net/tap.c

   1 /*
   2  * QEMU System Emulator
   3  *
   4  * Copyright (c) 2003-2008 Fabrice Bellard
   5  * Copyright (c) 2009 Red Hat, Inc.
   6  *
   7  * Permission is hereby granted, free of charge, to any person obtaining a copy
   8  * of this software and associated documentation files (the "Software"), to deal
   9  * in the Software without restriction, including without limitation the rights
  10  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  11  * copies of the Software, and to permit persons to whom the Software is
  12  * furnished to do so, subject to the following conditions:
  13  *
  14  * The above copyright notice and this permission notice shall be included in
  15  * all copies or substantial portions of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  23  * THE SOFTWARE.
  24  */
  25
  26 #include "qemu/osdep.h"
  27 #include "tap_int.h"
  28
  29
  30 #include <sys/ioctl.h>
  31 #include <sys/wait.h>
  32 #include <sys/socket.h>
  33 #include <net/if.h>
  34
  35 #include "net/net.h"
  36 #include "clients.h"
  37 #include "monitor/monitor.h"
  38 #include "sysemu/sysemu.h"
  39 #include "qapi/error.h"
  40 #include "qemu-common.h"
  41 #include "qemu/cutils.h"
  42 #include "qemu/error-report.h"
  43
  44 #include "net/tap.h"
  45
  46 #include "net/vhost_net.h"
  47
  48 typedef struct TAPState {
  49     NetClientState nc;
  50     int fd;
  51     char down_script[1024];
  52     char down_script_arg[128];
  53     uint8_t buf[NET_BUFSIZE];
  54     bool read_poll;
  55     bool write_poll;
  56     bool using_vnet_hdr;
  57     bool has_ufo;
  58     bool enabled;
  59     VHostNetState *vhost_net;
  60     unsigned host_vnet_hdr_len;
  61     Notifier exit;
  62 } TAPState;
  63
  64 static void launch_script(const char *setup_script, const char *ifname,
  65                           int fd, Error **errp);
  66
  67 static void tap_send(void *opaque);
  68 static void tap_writable(void *opaque);
  69
  70 static void tap_update_fd_handler(TAPState *s)
  71 {
  72     qemu_set_fd_handler(s->fd,
  73                         s->read_poll && s->enabled ? tap_send : NULL,
  74                         s->write_poll && s->enabled ? tap_writable : NULL,
  75                         s);
  76 }
  77
  78 static void tap_read_poll(TAPState *s, bool enable)
  79 {
  80     s->read_poll = enable;
  81     tap_update_fd_handler(s);
  82 }
  83
  84 static void tap_write_poll(TAPState *s, bool enable)
  85 {
  86     s->write_poll = enable;
  87     tap_update_fd_handler(s);
  88 }
  89
  90 static void tap_writable(void *opaque)
  91 {
  92     TAPState *s = opaque;
  93
  94     tap_write_poll(s, false);
  95
  96     qemu_flush_queued_packets(&s->nc);
  97 }
  98
  99 static ssize_t tap_write_packet(TAPState *s, const struct iovec *iov, int iovcnt)
 100 {
 101     ssize_t len;
 102
 103     do {
 104         len = writev(s->fd, iov, iovcnt);
 105     } while (len == -1 && errno == EINTR);
 106
 107     if (len == -1 && errno == EAGAIN) {
 108         tap_write_poll(s, true);
 109         return 0;
 110     }
 111
 112     return len;
 113 }
 114
 115 static ssize_t tap_receive_iov(NetClientState *nc, const struct iovec *iov,
 116                                int iovcnt)
 117 {
 118     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 119     const struct iovec *iovp = iov;
 120     struct iovec iov_copy[iovcnt + 1];
 121     struct virtio_net_hdr_mrg_rxbuf hdr = { };
 122
 123     if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
 124         iov_copy[0].iov_base = &hdr;
 125         iov_copy[0].iov_len =  s->host_vnet_hdr_len;
 126         memcpy(&iov_copy[1], iov, iovcnt * sizeof(*iov));
 127         iovp = iov_copy;
 128         iovcnt++;
 129     }
 130
 131     return tap_write_packet(s, iovp, iovcnt);
 132 }
 133
 134 static ssize_t tap_receive_raw(NetClientState *nc, const uint8_t *buf, size_t size)
 135 {
 136     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 137     struct iovec iov[2];
 138     int iovcnt = 0;
 139     struct virtio_net_hdr_mrg_rxbuf hdr = { };
 140
 141     if (s->host_vnet_hdr_len) {
 142         iov[iovcnt].iov_base = &hdr;
 143         iov[iovcnt].iov_len  = s->host_vnet_hdr_len;
 144         iovcnt++;
 145     }
 146
 147     iov[iovcnt].iov_base = (char *)buf;
 148     iov[iovcnt].iov_len  = size;
 149     iovcnt++;
 150
 151     return tap_write_packet(s, iov, iovcnt);
 152 }
 153
 154 static ssize_t tap_receive(NetClientState *nc, const uint8_t *buf, size_t size)
 155 {
 156     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 157     struct iovec iov[1];
 158
 159     if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
 160         return tap_receive_raw(nc, buf, size);
 161     }
 162
 163     iov[0].iov_base = (char *)buf;
 164     iov[0].iov_len  = size;
 165
 166     return tap_write_packet(s, iov, 1);
 167 }
 168
 169 #ifndef __sun__
 170 ssize_t tap_read_packet(int tapfd, uint8_t *buf, int maxlen)
 171 {
 172     return read(tapfd, buf, maxlen);
 173 }
 174 #endif
 175
 176 static void tap_send_completed(NetClientState *nc, ssize_t len)
 177 {
 178     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 179     tap_read_poll(s, true);
 180 }
 181
 182 static void tap_send(void *opaque)
 183 {
 184     TAPState *s = opaque;
 185     int size;
 186     int packets = 0;
 187
 188     while (true) {
 189         uint8_t *buf = s->buf;
 190
 191         size = tap_read_packet(s->fd, s->buf, sizeof(s->buf));
 192         if (size <= 0) {
 193             break;
 194         }
 195
 196         if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
 197             buf  += s->host_vnet_hdr_len;
 198             size -= s->host_vnet_hdr_len;
 199         }
 200
 201         size = qemu_send_packet_async(&s->nc, buf, size, tap_send_completed);
 202         if (size == 0) {
 203             tap_read_poll(s, false);
 204             break;
 205         } else if (size < 0) {
 206             break;
 207         }
 208
 209         /*
 210          * When the host keeps receiving more packets while tap_send() is
 211          * running we can hog the QEMU global mutex.  Limit the number of
 212          * packets that are processed per tap_send() callback to prevent
 213          * stalling the guest.
 214          */
 215         packets++;
 216         if (packets >= 50) {
 217             break;
 218         }
 219     }
 220 }
 221
 222 static bool tap_has_ufo(NetClientState *nc)
 223 {
 224     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 225
 226     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
 227
 228     return s->has_ufo;
 229 }
 230
 231 static bool tap_has_vnet_hdr(NetClientState *nc)
 232 {
 233     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 234
 235     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
 236
 237     return !!s->host_vnet_hdr_len;
 238 }
 239
 240 static bool tap_has_vnet_hdr_len(NetClientState *nc, int len)
 241 {
 242     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 243
 244     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
 245
 246     return !!tap_probe_vnet_hdr_len(s->fd, len);
 247 }
 248
 249 static void tap_set_vnet_hdr_len(NetClientState *nc, int len)
 250 {
 251     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 252
 253     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
 254     assert(len == sizeof(struct virtio_net_hdr_mrg_rxbuf) ||
 255            len == sizeof(struct virtio_net_hdr));
 256
 257     tap_fd_set_vnet_hdr_len(s->fd, len);
 258     s->host_vnet_hdr_len = len;
 259 }
 260
 261 static void tap_using_vnet_hdr(NetClientState *nc, bool using_vnet_hdr)
 262 {
 263     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 264
 265     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
 266     assert(!!s->host_vnet_hdr_len == using_vnet_hdr);
 267
 268     s->using_vnet_hdr = using_vnet_hdr;
 269 }
 270
 271 static int tap_set_vnet_le(NetClientState *nc, bool is_le)
 272 {
 273     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 274
 275     return tap_fd_set_vnet_le(s->fd, is_le);
 276 }
 277
 278 static int tap_set_vnet_be(NetClientState *nc, bool is_be)
 279 {
 280     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 281
 282     return tap_fd_set_vnet_be(s->fd, is_be);
 283 }
 284
 285 static void tap_set_offload(NetClientState *nc, int csum, int tso4,
 286                      int tso6, int ecn, int ufo)
 287 {
 288     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 289     if (s->fd < 0) {
 290         return;
 291     }
 292
 293     tap_fd_set_offload(s->fd, csum, tso4, tso6, ecn, ufo);
 294 }
 295
 296 static void tap_exit_notify(Notifier *notifier, void *data)
 297 {
 298     TAPState *s = container_of(notifier, TAPState, exit);
 299     Error *err = NULL;
 300
 301     if (s->down_script[0]) {
 302         launch_script(s->down_script, s->down_script_arg, s->fd, &err);
 303         if (err) {
 304             error_report_err(err);
 305         }
 306     }
 307 }
 308
 309 static void tap_cleanup(NetClientState *nc)
 310 {
 311     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 312
 313     if (s->vhost_net) {
 314         vhost_net_cleanup(s->vhost_net);
 315         s->vhost_net = NULL;
 316     }
 317
 318     qemu_purge_queued_packets(nc);
 319
 320     tap_exit_notify(&s->exit, NULL);
 321     qemu_remove_exit_notifier(&s->exit);
 322
 323     tap_read_poll(s, false);
 324     tap_write_poll(s, false);
 325     close(s->fd);
 326     s->fd = -1;
 327 }
 328
 329 static void tap_poll(NetClientState *nc, bool enable)
 330 {
 331     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 332     tap_read_poll(s, enable);
 333     tap_write_poll(s, enable);
 334 }
 335
 336 int tap_get_fd(NetClientState *nc)
 337 {
 338     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 339     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
 340     return s->fd;
 341 }
 342
 343 /* fd support */
 344
 345 static NetClientInfo net_tap_info = {
 346     .type = NET_CLIENT_OPTIONS_KIND_TAP,
 347     .size = sizeof(TAPState),
 348     .receive = tap_receive,
 349     .receive_raw = tap_receive_raw,
 350     .receive_iov = tap_receive_iov,
 351     .poll = tap_poll,
 352     .cleanup = tap_cleanup,
 353     .has_ufo = tap_has_ufo,
 354     .has_vnet_hdr = tap_has_vnet_hdr,
 355     .has_vnet_hdr_len = tap_has_vnet_hdr_len,
 356     .using_vnet_hdr = tap_using_vnet_hdr,
 357     .set_offload = tap_set_offload,
 358     .set_vnet_hdr_len = tap_set_vnet_hdr_len,
 359     .set_vnet_le = tap_set_vnet_le,
 360     .set_vnet_be = tap_set_vnet_be,
 361 };
 362
 363 static TAPState *net_tap_fd_init(NetClientState *peer,
 364                                  const char *model,
 365                                  const char *name,
 366                                  int fd,
 367                                  int vnet_hdr)
 368 {
 369     NetClientState *nc;
 370     TAPState *s;
 371
 372     nc = qemu_new_net_client(&net_tap_info, peer, model, name);
 373
 374     s = DO_UPCAST(TAPState, nc, nc);
 375
 376     s->fd = fd;
 377     s->host_vnet_hdr_len = vnet_hdr ? sizeof(struct virtio_net_hdr) : 0;
 378     s->using_vnet_hdr = false;
 379     s->has_ufo = tap_probe_has_ufo(s->fd);
 380     s->enabled = true;
 381     tap_set_offload(&s->nc, 0, 0, 0, 0, 0);
 382     /*
 383      * Make sure host header length is set correctly in tap:
 384      * it might have been modified by another instance of qemu.
 385      */
 386     if (tap_probe_vnet_hdr_len(s->fd, s->host_vnet_hdr_len)) {
 387         tap_fd_set_vnet_hdr_len(s->fd, s->host_vnet_hdr_len);
 388     }
 389     tap_read_poll(s, true);
 390     s->vhost_net = NULL;
 391
 392     s->exit.notify = tap_exit_notify;
 393     qemu_add_exit_notifier(&s->exit);
 394
 395     return s;
 396 }
 397
 398 static void launch_script(const char *setup_script, const char *ifname,
 399                           int fd, Error **errp)
 400 {
 401     int pid, status;
 402     char *args[3];
 403     char **parg;
 404
 405     /* try to launch network script */
 406     pid = fork();
 407     if (pid < 0) {
 408         error_setg_errno(errp, errno, "could not launch network script %s",
 409                          setup_script);
 410         return;
 411     }
 412     if (pid == 0) {
 413         int open_max = sysconf(_SC_OPEN_MAX), i;
 414
 415         for (i = 3; i < open_max; i++) {
 416             if (i != fd) {
 417                 close(i);
 418             }
 419         }
 420         parg = args;
 421         *parg++ = (char *)setup_script;
 422         *parg++ = (char *)ifname;
 423         *parg = NULL;
 424         execv(setup_script, args);
 425         _exit(1);
 426     } else {
 427         while (waitpid(pid, &status, 0) != pid) {
 428             /* loop */
 429         }
 430
 431         if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
 432             return;
 433         }
 434         error_setg(errp, "network script %s failed with status %d",
 435                    setup_script, status);
 436     }
 437 }
 438
 439 static int recv_fd(int c)
 440 {
 441     int fd;
 442     uint8_t msgbuf[CMSG_SPACE(sizeof(fd))];
 443     struct msghdr msg = {
 444         .msg_control = msgbuf,
 445         .msg_controllen = sizeof(msgbuf),
 446     };
 447     struct cmsghdr *cmsg;
 448     struct iovec iov;
 449     uint8_t req[1];
 450     ssize_t len;
 451
 452     cmsg = CMSG_FIRSTHDR(&msg);
 453     cmsg->cmsg_level = SOL_SOCKET;
 454     cmsg->cmsg_type = SCM_RIGHTS;
 455     cmsg->cmsg_len = CMSG_LEN(sizeof(fd));
 456     msg.msg_controllen = cmsg->cmsg_len;
 457
 458     iov.iov_base = req;
 459     iov.iov_len = sizeof(req);
 460
 461     msg.msg_iov = &iov;
 462     msg.msg_iovlen = 1;
 463
 464     len = recvmsg(c, &msg, 0);
 465     if (len > 0) {
 466         memcpy(&fd, CMSG_DATA(cmsg), sizeof(fd));
 467         return fd;
 468     }
 469
 470     return len;
 471 }
 472
 473 static int net_bridge_run_helper(const char *helper, const char *bridge,
 474                                  Error **errp)
 475 {
 476     sigset_t oldmask, mask;
 477     int pid, status;
 478     char *args[5];
 479     char **parg;
 480     int sv[2];
 481
 482     sigemptyset(&mask);
 483     sigaddset(&mask, SIGCHLD);
 484     sigprocmask(SIG_BLOCK, &mask, &oldmask);
 485
 486     if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) {
 487         error_setg_errno(errp, errno, "socketpair() failed");
 488         return -1;
 489     }
 490
 491     /* try to launch bridge helper */
 492     pid = fork();
 493     if (pid < 0) {
 494         error_setg_errno(errp, errno, "Can't fork bridge helper");
 495         return -1;
 496     }
 497     if (pid == 0) {
 498         int open_max = sysconf(_SC_OPEN_MAX), i;
 499         char fd_buf[6+10];
 500         char br_buf[6+IFNAMSIZ] = {0};
 501         char helper_cmd[PATH_MAX + sizeof(fd_buf) + sizeof(br_buf) + 15];
 502
 503         for (i = 3; i < open_max; i++) {
 504             if (i != sv[1]) {
 505                 close(i);
 506             }
 507         }
 508
 509         snprintf(fd_buf, sizeof(fd_buf), "%s%d", "--fd=", sv[1]);
 510
 511         if (strrchr(helper, ' ') || strrchr(helper, '\t')) {
 512             /* assume helper is a command */
 513
 514             if (strstr(helper, "--br=") == NULL) {
 515                 snprintf(br_buf, sizeof(br_buf), "%s%s", "--br=", bridge);
 516             }
 517
 518             snprintf(helper_cmd, sizeof(helper_cmd), "%s %s %s %s",
 519                      helper, "--use-vnet", fd_buf, br_buf);
 520
 521             parg = args;
 522             *parg++ = (char *)"sh";
 523             *parg++ = (char *)"-c";
 524             *parg++ = helper_cmd;
 525             *parg++ = NULL;
 526
 527             execv("/bin/sh", args);
 528         } else {
 529             /* assume helper is just the executable path name */
 530
 531             snprintf(br_buf, sizeof(br_buf), "%s%s", "--br=", bridge);
 532
 533             parg = args;
 534             *parg++ = (char *)helper;
 535             *parg++ = (char *)"--use-vnet";
 536             *parg++ = fd_buf;
 537             *parg++ = br_buf;
 538             *parg++ = NULL;
 539
 540             execv(helper, args);
 541         }
 542         _exit(1);
 543
 544     } else {
 545         int fd;
 546         int saved_errno;
 547
 548         close(sv[1]);
 549
 550         do {
 551             fd = recv_fd(sv[0]);
 552         } while (fd == -1 && errno == EINTR);
 553         saved_errno = errno;
 554
 555         close(sv[0]);
 556
 557         while (waitpid(pid, &status, 0) != pid) {
 558             /* loop */
 559         }
 560         sigprocmask(SIG_SETMASK, &oldmask, NULL);
 561         if (fd < 0) {
 562             error_setg_errno(errp, saved_errno,
 563                              "failed to recv file descriptor");
 564             return -1;
 565         }
 566         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 567             error_setg(errp, "bridge helper failed");
 568             return -1;
 569         }
 570         return fd;
 571     }
 572 }
 573
 574 int net_init_bridge(const NetClientOptions *opts, const char *name,
 575                     NetClientState *peer, Error **errp)
 576 {
 577     const NetdevBridgeOptions *bridge;
 578     const char *helper, *br;
 579     TAPState *s;
 580     int fd, vnet_hdr;
 581
 582     assert(opts->type == NET_CLIENT_OPTIONS_KIND_BRIDGE);
 583     bridge = opts->u.bridge.data;
 584
 585     helper = bridge->has_helper ? bridge->helper : DEFAULT_BRIDGE_HELPER;
 586     br     = bridge->has_br     ? bridge->br     : DEFAULT_BRIDGE_INTERFACE;
 587
 588     fd = net_bridge_run_helper(helper, br, errp);
 589     if (fd == -1) {
 590         return -1;
 591     }
 592
 593     fcntl(fd, F_SETFL, O_NONBLOCK);
 594     vnet_hdr = tap_probe_vnet_hdr(fd);
 595     s = net_tap_fd_init(peer, "bridge", name, fd, vnet_hdr);
 596
 597     snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s,br=%s", helper,
 598              br);
 599
 600     return 0;
 601 }
 602
 603 static int net_tap_init(const NetdevTapOptions *tap, int *vnet_hdr,
 604                         const char *setup_script, char *ifname,
 605                         size_t ifname_sz, int mq_required, Error **errp)
 606 {
 607     Error *err = NULL;
 608     int fd, vnet_hdr_required;
 609
 610     if (tap->has_vnet_hdr) {
 611         *vnet_hdr = tap->vnet_hdr;
 612         vnet_hdr_required = *vnet_hdr;
 613     } else {
 614         *vnet_hdr = 1;
 615         vnet_hdr_required = 0;
 616     }
 617
 618     TFR(fd = tap_open(ifname, ifname_sz, vnet_hdr, vnet_hdr_required,
 619                       mq_required, errp));
 620     if (fd < 0) {
 621         return -1;
 622     }
 623
 624     if (setup_script &&
 625         setup_script[0] != '\0' &&
 626         strcmp(setup_script, "no") != 0) {
 627         launch_script(setup_script, ifname, fd, &err);
 628         if (err) {
 629             error_propagate(errp, err);
 630             close(fd);
 631             return -1;
 632         }
 633     }
 634
 635     return fd;
 636 }
 637
 638 #define MAX_TAP_QUEUES 1024
 639
 640 static void net_init_tap_one(const NetdevTapOptions *tap, NetClientState *peer,
 641                              const char *model, const char *name,
 642                              const char *ifname, const char *script,
 643                              const char *downscript, const char *vhostfdname,
 644                              int vnet_hdr, int fd, Error **errp)
 645 {
 646     Error *err = NULL;
 647     TAPState *s = net_tap_fd_init(peer, model, name, fd, vnet_hdr);
 648     int vhostfd;
 649
 650     tap_set_sndbuf(s->fd, tap, &err);
 651     if (err) {
 652         error_propagate(errp, err);
 653         return;
 654     }
 655
 656     if (tap->has_fd || tap->has_fds) {
 657         snprintf(s->nc.info_str, sizeof(s->nc.info_str), "fd=%d", fd);
 658     } else if (tap->has_helper) {
 659         snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s",
 660                  tap->helper);
 661     } else {
 662         snprintf(s->nc.info_str, sizeof(s->nc.info_str),
 663                  "ifname=%s,script=%s,downscript=%s", ifname, script,
 664                  downscript);
 665
 666         if (strcmp(downscript, "no") != 0) {
 667             snprintf(s->down_script, sizeof(s->down_script), "%s", downscript);
 668             snprintf(s->down_script_arg, sizeof(s->down_script_arg),
 669                      "%s", ifname);
 670         }
 671     }
 672
 673     if (tap->has_vhost ? tap->vhost :
 674         vhostfdname || (tap->has_vhostforce && tap->vhostforce)) {
 675         VhostNetOptions options;
 676
 677         options.backend_type = VHOST_BACKEND_TYPE_KERNEL;
 678         options.net_backend = &s->nc;
 679         if (tap->has_poll_us) {
 680             options.busyloop_timeout = tap->poll_us;
 681         } else {
 682             options.busyloop_timeout = 0;
 683         }
 684
 685         if (vhostfdname) {
 686             vhostfd = monitor_fd_param(cur_mon, vhostfdname, &err);
 687             if (vhostfd == -1) {
 688                 error_propagate(errp, err);
 689                 return;
 690             }
 691         } else {
 692             vhostfd = open("/dev/vhost-net", O_RDWR);
 693             if (vhostfd < 0) {
 694                 error_setg_errno(errp, errno,
 695                                  "tap: open vhost char device failed");
 696                 return;
 697             }
 698         }
 699         options.opaque = (void *)(uintptr_t)vhostfd;
 700
 701         s->vhost_net = vhost_net_init(&options);
 702         if (!s->vhost_net) {
 703             error_setg(errp,
 704                        "vhost-net requested but could not be initialized");
 705             return;
 706         }
 707     } else if (vhostfdname) {
 708         error_setg(errp, "vhostfd(s)= is not valid without vhost");
 709     }
 710 }
 711
 712 static int get_fds(char *str, char *fds[], int max)
 713 {
 714     char *ptr = str, *this;
 715     size_t len = strlen(str);
 716     int i = 0;
 717
 718     while (i < max && ptr < str + len) {
 719         this = strchr(ptr, ':');
 720
 721         if (this == NULL) {
 722             fds[i] = g_strdup(ptr);
 723         } else {
 724             fds[i] = g_strndup(ptr, this - ptr);
 725         }
 726
 727         i++;
 728         if (this == NULL) {
 729             break;
 730         } else {
 731             ptr = this + 1;
 732         }
 733     }
 734
 735     return i;
 736 }
 737
 738 int net_init_tap(const NetClientOptions *opts, const char *name,
 739                  NetClientState *peer, Error **errp)
 740 {
 741     const NetdevTapOptions *tap;
 742     int fd, vnet_hdr = 0, i = 0, queues;
 743     /* for the no-fd, no-helper case */
 744     const char *script = NULL; /* suppress wrong "uninit'd use" gcc warning */
 745     const char *downscript = NULL;
 746     Error *err = NULL;
 747     const char *vhostfdname;
 748     char ifname[128];
 749
 750     assert(opts->type == NET_CLIENT_OPTIONS_KIND_TAP);
 751     tap = opts->u.tap.data;
 752     queues = tap->has_queues ? tap->queues : 1;
 753     vhostfdname = tap->has_vhostfd ? tap->vhostfd : NULL;
 754
 755     /* QEMU vlans does not support multiqueue tap, in this case peer is set.
 756      * For -netdev, peer is always NULL. */
 757     if (peer && (tap->has_queues || tap->has_fds || tap->has_vhostfds)) {
 758         error_setg(errp, "Multiqueue tap cannot be used with QEMU vlans");
 759         return -1;
 760     }
 761
 762     if (tap->has_fd) {
 763         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
 764             tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
 765             tap->has_fds || tap->has_vhostfds) {
 766             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 767                        "helper=, queues=, fds=, and vhostfds= "
 768                        "are invalid with fd=");
 769             return -1;
 770         }
 771
 772         fd = monitor_fd_param(cur_mon, tap->fd, &err);
 773         if (fd == -1) {
 774             error_propagate(errp, err);
 775             return -1;
 776         }
 777
 778         fcntl(fd, F_SETFL, O_NONBLOCK);
 779
 780         vnet_hdr = tap_probe_vnet_hdr(fd);
 781
 782         net_init_tap_one(tap, peer, "tap", name, NULL,
 783                          script, downscript,
 784                          vhostfdname, vnet_hdr, fd, &err);
 785         if (err) {
 786             error_propagate(errp, err);
 787             return -1;
 788         }
 789     } else if (tap->has_fds) {
 790         char **fds = g_new(char *, MAX_TAP_QUEUES);
 791         char **vhost_fds = g_new(char *, MAX_TAP_QUEUES);
 792         int nfds, nvhosts;
 793
 794         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
 795             tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
 796             tap->has_vhostfd) {
 797             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 798                        "helper=, queues=, and vhostfd= "
 799                        "are invalid with fds=");
 800             return -1;
 801         }
 802
 803         nfds = get_fds(tap->fds, fds, MAX_TAP_QUEUES);
 804         if (tap->has_vhostfds) {
 805             nvhosts = get_fds(tap->vhostfds, vhost_fds, MAX_TAP_QUEUES);
 806             if (nfds != nvhosts) {
 807                 error_setg(errp, "The number of fds passed does not match "
 808                            "the number of vhostfds passed");
 809                 return -1;
 810             }
 811         }
 812
 813         for (i = 0; i < nfds; i++) {
 814             fd = monitor_fd_param(cur_mon, fds[i], &err);
 815             if (fd == -1) {
 816                 error_propagate(errp, err);
 817                 return -1;
 818             }
 819
 820             fcntl(fd, F_SETFL, O_NONBLOCK);
 821
 822             if (i == 0) {
 823                 vnet_hdr = tap_probe_vnet_hdr(fd);
 824             } else if (vnet_hdr != tap_probe_vnet_hdr(fd)) {
 825                 error_setg(errp,
 826                            "vnet_hdr not consistent across given tap fds");
 827                 return -1;
 828             }
 829
 830             net_init_tap_one(tap, peer, "tap", name, ifname,
 831                              script, downscript,
 832                              tap->has_vhostfds ? vhost_fds[i] : NULL,
 833                              vnet_hdr, fd, &err);
 834             if (err) {
 835                 error_propagate(errp, err);
 836                 return -1;
 837             }
 838         }
 839         g_free(fds);
 840         g_free(vhost_fds);
 841     } else if (tap->has_helper) {
 842         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
 843             tap->has_vnet_hdr || tap->has_queues || tap->has_vhostfds) {
 844             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 845                        "queues=, and vhostfds= are invalid with helper=");
 846             return -1;
 847         }
 848
 849         fd = net_bridge_run_helper(tap->helper, DEFAULT_BRIDGE_INTERFACE,
 850                                    errp);
 851         if (fd == -1) {
 852             return -1;
 853         }
 854
 855         fcntl(fd, F_SETFL, O_NONBLOCK);
 856         vnet_hdr = tap_probe_vnet_hdr(fd);
 857
 858         net_init_tap_one(tap, peer, "bridge", name, ifname,
 859                          script, downscript, vhostfdname,
 860                          vnet_hdr, fd, &err);
 861         if (err) {
 862             error_propagate(errp, err);
 863             close(fd);
 864             return -1;
 865         }
 866     } else {
 867         if (tap->has_vhostfds) {
 868             error_setg(errp, "vhostfds= is invalid if fds= wasn't specified");
 869             return -1;
 870         }
 871         script = tap->has_script ? tap->script : DEFAULT_NETWORK_SCRIPT;
 872         downscript = tap->has_downscript ? tap->downscript :
 873             DEFAULT_NETWORK_DOWN_SCRIPT;
 874
 875         if (tap->has_ifname) {
 876             pstrcpy(ifname, sizeof ifname, tap->ifname);
 877         } else {
 878             ifname[0] = '\0';
 879         }
 880
 881         for (i = 0; i < queues; i++) {
 882             fd = net_tap_init(tap, &vnet_hdr, i >= 1 ? "no" : script,
 883                               ifname, sizeof ifname, queues > 1, errp);
 884             if (fd == -1) {
 885                 return -1;
 886             }
 887
 888             if (queues > 1 && i == 0 && !tap->has_ifname) {
 889                 if (tap_fd_get_ifname(fd, ifname)) {
 890                     error_setg(errp, "Fail to get ifname");
 891                     close(fd);
 892                     return -1;
 893                 }
 894             }
 895
 896             net_init_tap_one(tap, peer, "tap", name, ifname,
 897                              i >= 1 ? "no" : script,
 898                              i >= 1 ? "no" : downscript,
 899                              vhostfdname, vnet_hdr, fd, &err);
 900             if (err) {
 901                 error_propagate(errp, err);
 902                 close(fd);
 903                 return -1;
 904             }
 905         }
 906     }
 907
 908     return 0;
 909 }
 910
 911 VHostNetState *tap_get_vhost_net(NetClientState *nc)
 912 {
 913     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 914     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
 915     return s->vhost_net;
 916 }
 917
 918 int tap_enable(NetClientState *nc)
 919 {
 920     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 921     int ret;
 922
 923     if (s->enabled) {
 924         return 0;
 925     } else {
 926         ret = tap_fd_enable(s->fd);
 927         if (ret == 0) {
 928             s->enabled = true;
 929             tap_update_fd_handler(s);
 930         }
 931         return ret;
 932     }
 933 }
 934
 935 int tap_disable(NetClientState *nc)
 936 {
 937     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 938     int ret;
 939
 940     if (s->enabled == 0) {
 941         return 0;
 942     } else {
 943         ret = tap_fd_disable(s->fd);
 944         if (ret == 0) {
 945             qemu_purge_queued_packets(nc);
 946             s->enabled = false;
 947             tap_update_fd_handler(s);
 948         }
 949         return ret;
 950     }
 951 }