xref: /openbsd-src/sys/kern/vfs_subr.c (revision df930be708d50e9715f173caa26ffe1b7599b157)
1 /*	$NetBSD: vfs_subr.c,v 1.47 1995/10/07 06:28:48 mycroft Exp $	*/
2 
3 /*
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  * (c) UNIX System Laboratories, Inc.
7  * All or some portions of this file are derived from material licensed
8  * to the University of California by American Telephone and Telegraph
9  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10  * the permission of UNIX System Laboratories, Inc.
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  * 3. All advertising materials mentioning features or use of this software
21  *    must display the following acknowledgement:
22  *	This product includes software developed by the University of
23  *	California, Berkeley and its contributors.
24  * 4. Neither the name of the University nor the names of its contributors
25  *    may be used to endorse or promote products derived from this software
26  *    without specific prior written permission.
27  *
28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38  * SUCH DAMAGE.
39  *
40  *	@(#)vfs_subr.c	8.13 (Berkeley) 4/18/94
41  */
42 
43 /*
44  * External virtual filesystem routines
45  */
46 
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/proc.h>
50 #include <sys/mount.h>
51 #include <sys/time.h>
52 #include <sys/fcntl.h>
53 #include <sys/vnode.h>
54 #include <sys/stat.h>
55 #include <sys/namei.h>
56 #include <sys/ucred.h>
57 #include <sys/buf.h>
58 #include <sys/errno.h>
59 #include <sys/malloc.h>
60 #include <sys/domain.h>
61 #include <sys/mbuf.h>
62 
63 #include <vm/vm.h>
64 #include <sys/sysctl.h>
65 
66 #include <miscfs/specfs/specdev.h>
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 int doforce = 1;		/* 1 => permit forcible unmounting */
78 int prtactive = 0;		/* 1 => print out reclaim of active vnodes */
79 
80 /*
81  * Insq/Remq for the vnode usage lists.
82  */
83 #define	bufinsvn(bp, dp)	LIST_INSERT_HEAD(dp, bp, b_vnbufs)
84 #define	bufremvn(bp) {							\
85 	LIST_REMOVE(bp, b_vnbufs);					\
86 	(bp)->b_vnbufs.le_next = NOLIST;				\
87 }
88 TAILQ_HEAD(freelst, vnode) vnode_free_list;	/* vnode free list */
89 struct mntlist mountlist;			/* mounted filesystem list */
90 
91 /*
92  * Initialize the vnode management data structures.
93  */
94 vntblinit()
95 {
96 
97 	TAILQ_INIT(&vnode_free_list);
98 	CIRCLEQ_INIT(&mountlist);
99 }
100 
101 /*
102  * Lock a filesystem.
103  * Used to prevent access to it while mounting and unmounting.
104  */
105 vfs_lock(mp)
106 	register struct mount *mp;
107 {
108 
109 	while (mp->mnt_flag & MNT_MLOCK) {
110 		mp->mnt_flag |= MNT_MWAIT;
111 		tsleep((caddr_t)mp, PVFS, "vfslock", 0);
112 	}
113 	mp->mnt_flag |= MNT_MLOCK;
114 	return (0);
115 }
116 
117 /*
118  * Unlock a locked filesystem.
119  * Panic if filesystem is not locked.
120  */
121 void
122 vfs_unlock(mp)
123 	register struct mount *mp;
124 {
125 
126 	if ((mp->mnt_flag & MNT_MLOCK) == 0)
127 		panic("vfs_unlock: not locked");
128 	mp->mnt_flag &= ~MNT_MLOCK;
129 	if (mp->mnt_flag & MNT_MWAIT) {
130 		mp->mnt_flag &= ~MNT_MWAIT;
131 		wakeup((caddr_t)mp);
132 	}
133 }
134 
135 /*
136  * Mark a mount point as busy.
137  * Used to synchronize access and to delay unmounting.
138  */
139 vfs_busy(mp)
140 	register struct mount *mp;
141 {
142 
143 	while(mp->mnt_flag & MNT_MPBUSY) {
144 		mp->mnt_flag |= MNT_MPWANT;
145 		tsleep((caddr_t)&mp->mnt_flag, PVFS, "vfsbusy", 0);
146 	}
147 	if (mp->mnt_flag & MNT_UNMOUNT)
148 		return (1);
149 	mp->mnt_flag |= MNT_MPBUSY;
150 	return (0);
151 }
152 
153 /*
154  * Free a busy filesystem.
155  * Panic if filesystem is not busy.
156  */
157 vfs_unbusy(mp)
158 	register struct mount *mp;
159 {
160 
161 	if ((mp->mnt_flag & MNT_MPBUSY) == 0)
162 		panic("vfs_unbusy: not busy");
163 	mp->mnt_flag &= ~MNT_MPBUSY;
164 	if (mp->mnt_flag & MNT_MPWANT) {
165 		mp->mnt_flag &= ~MNT_MPWANT;
166 		wakeup((caddr_t)&mp->mnt_flag);
167 	}
168 }
169 
170 /*
171  * Lookup a mount point by filesystem identifier.
172  */
173 struct mount *
174 getvfs(fsid)
175 	fsid_t *fsid;
176 {
177 	register struct mount *mp;
178 
179 	for (mp = mountlist.cqh_first; mp != (void *)&mountlist;
180 	     mp = mp->mnt_list.cqe_next)
181 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
182 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1])
183 			return (mp);
184 	return ((struct mount *)0);
185 }
186 
187 /*
188  * Get a new unique fsid
189  */
190 void
191 getnewfsid(mp, mtype)
192 	struct mount *mp;
193 	int mtype;
194 {
195 	static u_short xxxfs_mntid;
196 
197 	fsid_t tfsid;
198 
199 	mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev + 11, 0);	/* XXX */
200 	mp->mnt_stat.f_fsid.val[1] = mtype;
201 	if (xxxfs_mntid == 0)
202 		++xxxfs_mntid;
203 	tfsid.val[0] = makedev((nblkdev + mtype) & 0xff, xxxfs_mntid);
204 	tfsid.val[1] = mtype;
205 	if (mountlist.cqh_first != (void *)&mountlist) {
206 		while (getvfs(&tfsid)) {
207 			tfsid.val[0]++;
208 			xxxfs_mntid++;
209 		}
210 	}
211 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
212 }
213 
214 /*
215  * Make a 'unique' number from a mount type name.
216  */
217 long
218 makefstype(type)
219 	char *type;
220 {
221 	long rv;
222 
223 	for (rv = 0; *type; type++) {
224 		rv <<= 2;
225 		rv ^= *type;
226 	}
227 	return rv;
228 }
229 
230 /*
231  * Set vnode attributes to VNOVAL
232  */
233 void
234 vattr_null(vap)
235 	register struct vattr *vap;
236 {
237 
238 	vap->va_type = VNON;
239 	/* XXX These next two used to be one line, but for a GCC bug. */
240 	vap->va_size = VNOVAL;
241 	vap->va_bytes = VNOVAL;
242 	vap->va_mode = vap->va_nlink = vap->va_uid = vap->va_gid =
243 		vap->va_fsid = vap->va_fileid =
244 		vap->va_blocksize = vap->va_rdev =
245 		vap->va_atime.ts_sec = vap->va_atime.ts_nsec =
246 		vap->va_mtime.ts_sec = vap->va_mtime.ts_nsec =
247 		vap->va_ctime.ts_sec = vap->va_ctime.ts_nsec =
248 		vap->va_flags = vap->va_gen = VNOVAL;
249 	vap->va_vaflags = 0;
250 }
251 
252 /*
253  * Routines having to do with the management of the vnode table.
254  */
255 extern int (**dead_vnodeop_p)();
256 extern void vclean();
257 long numvnodes;
258 
259 /*
260  * Return the next vnode from the free list.
261  */
262 getnewvnode(tag, mp, vops, vpp)
263 	enum vtagtype tag;
264 	struct mount *mp;
265 	int (**vops)();
266 	struct vnode **vpp;
267 {
268 	register struct vnode *vp;
269 	int s;
270 
271 	if ((vnode_free_list.tqh_first == NULL &&
272 	     numvnodes < 2 * desiredvnodes) ||
273 	    numvnodes < desiredvnodes) {
274 		vp = (struct vnode *)malloc((u_long)sizeof *vp,
275 		    M_VNODE, M_WAITOK);
276 		bzero((char *)vp, sizeof *vp);
277 		numvnodes++;
278 	} else {
279 		if ((vp = vnode_free_list.tqh_first) == NULL) {
280 			tablefull("vnode");
281 			*vpp = 0;
282 			return (ENFILE);
283 		}
284 		if (vp->v_usecount) {
285 			vprint("free vnode", vp);
286 			panic("free vnode isn't");
287 		}
288 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
289 		/* see comment on why 0xdeadb is set at end of vgone (below) */
290 		vp->v_freelist.tqe_prev = (struct vnode **)0xdeadb;
291 		vp->v_lease = NULL;
292 		if (vp->v_type != VBAD)
293 			vgone(vp);
294 #ifdef DIAGNOSTIC
295 		if (vp->v_data) {
296 			vprint("cleaned vnode", vp);
297 			panic("cleaned vnode isn't");
298 		}
299 		s = splbio();
300 		if (vp->v_numoutput)
301 			panic("Clean vnode has pending I/O's");
302 		splx(s);
303 #endif
304 		vp->v_flag = 0;
305 		vp->v_lastr = 0;
306 		vp->v_ralen = 0;
307 		vp->v_maxra = 0;
308 		vp->v_lastw = 0;
309 		vp->v_lasta = 0;
310 		vp->v_cstart = 0;
311 		vp->v_clen = 0;
312 		vp->v_socket = 0;
313 	}
314 	vp->v_type = VNON;
315 	cache_purge(vp);
316 	vp->v_tag = tag;
317 	vp->v_op = vops;
318 	insmntque(vp, mp);
319 	*vpp = vp;
320 	vp->v_usecount = 1;
321 	vp->v_data = 0;
322 	return (0);
323 }
324 
325 /*
326  * Move a vnode from one mount queue to another.
327  */
328 insmntque(vp, mp)
329 	register struct vnode *vp;
330 	register struct mount *mp;
331 {
332 
333 	/*
334 	 * Delete from old mount point vnode list, if on one.
335 	 */
336 	if (vp->v_mount != NULL)
337 		LIST_REMOVE(vp, v_mntvnodes);
338 	/*
339 	 * Insert into list of vnodes for the new mount point, if available.
340 	 */
341 	if ((vp->v_mount = mp) == NULL)
342 		return;
343 	LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes);
344 }
345 
346 /*
347  * Update outstanding I/O count and do wakeup if requested.
348  */
349 vwakeup(bp)
350 	register struct buf *bp;
351 {
352 	register struct vnode *vp;
353 
354 	bp->b_flags &= ~B_WRITEINPROG;
355 	if (vp = bp->b_vp) {
356 		if (--vp->v_numoutput < 0)
357 			panic("vwakeup: neg numoutput");
358 		if ((vp->v_flag & VBWAIT) && vp->v_numoutput <= 0) {
359 			vp->v_flag &= ~VBWAIT;
360 			wakeup((caddr_t)&vp->v_numoutput);
361 		}
362 	}
363 }
364 
365 /*
366  * Flush out and invalidate all buffers associated with a vnode.
367  * Called with the underlying object locked.
368  */
369 int
370 vinvalbuf(vp, flags, cred, p, slpflag, slptimeo)
371 	register struct vnode *vp;
372 	int flags;
373 	struct ucred *cred;
374 	struct proc *p;
375 	int slpflag, slptimeo;
376 {
377 	register struct buf *bp;
378 	struct buf *nbp, *blist;
379 	int s, error;
380 
381 	if (flags & V_SAVE) {
382 		if (error = VOP_FSYNC(vp, cred, MNT_WAIT, p))
383 			return (error);
384 		if (vp->v_dirtyblkhd.lh_first != NULL)
385 			panic("vinvalbuf: dirty bufs");
386 	}
387 	for (;;) {
388 		if ((blist = vp->v_cleanblkhd.lh_first) && flags & V_SAVEMETA)
389 			while (blist && blist->b_lblkno < 0)
390 				blist = blist->b_vnbufs.le_next;
391 		if (!blist && (blist = vp->v_dirtyblkhd.lh_first) &&
392 		    (flags & V_SAVEMETA))
393 			while (blist && blist->b_lblkno < 0)
394 				blist = blist->b_vnbufs.le_next;
395 		if (!blist)
396 			break;
397 
398 		for (bp = blist; bp; bp = nbp) {
399 			nbp = bp->b_vnbufs.le_next;
400 			if (flags & V_SAVEMETA && bp->b_lblkno < 0)
401 				continue;
402 			s = splbio();
403 			if (bp->b_flags & B_BUSY) {
404 				bp->b_flags |= B_WANTED;
405 				error = tsleep((caddr_t)bp,
406 					slpflag | (PRIBIO + 1), "vinvalbuf",
407 					slptimeo);
408 				splx(s);
409 				if (error)
410 					return (error);
411 				break;
412 			}
413 			bremfree(bp);
414 			bp->b_flags |= B_BUSY;
415 			splx(s);
416 			/*
417 			 * XXX Since there are no node locks for NFS, I believe
418 			 * there is a slight chance that a delayed write will
419 			 * occur while sleeping just above, so check for it.
420 			 */
421 			if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) {
422 				(void) VOP_BWRITE(bp);
423 				break;
424 			}
425 			bp->b_flags |= B_INVAL;
426 			brelse(bp);
427 		}
428 	}
429 	if (!(flags & V_SAVEMETA) &&
430 	    (vp->v_dirtyblkhd.lh_first || vp->v_cleanblkhd.lh_first))
431 		panic("vinvalbuf: flush failed");
432 	return (0);
433 }
434 
435 void
436 vflushbuf(vp, sync)
437 	register struct vnode *vp;
438 	int sync;
439 {
440 	register struct buf *bp, *nbp;
441 	int s;
442 
443 loop:
444 	s = splbio();
445 	for (bp = vp->v_dirtyblkhd.lh_first; bp; bp = nbp) {
446 		nbp = bp->b_vnbufs.le_next;
447 		if ((bp->b_flags & B_BUSY))
448 			continue;
449 		if ((bp->b_flags & B_DELWRI) == 0)
450 			panic("vflushbuf: not dirty");
451 		bremfree(bp);
452 		bp->b_flags |= B_BUSY;
453 		splx(s);
454 		/*
455 		 * Wait for I/O associated with indirect blocks to complete,
456 		 * since there is no way to quickly wait for them below.
457 		 */
458 		if (bp->b_vp == vp || sync == 0)
459 			(void) bawrite(bp);
460 		else
461 			(void) bwrite(bp);
462 		goto loop;
463 	}
464 	if (sync == 0) {
465 		splx(s);
466 		return;
467 	}
468 	while (vp->v_numoutput) {
469 		vp->v_flag |= VBWAIT;
470 		tsleep((caddr_t)&vp->v_numoutput, PRIBIO + 1, "vflushbuf", 0);
471 	}
472 	splx(s);
473 	if (vp->v_dirtyblkhd.lh_first != NULL) {
474 		vprint("vflushbuf: dirty", vp);
475 		goto loop;
476 	}
477 }
478 
479 /*
480  * Associate a buffer with a vnode.
481  */
482 bgetvp(vp, bp)
483 	register struct vnode *vp;
484 	register struct buf *bp;
485 {
486 
487 	if (bp->b_vp)
488 		panic("bgetvp: not free");
489 	VHOLD(vp);
490 	bp->b_vp = vp;
491 	if (vp->v_type == VBLK || vp->v_type == VCHR)
492 		bp->b_dev = vp->v_rdev;
493 	else
494 		bp->b_dev = NODEV;
495 	/*
496 	 * Insert onto list for new vnode.
497 	 */
498 	bufinsvn(bp, &vp->v_cleanblkhd);
499 }
500 
501 /*
502  * Disassociate a buffer from a vnode.
503  */
504 brelvp(bp)
505 	register struct buf *bp;
506 {
507 	struct vnode *vp;
508 
509 	if (bp->b_vp == (struct vnode *) 0)
510 		panic("brelvp: NULL");
511 	/*
512 	 * Delete from old vnode list, if on one.
513 	 */
514 	if (bp->b_vnbufs.le_next != NOLIST)
515 		bufremvn(bp);
516 	vp = bp->b_vp;
517 	bp->b_vp = (struct vnode *) 0;
518 	HOLDRELE(vp);
519 }
520 
521 /*
522  * Reassign a buffer from one vnode to another.
523  * Used to assign file specific control information
524  * (indirect blocks) to the vnode to which they belong.
525  */
526 reassignbuf(bp, newvp)
527 	register struct buf *bp;
528 	register struct vnode *newvp;
529 {
530 	register struct buflists *listheadp;
531 
532 	if (newvp == NULL) {
533 		printf("reassignbuf: NULL");
534 		return;
535 	}
536 	/*
537 	 * Delete from old vnode list, if on one.
538 	 */
539 	if (bp->b_vnbufs.le_next != NOLIST)
540 		bufremvn(bp);
541 	/*
542 	 * If dirty, put on list of dirty buffers;
543 	 * otherwise insert onto list of clean buffers.
544 	 */
545 	if (bp->b_flags & B_DELWRI)
546 		listheadp = &newvp->v_dirtyblkhd;
547 	else
548 		listheadp = &newvp->v_cleanblkhd;
549 	bufinsvn(bp, listheadp);
550 }
551 
552 /*
553  * Create a vnode for a block device.
554  * Used for root filesystem, argdev, and swap areas.
555  * Also used for memory file system special devices.
556  */
557 bdevvp(dev, vpp)
558 	dev_t dev;
559 	struct vnode **vpp;
560 {
561 
562 	return (getdevvp(dev, vpp, VBLK));
563 }
564 
565 /*
566  * Create a vnode for a character device.
567  * Used for kernfs and some console handling.
568  */
569 cdevvp(dev, vpp)
570 	dev_t dev;
571 	struct vnode **vpp;
572 {
573 
574 	return (getdevvp(dev, vpp, VCHR));
575 }
576 
577 /*
578  * Create a vnode for a device.
579  * Used by bdevvp (block device) for root file system etc.,
580  * and by cdevvp (character device) for console and kernfs.
581  */
582 getdevvp(dev, vpp, type)
583 	dev_t dev;
584 	struct vnode **vpp;
585 	enum vtype type;
586 {
587 	register struct vnode *vp;
588 	struct vnode *nvp;
589 	int error;
590 
591 	if (dev == NODEV)
592 		return (0);
593 	error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
594 	if (error) {
595 		*vpp = NULLVP;
596 		return (error);
597 	}
598 	vp = nvp;
599 	vp->v_type = type;
600 	if (nvp = checkalias(vp, dev, (struct mount *)0)) {
601 		vput(vp);
602 		vp = nvp;
603 	}
604 	*vpp = vp;
605 	return (0);
606 }
607 
608 /*
609  * Check to see if the new vnode represents a special device
610  * for which we already have a vnode (either because of
611  * bdevvp() or because of a different vnode representing
612  * the same block device). If such an alias exists, deallocate
613  * the existing contents and return the aliased vnode. The
614  * caller is responsible for filling it with its new contents.
615  */
616 struct vnode *
617 checkalias(nvp, nvp_rdev, mp)
618 	register struct vnode *nvp;
619 	dev_t nvp_rdev;
620 	struct mount *mp;
621 {
622 	register struct vnode *vp;
623 	struct vnode **vpp;
624 
625 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
626 		return (NULLVP);
627 
628 	vpp = &speclisth[SPECHASH(nvp_rdev)];
629 loop:
630 	for (vp = *vpp; vp; vp = vp->v_specnext) {
631 		if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type)
632 			continue;
633 		/*
634 		 * Alias, but not in use, so flush it out.
635 		 */
636 		if (vp->v_usecount == 0) {
637 			vgone(vp);
638 			goto loop;
639 		}
640 		if (vget(vp, 1))
641 			goto loop;
642 		break;
643 	}
644 	if (vp == NULL || vp->v_tag != VT_NON || vp->v_type != VBLK) {
645 		MALLOC(nvp->v_specinfo, struct specinfo *,
646 			sizeof(struct specinfo), M_VNODE, M_WAITOK);
647 		nvp->v_rdev = nvp_rdev;
648 		nvp->v_hashchain = vpp;
649 		nvp->v_specnext = *vpp;
650 		nvp->v_specflags = 0;
651 		*vpp = nvp;
652 		if (vp != NULL) {
653 			nvp->v_flag |= VALIASED;
654 			vp->v_flag |= VALIASED;
655 			vput(vp);
656 		}
657 		return (NULLVP);
658 	}
659 	VOP_UNLOCK(vp);
660 	vclean(vp, 0);
661 	vp->v_op = nvp->v_op;
662 	vp->v_tag = nvp->v_tag;
663 	nvp->v_type = VNON;
664 	insmntque(vp, mp);
665 	return (vp);
666 }
667 
668 /*
669  * Grab a particular vnode from the free list, increment its
670  * reference count and lock it. The vnode lock bit is set the
671  * vnode is being eliminated in vgone. The process is awakened
672  * when the transition is completed, and an error returned to
673  * indicate that the vnode is no longer usable (possibly having
674  * been changed to a new file system type).
675  */
676 int
677 vget(vp, lockflag)
678 	register struct vnode *vp;
679 	int lockflag;
680 {
681 
682 	/*
683 	 * If the vnode is in the process of being cleaned out for
684 	 * another use, we wait for the cleaning to finish and then
685 	 * return failure. Cleaning is determined either by checking
686 	 * that the VXLOCK flag is set, or that the use count is
687 	 * zero with the back pointer set to show that it has been
688 	 * removed from the free list by getnewvnode. The VXLOCK
689 	 * flag may not have been set yet because vclean is blocked in
690 	 * the VOP_LOCK call waiting for the VOP_INACTIVE to complete.
691 	 */
692 	if ((vp->v_flag & VXLOCK) ||
693 	    (vp->v_usecount == 0 &&
694 	     vp->v_freelist.tqe_prev == (struct vnode **)0xdeadb)) {
695 		vp->v_flag |= VXWANT;
696 		tsleep((caddr_t)vp, PINOD, "vget", 0);
697 		return (1);
698 	}
699 	if (vp->v_usecount == 0)
700 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
701 	vp->v_usecount++;
702 	if (lockflag)
703 		VOP_LOCK(vp);
704 	return (0);
705 }
706 
707 /*
708  * Vnode reference, just increment the count
709  */
710 void
711 vref(vp)
712 	struct vnode *vp;
713 {
714 
715 	if (vp->v_usecount <= 0)
716 		panic("vref used where vget required");
717 	vp->v_usecount++;
718 }
719 
720 /*
721  * vput(), just unlock and vrele()
722  */
723 void
724 vput(vp)
725 	register struct vnode *vp;
726 {
727 
728 	VOP_UNLOCK(vp);
729 	vrele(vp);
730 }
731 
732 /*
733  * Vnode release.
734  * If count drops to zero, call inactive routine and return to freelist.
735  */
736 void
737 vrele(vp)
738 	register struct vnode *vp;
739 {
740 
741 #ifdef DIAGNOSTIC
742 	if (vp == NULL)
743 		panic("vrele: null vp");
744 #endif
745 	vp->v_usecount--;
746 	if (vp->v_usecount > 0)
747 		return;
748 #ifdef DIAGNOSTIC
749 	if (vp->v_usecount != 0 || vp->v_writecount != 0) {
750 		vprint("vrele: bad ref count", vp);
751 		panic("vrele: ref cnt");
752 	}
753 #endif
754 	/*
755 	 * insert at tail of LRU list
756 	 */
757 	TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
758 	VOP_INACTIVE(vp);
759 }
760 
761 /*
762  * Page or buffer structure gets a reference.
763  */
764 void
765 vhold(vp)
766 	register struct vnode *vp;
767 {
768 
769 	vp->v_holdcnt++;
770 }
771 
772 /*
773  * Page or buffer structure frees a reference.
774  */
775 void
776 holdrele(vp)
777 	register struct vnode *vp;
778 {
779 
780 	if (vp->v_holdcnt <= 0)
781 		panic("holdrele: holdcnt");
782 	vp->v_holdcnt--;
783 }
784 
785 /*
786  * Remove any vnodes in the vnode table belonging to mount point mp.
787  *
788  * If MNT_NOFORCE is specified, there should not be any active ones,
789  * return error if any are found (nb: this is a user error, not a
790  * system error). If MNT_FORCE is specified, detach any active vnodes
791  * that are found.
792  */
793 #ifdef DEBUG
794 int busyprt = 0;	/* print out busy vnodes */
795 struct ctldebug debug1 = { "busyprt", &busyprt };
796 #endif
797 
798 vflush(mp, skipvp, flags)
799 	struct mount *mp;
800 	struct vnode *skipvp;
801 	int flags;
802 {
803 	register struct vnode *vp, *nvp;
804 	int busy = 0;
805 
806 	if ((mp->mnt_flag & MNT_MPBUSY) == 0)
807 		panic("vflush: not busy");
808 loop:
809 	for (vp = mp->mnt_vnodelist.lh_first; vp; vp = nvp) {
810 		if (vp->v_mount != mp)
811 			goto loop;
812 		nvp = vp->v_mntvnodes.le_next;
813 		/*
814 		 * Skip over a selected vnode.
815 		 */
816 		if (vp == skipvp)
817 			continue;
818 		/*
819 		 * Skip over a vnodes marked VSYSTEM.
820 		 */
821 		if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM))
822 			continue;
823 		/*
824 		 * If WRITECLOSE is set, only flush out regular file
825 		 * vnodes open for writing.
826 		 */
827 		if ((flags & WRITECLOSE) &&
828 		    (vp->v_writecount == 0 || vp->v_type != VREG))
829 			continue;
830 		/*
831 		 * With v_usecount == 0, all we need to do is clear
832 		 * out the vnode data structures and we are done.
833 		 */
834 		if (vp->v_usecount == 0) {
835 			vgone(vp);
836 			continue;
837 		}
838 		/*
839 		 * If FORCECLOSE is set, forcibly close the vnode.
840 		 * For block or character devices, revert to an
841 		 * anonymous device. For all other files, just kill them.
842 		 */
843 		if (flags & FORCECLOSE) {
844 			if (vp->v_type != VBLK && vp->v_type != VCHR) {
845 				vgone(vp);
846 			} else {
847 				vclean(vp, 0);
848 				vp->v_op = spec_vnodeop_p;
849 				insmntque(vp, (struct mount *)0);
850 			}
851 			continue;
852 		}
853 #ifdef DEBUG
854 		if (busyprt)
855 			vprint("vflush: busy vnode", vp);
856 #endif
857 		busy++;
858 	}
859 	if (busy)
860 		return (EBUSY);
861 	return (0);
862 }
863 
864 /*
865  * Disassociate the underlying file system from a vnode.
866  */
867 void
868 vclean(vp, flags)
869 	register struct vnode *vp;
870 	int flags;
871 {
872 	int active;
873 
874 	/*
875 	 * Check to see if the vnode is in use.
876 	 * If so we have to reference it before we clean it out
877 	 * so that its count cannot fall to zero and generate a
878 	 * race against ourselves to recycle it.
879 	 */
880 	if (active = vp->v_usecount)
881 		VREF(vp);
882 	/*
883 	 * Even if the count is zero, the VOP_INACTIVE routine may still
884 	 * have the object locked while it cleans it out. The VOP_LOCK
885 	 * ensures that the VOP_INACTIVE routine is done with its work.
886 	 * For active vnodes, it ensures that no other activity can
887 	 * occur while the underlying object is being cleaned out.
888 	 */
889 	VOP_LOCK(vp);
890 	/*
891 	 * Prevent the vnode from being recycled or
892 	 * brought into use while we clean it out.
893 	 */
894 	if (vp->v_flag & VXLOCK)
895 		panic("vclean: deadlock");
896 	vp->v_flag |= VXLOCK;
897 	/*
898 	 * Clean out any buffers associated with the vnode.
899 	 */
900 	if (flags & DOCLOSE)
901 		vinvalbuf(vp, V_SAVE, NOCRED, NULL, 0, 0);
902 	/*
903 	 * Any other processes trying to obtain this lock must first
904 	 * wait for VXLOCK to clear, then call the new lock operation.
905 	 */
906 	VOP_UNLOCK(vp);
907 	/*
908 	 * If purging an active vnode, it must be closed and
909 	 * deactivated before being reclaimed.
910 	 */
911 	if (active) {
912 		if (flags & DOCLOSE)
913 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, NULL);
914 		VOP_INACTIVE(vp);
915 	}
916 	/*
917 	 * Reclaim the vnode.
918 	 */
919 	if (VOP_RECLAIM(vp))
920 		panic("vclean: cannot reclaim");
921 	if (active)
922 		vrele(vp);
923 
924 	/*
925 	 * Done with purge, notify sleepers of the grim news.
926 	 */
927 	vp->v_op = dead_vnodeop_p;
928 	vp->v_tag = VT_NON;
929 	vp->v_flag &= ~VXLOCK;
930 	if (vp->v_flag & VXWANT) {
931 		vp->v_flag &= ~VXWANT;
932 		wakeup((caddr_t)vp);
933 	}
934 }
935 
936 /*
937  * Eliminate all activity associated with  the requested vnode
938  * and with all vnodes aliased to the requested vnode.
939  */
940 void
941 vgoneall(vp)
942 	register struct vnode *vp;
943 {
944 	register struct vnode *vq;
945 
946 	if (vp->v_flag & VALIASED) {
947 		/*
948 		 * If a vgone (or vclean) is already in progress,
949 		 * wait until it is done and return.
950 		 */
951 		if (vp->v_flag & VXLOCK) {
952 			vp->v_flag |= VXWANT;
953 			tsleep((caddr_t)vp, PINOD, "vgoneall", 0);
954 			return;
955 		}
956 		/*
957 		 * Ensure that vp will not be vgone'd while we
958 		 * are eliminating its aliases.
959 		 */
960 		vp->v_flag |= VXLOCK;
961 		while (vp->v_flag & VALIASED) {
962 			for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
963 				if (vq->v_rdev != vp->v_rdev ||
964 				    vq->v_type != vp->v_type || vp == vq)
965 					continue;
966 				vgone(vq);
967 				break;
968 			}
969 		}
970 		/*
971 		 * Remove the lock so that vgone below will
972 		 * really eliminate the vnode after which time
973 		 * vgone will awaken any sleepers.
974 		 */
975 		vp->v_flag &= ~VXLOCK;
976 	}
977 	vgone(vp);
978 }
979 
980 /*
981  * Eliminate all activity associated with a vnode
982  * in preparation for reuse.
983  */
984 void
985 vgone(vp)
986 	register struct vnode *vp;
987 {
988 	register struct vnode *vq;
989 	struct vnode *vx;
990 
991 	/*
992 	 * If a vgone (or vclean) is already in progress,
993 	 * wait until it is done and return.
994 	 */
995 	if (vp->v_flag & VXLOCK) {
996 		vp->v_flag |= VXWANT;
997 		tsleep((caddr_t)vp, PINOD, "vgone", 0);
998 		return;
999 	}
1000 	/*
1001 	 * Clean out the filesystem specific data.
1002 	 */
1003 	vclean(vp, DOCLOSE);
1004 	/*
1005 	 * Delete from old mount point vnode list, if on one.
1006 	 */
1007 	if (vp->v_mount != NULL) {
1008 		LIST_REMOVE(vp, v_mntvnodes);
1009 		vp->v_mount = NULL;
1010 	}
1011 	/*
1012 	 * If special device, remove it from special device alias list.
1013 	 */
1014 	if (vp->v_type == VBLK || vp->v_type == VCHR) {
1015 		if (*vp->v_hashchain == vp) {
1016 			*vp->v_hashchain = vp->v_specnext;
1017 		} else {
1018 			for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1019 				if (vq->v_specnext != vp)
1020 					continue;
1021 				vq->v_specnext = vp->v_specnext;
1022 				break;
1023 			}
1024 			if (vq == NULL)
1025 				panic("missing bdev");
1026 		}
1027 		if (vp->v_flag & VALIASED) {
1028 			vx = NULL;
1029 			for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1030 				if (vq->v_rdev != vp->v_rdev ||
1031 				    vq->v_type != vp->v_type)
1032 					continue;
1033 				if (vx)
1034 					break;
1035 				vx = vq;
1036 			}
1037 			if (vx == NULL)
1038 				panic("missing alias");
1039 			if (vq == NULL)
1040 				vx->v_flag &= ~VALIASED;
1041 			vp->v_flag &= ~VALIASED;
1042 		}
1043 		FREE(vp->v_specinfo, M_VNODE);
1044 		vp->v_specinfo = NULL;
1045 	}
1046 	/*
1047 	 * If it is on the freelist and not already at the head,
1048 	 * move it to the head of the list. The test of the back
1049 	 * pointer and the reference count of zero is because
1050 	 * it will be removed from the free list by getnewvnode,
1051 	 * but will not have its reference count incremented until
1052 	 * after calling vgone. If the reference count were
1053 	 * incremented first, vgone would (incorrectly) try to
1054 	 * close the previous instance of the underlying object.
1055 	 * So, the back pointer is explicitly set to `0xdeadb' in
1056 	 * getnewvnode after removing it from the freelist to ensure
1057 	 * that we do not try to move it here.
1058 	 */
1059 	if (vp->v_usecount == 0 &&
1060 	    vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb &&
1061 	    vnode_free_list.tqh_first != vp) {
1062 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1063 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
1064 	}
1065 	vp->v_type = VBAD;
1066 }
1067 
1068 /*
1069  * Lookup a vnode by device number.
1070  */
1071 vfinddev(dev, type, vpp)
1072 	dev_t dev;
1073 	enum vtype type;
1074 	struct vnode **vpp;
1075 {
1076 	register struct vnode *vp;
1077 
1078 	for (vp = speclisth[SPECHASH(dev)]; vp; vp = vp->v_specnext) {
1079 		if (dev != vp->v_rdev || type != vp->v_type)
1080 			continue;
1081 		*vpp = vp;
1082 		return (1);
1083 	}
1084 	return (0);
1085 }
1086 
1087 /*
1088  * Calculate the total number of references to a special device.
1089  */
1090 int
1091 vcount(vp)
1092 	register struct vnode *vp;
1093 {
1094 	register struct vnode *vq, *vnext;
1095 	int count;
1096 
1097 loop:
1098 	if ((vp->v_flag & VALIASED) == 0)
1099 		return (vp->v_usecount);
1100 	for (count = 0, vq = *vp->v_hashchain; vq; vq = vnext) {
1101 		vnext = vq->v_specnext;
1102 		if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type)
1103 			continue;
1104 		/*
1105 		 * Alias, but not in use, so flush it out.
1106 		 */
1107 		if (vq->v_usecount == 0 && vq != vp) {
1108 			vgone(vq);
1109 			goto loop;
1110 		}
1111 		count += vq->v_usecount;
1112 	}
1113 	return (count);
1114 }
1115 
1116 /*
1117  * Print out a description of a vnode.
1118  */
1119 static char *typename[] =
1120    { "VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD" };
1121 
1122 void
1123 vprint(label, vp)
1124 	char *label;
1125 	register struct vnode *vp;
1126 {
1127 	char buf[64];
1128 
1129 	if (label != NULL)
1130 		printf("%s: ", label);
1131 	printf("type %s, usecount %d, writecount %d, refcount %d,",
1132 		typename[vp->v_type], vp->v_usecount, vp->v_writecount,
1133 		vp->v_holdcnt);
1134 	buf[0] = '\0';
1135 	if (vp->v_flag & VROOT)
1136 		strcat(buf, "|VROOT");
1137 	if (vp->v_flag & VTEXT)
1138 		strcat(buf, "|VTEXT");
1139 	if (vp->v_flag & VSYSTEM)
1140 		strcat(buf, "|VSYSTEM");
1141 	if (vp->v_flag & VXLOCK)
1142 		strcat(buf, "|VXLOCK");
1143 	if (vp->v_flag & VXWANT)
1144 		strcat(buf, "|VXWANT");
1145 	if (vp->v_flag & VBWAIT)
1146 		strcat(buf, "|VBWAIT");
1147 	if (vp->v_flag & VALIASED)
1148 		strcat(buf, "|VALIASED");
1149 	if (buf[0] != '\0')
1150 		printf(" flags (%s)", &buf[1]);
1151 	if (vp->v_data == NULL) {
1152 		printf("\n");
1153 	} else {
1154 		printf("\n\t");
1155 		VOP_PRINT(vp);
1156 	}
1157 }
1158 
1159 #ifdef DEBUG
1160 /*
1161  * List all of the locked vnodes in the system.
1162  * Called when debugging the kernel.
1163  */
1164 printlockedvnodes()
1165 {
1166 	register struct mount *mp;
1167 	register struct vnode *vp;
1168 
1169 	printf("Locked vnodes\n");
1170 	for (mp = mountlist.cqh_first; mp != (void *)&mountlist;
1171 	     mp = mp->mnt_list.cqe_next) {
1172 		for (vp = mp->mnt_vnodelist.lh_first;
1173 		     vp != NULL;
1174 		     vp = vp->v_mntvnodes.le_next)
1175 			if (VOP_ISLOCKED(vp))
1176 				vprint((char *)0, vp);
1177 	}
1178 }
1179 #endif
1180 
1181 int kinfo_vdebug = 1;
1182 int kinfo_vgetfailed;
1183 #define KINFO_VNODESLOP	10
1184 /*
1185  * Dump vnode list (via sysctl).
1186  * Copyout address of vnode followed by vnode.
1187  */
1188 /* ARGSUSED */
1189 sysctl_vnode(where, sizep)
1190 	char *where;
1191 	size_t *sizep;
1192 {
1193 	register struct mount *mp, *nmp;
1194 	struct vnode *vp;
1195 	register char *bp = where, *savebp;
1196 	char *ewhere;
1197 	int error;
1198 
1199 #define VPTRSZ	sizeof (struct vnode *)
1200 #define VNODESZ	sizeof (struct vnode)
1201 	if (where == NULL) {
1202 		*sizep = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ);
1203 		return (0);
1204 	}
1205 	ewhere = where + *sizep;
1206 
1207 	for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) {
1208 		nmp = mp->mnt_list.cqe_next;
1209 		if (vfs_busy(mp))
1210 			continue;
1211 		savebp = bp;
1212 again:
1213 		for (vp = mp->mnt_vnodelist.lh_first;
1214 		     vp != NULL;
1215 		     vp = vp->v_mntvnodes.le_next) {
1216 			/*
1217 			 * Check that the vp is still associated with
1218 			 * this filesystem.  RACE: could have been
1219 			 * recycled onto the same filesystem.
1220 			 */
1221 			if (vp->v_mount != mp) {
1222 				if (kinfo_vdebug)
1223 					printf("kinfo: vp changed\n");
1224 				bp = savebp;
1225 				goto again;
1226 			}
1227 			if (bp + VPTRSZ + VNODESZ > ewhere) {
1228 				*sizep = bp - where;
1229 				return (ENOMEM);
1230 			}
1231 			if ((error = copyout((caddr_t)&vp, bp, VPTRSZ)) ||
1232 			   (error = copyout((caddr_t)vp, bp + VPTRSZ, VNODESZ)))
1233 				return (error);
1234 			bp += VPTRSZ + VNODESZ;
1235 		}
1236 		vfs_unbusy(mp);
1237 	}
1238 
1239 	*sizep = bp - where;
1240 	return (0);
1241 }
1242 
1243 /*
1244  * Check to see if a filesystem is mounted on a block device.
1245  */
1246 int
1247 vfs_mountedon(vp)
1248 	register struct vnode *vp;
1249 {
1250 	register struct vnode *vq;
1251 
1252 	if (vp->v_specflags & SI_MOUNTEDON)
1253 		return (EBUSY);
1254 	if (vp->v_flag & VALIASED) {
1255 		for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1256 			if (vq->v_rdev != vp->v_rdev ||
1257 			    vq->v_type != vp->v_type)
1258 				continue;
1259 			if (vq->v_specflags & SI_MOUNTEDON)
1260 				return (EBUSY);
1261 		}
1262 	}
1263 	return (0);
1264 }
1265 
1266 /*
1267  * Build hash lists of net addresses and hang them off the mount point.
1268  * Called by ufs_mount() to set up the lists of export addresses.
1269  */
1270 static int
1271 vfs_hang_addrlist(mp, nep, argp)
1272 	struct mount *mp;
1273 	struct netexport *nep;
1274 	struct export_args *argp;
1275 {
1276 	register struct netcred *np;
1277 	register struct radix_node_head *rnh;
1278 	register int i;
1279 	struct radix_node *rn;
1280 	struct sockaddr *saddr, *smask = 0;
1281 	struct domain *dom;
1282 	int error;
1283 
1284 	if (argp->ex_addrlen == 0) {
1285 		if (mp->mnt_flag & MNT_DEFEXPORTED)
1286 			return (EPERM);
1287 		np = &nep->ne_defexported;
1288 		np->netc_exflags = argp->ex_flags;
1289 		np->netc_anon = argp->ex_anon;
1290 		np->netc_anon.cr_ref = 1;
1291 		mp->mnt_flag |= MNT_DEFEXPORTED;
1292 		return (0);
1293 	}
1294 	i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
1295 	np = (struct netcred *)malloc(i, M_NETADDR, M_WAITOK);
1296 	bzero((caddr_t)np, i);
1297 	saddr = (struct sockaddr *)(np + 1);
1298 	if (error = copyin(argp->ex_addr, (caddr_t)saddr, argp->ex_addrlen))
1299 		goto out;
1300 	if (saddr->sa_len > argp->ex_addrlen)
1301 		saddr->sa_len = argp->ex_addrlen;
1302 	if (argp->ex_masklen) {
1303 		smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
1304 		error = copyin(argp->ex_addr, (caddr_t)smask, argp->ex_masklen);
1305 		if (error)
1306 			goto out;
1307 		if (smask->sa_len > argp->ex_masklen)
1308 			smask->sa_len = argp->ex_masklen;
1309 	}
1310 	i = saddr->sa_family;
1311 	if ((rnh = nep->ne_rtable[i]) == 0) {
1312 		/*
1313 		 * Seems silly to initialize every AF when most are not
1314 		 * used, do so on demand here
1315 		 */
1316 		for (dom = domains; dom; dom = dom->dom_next)
1317 			if (dom->dom_family == i && dom->dom_rtattach) {
1318 				dom->dom_rtattach((void **)&nep->ne_rtable[i],
1319 					dom->dom_rtoffset);
1320 				break;
1321 			}
1322 		if ((rnh = nep->ne_rtable[i]) == 0) {
1323 			error = ENOBUFS;
1324 			goto out;
1325 		}
1326 	}
1327 	rn = (*rnh->rnh_addaddr)((caddr_t)saddr, (caddr_t)smask, rnh,
1328 		np->netc_rnodes);
1329 	if (rn == 0 || np != (struct netcred *)rn) { /* already exists */
1330 		error = EPERM;
1331 		goto out;
1332 	}
1333 	np->netc_exflags = argp->ex_flags;
1334 	np->netc_anon = argp->ex_anon;
1335 	np->netc_anon.cr_ref = 1;
1336 	return (0);
1337 out:
1338 	free(np, M_NETADDR);
1339 	return (error);
1340 }
1341 
1342 /* ARGSUSED */
1343 static int
1344 vfs_free_netcred(rn, w)
1345 	struct radix_node *rn;
1346 	caddr_t w;
1347 {
1348 	register struct radix_node_head *rnh = (struct radix_node_head *)w;
1349 
1350 	(*rnh->rnh_deladdr)(rn->rn_key, rn->rn_mask, rnh);
1351 	free((caddr_t)rn, M_NETADDR);
1352 	return (0);
1353 }
1354 
1355 /*
1356  * Free the net address hash lists that are hanging off the mount points.
1357  */
1358 static void
1359 vfs_free_addrlist(nep)
1360 	struct netexport *nep;
1361 {
1362 	register int i;
1363 	register struct radix_node_head *rnh;
1364 
1365 	for (i = 0; i <= AF_MAX; i++)
1366 		if (rnh = nep->ne_rtable[i]) {
1367 			(*rnh->rnh_walktree)(rnh, vfs_free_netcred,
1368 			    (caddr_t)rnh);
1369 			free((caddr_t)rnh, M_RTABLE);
1370 			nep->ne_rtable[i] = 0;
1371 		}
1372 }
1373 
1374 int
1375 vfs_export(mp, nep, argp)
1376 	struct mount *mp;
1377 	struct netexport *nep;
1378 	struct export_args *argp;
1379 {
1380 	int error;
1381 
1382 	if (argp->ex_flags & MNT_DELEXPORT) {
1383 		vfs_free_addrlist(nep);
1384 		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
1385 	}
1386 	if (argp->ex_flags & MNT_EXPORTED) {
1387 		if (error = vfs_hang_addrlist(mp, nep, argp))
1388 			return (error);
1389 		mp->mnt_flag |= MNT_EXPORTED;
1390 	}
1391 	return (0);
1392 }
1393 
1394 struct netcred *
1395 vfs_export_lookup(mp, nep, nam)
1396 	register struct mount *mp;
1397 	struct netexport *nep;
1398 	struct mbuf *nam;
1399 {
1400 	register struct netcred *np;
1401 	register struct radix_node_head *rnh;
1402 	struct sockaddr *saddr;
1403 
1404 	np = NULL;
1405 	if (mp->mnt_flag & MNT_EXPORTED) {
1406 		/*
1407 		 * Lookup in the export list first.
1408 		 */
1409 		if (nam != NULL) {
1410 			saddr = mtod(nam, struct sockaddr *);
1411 			rnh = nep->ne_rtable[saddr->sa_family];
1412 			if (rnh != NULL) {
1413 				np = (struct netcred *)
1414 					(*rnh->rnh_matchaddr)((caddr_t)saddr,
1415 							      rnh);
1416 				if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
1417 					np = NULL;
1418 			}
1419 		}
1420 		/*
1421 		 * If no address match, use the default if it exists.
1422 		 */
1423 		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
1424 			np = &nep->ne_defexported;
1425 	}
1426 	return (np);
1427 }
1428 
1429 /*
1430  * Do the usual access checking.
1431  * file_mode, uid and gid are from the vnode in question,
1432  * while acc_mode and cred are from the VOP_ACCESS parameter list
1433  */
1434 int
1435 vaccess(file_mode, uid, gid, acc_mode, cred)
1436 	mode_t file_mode;
1437 	uid_t uid;
1438 	gid_t gid;
1439 	mode_t acc_mode;
1440 	struct ucred *cred;
1441 {
1442 	mode_t mask;
1443 	int i;
1444 	register gid_t *gp;
1445 
1446 	/* User id 0 always gets access. */
1447 	if (cred->cr_uid == 0)
1448 		return 0;
1449 
1450 	mask = 0;
1451 
1452 	/* Otherwise, check the owner. */
1453 	if (cred->cr_uid == uid) {
1454 		if (acc_mode & VEXEC)
1455 			mask |= S_IXUSR;
1456 		if (acc_mode & VREAD)
1457 			mask |= S_IRUSR;
1458 		if (acc_mode & VWRITE)
1459 			mask |= S_IWUSR;
1460 		return (file_mode & mask) == mask ? 0 : EACCES;
1461 	}
1462 
1463 	/* Otherwise, check the groups. */
1464 	if (cred->cr_gid == gid || groupmember(gid, cred)) {
1465 		if (acc_mode & VEXEC)
1466 			mask |= S_IXGRP;
1467 		if (acc_mode & VREAD)
1468 			mask |= S_IRGRP;
1469 		if (acc_mode & VWRITE)
1470 			mask |= S_IWGRP;
1471 		return (file_mode & mask) == mask ? 0 : EACCES;
1472 	}
1473 
1474 	/* Otherwise, check everyone else. */
1475 	if (acc_mode & VEXEC)
1476 		mask |= S_IXOTH;
1477 	if (acc_mode & VREAD)
1478 		mask |= S_IROTH;
1479 	if (acc_mode & VWRITE)
1480 		mask |= S_IWOTH;
1481 	return (file_mode & mask) == mask ? 0 : EACCES;
1482 }
1483 
1484 /*
1485  * Unmount all file systems.
1486  * We traverse the list in reverse order under the assumption that doing so
1487  * will avoid needing to worry about dependencies.
1488  */
1489 void
1490 vfs_unmountall()
1491 {
1492 	register struct mount *mp, *nmp;
1493 	int allerror, error;
1494 
1495 	for (allerror = 0,
1496 	     mp = mountlist.cqh_last; mp != (void *)&mountlist; mp = nmp) {
1497 		nmp = mp->mnt_list.cqe_prev;
1498 		if (error = dounmount(mp, MNT_FORCE, &proc0)) {
1499 			printf("unmount of %s failed with error %d\n",
1500 			    mp->mnt_stat.f_mntonname, error);
1501 			allerror = 1;
1502 		}
1503 	}
1504 	if (allerror)
1505 		printf("WARNING: some file systems would not unmount\n");
1506 }
1507 
1508 /*
1509  * Sync and unmount file systems before shutting down.
1510  */
1511 void
1512 vfs_shutdown()
1513 {
1514 	register struct buf *bp;
1515 	int iter, nbusy;
1516 
1517 	/* XXX Should suspend scheduling. */
1518 	(void) spl0();
1519 
1520 	printf("syncing disks... ");
1521 
1522 	if (panicstr == 0) {
1523 		/* Release inodes held by texts before update. */
1524 		vnode_pager_umount(NULL);
1525 #ifdef notdef
1526 		vnshutdown();
1527 #endif
1528 
1529 		/* Sync before unmount, in case we hang on something. */
1530 		sys_sync(&proc0, (void *)0, (int *)0);
1531 
1532 		/* Unmount file systems. */
1533 		vfs_unmountall();
1534 	}
1535 
1536 	/* Sync again after unmount, just in case. */
1537 	sys_sync(&proc0, (void *)0, (int *)0);
1538 
1539 	/* Wait for sync to finish. */
1540 	for (iter = 0; iter < 20; iter++) {
1541 		nbusy = 0;
1542 		for (bp = &buf[nbuf]; --bp >= buf; )
1543 			if ((bp->b_flags & (B_BUSY|B_INVAL)) == B_BUSY)
1544 				nbusy++;
1545 		if (nbusy == 0)
1546 			break;
1547 		printf("%d ", nbusy);
1548 		DELAY(40000 * iter);
1549 	}
1550 	if (nbusy)
1551 		printf("giving up\n");
1552 	else
1553 		printf("done\n");
1554 }
1555