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 <sys/types.h>
  21 #include <sys/stat.h>
  22 #include <sys/socket.h>
  23 #include <sys/ioctl.h>
  24 #include <sys/utsname.h>
  25 #include <sys/mman.h>
  26 #include <errno.h>
  27 #include <signal.h>
  28 #include <stdbool.h>
  29 #include <stdint.h>
  30 #include <sys/uio.h>
  31 #include <unistd.h>
  32 #include <arpa/inet.h>
  33 #include <net/if.h>
  34 #include <linux/if_tun.h>
  35 #include <linux/if_ether.h>
  36 #include <fcntl.h>
  37
  38 #include <rte_eth_tap.h>
  39 #include <tap_flow.h>
  40 #include <tap_netlink.h>
  41 #include <tap_tcmsgs.h>
  42
  43 /* Linux based path to the TUN device */
  44 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  45 #define DEFAULT_TAP_NAME        "dtap"
  46 #define DEFAULT_TUN_NAME        "dtun"
  47
  48 #define ETH_TAP_IFACE_ARG       "iface"
  49 #define ETH_TAP_REMOTE_ARG      "remote"
  50 #define ETH_TAP_MAC_ARG         "mac"
  51 #define ETH_TAP_MAC_FIXED       "fixed"
  52
  53 #define ETH_TAP_USR_MAC_FMT     "xx:xx:xx:xx:xx:xx"
  54 #define ETH_TAP_CMP_MAC_FMT     "0123456789ABCDEFabcdef"
  55 #define ETH_TAP_MAC_ARG_FMT     ETH_TAP_MAC_FIXED "|" ETH_TAP_USR_MAC_FMT
  56
  57 static struct rte_vdev_driver pmd_tap_drv;
  58 static struct rte_vdev_driver pmd_tun_drv;
  59
  60 static const char *valid_arguments[] = {
  61         ETH_TAP_IFACE_ARG,
  62         ETH_TAP_REMOTE_ARG,
  63         ETH_TAP_MAC_ARG,
  64         NULL
  65 };
  66
  67 static int tap_unit;
  68 static int tun_unit;
  69
  70 static int tap_type;
  71 static char tuntap_name[8];
  72
  73 static volatile uint32_t tap_trigger;   /* Rx trigger */
  74
  75 static struct rte_eth_link pmd_link = {
  76         .link_speed = ETH_SPEED_NUM_10G,
  77         .link_duplex = ETH_LINK_FULL_DUPLEX,
  78         .link_status = ETH_LINK_DOWN,
  79         .link_autoneg = ETH_LINK_AUTONEG
  80 };
  81
  82 static void
  83 tap_trigger_cb(int sig __rte_unused)
  84 {
  85         /* Valid trigger values are nonzero */
  86         tap_trigger = (tap_trigger + 1) | 0x80000000;
  87 }
  88
  89 /* Specifies on what netdevices the ioctl should be applied */
  90 enum ioctl_mode {
  91         LOCAL_AND_REMOTE,
  92         LOCAL_ONLY,
  93         REMOTE_ONLY,
  94 };
  95
  96 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
  97
  98 /* Tun/Tap allocation routine
  99  *
 100  * name is the number of the interface to use, unless NULL to take the host
 101  * supplied name.
 102  */
 103 static int
 104 tun_alloc(struct pmd_internals *pmd)
 105 {
 106         struct ifreq ifr;
 107 #ifdef IFF_MULTI_QUEUE
 108         unsigned int features;
 109 #endif
 110         int fd;
 111
 112         memset(&ifr, 0, sizeof(struct ifreq));
 113
 114         /*
 115          * Do not set IFF_NO_PI as packet information header will be needed
 116          * to check if a received packet has been truncated.
 117          */
 118         ifr.ifr_flags = (tap_type) ? IFF_TAP : IFF_TUN | IFF_POINTOPOINT;
 119         snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
 120
 121         RTE_LOG(DEBUG, PMD, "ifr_name '%s'\n", ifr.ifr_name);
 122
 123         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 124         if (fd < 0) {
 125                 RTE_LOG(ERR, PMD, "Unable to create %s interface\n",
 126                                 tuntap_name);
 127                 goto error;
 128         }
 129
 130 #ifdef IFF_MULTI_QUEUE
 131         /* Grab the TUN features to verify we can work multi-queue */
 132         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 133                 RTE_LOG(ERR, PMD, "%s unable to get TUN/TAP features\n",
 134                                 tuntap_name);
 135                 goto error;
 136         }
 137         RTE_LOG(DEBUG, PMD, "%s Features %08x\n", tuntap_name, features);
 138
 139         if (features & IFF_MULTI_QUEUE) {
 140                 RTE_LOG(DEBUG, PMD, "  Multi-queue support for %d queues\n",
 141                         RTE_PMD_TAP_MAX_QUEUES);
 142                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 143         } else
 144 #endif
 145         {
 146                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 147                 RTE_LOG(DEBUG, PMD, "  Single queue only support\n");
 148         }
 149
 150         /* Set the TUN/TAP configuration and set the name if needed */
 151         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 152                 RTE_LOG(WARNING, PMD,
 153                         "Unable to set TUNSETIFF for %s\n",
 154                         ifr.ifr_name);
 155                 perror("TUNSETIFF");
 156                 goto error;
 157         }
 158
 159         /* Always set the file descriptor to non-blocking */
 160         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 161                 RTE_LOG(WARNING, PMD,
 162                         "Unable to set %s to nonblocking\n",
 163                         ifr.ifr_name);
 164                 perror("F_SETFL, NONBLOCK");
 165                 goto error;
 166         }
 167
 168         /* Set up trigger to optimize empty Rx bursts */
 169         errno = 0;
 170         do {
 171                 struct sigaction sa;
 172                 int flags = fcntl(fd, F_GETFL);
 173
 174                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 175                         break;
 176                 if (sa.sa_handler != tap_trigger_cb) {
 177                         /*
 178                          * Make sure SIGIO is not already taken. This is done
 179                          * as late as possible to leave the application a
 180                          * chance to set up its own signal handler first.
 181                          */
 182                         if (sa.sa_handler != SIG_IGN &&
 183                             sa.sa_handler != SIG_DFL) {
 184                                 errno = EBUSY;
 185                                 break;
 186                         }
 187                         sa = (struct sigaction){
 188                                 .sa_flags = SA_RESTART,
 189                                 .sa_handler = tap_trigger_cb,
 190                         };
 191                         if (sigaction(SIGIO, &sa, NULL) == -1)
 192                                 break;
 193                 }
 194                 /* Enable SIGIO on file descriptor */
 195                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 196                 fcntl(fd, F_SETOWN, getpid());
 197         } while (0);
 198         if (errno) {
 199                 /* Disable trigger globally in case of error */
 200                 tap_trigger = 0;
 201                 RTE_LOG(WARNING, PMD, "Rx trigger disabled: %s\n",
 202                         strerror(errno));
 203         }
 204
 205         return fd;
 206
 207 error:
 208         if (fd > 0)
 209                 close(fd);
 210         return -1;
 211 }
 212
 213 static void
 214 tap_verify_csum(struct rte_mbuf *mbuf)
 215 {
 216         uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
 217         uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
 218         uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
 219         unsigned int l2_len = sizeof(struct ether_hdr);
 220         unsigned int l3_len;
 221         uint16_t cksum = 0;
 222         void *l3_hdr;
 223         void *l4_hdr;
 224
 225         if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
 226                 l2_len += 4;
 227         else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
 228                 l2_len += 8;
 229         /* Don't verify checksum for packets with discontinuous L2 header */
 230         if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
 231                      rte_pktmbuf_data_len(mbuf)))
 232                 return;
 233         l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
 234         if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 235                 struct ipv4_hdr *iph = l3_hdr;
 236
 237                 /* ihl contains the number of 4-byte words in the header */
 238                 l3_len = 4 * (iph->version_ihl & 0xf);
 239                 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
 240                         return;
 241
 242                 cksum = ~rte_raw_cksum(iph, l3_len);
 243                 mbuf->ol_flags |= cksum ?
 244                         PKT_RX_IP_CKSUM_BAD :
 245                         PKT_RX_IP_CKSUM_GOOD;
 246         } else if (l3 == RTE_PTYPE_L3_IPV6) {
 247                 l3_len = sizeof(struct ipv6_hdr);
 248         } else {
 249                 /* IPv6 extensions are not supported */
 250                 return;
 251         }
 252         if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
 253                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 254                 /* Don't verify checksum for multi-segment packets. */
 255                 if (mbuf->nb_segs > 1)
 256                         return;
 257                 if (l3 == RTE_PTYPE_L3_IPV4)
 258                         cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
 259                 else if (l3 == RTE_PTYPE_L3_IPV6)
 260                         cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
 261                 mbuf->ol_flags |= cksum ?
 262                         PKT_RX_L4_CKSUM_BAD :
 263                         PKT_RX_L4_CKSUM_GOOD;
 264         }
 265 }
 266
 267 static uint64_t
 268 tap_rx_offload_get_port_capa(void)
 269 {
 270         /*
 271          * In order to support legacy apps,
 272          * report capabilities also as port capabilities.
 273          */
 274         return DEV_RX_OFFLOAD_SCATTER |
 275                DEV_RX_OFFLOAD_IPV4_CKSUM |
 276                DEV_RX_OFFLOAD_UDP_CKSUM |
 277                DEV_RX_OFFLOAD_TCP_CKSUM |
 278                DEV_RX_OFFLOAD_CRC_STRIP;
 279 }
 280
 281 static uint64_t
 282 tap_rx_offload_get_queue_capa(void)
 283 {
 284         return DEV_RX_OFFLOAD_SCATTER |
 285                DEV_RX_OFFLOAD_IPV4_CKSUM |
 286                DEV_RX_OFFLOAD_UDP_CKSUM |
 287                DEV_RX_OFFLOAD_TCP_CKSUM |
 288                DEV_RX_OFFLOAD_CRC_STRIP;
 289 }
 290
 291 static bool
 292 tap_rxq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 293 {
 294         uint64_t port_offloads = dev->data->dev_conf.rxmode.offloads;
 295         uint64_t queue_supp_offloads = tap_rx_offload_get_queue_capa();
 296         uint64_t port_supp_offloads = tap_rx_offload_get_port_capa();
 297
 298         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 299             offloads)
 300                 return false;
 301         if ((port_offloads ^ offloads) & port_supp_offloads)
 302                 return false;
 303         return true;
 304 }
 305
 306 /* Callback to handle the rx burst of packets to the correct interface and
 307  * file descriptor(s) in a multi-queue setup.
 308  */
 309 static uint16_t
 310 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 311 {
 312         struct rx_queue *rxq = queue;
 313         uint16_t num_rx;
 314         unsigned long num_rx_bytes = 0;
 315         uint32_t trigger = tap_trigger;
 316
 317         if (trigger == rxq->trigger_seen)
 318                 return 0;
 319         if (trigger)
 320                 rxq->trigger_seen = trigger;
 321         rte_compiler_barrier();
 322         for (num_rx = 0; num_rx < nb_pkts; ) {
 323                 struct rte_mbuf *mbuf = rxq->pool;
 324                 struct rte_mbuf *seg = NULL;
 325                 struct rte_mbuf *new_tail = NULL;
 326                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 327                 int len;
 328
 329                 len = readv(rxq->fd, *rxq->iovecs,
 330                             1 +
 331                             (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
 332                              rxq->nb_rx_desc : 1));
 333                 if (len < (int)sizeof(struct tun_pi))
 334                         break;
 335
 336                 /* Packet couldn't fit in the provided mbuf */
 337                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 338                         rxq->stats.ierrors++;
 339                         continue;
 340                 }
 341
 342                 len -= sizeof(struct tun_pi);
 343
 344                 mbuf->pkt_len = len;
 345                 mbuf->port = rxq->in_port;
 346                 while (1) {
 347                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 348
 349                         if (unlikely(!buf)) {
 350                                 rxq->stats.rx_nombuf++;
 351                                 /* No new buf has been allocated: do nothing */
 352                                 if (!new_tail || !seg)
 353                                         goto end;
 354
 355                                 seg->next = NULL;
 356                                 rte_pktmbuf_free(mbuf);
 357
 358                                 goto end;
 359                         }
 360                         seg = seg ? seg->next : mbuf;
 361                         if (rxq->pool == mbuf)
 362                                 rxq->pool = buf;
 363                         if (new_tail)
 364                                 new_tail->next = buf;
 365                         new_tail = buf;
 366                         new_tail->next = seg->next;
 367
 368                         /* iovecs[0] is reserved for packet info (pi) */
 369                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 370                                 buf->buf_len - data_off;
 371                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 372                                 (char *)buf->buf_addr + data_off;
 373
 374                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 375                         seg->data_off = data_off;
 376
 377                         len -= seg->data_len;
 378                         if (len <= 0)
 379                                 break;
 380                         mbuf->nb_segs++;
 381                         /* First segment has headroom, not the others */
 382                         data_off = 0;
 383                 }
 384                 seg->next = NULL;
 385                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 386                                                       RTE_PTYPE_ALL_MASK);
 387                 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
 388                         tap_verify_csum(mbuf);
 389
 390                 /* account for the receive frame */
 391                 bufs[num_rx++] = mbuf;
 392                 num_rx_bytes += mbuf->pkt_len;
 393         }
 394 end:
 395         rxq->stats.ipackets += num_rx;
 396         rxq->stats.ibytes += num_rx_bytes;
 397
 398         return num_rx;
 399 }
 400
 401 static uint64_t
 402 tap_tx_offload_get_port_capa(void)
 403 {
 404         /*
 405          * In order to support legacy apps,
 406          * report capabilities also as port capabilities.
 407          */
 408         return DEV_TX_OFFLOAD_MULTI_SEGS |
 409                DEV_TX_OFFLOAD_IPV4_CKSUM |
 410                DEV_TX_OFFLOAD_UDP_CKSUM |
 411                DEV_TX_OFFLOAD_TCP_CKSUM;
 412 }
 413
 414 static uint64_t
 415 tap_tx_offload_get_queue_capa(void)
 416 {
 417         return DEV_TX_OFFLOAD_MULTI_SEGS |
 418                DEV_TX_OFFLOAD_IPV4_CKSUM |
 419                DEV_TX_OFFLOAD_UDP_CKSUM |
 420                DEV_TX_OFFLOAD_TCP_CKSUM;
 421 }
 422
 423 static bool
 424 tap_txq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 425 {
 426         uint64_t port_offloads = dev->data->dev_conf.txmode.offloads;
 427         uint64_t queue_supp_offloads = tap_tx_offload_get_queue_capa();
 428         uint64_t port_supp_offloads = tap_tx_offload_get_port_capa();
 429
 430         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 431             offloads)
 432                 return false;
 433         /* Verify we have no conflict with port offloads */
 434         if ((port_offloads ^ offloads) & port_supp_offloads)
 435                 return false;
 436         return true;
 437 }
 438
 439 static void
 440 tap_tx_offload(char *packet, uint64_t ol_flags, unsigned int l2_len,
 441                unsigned int l3_len)
 442 {
 443         void *l3_hdr = packet + l2_len;
 444
 445         if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
 446                 struct ipv4_hdr *iph = l3_hdr;
 447                 uint16_t cksum;
 448
 449                 iph->hdr_checksum = 0;
 450                 cksum = rte_raw_cksum(iph, l3_len);
 451                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 452         }
 453         if (ol_flags & PKT_TX_L4_MASK) {
 454                 uint16_t l4_len;
 455                 uint32_t cksum;
 456                 uint16_t *l4_cksum;
 457                 void *l4_hdr;
 458
 459                 l4_hdr = packet + l2_len + l3_len;
 460                 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
 461                         l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
 462                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 463                         l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
 464                 else
 465                         return;
 466                 *l4_cksum = 0;
 467                 if (ol_flags & PKT_TX_IPV4) {
 468                         struct ipv4_hdr *iph = l3_hdr;
 469
 470                         l4_len = rte_be_to_cpu_16(iph->total_length) - l3_len;
 471                         cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 472                 } else {
 473                         struct ipv6_hdr *ip6h = l3_hdr;
 474
 475                         /* payload_len does not include ext headers */
 476                         l4_len = rte_be_to_cpu_16(ip6h->payload_len) -
 477                                 l3_len + sizeof(struct ipv6_hdr);
 478                         cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 479                 }
 480                 cksum += rte_raw_cksum(l4_hdr, l4_len);
 481                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 482                 cksum = (~cksum) & 0xffff;
 483                 if (cksum == 0)
 484                         cksum = 0xffff;
 485                 *l4_cksum = cksum;
 486         }
 487 }
 488
 489 /* Callback to handle sending packets from the tap interface
 490  */
 491 static uint16_t
 492 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 493 {
 494         struct tx_queue *txq = queue;
 495         uint16_t num_tx = 0;
 496         unsigned long num_tx_bytes = 0;
 497         uint32_t max_size;
 498         int i;
 499
 500         if (unlikely(nb_pkts == 0))
 501                 return 0;
 502
 503         max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
 504         for (i = 0; i < nb_pkts; i++) {
 505                 struct rte_mbuf *mbuf = bufs[num_tx];
 506                 struct iovec iovecs[mbuf->nb_segs + 1];
 507                 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
 508                 struct rte_mbuf *seg = mbuf;
 509                 char m_copy[mbuf->data_len];
 510                 int n;
 511                 int j;
 512
 513                 /* stats.errs will be incremented */
 514                 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
 515                         break;
 516
 517                 /*
 518                  * TUN and TAP are created with IFF_NO_PI disabled.
 519                  * For TUN PMD this mandatory as fields are used by
 520                  * Kernel tun.c to determine whether its IP or non IP
 521                  * packets.
 522                  *
 523                  * The logic fetches the first byte of data from mbuf.
 524                  * compares whether its v4 or v6. If none matches default
 525                  * value 0x00 is taken for protocol field.
 526                  */
 527                 char *buff_data = rte_pktmbuf_mtod(seg, void *);
 528                 j = (*buff_data & 0xf0);
 529                 pi.proto = (j == 0x40) ? 0x0008 :
 530                                 (j == 0x60) ? 0xdd86 : 0x00;
 531
 532                 iovecs[0].iov_base = &pi;
 533                 iovecs[0].iov_len = sizeof(pi);
 534                 for (j = 1; j <= mbuf->nb_segs; j++) {
 535                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 536                         iovecs[j].iov_base =
 537                                 rte_pktmbuf_mtod(seg, void *);
 538                         seg = seg->next;
 539                 }
 540                 if (txq->csum &&
 541                     ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 542                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 543                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
 544                         /* Support only packets with all data in the same seg */
 545                         if (mbuf->nb_segs > 1)
 546                                 break;
 547                         /* To change checksums, work on a copy of data. */
 548                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 549                                    rte_pktmbuf_data_len(mbuf));
 550                         tap_tx_offload(m_copy, mbuf->ol_flags,
 551                                        mbuf->l2_len, mbuf->l3_len);
 552                         iovecs[1].iov_base = m_copy;
 553                 }
 554                 /* copy the tx frame data */
 555                 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
 556                 if (n <= 0)
 557                         break;
 558
 559                 num_tx++;
 560                 num_tx_bytes += mbuf->pkt_len;
 561                 rte_pktmbuf_free(mbuf);
 562         }
 563
 564         txq->stats.opackets += num_tx;
 565         txq->stats.errs += nb_pkts - num_tx;
 566         txq->stats.obytes += num_tx_bytes;
 567
 568         return num_tx;
 569 }
 570
 571 static const char *
 572 tap_ioctl_req2str(unsigned long request)
 573 {
 574         switch (request) {
 575         case SIOCSIFFLAGS:
 576                 return "SIOCSIFFLAGS";
 577         case SIOCGIFFLAGS:
 578                 return "SIOCGIFFLAGS";
 579         case SIOCGIFHWADDR:
 580                 return "SIOCGIFHWADDR";
 581         case SIOCSIFHWADDR:
 582                 return "SIOCSIFHWADDR";
 583         case SIOCSIFMTU:
 584                 return "SIOCSIFMTU";
 585         }
 586         return "UNKNOWN";
 587 }
 588
 589 static int
 590 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 591           struct ifreq *ifr, int set, enum ioctl_mode mode)
 592 {
 593         short req_flags = ifr->ifr_flags;
 594         int remote = pmd->remote_if_index &&
 595                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 596
 597         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 598                 return 0;
 599         /*
 600          * If there is a remote netdevice, apply ioctl on it, then apply it on
 601          * the tap netdevice.
 602          */
 603 apply:
 604         if (remote)
 605                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
 606         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 607                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 608         switch (request) {
 609         case SIOCSIFFLAGS:
 610                 /* fetch current flags to leave other flags untouched */
 611                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 612                         goto error;
 613                 if (set)
 614                         ifr->ifr_flags |= req_flags;
 615                 else
 616                         ifr->ifr_flags &= ~req_flags;
 617                 break;
 618         case SIOCGIFFLAGS:
 619         case SIOCGIFHWADDR:
 620         case SIOCSIFHWADDR:
 621         case SIOCSIFMTU:
 622                 break;
 623         default:
 624                 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
 625                         pmd->name);
 626                 return -EINVAL;
 627         }
 628         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 629                 goto error;
 630         if (remote-- && mode == LOCAL_AND_REMOTE)
 631                 goto apply;
 632         return 0;
 633
 634 error:
 635         RTE_LOG(DEBUG, PMD, "%s: %s(%s) failed: %s(%d)\n", ifr->ifr_name,
 636                 __func__, tap_ioctl_req2str(request), strerror(errno), errno);
 637         return -errno;
 638 }
 639
 640 static int
 641 tap_link_set_down(struct rte_eth_dev *dev)
 642 {
 643         struct pmd_internals *pmd = dev->data->dev_private;
 644         struct ifreq ifr = { .ifr_flags = IFF_UP };
 645
 646         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 647         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 648 }
 649
 650 static int
 651 tap_link_set_up(struct rte_eth_dev *dev)
 652 {
 653         struct pmd_internals *pmd = dev->data->dev_private;
 654         struct ifreq ifr = { .ifr_flags = IFF_UP };
 655
 656         dev->data->dev_link.link_status = ETH_LINK_UP;
 657         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 658 }
 659
 660 static int
 661 tap_dev_start(struct rte_eth_dev *dev)
 662 {
 663         int err;
 664
 665         err = tap_intr_handle_set(dev, 1);
 666         if (err)
 667                 return err;
 668         return tap_link_set_up(dev);
 669 }
 670
 671 /* This function gets called when the current port gets stopped.
 672  */
 673 static void
 674 tap_dev_stop(struct rte_eth_dev *dev)
 675 {
 676         tap_intr_handle_set(dev, 0);
 677         tap_link_set_down(dev);
 678 }
 679
 680 static int
 681 tap_dev_configure(struct rte_eth_dev *dev)
 682 {
 683         uint64_t supp_tx_offloads = tap_tx_offload_get_port_capa();
 684         uint64_t tx_offloads = dev->data->dev_conf.txmode.offloads;
 685
 686         if ((tx_offloads & supp_tx_offloads) != tx_offloads) {
 687                 rte_errno = ENOTSUP;
 688                 RTE_LOG(ERR, PMD,
 689                         "Some Tx offloads are not supported "
 690                         "requested 0x%" PRIx64 " supported 0x%" PRIx64 "\n",
 691                         tx_offloads, supp_tx_offloads);
 692                 return -rte_errno;
 693         }
 694         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 695                 RTE_LOG(ERR, PMD,
 696                         "%s: number of rx queues %d exceeds max num of queues %d\n",
 697                         dev->device->name,
 698                         dev->data->nb_rx_queues,
 699                         RTE_PMD_TAP_MAX_QUEUES);
 700                 return -1;
 701         }
 702         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 703                 RTE_LOG(ERR, PMD,
 704                         "%s: number of tx queues %d exceeds max num of queues %d\n",
 705                         dev->device->name,
 706                         dev->data->nb_tx_queues,
 707                         RTE_PMD_TAP_MAX_QUEUES);
 708                 return -1;
 709         }
 710
 711         RTE_LOG(INFO, PMD, "%s: %p: TX configured queues number: %u\n",
 712              dev->device->name, (void *)dev, dev->data->nb_tx_queues);
 713
 714         RTE_LOG(INFO, PMD, "%s: %p: RX configured queues number: %u\n",
 715              dev->device->name, (void *)dev, dev->data->nb_rx_queues);
 716
 717         return 0;
 718 }
 719
 720 static uint32_t
 721 tap_dev_speed_capa(void)
 722 {
 723         uint32_t speed = pmd_link.link_speed;
 724         uint32_t capa = 0;
 725
 726         if (speed >= ETH_SPEED_NUM_10M)
 727                 capa |= ETH_LINK_SPEED_10M;
 728         if (speed >= ETH_SPEED_NUM_100M)
 729                 capa |= ETH_LINK_SPEED_100M;
 730         if (speed >= ETH_SPEED_NUM_1G)
 731                 capa |= ETH_LINK_SPEED_1G;
 732         if (speed >= ETH_SPEED_NUM_5G)
 733                 capa |= ETH_LINK_SPEED_2_5G;
 734         if (speed >= ETH_SPEED_NUM_5G)
 735                 capa |= ETH_LINK_SPEED_5G;
 736         if (speed >= ETH_SPEED_NUM_10G)
 737                 capa |= ETH_LINK_SPEED_10G;
 738         if (speed >= ETH_SPEED_NUM_20G)
 739                 capa |= ETH_LINK_SPEED_20G;
 740         if (speed >= ETH_SPEED_NUM_25G)
 741                 capa |= ETH_LINK_SPEED_25G;
 742         if (speed >= ETH_SPEED_NUM_40G)
 743                 capa |= ETH_LINK_SPEED_40G;
 744         if (speed >= ETH_SPEED_NUM_50G)
 745                 capa |= ETH_LINK_SPEED_50G;
 746         if (speed >= ETH_SPEED_NUM_56G)
 747                 capa |= ETH_LINK_SPEED_56G;
 748         if (speed >= ETH_SPEED_NUM_100G)
 749                 capa |= ETH_LINK_SPEED_100G;
 750
 751         return capa;
 752 }
 753
 754 static void
 755 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 756 {
 757         struct pmd_internals *internals = dev->data->dev_private;
 758
 759         dev_info->if_index = internals->if_index;
 760         dev_info->max_mac_addrs = 1;
 761         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 762         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
 763         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
 764         dev_info->min_rx_bufsize = 0;
 765         dev_info->speed_capa = tap_dev_speed_capa();
 766         dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
 767         dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
 768                                     dev_info->rx_queue_offload_capa;
 769         dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
 770         dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
 771                                     dev_info->tx_queue_offload_capa;
 772 }
 773
 774 static int
 775 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 776 {
 777         unsigned int i, imax;
 778         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 779         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 780         unsigned long rx_nombuf = 0, ierrors = 0;
 781         const struct pmd_internals *pmd = dev->data->dev_private;
 782
 783         /* rx queue statistics */
 784         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 785                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 786         for (i = 0; i < imax; i++) {
 787                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 788                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 789                 rx_total += tap_stats->q_ipackets[i];
 790                 rx_bytes_total += tap_stats->q_ibytes[i];
 791                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
 792                 ierrors += pmd->rxq[i].stats.ierrors;
 793         }
 794
 795         /* tx queue statistics */
 796         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 797                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 798
 799         for (i = 0; i < imax; i++) {
 800                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 801                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 802                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 803                 tx_total += tap_stats->q_opackets[i];
 804                 tx_err_total += tap_stats->q_errors[i];
 805                 tx_bytes_total += tap_stats->q_obytes[i];
 806         }
 807
 808         tap_stats->ipackets = rx_total;
 809         tap_stats->ibytes = rx_bytes_total;
 810         tap_stats->ierrors = ierrors;
 811         tap_stats->rx_nombuf = rx_nombuf;
 812         tap_stats->opackets = tx_total;
 813         tap_stats->oerrors = tx_err_total;
 814         tap_stats->obytes = tx_bytes_total;
 815         return 0;
 816 }
 817
 818 static void
 819 tap_stats_reset(struct rte_eth_dev *dev)
 820 {
 821         int i;
 822         struct pmd_internals *pmd = dev->data->dev_private;
 823
 824         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 825                 pmd->rxq[i].stats.ipackets = 0;
 826                 pmd->rxq[i].stats.ibytes = 0;
 827                 pmd->rxq[i].stats.ierrors = 0;
 828                 pmd->rxq[i].stats.rx_nombuf = 0;
 829
 830                 pmd->txq[i].stats.opackets = 0;
 831                 pmd->txq[i].stats.errs = 0;
 832                 pmd->txq[i].stats.obytes = 0;
 833         }
 834 }
 835
 836 static void
 837 tap_dev_close(struct rte_eth_dev *dev)
 838 {
 839         int i;
 840         struct pmd_internals *internals = dev->data->dev_private;
 841
 842         tap_link_set_down(dev);
 843         tap_flow_flush(dev, NULL);
 844         tap_flow_implicit_flush(internals, NULL);
 845
 846         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 847                 if (internals->rxq[i].fd != -1) {
 848                         close(internals->rxq[i].fd);
 849                         internals->rxq[i].fd = -1;
 850                 }
 851                 if (internals->txq[i].fd != -1) {
 852                         close(internals->txq[i].fd);
 853                         internals->txq[i].fd = -1;
 854                 }
 855         }
 856
 857         if (internals->remote_if_index) {
 858                 /* Restore initial remote state */
 859                 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
 860                                 &internals->remote_initial_flags);
 861         }
 862 }
 863
 864 static void
 865 tap_rx_queue_release(void *queue)
 866 {
 867         struct rx_queue *rxq = queue;
 868
 869         if (rxq && (rxq->fd > 0)) {
 870                 close(rxq->fd);
 871                 rxq->fd = -1;
 872                 rte_pktmbuf_free(rxq->pool);
 873                 rte_free(rxq->iovecs);
 874                 rxq->pool = NULL;
 875                 rxq->iovecs = NULL;
 876         }
 877 }
 878
 879 static void
 880 tap_tx_queue_release(void *queue)
 881 {
 882         struct tx_queue *txq = queue;
 883
 884         if (txq && (txq->fd > 0)) {
 885                 close(txq->fd);
 886                 txq->fd = -1;
 887         }
 888 }
 889
 890 static int
 891 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
 892 {
 893         struct rte_eth_link *dev_link = &dev->data->dev_link;
 894         struct pmd_internals *pmd = dev->data->dev_private;
 895         struct ifreq ifr = { .ifr_flags = 0 };
 896
 897         if (pmd->remote_if_index) {
 898                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
 899                 if (!(ifr.ifr_flags & IFF_UP) ||
 900                     !(ifr.ifr_flags & IFF_RUNNING)) {
 901                         dev_link->link_status = ETH_LINK_DOWN;
 902                         return 0;
 903                 }
 904         }
 905         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
 906         dev_link->link_status =
 907                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
 908                  ETH_LINK_UP :
 909                  ETH_LINK_DOWN);
 910         return 0;
 911 }
 912
 913 static void
 914 tap_promisc_enable(struct rte_eth_dev *dev)
 915 {
 916         struct pmd_internals *pmd = dev->data->dev_private;
 917         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 918
 919         dev->data->promiscuous = 1;
 920         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 921         if (pmd->remote_if_index && !pmd->flow_isolate)
 922                 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
 923 }
 924
 925 static void
 926 tap_promisc_disable(struct rte_eth_dev *dev)
 927 {
 928         struct pmd_internals *pmd = dev->data->dev_private;
 929         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 930
 931         dev->data->promiscuous = 0;
 932         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 933         if (pmd->remote_if_index && !pmd->flow_isolate)
 934                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
 935 }
 936
 937 static void
 938 tap_allmulti_enable(struct rte_eth_dev *dev)
 939 {
 940         struct pmd_internals *pmd = dev->data->dev_private;
 941         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 942
 943         dev->data->all_multicast = 1;
 944         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 945         if (pmd->remote_if_index && !pmd->flow_isolate)
 946                 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
 947 }
 948
 949 static void
 950 tap_allmulti_disable(struct rte_eth_dev *dev)
 951 {
 952         struct pmd_internals *pmd = dev->data->dev_private;
 953         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 954
 955         dev->data->all_multicast = 0;
 956         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 957         if (pmd->remote_if_index && !pmd->flow_isolate)
 958                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
 959 }
 960
 961 static int
 962 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 963 {
 964         struct pmd_internals *pmd = dev->data->dev_private;
 965         enum ioctl_mode mode = LOCAL_ONLY;
 966         struct ifreq ifr;
 967         int ret;
 968
 969         if (is_zero_ether_addr(mac_addr)) {
 970                 RTE_LOG(ERR, PMD, "%s: can't set an empty MAC address\n",
 971                         dev->device->name);
 972                 return -EINVAL;
 973         }
 974         /* Check the actual current MAC address on the tap netdevice */
 975         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
 976         if (ret < 0)
 977                 return ret;
 978         if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 979                                mac_addr))
 980                 return 0;
 981         /* Check the current MAC address on the remote */
 982         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
 983         if (ret < 0)
 984                 return ret;
 985         if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 986                                mac_addr))
 987                 mode = LOCAL_AND_REMOTE;
 988         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
 989         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
 990         ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
 991         if (ret < 0)
 992                 return ret;
 993         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
 994         if (pmd->remote_if_index && !pmd->flow_isolate) {
 995                 /* Replace MAC redirection rule after a MAC change */
 996                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
 997                 if (ret < 0) {
 998                         RTE_LOG(ERR, PMD,
 999                                 "%s: Couldn't delete MAC redirection rule\n",
1000                                 dev->device->name);
1001                         return ret;
1002                 }
1003                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1004                 if (ret < 0) {
1005                         RTE_LOG(ERR, PMD,
1006                                 "%s: Couldn't add MAC redirection rule\n",
1007                                 dev->device->name);
1008                         return ret;
1009                 }
1010         }
1011
1012         return 0;
1013 }
1014
1015 static int
1016 tap_setup_queue(struct rte_eth_dev *dev,
1017                 struct pmd_internals *internals,
1018                 uint16_t qid,
1019                 int is_rx)
1020 {
1021         int *fd;
1022         int *other_fd;
1023         const char *dir;
1024         struct pmd_internals *pmd = dev->data->dev_private;
1025         struct rx_queue *rx = &internals->rxq[qid];
1026         struct tx_queue *tx = &internals->txq[qid];
1027
1028         if (is_rx) {
1029                 fd = &rx->fd;
1030                 other_fd = &tx->fd;
1031                 dir = "rx";
1032         } else {
1033                 fd = &tx->fd;
1034                 other_fd = &rx->fd;
1035                 dir = "tx";
1036         }
1037         if (*fd != -1) {
1038                 /* fd for this queue already exists */
1039                 RTE_LOG(DEBUG, PMD, "%s: fd %d for %s queue qid %d exists\n",
1040                         pmd->name, *fd, dir, qid);
1041         } else if (*other_fd != -1) {
1042                 /* Only other_fd exists. dup it */
1043                 *fd = dup(*other_fd);
1044                 if (*fd < 0) {
1045                         *fd = -1;
1046                         RTE_LOG(ERR, PMD, "%s: dup() failed.\n",
1047                                 pmd->name);
1048                         return -1;
1049                 }
1050                 RTE_LOG(DEBUG, PMD, "%s: dup fd %d for %s queue qid %d (%d)\n",
1051                         pmd->name, *other_fd, dir, qid, *fd);
1052         } else {
1053                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1054                 *fd = tun_alloc(pmd);
1055                 if (*fd < 0) {
1056                         *fd = -1; /* restore original value */
1057                         RTE_LOG(ERR, PMD, "%s: tun_alloc() failed.\n",
1058                                 pmd->name);
1059                         return -1;
1060                 }
1061                 RTE_LOG(DEBUG, PMD, "%s: add %s queue for qid %d fd %d\n",
1062                         pmd->name, dir, qid, *fd);
1063         }
1064
1065         tx->mtu = &dev->data->mtu;
1066         rx->rxmode = &dev->data->dev_conf.rxmode;
1067
1068         return *fd;
1069 }
1070
1071 static int
1072 tap_rx_queue_setup(struct rte_eth_dev *dev,
1073                    uint16_t rx_queue_id,
1074                    uint16_t nb_rx_desc,
1075                    unsigned int socket_id,
1076                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1077                    struct rte_mempool *mp)
1078 {
1079         struct pmd_internals *internals = dev->data->dev_private;
1080         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1081         struct rte_mbuf **tmp = &rxq->pool;
1082         long iov_max = sysconf(_SC_IOV_MAX);
1083         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1084         struct iovec (*iovecs)[nb_desc + 1];
1085         int data_off = RTE_PKTMBUF_HEADROOM;
1086         int ret = 0;
1087         int fd;
1088         int i;
1089
1090         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1091                 RTE_LOG(WARNING, PMD,
1092                         "nb_rx_queues %d too small or mempool NULL\n",
1093                         dev->data->nb_rx_queues);
1094                 return -1;
1095         }
1096
1097         /* Verify application offloads are valid for our port and queue. */
1098         if (!tap_rxq_are_offloads_valid(dev, rx_conf->offloads)) {
1099                 rte_errno = ENOTSUP;
1100                 RTE_LOG(ERR, PMD,
1101                         "%p: Rx queue offloads 0x%" PRIx64
1102                         " don't match port offloads 0x%" PRIx64
1103                         " or supported offloads 0x%" PRIx64 "\n",
1104                         (void *)dev, rx_conf->offloads,
1105                         dev->data->dev_conf.rxmode.offloads,
1106                         (tap_rx_offload_get_port_capa() |
1107                          tap_rx_offload_get_queue_capa()));
1108                 return -rte_errno;
1109         }
1110         rxq->mp = mp;
1111         rxq->trigger_seen = 1; /* force initial burst */
1112         rxq->in_port = dev->data->port_id;
1113         rxq->nb_rx_desc = nb_desc;
1114         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1115                                     socket_id);
1116         if (!iovecs) {
1117                 RTE_LOG(WARNING, PMD,
1118                         "%s: Couldn't allocate %d RX descriptors\n",
1119                         dev->device->name, nb_desc);
1120                 return -ENOMEM;
1121         }
1122         rxq->iovecs = iovecs;
1123
1124         dev->data->rx_queues[rx_queue_id] = rxq;
1125         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1126         if (fd == -1) {
1127                 ret = fd;
1128                 goto error;
1129         }
1130
1131         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1132         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1133
1134         for (i = 1; i <= nb_desc; i++) {
1135                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1136                 if (!*tmp) {
1137                         RTE_LOG(WARNING, PMD,
1138                                 "%s: couldn't allocate memory for queue %d\n",
1139                                 dev->device->name, rx_queue_id);
1140                         ret = -ENOMEM;
1141                         goto error;
1142                 }
1143                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1144                 (*rxq->iovecs)[i].iov_base =
1145                         (char *)(*tmp)->buf_addr + data_off;
1146                 data_off = 0;
1147                 tmp = &(*tmp)->next;
1148         }
1149
1150         RTE_LOG(DEBUG, PMD, "  RX TUNTAP device name %s, qid %d on fd %d\n",
1151                 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
1152
1153         return 0;
1154
1155 error:
1156         rte_pktmbuf_free(rxq->pool);
1157         rxq->pool = NULL;
1158         rte_free(rxq->iovecs);
1159         rxq->iovecs = NULL;
1160         return ret;
1161 }
1162
1163 static int
1164 tap_tx_queue_setup(struct rte_eth_dev *dev,
1165                    uint16_t tx_queue_id,
1166                    uint16_t nb_tx_desc __rte_unused,
1167                    unsigned int socket_id __rte_unused,
1168                    const struct rte_eth_txconf *tx_conf)
1169 {
1170         struct pmd_internals *internals = dev->data->dev_private;
1171         struct tx_queue *txq;
1172         int ret;
1173
1174         if (tx_queue_id >= dev->data->nb_tx_queues)
1175                 return -1;
1176         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1177         txq = dev->data->tx_queues[tx_queue_id];
1178         /*
1179          * Don't verify port offloads for application which
1180          * use the old API.
1181          */
1182         if (tx_conf != NULL &&
1183             !!(tx_conf->txq_flags & ETH_TXQ_FLAGS_IGNORE)) {
1184                 if (tap_txq_are_offloads_valid(dev, tx_conf->offloads)) {
1185                         txq->csum = !!(tx_conf->offloads &
1186                                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
1187                                          DEV_TX_OFFLOAD_UDP_CKSUM |
1188                                          DEV_TX_OFFLOAD_TCP_CKSUM));
1189                 } else {
1190                         rte_errno = ENOTSUP;
1191                         RTE_LOG(ERR, PMD,
1192                                 "%p: Tx queue offloads 0x%" PRIx64
1193                                 " don't match port offloads 0x%" PRIx64
1194                                 " or supported offloads 0x%" PRIx64,
1195                                 (void *)dev, tx_conf->offloads,
1196                                 dev->data->dev_conf.txmode.offloads,
1197                                 tap_tx_offload_get_port_capa());
1198                         return -rte_errno;
1199                 }
1200         }
1201         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1202         if (ret == -1)
1203                 return -1;
1204         RTE_LOG(DEBUG, PMD,
1205                 "  TX TUNTAP device name %s, qid %d on fd %d csum %s\n",
1206                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd,
1207                 txq->csum ? "on" : "off");
1208
1209         return 0;
1210 }
1211
1212 static int
1213 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1214 {
1215         struct pmd_internals *pmd = dev->data->dev_private;
1216         struct ifreq ifr = { .ifr_mtu = mtu };
1217         int err = 0;
1218
1219         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1220         if (!err)
1221                 dev->data->mtu = mtu;
1222
1223         return err;
1224 }
1225
1226 static int
1227 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1228                      struct ether_addr *mc_addr_set __rte_unused,
1229                      uint32_t nb_mc_addr __rte_unused)
1230 {
1231         /*
1232          * Nothing to do actually: the tap has no filtering whatsoever, every
1233          * packet is received.
1234          */
1235         return 0;
1236 }
1237
1238 static int
1239 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1240 {
1241         struct rte_eth_dev *dev = arg;
1242         struct pmd_internals *pmd = dev->data->dev_private;
1243         struct ifinfomsg *info = NLMSG_DATA(nh);
1244
1245         if (nh->nlmsg_type != RTM_NEWLINK ||
1246             (info->ifi_index != pmd->if_index &&
1247              info->ifi_index != pmd->remote_if_index))
1248                 return 0;
1249         return tap_link_update(dev, 0);
1250 }
1251
1252 static void
1253 tap_dev_intr_handler(void *cb_arg)
1254 {
1255         struct rte_eth_dev *dev = cb_arg;
1256         struct pmd_internals *pmd = dev->data->dev_private;
1257
1258         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1259 }
1260
1261 static int
1262 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1263 {
1264         struct pmd_internals *pmd = dev->data->dev_private;
1265
1266         /* In any case, disable interrupt if the conf is no longer there. */
1267         if (!dev->data->dev_conf.intr_conf.lsc) {
1268                 if (pmd->intr_handle.fd != -1) {
1269                         tap_nl_final(pmd->intr_handle.fd);
1270                         rte_intr_callback_unregister(&pmd->intr_handle,
1271                                 tap_dev_intr_handler, dev);
1272                 }
1273                 return 0;
1274         }
1275         if (set) {
1276                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1277                 if (unlikely(pmd->intr_handle.fd == -1))
1278                         return -EBADF;
1279                 return rte_intr_callback_register(
1280                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1281         }
1282         tap_nl_final(pmd->intr_handle.fd);
1283         return rte_intr_callback_unregister(&pmd->intr_handle,
1284                                             tap_dev_intr_handler, dev);
1285 }
1286
1287 static int
1288 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1289 {
1290         int err;
1291
1292         err = tap_lsc_intr_handle_set(dev, set);
1293         if (err)
1294                 return err;
1295         err = tap_rx_intr_vec_set(dev, set);
1296         if (err && set)
1297                 tap_lsc_intr_handle_set(dev, 0);
1298         return err;
1299 }
1300
1301 static const uint32_t*
1302 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1303 {
1304         static const uint32_t ptypes[] = {
1305                 RTE_PTYPE_INNER_L2_ETHER,
1306                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1307                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1308                 RTE_PTYPE_INNER_L3_IPV4,
1309                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1310                 RTE_PTYPE_INNER_L3_IPV6,
1311                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1312                 RTE_PTYPE_INNER_L4_FRAG,
1313                 RTE_PTYPE_INNER_L4_UDP,
1314                 RTE_PTYPE_INNER_L4_TCP,
1315                 RTE_PTYPE_INNER_L4_SCTP,
1316                 RTE_PTYPE_L2_ETHER,
1317                 RTE_PTYPE_L2_ETHER_VLAN,
1318                 RTE_PTYPE_L2_ETHER_QINQ,
1319                 RTE_PTYPE_L3_IPV4,
1320                 RTE_PTYPE_L3_IPV4_EXT,
1321                 RTE_PTYPE_L3_IPV6_EXT,
1322                 RTE_PTYPE_L3_IPV6,
1323                 RTE_PTYPE_L4_FRAG,
1324                 RTE_PTYPE_L4_UDP,
1325                 RTE_PTYPE_L4_TCP,
1326                 RTE_PTYPE_L4_SCTP,
1327         };
1328
1329         return ptypes;
1330 }
1331
1332 static int
1333 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1334                   struct rte_eth_fc_conf *fc_conf)
1335 {
1336         fc_conf->mode = RTE_FC_NONE;
1337         return 0;
1338 }
1339
1340 static int
1341 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1342                   struct rte_eth_fc_conf *fc_conf)
1343 {
1344         if (fc_conf->mode != RTE_FC_NONE)
1345                 return -ENOTSUP;
1346         return 0;
1347 }
1348
1349 static const struct eth_dev_ops ops = {
1350         .dev_start              = tap_dev_start,
1351         .dev_stop               = tap_dev_stop,
1352         .dev_close              = tap_dev_close,
1353         .dev_configure          = tap_dev_configure,
1354         .dev_infos_get          = tap_dev_info,
1355         .rx_queue_setup         = tap_rx_queue_setup,
1356         .tx_queue_setup         = tap_tx_queue_setup,
1357         .rx_queue_release       = tap_rx_queue_release,
1358         .tx_queue_release       = tap_tx_queue_release,
1359         .flow_ctrl_get          = tap_flow_ctrl_get,
1360         .flow_ctrl_set          = tap_flow_ctrl_set,
1361         .link_update            = tap_link_update,
1362         .dev_set_link_up        = tap_link_set_up,
1363         .dev_set_link_down      = tap_link_set_down,
1364         .promiscuous_enable     = tap_promisc_enable,
1365         .promiscuous_disable    = tap_promisc_disable,
1366         .allmulticast_enable    = tap_allmulti_enable,
1367         .allmulticast_disable   = tap_allmulti_disable,
1368         .mac_addr_set           = tap_mac_set,
1369         .mtu_set                = tap_mtu_set,
1370         .set_mc_addr_list       = tap_set_mc_addr_list,
1371         .stats_get              = tap_stats_get,
1372         .stats_reset            = tap_stats_reset,
1373         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1374         .filter_ctrl            = tap_dev_filter_ctrl,
1375 };
1376
1377 static int
1378 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1379                    char *remote_iface, struct ether_addr *mac_addr)
1380 {
1381         int numa_node = rte_socket_id();
1382         struct rte_eth_dev *dev;
1383         struct pmd_internals *pmd;
1384         struct rte_eth_dev_data *data;
1385         struct ifreq ifr;
1386         int i;
1387
1388         RTE_LOG(DEBUG, PMD, "%s device on numa %u\n",
1389                         tuntap_name, rte_socket_id());
1390
1391         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1392         if (!dev) {
1393                 RTE_LOG(ERR, PMD, "%s Unable to allocate device struct\n",
1394                                 tuntap_name);
1395                 goto error_exit_nodev;
1396         }
1397
1398         pmd = dev->data->dev_private;
1399         pmd->dev = dev;
1400         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1401
1402         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1403         if (pmd->ioctl_sock == -1) {
1404                 RTE_LOG(ERR, PMD,
1405                         "%s Unable to get a socket for management: %s\n",
1406                         tuntap_name, strerror(errno));
1407                 goto error_exit;
1408         }
1409
1410         /* Setup some default values */
1411         data = dev->data;
1412         data->dev_private = pmd;
1413         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1414         data->numa_node = numa_node;
1415
1416         data->dev_link = pmd_link;
1417         data->mac_addrs = &pmd->eth_addr;
1418         /* Set the number of RX and TX queues */
1419         data->nb_rx_queues = 0;
1420         data->nb_tx_queues = 0;
1421
1422         dev->dev_ops = &ops;
1423         dev->rx_pkt_burst = pmd_rx_burst;
1424         dev->tx_pkt_burst = pmd_tx_burst;
1425
1426         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1427         pmd->intr_handle.fd = -1;
1428         dev->intr_handle = &pmd->intr_handle;
1429
1430         /* Presetup the fds to -1 as being not valid */
1431         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1432                 pmd->rxq[i].fd = -1;
1433                 pmd->txq[i].fd = -1;
1434         }
1435
1436         if (tap_type) {
1437                 if (is_zero_ether_addr(mac_addr))
1438                         eth_random_addr((uint8_t *)&pmd->eth_addr);
1439                 else
1440                         rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1441         }
1442
1443         /* Immediately create the netdevice (this will create the 1st queue). */
1444         /* rx queue */
1445         if (tap_setup_queue(dev, pmd, 0, 1) == -1)
1446                 goto error_exit;
1447         /* tx queue */
1448         if (tap_setup_queue(dev, pmd, 0, 0) == -1)
1449                 goto error_exit;
1450
1451         ifr.ifr_mtu = dev->data->mtu;
1452         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1453                 goto error_exit;
1454
1455         if (tap_type) {
1456                 memset(&ifr, 0, sizeof(struct ifreq));
1457                 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1458                 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1459                                 ETHER_ADDR_LEN);
1460                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1461                         goto error_exit;
1462         }
1463
1464         /*
1465          * Set up everything related to rte_flow:
1466          * - netlink socket
1467          * - tap / remote if_index
1468          * - mandatory QDISCs
1469          * - rte_flow actual/implicit lists
1470          * - implicit rules
1471          */
1472         pmd->nlsk_fd = tap_nl_init(0);
1473         if (pmd->nlsk_fd == -1) {
1474                 RTE_LOG(WARNING, PMD, "%s: failed to create netlink socket.\n",
1475                         pmd->name);
1476                 goto disable_rte_flow;
1477         }
1478         pmd->if_index = if_nametoindex(pmd->name);
1479         if (!pmd->if_index) {
1480                 RTE_LOG(ERR, PMD, "%s: failed to get if_index.\n", pmd->name);
1481                 goto disable_rte_flow;
1482         }
1483         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1484                 RTE_LOG(ERR, PMD, "%s: failed to create multiq qdisc.\n",
1485                         pmd->name);
1486                 goto disable_rte_flow;
1487         }
1488         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1489                 RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1490                         pmd->name);
1491                 goto disable_rte_flow;
1492         }
1493         LIST_INIT(&pmd->flows);
1494
1495         if (strlen(remote_iface)) {
1496                 pmd->remote_if_index = if_nametoindex(remote_iface);
1497                 if (!pmd->remote_if_index) {
1498                         RTE_LOG(ERR, PMD, "%s: failed to get %s if_index.\n",
1499                                 pmd->name, remote_iface);
1500                         goto error_remote;
1501                 }
1502                 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1503                          "%s", remote_iface);
1504
1505                 /* Save state of remote device */
1506                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1507
1508                 /* Replicate remote MAC address */
1509                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1510                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1511                                 pmd->name, pmd->remote_iface);
1512                         goto error_remote;
1513                 }
1514                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1515                            ETHER_ADDR_LEN);
1516                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1517                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1518                         RTE_LOG(ERR, PMD, "%s: failed to get %s MAC address.\n",
1519                                 pmd->name, remote_iface);
1520                         goto error_remote;
1521                 }
1522
1523                 /*
1524                  * Flush usually returns negative value because it tries to
1525                  * delete every QDISC (and on a running device, one QDISC at
1526                  * least is needed). Ignore negative return value.
1527                  */
1528                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1529                 if (qdisc_create_ingress(pmd->nlsk_fd,
1530                                          pmd->remote_if_index) < 0) {
1531                         RTE_LOG(ERR, PMD, "%s: failed to create ingress qdisc.\n",
1532                                 pmd->remote_iface);
1533                         goto error_remote;
1534                 }
1535                 LIST_INIT(&pmd->implicit_flows);
1536                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1537                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1538                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1539                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1540                         RTE_LOG(ERR, PMD,
1541                                 "%s: failed to create implicit rules.\n",
1542                                 pmd->name);
1543                         goto error_remote;
1544                 }
1545         }
1546
1547         return 0;
1548
1549 disable_rte_flow:
1550         RTE_LOG(ERR, PMD, " Disabling rte flow support: %s(%d)\n",
1551                 strerror(errno), errno);
1552         if (strlen(remote_iface)) {
1553                 RTE_LOG(ERR, PMD, "Remote feature requires flow support.\n");
1554                 goto error_exit;
1555         }
1556         return 0;
1557
1558 error_remote:
1559         RTE_LOG(ERR, PMD, " Can't set up remote feature: %s(%d)\n",
1560                 strerror(errno), errno);
1561         tap_flow_implicit_flush(pmd, NULL);
1562
1563 error_exit:
1564         if (pmd->ioctl_sock > 0)
1565                 close(pmd->ioctl_sock);
1566         rte_eth_dev_release_port(dev);
1567
1568 error_exit_nodev:
1569         RTE_LOG(ERR, PMD, "%s Unable to initialize %s\n",
1570                 tuntap_name, rte_vdev_device_name(vdev));
1571
1572         return -EINVAL;
1573 }
1574
1575 static int
1576 set_interface_name(const char *key __rte_unused,
1577                    const char *value,
1578                    void *extra_args)
1579 {
1580         char *name = (char *)extra_args;
1581
1582         if (value)
1583                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN - 1);
1584         else
1585                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1586                          DEFAULT_TAP_NAME, (tap_unit - 1));
1587
1588         return 0;
1589 }
1590
1591 static int
1592 set_remote_iface(const char *key __rte_unused,
1593                  const char *value,
1594                  void *extra_args)
1595 {
1596         char *name = (char *)extra_args;
1597
1598         if (value)
1599                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
1600
1601         return 0;
1602 }
1603
1604 static int parse_user_mac(struct ether_addr *user_mac,
1605                 const char *value)
1606 {
1607         unsigned int index = 0;
1608         char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
1609
1610         if (user_mac == NULL || value == NULL)
1611                 return 0;
1612
1613         strlcpy(mac_temp, value, sizeof(mac_temp));
1614         mac_byte = strtok(mac_temp, ":");
1615
1616         while ((mac_byte != NULL) &&
1617                         (strlen(mac_byte) <= 2) &&
1618                         (strlen(mac_byte) == strspn(mac_byte,
1619                                         ETH_TAP_CMP_MAC_FMT))) {
1620                 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
1621                 mac_byte = strtok(NULL, ":");
1622         }
1623
1624         return index;
1625 }
1626
1627 static int
1628 set_mac_type(const char *key __rte_unused,
1629              const char *value,
1630              void *extra_args)
1631 {
1632         struct ether_addr *user_mac = extra_args;
1633
1634         if (!value)
1635                 return 0;
1636
1637         if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
1638                 static int iface_idx;
1639
1640                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1641                 memcpy((char *)user_mac->addr_bytes, "\0dtap", ETHER_ADDR_LEN);
1642                 user_mac->addr_bytes[ETHER_ADDR_LEN - 1] = iface_idx++ + '0';
1643                 goto success;
1644         }
1645
1646         if (parse_user_mac(user_mac, value) != 6)
1647                 goto error;
1648 success:
1649         RTE_LOG(DEBUG, PMD, "TAP user MAC param (%s)\n", value);
1650         return 0;
1651
1652 error:
1653         RTE_LOG(ERR, PMD, "TAP user MAC (%s) is not in format (%s|%s)\n",
1654                 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
1655         return -1;
1656 }
1657
1658 /*
1659  * Open a TUN interface device. TUN PMD
1660  * 1) sets tap_type as false
1661  * 2) intakes iface as argument.
1662  * 3) as interface is virtual set speed to 10G
1663  */
1664 static int
1665 rte_pmd_tun_probe(struct rte_vdev_device *dev)
1666 {
1667         const char *name, *params;
1668         int ret;
1669         struct rte_kvargs *kvlist = NULL;
1670         char tun_name[RTE_ETH_NAME_MAX_LEN];
1671         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1672
1673         tap_type = 0;
1674         strcpy(tuntap_name, "TUN");
1675
1676         name = rte_vdev_device_name(dev);
1677         params = rte_vdev_device_args(dev);
1678         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1679
1680         if (params && (params[0] != '\0')) {
1681                 RTE_LOG(DEBUG, PMD, "parameters (%s)\n", params);
1682
1683                 kvlist = rte_kvargs_parse(params, valid_arguments);
1684                 if (kvlist) {
1685                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1686                                 ret = rte_kvargs_process(kvlist,
1687                                         ETH_TAP_IFACE_ARG,
1688                                         &set_interface_name,
1689                                         tun_name);
1690
1691                                 if (ret == -1)
1692                                         goto leave;
1693                         }
1694                 }
1695         }
1696         pmd_link.link_speed = ETH_SPEED_NUM_10G;
1697
1698         RTE_LOG(NOTICE, PMD, "Initializing pmd_tun for %s as %s\n",
1699                 name, tun_name);
1700
1701         ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0);
1702
1703 leave:
1704         if (ret == -1) {
1705                 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1706                         name, tun_name);
1707                 tun_unit--; /* Restore the unit number */
1708         }
1709         rte_kvargs_free(kvlist);
1710
1711         return ret;
1712 }
1713
1714 /* Open a TAP interface device.
1715  */
1716 static int
1717 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1718 {
1719         const char *name, *params;
1720         int ret;
1721         struct rte_kvargs *kvlist = NULL;
1722         int speed;
1723         char tap_name[RTE_ETH_NAME_MAX_LEN];
1724         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1725         struct ether_addr user_mac = { .addr_bytes = {0} };
1726         struct rte_eth_dev *eth_dev;
1727
1728         tap_type = 1;
1729         strcpy(tuntap_name, "TAP");
1730
1731         name = rte_vdev_device_name(dev);
1732         params = rte_vdev_device_args(dev);
1733
1734         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
1735             strlen(params) == 0) {
1736                 eth_dev = rte_eth_dev_attach_secondary(name);
1737                 if (!eth_dev) {
1738                         RTE_LOG(ERR, PMD, "Failed to probe %s\n", name);
1739                         return -1;
1740                 }
1741                 /* TODO: request info from primary to set up Rx and Tx */
1742                 eth_dev->dev_ops = &ops;
1743                 return 0;
1744         }
1745
1746         speed = ETH_SPEED_NUM_10G;
1747         snprintf(tap_name, sizeof(tap_name), "%s%d",
1748                  DEFAULT_TAP_NAME, tap_unit++);
1749         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1750
1751         if (params && (params[0] != '\0')) {
1752                 RTE_LOG(DEBUG, PMD, "parameters (%s)\n", params);
1753
1754                 kvlist = rte_kvargs_parse(params, valid_arguments);
1755                 if (kvlist) {
1756                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1757                                 ret = rte_kvargs_process(kvlist,
1758                                                          ETH_TAP_IFACE_ARG,
1759                                                          &set_interface_name,
1760                                                          tap_name);
1761                                 if (ret == -1)
1762                                         goto leave;
1763                         }
1764
1765                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1766                                 ret = rte_kvargs_process(kvlist,
1767                                                          ETH_TAP_REMOTE_ARG,
1768                                                          &set_remote_iface,
1769                                                          remote_iface);
1770                                 if (ret == -1)
1771                                         goto leave;
1772                         }
1773
1774                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1775                                 ret = rte_kvargs_process(kvlist,
1776                                                          ETH_TAP_MAC_ARG,
1777                                                          &set_mac_type,
1778                                                          &user_mac);
1779                                 if (ret == -1)
1780                                         goto leave;
1781                         }
1782                 }
1783         }
1784         pmd_link.link_speed = speed;
1785
1786         RTE_LOG(NOTICE, PMD, "Initializing pmd_tap for %s as %s\n",
1787                 name, tap_name);
1788
1789         ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac);
1790
1791 leave:
1792         if (ret == -1) {
1793                 RTE_LOG(ERR, PMD, "Failed to create pmd for %s as %s\n",
1794                         name, tap_name);
1795                 tap_unit--;             /* Restore the unit number */
1796         }
1797         rte_kvargs_free(kvlist);
1798
1799         return ret;
1800 }
1801
1802 /* detach a TUNTAP device.
1803  */
1804 static int
1805 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1806 {
1807         struct rte_eth_dev *eth_dev = NULL;
1808         struct pmd_internals *internals;
1809         int i;
1810
1811         RTE_LOG(DEBUG, PMD, "Closing TUN/TAP Ethernet device on numa %u\n",
1812                 rte_socket_id());
1813
1814         /* find the ethdev entry */
1815         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1816         if (!eth_dev)
1817                 return 0;
1818
1819         internals = eth_dev->data->dev_private;
1820         if (internals->nlsk_fd) {
1821                 tap_flow_flush(eth_dev, NULL);
1822                 tap_flow_implicit_flush(internals, NULL);
1823                 tap_nl_final(internals->nlsk_fd);
1824         }
1825         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1826                 if (internals->rxq[i].fd != -1) {
1827                         close(internals->rxq[i].fd);
1828                         internals->rxq[i].fd = -1;
1829                 }
1830                 if (internals->txq[i].fd != -1) {
1831                         close(internals->txq[i].fd);
1832                         internals->txq[i].fd = -1;
1833                 }
1834         }
1835
1836         close(internals->ioctl_sock);
1837         rte_free(eth_dev->data->dev_private);
1838
1839         rte_eth_dev_release_port(eth_dev);
1840
1841         return 0;
1842 }
1843
1844 static struct rte_vdev_driver pmd_tun_drv = {
1845         .probe = rte_pmd_tun_probe,
1846         .remove = rte_pmd_tap_remove,
1847 };
1848
1849 static struct rte_vdev_driver pmd_tap_drv = {
1850         .probe = rte_pmd_tap_probe,
1851         .remove = rte_pmd_tap_remove,
1852 };
1853 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1854 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
1855 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1856 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
1857                               ETH_TAP_IFACE_ARG "=<string> ");
1858 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1859                               ETH_TAP_IFACE_ARG "=<string> "
1860                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
1861                               ETH_TAP_REMOTE_ARG "=<string>");