Home | History | Annotate | Line # | Download | only in kern
vfs_mount.c revision 1.79
      1 /*	$NetBSD: vfs_mount.c,v 1.79 2020/04/19 13:26:17 hannken Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 1997-2020 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * This code is derived from software contributed to The NetBSD Foundation
      8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
      9  * NASA Ames Research Center, by Charles M. Hannum, and by Andrew Doran.
     10  *
     11  * Redistribution and use in source and binary forms, with or without
     12  * modification, are permitted provided that the following conditions
     13  * are met:
     14  * 1. Redistributions of source code must retain the above copyright
     15  *    notice, this list of conditions and the following disclaimer.
     16  * 2. Redistributions in binary form must reproduce the above copyright
     17  *    notice, this list of conditions and the following disclaimer in the
     18  *    documentation and/or other materials provided with the distribution.
     19  *
     20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     30  * POSSIBILITY OF SUCH DAMAGE.
     31  */
     32 
     33 /*
     34  * Copyright (c) 1989, 1993
     35  *	The Regents of the University of California.  All rights reserved.
     36  * (c) UNIX System Laboratories, Inc.
     37  * All or some portions of this file are derived from material licensed
     38  * to the University of California by American Telephone and Telegraph
     39  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
     40  * the permission of UNIX System Laboratories, Inc.
     41  *
     42  * Redistribution and use in source and binary forms, with or without
     43  * modification, are permitted provided that the following conditions
     44  * are met:
     45  * 1. Redistributions of source code must retain the above copyright
     46  *    notice, this list of conditions and the following disclaimer.
     47  * 2. Redistributions in binary form must reproduce the above copyright
     48  *    notice, this list of conditions and the following disclaimer in the
     49  *    documentation and/or other materials provided with the distribution.
     50  * 3. Neither the name of the University nor the names of its contributors
     51  *    may be used to endorse or promote products derived from this software
     52  *    without specific prior written permission.
     53  *
     54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     64  * SUCH DAMAGE.
     65  *
     66  *	@(#)vfs_subr.c	8.13 (Berkeley) 4/18/94
     67  */
     68 
     69 #include <sys/cdefs.h>
     70 __KERNEL_RCSID(0, "$NetBSD: vfs_mount.c,v 1.79 2020/04/19 13:26:17 hannken Exp $");
     71 
     72 #include <sys/param.h>
     73 #include <sys/kernel.h>
     74 
     75 #include <sys/atomic.h>
     76 #include <sys/buf.h>
     77 #include <sys/conf.h>
     78 #include <sys/fcntl.h>
     79 #include <sys/filedesc.h>
     80 #include <sys/device.h>
     81 #include <sys/kauth.h>
     82 #include <sys/kmem.h>
     83 #include <sys/module.h>
     84 #include <sys/mount.h>
     85 #include <sys/fstrans.h>
     86 #include <sys/namei.h>
     87 #include <sys/extattr.h>
     88 #include <sys/syscallargs.h>
     89 #include <sys/sysctl.h>
     90 #include <sys/systm.h>
     91 #include <sys/vfs_syscalls.h>
     92 #include <sys/vnode_impl.h>
     93 #include <sys/xcall.h>
     94 
     95 #include <miscfs/genfs/genfs.h>
     96 #include <miscfs/specfs/specdev.h>
     97 
     98 enum mountlist_type {
     99 	ME_MOUNT,
    100 	ME_MARKER
    101 };
    102 struct mountlist_entry {
    103 	TAILQ_ENTRY(mountlist_entry) me_list;	/* Mount list. */
    104 	struct mount *me_mount;			/* Actual mount if ME_MOUNT,
    105 						   current mount else. */
    106 	enum mountlist_type me_type;		/* Mount or marker. */
    107 };
    108 struct mount_iterator {
    109 	struct mountlist_entry mi_entry;
    110 };
    111 
    112 static struct vnode *vfs_vnode_iterator_next1(struct vnode_iterator *,
    113     bool (*)(void *, struct vnode *), void *, bool);
    114 
    115 /* Root filesystem. */
    116 vnode_t *			rootvnode;
    117 
    118 extern struct mount		*dead_rootmount;
    119 
    120 /* Mounted filesystem list. */
    121 static TAILQ_HEAD(mountlist, mountlist_entry) mountlist;
    122 static kmutex_t			mountlist_lock __cacheline_aligned;
    123 int vnode_offset_next_by_lru	/* XXX: ugly hack for pstat.c */
    124     = offsetof(vnode_impl_t, vi_lrulist.tqe_next);
    125 
    126 kmutex_t			vfs_list_lock __cacheline_aligned;
    127 
    128 static specificdata_domain_t	mount_specificdata_domain;
    129 static kmutex_t			mntid_lock;
    130 
    131 static kmutex_t			mountgen_lock __cacheline_aligned;
    132 static uint64_t			mountgen;
    133 
    134 void
    135 vfs_mount_sysinit(void)
    136 {
    137 
    138 	TAILQ_INIT(&mountlist);
    139 	mutex_init(&mountlist_lock, MUTEX_DEFAULT, IPL_NONE);
    140 	mutex_init(&vfs_list_lock, MUTEX_DEFAULT, IPL_NONE);
    141 
    142 	mount_specificdata_domain = specificdata_domain_create();
    143 	mutex_init(&mntid_lock, MUTEX_DEFAULT, IPL_NONE);
    144 	mutex_init(&mountgen_lock, MUTEX_DEFAULT, IPL_NONE);
    145 	mountgen = 0;
    146 }
    147 
    148 struct mount *
    149 vfs_mountalloc(struct vfsops *vfsops, vnode_t *vp)
    150 {
    151 	struct mount *mp;
    152 	int error __diagused;
    153 
    154 	mp = kmem_zalloc(sizeof(*mp), KM_SLEEP);
    155 	mp->mnt_op = vfsops;
    156 	mp->mnt_refcnt = 1;
    157 	TAILQ_INIT(&mp->mnt_vnodelist);
    158 	mp->mnt_renamelock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
    159 	mp->mnt_vnodelock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
    160 	mp->mnt_updating = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
    161 	mp->mnt_vnodecovered = vp;
    162 	mount_initspecific(mp);
    163 
    164 	error = fstrans_mount(mp);
    165 	KASSERT(error == 0);
    166 
    167 	mutex_enter(&mountgen_lock);
    168 	mp->mnt_gen = mountgen++;
    169 	mutex_exit(&mountgen_lock);
    170 
    171 	return mp;
    172 }
    173 
    174 /*
    175  * vfs_rootmountalloc: lookup a filesystem type, and if found allocate and
    176  * initialize a mount structure for it.
    177  *
    178  * Devname is usually updated by mount(8) after booting.
    179  */
    180 int
    181 vfs_rootmountalloc(const char *fstypename, const char *devname,
    182     struct mount **mpp)
    183 {
    184 	struct vfsops *vfsp = NULL;
    185 	struct mount *mp;
    186 	int error __diagused;
    187 
    188 	mutex_enter(&vfs_list_lock);
    189 	LIST_FOREACH(vfsp, &vfs_list, vfs_list)
    190 		if (!strncmp(vfsp->vfs_name, fstypename,
    191 		    sizeof(mp->mnt_stat.f_fstypename)))
    192 			break;
    193 	if (vfsp == NULL) {
    194 		mutex_exit(&vfs_list_lock);
    195 		return (ENODEV);
    196 	}
    197 	vfsp->vfs_refcount++;
    198 	mutex_exit(&vfs_list_lock);
    199 
    200 	if ((mp = vfs_mountalloc(vfsp, NULL)) == NULL)
    201 		return ENOMEM;
    202 	error = vfs_busy(mp);
    203 	KASSERT(error == 0);
    204 	mp->mnt_flag = MNT_RDONLY;
    205 	(void)strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfs_name,
    206 	    sizeof(mp->mnt_stat.f_fstypename));
    207 	mp->mnt_stat.f_mntonname[0] = '/';
    208 	mp->mnt_stat.f_mntonname[1] = '\0';
    209 	mp->mnt_stat.f_mntfromname[sizeof(mp->mnt_stat.f_mntfromname) - 1] =
    210 	    '\0';
    211 	(void)copystr(devname, mp->mnt_stat.f_mntfromname,
    212 	    sizeof(mp->mnt_stat.f_mntfromname) - 1, 0);
    213 	*mpp = mp;
    214 	return 0;
    215 }
    216 
    217 /*
    218  * vfs_getnewfsid: get a new unique fsid.
    219  */
    220 void
    221 vfs_getnewfsid(struct mount *mp)
    222 {
    223 	static u_short xxxfs_mntid;
    224 	fsid_t tfsid;
    225 	int mtype;
    226 
    227 	mutex_enter(&mntid_lock);
    228 	mtype = makefstype(mp->mnt_op->vfs_name);
    229 	mp->mnt_stat.f_fsidx.__fsid_val[0] = makedev(mtype, 0);
    230 	mp->mnt_stat.f_fsidx.__fsid_val[1] = mtype;
    231 	mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
    232 	if (xxxfs_mntid == 0)
    233 		++xxxfs_mntid;
    234 	tfsid.__fsid_val[0] = makedev(mtype & 0xff, xxxfs_mntid);
    235 	tfsid.__fsid_val[1] = mtype;
    236 	while (vfs_getvfs(&tfsid)) {
    237 		tfsid.__fsid_val[0]++;
    238 		xxxfs_mntid++;
    239 	}
    240 	mp->mnt_stat.f_fsidx.__fsid_val[0] = tfsid.__fsid_val[0];
    241 	mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0];
    242 	mutex_exit(&mntid_lock);
    243 }
    244 
    245 /*
    246  * Lookup a mount point by filesystem identifier.
    247  *
    248  * XXX Needs to add a reference to the mount point.
    249  */
    250 struct mount *
    251 vfs_getvfs(fsid_t *fsid)
    252 {
    253 	mount_iterator_t *iter;
    254 	struct mount *mp;
    255 
    256 	mountlist_iterator_init(&iter);
    257 	while ((mp = mountlist_iterator_next(iter)) != NULL) {
    258 		if (mp->mnt_stat.f_fsidx.__fsid_val[0] == fsid->__fsid_val[0] &&
    259 		    mp->mnt_stat.f_fsidx.__fsid_val[1] == fsid->__fsid_val[1]) {
    260 			mountlist_iterator_destroy(iter);
    261 			return mp;
    262 		}
    263 	}
    264 	mountlist_iterator_destroy(iter);
    265 	return NULL;
    266 }
    267 
    268 /*
    269  * Take a reference to a mount structure.
    270  */
    271 void
    272 vfs_ref(struct mount *mp)
    273 {
    274 
    275 	KASSERT(mp->mnt_refcnt > 0 || mutex_owned(&mountlist_lock));
    276 
    277 	atomic_inc_uint(&mp->mnt_refcnt);
    278 }
    279 
    280 /*
    281  * Drop a reference to a mount structure, freeing if the last reference.
    282  */
    283 void
    284 vfs_rele(struct mount *mp)
    285 {
    286 
    287 	if (__predict_true((int)atomic_dec_uint_nv(&mp->mnt_refcnt) > 0)) {
    288 		return;
    289 	}
    290 
    291 	/*
    292 	 * Nothing else has visibility of the mount: we can now
    293 	 * free the data structures.
    294 	 */
    295 	KASSERT(mp->mnt_refcnt == 0);
    296 	specificdata_fini(mount_specificdata_domain, &mp->mnt_specdataref);
    297 	mutex_obj_free(mp->mnt_updating);
    298 	mutex_obj_free(mp->mnt_renamelock);
    299 	mutex_obj_free(mp->mnt_vnodelock);
    300 	if (mp->mnt_op != NULL) {
    301 		vfs_delref(mp->mnt_op);
    302 	}
    303 	fstrans_unmount(mp);
    304 	/*
    305 	 * Final free of mp gets done from fstrans_mount_dtor().
    306 	 *
    307 	 * Prevents this memory to be reused as a mount before
    308 	 * fstrans releases all references to it.
    309 	 */
    310 }
    311 
    312 /*
    313  * Mark a mount point as busy, and gain a new reference to it.  Used to
    314  * prevent the file system from being unmounted during critical sections.
    315  *
    316  * vfs_busy can be called multiple times and by multiple threads
    317  * and must be accompanied by the same number of vfs_unbusy calls.
    318  *
    319  * => The caller must hold a pre-existing reference to the mount.
    320  * => Will fail if the file system is being unmounted, or is unmounted.
    321  */
    322 static inline int
    323 _vfs_busy(struct mount *mp, bool wait)
    324 {
    325 
    326 	KASSERT(mp->mnt_refcnt > 0);
    327 
    328 	if (wait) {
    329 		fstrans_start(mp);
    330 	} else {
    331 		if (fstrans_start_nowait(mp))
    332 			return EBUSY;
    333 	}
    334 	if (__predict_false((mp->mnt_iflag & IMNT_GONE) != 0)) {
    335 		fstrans_done(mp);
    336 		return ENOENT;
    337 	}
    338 	vfs_ref(mp);
    339 	return 0;
    340 }
    341 
    342 int
    343 vfs_busy(struct mount *mp)
    344 {
    345 
    346 	return _vfs_busy(mp, true);
    347 }
    348 
    349 int
    350 vfs_trybusy(struct mount *mp)
    351 {
    352 
    353 	return _vfs_busy(mp, false);
    354 }
    355 
    356 /*
    357  * Unbusy a busy filesystem.
    358  *
    359  * Every successful vfs_busy() call must be undone by a vfs_unbusy() call.
    360  */
    361 void
    362 vfs_unbusy(struct mount *mp)
    363 {
    364 
    365 	KASSERT(mp->mnt_refcnt > 0);
    366 
    367 	fstrans_done(mp);
    368 	vfs_rele(mp);
    369 }
    370 
    371 struct vnode_iterator {
    372 	vnode_impl_t vi_vnode;
    373 };
    374 
    375 void
    376 vfs_vnode_iterator_init(struct mount *mp, struct vnode_iterator **vnip)
    377 {
    378 	vnode_t *vp;
    379 	vnode_impl_t *vip;
    380 
    381 	vp = vnalloc_marker(mp);
    382 	vip = VNODE_TO_VIMPL(vp);
    383 
    384 	mutex_enter(mp->mnt_vnodelock);
    385 	TAILQ_INSERT_HEAD(&mp->mnt_vnodelist, vip, vi_mntvnodes);
    386 	vp->v_usecount = 1;
    387 	mutex_exit(mp->mnt_vnodelock);
    388 
    389 	*vnip = (struct vnode_iterator *)vip;
    390 }
    391 
    392 void
    393 vfs_vnode_iterator_destroy(struct vnode_iterator *vni)
    394 {
    395 	vnode_impl_t *mvip = &vni->vi_vnode;
    396 	vnode_t *mvp = VIMPL_TO_VNODE(mvip);
    397 	kmutex_t *lock;
    398 
    399 	KASSERT(vnis_marker(mvp));
    400 	if (vrefcnt(mvp) != 0) {
    401 		lock = mvp->v_mount->mnt_vnodelock;
    402 		mutex_enter(lock);
    403 		TAILQ_REMOVE(&mvp->v_mount->mnt_vnodelist, mvip, vi_mntvnodes);
    404 		mvp->v_usecount = 0;
    405 		mutex_exit(lock);
    406 	}
    407 	vnfree_marker(mvp);
    408 }
    409 
    410 static struct vnode *
    411 vfs_vnode_iterator_next1(struct vnode_iterator *vni,
    412     bool (*f)(void *, struct vnode *), void *cl, bool do_wait)
    413 {
    414 	vnode_impl_t *mvip = &vni->vi_vnode;
    415 	struct mount *mp = VIMPL_TO_VNODE(mvip)->v_mount;
    416 	vnode_t *vp;
    417 	vnode_impl_t *vip;
    418 	kmutex_t *lock;
    419 	int error;
    420 
    421 	KASSERT(vnis_marker(VIMPL_TO_VNODE(mvip)));
    422 
    423 	lock = mp->mnt_vnodelock;
    424 	do {
    425 		mutex_enter(lock);
    426 		vip = TAILQ_NEXT(mvip, vi_mntvnodes);
    427 		TAILQ_REMOVE(&mp->mnt_vnodelist, mvip, vi_mntvnodes);
    428 		VIMPL_TO_VNODE(mvip)->v_usecount = 0;
    429 again:
    430 		if (vip == NULL) {
    431 			mutex_exit(lock);
    432 	       		return NULL;
    433 		}
    434 		vp = VIMPL_TO_VNODE(vip);
    435 		KASSERT(vp != NULL);
    436 		mutex_enter(vp->v_interlock);
    437 		if (vnis_marker(vp) ||
    438 		    vdead_check(vp, (do_wait ? 0 : VDEAD_NOWAIT)) ||
    439 		    (f && !(*f)(cl, vp))) {
    440 			mutex_exit(vp->v_interlock);
    441 			vip = TAILQ_NEXT(vip, vi_mntvnodes);
    442 			goto again;
    443 		}
    444 
    445 		TAILQ_INSERT_AFTER(&mp->mnt_vnodelist, vip, mvip, vi_mntvnodes);
    446 		VIMPL_TO_VNODE(mvip)->v_usecount = 1;
    447 		mutex_exit(lock);
    448 		error = vcache_vget(vp);
    449 		KASSERT(error == 0 || error == ENOENT);
    450 	} while (error != 0);
    451 
    452 	return vp;
    453 }
    454 
    455 struct vnode *
    456 vfs_vnode_iterator_next(struct vnode_iterator *vni,
    457     bool (*f)(void *, struct vnode *), void *cl)
    458 {
    459 
    460 	return vfs_vnode_iterator_next1(vni, f, cl, false);
    461 }
    462 
    463 /*
    464  * Move a vnode from one mount queue to another.
    465  */
    466 void
    467 vfs_insmntque(vnode_t *vp, struct mount *mp)
    468 {
    469 	vnode_impl_t *vip = VNODE_TO_VIMPL(vp);
    470 	struct mount *omp;
    471 	kmutex_t *lock;
    472 
    473 	KASSERT(mp == NULL || (mp->mnt_iflag & IMNT_UNMOUNT) == 0 ||
    474 	    vp->v_tag == VT_VFS);
    475 
    476 	/*
    477 	 * Delete from old mount point vnode list, if on one.
    478 	 */
    479 	if ((omp = vp->v_mount) != NULL) {
    480 		lock = omp->mnt_vnodelock;
    481 		mutex_enter(lock);
    482 		TAILQ_REMOVE(&vp->v_mount->mnt_vnodelist, vip, vi_mntvnodes);
    483 		mutex_exit(lock);
    484 	}
    485 
    486 	/*
    487 	 * Insert into list of vnodes for the new mount point, if
    488 	 * available.  The caller must take a reference on the mount
    489 	 * structure and donate to the vnode.
    490 	 */
    491 	if ((vp->v_mount = mp) != NULL) {
    492 		lock = mp->mnt_vnodelock;
    493 		mutex_enter(lock);
    494 		TAILQ_INSERT_TAIL(&mp->mnt_vnodelist, vip, vi_mntvnodes);
    495 		mutex_exit(lock);
    496 	}
    497 
    498 	if (omp != NULL) {
    499 		/* Release reference to old mount. */
    500 		vfs_rele(omp);
    501 	}
    502 }
    503 
    504 /*
    505  * Remove any vnodes in the vnode table belonging to mount point mp.
    506  *
    507  * If FORCECLOSE is not specified, there should not be any active ones,
    508  * return error if any are found (nb: this is a user error, not a
    509  * system error). If FORCECLOSE is specified, detach any active vnodes
    510  * that are found.
    511  *
    512  * If WRITECLOSE is set, only flush out regular file vnodes open for
    513  * writing.
    514  *
    515  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
    516  */
    517 #ifdef DEBUG
    518 int busyprt = 0;	/* print out busy vnodes */
    519 struct ctldebug debug1 = { "busyprt", &busyprt };
    520 #endif
    521 
    522 static vnode_t *
    523 vflushnext(struct vnode_iterator *marker, int *when)
    524 {
    525 	if (getticks() > *when) {
    526 		yield();
    527 		*when = getticks() + hz / 10;
    528 	}
    529 	return vfs_vnode_iterator_next1(marker, NULL, NULL, true);
    530 }
    531 
    532 /*
    533  * Flush one vnode.  Referenced on entry, unreferenced on return.
    534  */
    535 static int
    536 vflush_one(vnode_t *vp, vnode_t *skipvp, int flags)
    537 {
    538 	int error;
    539 	struct vattr vattr;
    540 
    541 	if (vp == skipvp ||
    542 	    ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM))) {
    543 		vrele(vp);
    544 		return 0;
    545 	}
    546 	/*
    547 	 * If WRITECLOSE is set, only flush out regular file
    548 	 * vnodes open for writing or open and unlinked.
    549 	 */
    550 	if ((flags & WRITECLOSE)) {
    551 		if (vp->v_type != VREG) {
    552 			vrele(vp);
    553 			return 0;
    554 		}
    555 		error = vn_lock(vp, LK_EXCLUSIVE);
    556 		if (error) {
    557 			KASSERT(error == ENOENT);
    558 			vrele(vp);
    559 			return 0;
    560 		}
    561 		error = VOP_FSYNC(vp, curlwp->l_cred, FSYNC_WAIT, 0, 0);
    562 		if (error == 0)
    563 			error = VOP_GETATTR(vp, &vattr, curlwp->l_cred);
    564 		VOP_UNLOCK(vp);
    565 		if (error) {
    566 			vrele(vp);
    567 			return error;
    568 		}
    569 		if (vp->v_writecount == 0 && vattr.va_nlink > 0) {
    570 			vrele(vp);
    571 			return 0;
    572 		}
    573 	}
    574 	/*
    575 	 * First try to recycle the vnode.
    576 	 */
    577 	if (vrecycle(vp))
    578 		return 0;
    579 	/*
    580 	 * If FORCECLOSE is set, forcibly close the vnode.
    581 	 * For block or character devices, revert to an
    582 	 * anonymous device.  For all other files, just
    583 	 * kill them.
    584 	 */
    585 	if (flags & FORCECLOSE) {
    586 		if (vrefcnt(vp) > 1 &&
    587 		    (vp->v_type == VBLK || vp->v_type == VCHR))
    588 			vcache_make_anon(vp);
    589 		else
    590 			vgone(vp);
    591 		return 0;
    592 	}
    593 	vrele(vp);
    594 	return EBUSY;
    595 }
    596 
    597 int
    598 vflush(struct mount *mp, vnode_t *skipvp, int flags)
    599 {
    600 	vnode_t *vp;
    601 	struct vnode_iterator *marker;
    602 	int busy, error, when, retries = 2;
    603 
    604 	do {
    605 		busy = error = when = 0;
    606 
    607 		/*
    608 		 * First, flush out any vnode references from the
    609 		 * deferred vrele list.
    610 		 */
    611 		vrele_flush(mp);
    612 
    613 		vfs_vnode_iterator_init(mp, &marker);
    614 
    615 		while ((vp = vflushnext(marker, &when)) != NULL) {
    616 			error = vflush_one(vp, skipvp, flags);
    617 			if (error == EBUSY) {
    618 				error = 0;
    619 				busy++;
    620 #ifdef DEBUG
    621 				if (busyprt && retries == 0)
    622 					vprint("vflush: busy vnode", vp);
    623 #endif
    624 			} else if (error != 0) {
    625 				break;
    626 			}
    627 		}
    628 
    629 		vfs_vnode_iterator_destroy(marker);
    630 	} while (error == 0 && busy > 0 && retries-- > 0);
    631 
    632 	if (error)
    633 		return error;
    634 	if (busy)
    635 		return EBUSY;
    636 	return 0;
    637 }
    638 
    639 /*
    640  * Mount a file system.
    641  */
    642 
    643 /*
    644  * Scan all active processes to see if any of them have a current or root
    645  * directory onto which the new filesystem has just been  mounted. If so,
    646  * replace them with the new mount point.
    647  */
    648 static void
    649 mount_checkdirs(vnode_t *olddp)
    650 {
    651 	vnode_t *newdp, *rele1, *rele2;
    652 	struct cwdinfo *cwdi;
    653 	struct proc *p;
    654 	bool retry;
    655 
    656 	if (vrefcnt(olddp) == 1) {
    657 		return;
    658 	}
    659 	if (VFS_ROOT(olddp->v_mountedhere, LK_EXCLUSIVE, &newdp))
    660 		panic("mount: lost mount");
    661 
    662 	do {
    663 		retry = false;
    664 		mutex_enter(proc_lock);
    665 		PROCLIST_FOREACH(p, &allproc) {
    666 			if ((cwdi = p->p_cwdi) == NULL)
    667 				continue;
    668 			/*
    669 			 * Cannot change to the old directory any more,
    670 			 * so even if we see a stale value it is not a
    671 			 * problem.
    672 			 */
    673 			if (cwdi->cwdi_cdir != olddp &&
    674 			    cwdi->cwdi_rdir != olddp)
    675 				continue;
    676 			retry = true;
    677 			rele1 = NULL;
    678 			rele2 = NULL;
    679 			atomic_inc_uint(&cwdi->cwdi_refcnt);
    680 			mutex_exit(proc_lock);
    681 			mutex_enter(&cwdi->cwdi_lock);
    682 			if (cwdi->cwdi_cdir == olddp ||
    683 			    cwdi->cwdi_rdir == olddp) {
    684 			    	/* XXX belongs in vfs_cwd.c, but rump. */
    685 			    	xc_barrier(0);
    686 			    	if (cwdi->cwdi_cdir == olddp) {
    687 					rele1 = cwdi->cwdi_cdir;
    688 					vref(newdp);
    689 					cwdi->cwdi_cdir = newdp;
    690 				}
    691 				if (cwdi->cwdi_rdir == olddp) {
    692 					rele2 = cwdi->cwdi_rdir;
    693 					vref(newdp);
    694 					cwdi->cwdi_rdir = newdp;
    695 				}
    696 			}
    697 			mutex_exit(&cwdi->cwdi_lock);
    698 			cwdfree(cwdi);
    699 			if (rele1 != NULL)
    700 				vrele(rele1);
    701 			if (rele2 != NULL)
    702 				vrele(rele2);
    703 			mutex_enter(proc_lock);
    704 			break;
    705 		}
    706 		mutex_exit(proc_lock);
    707 	} while (retry);
    708 
    709 	if (rootvnode == olddp) {
    710 		vrele(rootvnode);
    711 		vref(newdp);
    712 		rootvnode = newdp;
    713 	}
    714 	vput(newdp);
    715 }
    716 
    717 /*
    718  * Start extended attributes
    719  */
    720 static int
    721 start_extattr(struct mount *mp)
    722 {
    723 	int error;
    724 
    725 	error = VFS_EXTATTRCTL(mp, EXTATTR_CMD_START, NULL, 0, NULL);
    726 	if (error)
    727 		printf("%s: failed to start extattr: error = %d\n",
    728 		       mp->mnt_stat.f_mntonname, error);
    729 
    730 	return error;
    731 }
    732 
    733 int
    734 mount_domount(struct lwp *l, vnode_t **vpp, struct vfsops *vfsops,
    735     const char *path, int flags, void *data, size_t *data_len)
    736 {
    737 	vnode_t *vp = *vpp;
    738 	struct mount *mp;
    739 	struct pathbuf *pb;
    740 	struct nameidata nd;
    741 	int error;
    742 
    743 	error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_MOUNT,
    744 	    KAUTH_REQ_SYSTEM_MOUNT_NEW, vp, KAUTH_ARG(flags), data);
    745 	if (error) {
    746 		vfs_delref(vfsops);
    747 		return error;
    748 	}
    749 
    750 	/* Cannot make a non-dir a mount-point (from here anyway). */
    751 	if (vp->v_type != VDIR) {
    752 		vfs_delref(vfsops);
    753 		return ENOTDIR;
    754 	}
    755 
    756 	if (flags & MNT_EXPORTED) {
    757 		vfs_delref(vfsops);
    758 		return EINVAL;
    759 	}
    760 
    761 	if ((mp = vfs_mountalloc(vfsops, vp)) == NULL) {
    762 		vfs_delref(vfsops);
    763 		return ENOMEM;
    764 	}
    765 
    766 	mp->mnt_stat.f_owner = kauth_cred_geteuid(l->l_cred);
    767 
    768 	/*
    769 	 * The underlying file system may refuse the mount for
    770 	 * various reasons.  Allow the user to force it to happen.
    771 	 *
    772 	 * Set the mount level flags.
    773 	 */
    774 	mp->mnt_flag = flags & (MNT_BASIC_FLAGS | MNT_FORCE | MNT_IGNORE);
    775 
    776 	mutex_enter(mp->mnt_updating);
    777 	error = VFS_MOUNT(mp, path, data, data_len);
    778 	mp->mnt_flag &= ~MNT_OP_FLAGS;
    779 
    780 	if (error != 0)
    781 		goto err_unmounted;
    782 
    783 	/*
    784 	 * Validate and prepare the mount point.
    785 	 */
    786 	error = pathbuf_copyin(path, &pb);
    787 	if (error != 0) {
    788 		goto err_mounted;
    789 	}
    790 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb);
    791 	error = namei(&nd);
    792 	pathbuf_destroy(pb);
    793 	if (error != 0) {
    794 		goto err_mounted;
    795 	}
    796 	if (nd.ni_vp != vp) {
    797 		vput(nd.ni_vp);
    798 		error = EINVAL;
    799 		goto err_mounted;
    800 	}
    801 	if (vp->v_mountedhere != NULL) {
    802 		vput(nd.ni_vp);
    803 		error = EBUSY;
    804 		goto err_mounted;
    805 	}
    806 	error = vinvalbuf(vp, V_SAVE, l->l_cred, l, 0, 0);
    807 	if (error != 0) {
    808 		vput(nd.ni_vp);
    809 		goto err_mounted;
    810 	}
    811 
    812 	/*
    813 	 * Put the new filesystem on the mount list after root.
    814 	 */
    815 	cache_purge(vp);
    816 	mp->mnt_iflag &= ~IMNT_WANTRDWR;
    817 
    818 	mountlist_append(mp);
    819 	if ((mp->mnt_flag & (MNT_RDONLY | MNT_ASYNC)) == 0)
    820 		vfs_syncer_add_to_worklist(mp);
    821 	vp->v_mountedhere = mp;
    822 	vput(nd.ni_vp);
    823 
    824 	mount_checkdirs(vp);
    825 	mutex_exit(mp->mnt_updating);
    826 
    827 	/* Hold an additional reference to the mount across VFS_START(). */
    828 	vfs_ref(mp);
    829 	(void) VFS_STATVFS(mp, &mp->mnt_stat);
    830 	error = VFS_START(mp, 0);
    831 	if (error) {
    832 		vrele(vp);
    833 	} else if (flags & MNT_EXTATTR) {
    834 		if (start_extattr(mp) != 0)
    835 			mp->mnt_flag &= ~MNT_EXTATTR;
    836 	}
    837 	/* Drop reference held for VFS_START(). */
    838 	vfs_rele(mp);
    839 	*vpp = NULL;
    840 	return error;
    841 
    842 err_mounted:
    843 	if (VFS_UNMOUNT(mp, MNT_FORCE) != 0)
    844 		panic("Unmounting fresh file system failed");
    845 
    846 err_unmounted:
    847 	vp->v_mountedhere = NULL;
    848 	mutex_exit(mp->mnt_updating);
    849 	vfs_rele(mp);
    850 
    851 	return error;
    852 }
    853 
    854 /*
    855  * Do the actual file system unmount.  File system is assumed to have
    856  * been locked by the caller.
    857  *
    858  * => Caller hold reference to the mount, explicitly for dounmount().
    859  */
    860 int
    861 dounmount(struct mount *mp, int flags, struct lwp *l)
    862 {
    863 	vnode_t *coveredvp;
    864 	int error, async, used_syncer, used_extattr;
    865 	const bool was_suspended = fstrans_is_owner(mp);
    866 
    867 #if NVERIEXEC > 0
    868 	error = veriexec_unmountchk(mp);
    869 	if (error)
    870 		return (error);
    871 #endif /* NVERIEXEC > 0 */
    872 
    873 	if (!was_suspended) {
    874 		error = vfs_suspend(mp, 0);
    875 		if (error) {
    876 			return error;
    877 		}
    878 	}
    879 
    880 	KASSERT((mp->mnt_iflag & IMNT_GONE) == 0);
    881 
    882 	used_syncer = (mp->mnt_iflag & IMNT_ONWORKLIST) != 0;
    883 	used_extattr = mp->mnt_flag & MNT_EXTATTR;
    884 
    885 	mp->mnt_iflag |= IMNT_UNMOUNT;
    886 	mutex_enter(mp->mnt_updating);
    887 	async = mp->mnt_flag & MNT_ASYNC;
    888 	mp->mnt_flag &= ~MNT_ASYNC;
    889 	cache_purgevfs(mp);	/* remove cache entries for this file sys */
    890 	if (used_syncer)
    891 		vfs_syncer_remove_from_worklist(mp);
    892 	error = 0;
    893 	if (((mp->mnt_flag & MNT_RDONLY) == 0) && ((flags & MNT_FORCE) == 0)) {
    894 		error = VFS_SYNC(mp, MNT_WAIT, l->l_cred);
    895 	}
    896 	if (error == 0 || (flags & MNT_FORCE)) {
    897 		error = VFS_UNMOUNT(mp, flags);
    898 	}
    899 	if (error) {
    900 		mp->mnt_iflag &= ~IMNT_UNMOUNT;
    901 		if ((mp->mnt_flag & (MNT_RDONLY | MNT_ASYNC)) == 0)
    902 			vfs_syncer_add_to_worklist(mp);
    903 		mp->mnt_flag |= async;
    904 		mutex_exit(mp->mnt_updating);
    905 		if (!was_suspended)
    906 			vfs_resume(mp);
    907 		if (used_extattr) {
    908 			if (start_extattr(mp) != 0)
    909 				mp->mnt_flag &= ~MNT_EXTATTR;
    910 			else
    911 				mp->mnt_flag |= MNT_EXTATTR;
    912 		}
    913 		return (error);
    914 	}
    915 	mutex_exit(mp->mnt_updating);
    916 
    917 	/*
    918 	 * mark filesystem as gone to prevent further umounts
    919 	 * after mnt_umounting lock is gone, this also prevents
    920 	 * vfs_busy() from succeeding.
    921 	 */
    922 	mp->mnt_iflag |= IMNT_GONE;
    923 	if (!was_suspended)
    924 		vfs_resume(mp);
    925 
    926 	if ((coveredvp = mp->mnt_vnodecovered) != NULLVP) {
    927 		vn_lock(coveredvp, LK_EXCLUSIVE | LK_RETRY);
    928 		coveredvp->v_mountedhere = NULL;
    929 		VOP_UNLOCK(coveredvp);
    930 	}
    931 	mountlist_remove(mp);
    932 	if (TAILQ_FIRST(&mp->mnt_vnodelist) != NULL)
    933 		panic("unmount: dangling vnode");
    934 	vfs_hooks_unmount(mp);
    935 
    936 	vfs_rele(mp);	/* reference from mount() */
    937 	if (coveredvp != NULLVP) {
    938 		vrele(coveredvp);
    939 	}
    940 	return (0);
    941 }
    942 
    943 /*
    944  * Unmount all file systems.
    945  * We traverse the list in reverse order under the assumption that doing so
    946  * will avoid needing to worry about dependencies.
    947  */
    948 bool
    949 vfs_unmountall(struct lwp *l)
    950 {
    951 
    952 	printf("unmounting file systems...\n");
    953 	return vfs_unmountall1(l, true, true);
    954 }
    955 
    956 static void
    957 vfs_unmount_print(struct mount *mp, const char *pfx)
    958 {
    959 
    960 	aprint_verbose("%sunmounted %s on %s type %s\n", pfx,
    961 	    mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname,
    962 	    mp->mnt_stat.f_fstypename);
    963 }
    964 
    965 /*
    966  * Return the mount with the highest generation less than "gen".
    967  */
    968 static struct mount *
    969 vfs_unmount_next(uint64_t gen)
    970 {
    971 	mount_iterator_t *iter;
    972 	struct mount *mp, *nmp;
    973 
    974 	nmp = NULL;
    975 
    976 	mountlist_iterator_init(&iter);
    977 	while ((mp = mountlist_iterator_next(iter)) != NULL) {
    978 		if ((nmp == NULL || mp->mnt_gen > nmp->mnt_gen) &&
    979 		    mp->mnt_gen < gen) {
    980 			if (nmp != NULL)
    981 				vfs_rele(nmp);
    982 			nmp = mp;
    983 			vfs_ref(nmp);
    984 		}
    985 	}
    986 	mountlist_iterator_destroy(iter);
    987 
    988 	return nmp;
    989 }
    990 
    991 bool
    992 vfs_unmount_forceone(struct lwp *l)
    993 {
    994 	struct mount *mp;
    995 	int error;
    996 
    997 	mp = vfs_unmount_next(mountgen);
    998 	if (mp == NULL) {
    999 		return false;
   1000 	}
   1001 
   1002 #ifdef DEBUG
   1003 	printf("forcefully unmounting %s (%s)...\n",
   1004 	    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_mntfromname);
   1005 #endif
   1006 	if ((error = dounmount(mp, MNT_FORCE, l)) == 0) {
   1007 		vfs_unmount_print(mp, "forcefully ");
   1008 		return true;
   1009 	} else {
   1010 		vfs_rele(mp);
   1011 	}
   1012 
   1013 #ifdef DEBUG
   1014 	printf("forceful unmount of %s failed with error %d\n",
   1015 	    mp->mnt_stat.f_mntonname, error);
   1016 #endif
   1017 
   1018 	return false;
   1019 }
   1020 
   1021 bool
   1022 vfs_unmountall1(struct lwp *l, bool force, bool verbose)
   1023 {
   1024 	struct mount *mp;
   1025 	mount_iterator_t *iter;
   1026 	bool any_error = false, progress = false;
   1027 	uint64_t gen;
   1028 	int error;
   1029 
   1030 	gen = mountgen;
   1031 	for (;;) {
   1032 		mp = vfs_unmount_next(gen);
   1033 		if (mp == NULL)
   1034 			break;
   1035 		gen = mp->mnt_gen;
   1036 
   1037 #ifdef DEBUG
   1038 		printf("unmounting %p %s (%s)...\n",
   1039 		    (void *)mp, mp->mnt_stat.f_mntonname,
   1040 		    mp->mnt_stat.f_mntfromname);
   1041 #endif
   1042 		if ((error = dounmount(mp, force ? MNT_FORCE : 0, l)) == 0) {
   1043 			vfs_unmount_print(mp, "");
   1044 			progress = true;
   1045 		} else {
   1046 			vfs_rele(mp);
   1047 			if (verbose) {
   1048 				printf("unmount of %s failed with error %d\n",
   1049 				    mp->mnt_stat.f_mntonname, error);
   1050 			}
   1051 			any_error = true;
   1052 		}
   1053 	}
   1054 	if (verbose) {
   1055 		printf("unmounting done\n");
   1056 	}
   1057 	if (any_error && verbose) {
   1058 		printf("WARNING: some file systems would not unmount\n");
   1059 	}
   1060 
   1061 	/* If the mountlist is empty destroy anonymous device vnodes. */
   1062 	mountlist_iterator_init(&iter);
   1063 	if (mountlist_iterator_next(iter) == NULL) {
   1064 		struct vnode_iterator *marker;
   1065 		vnode_t *vp;
   1066 
   1067 		vfs_vnode_iterator_init(dead_rootmount, &marker);
   1068 		while ((vp = vfs_vnode_iterator_next(marker, NULL, NULL))) {
   1069 			if (vp->v_type == VCHR || vp->v_type == VBLK)
   1070 				vgone(vp);
   1071 			else
   1072 				vrele(vp);
   1073 		}
   1074 		vfs_vnode_iterator_destroy(marker);
   1075 	}
   1076 	mountlist_iterator_destroy(iter);
   1077 
   1078 	return progress;
   1079 }
   1080 
   1081 void
   1082 vfs_sync_all(struct lwp *l)
   1083 {
   1084 	printf("syncing disks... ");
   1085 
   1086 	/* remove user processes from run queue */
   1087 	suspendsched();
   1088 	(void)spl0();
   1089 
   1090 	/* avoid coming back this way again if we panic. */
   1091 	doing_shutdown = 1;
   1092 
   1093 	do_sys_sync(l);
   1094 
   1095 	/* Wait for sync to finish. */
   1096 	if (buf_syncwait() != 0) {
   1097 #if defined(DDB) && defined(DEBUG_HALT_BUSY)
   1098 		Debugger();
   1099 #endif
   1100 		printf("giving up\n");
   1101 		return;
   1102 	} else
   1103 		printf("done\n");
   1104 }
   1105 
   1106 /*
   1107  * Sync and unmount file systems before shutting down.
   1108  */
   1109 void
   1110 vfs_shutdown(void)
   1111 {
   1112 	lwp_t *l = curlwp;
   1113 
   1114 	vfs_sync_all(l);
   1115 
   1116 	/*
   1117 	 * If we have paniced - do not make the situation potentially
   1118 	 * worse by unmounting the file systems.
   1119 	 */
   1120 	if (panicstr != NULL) {
   1121 		return;
   1122 	}
   1123 
   1124 	/* Unmount file systems. */
   1125 	vfs_unmountall(l);
   1126 }
   1127 
   1128 /*
   1129  * Print a list of supported file system types (used by vfs_mountroot)
   1130  */
   1131 static void
   1132 vfs_print_fstypes(void)
   1133 {
   1134 	struct vfsops *v;
   1135 	int cnt = 0;
   1136 
   1137 	mutex_enter(&vfs_list_lock);
   1138 	LIST_FOREACH(v, &vfs_list, vfs_list)
   1139 		++cnt;
   1140 	mutex_exit(&vfs_list_lock);
   1141 
   1142 	if (cnt == 0) {
   1143 		printf("WARNING: No file system modules have been loaded.\n");
   1144 		return;
   1145 	}
   1146 
   1147 	printf("Supported file systems:");
   1148 	mutex_enter(&vfs_list_lock);
   1149 	LIST_FOREACH(v, &vfs_list, vfs_list) {
   1150 		printf(" %s", v->vfs_name);
   1151 	}
   1152 	mutex_exit(&vfs_list_lock);
   1153 	printf("\n");
   1154 }
   1155 
   1156 /*
   1157  * Mount the root file system.  If the operator didn't specify a
   1158  * file system to use, try all possible file systems until one
   1159  * succeeds.
   1160  */
   1161 int
   1162 vfs_mountroot(void)
   1163 {
   1164 	struct vfsops *v;
   1165 	int error = ENODEV;
   1166 
   1167 	if (root_device == NULL)
   1168 		panic("vfs_mountroot: root device unknown");
   1169 
   1170 	switch (device_class(root_device)) {
   1171 	case DV_IFNET:
   1172 		if (rootdev != NODEV)
   1173 			panic("vfs_mountroot: rootdev set for DV_IFNET "
   1174 			    "(0x%llx -> %llu,%llu)",
   1175 			    (unsigned long long)rootdev,
   1176 			    (unsigned long long)major(rootdev),
   1177 			    (unsigned long long)minor(rootdev));
   1178 		break;
   1179 
   1180 	case DV_DISK:
   1181 		if (rootdev == NODEV)
   1182 			panic("vfs_mountroot: rootdev not set for DV_DISK");
   1183 	        if (bdevvp(rootdev, &rootvp))
   1184 	                panic("vfs_mountroot: can't get vnode for rootdev");
   1185 		error = VOP_OPEN(rootvp, FREAD, FSCRED);
   1186 		if (error) {
   1187 			printf("vfs_mountroot: can't open root device\n");
   1188 			return (error);
   1189 		}
   1190 		break;
   1191 
   1192 	case DV_VIRTUAL:
   1193 		break;
   1194 
   1195 	default:
   1196 		printf("%s: inappropriate for root file system\n",
   1197 		    device_xname(root_device));
   1198 		return (ENODEV);
   1199 	}
   1200 
   1201 	/*
   1202 	 * If user specified a root fs type, use it.  Make sure the
   1203 	 * specified type exists and has a mount_root()
   1204 	 */
   1205 	if (strcmp(rootfstype, ROOT_FSTYPE_ANY) != 0) {
   1206 		v = vfs_getopsbyname(rootfstype);
   1207 		error = EFTYPE;
   1208 		if (v != NULL) {
   1209 			if (v->vfs_mountroot != NULL) {
   1210 				error = (v->vfs_mountroot)();
   1211 			}
   1212 			v->vfs_refcount--;
   1213 		}
   1214 		goto done;
   1215 	}
   1216 
   1217 	/*
   1218 	 * Try each file system currently configured into the kernel.
   1219 	 */
   1220 	mutex_enter(&vfs_list_lock);
   1221 	LIST_FOREACH(v, &vfs_list, vfs_list) {
   1222 		if (v->vfs_mountroot == NULL)
   1223 			continue;
   1224 #ifdef DEBUG
   1225 		aprint_normal("mountroot: trying %s...\n", v->vfs_name);
   1226 #endif
   1227 		v->vfs_refcount++;
   1228 		mutex_exit(&vfs_list_lock);
   1229 		error = (*v->vfs_mountroot)();
   1230 		mutex_enter(&vfs_list_lock);
   1231 		v->vfs_refcount--;
   1232 		if (!error) {
   1233 			aprint_normal("root file system type: %s\n",
   1234 			    v->vfs_name);
   1235 			break;
   1236 		}
   1237 	}
   1238 	mutex_exit(&vfs_list_lock);
   1239 
   1240 	if (v == NULL) {
   1241 		vfs_print_fstypes();
   1242 		printf("no file system for %s", device_xname(root_device));
   1243 		if (device_class(root_device) == DV_DISK)
   1244 			printf(" (dev 0x%llx)", (unsigned long long)rootdev);
   1245 		printf("\n");
   1246 		error = EFTYPE;
   1247 	}
   1248 
   1249 done:
   1250 	if (error && device_class(root_device) == DV_DISK) {
   1251 		VOP_CLOSE(rootvp, FREAD, FSCRED);
   1252 		vrele(rootvp);
   1253 	}
   1254 	if (error == 0) {
   1255 		mount_iterator_t *iter;
   1256 		struct mount *mp;
   1257 		extern struct cwdinfo cwdi0;
   1258 
   1259 		mountlist_iterator_init(&iter);
   1260 		mp = mountlist_iterator_next(iter);
   1261 		KASSERT(mp != NULL);
   1262 		mountlist_iterator_destroy(iter);
   1263 
   1264 		mp->mnt_flag |= MNT_ROOTFS;
   1265 		mp->mnt_op->vfs_refcount++;
   1266 
   1267 		/*
   1268 		 * Get the vnode for '/'.  Set cwdi0.cwdi_cdir to
   1269 		 * reference it, and donate it the reference grabbed
   1270 		 * with VFS_ROOT().
   1271 		 */
   1272 		error = VFS_ROOT(mp, LK_NONE, &rootvnode);
   1273 		if (error)
   1274 			panic("cannot find root vnode, error=%d", error);
   1275 		cwdi0.cwdi_cdir = rootvnode;
   1276 		cwdi0.cwdi_rdir = NULL;
   1277 
   1278 		/*
   1279 		 * Now that root is mounted, we can fixup initproc's CWD
   1280 		 * info.  All other processes are kthreads, which merely
   1281 		 * share proc0's CWD info.
   1282 		 */
   1283 		initproc->p_cwdi->cwdi_cdir = rootvnode;
   1284 		vref(initproc->p_cwdi->cwdi_cdir);
   1285 		initproc->p_cwdi->cwdi_rdir = NULL;
   1286 		/*
   1287 		 * Enable loading of modules from the filesystem
   1288 		 */
   1289 		module_load_vfs_init();
   1290 
   1291 	}
   1292 	return (error);
   1293 }
   1294 
   1295 /*
   1296  * mount_specific_key_create --
   1297  *	Create a key for subsystem mount-specific data.
   1298  */
   1299 int
   1300 mount_specific_key_create(specificdata_key_t *keyp, specificdata_dtor_t dtor)
   1301 {
   1302 
   1303 	return specificdata_key_create(mount_specificdata_domain, keyp, dtor);
   1304 }
   1305 
   1306 /*
   1307  * mount_specific_key_delete --
   1308  *	Delete a key for subsystem mount-specific data.
   1309  */
   1310 void
   1311 mount_specific_key_delete(specificdata_key_t key)
   1312 {
   1313 
   1314 	specificdata_key_delete(mount_specificdata_domain, key);
   1315 }
   1316 
   1317 /*
   1318  * mount_initspecific --
   1319  *	Initialize a mount's specificdata container.
   1320  */
   1321 void
   1322 mount_initspecific(struct mount *mp)
   1323 {
   1324 	int error __diagused;
   1325 
   1326 	error = specificdata_init(mount_specificdata_domain,
   1327 				  &mp->mnt_specdataref);
   1328 	KASSERT(error == 0);
   1329 }
   1330 
   1331 /*
   1332  * mount_finispecific --
   1333  *	Finalize a mount's specificdata container.
   1334  */
   1335 void
   1336 mount_finispecific(struct mount *mp)
   1337 {
   1338 
   1339 	specificdata_fini(mount_specificdata_domain, &mp->mnt_specdataref);
   1340 }
   1341 
   1342 /*
   1343  * mount_getspecific --
   1344  *	Return mount-specific data corresponding to the specified key.
   1345  */
   1346 void *
   1347 mount_getspecific(struct mount *mp, specificdata_key_t key)
   1348 {
   1349 
   1350 	return specificdata_getspecific(mount_specificdata_domain,
   1351 					 &mp->mnt_specdataref, key);
   1352 }
   1353 
   1354 /*
   1355  * mount_setspecific --
   1356  *	Set mount-specific data corresponding to the specified key.
   1357  */
   1358 void
   1359 mount_setspecific(struct mount *mp, specificdata_key_t key, void *data)
   1360 {
   1361 
   1362 	specificdata_setspecific(mount_specificdata_domain,
   1363 				 &mp->mnt_specdataref, key, data);
   1364 }
   1365 
   1366 /*
   1367  * Check to see if a filesystem is mounted on a block device.
   1368  */
   1369 int
   1370 vfs_mountedon(vnode_t *vp)
   1371 {
   1372 	vnode_t *vq;
   1373 	int error = 0;
   1374 
   1375 	if (vp->v_type != VBLK)
   1376 		return ENOTBLK;
   1377 	if (spec_node_getmountedfs(vp) != NULL)
   1378 		return EBUSY;
   1379 	if (spec_node_lookup_by_dev(vp->v_type, vp->v_rdev, &vq) == 0) {
   1380 		if (spec_node_getmountedfs(vq) != NULL)
   1381 			error = EBUSY;
   1382 		vrele(vq);
   1383 	}
   1384 
   1385 	return error;
   1386 }
   1387 
   1388 /*
   1389  * Check if a device pointed to by vp is mounted.
   1390  *
   1391  * Returns:
   1392  *   EINVAL	if it's not a disk
   1393  *   EBUSY	if it's a disk and mounted
   1394  *   0		if it's a disk and not mounted
   1395  */
   1396 int
   1397 rawdev_mounted(vnode_t *vp, vnode_t **bvpp)
   1398 {
   1399 	vnode_t *bvp;
   1400 	dev_t dev;
   1401 	int d_type;
   1402 
   1403 	bvp = NULL;
   1404 	d_type = D_OTHER;
   1405 
   1406 	if (iskmemvp(vp))
   1407 		return EINVAL;
   1408 
   1409 	switch (vp->v_type) {
   1410 	case VCHR: {
   1411 		const struct cdevsw *cdev;
   1412 
   1413 		dev = vp->v_rdev;
   1414 		cdev = cdevsw_lookup(dev);
   1415 		if (cdev != NULL) {
   1416 			dev_t blkdev;
   1417 
   1418 			blkdev = devsw_chr2blk(dev);
   1419 			if (blkdev != NODEV) {
   1420 				if (vfinddev(blkdev, VBLK, &bvp) != 0) {
   1421 					d_type = (cdev->d_flag & D_TYPEMASK);
   1422 					/* XXX: what if bvp disappears? */
   1423 					vrele(bvp);
   1424 				}
   1425 			}
   1426 		}
   1427 
   1428 		break;
   1429 		}
   1430 
   1431 	case VBLK: {
   1432 		const struct bdevsw *bdev;
   1433 
   1434 		dev = vp->v_rdev;
   1435 		bdev = bdevsw_lookup(dev);
   1436 		if (bdev != NULL)
   1437 			d_type = (bdev->d_flag & D_TYPEMASK);
   1438 
   1439 		bvp = vp;
   1440 
   1441 		break;
   1442 		}
   1443 
   1444 	default:
   1445 		break;
   1446 	}
   1447 
   1448 	if (d_type != D_DISK)
   1449 		return EINVAL;
   1450 
   1451 	if (bvpp != NULL)
   1452 		*bvpp = bvp;
   1453 
   1454 	/*
   1455 	 * XXX: This is bogus. We should be failing the request
   1456 	 * XXX: not only if this specific slice is mounted, but
   1457 	 * XXX: if it's on a disk with any other mounted slice.
   1458 	 */
   1459 	if (vfs_mountedon(bvp))
   1460 		return EBUSY;
   1461 
   1462 	return 0;
   1463 }
   1464 
   1465 /*
   1466  * Make a 'unique' number from a mount type name.
   1467  */
   1468 long
   1469 makefstype(const char *type)
   1470 {
   1471 	long rv;
   1472 
   1473 	for (rv = 0; *type; type++) {
   1474 		rv <<= 2;
   1475 		rv ^= *type;
   1476 	}
   1477 	return rv;
   1478 }
   1479 
   1480 static struct mountlist_entry *
   1481 mountlist_alloc(enum mountlist_type type, struct mount *mp)
   1482 {
   1483 	struct mountlist_entry *me;
   1484 
   1485 	me = kmem_zalloc(sizeof(*me), KM_SLEEP);
   1486 	me->me_mount = mp;
   1487 	me->me_type = type;
   1488 
   1489 	return me;
   1490 }
   1491 
   1492 static void
   1493 mountlist_free(struct mountlist_entry *me)
   1494 {
   1495 
   1496 	kmem_free(me, sizeof(*me));
   1497 }
   1498 
   1499 void
   1500 mountlist_iterator_init(mount_iterator_t **mip)
   1501 {
   1502 	struct mountlist_entry *me;
   1503 
   1504 	me = mountlist_alloc(ME_MARKER, NULL);
   1505 	mutex_enter(&mountlist_lock);
   1506 	TAILQ_INSERT_HEAD(&mountlist, me, me_list);
   1507 	mutex_exit(&mountlist_lock);
   1508 	*mip = (mount_iterator_t *)me;
   1509 }
   1510 
   1511 void
   1512 mountlist_iterator_destroy(mount_iterator_t *mi)
   1513 {
   1514 	struct mountlist_entry *marker = &mi->mi_entry;
   1515 
   1516 	if (marker->me_mount != NULL)
   1517 		vfs_unbusy(marker->me_mount);
   1518 
   1519 	mutex_enter(&mountlist_lock);
   1520 	TAILQ_REMOVE(&mountlist, marker, me_list);
   1521 	mutex_exit(&mountlist_lock);
   1522 
   1523 	mountlist_free(marker);
   1524 
   1525 }
   1526 
   1527 /*
   1528  * Return the next mount or NULL for this iterator.
   1529  * Mark it busy on success.
   1530  */
   1531 static inline struct mount *
   1532 _mountlist_iterator_next(mount_iterator_t *mi, bool wait)
   1533 {
   1534 	struct mountlist_entry *me, *marker = &mi->mi_entry;
   1535 	struct mount *mp;
   1536 	int error;
   1537 
   1538 	if (marker->me_mount != NULL) {
   1539 		vfs_unbusy(marker->me_mount);
   1540 		marker->me_mount = NULL;
   1541 	}
   1542 
   1543 	mutex_enter(&mountlist_lock);
   1544 	for (;;) {
   1545 		KASSERT(marker->me_type == ME_MARKER);
   1546 
   1547 		me = TAILQ_NEXT(marker, me_list);
   1548 		if (me == NULL) {
   1549 			/* End of list: keep marker and return. */
   1550 			mutex_exit(&mountlist_lock);
   1551 			return NULL;
   1552 		}
   1553 		TAILQ_REMOVE(&mountlist, marker, me_list);
   1554 		TAILQ_INSERT_AFTER(&mountlist, me, marker, me_list);
   1555 
   1556 		/* Skip other markers. */
   1557 		if (me->me_type != ME_MOUNT)
   1558 			continue;
   1559 
   1560 		/* Take an initial reference for vfs_busy() below. */
   1561 		mp = me->me_mount;
   1562 		KASSERT(mp != NULL);
   1563 		vfs_ref(mp);
   1564 		mutex_exit(&mountlist_lock);
   1565 
   1566 		/* Try to mark this mount busy and return on success. */
   1567 		if (wait)
   1568 			error = vfs_busy(mp);
   1569 		else
   1570 			error = vfs_trybusy(mp);
   1571 		if (error == 0) {
   1572 			vfs_rele(mp);
   1573 			marker->me_mount = mp;
   1574 			return mp;
   1575 		}
   1576 		vfs_rele(mp);
   1577 		mutex_enter(&mountlist_lock);
   1578 	}
   1579 }
   1580 
   1581 struct mount *
   1582 mountlist_iterator_next(mount_iterator_t *mi)
   1583 {
   1584 
   1585 	return _mountlist_iterator_next(mi, true);
   1586 }
   1587 
   1588 struct mount *
   1589 mountlist_iterator_trynext(mount_iterator_t *mi)
   1590 {
   1591 
   1592 	return _mountlist_iterator_next(mi, false);
   1593 }
   1594 
   1595 /*
   1596  * Attach new mount to the end of the mount list.
   1597  */
   1598 void
   1599 mountlist_append(struct mount *mp)
   1600 {
   1601 	struct mountlist_entry *me;
   1602 
   1603 	me = mountlist_alloc(ME_MOUNT, mp);
   1604 	mutex_enter(&mountlist_lock);
   1605 	TAILQ_INSERT_TAIL(&mountlist, me, me_list);
   1606 	mutex_exit(&mountlist_lock);
   1607 }
   1608 
   1609 /*
   1610  * Remove mount from mount list.
   1611  */void
   1612 mountlist_remove(struct mount *mp)
   1613 {
   1614 	struct mountlist_entry *me;
   1615 
   1616 	mutex_enter(&mountlist_lock);
   1617 	TAILQ_FOREACH(me, &mountlist, me_list)
   1618 		if (me->me_type == ME_MOUNT && me->me_mount == mp)
   1619 			break;
   1620 	KASSERT(me != NULL);
   1621 	TAILQ_REMOVE(&mountlist, me, me_list);
   1622 	mutex_exit(&mountlist_lock);
   1623 	mountlist_free(me);
   1624 }
   1625 
   1626 /*
   1627  * Unlocked variant to traverse the mountlist.
   1628  * To be used from DDB only.
   1629  */
   1630 struct mount *
   1631 _mountlist_next(struct mount *mp)
   1632 {
   1633 	struct mountlist_entry *me;
   1634 
   1635 	if (mp == NULL) {
   1636 		me = TAILQ_FIRST(&mountlist);
   1637 	} else {
   1638 		TAILQ_FOREACH(me, &mountlist, me_list)
   1639 			if (me->me_type == ME_MOUNT && me->me_mount == mp)
   1640 				break;
   1641 		if (me != NULL)
   1642 			me = TAILQ_NEXT(me, me_list);
   1643 	}
   1644 
   1645 	while (me != NULL && me->me_type != ME_MOUNT)
   1646 		me = TAILQ_NEXT(me, me_list);
   1647 
   1648 	return (me ? me->me_mount : NULL);
   1649 }
   1650