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/limits.h>
53 #include <sys/malloc.h>
54 #include <sys/mount.h>
55 #include <sys/mutex.h>
56 #include <sys/namei.h>
59 #include <sys/filedesc.h>
60 #include <sys/reboot.h>
62 #include <sys/syscallsubr.h>
63 #include <sys/sysproto.h>
65 #include <sys/sysctl.h>
66 #include <sys/sysent.h>
67 #include <sys/systm.h>
68 #include <sys/taskqueue.h>
69 #include <sys/vnode.h>
72 #include <geom/geom.h>
74 #include <machine/stdarg.h>
76 #include <security/audit/audit.h>
77 #include <security/mac/mac_framework.h>
79 #define VFS_MOUNTARG_SIZE_MAX (1024 * 64)
81 static int vfs_domount(struct thread *td, const char *fstype, char *fspath,
82 uint64_t fsflags, struct vfsoptlist **optlist);
83 static void free_mntarg(struct mntarg *ma);
85 static int usermount = 0;
86 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
87 "Unprivileged users may mount and unmount file systems");
89 static bool default_autoro = false;
90 SYSCTL_BOOL(_vfs, OID_AUTO, default_autoro, CTLFLAG_RW, &default_autoro, 0,
91 "Retry failed r/w mount as r/o if no explicit ro/rw option is specified");
93 static bool recursive_forced_unmount = false;
94 SYSCTL_BOOL(_vfs, OID_AUTO, recursive_forced_unmount, CTLFLAG_RW,
95 &recursive_forced_unmount, 0, "Recursively unmount stacked upper mounts"
96 " when a file system is forcibly unmounted");
98 static SYSCTL_NODE(_vfs, OID_AUTO, deferred_unmount,
99 CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "deferred unmount controls");
101 static unsigned int deferred_unmount_retry_limit = 10;
102 SYSCTL_UINT(_vfs_deferred_unmount, OID_AUTO, retry_limit, CTLFLAG_RW,
103 &deferred_unmount_retry_limit, 0,
104 "Maximum number of retries for deferred unmount failure");
106 static int deferred_unmount_retry_delay_hz;
107 SYSCTL_INT(_vfs_deferred_unmount, OID_AUTO, retry_delay_hz, CTLFLAG_RW,
108 &deferred_unmount_retry_delay_hz, 0,
109 "Delay in units of [1/kern.hz]s when retrying a failed deferred unmount");
111 static int deferred_unmount_total_retries = 0;
112 SYSCTL_INT(_vfs_deferred_unmount, OID_AUTO, total_retries, CTLFLAG_RD,
113 &deferred_unmount_total_retries, 0,
114 "Total number of retried deferred unmounts");
116 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
117 MALLOC_DEFINE(M_STATFS, "statfs", "statfs structure");
118 static uma_zone_t mount_zone;
120 /* List of mounted filesystems. */
121 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
123 /* For any iteration/modification of mountlist */
124 struct mtx_padalign __exclusive_cache_line mountlist_mtx;
126 EVENTHANDLER_LIST_DEFINE(vfs_mounted);
127 EVENTHANDLER_LIST_DEFINE(vfs_unmounted);
129 static void vfs_deferred_unmount(void *arg, int pending);
130 static struct timeout_task deferred_unmount_task;
131 static struct mtx deferred_unmount_lock;
132 MTX_SYSINIT(deferred_unmount, &deferred_unmount_lock, "deferred_unmount",
134 static STAILQ_HEAD(, mount) deferred_unmount_list =
135 STAILQ_HEAD_INITIALIZER(deferred_unmount_list);
136 TASKQUEUE_DEFINE_THREAD(deferred_unmount);
138 static void mount_devctl_event(const char *type, struct mount *mp, bool donew);
141 * Global opts, taken by all filesystems
143 static const char *global_opts[] = {
155 mount_init(void *mem, int size, int flags)
159 mp = (struct mount *)mem;
160 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
161 mtx_init(&mp->mnt_listmtx, "struct mount vlist mtx", NULL, MTX_DEF);
162 lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0);
163 mp->mnt_pcpu = uma_zalloc_pcpu(pcpu_zone_16, M_WAITOK | M_ZERO);
166 mp->mnt_rootvnode = NULL;
171 mount_fini(void *mem, int size)
175 mp = (struct mount *)mem;
176 uma_zfree_pcpu(pcpu_zone_16, mp->mnt_pcpu);
177 lockdestroy(&mp->mnt_explock);
178 mtx_destroy(&mp->mnt_listmtx);
179 mtx_destroy(&mp->mnt_mtx);
183 vfs_mount_init(void *dummy __unused)
185 TIMEOUT_TASK_INIT(taskqueue_deferred_unmount, &deferred_unmount_task,
186 0, vfs_deferred_unmount, NULL);
187 deferred_unmount_retry_delay_hz = hz;
188 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), NULL,
189 NULL, mount_init, mount_fini, UMA_ALIGN_CACHE, UMA_ZONE_NOFREE);
190 mtx_init(&mountlist_mtx, "mountlist", NULL, MTX_DEF);
192 SYSINIT(vfs_mount, SI_SUB_VFS, SI_ORDER_ANY, vfs_mount_init, NULL);
195 * ---------------------------------------------------------------------
196 * Functions for building and sanitizing the mount options
199 /* Remove one mount option. */
201 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
204 TAILQ_REMOVE(opts, opt, link);
205 free(opt->name, M_MOUNT);
206 if (opt->value != NULL)
207 free(opt->value, M_MOUNT);
211 /* Release all resources related to the mount options. */
213 vfs_freeopts(struct vfsoptlist *opts)
217 while (!TAILQ_EMPTY(opts)) {
218 opt = TAILQ_FIRST(opts);
219 vfs_freeopt(opts, opt);
225 vfs_deleteopt(struct vfsoptlist *opts, const char *name)
227 struct vfsopt *opt, *temp;
231 TAILQ_FOREACH_SAFE(opt, opts, link, temp) {
232 if (strcmp(opt->name, name) == 0)
233 vfs_freeopt(opts, opt);
238 vfs_isopt_ro(const char *opt)
241 if (strcmp(opt, "ro") == 0 || strcmp(opt, "rdonly") == 0 ||
242 strcmp(opt, "norw") == 0)
248 vfs_isopt_rw(const char *opt)
251 if (strcmp(opt, "rw") == 0 || strcmp(opt, "noro") == 0)
257 * Check if options are equal (with or without the "no" prefix).
260 vfs_equalopts(const char *opt1, const char *opt2)
264 /* "opt" vs. "opt" or "noopt" vs. "noopt" */
265 if (strcmp(opt1, opt2) == 0)
267 /* "noopt" vs. "opt" */
268 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
270 /* "opt" vs. "noopt" */
271 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
273 while ((p = strchr(opt1, '.')) != NULL &&
274 !strncmp(opt1, opt2, ++p - opt1)) {
277 /* "foo.noopt" vs. "foo.opt" */
278 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
280 /* "foo.opt" vs. "foo.noopt" */
281 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
284 /* "ro" / "rdonly" / "norw" / "rw" / "noro" */
285 if ((vfs_isopt_ro(opt1) || vfs_isopt_rw(opt1)) &&
286 (vfs_isopt_ro(opt2) || vfs_isopt_rw(opt2)))
292 * If a mount option is specified several times,
293 * (with or without the "no" prefix) only keep
294 * the last occurrence of it.
297 vfs_sanitizeopts(struct vfsoptlist *opts)
299 struct vfsopt *opt, *opt2, *tmp;
301 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
302 opt2 = TAILQ_PREV(opt, vfsoptlist, link);
303 while (opt2 != NULL) {
304 if (vfs_equalopts(opt->name, opt2->name)) {
305 tmp = TAILQ_PREV(opt2, vfsoptlist, link);
306 vfs_freeopt(opts, opt2);
309 opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
316 * Build a linked list of mount options from a struct uio.
319 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
321 struct vfsoptlist *opts;
323 size_t memused, namelen, optlen;
324 unsigned int i, iovcnt;
327 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
330 iovcnt = auio->uio_iovcnt;
331 for (i = 0; i < iovcnt; i += 2) {
332 namelen = auio->uio_iov[i].iov_len;
333 optlen = auio->uio_iov[i + 1].iov_len;
334 memused += sizeof(struct vfsopt) + optlen + namelen;
336 * Avoid consuming too much memory, and attempts to overflow
339 if (memused > VFS_MOUNTARG_SIZE_MAX ||
340 optlen > VFS_MOUNTARG_SIZE_MAX ||
341 namelen > VFS_MOUNTARG_SIZE_MAX) {
346 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
347 opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
354 * Do this early, so jumps to "bad" will free the current
357 TAILQ_INSERT_TAIL(opts, opt, link);
359 if (auio->uio_segflg == UIO_SYSSPACE) {
360 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
362 error = copyin(auio->uio_iov[i].iov_base, opt->name,
367 /* Ensure names are null-terminated strings. */
368 if (namelen == 0 || opt->name[namelen - 1] != '\0') {
374 opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
375 if (auio->uio_segflg == UIO_SYSSPACE) {
376 bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
379 error = copyin(auio->uio_iov[i + 1].iov_base,
386 vfs_sanitizeopts(opts);
395 * Merge the old mount options with the new ones passed
396 * in the MNT_UPDATE case.
398 * XXX: This function will keep a "nofoo" option in the new
399 * options. E.g, if the option's canonical name is "foo",
400 * "nofoo" ends up in the mount point's active options.
403 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *oldopts)
405 struct vfsopt *opt, *new;
407 TAILQ_FOREACH(opt, oldopts, link) {
408 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
409 new->name = strdup(opt->name, M_MOUNT);
411 new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
412 bcopy(opt->value, new->value, opt->len);
416 new->seen = opt->seen;
417 TAILQ_INSERT_HEAD(toopts, new, link);
419 vfs_sanitizeopts(toopts);
423 * Mount a filesystem.
425 #ifndef _SYS_SYSPROTO_H_
433 sys_nmount(struct thread *td, struct nmount_args *uap)
441 * Mount flags are now 64-bits. On 32-bit archtectures only
442 * 32-bits are passed in, but from here on everything handles
443 * 64-bit flags correctly.
447 AUDIT_ARG_FFLAGS(flags);
448 CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__,
449 uap->iovp, uap->iovcnt, flags);
452 * Filter out MNT_ROOTFS. We do not want clients of nmount() in
453 * userspace to set this flag, but we must filter it out if we want
454 * MNT_UPDATE on the root file system to work.
455 * MNT_ROOTFS should only be set by the kernel when mounting its
458 flags &= ~MNT_ROOTFS;
460 iovcnt = uap->iovcnt;
462 * Check that we have an even number of iovec's
463 * and that we have at least two options.
465 if ((iovcnt & 1) || (iovcnt < 4)) {
466 CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__,
471 error = copyinuio(uap->iovp, iovcnt, &auio);
473 CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno",
477 error = vfs_donmount(td, flags, auio);
484 * ---------------------------------------------------------------------
485 * Various utility functions
489 * Get a reference on a mount point from a vnode.
491 * The vnode is allowed to be passed unlocked and race against dooming. Note in
492 * such case there are no guarantees the referenced mount point will still be
493 * associated with it after the function returns.
496 vfs_ref_from_vp(struct vnode *vp)
499 struct mount_pcpu *mpcpu;
501 mp = atomic_load_ptr(&vp->v_mount);
502 if (__predict_false(mp == NULL)) {
505 if (vfs_op_thread_enter(mp, mpcpu)) {
506 if (__predict_true(mp == vp->v_mount)) {
507 vfs_mp_count_add_pcpu(mpcpu, ref, 1);
508 vfs_op_thread_exit(mp, mpcpu);
510 vfs_op_thread_exit(mp, mpcpu);
515 if (mp == vp->v_mount) {
527 vfs_ref(struct mount *mp)
529 struct mount_pcpu *mpcpu;
531 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
532 if (vfs_op_thread_enter(mp, mpcpu)) {
533 vfs_mp_count_add_pcpu(mpcpu, ref, 1);
534 vfs_op_thread_exit(mp, mpcpu);
544 * Register ump as an upper mount of the mount associated with
545 * vnode vp. This registration will be tracked through
546 * mount_upper_node upper, which should be allocated by the
547 * caller and stored in per-mount data associated with mp.
549 * If successful, this function will return the mount associated
550 * with vp, and will ensure that it cannot be unmounted until
551 * ump has been unregistered as one of its upper mounts.
553 * Upon failure this function will return NULL.
556 vfs_register_upper_from_vp(struct vnode *vp, struct mount *ump,
557 struct mount_upper_node *upper)
561 mp = atomic_load_ptr(&vp->v_mount);
565 if (mp != vp->v_mount ||
566 ((mp->mnt_kern_flag & (MNTK_UNMOUNT | MNTK_RECURSE)) != 0)) {
570 KASSERT(ump != mp, ("upper and lower mounts are identical"));
573 TAILQ_INSERT_TAIL(&mp->mnt_uppers, upper, mnt_upper_link);
579 * Register upper mount ump to receive vnode unlink/reclaim
580 * notifications from lower mount mp. This registration will
581 * be tracked through mount_upper_node upper, which should be
582 * allocated by the caller and stored in per-mount data
583 * associated with mp.
585 * ump must already be registered as an upper mount of mp
586 * through a call to vfs_register_upper_from_vp().
589 vfs_register_for_notification(struct mount *mp, struct mount *ump,
590 struct mount_upper_node *upper)
594 TAILQ_INSERT_TAIL(&mp->mnt_notify, upper, mnt_upper_link);
599 vfs_drain_upper_locked(struct mount *mp)
601 mtx_assert(MNT_MTX(mp), MA_OWNED);
602 while (mp->mnt_upper_pending != 0) {
603 mp->mnt_kern_flag |= MNTK_UPPER_WAITER;
604 msleep(&mp->mnt_uppers, MNT_MTX(mp), 0, "mntupw", 0);
609 * Undo a previous call to vfs_register_for_notification().
610 * The mount represented by upper must be currently registered
611 * as an upper mount for mp.
614 vfs_unregister_for_notification(struct mount *mp,
615 struct mount_upper_node *upper)
618 vfs_drain_upper_locked(mp);
619 TAILQ_REMOVE(&mp->mnt_notify, upper, mnt_upper_link);
624 * Undo a previous call to vfs_register_upper_from_vp().
625 * This must be done before mp can be unmounted.
628 vfs_unregister_upper(struct mount *mp, struct mount_upper_node *upper)
631 KASSERT((mp->mnt_kern_flag & MNTK_UNMOUNT) == 0,
632 ("registered upper with pending unmount"));
633 vfs_drain_upper_locked(mp);
634 TAILQ_REMOVE(&mp->mnt_uppers, upper, mnt_upper_link);
635 if ((mp->mnt_kern_flag & MNTK_TASKQUEUE_WAITER) != 0 &&
636 TAILQ_EMPTY(&mp->mnt_uppers)) {
637 mp->mnt_kern_flag &= ~MNTK_TASKQUEUE_WAITER;
638 wakeup(&mp->mnt_taskqueue_link);
645 vfs_rel(struct mount *mp)
647 struct mount_pcpu *mpcpu;
649 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
650 if (vfs_op_thread_enter(mp, mpcpu)) {
651 vfs_mp_count_sub_pcpu(mpcpu, ref, 1);
652 vfs_op_thread_exit(mp, mpcpu);
662 * Allocate and initialize the mount point struct.
665 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath,
670 mp = uma_zalloc(mount_zone, M_WAITOK);
671 bzero(&mp->mnt_startzero,
672 __rangeof(struct mount, mnt_startzero, mnt_endzero));
673 mp->mnt_kern_flag = 0;
675 mp->mnt_rootvnode = NULL;
676 mp->mnt_vnodecovered = NULL;
679 TAILQ_INIT(&mp->mnt_nvnodelist);
680 mp->mnt_nvnodelistsize = 0;
681 TAILQ_INIT(&mp->mnt_lazyvnodelist);
682 mp->mnt_lazyvnodelistsize = 0;
683 if (mp->mnt_ref != 0 || mp->mnt_lockref != 0 ||
684 mp->mnt_writeopcount != 0)
685 panic("%s: non-zero counters on new mp %p\n", __func__, mp);
686 if (mp->mnt_vfs_ops != 1)
687 panic("%s: vfs_ops should be 1 but %d found\n", __func__,
689 (void) vfs_busy(mp, MBF_NOWAIT);
690 atomic_add_acq_int(&vfsp->vfc_refcount, 1);
691 mp->mnt_op = vfsp->vfc_vfsops;
693 mp->mnt_stat.f_type = vfsp->vfc_typenum;
695 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
696 mp->mnt_vnodecovered = vp;
697 mp->mnt_cred = crdup(cred);
698 mp->mnt_stat.f_owner = cred->cr_uid;
699 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
700 mp->mnt_iosize_max = DFLTPHYS;
703 mac_mount_create(cred, mp);
705 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
706 mp->mnt_upper_pending = 0;
707 TAILQ_INIT(&mp->mnt_uppers);
708 TAILQ_INIT(&mp->mnt_notify);
709 mp->mnt_taskqueue_flags = 0;
710 mp->mnt_unmount_retries = 0;
715 * Destroy the mount struct previously allocated by vfs_mount_alloc().
718 vfs_mount_destroy(struct mount *mp)
721 if (mp->mnt_vfs_ops == 0)
722 panic("%s: entered with zero vfs_ops\n", __func__);
724 vfs_assert_mount_counters(mp);
727 mp->mnt_kern_flag |= MNTK_REFEXPIRE;
728 if (mp->mnt_kern_flag & MNTK_MWAIT) {
729 mp->mnt_kern_flag &= ~MNTK_MWAIT;
733 msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0);
734 KASSERT(mp->mnt_ref == 0,
735 ("%s: invalid refcount in the drain path @ %s:%d", __func__,
736 __FILE__, __LINE__));
737 if (mp->mnt_writeopcount != 0)
738 panic("vfs_mount_destroy: nonzero writeopcount");
739 if (mp->mnt_secondary_writes != 0)
740 panic("vfs_mount_destroy: nonzero secondary_writes");
741 atomic_subtract_rel_int(&mp->mnt_vfc->vfc_refcount, 1);
742 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
745 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
746 vn_printf(vp, "dangling vnode ");
747 panic("unmount: dangling vnode");
749 KASSERT(mp->mnt_upper_pending == 0, ("mnt_upper_pending"));
750 KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers"));
751 KASSERT(TAILQ_EMPTY(&mp->mnt_notify), ("mnt_notify"));
752 if (mp->mnt_nvnodelistsize != 0)
753 panic("vfs_mount_destroy: nonzero nvnodelistsize");
754 if (mp->mnt_lazyvnodelistsize != 0)
755 panic("vfs_mount_destroy: nonzero lazyvnodelistsize");
756 if (mp->mnt_lockref != 0)
757 panic("vfs_mount_destroy: nonzero lock refcount");
760 if (mp->mnt_vfs_ops != 1)
761 panic("%s: vfs_ops should be 1 but %d found\n", __func__,
764 if (mp->mnt_rootvnode != NULL)
765 panic("%s: mount point still has a root vnode %p\n", __func__,
768 if (mp->mnt_vnodecovered != NULL)
769 vrele(mp->mnt_vnodecovered);
771 mac_mount_destroy(mp);
773 if (mp->mnt_opt != NULL)
774 vfs_freeopts(mp->mnt_opt);
775 crfree(mp->mnt_cred);
776 uma_zfree(mount_zone, mp);
780 vfs_should_downgrade_to_ro_mount(uint64_t fsflags, int error)
782 /* This is an upgrade of an exisiting mount. */
783 if ((fsflags & MNT_UPDATE) != 0)
785 /* This is already an R/O mount. */
786 if ((fsflags & MNT_RDONLY) != 0)
790 case ENODEV: /* generic, geom, ... */
791 case EACCES: /* cam/scsi, ... */
792 case EROFS: /* md, mmcsd, ... */
794 * These errors can be returned by the storage layer to signal
795 * that the media is read-only. No harm in the R/O mount
796 * attempt if the error was returned for some other reason.
805 vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions)
807 struct vfsoptlist *optlist;
808 struct vfsopt *opt, *tmp_opt;
809 char *fstype, *fspath, *errmsg;
810 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
813 errmsg = fspath = NULL;
814 errmsg_len = fspathlen = 0;
816 autoro = default_autoro;
818 error = vfs_buildopts(fsoptions, &optlist);
822 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
823 errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
826 * We need these two options before the others,
827 * and they are mandatory for any filesystem.
828 * Ensure they are NUL terminated as well.
831 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
832 if (error || fstypelen <= 0 || fstype[fstypelen - 1] != '\0') {
835 strncpy(errmsg, "Invalid fstype", errmsg_len);
839 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
840 if (error || fspathlen <= 0 || fspath[fspathlen - 1] != '\0') {
843 strncpy(errmsg, "Invalid fspath", errmsg_len);
848 * We need to see if we have the "update" option
849 * before we call vfs_domount(), since vfs_domount() has special
850 * logic based on MNT_UPDATE. This is very important
851 * when we want to update the root filesystem.
853 TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) {
856 if (strcmp(opt->name, "update") == 0) {
857 fsflags |= MNT_UPDATE;
860 else if (strcmp(opt->name, "async") == 0)
861 fsflags |= MNT_ASYNC;
862 else if (strcmp(opt->name, "force") == 0) {
863 fsflags |= MNT_FORCE;
866 else if (strcmp(opt->name, "reload") == 0) {
867 fsflags |= MNT_RELOAD;
870 else if (strcmp(opt->name, "multilabel") == 0)
871 fsflags |= MNT_MULTILABEL;
872 else if (strcmp(opt->name, "noasync") == 0)
873 fsflags &= ~MNT_ASYNC;
874 else if (strcmp(opt->name, "noatime") == 0)
875 fsflags |= MNT_NOATIME;
876 else if (strcmp(opt->name, "atime") == 0) {
877 free(opt->name, M_MOUNT);
878 opt->name = strdup("nonoatime", M_MOUNT);
880 else if (strcmp(opt->name, "noclusterr") == 0)
881 fsflags |= MNT_NOCLUSTERR;
882 else if (strcmp(opt->name, "clusterr") == 0) {
883 free(opt->name, M_MOUNT);
884 opt->name = strdup("nonoclusterr", M_MOUNT);
886 else if (strcmp(opt->name, "noclusterw") == 0)
887 fsflags |= MNT_NOCLUSTERW;
888 else if (strcmp(opt->name, "clusterw") == 0) {
889 free(opt->name, M_MOUNT);
890 opt->name = strdup("nonoclusterw", M_MOUNT);
892 else if (strcmp(opt->name, "noexec") == 0)
893 fsflags |= MNT_NOEXEC;
894 else if (strcmp(opt->name, "exec") == 0) {
895 free(opt->name, M_MOUNT);
896 opt->name = strdup("nonoexec", M_MOUNT);
898 else if (strcmp(opt->name, "nosuid") == 0)
899 fsflags |= MNT_NOSUID;
900 else if (strcmp(opt->name, "suid") == 0) {
901 free(opt->name, M_MOUNT);
902 opt->name = strdup("nonosuid", M_MOUNT);
904 else if (strcmp(opt->name, "nosymfollow") == 0)
905 fsflags |= MNT_NOSYMFOLLOW;
906 else if (strcmp(opt->name, "symfollow") == 0) {
907 free(opt->name, M_MOUNT);
908 opt->name = strdup("nonosymfollow", M_MOUNT);
910 else if (strcmp(opt->name, "noro") == 0) {
911 fsflags &= ~MNT_RDONLY;
914 else if (strcmp(opt->name, "rw") == 0) {
915 fsflags &= ~MNT_RDONLY;
918 else if (strcmp(opt->name, "ro") == 0) {
919 fsflags |= MNT_RDONLY;
922 else if (strcmp(opt->name, "rdonly") == 0) {
923 free(opt->name, M_MOUNT);
924 opt->name = strdup("ro", M_MOUNT);
925 fsflags |= MNT_RDONLY;
928 else if (strcmp(opt->name, "autoro") == 0) {
932 else if (strcmp(opt->name, "suiddir") == 0)
933 fsflags |= MNT_SUIDDIR;
934 else if (strcmp(opt->name, "sync") == 0)
935 fsflags |= MNT_SYNCHRONOUS;
936 else if (strcmp(opt->name, "union") == 0)
937 fsflags |= MNT_UNION;
938 else if (strcmp(opt->name, "automounted") == 0) {
939 fsflags |= MNT_AUTOMOUNTED;
941 } else if (strcmp(opt->name, "nocover") == 0) {
942 fsflags |= MNT_NOCOVER;
944 } else if (strcmp(opt->name, "cover") == 0) {
945 fsflags &= ~MNT_NOCOVER;
947 } else if (strcmp(opt->name, "emptydir") == 0) {
948 fsflags |= MNT_EMPTYDIR;
950 } else if (strcmp(opt->name, "noemptydir") == 0) {
951 fsflags &= ~MNT_EMPTYDIR;
955 vfs_freeopt(optlist, opt);
959 * Be ultra-paranoid about making sure the type and fspath
960 * variables will fit in our mp buffers, including the
963 if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) {
964 error = ENAMETOOLONG;
968 error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
969 if (error == ENOENT) {
972 strncpy(errmsg, "Invalid fstype", errmsg_len);
977 * See if we can mount in the read-only mode if the error code suggests
978 * that it could be possible and the mount options allow for that.
979 * Never try it if "[no]{ro|rw}" has been explicitly requested and not
980 * overridden by "autoro".
982 if (autoro && vfs_should_downgrade_to_ro_mount(fsflags, error)) {
983 printf("%s: R/W mount failed, possibly R/O media,"
984 " trying R/O mount\n", __func__);
985 fsflags |= MNT_RDONLY;
986 error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
989 /* copyout the errmsg */
990 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
991 && errmsg_len > 0 && errmsg != NULL) {
992 if (fsoptions->uio_segflg == UIO_SYSSPACE) {
994 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
995 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
998 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
999 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
1003 if (optlist != NULL)
1004 vfs_freeopts(optlist);
1011 #ifndef _SYS_SYSPROTO_H_
1021 sys_mount(struct thread *td, struct mount_args *uap)
1024 struct vfsconf *vfsp = NULL;
1025 struct mntarg *ma = NULL;
1030 * Mount flags are now 64-bits. On 32-bit architectures only
1031 * 32-bits are passed in, but from here on everything handles
1032 * 64-bit flags correctly.
1036 AUDIT_ARG_FFLAGS(flags);
1039 * Filter out MNT_ROOTFS. We do not want clients of mount() in
1040 * userspace to set this flag, but we must filter it out if we want
1041 * MNT_UPDATE on the root file system to work.
1042 * MNT_ROOTFS should only be set by the kernel when mounting its
1045 flags &= ~MNT_ROOTFS;
1047 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
1048 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
1050 free(fstype, M_TEMP);
1054 AUDIT_ARG_TEXT(fstype);
1055 vfsp = vfs_byname_kld(fstype, td, &error);
1056 free(fstype, M_TEMP);
1059 if (((vfsp->vfc_flags & VFCF_SBDRY) != 0 &&
1060 vfsp->vfc_vfsops_sd->vfs_cmount == NULL) ||
1061 ((vfsp->vfc_flags & VFCF_SBDRY) == 0 &&
1062 vfsp->vfc_vfsops->vfs_cmount == NULL))
1063 return (EOPNOTSUPP);
1065 ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN);
1066 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
1067 ma = mount_argb(ma, flags & MNT_RDONLY, "noro");
1068 ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid");
1069 ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec");
1071 if ((vfsp->vfc_flags & VFCF_SBDRY) != 0)
1072 return (vfsp->vfc_vfsops_sd->vfs_cmount(ma, uap->data, flags));
1073 return (vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags));
1077 * vfs_domount_first(): first file system mount (not update)
1081 struct thread *td, /* Calling thread. */
1082 struct vfsconf *vfsp, /* File system type. */
1083 char *fspath, /* Mount path. */
1084 struct vnode *vp, /* Vnode to be covered. */
1085 uint64_t fsflags, /* Flags common to all filesystems. */
1086 struct vfsoptlist **optlist /* Options local to the filesystem. */
1091 struct vnode *newdp, *rootvp;
1095 ASSERT_VOP_ELOCKED(vp, __func__);
1096 KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here"));
1098 if ((fsflags & MNT_EMPTYDIR) != 0) {
1099 error = vfs_emptydir(vp);
1107 * If the jail of the calling thread lacks permission for this type of
1108 * file system, or is trying to cover its own root, deny immediately.
1110 if (jailed(td->td_ucred) && (!prison_allow(td->td_ucred,
1111 vfsp->vfc_prison_flag) || vp == td->td_ucred->cr_prison->pr_root)) {
1117 * If the user is not root, ensure that they own the directory
1118 * onto which we are attempting to mount.
1120 error = VOP_GETATTR(vp, &va, td->td_ucred);
1121 if (error == 0 && va.va_uid != td->td_ucred->cr_uid)
1122 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN);
1124 error = vinvalbuf(vp, V_SAVE, 0, 0);
1125 if (error == 0 && vp->v_type != VDIR)
1129 if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL)
1130 vp->v_iflag |= VI_MOUNT;
1139 vn_seqc_write_begin(vp);
1142 /* Allocate and initialize the filesystem. */
1143 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred);
1144 /* XXXMAC: pass to vfs_mount_alloc? */
1145 mp->mnt_optnew = *optlist;
1146 /* Set the mount level flags. */
1147 mp->mnt_flag = (fsflags & (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY));
1150 * Mount the filesystem.
1151 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1152 * get. No freeing of cn_pnbuf.
1156 if ((error = VFS_MOUNT(mp)) != 0 ||
1157 (error1 = VFS_STATFS(mp, &mp->mnt_stat)) != 0 ||
1158 (error1 = VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) != 0) {
1162 rootvp = vfs_cache_root_clear(mp);
1163 if (rootvp != NULL) {
1167 (void)vn_start_write(NULL, &mp, V_WAIT);
1169 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_UNMOUNTF;
1172 error = VFS_UNMOUNT(mp, 0);
1173 vn_finished_write(mp);
1176 "failed post-mount (%d): rollback unmount returned %d\n",
1183 mp->mnt_vnodecovered = NULL;
1185 /* XXXKIB wait for mnt_lockref drain? */
1186 vfs_mount_destroy(mp);
1189 vp->v_iflag &= ~VI_MOUNT;
1191 if (rootvp != NULL) {
1192 vn_seqc_write_end(rootvp);
1195 vn_seqc_write_end(vp);
1199 vn_seqc_write_begin(newdp);
1202 if (mp->mnt_opt != NULL)
1203 vfs_freeopts(mp->mnt_opt);
1204 mp->mnt_opt = mp->mnt_optnew;
1208 * Prevent external consumers of mount options from reading mnt_optnew.
1210 mp->mnt_optnew = NULL;
1213 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1214 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1215 mp->mnt_kern_flag |= MNTK_ASYNC;
1217 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1221 vn_irflag_set_locked(vp, VIRF_MOUNTPOINT);
1222 vp->v_mountedhere = mp;
1227 * We need to lock both vnodes.
1229 * Use vn_lock_pair to avoid establishing an ordering between vnodes
1230 * from different filesystems.
1232 vn_lock_pair(vp, false, newdp, false);
1235 vp->v_iflag &= ~VI_MOUNT;
1237 /* Place the new filesystem at the end of the mount list. */
1238 mtx_lock(&mountlist_mtx);
1239 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1240 mtx_unlock(&mountlist_mtx);
1241 vfs_event_signal(NULL, VQ_MOUNT, 0);
1243 EVENTHANDLER_DIRECT_INVOKE(vfs_mounted, mp, newdp, td);
1245 mount_devctl_event("MOUNT", mp, false);
1246 mountcheckdirs(vp, newdp);
1247 vn_seqc_write_end(vp);
1248 vn_seqc_write_end(newdp);
1250 if ((mp->mnt_flag & MNT_RDONLY) == 0)
1251 vfs_allocate_syncvnode(mp);
1258 * vfs_domount_update(): update of mounted file system
1262 struct thread *td, /* Calling thread. */
1263 struct vnode *vp, /* Mount point vnode. */
1264 uint64_t fsflags, /* Flags common to all filesystems. */
1265 struct vfsoptlist **optlist /* Options local to the filesystem. */
1268 struct export_args export;
1269 struct o2export_args o2export;
1270 struct vnode *rootvp;
1273 int error, export_error, i, len;
1277 ASSERT_VOP_ELOCKED(vp, __func__);
1278 KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here"));
1281 if ((vp->v_vflag & VV_ROOT) == 0) {
1282 if (vfs_copyopt(*optlist, "export", &export, sizeof(export))
1292 * We only allow the filesystem to be reloaded if it
1293 * is currently mounted read-only.
1295 flag = mp->mnt_flag;
1296 if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) {
1298 return (EOPNOTSUPP); /* Needs translation */
1301 * Only privileged root, or (if MNT_USER is set) the user that
1302 * did the original mount is permitted to update it.
1304 error = vfs_suser(mp, td);
1309 if (vfs_busy(mp, MBF_NOWAIT)) {
1314 if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) {
1320 vp->v_iflag |= VI_MOUNT;
1325 vn_seqc_write_begin(vp);
1329 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1334 mp->mnt_flag &= ~MNT_UPDATEMASK;
1335 mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE |
1336 MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY);
1337 if ((mp->mnt_flag & MNT_ASYNC) == 0)
1338 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1339 rootvp = vfs_cache_root_clear(mp);
1341 mp->mnt_optnew = *optlist;
1342 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
1345 * Mount the filesystem.
1346 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1347 * get. No freeing of cn_pnbuf.
1349 error = VFS_MOUNT(mp);
1352 /* Process the export option. */
1353 if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp,
1355 /* Assume that there is only 1 ABI for each length. */
1357 case (sizeof(struct oexport_args)):
1358 bzero(&o2export, sizeof(o2export));
1360 case (sizeof(o2export)):
1361 bcopy(bufp, &o2export, len);
1362 export.ex_flags = (uint64_t)o2export.ex_flags;
1363 export.ex_root = o2export.ex_root;
1364 export.ex_uid = o2export.ex_anon.cr_uid;
1365 export.ex_groups = NULL;
1366 export.ex_ngroups = o2export.ex_anon.cr_ngroups;
1367 if (export.ex_ngroups > 0) {
1368 if (export.ex_ngroups <= XU_NGROUPS) {
1369 export.ex_groups = malloc(
1370 export.ex_ngroups * sizeof(gid_t),
1372 for (i = 0; i < export.ex_ngroups; i++)
1373 export.ex_groups[i] =
1374 o2export.ex_anon.cr_groups[i];
1376 export_error = EINVAL;
1377 } else if (export.ex_ngroups < 0)
1378 export_error = EINVAL;
1379 export.ex_addr = o2export.ex_addr;
1380 export.ex_addrlen = o2export.ex_addrlen;
1381 export.ex_mask = o2export.ex_mask;
1382 export.ex_masklen = o2export.ex_masklen;
1383 export.ex_indexfile = o2export.ex_indexfile;
1384 export.ex_numsecflavors = o2export.ex_numsecflavors;
1385 if (export.ex_numsecflavors < MAXSECFLAVORS) {
1386 for (i = 0; i < export.ex_numsecflavors; i++)
1387 export.ex_secflavors[i] =
1388 o2export.ex_secflavors[i];
1390 export_error = EINVAL;
1391 if (export_error == 0)
1392 export_error = vfs_export(mp, &export);
1393 free(export.ex_groups, M_TEMP);
1395 case (sizeof(export)):
1396 bcopy(bufp, &export, len);
1398 if (export.ex_ngroups > 0) {
1399 if (export.ex_ngroups <= NGROUPS_MAX) {
1400 grps = malloc(export.ex_ngroups *
1401 sizeof(gid_t), M_TEMP, M_WAITOK);
1402 export_error = copyin(export.ex_groups,
1403 grps, export.ex_ngroups *
1405 if (export_error == 0)
1406 export.ex_groups = grps;
1408 export_error = EINVAL;
1409 } else if (export.ex_ngroups == 0)
1410 export.ex_groups = NULL;
1412 export_error = EINVAL;
1413 if (export_error == 0)
1414 export_error = vfs_export(mp, &export);
1418 export_error = EINVAL;
1425 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE |
1429 * If we fail, restore old mount flags. MNT_QUOTA is special,
1430 * because it is not part of MNT_UPDATEMASK, but it could have
1431 * changed in the meantime if quotactl(2) was called.
1432 * All in all we want current value of MNT_QUOTA, not the old
1435 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA);
1437 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1438 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1439 mp->mnt_kern_flag |= MNTK_ASYNC;
1441 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1447 mount_devctl_event("REMOUNT", mp, true);
1448 if (mp->mnt_opt != NULL)
1449 vfs_freeopts(mp->mnt_opt);
1450 mp->mnt_opt = mp->mnt_optnew;
1452 (void)VFS_STATFS(mp, &mp->mnt_stat);
1454 * Prevent external consumers of mount options from reading
1457 mp->mnt_optnew = NULL;
1459 if ((mp->mnt_flag & MNT_RDONLY) == 0)
1460 vfs_allocate_syncvnode(mp);
1462 vfs_deallocate_syncvnode(mp);
1465 if (rootvp != NULL) {
1466 vn_seqc_write_end(rootvp);
1469 vn_seqc_write_end(vp);
1472 vp->v_iflag &= ~VI_MOUNT;
1475 return (error != 0 ? error : export_error);
1479 * vfs_domount(): actually attempt a filesystem mount.
1483 struct thread *td, /* Calling thread. */
1484 const char *fstype, /* Filesystem type. */
1485 char *fspath, /* Mount path. */
1486 uint64_t fsflags, /* Flags common to all filesystems. */
1487 struct vfsoptlist **optlist /* Options local to the filesystem. */
1490 struct vfsconf *vfsp;
1491 struct nameidata nd;
1497 * Be ultra-paranoid about making sure the type and fspath
1498 * variables will fit in our mp buffers, including the
1501 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
1502 return (ENAMETOOLONG);
1504 if (jailed(td->td_ucred) || usermount == 0) {
1505 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
1510 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
1512 if (fsflags & MNT_EXPORTED) {
1513 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
1517 if (fsflags & MNT_SUIDDIR) {
1518 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
1523 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
1525 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
1526 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
1527 fsflags |= MNT_NOSUID | MNT_USER;
1530 /* Load KLDs before we lock the covered vnode to avoid reversals. */
1532 if ((fsflags & MNT_UPDATE) == 0) {
1533 /* Don't try to load KLDs if we're mounting the root. */
1534 if (fsflags & MNT_ROOTFS) {
1535 if ((vfsp = vfs_byname(fstype)) == NULL)
1538 if ((vfsp = vfs_byname_kld(fstype, td, &error)) == NULL)
1544 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE.
1546 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1547 UIO_SYSSPACE, fspath, td);
1551 NDFREE(&nd, NDF_ONLY_PNBUF);
1553 if ((fsflags & MNT_UPDATE) == 0) {
1554 if ((vp->v_vflag & VV_ROOT) != 0 &&
1555 (fsflags & MNT_NOCOVER) != 0) {
1559 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1560 strcpy(pathbuf, fspath);
1561 error = vn_path_to_global_path(td, vp, pathbuf, MNAMELEN);
1563 error = vfs_domount_first(td, vfsp, pathbuf, vp,
1566 free(pathbuf, M_TEMP);
1568 error = vfs_domount_update(td, vp, fsflags, optlist);
1574 * Unmount a filesystem.
1576 * Note: unmount takes a path to the vnode mounted on as argument, not
1577 * special file (as before).
1579 #ifndef _SYS_SYSPROTO_H_
1580 struct unmount_args {
1587 sys_unmount(struct thread *td, struct unmount_args *uap)
1590 return (kern_unmount(td, uap->path, uap->flags));
1594 kern_unmount(struct thread *td, const char *path, int flags)
1596 struct nameidata nd;
1598 char *fsidbuf, *pathbuf;
1602 AUDIT_ARG_VALUE(flags);
1603 if (jailed(td->td_ucred) || usermount == 0) {
1604 error = priv_check(td, PRIV_VFS_UNMOUNT);
1609 if (flags & MNT_BYFSID) {
1610 fsidbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1611 error = copyinstr(path, fsidbuf, MNAMELEN, NULL);
1613 free(fsidbuf, M_TEMP);
1617 AUDIT_ARG_TEXT(fsidbuf);
1618 /* Decode the filesystem ID. */
1619 if (sscanf(fsidbuf, "FSID:%d:%d", &fsid.val[0], &fsid.val[1]) != 2) {
1620 free(fsidbuf, M_TEMP);
1624 mp = vfs_getvfs(&fsid);
1625 free(fsidbuf, M_TEMP);
1630 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1631 error = copyinstr(path, pathbuf, MNAMELEN, NULL);
1633 free(pathbuf, M_TEMP);
1638 * Try to find global path for path argument.
1640 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1641 UIO_SYSSPACE, pathbuf, td);
1642 if (namei(&nd) == 0) {
1643 NDFREE(&nd, NDF_ONLY_PNBUF);
1644 error = vn_path_to_global_path(td, nd.ni_vp, pathbuf,
1649 mtx_lock(&mountlist_mtx);
1650 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1651 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) {
1656 mtx_unlock(&mountlist_mtx);
1657 free(pathbuf, M_TEMP);
1660 * Previously we returned ENOENT for a nonexistent path and
1661 * EINVAL for a non-mountpoint. We cannot tell these apart
1662 * now, so in the !MNT_BYFSID case return the more likely
1663 * EINVAL for compatibility.
1670 * Don't allow unmounting the root filesystem.
1672 if (mp->mnt_flag & MNT_ROOTFS) {
1676 error = dounmount(mp, flags, td);
1681 * Return error if any of the vnodes, ignoring the root vnode
1682 * and the syncer vnode, have non-zero usecount.
1684 * This function is purely advisory - it can return false positives
1688 vfs_check_usecounts(struct mount *mp)
1690 struct vnode *vp, *mvp;
1692 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1693 if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON &&
1694 vp->v_usecount != 0) {
1696 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1706 dounmount_cleanup(struct mount *mp, struct vnode *coveredvp, int mntkflags)
1709 mtx_assert(MNT_MTX(mp), MA_OWNED);
1710 mp->mnt_kern_flag &= ~mntkflags;
1711 if ((mp->mnt_kern_flag & MNTK_MWAIT) != 0) {
1712 mp->mnt_kern_flag &= ~MNTK_MWAIT;
1715 vfs_op_exit_locked(mp);
1717 if (coveredvp != NULL) {
1718 VOP_UNLOCK(coveredvp);
1721 vn_finished_write(mp);
1725 * There are various reference counters associated with the mount point.
1726 * Normally it is permitted to modify them without taking the mnt ilock,
1727 * but this behavior can be temporarily disabled if stable value is needed
1728 * or callers are expected to block (e.g. to not allow new users during
1732 vfs_op_enter(struct mount *mp)
1734 struct mount_pcpu *mpcpu;
1739 if (mp->mnt_vfs_ops > 1) {
1743 vfs_op_barrier_wait(mp);
1745 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1747 mp->mnt_ref += mpcpu->mntp_ref;
1748 mpcpu->mntp_ref = 0;
1750 mp->mnt_lockref += mpcpu->mntp_lockref;
1751 mpcpu->mntp_lockref = 0;
1753 mp->mnt_writeopcount += mpcpu->mntp_writeopcount;
1754 mpcpu->mntp_writeopcount = 0;
1756 if (mp->mnt_ref <= 0 || mp->mnt_lockref < 0 || mp->mnt_writeopcount < 0)
1757 panic("%s: invalid count(s) on mp %p: ref %d lockref %d writeopcount %d\n",
1758 __func__, mp, mp->mnt_ref, mp->mnt_lockref, mp->mnt_writeopcount);
1760 vfs_assert_mount_counters(mp);
1764 vfs_op_exit_locked(struct mount *mp)
1767 mtx_assert(MNT_MTX(mp), MA_OWNED);
1769 if (mp->mnt_vfs_ops <= 0)
1770 panic("%s: invalid vfs_ops count %d for mp %p\n",
1771 __func__, mp->mnt_vfs_ops, mp);
1776 vfs_op_exit(struct mount *mp)
1780 vfs_op_exit_locked(mp);
1784 struct vfs_op_barrier_ipi {
1786 struct smp_rendezvous_cpus_retry_arg srcra;
1790 vfs_op_action_func(void *arg)
1792 struct vfs_op_barrier_ipi *vfsopipi;
1795 vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1798 if (!vfs_op_thread_entered(mp))
1799 smp_rendezvous_cpus_done(arg);
1803 vfs_op_wait_func(void *arg, int cpu)
1805 struct vfs_op_barrier_ipi *vfsopipi;
1807 struct mount_pcpu *mpcpu;
1809 vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1812 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1813 while (atomic_load_int(&mpcpu->mntp_thread_in_ops))
1818 vfs_op_barrier_wait(struct mount *mp)
1820 struct vfs_op_barrier_ipi vfsopipi;
1824 smp_rendezvous_cpus_retry(all_cpus,
1825 smp_no_rendezvous_barrier,
1827 smp_no_rendezvous_barrier,
1834 vfs_assert_mount_counters(struct mount *mp)
1836 struct mount_pcpu *mpcpu;
1839 if (mp->mnt_vfs_ops == 0)
1843 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1844 if (mpcpu->mntp_ref != 0 ||
1845 mpcpu->mntp_lockref != 0 ||
1846 mpcpu->mntp_writeopcount != 0)
1847 vfs_dump_mount_counters(mp);
1852 vfs_dump_mount_counters(struct mount *mp)
1854 struct mount_pcpu *mpcpu;
1855 int ref, lockref, writeopcount;
1858 printf("%s: mp %p vfs_ops %d\n", __func__, mp, mp->mnt_vfs_ops);
1863 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1864 printf("%d ", mpcpu->mntp_ref);
1865 ref += mpcpu->mntp_ref;
1868 printf(" lockref : ");
1869 lockref = mp->mnt_lockref;
1871 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1872 printf("%d ", mpcpu->mntp_lockref);
1873 lockref += mpcpu->mntp_lockref;
1876 printf("writeopcount: ");
1877 writeopcount = mp->mnt_writeopcount;
1879 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1880 printf("%d ", mpcpu->mntp_writeopcount);
1881 writeopcount += mpcpu->mntp_writeopcount;
1885 printf("counter struct total\n");
1886 printf("ref %-5d %-5d\n", mp->mnt_ref, ref);
1887 printf("lockref %-5d %-5d\n", mp->mnt_lockref, lockref);
1888 printf("writeopcount %-5d %-5d\n", mp->mnt_writeopcount, writeopcount);
1890 panic("invalid counts on struct mount");
1895 vfs_mount_fetch_counter(struct mount *mp, enum mount_counter which)
1897 struct mount_pcpu *mpcpu;
1904 case MNT_COUNT_LOCKREF:
1905 sum = mp->mnt_lockref;
1907 case MNT_COUNT_WRITEOPCOUNT:
1908 sum = mp->mnt_writeopcount;
1913 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1916 sum += mpcpu->mntp_ref;
1918 case MNT_COUNT_LOCKREF:
1919 sum += mpcpu->mntp_lockref;
1921 case MNT_COUNT_WRITEOPCOUNT:
1922 sum += mpcpu->mntp_writeopcount;
1930 deferred_unmount_enqueue(struct mount *mp, uint64_t flags, bool requeue,
1936 mtx_lock(&deferred_unmount_lock);
1937 if ((mp->mnt_taskqueue_flags & MNT_DEFERRED) == 0 || requeue) {
1938 mp->mnt_taskqueue_flags = flags | MNT_DEFERRED;
1939 STAILQ_INSERT_TAIL(&deferred_unmount_list, mp,
1940 mnt_taskqueue_link);
1943 mtx_unlock(&deferred_unmount_lock);
1946 taskqueue_enqueue_timeout(taskqueue_deferred_unmount,
1947 &deferred_unmount_task, timeout_ticks);
1954 * Taskqueue handler for processing async/recursive unmounts
1957 vfs_deferred_unmount(void *argi __unused, int pending __unused)
1959 STAILQ_HEAD(, mount) local_unmounts;
1961 struct mount *mp, *tmp;
1963 unsigned int retries;
1966 STAILQ_INIT(&local_unmounts);
1967 mtx_lock(&deferred_unmount_lock);
1968 STAILQ_CONCAT(&local_unmounts, &deferred_unmount_list);
1969 mtx_unlock(&deferred_unmount_lock);
1971 STAILQ_FOREACH_SAFE(mp, &local_unmounts, mnt_taskqueue_link, tmp) {
1972 flags = mp->mnt_taskqueue_flags;
1973 KASSERT((flags & MNT_DEFERRED) != 0,
1974 ("taskqueue unmount without MNT_DEFERRED"));
1975 error = dounmount(mp, flags, curthread);
1978 unmounted = ((mp->mnt_kern_flag & MNTK_REFEXPIRE) != 0);
1982 * The deferred unmount thread is the only thread that
1983 * modifies the retry counts, so locking/atomics aren't
1986 retries = (mp->mnt_unmount_retries)++;
1987 deferred_unmount_total_retries++;
1988 if (!unmounted && retries < deferred_unmount_retry_limit) {
1989 deferred_unmount_enqueue(mp, flags, true,
1990 -deferred_unmount_retry_delay_hz);
1992 if (retries >= deferred_unmount_retry_limit) {
1993 printf("giving up on deferred unmount "
1994 "of %s after %d retries, error %d\n",
1995 mp->mnt_stat.f_mntonname, retries, error);
2004 * Do the actual filesystem unmount.
2007 dounmount(struct mount *mp, uint64_t flags, struct thread *td)
2009 struct mount_upper_node *upper;
2010 struct vnode *coveredvp, *rootvp;
2012 uint64_t async_flag;
2014 unsigned int retries;
2016 KASSERT((flags & MNT_DEFERRED) == 0 ||
2017 (flags & (MNT_RECURSE | MNT_FORCE)) == (MNT_RECURSE | MNT_FORCE),
2018 ("MNT_DEFERRED requires MNT_RECURSE | MNT_FORCE"));
2021 * If the caller has explicitly requested the unmount to be handled by
2022 * the taskqueue and we're not already in taskqueue context, queue
2023 * up the unmount request and exit. This is done prior to any
2024 * credential checks; MNT_DEFERRED should be used only for kernel-
2025 * initiated unmounts and will therefore be processed with the
2026 * (kernel) credentials of the taskqueue thread. Still, callers
2027 * should be sure this is the behavior they want.
2029 if ((flags & MNT_DEFERRED) != 0 &&
2030 taskqueue_member(taskqueue_deferred_unmount, curthread) == 0) {
2031 if (!deferred_unmount_enqueue(mp, flags, false, 0))
2033 return (EINPROGRESS);
2037 * Only privileged root, or (if MNT_USER is set) the user that did the
2038 * original mount is permitted to unmount this filesystem.
2039 * This check should be made prior to queueing up any recursive
2040 * unmounts of upper filesystems. Those unmounts will be executed
2041 * with kernel thread credentials and are expected to succeed, so
2042 * we must at least ensure the originating context has sufficient
2043 * privilege to unmount the base filesystem before proceeding with
2046 error = vfs_suser(mp, td);
2048 KASSERT((flags & MNT_DEFERRED) == 0,
2049 ("taskqueue unmount with insufficient privilege"));
2054 if (recursive_forced_unmount && ((flags & MNT_FORCE) != 0))
2055 flags |= MNT_RECURSE;
2057 if ((flags & MNT_RECURSE) != 0) {
2058 KASSERT((flags & MNT_FORCE) != 0,
2059 ("MNT_RECURSE requires MNT_FORCE"));
2063 * Set MNTK_RECURSE to prevent new upper mounts from being
2064 * added, and note that an operation on the uppers list is in
2065 * progress. This will ensure that unregistration from the
2066 * uppers list, and therefore any pending unmount of the upper
2067 * FS, can't complete until after we finish walking the list.
2069 mp->mnt_kern_flag |= MNTK_RECURSE;
2070 mp->mnt_upper_pending++;
2071 TAILQ_FOREACH(upper, &mp->mnt_uppers, mnt_upper_link) {
2072 retries = upper->mp->mnt_unmount_retries;
2073 if (retries > deferred_unmount_retry_limit) {
2080 if (!deferred_unmount_enqueue(upper->mp, flags,
2085 mp->mnt_upper_pending--;
2086 if ((mp->mnt_kern_flag & MNTK_UPPER_WAITER) != 0 &&
2087 mp->mnt_upper_pending == 0) {
2088 mp->mnt_kern_flag &= ~MNTK_UPPER_WAITER;
2089 wakeup(&mp->mnt_uppers);
2093 * If we're not on the taskqueue, wait until the uppers list
2094 * is drained before proceeding with unmount. Otherwise, if
2095 * we are on the taskqueue and there are still pending uppers,
2096 * just re-enqueue on the end of the taskqueue.
2098 if ((flags & MNT_DEFERRED) == 0) {
2099 while (error == 0 && !TAILQ_EMPTY(&mp->mnt_uppers)) {
2100 mp->mnt_kern_flag |= MNTK_TASKQUEUE_WAITER;
2101 error = msleep(&mp->mnt_taskqueue_link,
2102 MNT_MTX(mp), PCATCH, "umntqw", 0);
2109 } else if (!TAILQ_EMPTY(&mp->mnt_uppers)) {
2112 deferred_unmount_enqueue(mp, flags, true, 0);
2116 KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers not empty"));
2119 /* Allow the taskqueue to safely re-enqueue on failure */
2120 if ((flags & MNT_DEFERRED) != 0)
2123 if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
2124 mnt_gen_r = mp->mnt_gen;
2127 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
2129 * Check for mp being unmounted while waiting for the
2130 * covered vnode lock.
2132 if (coveredvp->v_mountedhere != mp ||
2133 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
2134 VOP_UNLOCK(coveredvp);
2143 vn_start_write(NULL, &mp, V_WAIT | V_MNTREF);
2145 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 ||
2146 (mp->mnt_flag & MNT_UPDATE) != 0 ||
2147 !TAILQ_EMPTY(&mp->mnt_uppers)) {
2148 dounmount_cleanup(mp, coveredvp, 0);
2151 mp->mnt_kern_flag |= MNTK_UNMOUNT;
2152 rootvp = vfs_cache_root_clear(mp);
2153 if (coveredvp != NULL)
2154 vn_seqc_write_begin(coveredvp);
2155 if (flags & MNT_NONBUSY) {
2157 error = vfs_check_usecounts(mp);
2160 vn_seqc_write_end(coveredvp);
2161 dounmount_cleanup(mp, coveredvp, MNTK_UNMOUNT);
2162 if (rootvp != NULL) {
2163 vn_seqc_write_end(rootvp);
2169 /* Allow filesystems to detect that a forced unmount is in progress. */
2170 if (flags & MNT_FORCE) {
2171 mp->mnt_kern_flag |= MNTK_UNMOUNTF;
2174 * Must be done after setting MNTK_UNMOUNTF and before
2175 * waiting for mnt_lockref to become 0.
2181 if (mp->mnt_lockref) {
2182 mp->mnt_kern_flag |= MNTK_DRAINING;
2183 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS,
2187 KASSERT(mp->mnt_lockref == 0,
2188 ("%s: invalid lock refcount in the drain path @ %s:%d",
2189 __func__, __FILE__, __LINE__));
2191 ("%s: invalid return value for msleep in the drain path @ %s:%d",
2192 __func__, __FILE__, __LINE__));
2195 * We want to keep the vnode around so that we can vn_seqc_write_end
2196 * after we are done with unmount. Downgrade our reference to a mere
2197 * hold count so that we don't interefere with anything.
2199 if (rootvp != NULL) {
2204 if (mp->mnt_flag & MNT_EXPUBLIC)
2205 vfs_setpublicfs(NULL, NULL, NULL);
2207 vfs_periodic(mp, MNT_WAIT);
2209 async_flag = mp->mnt_flag & MNT_ASYNC;
2210 mp->mnt_flag &= ~MNT_ASYNC;
2211 mp->mnt_kern_flag &= ~MNTK_ASYNC;
2213 vfs_deallocate_syncvnode(mp);
2214 error = VFS_UNMOUNT(mp, flags);
2215 vn_finished_write(mp);
2217 * If we failed to flush the dirty blocks for this mount point,
2218 * undo all the cdir/rdir and rootvnode changes we made above.
2219 * Unless we failed to do so because the device is reporting that
2220 * it doesn't exist anymore.
2222 if (error && error != ENXIO) {
2224 if ((mp->mnt_flag & MNT_RDONLY) == 0) {
2226 vfs_allocate_syncvnode(mp);
2229 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
2230 mp->mnt_flag |= async_flag;
2231 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
2232 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
2233 mp->mnt_kern_flag |= MNTK_ASYNC;
2234 if (mp->mnt_kern_flag & MNTK_MWAIT) {
2235 mp->mnt_kern_flag &= ~MNTK_MWAIT;
2238 vfs_op_exit_locked(mp);
2241 vn_seqc_write_end(coveredvp);
2242 VOP_UNLOCK(coveredvp);
2245 if (rootvp != NULL) {
2246 vn_seqc_write_end(rootvp);
2252 mtx_lock(&mountlist_mtx);
2253 TAILQ_REMOVE(&mountlist, mp, mnt_list);
2254 mtx_unlock(&mountlist_mtx);
2255 EVENTHANDLER_DIRECT_INVOKE(vfs_unmounted, mp, td);
2256 if (coveredvp != NULL) {
2258 vn_irflag_unset_locked(coveredvp, VIRF_MOUNTPOINT);
2259 coveredvp->v_mountedhere = NULL;
2260 vn_seqc_write_end_locked(coveredvp);
2261 VI_UNLOCK(coveredvp);
2262 VOP_UNLOCK(coveredvp);
2265 mount_devctl_event("UNMOUNT", mp, false);
2266 if (rootvp != NULL) {
2267 vn_seqc_write_end(rootvp);
2270 vfs_event_signal(NULL, VQ_UNMOUNT, 0);
2271 if (rootvnode != NULL && mp == rootvnode->v_mount) {
2275 if (mp == rootdevmp)
2277 if ((flags & MNT_DEFERRED) != 0)
2279 vfs_mount_destroy(mp);
2284 * Report errors during filesystem mounting.
2287 vfs_mount_error(struct mount *mp, const char *fmt, ...)
2289 struct vfsoptlist *moptlist = mp->mnt_optnew;
2294 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
2295 if (error || errmsg == NULL || len <= 0)
2299 vsnprintf(errmsg, (size_t)len, fmt, ap);
2304 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...)
2310 error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len);
2311 if (error || errmsg == NULL || len <= 0)
2315 vsnprintf(errmsg, (size_t)len, fmt, ap);
2320 * ---------------------------------------------------------------------
2321 * Functions for querying mount options/arguments from filesystems.
2325 * Check that no unknown options are given
2328 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
2332 const char **t, *p, *q;
2335 TAILQ_FOREACH(opt, opts, link) {
2338 if (p[0] == 'n' && p[1] == 'o')
2340 for(t = global_opts; *t != NULL; t++) {
2341 if (strcmp(*t, p) == 0)
2344 if (strcmp(*t, q) == 0)
2350 for(t = legal; *t != NULL; t++) {
2351 if (strcmp(*t, p) == 0)
2354 if (strcmp(*t, q) == 0)
2360 snprintf(errmsg, sizeof(errmsg),
2361 "mount option <%s> is unknown", p);
2365 TAILQ_FOREACH(opt, opts, link) {
2366 if (strcmp(opt->name, "errmsg") == 0) {
2367 strncpy((char *)opt->value, errmsg, opt->len);
2372 printf("%s\n", errmsg);
2378 * Get a mount option by its name.
2380 * Return 0 if the option was found, ENOENT otherwise.
2381 * If len is non-NULL it will be filled with the length
2382 * of the option. If buf is non-NULL, it will be filled
2383 * with the address of the option.
2386 vfs_getopt(struct vfsoptlist *opts, const char *name, void **buf, int *len)
2390 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2392 TAILQ_FOREACH(opt, opts, link) {
2393 if (strcmp(name, opt->name) == 0) {
2406 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
2413 TAILQ_FOREACH(opt, opts, link) {
2414 if (strcmp(name, opt->name) == 0) {
2423 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value)
2425 char *opt_value, *vtp;
2429 error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len);
2432 if (opt_len == 0 || opt_value == NULL)
2434 if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0')
2436 iv = strtoq(opt_value, &vtp, 0);
2437 if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0'))
2464 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
2469 TAILQ_FOREACH(opt, opts, link) {
2470 if (strcmp(name, opt->name) != 0)
2473 if (opt->len == 0 ||
2474 ((char *)opt->value)[opt->len - 1] != '\0') {
2478 return (opt->value);
2485 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w,
2490 TAILQ_FOREACH(opt, opts, link) {
2491 if (strcmp(name, opt->name) == 0) {
2504 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
2510 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2512 TAILQ_FOREACH(opt, opts, link) {
2513 if (strcmp(name, opt->name) != 0)
2516 if (opt->len == 0 || opt->value == NULL)
2518 if (((char *)opt->value)[opt->len - 1] != '\0')
2521 ret = vsscanf(opt->value, fmt, ap);
2529 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len)
2533 TAILQ_FOREACH(opt, opts, link) {
2534 if (strcmp(name, opt->name) != 0)
2537 if (opt->value == NULL)
2540 if (opt->len != len)
2542 bcopy(value, opt->value, len);
2550 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len)
2554 TAILQ_FOREACH(opt, opts, link) {
2555 if (strcmp(name, opt->name) != 0)
2558 if (opt->value == NULL)
2564 bcopy(value, opt->value, len);
2572 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value)
2576 TAILQ_FOREACH(opt, opts, link) {
2577 if (strcmp(name, opt->name) != 0)
2580 if (opt->value == NULL)
2581 opt->len = strlen(value) + 1;
2582 else if (strlcpy(opt->value, value, opt->len) >= opt->len)
2590 * Find and copy a mount option.
2592 * The size of the buffer has to be specified
2593 * in len, if it is not the same length as the
2594 * mount option, EINVAL is returned.
2595 * Returns ENOENT if the option is not found.
2598 vfs_copyopt(struct vfsoptlist *opts, const char *name, void *dest, int len)
2602 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
2604 TAILQ_FOREACH(opt, opts, link) {
2605 if (strcmp(name, opt->name) == 0) {
2607 if (len != opt->len)
2609 bcopy(opt->value, dest, opt->len);
2617 __vfs_statfs(struct mount *mp, struct statfs *sbp)
2621 * Filesystems only fill in part of the structure for updates, we
2622 * have to read the entirety first to get all content.
2624 if (sbp != &mp->mnt_stat)
2625 memcpy(sbp, &mp->mnt_stat, sizeof(*sbp));
2628 * Set these in case the underlying filesystem fails to do so.
2630 sbp->f_version = STATFS_VERSION;
2631 sbp->f_namemax = NAME_MAX;
2632 sbp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
2634 return (mp->mnt_op->vfs_statfs(mp, sbp));
2638 vfs_mountedfrom(struct mount *mp, const char *from)
2641 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
2642 strlcpy(mp->mnt_stat.f_mntfromname, from,
2643 sizeof mp->mnt_stat.f_mntfromname);
2647 * ---------------------------------------------------------------------
2648 * This is the api for building mount args and mounting filesystems from
2649 * inside the kernel.
2651 * The API works by accumulation of individual args. First error is
2654 * XXX: should be documented in new manpage kernel_mount(9)
2657 /* A memory allocation which must be freed when we are done */
2659 SLIST_ENTRY(mntaarg) next;
2662 /* The header for the mount arguments */
2667 SLIST_HEAD(, mntaarg) list;
2671 * Add a boolean argument.
2673 * flag is the boolean value.
2674 * name must start with "no".
2677 mount_argb(struct mntarg *ma, int flag, const char *name)
2680 KASSERT(name[0] == 'n' && name[1] == 'o',
2681 ("mount_argb(...,%s): name must start with 'no'", name));
2683 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
2687 * Add an argument printf style
2690 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
2693 struct mntaarg *maa;
2698 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2699 SLIST_INIT(&ma->list);
2704 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2706 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2707 ma->v[ma->len].iov_len = strlen(name) + 1;
2710 sb = sbuf_new_auto();
2712 sbuf_vprintf(sb, fmt, ap);
2715 len = sbuf_len(sb) + 1;
2716 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2717 SLIST_INSERT_HEAD(&ma->list, maa, next);
2718 bcopy(sbuf_data(sb), maa + 1, len);
2721 ma->v[ma->len].iov_base = maa + 1;
2722 ma->v[ma->len].iov_len = len;
2729 * Add an argument which is a userland string.
2732 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2734 struct mntaarg *maa;
2740 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2741 SLIST_INIT(&ma->list);
2745 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2746 SLIST_INSERT_HEAD(&ma->list, maa, next);
2747 tbuf = (void *)(maa + 1);
2748 ma->error = copyinstr(val, tbuf, len, NULL);
2749 return (mount_arg(ma, name, tbuf, -1));
2755 * If length is -1, treat value as a C string.
2758 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2762 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2763 SLIST_INIT(&ma->list);
2768 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2770 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2771 ma->v[ma->len].iov_len = strlen(name) + 1;
2774 ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2776 ma->v[ma->len].iov_len = strlen(val) + 1;
2778 ma->v[ma->len].iov_len = len;
2784 * Free a mntarg structure
2787 free_mntarg(struct mntarg *ma)
2789 struct mntaarg *maa;
2791 while (!SLIST_EMPTY(&ma->list)) {
2792 maa = SLIST_FIRST(&ma->list);
2793 SLIST_REMOVE_HEAD(&ma->list, next);
2796 free(ma->v, M_MOUNT);
2801 * Mount a filesystem
2804 kernel_mount(struct mntarg *ma, uint64_t flags)
2809 KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2810 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
2811 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2813 auio.uio_iov = ma->v;
2814 auio.uio_iovcnt = ma->len;
2815 auio.uio_segflg = UIO_SYSSPACE;
2819 error = vfs_donmount(curthread, flags, &auio);
2825 * A printflike function to mount a filesystem.
2828 kernel_vmount(int flags, ...)
2830 struct mntarg *ma = NULL;
2836 va_start(ap, flags);
2838 cp = va_arg(ap, const char *);
2841 vp = va_arg(ap, const void *);
2842 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0));
2846 error = kernel_mount(ma, flags);
2850 /* Map from mount options to printable formats. */
2851 static struct mntoptnames optnames[] = {
2855 #define DEVCTL_LEN 1024
2857 mount_devctl_event(const char *type, struct mount *mp, bool donew)
2860 struct mntoptnames *fp;
2862 struct statfs *sfp = &mp->mnt_stat;
2865 buf = malloc(DEVCTL_LEN, M_MOUNT, M_NOWAIT);
2868 sbuf_new(&sb, buf, DEVCTL_LEN, SBUF_FIXEDLEN);
2869 sbuf_cpy(&sb, "mount-point=\"");
2870 devctl_safe_quote_sb(&sb, sfp->f_mntonname);
2871 sbuf_cat(&sb, "\" mount-dev=\"");
2872 devctl_safe_quote_sb(&sb, sfp->f_mntfromname);
2873 sbuf_cat(&sb, "\" mount-type=\"");
2874 devctl_safe_quote_sb(&sb, sfp->f_fstypename);
2875 sbuf_cat(&sb, "\" fsid=0x");
2876 cp = (const uint8_t *)&sfp->f_fsid.val[0];
2877 for (int i = 0; i < sizeof(sfp->f_fsid); i++)
2878 sbuf_printf(&sb, "%02x", cp[i]);
2879 sbuf_printf(&sb, " owner=%u flags=\"", sfp->f_owner);
2880 for (fp = optnames; fp->o_opt != 0; fp++) {
2881 if ((mp->mnt_flag & fp->o_opt) != 0) {
2882 sbuf_cat(&sb, fp->o_name);
2883 sbuf_putc(&sb, ';');
2886 sbuf_putc(&sb, '"');
2890 * Options are not published because the form of the options depends on
2891 * the file system and may include binary data. In addition, they don't
2892 * necessarily provide enough useful information to be actionable when
2893 * devd processes them.
2896 if (sbuf_error(&sb) == 0)
2897 devctl_notify("VFS", "FS", type, sbuf_data(&sb));
2903 * Suspend write operations on all local writeable filesystems. Does
2904 * full sync of them in the process.
2906 * Iterate over the mount points in reverse order, suspending most
2907 * recently mounted filesystems first. It handles a case where a
2908 * filesystem mounted from a md(4) vnode-backed device should be
2909 * suspended before the filesystem that owns the vnode.
2912 suspend_all_fs(void)
2917 mtx_lock(&mountlist_mtx);
2918 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
2919 error = vfs_busy(mp, MBF_MNTLSTLOCK | MBF_NOWAIT);
2922 if ((mp->mnt_flag & (MNT_RDONLY | MNT_LOCAL)) != MNT_LOCAL ||
2923 (mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
2924 mtx_lock(&mountlist_mtx);
2928 error = vfs_write_suspend(mp, 0);
2931 MPASS((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0);
2932 mp->mnt_kern_flag |= MNTK_SUSPEND_ALL;
2934 mtx_lock(&mountlist_mtx);
2936 printf("suspend of %s failed, error %d\n",
2937 mp->mnt_stat.f_mntonname, error);
2938 mtx_lock(&mountlist_mtx);
2942 mtx_unlock(&mountlist_mtx);
2950 mtx_lock(&mountlist_mtx);
2951 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2952 if ((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0)
2954 mtx_unlock(&mountlist_mtx);
2956 MPASS((mp->mnt_kern_flag & MNTK_SUSPEND) != 0);
2957 mp->mnt_kern_flag &= ~MNTK_SUSPEND_ALL;
2959 vfs_write_resume(mp, 0);
2960 mtx_lock(&mountlist_mtx);
2963 mtx_unlock(&mountlist_mtx);