2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1999-2004 Poul-Henning Kamp
5 * Copyright (c) 1999 Michael Smith
6 * Copyright (c) 1989, 1993
7 * The Regents of the University of California. All rights reserved.
8 * (c) UNIX System Laboratories, Inc.
9 * All or some portions of this file are derived from material licensed
10 * to the University of California by American Telephone and Telegraph
11 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
12 * the permission of UNIX System Laboratories, Inc.
14 * Redistribution and use in source and binary forms, with or without
15 * modification, are permitted provided that the following conditions
17 * 1. Redistributions of source code must retain the above copyright
18 * notice, this list of conditions and the following disclaimer.
19 * 2. Redistributions in binary form must reproduce the above copyright
20 * notice, this list of conditions and the following disclaimer in the
21 * documentation and/or other materials provided with the distribution.
22 * 3. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
42 #include <sys/param.h>
45 #include <sys/devctl.h>
46 #include <sys/eventhandler.h>
47 #include <sys/fcntl.h>
49 #include <sys/kernel.h>
51 #include <sys/libkern.h>
52 #include <sys/malloc.h>
53 #include <sys/mount.h>
54 #include <sys/mutex.h>
55 #include <sys/namei.h>
58 #include <sys/filedesc.h>
59 #include <sys/reboot.h>
61 #include <sys/syscallsubr.h>
62 #include <sys/sysproto.h>
64 #include <sys/sysctl.h>
65 #include <sys/sysent.h>
66 #include <sys/systm.h>
67 #include <sys/vnode.h>
70 #include <geom/geom.h>
72 #include <machine/stdarg.h>
74 #include <security/audit/audit.h>
75 #include <security/mac/mac_framework.h>
77 #define VFS_MOUNTARG_SIZE_MAX (1024 * 64)
79 static int vfs_domount(struct thread *td, const char *fstype, char *fspath,
80 uint64_t fsflags, struct vfsoptlist **optlist);
81 static void free_mntarg(struct mntarg *ma);
83 static int usermount = 0;
84 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
85 "Unprivileged users may mount and unmount file systems");
87 static bool default_autoro = false;
88 SYSCTL_BOOL(_vfs, OID_AUTO, default_autoro, CTLFLAG_RW, &default_autoro, 0,
89 "Retry failed r/w mount as r/o if no explicit ro/rw option is specified");
91 MALLOC_DEFINE(M_MOUNT, "mount", "vfs mount structure");
92 MALLOC_DEFINE(M_STATFS, "statfs", "statfs structure");
93 static uma_zone_t mount_zone;
95 /* List of mounted filesystems. */
96 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
98 /* For any iteration/modification of mountlist */
99 struct mtx_padalign __exclusive_cache_line mountlist_mtx;
100 MTX_SYSINIT(mountlist, &mountlist_mtx, "mountlist", MTX_DEF);
102 EVENTHANDLER_LIST_DEFINE(vfs_mounted);
103 EVENTHANDLER_LIST_DEFINE(vfs_unmounted);
105 static void mount_devctl_event(const char *type, struct mount *mp, bool donew);
108 * Global opts, taken by all filesystems
110 static const char *global_opts[] = {
122 mount_init(void *mem, int size, int flags)
126 mp = (struct mount *)mem;
127 mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
128 mtx_init(&mp->mnt_listmtx, "struct mount vlist mtx", NULL, MTX_DEF);
129 lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0);
130 mp->mnt_pcpu = uma_zalloc_pcpu(pcpu_zone_16, M_WAITOK | M_ZERO);
133 mp->mnt_rootvnode = NULL;
138 mount_fini(void *mem, int size)
142 mp = (struct mount *)mem;
143 uma_zfree_pcpu(pcpu_zone_16, mp->mnt_pcpu);
144 lockdestroy(&mp->mnt_explock);
145 mtx_destroy(&mp->mnt_listmtx);
146 mtx_destroy(&mp->mnt_mtx);
150 vfs_mount_init(void *dummy __unused)
153 mount_zone = uma_zcreate("Mountpoints", sizeof(struct mount), NULL,
154 NULL, mount_init, mount_fini, UMA_ALIGN_CACHE, UMA_ZONE_NOFREE);
156 SYSINIT(vfs_mount, SI_SUB_VFS, SI_ORDER_ANY, vfs_mount_init, NULL);
159 * ---------------------------------------------------------------------
160 * Functions for building and sanitizing the mount options
163 /* Remove one mount option. */
165 vfs_freeopt(struct vfsoptlist *opts, struct vfsopt *opt)
168 TAILQ_REMOVE(opts, opt, link);
169 free(opt->name, M_MOUNT);
170 if (opt->value != NULL)
171 free(opt->value, M_MOUNT);
175 /* Release all resources related to the mount options. */
177 vfs_freeopts(struct vfsoptlist *opts)
181 while (!TAILQ_EMPTY(opts)) {
182 opt = TAILQ_FIRST(opts);
183 vfs_freeopt(opts, opt);
189 vfs_deleteopt(struct vfsoptlist *opts, const char *name)
191 struct vfsopt *opt, *temp;
195 TAILQ_FOREACH_SAFE(opt, opts, link, temp) {
196 if (strcmp(opt->name, name) == 0)
197 vfs_freeopt(opts, opt);
202 vfs_isopt_ro(const char *opt)
205 if (strcmp(opt, "ro") == 0 || strcmp(opt, "rdonly") == 0 ||
206 strcmp(opt, "norw") == 0)
212 vfs_isopt_rw(const char *opt)
215 if (strcmp(opt, "rw") == 0 || strcmp(opt, "noro") == 0)
221 * Check if options are equal (with or without the "no" prefix).
224 vfs_equalopts(const char *opt1, const char *opt2)
228 /* "opt" vs. "opt" or "noopt" vs. "noopt" */
229 if (strcmp(opt1, opt2) == 0)
231 /* "noopt" vs. "opt" */
232 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
234 /* "opt" vs. "noopt" */
235 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
237 while ((p = strchr(opt1, '.')) != NULL &&
238 !strncmp(opt1, opt2, ++p - opt1)) {
241 /* "foo.noopt" vs. "foo.opt" */
242 if (strncmp(opt1, "no", 2) == 0 && strcmp(opt1 + 2, opt2) == 0)
244 /* "foo.opt" vs. "foo.noopt" */
245 if (strncmp(opt2, "no", 2) == 0 && strcmp(opt1, opt2 + 2) == 0)
248 /* "ro" / "rdonly" / "norw" / "rw" / "noro" */
249 if ((vfs_isopt_ro(opt1) || vfs_isopt_rw(opt1)) &&
250 (vfs_isopt_ro(opt2) || vfs_isopt_rw(opt2)))
256 * If a mount option is specified several times,
257 * (with or without the "no" prefix) only keep
258 * the last occurrence of it.
261 vfs_sanitizeopts(struct vfsoptlist *opts)
263 struct vfsopt *opt, *opt2, *tmp;
265 TAILQ_FOREACH_REVERSE(opt, opts, vfsoptlist, link) {
266 opt2 = TAILQ_PREV(opt, vfsoptlist, link);
267 while (opt2 != NULL) {
268 if (vfs_equalopts(opt->name, opt2->name)) {
269 tmp = TAILQ_PREV(opt2, vfsoptlist, link);
270 vfs_freeopt(opts, opt2);
273 opt2 = TAILQ_PREV(opt2, vfsoptlist, link);
280 * Build a linked list of mount options from a struct uio.
283 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
285 struct vfsoptlist *opts;
287 size_t memused, namelen, optlen;
288 unsigned int i, iovcnt;
291 opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
294 iovcnt = auio->uio_iovcnt;
295 for (i = 0; i < iovcnt; i += 2) {
296 namelen = auio->uio_iov[i].iov_len;
297 optlen = auio->uio_iov[i + 1].iov_len;
298 memused += sizeof(struct vfsopt) + optlen + namelen;
300 * Avoid consuming too much memory, and attempts to overflow
303 if (memused > VFS_MOUNTARG_SIZE_MAX ||
304 optlen > VFS_MOUNTARG_SIZE_MAX ||
305 namelen > VFS_MOUNTARG_SIZE_MAX) {
310 opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
311 opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
318 * Do this early, so jumps to "bad" will free the current
321 TAILQ_INSERT_TAIL(opts, opt, link);
323 if (auio->uio_segflg == UIO_SYSSPACE) {
324 bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
326 error = copyin(auio->uio_iov[i].iov_base, opt->name,
331 /* Ensure names are null-terminated strings. */
332 if (namelen == 0 || opt->name[namelen - 1] != '\0') {
338 opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
339 if (auio->uio_segflg == UIO_SYSSPACE) {
340 bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
343 error = copyin(auio->uio_iov[i + 1].iov_base,
350 vfs_sanitizeopts(opts);
359 * Merge the old mount options with the new ones passed
360 * in the MNT_UPDATE case.
362 * XXX: This function will keep a "nofoo" option in the new
363 * options. E.g, if the option's canonical name is "foo",
364 * "nofoo" ends up in the mount point's active options.
367 vfs_mergeopts(struct vfsoptlist *toopts, struct vfsoptlist *oldopts)
369 struct vfsopt *opt, *new;
371 TAILQ_FOREACH(opt, oldopts, link) {
372 new = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
373 new->name = strdup(opt->name, M_MOUNT);
375 new->value = malloc(opt->len, M_MOUNT, M_WAITOK);
376 bcopy(opt->value, new->value, opt->len);
380 new->seen = opt->seen;
381 TAILQ_INSERT_HEAD(toopts, new, link);
383 vfs_sanitizeopts(toopts);
387 * Mount a filesystem.
389 #ifndef _SYS_SYSPROTO_H_
397 sys_nmount(struct thread *td, struct nmount_args *uap)
405 * Mount flags are now 64-bits. On 32-bit archtectures only
406 * 32-bits are passed in, but from here on everything handles
407 * 64-bit flags correctly.
411 AUDIT_ARG_FFLAGS(flags);
412 CTR4(KTR_VFS, "%s: iovp %p with iovcnt %d and flags %d", __func__,
413 uap->iovp, uap->iovcnt, flags);
416 * Filter out MNT_ROOTFS. We do not want clients of nmount() in
417 * userspace to set this flag, but we must filter it out if we want
418 * MNT_UPDATE on the root file system to work.
419 * MNT_ROOTFS should only be set by the kernel when mounting its
422 flags &= ~MNT_ROOTFS;
424 iovcnt = uap->iovcnt;
426 * Check that we have an even number of iovec's
427 * and that we have at least two options.
429 if ((iovcnt & 1) || (iovcnt < 4)) {
430 CTR2(KTR_VFS, "%s: failed for invalid iovcnt %d", __func__,
435 error = copyinuio(uap->iovp, iovcnt, &auio);
437 CTR2(KTR_VFS, "%s: failed for invalid uio op with %d errno",
441 error = vfs_donmount(td, flags, auio);
448 * ---------------------------------------------------------------------
449 * Various utility functions
453 * Get a reference on a mount point from a vnode.
455 * The vnode is allowed to be passed unlocked and race against dooming. Note in
456 * such case there are no guarantees the referenced mount point will still be
457 * associated with it after the function returns.
460 vfs_ref_from_vp(struct vnode *vp)
463 struct mount_pcpu *mpcpu;
465 mp = atomic_load_ptr(&vp->v_mount);
466 if (__predict_false(mp == NULL)) {
469 if (vfs_op_thread_enter(mp, mpcpu)) {
470 if (__predict_true(mp == vp->v_mount)) {
471 vfs_mp_count_add_pcpu(mpcpu, ref, 1);
472 vfs_op_thread_exit(mp, mpcpu);
474 vfs_op_thread_exit(mp, mpcpu);
479 if (mp == vp->v_mount) {
491 vfs_ref(struct mount *mp)
493 struct mount_pcpu *mpcpu;
495 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
496 if (vfs_op_thread_enter(mp, mpcpu)) {
497 vfs_mp_count_add_pcpu(mpcpu, ref, 1);
498 vfs_op_thread_exit(mp, mpcpu);
508 vfs_rel(struct mount *mp)
510 struct mount_pcpu *mpcpu;
512 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
513 if (vfs_op_thread_enter(mp, mpcpu)) {
514 vfs_mp_count_sub_pcpu(mpcpu, ref, 1);
515 vfs_op_thread_exit(mp, mpcpu);
525 * Allocate and initialize the mount point struct.
528 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp, const char *fspath,
533 mp = uma_zalloc(mount_zone, M_WAITOK);
534 bzero(&mp->mnt_startzero,
535 __rangeof(struct mount, mnt_startzero, mnt_endzero));
536 mp->mnt_kern_flag = 0;
538 mp->mnt_rootvnode = NULL;
539 mp->mnt_vnodecovered = NULL;
542 TAILQ_INIT(&mp->mnt_nvnodelist);
543 mp->mnt_nvnodelistsize = 0;
544 TAILQ_INIT(&mp->mnt_lazyvnodelist);
545 mp->mnt_lazyvnodelistsize = 0;
546 if (mp->mnt_ref != 0 || mp->mnt_lockref != 0 ||
547 mp->mnt_writeopcount != 0)
548 panic("%s: non-zero counters on new mp %p\n", __func__, mp);
549 if (mp->mnt_vfs_ops != 1)
550 panic("%s: vfs_ops should be 1 but %d found\n", __func__,
552 (void) vfs_busy(mp, MBF_NOWAIT);
553 atomic_add_acq_int(&vfsp->vfc_refcount, 1);
554 mp->mnt_op = vfsp->vfc_vfsops;
556 mp->mnt_stat.f_type = vfsp->vfc_typenum;
558 strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
559 mp->mnt_vnodecovered = vp;
560 mp->mnt_cred = crdup(cred);
561 mp->mnt_stat.f_owner = cred->cr_uid;
562 strlcpy(mp->mnt_stat.f_mntonname, fspath, MNAMELEN);
563 mp->mnt_iosize_max = DFLTPHYS;
566 mac_mount_create(cred, mp);
568 arc4rand(&mp->mnt_hashseed, sizeof mp->mnt_hashseed, 0);
569 TAILQ_INIT(&mp->mnt_uppers);
574 * Destroy the mount struct previously allocated by vfs_mount_alloc().
577 vfs_mount_destroy(struct mount *mp)
580 if (mp->mnt_vfs_ops == 0)
581 panic("%s: entered with zero vfs_ops\n", __func__);
583 vfs_assert_mount_counters(mp);
586 mp->mnt_kern_flag |= MNTK_REFEXPIRE;
587 if (mp->mnt_kern_flag & MNTK_MWAIT) {
588 mp->mnt_kern_flag &= ~MNTK_MWAIT;
592 msleep(mp, MNT_MTX(mp), PVFS, "mntref", 0);
593 KASSERT(mp->mnt_ref == 0,
594 ("%s: invalid refcount in the drain path @ %s:%d", __func__,
595 __FILE__, __LINE__));
596 if (mp->mnt_writeopcount != 0)
597 panic("vfs_mount_destroy: nonzero writeopcount");
598 if (mp->mnt_secondary_writes != 0)
599 panic("vfs_mount_destroy: nonzero secondary_writes");
600 atomic_subtract_rel_int(&mp->mnt_vfc->vfc_refcount, 1);
601 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) {
604 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes)
605 vn_printf(vp, "dangling vnode ");
606 panic("unmount: dangling vnode");
608 KASSERT(TAILQ_EMPTY(&mp->mnt_uppers), ("mnt_uppers"));
609 if (mp->mnt_nvnodelistsize != 0)
610 panic("vfs_mount_destroy: nonzero nvnodelistsize");
611 if (mp->mnt_lazyvnodelistsize != 0)
612 panic("vfs_mount_destroy: nonzero lazyvnodelistsize");
613 if (mp->mnt_lockref != 0)
614 panic("vfs_mount_destroy: nonzero lock refcount");
617 if (mp->mnt_vfs_ops != 1)
618 panic("%s: vfs_ops should be 1 but %d found\n", __func__,
621 if (mp->mnt_rootvnode != NULL)
622 panic("%s: mount point still has a root vnode %p\n", __func__,
625 if (mp->mnt_vnodecovered != NULL)
626 vrele(mp->mnt_vnodecovered);
628 mac_mount_destroy(mp);
630 if (mp->mnt_opt != NULL)
631 vfs_freeopts(mp->mnt_opt);
632 crfree(mp->mnt_cred);
633 uma_zfree(mount_zone, mp);
637 vfs_should_downgrade_to_ro_mount(uint64_t fsflags, int error)
639 /* This is an upgrade of an exisiting mount. */
640 if ((fsflags & MNT_UPDATE) != 0)
642 /* This is already an R/O mount. */
643 if ((fsflags & MNT_RDONLY) != 0)
647 case ENODEV: /* generic, geom, ... */
648 case EACCES: /* cam/scsi, ... */
649 case EROFS: /* md, mmcsd, ... */
651 * These errors can be returned by the storage layer to signal
652 * that the media is read-only. No harm in the R/O mount
653 * attempt if the error was returned for some other reason.
662 vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions)
664 struct vfsoptlist *optlist;
665 struct vfsopt *opt, *tmp_opt;
666 char *fstype, *fspath, *errmsg;
667 int error, fstypelen, fspathlen, errmsg_len, errmsg_pos;
670 errmsg = fspath = NULL;
671 errmsg_len = fspathlen = 0;
673 autoro = default_autoro;
675 error = vfs_buildopts(fsoptions, &optlist);
679 if (vfs_getopt(optlist, "errmsg", (void **)&errmsg, &errmsg_len) == 0)
680 errmsg_pos = vfs_getopt_pos(optlist, "errmsg");
683 * We need these two options before the others,
684 * and they are mandatory for any filesystem.
685 * Ensure they are NUL terminated as well.
688 error = vfs_getopt(optlist, "fstype", (void **)&fstype, &fstypelen);
689 if (error || fstypelen <= 0 || fstype[fstypelen - 1] != '\0') {
692 strncpy(errmsg, "Invalid fstype", errmsg_len);
696 error = vfs_getopt(optlist, "fspath", (void **)&fspath, &fspathlen);
697 if (error || fspathlen <= 0 || fspath[fspathlen - 1] != '\0') {
700 strncpy(errmsg, "Invalid fspath", errmsg_len);
705 * We need to see if we have the "update" option
706 * before we call vfs_domount(), since vfs_domount() has special
707 * logic based on MNT_UPDATE. This is very important
708 * when we want to update the root filesystem.
710 TAILQ_FOREACH_SAFE(opt, optlist, link, tmp_opt) {
713 if (strcmp(opt->name, "update") == 0) {
714 fsflags |= MNT_UPDATE;
717 else if (strcmp(opt->name, "async") == 0)
718 fsflags |= MNT_ASYNC;
719 else if (strcmp(opt->name, "force") == 0) {
720 fsflags |= MNT_FORCE;
723 else if (strcmp(opt->name, "reload") == 0) {
724 fsflags |= MNT_RELOAD;
727 else if (strcmp(opt->name, "multilabel") == 0)
728 fsflags |= MNT_MULTILABEL;
729 else if (strcmp(opt->name, "noasync") == 0)
730 fsflags &= ~MNT_ASYNC;
731 else if (strcmp(opt->name, "noatime") == 0)
732 fsflags |= MNT_NOATIME;
733 else if (strcmp(opt->name, "atime") == 0) {
734 free(opt->name, M_MOUNT);
735 opt->name = strdup("nonoatime", M_MOUNT);
737 else if (strcmp(opt->name, "noclusterr") == 0)
738 fsflags |= MNT_NOCLUSTERR;
739 else if (strcmp(opt->name, "clusterr") == 0) {
740 free(opt->name, M_MOUNT);
741 opt->name = strdup("nonoclusterr", M_MOUNT);
743 else if (strcmp(opt->name, "noclusterw") == 0)
744 fsflags |= MNT_NOCLUSTERW;
745 else if (strcmp(opt->name, "clusterw") == 0) {
746 free(opt->name, M_MOUNT);
747 opt->name = strdup("nonoclusterw", M_MOUNT);
749 else if (strcmp(opt->name, "noexec") == 0)
750 fsflags |= MNT_NOEXEC;
751 else if (strcmp(opt->name, "exec") == 0) {
752 free(opt->name, M_MOUNT);
753 opt->name = strdup("nonoexec", M_MOUNT);
755 else if (strcmp(opt->name, "nosuid") == 0)
756 fsflags |= MNT_NOSUID;
757 else if (strcmp(opt->name, "suid") == 0) {
758 free(opt->name, M_MOUNT);
759 opt->name = strdup("nonosuid", M_MOUNT);
761 else if (strcmp(opt->name, "nosymfollow") == 0)
762 fsflags |= MNT_NOSYMFOLLOW;
763 else if (strcmp(opt->name, "symfollow") == 0) {
764 free(opt->name, M_MOUNT);
765 opt->name = strdup("nonosymfollow", M_MOUNT);
767 else if (strcmp(opt->name, "noro") == 0) {
768 fsflags &= ~MNT_RDONLY;
771 else if (strcmp(opt->name, "rw") == 0) {
772 fsflags &= ~MNT_RDONLY;
775 else if (strcmp(opt->name, "ro") == 0) {
776 fsflags |= MNT_RDONLY;
779 else if (strcmp(opt->name, "rdonly") == 0) {
780 free(opt->name, M_MOUNT);
781 opt->name = strdup("ro", M_MOUNT);
782 fsflags |= MNT_RDONLY;
785 else if (strcmp(opt->name, "autoro") == 0) {
789 else if (strcmp(opt->name, "suiddir") == 0)
790 fsflags |= MNT_SUIDDIR;
791 else if (strcmp(opt->name, "sync") == 0)
792 fsflags |= MNT_SYNCHRONOUS;
793 else if (strcmp(opt->name, "union") == 0)
794 fsflags |= MNT_UNION;
795 else if (strcmp(opt->name, "automounted") == 0) {
796 fsflags |= MNT_AUTOMOUNTED;
798 } else if (strcmp(opt->name, "nocover") == 0) {
799 fsflags |= MNT_NOCOVER;
801 } else if (strcmp(opt->name, "cover") == 0) {
802 fsflags &= ~MNT_NOCOVER;
804 } else if (strcmp(opt->name, "emptydir") == 0) {
805 fsflags |= MNT_EMPTYDIR;
807 } else if (strcmp(opt->name, "noemptydir") == 0) {
808 fsflags &= ~MNT_EMPTYDIR;
812 vfs_freeopt(optlist, opt);
816 * Be ultra-paranoid about making sure the type and fspath
817 * variables will fit in our mp buffers, including the
820 if (fstypelen > MFSNAMELEN || fspathlen > MNAMELEN) {
821 error = ENAMETOOLONG;
825 error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
828 * See if we can mount in the read-only mode if the error code suggests
829 * that it could be possible and the mount options allow for that.
830 * Never try it if "[no]{ro|rw}" has been explicitly requested and not
831 * overridden by "autoro".
833 if (autoro && vfs_should_downgrade_to_ro_mount(fsflags, error)) {
834 printf("%s: R/W mount failed, possibly R/O media,"
835 " trying R/O mount\n", __func__);
836 fsflags |= MNT_RDONLY;
837 error = vfs_domount(td, fstype, fspath, fsflags, &optlist);
840 /* copyout the errmsg */
841 if (errmsg_pos != -1 && ((2 * errmsg_pos + 1) < fsoptions->uio_iovcnt)
842 && errmsg_len > 0 && errmsg != NULL) {
843 if (fsoptions->uio_segflg == UIO_SYSSPACE) {
845 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
846 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
849 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_base,
850 fsoptions->uio_iov[2 * errmsg_pos + 1].iov_len);
855 vfs_freeopts(optlist);
862 #ifndef _SYS_SYSPROTO_H_
872 sys_mount(struct thread *td, struct mount_args *uap)
875 struct vfsconf *vfsp = NULL;
876 struct mntarg *ma = NULL;
881 * Mount flags are now 64-bits. On 32-bit architectures only
882 * 32-bits are passed in, but from here on everything handles
883 * 64-bit flags correctly.
887 AUDIT_ARG_FFLAGS(flags);
890 * Filter out MNT_ROOTFS. We do not want clients of mount() in
891 * userspace to set this flag, but we must filter it out if we want
892 * MNT_UPDATE on the root file system to work.
893 * MNT_ROOTFS should only be set by the kernel when mounting its
896 flags &= ~MNT_ROOTFS;
898 fstype = malloc(MFSNAMELEN, M_TEMP, M_WAITOK);
899 error = copyinstr(uap->type, fstype, MFSNAMELEN, NULL);
901 free(fstype, M_TEMP);
905 AUDIT_ARG_TEXT(fstype);
906 vfsp = vfs_byname_kld(fstype, td, &error);
907 free(fstype, M_TEMP);
910 if (((vfsp->vfc_flags & VFCF_SBDRY) != 0 &&
911 vfsp->vfc_vfsops_sd->vfs_cmount == NULL) ||
912 ((vfsp->vfc_flags & VFCF_SBDRY) == 0 &&
913 vfsp->vfc_vfsops->vfs_cmount == NULL))
916 ma = mount_argsu(ma, "fstype", uap->type, MFSNAMELEN);
917 ma = mount_argsu(ma, "fspath", uap->path, MNAMELEN);
918 ma = mount_argb(ma, flags & MNT_RDONLY, "noro");
919 ma = mount_argb(ma, !(flags & MNT_NOSUID), "nosuid");
920 ma = mount_argb(ma, !(flags & MNT_NOEXEC), "noexec");
922 if ((vfsp->vfc_flags & VFCF_SBDRY) != 0)
923 return (vfsp->vfc_vfsops_sd->vfs_cmount(ma, uap->data, flags));
924 return (vfsp->vfc_vfsops->vfs_cmount(ma, uap->data, flags));
928 * vfs_domount_first(): first file system mount (not update)
932 struct thread *td, /* Calling thread. */
933 struct vfsconf *vfsp, /* File system type. */
934 char *fspath, /* Mount path. */
935 struct vnode *vp, /* Vnode to be covered. */
936 uint64_t fsflags, /* Flags common to all filesystems. */
937 struct vfsoptlist **optlist /* Options local to the filesystem. */
942 struct vnode *newdp, *rootvp;
946 ASSERT_VOP_ELOCKED(vp, __func__);
947 KASSERT((fsflags & MNT_UPDATE) == 0, ("MNT_UPDATE shouldn't be here"));
950 * If the jail of the calling thread lacks permission for this type of
951 * file system, or is trying to cover its own root, deny immediately.
953 if (jailed(td->td_ucred) && (!prison_allow(td->td_ucred,
954 vfsp->vfc_prison_flag) || vp == td->td_ucred->cr_prison->pr_root)) {
960 * If the user is not root, ensure that they own the directory
961 * onto which we are attempting to mount.
963 error = VOP_GETATTR(vp, &va, td->td_ucred);
964 if (error == 0 && va.va_uid != td->td_ucred->cr_uid)
965 error = priv_check_cred(td->td_ucred, PRIV_VFS_ADMIN);
967 error = vinvalbuf(vp, V_SAVE, 0, 0);
968 if (error == 0 && vp->v_type != VDIR)
970 if (error == 0 && (fsflags & MNT_EMPTYDIR) != 0)
971 error = vfs_emptydir(vp);
974 if ((vp->v_iflag & VI_MOUNT) == 0 && vp->v_mountedhere == NULL)
975 vp->v_iflag |= VI_MOUNT;
984 vn_seqc_write_begin(vp);
987 /* Allocate and initialize the filesystem. */
988 mp = vfs_mount_alloc(vp, vfsp, fspath, td->td_ucred);
989 /* XXXMAC: pass to vfs_mount_alloc? */
990 mp->mnt_optnew = *optlist;
991 /* Set the mount level flags. */
992 mp->mnt_flag = (fsflags & (MNT_UPDATEMASK | MNT_ROOTFS | MNT_RDONLY));
995 * Mount the filesystem.
996 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
997 * get. No freeing of cn_pnbuf.
1001 if ((error = VFS_MOUNT(mp)) != 0 ||
1002 (error1 = VFS_STATFS(mp, &mp->mnt_stat)) != 0 ||
1003 (error1 = VFS_ROOT(mp, LK_EXCLUSIVE, &newdp)) != 0) {
1007 rootvp = vfs_cache_root_clear(mp);
1008 if (rootvp != NULL) {
1012 (void)vn_start_write(NULL, &mp, V_WAIT);
1014 mp->mnt_kern_flag |= MNTK_UNMOUNT | MNTK_UNMOUNTF;
1017 error = VFS_UNMOUNT(mp, 0);
1018 vn_finished_write(mp);
1021 "failed post-mount (%d): rollback unmount returned %d\n",
1028 mp->mnt_vnodecovered = NULL;
1030 /* XXXKIB wait for mnt_lockref drain? */
1031 vfs_mount_destroy(mp);
1034 vp->v_iflag &= ~VI_MOUNT;
1036 if (rootvp != NULL) {
1037 vn_seqc_write_end(rootvp);
1040 vn_seqc_write_end(vp);
1044 vn_seqc_write_begin(newdp);
1047 if (mp->mnt_opt != NULL)
1048 vfs_freeopts(mp->mnt_opt);
1049 mp->mnt_opt = mp->mnt_optnew;
1053 * Prevent external consumers of mount options from reading mnt_optnew.
1055 mp->mnt_optnew = NULL;
1058 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1059 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1060 mp->mnt_kern_flag |= MNTK_ASYNC;
1062 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1065 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1068 vp->v_iflag &= ~VI_MOUNT;
1069 vn_irflag_set_locked(vp, VIRF_MOUNTPOINT);
1070 vp->v_mountedhere = mp;
1072 /* Place the new filesystem at the end of the mount list. */
1073 mtx_lock(&mountlist_mtx);
1074 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
1075 mtx_unlock(&mountlist_mtx);
1076 vfs_event_signal(NULL, VQ_MOUNT, 0);
1077 vn_lock(newdp, LK_EXCLUSIVE | LK_RETRY);
1079 EVENTHANDLER_DIRECT_INVOKE(vfs_mounted, mp, newdp, td);
1081 mount_devctl_event("MOUNT", mp, false);
1082 mountcheckdirs(vp, newdp);
1083 vn_seqc_write_end(vp);
1084 vn_seqc_write_end(newdp);
1086 if ((mp->mnt_flag & MNT_RDONLY) == 0)
1087 vfs_allocate_syncvnode(mp);
1094 * vfs_domount_update(): update of mounted file system
1098 struct thread *td, /* Calling thread. */
1099 struct vnode *vp, /* Mount point vnode. */
1100 uint64_t fsflags, /* Flags common to all filesystems. */
1101 struct vfsoptlist **optlist /* Options local to the filesystem. */
1104 struct export_args export;
1105 struct o2export_args o2export;
1106 struct vnode *rootvp;
1109 int error, export_error, i, len;
1113 ASSERT_VOP_ELOCKED(vp, __func__);
1114 KASSERT((fsflags & MNT_UPDATE) != 0, ("MNT_UPDATE should be here"));
1117 if ((vp->v_vflag & VV_ROOT) == 0) {
1118 if (vfs_copyopt(*optlist, "export", &export, sizeof(export))
1128 * We only allow the filesystem to be reloaded if it
1129 * is currently mounted read-only.
1131 flag = mp->mnt_flag;
1132 if ((fsflags & MNT_RELOAD) != 0 && (flag & MNT_RDONLY) == 0) {
1134 return (EOPNOTSUPP); /* Needs translation */
1137 * Only privileged root, or (if MNT_USER is set) the user that
1138 * did the original mount is permitted to update it.
1140 error = vfs_suser(mp, td);
1145 if (vfs_busy(mp, MBF_NOWAIT)) {
1150 if ((vp->v_iflag & VI_MOUNT) != 0 || vp->v_mountedhere != NULL) {
1156 vp->v_iflag |= VI_MOUNT;
1161 vn_seqc_write_begin(vp);
1165 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1170 mp->mnt_flag &= ~MNT_UPDATEMASK;
1171 mp->mnt_flag |= fsflags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE |
1172 MNT_SNAPSHOT | MNT_ROOTFS | MNT_UPDATEMASK | MNT_RDONLY);
1173 if ((mp->mnt_flag & MNT_ASYNC) == 0)
1174 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1175 rootvp = vfs_cache_root_clear(mp);
1177 mp->mnt_optnew = *optlist;
1178 vfs_mergeopts(mp->mnt_optnew, mp->mnt_opt);
1181 * Mount the filesystem.
1182 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
1183 * get. No freeing of cn_pnbuf.
1185 error = VFS_MOUNT(mp);
1188 /* Process the export option. */
1189 if (error == 0 && vfs_getopt(mp->mnt_optnew, "export", &bufp,
1191 /* Assume that there is only 1 ABI for each length. */
1193 case (sizeof(struct oexport_args)):
1194 bzero(&o2export, sizeof(o2export));
1196 case (sizeof(o2export)):
1197 bcopy(bufp, &o2export, len);
1198 export.ex_flags = (uint64_t)o2export.ex_flags;
1199 export.ex_root = o2export.ex_root;
1200 export.ex_uid = o2export.ex_anon.cr_uid;
1201 export.ex_groups = NULL;
1202 export.ex_ngroups = o2export.ex_anon.cr_ngroups;
1203 if (export.ex_ngroups > 0) {
1204 if (export.ex_ngroups <= XU_NGROUPS) {
1205 export.ex_groups = malloc(
1206 export.ex_ngroups * sizeof(gid_t),
1208 for (i = 0; i < export.ex_ngroups; i++)
1209 export.ex_groups[i] =
1210 o2export.ex_anon.cr_groups[i];
1212 export_error = EINVAL;
1213 } else if (export.ex_ngroups < 0)
1214 export_error = EINVAL;
1215 export.ex_addr = o2export.ex_addr;
1216 export.ex_addrlen = o2export.ex_addrlen;
1217 export.ex_mask = o2export.ex_mask;
1218 export.ex_masklen = o2export.ex_masklen;
1219 export.ex_indexfile = o2export.ex_indexfile;
1220 export.ex_numsecflavors = o2export.ex_numsecflavors;
1221 if (export.ex_numsecflavors < MAXSECFLAVORS) {
1222 for (i = 0; i < export.ex_numsecflavors; i++)
1223 export.ex_secflavors[i] =
1224 o2export.ex_secflavors[i];
1226 export_error = EINVAL;
1227 if (export_error == 0)
1228 export_error = vfs_export(mp, &export);
1229 free(export.ex_groups, M_TEMP);
1231 case (sizeof(export)):
1232 bcopy(bufp, &export, len);
1234 if (export.ex_ngroups > 0) {
1235 if (export.ex_ngroups <= NGROUPS_MAX) {
1236 grps = malloc(export.ex_ngroups *
1237 sizeof(gid_t), M_TEMP, M_WAITOK);
1238 export_error = copyin(export.ex_groups,
1239 grps, export.ex_ngroups *
1241 if (export_error == 0)
1242 export.ex_groups = grps;
1244 export_error = EINVAL;
1245 } else if (export.ex_ngroups == 0)
1246 export.ex_groups = NULL;
1248 export_error = EINVAL;
1249 if (export_error == 0)
1250 export_error = vfs_export(mp, &export);
1254 export_error = EINVAL;
1261 mp->mnt_flag &= ~(MNT_UPDATE | MNT_RELOAD | MNT_FORCE |
1265 * If we fail, restore old mount flags. MNT_QUOTA is special,
1266 * because it is not part of MNT_UPDATEMASK, but it could have
1267 * changed in the meantime if quotactl(2) was called.
1268 * All in all we want current value of MNT_QUOTA, not the old
1271 mp->mnt_flag = (mp->mnt_flag & MNT_QUOTA) | (flag & ~MNT_QUOTA);
1273 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1274 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1275 mp->mnt_kern_flag |= MNTK_ASYNC;
1277 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1283 mount_devctl_event("REMOUNT", mp, true);
1284 if (mp->mnt_opt != NULL)
1285 vfs_freeopts(mp->mnt_opt);
1286 mp->mnt_opt = mp->mnt_optnew;
1288 (void)VFS_STATFS(mp, &mp->mnt_stat);
1290 * Prevent external consumers of mount options from reading
1293 mp->mnt_optnew = NULL;
1295 if ((mp->mnt_flag & MNT_RDONLY) == 0)
1296 vfs_allocate_syncvnode(mp);
1298 vfs_deallocate_syncvnode(mp);
1301 if (rootvp != NULL) {
1302 vn_seqc_write_end(rootvp);
1305 vn_seqc_write_end(vp);
1308 vp->v_iflag &= ~VI_MOUNT;
1311 return (error != 0 ? error : export_error);
1315 * vfs_domount(): actually attempt a filesystem mount.
1319 struct thread *td, /* Calling thread. */
1320 const char *fstype, /* Filesystem type. */
1321 char *fspath, /* Mount path. */
1322 uint64_t fsflags, /* Flags common to all filesystems. */
1323 struct vfsoptlist **optlist /* Options local to the filesystem. */
1326 struct vfsconf *vfsp;
1327 struct nameidata nd;
1333 * Be ultra-paranoid about making sure the type and fspath
1334 * variables will fit in our mp buffers, including the
1337 if (strlen(fstype) >= MFSNAMELEN || strlen(fspath) >= MNAMELEN)
1338 return (ENAMETOOLONG);
1340 if (jailed(td->td_ucred) || usermount == 0) {
1341 if ((error = priv_check(td, PRIV_VFS_MOUNT)) != 0)
1346 * Do not allow NFS export or MNT_SUIDDIR by unprivileged users.
1348 if (fsflags & MNT_EXPORTED) {
1349 error = priv_check(td, PRIV_VFS_MOUNT_EXPORTED);
1353 if (fsflags & MNT_SUIDDIR) {
1354 error = priv_check(td, PRIV_VFS_MOUNT_SUIDDIR);
1359 * Silently enforce MNT_NOSUID and MNT_USER for unprivileged users.
1361 if ((fsflags & (MNT_NOSUID | MNT_USER)) != (MNT_NOSUID | MNT_USER)) {
1362 if (priv_check(td, PRIV_VFS_MOUNT_NONUSER) != 0)
1363 fsflags |= MNT_NOSUID | MNT_USER;
1366 /* Load KLDs before we lock the covered vnode to avoid reversals. */
1368 if ((fsflags & MNT_UPDATE) == 0) {
1369 /* Don't try to load KLDs if we're mounting the root. */
1370 if (fsflags & MNT_ROOTFS)
1371 vfsp = vfs_byname(fstype);
1373 vfsp = vfs_byname_kld(fstype, td, &error);
1379 * Get vnode to be covered or mount point's vnode in case of MNT_UPDATE.
1381 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1382 UIO_SYSSPACE, fspath, td);
1386 NDFREE(&nd, NDF_ONLY_PNBUF);
1388 if ((fsflags & MNT_UPDATE) == 0) {
1389 if ((vp->v_vflag & VV_ROOT) != 0 &&
1390 (fsflags & MNT_NOCOVER) != 0) {
1394 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1395 strcpy(pathbuf, fspath);
1396 error = vn_path_to_global_path(td, vp, pathbuf, MNAMELEN);
1398 error = vfs_domount_first(td, vfsp, pathbuf, vp,
1401 free(pathbuf, M_TEMP);
1403 error = vfs_domount_update(td, vp, fsflags, optlist);
1409 * Unmount a filesystem.
1411 * Note: unmount takes a path to the vnode mounted on as argument, not
1412 * special file (as before).
1414 #ifndef _SYS_SYSPROTO_H_
1415 struct unmount_args {
1422 sys_unmount(struct thread *td, struct unmount_args *uap)
1425 return (kern_unmount(td, uap->path, uap->flags));
1429 kern_unmount(struct thread *td, const char *path, int flags)
1431 struct nameidata nd;
1434 int error, id0, id1;
1436 AUDIT_ARG_VALUE(flags);
1437 if (jailed(td->td_ucred) || usermount == 0) {
1438 error = priv_check(td, PRIV_VFS_UNMOUNT);
1443 pathbuf = malloc(MNAMELEN, M_TEMP, M_WAITOK);
1444 error = copyinstr(path, pathbuf, MNAMELEN, NULL);
1446 free(pathbuf, M_TEMP);
1449 if (flags & MNT_BYFSID) {
1450 AUDIT_ARG_TEXT(pathbuf);
1451 /* Decode the filesystem ID. */
1452 if (sscanf(pathbuf, "FSID:%d:%d", &id0, &id1) != 2) {
1453 free(pathbuf, M_TEMP);
1457 mtx_lock(&mountlist_mtx);
1458 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1459 if (mp->mnt_stat.f_fsid.val[0] == id0 &&
1460 mp->mnt_stat.f_fsid.val[1] == id1) {
1465 mtx_unlock(&mountlist_mtx);
1468 * Try to find global path for path argument.
1470 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
1471 UIO_SYSSPACE, pathbuf, td);
1472 if (namei(&nd) == 0) {
1473 NDFREE(&nd, NDF_ONLY_PNBUF);
1474 error = vn_path_to_global_path(td, nd.ni_vp, pathbuf,
1479 mtx_lock(&mountlist_mtx);
1480 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
1481 if (strcmp(mp->mnt_stat.f_mntonname, pathbuf) == 0) {
1486 mtx_unlock(&mountlist_mtx);
1488 free(pathbuf, M_TEMP);
1491 * Previously we returned ENOENT for a nonexistent path and
1492 * EINVAL for a non-mountpoint. We cannot tell these apart
1493 * now, so in the !MNT_BYFSID case return the more likely
1494 * EINVAL for compatibility.
1496 return ((flags & MNT_BYFSID) ? ENOENT : EINVAL);
1500 * Don't allow unmounting the root filesystem.
1502 if (mp->mnt_flag & MNT_ROOTFS) {
1506 error = dounmount(mp, flags, td);
1511 * Return error if any of the vnodes, ignoring the root vnode
1512 * and the syncer vnode, have non-zero usecount.
1514 * This function is purely advisory - it can return false positives
1518 vfs_check_usecounts(struct mount *mp)
1520 struct vnode *vp, *mvp;
1522 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1523 if ((vp->v_vflag & VV_ROOT) == 0 && vp->v_type != VNON &&
1524 vp->v_usecount != 0) {
1526 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1536 dounmount_cleanup(struct mount *mp, struct vnode *coveredvp, int mntkflags)
1539 mtx_assert(MNT_MTX(mp), MA_OWNED);
1540 mp->mnt_kern_flag &= ~mntkflags;
1541 if ((mp->mnt_kern_flag & MNTK_MWAIT) != 0) {
1542 mp->mnt_kern_flag &= ~MNTK_MWAIT;
1545 vfs_op_exit_locked(mp);
1547 if (coveredvp != NULL) {
1548 VOP_UNLOCK(coveredvp);
1551 vn_finished_write(mp);
1555 * There are various reference counters associated with the mount point.
1556 * Normally it is permitted to modify them without taking the mnt ilock,
1557 * but this behavior can be temporarily disabled if stable value is needed
1558 * or callers are expected to block (e.g. to not allow new users during
1562 vfs_op_enter(struct mount *mp)
1564 struct mount_pcpu *mpcpu;
1569 if (mp->mnt_vfs_ops > 1) {
1573 vfs_op_barrier_wait(mp);
1575 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1577 mp->mnt_ref += mpcpu->mntp_ref;
1578 mpcpu->mntp_ref = 0;
1580 mp->mnt_lockref += mpcpu->mntp_lockref;
1581 mpcpu->mntp_lockref = 0;
1583 mp->mnt_writeopcount += mpcpu->mntp_writeopcount;
1584 mpcpu->mntp_writeopcount = 0;
1586 if (mp->mnt_ref <= 0 || mp->mnt_lockref < 0 || mp->mnt_writeopcount < 0)
1587 panic("%s: invalid count(s) on mp %p: ref %d lockref %d writeopcount %d\n",
1588 __func__, mp, mp->mnt_ref, mp->mnt_lockref, mp->mnt_writeopcount);
1590 vfs_assert_mount_counters(mp);
1594 vfs_op_exit_locked(struct mount *mp)
1597 mtx_assert(MNT_MTX(mp), MA_OWNED);
1599 if (mp->mnt_vfs_ops <= 0)
1600 panic("%s: invalid vfs_ops count %d for mp %p\n",
1601 __func__, mp->mnt_vfs_ops, mp);
1606 vfs_op_exit(struct mount *mp)
1610 vfs_op_exit_locked(mp);
1614 struct vfs_op_barrier_ipi {
1616 struct smp_rendezvous_cpus_retry_arg srcra;
1620 vfs_op_action_func(void *arg)
1622 struct vfs_op_barrier_ipi *vfsopipi;
1625 vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1628 if (!vfs_op_thread_entered(mp))
1629 smp_rendezvous_cpus_done(arg);
1633 vfs_op_wait_func(void *arg, int cpu)
1635 struct vfs_op_barrier_ipi *vfsopipi;
1637 struct mount_pcpu *mpcpu;
1639 vfsopipi = __containerof(arg, struct vfs_op_barrier_ipi, srcra);
1642 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1643 while (atomic_load_int(&mpcpu->mntp_thread_in_ops))
1648 vfs_op_barrier_wait(struct mount *mp)
1650 struct vfs_op_barrier_ipi vfsopipi;
1654 smp_rendezvous_cpus_retry(all_cpus,
1655 smp_no_rendezvous_barrier,
1657 smp_no_rendezvous_barrier,
1664 vfs_assert_mount_counters(struct mount *mp)
1666 struct mount_pcpu *mpcpu;
1669 if (mp->mnt_vfs_ops == 0)
1673 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1674 if (mpcpu->mntp_ref != 0 ||
1675 mpcpu->mntp_lockref != 0 ||
1676 mpcpu->mntp_writeopcount != 0)
1677 vfs_dump_mount_counters(mp);
1682 vfs_dump_mount_counters(struct mount *mp)
1684 struct mount_pcpu *mpcpu;
1685 int ref, lockref, writeopcount;
1688 printf("%s: mp %p vfs_ops %d\n", __func__, mp, mp->mnt_vfs_ops);
1693 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1694 printf("%d ", mpcpu->mntp_ref);
1695 ref += mpcpu->mntp_ref;
1698 printf(" lockref : ");
1699 lockref = mp->mnt_lockref;
1701 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1702 printf("%d ", mpcpu->mntp_lockref);
1703 lockref += mpcpu->mntp_lockref;
1706 printf("writeopcount: ");
1707 writeopcount = mp->mnt_writeopcount;
1709 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1710 printf("%d ", mpcpu->mntp_writeopcount);
1711 writeopcount += mpcpu->mntp_writeopcount;
1715 printf("counter struct total\n");
1716 printf("ref %-5d %-5d\n", mp->mnt_ref, ref);
1717 printf("lockref %-5d %-5d\n", mp->mnt_lockref, lockref);
1718 printf("writeopcount %-5d %-5d\n", mp->mnt_writeopcount, writeopcount);
1720 panic("invalid counts on struct mount");
1725 vfs_mount_fetch_counter(struct mount *mp, enum mount_counter which)
1727 struct mount_pcpu *mpcpu;
1734 case MNT_COUNT_LOCKREF:
1735 sum = mp->mnt_lockref;
1737 case MNT_COUNT_WRITEOPCOUNT:
1738 sum = mp->mnt_writeopcount;
1743 mpcpu = vfs_mount_pcpu_remote(mp, cpu);
1746 sum += mpcpu->mntp_ref;
1748 case MNT_COUNT_LOCKREF:
1749 sum += mpcpu->mntp_lockref;
1751 case MNT_COUNT_WRITEOPCOUNT:
1752 sum += mpcpu->mntp_writeopcount;
1760 * Do the actual filesystem unmount.
1763 dounmount(struct mount *mp, int flags, struct thread *td)
1765 struct vnode *coveredvp, *rootvp;
1767 uint64_t async_flag;
1770 if ((coveredvp = mp->mnt_vnodecovered) != NULL) {
1771 mnt_gen_r = mp->mnt_gen;
1774 vn_lock(coveredvp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY);
1776 * Check for mp being unmounted while waiting for the
1777 * covered vnode lock.
1779 if (coveredvp->v_mountedhere != mp ||
1780 coveredvp->v_mountedhere->mnt_gen != mnt_gen_r) {
1781 VOP_UNLOCK(coveredvp);
1789 * Only privileged root, or (if MNT_USER is set) the user that did the
1790 * original mount is permitted to unmount this filesystem.
1792 error = vfs_suser(mp, td);
1794 if (coveredvp != NULL) {
1795 VOP_UNLOCK(coveredvp);
1804 vn_start_write(NULL, &mp, V_WAIT | V_MNTREF);
1806 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 ||
1807 (mp->mnt_flag & MNT_UPDATE) != 0 ||
1808 !TAILQ_EMPTY(&mp->mnt_uppers)) {
1809 dounmount_cleanup(mp, coveredvp, 0);
1812 mp->mnt_kern_flag |= MNTK_UNMOUNT;
1813 rootvp = vfs_cache_root_clear(mp);
1814 if (coveredvp != NULL)
1815 vn_seqc_write_begin(coveredvp);
1816 if (flags & MNT_NONBUSY) {
1818 error = vfs_check_usecounts(mp);
1821 vn_seqc_write_end(coveredvp);
1822 dounmount_cleanup(mp, coveredvp, MNTK_UNMOUNT);
1823 if (rootvp != NULL) {
1824 vn_seqc_write_end(rootvp);
1830 /* Allow filesystems to detect that a forced unmount is in progress. */
1831 if (flags & MNT_FORCE) {
1832 mp->mnt_kern_flag |= MNTK_UNMOUNTF;
1835 * Must be done after setting MNTK_UNMOUNTF and before
1836 * waiting for mnt_lockref to become 0.
1842 if (mp->mnt_lockref) {
1843 mp->mnt_kern_flag |= MNTK_DRAINING;
1844 error = msleep(&mp->mnt_lockref, MNT_MTX(mp), PVFS,
1848 KASSERT(mp->mnt_lockref == 0,
1849 ("%s: invalid lock refcount in the drain path @ %s:%d",
1850 __func__, __FILE__, __LINE__));
1852 ("%s: invalid return value for msleep in the drain path @ %s:%d",
1853 __func__, __FILE__, __LINE__));
1856 * We want to keep the vnode around so that we can vn_seqc_write_end
1857 * after we are done with unmount. Downgrade our reference to a mere
1858 * hold count so that we don't interefere with anything.
1860 if (rootvp != NULL) {
1865 if (mp->mnt_flag & MNT_EXPUBLIC)
1866 vfs_setpublicfs(NULL, NULL, NULL);
1868 vfs_periodic(mp, MNT_WAIT);
1870 async_flag = mp->mnt_flag & MNT_ASYNC;
1871 mp->mnt_flag &= ~MNT_ASYNC;
1872 mp->mnt_kern_flag &= ~MNTK_ASYNC;
1874 vfs_deallocate_syncvnode(mp);
1875 error = VFS_UNMOUNT(mp, flags);
1876 vn_finished_write(mp);
1878 * If we failed to flush the dirty blocks for this mount point,
1879 * undo all the cdir/rdir and rootvnode changes we made above.
1880 * Unless we failed to do so because the device is reporting that
1881 * it doesn't exist anymore.
1883 if (error && error != ENXIO) {
1885 if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1887 vfs_allocate_syncvnode(mp);
1890 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
1891 mp->mnt_flag |= async_flag;
1892 if ((mp->mnt_flag & MNT_ASYNC) != 0 &&
1893 (mp->mnt_kern_flag & MNTK_NOASYNC) == 0)
1894 mp->mnt_kern_flag |= MNTK_ASYNC;
1895 if (mp->mnt_kern_flag & MNTK_MWAIT) {
1896 mp->mnt_kern_flag &= ~MNTK_MWAIT;
1899 vfs_op_exit_locked(mp);
1902 vn_seqc_write_end(coveredvp);
1903 VOP_UNLOCK(coveredvp);
1906 if (rootvp != NULL) {
1907 vn_seqc_write_end(rootvp);
1912 mtx_lock(&mountlist_mtx);
1913 TAILQ_REMOVE(&mountlist, mp, mnt_list);
1914 mtx_unlock(&mountlist_mtx);
1915 EVENTHANDLER_DIRECT_INVOKE(vfs_unmounted, mp, td);
1916 if (coveredvp != NULL) {
1918 vn_irflag_unset_locked(coveredvp, VIRF_MOUNTPOINT);
1919 coveredvp->v_mountedhere = NULL;
1920 vn_seqc_write_end_locked(coveredvp);
1921 VI_UNLOCK(coveredvp);
1922 VOP_UNLOCK(coveredvp);
1925 mount_devctl_event("UNMOUNT", mp, false);
1926 if (rootvp != NULL) {
1927 vn_seqc_write_end(rootvp);
1930 vfs_event_signal(NULL, VQ_UNMOUNT, 0);
1931 if (rootvnode != NULL && mp == rootvnode->v_mount) {
1935 if (mp == rootdevmp)
1937 vfs_mount_destroy(mp);
1942 * Report errors during filesystem mounting.
1945 vfs_mount_error(struct mount *mp, const char *fmt, ...)
1947 struct vfsoptlist *moptlist = mp->mnt_optnew;
1952 error = vfs_getopt(moptlist, "errmsg", (void **)&errmsg, &len);
1953 if (error || errmsg == NULL || len <= 0)
1957 vsnprintf(errmsg, (size_t)len, fmt, ap);
1962 vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...)
1968 error = vfs_getopt(opts, "errmsg", (void **)&errmsg, &len);
1969 if (error || errmsg == NULL || len <= 0)
1973 vsnprintf(errmsg, (size_t)len, fmt, ap);
1978 * ---------------------------------------------------------------------
1979 * Functions for querying mount options/arguments from filesystems.
1983 * Check that no unknown options are given
1986 vfs_filteropt(struct vfsoptlist *opts, const char **legal)
1990 const char **t, *p, *q;
1993 TAILQ_FOREACH(opt, opts, link) {
1996 if (p[0] == 'n' && p[1] == 'o')
1998 for(t = global_opts; *t != NULL; t++) {
1999 if (strcmp(*t, p) == 0)
2002 if (strcmp(*t, q) == 0)
2008 for(t = legal; *t != NULL; t++) {
2009 if (strcmp(*t, p) == 0)
2012 if (strcmp(*t, q) == 0)
2018 snprintf(errmsg, sizeof(errmsg),
2019 "mount option <%s> is unknown", p);
2023 TAILQ_FOREACH(opt, opts, link) {
2024 if (strcmp(opt->name, "errmsg") == 0) {
2025 strncpy((char *)opt->value, errmsg, opt->len);
2030 printf("%s\n", errmsg);
2036 * Get a mount option by its name.
2038 * Return 0 if the option was found, ENOENT otherwise.
2039 * If len is non-NULL it will be filled with the length
2040 * of the option. If buf is non-NULL, it will be filled
2041 * with the address of the option.
2044 vfs_getopt(struct vfsoptlist *opts, const char *name, void **buf, int *len)
2048 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2050 TAILQ_FOREACH(opt, opts, link) {
2051 if (strcmp(name, opt->name) == 0) {
2064 vfs_getopt_pos(struct vfsoptlist *opts, const char *name)
2071 TAILQ_FOREACH(opt, opts, link) {
2072 if (strcmp(name, opt->name) == 0) {
2081 vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value)
2083 char *opt_value, *vtp;
2087 error = vfs_getopt(opts, name, (void **)&opt_value, &opt_len);
2090 if (opt_len == 0 || opt_value == NULL)
2092 if (opt_value[0] == '\0' || opt_value[opt_len - 1] != '\0')
2094 iv = strtoq(opt_value, &vtp, 0);
2095 if (vtp == opt_value || (vtp[0] != '\0' && vtp[1] != '\0'))
2122 vfs_getopts(struct vfsoptlist *opts, const char *name, int *error)
2127 TAILQ_FOREACH(opt, opts, link) {
2128 if (strcmp(name, opt->name) != 0)
2131 if (opt->len == 0 ||
2132 ((char *)opt->value)[opt->len - 1] != '\0') {
2136 return (opt->value);
2143 vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w,
2148 TAILQ_FOREACH(opt, opts, link) {
2149 if (strcmp(name, opt->name) == 0) {
2162 vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...)
2168 KASSERT(opts != NULL, ("vfs_getopt: caller passed 'opts' as NULL"));
2170 TAILQ_FOREACH(opt, opts, link) {
2171 if (strcmp(name, opt->name) != 0)
2174 if (opt->len == 0 || opt->value == NULL)
2176 if (((char *)opt->value)[opt->len - 1] != '\0')
2179 ret = vsscanf(opt->value, fmt, ap);
2187 vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len)
2191 TAILQ_FOREACH(opt, opts, link) {
2192 if (strcmp(name, opt->name) != 0)
2195 if (opt->value == NULL)
2198 if (opt->len != len)
2200 bcopy(value, opt->value, len);
2208 vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len)
2212 TAILQ_FOREACH(opt, opts, link) {
2213 if (strcmp(name, opt->name) != 0)
2216 if (opt->value == NULL)
2222 bcopy(value, opt->value, len);
2230 vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value)
2234 TAILQ_FOREACH(opt, opts, link) {
2235 if (strcmp(name, opt->name) != 0)
2238 if (opt->value == NULL)
2239 opt->len = strlen(value) + 1;
2240 else if (strlcpy(opt->value, value, opt->len) >= opt->len)
2248 * Find and copy a mount option.
2250 * The size of the buffer has to be specified
2251 * in len, if it is not the same length as the
2252 * mount option, EINVAL is returned.
2253 * Returns ENOENT if the option is not found.
2256 vfs_copyopt(struct vfsoptlist *opts, const char *name, void *dest, int len)
2260 KASSERT(opts != NULL, ("vfs_copyopt: caller passed 'opts' as NULL"));
2262 TAILQ_FOREACH(opt, opts, link) {
2263 if (strcmp(name, opt->name) == 0) {
2265 if (len != opt->len)
2267 bcopy(opt->value, dest, opt->len);
2275 __vfs_statfs(struct mount *mp, struct statfs *sbp)
2279 * Filesystems only fill in part of the structure for updates, we
2280 * have to read the entirety first to get all content.
2282 if (sbp != &mp->mnt_stat)
2283 memcpy(sbp, &mp->mnt_stat, sizeof(*sbp));
2286 * Set these in case the underlying filesystem fails to do so.
2288 sbp->f_version = STATFS_VERSION;
2289 sbp->f_namemax = NAME_MAX;
2290 sbp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
2292 return (mp->mnt_op->vfs_statfs(mp, sbp));
2296 vfs_mountedfrom(struct mount *mp, const char *from)
2299 bzero(mp->mnt_stat.f_mntfromname, sizeof mp->mnt_stat.f_mntfromname);
2300 strlcpy(mp->mnt_stat.f_mntfromname, from,
2301 sizeof mp->mnt_stat.f_mntfromname);
2305 * ---------------------------------------------------------------------
2306 * This is the api for building mount args and mounting filesystems from
2307 * inside the kernel.
2309 * The API works by accumulation of individual args. First error is
2312 * XXX: should be documented in new manpage kernel_mount(9)
2315 /* A memory allocation which must be freed when we are done */
2317 SLIST_ENTRY(mntaarg) next;
2320 /* The header for the mount arguments */
2325 SLIST_HEAD(, mntaarg) list;
2329 * Add a boolean argument.
2331 * flag is the boolean value.
2332 * name must start with "no".
2335 mount_argb(struct mntarg *ma, int flag, const char *name)
2338 KASSERT(name[0] == 'n' && name[1] == 'o',
2339 ("mount_argb(...,%s): name must start with 'no'", name));
2341 return (mount_arg(ma, name + (flag ? 2 : 0), NULL, 0));
2345 * Add an argument printf style
2348 mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...)
2351 struct mntaarg *maa;
2356 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2357 SLIST_INIT(&ma->list);
2362 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2364 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2365 ma->v[ma->len].iov_len = strlen(name) + 1;
2368 sb = sbuf_new_auto();
2370 sbuf_vprintf(sb, fmt, ap);
2373 len = sbuf_len(sb) + 1;
2374 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2375 SLIST_INSERT_HEAD(&ma->list, maa, next);
2376 bcopy(sbuf_data(sb), maa + 1, len);
2379 ma->v[ma->len].iov_base = maa + 1;
2380 ma->v[ma->len].iov_len = len;
2387 * Add an argument which is a userland string.
2390 mount_argsu(struct mntarg *ma, const char *name, const void *val, int len)
2392 struct mntaarg *maa;
2398 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2399 SLIST_INIT(&ma->list);
2403 maa = malloc(sizeof *maa + len, M_MOUNT, M_WAITOK | M_ZERO);
2404 SLIST_INSERT_HEAD(&ma->list, maa, next);
2405 tbuf = (void *)(maa + 1);
2406 ma->error = copyinstr(val, tbuf, len, NULL);
2407 return (mount_arg(ma, name, tbuf, -1));
2413 * If length is -1, treat value as a C string.
2416 mount_arg(struct mntarg *ma, const char *name, const void *val, int len)
2420 ma = malloc(sizeof *ma, M_MOUNT, M_WAITOK | M_ZERO);
2421 SLIST_INIT(&ma->list);
2426 ma->v = realloc(ma->v, sizeof *ma->v * (ma->len + 2),
2428 ma->v[ma->len].iov_base = (void *)(uintptr_t)name;
2429 ma->v[ma->len].iov_len = strlen(name) + 1;
2432 ma->v[ma->len].iov_base = (void *)(uintptr_t)val;
2434 ma->v[ma->len].iov_len = strlen(val) + 1;
2436 ma->v[ma->len].iov_len = len;
2442 * Free a mntarg structure
2445 free_mntarg(struct mntarg *ma)
2447 struct mntaarg *maa;
2449 while (!SLIST_EMPTY(&ma->list)) {
2450 maa = SLIST_FIRST(&ma->list);
2451 SLIST_REMOVE_HEAD(&ma->list, next);
2454 free(ma->v, M_MOUNT);
2459 * Mount a filesystem
2462 kernel_mount(struct mntarg *ma, uint64_t flags)
2467 KASSERT(ma != NULL, ("kernel_mount NULL ma"));
2468 KASSERT(ma->v != NULL, ("kernel_mount NULL ma->v"));
2469 KASSERT(!(ma->len & 1), ("kernel_mount odd ma->len (%d)", ma->len));
2471 auio.uio_iov = ma->v;
2472 auio.uio_iovcnt = ma->len;
2473 auio.uio_segflg = UIO_SYSSPACE;
2477 error = vfs_donmount(curthread, flags, &auio);
2483 * A printflike function to mount a filesystem.
2486 kernel_vmount(int flags, ...)
2488 struct mntarg *ma = NULL;
2494 va_start(ap, flags);
2496 cp = va_arg(ap, const char *);
2499 vp = va_arg(ap, const void *);
2500 ma = mount_arg(ma, cp, vp, (vp != NULL ? -1 : 0));
2504 error = kernel_mount(ma, flags);
2508 /* Map from mount options to printable formats. */
2509 static struct mntoptnames optnames[] = {
2513 #define DEVCTL_LEN 1024
2515 mount_devctl_event(const char *type, struct mount *mp, bool donew)
2518 struct mntoptnames *fp;
2520 struct statfs *sfp = &mp->mnt_stat;
2523 buf = malloc(DEVCTL_LEN, M_MOUNT, M_NOWAIT);
2526 sbuf_new(&sb, buf, DEVCTL_LEN, SBUF_FIXEDLEN);
2527 sbuf_cpy(&sb, "mount-point=\"");
2528 devctl_safe_quote_sb(&sb, sfp->f_mntonname);
2529 sbuf_cat(&sb, "\" mount-dev=\"");
2530 devctl_safe_quote_sb(&sb, sfp->f_mntfromname);
2531 sbuf_cat(&sb, "\" mount-type=\"");
2532 devctl_safe_quote_sb(&sb, sfp->f_fstypename);
2533 sbuf_cat(&sb, "\" fsid=0x");
2534 cp = (const uint8_t *)&sfp->f_fsid.val[0];
2535 for (int i = 0; i < sizeof(sfp->f_fsid); i++)
2536 sbuf_printf(&sb, "%02x", cp[i]);
2537 sbuf_printf(&sb, " owner=%u flags=\"", sfp->f_owner);
2538 for (fp = optnames; fp->o_opt != 0; fp++) {
2539 if ((mp->mnt_flag & fp->o_opt) != 0) {
2540 sbuf_cat(&sb, fp->o_name);
2541 sbuf_putc(&sb, ';');
2544 sbuf_putc(&sb, '"');
2548 * Options are not published because the form of the options depends on
2549 * the file system and may include binary data. In addition, they don't
2550 * necessarily provide enough useful information to be actionable when
2551 * devd processes them.
2554 if (sbuf_error(&sb) == 0)
2555 devctl_notify("VFS", "FS", type, sbuf_data(&sb));
2561 * Suspend write operations on all local writeable filesystems. Does
2562 * full sync of them in the process.
2564 * Iterate over the mount points in reverse order, suspending most
2565 * recently mounted filesystems first. It handles a case where a
2566 * filesystem mounted from a md(4) vnode-backed device should be
2567 * suspended before the filesystem that owns the vnode.
2570 suspend_all_fs(void)
2575 mtx_lock(&mountlist_mtx);
2576 TAILQ_FOREACH_REVERSE(mp, &mountlist, mntlist, mnt_list) {
2577 error = vfs_busy(mp, MBF_MNTLSTLOCK | MBF_NOWAIT);
2580 if ((mp->mnt_flag & (MNT_RDONLY | MNT_LOCAL)) != MNT_LOCAL ||
2581 (mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
2582 mtx_lock(&mountlist_mtx);
2586 error = vfs_write_suspend(mp, 0);
2589 MPASS((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0);
2590 mp->mnt_kern_flag |= MNTK_SUSPEND_ALL;
2592 mtx_lock(&mountlist_mtx);
2594 printf("suspend of %s failed, error %d\n",
2595 mp->mnt_stat.f_mntonname, error);
2596 mtx_lock(&mountlist_mtx);
2600 mtx_unlock(&mountlist_mtx);
2608 mtx_lock(&mountlist_mtx);
2609 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2610 if ((mp->mnt_kern_flag & MNTK_SUSPEND_ALL) == 0)
2612 mtx_unlock(&mountlist_mtx);
2614 MPASS((mp->mnt_kern_flag & MNTK_SUSPEND) != 0);
2615 mp->mnt_kern_flag &= ~MNTK_SUSPEND_ALL;
2617 vfs_write_resume(mp, 0);
2618 mtx_lock(&mountlist_mtx);
2621 mtx_unlock(&mountlist_mtx);