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 #include "qemu/sockets.h"
  44
  45 #include "net/tap.h"
  46
  47 #include "net/vhost_net.h"
  48
  49 typedef struct TAPState {
  50     NetClientState nc;
  51     int fd;
  52     char down_script[1024];
  53     char down_script_arg[128];
  54     uint8_t buf[NET_BUFSIZE];
  55     bool read_poll;
  56     bool write_poll;
  57     bool using_vnet_hdr;
  58     bool has_ufo;
  59     bool enabled;
  60     VHostNetState *vhost_net;
  61     unsigned host_vnet_hdr_len;
  62     Notifier exit;
  63 } TAPState;
  64
  65 static void launch_script(const char *setup_script, const char *ifname,
  66                           int fd, Error **errp);
  67
  68 static void tap_send(void *opaque);
  69 static void tap_writable(void *opaque);
  70
  71 static void tap_update_fd_handler(TAPState *s)
  72 {
  73     qemu_set_fd_handler(s->fd,
  74                         s->read_poll && s->enabled ? tap_send : NULL,
  75                         s->write_poll && s->enabled ? tap_writable : NULL,
  76                         s);
  77 }
  78
  79 static void tap_read_poll(TAPState *s, bool enable)
  80 {
  81     s->read_poll = enable;
  82     tap_update_fd_handler(s);
  83 }
  84
  85 static void tap_write_poll(TAPState *s, bool enable)
  86 {
  87     s->write_poll = enable;
  88     tap_update_fd_handler(s);
  89 }
  90
  91 static void tap_writable(void *opaque)
  92 {
  93     TAPState *s = opaque;
  94
  95     tap_write_poll(s, false);
  96
  97     qemu_flush_queued_packets(&s->nc);
  98 }
  99
 100 static ssize_t tap_write_packet(TAPState *s, const struct iovec *iov, int iovcnt)
 101 {
 102     ssize_t len;
 103
 104     do {
 105         len = writev(s->fd, iov, iovcnt);
 106     } while (len == -1 && errno == EINTR);
 107
 108     if (len == -1 && errno == EAGAIN) {
 109         tap_write_poll(s, true);
 110         return 0;
 111     }
 112
 113     return len;
 114 }
 115
 116 static ssize_t tap_receive_iov(NetClientState *nc, const struct iovec *iov,
 117                                int iovcnt)
 118 {
 119     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 120     const struct iovec *iovp = iov;
 121     struct iovec iov_copy[iovcnt + 1];
 122     struct virtio_net_hdr_mrg_rxbuf hdr = { };
 123
 124     if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
 125         iov_copy[0].iov_base = &hdr;
 126         iov_copy[0].iov_len =  s->host_vnet_hdr_len;
 127         memcpy(&iov_copy[1], iov, iovcnt * sizeof(*iov));
 128         iovp = iov_copy;
 129         iovcnt++;
 130     }
 131
 132     return tap_write_packet(s, iovp, iovcnt);
 133 }
 134
 135 static ssize_t tap_receive_raw(NetClientState *nc, const uint8_t *buf, size_t size)
 136 {
 137     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 138     struct iovec iov[2];
 139     int iovcnt = 0;
 140     struct virtio_net_hdr_mrg_rxbuf hdr = { };
 141
 142     if (s->host_vnet_hdr_len) {
 143         iov[iovcnt].iov_base = &hdr;
 144         iov[iovcnt].iov_len  = s->host_vnet_hdr_len;
 145         iovcnt++;
 146     }
 147
 148     iov[iovcnt].iov_base = (char *)buf;
 149     iov[iovcnt].iov_len  = size;
 150     iovcnt++;
 151
 152     return tap_write_packet(s, iov, iovcnt);
 153 }
 154
 155 static ssize_t tap_receive(NetClientState *nc, const uint8_t *buf, size_t size)
 156 {
 157     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 158     struct iovec iov[1];
 159
 160     if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
 161         return tap_receive_raw(nc, buf, size);
 162     }
 163
 164     iov[0].iov_base = (char *)buf;
 165     iov[0].iov_len  = size;
 166
 167     return tap_write_packet(s, iov, 1);
 168 }
 169
 170 #ifndef __sun__
 171 ssize_t tap_read_packet(int tapfd, uint8_t *buf, int maxlen)
 172 {
 173     return read(tapfd, buf, maxlen);
 174 }
 175 #endif
 176
 177 static void tap_send_completed(NetClientState *nc, ssize_t len)
 178 {
 179     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 180     tap_read_poll(s, true);
 181 }
 182
 183 static void tap_send(void *opaque)
 184 {
 185     TAPState *s = opaque;
 186     int size;
 187     int packets = 0;
 188
 189     while (true) {
 190         uint8_t *buf = s->buf;
 191
 192         size = tap_read_packet(s->fd, s->buf, sizeof(s->buf));
 193         if (size <= 0) {
 194             break;
 195         }
 196
 197         if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
 198             buf  += s->host_vnet_hdr_len;
 199             size -= s->host_vnet_hdr_len;
 200         }
 201
 202         size = qemu_send_packet_async(&s->nc, buf, size, tap_send_completed);
 203         if (size == 0) {
 204             tap_read_poll(s, false);
 205             break;
 206         } else if (size < 0) {
 207             break;
 208         }
 209
 210         /*
 211          * When the host keeps receiving more packets while tap_send() is
 212          * running we can hog the QEMU global mutex.  Limit the number of
 213          * packets that are processed per tap_send() callback to prevent
 214          * stalling the guest.
 215          */
 216         packets++;
 217         if (packets >= 50) {
 218             break;
 219         }
 220     }
 221 }
 222
 223 static bool tap_has_ufo(NetClientState *nc)
 224 {
 225     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 226
 227     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 228
 229     return s->has_ufo;
 230 }
 231
 232 static bool tap_has_vnet_hdr(NetClientState *nc)
 233 {
 234     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 235
 236     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 237
 238     return !!s->host_vnet_hdr_len;
 239 }
 240
 241 static bool tap_has_vnet_hdr_len(NetClientState *nc, int len)
 242 {
 243     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 244
 245     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 246
 247     return !!tap_probe_vnet_hdr_len(s->fd, len);
 248 }
 249
 250 static void tap_set_vnet_hdr_len(NetClientState *nc, int len)
 251 {
 252     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 253
 254     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 255     assert(len == sizeof(struct virtio_net_hdr_mrg_rxbuf) ||
 256            len == sizeof(struct virtio_net_hdr));
 257
 258     tap_fd_set_vnet_hdr_len(s->fd, len);
 259     s->host_vnet_hdr_len = len;
 260 }
 261
 262 static void tap_using_vnet_hdr(NetClientState *nc, bool using_vnet_hdr)
 263 {
 264     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 265
 266     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 267     assert(!!s->host_vnet_hdr_len == using_vnet_hdr);
 268
 269     s->using_vnet_hdr = using_vnet_hdr;
 270 }
 271
 272 static int tap_set_vnet_le(NetClientState *nc, bool is_le)
 273 {
 274     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 275
 276     return tap_fd_set_vnet_le(s->fd, is_le);
 277 }
 278
 279 static int tap_set_vnet_be(NetClientState *nc, bool is_be)
 280 {
 281     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 282
 283     return tap_fd_set_vnet_be(s->fd, is_be);
 284 }
 285
 286 static void tap_set_offload(NetClientState *nc, int csum, int tso4,
 287                      int tso6, int ecn, int ufo)
 288 {
 289     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 290     if (s->fd < 0) {
 291         return;
 292     }
 293
 294     tap_fd_set_offload(s->fd, csum, tso4, tso6, ecn, ufo);
 295 }
 296
 297 static void tap_exit_notify(Notifier *notifier, void *data)
 298 {
 299     TAPState *s = container_of(notifier, TAPState, exit);
 300     Error *err = NULL;
 301
 302     if (s->down_script[0]) {
 303         launch_script(s->down_script, s->down_script_arg, s->fd, &err);
 304         if (err) {
 305             error_report_err(err);
 306         }
 307     }
 308 }
 309
 310 static void tap_cleanup(NetClientState *nc)
 311 {
 312     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 313
 314     if (s->vhost_net) {
 315         vhost_net_cleanup(s->vhost_net);
 316         g_free(s->vhost_net);
 317         s->vhost_net = NULL;
 318     }
 319
 320     qemu_purge_queued_packets(nc);
 321
 322     tap_exit_notify(&s->exit, NULL);
 323     qemu_remove_exit_notifier(&s->exit);
 324
 325     tap_read_poll(s, false);
 326     tap_write_poll(s, false);
 327     close(s->fd);
 328     s->fd = -1;
 329 }
 330
 331 static void tap_poll(NetClientState *nc, bool enable)
 332 {
 333     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 334     tap_read_poll(s, enable);
 335     tap_write_poll(s, enable);
 336 }
 337
 338 int tap_get_fd(NetClientState *nc)
 339 {
 340     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 341     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 342     return s->fd;
 343 }
 344
 345 /* fd support */
 346
 347 static NetClientInfo net_tap_info = {
 348     .type = NET_CLIENT_DRIVER_TAP,
 349     .size = sizeof(TAPState),
 350     .receive = tap_receive,
 351     .receive_raw = tap_receive_raw,
 352     .receive_iov = tap_receive_iov,
 353     .poll = tap_poll,
 354     .cleanup = tap_cleanup,
 355     .has_ufo = tap_has_ufo,
 356     .has_vnet_hdr = tap_has_vnet_hdr,
 357     .has_vnet_hdr_len = tap_has_vnet_hdr_len,
 358     .using_vnet_hdr = tap_using_vnet_hdr,
 359     .set_offload = tap_set_offload,
 360     .set_vnet_hdr_len = tap_set_vnet_hdr_len,
 361     .set_vnet_le = tap_set_vnet_le,
 362     .set_vnet_be = tap_set_vnet_be,
 363 };
 364
 365 static TAPState *net_tap_fd_init(NetClientState *peer,
 366                                  const char *model,
 367                                  const char *name,
 368                                  int fd,
 369                                  int vnet_hdr)
 370 {
 371     NetClientState *nc;
 372     TAPState *s;
 373
 374     nc = qemu_new_net_client(&net_tap_info, peer, model, name);
 375
 376     s = DO_UPCAST(TAPState, nc, nc);
 377
 378     s->fd = fd;
 379     s->host_vnet_hdr_len = vnet_hdr ? sizeof(struct virtio_net_hdr) : 0;
 380     s->using_vnet_hdr = false;
 381     s->has_ufo = tap_probe_has_ufo(s->fd);
 382     s->enabled = true;
 383     tap_set_offload(&s->nc, 0, 0, 0, 0, 0);
 384     /*
 385      * Make sure host header length is set correctly in tap:
 386      * it might have been modified by another instance of qemu.
 387      */
 388     if (tap_probe_vnet_hdr_len(s->fd, s->host_vnet_hdr_len)) {
 389         tap_fd_set_vnet_hdr_len(s->fd, s->host_vnet_hdr_len);
 390     }
 391     tap_read_poll(s, true);
 392     s->vhost_net = NULL;
 393
 394     s->exit.notify = tap_exit_notify;
 395     qemu_add_exit_notifier(&s->exit);
 396
 397     return s;
 398 }
 399
 400 static void launch_script(const char *setup_script, const char *ifname,
 401                           int fd, Error **errp)
 402 {
 403     int pid, status;
 404     char *args[3];
 405     char **parg;
 406
 407     /* try to launch network script */
 408     pid = fork();
 409     if (pid < 0) {
 410         error_setg_errno(errp, errno, "could not launch network script %s",
 411                          setup_script);
 412         return;
 413     }
 414     if (pid == 0) {
 415         int open_max = sysconf(_SC_OPEN_MAX), i;
 416
 417         for (i = 3; i < open_max; i++) {
 418             if (i != fd) {
 419                 close(i);
 420             }
 421         }
 422         parg = args;
 423         *parg++ = (char *)setup_script;
 424         *parg++ = (char *)ifname;
 425         *parg = NULL;
 426         execv(setup_script, args);
 427         _exit(1);
 428     } else {
 429         while (waitpid(pid, &status, 0) != pid) {
 430             /* loop */
 431         }
 432
 433         if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
 434             return;
 435         }
 436         error_setg(errp, "network script %s failed with status %d",
 437                    setup_script, status);
 438     }
 439 }
 440
 441 static int recv_fd(int c)
 442 {
 443     int fd;
 444     uint8_t msgbuf[CMSG_SPACE(sizeof(fd))];
 445     struct msghdr msg = {
 446         .msg_control = msgbuf,
 447         .msg_controllen = sizeof(msgbuf),
 448     };
 449     struct cmsghdr *cmsg;
 450     struct iovec iov;
 451     uint8_t req[1];
 452     ssize_t len;
 453
 454     cmsg = CMSG_FIRSTHDR(&msg);
 455     cmsg->cmsg_level = SOL_SOCKET;
 456     cmsg->cmsg_type = SCM_RIGHTS;
 457     cmsg->cmsg_len = CMSG_LEN(sizeof(fd));
 458     msg.msg_controllen = cmsg->cmsg_len;
 459
 460     iov.iov_base = req;
 461     iov.iov_len = sizeof(req);
 462
 463     msg.msg_iov = &iov;
 464     msg.msg_iovlen = 1;
 465
 466     len = recvmsg(c, &msg, 0);
 467     if (len > 0) {
 468         memcpy(&fd, CMSG_DATA(cmsg), sizeof(fd));
 469         return fd;
 470     }
 471
 472     return len;
 473 }
 474
 475 static int net_bridge_run_helper(const char *helper, const char *bridge,
 476                                  Error **errp)
 477 {
 478     sigset_t oldmask, mask;
 479     int pid, status;
 480     char *args[5];
 481     char **parg;
 482     int sv[2];
 483
 484     sigemptyset(&mask);
 485     sigaddset(&mask, SIGCHLD);
 486     sigprocmask(SIG_BLOCK, &mask, &oldmask);
 487
 488     if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) {
 489         error_setg_errno(errp, errno, "socketpair() failed");
 490         return -1;
 491     }
 492
 493     /* try to launch bridge helper */
 494     pid = fork();
 495     if (pid < 0) {
 496         error_setg_errno(errp, errno, "Can't fork bridge helper");
 497         return -1;
 498     }
 499     if (pid == 0) {
 500         int open_max = sysconf(_SC_OPEN_MAX), i;
 501         char *fd_buf = NULL;
 502         char *br_buf = NULL;
 503         char *helper_cmd = NULL;
 504
 505         for (i = 3; i < open_max; i++) {
 506             if (i != sv[1]) {
 507                 close(i);
 508             }
 509         }
 510
 511         fd_buf = g_strdup_printf("%s%d", "--fd=", sv[1]);
 512
 513         if (strrchr(helper, ' ') || strrchr(helper, '\t')) {
 514             /* assume helper is a command */
 515
 516             if (strstr(helper, "--br=") == NULL) {
 517                 br_buf = g_strdup_printf("%s%s", "--br=", bridge);
 518             }
 519
 520             helper_cmd = g_strdup_printf("%s %s %s %s", helper,
 521                             "--use-vnet", fd_buf, br_buf ? br_buf : "");
 522
 523             parg = args;
 524             *parg++ = (char *)"sh";
 525             *parg++ = (char *)"-c";
 526             *parg++ = helper_cmd;
 527             *parg++ = NULL;
 528
 529             execv("/bin/sh", args);
 530             g_free(helper_cmd);
 531         } else {
 532             /* assume helper is just the executable path name */
 533
 534             br_buf = g_strdup_printf("%s%s", "--br=", bridge);
 535
 536             parg = args;
 537             *parg++ = (char *)helper;
 538             *parg++ = (char *)"--use-vnet";
 539             *parg++ = fd_buf;
 540             *parg++ = br_buf;
 541             *parg++ = NULL;
 542
 543             execv(helper, args);
 544         }
 545         g_free(fd_buf);
 546         g_free(br_buf);
 547         _exit(1);
 548
 549     } else {
 550         int fd;
 551         int saved_errno;
 552
 553         close(sv[1]);
 554
 555         do {
 556             fd = recv_fd(sv[0]);
 557         } while (fd == -1 && errno == EINTR);
 558         saved_errno = errno;
 559
 560         close(sv[0]);
 561
 562         while (waitpid(pid, &status, 0) != pid) {
 563             /* loop */
 564         }
 565         sigprocmask(SIG_SETMASK, &oldmask, NULL);
 566         if (fd < 0) {
 567             error_setg_errno(errp, saved_errno,
 568                              "failed to recv file descriptor");
 569             return -1;
 570         }
 571         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 572             error_setg(errp, "bridge helper failed");
 573             return -1;
 574         }
 575         return fd;
 576     }
 577 }
 578
 579 int net_init_bridge(const Netdev *netdev, const char *name,
 580                     NetClientState *peer, Error **errp)
 581 {
 582     const NetdevBridgeOptions *bridge;
 583     const char *helper, *br;
 584     TAPState *s;
 585     int fd, vnet_hdr;
 586
 587     assert(netdev->type == NET_CLIENT_DRIVER_BRIDGE);
 588     bridge = &netdev->u.bridge;
 589
 590     helper = bridge->has_helper ? bridge->helper : DEFAULT_BRIDGE_HELPER;
 591     br     = bridge->has_br     ? bridge->br     : DEFAULT_BRIDGE_INTERFACE;
 592
 593     fd = net_bridge_run_helper(helper, br, errp);
 594     if (fd == -1) {
 595         return -1;
 596     }
 597
 598     qemu_set_nonblock(fd);
 599     vnet_hdr = tap_probe_vnet_hdr(fd);
 600     s = net_tap_fd_init(peer, "bridge", name, fd, vnet_hdr);
 601
 602     snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s,br=%s", helper,
 603              br);
 604
 605     return 0;
 606 }
 607
 608 static int net_tap_init(const NetdevTapOptions *tap, int *vnet_hdr,
 609                         const char *setup_script, char *ifname,
 610                         size_t ifname_sz, int mq_required, Error **errp)
 611 {
 612     Error *err = NULL;
 613     int fd, vnet_hdr_required;
 614
 615     if (tap->has_vnet_hdr) {
 616         *vnet_hdr = tap->vnet_hdr;
 617         vnet_hdr_required = *vnet_hdr;
 618     } else {
 619         *vnet_hdr = 1;
 620         vnet_hdr_required = 0;
 621     }
 622
 623     TFR(fd = tap_open(ifname, ifname_sz, vnet_hdr, vnet_hdr_required,
 624                       mq_required, errp));
 625     if (fd < 0) {
 626         return -1;
 627     }
 628
 629     if (setup_script &&
 630         setup_script[0] != '\0' &&
 631         strcmp(setup_script, "no") != 0) {
 632         launch_script(setup_script, ifname, fd, &err);
 633         if (err) {
 634             error_propagate(errp, err);
 635             close(fd);
 636             return -1;
 637         }
 638     }
 639
 640     return fd;
 641 }
 642
 643 #define MAX_TAP_QUEUES 1024
 644
 645 static void net_init_tap_one(const NetdevTapOptions *tap, NetClientState *peer,
 646                              const char *model, const char *name,
 647                              const char *ifname, const char *script,
 648                              const char *downscript, const char *vhostfdname,
 649                              int vnet_hdr, int fd, Error **errp)
 650 {
 651     Error *err = NULL;
 652     TAPState *s = net_tap_fd_init(peer, model, name, fd, vnet_hdr);
 653     int vhostfd;
 654
 655     tap_set_sndbuf(s->fd, tap, &err);
 656     if (err) {
 657         error_propagate(errp, err);
 658         return;
 659     }
 660
 661     if (tap->has_fd || tap->has_fds) {
 662         snprintf(s->nc.info_str, sizeof(s->nc.info_str), "fd=%d", fd);
 663     } else if (tap->has_helper) {
 664         snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s",
 665                  tap->helper);
 666     } else {
 667         snprintf(s->nc.info_str, sizeof(s->nc.info_str),
 668                  "ifname=%s,script=%s,downscript=%s", ifname, script,
 669                  downscript);
 670
 671         if (strcmp(downscript, "no") != 0) {
 672             snprintf(s->down_script, sizeof(s->down_script), "%s", downscript);
 673             snprintf(s->down_script_arg, sizeof(s->down_script_arg),
 674                      "%s", ifname);
 675         }
 676     }
 677
 678     if (tap->has_vhost ? tap->vhost :
 679         vhostfdname || (tap->has_vhostforce && tap->vhostforce)) {
 680         VhostNetOptions options;
 681
 682         options.backend_type = VHOST_BACKEND_TYPE_KERNEL;
 683         options.net_backend = &s->nc;
 684         if (tap->has_poll_us) {
 685             options.busyloop_timeout = tap->poll_us;
 686         } else {
 687             options.busyloop_timeout = 0;
 688         }
 689
 690         if (vhostfdname) {
 691             vhostfd = monitor_fd_param(cur_mon, vhostfdname, &err);
 692             if (vhostfd == -1) {
 693                 if (tap->has_vhostforce && tap->vhostforce) {
 694                     error_propagate(errp, err);
 695                 } else {
 696                     warn_report_err(err);
 697                 }
 698                 return;
 699             }
 700             qemu_set_nonblock(vhostfd);
 701         } else {
 702             vhostfd = open("/dev/vhost-net", O_RDWR);
 703             if (vhostfd < 0) {
 704                 if (tap->has_vhostforce && tap->vhostforce) {
 705                     error_setg_errno(errp, errno,
 706                                      "tap: open vhost char device failed");
 707                 } else {
 708                     warn_report("tap: open vhost char device failed: %s",
 709                                 strerror(errno));
 710                 }
 711                 return;
 712             }
 713             qemu_set_nonblock(vhostfd);
 714         }
 715         options.opaque = (void *)(uintptr_t)vhostfd;
 716
 717         s->vhost_net = vhost_net_init(&options);
 718         if (!s->vhost_net) {
 719             if (tap->has_vhostforce && tap->vhostforce) {
 720                 error_setg(errp, VHOST_NET_INIT_FAILED);
 721             } else {
 722                 warn_report(VHOST_NET_INIT_FAILED);
 723             }
 724             return;
 725         }
 726     } else if (vhostfdname) {
 727         error_setg(errp, "vhostfd(s)= is not valid without vhost");
 728     }
 729 }
 730
 731 static int get_fds(char *str, char *fds[], int max)
 732 {
 733     char *ptr = str, *this;
 734     size_t len = strlen(str);
 735     int i = 0;
 736
 737     while (i < max && ptr < str + len) {
 738         this = strchr(ptr, ':');
 739
 740         if (this == NULL) {
 741             fds[i] = g_strdup(ptr);
 742         } else {
 743             fds[i] = g_strndup(ptr, this - ptr);
 744         }
 745
 746         i++;
 747         if (this == NULL) {
 748             break;
 749         } else {
 750             ptr = this + 1;
 751         }
 752     }
 753
 754     return i;
 755 }
 756
 757 int net_init_tap(const Netdev *netdev, const char *name,
 758                  NetClientState *peer, Error **errp)
 759 {
 760     const NetdevTapOptions *tap;
 761     int fd, vnet_hdr = 0, i = 0, queues;
 762     /* for the no-fd, no-helper case */
 763     const char *script = NULL; /* suppress wrong "uninit'd use" gcc warning */
 764     const char *downscript = NULL;
 765     Error *err = NULL;
 766     const char *vhostfdname;
 767     char ifname[128];
 768
 769     assert(netdev->type == NET_CLIENT_DRIVER_TAP);
 770     tap = &netdev->u.tap;
 771     queues = tap->has_queues ? tap->queues : 1;
 772     vhostfdname = tap->has_vhostfd ? tap->vhostfd : NULL;
 773
 774     /* QEMU hubs do not support multiqueue tap, in this case peer is set.
 775      * For -netdev, peer is always NULL. */
 776     if (peer && (tap->has_queues || tap->has_fds || tap->has_vhostfds)) {
 777         error_setg(errp, "Multiqueue tap cannot be used with hubs");
 778         return -1;
 779     }
 780
 781     if (tap->has_fd) {
 782         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
 783             tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
 784             tap->has_fds || tap->has_vhostfds) {
 785             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 786                        "helper=, queues=, fds=, and vhostfds= "
 787                        "are invalid with fd=");
 788             return -1;
 789         }
 790
 791         fd = monitor_fd_param(cur_mon, tap->fd, &err);
 792         if (fd == -1) {
 793             error_propagate(errp, err);
 794             return -1;
 795         }
 796
 797         qemu_set_nonblock(fd);
 798
 799         vnet_hdr = tap_probe_vnet_hdr(fd);
 800
 801         net_init_tap_one(tap, peer, "tap", name, NULL,
 802                          script, downscript,
 803                          vhostfdname, vnet_hdr, fd, &err);
 804         if (err) {
 805             error_propagate(errp, err);
 806             return -1;
 807         }
 808     } else if (tap->has_fds) {
 809         char **fds;
 810         char **vhost_fds;
 811         int nfds = 0, nvhosts = 0;
 812         int ret = 0;
 813
 814         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
 815             tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
 816             tap->has_vhostfd) {
 817             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 818                        "helper=, queues=, and vhostfd= "
 819                        "are invalid with fds=");
 820             return -1;
 821         }
 822
 823         fds = g_new0(char *, MAX_TAP_QUEUES);
 824         vhost_fds = g_new0(char *, MAX_TAP_QUEUES);
 825
 826         nfds = get_fds(tap->fds, fds, MAX_TAP_QUEUES);
 827         if (tap->has_vhostfds) {
 828             nvhosts = get_fds(tap->vhostfds, vhost_fds, MAX_TAP_QUEUES);
 829             if (nfds != nvhosts) {
 830                 error_setg(errp, "The number of fds passed does not match "
 831                            "the number of vhostfds passed");
 832                 ret = -1;
 833                 goto free_fail;
 834             }
 835         }
 836
 837         for (i = 0; i < nfds; i++) {
 838             fd = monitor_fd_param(cur_mon, fds[i], &err);
 839             if (fd == -1) {
 840                 error_propagate(errp, err);
 841                 ret = -1;
 842                 goto free_fail;
 843             }
 844
 845             qemu_set_nonblock(fd);
 846
 847             if (i == 0) {
 848                 vnet_hdr = tap_probe_vnet_hdr(fd);
 849             } else if (vnet_hdr != tap_probe_vnet_hdr(fd)) {
 850                 error_setg(errp,
 851                            "vnet_hdr not consistent across given tap fds");
 852                 ret = -1;
 853                 goto free_fail;
 854             }
 855
 856             net_init_tap_one(tap, peer, "tap", name, ifname,
 857                              script, downscript,
 858                              tap->has_vhostfds ? vhost_fds[i] : NULL,
 859                              vnet_hdr, fd, &err);
 860             if (err) {
 861                 error_propagate(errp, err);
 862                 ret = -1;
 863                 goto free_fail;
 864             }
 865         }
 866
 867 free_fail:
 868         for (i = 0; i < nvhosts; i++) {
 869             g_free(vhost_fds[i]);
 870         }
 871         for (i = 0; i < nfds; i++) {
 872             g_free(fds[i]);
 873         }
 874         g_free(fds);
 875         g_free(vhost_fds);
 876         return ret;
 877     } else if (tap->has_helper) {
 878         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
 879             tap->has_vnet_hdr || tap->has_queues || tap->has_vhostfds) {
 880             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 881                        "queues=, and vhostfds= are invalid with helper=");
 882             return -1;
 883         }
 884
 885         fd = net_bridge_run_helper(tap->helper,
 886                                    tap->has_br ?
 887                                    tap->br : DEFAULT_BRIDGE_INTERFACE,
 888                                    errp);
 889         if (fd == -1) {
 890             return -1;
 891         }
 892
 893         qemu_set_nonblock(fd);
 894         vnet_hdr = tap_probe_vnet_hdr(fd);
 895
 896         net_init_tap_one(tap, peer, "bridge", name, ifname,
 897                          script, downscript, vhostfdname,
 898                          vnet_hdr, fd, &err);
 899         if (err) {
 900             error_propagate(errp, err);
 901             close(fd);
 902             return -1;
 903         }
 904     } else {
 905         if (tap->has_vhostfds) {
 906             error_setg(errp, "vhostfds= is invalid if fds= wasn't specified");
 907             return -1;
 908         }
 909         script = tap->has_script ? tap->script : DEFAULT_NETWORK_SCRIPT;
 910         downscript = tap->has_downscript ? tap->downscript :
 911             DEFAULT_NETWORK_DOWN_SCRIPT;
 912
 913         if (tap->has_ifname) {
 914             pstrcpy(ifname, sizeof ifname, tap->ifname);
 915         } else {
 916             ifname[0] = '\0';
 917         }
 918
 919         for (i = 0; i < queues; i++) {
 920             fd = net_tap_init(tap, &vnet_hdr, i >= 1 ? "no" : script,
 921                               ifname, sizeof ifname, queues > 1, errp);
 922             if (fd == -1) {
 923                 return -1;
 924             }
 925
 926             if (queues > 1 && i == 0 && !tap->has_ifname) {
 927                 if (tap_fd_get_ifname(fd, ifname)) {
 928                     error_setg(errp, "Fail to get ifname");
 929                     close(fd);
 930                     return -1;
 931                 }
 932             }
 933
 934             net_init_tap_one(tap, peer, "tap", name, ifname,
 935                              i >= 1 ? "no" : script,
 936                              i >= 1 ? "no" : downscript,
 937                              vhostfdname, vnet_hdr, fd, &err);
 938             if (err) {
 939                 error_propagate(errp, err);
 940                 close(fd);
 941                 return -1;
 942             }
 943         }
 944     }
 945
 946     return 0;
 947 }
 948
 949 VHostNetState *tap_get_vhost_net(NetClientState *nc)
 950 {
 951     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 952     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 953     return s->vhost_net;
 954 }
 955
 956 int tap_enable(NetClientState *nc)
 957 {
 958     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 959     int ret;
 960
 961     if (s->enabled) {
 962         return 0;
 963     } else {
 964         ret = tap_fd_enable(s->fd);
 965         if (ret == 0) {
 966             s->enabled = true;
 967             tap_update_fd_handler(s);
 968         }
 969         return ret;
 970     }
 971 }
 972
 973 int tap_disable(NetClientState *nc)
 974 {
 975     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 976     int ret;
 977
 978     if (s->enabled == 0) {
 979         return 0;
 980     } else {
 981         ret = tap_fd_disable(s->fd);
 982         if (ret == 0) {
 983             qemu_purge_queued_packets(nc);
 984             s->enabled = false;
 985             tap_update_fd_handler(s);
 986         }
 987         return ret;
 988     }
 989 }