drivers/net/tap/rte_eth_tap.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2016-2017 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <rte_atomic.h>
  35 #include <rte_branch_prediction.h>
  36 #include <rte_byteorder.h>
  37 #include <rte_common.h>
  38 #include <rte_mbuf.h>
  39 #include <rte_ethdev.h>
  40 #include <rte_ethdev_vdev.h>
  41 #include <rte_malloc.h>
  42 #include <rte_vdev.h>
  43 #include <rte_kvargs.h>
  44 #include <rte_net.h>
  45 #include <rte_debug.h>
  46 #include <rte_ip.h>
  47
  48 #include <sys/types.h>
  49 #include <sys/stat.h>
  50 #include <sys/socket.h>
  51 #include <sys/ioctl.h>
  52 #include <sys/utsname.h>
  53 #include <sys/mman.h>
  54 #include <errno.h>
  55 #include <signal.h>
  56 #include <stdint.h>
  57 #include <sys/uio.h>
  58 #include <unistd.h>
  59 #include <arpa/inet.h>
  60 #include <net/if.h>
  61 #include <linux/if_tun.h>
  62 #include <linux/if_ether.h>
  63 #include <linux/version.h>
  64 #include <fcntl.h>
  65
  66 #include <rte_eth_tap.h>
  67 #include <tap_flow.h>
  68 #include <tap_netlink.h>
  69 #include <tap_tcmsgs.h>
  70
  71 /* Linux based path to the TUN device */
  72 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  73 #define DEFAULT_TAP_NAME        "dtap"
  74
  75 #define ETH_TAP_IFACE_ARG       "iface"
  76 #define ETH_TAP_SPEED_ARG       "speed"
  77 #define ETH_TAP_REMOTE_ARG      "remote"
  78 #define ETH_TAP_MAC_ARG         "mac"
  79 #define ETH_TAP_MAC_FIXED       "fixed"
  80
  81 #define FLOWER_KERNEL_VERSION KERNEL_VERSION(4, 2, 0)
  82 #define FLOWER_VLAN_KERNEL_VERSION KERNEL_VERSION(4, 9, 0)
  83
  84 static struct rte_vdev_driver pmd_tap_drv;
  85
  86 static const char *valid_arguments[] = {
  87         ETH_TAP_IFACE_ARG,
  88         ETH_TAP_SPEED_ARG,
  89         ETH_TAP_REMOTE_ARG,
  90         ETH_TAP_MAC_ARG,
  91         NULL
  92 };
  93
  94 static int tap_unit;
  95
  96 static volatile uint32_t tap_trigger;   /* Rx trigger */
  97
  98 static struct rte_eth_link pmd_link = {
  99         .link_speed = ETH_SPEED_NUM_10G,
 100         .link_duplex = ETH_LINK_FULL_DUPLEX,
 101         .link_status = ETH_LINK_DOWN,
 102         .link_autoneg = ETH_LINK_SPEED_AUTONEG
 103 };
 104
 105 static void
 106 tap_trigger_cb(int sig __rte_unused)
 107 {
 108         /* Valid trigger values are nonzero */
 109         tap_trigger = (tap_trigger + 1) | 0x80000000;
 110 }
 111
 112 /* Specifies on what netdevices the ioctl should be applied */
 113 enum ioctl_mode {
 114         LOCAL_AND_REMOTE,
 115         LOCAL_ONLY,
 116         REMOTE_ONLY,
 117 };
 118
 119 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
 120
 121 /* Tun/Tap allocation routine
 122  *
 123  * name is the number of the interface to use, unless NULL to take the host
 124  * supplied name.
 125  */
 126 static int
 127 tun_alloc(struct pmd_internals *pmd)
 128 {
 129         struct ifreq ifr;
 130 #ifdef IFF_MULTI_QUEUE
 131         unsigned int features;
 132 #endif
 133         int fd;
 134
 135         memset(&ifr, 0, sizeof(struct ifreq));
 136
 137         /*
 138          * Do not set IFF_NO_PI as packet information header will be needed
 139          * to check if a received packet has been truncated.
 140          */
 141         ifr.ifr_flags = IFF_TAP;
 142         snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
 143
 144         RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
 145
 146         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 147         if (fd < 0) {
 148                 RTE_LOG(ERR, PMD, "Unable to create TAP interface");
 149                 goto error;
 150         }
 151
 152 #ifdef IFF_MULTI_QUEUE
 153         /* Grab the TUN features to verify we can work multi-queue */
 154         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 155                 RTE_LOG(ERR, PMD, "TAP unable to get TUN/TAP features\n");
 156                 goto error;
 157         }
 158         RTE_LOG(DEBUG, PMD, "  TAP Features %08x\n", features);
 159
 160         if (features & IFF_MULTI_QUEUE) {
 161                 RTE_LOG(DEBUG, PMD, "  Multi-queue support for %d queues\n",
 162                         RTE_PMD_TAP_MAX_QUEUES);
 163                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 164         } else
 165 #endif
 166         {
 167                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 168                 RTE_LOG(DEBUG, PMD, "  Single queue only support\n");
 169         }
 170
 171         /* Set the TUN/TAP configuration and set the name if needed */
 172         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 173                 RTE_LOG(WARNING, PMD,
 174                         "Unable to set TUNSETIFF for %s\n",
 175                         ifr.ifr_name);
 176                 perror("TUNSETIFF");
 177                 goto error;
 178         }
 179
 180         /* Always set the file descriptor to non-blocking */
 181         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 182                 RTE_LOG(WARNING, PMD,
 183                         "Unable to set %s to nonblocking\n",
 184                         ifr.ifr_name);
 185                 perror("F_SETFL, NONBLOCK");
 186                 goto error;
 187         }
 188
 189         /* Set up trigger to optimize empty Rx bursts */
 190         errno = 0;
 191         do {
 192                 struct sigaction sa;
 193                 int flags = fcntl(fd, F_GETFL);
 194
 195                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 196                         break;
 197                 if (sa.sa_handler != tap_trigger_cb) {
 198                         /*
 199                          * Make sure SIGIO is not already taken. This is done
 200                          * as late as possible to leave the application a
 201                          * chance to set up its own signal handler first.
 202                          */
 203                         if (sa.sa_handler != SIG_IGN &&
 204                             sa.sa_handler != SIG_DFL) {
 205                                 errno = EBUSY;
 206                                 break;
 207                         }
 208                         sa = (struct sigaction){
 209                                 .sa_flags = SA_RESTART,
 210                                 .sa_handler = tap_trigger_cb,
 211                         };
 212                         if (sigaction(SIGIO, &sa, NULL) == -1)
 213                                 break;
 214                 }
 215                 /* Enable SIGIO on file descriptor */
 216                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 217                 fcntl(fd, F_SETOWN, getpid());
 218         } while (0);
 219         if (errno) {
 220                 /* Disable trigger globally in case of error */
 221                 tap_trigger = 0;
 222                 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
 223                         strerror(errno));
 224         }
 225
 226         return fd;
 227
 228 error:
 229         if (fd > 0)
 230                 close(fd);
 231         return -1;
 232 }
 233
 234 static void
 235 tap_verify_csum(struct rte_mbuf *mbuf)
 236 {
 237         uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
 238         uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
 239         uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
 240         unsigned int l2_len = sizeof(struct ether_hdr);
 241         unsigned int l3_len;
 242         uint16_t cksum = 0;
 243         void *l3_hdr;
 244         void *l4_hdr;
 245
 246         if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
 247                 l2_len += 4;
 248         else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
 249                 l2_len += 8;
 250         /* Don't verify checksum for packets with discontinuous L2 header */
 251         if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
 252                      rte_pktmbuf_data_len(mbuf)))
 253                 return;
 254         l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
 255         if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 256                 struct ipv4_hdr *iph = l3_hdr;
 257
 258                 /* ihl contains the number of 4-byte words in the header */
 259                 l3_len = 4 * (iph->version_ihl & 0xf);
 260                 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
 261                         return;
 262
 263                 cksum = ~rte_raw_cksum(iph, l3_len);
 264                 mbuf->ol_flags |= cksum ?
 265                         PKT_RX_IP_CKSUM_BAD :
 266                         PKT_RX_IP_CKSUM_GOOD;
 267         } else if (l3 == RTE_PTYPE_L3_IPV6) {
 268                 l3_len = sizeof(struct ipv6_hdr);
 269         } else {
 270                 /* IPv6 extensions are not supported */
 271                 return;
 272         }
 273         if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
 274                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 275                 /* Don't verify checksum for multi-segment packets. */
 276                 if (mbuf->nb_segs > 1)
 277                         return;
 278                 if (l3 == RTE_PTYPE_L3_IPV4)
 279                         cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
 280                 else if (l3 == RTE_PTYPE_L3_IPV6)
 281                         cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
 282                 mbuf->ol_flags |= cksum ?
 283                         PKT_RX_L4_CKSUM_BAD :
 284                         PKT_RX_L4_CKSUM_GOOD;
 285         }
 286 }
 287
 288 /* Callback to handle the rx burst of packets to the correct interface and
 289  * file descriptor(s) in a multi-queue setup.
 290  */
 291 static uint16_t
 292 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 293 {
 294         struct rx_queue *rxq = queue;
 295         uint16_t num_rx;
 296         unsigned long num_rx_bytes = 0;
 297         uint32_t trigger = tap_trigger;
 298
 299         if (trigger == rxq->trigger_seen)
 300                 return 0;
 301         if (trigger)
 302                 rxq->trigger_seen = trigger;
 303         rte_compiler_barrier();
 304         for (num_rx = 0; num_rx < nb_pkts; ) {
 305                 struct rte_mbuf *mbuf = rxq->pool;
 306                 struct rte_mbuf *seg = NULL;
 307                 struct rte_mbuf *new_tail = NULL;
 308                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 309                 int len;
 310
 311                 len = readv(rxq->fd, *rxq->iovecs,
 312                             1 + (rxq->rxmode->enable_scatter ?
 313                                  rxq->nb_rx_desc : 1));
 314                 if (len < (int)sizeof(struct tun_pi))
 315                         break;
 316
 317                 /* Packet couldn't fit in the provided mbuf */
 318                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 319                         rxq->stats.ierrors++;
 320                         continue;
 321                 }
 322
 323                 len -= sizeof(struct tun_pi);
 324
 325                 mbuf->pkt_len = len;
 326                 mbuf->port = rxq->in_port;
 327                 while (1) {
 328                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 329
 330                         if (unlikely(!buf)) {
 331                                 rxq->stats.rx_nombuf++;
 332                                 /* No new buf has been allocated: do nothing */
 333                                 if (!new_tail || !seg)
 334                                         goto end;
 335
 336                                 seg->next = NULL;
 337                                 rte_pktmbuf_free(mbuf);
 338
 339                                 goto end;
 340                         }
 341                         seg = seg ? seg->next : mbuf;
 342                         if (rxq->pool == mbuf)
 343                                 rxq->pool = buf;
 344                         if (new_tail)
 345                                 new_tail->next = buf;
 346                         new_tail = buf;
 347                         new_tail->next = seg->next;
 348
 349                         /* iovecs[0] is reserved for packet info (pi) */
 350                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 351                                 buf->buf_len - data_off;
 352                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 353                                 (char *)buf->buf_addr + data_off;
 354
 355                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 356                         seg->data_off = data_off;
 357
 358                         len -= seg->data_len;
 359                         if (len <= 0)
 360                                 break;
 361                         mbuf->nb_segs++;
 362                         /* First segment has headroom, not the others */
 363                         data_off = 0;
 364                 }
 365                 seg->next = NULL;
 366                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 367                                                       RTE_PTYPE_ALL_MASK);
 368                 if (rxq->rxmode->hw_ip_checksum)
 369                         tap_verify_csum(mbuf);
 370
 371                 /* account for the receive frame */
 372                 bufs[num_rx++] = mbuf;
 373                 num_rx_bytes += mbuf->pkt_len;
 374         }
 375 end:
 376         rxq->stats.ipackets += num_rx;
 377         rxq->stats.ibytes += num_rx_bytes;
 378
 379         return num_rx;
 380 }
 381
 382 static void
 383 tap_tx_offload(char *packet, uint64_t ol_flags, unsigned int l2_len,
 384                unsigned int l3_len)
 385 {
 386         void *l3_hdr = packet + l2_len;
 387
 388         if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
 389                 struct ipv4_hdr *iph = l3_hdr;
 390                 uint16_t cksum;
 391
 392                 iph->hdr_checksum = 0;
 393                 cksum = rte_raw_cksum(iph, l3_len);
 394                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 395         }
 396         if (ol_flags & PKT_TX_L4_MASK) {
 397                 uint16_t l4_len;
 398                 uint32_t cksum;
 399                 uint16_t *l4_cksum;
 400                 void *l4_hdr;
 401
 402                 l4_hdr = packet + l2_len + l3_len;
 403                 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
 404                         l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
 405                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 406                         l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
 407                 else
 408                         return;
 409                 *l4_cksum = 0;
 410                 if (ol_flags & PKT_TX_IPV4) {
 411                         struct ipv4_hdr *iph = l3_hdr;
 412
 413                         l4_len = rte_be_to_cpu_16(iph->total_length) - l3_len;
 414                         cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 415                 } else {
 416                         struct ipv6_hdr *ip6h = l3_hdr;
 417
 418                         /* payload_len does not include ext headers */
 419                         l4_len = rte_be_to_cpu_16(ip6h->payload_len) -
 420                                 l3_len + sizeof(struct ipv6_hdr);
 421                         cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 422                 }
 423                 cksum += rte_raw_cksum(l4_hdr, l4_len);
 424                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 425                 cksum = (~cksum) & 0xffff;
 426                 if (cksum == 0)
 427                         cksum = 0xffff;
 428                 *l4_cksum = cksum;
 429         }
 430 }
 431
 432 /* Callback to handle sending packets from the tap interface
 433  */
 434 static uint16_t
 435 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 436 {
 437         struct tx_queue *txq = queue;
 438         uint16_t num_tx = 0;
 439         unsigned long num_tx_bytes = 0;
 440         uint32_t max_size;
 441         int i;
 442
 443         if (unlikely(nb_pkts == 0))
 444                 return 0;
 445
 446         max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
 447         for (i = 0; i < nb_pkts; i++) {
 448                 struct rte_mbuf *mbuf = bufs[num_tx];
 449                 struct iovec iovecs[mbuf->nb_segs + 1];
 450                 struct tun_pi pi = { .flags = 0 };
 451                 struct rte_mbuf *seg = mbuf;
 452                 char m_copy[mbuf->data_len];
 453                 int n;
 454                 int j;
 455
 456                 /* stats.errs will be incremented */
 457                 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
 458                         break;
 459
 460                 iovecs[0].iov_base = &pi;
 461                 iovecs[0].iov_len = sizeof(pi);
 462                 for (j = 1; j <= mbuf->nb_segs; j++) {
 463                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 464                         iovecs[j].iov_base =
 465                                 rte_pktmbuf_mtod(seg, void *);
 466                         seg = seg->next;
 467                 }
 468                 if (mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 469                     (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 470                     (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM) {
 471                         /* Support only packets with all data in the same seg */
 472                         if (mbuf->nb_segs > 1)
 473                                 break;
 474                         /* To change checksums, work on a copy of data. */
 475                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 476                                    rte_pktmbuf_data_len(mbuf));
 477                         tap_tx_offload(m_copy, mbuf->ol_flags,
 478                                        mbuf->l2_len, mbuf->l3_len);
 479                         iovecs[1].iov_base = m_copy;
 480                 }
 481                 /* copy the tx frame data */
 482                 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
 483                 if (n <= 0)
 484                         break;
 485
 486                 num_tx++;
 487                 num_tx_bytes += mbuf->pkt_len;
 488                 rte_pktmbuf_free(mbuf);
 489         }
 490
 491         txq->stats.opackets += num_tx;
 492         txq->stats.errs += nb_pkts - num_tx;
 493         txq->stats.obytes += num_tx_bytes;
 494
 495         return num_tx;
 496 }
 497
 498 static const char *
 499 tap_ioctl_req2str(unsigned long request)
 500 {
 501         switch (request) {
 502         case SIOCSIFFLAGS:
 503                 return "SIOCSIFFLAGS";
 504         case SIOCGIFFLAGS:
 505                 return "SIOCGIFFLAGS";
 506         case SIOCGIFHWADDR:
 507                 return "SIOCGIFHWADDR";
 508         case SIOCSIFHWADDR:
 509                 return "SIOCSIFHWADDR";
 510         case SIOCSIFMTU:
 511                 return "SIOCSIFMTU";
 512         }
 513         return "UNKNOWN";
 514 }
 515
 516 static int
 517 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 518           struct ifreq *ifr, int set, enum ioctl_mode mode)
 519 {
 520         short req_flags = ifr->ifr_flags;
 521         int remote = pmd->remote_if_index &&
 522                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 523
 524         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 525                 return 0;
 526         /*
 527          * If there is a remote netdevice, apply ioctl on it, then apply it on
 528          * the tap netdevice.
 529          */
 530 apply:
 531         if (remote)
 532                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
 533         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 534                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 535         switch (request) {
 536         case SIOCSIFFLAGS:
 537                 /* fetch current flags to leave other flags untouched */
 538                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 539                         goto error;
 540                 if (set)
 541                         ifr->ifr_flags |= req_flags;
 542                 else
 543                         ifr->ifr_flags &= ~req_flags;
 544                 break;
 545         case SIOCGIFFLAGS:
 546         case SIOCGIFHWADDR:
 547         case SIOCSIFHWADDR:
 548         case SIOCSIFMTU:
 549                 break;
 550         default:
 551                 RTE_ASSERT(!"unsupported request type: must not happen");
 552         }
 553         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 554                 goto error;
 555         if (remote-- && mode == LOCAL_AND_REMOTE)
 556                 goto apply;
 557         return 0;
 558
 559 error:
 560         RTE_LOG(DEBUG, PMD, "%s: %s(%s) failed: %s(%d)\n", ifr->ifr_name,
 561                 __func__, tap_ioctl_req2str(request), strerror(errno), errno);
 562         return -errno;
 563 }
 564
 565 static int
 566 tap_link_set_down(struct rte_eth_dev *dev)
 567 {
 568         struct pmd_internals *pmd = dev->data->dev_private;
 569         struct ifreq ifr = { .ifr_flags = IFF_UP };
 570
 571         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 572         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 573 }
 574
 575 static int
 576 tap_link_set_up(struct rte_eth_dev *dev)
 577 {
 578         struct pmd_internals *pmd = dev->data->dev_private;
 579         struct ifreq ifr = { .ifr_flags = IFF_UP };
 580
 581         dev->data->dev_link.link_status = ETH_LINK_UP;
 582         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 583 }
 584
 585 static int
 586 tap_dev_start(struct rte_eth_dev *dev)
 587 {
 588         int err;
 589
 590         err = tap_intr_handle_set(dev, 1);
 591         if (err)
 592                 return err;
 593         return tap_link_set_up(dev);
 594 }
 595
 596 /* This function gets called when the current port gets stopped.
 597  */
 598 static void
 599 tap_dev_stop(struct rte_eth_dev *dev)
 600 {
 601         tap_intr_handle_set(dev, 0);
 602         tap_link_set_down(dev);
 603 }
 604
 605 static int
 606 tap_dev_configure(struct rte_eth_dev *dev __rte_unused)
 607 {
 608         return 0;
 609 }
 610
 611 static uint32_t
 612 tap_dev_speed_capa(void)
 613 {
 614         uint32_t speed = pmd_link.link_speed;
 615         uint32_t capa = 0;
 616
 617         if (speed >= ETH_SPEED_NUM_10M)
 618                 capa |= ETH_LINK_SPEED_10M;
 619         if (speed >= ETH_SPEED_NUM_100M)
 620                 capa |= ETH_LINK_SPEED_100M;
 621         if (speed >= ETH_SPEED_NUM_1G)
 622                 capa |= ETH_LINK_SPEED_1G;
 623         if (speed >= ETH_SPEED_NUM_5G)
 624                 capa |= ETH_LINK_SPEED_2_5G;
 625         if (speed >= ETH_SPEED_NUM_5G)
 626                 capa |= ETH_LINK_SPEED_5G;
 627         if (speed >= ETH_SPEED_NUM_10G)
 628                 capa |= ETH_LINK_SPEED_10G;
 629         if (speed >= ETH_SPEED_NUM_20G)
 630                 capa |= ETH_LINK_SPEED_20G;
 631         if (speed >= ETH_SPEED_NUM_25G)
 632                 capa |= ETH_LINK_SPEED_25G;
 633         if (speed >= ETH_SPEED_NUM_40G)
 634                 capa |= ETH_LINK_SPEED_40G;
 635         if (speed >= ETH_SPEED_NUM_50G)
 636                 capa |= ETH_LINK_SPEED_50G;
 637         if (speed >= ETH_SPEED_NUM_56G)
 638                 capa |= ETH_LINK_SPEED_56G;
 639         if (speed >= ETH_SPEED_NUM_100G)
 640                 capa |= ETH_LINK_SPEED_100G;
 641
 642         return capa;
 643 }
 644
 645 static void
 646 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 647 {
 648         struct pmd_internals *internals = dev->data->dev_private;
 649
 650         dev_info->if_index = internals->if_index;
 651         dev_info->max_mac_addrs = 1;
 652         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 653         dev_info->max_rx_queues = internals->nb_queues;
 654         dev_info->max_tx_queues = internals->nb_queues;
 655         dev_info->min_rx_bufsize = 0;
 656         dev_info->pci_dev = NULL;
 657         dev_info->speed_capa = tap_dev_speed_capa();
 658         dev_info->rx_offload_capa = (DEV_RX_OFFLOAD_IPV4_CKSUM |
 659                                      DEV_RX_OFFLOAD_UDP_CKSUM |
 660                                      DEV_RX_OFFLOAD_TCP_CKSUM);
 661         dev_info->tx_offload_capa =
 662                 (DEV_TX_OFFLOAD_IPV4_CKSUM |
 663                  DEV_TX_OFFLOAD_UDP_CKSUM |
 664                  DEV_TX_OFFLOAD_TCP_CKSUM);
 665 }
 666
 667 static void
 668 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 669 {
 670         unsigned int i, imax;
 671         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 672         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 673         unsigned long rx_nombuf = 0, ierrors = 0;
 674         const struct pmd_internals *pmd = dev->data->dev_private;
 675
 676         imax = (pmd->nb_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 677                 pmd->nb_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 678
 679         for (i = 0; i < imax; i++) {
 680                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 681                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 682                 rx_total += tap_stats->q_ipackets[i];
 683                 rx_bytes_total += tap_stats->q_ibytes[i];
 684                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
 685                 ierrors += pmd->rxq[i].stats.ierrors;
 686
 687                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 688                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 689                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 690                 tx_total += tap_stats->q_opackets[i];
 691                 tx_err_total += tap_stats->q_errors[i];
 692                 tx_bytes_total += tap_stats->q_obytes[i];
 693         }
 694
 695         tap_stats->ipackets = rx_total;
 696         tap_stats->ibytes = rx_bytes_total;
 697         tap_stats->ierrors = ierrors;
 698         tap_stats->rx_nombuf = rx_nombuf;
 699         tap_stats->opackets = tx_total;
 700         tap_stats->oerrors = tx_err_total;
 701         tap_stats->obytes = tx_bytes_total;
 702 }
 703
 704 static void
 705 tap_stats_reset(struct rte_eth_dev *dev)
 706 {
 707         int i;
 708         struct pmd_internals *pmd = dev->data->dev_private;
 709
 710         for (i = 0; i < pmd->nb_queues; i++) {
 711                 pmd->rxq[i].stats.ipackets = 0;
 712                 pmd->rxq[i].stats.ibytes = 0;
 713                 pmd->rxq[i].stats.ierrors = 0;
 714                 pmd->rxq[i].stats.rx_nombuf = 0;
 715
 716                 pmd->txq[i].stats.opackets = 0;
 717                 pmd->txq[i].stats.errs = 0;
 718                 pmd->txq[i].stats.obytes = 0;
 719         }
 720 }
 721
 722 static void
 723 tap_dev_close(struct rte_eth_dev *dev)
 724 {
 725         int i;
 726         struct pmd_internals *internals = dev->data->dev_private;
 727
 728         tap_link_set_down(dev);
 729         tap_flow_flush(dev, NULL);
 730         tap_flow_implicit_flush(internals, NULL);
 731
 732         for (i = 0; i < internals->nb_queues; i++) {
 733                 if (internals->rxq[i].fd != -1)
 734                         close(internals->rxq[i].fd);
 735                 internals->rxq[i].fd = -1;
 736                 internals->txq[i].fd = -1;
 737         }
 738 }
 739
 740 static void
 741 tap_rx_queue_release(void *queue)
 742 {
 743         struct rx_queue *rxq = queue;
 744
 745         if (rxq && (rxq->fd > 0)) {
 746                 close(rxq->fd);
 747                 rxq->fd = -1;
 748                 rte_pktmbuf_free(rxq->pool);
 749                 rte_free(rxq->iovecs);
 750                 rxq->pool = NULL;
 751                 rxq->iovecs = NULL;
 752         }
 753 }
 754
 755 static void
 756 tap_tx_queue_release(void *queue)
 757 {
 758         struct tx_queue *txq = queue;
 759
 760         if (txq && (txq->fd > 0)) {
 761                 close(txq->fd);
 762                 txq->fd = -1;
 763         }
 764 }
 765
 766 static int
 767 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
 768 {
 769         struct rte_eth_link *dev_link = &dev->data->dev_link;
 770         struct pmd_internals *pmd = dev->data->dev_private;
 771         struct ifreq ifr = { .ifr_flags = 0 };
 772
 773         if (pmd->remote_if_index) {
 774                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
 775                 if (!(ifr.ifr_flags & IFF_UP) ||
 776                     !(ifr.ifr_flags & IFF_RUNNING)) {
 777                         dev_link->link_status = ETH_LINK_DOWN;
 778                         return 0;
 779                 }
 780         }
 781         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
 782         dev_link->link_status =
 783                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
 784                  ETH_LINK_UP :
 785                  ETH_LINK_DOWN);
 786         return 0;
 787 }
 788
 789 static void
 790 tap_promisc_enable(struct rte_eth_dev *dev)
 791 {
 792         struct pmd_internals *pmd = dev->data->dev_private;
 793         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 794
 795         dev->data->promiscuous = 1;
 796         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 797         if (pmd->remote_if_index)
 798                 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
 799 }
 800
 801 static void
 802 tap_promisc_disable(struct rte_eth_dev *dev)
 803 {
 804         struct pmd_internals *pmd = dev->data->dev_private;
 805         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 806
 807         dev->data->promiscuous = 0;
 808         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 809         if (pmd->remote_if_index)
 810                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
 811 }
 812
 813 static void
 814 tap_allmulti_enable(struct rte_eth_dev *dev)
 815 {
 816         struct pmd_internals *pmd = dev->data->dev_private;
 817         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 818
 819         dev->data->all_multicast = 1;
 820         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 821         if (pmd->remote_if_index)
 822                 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
 823 }
 824
 825 static void
 826 tap_allmulti_disable(struct rte_eth_dev *dev)
 827 {
 828         struct pmd_internals *pmd = dev->data->dev_private;
 829         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 830
 831         dev->data->all_multicast = 0;
 832         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 833         if (pmd->remote_if_index)
 834                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
 835 }
 836
 837 static void
 838 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 839 {
 840         struct pmd_internals *pmd = dev->data->dev_private;
 841         enum ioctl_mode mode = LOCAL_ONLY;
 842         struct ifreq ifr;
 843
 844         if (is_zero_ether_addr(mac_addr)) {
 845                 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
 846                         dev->data->name);
 847                 return;
 848         }
 849         /* Check the actual current MAC address on the tap netdevice */
 850         if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
 851                 return;
 852         if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 853                                mac_addr))
 854                 return;
 855         /* Check the current MAC address on the remote */
 856         if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0)
 857                 return;
 858         if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 859                                mac_addr))
 860                 mode = LOCAL_AND_REMOTE;
 861         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
 862         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
 863         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode) < 0)
 864                 return;
 865         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
 866         if (pmd->remote_if_index) {
 867                 /* Replace MAC redirection rule after a MAC change */
 868                 if (tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC) < 0) {
 869                         RTE_LOG(ERR, PMD,
 870                                 "%s: Couldn't delete MAC redirection rule\n",
 871                                 dev->data->name);
 872                         return;
 873                 }
 874                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0)
 875                         RTE_LOG(ERR, PMD,
 876                                 "%s: Couldn't add MAC redirection rule\n",
 877                                 dev->data->name);
 878         }
 879 }
 880
 881 static int
 882 tap_setup_queue(struct rte_eth_dev *dev,
 883                 struct pmd_internals *internals,
 884                 uint16_t qid)
 885 {
 886         struct pmd_internals *pmd = dev->data->dev_private;
 887         struct rx_queue *rx = &internals->rxq[qid];
 888         struct tx_queue *tx = &internals->txq[qid];
 889         int fd = rx->fd == -1 ? tx->fd : rx->fd;
 890
 891         if (fd == -1) {
 892                 RTE_LOG(INFO, PMD, "Add queue to TAP %s for qid %d\n",
 893                         pmd->name, qid);
 894                 fd = tun_alloc(pmd);
 895                 if (fd < 0) {
 896                         RTE_LOG(ERR, PMD, "%s: tun_alloc() failed.\n",
 897                                 pmd->name);
 898                         return -1;
 899                 }
 900         }
 901
 902         rx->fd = fd;
 903         tx->fd = fd;
 904         tx->mtu = &dev->data->mtu;
 905         rx->rxmode = &dev->data->dev_conf.rxmode;
 906
 907         return fd;
 908 }
 909
 910 static int
 911 tap_rx_queue_setup(struct rte_eth_dev *dev,
 912                    uint16_t rx_queue_id,
 913                    uint16_t nb_rx_desc,
 914                    unsigned int socket_id,
 915                    const struct rte_eth_rxconf *rx_conf __rte_unused,
 916                    struct rte_mempool *mp)
 917 {
 918         struct pmd_internals *internals = dev->data->dev_private;
 919         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
 920         struct rte_mbuf **tmp = &rxq->pool;
 921         long iov_max = sysconf(_SC_IOV_MAX);
 922         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
 923         struct iovec (*iovecs)[nb_desc + 1];
 924         int data_off = RTE_PKTMBUF_HEADROOM;
 925         int ret = 0;
 926         int fd;
 927         int i;
 928
 929         if ((rx_queue_id >= internals->nb_queues) || !mp) {
 930                 RTE_LOG(WARNING, PMD,
 931                         "nb_queues %d too small or mempool NULL\n",
 932                         internals->nb_queues);
 933                 return -1;
 934         }
 935
 936         rxq->mp = mp;
 937         rxq->trigger_seen = 1; /* force initial burst */
 938         rxq->in_port = dev->data->port_id;
 939         rxq->nb_rx_desc = nb_desc;
 940         iovecs = rte_zmalloc_socket(dev->data->name, sizeof(*iovecs), 0,
 941                                     socket_id);
 942         if (!iovecs) {
 943                 RTE_LOG(WARNING, PMD,
 944                         "%s: Couldn't allocate %d RX descriptors\n",
 945                         dev->data->name, nb_desc);
 946                 return -ENOMEM;
 947         }
 948         rxq->iovecs = iovecs;
 949
 950         dev->data->rx_queues[rx_queue_id] = rxq;
 951         fd = tap_setup_queue(dev, internals, rx_queue_id);
 952         if (fd == -1) {
 953                 ret = fd;
 954                 goto error;
 955         }
 956
 957         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
 958         (*rxq->iovecs)[0].iov_base = &rxq->pi;
 959
 960         for (i = 1; i <= nb_desc; i++) {
 961                 *tmp = rte_pktmbuf_alloc(rxq->mp);
 962                 if (!*tmp) {
 963                         RTE_LOG(WARNING, PMD,
 964                                 "%s: couldn't allocate memory for queue %d\n",
 965                                 dev->data->name, rx_queue_id);
 966                         ret = -ENOMEM;
 967                         goto error;
 968                 }
 969                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
 970                 (*rxq->iovecs)[i].iov_base =
 971                         (char *)(*tmp)->buf_addr + data_off;
 972                 data_off = 0;
 973                 tmp = &(*tmp)->next;
 974         }
 975
 976         RTE_LOG(DEBUG, PMD, "  RX TAP device name %s, qid %d on fd %d\n",
 977                 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
 978
 979         return 0;
 980
 981 error:
 982         rte_pktmbuf_free(rxq->pool);
 983         rxq->pool = NULL;
 984         rte_free(rxq->iovecs);
 985         rxq->iovecs = NULL;
 986         return ret;
 987 }
 988
 989 static int
 990 tap_tx_queue_setup(struct rte_eth_dev *dev,
 991                    uint16_t tx_queue_id,
 992                    uint16_t nb_tx_desc __rte_unused,
 993                    unsigned int socket_id __rte_unused,
 994                    const struct rte_eth_txconf *tx_conf __rte_unused)
 995 {
 996         struct pmd_internals *internals = dev->data->dev_private;
 997         int ret;
 998
 999         if (tx_queue_id >= internals->nb_queues)
1000                 return -1;
1001
1002         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1003         ret = tap_setup_queue(dev, internals, tx_queue_id);
1004         if (ret == -1)
1005                 return -1;
1006
1007         RTE_LOG(DEBUG, PMD, "  TX TAP device name %s, qid %d on fd %d\n",
1008                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd);
1009
1010         return 0;
1011 }
1012
1013 static int
1014 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1015 {
1016         struct pmd_internals *pmd = dev->data->dev_private;
1017         struct ifreq ifr = { .ifr_mtu = mtu };
1018         int err = 0;
1019
1020         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1021         if (!err)
1022                 dev->data->mtu = mtu;
1023
1024         return err;
1025 }
1026
1027 static int
1028 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1029                      struct ether_addr *mc_addr_set __rte_unused,
1030                      uint32_t nb_mc_addr __rte_unused)
1031 {
1032         /*
1033          * Nothing to do actually: the tap has no filtering whatsoever, every
1034          * packet is received.
1035          */
1036         return 0;
1037 }
1038
1039 static int
1040 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1041 {
1042         struct rte_eth_dev *dev = arg;
1043         struct pmd_internals *pmd = dev->data->dev_private;
1044         struct ifinfomsg *info = NLMSG_DATA(nh);
1045
1046         if (nh->nlmsg_type != RTM_NEWLINK ||
1047             (info->ifi_index != pmd->if_index &&
1048              info->ifi_index != pmd->remote_if_index))
1049                 return 0;
1050         return tap_link_update(dev, 0);
1051 }
1052
1053 static void
1054 tap_dev_intr_handler(void *cb_arg)
1055 {
1056         struct rte_eth_dev *dev = cb_arg;
1057         struct pmd_internals *pmd = dev->data->dev_private;
1058
1059         nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1060 }
1061
1062 static int
1063 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1064 {
1065         struct pmd_internals *pmd = dev->data->dev_private;
1066
1067         /* In any case, disable interrupt if the conf is no longer there. */
1068         if (!dev->data->dev_conf.intr_conf.lsc) {
1069                 if (pmd->intr_handle.fd != -1)
1070                         nl_final(pmd->intr_handle.fd);
1071                 rte_intr_callback_unregister(
1072                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1073                 return 0;
1074         }
1075         if (set) {
1076                 pmd->intr_handle.fd = nl_init(RTMGRP_LINK);
1077                 if (unlikely(pmd->intr_handle.fd == -1))
1078                         return -EBADF;
1079                 return rte_intr_callback_register(
1080                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1081         }
1082         nl_final(pmd->intr_handle.fd);
1083         return rte_intr_callback_unregister(&pmd->intr_handle,
1084                                             tap_dev_intr_handler, dev);
1085 }
1086
1087 static const uint32_t*
1088 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1089 {
1090         static const uint32_t ptypes[] = {
1091                 RTE_PTYPE_INNER_L2_ETHER,
1092                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1093                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1094                 RTE_PTYPE_INNER_L3_IPV4,
1095                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1096                 RTE_PTYPE_INNER_L3_IPV6,
1097                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1098                 RTE_PTYPE_INNER_L4_FRAG,
1099                 RTE_PTYPE_INNER_L4_UDP,
1100                 RTE_PTYPE_INNER_L4_TCP,
1101                 RTE_PTYPE_INNER_L4_SCTP,
1102                 RTE_PTYPE_L2_ETHER,
1103                 RTE_PTYPE_L2_ETHER_VLAN,
1104                 RTE_PTYPE_L2_ETHER_QINQ,
1105                 RTE_PTYPE_L3_IPV4,
1106                 RTE_PTYPE_L3_IPV4_EXT,
1107                 RTE_PTYPE_L3_IPV6_EXT,
1108                 RTE_PTYPE_L3_IPV6,
1109                 RTE_PTYPE_L4_FRAG,
1110                 RTE_PTYPE_L4_UDP,
1111                 RTE_PTYPE_L4_TCP,
1112                 RTE_PTYPE_L4_SCTP,
1113         };
1114
1115         return ptypes;
1116 }
1117
1118 static int
1119 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1120                   struct rte_eth_fc_conf *fc_conf)
1121 {
1122         fc_conf->mode = RTE_FC_NONE;
1123         return 0;
1124 }
1125
1126 static int
1127 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1128                   struct rte_eth_fc_conf *fc_conf)
1129 {
1130         if (fc_conf->mode != RTE_FC_NONE)
1131                 return -ENOTSUP;
1132         return 0;
1133 }
1134
1135 static const struct eth_dev_ops ops = {
1136         .dev_start              = tap_dev_start,
1137         .dev_stop               = tap_dev_stop,
1138         .dev_close              = tap_dev_close,
1139         .dev_configure          = tap_dev_configure,
1140         .dev_infos_get          = tap_dev_info,
1141         .rx_queue_setup         = tap_rx_queue_setup,
1142         .tx_queue_setup         = tap_tx_queue_setup,
1143         .rx_queue_release       = tap_rx_queue_release,
1144         .tx_queue_release       = tap_tx_queue_release,
1145         .flow_ctrl_get          = tap_flow_ctrl_get,
1146         .flow_ctrl_set          = tap_flow_ctrl_set,
1147         .link_update            = tap_link_update,
1148         .dev_set_link_up        = tap_link_set_up,
1149         .dev_set_link_down      = tap_link_set_down,
1150         .promiscuous_enable     = tap_promisc_enable,
1151         .promiscuous_disable    = tap_promisc_disable,
1152         .allmulticast_enable    = tap_allmulti_enable,
1153         .allmulticast_disable   = tap_allmulti_disable,
1154         .mac_addr_set           = tap_mac_set,
1155         .mtu_set                = tap_mtu_set,
1156         .set_mc_addr_list       = tap_set_mc_addr_list,
1157         .stats_get              = tap_stats_get,
1158         .stats_reset            = tap_stats_reset,
1159         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1160         .filter_ctrl            = tap_dev_filter_ctrl,
1161 };
1162
1163 static int
1164 tap_kernel_support(struct pmd_internals *pmd)
1165 {
1166         struct utsname utsname;
1167         int ver[3];
1168
1169         if (uname(&utsname) == -1 ||
1170             sscanf(utsname.release, "%d.%d.%d",
1171                    &ver[0], &ver[1], &ver[2]) != 3)
1172                 return 0;
1173         if (KERNEL_VERSION(ver[0], ver[1], ver[2]) >= FLOWER_KERNEL_VERSION)
1174                 pmd->flower_support = 1;
1175         if (KERNEL_VERSION(ver[0], ver[1], ver[2]) >=
1176             FLOWER_VLAN_KERNEL_VERSION)
1177                 pmd->flower_vlan_support = 1;
1178         return 1;
1179 }
1180
1181 static int
1182 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1183                    char *remote_iface, int fixed_mac_type)
1184 {
1185         int numa_node = rte_socket_id();
1186         struct rte_eth_dev *dev;
1187         struct pmd_internals *pmd;
1188         struct rte_eth_dev_data *data;
1189         struct ifreq ifr;
1190         int i;
1191
1192         RTE_LOG(DEBUG, PMD, "  TAP device on numa %u\n", rte_socket_id());
1193
1194         data = rte_zmalloc_socket(tap_name, sizeof(*data), 0, numa_node);
1195         if (!data) {
1196                 RTE_LOG(ERR, PMD, "TAP Failed to allocate data\n");
1197                 goto error_exit;
1198         }
1199
1200         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1201         if (!dev) {
1202                 RTE_LOG(ERR, PMD, "TAP Unable to allocate device struct\n");
1203                 goto error_exit;
1204         }
1205
1206         pmd = dev->data->dev_private;
1207         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1208         pmd->nb_queues = RTE_PMD_TAP_MAX_QUEUES;
1209
1210         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1211         if (pmd->ioctl_sock == -1) {
1212                 RTE_LOG(ERR, PMD,
1213                         "TAP Unable to get a socket for management: %s\n",
1214                         strerror(errno));
1215                 goto error_exit;
1216         }
1217
1218         /* Setup some default values */
1219         rte_memcpy(data, dev->data, sizeof(*data));
1220         data->dev_private = pmd;
1221         data->dev_flags = RTE_ETH_DEV_DETACHABLE | RTE_ETH_DEV_INTR_LSC;
1222         data->numa_node = numa_node;
1223
1224         data->dev_link = pmd_link;
1225         data->mac_addrs = &pmd->eth_addr;
1226         data->nb_rx_queues = pmd->nb_queues;
1227         data->nb_tx_queues = pmd->nb_queues;
1228
1229         dev->data = data;
1230         dev->dev_ops = &ops;
1231         dev->rx_pkt_burst = pmd_rx_burst;
1232         dev->tx_pkt_burst = pmd_tx_burst;
1233
1234         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1235         pmd->intr_handle.fd = -1;
1236
1237         /* Presetup the fds to -1 as being not valid */
1238         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1239                 pmd->rxq[i].fd = -1;
1240                 pmd->txq[i].fd = -1;
1241         }
1242
1243         if (fixed_mac_type) {
1244                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1245                 static int iface_idx;
1246                 char mac[ETHER_ADDR_LEN] = "\0dtap";
1247
1248                 mac[ETHER_ADDR_LEN - 1] = iface_idx++;
1249                 rte_memcpy(&pmd->eth_addr, mac, ETHER_ADDR_LEN);
1250         } else {
1251                 eth_random_addr((uint8_t *)&pmd->eth_addr);
1252         }
1253
1254         /* Immediately create the netdevice (this will create the 1st queue). */
1255         if (tap_setup_queue(dev, pmd, 0) == -1)
1256                 goto error_exit;
1257
1258         ifr.ifr_mtu = dev->data->mtu;
1259         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1260                 goto error_exit;
1261
1262         memset(&ifr, 0, sizeof(struct ifreq));
1263         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1264         rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr, ETHER_ADDR_LEN);
1265         if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1266                 goto error_exit;
1267
1268         tap_kernel_support(pmd);
1269         if (!pmd->flower_support) {
1270                 if (remote_iface[0]) {
1271                         RTE_LOG(ERR, PMD,
1272                                 "%s: kernel does not support TC rules, required for remote feature.",
1273                                 pmd->name);
1274                         goto error_exit;
1275                 } else {
1276                         RTE_LOG(INFO, PMD,
1277                                 "%s: kernel too old for Flow API support.\n",
1278                                 pmd->name);
1279                         return 0;
1280                 }
1281         }
1282
1283         /*
1284          * Set up everything related to rte_flow:
1285          * - netlink socket
1286          * - tap / remote if_index
1287          * - mandatory QDISCs
1288          * - rte_flow actual/implicit lists
1289          * - implicit rules
1290          */
1291         pmd->nlsk_fd = nl_init(0);
1292         if (pmd->nlsk_fd == -1) {
1293                 RTE_LOG(WARNING, PMD, "%s: failed to create netlink socket.",
1294                         pmd->name);
1295                 goto disable_rte_flow;
1296         }
1297         pmd->if_index = if_nametoindex(pmd->name);
1298         if (!pmd->if_index) {
1299                 RTE_LOG(ERR, PMD, "%s: failed to get if_index.", pmd->name);
1300                 goto disable_rte_flow;
1301         }
1302         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1303                 RTE_LOG(ERR, PMD, "%s: failed to create multiq qdisc.",
1304                         pmd->name);
1305                 goto disable_rte_flow;
1306         }
1307         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1308                 RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.",
1309                         pmd->name);
1310                 goto disable_rte_flow;
1311         }
1312         LIST_INIT(&pmd->flows);
1313
1314         if (strlen(remote_iface)) {
1315                 pmd->remote_if_index = if_nametoindex(remote_iface);
1316                 if (!pmd->remote_if_index) {
1317                         RTE_LOG(ERR, PMD, "%s: failed to get %s if_index.",
1318                                 pmd->name, remote_iface);
1319                         goto error_remote;
1320                 }
1321                 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1322                          "%s", remote_iface);
1323                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1324                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.",
1325                                 pmd->name, pmd->remote_iface);
1326                         goto error_remote;
1327                 }
1328                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1329                            ETHER_ADDR_LEN);
1330                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1331                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1332                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.",
1333                                 pmd->name, remote_iface);
1334                         goto error_remote;
1335                 }
1336
1337                 /*
1338                  * Flush usually returns negative value because it tries to
1339                  * delete every QDISC (and on a running device, one QDISC at
1340                  * least is needed). Ignore negative return value.
1341                  */
1342                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1343                 if (qdisc_create_ingress(pmd->nlsk_fd,
1344                                          pmd->remote_if_index) < 0) {
1345                         RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.",
1346                                 pmd->remote_iface);
1347                         goto error_remote;
1348                 }
1349                 LIST_INIT(&pmd->implicit_flows);
1350                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1351                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1352                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1353                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1354                         RTE_LOG(ERR, PMD,
1355                                 "%s: failed to create implicit rules.",
1356                                 pmd->name);
1357                         goto error_remote;
1358                 }
1359         }
1360
1361         return 0;
1362
1363 disable_rte_flow:
1364         RTE_LOG(ERR, PMD, " Disabling rte flow support: %s(%d)\n",
1365                 strerror(errno), errno);
1366         if (strlen(remote_iface)) {
1367                 RTE_LOG(ERR, PMD, "Remote feature requires flow support.\n");
1368                 goto error_exit;
1369         }
1370         pmd->flower_support = 0;
1371         return 0;
1372
1373 error_remote:
1374         RTE_LOG(ERR, PMD, " Can't set up remote feature: %s(%d)\n",
1375                 strerror(errno), errno);
1376         tap_flow_implicit_flush(pmd, NULL);
1377
1378 error_exit:
1379         RTE_LOG(ERR, PMD, "TAP Unable to initialize %s\n",
1380                 rte_vdev_device_name(vdev));
1381
1382         rte_free(data);
1383         return -EINVAL;
1384 }
1385
1386 static int
1387 set_interface_name(const char *key __rte_unused,
1388                    const char *value,
1389                    void *extra_args)
1390 {
1391         char *name = (char *)extra_args;
1392
1393         if (value)
1394                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s", value);
1395         else
1396                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1397                          DEFAULT_TAP_NAME, (tap_unit - 1));
1398
1399         return 0;
1400 }
1401
1402 static int
1403 set_interface_speed(const char *key __rte_unused,
1404                     const char *value,
1405                     void *extra_args)
1406 {
1407         *(int *)extra_args = (value) ? atoi(value) : ETH_SPEED_NUM_10G;
1408
1409         return 0;
1410 }
1411
1412 static int
1413 set_remote_iface(const char *key __rte_unused,
1414                  const char *value,
1415                  void *extra_args)
1416 {
1417         char *name = (char *)extra_args;
1418
1419         if (value)
1420                 snprintf(name, RTE_ETH_NAME_MAX_LEN, "%s", value);
1421
1422         return 0;
1423 }
1424
1425 static int
1426 set_mac_type(const char *key __rte_unused,
1427              const char *value,
1428              void *extra_args)
1429 {
1430         if (value &&
1431             !strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED)))
1432                 *(int *)extra_args = 1;
1433         return 0;
1434 }
1435
1436 /* Open a TAP interface device.
1437  */
1438 static int
1439 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1440 {
1441         const char *name, *params;
1442         int ret;
1443         struct rte_kvargs *kvlist = NULL;
1444         int speed;
1445         char tap_name[RTE_ETH_NAME_MAX_LEN];
1446         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1447         int fixed_mac_type = 0;
1448
1449         name = rte_vdev_device_name(dev);
1450         params = rte_vdev_device_args(dev);
1451
1452         speed = ETH_SPEED_NUM_10G;
1453         snprintf(tap_name, sizeof(tap_name), "%s%d",
1454                  DEFAULT_TAP_NAME, tap_unit++);
1455         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1456
1457         if (params && (params[0] != '\0')) {
1458                 RTE_LOG(DEBUG, PMD, "parameters (%s)\n", params);
1459
1460                 kvlist = rte_kvargs_parse(params, valid_arguments);
1461                 if (kvlist) {
1462                         if (rte_kvargs_count(kvlist, ETH_TAP_SPEED_ARG) == 1) {
1463                                 ret = rte_kvargs_process(kvlist,
1464                                                          ETH_TAP_SPEED_ARG,
1465                                                          &set_interface_speed,
1466                                                          &speed);
1467                                 if (ret == -1)
1468                                         goto leave;
1469                         }
1470
1471                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1472                                 ret = rte_kvargs_process(kvlist,
1473                                                          ETH_TAP_IFACE_ARG,
1474                                                          &set_interface_name,
1475                                                          tap_name);
1476                                 if (ret == -1)
1477                                         goto leave;
1478                         }
1479
1480                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1481                                 ret = rte_kvargs_process(kvlist,
1482                                                          ETH_TAP_REMOTE_ARG,
1483                                                          &set_remote_iface,
1484                                                          remote_iface);
1485                                 if (ret == -1)
1486                                         goto leave;
1487                         }
1488
1489                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1490                                 ret = rte_kvargs_process(kvlist,
1491                                                          ETH_TAP_MAC_ARG,
1492                                                          &set_mac_type,
1493                                                          &fixed_mac_type);
1494                                 if (ret == -1)
1495                                         goto leave;
1496                         }
1497                 }
1498         }
1499         pmd_link.link_speed = speed;
1500
1501         RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1502                 name, tap_name);
1503
1504         ret = eth_dev_tap_create(dev, tap_name, remote_iface, fixed_mac_type);
1505
1506 leave:
1507         if (ret == -1) {
1508                 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1509                         name, tap_name);
1510                 tap_unit--;             /* Restore the unit number */
1511         }
1512         rte_kvargs_free(kvlist);
1513
1514         return ret;
1515 }
1516
1517 /* detach a TAP device.
1518  */
1519 static int
1520 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1521 {
1522         struct rte_eth_dev *eth_dev = NULL;
1523         struct pmd_internals *internals;
1524         int i;
1525
1526         RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1527                 rte_socket_id());
1528
1529         /* find the ethdev entry */
1530         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1531         if (!eth_dev)
1532                 return 0;
1533
1534         internals = eth_dev->data->dev_private;
1535         if (internals->flower_support && internals->nlsk_fd) {
1536                 tap_flow_flush(eth_dev, NULL);
1537                 tap_flow_implicit_flush(internals, NULL);
1538                 nl_final(internals->nlsk_fd);
1539         }
1540         for (i = 0; i < internals->nb_queues; i++)
1541                 if (internals->rxq[i].fd != -1)
1542                         close(internals->rxq[i].fd);
1543
1544         close(internals->ioctl_sock);
1545         rte_free(eth_dev->data->dev_private);
1546         rte_free(eth_dev->data);
1547
1548         rte_eth_dev_release_port(eth_dev);
1549
1550         return 0;
1551 }
1552
1553 static struct rte_vdev_driver pmd_tap_drv = {
1554         .probe = rte_pmd_tap_probe,
1555         .remove = rte_pmd_tap_remove,
1556 };
1557 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1558 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1559 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1560                               ETH_TAP_IFACE_ARG "=<string> "
1561                               ETH_TAP_SPEED_ARG "=<int> "
1562                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_FIXED " "
1563                               ETH_TAP_REMOTE_ARG "=<string>");