xref: /openbsd-src/sys/kern/vfs_subr.c (revision 897fc685943471cf985a0fe38ba076ea6fe74fa5)
1 /*	$OpenBSD: vfs_subr.c,v 1.267 2018/03/07 18:30:23 bluhm 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/sysctl.h>
48 #include <sys/mount.h>
49 #include <sys/time.h>
50 #include <sys/fcntl.h>
51 #include <sys/kernel.h>
52 #include <sys/conf.h>
53 #include <sys/vnode.h>
54 #include <sys/lock.h>
55 #include <sys/stat.h>
56 #include <sys/acct.h>
57 #include <sys/namei.h>
58 #include <sys/ucred.h>
59 #include <sys/buf.h>
60 #include <sys/errno.h>
61 #include <sys/malloc.h>
62 #include <sys/mbuf.h>
63 #include <sys/syscallargs.h>
64 #include <sys/pool.h>
65 #include <sys/tree.h>
66 #include <sys/specdev.h>
67 
68 #include <netinet/in.h>
69 
70 #include <uvm/uvm_extern.h>
71 #include <uvm/uvm_vnode.h>
72 
73 #include "softraid.h"
74 
75 void sr_quiesce(void);
76 
77 enum vtype iftovt_tab[16] = {
78 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
79 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
80 };
81 
82 int	vttoif_tab[9] = {
83 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
84 	S_IFSOCK, S_IFIFO, S_IFMT,
85 };
86 
87 int prtactive = 0;		/* 1 => print out reclaim of active vnodes */
88 int suid_clear = 1;		/* 1 => clear SUID / SGID on owner change */
89 
90 /*
91  * Insq/Remq for the vnode usage lists.
92  */
93 #define	bufinsvn(bp, dp)	LIST_INSERT_HEAD(dp, bp, b_vnbufs)
94 #define	bufremvn(bp) {							\
95 	LIST_REMOVE(bp, b_vnbufs);					\
96 	LIST_NEXT(bp, b_vnbufs) = NOLIST;				\
97 }
98 
99 struct freelst vnode_hold_list;	/* list of vnodes referencing buffers */
100 struct freelst vnode_free_list;	/* vnode free list */
101 
102 struct mntlist mountlist;	/* mounted filesystem list */
103 
104 void	vclean(struct vnode *, int, struct proc *);
105 
106 void insmntque(struct vnode *, struct mount *);
107 int getdevvp(dev_t, struct vnode **, enum vtype);
108 
109 int vfs_hang_addrlist(struct mount *, struct netexport *,
110 				  struct export_args *);
111 int vfs_free_netcred(struct radix_node *, void *, u_int);
112 void vfs_free_addrlist(struct netexport *);
113 void vputonfreelist(struct vnode *);
114 
115 int vflush_vnode(struct vnode *, void *);
116 int maxvnodes;
117 
118 void vfs_unmountall(void);
119 
120 #ifdef DEBUG
121 void printlockedvnodes(void);
122 #endif
123 
124 struct pool vnode_pool;
125 struct pool uvm_vnode_pool;
126 
127 static inline int rb_buf_compare(const struct buf *b1, const struct buf *b2);
128 RBT_GENERATE(buf_rb_bufs, buf, b_rbbufs, rb_buf_compare);
129 
130 static inline int
131 rb_buf_compare(const struct buf *b1, const struct buf *b2)
132 {
133 	if (b1->b_lblkno < b2->b_lblkno)
134 		return(-1);
135 	if (b1->b_lblkno > b2->b_lblkno)
136 		return(1);
137 	return(0);
138 }
139 
140 /*
141  * Initialize the vnode management data structures.
142  */
143 void
144 vntblinit(void)
145 {
146 	/* buffer cache may need a vnode for each buffer */
147 	maxvnodes = 2 * initialvnodes;
148 	pool_init(&vnode_pool, sizeof(struct vnode), 0, IPL_NONE,
149 	    PR_WAITOK, "vnodes", NULL);
150 	pool_init(&uvm_vnode_pool, sizeof(struct uvm_vnode), 0, IPL_NONE,
151 	    PR_WAITOK, "uvmvnodes", NULL);
152 	TAILQ_INIT(&vnode_hold_list);
153 	TAILQ_INIT(&vnode_free_list);
154 	TAILQ_INIT(&mountlist);
155 	/*
156 	 * Initialize the filesystem syncer.
157 	 */
158 	vn_initialize_syncerd();
159 
160 #ifdef NFSSERVER
161 	rn_init(sizeof(struct sockaddr_in));
162 #endif /* NFSSERVER */
163 }
164 
165 /*
166  * Mark a mount point as busy. Used to synchronize access and to delay
167  * unmounting.
168  *
169  * Default behaviour is to attempt getting a READ lock and in case of an
170  * ongoing unmount, to wait for it to finish and then return failure.
171  */
172 int
173 vfs_busy(struct mount *mp, int flags)
174 {
175 	int rwflags = 0;
176 
177 	/* new mountpoints need their lock initialised */
178 	if (mp->mnt_lock.rwl_name == NULL)
179 		rw_init_flags(&mp->mnt_lock, "vfslock", RWL_IS_VNODE);
180 
181 	if (flags & VB_WRITE)
182 		rwflags |= RW_WRITE;
183 	else
184 		rwflags |= RW_READ;
185 
186 	if (flags & VB_WAIT)
187 		rwflags |= RW_SLEEPFAIL;
188 	else
189 		rwflags |= RW_NOSLEEP;
190 
191 	if (rw_enter(&mp->mnt_lock, rwflags))
192 		return (EBUSY);
193 
194 	return (0);
195 }
196 
197 /*
198  * Free a busy file system
199  */
200 void
201 vfs_unbusy(struct mount *mp)
202 {
203 	rw_exit(&mp->mnt_lock);
204 }
205 
206 int
207 vfs_isbusy(struct mount *mp)
208 {
209 	if (RWLOCK_OWNER(&mp->mnt_lock) > 0)
210 		return (1);
211 	else
212 		return (0);
213 }
214 
215 /*
216  * Lookup a filesystem type, and if found allocate and initialize
217  * a mount structure for it.
218  *
219  * Devname is usually updated by mount(8) after booting.
220  */
221 int
222 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
223 {
224 	struct vfsconf *vfsp;
225 	struct mount *mp;
226 
227 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
228 		if (!strcmp(vfsp->vfc_name, fstypename))
229 			break;
230 	if (vfsp == NULL)
231 		return (ENODEV);
232 	mp = malloc(sizeof(*mp), M_MOUNT, M_WAITOK|M_ZERO);
233 	(void)vfs_busy(mp, VB_READ|VB_NOWAIT);
234 	LIST_INIT(&mp->mnt_vnodelist);
235 	mp->mnt_vfc = vfsp;
236 	mp->mnt_op = vfsp->vfc_vfsops;
237 	mp->mnt_flag = MNT_RDONLY;
238 	mp->mnt_vnodecovered = NULLVP;
239 	vfsp->vfc_refcount++;
240 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
241 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
242 	mp->mnt_stat.f_mntonname[0] = '/';
243 	copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN, 0);
244 	copystr(devname, mp->mnt_stat.f_mntfromspec, MNAMELEN, 0);
245 	*mpp = mp;
246  	return (0);
247  }
248 
249 /*
250  * Lookup a mount point by filesystem identifier.
251  */
252 struct mount *
253 vfs_getvfs(fsid_t *fsid)
254 {
255 	struct mount *mp;
256 
257 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
258 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
259 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
260 			return (mp);
261 		}
262 	}
263 
264 	return (NULL);
265 }
266 
267 
268 /*
269  * Get a new unique fsid
270  */
271 void
272 vfs_getnewfsid(struct mount *mp)
273 {
274 	static u_short xxxfs_mntid;
275 
276 	fsid_t tfsid;
277 	int mtype;
278 
279 	mtype = mp->mnt_vfc->vfc_typenum;
280 	mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev + mtype, 0);
281 	mp->mnt_stat.f_fsid.val[1] = mtype;
282 	if (xxxfs_mntid == 0)
283 		++xxxfs_mntid;
284 	tfsid.val[0] = makedev(nblkdev + mtype, xxxfs_mntid);
285 	tfsid.val[1] = mtype;
286 	if (!TAILQ_EMPTY(&mountlist)) {
287 		while (vfs_getvfs(&tfsid)) {
288 			tfsid.val[0]++;
289 			xxxfs_mntid++;
290 		}
291 	}
292 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
293 }
294 
295 /*
296  * Set vnode attributes to VNOVAL
297  */
298 void
299 vattr_null(struct vattr *vap)
300 {
301 
302 	vap->va_type = VNON;
303 	/*
304 	 * Don't get fancy: u_quad_t = u_int = VNOVAL leaves the u_quad_t
305 	 * with 2^31-1 instead of 2^64-1.  Just write'm out and let
306 	 * the compiler do its job.
307 	 */
308 	vap->va_mode = VNOVAL;
309 	vap->va_nlink = VNOVAL;
310 	vap->va_uid = VNOVAL;
311 	vap->va_gid = VNOVAL;
312 	vap->va_fsid = VNOVAL;
313 	vap->va_fileid = VNOVAL;
314 	vap->va_size = VNOVAL;
315 	vap->va_blocksize = VNOVAL;
316 	vap->va_atime.tv_sec = VNOVAL;
317 	vap->va_atime.tv_nsec = VNOVAL;
318 	vap->va_mtime.tv_sec = VNOVAL;
319 	vap->va_mtime.tv_nsec = VNOVAL;
320 	vap->va_ctime.tv_sec = VNOVAL;
321 	vap->va_ctime.tv_nsec = VNOVAL;
322 	vap->va_gen = VNOVAL;
323 	vap->va_flags = VNOVAL;
324 	vap->va_rdev = VNOVAL;
325 	vap->va_bytes = VNOVAL;
326 	vap->va_filerev = VNOVAL;
327 	vap->va_vaflags = 0;
328 }
329 
330 /*
331  * Routines having to do with the management of the vnode table.
332  */
333 long numvnodes;
334 
335 /*
336  * Return the next vnode from the free list.
337  */
338 int
339 getnewvnode(enum vtagtype tag, struct mount *mp, struct vops *vops,
340     struct vnode **vpp)
341 {
342 	struct proc *p = curproc;
343 	struct freelst *listhd;
344 	static int toggle;
345 	struct vnode *vp;
346 	int s;
347 
348 	/*
349 	 * allow maxvnodes to increase if the buffer cache itself
350 	 * is big enough to justify it. (we don't shrink it ever)
351 	 */
352 	maxvnodes = maxvnodes < bcstats.numbufs ? bcstats.numbufs
353 	    : maxvnodes;
354 
355 	/*
356 	 * We must choose whether to allocate a new vnode or recycle an
357 	 * existing one. The criterion for allocating a new one is that
358 	 * the total number of vnodes is less than the number desired or
359 	 * there are no vnodes on either free list. Generally we only
360 	 * want to recycle vnodes that have no buffers associated with
361 	 * them, so we look first on the vnode_free_list. If it is empty,
362 	 * we next consider vnodes with referencing buffers on the
363 	 * vnode_hold_list. The toggle ensures that half the time we
364 	 * will use a buffer from the vnode_hold_list, and half the time
365 	 * we will allocate a new one unless the list has grown to twice
366 	 * the desired size. We are reticent to recycle vnodes from the
367 	 * vnode_hold_list because we will lose the identity of all its
368 	 * referencing buffers.
369 	 */
370 	toggle ^= 1;
371 	if (numvnodes / 2 > maxvnodes)
372 		toggle = 0;
373 
374 	s = splbio();
375 	if ((numvnodes < maxvnodes) ||
376 	    ((TAILQ_FIRST(listhd = &vnode_free_list) == NULL) &&
377 	    ((TAILQ_FIRST(listhd = &vnode_hold_list) == NULL) || toggle))) {
378 		splx(s);
379 		vp = pool_get(&vnode_pool, PR_WAITOK | PR_ZERO);
380 		vp->v_uvm = pool_get(&uvm_vnode_pool, PR_WAITOK | PR_ZERO);
381 		vp->v_uvm->u_vnode = vp;
382 		RBT_INIT(buf_rb_bufs, &vp->v_bufs_tree);
383 		cache_tree_init(&vp->v_nc_tree);
384 		TAILQ_INIT(&vp->v_cache_dst);
385 		numvnodes++;
386 	} else {
387 		TAILQ_FOREACH(vp, listhd, v_freelist) {
388 			if (VOP_ISLOCKED(vp) == 0)
389 				break;
390 		}
391 		/*
392 		 * Unless this is a bad time of the month, at most
393 		 * the first NCPUS items on the free list are
394 		 * locked, so this is close enough to being empty.
395 		 */
396 		if (vp == NULL) {
397 			splx(s);
398 			tablefull("vnode");
399 			*vpp = 0;
400 			return (ENFILE);
401 		}
402 
403 #ifdef DIAGNOSTIC
404 		if (vp->v_usecount) {
405 			vprint("free vnode", vp);
406 			panic("free vnode isn't");
407 		}
408 #endif
409 
410 		TAILQ_REMOVE(listhd, vp, v_freelist);
411 		vp->v_bioflag &= ~VBIOONFREELIST;
412 		splx(s);
413 
414 		if (vp->v_type != VBAD)
415 			vgonel(vp, p);
416 #ifdef DIAGNOSTIC
417 		if (vp->v_data) {
418 			vprint("cleaned vnode", vp);
419 			panic("cleaned vnode isn't");
420 		}
421 		s = splbio();
422 		if (vp->v_numoutput)
423 			panic("Clean vnode has pending I/O's");
424 		splx(s);
425 #endif
426 		vp->v_flag = 0;
427 		vp->v_socket = 0;
428 	}
429 	cache_purge(vp);
430 	vp->v_type = VNON;
431 	vp->v_tag = tag;
432 	vp->v_op = vops;
433 	insmntque(vp, mp);
434 	*vpp = vp;
435 	vp->v_usecount = 1;
436 	vp->v_data = 0;
437 	return (0);
438 }
439 
440 /*
441  * Move a vnode from one mount queue to another.
442  */
443 void
444 insmntque(struct vnode *vp, struct mount *mp)
445 {
446 	/*
447 	 * Delete from old mount point vnode list, if on one.
448 	 */
449 	if (vp->v_mount != NULL)
450 		LIST_REMOVE(vp, v_mntvnodes);
451 	/*
452 	 * Insert into list of vnodes for the new mount point, if available.
453 	 */
454 	if ((vp->v_mount = mp) != NULL)
455 		LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes);
456 }
457 
458 /*
459  * Create a vnode for a block device.
460  * Used for root filesystem, argdev, and swap areas.
461  * Also used for memory file system special devices.
462  */
463 int
464 bdevvp(dev_t dev, struct vnode **vpp)
465 {
466 	return (getdevvp(dev, vpp, VBLK));
467 }
468 
469 /*
470  * Create a vnode for a character device.
471  * Used for console handling.
472  */
473 int
474 cdevvp(dev_t dev, struct vnode **vpp)
475 {
476 	return (getdevvp(dev, vpp, VCHR));
477 }
478 
479 /*
480  * Create a vnode for a device.
481  * Used by bdevvp (block device) for root file system etc.,
482  * and by cdevvp (character device) for console.
483  */
484 int
485 getdevvp(dev_t dev, struct vnode **vpp, enum vtype type)
486 {
487 	struct vnode *vp;
488 	struct vnode *nvp;
489 	int error;
490 
491 	if (dev == NODEV) {
492 		*vpp = NULLVP;
493 		return (0);
494 	}
495 	error = getnewvnode(VT_NON, NULL, &spec_vops, &nvp);
496 	if (error) {
497 		*vpp = NULLVP;
498 		return (error);
499 	}
500 	vp = nvp;
501 	vp->v_type = type;
502 	if ((nvp = checkalias(vp, dev, NULL)) != 0) {
503 		vput(vp);
504 		vp = nvp;
505 	}
506 	if (vp->v_type == VCHR && cdevsw[major(vp->v_rdev)].d_type == D_TTY)
507 		vp->v_flag |= VISTTY;
508 	*vpp = vp;
509 	return (0);
510 }
511 
512 /*
513  * Check to see if the new vnode represents a special device
514  * for which we already have a vnode (either because of
515  * bdevvp() or because of a different vnode representing
516  * the same block device). If such an alias exists, deallocate
517  * the existing contents and return the aliased vnode. The
518  * caller is responsible for filling it with its new contents.
519  */
520 struct vnode *
521 checkalias(struct vnode *nvp, dev_t nvp_rdev, struct mount *mp)
522 {
523 	struct proc *p = curproc;
524 	struct vnode *vp;
525 	struct vnode **vpp;
526 
527 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
528 		return (NULLVP);
529 
530 	vpp = &speclisth[SPECHASH(nvp_rdev)];
531 loop:
532 	for (vp = *vpp; vp; vp = vp->v_specnext) {
533 		if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type) {
534 			continue;
535 		}
536 		/*
537 		 * Alias, but not in use, so flush it out.
538 		 */
539 		if (vp->v_usecount == 0) {
540 			vgonel(vp, p);
541 			goto loop;
542 		}
543 		if (vget(vp, LK_EXCLUSIVE, p)) {
544 			goto loop;
545 		}
546 		break;
547 	}
548 
549 	/*
550 	 * Common case is actually in the if statement
551 	 */
552 	if (vp == NULL || !(vp->v_tag == VT_NON && vp->v_type == VBLK)) {
553 		nvp->v_specinfo = malloc(sizeof(struct specinfo), M_VNODE,
554 			M_WAITOK);
555 		nvp->v_rdev = nvp_rdev;
556 		nvp->v_hashchain = vpp;
557 		nvp->v_specnext = *vpp;
558 		nvp->v_specmountpoint = NULL;
559 		nvp->v_speclockf = NULL;
560 		nvp->v_specbitmap = NULL;
561 		if (nvp->v_type == VCHR &&
562 		    (cdevsw[major(nvp_rdev)].d_flags & D_CLONE) &&
563 		    (minor(nvp_rdev) >> CLONE_SHIFT == 0)) {
564 			if (vp != NULLVP)
565 				nvp->v_specbitmap = vp->v_specbitmap;
566 			else
567 				nvp->v_specbitmap = malloc(CLONE_MAPSZ,
568 				    M_VNODE, M_WAITOK | M_ZERO);
569 		}
570 		*vpp = nvp;
571 		if (vp != NULLVP) {
572 			nvp->v_flag |= VALIASED;
573 			vp->v_flag |= VALIASED;
574 			vput(vp);
575 		}
576 		return (NULLVP);
577 	}
578 
579 	/*
580 	 * This code is the uncommon case. It is called in case
581 	 * we found an alias that was VT_NON && vtype of VBLK
582 	 * This means we found a block device that was created
583 	 * using bdevvp.
584 	 * An example of such a vnode is the root partition device vnode
585 	 * created in ffs_mountroot.
586 	 *
587 	 * The vnodes created by bdevvp should not be aliased (why?).
588 	 */
589 
590 	VOP_UNLOCK(vp, p);
591 	vclean(vp, 0, p);
592 	vp->v_op = nvp->v_op;
593 	vp->v_tag = nvp->v_tag;
594 	nvp->v_type = VNON;
595 	insmntque(vp, mp);
596 	return (vp);
597 }
598 
599 /*
600  * Grab a particular vnode from the free list, increment its
601  * reference count and lock it. If the vnode lock bit is set,
602  * the vnode is being eliminated in vgone. In that case, we
603  * cannot grab it, so the process is awakened when the
604  * transition is completed, and an error code is returned to
605  * indicate that the vnode is no longer usable, possibly
606  * having been changed to a new file system type.
607  */
608 int
609 vget(struct vnode *vp, int flags, struct proc *p)
610 {
611 	int error, s, onfreelist;
612 
613 	/*
614 	 * If the vnode is in the process of being cleaned out for
615 	 * another use, we wait for the cleaning to finish and then
616 	 * return failure. Cleaning is determined by checking that
617 	 * the VXLOCK flag is set.
618 	 */
619 
620 	if (vp->v_flag & VXLOCK) {
621 		if (flags & LK_NOWAIT) {
622 			return (EBUSY);
623 		}
624 
625 		vp->v_flag |= VXWANT;
626 		tsleep(vp, PINOD, "vget", 0);
627 		return (ENOENT);
628 	}
629 
630 	onfreelist = vp->v_bioflag & VBIOONFREELIST;
631 	if (vp->v_usecount == 0 && onfreelist) {
632 		s = splbio();
633 		if (vp->v_holdcnt > 0)
634 			TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist);
635 		else
636 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
637 		vp->v_bioflag &= ~VBIOONFREELIST;
638 		splx(s);
639 	}
640 
641  	vp->v_usecount++;
642 	if (flags & LK_TYPE_MASK) {
643 		if ((error = vn_lock(vp, flags, p)) != 0) {
644 			vp->v_usecount--;
645 			if (vp->v_usecount == 0 && onfreelist)
646 				vputonfreelist(vp);
647 		}
648 		return (error);
649 	}
650 
651 	return (0);
652 }
653 
654 
655 /* Vnode reference. */
656 void
657 vref(struct vnode *vp)
658 {
659 #ifdef DIAGNOSTIC
660 	if (vp->v_usecount == 0)
661 		panic("vref used where vget required");
662 	if (vp->v_type == VNON)
663 		panic("vref on a VNON vnode");
664 #endif
665 	vp->v_usecount++;
666 }
667 
668 void
669 vputonfreelist(struct vnode *vp)
670 {
671 	int s;
672 	struct freelst *lst;
673 
674 	s = splbio();
675 #ifdef DIAGNOSTIC
676 	if (vp->v_usecount != 0)
677 		panic("Use count is not zero!");
678 
679 	if (vp->v_bioflag & VBIOONFREELIST) {
680 		vprint("vnode already on free list: ", vp);
681 		panic("vnode already on free list");
682 	}
683 #endif
684 
685 	vp->v_bioflag |= VBIOONFREELIST;
686 
687 	if (vp->v_holdcnt > 0)
688 		lst = &vnode_hold_list;
689 	else
690 		lst = &vnode_free_list;
691 
692 	if (vp->v_type == VBAD)
693 		TAILQ_INSERT_HEAD(lst, vp, v_freelist);
694 	else
695 		TAILQ_INSERT_TAIL(lst, vp, v_freelist);
696 
697 	splx(s);
698 }
699 
700 /*
701  * vput(), just unlock and vrele()
702  */
703 void
704 vput(struct vnode *vp)
705 {
706 	struct proc *p = curproc;
707 
708 #ifdef DIAGNOSTIC
709 	if (vp == NULL)
710 		panic("vput: null vp");
711 #endif
712 
713 #ifdef DIAGNOSTIC
714 	if (vp->v_usecount == 0) {
715 		vprint("vput: bad ref count", vp);
716 		panic("vput: ref cnt");
717 	}
718 #endif
719 	vp->v_usecount--;
720 	if (vp->v_usecount > 0) {
721 		VOP_UNLOCK(vp, p);
722 		return;
723 	}
724 
725 #ifdef DIAGNOSTIC
726 	if (vp->v_writecount != 0) {
727 		vprint("vput: bad writecount", vp);
728 		panic("vput: v_writecount != 0");
729 	}
730 #endif
731 
732 	VOP_INACTIVE(vp, p);
733 
734 	if (vp->v_usecount == 0 && !(vp->v_bioflag & VBIOONFREELIST))
735 		vputonfreelist(vp);
736 }
737 
738 /*
739  * Vnode release - use for active VNODES.
740  * If count drops to zero, call inactive routine and return to freelist.
741  * Returns 0 if it did not sleep.
742  */
743 int
744 vrele(struct vnode *vp)
745 {
746 	struct proc *p = curproc;
747 
748 #ifdef DIAGNOSTIC
749 	if (vp == NULL)
750 		panic("vrele: null vp");
751 #endif
752 #ifdef DIAGNOSTIC
753 	if (vp->v_usecount == 0) {
754 		vprint("vrele: bad ref count", vp);
755 		panic("vrele: ref cnt");
756 	}
757 #endif
758 	vp->v_usecount--;
759 	if (vp->v_usecount > 0) {
760 		return (0);
761 	}
762 
763 #ifdef DIAGNOSTIC
764 	if (vp->v_writecount != 0) {
765 		vprint("vrele: bad writecount", vp);
766 		panic("vrele: v_writecount != 0");
767 	}
768 #endif
769 
770 	if (vn_lock(vp, LK_EXCLUSIVE, p)) {
771 #ifdef DIAGNOSTIC
772 		vprint("vrele: cannot lock", vp);
773 #endif
774 		return (1);
775 	}
776 
777 	VOP_INACTIVE(vp, p);
778 
779 	if (vp->v_usecount == 0 && !(vp->v_bioflag & VBIOONFREELIST))
780 		vputonfreelist(vp);
781 	return (1);
782 }
783 
784 /* Page or buffer structure gets a reference. */
785 void
786 vhold(struct vnode *vp)
787 {
788 	/*
789 	 * If it is on the freelist and the hold count is currently
790 	 * zero, move it to the hold list.
791 	 */
792 	if ((vp->v_bioflag & VBIOONFREELIST) &&
793 	    vp->v_holdcnt == 0 && vp->v_usecount == 0) {
794 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
795 		TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist);
796 	}
797 	vp->v_holdcnt++;
798 }
799 
800 /* Lose interest in a vnode. */
801 void
802 vdrop(struct vnode *vp)
803 {
804 #ifdef DIAGNOSTIC
805 	if (vp->v_holdcnt == 0)
806 		panic("vdrop: zero holdcnt");
807 #endif
808 
809 	vp->v_holdcnt--;
810 
811 	/*
812 	 * If it is on the holdlist and the hold count drops to
813 	 * zero, move it to the free list.
814 	 */
815 	if ((vp->v_bioflag & VBIOONFREELIST) &&
816 	    vp->v_holdcnt == 0 && vp->v_usecount == 0) {
817 		TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist);
818 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
819 	}
820 }
821 
822 /*
823  * Remove any vnodes in the vnode table belonging to mount point mp.
824  *
825  * If MNT_NOFORCE is specified, there should not be any active ones,
826  * return error if any are found (nb: this is a user error, not a
827  * system error). If MNT_FORCE is specified, detach any active vnodes
828  * that are found.
829  */
830 #ifdef DEBUG
831 int busyprt = 0;	/* print out busy vnodes */
832 struct ctldebug debug1 = { "busyprt", &busyprt };
833 #endif
834 
835 int
836 vfs_mount_foreach_vnode(struct mount *mp,
837     int (*func)(struct vnode *, void *), void *arg) {
838 	struct vnode *vp, *nvp;
839 	int error = 0;
840 
841 loop:
842 	LIST_FOREACH_SAFE(vp , &mp->mnt_vnodelist, v_mntvnodes, nvp) {
843 		if (vp->v_mount != mp)
844 			goto loop;
845 
846 		error = func(vp, arg);
847 
848 		if (error != 0)
849 			break;
850 	}
851 
852 	return (error);
853 }
854 
855 struct vflush_args {
856 	struct vnode *skipvp;
857 	int busy;
858 	int flags;
859 };
860 
861 int
862 vflush_vnode(struct vnode *vp, void *arg)
863 {
864 	struct vflush_args *va = arg;
865 	struct proc *p = curproc;
866 
867 	if (vp == va->skipvp) {
868 		return (0);
869 	}
870 
871 	if ((va->flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
872 		return (0);
873 	}
874 
875 	/*
876 	 * If WRITECLOSE is set, only flush out regular file
877 	 * vnodes open for writing.
878 	 */
879 	if ((va->flags & WRITECLOSE) &&
880 	    (vp->v_writecount == 0 || vp->v_type != VREG)) {
881 		return (0);
882 	}
883 
884 	/*
885 	 * With v_usecount == 0, all we need to do is clear
886 	 * out the vnode data structures and we are done.
887 	 */
888 	if (vp->v_usecount == 0) {
889 		vgonel(vp, p);
890 		return (0);
891 	}
892 
893 	/*
894 	 * If FORCECLOSE is set, forcibly close the vnode.
895 	 * For block or character devices, revert to an
896 	 * anonymous device. For all other files, just kill them.
897 	 */
898 	if (va->flags & FORCECLOSE) {
899 		if (vp->v_type != VBLK && vp->v_type != VCHR) {
900 			vgonel(vp, p);
901 		} else {
902 			vclean(vp, 0, p);
903 			vp->v_op = &spec_vops;
904 			insmntque(vp, NULL);
905 		}
906 		return (0);
907 	}
908 
909 	/*
910 	 * If set, this is allowed to ignore vnodes which don't
911 	 * have changes pending to disk.
912 	 * XXX Might be nice to check per-fs "inode" flags, but
913 	 * generally the filesystem is sync'd already, right?
914 	 */
915 	if ((va->flags & IGNORECLEAN) &&
916 	    LIST_EMPTY(&vp->v_dirtyblkhd))
917 		return (0);
918 
919 #ifdef DEBUG
920 	if (busyprt)
921 		vprint("vflush: busy vnode", vp);
922 #endif
923 	va->busy++;
924 	return (0);
925 }
926 
927 int
928 vflush(struct mount *mp, struct vnode *skipvp, int flags)
929 {
930 	struct vflush_args va;
931 	va.skipvp = skipvp;
932 	va.busy = 0;
933 	va.flags = flags;
934 
935 	vfs_mount_foreach_vnode(mp, vflush_vnode, &va);
936 
937 	if (va.busy)
938 		return (EBUSY);
939 	return (0);
940 }
941 
942 /*
943  * Disassociate the underlying file system from a vnode.
944  */
945 void
946 vclean(struct vnode *vp, int flags, struct proc *p)
947 {
948 	int active;
949 
950 	/*
951 	 * Check to see if the vnode is in use.
952 	 * If so we have to reference it before we clean it out
953 	 * so that its count cannot fall to zero and generate a
954 	 * race against ourselves to recycle it.
955 	 */
956 	if ((active = vp->v_usecount) != 0)
957 		vp->v_usecount++;
958 
959 	/*
960 	 * Prevent the vnode from being recycled or
961 	 * brought into use while we clean it out.
962 	 */
963 	if (vp->v_flag & VXLOCK)
964 		panic("vclean: deadlock");
965 	vp->v_flag |= VXLOCK;
966 	/*
967 	 * Even if the count is zero, the VOP_INACTIVE routine may still
968 	 * have the object locked while it cleans it out. The VOP_LOCK
969 	 * ensures that the VOP_INACTIVE routine is done with its work.
970 	 * For active vnodes, it ensures that no other activity can
971 	 * occur while the underlying object is being cleaned out.
972 	 */
973 	VOP_LOCK(vp, LK_DRAIN | LK_EXCLUSIVE, p);
974 
975 	/*
976 	 * Clean out any VM data associated with the vnode.
977 	 */
978 	uvm_vnp_terminate(vp);
979 	/*
980 	 * Clean out any buffers associated with the vnode.
981 	 */
982 	if (flags & DOCLOSE)
983 		vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0);
984 	/*
985 	 * If purging an active vnode, it must be closed and
986 	 * deactivated before being reclaimed. Note that the
987 	 * VOP_INACTIVE will unlock the vnode
988 	 */
989 	if (active) {
990 		if (flags & DOCLOSE)
991 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, p);
992 		VOP_INACTIVE(vp, p);
993 	} else {
994 		/*
995 		 * Any other processes trying to obtain this lock must first
996 		 * wait for VXLOCK to clear, then call the new lock operation.
997 		 */
998 		VOP_UNLOCK(vp, p);
999 	}
1000 
1001 	/*
1002 	 * Reclaim the vnode.
1003 	 */
1004 	if (VOP_RECLAIM(vp, p))
1005 		panic("vclean: cannot reclaim");
1006 	if (active) {
1007 		vp->v_usecount--;
1008 		if (vp->v_usecount == 0) {
1009 			if (vp->v_holdcnt > 0)
1010 				panic("vclean: not clean");
1011 			vputonfreelist(vp);
1012 		}
1013 	}
1014 	cache_purge(vp);
1015 
1016 	/*
1017 	 * Done with purge, notify sleepers of the grim news.
1018 	 */
1019 	vp->v_op = &dead_vops;
1020 	VN_KNOTE(vp, NOTE_REVOKE);
1021 	vp->v_tag = VT_NON;
1022 	vp->v_flag &= ~VXLOCK;
1023 #ifdef VFSLCKDEBUG
1024 	vp->v_flag &= ~VLOCKSWORK;
1025 #endif
1026 	if (vp->v_flag & VXWANT) {
1027 		vp->v_flag &= ~VXWANT;
1028 		wakeup(vp);
1029 	}
1030 }
1031 
1032 /*
1033  * Recycle an unused vnode to the front of the free list.
1034  */
1035 int
1036 vrecycle(struct vnode *vp, struct proc *p)
1037 {
1038 	if (vp->v_usecount == 0) {
1039 		vgonel(vp, p);
1040 		return (1);
1041 	}
1042 	return (0);
1043 }
1044 
1045 /*
1046  * Eliminate all activity associated with a vnode
1047  * in preparation for reuse.
1048  */
1049 void
1050 vgone(struct vnode *vp)
1051 {
1052 	struct proc *p = curproc;
1053 	vgonel(vp, p);
1054 }
1055 
1056 /*
1057  * vgone, with struct proc.
1058  */
1059 void
1060 vgonel(struct vnode *vp, struct proc *p)
1061 {
1062 	struct vnode *vq;
1063 	struct vnode *vx;
1064 
1065 	/*
1066 	 * If a vgone (or vclean) is already in progress,
1067 	 * wait until it is done and return.
1068 	 */
1069 	if (vp->v_flag & VXLOCK) {
1070 		vp->v_flag |= VXWANT;
1071 		tsleep(vp, PINOD, "vgone", 0);
1072 		return;
1073 	}
1074 
1075 	/*
1076 	 * Clean out the filesystem specific data.
1077 	 */
1078 	vclean(vp, DOCLOSE, p);
1079 	/*
1080 	 * Delete from old mount point vnode list, if on one.
1081 	 */
1082 	if (vp->v_mount != NULL)
1083 		insmntque(vp, NULL);
1084 	/*
1085 	 * If special device, remove it from special device alias list
1086 	 * if it is on one.
1087 	 */
1088 	if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_specinfo != 0) {
1089 		if ((vp->v_flag & VALIASED) == 0 && vp->v_type == VCHR &&
1090 		    (cdevsw[major(vp->v_rdev)].d_flags & D_CLONE) &&
1091 		    (minor(vp->v_rdev) >> CLONE_SHIFT == 0)) {
1092 			free(vp->v_specbitmap, M_VNODE, CLONE_MAPSZ);
1093 		}
1094 		if (*vp->v_hashchain == vp) {
1095 			*vp->v_hashchain = vp->v_specnext;
1096 		} else {
1097 			for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1098 				if (vq->v_specnext != vp)
1099 					continue;
1100 				vq->v_specnext = vp->v_specnext;
1101 				break;
1102 			}
1103 			if (vq == NULL)
1104 				panic("missing bdev");
1105 		}
1106 		if (vp->v_flag & VALIASED) {
1107 			vx = NULL;
1108 			for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1109 				if (vq->v_rdev != vp->v_rdev ||
1110 				    vq->v_type != vp->v_type)
1111 					continue;
1112 				if (vx)
1113 					break;
1114 				vx = vq;
1115 			}
1116 			if (vx == NULL)
1117 				panic("missing alias");
1118 			if (vq == NULL)
1119 				vx->v_flag &= ~VALIASED;
1120 			vp->v_flag &= ~VALIASED;
1121 		}
1122 		free(vp->v_specinfo, M_VNODE, sizeof(struct specinfo));
1123 		vp->v_specinfo = NULL;
1124 	}
1125 	/*
1126 	 * If it is on the freelist and not already at the head,
1127 	 * move it to the head of the list.
1128 	 */
1129 	vp->v_type = VBAD;
1130 
1131 	/*
1132 	 * Move onto the free list, unless we were called from
1133 	 * getnewvnode and we're not on any free list
1134 	 */
1135 	if (vp->v_usecount == 0 &&
1136 	    (vp->v_bioflag & VBIOONFREELIST)) {
1137 		int s;
1138 
1139 		s = splbio();
1140 
1141 		if (vp->v_holdcnt > 0)
1142 			panic("vgonel: not clean");
1143 
1144 		if (TAILQ_FIRST(&vnode_free_list) != vp) {
1145 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1146 			TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
1147 		}
1148 		splx(s);
1149 	}
1150 }
1151 
1152 /*
1153  * Lookup a vnode by device number.
1154  */
1155 int
1156 vfinddev(dev_t dev, enum vtype type, struct vnode **vpp)
1157 {
1158 	struct vnode *vp;
1159 	int rc =0;
1160 
1161 	for (vp = speclisth[SPECHASH(dev)]; vp; vp = vp->v_specnext) {
1162 		if (dev != vp->v_rdev || type != vp->v_type)
1163 			continue;
1164 		*vpp = vp;
1165 		rc = 1;
1166 		break;
1167 	}
1168 	return (rc);
1169 }
1170 
1171 /*
1172  * Revoke all the vnodes corresponding to the specified minor number
1173  * range (endpoints inclusive) of the specified major.
1174  */
1175 void
1176 vdevgone(int maj, int minl, int minh, enum vtype type)
1177 {
1178 	struct vnode *vp;
1179 	int mn;
1180 
1181 	for (mn = minl; mn <= minh; mn++)
1182 		if (vfinddev(makedev(maj, mn), type, &vp))
1183 			VOP_REVOKE(vp, REVOKEALL);
1184 }
1185 
1186 /*
1187  * Calculate the total number of references to a special device.
1188  */
1189 int
1190 vcount(struct vnode *vp)
1191 {
1192 	struct vnode *vq, *vnext;
1193 	int count;
1194 
1195 loop:
1196 	if ((vp->v_flag & VALIASED) == 0)
1197 		return (vp->v_usecount);
1198 	for (count = 0, vq = *vp->v_hashchain; vq; vq = vnext) {
1199 		vnext = vq->v_specnext;
1200 		if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type)
1201 			continue;
1202 		/*
1203 		 * Alias, but not in use, so flush it out.
1204 		 */
1205 		if (vq->v_usecount == 0 && vq != vp) {
1206 			vgone(vq);
1207 			goto loop;
1208 		}
1209 		count += vq->v_usecount;
1210 	}
1211 	return (count);
1212 }
1213 
1214 #if defined(DEBUG) || defined(DIAGNOSTIC)
1215 /*
1216  * Print out a description of a vnode.
1217  */
1218 static char *typename[] =
1219    { "VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD" };
1220 
1221 void
1222 vprint(char *label, struct vnode *vp)
1223 {
1224 	char buf[64];
1225 
1226 	if (label != NULL)
1227 		printf("%s: ", label);
1228 	printf("%p, type %s, use %u, write %u, hold %u,",
1229 		vp, typename[vp->v_type], vp->v_usecount, vp->v_writecount,
1230 		vp->v_holdcnt);
1231 	buf[0] = '\0';
1232 	if (vp->v_flag & VROOT)
1233 		strlcat(buf, "|VROOT", sizeof buf);
1234 	if (vp->v_flag & VTEXT)
1235 		strlcat(buf, "|VTEXT", sizeof buf);
1236 	if (vp->v_flag & VSYSTEM)
1237 		strlcat(buf, "|VSYSTEM", sizeof buf);
1238 	if (vp->v_flag & VXLOCK)
1239 		strlcat(buf, "|VXLOCK", sizeof buf);
1240 	if (vp->v_flag & VXWANT)
1241 		strlcat(buf, "|VXWANT", sizeof buf);
1242 	if (vp->v_bioflag & VBIOWAIT)
1243 		strlcat(buf, "|VBIOWAIT", sizeof buf);
1244 	if (vp->v_bioflag & VBIOONFREELIST)
1245 		strlcat(buf, "|VBIOONFREELIST", sizeof buf);
1246 	if (vp->v_bioflag & VBIOONSYNCLIST)
1247 		strlcat(buf, "|VBIOONSYNCLIST", sizeof buf);
1248 	if (vp->v_flag & VALIASED)
1249 		strlcat(buf, "|VALIASED", sizeof buf);
1250 	if (buf[0] != '\0')
1251 		printf(" flags (%s)", &buf[1]);
1252 	if (vp->v_data == NULL) {
1253 		printf("\n");
1254 	} else {
1255 		printf("\n\t");
1256 		VOP_PRINT(vp);
1257 	}
1258 }
1259 #endif /* DEBUG || DIAGNOSTIC */
1260 
1261 #ifdef DEBUG
1262 /*
1263  * List all of the locked vnodes in the system.
1264  * Called when debugging the kernel.
1265  */
1266 void
1267 printlockedvnodes(void)
1268 {
1269 	struct mount *mp;
1270 	struct vnode *vp;
1271 
1272 	printf("Locked vnodes\n");
1273 
1274 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
1275 		if (vfs_busy(mp, VB_READ|VB_NOWAIT))
1276 			continue;
1277 		LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
1278 			if (VOP_ISLOCKED(vp))
1279 				vprint(NULL, vp);
1280 		}
1281 		vfs_unbusy(mp);
1282  	}
1283 
1284 }
1285 #endif
1286 
1287 /*
1288  * Top level filesystem related information gathering.
1289  */
1290 int
1291 vfs_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp,
1292     size_t newlen, struct proc *p)
1293 {
1294 	struct vfsconf *vfsp, *tmpvfsp;
1295 	int ret;
1296 
1297 	/* all sysctl names at this level are at least name and field */
1298 	if (namelen < 2)
1299 		return (ENOTDIR);		/* overloaded */
1300 
1301 	if (name[0] != VFS_GENERIC) {
1302 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
1303 			if (vfsp->vfc_typenum == name[0])
1304 				break;
1305 
1306 		if (vfsp == NULL || vfsp->vfc_vfsops->vfs_sysctl == NULL)
1307 			return (EOPNOTSUPP);
1308 
1309 		return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
1310 		    oldp, oldlenp, newp, newlen, p));
1311 	}
1312 
1313 	switch (name[1]) {
1314 	case VFS_MAXTYPENUM:
1315 		return (sysctl_rdint(oldp, oldlenp, newp, maxvfsconf));
1316 
1317 	case VFS_CONF:
1318 		if (namelen < 3)
1319 			return (ENOTDIR);	/* overloaded */
1320 
1321 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
1322 			if (vfsp->vfc_typenum == name[2])
1323 				break;
1324 
1325 		if (vfsp == NULL)
1326 			return (EOPNOTSUPP);
1327 
1328 		/* Make a copy, clear out kernel pointers */
1329 		tmpvfsp = malloc(sizeof(*tmpvfsp), M_TEMP, M_WAITOK|M_ZERO);
1330 		memcpy(tmpvfsp, vfsp, sizeof(*tmpvfsp));
1331 		tmpvfsp->vfc_vfsops = NULL;
1332 		tmpvfsp->vfc_next = NULL;
1333 
1334 		ret = sysctl_rdstruct(oldp, oldlenp, newp, tmpvfsp,
1335 		    sizeof(struct vfsconf));
1336 
1337 		free(tmpvfsp, M_TEMP, sizeof(*tmpvfsp));
1338 		return (ret);
1339 	case VFS_BCACHESTAT:	/* buffer cache statistics */
1340 		ret = sysctl_rdstruct(oldp, oldlenp, newp, &bcstats,
1341 		    sizeof(struct bcachestats));
1342 		return(ret);
1343 	}
1344 	return (EOPNOTSUPP);
1345 }
1346 
1347 /*
1348  * Check to see if a filesystem is mounted on a block device.
1349  */
1350 int
1351 vfs_mountedon(struct vnode *vp)
1352 {
1353 	struct vnode *vq;
1354 	int error = 0;
1355 
1356  	if (vp->v_specmountpoint != NULL)
1357 		return (EBUSY);
1358 	if (vp->v_flag & VALIASED) {
1359 		for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
1360 			if (vq->v_rdev != vp->v_rdev ||
1361 			    vq->v_type != vp->v_type)
1362 				continue;
1363 			if (vq->v_specmountpoint != NULL) {
1364 				error = EBUSY;
1365 				break;
1366 			}
1367  		}
1368 	}
1369 	return (error);
1370 }
1371 
1372 #ifdef NFSSERVER
1373 /*
1374  * Build hash lists of net addresses and hang them off the mount point.
1375  * Called by vfs_export() to set up the lists of export addresses.
1376  */
1377 int
1378 vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
1379     struct export_args *argp)
1380 {
1381 	struct netcred *np;
1382 	struct radix_node_head *rnh;
1383 	int nplen, i;
1384 	struct radix_node *rn;
1385 	struct sockaddr *saddr, *smask = 0;
1386 	int error;
1387 
1388 	if (argp->ex_addrlen == 0) {
1389 		if (mp->mnt_flag & MNT_DEFEXPORTED)
1390 			return (EPERM);
1391 		np = &nep->ne_defexported;
1392 		/* fill in the kernel's ucred from userspace's xucred */
1393 		if ((error = crfromxucred(&np->netc_anon, &argp->ex_anon)))
1394 			return (error);
1395 		mp->mnt_flag |= MNT_DEFEXPORTED;
1396 		goto finish;
1397 	}
1398 	if (argp->ex_addrlen > MLEN || argp->ex_masklen > MLEN ||
1399 	    argp->ex_addrlen < 0 || argp->ex_masklen < 0)
1400 		return (EINVAL);
1401 	nplen = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
1402 	np = (struct netcred *)malloc(nplen, M_NETADDR, M_WAITOK|M_ZERO);
1403 	saddr = (struct sockaddr *)(np + 1);
1404 	error = copyin(argp->ex_addr, saddr, argp->ex_addrlen);
1405 	if (error)
1406 		goto out;
1407 	if (saddr->sa_len > argp->ex_addrlen)
1408 		saddr->sa_len = argp->ex_addrlen;
1409 	if (argp->ex_masklen) {
1410 		smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
1411 		error = copyin(argp->ex_mask, smask, argp->ex_masklen);
1412 		if (error)
1413 			goto out;
1414 		if (smask->sa_len > argp->ex_masklen)
1415 			smask->sa_len = argp->ex_masklen;
1416 	}
1417 	/* fill in the kernel's ucred from userspace's xucred */
1418 	if ((error = crfromxucred(&np->netc_anon, &argp->ex_anon)))
1419 		goto out;
1420 	i = saddr->sa_family;
1421 	switch (i) {
1422 	case AF_INET:
1423 		if ((rnh = nep->ne_rtable_inet) == NULL) {
1424 			if (!rn_inithead((void **)&nep->ne_rtable_inet,
1425 			    offsetof(struct sockaddr_in, sin_addr))) {
1426 				error = ENOBUFS;
1427 				goto out;
1428 			}
1429 			rnh = nep->ne_rtable_inet;
1430 		}
1431 		break;
1432 	default:
1433 		error = EINVAL;
1434 		goto out;
1435 	}
1436 	rn = rn_addroute(saddr, smask, rnh, np->netc_rnodes, 0);
1437 	if (rn == 0 || np != (struct netcred *)rn) { /* already exists */
1438 		error = EPERM;
1439 		goto out;
1440 	}
1441 finish:
1442 	np->netc_exflags = argp->ex_flags;
1443 	return (0);
1444 out:
1445 	free(np, M_NETADDR, nplen);
1446 	return (error);
1447 }
1448 
1449 int
1450 vfs_free_netcred(struct radix_node *rn, void *w, u_int id)
1451 {
1452 	struct radix_node_head *rnh = (struct radix_node_head *)w;
1453 
1454 	rn_delete(rn->rn_key, rn->rn_mask, rnh, NULL);
1455 	free(rn, M_NETADDR, 0);
1456 	return (0);
1457 }
1458 
1459 /*
1460  * Free the net address hash lists that are hanging off the mount points.
1461  */
1462 void
1463 vfs_free_addrlist(struct netexport *nep)
1464 {
1465 	struct radix_node_head *rnh;
1466 
1467 	if ((rnh = nep->ne_rtable_inet) != NULL) {
1468 		rn_walktree(rnh, vfs_free_netcred, rnh);
1469 		free(rnh, M_RTABLE, 0);
1470 		nep->ne_rtable_inet = NULL;
1471 	}
1472 }
1473 #endif /* NFSSERVER */
1474 
1475 int
1476 vfs_export(struct mount *mp, struct netexport *nep, struct export_args *argp)
1477 {
1478 #ifdef NFSSERVER
1479 	int error;
1480 
1481 	if (argp->ex_flags & MNT_DELEXPORT) {
1482 		vfs_free_addrlist(nep);
1483 		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
1484 	}
1485 	if (argp->ex_flags & MNT_EXPORTED) {
1486 		if ((error = vfs_hang_addrlist(mp, nep, argp)) != 0)
1487 			return (error);
1488 		mp->mnt_flag |= MNT_EXPORTED;
1489 	}
1490 	return (0);
1491 #else
1492 	return (ENOTSUP);
1493 #endif /* NFSSERVER */
1494 }
1495 
1496 struct netcred *
1497 vfs_export_lookup(struct mount *mp, struct netexport *nep, struct mbuf *nam)
1498 {
1499 #ifdef NFSSERVER
1500 	struct netcred *np;
1501 	struct radix_node_head *rnh;
1502 	struct sockaddr *saddr;
1503 
1504 	np = NULL;
1505 	if (mp->mnt_flag & MNT_EXPORTED) {
1506 		/*
1507 		 * Lookup in the export list first.
1508 		 */
1509 		if (nam != NULL) {
1510 			saddr = mtod(nam, struct sockaddr *);
1511 			switch(saddr->sa_family) {
1512 			case AF_INET:
1513 				rnh = nep->ne_rtable_inet;
1514 				break;
1515 			default:
1516 				rnh = NULL;
1517 				break;
1518 			}
1519 			if (rnh != NULL)
1520 				np = (struct netcred *)rn_match(saddr, rnh);
1521 		}
1522 		/*
1523 		 * If no address match, use the default if it exists.
1524 		 */
1525 		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
1526 			np = &nep->ne_defexported;
1527 	}
1528 	return (np);
1529 #else
1530 	return (NULL);
1531 #endif /* NFSSERVER */
1532 }
1533 
1534 /*
1535  * Do the usual access checking.
1536  * file_mode, uid and gid are from the vnode in question,
1537  * while acc_mode and cred are from the VOP_ACCESS parameter list
1538  */
1539 int
1540 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid,
1541     mode_t acc_mode, struct ucred *cred)
1542 {
1543 	mode_t mask;
1544 
1545 	/* User id 0 always gets read/write access. */
1546 	if (cred->cr_uid == 0) {
1547 		/* For VEXEC, at least one of the execute bits must be set. */
1548 		if ((acc_mode & VEXEC) && type != VDIR &&
1549 		    (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)
1550 			return EACCES;
1551 		return 0;
1552 	}
1553 
1554 	mask = 0;
1555 
1556 	/* Otherwise, check the owner. */
1557 	if (cred->cr_uid == uid) {
1558 		if (acc_mode & VEXEC)
1559 			mask |= S_IXUSR;
1560 		if (acc_mode & VREAD)
1561 			mask |= S_IRUSR;
1562 		if (acc_mode & VWRITE)
1563 			mask |= S_IWUSR;
1564 		return (file_mode & mask) == mask ? 0 : EACCES;
1565 	}
1566 
1567 	/* Otherwise, check the groups. */
1568 	if (groupmember(gid, cred)) {
1569 		if (acc_mode & VEXEC)
1570 			mask |= S_IXGRP;
1571 		if (acc_mode & VREAD)
1572 			mask |= S_IRGRP;
1573 		if (acc_mode & VWRITE)
1574 			mask |= S_IWGRP;
1575 		return (file_mode & mask) == mask ? 0 : EACCES;
1576 	}
1577 
1578 	/* Otherwise, check everyone else. */
1579 	if (acc_mode & VEXEC)
1580 		mask |= S_IXOTH;
1581 	if (acc_mode & VREAD)
1582 		mask |= S_IROTH;
1583 	if (acc_mode & VWRITE)
1584 		mask |= S_IWOTH;
1585 	return (file_mode & mask) == mask ? 0 : EACCES;
1586 }
1587 
1588 struct rwlock vfs_stall_lock = RWLOCK_INITIALIZER("vfs_stall");
1589 
1590 int
1591 vfs_stall(struct proc *p, int stall)
1592 {
1593 	struct mount *mp, *nmp;
1594 	int allerror = 0, error;
1595 
1596 	if (stall)
1597 		rw_enter_write(&vfs_stall_lock);
1598 
1599 	TAILQ_FOREACH_REVERSE_SAFE(mp, &mountlist, mntlist, mnt_list, nmp) {
1600 		if (stall) {
1601 			error = vfs_busy(mp, VB_WRITE|VB_WAIT);
1602 			if (error) {
1603 				printf("%s: busy\n", mp->mnt_stat.f_mntonname);
1604 				allerror = error;
1605 				continue;
1606 			}
1607 			uvm_vnp_sync(mp);
1608 			error = VFS_SYNC(mp, MNT_WAIT, stall, p->p_ucred, p);
1609 			if (error) {
1610 				printf("%s: failed to sync\n", mp->mnt_stat.f_mntonname);
1611 				vfs_unbusy(mp);
1612 				allerror = error;
1613 				continue;
1614 			}
1615 			mp->mnt_flag |= MNT_STALLED;
1616 		} else {
1617 			if (mp->mnt_flag & MNT_STALLED) {
1618 				vfs_unbusy(mp);
1619 				mp->mnt_flag &= ~MNT_STALLED;
1620 			}
1621 		}
1622 	}
1623 
1624 	if (!stall)
1625 		rw_exit_write(&vfs_stall_lock);
1626 
1627 	return (allerror);
1628 }
1629 
1630 /*
1631  * Unmount all file systems.
1632  * We traverse the list in reverse order under the assumption that doing so
1633  * will avoid needing to worry about dependencies.
1634  */
1635 void
1636 vfs_unmountall(void)
1637 {
1638 	struct mount *mp, *nmp;
1639 	int allerror, error, again = 1;
1640 
1641  retry:
1642 	allerror = 0;
1643 	TAILQ_FOREACH_REVERSE_SAFE(mp, &mountlist, mntlist, mnt_list, nmp) {
1644 		if (vfs_busy(mp, VB_WRITE|VB_NOWAIT))
1645 			continue;
1646 		/* XXX Here is a race, the next pointer is not locked. */
1647 		if ((error = dounmount(mp, MNT_FORCE, curproc)) != 0) {
1648 			printf("unmount of %s failed with error %d\n",
1649 			    mp->mnt_stat.f_mntonname, error);
1650 			allerror = 1;
1651 		}
1652 	}
1653 
1654 	if (allerror) {
1655 		printf("WARNING: some file systems would not unmount\n");
1656 		if (again) {
1657 			printf("retrying\n");
1658 			again = 0;
1659 			goto retry;
1660 		}
1661 	}
1662 }
1663 
1664 /*
1665  * Sync and unmount file systems before shutting down.
1666  */
1667 void
1668 vfs_shutdown(struct proc *p)
1669 {
1670 #ifdef ACCOUNTING
1671 	acct_shutdown();
1672 #endif
1673 
1674 	printf("syncing disks... ");
1675 
1676 	if (panicstr == 0) {
1677 		/* Sync before unmount, in case we hang on something. */
1678 		sys_sync(p, NULL, NULL);
1679 		vfs_unmountall();
1680 	}
1681 
1682 #if NSOFTRAID > 0
1683 	sr_quiesce();
1684 #endif
1685 
1686 	if (vfs_syncwait(p, 1))
1687 		printf("giving up\n");
1688 	else
1689 		printf("done\n");
1690 }
1691 
1692 /*
1693  * perform sync() operation and wait for buffers to flush.
1694  */
1695 int
1696 vfs_syncwait(struct proc *p, int verbose)
1697 {
1698 	struct buf *bp;
1699 	int iter, nbusy, dcount, s;
1700 #ifdef MULTIPROCESSOR
1701 	int hold_count;
1702 #endif
1703 
1704 	sys_sync(p, NULL, NULL);
1705 
1706 	/* Wait for sync to finish. */
1707 	dcount = 10000;
1708 	for (iter = 0; iter < 20; iter++) {
1709 		nbusy = 0;
1710 		LIST_FOREACH(bp, &bufhead, b_list) {
1711 			if ((bp->b_flags & (B_BUSY|B_INVAL|B_READ)) == B_BUSY)
1712 				nbusy++;
1713 			/*
1714 			 * With soft updates, some buffers that are
1715 			 * written will be remarked as dirty until other
1716 			 * buffers are written.
1717 			 */
1718 			if (bp->b_flags & B_DELWRI) {
1719 				s = splbio();
1720 				bremfree(bp);
1721 				buf_acquire(bp);
1722 				splx(s);
1723 				nbusy++;
1724 				bawrite(bp);
1725 				if (dcount-- <= 0) {
1726 					if (verbose)
1727 						printf("softdep ");
1728 					return 1;
1729 				}
1730 			}
1731 		}
1732 		if (nbusy == 0)
1733 			break;
1734 		if (verbose)
1735 			printf("%d ", nbusy);
1736 #ifdef MULTIPROCESSOR
1737 		if (_kernel_lock_held())
1738 			hold_count = __mp_release_all(&kernel_lock);
1739 		else
1740 			hold_count = 0;
1741 #endif
1742 		DELAY(40000 * iter);
1743 #ifdef MULTIPROCESSOR
1744 		if (hold_count)
1745 			__mp_acquire_count(&kernel_lock, hold_count);
1746 #endif
1747 	}
1748 
1749 	return nbusy;
1750 }
1751 
1752 /*
1753  * posix file system related system variables.
1754  */
1755 int
1756 fs_posix_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp,
1757     void *newp, size_t newlen, struct proc *p)
1758 {
1759 	/* all sysctl names at this level are terminal */
1760 	if (namelen != 1)
1761 		return (ENOTDIR);
1762 
1763 	switch (name[0]) {
1764 	case FS_POSIX_SETUID:
1765 		if (newp && securelevel > 0)
1766 			return (EPERM);
1767 		return(sysctl_int(oldp, oldlenp, newp, newlen, &suid_clear));
1768 	default:
1769 		return (EOPNOTSUPP);
1770 	}
1771 	/* NOTREACHED */
1772 }
1773 
1774 /*
1775  * file system related system variables.
1776  */
1777 int
1778 fs_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp,
1779     size_t newlen, struct proc *p)
1780 {
1781 	sysctlfn *fn;
1782 
1783 	switch (name[0]) {
1784 	case FS_POSIX:
1785 		fn = fs_posix_sysctl;
1786 		break;
1787 	default:
1788 		return (EOPNOTSUPP);
1789 	}
1790 	return (*fn)(name + 1, namelen - 1, oldp, oldlenp, newp, newlen, p);
1791 }
1792 
1793 
1794 /*
1795  * Routines dealing with vnodes and buffers
1796  */
1797 
1798 /*
1799  * Wait for all outstanding I/Os to complete
1800  *
1801  * Manipulates v_numoutput. Must be called at splbio()
1802  */
1803 int
1804 vwaitforio(struct vnode *vp, int slpflag, char *wmesg, int timeo)
1805 {
1806 	int error = 0;
1807 
1808 	splassert(IPL_BIO);
1809 
1810 	while (vp->v_numoutput) {
1811 		vp->v_bioflag |= VBIOWAIT;
1812 		error = tsleep(&vp->v_numoutput,
1813 		    slpflag | (PRIBIO + 1), wmesg, timeo);
1814 		if (error)
1815 			break;
1816 	}
1817 
1818 	return (error);
1819 }
1820 
1821 /*
1822  * Update outstanding I/O count and do wakeup if requested.
1823  *
1824  * Manipulates v_numoutput. Must be called at splbio()
1825  */
1826 void
1827 vwakeup(struct vnode *vp)
1828 {
1829 	splassert(IPL_BIO);
1830 
1831 	if (vp != NULL) {
1832 		if (vp->v_numoutput-- == 0)
1833 			panic("vwakeup: neg numoutput");
1834 		if ((vp->v_bioflag & VBIOWAIT) && vp->v_numoutput == 0) {
1835 			vp->v_bioflag &= ~VBIOWAIT;
1836 			wakeup(&vp->v_numoutput);
1837 		}
1838 	}
1839 }
1840 
1841 /*
1842  * Flush out and invalidate all buffers associated with a vnode.
1843  * Called with the underlying object locked.
1844  */
1845 int
1846 vinvalbuf(struct vnode *vp, int flags, struct ucred *cred, struct proc *p,
1847     int slpflag, int slptimeo)
1848 {
1849 	struct buf *bp;
1850 	struct buf *nbp, *blist;
1851 	int s, error;
1852 
1853 #ifdef VFSLCKDEBUG
1854 	if ((vp->v_flag & VLOCKSWORK) && !VOP_ISLOCKED(vp))
1855 		panic("vinvalbuf(): vp isn't locked");
1856 #endif
1857 
1858 	if (flags & V_SAVE) {
1859 		s = splbio();
1860 		vwaitforio(vp, 0, "vinvalbuf", 0);
1861 		if (!LIST_EMPTY(&vp->v_dirtyblkhd)) {
1862 			splx(s);
1863 			if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, p)) != 0)
1864 				return (error);
1865 			s = splbio();
1866 			if (vp->v_numoutput > 0 ||
1867 			    !LIST_EMPTY(&vp->v_dirtyblkhd))
1868 				panic("vinvalbuf: dirty bufs");
1869 		}
1870 		splx(s);
1871 	}
1872 loop:
1873 	s = splbio();
1874 	for (;;) {
1875 		if ((blist = LIST_FIRST(&vp->v_cleanblkhd)) &&
1876 		    (flags & V_SAVEMETA))
1877 			while (blist && blist->b_lblkno < 0)
1878 				blist = LIST_NEXT(blist, b_vnbufs);
1879 		if (blist == NULL &&
1880 		    (blist = LIST_FIRST(&vp->v_dirtyblkhd)) &&
1881 		    (flags & V_SAVEMETA))
1882 			while (blist && blist->b_lblkno < 0)
1883 				blist = LIST_NEXT(blist, b_vnbufs);
1884 		if (!blist)
1885 			break;
1886 
1887 		for (bp = blist; bp; bp = nbp) {
1888 			nbp = LIST_NEXT(bp, b_vnbufs);
1889 			if (flags & V_SAVEMETA && bp->b_lblkno < 0)
1890 				continue;
1891 			if (bp->b_flags & B_BUSY) {
1892 				bp->b_flags |= B_WANTED;
1893 				error = tsleep(bp, slpflag | (PRIBIO + 1),
1894 				    "vinvalbuf", slptimeo);
1895 				if (error) {
1896 					splx(s);
1897 					return (error);
1898 				}
1899 				break;
1900 			}
1901 			bremfree(bp);
1902 			/*
1903 			 * XXX Since there are no node locks for NFS, I believe
1904 			 * there is a slight chance that a delayed write will
1905 			 * occur while sleeping just above, so check for it.
1906 			 */
1907 			if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) {
1908 				buf_acquire(bp);
1909 				splx(s);
1910 				(void) VOP_BWRITE(bp);
1911 				goto loop;
1912 			}
1913 			buf_acquire_nomap(bp);
1914 			bp->b_flags |= B_INVAL;
1915 			brelse(bp);
1916 		}
1917 	}
1918 	if (!(flags & V_SAVEMETA) &&
1919 	    (!LIST_EMPTY(&vp->v_dirtyblkhd) || !LIST_EMPTY(&vp->v_cleanblkhd)))
1920 		panic("vinvalbuf: flush failed");
1921 	splx(s);
1922 	return (0);
1923 }
1924 
1925 void
1926 vflushbuf(struct vnode *vp, int sync)
1927 {
1928 	struct buf *bp, *nbp;
1929 	int s;
1930 
1931 loop:
1932 	s = splbio();
1933 	LIST_FOREACH_SAFE(bp, &vp->v_dirtyblkhd, b_vnbufs, nbp) {
1934 		if ((bp->b_flags & B_BUSY))
1935 			continue;
1936 		if ((bp->b_flags & B_DELWRI) == 0)
1937 			panic("vflushbuf: not dirty");
1938 		bremfree(bp);
1939 		buf_acquire(bp);
1940 		splx(s);
1941 		/*
1942 		 * Wait for I/O associated with indirect blocks to complete,
1943 		 * since there is no way to quickly wait for them below.
1944 		 */
1945 		if (bp->b_vp == vp || sync == 0)
1946 			(void) bawrite(bp);
1947 		else
1948 			(void) bwrite(bp);
1949 		goto loop;
1950 	}
1951 	if (sync == 0) {
1952 		splx(s);
1953 		return;
1954 	}
1955 	vwaitforio(vp, 0, "vflushbuf", 0);
1956 	if (!LIST_EMPTY(&vp->v_dirtyblkhd)) {
1957 		splx(s);
1958 #ifdef DIAGNOSTIC
1959 		vprint("vflushbuf: dirty", vp);
1960 #endif
1961 		goto loop;
1962 	}
1963 	splx(s);
1964 }
1965 
1966 /*
1967  * Associate a buffer with a vnode.
1968  *
1969  * Manipulates buffer vnode queues. Must be called at splbio().
1970  */
1971 void
1972 bgetvp(struct vnode *vp, struct buf *bp)
1973 {
1974 	splassert(IPL_BIO);
1975 
1976 
1977 	if (bp->b_vp)
1978 		panic("bgetvp: not free");
1979 	vhold(vp);
1980 	bp->b_vp = vp;
1981 	if (vp->v_type == VBLK || vp->v_type == VCHR)
1982 		bp->b_dev = vp->v_rdev;
1983 	else
1984 		bp->b_dev = NODEV;
1985 	/*
1986 	 * Insert onto list for new vnode.
1987 	 */
1988 	bufinsvn(bp, &vp->v_cleanblkhd);
1989 }
1990 
1991 /*
1992  * Disassociate a buffer from a vnode.
1993  *
1994  * Manipulates vnode buffer queues. Must be called at splbio().
1995  */
1996 void
1997 brelvp(struct buf *bp)
1998 {
1999 	struct vnode *vp;
2000 
2001 	splassert(IPL_BIO);
2002 
2003 	if ((vp = bp->b_vp) == (struct vnode *) 0)
2004 		panic("brelvp: NULL");
2005 	/*
2006 	 * Delete from old vnode list, if on one.
2007 	 */
2008 	if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
2009 		bufremvn(bp);
2010 	if ((vp->v_bioflag & VBIOONSYNCLIST) &&
2011 	    LIST_FIRST(&vp->v_dirtyblkhd) == NULL) {
2012 		vp->v_bioflag &= ~VBIOONSYNCLIST;
2013 		LIST_REMOVE(vp, v_synclist);
2014 	}
2015 	bp->b_vp = NULL;
2016 
2017 	vdrop(vp);
2018 }
2019 
2020 /*
2021  * Replaces the current vnode associated with the buffer, if any,
2022  * with a new vnode.
2023  *
2024  * If an output I/O is pending on the buffer, the old vnode
2025  * I/O count is adjusted.
2026  *
2027  * Ignores vnode buffer queues. Must be called at splbio().
2028  */
2029 void
2030 buf_replacevnode(struct buf *bp, struct vnode *newvp)
2031 {
2032 	struct vnode *oldvp = bp->b_vp;
2033 
2034 	splassert(IPL_BIO);
2035 
2036 	if (oldvp)
2037 		brelvp(bp);
2038 
2039 	if ((bp->b_flags & (B_READ | B_DONE)) == 0) {
2040 		newvp->v_numoutput++;	/* put it on swapdev */
2041 		vwakeup(oldvp);
2042 	}
2043 
2044 	bgetvp(newvp, bp);
2045 	bufremvn(bp);
2046 }
2047 
2048 /*
2049  * Used to assign buffers to the appropriate clean or dirty list on
2050  * the vnode and to add newly dirty vnodes to the appropriate
2051  * filesystem syncer list.
2052  *
2053  * Manipulates vnode buffer queues. Must be called at splbio().
2054  */
2055 void
2056 reassignbuf(struct buf *bp)
2057 {
2058 	struct buflists *listheadp;
2059 	int delay;
2060 	struct vnode *vp = bp->b_vp;
2061 
2062 	splassert(IPL_BIO);
2063 
2064 	/*
2065 	 * Delete from old vnode list, if on one.
2066 	 */
2067 	if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
2068 		bufremvn(bp);
2069 
2070 	/*
2071 	 * If dirty, put on list of dirty buffers;
2072 	 * otherwise insert onto list of clean buffers.
2073 	 */
2074 	if ((bp->b_flags & B_DELWRI) == 0) {
2075 		listheadp = &vp->v_cleanblkhd;
2076 		if ((vp->v_bioflag & VBIOONSYNCLIST) &&
2077 		    LIST_FIRST(&vp->v_dirtyblkhd) == NULL) {
2078 			vp->v_bioflag &= ~VBIOONSYNCLIST;
2079 			LIST_REMOVE(vp, v_synclist);
2080 		}
2081 	} else {
2082 		listheadp = &vp->v_dirtyblkhd;
2083 		if ((vp->v_bioflag & VBIOONSYNCLIST) == 0) {
2084 			switch (vp->v_type) {
2085 			case VDIR:
2086 				delay = syncdelay / 2;
2087 				break;
2088 			case VBLK:
2089 				if (vp->v_specmountpoint != NULL) {
2090 					delay = syncdelay / 3;
2091 					break;
2092 				}
2093 				/* FALLTHROUGH */
2094 			default:
2095 				delay = syncdelay;
2096 			}
2097 			vn_syncer_add_to_worklist(vp, delay);
2098 		}
2099 	}
2100 	bufinsvn(bp, listheadp);
2101 }
2102 
2103 int
2104 vfs_register(struct vfsconf *vfs)
2105 {
2106 	struct vfsconf *vfsp;
2107 	struct vfsconf **vfspp;
2108 
2109 #ifdef DIAGNOSTIC
2110 	/* Paranoia? */
2111 	if (vfs->vfc_refcount != 0)
2112 		printf("vfs_register called with vfc_refcount > 0\n");
2113 #endif
2114 
2115 	/* Check if filesystem already known */
2116 	for (vfspp = &vfsconf, vfsp = vfsconf; vfsp;
2117 	    vfspp = &vfsp->vfc_next, vfsp = vfsp->vfc_next)
2118 		if (strcmp(vfsp->vfc_name, vfs->vfc_name) == 0)
2119 			return (EEXIST);
2120 
2121 	if (vfs->vfc_typenum > maxvfsconf)
2122 		maxvfsconf = vfs->vfc_typenum;
2123 
2124 	vfs->vfc_next = NULL;
2125 
2126 	/* Add to the end of the list */
2127 	*vfspp = vfs;
2128 
2129 	/* Call vfs_init() */
2130 	if (vfs->vfc_vfsops->vfs_init)
2131 		(*(vfs->vfc_vfsops->vfs_init))(vfs);
2132 
2133 	return 0;
2134 }
2135 
2136 int
2137 vfs_unregister(struct vfsconf *vfs)
2138 {
2139 	struct vfsconf *vfsp;
2140 	struct vfsconf **vfspp;
2141 	int maxtypenum;
2142 
2143 	/* Find our vfsconf struct */
2144 	for (vfspp = &vfsconf, vfsp = vfsconf; vfsp;
2145 	    vfspp = &vfsp->vfc_next, vfsp = vfsp->vfc_next) {
2146 		if (strcmp(vfsp->vfc_name, vfs->vfc_name) == 0)
2147 			break;
2148 	}
2149 
2150 	if (!vfsp)			/* Not found */
2151 		return (ENOENT);
2152 
2153 	if (vfsp->vfc_refcount)		/* In use */
2154 		return (EBUSY);
2155 
2156 	/* Remove from list and free */
2157 	*vfspp = vfsp->vfc_next;
2158 
2159 	maxtypenum = 0;
2160 
2161 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2162 		if (vfsp->vfc_typenum > maxtypenum)
2163 			maxtypenum = vfsp->vfc_typenum;
2164 
2165 	maxvfsconf = maxtypenum;
2166 	return 0;
2167 }
2168 
2169 /*
2170  * Check if vnode represents a disk device
2171  */
2172 int
2173 vn_isdisk(struct vnode *vp, int *errp)
2174 {
2175 	if (vp->v_type != VBLK && vp->v_type != VCHR)
2176 		return (0);
2177 
2178 	return (1);
2179 }
2180 
2181 #ifdef DDB
2182 #include <machine/db_machdep.h>
2183 #include <ddb/db_interface.h>
2184 
2185 void
2186 vfs_buf_print(void *b, int full,
2187     int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2188 {
2189 	struct buf *bp = b;
2190 
2191 	(*pr)("  vp %p lblkno 0x%llx blkno 0x%llx dev 0x%x\n"
2192 	      "  proc %p error %d flags %lb\n",
2193 	    bp->b_vp, (int64_t)bp->b_lblkno, (int64_t)bp->b_blkno, bp->b_dev,
2194 	    bp->b_proc, bp->b_error, bp->b_flags, B_BITS);
2195 
2196 	(*pr)("  bufsize 0x%lx bcount 0x%lx resid 0x%lx\n"
2197 	      "  data %p saveaddr %p dep %p iodone %p\n",
2198 	    bp->b_bufsize, bp->b_bcount, (long)bp->b_resid,
2199 	    bp->b_data, bp->b_saveaddr,
2200 	    LIST_FIRST(&bp->b_dep), bp->b_iodone);
2201 
2202 	(*pr)("  dirty {off 0x%x end 0x%x} valid {off 0x%x end 0x%x}\n",
2203 	    bp->b_dirtyoff, bp->b_dirtyend, bp->b_validoff, bp->b_validend);
2204 
2205 #ifdef FFS_SOFTUPDATES
2206 	if (full)
2207 		softdep_print(bp, full, pr);
2208 #endif
2209 }
2210 
2211 const char *vtypes[] = { VTYPE_NAMES };
2212 const char *vtags[] = { VTAG_NAMES };
2213 
2214 void
2215 vfs_vnode_print(void *v, int full,
2216     int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2217 {
2218 	struct vnode *vp = v;
2219 
2220 	(*pr)("tag %s(%d) type %s(%d) mount %p typedata %p\n",
2221 	      (u_int)vp->v_tag >= nitems(vtags)? "<unk>":vtags[vp->v_tag],
2222 	      vp->v_tag,
2223 	      (u_int)vp->v_type >= nitems(vtypes)? "<unk>":vtypes[vp->v_type],
2224 	      vp->v_type, vp->v_mount, vp->v_mountedhere);
2225 
2226 	(*pr)("data %p usecount %d writecount %d holdcnt %d numoutput %d\n",
2227 	      vp->v_data, vp->v_usecount, vp->v_writecount,
2228 	      vp->v_holdcnt, vp->v_numoutput);
2229 
2230 	/* uvm_object_printit(&vp->v_uobj, full, pr); */
2231 
2232 	if (full) {
2233 		struct buf *bp;
2234 
2235 		(*pr)("clean bufs:\n");
2236 		LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) {
2237 			(*pr)(" bp %p\n", bp);
2238 			vfs_buf_print(bp, full, pr);
2239 		}
2240 
2241 		(*pr)("dirty bufs:\n");
2242 		LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) {
2243 			(*pr)(" bp %p\n", bp);
2244 			vfs_buf_print(bp, full, pr);
2245 		}
2246 	}
2247 }
2248 
2249 void
2250 vfs_mount_print(struct mount *mp, int full,
2251     int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2252 {
2253 	struct vfsconf *vfc = mp->mnt_vfc;
2254 	struct vnode *vp;
2255 	int cnt;
2256 
2257 	(*pr)("flags %b\nvnodecovered %p syncer %p data %p\n",
2258 	    mp->mnt_flag, MNT_BITS,
2259 	    mp->mnt_vnodecovered, mp->mnt_syncer, mp->mnt_data);
2260 
2261 	(*pr)("vfsconf: ops %p name \"%s\" num %d ref %d flags 0x%x\n",
2262             vfc->vfc_vfsops, vfc->vfc_name, vfc->vfc_typenum,
2263 	    vfc->vfc_refcount, vfc->vfc_flags);
2264 
2265 	(*pr)("statvfs cache: bsize %x iosize %x\nblocks %llu free %llu avail %lld\n",
2266 	    mp->mnt_stat.f_bsize, mp->mnt_stat.f_iosize, mp->mnt_stat.f_blocks,
2267 	    mp->mnt_stat.f_bfree, mp->mnt_stat.f_bavail);
2268 
2269 	(*pr)("  files %llu ffiles %llu favail %lld\n", mp->mnt_stat.f_files,
2270 	    mp->mnt_stat.f_ffree, mp->mnt_stat.f_favail);
2271 
2272 	(*pr)("  f_fsidx {0x%x, 0x%x} owner %u ctime 0x%llx\n",
2273 	    mp->mnt_stat.f_fsid.val[0], mp->mnt_stat.f_fsid.val[1],
2274 	    mp->mnt_stat.f_owner, mp->mnt_stat.f_ctime);
2275 
2276  	(*pr)("  syncwrites %llu asyncwrites = %llu\n",
2277 	    mp->mnt_stat.f_syncwrites, mp->mnt_stat.f_asyncwrites);
2278 
2279  	(*pr)("  syncreads %llu asyncreads = %llu\n",
2280 	    mp->mnt_stat.f_syncreads, mp->mnt_stat.f_asyncreads);
2281 
2282 	(*pr)("  fstype \"%s\" mnton \"%s\" mntfrom \"%s\" mntspec \"%s\"\n",
2283 	    mp->mnt_stat.f_fstypename, mp->mnt_stat.f_mntonname,
2284 	    mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntfromspec);
2285 
2286 	(*pr)("locked vnodes:");
2287 	/* XXX would take mountlist lock, except ddb has no context */
2288 	cnt = 0;
2289 	LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2290 		if (VOP_ISLOCKED(vp)) {
2291 			if (cnt == 0)
2292 				(*pr)("\n  %p", vp);
2293 			else if ((cnt % (72 / (sizeof(void *) * 2 + 4))) == 0)
2294 				(*pr)(",\n  %p", vp);
2295 			else
2296 				(*pr)(", %p", vp);
2297 			cnt++;
2298 		}
2299 	}
2300 	(*pr)("\n");
2301 
2302 	if (full) {
2303 		(*pr)("all vnodes:");
2304 		/* XXX would take mountlist lock, except ddb has no context */
2305 		cnt = 0;
2306 		LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2307 			if (cnt == 0)
2308 				(*pr)("\n  %p", vp);
2309 			else if ((cnt % (72 / (sizeof(void *) * 2 + 4))) == 0)
2310 				(*pr)(",\n  %p", vp);
2311 			else
2312 				(*pr)(", %p", vp);
2313 			cnt++;
2314 		}
2315 		(*pr)("\n");
2316 	}
2317 }
2318 #endif /* DDB */
2319 
2320 void
2321 copy_statfs_info(struct statfs *sbp, const struct mount *mp)
2322 {
2323 	const struct statfs *mbp;
2324 
2325 	strncpy(sbp->f_fstypename, mp->mnt_vfc->vfc_name, MFSNAMELEN);
2326 
2327 	if (sbp == (mbp = &mp->mnt_stat))
2328 		return;
2329 
2330 	sbp->f_fsid = mbp->f_fsid;
2331 	sbp->f_owner = mbp->f_owner;
2332 	sbp->f_flags = mbp->f_flags;
2333 	sbp->f_syncwrites = mbp->f_syncwrites;
2334 	sbp->f_asyncwrites = mbp->f_asyncwrites;
2335 	sbp->f_syncreads = mbp->f_syncreads;
2336 	sbp->f_asyncreads = mbp->f_asyncreads;
2337 	sbp->f_namemax = mbp->f_namemax;
2338 	memcpy(sbp->f_mntonname, mp->mnt_stat.f_mntonname, MNAMELEN);
2339 	memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname, MNAMELEN);
2340 	memcpy(sbp->f_mntfromspec, mp->mnt_stat.f_mntfromspec, MNAMELEN);
2341 	memcpy(&sbp->mount_info, &mp->mnt_stat.mount_info,
2342 	    sizeof(union mount_info));
2343 }
2344