drivers/net/tap/rte_eth_tap.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2016-2017 Intel Corporation
   3  */
   4
   5 #include <rte_atomic.h>
   6 #include <rte_branch_prediction.h>
   7 #include <rte_byteorder.h>
   8 #include <rte_common.h>
   9 #include <rte_mbuf.h>
  10 #include <rte_ethdev_driver.h>
  11 #include <rte_ethdev_vdev.h>
  12 #include <rte_malloc.h>
  13 #include <rte_bus_vdev.h>
  14 #include <rte_kvargs.h>
  15 #include <rte_net.h>
  16 #include <rte_debug.h>
  17 #include <rte_ip.h>
  18 #include <rte_string_fns.h>
  19 #include <rte_ethdev.h>
  20 #include <rte_errno.h>
  21
  22 #include <assert.h>
  23 #include <sys/types.h>
  24 #include <sys/stat.h>
  25 #include <sys/socket.h>
  26 #include <sys/ioctl.h>
  27 #include <sys/utsname.h>
  28 #include <sys/mman.h>
  29 #include <errno.h>
  30 #include <signal.h>
  31 #include <stdbool.h>
  32 #include <stdint.h>
  33 #include <sys/uio.h>
  34 #include <unistd.h>
  35 #include <arpa/inet.h>
  36 #include <net/if.h>
  37 #include <linux/if_tun.h>
  38 #include <linux/if_ether.h>
  39 #include <fcntl.h>
  40 #include <ctype.h>
  41
  42 #include <tap_rss.h>
  43 #include <rte_eth_tap.h>
  44 #include <tap_flow.h>
  45 #include <tap_netlink.h>
  46 #include <tap_tcmsgs.h>
  47
  48 /* Linux based path to the TUN device */
  49 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  50 #define DEFAULT_TAP_NAME        "dtap"
  51 #define DEFAULT_TUN_NAME        "dtun"
  52
  53 #define ETH_TAP_IFACE_ARG       "iface"
  54 #define ETH_TAP_REMOTE_ARG      "remote"
  55 #define ETH_TAP_MAC_ARG         "mac"
  56 #define ETH_TAP_MAC_FIXED       "fixed"
  57
  58 #define ETH_TAP_USR_MAC_FMT     "xx:xx:xx:xx:xx:xx"
  59 #define ETH_TAP_CMP_MAC_FMT     "0123456789ABCDEFabcdef"
  60 #define ETH_TAP_MAC_ARG_FMT     ETH_TAP_MAC_FIXED "|" ETH_TAP_USR_MAC_FMT
  61
  62 #define TAP_GSO_MBUFS_PER_CORE  128
  63 #define TAP_GSO_MBUF_SEG_SIZE   128
  64 #define TAP_GSO_MBUF_CACHE_SIZE 4
  65 #define TAP_GSO_MBUFS_NUM \
  66         (TAP_GSO_MBUFS_PER_CORE * TAP_GSO_MBUF_CACHE_SIZE)
  67
  68 /* IPC key for queue fds sync */
  69 #define TAP_MP_KEY "tap_mp_sync_queues"
  70
  71 static int tap_devices_count;
  72 static struct rte_vdev_driver pmd_tap_drv;
  73 static struct rte_vdev_driver pmd_tun_drv;
  74
  75 static const char *valid_arguments[] = {
  76         ETH_TAP_IFACE_ARG,
  77         ETH_TAP_REMOTE_ARG,
  78         ETH_TAP_MAC_ARG,
  79         NULL
  80 };
  81
  82 static char tuntap_name[8];
  83
  84 static volatile uint32_t tap_trigger;   /* Rx trigger */
  85
  86 static struct rte_eth_link pmd_link = {
  87         .link_speed = ETH_SPEED_NUM_10G,
  88         .link_duplex = ETH_LINK_FULL_DUPLEX,
  89         .link_status = ETH_LINK_DOWN,
  90         .link_autoneg = ETH_LINK_FIXED,
  91 };
  92
  93 static void
  94 tap_trigger_cb(int sig __rte_unused)
  95 {
  96         /* Valid trigger values are nonzero */
  97         tap_trigger = (tap_trigger + 1) | 0x80000000;
  98 }
  99
 100 /* Specifies on what netdevices the ioctl should be applied */
 101 enum ioctl_mode {
 102         LOCAL_AND_REMOTE,
 103         LOCAL_ONLY,
 104         REMOTE_ONLY,
 105 };
 106
 107 /* Message header to synchronize queues via IPC */
 108 struct ipc_queues {
 109         char port_name[RTE_DEV_NAME_MAX_LEN];
 110         int rxq_count;
 111         int txq_count;
 112         /*
 113          * The file descriptors are in the dedicated part
 114          * of the Unix message to be translated by the kernel.
 115          */
 116 };
 117
 118 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
 119
 120 /**
 121  * Tun/Tap allocation routine
 122  *
 123  * @param[in] pmd
 124  *   Pointer to private structure.
 125  *
 126  * @param[in] is_keepalive
 127  *   Keepalive flag
 128  *
 129  * @return
 130  *   -1 on failure, fd on success
 131  */
 132 static int
 133 tun_alloc(struct pmd_internals *pmd, int is_keepalive)
 134 {
 135         struct ifreq ifr;
 136 #ifdef IFF_MULTI_QUEUE
 137         unsigned int features;
 138 #endif
 139         int fd;
 140
 141         memset(&ifr, 0, sizeof(struct ifreq));
 142
 143         /*
 144          * Do not set IFF_NO_PI as packet information header will be needed
 145          * to check if a received packet has been truncated.
 146          */
 147         ifr.ifr_flags = (pmd->type == ETH_TUNTAP_TYPE_TAP) ?
 148                 IFF_TAP : IFF_TUN | IFF_POINTOPOINT;
 149         strlcpy(ifr.ifr_name, pmd->name, IFNAMSIZ);
 150
 151         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 152         if (fd < 0) {
 153                 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
 154                 goto error;
 155         }
 156
 157 #ifdef IFF_MULTI_QUEUE
 158         /* Grab the TUN features to verify we can work multi-queue */
 159         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 160                 TAP_LOG(ERR, "%s unable to get TUN/TAP features",
 161                         tuntap_name);
 162                 goto error;
 163         }
 164         TAP_LOG(DEBUG, "%s Features %08x", tuntap_name, features);
 165
 166         if (features & IFF_MULTI_QUEUE) {
 167                 TAP_LOG(DEBUG, "  Multi-queue support for %d queues",
 168                         RTE_PMD_TAP_MAX_QUEUES);
 169                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 170         } else
 171 #endif
 172         {
 173                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 174                 TAP_LOG(DEBUG, "  Single queue only support");
 175         }
 176
 177         /* Set the TUN/TAP configuration and set the name if needed */
 178         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 179                 TAP_LOG(WARNING, "Unable to set TUNSETIFF for %s: %s",
 180                         ifr.ifr_name, strerror(errno));
 181                 goto error;
 182         }
 183
 184         /*
 185          * Name passed to kernel might be wildcard like dtun%d
 186          * and need to find the resulting device.
 187          */
 188         TAP_LOG(DEBUG, "Device name is '%s'", ifr.ifr_name);
 189         strlcpy(pmd->name, ifr.ifr_name, RTE_ETH_NAME_MAX_LEN);
 190
 191         if (is_keepalive) {
 192                 /*
 193                  * Detach the TUN/TAP keep-alive queue
 194                  * to avoid traffic through it
 195                  */
 196                 ifr.ifr_flags = IFF_DETACH_QUEUE;
 197                 if (ioctl(fd, TUNSETQUEUE, (void *)&ifr) < 0) {
 198                         TAP_LOG(WARNING,
 199                                 "Unable to detach keep-alive queue for %s: %s",
 200                                 ifr.ifr_name, strerror(errno));
 201                         goto error;
 202                 }
 203         }
 204
 205         /* Always set the file descriptor to non-blocking */
 206         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 207                 TAP_LOG(WARNING,
 208                         "Unable to set %s to nonblocking: %s",
 209                         ifr.ifr_name, strerror(errno));
 210                 goto error;
 211         }
 212
 213         /* Set up trigger to optimize empty Rx bursts */
 214         errno = 0;
 215         do {
 216                 struct sigaction sa;
 217                 int flags = fcntl(fd, F_GETFL);
 218
 219                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 220                         break;
 221                 if (sa.sa_handler != tap_trigger_cb) {
 222                         /*
 223                          * Make sure SIGIO is not already taken. This is done
 224                          * as late as possible to leave the application a
 225                          * chance to set up its own signal handler first.
 226                          */
 227                         if (sa.sa_handler != SIG_IGN &&
 228                             sa.sa_handler != SIG_DFL) {
 229                                 errno = EBUSY;
 230                                 break;
 231                         }
 232                         sa = (struct sigaction){
 233                                 .sa_flags = SA_RESTART,
 234                                 .sa_handler = tap_trigger_cb,
 235                         };
 236                         if (sigaction(SIGIO, &sa, NULL) == -1)
 237                                 break;
 238                 }
 239                 /* Enable SIGIO on file descriptor */
 240                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 241                 fcntl(fd, F_SETOWN, getpid());
 242         } while (0);
 243
 244         if (errno) {
 245                 /* Disable trigger globally in case of error */
 246                 tap_trigger = 0;
 247                 TAP_LOG(WARNING, "Rx trigger disabled: %s",
 248                         strerror(errno));
 249         }
 250
 251         return fd;
 252
 253 error:
 254         if (fd >= 0)
 255                 close(fd);
 256         return -1;
 257 }
 258
 259 static void
 260 tap_verify_csum(struct rte_mbuf *mbuf)
 261 {
 262         uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
 263         uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
 264         uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
 265         unsigned int l2_len = sizeof(struct ether_hdr);
 266         unsigned int l3_len;
 267         uint16_t cksum = 0;
 268         void *l3_hdr;
 269         void *l4_hdr;
 270
 271         if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
 272                 l2_len += 4;
 273         else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
 274                 l2_len += 8;
 275         /* Don't verify checksum for packets with discontinuous L2 header */
 276         if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
 277                      rte_pktmbuf_data_len(mbuf)))
 278                 return;
 279         l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
 280         if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 281                 struct ipv4_hdr *iph = l3_hdr;
 282
 283                 /* ihl contains the number of 4-byte words in the header */
 284                 l3_len = 4 * (iph->version_ihl & 0xf);
 285                 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
 286                         return;
 287                 /* check that the total length reported by header is not
 288                  * greater than the total received size
 289                  */
 290                 if (l2_len + rte_be_to_cpu_16(iph->total_length) >
 291                                 rte_pktmbuf_data_len(mbuf))
 292                         return;
 293
 294                 cksum = ~rte_raw_cksum(iph, l3_len);
 295                 mbuf->ol_flags |= cksum ?
 296                         PKT_RX_IP_CKSUM_BAD :
 297                         PKT_RX_IP_CKSUM_GOOD;
 298         } else if (l3 == RTE_PTYPE_L3_IPV6) {
 299                 struct ipv6_hdr *iph = l3_hdr;
 300
 301                 l3_len = sizeof(struct ipv6_hdr);
 302                 /* check that the total length reported by header is not
 303                  * greater than the total received size
 304                  */
 305                 if (l2_len + l3_len + rte_be_to_cpu_16(iph->payload_len) >
 306                                 rte_pktmbuf_data_len(mbuf))
 307                         return;
 308         } else {
 309                 /* IPv6 extensions are not supported */
 310                 return;
 311         }
 312         if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
 313                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 314                 /* Don't verify checksum for multi-segment packets. */
 315                 if (mbuf->nb_segs > 1)
 316                         return;
 317                 if (l3 == RTE_PTYPE_L3_IPV4)
 318                         cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
 319                 else if (l3 == RTE_PTYPE_L3_IPV6)
 320                         cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
 321                 mbuf->ol_flags |= cksum ?
 322                         PKT_RX_L4_CKSUM_BAD :
 323                         PKT_RX_L4_CKSUM_GOOD;
 324         }
 325 }
 326
 327 static uint64_t
 328 tap_rx_offload_get_port_capa(void)
 329 {
 330         /*
 331          * No specific port Rx offload capabilities.
 332          */
 333         return 0;
 334 }
 335
 336 static uint64_t
 337 tap_rx_offload_get_queue_capa(void)
 338 {
 339         return DEV_RX_OFFLOAD_SCATTER |
 340                DEV_RX_OFFLOAD_IPV4_CKSUM |
 341                DEV_RX_OFFLOAD_UDP_CKSUM |
 342                DEV_RX_OFFLOAD_TCP_CKSUM;
 343 }
 344
 345 /* Callback to handle the rx burst of packets to the correct interface and
 346  * file descriptor(s) in a multi-queue setup.
 347  */
 348 static uint16_t
 349 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 350 {
 351         struct rx_queue *rxq = queue;
 352         struct pmd_process_private *process_private;
 353         uint16_t num_rx;
 354         unsigned long num_rx_bytes = 0;
 355         uint32_t trigger = tap_trigger;
 356
 357         if (trigger == rxq->trigger_seen)
 358                 return 0;
 359         if (trigger)
 360                 rxq->trigger_seen = trigger;
 361         process_private = rte_eth_devices[rxq->in_port].process_private;
 362         rte_compiler_barrier();
 363         for (num_rx = 0; num_rx < nb_pkts; ) {
 364                 struct rte_mbuf *mbuf = rxq->pool;
 365                 struct rte_mbuf *seg = NULL;
 366                 struct rte_mbuf *new_tail = NULL;
 367                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 368                 int len;
 369
 370                 len = readv(process_private->rxq_fds[rxq->queue_id],
 371                         *rxq->iovecs,
 372                         1 + (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
 373                              rxq->nb_rx_desc : 1));
 374                 if (len < (int)sizeof(struct tun_pi))
 375                         break;
 376
 377                 /* Packet couldn't fit in the provided mbuf */
 378                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 379                         rxq->stats.ierrors++;
 380                         continue;
 381                 }
 382
 383                 len -= sizeof(struct tun_pi);
 384
 385                 mbuf->pkt_len = len;
 386                 mbuf->port = rxq->in_port;
 387                 while (1) {
 388                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 389
 390                         if (unlikely(!buf)) {
 391                                 rxq->stats.rx_nombuf++;
 392                                 /* No new buf has been allocated: do nothing */
 393                                 if (!new_tail || !seg)
 394                                         goto end;
 395
 396                                 seg->next = NULL;
 397                                 rte_pktmbuf_free(mbuf);
 398
 399                                 goto end;
 400                         }
 401                         seg = seg ? seg->next : mbuf;
 402                         if (rxq->pool == mbuf)
 403                                 rxq->pool = buf;
 404                         if (new_tail)
 405                                 new_tail->next = buf;
 406                         new_tail = buf;
 407                         new_tail->next = seg->next;
 408
 409                         /* iovecs[0] is reserved for packet info (pi) */
 410                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 411                                 buf->buf_len - data_off;
 412                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 413                                 (char *)buf->buf_addr + data_off;
 414
 415                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 416                         seg->data_off = data_off;
 417
 418                         len -= seg->data_len;
 419                         if (len <= 0)
 420                                 break;
 421                         mbuf->nb_segs++;
 422                         /* First segment has headroom, not the others */
 423                         data_off = 0;
 424                 }
 425                 seg->next = NULL;
 426                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 427                                                       RTE_PTYPE_ALL_MASK);
 428                 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
 429                         tap_verify_csum(mbuf);
 430
 431                 /* account for the receive frame */
 432                 bufs[num_rx++] = mbuf;
 433                 num_rx_bytes += mbuf->pkt_len;
 434         }
 435 end:
 436         rxq->stats.ipackets += num_rx;
 437         rxq->stats.ibytes += num_rx_bytes;
 438
 439         return num_rx;
 440 }
 441
 442 static uint64_t
 443 tap_tx_offload_get_port_capa(void)
 444 {
 445         /*
 446          * No specific port Tx offload capabilities.
 447          */
 448         return 0;
 449 }
 450
 451 static uint64_t
 452 tap_tx_offload_get_queue_capa(void)
 453 {
 454         return DEV_TX_OFFLOAD_MULTI_SEGS |
 455                DEV_TX_OFFLOAD_IPV4_CKSUM |
 456                DEV_TX_OFFLOAD_UDP_CKSUM |
 457                DEV_TX_OFFLOAD_TCP_CKSUM |
 458                DEV_TX_OFFLOAD_TCP_TSO;
 459 }
 460
 461 /* Finalize l4 checksum calculation */
 462 static void
 463 tap_tx_l4_cksum(uint16_t *l4_cksum, uint16_t l4_phdr_cksum,
 464                 uint32_t l4_raw_cksum)
 465 {
 466         if (l4_cksum) {
 467                 uint32_t cksum;
 468
 469                 cksum = __rte_raw_cksum_reduce(l4_raw_cksum);
 470                 cksum += l4_phdr_cksum;
 471
 472                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 473                 cksum = (~cksum) & 0xffff;
 474                 if (cksum == 0)
 475                         cksum = 0xffff;
 476                 *l4_cksum = cksum;
 477         }
 478 }
 479
 480 /* Accumaulate L4 raw checksums */
 481 static void
 482 tap_tx_l4_add_rcksum(char *l4_data, unsigned int l4_len, uint16_t *l4_cksum,
 483                         uint32_t *l4_raw_cksum)
 484 {
 485         if (l4_cksum == NULL)
 486                 return;
 487
 488         *l4_raw_cksum = __rte_raw_cksum(l4_data, l4_len, *l4_raw_cksum);
 489 }
 490
 491 /* L3 and L4 pseudo headers checksum offloads */
 492 static void
 493 tap_tx_l3_cksum(char *packet, uint64_t ol_flags, unsigned int l2_len,
 494                 unsigned int l3_len, unsigned int l4_len, uint16_t **l4_cksum,
 495                 uint16_t *l4_phdr_cksum, uint32_t *l4_raw_cksum)
 496 {
 497         void *l3_hdr = packet + l2_len;
 498
 499         if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
 500                 struct ipv4_hdr *iph = l3_hdr;
 501                 uint16_t cksum;
 502
 503                 iph->hdr_checksum = 0;
 504                 cksum = rte_raw_cksum(iph, l3_len);
 505                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 506         }
 507         if (ol_flags & PKT_TX_L4_MASK) {
 508                 void *l4_hdr;
 509
 510                 l4_hdr = packet + l2_len + l3_len;
 511                 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
 512                         *l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
 513                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 514                         *l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
 515                 else
 516                         return;
 517                 **l4_cksum = 0;
 518                 if (ol_flags & PKT_TX_IPV4)
 519                         *l4_phdr_cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 520                 else
 521                         *l4_phdr_cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 522                 *l4_raw_cksum = __rte_raw_cksum(l4_hdr, l4_len, 0);
 523         }
 524 }
 525
 526 static inline void
 527 tap_write_mbufs(struct tx_queue *txq, uint16_t num_mbufs,
 528                         struct rte_mbuf **pmbufs,
 529                         uint16_t *num_packets, unsigned long *num_tx_bytes)
 530 {
 531         int i;
 532         uint16_t l234_hlen;
 533         struct pmd_process_private *process_private;
 534
 535         process_private = rte_eth_devices[txq->out_port].process_private;
 536
 537         for (i = 0; i < num_mbufs; i++) {
 538                 struct rte_mbuf *mbuf = pmbufs[i];
 539                 struct iovec iovecs[mbuf->nb_segs + 2];
 540                 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
 541                 struct rte_mbuf *seg = mbuf;
 542                 char m_copy[mbuf->data_len];
 543                 int proto;
 544                 int n;
 545                 int j;
 546                 int k; /* current index in iovecs for copying segments */
 547                 uint16_t seg_len; /* length of first segment */
 548                 uint16_t nb_segs;
 549                 uint16_t *l4_cksum; /* l4 checksum (pseudo header + payload) */
 550                 uint32_t l4_raw_cksum = 0; /* TCP/UDP payload raw checksum */
 551                 uint16_t l4_phdr_cksum = 0; /* TCP/UDP pseudo header checksum */
 552                 uint16_t is_cksum = 0; /* in case cksum should be offloaded */
 553
 554                 l4_cksum = NULL;
 555                 if (txq->type == ETH_TUNTAP_TYPE_TUN) {
 556                         /*
 557                          * TUN and TAP are created with IFF_NO_PI disabled.
 558                          * For TUN PMD this mandatory as fields are used by
 559                          * Kernel tun.c to determine whether its IP or non IP
 560                          * packets.
 561                          *
 562                          * The logic fetches the first byte of data from mbuf
 563                          * then compares whether its v4 or v6. If first byte
 564                          * is 4 or 6, then protocol field is updated.
 565                          */
 566                         char *buff_data = rte_pktmbuf_mtod(seg, void *);
 567                         proto = (*buff_data & 0xf0);
 568                         pi.proto = (proto == 0x40) ?
 569                                 rte_cpu_to_be_16(ETHER_TYPE_IPv4) :
 570                                 ((proto == 0x60) ?
 571                                         rte_cpu_to_be_16(ETHER_TYPE_IPv6) :
 572                                         0x00);
 573                 }
 574
 575                 k = 0;
 576                 iovecs[k].iov_base = &pi;
 577                 iovecs[k].iov_len = sizeof(pi);
 578                 k++;
 579
 580                 nb_segs = mbuf->nb_segs;
 581                 if (txq->csum &&
 582                     ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 583                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 584                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
 585                         is_cksum = 1;
 586
 587                         /* Support only packets with at least layer 4
 588                          * header included in the first segment
 589                          */
 590                         seg_len = rte_pktmbuf_data_len(mbuf);
 591                         l234_hlen = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
 592                         if (seg_len < l234_hlen)
 593                                 break;
 594
 595                         /* To change checksums, work on a * copy of l2, l3
 596                          * headers + l4 pseudo header
 597                          */
 598                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 599                                         l234_hlen);
 600                         tap_tx_l3_cksum(m_copy, mbuf->ol_flags,
 601                                        mbuf->l2_len, mbuf->l3_len, mbuf->l4_len,
 602                                        &l4_cksum, &l4_phdr_cksum,
 603                                        &l4_raw_cksum);
 604                         iovecs[k].iov_base = m_copy;
 605                         iovecs[k].iov_len = l234_hlen;
 606                         k++;
 607
 608                         /* Update next iovecs[] beyond l2, l3, l4 headers */
 609                         if (seg_len > l234_hlen) {
 610                                 iovecs[k].iov_len = seg_len - l234_hlen;
 611                                 iovecs[k].iov_base =
 612                                         rte_pktmbuf_mtod(seg, char *) +
 613                                                 l234_hlen;
 614                                 tap_tx_l4_add_rcksum(iovecs[k].iov_base,
 615                                         iovecs[k].iov_len, l4_cksum,
 616                                         &l4_raw_cksum);
 617                                 k++;
 618                                 nb_segs++;
 619                         }
 620                         seg = seg->next;
 621                 }
 622
 623                 for (j = k; j <= nb_segs; j++) {
 624                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 625                         iovecs[j].iov_base = rte_pktmbuf_mtod(seg, void *);
 626                         if (is_cksum)
 627                                 tap_tx_l4_add_rcksum(iovecs[j].iov_base,
 628                                         iovecs[j].iov_len, l4_cksum,
 629                                         &l4_raw_cksum);
 630                         seg = seg->next;
 631                 }
 632
 633                 if (is_cksum)
 634                         tap_tx_l4_cksum(l4_cksum, l4_phdr_cksum, l4_raw_cksum);
 635
 636                 /* copy the tx frame data */
 637                 n = writev(process_private->txq_fds[txq->queue_id], iovecs, j);
 638                 if (n <= 0)
 639                         break;
 640                 (*num_packets)++;
 641                 (*num_tx_bytes) += rte_pktmbuf_pkt_len(mbuf);
 642         }
 643 }
 644
 645 /* Callback to handle sending packets from the tap interface
 646  */
 647 static uint16_t
 648 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 649 {
 650         struct tx_queue *txq = queue;
 651         uint16_t num_tx = 0;
 652         uint16_t num_packets = 0;
 653         unsigned long num_tx_bytes = 0;
 654         uint32_t max_size;
 655         int i;
 656
 657         if (unlikely(nb_pkts == 0))
 658                 return 0;
 659
 660         struct rte_mbuf *gso_mbufs[MAX_GSO_MBUFS];
 661         max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
 662         for (i = 0; i < nb_pkts; i++) {
 663                 struct rte_mbuf *mbuf_in = bufs[num_tx];
 664                 struct rte_mbuf **mbuf;
 665                 uint16_t num_mbufs = 0;
 666                 uint16_t tso_segsz = 0;
 667                 int ret;
 668                 uint16_t hdrs_len;
 669                 int j;
 670                 uint64_t tso;
 671
 672                 tso = mbuf_in->ol_flags & PKT_TX_TCP_SEG;
 673                 if (tso) {
 674                         struct rte_gso_ctx *gso_ctx = &txq->gso_ctx;
 675
 676                         assert(gso_ctx != NULL);
 677
 678                         /* TCP segmentation implies TCP checksum offload */
 679                         mbuf_in->ol_flags |= PKT_TX_TCP_CKSUM;
 680
 681                         /* gso size is calculated without ETHER_CRC_LEN */
 682                         hdrs_len = mbuf_in->l2_len + mbuf_in->l3_len +
 683                                         mbuf_in->l4_len;
 684                         tso_segsz = mbuf_in->tso_segsz + hdrs_len;
 685                         if (unlikely(tso_segsz == hdrs_len) ||
 686                                 tso_segsz > *txq->mtu) {
 687                                 txq->stats.errs++;
 688                                 break;
 689                         }
 690                         gso_ctx->gso_size = tso_segsz;
 691                         ret = rte_gso_segment(mbuf_in, /* packet to segment */
 692                                 gso_ctx, /* gso control block */
 693                                 (struct rte_mbuf **)&gso_mbufs, /* out mbufs */
 694                                 RTE_DIM(gso_mbufs)); /* max tso mbufs */
 695
 696                         /* ret contains the number of new created mbufs */
 697                         if (ret < 0)
 698                                 break;
 699
 700                         mbuf = gso_mbufs;
 701                         num_mbufs = ret;
 702                 } else {
 703                         /* stats.errs will be incremented */
 704                         if (rte_pktmbuf_pkt_len(mbuf_in) > max_size)
 705                                 break;
 706
 707                         /* ret 0 indicates no new mbufs were created */
 708                         ret = 0;
 709                         mbuf = &mbuf_in;
 710                         num_mbufs = 1;
 711                 }
 712
 713                 tap_write_mbufs(txq, num_mbufs, mbuf,
 714                                 &num_packets, &num_tx_bytes);
 715                 num_tx++;
 716                 /* free original mbuf */
 717                 rte_pktmbuf_free(mbuf_in);
 718                 /* free tso mbufs */
 719                 for (j = 0; j < ret; j++)
 720                         rte_pktmbuf_free(mbuf[j]);
 721         }
 722
 723         txq->stats.opackets += num_packets;
 724         txq->stats.errs += nb_pkts - num_tx;
 725         txq->stats.obytes += num_tx_bytes;
 726
 727         return num_packets;
 728 }
 729
 730 static const char *
 731 tap_ioctl_req2str(unsigned long request)
 732 {
 733         switch (request) {
 734         case SIOCSIFFLAGS:
 735                 return "SIOCSIFFLAGS";
 736         case SIOCGIFFLAGS:
 737                 return "SIOCGIFFLAGS";
 738         case SIOCGIFHWADDR:
 739                 return "SIOCGIFHWADDR";
 740         case SIOCSIFHWADDR:
 741                 return "SIOCSIFHWADDR";
 742         case SIOCSIFMTU:
 743                 return "SIOCSIFMTU";
 744         }
 745         return "UNKNOWN";
 746 }
 747
 748 static int
 749 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 750           struct ifreq *ifr, int set, enum ioctl_mode mode)
 751 {
 752         short req_flags = ifr->ifr_flags;
 753         int remote = pmd->remote_if_index &&
 754                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 755
 756         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 757                 return 0;
 758         /*
 759          * If there is a remote netdevice, apply ioctl on it, then apply it on
 760          * the tap netdevice.
 761          */
 762 apply:
 763         if (remote)
 764                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
 765         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 766                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 767         switch (request) {
 768         case SIOCSIFFLAGS:
 769                 /* fetch current flags to leave other flags untouched */
 770                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 771                         goto error;
 772                 if (set)
 773                         ifr->ifr_flags |= req_flags;
 774                 else
 775                         ifr->ifr_flags &= ~req_flags;
 776                 break;
 777         case SIOCGIFFLAGS:
 778         case SIOCGIFHWADDR:
 779         case SIOCSIFHWADDR:
 780         case SIOCSIFMTU:
 781                 break;
 782         default:
 783                 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
 784                         pmd->name);
 785                 return -EINVAL;
 786         }
 787         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 788                 goto error;
 789         if (remote-- && mode == LOCAL_AND_REMOTE)
 790                 goto apply;
 791         return 0;
 792
 793 error:
 794         TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
 795                 tap_ioctl_req2str(request), strerror(errno), errno);
 796         return -errno;
 797 }
 798
 799 static int
 800 tap_link_set_down(struct rte_eth_dev *dev)
 801 {
 802         struct pmd_internals *pmd = dev->data->dev_private;
 803         struct ifreq ifr = { .ifr_flags = IFF_UP };
 804
 805         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 806         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 807 }
 808
 809 static int
 810 tap_link_set_up(struct rte_eth_dev *dev)
 811 {
 812         struct pmd_internals *pmd = dev->data->dev_private;
 813         struct ifreq ifr = { .ifr_flags = IFF_UP };
 814
 815         dev->data->dev_link.link_status = ETH_LINK_UP;
 816         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 817 }
 818
 819 static int
 820 tap_dev_start(struct rte_eth_dev *dev)
 821 {
 822         int err, i;
 823
 824         err = tap_intr_handle_set(dev, 1);
 825         if (err)
 826                 return err;
 827
 828         err = tap_link_set_up(dev);
 829         if (err)
 830                 return err;
 831
 832         for (i = 0; i < dev->data->nb_tx_queues; i++)
 833                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 834         for (i = 0; i < dev->data->nb_rx_queues; i++)
 835                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 836
 837         return err;
 838 }
 839
 840 /* This function gets called when the current port gets stopped.
 841  */
 842 static void
 843 tap_dev_stop(struct rte_eth_dev *dev)
 844 {
 845         int i;
 846
 847         for (i = 0; i < dev->data->nb_tx_queues; i++)
 848                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 849         for (i = 0; i < dev->data->nb_rx_queues; i++)
 850                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 851
 852         tap_intr_handle_set(dev, 0);
 853         tap_link_set_down(dev);
 854 }
 855
 856 static int
 857 tap_dev_configure(struct rte_eth_dev *dev)
 858 {
 859         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 860                 TAP_LOG(ERR,
 861                         "%s: number of rx queues %d exceeds max num of queues %d",
 862                         dev->device->name,
 863                         dev->data->nb_rx_queues,
 864                         RTE_PMD_TAP_MAX_QUEUES);
 865                 return -1;
 866         }
 867         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 868                 TAP_LOG(ERR,
 869                         "%s: number of tx queues %d exceeds max num of queues %d",
 870                         dev->device->name,
 871                         dev->data->nb_tx_queues,
 872                         RTE_PMD_TAP_MAX_QUEUES);
 873                 return -1;
 874         }
 875
 876         TAP_LOG(INFO, "%s: %p: TX configured queues number: %u",
 877                 dev->device->name, (void *)dev, dev->data->nb_tx_queues);
 878
 879         TAP_LOG(INFO, "%s: %p: RX configured queues number: %u",
 880                 dev->device->name, (void *)dev, dev->data->nb_rx_queues);
 881
 882         return 0;
 883 }
 884
 885 static uint32_t
 886 tap_dev_speed_capa(void)
 887 {
 888         uint32_t speed = pmd_link.link_speed;
 889         uint32_t capa = 0;
 890
 891         if (speed >= ETH_SPEED_NUM_10M)
 892                 capa |= ETH_LINK_SPEED_10M;
 893         if (speed >= ETH_SPEED_NUM_100M)
 894                 capa |= ETH_LINK_SPEED_100M;
 895         if (speed >= ETH_SPEED_NUM_1G)
 896                 capa |= ETH_LINK_SPEED_1G;
 897         if (speed >= ETH_SPEED_NUM_5G)
 898                 capa |= ETH_LINK_SPEED_2_5G;
 899         if (speed >= ETH_SPEED_NUM_5G)
 900                 capa |= ETH_LINK_SPEED_5G;
 901         if (speed >= ETH_SPEED_NUM_10G)
 902                 capa |= ETH_LINK_SPEED_10G;
 903         if (speed >= ETH_SPEED_NUM_20G)
 904                 capa |= ETH_LINK_SPEED_20G;
 905         if (speed >= ETH_SPEED_NUM_25G)
 906                 capa |= ETH_LINK_SPEED_25G;
 907         if (speed >= ETH_SPEED_NUM_40G)
 908                 capa |= ETH_LINK_SPEED_40G;
 909         if (speed >= ETH_SPEED_NUM_50G)
 910                 capa |= ETH_LINK_SPEED_50G;
 911         if (speed >= ETH_SPEED_NUM_56G)
 912                 capa |= ETH_LINK_SPEED_56G;
 913         if (speed >= ETH_SPEED_NUM_100G)
 914                 capa |= ETH_LINK_SPEED_100G;
 915
 916         return capa;
 917 }
 918
 919 static void
 920 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 921 {
 922         struct pmd_internals *internals = dev->data->dev_private;
 923
 924         dev_info->if_index = internals->if_index;
 925         dev_info->max_mac_addrs = 1;
 926         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 927         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
 928         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
 929         dev_info->min_rx_bufsize = 0;
 930         dev_info->speed_capa = tap_dev_speed_capa();
 931         dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
 932         dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
 933                                     dev_info->rx_queue_offload_capa;
 934         dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
 935         dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
 936                                     dev_info->tx_queue_offload_capa;
 937         dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
 938         /*
 939          * limitation: TAP supports all of IP, UDP and TCP hash
 940          * functions together and not in partial combinations
 941          */
 942         dev_info->flow_type_rss_offloads = ~TAP_RSS_HF_MASK;
 943 }
 944
 945 static int
 946 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 947 {
 948         unsigned int i, imax;
 949         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 950         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 951         unsigned long rx_nombuf = 0, ierrors = 0;
 952         const struct pmd_internals *pmd = dev->data->dev_private;
 953
 954         /* rx queue statistics */
 955         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 956                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 957         for (i = 0; i < imax; i++) {
 958                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 959                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 960                 rx_total += tap_stats->q_ipackets[i];
 961                 rx_bytes_total += tap_stats->q_ibytes[i];
 962                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
 963                 ierrors += pmd->rxq[i].stats.ierrors;
 964         }
 965
 966         /* tx queue statistics */
 967         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 968                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 969
 970         for (i = 0; i < imax; i++) {
 971                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 972                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 973                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 974                 tx_total += tap_stats->q_opackets[i];
 975                 tx_err_total += tap_stats->q_errors[i];
 976                 tx_bytes_total += tap_stats->q_obytes[i];
 977         }
 978
 979         tap_stats->ipackets = rx_total;
 980         tap_stats->ibytes = rx_bytes_total;
 981         tap_stats->ierrors = ierrors;
 982         tap_stats->rx_nombuf = rx_nombuf;
 983         tap_stats->opackets = tx_total;
 984         tap_stats->oerrors = tx_err_total;
 985         tap_stats->obytes = tx_bytes_total;
 986         return 0;
 987 }
 988
 989 static void
 990 tap_stats_reset(struct rte_eth_dev *dev)
 991 {
 992         int i;
 993         struct pmd_internals *pmd = dev->data->dev_private;
 994
 995         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 996                 pmd->rxq[i].stats.ipackets = 0;
 997                 pmd->rxq[i].stats.ibytes = 0;
 998                 pmd->rxq[i].stats.ierrors = 0;
 999                 pmd->rxq[i].stats.rx_nombuf = 0;
1000
1001                 pmd->txq[i].stats.opackets = 0;
1002                 pmd->txq[i].stats.errs = 0;
1003                 pmd->txq[i].stats.obytes = 0;
1004         }
1005 }
1006
1007 static void
1008 tap_dev_close(struct rte_eth_dev *dev)
1009 {
1010         int i;
1011         struct pmd_internals *internals = dev->data->dev_private;
1012         struct pmd_process_private *process_private = dev->process_private;
1013
1014         tap_link_set_down(dev);
1015         tap_flow_flush(dev, NULL);
1016         tap_flow_implicit_flush(internals, NULL);
1017
1018         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1019                 if (process_private->rxq_fds[i] != -1) {
1020                         close(process_private->rxq_fds[i]);
1021                         process_private->rxq_fds[i] = -1;
1022                 }
1023                 if (process_private->txq_fds[i] != -1) {
1024                         close(process_private->txq_fds[i]);
1025                         process_private->txq_fds[i] = -1;
1026                 }
1027         }
1028
1029         if (internals->remote_if_index) {
1030                 /* Restore initial remote state */
1031                 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
1032                                 &internals->remote_initial_flags);
1033         }
1034
1035         if (internals->ka_fd != -1) {
1036                 close(internals->ka_fd);
1037                 internals->ka_fd = -1;
1038         }
1039         /*
1040          * Since TUN device has no more opened file descriptors
1041          * it will be removed from kernel
1042          */
1043 }
1044
1045 static void
1046 tap_rx_queue_release(void *queue)
1047 {
1048         struct rx_queue *rxq = queue;
1049         struct pmd_process_private *process_private;
1050
1051         if (!rxq)
1052                 return;
1053         process_private = rte_eth_devices[rxq->in_port].process_private;
1054         if (process_private->rxq_fds[rxq->queue_id] > 0) {
1055                 close(process_private->rxq_fds[rxq->queue_id]);
1056                 process_private->rxq_fds[rxq->queue_id] = -1;
1057                 rte_pktmbuf_free(rxq->pool);
1058                 rte_free(rxq->iovecs);
1059                 rxq->pool = NULL;
1060                 rxq->iovecs = NULL;
1061         }
1062 }
1063
1064 static void
1065 tap_tx_queue_release(void *queue)
1066 {
1067         struct tx_queue *txq = queue;
1068         struct pmd_process_private *process_private;
1069
1070         if (!txq)
1071                 return;
1072         process_private = rte_eth_devices[txq->out_port].process_private;
1073
1074         if (process_private->txq_fds[txq->queue_id] > 0) {
1075                 close(process_private->txq_fds[txq->queue_id]);
1076                 process_private->txq_fds[txq->queue_id] = -1;
1077         }
1078 }
1079
1080 static int
1081 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1082 {
1083         struct rte_eth_link *dev_link = &dev->data->dev_link;
1084         struct pmd_internals *pmd = dev->data->dev_private;
1085         struct ifreq ifr = { .ifr_flags = 0 };
1086
1087         if (pmd->remote_if_index) {
1088                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
1089                 if (!(ifr.ifr_flags & IFF_UP) ||
1090                     !(ifr.ifr_flags & IFF_RUNNING)) {
1091                         dev_link->link_status = ETH_LINK_DOWN;
1092                         return 0;
1093                 }
1094         }
1095         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
1096         dev_link->link_status =
1097                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
1098                  ETH_LINK_UP :
1099                  ETH_LINK_DOWN);
1100         return 0;
1101 }
1102
1103 static void
1104 tap_promisc_enable(struct rte_eth_dev *dev)
1105 {
1106         struct pmd_internals *pmd = dev->data->dev_private;
1107         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1108
1109         dev->data->promiscuous = 1;
1110         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1111         if (pmd->remote_if_index && !pmd->flow_isolate)
1112                 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
1113 }
1114
1115 static void
1116 tap_promisc_disable(struct rte_eth_dev *dev)
1117 {
1118         struct pmd_internals *pmd = dev->data->dev_private;
1119         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1120
1121         dev->data->promiscuous = 0;
1122         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1123         if (pmd->remote_if_index && !pmd->flow_isolate)
1124                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
1125 }
1126
1127 static void
1128 tap_allmulti_enable(struct rte_eth_dev *dev)
1129 {
1130         struct pmd_internals *pmd = dev->data->dev_private;
1131         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1132
1133         dev->data->all_multicast = 1;
1134         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1135         if (pmd->remote_if_index && !pmd->flow_isolate)
1136                 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
1137 }
1138
1139 static void
1140 tap_allmulti_disable(struct rte_eth_dev *dev)
1141 {
1142         struct pmd_internals *pmd = dev->data->dev_private;
1143         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1144
1145         dev->data->all_multicast = 0;
1146         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1147         if (pmd->remote_if_index && !pmd->flow_isolate)
1148                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
1149 }
1150
1151 static int
1152 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1153 {
1154         struct pmd_internals *pmd = dev->data->dev_private;
1155         enum ioctl_mode mode = LOCAL_ONLY;
1156         struct ifreq ifr;
1157         int ret;
1158
1159         if (pmd->type == ETH_TUNTAP_TYPE_TUN) {
1160                 TAP_LOG(ERR, "%s: can't MAC address for TUN",
1161                         dev->device->name);
1162                 return -ENOTSUP;
1163         }
1164
1165         if (is_zero_ether_addr(mac_addr)) {
1166                 TAP_LOG(ERR, "%s: can't set an empty MAC address",
1167                         dev->device->name);
1168                 return -EINVAL;
1169         }
1170         /* Check the actual current MAC address on the tap netdevice */
1171         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
1172         if (ret < 0)
1173                 return ret;
1174         if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
1175                                mac_addr))
1176                 return 0;
1177         /* Check the current MAC address on the remote */
1178         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
1179         if (ret < 0)
1180                 return ret;
1181         if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
1182                                mac_addr))
1183                 mode = LOCAL_AND_REMOTE;
1184         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1185         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
1186         ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
1187         if (ret < 0)
1188                 return ret;
1189         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
1190         if (pmd->remote_if_index && !pmd->flow_isolate) {
1191                 /* Replace MAC redirection rule after a MAC change */
1192                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
1193                 if (ret < 0) {
1194                         TAP_LOG(ERR,
1195                                 "%s: Couldn't delete MAC redirection rule",
1196                                 dev->device->name);
1197                         return ret;
1198                 }
1199                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1200                 if (ret < 0) {
1201                         TAP_LOG(ERR,
1202                                 "%s: Couldn't add MAC redirection rule",
1203                                 dev->device->name);
1204                         return ret;
1205                 }
1206         }
1207
1208         return 0;
1209 }
1210
1211 static int
1212 tap_gso_ctx_setup(struct rte_gso_ctx *gso_ctx, struct rte_eth_dev *dev)
1213 {
1214         uint32_t gso_types;
1215         char pool_name[64];
1216
1217         /*
1218          * Create private mbuf pool with TAP_GSO_MBUF_SEG_SIZE bytes
1219          * size per mbuf use this pool for both direct and indirect mbufs
1220          */
1221
1222         struct rte_mempool *mp;      /* Mempool for GSO packets */
1223
1224         /* initialize GSO context */
1225         gso_types = DEV_TX_OFFLOAD_TCP_TSO;
1226         snprintf(pool_name, sizeof(pool_name), "mp_%s", dev->device->name);
1227         mp = rte_mempool_lookup((const char *)pool_name);
1228         if (!mp) {
1229                 mp = rte_pktmbuf_pool_create(pool_name, TAP_GSO_MBUFS_NUM,
1230                         TAP_GSO_MBUF_CACHE_SIZE, 0,
1231                         RTE_PKTMBUF_HEADROOM + TAP_GSO_MBUF_SEG_SIZE,
1232                         SOCKET_ID_ANY);
1233                 if (!mp) {
1234                         struct pmd_internals *pmd = dev->data->dev_private;
1235                         RTE_LOG(DEBUG, PMD, "%s: failed to create mbuf pool for device %s\n",
1236                                 pmd->name, dev->device->name);
1237                         return -1;
1238                 }
1239         }
1240
1241         gso_ctx->direct_pool = mp;
1242         gso_ctx->indirect_pool = mp;
1243         gso_ctx->gso_types = gso_types;
1244         gso_ctx->gso_size = 0; /* gso_size is set in tx_burst() per packet */
1245         gso_ctx->flag = 0;
1246
1247         return 0;
1248 }
1249
1250 static int
1251 tap_setup_queue(struct rte_eth_dev *dev,
1252                 struct pmd_internals *internals,
1253                 uint16_t qid,
1254                 int is_rx)
1255 {
1256         int ret;
1257         int *fd;
1258         int *other_fd;
1259         const char *dir;
1260         struct pmd_internals *pmd = dev->data->dev_private;
1261         struct pmd_process_private *process_private = dev->process_private;
1262         struct rx_queue *rx = &internals->rxq[qid];
1263         struct tx_queue *tx = &internals->txq[qid];
1264         struct rte_gso_ctx *gso_ctx;
1265
1266         if (is_rx) {
1267                 fd = &process_private->rxq_fds[qid];
1268                 other_fd = &process_private->txq_fds[qid];
1269                 dir = "rx";
1270                 gso_ctx = NULL;
1271         } else {
1272                 fd = &process_private->txq_fds[qid];
1273                 other_fd = &process_private->rxq_fds[qid];
1274                 dir = "tx";
1275                 gso_ctx = &tx->gso_ctx;
1276         }
1277         if (*fd != -1) {
1278                 /* fd for this queue already exists */
1279                 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1280                         pmd->name, *fd, dir, qid);
1281                 gso_ctx = NULL;
1282         } else if (*other_fd != -1) {
1283                 /* Only other_fd exists. dup it */
1284                 *fd = dup(*other_fd);
1285                 if (*fd < 0) {
1286                         *fd = -1;
1287                         TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1288                         return -1;
1289                 }
1290                 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1291                         pmd->name, *other_fd, dir, qid, *fd);
1292         } else {
1293                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1294                 *fd = tun_alloc(pmd, 0);
1295                 if (*fd < 0) {
1296                         *fd = -1; /* restore original value */
1297                         TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1298                         return -1;
1299                 }
1300                 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1301                         pmd->name, dir, qid, *fd);
1302         }
1303
1304         tx->mtu = &dev->data->mtu;
1305         rx->rxmode = &dev->data->dev_conf.rxmode;
1306         if (gso_ctx) {
1307                 ret = tap_gso_ctx_setup(gso_ctx, dev);
1308                 if (ret)
1309                         return -1;
1310         }
1311
1312         tx->type = pmd->type;
1313
1314         return *fd;
1315 }
1316
1317 static int
1318 tap_rx_queue_setup(struct rte_eth_dev *dev,
1319                    uint16_t rx_queue_id,
1320                    uint16_t nb_rx_desc,
1321                    unsigned int socket_id,
1322                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1323                    struct rte_mempool *mp)
1324 {
1325         struct pmd_internals *internals = dev->data->dev_private;
1326         struct pmd_process_private *process_private = dev->process_private;
1327         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1328         struct rte_mbuf **tmp = &rxq->pool;
1329         long iov_max = sysconf(_SC_IOV_MAX);
1330         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1331         struct iovec (*iovecs)[nb_desc + 1];
1332         int data_off = RTE_PKTMBUF_HEADROOM;
1333         int ret = 0;
1334         int fd;
1335         int i;
1336
1337         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1338                 TAP_LOG(WARNING,
1339                         "nb_rx_queues %d too small or mempool NULL",
1340                         dev->data->nb_rx_queues);
1341                 return -1;
1342         }
1343
1344         rxq->mp = mp;
1345         rxq->trigger_seen = 1; /* force initial burst */
1346         rxq->in_port = dev->data->port_id;
1347         rxq->queue_id = rx_queue_id;
1348         rxq->nb_rx_desc = nb_desc;
1349         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1350                                     socket_id);
1351         if (!iovecs) {
1352                 TAP_LOG(WARNING,
1353                         "%s: Couldn't allocate %d RX descriptors",
1354                         dev->device->name, nb_desc);
1355                 return -ENOMEM;
1356         }
1357         rxq->iovecs = iovecs;
1358
1359         dev->data->rx_queues[rx_queue_id] = rxq;
1360         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1361         if (fd == -1) {
1362                 ret = fd;
1363                 goto error;
1364         }
1365
1366         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1367         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1368
1369         for (i = 1; i <= nb_desc; i++) {
1370                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1371                 if (!*tmp) {
1372                         TAP_LOG(WARNING,
1373                                 "%s: couldn't allocate memory for queue %d",
1374                                 dev->device->name, rx_queue_id);
1375                         ret = -ENOMEM;
1376                         goto error;
1377                 }
1378                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1379                 (*rxq->iovecs)[i].iov_base =
1380                         (char *)(*tmp)->buf_addr + data_off;
1381                 data_off = 0;
1382                 tmp = &(*tmp)->next;
1383         }
1384
1385         TAP_LOG(DEBUG, "  RX TUNTAP device name %s, qid %d on fd %d",
1386                 internals->name, rx_queue_id,
1387                 process_private->rxq_fds[rx_queue_id]);
1388
1389         return 0;
1390
1391 error:
1392         rte_pktmbuf_free(rxq->pool);
1393         rxq->pool = NULL;
1394         rte_free(rxq->iovecs);
1395         rxq->iovecs = NULL;
1396         return ret;
1397 }
1398
1399 static int
1400 tap_tx_queue_setup(struct rte_eth_dev *dev,
1401                    uint16_t tx_queue_id,
1402                    uint16_t nb_tx_desc __rte_unused,
1403                    unsigned int socket_id __rte_unused,
1404                    const struct rte_eth_txconf *tx_conf)
1405 {
1406         struct pmd_internals *internals = dev->data->dev_private;
1407         struct pmd_process_private *process_private = dev->process_private;
1408         struct tx_queue *txq;
1409         int ret;
1410         uint64_t offloads;
1411
1412         if (tx_queue_id >= dev->data->nb_tx_queues)
1413                 return -1;
1414         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1415         txq = dev->data->tx_queues[tx_queue_id];
1416         txq->out_port = dev->data->port_id;
1417         txq->queue_id = tx_queue_id;
1418
1419         offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1420         txq->csum = !!(offloads &
1421                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
1422                          DEV_TX_OFFLOAD_UDP_CKSUM |
1423                          DEV_TX_OFFLOAD_TCP_CKSUM));
1424
1425         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1426         if (ret == -1)
1427                 return -1;
1428         TAP_LOG(DEBUG,
1429                 "  TX TUNTAP device name %s, qid %d on fd %d csum %s",
1430                 internals->name, tx_queue_id,
1431                 process_private->txq_fds[tx_queue_id],
1432                 txq->csum ? "on" : "off");
1433
1434         return 0;
1435 }
1436
1437 static int
1438 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1439 {
1440         struct pmd_internals *pmd = dev->data->dev_private;
1441         struct ifreq ifr = { .ifr_mtu = mtu };
1442         int err = 0;
1443
1444         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1445         if (!err)
1446                 dev->data->mtu = mtu;
1447
1448         return err;
1449 }
1450
1451 static int
1452 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1453                      struct ether_addr *mc_addr_set __rte_unused,
1454                      uint32_t nb_mc_addr __rte_unused)
1455 {
1456         /*
1457          * Nothing to do actually: the tap has no filtering whatsoever, every
1458          * packet is received.
1459          */
1460         return 0;
1461 }
1462
1463 static int
1464 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1465 {
1466         struct rte_eth_dev *dev = arg;
1467         struct pmd_internals *pmd = dev->data->dev_private;
1468         struct ifinfomsg *info = NLMSG_DATA(nh);
1469
1470         if (nh->nlmsg_type != RTM_NEWLINK ||
1471             (info->ifi_index != pmd->if_index &&
1472              info->ifi_index != pmd->remote_if_index))
1473                 return 0;
1474         return tap_link_update(dev, 0);
1475 }
1476
1477 static void
1478 tap_dev_intr_handler(void *cb_arg)
1479 {
1480         struct rte_eth_dev *dev = cb_arg;
1481         struct pmd_internals *pmd = dev->data->dev_private;
1482
1483         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1484 }
1485
1486 static int
1487 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1488 {
1489         struct pmd_internals *pmd = dev->data->dev_private;
1490
1491         /* In any case, disable interrupt if the conf is no longer there. */
1492         if (!dev->data->dev_conf.intr_conf.lsc) {
1493                 if (pmd->intr_handle.fd != -1) {
1494                         tap_nl_final(pmd->intr_handle.fd);
1495                         rte_intr_callback_unregister(&pmd->intr_handle,
1496                                 tap_dev_intr_handler, dev);
1497                 }
1498                 return 0;
1499         }
1500         if (set) {
1501                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1502                 if (unlikely(pmd->intr_handle.fd == -1))
1503                         return -EBADF;
1504                 return rte_intr_callback_register(
1505                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1506         }
1507         tap_nl_final(pmd->intr_handle.fd);
1508         return rte_intr_callback_unregister(&pmd->intr_handle,
1509                                             tap_dev_intr_handler, dev);
1510 }
1511
1512 static int
1513 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1514 {
1515         int err;
1516
1517         err = tap_lsc_intr_handle_set(dev, set);
1518         if (err)
1519                 return err;
1520         err = tap_rx_intr_vec_set(dev, set);
1521         if (err && set)
1522                 tap_lsc_intr_handle_set(dev, 0);
1523         return err;
1524 }
1525
1526 static const uint32_t*
1527 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1528 {
1529         static const uint32_t ptypes[] = {
1530                 RTE_PTYPE_INNER_L2_ETHER,
1531                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1532                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1533                 RTE_PTYPE_INNER_L3_IPV4,
1534                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1535                 RTE_PTYPE_INNER_L3_IPV6,
1536                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1537                 RTE_PTYPE_INNER_L4_FRAG,
1538                 RTE_PTYPE_INNER_L4_UDP,
1539                 RTE_PTYPE_INNER_L4_TCP,
1540                 RTE_PTYPE_INNER_L4_SCTP,
1541                 RTE_PTYPE_L2_ETHER,
1542                 RTE_PTYPE_L2_ETHER_VLAN,
1543                 RTE_PTYPE_L2_ETHER_QINQ,
1544                 RTE_PTYPE_L3_IPV4,
1545                 RTE_PTYPE_L3_IPV4_EXT,
1546                 RTE_PTYPE_L3_IPV6_EXT,
1547                 RTE_PTYPE_L3_IPV6,
1548                 RTE_PTYPE_L4_FRAG,
1549                 RTE_PTYPE_L4_UDP,
1550                 RTE_PTYPE_L4_TCP,
1551                 RTE_PTYPE_L4_SCTP,
1552         };
1553
1554         return ptypes;
1555 }
1556
1557 static int
1558 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1559                   struct rte_eth_fc_conf *fc_conf)
1560 {
1561         fc_conf->mode = RTE_FC_NONE;
1562         return 0;
1563 }
1564
1565 static int
1566 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1567                   struct rte_eth_fc_conf *fc_conf)
1568 {
1569         if (fc_conf->mode != RTE_FC_NONE)
1570                 return -ENOTSUP;
1571         return 0;
1572 }
1573
1574 /**
1575  * DPDK callback to update the RSS hash configuration.
1576  *
1577  * @param dev
1578  *   Pointer to Ethernet device structure.
1579  * @param[in] rss_conf
1580  *   RSS configuration data.
1581  *
1582  * @return
1583  *   0 on success, a negative errno value otherwise and rte_errno is set.
1584  */
1585 static int
1586 tap_rss_hash_update(struct rte_eth_dev *dev,
1587                 struct rte_eth_rss_conf *rss_conf)
1588 {
1589         if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1590                 rte_errno = EINVAL;
1591                 return -rte_errno;
1592         }
1593         if (rss_conf->rss_key && rss_conf->rss_key_len) {
1594                 /*
1595                  * Currently TAP RSS key is hard coded
1596                  * and cannot be updated
1597                  */
1598                 TAP_LOG(ERR,
1599                         "port %u RSS key cannot be updated",
1600                         dev->data->port_id);
1601                 rte_errno = EINVAL;
1602                 return -rte_errno;
1603         }
1604         return 0;
1605 }
1606
1607 static int
1608 tap_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1609 {
1610         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1611
1612         return 0;
1613 }
1614
1615 static int
1616 tap_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1617 {
1618         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1619
1620         return 0;
1621 }
1622
1623 static int
1624 tap_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1625 {
1626         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1627
1628         return 0;
1629 }
1630
1631 static int
1632 tap_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1633 {
1634         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1635
1636         return 0;
1637 }
1638 static const struct eth_dev_ops ops = {
1639         .dev_start              = tap_dev_start,
1640         .dev_stop               = tap_dev_stop,
1641         .dev_close              = tap_dev_close,
1642         .dev_configure          = tap_dev_configure,
1643         .dev_infos_get          = tap_dev_info,
1644         .rx_queue_setup         = tap_rx_queue_setup,
1645         .tx_queue_setup         = tap_tx_queue_setup,
1646         .rx_queue_start         = tap_rx_queue_start,
1647         .tx_queue_start         = tap_tx_queue_start,
1648         .rx_queue_stop          = tap_rx_queue_stop,
1649         .tx_queue_stop          = tap_tx_queue_stop,
1650         .rx_queue_release       = tap_rx_queue_release,
1651         .tx_queue_release       = tap_tx_queue_release,
1652         .flow_ctrl_get          = tap_flow_ctrl_get,
1653         .flow_ctrl_set          = tap_flow_ctrl_set,
1654         .link_update            = tap_link_update,
1655         .dev_set_link_up        = tap_link_set_up,
1656         .dev_set_link_down      = tap_link_set_down,
1657         .promiscuous_enable     = tap_promisc_enable,
1658         .promiscuous_disable    = tap_promisc_disable,
1659         .allmulticast_enable    = tap_allmulti_enable,
1660         .allmulticast_disable   = tap_allmulti_disable,
1661         .mac_addr_set           = tap_mac_set,
1662         .mtu_set                = tap_mtu_set,
1663         .set_mc_addr_list       = tap_set_mc_addr_list,
1664         .stats_get              = tap_stats_get,
1665         .stats_reset            = tap_stats_reset,
1666         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1667         .rss_hash_update        = tap_rss_hash_update,
1668         .filter_ctrl            = tap_dev_filter_ctrl,
1669 };
1670
1671 static int
1672 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1673                    char *remote_iface, struct ether_addr *mac_addr,
1674                    enum rte_tuntap_type type)
1675 {
1676         int numa_node = rte_socket_id();
1677         struct rte_eth_dev *dev;
1678         struct pmd_internals *pmd;
1679         struct pmd_process_private *process_private;
1680         struct rte_eth_dev_data *data;
1681         struct ifreq ifr;
1682         int i;
1683
1684         TAP_LOG(DEBUG, "%s device on numa %u",
1685                         tuntap_name, rte_socket_id());
1686
1687         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1688         if (!dev) {
1689                 TAP_LOG(ERR, "%s Unable to allocate device struct",
1690                                 tuntap_name);
1691                 goto error_exit_nodev;
1692         }
1693
1694         process_private = (struct pmd_process_private *)
1695                 rte_zmalloc_socket(tap_name, sizeof(struct pmd_process_private),
1696                         RTE_CACHE_LINE_SIZE, dev->device->numa_node);
1697
1698         if (process_private == NULL) {
1699                 TAP_LOG(ERR, "Failed to alloc memory for process private");
1700                 return -1;
1701         }
1702         pmd = dev->data->dev_private;
1703         dev->process_private = process_private;
1704         pmd->dev = dev;
1705         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1706         pmd->type = type;
1707
1708         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1709         if (pmd->ioctl_sock == -1) {
1710                 TAP_LOG(ERR,
1711                         "%s Unable to get a socket for management: %s",
1712                         tuntap_name, strerror(errno));
1713                 goto error_exit;
1714         }
1715
1716         /* Setup some default values */
1717         data = dev->data;
1718         data->dev_private = pmd;
1719         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1720         data->numa_node = numa_node;
1721
1722         data->dev_link = pmd_link;
1723         data->mac_addrs = &pmd->eth_addr;
1724         /* Set the number of RX and TX queues */
1725         data->nb_rx_queues = 0;
1726         data->nb_tx_queues = 0;
1727
1728         dev->dev_ops = &ops;
1729         dev->rx_pkt_burst = pmd_rx_burst;
1730         dev->tx_pkt_burst = pmd_tx_burst;
1731
1732         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1733         pmd->intr_handle.fd = -1;
1734         dev->intr_handle = &pmd->intr_handle;
1735
1736         /* Presetup the fds to -1 as being not valid */
1737         pmd->ka_fd = -1;
1738         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1739                 process_private->rxq_fds[i] = -1;
1740                 process_private->txq_fds[i] = -1;
1741         }
1742
1743         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1744                 if (is_zero_ether_addr(mac_addr))
1745                         eth_random_addr((uint8_t *)&pmd->eth_addr);
1746                 else
1747                         rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1748         }
1749
1750         /*
1751          * Allocate a TUN device keep-alive file descriptor that will only be
1752          * closed when the TUN device itself is closed or removed.
1753          * This keep-alive file descriptor will guarantee that the TUN device
1754          * exists even when all of its queues are closed
1755          */
1756         pmd->ka_fd = tun_alloc(pmd, 1);
1757         if (pmd->ka_fd == -1) {
1758                 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
1759                 goto error_exit;
1760         }
1761         TAP_LOG(DEBUG, "allocated %s", pmd->name);
1762
1763         ifr.ifr_mtu = dev->data->mtu;
1764         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1765                 goto error_exit;
1766
1767         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1768                 memset(&ifr, 0, sizeof(struct ifreq));
1769                 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1770                 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1771                                 ETHER_ADDR_LEN);
1772                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1773                         goto error_exit;
1774         }
1775
1776         /*
1777          * Set up everything related to rte_flow:
1778          * - netlink socket
1779          * - tap / remote if_index
1780          * - mandatory QDISCs
1781          * - rte_flow actual/implicit lists
1782          * - implicit rules
1783          */
1784         pmd->nlsk_fd = tap_nl_init(0);
1785         if (pmd->nlsk_fd == -1) {
1786                 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
1787                         pmd->name);
1788                 goto disable_rte_flow;
1789         }
1790         pmd->if_index = if_nametoindex(pmd->name);
1791         if (!pmd->if_index) {
1792                 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
1793                 goto disable_rte_flow;
1794         }
1795         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1796                 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
1797                         pmd->name);
1798                 goto disable_rte_flow;
1799         }
1800         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1801                 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1802                         pmd->name);
1803                 goto disable_rte_flow;
1804         }
1805         LIST_INIT(&pmd->flows);
1806
1807         if (strlen(remote_iface)) {
1808                 pmd->remote_if_index = if_nametoindex(remote_iface);
1809                 if (!pmd->remote_if_index) {
1810                         TAP_LOG(ERR, "%s: failed to get %s if_index.",
1811                                 pmd->name, remote_iface);
1812                         goto error_remote;
1813                 }
1814                 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1815                          "%s", remote_iface);
1816
1817                 /* Save state of remote device */
1818                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1819
1820                 /* Replicate remote MAC address */
1821                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1822                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1823                                 pmd->name, pmd->remote_iface);
1824                         goto error_remote;
1825                 }
1826                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1827                            ETHER_ADDR_LEN);
1828                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1829                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1830                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1831                                 pmd->name, remote_iface);
1832                         goto error_remote;
1833                 }
1834
1835                 /*
1836                  * Flush usually returns negative value because it tries to
1837                  * delete every QDISC (and on a running device, one QDISC at
1838                  * least is needed). Ignore negative return value.
1839                  */
1840                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1841                 if (qdisc_create_ingress(pmd->nlsk_fd,
1842                                          pmd->remote_if_index) < 0) {
1843                         TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1844                                 pmd->remote_iface);
1845                         goto error_remote;
1846                 }
1847                 LIST_INIT(&pmd->implicit_flows);
1848                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1849                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1850                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1851                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1852                         TAP_LOG(ERR,
1853                                 "%s: failed to create implicit rules.",
1854                                 pmd->name);
1855                         goto error_remote;
1856                 }
1857         }
1858
1859         rte_eth_dev_probing_finish(dev);
1860         return 0;
1861
1862 disable_rte_flow:
1863         TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
1864                 strerror(errno), errno);
1865         if (strlen(remote_iface)) {
1866                 TAP_LOG(ERR, "Remote feature requires flow support.");
1867                 goto error_exit;
1868         }
1869         rte_eth_dev_probing_finish(dev);
1870         return 0;
1871
1872 error_remote:
1873         TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
1874                 strerror(errno), errno);
1875         tap_flow_implicit_flush(pmd, NULL);
1876
1877 error_exit:
1878         if (pmd->ioctl_sock > 0)
1879                 close(pmd->ioctl_sock);
1880         /* mac_addrs must not be freed alone because part of dev_private */
1881         dev->data->mac_addrs = NULL;
1882         rte_eth_dev_release_port(dev);
1883
1884 error_exit_nodev:
1885         TAP_LOG(ERR, "%s Unable to initialize %s",
1886                 tuntap_name, rte_vdev_device_name(vdev));
1887
1888         return -EINVAL;
1889 }
1890
1891 /* make sure name is a possible Linux network device name */
1892 static bool
1893 is_valid_iface(const char *name)
1894 {
1895         if (*name == '\0')
1896                 return false;
1897
1898         if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1899                 return false;
1900
1901         while (*name) {
1902                 if (*name == '/' || *name == ':' || isspace(*name))
1903                         return false;
1904                 name++;
1905         }
1906         return true;
1907 }
1908
1909 static int
1910 set_interface_name(const char *key __rte_unused,
1911                    const char *value,
1912                    void *extra_args)
1913 {
1914         char *name = (char *)extra_args;
1915
1916         if (value) {
1917                 if (!is_valid_iface(value)) {
1918                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
1919                                 value);
1920                         return -1;
1921                 }
1922                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
1923         } else {
1924                 /* use tap%d which causes kernel to choose next available */
1925                 strlcpy(name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
1926         }
1927         return 0;
1928 }
1929
1930 static int
1931 set_remote_iface(const char *key __rte_unused,
1932                  const char *value,
1933                  void *extra_args)
1934 {
1935         char *name = (char *)extra_args;
1936
1937         if (value) {
1938                 if (!is_valid_iface(value)) {
1939                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
1940                                 value);
1941                         return -1;
1942                 }
1943                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
1944         }
1945
1946         return 0;
1947 }
1948
1949 static int parse_user_mac(struct ether_addr *user_mac,
1950                 const char *value)
1951 {
1952         unsigned int index = 0;
1953         char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
1954
1955         if (user_mac == NULL || value == NULL)
1956                 return 0;
1957
1958         strlcpy(mac_temp, value, sizeof(mac_temp));
1959         mac_byte = strtok(mac_temp, ":");
1960
1961         while ((mac_byte != NULL) &&
1962                         (strlen(mac_byte) <= 2) &&
1963                         (strlen(mac_byte) == strspn(mac_byte,
1964                                         ETH_TAP_CMP_MAC_FMT))) {
1965                 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
1966                 mac_byte = strtok(NULL, ":");
1967         }
1968
1969         return index;
1970 }
1971
1972 static int
1973 set_mac_type(const char *key __rte_unused,
1974              const char *value,
1975              void *extra_args)
1976 {
1977         struct ether_addr *user_mac = extra_args;
1978
1979         if (!value)
1980                 return 0;
1981
1982         if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
1983                 static int iface_idx;
1984
1985                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1986                 memcpy((char *)user_mac->addr_bytes, "\0dtap", ETHER_ADDR_LEN);
1987                 user_mac->addr_bytes[ETHER_ADDR_LEN - 1] = iface_idx++ + '0';
1988                 goto success;
1989         }
1990
1991         if (parse_user_mac(user_mac, value) != 6)
1992                 goto error;
1993 success:
1994         TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
1995         return 0;
1996
1997 error:
1998         TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
1999                 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
2000         return -1;
2001 }
2002
2003 /*
2004  * Open a TUN interface device. TUN PMD
2005  * 1) sets tap_type as false
2006  * 2) intakes iface as argument.
2007  * 3) as interface is virtual set speed to 10G
2008  */
2009 static int
2010 rte_pmd_tun_probe(struct rte_vdev_device *dev)
2011 {
2012         const char *name, *params;
2013         int ret;
2014         struct rte_kvargs *kvlist = NULL;
2015         char tun_name[RTE_ETH_NAME_MAX_LEN];
2016         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2017         struct rte_eth_dev *eth_dev;
2018
2019         strcpy(tuntap_name, "TUN");
2020
2021         name = rte_vdev_device_name(dev);
2022         params = rte_vdev_device_args(dev);
2023         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2024
2025         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
2026             strlen(params) == 0) {
2027                 eth_dev = rte_eth_dev_attach_secondary(name);
2028                 if (!eth_dev) {
2029                         TAP_LOG(ERR, "Failed to probe %s", name);
2030                         return -1;
2031                 }
2032                 eth_dev->dev_ops = &ops;
2033                 eth_dev->device = &dev->device;
2034                 rte_eth_dev_probing_finish(eth_dev);
2035                 return 0;
2036         }
2037
2038         /* use tun%d which causes kernel to choose next available */
2039         strlcpy(tun_name, DEFAULT_TUN_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2040
2041         if (params && (params[0] != '\0')) {
2042                 TAP_LOG(DEBUG, "parameters (%s)", params);
2043
2044                 kvlist = rte_kvargs_parse(params, valid_arguments);
2045                 if (kvlist) {
2046                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2047                                 ret = rte_kvargs_process(kvlist,
2048                                         ETH_TAP_IFACE_ARG,
2049                                         &set_interface_name,
2050                                         tun_name);
2051
2052                                 if (ret == -1)
2053                                         goto leave;
2054                         }
2055                 }
2056         }
2057         pmd_link.link_speed = ETH_SPEED_NUM_10G;
2058
2059         TAP_LOG(DEBUG, "Initializing pmd_tun for %s", name);
2060
2061         ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0,
2062                                  ETH_TUNTAP_TYPE_TUN);
2063
2064 leave:
2065         if (ret == -1) {
2066                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2067                         name, tun_name);
2068         }
2069         rte_kvargs_free(kvlist);
2070
2071         return ret;
2072 }
2073
2074 /* Request queue file descriptors from secondary to primary. */
2075 static int
2076 tap_mp_attach_queues(const char *port_name, struct rte_eth_dev *dev)
2077 {
2078         int ret;
2079         struct timespec timeout = {.tv_sec = 1, .tv_nsec = 0};
2080         struct rte_mp_msg request, *reply;
2081         struct rte_mp_reply replies;
2082         struct ipc_queues *request_param = (struct ipc_queues *)request.param;
2083         struct ipc_queues *reply_param;
2084         struct pmd_process_private *process_private = dev->process_private;
2085         int queue, fd_iterator;
2086
2087         /* Prepare the request */
2088         strlcpy(request.name, TAP_MP_KEY, sizeof(request.name));
2089         strlcpy(request_param->port_name, port_name,
2090                 sizeof(request_param->port_name));
2091         request.len_param = sizeof(*request_param);
2092         /* Send request and receive reply */
2093         ret = rte_mp_request_sync(&request, &replies, &timeout);
2094         if (ret < 0) {
2095                 TAP_LOG(ERR, "Failed to request queues from primary: %d",
2096                         rte_errno);
2097                 return -1;
2098         }
2099         reply = &replies.msgs[0];
2100         reply_param = (struct ipc_queues *)reply->param;
2101         TAP_LOG(DEBUG, "Received IPC reply for %s", reply_param->port_name);
2102
2103         /* Attach the queues from received file descriptors */
2104         dev->data->nb_rx_queues = reply_param->rxq_count;
2105         dev->data->nb_tx_queues = reply_param->txq_count;
2106         fd_iterator = 0;
2107         for (queue = 0; queue < reply_param->rxq_count; queue++)
2108                 process_private->rxq_fds[queue] = reply->fds[fd_iterator++];
2109         for (queue = 0; queue < reply_param->txq_count; queue++)
2110                 process_private->txq_fds[queue] = reply->fds[fd_iterator++];
2111
2112         return 0;
2113 }
2114
2115 /* Send the queue file descriptors from the primary process to secondary. */
2116 static int
2117 tap_mp_sync_queues(const struct rte_mp_msg *request, const void *peer)
2118 {
2119         struct rte_eth_dev *dev;
2120         struct pmd_process_private *process_private;
2121         struct rte_mp_msg reply;
2122         const struct ipc_queues *request_param =
2123                 (const struct ipc_queues *)request->param;
2124         struct ipc_queues *reply_param =
2125                 (struct ipc_queues *)reply.param;
2126         uint16_t port_id;
2127         int queue;
2128         int ret;
2129
2130         /* Get requested port */
2131         TAP_LOG(DEBUG, "Received IPC request for %s", request_param->port_name);
2132         ret = rte_eth_dev_get_port_by_name(request_param->port_name, &port_id);
2133         if (ret) {
2134                 TAP_LOG(ERR, "Failed to get port id for %s",
2135                         request_param->port_name);
2136                 return -1;
2137         }
2138         dev = &rte_eth_devices[port_id];
2139         process_private = dev->process_private;
2140
2141         /* Fill file descriptors for all queues */
2142         reply.num_fds = 0;
2143         reply_param->rxq_count = 0;
2144         for (queue = 0; queue < dev->data->nb_rx_queues; queue++) {
2145                 reply.fds[reply.num_fds++] = process_private->rxq_fds[queue];
2146                 reply_param->rxq_count++;
2147         }
2148         RTE_ASSERT(reply_param->rxq_count == dev->data->nb_rx_queues);
2149         RTE_ASSERT(reply_param->txq_count == dev->data->nb_tx_queues);
2150         RTE_ASSERT(reply.num_fds <= RTE_MP_MAX_FD_NUM);
2151
2152         reply_param->txq_count = 0;
2153         for (queue = 0; queue < dev->data->nb_tx_queues; queue++) {
2154                 reply.fds[reply.num_fds++] = process_private->txq_fds[queue];
2155                 reply_param->txq_count++;
2156         }
2157
2158         /* Send reply */
2159         strlcpy(reply.name, request->name, sizeof(reply.name));
2160         strlcpy(reply_param->port_name, request_param->port_name,
2161                 sizeof(reply_param->port_name));
2162         reply.len_param = sizeof(*reply_param);
2163         if (rte_mp_reply(&reply, peer) < 0) {
2164                 TAP_LOG(ERR, "Failed to reply an IPC request to sync queues");
2165                 return -1;
2166         }
2167         return 0;
2168 }
2169
2170 /* Open a TAP interface device.
2171  */
2172 static int
2173 rte_pmd_tap_probe(struct rte_vdev_device *dev)
2174 {
2175         const char *name, *params;
2176         int ret;
2177         struct rte_kvargs *kvlist = NULL;
2178         int speed;
2179         char tap_name[RTE_ETH_NAME_MAX_LEN];
2180         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2181         struct ether_addr user_mac = { .addr_bytes = {0} };
2182         struct rte_eth_dev *eth_dev;
2183         int tap_devices_count_increased = 0;
2184
2185         strcpy(tuntap_name, "TAP");
2186
2187         name = rte_vdev_device_name(dev);
2188         params = rte_vdev_device_args(dev);
2189
2190         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
2191                 eth_dev = rte_eth_dev_attach_secondary(name);
2192                 if (!eth_dev) {
2193                         TAP_LOG(ERR, "Failed to probe %s", name);
2194                         return -1;
2195                 }
2196                 eth_dev->dev_ops = &ops;
2197                 eth_dev->device = &dev->device;
2198                 eth_dev->rx_pkt_burst = pmd_rx_burst;
2199                 eth_dev->tx_pkt_burst = pmd_tx_burst;
2200                 if (!rte_eal_primary_proc_alive(NULL)) {
2201                         TAP_LOG(ERR, "Primary process is missing");
2202                         return -1;
2203                 }
2204                 eth_dev->process_private = (struct pmd_process_private *)
2205                         rte_zmalloc_socket(name,
2206                                 sizeof(struct pmd_process_private),
2207                                 RTE_CACHE_LINE_SIZE,
2208                                 eth_dev->device->numa_node);
2209                 if (eth_dev->process_private == NULL) {
2210                         TAP_LOG(ERR,
2211                                 "Failed to alloc memory for process private");
2212                         return -1;
2213                 }
2214
2215                 ret = tap_mp_attach_queues(name, eth_dev);
2216                 if (ret != 0)
2217                         return -1;
2218                 rte_eth_dev_probing_finish(eth_dev);
2219                 return 0;
2220         }
2221
2222         speed = ETH_SPEED_NUM_10G;
2223
2224         /* use tap%d which causes kernel to choose next available */
2225         strlcpy(tap_name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2226         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2227
2228         if (params && (params[0] != '\0')) {
2229                 TAP_LOG(DEBUG, "parameters (%s)", params);
2230
2231                 kvlist = rte_kvargs_parse(params, valid_arguments);
2232                 if (kvlist) {
2233                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2234                                 ret = rte_kvargs_process(kvlist,
2235                                                          ETH_TAP_IFACE_ARG,
2236                                                          &set_interface_name,
2237                                                          tap_name);
2238                                 if (ret == -1)
2239                                         goto leave;
2240                         }
2241
2242                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
2243                                 ret = rte_kvargs_process(kvlist,
2244                                                          ETH_TAP_REMOTE_ARG,
2245                                                          &set_remote_iface,
2246                                                          remote_iface);
2247                                 if (ret == -1)
2248                                         goto leave;
2249                         }
2250
2251                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
2252                                 ret = rte_kvargs_process(kvlist,
2253                                                          ETH_TAP_MAC_ARG,
2254                                                          &set_mac_type,
2255                                                          &user_mac);
2256                                 if (ret == -1)
2257                                         goto leave;
2258                         }
2259                 }
2260         }
2261         pmd_link.link_speed = speed;
2262
2263         TAP_LOG(DEBUG, "Initializing pmd_tap for %s", name);
2264
2265         /* Register IPC feed callback */
2266         if (!tap_devices_count) {
2267                 ret = rte_mp_action_register(TAP_MP_KEY, tap_mp_sync_queues);
2268                 if (ret < 0) {
2269                         TAP_LOG(ERR, "%s: Failed to register IPC callback: %s",
2270                                 tuntap_name, strerror(rte_errno));
2271                         goto leave;
2272                 }
2273         }
2274         tap_devices_count++;
2275         tap_devices_count_increased = 1;
2276         ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
2277                 ETH_TUNTAP_TYPE_TAP);
2278
2279 leave:
2280         if (ret == -1) {
2281                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2282                         name, tap_name);
2283                 if (tap_devices_count_increased == 1) {
2284                         if (tap_devices_count == 1)
2285                                 rte_mp_action_unregister(TAP_MP_KEY);
2286                         tap_devices_count--;
2287                 }
2288         }
2289         rte_kvargs_free(kvlist);
2290
2291         return ret;
2292 }
2293
2294 /* detach a TUNTAP device.
2295  */
2296 static int
2297 rte_pmd_tap_remove(struct rte_vdev_device *dev)
2298 {
2299         struct rte_eth_dev *eth_dev = NULL;
2300         struct pmd_internals *internals;
2301         struct pmd_process_private *process_private;
2302         int i;
2303
2304         /* find the ethdev entry */
2305         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
2306         if (!eth_dev)
2307                 return -ENODEV;
2308
2309         /* mac_addrs must not be freed alone because part of dev_private */
2310         eth_dev->data->mac_addrs = NULL;
2311
2312         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2313                 return rte_eth_dev_release_port(eth_dev);
2314
2315         internals = eth_dev->data->dev_private;
2316         process_private = eth_dev->process_private;
2317
2318         TAP_LOG(DEBUG, "Closing %s Ethernet device on numa %u",
2319                 (internals->type == ETH_TUNTAP_TYPE_TAP) ? "TAP" : "TUN",
2320                 rte_socket_id());
2321
2322         if (internals->nlsk_fd) {
2323                 tap_flow_flush(eth_dev, NULL);
2324                 tap_flow_implicit_flush(internals, NULL);
2325                 tap_nl_final(internals->nlsk_fd);
2326         }
2327         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
2328                 if (process_private->rxq_fds[i] != -1) {
2329                         close(process_private->rxq_fds[i]);
2330                         process_private->rxq_fds[i] = -1;
2331                 }
2332                 if (process_private->txq_fds[i] != -1) {
2333                         close(process_private->txq_fds[i]);
2334                         process_private->txq_fds[i] = -1;
2335                 }
2336         }
2337
2338         close(internals->ioctl_sock);
2339         rte_free(eth_dev->process_private);
2340         if (tap_devices_count == 1)
2341                 rte_mp_action_unregister(TAP_MP_KEY);
2342         tap_devices_count--;
2343         rte_eth_dev_release_port(eth_dev);
2344
2345         if (internals->ka_fd != -1) {
2346                 close(internals->ka_fd);
2347                 internals->ka_fd = -1;
2348         }
2349         return 0;
2350 }
2351
2352 static struct rte_vdev_driver pmd_tun_drv = {
2353         .probe = rte_pmd_tun_probe,
2354         .remove = rte_pmd_tap_remove,
2355 };
2356
2357 static struct rte_vdev_driver pmd_tap_drv = {
2358         .probe = rte_pmd_tap_probe,
2359         .remove = rte_pmd_tap_remove,
2360 };
2361
2362 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
2363 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
2364 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
2365 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
2366                               ETH_TAP_IFACE_ARG "=<string> ");
2367 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
2368                               ETH_TAP_IFACE_ARG "=<string> "
2369                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
2370                               ETH_TAP_REMOTE_ARG "=<string>");
2371 int tap_logtype;
2372
2373 RTE_INIT(tap_init_log)
2374 {
2375         tap_logtype = rte_log_register("pmd.net.tap");
2376         if (tap_logtype >= 0)
2377                 rte_log_set_level(tap_logtype, RTE_LOG_NOTICE);
2378 }