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     len = RETRY_ON_EINTR(writev(s->fd, iov, iovcnt));
 106
 107     if (len == -1 && errno == EAGAIN) {
 108         tap_write_poll(s, true);
 109         return 0;
 110     }
 111
 112     return len;
 113 }
 114
 115 static ssize_t tap_receive_iov(NetClientState *nc, const struct iovec *iov,
 116                                int iovcnt)
 117 {
 118     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 119     const struct iovec *iovp = iov;
 120     struct iovec iov_copy[iovcnt + 1];
 121     struct virtio_net_hdr_mrg_rxbuf hdr = { };
 122
 123     if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
 124         iov_copy[0].iov_base = &hdr;
 125         iov_copy[0].iov_len =  s->host_vnet_hdr_len;
 126         memcpy(&iov_copy[1], iov, iovcnt * sizeof(*iov));
 127         iovp = iov_copy;
 128         iovcnt++;
 129     }
 130
 131     return tap_write_packet(s, iovp, iovcnt);
 132 }
 133
 134 static ssize_t tap_receive_raw(NetClientState *nc, const uint8_t *buf, size_t size)
 135 {
 136     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 137     struct iovec iov[2];
 138     int iovcnt = 0;
 139     struct virtio_net_hdr_mrg_rxbuf hdr = { };
 140
 141     if (s->host_vnet_hdr_len) {
 142         iov[iovcnt].iov_base = &hdr;
 143         iov[iovcnt].iov_len  = s->host_vnet_hdr_len;
 144         iovcnt++;
 145     }
 146
 147     iov[iovcnt].iov_base = (char *)buf;
 148     iov[iovcnt].iov_len  = size;
 149     iovcnt++;
 150
 151     return tap_write_packet(s, iov, iovcnt);
 152 }
 153
 154 static ssize_t tap_receive(NetClientState *nc, const uint8_t *buf, size_t size)
 155 {
 156     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 157     struct iovec iov[1];
 158
 159     if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
 160         return tap_receive_raw(nc, buf, size);
 161     }
 162
 163     iov[0].iov_base = (char *)buf;
 164     iov[0].iov_len  = size;
 165
 166     return tap_write_packet(s, iov, 1);
 167 }
 168
 169 #ifndef __sun__
 170 ssize_t tap_read_packet(int tapfd, uint8_t *buf, int maxlen)
 171 {
 172     return read(tapfd, buf, maxlen);
 173 }
 174 #endif
 175
 176 static void tap_send_completed(NetClientState *nc, ssize_t len)
 177 {
 178     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 179     tap_read_poll(s, true);
 180 }
 181
 182 static void tap_send(void *opaque)
 183 {
 184     TAPState *s = opaque;
 185     int size;
 186     int packets = 0;
 187
 188     while (true) {
 189         uint8_t *buf = s->buf;
 190         uint8_t min_pkt[ETH_ZLEN];
 191         size_t min_pktsz = sizeof(min_pkt);
 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         if (net_peer_needs_padding(&s->nc)) {
 204             if (eth_pad_short_frame(min_pkt, &min_pktsz, buf, size)) {
 205                 buf = min_pkt;
 206                 size = min_pktsz;
 207             }
 208         }
 209
 210         size = qemu_send_packet_async(&s->nc, buf, size, tap_send_completed);
 211         if (size == 0) {
 212             tap_read_poll(s, false);
 213             break;
 214         } else if (size < 0) {
 215             break;
 216         }
 217
 218         /*
 219          * When the host keeps receiving more packets while tap_send() is
 220          * running we can hog the QEMU global mutex.  Limit the number of
 221          * packets that are processed per tap_send() callback to prevent
 222          * stalling the guest.
 223          */
 224         packets++;
 225         if (packets >= 50) {
 226             break;
 227         }
 228     }
 229 }
 230
 231 static bool tap_has_ufo(NetClientState *nc)
 232 {
 233     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 234
 235     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 236
 237     return s->has_ufo;
 238 }
 239
 240 static bool tap_has_vnet_hdr(NetClientState *nc)
 241 {
 242     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 243
 244     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 245
 246     return !!s->host_vnet_hdr_len;
 247 }
 248
 249 static bool tap_has_vnet_hdr_len(NetClientState *nc, int len)
 250 {
 251     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 252
 253     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 254
 255     return !!tap_probe_vnet_hdr_len(s->fd, len);
 256 }
 257
 258 static int tap_get_vnet_hdr_len(NetClientState *nc)
 259 {
 260     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 261
 262     return s->host_vnet_hdr_len;
 263 }
 264
 265 static void tap_set_vnet_hdr_len(NetClientState *nc, int len)
 266 {
 267     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 268
 269     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 270     assert(len == sizeof(struct virtio_net_hdr_mrg_rxbuf) ||
 271            len == sizeof(struct virtio_net_hdr) ||
 272            len == sizeof(struct virtio_net_hdr_v1_hash));
 273
 274     tap_fd_set_vnet_hdr_len(s->fd, len);
 275     s->host_vnet_hdr_len = len;
 276 }
 277
 278 static bool tap_get_using_vnet_hdr(NetClientState *nc)
 279 {
 280     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 281
 282     return s->using_vnet_hdr;
 283 }
 284
 285 static void tap_using_vnet_hdr(NetClientState *nc, bool using_vnet_hdr)
 286 {
 287     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 288
 289     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 290     assert(!!s->host_vnet_hdr_len == using_vnet_hdr);
 291
 292     s->using_vnet_hdr = using_vnet_hdr;
 293 }
 294
 295 static int tap_set_vnet_le(NetClientState *nc, bool is_le)
 296 {
 297     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 298
 299     return tap_fd_set_vnet_le(s->fd, is_le);
 300 }
 301
 302 static int tap_set_vnet_be(NetClientState *nc, bool is_be)
 303 {
 304     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 305
 306     return tap_fd_set_vnet_be(s->fd, is_be);
 307 }
 308
 309 static void tap_set_offload(NetClientState *nc, int csum, int tso4,
 310                      int tso6, int ecn, int ufo)
 311 {
 312     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 313     if (s->fd < 0) {
 314         return;
 315     }
 316
 317     tap_fd_set_offload(s->fd, csum, tso4, tso6, ecn, ufo);
 318 }
 319
 320 static void tap_exit_notify(Notifier *notifier, void *data)
 321 {
 322     TAPState *s = container_of(notifier, TAPState, exit);
 323     Error *err = NULL;
 324
 325     if (s->down_script[0]) {
 326         launch_script(s->down_script, s->down_script_arg, s->fd, &err);
 327         if (err) {
 328             error_report_err(err);
 329         }
 330     }
 331 }
 332
 333 static void tap_cleanup(NetClientState *nc)
 334 {
 335     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 336
 337     if (s->vhost_net) {
 338         vhost_net_cleanup(s->vhost_net);
 339         g_free(s->vhost_net);
 340         s->vhost_net = NULL;
 341     }
 342
 343     qemu_purge_queued_packets(nc);
 344
 345     tap_exit_notify(&s->exit, NULL);
 346     qemu_remove_exit_notifier(&s->exit);
 347
 348     tap_read_poll(s, false);
 349     tap_write_poll(s, false);
 350     close(s->fd);
 351     s->fd = -1;
 352 }
 353
 354 static void tap_poll(NetClientState *nc, bool enable)
 355 {
 356     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 357     tap_read_poll(s, enable);
 358     tap_write_poll(s, enable);
 359 }
 360
 361 static bool tap_set_steering_ebpf(NetClientState *nc, int prog_fd)
 362 {
 363     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 364     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 365
 366     return tap_fd_set_steering_ebpf(s->fd, prog_fd) == 0;
 367 }
 368
 369 int tap_get_fd(NetClientState *nc)
 370 {
 371     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 372     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 373     return s->fd;
 374 }
 375
 376 /* fd support */
 377
 378 static NetClientInfo net_tap_info = {
 379     .type = NET_CLIENT_DRIVER_TAP,
 380     .size = sizeof(TAPState),
 381     .receive = tap_receive,
 382     .receive_raw = tap_receive_raw,
 383     .receive_iov = tap_receive_iov,
 384     .poll = tap_poll,
 385     .cleanup = tap_cleanup,
 386     .has_ufo = tap_has_ufo,
 387     .has_vnet_hdr = tap_has_vnet_hdr,
 388     .has_vnet_hdr_len = tap_has_vnet_hdr_len,
 389     .get_using_vnet_hdr = tap_get_using_vnet_hdr,
 390     .using_vnet_hdr = tap_using_vnet_hdr,
 391     .set_offload = tap_set_offload,
 392     .get_vnet_hdr_len = tap_get_vnet_hdr_len,
 393     .set_vnet_hdr_len = tap_set_vnet_hdr_len,
 394     .set_vnet_le = tap_set_vnet_le,
 395     .set_vnet_be = tap_set_vnet_be,
 396     .set_steering_ebpf = tap_set_steering_ebpf,
 397 };
 398
 399 static TAPState *net_tap_fd_init(NetClientState *peer,
 400                                  const char *model,
 401                                  const char *name,
 402                                  int fd,
 403                                  int vnet_hdr)
 404 {
 405     NetClientState *nc;
 406     TAPState *s;
 407
 408     nc = qemu_new_net_client(&net_tap_info, peer, model, name);
 409
 410     s = DO_UPCAST(TAPState, nc, nc);
 411
 412     s->fd = fd;
 413     s->host_vnet_hdr_len = vnet_hdr ? sizeof(struct virtio_net_hdr) : 0;
 414     s->using_vnet_hdr = false;
 415     s->has_ufo = tap_probe_has_ufo(s->fd);
 416     s->enabled = true;
 417     tap_set_offload(&s->nc, 0, 0, 0, 0, 0);
 418     /*
 419      * Make sure host header length is set correctly in tap:
 420      * it might have been modified by another instance of qemu.
 421      */
 422     if (tap_probe_vnet_hdr_len(s->fd, s->host_vnet_hdr_len)) {
 423         tap_fd_set_vnet_hdr_len(s->fd, s->host_vnet_hdr_len);
 424     }
 425     tap_read_poll(s, true);
 426     s->vhost_net = NULL;
 427
 428     s->exit.notify = tap_exit_notify;
 429     qemu_add_exit_notifier(&s->exit);
 430
 431     return s;
 432 }
 433
 434 static void launch_script(const char *setup_script, const char *ifname,
 435                           int fd, Error **errp)
 436 {
 437     int pid, status;
 438     char *args[3];
 439     char **parg;
 440
 441     /* try to launch network script */
 442     pid = fork();
 443     if (pid < 0) {
 444         error_setg_errno(errp, errno, "could not launch network script %s",
 445                          setup_script);
 446         return;
 447     }
 448     if (pid == 0) {
 449         int open_max = sysconf(_SC_OPEN_MAX), i;
 450
 451         for (i = 3; i < open_max; i++) {
 452             if (i != fd) {
 453                 close(i);
 454             }
 455         }
 456         parg = args;
 457         *parg++ = (char *)setup_script;
 458         *parg++ = (char *)ifname;
 459         *parg = NULL;
 460         execv(setup_script, args);
 461         _exit(1);
 462     } else {
 463         while (waitpid(pid, &status, 0) != pid) {
 464             /* loop */
 465         }
 466
 467         if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
 468             return;
 469         }
 470         error_setg(errp, "network script %s failed with status %d",
 471                    setup_script, status);
 472     }
 473 }
 474
 475 static int recv_fd(int c)
 476 {
 477     int fd;
 478     uint8_t msgbuf[CMSG_SPACE(sizeof(fd))];
 479     struct msghdr msg = {
 480         .msg_control = msgbuf,
 481         .msg_controllen = sizeof(msgbuf),
 482     };
 483     struct cmsghdr *cmsg;
 484     struct iovec iov;
 485     uint8_t req[1];
 486     ssize_t len;
 487
 488     cmsg = CMSG_FIRSTHDR(&msg);
 489     cmsg->cmsg_level = SOL_SOCKET;
 490     cmsg->cmsg_type = SCM_RIGHTS;
 491     cmsg->cmsg_len = CMSG_LEN(sizeof(fd));
 492     msg.msg_controllen = cmsg->cmsg_len;
 493
 494     iov.iov_base = req;
 495     iov.iov_len = sizeof(req);
 496
 497     msg.msg_iov = &iov;
 498     msg.msg_iovlen = 1;
 499
 500     len = recvmsg(c, &msg, 0);
 501     if (len > 0) {
 502         memcpy(&fd, CMSG_DATA(cmsg), sizeof(fd));
 503         return fd;
 504     }
 505
 506     return len;
 507 }
 508
 509 static int net_bridge_run_helper(const char *helper, const char *bridge,
 510                                  Error **errp)
 511 {
 512     sigset_t oldmask, mask;
 513     g_autofree char *default_helper = NULL;
 514     int pid, status;
 515     char *args[5];
 516     char **parg;
 517     int sv[2];
 518
 519     sigemptyset(&mask);
 520     sigaddset(&mask, SIGCHLD);
 521     sigprocmask(SIG_BLOCK, &mask, &oldmask);
 522
 523     if (!helper) {
 524         helper = default_helper = get_relocated_path(DEFAULT_BRIDGE_HELPER);
 525     }
 526
 527     if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) {
 528         error_setg_errno(errp, errno, "socketpair() failed");
 529         return -1;
 530     }
 531
 532     /* try to launch bridge helper */
 533     pid = fork();
 534     if (pid < 0) {
 535         error_setg_errno(errp, errno, "Can't fork bridge helper");
 536         return -1;
 537     }
 538     if (pid == 0) {
 539         int open_max = sysconf(_SC_OPEN_MAX), i;
 540         char *fd_buf = NULL;
 541         char *br_buf = NULL;
 542         char *helper_cmd = NULL;
 543
 544         for (i = 3; i < open_max; i++) {
 545             if (i != sv[1]) {
 546                 close(i);
 547             }
 548         }
 549
 550         fd_buf = g_strdup_printf("%s%d", "--fd=", sv[1]);
 551
 552         if (strrchr(helper, ' ') || strrchr(helper, '\t')) {
 553             /* assume helper is a command */
 554
 555             if (strstr(helper, "--br=") == NULL) {
 556                 br_buf = g_strdup_printf("%s%s", "--br=", bridge);
 557             }
 558
 559             helper_cmd = g_strdup_printf("%s %s %s %s", helper,
 560                             "--use-vnet", fd_buf, br_buf ? br_buf : "");
 561
 562             parg = args;
 563             *parg++ = (char *)"sh";
 564             *parg++ = (char *)"-c";
 565             *parg++ = helper_cmd;
 566             *parg++ = NULL;
 567
 568             execv("/bin/sh", args);
 569             g_free(helper_cmd);
 570         } else {
 571             /* assume helper is just the executable path name */
 572
 573             br_buf = g_strdup_printf("%s%s", "--br=", bridge);
 574
 575             parg = args;
 576             *parg++ = (char *)helper;
 577             *parg++ = (char *)"--use-vnet";
 578             *parg++ = fd_buf;
 579             *parg++ = br_buf;
 580             *parg++ = NULL;
 581
 582             execv(helper, args);
 583         }
 584         g_free(fd_buf);
 585         g_free(br_buf);
 586         _exit(1);
 587
 588     } else {
 589         int fd;
 590         int saved_errno;
 591
 592         close(sv[1]);
 593
 594         fd = RETRY_ON_EINTR(recv_fd(sv[0]));
 595         saved_errno = errno;
 596
 597         close(sv[0]);
 598
 599         while (waitpid(pid, &status, 0) != pid) {
 600             /* loop */
 601         }
 602         sigprocmask(SIG_SETMASK, &oldmask, NULL);
 603         if (fd < 0) {
 604             error_setg_errno(errp, saved_errno,
 605                              "failed to recv file descriptor");
 606             return -1;
 607         }
 608         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 609             error_setg(errp, "bridge helper failed");
 610             return -1;
 611         }
 612         return fd;
 613     }
 614 }
 615
 616 int net_init_bridge(const Netdev *netdev, const char *name,
 617                     NetClientState *peer, Error **errp)
 618 {
 619     const NetdevBridgeOptions *bridge;
 620     const char *helper, *br;
 621     TAPState *s;
 622     int fd, vnet_hdr;
 623
 624     assert(netdev->type == NET_CLIENT_DRIVER_BRIDGE);
 625     bridge = &netdev->u.bridge;
 626     helper = bridge->helper;
 627     br     = bridge->br ?: DEFAULT_BRIDGE_INTERFACE;
 628
 629     fd = net_bridge_run_helper(helper, br, errp);
 630     if (fd == -1) {
 631         return -1;
 632     }
 633
 634     if (!g_unix_set_fd_nonblocking(fd, true, NULL)) {
 635         error_setg_errno(errp, errno, "Failed to set FD nonblocking");
 636         return -1;
 637     }
 638     vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 639     if (vnet_hdr < 0) {
 640         close(fd);
 641         return -1;
 642     }
 643     s = net_tap_fd_init(peer, "bridge", name, fd, vnet_hdr);
 644
 645     qemu_set_info_str(&s->nc, "helper=%s,br=%s", helper, br);
 646
 647     return 0;
 648 }
 649
 650 static int net_tap_init(const NetdevTapOptions *tap, int *vnet_hdr,
 651                         const char *setup_script, char *ifname,
 652                         size_t ifname_sz, int mq_required, Error **errp)
 653 {
 654     Error *err = NULL;
 655     int fd, vnet_hdr_required;
 656
 657     if (tap->has_vnet_hdr) {
 658         *vnet_hdr = tap->vnet_hdr;
 659         vnet_hdr_required = *vnet_hdr;
 660     } else {
 661         *vnet_hdr = 1;
 662         vnet_hdr_required = 0;
 663     }
 664
 665     fd = RETRY_ON_EINTR(tap_open(ifname, ifname_sz, vnet_hdr, vnet_hdr_required,
 666                       mq_required, errp));
 667     if (fd < 0) {
 668         return -1;
 669     }
 670
 671     if (setup_script &&
 672         setup_script[0] != '\0' &&
 673         strcmp(setup_script, "no") != 0) {
 674         launch_script(setup_script, ifname, fd, &err);
 675         if (err) {
 676             error_propagate(errp, err);
 677             close(fd);
 678             return -1;
 679         }
 680     }
 681
 682     return fd;
 683 }
 684
 685 #define MAX_TAP_QUEUES 1024
 686
 687 static void net_init_tap_one(const NetdevTapOptions *tap, NetClientState *peer,
 688                              const char *model, const char *name,
 689                              const char *ifname, const char *script,
 690                              const char *downscript, const char *vhostfdname,
 691                              int vnet_hdr, int fd, Error **errp)
 692 {
 693     Error *err = NULL;
 694     TAPState *s = net_tap_fd_init(peer, model, name, fd, vnet_hdr);
 695     int vhostfd;
 696
 697     tap_set_sndbuf(s->fd, tap, &err);
 698     if (err) {
 699         error_propagate(errp, err);
 700         goto failed;
 701     }
 702
 703     if (tap->fd || tap->fds) {
 704         qemu_set_info_str(&s->nc, "fd=%d", fd);
 705     } else if (tap->helper) {
 706         qemu_set_info_str(&s->nc, "helper=%s", tap->helper);
 707     } else {
 708         qemu_set_info_str(&s->nc, "ifname=%s,script=%s,downscript=%s", ifname,
 709                           script, downscript);
 710
 711         if (strcmp(downscript, "no") != 0) {
 712             snprintf(s->down_script, sizeof(s->down_script), "%s", downscript);
 713             snprintf(s->down_script_arg, sizeof(s->down_script_arg),
 714                      "%s", ifname);
 715         }
 716     }
 717
 718     if (tap->has_vhost ? tap->vhost :
 719         vhostfdname || (tap->has_vhostforce && tap->vhostforce)) {
 720         VhostNetOptions options;
 721
 722         options.backend_type = VHOST_BACKEND_TYPE_KERNEL;
 723         options.net_backend = &s->nc;
 724         if (tap->has_poll_us) {
 725             options.busyloop_timeout = tap->poll_us;
 726         } else {
 727             options.busyloop_timeout = 0;
 728         }
 729
 730         if (vhostfdname) {
 731             vhostfd = monitor_fd_param(monitor_cur(), vhostfdname, &err);
 732             if (vhostfd == -1) {
 733                 if (tap->has_vhostforce && tap->vhostforce) {
 734                     error_propagate(errp, err);
 735                 } else {
 736                     warn_report_err(err);
 737                 }
 738                 goto failed;
 739             }
 740             if (!g_unix_set_fd_nonblocking(vhostfd, true, NULL)) {
 741                 error_setg_errno(errp, errno, "%s: Can't use file descriptor %d",
 742                                  name, fd);
 743                 goto failed;
 744             }
 745         } else {
 746             vhostfd = open("/dev/vhost-net", O_RDWR);
 747             if (vhostfd < 0) {
 748                 if (tap->has_vhostforce && tap->vhostforce) {
 749                     error_setg_errno(errp, errno,
 750                                      "tap: open vhost char device failed");
 751                 } else {
 752                     warn_report("tap: open vhost char device failed: %s",
 753                                 strerror(errno));
 754                 }
 755                 goto failed;
 756             }
 757             if (!g_unix_set_fd_nonblocking(vhostfd, true, NULL)) {
 758                 error_setg_errno(errp, errno, "Failed to set FD nonblocking");
 759                 goto failed;
 760             }
 761         }
 762         options.opaque = (void *)(uintptr_t)vhostfd;
 763         options.nvqs = 2;
 764
 765         s->vhost_net = vhost_net_init(&options);
 766         if (!s->vhost_net) {
 767             if (tap->has_vhostforce && tap->vhostforce) {
 768                 error_setg(errp, VHOST_NET_INIT_FAILED);
 769             } else {
 770                 warn_report(VHOST_NET_INIT_FAILED);
 771             }
 772             goto failed;
 773         }
 774     } else if (vhostfdname) {
 775         error_setg(errp, "vhostfd(s)= is not valid without vhost");
 776         goto failed;
 777     }
 778
 779     return;
 780
 781 failed:
 782     qemu_del_net_client(&s->nc);
 783 }
 784
 785 static int get_fds(char *str, char *fds[], int max)
 786 {
 787     char *ptr = str, *this;
 788     size_t len = strlen(str);
 789     int i = 0;
 790
 791     while (i < max && ptr < str + len) {
 792         this = strchr(ptr, ':');
 793
 794         if (this == NULL) {
 795             fds[i] = g_strdup(ptr);
 796         } else {
 797             fds[i] = g_strndup(ptr, this - ptr);
 798         }
 799
 800         i++;
 801         if (this == NULL) {
 802             break;
 803         } else {
 804             ptr = this + 1;
 805         }
 806     }
 807
 808     return i;
 809 }
 810
 811 int net_init_tap(const Netdev *netdev, const char *name,
 812                  NetClientState *peer, Error **errp)
 813 {
 814     const NetdevTapOptions *tap;
 815     int fd, vnet_hdr = 0, i = 0, queues;
 816     /* for the no-fd, no-helper case */
 817     const char *script;
 818     const char *downscript;
 819     Error *err = NULL;
 820     const char *vhostfdname;
 821     char ifname[128];
 822     int ret = 0;
 823
 824     assert(netdev->type == NET_CLIENT_DRIVER_TAP);
 825     tap = &netdev->u.tap;
 826     queues = tap->has_queues ? tap->queues : 1;
 827     vhostfdname = tap->vhostfd;
 828     script = tap->script;
 829     downscript = tap->downscript;
 830
 831     /* QEMU hubs do not support multiqueue tap, in this case peer is set.
 832      * For -netdev, peer is always NULL. */
 833     if (peer && (tap->has_queues || tap->fds || tap->vhostfds)) {
 834         error_setg(errp, "Multiqueue tap cannot be used with hubs");
 835         return -1;
 836     }
 837
 838     if (tap->fd) {
 839         if (tap->ifname || tap->script || tap->downscript ||
 840             tap->has_vnet_hdr || tap->helper || tap->has_queues ||
 841             tap->fds || tap->vhostfds) {
 842             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 843                        "helper=, queues=, fds=, and vhostfds= "
 844                        "are invalid with fd=");
 845             return -1;
 846         }
 847
 848         fd = monitor_fd_param(monitor_cur(), tap->fd, errp);
 849         if (fd == -1) {
 850             return -1;
 851         }
 852
 853         if (!g_unix_set_fd_nonblocking(fd, true, NULL)) {
 854             error_setg_errno(errp, errno, "%s: Can't use file descriptor %d",
 855                              name, fd);
 856             close(fd);
 857             return -1;
 858         }
 859
 860         vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 861         if (vnet_hdr < 0) {
 862             close(fd);
 863             return -1;
 864         }
 865
 866         net_init_tap_one(tap, peer, "tap", name, NULL,
 867                          script, downscript,
 868                          vhostfdname, vnet_hdr, fd, &err);
 869         if (err) {
 870             error_propagate(errp, err);
 871             close(fd);
 872             return -1;
 873         }
 874     } else if (tap->fds) {
 875         char **fds;
 876         char **vhost_fds;
 877         int nfds = 0, nvhosts = 0;
 878
 879         if (tap->ifname || tap->script || tap->downscript ||
 880             tap->has_vnet_hdr || tap->helper || tap->has_queues ||
 881             tap->vhostfd) {
 882             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 883                        "helper=, queues=, and vhostfd= "
 884                        "are invalid with fds=");
 885             return -1;
 886         }
 887
 888         fds = g_new0(char *, MAX_TAP_QUEUES);
 889         vhost_fds = g_new0(char *, MAX_TAP_QUEUES);
 890
 891         nfds = get_fds(tap->fds, fds, MAX_TAP_QUEUES);
 892         if (tap->vhostfds) {
 893             nvhosts = get_fds(tap->vhostfds, vhost_fds, MAX_TAP_QUEUES);
 894             if (nfds != nvhosts) {
 895                 error_setg(errp, "The number of fds passed does not match "
 896                            "the number of vhostfds passed");
 897                 ret = -1;
 898                 goto free_fail;
 899             }
 900         }
 901
 902         for (i = 0; i < nfds; i++) {
 903             fd = monitor_fd_param(monitor_cur(), fds[i], errp);
 904             if (fd == -1) {
 905                 ret = -1;
 906                 goto free_fail;
 907             }
 908
 909             ret = g_unix_set_fd_nonblocking(fd, true, NULL);
 910             if (!ret) {
 911                 error_setg_errno(errp, errno, "%s: Can't use file descriptor %d",
 912                                  name, fd);
 913                 goto free_fail;
 914             }
 915
 916             if (i == 0) {
 917                 vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 918                 if (vnet_hdr < 0) {
 919                     ret = -1;
 920                     goto free_fail;
 921                 }
 922             } else if (vnet_hdr != tap_probe_vnet_hdr(fd, NULL)) {
 923                 error_setg(errp,
 924                            "vnet_hdr not consistent across given tap fds");
 925                 ret = -1;
 926                 goto free_fail;
 927             }
 928
 929             net_init_tap_one(tap, peer, "tap", name, ifname,
 930                              script, downscript,
 931                              tap->vhostfds ? vhost_fds[i] : NULL,
 932                              vnet_hdr, fd, &err);
 933             if (err) {
 934                 error_propagate(errp, err);
 935                 ret = -1;
 936                 goto free_fail;
 937             }
 938         }
 939
 940 free_fail:
 941         for (i = 0; i < nvhosts; i++) {
 942             g_free(vhost_fds[i]);
 943         }
 944         for (i = 0; i < nfds; i++) {
 945             g_free(fds[i]);
 946         }
 947         g_free(fds);
 948         g_free(vhost_fds);
 949         return ret;
 950     } else if (tap->helper) {
 951         if (tap->ifname || tap->script || tap->downscript ||
 952             tap->has_vnet_hdr || tap->has_queues || tap->vhostfds) {
 953             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 954                        "queues=, and vhostfds= are invalid with helper=");
 955             return -1;
 956         }
 957
 958         fd = net_bridge_run_helper(tap->helper,
 959                                    tap->br ?: DEFAULT_BRIDGE_INTERFACE,
 960                                    errp);
 961         if (fd == -1) {
 962             return -1;
 963         }
 964
 965         if (!g_unix_set_fd_nonblocking(fd, true, NULL)) {
 966             error_setg_errno(errp, errno, "Failed to set FD nonblocking");
 967             return -1;
 968         }
 969         vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 970         if (vnet_hdr < 0) {
 971             close(fd);
 972             return -1;
 973         }
 974
 975         net_init_tap_one(tap, peer, "bridge", name, ifname,
 976                          script, downscript, vhostfdname,
 977                          vnet_hdr, fd, &err);
 978         if (err) {
 979             error_propagate(errp, err);
 980             close(fd);
 981             return -1;
 982         }
 983     } else {
 984         g_autofree char *default_script = NULL;
 985         g_autofree char *default_downscript = NULL;
 986         if (tap->vhostfds) {
 987             error_setg(errp, "vhostfds= is invalid if fds= wasn't specified");
 988             return -1;
 989         }
 990
 991         if (!script) {
 992             script = default_script = get_relocated_path(DEFAULT_NETWORK_SCRIPT);
 993         }
 994         if (!downscript) {
 995             downscript = default_downscript =
 996                                  get_relocated_path(DEFAULT_NETWORK_DOWN_SCRIPT);
 997         }
 998
 999         if (tap->ifname) {
1000             pstrcpy(ifname, sizeof ifname, tap->ifname);
1001         } else {
1002             ifname[0] = '\0';
1003         }
1004
1005         for (i = 0; i < queues; i++) {
1006             fd = net_tap_init(tap, &vnet_hdr, i >= 1 ? "no" : script,
1007                               ifname, sizeof ifname, queues > 1, errp);
1008             if (fd == -1) {
1009                 return -1;
1010             }
1011
1012             if (queues > 1 && i == 0 && !tap->ifname) {
1013                 if (tap_fd_get_ifname(fd, ifname)) {
1014                     error_setg(errp, "Fail to get ifname");
1015                     close(fd);
1016                     return -1;
1017                 }
1018             }
1019
1020             net_init_tap_one(tap, peer, "tap", name, ifname,
1021                              i >= 1 ? "no" : script,
1022                              i >= 1 ? "no" : downscript,
1023                              vhostfdname, vnet_hdr, fd, &err);
1024             if (err) {
1025                 error_propagate(errp, err);
1026                 close(fd);
1027                 return -1;
1028             }
1029         }
1030     }
1031
1032     return 0;
1033 }
1034
1035 VHostNetState *tap_get_vhost_net(NetClientState *nc)
1036 {
1037     TAPState *s = DO_UPCAST(TAPState, nc, nc);
1038     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
1039     return s->vhost_net;
1040 }
1041
1042 int tap_enable(NetClientState *nc)
1043 {
1044     TAPState *s = DO_UPCAST(TAPState, nc, nc);
1045     int ret;
1046
1047     if (s->enabled) {
1048         return 0;
1049     } else {
1050         ret = tap_fd_enable(s->fd);
1051         if (ret == 0) {
1052             s->enabled = true;
1053             tap_update_fd_handler(s);
1054         }
1055         return ret;
1056     }
1057 }
1058
1059 int tap_disable(NetClientState *nc)
1060 {
1061     TAPState *s = DO_UPCAST(TAPState, nc, nc);
1062     int ret;
1063
1064     if (s->enabled == 0) {
1065         return 0;
1066     } else {
1067         ret = tap_fd_disable(s->fd);
1068         if (ret == 0) {
1069             qemu_purge_queued_packets(nc);
1070             s->enabled = false;
1071             tap_update_fd_handler(s);
1072         }
1073         return ret;
1074     }
1075 }