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
 594     assert(netdev->type == NET_CLIENT_DRIVER_BRIDGE);
 595     bridge = &netdev->u.bridge;
 596     helper = bridge->has_helper ? bridge->helper : NULL;
 597     br     = bridge->has_br     ? bridge->br     : DEFAULT_BRIDGE_INTERFACE;
 598
 599     fd = net_bridge_run_helper(helper, br, errp);
 600     if (fd == -1) {
 601         return -1;
 602     }
 603
 604     qemu_set_nonblock(fd);
 605     vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 606     if (vnet_hdr < 0) {
 607         close(fd);
 608         return -1;
 609     }
 610     s = net_tap_fd_init(peer, "bridge", name, fd, vnet_hdr);
 611
 612     snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s,br=%s", helper,
 613              br);
 614
 615     return 0;
 616 }
 617
 618 static int net_tap_init(const NetdevTapOptions *tap, int *vnet_hdr,
 619                         const char *setup_script, char *ifname,
 620                         size_t ifname_sz, int mq_required, Error **errp)
 621 {
 622     Error *err = NULL;
 623     int fd, vnet_hdr_required;
 624
 625     if (tap->has_vnet_hdr) {
 626         *vnet_hdr = tap->vnet_hdr;
 627         vnet_hdr_required = *vnet_hdr;
 628     } else {
 629         *vnet_hdr = 1;
 630         vnet_hdr_required = 0;
 631     }
 632
 633     TFR(fd = tap_open(ifname, ifname_sz, vnet_hdr, vnet_hdr_required,
 634                       mq_required, errp));
 635     if (fd < 0) {
 636         return -1;
 637     }
 638
 639     if (setup_script &&
 640         setup_script[0] != '\0' &&
 641         strcmp(setup_script, "no") != 0) {
 642         launch_script(setup_script, ifname, fd, &err);
 643         if (err) {
 644             error_propagate(errp, err);
 645             close(fd);
 646             return -1;
 647         }
 648     }
 649
 650     return fd;
 651 }
 652
 653 #define MAX_TAP_QUEUES 1024
 654
 655 static void net_init_tap_one(const NetdevTapOptions *tap, NetClientState *peer,
 656                              const char *model, const char *name,
 657                              const char *ifname, const char *script,
 658                              const char *downscript, const char *vhostfdname,
 659                              int vnet_hdr, int fd, Error **errp)
 660 {
 661     Error *err = NULL;
 662     TAPState *s = net_tap_fd_init(peer, model, name, fd, vnet_hdr);
 663     int vhostfd;
 664
 665     tap_set_sndbuf(s->fd, tap, &err);
 666     if (err) {
 667         error_propagate(errp, err);
 668         return;
 669     }
 670
 671     if (tap->has_fd || tap->has_fds) {
 672         snprintf(s->nc.info_str, sizeof(s->nc.info_str), "fd=%d", fd);
 673     } else if (tap->has_helper) {
 674         snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s",
 675                  tap->helper);
 676     } else {
 677         snprintf(s->nc.info_str, sizeof(s->nc.info_str),
 678                  "ifname=%s,script=%s,downscript=%s", ifname, script,
 679                  downscript);
 680
 681         if (strcmp(downscript, "no") != 0) {
 682             snprintf(s->down_script, sizeof(s->down_script), "%s", downscript);
 683             snprintf(s->down_script_arg, sizeof(s->down_script_arg),
 684                      "%s", ifname);
 685         }
 686     }
 687
 688     if (tap->has_vhost ? tap->vhost :
 689         vhostfdname || (tap->has_vhostforce && tap->vhostforce)) {
 690         VhostNetOptions options;
 691
 692         options.backend_type = VHOST_BACKEND_TYPE_KERNEL;
 693         options.net_backend = &s->nc;
 694         if (tap->has_poll_us) {
 695             options.busyloop_timeout = tap->poll_us;
 696         } else {
 697             options.busyloop_timeout = 0;
 698         }
 699
 700         if (vhostfdname) {
 701             int ret;
 702
 703             vhostfd = monitor_fd_param(monitor_cur(), vhostfdname, &err);
 704             if (vhostfd == -1) {
 705                 if (tap->has_vhostforce && tap->vhostforce) {
 706                     error_propagate(errp, err);
 707                 } else {
 708                     warn_report_err(err);
 709                 }
 710                 return;
 711             }
 712             ret = qemu_try_set_nonblock(vhostfd);
 713             if (ret < 0) {
 714                 error_setg_errno(errp, -ret, "%s: Can't use file descriptor %d",
 715                                  name, fd);
 716                 return;
 717             }
 718         } else {
 719             vhostfd = open("/dev/vhost-net", O_RDWR);
 720             if (vhostfd < 0) {
 721                 if (tap->has_vhostforce && tap->vhostforce) {
 722                     error_setg_errno(errp, errno,
 723                                      "tap: open vhost char device failed");
 724                 } else {
 725                     warn_report("tap: open vhost char device failed: %s",
 726                                 strerror(errno));
 727                 }
 728                 return;
 729             }
 730             qemu_set_nonblock(vhostfd);
 731         }
 732         options.opaque = (void *)(uintptr_t)vhostfd;
 733
 734         s->vhost_net = vhost_net_init(&options);
 735         if (!s->vhost_net) {
 736             if (tap->has_vhostforce && tap->vhostforce) {
 737                 error_setg(errp, VHOST_NET_INIT_FAILED);
 738             } else {
 739                 warn_report(VHOST_NET_INIT_FAILED);
 740             }
 741             return;
 742         }
 743     } else if (vhostfdname) {
 744         error_setg(errp, "vhostfd(s)= is not valid without vhost");
 745     }
 746 }
 747
 748 static int get_fds(char *str, char *fds[], int max)
 749 {
 750     char *ptr = str, *this;
 751     size_t len = strlen(str);
 752     int i = 0;
 753
 754     while (i < max && ptr < str + len) {
 755         this = strchr(ptr, ':');
 756
 757         if (this == NULL) {
 758             fds[i] = g_strdup(ptr);
 759         } else {
 760             fds[i] = g_strndup(ptr, this - ptr);
 761         }
 762
 763         i++;
 764         if (this == NULL) {
 765             break;
 766         } else {
 767             ptr = this + 1;
 768         }
 769     }
 770
 771     return i;
 772 }
 773
 774 int net_init_tap(const Netdev *netdev, const char *name,
 775                  NetClientState *peer, Error **errp)
 776 {
 777     const NetdevTapOptions *tap;
 778     int fd, vnet_hdr = 0, i = 0, queues;
 779     /* for the no-fd, no-helper case */
 780     const char *script;
 781     const char *downscript;
 782     Error *err = NULL;
 783     const char *vhostfdname;
 784     char ifname[128];
 785     int ret = 0;
 786
 787     assert(netdev->type == NET_CLIENT_DRIVER_TAP);
 788     tap = &netdev->u.tap;
 789     queues = tap->has_queues ? tap->queues : 1;
 790     vhostfdname = tap->has_vhostfd ? tap->vhostfd : NULL;
 791     script = tap->has_script ? tap->script : NULL;
 792     downscript = tap->has_downscript ? tap->downscript : NULL;
 793
 794     /* QEMU hubs do not support multiqueue tap, in this case peer is set.
 795      * For -netdev, peer is always NULL. */
 796     if (peer && (tap->has_queues || tap->has_fds || tap->has_vhostfds)) {
 797         error_setg(errp, "Multiqueue tap cannot be used with hubs");
 798         return -1;
 799     }
 800
 801     if (tap->has_fd) {
 802         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
 803             tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
 804             tap->has_fds || tap->has_vhostfds) {
 805             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 806                        "helper=, queues=, fds=, and vhostfds= "
 807                        "are invalid with fd=");
 808             return -1;
 809         }
 810
 811         fd = monitor_fd_param(monitor_cur(), tap->fd, errp);
 812         if (fd == -1) {
 813             return -1;
 814         }
 815
 816         ret = qemu_try_set_nonblock(fd);
 817         if (ret < 0) {
 818             error_setg_errno(errp, -ret, "%s: Can't use file descriptor %d",
 819                              name, fd);
 820             close(fd);
 821             return -1;
 822         }
 823
 824         vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 825         if (vnet_hdr < 0) {
 826             close(fd);
 827             return -1;
 828         }
 829
 830         net_init_tap_one(tap, peer, "tap", name, NULL,
 831                          script, downscript,
 832                          vhostfdname, vnet_hdr, fd, &err);
 833         if (err) {
 834             error_propagate(errp, err);
 835             close(fd);
 836             return -1;
 837         }
 838     } else if (tap->has_fds) {
 839         char **fds;
 840         char **vhost_fds;
 841         int nfds = 0, nvhosts = 0;
 842
 843         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
 844             tap->has_vnet_hdr || tap->has_helper || tap->has_queues ||
 845             tap->has_vhostfd) {
 846             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 847                        "helper=, queues=, and vhostfd= "
 848                        "are invalid with fds=");
 849             return -1;
 850         }
 851
 852         fds = g_new0(char *, MAX_TAP_QUEUES);
 853         vhost_fds = g_new0(char *, MAX_TAP_QUEUES);
 854
 855         nfds = get_fds(tap->fds, fds, MAX_TAP_QUEUES);
 856         if (tap->has_vhostfds) {
 857             nvhosts = get_fds(tap->vhostfds, vhost_fds, MAX_TAP_QUEUES);
 858             if (nfds != nvhosts) {
 859                 error_setg(errp, "The number of fds passed does not match "
 860                            "the number of vhostfds passed");
 861                 ret = -1;
 862                 goto free_fail;
 863             }
 864         }
 865
 866         for (i = 0; i < nfds; i++) {
 867             fd = monitor_fd_param(monitor_cur(), fds[i], errp);
 868             if (fd == -1) {
 869                 ret = -1;
 870                 goto free_fail;
 871             }
 872
 873             ret = qemu_try_set_nonblock(fd);
 874             if (ret < 0) {
 875                 error_setg_errno(errp, -ret, "%s: Can't use file descriptor %d",
 876                                  name, fd);
 877                 goto free_fail;
 878             }
 879
 880             if (i == 0) {
 881                 vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 882                 if (vnet_hdr < 0) {
 883                     goto free_fail;
 884                 }
 885             } else if (vnet_hdr != tap_probe_vnet_hdr(fd, NULL)) {
 886                 error_setg(errp,
 887                            "vnet_hdr not consistent across given tap fds");
 888                 ret = -1;
 889                 goto free_fail;
 890             }
 891
 892             net_init_tap_one(tap, peer, "tap", name, ifname,
 893                              script, downscript,
 894                              tap->has_vhostfds ? vhost_fds[i] : NULL,
 895                              vnet_hdr, fd, &err);
 896             if (err) {
 897                 error_propagate(errp, err);
 898                 ret = -1;
 899                 goto free_fail;
 900             }
 901         }
 902
 903 free_fail:
 904         for (i = 0; i < nvhosts; i++) {
 905             g_free(vhost_fds[i]);
 906         }
 907         for (i = 0; i < nfds; i++) {
 908             g_free(fds[i]);
 909         }
 910         g_free(fds);
 911         g_free(vhost_fds);
 912         return ret;
 913     } else if (tap->has_helper) {
 914         if (tap->has_ifname || tap->has_script || tap->has_downscript ||
 915             tap->has_vnet_hdr || tap->has_queues || tap->has_vhostfds) {
 916             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 917                        "queues=, and vhostfds= are invalid with helper=");
 918             return -1;
 919         }
 920
 921         fd = net_bridge_run_helper(tap->helper,
 922                                    tap->has_br ?
 923                                    tap->br : DEFAULT_BRIDGE_INTERFACE,
 924                                    errp);
 925         if (fd == -1) {
 926             return -1;
 927         }
 928
 929         qemu_set_nonblock(fd);
 930         vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 931         if (vnet_hdr < 0) {
 932             close(fd);
 933             return -1;
 934         }
 935
 936         net_init_tap_one(tap, peer, "bridge", name, ifname,
 937                          script, downscript, vhostfdname,
 938                          vnet_hdr, fd, &err);
 939         if (err) {
 940             error_propagate(errp, err);
 941             close(fd);
 942             return -1;
 943         }
 944     } else {
 945         g_autofree char *default_script = NULL;
 946         g_autofree char *default_downscript = NULL;
 947         if (tap->has_vhostfds) {
 948             error_setg(errp, "vhostfds= is invalid if fds= wasn't specified");
 949             return -1;
 950         }
 951
 952         if (!script) {
 953             script = default_script = get_relocated_path(DEFAULT_NETWORK_SCRIPT);
 954         }
 955         if (!downscript) {
 956             downscript = default_downscript =
 957                                  get_relocated_path(DEFAULT_NETWORK_DOWN_SCRIPT);
 958         }
 959
 960         if (tap->has_ifname) {
 961             pstrcpy(ifname, sizeof ifname, tap->ifname);
 962         } else {
 963             ifname[0] = '\0';
 964         }
 965
 966         for (i = 0; i < queues; i++) {
 967             fd = net_tap_init(tap, &vnet_hdr, i >= 1 ? "no" : script,
 968                               ifname, sizeof ifname, queues > 1, errp);
 969             if (fd == -1) {
 970                 return -1;
 971             }
 972
 973             if (queues > 1 && i == 0 && !tap->has_ifname) {
 974                 if (tap_fd_get_ifname(fd, ifname)) {
 975                     error_setg(errp, "Fail to get ifname");
 976                     close(fd);
 977                     return -1;
 978                 }
 979             }
 980
 981             net_init_tap_one(tap, peer, "tap", name, ifname,
 982                              i >= 1 ? "no" : script,
 983                              i >= 1 ? "no" : downscript,
 984                              vhostfdname, vnet_hdr, fd, &err);
 985             if (err) {
 986                 error_propagate(errp, err);
 987                 close(fd);
 988                 return -1;
 989             }
 990         }
 991     }
 992
 993     return 0;
 994 }
 995
 996 VHostNetState *tap_get_vhost_net(NetClientState *nc)
 997 {
 998     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 999     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
1000     return s->vhost_net;
1001 }
1002
1003 int tap_enable(NetClientState *nc)
1004 {
1005     TAPState *s = DO_UPCAST(TAPState, nc, nc);
1006     int ret;
1007
1008     if (s->enabled) {
1009         return 0;
1010     } else {
1011         ret = tap_fd_enable(s->fd);
1012         if (ret == 0) {
1013             s->enabled = true;
1014             tap_update_fd_handler(s);
1015         }
1016         return ret;
1017     }
1018 }
1019
1020 int tap_disable(NetClientState *nc)
1021 {
1022     TAPState *s = DO_UPCAST(TAPState, nc, nc);
1023     int ret;
1024
1025     if (s->enabled == 0) {
1026         return 0;
1027     } else {
1028         ret = tap_fd_disable(s->fd);
1029         if (ret == 0) {
1030             qemu_purge_queued_packets(nc);
1031             s->enabled = false;
1032             tap_update_fd_handler(s);
1033         }
1034         return ret;
1035     }
1036 }