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