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