2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1999-2004 Poul-Henning Kamp
5 * Copyright (c) 1999 Michael Smith
6 * Copyright (c) 1989, 1993
7 * The Regents of the University of California. All rights reserved.
8 * (c) UNIX System Laboratories, Inc.
9 * All or some portions of this file are derived from material licensed
10 * to the University of California by American Telephone and Telegraph
11 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
12 * the permission of UNIX System Laboratories, Inc.
14 * Redistribution and use in source and binary forms, with or without
15 * modification, are permitted provided that the following conditions
17 * 1. Redistributions of source code must retain the above copyright
18 * notice, this list of conditions and the following disclaimer.
19 * 2. Redistributions in binary form must reproduce the above copyright
20 * notice, this list of conditions and the following disclaimer in the
21 * documentation and/or other materials provided with the distribution.
22 * 3. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
42 #include <sys/param.h>
45 #include <sys/devctl.h>
46 #include <sys/eventhandler.h>
47 #include <sys/fcntl.h>
49 #include <sys/kernel.h>
51 #include <sys/libkern.h>
52 #include <sys/malloc.h>
53 #include <sys/mount.h>
54 #include <sys/mutex.h>
55 #include <sys/namei.h>
58 #include <sys/filedesc.h>
59 #include <sys/reboot.h>
61 #include <sys/syscallsubr.h>
62 #include <sys/sysproto.h>
64 #include <sys/sysctl.h>
65 #include <sys/systm.h>
66 #include <sys/vnode.h>
69 #include <geom/geom.h>
71 #include <machine/stdarg.h>
73 #include <security/audit/audit.h>
74 #include <security/mac/mac_framework.h>
76 #define VFS_MOUNTARG_SIZE_MAX (1024 * 64)
78 static int vfs_domount(struct thread *td, const char *fstype, char *fspath,
79 uint64_t fsflags, struct vfsoptlist **optlist);
80 static void free_mntarg(struct mntarg *ma);
82 static int usermount = 0;
83 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
84 "Unprivileged users may mount and unmount file systems");
86 static bool default_autoro = false;
87 SYSCTL_BOOL(_vfs, OID_AUTO, default_autoro, CTLFLAG_RW, &default_autoro, 0,
88 "Retry failed r/w mount as r/o if no explicit ro/rw option is specified");
90 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
91 MALLOC_DEFINE(M_STATFS, "statfs", "statfs structure");
92 static uma_zone_t mount_zone;
94 /* List of mounted filesystems. */
95 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
97 /* For any iteration/modification of mountlist */
98 struct mtx_padalign __exclusive_cache_line mountlist_mtx;
99 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF);
101 EVENTHANDLER_LIST_DEFINE(vfs_mounted);
102 EVENTHANDLER_LIST_DEFINE(vfs_unmounted);
104 static void mount_devctl_event(const char *type, struct mount *mp, bool donew);
107 * Global opts, taken by all filesystems
109 static const char *global_opts[] = {
121 mount_init(void *mem, int size, int flags)
125 mp = (struct mount *)mem;
126 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
127 mtx_init(&mp->mnt_listmtx, "struct mount vlist mtx", NULL, MTX_DEF);
128 lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0);
129 mp->mnt_pcpu = uma_zalloc_pcpu(pcpu_zone_16, M_WAITOK | M_ZERO);
132 mp->mnt_rootvnode = NULL;
137 mount_fini(void *mem, int size)
141 mp = (struct mount *)mem;
142 uma_zfree_pcpu(pcpu_zone_16, mp->mnt_pcpu);
143 lockdestroy(&mp->mnt_explock);
144 mtx_destroy(&mp->mnt_listmtx);
145 mtx_destroy(&mp->mnt_mtx);
149 vfs_mount_init(void *dummy __unused)
152 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), NULL,
153 NULL, mount_init, mount_fini, UMA_ALIGN_CACHE, UMA_ZONE_NOFREE);
155 SYSINIT(vfs_mount, SI_SUB_VFS, SI_ORDER_ANY, vfs_mount_init, NULL);
158 * ---------------------------------------------------------------------
159 * Functions for building and sanitizing the mount options
162 /* Remove one mount option. */
164 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
167 TAILQ_REMOVE(opts, opt, link);
168 free(opt->name, M_MOUNT);
169 if (opt->value != NULL)
170 free(opt->value, M_MOUNT);
174 /* Release all resources related to the mount options. */
176 vfs_freeopts(struct vfsoptlist *opts)
180 while (!TAILQ_EMPTY(opts)) {
181 opt = TAILQ_FIRST(opts);
182 vfs_freeopt(opts, opt);
188 vfs_deleteopt(struct vfsoptlist *opts, const char *name)
190 struct vfsopt *opt, *temp;
194 TAILQ_FOREACH_SAFE(opt, opts, link, temp) {
195 if (strcmp(opt->name, name) == 0)
196 vfs_freeopt(opts, opt);
201 vfs_isopt_ro(const char *opt)
204 if (strcmp(opt, "ro") == 0 || strcmp(opt, "rdonly") == 0 ||
205 strcmp(opt, "norw") == 0)
211 vfs_isopt_rw(const char *opt)
214 if (strcmp(opt, "rw") == 0 || strcmp(opt, "noro") == 0)
220 * Check if options are equal (with or without the "no" prefix).
223 vfs_equalopts(const char *opt1, const char *opt2)
227 /* "opt" vs. "opt" or "noopt" vs. "noopt" */
228 if (strcmp(opt1, opt2) == 0)
230 /* "noopt" vs. "opt" */
231 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
233 /* "opt" vs. "noopt" */
234 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
236 while ((p = strchr(opt1, '.')) != NULL &&
237 !strncmp(opt1, opt2, ++p - opt1)) {
240 /* "foo.noopt" vs. "foo.opt" */
241 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
243 /* "foo.opt" vs. "foo.noopt" */
244 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
247 /* "ro" / "rdonly" / "norw" / "rw" / "noro" */
248 if ((vfs_isopt_ro(opt1) || vfs_isopt_rw(opt1)) &&
249 (vfs_isopt_ro(opt2) || vfs_isopt_rw(opt2)))
255 * If a mount option is specified several times,
256 * (with or without the "no" prefix) only keep
257 * the last occurrence of it.
260 vfs_sanitizeopts(struct vfsoptlist *opts)
262 struct vfsopt *opt, *opt2, *tmp;
264 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
265 opt2 = TAILQ_PREV(opt, vfsoptlist, link);
266 while (opt2 != NULL) {
267 if (vfs_equalopts(opt->name, opt2->name)) {
268 tmp = TAILQ_PREV(opt2, vfsoptlist, link);
269 vfs_freeopt(opts, opt2);
272 opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
279 * Build a linked list of mount options from a struct uio.
282 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
284 struct vfsoptlist *opts;
286 size_t memused, namelen, optlen;
287 unsigned int i, iovcnt;
290 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
293 iovcnt = auio->uio_iovcnt;
294 for (i = 0; i < iovcnt; i += 2) {
295 namelen = auio->uio_iov[i].iov_len;
296 optlen = auio->uio_iov[i + 1].iov_len;
297 memused += sizeof(struct vfsopt) + optlen + namelen;
299 * Avoid consuming too much memory, and attempts to overflow
302 if (memused > VFS_MOUNTARG_SIZE_MAX ||
303 optlen > VFS_MOUNTARG_SIZE_MAX ||
304 namelen > VFS_MOUNTARG_SIZE_MAX) {
309 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
310 opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
317 * Do this early, so jumps to "bad" will free the current
320 TAILQ_INSERT_TAIL(opts, opt, link);
322 if (auio->uio_segflg == UIO_SYSSPACE) {
323 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
325 error = copyin(auio->uio_iov[i].iov_base, opt->name,
330 /* Ensure names are null-terminated strings. */
331 if (namelen == 0 || opt->name[namelen - 1] != '\0') {
337 opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
338 if (auio->uio_segflg == UIO_SYSSPACE) {
339 bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
342 error = copyin(auio->uio_iov[i + 1].iov_base,
349 vfs_sanitizeopts(opts);
358 * Merge the old mount options with the new ones passed
359 * in the MNT_UPDATE case.
361 * XXX: This function will keep a "nofoo" option in the new
362 * options. E.g, if the option's canonical name is "foo",
363 * "nofoo" ends up in the mount point's active options.
366 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *oldopts)
368 struct vfsopt *opt, *new;
370 TAILQ_FOREACH(opt, oldopts, link) {
371 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
372 new->name = strdup(opt->name, M_MOUNT);
374 new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
375 bcopy(opt->value, new->value, opt->len);
379 new->seen = opt->seen;
380 TAILQ_INSERT_HEAD(toopts, new, link);
382 vfs_sanitizeopts(toopts);
386 * Mount a filesystem.
388 #ifndef _SYS_SYSPROTO_H_
396 sys_nmount(struct thread *td, struct nmount_args *uap)
404 * Mount flags are now 64-bits. On 32-bit archtectures only
405 * 32-bits are passed in, but from here on everything handles
406 * 64-bit flags correctly.
410 AUDIT_ARG_FFLAGS(flags);
411 CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__,
412 uap->iovp, uap->iovcnt, flags);
415 * Filter out MNT_ROOTFS. We do not want clients of nmount() in
416 * userspace to set this flag, but we must filter it out if we want
417 * MNT_UPDATE on the root file system to work.
418 * MNT_ROOTFS should only be set by the kernel when mounting its
421 flags &= ~MNT_ROOTFS;
423 iovcnt = uap->iovcnt;
425 * Check that we have an even number of iovec's
426 * and that we have at least two options.
428 if ((iovcnt & 1) || (iovcnt < 4)) {
429 CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__,
434 error = copyinuio(uap->iovp, iovcnt, &auio);
436 CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno",
440 error = vfs_donmount(td, flags, auio);
447 * ---------------------------------------------------------------------
448 * Various utility functions
452 * Get a reference on a mount point from a vnode.
454 * The vnode is allowed to be passed unlocked and race against dooming. Note in
455 * such case there are no guarantees the referenced mount point will still be
456 * associated with it after the function returns.
459 vfs_ref_from_vp(struct vnode *vp)
462 struct mount_pcpu *mpcpu;
464 mp = atomic_load_ptr(&vp->v_mount);
465 if (__predict_false(mp == NULL)) {
468 if (vfs_op_thread_enter(mp, mpcpu)) {
469 if (__predict_true(mp == vp->v_mount)) {
470 vfs_mp_count_add_pcpu(mpcpu, ref, 1);
471 vfs_op_thread_exit(mp, mpcpu);
473 vfs_op_thread_exit(mp, mpcpu);
478 if (mp == vp->v_mount) {
490 vfs_ref(struct mount *mp)
492 struct mount_pcpu *mpcpu;
494 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
495 if (vfs_op_thread_enter(mp, mpcpu)) {
496 vfs_mp_count_add_pcpu(mpcpu, ref, 1);
497 vfs_op_thread_exit(mp, mpcpu);
507 vfs_rel(struct mount *mp)
509 struct mount_pcpu *mpcpu;
511 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
512 if (vfs_op_thread_enter(mp, mpcpu)) {
513 vfs_mp_count_sub_pcpu(mpcpu, ref, 1);
514 vfs_op_thread_exit(mp, mpcpu);
524 * Allocate and initialize the mount point struct.
527 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath,
532 mp = uma_zalloc(mount_zone, M_WAITOK);
533 bzero(&mp->mnt_startzero,
534 __rangeof(struct mount, mnt_startzero, mnt_endzero));
535 mp->mnt_kern_flag = 0;
537 mp->mnt_rootvnode = NULL;
538 mp->mnt_vnodecovered = NULL;
541 TAILQ_INIT(&mp->mnt_nvnodelist);
542 mp->mnt_nvnodelistsize = 0;
543 TAILQ_INIT(&mp->mnt_lazyvnodelist);
544 mp->mnt_lazyvnodelistsize = 0;
545 MPPASS(mp->mnt_ref == 0 && mp->mnt_lockref == 0 &&
546 mp->mnt_writeopcount == 0, mp);
547 MPASSERT(mp->mnt_vfs_ops == 1, mp,
548 ("vfs_ops should be 1 but %d found", mp->mnt_vfs_ops));
549 (void) vfs_busy(mp, MBF_NOWAIT);
550 atomic_add_acq_int(&vfsp->vfc_refcount, 1);
551 mp->mnt_op = vfsp->vfc_vfsops;
553 mp->mnt_stat.f_type = vfsp->vfc_typenum;
555 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
556 mp->mnt_vnodecovered = vp;
557 mp->mnt_cred = crdup(cred);
558 mp->mnt_stat.f_owner = cred->cr_uid;
559 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
560 mp->mnt_iosize_max = DFLTPHYS;
563 mac_mount_create(cred, mp);
565 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
566 TAILQ_INIT(&mp->mnt_uppers);
571 * Destroy the mount struct previously allocated by vfs_mount_alloc().
574 vfs_mount_destroy(struct mount *mp)
577 MPPASS(mp->mnt_vfs_ops != 0, mp);
579 vfs_assert_mount_counters(mp);
582 mp->mnt_kern_flag |= MNTK_REFEXPIRE;
583 if (mp->mnt_kern_flag & MNTK_MWAIT) {
584 mp->mnt_kern_flag &= ~MNTK_MWAIT;
588 msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0);
589 KASSERT(mp->mnt_ref == 0,
590 ("%s: invalid refcount in the drain path @ %s:%d", __func__,
591 __FILE__, __LINE__));
592 MPPASS(mp->mnt_writeopcount == 0, mp);
593 MPPASS(mp->mnt_secondary_writes == 0, mp);
594 atomic_subtract_rel_int(&mp->mnt_vfc->vfc_refcount, 1);
595 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
598 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
599 vn_printf(vp, "dangling vnode ");
600 panic("unmount: dangling vnode");
602 KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers"));
603 MPPASS(mp->mnt_nvnodelistsize == 0, mp);
604 MPPASS(mp->mnt_lazyvnodelistsize == 0, mp);
605 MPPASS(mp->mnt_lockref == 0, mp);
608 MPASSERT(mp->mnt_vfs_ops == 1, mp,
609 ("vfs_ops should be 1 but %d found", mp->mnt_vfs_ops));
611 MPASSERT(mp->mnt_rootvnode == NULL, mp,
612 ("mount point still has a root vnode %p", mp->mnt_rootvnode));
614 if (mp->mnt_vnodecovered != NULL)
615 vrele(mp->mnt_vnodecovered);
617 mac_mount_destroy(mp);
619 if (mp->mnt_opt != NULL)
620 vfs_freeopts(mp->mnt_opt);
621 crfree(mp->mnt_cred);
622 uma_zfree(mount_zone, mp);
626 vfs_should_downgrade_to_ro_mount(uint64_t fsflags, int error)
628 /* This is an upgrade of an exisiting mount. */
629 if ((fsflags & MNT_UPDATE) != 0)
631 /* This is already an R/O mount. */
632 if ((fsflags & MNT_RDONLY) != 0)
636 case ENODEV: /* generic, geom, ... */
637 case EACCES: /* cam/scsi, ... */
638 case EROFS: /* md, mmcsd, ... */
640 * These errors can be returned by the storage layer to signal
641 * that the media is read-only. No harm in the R/O mount
642 * attempt if the error was returned for some other reason.
651 vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions)
653 struct vfsoptlist *optlist;
654 struct vfsopt *opt, *tmp_opt;
655 char *fstype, *fspath, *errmsg;
656 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
659 errmsg = fspath = NULL;
660 errmsg_len = fspathlen = 0;
662 autoro = default_autoro;
664 error = vfs_buildopts(fsoptions, &optlist);
668 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
669 errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
672 * We need these two options before the others,
673 * and they are mandatory for any filesystem.
674 * Ensure they are NUL terminated as well.
677 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
678 if (error || fstypelen <= 0 || fstype[fstypelen - 1] != '\0') {
681 strncpy(errmsg, "Invalid fstype", errmsg_len);
685 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
686 if (error || fspathlen <= 0 || fspath[fspathlen - 1] != '\0') {
689 strncpy(errmsg, "Invalid fspath", errmsg_len);
694 * We need to see if we have the "update" option
695 * before we call vfs_domount(), since vfs_domount() has special
696 * logic based on MNT_UPDATE. This is very important
697 * when we want to update the root filesystem.
699 TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) {
702 if (strcmp(opt->name, "update") == 0) {
703 fsflags |= MNT_UPDATE;
706 else if (strcmp(opt->name, "async") == 0)
707 fsflags |= MNT_ASYNC;
708 else if (strcmp(opt->name, "force") == 0) {
709 fsflags |= MNT_FORCE;
712 else if (strcmp(opt->name, "reload") == 0) {
713 fsflags |= MNT_RELOAD;
716 else if (strcmp(opt->name, "multilabel") == 0)
717 fsflags |= MNT_MULTILABEL;
718 else if (strcmp(opt->name, "noasync") == 0)
719 fsflags &= ~MNT_ASYNC;
720 else if (strcmp(opt->name, "noatime") == 0)
721 fsflags |= MNT_NOATIME;
722 else if (strcmp(opt->name, "atime") == 0) {
723 free(opt->name, M_MOUNT);
724 opt->name = strdup("nonoatime", M_MOUNT);
726 else if (strcmp(opt->name, "noclusterr") == 0)
727 fsflags |= MNT_NOCLUSTERR;
728 else if (strcmp(opt->name, "clusterr") == 0) {
729 free(opt->name, M_MOUNT);
730 opt->name = strdup("nonoclusterr", M_MOUNT);
732 else if (strcmp(opt->name, "noclusterw") == 0)
733 fsflags |= MNT_NOCLUSTERW;
734 else if (strcmp(opt->name, "clusterw") == 0) {
735 free(opt->name, M_MOUNT);
736 opt->name = strdup("nonoclusterw", M_MOUNT);
738 else if (strcmp(opt->name, "noexec") == 0)
739 fsflags |= MNT_NOEXEC;
740 else if (strcmp(opt->name, "exec") == 0) {
741 free(opt->name, M_MOUNT);
742 opt->name = strdup("nonoexec", M_MOUNT);
744 else if (strcmp(opt->name, "nosuid") == 0)
745 fsflags |= MNT_NOSUID;
746 else if (strcmp(opt->name, "suid") == 0) {
747 free(opt->name, M_MOUNT);
748 opt->name = strdup("nonosuid", M_MOUNT);
750 else if (strcmp(opt->name, "nosymfollow") == 0)
751 fsflags |= MNT_NOSYMFOLLOW;
752 else if (strcmp(opt->name, "symfollow") == 0) {
753 free(opt->name, M_MOUNT);
754 opt->name = strdup("nonosymfollow", M_MOUNT);
756 else if (strcmp(opt->name, "noro") == 0) {
757 fsflags &= ~MNT_RDONLY;
760 else if (strcmp(opt->name, "rw") == 0) {
761 fsflags &= ~MNT_RDONLY;
764 else if (strcmp(opt->name, "ro") == 0) {
765 fsflags |= MNT_RDONLY;
768 else if (strcmp(opt->name, "rdonly") == 0) {
769 free(opt->name, M_MOUNT);
770 opt->name = strdup("ro", M_MOUNT);
771 fsflags |= MNT_RDONLY;
774 else if (strcmp(opt->name, "autoro") == 0) {
778 else if (strcmp(opt->name, "suiddir") == 0)
779 fsflags |= MNT_SUIDDIR;
780 else if (strcmp(opt->name, "sync") == 0)
781 fsflags |= MNT_SYNCHRONOUS;
782 else if (strcmp(opt->name, "union") == 0)
783 fsflags |= MNT_UNION;
784 else if (strcmp(opt->name, "export") == 0)
785 fsflags |= MNT_EXPORTED;
786 else if (strcmp(opt->name, "automounted") == 0) {
787 fsflags |= MNT_AUTOMOUNTED;
789 } else if (strcmp(opt->name, "nocover") == 0) {
790 fsflags |= MNT_NOCOVER;
792 } else if (strcmp(opt->name, "cover") == 0) {
793 fsflags &= ~MNT_NOCOVER;
795 } else if (strcmp(opt->name, "emptydir") == 0) {
796 fsflags |= MNT_EMPTYDIR;
798 } else if (strcmp(opt->name, "noemptydir") == 0) {
799 fsflags &= ~MNT_EMPTYDIR;
803 vfs_freeopt(optlist, opt);
807 * Be ultra-paranoid about making sure the type and fspath
808 * variables will fit in our mp buffers, including the
811 if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) {
812 error = ENAMETOOLONG;
816 error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
817 if (error == ENOENT) {
820 strncpy(errmsg, "Invalid fstype", errmsg_len);
825 * See if we can mount in the read-only mode if the error code suggests
826 * that it could be possible and the mount options allow for that.
827 * Never try it if "[no]{ro|rw}" has been explicitly requested and not
828 * overridden by "autoro".
830 if (autoro && vfs_should_downgrade_to_ro_mount(fsflags, error)) {
831 printf("%s: R/W mount failed, possibly R/O media,"
832 " trying R/O mount\n", __func__);
833 fsflags |= MNT_RDONLY;
834 error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
837 /* copyout the errmsg */
838 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
839 && errmsg_len > 0 && errmsg != NULL) {
840 if (fsoptions->uio_segflg == UIO_SYSSPACE) {
842 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
843 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
846 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
847 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
852 vfs_freeopts(optlist);
859 #ifndef _SYS_SYSPROTO_H_
869 sys_mount(struct thread *td, struct mount_args *uap)
872 struct vfsconf *vfsp = NULL;
873 struct mntarg *ma = NULL;
878 * Mount flags are now 64-bits. On 32-bit architectures only
879 * 32-bits are passed in, but from here on everything handles
880 * 64-bit flags correctly.
884 AUDIT_ARG_FFLAGS(flags);
887 * Filter out MNT_ROOTFS. We do not want clients of mount() in
888 * userspace to set this flag, but we must filter it out if we want
889 * MNT_UPDATE on the root file system to work.
890 * MNT_ROOTFS should only be set by the kernel when mounting its
893 flags &= ~MNT_ROOTFS;
895 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
896 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
898 free(fstype, M_TEMP);
902 AUDIT_ARG_TEXT(fstype);
903 vfsp = vfs_byname_kld(fstype, td, &error);
904 free(fstype, M_TEMP);
907 if (((vfsp->vfc_flags & VFCF_SBDRY) != 0 &&
908 vfsp->vfc_vfsops_sd->vfs_cmount == NULL) ||
909 ((vfsp->vfc_flags & VFCF_SBDRY) == 0 &&
910 vfsp->vfc_vfsops->vfs_cmount == NULL))
913 ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN);
914 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
915 ma = mount_argb(ma, flags & MNT_RDONLY, "noro");
916 ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid");
917 ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec");
919 if ((vfsp->vfc_flags & VFCF_SBDRY) != 0)
920 return (vfsp->vfc_vfsops_sd->vfs_cmount(ma, uap->data, flags));
921 return (vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags));
925 * vfs_domount_first(): first file system mount (not update)
929 struct thread *td, /* Calling thread. */
930 struct vfsconf *vfsp, /* File system type. */
931 char *fspath, /* Mount path. */
932 struct vnode *vp, /* Vnode to be covered. */
933 uint64_t fsflags, /* Flags common to all filesystems. */
934 struct vfsoptlist **optlist /* Options local to the filesystem. */
939 struct vnode *newdp, *rootvp;
943 ASSERT_VOP_ELOCKED(vp, __func__);
944 KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here"));
947 * If the jail of the calling thread lacks permission for this type of
948 * file system, or is trying to cover its own root, deny immediately.
950 if (jailed(td->td_ucred) && (!prison_allow(td->td_ucred,
951 vfsp->vfc_prison_flag) || vp == td->td_ucred->cr_prison->pr_root)) {
957 * If the user is not root, ensure that they own the directory
958 * onto which we are attempting to mount.
960 error = VOP_GETATTR(vp, &va, td->td_ucred);
961 if (error == 0 && va.va_uid != td->td_ucred->cr_uid)
962 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN);
964 error = vinvalbuf(vp, V_SAVE, 0, 0);
965 if (error == 0 && vp->v_type != VDIR)
967 if (error == 0 && (fsflags & MNT_EMPTYDIR) != 0)
968 error = vfs_emptydir(vp);
971 if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL)
972 vp->v_iflag |= VI_MOUNT;
981 vn_seqc_write_begin(vp);
984 /* Allocate and initialize the filesystem. */
985 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred);
986 /* XXXMAC: pass to vfs_mount_alloc? */
987 mp->mnt_optnew = *optlist;
988 /* Set the mount level flags. */
989 mp->mnt_flag = (fsflags & (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY));
992 * Mount the filesystem.
993 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
994 * get. No freeing of cn_pnbuf.
998 if ((error = VFS_MOUNT(mp)) != 0 ||
999 (error1 = VFS_STATFS(mp, &mp->mnt_stat)) != 0 ||
1000 (error1 = VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) != 0) {
1004 rootvp = vfs_cache_root_clear(mp);
1005 if (rootvp != NULL) {
1009 (void)vn_start_write(NULL, &mp, V_WAIT);
1011 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_UNMOUNTF;
1014 error = VFS_UNMOUNT(mp, 0);
1015 vn_finished_write(mp);
1018 "failed post-mount (%d): rollback unmount returned %d\n",
1025 mp->mnt_vnodecovered = NULL;
1027 /* XXXKIB wait for mnt_lockref drain? */
1028 vfs_mount_destroy(mp);
1031 vp->v_iflag &= ~VI_MOUNT;
1033 if (rootvp != NULL) {
1034 vn_seqc_write_end(rootvp);
1037 vn_seqc_write_end(vp);
1041 vn_seqc_write_begin(newdp);
1044 if (mp->mnt_opt != NULL)
1045 vfs_freeopts(mp->mnt_opt);
1046 mp->mnt_opt = mp->mnt_optnew;
1050 * Prevent external consumers of mount options from reading mnt_optnew.
1052 mp->mnt_optnew = NULL;
1055 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1056 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1057 mp->mnt_kern_flag |= MNTK_ASYNC;
1059 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1062 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1065 vp->v_iflag &= ~VI_MOUNT;
1066 vn_irflag_set_locked(vp, VIRF_MOUNTPOINT);
1067 vp->v_mountedhere = mp;
1069 /* Place the new filesystem at the end of the mount list. */
1070 mtx_lock(&mountlist_mtx);
1071 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1072 mtx_unlock(&mountlist_mtx);
1073 vfs_event_signal(NULL, VQ_MOUNT, 0);
1074 vn_lock(newdp, LK_EXCLUSIVE | LK_RETRY);
1076 EVENTHANDLER_DIRECT_INVOKE(vfs_mounted, mp, newdp, td);
1078 mount_devctl_event("MOUNT", mp, false);
1079 mountcheckdirs(vp, newdp);
1080 vn_seqc_write_end(vp);
1081 vn_seqc_write_end(newdp);
1083 if ((mp->mnt_flag & MNT_RDONLY) == 0)
1084 vfs_allocate_syncvnode(mp);
1091 * vfs_domount_update(): update of mounted file system
1095 struct thread *td, /* Calling thread. */
1096 struct vnode *vp, /* Mount point vnode. */
1097 uint64_t fsflags, /* Flags common to all filesystems. */
1098 struct vfsoptlist **optlist /* Options local to the filesystem. */
1101 struct export_args export;
1102 struct o2export_args o2export;
1103 struct vnode *rootvp;
1106 int error, export_error, i, len;
1110 ASSERT_VOP_ELOCKED(vp, __func__);
1111 KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here"));
1114 if ((vp->v_vflag & VV_ROOT) == 0) {
1115 if (vfs_copyopt(*optlist, "export", &export, sizeof(export))
1125 * We only allow the filesystem to be reloaded if it
1126 * is currently mounted read-only.
1128 flag = mp->mnt_flag;
1129 if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) {
1131 return (EOPNOTSUPP); /* Needs translation */
1134 * Only privileged root, or (if MNT_USER is set) the user that
1135 * did the original mount is permitted to update it.
1137 error = vfs_suser(mp, td);
1142 if (vfs_busy(mp, MBF_NOWAIT)) {
1147 if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) {
1153 vp->v_iflag |= VI_MOUNT;
1158 vn_seqc_write_begin(vp);
1162 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1167 mp->mnt_flag &= ~MNT_UPDATEMASK;
1168 mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE |
1169 MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY);
1170 if ((mp->mnt_flag & MNT_ASYNC) == 0)
1171 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1172 rootvp = vfs_cache_root_clear(mp);
1174 mp->mnt_optnew = *optlist;
1175 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
1178 * Mount the filesystem.
1179 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1180 * get. No freeing of cn_pnbuf.
1182 error = VFS_MOUNT(mp);
1185 /* Process the export option. */
1186 if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp,
1188 /* Assume that there is only 1 ABI for each length. */
1190 case (sizeof(struct oexport_args)):
1191 bzero(&o2export, sizeof(o2export));
1193 case (sizeof(o2export)):
1194 bcopy(bufp, &o2export, len);
1195 export.ex_flags = (uint64_t)o2export.ex_flags;
1196 export.ex_root = o2export.ex_root;
1197 export.ex_uid = o2export.ex_anon.cr_uid;
1198 export.ex_groups = NULL;
1199 export.ex_ngroups = o2export.ex_anon.cr_ngroups;
1200 if (export.ex_ngroups > 0) {
1201 if (export.ex_ngroups <= XU_NGROUPS) {
1202 export.ex_groups = malloc(
1203 export.ex_ngroups * sizeof(gid_t),
1205 for (i = 0; i < export.ex_ngroups; i++)
1206 export.ex_groups[i] =
1207 o2export.ex_anon.cr_groups[i];
1209 export_error = EINVAL;
1210 } else if (export.ex_ngroups < 0)
1211 export_error = EINVAL;
1212 export.ex_addr = o2export.ex_addr;
1213 export.ex_addrlen = o2export.ex_addrlen;
1214 export.ex_mask = o2export.ex_mask;
1215 export.ex_masklen = o2export.ex_masklen;
1216 export.ex_indexfile = o2export.ex_indexfile;
1217 export.ex_numsecflavors = o2export.ex_numsecflavors;
1218 if (export.ex_numsecflavors < MAXSECFLAVORS) {
1219 for (i = 0; i < export.ex_numsecflavors; i++)
1220 export.ex_secflavors[i] =
1221 o2export.ex_secflavors[i];
1223 export_error = EINVAL;
1224 if (export_error == 0)
1225 export_error = vfs_export(mp, &export);
1226 free(export.ex_groups, M_TEMP);
1228 case (sizeof(export)):
1229 bcopy(bufp, &export, len);
1231 if (export.ex_ngroups > 0) {
1232 if (export.ex_ngroups <= NGROUPS_MAX) {
1233 grps = malloc(export.ex_ngroups *
1234 sizeof(gid_t), M_TEMP, M_WAITOK);
1235 export_error = copyin(export.ex_groups,
1236 grps, export.ex_ngroups *
1238 if (export_error == 0)
1239 export.ex_groups = grps;
1241 export_error = EINVAL;
1242 } else if (export.ex_ngroups == 0)
1243 export.ex_groups = NULL;
1245 export_error = EINVAL;
1246 if (export_error == 0)
1247 export_error = vfs_export(mp, &export);
1251 export_error = EINVAL;
1258 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE |
1262 * If we fail, restore old mount flags. MNT_QUOTA is special,
1263 * because it is not part of MNT_UPDATEMASK, but it could have
1264 * changed in the meantime if quotactl(2) was called.
1265 * All in all we want current value of MNT_QUOTA, not the old
1268 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA);
1270 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1271 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1272 mp->mnt_kern_flag |= MNTK_ASYNC;
1274 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1280 mount_devctl_event("REMOUNT", mp, true);
1281 if (mp->mnt_opt != NULL)
1282 vfs_freeopts(mp->mnt_opt);
1283 mp->mnt_opt = mp->mnt_optnew;
1285 (void)VFS_STATFS(mp, &mp->mnt_stat);
1287 * Prevent external consumers of mount options from reading
1290 mp->mnt_optnew = NULL;
1292 if ((mp->mnt_flag & MNT_RDONLY) == 0)
1293 vfs_allocate_syncvnode(mp);
1295 vfs_deallocate_syncvnode(mp);
1298 if (rootvp != NULL) {
1299 vn_seqc_write_end(rootvp);
1302 vn_seqc_write_end(vp);
1305 vp->v_iflag &= ~VI_MOUNT;
1308 return (error != 0 ? error : export_error);
1312 * vfs_domount(): actually attempt a filesystem mount.
1316 struct thread *td, /* Calling thread. */
1317 const char *fstype, /* Filesystem type. */
1318 char *fspath, /* Mount path. */
1319 uint64_t fsflags, /* Flags common to all filesystems. */
1320 struct vfsoptlist **optlist /* Options local to the filesystem. */
1323 struct vfsconf *vfsp;
1324 struct nameidata nd;
1330 * Be ultra-paranoid about making sure the type and fspath
1331 * variables will fit in our mp buffers, including the
1334 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
1335 return (ENAMETOOLONG);
1337 if (jailed(td->td_ucred) || usermount == 0) {
1338 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
1343 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
1345 if (fsflags & MNT_EXPORTED) {
1346 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
1350 if (fsflags & MNT_SUIDDIR) {
1351 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
1356 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
1358 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
1359 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
1360 fsflags |= MNT_NOSUID | MNT_USER;
1363 /* Load KLDs before we lock the covered vnode to avoid reversals. */
1365 if ((fsflags & MNT_UPDATE) == 0) {
1366 /* Don't try to load KLDs if we're mounting the root. */
1367 if (fsflags & MNT_ROOTFS) {
1368 if ((vfsp = vfs_byname(fstype)) == NULL)
1371 if ((vfsp = vfs_byname_kld(fstype, td, &error)) == NULL)
1377 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE.
1379 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1380 UIO_SYSSPACE, fspath, td);
1384 NDFREE(&nd, NDF_ONLY_PNBUF);
1386 if ((fsflags & MNT_UPDATE) == 0) {
1387 if ((vp->v_vflag & VV_ROOT) != 0 &&
1388 (fsflags & MNT_NOCOVER) != 0) {
1392 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1393 strcpy(pathbuf, fspath);
1394 error = vn_path_to_global_path(td, vp, pathbuf, MNAMELEN);
1396 error = vfs_domount_first(td, vfsp, pathbuf, vp,
1399 free(pathbuf, M_TEMP);
1401 error = vfs_domount_update(td, vp, fsflags, optlist);
1407 * Unmount a filesystem.
1409 * Note: unmount takes a path to the vnode mounted on as argument, not
1410 * special file (as before).
1412 #ifndef _SYS_SYSPROTO_H_
1413 struct unmount_args {
1420 sys_unmount(struct thread *td, struct unmount_args *uap)
1423 return (kern_unmount(td, uap->path, uap->flags));
1427 kern_unmount(struct thread *td, const char *path, int flags)
1429 struct nameidata nd;
1432 int error, id0, id1;
1434 AUDIT_ARG_VALUE(flags);
1435 if (jailed(td->td_ucred) || usermount == 0) {
1436 error = priv_check(td, PRIV_VFS_UNMOUNT);
1441 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1442 error = copyinstr(path, pathbuf, MNAMELEN, NULL);
1444 free(pathbuf, M_TEMP);
1447 if (flags & MNT_BYFSID) {
1448 AUDIT_ARG_TEXT(pathbuf);
1449 /* Decode the filesystem ID. */
1450 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1451 free(pathbuf, M_TEMP);
1455 mtx_lock(&mountlist_mtx);
1456 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1457 if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1458 mp->mnt_stat.f_fsid.val[1] == id1) {
1463 mtx_unlock(&mountlist_mtx);
1466 * Try to find global path for path argument.
1468 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1469 UIO_SYSSPACE, pathbuf, td);
1470 if (namei(&nd) == 0) {
1471 NDFREE(&nd, NDF_ONLY_PNBUF);
1472 error = vn_path_to_global_path(td, nd.ni_vp, pathbuf,
1477 mtx_lock(&mountlist_mtx);
1478 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1479 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) {
1484 mtx_unlock(&mountlist_mtx);
1486 free(pathbuf, M_TEMP);
1489 * Previously we returned ENOENT for a nonexistent path and
1490 * EINVAL for a non-mountpoint. We cannot tell these apart
1491 * now, so in the !MNT_BYFSID case return the more likely
1492 * EINVAL for compatibility.
1494 return ((flags & MNT_BYFSID) ? ENOENT : EINVAL);
1498 * Don't allow unmounting the root filesystem.
1500 if (mp->mnt_flag & MNT_ROOTFS) {
1504 error = dounmount(mp, flags, td);
1509 * Return error if any of the vnodes, ignoring the root vnode
1510 * and the syncer vnode, have non-zero usecount.
1512 * This function is purely advisory - it can return false positives
1516 vfs_check_usecounts(struct mount *mp)
1518 struct vnode *vp, *mvp;
1520 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1521 if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON &&
1522 vp->v_usecount != 0) {
1524 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1534 dounmount_cleanup(struct mount *mp, struct vnode *coveredvp, int mntkflags)
1537 mtx_assert(MNT_MTX(mp), MA_OWNED);
1538 mp->mnt_kern_flag &= ~mntkflags;
1539 if ((mp->mnt_kern_flag & MNTK_MWAIT) != 0) {
1540 mp->mnt_kern_flag &= ~MNTK_MWAIT;
1543 vfs_op_exit_locked(mp);
1545 if (coveredvp != NULL) {
1546 VOP_UNLOCK(coveredvp);
1549 vn_finished_write(mp);
1553 * There are various reference counters associated with the mount point.
1554 * Normally it is permitted to modify them without taking the mnt ilock,
1555 * but this behavior can be temporarily disabled if stable value is needed
1556 * or callers are expected to block (e.g. to not allow new users during
1560 vfs_op_enter(struct mount *mp)
1562 struct mount_pcpu *mpcpu;
1567 if (mp->mnt_vfs_ops > 1) {
1571 vfs_op_barrier_wait(mp);
1573 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1575 mp->mnt_ref += mpcpu->mntp_ref;
1576 mpcpu->mntp_ref = 0;
1578 mp->mnt_lockref += mpcpu->mntp_lockref;
1579 mpcpu->mntp_lockref = 0;
1581 mp->mnt_writeopcount += mpcpu->mntp_writeopcount;
1582 mpcpu->mntp_writeopcount = 0;
1584 MPASSERT(mp->mnt_ref > 0 && mp->mnt_lockref >= 0 &&
1585 mp->mnt_writeopcount >= 0, mp,
1586 ("invalid count(s): ref %d lockref %d writeopcount %d",
1587 mp->mnt_ref, mp->mnt_lockref, mp->mnt_writeopcount));
1589 vfs_assert_mount_counters(mp);
1593 vfs_op_exit_locked(struct mount *mp)
1596 mtx_assert(MNT_MTX(mp), MA_OWNED);
1598 MPASSERT(mp->mnt_vfs_ops > 0, mp,
1599 ("invalid vfs_ops count %d", mp->mnt_vfs_ops));
1600 MPASSERT(mp->mnt_vfs_ops > 1 ||
1601 (mp->mnt_kern_flag & (MNTK_UNMOUNT | MNTK_SUSPEND)) == 0, mp,
1602 ("vfs_ops too low %d in unmount or suspend", mp->mnt_vfs_ops));
1607 vfs_op_exit(struct mount *mp)
1611 vfs_op_exit_locked(mp);
1615 struct vfs_op_barrier_ipi {
1617 struct smp_rendezvous_cpus_retry_arg srcra;
1621 vfs_op_action_func(void *arg)
1623 struct vfs_op_barrier_ipi *vfsopipi;
1626 vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1629 if (!vfs_op_thread_entered(mp))
1630 smp_rendezvous_cpus_done(arg);
1634 vfs_op_wait_func(void *arg, int cpu)
1636 struct vfs_op_barrier_ipi *vfsopipi;
1638 struct mount_pcpu *mpcpu;
1640 vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1643 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1644 while (atomic_load_int(&mpcpu->mntp_thread_in_ops))
1649 vfs_op_barrier_wait(struct mount *mp)
1651 struct vfs_op_barrier_ipi vfsopipi;
1655 smp_rendezvous_cpus_retry(all_cpus,
1656 smp_no_rendezvous_barrier,
1658 smp_no_rendezvous_barrier,
1665 vfs_assert_mount_counters(struct mount *mp)
1667 struct mount_pcpu *mpcpu;
1670 if (mp->mnt_vfs_ops == 0)
1674 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1675 if (mpcpu->mntp_ref != 0 ||
1676 mpcpu->mntp_lockref != 0 ||
1677 mpcpu->mntp_writeopcount != 0)
1678 vfs_dump_mount_counters(mp);
1683 vfs_dump_mount_counters(struct mount *mp)
1685 struct mount_pcpu *mpcpu;
1686 int ref, lockref, writeopcount;
1689 printf("%s: mp %p vfs_ops %d\n", __func__, mp, mp->mnt_vfs_ops);
1694 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1695 printf("%d ", mpcpu->mntp_ref);
1696 ref += mpcpu->mntp_ref;
1699 printf(" lockref : ");
1700 lockref = mp->mnt_lockref;
1702 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1703 printf("%d ", mpcpu->mntp_lockref);
1704 lockref += mpcpu->mntp_lockref;
1707 printf("writeopcount: ");
1708 writeopcount = mp->mnt_writeopcount;
1710 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1711 printf("%d ", mpcpu->mntp_writeopcount);
1712 writeopcount += mpcpu->mntp_writeopcount;
1716 printf("counter struct total\n");
1717 printf("ref %-5d %-5d\n", mp->mnt_ref, ref);
1718 printf("lockref %-5d %-5d\n", mp->mnt_lockref, lockref);
1719 printf("writeopcount %-5d %-5d\n", mp->mnt_writeopcount, writeopcount);
1721 panic("invalid counts on struct mount");
1726 vfs_mount_fetch_counter(struct mount *mp, enum mount_counter which)
1728 struct mount_pcpu *mpcpu;
1735 case MNT_COUNT_LOCKREF:
1736 sum = mp->mnt_lockref;
1738 case MNT_COUNT_WRITEOPCOUNT:
1739 sum = mp->mnt_writeopcount;
1744 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1747 sum += mpcpu->mntp_ref;
1749 case MNT_COUNT_LOCKREF:
1750 sum += mpcpu->mntp_lockref;
1752 case MNT_COUNT_WRITEOPCOUNT:
1753 sum += mpcpu->mntp_writeopcount;
1761 * Do the actual filesystem unmount.
1764 dounmount(struct mount *mp, int flags, struct thread *td)
1766 struct vnode *coveredvp, *rootvp;
1768 uint64_t async_flag;
1771 if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
1772 mnt_gen_r = mp->mnt_gen;
1775 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
1777 * Check for mp being unmounted while waiting for the
1778 * covered vnode lock.
1780 if (coveredvp->v_mountedhere != mp ||
1781 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
1782 VOP_UNLOCK(coveredvp);
1790 * Only privileged root, or (if MNT_USER is set) the user that did the
1791 * original mount is permitted to unmount this filesystem.
1793 error = vfs_suser(mp, td);
1795 if (coveredvp != NULL) {
1796 VOP_UNLOCK(coveredvp);
1805 vn_start_write(NULL, &mp, V_WAIT | V_MNTREF);
1807 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 ||
1808 (mp->mnt_flag & MNT_UPDATE) != 0 ||
1809 !TAILQ_EMPTY(&mp->mnt_uppers)) {
1810 dounmount_cleanup(mp, coveredvp, 0);
1813 mp->mnt_kern_flag |= MNTK_UNMOUNT;
1814 rootvp = vfs_cache_root_clear(mp);
1815 if (coveredvp != NULL)
1816 vn_seqc_write_begin(coveredvp);
1817 if (flags & MNT_NONBUSY) {
1819 error = vfs_check_usecounts(mp);
1822 vn_seqc_write_end(coveredvp);
1823 dounmount_cleanup(mp, coveredvp, MNTK_UNMOUNT);
1824 if (rootvp != NULL) {
1825 vn_seqc_write_end(rootvp);
1831 /* Allow filesystems to detect that a forced unmount is in progress. */
1832 if (flags & MNT_FORCE) {
1833 mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1836 * Must be done after setting MNTK_UNMOUNTF and before
1837 * waiting for mnt_lockref to become 0.
1843 if (mp->mnt_lockref) {
1844 mp->mnt_kern_flag |= MNTK_DRAINING;
1845 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS,
1849 KASSERT(mp->mnt_lockref == 0,
1850 ("%s: invalid lock refcount in the drain path @ %s:%d",
1851 __func__, __FILE__, __LINE__));
1853 ("%s: invalid return value for msleep in the drain path @ %s:%d",
1854 __func__, __FILE__, __LINE__));
1857 * We want to keep the vnode around so that we can vn_seqc_write_end
1858 * after we are done with unmount. Downgrade our reference to a mere
1859 * hold count so that we don't interefere with anything.
1861 if (rootvp != NULL) {
1866 if (mp->mnt_flag & MNT_EXPUBLIC)
1867 vfs_setpublicfs(NULL, NULL, NULL);
1869 vfs_periodic(mp, MNT_WAIT);
1871 async_flag = mp->mnt_flag & MNT_ASYNC;
1872 mp->mnt_flag &= ~MNT_ASYNC;
1873 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1875 vfs_deallocate_syncvnode(mp);
1876 error = VFS_UNMOUNT(mp, flags);
1877 vn_finished_write(mp);
1879 * If we failed to flush the dirty blocks for this mount point,
1880 * undo all the cdir/rdir and rootvnode changes we made above.
1881 * Unless we failed to do so because the device is reporting that
1882 * it doesn't exist anymore.
1884 if (error && error != ENXIO) {
1886 if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1888 vfs_allocate_syncvnode(mp);
1891 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1892 mp->mnt_flag |= async_flag;
1893 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1894 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1895 mp->mnt_kern_flag |= MNTK_ASYNC;
1896 if (mp->mnt_kern_flag & MNTK_MWAIT) {
1897 mp->mnt_kern_flag &= ~MNTK_MWAIT;
1900 vfs_op_exit_locked(mp);
1903 vn_seqc_write_end(coveredvp);
1904 VOP_UNLOCK(coveredvp);
1907 if (rootvp != NULL) {
1908 vn_seqc_write_end(rootvp);
1913 mtx_lock(&mountlist_mtx);
1914 TAILQ_REMOVE(&mountlist, mp, mnt_list);
1915 mtx_unlock(&mountlist_mtx);
1916 EVENTHANDLER_DIRECT_INVOKE(vfs_unmounted, mp, td);
1917 if (coveredvp != NULL) {
1919 vn_irflag_unset_locked(coveredvp, VIRF_MOUNTPOINT);
1920 coveredvp->v_mountedhere = NULL;
1921 vn_seqc_write_end_locked(coveredvp);
1922 VI_UNLOCK(coveredvp);
1923 VOP_UNLOCK(coveredvp);
1926 mount_devctl_event("UNMOUNT", mp, false);
1927 if (rootvp != NULL) {
1928 vn_seqc_write_end(rootvp);
1931 vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1932 if (rootvnode != NULL && mp == rootvnode->v_mount) {
1936 if (mp == rootdevmp)
1938 vfs_mount_destroy(mp);
1943 * Report errors during filesystem mounting.
1946 vfs_mount_error(struct mount *mp, const char *fmt, ...)
1948 struct vfsoptlist *moptlist = mp->mnt_optnew;
1953 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1954 if (error || errmsg == NULL || len <= 0)
1958 vsnprintf(errmsg, (size_t)len, fmt, ap);
1963 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...)
1969 error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len);
1970 if (error || errmsg == NULL || len <= 0)
1974 vsnprintf(errmsg, (size_t)len, fmt, ap);
1979 * ---------------------------------------------------------------------
1980 * Functions for querying mount options/arguments from filesystems.
1984 * Check that no unknown options are given
1987 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1991 const char **t, *p, *q;
1994 TAILQ_FOREACH(opt, opts, link) {
1997 if (p[0] == 'n' && p[1] == 'o')
1999 for(t = global_opts; *t != NULL; t++) {
2000 if (strcmp(*t, p) == 0)
2003 if (strcmp(*t, q) == 0)
2009 for(t = legal; *t != NULL; t++) {
2010 if (strcmp(*t, p) == 0)
2013 if (strcmp(*t, q) == 0)
2019 snprintf(errmsg, sizeof(errmsg),
2020 "mount option <%s> is unknown", p);
2024 TAILQ_FOREACH(opt, opts, link) {
2025 if (strcmp(opt->name, "errmsg") == 0) {
2026 strncpy((char *)opt->value, errmsg, opt->len);
2031 printf("%s\n", errmsg);
2037 * Get a mount option by its name.
2039 * Return 0 if the option was found, ENOENT otherwise.
2040 * If len is non-NULL it will be filled with the length
2041 * of the option. If buf is non-NULL, it will be filled
2042 * with the address of the option.
2045 vfs_getopt(struct vfsoptlist *opts, const char *name, void **buf, int *len)
2049 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2051 TAILQ_FOREACH(opt, opts, link) {
2052 if (strcmp(name, opt->name) == 0) {
2065 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
2072 TAILQ_FOREACH(opt, opts, link) {
2073 if (strcmp(name, opt->name) == 0) {
2082 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value)
2084 char *opt_value, *vtp;
2088 error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len);
2091 if (opt_len == 0 || opt_value == NULL)
2093 if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0')
2095 iv = strtoq(opt_value, &vtp, 0);
2096 if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0'))
2123 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
2128 TAILQ_FOREACH(opt, opts, link) {
2129 if (strcmp(name, opt->name) != 0)
2132 if (opt->len == 0 ||
2133 ((char *)opt->value)[opt->len - 1] != '\0') {
2137 return (opt->value);
2144 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w,
2149 TAILQ_FOREACH(opt, opts, link) {
2150 if (strcmp(name, opt->name) == 0) {
2163 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
2169 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2171 TAILQ_FOREACH(opt, opts, link) {
2172 if (strcmp(name, opt->name) != 0)
2175 if (opt->len == 0 || opt->value == NULL)
2177 if (((char *)opt->value)[opt->len - 1] != '\0')
2180 ret = vsscanf(opt->value, fmt, ap);
2188 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len)
2192 TAILQ_FOREACH(opt, opts, link) {
2193 if (strcmp(name, opt->name) != 0)
2196 if (opt->value == NULL)
2199 if (opt->len != len)
2201 bcopy(value, opt->value, len);
2209 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len)
2213 TAILQ_FOREACH(opt, opts, link) {
2214 if (strcmp(name, opt->name) != 0)
2217 if (opt->value == NULL)
2223 bcopy(value, opt->value, len);
2231 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value)
2235 TAILQ_FOREACH(opt, opts, link) {
2236 if (strcmp(name, opt->name) != 0)
2239 if (opt->value == NULL)
2240 opt->len = strlen(value) + 1;
2241 else if (strlcpy(opt->value, value, opt->len) >= opt->len)
2249 * Find and copy a mount option.
2251 * The size of the buffer has to be specified
2252 * in len, if it is not the same length as the
2253 * mount option, EINVAL is returned.
2254 * Returns ENOENT if the option is not found.
2257 vfs_copyopt(struct vfsoptlist *opts, const char *name, void *dest, int len)
2261 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
2263 TAILQ_FOREACH(opt, opts, link) {
2264 if (strcmp(name, opt->name) == 0) {
2266 if (len != opt->len)
2268 bcopy(opt->value, dest, opt->len);
2276 __vfs_statfs(struct mount *mp, struct statfs *sbp)
2280 * Filesystems only fill in part of the structure for updates, we
2281 * have to read the entirety first to get all content.
2283 if (sbp != &mp->mnt_stat)
2284 memcpy(sbp, &mp->mnt_stat, sizeof(*sbp));
2287 * Set these in case the underlying filesystem fails to do so.
2289 sbp->f_version = STATFS_VERSION;
2290 sbp->f_namemax = NAME_MAX;
2291 sbp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
2293 return (mp->mnt_op->vfs_statfs(mp, sbp));
2297 vfs_mountedfrom(struct mount *mp, const char *from)
2300 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
2301 strlcpy(mp->mnt_stat.f_mntfromname, from,
2302 sizeof mp->mnt_stat.f_mntfromname);
2306 * ---------------------------------------------------------------------
2307 * This is the api for building mount args and mounting filesystems from
2308 * inside the kernel.
2310 * The API works by accumulation of individual args. First error is
2313 * XXX: should be documented in new manpage kernel_mount(9)
2316 /* A memory allocation which must be freed when we are done */
2318 SLIST_ENTRY(mntaarg) next;
2321 /* The header for the mount arguments */
2326 SLIST_HEAD(, mntaarg) list;
2330 * Add a boolean argument.
2332 * flag is the boolean value.
2333 * name must start with "no".
2336 mount_argb(struct mntarg *ma, int flag, const char *name)
2339 KASSERT(name[0] == 'n' && name[1] == 'o',
2340 ("mount_argb(...,%s): name must start with 'no'", name));
2342 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
2346 * Add an argument printf style
2349 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
2352 struct mntaarg *maa;
2357 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2358 SLIST_INIT(&ma->list);
2363 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2365 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2366 ma->v[ma->len].iov_len = strlen(name) + 1;
2369 sb = sbuf_new_auto();
2371 sbuf_vprintf(sb, fmt, ap);
2374 len = sbuf_len(sb) + 1;
2375 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2376 SLIST_INSERT_HEAD(&ma->list, maa, next);
2377 bcopy(sbuf_data(sb), maa + 1, len);
2380 ma->v[ma->len].iov_base = maa + 1;
2381 ma->v[ma->len].iov_len = len;
2388 * Add an argument which is a userland string.
2391 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2393 struct mntaarg *maa;
2399 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2400 SLIST_INIT(&ma->list);
2404 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2405 SLIST_INSERT_HEAD(&ma->list, maa, next);
2406 tbuf = (void *)(maa + 1);
2407 ma->error = copyinstr(val, tbuf, len, NULL);
2408 return (mount_arg(ma, name, tbuf, -1));
2414 * If length is -1, treat value as a C string.
2417 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2421 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2422 SLIST_INIT(&ma->list);
2427 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2429 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2430 ma->v[ma->len].iov_len = strlen(name) + 1;
2433 ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2435 ma->v[ma->len].iov_len = strlen(val) + 1;
2437 ma->v[ma->len].iov_len = len;
2443 * Free a mntarg structure
2446 free_mntarg(struct mntarg *ma)
2448 struct mntaarg *maa;
2450 while (!SLIST_EMPTY(&ma->list)) {
2451 maa = SLIST_FIRST(&ma->list);
2452 SLIST_REMOVE_HEAD(&ma->list, next);
2455 free(ma->v, M_MOUNT);
2460 * Mount a filesystem
2463 kernel_mount(struct mntarg *ma, uint64_t flags)
2468 KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2469 KASSERT(ma->error != 0 || ma->v != NULL, ("kernel_mount NULL ma->v"));
2470 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2474 auio.uio_iov = ma->v;
2475 auio.uio_iovcnt = ma->len;
2476 auio.uio_segflg = UIO_SYSSPACE;
2477 error = vfs_donmount(curthread, flags, &auio);
2484 * A printflike function to mount a filesystem.
2487 kernel_vmount(int flags, ...)
2489 struct mntarg *ma = NULL;
2495 va_start(ap, flags);
2497 cp = va_arg(ap, const char *);
2500 vp = va_arg(ap, const void *);
2501 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0));
2505 error = kernel_mount(ma, flags);
2509 /* Map from mount options to printable formats. */
2510 static struct mntoptnames optnames[] = {
2514 #define DEVCTL_LEN 1024
2516 mount_devctl_event(const char *type, struct mount *mp, bool donew)
2519 struct mntoptnames *fp;
2521 struct statfs *sfp = &mp->mnt_stat;
2524 buf = malloc(DEVCTL_LEN, M_MOUNT, M_NOWAIT);
2527 sbuf_new(&sb, buf, DEVCTL_LEN, SBUF_FIXEDLEN);
2528 sbuf_cpy(&sb, "mount-point=\"");
2529 devctl_safe_quote_sb(&sb, sfp->f_mntonname);
2530 sbuf_cat(&sb, "\" mount-dev=\"");
2531 devctl_safe_quote_sb(&sb, sfp->f_mntfromname);
2532 sbuf_cat(&sb, "\" mount-type=\"");
2533 devctl_safe_quote_sb(&sb, sfp->f_fstypename);
2534 sbuf_cat(&sb, "\" fsid=0x");
2535 cp = (const uint8_t *)&sfp->f_fsid.val[0];
2536 for (int i = 0; i < sizeof(sfp->f_fsid); i++)
2537 sbuf_printf(&sb, "%02x", cp[i]);
2538 sbuf_printf(&sb, " owner=%u flags=\"", sfp->f_owner);
2539 for (fp = optnames; fp->o_opt != 0; fp++) {
2540 if ((mp->mnt_flag & fp->o_opt) != 0) {
2541 sbuf_cat(&sb, fp->o_name);
2542 sbuf_putc(&sb, ';');
2545 sbuf_putc(&sb, '"');
2549 * Options are not published because the form of the options depends on
2550 * the file system and may include binary data. In addition, they don't
2551 * necessarily provide enough useful information to be actionable when
2552 * devd processes them.
2555 if (sbuf_error(&sb) == 0)
2556 devctl_notify("VFS", "FS", type, sbuf_data(&sb));
2562 * Force remount specified mount point to read-only. The argument
2563 * must be busied to avoid parallel unmount attempts.
2565 * Intended use is to prevent further writes if some metadata
2566 * inconsistency is detected. Note that the function still flushes
2567 * all cached metadata and data for the mount point, which might be
2568 * not always suitable.
2571 vfs_remount_ro(struct mount *mp)
2573 struct vfsoptlist *opts;
2575 struct vnode *vp_covered, *rootvp;
2578 KASSERT(mp->mnt_lockref > 0,
2579 ("vfs_remount_ro: mp %p is not busied", mp));
2580 KASSERT((mp->mnt_kern_flag & MNTK_UNMOUNT) == 0,
2581 ("vfs_remount_ro: mp %p is being unmounted (and busy?)", mp));
2584 vp_covered = mp->mnt_vnodecovered;
2585 error = vget(vp_covered, LK_EXCLUSIVE | LK_NOWAIT);
2588 VI_LOCK(vp_covered);
2589 if ((vp_covered->v_iflag & VI_MOUNT) != 0) {
2590 VI_UNLOCK(vp_covered);
2594 vp_covered->v_iflag |= VI_MOUNT;
2595 VI_UNLOCK(vp_covered);
2597 vn_seqc_write_begin(vp_covered);
2600 if ((mp->mnt_flag & MNT_RDONLY) != 0) {
2605 mp->mnt_flag |= MNT_UPDATE | MNT_FORCE | MNT_RDONLY;
2606 rootvp = vfs_cache_root_clear(mp);
2609 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK | M_ZERO);
2611 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK | M_ZERO);
2612 opt->name = strdup("ro", M_MOUNT);
2614 TAILQ_INSERT_TAIL(opts, opt, link);
2615 vfs_mergeopts(opts, mp->mnt_opt);
2616 mp->mnt_optnew = opts;
2618 error = VFS_MOUNT(mp);
2622 mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE);
2624 vfs_deallocate_syncvnode(mp);
2625 if (mp->mnt_opt != NULL)
2626 vfs_freeopts(mp->mnt_opt);
2627 mp->mnt_opt = mp->mnt_optnew;
2630 mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE | MNT_RDONLY);
2632 vfs_freeopts(mp->mnt_optnew);
2634 mp->mnt_optnew = NULL;
2638 VI_LOCK(vp_covered);
2639 vp_covered->v_iflag &= ~VI_MOUNT;
2640 VI_UNLOCK(vp_covered);
2642 vn_seqc_write_end(vp_covered);
2643 if (rootvp != NULL) {
2644 vn_seqc_write_end(rootvp);
2651 * Suspend write operations on all local writeable filesystems. Does
2652 * full sync of them in the process.
2654 * Iterate over the mount points in reverse order, suspending most
2655 * recently mounted filesystems first. It handles a case where a
2656 * filesystem mounted from a md(4) vnode-backed device should be
2657 * suspended before the filesystem that owns the vnode.
2660 suspend_all_fs(void)
2665 mtx_lock(&mountlist_mtx);
2666 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
2667 error = vfs_busy(mp, MBF_MNTLSTLOCK | MBF_NOWAIT);
2670 if ((mp->mnt_flag & (MNT_RDONLY | MNT_LOCAL)) != MNT_LOCAL ||
2671 (mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
2672 mtx_lock(&mountlist_mtx);
2676 error = vfs_write_suspend(mp, 0);
2679 MPASS((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0);
2680 mp->mnt_kern_flag |= MNTK_SUSPEND_ALL;
2682 mtx_lock(&mountlist_mtx);
2684 printf("suspend of %s failed, error %d\n",
2685 mp->mnt_stat.f_mntonname, error);
2686 mtx_lock(&mountlist_mtx);
2690 mtx_unlock(&mountlist_mtx);
2698 mtx_lock(&mountlist_mtx);
2699 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2700 if ((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0)
2702 mtx_unlock(&mountlist_mtx);
2704 MPASS((mp->mnt_kern_flag & MNTK_SUSPEND) != 0);
2705 mp->mnt_kern_flag &= ~MNTK_SUSPEND_ALL;
2707 vfs_write_resume(mp, 0);
2708 mtx_lock(&mountlist_mtx);
2711 mtx_unlock(&mountlist_mtx);