1 /* $OpenBSD: config.c,v 1.58 2019/05/11 19:55:14 jasper Exp $ */ 2 3 /* 4 * Copyright (c) 2015 Reyk Floeter <reyk@openbsd.org> 5 * 6 * Permission to use, copy, modify, and distribute this software for any 7 * purpose with or without fee is hereby granted, provided that the above 8 * copyright notice and this permission notice appear in all copies. 9 * 10 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES 11 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF 12 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR 13 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES 14 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN 15 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF 16 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 17 */ 18 19 #include <sys/types.h> 20 #include <sys/queue.h> 21 #include <sys/time.h> 22 #include <sys/uio.h> 23 #include <sys/stat.h> 24 #include <sys/socket.h> 25 26 #include <net/if.h> 27 28 #include <stdio.h> 29 #include <stdlib.h> 30 #include <termios.h> 31 #include <unistd.h> 32 #include <limits.h> 33 #include <string.h> 34 #include <fcntl.h> 35 #include <util.h> 36 #include <errno.h> 37 #include <imsg.h> 38 39 #include "proc.h" 40 #include "vmd.h" 41 42 /* Supported bridge types */ 43 const char *vmd_descsw[] = { "switch", "bridge", NULL }; 44 45 static int config_init_localprefix(struct vmd_config *); 46 47 static int 48 config_init_localprefix(struct vmd_config *cfg) 49 { 50 struct sockaddr_in6 *sin6; 51 52 if (host(VMD_DHCP_PREFIX, &cfg->cfg_localprefix) == -1) 53 return (-1); 54 55 /* IPv6 is disabled by default */ 56 cfg->cfg_flags &= ~VMD_CFG_INET6; 57 58 /* Generate random IPv6 prefix only once */ 59 if (cfg->cfg_flags & VMD_CFG_AUTOINET6) 60 return (0); 61 if (host(VMD_ULA_PREFIX, &cfg->cfg_localprefix6) == -1) 62 return (-1); 63 /* Randomize the 56 bits "Global ID" and "Subnet ID" */ 64 sin6 = ss2sin6(&cfg->cfg_localprefix6.ss); 65 arc4random_buf(&sin6->sin6_addr.s6_addr[1], 7); 66 cfg->cfg_flags |= VMD_CFG_AUTOINET6; 67 68 return (0); 69 } 70 71 int 72 config_init(struct vmd *env) 73 { 74 struct privsep *ps = &env->vmd_ps; 75 unsigned int what; 76 77 /* Global configuration */ 78 ps->ps_what[PROC_PARENT] = CONFIG_ALL; 79 ps->ps_what[PROC_VMM] = CONFIG_VMS; 80 81 /* Local prefix */ 82 if (config_init_localprefix(&env->vmd_cfg) == -1) 83 return (-1); 84 85 /* Other configuration */ 86 what = ps->ps_what[privsep_process]; 87 if (what & CONFIG_VMS) { 88 if ((env->vmd_vms = calloc(1, sizeof(*env->vmd_vms))) == NULL) 89 return (-1); 90 if ((env->vmd_known = calloc(1, sizeof(*env->vmd_known))) == NULL) 91 return (-1); 92 TAILQ_INIT(env->vmd_vms); 93 TAILQ_INIT(env->vmd_known); 94 } 95 if (what & CONFIG_SWITCHES) { 96 if ((env->vmd_switches = calloc(1, 97 sizeof(*env->vmd_switches))) == NULL) 98 return (-1); 99 TAILQ_INIT(env->vmd_switches); 100 } 101 if (what & CONFIG_USERS) { 102 if ((env->vmd_users = calloc(1, 103 sizeof(*env->vmd_users))) == NULL) 104 return (-1); 105 TAILQ_INIT(env->vmd_users); 106 } 107 108 return (0); 109 } 110 111 void 112 config_purge(struct vmd *env, unsigned int reset) 113 { 114 struct privsep *ps = &env->vmd_ps; 115 struct name2id *n2i; 116 struct vmd_vm *vm; 117 struct vmd_switch *vsw; 118 unsigned int what; 119 120 DPRINTF("%s: %s purging vms and switches", 121 __func__, ps->ps_title[privsep_process]); 122 123 /* Reset global configuration (prefix was verified before) */ 124 config_init_localprefix(&env->vmd_cfg); 125 126 /* Reset other configuration */ 127 what = ps->ps_what[privsep_process] & reset; 128 if (what & CONFIG_VMS && env->vmd_vms != NULL) { 129 while ((vm = TAILQ_FIRST(env->vmd_vms)) != NULL) { 130 vm_remove(vm, __func__); 131 } 132 while ((n2i = TAILQ_FIRST(env->vmd_known)) != NULL) { 133 TAILQ_REMOVE(env->vmd_known, n2i, entry); 134 free(n2i); 135 } 136 env->vmd_nvm = 0; 137 } 138 if (what & CONFIG_SWITCHES && env->vmd_switches != NULL) { 139 while ((vsw = TAILQ_FIRST(env->vmd_switches)) != NULL) 140 switch_remove(vsw); 141 env->vmd_nswitches = 0; 142 } 143 } 144 145 int 146 config_setconfig(struct vmd *env) 147 { 148 struct privsep *ps = &env->vmd_ps; 149 unsigned int id; 150 151 DPRINTF("%s: setting config", __func__); 152 153 for (id = 0; id < PROC_MAX; id++) { 154 if (id == privsep_process) 155 continue; 156 proc_compose(ps, id, IMSG_VMDOP_CONFIG, &env->vmd_cfg, 157 sizeof(env->vmd_cfg)); 158 } 159 160 return (0); 161 } 162 163 int 164 config_getconfig(struct vmd *env, struct imsg *imsg) 165 { 166 struct privsep *ps = &env->vmd_ps; 167 168 log_debug("%s: %s retrieving config", 169 __func__, ps->ps_title[privsep_process]); 170 171 IMSG_SIZE_CHECK(imsg, &env->vmd_cfg); 172 memcpy(&env->vmd_cfg, imsg->data, sizeof(env->vmd_cfg)); 173 174 return (0); 175 } 176 177 int 178 config_setreset(struct vmd *env, unsigned int reset) 179 { 180 struct privsep *ps = &env->vmd_ps; 181 unsigned int id; 182 183 DPRINTF("%s: resetting state", __func__); 184 185 for (id = 0; id < PROC_MAX; id++) { 186 if ((reset & ps->ps_what[id]) == 0 || 187 id == privsep_process) 188 continue; 189 proc_compose(ps, id, IMSG_CTL_RESET, &reset, sizeof(reset)); 190 } 191 192 return (0); 193 } 194 195 int 196 config_getreset(struct vmd *env, struct imsg *imsg) 197 { 198 unsigned int mode; 199 200 IMSG_SIZE_CHECK(imsg, &mode); 201 memcpy(&mode, imsg->data, sizeof(mode)); 202 203 log_debug("%s: %s resetting state", 204 __func__, env->vmd_ps.ps_title[privsep_process]); 205 206 config_purge(env, mode); 207 208 return (0); 209 } 210 211 int 212 config_setvm(struct privsep *ps, struct vmd_vm *vm, uint32_t peerid, uid_t uid) 213 { 214 int diskfds[VMM_MAX_DISKS_PER_VM][VM_MAX_BASE_PER_DISK]; 215 struct vmd_if *vif; 216 struct vmop_create_params *vmc = &vm->vm_params; 217 struct vm_create_params *vcp = &vmc->vmc_params; 218 unsigned int i, j; 219 int fd = -1, vmboot = 0; 220 int kernfd = -1; 221 int *tapfds = NULL; 222 int cdromfd = -1; 223 int saved_errno = 0; 224 int n = 0, aflags, oflags; 225 char ifname[IF_NAMESIZE], *s; 226 char path[PATH_MAX]; 227 char base[PATH_MAX]; 228 unsigned int unit; 229 struct timeval tv, rate, since_last; 230 231 errno = 0; 232 233 if (vm->vm_state & VM_STATE_RUNNING) { 234 log_warnx("%s: vm is already running", __func__); 235 errno = EALREADY; 236 return (-1); 237 } 238 239 /* increase the user reference counter and check user limits */ 240 if (vm->vm_user != NULL && user_get(vm->vm_user->usr_id.uid) != NULL) { 241 user_inc(vcp, vm->vm_user, 1); 242 if (user_checklimit(vm->vm_user, vcp) == -1) { 243 errno = EPERM; 244 goto fail; 245 } 246 } 247 248 /* 249 * Rate-limit the VM so that it cannot restart in a loop: 250 * if the VM restarts after less than VM_START_RATE_SEC seconds, 251 * we increment the limit counter. After VM_START_RATE_LIMIT 252 * of suchs fast reboots the VM is stopped. 253 */ 254 getmonotime(&tv); 255 if (vm->vm_start_tv.tv_sec) { 256 timersub(&tv, &vm->vm_start_tv, &since_last); 257 258 rate.tv_sec = VM_START_RATE_SEC; 259 rate.tv_usec = 0; 260 if (timercmp(&since_last, &rate, <)) 261 vm->vm_start_limit++; 262 else { 263 /* Reset counter */ 264 vm->vm_start_limit = 0; 265 } 266 267 log_debug("%s: vm %u restarted after %lld.%ld seconds," 268 " limit %d/%d", __func__, vcp->vcp_id, since_last.tv_sec, 269 since_last.tv_usec, vm->vm_start_limit, 270 VM_START_RATE_LIMIT); 271 272 if (vm->vm_start_limit >= VM_START_RATE_LIMIT) { 273 log_warnx("%s: vm %u restarted too quickly", 274 __func__, vcp->vcp_id); 275 errno = EPERM; 276 goto fail; 277 } 278 } 279 vm->vm_start_tv = tv; 280 281 for (i = 0; i < VMM_MAX_DISKS_PER_VM; i++) 282 for (j = 0; j < VM_MAX_BASE_PER_DISK; j++) 283 diskfds[i][j] = -1; 284 285 tapfds = reallocarray(NULL, vcp->vcp_nnics, sizeof(*tapfds)); 286 if (tapfds == NULL) { 287 log_warn("%s: can't allocate tap fds", __func__); 288 goto fail; 289 } 290 for (i = 0; i < vcp->vcp_nnics; i++) 291 tapfds[i] = -1; 292 293 vm->vm_peerid = peerid; 294 vm->vm_uid = uid; 295 296 if (!(vm->vm_state & VM_STATE_RECEIVED)) { 297 if (strlen(vcp->vcp_kernel)) { 298 /* 299 * Boot kernel from disk image if path matches the 300 * root disk. 301 */ 302 if (vcp->vcp_ndisks && 303 strcmp(vcp->vcp_kernel, vcp->vcp_disks[0]) == 0) 304 vmboot = 1; 305 /* Open external kernel for child */ 306 else if ((kernfd = 307 open(vcp->vcp_kernel, O_RDONLY)) == -1) { 308 log_warn("%s: can't open kernel or BIOS " 309 "boot image %s", __func__, vcp->vcp_kernel); 310 goto fail; 311 } 312 } 313 314 /* 315 * Try to open the default BIOS image if no kernel/BIOS has been 316 * specified. The BIOS is an external firmware file that is 317 * typically distributed separately due to an incompatible 318 * license. 319 */ 320 if (kernfd == -1 && !vmboot && 321 (kernfd = open(VM_DEFAULT_BIOS, O_RDONLY)) == -1) { 322 log_warn("can't open %s", VM_DEFAULT_BIOS); 323 errno = VMD_BIOS_MISSING; 324 goto fail; 325 } 326 327 if (!vmboot && vm_checkaccess(kernfd, 328 vmc->vmc_checkaccess & VMOP_CREATE_KERNEL, 329 uid, R_OK) == -1) { 330 log_warnx("vm \"%s\" no read access to kernel %s", 331 vcp->vcp_name, vcp->vcp_kernel); 332 errno = EPERM; 333 goto fail; 334 } 335 } 336 337 /* Open CDROM image for child */ 338 if (strlen(vcp->vcp_cdrom)) { 339 /* Stat cdrom to ensure it is a regular file */ 340 if ((cdromfd = 341 open(vcp->vcp_cdrom, O_RDONLY)) == -1) { 342 log_warn("can't open cdrom %s", vcp->vcp_cdrom); 343 errno = VMD_CDROM_MISSING; 344 goto fail; 345 } 346 347 if (vm_checkaccess(cdromfd, 348 vmc->vmc_checkaccess & VMOP_CREATE_CDROM, 349 uid, R_OK) == -1) { 350 log_warnx("vm \"%s\" no read access to cdrom %s", 351 vcp->vcp_name, vcp->vcp_cdrom); 352 errno = EPERM; 353 goto fail; 354 } 355 } 356 357 /* Open disk images for child */ 358 for (i = 0 ; i < vcp->vcp_ndisks; i++) { 359 if (strlcpy(path, vcp->vcp_disks[i], sizeof(path)) 360 >= sizeof(path)) 361 log_warnx("disk path %s too long", vcp->vcp_disks[i]); 362 memset(vmc->vmc_diskbases, 0, sizeof(vmc->vmc_diskbases)); 363 oflags = O_RDWR|O_EXLOCK|O_NONBLOCK; 364 aflags = R_OK|W_OK; 365 for (j = 0; j < VM_MAX_BASE_PER_DISK; j++) { 366 /* Stat disk[i] to ensure it is a regular file */ 367 if ((diskfds[i][j] = open(path, oflags)) == -1) { 368 log_warn("can't open disk %s", 369 vcp->vcp_disks[i]); 370 errno = VMD_DISK_MISSING; 371 goto fail; 372 } 373 374 if (vm_checkaccess(diskfds[i][j], 375 vmc->vmc_checkaccess & VMOP_CREATE_DISK, 376 uid, aflags) == -1) { 377 log_warnx("vm \"%s\" unable to access " 378 "disk %s", vcp->vcp_name, path); 379 errno = EPERM; 380 goto fail; 381 } 382 383 /* 384 * Clear the write and exclusive flags for base images. 385 * All writes should go to the top image, allowing them 386 * to be shared. 387 */ 388 oflags = O_RDONLY|O_NONBLOCK; 389 aflags = R_OK; 390 n = virtio_get_base(diskfds[i][j], base, sizeof(base), 391 vmc->vmc_disktypes[i], path); 392 if (n == 0) 393 break; 394 if (n == -1) { 395 log_warnx("vm \"%s\" unable to read " 396 "base %s for disk %s", vcp->vcp_name, 397 base, vcp->vcp_disks[i]); 398 goto fail; 399 } 400 (void)strlcpy(path, base, sizeof(path)); 401 } 402 } 403 404 /* Open network interfaces */ 405 for (i = 0 ; i < vcp->vcp_nnics; i++) { 406 vif = &vm->vm_ifs[i]; 407 408 /* Check if the user has requested a specific tap(4) */ 409 s = vmc->vmc_ifnames[i]; 410 if (*s != '\0' && strcmp("tap", s) != 0) { 411 if (priv_getiftype(s, ifname, &unit) == -1 || 412 strcmp(ifname, "tap") != 0) { 413 log_warnx("%s: invalid tap name %s", 414 __func__, s); 415 errno = EINVAL; 416 goto fail; 417 } 418 } else 419 s = NULL; 420 421 /* 422 * Either open the requested tap(4) device or get 423 * the next available one. 424 */ 425 if (s != NULL) { 426 snprintf(path, PATH_MAX, "/dev/%s", s); 427 tapfds[i] = open(path, O_RDWR | O_NONBLOCK); 428 } else { 429 tapfds[i] = opentap(ifname); 430 s = ifname; 431 } 432 if (tapfds[i] == -1) { 433 log_warn("%s: can't open tap %s", __func__, s); 434 goto fail; 435 } 436 if ((vif->vif_name = strdup(s)) == NULL) { 437 log_warn("%s: can't save tap %s", __func__, s); 438 goto fail; 439 } 440 441 /* Check if the the interface is attached to a switch */ 442 s = vmc->vmc_ifswitch[i]; 443 if (*s != '\0') { 444 if ((vif->vif_switch = strdup(s)) == NULL) { 445 log_warn("%s: can't save switch %s", 446 __func__, s); 447 goto fail; 448 } 449 } 450 451 /* Check if the the interface is assigned to a group */ 452 s = vmc->vmc_ifgroup[i]; 453 if (*s != '\0') { 454 if ((vif->vif_group = strdup(s)) == NULL) { 455 log_warn("%s: can't save group %s", 456 __func__, s); 457 goto fail; 458 } 459 } 460 461 /* non-default rdomain (requires VMIFF_RDOMAIN below) */ 462 vif->vif_rdomain = vmc->vmc_ifrdomain[i]; 463 464 /* Set the interface status */ 465 vif->vif_flags = 466 vmc->vmc_ifflags[i] & (VMIFF_UP|VMIFF_OPTMASK); 467 } 468 469 /* Open TTY */ 470 if (vm->vm_ttyname == NULL) { 471 if (vm_opentty(vm) == -1) { 472 log_warn("%s: can't open tty %s", __func__, 473 vm->vm_ttyname == NULL ? "" : vm->vm_ttyname); 474 goto fail; 475 } 476 } 477 if ((fd = dup(vm->vm_tty)) == -1) { 478 log_warn("%s: can't re-open tty %s", __func__, vm->vm_ttyname); 479 goto fail; 480 } 481 482 /* Send VM information */ 483 if (vm->vm_state & VM_STATE_RECEIVED) 484 proc_compose_imsg(ps, PROC_VMM, -1, 485 IMSG_VMDOP_RECEIVE_VM_REQUEST, vm->vm_vmid, fd, vmc, 486 sizeof(struct vmop_create_params)); 487 else 488 proc_compose_imsg(ps, PROC_VMM, -1, 489 IMSG_VMDOP_START_VM_REQUEST, vm->vm_vmid, kernfd, 490 vmc, sizeof(*vmc)); 491 492 if (strlen(vcp->vcp_cdrom)) 493 proc_compose_imsg(ps, PROC_VMM, -1, 494 IMSG_VMDOP_START_VM_CDROM, vm->vm_vmid, cdromfd, 495 NULL, 0); 496 497 for (i = 0; i < vcp->vcp_ndisks; i++) { 498 for (j = 0; j < VM_MAX_BASE_PER_DISK; j++) { 499 if (diskfds[i][j] == -1) 500 break; 501 proc_compose_imsg(ps, PROC_VMM, -1, 502 IMSG_VMDOP_START_VM_DISK, vm->vm_vmid, 503 diskfds[i][j], &i, sizeof(i)); 504 } 505 } 506 for (i = 0; i < vcp->vcp_nnics; i++) { 507 proc_compose_imsg(ps, PROC_VMM, -1, 508 IMSG_VMDOP_START_VM_IF, vm->vm_vmid, tapfds[i], 509 &i, sizeof(i)); 510 } 511 512 if (!(vm->vm_state & VM_STATE_RECEIVED)) 513 proc_compose_imsg(ps, PROC_VMM, -1, 514 IMSG_VMDOP_START_VM_END, vm->vm_vmid, fd, NULL, 0); 515 516 free(tapfds); 517 518 vm->vm_state |= VM_STATE_RUNNING; 519 return (0); 520 521 fail: 522 saved_errno = errno; 523 log_warnx("failed to start vm %s", vcp->vcp_name); 524 525 if (kernfd != -1) 526 close(kernfd); 527 if (cdromfd != -1) 528 close(cdromfd); 529 for (i = 0; i < vcp->vcp_ndisks; i++) 530 for (j = 0; j < VM_MAX_BASE_PER_DISK; j++) 531 if (diskfds[i][j] != -1) 532 close(diskfds[i][j]); 533 if (tapfds != NULL) { 534 for (i = 0; i < vcp->vcp_nnics; i++) 535 close(tapfds[i]); 536 free(tapfds); 537 } 538 539 if (vm->vm_from_config) { 540 vm_stop(vm, 0, __func__); 541 } else { 542 vm_remove(vm, __func__); 543 } 544 errno = saved_errno; 545 if (errno == 0) 546 errno = EINVAL; 547 return (-1); 548 } 549 550 int 551 config_getvm(struct privsep *ps, struct imsg *imsg) 552 { 553 struct vmop_create_params vmc; 554 struct vmd_vm *vm; 555 556 IMSG_SIZE_CHECK(imsg, &vmc); 557 memcpy(&vmc, imsg->data, sizeof(vmc)); 558 559 errno = 0; 560 if (vm_register(ps, &vmc, &vm, imsg->hdr.peerid, 0) == -1) 561 goto fail; 562 563 /* If the fd is -1, the kernel will be searched on the disk */ 564 vm->vm_kernel = imsg->fd; 565 vm->vm_state |= VM_STATE_RUNNING; 566 vm->vm_peerid = (uint32_t)-1; 567 568 return (0); 569 570 fail: 571 if (imsg->fd != -1) { 572 close(imsg->fd); 573 imsg->fd = -1; 574 } 575 576 vm_remove(vm, __func__); 577 if (errno == 0) 578 errno = EINVAL; 579 580 return (-1); 581 } 582 583 int 584 config_getdisk(struct privsep *ps, struct imsg *imsg) 585 { 586 struct vmd_vm *vm; 587 unsigned int n, idx; 588 589 errno = 0; 590 if ((vm = vm_getbyvmid(imsg->hdr.peerid)) == NULL) { 591 errno = ENOENT; 592 return (-1); 593 } 594 595 IMSG_SIZE_CHECK(imsg, &n); 596 memcpy(&n, imsg->data, sizeof(n)); 597 598 if (n >= vm->vm_params.vmc_params.vcp_ndisks || imsg->fd == -1) { 599 log_warnx("invalid disk id"); 600 errno = EINVAL; 601 return (-1); 602 } 603 idx = vm->vm_params.vmc_diskbases[n]++; 604 if (idx >= VM_MAX_BASE_PER_DISK) { 605 log_warnx("too many bases for disk"); 606 errno = EINVAL; 607 return (-1); 608 } 609 vm->vm_disks[n][idx] = imsg->fd; 610 return (0); 611 } 612 613 int 614 config_getif(struct privsep *ps, struct imsg *imsg) 615 { 616 struct vmd_vm *vm; 617 unsigned int n; 618 619 errno = 0; 620 if ((vm = vm_getbyvmid(imsg->hdr.peerid)) == NULL) { 621 errno = ENOENT; 622 return (-1); 623 } 624 625 IMSG_SIZE_CHECK(imsg, &n); 626 memcpy(&n, imsg->data, sizeof(n)); 627 if (n >= vm->vm_params.vmc_params.vcp_nnics || 628 vm->vm_ifs[n].vif_fd != -1 || imsg->fd == -1) { 629 log_warnx("invalid interface id"); 630 goto fail; 631 } 632 vm->vm_ifs[n].vif_fd = imsg->fd; 633 return (0); 634 fail: 635 if (imsg->fd != -1) 636 close(imsg->fd); 637 errno = EINVAL; 638 return (-1); 639 } 640 641 int 642 config_getcdrom(struct privsep *ps, struct imsg *imsg) 643 { 644 struct vmd_vm *vm; 645 646 errno = 0; 647 if ((vm = vm_getbyvmid(imsg->hdr.peerid)) == NULL) { 648 errno = ENOENT; 649 return (-1); 650 } 651 652 if (imsg->fd == -1) { 653 log_warnx("invalid cdrom id"); 654 goto fail; 655 } 656 657 vm->vm_cdrom = imsg->fd; 658 return (0); 659 fail: 660 if (imsg->fd != -1) 661 close(imsg->fd); 662 errno = EINVAL; 663 return (-1); 664 } 665