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