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 #define TAP_IOV_DEFAULT_MAX 1024
  72
  73 static int tap_devices_count;
  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 rte_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 rte_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 rte_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 rte_ipv6_hdr *iph = l3_hdr;
 297
 298                 l3_len = sizeof(struct rte_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 rte_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 rte_udp_hdr *)l4_hdr)->dgram_cksum;
 510                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 511                         *l4_cksum = &((struct rte_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(RTE_ETHER_TYPE_IPV4) :
 567                                 ((proto == 0x60) ?
 568                                         rte_cpu_to_be_16(RTE_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 + (RTE_ETHER_HDR_LEN + RTE_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 RTE_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                 strlcpy(ifr->ifr_name, pmd->remote_iface, IFNAMSIZ);
 762         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 763                 strlcpy(ifr->ifr_name, pmd->name, IFNAMSIZ);
 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 int
 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)RTE_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         return 0;
 944 }
 945
 946 static int
 947 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 948 {
 949         unsigned int i, imax;
 950         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 951         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 952         unsigned long rx_nombuf = 0, ierrors = 0;
 953         const struct pmd_internals *pmd = dev->data->dev_private;
 954
 955         /* rx queue statistics */
 956         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 957                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 958         for (i = 0; i < imax; i++) {
 959                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 960                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 961                 rx_total += tap_stats->q_ipackets[i];
 962                 rx_bytes_total += tap_stats->q_ibytes[i];
 963                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
 964                 ierrors += pmd->rxq[i].stats.ierrors;
 965         }
 966
 967         /* tx queue statistics */
 968         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 969                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 970
 971         for (i = 0; i < imax; i++) {
 972                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 973                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 974                 tx_total += tap_stats->q_opackets[i];
 975                 tx_err_total += pmd->txq[i].stats.errs;
 976                 tx_bytes_total += tap_stats->q_obytes[i];
 977         }
 978
 979         tap_stats->ipackets = rx_total;
 980         tap_stats->ibytes = rx_bytes_total;
 981         tap_stats->ierrors = ierrors;
 982         tap_stats->rx_nombuf = rx_nombuf;
 983         tap_stats->opackets = tx_total;
 984         tap_stats->oerrors = tx_err_total;
 985         tap_stats->obytes = tx_bytes_total;
 986         return 0;
 987 }
 988
 989 static int
 990 tap_stats_reset(struct rte_eth_dev *dev)
 991 {
 992         int i;
 993         struct pmd_internals *pmd = dev->data->dev_private;
 994
 995         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 996                 pmd->rxq[i].stats.ipackets = 0;
 997                 pmd->rxq[i].stats.ibytes = 0;
 998                 pmd->rxq[i].stats.ierrors = 0;
 999                 pmd->rxq[i].stats.rx_nombuf = 0;
1000
1001                 pmd->txq[i].stats.opackets = 0;
1002                 pmd->txq[i].stats.errs = 0;
1003                 pmd->txq[i].stats.obytes = 0;
1004         }
1005
1006         return 0;
1007 }
1008
1009 static void
1010 tap_dev_close(struct rte_eth_dev *dev)
1011 {
1012         int i;
1013         struct pmd_internals *internals = dev->data->dev_private;
1014         struct pmd_process_private *process_private = dev->process_private;
1015
1016         tap_link_set_down(dev);
1017         tap_flow_flush(dev, NULL);
1018         tap_flow_implicit_flush(internals, NULL);
1019
1020         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1021                 if (process_private->rxq_fds[i] != -1) {
1022                         close(process_private->rxq_fds[i]);
1023                         process_private->rxq_fds[i] = -1;
1024                 }
1025                 if (process_private->txq_fds[i] != -1) {
1026                         close(process_private->txq_fds[i]);
1027                         process_private->txq_fds[i] = -1;
1028                 }
1029         }
1030
1031         if (internals->remote_if_index) {
1032                 /* Restore initial remote state */
1033                 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
1034                                 &internals->remote_initial_flags);
1035         }
1036
1037         if (internals->ka_fd != -1) {
1038                 close(internals->ka_fd);
1039                 internals->ka_fd = -1;
1040         }
1041         /*
1042          * Since TUN device has no more opened file descriptors
1043          * it will be removed from kernel
1044          */
1045 }
1046
1047 static void
1048 tap_rx_queue_release(void *queue)
1049 {
1050         struct rx_queue *rxq = queue;
1051         struct pmd_process_private *process_private;
1052
1053         if (!rxq)
1054                 return;
1055         process_private = rte_eth_devices[rxq->in_port].process_private;
1056         if (process_private->rxq_fds[rxq->queue_id] > 0) {
1057                 close(process_private->rxq_fds[rxq->queue_id]);
1058                 process_private->rxq_fds[rxq->queue_id] = -1;
1059                 rte_pktmbuf_free(rxq->pool);
1060                 rte_free(rxq->iovecs);
1061                 rxq->pool = NULL;
1062                 rxq->iovecs = NULL;
1063         }
1064 }
1065
1066 static void
1067 tap_tx_queue_release(void *queue)
1068 {
1069         struct tx_queue *txq = queue;
1070         struct pmd_process_private *process_private;
1071
1072         if (!txq)
1073                 return;
1074         process_private = rte_eth_devices[txq->out_port].process_private;
1075
1076         if (process_private->txq_fds[txq->queue_id] > 0) {
1077                 close(process_private->txq_fds[txq->queue_id]);
1078                 process_private->txq_fds[txq->queue_id] = -1;
1079         }
1080 }
1081
1082 static int
1083 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1084 {
1085         struct rte_eth_link *dev_link = &dev->data->dev_link;
1086         struct pmd_internals *pmd = dev->data->dev_private;
1087         struct ifreq ifr = { .ifr_flags = 0 };
1088
1089         if (pmd->remote_if_index) {
1090                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
1091                 if (!(ifr.ifr_flags & IFF_UP) ||
1092                     !(ifr.ifr_flags & IFF_RUNNING)) {
1093                         dev_link->link_status = ETH_LINK_DOWN;
1094                         return 0;
1095                 }
1096         }
1097         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
1098         dev_link->link_status =
1099                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
1100                  ETH_LINK_UP :
1101                  ETH_LINK_DOWN);
1102         return 0;
1103 }
1104
1105 static int
1106 tap_promisc_enable(struct rte_eth_dev *dev)
1107 {
1108         struct pmd_internals *pmd = dev->data->dev_private;
1109         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1110         int ret;
1111
1112         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1113         if (ret != 0)
1114                 return ret;
1115
1116         if (pmd->remote_if_index && !pmd->flow_isolate) {
1117                 dev->data->promiscuous = 1;
1118                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
1119                 if (ret != 0) {
1120                         /* Rollback promisc flag */
1121                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1122                         /*
1123                          * rte_eth_dev_promiscuous_enable() rollback
1124                          * dev->data->promiscuous in the case of failure.
1125                          */
1126                         return ret;
1127                 }
1128         }
1129
1130         return 0;
1131 }
1132
1133 static int
1134 tap_promisc_disable(struct rte_eth_dev *dev)
1135 {
1136         struct pmd_internals *pmd = dev->data->dev_private;
1137         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1138         int ret;
1139
1140         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1141         if (ret != 0)
1142                 return ret;
1143
1144         if (pmd->remote_if_index && !pmd->flow_isolate) {
1145                 dev->data->promiscuous = 0;
1146                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
1147                 if (ret != 0) {
1148                         /* Rollback promisc flag */
1149                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1150                         /*
1151                          * rte_eth_dev_promiscuous_disable() rollback
1152                          * dev->data->promiscuous in the case of failure.
1153                          */
1154                         return ret;
1155                 }
1156         }
1157
1158         return 0;
1159 }
1160
1161 static void
1162 tap_allmulti_enable(struct rte_eth_dev *dev)
1163 {
1164         struct pmd_internals *pmd = dev->data->dev_private;
1165         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1166
1167         dev->data->all_multicast = 1;
1168         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1169         if (pmd->remote_if_index && !pmd->flow_isolate)
1170                 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
1171 }
1172
1173 static void
1174 tap_allmulti_disable(struct rte_eth_dev *dev)
1175 {
1176         struct pmd_internals *pmd = dev->data->dev_private;
1177         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1178
1179         dev->data->all_multicast = 0;
1180         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1181         if (pmd->remote_if_index && !pmd->flow_isolate)
1182                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
1183 }
1184
1185 static int
1186 tap_mac_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1187 {
1188         struct pmd_internals *pmd = dev->data->dev_private;
1189         enum ioctl_mode mode = LOCAL_ONLY;
1190         struct ifreq ifr;
1191         int ret;
1192
1193         if (pmd->type == ETH_TUNTAP_TYPE_TUN) {
1194                 TAP_LOG(ERR, "%s: can't MAC address for TUN",
1195                         dev->device->name);
1196                 return -ENOTSUP;
1197         }
1198
1199         if (rte_is_zero_ether_addr(mac_addr)) {
1200                 TAP_LOG(ERR, "%s: can't set an empty MAC address",
1201                         dev->device->name);
1202                 return -EINVAL;
1203         }
1204         /* Check the actual current MAC address on the tap netdevice */
1205         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
1206         if (ret < 0)
1207                 return ret;
1208         if (rte_is_same_ether_addr(
1209                         (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1210                         mac_addr))
1211                 return 0;
1212         /* Check the current MAC address on the remote */
1213         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
1214         if (ret < 0)
1215                 return ret;
1216         if (!rte_is_same_ether_addr(
1217                         (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1218                         mac_addr))
1219                 mode = LOCAL_AND_REMOTE;
1220         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1221         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, RTE_ETHER_ADDR_LEN);
1222         ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
1223         if (ret < 0)
1224                 return ret;
1225         rte_memcpy(&pmd->eth_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1226         if (pmd->remote_if_index && !pmd->flow_isolate) {
1227                 /* Replace MAC redirection rule after a MAC change */
1228                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
1229                 if (ret < 0) {
1230                         TAP_LOG(ERR,
1231                                 "%s: Couldn't delete MAC redirection rule",
1232                                 dev->device->name);
1233                         return ret;
1234                 }
1235                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1236                 if (ret < 0) {
1237                         TAP_LOG(ERR,
1238                                 "%s: Couldn't add MAC redirection rule",
1239                                 dev->device->name);
1240                         return ret;
1241                 }
1242         }
1243
1244         return 0;
1245 }
1246
1247 static int
1248 tap_gso_ctx_setup(struct rte_gso_ctx *gso_ctx, struct rte_eth_dev *dev)
1249 {
1250         uint32_t gso_types;
1251         char pool_name[64];
1252
1253         /*
1254          * Create private mbuf pool with TAP_GSO_MBUF_SEG_SIZE bytes
1255          * size per mbuf use this pool for both direct and indirect mbufs
1256          */
1257
1258         struct rte_mempool *mp;      /* Mempool for GSO packets */
1259
1260         /* initialize GSO context */
1261         gso_types = DEV_TX_OFFLOAD_TCP_TSO;
1262         snprintf(pool_name, sizeof(pool_name), "mp_%s", dev->device->name);
1263         mp = rte_mempool_lookup((const char *)pool_name);
1264         if (!mp) {
1265                 mp = rte_pktmbuf_pool_create(pool_name, TAP_GSO_MBUFS_NUM,
1266                         TAP_GSO_MBUF_CACHE_SIZE, 0,
1267                         RTE_PKTMBUF_HEADROOM + TAP_GSO_MBUF_SEG_SIZE,
1268                         SOCKET_ID_ANY);
1269                 if (!mp) {
1270                         struct pmd_internals *pmd = dev->data->dev_private;
1271                         RTE_LOG(DEBUG, PMD, "%s: failed to create mbuf pool for device %s\n",
1272                                 pmd->name, dev->device->name);
1273                         return -1;
1274                 }
1275         }
1276
1277         gso_ctx->direct_pool = mp;
1278         gso_ctx->indirect_pool = mp;
1279         gso_ctx->gso_types = gso_types;
1280         gso_ctx->gso_size = 0; /* gso_size is set in tx_burst() per packet */
1281         gso_ctx->flag = 0;
1282
1283         return 0;
1284 }
1285
1286 static int
1287 tap_setup_queue(struct rte_eth_dev *dev,
1288                 struct pmd_internals *internals,
1289                 uint16_t qid,
1290                 int is_rx)
1291 {
1292         int ret;
1293         int *fd;
1294         int *other_fd;
1295         const char *dir;
1296         struct pmd_internals *pmd = dev->data->dev_private;
1297         struct pmd_process_private *process_private = dev->process_private;
1298         struct rx_queue *rx = &internals->rxq[qid];
1299         struct tx_queue *tx = &internals->txq[qid];
1300         struct rte_gso_ctx *gso_ctx;
1301
1302         if (is_rx) {
1303                 fd = &process_private->rxq_fds[qid];
1304                 other_fd = &process_private->txq_fds[qid];
1305                 dir = "rx";
1306                 gso_ctx = NULL;
1307         } else {
1308                 fd = &process_private->txq_fds[qid];
1309                 other_fd = &process_private->rxq_fds[qid];
1310                 dir = "tx";
1311                 gso_ctx = &tx->gso_ctx;
1312         }
1313         if (*fd != -1) {
1314                 /* fd for this queue already exists */
1315                 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1316                         pmd->name, *fd, dir, qid);
1317                 gso_ctx = NULL;
1318         } else if (*other_fd != -1) {
1319                 /* Only other_fd exists. dup it */
1320                 *fd = dup(*other_fd);
1321                 if (*fd < 0) {
1322                         *fd = -1;
1323                         TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1324                         return -1;
1325                 }
1326                 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1327                         pmd->name, *other_fd, dir, qid, *fd);
1328         } else {
1329                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1330                 *fd = tun_alloc(pmd, 0);
1331                 if (*fd < 0) {
1332                         *fd = -1; /* restore original value */
1333                         TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1334                         return -1;
1335                 }
1336                 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1337                         pmd->name, dir, qid, *fd);
1338         }
1339
1340         tx->mtu = &dev->data->mtu;
1341         rx->rxmode = &dev->data->dev_conf.rxmode;
1342         if (gso_ctx) {
1343                 ret = tap_gso_ctx_setup(gso_ctx, dev);
1344                 if (ret)
1345                         return -1;
1346         }
1347
1348         tx->type = pmd->type;
1349
1350         return *fd;
1351 }
1352
1353 static int
1354 tap_rx_queue_setup(struct rte_eth_dev *dev,
1355                    uint16_t rx_queue_id,
1356                    uint16_t nb_rx_desc,
1357                    unsigned int socket_id,
1358                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1359                    struct rte_mempool *mp)
1360 {
1361         struct pmd_internals *internals = dev->data->dev_private;
1362         struct pmd_process_private *process_private = dev->process_private;
1363         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1364         struct rte_mbuf **tmp = &rxq->pool;
1365         long iov_max = sysconf(_SC_IOV_MAX);
1366
1367         if (iov_max <= 0) {
1368                 TAP_LOG(WARNING,
1369                         "_SC_IOV_MAX is not defined. Using %d as default",
1370                         TAP_IOV_DEFAULT_MAX);
1371                 iov_max = TAP_IOV_DEFAULT_MAX;
1372         }
1373         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1374         struct iovec (*iovecs)[nb_desc + 1];
1375         int data_off = RTE_PKTMBUF_HEADROOM;
1376         int ret = 0;
1377         int fd;
1378         int i;
1379
1380         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1381                 TAP_LOG(WARNING,
1382                         "nb_rx_queues %d too small or mempool NULL",
1383                         dev->data->nb_rx_queues);
1384                 return -1;
1385         }
1386
1387         rxq->mp = mp;
1388         rxq->trigger_seen = 1; /* force initial burst */
1389         rxq->in_port = dev->data->port_id;
1390         rxq->queue_id = rx_queue_id;
1391         rxq->nb_rx_desc = nb_desc;
1392         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1393                                     socket_id);
1394         if (!iovecs) {
1395                 TAP_LOG(WARNING,
1396                         "%s: Couldn't allocate %d RX descriptors",
1397                         dev->device->name, nb_desc);
1398                 return -ENOMEM;
1399         }
1400         rxq->iovecs = iovecs;
1401
1402         dev->data->rx_queues[rx_queue_id] = rxq;
1403         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1404         if (fd == -1) {
1405                 ret = fd;
1406                 goto error;
1407         }
1408
1409         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1410         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1411
1412         for (i = 1; i <= nb_desc; i++) {
1413                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1414                 if (!*tmp) {
1415                         TAP_LOG(WARNING,
1416                                 "%s: couldn't allocate memory for queue %d",
1417                                 dev->device->name, rx_queue_id);
1418                         ret = -ENOMEM;
1419                         goto error;
1420                 }
1421                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1422                 (*rxq->iovecs)[i].iov_base =
1423                         (char *)(*tmp)->buf_addr + data_off;
1424                 data_off = 0;
1425                 tmp = &(*tmp)->next;
1426         }
1427
1428         TAP_LOG(DEBUG, "  RX TUNTAP device name %s, qid %d on fd %d",
1429                 internals->name, rx_queue_id,
1430                 process_private->rxq_fds[rx_queue_id]);
1431
1432         return 0;
1433
1434 error:
1435         rte_pktmbuf_free(rxq->pool);
1436         rxq->pool = NULL;
1437         rte_free(rxq->iovecs);
1438         rxq->iovecs = NULL;
1439         return ret;
1440 }
1441
1442 static int
1443 tap_tx_queue_setup(struct rte_eth_dev *dev,
1444                    uint16_t tx_queue_id,
1445                    uint16_t nb_tx_desc __rte_unused,
1446                    unsigned int socket_id __rte_unused,
1447                    const struct rte_eth_txconf *tx_conf)
1448 {
1449         struct pmd_internals *internals = dev->data->dev_private;
1450         struct pmd_process_private *process_private = dev->process_private;
1451         struct tx_queue *txq;
1452         int ret;
1453         uint64_t offloads;
1454
1455         if (tx_queue_id >= dev->data->nb_tx_queues)
1456                 return -1;
1457         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1458         txq = dev->data->tx_queues[tx_queue_id];
1459         txq->out_port = dev->data->port_id;
1460         txq->queue_id = tx_queue_id;
1461
1462         offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1463         txq->csum = !!(offloads &
1464                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
1465                          DEV_TX_OFFLOAD_UDP_CKSUM |
1466                          DEV_TX_OFFLOAD_TCP_CKSUM));
1467
1468         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1469         if (ret == -1)
1470                 return -1;
1471         TAP_LOG(DEBUG,
1472                 "  TX TUNTAP device name %s, qid %d on fd %d csum %s",
1473                 internals->name, tx_queue_id,
1474                 process_private->txq_fds[tx_queue_id],
1475                 txq->csum ? "on" : "off");
1476
1477         return 0;
1478 }
1479
1480 static int
1481 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1482 {
1483         struct pmd_internals *pmd = dev->data->dev_private;
1484         struct ifreq ifr = { .ifr_mtu = mtu };
1485         int err = 0;
1486
1487         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1488         if (!err)
1489                 dev->data->mtu = mtu;
1490
1491         return err;
1492 }
1493
1494 static int
1495 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1496                      struct rte_ether_addr *mc_addr_set __rte_unused,
1497                      uint32_t nb_mc_addr __rte_unused)
1498 {
1499         /*
1500          * Nothing to do actually: the tap has no filtering whatsoever, every
1501          * packet is received.
1502          */
1503         return 0;
1504 }
1505
1506 static int
1507 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1508 {
1509         struct rte_eth_dev *dev = arg;
1510         struct pmd_internals *pmd = dev->data->dev_private;
1511         struct ifinfomsg *info = NLMSG_DATA(nh);
1512
1513         if (nh->nlmsg_type != RTM_NEWLINK ||
1514             (info->ifi_index != pmd->if_index &&
1515              info->ifi_index != pmd->remote_if_index))
1516                 return 0;
1517         return tap_link_update(dev, 0);
1518 }
1519
1520 static void
1521 tap_dev_intr_handler(void *cb_arg)
1522 {
1523         struct rte_eth_dev *dev = cb_arg;
1524         struct pmd_internals *pmd = dev->data->dev_private;
1525
1526         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1527 }
1528
1529 static int
1530 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1531 {
1532         struct pmd_internals *pmd = dev->data->dev_private;
1533
1534         /* In any case, disable interrupt if the conf is no longer there. */
1535         if (!dev->data->dev_conf.intr_conf.lsc) {
1536                 if (pmd->intr_handle.fd != -1) {
1537                         tap_nl_final(pmd->intr_handle.fd);
1538                         rte_intr_callback_unregister(&pmd->intr_handle,
1539                                 tap_dev_intr_handler, dev);
1540                 }
1541                 return 0;
1542         }
1543         if (set) {
1544                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1545                 if (unlikely(pmd->intr_handle.fd == -1))
1546                         return -EBADF;
1547                 return rte_intr_callback_register(
1548                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1549         }
1550         tap_nl_final(pmd->intr_handle.fd);
1551         return rte_intr_callback_unregister(&pmd->intr_handle,
1552                                             tap_dev_intr_handler, dev);
1553 }
1554
1555 static int
1556 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1557 {
1558         int err;
1559
1560         err = tap_lsc_intr_handle_set(dev, set);
1561         if (err)
1562                 return err;
1563         err = tap_rx_intr_vec_set(dev, set);
1564         if (err && set)
1565                 tap_lsc_intr_handle_set(dev, 0);
1566         return err;
1567 }
1568
1569 static const uint32_t*
1570 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1571 {
1572         static const uint32_t ptypes[] = {
1573                 RTE_PTYPE_INNER_L2_ETHER,
1574                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1575                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1576                 RTE_PTYPE_INNER_L3_IPV4,
1577                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1578                 RTE_PTYPE_INNER_L3_IPV6,
1579                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1580                 RTE_PTYPE_INNER_L4_FRAG,
1581                 RTE_PTYPE_INNER_L4_UDP,
1582                 RTE_PTYPE_INNER_L4_TCP,
1583                 RTE_PTYPE_INNER_L4_SCTP,
1584                 RTE_PTYPE_L2_ETHER,
1585                 RTE_PTYPE_L2_ETHER_VLAN,
1586                 RTE_PTYPE_L2_ETHER_QINQ,
1587                 RTE_PTYPE_L3_IPV4,
1588                 RTE_PTYPE_L3_IPV4_EXT,
1589                 RTE_PTYPE_L3_IPV6_EXT,
1590                 RTE_PTYPE_L3_IPV6,
1591                 RTE_PTYPE_L4_FRAG,
1592                 RTE_PTYPE_L4_UDP,
1593                 RTE_PTYPE_L4_TCP,
1594                 RTE_PTYPE_L4_SCTP,
1595         };
1596
1597         return ptypes;
1598 }
1599
1600 static int
1601 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1602                   struct rte_eth_fc_conf *fc_conf)
1603 {
1604         fc_conf->mode = RTE_FC_NONE;
1605         return 0;
1606 }
1607
1608 static int
1609 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1610                   struct rte_eth_fc_conf *fc_conf)
1611 {
1612         if (fc_conf->mode != RTE_FC_NONE)
1613                 return -ENOTSUP;
1614         return 0;
1615 }
1616
1617 /**
1618  * DPDK callback to update the RSS hash configuration.
1619  *
1620  * @param dev
1621  *   Pointer to Ethernet device structure.
1622  * @param[in] rss_conf
1623  *   RSS configuration data.
1624  *
1625  * @return
1626  *   0 on success, a negative errno value otherwise and rte_errno is set.
1627  */
1628 static int
1629 tap_rss_hash_update(struct rte_eth_dev *dev,
1630                 struct rte_eth_rss_conf *rss_conf)
1631 {
1632         if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1633                 rte_errno = EINVAL;
1634                 return -rte_errno;
1635         }
1636         if (rss_conf->rss_key && rss_conf->rss_key_len) {
1637                 /*
1638                  * Currently TAP RSS key is hard coded
1639                  * and cannot be updated
1640                  */
1641                 TAP_LOG(ERR,
1642                         "port %u RSS key cannot be updated",
1643                         dev->data->port_id);
1644                 rte_errno = EINVAL;
1645                 return -rte_errno;
1646         }
1647         return 0;
1648 }
1649
1650 static int
1651 tap_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1652 {
1653         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1654
1655         return 0;
1656 }
1657
1658 static int
1659 tap_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1660 {
1661         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1662
1663         return 0;
1664 }
1665
1666 static int
1667 tap_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1668 {
1669         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1670
1671         return 0;
1672 }
1673
1674 static int
1675 tap_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1676 {
1677         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1678
1679         return 0;
1680 }
1681 static const struct eth_dev_ops ops = {
1682         .dev_start              = tap_dev_start,
1683         .dev_stop               = tap_dev_stop,
1684         .dev_close              = tap_dev_close,
1685         .dev_configure          = tap_dev_configure,
1686         .dev_infos_get          = tap_dev_info,
1687         .rx_queue_setup         = tap_rx_queue_setup,
1688         .tx_queue_setup         = tap_tx_queue_setup,
1689         .rx_queue_start         = tap_rx_queue_start,
1690         .tx_queue_start         = tap_tx_queue_start,
1691         .rx_queue_stop          = tap_rx_queue_stop,
1692         .tx_queue_stop          = tap_tx_queue_stop,
1693         .rx_queue_release       = tap_rx_queue_release,
1694         .tx_queue_release       = tap_tx_queue_release,
1695         .flow_ctrl_get          = tap_flow_ctrl_get,
1696         .flow_ctrl_set          = tap_flow_ctrl_set,
1697         .link_update            = tap_link_update,
1698         .dev_set_link_up        = tap_link_set_up,
1699         .dev_set_link_down      = tap_link_set_down,
1700         .promiscuous_enable     = tap_promisc_enable,
1701         .promiscuous_disable    = tap_promisc_disable,
1702         .allmulticast_enable    = tap_allmulti_enable,
1703         .allmulticast_disable   = tap_allmulti_disable,
1704         .mac_addr_set           = tap_mac_set,
1705         .mtu_set                = tap_mtu_set,
1706         .set_mc_addr_list       = tap_set_mc_addr_list,
1707         .stats_get              = tap_stats_get,
1708         .stats_reset            = tap_stats_reset,
1709         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1710         .rss_hash_update        = tap_rss_hash_update,
1711         .filter_ctrl            = tap_dev_filter_ctrl,
1712 };
1713
1714 static const char *tuntap_types[ETH_TUNTAP_TYPE_MAX] = {
1715         "UNKNOWN", "TUN", "TAP"
1716 };
1717
1718 static int
1719 eth_dev_tap_create(struct rte_vdev_device *vdev, const char *tap_name,
1720                    char *remote_iface, struct rte_ether_addr *mac_addr,
1721                    enum rte_tuntap_type type)
1722 {
1723         int numa_node = rte_socket_id();
1724         struct rte_eth_dev *dev;
1725         struct pmd_internals *pmd;
1726         struct pmd_process_private *process_private;
1727         const char *tuntap_name = tuntap_types[type];
1728         struct rte_eth_dev_data *data;
1729         struct ifreq ifr;
1730         int i;
1731
1732         TAP_LOG(DEBUG, "%s device on numa %u", tuntap_name, rte_socket_id());
1733
1734         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1735         if (!dev) {
1736                 TAP_LOG(ERR, "%s Unable to allocate device struct",
1737                                 tuntap_name);
1738                 goto error_exit_nodev;
1739         }
1740
1741         process_private = (struct pmd_process_private *)
1742                 rte_zmalloc_socket(tap_name, sizeof(struct pmd_process_private),
1743                         RTE_CACHE_LINE_SIZE, dev->device->numa_node);
1744
1745         if (process_private == NULL) {
1746                 TAP_LOG(ERR, "Failed to alloc memory for process private");
1747                 return -1;
1748         }
1749         pmd = dev->data->dev_private;
1750         dev->process_private = process_private;
1751         pmd->dev = dev;
1752         strlcpy(pmd->name, tap_name, sizeof(pmd->name));
1753         pmd->type = type;
1754
1755         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1756         if (pmd->ioctl_sock == -1) {
1757                 TAP_LOG(ERR,
1758                         "%s Unable to get a socket for management: %s",
1759                         tuntap_name, strerror(errno));
1760                 goto error_exit;
1761         }
1762
1763         /* Setup some default values */
1764         data = dev->data;
1765         data->dev_private = pmd;
1766         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1767         data->numa_node = numa_node;
1768
1769         data->dev_link = pmd_link;
1770         data->mac_addrs = &pmd->eth_addr;
1771         /* Set the number of RX and TX queues */
1772         data->nb_rx_queues = 0;
1773         data->nb_tx_queues = 0;
1774
1775         dev->dev_ops = &ops;
1776         dev->rx_pkt_burst = pmd_rx_burst;
1777         dev->tx_pkt_burst = pmd_tx_burst;
1778
1779         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1780         pmd->intr_handle.fd = -1;
1781         dev->intr_handle = &pmd->intr_handle;
1782
1783         /* Presetup the fds to -1 as being not valid */
1784         pmd->ka_fd = -1;
1785         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1786                 process_private->rxq_fds[i] = -1;
1787                 process_private->txq_fds[i] = -1;
1788         }
1789
1790         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1791                 if (rte_is_zero_ether_addr(mac_addr))
1792                         rte_eth_random_addr((uint8_t *)&pmd->eth_addr);
1793                 else
1794                         rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1795         }
1796
1797         /*
1798          * Allocate a TUN device keep-alive file descriptor that will only be
1799          * closed when the TUN device itself is closed or removed.
1800          * This keep-alive file descriptor will guarantee that the TUN device
1801          * exists even when all of its queues are closed
1802          */
1803         pmd->ka_fd = tun_alloc(pmd, 1);
1804         if (pmd->ka_fd == -1) {
1805                 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
1806                 goto error_exit;
1807         }
1808         TAP_LOG(DEBUG, "allocated %s", pmd->name);
1809
1810         ifr.ifr_mtu = dev->data->mtu;
1811         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1812                 goto error_exit;
1813
1814         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1815                 memset(&ifr, 0, sizeof(struct ifreq));
1816                 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1817                 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1818                                 RTE_ETHER_ADDR_LEN);
1819                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1820                         goto error_exit;
1821         }
1822
1823         /*
1824          * Set up everything related to rte_flow:
1825          * - netlink socket
1826          * - tap / remote if_index
1827          * - mandatory QDISCs
1828          * - rte_flow actual/implicit lists
1829          * - implicit rules
1830          */
1831         pmd->nlsk_fd = tap_nl_init(0);
1832         if (pmd->nlsk_fd == -1) {
1833                 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
1834                         pmd->name);
1835                 goto disable_rte_flow;
1836         }
1837         pmd->if_index = if_nametoindex(pmd->name);
1838         if (!pmd->if_index) {
1839                 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
1840                 goto disable_rte_flow;
1841         }
1842         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1843                 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
1844                         pmd->name);
1845                 goto disable_rte_flow;
1846         }
1847         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1848                 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1849                         pmd->name);
1850                 goto disable_rte_flow;
1851         }
1852         LIST_INIT(&pmd->flows);
1853
1854         if (strlen(remote_iface)) {
1855                 pmd->remote_if_index = if_nametoindex(remote_iface);
1856                 if (!pmd->remote_if_index) {
1857                         TAP_LOG(ERR, "%s: failed to get %s if_index.",
1858                                 pmd->name, remote_iface);
1859                         goto error_remote;
1860                 }
1861                 strlcpy(pmd->remote_iface, remote_iface, RTE_ETH_NAME_MAX_LEN);
1862
1863                 /* Save state of remote device */
1864                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1865
1866                 /* Replicate remote MAC address */
1867                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1868                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1869                                 pmd->name, pmd->remote_iface);
1870                         goto error_remote;
1871                 }
1872                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1873                            RTE_ETHER_ADDR_LEN);
1874                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1875                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1876                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1877                                 pmd->name, remote_iface);
1878                         goto error_remote;
1879                 }
1880
1881                 /*
1882                  * Flush usually returns negative value because it tries to
1883                  * delete every QDISC (and on a running device, one QDISC at
1884                  * least is needed). Ignore negative return value.
1885                  */
1886                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1887                 if (qdisc_create_ingress(pmd->nlsk_fd,
1888                                          pmd->remote_if_index) < 0) {
1889                         TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1890                                 pmd->remote_iface);
1891                         goto error_remote;
1892                 }
1893                 LIST_INIT(&pmd->implicit_flows);
1894                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1895                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1896                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1897                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1898                         TAP_LOG(ERR,
1899                                 "%s: failed to create implicit rules.",
1900                                 pmd->name);
1901                         goto error_remote;
1902                 }
1903         }
1904
1905         rte_eth_dev_probing_finish(dev);
1906         return 0;
1907
1908 disable_rte_flow:
1909         TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
1910                 strerror(errno), errno);
1911         if (strlen(remote_iface)) {
1912                 TAP_LOG(ERR, "Remote feature requires flow support.");
1913                 goto error_exit;
1914         }
1915         rte_eth_dev_probing_finish(dev);
1916         return 0;
1917
1918 error_remote:
1919         TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
1920                 strerror(errno), errno);
1921         tap_flow_implicit_flush(pmd, NULL);
1922
1923 error_exit:
1924         if (pmd->ioctl_sock > 0)
1925                 close(pmd->ioctl_sock);
1926         /* mac_addrs must not be freed alone because part of dev_private */
1927         dev->data->mac_addrs = NULL;
1928         rte_eth_dev_release_port(dev);
1929
1930 error_exit_nodev:
1931         TAP_LOG(ERR, "%s Unable to initialize %s",
1932                 tuntap_name, rte_vdev_device_name(vdev));
1933
1934         return -EINVAL;
1935 }
1936
1937 /* make sure name is a possible Linux network device name */
1938 static bool
1939 is_valid_iface(const char *name)
1940 {
1941         if (*name == '\0')
1942                 return false;
1943
1944         if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1945                 return false;
1946
1947         while (*name) {
1948                 if (*name == '/' || *name == ':' || isspace(*name))
1949                         return false;
1950                 name++;
1951         }
1952         return true;
1953 }
1954
1955 static int
1956 set_interface_name(const char *key __rte_unused,
1957                    const char *value,
1958                    void *extra_args)
1959 {
1960         char *name = (char *)extra_args;
1961
1962         if (value) {
1963                 if (!is_valid_iface(value)) {
1964                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
1965                                 value);
1966                         return -1;
1967                 }
1968                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
1969         } else {
1970                 /* use tap%d which causes kernel to choose next available */
1971                 strlcpy(name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
1972         }
1973         return 0;
1974 }
1975
1976 static int
1977 set_remote_iface(const char *key __rte_unused,
1978                  const char *value,
1979                  void *extra_args)
1980 {
1981         char *name = (char *)extra_args;
1982
1983         if (value) {
1984                 if (!is_valid_iface(value)) {
1985                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
1986                                 value);
1987                         return -1;
1988                 }
1989                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
1990         }
1991
1992         return 0;
1993 }
1994
1995 static int parse_user_mac(struct rte_ether_addr *user_mac,
1996                 const char *value)
1997 {
1998         unsigned int index = 0;
1999         char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
2000
2001         if (user_mac == NULL || value == NULL)
2002                 return 0;
2003
2004         strlcpy(mac_temp, value, sizeof(mac_temp));
2005         mac_byte = strtok(mac_temp, ":");
2006
2007         while ((mac_byte != NULL) &&
2008                         (strlen(mac_byte) <= 2) &&
2009                         (strlen(mac_byte) == strspn(mac_byte,
2010                                         ETH_TAP_CMP_MAC_FMT))) {
2011                 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
2012                 mac_byte = strtok(NULL, ":");
2013         }
2014
2015         return index;
2016 }
2017
2018 static int
2019 set_mac_type(const char *key __rte_unused,
2020              const char *value,
2021              void *extra_args)
2022 {
2023         struct rte_ether_addr *user_mac = extra_args;
2024
2025         if (!value)
2026                 return 0;
2027
2028         if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
2029                 static int iface_idx;
2030
2031                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
2032                 memcpy((char *)user_mac->addr_bytes, "\0dtap",
2033                         RTE_ETHER_ADDR_LEN);
2034                 user_mac->addr_bytes[RTE_ETHER_ADDR_LEN - 1] =
2035                         iface_idx++ + '0';
2036                 goto success;
2037         }
2038
2039         if (parse_user_mac(user_mac, value) != 6)
2040                 goto error;
2041 success:
2042         TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
2043         return 0;
2044
2045 error:
2046         TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
2047                 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
2048         return -1;
2049 }
2050
2051 /*
2052  * Open a TUN interface device. TUN PMD
2053  * 1) sets tap_type as false
2054  * 2) intakes iface as argument.
2055  * 3) as interface is virtual set speed to 10G
2056  */
2057 static int
2058 rte_pmd_tun_probe(struct rte_vdev_device *dev)
2059 {
2060         const char *name, *params;
2061         int ret;
2062         struct rte_kvargs *kvlist = NULL;
2063         char tun_name[RTE_ETH_NAME_MAX_LEN];
2064         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2065         struct rte_eth_dev *eth_dev;
2066
2067         name = rte_vdev_device_name(dev);
2068         params = rte_vdev_device_args(dev);
2069         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2070
2071         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
2072             strlen(params) == 0) {
2073                 eth_dev = rte_eth_dev_attach_secondary(name);
2074                 if (!eth_dev) {
2075                         TAP_LOG(ERR, "Failed to probe %s", name);
2076                         return -1;
2077                 }
2078                 eth_dev->dev_ops = &ops;
2079                 eth_dev->device = &dev->device;
2080                 rte_eth_dev_probing_finish(eth_dev);
2081                 return 0;
2082         }
2083
2084         /* use tun%d which causes kernel to choose next available */
2085         strlcpy(tun_name, DEFAULT_TUN_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2086
2087         if (params && (params[0] != '\0')) {
2088                 TAP_LOG(DEBUG, "parameters (%s)", params);
2089
2090                 kvlist = rte_kvargs_parse(params, valid_arguments);
2091                 if (kvlist) {
2092                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2093                                 ret = rte_kvargs_process(kvlist,
2094                                         ETH_TAP_IFACE_ARG,
2095                                         &set_interface_name,
2096                                         tun_name);
2097
2098                                 if (ret == -1)
2099                                         goto leave;
2100                         }
2101                 }
2102         }
2103         pmd_link.link_speed = ETH_SPEED_NUM_10G;
2104
2105         TAP_LOG(DEBUG, "Initializing pmd_tun for %s", name);
2106
2107         ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0,
2108                                  ETH_TUNTAP_TYPE_TUN);
2109
2110 leave:
2111         if (ret == -1) {
2112                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2113                         name, tun_name);
2114         }
2115         rte_kvargs_free(kvlist);
2116
2117         return ret;
2118 }
2119
2120 /* Request queue file descriptors from secondary to primary. */
2121 static int
2122 tap_mp_attach_queues(const char *port_name, struct rte_eth_dev *dev)
2123 {
2124         int ret;
2125         struct timespec timeout = {.tv_sec = 1, .tv_nsec = 0};
2126         struct rte_mp_msg request, *reply;
2127         struct rte_mp_reply replies;
2128         struct ipc_queues *request_param = (struct ipc_queues *)request.param;
2129         struct ipc_queues *reply_param;
2130         struct pmd_process_private *process_private = dev->process_private;
2131         int queue, fd_iterator;
2132
2133         /* Prepare the request */
2134         memset(&request, 0, sizeof(request));
2135         strlcpy(request.name, TAP_MP_KEY, sizeof(request.name));
2136         strlcpy(request_param->port_name, port_name,
2137                 sizeof(request_param->port_name));
2138         request.len_param = sizeof(*request_param);
2139         /* Send request and receive reply */
2140         ret = rte_mp_request_sync(&request, &replies, &timeout);
2141         if (ret < 0 || replies.nb_received != 1) {
2142                 TAP_LOG(ERR, "Failed to request queues from primary: %d",
2143                         rte_errno);
2144                 return -1;
2145         }
2146         reply = &replies.msgs[0];
2147         reply_param = (struct ipc_queues *)reply->param;
2148         TAP_LOG(DEBUG, "Received IPC reply for %s", reply_param->port_name);
2149
2150         /* Attach the queues from received file descriptors */
2151         if (reply_param->rxq_count + reply_param->txq_count != reply->num_fds) {
2152                 TAP_LOG(ERR, "Unexpected number of fds received");
2153                 return -1;
2154         }
2155
2156         dev->data->nb_rx_queues = reply_param->rxq_count;
2157         dev->data->nb_tx_queues = reply_param->txq_count;
2158         fd_iterator = 0;
2159         for (queue = 0; queue < reply_param->rxq_count; queue++)
2160                 process_private->rxq_fds[queue] = reply->fds[fd_iterator++];
2161         for (queue = 0; queue < reply_param->txq_count; queue++)
2162                 process_private->txq_fds[queue] = reply->fds[fd_iterator++];
2163         free(reply);
2164         return 0;
2165 }
2166
2167 /* Send the queue file descriptors from the primary process to secondary. */
2168 static int
2169 tap_mp_sync_queues(const struct rte_mp_msg *request, const void *peer)
2170 {
2171         struct rte_eth_dev *dev;
2172         struct pmd_process_private *process_private;
2173         struct rte_mp_msg reply;
2174         const struct ipc_queues *request_param =
2175                 (const struct ipc_queues *)request->param;
2176         struct ipc_queues *reply_param =
2177                 (struct ipc_queues *)reply.param;
2178         uint16_t port_id;
2179         int queue;
2180         int ret;
2181
2182         /* Get requested port */
2183         TAP_LOG(DEBUG, "Received IPC request for %s", request_param->port_name);
2184         ret = rte_eth_dev_get_port_by_name(request_param->port_name, &port_id);
2185         if (ret) {
2186                 TAP_LOG(ERR, "Failed to get port id for %s",
2187                         request_param->port_name);
2188                 return -1;
2189         }
2190         dev = &rte_eth_devices[port_id];
2191         process_private = dev->process_private;
2192
2193         /* Fill file descriptors for all queues */
2194         reply.num_fds = 0;
2195         reply_param->rxq_count = 0;
2196         if (dev->data->nb_rx_queues + dev->data->nb_tx_queues >
2197                         RTE_MP_MAX_FD_NUM){
2198                 TAP_LOG(ERR, "Number of rx/tx queues exceeds max number of fds");
2199                 return -1;
2200         }
2201
2202         for (queue = 0; queue < dev->data->nb_rx_queues; queue++) {
2203                 reply.fds[reply.num_fds++] = process_private->rxq_fds[queue];
2204                 reply_param->rxq_count++;
2205         }
2206         RTE_ASSERT(reply_param->rxq_count == dev->data->nb_rx_queues);
2207
2208         reply_param->txq_count = 0;
2209         for (queue = 0; queue < dev->data->nb_tx_queues; queue++) {
2210                 reply.fds[reply.num_fds++] = process_private->txq_fds[queue];
2211                 reply_param->txq_count++;
2212         }
2213         RTE_ASSERT(reply_param->txq_count == dev->data->nb_tx_queues);
2214
2215         /* Send reply */
2216         strlcpy(reply.name, request->name, sizeof(reply.name));
2217         strlcpy(reply_param->port_name, request_param->port_name,
2218                 sizeof(reply_param->port_name));
2219         reply.len_param = sizeof(*reply_param);
2220         if (rte_mp_reply(&reply, peer) < 0) {
2221                 TAP_LOG(ERR, "Failed to reply an IPC request to sync queues");
2222                 return -1;
2223         }
2224         return 0;
2225 }
2226
2227 /* Open a TAP interface device.
2228  */
2229 static int
2230 rte_pmd_tap_probe(struct rte_vdev_device *dev)
2231 {
2232         const char *name, *params;
2233         int ret;
2234         struct rte_kvargs *kvlist = NULL;
2235         int speed;
2236         char tap_name[RTE_ETH_NAME_MAX_LEN];
2237         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2238         struct rte_ether_addr user_mac = { .addr_bytes = {0} };
2239         struct rte_eth_dev *eth_dev;
2240         int tap_devices_count_increased = 0;
2241
2242         name = rte_vdev_device_name(dev);
2243         params = rte_vdev_device_args(dev);
2244
2245         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
2246                 eth_dev = rte_eth_dev_attach_secondary(name);
2247                 if (!eth_dev) {
2248                         TAP_LOG(ERR, "Failed to probe %s", name);
2249                         return -1;
2250                 }
2251                 eth_dev->dev_ops = &ops;
2252                 eth_dev->device = &dev->device;
2253                 eth_dev->rx_pkt_burst = pmd_rx_burst;
2254                 eth_dev->tx_pkt_burst = pmd_tx_burst;
2255                 if (!rte_eal_primary_proc_alive(NULL)) {
2256                         TAP_LOG(ERR, "Primary process is missing");
2257                         return -1;
2258                 }
2259                 eth_dev->process_private = (struct pmd_process_private *)
2260                         rte_zmalloc_socket(name,
2261                                 sizeof(struct pmd_process_private),
2262                                 RTE_CACHE_LINE_SIZE,
2263                                 eth_dev->device->numa_node);
2264                 if (eth_dev->process_private == NULL) {
2265                         TAP_LOG(ERR,
2266                                 "Failed to alloc memory for process private");
2267                         return -1;
2268                 }
2269
2270                 ret = tap_mp_attach_queues(name, eth_dev);
2271                 if (ret != 0)
2272                         return -1;
2273                 rte_eth_dev_probing_finish(eth_dev);
2274                 return 0;
2275         }
2276
2277         speed = ETH_SPEED_NUM_10G;
2278
2279         /* use tap%d which causes kernel to choose next available */
2280         strlcpy(tap_name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2281         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2282
2283         if (params && (params[0] != '\0')) {
2284                 TAP_LOG(DEBUG, "parameters (%s)", params);
2285
2286                 kvlist = rte_kvargs_parse(params, valid_arguments);
2287                 if (kvlist) {
2288                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2289                                 ret = rte_kvargs_process(kvlist,
2290                                                          ETH_TAP_IFACE_ARG,
2291                                                          &set_interface_name,
2292                                                          tap_name);
2293                                 if (ret == -1)
2294                                         goto leave;
2295                         }
2296
2297                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
2298                                 ret = rte_kvargs_process(kvlist,
2299                                                          ETH_TAP_REMOTE_ARG,
2300                                                          &set_remote_iface,
2301                                                          remote_iface);
2302                                 if (ret == -1)
2303                                         goto leave;
2304                         }
2305
2306                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
2307                                 ret = rte_kvargs_process(kvlist,
2308                                                          ETH_TAP_MAC_ARG,
2309                                                          &set_mac_type,
2310                                                          &user_mac);
2311                                 if (ret == -1)
2312                                         goto leave;
2313                         }
2314                 }
2315         }
2316         pmd_link.link_speed = speed;
2317
2318         TAP_LOG(DEBUG, "Initializing pmd_tap for %s", name);
2319
2320         /* Register IPC feed callback */
2321         if (!tap_devices_count) {
2322                 ret = rte_mp_action_register(TAP_MP_KEY, tap_mp_sync_queues);
2323                 if (ret < 0 && rte_errno != ENOTSUP) {
2324                         TAP_LOG(ERR, "tap: Failed to register IPC callback: %s",
2325                                 strerror(rte_errno));
2326                         goto leave;
2327                 }
2328         }
2329         tap_devices_count++;
2330         tap_devices_count_increased = 1;
2331         ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
2332                 ETH_TUNTAP_TYPE_TAP);
2333
2334 leave:
2335         if (ret == -1) {
2336                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2337                         name, tap_name);
2338                 if (tap_devices_count_increased == 1) {
2339                         if (tap_devices_count == 1)
2340                                 rte_mp_action_unregister(TAP_MP_KEY);
2341                         tap_devices_count--;
2342                 }
2343         }
2344         rte_kvargs_free(kvlist);
2345
2346         return ret;
2347 }
2348
2349 /* detach a TUNTAP device.
2350  */
2351 static int
2352 rte_pmd_tap_remove(struct rte_vdev_device *dev)
2353 {
2354         struct rte_eth_dev *eth_dev = NULL;
2355         struct pmd_internals *internals;
2356         struct pmd_process_private *process_private;
2357         int i;
2358
2359         /* find the ethdev entry */
2360         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
2361         if (!eth_dev)
2362                 return -ENODEV;
2363
2364         /* mac_addrs must not be freed alone because part of dev_private */
2365         eth_dev->data->mac_addrs = NULL;
2366
2367         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2368                 return rte_eth_dev_release_port(eth_dev);
2369
2370         internals = eth_dev->data->dev_private;
2371         process_private = eth_dev->process_private;
2372
2373         TAP_LOG(DEBUG, "Closing %s Ethernet device on numa %u",
2374                 tuntap_types[internals->type], rte_socket_id());
2375
2376         if (internals->nlsk_fd) {
2377                 tap_flow_flush(eth_dev, NULL);
2378                 tap_flow_implicit_flush(internals, NULL);
2379                 tap_nl_final(internals->nlsk_fd);
2380         }
2381         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
2382                 if (process_private->rxq_fds[i] != -1) {
2383                         close(process_private->rxq_fds[i]);
2384                         process_private->rxq_fds[i] = -1;
2385                 }
2386                 if (process_private->txq_fds[i] != -1) {
2387                         close(process_private->txq_fds[i]);
2388                         process_private->txq_fds[i] = -1;
2389                 }
2390         }
2391
2392         close(internals->ioctl_sock);
2393         rte_free(eth_dev->process_private);
2394         if (tap_devices_count == 1)
2395                 rte_mp_action_unregister(TAP_MP_KEY);
2396         tap_devices_count--;
2397         rte_eth_dev_release_port(eth_dev);
2398
2399         if (internals->ka_fd != -1) {
2400                 close(internals->ka_fd);
2401                 internals->ka_fd = -1;
2402         }
2403         return 0;
2404 }
2405
2406 static struct rte_vdev_driver pmd_tun_drv = {
2407         .probe = rte_pmd_tun_probe,
2408         .remove = rte_pmd_tap_remove,
2409 };
2410
2411 static struct rte_vdev_driver pmd_tap_drv = {
2412         .probe = rte_pmd_tap_probe,
2413         .remove = rte_pmd_tap_remove,
2414 };
2415
2416 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
2417 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
2418 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
2419 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
2420                               ETH_TAP_IFACE_ARG "=<string> ");
2421 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
2422                               ETH_TAP_IFACE_ARG "=<string> "
2423                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
2424                               ETH_TAP_REMOTE_ARG "=<string>");
2425 int tap_logtype;
2426
2427 RTE_INIT(tap_init_log)
2428 {
2429         tap_logtype = rte_log_register("pmd.net.tap");
2430         if (tap_logtype >= 0)
2431                 rte_log_set_level(tap_logtype, RTE_LOG_NOTICE);
2432 }