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 "net/tap.h" 27 28 #include "config-host.h" 29 30 #include <sys/ioctl.h> 31 #include <sys/stat.h> 32 #include <sys/wait.h> 33 #include <sys/socket.h> 34 #include <net/if.h> 35 36 #include "net.h" 37 #include "sysemu.h" 38 #include "qemu-char.h" 39 #include "qemu-common.h" 40 #include "qemu-error.h" 41 42 #include "net/tap-linux.h" 43 44 #include "hw/vhost_net.h" 45 46 /* Maximum GSO packet size (64k) plus plenty of room for 47 * the ethernet and virtio_net headers 48 */ 49 #define TAP_BUFSIZE (4096 + 65536) 50 51 typedef struct TAPState { 52 VLANClientState nc; 53 int fd; 54 char down_script[1024]; 55 char down_script_arg[128]; 56 uint8_t buf[TAP_BUFSIZE]; 57 unsigned int read_poll : 1; 58 unsigned int write_poll : 1; 59 unsigned int using_vnet_hdr : 1; 60 unsigned int has_ufo: 1; 61 VHostNetState *vhost_net; 62 unsigned host_vnet_hdr_len; 63 } TAPState; 64 65 static int launch_script(const char *setup_script, const char *ifname, int fd); 66 67 static int tap_can_send(void *opaque); 68 static void tap_send(void *opaque); 69 static void tap_writable(void *opaque); 70 71 static void tap_update_fd_handler(TAPState *s) 72 { 73 qemu_set_fd_handler2(s->fd, 74 s->read_poll ? tap_can_send : NULL, 75 s->read_poll ? tap_send : NULL, 76 s->write_poll ? tap_writable : NULL, 77 s); 78 } 79 80 static void tap_read_poll(TAPState *s, int enable) 81 { 82 s->read_poll = !!enable; 83 tap_update_fd_handler(s); 84 } 85 86 static void tap_write_poll(TAPState *s, int 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, 0); 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, 1); 111 return 0; 112 } 113 114 return len; 115 } 116 117 static ssize_t tap_receive_iov(VLANClientState *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(VLANClientState *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(VLANClientState *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 static int tap_can_send(void *opaque) 172 { 173 TAPState *s = opaque; 174 175 return qemu_can_send_packet(&s->nc); 176 } 177 178 #ifndef __sun__ 179 ssize_t tap_read_packet(int tapfd, uint8_t *buf, int maxlen) 180 { 181 return read(tapfd, buf, maxlen); 182 } 183 #endif 184 185 static void tap_send_completed(VLANClientState *nc, ssize_t len) 186 { 187 TAPState *s = DO_UPCAST(TAPState, nc, nc); 188 tap_read_poll(s, 1); 189 } 190 191 static void tap_send(void *opaque) 192 { 193 TAPState *s = opaque; 194 int size; 195 196 do { 197 uint8_t *buf = s->buf; 198 199 size = tap_read_packet(s->fd, s->buf, sizeof(s->buf)); 200 if (size <= 0) { 201 break; 202 } 203 204 if (s->host_vnet_hdr_len && !s->using_vnet_hdr) { 205 buf += s->host_vnet_hdr_len; 206 size -= s->host_vnet_hdr_len; 207 } 208 209 size = qemu_send_packet_async(&s->nc, buf, size, tap_send_completed); 210 if (size == 0) { 211 tap_read_poll(s, 0); 212 } 213 } while (size > 0 && qemu_can_send_packet(&s->nc)); 214 } 215 216 int tap_has_ufo(VLANClientState *nc) 217 { 218 TAPState *s = DO_UPCAST(TAPState, nc, nc); 219 220 assert(nc->info->type == NET_CLIENT_TYPE_TAP); 221 222 return s->has_ufo; 223 } 224 225 int tap_has_vnet_hdr(VLANClientState *nc) 226 { 227 TAPState *s = DO_UPCAST(TAPState, nc, nc); 228 229 assert(nc->info->type == NET_CLIENT_TYPE_TAP); 230 231 return !!s->host_vnet_hdr_len; 232 } 233 234 int tap_has_vnet_hdr_len(VLANClientState *nc, int len) 235 { 236 TAPState *s = DO_UPCAST(TAPState, nc, nc); 237 238 assert(nc->info->type == NET_CLIENT_TYPE_TAP); 239 240 return tap_probe_vnet_hdr_len(s->fd, len); 241 } 242 243 void tap_set_vnet_hdr_len(VLANClientState *nc, int len) 244 { 245 TAPState *s = DO_UPCAST(TAPState, nc, nc); 246 247 assert(nc->info->type == NET_CLIENT_TYPE_TAP); 248 assert(len == sizeof(struct virtio_net_hdr_mrg_rxbuf) || 249 len == sizeof(struct virtio_net_hdr)); 250 251 tap_fd_set_vnet_hdr_len(s->fd, len); 252 s->host_vnet_hdr_len = len; 253 } 254 255 void tap_using_vnet_hdr(VLANClientState *nc, int using_vnet_hdr) 256 { 257 TAPState *s = DO_UPCAST(TAPState, nc, nc); 258 259 using_vnet_hdr = using_vnet_hdr != 0; 260 261 assert(nc->info->type == NET_CLIENT_TYPE_TAP); 262 assert(!!s->host_vnet_hdr_len == using_vnet_hdr); 263 264 s->using_vnet_hdr = using_vnet_hdr; 265 } 266 267 void tap_set_offload(VLANClientState *nc, int csum, int tso4, 268 int tso6, int ecn, int ufo) 269 { 270 TAPState *s = DO_UPCAST(TAPState, nc, nc); 271 if (s->fd < 0) { 272 return; 273 } 274 275 tap_fd_set_offload(s->fd, csum, tso4, tso6, ecn, ufo); 276 } 277 278 static void tap_cleanup(VLANClientState *nc) 279 { 280 TAPState *s = DO_UPCAST(TAPState, nc, nc); 281 282 if (s->vhost_net) { 283 vhost_net_cleanup(s->vhost_net); 284 s->vhost_net = NULL; 285 } 286 287 qemu_purge_queued_packets(nc); 288 289 if (s->down_script[0]) 290 launch_script(s->down_script, s->down_script_arg, s->fd); 291 292 tap_read_poll(s, 0); 293 tap_write_poll(s, 0); 294 close(s->fd); 295 s->fd = -1; 296 } 297 298 static void tap_poll(VLANClientState *nc, bool enable) 299 { 300 TAPState *s = DO_UPCAST(TAPState, nc, nc); 301 tap_read_poll(s, enable); 302 tap_write_poll(s, enable); 303 } 304 305 int tap_get_fd(VLANClientState *nc) 306 { 307 TAPState *s = DO_UPCAST(TAPState, nc, nc); 308 assert(nc->info->type == NET_CLIENT_TYPE_TAP); 309 return s->fd; 310 } 311 312 /* fd support */ 313 314 static NetClientInfo net_tap_info = { 315 .type = NET_CLIENT_TYPE_TAP, 316 .size = sizeof(TAPState), 317 .receive = tap_receive, 318 .receive_raw = tap_receive_raw, 319 .receive_iov = tap_receive_iov, 320 .poll = tap_poll, 321 .cleanup = tap_cleanup, 322 }; 323 324 static TAPState *net_tap_fd_init(VLANState *vlan, 325 const char *model, 326 const char *name, 327 int fd, 328 int vnet_hdr) 329 { 330 VLANClientState *nc; 331 TAPState *s; 332 333 nc = qemu_new_net_client(&net_tap_info, vlan, NULL, model, name); 334 335 s = DO_UPCAST(TAPState, nc, nc); 336 337 s->fd = fd; 338 s->host_vnet_hdr_len = vnet_hdr ? sizeof(struct virtio_net_hdr) : 0; 339 s->using_vnet_hdr = 0; 340 s->has_ufo = tap_probe_has_ufo(s->fd); 341 tap_set_offload(&s->nc, 0, 0, 0, 0, 0); 342 tap_read_poll(s, 1); 343 s->vhost_net = NULL; 344 return s; 345 } 346 347 static int launch_script(const char *setup_script, const char *ifname, int fd) 348 { 349 int pid, status; 350 char *args[3]; 351 char **parg; 352 353 /* try to launch network script */ 354 pid = fork(); 355 if (pid == 0) { 356 int open_max = sysconf(_SC_OPEN_MAX), i; 357 358 for (i = 0; i < open_max; i++) { 359 if (i != STDIN_FILENO && 360 i != STDOUT_FILENO && 361 i != STDERR_FILENO && 362 i != fd) { 363 close(i); 364 } 365 } 366 parg = args; 367 *parg++ = (char *)setup_script; 368 *parg++ = (char *)ifname; 369 *parg = NULL; 370 execv(setup_script, args); 371 _exit(1); 372 } else if (pid > 0) { 373 while (waitpid(pid, &status, 0) != pid) { 374 /* loop */ 375 } 376 377 if (WIFEXITED(status) && WEXITSTATUS(status) == 0) { 378 return 0; 379 } 380 } 381 fprintf(stderr, "%s: could not launch network script\n", setup_script); 382 return -1; 383 } 384 385 static int recv_fd(int c) 386 { 387 int fd; 388 uint8_t msgbuf[CMSG_SPACE(sizeof(fd))]; 389 struct msghdr msg = { 390 .msg_control = msgbuf, 391 .msg_controllen = sizeof(msgbuf), 392 }; 393 struct cmsghdr *cmsg; 394 struct iovec iov; 395 uint8_t req[1]; 396 ssize_t len; 397 398 cmsg = CMSG_FIRSTHDR(&msg); 399 cmsg->cmsg_level = SOL_SOCKET; 400 cmsg->cmsg_type = SCM_RIGHTS; 401 cmsg->cmsg_len = CMSG_LEN(sizeof(fd)); 402 msg.msg_controllen = cmsg->cmsg_len; 403 404 iov.iov_base = req; 405 iov.iov_len = sizeof(req); 406 407 msg.msg_iov = &iov; 408 msg.msg_iovlen = 1; 409 410 len = recvmsg(c, &msg, 0); 411 if (len > 0) { 412 memcpy(&fd, CMSG_DATA(cmsg), sizeof(fd)); 413 return fd; 414 } 415 416 return len; 417 } 418 419 static int net_bridge_run_helper(const char *helper, const char *bridge) 420 { 421 sigset_t oldmask, mask; 422 int pid, status; 423 char *args[5]; 424 char **parg; 425 int sv[2]; 426 427 sigemptyset(&mask); 428 sigaddset(&mask, SIGCHLD); 429 sigprocmask(SIG_BLOCK, &mask, &oldmask); 430 431 if (socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) { 432 return -1; 433 } 434 435 /* try to launch bridge helper */ 436 pid = fork(); 437 if (pid == 0) { 438 int open_max = sysconf(_SC_OPEN_MAX), i; 439 char fd_buf[6+10]; 440 char br_buf[6+IFNAMSIZ] = {0}; 441 char helper_cmd[PATH_MAX + sizeof(fd_buf) + sizeof(br_buf) + 15]; 442 443 for (i = 0; i < open_max; i++) { 444 if (i != STDIN_FILENO && 445 i != STDOUT_FILENO && 446 i != STDERR_FILENO && 447 i != sv[1]) { 448 close(i); 449 } 450 } 451 452 snprintf(fd_buf, sizeof(fd_buf), "%s%d", "--fd=", sv[1]); 453 454 if (strrchr(helper, ' ') || strrchr(helper, '\t')) { 455 /* assume helper is a command */ 456 457 if (strstr(helper, "--br=") == NULL) { 458 snprintf(br_buf, sizeof(br_buf), "%s%s", "--br=", bridge); 459 } 460 461 snprintf(helper_cmd, sizeof(helper_cmd), "%s %s %s %s", 462 helper, "--use-vnet", fd_buf, br_buf); 463 464 parg = args; 465 *parg++ = (char *)"sh"; 466 *parg++ = (char *)"-c"; 467 *parg++ = helper_cmd; 468 *parg++ = NULL; 469 470 execv("/bin/sh", args); 471 } else { 472 /* assume helper is just the executable path name */ 473 474 snprintf(br_buf, sizeof(br_buf), "%s%s", "--br=", bridge); 475 476 parg = args; 477 *parg++ = (char *)helper; 478 *parg++ = (char *)"--use-vnet"; 479 *parg++ = fd_buf; 480 *parg++ = br_buf; 481 *parg++ = NULL; 482 483 execv(helper, args); 484 } 485 _exit(1); 486 487 } else if (pid > 0) { 488 int fd; 489 490 close(sv[1]); 491 492 do { 493 fd = recv_fd(sv[0]); 494 } while (fd == -1 && errno == EINTR); 495 496 close(sv[0]); 497 498 while (waitpid(pid, &status, 0) != pid) { 499 /* loop */ 500 } 501 sigprocmask(SIG_SETMASK, &oldmask, NULL); 502 if (fd < 0) { 503 fprintf(stderr, "failed to recv file descriptor\n"); 504 return -1; 505 } 506 507 if (WIFEXITED(status) && WEXITSTATUS(status) == 0) { 508 return fd; 509 } 510 } 511 fprintf(stderr, "failed to launch bridge helper\n"); 512 return -1; 513 } 514 515 int net_init_bridge(QemuOpts *opts, const char *name, VLANState *vlan) 516 { 517 TAPState *s; 518 int fd, vnet_hdr; 519 520 if (!qemu_opt_get(opts, "br")) { 521 qemu_opt_set(opts, "br", DEFAULT_BRIDGE_INTERFACE); 522 } 523 if (!qemu_opt_get(opts, "helper")) { 524 qemu_opt_set(opts, "helper", DEFAULT_BRIDGE_HELPER); 525 } 526 527 fd = net_bridge_run_helper(qemu_opt_get(opts, "helper"), 528 qemu_opt_get(opts, "br")); 529 if (fd == -1) { 530 return -1; 531 } 532 533 fcntl(fd, F_SETFL, O_NONBLOCK); 534 535 vnet_hdr = tap_probe_vnet_hdr(fd); 536 537 s = net_tap_fd_init(vlan, "bridge", name, fd, vnet_hdr); 538 if (!s) { 539 close(fd); 540 return -1; 541 } 542 543 snprintf(s->nc.info_str, sizeof(s->nc.info_str), "helper=%s,br=%s", 544 qemu_opt_get(opts, "helper"), qemu_opt_get(opts, "br")); 545 546 return 0; 547 } 548 549 static int net_tap_init(QemuOpts *opts, int *vnet_hdr) 550 { 551 int fd, vnet_hdr_required; 552 char ifname[128] = {0,}; 553 const char *setup_script; 554 555 if (qemu_opt_get(opts, "ifname")) { 556 pstrcpy(ifname, sizeof(ifname), qemu_opt_get(opts, "ifname")); 557 } 558 559 *vnet_hdr = qemu_opt_get_bool(opts, "vnet_hdr", 1); 560 if (qemu_opt_get(opts, "vnet_hdr")) { 561 vnet_hdr_required = *vnet_hdr; 562 } else { 563 vnet_hdr_required = 0; 564 } 565 566 TFR(fd = tap_open(ifname, sizeof(ifname), vnet_hdr, vnet_hdr_required)); 567 if (fd < 0) { 568 return -1; 569 } 570 571 setup_script = qemu_opt_get(opts, "script"); 572 if (setup_script && 573 setup_script[0] != '\0' && 574 strcmp(setup_script, "no") != 0 && 575 launch_script(setup_script, ifname, fd)) { 576 close(fd); 577 return -1; 578 } 579 580 qemu_opt_set(opts, "ifname", ifname); 581 582 return fd; 583 } 584 585 int net_init_tap(QemuOpts *opts, const char *name, VLANState *vlan) 586 { 587 TAPState *s; 588 int fd, vnet_hdr = 0; 589 const char *model; 590 591 if (qemu_opt_get(opts, "fd")) { 592 if (qemu_opt_get(opts, "ifname") || 593 qemu_opt_get(opts, "script") || 594 qemu_opt_get(opts, "downscript") || 595 qemu_opt_get(opts, "vnet_hdr") || 596 qemu_opt_get(opts, "helper")) { 597 error_report("ifname=, script=, downscript=, vnet_hdr=, " 598 "and helper= are invalid with fd="); 599 return -1; 600 } 601 602 fd = net_handle_fd_param(cur_mon, qemu_opt_get(opts, "fd")); 603 if (fd == -1) { 604 return -1; 605 } 606 607 fcntl(fd, F_SETFL, O_NONBLOCK); 608 609 vnet_hdr = tap_probe_vnet_hdr(fd); 610 611 model = "tap"; 612 613 } else if (qemu_opt_get(opts, "helper")) { 614 if (qemu_opt_get(opts, "ifname") || 615 qemu_opt_get(opts, "script") || 616 qemu_opt_get(opts, "downscript") || 617 qemu_opt_get(opts, "vnet_hdr")) { 618 error_report("ifname=, script=, downscript=, and vnet_hdr= " 619 "are invalid with helper="); 620 return -1; 621 } 622 623 fd = net_bridge_run_helper(qemu_opt_get(opts, "helper"), 624 DEFAULT_BRIDGE_INTERFACE); 625 if (fd == -1) { 626 return -1; 627 } 628 629 fcntl(fd, F_SETFL, O_NONBLOCK); 630 631 vnet_hdr = tap_probe_vnet_hdr(fd); 632 633 model = "bridge"; 634 635 } else { 636 if (!qemu_opt_get(opts, "script")) { 637 qemu_opt_set(opts, "script", DEFAULT_NETWORK_SCRIPT); 638 } 639 640 if (!qemu_opt_get(opts, "downscript")) { 641 qemu_opt_set(opts, "downscript", DEFAULT_NETWORK_DOWN_SCRIPT); 642 } 643 644 fd = net_tap_init(opts, &vnet_hdr); 645 if (fd == -1) { 646 return -1; 647 } 648 649 model = "tap"; 650 } 651 652 s = net_tap_fd_init(vlan, model, name, fd, vnet_hdr); 653 if (!s) { 654 close(fd); 655 return -1; 656 } 657 658 if (tap_set_sndbuf(s->fd, opts) < 0) { 659 return -1; 660 } 661 662 if (qemu_opt_get(opts, "fd")) { 663 snprintf(s->nc.info_str, sizeof(s->nc.info_str), "fd=%d", fd); 664 } else if (qemu_opt_get(opts, "helper")) { 665 snprintf(s->nc.info_str, sizeof(s->nc.info_str), 666 "helper=%s", qemu_opt_get(opts, "helper")); 667 } else { 668 const char *ifname, *script, *downscript; 669 670 ifname = qemu_opt_get(opts, "ifname"); 671 script = qemu_opt_get(opts, "script"); 672 downscript = qemu_opt_get(opts, "downscript"); 673 674 snprintf(s->nc.info_str, sizeof(s->nc.info_str), 675 "ifname=%s,script=%s,downscript=%s", 676 ifname, script, downscript); 677 678 if (strcmp(downscript, "no") != 0) { 679 snprintf(s->down_script, sizeof(s->down_script), "%s", downscript); 680 snprintf(s->down_script_arg, sizeof(s->down_script_arg), "%s", ifname); 681 } 682 } 683 684 if (qemu_opt_get_bool(opts, "vhost", !!qemu_opt_get(opts, "vhostfd") || 685 qemu_opt_get_bool(opts, "vhostforce", false))) { 686 int vhostfd, r; 687 bool force = qemu_opt_get_bool(opts, "vhostforce", false); 688 if (qemu_opt_get(opts, "vhostfd")) { 689 r = net_handle_fd_param(cur_mon, qemu_opt_get(opts, "vhostfd")); 690 if (r == -1) { 691 return -1; 692 } 693 vhostfd = r; 694 } else { 695 vhostfd = -1; 696 } 697 s->vhost_net = vhost_net_init(&s->nc, vhostfd, force); 698 if (!s->vhost_net) { 699 error_report("vhost-net requested but could not be initialized"); 700 return -1; 701 } 702 } else if (qemu_opt_get(opts, "vhostfd")) { 703 error_report("vhostfd= is not valid without vhost"); 704 return -1; 705 } 706 707 return 0; 708 } 709 710 VHostNetState *tap_get_vhost_net(VLANClientState *nc) 711 { 712 TAPState *s = DO_UPCAST(TAPState, nc, nc); 713 assert(nc->info->type == NET_CLIENT_TYPE_TAP); 714 return s->vhost_net; 715 } 716