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.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
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.
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
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
40 #include <sys/param.h>
42 #include <sys/fcntl.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>
52 #include <sys/filedesc.h>
53 #include <sys/reboot.h>
54 #include <sys/syscallsubr.h>
55 #include <sys/sysproto.h>
57 #include <sys/sysctl.h>
58 #include <sys/sysent.h>
59 #include <sys/systm.h>
60 #include <sys/vnode.h>
63 #include <geom/geom.h>
65 #include <machine/stdarg.h>
67 #include <security/audit/audit.h>
68 #include <security/mac/mac_framework.h>
70 #include "opt_rootdevname.h"
73 #define ROOTNAME "root_device"
74 #define VFS_MOUNTARG_SIZE_MAX (1024 * 64)
76 static int vfs_domount(struct thread *td, const char *fstype,
77 char *fspath, int fsflags, void *fsdata);
78 static int vfs_mountroot_ask(void);
79 static int vfs_mountroot_try(const char *mountfrom);
80 static int vfs_donmount(struct thread *td, int fsflags,
81 struct uio *fsoptions);
82 static void free_mntarg(struct mntarg *ma);
83 static int vfs_getopt_pos(struct vfsoptlist *opts, const char *name);
85 static int usermount = 0;
86 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
87 "Unprivileged users may mount and unmount file systems");
89 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
90 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
91 static uma_zone_t mount_zone;
93 /* List of mounted filesystems. */
94 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
96 /* For any iteration/modification of mountlist */
97 struct mtx mountlist_mtx;
98 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF);
100 TAILQ_HEAD(vfsoptlist, vfsopt);
102 TAILQ_ENTRY(vfsopt) link;
109 * The vnode of the system's root (/ in the filesystem, without chroot
112 struct vnode *rootvnode;
115 * The root filesystem is detailed in the kernel environment variable
116 * vfs.root.mountfrom, which is expected to be in the general format
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
126 * Global opts, taken by all filesystems
128 static const char *global_opts[] = {
141 * The root specifiers we will try if RB_CDROM is specified.
143 static char *cdrom_rootdevnames[] = {
149 /* legacy find-root code */
150 char *rootdevnames[2] = {NULL, NULL};
152 # define ROOTDEVNAME NULL
154 static const char *ctrootdevname = ROOTDEVNAME;
157 * ---------------------------------------------------------------------
158 * Functions for building and sanitizing the mount options
161 /* Remove one mount option. */
163 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
166 TAILQ_REMOVE(opts, opt, link);
167 free(opt->name, M_MOUNT);
168 if (opt->value != NULL)
169 free(opt->value, M_MOUNT);
171 else if (opt->len != 0)
172 panic("%s: mount option with NULL value but length != 0",
178 /* Release all resources related to the mount options. */
180 vfs_freeopts(struct vfsoptlist *opts)
184 while (!TAILQ_EMPTY(opts)) {
185 opt = TAILQ_FIRST(opts);
186 vfs_freeopt(opts, opt);
192 vfs_deleteopt(struct vfsoptlist *opts, const char *name)
194 struct vfsopt *opt, *temp;
196 TAILQ_FOREACH_SAFE(opt, opts, link, temp) {
197 if (strcmp(opt->name, name) == 0)
198 vfs_freeopt(opts, opt);
203 * Check if options are equal (with or without the "no" prefix).
206 vfs_equalopts(const char *opt1, const char *opt2)
209 /* "opt" vs. "opt" or "noopt" vs. "noopt" */
210 if (strcmp(opt1, opt2) == 0)
212 /* "noopt" vs. "opt" */
213 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
215 /* "opt" vs. "noopt" */
216 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
222 * If a mount option is specified several times,
223 * (with or without the "no" prefix) only keep
224 * the last occurence of it.
227 vfs_sanitizeopts(struct vfsoptlist *opts)
229 struct vfsopt *opt, *opt2, *tmp;
231 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
232 opt2 = TAILQ_PREV(opt, vfsoptlist, link);
233 while (opt2 != NULL) {
234 if (vfs_equalopts(opt->name, opt2->name)) {
235 tmp = TAILQ_PREV(opt2, vfsoptlist, link);
236 vfs_freeopt(opts, opt2);
239 opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
246 * Build a linked list of mount options from a struct uio.
249 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
251 struct vfsoptlist *opts;
254 unsigned int i, iovcnt;
255 int error, namelen, optlen;
257 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
260 iovcnt = auio->uio_iovcnt;
261 for (i = 0; i < iovcnt; i += 2) {
262 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
263 namelen = auio->uio_iov[i].iov_len;
264 optlen = auio->uio_iov[i + 1].iov_len;
265 opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
270 * Do this early, so jumps to "bad" will free the current
273 TAILQ_INSERT_TAIL(opts, opt, link);
274 memused += sizeof(struct vfsopt) + optlen + namelen;
277 * Avoid consuming too much memory, and attempts to overflow
280 if (memused > VFS_MOUNTARG_SIZE_MAX ||
281 optlen > VFS_MOUNTARG_SIZE_MAX ||
282 namelen > VFS_MOUNTARG_SIZE_MAX) {
287 if (auio->uio_segflg == UIO_SYSSPACE) {
288 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
290 error = copyin(auio->uio_iov[i].iov_base, opt->name,
295 /* Ensure names are null-terminated strings. */
296 if (opt->name[namelen - 1] != '\0') {
302 opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
303 if (auio->uio_segflg == UIO_SYSSPACE) {
304 bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
307 error = copyin(auio->uio_iov[i + 1].iov_base,
314 vfs_sanitizeopts(opts);
323 * Merge the old mount options with the new ones passed
324 * in the MNT_UPDATE case.
326 * XXX This function will keep a "nofoo" option in the
327 * new options if there is no matching "foo" option
328 * to be cancelled in the old options. This is a bug
329 * if the option's canonical name is "foo". E.g., "noro"
330 * shouldn't end up in the mount point's active options,
334 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts)
336 struct vfsopt *opt, *opt2, *new;
338 TAILQ_FOREACH(opt, opts, link) {
340 * Check that this option hasn't been redefined
341 * nor cancelled with a "no" mount option.
343 opt2 = TAILQ_FIRST(toopts);
344 while (opt2 != NULL) {
345 if (strcmp(opt2->name, opt->name) == 0)
347 if (strncmp(opt2->name, "no", 2) == 0 &&
348 strcmp(opt2->name + 2, opt->name) == 0) {
349 vfs_freeopt(toopts, opt2);
352 opt2 = TAILQ_NEXT(opt2, link);
354 /* We want this option, duplicate it. */
355 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
356 new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK);
357 strcpy(new->name, opt->name);
359 new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
360 bcopy(opt->value, new->value, opt->len);
365 TAILQ_INSERT_TAIL(toopts, new, link);
372 * Mount a filesystem.
377 struct nmount_args /* {
389 AUDIT_ARG(fflags, uap->flags);
392 * Filter out MNT_ROOTFS. We do not want clients of nmount() in
393 * userspace to set this flag, but we must filter it out if we want
394 * MNT_UPDATE on the root file system to work.
395 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try().
397 uap->flags &= ~MNT_ROOTFS;
399 iovcnt = uap->iovcnt;
401 * Check that we have an even number of iovec's
402 * and that we have at least two options.
404 if ((iovcnt & 1) || (iovcnt < 4))
407 error = copyinuio(uap->iovp, iovcnt, &auio);
411 for (i = 0; i < iovcnt; i++) {
412 if (iov->iov_len > MMAXOPTIONLEN) {
418 error = vfs_donmount(td, uap->flags, auio);
425 * ---------------------------------------------------------------------
426 * Various utility functions
430 vfs_ref(struct mount *mp)
439 vfs_rel(struct mount *mp)
448 mount_init(void *mem, int size, int flags)
452 mp = (struct mount *)mem;
453 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
454 lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0);
455 lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0);
460 mount_fini(void *mem, int size)
464 mp = (struct mount *)mem;
465 lockdestroy(&mp->mnt_explock);
466 lockdestroy(&mp->mnt_lock);
467 mtx_destroy(&mp->mnt_mtx);
471 * Allocate and initialize the mount point struct.
474 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp,
475 const char *fspath, struct thread *td)
479 mp = uma_zalloc(mount_zone, M_WAITOK);
480 bzero(&mp->mnt_startzero,
481 __rangeof(struct mount, mnt_startzero, mnt_endzero));
482 TAILQ_INIT(&mp->mnt_nvnodelist);
483 mp->mnt_nvnodelistsize = 0;
485 (void) vfs_busy(mp, LK_NOWAIT, 0, td);
486 mp->mnt_op = vfsp->vfc_vfsops;
488 vfsp->vfc_refcount++; /* XXX Unlocked */
489 mp->mnt_stat.f_type = vfsp->vfc_typenum;
491 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
492 mp->mnt_vnodecovered = vp;
493 mp->mnt_cred = crdup(td->td_ucred);
494 mp->mnt_stat.f_owner = td->td_ucred->cr_uid;
495 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
496 mp->mnt_iosize_max = DFLTPHYS;
499 mac_mount_create(td->td_ucred, mp);
501 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
506 * Destroy the mount struct previously allocated by vfs_mount_alloc().
509 vfs_mount_destroy(struct mount *mp)
514 for (i = 0; mp->mnt_ref && i < 3; i++)
515 msleep(mp, MNT_MTX(mp), PVFS, "mntref", hz);
517 * This will always cause a 3 second delay in rebooting due to
518 * refs on the root mountpoint that never go away. Most of these
519 * are held by init which never exits.
521 if (i == 3 && (!rebooting || bootverbose))
522 printf("Mount point %s had %d dangling refs\n",
523 mp->mnt_stat.f_mntonname, mp->mnt_ref);
524 if (mp->mnt_holdcnt != 0) {
525 printf("Waiting for mount point to be unheld\n");
526 while (mp->mnt_holdcnt != 0) {
527 mp->mnt_holdcntwaiters++;
528 msleep(&mp->mnt_holdcnt, MNT_MTX(mp),
529 PZERO, "mntdestroy", 0);
530 mp->mnt_holdcntwaiters--;
532 printf("mount point unheld\n");
534 if (mp->mnt_writeopcount > 0) {
535 printf("Waiting for mount point write ops\n");
536 while (mp->mnt_writeopcount > 0) {
537 mp->mnt_kern_flag |= MNTK_SUSPEND;
538 msleep(&mp->mnt_writeopcount,
540 PZERO, "mntdestroy2", 0);
542 printf("mount point write ops completed\n");
544 if (mp->mnt_secondary_writes > 0) {
545 printf("Waiting for mount point secondary write ops\n");
546 while (mp->mnt_secondary_writes > 0) {
547 mp->mnt_kern_flag |= MNTK_SUSPEND;
548 msleep(&mp->mnt_secondary_writes,
550 PZERO, "mntdestroy3", 0);
552 printf("mount point secondary write ops completed\n");
555 mp->mnt_vfc->vfc_refcount--;
556 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
559 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
561 panic("unmount: dangling vnode");
564 if (mp->mnt_kern_flag & MNTK_MWAIT)
566 if (mp->mnt_writeopcount != 0)
567 panic("vfs_mount_destroy: nonzero writeopcount");
568 if (mp->mnt_secondary_writes != 0)
569 panic("vfs_mount_destroy: nonzero secondary_writes");
570 if (mp->mnt_nvnodelistsize != 0)
571 panic("vfs_mount_destroy: nonzero nvnodelistsize");
572 mp->mnt_writeopcount = -1000;
573 mp->mnt_nvnodelistsize = -1000;
574 mp->mnt_secondary_writes = -1000;
577 mac_mount_destroy(mp);
579 if (mp->mnt_opt != NULL)
580 vfs_freeopts(mp->mnt_opt);
581 crfree(mp->mnt_cred);
582 uma_zfree(mount_zone, mp);
586 vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions)
588 struct vfsoptlist *optlist;
589 struct vfsopt *opt, *noro_opt;
590 char *fstype, *fspath, *errmsg;
591 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
592 int has_rw, has_noro;
600 error = vfs_buildopts(fsoptions, &optlist);
604 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
605 errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
608 * We need these two options before the others,
609 * and they are mandatory for any filesystem.
610 * Ensure they are NUL terminated as well.
613 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
614 if (error || fstype[fstypelen - 1] != '\0') {
617 strncpy(errmsg, "Invalid fstype", errmsg_len);
621 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
622 if (error || fspath[fspathlen - 1] != '\0') {
625 strncpy(errmsg, "Invalid fspath", errmsg_len);
630 * We need to see if we have the "update" option
631 * before we call vfs_domount(), since vfs_domount() has special
632 * logic based on MNT_UPDATE. This is very important
633 * when we want to update the root filesystem.
635 TAILQ_FOREACH(opt, optlist, link) {
636 if (strcmp(opt->name, "update") == 0)
637 fsflags |= MNT_UPDATE;
638 else if (strcmp(opt->name, "async") == 0)
639 fsflags |= MNT_ASYNC;
640 else if (strcmp(opt->name, "force") == 0)
641 fsflags |= MNT_FORCE;
642 else if (strcmp(opt->name, "multilabel") == 0)
643 fsflags |= MNT_MULTILABEL;
644 else if (strcmp(opt->name, "noasync") == 0)
645 fsflags &= ~MNT_ASYNC;
646 else if (strcmp(opt->name, "noatime") == 0)
647 fsflags |= MNT_NOATIME;
648 else if (strcmp(opt->name, "atime") == 0) {
649 free(opt->name, M_MOUNT);
650 opt->name = strdup("nonoatime", M_MOUNT);
652 else if (strcmp(opt->name, "noclusterr") == 0)
653 fsflags |= MNT_NOCLUSTERR;
654 else if (strcmp(opt->name, "clusterr") == 0) {
655 free(opt->name, M_MOUNT);
656 opt->name = strdup("nonoclusterr", M_MOUNT);
658 else if (strcmp(opt->name, "noclusterw") == 0)
659 fsflags |= MNT_NOCLUSTERW;
660 else if (strcmp(opt->name, "clusterw") == 0) {
661 free(opt->name, M_MOUNT);
662 opt->name = strdup("nonoclusterw", M_MOUNT);
664 else if (strcmp(opt->name, "noexec") == 0)
665 fsflags |= MNT_NOEXEC;
666 else if (strcmp(opt->name, "exec") == 0) {
667 free(opt->name, M_MOUNT);
668 opt->name = strdup("nonoexec", M_MOUNT);
670 else if (strcmp(opt->name, "nosuid") == 0)
671 fsflags |= MNT_NOSUID;
672 else if (strcmp(opt->name, "suid") == 0) {
673 free(opt->name, M_MOUNT);
674 opt->name = strdup("nonosuid", M_MOUNT);
676 else if (strcmp(opt->name, "nosymfollow") == 0)
677 fsflags |= MNT_NOSYMFOLLOW;
678 else if (strcmp(opt->name, "symfollow") == 0) {
679 free(opt->name, M_MOUNT);
680 opt->name = strdup("nonosymfollow", M_MOUNT);
682 else if (strcmp(opt->name, "noro") == 0) {
683 fsflags &= ~MNT_RDONLY;
686 else if (strcmp(opt->name, "rw") == 0) {
687 fsflags &= ~MNT_RDONLY;
690 else if (strcmp(opt->name, "ro") == 0)
691 fsflags |= MNT_RDONLY;
692 else if (strcmp(opt->name, "rdonly") == 0) {
693 free(opt->name, M_MOUNT);
694 opt->name = strdup("ro", M_MOUNT);
695 fsflags |= MNT_RDONLY;
697 else if (strcmp(opt->name, "suiddir") == 0)
698 fsflags |= MNT_SUIDDIR;
699 else if (strcmp(opt->name, "sync") == 0)
700 fsflags |= MNT_SYNCHRONOUS;
701 else if (strcmp(opt->name, "union") == 0)
702 fsflags |= MNT_UNION;
706 * If "rw" was specified as a mount option, and we
707 * are trying to update a mount-point from "ro" to "rw",
708 * we need a mount option "noro", since in vfs_mergeopts(),
709 * "noro" will cancel "ro", but "rw" will not do anything.
711 if (has_rw && !has_noro) {
712 noro_opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
713 noro_opt->name = strdup("noro", M_MOUNT);
714 noro_opt->value = NULL;
716 TAILQ_INSERT_TAIL(optlist, noro_opt, link);
720 * Be ultra-paranoid about making sure the type and fspath
721 * variables will fit in our mp buffers, including the
724 if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) {
725 error = ENAMETOOLONG;
730 error = vfs_domount(td, fstype, fspath, fsflags, optlist);
733 /* copyout the errmsg */
734 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
735 && errmsg_len > 0 && errmsg != NULL) {
736 if (fsoptions->uio_segflg == UIO_SYSSPACE) {
738 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
739 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
742 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
743 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
748 vfs_freeopts(optlist);
755 #ifndef _SYS_SYSPROTO_H_
767 struct mount_args /* {
775 struct vfsconf *vfsp = NULL;
776 struct mntarg *ma = NULL;
779 AUDIT_ARG(fflags, uap->flags);
782 * Filter out MNT_ROOTFS. We do not want clients of mount() in
783 * userspace to set this flag, but we must filter it out if we want
784 * MNT_UPDATE on the root file system to work.
785 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try().
787 uap->flags &= ~MNT_ROOTFS;
789 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
790 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
792 free(fstype, M_TEMP);
796 AUDIT_ARG(text, fstype);
798 vfsp = vfs_byname_kld(fstype, td, &error);
799 free(fstype, M_TEMP);
804 if (vfsp->vfc_vfsops->vfs_cmount == NULL) {
809 ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN);
810 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
811 ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro");
812 ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid");
813 ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec");
815 error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags, td);
822 * vfs_domount(): actually attempt a filesystem mount.
826 struct thread *td, /* Calling thread. */
827 const char *fstype, /* Filesystem type. */
828 char *fspath, /* Mount path. */
829 int fsflags, /* Flags common to all filesystems. */
830 void *fsdata /* Options local to the filesystem. */
835 struct vfsconf *vfsp;
836 struct export_args export;
841 mtx_assert(&Giant, MA_OWNED);
843 * Be ultra-paranoid about making sure the type and fspath
844 * variables will fit in our mp buffers, including the
847 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
848 return (ENAMETOOLONG);
850 if (jailed(td->td_ucred) || usermount == 0) {
851 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
856 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
858 if (fsflags & MNT_EXPORTED) {
859 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
863 if (fsflags & MNT_SUIDDIR) {
864 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
869 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
871 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
872 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
873 fsflags |= MNT_NOSUID | MNT_USER;
876 /* Load KLDs before we lock the covered vnode to avoid reversals. */
878 if ((fsflags & MNT_UPDATE) == 0) {
879 /* Don't try to load KLDs if we're mounting the root. */
880 if (fsflags & MNT_ROOTFS)
881 vfsp = vfs_byname(fstype);
883 vfsp = vfs_byname_kld(fstype, td, &error);
886 if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL))
890 * Get vnode to be covered
892 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE,
894 if ((error = namei(&nd)) != 0)
896 NDFREE(&nd, NDF_ONLY_PNBUF);
898 if (fsflags & MNT_UPDATE) {
899 if ((vp->v_vflag & VV_ROOT) == 0) {
907 * We only allow the filesystem to be reloaded if it
908 * is currently mounted read-only.
910 if ((fsflags & MNT_RELOAD) &&
911 ((mp->mnt_flag & MNT_RDONLY) == 0)) {
914 return (EOPNOTSUPP); /* Needs translation */
918 * Only privileged root, or (if MNT_USER is set) the user that
919 * did the original mount is permitted to update it.
921 error = vfs_suser(mp, td);
926 if (vfs_busy(mp, LK_NOWAIT, 0, td)) {
931 if ((vp->v_iflag & VI_MOUNT) != 0 ||
932 vp->v_mountedhere != NULL) {
938 vp->v_iflag |= VI_MOUNT;
941 mp->mnt_flag |= fsflags &
942 (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS);
945 mp->mnt_optnew = fsdata;
946 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
949 * If the user is not root, ensure that they own the directory
950 * onto which we are attempting to mount.
952 error = VOP_GETATTR(vp, &va, td->td_ucred, td);
957 if (va.va_uid != td->td_ucred->cr_uid) {
958 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN,
965 error = vinvalbuf(vp, V_SAVE, td, 0, 0);
970 if (vp->v_type != VDIR) {
975 if ((vp->v_iflag & VI_MOUNT) != 0 ||
976 vp->v_mountedhere != NULL) {
981 vp->v_iflag |= VI_MOUNT;
985 * Allocate and initialize the filesystem.
987 mp = vfs_mount_alloc(vp, vfsp, fspath, td);
990 /* XXXMAC: pass to vfs_mount_alloc? */
991 mp->mnt_optnew = fsdata;
995 * Set the mount level flags.
998 mp->mnt_flag = (mp->mnt_flag & ~MNT_UPDATEMASK) |
999 (fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS |
1001 if ((mp->mnt_flag & MNT_ASYNC) == 0)
1002 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1005 * Mount the filesystem.
1006 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1007 * get. No freeing of cn_pnbuf.
1009 error = VFS_MOUNT(mp, td);
1012 * Process the export option only if we are
1013 * updating mount options.
1015 if (!error && (fsflags & MNT_UPDATE)) {
1016 if (vfs_copyopt(mp->mnt_optnew, "export", &export,
1017 sizeof(export)) == 0)
1018 error = vfs_export(mp, &export);
1022 if (mp->mnt_opt != NULL)
1023 vfs_freeopts(mp->mnt_opt);
1024 mp->mnt_opt = mp->mnt_optnew;
1025 (void)VFS_STATFS(mp, &mp->mnt_stat, td);
1028 * Prevent external consumers of mount options from reading
1031 mp->mnt_optnew = NULL;
1032 if (mp->mnt_flag & MNT_UPDATE) {
1035 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) |
1036 (flag & ~MNT_QUOTA);
1038 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD |
1039 MNT_FORCE | MNT_SNAPSHOT);
1040 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
1041 mp->mnt_kern_flag |= MNTK_ASYNC;
1043 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1045 if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1046 if (mp->mnt_syncer == NULL)
1047 error = vfs_allocate_syncvnode(mp);
1049 if (mp->mnt_syncer != NULL)
1050 vrele(mp->mnt_syncer);
1051 mp->mnt_syncer = NULL;
1055 vp->v_iflag &= ~VI_MOUNT;
1061 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
1062 mp->mnt_kern_flag |= MNTK_ASYNC;
1064 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1066 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1068 * Put the new filesystem on the mount list after root.
1072 struct vnode *newdp;
1075 vp->v_iflag &= ~VI_MOUNT;
1077 vp->v_mountedhere = mp;
1078 mtx_lock(&mountlist_mtx);
1079 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1080 mtx_unlock(&mountlist_mtx);
1081 vfs_event_signal(NULL, VQ_MOUNT, 0);
1082 if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp, td))
1083 panic("mount: lost mount");
1084 mountcheckdirs(vp, newdp);
1087 if ((mp->mnt_flag & MNT_RDONLY) == 0)
1088 error = vfs_allocate_syncvnode(mp);
1094 vp->v_iflag &= ~VI_MOUNT;
1097 vfs_mount_destroy(mp);
1104 * Unmount a filesystem.
1106 * Note: unmount takes a path to the vnode mounted on as argument, not
1107 * special file (as before).
1109 #ifndef _SYS_SYSPROTO_H_
1110 struct unmount_args {
1119 register struct unmount_args /* {
1126 int error, id0, id1;
1128 if (jailed(td->td_ucred) || usermount == 0) {
1129 error = priv_check(td, PRIV_VFS_UNMOUNT);
1134 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1135 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL);
1137 free(pathbuf, M_TEMP);
1140 AUDIT_ARG(upath, td, pathbuf, ARG_UPATH1);
1142 if (uap->flags & MNT_BYFSID) {
1143 /* Decode the filesystem ID. */
1144 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1146 free(pathbuf, M_TEMP);
1150 mtx_lock(&mountlist_mtx);
1151 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1152 if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1153 mp->mnt_stat.f_fsid.val[1] == id1)
1156 mtx_unlock(&mountlist_mtx);
1158 mtx_lock(&mountlist_mtx);
1159 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1160 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0)
1163 mtx_unlock(&mountlist_mtx);
1165 free(pathbuf, M_TEMP);
1168 * Previously we returned ENOENT for a nonexistent path and
1169 * EINVAL for a non-mountpoint. We cannot tell these apart
1170 * now, so in the !MNT_BYFSID case return the more likely
1171 * EINVAL for compatibility.
1174 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL);
1178 * Don't allow unmounting the root filesystem.
1180 if (mp->mnt_flag & MNT_ROOTFS) {
1184 error = dounmount(mp, uap->flags, td);
1190 * Do the actual filesystem unmount.
1193 dounmount(mp, flags, td)
1198 struct vnode *coveredvp, *fsrootvp;
1203 mtx_assert(&Giant, MA_OWNED);
1205 if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
1206 mnt_gen_r = mp->mnt_gen;
1209 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
1212 * Check for mp being unmounted while waiting for the
1213 * covered vnode lock.
1215 if (coveredvp->v_mountedhere != mp ||
1216 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
1217 VOP_UNLOCK(coveredvp, 0);
1222 * Only privileged root, or (if MNT_USER is set) the user that did the
1223 * original mount is permitted to unmount this filesystem.
1225 error = vfs_suser(mp, td);
1228 VOP_UNLOCK(coveredvp, 0);
1233 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
1236 VOP_UNLOCK(coveredvp, 0);
1239 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ;
1240 /* Allow filesystems to detect that a forced unmount is in progress. */
1241 if (flags & MNT_FORCE)
1242 mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1243 error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK |
1244 ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), MNT_MTX(mp));
1247 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_NOINSMNTQ |
1249 if (mp->mnt_kern_flag & MNTK_MWAIT)
1253 VOP_UNLOCK(coveredvp, 0);
1256 vn_start_write(NULL, &mp, V_WAIT);
1258 if (mp->mnt_flag & MNT_EXPUBLIC)
1259 vfs_setpublicfs(NULL, NULL, NULL);
1261 vfs_msync(mp, MNT_WAIT);
1263 async_flag = mp->mnt_flag & MNT_ASYNC;
1264 mp->mnt_flag &= ~MNT_ASYNC;
1265 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1267 cache_purgevfs(mp); /* remove cache entries for this file sys */
1268 if (mp->mnt_syncer != NULL)
1269 vrele(mp->mnt_syncer);
1271 * For forced unmounts, move process cdir/rdir refs on the fs root
1272 * vnode to the covered vnode. For non-forced unmounts we want
1273 * such references to cause an EBUSY error.
1275 if ((flags & MNT_FORCE) &&
1276 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) {
1277 if (mp->mnt_vnodecovered != NULL)
1278 mountcheckdirs(fsrootvp, mp->mnt_vnodecovered);
1279 if (fsrootvp == rootvnode) {
1285 if (((mp->mnt_flag & MNT_RDONLY) ||
1286 (error = VFS_SYNC(mp, MNT_WAIT, td)) == 0) ||
1287 (flags & MNT_FORCE)) {
1288 error = VFS_UNMOUNT(mp, flags, td);
1290 vn_finished_write(mp);
1292 * If we failed to flush the dirty blocks for this mount point,
1293 * undo all the cdir/rdir and rootvnode changes we made above.
1294 * Unless we failed to do so because the device is reporting that
1295 * it doesn't exist anymore.
1297 if (error && error != ENXIO) {
1298 if ((flags & MNT_FORCE) &&
1299 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) {
1300 if (mp->mnt_vnodecovered != NULL)
1301 mountcheckdirs(mp->mnt_vnodecovered, fsrootvp);
1302 if (rootvnode == NULL) {
1303 rootvnode = fsrootvp;
1309 mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ;
1310 if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL) {
1312 (void) vfs_allocate_syncvnode(mp);
1315 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1316 mp->mnt_flag |= async_flag;
1317 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
1318 mp->mnt_kern_flag |= MNTK_ASYNC;
1319 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL);
1320 if (mp->mnt_kern_flag & MNTK_MWAIT)
1324 VOP_UNLOCK(coveredvp, 0);
1327 mtx_lock(&mountlist_mtx);
1328 TAILQ_REMOVE(&mountlist, mp, mnt_list);
1329 mtx_unlock(&mountlist_mtx);
1330 if (coveredvp != NULL) {
1331 coveredvp->v_mountedhere = NULL;
1334 vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1335 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL);
1336 vfs_mount_destroy(mp);
1341 * ---------------------------------------------------------------------
1342 * Mounting of root filesystem
1346 struct root_hold_token {
1348 LIST_ENTRY(root_hold_token) list;
1351 static LIST_HEAD(, root_hold_token) root_holds =
1352 LIST_HEAD_INITIALIZER(&root_holds);
1354 static int root_mount_complete;
1359 struct root_hold_token *
1360 root_mount_hold(const char *identifier)
1362 struct root_hold_token *h;
1364 h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK);
1365 h->who = identifier;
1366 mtx_lock(&mountlist_mtx);
1367 LIST_INSERT_HEAD(&root_holds, h, list);
1368 mtx_unlock(&mountlist_mtx);
1373 * Release root mount.
1376 root_mount_rel(struct root_hold_token *h)
1379 mtx_lock(&mountlist_mtx);
1380 LIST_REMOVE(h, list);
1381 wakeup(&root_holds);
1382 mtx_unlock(&mountlist_mtx);
1387 * Wait for all subsystems to release root mount.
1390 root_mount_prepare(void)
1392 struct root_hold_token *h;
1398 mtx_lock(&mountlist_mtx);
1399 if (LIST_EMPTY(&root_holds)) {
1400 mtx_unlock(&mountlist_mtx);
1403 printf("Root mount waiting for:");
1404 LIST_FOREACH(h, &root_holds, list)
1405 printf(" %s", h->who);
1407 msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold",
1413 * Root was mounted, share the good news.
1416 root_mount_done(void)
1420 * Use a mutex to prevent the wakeup being missed and waiting for
1421 * an extra 1 second sleep.
1423 mtx_lock(&mountlist_mtx);
1424 root_mount_complete = 1;
1425 wakeup(&root_mount_complete);
1426 mtx_unlock(&mountlist_mtx);
1430 * Return true if root is already mounted.
1436 /* No mutex is acquired here because int stores are atomic. */
1437 return (root_mount_complete);
1441 * Wait until root is mounted.
1444 root_mount_wait(void)
1448 * Panic on an obvious deadlock - the function can't be called from
1449 * a thread which is doing the whole SYSINIT stuff.
1451 KASSERT(curthread->td_proc->p_pid != 0,
1452 ("root_mount_wait: cannot be called from the swapper thread"));
1453 mtx_lock(&mountlist_mtx);
1454 while (!root_mount_complete) {
1455 msleep(&root_mount_complete, &mountlist_mtx, PZERO, "rootwait",
1458 mtx_unlock(&mountlist_mtx);
1462 set_rootvnode(struct thread *td)
1466 if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode, td))
1467 panic("Cannot find root vnode");
1470 FILEDESC_XLOCK(p->p_fd);
1472 if (p->p_fd->fd_cdir != NULL)
1473 vrele(p->p_fd->fd_cdir);
1474 p->p_fd->fd_cdir = rootvnode;
1477 if (p->p_fd->fd_rdir != NULL)
1478 vrele(p->p_fd->fd_rdir);
1479 p->p_fd->fd_rdir = rootvnode;
1482 FILEDESC_XUNLOCK(p->p_fd);
1484 VOP_UNLOCK(rootvnode, 0);
1486 EVENTHANDLER_INVOKE(mountroot);
1490 * Mount /devfs as our root filesystem, but do not put it on the mountlist
1491 * yet. Create a /dev -> / symlink so that absolute pathnames will lookup.
1497 struct thread *td = curthread;
1498 struct vfsoptlist *opts;
1499 struct vfsconf *vfsp;
1500 struct mount *mp = NULL;
1503 vfsp = vfs_byname("devfs");
1504 KASSERT(vfsp != NULL, ("Could not find devfs by name"));
1508 mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td);
1510 error = VFS_MOUNT(mp, td);
1511 KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error));
1515 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
1519 mtx_lock(&mountlist_mtx);
1520 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
1521 mtx_unlock(&mountlist_mtx);
1525 error = kern_symlink(td, "/", "dev", UIO_SYSSPACE);
1527 printf("kern_symlink /dev -> / returns %d\n", error);
1531 * Surgically move our devfs to be mounted on /dev.
1535 devfs_fixup(struct thread *td)
1537 struct nameidata nd;
1539 struct vnode *vp, *dvp;
1542 /* Remove our devfs mount from the mountlist and purge the cache */
1543 mtx_lock(&mountlist_mtx);
1544 mp = TAILQ_FIRST(&mountlist);
1545 TAILQ_REMOVE(&mountlist, mp, mnt_list);
1546 mtx_unlock(&mountlist_mtx);
1549 VFS_ROOT(mp, LK_EXCLUSIVE, &dvp, td);
1551 dvp->v_iflag &= ~VI_MOUNT;
1553 dvp->v_mountedhere = NULL;
1555 /* Set up the real rootvnode, and purge the cache */
1556 TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL;
1558 cache_purgevfs(rootvnode->v_mount);
1560 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
1563 printf("Lookup of /dev for devfs, error: %d\n", error);
1566 NDFREE(&nd, NDF_ONLY_PNBUF);
1568 if (vp->v_type != VDIR) {
1571 error = vinvalbuf(vp, V_SAVE, td, 0, 0);
1576 mp->mnt_vnodecovered = vp;
1577 vp->v_mountedhere = mp;
1578 mtx_lock(&mountlist_mtx);
1579 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1580 mtx_unlock(&mountlist_mtx);
1585 /* Unlink the no longer needed /dev/dev -> / symlink */
1586 kern_unlink(td, "/dev/dev", UIO_SYSSPACE);
1590 * Report errors during filesystem mounting.
1593 vfs_mount_error(struct mount *mp, const char *fmt, ...)
1595 struct vfsoptlist *moptlist = mp->mnt_optnew;
1600 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1601 if (error || errmsg == NULL || len <= 0)
1605 vsnprintf(errmsg, (size_t)len, fmt, ap);
1610 * Find and mount the root filesystem
1616 int error, i, asked = 0;
1618 root_mount_prepare();
1620 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount),
1621 NULL, NULL, mount_init, mount_fini,
1622 UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
1626 * We are booted with instructions to prompt for the root filesystem.
1628 if (boothowto & RB_ASKNAME) {
1629 if (!vfs_mountroot_ask())
1635 * The root filesystem information is compiled in, and we are
1636 * booted with instructions to use it.
1638 if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) {
1639 if (!vfs_mountroot_try(ctrootdevname))
1641 ctrootdevname = NULL;
1645 * We've been given the generic "use CDROM as root" flag. This is
1646 * necessary because one media may be used in many different
1647 * devices, so we need to search for them.
1649 if (boothowto & RB_CDROM) {
1650 for (i = 0; cdrom_rootdevnames[i] != NULL; i++) {
1651 if (!vfs_mountroot_try(cdrom_rootdevnames[i]))
1657 * Try to use the value read by the loader from /etc/fstab, or
1658 * supplied via some other means. This is the preferred
1661 cp = getenv("vfs.root.mountfrom");
1663 error = vfs_mountroot_try(cp);
1670 * Try values that may have been computed by code during boot
1672 if (!vfs_mountroot_try(rootdevnames[0]))
1674 if (!vfs_mountroot_try(rootdevnames[1]))
1678 * If we (still) have a compiled-in default, try it.
1680 if (ctrootdevname != NULL)
1681 if (!vfs_mountroot_try(ctrootdevname))
1684 * Everything so far has failed, prompt on the console if we haven't
1685 * already tried that.
1688 if (!vfs_mountroot_ask())
1691 panic("Root mount failed, startup aborted.");
1698 * Mount (mountfrom) as the root filesystem.
1701 vfs_mountroot_try(const char *mountfrom)
1704 char *vfsname, *path;
1714 if (mountfrom == NULL)
1715 return (error); /* don't complain */
1716 printf("Trying to mount root from %s\n", mountfrom);
1718 /* parse vfs name and path */
1719 vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK);
1720 path = malloc(MNAMELEN, M_MOUNT, M_WAITOK);
1721 vfsname[0] = path[0] = 0;
1722 sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN);
1723 if (sscanf(mountfrom, patt, vfsname, path) < 1)
1726 if (path[0] == '\0')
1727 strcpy(path, ROOTNAME);
1729 error = kernel_vmount(
1730 MNT_RDONLY | MNT_ROOTFS,
1737 * We mount devfs prior to mounting the / FS, so the first
1738 * entry will typically be devfs.
1740 mp = TAILQ_FIRST(&mountlist);
1741 KASSERT(mp != NULL, ("%s: mountlist is empty", __func__));
1744 * Iterate over all currently mounted file systems and use
1745 * the time stamp found to check and/or initialize the RTC.
1746 * Typically devfs has no time stamp and the only other FS
1747 * is the actual / FS.
1748 * Call inittodr() only once and pass it the largest of the
1749 * timestamps we encounter.
1753 if (mp->mnt_time > timebase)
1754 timebase = mp->mnt_time;
1755 mp = TAILQ_NEXT(mp, mnt_list);
1756 } while (mp != NULL);
1759 devfs_fixup(curthread);
1762 free(path, M_MOUNT);
1763 free(vfsname, M_MOUNT);
1768 * ---------------------------------------------------------------------
1769 * Interactive root filesystem selection code.
1773 vfs_mountroot_ask(void)
1778 printf("\nManual root filesystem specification:\n");
1779 printf(" <fstype>:<device> Mount <device> using filesystem <fstype>\n");
1780 #if defined(__amd64__) || defined(__i386__) || defined(__ia64__)
1781 printf(" eg. ufs:da0s1a\n");
1783 printf(" eg. ufs:/dev/da0a\n");
1785 printf(" ? List valid disk boot devices\n");
1786 printf(" <empty line> Abort manual input\n");
1787 printf("\nmountroot> ");
1788 gets(name, sizeof(name), 1);
1789 if (name[0] == '\0')
1791 if (name[0] == '?') {
1792 printf("\nList of GEOM managed disk devices:\n ");
1796 if (!vfs_mountroot_try(name))
1802 * ---------------------------------------------------------------------
1803 * Functions for querying mount options/arguments from filesystems.
1807 * Check that no unknown options are given
1810 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1814 const char **t, *p, *q;
1817 TAILQ_FOREACH(opt, opts, link) {
1820 if (p[0] == 'n' && p[1] == 'o')
1822 for(t = global_opts; *t != NULL; t++) {
1823 if (strcmp(*t, p) == 0)
1826 if (strcmp(*t, q) == 0)
1832 for(t = legal; *t != NULL; t++) {
1833 if (strcmp(*t, p) == 0)
1836 if (strcmp(*t, q) == 0)
1842 sprintf(errmsg, "mount option <%s> is unknown", p);
1843 printf("%s\n", errmsg);
1847 TAILQ_FOREACH(opt, opts, link) {
1848 if (strcmp(opt->name, "errmsg") == 0) {
1849 strncpy((char *)opt->value, errmsg, opt->len);
1857 * Get a mount option by its name.
1859 * Return 0 if the option was found, ENOENT otherwise.
1860 * If len is non-NULL it will be filled with the length
1861 * of the option. If buf is non-NULL, it will be filled
1862 * with the address of the option.
1865 vfs_getopt(opts, name, buf, len)
1866 struct vfsoptlist *opts;
1873 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1875 TAILQ_FOREACH(opt, opts, link) {
1876 if (strcmp(name, opt->name) == 0) {
1888 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
1897 TAILQ_FOREACH(opt, opts, link) {
1898 if (strcmp(name, opt->name) == 0)
1906 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
1911 TAILQ_FOREACH(opt, opts, link) {
1912 if (strcmp(name, opt->name) != 0)
1914 if (((char *)opt->value)[opt->len - 1] != '\0') {
1918 return (opt->value);
1925 vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val)
1929 TAILQ_FOREACH(opt, opts, link) {
1930 if (strcmp(name, opt->name) == 0) {
1942 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
1948 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1950 TAILQ_FOREACH(opt, opts, link) {
1951 if (strcmp(name, opt->name) != 0)
1953 if (opt->len == 0 || opt->value == NULL)
1955 if (((char *)opt->value)[opt->len - 1] != '\0')
1958 ret = vsscanf(opt->value, fmt, ap);
1966 * Find and copy a mount option.
1968 * The size of the buffer has to be specified
1969 * in len, if it is not the same length as the
1970 * mount option, EINVAL is returned.
1971 * Returns ENOENT if the option is not found.
1974 vfs_copyopt(opts, name, dest, len)
1975 struct vfsoptlist *opts;
1982 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
1984 TAILQ_FOREACH(opt, opts, link) {
1985 if (strcmp(name, opt->name) == 0) {
1986 if (len != opt->len)
1988 bcopy(opt->value, dest, opt->len);
1996 * This is a helper function for filesystems to traverse their
1997 * vnodes. See MNT_VNODE_FOREACH() in sys/mount.h
2001 __mnt_vnode_next(struct vnode **mvp, struct mount *mp)
2005 mtx_assert(MNT_MTX(mp), MA_OWNED);
2007 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
2008 if ((*mvp)->v_yield++ == 500) {
2010 (*mvp)->v_yield = 0;
2014 vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
2015 while (vp != NULL && vp->v_type == VMARKER)
2016 vp = TAILQ_NEXT(vp, v_nmntvnodes);
2018 /* Check if we are done */
2020 __mnt_vnode_markerfree(mvp, mp);
2023 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
2024 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
2029 __mnt_vnode_first(struct vnode **mvp, struct mount *mp)
2033 mtx_assert(MNT_MTX(mp), MA_OWNED);
2035 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2036 while (vp != NULL && vp->v_type == VMARKER)
2037 vp = TAILQ_NEXT(vp, v_nmntvnodes);
2039 /* Check if we are done */
2046 *mvp = (struct vnode *) malloc(sizeof(struct vnode),
2050 (*mvp)->v_type = VMARKER;
2052 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2053 while (vp != NULL && vp->v_type == VMARKER)
2054 vp = TAILQ_NEXT(vp, v_nmntvnodes);
2056 /* Check if we are done */
2059 free(*mvp, M_VNODE_MARKER);
2063 if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0)
2064 wakeup(&mp->mnt_holdcnt);
2067 mp->mnt_markercnt++;
2068 (*mvp)->v_mount = mp;
2069 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
2075 __mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp)
2081 mtx_assert(MNT_MTX(mp), MA_OWNED);
2083 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
2084 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
2086 free(*mvp, M_VNODE_MARKER);
2090 mp->mnt_markercnt--;
2092 if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0)
2093 wakeup(&mp->mnt_holdcnt);
2098 __vfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td)
2102 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat, td);
2103 if (sbp != &mp->mnt_stat)
2104 *sbp = mp->mnt_stat;
2109 vfs_mountedfrom(struct mount *mp, const char *from)
2112 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
2113 strlcpy(mp->mnt_stat.f_mntfromname, from,
2114 sizeof mp->mnt_stat.f_mntfromname);
2118 * ---------------------------------------------------------------------
2119 * This is the api for building mount args and mounting filesystems from
2120 * inside the kernel.
2122 * The API works by accumulation of individual args. First error is
2125 * XXX: should be documented in new manpage kernel_mount(9)
2128 /* A memory allocation which must be freed when we are done */
2130 SLIST_ENTRY(mntaarg) next;
2133 /* The header for the mount arguments */
2138 SLIST_HEAD(, mntaarg) list;
2142 * Add a boolean argument.
2144 * flag is the boolean value.
2145 * name must start with "no".
2148 mount_argb(struct mntarg *ma, int flag, const char *name)
2151 KASSERT(name[0] == 'n' && name[1] == 'o',
2152 ("mount_argb(...,%s): name must start with 'no'", name));
2154 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
2158 * Add an argument printf style
2161 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
2164 struct mntaarg *maa;
2169 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2170 SLIST_INIT(&ma->list);
2175 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2177 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2178 ma->v[ma->len].iov_len = strlen(name) + 1;
2181 sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND);
2183 sbuf_vprintf(sb, fmt, ap);
2186 len = sbuf_len(sb) + 1;
2187 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2188 SLIST_INSERT_HEAD(&ma->list, maa, next);
2189 bcopy(sbuf_data(sb), maa + 1, len);
2192 ma->v[ma->len].iov_base = maa + 1;
2193 ma->v[ma->len].iov_len = len;
2200 * Add an argument which is a userland string.
2203 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2205 struct mntaarg *maa;
2211 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2212 SLIST_INIT(&ma->list);
2216 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2217 SLIST_INSERT_HEAD(&ma->list, maa, next);
2218 tbuf = (void *)(maa + 1);
2219 ma->error = copyinstr(val, tbuf, len, NULL);
2220 return (mount_arg(ma, name, tbuf, -1));
2226 * If length is -1, treat value as a C string.
2229 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2233 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2234 SLIST_INIT(&ma->list);
2239 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2241 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2242 ma->v[ma->len].iov_len = strlen(name) + 1;
2245 ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2247 ma->v[ma->len].iov_len = strlen(val) + 1;
2249 ma->v[ma->len].iov_len = len;
2255 * Free a mntarg structure
2258 free_mntarg(struct mntarg *ma)
2260 struct mntaarg *maa;
2262 while (!SLIST_EMPTY(&ma->list)) {
2263 maa = SLIST_FIRST(&ma->list);
2264 SLIST_REMOVE_HEAD(&ma->list, next);
2267 free(ma->v, M_MOUNT);
2272 * Mount a filesystem
2275 kernel_mount(struct mntarg *ma, int flags)
2280 KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2281 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
2282 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2284 auio.uio_iov = ma->v;
2285 auio.uio_iovcnt = ma->len;
2286 auio.uio_segflg = UIO_SYSSPACE;
2290 error = vfs_donmount(curthread, flags, &auio);
2296 * A printflike function to mount a filesystem.
2299 kernel_vmount(int flags, ...)
2301 struct mntarg *ma = NULL;
2307 va_start(ap, flags);
2309 cp = va_arg(ap, const char *);
2312 vp = va_arg(ap, const void *);
2313 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0));
2317 error = kernel_mount(ma, flags);