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 "tap_int.h"
  27
  28 #include "config-host.h"
  29
  30 #include <sys/ioctl.h>
  31 #include <sys/stat.h>
  32 #include <sys/wait.h>
  33 #include <sys/socket.h>
  34 #include <net/if.h>
  35
  36 #include "net/net.h"
  37 #include "clients.h"
  38 #include "monitor/monitor.h"
  39 #include "sysemu/sysemu.h"
  40 #include "qemu-common.h"
  41 #include "qemu/error-report.h"
  42
  43 #include "net/tap.h"
  44
  45 #include "net/vhost_net.h"
  46
  47 typedef struct TAPState {
  48     NetClientState nc;
  49     int fd;
  50     char down_script[1024];
  51     char down_script_arg[128];
  52     uint8_t buf[NET_BUFSIZE];
  53     bool read_poll;
  54     bool write_poll;
  55     bool using_vnet_hdr;
  56     bool has_ufo;
  57     bool enabled;
  58     VHostNetState *vhost_net;
  59     unsigned host_vnet_hdr_len;
  60 } TAPState;
  61
  62 static void launch_script(const char *setup_script, const char *ifname,
  63                           int fd, Error **errp);
  64
  65 static int tap_can_send(void *opaque);
  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_handler2(s->fd,
  72                          s->read_poll && s->enabled ? tap_can_send : NULL,
  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 static int tap_can_send(void *opaque)
 170 {
 171     TAPState *s = opaque;
 172
 173     return qemu_can_send_packet(&s->nc);
 174 }
 175
 176 #ifndef __sun__
 177 ssize_t tap_read_packet(int tapfd, uint8_t *buf, int maxlen)
 178 {
 179     return read(tapfd, buf, maxlen);
 180 }
 181 #endif
 182
 183 static void tap_send_completed(NetClientState *nc, ssize_t len)
 184 {
 185     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 186     tap_read_poll(s, true);
 187 }
 188
 189 static void tap_send(void *opaque)
 190 {
 191     TAPState *s = opaque;
 192     int size;
 193     int packets = 0;
 194
 195     while (qemu_can_send_packet(&s->nc)) {
 196         uint8_t *buf = s->buf;
 197
 198         size = tap_read_packet(s->fd, s->buf, sizeof(s->buf));
 199         if (size <= 0) {
 200             break;
 201         }
 202
 203         if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
 204             buf  += s->host_vnet_hdr_len;
 205             size -= s->host_vnet_hdr_len;
 206         }
 207
 208         size = qemu_send_packet_async(&s->nc, buf, size, tap_send_completed);
 209         if (size == 0) {
 210             tap_read_poll(s, false);
 211             break;
 212         } else if (size < 0) {
 213             break;
 214         }
 215
 216         /*
 217          * When the host keeps receiving more packets while tap_send() is
 218          * running we can hog the QEMU global mutex.  Limit the number of
 219          * packets that are processed per tap_send() callback to prevent
 220          * stalling the guest.
 221          */
 222         packets++;
 223         if (packets >= 50) {
 224             break;
 225         }
 226     }
 227 }
 228
 229 static bool tap_has_ufo(NetClientState *nc)
 230 {
 231     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 232
 233     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
 234
 235     return s->has_ufo;
 236 }
 237
 238 static bool tap_has_vnet_hdr(NetClientState *nc)
 239 {
 240     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 241
 242     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
 243
 244     return !!s->host_vnet_hdr_len;
 245 }
 246
 247 static bool tap_has_vnet_hdr_len(NetClientState *nc, int len)
 248 {
 249     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 250
 251     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
 252
 253     return !!tap_probe_vnet_hdr_len(s->fd, len);
 254 }
 255
 256 static void tap_set_vnet_hdr_len(NetClientState *nc, int len)
 257 {
 258     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 259
 260     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
 261     assert(len == sizeof(struct virtio_net_hdr_mrg_rxbuf) ||
 262            len == sizeof(struct virtio_net_hdr));
 263
 264     tap_fd_set_vnet_hdr_len(s->fd, len);
 265     s->host_vnet_hdr_len = len;
 266 }
 267
 268 static void tap_using_vnet_hdr(NetClientState *nc, bool using_vnet_hdr)
 269 {
 270     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 271
 272     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
 273     assert(!!s->host_vnet_hdr_len == using_vnet_hdr);
 274
 275     s->using_vnet_hdr = using_vnet_hdr;
 276 }
 277
 278 static void tap_set_offload(NetClientState *nc, int csum, int tso4,
 279                      int tso6, int ecn, int ufo)
 280 {
 281     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 282     if (s->fd < 0) {
 283         return;
 284     }
 285
 286     tap_fd_set_offload(s->fd, csum, tso4, tso6, ecn, ufo);
 287 }
 288
 289 static void tap_cleanup(NetClientState *nc)
 290 {
 291     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 292     Error *err = NULL;
 293
 294     if (s->vhost_net) {
 295         vhost_net_cleanup(s->vhost_net);
 296         s->vhost_net = NULL;
 297     }
 298
 299     qemu_purge_queued_packets(nc);
 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     tap_read_poll(s, false);
 309     tap_write_poll(s, false);
 310     close(s->fd);
 311     s->fd = -1;
 312 }
 313
 314 static void tap_poll(NetClientState *nc, bool enable)
 315 {
 316     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 317     tap_read_poll(s, enable);
 318     tap_write_poll(s, enable);
 319 }
 320
 321 int tap_get_fd(NetClientState *nc)
 322 {
 323     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 324     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
 325     return s->fd;
 326 }
 327
 328 /* fd support */
 329
 330 static NetClientInfo net_tap_info = {
 331     .type = NET_CLIENT_OPTIONS_KIND_TAP,
 332     .size = sizeof(TAPState),
 333     .receive = tap_receive,
 334     .receive_raw = tap_receive_raw,
 335     .receive_iov = tap_receive_iov,
 336     .poll = tap_poll,
 337     .cleanup = tap_cleanup,
 338     .has_ufo = tap_has_ufo,
 339     .has_vnet_hdr = tap_has_vnet_hdr,
 340     .has_vnet_hdr_len = tap_has_vnet_hdr_len,
 341     .using_vnet_hdr = tap_using_vnet_hdr,
 342     .set_offload = tap_set_offload,
 343     .set_vnet_hdr_len = tap_set_vnet_hdr_len,
 344 };
 345
 346 static TAPState *net_tap_fd_init(NetClientState *peer,
 347                                  const char *model,
 348                                  const char *name,
 349                                  int fd,
 350                                  int vnet_hdr)
 351 {
 352     NetClientState *nc;
 353     TAPState *s;
 354
 355     nc = qemu_new_net_client(&net_tap_info, peer, model, name);
 356
 357     s = DO_UPCAST(TAPState, nc, nc);
 358
 359     s->fd = fd;
 360     s->host_vnet_hdr_len = vnet_hdr ? sizeof(struct virtio_net_hdr) : 0;
 361     s->using_vnet_hdr = false;
 362     s->has_ufo = tap_probe_has_ufo(s->fd);
 363     s->enabled = true;
 364     tap_set_offload(&s->nc, 0, 0, 0, 0, 0);
 365     /*
 366      * Make sure host header length is set correctly in tap:
 367      * it might have been modified by another instance of qemu.
 368      */
 369     if (tap_probe_vnet_hdr_len(s->fd, s->host_vnet_hdr_len)) {
 370         tap_fd_set_vnet_hdr_len(s->fd, s->host_vnet_hdr_len);
 371     }
 372     tap_read_poll(s, true);
 373     s->vhost_net = NULL;
 374     return s;
 375 }
 376
 377 static void launch_script(const char *setup_script, const char *ifname,
 378                           int fd, Error **errp)
 379 {
 380     int pid, status;
 381     char *args[3];
 382     char **parg;
 383
 384     /* try to launch network script */
 385     pid = fork();
 386     if (pid < 0) {
 387         error_setg_errno(errp, errno, "could not launch network script %s",
 388                          setup_script);
 389         return;
 390     }
 391     if (pid == 0) {
 392         int open_max = sysconf(_SC_OPEN_MAX), i;
 393
 394         for (i = 3; i < open_max; i++) {
 395             if (i != fd) {
 396                 close(i);
 397             }
 398         }
 399         parg = args;
 400         *parg++ = (char *)setup_script;
 401         *parg++ = (char *)ifname;
 402         *parg = NULL;
 403         execv(setup_script, args);
 404         _exit(1);
 405     } else {
 406         while (waitpid(pid, &status, 0) != pid) {
 407             /* loop */
 408         }
 409
 410         if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
 411             return;
 412         }
 413         error_setg(errp, "network script %s failed with status %d",
 414                    setup_script, status);
 415     }
 416 }
 417
 418 static int recv_fd(int c)
 419 {
 420     int fd;
 421     uint8_t msgbuf[CMSG_SPACE(sizeof(fd))];
 422     struct msghdr msg = {
 423         .msg_control = msgbuf,
 424         .msg_controllen = sizeof(msgbuf),
 425     };
 426     struct cmsghdr *cmsg;
 427     struct iovec iov;
 428     uint8_t req[1];
 429     ssize_t len;
 430
 431     cmsg = CMSG_FIRSTHDR(&msg);
 432     cmsg->cmsg_level = SOL_SOCKET;
 433     cmsg->cmsg_type = SCM_RIGHTS;
 434     cmsg->cmsg_len = CMSG_LEN(sizeof(fd));
 435     msg.msg_controllen = cmsg->cmsg_len;
 436
 437     iov.iov_base = req;
 438     iov.iov_len = sizeof(req);
 439
 440     msg.msg_iov = &iov;
 441     msg.msg_iovlen = 1;
 442
 443     len = recvmsg(c, &msg, 0);
 444     if (len > 0) {
 445         memcpy(&fd, CMSG_DATA(cmsg), sizeof(fd));
 446         return fd;
 447     }
 448
 449     return len;
 450 }
 451
 452 static int net_bridge_run_helper(const char *helper, const char *bridge,
 453                                  Error **errp)
 454 {
 455     sigset_t oldmask, mask;
 456     int pid, status;
 457     char *args[5];
 458     char **parg;
 459     int sv[2];
 460
 461     sigemptyset(&mask);
 462     sigaddset(&mask, SIGCHLD);
 463     sigprocmask(SIG_BLOCK, &mask, &oldmask);
 464
 465     if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) {
 466         error_setg_errno(errp, errno, "socketpair() failed");
 467         return -1;
 468     }
 469
 470     /* try to launch bridge helper */
 471     pid = fork();
 472     if (pid < 0) {
 473         error_setg_errno(errp, errno, "Can't fork bridge helper");
 474         return -1;
 475     }
 476     if (pid == 0) {
 477         int open_max = sysconf(_SC_OPEN_MAX), i;
 478         char fd_buf[6+10];
 479         char br_buf[6+IFNAMSIZ] = {0};
 480         char helper_cmd[PATH_MAX + sizeof(fd_buf) + sizeof(br_buf) + 15];
 481
 482         for (i = 3; i < open_max; i++) {
 483             if (i != sv[1]) {
 484                 close(i);
 485             }
 486         }
 487
 488         snprintf(fd_buf, sizeof(fd_buf), "%s%d", "--fd=", sv[1]);
 489
 490         if (strrchr(helper, ' ') || strrchr(helper, '\t')) {
 491             /* assume helper is a command */
 492
 493             if (strstr(helper, "--br=") == NULL) {
 494                 snprintf(br_buf, sizeof(br_buf), "%s%s", "--br=", bridge);
 495             }
 496
 497             snprintf(helper_cmd, sizeof(helper_cmd), "%s %s %s %s",
 498                      helper, "--use-vnet", fd_buf, br_buf);
 499
 500             parg = args;
 501             *parg++ = (char *)"sh";
 502             *parg++ = (char *)"-c";
 503             *parg++ = helper_cmd;
 504             *parg++ = NULL;
 505
 506             execv("/bin/sh", args);
 507         } else {
 508             /* assume helper is just the executable path name */
 509
 510             snprintf(br_buf, sizeof(br_buf), "%s%s", "--br=", bridge);
 511
 512             parg = args;
 513             *parg++ = (char *)helper;
 514             *parg++ = (char *)"--use-vnet";
 515             *parg++ = fd_buf;
 516             *parg++ = br_buf;
 517             *parg++ = NULL;
 518
 519             execv(helper, args);
 520         }
 521         _exit(1);
 522
 523     } else {
 524         int fd;
 525         int saved_errno;
 526
 527         close(sv[1]);
 528
 529         do {
 530             fd = recv_fd(sv[0]);
 531         } while (fd == -1 && errno == EINTR);
 532         saved_errno = errno;
 533
 534         close(sv[0]);
 535
 536         while (waitpid(pid, &status, 0) != pid) {
 537             /* loop */
 538         }
 539         sigprocmask(SIG_SETMASK, &oldmask, NULL);
 540         if (fd < 0) {
 541             error_setg_errno(errp, saved_errno,
 542                              "failed to recv file descriptor");
 543             return -1;
 544         }
 545         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 546             error_setg(errp, "bridge helper failed");
 547             return -1;
 548         }
 549         return fd;
 550     }
 551 }
 552
 553 int net_init_bridge(const NetClientOptions *opts, const char *name,
 554                     NetClientState *peer, Error **errp)
 555 {
 556     const NetdevBridgeOptions *bridge;
 557     const char *helper, *br;
 558     TAPState *s;
 559     int fd, vnet_hdr;
 560
 561     assert(opts->kind == NET_CLIENT_OPTIONS_KIND_BRIDGE);
 562     bridge = opts->bridge;
 563
 564     helper = bridge->has_helper ? bridge->helper : DEFAULT_BRIDGE_HELPER;
 565     br     = bridge->has_br     ? bridge->br     : DEFAULT_BRIDGE_INTERFACE;
 566
 567     fd = net_bridge_run_helper(helper, br, errp);
 568     if (fd == -1) {
 569         return -1;
 570     }
 571
 572     fcntl(fd, F_SETFL, O_NONBLOCK);
 573     vnet_hdr = tap_probe_vnet_hdr(fd);
 574     s = net_tap_fd_init(peer, "bridge", name, fd, vnet_hdr);
 575
 576     snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s,br=%s", helper,
 577              br);
 578
 579     return 0;
 580 }
 581
 582 static int net_tap_init(const NetdevTapOptions *tap, int *vnet_hdr,
 583                         const char *setup_script, char *ifname,
 584                         size_t ifname_sz, int mq_required, Error **errp)
 585 {
 586     Error *err = NULL;
 587     int fd, vnet_hdr_required;
 588
 589     if (tap->has_vnet_hdr) {
 590         *vnet_hdr = tap->vnet_hdr;
 591         vnet_hdr_required = *vnet_hdr;
 592     } else {
 593         *vnet_hdr = 1;
 594         vnet_hdr_required = 0;
 595     }
 596
 597     TFR(fd = tap_open(ifname, ifname_sz, vnet_hdr, vnet_hdr_required,
 598                       mq_required, errp));
 599     if (fd < 0) {
 600         return -1;
 601     }
 602
 603     if (setup_script &&
 604         setup_script[0] != '\0' &&
 605         strcmp(setup_script, "no") != 0) {
 606         launch_script(setup_script, ifname, fd, &err);
 607         if (err) {
 608             error_propagate(errp, err);
 609             close(fd);
 610             return -1;
 611         }
 612     }
 613
 614     return fd;
 615 }
 616
 617 #define MAX_TAP_QUEUES 1024
 618
 619 static void net_init_tap_one(const NetdevTapOptions *tap, NetClientState *peer,
 620                              const char *model, const char *name,
 621                              const char *ifname, const char *script,
 622                              const char *downscript, const char *vhostfdname,
 623                              int vnet_hdr, int fd, Error **errp)
 624 {
 625     Error *err = NULL;
 626     TAPState *s = net_tap_fd_init(peer, model, name, fd, vnet_hdr);
 627     int vhostfd;
 628
 629     tap_set_sndbuf(s->fd, tap, &err);
 630     if (err) {
 631         error_propagate(errp, err);
 632         return;
 633     }
 634
 635     if (tap->has_fd || tap->has_fds) {
 636         snprintf(s->nc.info_str, sizeof(s->nc.info_str), "fd=%d", fd);
 637     } else if (tap->has_helper) {
 638         snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s",
 639                  tap->helper);
 640     } else {
 641         snprintf(s->nc.info_str, sizeof(s->nc.info_str),
 642                  "ifname=%s,script=%s,downscript=%s", ifname, script,
 643                  downscript);
 644
 645         if (strcmp(downscript, "no") != 0) {
 646             snprintf(s->down_script, sizeof(s->down_script), "%s", downscript);
 647             snprintf(s->down_script_arg, sizeof(s->down_script_arg),
 648                      "%s", ifname);
 649         }
 650     }
 651
 652     if (tap->has_vhost ? tap->vhost :
 653         vhostfdname || (tap->has_vhostforce && tap->vhostforce)) {
 654         VhostNetOptions options;
 655
 656         options.backend_type = VHOST_BACKEND_TYPE_KERNEL;
 657         options.net_backend = &s->nc;
 658         options.force = tap->has_vhostforce && tap->vhostforce;
 659
 660         if (tap->has_vhostfd || tap->has_vhostfds) {
 661             vhostfd = monitor_fd_param(cur_mon, vhostfdname, &err);
 662             if (vhostfd == -1) {
 663                 error_propagate(errp, err);
 664                 return;
 665             }
 666         } else {
 667             vhostfd = open("/dev/vhost-net", O_RDWR);
 668             if (vhostfd < 0) {
 669                 error_setg_errno(errp, errno,
 670                                  "tap: open vhost char device failed");
 671                 return;
 672             }
 673         }
 674         options.opaque = (void *)(uintptr_t)vhostfd;
 675
 676         s->vhost_net = vhost_net_init(&options);
 677         if (!s->vhost_net) {
 678             error_setg(errp,
 679                        "vhost-net requested but could not be initialized");
 680             return;
 681         }
 682     } else if (tap->has_vhostfd || tap->has_vhostfds) {
 683         error_setg(errp, "vhostfd= is not valid without vhost");
 684     }
 685 }
 686
 687 static int get_fds(char *str, char *fds[], int max)
 688 {
 689     char *ptr = str, *this;
 690     size_t len = strlen(str);
 691     int i = 0;
 692
 693     while (i < max && ptr < str + len) {
 694         this = strchr(ptr, ':');
 695
 696         if (this == NULL) {
 697             fds[i] = g_strdup(ptr);
 698         } else {
 699             fds[i] = g_strndup(ptr, this - ptr);
 700         }
 701
 702         i++;
 703         if (this == NULL) {
 704             break;
 705         } else {
 706             ptr = this + 1;
 707         }
 708     }
 709
 710     return i;
 711 }
 712
 713 int net_init_tap(const NetClientOptions *opts, const char *name,
 714                  NetClientState *peer, Error **errp)
 715 {
 716     const NetdevTapOptions *tap;
 717     int fd, vnet_hdr = 0, i = 0, queues;
 718     /* for the no-fd, no-helper case */
 719     const char *script = NULL; /* suppress wrong "uninit'd use" gcc warning */
 720     const char *downscript = NULL;
 721     Error *err = NULL;
 722     const char *vhostfdname;
 723     char ifname[128];
 724
 725     assert(opts->kind == NET_CLIENT_OPTIONS_KIND_TAP);
 726     tap = opts->tap;
 727     queues = tap->has_queues ? tap->queues : 1;
 728     vhostfdname = tap->has_vhostfd ? tap->vhostfd : NULL;
 729
 730     /* QEMU vlans does not support multiqueue tap, in this case peer is set.
 731      * For -netdev, peer is always NULL. */
 732     if (peer && (tap->has_queues || tap->has_fds || tap->has_vhostfds)) {
 733         error_setg(errp, "Multiqueue tap cannot be used with QEMU vlans");
 734         return -1;
 735     }
 736
 737     if (tap->has_fd) {
 738         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
 739             tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
 740             tap->has_fds || tap->has_vhostfds) {
 741             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 742                        "helper=, queues=, fds=, and vhostfds= "
 743                        "are invalid with fd=");
 744             return -1;
 745         }
 746
 747         fd = monitor_fd_param(cur_mon, tap->fd, &err);
 748         if (fd == -1) {
 749             error_propagate(errp, err);
 750             return -1;
 751         }
 752
 753         fcntl(fd, F_SETFL, O_NONBLOCK);
 754
 755         vnet_hdr = tap_probe_vnet_hdr(fd);
 756
 757         net_init_tap_one(tap, peer, "tap", name, NULL,
 758                          script, downscript,
 759                          vhostfdname, vnet_hdr, fd, &err);
 760         if (err) {
 761             error_propagate(errp, err);
 762             return -1;
 763         }
 764     } else if (tap->has_fds) {
 765         char *fds[MAX_TAP_QUEUES];
 766         char *vhost_fds[MAX_TAP_QUEUES];
 767         int nfds, nvhosts;
 768
 769         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
 770             tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
 771             tap->has_vhostfd) {
 772             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 773                        "helper=, queues=, and vhostfd= "
 774                        "are invalid with fds=");
 775             return -1;
 776         }
 777
 778         nfds = get_fds(tap->fds, fds, MAX_TAP_QUEUES);
 779         if (tap->has_vhostfds) {
 780             nvhosts = get_fds(tap->vhostfds, vhost_fds, MAX_TAP_QUEUES);
 781             if (nfds != nvhosts) {
 782                 error_setg(errp, "The number of fds passed does not match "
 783                            "the number of vhostfds passed");
 784                 return -1;
 785             }
 786         }
 787
 788         for (i = 0; i < nfds; i++) {
 789             fd = monitor_fd_param(cur_mon, fds[i], &err);
 790             if (fd == -1) {
 791                 error_propagate(errp, err);
 792                 return -1;
 793             }
 794
 795             fcntl(fd, F_SETFL, O_NONBLOCK);
 796
 797             if (i == 0) {
 798                 vnet_hdr = tap_probe_vnet_hdr(fd);
 799             } else if (vnet_hdr != tap_probe_vnet_hdr(fd)) {
 800                 error_setg(errp,
 801                            "vnet_hdr not consistent across given tap fds");
 802                 return -1;
 803             }
 804
 805             net_init_tap_one(tap, peer, "tap", name, ifname,
 806                              script, downscript,
 807                              tap->has_vhostfds ? vhost_fds[i] : NULL,
 808                              vnet_hdr, fd, &err);
 809             if (err) {
 810                 error_propagate(errp, err);
 811                 return -1;
 812             }
 813         }
 814     } else if (tap->has_helper) {
 815         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
 816             tap->has_vnet_hdr || tap->has_queues || tap->has_vhostfds) {
 817             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 818                        "queues=, and vhostfds= are invalid with helper=");
 819             return -1;
 820         }
 821
 822         fd = net_bridge_run_helper(tap->helper, DEFAULT_BRIDGE_INTERFACE,
 823                                    errp);
 824         if (fd == -1) {
 825             return -1;
 826         }
 827
 828         fcntl(fd, F_SETFL, O_NONBLOCK);
 829         vnet_hdr = tap_probe_vnet_hdr(fd);
 830
 831         net_init_tap_one(tap, peer, "bridge", name, ifname,
 832                          script, downscript, vhostfdname,
 833                          vnet_hdr, fd, &err);
 834         if (err) {
 835             error_propagate(errp, err);
 836             close(fd);
 837             return -1;
 838         }
 839     } else {
 840         if (tap->has_vhostfds) {
 841             error_setg(errp, "vhostfds= is invalid if fds= wasn't specified");
 842             return -1;
 843         }
 844         script = tap->has_script ? tap->script : DEFAULT_NETWORK_SCRIPT;
 845         downscript = tap->has_downscript ? tap->downscript :
 846             DEFAULT_NETWORK_DOWN_SCRIPT;
 847
 848         if (tap->has_ifname) {
 849             pstrcpy(ifname, sizeof ifname, tap->ifname);
 850         } else {
 851             ifname[0] = '\0';
 852         }
 853
 854         for (i = 0; i < queues; i++) {
 855             fd = net_tap_init(tap, &vnet_hdr, i >= 1 ? "no" : script,
 856                               ifname, sizeof ifname, queues > 1, errp);
 857             if (fd == -1) {
 858                 return -1;
 859             }
 860
 861             if (queues > 1 && i == 0 && !tap->has_ifname) {
 862                 if (tap_fd_get_ifname(fd, ifname)) {
 863                     error_setg(errp, "Fail to get ifname");
 864                     close(fd);
 865                     return -1;
 866                 }
 867             }
 868
 869             net_init_tap_one(tap, peer, "tap", name, ifname,
 870                              i >= 1 ? "no" : script,
 871                              i >= 1 ? "no" : downscript,
 872                              vhostfdname, vnet_hdr, fd, &err);
 873             if (err) {
 874                 error_propagate(errp, err);
 875                 close(fd);
 876                 return -1;
 877             }
 878         }
 879     }
 880
 881     return 0;
 882 }
 883
 884 VHostNetState *tap_get_vhost_net(NetClientState *nc)
 885 {
 886     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 887     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
 888     return s->vhost_net;
 889 }
 890
 891 int tap_enable(NetClientState *nc)
 892 {
 893     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 894     int ret;
 895
 896     if (s->enabled) {
 897         return 0;
 898     } else {
 899         ret = tap_fd_enable(s->fd);
 900         if (ret == 0) {
 901             s->enabled = true;
 902             tap_update_fd_handler(s);
 903         }
 904         return ret;
 905     }
 906 }
 907
 908 int tap_disable(NetClientState *nc)
 909 {
 910     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 911     int ret;
 912
 913     if (s->enabled == 0) {
 914         return 0;
 915     } else {
 916         ret = tap_fd_disable(s->fd);
 917         if (ret == 0) {
 918             qemu_purge_queued_packets(nc);
 919             s->enabled = false;
 920             tap_update_fd_handler(s);
 921         }
 922         return ret;
 923     }
 924 }