xref: /openbsd/sys/kern/vfs_subr.c (revision 5a38ef86)
1 /*	$OpenBSD: vfs_subr.c,v 1.313 2021/10/25 10:24:54 claudio 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/lockf.h>
56 #include <sys/stat.h>
57 #include <sys/acct.h>
58 #include <sys/namei.h>
59 #include <sys/ucred.h>
60 #include <sys/buf.h>
61 #include <sys/errno.h>
62 #include <sys/malloc.h>
63 #include <sys/mbuf.h>
64 #include <sys/syscallargs.h>
65 #include <sys/pool.h>
66 #include <sys/tree.h>
67 #include <sys/specdev.h>
68 #include <sys/atomic.h>
69 
70 #include <netinet/in.h>
71 
72 #include <uvm/uvm_extern.h>
73 #include <uvm/uvm_vnode.h>
74 
75 #include "softraid.h"
76 
77 void sr_quiesce(void);
78 
79 enum vtype iftovt_tab[16] = {
80 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
81 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
82 };
83 
84 int	vttoif_tab[9] = {
85 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
86 	S_IFSOCK, S_IFIFO, S_IFMT,
87 };
88 
89 int prtactive = 0;		/* 1 => print out reclaim of active vnodes */
90 int suid_clear = 1;		/* 1 => clear SUID / SGID on owner change */
91 
92 /*
93  * Insq/Remq for the vnode usage lists.
94  */
95 #define	bufinsvn(bp, dp)	LIST_INSERT_HEAD(dp, bp, b_vnbufs)
96 #define	bufremvn(bp) {							\
97 	LIST_REMOVE(bp, b_vnbufs);					\
98 	LIST_NEXT(bp, b_vnbufs) = NOLIST;				\
99 }
100 
101 struct freelst vnode_hold_list;	/* list of vnodes referencing buffers */
102 struct freelst vnode_free_list;	/* vnode free list */
103 
104 struct mntlist mountlist;	/* mounted filesystem list */
105 
106 void	vclean(struct vnode *, int, struct proc *);
107 
108 void insmntque(struct vnode *, struct mount *);
109 int getdevvp(dev_t, struct vnode **, enum vtype);
110 
111 int vfs_hang_addrlist(struct mount *, struct netexport *,
112 				  struct export_args *);
113 int vfs_free_netcred(struct radix_node *, void *, u_int);
114 void vfs_free_addrlist(struct netexport *);
115 void vputonfreelist(struct vnode *);
116 
117 int vflush_vnode(struct vnode *, void *);
118 int maxvnodes;
119 
120 struct mutex vnode_mtx = MUTEX_INITIALIZER(IPL_BIO);
121 
122 void vfs_unmountall(void);
123 
124 #ifdef DEBUG
125 void printlockedvnodes(void);
126 #endif
127 
128 struct pool vnode_pool;
129 struct pool uvm_vnode_pool;
130 
131 static inline int rb_buf_compare(const struct buf *b1, const struct buf *b2);
132 RBT_GENERATE(buf_rb_bufs, buf, b_rbbufs, rb_buf_compare);
133 
134 static inline int
135 rb_buf_compare(const struct buf *b1, const struct buf *b2)
136 {
137 	if (b1->b_lblkno < b2->b_lblkno)
138 		return(-1);
139 	if (b1->b_lblkno > b2->b_lblkno)
140 		return(1);
141 	return(0);
142 }
143 
144 /*
145  * Initialize the vnode management data structures.
146  */
147 void
148 vntblinit(void)
149 {
150 	/* buffer cache may need a vnode for each buffer */
151 	maxvnodes = 2 * initialvnodes;
152 	pool_init(&vnode_pool, sizeof(struct vnode), 0, IPL_NONE,
153 	    PR_WAITOK, "vnodes", NULL);
154 	pool_init(&uvm_vnode_pool, sizeof(struct uvm_vnode), 0, IPL_NONE,
155 	    PR_WAITOK, "uvmvnodes", NULL);
156 	TAILQ_INIT(&vnode_hold_list);
157 	TAILQ_INIT(&vnode_free_list);
158 	TAILQ_INIT(&mountlist);
159 	/*
160 	 * Initialize the filesystem syncer.
161 	 */
162 	vn_initialize_syncerd();
163 
164 #ifdef NFSSERVER
165 	rn_init(sizeof(struct sockaddr_in));
166 #endif /* NFSSERVER */
167 }
168 
169 /*
170  * Allocate a mount point.
171  *
172  * The returned mount point is marked as busy.
173  */
174 struct mount *
175 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp)
176 {
177 	struct mount *mp;
178 
179 	mp = malloc(sizeof(*mp), M_MOUNT, M_WAITOK|M_ZERO);
180 	rw_init_flags(&mp->mnt_lock, "vfslock", RWL_IS_VNODE);
181 	(void)vfs_busy(mp, VB_READ|VB_NOWAIT);
182 
183 	TAILQ_INIT(&mp->mnt_vnodelist);
184 	mp->mnt_vnodecovered = vp;
185 
186 	atomic_inc_int(&vfsp->vfc_refcount);
187 	mp->mnt_vfc = vfsp;
188 	mp->mnt_op = vfsp->vfc_vfsops;
189 	mp->mnt_flag = vfsp->vfc_flags;
190 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
191 
192 	return (mp);
193 }
194 
195 /*
196  * Release a mount point.
197  */
198 void
199 vfs_mount_free(struct mount *mp)
200 {
201 	atomic_dec_int(&mp->mnt_vfc->vfc_refcount);
202 	free(mp, M_MOUNT, sizeof(*mp));
203 }
204 
205 /*
206  * Mark a mount point as busy. Used to synchronize access and to delay
207  * unmounting.
208  *
209  * Default behaviour is to attempt getting a READ lock and in case of an
210  * ongoing unmount, to wait for it to finish and then return failure.
211  */
212 int
213 vfs_busy(struct mount *mp, int flags)
214 {
215 	int rwflags = 0;
216 
217 	if (flags & VB_WRITE)
218 		rwflags |= RW_WRITE;
219 	else
220 		rwflags |= RW_READ;
221 
222 	if (flags & VB_WAIT)
223 		rwflags |= RW_SLEEPFAIL;
224 	else
225 		rwflags |= RW_NOSLEEP;
226 
227 #ifdef WITNESS
228 	if (flags & VB_DUPOK)
229 		rwflags |= RW_DUPOK;
230 #endif
231 
232 	if (rw_enter(&mp->mnt_lock, rwflags))
233 		return (EBUSY);
234 
235 	return (0);
236 }
237 
238 /*
239  * Free a busy file system
240  */
241 void
242 vfs_unbusy(struct mount *mp)
243 {
244 	rw_exit(&mp->mnt_lock);
245 }
246 
247 int
248 vfs_isbusy(struct mount *mp)
249 {
250 	if (RWLOCK_OWNER(&mp->mnt_lock) > 0)
251 		return (1);
252 	else
253 		return (0);
254 }
255 
256 /*
257  * Lookup a filesystem type, and if found allocate and initialize
258  * a mount structure for it.
259  *
260  * Devname is usually updated by mount(8) after booting.
261  */
262 int
263 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
264 {
265 	struct vfsconf *vfsp;
266 	struct mount *mp;
267 
268 	vfsp = vfs_byname(fstypename);
269 	if (vfsp == NULL)
270 		return (ENODEV);
271 	mp = vfs_mount_alloc(NULLVP, vfsp);
272 	mp->mnt_flag |= MNT_RDONLY;
273 	mp->mnt_stat.f_mntonname[0] = '/';
274 	copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN, NULL);
275 	copystr(devname, mp->mnt_stat.f_mntfromspec, MNAMELEN, NULL);
276 	*mpp = mp;
277 	return (0);
278  }
279 
280 /*
281  * Lookup a mount point by filesystem identifier.
282  */
283 struct mount *
284 vfs_getvfs(fsid_t *fsid)
285 {
286 	struct mount *mp;
287 
288 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
289 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
290 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
291 			return (mp);
292 		}
293 	}
294 
295 	return (NULL);
296 }
297 
298 
299 /*
300  * Get a new unique fsid
301  */
302 void
303 vfs_getnewfsid(struct mount *mp)
304 {
305 	static u_short xxxfs_mntid;
306 
307 	fsid_t tfsid;
308 	int mtype;
309 
310 	mtype = mp->mnt_vfc->vfc_typenum;
311 	mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev + mtype, 0);
312 	mp->mnt_stat.f_fsid.val[1] = mtype;
313 	if (xxxfs_mntid == 0)
314 		++xxxfs_mntid;
315 	tfsid.val[0] = makedev(nblkdev + mtype, xxxfs_mntid);
316 	tfsid.val[1] = mtype;
317 	if (!TAILQ_EMPTY(&mountlist)) {
318 		while (vfs_getvfs(&tfsid)) {
319 			tfsid.val[0]++;
320 			xxxfs_mntid++;
321 		}
322 	}
323 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
324 }
325 
326 /*
327  * Set vnode attributes to VNOVAL
328  */
329 void
330 vattr_null(struct vattr *vap)
331 {
332 
333 	vap->va_type = VNON;
334 	/*
335 	 * Don't get fancy: u_quad_t = u_int = VNOVAL leaves the u_quad_t
336 	 * with 2^31-1 instead of 2^64-1.  Just write'm out and let
337 	 * the compiler do its job.
338 	 */
339 	vap->va_mode = VNOVAL;
340 	vap->va_nlink = VNOVAL;
341 	vap->va_uid = VNOVAL;
342 	vap->va_gid = VNOVAL;
343 	vap->va_fsid = VNOVAL;
344 	vap->va_fileid = VNOVAL;
345 	vap->va_size = VNOVAL;
346 	vap->va_blocksize = VNOVAL;
347 	vap->va_atime.tv_sec = VNOVAL;
348 	vap->va_atime.tv_nsec = VNOVAL;
349 	vap->va_mtime.tv_sec = VNOVAL;
350 	vap->va_mtime.tv_nsec = VNOVAL;
351 	vap->va_ctime.tv_sec = VNOVAL;
352 	vap->va_ctime.tv_nsec = VNOVAL;
353 	vap->va_gen = VNOVAL;
354 	vap->va_flags = VNOVAL;
355 	vap->va_rdev = VNOVAL;
356 	vap->va_bytes = VNOVAL;
357 	vap->va_filerev = VNOVAL;
358 	vap->va_vaflags = 0;
359 }
360 
361 /*
362  * Routines having to do with the management of the vnode table.
363  */
364 long numvnodes;
365 
366 /*
367  * Return the next vnode from the free list.
368  */
369 int
370 getnewvnode(enum vtagtype tag, struct mount *mp, const struct vops *vops,
371     struct vnode **vpp)
372 {
373 	struct proc *p = curproc;
374 	struct freelst *listhd;
375 	static int toggle;
376 	struct vnode *vp;
377 	int s;
378 
379 	/*
380 	 * allow maxvnodes to increase if the buffer cache itself
381 	 * is big enough to justify it. (we don't shrink it ever)
382 	 */
383 	maxvnodes = maxvnodes < bcstats.numbufs ? bcstats.numbufs
384 	    : maxvnodes;
385 
386 	/*
387 	 * We must choose whether to allocate a new vnode or recycle an
388 	 * existing one. The criterion for allocating a new one is that
389 	 * the total number of vnodes is less than the number desired or
390 	 * there are no vnodes on either free list. Generally we only
391 	 * want to recycle vnodes that have no buffers associated with
392 	 * them, so we look first on the vnode_free_list. If it is empty,
393 	 * we next consider vnodes with referencing buffers on the
394 	 * vnode_hold_list. The toggle ensures that half the time we
395 	 * will use a buffer from the vnode_hold_list, and half the time
396 	 * we will allocate a new one unless the list has grown to twice
397 	 * the desired size. We are reticent to recycle vnodes from the
398 	 * vnode_hold_list because we will lose the identity of all its
399 	 * referencing buffers.
400 	 */
401 	toggle ^= 1;
402 	if (numvnodes / 2 > maxvnodes)
403 		toggle = 0;
404 
405 	s = splbio();
406 	if ((numvnodes < maxvnodes) ||
407 	    ((TAILQ_FIRST(listhd = &vnode_free_list) == NULL) &&
408 	    ((TAILQ_FIRST(listhd = &vnode_hold_list) == NULL) || toggle))) {
409 		splx(s);
410 		vp = pool_get(&vnode_pool, PR_WAITOK | PR_ZERO);
411 		vp->v_uvm = pool_get(&uvm_vnode_pool, PR_WAITOK | PR_ZERO);
412 		vp->v_uvm->u_vnode = vp;
413 		uvm_obj_init(&vp->v_uvm->u_obj, &uvm_vnodeops, 0);
414 		RBT_INIT(buf_rb_bufs, &vp->v_bufs_tree);
415 		cache_tree_init(&vp->v_nc_tree);
416 		TAILQ_INIT(&vp->v_cache_dst);
417 		numvnodes++;
418 	} else {
419 		TAILQ_FOREACH(vp, listhd, v_freelist) {
420 			if (VOP_ISLOCKED(vp) == 0)
421 				break;
422 		}
423 		/*
424 		 * Unless this is a bad time of the month, at most
425 		 * the first NCPUS items on the free list are
426 		 * locked, so this is close enough to being empty.
427 		 */
428 		if (vp == NULL) {
429 			splx(s);
430 			tablefull("vnode");
431 			*vpp = NULL;
432 			return (ENFILE);
433 		}
434 
435 #ifdef DIAGNOSTIC
436 		if (vp->v_usecount) {
437 			vprint("free vnode", vp);
438 			panic("free vnode isn't");
439 		}
440 #endif
441 
442 		TAILQ_REMOVE(listhd, vp, v_freelist);
443 		vp->v_bioflag &= ~VBIOONFREELIST;
444 		splx(s);
445 
446 		if (vp->v_type != VBAD)
447 			vgonel(vp, p);
448 #ifdef DIAGNOSTIC
449 		if (vp->v_data) {
450 			vprint("cleaned vnode", vp);
451 			panic("cleaned vnode isn't");
452 		}
453 		s = splbio();
454 		if (vp->v_numoutput)
455 			panic("Clean vnode has pending I/O's");
456 		splx(s);
457 #endif
458 		vp->v_flag = 0;
459 		vp->v_socket = NULL;
460 	}
461 	cache_purge(vp);
462 	vp->v_type = VNON;
463 	vp->v_tag = tag;
464 	vp->v_op = vops;
465 	insmntque(vp, mp);
466 	*vpp = vp;
467 	vp->v_usecount = 1;
468 	vp->v_data = NULL;
469 	return (0);
470 }
471 
472 /*
473  * Move a vnode from one mount queue to another.
474  */
475 void
476 insmntque(struct vnode *vp, struct mount *mp)
477 {
478 	/*
479 	 * Delete from old mount point vnode list, if on one.
480 	 */
481 	if (vp->v_mount != NULL)
482 		TAILQ_REMOVE(&vp->v_mount->mnt_vnodelist, vp, v_mntvnodes);
483 	/*
484 	 * Insert into list of vnodes for the new mount point, if available.
485 	 */
486 	if ((vp->v_mount = mp) != NULL)
487 		TAILQ_INSERT_TAIL(&mp->mnt_vnodelist, vp, v_mntvnodes);
488 }
489 
490 /*
491  * Create a vnode for a block device.
492  * Used for root filesystem, argdev, and swap areas.
493  * Also used for memory file system special devices.
494  */
495 int
496 bdevvp(dev_t dev, struct vnode **vpp)
497 {
498 	return (getdevvp(dev, vpp, VBLK));
499 }
500 
501 /*
502  * Create a vnode for a character device.
503  * Used for console handling.
504  */
505 int
506 cdevvp(dev_t dev, struct vnode **vpp)
507 {
508 	return (getdevvp(dev, vpp, VCHR));
509 }
510 
511 /*
512  * Create a vnode for a device.
513  * Used by bdevvp (block device) for root file system etc.,
514  * and by cdevvp (character device) for console.
515  */
516 int
517 getdevvp(dev_t dev, struct vnode **vpp, enum vtype type)
518 {
519 	struct vnode *vp;
520 	struct vnode *nvp;
521 	int error;
522 
523 	if (dev == NODEV) {
524 		*vpp = NULLVP;
525 		return (0);
526 	}
527 	error = getnewvnode(VT_NON, NULL, &spec_vops, &nvp);
528 	if (error) {
529 		*vpp = NULLVP;
530 		return (error);
531 	}
532 	vp = nvp;
533 	vp->v_type = type;
534 	if ((nvp = checkalias(vp, dev, NULL)) != NULL) {
535 		vput(vp);
536 		vp = nvp;
537 	}
538 	if (vp->v_type == VCHR && cdevsw[major(vp->v_rdev)].d_type == D_TTY)
539 		vp->v_flag |= VISTTY;
540 	*vpp = vp;
541 	return (0);
542 }
543 
544 /*
545  * Check to see if the new vnode represents a special device
546  * for which we already have a vnode (either because of
547  * bdevvp() or because of a different vnode representing
548  * the same block device). If such an alias exists, deallocate
549  * the existing contents and return the aliased vnode. The
550  * caller is responsible for filling it with its new contents.
551  */
552 struct vnode *
553 checkalias(struct vnode *nvp, dev_t nvp_rdev, struct mount *mp)
554 {
555 	struct proc *p = curproc;
556 	struct vnode *vp;
557 	struct vnodechain *vchain;
558 
559 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
560 		return (NULLVP);
561 
562 	vchain = &speclisth[SPECHASH(nvp_rdev)];
563 loop:
564 	SLIST_FOREACH(vp, vchain, v_specnext) {
565 		if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type) {
566 			continue;
567 		}
568 		/*
569 		 * Alias, but not in use, so flush it out.
570 		 */
571 		if (vp->v_usecount == 0) {
572 			vgonel(vp, p);
573 			goto loop;
574 		}
575 		if (vget(vp, LK_EXCLUSIVE)) {
576 			goto loop;
577 		}
578 		break;
579 	}
580 
581 	/*
582 	 * Common case is actually in the if statement
583 	 */
584 	if (vp == NULL || !(vp->v_tag == VT_NON && vp->v_type == VBLK)) {
585 		nvp->v_specinfo = malloc(sizeof(struct specinfo), M_VNODE,
586 			M_WAITOK);
587 		nvp->v_rdev = nvp_rdev;
588 		nvp->v_hashchain = vchain;
589 		nvp->v_specmountpoint = NULL;
590 		nvp->v_speclockf = NULL;
591 		nvp->v_specbitmap = NULL;
592 		if (nvp->v_type == VCHR &&
593 		    (cdevsw[major(nvp_rdev)].d_flags & D_CLONE) &&
594 		    (minor(nvp_rdev) >> CLONE_SHIFT == 0)) {
595 			if (vp != NULLVP)
596 				nvp->v_specbitmap = vp->v_specbitmap;
597 			else
598 				nvp->v_specbitmap = malloc(CLONE_MAPSZ,
599 				    M_VNODE, M_WAITOK | M_ZERO);
600 		}
601 		SLIST_INSERT_HEAD(vchain, nvp, v_specnext);
602 		if (vp != NULLVP) {
603 			nvp->v_flag |= VALIASED;
604 			vp->v_flag |= VALIASED;
605 			vput(vp);
606 		}
607 		return (NULLVP);
608 	}
609 
610 	/*
611 	 * This code is the uncommon case. It is called in case
612 	 * we found an alias that was VT_NON && vtype of VBLK
613 	 * This means we found a block device that was created
614 	 * using bdevvp.
615 	 * An example of such a vnode is the root partition device vnode
616 	 * created in ffs_mountroot.
617 	 *
618 	 * The vnodes created by bdevvp should not be aliased (why?).
619 	 */
620 
621 	VOP_UNLOCK(vp);
622 	vclean(vp, 0, p);
623 	vp->v_op = nvp->v_op;
624 	vp->v_tag = nvp->v_tag;
625 	nvp->v_type = VNON;
626 	insmntque(vp, mp);
627 	return (vp);
628 }
629 
630 /*
631  * Grab a particular vnode from the free list, increment its
632  * reference count and lock it. If the vnode lock bit is set,
633  * the vnode is being eliminated in vgone. In that case, we
634  * cannot grab it, so the process is awakened when the
635  * transition is completed, and an error code is returned to
636  * indicate that the vnode is no longer usable, possibly
637  * having been changed to a new file system type.
638  */
639 int
640 vget(struct vnode *vp, int flags)
641 {
642 	int error, s, onfreelist;
643 
644 	/*
645 	 * If the vnode is in the process of being cleaned out for
646 	 * another use, we wait for the cleaning to finish and then
647 	 * return failure. Cleaning is determined by checking that
648 	 * the VXLOCK flag is set.
649 	 */
650 	mtx_enter(&vnode_mtx);
651 	if (vp->v_lflag & VXLOCK) {
652 		if (flags & LK_NOWAIT) {
653 			mtx_leave(&vnode_mtx);
654 			return (EBUSY);
655 		}
656 
657 		vp->v_lflag |= VXWANT;
658 		msleep_nsec(vp, &vnode_mtx, PINOD, "vget", INFSLP);
659 		mtx_leave(&vnode_mtx);
660 		return (ENOENT);
661 	}
662 	mtx_leave(&vnode_mtx);
663 
664 	onfreelist = vp->v_bioflag & VBIOONFREELIST;
665 	if (vp->v_usecount == 0 && onfreelist) {
666 		s = splbio();
667 		if (vp->v_holdcnt > 0)
668 			TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist);
669 		else
670 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
671 		vp->v_bioflag &= ~VBIOONFREELIST;
672 		splx(s);
673 	}
674 
675 	vp->v_usecount++;
676 	if (flags & LK_TYPE_MASK) {
677 		if ((error = vn_lock(vp, flags)) != 0) {
678 			vp->v_usecount--;
679 			if (vp->v_usecount == 0 && onfreelist)
680 				vputonfreelist(vp);
681 		}
682 		return (error);
683 	}
684 
685 	return (0);
686 }
687 
688 
689 /* Vnode reference. */
690 void
691 vref(struct vnode *vp)
692 {
693 	KERNEL_ASSERT_LOCKED();
694 
695 #ifdef DIAGNOSTIC
696 	if (vp->v_usecount == 0)
697 		panic("vref used where vget required");
698 	if (vp->v_type == VNON)
699 		panic("vref on a VNON vnode");
700 #endif
701 	vp->v_usecount++;
702 }
703 
704 void
705 vputonfreelist(struct vnode *vp)
706 {
707 	int s;
708 	struct freelst *lst;
709 
710 	s = splbio();
711 #ifdef DIAGNOSTIC
712 	if (vp->v_usecount != 0)
713 		panic("Use count is not zero!");
714 
715 	/*
716 	 * If the hold count is still positive, one or many threads could still
717 	 * be waiting on the vnode lock inside uvn_io().
718 	 */
719 	if (vp->v_holdcnt == 0 && vp->v_lockcount != 0)
720 		panic("%s: lock count is not zero", __func__);
721 
722 	if (vp->v_bioflag & VBIOONFREELIST) {
723 		vprint("vnode already on free list: ", vp);
724 		panic("vnode already on free list");
725 	}
726 #endif
727 
728 	vp->v_bioflag |= VBIOONFREELIST;
729 	vp->v_bioflag &= ~VBIOERROR;
730 
731 	if (vp->v_holdcnt > 0)
732 		lst = &vnode_hold_list;
733 	else
734 		lst = &vnode_free_list;
735 
736 	if (vp->v_type == VBAD)
737 		TAILQ_INSERT_HEAD(lst, vp, v_freelist);
738 	else
739 		TAILQ_INSERT_TAIL(lst, vp, v_freelist);
740 
741 	splx(s);
742 }
743 
744 /*
745  * vput(), just unlock and vrele()
746  */
747 void
748 vput(struct vnode *vp)
749 {
750 	struct proc *p = curproc;
751 
752 #ifdef DIAGNOSTIC
753 	if (vp == NULL)
754 		panic("vput: null vp");
755 #endif
756 
757 #ifdef DIAGNOSTIC
758 	if (vp->v_usecount == 0) {
759 		vprint("vput: bad ref count", vp);
760 		panic("vput: ref cnt");
761 	}
762 #endif
763 	vp->v_usecount--;
764 	KASSERT(vp->v_usecount > 0 || vp->v_uvcount == 0);
765 	if (vp->v_usecount > 0) {
766 		VOP_UNLOCK(vp);
767 		return;
768 	}
769 
770 #ifdef DIAGNOSTIC
771 	if (vp->v_writecount != 0) {
772 		vprint("vput: bad writecount", vp);
773 		panic("vput: v_writecount != 0");
774 	}
775 #endif
776 
777 	VOP_INACTIVE(vp, p);
778 
779 	if (vp->v_usecount == 0 && !(vp->v_bioflag & VBIOONFREELIST))
780 		vputonfreelist(vp);
781 }
782 
783 /*
784  * Vnode release - use for active VNODES.
785  * If count drops to zero, call inactive routine and return to freelist.
786  * Returns 0 if it did not sleep.
787  */
788 int
789 vrele(struct vnode *vp)
790 {
791 	struct proc *p = curproc;
792 
793 #ifdef DIAGNOSTIC
794 	if (vp == NULL)
795 		panic("vrele: null vp");
796 #endif
797 #ifdef DIAGNOSTIC
798 	if (vp->v_usecount == 0) {
799 		vprint("vrele: bad ref count", vp);
800 		panic("vrele: ref cnt");
801 	}
802 #endif
803 	vp->v_usecount--;
804 	if (vp->v_usecount > 0) {
805 		return (0);
806 	}
807 
808 #ifdef DIAGNOSTIC
809 	if (vp->v_writecount != 0) {
810 		vprint("vrele: bad writecount", vp);
811 		panic("vrele: v_writecount != 0");
812 	}
813 #endif
814 
815 	if (vn_lock(vp, LK_EXCLUSIVE)) {
816 #ifdef DIAGNOSTIC
817 		vprint("vrele: cannot lock", vp);
818 #endif
819 		return (1);
820 	}
821 
822 	VOP_INACTIVE(vp, p);
823 
824 	if (vp->v_usecount == 0 && !(vp->v_bioflag & VBIOONFREELIST))
825 		vputonfreelist(vp);
826 	return (1);
827 }
828 
829 /* Page or buffer structure gets a reference. */
830 void
831 vhold(struct vnode *vp)
832 {
833 	/*
834 	 * If it is on the freelist and the hold count is currently
835 	 * zero, move it to the hold list.
836 	 */
837 	if ((vp->v_bioflag & VBIOONFREELIST) &&
838 	    vp->v_holdcnt == 0 && vp->v_usecount == 0) {
839 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
840 		TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist);
841 	}
842 	vp->v_holdcnt++;
843 }
844 
845 /* Lose interest in a vnode. */
846 void
847 vdrop(struct vnode *vp)
848 {
849 #ifdef DIAGNOSTIC
850 	if (vp->v_holdcnt == 0)
851 		panic("vdrop: zero holdcnt");
852 #endif
853 
854 	vp->v_holdcnt--;
855 
856 	/*
857 	 * If it is on the holdlist and the hold count drops to
858 	 * zero, move it to the free list.
859 	 */
860 	if ((vp->v_bioflag & VBIOONFREELIST) &&
861 	    vp->v_holdcnt == 0 && vp->v_usecount == 0) {
862 		TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist);
863 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
864 	}
865 }
866 
867 /*
868  * Remove any vnodes in the vnode table belonging to mount point mp.
869  *
870  * If MNT_NOFORCE is specified, there should not be any active ones,
871  * return error if any are found (nb: this is a user error, not a
872  * system error). If MNT_FORCE is specified, detach any active vnodes
873  * that are found.
874  */
875 #ifdef DEBUG_SYSCTL
876 int busyprt = 0;	/* print out busy vnodes */
877 struct ctldebug debug_vfs_busyprt = { "vfs_busyprt", &busyprt };
878 #endif
879 
880 int
881 vfs_mount_foreach_vnode(struct mount *mp,
882     int (*func)(struct vnode *, void *), void *arg) {
883 	struct vnode *vp, *nvp;
884 	int error = 0;
885 
886 loop:
887 	TAILQ_FOREACH_SAFE(vp , &mp->mnt_vnodelist, v_mntvnodes, nvp) {
888 		if (vp->v_mount != mp)
889 			goto loop;
890 
891 		error = func(vp, arg);
892 
893 		if (error != 0)
894 			break;
895 	}
896 
897 	return (error);
898 }
899 
900 struct vflush_args {
901 	struct vnode *skipvp;
902 	int busy;
903 	int flags;
904 };
905 
906 int
907 vflush_vnode(struct vnode *vp, void *arg)
908 {
909 	struct vflush_args *va = arg;
910 	struct proc *p = curproc;
911 
912 	if (vp == va->skipvp) {
913 		return (0);
914 	}
915 
916 	if ((va->flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
917 		return (0);
918 	}
919 
920 	/*
921 	 * If WRITECLOSE is set, only flush out regular file
922 	 * vnodes open for writing.
923 	 */
924 	if ((va->flags & WRITECLOSE) &&
925 	    (vp->v_writecount == 0 || vp->v_type != VREG)) {
926 		return (0);
927 	}
928 
929 	/*
930 	 * With v_usecount == 0, all we need to do is clear
931 	 * out the vnode data structures and we are done.
932 	 */
933 	if (vp->v_usecount == 0) {
934 		vgonel(vp, p);
935 		return (0);
936 	}
937 
938 	/*
939 	 * If FORCECLOSE is set, forcibly close the vnode.
940 	 * For block or character devices, revert to an
941 	 * anonymous device. For all other files, just kill them.
942 	 */
943 	if (va->flags & FORCECLOSE) {
944 		if (vp->v_type != VBLK && vp->v_type != VCHR) {
945 			vgonel(vp, p);
946 		} else {
947 			vclean(vp, 0, p);
948 			vp->v_op = &spec_vops;
949 			insmntque(vp, NULL);
950 		}
951 		return (0);
952 	}
953 
954 	/*
955 	 * If set, this is allowed to ignore vnodes which don't
956 	 * have changes pending to disk.
957 	 * XXX Might be nice to check per-fs "inode" flags, but
958 	 * generally the filesystem is sync'd already, right?
959 	 */
960 	if ((va->flags & IGNORECLEAN) &&
961 	    LIST_EMPTY(&vp->v_dirtyblkhd))
962 		return (0);
963 
964 #ifdef DEBUG_SYSCTL
965 	if (busyprt)
966 		vprint("vflush: busy vnode", vp);
967 #endif
968 	va->busy++;
969 	return (0);
970 }
971 
972 int
973 vflush(struct mount *mp, struct vnode *skipvp, int flags)
974 {
975 	struct vflush_args va;
976 	va.skipvp = skipvp;
977 	va.busy = 0;
978 	va.flags = flags;
979 
980 	vfs_mount_foreach_vnode(mp, vflush_vnode, &va);
981 
982 	if (va.busy)
983 		return (EBUSY);
984 	return (0);
985 }
986 
987 /*
988  * Disassociate the underlying file system from a vnode.
989  */
990 void
991 vclean(struct vnode *vp, int flags, struct proc *p)
992 {
993 	int active, do_wakeup = 0;
994 
995 	/*
996 	 * Check to see if the vnode is in use.
997 	 * If so we have to reference it before we clean it out
998 	 * so that its count cannot fall to zero and generate a
999 	 * race against ourselves to recycle it.
1000 	 */
1001 	if ((active = vp->v_usecount) != 0)
1002 		vp->v_usecount++;
1003 
1004 	/*
1005 	 * Prevent the vnode from being recycled or
1006 	 * brought into use while we clean it out.
1007 	 */
1008 	mtx_enter(&vnode_mtx);
1009 	if (vp->v_lflag & VXLOCK)
1010 		panic("vclean: deadlock");
1011 	vp->v_lflag |= VXLOCK;
1012 
1013 	if (vp->v_lockcount > 0) {
1014 		/*
1015 		 * Ensure that any thread currently waiting on the same lock has
1016 		 * observed that the vnode is about to be exclusively locked
1017 		 * before continuing.
1018 		 */
1019 		msleep_nsec(&vp->v_lockcount, &vnode_mtx, PINOD, "vop_lock",
1020 		    INFSLP);
1021 		KASSERT(vp->v_lockcount == 0);
1022 	}
1023 	mtx_leave(&vnode_mtx);
1024 
1025 	/*
1026 	 * Even if the count is zero, the VOP_INACTIVE routine may still
1027 	 * have the object locked while it cleans it out. The VOP_LOCK
1028 	 * ensures that the VOP_INACTIVE routine is done with its work.
1029 	 * For active vnodes, it ensures that no other activity can
1030 	 * occur while the underlying object is being cleaned out.
1031 	 */
1032 	VOP_LOCK(vp, LK_EXCLUSIVE | LK_DRAIN);
1033 
1034 	/*
1035 	 * Clean out any VM data associated with the vnode.
1036 	 */
1037 	uvm_vnp_terminate(vp);
1038 	/*
1039 	 * Clean out any buffers associated with the vnode.
1040 	 */
1041 	if (flags & DOCLOSE)
1042 		vinvalbuf(vp, V_SAVE, NOCRED, p, 0, INFSLP);
1043 	/*
1044 	 * If purging an active vnode, it must be closed and
1045 	 * deactivated before being reclaimed. Note that the
1046 	 * VOP_INACTIVE will unlock the vnode
1047 	 */
1048 	if (active) {
1049 		if (flags & DOCLOSE)
1050 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, p);
1051 		VOP_INACTIVE(vp, p);
1052 	} else {
1053 		/*
1054 		 * Any other processes trying to obtain this lock must first
1055 		 * wait for VXLOCK to clear, then call the new lock operation.
1056 		 */
1057 		VOP_UNLOCK(vp);
1058 	}
1059 
1060 	/*
1061 	 * Reclaim the vnode.
1062 	 */
1063 	if (VOP_RECLAIM(vp, p))
1064 		panic("vclean: cannot reclaim");
1065 	if (active) {
1066 		vp->v_usecount--;
1067 		if (vp->v_usecount == 0) {
1068 			if (vp->v_holdcnt > 0)
1069 				panic("vclean: not clean");
1070 			vputonfreelist(vp);
1071 		}
1072 	}
1073 	cache_purge(vp);
1074 
1075 	/*
1076 	 * Done with purge, notify sleepers of the grim news.
1077 	 */
1078 	vp->v_op = &dead_vops;
1079 	VN_KNOTE(vp, NOTE_REVOKE);
1080 	vp->v_tag = VT_NON;
1081 #ifdef VFSLCKDEBUG
1082 	vp->v_flag &= ~VLOCKSWORK;
1083 #endif
1084 	mtx_enter(&vnode_mtx);
1085 	vp->v_lflag &= ~VXLOCK;
1086 	if (vp->v_lflag & VXWANT) {
1087 		vp->v_lflag &= ~VXWANT;
1088 		do_wakeup = 1;
1089 	}
1090 	mtx_leave(&vnode_mtx);
1091 	if (do_wakeup)
1092 		wakeup(vp);
1093 }
1094 
1095 /*
1096  * Recycle an unused vnode to the front of the free list.
1097  */
1098 int
1099 vrecycle(struct vnode *vp, struct proc *p)
1100 {
1101 	if (vp->v_usecount == 0) {
1102 		vgonel(vp, p);
1103 		return (1);
1104 	}
1105 	return (0);
1106 }
1107 
1108 /*
1109  * Eliminate all activity associated with a vnode
1110  * in preparation for reuse.
1111  */
1112 void
1113 vgone(struct vnode *vp)
1114 {
1115 	struct proc *p = curproc;
1116 	vgonel(vp, p);
1117 }
1118 
1119 /*
1120  * vgone, with struct proc.
1121  */
1122 void
1123 vgonel(struct vnode *vp, struct proc *p)
1124 {
1125 	struct vnode *vq;
1126 	struct vnode *vx;
1127 
1128 	KASSERT(vp->v_uvcount == 0);
1129 
1130 	/*
1131 	 * If a vgone (or vclean) is already in progress,
1132 	 * wait until it is done and return.
1133 	 */
1134 	mtx_enter(&vnode_mtx);
1135 	if (vp->v_lflag & VXLOCK) {
1136 		vp->v_lflag |= VXWANT;
1137 		msleep_nsec(vp, &vnode_mtx, PINOD, "vgone", INFSLP);
1138 		mtx_leave(&vnode_mtx);
1139 		return;
1140 	}
1141 	mtx_leave(&vnode_mtx);
1142 
1143 	/*
1144 	 * Clean out the filesystem specific data.
1145 	 */
1146 	vclean(vp, DOCLOSE, p);
1147 	/*
1148 	 * Delete from old mount point vnode list, if on one.
1149 	 */
1150 	if (vp->v_mount != NULL)
1151 		insmntque(vp, NULL);
1152 	/*
1153 	 * If special device, remove it from special device alias list
1154 	 * if it is on one.
1155 	 */
1156 	if ((vp->v_type == VBLK || vp->v_type == VCHR) &&
1157 	    vp->v_specinfo != NULL) {
1158 		if ((vp->v_flag & VALIASED) == 0 && vp->v_type == VCHR &&
1159 		    (cdevsw[major(vp->v_rdev)].d_flags & D_CLONE) &&
1160 		    (minor(vp->v_rdev) >> CLONE_SHIFT == 0)) {
1161 			free(vp->v_specbitmap, M_VNODE, CLONE_MAPSZ);
1162 		}
1163 		SLIST_REMOVE(vp->v_hashchain, vp, vnode, v_specnext);
1164 		if (vp->v_flag & VALIASED) {
1165 			vx = NULL;
1166 			SLIST_FOREACH(vq, vp->v_hashchain, v_specnext) {
1167 				if (vq->v_rdev != vp->v_rdev ||
1168 				    vq->v_type != vp->v_type)
1169 					continue;
1170 				if (vx)
1171 					break;
1172 				vx = vq;
1173 			}
1174 			if (vx == NULL)
1175 				panic("missing alias");
1176 			if (vq == NULL)
1177 				vx->v_flag &= ~VALIASED;
1178 			vp->v_flag &= ~VALIASED;
1179 		}
1180 		lf_purgelocks(&vp->v_speclockf);
1181 		free(vp->v_specinfo, M_VNODE, sizeof(struct specinfo));
1182 		vp->v_specinfo = NULL;
1183 	}
1184 	/*
1185 	 * If it is on the freelist and not already at the head,
1186 	 * move it to the head of the list.
1187 	 */
1188 	vp->v_type = VBAD;
1189 
1190 	/*
1191 	 * Move onto the free list, unless we were called from
1192 	 * getnewvnode and we're not on any free list
1193 	 */
1194 	if (vp->v_usecount == 0 &&
1195 	    (vp->v_bioflag & VBIOONFREELIST)) {
1196 		int s;
1197 
1198 		s = splbio();
1199 
1200 		if (vp->v_holdcnt > 0)
1201 			panic("vgonel: not clean");
1202 
1203 		if (TAILQ_FIRST(&vnode_free_list) != vp) {
1204 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1205 			TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
1206 		}
1207 		splx(s);
1208 	}
1209 }
1210 
1211 /*
1212  * Lookup a vnode by device number.
1213  */
1214 int
1215 vfinddev(dev_t dev, enum vtype type, struct vnode **vpp)
1216 {
1217 	struct vnode *vp;
1218 	int rc =0;
1219 
1220 	SLIST_FOREACH(vp, &speclisth[SPECHASH(dev)], v_specnext) {
1221 		if (dev != vp->v_rdev || type != vp->v_type)
1222 			continue;
1223 		*vpp = vp;
1224 		rc = 1;
1225 		break;
1226 	}
1227 	return (rc);
1228 }
1229 
1230 /*
1231  * Revoke all the vnodes corresponding to the specified minor number
1232  * range (endpoints inclusive) of the specified major.
1233  */
1234 void
1235 vdevgone(int maj, int minl, int minh, enum vtype type)
1236 {
1237 	struct vnode *vp;
1238 	int mn;
1239 
1240 	for (mn = minl; mn <= minh; mn++)
1241 		if (vfinddev(makedev(maj, mn), type, &vp))
1242 			VOP_REVOKE(vp, REVOKEALL);
1243 }
1244 
1245 /*
1246  * Calculate the total number of references to a special device.
1247  */
1248 int
1249 vcount(struct vnode *vp)
1250 {
1251 	struct vnode *vq;
1252 	int count;
1253 
1254 loop:
1255 	if ((vp->v_flag & VALIASED) == 0)
1256 		return (vp->v_usecount);
1257 	count = 0;
1258 	SLIST_FOREACH(vq, vp->v_hashchain, v_specnext) {
1259 		if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type)
1260 			continue;
1261 		/*
1262 		 * Alias, but not in use, so flush it out.
1263 		 */
1264 		if (vq->v_usecount == 0 && vq != vp) {
1265 			vgone(vq);
1266 			goto loop;
1267 		}
1268 		count += vq->v_usecount;
1269 	}
1270 	return (count);
1271 }
1272 
1273 #if defined(DEBUG) || defined(DIAGNOSTIC)
1274 /*
1275  * Print out a description of a vnode.
1276  */
1277 static char *typename[] =
1278    { "VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD" };
1279 
1280 void
1281 vprint(char *label, struct vnode *vp)
1282 {
1283 	char buf[64];
1284 
1285 	if (label != NULL)
1286 		printf("%s: ", label);
1287 	printf("%p, type %s, use %u, write %u, hold %u,",
1288 		vp, typename[vp->v_type], vp->v_usecount, vp->v_writecount,
1289 		vp->v_holdcnt);
1290 	buf[0] = '\0';
1291 	if (vp->v_flag & VROOT)
1292 		strlcat(buf, "|VROOT", sizeof buf);
1293 	if (vp->v_flag & VTEXT)
1294 		strlcat(buf, "|VTEXT", sizeof buf);
1295 	if (vp->v_flag & VSYSTEM)
1296 		strlcat(buf, "|VSYSTEM", sizeof buf);
1297 	if (vp->v_lflag & VXLOCK)
1298 		strlcat(buf, "|VXLOCK", sizeof buf);
1299 	if (vp->v_lflag & VXWANT)
1300 		strlcat(buf, "|VXWANT", sizeof buf);
1301 	if (vp->v_bioflag & VBIOWAIT)
1302 		strlcat(buf, "|VBIOWAIT", sizeof buf);
1303 	if (vp->v_bioflag & VBIOONFREELIST)
1304 		strlcat(buf, "|VBIOONFREELIST", sizeof buf);
1305 	if (vp->v_bioflag & VBIOONSYNCLIST)
1306 		strlcat(buf, "|VBIOONSYNCLIST", sizeof buf);
1307 	if (vp->v_flag & VALIASED)
1308 		strlcat(buf, "|VALIASED", sizeof buf);
1309 	if (buf[0] != '\0')
1310 		printf(" flags (%s)", &buf[1]);
1311 	if (vp->v_data == NULL) {
1312 		printf("\n");
1313 	} else {
1314 		printf("\n\t");
1315 		VOP_PRINT(vp);
1316 	}
1317 }
1318 #endif /* DEBUG || DIAGNOSTIC */
1319 
1320 #ifdef DEBUG
1321 /*
1322  * List all of the locked vnodes in the system.
1323  * Called when debugging the kernel.
1324  */
1325 void
1326 printlockedvnodes(void)
1327 {
1328 	struct mount *mp;
1329 	struct vnode *vp;
1330 
1331 	printf("Locked vnodes\n");
1332 
1333 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
1334 		if (vfs_busy(mp, VB_READ|VB_NOWAIT))
1335 			continue;
1336 		TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
1337 			if (VOP_ISLOCKED(vp))
1338 				vprint(NULL, vp);
1339 		}
1340 		vfs_unbusy(mp);
1341 	}
1342 
1343 }
1344 #endif
1345 
1346 /*
1347  * Top level filesystem related information gathering.
1348  */
1349 int
1350 vfs_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp,
1351     size_t newlen, struct proc *p)
1352 {
1353 	struct vfsconf *vfsp, *tmpvfsp;
1354 	int ret;
1355 
1356 	/* all sysctl names at this level are at least name and field */
1357 	if (namelen < 2)
1358 		return (ENOTDIR);		/* overloaded */
1359 
1360 	if (name[0] != VFS_GENERIC) {
1361 		vfsp = vfs_bytypenum(name[0]);
1362 		if (vfsp == NULL || vfsp->vfc_vfsops->vfs_sysctl == NULL)
1363 			return (EOPNOTSUPP);
1364 
1365 		return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
1366 		    oldp, oldlenp, newp, newlen, p));
1367 	}
1368 
1369 	switch (name[1]) {
1370 	case VFS_MAXTYPENUM:
1371 		return (sysctl_rdint(oldp, oldlenp, newp, maxvfsconf));
1372 
1373 	case VFS_CONF:
1374 		if (namelen < 3)
1375 			return (ENOTDIR);	/* overloaded */
1376 
1377 		vfsp = vfs_bytypenum(name[2]);
1378 		if (vfsp == NULL)
1379 			return (EOPNOTSUPP);
1380 
1381 		/* Make a copy, clear out kernel pointers */
1382 		tmpvfsp = malloc(sizeof(*tmpvfsp), M_TEMP, M_WAITOK|M_ZERO);
1383 		memcpy(tmpvfsp, vfsp, sizeof(*tmpvfsp));
1384 		tmpvfsp->vfc_vfsops = NULL;
1385 
1386 		ret = sysctl_rdstruct(oldp, oldlenp, newp, tmpvfsp,
1387 		    sizeof(struct vfsconf));
1388 
1389 		free(tmpvfsp, M_TEMP, sizeof(*tmpvfsp));
1390 		return (ret);
1391 	case VFS_BCACHESTAT:	/* buffer cache statistics */
1392 		ret = sysctl_rdstruct(oldp, oldlenp, newp, &bcstats,
1393 		    sizeof(struct bcachestats));
1394 		return(ret);
1395 	}
1396 	return (EOPNOTSUPP);
1397 }
1398 
1399 /*
1400  * Check to see if a filesystem is mounted on a block device.
1401  */
1402 int
1403 vfs_mountedon(struct vnode *vp)
1404 {
1405 	struct vnode *vq;
1406 	int error = 0;
1407 
1408 	if (vp->v_specmountpoint != NULL)
1409 		return (EBUSY);
1410 	if (vp->v_flag & VALIASED) {
1411 		SLIST_FOREACH(vq, vp->v_hashchain, v_specnext) {
1412 			if (vq->v_rdev != vp->v_rdev ||
1413 			    vq->v_type != vp->v_type)
1414 				continue;
1415 			if (vq->v_specmountpoint != NULL) {
1416 				error = EBUSY;
1417 				break;
1418 			}
1419 		}
1420 	}
1421 	return (error);
1422 }
1423 
1424 #ifdef NFSSERVER
1425 /*
1426  * Build hash lists of net addresses and hang them off the mount point.
1427  * Called by vfs_export() to set up the lists of export addresses.
1428  */
1429 int
1430 vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
1431     struct export_args *argp)
1432 {
1433 	struct netcred *np;
1434 	struct radix_node_head *rnh;
1435 	int nplen, i;
1436 	struct radix_node *rn;
1437 	struct sockaddr *saddr, *smask = NULL;
1438 	int error;
1439 
1440 	if (argp->ex_addrlen == 0) {
1441 		if (mp->mnt_flag & MNT_DEFEXPORTED)
1442 			return (EPERM);
1443 		np = &nep->ne_defexported;
1444 		/* fill in the kernel's ucred from userspace's xucred */
1445 		if ((error = crfromxucred(&np->netc_anon, &argp->ex_anon)))
1446 			return (error);
1447 		mp->mnt_flag |= MNT_DEFEXPORTED;
1448 		goto finish;
1449 	}
1450 	if (argp->ex_addrlen > MLEN || argp->ex_masklen > MLEN ||
1451 	    argp->ex_addrlen < 0 || argp->ex_masklen < 0)
1452 		return (EINVAL);
1453 	nplen = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
1454 	np = (struct netcred *)malloc(nplen, M_NETADDR, M_WAITOK|M_ZERO);
1455 	np->netc_len = nplen;
1456 	saddr = (struct sockaddr *)(np + 1);
1457 	error = copyin(argp->ex_addr, saddr, argp->ex_addrlen);
1458 	if (error)
1459 		goto out;
1460 	if (saddr->sa_len > argp->ex_addrlen)
1461 		saddr->sa_len = argp->ex_addrlen;
1462 	if (argp->ex_masklen) {
1463 		smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
1464 		error = copyin(argp->ex_mask, smask, argp->ex_masklen);
1465 		if (error)
1466 			goto out;
1467 		if (smask->sa_len > argp->ex_masklen)
1468 			smask->sa_len = argp->ex_masklen;
1469 	}
1470 	/* fill in the kernel's ucred from userspace's xucred */
1471 	if ((error = crfromxucred(&np->netc_anon, &argp->ex_anon)))
1472 		goto out;
1473 	i = saddr->sa_family;
1474 	switch (i) {
1475 	case AF_INET:
1476 		if ((rnh = nep->ne_rtable_inet) == NULL) {
1477 			if (!rn_inithead((void **)&nep->ne_rtable_inet,
1478 			    offsetof(struct sockaddr_in, sin_addr))) {
1479 				error = ENOBUFS;
1480 				goto out;
1481 			}
1482 			rnh = nep->ne_rtable_inet;
1483 		}
1484 		break;
1485 	default:
1486 		error = EINVAL;
1487 		goto out;
1488 	}
1489 	rn = rn_addroute(saddr, smask, rnh, np->netc_rnodes, 0);
1490 	if (rn == NULL || np != (struct netcred *)rn) { /* already exists */
1491 		error = EPERM;
1492 		goto out;
1493 	}
1494 finish:
1495 	np->netc_exflags = argp->ex_flags;
1496 	return (0);
1497 out:
1498 	free(np, M_NETADDR, np->netc_len);
1499 	return (error);
1500 }
1501 
1502 int
1503 vfs_free_netcred(struct radix_node *rn, void *w, u_int id)
1504 {
1505 	struct radix_node_head *rnh = (struct radix_node_head *)w;
1506 	struct netcred * np = (struct netcred *)rn;
1507 
1508 	rn_delete(rn->rn_key, rn->rn_mask, rnh, NULL);
1509 	free(np, M_NETADDR, np->netc_len);
1510 	return (0);
1511 }
1512 
1513 /*
1514  * Free the net address hash lists that are hanging off the mount points.
1515  */
1516 void
1517 vfs_free_addrlist(struct netexport *nep)
1518 {
1519 	struct radix_node_head *rnh;
1520 
1521 	if ((rnh = nep->ne_rtable_inet) != NULL) {
1522 		rn_walktree(rnh, vfs_free_netcred, rnh);
1523 		free(rnh, M_RTABLE, sizeof(*rnh));
1524 		nep->ne_rtable_inet = NULL;
1525 	}
1526 }
1527 #endif /* NFSSERVER */
1528 
1529 int
1530 vfs_export(struct mount *mp, struct netexport *nep, struct export_args *argp)
1531 {
1532 #ifdef NFSSERVER
1533 	int error;
1534 
1535 	if (argp->ex_flags & MNT_DELEXPORT) {
1536 		vfs_free_addrlist(nep);
1537 		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
1538 	}
1539 	if (argp->ex_flags & MNT_EXPORTED) {
1540 		if ((error = vfs_hang_addrlist(mp, nep, argp)) != 0)
1541 			return (error);
1542 		mp->mnt_flag |= MNT_EXPORTED;
1543 	}
1544 	return (0);
1545 #else
1546 	return (ENOTSUP);
1547 #endif /* NFSSERVER */
1548 }
1549 
1550 struct netcred *
1551 vfs_export_lookup(struct mount *mp, struct netexport *nep, struct mbuf *nam)
1552 {
1553 #ifdef NFSSERVER
1554 	struct netcred *np;
1555 	struct radix_node_head *rnh;
1556 	struct sockaddr *saddr;
1557 
1558 	np = NULL;
1559 	if (mp->mnt_flag & MNT_EXPORTED) {
1560 		/*
1561 		 * Lookup in the export list first.
1562 		 */
1563 		if (nam != NULL) {
1564 			saddr = mtod(nam, struct sockaddr *);
1565 			switch(saddr->sa_family) {
1566 			case AF_INET:
1567 				rnh = nep->ne_rtable_inet;
1568 				break;
1569 			default:
1570 				rnh = NULL;
1571 				break;
1572 			}
1573 			if (rnh != NULL)
1574 				np = (struct netcred *)rn_match(saddr, rnh);
1575 		}
1576 		/*
1577 		 * If no address match, use the default if it exists.
1578 		 */
1579 		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
1580 			np = &nep->ne_defexported;
1581 	}
1582 	return (np);
1583 #else
1584 	return (NULL);
1585 #endif /* NFSSERVER */
1586 }
1587 
1588 /*
1589  * Do the usual access checking.
1590  * file_mode, uid and gid are from the vnode in question,
1591  * while acc_mode and cred are from the VOP_ACCESS parameter list
1592  */
1593 int
1594 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid,
1595     mode_t acc_mode, struct ucred *cred)
1596 {
1597 	mode_t mask;
1598 
1599 	/* User id 0 always gets read/write access. */
1600 	if (cred->cr_uid == 0) {
1601 		/* For VEXEC, at least one of the execute bits must be set. */
1602 		if ((acc_mode & VEXEC) && type != VDIR &&
1603 		    (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)
1604 			return EACCES;
1605 		return 0;
1606 	}
1607 
1608 	mask = 0;
1609 
1610 	/* Otherwise, check the owner. */
1611 	if (cred->cr_uid == uid) {
1612 		if (acc_mode & VEXEC)
1613 			mask |= S_IXUSR;
1614 		if (acc_mode & VREAD)
1615 			mask |= S_IRUSR;
1616 		if (acc_mode & VWRITE)
1617 			mask |= S_IWUSR;
1618 		return (file_mode & mask) == mask ? 0 : EACCES;
1619 	}
1620 
1621 	/* Otherwise, check the groups. */
1622 	if (groupmember(gid, cred)) {
1623 		if (acc_mode & VEXEC)
1624 			mask |= S_IXGRP;
1625 		if (acc_mode & VREAD)
1626 			mask |= S_IRGRP;
1627 		if (acc_mode & VWRITE)
1628 			mask |= S_IWGRP;
1629 		return (file_mode & mask) == mask ? 0 : EACCES;
1630 	}
1631 
1632 	/* Otherwise, check everyone else. */
1633 	if (acc_mode & VEXEC)
1634 		mask |= S_IXOTH;
1635 	if (acc_mode & VREAD)
1636 		mask |= S_IROTH;
1637 	if (acc_mode & VWRITE)
1638 		mask |= S_IWOTH;
1639 	return (file_mode & mask) == mask ? 0 : EACCES;
1640 }
1641 
1642 int
1643 vnoperm(struct vnode *vp)
1644 {
1645 	if (vp->v_flag & VROOT || vp->v_mount == NULL)
1646 		return 0;
1647 
1648 	return (vp->v_mount->mnt_flag & MNT_NOPERM);
1649 }
1650 
1651 struct rwlock vfs_stall_lock = RWLOCK_INITIALIZER("vfs_stall");
1652 unsigned int vfs_stalling = 0;
1653 
1654 int
1655 vfs_stall(struct proc *p, int stall)
1656 {
1657 	struct mount *mp;
1658 	int allerror = 0, error;
1659 
1660 	if (stall) {
1661 		atomic_inc_int(&vfs_stalling);
1662 		rw_enter_write(&vfs_stall_lock);
1663 	}
1664 
1665 	/*
1666 	 * The loop variable mp is protected by vfs_busy() so that it cannot
1667 	 * be unmounted while VFS_SYNC() sleeps.  Traverse forward to keep the
1668 	 * lock order consistent with dounmount().
1669 	 */
1670 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
1671 		if (stall) {
1672 			error = vfs_busy(mp, VB_WRITE|VB_WAIT|VB_DUPOK);
1673 			if (error) {
1674 				printf("%s: busy\n", mp->mnt_stat.f_mntonname);
1675 				allerror = error;
1676 				continue;
1677 			}
1678 			uvm_vnp_sync(mp);
1679 			error = VFS_SYNC(mp, MNT_WAIT, stall, p->p_ucred, p);
1680 			if (error) {
1681 				printf("%s: failed to sync\n",
1682 				    mp->mnt_stat.f_mntonname);
1683 				vfs_unbusy(mp);
1684 				allerror = error;
1685 				continue;
1686 			}
1687 			mp->mnt_flag |= MNT_STALLED;
1688 		} else {
1689 			if (mp->mnt_flag & MNT_STALLED) {
1690 				vfs_unbusy(mp);
1691 				mp->mnt_flag &= ~MNT_STALLED;
1692 			}
1693 		}
1694 	}
1695 
1696 	if (!stall) {
1697 		rw_exit_write(&vfs_stall_lock);
1698 		atomic_dec_int(&vfs_stalling);
1699 	}
1700 
1701 	return (allerror);
1702 }
1703 
1704 void
1705 vfs_stall_barrier(void)
1706 {
1707 	if (__predict_false(vfs_stalling)) {
1708 		rw_enter_read(&vfs_stall_lock);
1709 		rw_exit_read(&vfs_stall_lock);
1710 	}
1711 }
1712 
1713 /*
1714  * Unmount all file systems.
1715  * We traverse the list in reverse order under the assumption that doing so
1716  * will avoid needing to worry about dependencies.
1717  */
1718 void
1719 vfs_unmountall(void)
1720 {
1721 	struct mount *mp, *nmp;
1722 	int allerror, error, again = 1;
1723 
1724  retry:
1725 	allerror = 0;
1726 	TAILQ_FOREACH_REVERSE_SAFE(mp, &mountlist, mntlist, mnt_list, nmp) {
1727 		if (vfs_busy(mp, VB_WRITE|VB_NOWAIT))
1728 			continue;
1729 		/* XXX Here is a race, the next pointer is not locked. */
1730 		if ((error = dounmount(mp, MNT_FORCE, curproc)) != 0) {
1731 			printf("unmount of %s failed with error %d\n",
1732 			    mp->mnt_stat.f_mntonname, error);
1733 			allerror = 1;
1734 		}
1735 	}
1736 
1737 	if (allerror) {
1738 		printf("WARNING: some file systems would not unmount\n");
1739 		if (again) {
1740 			printf("retrying\n");
1741 			again = 0;
1742 			goto retry;
1743 		}
1744 	}
1745 }
1746 
1747 /*
1748  * Sync and unmount file systems before shutting down.
1749  */
1750 void
1751 vfs_shutdown(struct proc *p)
1752 {
1753 #ifdef ACCOUNTING
1754 	acct_shutdown();
1755 #endif
1756 
1757 	printf("syncing disks...");
1758 
1759 	if (panicstr == NULL) {
1760 		/* Sync before unmount, in case we hang on something. */
1761 		sys_sync(p, NULL, NULL);
1762 		vfs_unmountall();
1763 	}
1764 
1765 #if NSOFTRAID > 0
1766 	sr_quiesce();
1767 #endif
1768 
1769 	if (vfs_syncwait(p, 1))
1770 		printf(" giving up\n");
1771 	else
1772 		printf(" done\n");
1773 }
1774 
1775 /*
1776  * perform sync() operation and wait for buffers to flush.
1777  */
1778 int
1779 vfs_syncwait(struct proc *p, int verbose)
1780 {
1781 	struct buf *bp;
1782 	int iter, nbusy, dcount, s;
1783 #ifdef MULTIPROCESSOR
1784 	int hold_count;
1785 #endif
1786 
1787 	sys_sync(p, NULL, NULL);
1788 
1789 	/* Wait for sync to finish. */
1790 	dcount = 10000;
1791 	for (iter = 0; iter < 20; iter++) {
1792 		nbusy = 0;
1793 		LIST_FOREACH(bp, &bufhead, b_list) {
1794 			if ((bp->b_flags & (B_BUSY|B_INVAL|B_READ)) == B_BUSY)
1795 				nbusy++;
1796 			/*
1797 			 * With soft updates, some buffers that are
1798 			 * written will be remarked as dirty until other
1799 			 * buffers are written.
1800 			 */
1801 			if (bp->b_flags & B_DELWRI) {
1802 				s = splbio();
1803 				bremfree(bp);
1804 				buf_acquire(bp);
1805 				splx(s);
1806 				nbusy++;
1807 				bawrite(bp);
1808 				if (dcount-- <= 0) {
1809 					if (verbose)
1810 						printf("softdep ");
1811 					return 1;
1812 				}
1813 			}
1814 		}
1815 		if (nbusy == 0)
1816 			break;
1817 		if (verbose)
1818 			printf("%d ", nbusy);
1819 #ifdef MULTIPROCESSOR
1820 		if (_kernel_lock_held())
1821 			hold_count = __mp_release_all(&kernel_lock);
1822 		else
1823 			hold_count = 0;
1824 #endif
1825 		DELAY(40000 * iter);
1826 #ifdef MULTIPROCESSOR
1827 		if (hold_count)
1828 			__mp_acquire_count(&kernel_lock, hold_count);
1829 #endif
1830 	}
1831 
1832 	return nbusy;
1833 }
1834 
1835 /*
1836  * posix file system related system variables.
1837  */
1838 int
1839 fs_posix_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp,
1840     void *newp, size_t newlen, struct proc *p)
1841 {
1842 	/* all sysctl names at this level are terminal */
1843 	if (namelen != 1)
1844 		return (ENOTDIR);
1845 
1846 	switch (name[0]) {
1847 	case FS_POSIX_SETUID:
1848 		if (newp && securelevel > 0)
1849 			return (EPERM);
1850 		return(sysctl_int(oldp, oldlenp, newp, newlen, &suid_clear));
1851 	default:
1852 		return (EOPNOTSUPP);
1853 	}
1854 	/* NOTREACHED */
1855 }
1856 
1857 /*
1858  * file system related system variables.
1859  */
1860 int
1861 fs_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp,
1862     size_t newlen, struct proc *p)
1863 {
1864 	sysctlfn *fn;
1865 
1866 	switch (name[0]) {
1867 	case FS_POSIX:
1868 		fn = fs_posix_sysctl;
1869 		break;
1870 	default:
1871 		return (EOPNOTSUPP);
1872 	}
1873 	return (*fn)(name + 1, namelen - 1, oldp, oldlenp, newp, newlen, p);
1874 }
1875 
1876 
1877 /*
1878  * Routines dealing with vnodes and buffers
1879  */
1880 
1881 /*
1882  * Wait for all outstanding I/Os to complete
1883  *
1884  * Manipulates v_numoutput. Must be called at splbio()
1885  */
1886 int
1887 vwaitforio(struct vnode *vp, int slpflag, char *wmesg, uint64_t timeo)
1888 {
1889 	int error = 0;
1890 
1891 	splassert(IPL_BIO);
1892 
1893 	while (vp->v_numoutput) {
1894 		vp->v_bioflag |= VBIOWAIT;
1895 		error = tsleep_nsec(&vp->v_numoutput,
1896 		    slpflag | (PRIBIO + 1), wmesg, timeo);
1897 		if (error)
1898 			break;
1899 	}
1900 
1901 	return (error);
1902 }
1903 
1904 /*
1905  * Update outstanding I/O count and do wakeup if requested.
1906  *
1907  * Manipulates v_numoutput. Must be called at splbio()
1908  */
1909 void
1910 vwakeup(struct vnode *vp)
1911 {
1912 	splassert(IPL_BIO);
1913 
1914 	if (vp != NULL) {
1915 		if (vp->v_numoutput-- == 0)
1916 			panic("vwakeup: neg numoutput");
1917 		if ((vp->v_bioflag & VBIOWAIT) && vp->v_numoutput == 0) {
1918 			vp->v_bioflag &= ~VBIOWAIT;
1919 			wakeup(&vp->v_numoutput);
1920 		}
1921 	}
1922 }
1923 
1924 /*
1925  * Flush out and invalidate all buffers associated with a vnode.
1926  * Called with the underlying object locked.
1927  */
1928 int
1929 vinvalbuf(struct vnode *vp, int flags, struct ucred *cred, struct proc *p,
1930     int slpflag, uint64_t slptimeo)
1931 {
1932 	struct buf *bp;
1933 	struct buf *nbp, *blist;
1934 	int s, error;
1935 
1936 #ifdef VFSLCKDEBUG
1937 	if ((vp->v_flag & VLOCKSWORK) && !VOP_ISLOCKED(vp))
1938 		panic("%s: vp isn't locked, vp %p", __func__, vp);
1939 #endif
1940 
1941 	if (flags & V_SAVE) {
1942 		s = splbio();
1943 		vwaitforio(vp, 0, "vinvalbuf", INFSLP);
1944 		if (!LIST_EMPTY(&vp->v_dirtyblkhd)) {
1945 			splx(s);
1946 			if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, p)) != 0)
1947 				return (error);
1948 			s = splbio();
1949 			if (vp->v_numoutput > 0 ||
1950 			    !LIST_EMPTY(&vp->v_dirtyblkhd))
1951 				panic("%s: dirty bufs, vp %p", __func__, vp);
1952 		}
1953 		splx(s);
1954 	}
1955 loop:
1956 	s = splbio();
1957 	for (;;) {
1958 		int count = 0;
1959 		if ((blist = LIST_FIRST(&vp->v_cleanblkhd)) &&
1960 		    (flags & V_SAVEMETA))
1961 			while (blist && blist->b_lblkno < 0)
1962 				blist = LIST_NEXT(blist, b_vnbufs);
1963 		if (blist == NULL &&
1964 		    (blist = LIST_FIRST(&vp->v_dirtyblkhd)) &&
1965 		    (flags & V_SAVEMETA))
1966 			while (blist && blist->b_lblkno < 0)
1967 				blist = LIST_NEXT(blist, b_vnbufs);
1968 		if (!blist)
1969 			break;
1970 
1971 		for (bp = blist; bp; bp = nbp) {
1972 			nbp = LIST_NEXT(bp, b_vnbufs);
1973 			if (flags & V_SAVEMETA && bp->b_lblkno < 0)
1974 				continue;
1975 			if (bp->b_flags & B_BUSY) {
1976 				bp->b_flags |= B_WANTED;
1977 				error = tsleep_nsec(bp, slpflag | (PRIBIO + 1),
1978 				    "vinvalbuf", slptimeo);
1979 				if (error) {
1980 					splx(s);
1981 					return (error);
1982 				}
1983 				break;
1984 			}
1985 			bremfree(bp);
1986 			/*
1987 			 * XXX Since there are no node locks for NFS, I believe
1988 			 * there is a slight chance that a delayed write will
1989 			 * occur while sleeping just above, so check for it.
1990 			 */
1991 			if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) {
1992 				buf_acquire(bp);
1993 				splx(s);
1994 				(void) VOP_BWRITE(bp);
1995 				goto loop;
1996 			}
1997 			buf_acquire_nomap(bp);
1998 			bp->b_flags |= B_INVAL;
1999 			brelse(bp);
2000 			count++;
2001 			/*
2002 			 * XXX Temporary workaround XXX
2003 			 *
2004 			 * If this is a gigantisch vnode and we are
2005 			 * trashing a ton of buffers, drop the lock
2006 			 * and yield every so often. The longer term
2007 			 * fix is to add a separate list for these
2008 			 * invalid buffers so we don't have to do the
2009 			 * work to free these here.
2010 			 */
2011 			if (count > 100) {
2012 				splx(s);
2013 				sched_pause(yield);
2014 				goto loop;
2015 			}
2016 		}
2017 	}
2018 	if (!(flags & V_SAVEMETA) &&
2019 	    (!LIST_EMPTY(&vp->v_dirtyblkhd) || !LIST_EMPTY(&vp->v_cleanblkhd)))
2020 		panic("%s: flush failed, vp %p", __func__, vp);
2021 	splx(s);
2022 	return (0);
2023 }
2024 
2025 void
2026 vflushbuf(struct vnode *vp, int sync)
2027 {
2028 	struct buf *bp, *nbp;
2029 	int s;
2030 
2031 loop:
2032 	s = splbio();
2033 	LIST_FOREACH_SAFE(bp, &vp->v_dirtyblkhd, b_vnbufs, nbp) {
2034 		if ((bp->b_flags & B_BUSY))
2035 			continue;
2036 		if ((bp->b_flags & B_DELWRI) == 0)
2037 			panic("vflushbuf: not dirty");
2038 		bremfree(bp);
2039 		buf_acquire(bp);
2040 		splx(s);
2041 		/*
2042 		 * Wait for I/O associated with indirect blocks to complete,
2043 		 * since there is no way to quickly wait for them below.
2044 		 */
2045 		if (bp->b_vp == vp || sync == 0)
2046 			(void) bawrite(bp);
2047 		else
2048 			(void) bwrite(bp);
2049 		goto loop;
2050 	}
2051 	if (sync == 0) {
2052 		splx(s);
2053 		return;
2054 	}
2055 	vwaitforio(vp, 0, "vflushbuf", INFSLP);
2056 	if (!LIST_EMPTY(&vp->v_dirtyblkhd)) {
2057 		splx(s);
2058 #ifdef DIAGNOSTIC
2059 		vprint("vflushbuf: dirty", vp);
2060 #endif
2061 		goto loop;
2062 	}
2063 	splx(s);
2064 }
2065 
2066 /*
2067  * Associate a buffer with a vnode.
2068  *
2069  * Manipulates buffer vnode queues. Must be called at splbio().
2070  */
2071 void
2072 bgetvp(struct vnode *vp, struct buf *bp)
2073 {
2074 	splassert(IPL_BIO);
2075 
2076 
2077 	if (bp->b_vp)
2078 		panic("bgetvp: not free");
2079 	vhold(vp);
2080 	bp->b_vp = vp;
2081 	if (vp->v_type == VBLK || vp->v_type == VCHR)
2082 		bp->b_dev = vp->v_rdev;
2083 	else
2084 		bp->b_dev = NODEV;
2085 	/*
2086 	 * Insert onto list for new vnode.
2087 	 */
2088 	bufinsvn(bp, &vp->v_cleanblkhd);
2089 }
2090 
2091 /*
2092  * Disassociate a buffer from a vnode.
2093  *
2094  * Manipulates vnode buffer queues. Must be called at splbio().
2095  */
2096 void
2097 brelvp(struct buf *bp)
2098 {
2099 	struct vnode *vp;
2100 
2101 	splassert(IPL_BIO);
2102 
2103 	if ((vp = bp->b_vp) == (struct vnode *) 0)
2104 		panic("brelvp: NULL");
2105 	/*
2106 	 * Delete from old vnode list, if on one.
2107 	 */
2108 	if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
2109 		bufremvn(bp);
2110 	if ((vp->v_bioflag & VBIOONSYNCLIST) &&
2111 	    LIST_EMPTY(&vp->v_dirtyblkhd)) {
2112 		vp->v_bioflag &= ~VBIOONSYNCLIST;
2113 		LIST_REMOVE(vp, v_synclist);
2114 	}
2115 	bp->b_vp = NULL;
2116 
2117 	vdrop(vp);
2118 }
2119 
2120 /*
2121  * Replaces the current vnode associated with the buffer, if any,
2122  * with a new vnode.
2123  *
2124  * If an output I/O is pending on the buffer, the old vnode
2125  * I/O count is adjusted.
2126  *
2127  * Ignores vnode buffer queues. Must be called at splbio().
2128  */
2129 void
2130 buf_replacevnode(struct buf *bp, struct vnode *newvp)
2131 {
2132 	struct vnode *oldvp = bp->b_vp;
2133 
2134 	splassert(IPL_BIO);
2135 
2136 	if (oldvp)
2137 		brelvp(bp);
2138 
2139 	if ((bp->b_flags & (B_READ | B_DONE)) == 0) {
2140 		newvp->v_numoutput++;	/* put it on swapdev */
2141 		vwakeup(oldvp);
2142 	}
2143 
2144 	bgetvp(newvp, bp);
2145 	bufremvn(bp);
2146 }
2147 
2148 /*
2149  * Used to assign buffers to the appropriate clean or dirty list on
2150  * the vnode and to add newly dirty vnodes to the appropriate
2151  * filesystem syncer list.
2152  *
2153  * Manipulates vnode buffer queues. Must be called at splbio().
2154  */
2155 void
2156 reassignbuf(struct buf *bp)
2157 {
2158 	struct buflists *listheadp;
2159 	int delay;
2160 	struct vnode *vp = bp->b_vp;
2161 
2162 	splassert(IPL_BIO);
2163 
2164 	/*
2165 	 * Delete from old vnode list, if on one.
2166 	 */
2167 	if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
2168 		bufremvn(bp);
2169 
2170 	/*
2171 	 * If dirty, put on list of dirty buffers;
2172 	 * otherwise insert onto list of clean buffers.
2173 	 */
2174 	if ((bp->b_flags & B_DELWRI) == 0) {
2175 		listheadp = &vp->v_cleanblkhd;
2176 		if ((vp->v_bioflag & VBIOONSYNCLIST) &&
2177 		    LIST_EMPTY(&vp->v_dirtyblkhd)) {
2178 			vp->v_bioflag &= ~VBIOONSYNCLIST;
2179 			LIST_REMOVE(vp, v_synclist);
2180 		}
2181 	} else {
2182 		listheadp = &vp->v_dirtyblkhd;
2183 		if ((vp->v_bioflag & VBIOONSYNCLIST) == 0) {
2184 			switch (vp->v_type) {
2185 			case VDIR:
2186 				delay = syncdelay / 2;
2187 				break;
2188 			case VBLK:
2189 				if (vp->v_specmountpoint != NULL) {
2190 					delay = syncdelay / 3;
2191 					break;
2192 				}
2193 				/* FALLTHROUGH */
2194 			default:
2195 				delay = syncdelay;
2196 			}
2197 			vn_syncer_add_to_worklist(vp, delay);
2198 		}
2199 	}
2200 	bufinsvn(bp, listheadp);
2201 }
2202 
2203 /*
2204  * Check if vnode represents a disk device
2205  */
2206 int
2207 vn_isdisk(struct vnode *vp, int *errp)
2208 {
2209 	if (vp->v_type != VBLK && vp->v_type != VCHR)
2210 		return (0);
2211 
2212 	return (1);
2213 }
2214 
2215 #ifdef DDB
2216 #include <machine/db_machdep.h>
2217 #include <ddb/db_interface.h>
2218 
2219 void
2220 vfs_buf_print(void *b, int full,
2221     int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2222 {
2223 	struct buf *bp = b;
2224 
2225 	(*pr)("  vp %p lblkno 0x%llx blkno 0x%llx dev 0x%x\n"
2226 	      "  proc %p error %d flags %lb\n",
2227 	    bp->b_vp, (int64_t)bp->b_lblkno, (int64_t)bp->b_blkno, bp->b_dev,
2228 	    bp->b_proc, bp->b_error, bp->b_flags, B_BITS);
2229 
2230 	(*pr)("  bufsize 0x%lx bcount 0x%lx resid 0x%lx\n"
2231 	      "  data %p saveaddr %p dep %p iodone %p\n",
2232 	    bp->b_bufsize, bp->b_bcount, (long)bp->b_resid,
2233 	    bp->b_data, bp->b_saveaddr,
2234 	    LIST_FIRST(&bp->b_dep), bp->b_iodone);
2235 
2236 	(*pr)("  dirty {off 0x%x end 0x%x} valid {off 0x%x end 0x%x}\n",
2237 	    bp->b_dirtyoff, bp->b_dirtyend, bp->b_validoff, bp->b_validend);
2238 
2239 #ifdef FFS_SOFTUPDATES
2240 	if (full)
2241 		softdep_print(bp, full, pr);
2242 #endif
2243 }
2244 
2245 const char *vtypes[] = { VTYPE_NAMES };
2246 const char *vtags[] = { VTAG_NAMES };
2247 
2248 void
2249 vfs_vnode_print(void *v, int full,
2250     int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2251 {
2252 	struct vnode *vp = v;
2253 
2254 	(*pr)("tag %s(%d) type %s(%d) mount %p typedata %p\n",
2255 	      (u_int)vp->v_tag >= nitems(vtags)? "<unk>":vtags[vp->v_tag],
2256 	      vp->v_tag,
2257 	      (u_int)vp->v_type >= nitems(vtypes)? "<unk>":vtypes[vp->v_type],
2258 	      vp->v_type, vp->v_mount, vp->v_mountedhere);
2259 
2260 	(*pr)("data %p usecount %d writecount %d holdcnt %d numoutput %d\n",
2261 	      vp->v_data, vp->v_usecount, vp->v_writecount,
2262 	      vp->v_holdcnt, vp->v_numoutput);
2263 
2264 	/* uvm_object_printit(&vp->v_uobj, full, pr); */
2265 
2266 	if (full) {
2267 		struct buf *bp;
2268 
2269 		(*pr)("clean bufs:\n");
2270 		LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) {
2271 			(*pr)(" bp %p\n", bp);
2272 			vfs_buf_print(bp, full, pr);
2273 		}
2274 
2275 		(*pr)("dirty bufs:\n");
2276 		LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) {
2277 			(*pr)(" bp %p\n", bp);
2278 			vfs_buf_print(bp, full, pr);
2279 		}
2280 	}
2281 }
2282 
2283 void
2284 vfs_mount_print(struct mount *mp, int full,
2285     int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2))))
2286 {
2287 	struct vfsconf *vfc = mp->mnt_vfc;
2288 	struct vnode *vp;
2289 	int cnt;
2290 
2291 	(*pr)("flags %b\nvnodecovered %p syncer %p data %p\n",
2292 	    mp->mnt_flag, MNT_BITS,
2293 	    mp->mnt_vnodecovered, mp->mnt_syncer, mp->mnt_data);
2294 
2295 	(*pr)("vfsconf: ops %p name \"%s\" num %d ref %u flags 0x%x\n",
2296 	    vfc->vfc_vfsops, vfc->vfc_name, vfc->vfc_typenum,
2297 	    vfc->vfc_refcount, vfc->vfc_flags);
2298 
2299 	(*pr)("statvfs cache: bsize %x iosize %x\n"
2300 	    "blocks %llu free %llu avail %lld\n",
2301 	    mp->mnt_stat.f_bsize, mp->mnt_stat.f_iosize, mp->mnt_stat.f_blocks,
2302 	    mp->mnt_stat.f_bfree, mp->mnt_stat.f_bavail);
2303 
2304 	(*pr)("  files %llu ffiles %llu favail %lld\n", mp->mnt_stat.f_files,
2305 	    mp->mnt_stat.f_ffree, mp->mnt_stat.f_favail);
2306 
2307 	(*pr)("  f_fsidx {0x%x, 0x%x} owner %u ctime 0x%llx\n",
2308 	    mp->mnt_stat.f_fsid.val[0], mp->mnt_stat.f_fsid.val[1],
2309 	    mp->mnt_stat.f_owner, mp->mnt_stat.f_ctime);
2310 
2311 	(*pr)("  syncwrites %llu asyncwrites = %llu\n",
2312 	    mp->mnt_stat.f_syncwrites, mp->mnt_stat.f_asyncwrites);
2313 
2314 	(*pr)("  syncreads %llu asyncreads = %llu\n",
2315 	    mp->mnt_stat.f_syncreads, mp->mnt_stat.f_asyncreads);
2316 
2317 	(*pr)("  fstype \"%s\" mnton \"%s\" mntfrom \"%s\" mntspec \"%s\"\n",
2318 	    mp->mnt_stat.f_fstypename, mp->mnt_stat.f_mntonname,
2319 	    mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntfromspec);
2320 
2321 	(*pr)("locked vnodes:");
2322 	/* XXX would take mountlist lock, except ddb has no context */
2323 	cnt = 0;
2324 	TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2325 		if (VOP_ISLOCKED(vp)) {
2326 			if (cnt == 0)
2327 				(*pr)("\n  %p", vp);
2328 			else if ((cnt % (72 / (sizeof(void *) * 2 + 4))) == 0)
2329 				(*pr)(",\n  %p", vp);
2330 			else
2331 				(*pr)(", %p", vp);
2332 			cnt++;
2333 		}
2334 	}
2335 	(*pr)("\n");
2336 
2337 	if (full) {
2338 		(*pr)("all vnodes:");
2339 		/* XXX would take mountlist lock, except ddb has no context */
2340 		cnt = 0;
2341 		TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2342 			if (cnt == 0)
2343 				(*pr)("\n  %p", vp);
2344 			else if ((cnt % (72 / (sizeof(void *) * 2 + 4))) == 0)
2345 				(*pr)(",\n  %p", vp);
2346 			else
2347 				(*pr)(", %p", vp);
2348 			cnt++;
2349 		}
2350 		(*pr)("\n");
2351 	}
2352 }
2353 #endif /* DDB */
2354 
2355 void
2356 copy_statfs_info(struct statfs *sbp, const struct mount *mp)
2357 {
2358 	const struct statfs *mbp;
2359 
2360 	strncpy(sbp->f_fstypename, mp->mnt_vfc->vfc_name, MFSNAMELEN);
2361 
2362 	if (sbp == (mbp = &mp->mnt_stat))
2363 		return;
2364 
2365 	sbp->f_fsid = mbp->f_fsid;
2366 	sbp->f_owner = mbp->f_owner;
2367 	sbp->f_flags = mbp->f_flags;
2368 	sbp->f_syncwrites = mbp->f_syncwrites;
2369 	sbp->f_asyncwrites = mbp->f_asyncwrites;
2370 	sbp->f_syncreads = mbp->f_syncreads;
2371 	sbp->f_asyncreads = mbp->f_asyncreads;
2372 	sbp->f_namemax = mbp->f_namemax;
2373 	memcpy(sbp->f_mntonname, mp->mnt_stat.f_mntonname, MNAMELEN);
2374 	memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname, MNAMELEN);
2375 	memcpy(sbp->f_mntfromspec, mp->mnt_stat.f_mntfromspec, MNAMELEN);
2376 	memcpy(&sbp->mount_info, &mp->mnt_stat.mount_info,
2377 	    sizeof(union mount_info));
2378 }
2379