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/eth.h"
  36 #include "net/net.h"
  37 #include "clients.h"
  38 #include "monitor/monitor.h"
  39 #include "sysemu/sysemu.h"
  40 #include "qapi/error.h"
  41 #include "qemu/cutils.h"
  42 #include "qemu/error-report.h"
  43 #include "qemu/main-loop.h"
  44 #include "qemu/sockets.h"
  45
  46 #include "net/tap.h"
  47
  48 #include "net/vhost_net.h"
  49
  50 typedef struct TAPState {
  51     NetClientState nc;
  52     int fd;
  53     char down_script[1024];
  54     char down_script_arg[128];
  55     uint8_t buf[NET_BUFSIZE];
  56     bool read_poll;
  57     bool write_poll;
  58     bool using_vnet_hdr;
  59     bool has_ufo;
  60     bool has_uso;
  61     bool enabled;
  62     VHostNetState *vhost_net;
  63     unsigned host_vnet_hdr_len;
  64     Notifier exit;
  65 } TAPState;
  66
  67 static void launch_script(const char *setup_script, const char *ifname,
  68                           int fd, Error **errp);
  69
  70 static void tap_send(void *opaque);
  71 static void tap_writable(void *opaque);
  72
  73 static void tap_update_fd_handler(TAPState *s)
  74 {
  75     qemu_set_fd_handler(s->fd,
  76                         s->read_poll && s->enabled ? tap_send : NULL,
  77                         s->write_poll && s->enabled ? tap_writable : NULL,
  78                         s);
  79 }
  80
  81 static void tap_read_poll(TAPState *s, bool enable)
  82 {
  83     s->read_poll = enable;
  84     tap_update_fd_handler(s);
  85 }
  86
  87 static void tap_write_poll(TAPState *s, bool enable)
  88 {
  89     s->write_poll = enable;
  90     tap_update_fd_handler(s);
  91 }
  92
  93 static void tap_writable(void *opaque)
  94 {
  95     TAPState *s = opaque;
  96
  97     tap_write_poll(s, false);
  98
  99     qemu_flush_queued_packets(&s->nc);
 100 }
 101
 102 static ssize_t tap_write_packet(TAPState *s, const struct iovec *iov, int iovcnt)
 103 {
 104     ssize_t len;
 105
 106     len = RETRY_ON_EINTR(writev(s->fd, iov, iovcnt));
 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     g_autofree struct iovec *iov_copy = NULL;
 122     struct virtio_net_hdr hdr = { };
 123
 124     if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
 125         iov_copy = g_new(struct iovec, iovcnt + 1);
 126         iov_copy[0].iov_base = &hdr;
 127         iov_copy[0].iov_len =  s->host_vnet_hdr_len;
 128         memcpy(&iov_copy[1], iov, iovcnt * sizeof(*iov));
 129         iovp = iov_copy;
 130         iovcnt++;
 131     }
 132
 133     return tap_write_packet(s, iovp, iovcnt);
 134 }
 135
 136 static ssize_t tap_receive(NetClientState *nc, const uint8_t *buf, size_t size)
 137 {
 138     struct iovec iov = {
 139         .iov_base = (void *)buf,
 140         .iov_len = size
 141     };
 142
 143     return tap_receive_iov(nc, &iov, 1);
 144 }
 145
 146 #ifndef __sun__
 147 ssize_t tap_read_packet(int tapfd, uint8_t *buf, int maxlen)
 148 {
 149     return read(tapfd, buf, maxlen);
 150 }
 151 #endif
 152
 153 static void tap_send_completed(NetClientState *nc, ssize_t len)
 154 {
 155     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 156     tap_read_poll(s, true);
 157 }
 158
 159 static void tap_send(void *opaque)
 160 {
 161     TAPState *s = opaque;
 162     int size;
 163     int packets = 0;
 164
 165     while (true) {
 166         uint8_t *buf = s->buf;
 167         uint8_t min_pkt[ETH_ZLEN];
 168         size_t min_pktsz = sizeof(min_pkt);
 169
 170         size = tap_read_packet(s->fd, s->buf, sizeof(s->buf));
 171         if (size <= 0) {
 172             break;
 173         }
 174
 175         if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
 176             buf  += s->host_vnet_hdr_len;
 177             size -= s->host_vnet_hdr_len;
 178         }
 179
 180         if (net_peer_needs_padding(&s->nc)) {
 181             if (eth_pad_short_frame(min_pkt, &min_pktsz, buf, size)) {
 182                 buf = min_pkt;
 183                 size = min_pktsz;
 184             }
 185         }
 186
 187         size = qemu_send_packet_async(&s->nc, buf, size, tap_send_completed);
 188         if (size == 0) {
 189             tap_read_poll(s, false);
 190             break;
 191         } else if (size < 0) {
 192             break;
 193         }
 194
 195         /*
 196          * When the host keeps receiving more packets while tap_send() is
 197          * running we can hog the BQL.  Limit the number of
 198          * packets that are processed per tap_send() callback to prevent
 199          * stalling the guest.
 200          */
 201         packets++;
 202         if (packets >= 50) {
 203             break;
 204         }
 205     }
 206 }
 207
 208 static bool tap_has_ufo(NetClientState *nc)
 209 {
 210     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 211
 212     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 213
 214     return s->has_ufo;
 215 }
 216
 217 static bool tap_has_uso(NetClientState *nc)
 218 {
 219     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 220
 221     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 222
 223     return s->has_uso;
 224 }
 225
 226 static bool tap_has_vnet_hdr(NetClientState *nc)
 227 {
 228     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 229
 230     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 231
 232     return !!s->host_vnet_hdr_len;
 233 }
 234
 235 static bool tap_has_vnet_hdr_len(NetClientState *nc, int len)
 236 {
 237     return tap_has_vnet_hdr(nc);
 238 }
 239
 240 static void tap_set_vnet_hdr_len(NetClientState *nc, int len)
 241 {
 242     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 243
 244     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 245
 246     tap_fd_set_vnet_hdr_len(s->fd, len);
 247     s->host_vnet_hdr_len = len;
 248     s->using_vnet_hdr = true;
 249 }
 250
 251 static int tap_set_vnet_le(NetClientState *nc, bool is_le)
 252 {
 253     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 254
 255     return tap_fd_set_vnet_le(s->fd, is_le);
 256 }
 257
 258 static int tap_set_vnet_be(NetClientState *nc, bool is_be)
 259 {
 260     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 261
 262     return tap_fd_set_vnet_be(s->fd, is_be);
 263 }
 264
 265 static void tap_set_offload(NetClientState *nc, int csum, int tso4,
 266                      int tso6, int ecn, int ufo, int uso4, int uso6)
 267 {
 268     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 269     if (s->fd < 0) {
 270         return;
 271     }
 272
 273     tap_fd_set_offload(s->fd, csum, tso4, tso6, ecn, ufo, uso4, uso6);
 274 }
 275
 276 static void tap_exit_notify(Notifier *notifier, void *data)
 277 {
 278     TAPState *s = container_of(notifier, TAPState, exit);
 279     Error *err = NULL;
 280
 281     if (s->down_script[0]) {
 282         launch_script(s->down_script, s->down_script_arg, s->fd, &err);
 283         if (err) {
 284             error_report_err(err);
 285         }
 286     }
 287 }
 288
 289 static void tap_cleanup(NetClientState *nc)
 290 {
 291     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 292
 293     if (s->vhost_net) {
 294         vhost_net_cleanup(s->vhost_net);
 295         g_free(s->vhost_net);
 296         s->vhost_net = NULL;
 297     }
 298
 299     qemu_purge_queued_packets(nc);
 300
 301     tap_exit_notify(&s->exit, NULL);
 302     qemu_remove_exit_notifier(&s->exit);
 303
 304     tap_read_poll(s, false);
 305     tap_write_poll(s, false);
 306     close(s->fd);
 307     s->fd = -1;
 308 }
 309
 310 static void tap_poll(NetClientState *nc, bool enable)
 311 {
 312     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 313     tap_read_poll(s, enable);
 314     tap_write_poll(s, enable);
 315 }
 316
 317 static bool tap_set_steering_ebpf(NetClientState *nc, int prog_fd)
 318 {
 319     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 320     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 321
 322     return tap_fd_set_steering_ebpf(s->fd, prog_fd) == 0;
 323 }
 324
 325 int tap_get_fd(NetClientState *nc)
 326 {
 327     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 328     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 329     return s->fd;
 330 }
 331
 332 /* fd support */
 333
 334 static NetClientInfo net_tap_info = {
 335     .type = NET_CLIENT_DRIVER_TAP,
 336     .size = sizeof(TAPState),
 337     .receive = tap_receive,
 338     .receive_iov = tap_receive_iov,
 339     .poll = tap_poll,
 340     .cleanup = tap_cleanup,
 341     .has_ufo = tap_has_ufo,
 342     .has_uso = tap_has_uso,
 343     .has_vnet_hdr = tap_has_vnet_hdr,
 344     .has_vnet_hdr_len = tap_has_vnet_hdr_len,
 345     .set_offload = tap_set_offload,
 346     .set_vnet_hdr_len = tap_set_vnet_hdr_len,
 347     .set_vnet_le = tap_set_vnet_le,
 348     .set_vnet_be = tap_set_vnet_be,
 349     .set_steering_ebpf = tap_set_steering_ebpf,
 350 };
 351
 352 static TAPState *net_tap_fd_init(NetClientState *peer,
 353                                  const char *model,
 354                                  const char *name,
 355                                  int fd,
 356                                  int vnet_hdr)
 357 {
 358     NetClientState *nc;
 359     TAPState *s;
 360
 361     nc = qemu_new_net_client(&net_tap_info, peer, model, name);
 362
 363     s = DO_UPCAST(TAPState, nc, nc);
 364
 365     s->fd = fd;
 366     s->host_vnet_hdr_len = vnet_hdr ? sizeof(struct virtio_net_hdr) : 0;
 367     s->using_vnet_hdr = false;
 368     s->has_ufo = tap_probe_has_ufo(s->fd);
 369     s->has_uso = tap_probe_has_uso(s->fd);
 370     s->enabled = true;
 371     tap_set_offload(&s->nc, 0, 0, 0, 0, 0, 0, 0);
 372     /*
 373      * Make sure host header length is set correctly in tap:
 374      * it might have been modified by another instance of qemu.
 375      */
 376     if (vnet_hdr) {
 377         tap_fd_set_vnet_hdr_len(s->fd, s->host_vnet_hdr_len);
 378     }
 379     tap_read_poll(s, true);
 380     s->vhost_net = NULL;
 381
 382     s->exit.notify = tap_exit_notify;
 383     qemu_add_exit_notifier(&s->exit);
 384
 385     return s;
 386 }
 387
 388 static void launch_script(const char *setup_script, const char *ifname,
 389                           int fd, Error **errp)
 390 {
 391     int pid, status;
 392     char *args[3];
 393     char **parg;
 394
 395     /* try to launch network script */
 396     pid = fork();
 397     if (pid < 0) {
 398         error_setg_errno(errp, errno, "could not launch network script %s",
 399                          setup_script);
 400         return;
 401     }
 402     if (pid == 0) {
 403         int open_max = sysconf(_SC_OPEN_MAX), i;
 404
 405         for (i = 3; i < open_max; i++) {
 406             if (i != fd) {
 407                 close(i);
 408             }
 409         }
 410         parg = args;
 411         *parg++ = (char *)setup_script;
 412         *parg++ = (char *)ifname;
 413         *parg = NULL;
 414         execv(setup_script, args);
 415         _exit(1);
 416     } else {
 417         while (waitpid(pid, &status, 0) != pid) {
 418             /* loop */
 419         }
 420
 421         if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
 422             return;
 423         }
 424         error_setg(errp, "network script %s failed with status %d",
 425                    setup_script, status);
 426     }
 427 }
 428
 429 static int recv_fd(int c)
 430 {
 431     int fd;
 432     uint8_t msgbuf[CMSG_SPACE(sizeof(fd))];
 433     struct msghdr msg = {
 434         .msg_control = msgbuf,
 435         .msg_controllen = sizeof(msgbuf),
 436     };
 437     struct cmsghdr *cmsg;
 438     struct iovec iov;
 439     uint8_t req[1];
 440     ssize_t len;
 441
 442     cmsg = CMSG_FIRSTHDR(&msg);
 443     cmsg->cmsg_level = SOL_SOCKET;
 444     cmsg->cmsg_type = SCM_RIGHTS;
 445     cmsg->cmsg_len = CMSG_LEN(sizeof(fd));
 446     msg.msg_controllen = cmsg->cmsg_len;
 447
 448     iov.iov_base = req;
 449     iov.iov_len = sizeof(req);
 450
 451     msg.msg_iov = &iov;
 452     msg.msg_iovlen = 1;
 453
 454     len = recvmsg(c, &msg, 0);
 455     if (len > 0) {
 456         memcpy(&fd, CMSG_DATA(cmsg), sizeof(fd));
 457         return fd;
 458     }
 459
 460     return len;
 461 }
 462
 463 static int net_bridge_run_helper(const char *helper, const char *bridge,
 464                                  Error **errp)
 465 {
 466     sigset_t oldmask, mask;
 467     g_autofree char *default_helper = NULL;
 468     int pid, status;
 469     char *args[5];
 470     char **parg;
 471     int sv[2];
 472
 473     sigemptyset(&mask);
 474     sigaddset(&mask, SIGCHLD);
 475     sigprocmask(SIG_BLOCK, &mask, &oldmask);
 476
 477     if (!helper) {
 478         helper = default_helper = get_relocated_path(DEFAULT_BRIDGE_HELPER);
 479     }
 480
 481     if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) {
 482         error_setg_errno(errp, errno, "socketpair() failed");
 483         return -1;
 484     }
 485
 486     /* try to launch bridge helper */
 487     pid = fork();
 488     if (pid < 0) {
 489         error_setg_errno(errp, errno, "Can't fork bridge helper");
 490         return -1;
 491     }
 492     if (pid == 0) {
 493         int open_max = sysconf(_SC_OPEN_MAX), i;
 494         char *fd_buf = NULL;
 495         char *br_buf = NULL;
 496         char *helper_cmd = NULL;
 497
 498         for (i = 3; i < open_max; i++) {
 499             if (i != sv[1]) {
 500                 close(i);
 501             }
 502         }
 503
 504         fd_buf = g_strdup_printf("%s%d", "--fd=", sv[1]);
 505
 506         if (strrchr(helper, ' ') || strrchr(helper, '\t')) {
 507             /* assume helper is a command */
 508
 509             if (strstr(helper, "--br=") == NULL) {
 510                 br_buf = g_strdup_printf("%s%s", "--br=", bridge);
 511             }
 512
 513             helper_cmd = g_strdup_printf("%s %s %s %s", helper,
 514                             "--use-vnet", fd_buf, br_buf ? br_buf : "");
 515
 516             parg = args;
 517             *parg++ = (char *)"sh";
 518             *parg++ = (char *)"-c";
 519             *parg++ = helper_cmd;
 520             *parg++ = NULL;
 521
 522             execv("/bin/sh", args);
 523             g_free(helper_cmd);
 524         } else {
 525             /* assume helper is just the executable path name */
 526
 527             br_buf = g_strdup_printf("%s%s", "--br=", bridge);
 528
 529             parg = args;
 530             *parg++ = (char *)helper;
 531             *parg++ = (char *)"--use-vnet";
 532             *parg++ = fd_buf;
 533             *parg++ = br_buf;
 534             *parg++ = NULL;
 535
 536             execv(helper, args);
 537         }
 538         g_free(fd_buf);
 539         g_free(br_buf);
 540         _exit(1);
 541
 542     } else {
 543         int fd;
 544         int saved_errno;
 545
 546         close(sv[1]);
 547
 548         fd = RETRY_ON_EINTR(recv_fd(sv[0]));
 549         saved_errno = errno;
 550
 551         close(sv[0]);
 552
 553         while (waitpid(pid, &status, 0) != pid) {
 554             /* loop */
 555         }
 556         sigprocmask(SIG_SETMASK, &oldmask, NULL);
 557         if (fd < 0) {
 558             error_setg_errno(errp, saved_errno,
 559                              "failed to recv file descriptor");
 560             return -1;
 561         }
 562         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 563             error_setg(errp, "bridge helper failed");
 564             return -1;
 565         }
 566         return fd;
 567     }
 568 }
 569
 570 int net_init_bridge(const Netdev *netdev, const char *name,
 571                     NetClientState *peer, Error **errp)
 572 {
 573     const NetdevBridgeOptions *bridge;
 574     const char *helper, *br;
 575     TAPState *s;
 576     int fd, vnet_hdr;
 577
 578     assert(netdev->type == NET_CLIENT_DRIVER_BRIDGE);
 579     bridge = &netdev->u.bridge;
 580     helper = bridge->helper;
 581     br     = bridge->br ?: DEFAULT_BRIDGE_INTERFACE;
 582
 583     fd = net_bridge_run_helper(helper, br, errp);
 584     if (fd == -1) {
 585         return -1;
 586     }
 587
 588     if (!g_unix_set_fd_nonblocking(fd, true, NULL)) {
 589         error_setg_errno(errp, errno, "Failed to set FD nonblocking");
 590         return -1;
 591     }
 592     vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 593     if (vnet_hdr < 0) {
 594         close(fd);
 595         return -1;
 596     }
 597     s = net_tap_fd_init(peer, "bridge", name, fd, vnet_hdr);
 598
 599     qemu_set_info_str(&s->nc, "helper=%s,br=%s", helper, br);
 600
 601     return 0;
 602 }
 603
 604 static int net_tap_init(const NetdevTapOptions *tap, int *vnet_hdr,
 605                         const char *setup_script, char *ifname,
 606                         size_t ifname_sz, int mq_required, Error **errp)
 607 {
 608     Error *err = NULL;
 609     int fd, vnet_hdr_required;
 610
 611     if (tap->has_vnet_hdr) {
 612         *vnet_hdr = tap->vnet_hdr;
 613         vnet_hdr_required = *vnet_hdr;
 614     } else {
 615         *vnet_hdr = 1;
 616         vnet_hdr_required = 0;
 617     }
 618
 619     fd = RETRY_ON_EINTR(tap_open(ifname, ifname_sz, vnet_hdr, vnet_hdr_required,
 620                       mq_required, errp));
 621     if (fd < 0) {
 622         return -1;
 623     }
 624
 625     if (setup_script &&
 626         setup_script[0] != '\0' &&
 627         strcmp(setup_script, "no") != 0) {
 628         launch_script(setup_script, ifname, fd, &err);
 629         if (err) {
 630             error_propagate(errp, err);
 631             close(fd);
 632             return -1;
 633         }
 634     }
 635
 636     return fd;
 637 }
 638
 639 #define MAX_TAP_QUEUES 1024
 640
 641 static void net_init_tap_one(const NetdevTapOptions *tap, NetClientState *peer,
 642                              const char *model, const char *name,
 643                              const char *ifname, const char *script,
 644                              const char *downscript, const char *vhostfdname,
 645                              int vnet_hdr, int fd, Error **errp)
 646 {
 647     Error *err = NULL;
 648     TAPState *s = net_tap_fd_init(peer, model, name, fd, vnet_hdr);
 649     int vhostfd;
 650
 651     tap_set_sndbuf(s->fd, tap, &err);
 652     if (err) {
 653         error_propagate(errp, err);
 654         goto failed;
 655     }
 656
 657     if (tap->fd || tap->fds) {
 658         qemu_set_info_str(&s->nc, "fd=%d", fd);
 659     } else if (tap->helper) {
 660         qemu_set_info_str(&s->nc, "helper=%s", tap->helper);
 661     } else {
 662         qemu_set_info_str(&s->nc, "ifname=%s,script=%s,downscript=%s", ifname,
 663                           script, downscript);
 664
 665         if (strcmp(downscript, "no") != 0) {
 666             snprintf(s->down_script, sizeof(s->down_script), "%s", downscript);
 667             snprintf(s->down_script_arg, sizeof(s->down_script_arg),
 668                      "%s", ifname);
 669         }
 670     }
 671
 672     if (tap->has_vhost ? tap->vhost :
 673         vhostfdname || (tap->has_vhostforce && tap->vhostforce)) {
 674         VhostNetOptions options;
 675
 676         options.backend_type = VHOST_BACKEND_TYPE_KERNEL;
 677         options.net_backend = &s->nc;
 678         if (tap->has_poll_us) {
 679             options.busyloop_timeout = tap->poll_us;
 680         } else {
 681             options.busyloop_timeout = 0;
 682         }
 683
 684         if (vhostfdname) {
 685             vhostfd = monitor_fd_param(monitor_cur(), vhostfdname, &err);
 686             if (vhostfd == -1) {
 687                 error_propagate(errp, err);
 688                 goto failed;
 689             }
 690             if (!g_unix_set_fd_nonblocking(vhostfd, true, NULL)) {
 691                 error_setg_errno(errp, errno, "%s: Can't use file descriptor %d",
 692                                  name, fd);
 693                 goto failed;
 694             }
 695         } else {
 696             vhostfd = open("/dev/vhost-net", O_RDWR);
 697             if (vhostfd < 0) {
 698                 error_setg_errno(errp, errno,
 699                                  "tap: open vhost char device failed");
 700                 goto failed;
 701             }
 702             if (!g_unix_set_fd_nonblocking(vhostfd, true, NULL)) {
 703                 error_setg_errno(errp, errno, "Failed to set FD nonblocking");
 704                 goto failed;
 705             }
 706         }
 707         options.opaque = (void *)(uintptr_t)vhostfd;
 708         options.nvqs = 2;
 709
 710         s->vhost_net = vhost_net_init(&options);
 711         if (!s->vhost_net) {
 712             error_setg(errp,
 713                        "vhost-net requested but could not be initialized");
 714             goto failed;
 715         }
 716     } else if (vhostfdname) {
 717         error_setg(errp, "vhostfd(s)= is not valid without vhost");
 718         goto failed;
 719     }
 720
 721     return;
 722
 723 failed:
 724     qemu_del_net_client(&s->nc);
 725 }
 726
 727 static int get_fds(char *str, char *fds[], int max)
 728 {
 729     char *ptr = str, *this;
 730     size_t len = strlen(str);
 731     int i = 0;
 732
 733     while (i < max && ptr < str + len) {
 734         this = strchr(ptr, ':');
 735
 736         if (this == NULL) {
 737             fds[i] = g_strdup(ptr);
 738         } else {
 739             fds[i] = g_strndup(ptr, this - ptr);
 740         }
 741
 742         i++;
 743         if (this == NULL) {
 744             break;
 745         } else {
 746             ptr = this + 1;
 747         }
 748     }
 749
 750     return i;
 751 }
 752
 753 int net_init_tap(const Netdev *netdev, const char *name,
 754                  NetClientState *peer, Error **errp)
 755 {
 756     const NetdevTapOptions *tap;
 757     int fd, vnet_hdr = 0, i = 0, queues;
 758     /* for the no-fd, no-helper case */
 759     const char *script;
 760     const char *downscript;
 761     Error *err = NULL;
 762     const char *vhostfdname;
 763     char ifname[128];
 764     int ret = 0;
 765
 766     assert(netdev->type == NET_CLIENT_DRIVER_TAP);
 767     tap = &netdev->u.tap;
 768     queues = tap->has_queues ? tap->queues : 1;
 769     vhostfdname = tap->vhostfd;
 770     script = tap->script;
 771     downscript = tap->downscript;
 772
 773     /* QEMU hubs do not support multiqueue tap, in this case peer is set.
 774      * For -netdev, peer is always NULL. */
 775     if (peer && (tap->has_queues || tap->fds || tap->vhostfds)) {
 776         error_setg(errp, "Multiqueue tap cannot be used with hubs");
 777         return -1;
 778     }
 779
 780     if (tap->fd) {
 781         if (tap->ifname || tap->script || tap->downscript ||
 782             tap->has_vnet_hdr || tap->helper || tap->has_queues ||
 783             tap->fds || tap->vhostfds) {
 784             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 785                        "helper=, queues=, fds=, and vhostfds= "
 786                        "are invalid with fd=");
 787             return -1;
 788         }
 789
 790         fd = monitor_fd_param(monitor_cur(), tap->fd, errp);
 791         if (fd == -1) {
 792             return -1;
 793         }
 794
 795         if (!g_unix_set_fd_nonblocking(fd, true, NULL)) {
 796             error_setg_errno(errp, errno, "%s: Can't use file descriptor %d",
 797                              name, fd);
 798             close(fd);
 799             return -1;
 800         }
 801
 802         vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 803         if (vnet_hdr < 0) {
 804             close(fd);
 805             return -1;
 806         }
 807
 808         net_init_tap_one(tap, peer, "tap", name, NULL,
 809                          script, downscript,
 810                          vhostfdname, vnet_hdr, fd, &err);
 811         if (err) {
 812             error_propagate(errp, err);
 813             close(fd);
 814             return -1;
 815         }
 816     } else if (tap->fds) {
 817         char **fds;
 818         char **vhost_fds;
 819         int nfds = 0, nvhosts = 0;
 820
 821         if (tap->ifname || tap->script || tap->downscript ||
 822             tap->has_vnet_hdr || tap->helper || tap->has_queues ||
 823             tap->vhostfd) {
 824             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 825                        "helper=, queues=, and vhostfd= "
 826                        "are invalid with fds=");
 827             return -1;
 828         }
 829
 830         fds = g_new0(char *, MAX_TAP_QUEUES);
 831         vhost_fds = g_new0(char *, MAX_TAP_QUEUES);
 832
 833         nfds = get_fds(tap->fds, fds, MAX_TAP_QUEUES);
 834         if (tap->vhostfds) {
 835             nvhosts = get_fds(tap->vhostfds, vhost_fds, MAX_TAP_QUEUES);
 836             if (nfds != nvhosts) {
 837                 error_setg(errp, "The number of fds passed does not match "
 838                            "the number of vhostfds passed");
 839                 ret = -1;
 840                 goto free_fail;
 841             }
 842         }
 843
 844         for (i = 0; i < nfds; i++) {
 845             fd = monitor_fd_param(monitor_cur(), fds[i], errp);
 846             if (fd == -1) {
 847                 ret = -1;
 848                 goto free_fail;
 849             }
 850
 851             ret = g_unix_set_fd_nonblocking(fd, true, NULL);
 852             if (!ret) {
 853                 error_setg_errno(errp, errno, "%s: Can't use file descriptor %d",
 854                                  name, fd);
 855                 goto free_fail;
 856             }
 857
 858             if (i == 0) {
 859                 vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 860                 if (vnet_hdr < 0) {
 861                     ret = -1;
 862                     goto free_fail;
 863                 }
 864             } else if (vnet_hdr != tap_probe_vnet_hdr(fd, NULL)) {
 865                 error_setg(errp,
 866                            "vnet_hdr not consistent across given tap fds");
 867                 ret = -1;
 868                 goto free_fail;
 869             }
 870
 871             net_init_tap_one(tap, peer, "tap", name, ifname,
 872                              script, downscript,
 873                              tap->vhostfds ? vhost_fds[i] : NULL,
 874                              vnet_hdr, fd, &err);
 875             if (err) {
 876                 error_propagate(errp, err);
 877                 ret = -1;
 878                 goto free_fail;
 879             }
 880         }
 881
 882 free_fail:
 883         for (i = 0; i < nvhosts; i++) {
 884             g_free(vhost_fds[i]);
 885         }
 886         for (i = 0; i < nfds; i++) {
 887             g_free(fds[i]);
 888         }
 889         g_free(fds);
 890         g_free(vhost_fds);
 891         return ret;
 892     } else if (tap->helper) {
 893         if (tap->ifname || tap->script || tap->downscript ||
 894             tap->has_vnet_hdr || tap->has_queues || tap->vhostfds) {
 895             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 896                        "queues=, and vhostfds= are invalid with helper=");
 897             return -1;
 898         }
 899
 900         fd = net_bridge_run_helper(tap->helper,
 901                                    tap->br ?: DEFAULT_BRIDGE_INTERFACE,
 902                                    errp);
 903         if (fd == -1) {
 904             return -1;
 905         }
 906
 907         if (!g_unix_set_fd_nonblocking(fd, true, NULL)) {
 908             error_setg_errno(errp, errno, "Failed to set FD nonblocking");
 909             return -1;
 910         }
 911         vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 912         if (vnet_hdr < 0) {
 913             close(fd);
 914             return -1;
 915         }
 916
 917         net_init_tap_one(tap, peer, "bridge", name, ifname,
 918                          script, downscript, vhostfdname,
 919                          vnet_hdr, fd, &err);
 920         if (err) {
 921             error_propagate(errp, err);
 922             close(fd);
 923             return -1;
 924         }
 925     } else {
 926         g_autofree char *default_script = NULL;
 927         g_autofree char *default_downscript = NULL;
 928         if (tap->vhostfds) {
 929             error_setg(errp, "vhostfds= is invalid if fds= wasn't specified");
 930             return -1;
 931         }
 932
 933         if (!script) {
 934             script = default_script = get_relocated_path(DEFAULT_NETWORK_SCRIPT);
 935         }
 936         if (!downscript) {
 937             downscript = default_downscript =
 938                                  get_relocated_path(DEFAULT_NETWORK_DOWN_SCRIPT);
 939         }
 940
 941         if (tap->ifname) {
 942             pstrcpy(ifname, sizeof ifname, tap->ifname);
 943         } else {
 944             ifname[0] = '\0';
 945         }
 946
 947         for (i = 0; i < queues; i++) {
 948             fd = net_tap_init(tap, &vnet_hdr, i >= 1 ? "no" : script,
 949                               ifname, sizeof ifname, queues > 1, errp);
 950             if (fd == -1) {
 951                 return -1;
 952             }
 953
 954             if (queues > 1 && i == 0 && !tap->ifname) {
 955                 if (tap_fd_get_ifname(fd, ifname)) {
 956                     error_setg(errp, "Fail to get ifname");
 957                     close(fd);
 958                     return -1;
 959                 }
 960             }
 961
 962             net_init_tap_one(tap, peer, "tap", name, ifname,
 963                              i >= 1 ? "no" : script,
 964                              i >= 1 ? "no" : downscript,
 965                              vhostfdname, vnet_hdr, fd, &err);
 966             if (err) {
 967                 error_propagate(errp, err);
 968                 close(fd);
 969                 return -1;
 970             }
 971         }
 972     }
 973
 974     return 0;
 975 }
 976
 977 VHostNetState *tap_get_vhost_net(NetClientState *nc)
 978 {
 979     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 980     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 981     return s->vhost_net;
 982 }
 983
 984 int tap_enable(NetClientState *nc)
 985 {
 986     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 987     int ret;
 988
 989     if (s->enabled) {
 990         return 0;
 991     } else {
 992         ret = tap_fd_enable(s->fd);
 993         if (ret == 0) {
 994             s->enabled = true;
 995             tap_update_fd_handler(s);
 996         }
 997         return ret;
 998     }
 999 }
1000
1001 int tap_disable(NetClientState *nc)
1002 {
1003     TAPState *s = DO_UPCAST(TAPState, nc, nc);
1004     int ret;
1005
1006     if (s->enabled == 0) {
1007         return 0;
1008     } else {
1009         ret = tap_fd_disable(s->fd);
1010         if (ret == 0) {
1011             qemu_purge_queued_packets(nc);
1012             s->enabled = false;
1013             tap_update_fd_handler(s);
1014         }
1015         return ret;
1016     }
1017 }