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 if (mp->mnt_ref != 0 || mp->mnt_lockref != 0 ||
546 mp->mnt_writeopcount != 0)
547 panic("%s: non-zero counters on new mp %p\n", __func__, mp);
548 if (mp->mnt_vfs_ops != 1)
549 panic("%s: vfs_ops should be 1 but %d found\n", __func__,
551 (void) vfs_busy(mp, MBF_NOWAIT);
552 atomic_add_acq_int(&vfsp->vfc_refcount, 1);
553 mp->mnt_op = vfsp->vfc_vfsops;
555 mp->mnt_stat.f_type = vfsp->vfc_typenum;
557 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
558 mp->mnt_vnodecovered = vp;
559 mp->mnt_cred = crdup(cred);
560 mp->mnt_stat.f_owner = cred->cr_uid;
561 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
562 mp->mnt_iosize_max = DFLTPHYS;
565 mac_mount_create(cred, mp);
567 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
568 TAILQ_INIT(&mp->mnt_uppers);
573 * Destroy the mount struct previously allocated by vfs_mount_alloc().
576 vfs_mount_destroy(struct mount *mp)
579 if (mp->mnt_vfs_ops == 0)
580 panic("%s: entered with zero vfs_ops\n", __func__);
582 vfs_assert_mount_counters(mp);
585 mp->mnt_kern_flag |= MNTK_REFEXPIRE;
586 if (mp->mnt_kern_flag & MNTK_MWAIT) {
587 mp->mnt_kern_flag &= ~MNTK_MWAIT;
591 msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0);
592 KASSERT(mp->mnt_ref == 0,
593 ("%s: invalid refcount in the drain path @ %s:%d", __func__,
594 __FILE__, __LINE__));
595 if (mp->mnt_writeopcount != 0)
596 panic("vfs_mount_destroy: nonzero writeopcount");
597 if (mp->mnt_secondary_writes != 0)
598 panic("vfs_mount_destroy: nonzero secondary_writes");
599 atomic_subtract_rel_int(&mp->mnt_vfc->vfc_refcount, 1);
600 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
603 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
604 vn_printf(vp, "dangling vnode ");
605 panic("unmount: dangling vnode");
607 KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers"));
608 if (mp->mnt_nvnodelistsize != 0)
609 panic("vfs_mount_destroy: nonzero nvnodelistsize");
610 if (mp->mnt_lazyvnodelistsize != 0)
611 panic("vfs_mount_destroy: nonzero lazyvnodelistsize");
612 if (mp->mnt_lockref != 0)
613 panic("vfs_mount_destroy: nonzero lock refcount");
616 if (mp->mnt_vfs_ops != 1)
617 panic("%s: vfs_ops should be 1 but %d found\n", __func__,
620 if (mp->mnt_rootvnode != NULL)
621 panic("%s: mount point still has a root vnode %p\n", __func__,
624 if (mp->mnt_vnodecovered != NULL)
625 vrele(mp->mnt_vnodecovered);
627 mac_mount_destroy(mp);
629 if (mp->mnt_opt != NULL)
630 vfs_freeopts(mp->mnt_opt);
631 crfree(mp->mnt_cred);
632 uma_zfree(mount_zone, mp);
636 vfs_should_downgrade_to_ro_mount(uint64_t fsflags, int error)
638 /* This is an upgrade of an exisiting mount. */
639 if ((fsflags & MNT_UPDATE) != 0)
641 /* This is already an R/O mount. */
642 if ((fsflags & MNT_RDONLY) != 0)
646 case ENODEV: /* generic, geom, ... */
647 case EACCES: /* cam/scsi, ... */
648 case EROFS: /* md, mmcsd, ... */
650 * These errors can be returned by the storage layer to signal
651 * that the media is read-only. No harm in the R/O mount
652 * attempt if the error was returned for some other reason.
661 vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions)
663 struct vfsoptlist *optlist;
664 struct vfsopt *opt, *tmp_opt;
665 char *fstype, *fspath, *errmsg;
666 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
669 errmsg = fspath = NULL;
670 errmsg_len = fspathlen = 0;
672 autoro = default_autoro;
674 error = vfs_buildopts(fsoptions, &optlist);
678 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
679 errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
682 * We need these two options before the others,
683 * and they are mandatory for any filesystem.
684 * Ensure they are NUL terminated as well.
687 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
688 if (error || fstypelen <= 0 || fstype[fstypelen - 1] != '\0') {
691 strncpy(errmsg, "Invalid fstype", errmsg_len);
695 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
696 if (error || fspathlen <= 0 || fspath[fspathlen - 1] != '\0') {
699 strncpy(errmsg, "Invalid fspath", errmsg_len);
704 * We need to see if we have the "update" option
705 * before we call vfs_domount(), since vfs_domount() has special
706 * logic based on MNT_UPDATE. This is very important
707 * when we want to update the root filesystem.
709 TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) {
712 if (strcmp(opt->name, "update") == 0) {
713 fsflags |= MNT_UPDATE;
716 else if (strcmp(opt->name, "async") == 0)
717 fsflags |= MNT_ASYNC;
718 else if (strcmp(opt->name, "force") == 0) {
719 fsflags |= MNT_FORCE;
722 else if (strcmp(opt->name, "reload") == 0) {
723 fsflags |= MNT_RELOAD;
726 else if (strcmp(opt->name, "multilabel") == 0)
727 fsflags |= MNT_MULTILABEL;
728 else if (strcmp(opt->name, "noasync") == 0)
729 fsflags &= ~MNT_ASYNC;
730 else if (strcmp(opt->name, "noatime") == 0)
731 fsflags |= MNT_NOATIME;
732 else if (strcmp(opt->name, "atime") == 0) {
733 free(opt->name, M_MOUNT);
734 opt->name = strdup("nonoatime", M_MOUNT);
736 else if (strcmp(opt->name, "noclusterr") == 0)
737 fsflags |= MNT_NOCLUSTERR;
738 else if (strcmp(opt->name, "clusterr") == 0) {
739 free(opt->name, M_MOUNT);
740 opt->name = strdup("nonoclusterr", M_MOUNT);
742 else if (strcmp(opt->name, "noclusterw") == 0)
743 fsflags |= MNT_NOCLUSTERW;
744 else if (strcmp(opt->name, "clusterw") == 0) {
745 free(opt->name, M_MOUNT);
746 opt->name = strdup("nonoclusterw", M_MOUNT);
748 else if (strcmp(opt->name, "noexec") == 0)
749 fsflags |= MNT_NOEXEC;
750 else if (strcmp(opt->name, "exec") == 0) {
751 free(opt->name, M_MOUNT);
752 opt->name = strdup("nonoexec", M_MOUNT);
754 else if (strcmp(opt->name, "nosuid") == 0)
755 fsflags |= MNT_NOSUID;
756 else if (strcmp(opt->name, "suid") == 0) {
757 free(opt->name, M_MOUNT);
758 opt->name = strdup("nonosuid", M_MOUNT);
760 else if (strcmp(opt->name, "nosymfollow") == 0)
761 fsflags |= MNT_NOSYMFOLLOW;
762 else if (strcmp(opt->name, "symfollow") == 0) {
763 free(opt->name, M_MOUNT);
764 opt->name = strdup("nonosymfollow", M_MOUNT);
766 else if (strcmp(opt->name, "noro") == 0) {
767 fsflags &= ~MNT_RDONLY;
770 else if (strcmp(opt->name, "rw") == 0) {
771 fsflags &= ~MNT_RDONLY;
774 else if (strcmp(opt->name, "ro") == 0) {
775 fsflags |= MNT_RDONLY;
778 else if (strcmp(opt->name, "rdonly") == 0) {
779 free(opt->name, M_MOUNT);
780 opt->name = strdup("ro", M_MOUNT);
781 fsflags |= MNT_RDONLY;
784 else if (strcmp(opt->name, "autoro") == 0) {
788 else if (strcmp(opt->name, "suiddir") == 0)
789 fsflags |= MNT_SUIDDIR;
790 else if (strcmp(opt->name, "sync") == 0)
791 fsflags |= MNT_SYNCHRONOUS;
792 else if (strcmp(opt->name, "union") == 0)
793 fsflags |= MNT_UNION;
794 else if (strcmp(opt->name, "automounted") == 0) {
795 fsflags |= MNT_AUTOMOUNTED;
797 } else if (strcmp(opt->name, "nocover") == 0) {
798 fsflags |= MNT_NOCOVER;
800 } else if (strcmp(opt->name, "cover") == 0) {
801 fsflags &= ~MNT_NOCOVER;
803 } else if (strcmp(opt->name, "emptydir") == 0) {
804 fsflags |= MNT_EMPTYDIR;
806 } else if (strcmp(opt->name, "noemptydir") == 0) {
807 fsflags &= ~MNT_EMPTYDIR;
811 vfs_freeopt(optlist, opt);
815 * Be ultra-paranoid about making sure the type and fspath
816 * variables will fit in our mp buffers, including the
819 if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) {
820 error = ENAMETOOLONG;
824 error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
825 if (error == ENOENT) {
828 strncpy(errmsg, "Invalid fstype", errmsg_len);
833 * See if we can mount in the read-only mode if the error code suggests
834 * that it could be possible and the mount options allow for that.
835 * Never try it if "[no]{ro|rw}" has been explicitly requested and not
836 * overridden by "autoro".
838 if (autoro && vfs_should_downgrade_to_ro_mount(fsflags, error)) {
839 printf("%s: R/W mount failed, possibly R/O media,"
840 " trying R/O mount\n", __func__);
841 fsflags |= MNT_RDONLY;
842 error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
845 /* copyout the errmsg */
846 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
847 && errmsg_len > 0 && errmsg != NULL) {
848 if (fsoptions->uio_segflg == UIO_SYSSPACE) {
850 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
851 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
854 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
855 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
860 vfs_freeopts(optlist);
867 #ifndef _SYS_SYSPROTO_H_
877 sys_mount(struct thread *td, struct mount_args *uap)
880 struct vfsconf *vfsp = NULL;
881 struct mntarg *ma = NULL;
886 * Mount flags are now 64-bits. On 32-bit architectures only
887 * 32-bits are passed in, but from here on everything handles
888 * 64-bit flags correctly.
892 AUDIT_ARG_FFLAGS(flags);
895 * Filter out MNT_ROOTFS. We do not want clients of mount() in
896 * userspace to set this flag, but we must filter it out if we want
897 * MNT_UPDATE on the root file system to work.
898 * MNT_ROOTFS should only be set by the kernel when mounting its
901 flags &= ~MNT_ROOTFS;
903 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
904 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
906 free(fstype, M_TEMP);
910 AUDIT_ARG_TEXT(fstype);
911 vfsp = vfs_byname_kld(fstype, td, &error);
912 free(fstype, M_TEMP);
915 if (((vfsp->vfc_flags & VFCF_SBDRY) != 0 &&
916 vfsp->vfc_vfsops_sd->vfs_cmount == NULL) ||
917 ((vfsp->vfc_flags & VFCF_SBDRY) == 0 &&
918 vfsp->vfc_vfsops->vfs_cmount == NULL))
921 ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN);
922 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
923 ma = mount_argb(ma, flags & MNT_RDONLY, "noro");
924 ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid");
925 ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec");
927 if ((vfsp->vfc_flags & VFCF_SBDRY) != 0)
928 return (vfsp->vfc_vfsops_sd->vfs_cmount(ma, uap->data, flags));
929 return (vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags));
933 * vfs_domount_first(): first file system mount (not update)
937 struct thread *td, /* Calling thread. */
938 struct vfsconf *vfsp, /* File system type. */
939 char *fspath, /* Mount path. */
940 struct vnode *vp, /* Vnode to be covered. */
941 uint64_t fsflags, /* Flags common to all filesystems. */
942 struct vfsoptlist **optlist /* Options local to the filesystem. */
947 struct vnode *newdp, *rootvp;
951 ASSERT_VOP_ELOCKED(vp, __func__);
952 KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here"));
955 * If the jail of the calling thread lacks permission for this type of
956 * file system, or is trying to cover its own root, deny immediately.
958 if (jailed(td->td_ucred) && (!prison_allow(td->td_ucred,
959 vfsp->vfc_prison_flag) || vp == td->td_ucred->cr_prison->pr_root)) {
965 * If the user is not root, ensure that they own the directory
966 * onto which we are attempting to mount.
968 error = VOP_GETATTR(vp, &va, td->td_ucred);
969 if (error == 0 && va.va_uid != td->td_ucred->cr_uid)
970 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN);
972 error = vinvalbuf(vp, V_SAVE, 0, 0);
973 if (error == 0 && vp->v_type != VDIR)
975 if (error == 0 && (fsflags & MNT_EMPTYDIR) != 0)
976 error = vfs_emptydir(vp);
979 if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL)
980 vp->v_iflag |= VI_MOUNT;
989 vn_seqc_write_begin(vp);
992 /* Allocate and initialize the filesystem. */
993 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred);
994 /* XXXMAC: pass to vfs_mount_alloc? */
995 mp->mnt_optnew = *optlist;
996 /* Set the mount level flags. */
997 mp->mnt_flag = (fsflags & (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY));
1000 * Mount the filesystem.
1001 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1002 * get. No freeing of cn_pnbuf.
1006 if ((error = VFS_MOUNT(mp)) != 0 ||
1007 (error1 = VFS_STATFS(mp, &mp->mnt_stat)) != 0 ||
1008 (error1 = VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) != 0) {
1012 rootvp = vfs_cache_root_clear(mp);
1013 if (rootvp != NULL) {
1017 (void)vn_start_write(NULL, &mp, V_WAIT);
1019 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_UNMOUNTF;
1022 error = VFS_UNMOUNT(mp, 0);
1023 vn_finished_write(mp);
1026 "failed post-mount (%d): rollback unmount returned %d\n",
1033 mp->mnt_vnodecovered = NULL;
1035 /* XXXKIB wait for mnt_lockref drain? */
1036 vfs_mount_destroy(mp);
1039 vp->v_iflag &= ~VI_MOUNT;
1041 if (rootvp != NULL) {
1042 vn_seqc_write_end(rootvp);
1045 vn_seqc_write_end(vp);
1049 vn_seqc_write_begin(newdp);
1052 if (mp->mnt_opt != NULL)
1053 vfs_freeopts(mp->mnt_opt);
1054 mp->mnt_opt = mp->mnt_optnew;
1058 * Prevent external consumers of mount options from reading mnt_optnew.
1060 mp->mnt_optnew = NULL;
1063 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1064 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1065 mp->mnt_kern_flag |= MNTK_ASYNC;
1067 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1070 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1073 vp->v_iflag &= ~VI_MOUNT;
1074 vn_irflag_set_locked(vp, VIRF_MOUNTPOINT);
1075 vp->v_mountedhere = mp;
1077 /* Place the new filesystem at the end of the mount list. */
1078 mtx_lock(&mountlist_mtx);
1079 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1080 mtx_unlock(&mountlist_mtx);
1081 vfs_event_signal(NULL, VQ_MOUNT, 0);
1082 vn_lock(newdp, LK_EXCLUSIVE | LK_RETRY);
1084 EVENTHANDLER_DIRECT_INVOKE(vfs_mounted, mp, newdp, td);
1086 mount_devctl_event("MOUNT", mp, false);
1087 mountcheckdirs(vp, newdp);
1088 vn_seqc_write_end(vp);
1089 vn_seqc_write_end(newdp);
1091 if ((mp->mnt_flag & MNT_RDONLY) == 0)
1092 vfs_allocate_syncvnode(mp);
1099 * vfs_domount_update(): update of mounted file system
1103 struct thread *td, /* Calling thread. */
1104 struct vnode *vp, /* Mount point vnode. */
1105 uint64_t fsflags, /* Flags common to all filesystems. */
1106 struct vfsoptlist **optlist /* Options local to the filesystem. */
1109 struct export_args export;
1110 struct o2export_args o2export;
1111 struct vnode *rootvp;
1114 int error, export_error, i, len;
1118 ASSERT_VOP_ELOCKED(vp, __func__);
1119 KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here"));
1122 if ((vp->v_vflag & VV_ROOT) == 0) {
1123 if (vfs_copyopt(*optlist, "export", &export, sizeof(export))
1133 * We only allow the filesystem to be reloaded if it
1134 * is currently mounted read-only.
1136 flag = mp->mnt_flag;
1137 if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) {
1139 return (EOPNOTSUPP); /* Needs translation */
1142 * Only privileged root, or (if MNT_USER is set) the user that
1143 * did the original mount is permitted to update it.
1145 error = vfs_suser(mp, td);
1150 if (vfs_busy(mp, MBF_NOWAIT)) {
1155 if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) {
1161 vp->v_iflag |= VI_MOUNT;
1166 vn_seqc_write_begin(vp);
1170 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1175 mp->mnt_flag &= ~MNT_UPDATEMASK;
1176 mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE |
1177 MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY);
1178 if ((mp->mnt_flag & MNT_ASYNC) == 0)
1179 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1180 rootvp = vfs_cache_root_clear(mp);
1182 mp->mnt_optnew = *optlist;
1183 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
1186 * Mount the filesystem.
1187 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1188 * get. No freeing of cn_pnbuf.
1190 error = VFS_MOUNT(mp);
1193 /* Process the export option. */
1194 if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp,
1196 /* Assume that there is only 1 ABI for each length. */
1198 case (sizeof(struct oexport_args)):
1199 bzero(&o2export, sizeof(o2export));
1201 case (sizeof(o2export)):
1202 bcopy(bufp, &o2export, len);
1203 export.ex_flags = (uint64_t)o2export.ex_flags;
1204 export.ex_root = o2export.ex_root;
1205 export.ex_uid = o2export.ex_anon.cr_uid;
1206 export.ex_groups = NULL;
1207 export.ex_ngroups = o2export.ex_anon.cr_ngroups;
1208 if (export.ex_ngroups > 0) {
1209 if (export.ex_ngroups <= XU_NGROUPS) {
1210 export.ex_groups = malloc(
1211 export.ex_ngroups * sizeof(gid_t),
1213 for (i = 0; i < export.ex_ngroups; i++)
1214 export.ex_groups[i] =
1215 o2export.ex_anon.cr_groups[i];
1217 export_error = EINVAL;
1218 } else if (export.ex_ngroups < 0)
1219 export_error = EINVAL;
1220 export.ex_addr = o2export.ex_addr;
1221 export.ex_addrlen = o2export.ex_addrlen;
1222 export.ex_mask = o2export.ex_mask;
1223 export.ex_masklen = o2export.ex_masklen;
1224 export.ex_indexfile = o2export.ex_indexfile;
1225 export.ex_numsecflavors = o2export.ex_numsecflavors;
1226 if (export.ex_numsecflavors < MAXSECFLAVORS) {
1227 for (i = 0; i < export.ex_numsecflavors; i++)
1228 export.ex_secflavors[i] =
1229 o2export.ex_secflavors[i];
1231 export_error = EINVAL;
1232 if (export_error == 0)
1233 export_error = vfs_export(mp, &export);
1234 free(export.ex_groups, M_TEMP);
1236 case (sizeof(export)):
1237 bcopy(bufp, &export, len);
1239 if (export.ex_ngroups > 0) {
1240 if (export.ex_ngroups <= NGROUPS_MAX) {
1241 grps = malloc(export.ex_ngroups *
1242 sizeof(gid_t), M_TEMP, M_WAITOK);
1243 export_error = copyin(export.ex_groups,
1244 grps, export.ex_ngroups *
1246 if (export_error == 0)
1247 export.ex_groups = grps;
1249 export_error = EINVAL;
1250 } else if (export.ex_ngroups == 0)
1251 export.ex_groups = NULL;
1253 export_error = EINVAL;
1254 if (export_error == 0)
1255 export_error = vfs_export(mp, &export);
1259 export_error = EINVAL;
1266 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE |
1270 * If we fail, restore old mount flags. MNT_QUOTA is special,
1271 * because it is not part of MNT_UPDATEMASK, but it could have
1272 * changed in the meantime if quotactl(2) was called.
1273 * All in all we want current value of MNT_QUOTA, not the old
1276 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA);
1278 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1279 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1280 mp->mnt_kern_flag |= MNTK_ASYNC;
1282 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1288 mount_devctl_event("REMOUNT", mp, true);
1289 if (mp->mnt_opt != NULL)
1290 vfs_freeopts(mp->mnt_opt);
1291 mp->mnt_opt = mp->mnt_optnew;
1293 (void)VFS_STATFS(mp, &mp->mnt_stat);
1295 * Prevent external consumers of mount options from reading
1298 mp->mnt_optnew = NULL;
1300 if ((mp->mnt_flag & MNT_RDONLY) == 0)
1301 vfs_allocate_syncvnode(mp);
1303 vfs_deallocate_syncvnode(mp);
1306 if (rootvp != NULL) {
1307 vn_seqc_write_end(rootvp);
1310 vn_seqc_write_end(vp);
1313 vp->v_iflag &= ~VI_MOUNT;
1316 return (error != 0 ? error : export_error);
1320 * vfs_domount(): actually attempt a filesystem mount.
1324 struct thread *td, /* Calling thread. */
1325 const char *fstype, /* Filesystem type. */
1326 char *fspath, /* Mount path. */
1327 uint64_t fsflags, /* Flags common to all filesystems. */
1328 struct vfsoptlist **optlist /* Options local to the filesystem. */
1331 struct vfsconf *vfsp;
1332 struct nameidata nd;
1338 * Be ultra-paranoid about making sure the type and fspath
1339 * variables will fit in our mp buffers, including the
1342 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
1343 return (ENAMETOOLONG);
1345 if (jailed(td->td_ucred) || usermount == 0) {
1346 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
1351 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
1353 if (fsflags & MNT_EXPORTED) {
1354 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
1358 if (fsflags & MNT_SUIDDIR) {
1359 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
1364 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
1366 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
1367 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
1368 fsflags |= MNT_NOSUID | MNT_USER;
1371 /* Load KLDs before we lock the covered vnode to avoid reversals. */
1373 if ((fsflags & MNT_UPDATE) == 0) {
1374 /* Don't try to load KLDs if we're mounting the root. */
1375 if (fsflags & MNT_ROOTFS) {
1376 if ((vfsp = vfs_byname(fstype)) == NULL)
1379 if ((vfsp = vfs_byname_kld(fstype, td, &error)) == NULL)
1385 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE.
1387 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1388 UIO_SYSSPACE, fspath, td);
1392 NDFREE(&nd, NDF_ONLY_PNBUF);
1394 if ((fsflags & MNT_UPDATE) == 0) {
1395 if ((vp->v_vflag & VV_ROOT) != 0 &&
1396 (fsflags & MNT_NOCOVER) != 0) {
1400 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1401 strcpy(pathbuf, fspath);
1402 error = vn_path_to_global_path(td, vp, pathbuf, MNAMELEN);
1404 error = vfs_domount_first(td, vfsp, pathbuf, vp,
1407 free(pathbuf, M_TEMP);
1409 error = vfs_domount_update(td, vp, fsflags, optlist);
1415 * Unmount a filesystem.
1417 * Note: unmount takes a path to the vnode mounted on as argument, not
1418 * special file (as before).
1420 #ifndef _SYS_SYSPROTO_H_
1421 struct unmount_args {
1428 sys_unmount(struct thread *td, struct unmount_args *uap)
1431 return (kern_unmount(td, uap->path, uap->flags));
1435 kern_unmount(struct thread *td, const char *path, int flags)
1437 struct nameidata nd;
1440 int error, id0, id1;
1442 AUDIT_ARG_VALUE(flags);
1443 if (jailed(td->td_ucred) || usermount == 0) {
1444 error = priv_check(td, PRIV_VFS_UNMOUNT);
1449 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1450 error = copyinstr(path, pathbuf, MNAMELEN, NULL);
1452 free(pathbuf, M_TEMP);
1455 if (flags & MNT_BYFSID) {
1456 AUDIT_ARG_TEXT(pathbuf);
1457 /* Decode the filesystem ID. */
1458 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1459 free(pathbuf, M_TEMP);
1463 mtx_lock(&mountlist_mtx);
1464 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1465 if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1466 mp->mnt_stat.f_fsid.val[1] == id1) {
1471 mtx_unlock(&mountlist_mtx);
1474 * Try to find global path for path argument.
1476 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1477 UIO_SYSSPACE, pathbuf, td);
1478 if (namei(&nd) == 0) {
1479 NDFREE(&nd, NDF_ONLY_PNBUF);
1480 error = vn_path_to_global_path(td, nd.ni_vp, pathbuf,
1485 mtx_lock(&mountlist_mtx);
1486 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1487 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) {
1492 mtx_unlock(&mountlist_mtx);
1494 free(pathbuf, M_TEMP);
1497 * Previously we returned ENOENT for a nonexistent path and
1498 * EINVAL for a non-mountpoint. We cannot tell these apart
1499 * now, so in the !MNT_BYFSID case return the more likely
1500 * EINVAL for compatibility.
1502 return ((flags & MNT_BYFSID) ? ENOENT : EINVAL);
1506 * Don't allow unmounting the root filesystem.
1508 if (mp->mnt_flag & MNT_ROOTFS) {
1512 error = dounmount(mp, flags, td);
1517 * Return error if any of the vnodes, ignoring the root vnode
1518 * and the syncer vnode, have non-zero usecount.
1520 * This function is purely advisory - it can return false positives
1524 vfs_check_usecounts(struct mount *mp)
1526 struct vnode *vp, *mvp;
1528 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1529 if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON &&
1530 vp->v_usecount != 0) {
1532 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1542 dounmount_cleanup(struct mount *mp, struct vnode *coveredvp, int mntkflags)
1545 mtx_assert(MNT_MTX(mp), MA_OWNED);
1546 mp->mnt_kern_flag &= ~mntkflags;
1547 if ((mp->mnt_kern_flag & MNTK_MWAIT) != 0) {
1548 mp->mnt_kern_flag &= ~MNTK_MWAIT;
1551 vfs_op_exit_locked(mp);
1553 if (coveredvp != NULL) {
1554 VOP_UNLOCK(coveredvp);
1557 vn_finished_write(mp);
1561 * There are various reference counters associated with the mount point.
1562 * Normally it is permitted to modify them without taking the mnt ilock,
1563 * but this behavior can be temporarily disabled if stable value is needed
1564 * or callers are expected to block (e.g. to not allow new users during
1568 vfs_op_enter(struct mount *mp)
1570 struct mount_pcpu *mpcpu;
1575 if (mp->mnt_vfs_ops > 1) {
1579 vfs_op_barrier_wait(mp);
1581 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1583 mp->mnt_ref += mpcpu->mntp_ref;
1584 mpcpu->mntp_ref = 0;
1586 mp->mnt_lockref += mpcpu->mntp_lockref;
1587 mpcpu->mntp_lockref = 0;
1589 mp->mnt_writeopcount += mpcpu->mntp_writeopcount;
1590 mpcpu->mntp_writeopcount = 0;
1592 if (mp->mnt_ref <= 0 || mp->mnt_lockref < 0 || mp->mnt_writeopcount < 0)
1593 panic("%s: invalid count(s) on mp %p: ref %d lockref %d writeopcount %d\n",
1594 __func__, mp, mp->mnt_ref, mp->mnt_lockref, mp->mnt_writeopcount);
1596 vfs_assert_mount_counters(mp);
1600 vfs_op_exit_locked(struct mount *mp)
1603 mtx_assert(MNT_MTX(mp), MA_OWNED);
1605 if (mp->mnt_vfs_ops <= 0)
1606 panic("%s: invalid vfs_ops count %d for mp %p\n",
1607 __func__, mp->mnt_vfs_ops, mp);
1612 vfs_op_exit(struct mount *mp)
1616 vfs_op_exit_locked(mp);
1620 struct vfs_op_barrier_ipi {
1622 struct smp_rendezvous_cpus_retry_arg srcra;
1626 vfs_op_action_func(void *arg)
1628 struct vfs_op_barrier_ipi *vfsopipi;
1631 vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1634 if (!vfs_op_thread_entered(mp))
1635 smp_rendezvous_cpus_done(arg);
1639 vfs_op_wait_func(void *arg, int cpu)
1641 struct vfs_op_barrier_ipi *vfsopipi;
1643 struct mount_pcpu *mpcpu;
1645 vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1648 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1649 while (atomic_load_int(&mpcpu->mntp_thread_in_ops))
1654 vfs_op_barrier_wait(struct mount *mp)
1656 struct vfs_op_barrier_ipi vfsopipi;
1660 smp_rendezvous_cpus_retry(all_cpus,
1661 smp_no_rendezvous_barrier,
1663 smp_no_rendezvous_barrier,
1670 vfs_assert_mount_counters(struct mount *mp)
1672 struct mount_pcpu *mpcpu;
1675 if (mp->mnt_vfs_ops == 0)
1679 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1680 if (mpcpu->mntp_ref != 0 ||
1681 mpcpu->mntp_lockref != 0 ||
1682 mpcpu->mntp_writeopcount != 0)
1683 vfs_dump_mount_counters(mp);
1688 vfs_dump_mount_counters(struct mount *mp)
1690 struct mount_pcpu *mpcpu;
1691 int ref, lockref, writeopcount;
1694 printf("%s: mp %p vfs_ops %d\n", __func__, mp, mp->mnt_vfs_ops);
1699 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1700 printf("%d ", mpcpu->mntp_ref);
1701 ref += mpcpu->mntp_ref;
1704 printf(" lockref : ");
1705 lockref = mp->mnt_lockref;
1707 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1708 printf("%d ", mpcpu->mntp_lockref);
1709 lockref += mpcpu->mntp_lockref;
1712 printf("writeopcount: ");
1713 writeopcount = mp->mnt_writeopcount;
1715 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1716 printf("%d ", mpcpu->mntp_writeopcount);
1717 writeopcount += mpcpu->mntp_writeopcount;
1721 printf("counter struct total\n");
1722 printf("ref %-5d %-5d\n", mp->mnt_ref, ref);
1723 printf("lockref %-5d %-5d\n", mp->mnt_lockref, lockref);
1724 printf("writeopcount %-5d %-5d\n", mp->mnt_writeopcount, writeopcount);
1726 panic("invalid counts on struct mount");
1731 vfs_mount_fetch_counter(struct mount *mp, enum mount_counter which)
1733 struct mount_pcpu *mpcpu;
1740 case MNT_COUNT_LOCKREF:
1741 sum = mp->mnt_lockref;
1743 case MNT_COUNT_WRITEOPCOUNT:
1744 sum = mp->mnt_writeopcount;
1749 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1752 sum += mpcpu->mntp_ref;
1754 case MNT_COUNT_LOCKREF:
1755 sum += mpcpu->mntp_lockref;
1757 case MNT_COUNT_WRITEOPCOUNT:
1758 sum += mpcpu->mntp_writeopcount;
1766 * Do the actual filesystem unmount.
1769 dounmount(struct mount *mp, int flags, struct thread *td)
1771 struct vnode *coveredvp, *rootvp;
1773 uint64_t async_flag;
1776 if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
1777 mnt_gen_r = mp->mnt_gen;
1780 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
1782 * Check for mp being unmounted while waiting for the
1783 * covered vnode lock.
1785 if (coveredvp->v_mountedhere != mp ||
1786 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
1787 VOP_UNLOCK(coveredvp);
1795 * Only privileged root, or (if MNT_USER is set) the user that did the
1796 * original mount is permitted to unmount this filesystem.
1798 error = vfs_suser(mp, td);
1800 if (coveredvp != NULL) {
1801 VOP_UNLOCK(coveredvp);
1810 vn_start_write(NULL, &mp, V_WAIT | V_MNTREF);
1812 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 ||
1813 (mp->mnt_flag & MNT_UPDATE) != 0 ||
1814 !TAILQ_EMPTY(&mp->mnt_uppers)) {
1815 dounmount_cleanup(mp, coveredvp, 0);
1818 mp->mnt_kern_flag |= MNTK_UNMOUNT;
1819 rootvp = vfs_cache_root_clear(mp);
1820 if (coveredvp != NULL)
1821 vn_seqc_write_begin(coveredvp);
1822 if (flags & MNT_NONBUSY) {
1824 error = vfs_check_usecounts(mp);
1827 vn_seqc_write_end(coveredvp);
1828 dounmount_cleanup(mp, coveredvp, MNTK_UNMOUNT);
1829 if (rootvp != NULL) {
1830 vn_seqc_write_end(rootvp);
1836 /* Allow filesystems to detect that a forced unmount is in progress. */
1837 if (flags & MNT_FORCE) {
1838 mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1841 * Must be done after setting MNTK_UNMOUNTF and before
1842 * waiting for mnt_lockref to become 0.
1848 if (mp->mnt_lockref) {
1849 mp->mnt_kern_flag |= MNTK_DRAINING;
1850 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS,
1854 KASSERT(mp->mnt_lockref == 0,
1855 ("%s: invalid lock refcount in the drain path @ %s:%d",
1856 __func__, __FILE__, __LINE__));
1858 ("%s: invalid return value for msleep in the drain path @ %s:%d",
1859 __func__, __FILE__, __LINE__));
1862 * We want to keep the vnode around so that we can vn_seqc_write_end
1863 * after we are done with unmount. Downgrade our reference to a mere
1864 * hold count so that we don't interefere with anything.
1866 if (rootvp != NULL) {
1871 if (mp->mnt_flag & MNT_EXPUBLIC)
1872 vfs_setpublicfs(NULL, NULL, NULL);
1874 vfs_periodic(mp, MNT_WAIT);
1876 async_flag = mp->mnt_flag & MNT_ASYNC;
1877 mp->mnt_flag &= ~MNT_ASYNC;
1878 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1880 vfs_deallocate_syncvnode(mp);
1881 error = VFS_UNMOUNT(mp, flags);
1882 vn_finished_write(mp);
1884 * If we failed to flush the dirty blocks for this mount point,
1885 * undo all the cdir/rdir and rootvnode changes we made above.
1886 * Unless we failed to do so because the device is reporting that
1887 * it doesn't exist anymore.
1889 if (error && error != ENXIO) {
1891 if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1893 vfs_allocate_syncvnode(mp);
1896 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1897 mp->mnt_flag |= async_flag;
1898 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1899 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1900 mp->mnt_kern_flag |= MNTK_ASYNC;
1901 if (mp->mnt_kern_flag & MNTK_MWAIT) {
1902 mp->mnt_kern_flag &= ~MNTK_MWAIT;
1905 vfs_op_exit_locked(mp);
1908 vn_seqc_write_end(coveredvp);
1909 VOP_UNLOCK(coveredvp);
1912 if (rootvp != NULL) {
1913 vn_seqc_write_end(rootvp);
1918 mtx_lock(&mountlist_mtx);
1919 TAILQ_REMOVE(&mountlist, mp, mnt_list);
1920 mtx_unlock(&mountlist_mtx);
1921 EVENTHANDLER_DIRECT_INVOKE(vfs_unmounted, mp, td);
1922 if (coveredvp != NULL) {
1924 vn_irflag_unset_locked(coveredvp, VIRF_MOUNTPOINT);
1925 coveredvp->v_mountedhere = NULL;
1926 vn_seqc_write_end_locked(coveredvp);
1927 VI_UNLOCK(coveredvp);
1928 VOP_UNLOCK(coveredvp);
1931 mount_devctl_event("UNMOUNT", mp, false);
1932 if (rootvp != NULL) {
1933 vn_seqc_write_end(rootvp);
1936 vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1937 if (rootvnode != NULL && mp == rootvnode->v_mount) {
1941 if (mp == rootdevmp)
1943 vfs_mount_destroy(mp);
1948 * Report errors during filesystem mounting.
1951 vfs_mount_error(struct mount *mp, const char *fmt, ...)
1953 struct vfsoptlist *moptlist = mp->mnt_optnew;
1958 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1959 if (error || errmsg == NULL || len <= 0)
1963 vsnprintf(errmsg, (size_t)len, fmt, ap);
1968 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...)
1974 error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len);
1975 if (error || errmsg == NULL || len <= 0)
1979 vsnprintf(errmsg, (size_t)len, fmt, ap);
1984 * ---------------------------------------------------------------------
1985 * Functions for querying mount options/arguments from filesystems.
1989 * Check that no unknown options are given
1992 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1996 const char **t, *p, *q;
1999 TAILQ_FOREACH(opt, opts, link) {
2002 if (p[0] == 'n' && p[1] == 'o')
2004 for(t = global_opts; *t != NULL; t++) {
2005 if (strcmp(*t, p) == 0)
2008 if (strcmp(*t, q) == 0)
2014 for(t = legal; *t != NULL; t++) {
2015 if (strcmp(*t, p) == 0)
2018 if (strcmp(*t, q) == 0)
2024 snprintf(errmsg, sizeof(errmsg),
2025 "mount option <%s> is unknown", p);
2029 TAILQ_FOREACH(opt, opts, link) {
2030 if (strcmp(opt->name, "errmsg") == 0) {
2031 strncpy((char *)opt->value, errmsg, opt->len);
2036 printf("%s\n", errmsg);
2042 * Get a mount option by its name.
2044 * Return 0 if the option was found, ENOENT otherwise.
2045 * If len is non-NULL it will be filled with the length
2046 * of the option. If buf is non-NULL, it will be filled
2047 * with the address of the option.
2050 vfs_getopt(struct vfsoptlist *opts, const char *name, void **buf, int *len)
2054 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2056 TAILQ_FOREACH(opt, opts, link) {
2057 if (strcmp(name, opt->name) == 0) {
2070 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
2077 TAILQ_FOREACH(opt, opts, link) {
2078 if (strcmp(name, opt->name) == 0) {
2087 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value)
2089 char *opt_value, *vtp;
2093 error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len);
2096 if (opt_len == 0 || opt_value == NULL)
2098 if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0')
2100 iv = strtoq(opt_value, &vtp, 0);
2101 if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0'))
2128 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
2133 TAILQ_FOREACH(opt, opts, link) {
2134 if (strcmp(name, opt->name) != 0)
2137 if (opt->len == 0 ||
2138 ((char *)opt->value)[opt->len - 1] != '\0') {
2142 return (opt->value);
2149 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w,
2154 TAILQ_FOREACH(opt, opts, link) {
2155 if (strcmp(name, opt->name) == 0) {
2168 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
2174 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2176 TAILQ_FOREACH(opt, opts, link) {
2177 if (strcmp(name, opt->name) != 0)
2180 if (opt->len == 0 || opt->value == NULL)
2182 if (((char *)opt->value)[opt->len - 1] != '\0')
2185 ret = vsscanf(opt->value, fmt, ap);
2193 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len)
2197 TAILQ_FOREACH(opt, opts, link) {
2198 if (strcmp(name, opt->name) != 0)
2201 if (opt->value == NULL)
2204 if (opt->len != len)
2206 bcopy(value, opt->value, len);
2214 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len)
2218 TAILQ_FOREACH(opt, opts, link) {
2219 if (strcmp(name, opt->name) != 0)
2222 if (opt->value == NULL)
2228 bcopy(value, opt->value, len);
2236 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value)
2240 TAILQ_FOREACH(opt, opts, link) {
2241 if (strcmp(name, opt->name) != 0)
2244 if (opt->value == NULL)
2245 opt->len = strlen(value) + 1;
2246 else if (strlcpy(opt->value, value, opt->len) >= opt->len)
2254 * Find and copy a mount option.
2256 * The size of the buffer has to be specified
2257 * in len, if it is not the same length as the
2258 * mount option, EINVAL is returned.
2259 * Returns ENOENT if the option is not found.
2262 vfs_copyopt(struct vfsoptlist *opts, const char *name, void *dest, int len)
2266 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
2268 TAILQ_FOREACH(opt, opts, link) {
2269 if (strcmp(name, opt->name) == 0) {
2271 if (len != opt->len)
2273 bcopy(opt->value, dest, opt->len);
2281 __vfs_statfs(struct mount *mp, struct statfs *sbp)
2285 * Filesystems only fill in part of the structure for updates, we
2286 * have to read the entirety first to get all content.
2288 if (sbp != &mp->mnt_stat)
2289 memcpy(sbp, &mp->mnt_stat, sizeof(*sbp));
2292 * Set these in case the underlying filesystem fails to do so.
2294 sbp->f_version = STATFS_VERSION;
2295 sbp->f_namemax = NAME_MAX;
2296 sbp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
2298 return (mp->mnt_op->vfs_statfs(mp, sbp));
2302 vfs_mountedfrom(struct mount *mp, const char *from)
2305 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
2306 strlcpy(mp->mnt_stat.f_mntfromname, from,
2307 sizeof mp->mnt_stat.f_mntfromname);
2311 * ---------------------------------------------------------------------
2312 * This is the api for building mount args and mounting filesystems from
2313 * inside the kernel.
2315 * The API works by accumulation of individual args. First error is
2318 * XXX: should be documented in new manpage kernel_mount(9)
2321 /* A memory allocation which must be freed when we are done */
2323 SLIST_ENTRY(mntaarg) next;
2326 /* The header for the mount arguments */
2331 SLIST_HEAD(, mntaarg) list;
2335 * Add a boolean argument.
2337 * flag is the boolean value.
2338 * name must start with "no".
2341 mount_argb(struct mntarg *ma, int flag, const char *name)
2344 KASSERT(name[0] == 'n' && name[1] == 'o',
2345 ("mount_argb(...,%s): name must start with 'no'", name));
2347 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
2351 * Add an argument printf style
2354 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
2357 struct mntaarg *maa;
2362 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2363 SLIST_INIT(&ma->list);
2368 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2370 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2371 ma->v[ma->len].iov_len = strlen(name) + 1;
2374 sb = sbuf_new_auto();
2376 sbuf_vprintf(sb, fmt, ap);
2379 len = sbuf_len(sb) + 1;
2380 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2381 SLIST_INSERT_HEAD(&ma->list, maa, next);
2382 bcopy(sbuf_data(sb), maa + 1, len);
2385 ma->v[ma->len].iov_base = maa + 1;
2386 ma->v[ma->len].iov_len = len;
2393 * Add an argument which is a userland string.
2396 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2398 struct mntaarg *maa;
2404 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2405 SLIST_INIT(&ma->list);
2409 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2410 SLIST_INSERT_HEAD(&ma->list, maa, next);
2411 tbuf = (void *)(maa + 1);
2412 ma->error = copyinstr(val, tbuf, len, NULL);
2413 return (mount_arg(ma, name, tbuf, -1));
2419 * If length is -1, treat value as a C string.
2422 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2426 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2427 SLIST_INIT(&ma->list);
2432 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2434 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2435 ma->v[ma->len].iov_len = strlen(name) + 1;
2438 ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2440 ma->v[ma->len].iov_len = strlen(val) + 1;
2442 ma->v[ma->len].iov_len = len;
2448 * Free a mntarg structure
2451 free_mntarg(struct mntarg *ma)
2453 struct mntaarg *maa;
2455 while (!SLIST_EMPTY(&ma->list)) {
2456 maa = SLIST_FIRST(&ma->list);
2457 SLIST_REMOVE_HEAD(&ma->list, next);
2460 free(ma->v, M_MOUNT);
2465 * Mount a filesystem
2468 kernel_mount(struct mntarg *ma, uint64_t flags)
2473 KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2474 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
2475 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2477 auio.uio_iov = ma->v;
2478 auio.uio_iovcnt = ma->len;
2479 auio.uio_segflg = UIO_SYSSPACE;
2483 error = vfs_donmount(curthread, flags, &auio);
2489 * A printflike function to mount a filesystem.
2492 kernel_vmount(int flags, ...)
2494 struct mntarg *ma = NULL;
2500 va_start(ap, flags);
2502 cp = va_arg(ap, const char *);
2505 vp = va_arg(ap, const void *);
2506 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0));
2510 error = kernel_mount(ma, flags);
2514 /* Map from mount options to printable formats. */
2515 static struct mntoptnames optnames[] = {
2519 #define DEVCTL_LEN 1024
2521 mount_devctl_event(const char *type, struct mount *mp, bool donew)
2524 struct mntoptnames *fp;
2526 struct statfs *sfp = &mp->mnt_stat;
2529 buf = malloc(DEVCTL_LEN, M_MOUNT, M_NOWAIT);
2532 sbuf_new(&sb, buf, DEVCTL_LEN, SBUF_FIXEDLEN);
2533 sbuf_cpy(&sb, "mount-point=\"");
2534 devctl_safe_quote_sb(&sb, sfp->f_mntonname);
2535 sbuf_cat(&sb, "\" mount-dev=\"");
2536 devctl_safe_quote_sb(&sb, sfp->f_mntfromname);
2537 sbuf_cat(&sb, "\" mount-type=\"");
2538 devctl_safe_quote_sb(&sb, sfp->f_fstypename);
2539 sbuf_cat(&sb, "\" fsid=0x");
2540 cp = (const uint8_t *)&sfp->f_fsid.val[0];
2541 for (int i = 0; i < sizeof(sfp->f_fsid); i++)
2542 sbuf_printf(&sb, "%02x", cp[i]);
2543 sbuf_printf(&sb, " owner=%u flags=\"", sfp->f_owner);
2544 for (fp = optnames; fp->o_opt != 0; fp++) {
2545 if ((mp->mnt_flag & fp->o_opt) != 0) {
2546 sbuf_cat(&sb, fp->o_name);
2547 sbuf_putc(&sb, ';');
2550 sbuf_putc(&sb, '"');
2554 * Options are not published because the form of the options depends on
2555 * the file system and may include binary data. In addition, they don't
2556 * necessarily provide enough useful information to be actionable when
2557 * devd processes them.
2560 if (sbuf_error(&sb) == 0)
2561 devctl_notify("VFS", "FS", type, sbuf_data(&sb));
2567 * Force remount specified mount point to read-only. The argument
2568 * must be busied to avoid parallel unmount attempts.
2570 * Intended use is to prevent further writes if some metadata
2571 * inconsistency is detected. Note that the function still flushes
2572 * all cached metadata and data for the mount point, which might be
2573 * not always suitable.
2576 vfs_remount_ro(struct mount *mp)
2578 struct vfsoptlist *opts;
2580 struct vnode *vp_covered, *rootvp;
2583 KASSERT(mp->mnt_lockref > 0,
2584 ("vfs_remount_ro: mp %p is not busied", mp));
2585 KASSERT((mp->mnt_kern_flag & MNTK_UNMOUNT) == 0,
2586 ("vfs_remount_ro: mp %p is being unmounted (and busy?)", mp));
2589 vp_covered = mp->mnt_vnodecovered;
2590 error = vget(vp_covered, LK_EXCLUSIVE | LK_NOWAIT);
2593 VI_LOCK(vp_covered);
2594 if ((vp_covered->v_iflag & VI_MOUNT) != 0) {
2595 VI_UNLOCK(vp_covered);
2599 vp_covered->v_iflag |= VI_MOUNT;
2600 VI_UNLOCK(vp_covered);
2602 vn_seqc_write_begin(vp_covered);
2605 if ((mp->mnt_flag & MNT_RDONLY) != 0) {
2610 mp->mnt_flag |= MNT_UPDATE | MNT_FORCE | MNT_RDONLY;
2611 rootvp = vfs_cache_root_clear(mp);
2614 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK | M_ZERO);
2616 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK | M_ZERO);
2617 opt->name = strdup("ro", M_MOUNT);
2619 TAILQ_INSERT_TAIL(opts, opt, link);
2620 vfs_mergeopts(opts, mp->mnt_opt);
2621 mp->mnt_optnew = opts;
2623 error = VFS_MOUNT(mp);
2627 mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE);
2629 vfs_deallocate_syncvnode(mp);
2630 if (mp->mnt_opt != NULL)
2631 vfs_freeopts(mp->mnt_opt);
2632 mp->mnt_opt = mp->mnt_optnew;
2635 mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE | MNT_RDONLY);
2637 vfs_freeopts(mp->mnt_optnew);
2639 mp->mnt_optnew = NULL;
2643 VI_LOCK(vp_covered);
2644 vp_covered->v_iflag &= ~VI_MOUNT;
2645 VI_UNLOCK(vp_covered);
2647 vn_seqc_write_end(vp_covered);
2648 if (rootvp != NULL) {
2649 vn_seqc_write_end(rootvp);
2656 * Suspend write operations on all local writeable filesystems. Does
2657 * full sync of them in the process.
2659 * Iterate over the mount points in reverse order, suspending most
2660 * recently mounted filesystems first. It handles a case where a
2661 * filesystem mounted from a md(4) vnode-backed device should be
2662 * suspended before the filesystem that owns the vnode.
2665 suspend_all_fs(void)
2670 mtx_lock(&mountlist_mtx);
2671 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
2672 error = vfs_busy(mp, MBF_MNTLSTLOCK | MBF_NOWAIT);
2675 if ((mp->mnt_flag & (MNT_RDONLY | MNT_LOCAL)) != MNT_LOCAL ||
2676 (mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
2677 mtx_lock(&mountlist_mtx);
2681 error = vfs_write_suspend(mp, 0);
2684 MPASS((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0);
2685 mp->mnt_kern_flag |= MNTK_SUSPEND_ALL;
2687 mtx_lock(&mountlist_mtx);
2689 printf("suspend of %s failed, error %d\n",
2690 mp->mnt_stat.f_mntonname, error);
2691 mtx_lock(&mountlist_mtx);
2695 mtx_unlock(&mountlist_mtx);
2703 mtx_lock(&mountlist_mtx);
2704 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2705 if ((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0)
2707 mtx_unlock(&mountlist_mtx);
2709 MPASS((mp->mnt_kern_flag & MNTK_SUSPEND) != 0);
2710 mp->mnt_kern_flag &= ~MNTK_SUSPEND_ALL;
2712 vfs_write_resume(mp, 0);
2713 mtx_lock(&mountlist_mtx);
2716 mtx_unlock(&mountlist_mtx);