vfs_mount.c revision 212356
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 212356 2010-09-09 07:55:13Z pjd $"); 39 40#include <sys/param.h> 41#include <sys/conf.h> 42#include <sys/fcntl.h> 43#include <sys/jail.h> 44#include <sys/kernel.h> 45#include <sys/libkern.h> 46#include <sys/malloc.h> 47#include <sys/mount.h> 48#include <sys/mutex.h> 49#include <sys/namei.h> 50#include <sys/priv.h> 51#include <sys/proc.h> 52#include <sys/filedesc.h> 53#include <sys/reboot.h> 54#include <sys/syscallsubr.h> 55#include <sys/sysproto.h> 56#include <sys/sx.h> 57#include <sys/sysctl.h> 58#include <sys/sysent.h> 59#include <sys/systm.h> 60#include <sys/vnode.h> 61#include <vm/uma.h> 62 63#include <geom/geom.h> 64 65#include <machine/stdarg.h> 66 67#include <security/audit/audit.h> 68#include <security/mac/mac_framework.h> 69 70#include "opt_rootdevname.h" 71 72#define ROOTNAME "root_device" 73#define VFS_MOUNTARG_SIZE_MAX (1024 * 64) 74 75static void set_rootvnode(void); 76static int vfs_domount(struct thread *td, const char *fstype, 77 char *fspath, int fsflags, void *fsdata); 78static int vfs_mountroot_ask(void); 79static int vfs_mountroot_try(const char *mountfrom, const char *options); 80static void free_mntarg(struct mntarg *ma); 81 82static int usermount = 0; 83SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, 84 "Unprivileged users may mount and unmount file systems"); 85 86MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure"); 87MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker"); 88static uma_zone_t mount_zone; 89 90/* List of mounted filesystems. */ 91struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); 92 93/* For any iteration/modification of mountlist */ 94struct mtx mountlist_mtx; 95MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF); 96 97/* 98 * The vnode of the system's root (/ in the filesystem, without chroot 99 * active.) 100 */ 101struct vnode *rootvnode; 102 103/* 104 * The root filesystem is detailed in the kernel environment variable 105 * vfs.root.mountfrom, which is expected to be in the general format 106 * 107 * <vfsname>:[<path>][ <vfsname>:[<path>] ...] 108 * vfsname := the name of a VFS known to the kernel and capable 109 * of being mounted as root 110 * path := disk device name or other data used by the filesystem 111 * to locate its physical store 112 * 113 * If the environment variable vfs.root.mountfrom is a space separated list, 114 * each list element is tried in turn and the root filesystem will be mounted 115 * from the first one that suceeds. 116 * 117 * The environment variable vfs.root.mountfrom.options is a comma delimited 118 * set of string mount options. These mount options must be parseable 119 * by nmount() in the kernel. 120 */ 121 122/* 123 * Global opts, taken by all filesystems 124 */ 125static const char *global_opts[] = { 126 "errmsg", 127 "fstype", 128 "fspath", 129 "ro", 130 "rw", 131 "nosuid", 132 "noexec", 133 NULL 134}; 135 136/* 137 * The root specifiers we will try if RB_CDROM is specified. 138 */ 139static char *cdrom_rootdevnames[] = { 140 "cd9660:cd0", 141 "cd9660:acd0", 142 NULL 143}; 144 145/* legacy find-root code */ 146char *rootdevnames[2] = {NULL, NULL}; 147#ifndef ROOTDEVNAME 148# define ROOTDEVNAME NULL 149#endif 150static const char *ctrootdevname = ROOTDEVNAME; 151 152/* 153 * --------------------------------------------------------------------- 154 * Functions for building and sanitizing the mount options 155 */ 156 157/* Remove one mount option. */ 158static void 159vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt) 160{ 161 162 TAILQ_REMOVE(opts, opt, link); 163 free(opt->name, M_MOUNT); 164 if (opt->value != NULL) 165 free(opt->value, M_MOUNT); 166 free(opt, M_MOUNT); 167} 168 169/* Release all resources related to the mount options. */ 170void 171vfs_freeopts(struct vfsoptlist *opts) 172{ 173 struct vfsopt *opt; 174 175 while (!TAILQ_EMPTY(opts)) { 176 opt = TAILQ_FIRST(opts); 177 vfs_freeopt(opts, opt); 178 } 179 free(opts, M_MOUNT); 180} 181 182void 183vfs_deleteopt(struct vfsoptlist *opts, const char *name) 184{ 185 struct vfsopt *opt, *temp; 186 187 if (opts == NULL) 188 return; 189 TAILQ_FOREACH_SAFE(opt, opts, link, temp) { 190 if (strcmp(opt->name, name) == 0) 191 vfs_freeopt(opts, opt); 192 } 193} 194 195/* 196 * Check if options are equal (with or without the "no" prefix). 197 */ 198static int 199vfs_equalopts(const char *opt1, const char *opt2) 200{ 201 char *p; 202 203 /* "opt" vs. "opt" or "noopt" vs. "noopt" */ 204 if (strcmp(opt1, opt2) == 0) 205 return (1); 206 /* "noopt" vs. "opt" */ 207 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 208 return (1); 209 /* "opt" vs. "noopt" */ 210 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 211 return (1); 212 while ((p = strchr(opt1, '.')) != NULL && 213 !strncmp(opt1, opt2, ++p - opt1)) { 214 opt2 += p - opt1; 215 opt1 = p; 216 /* "foo.noopt" vs. "foo.opt" */ 217 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0) 218 return (1); 219 /* "foo.opt" vs. "foo.noopt" */ 220 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0) 221 return (1); 222 } 223 return (0); 224} 225 226/* 227 * If a mount option is specified several times, 228 * (with or without the "no" prefix) only keep 229 * the last occurence of it. 230 */ 231static void 232vfs_sanitizeopts(struct vfsoptlist *opts) 233{ 234 struct vfsopt *opt, *opt2, *tmp; 235 236 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) { 237 opt2 = TAILQ_PREV(opt, vfsoptlist, link); 238 while (opt2 != NULL) { 239 if (vfs_equalopts(opt->name, opt2->name)) { 240 tmp = TAILQ_PREV(opt2, vfsoptlist, link); 241 vfs_freeopt(opts, opt2); 242 opt2 = tmp; 243 } else { 244 opt2 = TAILQ_PREV(opt2, vfsoptlist, link); 245 } 246 } 247 } 248} 249 250/* 251 * Build a linked list of mount options from a struct uio. 252 */ 253int 254vfs_buildopts(struct uio *auio, struct vfsoptlist **options) 255{ 256 struct vfsoptlist *opts; 257 struct vfsopt *opt; 258 size_t memused, namelen, optlen; 259 unsigned int i, iovcnt; 260 int error; 261 262 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 263 TAILQ_INIT(opts); 264 memused = 0; 265 iovcnt = auio->uio_iovcnt; 266 for (i = 0; i < iovcnt; i += 2) { 267 namelen = auio->uio_iov[i].iov_len; 268 optlen = auio->uio_iov[i + 1].iov_len; 269 memused += sizeof(struct vfsopt) + optlen + namelen; 270 /* 271 * Avoid consuming too much memory, and attempts to overflow 272 * memused. 273 */ 274 if (memused > VFS_MOUNTARG_SIZE_MAX || 275 optlen > VFS_MOUNTARG_SIZE_MAX || 276 namelen > VFS_MOUNTARG_SIZE_MAX) { 277 error = EINVAL; 278 goto bad; 279 } 280 281 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 282 opt->name = malloc(namelen, M_MOUNT, M_WAITOK); 283 opt->value = NULL; 284 opt->len = 0; 285 opt->pos = i / 2; 286 opt->seen = 0; 287 288 /* 289 * Do this early, so jumps to "bad" will free the current 290 * option. 291 */ 292 TAILQ_INSERT_TAIL(opts, opt, link); 293 294 if (auio->uio_segflg == UIO_SYSSPACE) { 295 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen); 296 } else { 297 error = copyin(auio->uio_iov[i].iov_base, opt->name, 298 namelen); 299 if (error) 300 goto bad; 301 } 302 /* Ensure names are null-terminated strings. */ 303 if (namelen == 0 || opt->name[namelen - 1] != '\0') { 304 error = EINVAL; 305 goto bad; 306 } 307 if (optlen != 0) { 308 opt->len = optlen; 309 opt->value = malloc(optlen, M_MOUNT, M_WAITOK); 310 if (auio->uio_segflg == UIO_SYSSPACE) { 311 bcopy(auio->uio_iov[i + 1].iov_base, opt->value, 312 optlen); 313 } else { 314 error = copyin(auio->uio_iov[i + 1].iov_base, 315 opt->value, optlen); 316 if (error) 317 goto bad; 318 } 319 } 320 } 321 vfs_sanitizeopts(opts); 322 *options = opts; 323 return (0); 324bad: 325 vfs_freeopts(opts); 326 return (error); 327} 328 329/* 330 * Merge the old mount options with the new ones passed 331 * in the MNT_UPDATE case. 332 * 333 * XXX This function will keep a "nofoo" option in the 334 * new options if there is no matching "foo" option 335 * to be cancelled in the old options. This is a bug 336 * if the option's canonical name is "foo". E.g., "noro" 337 * shouldn't end up in the mount point's active options, 338 * but it can. 339 */ 340static void 341vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts) 342{ 343 struct vfsopt *opt, *opt2, *new; 344 345 TAILQ_FOREACH(opt, opts, link) { 346 /* 347 * Check that this option hasn't been redefined 348 * nor cancelled with a "no" mount option. 349 */ 350 opt2 = TAILQ_FIRST(toopts); 351 while (opt2 != NULL) { 352 if (strcmp(opt2->name, opt->name) == 0) 353 goto next; 354 if (strncmp(opt2->name, "no", 2) == 0 && 355 strcmp(opt2->name + 2, opt->name) == 0) { 356 vfs_freeopt(toopts, opt2); 357 goto next; 358 } 359 opt2 = TAILQ_NEXT(opt2, link); 360 } 361 /* We want this option, duplicate it. */ 362 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 363 new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK); 364 strcpy(new->name, opt->name); 365 if (opt->len != 0) { 366 new->value = malloc(opt->len, M_MOUNT, M_WAITOK); 367 bcopy(opt->value, new->value, opt->len); 368 } else { 369 new->value = NULL; 370 } 371 new->len = opt->len; 372 new->seen = opt->seen; 373 TAILQ_INSERT_TAIL(toopts, new, link); 374next: 375 continue; 376 } 377} 378 379/* 380 * Mount a filesystem. 381 */ 382int 383nmount(td, uap) 384 struct thread *td; 385 struct nmount_args /* { 386 struct iovec *iovp; 387 unsigned int iovcnt; 388 int flags; 389 } */ *uap; 390{ 391 struct uio *auio; 392 int error; 393 u_int iovcnt; 394 395 AUDIT_ARG_FFLAGS(uap->flags); 396 CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__, 397 uap->iovp, uap->iovcnt, uap->flags); 398 399 /* 400 * Filter out MNT_ROOTFS. We do not want clients of nmount() in 401 * userspace to set this flag, but we must filter it out if we want 402 * MNT_UPDATE on the root file system to work. 403 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try(). 404 */ 405 uap->flags &= ~MNT_ROOTFS; 406 407 iovcnt = uap->iovcnt; 408 /* 409 * Check that we have an even number of iovec's 410 * and that we have at least two options. 411 */ 412 if ((iovcnt & 1) || (iovcnt < 4)) { 413 CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__, 414 uap->iovcnt); 415 return (EINVAL); 416 } 417 418 error = copyinuio(uap->iovp, iovcnt, &auio); 419 if (error) { 420 CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno", 421 __func__, error); 422 return (error); 423 } 424 error = vfs_donmount(td, uap->flags, auio); 425 426 free(auio, M_IOV); 427 return (error); 428} 429 430/* 431 * --------------------------------------------------------------------- 432 * Various utility functions 433 */ 434 435void 436vfs_ref(struct mount *mp) 437{ 438 439 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 440 MNT_ILOCK(mp); 441 MNT_REF(mp); 442 MNT_IUNLOCK(mp); 443} 444 445void 446vfs_rel(struct mount *mp) 447{ 448 449 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 450 MNT_ILOCK(mp); 451 MNT_REL(mp); 452 MNT_IUNLOCK(mp); 453} 454 455static int 456mount_init(void *mem, int size, int flags) 457{ 458 struct mount *mp; 459 460 mp = (struct mount *)mem; 461 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF); 462 lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0); 463 return (0); 464} 465 466static void 467mount_fini(void *mem, int size) 468{ 469 struct mount *mp; 470 471 mp = (struct mount *)mem; 472 lockdestroy(&mp->mnt_explock); 473 mtx_destroy(&mp->mnt_mtx); 474} 475 476/* 477 * Allocate and initialize the mount point struct. 478 */ 479struct mount * 480vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath, 481 struct ucred *cred) 482{ 483 struct mount *mp; 484 485 mp = uma_zalloc(mount_zone, M_WAITOK); 486 bzero(&mp->mnt_startzero, 487 __rangeof(struct mount, mnt_startzero, mnt_endzero)); 488 TAILQ_INIT(&mp->mnt_nvnodelist); 489 mp->mnt_nvnodelistsize = 0; 490 mp->mnt_ref = 0; 491 (void) vfs_busy(mp, MBF_NOWAIT); 492 mp->mnt_op = vfsp->vfc_vfsops; 493 mp->mnt_vfc = vfsp; 494 vfsp->vfc_refcount++; /* XXX Unlocked */ 495 mp->mnt_stat.f_type = vfsp->vfc_typenum; 496 mp->mnt_gen++; 497 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 498 mp->mnt_vnodecovered = vp; 499 mp->mnt_cred = crdup(cred); 500 mp->mnt_stat.f_owner = cred->cr_uid; 501 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN); 502 mp->mnt_iosize_max = DFLTPHYS; 503#ifdef MAC 504 mac_mount_init(mp); 505 mac_mount_create(cred, mp); 506#endif 507 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0); 508 return (mp); 509} 510 511/* 512 * Destroy the mount struct previously allocated by vfs_mount_alloc(). 513 */ 514void 515vfs_mount_destroy(struct mount *mp) 516{ 517 518 MNT_ILOCK(mp); 519 mp->mnt_kern_flag |= MNTK_REFEXPIRE; 520 if (mp->mnt_kern_flag & MNTK_MWAIT) { 521 mp->mnt_kern_flag &= ~MNTK_MWAIT; 522 wakeup(mp); 523 } 524 while (mp->mnt_ref) 525 msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0); 526 KASSERT(mp->mnt_ref == 0, 527 ("%s: invalid refcount in the drain path @ %s:%d", __func__, 528 __FILE__, __LINE__)); 529 if (mp->mnt_writeopcount != 0) 530 panic("vfs_mount_destroy: nonzero writeopcount"); 531 if (mp->mnt_secondary_writes != 0) 532 panic("vfs_mount_destroy: nonzero secondary_writes"); 533 mp->mnt_vfc->vfc_refcount--; 534 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) { 535 struct vnode *vp; 536 537 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) 538 vprint("", vp); 539 panic("unmount: dangling vnode"); 540 } 541 if (mp->mnt_nvnodelistsize != 0) 542 panic("vfs_mount_destroy: nonzero nvnodelistsize"); 543 if (mp->mnt_lockref != 0) 544 panic("vfs_mount_destroy: nonzero lock refcount"); 545 MNT_IUNLOCK(mp); 546#ifdef MAC 547 mac_mount_destroy(mp); 548#endif 549 if (mp->mnt_opt != NULL) 550 vfs_freeopts(mp->mnt_opt); 551 crfree(mp->mnt_cred); 552 uma_zfree(mount_zone, mp); 553} 554 555int 556vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions) 557{ 558 struct vfsoptlist *optlist; 559 struct vfsopt *opt, *noro_opt, *tmp_opt; 560 char *fstype, *fspath, *errmsg; 561 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos; 562 int has_rw, has_noro; 563 564 errmsg = fspath = NULL; 565 errmsg_len = has_noro = has_rw = fspathlen = 0; 566 errmsg_pos = -1; 567 568 error = vfs_buildopts(fsoptions, &optlist); 569 if (error) 570 return (error); 571 572 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0) 573 errmsg_pos = vfs_getopt_pos(optlist, "errmsg"); 574 575 /* 576 * We need these two options before the others, 577 * and they are mandatory for any filesystem. 578 * Ensure they are NUL terminated as well. 579 */ 580 fstypelen = 0; 581 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen); 582 if (error || fstype[fstypelen - 1] != '\0') { 583 error = EINVAL; 584 if (errmsg != NULL) 585 strncpy(errmsg, "Invalid fstype", errmsg_len); 586 goto bail; 587 } 588 fspathlen = 0; 589 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen); 590 if (error || fspath[fspathlen - 1] != '\0') { 591 error = EINVAL; 592 if (errmsg != NULL) 593 strncpy(errmsg, "Invalid fspath", errmsg_len); 594 goto bail; 595 } 596 597 /* 598 * We need to see if we have the "update" option 599 * before we call vfs_domount(), since vfs_domount() has special 600 * logic based on MNT_UPDATE. This is very important 601 * when we want to update the root filesystem. 602 */ 603 TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) { 604 if (strcmp(opt->name, "update") == 0) { 605 fsflags |= MNT_UPDATE; 606 vfs_freeopt(optlist, opt); 607 } 608 else if (strcmp(opt->name, "async") == 0) 609 fsflags |= MNT_ASYNC; 610 else if (strcmp(opt->name, "force") == 0) { 611 fsflags |= MNT_FORCE; 612 vfs_freeopt(optlist, opt); 613 } 614 else if (strcmp(opt->name, "reload") == 0) { 615 fsflags |= MNT_RELOAD; 616 vfs_freeopt(optlist, opt); 617 } 618 else if (strcmp(opt->name, "multilabel") == 0) 619 fsflags |= MNT_MULTILABEL; 620 else if (strcmp(opt->name, "noasync") == 0) 621 fsflags &= ~MNT_ASYNC; 622 else if (strcmp(opt->name, "noatime") == 0) 623 fsflags |= MNT_NOATIME; 624 else if (strcmp(opt->name, "atime") == 0) { 625 free(opt->name, M_MOUNT); 626 opt->name = strdup("nonoatime", M_MOUNT); 627 } 628 else if (strcmp(opt->name, "noclusterr") == 0) 629 fsflags |= MNT_NOCLUSTERR; 630 else if (strcmp(opt->name, "clusterr") == 0) { 631 free(opt->name, M_MOUNT); 632 opt->name = strdup("nonoclusterr", M_MOUNT); 633 } 634 else if (strcmp(opt->name, "noclusterw") == 0) 635 fsflags |= MNT_NOCLUSTERW; 636 else if (strcmp(opt->name, "clusterw") == 0) { 637 free(opt->name, M_MOUNT); 638 opt->name = strdup("nonoclusterw", M_MOUNT); 639 } 640 else if (strcmp(opt->name, "noexec") == 0) 641 fsflags |= MNT_NOEXEC; 642 else if (strcmp(opt->name, "exec") == 0) { 643 free(opt->name, M_MOUNT); 644 opt->name = strdup("nonoexec", M_MOUNT); 645 } 646 else if (strcmp(opt->name, "nosuid") == 0) 647 fsflags |= MNT_NOSUID; 648 else if (strcmp(opt->name, "suid") == 0) { 649 free(opt->name, M_MOUNT); 650 opt->name = strdup("nonosuid", M_MOUNT); 651 } 652 else if (strcmp(opt->name, "nosymfollow") == 0) 653 fsflags |= MNT_NOSYMFOLLOW; 654 else if (strcmp(opt->name, "symfollow") == 0) { 655 free(opt->name, M_MOUNT); 656 opt->name = strdup("nonosymfollow", M_MOUNT); 657 } 658 else if (strcmp(opt->name, "noro") == 0) { 659 fsflags &= ~MNT_RDONLY; 660 has_noro = 1; 661 } 662 else if (strcmp(opt->name, "rw") == 0) { 663 fsflags &= ~MNT_RDONLY; 664 has_rw = 1; 665 } 666 else if (strcmp(opt->name, "ro") == 0) 667 fsflags |= MNT_RDONLY; 668 else if (strcmp(opt->name, "rdonly") == 0) { 669 free(opt->name, M_MOUNT); 670 opt->name = strdup("ro", M_MOUNT); 671 fsflags |= MNT_RDONLY; 672 } 673 else if (strcmp(opt->name, "suiddir") == 0) 674 fsflags |= MNT_SUIDDIR; 675 else if (strcmp(opt->name, "sync") == 0) 676 fsflags |= MNT_SYNCHRONOUS; 677 else if (strcmp(opt->name, "union") == 0) 678 fsflags |= MNT_UNION; 679 } 680 681 /* 682 * If "rw" was specified as a mount option, and we 683 * are trying to update a mount-point from "ro" to "rw", 684 * we need a mount option "noro", since in vfs_mergeopts(), 685 * "noro" will cancel "ro", but "rw" will not do anything. 686 */ 687 if (has_rw && !has_noro) { 688 noro_opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK); 689 noro_opt->name = strdup("noro", M_MOUNT); 690 noro_opt->value = NULL; 691 noro_opt->len = 0; 692 noro_opt->pos = -1; 693 noro_opt->seen = 1; 694 TAILQ_INSERT_TAIL(optlist, noro_opt, link); 695 } 696 697 /* 698 * Be ultra-paranoid about making sure the type and fspath 699 * variables will fit in our mp buffers, including the 700 * terminating NUL. 701 */ 702 if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) { 703 error = ENAMETOOLONG; 704 goto bail; 705 } 706 707 error = vfs_domount(td, fstype, fspath, fsflags, optlist); 708bail: 709 /* copyout the errmsg */ 710 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt) 711 && errmsg_len > 0 && errmsg != NULL) { 712 if (fsoptions->uio_segflg == UIO_SYSSPACE) { 713 bcopy(errmsg, 714 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 715 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 716 } else { 717 copyout(errmsg, 718 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base, 719 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len); 720 } 721 } 722 723 if (error != 0) 724 vfs_freeopts(optlist); 725 return (error); 726} 727 728/* 729 * Old mount API. 730 */ 731#ifndef _SYS_SYSPROTO_H_ 732struct mount_args { 733 char *type; 734 char *path; 735 int flags; 736 caddr_t data; 737}; 738#endif 739/* ARGSUSED */ 740int 741mount(td, uap) 742 struct thread *td; 743 struct mount_args /* { 744 char *type; 745 char *path; 746 int flags; 747 caddr_t data; 748 } */ *uap; 749{ 750 char *fstype; 751 struct vfsconf *vfsp = NULL; 752 struct mntarg *ma = NULL; 753 int error; 754 755 AUDIT_ARG_FFLAGS(uap->flags); 756 757 /* 758 * Filter out MNT_ROOTFS. We do not want clients of mount() in 759 * userspace to set this flag, but we must filter it out if we want 760 * MNT_UPDATE on the root file system to work. 761 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try(). 762 */ 763 uap->flags &= ~MNT_ROOTFS; 764 765 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK); 766 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL); 767 if (error) { 768 free(fstype, M_TEMP); 769 return (error); 770 } 771 772 AUDIT_ARG_TEXT(fstype); 773 mtx_lock(&Giant); 774 vfsp = vfs_byname_kld(fstype, td, &error); 775 free(fstype, M_TEMP); 776 if (vfsp == NULL) { 777 mtx_unlock(&Giant); 778 return (ENOENT); 779 } 780 if (vfsp->vfc_vfsops->vfs_cmount == NULL) { 781 mtx_unlock(&Giant); 782 return (EOPNOTSUPP); 783 } 784 785 ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN); 786 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN); 787 ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro"); 788 ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid"); 789 ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec"); 790 791 error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags); 792 mtx_unlock(&Giant); 793 return (error); 794} 795 796/* 797 * vfs_domount_first(): first file system mount (not update) 798 */ 799static int 800vfs_domount_first( 801 struct thread *td, /* Calling thread. */ 802 struct vfsconf *vfsp, /* File system type. */ 803 char *fspath, /* Mount path. */ 804 struct vnode *vp, /* Vnode to be covered. */ 805 int fsflags, /* Flags common to all filesystems. */ 806 void *fsdata /* Options local to the filesystem. */ 807 ) 808{ 809 struct vattr va; 810 struct mount *mp; 811 struct vnode *newdp; 812 int error; 813 814 mtx_assert(&Giant, MA_OWNED); 815 ASSERT_VOP_ELOCKED(vp, __func__); 816 KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here")); 817 818 /* 819 * If the user is not root, ensure that they own the directory 820 * onto which we are attempting to mount. 821 */ 822 error = VOP_GETATTR(vp, &va, td->td_ucred); 823 if (error == 0 && va.va_uid != td->td_ucred->cr_uid) 824 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN, 0); 825 if (error == 0) 826 error = vinvalbuf(vp, V_SAVE, 0, 0); 827 if (error == 0 && vp->v_type != VDIR) 828 error = ENOTDIR; 829 if (error == 0) { 830 VI_LOCK(vp); 831 if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL) 832 vp->v_iflag |= VI_MOUNT; 833 else 834 error = EBUSY; 835 VI_UNLOCK(vp); 836 } 837 if (error != 0) { 838 vput(vp); 839 return (error); 840 } 841 VOP_UNLOCK(vp, 0); 842 843 /* Allocate and initialize the filesystem. */ 844 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred); 845 /* XXXMAC: pass to vfs_mount_alloc? */ 846 mp->mnt_optnew = fsdata; 847 /* Set the mount level flags. */ 848 mp->mnt_flag = (fsflags & (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY)); 849 850 /* 851 * Mount the filesystem. 852 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 853 * get. No freeing of cn_pnbuf. 854 */ 855 error = VFS_MOUNT(mp); 856 if (error != 0) { 857 vfs_unbusy(mp); 858 vfs_mount_destroy(mp); 859 VI_LOCK(vp); 860 vp->v_iflag &= ~VI_MOUNT; 861 VI_UNLOCK(vp); 862 vrele(vp); 863 return (error); 864 } 865 866 if (mp->mnt_opt != NULL) 867 vfs_freeopts(mp->mnt_opt); 868 mp->mnt_opt = mp->mnt_optnew; 869 (void)VFS_STATFS(mp, &mp->mnt_stat); 870 871 /* 872 * Prevent external consumers of mount options from reading mnt_optnew. 873 */ 874 mp->mnt_optnew = NULL; 875 876 MNT_ILOCK(mp); 877 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 878 mp->mnt_kern_flag |= MNTK_ASYNC; 879 else 880 mp->mnt_kern_flag &= ~MNTK_ASYNC; 881 MNT_IUNLOCK(mp); 882 883 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 884 cache_purge(vp); 885 VI_LOCK(vp); 886 vp->v_iflag &= ~VI_MOUNT; 887 VI_UNLOCK(vp); 888 vp->v_mountedhere = mp; 889 /* Place the new filesystem at the end of the mount list. */ 890 mtx_lock(&mountlist_mtx); 891 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 892 mtx_unlock(&mountlist_mtx); 893 vfs_event_signal(NULL, VQ_MOUNT, 0); 894 if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) 895 panic("mount: lost mount"); 896 VOP_UNLOCK(newdp, 0); 897 VOP_UNLOCK(vp, 0); 898 mountcheckdirs(vp, newdp); 899 vrele(newdp); 900 if ((mp->mnt_flag & MNT_RDONLY) == 0) 901 vfs_allocate_syncvnode(mp); 902 vfs_unbusy(mp); 903 return (0); 904} 905 906/* 907 * vfs_domount_update(): update of mounted file system 908 */ 909static int 910vfs_domount_update( 911 struct thread *td, /* Calling thread. */ 912 struct vnode *vp, /* Mount point vnode. */ 913 int fsflags, /* Flags common to all filesystems. */ 914 void *fsdata /* Options local to the filesystem. */ 915 ) 916{ 917 struct oexport_args oexport; 918 struct export_args export; 919 struct mount *mp; 920 int error, flag; 921 922 mtx_assert(&Giant, MA_OWNED); 923 ASSERT_VOP_ELOCKED(vp, __func__); 924 KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here")); 925 926 if ((vp->v_vflag & VV_ROOT) == 0) { 927 vput(vp); 928 return (EINVAL); 929 } 930 mp = vp->v_mount; 931 /* 932 * We only allow the filesystem to be reloaded if it 933 * is currently mounted read-only. 934 */ 935 flag = mp->mnt_flag; 936 if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) { 937 vput(vp); 938 return (EOPNOTSUPP); /* Needs translation */ 939 } 940 /* 941 * Only privileged root, or (if MNT_USER is set) the user that 942 * did the original mount is permitted to update it. 943 */ 944 error = vfs_suser(mp, td); 945 if (error != 0) { 946 vput(vp); 947 return (error); 948 } 949 if (vfs_busy(mp, MBF_NOWAIT)) { 950 vput(vp); 951 return (EBUSY); 952 } 953 VI_LOCK(vp); 954 if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) { 955 VI_UNLOCK(vp); 956 vfs_unbusy(mp); 957 vput(vp); 958 return (EBUSY); 959 } 960 vp->v_iflag |= VI_MOUNT; 961 VI_UNLOCK(vp); 962 VOP_UNLOCK(vp, 0); 963 964 MNT_ILOCK(mp); 965 mp->mnt_flag &= ~MNT_UPDATEMASK; 966 mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | 967 MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY); 968 if ((mp->mnt_flag & MNT_ASYNC) == 0) 969 mp->mnt_kern_flag &= ~MNTK_ASYNC; 970 MNT_IUNLOCK(mp); 971 mp->mnt_optnew = fsdata; 972 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt); 973 974 /* 975 * Mount the filesystem. 976 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 977 * get. No freeing of cn_pnbuf. 978 */ 979 error = VFS_MOUNT(mp); 980 981 if (error == 0) { 982 /* Process the export option. */ 983 if (vfs_copyopt(mp->mnt_optnew, "export", &export, 984 sizeof(export)) == 0) { 985 error = vfs_export(mp, &export); 986 } else if (vfs_copyopt(mp->mnt_optnew, "export", &oexport, 987 sizeof(oexport)) == 0) { 988 export.ex_flags = oexport.ex_flags; 989 export.ex_root = oexport.ex_root; 990 export.ex_anon = oexport.ex_anon; 991 export.ex_addr = oexport.ex_addr; 992 export.ex_addrlen = oexport.ex_addrlen; 993 export.ex_mask = oexport.ex_mask; 994 export.ex_masklen = oexport.ex_masklen; 995 export.ex_indexfile = oexport.ex_indexfile; 996 export.ex_numsecflavors = 0; 997 error = vfs_export(mp, &export); 998 } 999 } 1000 1001 MNT_ILOCK(mp); 1002 if (error == 0) { 1003 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE | 1004 MNT_SNAPSHOT); 1005 } else { 1006 /* 1007 * If we fail, restore old mount flags. MNT_QUOTA is special, 1008 * because it is not part of MNT_UPDATEMASK, but it could have 1009 * changed in the meantime if quotactl(2) was called. 1010 * All in all we want current value of MNT_QUOTA, not the old 1011 * one. 1012 */ 1013 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA); 1014 } 1015 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1016 mp->mnt_kern_flag |= MNTK_ASYNC; 1017 else 1018 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1019 MNT_IUNLOCK(mp); 1020 1021 if (error != 0) 1022 goto end; 1023 1024 if (mp->mnt_opt != NULL) 1025 vfs_freeopts(mp->mnt_opt); 1026 mp->mnt_opt = mp->mnt_optnew; 1027 (void)VFS_STATFS(mp, &mp->mnt_stat); 1028 /* 1029 * Prevent external consumers of mount options from reading 1030 * mnt_optnew. 1031 */ 1032 mp->mnt_optnew = NULL; 1033 1034 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 1035 if (mp->mnt_syncer == NULL) 1036 vfs_allocate_syncvnode(mp); 1037 } else { 1038 if (mp->mnt_syncer != NULL) 1039 vrele(mp->mnt_syncer); 1040 mp->mnt_syncer = NULL; 1041 } 1042end: 1043 vfs_unbusy(mp); 1044 VI_LOCK(vp); 1045 vp->v_iflag &= ~VI_MOUNT; 1046 VI_UNLOCK(vp); 1047 vrele(vp); 1048 return (error); 1049} 1050 1051/* 1052 * vfs_domount(): actually attempt a filesystem mount. 1053 */ 1054static int 1055vfs_domount( 1056 struct thread *td, /* Calling thread. */ 1057 const char *fstype, /* Filesystem type. */ 1058 char *fspath, /* Mount path. */ 1059 int fsflags, /* Flags common to all filesystems. */ 1060 void *fsdata /* Options local to the filesystem. */ 1061 ) 1062{ 1063 struct vfsconf *vfsp; 1064 struct nameidata nd; 1065 struct vnode *vp; 1066 int error; 1067 1068 /* 1069 * Be ultra-paranoid about making sure the type and fspath 1070 * variables will fit in our mp buffers, including the 1071 * terminating NUL. 1072 */ 1073 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN) 1074 return (ENAMETOOLONG); 1075 1076 if (jailed(td->td_ucred) || usermount == 0) { 1077 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0) 1078 return (error); 1079 } 1080 1081 /* 1082 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users. 1083 */ 1084 if (fsflags & MNT_EXPORTED) { 1085 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED); 1086 if (error) 1087 return (error); 1088 } 1089 if (fsflags & MNT_SUIDDIR) { 1090 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR); 1091 if (error) 1092 return (error); 1093 } 1094 /* 1095 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users. 1096 */ 1097 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) { 1098 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0) 1099 fsflags |= MNT_NOSUID | MNT_USER; 1100 } 1101 1102 /* Load KLDs before we lock the covered vnode to avoid reversals. */ 1103 vfsp = NULL; 1104 if ((fsflags & MNT_UPDATE) == 0) { 1105 /* Don't try to load KLDs if we're mounting the root. */ 1106 if (fsflags & MNT_ROOTFS) 1107 vfsp = vfs_byname(fstype); 1108 else 1109 vfsp = vfs_byname_kld(fstype, td, &error); 1110 if (vfsp == NULL) 1111 return (ENODEV); 1112 if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL)) 1113 return (EPERM); 1114 } 1115 1116 /* 1117 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE. 1118 */ 1119 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | MPSAFE | AUDITVNODE1, 1120 UIO_SYSSPACE, fspath, td); 1121 error = namei(&nd); 1122 if (error != 0) 1123 return (error); 1124 if (!NDHASGIANT(&nd)) 1125 mtx_lock(&Giant); 1126 NDFREE(&nd, NDF_ONLY_PNBUF); 1127 vp = nd.ni_vp; 1128 if ((fsflags & MNT_UPDATE) == 0) { 1129 error = vfs_domount_first(td, vfsp, fspath, vp, fsflags, 1130 fsdata); 1131 } else { 1132 error = vfs_domount_update(td, vp, fsflags, fsdata); 1133 } 1134 mtx_unlock(&Giant); 1135 1136 ASSERT_VI_UNLOCKED(vp, __func__); 1137 ASSERT_VOP_UNLOCKED(vp, __func__); 1138 1139 return (error); 1140} 1141 1142/* 1143 * Unmount a filesystem. 1144 * 1145 * Note: unmount takes a path to the vnode mounted on as argument, not 1146 * special file (as before). 1147 */ 1148#ifndef _SYS_SYSPROTO_H_ 1149struct unmount_args { 1150 char *path; 1151 int flags; 1152}; 1153#endif 1154/* ARGSUSED */ 1155int 1156unmount(td, uap) 1157 struct thread *td; 1158 register struct unmount_args /* { 1159 char *path; 1160 int flags; 1161 } */ *uap; 1162{ 1163 struct mount *mp; 1164 char *pathbuf; 1165 int error, id0, id1; 1166 1167 AUDIT_ARG_VALUE(uap->flags); 1168 if (jailed(td->td_ucred) || usermount == 0) { 1169 error = priv_check(td, PRIV_VFS_UNMOUNT); 1170 if (error) 1171 return (error); 1172 } 1173 1174 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK); 1175 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL); 1176 if (error) { 1177 free(pathbuf, M_TEMP); 1178 return (error); 1179 } 1180 mtx_lock(&Giant); 1181 if (uap->flags & MNT_BYFSID) { 1182 AUDIT_ARG_TEXT(pathbuf); 1183 /* Decode the filesystem ID. */ 1184 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) { 1185 mtx_unlock(&Giant); 1186 free(pathbuf, M_TEMP); 1187 return (EINVAL); 1188 } 1189 1190 mtx_lock(&mountlist_mtx); 1191 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1192 if (mp->mnt_stat.f_fsid.val[0] == id0 && 1193 mp->mnt_stat.f_fsid.val[1] == id1) 1194 break; 1195 } 1196 mtx_unlock(&mountlist_mtx); 1197 } else { 1198 AUDIT_ARG_UPATH1(td, pathbuf); 1199 mtx_lock(&mountlist_mtx); 1200 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) { 1201 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) 1202 break; 1203 } 1204 mtx_unlock(&mountlist_mtx); 1205 } 1206 free(pathbuf, M_TEMP); 1207 if (mp == NULL) { 1208 /* 1209 * Previously we returned ENOENT for a nonexistent path and 1210 * EINVAL for a non-mountpoint. We cannot tell these apart 1211 * now, so in the !MNT_BYFSID case return the more likely 1212 * EINVAL for compatibility. 1213 */ 1214 mtx_unlock(&Giant); 1215 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL); 1216 } 1217 1218 /* 1219 * Don't allow unmounting the root filesystem. 1220 */ 1221 if (mp->mnt_flag & MNT_ROOTFS) { 1222 mtx_unlock(&Giant); 1223 return (EINVAL); 1224 } 1225 error = dounmount(mp, uap->flags, td); 1226 mtx_unlock(&Giant); 1227 return (error); 1228} 1229 1230/* 1231 * Do the actual filesystem unmount. 1232 */ 1233int 1234dounmount(mp, flags, td) 1235 struct mount *mp; 1236 int flags; 1237 struct thread *td; 1238{ 1239 struct vnode *coveredvp, *fsrootvp; 1240 int error; 1241 int async_flag; 1242 int mnt_gen_r; 1243 1244 mtx_assert(&Giant, MA_OWNED); 1245 1246 if ((coveredvp = mp->mnt_vnodecovered) != NULL) { 1247 mnt_gen_r = mp->mnt_gen; 1248 VI_LOCK(coveredvp); 1249 vholdl(coveredvp); 1250 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY); 1251 vdrop(coveredvp); 1252 /* 1253 * Check for mp being unmounted while waiting for the 1254 * covered vnode lock. 1255 */ 1256 if (coveredvp->v_mountedhere != mp || 1257 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) { 1258 VOP_UNLOCK(coveredvp, 0); 1259 return (EBUSY); 1260 } 1261 } 1262 /* 1263 * Only privileged root, or (if MNT_USER is set) the user that did the 1264 * original mount is permitted to unmount this filesystem. 1265 */ 1266 error = vfs_suser(mp, td); 1267 if (error) { 1268 if (coveredvp) 1269 VOP_UNLOCK(coveredvp, 0); 1270 return (error); 1271 } 1272 1273 MNT_ILOCK(mp); 1274 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 1275 MNT_IUNLOCK(mp); 1276 if (coveredvp) 1277 VOP_UNLOCK(coveredvp, 0); 1278 return (EBUSY); 1279 } 1280 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ; 1281 /* Allow filesystems to detect that a forced unmount is in progress. */ 1282 if (flags & MNT_FORCE) 1283 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 1284 error = 0; 1285 if (mp->mnt_lockref) { 1286 if ((flags & MNT_FORCE) == 0) { 1287 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_NOINSMNTQ | 1288 MNTK_UNMOUNTF); 1289 if (mp->mnt_kern_flag & MNTK_MWAIT) { 1290 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1291 wakeup(mp); 1292 } 1293 MNT_IUNLOCK(mp); 1294 if (coveredvp) 1295 VOP_UNLOCK(coveredvp, 0); 1296 return (EBUSY); 1297 } 1298 mp->mnt_kern_flag |= MNTK_DRAINING; 1299 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS, 1300 "mount drain", 0); 1301 } 1302 MNT_IUNLOCK(mp); 1303 KASSERT(mp->mnt_lockref == 0, 1304 ("%s: invalid lock refcount in the drain path @ %s:%d", 1305 __func__, __FILE__, __LINE__)); 1306 KASSERT(error == 0, 1307 ("%s: invalid return value for msleep in the drain path @ %s:%d", 1308 __func__, __FILE__, __LINE__)); 1309 vn_start_write(NULL, &mp, V_WAIT); 1310 1311 if (mp->mnt_flag & MNT_EXPUBLIC) 1312 vfs_setpublicfs(NULL, NULL, NULL); 1313 1314 vfs_msync(mp, MNT_WAIT); 1315 MNT_ILOCK(mp); 1316 async_flag = mp->mnt_flag & MNT_ASYNC; 1317 mp->mnt_flag &= ~MNT_ASYNC; 1318 mp->mnt_kern_flag &= ~MNTK_ASYNC; 1319 MNT_IUNLOCK(mp); 1320 cache_purgevfs(mp); /* remove cache entries for this file sys */ 1321 if (mp->mnt_syncer != NULL) 1322 vrele(mp->mnt_syncer); 1323 /* 1324 * For forced unmounts, move process cdir/rdir refs on the fs root 1325 * vnode to the covered vnode. For non-forced unmounts we want 1326 * such references to cause an EBUSY error. 1327 */ 1328 if ((flags & MNT_FORCE) && 1329 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) { 1330 if (mp->mnt_vnodecovered != NULL) 1331 mountcheckdirs(fsrootvp, mp->mnt_vnodecovered); 1332 if (fsrootvp == rootvnode) { 1333 vrele(rootvnode); 1334 rootvnode = NULL; 1335 } 1336 vput(fsrootvp); 1337 } 1338 if (((mp->mnt_flag & MNT_RDONLY) || 1339 (error = VFS_SYNC(mp, MNT_WAIT)) == 0) || (flags & MNT_FORCE) != 0) 1340 error = VFS_UNMOUNT(mp, flags); 1341 vn_finished_write(mp); 1342 /* 1343 * If we failed to flush the dirty blocks for this mount point, 1344 * undo all the cdir/rdir and rootvnode changes we made above. 1345 * Unless we failed to do so because the device is reporting that 1346 * it doesn't exist anymore. 1347 */ 1348 if (error && error != ENXIO) { 1349 if ((flags & MNT_FORCE) && 1350 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp) == 0) { 1351 if (mp->mnt_vnodecovered != NULL) 1352 mountcheckdirs(mp->mnt_vnodecovered, fsrootvp); 1353 if (rootvnode == NULL) { 1354 rootvnode = fsrootvp; 1355 vref(rootvnode); 1356 } 1357 vput(fsrootvp); 1358 } 1359 MNT_ILOCK(mp); 1360 mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ; 1361 if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL) { 1362 MNT_IUNLOCK(mp); 1363 vfs_allocate_syncvnode(mp); 1364 MNT_ILOCK(mp); 1365 } 1366 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 1367 mp->mnt_flag |= async_flag; 1368 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 1369 mp->mnt_kern_flag |= MNTK_ASYNC; 1370 if (mp->mnt_kern_flag & MNTK_MWAIT) { 1371 mp->mnt_kern_flag &= ~MNTK_MWAIT; 1372 wakeup(mp); 1373 } 1374 MNT_IUNLOCK(mp); 1375 if (coveredvp) 1376 VOP_UNLOCK(coveredvp, 0); 1377 return (error); 1378 } 1379 mtx_lock(&mountlist_mtx); 1380 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1381 mtx_unlock(&mountlist_mtx); 1382 if (coveredvp != NULL) { 1383 coveredvp->v_mountedhere = NULL; 1384 vput(coveredvp); 1385 } 1386 vfs_event_signal(NULL, VQ_UNMOUNT, 0); 1387 vfs_mount_destroy(mp); 1388 return (0); 1389} 1390 1391/* 1392 * --------------------------------------------------------------------- 1393 * Mounting of root filesystem 1394 * 1395 */ 1396 1397struct root_hold_token { 1398 const char *who; 1399 LIST_ENTRY(root_hold_token) list; 1400}; 1401 1402static LIST_HEAD(, root_hold_token) root_holds = 1403 LIST_HEAD_INITIALIZER(root_holds); 1404 1405static int root_mount_complete; 1406 1407/* 1408 * Hold root mount. 1409 */ 1410struct root_hold_token * 1411root_mount_hold(const char *identifier) 1412{ 1413 struct root_hold_token *h; 1414 1415 if (root_mounted()) 1416 return (NULL); 1417 1418 h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK); 1419 h->who = identifier; 1420 mtx_lock(&mountlist_mtx); 1421 LIST_INSERT_HEAD(&root_holds, h, list); 1422 mtx_unlock(&mountlist_mtx); 1423 return (h); 1424} 1425 1426/* 1427 * Release root mount. 1428 */ 1429void 1430root_mount_rel(struct root_hold_token *h) 1431{ 1432 1433 if (h == NULL) 1434 return; 1435 mtx_lock(&mountlist_mtx); 1436 LIST_REMOVE(h, list); 1437 wakeup(&root_holds); 1438 mtx_unlock(&mountlist_mtx); 1439 free(h, M_DEVBUF); 1440} 1441 1442/* 1443 * Wait for all subsystems to release root mount. 1444 */ 1445static void 1446root_mount_prepare(void) 1447{ 1448 struct root_hold_token *h; 1449 struct timeval lastfail; 1450 int curfail = 0; 1451 1452 for (;;) { 1453 DROP_GIANT(); 1454 g_waitidle(); 1455 PICKUP_GIANT(); 1456 mtx_lock(&mountlist_mtx); 1457 if (LIST_EMPTY(&root_holds)) { 1458 mtx_unlock(&mountlist_mtx); 1459 break; 1460 } 1461 if (ppsratecheck(&lastfail, &curfail, 1)) { 1462 printf("Root mount waiting for:"); 1463 LIST_FOREACH(h, &root_holds, list) 1464 printf(" %s", h->who); 1465 printf("\n"); 1466 } 1467 msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold", 1468 hz); 1469 } 1470} 1471 1472/* 1473 * Root was mounted, share the good news. 1474 */ 1475static void 1476root_mount_done(void) 1477{ 1478 1479 /* Keep prison0's root in sync with the global rootvnode. */ 1480 mtx_lock(&prison0.pr_mtx); 1481 prison0.pr_root = rootvnode; 1482 vref(prison0.pr_root); 1483 mtx_unlock(&prison0.pr_mtx); 1484 /* 1485 * Use a mutex to prevent the wakeup being missed and waiting for 1486 * an extra 1 second sleep. 1487 */ 1488 mtx_lock(&mountlist_mtx); 1489 root_mount_complete = 1; 1490 wakeup(&root_mount_complete); 1491 mtx_unlock(&mountlist_mtx); 1492} 1493 1494/* 1495 * Return true if root is already mounted. 1496 */ 1497int 1498root_mounted(void) 1499{ 1500 1501 /* No mutex is acquired here because int stores are atomic. */ 1502 return (root_mount_complete); 1503} 1504 1505/* 1506 * Wait until root is mounted. 1507 */ 1508void 1509root_mount_wait(void) 1510{ 1511 1512 /* 1513 * Panic on an obvious deadlock - the function can't be called from 1514 * a thread which is doing the whole SYSINIT stuff. 1515 */ 1516 KASSERT(curthread->td_proc->p_pid != 0, 1517 ("root_mount_wait: cannot be called from the swapper thread")); 1518 mtx_lock(&mountlist_mtx); 1519 while (!root_mount_complete) { 1520 msleep(&root_mount_complete, &mountlist_mtx, PZERO, "rootwait", 1521 hz); 1522 } 1523 mtx_unlock(&mountlist_mtx); 1524} 1525 1526static void 1527set_rootvnode() 1528{ 1529 struct proc *p; 1530 1531 if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode)) 1532 panic("Cannot find root vnode"); 1533 1534 VOP_UNLOCK(rootvnode, 0); 1535 1536 p = curthread->td_proc; 1537 FILEDESC_XLOCK(p->p_fd); 1538 1539 if (p->p_fd->fd_cdir != NULL) 1540 vrele(p->p_fd->fd_cdir); 1541 p->p_fd->fd_cdir = rootvnode; 1542 VREF(rootvnode); 1543 1544 if (p->p_fd->fd_rdir != NULL) 1545 vrele(p->p_fd->fd_rdir); 1546 p->p_fd->fd_rdir = rootvnode; 1547 VREF(rootvnode); 1548 1549 FILEDESC_XUNLOCK(p->p_fd); 1550 1551 EVENTHANDLER_INVOKE(mountroot); 1552} 1553 1554/* 1555 * Mount /devfs as our root filesystem, but do not put it on the mountlist 1556 * yet. Create a /dev -> / symlink so that absolute pathnames will lookup. 1557 */ 1558 1559static void 1560devfs_first(void) 1561{ 1562 struct thread *td = curthread; 1563 struct vfsoptlist *opts; 1564 struct vfsconf *vfsp; 1565 struct mount *mp = NULL; 1566 int error; 1567 1568 vfsp = vfs_byname("devfs"); 1569 KASSERT(vfsp != NULL, ("Could not find devfs by name")); 1570 if (vfsp == NULL) 1571 return; 1572 1573 mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td->td_ucred); 1574 1575 error = VFS_MOUNT(mp); 1576 KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error)); 1577 if (error) 1578 return; 1579 1580 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK); 1581 TAILQ_INIT(opts); 1582 mp->mnt_opt = opts; 1583 1584 mtx_lock(&mountlist_mtx); 1585 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list); 1586 mtx_unlock(&mountlist_mtx); 1587 1588 set_rootvnode(); 1589 1590 error = kern_symlink(td, "/", "dev", UIO_SYSSPACE); 1591 if (error) 1592 printf("kern_symlink /dev -> / returns %d\n", error); 1593} 1594 1595/* 1596 * Surgically move our devfs to be mounted on /dev. 1597 */ 1598 1599static void 1600devfs_fixup(struct thread *td) 1601{ 1602 struct nameidata nd; 1603 struct vnode *vp, *dvp; 1604 struct mount *mp; 1605 int error; 1606 1607 /* Remove our devfs mount from the mountlist and purge the cache */ 1608 mtx_lock(&mountlist_mtx); 1609 mp = TAILQ_FIRST(&mountlist); 1610 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1611 mtx_unlock(&mountlist_mtx); 1612 cache_purgevfs(mp); 1613 1614 VFS_ROOT(mp, LK_EXCLUSIVE, &dvp); 1615 VI_LOCK(dvp); 1616 dvp->v_iflag &= ~VI_MOUNT; 1617 VI_UNLOCK(dvp); 1618 dvp->v_mountedhere = NULL; 1619 1620 /* Set up the real rootvnode, and purge the cache */ 1621 TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL; 1622 set_rootvnode(); 1623 cache_purgevfs(rootvnode->v_mount); 1624 1625 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td); 1626 error = namei(&nd); 1627 if (error) { 1628 printf("Lookup of /dev for devfs, error: %d\n", error); 1629 vput(dvp); 1630 vfs_unbusy(mp); 1631 return; 1632 } 1633 NDFREE(&nd, NDF_ONLY_PNBUF); 1634 vp = nd.ni_vp; 1635 if (vp->v_type != VDIR) { 1636 printf("/dev is not a directory\n"); 1637 vput(dvp); 1638 vput(vp); 1639 vfs_unbusy(mp); 1640 return; 1641 } 1642 error = vinvalbuf(vp, V_SAVE, 0, 0); 1643 if (error) { 1644 printf("vinvalbuf() of /dev failed, error: %d\n", error); 1645 vput(dvp); 1646 vput(vp); 1647 vfs_unbusy(mp); 1648 return; 1649 } 1650 cache_purge(vp); 1651 mp->mnt_vnodecovered = vp; 1652 vp->v_mountedhere = mp; 1653 mtx_lock(&mountlist_mtx); 1654 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 1655 mtx_unlock(&mountlist_mtx); 1656 VOP_UNLOCK(vp, 0); 1657 vput(dvp); 1658 vfs_unbusy(mp); 1659 1660 /* Unlink the no longer needed /dev/dev -> / symlink */ 1661 error = kern_unlink(td, "/dev/dev", UIO_SYSSPACE); 1662 if (error) 1663 printf("kern_unlink of /dev/dev failed, error: %d\n", error); 1664} 1665 1666/* 1667 * Report errors during filesystem mounting. 1668 */ 1669void 1670vfs_mount_error(struct mount *mp, const char *fmt, ...) 1671{ 1672 struct vfsoptlist *moptlist = mp->mnt_optnew; 1673 va_list ap; 1674 int error, len; 1675 char *errmsg; 1676 1677 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len); 1678 if (error || errmsg == NULL || len <= 0) 1679 return; 1680 1681 va_start(ap, fmt); 1682 vsnprintf(errmsg, (size_t)len, fmt, ap); 1683 va_end(ap); 1684} 1685 1686void 1687vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...) 1688{ 1689 va_list ap; 1690 int error, len; 1691 char *errmsg; 1692 1693 error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len); 1694 if (error || errmsg == NULL || len <= 0) 1695 return; 1696 1697 va_start(ap, fmt); 1698 vsnprintf(errmsg, (size_t)len, fmt, ap); 1699 va_end(ap); 1700} 1701 1702/* 1703 * Find and mount the root filesystem 1704 */ 1705void 1706vfs_mountroot(void) 1707{ 1708 char *cp, *cpt, *options, *tmpdev; 1709 int error, i, asked = 0; 1710 1711 options = NULL; 1712 1713 root_mount_prepare(); 1714 1715 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), 1716 NULL, NULL, mount_init, mount_fini, 1717 UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1718 devfs_first(); 1719 1720 /* 1721 * We are booted with instructions to prompt for the root filesystem. 1722 */ 1723 if (boothowto & RB_ASKNAME) { 1724 if (!vfs_mountroot_ask()) 1725 goto mounted; 1726 asked = 1; 1727 } 1728 1729 options = getenv("vfs.root.mountfrom.options"); 1730 1731 /* 1732 * The root filesystem information is compiled in, and we are 1733 * booted with instructions to use it. 1734 */ 1735 if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) { 1736 if (!vfs_mountroot_try(ctrootdevname, options)) 1737 goto mounted; 1738 ctrootdevname = NULL; 1739 } 1740 1741 /* 1742 * We've been given the generic "use CDROM as root" flag. This is 1743 * necessary because one media may be used in many different 1744 * devices, so we need to search for them. 1745 */ 1746 if (boothowto & RB_CDROM) { 1747 for (i = 0; cdrom_rootdevnames[i] != NULL; i++) { 1748 if (!vfs_mountroot_try(cdrom_rootdevnames[i], options)) 1749 goto mounted; 1750 } 1751 } 1752 1753 /* 1754 * Try to use the value read by the loader from /etc/fstab, or 1755 * supplied via some other means. This is the preferred 1756 * mechanism. 1757 */ 1758 cp = getenv("vfs.root.mountfrom"); 1759 if (cp != NULL) { 1760 cpt = cp; 1761 while ((tmpdev = strsep(&cpt, " \t")) != NULL) { 1762 error = vfs_mountroot_try(tmpdev, options); 1763 if (error == 0) { 1764 freeenv(cp); 1765 goto mounted; 1766 } 1767 } 1768 freeenv(cp); 1769 } 1770 1771 /* 1772 * Try values that may have been computed by code during boot 1773 */ 1774 if (!vfs_mountroot_try(rootdevnames[0], options)) 1775 goto mounted; 1776 if (!vfs_mountroot_try(rootdevnames[1], options)) 1777 goto mounted; 1778 1779 /* 1780 * If we (still) have a compiled-in default, try it. 1781 */ 1782 if (ctrootdevname != NULL) 1783 if (!vfs_mountroot_try(ctrootdevname, options)) 1784 goto mounted; 1785 /* 1786 * Everything so far has failed, prompt on the console if we haven't 1787 * already tried that. 1788 */ 1789 if (!asked) 1790 if (!vfs_mountroot_ask()) 1791 goto mounted; 1792 1793 panic("Root mount failed, startup aborted."); 1794 1795mounted: 1796 root_mount_done(); 1797 freeenv(options); 1798} 1799 1800static struct mntarg * 1801parse_mountroot_options(struct mntarg *ma, const char *options) 1802{ 1803 char *p; 1804 char *name, *name_arg; 1805 char *val, *val_arg; 1806 char *opts; 1807 1808 if (options == NULL || options[0] == '\0') 1809 return (ma); 1810 1811 p = opts = strdup(options, M_MOUNT); 1812 if (opts == NULL) { 1813 return (ma); 1814 } 1815 1816 while((name = strsep(&p, ",")) != NULL) { 1817 if (name[0] == '\0') 1818 break; 1819 1820 val = strchr(name, '='); 1821 if (val != NULL) { 1822 *val = '\0'; 1823 ++val; 1824 } 1825 if( strcmp(name, "rw") == 0 || 1826 strcmp(name, "noro") == 0) { 1827 /* 1828 * The first time we mount the root file system, 1829 * we need to mount 'ro', so We need to ignore 1830 * 'rw' and 'noro' mount options. 1831 */ 1832 continue; 1833 } 1834 name_arg = strdup(name, M_MOUNT); 1835 val_arg = NULL; 1836 if (val != NULL) 1837 val_arg = strdup(val, M_MOUNT); 1838 1839 ma = mount_arg(ma, name_arg, val_arg, 1840 (val_arg != NULL ? -1 : 0)); 1841 } 1842 free(opts, M_MOUNT); 1843 return (ma); 1844} 1845 1846/* 1847 * Mount (mountfrom) as the root filesystem. 1848 */ 1849static int 1850vfs_mountroot_try(const char *mountfrom, const char *options) 1851{ 1852 struct mount *mp; 1853 struct mntarg *ma; 1854 char *vfsname, *path; 1855 time_t timebase; 1856 int error; 1857 char patt[32]; 1858 char errmsg[255]; 1859 1860 vfsname = NULL; 1861 path = NULL; 1862 mp = NULL; 1863 ma = NULL; 1864 error = EINVAL; 1865 bzero(errmsg, sizeof(errmsg)); 1866 1867 if (mountfrom == NULL) 1868 return (error); /* don't complain */ 1869 printf("Trying to mount root from %s\n", mountfrom); 1870 1871 /* parse vfs name and path */ 1872 vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK); 1873 path = malloc(MNAMELEN, M_MOUNT, M_WAITOK); 1874 vfsname[0] = path[0] = 0; 1875 sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN); 1876 if (sscanf(mountfrom, patt, vfsname, path) < 1) 1877 goto out; 1878 1879 if (path[0] == '\0') 1880 strcpy(path, ROOTNAME); 1881 1882 ma = mount_arg(ma, "fstype", vfsname, -1); 1883 ma = mount_arg(ma, "fspath", "/", -1); 1884 ma = mount_arg(ma, "from", path, -1); 1885 ma = mount_arg(ma, "errmsg", errmsg, sizeof(errmsg)); 1886 ma = mount_arg(ma, "ro", NULL, 0); 1887 ma = parse_mountroot_options(ma, options); 1888 error = kernel_mount(ma, MNT_ROOTFS); 1889 1890 if (error == 0) { 1891 /* 1892 * We mount devfs prior to mounting the / FS, so the first 1893 * entry will typically be devfs. 1894 */ 1895 mp = TAILQ_FIRST(&mountlist); 1896 KASSERT(mp != NULL, ("%s: mountlist is empty", __func__)); 1897 1898 /* 1899 * Iterate over all currently mounted file systems and use 1900 * the time stamp found to check and/or initialize the RTC. 1901 * Typically devfs has no time stamp and the only other FS 1902 * is the actual / FS. 1903 * Call inittodr() only once and pass it the largest of the 1904 * timestamps we encounter. 1905 */ 1906 timebase = 0; 1907 do { 1908 if (mp->mnt_time > timebase) 1909 timebase = mp->mnt_time; 1910 mp = TAILQ_NEXT(mp, mnt_list); 1911 } while (mp != NULL); 1912 inittodr(timebase); 1913 1914 devfs_fixup(curthread); 1915 } 1916 1917 if (error != 0 ) { 1918 printf("ROOT MOUNT ERROR: %s\n", errmsg); 1919 printf("If you have invalid mount options, reboot, and "); 1920 printf("first try the following from\n"); 1921 printf("the loader prompt:\n\n"); 1922 printf(" set vfs.root.mountfrom.options=rw\n\n"); 1923 printf("and then remove invalid mount options from "); 1924 printf("/etc/fstab.\n\n"); 1925 } 1926out: 1927 free(path, M_MOUNT); 1928 free(vfsname, M_MOUNT); 1929 return (error); 1930} 1931 1932/* 1933 * --------------------------------------------------------------------- 1934 * Interactive root filesystem selection code. 1935 */ 1936 1937static int 1938vfs_mountroot_ask(void) 1939{ 1940 char name[128]; 1941 char *mountfrom; 1942 char *options; 1943 1944 for(;;) { 1945 printf("Loader variables:\n"); 1946 printf("vfs.root.mountfrom="); 1947 mountfrom = getenv("vfs.root.mountfrom"); 1948 if (mountfrom != NULL) { 1949 printf("%s", mountfrom); 1950 } 1951 printf("\n"); 1952 printf("vfs.root.mountfrom.options="); 1953 options = getenv("vfs.root.mountfrom.options"); 1954 if (options != NULL) { 1955 printf("%s", options); 1956 } 1957 printf("\n"); 1958 freeenv(mountfrom); 1959 freeenv(options); 1960 printf("\nManual root filesystem specification:\n"); 1961 printf(" <fstype>:<device> Mount <device> using filesystem <fstype>\n"); 1962 printf(" eg. zfs:tank\n"); 1963 printf(" eg. ufs:/dev/da0s1a\n"); 1964 printf(" eg. cd9660:/dev/acd0\n"); 1965 printf(" This is equivalent to: "); 1966 printf("mount -t cd9660 /dev/acd0 /\n"); 1967 printf("\n"); 1968 printf(" ? List valid disk boot devices\n"); 1969 printf(" <empty line> Abort manual input\n"); 1970 printf("\nmountroot> "); 1971 gets(name, sizeof(name), 1); 1972 if (name[0] == '\0') 1973 return (1); 1974 if (name[0] == '?') { 1975 printf("\nList of GEOM managed disk devices:\n "); 1976 g_dev_print(); 1977 continue; 1978 } 1979 if (!vfs_mountroot_try(name, NULL)) 1980 return (0); 1981 } 1982} 1983 1984/* 1985 * --------------------------------------------------------------------- 1986 * Functions for querying mount options/arguments from filesystems. 1987 */ 1988 1989/* 1990 * Check that no unknown options are given 1991 */ 1992int 1993vfs_filteropt(struct vfsoptlist *opts, const char **legal) 1994{ 1995 struct vfsopt *opt; 1996 char errmsg[255]; 1997 const char **t, *p, *q; 1998 int ret = 0; 1999 2000 TAILQ_FOREACH(opt, opts, link) { 2001 p = opt->name; 2002 q = NULL; 2003 if (p[0] == 'n' && p[1] == 'o') 2004 q = p + 2; 2005 for(t = global_opts; *t != NULL; t++) { 2006 if (strcmp(*t, p) == 0) 2007 break; 2008 if (q != NULL) { 2009 if (strcmp(*t, q) == 0) 2010 break; 2011 } 2012 } 2013 if (*t != NULL) 2014 continue; 2015 for(t = legal; *t != NULL; t++) { 2016 if (strcmp(*t, p) == 0) 2017 break; 2018 if (q != NULL) { 2019 if (strcmp(*t, q) == 0) 2020 break; 2021 } 2022 } 2023 if (*t != NULL) 2024 continue; 2025 snprintf(errmsg, sizeof(errmsg), 2026 "mount option <%s> is unknown", p); 2027 printf("%s\n", errmsg); 2028 ret = EINVAL; 2029 } 2030 if (ret != 0) { 2031 TAILQ_FOREACH(opt, opts, link) { 2032 if (strcmp(opt->name, "errmsg") == 0) { 2033 strncpy((char *)opt->value, errmsg, opt->len); 2034 } 2035 } 2036 } 2037 return (ret); 2038} 2039 2040/* 2041 * Get a mount option by its name. 2042 * 2043 * Return 0 if the option was found, ENOENT otherwise. 2044 * If len is non-NULL it will be filled with the length 2045 * of the option. If buf is non-NULL, it will be filled 2046 * with the address of the option. 2047 */ 2048int 2049vfs_getopt(opts, name, buf, len) 2050 struct vfsoptlist *opts; 2051 const char *name; 2052 void **buf; 2053 int *len; 2054{ 2055 struct vfsopt *opt; 2056 2057 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 2058 2059 TAILQ_FOREACH(opt, opts, link) { 2060 if (strcmp(name, opt->name) == 0) { 2061 opt->seen = 1; 2062 if (len != NULL) 2063 *len = opt->len; 2064 if (buf != NULL) 2065 *buf = opt->value; 2066 return (0); 2067 } 2068 } 2069 return (ENOENT); 2070} 2071 2072int 2073vfs_getopt_pos(struct vfsoptlist *opts, const char *name) 2074{ 2075 struct vfsopt *opt; 2076 2077 if (opts == NULL) 2078 return (-1); 2079 2080 TAILQ_FOREACH(opt, opts, link) { 2081 if (strcmp(name, opt->name) == 0) { 2082 opt->seen = 1; 2083 return (opt->pos); 2084 } 2085 } 2086 return (-1); 2087} 2088 2089char * 2090vfs_getopts(struct vfsoptlist *opts, const char *name, int *error) 2091{ 2092 struct vfsopt *opt; 2093 2094 *error = 0; 2095 TAILQ_FOREACH(opt, opts, link) { 2096 if (strcmp(name, opt->name) != 0) 2097 continue; 2098 opt->seen = 1; 2099 if (opt->len == 0 || 2100 ((char *)opt->value)[opt->len - 1] != '\0') { 2101 *error = EINVAL; 2102 return (NULL); 2103 } 2104 return (opt->value); 2105 } 2106 *error = ENOENT; 2107 return (NULL); 2108} 2109 2110int 2111vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val) 2112{ 2113 struct vfsopt *opt; 2114 2115 TAILQ_FOREACH(opt, opts, link) { 2116 if (strcmp(name, opt->name) == 0) { 2117 opt->seen = 1; 2118 if (w != NULL) 2119 *w |= val; 2120 return (1); 2121 } 2122 } 2123 if (w != NULL) 2124 *w &= ~val; 2125 return (0); 2126} 2127 2128int 2129vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...) 2130{ 2131 va_list ap; 2132 struct vfsopt *opt; 2133 int ret; 2134 2135 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL")); 2136 2137 TAILQ_FOREACH(opt, opts, link) { 2138 if (strcmp(name, opt->name) != 0) 2139 continue; 2140 opt->seen = 1; 2141 if (opt->len == 0 || opt->value == NULL) 2142 return (0); 2143 if (((char *)opt->value)[opt->len - 1] != '\0') 2144 return (0); 2145 va_start(ap, fmt); 2146 ret = vsscanf(opt->value, fmt, ap); 2147 va_end(ap); 2148 return (ret); 2149 } 2150 return (0); 2151} 2152 2153int 2154vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len) 2155{ 2156 struct vfsopt *opt; 2157 2158 TAILQ_FOREACH(opt, opts, link) { 2159 if (strcmp(name, opt->name) != 0) 2160 continue; 2161 opt->seen = 1; 2162 if (opt->value == NULL) 2163 opt->len = len; 2164 else { 2165 if (opt->len != len) 2166 return (EINVAL); 2167 bcopy(value, opt->value, len); 2168 } 2169 return (0); 2170 } 2171 return (ENOENT); 2172} 2173 2174int 2175vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len) 2176{ 2177 struct vfsopt *opt; 2178 2179 TAILQ_FOREACH(opt, opts, link) { 2180 if (strcmp(name, opt->name) != 0) 2181 continue; 2182 opt->seen = 1; 2183 if (opt->value == NULL) 2184 opt->len = len; 2185 else { 2186 if (opt->len < len) 2187 return (EINVAL); 2188 opt->len = len; 2189 bcopy(value, opt->value, len); 2190 } 2191 return (0); 2192 } 2193 return (ENOENT); 2194} 2195 2196int 2197vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value) 2198{ 2199 struct vfsopt *opt; 2200 2201 TAILQ_FOREACH(opt, opts, link) { 2202 if (strcmp(name, opt->name) != 0) 2203 continue; 2204 opt->seen = 1; 2205 if (opt->value == NULL) 2206 opt->len = strlen(value) + 1; 2207 else if (strlcpy(opt->value, value, opt->len) >= opt->len) 2208 return (EINVAL); 2209 return (0); 2210 } 2211 return (ENOENT); 2212} 2213 2214/* 2215 * Find and copy a mount option. 2216 * 2217 * The size of the buffer has to be specified 2218 * in len, if it is not the same length as the 2219 * mount option, EINVAL is returned. 2220 * Returns ENOENT if the option is not found. 2221 */ 2222int 2223vfs_copyopt(opts, name, dest, len) 2224 struct vfsoptlist *opts; 2225 const char *name; 2226 void *dest; 2227 int len; 2228{ 2229 struct vfsopt *opt; 2230 2231 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL")); 2232 2233 TAILQ_FOREACH(opt, opts, link) { 2234 if (strcmp(name, opt->name) == 0) { 2235 opt->seen = 1; 2236 if (len != opt->len) 2237 return (EINVAL); 2238 bcopy(opt->value, dest, opt->len); 2239 return (0); 2240 } 2241 } 2242 return (ENOENT); 2243} 2244 2245/* 2246 * This is a helper function for filesystems to traverse their 2247 * vnodes. See MNT_VNODE_FOREACH() in sys/mount.h 2248 */ 2249 2250struct vnode * 2251__mnt_vnode_next(struct vnode **mvp, struct mount *mp) 2252{ 2253 struct vnode *vp; 2254 2255 mtx_assert(MNT_MTX(mp), MA_OWNED); 2256 2257 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 2258 if ((*mvp)->v_yield++ == 500) { 2259 MNT_IUNLOCK(mp); 2260 (*mvp)->v_yield = 0; 2261 uio_yield(); 2262 MNT_ILOCK(mp); 2263 } 2264 vp = TAILQ_NEXT(*mvp, v_nmntvnodes); 2265 while (vp != NULL && vp->v_type == VMARKER) 2266 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2267 2268 /* Check if we are done */ 2269 if (vp == NULL) { 2270 __mnt_vnode_markerfree(mvp, mp); 2271 return (NULL); 2272 } 2273 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 2274 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 2275 return (vp); 2276} 2277 2278struct vnode * 2279__mnt_vnode_first(struct vnode **mvp, struct mount *mp) 2280{ 2281 struct vnode *vp; 2282 2283 mtx_assert(MNT_MTX(mp), MA_OWNED); 2284 2285 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2286 while (vp != NULL && vp->v_type == VMARKER) 2287 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2288 2289 /* Check if we are done */ 2290 if (vp == NULL) { 2291 *mvp = NULL; 2292 return (NULL); 2293 } 2294 MNT_REF(mp); 2295 MNT_IUNLOCK(mp); 2296 *mvp = (struct vnode *) malloc(sizeof(struct vnode), 2297 M_VNODE_MARKER, 2298 M_WAITOK | M_ZERO); 2299 MNT_ILOCK(mp); 2300 (*mvp)->v_type = VMARKER; 2301 2302 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2303 while (vp != NULL && vp->v_type == VMARKER) 2304 vp = TAILQ_NEXT(vp, v_nmntvnodes); 2305 2306 /* Check if we are done */ 2307 if (vp == NULL) { 2308 MNT_IUNLOCK(mp); 2309 free(*mvp, M_VNODE_MARKER); 2310 MNT_ILOCK(mp); 2311 *mvp = NULL; 2312 MNT_REL(mp); 2313 return (NULL); 2314 } 2315 (*mvp)->v_mount = mp; 2316 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 2317 return (vp); 2318} 2319 2320 2321void 2322__mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp) 2323{ 2324 2325 if (*mvp == NULL) 2326 return; 2327 2328 mtx_assert(MNT_MTX(mp), MA_OWNED); 2329 2330 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 2331 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 2332 MNT_IUNLOCK(mp); 2333 free(*mvp, M_VNODE_MARKER); 2334 MNT_ILOCK(mp); 2335 *mvp = NULL; 2336 MNT_REL(mp); 2337} 2338 2339 2340int 2341__vfs_statfs(struct mount *mp, struct statfs *sbp) 2342{ 2343 int error; 2344 2345 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat); 2346 if (sbp != &mp->mnt_stat) 2347 *sbp = mp->mnt_stat; 2348 return (error); 2349} 2350 2351void 2352vfs_mountedfrom(struct mount *mp, const char *from) 2353{ 2354 2355 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname); 2356 strlcpy(mp->mnt_stat.f_mntfromname, from, 2357 sizeof mp->mnt_stat.f_mntfromname); 2358} 2359 2360/* 2361 * --------------------------------------------------------------------- 2362 * This is the api for building mount args and mounting filesystems from 2363 * inside the kernel. 2364 * 2365 * The API works by accumulation of individual args. First error is 2366 * latched. 2367 * 2368 * XXX: should be documented in new manpage kernel_mount(9) 2369 */ 2370 2371/* A memory allocation which must be freed when we are done */ 2372struct mntaarg { 2373 SLIST_ENTRY(mntaarg) next; 2374}; 2375 2376/* The header for the mount arguments */ 2377struct mntarg { 2378 struct iovec *v; 2379 int len; 2380 int error; 2381 SLIST_HEAD(, mntaarg) list; 2382}; 2383 2384/* 2385 * Add a boolean argument. 2386 * 2387 * flag is the boolean value. 2388 * name must start with "no". 2389 */ 2390struct mntarg * 2391mount_argb(struct mntarg *ma, int flag, const char *name) 2392{ 2393 2394 KASSERT(name[0] == 'n' && name[1] == 'o', 2395 ("mount_argb(...,%s): name must start with 'no'", name)); 2396 2397 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0)); 2398} 2399 2400/* 2401 * Add an argument printf style 2402 */ 2403struct mntarg * 2404mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...) 2405{ 2406 va_list ap; 2407 struct mntaarg *maa; 2408 struct sbuf *sb; 2409 int len; 2410 2411 if (ma == NULL) { 2412 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2413 SLIST_INIT(&ma->list); 2414 } 2415 if (ma->error) 2416 return (ma); 2417 2418 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2419 M_MOUNT, M_WAITOK); 2420 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2421 ma->v[ma->len].iov_len = strlen(name) + 1; 2422 ma->len++; 2423 2424 sb = sbuf_new_auto(); 2425 va_start(ap, fmt); 2426 sbuf_vprintf(sb, fmt, ap); 2427 va_end(ap); 2428 sbuf_finish(sb); 2429 len = sbuf_len(sb) + 1; 2430 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2431 SLIST_INSERT_HEAD(&ma->list, maa, next); 2432 bcopy(sbuf_data(sb), maa + 1, len); 2433 sbuf_delete(sb); 2434 2435 ma->v[ma->len].iov_base = maa + 1; 2436 ma->v[ma->len].iov_len = len; 2437 ma->len++; 2438 2439 return (ma); 2440} 2441 2442/* 2443 * Add an argument which is a userland string. 2444 */ 2445struct mntarg * 2446mount_argsu(struct mntarg *ma, const char *name, const void *val, int len) 2447{ 2448 struct mntaarg *maa; 2449 char *tbuf; 2450 2451 if (val == NULL) 2452 return (ma); 2453 if (ma == NULL) { 2454 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2455 SLIST_INIT(&ma->list); 2456 } 2457 if (ma->error) 2458 return (ma); 2459 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO); 2460 SLIST_INSERT_HEAD(&ma->list, maa, next); 2461 tbuf = (void *)(maa + 1); 2462 ma->error = copyinstr(val, tbuf, len, NULL); 2463 return (mount_arg(ma, name, tbuf, -1)); 2464} 2465 2466/* 2467 * Plain argument. 2468 * 2469 * If length is -1, treat value as a C string. 2470 */ 2471struct mntarg * 2472mount_arg(struct mntarg *ma, const char *name, const void *val, int len) 2473{ 2474 2475 if (ma == NULL) { 2476 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO); 2477 SLIST_INIT(&ma->list); 2478 } 2479 if (ma->error) 2480 return (ma); 2481 2482 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2), 2483 M_MOUNT, M_WAITOK); 2484 ma->v[ma->len].iov_base = (void *)(uintptr_t)name; 2485 ma->v[ma->len].iov_len = strlen(name) + 1; 2486 ma->len++; 2487 2488 ma->v[ma->len].iov_base = (void *)(uintptr_t)val; 2489 if (len < 0) 2490 ma->v[ma->len].iov_len = strlen(val) + 1; 2491 else 2492 ma->v[ma->len].iov_len = len; 2493 ma->len++; 2494 return (ma); 2495} 2496 2497/* 2498 * Free a mntarg structure 2499 */ 2500static void 2501free_mntarg(struct mntarg *ma) 2502{ 2503 struct mntaarg *maa; 2504 2505 while (!SLIST_EMPTY(&ma->list)) { 2506 maa = SLIST_FIRST(&ma->list); 2507 SLIST_REMOVE_HEAD(&ma->list, next); 2508 free(maa, M_MOUNT); 2509 } 2510 free(ma->v, M_MOUNT); 2511 free(ma, M_MOUNT); 2512} 2513 2514/* 2515 * Mount a filesystem 2516 */ 2517int 2518kernel_mount(struct mntarg *ma, int flags) 2519{ 2520 struct uio auio; 2521 int error; 2522 2523 KASSERT(ma != NULL, ("kernel_mount NULL ma")); 2524 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v")); 2525 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len)); 2526 2527 auio.uio_iov = ma->v; 2528 auio.uio_iovcnt = ma->len; 2529 auio.uio_segflg = UIO_SYSSPACE; 2530 2531 error = ma->error; 2532 if (!error) 2533 error = vfs_donmount(curthread, flags, &auio); 2534 free_mntarg(ma); 2535 return (error); 2536} 2537 2538/* 2539 * A printflike function to mount a filesystem. 2540 */ 2541int 2542kernel_vmount(int flags, ...) 2543{ 2544 struct mntarg *ma = NULL; 2545 va_list ap; 2546 const char *cp; 2547 const void *vp; 2548 int error; 2549 2550 va_start(ap, flags); 2551 for (;;) { 2552 cp = va_arg(ap, const char *); 2553 if (cp == NULL) 2554 break; 2555 vp = va_arg(ap, const void *); 2556 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0)); 2557 } 2558 va_end(ap); 2559 2560 error = kernel_mount(ma, flags); 2561 return (error); 2562} 2563