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1167                   struct rte_eth_fc_conf *fc_conf)
1168 {
1169         if (fc_conf->mode != RTE_FC_NONE)
1170                 return -ENOTSUP;
1171         return 0;
1172 }
1173
1174 static const struct eth_dev_ops ops = {
1175         .dev_start              = tap_dev_start,
1176         .dev_stop               = tap_dev_stop,
1177         .dev_close              = tap_dev_close,
1178         .dev_configure          = tap_dev_configure,
1179         .dev_infos_get          = tap_dev_info,
1180         .rx_queue_setup         = tap_rx_queue_setup,
1181         .tx_queue_setup         = tap_tx_queue_setup,
1182         .rx_queue_release       = tap_rx_queue_release,
1183         .tx_queue_release       = tap_tx_queue_release,
1184         .flow_ctrl_get          = tap_flow_ctrl_get,
1185         .flow_ctrl_set          = tap_flow_ctrl_set,
1186         .link_update            = tap_link_update,
1187         .dev_set_link_up        = tap_link_set_up,
1188         .dev_set_link_down      = tap_link_set_down,
1189         .promiscuous_enable     = tap_promisc_enable,
1190         .promiscuous_disable    = tap_promisc_disable,
1191         .allmulticast_enable    = tap_allmulti_enable,
1192         .allmulticast_disable   = tap_allmulti_disable,
1193         .mac_addr_set           = tap_mac_set,
1194         .mtu_set                = tap_mtu_set,
1195         .set_mc_addr_list       = tap_set_mc_addr_list,
1196         .stats_get              = tap_stats_get,
1197         .stats_reset            = tap_stats_reset,
1198         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1199         .filter_ctrl            = tap_dev_filter_ctrl,
1200 };
1201
1202 static int
1203 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1204                    char *remote_iface, int fixed_mac_type)
1205 {
1206         int numa_node = rte_socket_id();
1207         struct rte_eth_dev *dev;
1208         struct pmd_internals *pmd;
1209         struct rte_eth_dev_data *data;
1210         struct ifreq ifr;
1211         int i;
1212
1213         RTE_LOG(DEBUG, PMD, "  TAP device on numa %u\n", rte_socket_id());
1214
1215         data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
1216         if (!data) {
1217                 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
1218                 goto error_exit;
1219         }
1220
1221         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1222         if (!dev) {
1223                 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
1224                 goto error_exit;
1225         }
1226
1227         pmd = dev->data->dev_private;
1228         pmd->dev = dev;
1229         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1230
1231         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1232         if (pmd->ioctl_sock == -1) {
1233                 RTE_LOG(ERR, PMD,
1234                         "TAP Unable to get a socket for management: %s\n",
1235                         strerror(errno));
1236                 goto error_exit;
1237         }
1238
1239         /* Setup some default values */
1240         rte_memcpy(data, dev->data, sizeof(*data));
1241         data->dev_private = pmd;
1242         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1243         data->numa_node = numa_node;
1244
1245         data->dev_link = pmd_link;
1246         data->mac_addrs = &pmd->eth_addr;
1247         /* Set the number of RX and TX queues */
1248         data->nb_rx_queues = 0;
1249         data->nb_tx_queues = 0;
1250
1251         dev->data = data;
1252         dev->dev_ops = &ops;
1253         dev->rx_pkt_burst = pmd_rx_burst;
1254         dev->tx_pkt_burst = pmd_tx_burst;
1255
1256         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1257         pmd->intr_handle.fd = -1;
1258
1259         /* Presetup the fds to -1 as being not valid */
1260         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1261                 pmd->rxq[i].fd = -1;
1262                 pmd->txq[i].fd = -1;
1263         }
1264
1265         if (fixed_mac_type) {
1266                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1267                 static int iface_idx;
1268                 char mac[ETHER_ADDR_LEN] = "\0dtap";
1269
1270                 mac[ETHER_ADDR_LEN - 1] = iface_idx++;
1271                 rte_memcpy(&pmd->eth_addr, mac, ETHER_ADDR_LEN);
1272         } else {
1273                 eth_random_addr((uint8_t *)&pmd->eth_addr);
1274         }
1275
1276         /* Immediately create the netdevice (this will create the 1st queue). */
1277         /* rx queue */
1278         if (tap_setup_queue(dev, pmd, 0, 1) == -1)
1279                 goto error_exit;
1280         /* tx queue */
1281         if (tap_setup_queue(dev, pmd, 0, 0) == -1)
1282                 goto error_exit;
1283
1284         ifr.ifr_mtu = dev->data->mtu;
1285         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1286                 goto error_exit;
1287
1288         memset(&ifr, 0, sizeof(struct ifreq));
1289         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1290         rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr, ETHER_ADDR_LEN);
1291         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1292                 goto error_exit;
1293
1294         /*
1295          * Set up everything related to rte_flow:
1296          * - netlink socket
1297          * - tap / remote if_index
1298          * - mandatory QDISCs
1299          * - rte_flow actual/implicit lists
1300          * - implicit rules
1301          */
1302         pmd->nlsk_fd = tap_nl_init(0);
1303         if (pmd->nlsk_fd == -1) {
1304                 RTE_LOG(WARNING, PMD, "%s: failed to create netlink socket.\n",
1305                         pmd->name);
1306                 goto disable_rte_flow;
1307         }
1308         pmd->if_index = if_nametoindex(pmd->name);
1309         if (!pmd->if_index) {
1310                 RTE_LOG(ERR, PMD, "%s: failed to get if_index.\n", pmd->name);
1311                 goto disable_rte_flow;
1312         }
1313         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1314                 RTE_LOG(ERR, PMD, "%s: failed to create multiq qdisc.\n",
1315                         pmd->name);
1316                 goto disable_rte_flow;
1317         }
1318         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1319                 RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1320                         pmd->name);
1321                 goto disable_rte_flow;
1322         }
1323         LIST_INIT(&pmd->flows);
1324
1325         if (strlen(remote_iface)) {
1326                 pmd->remote_if_index = if_nametoindex(remote_iface);
1327                 if (!pmd->remote_if_index) {
1328                         RTE_LOG(ERR, PMD, "%s: failed to get %s if_index.\n",
1329                                 pmd->name, remote_iface);
1330                         goto error_remote;
1331                 }
1332                 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1333                          "%s", remote_iface);
1334
1335                 /* Save state of remote device */
1336                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1337
1338                 /* Replicate remote MAC address */
1339                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1340                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1341                                 pmd->name, pmd->remote_iface);
1342                         goto error_remote;
1343                 }
1344                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1345                            ETHER_ADDR_LEN);
1346                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1347                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1348                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1349                                 pmd->name, remote_iface);
1350                         goto error_remote;
1351                 }
1352
1353                 /*
1354                  * Flush usually returns negative value because it tries to
1355                  * delete every QDISC (and on a running device, one QDISC at
1356                  * least is needed). Ignore negative return value.
1357                  */
1358                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1359                 if (qdisc_create_ingress(pmd->nlsk_fd,
1360                                          pmd->remote_if_index) < 0) {
1361                         RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1362                                 pmd->remote_iface);
1363                         goto error_remote;
1364                 }
1365                 LIST_INIT(&pmd->implicit_flows);
1366                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1367                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1368                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1369                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1370                         RTE_LOG(ERR, PMD,
1371                                 "%s: failed to create implicit rules.\n",
1372                                 pmd->name);
1373                         goto error_remote;
1374                 }
1375         }
1376
1377         return 0;
1378
1379 disable_rte_flow:
1380         RTE_LOG(ERR, PMD, " Disabling rte flow support: %s(%d)\n",
1381                 strerror(errno), errno);
1382         if (strlen(remote_iface)) {
1383                 RTE_LOG(ERR, PMD, "Remote feature requires flow support.\n");
1384                 goto error_exit;
1385         }
1386         return 0;
1387
1388 error_remote:
1389         RTE_LOG(ERR, PMD, " Can't set up remote feature: %s(%d)\n",
1390                 strerror(errno), errno);
1391         tap_flow_implicit_flush(pmd, NULL);
1392
1393 error_exit:
1394         RTE_LOG(ERR, PMD, "TAP Unable to initialize %s\n",
1395                 rte_vdev_device_name(vdev));
1396
1397         rte_free(data);
1398         return -EINVAL;
1399 }
1400
1401 static int
1402 set_interface_name(const char *key __rte_unused,
1403                    const char *value,
1404                    void *extra_args)
1405 {
1406         char *name = (char *)extra_args;
1407
1408         if (value)
1409                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
1410         else
1411                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1412                          DEFAULT_TAP_NAME, (tap_unit - 1));
1413
1414         return 0;
1415 }
1416
1417 static int
1418 set_interface_speed(const char *key __rte_unused,
1419                     const char *value,
1420                     void *extra_args)
1421 {
1422         *(int *)extra_args = (value) ? atoi(value) : ETH_SPEED_NUM_10G;
1423
1424         return 0;
1425 }
1426
1427 static int
1428 set_remote_iface(const char *key __rte_unused,
1429                  const char *value,
1430                  void *extra_args)
1431 {
1432         char *name = (char *)extra_args;
1433
1434         if (value)
1435                 snprintf(name, RTE_ETH_NAME_MAX_LEN, "%s", value);
1436
1437         return 0;
1438 }
1439
1440 static int
1441 set_mac_type(const char *key __rte_unused,
1442              const char *value,
1443              void *extra_args)
1444 {
1445         if (value &&
1446             !strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED)))
1447                 *(int *)extra_args = 1;
1448         return 0;
1449 }
1450
1451 /* Open a TAP interface device.
1452  */
1453 static int
1454 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1455 {
1456         const char *name, *params;
1457         int ret;
1458         struct rte_kvargs *kvlist = NULL;
1459         int speed;
1460         char tap_name[RTE_ETH_NAME_MAX_LEN];
1461         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1462         int fixed_mac_type = 0;
1463
1464         name = rte_vdev_device_name(dev);
1465         params = rte_vdev_device_args(dev);
1466
1467         speed = ETH_SPEED_NUM_10G;
1468         snprintf(tap_name, sizeof(tap_name), "%s%d",
1469                  DEFAULT_TAP_NAME, tap_unit++);
1470         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1471
1472         if (params && (params[0] != '\0')) {
1473                 RTE_LOG(DEBUG, PMD, "parameters (%s)\n", params);
1474
1475                 kvlist = rte_kvargs_parse(params, valid_arguments);
1476                 if (kvlist) {
1477                         if (rte_kvargs_count(kvlist, ETH_TAP_SPEED_ARG) == 1) {
1478                                 ret = rte_kvargs_process(kvlist,
1479                                                          ETH_TAP_SPEED_ARG,
1480                                                          &set_interface_speed,
1481                                                          &speed);
1482                                 if (ret == -1)
1483                                         goto leave;
1484                         }
1485
1486                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1487                                 ret = rte_kvargs_process(kvlist,
1488                                                          ETH_TAP_IFACE_ARG,
1489                                                          &set_interface_name,
1490                                                          tap_name);
1491                                 if (ret == -1)
1492                                         goto leave;
1493                         }
1494
1495                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1496                                 ret = rte_kvargs_process(kvlist,
1497                                                          ETH_TAP_REMOTE_ARG,
1498                                                          &set_remote_iface,
1499                                                          remote_iface);
1500                                 if (ret == -1)
1501                                         goto leave;
1502                         }
1503
1504                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1505                                 ret = rte_kvargs_process(kvlist,
1506                                                          ETH_TAP_MAC_ARG,
1507                                                          &set_mac_type,
1508                                                          &fixed_mac_type);
1509                                 if (ret == -1)
1510                                         goto leave;
1511                         }
1512                 }
1513         }
1514         pmd_link.link_speed = speed;
1515
1516         RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1517                 name, tap_name);
1518
1519         ret = eth_dev_tap_create(dev, tap_name, remote_iface, fixed_mac_type);
1520
1521 leave:
1522         if (ret == -1) {
1523                 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1524                         name, tap_name);
1525                 tap_unit--;             /* Restore the unit number */
1526         }
1527         rte_kvargs_free(kvlist);
1528
1529         return ret;
1530 }
1531
1532 /* detach a TAP device.
1533  */
1534 static int
1535 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1536 {
1537         struct rte_eth_dev *eth_dev = NULL;
1538         struct pmd_internals *internals;
1539         int i;
1540
1541         RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1542                 rte_socket_id());
1543
1544         /* find the ethdev entry */
1545         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1546         if (!eth_dev)
1547                 return 0;
1548
1549         internals = eth_dev->data->dev_private;
1550         if (internals->nlsk_fd) {
1551                 tap_flow_flush(eth_dev, NULL);
1552                 tap_flow_implicit_flush(internals, NULL);
1553                 tap_nl_final(internals->nlsk_fd);
1554         }
1555         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1556                 if (internals->rxq[i].fd != -1) {
1557                         close(internals->rxq[i].fd);
1558                         internals->rxq[i].fd = -1;
1559                 }
1560                 if (internals->txq[i].fd != -1) {
1561                         close(internals->txq[i].fd);
1562                         internals->txq[i].fd = -1;
1563                 }
1564         }
1565
1566         close(internals->ioctl_sock);
1567         rte_free(eth_dev->data->dev_private);
1568         rte_free(eth_dev->data);
1569
1570         rte_eth_dev_release_port(eth_dev);
1571
1572         return 0;
1573 }
1574
1575 static struct rte_vdev_driver pmd_tap_drv = {
1576         .probe = rte_pmd_tap_probe,
1577         .remove = rte_pmd_tap_remove,
1578 };
1579 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1580 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1581 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1582                               ETH_TAP_IFACE_ARG "=<string> "
1583                               ETH_TAP_SPEED_ARG "=<int> "
1584                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_FIXED " "
1585                               ETH_TAP_REMOTE_ARG "=<string>");