net/tap.c

   1 /*
   2  * QEMU System Emulator
   3  *
   4  * Copyright (c) 2003-2008 Fabrice Bellard
   5  * Copyright (c) 2009 Red Hat, Inc.
   6  *
   7  * Permission is hereby granted, free of charge, to any person obtaining a copy
   8  * of this software and associated documentation files (the "Software"), to deal
   9  * in the Software without restriction, including without limitation the rights
  10  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  11  * copies of the Software, and to permit persons to whom the Software is
  12  * furnished to do so, subject to the following conditions:
  13  *
  14  * The above copyright notice and this permission notice shall be included in
  15  * all copies or substantial portions of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  23  * THE SOFTWARE.
  24  */
  25
  26 #include "qemu/osdep.h"
  27 #include "tap_int.h"
  28
  29
  30 #include <sys/ioctl.h>
  31 #include <sys/wait.h>
  32 #include <sys/socket.h>
  33 #include <net/if.h>
  34
  35 #include "net/net.h"
  36 #include "clients.h"
  37 #include "monitor/monitor.h"
  38 #include "sysemu/sysemu.h"
  39 #include "qapi/error.h"
  40 #include "qemu-common.h"
  41 #include "qemu/cutils.h"
  42 #include "qemu/error-report.h"
  43 #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
 193         size = tap_read_packet(s->fd, s->buf, sizeof(s->buf));
 194         if (size <= 0) {
 195             break;
 196         }
 197
 198         if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
 199             buf  += s->host_vnet_hdr_len;
 200             size -= s->host_vnet_hdr_len;
 201         }
 202
 203         size = qemu_send_packet_async(&s->nc, buf, size, tap_send_completed);
 204         if (size == 0) {
 205             tap_read_poll(s, false);
 206             break;
 207         } else if (size < 0) {
 208             break;
 209         }
 210
 211         /*
 212          * When the host keeps receiving more packets while tap_send() is
 213          * running we can hog the QEMU global mutex.  Limit the number of
 214          * packets that are processed per tap_send() callback to prevent
 215          * stalling the guest.
 216          */
 217         packets++;
 218         if (packets >= 50) {
 219             break;
 220         }
 221     }
 222 }
 223
 224 static bool tap_has_ufo(NetClientState *nc)
 225 {
 226     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 227
 228     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 229
 230     return s->has_ufo;
 231 }
 232
 233 static bool tap_has_vnet_hdr(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->host_vnet_hdr_len;
 240 }
 241
 242 static bool tap_has_vnet_hdr_len(NetClientState *nc, int len)
 243 {
 244     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 245
 246     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 247
 248     return !!tap_probe_vnet_hdr_len(s->fd, len);
 249 }
 250
 251 static void tap_set_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     assert(len == sizeof(struct virtio_net_hdr_mrg_rxbuf) ||
 257            len == sizeof(struct virtio_net_hdr) ||
 258            len == sizeof(struct virtio_net_hdr_v1_hash));
 259
 260     tap_fd_set_vnet_hdr_len(s->fd, len);
 261     s->host_vnet_hdr_len = len;
 262 }
 263
 264 static void tap_using_vnet_hdr(NetClientState *nc, bool using_vnet_hdr)
 265 {
 266     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 267
 268     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 269     assert(!!s->host_vnet_hdr_len == using_vnet_hdr);
 270
 271     s->using_vnet_hdr = using_vnet_hdr;
 272 }
 273
 274 static int tap_set_vnet_le(NetClientState *nc, bool is_le)
 275 {
 276     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 277
 278     return tap_fd_set_vnet_le(s->fd, is_le);
 279 }
 280
 281 static int tap_set_vnet_be(NetClientState *nc, bool is_be)
 282 {
 283     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 284
 285     return tap_fd_set_vnet_be(s->fd, is_be);
 286 }
 287
 288 static void tap_set_offload(NetClientState *nc, int csum, int tso4,
 289                      int tso6, int ecn, int ufo)
 290 {
 291     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 292     if (s->fd < 0) {
 293         return;
 294     }
 295
 296     tap_fd_set_offload(s->fd, csum, tso4, tso6, ecn, ufo);
 297 }
 298
 299 static void tap_exit_notify(Notifier *notifier, void *data)
 300 {
 301     TAPState *s = container_of(notifier, TAPState, exit);
 302     Error *err = NULL;
 303
 304     if (s->down_script[0]) {
 305         launch_script(s->down_script, s->down_script_arg, s->fd, &err);
 306         if (err) {
 307             error_report_err(err);
 308         }
 309     }
 310 }
 311
 312 static void tap_cleanup(NetClientState *nc)
 313 {
 314     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 315
 316     if (s->vhost_net) {
 317         vhost_net_cleanup(s->vhost_net);
 318         g_free(s->vhost_net);
 319         s->vhost_net = NULL;
 320     }
 321
 322     qemu_purge_queued_packets(nc);
 323
 324     tap_exit_notify(&s->exit, NULL);
 325     qemu_remove_exit_notifier(&s->exit);
 326
 327     tap_read_poll(s, false);
 328     tap_write_poll(s, false);
 329     close(s->fd);
 330     s->fd = -1;
 331 }
 332
 333 static void tap_poll(NetClientState *nc, bool enable)
 334 {
 335     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 336     tap_read_poll(s, enable);
 337     tap_write_poll(s, enable);
 338 }
 339
 340 int tap_get_fd(NetClientState *nc)
 341 {
 342     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 343     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 344     return s->fd;
 345 }
 346
 347 /* fd support */
 348
 349 static NetClientInfo net_tap_info = {
 350     .type = NET_CLIENT_DRIVER_TAP,
 351     .size = sizeof(TAPState),
 352     .receive = tap_receive,
 353     .receive_raw = tap_receive_raw,
 354     .receive_iov = tap_receive_iov,
 355     .poll = tap_poll,
 356     .cleanup = tap_cleanup,
 357     .has_ufo = tap_has_ufo,
 358     .has_vnet_hdr = tap_has_vnet_hdr,
 359     .has_vnet_hdr_len = tap_has_vnet_hdr_len,
 360     .using_vnet_hdr = tap_using_vnet_hdr,
 361     .set_offload = tap_set_offload,
 362     .set_vnet_hdr_len = tap_set_vnet_hdr_len,
 363     .set_vnet_le = tap_set_vnet_le,
 364     .set_vnet_be = tap_set_vnet_be,
 365 };
 366
 367 static TAPState *net_tap_fd_init(NetClientState *peer,
 368                                  const char *model,
 369                                  const char *name,
 370                                  int fd,
 371                                  int vnet_hdr)
 372 {
 373     NetClientState *nc;
 374     TAPState *s;
 375
 376     nc = qemu_new_net_client(&net_tap_info, peer, model, name);
 377
 378     s = DO_UPCAST(TAPState, nc, nc);
 379
 380     s->fd = fd;
 381     s->host_vnet_hdr_len = vnet_hdr ? sizeof(struct virtio_net_hdr) : 0;
 382     s->using_vnet_hdr = false;
 383     s->has_ufo = tap_probe_has_ufo(s->fd);
 384     s->enabled = true;
 385     tap_set_offload(&s->nc, 0, 0, 0, 0, 0);
 386     /*
 387      * Make sure host header length is set correctly in tap:
 388      * it might have been modified by another instance of qemu.
 389      */
 390     if (tap_probe_vnet_hdr_len(s->fd, s->host_vnet_hdr_len)) {
 391         tap_fd_set_vnet_hdr_len(s->fd, s->host_vnet_hdr_len);
 392     }
 393     tap_read_poll(s, true);
 394     s->vhost_net = NULL;
 395
 396     s->exit.notify = tap_exit_notify;
 397     qemu_add_exit_notifier(&s->exit);
 398
 399     return s;
 400 }
 401
 402 static void launch_script(const char *setup_script, const char *ifname,
 403                           int fd, Error **errp)
 404 {
 405     int pid, status;
 406     char *args[3];
 407     char **parg;
 408
 409     /* try to launch network script */
 410     pid = fork();
 411     if (pid < 0) {
 412         error_setg_errno(errp, errno, "could not launch network script %s",
 413                          setup_script);
 414         return;
 415     }
 416     if (pid == 0) {
 417         int open_max = sysconf(_SC_OPEN_MAX), i;
 418
 419         for (i = 3; i < open_max; i++) {
 420             if (i != fd) {
 421                 close(i);
 422             }
 423         }
 424         parg = args;
 425         *parg++ = (char *)setup_script;
 426         *parg++ = (char *)ifname;
 427         *parg = NULL;
 428         execv(setup_script, args);
 429         _exit(1);
 430     } else {
 431         while (waitpid(pid, &status, 0) != pid) {
 432             /* loop */
 433         }
 434
 435         if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
 436             return;
 437         }
 438         error_setg(errp, "network script %s failed with status %d",
 439                    setup_script, status);
 440     }
 441 }
 442
 443 static int recv_fd(int c)
 444 {
 445     int fd;
 446     uint8_t msgbuf[CMSG_SPACE(sizeof(fd))];
 447     struct msghdr msg = {
 448         .msg_control = msgbuf,
 449         .msg_controllen = sizeof(msgbuf),
 450     };
 451     struct cmsghdr *cmsg;
 452     struct iovec iov;
 453     uint8_t req[1];
 454     ssize_t len;
 455
 456     cmsg = CMSG_FIRSTHDR(&msg);
 457     cmsg->cmsg_level = SOL_SOCKET;
 458     cmsg->cmsg_type = SCM_RIGHTS;
 459     cmsg->cmsg_len = CMSG_LEN(sizeof(fd));
 460     msg.msg_controllen = cmsg->cmsg_len;
 461
 462     iov.iov_base = req;
 463     iov.iov_len = sizeof(req);
 464
 465     msg.msg_iov = &iov;
 466     msg.msg_iovlen = 1;
 467
 468     len = recvmsg(c, &msg, 0);
 469     if (len > 0) {
 470         memcpy(&fd, CMSG_DATA(cmsg), sizeof(fd));
 471         return fd;
 472     }
 473
 474     return len;
 475 }
 476
 477 static int net_bridge_run_helper(const char *helper, const char *bridge,
 478                                  Error **errp)
 479 {
 480     sigset_t oldmask, mask;
 481     int pid, status;
 482     char *args[5];
 483     char **parg;
 484     int sv[2];
 485
 486     sigemptyset(&mask);
 487     sigaddset(&mask, SIGCHLD);
 488     sigprocmask(SIG_BLOCK, &mask, &oldmask);
 489
 490     if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) {
 491         error_setg_errno(errp, errno, "socketpair() failed");
 492         return -1;
 493     }
 494
 495     /* try to launch bridge helper */
 496     pid = fork();
 497     if (pid < 0) {
 498         error_setg_errno(errp, errno, "Can't fork bridge helper");
 499         return -1;
 500     }
 501     if (pid == 0) {
 502         int open_max = sysconf(_SC_OPEN_MAX), i;
 503         char *fd_buf = NULL;
 504         char *br_buf = NULL;
 505         char *helper_cmd = NULL;
 506
 507         for (i = 3; i < open_max; i++) {
 508             if (i != sv[1]) {
 509                 close(i);
 510             }
 511         }
 512
 513         fd_buf = g_strdup_printf("%s%d", "--fd=", sv[1]);
 514
 515         if (strrchr(helper, ' ') || strrchr(helper, '\t')) {
 516             /* assume helper is a command */
 517
 518             if (strstr(helper, "--br=") == NULL) {
 519                 br_buf = g_strdup_printf("%s%s", "--br=", bridge);
 520             }
 521
 522             helper_cmd = g_strdup_printf("%s %s %s %s", helper,
 523                             "--use-vnet", fd_buf, br_buf ? br_buf : "");
 524
 525             parg = args;
 526             *parg++ = (char *)"sh";
 527             *parg++ = (char *)"-c";
 528             *parg++ = helper_cmd;
 529             *parg++ = NULL;
 530
 531             execv("/bin/sh", args);
 532             g_free(helper_cmd);
 533         } else {
 534             /* assume helper is just the executable path name */
 535
 536             br_buf = g_strdup_printf("%s%s", "--br=", bridge);
 537
 538             parg = args;
 539             *parg++ = (char *)helper;
 540             *parg++ = (char *)"--use-vnet";
 541             *parg++ = fd_buf;
 542             *parg++ = br_buf;
 543             *parg++ = NULL;
 544
 545             execv(helper, args);
 546         }
 547         g_free(fd_buf);
 548         g_free(br_buf);
 549         _exit(1);
 550
 551     } else {
 552         int fd;
 553         int saved_errno;
 554
 555         close(sv[1]);
 556
 557         do {
 558             fd = recv_fd(sv[0]);
 559         } while (fd == -1 && errno == EINTR);
 560         saved_errno = errno;
 561
 562         close(sv[0]);
 563
 564         while (waitpid(pid, &status, 0) != pid) {
 565             /* loop */
 566         }
 567         sigprocmask(SIG_SETMASK, &oldmask, NULL);
 568         if (fd < 0) {
 569             error_setg_errno(errp, saved_errno,
 570                              "failed to recv file descriptor");
 571             return -1;
 572         }
 573         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 574             error_setg(errp, "bridge helper failed");
 575             return -1;
 576         }
 577         return fd;
 578     }
 579 }
 580
 581 int net_init_bridge(const Netdev *netdev, const char *name,
 582                     NetClientState *peer, Error **errp)
 583 {
 584     const NetdevBridgeOptions *bridge;
 585     const char *helper, *br;
 586     TAPState *s;
 587     int fd, vnet_hdr;
 588
 589     assert(netdev->type == NET_CLIENT_DRIVER_BRIDGE);
 590     bridge = &netdev->u.bridge;
 591
 592     helper = bridge->has_helper ? bridge->helper : DEFAULT_BRIDGE_HELPER;
 593     br     = bridge->has_br     ? bridge->br     : DEFAULT_BRIDGE_INTERFACE;
 594
 595     fd = net_bridge_run_helper(helper, br, errp);
 596     if (fd == -1) {
 597         return -1;
 598     }
 599
 600     qemu_set_nonblock(fd);
 601     vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 602     if (vnet_hdr < 0) {
 603         close(fd);
 604         return -1;
 605     }
 606     s = net_tap_fd_init(peer, "bridge", name, fd, vnet_hdr);
 607
 608     snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s,br=%s", helper,
 609              br);
 610
 611     return 0;
 612 }
 613
 614 static int net_tap_init(const NetdevTapOptions *tap, int *vnet_hdr,
 615                         const char *setup_script, char *ifname,
 616                         size_t ifname_sz, int mq_required, Error **errp)
 617 {
 618     Error *err = NULL;
 619     int fd, vnet_hdr_required;
 620
 621     if (tap->has_vnet_hdr) {
 622         *vnet_hdr = tap->vnet_hdr;
 623         vnet_hdr_required = *vnet_hdr;
 624     } else {
 625         *vnet_hdr = 1;
 626         vnet_hdr_required = 0;
 627     }
 628
 629     TFR(fd = tap_open(ifname, ifname_sz, vnet_hdr, vnet_hdr_required,
 630                       mq_required, errp));
 631     if (fd < 0) {
 632         return -1;
 633     }
 634
 635     if (setup_script &&
 636         setup_script[0] != '\0' &&
 637         strcmp(setup_script, "no") != 0) {
 638         launch_script(setup_script, ifname, fd, &err);
 639         if (err) {
 640             error_propagate(errp, err);
 641             close(fd);
 642             return -1;
 643         }
 644     }
 645
 646     return fd;
 647 }
 648
 649 #define MAX_TAP_QUEUES 1024
 650
 651 static void net_init_tap_one(const NetdevTapOptions *tap, NetClientState *peer,
 652                              const char *model, const char *name,
 653                              const char *ifname, const char *script,
 654                              const char *downscript, const char *vhostfdname,
 655                              int vnet_hdr, int fd, Error **errp)
 656 {
 657     Error *err = NULL;
 658     TAPState *s = net_tap_fd_init(peer, model, name, fd, vnet_hdr);
 659     int vhostfd;
 660
 661     tap_set_sndbuf(s->fd, tap, &err);
 662     if (err) {
 663         error_propagate(errp, err);
 664         return;
 665     }
 666
 667     if (tap->has_fd || tap->has_fds) {
 668         snprintf(s->nc.info_str, sizeof(s->nc.info_str), "fd=%d", fd);
 669     } else if (tap->has_helper) {
 670         snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s",
 671                  tap->helper);
 672     } else {
 673         snprintf(s->nc.info_str, sizeof(s->nc.info_str),
 674                  "ifname=%s,script=%s,downscript=%s", ifname, script,
 675                  downscript);
 676
 677         if (strcmp(downscript, "no") != 0) {
 678             snprintf(s->down_script, sizeof(s->down_script), "%s", downscript);
 679             snprintf(s->down_script_arg, sizeof(s->down_script_arg),
 680                      "%s", ifname);
 681         }
 682     }
 683
 684     if (tap->has_vhost ? tap->vhost :
 685         vhostfdname || (tap->has_vhostforce && tap->vhostforce)) {
 686         VhostNetOptions options;
 687
 688         options.backend_type = VHOST_BACKEND_TYPE_KERNEL;
 689         options.net_backend = &s->nc;
 690         if (tap->has_poll_us) {
 691             options.busyloop_timeout = tap->poll_us;
 692         } else {
 693             options.busyloop_timeout = 0;
 694         }
 695
 696         if (vhostfdname) {
 697             int ret;
 698
 699             vhostfd = monitor_fd_param(cur_mon, vhostfdname, &err);
 700             if (vhostfd == -1) {
 701                 if (tap->has_vhostforce && tap->vhostforce) {
 702                     error_propagate(errp, err);
 703                 } else {
 704                     warn_report_err(err);
 705                 }
 706                 return;
 707             }
 708             ret = qemu_try_set_nonblock(vhostfd);
 709             if (ret < 0) {
 710                 error_setg_errno(errp, -ret, "%s: Can't use file descriptor %d",
 711                                  name, fd);
 712                 return;
 713             }
 714         } else {
 715             vhostfd = open("/dev/vhost-net", O_RDWR);
 716             if (vhostfd < 0) {
 717                 if (tap->has_vhostforce && tap->vhostforce) {
 718                     error_setg_errno(errp, errno,
 719                                      "tap: open vhost char device failed");
 720                 } else {
 721                     warn_report("tap: open vhost char device failed: %s",
 722                                 strerror(errno));
 723                 }
 724                 return;
 725             }
 726             qemu_set_nonblock(vhostfd);
 727         }
 728         options.opaque = (void *)(uintptr_t)vhostfd;
 729
 730         s->vhost_net = vhost_net_init(&options);
 731         if (!s->vhost_net) {
 732             if (tap->has_vhostforce && tap->vhostforce) {
 733                 error_setg(errp, VHOST_NET_INIT_FAILED);
 734             } else {
 735                 warn_report(VHOST_NET_INIT_FAILED);
 736             }
 737             return;
 738         }
 739     } else if (vhostfdname) {
 740         error_setg(errp, "vhostfd(s)= is not valid without vhost");
 741     }
 742 }
 743
 744 static int get_fds(char *str, char *fds[], int max)
 745 {
 746     char *ptr = str, *this;
 747     size_t len = strlen(str);
 748     int i = 0;
 749
 750     while (i < max && ptr < str + len) {
 751         this = strchr(ptr, ':');
 752
 753         if (this == NULL) {
 754             fds[i] = g_strdup(ptr);
 755         } else {
 756             fds[i] = g_strndup(ptr, this - ptr);
 757         }
 758
 759         i++;
 760         if (this == NULL) {
 761             break;
 762         } else {
 763             ptr = this + 1;
 764         }
 765     }
 766
 767     return i;
 768 }
 769
 770 int net_init_tap(const Netdev *netdev, const char *name,
 771                  NetClientState *peer, Error **errp)
 772 {
 773     const NetdevTapOptions *tap;
 774     int fd, vnet_hdr = 0, i = 0, queues;
 775     /* for the no-fd, no-helper case */
 776     const char *script = NULL; /* suppress wrong "uninit'd use" gcc warning */
 777     const char *downscript = NULL;
 778     Error *err = NULL;
 779     const char *vhostfdname;
 780     char ifname[128];
 781     int ret = 0;
 782
 783     assert(netdev->type == NET_CLIENT_DRIVER_TAP);
 784     tap = &netdev->u.tap;
 785     queues = tap->has_queues ? tap->queues : 1;
 786     vhostfdname = tap->has_vhostfd ? tap->vhostfd : NULL;
 787
 788     /* QEMU hubs do not support multiqueue tap, in this case peer is set.
 789      * For -netdev, peer is always NULL. */
 790     if (peer && (tap->has_queues || tap->has_fds || tap->has_vhostfds)) {
 791         error_setg(errp, "Multiqueue tap cannot be used with hubs");
 792         return -1;
 793     }
 794
 795     if (tap->has_fd) {
 796         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
 797             tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
 798             tap->has_fds || tap->has_vhostfds) {
 799             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 800                        "helper=, queues=, fds=, and vhostfds= "
 801                        "are invalid with fd=");
 802             return -1;
 803         }
 804
 805         fd = monitor_fd_param(cur_mon, tap->fd, errp);
 806         if (fd == -1) {
 807             return -1;
 808         }
 809
 810         ret = qemu_try_set_nonblock(fd);
 811         if (ret < 0) {
 812             error_setg_errno(errp, -ret, "%s: Can't use file descriptor %d",
 813                              name, fd);
 814             return -1;
 815         }
 816
 817         vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 818         if (vnet_hdr < 0) {
 819             close(fd);
 820             return -1;
 821         }
 822
 823         net_init_tap_one(tap, peer, "tap", name, NULL,
 824                          script, downscript,
 825                          vhostfdname, vnet_hdr, fd, &err);
 826         if (err) {
 827             error_propagate(errp, err);
 828             return -1;
 829         }
 830     } else if (tap->has_fds) {
 831         char **fds;
 832         char **vhost_fds;
 833         int nfds = 0, nvhosts = 0;
 834
 835         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
 836             tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
 837             tap->has_vhostfd) {
 838             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 839                        "helper=, queues=, and vhostfd= "
 840                        "are invalid with fds=");
 841             return -1;
 842         }
 843
 844         fds = g_new0(char *, MAX_TAP_QUEUES);
 845         vhost_fds = g_new0(char *, MAX_TAP_QUEUES);
 846
 847         nfds = get_fds(tap->fds, fds, MAX_TAP_QUEUES);
 848         if (tap->has_vhostfds) {
 849             nvhosts = get_fds(tap->vhostfds, vhost_fds, MAX_TAP_QUEUES);
 850             if (nfds != nvhosts) {
 851                 error_setg(errp, "The number of fds passed does not match "
 852                            "the number of vhostfds passed");
 853                 ret = -1;
 854                 goto free_fail;
 855             }
 856         }
 857
 858         for (i = 0; i < nfds; i++) {
 859             fd = monitor_fd_param(cur_mon, fds[i], errp);
 860             if (fd == -1) {
 861                 ret = -1;
 862                 goto free_fail;
 863             }
 864
 865             ret = qemu_try_set_nonblock(fd);
 866             if (ret < 0) {
 867                 error_setg_errno(errp, -ret, "%s: Can't use file descriptor %d",
 868                                  name, fd);
 869                 goto free_fail;
 870             }
 871
 872             if (i == 0) {
 873                 vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 874                 if (vnet_hdr < 0) {
 875                     goto free_fail;
 876                 }
 877             } else if (vnet_hdr != tap_probe_vnet_hdr(fd, NULL)) {
 878                 error_setg(errp,
 879                            "vnet_hdr not consistent across given tap fds");
 880                 ret = -1;
 881                 goto free_fail;
 882             }
 883
 884             net_init_tap_one(tap, peer, "tap", name, ifname,
 885                              script, downscript,
 886                              tap->has_vhostfds ? vhost_fds[i] : NULL,
 887                              vnet_hdr, fd, &err);
 888             if (err) {
 889                 error_propagate(errp, err);
 890                 ret = -1;
 891                 goto free_fail;
 892             }
 893         }
 894
 895 free_fail:
 896         for (i = 0; i < nvhosts; i++) {
 897             g_free(vhost_fds[i]);
 898         }
 899         for (i = 0; i < nfds; i++) {
 900             g_free(fds[i]);
 901         }
 902         g_free(fds);
 903         g_free(vhost_fds);
 904         return ret;
 905     } else if (tap->has_helper) {
 906         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
 907             tap->has_vnet_hdr || tap->has_queues || tap->has_vhostfds) {
 908             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 909                        "queues=, and vhostfds= are invalid with helper=");
 910             return -1;
 911         }
 912
 913         fd = net_bridge_run_helper(tap->helper,
 914                                    tap->has_br ?
 915                                    tap->br : DEFAULT_BRIDGE_INTERFACE,
 916                                    errp);
 917         if (fd == -1) {
 918             return -1;
 919         }
 920
 921         qemu_set_nonblock(fd);
 922         vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 923         if (vnet_hdr < 0) {
 924             close(fd);
 925             return -1;
 926         }
 927
 928         net_init_tap_one(tap, peer, "bridge", name, ifname,
 929                          script, downscript, vhostfdname,
 930                          vnet_hdr, fd, &err);
 931         if (err) {
 932             error_propagate(errp, err);
 933             close(fd);
 934             return -1;
 935         }
 936     } else {
 937         if (tap->has_vhostfds) {
 938             error_setg(errp, "vhostfds= is invalid if fds= wasn't specified");
 939             return -1;
 940         }
 941         script = tap->has_script ? tap->script : DEFAULT_NETWORK_SCRIPT;
 942         downscript = tap->has_downscript ? tap->downscript :
 943             DEFAULT_NETWORK_DOWN_SCRIPT;
 944
 945         if (tap->has_ifname) {
 946             pstrcpy(ifname, sizeof ifname, tap->ifname);
 947         } else {
 948             ifname[0] = '\0';
 949         }
 950
 951         for (i = 0; i < queues; i++) {
 952             fd = net_tap_init(tap, &vnet_hdr, i >= 1 ? "no" : script,
 953                               ifname, sizeof ifname, queues > 1, errp);
 954             if (fd == -1) {
 955                 return -1;
 956             }
 957
 958             if (queues > 1 && i == 0 && !tap->has_ifname) {
 959                 if (tap_fd_get_ifname(fd, ifname)) {
 960                     error_setg(errp, "Fail to get ifname");
 961                     close(fd);
 962                     return -1;
 963                 }
 964             }
 965
 966             net_init_tap_one(tap, peer, "tap", name, ifname,
 967                              i >= 1 ? "no" : script,
 968                              i >= 1 ? "no" : downscript,
 969                              vhostfdname, vnet_hdr, fd, &err);
 970             if (err) {
 971                 error_propagate(errp, err);
 972                 close(fd);
 973                 return -1;
 974             }
 975         }
 976     }
 977
 978     return 0;
 979 }
 980
 981 VHostNetState *tap_get_vhost_net(NetClientState *nc)
 982 {
 983     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 984     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 985     return s->vhost_net;
 986 }
 987
 988 int tap_enable(NetClientState *nc)
 989 {
 990     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 991     int ret;
 992
 993     if (s->enabled) {
 994         return 0;
 995     } else {
 996         ret = tap_fd_enable(s->fd);
 997         if (ret == 0) {
 998             s->enabled = true;
 999             tap_update_fd_handler(s);
1000         }
1001         return ret;
1002     }
1003 }
1004
1005 int tap_disable(NetClientState *nc)
1006 {
1007     TAPState *s = DO_UPCAST(TAPState, nc, nc);
1008     int ret;
1009
1010     if (s->enabled == 0) {
1011         return 0;
1012     } else {
1013         ret = tap_fd_disable(s->fd);
1014         if (ret == 0) {
1015             qemu_purge_queued_packets(nc);
1016             s->enabled = false;
1017             tap_update_fd_handler(s);
1018         }
1019         return ret;
1020     }
1021 }