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 #include <sys/param.h>
43 #include <sys/devctl.h>
44 #include <sys/eventhandler.h>
45 #include <sys/fcntl.h>
47 #include <sys/kernel.h>
49 #include <sys/libkern.h>
50 #include <sys/limits.h>
51 #include <sys/malloc.h>
52 #include <sys/mount.h>
53 #include <sys/mutex.h>
54 #include <sys/namei.h>
57 #include <sys/filedesc.h>
58 #include <sys/reboot.h>
60 #include <sys/syscallsubr.h>
61 #include <sys/sysproto.h>
63 #include <sys/sysctl.h>
64 #include <sys/systm.h>
65 #include <sys/taskqueue.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, bool jail_export,
80 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 static bool recursive_forced_unmount = false;
92 SYSCTL_BOOL(_vfs, OID_AUTO, recursive_forced_unmount, CTLFLAG_RW,
93 &recursive_forced_unmount, 0, "Recursively unmount stacked upper mounts"
94 " when a file system is forcibly unmounted");
96 static SYSCTL_NODE(_vfs, OID_AUTO, deferred_unmount,
97 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "deferred unmount controls");
99 static unsigned int deferred_unmount_retry_limit = 10;
100 SYSCTL_UINT(_vfs_deferred_unmount, OID_AUTO, retry_limit, CTLFLAG_RW,
101 &deferred_unmount_retry_limit, 0,
102 "Maximum number of retries for deferred unmount failure");
104 static int deferred_unmount_retry_delay_hz;
105 SYSCTL_INT(_vfs_deferred_unmount, OID_AUTO, retry_delay_hz, CTLFLAG_RW,
106 &deferred_unmount_retry_delay_hz, 0,
107 "Delay in units of [1/kern.hz]s when retrying a failed deferred unmount");
109 static int deferred_unmount_total_retries = 0;
110 SYSCTL_INT(_vfs_deferred_unmount, OID_AUTO, total_retries, CTLFLAG_RD,
111 &deferred_unmount_total_retries, 0,
112 "Total number of retried deferred unmounts");
114 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
115 MALLOC_DEFINE(M_STATFS, "statfs", "statfs structure");
116 static uma_zone_t mount_zone;
118 /* List of mounted filesystems. */
119 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
121 /* For any iteration/modification of mountlist */
122 struct mtx_padalign __exclusive_cache_line mountlist_mtx;
124 EVENTHANDLER_LIST_DEFINE(vfs_mounted);
125 EVENTHANDLER_LIST_DEFINE(vfs_unmounted);
127 static void vfs_deferred_unmount(void *arg, int pending);
128 static struct timeout_task deferred_unmount_task;
129 static struct mtx deferred_unmount_lock;
130 MTX_SYSINIT(deferred_unmount, &deferred_unmount_lock, "deferred_unmount",
132 static STAILQ_HEAD(, mount) deferred_unmount_list =
133 STAILQ_HEAD_INITIALIZER(deferred_unmount_list);
134 TASKQUEUE_DEFINE_THREAD(deferred_unmount);
136 static void mount_devctl_event(const char *type, struct mount *mp, bool donew);
139 * Global opts, taken by all filesystems
141 static const char *global_opts[] = {
153 mount_init(void *mem, int size, int flags)
157 mp = (struct mount *)mem;
158 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
159 mtx_init(&mp->mnt_listmtx, "struct mount vlist mtx", NULL, MTX_DEF);
160 lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0);
161 mp->mnt_pcpu = uma_zalloc_pcpu(pcpu_zone_16, M_WAITOK | M_ZERO);
164 mp->mnt_rootvnode = NULL;
169 mount_fini(void *mem, int size)
173 mp = (struct mount *)mem;
174 uma_zfree_pcpu(pcpu_zone_16, mp->mnt_pcpu);
175 lockdestroy(&mp->mnt_explock);
176 mtx_destroy(&mp->mnt_listmtx);
177 mtx_destroy(&mp->mnt_mtx);
181 vfs_mount_init(void *dummy __unused)
183 TIMEOUT_TASK_INIT(taskqueue_deferred_unmount, &deferred_unmount_task,
184 0, vfs_deferred_unmount, NULL);
185 deferred_unmount_retry_delay_hz = hz;
186 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), NULL,
187 NULL, mount_init, mount_fini, UMA_ALIGN_CACHE, UMA_ZONE_NOFREE);
188 mtx_init(&mountlist_mtx, "mountlist", NULL, MTX_DEF);
190 SYSINIT(vfs_mount, SI_SUB_VFS, SI_ORDER_ANY, vfs_mount_init, NULL);
193 * ---------------------------------------------------------------------
194 * Functions for building and sanitizing the mount options
197 /* Remove one mount option. */
199 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
202 TAILQ_REMOVE(opts, opt, link);
203 free(opt->name, M_MOUNT);
204 if (opt->value != NULL)
205 free(opt->value, M_MOUNT);
209 /* Release all resources related to the mount options. */
211 vfs_freeopts(struct vfsoptlist *opts)
215 while (!TAILQ_EMPTY(opts)) {
216 opt = TAILQ_FIRST(opts);
217 vfs_freeopt(opts, opt);
223 vfs_deleteopt(struct vfsoptlist *opts, const char *name)
225 struct vfsopt *opt, *temp;
229 TAILQ_FOREACH_SAFE(opt, opts, link, temp) {
230 if (strcmp(opt->name, name) == 0)
231 vfs_freeopt(opts, opt);
236 vfs_isopt_ro(const char *opt)
239 if (strcmp(opt, "ro") == 0 || strcmp(opt, "rdonly") == 0 ||
240 strcmp(opt, "norw") == 0)
246 vfs_isopt_rw(const char *opt)
249 if (strcmp(opt, "rw") == 0 || strcmp(opt, "noro") == 0)
255 * Check if options are equal (with or without the "no" prefix).
258 vfs_equalopts(const char *opt1, const char *opt2)
262 /* "opt" vs. "opt" or "noopt" vs. "noopt" */
263 if (strcmp(opt1, opt2) == 0)
265 /* "noopt" vs. "opt" */
266 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
268 /* "opt" vs. "noopt" */
269 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
271 while ((p = strchr(opt1, '.')) != NULL &&
272 !strncmp(opt1, opt2, ++p - opt1)) {
275 /* "foo.noopt" vs. "foo.opt" */
276 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
278 /* "foo.opt" vs. "foo.noopt" */
279 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
282 /* "ro" / "rdonly" / "norw" / "rw" / "noro" */
283 if ((vfs_isopt_ro(opt1) || vfs_isopt_rw(opt1)) &&
284 (vfs_isopt_ro(opt2) || vfs_isopt_rw(opt2)))
290 * If a mount option is specified several times,
291 * (with or without the "no" prefix) only keep
292 * the last occurrence of it.
295 vfs_sanitizeopts(struct vfsoptlist *opts)
297 struct vfsopt *opt, *opt2, *tmp;
299 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
300 opt2 = TAILQ_PREV(opt, vfsoptlist, link);
301 while (opt2 != NULL) {
302 if (vfs_equalopts(opt->name, opt2->name)) {
303 tmp = TAILQ_PREV(opt2, vfsoptlist, link);
304 vfs_freeopt(opts, opt2);
307 opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
314 * Build a linked list of mount options from a struct uio.
317 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
319 struct vfsoptlist *opts;
321 size_t memused, namelen, optlen;
322 unsigned int i, iovcnt;
325 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
328 iovcnt = auio->uio_iovcnt;
329 for (i = 0; i < iovcnt; i += 2) {
330 namelen = auio->uio_iov[i].iov_len;
331 optlen = auio->uio_iov[i + 1].iov_len;
332 memused += sizeof(struct vfsopt) + optlen + namelen;
334 * Avoid consuming too much memory, and attempts to overflow
337 if (memused > VFS_MOUNTARG_SIZE_MAX ||
338 optlen > VFS_MOUNTARG_SIZE_MAX ||
339 namelen > VFS_MOUNTARG_SIZE_MAX) {
344 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
345 opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
352 * Do this early, so jumps to "bad" will free the current
355 TAILQ_INSERT_TAIL(opts, opt, link);
357 if (auio->uio_segflg == UIO_SYSSPACE) {
358 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
360 error = copyin(auio->uio_iov[i].iov_base, opt->name,
365 /* Ensure names are null-terminated strings. */
366 if (namelen == 0 || opt->name[namelen - 1] != '\0') {
372 opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
373 if (auio->uio_segflg == UIO_SYSSPACE) {
374 bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
377 error = copyin(auio->uio_iov[i + 1].iov_base,
384 vfs_sanitizeopts(opts);
393 * Merge the old mount options with the new ones passed
394 * in the MNT_UPDATE case.
396 * XXX: This function will keep a "nofoo" option in the new
397 * options. E.g, if the option's canonical name is "foo",
398 * "nofoo" ends up in the mount point's active options.
401 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *oldopts)
403 struct vfsopt *opt, *new;
405 TAILQ_FOREACH(opt, oldopts, link) {
406 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
407 new->name = strdup(opt->name, M_MOUNT);
409 new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
410 bcopy(opt->value, new->value, opt->len);
414 new->seen = opt->seen;
415 TAILQ_INSERT_HEAD(toopts, new, link);
417 vfs_sanitizeopts(toopts);
421 * Mount a filesystem.
423 #ifndef _SYS_SYSPROTO_H_
431 sys_nmount(struct thread *td, struct nmount_args *uap)
439 * Mount flags are now 64-bits. On 32-bit archtectures only
440 * 32-bits are passed in, but from here on everything handles
441 * 64-bit flags correctly.
445 AUDIT_ARG_FFLAGS(flags);
446 CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__,
447 uap->iovp, uap->iovcnt, flags);
450 * Filter out MNT_ROOTFS. We do not want clients of nmount() in
451 * userspace to set this flag, but we must filter it out if we want
452 * MNT_UPDATE on the root file system to work.
453 * MNT_ROOTFS should only be set by the kernel when mounting its
456 flags &= ~MNT_ROOTFS;
458 iovcnt = uap->iovcnt;
460 * Check that we have an even number of iovec's
461 * and that we have at least two options.
463 if ((iovcnt & 1) || (iovcnt < 4)) {
464 CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__,
469 error = copyinuio(uap->iovp, iovcnt, &auio);
471 CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno",
475 error = vfs_donmount(td, flags, auio);
482 * ---------------------------------------------------------------------
483 * Various utility functions
487 * Get a reference on a mount point from a vnode.
489 * The vnode is allowed to be passed unlocked and race against dooming. Note in
490 * such case there are no guarantees the referenced mount point will still be
491 * associated with it after the function returns.
494 vfs_ref_from_vp(struct vnode *vp)
497 struct mount_pcpu *mpcpu;
499 mp = atomic_load_ptr(&vp->v_mount);
500 if (__predict_false(mp == NULL)) {
503 if (vfs_op_thread_enter(mp, mpcpu)) {
504 if (__predict_true(mp == vp->v_mount)) {
505 vfs_mp_count_add_pcpu(mpcpu, ref, 1);
506 vfs_op_thread_exit(mp, mpcpu);
508 vfs_op_thread_exit(mp, mpcpu);
513 if (mp == vp->v_mount) {
525 vfs_ref(struct mount *mp)
527 struct mount_pcpu *mpcpu;
529 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
530 if (vfs_op_thread_enter(mp, mpcpu)) {
531 vfs_mp_count_add_pcpu(mpcpu, ref, 1);
532 vfs_op_thread_exit(mp, mpcpu);
542 * Register ump as an upper mount of the mount associated with
543 * vnode vp. This registration will be tracked through
544 * mount_upper_node upper, which should be allocated by the
545 * caller and stored in per-mount data associated with mp.
547 * If successful, this function will return the mount associated
548 * with vp, and will ensure that it cannot be unmounted until
549 * ump has been unregistered as one of its upper mounts.
551 * Upon failure this function will return NULL.
554 vfs_register_upper_from_vp(struct vnode *vp, struct mount *ump,
555 struct mount_upper_node *upper)
559 mp = atomic_load_ptr(&vp->v_mount);
563 if (mp != vp->v_mount ||
564 ((mp->mnt_kern_flag & (MNTK_UNMOUNT | MNTK_RECURSE)) != 0)) {
568 KASSERT(ump != mp, ("upper and lower mounts are identical"));
571 TAILQ_INSERT_TAIL(&mp->mnt_uppers, upper, mnt_upper_link);
577 * Register upper mount ump to receive vnode unlink/reclaim
578 * notifications from lower mount mp. This registration will
579 * be tracked through mount_upper_node upper, which should be
580 * allocated by the caller and stored in per-mount data
581 * associated with mp.
583 * ump must already be registered as an upper mount of mp
584 * through a call to vfs_register_upper_from_vp().
587 vfs_register_for_notification(struct mount *mp, struct mount *ump,
588 struct mount_upper_node *upper)
592 TAILQ_INSERT_TAIL(&mp->mnt_notify, upper, mnt_upper_link);
597 vfs_drain_upper_locked(struct mount *mp)
599 mtx_assert(MNT_MTX(mp), MA_OWNED);
600 while (mp->mnt_upper_pending != 0) {
601 mp->mnt_kern_flag |= MNTK_UPPER_WAITER;
602 msleep(&mp->mnt_uppers, MNT_MTX(mp), 0, "mntupw", 0);
607 * Undo a previous call to vfs_register_for_notification().
608 * The mount represented by upper must be currently registered
609 * as an upper mount for mp.
612 vfs_unregister_for_notification(struct mount *mp,
613 struct mount_upper_node *upper)
616 vfs_drain_upper_locked(mp);
617 TAILQ_REMOVE(&mp->mnt_notify, upper, mnt_upper_link);
622 * Undo a previous call to vfs_register_upper_from_vp().
623 * This must be done before mp can be unmounted.
626 vfs_unregister_upper(struct mount *mp, struct mount_upper_node *upper)
629 KASSERT((mp->mnt_kern_flag & MNTK_UNMOUNT) == 0,
630 ("registered upper with pending unmount"));
631 vfs_drain_upper_locked(mp);
632 TAILQ_REMOVE(&mp->mnt_uppers, upper, mnt_upper_link);
633 if ((mp->mnt_kern_flag & MNTK_TASKQUEUE_WAITER) != 0 &&
634 TAILQ_EMPTY(&mp->mnt_uppers)) {
635 mp->mnt_kern_flag &= ~MNTK_TASKQUEUE_WAITER;
636 wakeup(&mp->mnt_taskqueue_link);
643 vfs_rel(struct mount *mp)
645 struct mount_pcpu *mpcpu;
647 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
648 if (vfs_op_thread_enter(mp, mpcpu)) {
649 vfs_mp_count_sub_pcpu(mpcpu, ref, 1);
650 vfs_op_thread_exit(mp, mpcpu);
660 * Allocate and initialize the mount point struct.
663 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath,
668 mp = uma_zalloc(mount_zone, M_WAITOK);
669 bzero(&mp->mnt_startzero,
670 __rangeof(struct mount, mnt_startzero, mnt_endzero));
671 mp->mnt_kern_flag = 0;
673 mp->mnt_rootvnode = NULL;
674 mp->mnt_vnodecovered = NULL;
677 TAILQ_INIT(&mp->mnt_nvnodelist);
678 mp->mnt_nvnodelistsize = 0;
679 TAILQ_INIT(&mp->mnt_lazyvnodelist);
680 mp->mnt_lazyvnodelistsize = 0;
681 MPPASS(mp->mnt_ref == 0 && mp->mnt_lockref == 0 &&
682 mp->mnt_writeopcount == 0, mp);
683 MPASSERT(mp->mnt_vfs_ops == 1, mp,
684 ("vfs_ops should be 1 but %d found", mp->mnt_vfs_ops));
685 (void) vfs_busy(mp, MBF_NOWAIT);
686 atomic_add_acq_int(&vfsp->vfc_refcount, 1);
687 mp->mnt_op = vfsp->vfc_vfsops;
689 mp->mnt_stat.f_type = vfsp->vfc_typenum;
691 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
692 mp->mnt_vnodecovered = vp;
693 mp->mnt_cred = crdup(cred);
694 mp->mnt_stat.f_owner = cred->cr_uid;
695 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
696 mp->mnt_iosize_max = DFLTPHYS;
699 mac_mount_create(cred, mp);
701 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
702 mp->mnt_upper_pending = 0;
703 TAILQ_INIT(&mp->mnt_uppers);
704 TAILQ_INIT(&mp->mnt_notify);
705 mp->mnt_taskqueue_flags = 0;
706 mp->mnt_unmount_retries = 0;
711 * Destroy the mount struct previously allocated by vfs_mount_alloc().
714 vfs_mount_destroy(struct mount *mp)
717 MPPASS(mp->mnt_vfs_ops != 0, mp);
719 vfs_assert_mount_counters(mp);
722 mp->mnt_kern_flag |= MNTK_REFEXPIRE;
723 if (mp->mnt_kern_flag & MNTK_MWAIT) {
724 mp->mnt_kern_flag &= ~MNTK_MWAIT;
728 msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0);
729 KASSERT(mp->mnt_ref == 0,
730 ("%s: invalid refcount in the drain path @ %s:%d", __func__,
731 __FILE__, __LINE__));
732 MPPASS(mp->mnt_writeopcount == 0, mp);
733 MPPASS(mp->mnt_secondary_writes == 0, mp);
734 atomic_subtract_rel_int(&mp->mnt_vfc->vfc_refcount, 1);
735 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
738 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
739 vn_printf(vp, "dangling vnode ");
740 panic("unmount: dangling vnode");
742 KASSERT(mp->mnt_upper_pending == 0, ("mnt_upper_pending"));
743 KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers"));
744 KASSERT(TAILQ_EMPTY(&mp->mnt_notify), ("mnt_notify"));
745 MPPASS(mp->mnt_nvnodelistsize == 0, mp);
746 MPPASS(mp->mnt_lazyvnodelistsize == 0, mp);
747 MPPASS(mp->mnt_lockref == 0, mp);
750 MPASSERT(mp->mnt_vfs_ops == 1, mp,
751 ("vfs_ops should be 1 but %d found", mp->mnt_vfs_ops));
753 MPASSERT(mp->mnt_rootvnode == NULL, mp,
754 ("mount point still has a root vnode %p", mp->mnt_rootvnode));
756 if (mp->mnt_vnodecovered != NULL)
757 vrele(mp->mnt_vnodecovered);
759 mac_mount_destroy(mp);
761 if (mp->mnt_opt != NULL)
762 vfs_freeopts(mp->mnt_opt);
763 if (mp->mnt_exjail != NULL) {
764 atomic_subtract_int(&mp->mnt_exjail->cr_prison->pr_exportcnt,
766 crfree(mp->mnt_exjail);
768 if (mp->mnt_export != NULL) {
769 vfs_free_addrlist(mp->mnt_export);
770 free(mp->mnt_export, M_MOUNT);
772 crfree(mp->mnt_cred);
773 uma_zfree(mount_zone, mp);
777 vfs_should_downgrade_to_ro_mount(uint64_t fsflags, int error)
779 /* This is an upgrade of an exisiting mount. */
780 if ((fsflags & MNT_UPDATE) != 0)
782 /* This is already an R/O mount. */
783 if ((fsflags & MNT_RDONLY) != 0)
787 case ENODEV: /* generic, geom, ... */
788 case EACCES: /* cam/scsi, ... */
789 case EROFS: /* md, mmcsd, ... */
791 * These errors can be returned by the storage layer to signal
792 * that the media is read-only. No harm in the R/O mount
793 * attempt if the error was returned for some other reason.
802 vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions)
804 struct vfsoptlist *optlist;
805 struct vfsopt *opt, *tmp_opt;
806 char *fstype, *fspath, *errmsg;
807 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
808 bool autoro, has_nonexport, jail_export;
810 errmsg = fspath = NULL;
811 errmsg_len = fspathlen = 0;
813 autoro = default_autoro;
815 error = vfs_buildopts(fsoptions, &optlist);
819 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
820 errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
823 * We need these two options before the others,
824 * and they are mandatory for any filesystem.
825 * Ensure they are NUL terminated as well.
828 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
829 if (error || fstypelen <= 0 || fstype[fstypelen - 1] != '\0') {
832 strncpy(errmsg, "Invalid fstype", errmsg_len);
836 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
837 if (error || fspathlen <= 0 || fspath[fspathlen - 1] != '\0') {
840 strncpy(errmsg, "Invalid fspath", errmsg_len);
845 * Check to see that "export" is only used with the "update", "fstype",
846 * "fspath", "from" and "errmsg" options when in a vnet jail.
847 * These are the ones used to set/update exports by mountd(8).
848 * If only the above options are set in a jail that can run mountd(8),
849 * then the jail_export argument of vfs_domount() will be true.
850 * When jail_export is true, the vfs_suser() check does not cause
851 * failure, but limits the update to exports only.
852 * This allows mountd(8) running within the vnet jail
853 * to export file systems visible within the jail, but
854 * mounted outside of the jail.
857 * We need to see if we have the "update" option
858 * before we call vfs_domount(), since vfs_domount() has special
859 * logic based on MNT_UPDATE. This is very important
860 * when we want to update the root filesystem.
862 has_nonexport = false;
864 TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) {
867 if (jailed(td->td_ucred) &&
868 strcmp(opt->name, "export") != 0 &&
869 strcmp(opt->name, "update") != 0 &&
870 strcmp(opt->name, "fstype") != 0 &&
871 strcmp(opt->name, "fspath") != 0 &&
872 strcmp(opt->name, "from") != 0 &&
873 strcmp(opt->name, "errmsg") != 0)
874 has_nonexport = true;
875 if (strcmp(opt->name, "update") == 0) {
876 fsflags |= MNT_UPDATE;
879 else if (strcmp(opt->name, "async") == 0)
880 fsflags |= MNT_ASYNC;
881 else if (strcmp(opt->name, "force") == 0) {
882 fsflags |= MNT_FORCE;
885 else if (strcmp(opt->name, "reload") == 0) {
886 fsflags |= MNT_RELOAD;
889 else if (strcmp(opt->name, "multilabel") == 0)
890 fsflags |= MNT_MULTILABEL;
891 else if (strcmp(opt->name, "noasync") == 0)
892 fsflags &= ~MNT_ASYNC;
893 else if (strcmp(opt->name, "noatime") == 0)
894 fsflags |= MNT_NOATIME;
895 else if (strcmp(opt->name, "atime") == 0) {
896 free(opt->name, M_MOUNT);
897 opt->name = strdup("nonoatime", M_MOUNT);
899 else if (strcmp(opt->name, "noclusterr") == 0)
900 fsflags |= MNT_NOCLUSTERR;
901 else if (strcmp(opt->name, "clusterr") == 0) {
902 free(opt->name, M_MOUNT);
903 opt->name = strdup("nonoclusterr", M_MOUNT);
905 else if (strcmp(opt->name, "noclusterw") == 0)
906 fsflags |= MNT_NOCLUSTERW;
907 else if (strcmp(opt->name, "clusterw") == 0) {
908 free(opt->name, M_MOUNT);
909 opt->name = strdup("nonoclusterw", M_MOUNT);
911 else if (strcmp(opt->name, "noexec") == 0)
912 fsflags |= MNT_NOEXEC;
913 else if (strcmp(opt->name, "exec") == 0) {
914 free(opt->name, M_MOUNT);
915 opt->name = strdup("nonoexec", M_MOUNT);
917 else if (strcmp(opt->name, "nosuid") == 0)
918 fsflags |= MNT_NOSUID;
919 else if (strcmp(opt->name, "suid") == 0) {
920 free(opt->name, M_MOUNT);
921 opt->name = strdup("nonosuid", M_MOUNT);
923 else if (strcmp(opt->name, "nosymfollow") == 0)
924 fsflags |= MNT_NOSYMFOLLOW;
925 else if (strcmp(opt->name, "symfollow") == 0) {
926 free(opt->name, M_MOUNT);
927 opt->name = strdup("nonosymfollow", M_MOUNT);
929 else if (strcmp(opt->name, "noro") == 0) {
930 fsflags &= ~MNT_RDONLY;
933 else if (strcmp(opt->name, "rw") == 0) {
934 fsflags &= ~MNT_RDONLY;
937 else if (strcmp(opt->name, "ro") == 0) {
938 fsflags |= MNT_RDONLY;
941 else if (strcmp(opt->name, "rdonly") == 0) {
942 free(opt->name, M_MOUNT);
943 opt->name = strdup("ro", M_MOUNT);
944 fsflags |= MNT_RDONLY;
947 else if (strcmp(opt->name, "autoro") == 0) {
951 else if (strcmp(opt->name, "suiddir") == 0)
952 fsflags |= MNT_SUIDDIR;
953 else if (strcmp(opt->name, "sync") == 0)
954 fsflags |= MNT_SYNCHRONOUS;
955 else if (strcmp(opt->name, "union") == 0)
956 fsflags |= MNT_UNION;
957 else if (strcmp(opt->name, "export") == 0) {
958 fsflags |= MNT_EXPORTED;
960 } else if (strcmp(opt->name, "automounted") == 0) {
961 fsflags |= MNT_AUTOMOUNTED;
963 } else if (strcmp(opt->name, "nocover") == 0) {
964 fsflags |= MNT_NOCOVER;
966 } else if (strcmp(opt->name, "cover") == 0) {
967 fsflags &= ~MNT_NOCOVER;
969 } else if (strcmp(opt->name, "emptydir") == 0) {
970 fsflags |= MNT_EMPTYDIR;
972 } else if (strcmp(opt->name, "noemptydir") == 0) {
973 fsflags &= ~MNT_EMPTYDIR;
977 vfs_freeopt(optlist, opt);
981 * Be ultra-paranoid about making sure the type and fspath
982 * variables will fit in our mp buffers, including the
985 if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) {
986 error = ENAMETOOLONG;
991 * If has_nonexport is true or the caller is not running within a
992 * vnet prison that can run mountd(8), set jail_export false.
994 if (has_nonexport || !jailed(td->td_ucred) ||
995 !prison_check_nfsd(td->td_ucred))
998 error = vfs_domount(td, fstype, fspath, fsflags, jail_export, &optlist);
999 if (error == ENOENT) {
1002 strncpy(errmsg, "Invalid fstype", errmsg_len);
1007 * See if we can mount in the read-only mode if the error code suggests
1008 * that it could be possible and the mount options allow for that.
1009 * Never try it if "[no]{ro|rw}" has been explicitly requested and not
1010 * overridden by "autoro".
1012 if (autoro && vfs_should_downgrade_to_ro_mount(fsflags, error)) {
1013 printf("%s: R/W mount failed, possibly R/O media,"
1014 " trying R/O mount\n", __func__);
1015 fsflags |= MNT_RDONLY;
1016 error = vfs_domount(td, fstype, fspath, fsflags, jail_export,
1020 /* copyout the errmsg */
1021 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
1022 && errmsg_len > 0 && errmsg != NULL) {
1023 if (fsoptions->uio_segflg == UIO_SYSSPACE) {
1025 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
1026 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
1029 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
1030 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
1034 if (optlist != NULL)
1035 vfs_freeopts(optlist);
1042 #ifndef _SYS_SYSPROTO_H_
1052 sys_mount(struct thread *td, struct mount_args *uap)
1055 struct vfsconf *vfsp = NULL;
1056 struct mntarg *ma = NULL;
1061 * Mount flags are now 64-bits. On 32-bit architectures only
1062 * 32-bits are passed in, but from here on everything handles
1063 * 64-bit flags correctly.
1067 AUDIT_ARG_FFLAGS(flags);
1070 * Filter out MNT_ROOTFS. We do not want clients of mount() in
1071 * userspace to set this flag, but we must filter it out if we want
1072 * MNT_UPDATE on the root file system to work.
1073 * MNT_ROOTFS should only be set by the kernel when mounting its
1076 flags &= ~MNT_ROOTFS;
1078 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
1079 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
1081 free(fstype, M_TEMP);
1085 AUDIT_ARG_TEXT(fstype);
1086 vfsp = vfs_byname_kld(fstype, td, &error);
1087 free(fstype, M_TEMP);
1090 if (((vfsp->vfc_flags & VFCF_SBDRY) != 0 &&
1091 vfsp->vfc_vfsops_sd->vfs_cmount == NULL) ||
1092 ((vfsp->vfc_flags & VFCF_SBDRY) == 0 &&
1093 vfsp->vfc_vfsops->vfs_cmount == NULL))
1094 return (EOPNOTSUPP);
1096 ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN);
1097 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
1098 ma = mount_argb(ma, flags & MNT_RDONLY, "noro");
1099 ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid");
1100 ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec");
1102 if ((vfsp->vfc_flags & VFCF_SBDRY) != 0)
1103 return (vfsp->vfc_vfsops_sd->vfs_cmount(ma, uap->data, flags));
1104 return (vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags));
1108 * vfs_domount_first(): first file system mount (not update)
1112 struct thread *td, /* Calling thread. */
1113 struct vfsconf *vfsp, /* File system type. */
1114 char *fspath, /* Mount path. */
1115 struct vnode *vp, /* Vnode to be covered. */
1116 uint64_t fsflags, /* Flags common to all filesystems. */
1117 struct vfsoptlist **optlist /* Options local to the filesystem. */
1122 struct vnode *newdp, *rootvp;
1126 ASSERT_VOP_ELOCKED(vp, __func__);
1127 KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here"));
1130 * If the jail of the calling thread lacks permission for this type of
1131 * file system, or is trying to cover its own root, deny immediately.
1133 if (jailed(td->td_ucred) && (!prison_allow(td->td_ucred,
1134 vfsp->vfc_prison_flag) || vp == td->td_ucred->cr_prison->pr_root)) {
1140 * If the user is not root, ensure that they own the directory
1141 * onto which we are attempting to mount.
1143 error = VOP_GETATTR(vp, &va, td->td_ucred);
1144 if (error == 0 && va.va_uid != td->td_ucred->cr_uid)
1145 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN);
1147 error = vinvalbuf(vp, V_SAVE, 0, 0);
1148 if (vfsp->vfc_flags & VFCF_FILEMOUNT) {
1149 if (error == 0 && vp->v_type != VDIR && vp->v_type != VREG)
1152 * For file mounts, ensure that there is only one hardlink to the file.
1154 if (error == 0 && vp->v_type == VREG && va.va_nlink != 1)
1157 if (error == 0 && vp->v_type != VDIR)
1160 if (error == 0 && (fsflags & MNT_EMPTYDIR) != 0)
1161 error = vn_dir_check_empty(vp);
1164 if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL)
1165 vp->v_iflag |= VI_MOUNT;
1174 vn_seqc_write_begin(vp);
1177 /* Allocate and initialize the filesystem. */
1178 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred);
1179 /* XXXMAC: pass to vfs_mount_alloc? */
1180 mp->mnt_optnew = *optlist;
1181 /* Set the mount level flags. */
1182 mp->mnt_flag = (fsflags &
1183 (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY | MNT_FORCE));
1186 * Mount the filesystem.
1187 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1188 * get. No freeing of cn_pnbuf.
1192 if ((error = VFS_MOUNT(mp)) != 0 ||
1193 (error1 = VFS_STATFS(mp, &mp->mnt_stat)) != 0 ||
1194 (error1 = VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) != 0) {
1198 rootvp = vfs_cache_root_clear(mp);
1199 if (rootvp != NULL) {
1203 (void)vn_start_write(NULL, &mp, V_WAIT);
1205 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_UNMOUNTF;
1208 error = VFS_UNMOUNT(mp, 0);
1209 vn_finished_write(mp);
1212 "failed post-mount (%d): rollback unmount returned %d\n",
1219 mp->mnt_vnodecovered = NULL;
1221 /* XXXKIB wait for mnt_lockref drain? */
1222 vfs_mount_destroy(mp);
1225 vp->v_iflag &= ~VI_MOUNT;
1227 if (rootvp != NULL) {
1228 vn_seqc_write_end(rootvp);
1231 vn_seqc_write_end(vp);
1235 vn_seqc_write_begin(newdp);
1238 if (mp->mnt_opt != NULL)
1239 vfs_freeopts(mp->mnt_opt);
1240 mp->mnt_opt = mp->mnt_optnew;
1244 * Prevent external consumers of mount options from reading mnt_optnew.
1246 mp->mnt_optnew = NULL;
1249 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1250 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1251 mp->mnt_kern_flag |= MNTK_ASYNC;
1253 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1257 * VIRF_MOUNTPOINT and v_mountedhere need to be set under the
1258 * vp lock to satisfy vfs_lookup() requirements.
1260 VOP_LOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1262 vn_irflag_set_locked(vp, VIRF_MOUNTPOINT);
1263 vp->v_mountedhere = mp;
1269 * We need to lock both vnodes.
1271 * Use vn_lock_pair to avoid establishing an ordering between vnodes
1272 * from different filesystems.
1274 vn_lock_pair(vp, false, LK_EXCLUSIVE, newdp, false, LK_EXCLUSIVE);
1277 vp->v_iflag &= ~VI_MOUNT;
1279 /* Place the new filesystem at the end of the mount list. */
1280 mtx_lock(&mountlist_mtx);
1281 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1282 mtx_unlock(&mountlist_mtx);
1283 vfs_event_signal(NULL, VQ_MOUNT, 0);
1285 EVENTHANDLER_DIRECT_INVOKE(vfs_mounted, mp, newdp, td);
1287 mount_devctl_event("MOUNT", mp, false);
1288 mountcheckdirs(vp, newdp);
1289 vn_seqc_write_end(vp);
1290 vn_seqc_write_end(newdp);
1292 if ((mp->mnt_flag & MNT_RDONLY) == 0)
1293 vfs_allocate_syncvnode(mp);
1300 * vfs_domount_update(): update of mounted file system
1304 struct thread *td, /* Calling thread. */
1305 struct vnode *vp, /* Mount point vnode. */
1306 uint64_t fsflags, /* Flags common to all filesystems. */
1307 bool jail_export, /* Got export option in vnet prison. */
1308 struct vfsoptlist **optlist /* Options local to the filesystem. */
1311 struct export_args export;
1312 struct o2export_args o2export;
1313 struct vnode *rootvp;
1316 int error, export_error, i, len;
1319 bool vfs_suser_failed;
1321 ASSERT_VOP_ELOCKED(vp, __func__);
1322 KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here"));
1325 if ((vp->v_vflag & VV_ROOT) == 0) {
1326 if (vfs_copyopt(*optlist, "export", &export, sizeof(export))
1336 * We only allow the filesystem to be reloaded if it
1337 * is currently mounted read-only.
1339 flag = mp->mnt_flag;
1340 if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) {
1342 return (EOPNOTSUPP); /* Needs translation */
1345 * Only privileged root, or (if MNT_USER is set) the user that
1346 * did the original mount is permitted to update it.
1349 * For the case of mountd(8) doing exports in a jail, the vfs_suser()
1350 * call does not cause failure. vfs_domount() has already checked
1351 * that "root" is doing this and vfs_suser() will fail when
1352 * the file system has been mounted outside the jail.
1353 * jail_export set true indicates that "export" is not mixed
1354 * with other options that change mount behaviour.
1356 vfs_suser_failed = false;
1357 error = vfs_suser(mp, td);
1358 if (jail_export && error != 0) {
1360 vfs_suser_failed = true;
1366 if (vfs_busy(mp, MBF_NOWAIT)) {
1371 if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) {
1377 vp->v_iflag |= VI_MOUNT;
1382 vn_seqc_write_begin(vp);
1386 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1391 if (vfs_suser_failed) {
1392 KASSERT((fsflags & (MNT_EXPORTED | MNT_UPDATE)) ==
1393 (MNT_EXPORTED | MNT_UPDATE),
1394 ("%s: jailed export did not set expected fsflags",
1397 * For this case, only MNT_UPDATE and
1398 * MNT_EXPORTED have been set in fsflags
1399 * by the options. Only set MNT_UPDATE,
1400 * since that is the one that would be set
1401 * when set in fsflags, below.
1403 mp->mnt_flag |= MNT_UPDATE;
1405 mp->mnt_flag &= ~MNT_UPDATEMASK;
1406 mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE |
1407 MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY);
1408 if ((mp->mnt_flag & MNT_ASYNC) == 0)
1409 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1411 rootvp = vfs_cache_root_clear(mp);
1413 mp->mnt_optnew = *optlist;
1414 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
1417 * Mount the filesystem.
1418 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1419 * get. No freeing of cn_pnbuf.
1422 * For the case of mountd(8) doing exports from within a vnet jail,
1423 * "from" is typically not set correctly such that VFS_MOUNT() will
1424 * return ENOENT. It is not obvious that VFS_MOUNT() ever needs to be
1425 * called when mountd is doing exports, but this check only applies to
1426 * the specific case where it is running inside a vnet jail, to
1427 * avoid any POLA violation.
1431 error = VFS_MOUNT(mp);
1434 /* Process the export option. */
1435 if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp,
1437 /* Assume that there is only 1 ABI for each length. */
1439 case (sizeof(struct oexport_args)):
1440 bzero(&o2export, sizeof(o2export));
1442 case (sizeof(o2export)):
1443 bcopy(bufp, &o2export, len);
1444 export.ex_flags = (uint64_t)o2export.ex_flags;
1445 export.ex_root = o2export.ex_root;
1446 export.ex_uid = o2export.ex_anon.cr_uid;
1447 export.ex_groups = NULL;
1448 export.ex_ngroups = o2export.ex_anon.cr_ngroups;
1449 if (export.ex_ngroups > 0) {
1450 if (export.ex_ngroups <= XU_NGROUPS) {
1451 export.ex_groups = malloc(
1452 export.ex_ngroups * sizeof(gid_t),
1454 for (i = 0; i < export.ex_ngroups; i++)
1455 export.ex_groups[i] =
1456 o2export.ex_anon.cr_groups[i];
1458 export_error = EINVAL;
1459 } else if (export.ex_ngroups < 0)
1460 export_error = EINVAL;
1461 export.ex_addr = o2export.ex_addr;
1462 export.ex_addrlen = o2export.ex_addrlen;
1463 export.ex_mask = o2export.ex_mask;
1464 export.ex_masklen = o2export.ex_masklen;
1465 export.ex_indexfile = o2export.ex_indexfile;
1466 export.ex_numsecflavors = o2export.ex_numsecflavors;
1467 if (export.ex_numsecflavors < MAXSECFLAVORS) {
1468 for (i = 0; i < export.ex_numsecflavors; i++)
1469 export.ex_secflavors[i] =
1470 o2export.ex_secflavors[i];
1472 export_error = EINVAL;
1473 if (export_error == 0)
1474 export_error = vfs_export(mp, &export, true);
1475 free(export.ex_groups, M_TEMP);
1477 case (sizeof(export)):
1478 bcopy(bufp, &export, len);
1480 if (export.ex_ngroups > 0) {
1481 if (export.ex_ngroups <= NGROUPS_MAX) {
1482 grps = malloc(export.ex_ngroups *
1483 sizeof(gid_t), M_TEMP, M_WAITOK);
1484 export_error = copyin(export.ex_groups,
1485 grps, export.ex_ngroups *
1487 if (export_error == 0)
1488 export.ex_groups = grps;
1490 export_error = EINVAL;
1491 } else if (export.ex_ngroups == 0)
1492 export.ex_groups = NULL;
1494 export_error = EINVAL;
1495 if (export_error == 0)
1496 export_error = vfs_export(mp, &export, true);
1500 export_error = EINVAL;
1507 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE |
1511 * If we fail, restore old mount flags. MNT_QUOTA is special,
1512 * because it is not part of MNT_UPDATEMASK, but it could have
1513 * changed in the meantime if quotactl(2) was called.
1514 * All in all we want current value of MNT_QUOTA, not the old
1517 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA);
1519 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1520 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1521 mp->mnt_kern_flag |= MNTK_ASYNC;
1523 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1529 mount_devctl_event("REMOUNT", mp, true);
1530 if (mp->mnt_opt != NULL)
1531 vfs_freeopts(mp->mnt_opt);
1532 mp->mnt_opt = mp->mnt_optnew;
1534 (void)VFS_STATFS(mp, &mp->mnt_stat);
1536 * Prevent external consumers of mount options from reading
1539 mp->mnt_optnew = NULL;
1541 if ((mp->mnt_flag & MNT_RDONLY) == 0)
1542 vfs_allocate_syncvnode(mp);
1544 vfs_deallocate_syncvnode(mp);
1547 if (rootvp != NULL) {
1548 vn_seqc_write_end(rootvp);
1551 vn_seqc_write_end(vp);
1554 vp->v_iflag &= ~VI_MOUNT;
1557 return (error != 0 ? error : export_error);
1561 * vfs_domount(): actually attempt a filesystem mount.
1565 struct thread *td, /* Calling thread. */
1566 const char *fstype, /* Filesystem type. */
1567 char *fspath, /* Mount path. */
1568 uint64_t fsflags, /* Flags common to all filesystems. */
1569 bool jail_export, /* Got export option in vnet prison. */
1570 struct vfsoptlist **optlist /* Options local to the filesystem. */
1573 struct vfsconf *vfsp;
1574 struct nameidata nd;
1580 * Be ultra-paranoid about making sure the type and fspath
1581 * variables will fit in our mp buffers, including the
1584 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
1585 return (ENAMETOOLONG);
1588 error = priv_check(td, PRIV_NFS_DAEMON);
1591 } else if (jailed(td->td_ucred) || usermount == 0) {
1592 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
1597 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
1599 if (fsflags & MNT_EXPORTED) {
1600 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
1604 if (fsflags & MNT_SUIDDIR) {
1605 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
1610 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
1612 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
1613 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
1614 fsflags |= MNT_NOSUID | MNT_USER;
1617 /* Load KLDs before we lock the covered vnode to avoid reversals. */
1619 if ((fsflags & MNT_UPDATE) == 0) {
1620 /* Don't try to load KLDs if we're mounting the root. */
1621 if (fsflags & MNT_ROOTFS) {
1622 if ((vfsp = vfs_byname(fstype)) == NULL)
1625 if ((vfsp = vfs_byname_kld(fstype, td, &error)) == NULL)
1631 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE.
1633 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1 | WANTPARENT,
1634 UIO_SYSSPACE, fspath);
1640 * Don't allow stacking file mounts to work around problems with the way
1641 * that namei sets nd.ni_dvp to vp_crossmp for these.
1643 if (vp->v_type == VREG)
1644 fsflags |= MNT_NOCOVER;
1645 if ((fsflags & MNT_UPDATE) == 0) {
1646 if ((vp->v_vflag & VV_ROOT) != 0 &&
1647 (fsflags & MNT_NOCOVER) != 0) {
1652 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1653 strcpy(pathbuf, fspath);
1655 * Note: we allow any vnode type here. If the path sanity check
1656 * succeeds, the type will be validated in vfs_domount_first
1659 if (vp->v_type == VDIR)
1660 error = vn_path_to_global_path(td, vp, pathbuf,
1663 error = vn_path_to_global_path_hardlink(td, vp,
1664 nd.ni_dvp, pathbuf, MNAMELEN,
1665 nd.ni_cnd.cn_nameptr, nd.ni_cnd.cn_namelen);
1667 error = vfs_domount_first(td, vfsp, pathbuf, vp,
1670 free(pathbuf, M_TEMP);
1672 error = vfs_domount_update(td, vp, fsflags, jail_export,
1683 * Unmount a filesystem.
1685 * Note: unmount takes a path to the vnode mounted on as argument, not
1686 * special file (as before).
1688 #ifndef _SYS_SYSPROTO_H_
1689 struct unmount_args {
1696 sys_unmount(struct thread *td, struct unmount_args *uap)
1699 return (kern_unmount(td, uap->path, uap->flags));
1703 kern_unmount(struct thread *td, const char *path, int flags)
1705 struct nameidata nd;
1707 char *fsidbuf, *pathbuf;
1711 AUDIT_ARG_VALUE(flags);
1712 if (jailed(td->td_ucred) || usermount == 0) {
1713 error = priv_check(td, PRIV_VFS_UNMOUNT);
1718 if (flags & MNT_BYFSID) {
1719 fsidbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1720 error = copyinstr(path, fsidbuf, MNAMELEN, NULL);
1722 free(fsidbuf, M_TEMP);
1726 AUDIT_ARG_TEXT(fsidbuf);
1727 /* Decode the filesystem ID. */
1728 if (sscanf(fsidbuf, "FSID:%d:%d", &fsid.val[0], &fsid.val[1]) != 2) {
1729 free(fsidbuf, M_TEMP);
1733 mp = vfs_getvfs(&fsid);
1734 free(fsidbuf, M_TEMP);
1739 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1740 error = copyinstr(path, pathbuf, MNAMELEN, NULL);
1742 free(pathbuf, M_TEMP);
1747 * Try to find global path for path argument.
1749 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1750 UIO_SYSSPACE, pathbuf);
1751 if (namei(&nd) == 0) {
1753 error = vn_path_to_global_path(td, nd.ni_vp, pathbuf,
1758 mtx_lock(&mountlist_mtx);
1759 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1760 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) {
1765 mtx_unlock(&mountlist_mtx);
1766 free(pathbuf, M_TEMP);
1769 * Previously we returned ENOENT for a nonexistent path and
1770 * EINVAL for a non-mountpoint. We cannot tell these apart
1771 * now, so in the !MNT_BYFSID case return the more likely
1772 * EINVAL for compatibility.
1779 * Don't allow unmounting the root filesystem.
1781 if (mp->mnt_flag & MNT_ROOTFS) {
1785 error = dounmount(mp, flags, td);
1790 * Return error if any of the vnodes, ignoring the root vnode
1791 * and the syncer vnode, have non-zero usecount.
1793 * This function is purely advisory - it can return false positives
1797 vfs_check_usecounts(struct mount *mp)
1799 struct vnode *vp, *mvp;
1801 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1802 if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON &&
1803 vp->v_usecount != 0) {
1805 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1815 dounmount_cleanup(struct mount *mp, struct vnode *coveredvp, int mntkflags)
1818 mtx_assert(MNT_MTX(mp), MA_OWNED);
1819 mp->mnt_kern_flag &= ~mntkflags;
1820 if ((mp->mnt_kern_flag & MNTK_MWAIT) != 0) {
1821 mp->mnt_kern_flag &= ~MNTK_MWAIT;
1824 vfs_op_exit_locked(mp);
1826 if (coveredvp != NULL) {
1827 VOP_UNLOCK(coveredvp);
1830 vn_finished_write(mp);
1835 * There are various reference counters associated with the mount point.
1836 * Normally it is permitted to modify them without taking the mnt ilock,
1837 * but this behavior can be temporarily disabled if stable value is needed
1838 * or callers are expected to block (e.g. to not allow new users during
1842 vfs_op_enter(struct mount *mp)
1844 struct mount_pcpu *mpcpu;
1849 if (mp->mnt_vfs_ops > 1) {
1853 vfs_op_barrier_wait(mp);
1855 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1857 mp->mnt_ref += mpcpu->mntp_ref;
1858 mpcpu->mntp_ref = 0;
1860 mp->mnt_lockref += mpcpu->mntp_lockref;
1861 mpcpu->mntp_lockref = 0;
1863 mp->mnt_writeopcount += mpcpu->mntp_writeopcount;
1864 mpcpu->mntp_writeopcount = 0;
1866 MPASSERT(mp->mnt_ref > 0 && mp->mnt_lockref >= 0 &&
1867 mp->mnt_writeopcount >= 0, mp,
1868 ("invalid count(s): ref %d lockref %d writeopcount %d",
1869 mp->mnt_ref, mp->mnt_lockref, mp->mnt_writeopcount));
1871 vfs_assert_mount_counters(mp);
1875 vfs_op_exit_locked(struct mount *mp)
1878 mtx_assert(MNT_MTX(mp), MA_OWNED);
1880 MPASSERT(mp->mnt_vfs_ops > 0, mp,
1881 ("invalid vfs_ops count %d", mp->mnt_vfs_ops));
1882 MPASSERT(mp->mnt_vfs_ops > 1 ||
1883 (mp->mnt_kern_flag & (MNTK_UNMOUNT | MNTK_SUSPEND)) == 0, mp,
1884 ("vfs_ops too low %d in unmount or suspend", mp->mnt_vfs_ops));
1889 vfs_op_exit(struct mount *mp)
1893 vfs_op_exit_locked(mp);
1897 struct vfs_op_barrier_ipi {
1899 struct smp_rendezvous_cpus_retry_arg srcra;
1903 vfs_op_action_func(void *arg)
1905 struct vfs_op_barrier_ipi *vfsopipi;
1908 vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1911 if (!vfs_op_thread_entered(mp))
1912 smp_rendezvous_cpus_done(arg);
1916 vfs_op_wait_func(void *arg, int cpu)
1918 struct vfs_op_barrier_ipi *vfsopipi;
1920 struct mount_pcpu *mpcpu;
1922 vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1925 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1926 while (atomic_load_int(&mpcpu->mntp_thread_in_ops))
1931 vfs_op_barrier_wait(struct mount *mp)
1933 struct vfs_op_barrier_ipi vfsopipi;
1937 smp_rendezvous_cpus_retry(all_cpus,
1938 smp_no_rendezvous_barrier,
1940 smp_no_rendezvous_barrier,
1947 vfs_assert_mount_counters(struct mount *mp)
1949 struct mount_pcpu *mpcpu;
1952 if (mp->mnt_vfs_ops == 0)
1956 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1957 if (mpcpu->mntp_ref != 0 ||
1958 mpcpu->mntp_lockref != 0 ||
1959 mpcpu->mntp_writeopcount != 0)
1960 vfs_dump_mount_counters(mp);
1965 vfs_dump_mount_counters(struct mount *mp)
1967 struct mount_pcpu *mpcpu;
1968 int ref, lockref, writeopcount;
1971 printf("%s: mp %p vfs_ops %d\n", __func__, mp, mp->mnt_vfs_ops);
1976 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1977 printf("%d ", mpcpu->mntp_ref);
1978 ref += mpcpu->mntp_ref;
1981 printf(" lockref : ");
1982 lockref = mp->mnt_lockref;
1984 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1985 printf("%d ", mpcpu->mntp_lockref);
1986 lockref += mpcpu->mntp_lockref;
1989 printf("writeopcount: ");
1990 writeopcount = mp->mnt_writeopcount;
1992 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1993 printf("%d ", mpcpu->mntp_writeopcount);
1994 writeopcount += mpcpu->mntp_writeopcount;
1998 printf("counter struct total\n");
1999 printf("ref %-5d %-5d\n", mp->mnt_ref, ref);
2000 printf("lockref %-5d %-5d\n", mp->mnt_lockref, lockref);
2001 printf("writeopcount %-5d %-5d\n", mp->mnt_writeopcount, writeopcount);
2003 panic("invalid counts on struct mount");
2008 vfs_mount_fetch_counter(struct mount *mp, enum mount_counter which)
2010 struct mount_pcpu *mpcpu;
2017 case MNT_COUNT_LOCKREF:
2018 sum = mp->mnt_lockref;
2020 case MNT_COUNT_WRITEOPCOUNT:
2021 sum = mp->mnt_writeopcount;
2026 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
2029 sum += mpcpu->mntp_ref;
2031 case MNT_COUNT_LOCKREF:
2032 sum += mpcpu->mntp_lockref;
2034 case MNT_COUNT_WRITEOPCOUNT:
2035 sum += mpcpu->mntp_writeopcount;
2043 deferred_unmount_enqueue(struct mount *mp, uint64_t flags, bool requeue,
2049 mtx_lock(&deferred_unmount_lock);
2050 if ((mp->mnt_taskqueue_flags & MNT_DEFERRED) == 0 || requeue) {
2051 mp->mnt_taskqueue_flags = flags | MNT_DEFERRED;
2052 STAILQ_INSERT_TAIL(&deferred_unmount_list, mp,
2053 mnt_taskqueue_link);
2056 mtx_unlock(&deferred_unmount_lock);
2059 taskqueue_enqueue_timeout(taskqueue_deferred_unmount,
2060 &deferred_unmount_task, timeout_ticks);
2067 * Taskqueue handler for processing async/recursive unmounts
2070 vfs_deferred_unmount(void *argi __unused, int pending __unused)
2072 STAILQ_HEAD(, mount) local_unmounts;
2074 struct mount *mp, *tmp;
2076 unsigned int retries;
2079 STAILQ_INIT(&local_unmounts);
2080 mtx_lock(&deferred_unmount_lock);
2081 STAILQ_CONCAT(&local_unmounts, &deferred_unmount_list);
2082 mtx_unlock(&deferred_unmount_lock);
2084 STAILQ_FOREACH_SAFE(mp, &local_unmounts, mnt_taskqueue_link, tmp) {
2085 flags = mp->mnt_taskqueue_flags;
2086 KASSERT((flags & MNT_DEFERRED) != 0,
2087 ("taskqueue unmount without MNT_DEFERRED"));
2088 error = dounmount(mp, flags, curthread);
2091 unmounted = ((mp->mnt_kern_flag & MNTK_REFEXPIRE) != 0);
2095 * The deferred unmount thread is the only thread that
2096 * modifies the retry counts, so locking/atomics aren't
2099 retries = (mp->mnt_unmount_retries)++;
2100 deferred_unmount_total_retries++;
2101 if (!unmounted && retries < deferred_unmount_retry_limit) {
2102 deferred_unmount_enqueue(mp, flags, true,
2103 -deferred_unmount_retry_delay_hz);
2105 if (retries >= deferred_unmount_retry_limit) {
2106 printf("giving up on deferred unmount "
2107 "of %s after %d retries, error %d\n",
2108 mp->mnt_stat.f_mntonname, retries, error);
2117 * Do the actual filesystem unmount.
2120 dounmount(struct mount *mp, uint64_t flags, struct thread *td)
2122 struct mount_upper_node *upper;
2123 struct vnode *coveredvp, *rootvp;
2125 uint64_t async_flag;
2127 unsigned int retries;
2129 KASSERT((flags & MNT_DEFERRED) == 0 ||
2130 (flags & (MNT_RECURSE | MNT_FORCE)) == (MNT_RECURSE | MNT_FORCE),
2131 ("MNT_DEFERRED requires MNT_RECURSE | MNT_FORCE"));
2134 * If the caller has explicitly requested the unmount to be handled by
2135 * the taskqueue and we're not already in taskqueue context, queue
2136 * up the unmount request and exit. This is done prior to any
2137 * credential checks; MNT_DEFERRED should be used only for kernel-
2138 * initiated unmounts and will therefore be processed with the
2139 * (kernel) credentials of the taskqueue thread. Still, callers
2140 * should be sure this is the behavior they want.
2142 if ((flags & MNT_DEFERRED) != 0 &&
2143 taskqueue_member(taskqueue_deferred_unmount, curthread) == 0) {
2144 if (!deferred_unmount_enqueue(mp, flags, false, 0))
2146 return (EINPROGRESS);
2150 * Only privileged root, or (if MNT_USER is set) the user that did the
2151 * original mount is permitted to unmount this filesystem.
2152 * This check should be made prior to queueing up any recursive
2153 * unmounts of upper filesystems. Those unmounts will be executed
2154 * with kernel thread credentials and are expected to succeed, so
2155 * we must at least ensure the originating context has sufficient
2156 * privilege to unmount the base filesystem before proceeding with
2159 error = vfs_suser(mp, td);
2161 KASSERT((flags & MNT_DEFERRED) == 0,
2162 ("taskqueue unmount with insufficient privilege"));
2167 if (recursive_forced_unmount && ((flags & MNT_FORCE) != 0))
2168 flags |= MNT_RECURSE;
2170 if ((flags & MNT_RECURSE) != 0) {
2171 KASSERT((flags & MNT_FORCE) != 0,
2172 ("MNT_RECURSE requires MNT_FORCE"));
2176 * Set MNTK_RECURSE to prevent new upper mounts from being
2177 * added, and note that an operation on the uppers list is in
2178 * progress. This will ensure that unregistration from the
2179 * uppers list, and therefore any pending unmount of the upper
2180 * FS, can't complete until after we finish walking the list.
2182 mp->mnt_kern_flag |= MNTK_RECURSE;
2183 mp->mnt_upper_pending++;
2184 TAILQ_FOREACH(upper, &mp->mnt_uppers, mnt_upper_link) {
2185 retries = upper->mp->mnt_unmount_retries;
2186 if (retries > deferred_unmount_retry_limit) {
2193 if (!deferred_unmount_enqueue(upper->mp, flags,
2198 mp->mnt_upper_pending--;
2199 if ((mp->mnt_kern_flag & MNTK_UPPER_WAITER) != 0 &&
2200 mp->mnt_upper_pending == 0) {
2201 mp->mnt_kern_flag &= ~MNTK_UPPER_WAITER;
2202 wakeup(&mp->mnt_uppers);
2206 * If we're not on the taskqueue, wait until the uppers list
2207 * is drained before proceeding with unmount. Otherwise, if
2208 * we are on the taskqueue and there are still pending uppers,
2209 * just re-enqueue on the end of the taskqueue.
2211 if ((flags & MNT_DEFERRED) == 0) {
2212 while (error == 0 && !TAILQ_EMPTY(&mp->mnt_uppers)) {
2213 mp->mnt_kern_flag |= MNTK_TASKQUEUE_WAITER;
2214 error = msleep(&mp->mnt_taskqueue_link,
2215 MNT_MTX(mp), PCATCH, "umntqw", 0);
2222 } else if (!TAILQ_EMPTY(&mp->mnt_uppers)) {
2225 deferred_unmount_enqueue(mp, flags, true, 0);
2229 KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers not empty"));
2232 /* Allow the taskqueue to safely re-enqueue on failure */
2233 if ((flags & MNT_DEFERRED) != 0)
2236 if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
2237 mnt_gen_r = mp->mnt_gen;
2240 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
2242 * Check for mp being unmounted while waiting for the
2243 * covered vnode lock.
2245 if (coveredvp->v_mountedhere != mp ||
2246 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
2247 VOP_UNLOCK(coveredvp);
2256 vn_start_write(NULL, &mp, V_WAIT);
2258 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 ||
2259 (mp->mnt_flag & MNT_UPDATE) != 0 ||
2260 !TAILQ_EMPTY(&mp->mnt_uppers)) {
2261 dounmount_cleanup(mp, coveredvp, 0);
2264 mp->mnt_kern_flag |= MNTK_UNMOUNT;
2265 rootvp = vfs_cache_root_clear(mp);
2266 if (coveredvp != NULL)
2267 vn_seqc_write_begin(coveredvp);
2268 if (flags & MNT_NONBUSY) {
2270 error = vfs_check_usecounts(mp);
2273 vn_seqc_write_end(coveredvp);
2274 dounmount_cleanup(mp, coveredvp, MNTK_UNMOUNT);
2275 if (rootvp != NULL) {
2276 vn_seqc_write_end(rootvp);
2282 /* Allow filesystems to detect that a forced unmount is in progress. */
2283 if (flags & MNT_FORCE) {
2284 mp->mnt_kern_flag |= MNTK_UNMOUNTF;
2287 * Must be done after setting MNTK_UNMOUNTF and before
2288 * waiting for mnt_lockref to become 0.
2294 if (mp->mnt_lockref) {
2295 mp->mnt_kern_flag |= MNTK_DRAINING;
2296 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS,
2300 KASSERT(mp->mnt_lockref == 0,
2301 ("%s: invalid lock refcount in the drain path @ %s:%d",
2302 __func__, __FILE__, __LINE__));
2304 ("%s: invalid return value for msleep in the drain path @ %s:%d",
2305 __func__, __FILE__, __LINE__));
2308 * We want to keep the vnode around so that we can vn_seqc_write_end
2309 * after we are done with unmount. Downgrade our reference to a mere
2310 * hold count so that we don't interefere with anything.
2312 if (rootvp != NULL) {
2317 if (mp->mnt_flag & MNT_EXPUBLIC)
2318 vfs_setpublicfs(NULL, NULL, NULL);
2320 vfs_periodic(mp, MNT_WAIT);
2322 async_flag = mp->mnt_flag & MNT_ASYNC;
2323 mp->mnt_flag &= ~MNT_ASYNC;
2324 mp->mnt_kern_flag &= ~MNTK_ASYNC;
2326 vfs_deallocate_syncvnode(mp);
2327 error = VFS_UNMOUNT(mp, flags);
2328 vn_finished_write(mp);
2331 * If we failed to flush the dirty blocks for this mount point,
2332 * undo all the cdir/rdir and rootvnode changes we made above.
2333 * Unless we failed to do so because the device is reporting that
2334 * it doesn't exist anymore.
2336 if (error && error != ENXIO) {
2338 if ((mp->mnt_flag & MNT_RDONLY) == 0) {
2340 vfs_allocate_syncvnode(mp);
2343 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
2344 mp->mnt_flag |= async_flag;
2345 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
2346 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
2347 mp->mnt_kern_flag |= MNTK_ASYNC;
2348 if (mp->mnt_kern_flag & MNTK_MWAIT) {
2349 mp->mnt_kern_flag &= ~MNTK_MWAIT;
2352 vfs_op_exit_locked(mp);
2355 vn_seqc_write_end(coveredvp);
2356 VOP_UNLOCK(coveredvp);
2359 if (rootvp != NULL) {
2360 vn_seqc_write_end(rootvp);
2366 mtx_lock(&mountlist_mtx);
2367 TAILQ_REMOVE(&mountlist, mp, mnt_list);
2368 mtx_unlock(&mountlist_mtx);
2369 EVENTHANDLER_DIRECT_INVOKE(vfs_unmounted, mp, td);
2370 if (coveredvp != NULL) {
2372 vn_irflag_unset_locked(coveredvp, VIRF_MOUNTPOINT);
2373 coveredvp->v_mountedhere = NULL;
2374 vn_seqc_write_end_locked(coveredvp);
2375 VI_UNLOCK(coveredvp);
2376 VOP_UNLOCK(coveredvp);
2379 mount_devctl_event("UNMOUNT", mp, false);
2380 if (rootvp != NULL) {
2381 vn_seqc_write_end(rootvp);
2384 vfs_event_signal(NULL, VQ_UNMOUNT, 0);
2385 if (rootvnode != NULL && mp == rootvnode->v_mount) {
2389 if (mp == rootdevmp)
2391 if ((flags & MNT_DEFERRED) != 0)
2393 vfs_mount_destroy(mp);
2398 * Report errors during filesystem mounting.
2401 vfs_mount_error(struct mount *mp, const char *fmt, ...)
2403 struct vfsoptlist *moptlist = mp->mnt_optnew;
2408 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
2409 if (error || errmsg == NULL || len <= 0)
2413 vsnprintf(errmsg, (size_t)len, fmt, ap);
2418 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...)
2424 error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len);
2425 if (error || errmsg == NULL || len <= 0)
2429 vsnprintf(errmsg, (size_t)len, fmt, ap);
2434 * ---------------------------------------------------------------------
2435 * Functions for querying mount options/arguments from filesystems.
2439 * Check that no unknown options are given
2442 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
2446 const char **t, *p, *q;
2449 TAILQ_FOREACH(opt, opts, link) {
2452 if (p[0] == 'n' && p[1] == 'o')
2454 for(t = global_opts; *t != NULL; t++) {
2455 if (strcmp(*t, p) == 0)
2458 if (strcmp(*t, q) == 0)
2464 for(t = legal; *t != NULL; t++) {
2465 if (strcmp(*t, p) == 0)
2468 if (strcmp(*t, q) == 0)
2474 snprintf(errmsg, sizeof(errmsg),
2475 "mount option <%s> is unknown", p);
2479 TAILQ_FOREACH(opt, opts, link) {
2480 if (strcmp(opt->name, "errmsg") == 0) {
2481 strncpy((char *)opt->value, errmsg, opt->len);
2486 printf("%s\n", errmsg);
2492 * Get a mount option by its name.
2494 * Return 0 if the option was found, ENOENT otherwise.
2495 * If len is non-NULL it will be filled with the length
2496 * of the option. If buf is non-NULL, it will be filled
2497 * with the address of the option.
2500 vfs_getopt(struct vfsoptlist *opts, const char *name, void **buf, int *len)
2504 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2506 TAILQ_FOREACH(opt, opts, link) {
2507 if (strcmp(name, opt->name) == 0) {
2520 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
2527 TAILQ_FOREACH(opt, opts, link) {
2528 if (strcmp(name, opt->name) == 0) {
2537 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value)
2539 char *opt_value, *vtp;
2543 error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len);
2546 if (opt_len == 0 || opt_value == NULL)
2548 if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0')
2550 iv = strtoq(opt_value, &vtp, 0);
2551 if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0'))
2578 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
2583 TAILQ_FOREACH(opt, opts, link) {
2584 if (strcmp(name, opt->name) != 0)
2587 if (opt->len == 0 ||
2588 ((char *)opt->value)[opt->len - 1] != '\0') {
2592 return (opt->value);
2599 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w,
2604 TAILQ_FOREACH(opt, opts, link) {
2605 if (strcmp(name, opt->name) == 0) {
2618 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
2624 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2626 TAILQ_FOREACH(opt, opts, link) {
2627 if (strcmp(name, opt->name) != 0)
2630 if (opt->len == 0 || opt->value == NULL)
2632 if (((char *)opt->value)[opt->len - 1] != '\0')
2635 ret = vsscanf(opt->value, fmt, ap);
2643 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len)
2647 TAILQ_FOREACH(opt, opts, link) {
2648 if (strcmp(name, opt->name) != 0)
2651 if (opt->value == NULL)
2654 if (opt->len != len)
2656 bcopy(value, opt->value, len);
2664 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len)
2668 TAILQ_FOREACH(opt, opts, link) {
2669 if (strcmp(name, opt->name) != 0)
2672 if (opt->value == NULL)
2678 bcopy(value, opt->value, len);
2686 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value)
2690 TAILQ_FOREACH(opt, opts, link) {
2691 if (strcmp(name, opt->name) != 0)
2694 if (opt->value == NULL)
2695 opt->len = strlen(value) + 1;
2696 else if (strlcpy(opt->value, value, opt->len) >= opt->len)
2704 * Find and copy a mount option.
2706 * The size of the buffer has to be specified
2707 * in len, if it is not the same length as the
2708 * mount option, EINVAL is returned.
2709 * Returns ENOENT if the option is not found.
2712 vfs_copyopt(struct vfsoptlist *opts, const char *name, void *dest, int len)
2716 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
2718 TAILQ_FOREACH(opt, opts, link) {
2719 if (strcmp(name, opt->name) == 0) {
2721 if (len != opt->len)
2723 bcopy(opt->value, dest, opt->len);
2731 __vfs_statfs(struct mount *mp, struct statfs *sbp)
2734 * Filesystems only fill in part of the structure for updates, we
2735 * have to read the entirety first to get all content.
2737 if (sbp != &mp->mnt_stat)
2738 memcpy(sbp, &mp->mnt_stat, sizeof(*sbp));
2741 * Set these in case the underlying filesystem fails to do so.
2743 sbp->f_version = STATFS_VERSION;
2744 sbp->f_namemax = NAME_MAX;
2745 sbp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
2746 sbp->f_nvnodelistsize = mp->mnt_nvnodelistsize;
2748 return (mp->mnt_op->vfs_statfs(mp, sbp));
2752 vfs_mountedfrom(struct mount *mp, const char *from)
2755 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
2756 strlcpy(mp->mnt_stat.f_mntfromname, from,
2757 sizeof mp->mnt_stat.f_mntfromname);
2761 * ---------------------------------------------------------------------
2762 * This is the api for building mount args and mounting filesystems from
2763 * inside the kernel.
2765 * The API works by accumulation of individual args. First error is
2768 * XXX: should be documented in new manpage kernel_mount(9)
2771 /* A memory allocation which must be freed when we are done */
2773 SLIST_ENTRY(mntaarg) next;
2776 /* The header for the mount arguments */
2781 SLIST_HEAD(, mntaarg) list;
2785 * Add a boolean argument.
2787 * flag is the boolean value.
2788 * name must start with "no".
2791 mount_argb(struct mntarg *ma, int flag, const char *name)
2794 KASSERT(name[0] == 'n' && name[1] == 'o',
2795 ("mount_argb(...,%s): name must start with 'no'", name));
2797 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
2801 * Add an argument printf style
2804 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
2807 struct mntaarg *maa;
2812 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2813 SLIST_INIT(&ma->list);
2818 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2820 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2821 ma->v[ma->len].iov_len = strlen(name) + 1;
2824 sb = sbuf_new_auto();
2826 sbuf_vprintf(sb, fmt, ap);
2829 len = sbuf_len(sb) + 1;
2830 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2831 SLIST_INSERT_HEAD(&ma->list, maa, next);
2832 bcopy(sbuf_data(sb), maa + 1, len);
2835 ma->v[ma->len].iov_base = maa + 1;
2836 ma->v[ma->len].iov_len = len;
2843 * Add an argument which is a userland string.
2846 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2848 struct mntaarg *maa;
2854 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2855 SLIST_INIT(&ma->list);
2859 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2860 SLIST_INSERT_HEAD(&ma->list, maa, next);
2861 tbuf = (void *)(maa + 1);
2862 ma->error = copyinstr(val, tbuf, len, NULL);
2863 return (mount_arg(ma, name, tbuf, -1));
2869 * If length is -1, treat value as a C string.
2872 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2876 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2877 SLIST_INIT(&ma->list);
2882 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2884 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2885 ma->v[ma->len].iov_len = strlen(name) + 1;
2888 ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2890 ma->v[ma->len].iov_len = strlen(val) + 1;
2892 ma->v[ma->len].iov_len = len;
2898 * Free a mntarg structure
2901 free_mntarg(struct mntarg *ma)
2903 struct mntaarg *maa;
2905 while (!SLIST_EMPTY(&ma->list)) {
2906 maa = SLIST_FIRST(&ma->list);
2907 SLIST_REMOVE_HEAD(&ma->list, next);
2910 free(ma->v, M_MOUNT);
2915 * Mount a filesystem
2918 kernel_mount(struct mntarg *ma, uint64_t flags)
2923 KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2924 KASSERT(ma->error != 0 || ma->v != NULL, ("kernel_mount NULL ma->v"));
2925 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2929 auio.uio_iov = ma->v;
2930 auio.uio_iovcnt = ma->len;
2931 auio.uio_segflg = UIO_SYSSPACE;
2932 error = vfs_donmount(curthread, flags, &auio);
2938 /* Map from mount options to printable formats. */
2939 static struct mntoptnames optnames[] = {
2943 #define DEVCTL_LEN 1024
2945 mount_devctl_event(const char *type, struct mount *mp, bool donew)
2948 struct mntoptnames *fp;
2950 struct statfs *sfp = &mp->mnt_stat;
2953 buf = malloc(DEVCTL_LEN, M_MOUNT, M_NOWAIT);
2956 sbuf_new(&sb, buf, DEVCTL_LEN, SBUF_FIXEDLEN);
2957 sbuf_cpy(&sb, "mount-point=\"");
2958 devctl_safe_quote_sb(&sb, sfp->f_mntonname);
2959 sbuf_cat(&sb, "\" mount-dev=\"");
2960 devctl_safe_quote_sb(&sb, sfp->f_mntfromname);
2961 sbuf_cat(&sb, "\" mount-type=\"");
2962 devctl_safe_quote_sb(&sb, sfp->f_fstypename);
2963 sbuf_cat(&sb, "\" fsid=0x");
2964 cp = (const uint8_t *)&sfp->f_fsid.val[0];
2965 for (int i = 0; i < sizeof(sfp->f_fsid); i++)
2966 sbuf_printf(&sb, "%02x", cp[i]);
2967 sbuf_printf(&sb, " owner=%u flags=\"", sfp->f_owner);
2968 for (fp = optnames; fp->o_opt != 0; fp++) {
2969 if ((mp->mnt_flag & fp->o_opt) != 0) {
2970 sbuf_cat(&sb, fp->o_name);
2971 sbuf_putc(&sb, ';');
2974 sbuf_putc(&sb, '"');
2978 * Options are not published because the form of the options depends on
2979 * the file system and may include binary data. In addition, they don't
2980 * necessarily provide enough useful information to be actionable when
2981 * devd processes them.
2984 if (sbuf_error(&sb) == 0)
2985 devctl_notify("VFS", "FS", type, sbuf_data(&sb));
2991 * Force remount specified mount point to read-only. The argument
2992 * must be busied to avoid parallel unmount attempts.
2994 * Intended use is to prevent further writes if some metadata
2995 * inconsistency is detected. Note that the function still flushes
2996 * all cached metadata and data for the mount point, which might be
2997 * not always suitable.
3000 vfs_remount_ro(struct mount *mp)
3002 struct vfsoptlist *opts;
3004 struct vnode *vp_covered, *rootvp;
3007 KASSERT(mp->mnt_lockref > 0,
3008 ("vfs_remount_ro: mp %p is not busied", mp));
3009 KASSERT((mp->mnt_kern_flag & MNTK_UNMOUNT) == 0,
3010 ("vfs_remount_ro: mp %p is being unmounted (and busy?)", mp));
3013 vp_covered = mp->mnt_vnodecovered;
3014 error = vget(vp_covered, LK_EXCLUSIVE | LK_NOWAIT);
3017 VI_LOCK(vp_covered);
3018 if ((vp_covered->v_iflag & VI_MOUNT) != 0) {
3019 VI_UNLOCK(vp_covered);
3023 vp_covered->v_iflag |= VI_MOUNT;
3024 VI_UNLOCK(vp_covered);
3026 vn_seqc_write_begin(vp_covered);
3029 if ((mp->mnt_flag & MNT_RDONLY) != 0) {
3034 mp->mnt_flag |= MNT_UPDATE | MNT_FORCE | MNT_RDONLY;
3035 rootvp = vfs_cache_root_clear(mp);
3038 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK | M_ZERO);
3040 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK | M_ZERO);
3041 opt->name = strdup("ro", M_MOUNT);
3043 TAILQ_INSERT_TAIL(opts, opt, link);
3044 vfs_mergeopts(opts, mp->mnt_opt);
3045 mp->mnt_optnew = opts;
3047 error = VFS_MOUNT(mp);
3051 mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE);
3053 vfs_deallocate_syncvnode(mp);
3054 if (mp->mnt_opt != NULL)
3055 vfs_freeopts(mp->mnt_opt);
3056 mp->mnt_opt = mp->mnt_optnew;
3059 mp->mnt_flag &= ~(MNT_UPDATE | MNT_FORCE | MNT_RDONLY);
3061 vfs_freeopts(mp->mnt_optnew);
3063 mp->mnt_optnew = NULL;
3067 VI_LOCK(vp_covered);
3068 vp_covered->v_iflag &= ~VI_MOUNT;
3069 VI_UNLOCK(vp_covered);
3071 vn_seqc_write_end(vp_covered);
3072 if (rootvp != NULL) {
3073 vn_seqc_write_end(rootvp);
3080 * Suspend write operations on all local writeable filesystems. Does
3081 * full sync of them in the process.
3083 * Iterate over the mount points in reverse order, suspending most
3084 * recently mounted filesystems first. It handles a case where a
3085 * filesystem mounted from a md(4) vnode-backed device should be
3086 * suspended before the filesystem that owns the vnode.
3089 suspend_all_fs(void)
3094 mtx_lock(&mountlist_mtx);
3095 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
3096 error = vfs_busy(mp, MBF_MNTLSTLOCK | MBF_NOWAIT);
3099 if ((mp->mnt_flag & (MNT_RDONLY | MNT_LOCAL)) != MNT_LOCAL ||
3100 (mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
3101 mtx_lock(&mountlist_mtx);
3105 error = vfs_write_suspend(mp, 0);
3108 MPASS((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0);
3109 mp->mnt_kern_flag |= MNTK_SUSPEND_ALL;
3111 mtx_lock(&mountlist_mtx);
3113 printf("suspend of %s failed, error %d\n",
3114 mp->mnt_stat.f_mntonname, error);
3115 mtx_lock(&mountlist_mtx);
3119 mtx_unlock(&mountlist_mtx);
3127 mtx_lock(&mountlist_mtx);
3128 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3129 if ((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0)
3131 mtx_unlock(&mountlist_mtx);
3133 MPASS((mp->mnt_kern_flag & MNTK_SUSPEND) != 0);
3134 mp->mnt_kern_flag &= ~MNTK_SUSPEND_ALL;
3136 vfs_write_resume(mp, 0);
3137 mtx_lock(&mountlist_mtx);
3140 mtx_unlock(&mountlist_mtx);