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 enabled;
  61     VHostNetState *vhost_net;
  62     unsigned host_vnet_hdr_len;
  63     Notifier exit;
  64 } TAPState;
  65
  66 static void launch_script(const char *setup_script, const char *ifname,
  67                           int fd, Error **errp);
  68
  69 static void tap_send(void *opaque);
  70 static void tap_writable(void *opaque);
  71
  72 static void tap_update_fd_handler(TAPState *s)
  73 {
  74     qemu_set_fd_handler(s->fd,
  75                         s->read_poll && s->enabled ? tap_send : NULL,
  76                         s->write_poll && s->enabled ? tap_writable : NULL,
  77                         s);
  78 }
  79
  80 static void tap_read_poll(TAPState *s, bool enable)
  81 {
  82     s->read_poll = enable;
  83     tap_update_fd_handler(s);
  84 }
  85
  86 static void tap_write_poll(TAPState *s, bool enable)
  87 {
  88     s->write_poll = enable;
  89     tap_update_fd_handler(s);
  90 }
  91
  92 static void tap_writable(void *opaque)
  93 {
  94     TAPState *s = opaque;
  95
  96     tap_write_poll(s, false);
  97
  98     qemu_flush_queued_packets(&s->nc);
  99 }
 100
 101 static ssize_t tap_write_packet(TAPState *s, const struct iovec *iov, int iovcnt)
 102 {
 103     ssize_t len;
 104
 105     do {
 106         len = writev(s->fd, iov, iovcnt);
 107     } while (len == -1 && errno == EINTR);
 108
 109     if (len == -1 && errno == EAGAIN) {
 110         tap_write_poll(s, true);
 111         return 0;
 112     }
 113
 114     return len;
 115 }
 116
 117 static ssize_t tap_receive_iov(NetClientState *nc, const struct iovec *iov,
 118                                int iovcnt)
 119 {
 120     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 121     const struct iovec *iovp = iov;
 122     struct iovec iov_copy[iovcnt + 1];
 123     struct virtio_net_hdr_mrg_rxbuf hdr = { };
 124
 125     if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
 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_raw(NetClientState *nc, const uint8_t *buf, size_t size)
 137 {
 138     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 139     struct iovec iov[2];
 140     int iovcnt = 0;
 141     struct virtio_net_hdr_mrg_rxbuf hdr = { };
 142
 143     if (s->host_vnet_hdr_len) {
 144         iov[iovcnt].iov_base = &hdr;
 145         iov[iovcnt].iov_len  = s->host_vnet_hdr_len;
 146         iovcnt++;
 147     }
 148
 149     iov[iovcnt].iov_base = (char *)buf;
 150     iov[iovcnt].iov_len  = size;
 151     iovcnt++;
 152
 153     return tap_write_packet(s, iov, iovcnt);
 154 }
 155
 156 static ssize_t tap_receive(NetClientState *nc, const uint8_t *buf, size_t size)
 157 {
 158     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 159     struct iovec iov[1];
 160
 161     if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
 162         return tap_receive_raw(nc, buf, size);
 163     }
 164
 165     iov[0].iov_base = (char *)buf;
 166     iov[0].iov_len  = size;
 167
 168     return tap_write_packet(s, iov, 1);
 169 }
 170
 171 #ifndef __sun__
 172 ssize_t tap_read_packet(int tapfd, uint8_t *buf, int maxlen)
 173 {
 174     return read(tapfd, buf, maxlen);
 175 }
 176 #endif
 177
 178 static void tap_send_completed(NetClientState *nc, ssize_t len)
 179 {
 180     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 181     tap_read_poll(s, true);
 182 }
 183
 184 static void tap_send(void *opaque)
 185 {
 186     TAPState *s = opaque;
 187     int size;
 188     int packets = 0;
 189
 190     while (true) {
 191         uint8_t *buf = s->buf;
 192         uint8_t min_pkt[ETH_ZLEN];
 193         size_t min_pktsz = sizeof(min_pkt);
 194
 195         size = tap_read_packet(s->fd, s->buf, sizeof(s->buf));
 196         if (size <= 0) {
 197             break;
 198         }
 199
 200         if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
 201             buf  += s->host_vnet_hdr_len;
 202             size -= s->host_vnet_hdr_len;
 203         }
 204
 205         if (net_peer_needs_padding(&s->nc)) {
 206             if (eth_pad_short_frame(min_pkt, &min_pktsz, buf, size)) {
 207                 buf = min_pkt;
 208                 size = min_pktsz;
 209             }
 210         }
 211
 212         size = qemu_send_packet_async(&s->nc, buf, size, tap_send_completed);
 213         if (size == 0) {
 214             tap_read_poll(s, false);
 215             break;
 216         } else if (size < 0) {
 217             break;
 218         }
 219
 220         /*
 221          * When the host keeps receiving more packets while tap_send() is
 222          * running we can hog the QEMU global mutex.  Limit the number of
 223          * packets that are processed per tap_send() callback to prevent
 224          * stalling the guest.
 225          */
 226         packets++;
 227         if (packets >= 50) {
 228             break;
 229         }
 230     }
 231 }
 232
 233 static bool tap_has_ufo(NetClientState *nc)
 234 {
 235     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 236
 237     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 238
 239     return s->has_ufo;
 240 }
 241
 242 static bool tap_has_vnet_hdr(NetClientState *nc)
 243 {
 244     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 245
 246     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 247
 248     return !!s->host_vnet_hdr_len;
 249 }
 250
 251 static bool tap_has_vnet_hdr_len(NetClientState *nc, int len)
 252 {
 253     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 254
 255     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 256
 257     return !!tap_probe_vnet_hdr_len(s->fd, len);
 258 }
 259
 260 static void tap_set_vnet_hdr_len(NetClientState *nc, int len)
 261 {
 262     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 263
 264     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 265     assert(len == sizeof(struct virtio_net_hdr_mrg_rxbuf) ||
 266            len == sizeof(struct virtio_net_hdr) ||
 267            len == sizeof(struct virtio_net_hdr_v1_hash));
 268
 269     tap_fd_set_vnet_hdr_len(s->fd, len);
 270     s->host_vnet_hdr_len = len;
 271 }
 272
 273 static void tap_using_vnet_hdr(NetClientState *nc, bool using_vnet_hdr)
 274 {
 275     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 276
 277     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 278     assert(!!s->host_vnet_hdr_len == using_vnet_hdr);
 279
 280     s->using_vnet_hdr = using_vnet_hdr;
 281 }
 282
 283 static int tap_set_vnet_le(NetClientState *nc, bool is_le)
 284 {
 285     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 286
 287     return tap_fd_set_vnet_le(s->fd, is_le);
 288 }
 289
 290 static int tap_set_vnet_be(NetClientState *nc, bool is_be)
 291 {
 292     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 293
 294     return tap_fd_set_vnet_be(s->fd, is_be);
 295 }
 296
 297 static void tap_set_offload(NetClientState *nc, int csum, int tso4,
 298                      int tso6, int ecn, int ufo)
 299 {
 300     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 301     if (s->fd < 0) {
 302         return;
 303     }
 304
 305     tap_fd_set_offload(s->fd, csum, tso4, tso6, ecn, ufo);
 306 }
 307
 308 static void tap_exit_notify(Notifier *notifier, void *data)
 309 {
 310     TAPState *s = container_of(notifier, TAPState, exit);
 311     Error *err = NULL;
 312
 313     if (s->down_script[0]) {
 314         launch_script(s->down_script, s->down_script_arg, s->fd, &err);
 315         if (err) {
 316             error_report_err(err);
 317         }
 318     }
 319 }
 320
 321 static void tap_cleanup(NetClientState *nc)
 322 {
 323     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 324
 325     if (s->vhost_net) {
 326         vhost_net_cleanup(s->vhost_net);
 327         g_free(s->vhost_net);
 328         s->vhost_net = NULL;
 329     }
 330
 331     qemu_purge_queued_packets(nc);
 332
 333     tap_exit_notify(&s->exit, NULL);
 334     qemu_remove_exit_notifier(&s->exit);
 335
 336     tap_read_poll(s, false);
 337     tap_write_poll(s, false);
 338     close(s->fd);
 339     s->fd = -1;
 340 }
 341
 342 static void tap_poll(NetClientState *nc, bool enable)
 343 {
 344     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 345     tap_read_poll(s, enable);
 346     tap_write_poll(s, enable);
 347 }
 348
 349 static bool tap_set_steering_ebpf(NetClientState *nc, int prog_fd)
 350 {
 351     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 352     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 353
 354     return tap_fd_set_steering_ebpf(s->fd, prog_fd) == 0;
 355 }
 356
 357 int tap_get_fd(NetClientState *nc)
 358 {
 359     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 360     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 361     return s->fd;
 362 }
 363
 364 /* fd support */
 365
 366 static NetClientInfo net_tap_info = {
 367     .type = NET_CLIENT_DRIVER_TAP,
 368     .size = sizeof(TAPState),
 369     .receive = tap_receive,
 370     .receive_raw = tap_receive_raw,
 371     .receive_iov = tap_receive_iov,
 372     .poll = tap_poll,
 373     .cleanup = tap_cleanup,
 374     .has_ufo = tap_has_ufo,
 375     .has_vnet_hdr = tap_has_vnet_hdr,
 376     .has_vnet_hdr_len = tap_has_vnet_hdr_len,
 377     .using_vnet_hdr = tap_using_vnet_hdr,
 378     .set_offload = tap_set_offload,
 379     .set_vnet_hdr_len = tap_set_vnet_hdr_len,
 380     .set_vnet_le = tap_set_vnet_le,
 381     .set_vnet_be = tap_set_vnet_be,
 382     .set_steering_ebpf = tap_set_steering_ebpf,
 383 };
 384
 385 static TAPState *net_tap_fd_init(NetClientState *peer,
 386                                  const char *model,
 387                                  const char *name,
 388                                  int fd,
 389                                  int vnet_hdr)
 390 {
 391     NetClientState *nc;
 392     TAPState *s;
 393
 394     nc = qemu_new_net_client(&net_tap_info, peer, model, name);
 395
 396     s = DO_UPCAST(TAPState, nc, nc);
 397
 398     s->fd = fd;
 399     s->host_vnet_hdr_len = vnet_hdr ? sizeof(struct virtio_net_hdr) : 0;
 400     s->using_vnet_hdr = false;
 401     s->has_ufo = tap_probe_has_ufo(s->fd);
 402     s->enabled = true;
 403     tap_set_offload(&s->nc, 0, 0, 0, 0, 0);
 404     /*
 405      * Make sure host header length is set correctly in tap:
 406      * it might have been modified by another instance of qemu.
 407      */
 408     if (tap_probe_vnet_hdr_len(s->fd, s->host_vnet_hdr_len)) {
 409         tap_fd_set_vnet_hdr_len(s->fd, s->host_vnet_hdr_len);
 410     }
 411     tap_read_poll(s, true);
 412     s->vhost_net = NULL;
 413
 414     s->exit.notify = tap_exit_notify;
 415     qemu_add_exit_notifier(&s->exit);
 416
 417     return s;
 418 }
 419
 420 static void launch_script(const char *setup_script, const char *ifname,
 421                           int fd, Error **errp)
 422 {
 423     int pid, status;
 424     char *args[3];
 425     char **parg;
 426
 427     /* try to launch network script */
 428     pid = fork();
 429     if (pid < 0) {
 430         error_setg_errno(errp, errno, "could not launch network script %s",
 431                          setup_script);
 432         return;
 433     }
 434     if (pid == 0) {
 435         int open_max = sysconf(_SC_OPEN_MAX), i;
 436
 437         for (i = 3; i < open_max; i++) {
 438             if (i != fd) {
 439                 close(i);
 440             }
 441         }
 442         parg = args;
 443         *parg++ = (char *)setup_script;
 444         *parg++ = (char *)ifname;
 445         *parg = NULL;
 446         execv(setup_script, args);
 447         _exit(1);
 448     } else {
 449         while (waitpid(pid, &status, 0) != pid) {
 450             /* loop */
 451         }
 452
 453         if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
 454             return;
 455         }
 456         error_setg(errp, "network script %s failed with status %d",
 457                    setup_script, status);
 458     }
 459 }
 460
 461 static int recv_fd(int c)
 462 {
 463     int fd;
 464     uint8_t msgbuf[CMSG_SPACE(sizeof(fd))];
 465     struct msghdr msg = {
 466         .msg_control = msgbuf,
 467         .msg_controllen = sizeof(msgbuf),
 468     };
 469     struct cmsghdr *cmsg;
 470     struct iovec iov;
 471     uint8_t req[1];
 472     ssize_t len;
 473
 474     cmsg = CMSG_FIRSTHDR(&msg);
 475     cmsg->cmsg_level = SOL_SOCKET;
 476     cmsg->cmsg_type = SCM_RIGHTS;
 477     cmsg->cmsg_len = CMSG_LEN(sizeof(fd));
 478     msg.msg_controllen = cmsg->cmsg_len;
 479
 480     iov.iov_base = req;
 481     iov.iov_len = sizeof(req);
 482
 483     msg.msg_iov = &iov;
 484     msg.msg_iovlen = 1;
 485
 486     len = recvmsg(c, &msg, 0);
 487     if (len > 0) {
 488         memcpy(&fd, CMSG_DATA(cmsg), sizeof(fd));
 489         return fd;
 490     }
 491
 492     return len;
 493 }
 494
 495 static int net_bridge_run_helper(const char *helper, const char *bridge,
 496                                  Error **errp)
 497 {
 498     sigset_t oldmask, mask;
 499     g_autofree char *default_helper = NULL;
 500     int pid, status;
 501     char *args[5];
 502     char **parg;
 503     int sv[2];
 504
 505     sigemptyset(&mask);
 506     sigaddset(&mask, SIGCHLD);
 507     sigprocmask(SIG_BLOCK, &mask, &oldmask);
 508
 509     if (!helper) {
 510         helper = default_helper = get_relocated_path(DEFAULT_BRIDGE_HELPER);
 511     }
 512
 513     if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) {
 514         error_setg_errno(errp, errno, "socketpair() failed");
 515         return -1;
 516     }
 517
 518     /* try to launch bridge helper */
 519     pid = fork();
 520     if (pid < 0) {
 521         error_setg_errno(errp, errno, "Can't fork bridge helper");
 522         return -1;
 523     }
 524     if (pid == 0) {
 525         int open_max = sysconf(_SC_OPEN_MAX), i;
 526         char *fd_buf = NULL;
 527         char *br_buf = NULL;
 528         char *helper_cmd = NULL;
 529
 530         for (i = 3; i < open_max; i++) {
 531             if (i != sv[1]) {
 532                 close(i);
 533             }
 534         }
 535
 536         fd_buf = g_strdup_printf("%s%d", "--fd=", sv[1]);
 537
 538         if (strrchr(helper, ' ') || strrchr(helper, '\t')) {
 539             /* assume helper is a command */
 540
 541             if (strstr(helper, "--br=") == NULL) {
 542                 br_buf = g_strdup_printf("%s%s", "--br=", bridge);
 543             }
 544
 545             helper_cmd = g_strdup_printf("%s %s %s %s", helper,
 546                             "--use-vnet", fd_buf, br_buf ? br_buf : "");
 547
 548             parg = args;
 549             *parg++ = (char *)"sh";
 550             *parg++ = (char *)"-c";
 551             *parg++ = helper_cmd;
 552             *parg++ = NULL;
 553
 554             execv("/bin/sh", args);
 555             g_free(helper_cmd);
 556         } else {
 557             /* assume helper is just the executable path name */
 558
 559             br_buf = g_strdup_printf("%s%s", "--br=", bridge);
 560
 561             parg = args;
 562             *parg++ = (char *)helper;
 563             *parg++ = (char *)"--use-vnet";
 564             *parg++ = fd_buf;
 565             *parg++ = br_buf;
 566             *parg++ = NULL;
 567
 568             execv(helper, args);
 569         }
 570         g_free(fd_buf);
 571         g_free(br_buf);
 572         _exit(1);
 573
 574     } else {
 575         int fd;
 576         int saved_errno;
 577
 578         close(sv[1]);
 579
 580         do {
 581             fd = recv_fd(sv[0]);
 582         } while (fd == -1 && errno == EINTR);
 583         saved_errno = errno;
 584
 585         close(sv[0]);
 586
 587         while (waitpid(pid, &status, 0) != pid) {
 588             /* loop */
 589         }
 590         sigprocmask(SIG_SETMASK, &oldmask, NULL);
 591         if (fd < 0) {
 592             error_setg_errno(errp, saved_errno,
 593                              "failed to recv file descriptor");
 594             return -1;
 595         }
 596         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 597             error_setg(errp, "bridge helper failed");
 598             return -1;
 599         }
 600         return fd;
 601     }
 602 }
 603
 604 int net_init_bridge(const Netdev *netdev, const char *name,
 605                     NetClientState *peer, Error **errp)
 606 {
 607     const NetdevBridgeOptions *bridge;
 608     const char *helper, *br;
 609     TAPState *s;
 610     int fd, vnet_hdr;
 611
 612     assert(netdev->type == NET_CLIENT_DRIVER_BRIDGE);
 613     bridge = &netdev->u.bridge;
 614     helper = bridge->helper;
 615     br     = bridge->br ?: DEFAULT_BRIDGE_INTERFACE;
 616
 617     fd = net_bridge_run_helper(helper, br, errp);
 618     if (fd == -1) {
 619         return -1;
 620     }
 621
 622     if (!g_unix_set_fd_nonblocking(fd, true, NULL)) {
 623         error_setg_errno(errp, errno, "Failed to set FD nonblocking");
 624         return -1;
 625     }
 626     vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 627     if (vnet_hdr < 0) {
 628         close(fd);
 629         return -1;
 630     }
 631     s = net_tap_fd_init(peer, "bridge", name, fd, vnet_hdr);
 632
 633     qemu_set_info_str(&s->nc, "helper=%s,br=%s", helper, br);
 634
 635     return 0;
 636 }
 637
 638 static int net_tap_init(const NetdevTapOptions *tap, int *vnet_hdr,
 639                         const char *setup_script, char *ifname,
 640                         size_t ifname_sz, int mq_required, Error **errp)
 641 {
 642     Error *err = NULL;
 643     int fd, vnet_hdr_required;
 644
 645     if (tap->has_vnet_hdr) {
 646         *vnet_hdr = tap->vnet_hdr;
 647         vnet_hdr_required = *vnet_hdr;
 648     } else {
 649         *vnet_hdr = 1;
 650         vnet_hdr_required = 0;
 651     }
 652
 653     TFR(fd = tap_open(ifname, ifname_sz, vnet_hdr, vnet_hdr_required,
 654                       mq_required, errp));
 655     if (fd < 0) {
 656         return -1;
 657     }
 658
 659     if (setup_script &&
 660         setup_script[0] != '\0' &&
 661         strcmp(setup_script, "no") != 0) {
 662         launch_script(setup_script, ifname, fd, &err);
 663         if (err) {
 664             error_propagate(errp, err);
 665             close(fd);
 666             return -1;
 667         }
 668     }
 669
 670     return fd;
 671 }
 672
 673 #define MAX_TAP_QUEUES 1024
 674
 675 static void net_init_tap_one(const NetdevTapOptions *tap, NetClientState *peer,
 676                              const char *model, const char *name,
 677                              const char *ifname, const char *script,
 678                              const char *downscript, const char *vhostfdname,
 679                              int vnet_hdr, int fd, Error **errp)
 680 {
 681     Error *err = NULL;
 682     TAPState *s = net_tap_fd_init(peer, model, name, fd, vnet_hdr);
 683     int vhostfd;
 684
 685     tap_set_sndbuf(s->fd, tap, &err);
 686     if (err) {
 687         error_propagate(errp, err);
 688         goto failed;
 689     }
 690
 691     if (tap->fd || tap->fds) {
 692         qemu_set_info_str(&s->nc, "fd=%d", fd);
 693     } else if (tap->helper) {
 694         qemu_set_info_str(&s->nc, "helper=%s", tap->helper);
 695     } else {
 696         qemu_set_info_str(&s->nc, "ifname=%s,script=%s,downscript=%s", ifname,
 697                           script, downscript);
 698
 699         if (strcmp(downscript, "no") != 0) {
 700             snprintf(s->down_script, sizeof(s->down_script), "%s", downscript);
 701             snprintf(s->down_script_arg, sizeof(s->down_script_arg),
 702                      "%s", ifname);
 703         }
 704     }
 705
 706     if (tap->has_vhost ? tap->vhost :
 707         vhostfdname || (tap->has_vhostforce && tap->vhostforce)) {
 708         VhostNetOptions options;
 709
 710         options.backend_type = VHOST_BACKEND_TYPE_KERNEL;
 711         options.net_backend = &s->nc;
 712         if (tap->has_poll_us) {
 713             options.busyloop_timeout = tap->poll_us;
 714         } else {
 715             options.busyloop_timeout = 0;
 716         }
 717
 718         if (vhostfdname) {
 719             vhostfd = monitor_fd_param(monitor_cur(), vhostfdname, &err);
 720             if (vhostfd == -1) {
 721                 if (tap->has_vhostforce && tap->vhostforce) {
 722                     error_propagate(errp, err);
 723                 } else {
 724                     warn_report_err(err);
 725                 }
 726                 goto failed;
 727             }
 728             if (!g_unix_set_fd_nonblocking(vhostfd, true, NULL)) {
 729                 error_setg_errno(errp, errno, "%s: Can't use file descriptor %d",
 730                                  name, fd);
 731                 goto failed;
 732             }
 733         } else {
 734             vhostfd = open("/dev/vhost-net", O_RDWR);
 735             if (vhostfd < 0) {
 736                 if (tap->has_vhostforce && tap->vhostforce) {
 737                     error_setg_errno(errp, errno,
 738                                      "tap: open vhost char device failed");
 739                 } else {
 740                     warn_report("tap: open vhost char device failed: %s",
 741                                 strerror(errno));
 742                 }
 743                 goto failed;
 744             }
 745             if (!g_unix_set_fd_nonblocking(vhostfd, true, NULL)) {
 746                 error_setg_errno(errp, errno, "Failed to set FD nonblocking");
 747                 goto failed;
 748             }
 749         }
 750         options.opaque = (void *)(uintptr_t)vhostfd;
 751         options.nvqs = 2;
 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             goto failed;
 761         }
 762     } else if (vhostfdname) {
 763         error_setg(errp, "vhostfd(s)= is not valid without vhost");
 764         goto failed;
 765     }
 766
 767     return;
 768
 769 failed:
 770     qemu_del_net_client(&s->nc);
 771 }
 772
 773 static int get_fds(char *str, char *fds[], int max)
 774 {
 775     char *ptr = str, *this;
 776     size_t len = strlen(str);
 777     int i = 0;
 778
 779     while (i < max && ptr < str + len) {
 780         this = strchr(ptr, ':');
 781
 782         if (this == NULL) {
 783             fds[i] = g_strdup(ptr);
 784         } else {
 785             fds[i] = g_strndup(ptr, this - ptr);
 786         }
 787
 788         i++;
 789         if (this == NULL) {
 790             break;
 791         } else {
 792             ptr = this + 1;
 793         }
 794     }
 795
 796     return i;
 797 }
 798
 799 int net_init_tap(const Netdev *netdev, const char *name,
 800                  NetClientState *peer, Error **errp)
 801 {
 802     const NetdevTapOptions *tap;
 803     int fd, vnet_hdr = 0, i = 0, queues;
 804     /* for the no-fd, no-helper case */
 805     const char *script;
 806     const char *downscript;
 807     Error *err = NULL;
 808     const char *vhostfdname;
 809     char ifname[128];
 810     int ret = 0;
 811
 812     assert(netdev->type == NET_CLIENT_DRIVER_TAP);
 813     tap = &netdev->u.tap;
 814     queues = tap->has_queues ? tap->queues : 1;
 815     vhostfdname = tap->vhostfd;
 816     script = tap->script;
 817     downscript = tap->downscript;
 818
 819     /* QEMU hubs do not support multiqueue tap, in this case peer is set.
 820      * For -netdev, peer is always NULL. */
 821     if (peer && (tap->has_queues || tap->fds || tap->vhostfds)) {
 822         error_setg(errp, "Multiqueue tap cannot be used with hubs");
 823         return -1;
 824     }
 825
 826     if (tap->fd) {
 827         if (tap->ifname || tap->script || tap->downscript ||
 828             tap->has_vnet_hdr || tap->helper || tap->has_queues ||
 829             tap->fds || tap->vhostfds) {
 830             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 831                        "helper=, queues=, fds=, and vhostfds= "
 832                        "are invalid with fd=");
 833             return -1;
 834         }
 835
 836         fd = monitor_fd_param(monitor_cur(), tap->fd, errp);
 837         if (fd == -1) {
 838             return -1;
 839         }
 840
 841         if (!g_unix_set_fd_nonblocking(fd, true, NULL)) {
 842             error_setg_errno(errp, errno, "%s: Can't use file descriptor %d",
 843                              name, fd);
 844             close(fd);
 845             return -1;
 846         }
 847
 848         vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 849         if (vnet_hdr < 0) {
 850             close(fd);
 851             return -1;
 852         }
 853
 854         net_init_tap_one(tap, peer, "tap", name, NULL,
 855                          script, downscript,
 856                          vhostfdname, vnet_hdr, fd, &err);
 857         if (err) {
 858             error_propagate(errp, err);
 859             close(fd);
 860             return -1;
 861         }
 862     } else if (tap->fds) {
 863         char **fds;
 864         char **vhost_fds;
 865         int nfds = 0, nvhosts = 0;
 866
 867         if (tap->ifname || tap->script || tap->downscript ||
 868             tap->has_vnet_hdr || tap->helper || tap->has_queues ||
 869             tap->vhostfd) {
 870             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 871                        "helper=, queues=, and vhostfd= "
 872                        "are invalid with fds=");
 873             return -1;
 874         }
 875
 876         fds = g_new0(char *, MAX_TAP_QUEUES);
 877         vhost_fds = g_new0(char *, MAX_TAP_QUEUES);
 878
 879         nfds = get_fds(tap->fds, fds, MAX_TAP_QUEUES);
 880         if (tap->vhostfds) {
 881             nvhosts = get_fds(tap->vhostfds, vhost_fds, MAX_TAP_QUEUES);
 882             if (nfds != nvhosts) {
 883                 error_setg(errp, "The number of fds passed does not match "
 884                            "the number of vhostfds passed");
 885                 ret = -1;
 886                 goto free_fail;
 887             }
 888         }
 889
 890         for (i = 0; i < nfds; i++) {
 891             fd = monitor_fd_param(monitor_cur(), fds[i], errp);
 892             if (fd == -1) {
 893                 ret = -1;
 894                 goto free_fail;
 895             }
 896
 897             ret = g_unix_set_fd_nonblocking(fd, true, NULL);
 898             if (!ret) {
 899                 error_setg_errno(errp, errno, "%s: Can't use file descriptor %d",
 900                                  name, fd);
 901                 goto free_fail;
 902             }
 903
 904             if (i == 0) {
 905                 vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 906                 if (vnet_hdr < 0) {
 907                     ret = -1;
 908                     goto free_fail;
 909                 }
 910             } else if (vnet_hdr != tap_probe_vnet_hdr(fd, NULL)) {
 911                 error_setg(errp,
 912                            "vnet_hdr not consistent across given tap fds");
 913                 ret = -1;
 914                 goto free_fail;
 915             }
 916
 917             net_init_tap_one(tap, peer, "tap", name, ifname,
 918                              script, downscript,
 919                              tap->vhostfds ? vhost_fds[i] : NULL,
 920                              vnet_hdr, fd, &err);
 921             if (err) {
 922                 error_propagate(errp, err);
 923                 ret = -1;
 924                 goto free_fail;
 925             }
 926         }
 927
 928 free_fail:
 929         for (i = 0; i < nvhosts; i++) {
 930             g_free(vhost_fds[i]);
 931         }
 932         for (i = 0; i < nfds; i++) {
 933             g_free(fds[i]);
 934         }
 935         g_free(fds);
 936         g_free(vhost_fds);
 937         return ret;
 938     } else if (tap->helper) {
 939         if (tap->ifname || tap->script || tap->downscript ||
 940             tap->has_vnet_hdr || tap->has_queues || tap->vhostfds) {
 941             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 942                        "queues=, and vhostfds= are invalid with helper=");
 943             return -1;
 944         }
 945
 946         fd = net_bridge_run_helper(tap->helper,
 947                                    tap->br ?: DEFAULT_BRIDGE_INTERFACE,
 948                                    errp);
 949         if (fd == -1) {
 950             return -1;
 951         }
 952
 953         if (!g_unix_set_fd_nonblocking(fd, true, NULL)) {
 954             error_setg_errno(errp, errno, "Failed to set FD nonblocking");
 955             return -1;
 956         }
 957         vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 958         if (vnet_hdr < 0) {
 959             close(fd);
 960             return -1;
 961         }
 962
 963         net_init_tap_one(tap, peer, "bridge", name, ifname,
 964                          script, downscript, vhostfdname,
 965                          vnet_hdr, fd, &err);
 966         if (err) {
 967             error_propagate(errp, err);
 968             close(fd);
 969             return -1;
 970         }
 971     } else {
 972         g_autofree char *default_script = NULL;
 973         g_autofree char *default_downscript = NULL;
 974         if (tap->vhostfds) {
 975             error_setg(errp, "vhostfds= is invalid if fds= wasn't specified");
 976             return -1;
 977         }
 978
 979         if (!script) {
 980             script = default_script = get_relocated_path(DEFAULT_NETWORK_SCRIPT);
 981         }
 982         if (!downscript) {
 983             downscript = default_downscript =
 984                                  get_relocated_path(DEFAULT_NETWORK_DOWN_SCRIPT);
 985         }
 986
 987         if (tap->ifname) {
 988             pstrcpy(ifname, sizeof ifname, tap->ifname);
 989         } else {
 990             ifname[0] = '\0';
 991         }
 992
 993         for (i = 0; i < queues; i++) {
 994             fd = net_tap_init(tap, &vnet_hdr, i >= 1 ? "no" : script,
 995                               ifname, sizeof ifname, queues > 1, errp);
 996             if (fd == -1) {
 997                 return -1;
 998             }
 999
1000             if (queues > 1 && i == 0 && !tap->ifname) {
1001                 if (tap_fd_get_ifname(fd, ifname)) {
1002                     error_setg(errp, "Fail to get ifname");
1003                     close(fd);
1004                     return -1;
1005                 }
1006             }
1007
1008             net_init_tap_one(tap, peer, "tap", name, ifname,
1009                              i >= 1 ? "no" : script,
1010                              i >= 1 ? "no" : downscript,
1011                              vhostfdname, vnet_hdr, fd, &err);
1012             if (err) {
1013                 error_propagate(errp, err);
1014                 close(fd);
1015                 return -1;
1016             }
1017         }
1018     }
1019
1020     return 0;
1021 }
1022
1023 VHostNetState *tap_get_vhost_net(NetClientState *nc)
1024 {
1025     TAPState *s = DO_UPCAST(TAPState, nc, nc);
1026     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
1027     return s->vhost_net;
1028 }
1029
1030 int tap_enable(NetClientState *nc)
1031 {
1032     TAPState *s = DO_UPCAST(TAPState, nc, nc);
1033     int ret;
1034
1035     if (s->enabled) {
1036         return 0;
1037     } else {
1038         ret = tap_fd_enable(s->fd);
1039         if (ret == 0) {
1040             s->enabled = true;
1041             tap_update_fd_handler(s);
1042         }
1043         return ret;
1044     }
1045 }
1046
1047 int tap_disable(NetClientState *nc)
1048 {
1049     TAPState *s = DO_UPCAST(TAPState, nc, nc);
1050     int ret;
1051
1052     if (s->enabled == 0) {
1053         return 0;
1054     } else {
1055         ret = tap_fd_disable(s->fd);
1056         if (ret == 0) {
1057             qemu_purge_queued_packets(nc);
1058             s->enabled = false;
1059             tap_update_fd_handler(s);
1060         }
1061         return ret;
1062     }
1063 }