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 <ethdev_driver.h>
  11 #include <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 #include <rte_ethdev.h>
  20 #include <rte_errno.h>
  21 #include <rte_cycles.h>
  22
  23 #include <sys/types.h>
  24 #include <sys/stat.h>
  25 #include <sys/socket.h>
  26 #include <sys/ioctl.h>
  27 #include <sys/utsname.h>
  28 #include <sys/mman.h>
  29 #include <errno.h>
  30 #include <signal.h>
  31 #include <stdbool.h>
  32 #include <stdint.h>
  33 #include <sys/uio.h>
  34 #include <unistd.h>
  35 #include <arpa/inet.h>
  36 #include <net/if.h>
  37 #include <linux/if_tun.h>
  38 #include <linux/if_ether.h>
  39 #include <fcntl.h>
  40 #include <ctype.h>
  41
  42 #include <tap_rss.h>
  43 #include <rte_eth_tap.h>
  44 #include <tap_flow.h>
  45 #include <tap_netlink.h>
  46 #include <tap_tcmsgs.h>
  47
  48 /* Linux based path to the TUN device */
  49 #define TUN_TAP_DEV_PATH        "/dev/net/tun"
  50 #define DEFAULT_TAP_NAME        "dtap"
  51 #define DEFAULT_TUN_NAME        "dtun"
  52
  53 #define ETH_TAP_IFACE_ARG       "iface"
  54 #define ETH_TAP_REMOTE_ARG      "remote"
  55 #define ETH_TAP_MAC_ARG         "mac"
  56 #define ETH_TAP_MAC_FIXED       "fixed"
  57
  58 #define ETH_TAP_USR_MAC_FMT     "xx:xx:xx:xx:xx:xx"
  59 #define ETH_TAP_CMP_MAC_FMT     "0123456789ABCDEFabcdef"
  60 #define ETH_TAP_MAC_ARG_FMT     ETH_TAP_MAC_FIXED "|" ETH_TAP_USR_MAC_FMT
  61
  62 #define TAP_GSO_MBUFS_PER_CORE  128
  63 #define TAP_GSO_MBUF_SEG_SIZE   128
  64 #define TAP_GSO_MBUF_CACHE_SIZE 4
  65 #define TAP_GSO_MBUFS_NUM \
  66         (TAP_GSO_MBUFS_PER_CORE * TAP_GSO_MBUF_CACHE_SIZE)
  67
  68 /* IPC key for queue fds sync */
  69 #define TAP_MP_KEY "tap_mp_sync_queues"
  70
  71 #define TAP_IOV_DEFAULT_MAX 1024
  72
  73 #define TAP_RX_OFFLOAD (RTE_ETH_RX_OFFLOAD_SCATTER |    \
  74                         RTE_ETH_RX_OFFLOAD_IPV4_CKSUM | \
  75                         RTE_ETH_RX_OFFLOAD_UDP_CKSUM |  \
  76                         RTE_ETH_RX_OFFLOAD_TCP_CKSUM)
  77
  78 #define TAP_TX_OFFLOAD (RTE_ETH_TX_OFFLOAD_MULTI_SEGS | \
  79                         RTE_ETH_TX_OFFLOAD_IPV4_CKSUM | \
  80                         RTE_ETH_TX_OFFLOAD_UDP_CKSUM |  \
  81                         RTE_ETH_TX_OFFLOAD_TCP_CKSUM |  \
  82                         RTE_ETH_TX_OFFLOAD_TCP_TSO)
  83
  84 static int tap_devices_count;
  85
  86 static const char *tuntap_types[ETH_TUNTAP_TYPE_MAX] = {
  87         "UNKNOWN", "TUN", "TAP"
  88 };
  89
  90 static const char *valid_arguments[] = {
  91         ETH_TAP_IFACE_ARG,
  92         ETH_TAP_REMOTE_ARG,
  93         ETH_TAP_MAC_ARG,
  94         NULL
  95 };
  96
  97 static volatile uint32_t tap_trigger;   /* Rx trigger */
  98
  99 static struct rte_eth_link pmd_link = {
 100         .link_speed = RTE_ETH_SPEED_NUM_10G,
 101         .link_duplex = RTE_ETH_LINK_FULL_DUPLEX,
 102         .link_status = RTE_ETH_LINK_DOWN,
 103         .link_autoneg = RTE_ETH_LINK_FIXED,
 104 };
 105
 106 static void
 107 tap_trigger_cb(int sig __rte_unused)
 108 {
 109         /* Valid trigger values are nonzero */
 110         tap_trigger = (tap_trigger + 1) | 0x80000000;
 111 }
 112
 113 /* Specifies on what netdevices the ioctl should be applied */
 114 enum ioctl_mode {
 115         LOCAL_AND_REMOTE,
 116         LOCAL_ONLY,
 117         REMOTE_ONLY,
 118 };
 119
 120 /* Message header to synchronize queues via IPC */
 121 struct ipc_queues {
 122         char port_name[RTE_DEV_NAME_MAX_LEN];
 123         int rxq_count;
 124         int txq_count;
 125         /*
 126          * The file descriptors are in the dedicated part
 127          * of the Unix message to be translated by the kernel.
 128          */
 129 };
 130
 131 static int tap_intr_handle_set(struct rte_eth_dev *dev, int set);
 132
 133 /**
 134  * Tun/Tap allocation routine
 135  *
 136  * @param[in] pmd
 137  *   Pointer to private structure.
 138  *
 139  * @param[in] is_keepalive
 140  *   Keepalive flag
 141  *
 142  * @return
 143  *   -1 on failure, fd on success
 144  */
 145 static int
 146 tun_alloc(struct pmd_internals *pmd, int is_keepalive)
 147 {
 148         struct ifreq ifr;
 149 #ifdef IFF_MULTI_QUEUE
 150         unsigned int features;
 151 #endif
 152         int fd, signo, flags;
 153
 154         memset(&ifr, 0, sizeof(struct ifreq));
 155
 156         /*
 157          * Do not set IFF_NO_PI as packet information header will be needed
 158          * to check if a received packet has been truncated.
 159          */
 160         ifr.ifr_flags = (pmd->type == ETH_TUNTAP_TYPE_TAP) ?
 161                 IFF_TAP : IFF_TUN | IFF_POINTOPOINT;
 162         strlcpy(ifr.ifr_name, pmd->name, IFNAMSIZ);
 163
 164         fd = open(TUN_TAP_DEV_PATH, O_RDWR);
 165         if (fd < 0) {
 166                 TAP_LOG(ERR, "Unable to open %s interface", TUN_TAP_DEV_PATH);
 167                 goto error;
 168         }
 169
 170 #ifdef IFF_MULTI_QUEUE
 171         /* Grab the TUN features to verify we can work multi-queue */
 172         if (ioctl(fd, TUNGETFEATURES, &features) < 0) {
 173                 TAP_LOG(ERR, "unable to get TUN/TAP features");
 174                 goto error;
 175         }
 176         TAP_LOG(DEBUG, "%s Features %08x", TUN_TAP_DEV_PATH, features);
 177
 178         if (features & IFF_MULTI_QUEUE) {
 179                 TAP_LOG(DEBUG, "  Multi-queue support for %d queues",
 180                         RTE_PMD_TAP_MAX_QUEUES);
 181                 ifr.ifr_flags |= IFF_MULTI_QUEUE;
 182         } else
 183 #endif
 184         {
 185                 ifr.ifr_flags |= IFF_ONE_QUEUE;
 186                 TAP_LOG(DEBUG, "  Single queue only support");
 187         }
 188
 189         /* Set the TUN/TAP configuration and set the name if needed */
 190         if (ioctl(fd, TUNSETIFF, (void *)&ifr) < 0) {
 191                 TAP_LOG(WARNING, "Unable to set TUNSETIFF for %s: %s",
 192                         ifr.ifr_name, strerror(errno));
 193                 goto error;
 194         }
 195
 196         /*
 197          * Name passed to kernel might be wildcard like dtun%d
 198          * and need to find the resulting device.
 199          */
 200         TAP_LOG(DEBUG, "Device name is '%s'", ifr.ifr_name);
 201         strlcpy(pmd->name, ifr.ifr_name, RTE_ETH_NAME_MAX_LEN);
 202
 203         if (is_keepalive) {
 204                 /*
 205                  * Detach the TUN/TAP keep-alive queue
 206                  * to avoid traffic through it
 207                  */
 208                 ifr.ifr_flags = IFF_DETACH_QUEUE;
 209                 if (ioctl(fd, TUNSETQUEUE, (void *)&ifr) < 0) {
 210                         TAP_LOG(WARNING,
 211                                 "Unable to detach keep-alive queue for %s: %s",
 212                                 ifr.ifr_name, strerror(errno));
 213                         goto error;
 214                 }
 215         }
 216
 217         flags = fcntl(fd, F_GETFL);
 218         if (flags == -1) {
 219                 TAP_LOG(WARNING,
 220                         "Unable to get %s current flags\n",
 221                         ifr.ifr_name);
 222                 goto error;
 223         }
 224
 225         /* Always set the file descriptor to non-blocking */
 226         flags |= O_NONBLOCK;
 227         if (fcntl(fd, F_SETFL, flags) < 0) {
 228                 TAP_LOG(WARNING,
 229                         "Unable to set %s to nonblocking: %s",
 230                         ifr.ifr_name, strerror(errno));
 231                 goto error;
 232         }
 233
 234         /* Find a free realtime signal */
 235         for (signo = SIGRTMIN + 1; signo < SIGRTMAX; signo++) {
 236                 struct sigaction sa;
 237
 238                 if (sigaction(signo, NULL, &sa) == -1) {
 239                         TAP_LOG(WARNING,
 240                                 "Unable to get current rt-signal %d handler",
 241                                 signo);
 242                         goto error;
 243                 }
 244
 245                 /* Already have the handler we want on this signal  */
 246                 if (sa.sa_handler == tap_trigger_cb)
 247                         break;
 248
 249                 /* Is handler in use by application */
 250                 if (sa.sa_handler != SIG_DFL) {
 251                         TAP_LOG(DEBUG,
 252                                 "Skipping used rt-signal %d", signo);
 253                         continue;
 254                 }
 255
 256                 sa = (struct sigaction) {
 257                         .sa_flags = SA_RESTART,
 258                         .sa_handler = tap_trigger_cb,
 259                 };
 260
 261                 if (sigaction(signo, &sa, NULL) == -1) {
 262                         TAP_LOG(WARNING,
 263                                 "Unable to set rt-signal %d handler\n", signo);
 264                         goto error;
 265                 }
 266
 267                 /* Found a good signal to use */
 268                 TAP_LOG(DEBUG,
 269                         "Using rt-signal %d", signo);
 270                 break;
 271         }
 272
 273         if (signo == SIGRTMAX) {
 274                 TAP_LOG(WARNING, "All rt-signals are in use\n");
 275
 276                 /* Disable trigger globally in case of error */
 277                 tap_trigger = 0;
 278                 TAP_LOG(NOTICE, "No Rx trigger signal available\n");
 279         } else {
 280                 /* Enable signal on file descriptor */
 281                 if (fcntl(fd, F_SETSIG, signo) < 0) {
 282                         TAP_LOG(WARNING, "Unable to set signo %d for fd %d: %s",
 283                                 signo, fd, strerror(errno));
 284                         goto error;
 285                 }
 286                 if (fcntl(fd, F_SETFL, flags | O_ASYNC) < 0) {
 287                         TAP_LOG(WARNING, "Unable to set fcntl flags: %s",
 288                                 strerror(errno));
 289                         goto error;
 290                 }
 291
 292                 if (fcntl(fd, F_SETOWN, getpid()) < 0) {
 293                         TAP_LOG(WARNING, "Unable to set fcntl owner: %s",
 294                                 strerror(errno));
 295                         goto error;
 296                 }
 297         }
 298         return fd;
 299
 300 error:
 301         if (fd >= 0)
 302                 close(fd);
 303         return -1;
 304 }
 305
 306 static void
 307 tap_verify_csum(struct rte_mbuf *mbuf)
 308 {
 309         uint32_t l2 = mbuf->packet_type & RTE_PTYPE_L2_MASK;
 310         uint32_t l3 = mbuf->packet_type & RTE_PTYPE_L3_MASK;
 311         uint32_t l4 = mbuf->packet_type & RTE_PTYPE_L4_MASK;
 312         unsigned int l2_len = sizeof(struct rte_ether_hdr);
 313         unsigned int l3_len;
 314         uint16_t cksum = 0;
 315         void *l3_hdr;
 316         void *l4_hdr;
 317         struct rte_udp_hdr *udp_hdr;
 318
 319         if (l2 == RTE_PTYPE_L2_ETHER_VLAN)
 320                 l2_len += 4;
 321         else if (l2 == RTE_PTYPE_L2_ETHER_QINQ)
 322                 l2_len += 8;
 323         /* Don't verify checksum for packets with discontinuous L2 header */
 324         if (unlikely(l2_len + sizeof(struct rte_ipv4_hdr) >
 325                      rte_pktmbuf_data_len(mbuf)))
 326                 return;
 327         l3_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len);
 328         if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 329                 struct rte_ipv4_hdr *iph = l3_hdr;
 330
 331                 l3_len = rte_ipv4_hdr_len(iph);
 332                 if (unlikely(l2_len + l3_len > rte_pktmbuf_data_len(mbuf)))
 333                         return;
 334                 /* check that the total length reported by header is not
 335                  * greater than the total received size
 336                  */
 337                 if (l2_len + rte_be_to_cpu_16(iph->total_length) >
 338                                 rte_pktmbuf_data_len(mbuf))
 339                         return;
 340
 341                 cksum = ~rte_raw_cksum(iph, l3_len);
 342                 mbuf->ol_flags |= cksum ?
 343                         RTE_MBUF_F_RX_IP_CKSUM_BAD :
 344                         RTE_MBUF_F_RX_IP_CKSUM_GOOD;
 345         } else if (l3 == RTE_PTYPE_L3_IPV6) {
 346                 struct rte_ipv6_hdr *iph = l3_hdr;
 347
 348                 l3_len = sizeof(struct rte_ipv6_hdr);
 349                 /* check that the total length reported by header is not
 350                  * greater than the total received size
 351                  */
 352                 if (l2_len + l3_len + rte_be_to_cpu_16(iph->payload_len) >
 353                                 rte_pktmbuf_data_len(mbuf))
 354                         return;
 355         } else {
 356                 /* - RTE_PTYPE_L3_IPV4_EXT_UNKNOWN cannot happen because
 357                  *   mbuf->packet_type is filled by rte_net_get_ptype() which
 358                  *   never returns this value.
 359                  * - IPv6 extensions are not supported.
 360                  */
 361                 return;
 362         }
 363         if (l4 == RTE_PTYPE_L4_UDP || l4 == RTE_PTYPE_L4_TCP) {
 364                 int cksum_ok;
 365
 366                 l4_hdr = rte_pktmbuf_mtod_offset(mbuf, void *, l2_len + l3_len);
 367                 /* Don't verify checksum for multi-segment packets. */
 368                 if (mbuf->nb_segs > 1)
 369                         return;
 370                 if (l3 == RTE_PTYPE_L3_IPV4 || l3 == RTE_PTYPE_L3_IPV4_EXT) {
 371                         if (l4 == RTE_PTYPE_L4_UDP) {
 372                                 udp_hdr = (struct rte_udp_hdr *)l4_hdr;
 373                                 if (udp_hdr->dgram_cksum == 0) {
 374                                         /*
 375                                          * For IPv4, a zero UDP checksum
 376                                          * indicates that the sender did not
 377                                          * generate one [RFC 768].
 378                                          */
 379                                         mbuf->ol_flags |= RTE_MBUF_F_RX_L4_CKSUM_NONE;
 380                                         return;
 381                                 }
 382                         }
 383                         cksum_ok = !rte_ipv4_udptcp_cksum_verify(l3_hdr,
 384                                                                  l4_hdr);
 385                 } else { /* l3 == RTE_PTYPE_L3_IPV6, checked above */
 386                         cksum_ok = !rte_ipv6_udptcp_cksum_verify(l3_hdr,
 387                                                                  l4_hdr);
 388                 }
 389                 mbuf->ol_flags |= cksum_ok ?
 390                         RTE_MBUF_F_RX_L4_CKSUM_GOOD : RTE_MBUF_F_RX_L4_CKSUM_BAD;
 391         }
 392 }
 393
 394 static void
 395 tap_rxq_pool_free(struct rte_mbuf *pool)
 396 {
 397         struct rte_mbuf *mbuf = pool;
 398         uint16_t nb_segs = 1;
 399
 400         if (mbuf == NULL)
 401                 return;
 402
 403         while (mbuf->next) {
 404                 mbuf = mbuf->next;
 405                 nb_segs++;
 406         }
 407         pool->nb_segs = nb_segs;
 408         rte_pktmbuf_free(pool);
 409 }
 410
 411 /* Callback to handle the rx burst of packets to the correct interface and
 412  * file descriptor(s) in a multi-queue setup.
 413  */
 414 static uint16_t
 415 pmd_rx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 416 {
 417         struct rx_queue *rxq = queue;
 418         struct pmd_process_private *process_private;
 419         uint16_t num_rx;
 420         unsigned long num_rx_bytes = 0;
 421         uint32_t trigger = tap_trigger;
 422
 423         if (trigger == rxq->trigger_seen)
 424                 return 0;
 425
 426         process_private = rte_eth_devices[rxq->in_port].process_private;
 427         for (num_rx = 0; num_rx < nb_pkts; ) {
 428                 struct rte_mbuf *mbuf = rxq->pool;
 429                 struct rte_mbuf *seg = NULL;
 430                 struct rte_mbuf *new_tail = NULL;
 431                 uint16_t data_off = rte_pktmbuf_headroom(mbuf);
 432                 int len;
 433
 434                 len = readv(process_private->rxq_fds[rxq->queue_id],
 435                         *rxq->iovecs,
 436                         1 + (rxq->rxmode->offloads & RTE_ETH_RX_OFFLOAD_SCATTER ?
 437                              rxq->nb_rx_desc : 1));
 438                 if (len < (int)sizeof(struct tun_pi))
 439                         break;
 440
 441                 /* Packet couldn't fit in the provided mbuf */
 442                 if (unlikely(rxq->pi.flags & TUN_PKT_STRIP)) {
 443                         rxq->stats.ierrors++;
 444                         continue;
 445                 }
 446
 447                 len -= sizeof(struct tun_pi);
 448
 449                 mbuf->pkt_len = len;
 450                 mbuf->port = rxq->in_port;
 451                 while (1) {
 452                         struct rte_mbuf *buf = rte_pktmbuf_alloc(rxq->mp);
 453
 454                         if (unlikely(!buf)) {
 455                                 rxq->stats.rx_nombuf++;
 456                                 /* No new buf has been allocated: do nothing */
 457                                 if (!new_tail || !seg)
 458                                         goto end;
 459
 460                                 seg->next = NULL;
 461                                 tap_rxq_pool_free(mbuf);
 462
 463                                 goto end;
 464                         }
 465                         seg = seg ? seg->next : mbuf;
 466                         if (rxq->pool == mbuf)
 467                                 rxq->pool = buf;
 468                         if (new_tail)
 469                                 new_tail->next = buf;
 470                         new_tail = buf;
 471                         new_tail->next = seg->next;
 472
 473                         /* iovecs[0] is reserved for packet info (pi) */
 474                         (*rxq->iovecs)[mbuf->nb_segs].iov_len =
 475                                 buf->buf_len - data_off;
 476                         (*rxq->iovecs)[mbuf->nb_segs].iov_base =
 477                                 (char *)buf->buf_addr + data_off;
 478
 479                         seg->data_len = RTE_MIN(seg->buf_len - data_off, len);
 480                         seg->data_off = data_off;
 481
 482                         len -= seg->data_len;
 483                         if (len <= 0)
 484                                 break;
 485                         mbuf->nb_segs++;
 486                         /* First segment has headroom, not the others */
 487                         data_off = 0;
 488                 }
 489                 seg->next = NULL;
 490                 mbuf->packet_type = rte_net_get_ptype(mbuf, NULL,
 491                                                       RTE_PTYPE_ALL_MASK);
 492                 if (rxq->rxmode->offloads & RTE_ETH_RX_OFFLOAD_CHECKSUM)
 493                         tap_verify_csum(mbuf);
 494
 495                 /* account for the receive frame */
 496                 bufs[num_rx++] = mbuf;
 497                 num_rx_bytes += mbuf->pkt_len;
 498         }
 499 end:
 500         rxq->stats.ipackets += num_rx;
 501         rxq->stats.ibytes += num_rx_bytes;
 502
 503         if (trigger && num_rx < nb_pkts)
 504                 rxq->trigger_seen = trigger;
 505
 506         return num_rx;
 507 }
 508
 509 /* Finalize l4 checksum calculation */
 510 static void
 511 tap_tx_l4_cksum(uint16_t *l4_cksum, uint16_t l4_phdr_cksum,
 512                 uint32_t l4_raw_cksum)
 513 {
 514         if (l4_cksum) {
 515                 uint32_t cksum;
 516
 517                 cksum = __rte_raw_cksum_reduce(l4_raw_cksum);
 518                 cksum += l4_phdr_cksum;
 519
 520                 cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 521                 cksum = (~cksum) & 0xffff;
 522                 if (cksum == 0)
 523                         cksum = 0xffff;
 524                 *l4_cksum = cksum;
 525         }
 526 }
 527
 528 /* Accumulate L4 raw checksums */
 529 static void
 530 tap_tx_l4_add_rcksum(char *l4_data, unsigned int l4_len, uint16_t *l4_cksum,
 531                         uint32_t *l4_raw_cksum)
 532 {
 533         if (l4_cksum == NULL)
 534                 return;
 535
 536         *l4_raw_cksum = __rte_raw_cksum(l4_data, l4_len, *l4_raw_cksum);
 537 }
 538
 539 /* L3 and L4 pseudo headers checksum offloads */
 540 static void
 541 tap_tx_l3_cksum(char *packet, uint64_t ol_flags, unsigned int l2_len,
 542                 unsigned int l3_len, unsigned int l4_len, uint16_t **l4_cksum,
 543                 uint16_t *l4_phdr_cksum, uint32_t *l4_raw_cksum)
 544 {
 545         void *l3_hdr = packet + l2_len;
 546
 547         if (ol_flags & (RTE_MBUF_F_TX_IP_CKSUM | RTE_MBUF_F_TX_IPV4)) {
 548                 struct rte_ipv4_hdr *iph = l3_hdr;
 549                 uint16_t cksum;
 550
 551                 iph->hdr_checksum = 0;
 552                 cksum = rte_raw_cksum(iph, l3_len);
 553                 iph->hdr_checksum = (cksum == 0xffff) ? cksum : ~cksum;
 554         }
 555         if (ol_flags & RTE_MBUF_F_TX_L4_MASK) {
 556                 void *l4_hdr;
 557
 558                 l4_hdr = packet + l2_len + l3_len;
 559                 if ((ol_flags & RTE_MBUF_F_TX_L4_MASK) == RTE_MBUF_F_TX_UDP_CKSUM)
 560                         *l4_cksum = &((struct rte_udp_hdr *)l4_hdr)->dgram_cksum;
 561                 else if ((ol_flags & RTE_MBUF_F_TX_L4_MASK) == RTE_MBUF_F_TX_TCP_CKSUM)
 562                         *l4_cksum = &((struct rte_tcp_hdr *)l4_hdr)->cksum;
 563                 else
 564                         return;
 565                 **l4_cksum = 0;
 566                 if (ol_flags & RTE_MBUF_F_TX_IPV4)
 567                         *l4_phdr_cksum = rte_ipv4_phdr_cksum(l3_hdr, 0);
 568                 else
 569                         *l4_phdr_cksum = rte_ipv6_phdr_cksum(l3_hdr, 0);
 570                 *l4_raw_cksum = __rte_raw_cksum(l4_hdr, l4_len, 0);
 571         }
 572 }
 573
 574 static inline int
 575 tap_write_mbufs(struct tx_queue *txq, uint16_t num_mbufs,
 576                         struct rte_mbuf **pmbufs,
 577                         uint16_t *num_packets, unsigned long *num_tx_bytes)
 578 {
 579         int i;
 580         uint16_t l234_hlen;
 581         struct pmd_process_private *process_private;
 582
 583         process_private = rte_eth_devices[txq->out_port].process_private;
 584
 585         for (i = 0; i < num_mbufs; i++) {
 586                 struct rte_mbuf *mbuf = pmbufs[i];
 587                 struct iovec iovecs[mbuf->nb_segs + 2];
 588                 struct tun_pi pi = { .flags = 0, .proto = 0x00 };
 589                 struct rte_mbuf *seg = mbuf;
 590                 char m_copy[mbuf->data_len];
 591                 int proto;
 592                 int n;
 593                 int j;
 594                 int k; /* current index in iovecs for copying segments */
 595                 uint16_t seg_len; /* length of first segment */
 596                 uint16_t nb_segs;
 597                 uint16_t *l4_cksum; /* l4 checksum (pseudo header + payload) */
 598                 uint32_t l4_raw_cksum = 0; /* TCP/UDP payload raw checksum */
 599                 uint16_t l4_phdr_cksum = 0; /* TCP/UDP pseudo header checksum */
 600                 uint16_t is_cksum = 0; /* in case cksum should be offloaded */
 601
 602                 l4_cksum = NULL;
 603                 if (txq->type == ETH_TUNTAP_TYPE_TUN) {
 604                         /*
 605                          * TUN and TAP are created with IFF_NO_PI disabled.
 606                          * For TUN PMD this mandatory as fields are used by
 607                          * Kernel tun.c to determine whether its IP or non IP
 608                          * packets.
 609                          *
 610                          * The logic fetches the first byte of data from mbuf
 611                          * then compares whether its v4 or v6. If first byte
 612                          * is 4 or 6, then protocol field is updated.
 613                          */
 614                         char *buff_data = rte_pktmbuf_mtod(seg, void *);
 615                         proto = (*buff_data & 0xf0);
 616                         pi.proto = (proto == 0x40) ?
 617                                 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4) :
 618                                 ((proto == 0x60) ?
 619                                         rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6) :
 620                                         0x00);
 621                 }
 622
 623                 k = 0;
 624                 iovecs[k].iov_base = &pi;
 625                 iovecs[k].iov_len = sizeof(pi);
 626                 k++;
 627
 628                 nb_segs = mbuf->nb_segs;
 629                 if (txq->csum &&
 630                     ((mbuf->ol_flags & (RTE_MBUF_F_TX_IP_CKSUM | RTE_MBUF_F_TX_IPV4) ||
 631                       (mbuf->ol_flags & RTE_MBUF_F_TX_L4_MASK) == RTE_MBUF_F_TX_UDP_CKSUM ||
 632                       (mbuf->ol_flags & RTE_MBUF_F_TX_L4_MASK) == RTE_MBUF_F_TX_TCP_CKSUM))) {
 633                         is_cksum = 1;
 634
 635                         /* Support only packets with at least layer 4
 636                          * header included in the first segment
 637                          */
 638                         seg_len = rte_pktmbuf_data_len(mbuf);
 639                         l234_hlen = mbuf->l2_len + mbuf->l3_len + mbuf->l4_len;
 640                         if (seg_len < l234_hlen)
 641                                 return -1;
 642
 643                         /* To change checksums, work on a * copy of l2, l3
 644                          * headers + l4 pseudo header
 645                          */
 646                         rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
 647                                         l234_hlen);
 648                         tap_tx_l3_cksum(m_copy, mbuf->ol_flags,
 649                                        mbuf->l2_len, mbuf->l3_len, mbuf->l4_len,
 650                                        &l4_cksum, &l4_phdr_cksum,
 651                                        &l4_raw_cksum);
 652                         iovecs[k].iov_base = m_copy;
 653                         iovecs[k].iov_len = l234_hlen;
 654                         k++;
 655
 656                         /* Update next iovecs[] beyond l2, l3, l4 headers */
 657                         if (seg_len > l234_hlen) {
 658                                 iovecs[k].iov_len = seg_len - l234_hlen;
 659                                 iovecs[k].iov_base =
 660                                         rte_pktmbuf_mtod(seg, char *) +
 661                                                 l234_hlen;
 662                                 tap_tx_l4_add_rcksum(iovecs[k].iov_base,
 663                                         iovecs[k].iov_len, l4_cksum,
 664                                         &l4_raw_cksum);
 665                                 k++;
 666                                 nb_segs++;
 667                         }
 668                         seg = seg->next;
 669                 }
 670
 671                 for (j = k; j <= nb_segs; j++) {
 672                         iovecs[j].iov_len = rte_pktmbuf_data_len(seg);
 673                         iovecs[j].iov_base = rte_pktmbuf_mtod(seg, void *);
 674                         if (is_cksum)
 675                                 tap_tx_l4_add_rcksum(iovecs[j].iov_base,
 676                                         iovecs[j].iov_len, l4_cksum,
 677                                         &l4_raw_cksum);
 678                         seg = seg->next;
 679                 }
 680
 681                 if (is_cksum)
 682                         tap_tx_l4_cksum(l4_cksum, l4_phdr_cksum, l4_raw_cksum);
 683
 684                 /* copy the tx frame data */
 685                 n = writev(process_private->txq_fds[txq->queue_id], iovecs, j);
 686                 if (n <= 0)
 687                         return -1;
 688
 689                 (*num_packets)++;
 690                 (*num_tx_bytes) += rte_pktmbuf_pkt_len(mbuf);
 691         }
 692         return 0;
 693 }
 694
 695 /* Callback to handle sending packets from the tap interface
 696  */
 697 static uint16_t
 698 pmd_tx_burst(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 699 {
 700         struct tx_queue *txq = queue;
 701         uint16_t num_tx = 0;
 702         uint16_t num_packets = 0;
 703         unsigned long num_tx_bytes = 0;
 704         uint32_t max_size;
 705         int i;
 706
 707         if (unlikely(nb_pkts == 0))
 708                 return 0;
 709
 710         struct rte_mbuf *gso_mbufs[MAX_GSO_MBUFS];
 711         max_size = *txq->mtu + (RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN + 4);
 712         for (i = 0; i < nb_pkts; i++) {
 713                 struct rte_mbuf *mbuf_in = bufs[num_tx];
 714                 struct rte_mbuf **mbuf;
 715                 uint16_t num_mbufs = 0;
 716                 uint16_t tso_segsz = 0;
 717                 int ret;
 718                 int num_tso_mbufs;
 719                 uint16_t hdrs_len;
 720                 uint64_t tso;
 721
 722                 tso = mbuf_in->ol_flags & RTE_MBUF_F_TX_TCP_SEG;
 723                 if (tso) {
 724                         struct rte_gso_ctx *gso_ctx = &txq->gso_ctx;
 725
 726                         /* TCP segmentation implies TCP checksum offload */
 727                         mbuf_in->ol_flags |= RTE_MBUF_F_TX_TCP_CKSUM;
 728
 729                         /* gso size is calculated without RTE_ETHER_CRC_LEN */
 730                         hdrs_len = mbuf_in->l2_len + mbuf_in->l3_len +
 731                                         mbuf_in->l4_len;
 732                         tso_segsz = mbuf_in->tso_segsz + hdrs_len;
 733                         if (unlikely(tso_segsz == hdrs_len) ||
 734                                 tso_segsz > *txq->mtu) {
 735                                 txq->stats.errs++;
 736                                 break;
 737                         }
 738                         gso_ctx->gso_size = tso_segsz;
 739                         /* 'mbuf_in' packet to segment */
 740                         num_tso_mbufs = rte_gso_segment(mbuf_in,
 741                                 gso_ctx, /* gso control block */
 742                                 (struct rte_mbuf **)&gso_mbufs, /* out mbufs */
 743                                 RTE_DIM(gso_mbufs)); /* max tso mbufs */
 744
 745                         /* ret contains the number of new created mbufs */
 746                         if (num_tso_mbufs < 0)
 747                                 break;
 748
 749                         if (num_tso_mbufs >= 1) {
 750                                 mbuf = gso_mbufs;
 751                                 num_mbufs = num_tso_mbufs;
 752                         } else {
 753                                 /* 0 means it can be transmitted directly
 754                                  * without gso.
 755                                  */
 756                                 mbuf = &mbuf_in;
 757                                 num_mbufs = 1;
 758                         }
 759                 } else {
 760                         /* stats.errs will be incremented */
 761                         if (rte_pktmbuf_pkt_len(mbuf_in) > max_size)
 762                                 break;
 763
 764                         /* ret 0 indicates no new mbufs were created */
 765                         num_tso_mbufs = 0;
 766                         mbuf = &mbuf_in;
 767                         num_mbufs = 1;
 768                 }
 769
 770                 ret = tap_write_mbufs(txq, num_mbufs, mbuf,
 771                                 &num_packets, &num_tx_bytes);
 772                 if (ret == -1) {
 773                         txq->stats.errs++;
 774                         /* free tso mbufs */
 775                         if (num_tso_mbufs > 0)
 776                                 rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
 777                         break;
 778                 }
 779                 num_tx++;
 780                 /* free original mbuf */
 781                 rte_pktmbuf_free(mbuf_in);
 782                 /* free tso mbufs */
 783                 if (num_tso_mbufs > 0)
 784                         rte_pktmbuf_free_bulk(mbuf, num_tso_mbufs);
 785         }
 786
 787         txq->stats.opackets += num_packets;
 788         txq->stats.errs += nb_pkts - num_tx;
 789         txq->stats.obytes += num_tx_bytes;
 790
 791         return num_tx;
 792 }
 793
 794 static const char *
 795 tap_ioctl_req2str(unsigned long request)
 796 {
 797         switch (request) {
 798         case SIOCSIFFLAGS:
 799                 return "SIOCSIFFLAGS";
 800         case SIOCGIFFLAGS:
 801                 return "SIOCGIFFLAGS";
 802         case SIOCGIFHWADDR:
 803                 return "SIOCGIFHWADDR";
 804         case SIOCSIFHWADDR:
 805                 return "SIOCSIFHWADDR";
 806         case SIOCSIFMTU:
 807                 return "SIOCSIFMTU";
 808         }
 809         return "UNKNOWN";
 810 }
 811
 812 static int
 813 tap_ioctl(struct pmd_internals *pmd, unsigned long request,
 814           struct ifreq *ifr, int set, enum ioctl_mode mode)
 815 {
 816         short req_flags = ifr->ifr_flags;
 817         int remote = pmd->remote_if_index &&
 818                 (mode == REMOTE_ONLY || mode == LOCAL_AND_REMOTE);
 819
 820         if (!pmd->remote_if_index && mode == REMOTE_ONLY)
 821                 return 0;
 822         /*
 823          * If there is a remote netdevice, apply ioctl on it, then apply it on
 824          * the tap netdevice.
 825          */
 826 apply:
 827         if (remote)
 828                 strlcpy(ifr->ifr_name, pmd->remote_iface, IFNAMSIZ);
 829         else if (mode == LOCAL_ONLY || mode == LOCAL_AND_REMOTE)
 830                 strlcpy(ifr->ifr_name, pmd->name, IFNAMSIZ);
 831         switch (request) {
 832         case SIOCSIFFLAGS:
 833                 /* fetch current flags to leave other flags untouched */
 834                 if (ioctl(pmd->ioctl_sock, SIOCGIFFLAGS, ifr) < 0)
 835                         goto error;
 836                 if (set)
 837                         ifr->ifr_flags |= req_flags;
 838                 else
 839                         ifr->ifr_flags &= ~req_flags;
 840                 break;
 841         case SIOCGIFFLAGS:
 842         case SIOCGIFHWADDR:
 843         case SIOCSIFHWADDR:
 844         case SIOCSIFMTU:
 845                 break;
 846         default:
 847                 TAP_LOG(WARNING, "%s: ioctl() called with wrong arg",
 848                         pmd->name);
 849                 return -EINVAL;
 850         }
 851         if (ioctl(pmd->ioctl_sock, request, ifr) < 0)
 852                 goto error;
 853         if (remote-- && mode == LOCAL_AND_REMOTE)
 854                 goto apply;
 855         return 0;
 856
 857 error:
 858         TAP_LOG(DEBUG, "%s(%s) failed: %s(%d)", ifr->ifr_name,
 859                 tap_ioctl_req2str(request), strerror(errno), errno);
 860         return -errno;
 861 }
 862
 863 static int
 864 tap_link_set_down(struct rte_eth_dev *dev)
 865 {
 866         struct pmd_internals *pmd = dev->data->dev_private;
 867         struct ifreq ifr = { .ifr_flags = IFF_UP };
 868
 869         dev->data->dev_link.link_status = RTE_ETH_LINK_DOWN;
 870         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_ONLY);
 871 }
 872
 873 static int
 874 tap_link_set_up(struct rte_eth_dev *dev)
 875 {
 876         struct pmd_internals *pmd = dev->data->dev_private;
 877         struct ifreq ifr = { .ifr_flags = IFF_UP };
 878
 879         dev->data->dev_link.link_status = RTE_ETH_LINK_UP;
 880         return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
 881 }
 882
 883 static int
 884 tap_dev_start(struct rte_eth_dev *dev)
 885 {
 886         int err, i;
 887
 888         err = tap_intr_handle_set(dev, 1);
 889         if (err)
 890                 return err;
 891
 892         err = tap_link_set_up(dev);
 893         if (err)
 894                 return err;
 895
 896         for (i = 0; i < dev->data->nb_tx_queues; i++)
 897                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 898         for (i = 0; i < dev->data->nb_rx_queues; i++)
 899                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STARTED;
 900
 901         return err;
 902 }
 903
 904 /* This function gets called when the current port gets stopped.
 905  */
 906 static int
 907 tap_dev_stop(struct rte_eth_dev *dev)
 908 {
 909         int i;
 910
 911         for (i = 0; i < dev->data->nb_tx_queues; i++)
 912                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 913         for (i = 0; i < dev->data->nb_rx_queues; i++)
 914                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
 915
 916         tap_intr_handle_set(dev, 0);
 917         tap_link_set_down(dev);
 918
 919         return 0;
 920 }
 921
 922 static int
 923 tap_dev_configure(struct rte_eth_dev *dev)
 924 {
 925         struct pmd_internals *pmd = dev->data->dev_private;
 926
 927         if (dev->data->nb_rx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 928                 TAP_LOG(ERR,
 929                         "%s: number of rx queues %d exceeds max num of queues %d",
 930                         dev->device->name,
 931                         dev->data->nb_rx_queues,
 932                         RTE_PMD_TAP_MAX_QUEUES);
 933                 return -1;
 934         }
 935         if (dev->data->nb_tx_queues > RTE_PMD_TAP_MAX_QUEUES) {
 936                 TAP_LOG(ERR,
 937                         "%s: number of tx queues %d exceeds max num of queues %d",
 938                         dev->device->name,
 939                         dev->data->nb_tx_queues,
 940                         RTE_PMD_TAP_MAX_QUEUES);
 941                 return -1;
 942         }
 943
 944         TAP_LOG(INFO, "%s: %s: TX configured queues number: %u",
 945                 dev->device->name, pmd->name, dev->data->nb_tx_queues);
 946
 947         TAP_LOG(INFO, "%s: %s: RX configured queues number: %u",
 948                 dev->device->name, pmd->name, dev->data->nb_rx_queues);
 949
 950         return 0;
 951 }
 952
 953 static uint32_t
 954 tap_dev_speed_capa(void)
 955 {
 956         uint32_t speed = pmd_link.link_speed;
 957         uint32_t capa = 0;
 958
 959         if (speed >= RTE_ETH_SPEED_NUM_10M)
 960                 capa |= RTE_ETH_LINK_SPEED_10M;
 961         if (speed >= RTE_ETH_SPEED_NUM_100M)
 962                 capa |= RTE_ETH_LINK_SPEED_100M;
 963         if (speed >= RTE_ETH_SPEED_NUM_1G)
 964                 capa |= RTE_ETH_LINK_SPEED_1G;
 965         if (speed >= RTE_ETH_SPEED_NUM_5G)
 966                 capa |= RTE_ETH_LINK_SPEED_2_5G;
 967         if (speed >= RTE_ETH_SPEED_NUM_5G)
 968                 capa |= RTE_ETH_LINK_SPEED_5G;
 969         if (speed >= RTE_ETH_SPEED_NUM_10G)
 970                 capa |= RTE_ETH_LINK_SPEED_10G;
 971         if (speed >= RTE_ETH_SPEED_NUM_20G)
 972                 capa |= RTE_ETH_LINK_SPEED_20G;
 973         if (speed >= RTE_ETH_SPEED_NUM_25G)
 974                 capa |= RTE_ETH_LINK_SPEED_25G;
 975         if (speed >= RTE_ETH_SPEED_NUM_40G)
 976                 capa |= RTE_ETH_LINK_SPEED_40G;
 977         if (speed >= RTE_ETH_SPEED_NUM_50G)
 978                 capa |= RTE_ETH_LINK_SPEED_50G;
 979         if (speed >= RTE_ETH_SPEED_NUM_56G)
 980                 capa |= RTE_ETH_LINK_SPEED_56G;
 981         if (speed >= RTE_ETH_SPEED_NUM_100G)
 982                 capa |= RTE_ETH_LINK_SPEED_100G;
 983
 984         return capa;
 985 }
 986
 987 static int
 988 tap_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 989 {
 990         struct pmd_internals *internals = dev->data->dev_private;
 991
 992         dev_info->if_index = internals->if_index;
 993         dev_info->max_mac_addrs = 1;
 994         dev_info->max_rx_pktlen = (uint32_t)RTE_ETHER_MAX_VLAN_FRAME_LEN;
 995         dev_info->max_rx_queues = RTE_PMD_TAP_MAX_QUEUES;
 996         dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES;
 997         dev_info->min_rx_bufsize = 0;
 998         dev_info->speed_capa = tap_dev_speed_capa();
 999         dev_info->rx_queue_offload_capa = TAP_RX_OFFLOAD;
1000         dev_info->rx_offload_capa = dev_info->rx_queue_offload_capa;
1001         dev_info->tx_queue_offload_capa = TAP_TX_OFFLOAD;
1002         dev_info->tx_offload_capa = dev_info->tx_queue_offload_capa;
1003         dev_info->hash_key_size = TAP_RSS_HASH_KEY_SIZE;
1004         /*
1005          * limitation: TAP supports all of IP, UDP and TCP hash
1006          * functions together and not in partial combinations
1007          */
1008         dev_info->flow_type_rss_offloads = ~TAP_RSS_HF_MASK;
1009         dev_info->dev_capa &= ~RTE_ETH_DEV_CAPA_FLOW_RULE_KEEP;
1010
1011         return 0;
1012 }
1013
1014 static int
1015 tap_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *tap_stats)
1016 {
1017         unsigned int i, imax;
1018         unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
1019         unsigned long rx_bytes_total = 0, tx_bytes_total = 0;
1020         unsigned long rx_nombuf = 0, ierrors = 0;
1021         const struct pmd_internals *pmd = dev->data->dev_private;
1022
1023         /* rx queue statistics */
1024         imax = (dev->data->nb_rx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
1025                 dev->data->nb_rx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
1026         for (i = 0; i < imax; i++) {
1027                 tap_stats->q_ipackets[i] = pmd->rxq[i].stats.ipackets;
1028                 tap_stats->q_ibytes[i] = pmd->rxq[i].stats.ibytes;
1029                 rx_total += tap_stats->q_ipackets[i];
1030                 rx_bytes_total += tap_stats->q_ibytes[i];
1031                 rx_nombuf += pmd->rxq[i].stats.rx_nombuf;
1032                 ierrors += pmd->rxq[i].stats.ierrors;
1033         }
1034
1035         /* tx queue statistics */
1036         imax = (dev->data->nb_tx_queues < RTE_ETHDEV_QUEUE_STAT_CNTRS) ?
1037                 dev->data->nb_tx_queues : RTE_ETHDEV_QUEUE_STAT_CNTRS;
1038
1039         for (i = 0; i < imax; i++) {
1040                 tap_stats->q_opackets[i] = pmd->txq[i].stats.opackets;
1041                 tap_stats->q_obytes[i] = pmd->txq[i].stats.obytes;
1042                 tx_total += tap_stats->q_opackets[i];
1043                 tx_err_total += pmd->txq[i].stats.errs;
1044                 tx_bytes_total += tap_stats->q_obytes[i];
1045         }
1046
1047         tap_stats->ipackets = rx_total;
1048         tap_stats->ibytes = rx_bytes_total;
1049         tap_stats->ierrors = ierrors;
1050         tap_stats->rx_nombuf = rx_nombuf;
1051         tap_stats->opackets = tx_total;
1052         tap_stats->oerrors = tx_err_total;
1053         tap_stats->obytes = tx_bytes_total;
1054         return 0;
1055 }
1056
1057 static int
1058 tap_stats_reset(struct rte_eth_dev *dev)
1059 {
1060         int i;
1061         struct pmd_internals *pmd = dev->data->dev_private;
1062
1063         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1064                 pmd->rxq[i].stats.ipackets = 0;
1065                 pmd->rxq[i].stats.ibytes = 0;
1066                 pmd->rxq[i].stats.ierrors = 0;
1067                 pmd->rxq[i].stats.rx_nombuf = 0;
1068
1069                 pmd->txq[i].stats.opackets = 0;
1070                 pmd->txq[i].stats.errs = 0;
1071                 pmd->txq[i].stats.obytes = 0;
1072         }
1073
1074         return 0;
1075 }
1076
1077 static int
1078 tap_dev_close(struct rte_eth_dev *dev)
1079 {
1080         int i;
1081         struct pmd_internals *internals = dev->data->dev_private;
1082         struct pmd_process_private *process_private = dev->process_private;
1083         struct rx_queue *rxq;
1084
1085         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1086                 rte_free(dev->process_private);
1087                 return 0;
1088         }
1089
1090         tap_link_set_down(dev);
1091         if (internals->nlsk_fd != -1) {
1092                 tap_flow_flush(dev, NULL);
1093                 tap_flow_implicit_flush(internals, NULL);
1094                 tap_nl_final(internals->nlsk_fd);
1095                 internals->nlsk_fd = -1;
1096         }
1097
1098         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1099                 if (process_private->rxq_fds[i] != -1) {
1100                         rxq = &internals->rxq[i];
1101                         close(process_private->rxq_fds[i]);
1102                         process_private->rxq_fds[i] = -1;
1103                         tap_rxq_pool_free(rxq->pool);
1104                         rte_free(rxq->iovecs);
1105                         rxq->pool = NULL;
1106                         rxq->iovecs = NULL;
1107                 }
1108                 if (process_private->txq_fds[i] != -1) {
1109                         close(process_private->txq_fds[i]);
1110                         process_private->txq_fds[i] = -1;
1111                 }
1112         }
1113
1114         if (internals->remote_if_index) {
1115                 /* Restore initial remote state */
1116                 int ret = ioctl(internals->ioctl_sock, SIOCSIFFLAGS,
1117                                 &internals->remote_initial_flags);
1118                 if (ret)
1119                         TAP_LOG(ERR, "restore remote state failed: %d", ret);
1120
1121         }
1122
1123         rte_mempool_free(internals->gso_ctx_mp);
1124         internals->gso_ctx_mp = NULL;
1125
1126         if (internals->ka_fd != -1) {
1127                 close(internals->ka_fd);
1128                 internals->ka_fd = -1;
1129         }
1130
1131         /* mac_addrs must not be freed alone because part of dev_private */
1132         dev->data->mac_addrs = NULL;
1133
1134         internals = dev->data->dev_private;
1135         TAP_LOG(DEBUG, "Closing %s Ethernet device on numa %u",
1136                 tuntap_types[internals->type], rte_socket_id());
1137
1138         if (internals->ioctl_sock != -1) {
1139                 close(internals->ioctl_sock);
1140                 internals->ioctl_sock = -1;
1141         }
1142         rte_free(dev->process_private);
1143         if (tap_devices_count == 1)
1144                 rte_mp_action_unregister(TAP_MP_KEY);
1145         tap_devices_count--;
1146         /*
1147          * Since TUN device has no more opened file descriptors
1148          * it will be removed from kernel
1149          */
1150
1151         return 0;
1152 }
1153
1154 static void
1155 tap_rx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
1156 {
1157         struct rx_queue *rxq = dev->data->rx_queues[qid];
1158         struct pmd_process_private *process_private;
1159
1160         if (!rxq)
1161                 return;
1162         process_private = rte_eth_devices[rxq->in_port].process_private;
1163         if (process_private->rxq_fds[rxq->queue_id] != -1) {
1164                 close(process_private->rxq_fds[rxq->queue_id]);
1165                 process_private->rxq_fds[rxq->queue_id] = -1;
1166                 tap_rxq_pool_free(rxq->pool);
1167                 rte_free(rxq->iovecs);
1168                 rxq->pool = NULL;
1169                 rxq->iovecs = NULL;
1170         }
1171 }
1172
1173 static void
1174 tap_tx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
1175 {
1176         struct tx_queue *txq = dev->data->tx_queues[qid];
1177         struct pmd_process_private *process_private;
1178
1179         if (!txq)
1180                 return;
1181         process_private = rte_eth_devices[txq->out_port].process_private;
1182
1183         if (process_private->txq_fds[txq->queue_id] != -1) {
1184                 close(process_private->txq_fds[txq->queue_id]);
1185                 process_private->txq_fds[txq->queue_id] = -1;
1186         }
1187 }
1188
1189 static int
1190 tap_link_update(struct rte_eth_dev *dev, int wait_to_complete __rte_unused)
1191 {
1192         struct rte_eth_link *dev_link = &dev->data->dev_link;
1193         struct pmd_internals *pmd = dev->data->dev_private;
1194         struct ifreq ifr = { .ifr_flags = 0 };
1195
1196         if (pmd->remote_if_index) {
1197                 tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, REMOTE_ONLY);
1198                 if (!(ifr.ifr_flags & IFF_UP) ||
1199                     !(ifr.ifr_flags & IFF_RUNNING)) {
1200                         dev_link->link_status = RTE_ETH_LINK_DOWN;
1201                         return 0;
1202                 }
1203         }
1204         tap_ioctl(pmd, SIOCGIFFLAGS, &ifr, 0, LOCAL_ONLY);
1205         dev_link->link_status =
1206                 ((ifr.ifr_flags & IFF_UP) && (ifr.ifr_flags & IFF_RUNNING) ?
1207                  RTE_ETH_LINK_UP :
1208                  RTE_ETH_LINK_DOWN);
1209         return 0;
1210 }
1211
1212 static int
1213 tap_promisc_enable(struct rte_eth_dev *dev)
1214 {
1215         struct pmd_internals *pmd = dev->data->dev_private;
1216         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1217         int ret;
1218
1219         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1220         if (ret != 0)
1221                 return ret;
1222
1223         if (pmd->remote_if_index && !pmd->flow_isolate) {
1224                 dev->data->promiscuous = 1;
1225                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_PROMISC);
1226                 if (ret != 0) {
1227                         /* Rollback promisc flag */
1228                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1229                         /*
1230                          * rte_eth_dev_promiscuous_enable() rollback
1231                          * dev->data->promiscuous in the case of failure.
1232                          */
1233                         return ret;
1234                 }
1235         }
1236
1237         return 0;
1238 }
1239
1240 static int
1241 tap_promisc_disable(struct rte_eth_dev *dev)
1242 {
1243         struct pmd_internals *pmd = dev->data->dev_private;
1244         struct ifreq ifr = { .ifr_flags = IFF_PROMISC };
1245         int ret;
1246
1247         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1248         if (ret != 0)
1249                 return ret;
1250
1251         if (pmd->remote_if_index && !pmd->flow_isolate) {
1252                 dev->data->promiscuous = 0;
1253                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_PROMISC);
1254                 if (ret != 0) {
1255                         /* Rollback promisc flag */
1256                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1257                         /*
1258                          * rte_eth_dev_promiscuous_disable() rollback
1259                          * dev->data->promiscuous in the case of failure.
1260                          */
1261                         return ret;
1262                 }
1263         }
1264
1265         return 0;
1266 }
1267
1268 static int
1269 tap_allmulti_enable(struct rte_eth_dev *dev)
1270 {
1271         struct pmd_internals *pmd = dev->data->dev_private;
1272         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1273         int ret;
1274
1275         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1276         if (ret != 0)
1277                 return ret;
1278
1279         if (pmd->remote_if_index && !pmd->flow_isolate) {
1280                 dev->data->all_multicast = 1;
1281                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_ALLMULTI);
1282                 if (ret != 0) {
1283                         /* Rollback allmulti flag */
1284                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1285                         /*
1286                          * rte_eth_dev_allmulticast_enable() rollback
1287                          * dev->data->all_multicast in the case of failure.
1288                          */
1289                         return ret;
1290                 }
1291         }
1292
1293         return 0;
1294 }
1295
1296 static int
1297 tap_allmulti_disable(struct rte_eth_dev *dev)
1298 {
1299         struct pmd_internals *pmd = dev->data->dev_private;
1300         struct ifreq ifr = { .ifr_flags = IFF_ALLMULTI };
1301         int ret;
1302
1303         ret = tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 0, LOCAL_AND_REMOTE);
1304         if (ret != 0)
1305                 return ret;
1306
1307         if (pmd->remote_if_index && !pmd->flow_isolate) {
1308                 dev->data->all_multicast = 0;
1309                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_ALLMULTI);
1310                 if (ret != 0) {
1311                         /* Rollback allmulti flag */
1312                         tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
1313                         /*
1314                          * rte_eth_dev_allmulticast_disable() rollback
1315                          * dev->data->all_multicast in the case of failure.
1316                          */
1317                         return ret;
1318                 }
1319         }
1320
1321         return 0;
1322 }
1323
1324 static int
1325 tap_mac_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1326 {
1327         struct pmd_internals *pmd = dev->data->dev_private;
1328         enum ioctl_mode mode = LOCAL_ONLY;
1329         struct ifreq ifr;
1330         int ret;
1331
1332         if (pmd->type == ETH_TUNTAP_TYPE_TUN) {
1333                 TAP_LOG(ERR, "%s: can't MAC address for TUN",
1334                         dev->device->name);
1335                 return -ENOTSUP;
1336         }
1337
1338         if (rte_is_zero_ether_addr(mac_addr)) {
1339                 TAP_LOG(ERR, "%s: can't set an empty MAC address",
1340                         dev->device->name);
1341                 return -EINVAL;
1342         }
1343         /* Check the actual current MAC address on the tap netdevice */
1344         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, LOCAL_ONLY);
1345         if (ret < 0)
1346                 return ret;
1347         if (rte_is_same_ether_addr(
1348                         (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1349                         mac_addr))
1350                 return 0;
1351         /* Check the current MAC address on the remote */
1352         ret = tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY);
1353         if (ret < 0)
1354                 return ret;
1355         if (!rte_is_same_ether_addr(
1356                         (struct rte_ether_addr *)&ifr.ifr_hwaddr.sa_data,
1357                         mac_addr))
1358                 mode = LOCAL_AND_REMOTE;
1359         ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1360         rte_memcpy(ifr.ifr_hwaddr.sa_data, mac_addr, RTE_ETHER_ADDR_LEN);
1361         ret = tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 1, mode);
1362         if (ret < 0)
1363                 return ret;
1364         rte_memcpy(&pmd->eth_addr, mac_addr, RTE_ETHER_ADDR_LEN);
1365         if (pmd->remote_if_index && !pmd->flow_isolate) {
1366                 /* Replace MAC redirection rule after a MAC change */
1367                 ret = tap_flow_implicit_destroy(pmd, TAP_REMOTE_LOCAL_MAC);
1368                 if (ret < 0) {
1369                         TAP_LOG(ERR,
1370                                 "%s: Couldn't delete MAC redirection rule",
1371                                 dev->device->name);
1372                         return ret;
1373                 }
1374                 ret = tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC);
1375                 if (ret < 0) {
1376                         TAP_LOG(ERR,
1377                                 "%s: Couldn't add MAC redirection rule",
1378                                 dev->device->name);
1379                         return ret;
1380                 }
1381         }
1382
1383         return 0;
1384 }
1385
1386 static int
1387 tap_gso_ctx_setup(struct rte_gso_ctx *gso_ctx, struct rte_eth_dev *dev)
1388 {
1389         uint32_t gso_types;
1390         char pool_name[64];
1391         struct pmd_internals *pmd = dev->data->dev_private;
1392         int ret;
1393
1394         /* initialize GSO context */
1395         gso_types = RTE_ETH_TX_OFFLOAD_TCP_TSO;
1396         if (!pmd->gso_ctx_mp) {
1397                 /*
1398                  * Create private mbuf pool with TAP_GSO_MBUF_SEG_SIZE
1399                  * bytes size per mbuf use this pool for both direct and
1400                  * indirect mbufs
1401                  */
1402                 ret = snprintf(pool_name, sizeof(pool_name), "mp_%s",
1403                                 dev->device->name);
1404                 if (ret < 0 || ret >= (int)sizeof(pool_name)) {
1405                         TAP_LOG(ERR,
1406                                 "%s: failed to create mbuf pool name for device %s,"
1407                                 "device name too long or output error, ret: %d\n",
1408                                 pmd->name, dev->device->name, ret);
1409                         return -ENAMETOOLONG;
1410                 }
1411                 pmd->gso_ctx_mp = rte_pktmbuf_pool_create(pool_name,
1412                         TAP_GSO_MBUFS_NUM, TAP_GSO_MBUF_CACHE_SIZE, 0,
1413                         RTE_PKTMBUF_HEADROOM + TAP_GSO_MBUF_SEG_SIZE,
1414                         SOCKET_ID_ANY);
1415                 if (!pmd->gso_ctx_mp) {
1416                         TAP_LOG(ERR,
1417                                 "%s: failed to create mbuf pool for device %s\n",
1418                                 pmd->name, dev->device->name);
1419                         return -1;
1420                 }
1421         }
1422
1423         gso_ctx->direct_pool = pmd->gso_ctx_mp;
1424         gso_ctx->indirect_pool = pmd->gso_ctx_mp;
1425         gso_ctx->gso_types = gso_types;
1426         gso_ctx->gso_size = 0; /* gso_size is set in tx_burst() per packet */
1427         gso_ctx->flag = 0;
1428
1429         return 0;
1430 }
1431
1432 static int
1433 tap_setup_queue(struct rte_eth_dev *dev,
1434                 struct pmd_internals *internals,
1435                 uint16_t qid,
1436                 int is_rx)
1437 {
1438         int ret;
1439         int *fd;
1440         int *other_fd;
1441         const char *dir;
1442         struct pmd_internals *pmd = dev->data->dev_private;
1443         struct pmd_process_private *process_private = dev->process_private;
1444         struct rx_queue *rx = &internals->rxq[qid];
1445         struct tx_queue *tx = &internals->txq[qid];
1446         struct rte_gso_ctx *gso_ctx;
1447
1448         if (is_rx) {
1449                 fd = &process_private->rxq_fds[qid];
1450                 other_fd = &process_private->txq_fds[qid];
1451                 dir = "rx";
1452                 gso_ctx = NULL;
1453         } else {
1454                 fd = &process_private->txq_fds[qid];
1455                 other_fd = &process_private->rxq_fds[qid];
1456                 dir = "tx";
1457                 gso_ctx = &tx->gso_ctx;
1458         }
1459         if (*fd != -1) {
1460                 /* fd for this queue already exists */
1461                 TAP_LOG(DEBUG, "%s: fd %d for %s queue qid %d exists",
1462                         pmd->name, *fd, dir, qid);
1463                 gso_ctx = NULL;
1464         } else if (*other_fd != -1) {
1465                 /* Only other_fd exists. dup it */
1466                 *fd = dup(*other_fd);
1467                 if (*fd < 0) {
1468                         *fd = -1;
1469                         TAP_LOG(ERR, "%s: dup() failed.", pmd->name);
1470                         return -1;
1471                 }
1472                 TAP_LOG(DEBUG, "%s: dup fd %d for %s queue qid %d (%d)",
1473                         pmd->name, *other_fd, dir, qid, *fd);
1474         } else {
1475                 /* Both RX and TX fds do not exist (equal -1). Create fd */
1476                 *fd = tun_alloc(pmd, 0);
1477                 if (*fd < 0) {
1478                         *fd = -1; /* restore original value */
1479                         TAP_LOG(ERR, "%s: tun_alloc() failed.", pmd->name);
1480                         return -1;
1481                 }
1482                 TAP_LOG(DEBUG, "%s: add %s queue for qid %d fd %d",
1483                         pmd->name, dir, qid, *fd);
1484         }
1485
1486         tx->mtu = &dev->data->mtu;
1487         rx->rxmode = &dev->data->dev_conf.rxmode;
1488         if (gso_ctx) {
1489                 ret = tap_gso_ctx_setup(gso_ctx, dev);
1490                 if (ret)
1491                         return -1;
1492         }
1493
1494         tx->type = pmd->type;
1495
1496         return *fd;
1497 }
1498
1499 static int
1500 tap_rx_queue_setup(struct rte_eth_dev *dev,
1501                    uint16_t rx_queue_id,
1502                    uint16_t nb_rx_desc,
1503                    unsigned int socket_id,
1504                    const struct rte_eth_rxconf *rx_conf __rte_unused,
1505                    struct rte_mempool *mp)
1506 {
1507         struct pmd_internals *internals = dev->data->dev_private;
1508         struct pmd_process_private *process_private = dev->process_private;
1509         struct rx_queue *rxq = &internals->rxq[rx_queue_id];
1510         struct rte_mbuf **tmp = &rxq->pool;
1511         long iov_max = sysconf(_SC_IOV_MAX);
1512
1513         if (iov_max <= 0) {
1514                 TAP_LOG(WARNING,
1515                         "_SC_IOV_MAX is not defined. Using %d as default",
1516                         TAP_IOV_DEFAULT_MAX);
1517                 iov_max = TAP_IOV_DEFAULT_MAX;
1518         }
1519         uint16_t nb_desc = RTE_MIN(nb_rx_desc, iov_max - 1);
1520         struct iovec (*iovecs)[nb_desc + 1];
1521         int data_off = RTE_PKTMBUF_HEADROOM;
1522         int ret = 0;
1523         int fd;
1524         int i;
1525
1526         if (rx_queue_id >= dev->data->nb_rx_queues || !mp) {
1527                 TAP_LOG(WARNING,
1528                         "nb_rx_queues %d too small or mempool NULL",
1529                         dev->data->nb_rx_queues);
1530                 return -1;
1531         }
1532
1533         rxq->mp = mp;
1534         rxq->trigger_seen = 1; /* force initial burst */
1535         rxq->in_port = dev->data->port_id;
1536         rxq->queue_id = rx_queue_id;
1537         rxq->nb_rx_desc = nb_desc;
1538         iovecs = rte_zmalloc_socket(dev->device->name, sizeof(*iovecs), 0,
1539                                     socket_id);
1540         if (!iovecs) {
1541                 TAP_LOG(WARNING,
1542                         "%s: Couldn't allocate %d RX descriptors",
1543                         dev->device->name, nb_desc);
1544                 return -ENOMEM;
1545         }
1546         rxq->iovecs = iovecs;
1547
1548         dev->data->rx_queues[rx_queue_id] = rxq;
1549         fd = tap_setup_queue(dev, internals, rx_queue_id, 1);
1550         if (fd == -1) {
1551                 ret = fd;
1552                 goto error;
1553         }
1554
1555         (*rxq->iovecs)[0].iov_len = sizeof(struct tun_pi);
1556         (*rxq->iovecs)[0].iov_base = &rxq->pi;
1557
1558         for (i = 1; i <= nb_desc; i++) {
1559                 *tmp = rte_pktmbuf_alloc(rxq->mp);
1560                 if (!*tmp) {
1561                         TAP_LOG(WARNING,
1562                                 "%s: couldn't allocate memory for queue %d",
1563                                 dev->device->name, rx_queue_id);
1564                         ret = -ENOMEM;
1565                         goto error;
1566                 }
1567                 (*rxq->iovecs)[i].iov_len = (*tmp)->buf_len - data_off;
1568                 (*rxq->iovecs)[i].iov_base =
1569                         (char *)(*tmp)->buf_addr + data_off;
1570                 data_off = 0;
1571                 tmp = &(*tmp)->next;
1572         }
1573
1574         TAP_LOG(DEBUG, "  RX TUNTAP device name %s, qid %d on fd %d",
1575                 internals->name, rx_queue_id,
1576                 process_private->rxq_fds[rx_queue_id]);
1577
1578         return 0;
1579
1580 error:
1581         tap_rxq_pool_free(rxq->pool);
1582         rxq->pool = NULL;
1583         rte_free(rxq->iovecs);
1584         rxq->iovecs = NULL;
1585         return ret;
1586 }
1587
1588 static int
1589 tap_tx_queue_setup(struct rte_eth_dev *dev,
1590                    uint16_t tx_queue_id,
1591                    uint16_t nb_tx_desc __rte_unused,
1592                    unsigned int socket_id __rte_unused,
1593                    const struct rte_eth_txconf *tx_conf)
1594 {
1595         struct pmd_internals *internals = dev->data->dev_private;
1596         struct pmd_process_private *process_private = dev->process_private;
1597         struct tx_queue *txq;
1598         int ret;
1599         uint64_t offloads;
1600
1601         if (tx_queue_id >= dev->data->nb_tx_queues)
1602                 return -1;
1603         dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id];
1604         txq = dev->data->tx_queues[tx_queue_id];
1605         txq->out_port = dev->data->port_id;
1606         txq->queue_id = tx_queue_id;
1607
1608         offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1609         txq->csum = !!(offloads &
1610                         (RTE_ETH_TX_OFFLOAD_IPV4_CKSUM |
1611                          RTE_ETH_TX_OFFLOAD_UDP_CKSUM |
1612                          RTE_ETH_TX_OFFLOAD_TCP_CKSUM));
1613
1614         ret = tap_setup_queue(dev, internals, tx_queue_id, 0);
1615         if (ret == -1)
1616                 return -1;
1617         TAP_LOG(DEBUG,
1618                 "  TX TUNTAP device name %s, qid %d on fd %d csum %s",
1619                 internals->name, tx_queue_id,
1620                 process_private->txq_fds[tx_queue_id],
1621                 txq->csum ? "on" : "off");
1622
1623         return 0;
1624 }
1625
1626 static int
1627 tap_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1628 {
1629         struct pmd_internals *pmd = dev->data->dev_private;
1630         struct ifreq ifr = { .ifr_mtu = mtu };
1631
1632         return tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE);
1633 }
1634
1635 static int
1636 tap_set_mc_addr_list(struct rte_eth_dev *dev __rte_unused,
1637                      struct rte_ether_addr *mc_addr_set __rte_unused,
1638                      uint32_t nb_mc_addr __rte_unused)
1639 {
1640         /*
1641          * Nothing to do actually: the tap has no filtering whatsoever, every
1642          * packet is received.
1643          */
1644         return 0;
1645 }
1646
1647 static int
1648 tap_nl_msg_handler(struct nlmsghdr *nh, void *arg)
1649 {
1650         struct rte_eth_dev *dev = arg;
1651         struct pmd_internals *pmd = dev->data->dev_private;
1652         struct ifinfomsg *info = NLMSG_DATA(nh);
1653
1654         if (nh->nlmsg_type != RTM_NEWLINK ||
1655             (info->ifi_index != pmd->if_index &&
1656              info->ifi_index != pmd->remote_if_index))
1657                 return 0;
1658         return tap_link_update(dev, 0);
1659 }
1660
1661 static void
1662 tap_dev_intr_handler(void *cb_arg)
1663 {
1664         struct rte_eth_dev *dev = cb_arg;
1665         struct pmd_internals *pmd = dev->data->dev_private;
1666
1667         if (rte_intr_fd_get(pmd->intr_handle) >= 0)
1668                 tap_nl_recv(rte_intr_fd_get(pmd->intr_handle),
1669                             tap_nl_msg_handler, dev);
1670 }
1671
1672 static int
1673 tap_lsc_intr_handle_set(struct rte_eth_dev *dev, int set)
1674 {
1675         struct pmd_internals *pmd = dev->data->dev_private;
1676         int ret;
1677
1678         /* In any case, disable interrupt if the conf is no longer there. */
1679         if (!dev->data->dev_conf.intr_conf.lsc) {
1680                 if (rte_intr_fd_get(pmd->intr_handle) != -1)
1681                         goto clean;
1682
1683                 return 0;
1684         }
1685         if (set) {
1686                 rte_intr_fd_set(pmd->intr_handle, tap_nl_init(RTMGRP_LINK));
1687                 if (unlikely(rte_intr_fd_get(pmd->intr_handle) == -1))
1688                         return -EBADF;
1689                 return rte_intr_callback_register(
1690                         pmd->intr_handle, tap_dev_intr_handler, dev);
1691         }
1692
1693 clean:
1694         do {
1695                 ret = rte_intr_callback_unregister(pmd->intr_handle,
1696                         tap_dev_intr_handler, dev);
1697                 if (ret >= 0) {
1698                         break;
1699                 } else if (ret == -EAGAIN) {
1700                         rte_delay_ms(100);
1701                 } else {
1702                         TAP_LOG(ERR, "intr callback unregister failed: %d",
1703                                      ret);
1704                         break;
1705                 }
1706         } while (true);
1707
1708         if (rte_intr_fd_get(pmd->intr_handle) >= 0) {
1709                 tap_nl_final(rte_intr_fd_get(pmd->intr_handle));
1710                 rte_intr_fd_set(pmd->intr_handle, -1);
1711         }
1712
1713         return 0;
1714 }
1715
1716 static int
1717 tap_intr_handle_set(struct rte_eth_dev *dev, int set)
1718 {
1719         int err;
1720
1721         err = tap_lsc_intr_handle_set(dev, set);
1722         if (err < 0) {
1723                 if (!set)
1724                         tap_rx_intr_vec_set(dev, 0);
1725                 return err;
1726         }
1727         err = tap_rx_intr_vec_set(dev, set);
1728         if (err && set)
1729                 tap_lsc_intr_handle_set(dev, 0);
1730         return err;
1731 }
1732
1733 static const uint32_t*
1734 tap_dev_supported_ptypes_get(struct rte_eth_dev *dev __rte_unused)
1735 {
1736         static const uint32_t ptypes[] = {
1737                 RTE_PTYPE_INNER_L2_ETHER,
1738                 RTE_PTYPE_INNER_L2_ETHER_VLAN,
1739                 RTE_PTYPE_INNER_L2_ETHER_QINQ,
1740                 RTE_PTYPE_INNER_L3_IPV4,
1741                 RTE_PTYPE_INNER_L3_IPV4_EXT,
1742                 RTE_PTYPE_INNER_L3_IPV6,
1743                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1744                 RTE_PTYPE_INNER_L4_FRAG,
1745                 RTE_PTYPE_INNER_L4_UDP,
1746                 RTE_PTYPE_INNER_L4_TCP,
1747                 RTE_PTYPE_INNER_L4_SCTP,
1748                 RTE_PTYPE_L2_ETHER,
1749                 RTE_PTYPE_L2_ETHER_VLAN,
1750                 RTE_PTYPE_L2_ETHER_QINQ,
1751                 RTE_PTYPE_L3_IPV4,
1752                 RTE_PTYPE_L3_IPV4_EXT,
1753                 RTE_PTYPE_L3_IPV6_EXT,
1754                 RTE_PTYPE_L3_IPV6,
1755                 RTE_PTYPE_L4_FRAG,
1756                 RTE_PTYPE_L4_UDP,
1757                 RTE_PTYPE_L4_TCP,
1758                 RTE_PTYPE_L4_SCTP,
1759         };
1760
1761         return ptypes;
1762 }
1763
1764 static int
1765 tap_flow_ctrl_get(struct rte_eth_dev *dev __rte_unused,
1766                   struct rte_eth_fc_conf *fc_conf)
1767 {
1768         fc_conf->mode = RTE_ETH_FC_NONE;
1769         return 0;
1770 }
1771
1772 static int
1773 tap_flow_ctrl_set(struct rte_eth_dev *dev __rte_unused,
1774                   struct rte_eth_fc_conf *fc_conf)
1775 {
1776         if (fc_conf->mode != RTE_ETH_FC_NONE)
1777                 return -ENOTSUP;
1778         return 0;
1779 }
1780
1781 /**
1782  * DPDK callback to update the RSS hash configuration.
1783  *
1784  * @param dev
1785  *   Pointer to Ethernet device structure.
1786  * @param[in] rss_conf
1787  *   RSS configuration data.
1788  *
1789  * @return
1790  *   0 on success, a negative errno value otherwise and rte_errno is set.
1791  */
1792 static int
1793 tap_rss_hash_update(struct rte_eth_dev *dev,
1794                 struct rte_eth_rss_conf *rss_conf)
1795 {
1796         if (rss_conf->rss_hf & TAP_RSS_HF_MASK) {
1797                 rte_errno = EINVAL;
1798                 return -rte_errno;
1799         }
1800         if (rss_conf->rss_key && rss_conf->rss_key_len) {
1801                 /*
1802                  * Currently TAP RSS key is hard coded
1803                  * and cannot be updated
1804                  */
1805                 TAP_LOG(ERR,
1806                         "port %u RSS key cannot be updated",
1807                         dev->data->port_id);
1808                 rte_errno = EINVAL;
1809                 return -rte_errno;
1810         }
1811         return 0;
1812 }
1813
1814 static int
1815 tap_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1816 {
1817         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1818
1819         return 0;
1820 }
1821
1822 static int
1823 tap_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1824 {
1825         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
1826
1827         return 0;
1828 }
1829
1830 static int
1831 tap_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1832 {
1833         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1834
1835         return 0;
1836 }
1837
1838 static int
1839 tap_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1840 {
1841         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
1842
1843         return 0;
1844 }
1845 static const struct eth_dev_ops ops = {
1846         .dev_start              = tap_dev_start,
1847         .dev_stop               = tap_dev_stop,
1848         .dev_close              = tap_dev_close,
1849         .dev_configure          = tap_dev_configure,
1850         .dev_infos_get          = tap_dev_info,
1851         .rx_queue_setup         = tap_rx_queue_setup,
1852         .tx_queue_setup         = tap_tx_queue_setup,
1853         .rx_queue_start         = tap_rx_queue_start,
1854         .tx_queue_start         = tap_tx_queue_start,
1855         .rx_queue_stop          = tap_rx_queue_stop,
1856         .tx_queue_stop          = tap_tx_queue_stop,
1857         .rx_queue_release       = tap_rx_queue_release,
1858         .tx_queue_release       = tap_tx_queue_release,
1859         .flow_ctrl_get          = tap_flow_ctrl_get,
1860         .flow_ctrl_set          = tap_flow_ctrl_set,
1861         .link_update            = tap_link_update,
1862         .dev_set_link_up        = tap_link_set_up,
1863         .dev_set_link_down      = tap_link_set_down,
1864         .promiscuous_enable     = tap_promisc_enable,
1865         .promiscuous_disable    = tap_promisc_disable,
1866         .allmulticast_enable    = tap_allmulti_enable,
1867         .allmulticast_disable   = tap_allmulti_disable,
1868         .mac_addr_set           = tap_mac_set,
1869         .mtu_set                = tap_mtu_set,
1870         .set_mc_addr_list       = tap_set_mc_addr_list,
1871         .stats_get              = tap_stats_get,
1872         .stats_reset            = tap_stats_reset,
1873         .dev_supported_ptypes_get = tap_dev_supported_ptypes_get,
1874         .rss_hash_update        = tap_rss_hash_update,
1875         .flow_ops_get           = tap_dev_flow_ops_get,
1876 };
1877
1878 static int
1879 eth_dev_tap_create(struct rte_vdev_device *vdev, const char *tap_name,
1880                    char *remote_iface, struct rte_ether_addr *mac_addr,
1881                    enum rte_tuntap_type type)
1882 {
1883         int numa_node = rte_socket_id();
1884         struct rte_eth_dev *dev;
1885         struct pmd_internals *pmd;
1886         struct pmd_process_private *process_private;
1887         const char *tuntap_name = tuntap_types[type];
1888         struct rte_eth_dev_data *data;
1889         struct ifreq ifr;
1890         int i;
1891
1892         TAP_LOG(DEBUG, "%s device on numa %u", tuntap_name, rte_socket_id());
1893
1894         dev = rte_eth_vdev_allocate(vdev, sizeof(*pmd));
1895         if (!dev) {
1896                 TAP_LOG(ERR, "%s Unable to allocate device struct",
1897                                 tuntap_name);
1898                 goto error_exit_nodev;
1899         }
1900
1901         process_private = (struct pmd_process_private *)
1902                 rte_zmalloc_socket(tap_name, sizeof(struct pmd_process_private),
1903                         RTE_CACHE_LINE_SIZE, dev->device->numa_node);
1904
1905         if (process_private == NULL) {
1906                 TAP_LOG(ERR, "Failed to alloc memory for process private");
1907                 return -1;
1908         }
1909         pmd = dev->data->dev_private;
1910         dev->process_private = process_private;
1911         pmd->dev = dev;
1912         strlcpy(pmd->name, tap_name, sizeof(pmd->name));
1913         pmd->type = type;
1914         pmd->ka_fd = -1;
1915         pmd->nlsk_fd = -1;
1916         pmd->gso_ctx_mp = NULL;
1917
1918         pmd->ioctl_sock = socket(AF_INET, SOCK_DGRAM, 0);
1919         if (pmd->ioctl_sock == -1) {
1920                 TAP_LOG(ERR,
1921                         "%s Unable to get a socket for management: %s",
1922                         tuntap_name, strerror(errno));
1923                 goto error_exit;
1924         }
1925
1926         /* Allocate interrupt instance */
1927         pmd->intr_handle = rte_intr_instance_alloc(RTE_INTR_INSTANCE_F_SHARED);
1928         if (pmd->intr_handle == NULL) {
1929                 TAP_LOG(ERR, "Failed to allocate intr handle");
1930                 goto error_exit;
1931         }
1932
1933         /* Setup some default values */
1934         data = dev->data;
1935         data->dev_private = pmd;
1936         data->dev_flags = RTE_ETH_DEV_INTR_LSC |
1937                                 RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1938         data->numa_node = numa_node;
1939
1940         data->dev_link = pmd_link;
1941         data->mac_addrs = &pmd->eth_addr;
1942         /* Set the number of RX and TX queues */
1943         data->nb_rx_queues = 0;
1944         data->nb_tx_queues = 0;
1945
1946         dev->dev_ops = &ops;
1947         dev->rx_pkt_burst = pmd_rx_burst;
1948         dev->tx_pkt_burst = pmd_tx_burst;
1949
1950         rte_intr_type_set(pmd->intr_handle, RTE_INTR_HANDLE_EXT);
1951         rte_intr_fd_set(pmd->intr_handle, -1);
1952         dev->intr_handle = pmd->intr_handle;
1953
1954         /* Presetup the fds to -1 as being not valid */
1955         for (i = 0; i < RTE_PMD_TAP_MAX_QUEUES; i++) {
1956                 process_private->rxq_fds[i] = -1;
1957                 process_private->txq_fds[i] = -1;
1958         }
1959
1960         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1961                 if (rte_is_zero_ether_addr(mac_addr))
1962                         rte_eth_random_addr((uint8_t *)&pmd->eth_addr);
1963                 else
1964                         rte_memcpy(&pmd->eth_addr, mac_addr, sizeof(*mac_addr));
1965         }
1966
1967         /*
1968          * Allocate a TUN device keep-alive file descriptor that will only be
1969          * closed when the TUN device itself is closed or removed.
1970          * This keep-alive file descriptor will guarantee that the TUN device
1971          * exists even when all of its queues are closed
1972          */
1973         pmd->ka_fd = tun_alloc(pmd, 1);
1974         if (pmd->ka_fd == -1) {
1975                 TAP_LOG(ERR, "Unable to create %s interface", tuntap_name);
1976                 goto error_exit;
1977         }
1978         TAP_LOG(DEBUG, "allocated %s", pmd->name);
1979
1980         ifr.ifr_mtu = dev->data->mtu;
1981         if (tap_ioctl(pmd, SIOCSIFMTU, &ifr, 1, LOCAL_AND_REMOTE) < 0)
1982                 goto error_exit;
1983
1984         if (pmd->type == ETH_TUNTAP_TYPE_TAP) {
1985                 memset(&ifr, 0, sizeof(struct ifreq));
1986                 ifr.ifr_hwaddr.sa_family = AF_LOCAL;
1987                 rte_memcpy(ifr.ifr_hwaddr.sa_data, &pmd->eth_addr,
1988                                 RTE_ETHER_ADDR_LEN);
1989                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0)
1990                         goto error_exit;
1991         }
1992
1993         /*
1994          * Set up everything related to rte_flow:
1995          * - netlink socket
1996          * - tap / remote if_index
1997          * - mandatory QDISCs
1998          * - rte_flow actual/implicit lists
1999          * - implicit rules
2000          */
2001         pmd->nlsk_fd = tap_nl_init(0);
2002         if (pmd->nlsk_fd == -1) {
2003                 TAP_LOG(WARNING, "%s: failed to create netlink socket.",
2004                         pmd->name);
2005                 goto disable_rte_flow;
2006         }
2007         pmd->if_index = if_nametoindex(pmd->name);
2008         if (!pmd->if_index) {
2009                 TAP_LOG(ERR, "%s: failed to get if_index.", pmd->name);
2010                 goto disable_rte_flow;
2011         }
2012         if (qdisc_create_multiq(pmd->nlsk_fd, pmd->if_index) < 0) {
2013                 TAP_LOG(ERR, "%s: failed to create multiq qdisc.",
2014                         pmd->name);
2015                 goto disable_rte_flow;
2016         }
2017         if (qdisc_create_ingress(pmd->nlsk_fd, pmd->if_index) < 0) {
2018                 TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
2019                         pmd->name);
2020                 goto disable_rte_flow;
2021         }
2022         LIST_INIT(&pmd->flows);
2023
2024         if (strlen(remote_iface)) {
2025                 pmd->remote_if_index = if_nametoindex(remote_iface);
2026                 if (!pmd->remote_if_index) {
2027                         TAP_LOG(ERR, "%s: failed to get %s if_index.",
2028                                 pmd->name, remote_iface);
2029                         goto error_remote;
2030                 }
2031                 strlcpy(pmd->remote_iface, remote_iface, RTE_ETH_NAME_MAX_LEN);
2032
2033                 /* Save state of remote device */
2034                 tap_ioctl(pmd, SIOCGIFFLAGS, &pmd->remote_initial_flags, 0, REMOTE_ONLY);
2035
2036                 /* Replicate remote MAC address */
2037                 if (tap_ioctl(pmd, SIOCGIFHWADDR, &ifr, 0, REMOTE_ONLY) < 0) {
2038                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
2039                                 pmd->name, pmd->remote_iface);
2040                         goto error_remote;
2041                 }
2042                 rte_memcpy(&pmd->eth_addr, ifr.ifr_hwaddr.sa_data,
2043                            RTE_ETHER_ADDR_LEN);
2044                 /* The desired MAC is already in ifreq after SIOCGIFHWADDR. */
2045                 if (tap_ioctl(pmd, SIOCSIFHWADDR, &ifr, 0, LOCAL_ONLY) < 0) {
2046                         TAP_LOG(ERR, "%s: failed to get %s MAC address.",
2047                                 pmd->name, remote_iface);
2048                         goto error_remote;
2049                 }
2050
2051                 /*
2052                  * Flush usually returns negative value because it tries to
2053                  * delete every QDISC (and on a running device, one QDISC at
2054                  * least is needed). Ignore negative return value.
2055                  */
2056                 qdisc_flush(pmd->nlsk_fd, pmd->remote_if_index);
2057                 if (qdisc_create_ingress(pmd->nlsk_fd,
2058                                          pmd->remote_if_index) < 0) {
2059                         TAP_LOG(ERR, "%s: failed to create ingress qdisc.",
2060                                 pmd->remote_iface);
2061                         goto error_remote;
2062                 }
2063                 LIST_INIT(&pmd->implicit_flows);
2064                 if (tap_flow_implicit_create(pmd, TAP_REMOTE_TX) < 0 ||
2065                     tap_flow_implicit_create(pmd, TAP_REMOTE_LOCAL_MAC) < 0 ||
2066                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCAST) < 0 ||
2067                     tap_flow_implicit_create(pmd, TAP_REMOTE_BROADCASTV6) < 0) {
2068                         TAP_LOG(ERR,
2069                                 "%s: failed to create implicit rules.",
2070                                 pmd->name);
2071                         goto error_remote;
2072                 }
2073         }
2074
2075         rte_eth_dev_probing_finish(dev);
2076         return 0;
2077
2078 disable_rte_flow:
2079         TAP_LOG(ERR, " Disabling rte flow support: %s(%d)",
2080                 strerror(errno), errno);
2081         if (strlen(remote_iface)) {
2082                 TAP_LOG(ERR, "Remote feature requires flow support.");
2083                 goto error_exit;
2084         }
2085         rte_eth_dev_probing_finish(dev);
2086         return 0;
2087
2088 error_remote:
2089         TAP_LOG(ERR, " Can't set up remote feature: %s(%d)",
2090                 strerror(errno), errno);
2091         tap_flow_implicit_flush(pmd, NULL);
2092
2093 error_exit:
2094         if (pmd->nlsk_fd != -1)
2095                 close(pmd->nlsk_fd);
2096         if (pmd->ka_fd != -1)
2097                 close(pmd->ka_fd);
2098         if (pmd->ioctl_sock != -1)
2099                 close(pmd->ioctl_sock);
2100         /* mac_addrs must not be freed alone because part of dev_private */
2101         dev->data->mac_addrs = NULL;
2102         rte_eth_dev_release_port(dev);
2103         rte_intr_instance_free(pmd->intr_handle);
2104
2105 error_exit_nodev:
2106         TAP_LOG(ERR, "%s Unable to initialize %s",
2107                 tuntap_name, rte_vdev_device_name(vdev));
2108
2109         return -EINVAL;
2110 }
2111
2112 /* make sure name is a possible Linux network device name */
2113 static bool
2114 is_valid_iface(const char *name)
2115 {
2116         if (*name == '\0')
2117                 return false;
2118
2119         if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
2120                 return false;
2121
2122         while (*name) {
2123                 if (*name == '/' || *name == ':' || isspace(*name))
2124                         return false;
2125                 name++;
2126         }
2127         return true;
2128 }
2129
2130 static int
2131 set_interface_name(const char *key __rte_unused,
2132                    const char *value,
2133                    void *extra_args)
2134 {
2135         char *name = (char *)extra_args;
2136
2137         if (value) {
2138                 if (!is_valid_iface(value)) {
2139                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2140                                 value);
2141                         return -1;
2142                 }
2143                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2144         } else {
2145                 /* use tap%d which causes kernel to choose next available */
2146                 strlcpy(name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2147         }
2148         return 0;
2149 }
2150
2151 static int
2152 set_remote_iface(const char *key __rte_unused,
2153                  const char *value,
2154                  void *extra_args)
2155 {
2156         char *name = (char *)extra_args;
2157
2158         if (value) {
2159                 if (!is_valid_iface(value)) {
2160                         TAP_LOG(ERR, "TAP invalid remote interface name (%s)",
2161                                 value);
2162                         return -1;
2163                 }
2164                 strlcpy(name, value, RTE_ETH_NAME_MAX_LEN);
2165         }
2166
2167         return 0;
2168 }
2169
2170 static int parse_user_mac(struct rte_ether_addr *user_mac,
2171                 const char *value)
2172 {
2173         unsigned int index = 0;
2174         char mac_temp[strlen(ETH_TAP_USR_MAC_FMT) + 1], *mac_byte = NULL;
2175
2176         if (user_mac == NULL || value == NULL)
2177                 return 0;
2178
2179         strlcpy(mac_temp, value, sizeof(mac_temp));
2180         mac_byte = strtok(mac_temp, ":");
2181
2182         while ((mac_byte != NULL) &&
2183                         (strlen(mac_byte) <= 2) &&
2184                         (strlen(mac_byte) == strspn(mac_byte,
2185                                         ETH_TAP_CMP_MAC_FMT))) {
2186                 user_mac->addr_bytes[index++] = strtoul(mac_byte, NULL, 16);
2187                 mac_byte = strtok(NULL, ":");
2188         }
2189
2190         return index;
2191 }
2192
2193 static int
2194 set_mac_type(const char *key __rte_unused,
2195              const char *value,
2196              void *extra_args)
2197 {
2198         struct rte_ether_addr *user_mac = extra_args;
2199
2200         if (!value)
2201                 return 0;
2202
2203         if (!strncasecmp(ETH_TAP_MAC_FIXED, value, strlen(ETH_TAP_MAC_FIXED))) {
2204                 static int iface_idx;
2205
2206                 /* fixed mac = 00:64:74:61:70:<iface_idx> */
2207                 memcpy((char *)user_mac->addr_bytes, "\0dtap",
2208                         RTE_ETHER_ADDR_LEN);
2209                 user_mac->addr_bytes[RTE_ETHER_ADDR_LEN - 1] =
2210                         iface_idx++ + '0';
2211                 goto success;
2212         }
2213
2214         if (parse_user_mac(user_mac, value) != 6)
2215                 goto error;
2216 success:
2217         TAP_LOG(DEBUG, "TAP user MAC param (%s)", value);
2218         return 0;
2219
2220 error:
2221         TAP_LOG(ERR, "TAP user MAC (%s) is not in format (%s|%s)",
2222                 value, ETH_TAP_MAC_FIXED, ETH_TAP_USR_MAC_FMT);
2223         return -1;
2224 }
2225
2226 /*
2227  * Open a TUN interface device. TUN PMD
2228  * 1) sets tap_type as false
2229  * 2) intakes iface as argument.
2230  * 3) as interface is virtual set speed to 10G
2231  */
2232 static int
2233 rte_pmd_tun_probe(struct rte_vdev_device *dev)
2234 {
2235         const char *name, *params;
2236         int ret;
2237         struct rte_kvargs *kvlist = NULL;
2238         char tun_name[RTE_ETH_NAME_MAX_LEN];
2239         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2240         struct rte_eth_dev *eth_dev;
2241
2242         name = rte_vdev_device_name(dev);
2243         params = rte_vdev_device_args(dev);
2244         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2245
2246         if (rte_eal_process_type() == RTE_PROC_SECONDARY &&
2247             strlen(params) == 0) {
2248                 eth_dev = rte_eth_dev_attach_secondary(name);
2249                 if (!eth_dev) {
2250                         TAP_LOG(ERR, "Failed to probe %s", name);
2251                         return -1;
2252                 }
2253                 eth_dev->dev_ops = &ops;
2254                 eth_dev->device = &dev->device;
2255                 rte_eth_dev_probing_finish(eth_dev);
2256                 return 0;
2257         }
2258
2259         /* use tun%d which causes kernel to choose next available */
2260         strlcpy(tun_name, DEFAULT_TUN_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2261
2262         if (params && (params[0] != '\0')) {
2263                 TAP_LOG(DEBUG, "parameters (%s)", params);
2264
2265                 kvlist = rte_kvargs_parse(params, valid_arguments);
2266                 if (kvlist) {
2267                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2268                                 ret = rte_kvargs_process(kvlist,
2269                                         ETH_TAP_IFACE_ARG,
2270                                         &set_interface_name,
2271                                         tun_name);
2272
2273                                 if (ret == -1)
2274                                         goto leave;
2275                         }
2276                 }
2277         }
2278         pmd_link.link_speed = RTE_ETH_SPEED_NUM_10G;
2279
2280         TAP_LOG(DEBUG, "Initializing pmd_tun for %s", name);
2281
2282         ret = eth_dev_tap_create(dev, tun_name, remote_iface, 0,
2283                                  ETH_TUNTAP_TYPE_TUN);
2284
2285 leave:
2286         if (ret == -1) {
2287                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2288                         name, tun_name);
2289         }
2290         rte_kvargs_free(kvlist);
2291
2292         return ret;
2293 }
2294
2295 /* Request queue file descriptors from secondary to primary. */
2296 static int
2297 tap_mp_attach_queues(const char *port_name, struct rte_eth_dev *dev)
2298 {
2299         int ret;
2300         struct timespec timeout = {.tv_sec = 1, .tv_nsec = 0};
2301         struct rte_mp_msg request, *reply;
2302         struct rte_mp_reply replies;
2303         struct ipc_queues *request_param = (struct ipc_queues *)request.param;
2304         struct ipc_queues *reply_param;
2305         struct pmd_process_private *process_private = dev->process_private;
2306         int queue, fd_iterator;
2307
2308         /* Prepare the request */
2309         memset(&request, 0, sizeof(request));
2310         strlcpy(request.name, TAP_MP_KEY, sizeof(request.name));
2311         strlcpy(request_param->port_name, port_name,
2312                 sizeof(request_param->port_name));
2313         request.len_param = sizeof(*request_param);
2314         /* Send request and receive reply */
2315         ret = rte_mp_request_sync(&request, &replies, &timeout);
2316         if (ret < 0 || replies.nb_received != 1) {
2317                 TAP_LOG(ERR, "Failed to request queues from primary: %d",
2318                         rte_errno);
2319                 return -1;
2320         }
2321         reply = &replies.msgs[0];
2322         reply_param = (struct ipc_queues *)reply->param;
2323         TAP_LOG(DEBUG, "Received IPC reply for %s", reply_param->port_name);
2324
2325         /* Attach the queues from received file descriptors */
2326         if (reply_param->rxq_count + reply_param->txq_count != reply->num_fds) {
2327                 TAP_LOG(ERR, "Unexpected number of fds received");
2328                 return -1;
2329         }
2330
2331         dev->data->nb_rx_queues = reply_param->rxq_count;
2332         dev->data->nb_tx_queues = reply_param->txq_count;
2333         fd_iterator = 0;
2334         for (queue = 0; queue < reply_param->rxq_count; queue++)
2335                 process_private->rxq_fds[queue] = reply->fds[fd_iterator++];
2336         for (queue = 0; queue < reply_param->txq_count; queue++)
2337                 process_private->txq_fds[queue] = reply->fds[fd_iterator++];
2338         free(reply);
2339         return 0;
2340 }
2341
2342 /* Send the queue file descriptors from the primary process to secondary. */
2343 static int
2344 tap_mp_sync_queues(const struct rte_mp_msg *request, const void *peer)
2345 {
2346         struct rte_eth_dev *dev;
2347         struct pmd_process_private *process_private;
2348         struct rte_mp_msg reply;
2349         const struct ipc_queues *request_param =
2350                 (const struct ipc_queues *)request->param;
2351         struct ipc_queues *reply_param =
2352                 (struct ipc_queues *)reply.param;
2353         uint16_t port_id;
2354         int queue;
2355         int ret;
2356
2357         /* Get requested port */
2358         TAP_LOG(DEBUG, "Received IPC request for %s", request_param->port_name);
2359         ret = rte_eth_dev_get_port_by_name(request_param->port_name, &port_id);
2360         if (ret) {
2361                 TAP_LOG(ERR, "Failed to get port id for %s",
2362                         request_param->port_name);
2363                 return -1;
2364         }
2365         dev = &rte_eth_devices[port_id];
2366         process_private = dev->process_private;
2367
2368         /* Fill file descriptors for all queues */
2369         reply.num_fds = 0;
2370         reply_param->rxq_count = 0;
2371         if (dev->data->nb_rx_queues + dev->data->nb_tx_queues >
2372                         RTE_MP_MAX_FD_NUM){
2373                 TAP_LOG(ERR, "Number of rx/tx queues exceeds max number of fds");
2374                 return -1;
2375         }
2376
2377         for (queue = 0; queue < dev->data->nb_rx_queues; queue++) {
2378                 reply.fds[reply.num_fds++] = process_private->rxq_fds[queue];
2379                 reply_param->rxq_count++;
2380         }
2381         RTE_ASSERT(reply_param->rxq_count == dev->data->nb_rx_queues);
2382
2383         reply_param->txq_count = 0;
2384         for (queue = 0; queue < dev->data->nb_tx_queues; queue++) {
2385                 reply.fds[reply.num_fds++] = process_private->txq_fds[queue];
2386                 reply_param->txq_count++;
2387         }
2388         RTE_ASSERT(reply_param->txq_count == dev->data->nb_tx_queues);
2389
2390         /* Send reply */
2391         strlcpy(reply.name, request->name, sizeof(reply.name));
2392         strlcpy(reply_param->port_name, request_param->port_name,
2393                 sizeof(reply_param->port_name));
2394         reply.len_param = sizeof(*reply_param);
2395         if (rte_mp_reply(&reply, peer) < 0) {
2396                 TAP_LOG(ERR, "Failed to reply an IPC request to sync queues");
2397                 return -1;
2398         }
2399         return 0;
2400 }
2401
2402 /* Open a TAP interface device.
2403  */
2404 static int
2405 rte_pmd_tap_probe(struct rte_vdev_device *dev)
2406 {
2407         const char *name, *params;
2408         int ret;
2409         struct rte_kvargs *kvlist = NULL;
2410         int speed;
2411         char tap_name[RTE_ETH_NAME_MAX_LEN];
2412         char remote_iface[RTE_ETH_NAME_MAX_LEN];
2413         struct rte_ether_addr user_mac = { .addr_bytes = {0} };
2414         struct rte_eth_dev *eth_dev;
2415         int tap_devices_count_increased = 0;
2416
2417         name = rte_vdev_device_name(dev);
2418         params = rte_vdev_device_args(dev);
2419
2420         if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
2421                 eth_dev = rte_eth_dev_attach_secondary(name);
2422                 if (!eth_dev) {
2423                         TAP_LOG(ERR, "Failed to probe %s", name);
2424                         return -1;
2425                 }
2426                 eth_dev->dev_ops = &ops;
2427                 eth_dev->device = &dev->device;
2428                 eth_dev->rx_pkt_burst = pmd_rx_burst;
2429                 eth_dev->tx_pkt_burst = pmd_tx_burst;
2430                 if (!rte_eal_primary_proc_alive(NULL)) {
2431                         TAP_LOG(ERR, "Primary process is missing");
2432                         return -1;
2433                 }
2434                 eth_dev->process_private = (struct pmd_process_private *)
2435                         rte_zmalloc_socket(name,
2436                                 sizeof(struct pmd_process_private),
2437                                 RTE_CACHE_LINE_SIZE,
2438                                 eth_dev->device->numa_node);
2439                 if (eth_dev->process_private == NULL) {
2440                         TAP_LOG(ERR,
2441                                 "Failed to alloc memory for process private");
2442                         return -1;
2443                 }
2444
2445                 ret = tap_mp_attach_queues(name, eth_dev);
2446                 if (ret != 0)
2447                         return -1;
2448                 rte_eth_dev_probing_finish(eth_dev);
2449                 return 0;
2450         }
2451
2452         speed = RTE_ETH_SPEED_NUM_10G;
2453
2454         /* use tap%d which causes kernel to choose next available */
2455         strlcpy(tap_name, DEFAULT_TAP_NAME "%d", RTE_ETH_NAME_MAX_LEN);
2456         memset(remote_iface, 0, RTE_ETH_NAME_MAX_LEN);
2457
2458         if (params && (params[0] != '\0')) {
2459                 TAP_LOG(DEBUG, "parameters (%s)", params);
2460
2461                 kvlist = rte_kvargs_parse(params, valid_arguments);
2462                 if (kvlist) {
2463                         if (rte_kvargs_count(kvlist, ETH_TAP_IFACE_ARG) == 1) {
2464                                 ret = rte_kvargs_process(kvlist,
2465                                                          ETH_TAP_IFACE_ARG,
2466                                                          &set_interface_name,
2467                                                          tap_name);
2468                                 if (ret == -1)
2469                                         goto leave;
2470                         }
2471
2472                         if (rte_kvargs_count(kvlist, ETH_TAP_REMOTE_ARG) == 1) {
2473                                 ret = rte_kvargs_process(kvlist,
2474                                                          ETH_TAP_REMOTE_ARG,
2475                                                          &set_remote_iface,
2476                                                          remote_iface);
2477                                 if (ret == -1)
2478                                         goto leave;
2479                         }
2480
2481                         if (rte_kvargs_count(kvlist, ETH_TAP_MAC_ARG) == 1) {
2482                                 ret = rte_kvargs_process(kvlist,
2483                                                          ETH_TAP_MAC_ARG,
2484                                                          &set_mac_type,
2485                                                          &user_mac);
2486                                 if (ret == -1)
2487                                         goto leave;
2488                         }
2489                 }
2490         }
2491         pmd_link.link_speed = speed;
2492
2493         TAP_LOG(DEBUG, "Initializing pmd_tap for %s", name);
2494
2495         /* Register IPC feed callback */
2496         if (!tap_devices_count) {
2497                 ret = rte_mp_action_register(TAP_MP_KEY, tap_mp_sync_queues);
2498                 if (ret < 0 && rte_errno != ENOTSUP) {
2499                         TAP_LOG(ERR, "tap: Failed to register IPC callback: %s",
2500                                 strerror(rte_errno));
2501                         goto leave;
2502                 }
2503         }
2504         tap_devices_count++;
2505         tap_devices_count_increased = 1;
2506         ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
2507                 ETH_TUNTAP_TYPE_TAP);
2508
2509 leave:
2510         if (ret == -1) {
2511                 TAP_LOG(ERR, "Failed to create pmd for %s as %s",
2512                         name, tap_name);
2513                 if (tap_devices_count_increased == 1) {
2514                         if (tap_devices_count == 1)
2515                                 rte_mp_action_unregister(TAP_MP_KEY);
2516                         tap_devices_count--;
2517                 }
2518         }
2519         rte_kvargs_free(kvlist);
2520
2521         return ret;
2522 }
2523
2524 /* detach a TUNTAP device.
2525  */
2526 static int
2527 rte_pmd_tap_remove(struct rte_vdev_device *dev)
2528 {
2529         struct rte_eth_dev *eth_dev = NULL;
2530
2531         /* find the ethdev entry */
2532         eth_dev = rte_eth_dev_allocated(rte_vdev_device_name(dev));
2533         if (!eth_dev)
2534                 return 0;
2535
2536         tap_dev_close(eth_dev);
2537         rte_eth_dev_release_port(eth_dev);
2538
2539         return 0;
2540 }
2541
2542 static struct rte_vdev_driver pmd_tun_drv = {
2543         .probe = rte_pmd_tun_probe,
2544         .remove = rte_pmd_tap_remove,
2545 };
2546
2547 static struct rte_vdev_driver pmd_tap_drv = {
2548         .probe = rte_pmd_tap_probe,
2549         .remove = rte_pmd_tap_remove,
2550 };
2551
2552 RTE_PMD_REGISTER_VDEV(net_tap, pmd_tap_drv);
2553 RTE_PMD_REGISTER_VDEV(net_tun, pmd_tun_drv);
2554 RTE_PMD_REGISTER_ALIAS(net_tap, eth_tap);
2555 RTE_PMD_REGISTER_PARAM_STRING(net_tun,
2556                               ETH_TAP_IFACE_ARG "=<string> ");
2557 RTE_PMD_REGISTER_PARAM_STRING(net_tap,
2558                               ETH_TAP_IFACE_ARG "=<string> "
2559                               ETH_TAP_MAC_ARG "=" ETH_TAP_MAC_ARG_FMT " "
2560                               ETH_TAP_REMOTE_ARG "=<string>");
2561 RTE_LOG_REGISTER_DEFAULT(tap_logtype, NOTICE);