vfs_mount.c revision 152620
1/*- 2 * Copyright (c) 1999-2004 Poul-Henning Kamp 3 * Copyright (c) 1999 Michael Smith 4 * Copyright (c) 1989, 1993 5 * The Regents of the University of California. All rights reserved. 6 * (c) UNIX System Laboratories, Inc. 7 * All or some portions of this file are derived from material licensed 8 * to the University of California by American Telephone and Telegraph 9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 10 * the permission of UNIX System Laboratories, Inc. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 */ 36 37#include <sys/cdefs.h> 38__FBSDID("$FreeBSD: head/sys/kern/vfs_mount.c 152620 2005-11-19 21:51:45Z marcel $"); 39 40#include <sys/param.h> 41#include <sys/conf.h> 42#include <sys/jail.h> 43#include <sys/kernel.h> 44#include <sys/libkern.h> 45#include <sys/mac.h> 46#include <sys/malloc.h> 47#include <sys/mount.h> 48#include <sys/mutex.h> 49#include <sys/namei.h> 50#include <sys/proc.h> 51#include <sys/filedesc.h> 52#include <sys/reboot.h> 53#include <sys/syscallsubr.h> 54#include <sys/sysproto.h> 55#include <sys/sx.h> 56#include <sys/sysctl.h> 57#include <sys/sysent.h> 58#include <sys/systm.h> 59#include <sys/vnode.h> 60 61#include <geom/geom.h> 62 63#include <machine/stdarg.h> 64 65#include "opt_rootdevname.h" 66#include "opt_ddb.h" 67#include "opt_mac.h" 68 69#ifdef DDB 70#include <ddb/ddb.h> 71#endif 72 73#define ROOTNAME "root_device" 74#define VFS_MOUNTARG_SIZE_MAX (1024 * 64) 75 76static int vfs_domount(struct thread *td, const char *fstype, 77 char *fspath, int fsflags, void *fsdata); 78static int vfs_mount_alloc(struct vnode *dvp, struct vfsconf *vfsp, 79 const char *fspath, struct thread *td, struct mount **mpp); 80static int vfs_mountroot_ask(void); 81static int vfs_mountroot_try(const char *mountfrom); 82static int vfs_donmount(struct thread *td, int fsflags, 83 struct uio *fsoptions); 84static void free_mntarg(struct mntarg *ma); 85static void vfs_mount_destroy(struct mount *, struct thread *); 86 87static int usermount = 0; 88SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, 89 "Unprivileged users may mount and unmount file systems"); 90 91MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure"); 92 93/* List of mounted filesystems. */ 94struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); 95 96/* For any iteration/modification of mountlist */ 97struct mtx mountlist_mtx; 98MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF); 99 100TAILQ_HEAD(vfsoptlist, vfsopt); 101struct vfsopt { 102 TAILQ_ENTRY(vfsopt) link; 103 char *name; 104 void *value; 105 int len; 106}; 107 108/* 109 * The vnode of the system's root (/ in the filesystem, without chroot 110 * active.) 111 */ 112struct vnode *rootvnode; 113 114/* 115 * The root filesystem is detailed in the kernel environment variable 116 * vfs.root.mountfrom, which is expected to be in the general format 117 * 118 * <vfsname>:[<path>] 119 * vfsname := the name of a VFS known to the kernel and capable 120 * of being mounted as root 121 * path := disk device name or other data used by the filesystem 122 * to locate its physical store 123 */ 124 125/* 126 * Global opts, taken by all filesystems 127 */ 128static const char *global_opts[] = { 129 "fstype", 130 "fspath", 131 "ro", 132 "suid", 133 "exec", 134 NULL 135}; 136 137/* 138 * The root specifiers we will try if RB_CDROM is specified. 139 */ 140static char *cdrom_rootdevnames[] = { 141 "cd9660:cd0", 142 "cd9660:acd0", 143 NULL 144}; 145 146/* legacy find-root code */ 147char *rootdevnames[2] = {NULL, NULL}; 148#ifndef ROOTDEVNAME 149# define ROOTDEVNAME NULL 150#endif 151static const char *ctrootdevname = ROOTDEVNAME; 152 153/* 154 * --------------------------------------------------------------------- 155 * Functions for building and sanitizing the mount options 156 */ 157 158/* Remove one mount option. */ 159static void 160vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt) 161{ 162 163 TAILQ_REMOVE(opts, opt, link); 164 free(opt->name, M_MOUNT); 165 if (opt->value != NULL) 166 free(opt->value, M_MOUNT); 167#ifdef INVARIANTS 168 else if (opt->len != 0) 169 panic("%s: mount option with NULL value but length != 0", 170 __func__); 171#endif 172 free(opt, M_MOUNT); 173} 174 175/* Release all resources related to the mount options. */ 176static void 177vfs_freeopts(struct vfsoptlist *opts) 178{ 179 struct vfsopt *opt; 180 181 while (!TAILQ_EMPTY(opts)) { 182 opt = TAILQ_FIRST(opts); 183 vfs_freeopt(opts, opt); 184 } 185 free(opts, M_MOUNT); 186} 187 188/* 189 * Check if options are equal (with or without the "no" prefix). 190 */ 191static int 192vfs_equalopts(const char *opt1, const char *opt2) 193{ 194 195 /* "opt" vs. "opt" or "noopt" vs. "noopt" */ 196 if (strcmp(opt1, opt2) == 0) 197 return (1); 198 /* "noopt" vs. "opt" */ 199 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 200 return (1); 201 /* "opt" vs. "noopt" */ 202 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 203 return (1); 204 return (0); 205} 206 207/* 208 * If a mount option is specified several times, 209 * (with or without the "no" prefix) only keep 210 * the last occurence of it. 211 */ 212static void 213vfs_sanitizeopts(struct vfsoptlist *opts) 214{ 215 struct vfsopt *opt, *opt2, *tmp; 216 217 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) { 218 opt2 = TAILQ_PREV(opt, vfsoptlist, link); 219 while (opt2 != NULL) { 220 if (vfs_equalopts(opt->name, opt2->name)) { 221 tmp = TAILQ_PREV(opt2, vfsoptlist, link); 222 vfs_freeopt(opts, opt2); 223 opt2 = tmp; 224 } else { 225 opt2 = TAILQ_PREV(opt2, vfsoptlist, link); 226 } 227 } 228 } 229} 230 231/* 232 * Build a linked list of mount options from a struct uio. 233 */ 234static int 235vfs_buildopts(struct uio *auio, struct vfsoptlist **options) 236{ 237 struct vfsoptlist *opts; 238 struct vfsopt *opt; 239 size_t memused; 240 unsigned int i, iovcnt; 241 int error, namelen, optlen; 242 243 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 244 TAILQ_INIT(opts); 245 memused = 0; 246 iovcnt = auio->uio_iovcnt; 247 for (i = 0; i < iovcnt; i += 2) { 248 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 249 namelen = auio->uio_iov[i].iov_len; 250 optlen = auio->uio_iov[i + 1].iov_len; 251 opt->name = malloc(namelen, M_MOUNT, M_WAITOK); 252 opt->value = NULL; 253 opt->len = 0; 254 255 /* 256 * Do this early, so jumps to "bad" will free the current 257 * option. 258 */ 259 TAILQ_INSERT_TAIL(opts, opt, link); 260 memused += sizeof(struct vfsopt) + optlen + namelen; 261 262 /* 263 * Avoid consuming too much memory, and attempts to overflow 264 * memused. 265 */ 266 if (memused > VFS_MOUNTARG_SIZE_MAX || 267 optlen > VFS_MOUNTARG_SIZE_MAX || 268 namelen > VFS_MOUNTARG_SIZE_MAX) { 269 error = EINVAL; 270 goto bad; 271 } 272 273 if (auio->uio_segflg == UIO_SYSSPACE) { 274 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen); 275 } else { 276 error = copyin(auio->uio_iov[i].iov_base, opt->name, 277 namelen); 278 if (error) 279 goto bad; 280 } 281 /* Ensure names are null-terminated strings. */ 282 if (opt->name[namelen - 1] != '\0') { 283 error = EINVAL; 284 goto bad; 285 } 286 if (optlen != 0) { 287 opt->len = optlen; 288 opt->value = malloc(optlen, M_MOUNT, M_WAITOK); 289 if (auio->uio_segflg == UIO_SYSSPACE) { 290 bcopy(auio->uio_iov[i + 1].iov_base, opt->value, 291 optlen); 292 } else { 293 error = copyin(auio->uio_iov[i + 1].iov_base, 294 opt->value, optlen); 295 if (error) 296 goto bad; 297 } 298 } 299 } 300 vfs_sanitizeopts(opts); 301 *options = opts; 302 return (0); 303bad: 304 vfs_freeopts(opts); 305 return (error); 306} 307 308/* 309 * Merge the old mount options with the new ones passed 310 * in the MNT_UPDATE case. 311 */ 312static void 313vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts) 314{ 315 struct vfsopt *opt, *opt2, *new; 316 317 TAILQ_FOREACH(opt, opts, link) { 318 /* 319 * Check that this option hasn't been redefined 320 * nor cancelled with a "no" mount option. 321 */ 322 opt2 = TAILQ_FIRST(toopts); 323 while (opt2 != NULL) { 324 if (strcmp(opt2->name, opt->name) == 0) 325 goto next; 326 if (strncmp(opt2->name, "no", 2) == 0 && 327 strcmp(opt2->name + 2, opt->name) == 0) { 328 vfs_freeopt(toopts, opt2); 329 goto next; 330 } 331 opt2 = TAILQ_NEXT(opt2, link); 332 } 333 /* We want this option, duplicate it. */ 334 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 335 new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK); 336 strcpy(new->name, opt->name); 337 if (opt->len != 0) { 338 new->value = malloc(opt->len, M_MOUNT, M_WAITOK); 339 bcopy(opt->value, new->value, opt->len); 340 } else { 341 new->value = NULL; 342 } 343 new->len = opt->len; 344 TAILQ_INSERT_TAIL(toopts, new, link); 345next: 346 continue; 347 } 348} 349 350/* 351 * --------------------------------------------------------------------- 352 * Mount a filesystem 353 */ 354int 355nmount(td, uap) 356 struct thread *td; 357 struct nmount_args /* { 358 struct iovec *iovp; 359 unsigned int iovcnt; 360 int flags; 361 } */ *uap; 362{ 363 struct uio *auio; 364 struct iovec *iov; 365 unsigned int i; 366 int error; 367 u_int iovcnt; 368 const char *name; 369 370 /* Kick out MNT_ROOTFS early as it is legal internally */ 371 if (uap->flags & MNT_ROOTFS) 372 return (EINVAL); 373 374 iovcnt = uap->iovcnt; 375 /* 376 * Check that we have an even number of iovec's 377 * and that we have at least two options. 378 */ 379 if ((iovcnt & 1) || (iovcnt < 4)) 380 return (EINVAL); 381 382 error = copyinuio(uap->iovp, iovcnt, &auio); 383 if (error) 384 return (error); 385 iov = auio->uio_iov; 386 for (i = 0; i < iovcnt; i++) { 387 if (iov->iov_len > MMAXOPTIONLEN) { 388 free(auio, M_IOV); 389 return (EINVAL); 390 } 391 iov++; 392 } 393 error = vfs_donmount(td, uap->flags, auio); 394 395 /* copyout the errmsg */ 396 for (i = 0; error != 0 && i < iovcnt; i += 2) { 397 name = (const char *)auio->uio_iov[i].iov_base; 398 if (strcmp(name, "errmsg") == 0) { 399 copyout(auio->uio_iov[i + 1].iov_base, 400 uap->iovp[i + 1].iov_base, uap->iovp[i + 1].iov_len); 401 break; 402 } 403 } 404 405 free(auio, M_IOV); 406 return (error); 407} 408 409/* 410 * --------------------------------------------------------------------- 411 * Various utility functions 412 */ 413 414/* 415 * Allocate and initialize the mount point struct. 416 */ 417static int 418vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, 419 const char *fspath, struct thread *td, struct mount **mpp) 420{ 421 struct mount *mp; 422 423 mp = malloc(sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO); 424 TAILQ_INIT(&mp->mnt_nvnodelist); 425 mp->mnt_nvnodelistsize = 0; 426 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF); 427 lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0); 428 vfs_busy(mp, LK_NOWAIT, 0, td); 429 mp->mnt_op = vfsp->vfc_vfsops; 430 mp->mnt_vfc = vfsp; 431 vfsp->vfc_refcount++; 432 mp->mnt_stat.f_type = vfsp->vfc_typenum; 433 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK; 434 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 435 mp->mnt_vnodecovered = vp; 436 mp->mnt_cred = crdup(td->td_ucred); 437 mp->mnt_stat.f_owner = td->td_ucred->cr_uid; 438 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN); 439 mp->mnt_iosize_max = DFLTPHYS; 440#ifdef MAC 441 mac_init_mount(mp); 442 mac_create_mount(td->td_ucred, mp); 443#endif 444 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0); 445 *mpp = mp; 446 return (0); 447} 448 449/* 450 * Destroy the mount struct previously allocated by vfs_mount_alloc(). 451 */ 452static void 453vfs_mount_destroy(struct mount *mp, struct thread *td) 454{ 455 456 mp->mnt_vfc->vfc_refcount--; 457 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) 458 panic("unmount: dangling vnode"); 459 vfs_unbusy(mp,td); 460 lockdestroy(&mp->mnt_lock); 461 MNT_ILOCK(mp); 462 if (mp->mnt_kern_flag & MNTK_MWAIT) 463 wakeup(mp); 464 MNT_IUNLOCK(mp); 465 mtx_destroy(&mp->mnt_mtx); 466#ifdef MAC 467 mac_destroy_mount(mp); 468#endif 469 if (mp->mnt_opt != NULL) 470 vfs_freeopts(mp->mnt_opt); 471 crfree(mp->mnt_cred); 472 free(mp, M_MOUNT); 473} 474 475static int 476vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions) 477{ 478 struct vfsoptlist *optlist; 479 struct iovec *iov_errmsg; 480 char *fstype, *fspath; 481 int error, fstypelen, fspathlen, i; 482 483 iov_errmsg = NULL; 484 485 for (i = 0; i < fsoptions->uio_iovcnt; i += 2) { 486 if (strcmp((char *)fsoptions->uio_iov[i].iov_base, "errmsg") == 0) 487 iov_errmsg = &fsoptions->uio_iov[i + 1]; 488 } 489 490 error = vfs_buildopts(fsoptions, &optlist); 491 if (error) 492 return (error); 493 494 /* 495 * We need these two options before the others, 496 * and they are mandatory for any filesystem. 497 * Ensure they are NUL terminated as well. 498 */ 499 fstypelen = 0; 500 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen); 501 if (error || fstype[fstypelen - 1] != '\0') { 502 error = EINVAL; 503 if (iov_errmsg != NULL) 504 strncpy((char *)iov_errmsg->iov_base, "Invalid fstype", 505 iov_errmsg->iov_len); 506 goto bail; 507 } 508 fspathlen = 0; 509 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen); 510 if (error || fspath[fspathlen - 1] != '\0') { 511 error = EINVAL; 512 if (iov_errmsg != NULL) 513 strncpy((char *)iov_errmsg->iov_base, "Invalid fspath", 514 iov_errmsg->iov_len); 515 goto bail; 516 } 517 518 /* 519 * We need to see if we have the "update" option 520 * before we call vfs_domount(), since vfs_domount() has special 521 * logic based on MNT_UPDATE. This is very important 522 * when we want to update the root filesystem. 523 */ 524 if (vfs_getopt(optlist, "update", NULL, NULL) == 0) 525 fsflags |= MNT_UPDATE; 526 527 if (vfs_getopt(optlist, "async", NULL, NULL) == 0) 528 fsflags |= MNT_ASYNC; 529 530 if (vfs_getopt(optlist, "force", NULL, NULL) == 0) 531 fsflags |= MNT_FORCE; 532 533 if (vfs_getopt(optlist, "multilabel", NULL, NULL) == 0) 534 fsflags |= MNT_MULTILABEL; 535 536 if (vfs_getopt(optlist, "noasync", NULL, NULL) == 0) 537 fsflags &= ~MNT_ASYNC; 538 539 if (vfs_getopt(optlist, "noatime", NULL, NULL) == 0) 540 fsflags |= MNT_NOATIME; 541 542 if (vfs_getopt(optlist, "noclusterr", NULL, NULL) == 0) 543 fsflags |= MNT_NOCLUSTERR; 544 545 if (vfs_getopt(optlist, "noclusterw", NULL, NULL) == 0) 546 fsflags |= MNT_NOCLUSTERW; 547 548 if (vfs_getopt(optlist, "noexec", NULL, NULL) == 0) 549 fsflags |= MNT_NOEXEC; 550 551 if (vfs_getopt(optlist, "nosuid", NULL, NULL) == 0) 552 fsflags |= MNT_NOSUID; 553 554 if (vfs_getopt(optlist, "nosymfollow", NULL, NULL) == 0) 555 fsflags |= MNT_NOSYMFOLLOW; 556 557 if (vfs_getopt(optlist, "snapshot", NULL, NULL) == 0) 558 fsflags |= MNT_SNAPSHOT; 559 560 if (vfs_getopt(optlist, "suiddir", NULL, NULL) == 0) 561 fsflags |= MNT_SUIDDIR; 562 563 if (vfs_getopt(optlist, "sync", NULL, NULL) == 0) 564 fsflags |= MNT_SYNCHRONOUS; 565 566 if (vfs_getopt(optlist, "union", NULL, NULL) == 0) 567 fsflags |= MNT_UNION; 568 569 /* 570 * Be ultra-paranoid about making sure the type and fspath 571 * variables will fit in our mp buffers, including the 572 * terminating NUL. 573 */ 574 if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) { 575 error = ENAMETOOLONG; 576 goto bail; 577 } 578 579 mtx_lock(&Giant); 580 error = vfs_domount(td, fstype, fspath, fsflags, optlist); 581 mtx_unlock(&Giant); 582bail: 583 if (error != 0 && iov_errmsg != NULL) { 584 /* Save the errmsg so we can return it to userspace. */ 585 char *errmsg; 586 int len, ret; 587 ret = vfs_getopt(optlist, "errmsg", (void **)&errmsg, &len); 588 if (ret == 0 && len > 0) 589 strncpy((char *)iov_errmsg->iov_base, errmsg, 590 iov_errmsg->iov_len); 591 592 } 593 594 if (error != 0) 595 vfs_freeopts(optlist); 596 return (error); 597} 598 599/* 600 * --------------------------------------------------------------------- 601 * Old mount API. 602 */ 603#ifndef _SYS_SYSPROTO_H_ 604struct mount_args { 605 char *type; 606 char *path; 607 int flags; 608 caddr_t data; 609}; 610#endif 611/* ARGSUSED */ 612int 613mount(td, uap) 614 struct thread *td; 615 struct mount_args /* { 616 char *type; 617 char *path; 618 int flags; 619 caddr_t data; 620 } */ *uap; 621{ 622 char *fstype; 623 struct vfsconf *vfsp = NULL; 624 struct mntarg *ma = NULL; 625 int error; 626 627 /* Kick out MNT_ROOTFS early as it is legal internally */ 628 uap->flags &= ~MNT_ROOTFS; 629 630 if (uap->data == NULL) 631 return (EINVAL); 632 633 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 634 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 635 if (!error) { 636 mtx_lock(&Giant); /* XXX ? */ 637 vfsp = vfs_byname_kld(fstype, td, &error); 638 mtx_unlock(&Giant); 639 } 640 free(fstype, M_TEMP); 641 if (error) 642 return (error); 643 if (vfsp == NULL) 644 return (ENOENT); 645 if (vfsp->vfc_vfsops->vfs_cmount == NULL) 646 return (EOPNOTSUPP); 647 648 ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN); 649 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 650 ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro"); 651 ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid"); 652 ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec"); 653 654 error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags, td); 655 return (error); 656} 657 658 659/* 660 * vfs_domount(): actually attempt a filesystem mount. 661 */ 662static int 663vfs_domount( 664 struct thread *td, /* Flags common to all filesystems. */ 665 const char *fstype, /* Filesystem type. */ 666 char *fspath, /* Mount path. */ 667 int fsflags, /* Flags common to all filesystems. */ 668 void *fsdata /* Options local to the filesystem. */ 669 ) 670{ 671 struct vnode *vp; 672 struct mount *mp; 673 struct vfsconf *vfsp; 674 int error, flag = 0, kern_flag = 0; 675 struct vattr va; 676 struct nameidata nd; 677 678 mtx_assert(&Giant, MA_OWNED); 679 680 /* 681 * Be ultra-paranoid about making sure the type and fspath 682 * variables will fit in our mp buffers, including the 683 * terminating NUL. 684 */ 685 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 686 return (ENAMETOOLONG); 687 688 if (jailed(td->td_ucred)) 689 return (EPERM); 690 if (usermount == 0) { 691 if ((error = suser(td)) != 0) 692 return (error); 693 } 694 695 /* 696 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 697 */ 698 if (fsflags & (MNT_EXPORTED | MNT_SUIDDIR)) { 699 if ((error = suser(td)) != 0) 700 return (error); 701 } 702 /* 703 * Silently enforce MNT_NOSUID and MNT_USER for 704 * unprivileged users. 705 */ 706 if (suser(td) != 0) 707 fsflags |= MNT_NOSUID | MNT_USER; 708 /* 709 * Get vnode to be covered 710 */ 711 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspath, td); 712 if ((error = namei(&nd)) != 0) 713 return (error); 714 NDFREE(&nd, NDF_ONLY_PNBUF); 715 vp = nd.ni_vp; 716 if (fsflags & MNT_UPDATE) { 717 if ((vp->v_vflag & VV_ROOT) == 0) { 718 vput(vp); 719 return (EINVAL); 720 } 721 mp = vp->v_mount; 722 flag = mp->mnt_flag; 723 kern_flag = mp->mnt_kern_flag; 724 /* 725 * We only allow the filesystem to be reloaded if it 726 * is currently mounted read-only. 727 */ 728 if ((fsflags & MNT_RELOAD) && 729 ((mp->mnt_flag & MNT_RDONLY) == 0)) { 730 vput(vp); 731 return (EOPNOTSUPP); /* Needs translation */ 732 } 733 /* 734 * Only privileged root, or (if MNT_USER is set) the user that 735 * did the original mount is permitted to update it. 736 */ 737 error = vfs_suser(mp, td); 738 if (error) { 739 vput(vp); 740 return (error); 741 } 742 if (vfs_busy(mp, LK_NOWAIT, 0, td)) { 743 vput(vp); 744 return (EBUSY); 745 } 746 VI_LOCK(vp); 747 if ((vp->v_iflag & VI_MOUNT) != 0 || 748 vp->v_mountedhere != NULL) { 749 VI_UNLOCK(vp); 750 vfs_unbusy(mp, td); 751 vput(vp); 752 return (EBUSY); 753 } 754 vp->v_iflag |= VI_MOUNT; 755 VI_UNLOCK(vp); 756 mp->mnt_flag |= fsflags & 757 (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS); 758 VOP_UNLOCK(vp, 0, td); 759 mp->mnt_optnew = fsdata; 760 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 761 } else { 762 /* 763 * If the user is not root, ensure that they own the directory 764 * onto which we are attempting to mount. 765 */ 766 error = VOP_GETATTR(vp, &va, td->td_ucred, td); 767 if (error) { 768 vput(vp); 769 return (error); 770 } 771 if (va.va_uid != td->td_ucred->cr_uid) { 772 if ((error = suser(td)) != 0) { 773 vput(vp); 774 return (error); 775 } 776 } 777 error = vinvalbuf(vp, V_SAVE, td, 0, 0); 778 if (error != 0) { 779 vput(vp); 780 return (error); 781 } 782 if (vp->v_type != VDIR) { 783 vput(vp); 784 return (ENOTDIR); 785 } 786 vfsp = vfs_byname_kld(fstype, td, &error); 787 if (vfsp == NULL) { 788 vput(vp); 789 return (error); 790 } 791 VI_LOCK(vp); 792 if ((vp->v_iflag & VI_MOUNT) != 0 || 793 vp->v_mountedhere != NULL) { 794 VI_UNLOCK(vp); 795 vput(vp); 796 return (EBUSY); 797 } 798 vp->v_iflag |= VI_MOUNT; 799 VI_UNLOCK(vp); 800 801 /* 802 * Allocate and initialize the filesystem. 803 */ 804 error = vfs_mount_alloc(vp, vfsp, fspath, td, &mp); 805 if (error) { 806 vput(vp); 807 return (error); 808 } 809 VOP_UNLOCK(vp, 0, td); 810 811 /* XXXMAC: pass to vfs_mount_alloc? */ 812 mp->mnt_optnew = fsdata; 813 } 814 815 /* 816 * Set the mount level flags. 817 */ 818 if (fsflags & MNT_RDONLY) 819 mp->mnt_flag |= MNT_RDONLY; 820 mp->mnt_flag &=~ MNT_UPDATEMASK; 821 mp->mnt_flag |= fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS); 822 /* 823 * Mount the filesystem. 824 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 825 * get. No freeing of cn_pnbuf. 826 */ 827 error = VFS_MOUNT(mp, td); 828 if (!error) { 829 if (mp->mnt_opt != NULL) 830 vfs_freeopts(mp->mnt_opt); 831 mp->mnt_opt = mp->mnt_optnew; 832 VFS_STATFS(mp, &mp->mnt_stat, td); 833 } 834 /* 835 * Prevent external consumers of mount options from reading 836 * mnt_optnew. 837 */ 838 mp->mnt_optnew = NULL; 839 if (mp->mnt_flag & MNT_UPDATE) { 840 mp->mnt_flag &= 841 ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE | MNT_SNAPSHOT); 842 if (error) { 843 mp->mnt_flag = flag; 844 mp->mnt_kern_flag = kern_flag; 845 } 846 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 847 if (mp->mnt_syncer == NULL) 848 error = vfs_allocate_syncvnode(mp); 849 } else { 850 if (mp->mnt_syncer != NULL) 851 vrele(mp->mnt_syncer); 852 mp->mnt_syncer = NULL; 853 } 854 vfs_unbusy(mp, td); 855 VI_LOCK(vp); 856 vp->v_iflag &= ~VI_MOUNT; 857 VI_UNLOCK(vp); 858 vrele(vp); 859 return (error); 860 } 861 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 862 /* 863 * Put the new filesystem on the mount list after root. 864 */ 865 cache_purge(vp); 866 if (!error) { 867 struct vnode *newdp; 868 869 VI_LOCK(vp); 870 vp->v_iflag &= ~VI_MOUNT; 871 VI_UNLOCK(vp); 872 vp->v_mountedhere = mp; 873 mtx_lock(&mountlist_mtx); 874 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 875 mtx_unlock(&mountlist_mtx); 876 vfs_event_signal(NULL, VQ_MOUNT, 0); 877 if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp, td)) 878 panic("mount: lost mount"); 879 mountcheckdirs(vp, newdp); 880 vput(newdp); 881 VOP_UNLOCK(vp, 0, td); 882 if ((mp->mnt_flag & MNT_RDONLY) == 0) 883 error = vfs_allocate_syncvnode(mp); 884 vfs_unbusy(mp, td); 885 if (error) 886 vrele(vp); 887 } else { 888 VI_LOCK(vp); 889 vp->v_iflag &= ~VI_MOUNT; 890 VI_UNLOCK(vp); 891 vfs_mount_destroy(mp, td); 892 vput(vp); 893 } 894 return (error); 895} 896 897/* 898 * --------------------------------------------------------------------- 899 * Unmount a filesystem. 900 * 901 * Note: unmount takes a path to the vnode mounted on as argument, 902 * not special file (as before). 903 */ 904#ifndef _SYS_SYSPROTO_H_ 905struct unmount_args { 906 char *path; 907 int flags; 908}; 909#endif 910/* ARGSUSED */ 911int 912unmount(td, uap) 913 struct thread *td; 914 register struct unmount_args /* { 915 char *path; 916 int flags; 917 } */ *uap; 918{ 919 struct mount *mp; 920 char *pathbuf; 921 int error, id0, id1; 922 923 if (jailed(td->td_ucred)) 924 return (EPERM); 925 if (usermount == 0) { 926 if ((error = suser(td)) != 0) 927 return (error); 928 } 929 930 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 931 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL); 932 if (error) { 933 free(pathbuf, M_TEMP); 934 return (error); 935 } 936 if (uap->flags & MNT_BYFSID) { 937 /* Decode the filesystem ID. */ 938 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) { 939 free(pathbuf, M_TEMP); 940 return (EINVAL); 941 } 942 943 mtx_lock(&mountlist_mtx); 944 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 945 if (mp->mnt_stat.f_fsid.val[0] == id0 && 946 mp->mnt_stat.f_fsid.val[1] == id1) 947 break; 948 } 949 mtx_unlock(&mountlist_mtx); 950 } else { 951 mtx_lock(&mountlist_mtx); 952 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 953 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) 954 break; 955 } 956 mtx_unlock(&mountlist_mtx); 957 } 958 free(pathbuf, M_TEMP); 959 if (mp == NULL) { 960 /* 961 * Previously we returned ENOENT for a nonexistent path and 962 * EINVAL for a non-mountpoint. We cannot tell these apart 963 * now, so in the !MNT_BYFSID case return the more likely 964 * EINVAL for compatibility. 965 */ 966 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL); 967 } 968 969 /* 970 * Only privileged root, or (if MNT_USER is set) the user that did the 971 * original mount is permitted to unmount this filesystem. 972 */ 973 error = vfs_suser(mp, td); 974 if (error) 975 return (error); 976 977 /* 978 * Don't allow unmounting the root filesystem. 979 */ 980 if (mp->mnt_flag & MNT_ROOTFS) 981 return (EINVAL); 982 mtx_lock(&Giant); 983 error = dounmount(mp, uap->flags, td); 984 mtx_unlock(&Giant); 985 return (error); 986} 987 988/* 989 * Do the actual filesystem unmount. 990 */ 991int 992dounmount(mp, flags, td) 993 struct mount *mp; 994 int flags; 995 struct thread *td; 996{ 997 struct vnode *coveredvp, *fsrootvp; 998 int error; 999 int async_flag; 1000 1001 mtx_assert(&Giant, MA_OWNED); 1002 1003 MNT_ILOCK(mp); 1004 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 1005 MNT_IUNLOCK(mp); 1006 return (EBUSY); 1007 } 1008 mp->mnt_kern_flag |= MNTK_UNMOUNT; 1009 /* Allow filesystems to detect that a forced unmount is in progress. */ 1010 if (flags & MNT_FORCE) 1011 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 1012 error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK | 1013 ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), MNT_MTX(mp), td); 1014 if (error) { 1015 MNT_ILOCK(mp); 1016 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1017 if (mp->mnt_kern_flag & MNTK_MWAIT) 1018 wakeup(mp); 1019 MNT_IUNLOCK(mp); 1020 return (error); 1021 } 1022 vn_start_write(NULL, &mp, V_WAIT); 1023 1024 if (mp->mnt_flag & MNT_EXPUBLIC) 1025 vfs_setpublicfs(NULL, NULL, NULL); 1026 1027 vfs_msync(mp, MNT_WAIT); 1028 async_flag = mp->mnt_flag & MNT_ASYNC; 1029 mp->mnt_flag &= ~MNT_ASYNC; 1030 cache_purgevfs(mp); /* remove cache entries for this file sys */ 1031 if (mp->mnt_syncer != NULL) 1032 vrele(mp->mnt_syncer); 1033 /* 1034 * For forced unmounts, move process cdir/rdir refs on the fs root 1035 * vnode to the covered vnode. For non-forced unmounts we want 1036 * such references to cause an EBUSY error. 1037 */ 1038 if ((flags & MNT_FORCE) && 1039 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) { 1040 if (mp->mnt_vnodecovered != NULL) 1041 mountcheckdirs(fsrootvp, mp->mnt_vnodecovered); 1042 if (fsrootvp == rootvnode) { 1043 vrele(rootvnode); 1044 rootvnode = NULL; 1045 } 1046 vput(fsrootvp); 1047 } 1048 if (((mp->mnt_flag & MNT_RDONLY) || 1049 (error = VFS_SYNC(mp, MNT_WAIT, td)) == 0) || 1050 (flags & MNT_FORCE)) { 1051 error = VFS_UNMOUNT(mp, flags, td); 1052 } 1053 vn_finished_write(mp); 1054 if (error) { 1055 /* Undo cdir/rdir and rootvnode changes made above. */ 1056 if ((flags & MNT_FORCE) && 1057 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) { 1058 if (mp->mnt_vnodecovered != NULL) 1059 mountcheckdirs(mp->mnt_vnodecovered, fsrootvp); 1060 if (rootvnode == NULL) { 1061 rootvnode = fsrootvp; 1062 vref(rootvnode); 1063 } 1064 vput(fsrootvp); 1065 } 1066 if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL) 1067 (void) vfs_allocate_syncvnode(mp); 1068 MNT_ILOCK(mp); 1069 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1070 mp->mnt_flag |= async_flag; 1071 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td); 1072 if (mp->mnt_kern_flag & MNTK_MWAIT) 1073 wakeup(mp); 1074 MNT_IUNLOCK(mp); 1075 return (error); 1076 } 1077 mtx_lock(&mountlist_mtx); 1078 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1079 if ((coveredvp = mp->mnt_vnodecovered) != NULL) 1080 coveredvp->v_mountedhere = NULL; 1081 mtx_unlock(&mountlist_mtx); 1082 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 1083 vfs_mount_destroy(mp, td); 1084 if (coveredvp != NULL) 1085 vrele(coveredvp); 1086 return (0); 1087} 1088 1089/* 1090 * --------------------------------------------------------------------- 1091 * Mounting of root filesystem 1092 * 1093 */ 1094 1095struct root_hold_token { 1096 const char *who; 1097 LIST_ENTRY(root_hold_token) list; 1098}; 1099 1100static LIST_HEAD(, root_hold_token) root_holds = 1101 LIST_HEAD_INITIALIZER(&root_holds); 1102 1103struct root_hold_token * 1104root_mount_hold(const char *identifier) 1105{ 1106 struct root_hold_token *h; 1107 1108 h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK); 1109 h->who = identifier; 1110 mtx_lock(&mountlist_mtx); 1111 LIST_INSERT_HEAD(&root_holds, h, list); 1112 mtx_unlock(&mountlist_mtx); 1113 return (h); 1114} 1115 1116void 1117root_mount_rel(struct root_hold_token *h) 1118{ 1119 1120 mtx_lock(&mountlist_mtx); 1121 LIST_REMOVE(h, list); 1122 wakeup(&root_holds); 1123 mtx_unlock(&mountlist_mtx); 1124 free(h, M_DEVBUF); 1125} 1126 1127static void 1128root_mount_wait(void) 1129{ 1130 struct root_hold_token *h; 1131 1132 for (;;) { 1133 DROP_GIANT(); 1134 g_waitidle(); 1135 PICKUP_GIANT(); 1136 mtx_lock(&mountlist_mtx); 1137 if (LIST_EMPTY(&root_holds)) { 1138 mtx_unlock(&mountlist_mtx); 1139 break; 1140 } 1141 printf("Root mount waiting for:"); 1142 LIST_FOREACH(h, &root_holds, list) 1143 printf(" %s", h->who); 1144 printf("\n"); 1145 msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold", 1146 hz); 1147 } 1148} 1149 1150static void 1151set_rootvnode(struct thread *td) 1152{ 1153 struct proc *p; 1154 1155 if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode, td)) 1156 panic("Cannot find root vnode"); 1157 1158 p = td->td_proc; 1159 FILEDESC_LOCK(p->p_fd); 1160 1161 if (p->p_fd->fd_cdir != NULL) 1162 vrele(p->p_fd->fd_cdir); 1163 p->p_fd->fd_cdir = rootvnode; 1164 VREF(rootvnode); 1165 1166 if (p->p_fd->fd_rdir != NULL) 1167 vrele(p->p_fd->fd_rdir); 1168 p->p_fd->fd_rdir = rootvnode; 1169 VREF(rootvnode); 1170 1171 FILEDESC_UNLOCK(p->p_fd); 1172 1173 VOP_UNLOCK(rootvnode, 0, td); 1174} 1175 1176/* 1177 * Mount /devfs as our root filesystem, but do not put it on the mountlist 1178 * yet. Create a /dev -> / symlink so that absolute pathnames will lookup. 1179 */ 1180 1181static void 1182devfs_first(void) 1183{ 1184 struct thread *td = curthread; 1185 struct vfsconf *vfsp; 1186 struct mount *mp = NULL; 1187 int error; 1188 1189 vfsp = vfs_byname("devfs"); 1190 KASSERT(vfsp != NULL, ("Could not find devfs by name")); 1191 if (vfsp == NULL) 1192 return; 1193 1194 error = vfs_mount_alloc(NULLVP, vfsp, "/dev", td, &mp); 1195 KASSERT(error == 0, ("vfs_mount_alloc failed %d", error)); 1196 if (error) 1197 return; 1198 1199 error = VFS_MOUNT(mp, curthread); 1200 KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error)); 1201 if (error) 1202 return; 1203 1204 mtx_lock(&mountlist_mtx); 1205 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list); 1206 mtx_unlock(&mountlist_mtx); 1207 1208 set_rootvnode(td); 1209 1210 error = kern_symlink(td, "/", "dev", UIO_SYSSPACE); 1211 if (error) 1212 printf("kern_symlink /dev -> / returns %d\n", error); 1213} 1214 1215/* 1216 * Surgically move our devfs to be mounted on /dev. 1217 */ 1218 1219static void 1220devfs_fixup(struct thread *td) 1221{ 1222 struct nameidata nd; 1223 int error; 1224 struct vnode *vp, *dvp; 1225 struct mount *mp; 1226 1227 /* Remove our devfs mount from the mountlist and purge the cache */ 1228 mtx_lock(&mountlist_mtx); 1229 mp = TAILQ_FIRST(&mountlist); 1230 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1231 mtx_unlock(&mountlist_mtx); 1232 cache_purgevfs(mp); 1233 1234 VFS_ROOT(mp, LK_EXCLUSIVE, &dvp, td); 1235 VI_LOCK(dvp); 1236 dvp->v_iflag &= ~VI_MOUNT; 1237 dvp->v_mountedhere = NULL; 1238 VI_UNLOCK(dvp); 1239 1240 /* Set up the real rootvnode, and purge the cache */ 1241 TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL; 1242 set_rootvnode(td); 1243 cache_purgevfs(rootvnode->v_mount); 1244 1245 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td); 1246 error = namei(&nd); 1247 if (error) { 1248 printf("Lookup of /dev for devfs, error: %d\n", error); 1249 return; 1250 } 1251 NDFREE(&nd, NDF_ONLY_PNBUF); 1252 vp = nd.ni_vp; 1253 if (vp->v_type != VDIR) { 1254 vput(vp); 1255 } 1256 error = vinvalbuf(vp, V_SAVE, td, 0, 0); 1257 if (error) { 1258 vput(vp); 1259 } 1260 cache_purge(vp); 1261 mp->mnt_vnodecovered = vp; 1262 vp->v_mountedhere = mp; 1263 mtx_lock(&mountlist_mtx); 1264 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1265 mtx_unlock(&mountlist_mtx); 1266 VOP_UNLOCK(vp, 0, td); 1267 vfs_unbusy(mp, td); 1268 vput(dvp); 1269 1270 /* Unlink the no longer needed /dev/dev -> / symlink */ 1271 kern_unlink(td, "/dev/dev", UIO_SYSSPACE); 1272} 1273 1274/* 1275 * Report errors during filesystem mounting. 1276 */ 1277void 1278vfs_mount_error(struct mount *mp, const char *fmt, ...) 1279{ 1280 struct vfsoptlist *moptlist = mp->mnt_optnew; 1281 va_list ap; 1282 int error, len; 1283 char *errmsg; 1284 1285 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len); 1286 if (error || errmsg == NULL || len <= 0) 1287 return; 1288 1289 va_start(ap, fmt); 1290 vsnprintf(errmsg, (size_t)len, fmt, ap); 1291 va_end(ap); 1292} 1293 1294/* 1295 * Find and mount the root filesystem 1296 */ 1297void 1298vfs_mountroot(void) 1299{ 1300 char *cp; 1301 int error, i, asked = 0; 1302 1303 root_mount_wait(); 1304 1305 devfs_first(); 1306 1307 /* 1308 * We are booted with instructions to prompt for the root filesystem. 1309 */ 1310 if (boothowto & RB_ASKNAME) { 1311 if (!vfs_mountroot_ask()) 1312 return; 1313 asked = 1; 1314 } 1315 1316 /* 1317 * The root filesystem information is compiled in, and we are 1318 * booted with instructions to use it. 1319 */ 1320 if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) { 1321 if (!vfs_mountroot_try(ctrootdevname)) 1322 return; 1323 ctrootdevname = NULL; 1324 } 1325 1326 /* 1327 * We've been given the generic "use CDROM as root" flag. This is 1328 * necessary because one media may be used in many different 1329 * devices, so we need to search for them. 1330 */ 1331 if (boothowto & RB_CDROM) { 1332 for (i = 0; cdrom_rootdevnames[i] != NULL; i++) { 1333 if (!vfs_mountroot_try(cdrom_rootdevnames[i])) 1334 return; 1335 } 1336 } 1337 1338 /* 1339 * Try to use the value read by the loader from /etc/fstab, or 1340 * supplied via some other means. This is the preferred 1341 * mechanism. 1342 */ 1343 cp = getenv("vfs.root.mountfrom"); 1344 if (cp != NULL) { 1345 error = vfs_mountroot_try(cp); 1346 freeenv(cp); 1347 if (!error) 1348 return; 1349 } 1350 1351 /* 1352 * Try values that may have been computed by code during boot 1353 */ 1354 if (!vfs_mountroot_try(rootdevnames[0])) 1355 return; 1356 if (!vfs_mountroot_try(rootdevnames[1])) 1357 return; 1358 1359 /* 1360 * If we (still) have a compiled-in default, try it. 1361 */ 1362 if (ctrootdevname != NULL) 1363 if (!vfs_mountroot_try(ctrootdevname)) 1364 return; 1365 /* 1366 * Everything so far has failed, prompt on the console if we haven't 1367 * already tried that. 1368 */ 1369 if (!asked) 1370 if (!vfs_mountroot_ask()) 1371 return; 1372 1373 panic("Root mount failed, startup aborted."); 1374} 1375 1376/* 1377 * Mount (mountfrom) as the root filesystem. 1378 */ 1379static int 1380vfs_mountroot_try(const char *mountfrom) 1381{ 1382 struct mount *mp; 1383 char *vfsname, *path; 1384 time_t timebase; 1385 int error; 1386 char patt[32]; 1387 1388 vfsname = NULL; 1389 path = NULL; 1390 mp = NULL; 1391 error = EINVAL; 1392 1393 if (mountfrom == NULL) 1394 return (error); /* don't complain */ 1395 printf("Trying to mount root from %s\n", mountfrom); 1396 1397 /* parse vfs name and path */ 1398 vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK); 1399 path = malloc(MNAMELEN, M_MOUNT, M_WAITOK); 1400 vfsname[0] = path[0] = 0; 1401 sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN); 1402 if (sscanf(mountfrom, patt, vfsname, path) < 1) 1403 goto out; 1404 1405 if (path[0] == '\0') 1406 strcpy(path, ROOTNAME); 1407 1408 error = kernel_vmount( 1409 MNT_RDONLY | MNT_ROOTFS, 1410 "fstype", vfsname, 1411 "fspath", "/", 1412 "from", path, 1413 NULL); 1414 if (error == 0) { 1415 /* 1416 * We mount devfs prior to mounting the / FS, so the first 1417 * entry will typically be devfs. 1418 */ 1419 mp = TAILQ_FIRST(&mountlist); 1420 KASSERT(mp != NULL, ("%s: mountlist is empty", __func__)); 1421 1422 /* 1423 * Iterate over all currently mounted file systems and use 1424 * the time stamp found to check and/or initialize the RTC. 1425 * Typically devfs has no time stamp and the only other FS 1426 * is the actual / FS. 1427 * Call inittodr() only once and pass it the largest of the 1428 * timestamps we encounter. 1429 */ 1430 timebase = 0; 1431 do { 1432 if (mp->mnt_time > timebase) 1433 timebase = mp->mnt_time; 1434 mp = TAILQ_NEXT(mp, mnt_list); 1435 } while (mp != NULL); 1436 inittodr(timebase); 1437 1438 devfs_fixup(curthread); 1439 } 1440out: 1441 free(path, M_MOUNT); 1442 free(vfsname, M_MOUNT); 1443 return (error); 1444} 1445 1446/* 1447 * --------------------------------------------------------------------- 1448 * Interactive root filesystem selection code. 1449 */ 1450 1451static int 1452vfs_mountroot_ask(void) 1453{ 1454 char name[128]; 1455 1456 for(;;) { 1457 printf("\nManual root filesystem specification:\n"); 1458 printf(" <fstype>:<device> Mount <device> using filesystem <fstype>\n"); 1459#if defined(__i386__) || defined(__ia64__) 1460 printf(" eg. ufs:da0s1a\n"); 1461#else 1462 printf(" eg. ufs:/dev/da0a\n"); 1463#endif 1464 printf(" ? List valid disk boot devices\n"); 1465 printf(" <empty line> Abort manual input\n"); 1466 printf("\nmountroot> "); 1467 gets(name, sizeof(name), 1); 1468 if (name[0] == '\0') 1469 return (1); 1470 if (name[0] == '?') { 1471 printf("\nList of GEOM managed disk devices:\n "); 1472 g_dev_print(); 1473 continue; 1474 } 1475 if (!vfs_mountroot_try(name)) 1476 return (0); 1477 } 1478} 1479 1480/* 1481 * --------------------------------------------------------------------- 1482 * Functions for querying mount options/arguments from filesystems. 1483 */ 1484 1485/* 1486 * Check that no unknown options are given 1487 */ 1488int 1489vfs_filteropt(struct vfsoptlist *opts, const char **legal) 1490{ 1491 struct vfsopt *opt; 1492 const char **t, *p; 1493 1494 1495 TAILQ_FOREACH(opt, opts, link) { 1496 p = opt->name; 1497 if (p[0] == 'n' && p[1] == 'o') 1498 p += 2; 1499 for(t = global_opts; *t != NULL; t++) 1500 if (!strcmp(*t, p)) 1501 break; 1502 if (*t != NULL) 1503 continue; 1504 for(t = legal; *t != NULL; t++) 1505 if (!strcmp(*t, p)) 1506 break; 1507 if (*t != NULL) 1508 continue; 1509 printf("mount option <%s> is unknown\n", p); 1510 return (EINVAL); 1511 } 1512 return (0); 1513} 1514 1515/* 1516 * Get a mount option by its name. 1517 * 1518 * Return 0 if the option was found, ENOENT otherwise. 1519 * If len is non-NULL it will be filled with the length 1520 * of the option. If buf is non-NULL, it will be filled 1521 * with the address of the option. 1522 */ 1523int 1524vfs_getopt(opts, name, buf, len) 1525 struct vfsoptlist *opts; 1526 const char *name; 1527 void **buf; 1528 int *len; 1529{ 1530 struct vfsopt *opt; 1531 1532 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1533 1534 TAILQ_FOREACH(opt, opts, link) { 1535 if (strcmp(name, opt->name) == 0) { 1536 if (len != NULL) 1537 *len = opt->len; 1538 if (buf != NULL) 1539 *buf = opt->value; 1540 return (0); 1541 } 1542 } 1543 return (ENOENT); 1544} 1545 1546char * 1547vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) 1548{ 1549 struct vfsopt *opt; 1550 1551 *error = 0; 1552 TAILQ_FOREACH(opt, opts, link) { 1553 if (strcmp(name, opt->name) != 0) 1554 continue; 1555 if (((char *)opt->value)[opt->len - 1] != '\0') { 1556 *error = EINVAL; 1557 return (NULL); 1558 } 1559 return (opt->value); 1560 } 1561 return (NULL); 1562} 1563 1564int 1565vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val) 1566{ 1567 struct vfsopt *opt; 1568 1569 TAILQ_FOREACH(opt, opts, link) { 1570 if (strcmp(name, opt->name) == 0) { 1571 if (w != NULL) 1572 *w |= val; 1573 return (1); 1574 } 1575 } 1576 if (w != NULL) 1577 *w &= ~val; 1578 return (0); 1579} 1580 1581int 1582vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) 1583{ 1584 va_list ap; 1585 struct vfsopt *opt; 1586 int ret; 1587 1588 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 1589 1590 TAILQ_FOREACH(opt, opts, link) { 1591 if (strcmp(name, opt->name) != 0) 1592 continue; 1593 if (((char *)opt->value)[opt->len - 1] != '\0') 1594 return (0); 1595 va_start(ap, fmt); 1596 ret = vsscanf(opt->value, fmt, ap); 1597 va_end(ap); 1598 return (ret); 1599 } 1600 return (0); 1601} 1602 1603/* 1604 * Find and copy a mount option. 1605 * 1606 * The size of the buffer has to be specified 1607 * in len, if it is not the same length as the 1608 * mount option, EINVAL is returned. 1609 * Returns ENOENT if the option is not found. 1610 */ 1611int 1612vfs_copyopt(opts, name, dest, len) 1613 struct vfsoptlist *opts; 1614 const char *name; 1615 void *dest; 1616 int len; 1617{ 1618 struct vfsopt *opt; 1619 1620 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 1621 1622 TAILQ_FOREACH(opt, opts, link) { 1623 if (strcmp(name, opt->name) == 0) { 1624 if (len != opt->len) 1625 return (EINVAL); 1626 bcopy(opt->value, dest, opt->len); 1627 return (0); 1628 } 1629 } 1630 return (ENOENT); 1631} 1632 1633/* 1634 * This is a helper function for filesystems to traverse their 1635 * vnodes. See MNT_VNODE_FOREACH() in sys/mount.h 1636 */ 1637 1638struct vnode * 1639__mnt_vnode_next(struct vnode **nvp, struct mount *mp) 1640{ 1641 struct vnode *vp; 1642 1643 mtx_assert(&mp->mnt_mtx, MA_OWNED); 1644 1645 vp = *nvp; 1646 /* Check if we are done */ 1647 if (vp == NULL) 1648 return (NULL); 1649 /* If our next vnode is no longer ours, start over */ 1650 if (vp->v_mount != mp) 1651 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 1652 /* Save pointer to next vnode in list */ 1653 if (vp != NULL) 1654 *nvp = TAILQ_NEXT(vp, v_nmntvnodes); 1655 else 1656 *nvp = NULL; 1657 return (vp); 1658} 1659 1660int 1661__vfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td) 1662{ 1663 int error; 1664 1665 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat, td); 1666 if (sbp != &mp->mnt_stat) 1667 *sbp = mp->mnt_stat; 1668 return (error); 1669} 1670 1671void 1672vfs_mountedfrom(struct mount *mp, const char *from) 1673{ 1674 1675 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); 1676 strlcpy(mp->mnt_stat.f_mntfromname, from, 1677 sizeof mp->mnt_stat.f_mntfromname); 1678} 1679 1680/* 1681 * --------------------------------------------------------------------- 1682 * This is the api for building mount args and mounting filesystems from 1683 * inside the kernel. 1684 * 1685 * The API works by accumulation of individual args. First error is 1686 * latched. 1687 * 1688 * XXX: should be documented in new manpage kernel_mount(9) 1689 */ 1690 1691/* A memory allocation which must be freed when we are done */ 1692struct mntaarg { 1693 SLIST_ENTRY(mntaarg) next; 1694}; 1695 1696/* The header for the mount arguments */ 1697struct mntarg { 1698 struct iovec *v; 1699 int len; 1700 int error; 1701 SLIST_HEAD(, mntaarg) list; 1702}; 1703 1704/* 1705 * Add a boolean argument. 1706 * 1707 * flag is the boolean value. 1708 * name must start with "no". 1709 */ 1710struct mntarg * 1711mount_argb(struct mntarg *ma, int flag, const char *name) 1712{ 1713 1714 KASSERT(name[0] == 'n' && name[1] == 'o', 1715 ("mount_argb(...,%s): name must start with 'no'", name)); 1716 1717 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 1718} 1719 1720/* 1721 * Add an argument printf style 1722 */ 1723struct mntarg * 1724mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 1725{ 1726 va_list ap; 1727 struct mntaarg *maa; 1728 struct sbuf *sb; 1729 int len; 1730 1731 if (ma == NULL) { 1732 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 1733 SLIST_INIT(&ma->list); 1734 } 1735 if (ma->error) 1736 return (ma); 1737 1738 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 1739 M_MOUNT, M_WAITOK); 1740 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 1741 ma->v[ma->len].iov_len = strlen(name) + 1; 1742 ma->len++; 1743 1744 sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND); 1745 va_start(ap, fmt); 1746 sbuf_vprintf(sb, fmt, ap); 1747 va_end(ap); 1748 sbuf_finish(sb); 1749 len = sbuf_len(sb) + 1; 1750 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 1751 SLIST_INSERT_HEAD(&ma->list, maa, next); 1752 bcopy(sbuf_data(sb), maa + 1, len); 1753 sbuf_delete(sb); 1754 1755 ma->v[ma->len].iov_base = maa + 1; 1756 ma->v[ma->len].iov_len = len; 1757 ma->len++; 1758 1759 return (ma); 1760} 1761 1762/* 1763 * Add an argument which is a userland string. 1764 */ 1765struct mntarg * 1766mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 1767{ 1768 struct mntaarg *maa; 1769 char *tbuf; 1770 1771 if (val == NULL) 1772 return (ma); 1773 if (ma == NULL) { 1774 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 1775 SLIST_INIT(&ma->list); 1776 } 1777 if (ma->error) 1778 return (ma); 1779 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 1780 SLIST_INSERT_HEAD(&ma->list, maa, next); 1781 tbuf = (void *)(maa + 1); 1782 ma->error = copyinstr(val, tbuf, len, NULL); 1783 return (mount_arg(ma, name, tbuf, -1)); 1784} 1785 1786/* 1787 * Plain argument. 1788 * 1789 * If length is -1, use printf. 1790 */ 1791struct mntarg * 1792mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 1793{ 1794 1795 if (ma == NULL) { 1796 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 1797 SLIST_INIT(&ma->list); 1798 } 1799 if (ma->error) 1800 return (ma); 1801 1802 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 1803 M_MOUNT, M_WAITOK); 1804 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 1805 ma->v[ma->len].iov_len = strlen(name) + 1; 1806 ma->len++; 1807 1808 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 1809 if (len < 0) 1810 ma->v[ma->len].iov_len = strlen(val) + 1; 1811 else 1812 ma->v[ma->len].iov_len = len; 1813 ma->len++; 1814 return (ma); 1815} 1816 1817/* 1818 * Free a mntarg structure 1819 */ 1820static void 1821free_mntarg(struct mntarg *ma) 1822{ 1823 struct mntaarg *maa; 1824 1825 while (!SLIST_EMPTY(&ma->list)) { 1826 maa = SLIST_FIRST(&ma->list); 1827 SLIST_REMOVE_HEAD(&ma->list, next); 1828 free(maa, M_MOUNT); 1829 } 1830 free(ma->v, M_MOUNT); 1831 free(ma, M_MOUNT); 1832} 1833 1834/* 1835 * Mount a filesystem 1836 */ 1837int 1838kernel_mount(struct mntarg *ma, int flags) 1839{ 1840 struct uio auio; 1841 int error; 1842 1843 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 1844 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v")); 1845 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 1846 1847 auio.uio_iov = ma->v; 1848 auio.uio_iovcnt = ma->len; 1849 auio.uio_segflg = UIO_SYSSPACE; 1850 1851 error = ma->error; 1852 if (!error) 1853 error = vfs_donmount(curthread, flags, &auio); 1854 free_mntarg(ma); 1855 return (error); 1856} 1857 1858/* 1859 * A printflike function to mount a filesystem. 1860 */ 1861int 1862kernel_vmount(int flags, ...) 1863{ 1864 struct mntarg *ma = NULL; 1865 va_list ap; 1866 const char *cp; 1867 const void *vp; 1868 int error; 1869 1870 va_start(ap, flags); 1871 for (;;) { 1872 cp = va_arg(ap, const char *); 1873 if (cp == NULL) 1874 break; 1875 vp = va_arg(ap, const void *); 1876 ma = mount_arg(ma, cp, vp, -1); 1877 } 1878 va_end(ap); 1879 1880 error = kernel_mount(ma, flags); 1881 return (error); 1882} 1883