net/tap.c

   1 /*
   2  * QEMU System Emulator
   3  *
   4  * Copyright (c) 2003-2008 Fabrice Bellard
   5  * Copyright (c) 2009 Red Hat, Inc.
   6  *
   7  * Permission is hereby granted, free of charge, to any person obtaining a copy
   8  * of this software and associated documentation files (the "Software"), to deal
   9  * in the Software without restriction, including without limitation the rights
  10  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  11  * copies of the Software, and to permit persons to whom the Software is
  12  * furnished to do so, subject to the following conditions:
  13  *
  14  * The above copyright notice and this permission notice shall be included in
  15  * all copies or substantial portions of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  23  * THE SOFTWARE.
  24  */
  25
  26 #include "qemu/osdep.h"
  27 #include "tap_int.h"
  28
  29
  30 #include <sys/ioctl.h>
  31 #include <sys/wait.h>
  32 #include <sys/socket.h>
  33 #include <net/if.h>
  34
  35 #include "net/eth.h"
  36 #include "net/net.h"
  37 #include "clients.h"
  38 #include "monitor/monitor.h"
  39 #include "sysemu/sysemu.h"
  40 #include "qapi/error.h"
  41 #include "qemu/cutils.h"
  42 #include "qemu/error-report.h"
  43 #include "qemu/main-loop.h"
  44 #include "qemu/sockets.h"
  45
  46 #include "net/tap.h"
  47
  48 #include "net/vhost_net.h"
  49
  50 typedef struct TAPState {
  51     NetClientState nc;
  52     int fd;
  53     char down_script[1024];
  54     char down_script_arg[128];
  55     uint8_t buf[NET_BUFSIZE];
  56     bool read_poll;
  57     bool write_poll;
  58     bool using_vnet_hdr;
  59     bool has_ufo;
  60     bool has_uso;
  61     bool enabled;
  62     VHostNetState *vhost_net;
  63     unsigned host_vnet_hdr_len;
  64     Notifier exit;
  65 } TAPState;
  66
  67 static void launch_script(const char *setup_script, const char *ifname,
  68                           int fd, Error **errp);
  69
  70 static void tap_send(void *opaque);
  71 static void tap_writable(void *opaque);
  72
  73 static void tap_update_fd_handler(TAPState *s)
  74 {
  75     qemu_set_fd_handler(s->fd,
  76                         s->read_poll && s->enabled ? tap_send : NULL,
  77                         s->write_poll && s->enabled ? tap_writable : NULL,
  78                         s);
  79 }
  80
  81 static void tap_read_poll(TAPState *s, bool enable)
  82 {
  83     s->read_poll = enable;
  84     tap_update_fd_handler(s);
  85 }
  86
  87 static void tap_write_poll(TAPState *s, bool enable)
  88 {
  89     s->write_poll = enable;
  90     tap_update_fd_handler(s);
  91 }
  92
  93 static void tap_writable(void *opaque)
  94 {
  95     TAPState *s = opaque;
  96
  97     tap_write_poll(s, false);
  98
  99     qemu_flush_queued_packets(&s->nc);
 100 }
 101
 102 static ssize_t tap_write_packet(TAPState *s, const struct iovec *iov, int iovcnt)
 103 {
 104     ssize_t len;
 105
 106     len = RETRY_ON_EINTR(writev(s->fd, iov, iovcnt));
 107
 108     if (len == -1 && errno == EAGAIN) {
 109         tap_write_poll(s, true);
 110         return 0;
 111     }
 112
 113     return len;
 114 }
 115
 116 static ssize_t tap_receive_iov(NetClientState *nc, const struct iovec *iov,
 117                                int iovcnt)
 118 {
 119     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 120     const struct iovec *iovp = iov;
 121     g_autofree struct iovec *iov_copy = NULL;
 122     struct virtio_net_hdr hdr = { };
 123
 124     if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
 125         iov_copy = g_new(struct iovec, iovcnt + 1);
 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(NetClientState *nc, const uint8_t *buf, size_t size)
 137 {
 138     struct iovec iov = {
 139         .iov_base = (void *)buf,
 140         .iov_len = size
 141     };
 142
 143     return tap_receive_iov(nc, &iov, 1);
 144 }
 145
 146 #ifndef __sun__
 147 ssize_t tap_read_packet(int tapfd, uint8_t *buf, int maxlen)
 148 {
 149     return read(tapfd, buf, maxlen);
 150 }
 151 #endif
 152
 153 static void tap_send_completed(NetClientState *nc, ssize_t len)
 154 {
 155     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 156     tap_read_poll(s, true);
 157 }
 158
 159 static void tap_send(void *opaque)
 160 {
 161     TAPState *s = opaque;
 162     int size;
 163     int packets = 0;
 164
 165     while (true) {
 166         uint8_t *buf = s->buf;
 167         uint8_t min_pkt[ETH_ZLEN];
 168         size_t min_pktsz = sizeof(min_pkt);
 169
 170         size = tap_read_packet(s->fd, s->buf, sizeof(s->buf));
 171         if (size <= 0) {
 172             break;
 173         }
 174
 175         if (s->host_vnet_hdr_len && !s->using_vnet_hdr) {
 176             buf  += s->host_vnet_hdr_len;
 177             size -= s->host_vnet_hdr_len;
 178         }
 179
 180         if (net_peer_needs_padding(&s->nc)) {
 181             if (eth_pad_short_frame(min_pkt, &min_pktsz, buf, size)) {
 182                 buf = min_pkt;
 183                 size = min_pktsz;
 184             }
 185         }
 186
 187         size = qemu_send_packet_async(&s->nc, buf, size, tap_send_completed);
 188         if (size == 0) {
 189             tap_read_poll(s, false);
 190             break;
 191         } else if (size < 0) {
 192             break;
 193         }
 194
 195         /*
 196          * When the host keeps receiving more packets while tap_send() is
 197          * running we can hog the BQL.  Limit the number of
 198          * packets that are processed per tap_send() callback to prevent
 199          * stalling the guest.
 200          */
 201         packets++;
 202         if (packets >= 50) {
 203             break;
 204         }
 205     }
 206 }
 207
 208 static bool tap_has_ufo(NetClientState *nc)
 209 {
 210     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 211
 212     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 213
 214     return s->has_ufo;
 215 }
 216
 217 static bool tap_has_uso(NetClientState *nc)
 218 {
 219     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 220
 221     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 222
 223     return s->has_uso;
 224 }
 225
 226 static bool tap_has_vnet_hdr(NetClientState *nc)
 227 {
 228     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 229
 230     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 231
 232     return !!s->host_vnet_hdr_len;
 233 }
 234
 235 static bool tap_has_vnet_hdr_len(NetClientState *nc, int len)
 236 {
 237     return tap_has_vnet_hdr(nc);
 238 }
 239
 240 static void tap_set_vnet_hdr_len(NetClientState *nc, int len)
 241 {
 242     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 243
 244     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 245
 246     tap_fd_set_vnet_hdr_len(s->fd, len);
 247     s->host_vnet_hdr_len = len;
 248     s->using_vnet_hdr = true;
 249 }
 250
 251 static int tap_set_vnet_le(NetClientState *nc, bool is_le)
 252 {
 253     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 254
 255     return tap_fd_set_vnet_le(s->fd, is_le);
 256 }
 257
 258 static int tap_set_vnet_be(NetClientState *nc, bool is_be)
 259 {
 260     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 261
 262     return tap_fd_set_vnet_be(s->fd, is_be);
 263 }
 264
 265 static void tap_set_offload(NetClientState *nc, int csum, int tso4,
 266                      int tso6, int ecn, int ufo, int uso4, int uso6)
 267 {
 268     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 269     if (s->fd < 0) {
 270         return;
 271     }
 272
 273     tap_fd_set_offload(s->fd, csum, tso4, tso6, ecn, ufo, uso4, uso6);
 274 }
 275
 276 static void tap_exit_notify(Notifier *notifier, void *data)
 277 {
 278     TAPState *s = container_of(notifier, TAPState, exit);
 279     Error *err = NULL;
 280
 281     if (s->down_script[0]) {
 282         launch_script(s->down_script, s->down_script_arg, s->fd, &err);
 283         if (err) {
 284             error_report_err(err);
 285         }
 286     }
 287 }
 288
 289 static void tap_cleanup(NetClientState *nc)
 290 {
 291     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 292
 293     if (s->vhost_net) {
 294         vhost_net_cleanup(s->vhost_net);
 295         g_free(s->vhost_net);
 296         s->vhost_net = NULL;
 297     }
 298
 299     qemu_purge_queued_packets(nc);
 300
 301     tap_exit_notify(&s->exit, NULL);
 302     qemu_remove_exit_notifier(&s->exit);
 303
 304     tap_read_poll(s, false);
 305     tap_write_poll(s, false);
 306     close(s->fd);
 307     s->fd = -1;
 308 }
 309
 310 static void tap_poll(NetClientState *nc, bool enable)
 311 {
 312     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 313     tap_read_poll(s, enable);
 314     tap_write_poll(s, enable);
 315 }
 316
 317 static bool tap_set_steering_ebpf(NetClientState *nc, int prog_fd)
 318 {
 319     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 320     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 321
 322     return tap_fd_set_steering_ebpf(s->fd, prog_fd) == 0;
 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_DRIVER_TAP);
 329     return s->fd;
 330 }
 331
 332 /* fd support */
 333
 334 static NetClientInfo net_tap_info = {
 335     .type = NET_CLIENT_DRIVER_TAP,
 336     .size = sizeof(TAPState),
 337     .receive = tap_receive,
 338     .receive_iov = tap_receive_iov,
 339     .poll = tap_poll,
 340     .cleanup = tap_cleanup,
 341     .has_ufo = tap_has_ufo,
 342     .has_uso = tap_has_uso,
 343     .has_vnet_hdr = tap_has_vnet_hdr,
 344     .has_vnet_hdr_len = tap_has_vnet_hdr_len,
 345     .set_offload = tap_set_offload,
 346     .set_vnet_hdr_len = tap_set_vnet_hdr_len,
 347     .set_vnet_le = tap_set_vnet_le,
 348     .set_vnet_be = tap_set_vnet_be,
 349     .set_steering_ebpf = tap_set_steering_ebpf,
 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->has_uso = tap_probe_has_uso(s->fd);
 370     s->enabled = true;
 371     tap_set_offload(&s->nc, 0, 0, 0, 0, 0, 0, 0);
 372     /*
 373      * Make sure host header length is set correctly in tap:
 374      * it might have been modified by another instance of qemu.
 375      */
 376     if (vnet_hdr) {
 377         tap_fd_set_vnet_hdr_len(s->fd, s->host_vnet_hdr_len);
 378     }
 379     tap_read_poll(s, true);
 380     s->vhost_net = NULL;
 381
 382     s->exit.notify = tap_exit_notify;
 383     qemu_add_exit_notifier(&s->exit);
 384
 385     return s;
 386 }
 387
 388 static void close_all_fds_after_fork(int excluded_fd)
 389 {
 390     const int skip_fd[] = {STDIN_FILENO, STDOUT_FILENO, STDERR_FILENO,
 391                            excluded_fd};
 392     unsigned int nskip = ARRAY_SIZE(skip_fd);
 393
 394     /*
 395      * skip_fd must be an ordered array of distinct fds, exclude
 396      * excluded_fd if already included in the [STDIN_FILENO - STDERR_FILENO]
 397      * range
 398      */
 399     if (excluded_fd <= STDERR_FILENO) {
 400         nskip--;
 401     }
 402
 403     qemu_close_all_open_fd(skip_fd, nskip);
 404 }
 405
 406 static void launch_script(const char *setup_script, const char *ifname,
 407                           int fd, Error **errp)
 408 {
 409     int pid, status;
 410     char *args[3];
 411     char **parg;
 412
 413     /* try to launch network script */
 414     pid = fork();
 415     if (pid < 0) {
 416         error_setg_errno(errp, errno, "could not launch network script %s",
 417                          setup_script);
 418         return;
 419     }
 420     if (pid == 0) {
 421         close_all_fds_after_fork(fd);
 422         parg = args;
 423         *parg++ = (char *)setup_script;
 424         *parg++ = (char *)ifname;
 425         *parg = NULL;
 426         execv(setup_script, args);
 427         _exit(1);
 428     } else {
 429         while (waitpid(pid, &status, 0) != pid) {
 430             /* loop */
 431         }
 432
 433         if (WIFEXITED(status) && WEXITSTATUS(status) == 0) {
 434             return;
 435         }
 436         error_setg(errp, "network script %s failed with status %d",
 437                    setup_script, status);
 438     }
 439 }
 440
 441 static int recv_fd(int c)
 442 {
 443     int fd;
 444     uint8_t msgbuf[CMSG_SPACE(sizeof(fd))];
 445     struct msghdr msg = {
 446         .msg_control = msgbuf,
 447         .msg_controllen = sizeof(msgbuf),
 448     };
 449     struct cmsghdr *cmsg;
 450     struct iovec iov;
 451     uint8_t req[1];
 452     ssize_t len;
 453
 454     cmsg = CMSG_FIRSTHDR(&msg);
 455     cmsg->cmsg_level = SOL_SOCKET;
 456     cmsg->cmsg_type = SCM_RIGHTS;
 457     cmsg->cmsg_len = CMSG_LEN(sizeof(fd));
 458     msg.msg_controllen = cmsg->cmsg_len;
 459
 460     iov.iov_base = req;
 461     iov.iov_len = sizeof(req);
 462
 463     msg.msg_iov = &iov;
 464     msg.msg_iovlen = 1;
 465
 466     len = recvmsg(c, &msg, 0);
 467     if (len > 0) {
 468         memcpy(&fd, CMSG_DATA(cmsg), sizeof(fd));
 469         return fd;
 470     }
 471
 472     return len;
 473 }
 474
 475 static int net_bridge_run_helper(const char *helper, const char *bridge,
 476                                  Error **errp)
 477 {
 478     sigset_t oldmask, mask;
 479     g_autofree char *default_helper = NULL;
 480     int pid, status;
 481     char *args[5];
 482     char **parg;
 483     int sv[2];
 484
 485     sigemptyset(&mask);
 486     sigaddset(&mask, SIGCHLD);
 487     sigprocmask(SIG_BLOCK, &mask, &oldmask);
 488
 489     if (!helper) {
 490         helper = default_helper = get_relocated_path(DEFAULT_BRIDGE_HELPER);
 491     }
 492
 493     if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) {
 494         error_setg_errno(errp, errno, "socketpair() failed");
 495         return -1;
 496     }
 497
 498     /* try to launch bridge helper */
 499     pid = fork();
 500     if (pid < 0) {
 501         error_setg_errno(errp, errno, "Can't fork bridge helper");
 502         return -1;
 503     }
 504     if (pid == 0) {
 505         char *fd_buf = NULL;
 506         char *br_buf = NULL;
 507         char *helper_cmd = NULL;
 508
 509         close_all_fds_after_fork(sv[1]);
 510         fd_buf = g_strdup_printf("%s%d", "--fd=", sv[1]);
 511
 512         if (strrchr(helper, ' ') || strrchr(helper, '\t')) {
 513             /* assume helper is a command */
 514
 515             if (strstr(helper, "--br=") == NULL) {
 516                 br_buf = g_strdup_printf("%s%s", "--br=", bridge);
 517             }
 518
 519             helper_cmd = g_strdup_printf("%s %s %s %s", helper,
 520                             "--use-vnet", fd_buf, br_buf ? br_buf : "");
 521
 522             parg = args;
 523             *parg++ = (char *)"sh";
 524             *parg++ = (char *)"-c";
 525             *parg++ = helper_cmd;
 526             *parg++ = NULL;
 527
 528             execv("/bin/sh", args);
 529             g_free(helper_cmd);
 530         } else {
 531             /* assume helper is just the executable path name */
 532
 533             br_buf = g_strdup_printf("%s%s", "--br=", bridge);
 534
 535             parg = args;
 536             *parg++ = (char *)helper;
 537             *parg++ = (char *)"--use-vnet";
 538             *parg++ = fd_buf;
 539             *parg++ = br_buf;
 540             *parg++ = NULL;
 541
 542             execv(helper, args);
 543         }
 544         g_free(fd_buf);
 545         g_free(br_buf);
 546         _exit(1);
 547
 548     } else {
 549         int fd;
 550         int saved_errno;
 551
 552         close(sv[1]);
 553
 554         fd = RETRY_ON_EINTR(recv_fd(sv[0]));
 555         saved_errno = errno;
 556
 557         close(sv[0]);
 558
 559         while (waitpid(pid, &status, 0) != pid) {
 560             /* loop */
 561         }
 562         sigprocmask(SIG_SETMASK, &oldmask, NULL);
 563         if (fd < 0) {
 564             error_setg_errno(errp, saved_errno,
 565                              "failed to recv file descriptor");
 566             return -1;
 567         }
 568         if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 569             error_setg(errp, "bridge helper failed");
 570             return -1;
 571         }
 572         return fd;
 573     }
 574 }
 575
 576 int net_init_bridge(const Netdev *netdev, const char *name,
 577                     NetClientState *peer, Error **errp)
 578 {
 579     const NetdevBridgeOptions *bridge;
 580     const char *helper, *br;
 581     TAPState *s;
 582     int fd, vnet_hdr;
 583
 584     assert(netdev->type == NET_CLIENT_DRIVER_BRIDGE);
 585     bridge = &netdev->u.bridge;
 586     helper = bridge->helper;
 587     br     = bridge->br ?: DEFAULT_BRIDGE_INTERFACE;
 588
 589     fd = net_bridge_run_helper(helper, br, errp);
 590     if (fd == -1) {
 591         return -1;
 592     }
 593
 594     if (!g_unix_set_fd_nonblocking(fd, true, NULL)) {
 595         error_setg_errno(errp, errno, "Failed to set FD nonblocking");
 596         return -1;
 597     }
 598     vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 599     if (vnet_hdr < 0) {
 600         close(fd);
 601         return -1;
 602     }
 603     s = net_tap_fd_init(peer, "bridge", name, fd, vnet_hdr);
 604
 605     qemu_set_info_str(&s->nc, "helper=%s,br=%s", helper, br);
 606
 607     return 0;
 608 }
 609
 610 static int net_tap_init(const NetdevTapOptions *tap, int *vnet_hdr,
 611                         const char *setup_script, char *ifname,
 612                         size_t ifname_sz, int mq_required, Error **errp)
 613 {
 614     Error *err = NULL;
 615     int fd, vnet_hdr_required;
 616
 617     if (tap->has_vnet_hdr) {
 618         *vnet_hdr = tap->vnet_hdr;
 619         vnet_hdr_required = *vnet_hdr;
 620     } else {
 621         *vnet_hdr = 1;
 622         vnet_hdr_required = 0;
 623     }
 624
 625     fd = RETRY_ON_EINTR(tap_open(ifname, ifname_sz, vnet_hdr, vnet_hdr_required,
 626                       mq_required, errp));
 627     if (fd < 0) {
 628         return -1;
 629     }
 630
 631     if (setup_script &&
 632         setup_script[0] != '\0' &&
 633         strcmp(setup_script, "no") != 0) {
 634         launch_script(setup_script, ifname, fd, &err);
 635         if (err) {
 636             error_propagate(errp, err);
 637             close(fd);
 638             return -1;
 639         }
 640     }
 641
 642     return fd;
 643 }
 644
 645 #define MAX_TAP_QUEUES 1024
 646
 647 static void net_init_tap_one(const NetdevTapOptions *tap, NetClientState *peer,
 648                              const char *model, const char *name,
 649                              const char *ifname, const char *script,
 650                              const char *downscript, const char *vhostfdname,
 651                              int vnet_hdr, int fd, Error **errp)
 652 {
 653     Error *err = NULL;
 654     TAPState *s = net_tap_fd_init(peer, model, name, fd, vnet_hdr);
 655     int vhostfd;
 656
 657     tap_set_sndbuf(s->fd, tap, &err);
 658     if (err) {
 659         error_propagate(errp, err);
 660         goto failed;
 661     }
 662
 663     if (tap->fd || tap->fds) {
 664         qemu_set_info_str(&s->nc, "fd=%d", fd);
 665     } else if (tap->helper) {
 666         qemu_set_info_str(&s->nc, "helper=%s", tap->helper);
 667     } else {
 668         qemu_set_info_str(&s->nc, "ifname=%s,script=%s,downscript=%s", ifname,
 669                           script, downscript);
 670
 671         if (strcmp(downscript, "no") != 0) {
 672             snprintf(s->down_script, sizeof(s->down_script), "%s", downscript);
 673             snprintf(s->down_script_arg, sizeof(s->down_script_arg),
 674                      "%s", ifname);
 675         }
 676     }
 677
 678     if (tap->has_vhost ? tap->vhost :
 679         vhostfdname || (tap->has_vhostforce && tap->vhostforce)) {
 680         VhostNetOptions options;
 681
 682         options.backend_type = VHOST_BACKEND_TYPE_KERNEL;
 683         options.net_backend = &s->nc;
 684         if (tap->has_poll_us) {
 685             options.busyloop_timeout = tap->poll_us;
 686         } else {
 687             options.busyloop_timeout = 0;
 688         }
 689
 690         if (vhostfdname) {
 691             vhostfd = monitor_fd_param(monitor_cur(), vhostfdname, &err);
 692             if (vhostfd == -1) {
 693                 error_propagate(errp, err);
 694                 goto failed;
 695             }
 696             if (!g_unix_set_fd_nonblocking(vhostfd, true, NULL)) {
 697                 error_setg_errno(errp, errno, "%s: Can't use file descriptor %d",
 698                                  name, fd);
 699                 goto failed;
 700             }
 701         } else {
 702             vhostfd = open("/dev/vhost-net", O_RDWR);
 703             if (vhostfd < 0) {
 704                 error_setg_errno(errp, errno,
 705                                  "tap: open vhost char device failed");
 706                 goto failed;
 707             }
 708             if (!g_unix_set_fd_nonblocking(vhostfd, true, NULL)) {
 709                 error_setg_errno(errp, errno, "Failed to set FD nonblocking");
 710                 goto failed;
 711             }
 712         }
 713         options.opaque = (void *)(uintptr_t)vhostfd;
 714         options.nvqs = 2;
 715
 716         s->vhost_net = vhost_net_init(&options);
 717         if (!s->vhost_net) {
 718             error_setg(errp,
 719                        "vhost-net requested but could not be initialized");
 720             goto failed;
 721         }
 722     } else if (vhostfdname) {
 723         error_setg(errp, "vhostfd(s)= is not valid without vhost");
 724         goto failed;
 725     }
 726
 727     return;
 728
 729 failed:
 730     qemu_del_net_client(&s->nc);
 731 }
 732
 733 static int get_fds(char *str, char *fds[], int max)
 734 {
 735     char *ptr = str, *this;
 736     size_t len = strlen(str);
 737     int i = 0;
 738
 739     while (i < max && ptr < str + len) {
 740         this = strchr(ptr, ':');
 741
 742         if (this == NULL) {
 743             fds[i] = g_strdup(ptr);
 744         } else {
 745             fds[i] = g_strndup(ptr, this - ptr);
 746         }
 747
 748         i++;
 749         if (this == NULL) {
 750             break;
 751         } else {
 752             ptr = this + 1;
 753         }
 754     }
 755
 756     return i;
 757 }
 758
 759 int net_init_tap(const Netdev *netdev, const char *name,
 760                  NetClientState *peer, Error **errp)
 761 {
 762     const NetdevTapOptions *tap;
 763     int fd, vnet_hdr = 0, i = 0, queues;
 764     /* for the no-fd, no-helper case */
 765     const char *script;
 766     const char *downscript;
 767     Error *err = NULL;
 768     const char *vhostfdname;
 769     char ifname[128];
 770     int ret = 0;
 771
 772     assert(netdev->type == NET_CLIENT_DRIVER_TAP);
 773     tap = &netdev->u.tap;
 774     queues = tap->has_queues ? tap->queues : 1;
 775     vhostfdname = tap->vhostfd;
 776     script = tap->script;
 777     downscript = tap->downscript;
 778
 779     /* QEMU hubs do not support multiqueue tap, in this case peer is set.
 780      * For -netdev, peer is always NULL. */
 781     if (peer && (tap->has_queues || tap->fds || tap->vhostfds)) {
 782         error_setg(errp, "Multiqueue tap cannot be used with hubs");
 783         return -1;
 784     }
 785
 786     if (tap->fd) {
 787         if (tap->ifname || tap->script || tap->downscript ||
 788             tap->has_vnet_hdr || tap->helper || tap->has_queues ||
 789             tap->fds || tap->vhostfds) {
 790             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 791                        "helper=, queues=, fds=, and vhostfds= "
 792                        "are invalid with fd=");
 793             return -1;
 794         }
 795
 796         fd = monitor_fd_param(monitor_cur(), tap->fd, errp);
 797         if (fd == -1) {
 798             return -1;
 799         }
 800
 801         if (!g_unix_set_fd_nonblocking(fd, true, NULL)) {
 802             error_setg_errno(errp, errno, "%s: Can't use file descriptor %d",
 803                              name, fd);
 804             close(fd);
 805             return -1;
 806         }
 807
 808         vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 809         if (vnet_hdr < 0) {
 810             close(fd);
 811             return -1;
 812         }
 813
 814         net_init_tap_one(tap, peer, "tap", name, NULL,
 815                          script, downscript,
 816                          vhostfdname, vnet_hdr, fd, &err);
 817         if (err) {
 818             error_propagate(errp, err);
 819             close(fd);
 820             return -1;
 821         }
 822     } else if (tap->fds) {
 823         char **fds;
 824         char **vhost_fds;
 825         int nfds = 0, nvhosts = 0;
 826
 827         if (tap->ifname || tap->script || tap->downscript ||
 828             tap->has_vnet_hdr || tap->helper || tap->has_queues ||
 829             tap->vhostfd) {
 830             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 831                        "helper=, queues=, and vhostfd= "
 832                        "are invalid with fds=");
 833             return -1;
 834         }
 835
 836         fds = g_new0(char *, MAX_TAP_QUEUES);
 837         vhost_fds = g_new0(char *, MAX_TAP_QUEUES);
 838
 839         nfds = get_fds(tap->fds, fds, MAX_TAP_QUEUES);
 840         if (tap->vhostfds) {
 841             nvhosts = get_fds(tap->vhostfds, vhost_fds, MAX_TAP_QUEUES);
 842             if (nfds != nvhosts) {
 843                 error_setg(errp, "The number of fds passed does not match "
 844                            "the number of vhostfds passed");
 845                 ret = -1;
 846                 goto free_fail;
 847             }
 848         }
 849
 850         for (i = 0; i < nfds; i++) {
 851             fd = monitor_fd_param(monitor_cur(), fds[i], errp);
 852             if (fd == -1) {
 853                 ret = -1;
 854                 goto free_fail;
 855             }
 856
 857             ret = g_unix_set_fd_nonblocking(fd, true, NULL);
 858             if (!ret) {
 859                 error_setg_errno(errp, errno, "%s: Can't use file descriptor %d",
 860                                  name, fd);
 861                 goto free_fail;
 862             }
 863
 864             if (i == 0) {
 865                 vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 866                 if (vnet_hdr < 0) {
 867                     ret = -1;
 868                     goto free_fail;
 869                 }
 870             } else if (vnet_hdr != tap_probe_vnet_hdr(fd, NULL)) {
 871                 error_setg(errp,
 872                            "vnet_hdr not consistent across given tap fds");
 873                 ret = -1;
 874                 goto free_fail;
 875             }
 876
 877             net_init_tap_one(tap, peer, "tap", name, ifname,
 878                              script, downscript,
 879                              tap->vhostfds ? vhost_fds[i] : NULL,
 880                              vnet_hdr, fd, &err);
 881             if (err) {
 882                 error_propagate(errp, err);
 883                 ret = -1;
 884                 goto free_fail;
 885             }
 886         }
 887
 888 free_fail:
 889         for (i = 0; i < nvhosts; i++) {
 890             g_free(vhost_fds[i]);
 891         }
 892         for (i = 0; i < nfds; i++) {
 893             g_free(fds[i]);
 894         }
 895         g_free(fds);
 896         g_free(vhost_fds);
 897         return ret;
 898     } else if (tap->helper) {
 899         if (tap->ifname || tap->script || tap->downscript ||
 900             tap->has_vnet_hdr || tap->has_queues || tap->vhostfds) {
 901             error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, "
 902                        "queues=, and vhostfds= are invalid with helper=");
 903             return -1;
 904         }
 905
 906         fd = net_bridge_run_helper(tap->helper,
 907                                    tap->br ?: DEFAULT_BRIDGE_INTERFACE,
 908                                    errp);
 909         if (fd == -1) {
 910             return -1;
 911         }
 912
 913         if (!g_unix_set_fd_nonblocking(fd, true, NULL)) {
 914             error_setg_errno(errp, errno, "Failed to set FD nonblocking");
 915             return -1;
 916         }
 917         vnet_hdr = tap_probe_vnet_hdr(fd, errp);
 918         if (vnet_hdr < 0) {
 919             close(fd);
 920             return -1;
 921         }
 922
 923         net_init_tap_one(tap, peer, "bridge", name, ifname,
 924                          script, downscript, vhostfdname,
 925                          vnet_hdr, fd, &err);
 926         if (err) {
 927             error_propagate(errp, err);
 928             close(fd);
 929             return -1;
 930         }
 931     } else {
 932         g_autofree char *default_script = NULL;
 933         g_autofree char *default_downscript = NULL;
 934         if (tap->vhostfds) {
 935             error_setg(errp, "vhostfds= is invalid if fds= wasn't specified");
 936             return -1;
 937         }
 938
 939         if (!script) {
 940             script = default_script = get_relocated_path(DEFAULT_NETWORK_SCRIPT);
 941         }
 942         if (!downscript) {
 943             downscript = default_downscript =
 944                                  get_relocated_path(DEFAULT_NETWORK_DOWN_SCRIPT);
 945         }
 946
 947         if (tap->ifname) {
 948             pstrcpy(ifname, sizeof ifname, tap->ifname);
 949         } else {
 950             ifname[0] = '\0';
 951         }
 952
 953         for (i = 0; i < queues; i++) {
 954             fd = net_tap_init(tap, &vnet_hdr, i >= 1 ? "no" : script,
 955                               ifname, sizeof ifname, queues > 1, errp);
 956             if (fd == -1) {
 957                 return -1;
 958             }
 959
 960             if (queues > 1 && i == 0 && !tap->ifname) {
 961                 if (tap_fd_get_ifname(fd, ifname)) {
 962                     error_setg(errp, "Fail to get ifname");
 963                     close(fd);
 964                     return -1;
 965                 }
 966             }
 967
 968             net_init_tap_one(tap, peer, "tap", name, ifname,
 969                              i >= 1 ? "no" : script,
 970                              i >= 1 ? "no" : downscript,
 971                              vhostfdname, vnet_hdr, fd, &err);
 972             if (err) {
 973                 error_propagate(errp, err);
 974                 close(fd);
 975                 return -1;
 976             }
 977         }
 978     }
 979
 980     return 0;
 981 }
 982
 983 VHostNetState *tap_get_vhost_net(NetClientState *nc)
 984 {
 985     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 986     assert(nc->info->type == NET_CLIENT_DRIVER_TAP);
 987     return s->vhost_net;
 988 }
 989
 990 int tap_enable(NetClientState *nc)
 991 {
 992     TAPState *s = DO_UPCAST(TAPState, nc, nc);
 993     int ret;
 994
 995     if (s->enabled) {
 996         return 0;
 997     } else {
 998         ret = tap_fd_enable(s->fd);
 999         if (ret == 0) {
1000             s->enabled = true;
1001             tap_update_fd_handler(s);
1002         }
1003         return ret;
1004     }
1005 }
1006
1007 int tap_disable(NetClientState *nc)
1008 {
1009     TAPState *s = DO_UPCAST(TAPState, nc, nc);
1010     int ret;
1011
1012     if (s->enabled == 0) {
1013         return 0;
1014     } else {
1015         ret = tap_fd_disable(s->fd);
1016         if (ret == 0) {
1017             qemu_purge_queued_packets(nc);
1018             s->enabled = false;
1019             tap_update_fd_handler(s);
1020         }
1021         return ret;
1022     }
1023 }