vfs_mountroot.c revision 293742
160786Sps/*- 260786Sps * Copyright (c) 2010 Marcel Moolenaar 360786Sps * Copyright (c) 1999-2004 Poul-Henning Kamp 460786Sps * Copyright (c) 1999 Michael Smith 560786Sps * Copyright (c) 1989, 1993 660786Sps * The Regents of the University of California. All rights reserved. 760786Sps * (c) UNIX System Laboratories, Inc. 860786Sps * All or some portions of this file are derived from material licensed 960786Sps * to the University of California by American Telephone and Telegraph 1060786Sps * Co. or Unix System Laboratories, Inc. and are reproduced herein with 1160786Sps * the permission of UNIX System Laboratories, Inc. 1260786Sps * 1360786Sps * Redistribution and use in source and binary forms, with or without 1460786Sps * modification, are permitted provided that the following conditions 1560786Sps * are met: 1660786Sps * 1. Redistributions of source code must retain the above copyright 1760786Sps * notice, this list of conditions and the following disclaimer. 1860786Sps * 2. Redistributions in binary form must reproduce the above copyright 1960786Sps * notice, this list of conditions and the following disclaimer in the 2060786Sps * documentation and/or other materials provided with the distribution. 2160786Sps * 4. Neither the name of the University nor the names of its contributors 2260786Sps * may be used to endorse or promote products derived from this software 2360786Sps * without specific prior written permission. 2460786Sps * 25128345Stjr * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 26128345Stjr * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 2760786Sps * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 2860786Sps * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 2960786Sps * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 3060786Sps * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 3160786Sps * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 3260786Sps * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 3360786Sps * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 3460786Sps * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 3560786Sps * SUCH DAMAGE. 3660786Sps */ 3760786Sps 3860786Sps#include "opt_rootdevname.h" 3960786Sps 4060786Sps#include <sys/cdefs.h> 4160786Sps__FBSDID("$FreeBSD: stable/10/sys/kern/vfs_mountroot.c 293742 2016-01-12 10:09:03Z trasz $"); 4260786Sps 4360786Sps#include <sys/param.h> 4460786Sps#include <sys/conf.h> 4560786Sps#include <sys/cons.h> 4660786Sps#include <sys/fcntl.h> 4760786Sps#include <sys/jail.h> 4860786Sps#include <sys/kernel.h> 4960786Sps#include <sys/malloc.h> 5060786Sps#include <sys/mdioctl.h> 5160786Sps#include <sys/mount.h> 5260786Sps#include <sys/mutex.h> 5360786Sps#include <sys/namei.h> 5460786Sps#include <sys/priv.h> 5560786Sps#include <sys/proc.h> 5660786Sps#include <sys/filedesc.h> 5760786Sps#include <sys/reboot.h> 5889019Sps#include <sys/sbuf.h> 5989019Sps#include <sys/stat.h> 6060786Sps#include <sys/syscallsubr.h> 6160786Sps#include <sys/sysproto.h> 6260786Sps#include <sys/sx.h> 6360786Sps#include <sys/sysctl.h> 6460786Sps#include <sys/sysent.h> 6560786Sps#include <sys/systm.h> 6660786Sps#include <sys/vnode.h> 6760786Sps 6860786Sps#include <geom/geom.h> 6960786Sps 7060786Sps/* 7160786Sps * The root filesystem is detailed in the kernel environment variable 7260786Sps * vfs.root.mountfrom, which is expected to be in the general format 7360786Sps * 7460786Sps * <vfsname>:[<path>][ <vfsname>:[<path>] ...] 7560786Sps * vfsname := the name of a VFS known to the kernel and capable 7660786Sps * of being mounted as root 7760786Sps * path := disk device name or other data used by the filesystem 7860786Sps * to locate its physical store 7960786Sps * 8060786Sps * If the environment variable vfs.root.mountfrom is a space separated list, 8160786Sps * each list element is tried in turn and the root filesystem will be mounted 8260786Sps * from the first one that suceeds. 8360786Sps * 8460786Sps * The environment variable vfs.root.mountfrom.options is a comma delimited 8560786Sps * set of string mount options. These mount options must be parseable 8660786Sps * by nmount() in the kernel. 8760786Sps */ 8860786Sps 8960786Spsstatic int parse_mount(char **); 9060786Spsstatic struct mntarg *parse_mountroot_options(struct mntarg *, const char *); 9160786Sps 9260786Sps/* 9360786Sps * The vnode of the system's root (/ in the filesystem, without chroot 9460786Sps * active.) 9560786Sps */ 9660786Spsstruct vnode *rootvnode; 9760786Sps 9860786Sps/* 9960786Sps * Mount of the system's /dev. 10060786Sps */ 10160786Spsstruct mount *rootdevmp; 10260786Sps 10360786Spschar *rootdevnames[2] = {NULL, NULL}; 10460786Sps 10560786Spsstruct mtx root_holds_mtx; 10660786SpsMTX_SYSINIT(root_holds, &root_holds_mtx, "root_holds", MTX_DEF); 10760786Sps 10860786Spsstruct root_hold_token { 10960786Sps const char *who; 11060786Sps LIST_ENTRY(root_hold_token) list; 11160786Sps}; 11260786Sps 11360786Spsstatic LIST_HEAD(, root_hold_token) root_holds = 11460786Sps LIST_HEAD_INITIALIZER(root_holds); 11560786Sps 11660786Spsenum action { 11760786Sps A_CONTINUE, 11860786Sps A_PANIC, 11960786Sps A_REBOOT, 12060786Sps A_RETRY 121170256Sdelphij}; 122170256Sdelphij 12360786Spsstatic enum action root_mount_onfail = A_CONTINUE; 12460786Sps 12560786Spsstatic int root_mount_mddev; 12660786Spsstatic int root_mount_complete; 12760786Sps 12860786Sps/* By default wait up to 3 seconds for devices to appear. */ 12960786Spsstatic int root_mount_timeout = 3; 13060786SpsTUNABLE_INT("vfs.mountroot.timeout", &root_mount_timeout); 13160786Sps 13260786Spsstruct root_hold_token * 13360786Spsroot_mount_hold(const char *identifier) 13460786Sps{ 13560786Sps struct root_hold_token *h; 13660786Sps 13760786Sps if (root_mounted()) 13860786Sps return (NULL); 13960786Sps 14060786Sps h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK); 14160786Sps h->who = identifier; 14260786Sps mtx_lock(&root_holds_mtx); 14360786Sps LIST_INSERT_HEAD(&root_holds, h, list); 14460786Sps mtx_unlock(&root_holds_mtx); 14563128Sps return (h); 14663128Sps} 14760786Sps 14860786Spsvoid 149128345Stjrroot_mount_rel(struct root_hold_token *h) 150128345Stjr{ 15160786Sps 15260786Sps if (h == NULL) 15360786Sps return; 154170256Sdelphij mtx_lock(&root_holds_mtx); 15560786Sps LIST_REMOVE(h, list); 15660786Sps wakeup(&root_holds); 15760786Sps mtx_unlock(&root_holds_mtx); 15860786Sps free(h, M_DEVBUF); 15960786Sps} 16060786Sps 16160786Spsint 16260786Spsroot_mounted(void) 16360786Sps{ 16460786Sps 16560786Sps /* No mutex is acquired here because int stores are atomic. */ 16660786Sps return (root_mount_complete); 16760786Sps} 16860786Sps 16960786Spsvoid 17060786Spsroot_mount_wait(void) 17160786Sps{ 17260786Sps 17360786Sps /* 17460786Sps * Panic on an obvious deadlock - the function can't be called from 17560786Sps * a thread which is doing the whole SYSINIT stuff. 17660786Sps */ 17760786Sps KASSERT(curthread->td_proc->p_pid != 0, 17860786Sps ("root_mount_wait: cannot be called from the swapper thread")); 17960786Sps mtx_lock(&root_holds_mtx); 18060786Sps while (!root_mount_complete) { 18160786Sps msleep(&root_mount_complete, &root_holds_mtx, PZERO, "rootwait", 18260786Sps hz); 18389019Sps } 18460786Sps mtx_unlock(&root_holds_mtx); 18560786Sps} 18660786Sps 18789019Spsstatic void 18889019Spsset_rootvnode(void) 18960786Sps{ 19060786Sps struct proc *p; 19160786Sps 19260786Sps if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode)) 19360786Sps panic("Cannot find root vnode"); 19460786Sps 19560786Sps VOP_UNLOCK(rootvnode, 0); 19660786Sps 19763128Sps p = curthread->td_proc; 19863128Sps FILEDESC_XLOCK(p->p_fd); 19963128Sps 20060786Sps if (p->p_fd->fd_cdir != NULL) 20160786Sps vrele(p->p_fd->fd_cdir); 20260786Sps p->p_fd->fd_cdir = rootvnode; 20360786Sps VREF(rootvnode); 20460786Sps 20560786Sps if (p->p_fd->fd_rdir != NULL) 20660786Sps vrele(p->p_fd->fd_rdir); 20760786Sps p->p_fd->fd_rdir = rootvnode; 20860786Sps VREF(rootvnode); 20960786Sps 21060786Sps FILEDESC_XUNLOCK(p->p_fd); 21160786Sps} 21260786Sps 21360786Spsstatic int 21460786Spsvfs_mountroot_devfs(struct thread *td, struct mount **mpp) 21560786Sps{ 21660786Sps struct vfsoptlist *opts; 21760786Sps struct vfsconf *vfsp; 21860786Sps struct mount *mp; 21960786Sps int error; 22060786Sps 22160786Sps *mpp = NULL; 22260786Sps 22360786Sps vfsp = vfs_byname("devfs"); 224170256Sdelphij KASSERT(vfsp != NULL, ("Could not find devfs by name")); 22560786Sps if (vfsp == NULL) 22660786Sps return (ENOENT); 227 228 mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td->td_ucred); 229 230 error = VFS_MOUNT(mp); 231 KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error)); 232 if (error) 233 return (error); 234 235 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 236 TAILQ_INIT(opts); 237 mp->mnt_opt = opts; 238 239 mtx_lock(&mountlist_mtx); 240 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list); 241 mtx_unlock(&mountlist_mtx); 242 243 *mpp = mp; 244 rootdevmp = mp; 245 set_rootvnode(); 246 247 error = kern_symlink(td, "/", "dev", UIO_SYSSPACE); 248 if (error) 249 printf("kern_symlink /dev -> / returns %d\n", error); 250 251 return (error); 252} 253 254static void 255vfs_mountroot_shuffle(struct thread *td, struct mount *mpdevfs) 256{ 257 struct nameidata nd; 258 struct mount *mporoot, *mpnroot; 259 struct vnode *vp, *vporoot, *vpdevfs; 260 char *fspath; 261 int error; 262 263 mpnroot = TAILQ_NEXT(mpdevfs, mnt_list); 264 265 /* Shuffle the mountlist. */ 266 mtx_lock(&mountlist_mtx); 267 mporoot = TAILQ_FIRST(&mountlist); 268 TAILQ_REMOVE(&mountlist, mpdevfs, mnt_list); 269 if (mporoot != mpdevfs) { 270 TAILQ_REMOVE(&mountlist, mpnroot, mnt_list); 271 TAILQ_INSERT_HEAD(&mountlist, mpnroot, mnt_list); 272 } 273 TAILQ_INSERT_TAIL(&mountlist, mpdevfs, mnt_list); 274 mtx_unlock(&mountlist_mtx); 275 276 cache_purgevfs(mporoot); 277 if (mporoot != mpdevfs) 278 cache_purgevfs(mpdevfs); 279 280 VFS_ROOT(mporoot, LK_EXCLUSIVE, &vporoot); 281 282 VI_LOCK(vporoot); 283 vporoot->v_iflag &= ~VI_MOUNT; 284 VI_UNLOCK(vporoot); 285 vporoot->v_mountedhere = NULL; 286 mporoot->mnt_flag &= ~MNT_ROOTFS; 287 mporoot->mnt_vnodecovered = NULL; 288 vput(vporoot); 289 290 /* Set up the new rootvnode, and purge the cache */ 291 mpnroot->mnt_vnodecovered = NULL; 292 set_rootvnode(); 293 cache_purgevfs(rootvnode->v_mount); 294 295 if (mporoot != mpdevfs) { 296 /* Remount old root under /.mount or /mnt */ 297 fspath = "/.mount"; 298 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, 299 fspath, td); 300 error = namei(&nd); 301 if (error) { 302 NDFREE(&nd, NDF_ONLY_PNBUF); 303 fspath = "/mnt"; 304 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, 305 fspath, td); 306 error = namei(&nd); 307 } 308 if (!error) { 309 vp = nd.ni_vp; 310 error = (vp->v_type == VDIR) ? 0 : ENOTDIR; 311 if (!error) 312 error = vinvalbuf(vp, V_SAVE, 0, 0); 313 if (!error) { 314 cache_purge(vp); 315 mporoot->mnt_vnodecovered = vp; 316 vp->v_mountedhere = mporoot; 317 strlcpy(mporoot->mnt_stat.f_mntonname, 318 fspath, MNAMELEN); 319 VOP_UNLOCK(vp, 0); 320 } else 321 vput(vp); 322 } 323 NDFREE(&nd, NDF_ONLY_PNBUF); 324 325 if (error && bootverbose) 326 printf("mountroot: unable to remount previous root " 327 "under /.mount or /mnt (error %d).\n", error); 328 } 329 330 /* Remount devfs under /dev */ 331 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td); 332 error = namei(&nd); 333 if (!error) { 334 vp = nd.ni_vp; 335 error = (vp->v_type == VDIR) ? 0 : ENOTDIR; 336 if (!error) 337 error = vinvalbuf(vp, V_SAVE, 0, 0); 338 if (!error) { 339 vpdevfs = mpdevfs->mnt_vnodecovered; 340 if (vpdevfs != NULL) { 341 cache_purge(vpdevfs); 342 vpdevfs->v_mountedhere = NULL; 343 vrele(vpdevfs); 344 } 345 mpdevfs->mnt_vnodecovered = vp; 346 vp->v_mountedhere = mpdevfs; 347 VOP_UNLOCK(vp, 0); 348 } else 349 vput(vp); 350 } 351 if (error && bootverbose) 352 printf("mountroot: unable to remount devfs under /dev " 353 "(error %d).\n", error); 354 NDFREE(&nd, NDF_ONLY_PNBUF); 355 356 if (mporoot == mpdevfs) { 357 vfs_unbusy(mpdevfs); 358 /* Unlink the no longer needed /dev/dev -> / symlink */ 359 error = kern_unlink(td, "/dev/dev", UIO_SYSSPACE); 360 if (error && bootverbose) 361 printf("mountroot: unable to unlink /dev/dev " 362 "(error %d)\n", error); 363 } 364} 365 366/* 367 * Configuration parser. 368 */ 369 370/* Parser character classes. */ 371#define CC_WHITESPACE -1 372#define CC_NONWHITESPACE -2 373 374/* Parse errors. */ 375#define PE_EOF -1 376#define PE_EOL -2 377 378static __inline int 379parse_peek(char **conf) 380{ 381 382 return (**conf); 383} 384 385static __inline void 386parse_poke(char **conf, int c) 387{ 388 389 **conf = c; 390} 391 392static __inline void 393parse_advance(char **conf) 394{ 395 396 (*conf)++; 397} 398 399static int 400parse_skipto(char **conf, int mc) 401{ 402 int c, match; 403 404 while (1) { 405 c = parse_peek(conf); 406 if (c == 0) 407 return (PE_EOF); 408 switch (mc) { 409 case CC_WHITESPACE: 410 match = (c == ' ' || c == '\t' || c == '\n') ? 1 : 0; 411 break; 412 case CC_NONWHITESPACE: 413 if (c == '\n') 414 return (PE_EOL); 415 match = (c != ' ' && c != '\t') ? 1 : 0; 416 break; 417 default: 418 match = (c == mc) ? 1 : 0; 419 break; 420 } 421 if (match) 422 break; 423 parse_advance(conf); 424 } 425 return (0); 426} 427 428static int 429parse_token(char **conf, char **tok) 430{ 431 char *p; 432 size_t len; 433 int error; 434 435 *tok = NULL; 436 error = parse_skipto(conf, CC_NONWHITESPACE); 437 if (error) 438 return (error); 439 p = *conf; 440 error = parse_skipto(conf, CC_WHITESPACE); 441 len = *conf - p; 442 *tok = malloc(len + 1, M_TEMP, M_WAITOK | M_ZERO); 443 bcopy(p, *tok, len); 444 return (0); 445} 446 447static void 448parse_dir_ask_printenv(const char *var) 449{ 450 char *val; 451 452 val = getenv(var); 453 if (val != NULL) { 454 printf(" %s=%s\n", var, val); 455 freeenv(val); 456 } 457} 458 459static int 460parse_dir_ask(char **conf) 461{ 462 char name[80]; 463 char *mnt; 464 int error; 465 466 printf("\nLoader variables:\n"); 467 parse_dir_ask_printenv("vfs.root.mountfrom"); 468 parse_dir_ask_printenv("vfs.root.mountfrom.options"); 469 470 printf("\nManual root filesystem specification:\n"); 471 printf(" <fstype>:<device> [options]\n"); 472 printf(" Mount <device> using filesystem <fstype>\n"); 473 printf(" and with the specified (optional) option list.\n"); 474 printf("\n"); 475 printf(" eg. ufs:/dev/da0s1a\n"); 476 printf(" zfs:tank\n"); 477 printf(" cd9660:/dev/acd0 ro\n"); 478 printf(" (which is equivalent to: "); 479 printf("mount -t cd9660 -o ro /dev/acd0 /)\n"); 480 printf("\n"); 481 printf(" ? List valid disk boot devices\n"); 482 printf(" . Yield 1 second (for background tasks)\n"); 483 printf(" <empty line> Abort manual input\n"); 484 485 do { 486 error = EINVAL; 487 printf("\nmountroot> "); 488 cngets(name, sizeof(name), GETS_ECHO); 489 if (name[0] == '\0') 490 break; 491 if (name[0] == '?' && name[1] == '\0') { 492 printf("\nList of GEOM managed disk devices:\n "); 493 g_dev_print(); 494 continue; 495 } 496 if (name[0] == '.' && name[1] == '\0') { 497 pause("rmask", hz); 498 continue; 499 } 500 mnt = name; 501 error = parse_mount(&mnt); 502 if (error == -1) 503 printf("Invalid file system specification.\n"); 504 } while (error != 0); 505 506 return (error); 507} 508 509static int 510parse_dir_md(char **conf) 511{ 512 struct stat sb; 513 struct thread *td; 514 struct md_ioctl *mdio; 515 char *path, *tok; 516 int error, fd, len; 517 518 td = curthread; 519 520 error = parse_token(conf, &tok); 521 if (error) 522 return (error); 523 524 len = strlen(tok); 525 mdio = malloc(sizeof(*mdio) + len + 1, M_TEMP, M_WAITOK | M_ZERO); 526 path = (void *)(mdio + 1); 527 bcopy(tok, path, len); 528 free(tok, M_TEMP); 529 530 /* Get file status. */ 531 error = kern_stat(td, path, UIO_SYSSPACE, &sb); 532 if (error) 533 goto out; 534 535 /* Open /dev/mdctl so that we can attach/detach. */ 536 error = kern_open(td, "/dev/" MDCTL_NAME, UIO_SYSSPACE, O_RDWR, 0); 537 if (error) 538 goto out; 539 540 fd = td->td_retval[0]; 541 mdio->md_version = MDIOVERSION; 542 mdio->md_type = MD_VNODE; 543 544 if (root_mount_mddev != -1) { 545 mdio->md_unit = root_mount_mddev; 546 DROP_GIANT(); 547 error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio); 548 PICKUP_GIANT(); 549 /* Ignore errors. We don't care. */ 550 root_mount_mddev = -1; 551 } 552 553 mdio->md_file = (void *)(mdio + 1); 554 mdio->md_options = MD_AUTOUNIT | MD_READONLY; 555 mdio->md_mediasize = sb.st_size; 556 mdio->md_unit = 0; 557 DROP_GIANT(); 558 error = kern_ioctl(td, fd, MDIOCATTACH, (void *)mdio); 559 PICKUP_GIANT(); 560 if (error) 561 goto out; 562 563 if (mdio->md_unit > 9) { 564 printf("rootmount: too many md units\n"); 565 mdio->md_file = NULL; 566 mdio->md_options = 0; 567 mdio->md_mediasize = 0; 568 DROP_GIANT(); 569 error = kern_ioctl(td, fd, MDIOCDETACH, (void *)mdio); 570 PICKUP_GIANT(); 571 /* Ignore errors. We don't care. */ 572 error = ERANGE; 573 goto out; 574 } 575 576 root_mount_mddev = mdio->md_unit; 577 printf(MD_NAME "%u attached to %s\n", root_mount_mddev, mdio->md_file); 578 579 error = kern_close(td, fd); 580 581 out: 582 free(mdio, M_TEMP); 583 return (error); 584} 585 586static int 587parse_dir_onfail(char **conf) 588{ 589 char *action; 590 int error; 591 592 error = parse_token(conf, &action); 593 if (error) 594 return (error); 595 596 if (!strcmp(action, "continue")) 597 root_mount_onfail = A_CONTINUE; 598 else if (!strcmp(action, "panic")) 599 root_mount_onfail = A_PANIC; 600 else if (!strcmp(action, "reboot")) 601 root_mount_onfail = A_REBOOT; 602 else if (!strcmp(action, "retry")) 603 root_mount_onfail = A_RETRY; 604 else { 605 printf("rootmount: %s: unknown action\n", action); 606 error = EINVAL; 607 } 608 609 free(action, M_TEMP); 610 return (0); 611} 612 613static int 614parse_dir_timeout(char **conf) 615{ 616 char *tok, *endtok; 617 long secs; 618 int error; 619 620 error = parse_token(conf, &tok); 621 if (error) 622 return (error); 623 624 secs = strtol(tok, &endtok, 0); 625 error = (secs < 0 || *endtok != '\0') ? EINVAL : 0; 626 if (!error) 627 root_mount_timeout = secs; 628 free(tok, M_TEMP); 629 return (error); 630} 631 632static int 633parse_directive(char **conf) 634{ 635 char *dir; 636 int error; 637 638 error = parse_token(conf, &dir); 639 if (error) 640 return (error); 641 642 if (strcmp(dir, ".ask") == 0) 643 error = parse_dir_ask(conf); 644 else if (strcmp(dir, ".md") == 0) 645 error = parse_dir_md(conf); 646 else if (strcmp(dir, ".onfail") == 0) 647 error = parse_dir_onfail(conf); 648 else if (strcmp(dir, ".timeout") == 0) 649 error = parse_dir_timeout(conf); 650 else { 651 printf("mountroot: invalid directive `%s'\n", dir); 652 /* Ignore the rest of the line. */ 653 (void)parse_skipto(conf, '\n'); 654 error = EINVAL; 655 } 656 free(dir, M_TEMP); 657 return (error); 658} 659 660static int 661parse_mount_dev_present(const char *dev) 662{ 663 struct nameidata nd; 664 int error; 665 666 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, dev, curthread); 667 error = namei(&nd); 668 if (!error) 669 vput(nd.ni_vp); 670 NDFREE(&nd, NDF_ONLY_PNBUF); 671 return (error != 0) ? 0 : 1; 672} 673 674#define ERRMSGL 255 675static int 676parse_mount(char **conf) 677{ 678 char *errmsg; 679 struct mntarg *ma; 680 char *dev, *fs, *opts, *tok; 681 int delay, error, timeout; 682 683 error = parse_token(conf, &tok); 684 if (error) 685 return (error); 686 fs = tok; 687 error = parse_skipto(&tok, ':'); 688 if (error) { 689 free(fs, M_TEMP); 690 return (error); 691 } 692 parse_poke(&tok, '\0'); 693 parse_advance(&tok); 694 dev = tok; 695 696 if (root_mount_mddev != -1) { 697 /* Handle substitution for the md unit number. */ 698 tok = strstr(dev, "md#"); 699 if (tok != NULL) 700 tok[2] = '0' + root_mount_mddev; 701 } 702 703 /* Parse options. */ 704 error = parse_token(conf, &tok); 705 opts = (error == 0) ? tok : NULL; 706 707 printf("Trying to mount root from %s:%s [%s]...\n", fs, dev, 708 (opts != NULL) ? opts : ""); 709 710 errmsg = malloc(ERRMSGL, M_TEMP, M_WAITOK | M_ZERO); 711 712 if (vfs_byname(fs) == NULL) { 713 strlcpy(errmsg, "unknown file system", ERRMSGL); 714 error = ENOENT; 715 goto out; 716 } 717 718 if (strcmp(fs, "zfs") != 0 && strstr(fs, "nfs") == NULL && 719 dev[0] != '\0' && !parse_mount_dev_present(dev)) { 720 printf("mountroot: waiting for device %s ...\n", dev); 721 delay = hz / 10; 722 timeout = root_mount_timeout * hz; 723 do { 724 pause("rmdev", delay); 725 timeout -= delay; 726 } while (timeout > 0 && !parse_mount_dev_present(dev)); 727 if (timeout <= 0) { 728 error = ENODEV; 729 goto out; 730 } 731 } 732 733 ma = NULL; 734 ma = mount_arg(ma, "fstype", fs, -1); 735 ma = mount_arg(ma, "fspath", "/", -1); 736 ma = mount_arg(ma, "from", dev, -1); 737 ma = mount_arg(ma, "errmsg", errmsg, ERRMSGL); 738 ma = mount_arg(ma, "ro", NULL, 0); 739 ma = parse_mountroot_options(ma, opts); 740 error = kernel_mount(ma, MNT_ROOTFS); 741 742 out: 743 if (error) { 744 printf("Mounting from %s:%s failed with error %d", 745 fs, dev, error); 746 if (errmsg[0] != '\0') 747 printf(": %s", errmsg); 748 printf(".\n"); 749 } 750 free(fs, M_TEMP); 751 free(errmsg, M_TEMP); 752 if (opts != NULL) 753 free(opts, M_TEMP); 754 /* kernel_mount can return -1 on error. */ 755 return ((error < 0) ? EDOOFUS : error); 756} 757#undef ERRMSGL 758 759static int 760vfs_mountroot_parse(struct sbuf *sb, struct mount *mpdevfs) 761{ 762 struct mount *mp; 763 char *conf; 764 int error; 765 766 root_mount_mddev = -1; 767 768retry: 769 conf = sbuf_data(sb); 770 mp = TAILQ_NEXT(mpdevfs, mnt_list); 771 error = (mp == NULL) ? 0 : EDOOFUS; 772 root_mount_onfail = A_CONTINUE; 773 while (mp == NULL) { 774 error = parse_skipto(&conf, CC_NONWHITESPACE); 775 if (error == PE_EOL) { 776 parse_advance(&conf); 777 continue; 778 } 779 if (error < 0) 780 break; 781 switch (parse_peek(&conf)) { 782 case '#': 783 error = parse_skipto(&conf, '\n'); 784 break; 785 case '.': 786 error = parse_directive(&conf); 787 break; 788 default: 789 error = parse_mount(&conf); 790 break; 791 } 792 if (error < 0) 793 break; 794 /* Ignore any trailing garbage on the line. */ 795 if (parse_peek(&conf) != '\n') { 796 printf("mountroot: advancing to next directive...\n"); 797 (void)parse_skipto(&conf, '\n'); 798 } 799 mp = TAILQ_NEXT(mpdevfs, mnt_list); 800 } 801 if (mp != NULL) 802 return (0); 803 804 /* 805 * We failed to mount (a new) root. 806 */ 807 switch (root_mount_onfail) { 808 case A_CONTINUE: 809 break; 810 case A_PANIC: 811 panic("mountroot: unable to (re-)mount root."); 812 /* NOTREACHED */ 813 case A_RETRY: 814 goto retry; 815 case A_REBOOT: 816 kern_reboot(RB_NOSYNC); 817 /* NOTREACHED */ 818 } 819 820 return (error); 821} 822 823static void 824vfs_mountroot_conf0(struct sbuf *sb) 825{ 826 char *s, *tok, *mnt, *opt; 827 int error; 828 829 sbuf_printf(sb, ".onfail panic\n"); 830 sbuf_printf(sb, ".timeout %d\n", root_mount_timeout); 831 if (boothowto & RB_ASKNAME) 832 sbuf_printf(sb, ".ask\n"); 833#ifdef ROOTDEVNAME 834 if (boothowto & RB_DFLTROOT) 835 sbuf_printf(sb, "%s\n", ROOTDEVNAME); 836#endif 837 if (boothowto & RB_CDROM) { 838 sbuf_printf(sb, "cd9660:/dev/cd0 ro\n"); 839 sbuf_printf(sb, ".timeout 0\n"); 840 sbuf_printf(sb, "cd9660:/dev/acd0 ro\n"); 841 sbuf_printf(sb, ".timeout %d\n", root_mount_timeout); 842 } 843 s = getenv("vfs.root.mountfrom"); 844 if (s != NULL) { 845 opt = getenv("vfs.root.mountfrom.options"); 846 tok = s; 847 error = parse_token(&tok, &mnt); 848 while (!error) { 849 sbuf_printf(sb, "%s %s\n", mnt, 850 (opt != NULL) ? opt : ""); 851 free(mnt, M_TEMP); 852 error = parse_token(&tok, &mnt); 853 } 854 if (opt != NULL) 855 freeenv(opt); 856 freeenv(s); 857 } 858 if (rootdevnames[0] != NULL) 859 sbuf_printf(sb, "%s\n", rootdevnames[0]); 860 if (rootdevnames[1] != NULL) 861 sbuf_printf(sb, "%s\n", rootdevnames[1]); 862#ifdef ROOTDEVNAME 863 if (!(boothowto & RB_DFLTROOT)) 864 sbuf_printf(sb, "%s\n", ROOTDEVNAME); 865#endif 866 if (!(boothowto & RB_ASKNAME)) 867 sbuf_printf(sb, ".ask\n"); 868} 869 870static int 871vfs_mountroot_readconf(struct thread *td, struct sbuf *sb) 872{ 873 static char buf[128]; 874 struct nameidata nd; 875 off_t ofs; 876 ssize_t resid; 877 int error, flags, len; 878 879 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, "/.mount.conf", td); 880 flags = FREAD; 881 error = vn_open(&nd, &flags, 0, NULL); 882 if (error) 883 return (error); 884 885 NDFREE(&nd, NDF_ONLY_PNBUF); 886 ofs = 0; 887 len = sizeof(buf) - 1; 888 while (1) { 889 error = vn_rdwr(UIO_READ, nd.ni_vp, buf, len, ofs, 890 UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, 891 NOCRED, &resid, td); 892 if (error) 893 break; 894 if (resid == len) 895 break; 896 buf[len - resid] = 0; 897 sbuf_printf(sb, "%s", buf); 898 ofs += len - resid; 899 } 900 901 VOP_UNLOCK(nd.ni_vp, 0); 902 vn_close(nd.ni_vp, FREAD, td->td_ucred, td); 903 return (error); 904} 905 906static void 907vfs_mountroot_wait(void) 908{ 909 struct root_hold_token *h; 910 struct timeval lastfail; 911 int curfail; 912 913 curfail = 0; 914 while (1) { 915 DROP_GIANT(); 916 g_waitidle(); 917 PICKUP_GIANT(); 918 mtx_lock(&root_holds_mtx); 919 if (LIST_EMPTY(&root_holds)) { 920 mtx_unlock(&root_holds_mtx); 921 break; 922 } 923 if (ppsratecheck(&lastfail, &curfail, 1)) { 924 printf("Root mount waiting for:"); 925 LIST_FOREACH(h, &root_holds, list) 926 printf(" %s", h->who); 927 printf("\n"); 928 } 929 msleep(&root_holds, &root_holds_mtx, PZERO | PDROP, "roothold", 930 hz); 931 } 932} 933 934void 935vfs_mountroot(void) 936{ 937 struct mount *mp; 938 struct sbuf *sb; 939 struct thread *td; 940 time_t timebase; 941 int error; 942 943 td = curthread; 944 945 vfs_mountroot_wait(); 946 947 sb = sbuf_new_auto(); 948 vfs_mountroot_conf0(sb); 949 sbuf_finish(sb); 950 951 error = vfs_mountroot_devfs(td, &mp); 952 while (!error) { 953 error = vfs_mountroot_parse(sb, mp); 954 if (!error) { 955 vfs_mountroot_shuffle(td, mp); 956 sbuf_clear(sb); 957 error = vfs_mountroot_readconf(td, sb); 958 sbuf_finish(sb); 959 } 960 } 961 962 sbuf_delete(sb); 963 964 /* 965 * Iterate over all currently mounted file systems and use 966 * the time stamp found to check and/or initialize the RTC. 967 * Call inittodr() only once and pass it the largest of the 968 * timestamps we encounter. 969 */ 970 timebase = 0; 971 mtx_lock(&mountlist_mtx); 972 mp = TAILQ_FIRST(&mountlist); 973 while (mp != NULL) { 974 if (mp->mnt_time > timebase) 975 timebase = mp->mnt_time; 976 mp = TAILQ_NEXT(mp, mnt_list); 977 } 978 mtx_unlock(&mountlist_mtx); 979 inittodr(timebase); 980 981 /* Keep prison0's root in sync with the global rootvnode. */ 982 mtx_lock(&prison0.pr_mtx); 983 prison0.pr_root = rootvnode; 984 vref(prison0.pr_root); 985 mtx_unlock(&prison0.pr_mtx); 986 987 mtx_lock(&root_holds_mtx); 988 atomic_store_rel_int(&root_mount_complete, 1); 989 wakeup(&root_mount_complete); 990 mtx_unlock(&root_holds_mtx); 991 992 EVENTHANDLER_INVOKE(mountroot); 993} 994 995static struct mntarg * 996parse_mountroot_options(struct mntarg *ma, const char *options) 997{ 998 char *p; 999 char *name, *name_arg; 1000 char *val, *val_arg; 1001 char *opts; 1002 1003 if (options == NULL || options[0] == '\0') 1004 return (ma); 1005 1006 p = opts = strdup(options, M_MOUNT); 1007 if (opts == NULL) { 1008 return (ma); 1009 } 1010 1011 while((name = strsep(&p, ",")) != NULL) { 1012 if (name[0] == '\0') 1013 break; 1014 1015 val = strchr(name, '='); 1016 if (val != NULL) { 1017 *val = '\0'; 1018 ++val; 1019 } 1020 if( strcmp(name, "rw") == 0 || 1021 strcmp(name, "noro") == 0) { 1022 /* 1023 * The first time we mount the root file system, 1024 * we need to mount 'ro', so We need to ignore 1025 * 'rw' and 'noro' mount options. 1026 */ 1027 continue; 1028 } 1029 name_arg = strdup(name, M_MOUNT); 1030 val_arg = NULL; 1031 if (val != NULL) 1032 val_arg = strdup(val, M_MOUNT); 1033 1034 ma = mount_arg(ma, name_arg, val_arg, 1035 (val_arg != NULL ? -1 : 0)); 1036 } 1037 free(opts, M_MOUNT); 1038 return (ma); 1039} 1040