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 (!s->nc.peer->do_not_pad) {
 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 int tap_get_fd(NetClientState *nc)
 351 {
 352     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 353     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 354     return s->fd;
 355 }
 356
 357 /* fd support */
 358
 359 static NetClientInfo net_tap_info = {
 360     .type = NET_CLIENT_DRIVER_TAP,
 361     .size = sizeof(TAPState),
 362     .receive = tap_receive,
 363     .receive_raw = tap_receive_raw,
 364     .receive_iov = tap_receive_iov,
 365     .poll = tap_poll,
 366     .cleanup = tap_cleanup,
 367     .has_ufo = tap_has_ufo,
 368     .has_vnet_hdr = tap_has_vnet_hdr,
 369     .has_vnet_hdr_len = tap_has_vnet_hdr_len,
 370     .using_vnet_hdr = tap_using_vnet_hdr,
 371     .set_offload = tap_set_offload,
 372     .set_vnet_hdr_len = tap_set_vnet_hdr_len,
 373     .set_vnet_le = tap_set_vnet_le,
 374     .set_vnet_be = tap_set_vnet_be,
 375 };
 376
 377 static TAPState *net_tap_fd_init(NetClientState *peer,
 378                                  const char *model,
 379                                  const char *name,
 380                                  int fd,
 381                                  int vnet_hdr)
 382 {
 383     NetClientState *nc;
 384     TAPState *s;
 385
 386     nc = qemu_new_net_client(&net_tap_info, peer, model, name);
 387
 388     s = DO_UPCAST(TAPState, nc, nc);
 389
 390     s->fd = fd;
 391     s->host_vnet_hdr_len = vnet_hdr ? sizeof(struct virtio_net_hdr) : 0;
 392     s->using_vnet_hdr = false;
 393     s->has_ufo = tap_probe_has_ufo(s->fd);
 394     s->enabled = true;
 395     tap_set_offload(&s->nc, 0, 0, 0, 0, 0);
 396     /*
 397      * Make sure host header length is set correctly in tap:
 398      * it might have been modified by another instance of qemu.
 399      */
 400     if (tap_probe_vnet_hdr_len(s->fd, s->host_vnet_hdr_len)) {
 401         tap_fd_set_vnet_hdr_len(s->fd, s->host_vnet_hdr_len);
 402     }
 403     tap_read_poll(s, true);
 404     s->vhost_net = NULL;
 405
 406     s->exit.notify = tap_exit_notify;
 407     qemu_add_exit_notifier(&s->exit);
 408
 409     return s;
 410 }
 411
 412 static void launch_script(const char *setup_script, const char *ifname,
 413                           int fd, Error **errp)
 414 {
 415     int pid, status;
 416     char *args[3];
 417     char **parg;
 418
 419     /* try to launch network script */
 420     pid = fork();
 421     if (pid < 0) {
 422         error_setg_errno(errp, errno, "could not launch network script %s",
 423                          setup_script);
 424         return;
 425     }
 426     if (pid == 0) {
 427         int open_max = sysconf(_SC_OPEN_MAX), i;
 428
 429         for (i = 3; i < open_max; i++) {
 430             if (i != fd) {
 431                 close(i);
 432             }
 433         }
 434         parg = args;
 435         *parg++ = (char *)setup_script;
 436         *parg++ = (char *)ifname;
 437         *parg = NULL;
 438         execv(setup_script, args);
 439         _exit(1);
 440     } else {
 441         while (waitpid(pid, &status, 0) != pid) {
 442             /* loop */
 443         }
 444
 445         if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
 446             return;
 447         }
 448         error_setg(errp, "network script %s failed with status %d",
 449                    setup_script, status);
 450     }
 451 }
 452
 453 static int recv_fd(int c)
 454 {
 455     int fd;
 456     uint8_t msgbuf[CMSG_SPACE(sizeof(fd))];
 457     struct msghdr msg = {
 458         .msg_control = msgbuf,
 459         .msg_controllen = sizeof(msgbuf),
 460     };
 461     struct cmsghdr *cmsg;
 462     struct iovec iov;
 463     uint8_t req[1];
 464     ssize_t len;
 465
 466     cmsg = CMSG_FIRSTHDR(&msg);
 467     cmsg->cmsg_level = SOL_SOCKET;
 468     cmsg->cmsg_type = SCM_RIGHTS;
 469     cmsg->cmsg_len = CMSG_LEN(sizeof(fd));
 470     msg.msg_controllen = cmsg->cmsg_len;
 471
 472     iov.iov_base = req;
 473     iov.iov_len = sizeof(req);
 474
 475     msg.msg_iov = &iov;
 476     msg.msg_iovlen = 1;
 477
 478     len = recvmsg(c, &msg, 0);
 479     if (len > 0) {
 480         memcpy(&fd, CMSG_DATA(cmsg), sizeof(fd));
 481         return fd;
 482     }
 483
 484     return len;
 485 }
 486
 487 static int net_bridge_run_helper(const char *helper, const char *bridge,
 488                                  Error **errp)
 489 {
 490     sigset_t oldmask, mask;
 491     g_autofree char *default_helper = NULL;
 492     int pid, status;
 493     char *args[5];
 494     char **parg;
 495     int sv[2];
 496
 497     sigemptyset(&mask);
 498     sigaddset(&mask, SIGCHLD);
 499     sigprocmask(SIG_BLOCK, &mask, &oldmask);
 500
 501     if (!helper) {
 502         helper = default_helper = get_relocated_path(DEFAULT_BRIDGE_HELPER);
 503     }
 504
 505     if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) {
 506         error_setg_errno(errp, errno, "socketpair() failed");
 507         return -1;
 508     }
 509
 510     /* try to launch bridge helper */
 511     pid = fork();
 512     if (pid < 0) {
 513         error_setg_errno(errp, errno, "Can't fork bridge helper");
 514         return -1;
 515     }
 516     if (pid == 0) {
 517         int open_max = sysconf(_SC_OPEN_MAX), i;
 518         char *fd_buf = NULL;
 519         char *br_buf = NULL;
 520         char *helper_cmd = NULL;
 521
 522         for (i = 3; i < open_max; i++) {
 523             if (i != sv[1]) {
 524                 close(i);
 525             }
 526         }
 527
 528         fd_buf = g_strdup_printf("%s%d", "--fd=", sv[1]);
 529
 530         if (strrchr(helper, ' ') || strrchr(helper, '\t')) {
 531             /* assume helper is a command */
 532
 533             if (strstr(helper, "--br=") == NULL) {
 534                 br_buf = g_strdup_printf("%s%s", "--br=", bridge);
 535             }
 536
 537             helper_cmd = g_strdup_printf("%s %s %s %s", helper,
 538                             "--use-vnet", fd_buf, br_buf ? br_buf : "");
 539
 540             parg = args;
 541             *parg++ = (char *)"sh";
 542             *parg++ = (char *)"-c";
 543             *parg++ = helper_cmd;
 544             *parg++ = NULL;
 545
 546             execv("/bin/sh", args);
 547             g_free(helper_cmd);
 548         } else {
 549             /* assume helper is just the executable path name */
 550
 551             br_buf = g_strdup_printf("%s%s", "--br=", bridge);
 552
 553             parg = args;
 554             *parg++ = (char *)helper;
 555             *parg++ = (char *)"--use-vnet";
 556             *parg++ = fd_buf;
 557             *parg++ = br_buf;
 558             *parg++ = NULL;
 559
 560             execv(helper, args);
 561         }
 562         g_free(fd_buf);
 563         g_free(br_buf);
 564         _exit(1);
 565
 566     } else {
 567         int fd;
 568         int saved_errno;
 569
 570         close(sv[1]);
 571
 572         do {
 573             fd = recv_fd(sv[0]);
 574         } while (fd == -1 && errno == EINTR);
 575         saved_errno = errno;
 576
 577         close(sv[0]);
 578
 579         while (waitpid(pid, &status, 0) != pid) {
 580             /* loop */
 581         }
 582         sigprocmask(SIG_SETMASK, &oldmask, NULL);
 583         if (fd < 0) {
 584             error_setg_errno(errp, saved_errno,
 585                              "failed to recv file descriptor");
 586             return -1;
 587         }
 588         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 589             error_setg(errp, "bridge helper failed");
 590             return -1;
 591         }
 592         return fd;
 593     }
 594 }
 595
 596 int net_init_bridge(const Netdev *netdev, const char *name,
 597                     NetClientState *peer, Error **errp)
 598 {
 599     const NetdevBridgeOptions *bridge;
 600     const char *helper, *br;
 601     TAPState *s;
 602     int fd, vnet_hdr;
 603
 604     assert(netdev->type == NET_CLIENT_DRIVER_BRIDGE);
 605     bridge = &netdev->u.bridge;
 606     helper = bridge->has_helper ? bridge->helper : NULL;
 607     br     = bridge->has_br     ? bridge->br     : DEFAULT_BRIDGE_INTERFACE;
 608
 609     fd = net_bridge_run_helper(helper, br, errp);
 610     if (fd == -1) {
 611         return -1;
 612     }
 613
 614     qemu_set_nonblock(fd);
 615     vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 616     if (vnet_hdr < 0) {
 617         close(fd);
 618         return -1;
 619     }
 620     s = net_tap_fd_init(peer, "bridge", name, fd, vnet_hdr);
 621
 622     snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s,br=%s", helper,
 623              br);
 624
 625     return 0;
 626 }
 627
 628 static int net_tap_init(const NetdevTapOptions *tap, int *vnet_hdr,
 629                         const char *setup_script, char *ifname,
 630                         size_t ifname_sz, int mq_required, Error **errp)
 631 {
 632     Error *err = NULL;
 633     int fd, vnet_hdr_required;
 634
 635     if (tap->has_vnet_hdr) {
 636         *vnet_hdr = tap->vnet_hdr;
 637         vnet_hdr_required = *vnet_hdr;
 638     } else {
 639         *vnet_hdr = 1;
 640         vnet_hdr_required = 0;
 641     }
 642
 643     TFR(fd = tap_open(ifname, ifname_sz, vnet_hdr, vnet_hdr_required,
 644                       mq_required, errp));
 645     if (fd < 0) {
 646         return -1;
 647     }
 648
 649     if (setup_script &&
 650         setup_script[0] != '\0' &&
 651         strcmp(setup_script, "no") != 0) {
 652         launch_script(setup_script, ifname, fd, &err);
 653         if (err) {
 654             error_propagate(errp, err);
 655             close(fd);
 656             return -1;
 657         }
 658     }
 659
 660     return fd;
 661 }
 662
 663 #define MAX_TAP_QUEUES 1024
 664
 665 static void net_init_tap_one(const NetdevTapOptions *tap, NetClientState *peer,
 666                              const char *model, const char *name,
 667                              const char *ifname, const char *script,
 668                              const char *downscript, const char *vhostfdname,
 669                              int vnet_hdr, int fd, Error **errp)
 670 {
 671     Error *err = NULL;
 672     TAPState *s = net_tap_fd_init(peer, model, name, fd, vnet_hdr);
 673     int vhostfd;
 674
 675     tap_set_sndbuf(s->fd, tap, &err);
 676     if (err) {
 677         error_propagate(errp, err);
 678         return;
 679     }
 680
 681     if (tap->has_fd || tap->has_fds) {
 682         snprintf(s->nc.info_str, sizeof(s->nc.info_str), "fd=%d", fd);
 683     } else if (tap->has_helper) {
 684         snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s",
 685                  tap->helper);
 686     } else {
 687         snprintf(s->nc.info_str, sizeof(s->nc.info_str),
 688                  "ifname=%s,script=%s,downscript=%s", ifname, script,
 689                  downscript);
 690
 691         if (strcmp(downscript, "no") != 0) {
 692             snprintf(s->down_script, sizeof(s->down_script), "%s", downscript);
 693             snprintf(s->down_script_arg, sizeof(s->down_script_arg),
 694                      "%s", ifname);
 695         }
 696     }
 697
 698     if (tap->has_vhost ? tap->vhost :
 699         vhostfdname || (tap->has_vhostforce && tap->vhostforce)) {
 700         VhostNetOptions options;
 701
 702         options.backend_type = VHOST_BACKEND_TYPE_KERNEL;
 703         options.net_backend = &s->nc;
 704         if (tap->has_poll_us) {
 705             options.busyloop_timeout = tap->poll_us;
 706         } else {
 707             options.busyloop_timeout = 0;
 708         }
 709
 710         if (vhostfdname) {
 711             int ret;
 712
 713             vhostfd = monitor_fd_param(monitor_cur(), vhostfdname, &err);
 714             if (vhostfd == -1) {
 715                 if (tap->has_vhostforce && tap->vhostforce) {
 716                     error_propagate(errp, err);
 717                 } else {
 718                     warn_report_err(err);
 719                 }
 720                 return;
 721             }
 722             ret = qemu_try_set_nonblock(vhostfd);
 723             if (ret < 0) {
 724                 error_setg_errno(errp, -ret, "%s: Can't use file descriptor %d",
 725                                  name, fd);
 726                 return;
 727             }
 728         } else {
 729             vhostfd = open("/dev/vhost-net", O_RDWR);
 730             if (vhostfd < 0) {
 731                 if (tap->has_vhostforce && tap->vhostforce) {
 732                     error_setg_errno(errp, errno,
 733                                      "tap: open vhost char device failed");
 734                 } else {
 735                     warn_report("tap: open vhost char device failed: %s",
 736                                 strerror(errno));
 737                 }
 738                 return;
 739             }
 740             qemu_set_nonblock(vhostfd);
 741         }
 742         options.opaque = (void *)(uintptr_t)vhostfd;
 743
 744         s->vhost_net = vhost_net_init(&options);
 745         if (!s->vhost_net) {
 746             if (tap->has_vhostforce && tap->vhostforce) {
 747                 error_setg(errp, VHOST_NET_INIT_FAILED);
 748             } else {
 749                 warn_report(VHOST_NET_INIT_FAILED);
 750             }
 751             return;
 752         }
 753     } else if (vhostfdname) {
 754         error_setg(errp, "vhostfd(s)= is not valid without vhost");
 755     }
 756 }
 757
 758 static int get_fds(char *str, char *fds[], int max)
 759 {
 760     char *ptr = str, *this;
 761     size_t len = strlen(str);
 762     int i = 0;
 763
 764     while (i < max && ptr < str + len) {
 765         this = strchr(ptr, ':');
 766
 767         if (this == NULL) {
 768             fds[i] = g_strdup(ptr);
 769         } else {
 770             fds[i] = g_strndup(ptr, this - ptr);
 771         }
 772
 773         i++;
 774         if (this == NULL) {
 775             break;
 776         } else {
 777             ptr = this + 1;
 778         }
 779     }
 780
 781     return i;
 782 }
 783
 784 int net_init_tap(const Netdev *netdev, const char *name,
 785                  NetClientState *peer, Error **errp)
 786 {
 787     const NetdevTapOptions *tap;
 788     int fd, vnet_hdr = 0, i = 0, queues;
 789     /* for the no-fd, no-helper case */
 790     const char *script;
 791     const char *downscript;
 792     Error *err = NULL;
 793     const char *vhostfdname;
 794     char ifname[128];
 795     int ret = 0;
 796
 797     assert(netdev->type == NET_CLIENT_DRIVER_TAP);
 798     tap = &netdev->u.tap;
 799     queues = tap->has_queues ? tap->queues : 1;
 800     vhostfdname = tap->has_vhostfd ? tap->vhostfd : NULL;
 801     script = tap->has_script ? tap->script : NULL;
 802     downscript = tap->has_downscript ? tap->downscript : NULL;
 803
 804     /* QEMU hubs do not support multiqueue tap, in this case peer is set.
 805      * For -netdev, peer is always NULL. */
 806     if (peer && (tap->has_queues || tap->has_fds || tap->has_vhostfds)) {
 807         error_setg(errp, "Multiqueue tap cannot be used with hubs");
 808         return -1;
 809     }
 810
 811     if (tap->has_fd) {
 812         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
 813             tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
 814             tap->has_fds || tap->has_vhostfds) {
 815             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 816                        "helper=, queues=, fds=, and vhostfds= "
 817                        "are invalid with fd=");
 818             return -1;
 819         }
 820
 821         fd = monitor_fd_param(monitor_cur(), tap->fd, errp);
 822         if (fd == -1) {
 823             return -1;
 824         }
 825
 826         ret = qemu_try_set_nonblock(fd);
 827         if (ret < 0) {
 828             error_setg_errno(errp, -ret, "%s: Can't use file descriptor %d",
 829                              name, fd);
 830             close(fd);
 831             return -1;
 832         }
 833
 834         vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 835         if (vnet_hdr < 0) {
 836             close(fd);
 837             return -1;
 838         }
 839
 840         net_init_tap_one(tap, peer, "tap", name, NULL,
 841                          script, downscript,
 842                          vhostfdname, vnet_hdr, fd, &err);
 843         if (err) {
 844             error_propagate(errp, err);
 845             close(fd);
 846             return -1;
 847         }
 848     } else if (tap->has_fds) {
 849         char **fds;
 850         char **vhost_fds;
 851         int nfds = 0, nvhosts = 0;
 852
 853         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
 854             tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
 855             tap->has_vhostfd) {
 856             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 857                        "helper=, queues=, and vhostfd= "
 858                        "are invalid with fds=");
 859             return -1;
 860         }
 861
 862         fds = g_new0(char *, MAX_TAP_QUEUES);
 863         vhost_fds = g_new0(char *, MAX_TAP_QUEUES);
 864
 865         nfds = get_fds(tap->fds, fds, MAX_TAP_QUEUES);
 866         if (tap->has_vhostfds) {
 867             nvhosts = get_fds(tap->vhostfds, vhost_fds, MAX_TAP_QUEUES);
 868             if (nfds != nvhosts) {
 869                 error_setg(errp, "The number of fds passed does not match "
 870                            "the number of vhostfds passed");
 871                 ret = -1;
 872                 goto free_fail;
 873             }
 874         }
 875
 876         for (i = 0; i < nfds; i++) {
 877             fd = monitor_fd_param(monitor_cur(), fds[i], errp);
 878             if (fd == -1) {
 879                 ret = -1;
 880                 goto free_fail;
 881             }
 882
 883             ret = qemu_try_set_nonblock(fd);
 884             if (ret < 0) {
 885                 error_setg_errno(errp, -ret, "%s: Can't use file descriptor %d",
 886                                  name, fd);
 887                 goto free_fail;
 888             }
 889
 890             if (i == 0) {
 891                 vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 892                 if (vnet_hdr < 0) {
 893                     goto free_fail;
 894                 }
 895             } else if (vnet_hdr != tap_probe_vnet_hdr(fd, NULL)) {
 896                 error_setg(errp,
 897                            "vnet_hdr not consistent across given tap fds");
 898                 ret = -1;
 899                 goto free_fail;
 900             }
 901
 902             net_init_tap_one(tap, peer, "tap", name, ifname,
 903                              script, downscript,
 904                              tap->has_vhostfds ? vhost_fds[i] : NULL,
 905                              vnet_hdr, fd, &err);
 906             if (err) {
 907                 error_propagate(errp, err);
 908                 ret = -1;
 909                 goto free_fail;
 910             }
 911         }
 912
 913 free_fail:
 914         for (i = 0; i < nvhosts; i++) {
 915             g_free(vhost_fds[i]);
 916         }
 917         for (i = 0; i < nfds; i++) {
 918             g_free(fds[i]);
 919         }
 920         g_free(fds);
 921         g_free(vhost_fds);
 922         return ret;
 923     } else if (tap->has_helper) {
 924         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
 925             tap->has_vnet_hdr || tap->has_queues || tap->has_vhostfds) {
 926             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 927                        "queues=, and vhostfds= are invalid with helper=");
 928             return -1;
 929         }
 930
 931         fd = net_bridge_run_helper(tap->helper,
 932                                    tap->has_br ?
 933                                    tap->br : DEFAULT_BRIDGE_INTERFACE,
 934                                    errp);
 935         if (fd == -1) {
 936             return -1;
 937         }
 938
 939         qemu_set_nonblock(fd);
 940         vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 941         if (vnet_hdr < 0) {
 942             close(fd);
 943             return -1;
 944         }
 945
 946         net_init_tap_one(tap, peer, "bridge", name, ifname,
 947                          script, downscript, vhostfdname,
 948                          vnet_hdr, fd, &err);
 949         if (err) {
 950             error_propagate(errp, err);
 951             close(fd);
 952             return -1;
 953         }
 954     } else {
 955         g_autofree char *default_script = NULL;
 956         g_autofree char *default_downscript = NULL;
 957         if (tap->has_vhostfds) {
 958             error_setg(errp, "vhostfds= is invalid if fds= wasn't specified");
 959             return -1;
 960         }
 961
 962         if (!script) {
 963             script = default_script = get_relocated_path(DEFAULT_NETWORK_SCRIPT);
 964         }
 965         if (!downscript) {
 966             downscript = default_downscript =
 967                                  get_relocated_path(DEFAULT_NETWORK_DOWN_SCRIPT);
 968         }
 969
 970         if (tap->has_ifname) {
 971             pstrcpy(ifname, sizeof ifname, tap->ifname);
 972         } else {
 973             ifname[0] = '\0';
 974         }
 975
 976         for (i = 0; i < queues; i++) {
 977             fd = net_tap_init(tap, &vnet_hdr, i >= 1 ? "no" : script,
 978                               ifname, sizeof ifname, queues > 1, errp);
 979             if (fd == -1) {
 980                 return -1;
 981             }
 982
 983             if (queues > 1 && i == 0 && !tap->has_ifname) {
 984                 if (tap_fd_get_ifname(fd, ifname)) {
 985                     error_setg(errp, "Fail to get ifname");
 986                     close(fd);
 987                     return -1;
 988                 }
 989             }
 990
 991             net_init_tap_one(tap, peer, "tap", name, ifname,
 992                              i >= 1 ? "no" : script,
 993                              i >= 1 ? "no" : downscript,
 994                              vhostfdname, vnet_hdr, fd, &err);
 995             if (err) {
 996                 error_propagate(errp, err);
 997                 close(fd);
 998                 return -1;
 999             }
1000         }
1001     }
1002
1003     return 0;
1004 }
1005
1006 VHostNetState *tap_get_vhost_net(NetClientState *nc)
1007 {
1008     TAPState *s = DO_UPCAST(TAPState, nc, nc);
1009     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
1010     return s->vhost_net;
1011 }
1012
1013 int tap_enable(NetClientState *nc)
1014 {
1015     TAPState *s = DO_UPCAST(TAPState, nc, nc);
1016     int ret;
1017
1018     if (s->enabled) {
1019         return 0;
1020     } else {
1021         ret = tap_fd_enable(s->fd);
1022         if (ret == 0) {
1023             s->enabled = true;
1024             tap_update_fd_handler(s);
1025         }
1026         return ret;
1027     }
1028 }
1029
1030 int tap_disable(NetClientState *nc)
1031 {
1032     TAPState *s = DO_UPCAST(TAPState, nc, nc);
1033     int ret;
1034
1035     if (s->enabled == 0) {
1036         return 0;
1037     } else {
1038         ret = tap_fd_disable(s->fd);
1039         if (ret == 0) {
1040             qemu_purge_queued_packets(nc);
1041             s->enabled = false;
1042             tap_update_fd_handler(s);
1043         }
1044         return ret;
1045     }
1046 }