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_mrg_rxbuf 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_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 uint8_t min_pkt[ETH_ZLEN]; 193 size_t min_pktsz = sizeof(min_pkt); 194 195 size = tap_read_packet(s->fd, s->buf, sizeof(s->buf)); 196 if (size <= 0) { 197 break; 198 } 199 200 if (s->host_vnet_hdr_len && !s->using_vnet_hdr) { 201 buf += s->host_vnet_hdr_len; 202 size -= s->host_vnet_hdr_len; 203 } 204 205 if (net_peer_needs_padding(&s->nc)) { 206 if (eth_pad_short_frame(min_pkt, &min_pktsz, buf, size)) { 207 buf = min_pkt; 208 size = min_pktsz; 209 } 210 } 211 212 size = qemu_send_packet_async(&s->nc, buf, size, tap_send_completed); 213 if (size == 0) { 214 tap_read_poll(s, false); 215 break; 216 } else if (size < 0) { 217 break; 218 } 219 220 /* 221 * When the host keeps receiving more packets while tap_send() is 222 * running we can hog the BQL. Limit the number of 223 * packets that are processed per tap_send() callback to prevent 224 * stalling the guest. 225 */ 226 packets++; 227 if (packets >= 50) { 228 break; 229 } 230 } 231 } 232 233 static bool tap_has_ufo(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->has_ufo; 240 } 241 242 static bool tap_has_uso(NetClientState *nc) 243 { 244 TAPState *s = DO_UPCAST(TAPState, nc, nc); 245 246 assert(nc->info->type == NET_CLIENT_DRIVER_TAP); 247 248 return s->has_uso; 249 } 250 251 static bool tap_has_vnet_hdr(NetClientState *nc) 252 { 253 TAPState *s = DO_UPCAST(TAPState, nc, nc); 254 255 assert(nc->info->type == NET_CLIENT_DRIVER_TAP); 256 257 return !!s->host_vnet_hdr_len; 258 } 259 260 static bool tap_has_vnet_hdr_len(NetClientState *nc, int len) 261 { 262 return tap_has_vnet_hdr(nc); 263 } 264 265 static void tap_set_vnet_hdr_len(NetClientState *nc, int len) 266 { 267 TAPState *s = DO_UPCAST(TAPState, nc, nc); 268 269 assert(nc->info->type == NET_CLIENT_DRIVER_TAP); 270 271 tap_fd_set_vnet_hdr_len(s->fd, len); 272 s->host_vnet_hdr_len = len; 273 s->using_vnet_hdr = true; 274 } 275 276 static int tap_set_vnet_le(NetClientState *nc, bool is_le) 277 { 278 TAPState *s = DO_UPCAST(TAPState, nc, nc); 279 280 return tap_fd_set_vnet_le(s->fd, is_le); 281 } 282 283 static int tap_set_vnet_be(NetClientState *nc, bool is_be) 284 { 285 TAPState *s = DO_UPCAST(TAPState, nc, nc); 286 287 return tap_fd_set_vnet_be(s->fd, is_be); 288 } 289 290 static void tap_set_offload(NetClientState *nc, int csum, int tso4, 291 int tso6, int ecn, int ufo, int uso4, int uso6) 292 { 293 TAPState *s = DO_UPCAST(TAPState, nc, nc); 294 if (s->fd < 0) { 295 return; 296 } 297 298 tap_fd_set_offload(s->fd, csum, tso4, tso6, ecn, ufo, uso4, uso6); 299 } 300 301 static void tap_exit_notify(Notifier *notifier, void *data) 302 { 303 TAPState *s = container_of(notifier, TAPState, exit); 304 Error *err = NULL; 305 306 if (s->down_script[0]) { 307 launch_script(s->down_script, s->down_script_arg, s->fd, &err); 308 if (err) { 309 error_report_err(err); 310 } 311 } 312 } 313 314 static void tap_cleanup(NetClientState *nc) 315 { 316 TAPState *s = DO_UPCAST(TAPState, nc, nc); 317 318 if (s->vhost_net) { 319 vhost_net_cleanup(s->vhost_net); 320 g_free(s->vhost_net); 321 s->vhost_net = NULL; 322 } 323 324 qemu_purge_queued_packets(nc); 325 326 tap_exit_notify(&s->exit, NULL); 327 qemu_remove_exit_notifier(&s->exit); 328 329 tap_read_poll(s, false); 330 tap_write_poll(s, false); 331 close(s->fd); 332 s->fd = -1; 333 } 334 335 static void tap_poll(NetClientState *nc, bool enable) 336 { 337 TAPState *s = DO_UPCAST(TAPState, nc, nc); 338 tap_read_poll(s, enable); 339 tap_write_poll(s, enable); 340 } 341 342 static bool tap_set_steering_ebpf(NetClientState *nc, int prog_fd) 343 { 344 TAPState *s = DO_UPCAST(TAPState, nc, nc); 345 assert(nc->info->type == NET_CLIENT_DRIVER_TAP); 346 347 return tap_fd_set_steering_ebpf(s->fd, prog_fd) == 0; 348 } 349 350 int tap_get_fd(NetClientState *nc) 351 { 352 TAPState *s = DO_UPCAST(TAPState, nc, nc); 353 assert(nc->info->type == NET_CLIENT_DRIVER_TAP); 354 return s->fd; 355 } 356 357 /* fd support */ 358 359 static NetClientInfo net_tap_info = { 360 .type = NET_CLIENT_DRIVER_TAP, 361 .size = sizeof(TAPState), 362 .receive = tap_receive, 363 .receive_raw = tap_receive_raw, 364 .receive_iov = tap_receive_iov, 365 .poll = tap_poll, 366 .cleanup = tap_cleanup, 367 .has_ufo = tap_has_ufo, 368 .has_uso = tap_has_uso, 369 .has_vnet_hdr = tap_has_vnet_hdr, 370 .has_vnet_hdr_len = tap_has_vnet_hdr_len, 371 .set_offload = tap_set_offload, 372 .set_vnet_hdr_len = tap_set_vnet_hdr_len, 373 .set_vnet_le = tap_set_vnet_le, 374 .set_vnet_be = tap_set_vnet_be, 375 .set_steering_ebpf = tap_set_steering_ebpf, 376 }; 377 378 static TAPState *net_tap_fd_init(NetClientState *peer, 379 const char *model, 380 const char *name, 381 int fd, 382 int vnet_hdr) 383 { 384 NetClientState *nc; 385 TAPState *s; 386 387 nc = qemu_new_net_client(&net_tap_info, peer, model, name); 388 389 s = DO_UPCAST(TAPState, nc, nc); 390 391 s->fd = fd; 392 s->host_vnet_hdr_len = vnet_hdr ? sizeof(struct virtio_net_hdr) : 0; 393 s->using_vnet_hdr = false; 394 s->has_ufo = tap_probe_has_ufo(s->fd); 395 s->has_uso = tap_probe_has_uso(s->fd); 396 s->enabled = true; 397 tap_set_offload(&s->nc, 0, 0, 0, 0, 0, 0, 0); 398 /* 399 * Make sure host header length is set correctly in tap: 400 * it might have been modified by another instance of qemu. 401 */ 402 if (vnet_hdr) { 403 tap_fd_set_vnet_hdr_len(s->fd, s->host_vnet_hdr_len); 404 } 405 tap_read_poll(s, true); 406 s->vhost_net = NULL; 407 408 s->exit.notify = tap_exit_notify; 409 qemu_add_exit_notifier(&s->exit); 410 411 return s; 412 } 413 414 static void launch_script(const char *setup_script, const char *ifname, 415 int fd, Error **errp) 416 { 417 int pid, status; 418 char *args[3]; 419 char **parg; 420 421 /* try to launch network script */ 422 pid = fork(); 423 if (pid < 0) { 424 error_setg_errno(errp, errno, "could not launch network script %s", 425 setup_script); 426 return; 427 } 428 if (pid == 0) { 429 int open_max = sysconf(_SC_OPEN_MAX), i; 430 431 for (i = 3; i < open_max; i++) { 432 if (i != fd) { 433 close(i); 434 } 435 } 436 parg = args; 437 *parg++ = (char *)setup_script; 438 *parg++ = (char *)ifname; 439 *parg = NULL; 440 execv(setup_script, args); 441 _exit(1); 442 } else { 443 while (waitpid(pid, &status, 0) != pid) { 444 /* loop */ 445 } 446 447 if (WIFEXITED(status) && WEXITSTATUS(status) == 0) { 448 return; 449 } 450 error_setg(errp, "network script %s failed with status %d", 451 setup_script, status); 452 } 453 } 454 455 static int recv_fd(int c) 456 { 457 int fd; 458 uint8_t msgbuf[CMSG_SPACE(sizeof(fd))]; 459 struct msghdr msg = { 460 .msg_control = msgbuf, 461 .msg_controllen = sizeof(msgbuf), 462 }; 463 struct cmsghdr *cmsg; 464 struct iovec iov; 465 uint8_t req[1]; 466 ssize_t len; 467 468 cmsg = CMSG_FIRSTHDR(&msg); 469 cmsg->cmsg_level = SOL_SOCKET; 470 cmsg->cmsg_type = SCM_RIGHTS; 471 cmsg->cmsg_len = CMSG_LEN(sizeof(fd)); 472 msg.msg_controllen = cmsg->cmsg_len; 473 474 iov.iov_base = req; 475 iov.iov_len = sizeof(req); 476 477 msg.msg_iov = &iov; 478 msg.msg_iovlen = 1; 479 480 len = recvmsg(c, &msg, 0); 481 if (len > 0) { 482 memcpy(&fd, CMSG_DATA(cmsg), sizeof(fd)); 483 return fd; 484 } 485 486 return len; 487 } 488 489 static int net_bridge_run_helper(const char *helper, const char *bridge, 490 Error **errp) 491 { 492 sigset_t oldmask, mask; 493 g_autofree char *default_helper = NULL; 494 int pid, status; 495 char *args[5]; 496 char **parg; 497 int sv[2]; 498 499 sigemptyset(&mask); 500 sigaddset(&mask, SIGCHLD); 501 sigprocmask(SIG_BLOCK, &mask, &oldmask); 502 503 if (!helper) { 504 helper = default_helper = get_relocated_path(DEFAULT_BRIDGE_HELPER); 505 } 506 507 if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) { 508 error_setg_errno(errp, errno, "socketpair() failed"); 509 return -1; 510 } 511 512 /* try to launch bridge helper */ 513 pid = fork(); 514 if (pid < 0) { 515 error_setg_errno(errp, errno, "Can't fork bridge helper"); 516 return -1; 517 } 518 if (pid == 0) { 519 int open_max = sysconf(_SC_OPEN_MAX), i; 520 char *fd_buf = NULL; 521 char *br_buf = NULL; 522 char *helper_cmd = NULL; 523 524 for (i = 3; i < open_max; i++) { 525 if (i != sv[1]) { 526 close(i); 527 } 528 } 529 530 fd_buf = g_strdup_printf("%s%d", "--fd=", sv[1]); 531 532 if (strrchr(helper, ' ') || strrchr(helper, '\t')) { 533 /* assume helper is a command */ 534 535 if (strstr(helper, "--br=") == NULL) { 536 br_buf = g_strdup_printf("%s%s", "--br=", bridge); 537 } 538 539 helper_cmd = g_strdup_printf("%s %s %s %s", helper, 540 "--use-vnet", fd_buf, br_buf ? br_buf : ""); 541 542 parg = args; 543 *parg++ = (char *)"sh"; 544 *parg++ = (char *)"-c"; 545 *parg++ = helper_cmd; 546 *parg++ = NULL; 547 548 execv("/bin/sh", args); 549 g_free(helper_cmd); 550 } else { 551 /* assume helper is just the executable path name */ 552 553 br_buf = g_strdup_printf("%s%s", "--br=", bridge); 554 555 parg = args; 556 *parg++ = (char *)helper; 557 *parg++ = (char *)"--use-vnet"; 558 *parg++ = fd_buf; 559 *parg++ = br_buf; 560 *parg++ = NULL; 561 562 execv(helper, args); 563 } 564 g_free(fd_buf); 565 g_free(br_buf); 566 _exit(1); 567 568 } else { 569 int fd; 570 int saved_errno; 571 572 close(sv[1]); 573 574 fd = RETRY_ON_EINTR(recv_fd(sv[0])); 575 saved_errno = errno; 576 577 close(sv[0]); 578 579 while (waitpid(pid, &status, 0) != pid) { 580 /* loop */ 581 } 582 sigprocmask(SIG_SETMASK, &oldmask, NULL); 583 if (fd < 0) { 584 error_setg_errno(errp, saved_errno, 585 "failed to recv file descriptor"); 586 return -1; 587 } 588 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) { 589 error_setg(errp, "bridge helper failed"); 590 return -1; 591 } 592 return fd; 593 } 594 } 595 596 int net_init_bridge(const Netdev *netdev, const char *name, 597 NetClientState *peer, Error **errp) 598 { 599 const NetdevBridgeOptions *bridge; 600 const char *helper, *br; 601 TAPState *s; 602 int fd, vnet_hdr; 603 604 assert(netdev->type == NET_CLIENT_DRIVER_BRIDGE); 605 bridge = &netdev->u.bridge; 606 helper = bridge->helper; 607 br = bridge->br ?: DEFAULT_BRIDGE_INTERFACE; 608 609 fd = net_bridge_run_helper(helper, br, errp); 610 if (fd == -1) { 611 return -1; 612 } 613 614 if (!g_unix_set_fd_nonblocking(fd, true, NULL)) { 615 error_setg_errno(errp, errno, "Failed to set FD nonblocking"); 616 return -1; 617 } 618 vnet_hdr = tap_probe_vnet_hdr(fd, errp); 619 if (vnet_hdr < 0) { 620 close(fd); 621 return -1; 622 } 623 s = net_tap_fd_init(peer, "bridge", name, fd, vnet_hdr); 624 625 qemu_set_info_str(&s->nc, "helper=%s,br=%s", helper, br); 626 627 return 0; 628 } 629 630 static int net_tap_init(const NetdevTapOptions *tap, int *vnet_hdr, 631 const char *setup_script, char *ifname, 632 size_t ifname_sz, int mq_required, Error **errp) 633 { 634 Error *err = NULL; 635 int fd, vnet_hdr_required; 636 637 if (tap->has_vnet_hdr) { 638 *vnet_hdr = tap->vnet_hdr; 639 vnet_hdr_required = *vnet_hdr; 640 } else { 641 *vnet_hdr = 1; 642 vnet_hdr_required = 0; 643 } 644 645 fd = RETRY_ON_EINTR(tap_open(ifname, ifname_sz, vnet_hdr, vnet_hdr_required, 646 mq_required, errp)); 647 if (fd < 0) { 648 return -1; 649 } 650 651 if (setup_script && 652 setup_script[0] != '\0' && 653 strcmp(setup_script, "no") != 0) { 654 launch_script(setup_script, ifname, fd, &err); 655 if (err) { 656 error_propagate(errp, err); 657 close(fd); 658 return -1; 659 } 660 } 661 662 return fd; 663 } 664 665 #define MAX_TAP_QUEUES 1024 666 667 static void net_init_tap_one(const NetdevTapOptions *tap, NetClientState *peer, 668 const char *model, const char *name, 669 const char *ifname, const char *script, 670 const char *downscript, const char *vhostfdname, 671 int vnet_hdr, int fd, Error **errp) 672 { 673 Error *err = NULL; 674 TAPState *s = net_tap_fd_init(peer, model, name, fd, vnet_hdr); 675 int vhostfd; 676 677 tap_set_sndbuf(s->fd, tap, &err); 678 if (err) { 679 error_propagate(errp, err); 680 goto failed; 681 } 682 683 if (tap->fd || tap->fds) { 684 qemu_set_info_str(&s->nc, "fd=%d", fd); 685 } else if (tap->helper) { 686 qemu_set_info_str(&s->nc, "helper=%s", tap->helper); 687 } else { 688 qemu_set_info_str(&s->nc, "ifname=%s,script=%s,downscript=%s", ifname, 689 script, downscript); 690 691 if (strcmp(downscript, "no") != 0) { 692 snprintf(s->down_script, sizeof(s->down_script), "%s", downscript); 693 snprintf(s->down_script_arg, sizeof(s->down_script_arg), 694 "%s", ifname); 695 } 696 } 697 698 if (tap->has_vhost ? tap->vhost : 699 vhostfdname || (tap->has_vhostforce && tap->vhostforce)) { 700 VhostNetOptions options; 701 702 options.backend_type = VHOST_BACKEND_TYPE_KERNEL; 703 options.net_backend = &s->nc; 704 if (tap->has_poll_us) { 705 options.busyloop_timeout = tap->poll_us; 706 } else { 707 options.busyloop_timeout = 0; 708 } 709 710 if (vhostfdname) { 711 vhostfd = monitor_fd_param(monitor_cur(), vhostfdname, &err); 712 if (vhostfd == -1) { 713 error_propagate(errp, err); 714 goto failed; 715 } 716 if (!g_unix_set_fd_nonblocking(vhostfd, true, NULL)) { 717 error_setg_errno(errp, errno, "%s: Can't use file descriptor %d", 718 name, fd); 719 goto failed; 720 } 721 } else { 722 vhostfd = open("/dev/vhost-net", O_RDWR); 723 if (vhostfd < 0) { 724 error_setg_errno(errp, errno, 725 "tap: open vhost char device failed"); 726 goto failed; 727 } 728 if (!g_unix_set_fd_nonblocking(vhostfd, true, NULL)) { 729 error_setg_errno(errp, errno, "Failed to set FD nonblocking"); 730 goto failed; 731 } 732 } 733 options.opaque = (void *)(uintptr_t)vhostfd; 734 options.nvqs = 2; 735 736 s->vhost_net = vhost_net_init(&options); 737 if (!s->vhost_net) { 738 error_setg(errp, 739 "vhost-net requested but could not be initialized"); 740 goto failed; 741 } 742 } else if (vhostfdname) { 743 error_setg(errp, "vhostfd(s)= is not valid without vhost"); 744 goto failed; 745 } 746 747 return; 748 749 failed: 750 qemu_del_net_client(&s->nc); 751 } 752 753 static int get_fds(char *str, char *fds[], int max) 754 { 755 char *ptr = str, *this; 756 size_t len = strlen(str); 757 int i = 0; 758 759 while (i < max && ptr < str + len) { 760 this = strchr(ptr, ':'); 761 762 if (this == NULL) { 763 fds[i] = g_strdup(ptr); 764 } else { 765 fds[i] = g_strndup(ptr, this - ptr); 766 } 767 768 i++; 769 if (this == NULL) { 770 break; 771 } else { 772 ptr = this + 1; 773 } 774 } 775 776 return i; 777 } 778 779 int net_init_tap(const Netdev *netdev, const char *name, 780 NetClientState *peer, Error **errp) 781 { 782 const NetdevTapOptions *tap; 783 int fd, vnet_hdr = 0, i = 0, queues; 784 /* for the no-fd, no-helper case */ 785 const char *script; 786 const char *downscript; 787 Error *err = NULL; 788 const char *vhostfdname; 789 char ifname[128]; 790 int ret = 0; 791 792 assert(netdev->type == NET_CLIENT_DRIVER_TAP); 793 tap = &netdev->u.tap; 794 queues = tap->has_queues ? tap->queues : 1; 795 vhostfdname = tap->vhostfd; 796 script = tap->script; 797 downscript = tap->downscript; 798 799 /* QEMU hubs do not support multiqueue tap, in this case peer is set. 800 * For -netdev, peer is always NULL. */ 801 if (peer && (tap->has_queues || tap->fds || tap->vhostfds)) { 802 error_setg(errp, "Multiqueue tap cannot be used with hubs"); 803 return -1; 804 } 805 806 if (tap->fd) { 807 if (tap->ifname || tap->script || tap->downscript || 808 tap->has_vnet_hdr || tap->helper || tap->has_queues || 809 tap->fds || tap->vhostfds) { 810 error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, " 811 "helper=, queues=, fds=, and vhostfds= " 812 "are invalid with fd="); 813 return -1; 814 } 815 816 fd = monitor_fd_param(monitor_cur(), tap->fd, errp); 817 if (fd == -1) { 818 return -1; 819 } 820 821 if (!g_unix_set_fd_nonblocking(fd, true, NULL)) { 822 error_setg_errno(errp, errno, "%s: Can't use file descriptor %d", 823 name, fd); 824 close(fd); 825 return -1; 826 } 827 828 vnet_hdr = tap_probe_vnet_hdr(fd, errp); 829 if (vnet_hdr < 0) { 830 close(fd); 831 return -1; 832 } 833 834 net_init_tap_one(tap, peer, "tap", name, NULL, 835 script, downscript, 836 vhostfdname, vnet_hdr, fd, &err); 837 if (err) { 838 error_propagate(errp, err); 839 close(fd); 840 return -1; 841 } 842 } else if (tap->fds) { 843 char **fds; 844 char **vhost_fds; 845 int nfds = 0, nvhosts = 0; 846 847 if (tap->ifname || tap->script || tap->downscript || 848 tap->has_vnet_hdr || tap->helper || tap->has_queues || 849 tap->vhostfd) { 850 error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, " 851 "helper=, queues=, and vhostfd= " 852 "are invalid with fds="); 853 return -1; 854 } 855 856 fds = g_new0(char *, MAX_TAP_QUEUES); 857 vhost_fds = g_new0(char *, MAX_TAP_QUEUES); 858 859 nfds = get_fds(tap->fds, fds, MAX_TAP_QUEUES); 860 if (tap->vhostfds) { 861 nvhosts = get_fds(tap->vhostfds, vhost_fds, MAX_TAP_QUEUES); 862 if (nfds != nvhosts) { 863 error_setg(errp, "The number of fds passed does not match " 864 "the number of vhostfds passed"); 865 ret = -1; 866 goto free_fail; 867 } 868 } 869 870 for (i = 0; i < nfds; i++) { 871 fd = monitor_fd_param(monitor_cur(), fds[i], errp); 872 if (fd == -1) { 873 ret = -1; 874 goto free_fail; 875 } 876 877 ret = g_unix_set_fd_nonblocking(fd, true, NULL); 878 if (!ret) { 879 error_setg_errno(errp, errno, "%s: Can't use file descriptor %d", 880 name, fd); 881 goto free_fail; 882 } 883 884 if (i == 0) { 885 vnet_hdr = tap_probe_vnet_hdr(fd, errp); 886 if (vnet_hdr < 0) { 887 ret = -1; 888 goto free_fail; 889 } 890 } else if (vnet_hdr != tap_probe_vnet_hdr(fd, NULL)) { 891 error_setg(errp, 892 "vnet_hdr not consistent across given tap fds"); 893 ret = -1; 894 goto free_fail; 895 } 896 897 net_init_tap_one(tap, peer, "tap", name, ifname, 898 script, downscript, 899 tap->vhostfds ? vhost_fds[i] : NULL, 900 vnet_hdr, fd, &err); 901 if (err) { 902 error_propagate(errp, err); 903 ret = -1; 904 goto free_fail; 905 } 906 } 907 908 free_fail: 909 for (i = 0; i < nvhosts; i++) { 910 g_free(vhost_fds[i]); 911 } 912 for (i = 0; i < nfds; i++) { 913 g_free(fds[i]); 914 } 915 g_free(fds); 916 g_free(vhost_fds); 917 return ret; 918 } else if (tap->helper) { 919 if (tap->ifname || tap->script || tap->downscript || 920 tap->has_vnet_hdr || tap->has_queues || tap->vhostfds) { 921 error_setg(errp, "ifname=, script=, downscript=, vnet_hdr=, " 922 "queues=, and vhostfds= are invalid with helper="); 923 return -1; 924 } 925 926 fd = net_bridge_run_helper(tap->helper, 927 tap->br ?: DEFAULT_BRIDGE_INTERFACE, 928 errp); 929 if (fd == -1) { 930 return -1; 931 } 932 933 if (!g_unix_set_fd_nonblocking(fd, true, NULL)) { 934 error_setg_errno(errp, errno, "Failed to set FD nonblocking"); 935 return -1; 936 } 937 vnet_hdr = tap_probe_vnet_hdr(fd, errp); 938 if (vnet_hdr < 0) { 939 close(fd); 940 return -1; 941 } 942 943 net_init_tap_one(tap, peer, "bridge", name, ifname, 944 script, downscript, vhostfdname, 945 vnet_hdr, fd, &err); 946 if (err) { 947 error_propagate(errp, err); 948 close(fd); 949 return -1; 950 } 951 } else { 952 g_autofree char *default_script = NULL; 953 g_autofree char *default_downscript = NULL; 954 if (tap->vhostfds) { 955 error_setg(errp, "vhostfds= is invalid if fds= wasn't specified"); 956 return -1; 957 } 958 959 if (!script) { 960 script = default_script = get_relocated_path(DEFAULT_NETWORK_SCRIPT); 961 } 962 if (!downscript) { 963 downscript = default_downscript = 964 get_relocated_path(DEFAULT_NETWORK_DOWN_SCRIPT); 965 } 966 967 if (tap->ifname) { 968 pstrcpy(ifname, sizeof ifname, tap->ifname); 969 } else { 970 ifname[0] = '\0'; 971 } 972 973 for (i = 0; i < queues; i++) { 974 fd = net_tap_init(tap, &vnet_hdr, i >= 1 ? "no" : script, 975 ifname, sizeof ifname, queues > 1, errp); 976 if (fd == -1) { 977 return -1; 978 } 979 980 if (queues > 1 && i == 0 && !tap->ifname) { 981 if (tap_fd_get_ifname(fd, ifname)) { 982 error_setg(errp, "Fail to get ifname"); 983 close(fd); 984 return -1; 985 } 986 } 987 988 net_init_tap_one(tap, peer, "tap", name, ifname, 989 i >= 1 ? "no" : script, 990 i >= 1 ? "no" : downscript, 991 vhostfdname, vnet_hdr, fd, &err); 992 if (err) { 993 error_propagate(errp, err); 994 close(fd); 995 return -1; 996 } 997 } 998 } 999 1000 return 0; 1001 } 1002 1003 VHostNetState *tap_get_vhost_net(NetClientState *nc) 1004 { 1005 TAPState *s = DO_UPCAST(TAPState, nc, nc); 1006 assert(nc->info->type == NET_CLIENT_DRIVER_TAP); 1007 return s->vhost_net; 1008 } 1009 1010 int tap_enable(NetClientState *nc) 1011 { 1012 TAPState *s = DO_UPCAST(TAPState, nc, nc); 1013 int ret; 1014 1015 if (s->enabled) { 1016 return 0; 1017 } else { 1018 ret = tap_fd_enable(s->fd); 1019 if (ret == 0) { 1020 s->enabled = true; 1021 tap_update_fd_handler(s); 1022 } 1023 return ret; 1024 } 1025 } 1026 1027 int tap_disable(NetClientState *nc) 1028 { 1029 TAPState *s = DO_UPCAST(TAPState, nc, nc); 1030 int ret; 1031 1032 if (s->enabled == 0) { 1033 return 0; 1034 } else { 1035 ret = tap_fd_disable(s->fd); 1036 if (ret == 0) { 1037 qemu_purge_queued_packets(nc); 1038 s->enabled = false; 1039 tap_update_fd_handler(s); 1040 } 1041 return ret; 1042 } 1043 } 1044