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/sysent.h>
66 #include <sys/systm.h>
67 #include <sys/vnode.h>
70 #include <geom/geom.h>
72 #include <machine/stdarg.h>
74 #include <security/audit/audit.h>
75 #include <security/mac/mac_framework.h>
77 #define VFS_MOUNTARG_SIZE_MAX (1024 * 64)
79 static int vfs_domount(struct thread *td, const char *fstype, char *fspath,
80 uint64_t fsflags, struct vfsoptlist **optlist);
81 static void free_mntarg(struct mntarg *ma);
83 static int usermount = 0;
84 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
85 "Unprivileged users may mount and unmount file systems");
87 static bool default_autoro = false;
88 SYSCTL_BOOL(_vfs, OID_AUTO, default_autoro, CTLFLAG_RW, &default_autoro, 0,
89 "Retry failed r/w mount as r/o if no explicit ro/rw option is specified");
91 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
92 MALLOC_DEFINE(M_STATFS, "statfs", "statfs structure");
93 static uma_zone_t mount_zone;
95 /* List of mounted filesystems. */
96 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
98 /* For any iteration/modification of mountlist */
99 struct mtx_padalign __exclusive_cache_line mountlist_mtx;
100 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF);
102 EVENTHANDLER_LIST_DEFINE(vfs_mounted);
103 EVENTHANDLER_LIST_DEFINE(vfs_unmounted);
105 static void mount_devctl_event(const char *type, struct mount *mp, bool donew);
108 * Global opts, taken by all filesystems
110 static const char *global_opts[] = {
122 mount_init(void *mem, int size, int flags)
126 mp = (struct mount *)mem;
127 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
128 mtx_init(&mp->mnt_listmtx, "struct mount vlist mtx", NULL, MTX_DEF);
129 lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0);
130 mp->mnt_pcpu = uma_zalloc_pcpu(pcpu_zone_16, M_WAITOK | M_ZERO);
133 mp->mnt_rootvnode = NULL;
138 mount_fini(void *mem, int size)
142 mp = (struct mount *)mem;
143 uma_zfree_pcpu(pcpu_zone_16, mp->mnt_pcpu);
144 lockdestroy(&mp->mnt_explock);
145 mtx_destroy(&mp->mnt_listmtx);
146 mtx_destroy(&mp->mnt_mtx);
150 vfs_mount_init(void *dummy __unused)
153 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), NULL,
154 NULL, mount_init, mount_fini, UMA_ALIGN_CACHE, UMA_ZONE_NOFREE);
156 SYSINIT(vfs_mount, SI_SUB_VFS, SI_ORDER_ANY, vfs_mount_init, NULL);
159 * ---------------------------------------------------------------------
160 * Functions for building and sanitizing the mount options
163 /* Remove one mount option. */
165 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
168 TAILQ_REMOVE(opts, opt, link);
169 free(opt->name, M_MOUNT);
170 if (opt->value != NULL)
171 free(opt->value, M_MOUNT);
175 /* Release all resources related to the mount options. */
177 vfs_freeopts(struct vfsoptlist *opts)
181 while (!TAILQ_EMPTY(opts)) {
182 opt = TAILQ_FIRST(opts);
183 vfs_freeopt(opts, opt);
189 vfs_deleteopt(struct vfsoptlist *opts, const char *name)
191 struct vfsopt *opt, *temp;
195 TAILQ_FOREACH_SAFE(opt, opts, link, temp) {
196 if (strcmp(opt->name, name) == 0)
197 vfs_freeopt(opts, opt);
202 vfs_isopt_ro(const char *opt)
205 if (strcmp(opt, "ro") == 0 || strcmp(opt, "rdonly") == 0 ||
206 strcmp(opt, "norw") == 0)
212 vfs_isopt_rw(const char *opt)
215 if (strcmp(opt, "rw") == 0 || strcmp(opt, "noro") == 0)
221 * Check if options are equal (with or without the "no" prefix).
224 vfs_equalopts(const char *opt1, const char *opt2)
228 /* "opt" vs. "opt" or "noopt" vs. "noopt" */
229 if (strcmp(opt1, opt2) == 0)
231 /* "noopt" vs. "opt" */
232 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
234 /* "opt" vs. "noopt" */
235 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
237 while ((p = strchr(opt1, '.')) != NULL &&
238 !strncmp(opt1, opt2, ++p - opt1)) {
241 /* "foo.noopt" vs. "foo.opt" */
242 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
244 /* "foo.opt" vs. "foo.noopt" */
245 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
248 /* "ro" / "rdonly" / "norw" / "rw" / "noro" */
249 if ((vfs_isopt_ro(opt1) || vfs_isopt_rw(opt1)) &&
250 (vfs_isopt_ro(opt2) || vfs_isopt_rw(opt2)))
256 * If a mount option is specified several times,
257 * (with or without the "no" prefix) only keep
258 * the last occurrence of it.
261 vfs_sanitizeopts(struct vfsoptlist *opts)
263 struct vfsopt *opt, *opt2, *tmp;
265 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
266 opt2 = TAILQ_PREV(opt, vfsoptlist, link);
267 while (opt2 != NULL) {
268 if (vfs_equalopts(opt->name, opt2->name)) {
269 tmp = TAILQ_PREV(opt2, vfsoptlist, link);
270 vfs_freeopt(opts, opt2);
273 opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
280 * Build a linked list of mount options from a struct uio.
283 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
285 struct vfsoptlist *opts;
287 size_t memused, namelen, optlen;
288 unsigned int i, iovcnt;
291 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
294 iovcnt = auio->uio_iovcnt;
295 for (i = 0; i < iovcnt; i += 2) {
296 namelen = auio->uio_iov[i].iov_len;
297 optlen = auio->uio_iov[i + 1].iov_len;
298 memused += sizeof(struct vfsopt) + optlen + namelen;
300 * Avoid consuming too much memory, and attempts to overflow
303 if (memused > VFS_MOUNTARG_SIZE_MAX ||
304 optlen > VFS_MOUNTARG_SIZE_MAX ||
305 namelen > VFS_MOUNTARG_SIZE_MAX) {
310 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
311 opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
318 * Do this early, so jumps to "bad" will free the current
321 TAILQ_INSERT_TAIL(opts, opt, link);
323 if (auio->uio_segflg == UIO_SYSSPACE) {
324 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
326 error = copyin(auio->uio_iov[i].iov_base, opt->name,
331 /* Ensure names are null-terminated strings. */
332 if (namelen == 0 || opt->name[namelen - 1] != '\0') {
338 opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
339 if (auio->uio_segflg == UIO_SYSSPACE) {
340 bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
343 error = copyin(auio->uio_iov[i + 1].iov_base,
350 vfs_sanitizeopts(opts);
359 * Merge the old mount options with the new ones passed
360 * in the MNT_UPDATE case.
362 * XXX: This function will keep a "nofoo" option in the new
363 * options. E.g, if the option's canonical name is "foo",
364 * "nofoo" ends up in the mount point's active options.
367 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *oldopts)
369 struct vfsopt *opt, *new;
371 TAILQ_FOREACH(opt, oldopts, link) {
372 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
373 new->name = strdup(opt->name, M_MOUNT);
375 new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
376 bcopy(opt->value, new->value, opt->len);
380 new->seen = opt->seen;
381 TAILQ_INSERT_HEAD(toopts, new, link);
383 vfs_sanitizeopts(toopts);
387 * Mount a filesystem.
389 #ifndef _SYS_SYSPROTO_H_
397 sys_nmount(struct thread *td, struct nmount_args *uap)
405 * Mount flags are now 64-bits. On 32-bit archtectures only
406 * 32-bits are passed in, but from here on everything handles
407 * 64-bit flags correctly.
411 AUDIT_ARG_FFLAGS(flags);
412 CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__,
413 uap->iovp, uap->iovcnt, flags);
416 * Filter out MNT_ROOTFS. We do not want clients of nmount() in
417 * userspace to set this flag, but we must filter it out if we want
418 * MNT_UPDATE on the root file system to work.
419 * MNT_ROOTFS should only be set by the kernel when mounting its
422 flags &= ~MNT_ROOTFS;
424 iovcnt = uap->iovcnt;
426 * Check that we have an even number of iovec's
427 * and that we have at least two options.
429 if ((iovcnt & 1) || (iovcnt < 4)) {
430 CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__,
435 error = copyinuio(uap->iovp, iovcnt, &auio);
437 CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno",
441 error = vfs_donmount(td, flags, auio);
448 * ---------------------------------------------------------------------
449 * Various utility functions
453 * Get a reference on a mount point from a vnode.
455 * The vnode is allowed to be passed unlocked and race against dooming. Note in
456 * such case there are no guarantees the referenced mount point will still be
457 * associated with it after the function returns.
460 vfs_ref_from_vp(struct vnode *vp)
463 struct mount_pcpu *mpcpu;
465 mp = atomic_load_ptr(&vp->v_mount);
466 if (__predict_false(mp == NULL)) {
469 if (vfs_op_thread_enter(mp, mpcpu)) {
470 if (__predict_true(mp == vp->v_mount)) {
471 vfs_mp_count_add_pcpu(mpcpu, ref, 1);
472 vfs_op_thread_exit(mp, mpcpu);
474 vfs_op_thread_exit(mp, mpcpu);
479 if (mp == vp->v_mount) {
491 vfs_ref(struct mount *mp)
493 struct mount_pcpu *mpcpu;
495 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
496 if (vfs_op_thread_enter(mp, mpcpu)) {
497 vfs_mp_count_add_pcpu(mpcpu, ref, 1);
498 vfs_op_thread_exit(mp, mpcpu);
508 vfs_rel(struct mount *mp)
510 struct mount_pcpu *mpcpu;
512 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
513 if (vfs_op_thread_enter(mp, mpcpu)) {
514 vfs_mp_count_sub_pcpu(mpcpu, ref, 1);
515 vfs_op_thread_exit(mp, mpcpu);
525 * Allocate and initialize the mount point struct.
528 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath,
533 mp = uma_zalloc(mount_zone, M_WAITOK);
534 bzero(&mp->mnt_startzero,
535 __rangeof(struct mount, mnt_startzero, mnt_endzero));
536 mp->mnt_kern_flag = 0;
538 mp->mnt_rootvnode = NULL;
539 mp->mnt_vnodecovered = NULL;
542 TAILQ_INIT(&mp->mnt_nvnodelist);
543 mp->mnt_nvnodelistsize = 0;
544 TAILQ_INIT(&mp->mnt_lazyvnodelist);
545 mp->mnt_lazyvnodelistsize = 0;
546 if (mp->mnt_ref != 0 || mp->mnt_lockref != 0 ||
547 mp->mnt_writeopcount != 0)
548 panic("%s: non-zero counters on new mp %p\n", __func__, mp);
549 if (mp->mnt_vfs_ops != 1)
550 panic("%s: vfs_ops should be 1 but %d found\n", __func__,
552 (void) vfs_busy(mp, MBF_NOWAIT);
553 atomic_add_acq_int(&vfsp->vfc_refcount, 1);
554 mp->mnt_op = vfsp->vfc_vfsops;
556 mp->mnt_stat.f_type = vfsp->vfc_typenum;
558 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
559 mp->mnt_vnodecovered = vp;
560 mp->mnt_cred = crdup(cred);
561 mp->mnt_stat.f_owner = cred->cr_uid;
562 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
563 mp->mnt_iosize_max = DFLTPHYS;
566 mac_mount_create(cred, mp);
568 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
569 TAILQ_INIT(&mp->mnt_uppers);
574 * Destroy the mount struct previously allocated by vfs_mount_alloc().
577 vfs_mount_destroy(struct mount *mp)
580 if (mp->mnt_vfs_ops == 0)
581 panic("%s: entered with zero vfs_ops\n", __func__);
583 vfs_assert_mount_counters(mp);
586 mp->mnt_kern_flag |= MNTK_REFEXPIRE;
587 if (mp->mnt_kern_flag & MNTK_MWAIT) {
588 mp->mnt_kern_flag &= ~MNTK_MWAIT;
592 msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0);
593 KASSERT(mp->mnt_ref == 0,
594 ("%s: invalid refcount in the drain path @ %s:%d", __func__,
595 __FILE__, __LINE__));
596 if (mp->mnt_writeopcount != 0)
597 panic("vfs_mount_destroy: nonzero writeopcount");
598 if (mp->mnt_secondary_writes != 0)
599 panic("vfs_mount_destroy: nonzero secondary_writes");
600 atomic_subtract_rel_int(&mp->mnt_vfc->vfc_refcount, 1);
601 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
604 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
605 vn_printf(vp, "dangling vnode ");
606 panic("unmount: dangling vnode");
608 KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers"));
609 if (mp->mnt_nvnodelistsize != 0)
610 panic("vfs_mount_destroy: nonzero nvnodelistsize");
611 if (mp->mnt_lazyvnodelistsize != 0)
612 panic("vfs_mount_destroy: nonzero lazyvnodelistsize");
613 if (mp->mnt_lockref != 0)
614 panic("vfs_mount_destroy: nonzero lock refcount");
617 if (mp->mnt_vfs_ops != 1)
618 panic("%s: vfs_ops should be 1 but %d found\n", __func__,
621 if (mp->mnt_rootvnode != NULL)
622 panic("%s: mount point still has a root vnode %p\n", __func__,
625 if (mp->mnt_vnodecovered != NULL)
626 vrele(mp->mnt_vnodecovered);
628 mac_mount_destroy(mp);
630 if (mp->mnt_opt != NULL)
631 vfs_freeopts(mp->mnt_opt);
632 crfree(mp->mnt_cred);
633 uma_zfree(mount_zone, mp);
637 vfs_should_downgrade_to_ro_mount(uint64_t fsflags, int error)
639 /* This is an upgrade of an exisiting mount. */
640 if ((fsflags & MNT_UPDATE) != 0)
642 /* This is already an R/O mount. */
643 if ((fsflags & MNT_RDONLY) != 0)
647 case ENODEV: /* generic, geom, ... */
648 case EACCES: /* cam/scsi, ... */
649 case EROFS: /* md, mmcsd, ... */
651 * These errors can be returned by the storage layer to signal
652 * that the media is read-only. No harm in the R/O mount
653 * attempt if the error was returned for some other reason.
662 vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions)
664 struct vfsoptlist *optlist;
665 struct vfsopt *opt, *tmp_opt;
666 char *fstype, *fspath, *errmsg;
667 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
670 errmsg = fspath = NULL;
671 errmsg_len = fspathlen = 0;
673 autoro = default_autoro;
675 error = vfs_buildopts(fsoptions, &optlist);
679 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
680 errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
683 * We need these two options before the others,
684 * and they are mandatory for any filesystem.
685 * Ensure they are NUL terminated as well.
688 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
689 if (error || fstypelen <= 0 || fstype[fstypelen - 1] != '\0') {
692 strncpy(errmsg, "Invalid fstype", errmsg_len);
696 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
697 if (error || fspathlen <= 0 || fspath[fspathlen - 1] != '\0') {
700 strncpy(errmsg, "Invalid fspath", errmsg_len);
705 * We need to see if we have the "update" option
706 * before we call vfs_domount(), since vfs_domount() has special
707 * logic based on MNT_UPDATE. This is very important
708 * when we want to update the root filesystem.
710 TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) {
713 if (strcmp(opt->name, "update") == 0) {
714 fsflags |= MNT_UPDATE;
717 else if (strcmp(opt->name, "async") == 0)
718 fsflags |= MNT_ASYNC;
719 else if (strcmp(opt->name, "force") == 0) {
720 fsflags |= MNT_FORCE;
723 else if (strcmp(opt->name, "reload") == 0) {
724 fsflags |= MNT_RELOAD;
727 else if (strcmp(opt->name, "multilabel") == 0)
728 fsflags |= MNT_MULTILABEL;
729 else if (strcmp(opt->name, "noasync") == 0)
730 fsflags &= ~MNT_ASYNC;
731 else if (strcmp(opt->name, "noatime") == 0)
732 fsflags |= MNT_NOATIME;
733 else if (strcmp(opt->name, "atime") == 0) {
734 free(opt->name, M_MOUNT);
735 opt->name = strdup("nonoatime", M_MOUNT);
737 else if (strcmp(opt->name, "noclusterr") == 0)
738 fsflags |= MNT_NOCLUSTERR;
739 else if (strcmp(opt->name, "clusterr") == 0) {
740 free(opt->name, M_MOUNT);
741 opt->name = strdup("nonoclusterr", M_MOUNT);
743 else if (strcmp(opt->name, "noclusterw") == 0)
744 fsflags |= MNT_NOCLUSTERW;
745 else if (strcmp(opt->name, "clusterw") == 0) {
746 free(opt->name, M_MOUNT);
747 opt->name = strdup("nonoclusterw", M_MOUNT);
749 else if (strcmp(opt->name, "noexec") == 0)
750 fsflags |= MNT_NOEXEC;
751 else if (strcmp(opt->name, "exec") == 0) {
752 free(opt->name, M_MOUNT);
753 opt->name = strdup("nonoexec", M_MOUNT);
755 else if (strcmp(opt->name, "nosuid") == 0)
756 fsflags |= MNT_NOSUID;
757 else if (strcmp(opt->name, "suid") == 0) {
758 free(opt->name, M_MOUNT);
759 opt->name = strdup("nonosuid", M_MOUNT);
761 else if (strcmp(opt->name, "nosymfollow") == 0)
762 fsflags |= MNT_NOSYMFOLLOW;
763 else if (strcmp(opt->name, "symfollow") == 0) {
764 free(opt->name, M_MOUNT);
765 opt->name = strdup("nonosymfollow", M_MOUNT);
767 else if (strcmp(opt->name, "noro") == 0) {
768 fsflags &= ~MNT_RDONLY;
771 else if (strcmp(opt->name, "rw") == 0) {
772 fsflags &= ~MNT_RDONLY;
775 else if (strcmp(opt->name, "ro") == 0) {
776 fsflags |= MNT_RDONLY;
779 else if (strcmp(opt->name, "rdonly") == 0) {
780 free(opt->name, M_MOUNT);
781 opt->name = strdup("ro", M_MOUNT);
782 fsflags |= MNT_RDONLY;
785 else if (strcmp(opt->name, "autoro") == 0) {
789 else if (strcmp(opt->name, "suiddir") == 0)
790 fsflags |= MNT_SUIDDIR;
791 else if (strcmp(opt->name, "sync") == 0)
792 fsflags |= MNT_SYNCHRONOUS;
793 else if (strcmp(opt->name, "union") == 0)
794 fsflags |= MNT_UNION;
795 else if (strcmp(opt->name, "automounted") == 0) {
796 fsflags |= MNT_AUTOMOUNTED;
798 } else if (strcmp(opt->name, "nocover") == 0) {
799 fsflags |= MNT_NOCOVER;
801 } else if (strcmp(opt->name, "cover") == 0) {
802 fsflags &= ~MNT_NOCOVER;
804 } else if (strcmp(opt->name, "emptydir") == 0) {
805 fsflags |= MNT_EMPTYDIR;
807 } else if (strcmp(opt->name, "noemptydir") == 0) {
808 fsflags &= ~MNT_EMPTYDIR;
812 vfs_freeopt(optlist, opt);
816 * Be ultra-paranoid about making sure the type and fspath
817 * variables will fit in our mp buffers, including the
820 if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) {
821 error = ENAMETOOLONG;
825 error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
826 if (error == ENOENT) {
829 strncpy(errmsg, "Invalid fstype", errmsg_len);
834 * See if we can mount in the read-only mode if the error code suggests
835 * that it could be possible and the mount options allow for that.
836 * Never try it if "[no]{ro|rw}" has been explicitly requested and not
837 * overridden by "autoro".
839 if (autoro && vfs_should_downgrade_to_ro_mount(fsflags, error)) {
840 printf("%s: R/W mount failed, possibly R/O media,"
841 " trying R/O mount\n", __func__);
842 fsflags |= MNT_RDONLY;
843 error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
846 /* copyout the errmsg */
847 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
848 && errmsg_len > 0 && errmsg != NULL) {
849 if (fsoptions->uio_segflg == UIO_SYSSPACE) {
851 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
852 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
855 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
856 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
861 vfs_freeopts(optlist);
868 #ifndef _SYS_SYSPROTO_H_
878 sys_mount(struct thread *td, struct mount_args *uap)
881 struct vfsconf *vfsp = NULL;
882 struct mntarg *ma = NULL;
887 * Mount flags are now 64-bits. On 32-bit architectures only
888 * 32-bits are passed in, but from here on everything handles
889 * 64-bit flags correctly.
893 AUDIT_ARG_FFLAGS(flags);
896 * Filter out MNT_ROOTFS. We do not want clients of mount() in
897 * userspace to set this flag, but we must filter it out if we want
898 * MNT_UPDATE on the root file system to work.
899 * MNT_ROOTFS should only be set by the kernel when mounting its
902 flags &= ~MNT_ROOTFS;
904 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
905 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
907 free(fstype, M_TEMP);
911 AUDIT_ARG_TEXT(fstype);
912 vfsp = vfs_byname_kld(fstype, td, &error);
913 free(fstype, M_TEMP);
916 if (((vfsp->vfc_flags & VFCF_SBDRY) != 0 &&
917 vfsp->vfc_vfsops_sd->vfs_cmount == NULL) ||
918 ((vfsp->vfc_flags & VFCF_SBDRY) == 0 &&
919 vfsp->vfc_vfsops->vfs_cmount == NULL))
922 ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN);
923 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
924 ma = mount_argb(ma, flags & MNT_RDONLY, "noro");
925 ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid");
926 ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec");
928 if ((vfsp->vfc_flags & VFCF_SBDRY) != 0)
929 return (vfsp->vfc_vfsops_sd->vfs_cmount(ma, uap->data, flags));
930 return (vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags));
934 * vfs_domount_first(): first file system mount (not update)
938 struct thread *td, /* Calling thread. */
939 struct vfsconf *vfsp, /* File system type. */
940 char *fspath, /* Mount path. */
941 struct vnode *vp, /* Vnode to be covered. */
942 uint64_t fsflags, /* Flags common to all filesystems. */
943 struct vfsoptlist **optlist /* Options local to the filesystem. */
948 struct vnode *newdp, *rootvp;
952 ASSERT_VOP_ELOCKED(vp, __func__);
953 KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here"));
956 * If the jail of the calling thread lacks permission for this type of
957 * file system, or is trying to cover its own root, deny immediately.
959 if (jailed(td->td_ucred) && (!prison_allow(td->td_ucred,
960 vfsp->vfc_prison_flag) || vp == td->td_ucred->cr_prison->pr_root)) {
966 * If the user is not root, ensure that they own the directory
967 * onto which we are attempting to mount.
969 error = VOP_GETATTR(vp, &va, td->td_ucred);
970 if (error == 0 && va.va_uid != td->td_ucred->cr_uid)
971 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN);
973 error = vinvalbuf(vp, V_SAVE, 0, 0);
974 if (error == 0 && vp->v_type != VDIR)
976 if (error == 0 && (fsflags & MNT_EMPTYDIR) != 0)
977 error = vfs_emptydir(vp);
980 if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL)
981 vp->v_iflag |= VI_MOUNT;
990 vn_seqc_write_begin(vp);
993 /* Allocate and initialize the filesystem. */
994 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred);
995 /* XXXMAC: pass to vfs_mount_alloc? */
996 mp->mnt_optnew = *optlist;
997 /* Set the mount level flags. */
998 mp->mnt_flag = (fsflags & (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY));
1001 * Mount the filesystem.
1002 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1003 * get. No freeing of cn_pnbuf.
1007 if ((error = VFS_MOUNT(mp)) != 0 ||
1008 (error1 = VFS_STATFS(mp, &mp->mnt_stat)) != 0 ||
1009 (error1 = VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) != 0) {
1013 rootvp = vfs_cache_root_clear(mp);
1014 if (rootvp != NULL) {
1018 (void)vn_start_write(NULL, &mp, V_WAIT);
1020 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_UNMOUNTF;
1023 error = VFS_UNMOUNT(mp, 0);
1024 vn_finished_write(mp);
1027 "failed post-mount (%d): rollback unmount returned %d\n",
1034 mp->mnt_vnodecovered = NULL;
1036 /* XXXKIB wait for mnt_lockref drain? */
1037 vfs_mount_destroy(mp);
1040 vp->v_iflag &= ~VI_MOUNT;
1042 if (rootvp != NULL) {
1043 vn_seqc_write_end(rootvp);
1046 vn_seqc_write_end(vp);
1050 vn_seqc_write_begin(newdp);
1053 if (mp->mnt_opt != NULL)
1054 vfs_freeopts(mp->mnt_opt);
1055 mp->mnt_opt = mp->mnt_optnew;
1059 * Prevent external consumers of mount options from reading mnt_optnew.
1061 mp->mnt_optnew = NULL;
1064 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1065 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1066 mp->mnt_kern_flag |= MNTK_ASYNC;
1068 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1071 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1074 vp->v_iflag &= ~VI_MOUNT;
1075 vn_irflag_set_locked(vp, VIRF_MOUNTPOINT);
1076 vp->v_mountedhere = mp;
1078 /* Place the new filesystem at the end of the mount list. */
1079 mtx_lock(&mountlist_mtx);
1080 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1081 mtx_unlock(&mountlist_mtx);
1082 vfs_event_signal(NULL, VQ_MOUNT, 0);
1083 vn_lock(newdp, LK_EXCLUSIVE | LK_RETRY);
1085 EVENTHANDLER_DIRECT_INVOKE(vfs_mounted, mp, newdp, td);
1087 mount_devctl_event("MOUNT", mp, false);
1088 mountcheckdirs(vp, newdp);
1089 vn_seqc_write_end(vp);
1090 vn_seqc_write_end(newdp);
1092 if ((mp->mnt_flag & MNT_RDONLY) == 0)
1093 vfs_allocate_syncvnode(mp);
1100 * vfs_domount_update(): update of mounted file system
1104 struct thread *td, /* Calling thread. */
1105 struct vnode *vp, /* Mount point vnode. */
1106 uint64_t fsflags, /* Flags common to all filesystems. */
1107 struct vfsoptlist **optlist /* Options local to the filesystem. */
1110 struct export_args export;
1111 struct o2export_args o2export;
1112 struct vnode *rootvp;
1115 int error, export_error, i, len;
1119 ASSERT_VOP_ELOCKED(vp, __func__);
1120 KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here"));
1123 if ((vp->v_vflag & VV_ROOT) == 0) {
1124 if (vfs_copyopt(*optlist, "export", &export, sizeof(export))
1134 * We only allow the filesystem to be reloaded if it
1135 * is currently mounted read-only.
1137 flag = mp->mnt_flag;
1138 if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) {
1140 return (EOPNOTSUPP); /* Needs translation */
1143 * Only privileged root, or (if MNT_USER is set) the user that
1144 * did the original mount is permitted to update it.
1146 error = vfs_suser(mp, td);
1151 if (vfs_busy(mp, MBF_NOWAIT)) {
1156 if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) {
1162 vp->v_iflag |= VI_MOUNT;
1167 vn_seqc_write_begin(vp);
1171 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1176 mp->mnt_flag &= ~MNT_UPDATEMASK;
1177 mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE |
1178 MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY);
1179 if ((mp->mnt_flag & MNT_ASYNC) == 0)
1180 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1181 rootvp = vfs_cache_root_clear(mp);
1183 mp->mnt_optnew = *optlist;
1184 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
1187 * Mount the filesystem.
1188 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1189 * get. No freeing of cn_pnbuf.
1191 error = VFS_MOUNT(mp);
1194 /* Process the export option. */
1195 if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp,
1197 /* Assume that there is only 1 ABI for each length. */
1199 case (sizeof(struct oexport_args)):
1200 bzero(&o2export, sizeof(o2export));
1202 case (sizeof(o2export)):
1203 bcopy(bufp, &o2export, len);
1204 export.ex_flags = (uint64_t)o2export.ex_flags;
1205 export.ex_root = o2export.ex_root;
1206 export.ex_uid = o2export.ex_anon.cr_uid;
1207 export.ex_groups = NULL;
1208 export.ex_ngroups = o2export.ex_anon.cr_ngroups;
1209 if (export.ex_ngroups > 0) {
1210 if (export.ex_ngroups <= XU_NGROUPS) {
1211 export.ex_groups = malloc(
1212 export.ex_ngroups * sizeof(gid_t),
1214 for (i = 0; i < export.ex_ngroups; i++)
1215 export.ex_groups[i] =
1216 o2export.ex_anon.cr_groups[i];
1218 export_error = EINVAL;
1219 } else if (export.ex_ngroups < 0)
1220 export_error = EINVAL;
1221 export.ex_addr = o2export.ex_addr;
1222 export.ex_addrlen = o2export.ex_addrlen;
1223 export.ex_mask = o2export.ex_mask;
1224 export.ex_masklen = o2export.ex_masklen;
1225 export.ex_indexfile = o2export.ex_indexfile;
1226 export.ex_numsecflavors = o2export.ex_numsecflavors;
1227 if (export.ex_numsecflavors < MAXSECFLAVORS) {
1228 for (i = 0; i < export.ex_numsecflavors; i++)
1229 export.ex_secflavors[i] =
1230 o2export.ex_secflavors[i];
1232 export_error = EINVAL;
1233 if (export_error == 0)
1234 export_error = vfs_export(mp, &export);
1235 free(export.ex_groups, M_TEMP);
1237 case (sizeof(export)):
1238 bcopy(bufp, &export, len);
1240 if (export.ex_ngroups > 0) {
1241 if (export.ex_ngroups <= NGROUPS_MAX) {
1242 grps = malloc(export.ex_ngroups *
1243 sizeof(gid_t), M_TEMP, M_WAITOK);
1244 export_error = copyin(export.ex_groups,
1245 grps, export.ex_ngroups *
1247 if (export_error == 0)
1248 export.ex_groups = grps;
1250 export_error = EINVAL;
1251 } else if (export.ex_ngroups == 0)
1252 export.ex_groups = NULL;
1254 export_error = EINVAL;
1255 if (export_error == 0)
1256 export_error = vfs_export(mp, &export);
1260 export_error = EINVAL;
1267 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE |
1271 * If we fail, restore old mount flags. MNT_QUOTA is special,
1272 * because it is not part of MNT_UPDATEMASK, but it could have
1273 * changed in the meantime if quotactl(2) was called.
1274 * All in all we want current value of MNT_QUOTA, not the old
1277 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA);
1279 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1280 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1281 mp->mnt_kern_flag |= MNTK_ASYNC;
1283 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1289 mount_devctl_event("REMOUNT", mp, true);
1290 if (mp->mnt_opt != NULL)
1291 vfs_freeopts(mp->mnt_opt);
1292 mp->mnt_opt = mp->mnt_optnew;
1294 (void)VFS_STATFS(mp, &mp->mnt_stat);
1296 * Prevent external consumers of mount options from reading
1299 mp->mnt_optnew = NULL;
1301 if ((mp->mnt_flag & MNT_RDONLY) == 0)
1302 vfs_allocate_syncvnode(mp);
1304 vfs_deallocate_syncvnode(mp);
1307 if (rootvp != NULL) {
1308 vn_seqc_write_end(rootvp);
1311 vn_seqc_write_end(vp);
1314 vp->v_iflag &= ~VI_MOUNT;
1317 return (error != 0 ? error : export_error);
1321 * vfs_domount(): actually attempt a filesystem mount.
1325 struct thread *td, /* Calling thread. */
1326 const char *fstype, /* Filesystem type. */
1327 char *fspath, /* Mount path. */
1328 uint64_t fsflags, /* Flags common to all filesystems. */
1329 struct vfsoptlist **optlist /* Options local to the filesystem. */
1332 struct vfsconf *vfsp;
1333 struct nameidata nd;
1339 * Be ultra-paranoid about making sure the type and fspath
1340 * variables will fit in our mp buffers, including the
1343 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
1344 return (ENAMETOOLONG);
1346 if (jailed(td->td_ucred) || usermount == 0) {
1347 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
1352 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
1354 if (fsflags & MNT_EXPORTED) {
1355 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
1359 if (fsflags & MNT_SUIDDIR) {
1360 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
1365 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
1367 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
1368 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
1369 fsflags |= MNT_NOSUID | MNT_USER;
1372 /* Load KLDs before we lock the covered vnode to avoid reversals. */
1374 if ((fsflags & MNT_UPDATE) == 0) {
1375 /* Don't try to load KLDs if we're mounting the root. */
1376 if (fsflags & MNT_ROOTFS) {
1377 if ((vfsp = vfs_byname(fstype)) == NULL)
1380 if ((vfsp = vfs_byname_kld(fstype, td, &error)) == NULL)
1386 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE.
1388 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1389 UIO_SYSSPACE, fspath, td);
1393 NDFREE(&nd, NDF_ONLY_PNBUF);
1395 if ((fsflags & MNT_UPDATE) == 0) {
1396 if ((vp->v_vflag & VV_ROOT) != 0 &&
1397 (fsflags & MNT_NOCOVER) != 0) {
1401 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1402 strcpy(pathbuf, fspath);
1403 error = vn_path_to_global_path(td, vp, pathbuf, MNAMELEN);
1405 error = vfs_domount_first(td, vfsp, pathbuf, vp,
1408 free(pathbuf, M_TEMP);
1410 error = vfs_domount_update(td, vp, fsflags, optlist);
1416 * Unmount a filesystem.
1418 * Note: unmount takes a path to the vnode mounted on as argument, not
1419 * special file (as before).
1421 #ifndef _SYS_SYSPROTO_H_
1422 struct unmount_args {
1429 sys_unmount(struct thread *td, struct unmount_args *uap)
1432 return (kern_unmount(td, uap->path, uap->flags));
1436 kern_unmount(struct thread *td, const char *path, int flags)
1438 struct nameidata nd;
1441 int error, id0, id1;
1443 AUDIT_ARG_VALUE(flags);
1444 if (jailed(td->td_ucred) || usermount == 0) {
1445 error = priv_check(td, PRIV_VFS_UNMOUNT);
1450 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1451 error = copyinstr(path, pathbuf, MNAMELEN, NULL);
1453 free(pathbuf, M_TEMP);
1456 if (flags & MNT_BYFSID) {
1457 AUDIT_ARG_TEXT(pathbuf);
1458 /* Decode the filesystem ID. */
1459 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1460 free(pathbuf, M_TEMP);
1464 mtx_lock(&mountlist_mtx);
1465 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1466 if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1467 mp->mnt_stat.f_fsid.val[1] == id1) {
1472 mtx_unlock(&mountlist_mtx);
1475 * Try to find global path for path argument.
1477 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1478 UIO_SYSSPACE, pathbuf, td);
1479 if (namei(&nd) == 0) {
1480 NDFREE(&nd, NDF_ONLY_PNBUF);
1481 error = vn_path_to_global_path(td, nd.ni_vp, pathbuf,
1486 mtx_lock(&mountlist_mtx);
1487 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1488 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) {
1493 mtx_unlock(&mountlist_mtx);
1495 free(pathbuf, M_TEMP);
1498 * Previously we returned ENOENT for a nonexistent path and
1499 * EINVAL for a non-mountpoint. We cannot tell these apart
1500 * now, so in the !MNT_BYFSID case return the more likely
1501 * EINVAL for compatibility.
1503 return ((flags & MNT_BYFSID) ? ENOENT : EINVAL);
1507 * Don't allow unmounting the root filesystem.
1509 if (mp->mnt_flag & MNT_ROOTFS) {
1513 error = dounmount(mp, flags, td);
1518 * Return error if any of the vnodes, ignoring the root vnode
1519 * and the syncer vnode, have non-zero usecount.
1521 * This function is purely advisory - it can return false positives
1525 vfs_check_usecounts(struct mount *mp)
1527 struct vnode *vp, *mvp;
1529 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1530 if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON &&
1531 vp->v_usecount != 0) {
1533 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1543 dounmount_cleanup(struct mount *mp, struct vnode *coveredvp, int mntkflags)
1546 mtx_assert(MNT_MTX(mp), MA_OWNED);
1547 mp->mnt_kern_flag &= ~mntkflags;
1548 if ((mp->mnt_kern_flag & MNTK_MWAIT) != 0) {
1549 mp->mnt_kern_flag &= ~MNTK_MWAIT;
1552 vfs_op_exit_locked(mp);
1554 if (coveredvp != NULL) {
1555 VOP_UNLOCK(coveredvp);
1558 vn_finished_write(mp);
1562 * There are various reference counters associated with the mount point.
1563 * Normally it is permitted to modify them without taking the mnt ilock,
1564 * but this behavior can be temporarily disabled if stable value is needed
1565 * or callers are expected to block (e.g. to not allow new users during
1569 vfs_op_enter(struct mount *mp)
1571 struct mount_pcpu *mpcpu;
1576 if (mp->mnt_vfs_ops > 1) {
1580 vfs_op_barrier_wait(mp);
1582 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1584 mp->mnt_ref += mpcpu->mntp_ref;
1585 mpcpu->mntp_ref = 0;
1587 mp->mnt_lockref += mpcpu->mntp_lockref;
1588 mpcpu->mntp_lockref = 0;
1590 mp->mnt_writeopcount += mpcpu->mntp_writeopcount;
1591 mpcpu->mntp_writeopcount = 0;
1593 if (mp->mnt_ref <= 0 || mp->mnt_lockref < 0 || mp->mnt_writeopcount < 0)
1594 panic("%s: invalid count(s) on mp %p: ref %d lockref %d writeopcount %d\n",
1595 __func__, mp, mp->mnt_ref, mp->mnt_lockref, mp->mnt_writeopcount);
1597 vfs_assert_mount_counters(mp);
1601 vfs_op_exit_locked(struct mount *mp)
1604 mtx_assert(MNT_MTX(mp), MA_OWNED);
1606 if (mp->mnt_vfs_ops <= 0)
1607 panic("%s: invalid vfs_ops count %d for mp %p\n",
1608 __func__, mp->mnt_vfs_ops, mp);
1613 vfs_op_exit(struct mount *mp)
1617 vfs_op_exit_locked(mp);
1621 struct vfs_op_barrier_ipi {
1623 struct smp_rendezvous_cpus_retry_arg srcra;
1627 vfs_op_action_func(void *arg)
1629 struct vfs_op_barrier_ipi *vfsopipi;
1632 vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1635 if (!vfs_op_thread_entered(mp))
1636 smp_rendezvous_cpus_done(arg);
1640 vfs_op_wait_func(void *arg, int cpu)
1642 struct vfs_op_barrier_ipi *vfsopipi;
1644 struct mount_pcpu *mpcpu;
1646 vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1649 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1650 while (atomic_load_int(&mpcpu->mntp_thread_in_ops))
1655 vfs_op_barrier_wait(struct mount *mp)
1657 struct vfs_op_barrier_ipi vfsopipi;
1661 smp_rendezvous_cpus_retry(all_cpus,
1662 smp_no_rendezvous_barrier,
1664 smp_no_rendezvous_barrier,
1671 vfs_assert_mount_counters(struct mount *mp)
1673 struct mount_pcpu *mpcpu;
1676 if (mp->mnt_vfs_ops == 0)
1680 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1681 if (mpcpu->mntp_ref != 0 ||
1682 mpcpu->mntp_lockref != 0 ||
1683 mpcpu->mntp_writeopcount != 0)
1684 vfs_dump_mount_counters(mp);
1689 vfs_dump_mount_counters(struct mount *mp)
1691 struct mount_pcpu *mpcpu;
1692 int ref, lockref, writeopcount;
1695 printf("%s: mp %p vfs_ops %d\n", __func__, mp, mp->mnt_vfs_ops);
1700 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1701 printf("%d ", mpcpu->mntp_ref);
1702 ref += mpcpu->mntp_ref;
1705 printf(" lockref : ");
1706 lockref = mp->mnt_lockref;
1708 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1709 printf("%d ", mpcpu->mntp_lockref);
1710 lockref += mpcpu->mntp_lockref;
1713 printf("writeopcount: ");
1714 writeopcount = mp->mnt_writeopcount;
1716 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1717 printf("%d ", mpcpu->mntp_writeopcount);
1718 writeopcount += mpcpu->mntp_writeopcount;
1722 printf("counter struct total\n");
1723 printf("ref %-5d %-5d\n", mp->mnt_ref, ref);
1724 printf("lockref %-5d %-5d\n", mp->mnt_lockref, lockref);
1725 printf("writeopcount %-5d %-5d\n", mp->mnt_writeopcount, writeopcount);
1727 panic("invalid counts on struct mount");
1732 vfs_mount_fetch_counter(struct mount *mp, enum mount_counter which)
1734 struct mount_pcpu *mpcpu;
1741 case MNT_COUNT_LOCKREF:
1742 sum = mp->mnt_lockref;
1744 case MNT_COUNT_WRITEOPCOUNT:
1745 sum = mp->mnt_writeopcount;
1750 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1753 sum += mpcpu->mntp_ref;
1755 case MNT_COUNT_LOCKREF:
1756 sum += mpcpu->mntp_lockref;
1758 case MNT_COUNT_WRITEOPCOUNT:
1759 sum += mpcpu->mntp_writeopcount;
1767 * Do the actual filesystem unmount.
1770 dounmount(struct mount *mp, int flags, struct thread *td)
1772 struct vnode *coveredvp, *rootvp;
1774 uint64_t async_flag;
1777 if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
1778 mnt_gen_r = mp->mnt_gen;
1781 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
1783 * Check for mp being unmounted while waiting for the
1784 * covered vnode lock.
1786 if (coveredvp->v_mountedhere != mp ||
1787 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
1788 VOP_UNLOCK(coveredvp);
1796 * Only privileged root, or (if MNT_USER is set) the user that did the
1797 * original mount is permitted to unmount this filesystem.
1799 error = vfs_suser(mp, td);
1801 if (coveredvp != NULL) {
1802 VOP_UNLOCK(coveredvp);
1811 vn_start_write(NULL, &mp, V_WAIT | V_MNTREF);
1813 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 ||
1814 (mp->mnt_flag & MNT_UPDATE) != 0 ||
1815 !TAILQ_EMPTY(&mp->mnt_uppers)) {
1816 dounmount_cleanup(mp, coveredvp, 0);
1819 mp->mnt_kern_flag |= MNTK_UNMOUNT;
1820 rootvp = vfs_cache_root_clear(mp);
1821 if (coveredvp != NULL)
1822 vn_seqc_write_begin(coveredvp);
1823 if (flags & MNT_NONBUSY) {
1825 error = vfs_check_usecounts(mp);
1828 vn_seqc_write_end(coveredvp);
1829 dounmount_cleanup(mp, coveredvp, MNTK_UNMOUNT);
1830 if (rootvp != NULL) {
1831 vn_seqc_write_end(rootvp);
1837 /* Allow filesystems to detect that a forced unmount is in progress. */
1838 if (flags & MNT_FORCE) {
1839 mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1842 * Must be done after setting MNTK_UNMOUNTF and before
1843 * waiting for mnt_lockref to become 0.
1849 if (mp->mnt_lockref) {
1850 mp->mnt_kern_flag |= MNTK_DRAINING;
1851 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS,
1855 KASSERT(mp->mnt_lockref == 0,
1856 ("%s: invalid lock refcount in the drain path @ %s:%d",
1857 __func__, __FILE__, __LINE__));
1859 ("%s: invalid return value for msleep in the drain path @ %s:%d",
1860 __func__, __FILE__, __LINE__));
1863 * We want to keep the vnode around so that we can vn_seqc_write_end
1864 * after we are done with unmount. Downgrade our reference to a mere
1865 * hold count so that we don't interefere with anything.
1867 if (rootvp != NULL) {
1872 if (mp->mnt_flag & MNT_EXPUBLIC)
1873 vfs_setpublicfs(NULL, NULL, NULL);
1875 vfs_periodic(mp, MNT_WAIT);
1877 async_flag = mp->mnt_flag & MNT_ASYNC;
1878 mp->mnt_flag &= ~MNT_ASYNC;
1879 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1881 vfs_deallocate_syncvnode(mp);
1882 error = VFS_UNMOUNT(mp, flags);
1883 vn_finished_write(mp);
1885 * If we failed to flush the dirty blocks for this mount point,
1886 * undo all the cdir/rdir and rootvnode changes we made above.
1887 * Unless we failed to do so because the device is reporting that
1888 * it doesn't exist anymore.
1890 if (error && error != ENXIO) {
1892 if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1894 vfs_allocate_syncvnode(mp);
1897 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1898 mp->mnt_flag |= async_flag;
1899 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1900 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1901 mp->mnt_kern_flag |= MNTK_ASYNC;
1902 if (mp->mnt_kern_flag & MNTK_MWAIT) {
1903 mp->mnt_kern_flag &= ~MNTK_MWAIT;
1906 vfs_op_exit_locked(mp);
1909 vn_seqc_write_end(coveredvp);
1910 VOP_UNLOCK(coveredvp);
1913 if (rootvp != NULL) {
1914 vn_seqc_write_end(rootvp);
1919 mtx_lock(&mountlist_mtx);
1920 TAILQ_REMOVE(&mountlist, mp, mnt_list);
1921 mtx_unlock(&mountlist_mtx);
1922 EVENTHANDLER_DIRECT_INVOKE(vfs_unmounted, mp, td);
1923 if (coveredvp != NULL) {
1925 vn_irflag_unset_locked(coveredvp, VIRF_MOUNTPOINT);
1926 coveredvp->v_mountedhere = NULL;
1927 vn_seqc_write_end_locked(coveredvp);
1928 VI_UNLOCK(coveredvp);
1929 VOP_UNLOCK(coveredvp);
1932 mount_devctl_event("UNMOUNT", mp, false);
1933 if (rootvp != NULL) {
1934 vn_seqc_write_end(rootvp);
1937 vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1938 if (rootvnode != NULL && mp == rootvnode->v_mount) {
1942 if (mp == rootdevmp)
1944 vfs_mount_destroy(mp);
1949 * Report errors during filesystem mounting.
1952 vfs_mount_error(struct mount *mp, const char *fmt, ...)
1954 struct vfsoptlist *moptlist = mp->mnt_optnew;
1959 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1960 if (error || errmsg == NULL || len <= 0)
1964 vsnprintf(errmsg, (size_t)len, fmt, ap);
1969 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...)
1975 error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len);
1976 if (error || errmsg == NULL || len <= 0)
1980 vsnprintf(errmsg, (size_t)len, fmt, ap);
1985 * ---------------------------------------------------------------------
1986 * Functions for querying mount options/arguments from filesystems.
1990 * Check that no unknown options are given
1993 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1997 const char **t, *p, *q;
2000 TAILQ_FOREACH(opt, opts, link) {
2003 if (p[0] == 'n' && p[1] == 'o')
2005 for(t = global_opts; *t != NULL; t++) {
2006 if (strcmp(*t, p) == 0)
2009 if (strcmp(*t, q) == 0)
2015 for(t = legal; *t != NULL; t++) {
2016 if (strcmp(*t, p) == 0)
2019 if (strcmp(*t, q) == 0)
2025 snprintf(errmsg, sizeof(errmsg),
2026 "mount option <%s> is unknown", p);
2030 TAILQ_FOREACH(opt, opts, link) {
2031 if (strcmp(opt->name, "errmsg") == 0) {
2032 strncpy((char *)opt->value, errmsg, opt->len);
2037 printf("%s\n", errmsg);
2043 * Get a mount option by its name.
2045 * Return 0 if the option was found, ENOENT otherwise.
2046 * If len is non-NULL it will be filled with the length
2047 * of the option. If buf is non-NULL, it will be filled
2048 * with the address of the option.
2051 vfs_getopt(struct vfsoptlist *opts, const char *name, void **buf, int *len)
2055 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2057 TAILQ_FOREACH(opt, opts, link) {
2058 if (strcmp(name, opt->name) == 0) {
2071 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
2078 TAILQ_FOREACH(opt, opts, link) {
2079 if (strcmp(name, opt->name) == 0) {
2088 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value)
2090 char *opt_value, *vtp;
2094 error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len);
2097 if (opt_len == 0 || opt_value == NULL)
2099 if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0')
2101 iv = strtoq(opt_value, &vtp, 0);
2102 if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0'))
2129 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
2134 TAILQ_FOREACH(opt, opts, link) {
2135 if (strcmp(name, opt->name) != 0)
2138 if (opt->len == 0 ||
2139 ((char *)opt->value)[opt->len - 1] != '\0') {
2143 return (opt->value);
2150 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w,
2155 TAILQ_FOREACH(opt, opts, link) {
2156 if (strcmp(name, opt->name) == 0) {
2169 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
2175 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2177 TAILQ_FOREACH(opt, opts, link) {
2178 if (strcmp(name, opt->name) != 0)
2181 if (opt->len == 0 || opt->value == NULL)
2183 if (((char *)opt->value)[opt->len - 1] != '\0')
2186 ret = vsscanf(opt->value, fmt, ap);
2194 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len)
2198 TAILQ_FOREACH(opt, opts, link) {
2199 if (strcmp(name, opt->name) != 0)
2202 if (opt->value == NULL)
2205 if (opt->len != len)
2207 bcopy(value, opt->value, len);
2215 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len)
2219 TAILQ_FOREACH(opt, opts, link) {
2220 if (strcmp(name, opt->name) != 0)
2223 if (opt->value == NULL)
2229 bcopy(value, opt->value, len);
2237 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value)
2241 TAILQ_FOREACH(opt, opts, link) {
2242 if (strcmp(name, opt->name) != 0)
2245 if (opt->value == NULL)
2246 opt->len = strlen(value) + 1;
2247 else if (strlcpy(opt->value, value, opt->len) >= opt->len)
2255 * Find and copy a mount option.
2257 * The size of the buffer has to be specified
2258 * in len, if it is not the same length as the
2259 * mount option, EINVAL is returned.
2260 * Returns ENOENT if the option is not found.
2263 vfs_copyopt(struct vfsoptlist *opts, const char *name, void *dest, int len)
2267 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
2269 TAILQ_FOREACH(opt, opts, link) {
2270 if (strcmp(name, opt->name) == 0) {
2272 if (len != opt->len)
2274 bcopy(opt->value, dest, opt->len);
2282 __vfs_statfs(struct mount *mp, struct statfs *sbp)
2286 * Filesystems only fill in part of the structure for updates, we
2287 * have to read the entirety first to get all content.
2289 if (sbp != &mp->mnt_stat)
2290 memcpy(sbp, &mp->mnt_stat, sizeof(*sbp));
2293 * Set these in case the underlying filesystem fails to do so.
2295 sbp->f_version = STATFS_VERSION;
2296 sbp->f_namemax = NAME_MAX;
2297 sbp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
2299 return (mp->mnt_op->vfs_statfs(mp, sbp));
2303 vfs_mountedfrom(struct mount *mp, const char *from)
2306 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
2307 strlcpy(mp->mnt_stat.f_mntfromname, from,
2308 sizeof mp->mnt_stat.f_mntfromname);
2312 * ---------------------------------------------------------------------
2313 * This is the api for building mount args and mounting filesystems from
2314 * inside the kernel.
2316 * The API works by accumulation of individual args. First error is
2319 * XXX: should be documented in new manpage kernel_mount(9)
2322 /* A memory allocation which must be freed when we are done */
2324 SLIST_ENTRY(mntaarg) next;
2327 /* The header for the mount arguments */
2332 SLIST_HEAD(, mntaarg) list;
2336 * Add a boolean argument.
2338 * flag is the boolean value.
2339 * name must start with "no".
2342 mount_argb(struct mntarg *ma, int flag, const char *name)
2345 KASSERT(name[0] == 'n' && name[1] == 'o',
2346 ("mount_argb(...,%s): name must start with 'no'", name));
2348 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
2352 * Add an argument printf style
2355 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
2358 struct mntaarg *maa;
2363 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2364 SLIST_INIT(&ma->list);
2369 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2371 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2372 ma->v[ma->len].iov_len = strlen(name) + 1;
2375 sb = sbuf_new_auto();
2377 sbuf_vprintf(sb, fmt, ap);
2380 len = sbuf_len(sb) + 1;
2381 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2382 SLIST_INSERT_HEAD(&ma->list, maa, next);
2383 bcopy(sbuf_data(sb), maa + 1, len);
2386 ma->v[ma->len].iov_base = maa + 1;
2387 ma->v[ma->len].iov_len = len;
2394 * Add an argument which is a userland string.
2397 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2399 struct mntaarg *maa;
2405 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2406 SLIST_INIT(&ma->list);
2410 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2411 SLIST_INSERT_HEAD(&ma->list, maa, next);
2412 tbuf = (void *)(maa + 1);
2413 ma->error = copyinstr(val, tbuf, len, NULL);
2414 return (mount_arg(ma, name, tbuf, -1));
2420 * If length is -1, treat value as a C string.
2423 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2427 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2428 SLIST_INIT(&ma->list);
2433 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2435 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2436 ma->v[ma->len].iov_len = strlen(name) + 1;
2439 ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2441 ma->v[ma->len].iov_len = strlen(val) + 1;
2443 ma->v[ma->len].iov_len = len;
2449 * Free a mntarg structure
2452 free_mntarg(struct mntarg *ma)
2454 struct mntaarg *maa;
2456 while (!SLIST_EMPTY(&ma->list)) {
2457 maa = SLIST_FIRST(&ma->list);
2458 SLIST_REMOVE_HEAD(&ma->list, next);
2461 free(ma->v, M_MOUNT);
2466 * Mount a filesystem
2469 kernel_mount(struct mntarg *ma, uint64_t flags)
2474 KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2475 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
2476 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2478 auio.uio_iov = ma->v;
2479 auio.uio_iovcnt = ma->len;
2480 auio.uio_segflg = UIO_SYSSPACE;
2484 error = vfs_donmount(curthread, flags, &auio);
2490 * A printflike function to mount a filesystem.
2493 kernel_vmount(int flags, ...)
2495 struct mntarg *ma = NULL;
2501 va_start(ap, flags);
2503 cp = va_arg(ap, const char *);
2506 vp = va_arg(ap, const void *);
2507 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0));
2511 error = kernel_mount(ma, flags);
2515 /* Map from mount options to printable formats. */
2516 static struct mntoptnames optnames[] = {
2520 #define DEVCTL_LEN 1024
2522 mount_devctl_event(const char *type, struct mount *mp, bool donew)
2525 struct mntoptnames *fp;
2527 struct statfs *sfp = &mp->mnt_stat;
2530 buf = malloc(DEVCTL_LEN, M_MOUNT, M_NOWAIT);
2533 sbuf_new(&sb, buf, DEVCTL_LEN, SBUF_FIXEDLEN);
2534 sbuf_cpy(&sb, "mount-point=\"");
2535 devctl_safe_quote_sb(&sb, sfp->f_mntonname);
2536 sbuf_cat(&sb, "\" mount-dev=\"");
2537 devctl_safe_quote_sb(&sb, sfp->f_mntfromname);
2538 sbuf_cat(&sb, "\" mount-type=\"");
2539 devctl_safe_quote_sb(&sb, sfp->f_fstypename);
2540 sbuf_cat(&sb, "\" fsid=0x");
2541 cp = (const uint8_t *)&sfp->f_fsid.val[0];
2542 for (int i = 0; i < sizeof(sfp->f_fsid); i++)
2543 sbuf_printf(&sb, "%02x", cp[i]);
2544 sbuf_printf(&sb, " owner=%u flags=\"", sfp->f_owner);
2545 for (fp = optnames; fp->o_opt != 0; fp++) {
2546 if ((mp->mnt_flag & fp->o_opt) != 0) {
2547 sbuf_cat(&sb, fp->o_name);
2548 sbuf_putc(&sb, ';');
2551 sbuf_putc(&sb, '"');
2555 * Options are not published because the form of the options depends on
2556 * the file system and may include binary data. In addition, they don't
2557 * necessarily provide enough useful information to be actionable when
2558 * devd processes them.
2561 if (sbuf_error(&sb) == 0)
2562 devctl_notify("VFS", "FS", type, sbuf_data(&sb));
2568 * Force remount specified mount point to read-only. The argument
2569 * must be busied to avoid parallel unmount attempts.
2571 * Intended use is to prevent further writes if some metadata
2572 * inconsistency is detected. Note that the function still flushes
2573 * all cached metadata and data for the mount point, which might be
2574 * not always suitable.
2577 vfs_remount_ro(struct mount *mp)
2579 struct vfsoptlist *opts;
2581 struct vnode *vp_covered, *rootvp;
2584 KASSERT(mp->mnt_lockref > 0,
2585 ("vfs_remount_ro: mp %p is not busied", mp));
2586 KASSERT((mp->mnt_kern_flag & MNTK_UNMOUNT) == 0,
2587 ("vfs_remount_ro: mp %p is being unmounted (and busy?)", mp));
2590 vp_covered = mp->mnt_vnodecovered;
2591 error = vget(vp_covered, LK_EXCLUSIVE | LK_NOWAIT);
2594 VI_LOCK(vp_covered);
2595 if ((vp_covered->v_iflag & VI_MOUNT) != 0) {
2596 VI_UNLOCK(vp_covered);
2600 vp_covered->v_iflag |= VI_MOUNT;
2601 VI_UNLOCK(vp_covered);
2603 vn_seqc_write_begin(vp_covered);
2606 if ((mp->mnt_flag & MNT_RDONLY) != 0) {
2611 mp->mnt_flag |= MNT_UPDATE | MNT_FORCE | MNT_RDONLY;
2612 rootvp = vfs_cache_root_clear(mp);
2615 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK | M_ZERO);
2617 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK | M_ZERO);
2618 opt->name = strdup("ro", M_MOUNT);
2620 TAILQ_INSERT_TAIL(opts, opt, link);
2621 vfs_mergeopts(opts, mp->mnt_opt);
2622 mp->mnt_optnew = opts;
2624 error = VFS_MOUNT(mp);
2628 mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE);
2630 vfs_deallocate_syncvnode(mp);
2631 if (mp->mnt_opt != NULL)
2632 vfs_freeopts(mp->mnt_opt);
2633 mp->mnt_opt = mp->mnt_optnew;
2636 mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE | MNT_RDONLY);
2638 vfs_freeopts(mp->mnt_optnew);
2640 mp->mnt_optnew = NULL;
2644 VI_LOCK(vp_covered);
2645 vp_covered->v_iflag &= ~VI_MOUNT;
2646 VI_UNLOCK(vp_covered);
2648 vn_seqc_write_end(vp_covered);
2649 if (rootvp != NULL) {
2650 vn_seqc_write_end(rootvp);
2657 * Suspend write operations on all local writeable filesystems. Does
2658 * full sync of them in the process.
2660 * Iterate over the mount points in reverse order, suspending most
2661 * recently mounted filesystems first. It handles a case where a
2662 * filesystem mounted from a md(4) vnode-backed device should be
2663 * suspended before the filesystem that owns the vnode.
2666 suspend_all_fs(void)
2671 mtx_lock(&mountlist_mtx);
2672 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
2673 error = vfs_busy(mp, MBF_MNTLSTLOCK | MBF_NOWAIT);
2676 if ((mp->mnt_flag & (MNT_RDONLY | MNT_LOCAL)) != MNT_LOCAL ||
2677 (mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
2678 mtx_lock(&mountlist_mtx);
2682 error = vfs_write_suspend(mp, 0);
2685 MPASS((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0);
2686 mp->mnt_kern_flag |= MNTK_SUSPEND_ALL;
2688 mtx_lock(&mountlist_mtx);
2690 printf("suspend of %s failed, error %d\n",
2691 mp->mnt_stat.f_mntonname, error);
2692 mtx_lock(&mountlist_mtx);
2696 mtx_unlock(&mountlist_mtx);
2704 mtx_lock(&mountlist_mtx);
2705 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2706 if ((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0)
2708 mtx_unlock(&mountlist_mtx);
2710 MPASS((mp->mnt_kern_flag & MNTK_SUSPEND) != 0);
2711 mp->mnt_kern_flag &= ~MNTK_SUSPEND_ALL;
2713 vfs_write_resume(mp, 0);
2714 mtx_lock(&mountlist_mtx);
2717 mtx_unlock(&mountlist_mtx);