vfs_export.c revision 21673
1/*
2 * Copyright (c) 1989, 1993
3 *	The Regents of the University of California.  All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 *    notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 *    notice, this list of conditions and the following disclaimer in the
17 *    documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 *    must display the following acknowledgement:
20 *	This product includes software developed by the University of
21 *	California, Berkeley and its contributors.
22 * 4. 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.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS 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 REGENTS 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
36 * SUCH DAMAGE.
37 *
38 *	@(#)vfs_subr.c	8.13 (Berkeley) 4/18/94
39 * $FreeBSD: head/sys/kern/vfs_export.c 21673 1997-01-14 07:20:47Z jkh $
40 */
41
42/*
43 * External virtual filesystem routines
44 */
45#include "opt_ddb.h"
46#include "opt_devfs.h"
47
48#include <sys/param.h>
49#include <sys/systm.h>
50#include <sys/kernel.h>
51#include <sys/file.h>
52#include <sys/proc.h>
53#include <sys/mount.h>
54#include <sys/time.h>
55#include <sys/vnode.h>
56#include <sys/stat.h>
57#include <sys/namei.h>
58#include <sys/ucred.h>
59#include <sys/buf.h>
60#include <sys/errno.h>
61#include <sys/malloc.h>
62#include <sys/domain.h>
63#include <sys/mbuf.h>
64
65#include <vm/vm.h>
66#include <vm/vm_param.h>
67#include <vm/vm_object.h>
68#include <vm/vm_extern.h>
69#include <vm/vm_pager.h>
70#include <vm/vnode_pager.h>
71#include <sys/sysctl.h>
72
73#include <miscfs/specfs/specdev.h>
74
75#ifdef DDB
76extern void	printlockedvnodes __P((void));
77#endif
78extern void	vclean __P((struct vnode *vp, int flags));
79extern void	vfs_unmountroot __P((struct mount *rootfs));
80
81enum vtype iftovt_tab[16] = {
82	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
83	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
84};
85int vttoif_tab[9] = {
86	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
87	S_IFSOCK, S_IFIFO, S_IFMT,
88};
89
90/*
91 * Insq/Remq for the vnode usage lists.
92 */
93#define	bufinsvn(bp, dp)	LIST_INSERT_HEAD(dp, bp, b_vnbufs)
94#define	bufremvn(bp) {  \
95	LIST_REMOVE(bp, b_vnbufs); \
96	(bp)->b_vnbufs.le_next = NOLIST; \
97}
98
99TAILQ_HEAD(freelst, vnode) vnode_free_list;	/* vnode free list */
100static u_long freevnodes = 0;
101
102struct mntlist mountlist;	/* mounted filesystem list */
103
104int desiredvnodes;
105SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, &desiredvnodes, 0, "");
106
107static void	vfs_free_addrlist __P((struct netexport *nep));
108static int	vfs_free_netcred __P((struct radix_node *rn, void *w));
109static int	vfs_hang_addrlist __P((struct mount *mp, struct netexport *nep,
110				       struct export_args *argp));
111
112/*
113 * Initialize the vnode management data structures.
114 */
115void
116vntblinit()
117{
118	desiredvnodes = maxproc + vm_object_cache_max + extravnodes;
119
120	TAILQ_INIT(&vnode_free_list);
121	CIRCLEQ_INIT(&mountlist);
122}
123
124/*
125 * Lock a filesystem.
126 * Used to prevent access to it while mounting and unmounting.
127 */
128int
129vfs_lock(mp)
130	register struct mount *mp;
131{
132
133	while (mp->mnt_flag & MNT_MLOCK) {
134		mp->mnt_flag |= MNT_MWAIT;
135		(void) tsleep((caddr_t) mp, PVFS, "vfslck", 0);
136	}
137	mp->mnt_flag |= MNT_MLOCK;
138	return (0);
139}
140
141/*
142 * Unlock a locked filesystem.
143 * Panic if filesystem is not locked.
144 */
145void
146vfs_unlock(mp)
147	register struct mount *mp;
148{
149
150	if ((mp->mnt_flag & MNT_MLOCK) == 0)
151		panic("vfs_unlock: not locked");
152	mp->mnt_flag &= ~MNT_MLOCK;
153	if (mp->mnt_flag & MNT_MWAIT) {
154		mp->mnt_flag &= ~MNT_MWAIT;
155		wakeup((caddr_t) mp);
156	}
157}
158
159/*
160 * Mark a mount point as busy.
161 * Used to synchronize access and to delay unmounting.
162 */
163int
164vfs_busy(mp)
165	register struct mount *mp;
166{
167
168	while (mp->mnt_flag & MNT_MPBUSY) {
169		mp->mnt_flag |= MNT_MPWANT;
170		(void) tsleep((caddr_t) &mp->mnt_flag, PVFS, "vfsbsy", 0);
171	}
172	if (mp->mnt_flag & MNT_UNMOUNT)
173		return (1);
174	mp->mnt_flag |= MNT_MPBUSY;
175	return (0);
176}
177
178/*
179 * Free a busy filesystem.
180 * Panic if filesystem is not busy.
181 */
182void
183vfs_unbusy(mp)
184	register struct mount *mp;
185{
186
187	if ((mp->mnt_flag & MNT_MPBUSY) == 0)
188		panic("vfs_unbusy: not busy");
189	mp->mnt_flag &= ~MNT_MPBUSY;
190	if (mp->mnt_flag & MNT_MPWANT) {
191		mp->mnt_flag &= ~MNT_MPWANT;
192		wakeup((caddr_t) &mp->mnt_flag);
193	}
194}
195
196void
197vfs_unmountroot(struct mount *rootfs)
198{
199	struct mount *mp = rootfs;
200	int error;
201
202	if (vfs_busy(mp)) {
203		printf("failed to unmount root\n");
204		return;
205	}
206	mp->mnt_flag |= MNT_UNMOUNT;
207	if ((error = vfs_lock(mp))) {
208		printf("lock of root filesystem failed (%d)\n", error);
209		return;
210	}
211	vnode_pager_umount(mp);	/* release cached vnodes */
212	cache_purgevfs(mp);	/* remove cache entries for this file sys */
213
214	if ((error = VFS_SYNC(mp, MNT_WAIT, initproc->p_ucred, initproc)))
215		printf("sync of root filesystem failed (%d)\n", error);
216
217	if ((error = VFS_UNMOUNT(mp, MNT_FORCE, initproc))) {
218		printf("unmount of root filesystem failed (");
219		if (error == EBUSY)
220			printf("BUSY)\n");
221		else
222			printf("%d)\n", error);
223	}
224	mp->mnt_flag &= ~MNT_UNMOUNT;
225	vfs_unbusy(mp);
226}
227
228/*
229 * Unmount all filesystems.  Should only be called by halt().
230 */
231void
232vfs_unmountall()
233{
234	struct mount *mp, *nmp, *rootfs = NULL;
235	int error;
236
237	/* unmount all but rootfs */
238	for (mp = mountlist.cqh_last; mp != (void *)&mountlist; mp = nmp) {
239		nmp = mp->mnt_list.cqe_prev;
240
241		if (mp->mnt_flag & MNT_ROOTFS) {
242			rootfs = mp;
243			continue;
244		}
245		error = dounmount(mp, MNT_FORCE, initproc);
246		if (error) {
247			printf("unmount of %s failed (", mp->mnt_stat.f_mntonname);
248			if (error == EBUSY)
249				printf("BUSY)\n");
250			else
251				printf("%d)\n", error);
252		}
253	}
254
255	/* and finally... */
256	if (rootfs) {
257		vfs_unmountroot(rootfs);
258	} else {
259		printf("no root filesystem\n");
260	}
261}
262
263/*
264 * Lookup a mount point by filesystem identifier.
265 */
266struct mount *
267getvfs(fsid)
268	fsid_t *fsid;
269{
270	register struct mount *mp;
271
272	for (mp = mountlist.cqh_first; mp != (void *)&mountlist;
273	    mp = mp->mnt_list.cqe_next) {
274		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
275		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1])
276			return (mp);
277	}
278	return ((struct mount *) 0);
279}
280
281/*
282 * Get a new unique fsid
283 */
284void
285getnewfsid(mp, mtype)
286	struct mount *mp;
287	int mtype;
288{
289	static u_short xxxfs_mntid;
290
291	fsid_t tfsid;
292
293	mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev + mtype, 0);
294	mp->mnt_stat.f_fsid.val[1] = mtype;
295	if (xxxfs_mntid == 0)
296		++xxxfs_mntid;
297	tfsid.val[0] = makedev(nblkdev + mtype, xxxfs_mntid);
298	tfsid.val[1] = mtype;
299	if (mountlist.cqh_first != (void *)&mountlist) {
300		while (getvfs(&tfsid)) {
301			tfsid.val[0]++;
302			xxxfs_mntid++;
303		}
304	}
305	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
306}
307
308/*
309 * Set vnode attributes to VNOVAL
310 */
311void
312vattr_null(vap)
313	register struct vattr *vap;
314{
315
316	vap->va_type = VNON;
317	vap->va_size = VNOVAL;
318	vap->va_bytes = VNOVAL;
319	vap->va_mode = vap->va_nlink = vap->va_uid = vap->va_gid =
320	    vap->va_fsid = vap->va_fileid =
321	    vap->va_blocksize = vap->va_rdev =
322	    vap->va_atime.tv_sec = vap->va_atime.tv_nsec =
323	    vap->va_mtime.tv_sec = vap->va_mtime.tv_nsec =
324	    vap->va_ctime.tv_sec = vap->va_ctime.tv_nsec =
325	    vap->va_flags = vap->va_gen = VNOVAL;
326	vap->va_vaflags = 0;
327}
328
329/*
330 * Routines having to do with the management of the vnode table.
331 */
332extern vop_t **dead_vnodeop_p;
333
334/*
335 * Return the next vnode from the free list.
336 */
337int
338getnewvnode(tag, mp, vops, vpp)
339	enum vtagtype tag;
340	struct mount *mp;
341	vop_t **vops;
342	struct vnode **vpp;
343{
344	register struct vnode *vp;
345
346retry:
347	vp = vnode_free_list.tqh_first;
348	/*
349	 * we allocate a new vnode if
350	 * 	1. we don't have any free
351	 *		Pretty obvious, we actually used to panic, but that
352	 *		is a silly thing to do.
353	 *	2. we havn't filled our pool yet
354	 *		We don't want to trash the incore (VM-)vnodecache.
355	 *	3. if less that 1/4th of our vnodes are free.
356	 *		We don't want to trash the namei cache either.
357	 */
358	if (freevnodes < (numvnodes >> 2) ||
359	    numvnodes < desiredvnodes ||
360	    vp == NULL) {
361		vp = (struct vnode *) malloc((u_long) sizeof *vp,
362		    M_VNODE, M_WAITOK);
363		bzero((char *) vp, sizeof *vp);
364		numvnodes++;
365	} else {
366		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
367		if (vp->v_usage > 0) {
368			--vp->v_usage;
369			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
370			goto retry;
371		}
372		freevnodes--;
373		if (vp->v_usecount)
374			panic("free vnode isn't");
375
376		/* see comment on why 0xdeadb is set at end of vgone (below) */
377		vp->v_freelist.tqe_prev = (struct vnode **) 0xdeadb;
378		vp->v_lease = NULL;
379		if (vp->v_type != VBAD)
380			vgone(vp);
381
382#ifdef DIAGNOSTIC
383		{
384			int s;
385
386			if (vp->v_data)
387				panic("cleaned vnode isn't");
388			s = splbio();
389			if (vp->v_numoutput)
390				panic("Clean vnode has pending I/O's");
391			splx(s);
392		}
393#endif
394		vp->v_flag = 0;
395		vp->v_lastr = 0;
396		vp->v_lastw = 0;
397		vp->v_lasta = 0;
398		vp->v_cstart = 0;
399		vp->v_clen = 0;
400		vp->v_socket = 0;
401		vp->v_writecount = 0;	/* XXX */
402		vp->v_usage = 0;
403	}
404	vp->v_type = VNON;
405	cache_purge(vp);
406	vp->v_tag = tag;
407	vp->v_op = vops;
408	insmntque(vp, mp);
409	*vpp = vp;
410	vp->v_usecount = 1;
411	vp->v_data = 0;
412	return (0);
413}
414
415/*
416 * Move a vnode from one mount queue to another.
417 */
418void
419insmntque(vp, mp)
420	register struct vnode *vp;
421	register struct mount *mp;
422{
423
424	/*
425	 * Delete from old mount point vnode list, if on one.
426	 */
427	if (vp->v_mount != NULL)
428		LIST_REMOVE(vp, v_mntvnodes);
429	/*
430	 * Insert into list of vnodes for the new mount point, if available.
431	 */
432	if ((vp->v_mount = mp) == NULL)
433		return;
434	LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes);
435}
436
437/*
438 * Update outstanding I/O count and do wakeup if requested.
439 */
440void
441vwakeup(bp)
442	register struct buf *bp;
443{
444	register struct vnode *vp;
445
446	bp->b_flags &= ~B_WRITEINPROG;
447	if ((vp = bp->b_vp)) {
448		vp->v_numoutput--;
449		if (vp->v_numoutput < 0)
450			panic("vwakeup: neg numoutput");
451		if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
452			vp->v_flag &= ~VBWAIT;
453			wakeup((caddr_t) &vp->v_numoutput);
454		}
455	}
456}
457
458/*
459 * Flush out and invalidate all buffers associated with a vnode.
460 * Called with the underlying object locked.
461 */
462int
463vinvalbuf(vp, flags, cred, p, slpflag, slptimeo)
464	register struct vnode *vp;
465	int flags;
466	struct ucred *cred;
467	struct proc *p;
468	int slpflag, slptimeo;
469{
470	register struct buf *bp;
471	struct buf *nbp, *blist;
472	int s, error;
473	vm_object_t object;
474
475	if (flags & V_SAVE) {
476		if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, p)))
477			return (error);
478		if (vp->v_dirtyblkhd.lh_first != NULL)
479			panic("vinvalbuf: dirty bufs");
480	}
481
482	s = splbio();
483	for (;;) {
484		if ((blist = vp->v_cleanblkhd.lh_first) && (flags & V_SAVEMETA))
485			while (blist && blist->b_lblkno < 0)
486				blist = blist->b_vnbufs.le_next;
487		if (!blist && (blist = vp->v_dirtyblkhd.lh_first) &&
488		    (flags & V_SAVEMETA))
489			while (blist && blist->b_lblkno < 0)
490				blist = blist->b_vnbufs.le_next;
491		if (!blist)
492			break;
493
494		for (bp = blist; bp; bp = nbp) {
495			nbp = bp->b_vnbufs.le_next;
496			if ((flags & V_SAVEMETA) && bp->b_lblkno < 0)
497				continue;
498			if (bp->b_flags & B_BUSY) {
499				bp->b_flags |= B_WANTED;
500				error = tsleep((caddr_t) bp,
501				    slpflag | (PRIBIO + 1), "vinvalbuf",
502				    slptimeo);
503				splx(s);
504				if (error)
505					return (error);
506				break;
507			}
508			bremfree(bp);
509			bp->b_flags |= B_BUSY;
510			/*
511			 * XXX Since there are no node locks for NFS, I
512			 * believe there is a slight chance that a delayed
513			 * write will occur while sleeping just above, so
514			 * check for it.
515			 */
516			if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) {
517				(void) VOP_BWRITE(bp);
518				break;
519			}
520			bp->b_flags |= (B_INVAL|B_NOCACHE|B_RELBUF);
521			brelse(bp);
522		}
523	}
524	splx(s);
525
526	s = splbio();
527	while (vp->v_numoutput > 0) {
528		vp->v_flag |= VBWAIT;
529		tsleep(&vp->v_numoutput, PVM, "vnvlbv", 0);
530	}
531	splx(s);
532
533	/*
534	 * Destroy the copy in the VM cache, too.
535	 */
536	object = vp->v_object;
537	if (object != NULL) {
538		vm_object_page_remove(object, 0, object->size,
539		    (flags & V_SAVE) ? TRUE : FALSE);
540	}
541	if (!(flags & V_SAVEMETA) &&
542	    (vp->v_dirtyblkhd.lh_first || vp->v_cleanblkhd.lh_first))
543		panic("vinvalbuf: flush failed");
544	return (0);
545}
546
547/*
548 * Associate a buffer with a vnode.
549 */
550void
551bgetvp(vp, bp)
552	register struct vnode *vp;
553	register struct buf *bp;
554{
555	int s;
556
557	if (bp->b_vp)
558		panic("bgetvp: not free");
559	VHOLD(vp);
560	bp->b_vp = vp;
561	if (vp->v_type == VBLK || vp->v_type == VCHR)
562		bp->b_dev = vp->v_rdev;
563	else
564		bp->b_dev = NODEV;
565	/*
566	 * Insert onto list for new vnode.
567	 */
568	s = splbio();
569	bufinsvn(bp, &vp->v_cleanblkhd);
570	splx(s);
571}
572
573/*
574 * Disassociate a buffer from a vnode.
575 */
576void
577brelvp(bp)
578	register struct buf *bp;
579{
580	struct vnode *vp;
581	int s;
582
583	if (bp->b_vp == (struct vnode *) 0)
584		panic("brelvp: NULL");
585	/*
586	 * Delete from old vnode list, if on one.
587	 */
588	s = splbio();
589	if (bp->b_vnbufs.le_next != NOLIST)
590		bufremvn(bp);
591	splx(s);
592
593	vp = bp->b_vp;
594	bp->b_vp = (struct vnode *) 0;
595	HOLDRELE(vp);
596}
597
598/*
599 * Associate a p-buffer with a vnode.
600 */
601void
602pbgetvp(vp, bp)
603	register struct vnode *vp;
604	register struct buf *bp;
605{
606	if (bp->b_vp)
607		panic("pbgetvp: not free");
608	VHOLD(vp);
609	bp->b_vp = vp;
610	if (vp->v_type == VBLK || vp->v_type == VCHR)
611		bp->b_dev = vp->v_rdev;
612	else
613		bp->b_dev = NODEV;
614}
615
616/*
617 * Disassociate a p-buffer from a vnode.
618 */
619void
620pbrelvp(bp)
621	register struct buf *bp;
622{
623	struct vnode *vp;
624
625	if (bp->b_vp == (struct vnode *) 0)
626		panic("brelvp: NULL");
627
628	vp = bp->b_vp;
629	bp->b_vp = (struct vnode *) 0;
630	HOLDRELE(vp);
631}
632
633/*
634 * Reassign a buffer from one vnode to another.
635 * Used to assign file specific control information
636 * (indirect blocks) to the vnode to which they belong.
637 */
638void
639reassignbuf(bp, newvp)
640	register struct buf *bp;
641	register struct vnode *newvp;
642{
643	int s;
644
645	if (newvp == NULL) {
646		printf("reassignbuf: NULL");
647		return;
648	}
649
650	s = splbio();
651	/*
652	 * Delete from old vnode list, if on one.
653	 */
654	if (bp->b_vnbufs.le_next != NOLIST)
655		bufremvn(bp);
656	/*
657	 * If dirty, put on list of dirty buffers; otherwise insert onto list
658	 * of clean buffers.
659	 */
660	if (bp->b_flags & B_DELWRI) {
661		struct buf *tbp;
662
663		tbp = newvp->v_dirtyblkhd.lh_first;
664		if (!tbp || (tbp->b_lblkno > bp->b_lblkno)) {
665			bufinsvn(bp, &newvp->v_dirtyblkhd);
666		} else {
667			while (tbp->b_vnbufs.le_next &&
668				(tbp->b_vnbufs.le_next->b_lblkno < bp->b_lblkno)) {
669				tbp = tbp->b_vnbufs.le_next;
670			}
671			LIST_INSERT_AFTER(tbp, bp, b_vnbufs);
672		}
673	} else {
674		bufinsvn(bp, &newvp->v_cleanblkhd);
675	}
676	splx(s);
677}
678
679#ifndef DEVFS_ROOT
680/*
681 * Create a vnode for a block device.
682 * Used for root filesystem, argdev, and swap areas.
683 * Also used for memory file system special devices.
684 */
685int
686bdevvp(dev, vpp)
687	dev_t dev;
688	struct vnode **vpp;
689{
690	register struct vnode *vp;
691	struct vnode *nvp;
692	int error;
693
694	if (dev == NODEV)
695		return (0);
696	error = getnewvnode(VT_NON, (struct mount *) 0, spec_vnodeop_p, &nvp);
697	if (error) {
698		*vpp = 0;
699		return (error);
700	}
701	vp = nvp;
702	vp->v_type = VBLK;
703	if ((nvp = checkalias(vp, dev, (struct mount *) 0))) {
704		vput(vp);
705		vp = nvp;
706	}
707	*vpp = vp;
708	return (0);
709}
710#endif /* !DEVFS_ROOT */
711
712/*
713 * Check to see if the new vnode represents a special device
714 * for which we already have a vnode (either because of
715 * bdevvp() or because of a different vnode representing
716 * the same block device). If such an alias exists, deallocate
717 * the existing contents and return the aliased vnode. The
718 * caller is responsible for filling it with its new contents.
719 */
720struct vnode *
721checkalias(nvp, nvp_rdev, mp)
722	register struct vnode *nvp;
723	dev_t nvp_rdev;
724	struct mount *mp;
725{
726	register struct vnode *vp;
727	struct vnode **vpp;
728
729	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
730		return (NULLVP);
731
732	vpp = &speclisth[SPECHASH(nvp_rdev)];
733loop:
734	for (vp = *vpp; vp; vp = vp->v_specnext) {
735		if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type)
736			continue;
737		/*
738		 * Alias, but not in use, so flush it out.
739		 */
740		if (vp->v_usecount == 0) {
741			vgone(vp);
742			goto loop;
743		}
744		if (vget(vp, 1))
745			goto loop;
746		break;
747	}
748
749	if (vp == NULL || vp->v_tag != VT_NON) {
750		MALLOC(nvp->v_specinfo, struct specinfo *,
751		    sizeof(struct specinfo), M_VNODE, M_WAITOK);
752		nvp->v_rdev = nvp_rdev;
753		nvp->v_hashchain = vpp;
754		nvp->v_specnext = *vpp;
755		nvp->v_specflags = 0;
756		*vpp = nvp;
757		if (vp != NULL) {
758			nvp->v_flag |= VALIASED;
759			vp->v_flag |= VALIASED;
760			vput(vp);
761		}
762		return (NULLVP);
763	}
764	VOP_UNLOCK(vp);
765	vclean(vp, 0);
766	vp->v_op = nvp->v_op;
767	vp->v_tag = nvp->v_tag;
768	nvp->v_type = VNON;
769	insmntque(vp, mp);
770	return (vp);
771}
772
773/*
774 * Grab a particular vnode from the free list, increment its
775 * reference count and lock it. The vnode lock bit is set the
776 * vnode is being eliminated in vgone. The process is awakened
777 * when the transition is completed, and an error returned to
778 * indicate that the vnode is no longer usable (possibly having
779 * been changed to a new file system type).
780 */
781int
782vget(vp, lockflag)
783	register struct vnode *vp;
784	int lockflag;
785{
786
787	/*
788	 * If the vnode is in the process of being cleaned out for another
789	 * use, we wait for the cleaning to finish and then return failure.
790	 * Cleaning is determined either by checking that the VXLOCK flag is
791	 * set, or that the use count is zero with the back pointer set to
792	 * show that it has been removed from the free list by getnewvnode.
793	 * The VXLOCK flag may not have been set yet because vclean is blocked
794	 * in the VOP_LOCK call waiting for the VOP_INACTIVE to complete.
795	 */
796	if ((vp->v_flag & VXLOCK) ||
797	    (vp->v_usecount == 0 &&
798		vp->v_freelist.tqe_prev == (struct vnode **) 0xdeadb)) {
799		vp->v_flag |= VXWANT;
800		(void) tsleep((caddr_t) vp, PINOD, "vget", 0);
801		return (1);
802	}
803	if (vp->v_usecount == 0) {
804		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
805		freevnodes--;
806	}
807	vp->v_usecount++;
808
809	/*
810	 * Create the VM object, if needed
811	 */
812	if ((vp->v_type == VREG) &&
813		((vp->v_object == NULL) ||
814			(vp->v_object->flags & OBJ_VFS_REF) == 0)) {
815		vfs_object_create(vp, curproc, curproc->p_ucred, 0);
816	}
817	if (lockflag)
818		VOP_LOCK(vp);
819
820	return (0);
821}
822
823/*
824 * Vnode reference, just increment the count
825 */
826void
827vref(vp)
828	struct vnode *vp;
829{
830	if (vp->v_usecount <= 0)
831		panic("vref used where vget required");
832
833	vp->v_usecount++;
834
835	if ((vp->v_type == VREG) &&
836		((vp->v_object == NULL) ||
837			((vp->v_object->flags & OBJ_VFS_REF) == 0)) ) {
838		/*
839		 * We need to lock to VP during the time that
840		 * the object is created.  This is necessary to
841		 * keep the system from re-entrantly doing it
842		 * multiple times.
843		 */
844		vfs_object_create(vp, curproc, curproc->p_ucred, 0);
845	}
846}
847
848/*
849 * vput(), just unlock and vrele()
850 */
851void
852vput(vp)
853	register struct vnode *vp;
854{
855	VOP_UNLOCK(vp);
856	vrele(vp);
857}
858
859/*
860 * Vnode release.
861 * If count drops to zero, call inactive routine and return to freelist.
862 */
863void
864vrele(vp)
865	register struct vnode *vp;
866{
867
868#ifdef DIAGNOSTIC
869	if (vp == NULL)
870		panic("vrele: null vp");
871#endif
872
873	vp->v_usecount--;
874
875	if ((vp->v_usecount == 1) &&
876		vp->v_object &&
877		(vp->v_object->flags & OBJ_VFS_REF)) {
878		vp->v_object->flags &= ~OBJ_VFS_REF;
879		vm_object_deallocate(vp->v_object);
880		return;
881	}
882
883	if (vp->v_usecount > 0)
884		return;
885
886	if (vp->v_usecount < 0) {
887#ifdef DIAGNOSTIC
888		vprint("vrele: negative ref count", vp);
889#endif
890		panic("vrele: negative reference cnt");
891	}
892	if (vp->v_flag & VAGE) {
893		if(vp->v_tag != VT_TFS)
894			TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
895		vp->v_flag &= ~VAGE;
896		vp->v_usage = 0;
897	} else {
898		if(vp->v_tag != VT_TFS)
899			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
900	}
901	freevnodes++;
902
903	VOP_INACTIVE(vp);
904}
905
906#ifdef DIAGNOSTIC
907/*
908 * Page or buffer structure gets a reference.
909 */
910void
911vhold(vp)
912	register struct vnode *vp;
913{
914
915	vp->v_holdcnt++;
916}
917
918/*
919 * Page or buffer structure frees a reference.
920 */
921void
922holdrele(vp)
923	register struct vnode *vp;
924{
925
926	if (vp->v_holdcnt <= 0)
927		panic("holdrele: holdcnt");
928	vp->v_holdcnt--;
929}
930#endif /* DIAGNOSTIC */
931
932/*
933 * Remove any vnodes in the vnode table belonging to mount point mp.
934 *
935 * If MNT_NOFORCE is specified, there should not be any active ones,
936 * return error if any are found (nb: this is a user error, not a
937 * system error). If MNT_FORCE is specified, detach any active vnodes
938 * that are found.
939 */
940#ifdef DIAGNOSTIC
941static int busyprt = 0;		/* print out busy vnodes */
942SYSCTL_INT(_debug, 1, busyprt, CTLFLAG_RW, &busyprt, 0, "");
943#endif
944
945int
946vflush(mp, skipvp, flags)
947	struct mount *mp;
948	struct vnode *skipvp;
949	int flags;
950{
951	register struct vnode *vp, *nvp;
952	int busy = 0;
953
954	if ((mp->mnt_flag & MNT_MPBUSY) == 0)
955		panic("vflush: not busy");
956loop:
957	for (vp = mp->mnt_vnodelist.lh_first; vp; vp = nvp) {
958		/*
959		 * Make sure this vnode wasn't reclaimed in getnewvnode().
960		 * Start over if it has (it won't be on the list anymore).
961		 */
962		if (vp->v_mount != mp)
963			goto loop;
964		nvp = vp->v_mntvnodes.le_next;
965		/*
966		 * Skip over a selected vnode.
967		 */
968		if (vp == skipvp)
969			continue;
970		/*
971		 * Skip over a vnodes marked VSYSTEM.
972		 */
973		if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM))
974			continue;
975		/*
976		 * If WRITECLOSE is set, only flush out regular file vnodes
977		 * open for writing.
978		 */
979		if ((flags & WRITECLOSE) &&
980		    (vp->v_writecount == 0 || vp->v_type != VREG))
981			continue;
982
983		if (vp->v_object && (vp->v_object->flags & OBJ_VFS_REF)) {
984			vm_object_reference(vp->v_object);
985			pager_cache(vp->v_object, FALSE);
986			vp->v_object->flags &= ~OBJ_VFS_REF;
987			vm_object_deallocate(vp->v_object);
988		}
989
990		/*
991		 * With v_usecount == 0, all we need to do is clear out the
992		 * vnode data structures and we are done.
993		 */
994		if (vp->v_usecount == 0) {
995			vgone(vp);
996			continue;
997		}
998
999		/*
1000		 * If FORCECLOSE is set, forcibly close the vnode. For block
1001		 * or character devices, revert to an anonymous device. For
1002		 * all other files, just kill them.
1003		 */
1004		if (flags & FORCECLOSE) {
1005			if (vp->v_type != VBLK && vp->v_type != VCHR) {
1006				vgone(vp);
1007			} else {
1008				vclean(vp, 0);
1009				vp->v_op = spec_vnodeop_p;
1010				insmntque(vp, (struct mount *) 0);
1011			}
1012			continue;
1013		}
1014#ifdef DIAGNOSTIC
1015		if (busyprt)
1016			vprint("vflush: busy vnode", vp);
1017#endif
1018		busy++;
1019	}
1020	if (busy)
1021		return (EBUSY);
1022	return (0);
1023}
1024
1025/*
1026 * Disassociate the underlying file system from a vnode.
1027 */
1028void
1029vclean(struct vnode *vp, int flags)
1030{
1031	int active;
1032
1033	/*
1034	 * Check to see if the vnode is in use. If so we have to reference it
1035	 * before we clean it out so that its count cannot fall to zero and
1036	 * generate a race against ourselves to recycle it.
1037	 */
1038	if ((active = vp->v_usecount))
1039		VREF(vp);
1040	/*
1041	 * Even if the count is zero, the VOP_INACTIVE routine may still have
1042	 * the object locked while it cleans it out. The VOP_LOCK ensures that
1043	 * the VOP_INACTIVE routine is done with its work. For active vnodes,
1044	 * it ensures that no other activity can occur while the underlying
1045	 * object is being cleaned out.
1046	 */
1047	VOP_LOCK(vp);
1048	/*
1049	 * Prevent the vnode from being recycled or brought into use while we
1050	 * clean it out.
1051	 */
1052	if (vp->v_flag & VXLOCK)
1053		panic("vclean: deadlock");
1054	vp->v_flag |= VXLOCK;
1055	/*
1056	 * Clean out any buffers associated with the vnode.
1057	 */
1058	if (flags & DOCLOSE)
1059		vinvalbuf(vp, V_SAVE, NOCRED, NULL, 0, 0);
1060	/*
1061	 * Any other processes trying to obtain this lock must first wait for
1062	 * VXLOCK to clear, then call the new lock operation.
1063	 */
1064	VOP_UNLOCK(vp);
1065	/*
1066	 * If purging an active vnode, it must be closed and deactivated
1067	 * before being reclaimed.
1068	 */
1069	if (active) {
1070		if (flags & DOCLOSE)
1071			VOP_CLOSE(vp, FNONBLOCK, NOCRED, NULL);
1072		VOP_INACTIVE(vp);
1073	}
1074	/*
1075	 * Reclaim the vnode.
1076	 */
1077	if (VOP_RECLAIM(vp))
1078		panic("vclean: cannot reclaim");
1079	if (active)
1080		vrele(vp);
1081
1082	/*
1083	 * Done with purge, notify sleepers of the grim news.
1084	 */
1085	vp->v_op = dead_vnodeop_p;
1086	vp->v_tag = VT_NON;
1087	vp->v_flag &= ~VXLOCK;
1088	if (vp->v_flag & VXWANT) {
1089		vp->v_flag &= ~VXWANT;
1090		wakeup((caddr_t) vp);
1091	}
1092}
1093
1094/*
1095 * Eliminate all activity associated with  the requested vnode
1096 * and with all vnodes aliased to the requested vnode.
1097 */
1098void
1099vgoneall(vp)
1100	register struct vnode *vp;
1101{
1102	register struct vnode *vq;
1103
1104	if (vp->v_flag & VALIASED) {
1105		/*
1106		 * If a vgone (or vclean) is already in progress, wait until
1107		 * it is done and return.
1108		 */
1109		if (vp->v_flag & VXLOCK) {
1110			vp->v_flag |= VXWANT;
1111			(void) tsleep((caddr_t) vp, PINOD, "vgall", 0);
1112			return;
1113		}
1114		/*
1115		 * Ensure that vp will not be vgone'd while we are eliminating
1116		 * its aliases.
1117		 */
1118		vp->v_flag |= VXLOCK;
1119		while (vp->v_flag & VALIASED) {
1120			for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1121				if (vq->v_rdev != vp->v_rdev ||
1122				    vq->v_type != vp->v_type || vp == vq)
1123					continue;
1124				vgone(vq);
1125				break;
1126			}
1127		}
1128		/*
1129		 * Remove the lock so that vgone below will really eliminate
1130		 * the vnode after which time vgone will awaken any sleepers.
1131		 */
1132		vp->v_flag &= ~VXLOCK;
1133	}
1134	vgone(vp);
1135}
1136
1137/*
1138 * Eliminate all activity associated with a vnode
1139 * in preparation for reuse.
1140 */
1141void
1142vgone(vp)
1143	register struct vnode *vp;
1144{
1145	register struct vnode *vq;
1146	struct vnode *vx;
1147
1148	/*
1149	 * If a vgone (or vclean) is already in progress, wait until it is
1150	 * done and return.
1151	 */
1152	if (vp->v_flag & VXLOCK) {
1153		vp->v_flag |= VXWANT;
1154		(void) tsleep((caddr_t) vp, PINOD, "vgone", 0);
1155		return;
1156	}
1157
1158	if (vp->v_object) {
1159		vp->v_object->flags |= OBJ_VNODE_GONE;
1160	}
1161
1162	/*
1163	 * Clean out the filesystem specific data.
1164	 */
1165	vclean(vp, DOCLOSE);
1166	/*
1167	 * Delete from old mount point vnode list, if on one.
1168	 */
1169	if (vp->v_mount != NULL) {
1170		LIST_REMOVE(vp, v_mntvnodes);
1171		vp->v_mount = NULL;
1172	}
1173	/*
1174	 * If special device, remove it from special device alias list.
1175	 */
1176	if (vp->v_type == VBLK || vp->v_type == VCHR) {
1177		if (*vp->v_hashchain == vp) {
1178			*vp->v_hashchain = vp->v_specnext;
1179		} else {
1180			for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1181				if (vq->v_specnext != vp)
1182					continue;
1183				vq->v_specnext = vp->v_specnext;
1184				break;
1185			}
1186			if (vq == NULL)
1187				panic("missing bdev");
1188		}
1189		if (vp->v_flag & VALIASED) {
1190			vx = NULL;
1191			for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1192				if (vq->v_rdev != vp->v_rdev ||
1193				    vq->v_type != vp->v_type)
1194					continue;
1195				if (vx)
1196					break;
1197				vx = vq;
1198			}
1199			if (vx == NULL)
1200				panic("missing alias");
1201			if (vq == NULL)
1202				vx->v_flag &= ~VALIASED;
1203			vp->v_flag &= ~VALIASED;
1204		}
1205		FREE(vp->v_specinfo, M_VNODE);
1206		vp->v_specinfo = NULL;
1207	}
1208	/*
1209	 * If it is on the freelist and not already at the head, move it to
1210	 * the head of the list. The test of the back pointer and the
1211	 * reference count of zero is because it will be removed from the free
1212	 * list by getnewvnode, but will not have its reference count
1213	 * incremented until after calling vgone. If the reference count were
1214	 * incremented first, vgone would (incorrectly) try to close the
1215	 * previous instance of the underlying object. So, the back pointer is
1216	 * explicitly set to `0xdeadb' in getnewvnode after removing it from
1217	 * the freelist to ensure that we do not try to move it here.
1218	 */
1219	if (vp->v_usecount == 0 &&
1220	    vp->v_freelist.tqe_prev != (struct vnode **) 0xdeadb &&
1221	    vnode_free_list.tqh_first != vp) {
1222		if(vp->v_tag != VT_TFS) {
1223			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1224			TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
1225		}
1226	}
1227	vp->v_type = VBAD;
1228}
1229
1230/*
1231 * Lookup a vnode by device number.
1232 */
1233int
1234vfinddev(dev, type, vpp)
1235	dev_t dev;
1236	enum vtype type;
1237	struct vnode **vpp;
1238{
1239	register struct vnode *vp;
1240
1241	for (vp = speclisth[SPECHASH(dev)]; vp; vp = vp->v_specnext) {
1242		if (dev != vp->v_rdev || type != vp->v_type)
1243			continue;
1244		*vpp = vp;
1245		return (1);
1246	}
1247	return (0);
1248}
1249
1250/*
1251 * Calculate the total number of references to a special device.
1252 */
1253int
1254vcount(vp)
1255	register struct vnode *vp;
1256{
1257	register struct vnode *vq, *vnext;
1258	int count;
1259
1260loop:
1261	if ((vp->v_flag & VALIASED) == 0)
1262		return (vp->v_usecount);
1263	for (count = 0, vq = *vp->v_hashchain; vq; vq = vnext) {
1264		vnext = vq->v_specnext;
1265		if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type)
1266			continue;
1267		/*
1268		 * Alias, but not in use, so flush it out.
1269		 */
1270		if (vq->v_usecount == 0 && vq != vp) {
1271			vgone(vq);
1272			goto loop;
1273		}
1274		count += vq->v_usecount;
1275	}
1276	return (count);
1277}
1278
1279/*
1280 * Print out a description of a vnode.
1281 */
1282static char *typename[] =
1283{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1284
1285void
1286vprint(label, vp)
1287	char *label;
1288	register struct vnode *vp;
1289{
1290	char buf[64];
1291
1292	if (label != NULL)
1293		printf("%s: ", label);
1294	printf("type %s, usecount %d, writecount %d, refcount %ld,",
1295	    typename[vp->v_type], vp->v_usecount, vp->v_writecount,
1296	    vp->v_holdcnt);
1297	buf[0] = '\0';
1298	if (vp->v_flag & VROOT)
1299		strcat(buf, "|VROOT");
1300	if (vp->v_flag & VTEXT)
1301		strcat(buf, "|VTEXT");
1302	if (vp->v_flag & VSYSTEM)
1303		strcat(buf, "|VSYSTEM");
1304	if (vp->v_flag & VXLOCK)
1305		strcat(buf, "|VXLOCK");
1306	if (vp->v_flag & VXWANT)
1307		strcat(buf, "|VXWANT");
1308	if (vp->v_flag & VBWAIT)
1309		strcat(buf, "|VBWAIT");
1310	if (vp->v_flag & VALIASED)
1311		strcat(buf, "|VALIASED");
1312	if (buf[0] != '\0')
1313		printf(" flags (%s)", &buf[1]);
1314	if (vp->v_data == NULL) {
1315		printf("\n");
1316	} else {
1317		printf("\n\t");
1318		VOP_PRINT(vp);
1319	}
1320}
1321
1322#ifdef DDB
1323/*
1324 * List all of the locked vnodes in the system.
1325 * Called when debugging the kernel.
1326 */
1327void
1328printlockedvnodes(void)
1329{
1330	register struct mount *mp;
1331	register struct vnode *vp;
1332
1333	printf("Locked vnodes\n");
1334	for (mp = mountlist.cqh_first; mp != (void *)&mountlist;
1335	    mp = mp->mnt_list.cqe_next) {
1336		for (vp = mp->mnt_vnodelist.lh_first;
1337		    vp != NULL;
1338		    vp = vp->v_mntvnodes.le_next)
1339			if (VOP_ISLOCKED(vp))
1340				vprint((char *) 0, vp);
1341	}
1342}
1343#endif
1344
1345int kinfo_vdebug = 1;
1346int kinfo_vgetfailed;
1347
1348#define KINFO_VNODESLOP	10
1349/*
1350 * Dump vnode list (via sysctl).
1351 * Copyout address of vnode followed by vnode.
1352 */
1353/* ARGSUSED */
1354static int
1355sysctl_vnode SYSCTL_HANDLER_ARGS
1356{
1357	register struct mount *mp, *nmp;
1358	struct vnode *vp;
1359	int error;
1360
1361#define VPTRSZ	sizeof (struct vnode *)
1362#define VNODESZ	sizeof (struct vnode)
1363
1364	req->lock = 0;
1365	if (!req->oldptr) /* Make an estimate */
1366		return (SYSCTL_OUT(req, 0,
1367			(numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
1368
1369	for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) {
1370		nmp = mp->mnt_list.cqe_next;
1371		if (vfs_busy(mp))
1372			continue;
1373again:
1374		for (vp = mp->mnt_vnodelist.lh_first;
1375		    vp != NULL;
1376		    vp = vp->v_mntvnodes.le_next) {
1377			/*
1378			 * Check that the vp is still associated with this
1379			 * filesystem.  RACE: could have been recycled onto
1380			 * the same filesystem.
1381			 */
1382			if (vp->v_mount != mp) {
1383				if (kinfo_vdebug)
1384					printf("kinfo: vp changed\n");
1385				goto again;
1386			}
1387			if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
1388			    (error = SYSCTL_OUT(req, vp, VNODESZ))) {
1389				vfs_unbusy(mp);
1390				return (error);
1391			}
1392		}
1393		vfs_unbusy(mp);
1394	}
1395
1396	return (0);
1397}
1398
1399SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
1400	0, 0, sysctl_vnode, "S,vnode", "");
1401
1402/*
1403 * Check to see if a filesystem is mounted on a block device.
1404 */
1405int
1406vfs_mountedon(vp)
1407	register struct vnode *vp;
1408{
1409	register struct vnode *vq;
1410
1411	if (vp->v_specflags & SI_MOUNTEDON)
1412		return (EBUSY);
1413	if (vp->v_flag & VALIASED) {
1414		for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1415			if (vq->v_rdev != vp->v_rdev ||
1416			    vq->v_type != vp->v_type)
1417				continue;
1418			if (vq->v_specflags & SI_MOUNTEDON)
1419				return (EBUSY);
1420		}
1421	}
1422	return (0);
1423}
1424
1425/*
1426 * Build hash lists of net addresses and hang them off the mount point.
1427 * Called by ufs_mount() to set up the lists of export addresses.
1428 */
1429static int
1430vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
1431	struct export_args *argp)
1432{
1433	register struct netcred *np;
1434	register struct radix_node_head *rnh;
1435	register int i;
1436	struct radix_node *rn;
1437	struct sockaddr *saddr, *smask = 0;
1438	struct domain *dom;
1439	int error;
1440
1441	if (argp->ex_addrlen == 0) {
1442		if (mp->mnt_flag & MNT_DEFEXPORTED)
1443			return (EPERM);
1444		np = &nep->ne_defexported;
1445		np->netc_exflags = argp->ex_flags;
1446		np->netc_anon = argp->ex_anon;
1447		np->netc_anon.cr_ref = 1;
1448		mp->mnt_flag |= MNT_DEFEXPORTED;
1449		return (0);
1450	}
1451	i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
1452	np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
1453	bzero((caddr_t) np, i);
1454	saddr = (struct sockaddr *) (np + 1);
1455	if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
1456		goto out;
1457	if (saddr->sa_len > argp->ex_addrlen)
1458		saddr->sa_len = argp->ex_addrlen;
1459	if (argp->ex_masklen) {
1460		smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen);
1461		error = copyin(argp->ex_addr, (caddr_t) smask, argp->ex_masklen);
1462		if (error)
1463			goto out;
1464		if (smask->sa_len > argp->ex_masklen)
1465			smask->sa_len = argp->ex_masklen;
1466	}
1467	i = saddr->sa_family;
1468	if ((rnh = nep->ne_rtable[i]) == 0) {
1469		/*
1470		 * Seems silly to initialize every AF when most are not used,
1471		 * do so on demand here
1472		 */
1473		for (dom = domains; dom; dom = dom->dom_next)
1474			if (dom->dom_family == i && dom->dom_rtattach) {
1475				dom->dom_rtattach((void **) &nep->ne_rtable[i],
1476				    dom->dom_rtoffset);
1477				break;
1478			}
1479		if ((rnh = nep->ne_rtable[i]) == 0) {
1480			error = ENOBUFS;
1481			goto out;
1482		}
1483	}
1484	rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh,
1485	    np->netc_rnodes);
1486	if (rn == 0 || np != (struct netcred *) rn) {	/* already exists */
1487		error = EPERM;
1488		goto out;
1489	}
1490	np->netc_exflags = argp->ex_flags;
1491	np->netc_anon = argp->ex_anon;
1492	np->netc_anon.cr_ref = 1;
1493	return (0);
1494out:
1495	free(np, M_NETADDR);
1496	return (error);
1497}
1498
1499/* ARGSUSED */
1500static int
1501vfs_free_netcred(struct radix_node *rn, void *w)
1502{
1503	register struct radix_node_head *rnh = (struct radix_node_head *) w;
1504
1505	(*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
1506	free((caddr_t) rn, M_NETADDR);
1507	return (0);
1508}
1509
1510/*
1511 * Free the net address hash lists that are hanging off the mount points.
1512 */
1513static void
1514vfs_free_addrlist(struct netexport *nep)
1515{
1516	register int i;
1517	register struct radix_node_head *rnh;
1518
1519	for (i = 0; i <= AF_MAX; i++)
1520		if ((rnh = nep->ne_rtable[i])) {
1521			(*rnh->rnh_walktree) (rnh, vfs_free_netcred,
1522			    (caddr_t) rnh);
1523			free((caddr_t) rnh, M_RTABLE);
1524			nep->ne_rtable[i] = 0;
1525		}
1526}
1527
1528int
1529vfs_export(mp, nep, argp)
1530	struct mount *mp;
1531	struct netexport *nep;
1532	struct export_args *argp;
1533{
1534	int error;
1535
1536	if (argp->ex_flags & MNT_DELEXPORT) {
1537		vfs_free_addrlist(nep);
1538		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
1539	}
1540	if (argp->ex_flags & MNT_EXPORTED) {
1541		if ((error = vfs_hang_addrlist(mp, nep, argp)))
1542			return (error);
1543		mp->mnt_flag |= MNT_EXPORTED;
1544	}
1545	return (0);
1546}
1547
1548struct netcred *
1549vfs_export_lookup(mp, nep, nam)
1550	register struct mount *mp;
1551	struct netexport *nep;
1552	struct mbuf *nam;
1553{
1554	register struct netcred *np;
1555	register struct radix_node_head *rnh;
1556	struct sockaddr *saddr;
1557
1558	np = NULL;
1559	if (mp->mnt_flag & MNT_EXPORTED) {
1560		/*
1561		 * Lookup in the export list first.
1562		 */
1563		if (nam != NULL) {
1564			saddr = mtod(nam, struct sockaddr *);
1565			rnh = nep->ne_rtable[saddr->sa_family];
1566			if (rnh != NULL) {
1567				np = (struct netcred *)
1568				    (*rnh->rnh_matchaddr) ((caddr_t) saddr,
1569				    rnh);
1570				if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
1571					np = NULL;
1572			}
1573		}
1574		/*
1575		 * If no address match, use the default if it exists.
1576		 */
1577		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
1578			np = &nep->ne_defexported;
1579	}
1580	return (np);
1581}
1582
1583
1584/*
1585 * perform msync on all vnodes under a mount point
1586 * the mount point must be locked.
1587 */
1588void
1589vfs_msync(struct mount *mp, int flags) {
1590	struct vnode *vp, *nvp;
1591loop:
1592	for (vp = mp->mnt_vnodelist.lh_first; vp != NULL; vp = nvp) {
1593
1594		if (vp->v_mount != mp)
1595			goto loop;
1596		nvp = vp->v_mntvnodes.le_next;
1597		if (VOP_ISLOCKED(vp) && (flags != MNT_WAIT))
1598			continue;
1599		if (vp->v_object &&
1600		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1601			vm_object_page_clean(vp->v_object, 0, 0, TRUE, TRUE);
1602		}
1603	}
1604}
1605
1606/*
1607 * Create the VM object needed for VMIO and mmap support.  This
1608 * is done for all VREG files in the system.  Some filesystems might
1609 * afford the additional metadata buffering capability of the
1610 * VMIO code by making the device node be VMIO mode also.
1611 */
1612int
1613vfs_object_create(vp, p, cred, waslocked)
1614	struct vnode *vp;
1615	struct proc *p;
1616	struct ucred *cred;
1617	int waslocked;
1618{
1619	struct vattr vat;
1620	vm_object_t object;
1621	int error = 0;
1622
1623retry:
1624	if ((object = vp->v_object) == NULL) {
1625		if (vp->v_type == VREG) {
1626			if ((error = VOP_GETATTR(vp, &vat, cred, p)) != 0)
1627				goto retn;
1628			(void) vnode_pager_alloc(vp,
1629				OFF_TO_IDX(round_page(vat.va_size)), 0, 0);
1630		} else {
1631			/*
1632			 * This simply allocates the biggest object possible
1633			 * for a VBLK vnode.  This should be fixed, but doesn't
1634			 * cause any problems (yet).
1635			 */
1636			(void) vnode_pager_alloc(vp, INT_MAX, 0, 0);
1637		}
1638		vp->v_object->flags |= OBJ_VFS_REF;
1639	} else {
1640		if (object->flags & OBJ_DEAD) {
1641			if (waslocked)
1642				VOP_UNLOCK(vp);
1643			tsleep(object, PVM, "vodead", 0);
1644			if (waslocked)
1645				VOP_LOCK(vp);
1646			goto retry;
1647		}
1648		if ((object->flags & OBJ_VFS_REF) == 0) {
1649			object->flags |= OBJ_VFS_REF;
1650			vm_object_reference(object);
1651		}
1652	}
1653	if (vp->v_object)
1654		vp->v_flag |= VVMIO;
1655
1656retn:
1657	return error;
1658}
1659