2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1999-2004 Poul-Henning Kamp
5 * Copyright (c) 1999 Michael Smith
6 * Copyright (c) 1989, 1993
7 * The Regents of the University of California. All rights reserved.
8 * (c) UNIX System Laboratories, Inc.
9 * All or some portions of this file are derived from material licensed
10 * to the University of California by American Telephone and Telegraph
11 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
12 * the permission of UNIX System Laboratories, Inc.
14 * Redistribution and use in source and binary forms, with or without
15 * modification, are permitted provided that the following conditions
17 * 1. Redistributions of source code must retain the above copyright
18 * notice, this list of conditions and the following disclaimer.
19 * 2. Redistributions in binary form must reproduce the above copyright
20 * notice, this list of conditions and the following disclaimer in the
21 * documentation and/or other materials provided with the distribution.
22 * 3. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
42 #include <sys/param.h>
45 #include <sys/devctl.h>
46 #include <sys/eventhandler.h>
47 #include <sys/fcntl.h>
49 #include <sys/kernel.h>
51 #include <sys/libkern.h>
52 #include <sys/malloc.h>
53 #include <sys/mount.h>
54 #include <sys/mutex.h>
55 #include <sys/namei.h>
58 #include <sys/filedesc.h>
59 #include <sys/reboot.h>
61 #include <sys/syscallsubr.h>
62 #include <sys/sysproto.h>
64 #include <sys/sysctl.h>
65 #include <sys/systm.h>
66 #include <sys/vnode.h>
69 #include <geom/geom.h>
71 #include <machine/stdarg.h>
73 #include <security/audit/audit.h>
74 #include <security/mac/mac_framework.h>
76 #define VFS_MOUNTARG_SIZE_MAX (1024 * 64)
78 static int vfs_domount(struct thread *td, const char *fstype, char *fspath,
79 uint64_t fsflags, struct vfsoptlist **optlist);
80 static void free_mntarg(struct mntarg *ma);
82 static int usermount = 0;
83 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
84 "Unprivileged users may mount and unmount file systems");
86 static bool default_autoro = false;
87 SYSCTL_BOOL(_vfs, OID_AUTO, default_autoro, CTLFLAG_RW, &default_autoro, 0,
88 "Retry failed r/w mount as r/o if no explicit ro/rw option is specified");
90 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
91 MALLOC_DEFINE(M_STATFS, "statfs", "statfs structure");
92 static uma_zone_t mount_zone;
94 /* List of mounted filesystems. */
95 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
97 /* For any iteration/modification of mountlist */
98 struct mtx_padalign __exclusive_cache_line mountlist_mtx;
99 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF);
101 EVENTHANDLER_LIST_DEFINE(vfs_mounted);
102 EVENTHANDLER_LIST_DEFINE(vfs_unmounted);
104 static void mount_devctl_event(const char *type, struct mount *mp, bool donew);
107 * Global opts, taken by all filesystems
109 static const char *global_opts[] = {
121 mount_init(void *mem, int size, int flags)
125 mp = (struct mount *)mem;
126 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
127 mtx_init(&mp->mnt_listmtx, "struct mount vlist mtx", NULL, MTX_DEF);
128 lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0);
129 mp->mnt_pcpu = uma_zalloc_pcpu(pcpu_zone_16, M_WAITOK | M_ZERO);
132 mp->mnt_rootvnode = NULL;
137 mount_fini(void *mem, int size)
141 mp = (struct mount *)mem;
142 uma_zfree_pcpu(pcpu_zone_16, mp->mnt_pcpu);
143 lockdestroy(&mp->mnt_explock);
144 mtx_destroy(&mp->mnt_listmtx);
145 mtx_destroy(&mp->mnt_mtx);
149 vfs_mount_init(void *dummy __unused)
152 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), NULL,
153 NULL, mount_init, mount_fini, UMA_ALIGN_CACHE, UMA_ZONE_NOFREE);
155 SYSINIT(vfs_mount, SI_SUB_VFS, SI_ORDER_ANY, vfs_mount_init, NULL);
158 * ---------------------------------------------------------------------
159 * Functions for building and sanitizing the mount options
162 /* Remove one mount option. */
164 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
167 TAILQ_REMOVE(opts, opt, link);
168 free(opt->name, M_MOUNT);
169 if (opt->value != NULL)
170 free(opt->value, M_MOUNT);
174 /* Release all resources related to the mount options. */
176 vfs_freeopts(struct vfsoptlist *opts)
180 while (!TAILQ_EMPTY(opts)) {
181 opt = TAILQ_FIRST(opts);
182 vfs_freeopt(opts, opt);
188 vfs_deleteopt(struct vfsoptlist *opts, const char *name)
190 struct vfsopt *opt, *temp;
194 TAILQ_FOREACH_SAFE(opt, opts, link, temp) {
195 if (strcmp(opt->name, name) == 0)
196 vfs_freeopt(opts, opt);
201 vfs_isopt_ro(const char *opt)
204 if (strcmp(opt, "ro") == 0 || strcmp(opt, "rdonly") == 0 ||
205 strcmp(opt, "norw") == 0)
211 vfs_isopt_rw(const char *opt)
214 if (strcmp(opt, "rw") == 0 || strcmp(opt, "noro") == 0)
220 * Check if options are equal (with or without the "no" prefix).
223 vfs_equalopts(const char *opt1, const char *opt2)
227 /* "opt" vs. "opt" or "noopt" vs. "noopt" */
228 if (strcmp(opt1, opt2) == 0)
230 /* "noopt" vs. "opt" */
231 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
233 /* "opt" vs. "noopt" */
234 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
236 while ((p = strchr(opt1, '.')) != NULL &&
237 !strncmp(opt1, opt2, ++p - opt1)) {
240 /* "foo.noopt" vs. "foo.opt" */
241 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
243 /* "foo.opt" vs. "foo.noopt" */
244 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
247 /* "ro" / "rdonly" / "norw" / "rw" / "noro" */
248 if ((vfs_isopt_ro(opt1) || vfs_isopt_rw(opt1)) &&
249 (vfs_isopt_ro(opt2) || vfs_isopt_rw(opt2)))
255 * If a mount option is specified several times,
256 * (with or without the "no" prefix) only keep
257 * the last occurrence of it.
260 vfs_sanitizeopts(struct vfsoptlist *opts)
262 struct vfsopt *opt, *opt2, *tmp;
264 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
265 opt2 = TAILQ_PREV(opt, vfsoptlist, link);
266 while (opt2 != NULL) {
267 if (vfs_equalopts(opt->name, opt2->name)) {
268 tmp = TAILQ_PREV(opt2, vfsoptlist, link);
269 vfs_freeopt(opts, opt2);
272 opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
279 * Build a linked list of mount options from a struct uio.
282 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
284 struct vfsoptlist *opts;
286 size_t memused, namelen, optlen;
287 unsigned int i, iovcnt;
290 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
293 iovcnt = auio->uio_iovcnt;
294 for (i = 0; i < iovcnt; i += 2) {
295 namelen = auio->uio_iov[i].iov_len;
296 optlen = auio->uio_iov[i + 1].iov_len;
297 memused += sizeof(struct vfsopt) + optlen + namelen;
299 * Avoid consuming too much memory, and attempts to overflow
302 if (memused > VFS_MOUNTARG_SIZE_MAX ||
303 optlen > VFS_MOUNTARG_SIZE_MAX ||
304 namelen > VFS_MOUNTARG_SIZE_MAX) {
309 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
310 opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
317 * Do this early, so jumps to "bad" will free the current
320 TAILQ_INSERT_TAIL(opts, opt, link);
322 if (auio->uio_segflg == UIO_SYSSPACE) {
323 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
325 error = copyin(auio->uio_iov[i].iov_base, opt->name,
330 /* Ensure names are null-terminated strings. */
331 if (namelen == 0 || opt->name[namelen - 1] != '\0') {
337 opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
338 if (auio->uio_segflg == UIO_SYSSPACE) {
339 bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
342 error = copyin(auio->uio_iov[i + 1].iov_base,
349 vfs_sanitizeopts(opts);
358 * Merge the old mount options with the new ones passed
359 * in the MNT_UPDATE case.
361 * XXX: This function will keep a "nofoo" option in the new
362 * options. E.g, if the option's canonical name is "foo",
363 * "nofoo" ends up in the mount point's active options.
366 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *oldopts)
368 struct vfsopt *opt, *new;
370 TAILQ_FOREACH(opt, oldopts, link) {
371 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
372 new->name = strdup(opt->name, M_MOUNT);
374 new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
375 bcopy(opt->value, new->value, opt->len);
379 new->seen = opt->seen;
380 TAILQ_INSERT_HEAD(toopts, new, link);
382 vfs_sanitizeopts(toopts);
386 * Mount a filesystem.
388 #ifndef _SYS_SYSPROTO_H_
396 sys_nmount(struct thread *td, struct nmount_args *uap)
404 * Mount flags are now 64-bits. On 32-bit archtectures only
405 * 32-bits are passed in, but from here on everything handles
406 * 64-bit flags correctly.
410 AUDIT_ARG_FFLAGS(flags);
411 CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__,
412 uap->iovp, uap->iovcnt, flags);
415 * Filter out MNT_ROOTFS. We do not want clients of nmount() in
416 * userspace to set this flag, but we must filter it out if we want
417 * MNT_UPDATE on the root file system to work.
418 * MNT_ROOTFS should only be set by the kernel when mounting its
421 flags &= ~MNT_ROOTFS;
423 iovcnt = uap->iovcnt;
425 * Check that we have an even number of iovec's
426 * and that we have at least two options.
428 if ((iovcnt & 1) || (iovcnt < 4)) {
429 CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__,
434 error = copyinuio(uap->iovp, iovcnt, &auio);
436 CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno",
440 error = vfs_donmount(td, flags, auio);
447 * ---------------------------------------------------------------------
448 * Various utility functions
452 * Get a reference on a mount point from a vnode.
454 * The vnode is allowed to be passed unlocked and race against dooming. Note in
455 * such case there are no guarantees the referenced mount point will still be
456 * associated with it after the function returns.
459 vfs_ref_from_vp(struct vnode *vp)
462 struct mount_pcpu *mpcpu;
464 mp = atomic_load_ptr(&vp->v_mount);
465 if (__predict_false(mp == NULL)) {
468 if (vfs_op_thread_enter(mp, mpcpu)) {
469 if (__predict_true(mp == vp->v_mount)) {
470 vfs_mp_count_add_pcpu(mpcpu, ref, 1);
471 vfs_op_thread_exit(mp, mpcpu);
473 vfs_op_thread_exit(mp, mpcpu);
478 if (mp == vp->v_mount) {
490 vfs_ref(struct mount *mp)
492 struct mount_pcpu *mpcpu;
494 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
495 if (vfs_op_thread_enter(mp, mpcpu)) {
496 vfs_mp_count_add_pcpu(mpcpu, ref, 1);
497 vfs_op_thread_exit(mp, mpcpu);
507 vfs_rel(struct mount *mp)
509 struct mount_pcpu *mpcpu;
511 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
512 if (vfs_op_thread_enter(mp, mpcpu)) {
513 vfs_mp_count_sub_pcpu(mpcpu, ref, 1);
514 vfs_op_thread_exit(mp, mpcpu);
524 * Allocate and initialize the mount point struct.
527 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath,
532 mp = uma_zalloc(mount_zone, M_WAITOK);
533 bzero(&mp->mnt_startzero,
534 __rangeof(struct mount, mnt_startzero, mnt_endzero));
535 mp->mnt_kern_flag = 0;
537 mp->mnt_rootvnode = NULL;
538 mp->mnt_vnodecovered = NULL;
541 TAILQ_INIT(&mp->mnt_nvnodelist);
542 mp->mnt_nvnodelistsize = 0;
543 TAILQ_INIT(&mp->mnt_lazyvnodelist);
544 mp->mnt_lazyvnodelistsize = 0;
545 MPPASS(mp->mnt_ref == 0 && mp->mnt_lockref == 0 &&
546 mp->mnt_writeopcount == 0, mp);
547 MPASSERT(mp->mnt_vfs_ops == 1, mp,
548 ("vfs_ops should be 1 but %d found", mp->mnt_vfs_ops));
549 (void) vfs_busy(mp, MBF_NOWAIT);
550 atomic_add_acq_int(&vfsp->vfc_refcount, 1);
551 mp->mnt_op = vfsp->vfc_vfsops;
553 mp->mnt_stat.f_type = vfsp->vfc_typenum;
555 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
556 mp->mnt_vnodecovered = vp;
557 mp->mnt_cred = crdup(cred);
558 mp->mnt_stat.f_owner = cred->cr_uid;
559 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
560 mp->mnt_iosize_max = DFLTPHYS;
563 mac_mount_create(cred, mp);
565 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
566 TAILQ_INIT(&mp->mnt_uppers);
571 * Destroy the mount struct previously allocated by vfs_mount_alloc().
574 vfs_mount_destroy(struct mount *mp)
577 MPPASS(mp->mnt_vfs_ops != 0, mp);
579 vfs_assert_mount_counters(mp);
582 mp->mnt_kern_flag |= MNTK_REFEXPIRE;
583 if (mp->mnt_kern_flag & MNTK_MWAIT) {
584 mp->mnt_kern_flag &= ~MNTK_MWAIT;
588 msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0);
589 KASSERT(mp->mnt_ref == 0,
590 ("%s: invalid refcount in the drain path @ %s:%d", __func__,
591 __FILE__, __LINE__));
592 MPPASS(mp->mnt_writeopcount == 0, mp);
593 MPPASS(mp->mnt_secondary_writes == 0, mp);
594 atomic_subtract_rel_int(&mp->mnt_vfc->vfc_refcount, 1);
595 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
598 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
599 vn_printf(vp, "dangling vnode ");
600 panic("unmount: dangling vnode");
602 KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers"));
603 MPPASS(mp->mnt_nvnodelistsize == 0, mp);
604 MPPASS(mp->mnt_lazyvnodelistsize == 0, mp);
605 MPPASS(mp->mnt_lockref == 0, mp);
608 MPASSERT(mp->mnt_vfs_ops == 1, mp,
609 ("vfs_ops should be 1 but %d found", mp->mnt_vfs_ops));
611 MPASSERT(mp->mnt_rootvnode == NULL, mp,
612 ("mount point still has a root vnode %p", mp->mnt_rootvnode));
614 if (mp->mnt_vnodecovered != NULL)
615 vrele(mp->mnt_vnodecovered);
617 mac_mount_destroy(mp);
619 if (mp->mnt_opt != NULL)
620 vfs_freeopts(mp->mnt_opt);
621 if (mp->mnt_exjail != NULL) {
622 atomic_subtract_int(&mp->mnt_exjail->cr_prison->pr_exportcnt,
624 crfree(mp->mnt_exjail);
626 if (mp->mnt_export != NULL) {
627 vfs_free_addrlist(mp->mnt_export);
628 free(mp->mnt_export, M_MOUNT);
630 crfree(mp->mnt_cred);
631 uma_zfree(mount_zone, mp);
635 vfs_should_downgrade_to_ro_mount(uint64_t fsflags, int error)
637 /* This is an upgrade of an exisiting mount. */
638 if ((fsflags & MNT_UPDATE) != 0)
640 /* This is already an R/O mount. */
641 if ((fsflags & MNT_RDONLY) != 0)
645 case ENODEV: /* generic, geom, ... */
646 case EACCES: /* cam/scsi, ... */
647 case EROFS: /* md, mmcsd, ... */
649 * These errors can be returned by the storage layer to signal
650 * that the media is read-only. No harm in the R/O mount
651 * attempt if the error was returned for some other reason.
660 vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions)
662 struct vfsoptlist *optlist;
663 struct vfsopt *opt, *tmp_opt;
664 char *fstype, *fspath, *errmsg;
665 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
668 errmsg = fspath = NULL;
669 errmsg_len = fspathlen = 0;
671 autoro = default_autoro;
673 error = vfs_buildopts(fsoptions, &optlist);
677 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
678 errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
681 * We need these two options before the others,
682 * and they are mandatory for any filesystem.
683 * Ensure they are NUL terminated as well.
686 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
687 if (error || fstypelen <= 0 || fstype[fstypelen - 1] != '\0') {
690 strncpy(errmsg, "Invalid fstype", errmsg_len);
694 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
695 if (error || fspathlen <= 0 || fspath[fspathlen - 1] != '\0') {
698 strncpy(errmsg, "Invalid fspath", errmsg_len);
703 * We need to see if we have the "update" option
704 * before we call vfs_domount(), since vfs_domount() has special
705 * logic based on MNT_UPDATE. This is very important
706 * when we want to update the root filesystem.
708 TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) {
711 if (strcmp(opt->name, "update") == 0) {
712 fsflags |= MNT_UPDATE;
715 else if (strcmp(opt->name, "async") == 0)
716 fsflags |= MNT_ASYNC;
717 else if (strcmp(opt->name, "force") == 0) {
718 fsflags |= MNT_FORCE;
721 else if (strcmp(opt->name, "reload") == 0) {
722 fsflags |= MNT_RELOAD;
725 else if (strcmp(opt->name, "multilabel") == 0)
726 fsflags |= MNT_MULTILABEL;
727 else if (strcmp(opt->name, "noasync") == 0)
728 fsflags &= ~MNT_ASYNC;
729 else if (strcmp(opt->name, "noatime") == 0)
730 fsflags |= MNT_NOATIME;
731 else if (strcmp(opt->name, "atime") == 0) {
732 free(opt->name, M_MOUNT);
733 opt->name = strdup("nonoatime", M_MOUNT);
735 else if (strcmp(opt->name, "noclusterr") == 0)
736 fsflags |= MNT_NOCLUSTERR;
737 else if (strcmp(opt->name, "clusterr") == 0) {
738 free(opt->name, M_MOUNT);
739 opt->name = strdup("nonoclusterr", M_MOUNT);
741 else if (strcmp(opt->name, "noclusterw") == 0)
742 fsflags |= MNT_NOCLUSTERW;
743 else if (strcmp(opt->name, "clusterw") == 0) {
744 free(opt->name, M_MOUNT);
745 opt->name = strdup("nonoclusterw", M_MOUNT);
747 else if (strcmp(opt->name, "noexec") == 0)
748 fsflags |= MNT_NOEXEC;
749 else if (strcmp(opt->name, "exec") == 0) {
750 free(opt->name, M_MOUNT);
751 opt->name = strdup("nonoexec", M_MOUNT);
753 else if (strcmp(opt->name, "nosuid") == 0)
754 fsflags |= MNT_NOSUID;
755 else if (strcmp(opt->name, "suid") == 0) {
756 free(opt->name, M_MOUNT);
757 opt->name = strdup("nonosuid", M_MOUNT);
759 else if (strcmp(opt->name, "nosymfollow") == 0)
760 fsflags |= MNT_NOSYMFOLLOW;
761 else if (strcmp(opt->name, "symfollow") == 0) {
762 free(opt->name, M_MOUNT);
763 opt->name = strdup("nonosymfollow", M_MOUNT);
765 else if (strcmp(opt->name, "noro") == 0) {
766 fsflags &= ~MNT_RDONLY;
769 else if (strcmp(opt->name, "rw") == 0) {
770 fsflags &= ~MNT_RDONLY;
773 else if (strcmp(opt->name, "ro") == 0) {
774 fsflags |= MNT_RDONLY;
777 else if (strcmp(opt->name, "rdonly") == 0) {
778 free(opt->name, M_MOUNT);
779 opt->name = strdup("ro", M_MOUNT);
780 fsflags |= MNT_RDONLY;
783 else if (strcmp(opt->name, "autoro") == 0) {
787 else if (strcmp(opt->name, "suiddir") == 0)
788 fsflags |= MNT_SUIDDIR;
789 else if (strcmp(opt->name, "sync") == 0)
790 fsflags |= MNT_SYNCHRONOUS;
791 else if (strcmp(opt->name, "union") == 0)
792 fsflags |= MNT_UNION;
793 else if (strcmp(opt->name, "export") == 0)
794 fsflags |= MNT_EXPORTED;
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 (vfsp->vfc_flags & VFCF_FILEMOUNT) {
975 if (error == 0 && vp->v_type != VDIR && vp->v_type != VREG)
978 * For file mounts, ensure that there is only one hardlink to the file.
980 if (error == 0 && vp->v_type == VREG && va.va_nlink != 1)
983 if (error == 0 && vp->v_type != VDIR)
986 if (error == 0 && (fsflags & MNT_EMPTYDIR) != 0)
987 error = vn_dir_check_empty(vp);
990 if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL)
991 vp->v_iflag |= VI_MOUNT;
1000 vn_seqc_write_begin(vp);
1003 /* Allocate and initialize the filesystem. */
1004 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred);
1005 /* XXXMAC: pass to vfs_mount_alloc? */
1006 mp->mnt_optnew = *optlist;
1007 /* Set the mount level flags. */
1008 mp->mnt_flag = (fsflags & (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY));
1011 * Mount the filesystem.
1012 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1013 * get. No freeing of cn_pnbuf.
1017 if ((error = VFS_MOUNT(mp)) != 0 ||
1018 (error1 = VFS_STATFS(mp, &mp->mnt_stat)) != 0 ||
1019 (error1 = VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) != 0) {
1023 rootvp = vfs_cache_root_clear(mp);
1024 if (rootvp != NULL) {
1028 (void)vn_start_write(NULL, &mp, V_WAIT);
1030 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_UNMOUNTF;
1033 error = VFS_UNMOUNT(mp, 0);
1034 vn_finished_write(mp);
1037 "failed post-mount (%d): rollback unmount returned %d\n",
1044 mp->mnt_vnodecovered = NULL;
1046 /* XXXKIB wait for mnt_lockref drain? */
1047 vfs_mount_destroy(mp);
1050 vp->v_iflag &= ~VI_MOUNT;
1052 if (rootvp != NULL) {
1053 vn_seqc_write_end(rootvp);
1056 vn_seqc_write_end(vp);
1060 vn_seqc_write_begin(newdp);
1063 if (mp->mnt_opt != NULL)
1064 vfs_freeopts(mp->mnt_opt);
1065 mp->mnt_opt = mp->mnt_optnew;
1069 * Prevent external consumers of mount options from reading mnt_optnew.
1071 mp->mnt_optnew = NULL;
1074 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1075 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1076 mp->mnt_kern_flag |= MNTK_ASYNC;
1078 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1081 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1084 vp->v_iflag &= ~VI_MOUNT;
1085 vn_irflag_set_locked(vp, VIRF_MOUNTPOINT);
1086 vp->v_mountedhere = mp;
1088 /* Place the new filesystem at the end of the mount list. */
1089 mtx_lock(&mountlist_mtx);
1090 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1091 mtx_unlock(&mountlist_mtx);
1092 vfs_event_signal(NULL, VQ_MOUNT, 0);
1093 vn_lock(newdp, LK_EXCLUSIVE | LK_RETRY);
1095 EVENTHANDLER_DIRECT_INVOKE(vfs_mounted, mp, newdp, td);
1097 mount_devctl_event("MOUNT", mp, false);
1098 mountcheckdirs(vp, newdp);
1099 vn_seqc_write_end(vp);
1100 vn_seqc_write_end(newdp);
1102 if ((mp->mnt_flag & MNT_RDONLY) == 0)
1103 vfs_allocate_syncvnode(mp);
1110 * vfs_domount_update(): update of mounted file system
1114 struct thread *td, /* Calling thread. */
1115 struct vnode *vp, /* Mount point vnode. */
1116 uint64_t fsflags, /* Flags common to all filesystems. */
1117 struct vfsoptlist **optlist /* Options local to the filesystem. */
1120 struct export_args export;
1121 struct o2export_args o2export;
1122 struct vnode *rootvp;
1125 int error, export_error, i, len;
1129 ASSERT_VOP_ELOCKED(vp, __func__);
1130 KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here"));
1133 if ((vp->v_vflag & VV_ROOT) == 0) {
1134 if (vfs_copyopt(*optlist, "export", &export, sizeof(export))
1144 * We only allow the filesystem to be reloaded if it
1145 * is currently mounted read-only.
1147 flag = mp->mnt_flag;
1148 if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) {
1150 return (EOPNOTSUPP); /* Needs translation */
1153 * Only privileged root, or (if MNT_USER is set) the user that
1154 * did the original mount is permitted to update it.
1156 error = vfs_suser(mp, td);
1161 if (vfs_busy(mp, MBF_NOWAIT)) {
1166 if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) {
1172 vp->v_iflag |= VI_MOUNT;
1177 vn_seqc_write_begin(vp);
1181 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1186 mp->mnt_flag &= ~MNT_UPDATEMASK;
1187 mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE |
1188 MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY);
1189 if ((mp->mnt_flag & MNT_ASYNC) == 0)
1190 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1191 rootvp = vfs_cache_root_clear(mp);
1193 mp->mnt_optnew = *optlist;
1194 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
1197 * Mount the filesystem.
1198 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1199 * get. No freeing of cn_pnbuf.
1201 error = VFS_MOUNT(mp);
1204 /* Process the export option. */
1205 if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp,
1207 /* Assume that there is only 1 ABI for each length. */
1209 case (sizeof(struct oexport_args)):
1210 bzero(&o2export, sizeof(o2export));
1212 case (sizeof(o2export)):
1213 bcopy(bufp, &o2export, len);
1214 export.ex_flags = (uint64_t)o2export.ex_flags;
1215 export.ex_root = o2export.ex_root;
1216 export.ex_uid = o2export.ex_anon.cr_uid;
1217 export.ex_groups = NULL;
1218 export.ex_ngroups = o2export.ex_anon.cr_ngroups;
1219 if (export.ex_ngroups > 0) {
1220 if (export.ex_ngroups <= XU_NGROUPS) {
1221 export.ex_groups = malloc(
1222 export.ex_ngroups * sizeof(gid_t),
1224 for (i = 0; i < export.ex_ngroups; i++)
1225 export.ex_groups[i] =
1226 o2export.ex_anon.cr_groups[i];
1228 export_error = EINVAL;
1229 } else if (export.ex_ngroups < 0)
1230 export_error = EINVAL;
1231 export.ex_addr = o2export.ex_addr;
1232 export.ex_addrlen = o2export.ex_addrlen;
1233 export.ex_mask = o2export.ex_mask;
1234 export.ex_masklen = o2export.ex_masklen;
1235 export.ex_indexfile = o2export.ex_indexfile;
1236 export.ex_numsecflavors = o2export.ex_numsecflavors;
1237 if (export.ex_numsecflavors < MAXSECFLAVORS) {
1238 for (i = 0; i < export.ex_numsecflavors; i++)
1239 export.ex_secflavors[i] =
1240 o2export.ex_secflavors[i];
1242 export_error = EINVAL;
1243 if (export_error == 0)
1244 export_error = vfs_export(mp, &export, true);
1245 free(export.ex_groups, M_TEMP);
1247 case (sizeof(export)):
1248 bcopy(bufp, &export, len);
1250 if (export.ex_ngroups > 0) {
1251 if (export.ex_ngroups <= NGROUPS_MAX) {
1252 grps = malloc(export.ex_ngroups *
1253 sizeof(gid_t), M_TEMP, M_WAITOK);
1254 export_error = copyin(export.ex_groups,
1255 grps, export.ex_ngroups *
1257 if (export_error == 0)
1258 export.ex_groups = grps;
1260 export_error = EINVAL;
1261 } else if (export.ex_ngroups == 0)
1262 export.ex_groups = NULL;
1264 export_error = EINVAL;
1265 if (export_error == 0)
1266 export_error = vfs_export(mp, &export, true);
1270 export_error = EINVAL;
1277 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE |
1281 * If we fail, restore old mount flags. MNT_QUOTA is special,
1282 * because it is not part of MNT_UPDATEMASK, but it could have
1283 * changed in the meantime if quotactl(2) was called.
1284 * All in all we want current value of MNT_QUOTA, not the old
1287 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA);
1289 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1290 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1291 mp->mnt_kern_flag |= MNTK_ASYNC;
1293 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1299 mount_devctl_event("REMOUNT", mp, true);
1300 if (mp->mnt_opt != NULL)
1301 vfs_freeopts(mp->mnt_opt);
1302 mp->mnt_opt = mp->mnt_optnew;
1304 (void)VFS_STATFS(mp, &mp->mnt_stat);
1306 * Prevent external consumers of mount options from reading
1309 mp->mnt_optnew = NULL;
1311 if ((mp->mnt_flag & MNT_RDONLY) == 0)
1312 vfs_allocate_syncvnode(mp);
1314 vfs_deallocate_syncvnode(mp);
1317 if (rootvp != NULL) {
1318 vn_seqc_write_end(rootvp);
1321 vn_seqc_write_end(vp);
1324 vp->v_iflag &= ~VI_MOUNT;
1327 return (error != 0 ? error : export_error);
1331 * vfs_domount(): actually attempt a filesystem mount.
1335 struct thread *td, /* Calling thread. */
1336 const char *fstype, /* Filesystem type. */
1337 char *fspath, /* Mount path. */
1338 uint64_t fsflags, /* Flags common to all filesystems. */
1339 struct vfsoptlist **optlist /* Options local to the filesystem. */
1342 struct vfsconf *vfsp;
1343 struct nameidata nd;
1349 * Be ultra-paranoid about making sure the type and fspath
1350 * variables will fit in our mp buffers, including the
1353 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
1354 return (ENAMETOOLONG);
1356 if (jailed(td->td_ucred) || usermount == 0) {
1357 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
1362 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
1364 if (fsflags & MNT_EXPORTED) {
1365 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
1369 if (fsflags & MNT_SUIDDIR) {
1370 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
1375 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
1377 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
1378 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
1379 fsflags |= MNT_NOSUID | MNT_USER;
1382 /* Load KLDs before we lock the covered vnode to avoid reversals. */
1384 if ((fsflags & MNT_UPDATE) == 0) {
1385 /* Don't try to load KLDs if we're mounting the root. */
1386 if (fsflags & MNT_ROOTFS) {
1387 if ((vfsp = vfs_byname(fstype)) == NULL)
1390 if ((vfsp = vfs_byname_kld(fstype, td, &error)) == NULL)
1396 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE.
1398 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1 | WANTPARENT,
1399 UIO_SYSSPACE, fspath, td);
1405 * Don't allow stacking file mounts to work around problems with the way
1406 * that namei sets nd.ni_dvp to vp_crossmp for these.
1408 if (vp->v_type == VREG)
1409 fsflags |= MNT_NOCOVER;
1410 if ((fsflags & MNT_UPDATE) == 0) {
1411 if ((vp->v_vflag & VV_ROOT) != 0 &&
1412 (fsflags & MNT_NOCOVER) != 0) {
1417 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1418 strcpy(pathbuf, fspath);
1420 * Note: we allow any vnode type here. If the path sanity check
1421 * succeeds, the type will be validated in vfs_domount_first
1424 if (vp->v_type == VDIR)
1425 error = vn_path_to_global_path(td, vp, pathbuf,
1428 error = vn_path_to_global_path_hardlink(td, vp,
1429 nd.ni_dvp, pathbuf, MNAMELEN,
1430 nd.ni_cnd.cn_nameptr, nd.ni_cnd.cn_namelen);
1432 error = vfs_domount_first(td, vfsp, pathbuf, vp,
1435 free(pathbuf, M_TEMP);
1437 error = vfs_domount_update(td, vp, fsflags, optlist);
1440 NDFREE(&nd, NDF_ONLY_PNBUF);
1447 * Unmount a filesystem.
1449 * Note: unmount takes a path to the vnode mounted on as argument, not
1450 * special file (as before).
1452 #ifndef _SYS_SYSPROTO_H_
1453 struct unmount_args {
1460 sys_unmount(struct thread *td, struct unmount_args *uap)
1463 return (kern_unmount(td, uap->path, uap->flags));
1467 kern_unmount(struct thread *td, const char *path, int flags)
1469 struct nameidata nd;
1472 int error, id0, id1;
1474 AUDIT_ARG_VALUE(flags);
1475 if (jailed(td->td_ucred) || usermount == 0) {
1476 error = priv_check(td, PRIV_VFS_UNMOUNT);
1481 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1482 error = copyinstr(path, pathbuf, MNAMELEN, NULL);
1484 free(pathbuf, M_TEMP);
1487 if (flags & MNT_BYFSID) {
1488 AUDIT_ARG_TEXT(pathbuf);
1489 /* Decode the filesystem ID. */
1490 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1491 free(pathbuf, M_TEMP);
1495 mtx_lock(&mountlist_mtx);
1496 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1497 if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1498 mp->mnt_stat.f_fsid.val[1] == id1) {
1503 mtx_unlock(&mountlist_mtx);
1506 * Try to find global path for path argument.
1508 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1509 UIO_SYSSPACE, pathbuf, td);
1510 if (namei(&nd) == 0) {
1511 NDFREE(&nd, NDF_ONLY_PNBUF);
1512 error = vn_path_to_global_path(td, nd.ni_vp, pathbuf,
1517 mtx_lock(&mountlist_mtx);
1518 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1519 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) {
1524 mtx_unlock(&mountlist_mtx);
1526 free(pathbuf, M_TEMP);
1529 * Previously we returned ENOENT for a nonexistent path and
1530 * EINVAL for a non-mountpoint. We cannot tell these apart
1531 * now, so in the !MNT_BYFSID case return the more likely
1532 * EINVAL for compatibility.
1534 return ((flags & MNT_BYFSID) ? ENOENT : EINVAL);
1538 * Don't allow unmounting the root filesystem.
1540 if (mp->mnt_flag & MNT_ROOTFS) {
1544 error = dounmount(mp, flags, td);
1549 * Return error if any of the vnodes, ignoring the root vnode
1550 * and the syncer vnode, have non-zero usecount.
1552 * This function is purely advisory - it can return false positives
1556 vfs_check_usecounts(struct mount *mp)
1558 struct vnode *vp, *mvp;
1560 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1561 if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON &&
1562 vp->v_usecount != 0) {
1564 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1574 dounmount_cleanup(struct mount *mp, struct vnode *coveredvp, int mntkflags)
1577 mtx_assert(MNT_MTX(mp), MA_OWNED);
1578 mp->mnt_kern_flag &= ~mntkflags;
1579 if ((mp->mnt_kern_flag & MNTK_MWAIT) != 0) {
1580 mp->mnt_kern_flag &= ~MNTK_MWAIT;
1583 vfs_op_exit_locked(mp);
1585 if (coveredvp != NULL) {
1586 VOP_UNLOCK(coveredvp);
1589 vn_finished_write(mp);
1593 * There are various reference counters associated with the mount point.
1594 * Normally it is permitted to modify them without taking the mnt ilock,
1595 * but this behavior can be temporarily disabled if stable value is needed
1596 * or callers are expected to block (e.g. to not allow new users during
1600 vfs_op_enter(struct mount *mp)
1602 struct mount_pcpu *mpcpu;
1607 if (mp->mnt_vfs_ops > 1) {
1611 vfs_op_barrier_wait(mp);
1613 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1615 mp->mnt_ref += mpcpu->mntp_ref;
1616 mpcpu->mntp_ref = 0;
1618 mp->mnt_lockref += mpcpu->mntp_lockref;
1619 mpcpu->mntp_lockref = 0;
1621 mp->mnt_writeopcount += mpcpu->mntp_writeopcount;
1622 mpcpu->mntp_writeopcount = 0;
1624 MPASSERT(mp->mnt_ref > 0 && mp->mnt_lockref >= 0 &&
1625 mp->mnt_writeopcount >= 0, mp,
1626 ("invalid count(s): ref %d lockref %d writeopcount %d",
1627 mp->mnt_ref, mp->mnt_lockref, mp->mnt_writeopcount));
1629 vfs_assert_mount_counters(mp);
1633 vfs_op_exit_locked(struct mount *mp)
1636 mtx_assert(MNT_MTX(mp), MA_OWNED);
1638 MPASSERT(mp->mnt_vfs_ops > 0, mp,
1639 ("invalid vfs_ops count %d", mp->mnt_vfs_ops));
1640 MPASSERT(mp->mnt_vfs_ops > 1 ||
1641 (mp->mnt_kern_flag & (MNTK_UNMOUNT | MNTK_SUSPEND)) == 0, mp,
1642 ("vfs_ops too low %d in unmount or suspend", mp->mnt_vfs_ops));
1647 vfs_op_exit(struct mount *mp)
1651 vfs_op_exit_locked(mp);
1655 struct vfs_op_barrier_ipi {
1657 struct smp_rendezvous_cpus_retry_arg srcra;
1661 vfs_op_action_func(void *arg)
1663 struct vfs_op_barrier_ipi *vfsopipi;
1666 vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1669 if (!vfs_op_thread_entered(mp))
1670 smp_rendezvous_cpus_done(arg);
1674 vfs_op_wait_func(void *arg, int cpu)
1676 struct vfs_op_barrier_ipi *vfsopipi;
1678 struct mount_pcpu *mpcpu;
1680 vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1683 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1684 while (atomic_load_int(&mpcpu->mntp_thread_in_ops))
1689 vfs_op_barrier_wait(struct mount *mp)
1691 struct vfs_op_barrier_ipi vfsopipi;
1695 smp_rendezvous_cpus_retry(all_cpus,
1696 smp_no_rendezvous_barrier,
1698 smp_no_rendezvous_barrier,
1705 vfs_assert_mount_counters(struct mount *mp)
1707 struct mount_pcpu *mpcpu;
1710 if (mp->mnt_vfs_ops == 0)
1714 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1715 if (mpcpu->mntp_ref != 0 ||
1716 mpcpu->mntp_lockref != 0 ||
1717 mpcpu->mntp_writeopcount != 0)
1718 vfs_dump_mount_counters(mp);
1723 vfs_dump_mount_counters(struct mount *mp)
1725 struct mount_pcpu *mpcpu;
1726 int ref, lockref, writeopcount;
1729 printf("%s: mp %p vfs_ops %d\n", __func__, mp, mp->mnt_vfs_ops);
1734 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1735 printf("%d ", mpcpu->mntp_ref);
1736 ref += mpcpu->mntp_ref;
1739 printf(" lockref : ");
1740 lockref = mp->mnt_lockref;
1742 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1743 printf("%d ", mpcpu->mntp_lockref);
1744 lockref += mpcpu->mntp_lockref;
1747 printf("writeopcount: ");
1748 writeopcount = mp->mnt_writeopcount;
1750 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1751 printf("%d ", mpcpu->mntp_writeopcount);
1752 writeopcount += mpcpu->mntp_writeopcount;
1756 printf("counter struct total\n");
1757 printf("ref %-5d %-5d\n", mp->mnt_ref, ref);
1758 printf("lockref %-5d %-5d\n", mp->mnt_lockref, lockref);
1759 printf("writeopcount %-5d %-5d\n", mp->mnt_writeopcount, writeopcount);
1761 panic("invalid counts on struct mount");
1766 vfs_mount_fetch_counter(struct mount *mp, enum mount_counter which)
1768 struct mount_pcpu *mpcpu;
1775 case MNT_COUNT_LOCKREF:
1776 sum = mp->mnt_lockref;
1778 case MNT_COUNT_WRITEOPCOUNT:
1779 sum = mp->mnt_writeopcount;
1784 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1787 sum += mpcpu->mntp_ref;
1789 case MNT_COUNT_LOCKREF:
1790 sum += mpcpu->mntp_lockref;
1792 case MNT_COUNT_WRITEOPCOUNT:
1793 sum += mpcpu->mntp_writeopcount;
1801 * Do the actual filesystem unmount.
1804 dounmount(struct mount *mp, int flags, struct thread *td)
1806 struct vnode *coveredvp, *rootvp;
1808 uint64_t async_flag;
1811 if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
1812 mnt_gen_r = mp->mnt_gen;
1815 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
1817 * Check for mp being unmounted while waiting for the
1818 * covered vnode lock.
1820 if (coveredvp->v_mountedhere != mp ||
1821 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
1822 VOP_UNLOCK(coveredvp);
1830 * Only privileged root, or (if MNT_USER is set) the user that did the
1831 * original mount is permitted to unmount this filesystem.
1833 error = vfs_suser(mp, td);
1835 if (coveredvp != NULL) {
1836 VOP_UNLOCK(coveredvp);
1845 vn_start_write(NULL, &mp, V_WAIT | V_MNTREF);
1847 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 ||
1848 (mp->mnt_flag & MNT_UPDATE) != 0 ||
1849 !TAILQ_EMPTY(&mp->mnt_uppers)) {
1850 dounmount_cleanup(mp, coveredvp, 0);
1853 mp->mnt_kern_flag |= MNTK_UNMOUNT;
1854 rootvp = vfs_cache_root_clear(mp);
1855 if (coveredvp != NULL)
1856 vn_seqc_write_begin(coveredvp);
1857 if (flags & MNT_NONBUSY) {
1859 error = vfs_check_usecounts(mp);
1862 vn_seqc_write_end(coveredvp);
1863 dounmount_cleanup(mp, coveredvp, MNTK_UNMOUNT);
1864 if (rootvp != NULL) {
1865 vn_seqc_write_end(rootvp);
1871 /* Allow filesystems to detect that a forced unmount is in progress. */
1872 if (flags & MNT_FORCE) {
1873 mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1876 * Must be done after setting MNTK_UNMOUNTF and before
1877 * waiting for mnt_lockref to become 0.
1883 if (mp->mnt_lockref) {
1884 mp->mnt_kern_flag |= MNTK_DRAINING;
1885 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS,
1889 KASSERT(mp->mnt_lockref == 0,
1890 ("%s: invalid lock refcount in the drain path @ %s:%d",
1891 __func__, __FILE__, __LINE__));
1893 ("%s: invalid return value for msleep in the drain path @ %s:%d",
1894 __func__, __FILE__, __LINE__));
1897 * We want to keep the vnode around so that we can vn_seqc_write_end
1898 * after we are done with unmount. Downgrade our reference to a mere
1899 * hold count so that we don't interefere with anything.
1901 if (rootvp != NULL) {
1906 if (mp->mnt_flag & MNT_EXPUBLIC)
1907 vfs_setpublicfs(NULL, NULL, NULL);
1909 vfs_periodic(mp, MNT_WAIT);
1911 async_flag = mp->mnt_flag & MNT_ASYNC;
1912 mp->mnt_flag &= ~MNT_ASYNC;
1913 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1915 vfs_deallocate_syncvnode(mp);
1916 error = VFS_UNMOUNT(mp, flags);
1917 vn_finished_write(mp);
1919 * If we failed to flush the dirty blocks for this mount point,
1920 * undo all the cdir/rdir and rootvnode changes we made above.
1921 * Unless we failed to do so because the device is reporting that
1922 * it doesn't exist anymore.
1924 if (error && error != ENXIO) {
1926 if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1928 vfs_allocate_syncvnode(mp);
1931 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1932 mp->mnt_flag |= async_flag;
1933 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1934 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1935 mp->mnt_kern_flag |= MNTK_ASYNC;
1936 if (mp->mnt_kern_flag & MNTK_MWAIT) {
1937 mp->mnt_kern_flag &= ~MNTK_MWAIT;
1940 vfs_op_exit_locked(mp);
1943 vn_seqc_write_end(coveredvp);
1944 VOP_UNLOCK(coveredvp);
1947 if (rootvp != NULL) {
1948 vn_seqc_write_end(rootvp);
1953 mtx_lock(&mountlist_mtx);
1954 TAILQ_REMOVE(&mountlist, mp, mnt_list);
1955 mtx_unlock(&mountlist_mtx);
1956 EVENTHANDLER_DIRECT_INVOKE(vfs_unmounted, mp, td);
1957 if (coveredvp != NULL) {
1959 vn_irflag_unset_locked(coveredvp, VIRF_MOUNTPOINT);
1960 coveredvp->v_mountedhere = NULL;
1961 vn_seqc_write_end_locked(coveredvp);
1962 VI_UNLOCK(coveredvp);
1963 VOP_UNLOCK(coveredvp);
1966 mount_devctl_event("UNMOUNT", mp, false);
1967 if (rootvp != NULL) {
1968 vn_seqc_write_end(rootvp);
1971 vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1972 if (rootvnode != NULL && mp == rootvnode->v_mount) {
1976 if (mp == rootdevmp)
1978 vfs_mount_destroy(mp);
1983 * Report errors during filesystem mounting.
1986 vfs_mount_error(struct mount *mp, const char *fmt, ...)
1988 struct vfsoptlist *moptlist = mp->mnt_optnew;
1993 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1994 if (error || errmsg == NULL || len <= 0)
1998 vsnprintf(errmsg, (size_t)len, fmt, ap);
2003 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...)
2009 error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len);
2010 if (error || errmsg == NULL || len <= 0)
2014 vsnprintf(errmsg, (size_t)len, fmt, ap);
2019 * ---------------------------------------------------------------------
2020 * Functions for querying mount options/arguments from filesystems.
2024 * Check that no unknown options are given
2027 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
2031 const char **t, *p, *q;
2034 TAILQ_FOREACH(opt, opts, link) {
2037 if (p[0] == 'n' && p[1] == 'o')
2039 for(t = global_opts; *t != NULL; t++) {
2040 if (strcmp(*t, p) == 0)
2043 if (strcmp(*t, q) == 0)
2049 for(t = legal; *t != NULL; t++) {
2050 if (strcmp(*t, p) == 0)
2053 if (strcmp(*t, q) == 0)
2059 snprintf(errmsg, sizeof(errmsg),
2060 "mount option <%s> is unknown", p);
2064 TAILQ_FOREACH(opt, opts, link) {
2065 if (strcmp(opt->name, "errmsg") == 0) {
2066 strncpy((char *)opt->value, errmsg, opt->len);
2071 printf("%s\n", errmsg);
2077 * Get a mount option by its name.
2079 * Return 0 if the option was found, ENOENT otherwise.
2080 * If len is non-NULL it will be filled with the length
2081 * of the option. If buf is non-NULL, it will be filled
2082 * with the address of the option.
2085 vfs_getopt(struct vfsoptlist *opts, const char *name, void **buf, int *len)
2089 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2091 TAILQ_FOREACH(opt, opts, link) {
2092 if (strcmp(name, opt->name) == 0) {
2105 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
2112 TAILQ_FOREACH(opt, opts, link) {
2113 if (strcmp(name, opt->name) == 0) {
2122 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value)
2124 char *opt_value, *vtp;
2128 error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len);
2131 if (opt_len == 0 || opt_value == NULL)
2133 if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0')
2135 iv = strtoq(opt_value, &vtp, 0);
2136 if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0'))
2163 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
2168 TAILQ_FOREACH(opt, opts, link) {
2169 if (strcmp(name, opt->name) != 0)
2172 if (opt->len == 0 ||
2173 ((char *)opt->value)[opt->len - 1] != '\0') {
2177 return (opt->value);
2184 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w,
2189 TAILQ_FOREACH(opt, opts, link) {
2190 if (strcmp(name, opt->name) == 0) {
2203 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
2209 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2211 TAILQ_FOREACH(opt, opts, link) {
2212 if (strcmp(name, opt->name) != 0)
2215 if (opt->len == 0 || opt->value == NULL)
2217 if (((char *)opt->value)[opt->len - 1] != '\0')
2220 ret = vsscanf(opt->value, fmt, ap);
2228 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len)
2232 TAILQ_FOREACH(opt, opts, link) {
2233 if (strcmp(name, opt->name) != 0)
2236 if (opt->value == NULL)
2239 if (opt->len != len)
2241 bcopy(value, opt->value, len);
2249 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len)
2253 TAILQ_FOREACH(opt, opts, link) {
2254 if (strcmp(name, opt->name) != 0)
2257 if (opt->value == NULL)
2263 bcopy(value, opt->value, len);
2271 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value)
2275 TAILQ_FOREACH(opt, opts, link) {
2276 if (strcmp(name, opt->name) != 0)
2279 if (opt->value == NULL)
2280 opt->len = strlen(value) + 1;
2281 else if (strlcpy(opt->value, value, opt->len) >= opt->len)
2289 * Find and copy a mount option.
2291 * The size of the buffer has to be specified
2292 * in len, if it is not the same length as the
2293 * mount option, EINVAL is returned.
2294 * Returns ENOENT if the option is not found.
2297 vfs_copyopt(struct vfsoptlist *opts, const char *name, void *dest, int len)
2301 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
2303 TAILQ_FOREACH(opt, opts, link) {
2304 if (strcmp(name, opt->name) == 0) {
2306 if (len != opt->len)
2308 bcopy(opt->value, dest, opt->len);
2316 __vfs_statfs(struct mount *mp, struct statfs *sbp)
2320 * Filesystems only fill in part of the structure for updates, we
2321 * have to read the entirety first to get all content.
2323 if (sbp != &mp->mnt_stat)
2324 memcpy(sbp, &mp->mnt_stat, sizeof(*sbp));
2327 * Set these in case the underlying filesystem fails to do so.
2329 sbp->f_version = STATFS_VERSION;
2330 sbp->f_namemax = NAME_MAX;
2331 sbp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
2333 return (mp->mnt_op->vfs_statfs(mp, sbp));
2337 vfs_mountedfrom(struct mount *mp, const char *from)
2340 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
2341 strlcpy(mp->mnt_stat.f_mntfromname, from,
2342 sizeof mp->mnt_stat.f_mntfromname);
2346 * ---------------------------------------------------------------------
2347 * This is the api for building mount args and mounting filesystems from
2348 * inside the kernel.
2350 * The API works by accumulation of individual args. First error is
2353 * XXX: should be documented in new manpage kernel_mount(9)
2356 /* A memory allocation which must be freed when we are done */
2358 SLIST_ENTRY(mntaarg) next;
2361 /* The header for the mount arguments */
2366 SLIST_HEAD(, mntaarg) list;
2370 * Add a boolean argument.
2372 * flag is the boolean value.
2373 * name must start with "no".
2376 mount_argb(struct mntarg *ma, int flag, const char *name)
2379 KASSERT(name[0] == 'n' && name[1] == 'o',
2380 ("mount_argb(...,%s): name must start with 'no'", name));
2382 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
2386 * Add an argument printf style
2389 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
2392 struct mntaarg *maa;
2397 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2398 SLIST_INIT(&ma->list);
2403 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2405 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2406 ma->v[ma->len].iov_len = strlen(name) + 1;
2409 sb = sbuf_new_auto();
2411 sbuf_vprintf(sb, fmt, ap);
2414 len = sbuf_len(sb) + 1;
2415 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2416 SLIST_INSERT_HEAD(&ma->list, maa, next);
2417 bcopy(sbuf_data(sb), maa + 1, len);
2420 ma->v[ma->len].iov_base = maa + 1;
2421 ma->v[ma->len].iov_len = len;
2428 * Add an argument which is a userland string.
2431 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2433 struct mntaarg *maa;
2439 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2440 SLIST_INIT(&ma->list);
2444 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2445 SLIST_INSERT_HEAD(&ma->list, maa, next);
2446 tbuf = (void *)(maa + 1);
2447 ma->error = copyinstr(val, tbuf, len, NULL);
2448 return (mount_arg(ma, name, tbuf, -1));
2454 * If length is -1, treat value as a C string.
2457 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2461 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2462 SLIST_INIT(&ma->list);
2467 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2469 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2470 ma->v[ma->len].iov_len = strlen(name) + 1;
2473 ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2475 ma->v[ma->len].iov_len = strlen(val) + 1;
2477 ma->v[ma->len].iov_len = len;
2483 * Free a mntarg structure
2486 free_mntarg(struct mntarg *ma)
2488 struct mntaarg *maa;
2490 while (!SLIST_EMPTY(&ma->list)) {
2491 maa = SLIST_FIRST(&ma->list);
2492 SLIST_REMOVE_HEAD(&ma->list, next);
2495 free(ma->v, M_MOUNT);
2500 * Mount a filesystem
2503 kernel_mount(struct mntarg *ma, uint64_t flags)
2508 KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2509 KASSERT(ma->error != 0 || ma->v != NULL, ("kernel_mount NULL ma->v"));
2510 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2514 auio.uio_iov = ma->v;
2515 auio.uio_iovcnt = ma->len;
2516 auio.uio_segflg = UIO_SYSSPACE;
2517 error = vfs_donmount(curthread, flags, &auio);
2524 * A printflike function to mount a filesystem.
2527 kernel_vmount(int flags, ...)
2529 struct mntarg *ma = NULL;
2535 va_start(ap, flags);
2537 cp = va_arg(ap, const char *);
2540 vp = va_arg(ap, const void *);
2541 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0));
2545 error = kernel_mount(ma, flags);
2549 /* Map from mount options to printable formats. */
2550 static struct mntoptnames optnames[] = {
2554 #define DEVCTL_LEN 1024
2556 mount_devctl_event(const char *type, struct mount *mp, bool donew)
2559 struct mntoptnames *fp;
2561 struct statfs *sfp = &mp->mnt_stat;
2564 buf = malloc(DEVCTL_LEN, M_MOUNT, M_NOWAIT);
2567 sbuf_new(&sb, buf, DEVCTL_LEN, SBUF_FIXEDLEN);
2568 sbuf_cpy(&sb, "mount-point=\"");
2569 devctl_safe_quote_sb(&sb, sfp->f_mntonname);
2570 sbuf_cat(&sb, "\" mount-dev=\"");
2571 devctl_safe_quote_sb(&sb, sfp->f_mntfromname);
2572 sbuf_cat(&sb, "\" mount-type=\"");
2573 devctl_safe_quote_sb(&sb, sfp->f_fstypename);
2574 sbuf_cat(&sb, "\" fsid=0x");
2575 cp = (const uint8_t *)&sfp->f_fsid.val[0];
2576 for (int i = 0; i < sizeof(sfp->f_fsid); i++)
2577 sbuf_printf(&sb, "%02x", cp[i]);
2578 sbuf_printf(&sb, " owner=%u flags=\"", sfp->f_owner);
2579 for (fp = optnames; fp->o_opt != 0; fp++) {
2580 if ((mp->mnt_flag & fp->o_opt) != 0) {
2581 sbuf_cat(&sb, fp->o_name);
2582 sbuf_putc(&sb, ';');
2585 sbuf_putc(&sb, '"');
2589 * Options are not published because the form of the options depends on
2590 * the file system and may include binary data. In addition, they don't
2591 * necessarily provide enough useful information to be actionable when
2592 * devd processes them.
2595 if (sbuf_error(&sb) == 0)
2596 devctl_notify("VFS", "FS", type, sbuf_data(&sb));
2602 * Force remount specified mount point to read-only. The argument
2603 * must be busied to avoid parallel unmount attempts.
2605 * Intended use is to prevent further writes if some metadata
2606 * inconsistency is detected. Note that the function still flushes
2607 * all cached metadata and data for the mount point, which might be
2608 * not always suitable.
2611 vfs_remount_ro(struct mount *mp)
2613 struct vfsoptlist *opts;
2615 struct vnode *vp_covered, *rootvp;
2618 KASSERT(mp->mnt_lockref > 0,
2619 ("vfs_remount_ro: mp %p is not busied", mp));
2620 KASSERT((mp->mnt_kern_flag & MNTK_UNMOUNT) == 0,
2621 ("vfs_remount_ro: mp %p is being unmounted (and busy?)", mp));
2624 vp_covered = mp->mnt_vnodecovered;
2625 error = vget(vp_covered, LK_EXCLUSIVE | LK_NOWAIT);
2628 VI_LOCK(vp_covered);
2629 if ((vp_covered->v_iflag & VI_MOUNT) != 0) {
2630 VI_UNLOCK(vp_covered);
2634 vp_covered->v_iflag |= VI_MOUNT;
2635 VI_UNLOCK(vp_covered);
2637 vn_seqc_write_begin(vp_covered);
2640 if ((mp->mnt_flag & MNT_RDONLY) != 0) {
2645 mp->mnt_flag |= MNT_UPDATE | MNT_FORCE | MNT_RDONLY;
2646 rootvp = vfs_cache_root_clear(mp);
2649 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK | M_ZERO);
2651 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK | M_ZERO);
2652 opt->name = strdup("ro", M_MOUNT);
2654 TAILQ_INSERT_TAIL(opts, opt, link);
2655 vfs_mergeopts(opts, mp->mnt_opt);
2656 mp->mnt_optnew = opts;
2658 error = VFS_MOUNT(mp);
2662 mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE);
2664 vfs_deallocate_syncvnode(mp);
2665 if (mp->mnt_opt != NULL)
2666 vfs_freeopts(mp->mnt_opt);
2667 mp->mnt_opt = mp->mnt_optnew;
2670 mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE | MNT_RDONLY);
2672 vfs_freeopts(mp->mnt_optnew);
2674 mp->mnt_optnew = NULL;
2678 VI_LOCK(vp_covered);
2679 vp_covered->v_iflag &= ~VI_MOUNT;
2680 VI_UNLOCK(vp_covered);
2682 vn_seqc_write_end(vp_covered);
2683 if (rootvp != NULL) {
2684 vn_seqc_write_end(rootvp);
2691 * Suspend write operations on all local writeable filesystems. Does
2692 * full sync of them in the process.
2694 * Iterate over the mount points in reverse order, suspending most
2695 * recently mounted filesystems first. It handles a case where a
2696 * filesystem mounted from a md(4) vnode-backed device should be
2697 * suspended before the filesystem that owns the vnode.
2700 suspend_all_fs(void)
2705 mtx_lock(&mountlist_mtx);
2706 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
2707 error = vfs_busy(mp, MBF_MNTLSTLOCK | MBF_NOWAIT);
2710 if ((mp->mnt_flag & (MNT_RDONLY | MNT_LOCAL)) != MNT_LOCAL ||
2711 (mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
2712 mtx_lock(&mountlist_mtx);
2716 error = vfs_write_suspend(mp, 0);
2719 MPASS((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0);
2720 mp->mnt_kern_flag |= MNTK_SUSPEND_ALL;
2722 mtx_lock(&mountlist_mtx);
2724 printf("suspend of %s failed, error %d\n",
2725 mp->mnt_stat.f_mntonname, error);
2726 mtx_lock(&mountlist_mtx);
2730 mtx_unlock(&mountlist_mtx);
2738 mtx_lock(&mountlist_mtx);
2739 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2740 if ((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0)
2742 mtx_unlock(&mountlist_mtx);
2744 MPASS((mp->mnt_kern_flag & MNTK_SUSPEND) != 0);
2745 mp->mnt_kern_flag &= ~MNTK_SUSPEND_ALL;
2747 vfs_write_resume(mp, 0);
2748 mtx_lock(&mountlist_mtx);
2751 mtx_unlock(&mountlist_mtx);