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