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