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-common.h"
  42 #include "qemu/cutils.h"
  43 #include "qemu/error-report.h"
  44 #include "qemu/main-loop.h"
  45 #include "qemu/sockets.h"
  46
  47 #include "net/tap.h"
  48
  49 #include "net/vhost_net.h"
  50
  51 typedef struct TAPState {
  52     NetClientState nc;
  53     int fd;
  54     char down_script[1024];
  55     char down_script_arg[128];
  56     uint8_t buf[NET_BUFSIZE];
  57     bool read_poll;
  58     bool write_poll;
  59     bool using_vnet_hdr;
  60     bool has_ufo;
  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     do {
 107         len = writev(s->fd, iov, iovcnt);
 108     } while (len == -1 && errno == EINTR);
 109
 110     if (len == -1 && errno == EAGAIN) {
 111         tap_write_poll(s, true);
 112         return 0;
 113     }
 114
 115     return len;
 116 }
 117
 118 static ssize_t tap_receive_iov(NetClientState *nc, const struct iovec *iov,
 119                                int iovcnt)
 120 {
 121     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 122     const struct iovec *iovp = iov;
 123     struct iovec iov_copy[iovcnt + 1];
 124     struct virtio_net_hdr_mrg_rxbuf hdr = { };
 125
 126     if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
 127         iov_copy[0].iov_base = &hdr;
 128         iov_copy[0].iov_len =  s->host_vnet_hdr_len;
 129         memcpy(&iov_copy[1], iov, iovcnt * sizeof(*iov));
 130         iovp = iov_copy;
 131         iovcnt++;
 132     }
 133
 134     return tap_write_packet(s, iovp, iovcnt);
 135 }
 136
 137 static ssize_t tap_receive_raw(NetClientState *nc, const uint8_t *buf, size_t size)
 138 {
 139     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 140     struct iovec iov[2];
 141     int iovcnt = 0;
 142     struct virtio_net_hdr_mrg_rxbuf hdr = { };
 143
 144     if (s->host_vnet_hdr_len) {
 145         iov[iovcnt].iov_base = &hdr;
 146         iov[iovcnt].iov_len  = s->host_vnet_hdr_len;
 147         iovcnt++;
 148     }
 149
 150     iov[iovcnt].iov_base = (char *)buf;
 151     iov[iovcnt].iov_len  = size;
 152     iovcnt++;
 153
 154     return tap_write_packet(s, iov, iovcnt);
 155 }
 156
 157 static ssize_t tap_receive(NetClientState *nc, const uint8_t *buf, size_t size)
 158 {
 159     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 160     struct iovec iov[1];
 161
 162     if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
 163         return tap_receive_raw(nc, buf, size);
 164     }
 165
 166     iov[0].iov_base = (char *)buf;
 167     iov[0].iov_len  = size;
 168
 169     return tap_write_packet(s, iov, 1);
 170 }
 171
 172 #ifndef __sun__
 173 ssize_t tap_read_packet(int tapfd, uint8_t *buf, int maxlen)
 174 {
 175     return read(tapfd, buf, maxlen);
 176 }
 177 #endif
 178
 179 static void tap_send_completed(NetClientState *nc, ssize_t len)
 180 {
 181     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 182     tap_read_poll(s, true);
 183 }
 184
 185 static void tap_send(void *opaque)
 186 {
 187     TAPState *s = opaque;
 188     int size;
 189     int packets = 0;
 190
 191     while (true) {
 192         uint8_t *buf = s->buf;
 193         uint8_t min_pkt[ETH_ZLEN];
 194         size_t min_pktsz = sizeof(min_pkt);
 195
 196         size = tap_read_packet(s->fd, s->buf, sizeof(s->buf));
 197         if (size <= 0) {
 198             break;
 199         }
 200
 201         if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
 202             buf  += s->host_vnet_hdr_len;
 203             size -= s->host_vnet_hdr_len;
 204         }
 205
 206         if (net_peer_needs_padding(&s->nc)) {
 207             if (eth_pad_short_frame(min_pkt, &min_pktsz, buf, size)) {
 208                 buf = min_pkt;
 209                 size = min_pktsz;
 210             }
 211         }
 212
 213         size = qemu_send_packet_async(&s->nc, buf, size, tap_send_completed);
 214         if (size == 0) {
 215             tap_read_poll(s, false);
 216             break;
 217         } else if (size < 0) {
 218             break;
 219         }
 220
 221         /*
 222          * When the host keeps receiving more packets while tap_send() is
 223          * running we can hog the QEMU global mutex.  Limit the number of
 224          * packets that are processed per tap_send() callback to prevent
 225          * stalling the guest.
 226          */
 227         packets++;
 228         if (packets >= 50) {
 229             break;
 230         }
 231     }
 232 }
 233
 234 static bool tap_has_ufo(NetClientState *nc)
 235 {
 236     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 237
 238     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 239
 240     return s->has_ufo;
 241 }
 242
 243 static bool tap_has_vnet_hdr(NetClientState *nc)
 244 {
 245     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 246
 247     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 248
 249     return !!s->host_vnet_hdr_len;
 250 }
 251
 252 static bool tap_has_vnet_hdr_len(NetClientState *nc, int len)
 253 {
 254     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 255
 256     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 257
 258     return !!tap_probe_vnet_hdr_len(s->fd, len);
 259 }
 260
 261 static void tap_set_vnet_hdr_len(NetClientState *nc, int len)
 262 {
 263     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 264
 265     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 266     assert(len == sizeof(struct virtio_net_hdr_mrg_rxbuf) ||
 267            len == sizeof(struct virtio_net_hdr) ||
 268            len == sizeof(struct virtio_net_hdr_v1_hash));
 269
 270     tap_fd_set_vnet_hdr_len(s->fd, len);
 271     s->host_vnet_hdr_len = len;
 272 }
 273
 274 static void tap_using_vnet_hdr(NetClientState *nc, bool using_vnet_hdr)
 275 {
 276     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 277
 278     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 279     assert(!!s->host_vnet_hdr_len == using_vnet_hdr);
 280
 281     s->using_vnet_hdr = using_vnet_hdr;
 282 }
 283
 284 static int tap_set_vnet_le(NetClientState *nc, bool is_le)
 285 {
 286     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 287
 288     return tap_fd_set_vnet_le(s->fd, is_le);
 289 }
 290
 291 static int tap_set_vnet_be(NetClientState *nc, bool is_be)
 292 {
 293     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 294
 295     return tap_fd_set_vnet_be(s->fd, is_be);
 296 }
 297
 298 static void tap_set_offload(NetClientState *nc, int csum, int tso4,
 299                      int tso6, int ecn, int ufo)
 300 {
 301     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 302     if (s->fd < 0) {
 303         return;
 304     }
 305
 306     tap_fd_set_offload(s->fd, csum, tso4, tso6, ecn, ufo);
 307 }
 308
 309 static void tap_exit_notify(Notifier *notifier, void *data)
 310 {
 311     TAPState *s = container_of(notifier, TAPState, exit);
 312     Error *err = NULL;
 313
 314     if (s->down_script[0]) {
 315         launch_script(s->down_script, s->down_script_arg, s->fd, &err);
 316         if (err) {
 317             error_report_err(err);
 318         }
 319     }
 320 }
 321
 322 static void tap_cleanup(NetClientState *nc)
 323 {
 324     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 325
 326     if (s->vhost_net) {
 327         vhost_net_cleanup(s->vhost_net);
 328         g_free(s->vhost_net);
 329         s->vhost_net = NULL;
 330     }
 331
 332     qemu_purge_queued_packets(nc);
 333
 334     tap_exit_notify(&s->exit, NULL);
 335     qemu_remove_exit_notifier(&s->exit);
 336
 337     tap_read_poll(s, false);
 338     tap_write_poll(s, false);
 339     close(s->fd);
 340     s->fd = -1;
 341 }
 342
 343 static void tap_poll(NetClientState *nc, bool enable)
 344 {
 345     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 346     tap_read_poll(s, enable);
 347     tap_write_poll(s, enable);
 348 }
 349
 350 static bool tap_set_steering_ebpf(NetClientState *nc, int prog_fd)
 351 {
 352     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 353     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 354
 355     return tap_fd_set_steering_ebpf(s->fd, prog_fd) == 0;
 356 }
 357
 358 int tap_get_fd(NetClientState *nc)
 359 {
 360     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 361     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 362     return s->fd;
 363 }
 364
 365 /* fd support */
 366
 367 static NetClientInfo net_tap_info = {
 368     .type = NET_CLIENT_DRIVER_TAP,
 369     .size = sizeof(TAPState),
 370     .receive = tap_receive,
 371     .receive_raw = tap_receive_raw,
 372     .receive_iov = tap_receive_iov,
 373     .poll = tap_poll,
 374     .cleanup = tap_cleanup,
 375     .has_ufo = tap_has_ufo,
 376     .has_vnet_hdr = tap_has_vnet_hdr,
 377     .has_vnet_hdr_len = tap_has_vnet_hdr_len,
 378     .using_vnet_hdr = tap_using_vnet_hdr,
 379     .set_offload = tap_set_offload,
 380     .set_vnet_hdr_len = tap_set_vnet_hdr_len,
 381     .set_vnet_le = tap_set_vnet_le,
 382     .set_vnet_be = tap_set_vnet_be,
 383     .set_steering_ebpf = tap_set_steering_ebpf,
 384 };
 385
 386 static TAPState *net_tap_fd_init(NetClientState *peer,
 387                                  const char *model,
 388                                  const char *name,
 389                                  int fd,
 390                                  int vnet_hdr)
 391 {
 392     NetClientState *nc;
 393     TAPState *s;
 394
 395     nc = qemu_new_net_client(&net_tap_info, peer, model, name);
 396
 397     s = DO_UPCAST(TAPState, nc, nc);
 398
 399     s->fd = fd;
 400     s->host_vnet_hdr_len = vnet_hdr ? sizeof(struct virtio_net_hdr) : 0;
 401     s->using_vnet_hdr = false;
 402     s->has_ufo = tap_probe_has_ufo(s->fd);
 403     s->enabled = true;
 404     tap_set_offload(&s->nc, 0, 0, 0, 0, 0);
 405     /*
 406      * Make sure host header length is set correctly in tap:
 407      * it might have been modified by another instance of qemu.
 408      */
 409     if (tap_probe_vnet_hdr_len(s->fd, s->host_vnet_hdr_len)) {
 410         tap_fd_set_vnet_hdr_len(s->fd, s->host_vnet_hdr_len);
 411     }
 412     tap_read_poll(s, true);
 413     s->vhost_net = NULL;
 414
 415     s->exit.notify = tap_exit_notify;
 416     qemu_add_exit_notifier(&s->exit);
 417
 418     return s;
 419 }
 420
 421 static void launch_script(const char *setup_script, const char *ifname,
 422                           int fd, Error **errp)
 423 {
 424     int pid, status;
 425     char *args[3];
 426     char **parg;
 427
 428     /* try to launch network script */
 429     pid = fork();
 430     if (pid < 0) {
 431         error_setg_errno(errp, errno, "could not launch network script %s",
 432                          setup_script);
 433         return;
 434     }
 435     if (pid == 0) {
 436         int open_max = sysconf(_SC_OPEN_MAX), i;
 437
 438         for (i = 3; i < open_max; i++) {
 439             if (i != fd) {
 440                 close(i);
 441             }
 442         }
 443         parg = args;
 444         *parg++ = (char *)setup_script;
 445         *parg++ = (char *)ifname;
 446         *parg = NULL;
 447         execv(setup_script, args);
 448         _exit(1);
 449     } else {
 450         while (waitpid(pid, &status, 0) != pid) {
 451             /* loop */
 452         }
 453
 454         if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
 455             return;
 456         }
 457         error_setg(errp, "network script %s failed with status %d",
 458                    setup_script, status);
 459     }
 460 }
 461
 462 static int recv_fd(int c)
 463 {
 464     int fd;
 465     uint8_t msgbuf[CMSG_SPACE(sizeof(fd))];
 466     struct msghdr msg = {
 467         .msg_control = msgbuf,
 468         .msg_controllen = sizeof(msgbuf),
 469     };
 470     struct cmsghdr *cmsg;
 471     struct iovec iov;
 472     uint8_t req[1];
 473     ssize_t len;
 474
 475     cmsg = CMSG_FIRSTHDR(&msg);
 476     cmsg->cmsg_level = SOL_SOCKET;
 477     cmsg->cmsg_type = SCM_RIGHTS;
 478     cmsg->cmsg_len = CMSG_LEN(sizeof(fd));
 479     msg.msg_controllen = cmsg->cmsg_len;
 480
 481     iov.iov_base = req;
 482     iov.iov_len = sizeof(req);
 483
 484     msg.msg_iov = &iov;
 485     msg.msg_iovlen = 1;
 486
 487     len = recvmsg(c, &msg, 0);
 488     if (len > 0) {
 489         memcpy(&fd, CMSG_DATA(cmsg), sizeof(fd));
 490         return fd;
 491     }
 492
 493     return len;
 494 }
 495
 496 static int net_bridge_run_helper(const char *helper, const char *bridge,
 497                                  Error **errp)
 498 {
 499     sigset_t oldmask, mask;
 500     g_autofree char *default_helper = NULL;
 501     int pid, status;
 502     char *args[5];
 503     char **parg;
 504     int sv[2];
 505
 506     sigemptyset(&mask);
 507     sigaddset(&mask, SIGCHLD);
 508     sigprocmask(SIG_BLOCK, &mask, &oldmask);
 509
 510     if (!helper) {
 511         helper = default_helper = get_relocated_path(DEFAULT_BRIDGE_HELPER);
 512     }
 513
 514     if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) {
 515         error_setg_errno(errp, errno, "socketpair() failed");
 516         return -1;
 517     }
 518
 519     /* try to launch bridge helper */
 520     pid = fork();
 521     if (pid < 0) {
 522         error_setg_errno(errp, errno, "Can't fork bridge helper");
 523         return -1;
 524     }
 525     if (pid == 0) {
 526         int open_max = sysconf(_SC_OPEN_MAX), i;
 527         char *fd_buf = NULL;
 528         char *br_buf = NULL;
 529         char *helper_cmd = NULL;
 530
 531         for (i = 3; i < open_max; i++) {
 532             if (i != sv[1]) {
 533                 close(i);
 534             }
 535         }
 536
 537         fd_buf = g_strdup_printf("%s%d", "--fd=", sv[1]);
 538
 539         if (strrchr(helper, ' ') || strrchr(helper, '\t')) {
 540             /* assume helper is a command */
 541
 542             if (strstr(helper, "--br=") == NULL) {
 543                 br_buf = g_strdup_printf("%s%s", "--br=", bridge);
 544             }
 545
 546             helper_cmd = g_strdup_printf("%s %s %s %s", helper,
 547                             "--use-vnet", fd_buf, br_buf ? br_buf : "");
 548
 549             parg = args;
 550             *parg++ = (char *)"sh";
 551             *parg++ = (char *)"-c";
 552             *parg++ = helper_cmd;
 553             *parg++ = NULL;
 554
 555             execv("/bin/sh", args);
 556             g_free(helper_cmd);
 557         } else {
 558             /* assume helper is just the executable path name */
 559
 560             br_buf = g_strdup_printf("%s%s", "--br=", bridge);
 561
 562             parg = args;
 563             *parg++ = (char *)helper;
 564             *parg++ = (char *)"--use-vnet";
 565             *parg++ = fd_buf;
 566             *parg++ = br_buf;
 567             *parg++ = NULL;
 568
 569             execv(helper, args);
 570         }
 571         g_free(fd_buf);
 572         g_free(br_buf);
 573         _exit(1);
 574
 575     } else {
 576         int fd;
 577         int saved_errno;
 578
 579         close(sv[1]);
 580
 581         do {
 582             fd = recv_fd(sv[0]);
 583         } while (fd == -1 && errno == EINTR);
 584         saved_errno = errno;
 585
 586         close(sv[0]);
 587
 588         while (waitpid(pid, &status, 0) != pid) {
 589             /* loop */
 590         }
 591         sigprocmask(SIG_SETMASK, &oldmask, NULL);
 592         if (fd < 0) {
 593             error_setg_errno(errp, saved_errno,
 594                              "failed to recv file descriptor");
 595             return -1;
 596         }
 597         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 598             error_setg(errp, "bridge helper failed");
 599             return -1;
 600         }
 601         return fd;
 602     }
 603 }
 604
 605 int net_init_bridge(const Netdev *netdev, const char *name,
 606                     NetClientState *peer, Error **errp)
 607 {
 608     const NetdevBridgeOptions *bridge;
 609     const char *helper, *br;
 610     TAPState *s;
 611     int fd, vnet_hdr;
 612
 613     assert(netdev->type == NET_CLIENT_DRIVER_BRIDGE);
 614     bridge = &netdev->u.bridge;
 615     helper = bridge->has_helper ? bridge->helper : NULL;
 616     br     = bridge->has_br     ? bridge->br     : DEFAULT_BRIDGE_INTERFACE;
 617
 618     fd = net_bridge_run_helper(helper, br, errp);
 619     if (fd == -1) {
 620         return -1;
 621     }
 622
 623     qemu_set_nonblock(fd);
 624     vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 625     if (vnet_hdr < 0) {
 626         close(fd);
 627         return -1;
 628     }
 629     s = net_tap_fd_init(peer, "bridge", name, fd, vnet_hdr);
 630
 631     snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s,br=%s", helper,
 632              br);
 633
 634     return 0;
 635 }
 636
 637 static int net_tap_init(const NetdevTapOptions *tap, int *vnet_hdr,
 638                         const char *setup_script, char *ifname,
 639                         size_t ifname_sz, int mq_required, Error **errp)
 640 {
 641     Error *err = NULL;
 642     int fd, vnet_hdr_required;
 643
 644     if (tap->has_vnet_hdr) {
 645         *vnet_hdr = tap->vnet_hdr;
 646         vnet_hdr_required = *vnet_hdr;
 647     } else {
 648         *vnet_hdr = 1;
 649         vnet_hdr_required = 0;
 650     }
 651
 652     TFR(fd = tap_open(ifname, ifname_sz, vnet_hdr, vnet_hdr_required,
 653                       mq_required, errp));
 654     if (fd < 0) {
 655         return -1;
 656     }
 657
 658     if (setup_script &&
 659         setup_script[0] != '\0' &&
 660         strcmp(setup_script, "no") != 0) {
 661         launch_script(setup_script, ifname, fd, &err);
 662         if (err) {
 663             error_propagate(errp, err);
 664             close(fd);
 665             return -1;
 666         }
 667     }
 668
 669     return fd;
 670 }
 671
 672 #define MAX_TAP_QUEUES 1024
 673
 674 static void net_init_tap_one(const NetdevTapOptions *tap, NetClientState *peer,
 675                              const char *model, const char *name,
 676                              const char *ifname, const char *script,
 677                              const char *downscript, const char *vhostfdname,
 678                              int vnet_hdr, int fd, Error **errp)
 679 {
 680     Error *err = NULL;
 681     TAPState *s = net_tap_fd_init(peer, model, name, fd, vnet_hdr);
 682     int vhostfd;
 683
 684     tap_set_sndbuf(s->fd, tap, &err);
 685     if (err) {
 686         error_propagate(errp, err);
 687         return;
 688     }
 689
 690     if (tap->has_fd || tap->has_fds) {
 691         snprintf(s->nc.info_str, sizeof(s->nc.info_str), "fd=%d", fd);
 692     } else if (tap->has_helper) {
 693         snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s",
 694                  tap->helper);
 695     } else {
 696         snprintf(s->nc.info_str, sizeof(s->nc.info_str),
 697                  "ifname=%s,script=%s,downscript=%s", ifname, script,
 698                  downscript);
 699
 700         if (strcmp(downscript, "no") != 0) {
 701             snprintf(s->down_script, sizeof(s->down_script), "%s", downscript);
 702             snprintf(s->down_script_arg, sizeof(s->down_script_arg),
 703                      "%s", ifname);
 704         }
 705     }
 706
 707     if (tap->has_vhost ? tap->vhost :
 708         vhostfdname || (tap->has_vhostforce && tap->vhostforce)) {
 709         VhostNetOptions options;
 710
 711         options.backend_type = VHOST_BACKEND_TYPE_KERNEL;
 712         options.net_backend = &s->nc;
 713         if (tap->has_poll_us) {
 714             options.busyloop_timeout = tap->poll_us;
 715         } else {
 716             options.busyloop_timeout = 0;
 717         }
 718
 719         if (vhostfdname) {
 720             int ret;
 721
 722             vhostfd = monitor_fd_param(monitor_cur(), vhostfdname, &err);
 723             if (vhostfd == -1) {
 724                 if (tap->has_vhostforce && tap->vhostforce) {
 725                     error_propagate(errp, err);
 726                 } else {
 727                     warn_report_err(err);
 728                 }
 729                 return;
 730             }
 731             ret = qemu_try_set_nonblock(vhostfd);
 732             if (ret < 0) {
 733                 error_setg_errno(errp, -ret, "%s: Can't use file descriptor %d",
 734                                  name, fd);
 735                 return;
 736             }
 737         } else {
 738             vhostfd = open("/dev/vhost-net", O_RDWR);
 739             if (vhostfd < 0) {
 740                 if (tap->has_vhostforce && tap->vhostforce) {
 741                     error_setg_errno(errp, errno,
 742                                      "tap: open vhost char device failed");
 743                 } else {
 744                     warn_report("tap: open vhost char device failed: %s",
 745                                 strerror(errno));
 746                 }
 747                 return;
 748             }
 749             qemu_set_nonblock(vhostfd);
 750         }
 751         options.opaque = (void *)(uintptr_t)vhostfd;
 752
 753         s->vhost_net = vhost_net_init(&options);
 754         if (!s->vhost_net) {
 755             if (tap->has_vhostforce && tap->vhostforce) {
 756                 error_setg(errp, VHOST_NET_INIT_FAILED);
 757             } else {
 758                 warn_report(VHOST_NET_INIT_FAILED);
 759             }
 760             return;
 761         }
 762     } else if (vhostfdname) {
 763         error_setg(errp, "vhostfd(s)= is not valid without vhost");
 764     }
 765 }
 766
 767 static int get_fds(char *str, char *fds[], int max)
 768 {
 769     char *ptr = str, *this;
 770     size_t len = strlen(str);
 771     int i = 0;
 772
 773     while (i < max && ptr < str + len) {
 774         this = strchr(ptr, ':');
 775
 776         if (this == NULL) {
 777             fds[i] = g_strdup(ptr);
 778         } else {
 779             fds[i] = g_strndup(ptr, this - ptr);
 780         }
 781
 782         i++;
 783         if (this == NULL) {
 784             break;
 785         } else {
 786             ptr = this + 1;
 787         }
 788     }
 789
 790     return i;
 791 }
 792
 793 int net_init_tap(const Netdev *netdev, const char *name,
 794                  NetClientState *peer, Error **errp)
 795 {
 796     const NetdevTapOptions *tap;
 797     int fd, vnet_hdr = 0, i = 0, queues;
 798     /* for the no-fd, no-helper case */
 799     const char *script;
 800     const char *downscript;
 801     Error *err = NULL;
 802     const char *vhostfdname;
 803     char ifname[128];
 804     int ret = 0;
 805
 806     assert(netdev->type == NET_CLIENT_DRIVER_TAP);
 807     tap = &netdev->u.tap;
 808     queues = tap->has_queues ? tap->queues : 1;
 809     vhostfdname = tap->has_vhostfd ? tap->vhostfd : NULL;
 810     script = tap->has_script ? tap->script : NULL;
 811     downscript = tap->has_downscript ? tap->downscript : NULL;
 812
 813     /* QEMU hubs do not support multiqueue tap, in this case peer is set.
 814      * For -netdev, peer is always NULL. */
 815     if (peer && (tap->has_queues || tap->has_fds || tap->has_vhostfds)) {
 816         error_setg(errp, "Multiqueue tap cannot be used with hubs");
 817         return -1;
 818     }
 819
 820     if (tap->has_fd) {
 821         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
 822             tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
 823             tap->has_fds || tap->has_vhostfds) {
 824             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 825                        "helper=, queues=, fds=, and vhostfds= "
 826                        "are invalid with fd=");
 827             return -1;
 828         }
 829
 830         fd = monitor_fd_param(monitor_cur(), tap->fd, errp);
 831         if (fd == -1) {
 832             return -1;
 833         }
 834
 835         ret = qemu_try_set_nonblock(fd);
 836         if (ret < 0) {
 837             error_setg_errno(errp, -ret, "%s: Can't use file descriptor %d",
 838                              name, fd);
 839             close(fd);
 840             return -1;
 841         }
 842
 843         vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 844         if (vnet_hdr < 0) {
 845             close(fd);
 846             return -1;
 847         }
 848
 849         net_init_tap_one(tap, peer, "tap", name, NULL,
 850                          script, downscript,
 851                          vhostfdname, vnet_hdr, fd, &err);
 852         if (err) {
 853             error_propagate(errp, err);
 854             close(fd);
 855             return -1;
 856         }
 857     } else if (tap->has_fds) {
 858         char **fds;
 859         char **vhost_fds;
 860         int nfds = 0, nvhosts = 0;
 861
 862         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
 863             tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
 864             tap->has_vhostfd) {
 865             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 866                        "helper=, queues=, and vhostfd= "
 867                        "are invalid with fds=");
 868             return -1;
 869         }
 870
 871         fds = g_new0(char *, MAX_TAP_QUEUES);
 872         vhost_fds = g_new0(char *, MAX_TAP_QUEUES);
 873
 874         nfds = get_fds(tap->fds, fds, MAX_TAP_QUEUES);
 875         if (tap->has_vhostfds) {
 876             nvhosts = get_fds(tap->vhostfds, vhost_fds, MAX_TAP_QUEUES);
 877             if (nfds != nvhosts) {
 878                 error_setg(errp, "The number of fds passed does not match "
 879                            "the number of vhostfds passed");
 880                 ret = -1;
 881                 goto free_fail;
 882             }
 883         }
 884
 885         for (i = 0; i < nfds; i++) {
 886             fd = monitor_fd_param(monitor_cur(), fds[i], errp);
 887             if (fd == -1) {
 888                 ret = -1;
 889                 goto free_fail;
 890             }
 891
 892             ret = qemu_try_set_nonblock(fd);
 893             if (ret < 0) {
 894                 error_setg_errno(errp, -ret, "%s: Can't use file descriptor %d",
 895                                  name, fd);
 896                 goto free_fail;
 897             }
 898
 899             if (i == 0) {
 900                 vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 901                 if (vnet_hdr < 0) {
 902                     goto free_fail;
 903                 }
 904             } else if (vnet_hdr != tap_probe_vnet_hdr(fd, NULL)) {
 905                 error_setg(errp,
 906                            "vnet_hdr not consistent across given tap fds");
 907                 ret = -1;
 908                 goto free_fail;
 909             }
 910
 911             net_init_tap_one(tap, peer, "tap", name, ifname,
 912                              script, downscript,
 913                              tap->has_vhostfds ? vhost_fds[i] : NULL,
 914                              vnet_hdr, fd, &err);
 915             if (err) {
 916                 error_propagate(errp, err);
 917                 ret = -1;
 918                 goto free_fail;
 919             }
 920         }
 921
 922 free_fail:
 923         for (i = 0; i < nvhosts; i++) {
 924             g_free(vhost_fds[i]);
 925         }
 926         for (i = 0; i < nfds; i++) {
 927             g_free(fds[i]);
 928         }
 929         g_free(fds);
 930         g_free(vhost_fds);
 931         return ret;
 932     } else if (tap->has_helper) {
 933         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
 934             tap->has_vnet_hdr || tap->has_queues || tap->has_vhostfds) {
 935             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 936                        "queues=, and vhostfds= are invalid with helper=");
 937             return -1;
 938         }
 939
 940         fd = net_bridge_run_helper(tap->helper,
 941                                    tap->has_br ?
 942                                    tap->br : DEFAULT_BRIDGE_INTERFACE,
 943                                    errp);
 944         if (fd == -1) {
 945             return -1;
 946         }
 947
 948         qemu_set_nonblock(fd);
 949         vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 950         if (vnet_hdr < 0) {
 951             close(fd);
 952             return -1;
 953         }
 954
 955         net_init_tap_one(tap, peer, "bridge", name, ifname,
 956                          script, downscript, vhostfdname,
 957                          vnet_hdr, fd, &err);
 958         if (err) {
 959             error_propagate(errp, err);
 960             close(fd);
 961             return -1;
 962         }
 963     } else {
 964         g_autofree char *default_script = NULL;
 965         g_autofree char *default_downscript = NULL;
 966         if (tap->has_vhostfds) {
 967             error_setg(errp, "vhostfds= is invalid if fds= wasn't specified");
 968             return -1;
 969         }
 970
 971         if (!script) {
 972             script = default_script = get_relocated_path(DEFAULT_NETWORK_SCRIPT);
 973         }
 974         if (!downscript) {
 975             downscript = default_downscript =
 976                                  get_relocated_path(DEFAULT_NETWORK_DOWN_SCRIPT);
 977         }
 978
 979         if (tap->has_ifname) {
 980             pstrcpy(ifname, sizeof ifname, tap->ifname);
 981         } else {
 982             ifname[0] = '\0';
 983         }
 984
 985         for (i = 0; i < queues; i++) {
 986             fd = net_tap_init(tap, &vnet_hdr, i >= 1 ? "no" : script,
 987                               ifname, sizeof ifname, queues > 1, errp);
 988             if (fd == -1) {
 989                 return -1;
 990             }
 991
 992             if (queues > 1 && i == 0 && !tap->has_ifname) {
 993                 if (tap_fd_get_ifname(fd, ifname)) {
 994                     error_setg(errp, "Fail to get ifname");
 995                     close(fd);
 996                     return -1;
 997                 }
 998             }
 999
1000             net_init_tap_one(tap, peer, "tap", name, ifname,
1001                              i >= 1 ? "no" : script,
1002                              i >= 1 ? "no" : downscript,
1003                              vhostfdname, vnet_hdr, fd, &err);
1004             if (err) {
1005                 error_propagate(errp, err);
1006                 close(fd);
1007                 return -1;
1008             }
1009         }
1010     }
1011
1012     return 0;
1013 }
1014
1015 VHostNetState *tap_get_vhost_net(NetClientState *nc)
1016 {
1017     TAPState *s = DO_UPCAST(TAPState, nc, nc);
1018     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
1019     return s->vhost_net;
1020 }
1021
1022 int tap_enable(NetClientState *nc)
1023 {
1024     TAPState *s = DO_UPCAST(TAPState, nc, nc);
1025     int ret;
1026
1027     if (s->enabled) {
1028         return 0;
1029     } else {
1030         ret = tap_fd_enable(s->fd);
1031         if (ret == 0) {
1032             s->enabled = true;
1033             tap_update_fd_handler(s);
1034         }
1035         return ret;
1036     }
1037 }
1038
1039 int tap_disable(NetClientState *nc)
1040 {
1041     TAPState *s = DO_UPCAST(TAPState, nc, nc);
1042     int ret;
1043
1044     if (s->enabled == 0) {
1045         return 0;
1046     } else {
1047         ret = tap_fd_disable(s->fd);
1048         if (ret == 0) {
1049             qemu_purge_queued_packets(nc);
1050             s->enabled = false;
1051             tap_update_fd_handler(s);
1052         }
1053         return ret;
1054     }
1055 }