net/tap.c

   1 /*
   2  * QEMU System Emulator
   3  *
   4  * Copyright (c) 2003-2008 Fabrice Bellard
   5  * Copyright (c) 2009 Red Hat, Inc.
   6  *
   7  * Permission is hereby granted, free of charge, to any person obtaining a copy
   8  * of this software and associated documentation files (the "Software"), to deal
   9  * in the Software without restriction, including without limitation the rights
  10  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  11  * copies of the Software, and to permit persons to whom the Software is
  12  * furnished to do so, subject to the following conditions:
  13  *
  14  * The above copyright notice and this permission notice shall be included in
  15  * all copies or substantial portions of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  23  * THE SOFTWARE.
  24  */
  25
  26 #include "qemu/osdep.h"
  27 #include "tap_int.h"
  28
  29
  30 #include <sys/ioctl.h>
  31 #include <sys/wait.h>
  32 #include <sys/socket.h>
  33 #include <net/if.h>
  34
  35 #include "net/net.h"
  36 #include "clients.h"
  37 #include "monitor/monitor.h"
  38 #include "sysemu/sysemu.h"
  39 #include "qapi/error.h"
  40 #include "qemu-common.h"
  41 #include "qemu/cutils.h"
  42 #include "qemu/error-report.h"
  43 #include "qemu/main-loop.h"
  44 #include "qemu/sockets.h"
  45
  46 #include "net/tap.h"
  47
  48 #include "net/vhost_net.h"
  49
  50 typedef struct TAPState {
  51     NetClientState nc;
  52     int fd;
  53     char down_script[1024];
  54     char down_script_arg[128];
  55     uint8_t buf[NET_BUFSIZE];
  56     bool read_poll;
  57     bool write_poll;
  58     bool using_vnet_hdr;
  59     bool has_ufo;
  60     bool enabled;
  61     VHostNetState *vhost_net;
  62     unsigned host_vnet_hdr_len;
  63     Notifier exit;
  64 } TAPState;
  65
  66 static void launch_script(const char *setup_script, const char *ifname,
  67                           int fd, Error **errp);
  68
  69 static void tap_send(void *opaque);
  70 static void tap_writable(void *opaque);
  71
  72 static void tap_update_fd_handler(TAPState *s)
  73 {
  74     qemu_set_fd_handler(s->fd,
  75                         s->read_poll && s->enabled ? tap_send : NULL,
  76                         s->write_poll && s->enabled ? tap_writable : NULL,
  77                         s);
  78 }
  79
  80 static void tap_read_poll(TAPState *s, bool enable)
  81 {
  82     s->read_poll = enable;
  83     tap_update_fd_handler(s);
  84 }
  85
  86 static void tap_write_poll(TAPState *s, bool enable)
  87 {
  88     s->write_poll = enable;
  89     tap_update_fd_handler(s);
  90 }
  91
  92 static void tap_writable(void *opaque)
  93 {
  94     TAPState *s = opaque;
  95
  96     tap_write_poll(s, false);
  97
  98     qemu_flush_queued_packets(&s->nc);
  99 }
 100
 101 static ssize_t tap_write_packet(TAPState *s, const struct iovec *iov, int iovcnt)
 102 {
 103     ssize_t len;
 104
 105     do {
 106         len = writev(s->fd, iov, iovcnt);
 107     } while (len == -1 && errno == EINTR);
 108
 109     if (len == -1 && errno == EAGAIN) {
 110         tap_write_poll(s, true);
 111         return 0;
 112     }
 113
 114     return len;
 115 }
 116
 117 static ssize_t tap_receive_iov(NetClientState *nc, const struct iovec *iov,
 118                                int iovcnt)
 119 {
 120     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 121     const struct iovec *iovp = iov;
 122     struct iovec iov_copy[iovcnt + 1];
 123     struct virtio_net_hdr_mrg_rxbuf hdr = { };
 124
 125     if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
 126         iov_copy[0].iov_base = &hdr;
 127         iov_copy[0].iov_len =  s->host_vnet_hdr_len;
 128         memcpy(&iov_copy[1], iov, iovcnt * sizeof(*iov));
 129         iovp = iov_copy;
 130         iovcnt++;
 131     }
 132
 133     return tap_write_packet(s, iovp, iovcnt);
 134 }
 135
 136 static ssize_t tap_receive_raw(NetClientState *nc, const uint8_t *buf, size_t size)
 137 {
 138     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 139     struct iovec iov[2];
 140     int iovcnt = 0;
 141     struct virtio_net_hdr_mrg_rxbuf hdr = { };
 142
 143     if (s->host_vnet_hdr_len) {
 144         iov[iovcnt].iov_base = &hdr;
 145         iov[iovcnt].iov_len  = s->host_vnet_hdr_len;
 146         iovcnt++;
 147     }
 148
 149     iov[iovcnt].iov_base = (char *)buf;
 150     iov[iovcnt].iov_len  = size;
 151     iovcnt++;
 152
 153     return tap_write_packet(s, iov, iovcnt);
 154 }
 155
 156 static ssize_t tap_receive(NetClientState *nc, const uint8_t *buf, size_t size)
 157 {
 158     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 159     struct iovec iov[1];
 160
 161     if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
 162         return tap_receive_raw(nc, buf, size);
 163     }
 164
 165     iov[0].iov_base = (char *)buf;
 166     iov[0].iov_len  = size;
 167
 168     return tap_write_packet(s, iov, 1);
 169 }
 170
 171 #ifndef __sun__
 172 ssize_t tap_read_packet(int tapfd, uint8_t *buf, int maxlen)
 173 {
 174     return read(tapfd, buf, maxlen);
 175 }
 176 #endif
 177
 178 static void tap_send_completed(NetClientState *nc, ssize_t len)
 179 {
 180     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 181     tap_read_poll(s, true);
 182 }
 183
 184 static void tap_send(void *opaque)
 185 {
 186     TAPState *s = opaque;
 187     int size;
 188     int packets = 0;
 189
 190     while (true) {
 191         uint8_t *buf = s->buf;
 192
 193         size = tap_read_packet(s->fd, s->buf, sizeof(s->buf));
 194         if (size <= 0) {
 195             break;
 196         }
 197
 198         if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
 199             buf  += s->host_vnet_hdr_len;
 200             size -= s->host_vnet_hdr_len;
 201         }
 202
 203         size = qemu_send_packet_async(&s->nc, buf, size, tap_send_completed);
 204         if (size == 0) {
 205             tap_read_poll(s, false);
 206             break;
 207         } else if (size < 0) {
 208             break;
 209         }
 210
 211         /*
 212          * When the host keeps receiving more packets while tap_send() is
 213          * running we can hog the QEMU global mutex.  Limit the number of
 214          * packets that are processed per tap_send() callback to prevent
 215          * stalling the guest.
 216          */
 217         packets++;
 218         if (packets >= 50) {
 219             break;
 220         }
 221     }
 222 }
 223
 224 static bool tap_has_ufo(NetClientState *nc)
 225 {
 226     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 227
 228     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 229
 230     return s->has_ufo;
 231 }
 232
 233 static bool tap_has_vnet_hdr(NetClientState *nc)
 234 {
 235     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 236
 237     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 238
 239     return !!s->host_vnet_hdr_len;
 240 }
 241
 242 static bool tap_has_vnet_hdr_len(NetClientState *nc, int len)
 243 {
 244     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 245
 246     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 247
 248     return !!tap_probe_vnet_hdr_len(s->fd, len);
 249 }
 250
 251 static void tap_set_vnet_hdr_len(NetClientState *nc, int len)
 252 {
 253     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 254
 255     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 256     assert(len == sizeof(struct virtio_net_hdr_mrg_rxbuf) ||
 257            len == sizeof(struct virtio_net_hdr) ||
 258            len == sizeof(struct virtio_net_hdr_v1_hash));
 259
 260     tap_fd_set_vnet_hdr_len(s->fd, len);
 261     s->host_vnet_hdr_len = len;
 262 }
 263
 264 static void tap_using_vnet_hdr(NetClientState *nc, bool using_vnet_hdr)
 265 {
 266     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 267
 268     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 269     assert(!!s->host_vnet_hdr_len == using_vnet_hdr);
 270
 271     s->using_vnet_hdr = using_vnet_hdr;
 272 }
 273
 274 static int tap_set_vnet_le(NetClientState *nc, bool is_le)
 275 {
 276     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 277
 278     return tap_fd_set_vnet_le(s->fd, is_le);
 279 }
 280
 281 static int tap_set_vnet_be(NetClientState *nc, bool is_be)
 282 {
 283     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 284
 285     return tap_fd_set_vnet_be(s->fd, is_be);
 286 }
 287
 288 static void tap_set_offload(NetClientState *nc, int csum, int tso4,
 289                      int tso6, int ecn, int ufo)
 290 {
 291     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 292     if (s->fd < 0) {
 293         return;
 294     }
 295
 296     tap_fd_set_offload(s->fd, csum, tso4, tso6, ecn, ufo);
 297 }
 298
 299 static void tap_exit_notify(Notifier *notifier, void *data)
 300 {
 301     TAPState *s = container_of(notifier, TAPState, exit);
 302     Error *err = NULL;
 303
 304     if (s->down_script[0]) {
 305         launch_script(s->down_script, s->down_script_arg, s->fd, &err);
 306         if (err) {
 307             error_report_err(err);
 308         }
 309     }
 310 }
 311
 312 static void tap_cleanup(NetClientState *nc)
 313 {
 314     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 315
 316     if (s->vhost_net) {
 317         vhost_net_cleanup(s->vhost_net);
 318         g_free(s->vhost_net);
 319         s->vhost_net = NULL;
 320     }
 321
 322     qemu_purge_queued_packets(nc);
 323
 324     tap_exit_notify(&s->exit, NULL);
 325     qemu_remove_exit_notifier(&s->exit);
 326
 327     tap_read_poll(s, false);
 328     tap_write_poll(s, false);
 329     close(s->fd);
 330     s->fd = -1;
 331 }
 332
 333 static void tap_poll(NetClientState *nc, bool enable)
 334 {
 335     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 336     tap_read_poll(s, enable);
 337     tap_write_poll(s, enable);
 338 }
 339
 340 int tap_get_fd(NetClientState *nc)
 341 {
 342     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 343     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 344     return s->fd;
 345 }
 346
 347 /* fd support */
 348
 349 static NetClientInfo net_tap_info = {
 350     .type = NET_CLIENT_DRIVER_TAP,
 351     .size = sizeof(TAPState),
 352     .receive = tap_receive,
 353     .receive_raw = tap_receive_raw,
 354     .receive_iov = tap_receive_iov,
 355     .poll = tap_poll,
 356     .cleanup = tap_cleanup,
 357     .has_ufo = tap_has_ufo,
 358     .has_vnet_hdr = tap_has_vnet_hdr,
 359     .has_vnet_hdr_len = tap_has_vnet_hdr_len,
 360     .using_vnet_hdr = tap_using_vnet_hdr,
 361     .set_offload = tap_set_offload,
 362     .set_vnet_hdr_len = tap_set_vnet_hdr_len,
 363     .set_vnet_le = tap_set_vnet_le,
 364     .set_vnet_be = tap_set_vnet_be,
 365 };
 366
 367 static TAPState *net_tap_fd_init(NetClientState *peer,
 368                                  const char *model,
 369                                  const char *name,
 370                                  int fd,
 371                                  int vnet_hdr)
 372 {
 373     NetClientState *nc;
 374     TAPState *s;
 375
 376     nc = qemu_new_net_client(&net_tap_info, peer, model, name);
 377
 378     s = DO_UPCAST(TAPState, nc, nc);
 379
 380     s->fd = fd;
 381     s->host_vnet_hdr_len = vnet_hdr ? sizeof(struct virtio_net_hdr) : 0;
 382     s->using_vnet_hdr = false;
 383     s->has_ufo = tap_probe_has_ufo(s->fd);
 384     s->enabled = true;
 385     tap_set_offload(&s->nc, 0, 0, 0, 0, 0);
 386     /*
 387      * Make sure host header length is set correctly in tap:
 388      * it might have been modified by another instance of qemu.
 389      */
 390     if (tap_probe_vnet_hdr_len(s->fd, s->host_vnet_hdr_len)) {
 391         tap_fd_set_vnet_hdr_len(s->fd, s->host_vnet_hdr_len);
 392     }
 393     tap_read_poll(s, true);
 394     s->vhost_net = NULL;
 395
 396     s->exit.notify = tap_exit_notify;
 397     qemu_add_exit_notifier(&s->exit);
 398
 399     return s;
 400 }
 401
 402 static void launch_script(const char *setup_script, const char *ifname,
 403                           int fd, Error **errp)
 404 {
 405     int pid, status;
 406     char *args[3];
 407     char **parg;
 408
 409     /* try to launch network script */
 410     pid = fork();
 411     if (pid < 0) {
 412         error_setg_errno(errp, errno, "could not launch network script %s",
 413                          setup_script);
 414         return;
 415     }
 416     if (pid == 0) {
 417         int open_max = sysconf(_SC_OPEN_MAX), i;
 418
 419         for (i = 3; i < open_max; i++) {
 420             if (i != fd) {
 421                 close(i);
 422             }
 423         }
 424         parg = args;
 425         *parg++ = (char *)setup_script;
 426         *parg++ = (char *)ifname;
 427         *parg = NULL;
 428         execv(setup_script, args);
 429         _exit(1);
 430     } else {
 431         while (waitpid(pid, &status, 0) != pid) {
 432             /* loop */
 433         }
 434
 435         if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
 436             return;
 437         }
 438         error_setg(errp, "network script %s failed with status %d",
 439                    setup_script, status);
 440     }
 441 }
 442
 443 static int recv_fd(int c)
 444 {
 445     int fd;
 446     uint8_t msgbuf[CMSG_SPACE(sizeof(fd))];
 447     struct msghdr msg = {
 448         .msg_control = msgbuf,
 449         .msg_controllen = sizeof(msgbuf),
 450     };
 451     struct cmsghdr *cmsg;
 452     struct iovec iov;
 453     uint8_t req[1];
 454     ssize_t len;
 455
 456     cmsg = CMSG_FIRSTHDR(&msg);
 457     cmsg->cmsg_level = SOL_SOCKET;
 458     cmsg->cmsg_type = SCM_RIGHTS;
 459     cmsg->cmsg_len = CMSG_LEN(sizeof(fd));
 460     msg.msg_controllen = cmsg->cmsg_len;
 461
 462     iov.iov_base = req;
 463     iov.iov_len = sizeof(req);
 464
 465     msg.msg_iov = &iov;
 466     msg.msg_iovlen = 1;
 467
 468     len = recvmsg(c, &msg, 0);
 469     if (len > 0) {
 470         memcpy(&fd, CMSG_DATA(cmsg), sizeof(fd));
 471         return fd;
 472     }
 473
 474     return len;
 475 }
 476
 477 static int net_bridge_run_helper(const char *helper, const char *bridge,
 478                                  Error **errp)
 479 {
 480     sigset_t oldmask, mask;
 481     g_autofree char *default_helper = NULL;
 482     int pid, status;
 483     char *args[5];
 484     char **parg;
 485     int sv[2];
 486
 487     sigemptyset(&mask);
 488     sigaddset(&mask, SIGCHLD);
 489     sigprocmask(SIG_BLOCK, &mask, &oldmask);
 490
 491     if (!helper) {
 492         helper = default_helper = get_relocated_path(DEFAULT_BRIDGE_HELPER);
 493     }
 494
 495     if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) {
 496         error_setg_errno(errp, errno, "socketpair() failed");
 497         return -1;
 498     }
 499
 500     /* try to launch bridge helper */
 501     pid = fork();
 502     if (pid < 0) {
 503         error_setg_errno(errp, errno, "Can't fork bridge helper");
 504         return -1;
 505     }
 506     if (pid == 0) {
 507         int open_max = sysconf(_SC_OPEN_MAX), i;
 508         char *fd_buf = NULL;
 509         char *br_buf = NULL;
 510         char *helper_cmd = NULL;
 511
 512         for (i = 3; i < open_max; i++) {
 513             if (i != sv[1]) {
 514                 close(i);
 515             }
 516         }
 517
 518         fd_buf = g_strdup_printf("%s%d", "--fd=", sv[1]);
 519
 520         if (strrchr(helper, ' ') || strrchr(helper, '\t')) {
 521             /* assume helper is a command */
 522
 523             if (strstr(helper, "--br=") == NULL) {
 524                 br_buf = g_strdup_printf("%s%s", "--br=", bridge);
 525             }
 526
 527             helper_cmd = g_strdup_printf("%s %s %s %s", helper,
 528                             "--use-vnet", fd_buf, br_buf ? br_buf : "");
 529
 530             parg = args;
 531             *parg++ = (char *)"sh";
 532             *parg++ = (char *)"-c";
 533             *parg++ = helper_cmd;
 534             *parg++ = NULL;
 535
 536             execv("/bin/sh", args);
 537             g_free(helper_cmd);
 538         } else {
 539             /* assume helper is just the executable path name */
 540
 541             br_buf = g_strdup_printf("%s%s", "--br=", bridge);
 542
 543             parg = args;
 544             *parg++ = (char *)helper;
 545             *parg++ = (char *)"--use-vnet";
 546             *parg++ = fd_buf;
 547             *parg++ = br_buf;
 548             *parg++ = NULL;
 549
 550             execv(helper, args);
 551         }
 552         g_free(fd_buf);
 553         g_free(br_buf);
 554         _exit(1);
 555
 556     } else {
 557         int fd;
 558         int saved_errno;
 559
 560         close(sv[1]);
 561
 562         do {
 563             fd = recv_fd(sv[0]);
 564         } while (fd == -1 && errno == EINTR);
 565         saved_errno = errno;
 566
 567         close(sv[0]);
 568
 569         while (waitpid(pid, &status, 0) != pid) {
 570             /* loop */
 571         }
 572         sigprocmask(SIG_SETMASK, &oldmask, NULL);
 573         if (fd < 0) {
 574             error_setg_errno(errp, saved_errno,
 575                              "failed to recv file descriptor");
 576             return -1;
 577         }
 578         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 579             error_setg(errp, "bridge helper failed");
 580             return -1;
 581         }
 582         return fd;
 583     }
 584 }
 585
 586 int net_init_bridge(const Netdev *netdev, const char *name,
 587                     NetClientState *peer, Error **errp)
 588 {
 589     const NetdevBridgeOptions *bridge;
 590     const char *helper, *br;
 591     TAPState *s;
 592     int fd, vnet_hdr;
 593     NetdevBridgeOptions *stored;
 594
 595     assert(netdev->type == NET_CLIENT_DRIVER_BRIDGE);
 596     bridge = &netdev->u.bridge;
 597     helper = bridge->has_helper ? bridge->helper : NULL;
 598     br     = bridge->has_br     ? bridge->br     : DEFAULT_BRIDGE_INTERFACE;
 599
 600     fd = net_bridge_run_helper(helper, br, errp);
 601     if (fd == -1) {
 602         return -1;
 603     }
 604
 605     qemu_set_nonblock(fd);
 606     vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 607     if (vnet_hdr < 0) {
 608         close(fd);
 609         return -1;
 610     }
 611     s = net_tap_fd_init(peer, "bridge", name, fd, vnet_hdr);
 612
 613     /* Store startup parameters */
 614     s->nc.stored_config = g_new0(NetdevInfo, 1);
 615     s->nc.stored_config->type = NET_BACKEND_BRIDGE;
 616     stored = &s->nc.stored_config->u.bridge;
 617
 618     if (br) {
 619         stored->has_br = true;
 620         stored->br = g_strdup(br);
 621     }
 622
 623     if (helper) {
 624         stored->has_helper = true;
 625         stored->helper = g_strdup(helper);
 626     }
 627
 628     snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s,br=%s", helper,
 629              br);
 630
 631     return 0;
 632 }
 633
 634 static int net_tap_init(const NetdevTapOptions *tap, int *vnet_hdr,
 635                         const char *setup_script, char *ifname,
 636                         size_t ifname_sz, int mq_required, Error **errp)
 637 {
 638     Error *err = NULL;
 639     int fd, vnet_hdr_required;
 640
 641     if (tap->has_vnet_hdr) {
 642         *vnet_hdr = tap->vnet_hdr;
 643         vnet_hdr_required = *vnet_hdr;
 644     } else {
 645         *vnet_hdr = 1;
 646         vnet_hdr_required = 0;
 647     }
 648
 649     TFR(fd = tap_open(ifname, ifname_sz, vnet_hdr, vnet_hdr_required,
 650                       mq_required, errp));
 651     if (fd < 0) {
 652         return -1;
 653     }
 654
 655     if (setup_script &&
 656         setup_script[0] != '\0' &&
 657         strcmp(setup_script, "no") != 0) {
 658         launch_script(setup_script, ifname, fd, &err);
 659         if (err) {
 660             error_propagate(errp, err);
 661             close(fd);
 662             return -1;
 663         }
 664     }
 665
 666     return fd;
 667 }
 668
 669 #define MAX_TAP_QUEUES 1024
 670
 671 static void net_init_tap_one(const NetdevTapOptions *tap, NetClientState *peer,
 672                              const char *model, const char *name,
 673                              const char *ifname, const char *script,
 674                              const char *downscript, const char *vhostfdname,
 675                              int vnet_hdr, int fd, NetdevInfo **common_stored,
 676                              Error **errp)
 677 {
 678     Error *err = NULL;
 679     TAPState *s = net_tap_fd_init(peer, model, name, fd, vnet_hdr);
 680     int vhostfd;
 681     NetdevTapOptions *stored;
 682
 683     tap_set_sndbuf(s->fd, tap, &err);
 684     if (err) {
 685         error_propagate(errp, err);
 686         return;
 687     }
 688
 689     /* Store startup parameters */
 690     if (!*common_stored) {
 691         *common_stored = g_new0(NetdevInfo, 1);
 692         (*common_stored)->type = NET_BACKEND_TAP;
 693         s->nc.stored_config = *common_stored;
 694     }
 695     stored = &(*common_stored)->u.tap;
 696
 697     if (tap->has_sndbuf && !stored->has_sndbuf) {
 698         stored->has_sndbuf = true;
 699         stored->sndbuf = tap->sndbuf;
 700     }
 701
 702     if (vnet_hdr && !stored->has_vnet_hdr) {
 703         stored->has_vnet_hdr = true;
 704         stored->vnet_hdr = true;
 705     }
 706
 707     if (tap->has_fd || tap->has_fds) {
 708         if (!stored->has_fds) {
 709             stored->has_fds = true;
 710             stored->fds = g_strdup_printf("%d", fd);
 711         } else {
 712             char *tmp_s = stored->fds;
 713             stored->fds = g_strdup_printf("%s:%d", stored->fds, fd);
 714             g_free(tmp_s);
 715         }
 716
 717         snprintf(s->nc.info_str, sizeof(s->nc.info_str), "fd=%d", fd);
 718     } else if (tap->has_helper) {
 719         if (!stored->has_helper) {
 720             stored->has_helper = true;
 721             stored->helper = g_strdup(tap->helper);
 722         }
 723
 724         if (!stored->has_br) {
 725             stored->has_br = true;
 726             stored->br = tap->has_br ? g_strdup(tap->br) :
 727                                        g_strdup(DEFAULT_BRIDGE_INTERFACE);
 728         }
 729
 730         snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s",
 731                  tap->helper);
 732     } else {
 733         if (ifname && !stored->has_ifname) {
 734             stored->has_ifname = true;
 735             stored->ifname = g_strdup(ifname);
 736         }
 737
 738         if (script && !stored->has_script) {
 739             stored->has_script = true;
 740             stored->script = g_strdup(script);
 741         }
 742
 743         if (downscript && !stored->has_downscript) {
 744             stored->has_downscript = true;
 745             stored->downscript = g_strdup(downscript);
 746         }
 747
 748         snprintf(s->nc.info_str, sizeof(s->nc.info_str),
 749                  "ifname=%s,script=%s,downscript=%s", ifname, script,
 750                  downscript);
 751
 752         if (strcmp(downscript, "no") != 0) {
 753             snprintf(s->down_script, sizeof(s->down_script), "%s", downscript);
 754             snprintf(s->down_script_arg, sizeof(s->down_script_arg),
 755                      "%s", ifname);
 756         }
 757     }
 758
 759     if (tap->has_vhost ? tap->vhost :
 760         vhostfdname || (tap->has_vhostforce && tap->vhostforce)) {
 761         VhostNetOptions options;
 762
 763         stored->has_vhost = true;
 764         stored->vhost = true;
 765
 766         if (tap->has_vhostforce && tap->vhostforce) {
 767             stored->has_vhostforce = true;
 768             stored->vhostforce = true;
 769         }
 770
 771         options.backend_type = VHOST_BACKEND_TYPE_KERNEL;
 772         options.net_backend = &s->nc;
 773         if (tap->has_poll_us) {
 774             stored->has_poll_us = true;
 775             stored->poll_us = tap->poll_us;
 776
 777             options.busyloop_timeout = tap->poll_us;
 778         } else {
 779             options.busyloop_timeout = 0;
 780         }
 781
 782         if (vhostfdname) {
 783             int ret;
 784
 785             vhostfd = monitor_fd_param(monitor_cur(), vhostfdname, &err);
 786             if (vhostfd == -1) {
 787                 if (tap->has_vhostforce && tap->vhostforce) {
 788                     error_propagate(errp, err);
 789                 } else {
 790                     warn_report_err(err);
 791                 }
 792                 return;
 793             }
 794             ret = qemu_try_set_nonblock(vhostfd);
 795             if (ret < 0) {
 796                 error_setg_errno(errp, -ret, "%s: Can't use file descriptor %d",
 797                                  name, fd);
 798                 return;
 799             }
 800         } else {
 801             vhostfd = open("/dev/vhost-net", O_RDWR);
 802             if (vhostfd < 0) {
 803                 if (tap->has_vhostforce && tap->vhostforce) {
 804                     error_setg_errno(errp, errno,
 805                                      "tap: open vhost char device failed");
 806                 } else {
 807                     warn_report("tap: open vhost char device failed: %s",
 808                                 strerror(errno));
 809                 }
 810                 return;
 811             }
 812             qemu_set_nonblock(vhostfd);
 813         }
 814         options.opaque = (void *)(uintptr_t)vhostfd;
 815
 816         if (!stored->has_vhostfds) {
 817             stored->has_vhostfds = true;
 818             stored->vhostfds = g_strdup_printf("%d", vhostfd);
 819         } else {
 820             char *tmp_s = stored->vhostfds;
 821             stored->vhostfds = g_strdup_printf("%s:%d", stored->fds, vhostfd);
 822             g_free(tmp_s);
 823         }
 824
 825         s->vhost_net = vhost_net_init(&options);
 826         if (!s->vhost_net) {
 827             if (tap->has_vhostforce && tap->vhostforce) {
 828                 error_setg(errp, VHOST_NET_INIT_FAILED);
 829             } else {
 830                 warn_report(VHOST_NET_INIT_FAILED);
 831             }
 832             return;
 833         }
 834     } else if (vhostfdname) {
 835         error_setg(errp, "vhostfd(s)= is not valid without vhost");
 836     }
 837 }
 838
 839 static int get_fds(char *str, char *fds[], int max)
 840 {
 841     char *ptr = str, *this;
 842     size_t len = strlen(str);
 843     int i = 0;
 844
 845     while (i < max && ptr < str + len) {
 846         this = strchr(ptr, ':');
 847
 848         if (this == NULL) {
 849             fds[i] = g_strdup(ptr);
 850         } else {
 851             fds[i] = g_strndup(ptr, this - ptr);
 852         }
 853
 854         i++;
 855         if (this == NULL) {
 856             break;
 857         } else {
 858             ptr = this + 1;
 859         }
 860     }
 861
 862     return i;
 863 }
 864
 865 int net_init_tap(const Netdev *netdev, const char *name,
 866                  NetClientState *peer, Error **errp)
 867 {
 868     const NetdevTapOptions *tap;
 869     int fd, vnet_hdr = 0, i = 0, queues;
 870     /* for the no-fd, no-helper case */
 871     const char *script;
 872     const char *downscript;
 873     Error *err = NULL;
 874     const char *vhostfdname;
 875     char ifname[128];
 876     int ret = 0;
 877     NetdevInfo *common_stored = NULL; /* will store configuration */
 878
 879     assert(netdev->type == NET_CLIENT_DRIVER_TAP);
 880     tap = &netdev->u.tap;
 881     queues = tap->has_queues ? tap->queues : 1;
 882     vhostfdname = tap->has_vhostfd ? tap->vhostfd : NULL;
 883     script = tap->has_script ? tap->script : NULL;
 884     downscript = tap->has_downscript ? tap->downscript : NULL;
 885
 886     /* QEMU hubs do not support multiqueue tap, in this case peer is set.
 887      * For -netdev, peer is always NULL. */
 888     if (peer && (tap->has_queues || tap->has_fds || tap->has_vhostfds)) {
 889         error_setg(errp, "Multiqueue tap cannot be used with hubs");
 890         return -1;
 891     }
 892
 893     if (tap->has_fd) {
 894         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
 895             tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
 896             tap->has_fds || tap->has_vhostfds) {
 897             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 898                        "helper=, queues=, fds=, and vhostfds= "
 899                        "are invalid with fd=");
 900             return -1;
 901         }
 902
 903         fd = monitor_fd_param(monitor_cur(), tap->fd, errp);
 904         if (fd == -1) {
 905             return -1;
 906         }
 907
 908         ret = qemu_try_set_nonblock(fd);
 909         if (ret < 0) {
 910             error_setg_errno(errp, -ret, "%s: Can't use file descriptor %d",
 911                              name, fd);
 912             close(fd);
 913             return -1;
 914         }
 915
 916         vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 917         if (vnet_hdr < 0) {
 918             close(fd);
 919             return -1;
 920         }
 921
 922         net_init_tap_one(tap, peer, "tap", name, NULL,
 923                          script, downscript,
 924                          vhostfdname, vnet_hdr, fd, &common_stored, &err);
 925         if (err) {
 926             error_propagate(errp, err);
 927             close(fd);
 928             return -1;
 929         }
 930     } else if (tap->has_fds) {
 931         char **fds;
 932         char **vhost_fds;
 933         int nfds = 0, nvhosts = 0;
 934
 935         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
 936             tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
 937             tap->has_vhostfd) {
 938             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 939                        "helper=, queues=, and vhostfd= "
 940                        "are invalid with fds=");
 941             return -1;
 942         }
 943
 944         fds = g_new0(char *, MAX_TAP_QUEUES);
 945         vhost_fds = g_new0(char *, MAX_TAP_QUEUES);
 946
 947         nfds = get_fds(tap->fds, fds, MAX_TAP_QUEUES);
 948         if (tap->has_vhostfds) {
 949             nvhosts = get_fds(tap->vhostfds, vhost_fds, MAX_TAP_QUEUES);
 950             if (nfds != nvhosts) {
 951                 error_setg(errp, "The number of fds passed does not match "
 952                            "the number of vhostfds passed");
 953                 ret = -1;
 954                 goto free_fail;
 955             }
 956         }
 957
 958         for (i = 0; i < nfds; i++) {
 959             fd = monitor_fd_param(monitor_cur(), fds[i], errp);
 960             if (fd == -1) {
 961                 ret = -1;
 962                 goto free_fail;
 963             }
 964
 965             ret = qemu_try_set_nonblock(fd);
 966             if (ret < 0) {
 967                 error_setg_errno(errp, -ret, "%s: Can't use file descriptor %d",
 968                                  name, fd);
 969                 goto free_fail;
 970             }
 971
 972             if (i == 0) {
 973                 vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 974                 if (vnet_hdr < 0) {
 975                     goto free_fail;
 976                 }
 977             } else if (vnet_hdr != tap_probe_vnet_hdr(fd, NULL)) {
 978                 error_setg(errp,
 979                            "vnet_hdr not consistent across given tap fds");
 980                 ret = -1;
 981                 goto free_fail;
 982             }
 983
 984             net_init_tap_one(tap, peer, "tap", name, ifname,
 985                              script, downscript,
 986                              tap->has_vhostfds ? vhost_fds[i] : NULL,
 987                              vnet_hdr, fd, &common_stored, &err);
 988             if (err) {
 989                 error_propagate(errp, err);
 990                 ret = -1;
 991                 goto free_fail;
 992             }
 993         }
 994
 995 free_fail:
 996         for (i = 0; i < nvhosts; i++) {
 997             g_free(vhost_fds[i]);
 998         }
 999         for (i = 0; i < nfds; i++) {
1000             g_free(fds[i]);
1001         }
1002         g_free(fds);
1003         g_free(vhost_fds);
1004         return ret;
1005     } else if (tap->has_helper) {
1006         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
1007             tap->has_vnet_hdr || tap->has_queues || tap->has_vhostfds) {
1008             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
1009                        "queues=, and vhostfds= are invalid with helper=");
1010             return -1;
1011         }
1012
1013         fd = net_bridge_run_helper(tap->helper,
1014                                    tap->has_br ?
1015                                    tap->br : DEFAULT_BRIDGE_INTERFACE,
1016                                    errp);
1017         if (fd == -1) {
1018             return -1;
1019         }
1020
1021         qemu_set_nonblock(fd);
1022         vnet_hdr = tap_probe_vnet_hdr(fd, errp);
1023         if (vnet_hdr < 0) {
1024             close(fd);
1025             return -1;
1026         }
1027
1028         net_init_tap_one(tap, peer, "bridge", name, ifname,
1029                          script, downscript, vhostfdname,
1030                          vnet_hdr, fd, &common_stored, &err);
1031         if (err) {
1032             error_propagate(errp, err);
1033             close(fd);
1034             return -1;
1035         }
1036     } else {
1037         g_autofree char *default_script = NULL;
1038         g_autofree char *default_downscript = NULL;
1039         if (tap->has_vhostfds) {
1040             error_setg(errp, "vhostfds= is invalid if fds= wasn't specified");
1041             return -1;
1042         }
1043
1044         if (!script) {
1045             script = default_script = get_relocated_path(DEFAULT_NETWORK_SCRIPT);
1046         }
1047         if (!downscript) {
1048             downscript = default_downscript =
1049                                  get_relocated_path(DEFAULT_NETWORK_DOWN_SCRIPT);
1050         }
1051
1052         if (tap->has_ifname) {
1053             pstrcpy(ifname, sizeof ifname, tap->ifname);
1054         } else {
1055             ifname[0] = '\0';
1056         }
1057
1058         for (i = 0; i < queues; i++) {
1059             fd = net_tap_init(tap, &vnet_hdr, i >= 1 ? "no" : script,
1060                               ifname, sizeof ifname, queues > 1, errp);
1061             if (fd == -1) {
1062                 return -1;
1063             }
1064
1065             if (queues > 1 && i == 0 && !tap->has_ifname) {
1066                 if (tap_fd_get_ifname(fd, ifname)) {
1067                     error_setg(errp, "Fail to get ifname");
1068                     close(fd);
1069                     return -1;
1070                 }
1071             }
1072
1073             net_init_tap_one(tap, peer, "tap", name, ifname,
1074                              i >= 1 ? "no" : script,
1075                              i >= 1 ? "no" : downscript,
1076                              vhostfdname, vnet_hdr, fd,
1077                              &common_stored, &err);
1078             if (err) {
1079                 error_propagate(errp, err);
1080                 close(fd);
1081                 return -1;
1082             }
1083         }
1084     }
1085
1086     return 0;
1087 }
1088
1089 VHostNetState *tap_get_vhost_net(NetClientState *nc)
1090 {
1091     TAPState *s = DO_UPCAST(TAPState, nc, nc);
1092     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
1093     return s->vhost_net;
1094 }
1095
1096 int tap_enable(NetClientState *nc)
1097 {
1098     TAPState *s = DO_UPCAST(TAPState, nc, nc);
1099     int ret;
1100
1101     if (s->enabled) {
1102         return 0;
1103     } else {
1104         ret = tap_fd_enable(s->fd);
1105         if (ret == 0) {
1106             s->enabled = true;
1107             tap_update_fd_handler(s);
1108         }
1109         return ret;
1110     }
1111 }
1112
1113 int tap_disable(NetClientState *nc)
1114 {
1115     TAPState *s = DO_UPCAST(TAPState, nc, nc);
1116     int ret;
1117
1118     if (s->enabled == 0) {
1119         return 0;
1120     } else {
1121         ret = tap_fd_disable(s->fd);
1122         if (ret == 0) {
1123             qemu_purge_queued_packets(nc);
1124             s->enabled = false;
1125             tap_update_fd_handler(s);
1126         }
1127         return ret;
1128     }
1129 }