xref: /openbsd-src/sys/kern/vfs_subr.c (revision a28daedfc357b214be5c701aa8ba8adb29a7f1c2)
1 /*	$OpenBSD: vfs_subr.c,v 1.175 2008/11/10 11:53:16 pedro Exp $	*/
2 /*	$NetBSD: vfs_subr.c,v 1.53 1996/04/22 01:39:13 christos Exp $	*/
3 
4 /*
5  * Copyright (c) 1989, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  * (c) UNIX System Laboratories, Inc.
8  * All or some portions of this file are derived from material licensed
9  * to the University of California by American Telephone and Telegraph
10  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
11  * the permission of UNIX System Laboratories, Inc.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  *	@(#)vfs_subr.c	8.13 (Berkeley) 4/18/94
38  */
39 
40 /*
41  * External virtual filesystem routines
42  */
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/proc.h>
47 #include <sys/mount.h>
48 #include <sys/time.h>
49 #include <sys/fcntl.h>
50 #include <sys/kernel.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 #include <sys/syscallargs.h>
61 #include <sys/pool.h>
62 
63 #include <uvm/uvm_extern.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 
73 int	vttoif_tab[9] = {
74 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
75 	S_IFSOCK, S_IFIFO, S_IFMT,
76 };
77 
78 int doforce = 1;		/* 1 => permit forcible unmounting */
79 int prtactive = 0;		/* 1 => print out reclaim of active vnodes */
80 int suid_clear = 1;		/* 1 => clear SUID / SGID on owner change */
81 
82 /*
83  * Insq/Remq for the vnode usage lists.
84  */
85 #define	bufinsvn(bp, dp)	LIST_INSERT_HEAD(dp, bp, b_vnbufs)
86 #define	bufremvn(bp) {							\
87 	LIST_REMOVE(bp, b_vnbufs);					\
88 	LIST_NEXT(bp, b_vnbufs) = NOLIST;				\
89 }
90 
91 struct freelst vnode_hold_list;	/* list of vnodes referencing buffers */
92 struct freelst vnode_free_list;	/* vnode free list */
93 
94 struct mntlist mountlist;	/* mounted filesystem list */
95 
96 void	vclean(struct vnode *, int, struct proc *);
97 void	vhold(struct vnode *);
98 void	vdrop(struct vnode *);
99 
100 void insmntque(struct vnode *, struct mount *);
101 int getdevvp(dev_t, struct vnode **, enum vtype);
102 
103 int vfs_hang_addrlist(struct mount *, struct netexport *,
104 				  struct export_args *);
105 int vfs_free_netcred(struct radix_node *, void *);
106 void vfs_free_addrlist(struct netexport *);
107 void vputonfreelist(struct vnode *);
108 
109 int vflush_vnode(struct vnode *, void *);
110 int maxvnodes;
111 
112 #ifdef DEBUG
113 void printlockedvnodes(void);
114 #endif
115 
116 struct pool vnode_pool;
117 
118 /*
119  * Initialize the vnode management data structures.
120  */
121 void
122 vntblinit(void)
123 {
124 	/* buffer cache may need a vnode for each buffer */
125 	maxvnodes = desiredvnodes;
126 	pool_init(&vnode_pool, sizeof(struct vnode), 0, 0, 0, "vnodes",
127 	    &pool_allocator_nointr);
128 	TAILQ_INIT(&vnode_hold_list);
129 	TAILQ_INIT(&vnode_free_list);
130 	CIRCLEQ_INIT(&mountlist);
131 	/*
132 	 * Initialize the filesystem syncer.
133 	 */
134 	vn_initialize_syncerd();
135 }
136 
137 /*
138  * Mark a mount point as busy. Used to synchronize access and to delay
139  * unmounting.
140  *
141  * Default behaviour is to attempt getting a READ lock and in case of an
142  * ongoing unmount, to wait for it to finish and then return failure.
143  */
144 int
145 vfs_busy(struct mount *mp, int flags)
146 {
147 	int rwflags = 0;
148 
149 	/* new mountpoints need their lock initialised */
150 	if (mp->mnt_lock.rwl_name == NULL)
151 		rw_init(&mp->mnt_lock, "vfslock");
152 
153 	if (flags & VB_WRITE)
154 		rwflags |= RW_WRITE;
155 	else
156 		rwflags |= RW_READ;
157 
158 	if (flags & VB_WAIT)
159 		rwflags |= RW_SLEEPFAIL;
160 	else
161 		rwflags |= RW_NOSLEEP;
162 
163 	if (rw_enter(&mp->mnt_lock, rwflags))
164 		return (EBUSY);
165 
166 	return (0);
167 }
168 
169 /*
170  * Free a busy file system
171  */
172 void
173 vfs_unbusy(struct mount *mp)
174 {
175 	rw_exit(&mp->mnt_lock);
176 }
177 
178 int
179 vfs_isbusy(struct mount *mp)
180 {
181 	if (RWLOCK_OWNER(&mp->mnt_lock) > 0)
182 		return (1);
183 	else
184 		return (0);
185 }
186 
187 /*
188  * Lookup a filesystem type, and if found allocate and initialize
189  * a mount structure for it.
190  *
191  * Devname is usually updated by mount(8) after booting.
192  */
193 int
194 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
195 {
196 	struct vfsconf *vfsp;
197 	struct mount *mp;
198 
199 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
200 		if (!strcmp(vfsp->vfc_name, fstypename))
201 			break;
202 	if (vfsp == NULL)
203 		return (ENODEV);
204 	mp = malloc(sizeof(struct mount), M_MOUNT, M_WAITOK|M_ZERO);
205 	(void)vfs_busy(mp, VB_READ|VB_NOWAIT);
206 	LIST_INIT(&mp->mnt_vnodelist);
207 	mp->mnt_vfc = vfsp;
208 	mp->mnt_op = vfsp->vfc_vfsops;
209 	mp->mnt_flag = MNT_RDONLY;
210 	mp->mnt_vnodecovered = NULLVP;
211 	vfsp->vfc_refcount++;
212 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
213 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
214 	mp->mnt_stat.f_mntonname[0] = '/';
215 	(void)copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
216 	*mpp = mp;
217  	return (0);
218  }
219 
220 /*
221  * Lookup a mount point by filesystem identifier.
222  */
223 struct mount *
224 vfs_getvfs(fsid_t *fsid)
225 {
226 	struct mount *mp;
227 
228 	CIRCLEQ_FOREACH(mp, &mountlist, mnt_list) {
229 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
230 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
231 			return (mp);
232 		}
233 	}
234 
235 	return (NULL);
236 }
237 
238 
239 /*
240  * Get a new unique fsid
241  */
242 void
243 vfs_getnewfsid(struct mount *mp)
244 {
245 	static u_short xxxfs_mntid;
246 
247 	fsid_t tfsid;
248 	int mtype;
249 
250 	mtype = mp->mnt_vfc->vfc_typenum;
251 	mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev + mtype, 0);
252 	mp->mnt_stat.f_fsid.val[1] = mtype;
253 	if (xxxfs_mntid == 0)
254 		++xxxfs_mntid;
255 	tfsid.val[0] = makedev(nblkdev + mtype, xxxfs_mntid);
256 	tfsid.val[1] = mtype;
257 	if (!CIRCLEQ_EMPTY(&mountlist)) {
258 		while (vfs_getvfs(&tfsid)) {
259 			tfsid.val[0]++;
260 			xxxfs_mntid++;
261 		}
262 	}
263 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
264 }
265 
266 /*
267  * Make a 'unique' number from a mount type name.
268  * Note that this is no longer used for ffs which
269  * now has an on-disk filesystem id.
270  */
271 long
272 makefstype(char *type)
273 {
274 	long rv;
275 
276 	for (rv = 0; *type; type++) {
277 		rv <<= 2;
278 		rv ^= *type;
279 	}
280 	return rv;
281 }
282 
283 /*
284  * Set vnode attributes to VNOVAL
285  */
286 void
287 vattr_null(struct vattr *vap)
288 {
289 
290 	vap->va_type = VNON;
291 	/* XXX These next two used to be one line, but for a GCC bug. */
292 	vap->va_size = VNOVAL;
293 	vap->va_bytes = VNOVAL;
294 	vap->va_mode = vap->va_nlink = vap->va_uid = vap->va_gid =
295 		vap->va_fsid = vap->va_fileid =
296 		vap->va_blocksize = vap->va_rdev =
297 		vap->va_atime.tv_sec = vap->va_atime.tv_nsec =
298 		vap->va_mtime.tv_sec = vap->va_mtime.tv_nsec =
299 		vap->va_ctime.tv_sec = vap->va_ctime.tv_nsec =
300 		vap->va_flags = vap->va_gen = VNOVAL;
301 	vap->va_vaflags = 0;
302 }
303 
304 /*
305  * Routines having to do with the management of the vnode table.
306  */
307 extern int (**dead_vnodeop_p)(void *);
308 long numvnodes;
309 
310 /*
311  * Return the next vnode from the free list.
312  */
313 int
314 getnewvnode(enum vtagtype tag, struct mount *mp, int (**vops)(void *),
315     struct vnode **vpp)
316 {
317 	struct proc *p = curproc;
318 	struct freelst *listhd;
319 	static int toggle;
320 	struct vnode *vp;
321 	int s;
322 
323 	/*
324 	 * We must choose whether to allocate a new vnode or recycle an
325 	 * existing one. The criterion for allocating a new one is that
326 	 * the total number of vnodes is less than the number desired or
327 	 * there are no vnodes on either free list. Generally we only
328 	 * want to recycle vnodes that have no buffers associated with
329 	 * them, so we look first on the vnode_free_list. If it is empty,
330 	 * we next consider vnodes with referencing buffers on the
331 	 * vnode_hold_list. The toggle ensures that half the time we
332 	 * will use a buffer from the vnode_hold_list, and half the time
333 	 * we will allocate a new one unless the list has grown to twice
334 	 * the desired size. We are reticent to recycle vnodes from the
335 	 * vnode_hold_list because we will lose the identity of all its
336 	 * referencing buffers.
337 	 */
338 	toggle ^= 1;
339 	if (numvnodes > 2 * maxvnodes)
340 		toggle = 0;
341 
342 	s = splbio();
343 	if ((numvnodes < maxvnodes) ||
344 	    ((TAILQ_FIRST(listhd = &vnode_free_list) == NULL) &&
345 	    ((TAILQ_FIRST(listhd = &vnode_hold_list) == NULL) || toggle))) {
346 		splx(s);
347 		vp = pool_get(&vnode_pool, PR_WAITOK | PR_ZERO);
348 		numvnodes++;
349 	} else {
350 		for (vp = TAILQ_FIRST(listhd); vp != NULLVP;
351 		    vp = TAILQ_NEXT(vp, v_freelist)) {
352 			if (VOP_ISLOCKED(vp) == 0)
353 				break;
354 		}
355 		/*
356 		 * Unless this is a bad time of the month, at most
357 		 * the first NCPUS items on the free list are
358 		 * locked, so this is close enough to being empty.
359 		 */
360 		if (vp == NULL) {
361 			splx(s);
362 			tablefull("vnode");
363 			*vpp = 0;
364 			return (ENFILE);
365 		}
366 
367 #ifdef DIAGNOSTIC
368 		if (vp->v_usecount) {
369 			vprint("free vnode", vp);
370 			panic("free vnode isn't");
371 		}
372 #endif
373 
374 		TAILQ_REMOVE(listhd, vp, v_freelist);
375 		vp->v_bioflag &= ~VBIOONFREELIST;
376 		splx(s);
377 
378 		if (vp->v_type != VBAD)
379 			vgonel(vp, p);
380 #ifdef DIAGNOSTIC
381 		if (vp->v_data) {
382 			vprint("cleaned vnode", vp);
383 			panic("cleaned vnode isn't");
384 		}
385 		s = splbio();
386 		if (vp->v_numoutput)
387 			panic("Clean vnode has pending I/O's");
388 		splx(s);
389 #endif
390 		vp->v_flag = 0;
391 		vp->v_socket = 0;
392 	}
393 	vp->v_type = VNON;
394 	cache_purge(vp);
395 	vp->v_tag = tag;
396 	vp->v_op = vops;
397 	insmntque(vp, mp);
398 	*vpp = vp;
399 	vp->v_usecount = 1;
400 	vp->v_data = 0;
401 	simple_lock_init(&vp->v_uvm.u_obj.vmobjlock);
402 	return (0);
403 }
404 
405 /*
406  * Move a vnode from one mount queue to another.
407  */
408 void
409 insmntque(struct vnode *vp, struct mount *mp)
410 {
411 	/*
412 	 * Delete from old mount point vnode list, if on one.
413 	 */
414 	if (vp->v_mount != NULL)
415 		LIST_REMOVE(vp, v_mntvnodes);
416 	/*
417 	 * Insert into list of vnodes for the new mount point, if available.
418 	 */
419 	if ((vp->v_mount = mp) != NULL)
420 		LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes);
421 }
422 
423 /*
424  * Create a vnode for a block device.
425  * Used for root filesystem, argdev, and swap areas.
426  * Also used for memory file system special devices.
427  */
428 int
429 bdevvp(dev_t dev, struct vnode **vpp)
430 {
431 	return (getdevvp(dev, vpp, VBLK));
432 }
433 
434 /*
435  * Create a vnode for a character device.
436  * Used for console handling.
437  */
438 int
439 cdevvp(dev_t dev, struct vnode **vpp)
440 {
441 	return (getdevvp(dev, vpp, VCHR));
442 }
443 
444 /*
445  * Create a vnode for a device.
446  * Used by bdevvp (block device) for root file system etc.,
447  * and by cdevvp (character device) for console.
448  */
449 int
450 getdevvp(dev_t dev, struct vnode **vpp, enum vtype type)
451 {
452 	struct vnode *vp;
453 	struct vnode *nvp;
454 	int error;
455 
456 	if (dev == NODEV) {
457 		*vpp = NULLVP;
458 		return (0);
459 	}
460 	error = getnewvnode(VT_NON, NULL, spec_vnodeop_p, &nvp);
461 	if (error) {
462 		*vpp = NULLVP;
463 		return (error);
464 	}
465 	vp = nvp;
466 	vp->v_type = type;
467 	if ((nvp = checkalias(vp, dev, NULL)) != 0) {
468 		vput(vp);
469 		vp = nvp;
470 	}
471 	*vpp = vp;
472 	return (0);
473 }
474 
475 /*
476  * Check to see if the new vnode represents a special device
477  * for which we already have a vnode (either because of
478  * bdevvp() or because of a different vnode representing
479  * the same block device). If such an alias exists, deallocate
480  * the existing contents and return the aliased vnode. The
481  * caller is responsible for filling it with its new contents.
482  */
483 struct vnode *
484 checkalias(struct vnode *nvp, dev_t nvp_rdev, struct mount *mp)
485 {
486 	struct proc *p = curproc;
487 	struct vnode *vp;
488 	struct vnode **vpp;
489 
490 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
491 		return (NULLVP);
492 
493 	vpp = &speclisth[SPECHASH(nvp_rdev)];
494 loop:
495 	for (vp = *vpp; vp; vp = vp->v_specnext) {
496 		if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type) {
497 			continue;
498 		}
499 		/*
500 		 * Alias, but not in use, so flush it out.
501 		 */
502 		if (vp->v_usecount == 0) {
503 			vgonel(vp, p);
504 			goto loop;
505 		}
506 		if (vget(vp, LK_EXCLUSIVE, p)) {
507 			goto loop;
508 		}
509 		break;
510 	}
511 
512 	/*
513 	 * Common case is actually in the if statement
514 	 */
515 	if (vp == NULL || !(vp->v_tag == VT_NON && vp->v_type == VBLK)) {
516 		nvp->v_specinfo = malloc(sizeof(struct specinfo), M_VNODE,
517 			M_WAITOK);
518 		nvp->v_rdev = nvp_rdev;
519 		nvp->v_hashchain = vpp;
520 		nvp->v_specnext = *vpp;
521 		nvp->v_specmountpoint = NULL;
522 		nvp->v_speclockf = NULL;
523 		bzero(nvp->v_specbitmap, sizeof(nvp->v_specbitmap));
524 		*vpp = nvp;
525 		if (vp != NULLVP) {
526 			nvp->v_flag |= VALIASED;
527 			vp->v_flag |= VALIASED;
528 			vput(vp);
529 		}
530 		return (NULLVP);
531 	}
532 
533 	/*
534 	 * This code is the uncommon case. It is called in case
535 	 * we found an alias that was VT_NON && vtype of VBLK
536 	 * This means we found a block device that was created
537 	 * using bdevvp.
538 	 * An example of such a vnode is the root partition device vnode
539 	 * created in ffs_mountroot.
540 	 *
541 	 * The vnodes created by bdevvp should not be aliased (why?).
542 	 */
543 
544 	VOP_UNLOCK(vp, 0, p);
545 	vclean(vp, 0, p);
546 	vp->v_op = nvp->v_op;
547 	vp->v_tag = nvp->v_tag;
548 	nvp->v_type = VNON;
549 	insmntque(vp, mp);
550 	return (vp);
551 }
552 
553 /*
554  * Grab a particular vnode from the free list, increment its
555  * reference count and lock it. If the vnode lock bit is set,
556  * the vnode is being eliminated in vgone. In that case, we
557  * cannot grab it, so the process is awakened when the
558  * transition is completed, and an error code is returned to
559  * indicate that the vnode is no longer usable, possibly
560  * having been changed to a new file system type.
561  */
562 int
563 vget(struct vnode *vp, int flags, struct proc *p)
564 {
565 	int error, s, onfreelist;
566 
567 	/*
568 	 * If the vnode is in the process of being cleaned out for
569 	 * another use, we wait for the cleaning to finish and then
570 	 * return failure. Cleaning is determined by checking that
571 	 * the VXLOCK flag is set.
572 	 */
573 
574 	if (vp->v_flag & VXLOCK) {
575 		if (flags & LK_NOWAIT) {
576 			return (EBUSY);
577 		}
578 
579 		vp->v_flag |= VXWANT;
580 		tsleep(vp, PINOD, "vget", 0);
581 		return (ENOENT);
582 	}
583 
584 	onfreelist = vp->v_bioflag & VBIOONFREELIST;
585 	if (vp->v_usecount == 0 && onfreelist) {
586 		s = splbio();
587 		if (vp->v_holdcnt > 0)
588 			TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist);
589 		else
590 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
591 		vp->v_bioflag &= ~VBIOONFREELIST;
592 		splx(s);
593 	}
594 
595  	vp->v_usecount++;
596 	if (flags & LK_TYPE_MASK) {
597 		if ((error = vn_lock(vp, flags, p)) != 0) {
598 			vp->v_usecount--;
599 			if (vp->v_usecount == 0 && onfreelist)
600 				vputonfreelist(vp);
601 		}
602 		return (error);
603 	}
604 
605 	return (0);
606 }
607 
608 
609 /* Vnode reference. */
610 void
611 vref(struct vnode *vp)
612 {
613 #ifdef DIAGNOSTIC
614 	if (vp->v_usecount == 0)
615 		panic("vref used where vget required");
616 #endif
617 	vp->v_usecount++;
618 }
619 
620 void
621 vputonfreelist(struct vnode *vp)
622 {
623 	int s;
624 	struct freelst *lst;
625 
626 	s = splbio();
627 #ifdef DIAGNOSTIC
628 	if (vp->v_usecount != 0)
629 		panic("Use count is not zero!");
630 
631 	if (vp->v_bioflag & VBIOONFREELIST) {
632 		vprint("vnode already on free list: ", vp);
633 		panic("vnode already on free list");
634 	}
635 #endif
636 
637 	vp->v_bioflag |= VBIOONFREELIST;
638 
639 	if (vp->v_holdcnt > 0)
640 		lst = &vnode_hold_list;
641 	else
642 		lst = &vnode_free_list;
643 
644 	if (vp->v_type == VBAD)
645 		TAILQ_INSERT_HEAD(lst, vp, v_freelist);
646 	else
647 		TAILQ_INSERT_TAIL(lst, vp, v_freelist);
648 
649 	splx(s);
650 }
651 
652 /*
653  * vput(), just unlock and vrele()
654  */
655 void
656 vput(struct vnode *vp)
657 {
658 	struct proc *p = curproc;
659 
660 #ifdef DIAGNOSTIC
661 	if (vp == NULL)
662 		panic("vput: null vp");
663 #endif
664 
665 #ifdef DIAGNOSTIC
666 	if (vp->v_usecount == 0) {
667 		vprint("vput: bad ref count", vp);
668 		panic("vput: ref cnt");
669 	}
670 #endif
671 	vp->v_usecount--;
672 	if (vp->v_usecount > 0) {
673 		VOP_UNLOCK(vp, 0, p);
674 		return;
675 	}
676 
677 #ifdef DIAGNOSTIC
678 	if (vp->v_writecount != 0) {
679 		vprint("vput: bad writecount", vp);
680 		panic("vput: v_writecount != 0");
681 	}
682 #endif
683 
684 	VOP_INACTIVE(vp, p);
685 
686 	if (vp->v_usecount == 0 && !(vp->v_bioflag & VBIOONFREELIST))
687 		vputonfreelist(vp);
688 }
689 
690 /*
691  * Vnode release - use for active VNODES.
692  * If count drops to zero, call inactive routine and return to freelist.
693  * Returns 0 if it did not sleep.
694  */
695 int
696 vrele(struct vnode *vp)
697 {
698 	struct proc *p = curproc;
699 
700 #ifdef DIAGNOSTIC
701 	if (vp == NULL)
702 		panic("vrele: null vp");
703 #endif
704 #ifdef DIAGNOSTIC
705 	if (vp->v_usecount == 0) {
706 		vprint("vrele: bad ref count", vp);
707 		panic("vrele: ref cnt");
708 	}
709 #endif
710 	vp->v_usecount--;
711 	if (vp->v_usecount > 0) {
712 		return (0);
713 	}
714 
715 #ifdef DIAGNOSTIC
716 	if (vp->v_writecount != 0) {
717 		vprint("vrele: bad writecount", vp);
718 		panic("vrele: v_writecount != 0");
719 	}
720 #endif
721 
722 	if (vn_lock(vp, LK_EXCLUSIVE, p)) {
723 #ifdef DIAGNOSTIC
724 		vprint("vrele: cannot lock", vp);
725 #endif
726 		return (1);
727 	}
728 
729 	VOP_INACTIVE(vp, p);
730 
731 	if (vp->v_usecount == 0 && !(vp->v_bioflag & VBIOONFREELIST))
732 		vputonfreelist(vp);
733 	return (1);
734 }
735 
736 /* Page or buffer structure gets a reference. */
737 void
738 vhold(struct vnode *vp)
739 {
740 	/*
741 	 * If it is on the freelist and the hold count is currently
742 	 * zero, move it to the hold list.
743 	 */
744 	if ((vp->v_bioflag & VBIOONFREELIST) &&
745 	    vp->v_holdcnt == 0 && vp->v_usecount == 0) {
746 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
747 		TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist);
748 	}
749 	vp->v_holdcnt++;
750 }
751 
752 /* Lose interest in a vnode. */
753 void
754 vdrop(struct vnode *vp)
755 {
756 #ifdef DIAGNOSTIC
757 	if (vp->v_holdcnt == 0)
758 		panic("vdrop: zero holdcnt");
759 #endif
760 
761 	vp->v_holdcnt--;
762 
763 	/*
764 	 * If it is on the holdlist and the hold count drops to
765 	 * zero, move it to the free list.
766 	 */
767 	if ((vp->v_bioflag & VBIOONFREELIST) &&
768 	    vp->v_holdcnt == 0 && vp->v_usecount == 0) {
769 		TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist);
770 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
771 	}
772 }
773 
774 /*
775  * Remove any vnodes in the vnode table belonging to mount point mp.
776  *
777  * If MNT_NOFORCE is specified, there should not be any active ones,
778  * return error if any are found (nb: this is a user error, not a
779  * system error). If MNT_FORCE is specified, detach any active vnodes
780  * that are found.
781  */
782 #ifdef DEBUG
783 int busyprt = 0;	/* print out busy vnodes */
784 struct ctldebug debug1 = { "busyprt", &busyprt };
785 #endif
786 
787 int
788 vfs_mount_foreach_vnode(struct mount *mp,
789     int (*func)(struct vnode *, void *), void *arg) {
790 	struct vnode *vp, *nvp;
791 	int error = 0;
792 
793 loop:
794 	for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp != NULL; vp = nvp) {
795 		if (vp->v_mount != mp)
796 			goto loop;
797 		nvp = LIST_NEXT(vp, v_mntvnodes);
798 
799 		error = func(vp, arg);
800 
801 		if (error != 0)
802 			break;
803 	}
804 
805 	return (error);
806 }
807 
808 struct vflush_args {
809 	struct vnode *skipvp;
810 	int busy;
811 	int flags;
812 };
813 
814 int
815 vflush_vnode(struct vnode *vp, void *arg) {
816 	struct vflush_args *va = arg;
817 	struct proc *p = curproc;
818 
819 	if (vp == va->skipvp) {
820 		return (0);
821 	}
822 
823 	if ((va->flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
824 		return (0);
825 	}
826 
827 	/*
828 	 * If WRITECLOSE is set, only flush out regular file
829 	 * vnodes open for writing.
830 	 */
831 	if ((va->flags & WRITECLOSE) &&
832 	    (vp->v_writecount == 0 || vp->v_type != VREG)) {
833 		return (0);
834 	}
835 
836 	/*
837 	 * With v_usecount == 0, all we need to do is clear
838 	 * out the vnode data structures and we are done.
839 	 */
840 	if (vp->v_usecount == 0) {
841 		vgonel(vp, p);
842 		return (0);
843 	}
844 
845 	/*
846 	 * If FORCECLOSE is set, forcibly close the vnode.
847 	 * For block or character devices, revert to an
848 	 * anonymous device. For all other files, just kill them.
849 	 */
850 	if (va->flags & FORCECLOSE) {
851 		if (vp->v_type != VBLK && vp->v_type != VCHR) {
852 			vgonel(vp, p);
853 		} else {
854 			vclean(vp, 0, p);
855 			vp->v_op = spec_vnodeop_p;
856 			insmntque(vp, (struct mount *)0);
857 		}
858 		return (0);
859 	}
860 
861 #ifdef DEBUG
862 	if (busyprt)
863 		vprint("vflush: busy vnode", vp);
864 #endif
865 	va->busy++;
866 	return (0);
867 }
868 
869 int
870 vflush(struct mount *mp, struct vnode *skipvp, int flags)
871 {
872 	struct vflush_args va;
873 	va.skipvp = skipvp;
874 	va.busy = 0;
875 	va.flags = flags;
876 
877 	vfs_mount_foreach_vnode(mp, vflush_vnode, &va);
878 
879 	if (va.busy)
880 		return (EBUSY);
881 	return (0);
882 }
883 
884 /*
885  * Disassociate the underlying file system from a vnode.
886  */
887 void
888 vclean(struct vnode *vp, int flags, struct proc *p)
889 {
890 	int active;
891 
892 	/*
893 	 * Check to see if the vnode is in use.
894 	 * If so we have to reference it before we clean it out
895 	 * so that its count cannot fall to zero and generate a
896 	 * race against ourselves to recycle it.
897 	 */
898 	if ((active = vp->v_usecount) != 0)
899 		vp->v_usecount++;
900 
901 	/*
902 	 * Prevent the vnode from being recycled or
903 	 * brought into use while we clean it out.
904 	 */
905 	if (vp->v_flag & VXLOCK)
906 		panic("vclean: deadlock");
907 	vp->v_flag |= VXLOCK;
908 	/*
909 	 * Even if the count is zero, the VOP_INACTIVE routine may still
910 	 * have the object locked while it cleans it out. The VOP_LOCK
911 	 * ensures that the VOP_INACTIVE routine is done with its work.
912 	 * For active vnodes, it ensures that no other activity can
913 	 * occur while the underlying object is being cleaned out.
914 	 */
915 	VOP_LOCK(vp, LK_DRAIN, p);
916 
917 	/*
918 	 * Clean out any VM data associated with the vnode.
919 	 */
920 	uvm_vnp_terminate(vp);
921 	/*
922 	 * Clean out any buffers associated with the vnode.
923 	 */
924 	if (flags & DOCLOSE)
925 		vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0);
926 	/*
927 	 * If purging an active vnode, it must be closed and
928 	 * deactivated before being reclaimed. Note that the
929 	 * VOP_INACTIVE will unlock the vnode
930 	 */
931 	if (active) {
932 		if (flags & DOCLOSE)
933 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, p);
934 		VOP_INACTIVE(vp, p);
935 	} else {
936 		/*
937 		 * Any other processes trying to obtain this lock must first
938 		 * wait for VXLOCK to clear, then call the new lock operation.
939 		 */
940 		VOP_UNLOCK(vp, 0, p);
941 	}
942 
943 	/*
944 	 * Reclaim the vnode.
945 	 */
946 	if (VOP_RECLAIM(vp, p))
947 		panic("vclean: cannot reclaim");
948 	if (active) {
949 		vp->v_usecount--;
950 		if (vp->v_usecount == 0) {
951 			if (vp->v_holdcnt > 0)
952 				panic("vclean: not clean");
953 			vputonfreelist(vp);
954 		}
955 	}
956 	cache_purge(vp);
957 
958 	/*
959 	 * Done with purge, notify sleepers of the grim news.
960 	 */
961 	vp->v_op = dead_vnodeop_p;
962 	VN_KNOTE(vp, NOTE_REVOKE);
963 	vp->v_tag = VT_NON;
964 	vp->v_flag &= ~VXLOCK;
965 #ifdef VFSDEBUG
966 	vp->v_flag &= ~VLOCKSWORK;
967 #endif
968 	if (vp->v_flag & VXWANT) {
969 		vp->v_flag &= ~VXWANT;
970 		wakeup(vp);
971 	}
972 }
973 
974 /*
975  * Recycle an unused vnode to the front of the free list.
976  */
977 int
978 vrecycle(struct vnode *vp, struct proc *p)
979 {
980 	if (vp->v_usecount == 0) {
981 		vgonel(vp, p);
982 		return (1);
983 	}
984 	return (0);
985 }
986 
987 /*
988  * Eliminate all activity associated with a vnode
989  * in preparation for reuse.
990  */
991 void
992 vgone(struct vnode *vp)
993 {
994 	struct proc *p = curproc;
995 	vgonel(vp, p);
996 }
997 
998 /*
999  * vgone, with struct proc.
1000  */
1001 void
1002 vgonel(struct vnode *vp, struct proc *p)
1003 {
1004 	struct vnode *vq;
1005 	struct vnode *vx;
1006 
1007 	/*
1008 	 * If a vgone (or vclean) is already in progress,
1009 	 * wait until it is done and return.
1010 	 */
1011 	if (vp->v_flag & VXLOCK) {
1012 		vp->v_flag |= VXWANT;
1013 		tsleep(vp, PINOD, "vgone", 0);
1014 		return;
1015 	}
1016 
1017 	/*
1018 	 * Clean out the filesystem specific data.
1019 	 */
1020 	vclean(vp, DOCLOSE, p);
1021 	/*
1022 	 * Delete from old mount point vnode list, if on one.
1023 	 */
1024 	if (vp->v_mount != NULL)
1025 		insmntque(vp, (struct mount *)0);
1026 	/*
1027 	 * If special device, remove it from special device alias list
1028 	 * if it is on one.
1029 	 */
1030 	if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_specinfo != 0) {
1031 		if (*vp->v_hashchain == vp) {
1032 			*vp->v_hashchain = vp->v_specnext;
1033 		} else {
1034 			for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1035 				if (vq->v_specnext != vp)
1036 					continue;
1037 				vq->v_specnext = vp->v_specnext;
1038 				break;
1039 			}
1040 			if (vq == NULL)
1041 				panic("missing bdev");
1042 		}
1043 		if (vp->v_flag & VALIASED) {
1044 			vx = NULL;
1045 			for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1046 				if (vq->v_rdev != vp->v_rdev ||
1047 				    vq->v_type != vp->v_type)
1048 					continue;
1049 				if (vx)
1050 					break;
1051 				vx = vq;
1052 			}
1053 			if (vx == NULL)
1054 				panic("missing alias");
1055 			if (vq == NULL)
1056 				vx->v_flag &= ~VALIASED;
1057 			vp->v_flag &= ~VALIASED;
1058 		}
1059 		free(vp->v_specinfo, M_VNODE);
1060 		vp->v_specinfo = NULL;
1061 	}
1062 	/*
1063 	 * If it is on the freelist and not already at the head,
1064 	 * move it to the head of the list.
1065 	 */
1066 	vp->v_type = VBAD;
1067 
1068 	/*
1069 	 * Move onto the free list, unless we were called from
1070 	 * getnewvnode and we're not on any free list
1071 	 */
1072 	if (vp->v_usecount == 0 &&
1073 	    (vp->v_bioflag & VBIOONFREELIST)) {
1074 		int s;
1075 
1076 		s = splbio();
1077 
1078 		if (vp->v_holdcnt > 0)
1079 			panic("vgonel: not clean");
1080 
1081 		if (TAILQ_FIRST(&vnode_free_list) != vp) {
1082 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1083 			TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
1084 		}
1085 		splx(s);
1086 	}
1087 }
1088 
1089 /*
1090  * Lookup a vnode by device number.
1091  */
1092 int
1093 vfinddev(dev_t dev, enum vtype type, struct vnode **vpp)
1094 {
1095 	struct vnode *vp;
1096 	int rc =0;
1097 
1098 	for (vp = speclisth[SPECHASH(dev)]; vp; vp = vp->v_specnext) {
1099 		if (dev != vp->v_rdev || type != vp->v_type)
1100 			continue;
1101 		*vpp = vp;
1102 		rc = 1;
1103 		break;
1104 	}
1105 	return (rc);
1106 }
1107 
1108 /*
1109  * Revoke all the vnodes corresponding to the specified minor number
1110  * range (endpoints inclusive) of the specified major.
1111  */
1112 void
1113 vdevgone(int maj, int minl, int minh, enum vtype type)
1114 {
1115 	struct vnode *vp;
1116 	int mn;
1117 
1118 	for (mn = minl; mn <= minh; mn++)
1119 		if (vfinddev(makedev(maj, mn), type, &vp))
1120 			VOP_REVOKE(vp, REVOKEALL);
1121 }
1122 
1123 /*
1124  * Calculate the total number of references to a special device.
1125  */
1126 int
1127 vcount(struct vnode *vp)
1128 {
1129 	struct vnode *vq, *vnext;
1130 	int count;
1131 
1132 loop:
1133 	if ((vp->v_flag & VALIASED) == 0)
1134 		return (vp->v_usecount);
1135 	for (count = 0, vq = *vp->v_hashchain; vq; vq = vnext) {
1136 		vnext = vq->v_specnext;
1137 		if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type)
1138 			continue;
1139 		/*
1140 		 * Alias, but not in use, so flush it out.
1141 		 */
1142 		if (vq->v_usecount == 0 && vq != vp) {
1143 			vgone(vq);
1144 			goto loop;
1145 		}
1146 		count += vq->v_usecount;
1147 	}
1148 	return (count);
1149 }
1150 
1151 #if defined(DEBUG) || defined(DIAGNOSTIC)
1152 /*
1153  * Print out a description of a vnode.
1154  */
1155 static char *typename[] =
1156    { "VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD" };
1157 
1158 void
1159 vprint(char *label, struct vnode *vp)
1160 {
1161 	char buf[64];
1162 
1163 	if (label != NULL)
1164 		printf("%s: ", label);
1165 	printf("%p, type %s, use %u, write %u, hold %u,",
1166 		vp, typename[vp->v_type], vp->v_usecount, vp->v_writecount,
1167 		vp->v_holdcnt);
1168 	buf[0] = '\0';
1169 	if (vp->v_flag & VROOT)
1170 		strlcat(buf, "|VROOT", sizeof buf);
1171 	if (vp->v_flag & VTEXT)
1172 		strlcat(buf, "|VTEXT", sizeof buf);
1173 	if (vp->v_flag & VSYSTEM)
1174 		strlcat(buf, "|VSYSTEM", sizeof buf);
1175 	if (vp->v_flag & VXLOCK)
1176 		strlcat(buf, "|VXLOCK", sizeof buf);
1177 	if (vp->v_flag & VXWANT)
1178 		strlcat(buf, "|VXWANT", sizeof buf);
1179 	if (vp->v_bioflag & VBIOWAIT)
1180 		strlcat(buf, "|VBIOWAIT", sizeof buf);
1181 	if (vp->v_bioflag & VBIOONFREELIST)
1182 		strlcat(buf, "|VBIOONFREELIST", sizeof buf);
1183 	if (vp->v_bioflag & VBIOONSYNCLIST)
1184 		strlcat(buf, "|VBIOONSYNCLIST", sizeof buf);
1185 	if (vp->v_flag & VALIASED)
1186 		strlcat(buf, "|VALIASED", sizeof buf);
1187 	if (buf[0] != '\0')
1188 		printf(" flags (%s)", &buf[1]);
1189 	if (vp->v_data == NULL) {
1190 		printf("\n");
1191 	} else {
1192 		printf("\n\t");
1193 		VOP_PRINT(vp);
1194 	}
1195 }
1196 #endif /* DEBUG || DIAGNOSTIC */
1197 
1198 #ifdef DEBUG
1199 /*
1200  * List all of the locked vnodes in the system.
1201  * Called when debugging the kernel.
1202  */
1203 void
1204 printlockedvnodes(void)
1205 {
1206 	struct mount *mp, *nmp;
1207 	struct vnode *vp;
1208 
1209 	printf("Locked vnodes\n");
1210 
1211 	for (mp = CIRCLEQ_FIRST(&mountlist); mp != CIRCLEQ_END(&mountlist);
1212 	    mp = nmp) {
1213 		if (vfs_busy(mp, VB_READ|VB_NOWAIT)) {
1214 			nmp = CIRCLEQ_NEXT(mp, mnt_list);
1215 			continue;
1216 		}
1217 		LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
1218 			if (VOP_ISLOCKED(vp))
1219 				vprint((char *)0, vp);
1220 		}
1221 		nmp = CIRCLEQ_NEXT(mp, mnt_list);
1222 		vfs_unbusy(mp);
1223  	}
1224 
1225 }
1226 #endif
1227 
1228 /*
1229  * Top level filesystem related information gathering.
1230  */
1231 int
1232 vfs_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp,
1233     size_t newlen, struct proc *p)
1234 {
1235 	struct vfsconf *vfsp, *tmpvfsp;
1236 	int ret;
1237 
1238 	/* all sysctl names at this level are at least name and field */
1239 	if (namelen < 2)
1240 		return (ENOTDIR);		/* overloaded */
1241 
1242 	if (name[0] != VFS_GENERIC) {
1243 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
1244 			if (vfsp->vfc_typenum == name[0])
1245 				break;
1246 
1247 		if (vfsp == NULL)
1248 			return (EOPNOTSUPP);
1249 
1250 		return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
1251 		    oldp, oldlenp, newp, newlen, p));
1252 	}
1253 
1254 	switch (name[1]) {
1255 	case VFS_MAXTYPENUM:
1256 		return (sysctl_rdint(oldp, oldlenp, newp, maxvfsconf));
1257 
1258 	case VFS_CONF:
1259 		if (namelen < 3)
1260 			return (ENOTDIR);	/* overloaded */
1261 
1262 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
1263 			if (vfsp->vfc_typenum == name[2])
1264 				break;
1265 
1266 		if (vfsp == NULL)
1267 			return (EOPNOTSUPP);
1268 
1269 		/* Make a copy, clear out kernel pointers */
1270 		tmpvfsp = malloc(sizeof(*tmpvfsp), M_TEMP, M_WAITOK);
1271 		bcopy(vfsp, tmpvfsp, sizeof(*tmpvfsp));
1272 		tmpvfsp->vfc_vfsops = NULL;
1273 		tmpvfsp->vfc_next = NULL;
1274 
1275 		ret = sysctl_rdstruct(oldp, oldlenp, newp, tmpvfsp,
1276 		    sizeof(struct vfsconf));
1277 
1278 		free(tmpvfsp, M_TEMP);
1279 		return (ret);
1280 	case VFS_BCACHESTAT:	/* buffer cache statistics */
1281 		ret = sysctl_rdstruct(oldp, oldlenp, newp, &bcstats,
1282 		    sizeof(struct bcachestats));
1283 		return(ret);
1284 	}
1285 	return (EOPNOTSUPP);
1286 }
1287 
1288 int kinfo_vdebug = 1;
1289 #define KINFO_VNODESLOP	10
1290 /*
1291  * Dump vnode list (via sysctl).
1292  * Copyout address of vnode followed by vnode.
1293  */
1294 /* ARGSUSED */
1295 int
1296 sysctl_vnode(char *where, size_t *sizep, struct proc *p)
1297 {
1298 	struct mount *mp, *nmp;
1299 	struct vnode *vp, *nvp;
1300 	char *bp = where, *savebp;
1301 	char *ewhere;
1302 	int error;
1303 
1304 	if (where == NULL) {
1305 		*sizep = (numvnodes + KINFO_VNODESLOP) * sizeof(struct e_vnode);
1306 		return (0);
1307 	}
1308 	ewhere = where + *sizep;
1309 
1310 	for (mp = CIRCLEQ_FIRST(&mountlist); mp != CIRCLEQ_END(&mountlist);
1311 	    mp = nmp) {
1312 		if (vfs_busy(mp, VB_READ|VB_NOWAIT)) {
1313 			nmp = CIRCLEQ_NEXT(mp, mnt_list);
1314 			continue;
1315 		}
1316 		savebp = bp;
1317 again:
1318 		for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp != NULL;
1319 		    vp = nvp) {
1320 			/*
1321 			 * Check that the vp is still associated with
1322 			 * this filesystem.  RACE: could have been
1323 			 * recycled onto the same filesystem.
1324 			 */
1325 			if (vp->v_mount != mp) {
1326 				if (kinfo_vdebug)
1327 					printf("kinfo: vp changed\n");
1328 				bp = savebp;
1329 				goto again;
1330 			}
1331 			nvp = LIST_NEXT(vp, v_mntvnodes);
1332 			if (bp + sizeof(struct e_vnode) > ewhere) {
1333 				*sizep = bp - where;
1334 				vfs_unbusy(mp);
1335 				return (ENOMEM);
1336 			}
1337 			if ((error = copyout(&vp,
1338 			    &((struct e_vnode *)bp)->vptr,
1339 			    sizeof(struct vnode *))) ||
1340 			   (error = copyout(vp,
1341 			    &((struct e_vnode *)bp)->vnode,
1342 			    sizeof(struct vnode)))) {
1343 				vfs_unbusy(mp);
1344 				return (error);
1345 			}
1346 			bp += sizeof(struct e_vnode);
1347 		}
1348 
1349 		nmp = CIRCLEQ_NEXT(mp, mnt_list);
1350 		vfs_unbusy(mp);
1351 	}
1352 
1353 	*sizep = bp - where;
1354 
1355 	return (0);
1356 }
1357 
1358 /*
1359  * Check to see if a filesystem is mounted on a block device.
1360  */
1361 int
1362 vfs_mountedon(struct vnode *vp)
1363 {
1364 	struct vnode *vq;
1365 	int error = 0;
1366 
1367  	if (vp->v_specmountpoint != NULL)
1368 		return (EBUSY);
1369 	if (vp->v_flag & VALIASED) {
1370 		for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1371 			if (vq->v_rdev != vp->v_rdev ||
1372 			    vq->v_type != vp->v_type)
1373 				continue;
1374 			if (vq->v_specmountpoint != NULL) {
1375 				error = EBUSY;
1376 				break;
1377 			}
1378  		}
1379 	}
1380 	return (error);
1381 }
1382 
1383 /*
1384  * Build hash lists of net addresses and hang them off the mount point.
1385  * Called by ufs_mount() to set up the lists of export addresses.
1386  */
1387 int
1388 vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
1389     struct export_args *argp)
1390 {
1391 	struct netcred *np;
1392 	struct radix_node_head *rnh;
1393 	int i;
1394 	struct radix_node *rn;
1395 	struct sockaddr *saddr, *smask = 0;
1396 	struct domain *dom;
1397 	int error;
1398 
1399 	if (argp->ex_addrlen == 0) {
1400 		if (mp->mnt_flag & MNT_DEFEXPORTED)
1401 			return (EPERM);
1402 		np = &nep->ne_defexported;
1403 		np->netc_exflags = argp->ex_flags;
1404 		np->netc_anon = argp->ex_anon;
1405 		np->netc_anon.cr_ref = 1;
1406 		mp->mnt_flag |= MNT_DEFEXPORTED;
1407 		return (0);
1408 	}
1409 	if (argp->ex_addrlen > MLEN || argp->ex_masklen > MLEN ||
1410 	    argp->ex_addrlen < 0 || argp->ex_masklen < 0)
1411 		return (EINVAL);
1412 	i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
1413 	np = (struct netcred *)malloc(i, M_NETADDR, M_WAITOK|M_ZERO);
1414 	saddr = (struct sockaddr *)(np + 1);
1415 	error = copyin(argp->ex_addr, saddr, argp->ex_addrlen);
1416 	if (error)
1417 		goto out;
1418 	if (saddr->sa_len > argp->ex_addrlen)
1419 		saddr->sa_len = argp->ex_addrlen;
1420 	if (argp->ex_masklen) {
1421 		smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
1422 		error = copyin(argp->ex_mask, smask, argp->ex_masklen);
1423 		if (error)
1424 			goto out;
1425 		if (smask->sa_len > argp->ex_masklen)
1426 			smask->sa_len = argp->ex_masklen;
1427 	}
1428 	i = saddr->sa_family;
1429 	if (i < 0 || i > AF_MAX) {
1430 		error = EINVAL;
1431 		goto out;
1432 	}
1433 	if ((rnh = nep->ne_rtable[i]) == 0) {
1434 		/*
1435 		 * Seems silly to initialize every AF when most are not
1436 		 * used, do so on demand here
1437 		 */
1438 		for (dom = domains; dom; dom = dom->dom_next)
1439 			if (dom->dom_family == i && dom->dom_rtattach) {
1440 				dom->dom_rtattach((void **)&nep->ne_rtable[i],
1441 					dom->dom_rtoffset);
1442 				break;
1443 			}
1444 		if ((rnh = nep->ne_rtable[i]) == 0) {
1445 			error = ENOBUFS;
1446 			goto out;
1447 		}
1448 	}
1449 	rn = (*rnh->rnh_addaddr)((caddr_t)saddr, (caddr_t)smask, rnh,
1450 		np->netc_rnodes, 0);
1451 	if (rn == 0 || np != (struct netcred *)rn) { /* already exists */
1452 		error = EPERM;
1453 		goto out;
1454 	}
1455 	np->netc_exflags = argp->ex_flags;
1456 	np->netc_anon = argp->ex_anon;
1457 	np->netc_anon.cr_ref = 1;
1458 	return (0);
1459 out:
1460 	free(np, M_NETADDR);
1461 	return (error);
1462 }
1463 
1464 /* ARGSUSED */
1465 int
1466 vfs_free_netcred(struct radix_node *rn, void *w)
1467 {
1468 	struct radix_node_head *rnh = (struct radix_node_head *)w;
1469 
1470 	(*rnh->rnh_deladdr)(rn->rn_key, rn->rn_mask, rnh, NULL);
1471 	free(rn, M_NETADDR);
1472 	return (0);
1473 }
1474 
1475 /*
1476  * Free the net address hash lists that are hanging off the mount points.
1477  */
1478 void
1479 vfs_free_addrlist(struct netexport *nep)
1480 {
1481 	int i;
1482 	struct radix_node_head *rnh;
1483 
1484 	for (i = 0; i <= AF_MAX; i++)
1485 		if ((rnh = nep->ne_rtable[i]) != NULL) {
1486 			(*rnh->rnh_walktree)(rnh, vfs_free_netcred, rnh);
1487 			free(rnh, M_RTABLE);
1488 			nep->ne_rtable[i] = 0;
1489 		}
1490 }
1491 
1492 int
1493 vfs_export(struct mount *mp, struct netexport *nep, struct export_args *argp)
1494 {
1495 	int error;
1496 
1497 	if (argp->ex_flags & MNT_DELEXPORT) {
1498 		vfs_free_addrlist(nep);
1499 		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
1500 	}
1501 	if (argp->ex_flags & MNT_EXPORTED) {
1502 		if ((error = vfs_hang_addrlist(mp, nep, argp)) != 0)
1503 			return (error);
1504 		mp->mnt_flag |= MNT_EXPORTED;
1505 	}
1506 	return (0);
1507 }
1508 
1509 struct netcred *
1510 vfs_export_lookup(struct mount *mp, struct netexport *nep, struct mbuf *nam)
1511 {
1512 	struct netcred *np;
1513 	struct radix_node_head *rnh;
1514 	struct sockaddr *saddr;
1515 
1516 	np = NULL;
1517 	if (mp->mnt_flag & MNT_EXPORTED) {
1518 		/*
1519 		 * Lookup in the export list first.
1520 		 */
1521 		if (nam != NULL) {
1522 			saddr = mtod(nam, struct sockaddr *);
1523 			rnh = nep->ne_rtable[saddr->sa_family];
1524 			if (rnh != NULL) {
1525 				np = (struct netcred *)
1526 					(*rnh->rnh_matchaddr)((caddr_t)saddr,
1527 					    rnh);
1528 				if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
1529 					np = NULL;
1530 			}
1531 		}
1532 		/*
1533 		 * If no address match, use the default if it exists.
1534 		 */
1535 		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
1536 			np = &nep->ne_defexported;
1537 	}
1538 	return (np);
1539 }
1540 
1541 /*
1542  * Do the usual access checking.
1543  * file_mode, uid and gid are from the vnode in question,
1544  * while acc_mode and cred are from the VOP_ACCESS parameter list
1545  */
1546 int
1547 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid,
1548     mode_t acc_mode, struct ucred *cred)
1549 {
1550 	mode_t mask;
1551 
1552 	/* User id 0 always gets read/write access. */
1553 	if (cred->cr_uid == 0) {
1554 		/* For VEXEC, at least one of the execute bits must be set. */
1555 		if ((acc_mode & VEXEC) && type != VDIR &&
1556 		    (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)
1557 			return EACCES;
1558 		return 0;
1559 	}
1560 
1561 	mask = 0;
1562 
1563 	/* Otherwise, check the owner. */
1564 	if (cred->cr_uid == uid) {
1565 		if (acc_mode & VEXEC)
1566 			mask |= S_IXUSR;
1567 		if (acc_mode & VREAD)
1568 			mask |= S_IRUSR;
1569 		if (acc_mode & VWRITE)
1570 			mask |= S_IWUSR;
1571 		return (file_mode & mask) == mask ? 0 : EACCES;
1572 	}
1573 
1574 	/* Otherwise, check the groups. */
1575 	if (cred->cr_gid == gid || groupmember(gid, cred)) {
1576 		if (acc_mode & VEXEC)
1577 			mask |= S_IXGRP;
1578 		if (acc_mode & VREAD)
1579 			mask |= S_IRGRP;
1580 		if (acc_mode & VWRITE)
1581 			mask |= S_IWGRP;
1582 		return (file_mode & mask) == mask ? 0 : EACCES;
1583 	}
1584 
1585 	/* Otherwise, check everyone else. */
1586 	if (acc_mode & VEXEC)
1587 		mask |= S_IXOTH;
1588 	if (acc_mode & VREAD)
1589 		mask |= S_IROTH;
1590 	if (acc_mode & VWRITE)
1591 		mask |= S_IWOTH;
1592 	return (file_mode & mask) == mask ? 0 : EACCES;
1593 }
1594 
1595 /*
1596  * Unmount all file systems.
1597  * We traverse the list in reverse order under the assumption that doing so
1598  * will avoid needing to worry about dependencies.
1599  */
1600 void
1601 vfs_unmountall(void)
1602 {
1603 	struct mount *mp, *nmp;
1604 	int allerror, error, again = 1;
1605 
1606  retry:
1607 	allerror = 0;
1608 	for (mp = CIRCLEQ_LAST(&mountlist); mp != CIRCLEQ_END(&mountlist);
1609 	    mp = nmp) {
1610 		nmp = CIRCLEQ_PREV(mp, mnt_list);
1611 		if ((vfs_busy(mp, VB_WRITE|VB_NOWAIT)) != 0)
1612 			continue;
1613 		if ((error = dounmount(mp, MNT_FORCE, curproc, NULL)) != 0) {
1614 			printf("unmount of %s failed with error %d\n",
1615 			    mp->mnt_stat.f_mntonname, error);
1616 			allerror = 1;
1617 		}
1618 	}
1619 
1620 	if (allerror) {
1621 		printf("WARNING: some file systems would not unmount\n");
1622 		if (again) {
1623 			printf("retrying\n");
1624 			again = 0;
1625 			goto retry;
1626 		}
1627 	}
1628 }
1629 
1630 /*
1631  * Sync and unmount file systems before shutting down.
1632  */
1633 void
1634 vfs_shutdown(void)
1635 {
1636 #ifdef ACCOUNTING
1637 	extern void acct_shutdown(void);
1638 
1639 	acct_shutdown();
1640 #endif
1641 
1642 	/* XXX Should suspend scheduling. */
1643 	(void) spl0();
1644 
1645 	printf("syncing disks... ");
1646 
1647 	if (panicstr == 0) {
1648 		/* Sync before unmount, in case we hang on something. */
1649 		sys_sync(&proc0, (void *)0, (register_t *)0);
1650 
1651 		/* Unmount file systems. */
1652 		vfs_unmountall();
1653 	}
1654 
1655 	if (vfs_syncwait(1))
1656 		printf("giving up\n");
1657 	else
1658 		printf("done\n");
1659 }
1660 
1661 /*
1662  * perform sync() operation and wait for buffers to flush.
1663  * assumptions: called w/ scheduler disabled and physical io enabled
1664  * for now called at spl0() XXX
1665  */
1666 int
1667 vfs_syncwait(int verbose)
1668 {
1669 	struct buf *bp;
1670 	int iter, nbusy, dcount, s;
1671 	struct proc *p;
1672 
1673 	p = curproc? curproc : &proc0;
1674 	sys_sync(p, (void *)0, (register_t *)0);
1675 
1676 	/* Wait for sync to finish. */
1677 	dcount = 10000;
1678 	for (iter = 0; iter < 20; iter++) {
1679 		nbusy = 0;
1680 		LIST_FOREACH(bp, &bufhead, b_list) {
1681 			if ((bp->b_flags & (B_BUSY|B_INVAL|B_READ)) == B_BUSY)
1682 				nbusy++;
1683 			/*
1684 			 * With soft updates, some buffers that are
1685 			 * written will be remarked as dirty until other
1686 			 * buffers are written.
1687 			 */
1688 			if (bp->b_flags & B_DELWRI) {
1689 				s = splbio();
1690 				bremfree(bp);
1691 				buf_acquire(bp);
1692 				splx(s);
1693 				nbusy++;
1694 				bawrite(bp);
1695 				if (dcount-- <= 0) {
1696 					if (verbose)
1697 						printf("softdep ");
1698 					return 1;
1699 				}
1700 			}
1701 		}
1702 		if (nbusy == 0)
1703 			break;
1704 		if (verbose)
1705 			printf("%d ", nbusy);
1706 		DELAY(40000 * iter);
1707 	}
1708 
1709 	return nbusy;
1710 }
1711 
1712 /*
1713  * posix file system related system variables.
1714  */
1715 int
1716 fs_posix_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp,
1717     void *newp, size_t newlen, struct proc *p)
1718 {
1719 	/* all sysctl names at this level are terminal */
1720 	if (namelen != 1)
1721 		return (ENOTDIR);
1722 
1723 	switch (name[0]) {
1724 	case FS_POSIX_SETUID:
1725 		if (newp && securelevel > 0)
1726 			return (EPERM);
1727 		return(sysctl_int(oldp, oldlenp, newp, newlen, &suid_clear));
1728 	default:
1729 		return (EOPNOTSUPP);
1730 	}
1731 	/* NOTREACHED */
1732 }
1733 
1734 /*
1735  * file system related system variables.
1736  */
1737 int
1738 fs_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp,
1739     size_t newlen, struct proc *p)
1740 {
1741 	sysctlfn *fn;
1742 
1743 	switch (name[0]) {
1744 	case FS_POSIX:
1745 		fn = fs_posix_sysctl;
1746 		break;
1747 	default:
1748 		return (EOPNOTSUPP);
1749 	}
1750 	return (*fn)(name + 1, namelen - 1, oldp, oldlenp, newp, newlen, p);
1751 }
1752 
1753 
1754 /*
1755  * Routines dealing with vnodes and buffers
1756  */
1757 
1758 /*
1759  * Wait for all outstanding I/Os to complete
1760  *
1761  * Manipulates v_numoutput. Must be called at splbio()
1762  */
1763 int
1764 vwaitforio(struct vnode *vp, int slpflag, char *wmesg, int timeo)
1765 {
1766 	int error = 0;
1767 
1768 	splassert(IPL_BIO);
1769 
1770 	while (vp->v_numoutput) {
1771 		vp->v_bioflag |= VBIOWAIT;
1772 		error = tsleep(&vp->v_numoutput,
1773 		    slpflag | (PRIBIO + 1), wmesg, timeo);
1774 		if (error)
1775 			break;
1776 	}
1777 
1778 	return (error);
1779 }
1780 
1781 /*
1782  * Update outstanding I/O count and do wakeup if requested.
1783  *
1784  * Manipulates v_numoutput. Must be called at splbio()
1785  */
1786 void
1787 vwakeup(struct vnode *vp)
1788 {
1789 	splassert(IPL_BIO);
1790 
1791 	if (vp != NULL) {
1792 		if (vp->v_numoutput-- == 0)
1793 			panic("vwakeup: neg numoutput");
1794 		if ((vp->v_bioflag & VBIOWAIT) && vp->v_numoutput == 0) {
1795 			vp->v_bioflag &= ~VBIOWAIT;
1796 			wakeup(&vp->v_numoutput);
1797 		}
1798 	}
1799 }
1800 
1801 /*
1802  * Flush out and invalidate all buffers associated with a vnode.
1803  * Called with the underlying object locked.
1804  */
1805 int
1806 vinvalbuf(struct vnode *vp, int flags, struct ucred *cred, struct proc *p,
1807     int slpflag, int slptimeo)
1808 {
1809 	struct buf *bp;
1810 	struct buf *nbp, *blist;
1811 	int s, error;
1812 
1813 #ifdef VFSDEBUG
1814 	if ((vp->v_flag & VLOCKSWORK) && !VOP_ISLOCKED(vp))
1815 		panic("vinvalbuf(): vp isn't locked");
1816 #endif
1817 
1818 	if (flags & V_SAVE) {
1819 		s = splbio();
1820 		vwaitforio(vp, 0, "vinvalbuf", 0);
1821 		if (!LIST_EMPTY(&vp->v_dirtyblkhd)) {
1822 			splx(s);
1823 			if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, p)) != 0)
1824 				return (error);
1825 			s = splbio();
1826 			if (vp->v_numoutput > 0 ||
1827 			    !LIST_EMPTY(&vp->v_dirtyblkhd))
1828 				panic("vinvalbuf: dirty bufs");
1829 		}
1830 		splx(s);
1831 	}
1832 loop:
1833 	s = splbio();
1834 	for (;;) {
1835 		if ((blist = LIST_FIRST(&vp->v_cleanblkhd)) &&
1836 		    (flags & V_SAVEMETA))
1837 			while (blist && blist->b_lblkno < 0)
1838 				blist = LIST_NEXT(blist, b_vnbufs);
1839 		if (blist == NULL &&
1840 		    (blist = LIST_FIRST(&vp->v_dirtyblkhd)) &&
1841 		    (flags & V_SAVEMETA))
1842 			while (blist && blist->b_lblkno < 0)
1843 				blist = LIST_NEXT(blist, b_vnbufs);
1844 		if (!blist)
1845 			break;
1846 
1847 		for (bp = blist; bp; bp = nbp) {
1848 			nbp = LIST_NEXT(bp, b_vnbufs);
1849 			if (flags & V_SAVEMETA && bp->b_lblkno < 0)
1850 				continue;
1851 			if (bp->b_flags & B_BUSY) {
1852 				bp->b_flags |= B_WANTED;
1853 				error = tsleep(bp, slpflag | (PRIBIO + 1),
1854 				    "vinvalbuf", slptimeo);
1855 				if (error) {
1856 					splx(s);
1857 					return (error);
1858 				}
1859 				break;
1860 			}
1861 			bremfree(bp);
1862 			buf_acquire(bp);
1863 			/*
1864 			 * XXX Since there are no node locks for NFS, I believe
1865 			 * there is a slight chance that a delayed write will
1866 			 * occur while sleeping just above, so check for it.
1867 			 */
1868 			if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) {
1869 				splx(s);
1870 				(void) VOP_BWRITE(bp);
1871 				goto loop;
1872 			}
1873 			bp->b_flags |= B_INVAL;
1874 			brelse(bp);
1875 		}
1876 	}
1877 	if (!(flags & V_SAVEMETA) &&
1878 	    (!LIST_EMPTY(&vp->v_dirtyblkhd) || !LIST_EMPTY(&vp->v_cleanblkhd)))
1879 		panic("vinvalbuf: flush failed");
1880 	splx(s);
1881 	return (0);
1882 }
1883 
1884 void
1885 vflushbuf(struct vnode *vp, int sync)
1886 {
1887 	struct buf *bp, *nbp;
1888 	int s;
1889 
1890 loop:
1891 	s = splbio();
1892 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd);
1893 	    bp != LIST_END(&vp->v_dirtyblkhd); bp = nbp) {
1894 		nbp = LIST_NEXT(bp, b_vnbufs);
1895 		if ((bp->b_flags & B_BUSY))
1896 			continue;
1897 		if ((bp->b_flags & B_DELWRI) == 0)
1898 			panic("vflushbuf: not dirty");
1899 		bremfree(bp);
1900 		buf_acquire(bp);
1901 		splx(s);
1902 		/*
1903 		 * Wait for I/O associated with indirect blocks to complete,
1904 		 * since there is no way to quickly wait for them below.
1905 		 */
1906 		if (bp->b_vp == vp || sync == 0)
1907 			(void) bawrite(bp);
1908 		else
1909 			(void) bwrite(bp);
1910 		goto loop;
1911 	}
1912 	if (sync == 0) {
1913 		splx(s);
1914 		return;
1915 	}
1916 	vwaitforio(vp, 0, "vflushbuf", 0);
1917 	if (!LIST_EMPTY(&vp->v_dirtyblkhd)) {
1918 		splx(s);
1919 #ifdef DIAGNOSTIC
1920 		vprint("vflushbuf: dirty", vp);
1921 #endif
1922 		goto loop;
1923 	}
1924 	splx(s);
1925 }
1926 
1927 /*
1928  * Associate a buffer with a vnode.
1929  *
1930  * Manipulates buffer vnode queues. Must be called at splbio().
1931  */
1932 void
1933 bgetvp(struct vnode *vp, struct buf *bp)
1934 {
1935 	splassert(IPL_BIO);
1936 
1937 
1938 	if (bp->b_vp)
1939 		panic("bgetvp: not free");
1940 	vhold(vp);
1941 	bp->b_vp = vp;
1942 	if (vp->v_type == VBLK || vp->v_type == VCHR)
1943 		bp->b_dev = vp->v_rdev;
1944 	else
1945 		bp->b_dev = NODEV;
1946 	/*
1947 	 * Insert onto list for new vnode.
1948 	 */
1949 	bufinsvn(bp, &vp->v_cleanblkhd);
1950 }
1951 
1952 /*
1953  * Disassociate a buffer from a vnode.
1954  *
1955  * Manipulates vnode buffer queues. Must be called at splbio().
1956  */
1957 void
1958 brelvp(struct buf *bp)
1959 {
1960 	struct vnode *vp;
1961 
1962 	splassert(IPL_BIO);
1963 
1964 	if ((vp = bp->b_vp) == (struct vnode *) 0)
1965 		panic("brelvp: NULL");
1966 	/*
1967 	 * Delete from old vnode list, if on one.
1968 	 */
1969 	if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
1970 		bufremvn(bp);
1971 	if ((vp->v_bioflag & VBIOONSYNCLIST) &&
1972 	    LIST_FIRST(&vp->v_dirtyblkhd) == NULL) {
1973 		vp->v_bioflag &= ~VBIOONSYNCLIST;
1974 		LIST_REMOVE(vp, v_synclist);
1975 	}
1976 	bp->b_vp = NULL;
1977 
1978 	vdrop(vp);
1979 }
1980 
1981 /*
1982  * Replaces the current vnode associated with the buffer, if any,
1983  * with a new vnode.
1984  *
1985  * If an output I/O is pending on the buffer, the old vnode
1986  * I/O count is adjusted.
1987  *
1988  * Ignores vnode buffer queues. Must be called at splbio().
1989  */
1990 void
1991 buf_replacevnode(struct buf *bp, struct vnode *newvp)
1992 {
1993 	struct vnode *oldvp = bp->b_vp;
1994 
1995 	splassert(IPL_BIO);
1996 
1997 	if (oldvp)
1998 		brelvp(bp);
1999 
2000 	if ((bp->b_flags & (B_READ | B_DONE)) == 0) {
2001 		newvp->v_numoutput++;	/* put it on swapdev */
2002 		vwakeup(oldvp);
2003 	}
2004 
2005 	bgetvp(newvp, bp);
2006 	bufremvn(bp);
2007 }
2008 
2009 /*
2010  * Used to assign buffers to the appropriate clean or dirty list on
2011  * the vnode and to add newly dirty vnodes to the appropriate
2012  * filesystem syncer list.
2013  *
2014  * Manipulates vnode buffer queues. Must be called at splbio().
2015  */
2016 void
2017 reassignbuf(struct buf *bp)
2018 {
2019 	struct buflists *listheadp;
2020 	int delay;
2021 	struct vnode *vp = bp->b_vp;
2022 
2023 	splassert(IPL_BIO);
2024 
2025 	/*
2026 	 * Delete from old vnode list, if on one.
2027 	 */
2028 	if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
2029 		bufremvn(bp);
2030 
2031 	/*
2032 	 * If dirty, put on list of dirty buffers;
2033 	 * otherwise insert onto list of clean buffers.
2034 	 */
2035 	if ((bp->b_flags & B_DELWRI) == 0) {
2036 		listheadp = &vp->v_cleanblkhd;
2037 		if ((vp->v_bioflag & VBIOONSYNCLIST) &&
2038 		    LIST_FIRST(&vp->v_dirtyblkhd) == NULL) {
2039 			vp->v_bioflag &= ~VBIOONSYNCLIST;
2040 			LIST_REMOVE(vp, v_synclist);
2041 		}
2042 	} else {
2043 		listheadp = &vp->v_dirtyblkhd;
2044 		if ((vp->v_bioflag & VBIOONSYNCLIST) == 0) {
2045 			switch (vp->v_type) {
2046 			case VDIR:
2047 				delay = syncdelay / 2;
2048 				break;
2049 			case VBLK:
2050 				if (vp->v_specmountpoint != NULL) {
2051 					delay = syncdelay / 3;
2052 					break;
2053 				}
2054 				/* FALLTHROUGH */
2055 			default:
2056 				delay = syncdelay;
2057 			}
2058 			vn_syncer_add_to_worklist(vp, delay);
2059 		}
2060 	}
2061 	bufinsvn(bp, listheadp);
2062 }
2063 
2064 int
2065 vfs_register(struct vfsconf *vfs)
2066 {
2067 	struct vfsconf *vfsp;
2068 	struct vfsconf **vfspp;
2069 
2070 #ifdef DIAGNOSTIC
2071 	/* Paranoia? */
2072 	if (vfs->vfc_refcount != 0)
2073 		printf("vfs_register called with vfc_refcount > 0\n");
2074 #endif
2075 
2076 	/* Check if filesystem already known */
2077 	for (vfspp = &vfsconf, vfsp = vfsconf; vfsp;
2078 	    vfspp = &vfsp->vfc_next, vfsp = vfsp->vfc_next)
2079 		if (strcmp(vfsp->vfc_name, vfs->vfc_name) == 0)
2080 			return (EEXIST);
2081 
2082 	if (vfs->vfc_typenum > maxvfsconf)
2083 		maxvfsconf = vfs->vfc_typenum;
2084 
2085 	vfs->vfc_next = NULL;
2086 
2087 	/* Add to the end of the list */
2088 	*vfspp = vfs;
2089 
2090 	/* Call vfs_init() */
2091 	if (vfs->vfc_vfsops->vfs_init)
2092 		(*(vfs->vfc_vfsops->vfs_init))(vfs);
2093 
2094 	return 0;
2095 }
2096 
2097 int
2098 vfs_unregister(struct vfsconf *vfs)
2099 {
2100 	struct vfsconf *vfsp;
2101 	struct vfsconf **vfspp;
2102 	int maxtypenum;
2103 
2104 	/* Find our vfsconf struct */
2105 	for (vfspp = &vfsconf, vfsp = vfsconf; vfsp;
2106 	    vfspp = &vfsp->vfc_next, vfsp = vfsp->vfc_next) {
2107 		if (strcmp(vfsp->vfc_name, vfs->vfc_name) == 0)
2108 			break;
2109 	}
2110 
2111 	if (!vfsp)			/* Not found */
2112 		return (ENOENT);
2113 
2114 	if (vfsp->vfc_refcount)		/* In use */
2115 		return (EBUSY);
2116 
2117 	/* Remove from list and free */
2118 	*vfspp = vfsp->vfc_next;
2119 
2120 	maxtypenum = 0;
2121 
2122 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2123 		if (vfsp->vfc_typenum > maxtypenum)
2124 			maxtypenum = vfsp->vfc_typenum;
2125 
2126 	maxvfsconf = maxtypenum;
2127 	return 0;
2128 }
2129 
2130 /*
2131  * Check if vnode represents a disk device
2132  */
2133 int
2134 vn_isdisk(struct vnode *vp, int *errp)
2135 {
2136 	if (vp->v_type != VBLK && vp->v_type != VCHR)
2137 		return (0);
2138 
2139 	return (1);
2140 }
2141 
2142 #ifdef DDB
2143 #include <machine/db_machdep.h>
2144 #include <ddb/db_interface.h>
2145 #include <ddb/db_output.h>
2146 
2147 void
2148 vfs_buf_print(struct buf *bp, int full, int (*pr)(const char *, ...))
2149 {
2150 
2151 	(*pr)("  vp %p lblkno 0x%llx blkno 0x%llx dev 0x%x\n"
2152 	      "  proc %p error %d flags %b\n",
2153 	    bp->b_vp, (int64_t)bp->b_lblkno, (int64_t)bp->b_blkno, bp->b_dev,
2154 	    bp->b_proc, bp->b_error, bp->b_flags, B_BITS);
2155 
2156 	(*pr)("  bufsize 0x%lx bcount 0x%lx resid 0x%lx sync 0x%x\n"
2157 	      "  data %p saveaddr %p dep %p iodone %p\n",
2158 	    bp->b_bufsize, bp->b_bcount, (long)bp->b_resid, bp->b_synctime,
2159 	    bp->b_data, bp->b_saveaddr, LIST_FIRST(&bp->b_dep), bp->b_iodone);
2160 
2161 	(*pr)("  dirty {off 0x%x end 0x%x} valid {off 0x%x end 0x%x}\n",
2162 	    bp->b_dirtyoff, bp->b_dirtyend, bp->b_validoff, bp->b_validend);
2163 
2164 #ifdef FFS_SOFTUPDATES
2165 	if (full)
2166 		softdep_print(bp, full, pr);
2167 #endif
2168 }
2169 
2170 const char *vtypes[] = { VTYPE_NAMES };
2171 const char *vtags[] = { VTAG_NAMES };
2172 
2173 void
2174 vfs_vnode_print(struct vnode *vp, int full, int (*pr)(const char *, ...))
2175 {
2176 
2177 #define	NENTS(n)	(sizeof n / sizeof(n[0]))
2178 	(*pr)("tag %s(%d) type %s(%d) mount %p typedata %p\n",
2179 	      vp->v_tag > NENTS(vtags)? "<unk>":vtags[vp->v_tag], vp->v_tag,
2180 	      vp->v_type > NENTS(vtypes)? "<unk>":vtypes[vp->v_type],
2181 	      vp->v_type, vp->v_mount, vp->v_mountedhere);
2182 
2183 	(*pr)("data %p usecount %d writecount %ld holdcnt %ld numoutput %d\n",
2184 	      vp->v_data, vp->v_usecount, vp->v_writecount,
2185 	      vp->v_holdcnt, vp->v_numoutput);
2186 
2187 	/* uvm_object_printit(&vp->v_uobj, full, pr); */
2188 
2189 	if (full) {
2190 		struct buf *bp;
2191 
2192 		(*pr)("clean bufs:\n");
2193 		LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) {
2194 			(*pr)(" bp %p\n", bp);
2195 			vfs_buf_print(bp, full, pr);
2196 		}
2197 
2198 		(*pr)("dirty bufs:\n");
2199 		LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) {
2200 			(*pr)(" bp %p\n", bp);
2201 			vfs_buf_print(bp, full, pr);
2202 		}
2203 	}
2204 }
2205 
2206 void
2207 vfs_mount_print(struct mount *mp, int full, int (*pr)(const char *, ...))
2208 {
2209 	struct vfsconf *vfc = mp->mnt_vfc;
2210 	struct vnode *vp;
2211 	int cnt = 0;
2212 
2213 	(*pr)("flags %b\nvnodecovered %p syncer %p data %p\n",
2214 	    mp->mnt_flag, MNT_BITS,
2215 	    mp->mnt_vnodecovered, mp->mnt_syncer, mp->mnt_data);
2216 
2217 	(*pr)("vfsconf: ops %p name \"%s\" num %d ref %d flags 0x%x\n",
2218             vfc->vfc_vfsops, vfc->vfc_name, vfc->vfc_typenum,
2219 	    vfc->vfc_refcount, vfc->vfc_flags);
2220 
2221 	(*pr)("statvfs cache: bsize %x iosize %x\nblocks %llu free %llu avail %lld\n",
2222 	    mp->mnt_stat.f_bsize, mp->mnt_stat.f_iosize, mp->mnt_stat.f_blocks,
2223 	    mp->mnt_stat.f_bfree, mp->mnt_stat.f_bavail);
2224 
2225 	(*pr)("  files %llu ffiles %llu favail $lld\n", mp->mnt_stat.f_files,
2226 	    mp->mnt_stat.f_ffree, mp->mnt_stat.f_favail);
2227 
2228 	(*pr)("  f_fsidx {0x%x, 0x%x} owner %u ctime 0x%x\n",
2229 	    mp->mnt_stat.f_fsid.val[0], mp->mnt_stat.f_fsid.val[1],
2230 	    mp->mnt_stat.f_owner, mp->mnt_stat.f_ctime);
2231 
2232  	(*pr)("  syncwrites %llu asyncwrites = %llu\n",
2233 	    mp->mnt_stat.f_syncwrites, mp->mnt_stat.f_asyncwrites);
2234 
2235  	(*pr)("  syncreads %llu asyncreads = %llu\n",
2236 	    mp->mnt_stat.f_syncreads, mp->mnt_stat.f_asyncreads);
2237 
2238 	(*pr)("  fstype \"%s\" mnton \"%s\" mntfrom \"%s\"\n",
2239 	    mp->mnt_stat.f_fstypename, mp->mnt_stat.f_mntonname,
2240 	    mp->mnt_stat.f_mntfromname);
2241 
2242 	(*pr)("locked vnodes:");
2243 	/* XXX would take mountlist lock, except ddb has no context */
2244 	LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes)
2245 		if (VOP_ISLOCKED(vp)) {
2246 			if (!LIST_NEXT(vp, v_mntvnodes))
2247 				(*pr)(" %p", vp);
2248 			else if (!(cnt++ % (72 / (sizeof(void *) * 2 + 4))))
2249 				(*pr)("\n\t%p", vp);
2250 			else
2251 				(*pr)(", %p", vp);
2252 		}
2253 	(*pr)("\n");
2254 
2255 	if (full) {
2256 		(*pr)("all vnodes:\n\t");
2257 		/* XXX would take mountlist lock, except ddb has no context */
2258 		LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes)
2259 			if (!LIST_NEXT(vp, v_mntvnodes))
2260 				(*pr)(" %p", vp);
2261 			else if (!(cnt++ % (72 / (sizeof(void *) * 2 + 4))))
2262 				(*pr)(" %p,\n\t", vp);
2263 			else
2264 				(*pr)(" %p,", vp);
2265 		(*pr)("\n");
2266 	}
2267 }
2268 #endif /* DDB */
2269 
2270 void
2271 copy_statfs_info(struct statfs *sbp, const struct mount *mp)
2272 {
2273 	const struct statfs *mbp;
2274 
2275 	strncpy(sbp->f_fstypename, mp->mnt_vfc->vfc_name, MFSNAMELEN);
2276 
2277 	if (sbp == (mbp = &mp->mnt_stat))
2278 		return;
2279 
2280 	sbp->f_fsid = mbp->f_fsid;
2281 	sbp->f_owner = mbp->f_owner;
2282 	sbp->f_flags = mbp->f_flags;
2283 	sbp->f_syncwrites = mbp->f_syncwrites;
2284 	sbp->f_asyncwrites = mbp->f_asyncwrites;
2285 	sbp->f_syncreads = mbp->f_syncreads;
2286 	sbp->f_asyncreads = mbp->f_asyncreads;
2287 	sbp->f_namemax = mbp->f_namemax;
2288 	bcopy(mp->mnt_stat.f_mntonname, sbp->f_mntonname, MNAMELEN);
2289 	bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
2290 	bcopy(&mp->mnt_stat.mount_info.ufs_args, &sbp->mount_info.ufs_args,
2291 	    sizeof(struct ufs_args));
2292 }
2293 
2294