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 <tap_rss.h>
  39 #include <rte_eth_tap.h>
  40 #include <tap_flow.h>
  41 #include <tap_netlink.h>
  42 #include <tap_tcmsgs.h>
  43
  44 /* Linux based path to the TUN device */
  45 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  46 #define DEFAULT_TAP_NAME        "dtap"
  47 #define DEFAULT_TUN_NAME        "dtun"
  48
  49 #define ETH_TAP_IFACE_ARG       "iface"
  50 #define ETH_TAP_REMOTE_ARG      "remote"
  51 #define ETH_TAP_MAC_ARG         "mac"
  52 #define ETH_TAP_MAC_FIXED       "fixed"
  53
  54 #define ETH_TAP_USR_MAC_FMT     "xx:xx:xx:xx:xx:xx"
  55 #define ETH_TAP_CMP_MAC_FMT     "0123456789ABCDEFabcdef"
  56 #define ETH_TAP_MAC_ARG_FMT     ETH_TAP_MAC_FIXED "|" ETH_TAP_USR_MAC_FMT
  57
  58 static struct rte_vdev_driver pmd_tap_drv;
  59 static struct rte_vdev_driver pmd_tun_drv;
  60
  61 static const char *valid_arguments[] = {
  62         ETH_TAP_IFACE_ARG,
  63         ETH_TAP_REMOTE_ARG,
  64         ETH_TAP_MAC_ARG,
  65         NULL
  66 };
  67
  68 static int tap_unit;
  69 static int tun_unit;
  70
  71 static int tap_type;
  72 static char tuntap_name[8];
  73
  74 static volatile uint32_t tap_trigger;   /* Rx trigger */
  75
  76 static struct rte_eth_link pmd_link = {
  77         .link_speed = ETH_SPEED_NUM_10G,
  78         .link_duplex = ETH_LINK_FULL_DUPLEX,
  79         .link_status = ETH_LINK_DOWN,
  80         .link_autoneg = ETH_LINK_FIXED,
  81 };
  82
  83 static void
  84 tap_trigger_cb(int sig __rte_unused)
  85 {
  86         /* Valid trigger values are nonzero */
  87         tap_trigger = (tap_trigger + 1) | 0x80000000;
  88 }
  89
  90 /* Specifies on what netdevices the ioctl should be applied */
  91 enum ioctl_mode {
  92         LOCAL_AND_REMOTE,
  93         LOCAL_ONLY,
  94         REMOTE_ONLY,
  95 };
  96
  97 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
  98
  99 /* Tun/Tap allocation routine
 100  *
 101  * name is the number of the interface to use, unless NULL to take the host
 102  * supplied name.
 103  */
 104 static int
 105 tun_alloc(struct pmd_internals *pmd)
 106 {
 107         struct ifreq ifr;
 108 #ifdef IFF_MULTI_QUEUE
 109         unsigned int features;
 110 #endif
 111         int fd;
 112
 113         memset(&ifr, 0, sizeof(struct ifreq));
 114
 115         /*
 116          * Do not set IFF_NO_PI as packet information header will be needed
 117          * to check if a received packet has been truncated.
 118          */
 119         ifr.ifr_flags = (tap_type) ? IFF_TAP : IFF_TUN | IFF_POINTOPOINT;
 120         snprintf(ifr.ifr_name, IFNAMSIZ, "%s", pmd->name);
 121
 122         TAP_LOG(DEBUG, "ifr_name '%s'", ifr.ifr_name);
 123
 124         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 125         if (fd < 0) {
 126                 TAP_LOG(ERR, "Unable to create %s interface", 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                 TAP_LOG(ERR, "%s unable to get TUN/TAP features",
 134                         tuntap_name);
 135                 goto error;
 136         }
 137         TAP_LOG(DEBUG, "%s Features %08x", tuntap_name, features);
 138
 139         if (features & IFF_MULTI_QUEUE) {
 140                 TAP_LOG(DEBUG, "  Multi-queue support for %d queues",
 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                 TAP_LOG(DEBUG, "  Single queue only support");
 148         }
 149
 150         /* Set the TUN/TAP configuration and set the name if needed */
 151         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 152                 TAP_LOG(WARNING, "Unable to set TUNSETIFF for %s: %s",
 153                         ifr.ifr_name, strerror(errno));
 154                 goto error;
 155         }
 156
 157         /* Always set the file descriptor to non-blocking */
 158         if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) {
 159                 TAP_LOG(WARNING,
 160                         "Unable to set %s to nonblocking: %s",
 161                         ifr.ifr_name, strerror(errno));
 162                 goto error;
 163         }
 164
 165         /* Set up trigger to optimize empty Rx bursts */
 166         errno = 0;
 167         do {
 168                 struct sigaction sa;
 169                 int flags = fcntl(fd, F_GETFL);
 170
 171                 if (flags == -1 || sigaction(SIGIO, NULL, &sa) == -1)
 172                         break;
 173                 if (sa.sa_handler != tap_trigger_cb) {
 174                         /*
 175                          * Make sure SIGIO is not already taken. This is done
 176                          * as late as possible to leave the application a
 177                          * chance to set up its own signal handler first.
 178                          */
 179                         if (sa.sa_handler != SIG_IGN &&
 180                             sa.sa_handler != SIG_DFL) {
 181                                 errno = EBUSY;
 182                                 break;
 183                         }
 184                         sa = (struct sigaction){
 185                                 .sa_flags = SA_RESTART,
 186                                 .sa_handler = tap_trigger_cb,
 187                         };
 188                         if (sigaction(SIGIO, &sa, NULL) == -1)
 189                                 break;
 190                 }
 191                 /* Enable SIGIO on file descriptor */
 192                 fcntl(fd, F_SETFL, flags | O_ASYNC);
 193                 fcntl(fd, F_SETOWN, getpid());
 194         } while (0);
 195
 196         if (errno) {
 197                 /* Disable trigger globally in case of error */
 198                 tap_trigger = 0;
 199                 TAP_LOG(WARNING, "Rx trigger disabled: %s",
 200                         strerror(errno));
 201         }
 202
 203         return fd;
 204
 205 error:
 206         if (fd > 0)
 207                 close(fd);
 208         return -1;
 209 }
 210
 211 static void
 212 tap_verify_csum(struct rte_mbuf *mbuf)
 213 {
 214         uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
 215         uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
 216         uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
 217         unsigned int l2_len = sizeof(struct ether_hdr);
 218         unsigned int l3_len;
 219         uint16_t cksum = 0;
 220         void *l3_hdr;
 221         void *l4_hdr;
 222
 223         if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
 224                 l2_len += 4;
 225         else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
 226                 l2_len += 8;
 227         /* Don't verify checksum for packets with discontinuous L2 header */
 228         if (unlikely(l2_len + sizeof(struct ipv4_hdr) >
 229                      rte_pktmbuf_data_len(mbuf)))
 230                 return;
 231         l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
 232         if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 233                 struct ipv4_hdr *iph = l3_hdr;
 234
 235                 /* ihl contains the number of 4-byte words in the header */
 236                 l3_len = 4 * (iph->version_ihl & 0xf);
 237                 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
 238                         return;
 239
 240                 cksum = ~rte_raw_cksum(iph, l3_len);
 241                 mbuf->ol_flags |= cksum ?
 242                         PKT_RX_IP_CKSUM_BAD :
 243                         PKT_RX_IP_CKSUM_GOOD;
 244         } else if (l3 == RTE_PTYPE_L3_IPV6) {
 245                 l3_len = sizeof(struct ipv6_hdr);
 246         } else {
 247                 /* IPv6 extensions are not supported */
 248                 return;
 249         }
 250         if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
 251                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 252                 /* Don't verify checksum for multi-segment packets. */
 253                 if (mbuf->nb_segs > 1)
 254                         return;
 255                 if (l3 == RTE_PTYPE_L3_IPV4)
 256                         cksum = ~rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
 257                 else if (l3 == RTE_PTYPE_L3_IPV6)
 258                         cksum = ~rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
 259                 mbuf->ol_flags |= cksum ?
 260                         PKT_RX_L4_CKSUM_BAD :
 261                         PKT_RX_L4_CKSUM_GOOD;
 262         }
 263 }
 264
 265 static uint64_t
 266 tap_rx_offload_get_port_capa(void)
 267 {
 268         /*
 269          * No specific port Rx offload capabilities.
 270          */
 271         return 0;
 272 }
 273
 274 static uint64_t
 275 tap_rx_offload_get_queue_capa(void)
 276 {
 277         return DEV_RX_OFFLOAD_SCATTER |
 278                DEV_RX_OFFLOAD_IPV4_CKSUM |
 279                DEV_RX_OFFLOAD_UDP_CKSUM |
 280                DEV_RX_OFFLOAD_TCP_CKSUM |
 281                DEV_RX_OFFLOAD_CRC_STRIP;
 282 }
 283
 284 static bool
 285 tap_rxq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 286 {
 287         uint64_t port_offloads = dev->data->dev_conf.rxmode.offloads;
 288         uint64_t queue_supp_offloads = tap_rx_offload_get_queue_capa();
 289         uint64_t port_supp_offloads = tap_rx_offload_get_port_capa();
 290
 291         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 292             offloads)
 293                 return false;
 294         if ((port_offloads ^ offloads) & port_supp_offloads)
 295                 return false;
 296         return true;
 297 }
 298
 299 /* Callback to handle the rx burst of packets to the correct interface and
 300  * file descriptor(s) in a multi-queue setup.
 301  */
 302 static uint16_t
 303 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 304 {
 305         struct rx_queue *rxq = queue;
 306         uint16_t num_rx;
 307         unsigned long num_rx_bytes = 0;
 308         uint32_t trigger = tap_trigger;
 309
 310         if (trigger == rxq->trigger_seen)
 311                 return 0;
 312         if (trigger)
 313                 rxq->trigger_seen = trigger;
 314         rte_compiler_barrier();
 315         for (num_rx = 0; num_rx < nb_pkts; ) {
 316                 struct rte_mbuf *mbuf = rxq->pool;
 317                 struct rte_mbuf *seg = NULL;
 318                 struct rte_mbuf *new_tail = NULL;
 319                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 320                 int len;
 321
 322                 len = readv(rxq->fd, *rxq->iovecs,
 323                             1 +
 324                             (rxq->rxmode->offloads & DEV_RX_OFFLOAD_SCATTER ?
 325                              rxq->nb_rx_desc : 1));
 326                 if (len < (int)sizeof(struct tun_pi))
 327                         break;
 328
 329                 /* Packet couldn't fit in the provided mbuf */
 330                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 331                         rxq->stats.ierrors++;
 332                         continue;
 333                 }
 334
 335                 len -= sizeof(struct tun_pi);
 336
 337                 mbuf->pkt_len = len;
 338                 mbuf->port = rxq->in_port;
 339                 while (1) {
 340                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 341
 342                         if (unlikely(!buf)) {
 343                                 rxq->stats.rx_nombuf++;
 344                                 /* No new buf has been allocated: do nothing */
 345                                 if (!new_tail || !seg)
 346                                         goto end;
 347
 348                                 seg->next = NULL;
 349                                 rte_pktmbuf_free(mbuf);
 350
 351                                 goto end;
 352                         }
 353                         seg = seg ? seg->next : mbuf;
 354                         if (rxq->pool == mbuf)
 355                                 rxq->pool = buf;
 356                         if (new_tail)
 357                                 new_tail->next = buf;
 358                         new_tail = buf;
 359                         new_tail->next = seg->next;
 360
 361                         /* iovecs[0] is reserved for packet info (pi) */
 362                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 363                                 buf->buf_len - data_off;
 364                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 365                                 (char *)buf->buf_addr + data_off;
 366
 367                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 368                         seg->data_off = data_off;
 369
 370                         len -= seg->data_len;
 371                         if (len <= 0)
 372                                 break;
 373                         mbuf->nb_segs++;
 374                         /* First segment has headroom, not the others */
 375                         data_off = 0;
 376                 }
 377                 seg->next = NULL;
 378                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 379                                                       RTE_PTYPE_ALL_MASK);
 380                 if (rxq->rxmode->offloads & DEV_RX_OFFLOAD_CHECKSUM)
 381                         tap_verify_csum(mbuf);
 382
 383                 /* account for the receive frame */
 384                 bufs[num_rx++] = mbuf;
 385                 num_rx_bytes += mbuf->pkt_len;
 386         }
 387 end:
 388         rxq->stats.ipackets += num_rx;
 389         rxq->stats.ibytes += num_rx_bytes;
 390
 391         return num_rx;
 392 }
 393
 394 static uint64_t
 395 tap_tx_offload_get_port_capa(void)
 396 {
 397         /*
 398          * No specific port Tx offload capabilities.
 399          */
 400         return 0;
 401 }
 402
 403 static uint64_t
 404 tap_tx_offload_get_queue_capa(void)
 405 {
 406         return DEV_TX_OFFLOAD_MULTI_SEGS |
 407                DEV_TX_OFFLOAD_IPV4_CKSUM |
 408                DEV_TX_OFFLOAD_UDP_CKSUM |
 409                DEV_TX_OFFLOAD_TCP_CKSUM;
 410 }
 411
 412 static bool
 413 tap_txq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads)
 414 {
 415         uint64_t port_offloads = dev->data->dev_conf.txmode.offloads;
 416         uint64_t queue_supp_offloads = tap_tx_offload_get_queue_capa();
 417         uint64_t port_supp_offloads = tap_tx_offload_get_port_capa();
 418
 419         if ((offloads & (queue_supp_offloads | port_supp_offloads)) !=
 420             offloads)
 421                 return false;
 422         /* Verify we have no conflict with port offloads */
 423         if ((port_offloads ^ offloads) & port_supp_offloads)
 424                 return false;
 425         return true;
 426 }
 427
 428 static void
 429 tap_tx_offload(char *packet, uint64_t ol_flags, unsigned int l2_len,
 430                unsigned int l3_len)
 431 {
 432         void *l3_hdr = packet + l2_len;
 433
 434         if (ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4)) {
 435                 struct ipv4_hdr *iph = l3_hdr;
 436                 uint16_t cksum;
 437
 438                 iph->hdr_checksum = 0;
 439                 cksum = rte_raw_cksum(iph, l3_len);
 440                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 441         }
 442         if (ol_flags & PKT_TX_L4_MASK) {
 443                 uint16_t l4_len;
 444                 uint32_t cksum;
 445                 uint16_t *l4_cksum;
 446                 void *l4_hdr;
 447
 448                 l4_hdr = packet + l2_len + l3_len;
 449                 if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM)
 450                         l4_cksum = &((struct udp_hdr *)l4_hdr)->dgram_cksum;
 451                 else if ((ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM)
 452                         l4_cksum = &((struct tcp_hdr *)l4_hdr)->cksum;
 453                 else
 454                         return;
 455                 *l4_cksum = 0;
 456                 if (ol_flags & PKT_TX_IPV4) {
 457                         struct ipv4_hdr *iph = l3_hdr;
 458
 459                         l4_len = rte_be_to_cpu_16(iph->total_length) - l3_len;
 460                         cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 461                 } else {
 462                         struct ipv6_hdr *ip6h = l3_hdr;
 463
 464                         /* payload_len does not include ext headers */
 465                         l4_len = rte_be_to_cpu_16(ip6h->payload_len) -
 466                                 l3_len + sizeof(struct ipv6_hdr);
 467                         cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 468                 }
 469                 cksum += rte_raw_cksum(l4_hdr, l4_len);
 470                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 471                 cksum = (~cksum) & 0xffff;
 472                 if (cksum == 0)
 473                         cksum = 0xffff;
 474                 *l4_cksum = cksum;
 475         }
 476 }
 477
 478 /* Callback to handle sending packets from the tap interface
 479  */
 480 static uint16_t
 481 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 482 {
 483         struct tx_queue *txq = queue;
 484         uint16_t num_tx = 0;
 485         unsigned long num_tx_bytes = 0;
 486         uint32_t max_size;
 487         int i;
 488
 489         if (unlikely(nb_pkts == 0))
 490                 return 0;
 491
 492         max_size = *txq->mtu + (ETHER_HDR_LEN + ETHER_CRC_LEN + 4);
 493         for (i = 0; i < nb_pkts; i++) {
 494                 struct rte_mbuf *mbuf = bufs[num_tx];
 495                 struct iovec iovecs[mbuf->nb_segs + 1];
 496                 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
 497                 struct rte_mbuf *seg = mbuf;
 498                 char m_copy[mbuf->data_len];
 499                 int n;
 500                 int j;
 501
 502                 /* stats.errs will be incremented */
 503                 if (rte_pktmbuf_pkt_len(mbuf) > max_size)
 504                         break;
 505
 506                 /*
 507                  * TUN and TAP are created with IFF_NO_PI disabled.
 508                  * For TUN PMD this mandatory as fields are used by
 509                  * Kernel tun.c to determine whether its IP or non IP
 510                  * packets.
 511                  *
 512                  * The logic fetches the first byte of data from mbuf.
 513                  * compares whether its v4 or v6. If none matches default
 514                  * value 0x00 is taken for protocol field.
 515                  */
 516                 char *buff_data = rte_pktmbuf_mtod(seg, void *);
 517                 j = (*buff_data & 0xf0);
 518                 pi.proto = (j == 0x40) ? 0x0008 :
 519                                 (j == 0x60) ? 0xdd86 : 0x00;
 520
 521                 iovecs[0].iov_base = &pi;
 522                 iovecs[0].iov_len = sizeof(pi);
 523                 for (j = 1; j <= mbuf->nb_segs; j++) {
 524                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 525                         iovecs[j].iov_base =
 526                                 rte_pktmbuf_mtod(seg, void *);
 527                         seg = seg->next;
 528                 }
 529                 if (txq->csum &&
 530                     ((mbuf->ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_IPV4) ||
 531                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_UDP_CKSUM ||
 532                      (mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM))) {
 533                         /* Support only packets with all data in the same seg */
 534                         if (mbuf->nb_segs > 1)
 535                                 break;
 536                         /* To change checksums, work on a copy of data. */
 537                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 538                                    rte_pktmbuf_data_len(mbuf));
 539                         tap_tx_offload(m_copy, mbuf->ol_flags,
 540                                        mbuf->l2_len, mbuf->l3_len);
 541                         iovecs[1].iov_base = m_copy;
 542                 }
 543                 /* copy the tx frame data */
 544                 n = writev(txq->fd, iovecs, mbuf->nb_segs + 1);
 545                 if (n <= 0)
 546                         break;
 547
 548                 num_tx++;
 549                 num_tx_bytes += mbuf->pkt_len;
 550                 rte_pktmbuf_free(mbuf);
 551         }
 552
 553         txq->stats.opackets += num_tx;
 554         txq->stats.errs += nb_pkts - num_tx;
 555         txq->stats.obytes += num_tx_bytes;
 556
 557         return num_tx;
 558 }
 559
 560 static const char *
 561 tap_ioctl_req2str(unsigned long request)
 562 {
 563         switch (request) {
 564         case SIOCSIFFLAGS:
 565                 return "SIOCSIFFLAGS";
 566         case SIOCGIFFLAGS:
 567                 return "SIOCGIFFLAGS";
 568         case SIOCGIFHWADDR:
 569                 return "SIOCGIFHWADDR";
 570         case SIOCSIFHWADDR:
 571                 return "SIOCSIFHWADDR";
 572         case SIOCSIFMTU:
 573                 return "SIOCSIFMTU";
 574         }
 575         return "UNKNOWN";
 576 }
 577
 578 static int
 579 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 580           struct ifreq *ifr, int set, enum ioctl_mode mode)
 581 {
 582         short req_flags = ifr->ifr_flags;
 583         int remote = pmd->remote_if_index &&
 584                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 585
 586         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 587                 return 0;
 588         /*
 589          * If there is a remote netdevice, apply ioctl on it, then apply it on
 590          * the tap netdevice.
 591          */
 592 apply:
 593         if (remote)
 594                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->remote_iface);
 595         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 596                 snprintf(ifr->ifr_name, IFNAMSIZ, "%s", pmd->name);
 597         switch (request) {
 598         case SIOCSIFFLAGS:
 599                 /* fetch current flags to leave other flags untouched */
 600                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 601                         goto error;
 602                 if (set)
 603                         ifr->ifr_flags |= req_flags;
 604                 else
 605                         ifr->ifr_flags &= ~req_flags;
 606                 break;
 607         case SIOCGIFFLAGS:
 608         case SIOCGIFHWADDR:
 609         case SIOCSIFHWADDR:
 610         case SIOCSIFMTU:
 611                 break;
 612         default:
 613                 RTE_LOG(WARNING, PMD, "%s: ioctl() called with wrong arg\n",
 614                         pmd->name);
 615                 return -EINVAL;
 616         }
 617         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 618                 goto error;
 619         if (remote-- && mode == LOCAL_AND_REMOTE)
 620                 goto apply;
 621         return 0;
 622
 623 error:
 624         TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
 625                 tap_ioctl_req2str(request), strerror(errno), errno);
 626         return -errno;
 627 }
 628
 629 static int
 630 tap_link_set_down(struct rte_eth_dev *dev)
 631 {
 632         struct pmd_internals *pmd = dev->data->dev_private;
 633         struct ifreq ifr = { .ifr_flags = IFF_UP };
 634
 635         dev->data->dev_link.link_status = ETH_LINK_DOWN;
 636         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 637 }
 638
 639 static int
 640 tap_link_set_up(struct rte_eth_dev *dev)
 641 {
 642         struct pmd_internals *pmd = dev->data->dev_private;
 643         struct ifreq ifr = { .ifr_flags = IFF_UP };
 644
 645         dev->data->dev_link.link_status = ETH_LINK_UP;
 646         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 647 }
 648
 649 static int
 650 tap_dev_start(struct rte_eth_dev *dev)
 651 {
 652         int err;
 653
 654         err = tap_intr_handle_set(dev, 1);
 655         if (err)
 656                 return err;
 657         return tap_link_set_up(dev);
 658 }
 659
 660 /* This function gets called when the current port gets stopped.
 661  */
 662 static void
 663 tap_dev_stop(struct rte_eth_dev *dev)
 664 {
 665         tap_intr_handle_set(dev, 0);
 666         tap_link_set_down(dev);
 667 }
 668
 669 static int
 670 tap_dev_configure(struct rte_eth_dev *dev)
 671 {
 672         uint64_t supp_tx_offloads = tap_tx_offload_get_port_capa() |
 673                                 tap_tx_offload_get_queue_capa();
 674         uint64_t tx_offloads = dev->data->dev_conf.txmode.offloads;
 675
 676         if ((tx_offloads & supp_tx_offloads) != tx_offloads) {
 677                 rte_errno = ENOTSUP;
 678                 TAP_LOG(ERR,
 679                         "Some Tx offloads are not supported "
 680                         "requested 0x%" PRIx64 " supported 0x%" PRIx64,
 681                         tx_offloads, supp_tx_offloads);
 682                 return -rte_errno;
 683         }
 684         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 685                 TAP_LOG(ERR,
 686                         "%s: number of rx queues %d exceeds max num of queues %d",
 687                         dev->device->name,
 688                         dev->data->nb_rx_queues,
 689                         RTE_PMD_TAP_MAX_QUEUES);
 690                 return -1;
 691         }
 692         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 693                 TAP_LOG(ERR,
 694                         "%s: number of tx queues %d exceeds max num of queues %d",
 695                         dev->device->name,
 696                         dev->data->nb_tx_queues,
 697                         RTE_PMD_TAP_MAX_QUEUES);
 698                 return -1;
 699         }
 700
 701         TAP_LOG(INFO, "%s: %p: TX configured queues number: %u",
 702                 dev->device->name, (void *)dev, dev->data->nb_tx_queues);
 703
 704         TAP_LOG(INFO, "%s: %p: RX configured queues number: %u",
 705                 dev->device->name, (void *)dev, dev->data->nb_rx_queues);
 706
 707         return 0;
 708 }
 709
 710 static uint32_t
 711 tap_dev_speed_capa(void)
 712 {
 713         uint32_t speed = pmd_link.link_speed;
 714         uint32_t capa = 0;
 715
 716         if (speed >= ETH_SPEED_NUM_10M)
 717                 capa |= ETH_LINK_SPEED_10M;
 718         if (speed >= ETH_SPEED_NUM_100M)
 719                 capa |= ETH_LINK_SPEED_100M;
 720         if (speed >= ETH_SPEED_NUM_1G)
 721                 capa |= ETH_LINK_SPEED_1G;
 722         if (speed >= ETH_SPEED_NUM_5G)
 723                 capa |= ETH_LINK_SPEED_2_5G;
 724         if (speed >= ETH_SPEED_NUM_5G)
 725                 capa |= ETH_LINK_SPEED_5G;
 726         if (speed >= ETH_SPEED_NUM_10G)
 727                 capa |= ETH_LINK_SPEED_10G;
 728         if (speed >= ETH_SPEED_NUM_20G)
 729                 capa |= ETH_LINK_SPEED_20G;
 730         if (speed >= ETH_SPEED_NUM_25G)
 731                 capa |= ETH_LINK_SPEED_25G;
 732         if (speed >= ETH_SPEED_NUM_40G)
 733                 capa |= ETH_LINK_SPEED_40G;
 734         if (speed >= ETH_SPEED_NUM_50G)
 735                 capa |= ETH_LINK_SPEED_50G;
 736         if (speed >= ETH_SPEED_NUM_56G)
 737                 capa |= ETH_LINK_SPEED_56G;
 738         if (speed >= ETH_SPEED_NUM_100G)
 739                 capa |= ETH_LINK_SPEED_100G;
 740
 741         return capa;
 742 }
 743
 744 static void
 745 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 746 {
 747         struct pmd_internals *internals = dev->data->dev_private;
 748
 749         dev_info->if_index = internals->if_index;
 750         dev_info->max_mac_addrs = 1;
 751         dev_info->max_rx_pktlen = (uint32_t)ETHER_MAX_VLAN_FRAME_LEN;
 752         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
 753         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
 754         dev_info->min_rx_bufsize = 0;
 755         dev_info->speed_capa = tap_dev_speed_capa();
 756         dev_info->rx_queue_offload_capa = tap_rx_offload_get_queue_capa();
 757         dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() |
 758                                     dev_info->rx_queue_offload_capa;
 759         dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa();
 760         dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() |
 761                                     dev_info->tx_queue_offload_capa;
 762         dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
 763         /*
 764          * limitation: TAP suppors all of the following hash
 765          * functions together and not in partial combinations
 766          */
 767         dev_info->flow_type_rss_offloads =
 768                 ETH_RSS_IP | ETH_RSS_UDP | ETH_RSS_TCP;
 769 }
 770
 771 static int
 772 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
 773 {
 774         unsigned int i, imax;
 775         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
 776         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
 777         unsigned long rx_nombuf = 0, ierrors = 0;
 778         const struct pmd_internals *pmd = dev->data->dev_private;
 779
 780         /* rx queue statistics */
 781         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 782                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 783         for (i = 0; i < imax; i++) {
 784                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
 785                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
 786                 rx_total += tap_stats->q_ipackets[i];
 787                 rx_bytes_total += tap_stats->q_ibytes[i];
 788                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
 789                 ierrors += pmd->rxq[i].stats.ierrors;
 790         }
 791
 792         /* tx queue statistics */
 793         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
 794                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
 795
 796         for (i = 0; i < imax; i++) {
 797                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
 798                 tap_stats->q_errors[i] = pmd->txq[i].stats.errs;
 799                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
 800                 tx_total += tap_stats->q_opackets[i];
 801                 tx_err_total += tap_stats->q_errors[i];
 802                 tx_bytes_total += tap_stats->q_obytes[i];
 803         }
 804
 805         tap_stats->ipackets = rx_total;
 806         tap_stats->ibytes = rx_bytes_total;
 807         tap_stats->ierrors = ierrors;
 808         tap_stats->rx_nombuf = rx_nombuf;
 809         tap_stats->opackets = tx_total;
 810         tap_stats->oerrors = tx_err_total;
 811         tap_stats->obytes = tx_bytes_total;
 812         return 0;
 813 }
 814
 815 static void
 816 tap_stats_reset(struct rte_eth_dev *dev)
 817 {
 818         int i;
 819         struct pmd_internals *pmd = dev->data->dev_private;
 820
 821         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 822                 pmd->rxq[i].stats.ipackets = 0;
 823                 pmd->rxq[i].stats.ibytes = 0;
 824                 pmd->rxq[i].stats.ierrors = 0;
 825                 pmd->rxq[i].stats.rx_nombuf = 0;
 826
 827                 pmd->txq[i].stats.opackets = 0;
 828                 pmd->txq[i].stats.errs = 0;
 829                 pmd->txq[i].stats.obytes = 0;
 830         }
 831 }
 832
 833 static void
 834 tap_dev_close(struct rte_eth_dev *dev)
 835 {
 836         int i;
 837         struct pmd_internals *internals = dev->data->dev_private;
 838
 839         tap_link_set_down(dev);
 840         tap_flow_flush(dev, NULL);
 841         tap_flow_implicit_flush(internals, NULL);
 842
 843         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
 844                 if (internals->rxq[i].fd != -1) {
 845                         close(internals->rxq[i].fd);
 846                         internals->rxq[i].fd = -1;
 847                 }
 848                 if (internals->txq[i].fd != -1) {
 849                         close(internals->txq[i].fd);
 850                         internals->txq[i].fd = -1;
 851                 }
 852         }
 853
 854         if (internals->remote_if_index) {
 855                 /* Restore initial remote state */
 856                 ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
 857                                 &internals->remote_initial_flags);
 858         }
 859 }
 860
 861 static void
 862 tap_rx_queue_release(void *queue)
 863 {
 864         struct rx_queue *rxq = queue;
 865
 866         if (rxq && (rxq->fd > 0)) {
 867                 close(rxq->fd);
 868                 rxq->fd = -1;
 869                 rte_pktmbuf_free(rxq->pool);
 870                 rte_free(rxq->iovecs);
 871                 rxq->pool = NULL;
 872                 rxq->iovecs = NULL;
 873         }
 874 }
 875
 876 static void
 877 tap_tx_queue_release(void *queue)
 878 {
 879         struct tx_queue *txq = queue;
 880
 881         if (txq && (txq->fd > 0)) {
 882                 close(txq->fd);
 883                 txq->fd = -1;
 884         }
 885 }
 886
 887 static int
 888 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
 889 {
 890         struct rte_eth_link *dev_link = &dev->data->dev_link;
 891         struct pmd_internals *pmd = dev->data->dev_private;
 892         struct ifreq ifr = { .ifr_flags = 0 };
 893
 894         if (pmd->remote_if_index) {
 895                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
 896                 if (!(ifr.ifr_flags & IFF_UP) ||
 897                     !(ifr.ifr_flags & IFF_RUNNING)) {
 898                         dev_link->link_status = ETH_LINK_DOWN;
 899                         return 0;
 900                 }
 901         }
 902         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
 903         dev_link->link_status =
 904                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
 905                  ETH_LINK_UP :
 906                  ETH_LINK_DOWN);
 907         return 0;
 908 }
 909
 910 static void
 911 tap_promisc_enable(struct rte_eth_dev *dev)
 912 {
 913         struct pmd_internals *pmd = dev->data->dev_private;
 914         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 915
 916         dev->data->promiscuous = 1;
 917         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 918         if (pmd->remote_if_index && !pmd->flow_isolate)
 919                 tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
 920 }
 921
 922 static void
 923 tap_promisc_disable(struct rte_eth_dev *dev)
 924 {
 925         struct pmd_internals *pmd = dev->data->dev_private;
 926         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
 927
 928         dev->data->promiscuous = 0;
 929         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 930         if (pmd->remote_if_index && !pmd->flow_isolate)
 931                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
 932 }
 933
 934 static void
 935 tap_allmulti_enable(struct rte_eth_dev *dev)
 936 {
 937         struct pmd_internals *pmd = dev->data->dev_private;
 938         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 939
 940         dev->data->all_multicast = 1;
 941         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 942         if (pmd->remote_if_index && !pmd->flow_isolate)
 943                 tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
 944 }
 945
 946 static void
 947 tap_allmulti_disable(struct rte_eth_dev *dev)
 948 {
 949         struct pmd_internals *pmd = dev->data->dev_private;
 950         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
 951
 952         dev->data->all_multicast = 0;
 953         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
 954         if (pmd->remote_if_index && !pmd->flow_isolate)
 955                 tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
 956 }
 957
 958 static int
 959 tap_mac_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 960 {
 961         struct pmd_internals *pmd = dev->data->dev_private;
 962         enum ioctl_mode mode = LOCAL_ONLY;
 963         struct ifreq ifr;
 964         int ret;
 965
 966         if (is_zero_ether_addr(mac_addr)) {
 967                 TAP_LOG(ERR, "%s: can't set an empty MAC address",
 968                         dev->device->name);
 969                 return -EINVAL;
 970         }
 971         /* Check the actual current MAC address on the tap netdevice */
 972         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
 973         if (ret < 0)
 974                 return ret;
 975         if (is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 976                                mac_addr))
 977                 return 0;
 978         /* Check the current MAC address on the remote */
 979         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
 980         if (ret < 0)
 981                 return ret;
 982         if (!is_same_ether_addr((struct ether_addr *)&ifr.ifr_hwaddr.sa_data,
 983                                mac_addr))
 984                 mode = LOCAL_AND_REMOTE;
 985         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
 986         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, ETHER_ADDR_LEN);
 987         ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
 988         if (ret < 0)
 989                 return ret;
 990         rte_memcpy(&pmd->eth_addr, mac_addr, ETHER_ADDR_LEN);
 991         if (pmd->remote_if_index && !pmd->flow_isolate) {
 992                 /* Replace MAC redirection rule after a MAC change */
 993                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
 994                 if (ret < 0) {
 995                         TAP_LOG(ERR,
 996                                 "%s: Couldn't delete MAC redirection rule",
 997                                 dev->device->name);
 998                         return ret;
 999                 }
1000                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1001                 if (ret < 0) {
1002                         TAP_LOG(ERR,
1003                                 "%s: Couldn't add MAC redirection rule",
1004                                 dev->device->name);
1005                         return ret;
1006                 }
1007         }
1008
1009         return 0;
1010 }
1011
1012 static int
1013 tap_setup_queue(struct rte_eth_dev *dev,
1014                 struct pmd_internals *internals,
1015                 uint16_t qid,
1016                 int is_rx)
1017 {
1018         int *fd;
1019         int *other_fd;
1020         const char *dir;
1021         struct pmd_internals *pmd = dev->data->dev_private;
1022         struct rx_queue *rx = &internals->rxq[qid];
1023         struct tx_queue *tx = &internals->txq[qid];
1024
1025         if (is_rx) {
1026                 fd = &rx->fd;
1027                 other_fd = &tx->fd;
1028                 dir = "rx";
1029         } else {
1030                 fd = &tx->fd;
1031                 other_fd = &rx->fd;
1032                 dir = "tx";
1033         }
1034         if (*fd != -1) {
1035                 /* fd for this queue already exists */
1036                 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1037                         pmd->name, *fd, dir, qid);
1038         } else if (*other_fd != -1) {
1039                 /* Only other_fd exists. dup it */
1040                 *fd = dup(*other_fd);
1041                 if (*fd < 0) {
1042                         *fd = -1;
1043                         TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1044                         return -1;
1045                 }
1046                 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1047                         pmd->name, *other_fd, dir, qid, *fd);
1048         } else {
1049                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1050                 *fd = tun_alloc(pmd);
1051                 if (*fd < 0) {
1052                         *fd = -1; /* restore original value */
1053                         TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1054                         return -1;
1055                 }
1056                 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1057                         pmd->name, dir, qid, *fd);
1058         }
1059
1060         tx->mtu = &dev->data->mtu;
1061         rx->rxmode = &dev->data->dev_conf.rxmode;
1062
1063         return *fd;
1064 }
1065
1066 static int
1067 tap_rx_queue_setup(struct rte_eth_dev *dev,
1068                    uint16_t rx_queue_id,
1069                    uint16_t nb_rx_desc,
1070                    unsigned int socket_id,
1071                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1072                    struct rte_mempool *mp)
1073 {
1074         struct pmd_internals *internals = dev->data->dev_private;
1075         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1076         struct rte_mbuf **tmp = &rxq->pool;
1077         long iov_max = sysconf(_SC_IOV_MAX);
1078         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1079         struct iovec (*iovecs)[nb_desc + 1];
1080         int data_off = RTE_PKTMBUF_HEADROOM;
1081         int ret = 0;
1082         int fd;
1083         int i;
1084
1085         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1086                 TAP_LOG(WARNING,
1087                         "nb_rx_queues %d too small or mempool NULL",
1088                         dev->data->nb_rx_queues);
1089                 return -1;
1090         }
1091
1092         /* Verify application offloads are valid for our port and queue. */
1093         if (!tap_rxq_are_offloads_valid(dev, rx_conf->offloads)) {
1094                 rte_errno = ENOTSUP;
1095                 TAP_LOG(ERR,
1096                         "%p: Rx queue offloads 0x%" PRIx64
1097                         " don't match port offloads 0x%" PRIx64
1098                         " or supported offloads 0x%" PRIx64,
1099                         (void *)dev, rx_conf->offloads,
1100                         dev->data->dev_conf.rxmode.offloads,
1101                         (tap_rx_offload_get_port_capa() |
1102                          tap_rx_offload_get_queue_capa()));
1103                 return -rte_errno;
1104         }
1105         rxq->mp = mp;
1106         rxq->trigger_seen = 1; /* force initial burst */
1107         rxq->in_port = dev->data->port_id;
1108         rxq->nb_rx_desc = nb_desc;
1109         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1110                                     socket_id);
1111         if (!iovecs) {
1112                 TAP_LOG(WARNING,
1113                         "%s: Couldn't allocate %d RX descriptors",
1114                         dev->device->name, nb_desc);
1115                 return -ENOMEM;
1116         }
1117         rxq->iovecs = iovecs;
1118
1119         dev->data->rx_queues[rx_queue_id] = rxq;
1120         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1121         if (fd == -1) {
1122                 ret = fd;
1123                 goto error;
1124         }
1125
1126         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1127         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1128
1129         for (i = 1; i <= nb_desc; i++) {
1130                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1131                 if (!*tmp) {
1132                         TAP_LOG(WARNING,
1133                                 "%s: couldn't allocate memory for queue %d",
1134                                 dev->device->name, rx_queue_id);
1135                         ret = -ENOMEM;
1136                         goto error;
1137                 }
1138                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1139                 (*rxq->iovecs)[i].iov_base =
1140                         (char *)(*tmp)->buf_addr + data_off;
1141                 data_off = 0;
1142                 tmp = &(*tmp)->next;
1143         }
1144
1145         TAP_LOG(DEBUG, "  RX TUNTAP device name %s, qid %d on fd %d",
1146                 internals->name, rx_queue_id, internals->rxq[rx_queue_id].fd);
1147
1148         return 0;
1149
1150 error:
1151         rte_pktmbuf_free(rxq->pool);
1152         rxq->pool = NULL;
1153         rte_free(rxq->iovecs);
1154         rxq->iovecs = NULL;
1155         return ret;
1156 }
1157
1158 static int
1159 tap_tx_queue_setup(struct rte_eth_dev *dev,
1160                    uint16_t tx_queue_id,
1161                    uint16_t nb_tx_desc __rte_unused,
1162                    unsigned int socket_id __rte_unused,
1163                    const struct rte_eth_txconf *tx_conf)
1164 {
1165         struct pmd_internals *internals = dev->data->dev_private;
1166         struct tx_queue *txq;
1167         int ret;
1168
1169         if (tx_queue_id >= dev->data->nb_tx_queues)
1170                 return -1;
1171         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1172         txq = dev->data->tx_queues[tx_queue_id];
1173         /*
1174          * Don't verify port offloads for application which
1175          * use the old API.
1176          */
1177         if (tx_conf != NULL &&
1178             !!(tx_conf->txq_flags & ETH_TXQ_FLAGS_IGNORE)) {
1179                 if (tap_txq_are_offloads_valid(dev, tx_conf->offloads)) {
1180                         txq->csum = !!(tx_conf->offloads &
1181                                         (DEV_TX_OFFLOAD_IPV4_CKSUM |
1182                                          DEV_TX_OFFLOAD_UDP_CKSUM |
1183                                          DEV_TX_OFFLOAD_TCP_CKSUM));
1184                 } else {
1185                         rte_errno = ENOTSUP;
1186                         TAP_LOG(ERR,
1187                                 "%p: Tx queue offloads 0x%" PRIx64
1188                                 " don't match port offloads 0x%" PRIx64
1189                                 " or supported offloads 0x%" PRIx64,
1190                                 (void *)dev, tx_conf->offloads,
1191                                 dev->data->dev_conf.txmode.offloads,
1192                                 (tap_tx_offload_get_port_capa() |
1193                                 tap_tx_offload_get_queue_capa()));
1194                         return -rte_errno;
1195                 }
1196         }
1197         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1198         if (ret == -1)
1199                 return -1;
1200         TAP_LOG(DEBUG,
1201                 "  TX TUNTAP device name %s, qid %d on fd %d csum %s",
1202                 internals->name, tx_queue_id, internals->txq[tx_queue_id].fd,
1203                 txq->csum ? "on" : "off");
1204
1205         return 0;
1206 }
1207
1208 static int
1209 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1210 {
1211         struct pmd_internals *pmd = dev->data->dev_private;
1212         struct ifreq ifr = { .ifr_mtu = mtu };
1213         int err = 0;
1214
1215         err = tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1216         if (!err)
1217                 dev->data->mtu = mtu;
1218
1219         return err;
1220 }
1221
1222 static int
1223 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1224                      struct ether_addr *mc_addr_set __rte_unused,
1225                      uint32_t nb_mc_addr __rte_unused)
1226 {
1227         /*
1228          * Nothing to do actually: the tap has no filtering whatsoever, every
1229          * packet is received.
1230          */
1231         return 0;
1232 }
1233
1234 static int
1235 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1236 {
1237         struct rte_eth_dev *dev = arg;
1238         struct pmd_internals *pmd = dev->data->dev_private;
1239         struct ifinfomsg *info = NLMSG_DATA(nh);
1240
1241         if (nh->nlmsg_type != RTM_NEWLINK ||
1242             (info->ifi_index != pmd->if_index &&
1243              info->ifi_index != pmd->remote_if_index))
1244                 return 0;
1245         return tap_link_update(dev, 0);
1246 }
1247
1248 static void
1249 tap_dev_intr_handler(void *cb_arg)
1250 {
1251         struct rte_eth_dev *dev = cb_arg;
1252         struct pmd_internals *pmd = dev->data->dev_private;
1253
1254         tap_nl_recv(pmd->intr_handle.fd, tap_nl_msg_handler, dev);
1255 }
1256
1257 static int
1258 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1259 {
1260         struct pmd_internals *pmd = dev->data->dev_private;
1261
1262         /* In any case, disable interrupt if the conf is no longer there. */
1263         if (!dev->data->dev_conf.intr_conf.lsc) {
1264                 if (pmd->intr_handle.fd != -1) {
1265                         tap_nl_final(pmd->intr_handle.fd);
1266                         rte_intr_callback_unregister(&pmd->intr_handle,
1267                                 tap_dev_intr_handler, dev);
1268                 }
1269                 return 0;
1270         }
1271         if (set) {
1272                 pmd->intr_handle.fd = tap_nl_init(RTMGRP_LINK);
1273                 if (unlikely(pmd->intr_handle.fd == -1))
1274                         return -EBADF;
1275                 return rte_intr_callback_register(
1276                         &pmd->intr_handle, tap_dev_intr_handler, dev);
1277         }
1278         tap_nl_final(pmd->intr_handle.fd);
1279         return rte_intr_callback_unregister(&pmd->intr_handle,
1280                                             tap_dev_intr_handler, dev);
1281 }
1282
1283 static int
1284 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1285 {
1286         int err;
1287
1288         err = tap_lsc_intr_handle_set(dev, set);
1289         if (err)
1290                 return err;
1291         err = tap_rx_intr_vec_set(dev, set);
1292         if (err && set)
1293                 tap_lsc_intr_handle_set(dev, 0);
1294         return err;
1295 }
1296
1297 static const uint32_t*
1298 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1299 {
1300         static const uint32_t ptypes[] = {
1301                 RTE_PTYPE_INNER_L2_ETHER,
1302                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1303                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1304                 RTE_PTYPE_INNER_L3_IPV4,
1305                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1306                 RTE_PTYPE_INNER_L3_IPV6,
1307                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1308                 RTE_PTYPE_INNER_L4_FRAG,
1309                 RTE_PTYPE_INNER_L4_UDP,
1310                 RTE_PTYPE_INNER_L4_TCP,
1311                 RTE_PTYPE_INNER_L4_SCTP,
1312                 RTE_PTYPE_L2_ETHER,
1313                 RTE_PTYPE_L2_ETHER_VLAN,
1314                 RTE_PTYPE_L2_ETHER_QINQ,
1315                 RTE_PTYPE_L3_IPV4,
1316                 RTE_PTYPE_L3_IPV4_EXT,
1317                 RTE_PTYPE_L3_IPV6_EXT,
1318                 RTE_PTYPE_L3_IPV6,
1319                 RTE_PTYPE_L4_FRAG,
1320                 RTE_PTYPE_L4_UDP,
1321                 RTE_PTYPE_L4_TCP,
1322                 RTE_PTYPE_L4_SCTP,
1323         };
1324
1325         return ptypes;
1326 }
1327
1328 static int
1329 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1330                   struct rte_eth_fc_conf *fc_conf)
1331 {
1332         fc_conf->mode = RTE_FC_NONE;
1333         return 0;
1334 }
1335
1336 static int
1337 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1338                   struct rte_eth_fc_conf *fc_conf)
1339 {
1340         if (fc_conf->mode != RTE_FC_NONE)
1341                 return -ENOTSUP;
1342         return 0;
1343 }
1344
1345 static const struct eth_dev_ops ops = {
1346         .dev_start              = tap_dev_start,
1347         .dev_stop               = tap_dev_stop,
1348         .dev_close              = tap_dev_close,
1349         .dev_configure          = tap_dev_configure,
1350         .dev_infos_get          = tap_dev_info,
1351         .rx_queue_setup         = tap_rx_queue_setup,
1352         .tx_queue_setup         = tap_tx_queue_setup,
1353         .rx_queue_release       = tap_rx_queue_release,
1354         .tx_queue_release       = tap_tx_queue_release,
1355         .flow_ctrl_get          = tap_flow_ctrl_get,
1356         .flow_ctrl_set          = tap_flow_ctrl_set,
1357         .link_update            = tap_link_update,
1358         .dev_set_link_up        = tap_link_set_up,
1359         .dev_set_link_down      = tap_link_set_down,
1360         .promiscuous_enable     = tap_promisc_enable,
1361         .promiscuous_disable    = tap_promisc_disable,
1362         .allmulticast_enable    = tap_allmulti_enable,
1363         .allmulticast_disable   = tap_allmulti_disable,
1364         .mac_addr_set           = tap_mac_set,
1365         .mtu_set                = tap_mtu_set,
1366         .set_mc_addr_list       = tap_set_mc_addr_list,
1367         .stats_get              = tap_stats_get,
1368         .stats_reset            = tap_stats_reset,
1369         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1370         .filter_ctrl            = tap_dev_filter_ctrl,
1371 };
1372
1373 static int
1374 eth_dev_tap_create(struct rte_vdev_device *vdev, char *tap_name,
1375                    char *remote_iface, struct ether_addr *mac_addr)
1376 {
1377         int numa_node = rte_socket_id();
1378         struct rte_eth_dev *dev;
1379         struct pmd_internals *pmd;
1380         struct rte_eth_dev_data *data;
1381         struct ifreq ifr;
1382         int i;
1383
1384         TAP_LOG(DEBUG, "%s device on numa %u",
1385                         tuntap_name, rte_socket_id());
1386
1387         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1388         if (!dev) {
1389                 TAP_LOG(ERR, "%s Unable to allocate device struct",
1390                                 tuntap_name);
1391                 goto error_exit_nodev;
1392         }
1393
1394         pmd = dev->data->dev_private;
1395         pmd->dev = dev;
1396         snprintf(pmd->name, sizeof(pmd->name), "%s", tap_name);
1397
1398         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1399         if (pmd->ioctl_sock == -1) {
1400                 TAP_LOG(ERR,
1401                         "%s Unable to get a socket for management: %s",
1402                         tuntap_name, strerror(errno));
1403                 goto error_exit;
1404         }
1405
1406         /* Setup some default values */
1407         data = dev->data;
1408         data->dev_private = pmd;
1409         data->dev_flags = RTE_ETH_DEV_INTR_LSC;
1410         data->numa_node = numa_node;
1411
1412         data->dev_link = pmd_link;
1413         data->mac_addrs = &pmd->eth_addr;
1414         /* Set the number of RX and TX queues */
1415         data->nb_rx_queues = 0;
1416         data->nb_tx_queues = 0;
1417
1418         dev->dev_ops = &ops;
1419         dev->rx_pkt_burst = pmd_rx_burst;
1420         dev->tx_pkt_burst = pmd_tx_burst;
1421
1422         pmd->intr_handle.type = RTE_INTR_HANDLE_EXT;
1423         pmd->intr_handle.fd = -1;
1424         dev->intr_handle = &pmd->intr_handle;
1425
1426         /* Presetup the fds to -1 as being not valid */
1427         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1428                 pmd->rxq[i].fd = -1;
1429                 pmd->txq[i].fd = -1;
1430         }
1431
1432         if (tap_type) {
1433                 if (is_zero_ether_addr(mac_addr))
1434                         eth_random_addr((uint8_t *)&pmd->eth_addr);
1435                 else
1436                         rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1437         }
1438
1439         /* Immediately create the netdevice (this will create the 1st queue). */
1440         /* rx queue */
1441         if (tap_setup_queue(dev, pmd, 0, 1) == -1)
1442                 goto error_exit;
1443         /* tx queue */
1444         if (tap_setup_queue(dev, pmd, 0, 0) == -1)
1445                 goto error_exit;
1446
1447         ifr.ifr_mtu = dev->data->mtu;
1448         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1449                 goto error_exit;
1450
1451         if (tap_type) {
1452                 memset(&ifr, 0, sizeof(struct ifreq));
1453                 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1454                 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1455                                 ETHER_ADDR_LEN);
1456                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1457                         goto error_exit;
1458         }
1459
1460         /*
1461          * Set up everything related to rte_flow:
1462          * - netlink socket
1463          * - tap / remote if_index
1464          * - mandatory QDISCs
1465          * - rte_flow actual/implicit lists
1466          * - implicit rules
1467          */
1468         pmd->nlsk_fd = tap_nl_init(0);
1469         if (pmd->nlsk_fd == -1) {
1470                 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
1471                         pmd->name);
1472                 goto disable_rte_flow;
1473         }
1474         pmd->if_index = if_nametoindex(pmd->name);
1475         if (!pmd->if_index) {
1476                 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
1477                 goto disable_rte_flow;
1478         }
1479         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
1480                 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
1481                         pmd->name);
1482                 goto disable_rte_flow;
1483         }
1484         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
1485                 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1486                         pmd->name);
1487                 goto disable_rte_flow;
1488         }
1489         LIST_INIT(&pmd->flows);
1490
1491         if (strlen(remote_iface)) {
1492                 pmd->remote_if_index = if_nametoindex(remote_iface);
1493                 if (!pmd->remote_if_index) {
1494                         TAP_LOG(ERR, "%s: failed to get %s if_index.",
1495                                 pmd->name, remote_iface);
1496                         goto error_remote;
1497                 }
1498                 snprintf(pmd->remote_iface, RTE_ETH_NAME_MAX_LEN,
1499                          "%s", remote_iface);
1500
1501                 /* Save state of remote device */
1502                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
1503
1504                 /* Replicate remote MAC address */
1505                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
1506                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1507                                 pmd->name, pmd->remote_iface);
1508                         goto error_remote;
1509                 }
1510                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
1511                            ETHER_ADDR_LEN);
1512                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
1513                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
1514                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
1515                                 pmd->name, remote_iface);
1516                         goto error_remote;
1517                 }
1518
1519                 /*
1520                  * Flush usually returns negative value because it tries to
1521                  * delete every QDISC (and on a running device, one QDISC at
1522                  * least is needed). Ignore negative return value.
1523                  */
1524                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
1525                 if (qdisc_create_ingress(pmd->nlsk_fd,
1526                                          pmd->remote_if_index) < 0) {
1527                         TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
1528                                 pmd->remote_iface);
1529                         goto error_remote;
1530                 }
1531                 LIST_INIT(&pmd->implicit_flows);
1532                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
1533                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
1534                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
1535                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
1536                         TAP_LOG(ERR,
1537                                 "%s: failed to create implicit rules.",
1538                                 pmd->name);
1539                         goto error_remote;
1540                 }
1541         }
1542
1543         return 0;
1544
1545 disable_rte_flow:
1546         TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
1547                 strerror(errno), errno);
1548         if (strlen(remote_iface)) {
1549                 TAP_LOG(ERR, "Remote feature requires flow support.");
1550                 goto error_exit;
1551         }
1552         return 0;
1553
1554 error_remote:
1555         TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
1556                 strerror(errno), errno);
1557         tap_flow_implicit_flush(pmd, NULL);
1558
1559 error_exit:
1560         if (pmd->ioctl_sock > 0)
1561                 close(pmd->ioctl_sock);
1562         rte_eth_dev_release_port(dev);
1563
1564 error_exit_nodev:
1565         TAP_LOG(ERR, "%s Unable to initialize %s",
1566                 tuntap_name, rte_vdev_device_name(vdev));
1567
1568         return -EINVAL;
1569 }
1570
1571 static int
1572 set_interface_name(const char *key __rte_unused,
1573                    const char *value,
1574                    void *extra_args)
1575 {
1576         char *name = (char *)extra_args;
1577
1578         if (value)
1579                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN - 1);
1580         else
1581                 snprintf(name, RTE_ETH_NAME_MAX_LEN - 1, "%s%d",
1582                          DEFAULT_TAP_NAME, (tap_unit - 1));
1583
1584         return 0;
1585 }
1586
1587 static int
1588 set_remote_iface(const char *key __rte_unused,
1589                  const char *value,
1590                  void *extra_args)
1591 {
1592         char *name = (char *)extra_args;
1593
1594         if (value)
1595                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
1596
1597         return 0;
1598 }
1599
1600 static int parse_user_mac(struct ether_addr *user_mac,
1601                 const char *value)
1602 {
1603         unsigned int index = 0;
1604         char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
1605
1606         if (user_mac == NULL || value == NULL)
1607                 return 0;
1608
1609         strlcpy(mac_temp, value, sizeof(mac_temp));
1610         mac_byte = strtok(mac_temp, ":");
1611
1612         while ((mac_byte != NULL) &&
1613                         (strlen(mac_byte) <= 2) &&
1614                         (strlen(mac_byte) == strspn(mac_byte,
1615                                         ETH_TAP_CMP_MAC_FMT))) {
1616                 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
1617                 mac_byte = strtok(NULL, ":");
1618         }
1619
1620         return index;
1621 }
1622
1623 static int
1624 set_mac_type(const char *key __rte_unused,
1625              const char *value,
1626              void *extra_args)
1627 {
1628         struct ether_addr *user_mac = extra_args;
1629
1630         if (!value)
1631                 return 0;
1632
1633         if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
1634                 static int iface_idx;
1635
1636                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
1637                 memcpy((char *)user_mac->addr_bytes, "\0dtap", ETHER_ADDR_LEN);
1638                 user_mac->addr_bytes[ETHER_ADDR_LEN - 1] = iface_idx++ + '0';
1639                 goto success;
1640         }
1641
1642         if (parse_user_mac(user_mac, value) != 6)
1643                 goto error;
1644 success:
1645         TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
1646         return 0;
1647
1648 error:
1649         TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
1650                 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
1651         return -1;
1652 }
1653
1654 /*
1655  * Open a TUN interface device. TUN PMD
1656  * 1) sets tap_type as false
1657  * 2) intakes iface as argument.
1658  * 3) as interface is virtual set speed to 10G
1659  */
1660 static int
1661 rte_pmd_tun_probe(struct rte_vdev_device *dev)
1662 {
1663         const char *name, *params;
1664         int ret;
1665         struct rte_kvargs *kvlist = NULL;
1666         char tun_name[RTE_ETH_NAME_MAX_LEN];
1667         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1668
1669         tap_type = 0;
1670         strcpy(tuntap_name, "TUN");
1671
1672         name = rte_vdev_device_name(dev);
1673         params = rte_vdev_device_args(dev);
1674         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1675
1676         if (params && (params[0] != '\0')) {
1677                 TAP_LOG(DEBUG, "parameters (%s)", params);
1678
1679                 kvlist = rte_kvargs_parse(params, valid_arguments);
1680                 if (kvlist) {
1681                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1682                                 ret = rte_kvargs_process(kvlist,
1683                                         ETH_TAP_IFACE_ARG,
1684                                         &set_interface_name,
1685                                         tun_name);
1686
1687                                 if (ret == -1)
1688                                         goto leave;
1689                         }
1690                 }
1691         }
1692         pmd_link.link_speed = ETH_SPEED_NUM_10G;
1693
1694         TAP_LOG(NOTICE, "Initializing pmd_tun for %s as %s",
1695                 name, tun_name);
1696
1697         ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0);
1698
1699 leave:
1700         if (ret == -1) {
1701                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
1702                         name, tun_name);
1703                 tun_unit--; /* Restore the unit number */
1704         }
1705         rte_kvargs_free(kvlist);
1706
1707         return ret;
1708 }
1709
1710 /* Open a TAP interface device.
1711  */
1712 static int
1713 rte_pmd_tap_probe(struct rte_vdev_device *dev)
1714 {
1715         const char *name, *params;
1716         int ret;
1717         struct rte_kvargs *kvlist = NULL;
1718         int speed;
1719         char tap_name[RTE_ETH_NAME_MAX_LEN];
1720         char remote_iface[RTE_ETH_NAME_MAX_LEN];
1721         struct ether_addr user_mac = { .addr_bytes = {0} };
1722         struct rte_eth_dev *eth_dev;
1723
1724         tap_type = 1;
1725         strcpy(tuntap_name, "TAP");
1726
1727         name = rte_vdev_device_name(dev);
1728         params = rte_vdev_device_args(dev);
1729
1730         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
1731             strlen(params) == 0) {
1732                 eth_dev = rte_eth_dev_attach_secondary(name);
1733                 if (!eth_dev) {
1734                         TAP_LOG(ERR, "Failed to probe %s", name);
1735                         return -1;
1736                 }
1737                 /* TODO: request info from primary to set up Rx and Tx */
1738                 eth_dev->dev_ops = &ops;
1739                 return 0;
1740         }
1741
1742         speed = ETH_SPEED_NUM_10G;
1743         snprintf(tap_name, sizeof(tap_name), "%s%d",
1744                  DEFAULT_TAP_NAME, tap_unit++);
1745         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
1746
1747         if (params && (params[0] != '\0')) {
1748                 TAP_LOG(DEBUG, "parameters (%s)", params);
1749
1750                 kvlist = rte_kvargs_parse(params, valid_arguments);
1751                 if (kvlist) {
1752                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
1753                                 ret = rte_kvargs_process(kvlist,
1754                                                          ETH_TAP_IFACE_ARG,
1755                                                          &set_interface_name,
1756                                                          tap_name);
1757                                 if (ret == -1)
1758                                         goto leave;
1759                         }
1760
1761                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
1762                                 ret = rte_kvargs_process(kvlist,
1763                                                          ETH_TAP_REMOTE_ARG,
1764                                                          &set_remote_iface,
1765                                                          remote_iface);
1766                                 if (ret == -1)
1767                                         goto leave;
1768                         }
1769
1770                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
1771                                 ret = rte_kvargs_process(kvlist,
1772                                                          ETH_TAP_MAC_ARG,
1773                                                          &set_mac_type,
1774                                                          &user_mac);
1775                                 if (ret == -1)
1776                                         goto leave;
1777                         }
1778                 }
1779         }
1780         pmd_link.link_speed = speed;
1781
1782         TAP_LOG(NOTICE, "Initializing pmd_tap for %s as %s",
1783                 name, tap_name);
1784
1785         ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac);
1786
1787 leave:
1788         if (ret == -1) {
1789                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
1790                         name, tap_name);
1791                 tap_unit--;             /* Restore the unit number */
1792         }
1793         rte_kvargs_free(kvlist);
1794
1795         return ret;
1796 }
1797
1798 /* detach a TUNTAP device.
1799  */
1800 static int
1801 rte_pmd_tap_remove(struct rte_vdev_device *dev)
1802 {
1803         struct rte_eth_dev *eth_dev = NULL;
1804         struct pmd_internals *internals;
1805         int i;
1806
1807         TAP_LOG(DEBUG, "Closing TUN/TAP Ethernet device on numa %u",
1808                 rte_socket_id());
1809
1810         /* find the ethdev entry */
1811         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
1812         if (!eth_dev)
1813                 return 0;
1814
1815         internals = eth_dev->data->dev_private;
1816         if (internals->nlsk_fd) {
1817                 tap_flow_flush(eth_dev, NULL);
1818                 tap_flow_implicit_flush(internals, NULL);
1819                 tap_nl_final(internals->nlsk_fd);
1820         }
1821         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1822                 if (internals->rxq[i].fd != -1) {
1823                         close(internals->rxq[i].fd);
1824                         internals->rxq[i].fd = -1;
1825                 }
1826                 if (internals->txq[i].fd != -1) {
1827                         close(internals->txq[i].fd);
1828                         internals->txq[i].fd = -1;
1829                 }
1830         }
1831
1832         close(internals->ioctl_sock);
1833         rte_free(eth_dev->data->dev_private);
1834
1835         rte_eth_dev_release_port(eth_dev);
1836
1837         return 0;
1838 }
1839
1840 static struct rte_vdev_driver pmd_tun_drv = {
1841         .probe = rte_pmd_tun_probe,
1842         .remove = rte_pmd_tap_remove,
1843 };
1844
1845 static struct rte_vdev_driver pmd_tap_drv = {
1846         .probe = rte_pmd_tap_probe,
1847         .remove = rte_pmd_tap_remove,
1848 };
1849
1850 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
1851 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
1852 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
1853 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
1854                               ETH_TAP_IFACE_ARG "=<string> ");
1855 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
1856                               ETH_TAP_IFACE_ARG "=<string> "
1857                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
1858                               ETH_TAP_REMOTE_ARG "=<string>");
1859 int tap_logtype;
1860
1861 RTE_INIT(tap_init_log);
1862 static void
1863 tap_init_log(void)
1864 {
1865         tap_logtype = rte_log_register("pmd.net.tap");
1866         if (tap_logtype >= 0)
1867                 rte_log_set_level(tap_logtype, RTE_LOG_NOTICE);
1868 }