xref: /freebsd/sys/kern/vfs_subr.c (revision 417ed37975261df51f61d13e179ad04d8f4839c7)
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  * $Id: vfs_subr.c,v 1.14 1995/01/09 16:04:54 davidg Exp $
40  */
41 
42 /*
43  * External virtual filesystem routines
44  */
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/proc.h>
49 #include <sys/mount.h>
50 #include <sys/time.h>
51 #include <sys/vnode.h>
52 #include <sys/stat.h>
53 #include <sys/namei.h>
54 #include <sys/ucred.h>
55 #include <sys/buf.h>
56 #include <sys/errno.h>
57 #include <sys/malloc.h>
58 #include <sys/domain.h>
59 #include <sys/mbuf.h>
60 
61 #include <vm/vm.h>
62 #include <sys/sysctl.h>
63 
64 #include <miscfs/specfs/specdev.h>
65 
66 void insmntque __P((struct vnode *, struct mount *));
67 
68 enum vtype iftovt_tab[16] = {
69 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
70 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
71 };
72 int vttoif_tab[9] = {
73 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
74 	S_IFSOCK, S_IFIFO, S_IFMT,
75 };
76 
77 /*
78  * Insq/Remq for the vnode usage lists.
79  */
80 #define	bufinsvn(bp, dp)	LIST_INSERT_HEAD(dp, bp, b_vnbufs)
81 #define	bufremvn(bp) {  \
82 	LIST_REMOVE(bp, b_vnbufs); \
83 	(bp)->b_vnbufs.le_next = NOLIST; \
84 }
85 
86 TAILQ_HEAD(freelst, vnode) vnode_free_list;	/* vnode free list */
87 struct mntlist mountlist;	/* mounted filesystem list */
88 
89 int desiredvnodes;
90 
91 /*
92  * Initialize the vnode management data structures.
93  */
94 void
95 vntblinit()
96 {
97 	extern int vm_object_cache_max;
98 
99 	desiredvnodes = maxproc + vm_object_cache_max;
100 
101 	TAILQ_INIT(&vnode_free_list);
102 	TAILQ_INIT(&mountlist);
103 }
104 
105 /*
106  * Lock a filesystem.
107  * Used to prevent access to it while mounting and unmounting.
108  */
109 int
110 vfs_lock(mp)
111 	register struct mount *mp;
112 {
113 
114 	while (mp->mnt_flag & MNT_MLOCK) {
115 		mp->mnt_flag |= MNT_MWAIT;
116 		(void) tsleep((caddr_t) mp, PVFS, "vfslck", 0);
117 	}
118 	mp->mnt_flag |= MNT_MLOCK;
119 	return (0);
120 }
121 
122 /*
123  * Unlock a locked filesystem.
124  * Panic if filesystem is not locked.
125  */
126 void
127 vfs_unlock(mp)
128 	register struct mount *mp;
129 {
130 
131 	if ((mp->mnt_flag & MNT_MLOCK) == 0)
132 		panic("vfs_unlock: not locked");
133 	mp->mnt_flag &= ~MNT_MLOCK;
134 	if (mp->mnt_flag & MNT_MWAIT) {
135 		mp->mnt_flag &= ~MNT_MWAIT;
136 		wakeup((caddr_t) mp);
137 	}
138 }
139 
140 /*
141  * Mark a mount point as busy.
142  * Used to synchronize access and to delay unmounting.
143  */
144 int
145 vfs_busy(mp)
146 	register struct mount *mp;
147 {
148 
149 	while (mp->mnt_flag & MNT_MPBUSY) {
150 		mp->mnt_flag |= MNT_MPWANT;
151 		(void) tsleep((caddr_t) &mp->mnt_flag, PVFS, "vfsbsy", 0);
152 	}
153 	if (mp->mnt_flag & MNT_UNMOUNT)
154 		return (1);
155 	mp->mnt_flag |= MNT_MPBUSY;
156 	return (0);
157 }
158 
159 /*
160  * Free a busy filesystem.
161  * Panic if filesystem is not busy.
162  */
163 void
164 vfs_unbusy(mp)
165 	register struct mount *mp;
166 {
167 
168 	if ((mp->mnt_flag & MNT_MPBUSY) == 0)
169 		panic("vfs_unbusy: not busy");
170 	mp->mnt_flag &= ~MNT_MPBUSY;
171 	if (mp->mnt_flag & MNT_MPWANT) {
172 		mp->mnt_flag &= ~MNT_MPWANT;
173 		wakeup((caddr_t) &mp->mnt_flag);
174 	}
175 }
176 
177 void
178 vfs_unmountroot(rootfs)
179 	struct mount *rootfs;
180 {
181 	struct mount *mp = rootfs;
182 	int error;
183 
184 	if (vfs_busy(mp)) {
185 		printf("failed to unmount root\n");
186 		return;
187 	}
188 	mp->mnt_flag |= MNT_UNMOUNT;
189 	if ((error = vfs_lock(mp))) {
190 		printf("lock of root filesystem failed (%d)\n", error);
191 		return;
192 	}
193 	vnode_pager_umount(mp);	/* release cached vnodes */
194 	cache_purgevfs(mp);	/* remove cache entries for this file sys */
195 
196 	if ((error = VFS_SYNC(mp, MNT_WAIT, initproc->p_ucred, initproc)))
197 		printf("sync of root filesystem failed (%d)\n", error);
198 
199 	if ((error = VFS_UNMOUNT(mp, MNT_FORCE, initproc))) {
200 		printf("unmount of root filesystem failed (");
201 		if (error == EBUSY)
202 			printf("BUSY)\n");
203 		else
204 			printf("%d)\n", error);
205 	}
206 	mp->mnt_flag &= ~MNT_UNMOUNT;
207 	vfs_unbusy(mp);
208 }
209 
210 /*
211  * Unmount all filesystems.  Should only be called by halt().
212  */
213 void
214 vfs_unmountall()
215 {
216 	struct mount *mp, *mp_next, *rootfs = NULL;
217 	int error;
218 
219 	/* unmount all but rootfs */
220 	for (mp = mountlist.tqh_first; mp != NULL; mp = mp_next) {
221 		mp_next = mp->mnt_list.tqe_next;
222 
223 		if (mp->mnt_flag & MNT_ROOTFS) {
224 			rootfs = mp;
225 			continue;
226 		}
227 		error = dounmount(mp, MNT_FORCE, initproc);
228 		if (error) {
229 			printf("unmount of %s failed (", mp->mnt_stat.f_mntonname);
230 			if (error == EBUSY)
231 				printf("BUSY)\n");
232 			else
233 				printf("%d)\n", error);
234 		}
235 	}
236 
237 	/* and finally... */
238 	if (rootfs) {
239 		vfs_unmountroot(rootfs);
240 	} else {
241 		printf("no root filesystem\n");
242 	}
243 }
244 
245 /*
246  * Lookup a mount point by filesystem identifier.
247  */
248 struct mount *
249 getvfs(fsid)
250 	fsid_t *fsid;
251 {
252 	register struct mount *mp;
253 
254 	for (mp = mountlist.tqh_first; mp != NULL; mp = mp->mnt_list.tqe_next) {
255 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
256 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1])
257 			return (mp);
258 	}
259 	return ((struct mount *) 0);
260 }
261 
262 /*
263  * Get a new unique fsid
264  */
265 void
266 getnewfsid(mp, mtype)
267 	struct mount *mp;
268 	int mtype;
269 {
270 	static u_short xxxfs_mntid;
271 
272 	fsid_t tfsid;
273 
274 	mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev + mtype, 0);
275 	mp->mnt_stat.f_fsid.val[1] = mtype;
276 	if (xxxfs_mntid == 0)
277 		++xxxfs_mntid;
278 	tfsid.val[0] = makedev(nblkdev + mtype, xxxfs_mntid);
279 	tfsid.val[1] = mtype;
280 	if (mountlist.tqh_first != NULL) {
281 		while (getvfs(&tfsid)) {
282 			tfsid.val[0]++;
283 			xxxfs_mntid++;
284 		}
285 	}
286 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
287 }
288 
289 /*
290  * Set vnode attributes to VNOVAL
291  */
292 void
293 vattr_null(vap)
294 	register struct vattr *vap;
295 {
296 
297 	vap->va_type = VNON;
298 	vap->va_size = VNOVAL;
299 	vap->va_bytes = VNOVAL;
300 	vap->va_mode = vap->va_nlink = vap->va_uid = vap->va_gid =
301 	    vap->va_fsid = vap->va_fileid =
302 	    vap->va_blocksize = vap->va_rdev =
303 	    vap->va_atime.ts_sec = vap->va_atime.ts_nsec =
304 	    vap->va_mtime.ts_sec = vap->va_mtime.ts_nsec =
305 	    vap->va_ctime.ts_sec = vap->va_ctime.ts_nsec =
306 	    vap->va_flags = vap->va_gen = VNOVAL;
307 	vap->va_vaflags = 0;
308 }
309 
310 /*
311  * Routines having to do with the management of the vnode table.
312  */
313 extern int (**dead_vnodeop_p) ();
314 extern void vclean();
315 long numvnodes;
316 
317 /*
318  * Return the next vnode from the free list.
319  */
320 int
321 getnewvnode(tag, mp, vops, vpp)
322 	enum vtagtype tag;
323 	struct mount *mp;
324 	int (**vops) ();
325 	struct vnode **vpp;
326 {
327 	register struct vnode *vp;
328 
329 	if (vnode_free_list.tqh_first == NULL ||
330 	    numvnodes < desiredvnodes) {
331 		vp = (struct vnode *) malloc((u_long) sizeof *vp,
332 		    M_VNODE, M_WAITOK);
333 		bzero((char *) vp, sizeof *vp);
334 		numvnodes++;
335 	} else {
336 		if ((vp = vnode_free_list.tqh_first) == NULL) {
337 			tablefull("vnode");
338 			*vpp = 0;
339 			return (ENFILE);
340 		}
341 		if (vp->v_usecount)
342 			panic("free vnode isn't");
343 
344 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
345 		/* see comment on why 0xdeadb is set at end of vgone (below) */
346 		vp->v_freelist.tqe_prev = (struct vnode **) 0xdeadb;
347 		vp->v_lease = NULL;
348 		if (vp->v_type != VBAD)
349 			vgone(vp);
350 #ifdef DIAGNOSTIC
351 		{
352 			int s;
353 
354 			if (vp->v_data)
355 				panic("cleaned vnode isn't");
356 			s = splbio();
357 			if (vp->v_numoutput)
358 				panic("Clean vnode has pending I/O's");
359 			splx(s);
360 		}
361 #endif
362 		vp->v_flag = 0;
363 		vp->v_lastr = 0;
364 		vp->v_ralen = 0;
365 		vp->v_maxra = 0;
366 		vp->v_lastw = 0;
367 		vp->v_lasta = 0;
368 		vp->v_cstart = 0;
369 		vp->v_clen = 0;
370 		vp->v_socket = 0;
371 		vp->v_writecount = 0;	/* XXX */
372 	}
373 	vp->v_type = VNON;
374 	cache_purge(vp);
375 	vp->v_tag = tag;
376 	vp->v_op = vops;
377 	insmntque(vp, mp);
378 	*vpp = vp;
379 	vp->v_usecount = 1;
380 	vp->v_data = 0;
381 	return (0);
382 }
383 
384 /*
385  * Move a vnode from one mount queue to another.
386  */
387 void
388 insmntque(vp, mp)
389 	register struct vnode *vp;
390 	register struct mount *mp;
391 {
392 
393 	/*
394 	 * Delete from old mount point vnode list, if on one.
395 	 */
396 	if (vp->v_mount != NULL)
397 		LIST_REMOVE(vp, v_mntvnodes);
398 	/*
399 	 * Insert into list of vnodes for the new mount point, if available.
400 	 */
401 	if ((vp->v_mount = mp) == NULL)
402 		return;
403 	LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes);
404 }
405 
406 /*
407  * Update outstanding I/O count and do wakeup if requested.
408  */
409 void
410 vwakeup(bp)
411 	register struct buf *bp;
412 {
413 	register struct vnode *vp;
414 
415 	bp->b_flags &= ~B_WRITEINPROG;
416 	if ((vp = bp->b_vp)) {
417 		vp->v_numoutput--;
418 		if (vp->v_numoutput < 0)
419 			panic("vwakeup: neg numoutput");
420 		if (vp->v_flag & VBWAIT) {
421 			vp->v_flag &= ~VBWAIT;
422 			wakeup((caddr_t) &vp->v_numoutput);
423 		}
424 	}
425 }
426 
427 /*
428  * Flush out and invalidate all buffers associated with a vnode.
429  * Called with the underlying object locked.
430  */
431 int
432 vinvalbuf(vp, flags, cred, p, slpflag, slptimeo)
433 	register struct vnode *vp;
434 	int flags;
435 	struct ucred *cred;
436 	struct proc *p;
437 	int slpflag, slptimeo;
438 {
439 	register struct buf *bp;
440 	struct buf *nbp, *blist;
441 	int s, error;
442 	vm_pager_t pager;
443 	vm_object_t object;
444 
445 	if (flags & V_SAVE) {
446 		if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, p)))
447 			return (error);
448 		if (vp->v_dirtyblkhd.lh_first != NULL)
449 			panic("vinvalbuf: dirty bufs");
450 	}
451 	for (;;) {
452 		if ((blist = vp->v_cleanblkhd.lh_first) && (flags & V_SAVEMETA))
453 			while (blist && blist->b_lblkno < 0)
454 				blist = blist->b_vnbufs.le_next;
455 		if (!blist && (blist = vp->v_dirtyblkhd.lh_first) &&
456 		    (flags & V_SAVEMETA))
457 			while (blist && blist->b_lblkno < 0)
458 				blist = blist->b_vnbufs.le_next;
459 		if (!blist)
460 			break;
461 
462 		for (bp = blist; bp; bp = nbp) {
463 			nbp = bp->b_vnbufs.le_next;
464 			if ((flags & V_SAVEMETA) && bp->b_lblkno < 0)
465 				continue;
466 			s = splbio();
467 			if (bp->b_flags & B_BUSY) {
468 				bp->b_flags |= B_WANTED;
469 				error = tsleep((caddr_t) bp,
470 				    slpflag | (PRIBIO + 1), "vinvalbuf",
471 				    slptimeo);
472 				splx(s);
473 				if (error)
474 					return (error);
475 				break;
476 			}
477 			bremfree(bp);
478 			bp->b_flags |= B_BUSY;
479 			splx(s);
480 			/*
481 			 * XXX Since there are no node locks for NFS, I
482 			 * believe there is a slight chance that a delayed
483 			 * write will occur while sleeping just above, so
484 			 * check for it.
485 			 */
486 			if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) {
487 				(void) VOP_BWRITE(bp);
488 				break;
489 			}
490 			bp->b_flags |= B_INVAL;
491 			brelse(bp);
492 		}
493 	}
494 
495 
496 	s = splbio();
497 	while (vp->v_numoutput > 0) {
498 		vp->v_flag |= VBWAIT;
499 		tsleep(&vp->v_numoutput, PVM, "vnvlbv", 0);
500 	}
501 	splx(s);
502 
503 	pager = NULL;
504 	object = (vm_object_t) vp->v_vmdata;
505 	if (object != NULL)
506 		pager = object->pager;
507 	if (pager != NULL) {
508 		object = vm_object_lookup(pager);
509 		if (object) {
510 			vm_object_lock(object);
511 			if (flags & V_SAVE)
512 				vm_object_page_clean(object, 0, 0, TRUE, FALSE);
513 			vm_object_page_remove(object, 0, object->size);
514 			vm_object_unlock(object);
515 			vm_object_deallocate(object);
516 		}
517 	}
518 	if (!(flags & V_SAVEMETA) &&
519 	    (vp->v_dirtyblkhd.lh_first || vp->v_cleanblkhd.lh_first))
520 		panic("vinvalbuf: flush failed");
521 	return (0);
522 }
523 
524 /*
525  * Associate a buffer with a vnode.
526  */
527 void
528 bgetvp(vp, bp)
529 	register struct vnode *vp;
530 	register struct buf *bp;
531 {
532 	int s;
533 
534 	if (bp->b_vp)
535 		panic("bgetvp: not free");
536 	VHOLD(vp);
537 	bp->b_vp = vp;
538 	if (vp->v_type == VBLK || vp->v_type == VCHR)
539 		bp->b_dev = vp->v_rdev;
540 	else
541 		bp->b_dev = NODEV;
542 	/*
543 	 * Insert onto list for new vnode.
544 	 */
545 	s = splbio();
546 	bufinsvn(bp, &vp->v_cleanblkhd);
547 	splx(s);
548 }
549 
550 /*
551  * Disassociate a buffer from a vnode.
552  */
553 void
554 brelvp(bp)
555 	register struct buf *bp;
556 {
557 	struct vnode *vp;
558 	int s;
559 
560 	if (bp->b_vp == (struct vnode *) 0)
561 		panic("brelvp: NULL");
562 	/*
563 	 * Delete from old vnode list, if on one.
564 	 */
565 	s = splbio();
566 	if (bp->b_vnbufs.le_next != NOLIST)
567 		bufremvn(bp);
568 	splx(s);
569 
570 	vp = bp->b_vp;
571 	bp->b_vp = (struct vnode *) 0;
572 	HOLDRELE(vp);
573 }
574 
575 /*
576  * Associate a p-buffer with a vnode.
577  */
578 void
579 pbgetvp(vp, bp)
580 	register struct vnode *vp;
581 	register struct buf *bp;
582 {
583 	if (bp->b_vp)
584 		panic("pbgetvp: not free");
585 	VHOLD(vp);
586 	bp->b_vp = vp;
587 	if (vp->v_type == VBLK || vp->v_type == VCHR)
588 		bp->b_dev = vp->v_rdev;
589 	else
590 		bp->b_dev = NODEV;
591 }
592 
593 /*
594  * Disassociate a p-buffer from a vnode.
595  */
596 void
597 pbrelvp(bp)
598 	register struct buf *bp;
599 {
600 	struct vnode *vp;
601 
602 	if (bp->b_vp == (struct vnode *) 0)
603 		panic("brelvp: NULL");
604 
605 	vp = bp->b_vp;
606 	bp->b_vp = (struct vnode *) 0;
607 	HOLDRELE(vp);
608 }
609 
610 /*
611  * Reassign a buffer from one vnode to another.
612  * Used to assign file specific control information
613  * (indirect blocks) to the vnode to which they belong.
614  */
615 void
616 reassignbuf(bp, newvp)
617 	register struct buf *bp;
618 	register struct vnode *newvp;
619 {
620 	register struct buflists *listheadp;
621 
622 	if (newvp == NULL) {
623 		printf("reassignbuf: NULL");
624 		return;
625 	}
626 	/*
627 	 * Delete from old vnode list, if on one.
628 	 */
629 	if (bp->b_vnbufs.le_next != NOLIST)
630 		bufremvn(bp);
631 	/*
632 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
633 	 * of clean buffers.
634 	 */
635 	if (bp->b_flags & B_DELWRI) {
636 		struct buf *tbp;
637 
638 		tbp = newvp->v_dirtyblkhd.lh_first;
639 		if (!tbp || (tbp->b_lblkno > bp->b_lblkno)) {
640 			bufinsvn(bp, &newvp->v_dirtyblkhd);
641 		} else {
642 			while (tbp->b_vnbufs.le_next && (tbp->b_vnbufs.le_next->b_lblkno < bp->b_lblkno)) {
643 				tbp = tbp->b_vnbufs.le_next;
644 			}
645 			LIST_INSERT_AFTER(tbp, bp, b_vnbufs);
646 		}
647 	} else {
648 		listheadp = &newvp->v_cleanblkhd;
649 		bufinsvn(bp, listheadp);
650 	}
651 }
652 
653 /*
654  * Create a vnode for a block device.
655  * Used for root filesystem, argdev, and swap areas.
656  * Also used for memory file system special devices.
657  */
658 int
659 bdevvp(dev, vpp)
660 	dev_t dev;
661 	struct vnode **vpp;
662 {
663 	register struct vnode *vp;
664 	struct vnode *nvp;
665 	int error;
666 
667 	if (dev == NODEV)
668 		return (0);
669 	error = getnewvnode(VT_NON, (struct mount *) 0, spec_vnodeop_p, &nvp);
670 	if (error) {
671 		*vpp = 0;
672 		return (error);
673 	}
674 	vp = nvp;
675 	vp->v_type = VBLK;
676 	if ((nvp = checkalias(vp, dev, (struct mount *) 0))) {
677 		vput(vp);
678 		vp = nvp;
679 	}
680 	*vpp = vp;
681 	return (0);
682 }
683 
684 /*
685  * Check to see if the new vnode represents a special device
686  * for which we already have a vnode (either because of
687  * bdevvp() or because of a different vnode representing
688  * the same block device). If such an alias exists, deallocate
689  * the existing contents and return the aliased vnode. The
690  * caller is responsible for filling it with its new contents.
691  */
692 struct vnode *
693 checkalias(nvp, nvp_rdev, mp)
694 	register struct vnode *nvp;
695 	dev_t nvp_rdev;
696 	struct mount *mp;
697 {
698 	register struct vnode *vp;
699 	struct vnode **vpp;
700 
701 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
702 		return (NULLVP);
703 
704 	vpp = &speclisth[SPECHASH(nvp_rdev)];
705 loop:
706 	for (vp = *vpp; vp; vp = vp->v_specnext) {
707 		if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type)
708 			continue;
709 		/*
710 		 * Alias, but not in use, so flush it out.
711 		 */
712 		if (vp->v_usecount == 0) {
713 			vgone(vp);
714 			goto loop;
715 		}
716 		if (vget(vp, 1))
717 			goto loop;
718 		break;
719 	}
720 	if (vp == NULL || vp->v_tag != VT_NON) {
721 		MALLOC(nvp->v_specinfo, struct specinfo *,
722 		    sizeof(struct specinfo), M_VNODE, M_WAITOK);
723 		nvp->v_rdev = nvp_rdev;
724 		nvp->v_hashchain = vpp;
725 		nvp->v_specnext = *vpp;
726 		nvp->v_specflags = 0;
727 		*vpp = nvp;
728 		if (vp != NULL) {
729 			nvp->v_flag |= VALIASED;
730 			vp->v_flag |= VALIASED;
731 			vput(vp);
732 		}
733 		return (NULLVP);
734 	}
735 	VOP_UNLOCK(vp);
736 	vclean(vp, 0);
737 	vp->v_op = nvp->v_op;
738 	vp->v_tag = nvp->v_tag;
739 	nvp->v_type = VNON;
740 	insmntque(vp, mp);
741 	return (vp);
742 }
743 
744 /*
745  * Grab a particular vnode from the free list, increment its
746  * reference count and lock it. The vnode lock bit is set the
747  * vnode is being eliminated in vgone. The process is awakened
748  * when the transition is completed, and an error returned to
749  * indicate that the vnode is no longer usable (possibly having
750  * been changed to a new file system type).
751  */
752 int
753 vget(vp, lockflag)
754 	register struct vnode *vp;
755 	int lockflag;
756 {
757 
758 	/*
759 	 * If the vnode is in the process of being cleaned out for another
760 	 * use, we wait for the cleaning to finish and then return failure.
761 	 * Cleaning is determined either by checking that the VXLOCK flag is
762 	 * set, or that the use count is zero with the back pointer set to
763 	 * show that it has been removed from the free list by getnewvnode.
764 	 * The VXLOCK flag may not have been set yet because vclean is blocked
765 	 * in the VOP_LOCK call waiting for the VOP_INACTIVE to complete.
766 	 */
767 	if ((vp->v_flag & VXLOCK) ||
768 	    (vp->v_usecount == 0 &&
769 		vp->v_freelist.tqe_prev == (struct vnode **) 0xdeadb)) {
770 		vp->v_flag |= VXWANT;
771 		(void) tsleep((caddr_t) vp, PINOD, "vget", 0);
772 		return (1);
773 	}
774 	if (vp->v_usecount == 0)
775 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
776 	vp->v_usecount++;
777 	if (lockflag)
778 		VOP_LOCK(vp);
779 	return (0);
780 }
781 
782 /*
783  * Vnode reference, just increment the count
784  */
785 void
786 vref(vp)
787 	struct vnode *vp;
788 {
789 
790 	if (vp->v_usecount <= 0)
791 		panic("vref used where vget required");
792 	vp->v_usecount++;
793 }
794 
795 /*
796  * vput(), just unlock and vrele()
797  */
798 void
799 vput(vp)
800 	register struct vnode *vp;
801 {
802 
803 	VOP_UNLOCK(vp);
804 	vrele(vp);
805 }
806 
807 /*
808  * Vnode release.
809  * If count drops to zero, call inactive routine and return to freelist.
810  */
811 void
812 vrele(vp)
813 	register struct vnode *vp;
814 {
815 
816 #ifdef DIAGNOSTIC
817 	if (vp == NULL)
818 		panic("vrele: null vp");
819 #endif
820 	vp->v_usecount--;
821 	if (vp->v_usecount > 0)
822 		return;
823 #ifdef DIAGNOSTIC
824 	if (vp->v_usecount != 0 /* || vp->v_writecount != 0 */ ) {
825 		vprint("vrele: bad ref count", vp);
826 		panic("vrele: ref cnt");
827 	}
828 #endif
829 	/*
830 	 * insert at tail of LRU list
831 	 */
832 	TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
833 	VOP_INACTIVE(vp);
834 }
835 
836 /*
837  * Page or buffer structure gets a reference.
838  */
839 void
840 vhold(vp)
841 	register struct vnode *vp;
842 {
843 
844 	vp->v_holdcnt++;
845 }
846 
847 /*
848  * Page or buffer structure frees a reference.
849  */
850 void
851 holdrele(vp)
852 	register struct vnode *vp;
853 {
854 
855 	if (vp->v_holdcnt <= 0)
856 		panic("holdrele: holdcnt");
857 	vp->v_holdcnt--;
858 }
859 
860 /*
861  * Remove any vnodes in the vnode table belonging to mount point mp.
862  *
863  * If MNT_NOFORCE is specified, there should not be any active ones,
864  * return error if any are found (nb: this is a user error, not a
865  * system error). If MNT_FORCE is specified, detach any active vnodes
866  * that are found.
867  */
868 #ifdef DIAGNOSTIC
869 int busyprt = 0;		/* print out busy vnodes */
870 struct ctldebug debug1 = {"busyprt", &busyprt};
871 
872 #endif
873 
874 int
875 vflush(mp, skipvp, flags)
876 	struct mount *mp;
877 	struct vnode *skipvp;
878 	int flags;
879 {
880 	register struct vnode *vp, *nvp;
881 	int busy = 0;
882 
883 	if ((mp->mnt_flag & MNT_MPBUSY) == 0)
884 		panic("vflush: not busy");
885 loop:
886 	for (vp = mp->mnt_vnodelist.lh_first; vp; vp = nvp) {
887 		if (vp->v_mount != mp)
888 			goto loop;
889 		nvp = vp->v_mntvnodes.le_next;
890 		/*
891 		 * Skip over a selected vnode.
892 		 */
893 		if (vp == skipvp)
894 			continue;
895 		/*
896 		 * Skip over a vnodes marked VSYSTEM.
897 		 */
898 		if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM))
899 			continue;
900 		/*
901 		 * If WRITECLOSE is set, only flush out regular file vnodes
902 		 * open for writing.
903 		 */
904 		if ((flags & WRITECLOSE) &&
905 		    (vp->v_writecount == 0 || vp->v_type != VREG))
906 			continue;
907 		/*
908 		 * With v_usecount == 0, all we need to do is clear out the
909 		 * vnode data structures and we are done.
910 		 */
911 		if (vp->v_usecount == 0) {
912 			vgone(vp);
913 			continue;
914 		}
915 		/*
916 		 * If FORCECLOSE is set, forcibly close the vnode. For block
917 		 * or character devices, revert to an anonymous device. For
918 		 * all other files, just kill them.
919 		 */
920 		if (flags & FORCECLOSE) {
921 			if (vp->v_type != VBLK && vp->v_type != VCHR) {
922 				vgone(vp);
923 			} else {
924 				vclean(vp, 0);
925 				vp->v_op = spec_vnodeop_p;
926 				insmntque(vp, (struct mount *) 0);
927 			}
928 			continue;
929 		}
930 #ifdef DIAGNOSTIC
931 		if (busyprt)
932 			vprint("vflush: busy vnode", vp);
933 #endif
934 		busy++;
935 	}
936 	if (busy)
937 		return (EBUSY);
938 	return (0);
939 }
940 
941 /*
942  * Disassociate the underlying file system from a vnode.
943  */
944 void
945 vclean(vp, flags)
946 	register struct vnode *vp;
947 	int flags;
948 {
949 	int active;
950 
951 	/*
952 	 * Check to see if the vnode is in use. If so we have to reference it
953 	 * before we clean it out so that its count cannot fall to zero and
954 	 * generate a race against ourselves to recycle it.
955 	 */
956 	if ((active = vp->v_usecount))
957 		VREF(vp);
958 	/*
959 	 * Even if the count is zero, the VOP_INACTIVE routine may still have
960 	 * the object locked while it cleans it out. The VOP_LOCK ensures that
961 	 * the VOP_INACTIVE routine is done with its work. For active vnodes,
962 	 * it ensures that no other activity can occur while the underlying
963 	 * object is being cleaned out.
964 	 */
965 	VOP_LOCK(vp);
966 	/*
967 	 * Prevent the vnode from being recycled or brought into use while we
968 	 * clean it out.
969 	 */
970 	if (vp->v_flag & VXLOCK)
971 		panic("vclean: deadlock");
972 	vp->v_flag |= VXLOCK;
973 	/*
974 	 * Clean out any buffers associated with the vnode.
975 	 */
976 	if (flags & DOCLOSE)
977 		vinvalbuf(vp, V_SAVE, NOCRED, NULL, 0, 0);
978 	/*
979 	 * Any other processes trying to obtain this lock must first wait for
980 	 * VXLOCK to clear, then call the new lock operation.
981 	 */
982 	VOP_UNLOCK(vp);
983 	/*
984 	 * If purging an active vnode, it must be closed and deactivated
985 	 * before being reclaimed.
986 	 */
987 	if (active) {
988 		if (flags & DOCLOSE)
989 			VOP_CLOSE(vp, IO_NDELAY, NOCRED, NULL);
990 		VOP_INACTIVE(vp);
991 	}
992 	/*
993 	 * Reclaim the vnode.
994 	 */
995 	if (VOP_RECLAIM(vp))
996 		panic("vclean: cannot reclaim");
997 	if (active)
998 		vrele(vp);
999 
1000 	/*
1001 	 * Done with purge, notify sleepers of the grim news.
1002 	 */
1003 	vp->v_op = dead_vnodeop_p;
1004 	vp->v_tag = VT_NON;
1005 	vp->v_flag &= ~VXLOCK;
1006 	if (vp->v_flag & VXWANT) {
1007 		vp->v_flag &= ~VXWANT;
1008 		wakeup((caddr_t) vp);
1009 	}
1010 }
1011 
1012 /*
1013  * Eliminate all activity associated with  the requested vnode
1014  * and with all vnodes aliased to the requested vnode.
1015  */
1016 void
1017 vgoneall(vp)
1018 	register struct vnode *vp;
1019 {
1020 	register struct vnode *vq;
1021 
1022 	if (vp->v_flag & VALIASED) {
1023 		/*
1024 		 * If a vgone (or vclean) is already in progress, wait until
1025 		 * it is done and return.
1026 		 */
1027 		if (vp->v_flag & VXLOCK) {
1028 			vp->v_flag |= VXWANT;
1029 			(void) tsleep((caddr_t) vp, PINOD, "vgall", 0);
1030 			return;
1031 		}
1032 		/*
1033 		 * Ensure that vp will not be vgone'd while we are eliminating
1034 		 * its aliases.
1035 		 */
1036 		vp->v_flag |= VXLOCK;
1037 		while (vp->v_flag & VALIASED) {
1038 			for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1039 				if (vq->v_rdev != vp->v_rdev ||
1040 				    vq->v_type != vp->v_type || vp == vq)
1041 					continue;
1042 				vgone(vq);
1043 				break;
1044 			}
1045 		}
1046 		/*
1047 		 * Remove the lock so that vgone below will really eliminate
1048 		 * the vnode after which time vgone will awaken any sleepers.
1049 		 */
1050 		vp->v_flag &= ~VXLOCK;
1051 	}
1052 	vgone(vp);
1053 }
1054 
1055 /*
1056  * Eliminate all activity associated with a vnode
1057  * in preparation for reuse.
1058  */
1059 void
1060 vgone(vp)
1061 	register struct vnode *vp;
1062 {
1063 	register struct vnode *vq;
1064 	struct vnode *vx;
1065 
1066 	/*
1067 	 * If a vgone (or vclean) is already in progress, wait until it is
1068 	 * done and return.
1069 	 */
1070 	if (vp->v_flag & VXLOCK) {
1071 		vp->v_flag |= VXWANT;
1072 		(void) tsleep((caddr_t) vp, PINOD, "vgone", 0);
1073 		return;
1074 	}
1075 	/*
1076 	 * Clean out the filesystem specific data.
1077 	 */
1078 	vclean(vp, DOCLOSE);
1079 	/*
1080 	 * Delete from old mount point vnode list, if on one.
1081 	 */
1082 	if (vp->v_mount != NULL) {
1083 		LIST_REMOVE(vp, v_mntvnodes);
1084 		vp->v_mount = NULL;
1085 	}
1086 	/*
1087 	 * If special device, remove it from special device alias list.
1088 	 */
1089 	if (vp->v_type == VBLK || vp->v_type == VCHR) {
1090 		if (*vp->v_hashchain == vp) {
1091 			*vp->v_hashchain = vp->v_specnext;
1092 		} else {
1093 			for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1094 				if (vq->v_specnext != vp)
1095 					continue;
1096 				vq->v_specnext = vp->v_specnext;
1097 				break;
1098 			}
1099 			if (vq == NULL)
1100 				panic("missing bdev");
1101 		}
1102 		if (vp->v_flag & VALIASED) {
1103 			vx = NULL;
1104 			for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1105 				if (vq->v_rdev != vp->v_rdev ||
1106 				    vq->v_type != vp->v_type)
1107 					continue;
1108 				if (vx)
1109 					break;
1110 				vx = vq;
1111 			}
1112 			if (vx == NULL)
1113 				panic("missing alias");
1114 			if (vq == NULL)
1115 				vx->v_flag &= ~VALIASED;
1116 			vp->v_flag &= ~VALIASED;
1117 		}
1118 		FREE(vp->v_specinfo, M_VNODE);
1119 		vp->v_specinfo = NULL;
1120 	}
1121 	/*
1122 	 * If it is on the freelist and not already at the head, move it to
1123 	 * the head of the list. The test of the back pointer and the
1124 	 * reference count of zero is because it will be removed from the free
1125 	 * list by getnewvnode, but will not have its reference count
1126 	 * incremented until after calling vgone. If the reference count were
1127 	 * incremented first, vgone would (incorrectly) try to close the
1128 	 * previous instance of the underlying object. So, the back pointer is
1129 	 * explicitly set to `0xdeadb' in getnewvnode after removing it from
1130 	 * the freelist to ensure that we do not try to move it here.
1131 	 */
1132 	if (vp->v_usecount == 0 &&
1133 	    vp->v_freelist.tqe_prev != (struct vnode **) 0xdeadb &&
1134 	    vnode_free_list.tqh_first != vp) {
1135 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1136 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
1137 	}
1138 	vp->v_type = VBAD;
1139 }
1140 
1141 /*
1142  * Lookup a vnode by device number.
1143  */
1144 int
1145 vfinddev(dev, type, vpp)
1146 	dev_t dev;
1147 	enum vtype type;
1148 	struct vnode **vpp;
1149 {
1150 	register struct vnode *vp;
1151 
1152 	for (vp = speclisth[SPECHASH(dev)]; vp; vp = vp->v_specnext) {
1153 		if (dev != vp->v_rdev || type != vp->v_type)
1154 			continue;
1155 		*vpp = vp;
1156 		return (1);
1157 	}
1158 	return (0);
1159 }
1160 
1161 /*
1162  * Calculate the total number of references to a special device.
1163  */
1164 int
1165 vcount(vp)
1166 	register struct vnode *vp;
1167 {
1168 	register struct vnode *vq, *vnext;
1169 	int count;
1170 
1171 loop:
1172 	if ((vp->v_flag & VALIASED) == 0)
1173 		return (vp->v_usecount);
1174 	for (count = 0, vq = *vp->v_hashchain; vq; vq = vnext) {
1175 		vnext = vq->v_specnext;
1176 		if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type)
1177 			continue;
1178 		/*
1179 		 * Alias, but not in use, so flush it out.
1180 		 */
1181 		if (vq->v_usecount == 0 && vq != vp) {
1182 			vgone(vq);
1183 			goto loop;
1184 		}
1185 		count += vq->v_usecount;
1186 	}
1187 	return (count);
1188 }
1189 
1190 /*
1191  * Print out a description of a vnode.
1192  */
1193 static char *typename[] =
1194 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1195 
1196 void
1197 vprint(label, vp)
1198 	char *label;
1199 	register struct vnode *vp;
1200 {
1201 	char buf[64];
1202 
1203 	if (label != NULL)
1204 		printf("%s: ", label);
1205 	printf("type %s, usecount %d, writecount %d, refcount %ld,",
1206 	    typename[vp->v_type], vp->v_usecount, vp->v_writecount,
1207 	    vp->v_holdcnt);
1208 	buf[0] = '\0';
1209 	if (vp->v_flag & VROOT)
1210 		strcat(buf, "|VROOT");
1211 	if (vp->v_flag & VTEXT)
1212 		strcat(buf, "|VTEXT");
1213 	if (vp->v_flag & VSYSTEM)
1214 		strcat(buf, "|VSYSTEM");
1215 	if (vp->v_flag & VXLOCK)
1216 		strcat(buf, "|VXLOCK");
1217 	if (vp->v_flag & VXWANT)
1218 		strcat(buf, "|VXWANT");
1219 	if (vp->v_flag & VBWAIT)
1220 		strcat(buf, "|VBWAIT");
1221 	if (vp->v_flag & VALIASED)
1222 		strcat(buf, "|VALIASED");
1223 	if (buf[0] != '\0')
1224 		printf(" flags (%s)", &buf[1]);
1225 	if (vp->v_data == NULL) {
1226 		printf("\n");
1227 	} else {
1228 		printf("\n\t");
1229 		VOP_PRINT(vp);
1230 	}
1231 }
1232 
1233 #ifdef DEBUG
1234 /*
1235  * List all of the locked vnodes in the system.
1236  * Called when debugging the kernel.
1237  */
1238 void
1239 printlockedvnodes()
1240 {
1241 	register struct mount *mp;
1242 	register struct vnode *vp;
1243 
1244 	printf("Locked vnodes\n");
1245 	for (mp = mountlist.tqh_first; mp != NULL; mp = mp->mnt_list.tqe_next) {
1246 		for (vp = mp->mnt_vnodelist.lh_first;
1247 		    vp != NULL;
1248 		    vp = vp->v_mntvnodes.le_next)
1249 			if (VOP_ISLOCKED(vp))
1250 				vprint((char *) 0, vp);
1251 	}
1252 }
1253 #endif
1254 
1255 int kinfo_vdebug = 1;
1256 int kinfo_vgetfailed;
1257 
1258 #define KINFO_VNODESLOP	10
1259 /*
1260  * Dump vnode list (via sysctl).
1261  * Copyout address of vnode followed by vnode.
1262  */
1263 /* ARGSUSED */
1264 int
1265 sysctl_vnode(where, sizep)
1266 	char *where;
1267 	size_t *sizep;
1268 {
1269 	register struct mount *mp, *nmp;
1270 	struct vnode *vp;
1271 	register char *bp = where, *savebp;
1272 	char *ewhere;
1273 	int error;
1274 
1275 #define VPTRSZ	sizeof (struct vnode *)
1276 #define VNODESZ	sizeof (struct vnode)
1277 	if (where == NULL) {
1278 		*sizep = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ);
1279 		return (0);
1280 	}
1281 	ewhere = where + *sizep;
1282 
1283 	for (mp = mountlist.tqh_first; mp != NULL; mp = nmp) {
1284 		nmp = mp->mnt_list.tqe_next;
1285 		if (vfs_busy(mp))
1286 			continue;
1287 		savebp = bp;
1288 again:
1289 		for (vp = mp->mnt_vnodelist.lh_first;
1290 		    vp != NULL;
1291 		    vp = vp->v_mntvnodes.le_next) {
1292 			/*
1293 			 * Check that the vp is still associated with this
1294 			 * filesystem.  RACE: could have been recycled onto
1295 			 * the same filesystem.
1296 			 */
1297 			if (vp->v_mount != mp) {
1298 				if (kinfo_vdebug)
1299 					printf("kinfo: vp changed\n");
1300 				bp = savebp;
1301 				goto again;
1302 			}
1303 			if (bp + VPTRSZ + VNODESZ > ewhere) {
1304 				*sizep = bp - where;
1305 				return (ENOMEM);
1306 			}
1307 			if ((error = copyout((caddr_t) &vp, bp, VPTRSZ)) ||
1308 			    (error = copyout((caddr_t) vp, bp + VPTRSZ, VNODESZ)))
1309 				return (error);
1310 			bp += VPTRSZ + VNODESZ;
1311 		}
1312 		vfs_unbusy(mp);
1313 	}
1314 
1315 	*sizep = bp - where;
1316 	return (0);
1317 }
1318 
1319 /*
1320  * Check to see if a filesystem is mounted on a block device.
1321  */
1322 int
1323 vfs_mountedon(vp)
1324 	register struct vnode *vp;
1325 {
1326 	register struct vnode *vq;
1327 
1328 	if (vp->v_specflags & SI_MOUNTEDON)
1329 		return (EBUSY);
1330 	if (vp->v_flag & VALIASED) {
1331 		for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1332 			if (vq->v_rdev != vp->v_rdev ||
1333 			    vq->v_type != vp->v_type)
1334 				continue;
1335 			if (vq->v_specflags & SI_MOUNTEDON)
1336 				return (EBUSY);
1337 		}
1338 	}
1339 	return (0);
1340 }
1341 
1342 /*
1343  * Build hash lists of net addresses and hang them off the mount point.
1344  * Called by ufs_mount() to set up the lists of export addresses.
1345  */
1346 static int
1347 vfs_hang_addrlist(mp, nep, argp)
1348 	struct mount *mp;
1349 	struct netexport *nep;
1350 	struct export_args *argp;
1351 {
1352 	register struct netcred *np;
1353 	register struct radix_node_head *rnh;
1354 	register int i;
1355 	struct radix_node *rn;
1356 	struct sockaddr *saddr, *smask = 0;
1357 	struct domain *dom;
1358 	int error;
1359 
1360 	if (argp->ex_addrlen == 0) {
1361 		if (mp->mnt_flag & MNT_DEFEXPORTED)
1362 			return (EPERM);
1363 		np = &nep->ne_defexported;
1364 		np->netc_exflags = argp->ex_flags;
1365 		np->netc_anon = argp->ex_anon;
1366 		np->netc_anon.cr_ref = 1;
1367 		mp->mnt_flag |= MNT_DEFEXPORTED;
1368 		return (0);
1369 	}
1370 	i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
1371 	np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK);
1372 	bzero((caddr_t) np, i);
1373 	saddr = (struct sockaddr *) (np + 1);
1374 	if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
1375 		goto out;
1376 	if (saddr->sa_len > argp->ex_addrlen)
1377 		saddr->sa_len = argp->ex_addrlen;
1378 	if (argp->ex_masklen) {
1379 		smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen);
1380 		error = copyin(argp->ex_addr, (caddr_t) smask, argp->ex_masklen);
1381 		if (error)
1382 			goto out;
1383 		if (smask->sa_len > argp->ex_masklen)
1384 			smask->sa_len = argp->ex_masklen;
1385 	}
1386 	i = saddr->sa_family;
1387 	if ((rnh = nep->ne_rtable[i]) == 0) {
1388 		/*
1389 		 * Seems silly to initialize every AF when most are not used,
1390 		 * do so on demand here
1391 		 */
1392 		for (dom = domains; dom; dom = dom->dom_next)
1393 			if (dom->dom_family == i && dom->dom_rtattach) {
1394 				dom->dom_rtattach((void **) &nep->ne_rtable[i],
1395 				    dom->dom_rtoffset);
1396 				break;
1397 			}
1398 		if ((rnh = nep->ne_rtable[i]) == 0) {
1399 			error = ENOBUFS;
1400 			goto out;
1401 		}
1402 	}
1403 	rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh,
1404 	    np->netc_rnodes);
1405 	if (rn == 0 || np != (struct netcred *) rn) {	/* already exists */
1406 		error = EPERM;
1407 		goto out;
1408 	}
1409 	np->netc_exflags = argp->ex_flags;
1410 	np->netc_anon = argp->ex_anon;
1411 	np->netc_anon.cr_ref = 1;
1412 	return (0);
1413 out:
1414 	free(np, M_NETADDR);
1415 	return (error);
1416 }
1417 
1418 /* ARGSUSED */
1419 static int
1420 vfs_free_netcred(rn, w)
1421 	struct radix_node *rn;
1422 	caddr_t w;
1423 {
1424 	register struct radix_node_head *rnh = (struct radix_node_head *) w;
1425 
1426 	(*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
1427 	free((caddr_t) rn, M_NETADDR);
1428 	return (0);
1429 }
1430 
1431 /*
1432  * Free the net address hash lists that are hanging off the mount points.
1433  */
1434 static void
1435 vfs_free_addrlist(nep)
1436 	struct netexport *nep;
1437 {
1438 	register int i;
1439 	register struct radix_node_head *rnh;
1440 
1441 	for (i = 0; i <= AF_MAX; i++)
1442 		if ((rnh = nep->ne_rtable[i])) {
1443 			(*rnh->rnh_walktree) (rnh, vfs_free_netcred,
1444 			    (caddr_t) rnh);
1445 			free((caddr_t) rnh, M_RTABLE);
1446 			nep->ne_rtable[i] = 0;
1447 		}
1448 }
1449 
1450 int
1451 vfs_export(mp, nep, argp)
1452 	struct mount *mp;
1453 	struct netexport *nep;
1454 	struct export_args *argp;
1455 {
1456 	int error;
1457 
1458 	if (argp->ex_flags & MNT_DELEXPORT) {
1459 		vfs_free_addrlist(nep);
1460 		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
1461 	}
1462 	if (argp->ex_flags & MNT_EXPORTED) {
1463 		if ((error = vfs_hang_addrlist(mp, nep, argp)))
1464 			return (error);
1465 		mp->mnt_flag |= MNT_EXPORTED;
1466 	}
1467 	return (0);
1468 }
1469 
1470 struct netcred *
1471 vfs_export_lookup(mp, nep, nam)
1472 	register struct mount *mp;
1473 	struct netexport *nep;
1474 	struct mbuf *nam;
1475 {
1476 	register struct netcred *np;
1477 	register struct radix_node_head *rnh;
1478 	struct sockaddr *saddr;
1479 
1480 	np = NULL;
1481 	if (mp->mnt_flag & MNT_EXPORTED) {
1482 		/*
1483 		 * Lookup in the export list first.
1484 		 */
1485 		if (nam != NULL) {
1486 			saddr = mtod(nam, struct sockaddr *);
1487 			rnh = nep->ne_rtable[saddr->sa_family];
1488 			if (rnh != NULL) {
1489 				np = (struct netcred *)
1490 				    (*rnh->rnh_matchaddr) ((caddr_t) saddr,
1491 				    rnh);
1492 				if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
1493 					np = NULL;
1494 			}
1495 		}
1496 		/*
1497 		 * If no address match, use the default if it exists.
1498 		 */
1499 		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
1500 			np = &nep->ne_defexported;
1501 	}
1502 	return (np);
1503 }
1504