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 "qemu-common.h"
  40 #include "qemu/error-report.h"
  41
  42 #include "net/tap.h"
  43
  44 #include "net/vhost_net.h"
  45
  46 typedef struct TAPState {
  47     NetClientState nc;
  48     int fd;
  49     char down_script[1024];
  50     char down_script_arg[128];
  51     uint8_t buf[NET_BUFSIZE];
  52     bool read_poll;
  53     bool write_poll;
  54     bool using_vnet_hdr;
  55     bool has_ufo;
  56     bool enabled;
  57     VHostNetState *vhost_net;
  58     unsigned host_vnet_hdr_len;
  59 } TAPState;
  60
  61 static void launch_script(const char *setup_script, const char *ifname,
  62                           int fd, Error **errp);
  63
  64 static void tap_send(void *opaque);
  65 static void tap_writable(void *opaque);
  66
  67 static void tap_update_fd_handler(TAPState *s)
  68 {
  69     qemu_set_fd_handler(s->fd,
  70                         s->read_poll && s->enabled ? tap_send : NULL,
  71                         s->write_poll && s->enabled ? tap_writable : NULL,
  72                         s);
  73 }
  74
  75 static void tap_read_poll(TAPState *s, bool enable)
  76 {
  77     s->read_poll = enable;
  78     tap_update_fd_handler(s);
  79 }
  80
  81 static void tap_write_poll(TAPState *s, bool enable)
  82 {
  83     s->write_poll = enable;
  84     tap_update_fd_handler(s);
  85 }
  86
  87 static void tap_writable(void *opaque)
  88 {
  89     TAPState *s = opaque;
  90
  91     tap_write_poll(s, false);
  92
  93     qemu_flush_queued_packets(&s->nc);
  94 }
  95
  96 static ssize_t tap_write_packet(TAPState *s, const struct iovec *iov, int iovcnt)
  97 {
  98     ssize_t len;
  99
 100     do {
 101         len = writev(s->fd, iov, iovcnt);
 102     } while (len == -1 && errno == EINTR);
 103
 104     if (len == -1 && errno == EAGAIN) {
 105         tap_write_poll(s, true);
 106         return 0;
 107     }
 108
 109     return len;
 110 }
 111
 112 static ssize_t tap_receive_iov(NetClientState *nc, const struct iovec *iov,
 113                                int iovcnt)
 114 {
 115     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 116     const struct iovec *iovp = iov;
 117     struct iovec iov_copy[iovcnt + 1];
 118     struct virtio_net_hdr_mrg_rxbuf hdr = { };
 119
 120     if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
 121         iov_copy[0].iov_base = &hdr;
 122         iov_copy[0].iov_len =  s->host_vnet_hdr_len;
 123         memcpy(&iov_copy[1], iov, iovcnt * sizeof(*iov));
 124         iovp = iov_copy;
 125         iovcnt++;
 126     }
 127
 128     return tap_write_packet(s, iovp, iovcnt);
 129 }
 130
 131 static ssize_t tap_receive_raw(NetClientState *nc, const uint8_t *buf, size_t size)
 132 {
 133     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 134     struct iovec iov[2];
 135     int iovcnt = 0;
 136     struct virtio_net_hdr_mrg_rxbuf hdr = { };
 137
 138     if (s->host_vnet_hdr_len) {
 139         iov[iovcnt].iov_base = &hdr;
 140         iov[iovcnt].iov_len  = s->host_vnet_hdr_len;
 141         iovcnt++;
 142     }
 143
 144     iov[iovcnt].iov_base = (char *)buf;
 145     iov[iovcnt].iov_len  = size;
 146     iovcnt++;
 147
 148     return tap_write_packet(s, iov, iovcnt);
 149 }
 150
 151 static ssize_t tap_receive(NetClientState *nc, const uint8_t *buf, size_t size)
 152 {
 153     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 154     struct iovec iov[1];
 155
 156     if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
 157         return tap_receive_raw(nc, buf, size);
 158     }
 159
 160     iov[0].iov_base = (char *)buf;
 161     iov[0].iov_len  = size;
 162
 163     return tap_write_packet(s, iov, 1);
 164 }
 165
 166 #ifndef __sun__
 167 ssize_t tap_read_packet(int tapfd, uint8_t *buf, int maxlen)
 168 {
 169     return read(tapfd, buf, maxlen);
 170 }
 171 #endif
 172
 173 static void tap_send_completed(NetClientState *nc, ssize_t len)
 174 {
 175     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 176     tap_read_poll(s, true);
 177 }
 178
 179 static void tap_send(void *opaque)
 180 {
 181     TAPState *s = opaque;
 182     int size;
 183     int packets = 0;
 184
 185     while (true) {
 186         uint8_t *buf = s->buf;
 187
 188         size = tap_read_packet(s->fd, s->buf, sizeof(s->buf));
 189         if (size <= 0) {
 190             break;
 191         }
 192
 193         if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
 194             buf  += s->host_vnet_hdr_len;
 195             size -= s->host_vnet_hdr_len;
 196         }
 197
 198         size = qemu_send_packet_async(&s->nc, buf, size, tap_send_completed);
 199         if (size == 0) {
 200             tap_read_poll(s, false);
 201             break;
 202         } else if (size < 0) {
 203             break;
 204         }
 205
 206         /*
 207          * When the host keeps receiving more packets while tap_send() is
 208          * running we can hog the QEMU global mutex.  Limit the number of
 209          * packets that are processed per tap_send() callback to prevent
 210          * stalling the guest.
 211          */
 212         packets++;
 213         if (packets >= 50) {
 214             break;
 215         }
 216     }
 217 }
 218
 219 static bool tap_has_ufo(NetClientState *nc)
 220 {
 221     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 222
 223     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
 224
 225     return s->has_ufo;
 226 }
 227
 228 static bool tap_has_vnet_hdr(NetClientState *nc)
 229 {
 230     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 231
 232     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
 233
 234     return !!s->host_vnet_hdr_len;
 235 }
 236
 237 static bool tap_has_vnet_hdr_len(NetClientState *nc, int len)
 238 {
 239     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 240
 241     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
 242
 243     return !!tap_probe_vnet_hdr_len(s->fd, len);
 244 }
 245
 246 static void tap_set_vnet_hdr_len(NetClientState *nc, int len)
 247 {
 248     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 249
 250     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
 251     assert(len == sizeof(struct virtio_net_hdr_mrg_rxbuf) ||
 252            len == sizeof(struct virtio_net_hdr));
 253
 254     tap_fd_set_vnet_hdr_len(s->fd, len);
 255     s->host_vnet_hdr_len = len;
 256 }
 257
 258 static void tap_using_vnet_hdr(NetClientState *nc, bool using_vnet_hdr)
 259 {
 260     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 261
 262     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
 263     assert(!!s->host_vnet_hdr_len == using_vnet_hdr);
 264
 265     s->using_vnet_hdr = using_vnet_hdr;
 266 }
 267
 268 static int tap_set_vnet_le(NetClientState *nc, bool is_le)
 269 {
 270     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 271
 272     return tap_fd_set_vnet_le(s->fd, is_le);
 273 }
 274
 275 static int tap_set_vnet_be(NetClientState *nc, bool is_be)
 276 {
 277     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 278
 279     return tap_fd_set_vnet_be(s->fd, is_be);
 280 }
 281
 282 static void tap_set_offload(NetClientState *nc, int csum, int tso4,
 283                      int tso6, int ecn, int ufo)
 284 {
 285     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 286     if (s->fd < 0) {
 287         return;
 288     }
 289
 290     tap_fd_set_offload(s->fd, csum, tso4, tso6, ecn, ufo);
 291 }
 292
 293 static void tap_cleanup(NetClientState *nc)
 294 {
 295     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 296     Error *err = NULL;
 297
 298     if (s->vhost_net) {
 299         vhost_net_cleanup(s->vhost_net);
 300         s->vhost_net = NULL;
 301     }
 302
 303     qemu_purge_queued_packets(nc);
 304
 305     if (s->down_script[0]) {
 306         launch_script(s->down_script, s->down_script_arg, s->fd, &err);
 307         if (err) {
 308             error_report_err(err);
 309         }
 310     }
 311
 312     tap_read_poll(s, false);
 313     tap_write_poll(s, false);
 314     close(s->fd);
 315     s->fd = -1;
 316 }
 317
 318 static void tap_poll(NetClientState *nc, bool enable)
 319 {
 320     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 321     tap_read_poll(s, enable);
 322     tap_write_poll(s, enable);
 323 }
 324
 325 int tap_get_fd(NetClientState *nc)
 326 {
 327     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 328     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
 329     return s->fd;
 330 }
 331
 332 /* fd support */
 333
 334 static NetClientInfo net_tap_info = {
 335     .type = NET_CLIENT_OPTIONS_KIND_TAP,
 336     .size = sizeof(TAPState),
 337     .receive = tap_receive,
 338     .receive_raw = tap_receive_raw,
 339     .receive_iov = tap_receive_iov,
 340     .poll = tap_poll,
 341     .cleanup = tap_cleanup,
 342     .has_ufo = tap_has_ufo,
 343     .has_vnet_hdr = tap_has_vnet_hdr,
 344     .has_vnet_hdr_len = tap_has_vnet_hdr_len,
 345     .using_vnet_hdr = tap_using_vnet_hdr,
 346     .set_offload = tap_set_offload,
 347     .set_vnet_hdr_len = tap_set_vnet_hdr_len,
 348     .set_vnet_le = tap_set_vnet_le,
 349     .set_vnet_be = tap_set_vnet_be,
 350 };
 351
 352 static TAPState *net_tap_fd_init(NetClientState *peer,
 353                                  const char *model,
 354                                  const char *name,
 355                                  int fd,
 356                                  int vnet_hdr)
 357 {
 358     NetClientState *nc;
 359     TAPState *s;
 360
 361     nc = qemu_new_net_client(&net_tap_info, peer, model, name);
 362
 363     s = DO_UPCAST(TAPState, nc, nc);
 364
 365     s->fd = fd;
 366     s->host_vnet_hdr_len = vnet_hdr ? sizeof(struct virtio_net_hdr) : 0;
 367     s->using_vnet_hdr = false;
 368     s->has_ufo = tap_probe_has_ufo(s->fd);
 369     s->enabled = true;
 370     tap_set_offload(&s->nc, 0, 0, 0, 0, 0);
 371     /*
 372      * Make sure host header length is set correctly in tap:
 373      * it might have been modified by another instance of qemu.
 374      */
 375     if (tap_probe_vnet_hdr_len(s->fd, s->host_vnet_hdr_len)) {
 376         tap_fd_set_vnet_hdr_len(s->fd, s->host_vnet_hdr_len);
 377     }
 378     tap_read_poll(s, true);
 379     s->vhost_net = NULL;
 380     return s;
 381 }
 382
 383 static void launch_script(const char *setup_script, const char *ifname,
 384                           int fd, Error **errp)
 385 {
 386     int pid, status;
 387     char *args[3];
 388     char **parg;
 389
 390     /* try to launch network script */
 391     pid = fork();
 392     if (pid < 0) {
 393         error_setg_errno(errp, errno, "could not launch network script %s",
 394                          setup_script);
 395         return;
 396     }
 397     if (pid == 0) {
 398         int open_max = sysconf(_SC_OPEN_MAX), i;
 399
 400         for (i = 3; i < open_max; i++) {
 401             if (i != fd) {
 402                 close(i);
 403             }
 404         }
 405         parg = args;
 406         *parg++ = (char *)setup_script;
 407         *parg++ = (char *)ifname;
 408         *parg = NULL;
 409         execv(setup_script, args);
 410         _exit(1);
 411     } else {
 412         while (waitpid(pid, &status, 0) != pid) {
 413             /* loop */
 414         }
 415
 416         if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
 417             return;
 418         }
 419         error_setg(errp, "network script %s failed with status %d",
 420                    setup_script, status);
 421     }
 422 }
 423
 424 static int recv_fd(int c)
 425 {
 426     int fd;
 427     uint8_t msgbuf[CMSG_SPACE(sizeof(fd))];
 428     struct msghdr msg = {
 429         .msg_control = msgbuf,
 430         .msg_controllen = sizeof(msgbuf),
 431     };
 432     struct cmsghdr *cmsg;
 433     struct iovec iov;
 434     uint8_t req[1];
 435     ssize_t len;
 436
 437     cmsg = CMSG_FIRSTHDR(&msg);
 438     cmsg->cmsg_level = SOL_SOCKET;
 439     cmsg->cmsg_type = SCM_RIGHTS;
 440     cmsg->cmsg_len = CMSG_LEN(sizeof(fd));
 441     msg.msg_controllen = cmsg->cmsg_len;
 442
 443     iov.iov_base = req;
 444     iov.iov_len = sizeof(req);
 445
 446     msg.msg_iov = &iov;
 447     msg.msg_iovlen = 1;
 448
 449     len = recvmsg(c, &msg, 0);
 450     if (len > 0) {
 451         memcpy(&fd, CMSG_DATA(cmsg), sizeof(fd));
 452         return fd;
 453     }
 454
 455     return len;
 456 }
 457
 458 static int net_bridge_run_helper(const char *helper, const char *bridge,
 459                                  Error **errp)
 460 {
 461     sigset_t oldmask, mask;
 462     int pid, status;
 463     char *args[5];
 464     char **parg;
 465     int sv[2];
 466
 467     sigemptyset(&mask);
 468     sigaddset(&mask, SIGCHLD);
 469     sigprocmask(SIG_BLOCK, &mask, &oldmask);
 470
 471     if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) {
 472         error_setg_errno(errp, errno, "socketpair() failed");
 473         return -1;
 474     }
 475
 476     /* try to launch bridge helper */
 477     pid = fork();
 478     if (pid < 0) {
 479         error_setg_errno(errp, errno, "Can't fork bridge helper");
 480         return -1;
 481     }
 482     if (pid == 0) {
 483         int open_max = sysconf(_SC_OPEN_MAX), i;
 484         char fd_buf[6+10];
 485         char br_buf[6+IFNAMSIZ] = {0};
 486         char helper_cmd[PATH_MAX + sizeof(fd_buf) + sizeof(br_buf) + 15];
 487
 488         for (i = 3; i < open_max; i++) {
 489             if (i != sv[1]) {
 490                 close(i);
 491             }
 492         }
 493
 494         snprintf(fd_buf, sizeof(fd_buf), "%s%d", "--fd=", sv[1]);
 495
 496         if (strrchr(helper, ' ') || strrchr(helper, '\t')) {
 497             /* assume helper is a command */
 498
 499             if (strstr(helper, "--br=") == NULL) {
 500                 snprintf(br_buf, sizeof(br_buf), "%s%s", "--br=", bridge);
 501             }
 502
 503             snprintf(helper_cmd, sizeof(helper_cmd), "%s %s %s %s",
 504                      helper, "--use-vnet", fd_buf, br_buf);
 505
 506             parg = args;
 507             *parg++ = (char *)"sh";
 508             *parg++ = (char *)"-c";
 509             *parg++ = helper_cmd;
 510             *parg++ = NULL;
 511
 512             execv("/bin/sh", args);
 513         } else {
 514             /* assume helper is just the executable path name */
 515
 516             snprintf(br_buf, sizeof(br_buf), "%s%s", "--br=", bridge);
 517
 518             parg = args;
 519             *parg++ = (char *)helper;
 520             *parg++ = (char *)"--use-vnet";
 521             *parg++ = fd_buf;
 522             *parg++ = br_buf;
 523             *parg++ = NULL;
 524
 525             execv(helper, args);
 526         }
 527         _exit(1);
 528
 529     } else {
 530         int fd;
 531         int saved_errno;
 532
 533         close(sv[1]);
 534
 535         do {
 536             fd = recv_fd(sv[0]);
 537         } while (fd == -1 && errno == EINTR);
 538         saved_errno = errno;
 539
 540         close(sv[0]);
 541
 542         while (waitpid(pid, &status, 0) != pid) {
 543             /* loop */
 544         }
 545         sigprocmask(SIG_SETMASK, &oldmask, NULL);
 546         if (fd < 0) {
 547             error_setg_errno(errp, saved_errno,
 548                              "failed to recv file descriptor");
 549             return -1;
 550         }
 551         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 552             error_setg(errp, "bridge helper failed");
 553             return -1;
 554         }
 555         return fd;
 556     }
 557 }
 558
 559 int net_init_bridge(const NetClientOptions *opts, const char *name,
 560                     NetClientState *peer, Error **errp)
 561 {
 562     const NetdevBridgeOptions *bridge;
 563     const char *helper, *br;
 564     TAPState *s;
 565     int fd, vnet_hdr;
 566
 567     assert(opts->type == NET_CLIENT_OPTIONS_KIND_BRIDGE);
 568     bridge = opts->u.bridge;
 569
 570     helper = bridge->has_helper ? bridge->helper : DEFAULT_BRIDGE_HELPER;
 571     br     = bridge->has_br     ? bridge->br     : DEFAULT_BRIDGE_INTERFACE;
 572
 573     fd = net_bridge_run_helper(helper, br, errp);
 574     if (fd == -1) {
 575         return -1;
 576     }
 577
 578     fcntl(fd, F_SETFL, O_NONBLOCK);
 579     vnet_hdr = tap_probe_vnet_hdr(fd);
 580     s = net_tap_fd_init(peer, "bridge", name, fd, vnet_hdr);
 581
 582     snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s,br=%s", helper,
 583              br);
 584
 585     return 0;
 586 }
 587
 588 static int net_tap_init(const NetdevTapOptions *tap, int *vnet_hdr,
 589                         const char *setup_script, char *ifname,
 590                         size_t ifname_sz, int mq_required, Error **errp)
 591 {
 592     Error *err = NULL;
 593     int fd, vnet_hdr_required;
 594
 595     if (tap->has_vnet_hdr) {
 596         *vnet_hdr = tap->vnet_hdr;
 597         vnet_hdr_required = *vnet_hdr;
 598     } else {
 599         *vnet_hdr = 1;
 600         vnet_hdr_required = 0;
 601     }
 602
 603     TFR(fd = tap_open(ifname, ifname_sz, vnet_hdr, vnet_hdr_required,
 604                       mq_required, errp));
 605     if (fd < 0) {
 606         return -1;
 607     }
 608
 609     if (setup_script &&
 610         setup_script[0] != '\0' &&
 611         strcmp(setup_script, "no") != 0) {
 612         launch_script(setup_script, ifname, fd, &err);
 613         if (err) {
 614             error_propagate(errp, err);
 615             close(fd);
 616             return -1;
 617         }
 618     }
 619
 620     return fd;
 621 }
 622
 623 #define MAX_TAP_QUEUES 1024
 624
 625 static void net_init_tap_one(const NetdevTapOptions *tap, NetClientState *peer,
 626                              const char *model, const char *name,
 627                              const char *ifname, const char *script,
 628                              const char *downscript, const char *vhostfdname,
 629                              int vnet_hdr, int fd, Error **errp)
 630 {
 631     Error *err = NULL;
 632     TAPState *s = net_tap_fd_init(peer, model, name, fd, vnet_hdr);
 633     int vhostfd;
 634
 635     tap_set_sndbuf(s->fd, tap, &err);
 636     if (err) {
 637         error_propagate(errp, err);
 638         return;
 639     }
 640
 641     if (tap->has_fd || tap->has_fds) {
 642         snprintf(s->nc.info_str, sizeof(s->nc.info_str), "fd=%d", fd);
 643     } else if (tap->has_helper) {
 644         snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s",
 645                  tap->helper);
 646     } else {
 647         snprintf(s->nc.info_str, sizeof(s->nc.info_str),
 648                  "ifname=%s,script=%s,downscript=%s", ifname, script,
 649                  downscript);
 650
 651         if (strcmp(downscript, "no") != 0) {
 652             snprintf(s->down_script, sizeof(s->down_script), "%s", downscript);
 653             snprintf(s->down_script_arg, sizeof(s->down_script_arg),
 654                      "%s", ifname);
 655         }
 656     }
 657
 658     if (tap->has_vhost ? tap->vhost :
 659         vhostfdname || (tap->has_vhostforce && tap->vhostforce)) {
 660         VhostNetOptions options;
 661
 662         options.backend_type = VHOST_BACKEND_TYPE_KERNEL;
 663         options.net_backend = &s->nc;
 664
 665         if (vhostfdname) {
 666             vhostfd = monitor_fd_param(cur_mon, vhostfdname, &err);
 667             if (vhostfd == -1) {
 668                 error_propagate(errp, err);
 669                 return;
 670             }
 671         } else {
 672             vhostfd = open("/dev/vhost-net", O_RDWR);
 673             if (vhostfd < 0) {
 674                 error_setg_errno(errp, errno,
 675                                  "tap: open vhost char device failed");
 676                 return;
 677             }
 678         }
 679         options.opaque = (void *)(uintptr_t)vhostfd;
 680
 681         s->vhost_net = vhost_net_init(&options);
 682         if (!s->vhost_net) {
 683             error_setg(errp,
 684                        "vhost-net requested but could not be initialized");
 685             return;
 686         }
 687     } else if (vhostfdname) {
 688         error_setg(errp, "vhostfd= is not valid without vhost");
 689     }
 690 }
 691
 692 static int get_fds(char *str, char *fds[], int max)
 693 {
 694     char *ptr = str, *this;
 695     size_t len = strlen(str);
 696     int i = 0;
 697
 698     while (i < max && ptr < str + len) {
 699         this = strchr(ptr, ':');
 700
 701         if (this == NULL) {
 702             fds[i] = g_strdup(ptr);
 703         } else {
 704             fds[i] = g_strndup(ptr, this - ptr);
 705         }
 706
 707         i++;
 708         if (this == NULL) {
 709             break;
 710         } else {
 711             ptr = this + 1;
 712         }
 713     }
 714
 715     return i;
 716 }
 717
 718 int net_init_tap(const NetClientOptions *opts, const char *name,
 719                  NetClientState *peer, Error **errp)
 720 {
 721     const NetdevTapOptions *tap;
 722     int fd, vnet_hdr = 0, i = 0, queues;
 723     /* for the no-fd, no-helper case */
 724     const char *script = NULL; /* suppress wrong "uninit'd use" gcc warning */
 725     const char *downscript = NULL;
 726     Error *err = NULL;
 727     const char *vhostfdname;
 728     char ifname[128];
 729
 730     assert(opts->type == NET_CLIENT_OPTIONS_KIND_TAP);
 731     tap = opts->u.tap;
 732     queues = tap->has_queues ? tap->queues : 1;
 733     vhostfdname = tap->has_vhostfd ? tap->vhostfd : NULL;
 734
 735     /* QEMU vlans does not support multiqueue tap, in this case peer is set.
 736      * For -netdev, peer is always NULL. */
 737     if (peer && (tap->has_queues || tap->has_fds || tap->has_vhostfds)) {
 738         error_setg(errp, "Multiqueue tap cannot be used with QEMU vlans");
 739         return -1;
 740     }
 741
 742     if (tap->has_fd) {
 743         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
 744             tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
 745             tap->has_fds || tap->has_vhostfds) {
 746             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 747                        "helper=, queues=, fds=, and vhostfds= "
 748                        "are invalid with fd=");
 749             return -1;
 750         }
 751
 752         fd = monitor_fd_param(cur_mon, tap->fd, &err);
 753         if (fd == -1) {
 754             error_propagate(errp, err);
 755             return -1;
 756         }
 757
 758         fcntl(fd, F_SETFL, O_NONBLOCK);
 759
 760         vnet_hdr = tap_probe_vnet_hdr(fd);
 761
 762         net_init_tap_one(tap, peer, "tap", name, NULL,
 763                          script, downscript,
 764                          vhostfdname, vnet_hdr, fd, &err);
 765         if (err) {
 766             error_propagate(errp, err);
 767             return -1;
 768         }
 769     } else if (tap->has_fds) {
 770         char *fds[MAX_TAP_QUEUES];
 771         char *vhost_fds[MAX_TAP_QUEUES];
 772         int nfds, nvhosts;
 773
 774         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
 775             tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
 776             tap->has_vhostfd) {
 777             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 778                        "helper=, queues=, and vhostfd= "
 779                        "are invalid with fds=");
 780             return -1;
 781         }
 782
 783         nfds = get_fds(tap->fds, fds, MAX_TAP_QUEUES);
 784         if (tap->has_vhostfds) {
 785             nvhosts = get_fds(tap->vhostfds, vhost_fds, MAX_TAP_QUEUES);
 786             if (nfds != nvhosts) {
 787                 error_setg(errp, "The number of fds passed does not match "
 788                            "the number of vhostfds passed");
 789                 return -1;
 790             }
 791         }
 792
 793         for (i = 0; i < nfds; i++) {
 794             fd = monitor_fd_param(cur_mon, fds[i], &err);
 795             if (fd == -1) {
 796                 error_propagate(errp, err);
 797                 return -1;
 798             }
 799
 800             fcntl(fd, F_SETFL, O_NONBLOCK);
 801
 802             if (i == 0) {
 803                 vnet_hdr = tap_probe_vnet_hdr(fd);
 804             } else if (vnet_hdr != tap_probe_vnet_hdr(fd)) {
 805                 error_setg(errp,
 806                            "vnet_hdr not consistent across given tap fds");
 807                 return -1;
 808             }
 809
 810             net_init_tap_one(tap, peer, "tap", name, ifname,
 811                              script, downscript,
 812                              tap->has_vhostfds ? vhost_fds[i] : NULL,
 813                              vnet_hdr, fd, &err);
 814             if (err) {
 815                 error_propagate(errp, err);
 816                 return -1;
 817             }
 818         }
 819     } else if (tap->has_helper) {
 820         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
 821             tap->has_vnet_hdr || tap->has_queues || tap->has_vhostfds) {
 822             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 823                        "queues=, and vhostfds= are invalid with helper=");
 824             return -1;
 825         }
 826
 827         fd = net_bridge_run_helper(tap->helper, DEFAULT_BRIDGE_INTERFACE,
 828                                    errp);
 829         if (fd == -1) {
 830             return -1;
 831         }
 832
 833         fcntl(fd, F_SETFL, O_NONBLOCK);
 834         vnet_hdr = tap_probe_vnet_hdr(fd);
 835
 836         net_init_tap_one(tap, peer, "bridge", name, ifname,
 837                          script, downscript, vhostfdname,
 838                          vnet_hdr, fd, &err);
 839         if (err) {
 840             error_propagate(errp, err);
 841             close(fd);
 842             return -1;
 843         }
 844     } else {
 845         if (tap->has_vhostfds) {
 846             error_setg(errp, "vhostfds= is invalid if fds= wasn't specified");
 847             return -1;
 848         }
 849         script = tap->has_script ? tap->script : DEFAULT_NETWORK_SCRIPT;
 850         downscript = tap->has_downscript ? tap->downscript :
 851             DEFAULT_NETWORK_DOWN_SCRIPT;
 852
 853         if (tap->has_ifname) {
 854             pstrcpy(ifname, sizeof ifname, tap->ifname);
 855         } else {
 856             ifname[0] = '\0';
 857         }
 858
 859         for (i = 0; i < queues; i++) {
 860             fd = net_tap_init(tap, &vnet_hdr, i >= 1 ? "no" : script,
 861                               ifname, sizeof ifname, queues > 1, errp);
 862             if (fd == -1) {
 863                 return -1;
 864             }
 865
 866             if (queues > 1 && i == 0 && !tap->has_ifname) {
 867                 if (tap_fd_get_ifname(fd, ifname)) {
 868                     error_setg(errp, "Fail to get ifname");
 869                     close(fd);
 870                     return -1;
 871                 }
 872             }
 873
 874             net_init_tap_one(tap, peer, "tap", name, ifname,
 875                              i >= 1 ? "no" : script,
 876                              i >= 1 ? "no" : downscript,
 877                              vhostfdname, vnet_hdr, fd, &err);
 878             if (err) {
 879                 error_propagate(errp, err);
 880                 close(fd);
 881                 return -1;
 882             }
 883         }
 884     }
 885
 886     return 0;
 887 }
 888
 889 VHostNetState *tap_get_vhost_net(NetClientState *nc)
 890 {
 891     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 892     assert(nc->info->type == NET_CLIENT_OPTIONS_KIND_TAP);
 893     return s->vhost_net;
 894 }
 895
 896 int tap_enable(NetClientState *nc)
 897 {
 898     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 899     int ret;
 900
 901     if (s->enabled) {
 902         return 0;
 903     } else {
 904         ret = tap_fd_enable(s->fd);
 905         if (ret == 0) {
 906             s->enabled = true;
 907             tap_update_fd_handler(s);
 908         }
 909         return ret;
 910     }
 911 }
 912
 913 int tap_disable(NetClientState *nc)
 914 {
 915     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 916     int ret;
 917
 918     if (s->enabled == 0) {
 919         return 0;
 920     } else {
 921         ret = tap_fd_disable(s->fd);
 922         if (ret == 0) {
 923             qemu_purge_queued_packets(nc);
 924             s->enabled = false;
 925             tap_update_fd_handler(s);
 926         }
 927         return ret;
 928     }
 929 }