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/clock.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"
78 #define ROOTNAME "root_device"
79 #define VFS_MOUNTARG_SIZE_MAX (1024 * 64)
81 static int vfs_domount(struct thread *td, const char *fstype,
82 char *fspath, int fsflags, void *fsdata);
83 static int vfs_mountroot_ask(void);
84 static int vfs_mountroot_try(const char *mountfrom);
85 static int vfs_donmount(struct thread *td, int fsflags,
86 struct uio *fsoptions);
87 static void free_mntarg(struct mntarg *ma);
88 static int vfs_getopt_pos(struct vfsoptlist *opts, const char *name);
90 static int usermount = 0;
91 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
92 "Unprivileged users may mount and unmount file systems");
94 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
95 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
96 static uma_zone_t mount_zone;
98 /* List of mounted filesystems. */
99 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
101 /* For any iteration/modification of mountlist */
102 struct mtx mountlist_mtx;
103 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF);
105 TAILQ_HEAD(vfsoptlist, vfsopt);
107 TAILQ_ENTRY(vfsopt) link;
114 * The vnode of the system's root (/ in the filesystem, without chroot
117 struct vnode *rootvnode;
120 * The root filesystem is detailed in the kernel environment variable
121 * vfs.root.mountfrom, which is expected to be in the general format
124 * vfsname := the name of a VFS known to the kernel and capable
125 * of being mounted as root
126 * path := disk device name or other data used by the filesystem
127 * to locate its physical store
131 * Global opts, taken by all filesystems
133 static const char *global_opts[] = {
146 * The root specifiers we will try if RB_CDROM is specified.
148 static char *cdrom_rootdevnames[] = {
154 /* legacy find-root code */
155 char *rootdevnames[2] = {NULL, NULL};
157 # define ROOTDEVNAME NULL
159 static const char *ctrootdevname = ROOTDEVNAME;
162 * ---------------------------------------------------------------------
163 * Functions for building and sanitizing the mount options
166 /* Remove one mount option. */
168 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
171 TAILQ_REMOVE(opts, opt, link);
172 free(opt->name, M_MOUNT);
173 if (opt->value != NULL)
174 free(opt->value, M_MOUNT);
176 else if (opt->len != 0)
177 panic("%s: mount option with NULL value but length != 0",
183 /* Release all resources related to the mount options. */
185 vfs_freeopts(struct vfsoptlist *opts)
189 while (!TAILQ_EMPTY(opts)) {
190 opt = TAILQ_FIRST(opts);
191 vfs_freeopt(opts, opt);
197 vfs_deleteopt(struct vfsoptlist *opts, const char *name)
199 struct vfsopt *opt, *temp;
201 TAILQ_FOREACH_SAFE(opt, opts, link, temp) {
202 if (strcmp(opt->name, name) == 0)
203 vfs_freeopt(opts, opt);
208 * Check if options are equal (with or without the "no" prefix).
211 vfs_equalopts(const char *opt1, const char *opt2)
214 /* "opt" vs. "opt" or "noopt" vs. "noopt" */
215 if (strcmp(opt1, opt2) == 0)
217 /* "noopt" vs. "opt" */
218 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
220 /* "opt" vs. "noopt" */
221 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
227 * If a mount option is specified several times,
228 * (with or without the "no" prefix) only keep
229 * the last occurence of it.
232 vfs_sanitizeopts(struct vfsoptlist *opts)
234 struct vfsopt *opt, *opt2, *tmp;
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);
244 opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
251 * Build a linked list of mount options from a struct uio.
254 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
256 struct vfsoptlist *opts;
259 unsigned int i, iovcnt;
260 int error, namelen, optlen;
262 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
265 iovcnt = auio->uio_iovcnt;
266 for (i = 0; i < iovcnt; i += 2) {
267 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
268 namelen = auio->uio_iov[i].iov_len;
269 optlen = auio->uio_iov[i + 1].iov_len;
270 opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
275 * Do this early, so jumps to "bad" will free the current
278 TAILQ_INSERT_TAIL(opts, opt, link);
279 memused += sizeof(struct vfsopt) + optlen + namelen;
282 * Avoid consuming too much memory, and attempts to overflow
285 if (memused > VFS_MOUNTARG_SIZE_MAX ||
286 optlen > VFS_MOUNTARG_SIZE_MAX ||
287 namelen > VFS_MOUNTARG_SIZE_MAX) {
292 if (auio->uio_segflg == UIO_SYSSPACE) {
293 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
295 error = copyin(auio->uio_iov[i].iov_base, opt->name,
300 /* Ensure names are null-terminated strings. */
301 if (opt->name[namelen - 1] != '\0') {
307 opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
308 if (auio->uio_segflg == UIO_SYSSPACE) {
309 bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
312 error = copyin(auio->uio_iov[i + 1].iov_base,
319 vfs_sanitizeopts(opts);
328 * Merge the old mount options with the new ones passed
329 * in the MNT_UPDATE case.
331 * XXX This function will keep a "nofoo" option in the
332 * new options if there is no matching "foo" option
333 * to be cancelled in the old options. This is a bug
334 * if the option's canonical name is "foo". E.g., "noro"
335 * shouldn't end up in the mount point's active options,
339 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *opts)
341 struct vfsopt *opt, *opt2, *new;
343 TAILQ_FOREACH(opt, opts, link) {
345 * Check that this option hasn't been redefined
346 * nor cancelled with a "no" mount option.
348 opt2 = TAILQ_FIRST(toopts);
349 while (opt2 != NULL) {
350 if (strcmp(opt2->name, opt->name) == 0)
352 if (strncmp(opt2->name, "no", 2) == 0 &&
353 strcmp(opt2->name + 2, opt->name) == 0) {
354 vfs_freeopt(toopts, opt2);
357 opt2 = TAILQ_NEXT(opt2, link);
359 /* We want this option, duplicate it. */
360 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
361 new->name = malloc(strlen(opt->name) + 1, M_MOUNT, M_WAITOK);
362 strcpy(new->name, opt->name);
364 new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
365 bcopy(opt->value, new->value, opt->len);
370 TAILQ_INSERT_TAIL(toopts, new, link);
377 * Mount a filesystem.
382 struct nmount_args /* {
394 AUDIT_ARG(fflags, uap->flags);
397 * Filter out MNT_ROOTFS. We do not want clients of nmount() in
398 * userspace to set this flag, but we must filter it out if we want
399 * MNT_UPDATE on the root file system to work.
400 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try().
402 uap->flags &= ~MNT_ROOTFS;
404 iovcnt = uap->iovcnt;
406 * Check that we have an even number of iovec's
407 * and that we have at least two options.
409 if ((iovcnt & 1) || (iovcnt < 4))
412 error = copyinuio(uap->iovp, iovcnt, &auio);
416 for (i = 0; i < iovcnt; i++) {
417 if (iov->iov_len > MMAXOPTIONLEN) {
423 error = vfs_donmount(td, uap->flags, auio);
430 * ---------------------------------------------------------------------
431 * Various utility functions
435 vfs_ref(struct mount *mp)
444 vfs_rel(struct mount *mp)
453 mount_init(void *mem, int size, int flags)
457 mp = (struct mount *)mem;
458 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
459 lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0);
464 mount_fini(void *mem, int size)
468 mp = (struct mount *)mem;
469 lockdestroy(&mp->mnt_lock);
470 mtx_destroy(&mp->mnt_mtx);
474 * Allocate and initialize the mount point struct.
477 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp,
478 const char *fspath, struct thread *td)
482 mp = uma_zalloc(mount_zone, M_WAITOK);
483 bzero(&mp->mnt_startzero,
484 __rangeof(struct mount, mnt_startzero, mnt_endzero));
485 TAILQ_INIT(&mp->mnt_nvnodelist);
486 mp->mnt_nvnodelistsize = 0;
488 (void) vfs_busy(mp, LK_NOWAIT, 0, td);
489 mp->mnt_op = vfsp->vfc_vfsops;
491 vfsp->vfc_refcount++; /* XXX Unlocked */
492 mp->mnt_stat.f_type = vfsp->vfc_typenum;
494 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
495 mp->mnt_vnodecovered = vp;
496 mp->mnt_cred = crdup(td->td_ucred);
497 mp->mnt_stat.f_owner = td->td_ucred->cr_uid;
498 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
499 mp->mnt_iosize_max = DFLTPHYS;
502 mac_mount_create(td->td_ucred, mp);
504 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
509 * Destroy the mount struct previously allocated by vfs_mount_alloc().
512 vfs_mount_destroy(struct mount *mp)
517 for (i = 0; mp->mnt_ref && i < 3; i++)
518 msleep(mp, MNT_MTX(mp), PVFS, "mntref", hz);
520 * This will always cause a 3 second delay in rebooting due to
521 * refs on the root mountpoint that never go away. Most of these
522 * are held by init which never exits.
524 if (i == 3 && (!rebooting || bootverbose))
525 printf("Mount point %s had %d dangling refs\n",
526 mp->mnt_stat.f_mntonname, mp->mnt_ref);
527 if (mp->mnt_holdcnt != 0) {
528 printf("Waiting for mount point to be unheld\n");
529 while (mp->mnt_holdcnt != 0) {
530 mp->mnt_holdcntwaiters++;
531 msleep(&mp->mnt_holdcnt, MNT_MTX(mp),
532 PZERO, "mntdestroy", 0);
533 mp->mnt_holdcntwaiters--;
535 printf("mount point unheld\n");
537 if (mp->mnt_writeopcount > 0) {
538 printf("Waiting for mount point write ops\n");
539 while (mp->mnt_writeopcount > 0) {
540 mp->mnt_kern_flag |= MNTK_SUSPEND;
541 msleep(&mp->mnt_writeopcount,
543 PZERO, "mntdestroy2", 0);
545 printf("mount point write ops completed\n");
547 if (mp->mnt_secondary_writes > 0) {
548 printf("Waiting for mount point secondary write ops\n");
549 while (mp->mnt_secondary_writes > 0) {
550 mp->mnt_kern_flag |= MNTK_SUSPEND;
551 msleep(&mp->mnt_secondary_writes,
553 PZERO, "mntdestroy3", 0);
555 printf("mount point secondary write ops completed\n");
558 mp->mnt_vfc->vfc_refcount--;
559 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
562 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
564 panic("unmount: dangling vnode");
567 if (mp->mnt_kern_flag & MNTK_MWAIT)
569 if (mp->mnt_writeopcount != 0)
570 panic("vfs_mount_destroy: nonzero writeopcount");
571 if (mp->mnt_secondary_writes != 0)
572 panic("vfs_mount_destroy: nonzero secondary_writes");
573 if (mp->mnt_nvnodelistsize != 0)
574 panic("vfs_mount_destroy: nonzero nvnodelistsize");
575 mp->mnt_writeopcount = -1000;
576 mp->mnt_nvnodelistsize = -1000;
577 mp->mnt_secondary_writes = -1000;
580 mac_mount_destroy(mp);
582 if (mp->mnt_opt != NULL)
583 vfs_freeopts(mp->mnt_opt);
584 crfree(mp->mnt_cred);
585 uma_zfree(mount_zone, mp);
589 vfs_donmount(struct thread *td, int fsflags, struct uio *fsoptions)
591 struct vfsoptlist *optlist;
592 struct vfsopt *opt, *noro_opt;
593 char *fstype, *fspath, *errmsg;
594 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
595 int has_rw, has_noro;
603 error = vfs_buildopts(fsoptions, &optlist);
607 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
608 errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
611 * We need these two options before the others,
612 * and they are mandatory for any filesystem.
613 * Ensure they are NUL terminated as well.
616 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
617 if (error || fstype[fstypelen - 1] != '\0') {
620 strncpy(errmsg, "Invalid fstype", errmsg_len);
624 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
625 if (error || fspath[fspathlen - 1] != '\0') {
628 strncpy(errmsg, "Invalid fspath", errmsg_len);
633 * We need to see if we have the "update" option
634 * before we call vfs_domount(), since vfs_domount() has special
635 * logic based on MNT_UPDATE. This is very important
636 * when we want to update the root filesystem.
638 TAILQ_FOREACH(opt, optlist, link) {
639 if (strcmp(opt->name, "update") == 0)
640 fsflags |= MNT_UPDATE;
641 else if (strcmp(opt->name, "async") == 0)
642 fsflags |= MNT_ASYNC;
643 else if (strcmp(opt->name, "force") == 0)
644 fsflags |= MNT_FORCE;
645 else if (strcmp(opt->name, "multilabel") == 0)
646 fsflags |= MNT_MULTILABEL;
647 else if (strcmp(opt->name, "noasync") == 0)
648 fsflags &= ~MNT_ASYNC;
649 else if (strcmp(opt->name, "noatime") == 0)
650 fsflags |= MNT_NOATIME;
651 else if (strcmp(opt->name, "atime") == 0) {
652 free(opt->name, M_MOUNT);
653 opt->name = strdup("nonoatime", M_MOUNT);
655 else if (strcmp(opt->name, "noclusterr") == 0)
656 fsflags |= MNT_NOCLUSTERR;
657 else if (strcmp(opt->name, "clusterr") == 0) {
658 free(opt->name, M_MOUNT);
659 opt->name = strdup("nonoclusterr", M_MOUNT);
661 else if (strcmp(opt->name, "noclusterw") == 0)
662 fsflags |= MNT_NOCLUSTERW;
663 else if (strcmp(opt->name, "clusterw") == 0) {
664 free(opt->name, M_MOUNT);
665 opt->name = strdup("nonoclusterw", M_MOUNT);
667 else if (strcmp(opt->name, "noexec") == 0)
668 fsflags |= MNT_NOEXEC;
669 else if (strcmp(opt->name, "exec") == 0) {
670 free(opt->name, M_MOUNT);
671 opt->name = strdup("nonoexec", M_MOUNT);
673 else if (strcmp(opt->name, "nosuid") == 0)
674 fsflags |= MNT_NOSUID;
675 else if (strcmp(opt->name, "suid") == 0) {
676 free(opt->name, M_MOUNT);
677 opt->name = strdup("nonosuid", M_MOUNT);
679 else if (strcmp(opt->name, "nosymfollow") == 0)
680 fsflags |= MNT_NOSYMFOLLOW;
681 else if (strcmp(opt->name, "symfollow") == 0) {
682 free(opt->name, M_MOUNT);
683 opt->name = strdup("nonosymfollow", M_MOUNT);
685 else if (strcmp(opt->name, "noro") == 0) {
686 fsflags &= ~MNT_RDONLY;
689 else if (strcmp(opt->name, "rw") == 0) {
690 fsflags &= ~MNT_RDONLY;
693 else if (strcmp(opt->name, "ro") == 0)
694 fsflags |= MNT_RDONLY;
695 else if (strcmp(opt->name, "rdonly") == 0) {
696 free(opt->name, M_MOUNT);
697 opt->name = strdup("ro", M_MOUNT);
698 fsflags |= MNT_RDONLY;
700 else if (strcmp(opt->name, "snapshot") == 0)
701 fsflags |= MNT_SNAPSHOT;
702 else if (strcmp(opt->name, "suiddir") == 0)
703 fsflags |= MNT_SUIDDIR;
704 else if (strcmp(opt->name, "sync") == 0)
705 fsflags |= MNT_SYNCHRONOUS;
706 else if (strcmp(opt->name, "union") == 0)
707 fsflags |= MNT_UNION;
711 * If "rw" was specified as a mount option, and we
712 * are trying to update a mount-point from "ro" to "rw",
713 * we need a mount option "noro", since in vfs_mergeopts(),
714 * "noro" will cancel "ro", but "rw" will not do anything.
716 if (has_rw && !has_noro) {
717 noro_opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
718 noro_opt->name = strdup("noro", M_MOUNT);
719 noro_opt->value = NULL;
721 TAILQ_INSERT_TAIL(optlist, noro_opt, link);
725 * Be ultra-paranoid about making sure the type and fspath
726 * variables will fit in our mp buffers, including the
729 if (fstypelen >= MFSNAMELEN - 1 || fspathlen >= MNAMELEN - 1) {
730 error = ENAMETOOLONG;
735 error = vfs_domount(td, fstype, fspath, fsflags, optlist);
738 /* copyout the errmsg */
739 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
740 && errmsg_len > 0 && errmsg != NULL) {
741 if (fsoptions->uio_segflg == UIO_SYSSPACE) {
743 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
744 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
747 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
748 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
753 vfs_freeopts(optlist);
760 #ifndef _SYS_SYSPROTO_H_
772 struct mount_args /* {
780 struct vfsconf *vfsp = NULL;
781 struct mntarg *ma = NULL;
784 AUDIT_ARG(fflags, uap->flags);
787 * Filter out MNT_ROOTFS. We do not want clients of mount() in
788 * userspace to set this flag, but we must filter it out if we want
789 * MNT_UPDATE on the root file system to work.
790 * MNT_ROOTFS should only be set in the kernel in vfs_mountroot_try().
792 uap->flags &= ~MNT_ROOTFS;
794 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
795 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
797 free(fstype, M_TEMP);
801 AUDIT_ARG(text, fstype);
803 vfsp = vfs_byname_kld(fstype, td, &error);
804 free(fstype, M_TEMP);
809 if (vfsp->vfc_vfsops->vfs_cmount == NULL) {
814 ma = mount_argsu(ma, "fstype", uap->type, MNAMELEN);
815 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
816 ma = mount_argb(ma, uap->flags & MNT_RDONLY, "noro");
817 ma = mount_argb(ma, !(uap->flags & MNT_NOSUID), "nosuid");
818 ma = mount_argb(ma, !(uap->flags & MNT_NOEXEC), "noexec");
820 error = vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, uap->flags, td);
827 * vfs_domount(): actually attempt a filesystem mount.
831 struct thread *td, /* Calling thread. */
832 const char *fstype, /* Filesystem type. */
833 char *fspath, /* Mount path. */
834 int fsflags, /* Flags common to all filesystems. */
835 void *fsdata /* Options local to the filesystem. */
840 struct vfsconf *vfsp;
841 struct export_args export;
846 mtx_assert(&Giant, MA_OWNED);
848 * Be ultra-paranoid about making sure the type and fspath
849 * variables will fit in our mp buffers, including the
852 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
853 return (ENAMETOOLONG);
855 if (jailed(td->td_ucred) || usermount == 0) {
856 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
861 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
863 if (fsflags & MNT_EXPORTED) {
864 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
868 if (fsflags & MNT_SUIDDIR) {
869 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
874 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
876 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
877 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
878 fsflags |= MNT_NOSUID | MNT_USER;
881 /* Load KLDs before we lock the covered vnode to avoid reversals. */
883 if ((fsflags & MNT_UPDATE) == 0) {
884 /* Don't try to load KLDs if we're mounting the root. */
885 if (fsflags & MNT_ROOTFS)
886 vfsp = vfs_byname(fstype);
888 vfsp = vfs_byname_kld(fstype, td, &error);
891 if (jailed(td->td_ucred) && !(vfsp->vfc_flags & VFCF_JAIL))
895 * Get vnode to be covered
897 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE,
899 if ((error = namei(&nd)) != 0)
901 NDFREE(&nd, NDF_ONLY_PNBUF);
903 if (fsflags & MNT_UPDATE) {
904 if ((vp->v_vflag & VV_ROOT) == 0) {
912 * We only allow the filesystem to be reloaded if it
913 * is currently mounted read-only.
915 if ((fsflags & MNT_RELOAD) &&
916 ((mp->mnt_flag & MNT_RDONLY) == 0)) {
919 return (EOPNOTSUPP); /* Needs translation */
923 * Only privileged root, or (if MNT_USER is set) the user that
924 * did the original mount is permitted to update it.
926 error = vfs_suser(mp, td);
931 if (vfs_busy(mp, LK_NOWAIT, 0, td)) {
936 if ((vp->v_iflag & VI_MOUNT) != 0 ||
937 vp->v_mountedhere != NULL) {
943 vp->v_iflag |= VI_MOUNT;
946 mp->mnt_flag |= fsflags &
947 (MNT_RELOAD | MNT_FORCE | MNT_UPDATE | MNT_SNAPSHOT | MNT_ROOTFS);
950 mp->mnt_optnew = fsdata;
951 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
954 * If the user is not root, ensure that they own the directory
955 * onto which we are attempting to mount.
957 error = VOP_GETATTR(vp, &va, td->td_ucred, td);
962 if (va.va_uid != td->td_ucred->cr_uid) {
963 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN,
970 error = vinvalbuf(vp, V_SAVE, td, 0, 0);
975 if (vp->v_type != VDIR) {
980 if ((vp->v_iflag & VI_MOUNT) != 0 ||
981 vp->v_mountedhere != NULL) {
986 vp->v_iflag |= VI_MOUNT;
990 * Allocate and initialize the filesystem.
992 mp = vfs_mount_alloc(vp, vfsp, fspath, td);
995 /* XXXMAC: pass to vfs_mount_alloc? */
996 mp->mnt_optnew = fsdata;
1000 * Set the mount level flags.
1003 mp->mnt_flag = (mp->mnt_flag & ~MNT_UPDATEMASK) |
1004 (fsflags & (MNT_UPDATEMASK | MNT_FORCE | MNT_ROOTFS |
1006 if ((mp->mnt_flag & MNT_ASYNC) == 0)
1007 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1010 * Mount the filesystem.
1011 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1012 * get. No freeing of cn_pnbuf.
1014 error = VFS_MOUNT(mp, td);
1017 * Process the export option only if we are
1018 * updating mount options.
1020 if (!error && (fsflags & MNT_UPDATE)) {
1021 if (vfs_copyopt(mp->mnt_optnew, "export", &export,
1022 sizeof(export)) == 0)
1023 error = vfs_export(mp, &export);
1027 if (mp->mnt_opt != NULL)
1028 vfs_freeopts(mp->mnt_opt);
1029 mp->mnt_opt = mp->mnt_optnew;
1030 (void)VFS_STATFS(mp, &mp->mnt_stat, td);
1033 * Prevent external consumers of mount options from reading
1036 mp->mnt_optnew = NULL;
1037 if (mp->mnt_flag & MNT_UPDATE) {
1040 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) |
1041 (flag & ~MNT_QUOTA);
1043 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD |
1044 MNT_FORCE | MNT_SNAPSHOT);
1045 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
1046 mp->mnt_kern_flag |= MNTK_ASYNC;
1048 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1050 if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1051 if (mp->mnt_syncer == NULL)
1052 error = vfs_allocate_syncvnode(mp);
1054 if (mp->mnt_syncer != NULL)
1055 vrele(mp->mnt_syncer);
1056 mp->mnt_syncer = NULL;
1060 vp->v_iflag &= ~VI_MOUNT;
1066 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
1067 mp->mnt_kern_flag |= MNTK_ASYNC;
1069 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1071 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1073 * Put the new filesystem on the mount list after root.
1077 struct vnode *newdp;
1080 vp->v_iflag &= ~VI_MOUNT;
1082 vp->v_mountedhere = mp;
1083 mtx_lock(&mountlist_mtx);
1084 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1085 mtx_unlock(&mountlist_mtx);
1086 vfs_event_signal(NULL, VQ_MOUNT, 0);
1087 if (VFS_ROOT(mp, LK_EXCLUSIVE, &newdp, td))
1088 panic("mount: lost mount");
1089 mountcheckdirs(vp, newdp);
1092 if ((mp->mnt_flag & MNT_RDONLY) == 0)
1093 error = vfs_allocate_syncvnode(mp);
1099 vp->v_iflag &= ~VI_MOUNT;
1102 vfs_mount_destroy(mp);
1109 * Unmount a filesystem.
1111 * Note: unmount takes a path to the vnode mounted on as argument, not
1112 * special file (as before).
1114 #ifndef _SYS_SYSPROTO_H_
1115 struct unmount_args {
1124 register struct unmount_args /* {
1131 int error, id0, id1;
1133 if (jailed(td->td_ucred) || usermount == 0) {
1134 error = priv_check(td, PRIV_VFS_UNMOUNT);
1139 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1140 error = copyinstr(uap->path, pathbuf, MNAMELEN, NULL);
1142 free(pathbuf, M_TEMP);
1145 AUDIT_ARG(upath, td, pathbuf, ARG_UPATH1);
1147 if (uap->flags & MNT_BYFSID) {
1148 /* Decode the filesystem ID. */
1149 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1151 free(pathbuf, M_TEMP);
1155 mtx_lock(&mountlist_mtx);
1156 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1157 if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1158 mp->mnt_stat.f_fsid.val[1] == id1)
1161 mtx_unlock(&mountlist_mtx);
1163 mtx_lock(&mountlist_mtx);
1164 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1165 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0)
1168 mtx_unlock(&mountlist_mtx);
1170 free(pathbuf, M_TEMP);
1173 * Previously we returned ENOENT for a nonexistent path and
1174 * EINVAL for a non-mountpoint. We cannot tell these apart
1175 * now, so in the !MNT_BYFSID case return the more likely
1176 * EINVAL for compatibility.
1179 return ((uap->flags & MNT_BYFSID) ? ENOENT : EINVAL);
1183 * Don't allow unmounting the root filesystem.
1185 if (mp->mnt_flag & MNT_ROOTFS) {
1189 error = dounmount(mp, uap->flags, td);
1195 * Do the actual filesystem unmount.
1198 dounmount(mp, flags, td)
1203 struct vnode *coveredvp, *fsrootvp;
1208 mtx_assert(&Giant, MA_OWNED);
1210 if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
1211 mnt_gen_r = mp->mnt_gen;
1214 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
1217 * Check for mp being unmounted while waiting for the
1218 * covered vnode lock.
1220 if (coveredvp->v_mountedhere != mp ||
1221 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
1222 VOP_UNLOCK(coveredvp, 0);
1227 * Only privileged root, or (if MNT_USER is set) the user that did the
1228 * original mount is permitted to unmount this filesystem.
1230 error = vfs_suser(mp, td);
1233 VOP_UNLOCK(coveredvp, 0);
1238 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
1241 VOP_UNLOCK(coveredvp, 0);
1244 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_NOINSMNTQ;
1245 /* Allow filesystems to detect that a forced unmount is in progress. */
1246 if (flags & MNT_FORCE)
1247 mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1248 error = lockmgr(&mp->mnt_lock, LK_DRAIN | LK_INTERLOCK |
1249 ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), MNT_MTX(mp));
1252 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_NOINSMNTQ |
1254 if (mp->mnt_kern_flag & MNTK_MWAIT)
1258 VOP_UNLOCK(coveredvp, 0);
1261 vn_start_write(NULL, &mp, V_WAIT);
1263 if (mp->mnt_flag & MNT_EXPUBLIC)
1264 vfs_setpublicfs(NULL, NULL, NULL);
1266 vfs_msync(mp, MNT_WAIT);
1268 async_flag = mp->mnt_flag & MNT_ASYNC;
1269 mp->mnt_flag &= ~MNT_ASYNC;
1270 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1272 cache_purgevfs(mp); /* remove cache entries for this file sys */
1273 if (mp->mnt_syncer != NULL)
1274 vrele(mp->mnt_syncer);
1276 * For forced unmounts, move process cdir/rdir refs on the fs root
1277 * vnode to the covered vnode. For non-forced unmounts we want
1278 * such references to cause an EBUSY error.
1280 if ((flags & MNT_FORCE) &&
1281 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) {
1282 if (mp->mnt_vnodecovered != NULL)
1283 mountcheckdirs(fsrootvp, mp->mnt_vnodecovered);
1284 if (fsrootvp == rootvnode) {
1290 if (((mp->mnt_flag & MNT_RDONLY) ||
1291 (error = VFS_SYNC(mp, MNT_WAIT, td)) == 0) ||
1292 (flags & MNT_FORCE)) {
1293 error = VFS_UNMOUNT(mp, flags, td);
1295 vn_finished_write(mp);
1297 * If we failed to flush the dirty blocks for this mount point,
1298 * undo all the cdir/rdir and rootvnode changes we made above.
1299 * Unless we failed to do so because the device is reporting that
1300 * it doesn't exist anymore.
1302 if (error && error != ENXIO) {
1303 if ((flags & MNT_FORCE) &&
1304 VFS_ROOT(mp, LK_EXCLUSIVE, &fsrootvp, td) == 0) {
1305 if (mp->mnt_vnodecovered != NULL)
1306 mountcheckdirs(mp->mnt_vnodecovered, fsrootvp);
1307 if (rootvnode == NULL) {
1308 rootvnode = fsrootvp;
1314 mp->mnt_kern_flag &= ~MNTK_NOINSMNTQ;
1315 if ((mp->mnt_flag & MNT_RDONLY) == 0 && mp->mnt_syncer == NULL) {
1317 (void) vfs_allocate_syncvnode(mp);
1320 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1321 mp->mnt_flag |= async_flag;
1322 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
1323 mp->mnt_kern_flag |= MNTK_ASYNC;
1324 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL);
1325 if (mp->mnt_kern_flag & MNTK_MWAIT)
1329 VOP_UNLOCK(coveredvp, 0);
1332 mtx_lock(&mountlist_mtx);
1333 TAILQ_REMOVE(&mountlist, mp, mnt_list);
1334 mtx_unlock(&mountlist_mtx);
1335 if (coveredvp != NULL) {
1336 coveredvp->v_mountedhere = NULL;
1339 vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1340 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL);
1341 vfs_mount_destroy(mp);
1346 * ---------------------------------------------------------------------
1347 * Mounting of root filesystem
1351 struct root_hold_token {
1353 LIST_ENTRY(root_hold_token) list;
1356 static LIST_HEAD(, root_hold_token) root_holds =
1357 LIST_HEAD_INITIALIZER(&root_holds);
1359 static int root_mount_complete;
1364 struct root_hold_token *
1365 root_mount_hold(const char *identifier)
1367 struct root_hold_token *h;
1369 h = malloc(sizeof *h, M_DEVBUF, M_ZERO | M_WAITOK);
1370 h->who = identifier;
1371 mtx_lock(&mountlist_mtx);
1372 LIST_INSERT_HEAD(&root_holds, h, list);
1373 mtx_unlock(&mountlist_mtx);
1378 * Release root mount.
1381 root_mount_rel(struct root_hold_token *h)
1384 mtx_lock(&mountlist_mtx);
1385 LIST_REMOVE(h, list);
1386 wakeup(&root_holds);
1387 mtx_unlock(&mountlist_mtx);
1392 * Wait for all subsystems to release root mount.
1395 root_mount_prepare(void)
1397 struct root_hold_token *h;
1403 mtx_lock(&mountlist_mtx);
1404 if (LIST_EMPTY(&root_holds)) {
1405 mtx_unlock(&mountlist_mtx);
1408 printf("Root mount waiting for:");
1409 LIST_FOREACH(h, &root_holds, list)
1410 printf(" %s", h->who);
1412 msleep(&root_holds, &mountlist_mtx, PZERO | PDROP, "roothold",
1418 * Root was mounted, share the good news.
1421 root_mount_done(void)
1425 * Use a mutex to prevent the wakeup being missed and waiting for
1426 * an extra 1 second sleep.
1428 mtx_lock(&mountlist_mtx);
1429 root_mount_complete = 1;
1430 wakeup(&root_mount_complete);
1431 mtx_unlock(&mountlist_mtx);
1435 * Return true if root is already mounted.
1441 /* No mutex is acquired here because int stores are atomic. */
1442 return (root_mount_complete);
1446 * Wait until root is mounted.
1449 root_mount_wait(void)
1453 * Panic on an obvious deadlock - the function can't be called from
1454 * a thread which is doing the whole SYSINIT stuff.
1456 KASSERT(curthread->td_proc->p_pid != 0,
1457 ("root_mount_wait: cannot be called from the swapper thread"));
1458 mtx_lock(&mountlist_mtx);
1459 while (!root_mount_complete) {
1460 msleep(&root_mount_complete, &mountlist_mtx, PZERO, "rootwait",
1463 mtx_unlock(&mountlist_mtx);
1467 set_rootvnode(struct thread *td)
1471 if (VFS_ROOT(TAILQ_FIRST(&mountlist), LK_EXCLUSIVE, &rootvnode, td))
1472 panic("Cannot find root vnode");
1475 FILEDESC_SLOCK(p->p_fd);
1477 if (p->p_fd->fd_cdir != NULL)
1478 vrele(p->p_fd->fd_cdir);
1479 p->p_fd->fd_cdir = rootvnode;
1482 if (p->p_fd->fd_rdir != NULL)
1483 vrele(p->p_fd->fd_rdir);
1484 p->p_fd->fd_rdir = rootvnode;
1487 FILEDESC_SUNLOCK(p->p_fd);
1489 VOP_UNLOCK(rootvnode, 0);
1493 * Mount /devfs as our root filesystem, but do not put it on the mountlist
1494 * yet. Create a /dev -> / symlink so that absolute pathnames will lookup.
1500 struct thread *td = curthread;
1501 struct vfsoptlist *opts;
1502 struct vfsconf *vfsp;
1503 struct mount *mp = NULL;
1506 vfsp = vfs_byname("devfs");
1507 KASSERT(vfsp != NULL, ("Could not find devfs by name"));
1511 mp = vfs_mount_alloc(NULLVP, vfsp, "/dev", td);
1513 error = VFS_MOUNT(mp, td);
1514 KASSERT(error == 0, ("VFS_MOUNT(devfs) failed %d", error));
1518 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
1522 mtx_lock(&mountlist_mtx);
1523 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
1524 mtx_unlock(&mountlist_mtx);
1528 error = kern_symlink(td, "/", "dev", UIO_SYSSPACE);
1530 printf("kern_symlink /dev -> / returns %d\n", error);
1534 * Surgically move our devfs to be mounted on /dev.
1538 devfs_fixup(struct thread *td)
1540 struct nameidata nd;
1542 struct vnode *vp, *dvp;
1545 /* Remove our devfs mount from the mountlist and purge the cache */
1546 mtx_lock(&mountlist_mtx);
1547 mp = TAILQ_FIRST(&mountlist);
1548 TAILQ_REMOVE(&mountlist, mp, mnt_list);
1549 mtx_unlock(&mountlist_mtx);
1552 VFS_ROOT(mp, LK_EXCLUSIVE, &dvp, td);
1554 dvp->v_iflag &= ~VI_MOUNT;
1556 dvp->v_mountedhere = NULL;
1558 /* Set up the real rootvnode, and purge the cache */
1559 TAILQ_FIRST(&mountlist)->mnt_vnodecovered = NULL;
1561 cache_purgevfs(rootvnode->v_mount);
1563 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, "/dev", td);
1566 printf("Lookup of /dev for devfs, error: %d\n", error);
1569 NDFREE(&nd, NDF_ONLY_PNBUF);
1571 if (vp->v_type != VDIR) {
1574 error = vinvalbuf(vp, V_SAVE, td, 0, 0);
1579 mp->mnt_vnodecovered = vp;
1580 vp->v_mountedhere = mp;
1581 mtx_lock(&mountlist_mtx);
1582 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1583 mtx_unlock(&mountlist_mtx);
1588 /* Unlink the no longer needed /dev/dev -> / symlink */
1589 kern_unlink(td, "/dev/dev", UIO_SYSSPACE);
1593 * Report errors during filesystem mounting.
1596 vfs_mount_error(struct mount *mp, const char *fmt, ...)
1598 struct vfsoptlist *moptlist = mp->mnt_optnew;
1603 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1604 if (error || errmsg == NULL || len <= 0)
1608 vsnprintf(errmsg, (size_t)len, fmt, ap);
1613 * Find and mount the root filesystem
1619 int error, i, asked = 0;
1621 root_mount_prepare();
1623 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount),
1624 NULL, NULL, mount_init, mount_fini,
1625 UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
1629 * We are booted with instructions to prompt for the root filesystem.
1631 if (boothowto & RB_ASKNAME) {
1632 if (!vfs_mountroot_ask())
1638 * The root filesystem information is compiled in, and we are
1639 * booted with instructions to use it.
1641 if (ctrootdevname != NULL && (boothowto & RB_DFLTROOT)) {
1642 if (!vfs_mountroot_try(ctrootdevname))
1644 ctrootdevname = NULL;
1648 * We've been given the generic "use CDROM as root" flag. This is
1649 * necessary because one media may be used in many different
1650 * devices, so we need to search for them.
1652 if (boothowto & RB_CDROM) {
1653 for (i = 0; cdrom_rootdevnames[i] != NULL; i++) {
1654 if (!vfs_mountroot_try(cdrom_rootdevnames[i]))
1660 * Try to use the value read by the loader from /etc/fstab, or
1661 * supplied via some other means. This is the preferred
1664 cp = getenv("vfs.root.mountfrom");
1666 error = vfs_mountroot_try(cp);
1673 * Try values that may have been computed by code during boot
1675 if (!vfs_mountroot_try(rootdevnames[0]))
1677 if (!vfs_mountroot_try(rootdevnames[1]))
1681 * If we (still) have a compiled-in default, try it.
1683 if (ctrootdevname != NULL)
1684 if (!vfs_mountroot_try(ctrootdevname))
1687 * Everything so far has failed, prompt on the console if we haven't
1688 * already tried that.
1691 if (!vfs_mountroot_ask())
1694 panic("Root mount failed, startup aborted.");
1701 * Mount (mountfrom) as the root filesystem.
1704 vfs_mountroot_try(const char *mountfrom)
1707 char *vfsname, *path;
1717 if (mountfrom == NULL)
1718 return (error); /* don't complain */
1719 printf("Trying to mount root from %s\n", mountfrom);
1721 /* parse vfs name and path */
1722 vfsname = malloc(MFSNAMELEN, M_MOUNT, M_WAITOK);
1723 path = malloc(MNAMELEN, M_MOUNT, M_WAITOK);
1724 vfsname[0] = path[0] = 0;
1725 sprintf(patt, "%%%d[a-z0-9]:%%%ds", MFSNAMELEN, MNAMELEN);
1726 if (sscanf(mountfrom, patt, vfsname, path) < 1)
1729 if (path[0] == '\0')
1730 strcpy(path, ROOTNAME);
1732 error = kernel_vmount(
1733 MNT_RDONLY | MNT_ROOTFS,
1740 * We mount devfs prior to mounting the / FS, so the first
1741 * entry will typically be devfs.
1743 mp = TAILQ_FIRST(&mountlist);
1744 KASSERT(mp != NULL, ("%s: mountlist is empty", __func__));
1747 * Iterate over all currently mounted file systems and use
1748 * the time stamp found to check and/or initialize the RTC.
1749 * Typically devfs has no time stamp and the only other FS
1750 * is the actual / FS.
1751 * Call inittodr() only once and pass it the largest of the
1752 * timestamps we encounter.
1756 if (mp->mnt_time > timebase)
1757 timebase = mp->mnt_time;
1758 mp = TAILQ_NEXT(mp, mnt_list);
1759 } while (mp != NULL);
1762 devfs_fixup(curthread);
1765 free(path, M_MOUNT);
1766 free(vfsname, M_MOUNT);
1771 * ---------------------------------------------------------------------
1772 * Interactive root filesystem selection code.
1776 vfs_mountroot_ask(void)
1781 printf("\nManual root filesystem specification:\n");
1782 printf(" <fstype>:<device> Mount <device> using filesystem <fstype>\n");
1783 #if defined(__amd64__) || defined(__i386__) || defined(__ia64__)
1784 printf(" eg. ufs:da0s1a\n");
1786 printf(" eg. ufs:/dev/da0a\n");
1788 printf(" ? List valid disk boot devices\n");
1789 printf(" <empty line> Abort manual input\n");
1790 printf("\nmountroot> ");
1791 gets(name, sizeof(name), 1);
1792 if (name[0] == '\0')
1794 if (name[0] == '?') {
1795 printf("\nList of GEOM managed disk devices:\n ");
1799 if (!vfs_mountroot_try(name))
1805 * ---------------------------------------------------------------------
1806 * Functions for querying mount options/arguments from filesystems.
1810 * Check that no unknown options are given
1813 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1817 const char **t, *p, *q;
1820 TAILQ_FOREACH(opt, opts, link) {
1823 if (p[0] == 'n' && p[1] == 'o')
1825 for(t = global_opts; *t != NULL; t++) {
1826 if (strcmp(*t, p) == 0)
1829 if (strcmp(*t, q) == 0)
1835 for(t = legal; *t != NULL; t++) {
1836 if (strcmp(*t, p) == 0)
1839 if (strcmp(*t, q) == 0)
1845 sprintf(errmsg, "mount option <%s> is unknown", p);
1846 printf("%s\n", errmsg);
1850 TAILQ_FOREACH(opt, opts, link) {
1851 if (strcmp(opt->name, "errmsg") == 0) {
1852 strncpy((char *)opt->value, errmsg, opt->len);
1860 * Get a mount option by its name.
1862 * Return 0 if the option was found, ENOENT otherwise.
1863 * If len is non-NULL it will be filled with the length
1864 * of the option. If buf is non-NULL, it will be filled
1865 * with the address of the option.
1868 vfs_getopt(opts, name, buf, len)
1869 struct vfsoptlist *opts;
1876 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1878 TAILQ_FOREACH(opt, opts, link) {
1879 if (strcmp(name, opt->name) == 0) {
1891 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
1900 TAILQ_FOREACH(opt, opts, link) {
1901 if (strcmp(name, opt->name) == 0)
1909 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
1914 TAILQ_FOREACH(opt, opts, link) {
1915 if (strcmp(name, opt->name) != 0)
1917 if (((char *)opt->value)[opt->len - 1] != '\0') {
1921 return (opt->value);
1928 vfs_flagopt(struct vfsoptlist *opts, const char *name, u_int *w, u_int val)
1932 TAILQ_FOREACH(opt, opts, link) {
1933 if (strcmp(name, opt->name) == 0) {
1945 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
1951 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
1953 TAILQ_FOREACH(opt, opts, link) {
1954 if (strcmp(name, opt->name) != 0)
1956 if (opt->len == 0 || opt->value == NULL)
1958 if (((char *)opt->value)[opt->len - 1] != '\0')
1961 ret = vsscanf(opt->value, fmt, ap);
1969 * Find and copy a mount option.
1971 * The size of the buffer has to be specified
1972 * in len, if it is not the same length as the
1973 * mount option, EINVAL is returned.
1974 * Returns ENOENT if the option is not found.
1977 vfs_copyopt(opts, name, dest, len)
1978 struct vfsoptlist *opts;
1985 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
1987 TAILQ_FOREACH(opt, opts, link) {
1988 if (strcmp(name, opt->name) == 0) {
1989 if (len != opt->len)
1991 bcopy(opt->value, dest, opt->len);
1999 * This is a helper function for filesystems to traverse their
2000 * vnodes. See MNT_VNODE_FOREACH() in sys/mount.h
2004 __mnt_vnode_next(struct vnode **mvp, struct mount *mp)
2008 mtx_assert(MNT_MTX(mp), MA_OWNED);
2010 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
2011 vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
2012 while (vp != NULL && vp->v_type == VMARKER)
2013 vp = TAILQ_NEXT(vp, v_nmntvnodes);
2015 /* Check if we are done */
2017 __mnt_vnode_markerfree(mvp, mp);
2020 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
2021 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
2026 __mnt_vnode_first(struct vnode **mvp, struct mount *mp)
2030 mtx_assert(MNT_MTX(mp), MA_OWNED);
2032 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2033 while (vp != NULL && vp->v_type == VMARKER)
2034 vp = TAILQ_NEXT(vp, v_nmntvnodes);
2036 /* Check if we are done */
2043 *mvp = (struct vnode *) malloc(sizeof(struct vnode),
2047 (*mvp)->v_type = VMARKER;
2049 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2050 while (vp != NULL && vp->v_type == VMARKER)
2051 vp = TAILQ_NEXT(vp, v_nmntvnodes);
2053 /* Check if we are done */
2056 free(*mvp, M_VNODE_MARKER);
2060 if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0)
2061 wakeup(&mp->mnt_holdcnt);
2064 mp->mnt_markercnt++;
2065 (*mvp)->v_mount = mp;
2066 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
2072 __mnt_vnode_markerfree(struct vnode **mvp, struct mount *mp)
2078 mtx_assert(MNT_MTX(mp), MA_OWNED);
2080 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
2081 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
2083 free(*mvp, M_VNODE_MARKER);
2087 mp->mnt_markercnt--;
2089 if (mp->mnt_holdcnt == 0 && mp->mnt_holdcntwaiters != 0)
2090 wakeup(&mp->mnt_holdcnt);
2095 __vfs_statfs(struct mount *mp, struct statfs *sbp, struct thread *td)
2099 error = mp->mnt_op->vfs_statfs(mp, &mp->mnt_stat, td);
2100 if (sbp != &mp->mnt_stat)
2101 *sbp = mp->mnt_stat;
2106 vfs_mountedfrom(struct mount *mp, const char *from)
2109 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
2110 strlcpy(mp->mnt_stat.f_mntfromname, from,
2111 sizeof mp->mnt_stat.f_mntfromname);
2115 * ---------------------------------------------------------------------
2116 * This is the api for building mount args and mounting filesystems from
2117 * inside the kernel.
2119 * The API works by accumulation of individual args. First error is
2122 * XXX: should be documented in new manpage kernel_mount(9)
2125 /* A memory allocation which must be freed when we are done */
2127 SLIST_ENTRY(mntaarg) next;
2130 /* The header for the mount arguments */
2135 SLIST_HEAD(, mntaarg) list;
2139 * Add a boolean argument.
2141 * flag is the boolean value.
2142 * name must start with "no".
2145 mount_argb(struct mntarg *ma, int flag, const char *name)
2148 KASSERT(name[0] == 'n' && name[1] == 'o',
2149 ("mount_argb(...,%s): name must start with 'no'", name));
2151 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
2155 * Add an argument printf style
2158 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
2161 struct mntaarg *maa;
2166 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2167 SLIST_INIT(&ma->list);
2172 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2174 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2175 ma->v[ma->len].iov_len = strlen(name) + 1;
2178 sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND);
2180 sbuf_vprintf(sb, fmt, ap);
2183 len = sbuf_len(sb) + 1;
2184 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2185 SLIST_INSERT_HEAD(&ma->list, maa, next);
2186 bcopy(sbuf_data(sb), maa + 1, len);
2189 ma->v[ma->len].iov_base = maa + 1;
2190 ma->v[ma->len].iov_len = len;
2197 * Add an argument which is a userland string.
2200 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2202 struct mntaarg *maa;
2208 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2209 SLIST_INIT(&ma->list);
2213 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2214 SLIST_INSERT_HEAD(&ma->list, maa, next);
2215 tbuf = (void *)(maa + 1);
2216 ma->error = copyinstr(val, tbuf, len, NULL);
2217 return (mount_arg(ma, name, tbuf, -1));
2223 * If length is -1, treat value as a C string.
2226 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2230 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2231 SLIST_INIT(&ma->list);
2236 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2238 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2239 ma->v[ma->len].iov_len = strlen(name) + 1;
2242 ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2244 ma->v[ma->len].iov_len = strlen(val) + 1;
2246 ma->v[ma->len].iov_len = len;
2252 * Free a mntarg structure
2255 free_mntarg(struct mntarg *ma)
2257 struct mntaarg *maa;
2259 while (!SLIST_EMPTY(&ma->list)) {
2260 maa = SLIST_FIRST(&ma->list);
2261 SLIST_REMOVE_HEAD(&ma->list, next);
2264 free(ma->v, M_MOUNT);
2269 * Mount a filesystem
2272 kernel_mount(struct mntarg *ma, int flags)
2277 KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2278 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
2279 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2281 auio.uio_iov = ma->v;
2282 auio.uio_iovcnt = ma->len;
2283 auio.uio_segflg = UIO_SYSSPACE;
2287 error = vfs_donmount(curthread, flags, &auio);
2293 * A printflike function to mount a filesystem.
2296 kernel_vmount(int flags, ...)
2298 struct mntarg *ma = NULL;
2304 va_start(ap, flags);
2306 cp = va_arg(ap, const char *);
2309 vp = va_arg(ap, const void *);
2310 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0));
2314 error = kernel_mount(ma, flags);