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