xref: /openbsd-src/sys/kern/vfs_subr.c (revision c1a45aed656e7d5627c30c92421893a76f370ccb)
1 /*	$OpenBSD: vfs_subr.c,v 1.315 2022/03/27 16:19:39 semarie 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 TAILQ_HEAD(freelst, vnode);
102 struct freelst vnode_hold_list;	/* list of vnodes referencing buffers */
103 struct freelst vnode_free_list;	/* vnode free list */
104 
105 struct mntlist mountlist;	/* mounted filesystem list */
106 
107 void	vclean(struct vnode *, int, struct proc *);
108 
109 void insmntque(struct vnode *, struct mount *);
110 int getdevvp(dev_t, struct vnode **, enum vtype);
111 
112 int vfs_hang_addrlist(struct mount *, struct netexport *,
113 				  struct export_args *);
114 int vfs_free_netcred(struct radix_node *, void *, u_int);
115 void vfs_free_addrlist(struct netexport *);
116 void vputonfreelist(struct vnode *);
117 
118 int vflush_vnode(struct vnode *, void *);
119 int maxvnodes;
120 
121 struct mutex vnode_mtx = MUTEX_INITIALIZER(IPL_BIO);
122 
123 void vfs_unmountall(void);
124 
125 #ifdef DEBUG
126 void printlockedvnodes(void);
127 #endif
128 
129 struct pool vnode_pool;
130 struct pool uvm_vnode_pool;
131 
132 static inline int rb_buf_compare(const struct buf *b1, const struct buf *b2);
133 RBT_GENERATE(buf_rb_bufs, buf, b_rbbufs, rb_buf_compare);
134 
135 static inline int
136 rb_buf_compare(const struct buf *b1, const struct buf *b2)
137 {
138 	if (b1->b_lblkno < b2->b_lblkno)
139 		return(-1);
140 	if (b1->b_lblkno > b2->b_lblkno)
141 		return(1);
142 	return(0);
143 }
144 
145 /*
146  * Initialize the vnode management data structures.
147  */
148 void
149 vntblinit(void)
150 {
151 	/* buffer cache may need a vnode for each buffer */
152 	maxvnodes = 2 * initialvnodes;
153 	pool_init(&vnode_pool, sizeof(struct vnode), 0, IPL_NONE,
154 	    PR_WAITOK, "vnodes", NULL);
155 	pool_init(&uvm_vnode_pool, sizeof(struct uvm_vnode), 0, IPL_NONE,
156 	    PR_WAITOK, "uvmvnodes", NULL);
157 	TAILQ_INIT(&vnode_hold_list);
158 	TAILQ_INIT(&vnode_free_list);
159 	TAILQ_INIT(&mountlist);
160 	/*
161 	 * Initialize the filesystem syncer.
162 	 */
163 	vn_initialize_syncerd();
164 
165 #ifdef NFSSERVER
166 	rn_init(sizeof(struct sockaddr_in));
167 #endif /* NFSSERVER */
168 }
169 
170 /*
171  * Allocate a mount point.
172  *
173  * The returned mount point is marked as busy.
174  */
175 struct mount *
176 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp)
177 {
178 	struct mount *mp;
179 
180 	mp = malloc(sizeof(*mp), M_MOUNT, M_WAITOK|M_ZERO);
181 	rw_init_flags(&mp->mnt_lock, "vfslock", RWL_IS_VNODE);
182 	(void)vfs_busy(mp, VB_READ|VB_NOWAIT);
183 
184 	TAILQ_INIT(&mp->mnt_vnodelist);
185 	mp->mnt_vnodecovered = vp;
186 
187 	atomic_inc_int(&vfsp->vfc_refcount);
188 	mp->mnt_vfc = vfsp;
189 	mp->mnt_op = vfsp->vfc_vfsops;
190 	mp->mnt_flag = vfsp->vfc_flags;
191 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
192 
193 	return (mp);
194 }
195 
196 /*
197  * Release a mount point.
198  */
199 void
200 vfs_mount_free(struct mount *mp)
201 {
202 	atomic_dec_int(&mp->mnt_vfc->vfc_refcount);
203 	free(mp, M_MOUNT, sizeof(*mp));
204 }
205 
206 /*
207  * Mark a mount point as busy. Used to synchronize access and to delay
208  * unmounting.
209  *
210  * Default behaviour is to attempt getting a READ lock and in case of an
211  * ongoing unmount, to wait for it to finish and then return failure.
212  */
213 int
214 vfs_busy(struct mount *mp, int flags)
215 {
216 	int rwflags = 0;
217 
218 	if (flags & VB_WRITE)
219 		rwflags |= RW_WRITE;
220 	else
221 		rwflags |= RW_READ;
222 
223 	if (flags & VB_WAIT)
224 		rwflags |= RW_SLEEPFAIL;
225 	else
226 		rwflags |= RW_NOSLEEP;
227 
228 #ifdef WITNESS
229 	if (flags & VB_DUPOK)
230 		rwflags |= RW_DUPOK;
231 #endif
232 
233 	if (rw_enter(&mp->mnt_lock, rwflags))
234 		return (EBUSY);
235 
236 	return (0);
237 }
238 
239 /*
240  * Free a busy file system
241  */
242 void
243 vfs_unbusy(struct mount *mp)
244 {
245 	rw_exit(&mp->mnt_lock);
246 }
247 
248 int
249 vfs_isbusy(struct mount *mp)
250 {
251 	if (RWLOCK_OWNER(&mp->mnt_lock) > 0)
252 		return (1);
253 	else
254 		return (0);
255 }
256 
257 /*
258  * Lookup a filesystem type, and if found allocate and initialize
259  * a mount structure for it.
260  *
261  * Devname is usually updated by mount(8) after booting.
262  */
263 int
264 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
265 {
266 	struct vfsconf *vfsp;
267 	struct mount *mp;
268 
269 	vfsp = vfs_byname(fstypename);
270 	if (vfsp == NULL)
271 		return (ENODEV);
272 	mp = vfs_mount_alloc(NULLVP, vfsp);
273 	mp->mnt_flag |= MNT_RDONLY;
274 	mp->mnt_stat.f_mntonname[0] = '/';
275 	copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN, NULL);
276 	copystr(devname, mp->mnt_stat.f_mntfromspec, MNAMELEN, NULL);
277 	*mpp = mp;
278 	return (0);
279  }
280 
281 /*
282  * Lookup a mount point by filesystem identifier.
283  */
284 struct mount *
285 vfs_getvfs(fsid_t *fsid)
286 {
287 	struct mount *mp;
288 
289 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
290 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
291 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
292 			return (mp);
293 		}
294 	}
295 
296 	return (NULL);
297 }
298 
299 
300 /*
301  * Get a new unique fsid
302  */
303 void
304 vfs_getnewfsid(struct mount *mp)
305 {
306 	static u_short xxxfs_mntid;
307 
308 	fsid_t tfsid;
309 	int mtype;
310 
311 	mtype = mp->mnt_vfc->vfc_typenum;
312 	mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev + mtype, 0);
313 	mp->mnt_stat.f_fsid.val[1] = mtype;
314 	if (xxxfs_mntid == 0)
315 		++xxxfs_mntid;
316 	tfsid.val[0] = makedev(nblkdev + mtype, xxxfs_mntid);
317 	tfsid.val[1] = mtype;
318 	if (!TAILQ_EMPTY(&mountlist)) {
319 		while (vfs_getvfs(&tfsid)) {
320 			tfsid.val[0]++;
321 			xxxfs_mntid++;
322 		}
323 	}
324 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
325 }
326 
327 /*
328  * Set vnode attributes to VNOVAL
329  */
330 void
331 vattr_null(struct vattr *vap)
332 {
333 
334 	vap->va_type = VNON;
335 	/*
336 	 * Don't get fancy: u_quad_t = u_int = VNOVAL leaves the u_quad_t
337 	 * with 2^31-1 instead of 2^64-1.  Just write'm out and let
338 	 * the compiler do its job.
339 	 */
340 	vap->va_mode = VNOVAL;
341 	vap->va_nlink = VNOVAL;
342 	vap->va_uid = VNOVAL;
343 	vap->va_gid = VNOVAL;
344 	vap->va_fsid = VNOVAL;
345 	vap->va_fileid = VNOVAL;
346 	vap->va_size = VNOVAL;
347 	vap->va_blocksize = VNOVAL;
348 	vap->va_atime.tv_sec = VNOVAL;
349 	vap->va_atime.tv_nsec = VNOVAL;
350 	vap->va_mtime.tv_sec = VNOVAL;
351 	vap->va_mtime.tv_nsec = VNOVAL;
352 	vap->va_ctime.tv_sec = VNOVAL;
353 	vap->va_ctime.tv_nsec = VNOVAL;
354 	vap->va_gen = VNOVAL;
355 	vap->va_flags = VNOVAL;
356 	vap->va_rdev = VNOVAL;
357 	vap->va_bytes = VNOVAL;
358 	vap->va_filerev = VNOVAL;
359 	vap->va_vaflags = 0;
360 }
361 
362 /*
363  * Routines having to do with the management of the vnode table.
364  */
365 long numvnodes;
366 
367 /*
368  * Return the next vnode from the free list.
369  */
370 int
371 getnewvnode(enum vtagtype tag, struct mount *mp, const struct vops *vops,
372     struct vnode **vpp)
373 {
374 	struct proc *p = curproc;
375 	struct freelst *listhd;
376 	static int toggle;
377 	struct vnode *vp;
378 	int s;
379 
380 	/*
381 	 * allow maxvnodes to increase if the buffer cache itself
382 	 * is big enough to justify it. (we don't shrink it ever)
383 	 */
384 	maxvnodes = maxvnodes < bcstats.numbufs ? bcstats.numbufs
385 	    : maxvnodes;
386 
387 	/*
388 	 * We must choose whether to allocate a new vnode or recycle an
389 	 * existing one. The criterion for allocating a new one is that
390 	 * the total number of vnodes is less than the number desired or
391 	 * there are no vnodes on either free list. Generally we only
392 	 * want to recycle vnodes that have no buffers associated with
393 	 * them, so we look first on the vnode_free_list. If it is empty,
394 	 * we next consider vnodes with referencing buffers on the
395 	 * vnode_hold_list. The toggle ensures that half the time we
396 	 * will use a buffer from the vnode_hold_list, and half the time
397 	 * we will allocate a new one unless the list has grown to twice
398 	 * the desired size. We are reticent to recycle vnodes from the
399 	 * vnode_hold_list because we will lose the identity of all its
400 	 * referencing buffers.
401 	 */
402 	toggle ^= 1;
403 	if (numvnodes / 2 > maxvnodes)
404 		toggle = 0;
405 
406 	s = splbio();
407 	if ((numvnodes < maxvnodes) ||
408 	    ((TAILQ_FIRST(listhd = &vnode_free_list) == NULL) &&
409 	    ((TAILQ_FIRST(listhd = &vnode_hold_list) == NULL) || toggle))) {
410 		splx(s);
411 		vp = pool_get(&vnode_pool, PR_WAITOK | PR_ZERO);
412 		vp->v_uvm = pool_get(&uvm_vnode_pool, PR_WAITOK | PR_ZERO);
413 		vp->v_uvm->u_vnode = vp;
414 		uvm_obj_init(&vp->v_uvm->u_obj, &uvm_vnodeops, 0);
415 		RBT_INIT(buf_rb_bufs, &vp->v_bufs_tree);
416 		cache_tree_init(&vp->v_nc_tree);
417 		TAILQ_INIT(&vp->v_cache_dst);
418 		numvnodes++;
419 	} else {
420 		TAILQ_FOREACH(vp, listhd, v_freelist) {
421 			if (VOP_ISLOCKED(vp) == 0)
422 				break;
423 		}
424 		/*
425 		 * Unless this is a bad time of the month, at most
426 		 * the first NCPUS items on the free list are
427 		 * locked, so this is close enough to being empty.
428 		 */
429 		if (vp == NULL) {
430 			splx(s);
431 			tablefull("vnode");
432 			*vpp = NULL;
433 			return (ENFILE);
434 		}
435 
436 #ifdef DIAGNOSTIC
437 		if (vp->v_usecount) {
438 			vprint("free vnode", vp);
439 			panic("free vnode isn't");
440 		}
441 #endif
442 
443 		TAILQ_REMOVE(listhd, vp, v_freelist);
444 		vp->v_bioflag &= ~VBIOONFREELIST;
445 		splx(s);
446 
447 		if (vp->v_type != VBAD)
448 			vgonel(vp, p);
449 #ifdef DIAGNOSTIC
450 		if (vp->v_data) {
451 			vprint("cleaned vnode", vp);
452 			panic("cleaned vnode isn't");
453 		}
454 		s = splbio();
455 		if (vp->v_numoutput)
456 			panic("Clean vnode has pending I/O's");
457 		splx(s);
458 #endif
459 		vp->v_flag = 0;
460 		vp->v_socket = NULL;
461 	}
462 	cache_purge(vp);
463 	vp->v_type = VNON;
464 	vp->v_tag = tag;
465 	vp->v_op = vops;
466 	insmntque(vp, mp);
467 	*vpp = vp;
468 	vp->v_usecount = 1;
469 	vp->v_data = NULL;
470 	return (0);
471 }
472 
473 /*
474  * Move a vnode from one mount queue to another.
475  */
476 void
477 insmntque(struct vnode *vp, struct mount *mp)
478 {
479 	/*
480 	 * Delete from old mount point vnode list, if on one.
481 	 */
482 	if (vp->v_mount != NULL)
483 		TAILQ_REMOVE(&vp->v_mount->mnt_vnodelist, vp, v_mntvnodes);
484 	/*
485 	 * Insert into list of vnodes for the new mount point, if available.
486 	 */
487 	if ((vp->v_mount = mp) != NULL)
488 		TAILQ_INSERT_TAIL(&mp->mnt_vnodelist, vp, v_mntvnodes);
489 }
490 
491 /*
492  * Create a vnode for a block device.
493  * Used for root filesystem, argdev, and swap areas.
494  * Also used for memory file system special devices.
495  */
496 int
497 bdevvp(dev_t dev, struct vnode **vpp)
498 {
499 	return (getdevvp(dev, vpp, VBLK));
500 }
501 
502 /*
503  * Create a vnode for a character device.
504  * Used for console handling.
505  */
506 int
507 cdevvp(dev_t dev, struct vnode **vpp)
508 {
509 	return (getdevvp(dev, vpp, VCHR));
510 }
511 
512 /*
513  * Create a vnode for a device.
514  * Used by bdevvp (block device) for root file system etc.,
515  * and by cdevvp (character device) for console.
516  */
517 int
518 getdevvp(dev_t dev, struct vnode **vpp, enum vtype type)
519 {
520 	struct vnode *vp;
521 	struct vnode *nvp;
522 	int error;
523 
524 	if (dev == NODEV) {
525 		*vpp = NULLVP;
526 		return (0);
527 	}
528 	error = getnewvnode(VT_NON, NULL, &spec_vops, &nvp);
529 	if (error) {
530 		*vpp = NULLVP;
531 		return (error);
532 	}
533 	vp = nvp;
534 	vp->v_type = type;
535 	if ((nvp = checkalias(vp, dev, NULL)) != NULL) {
536 		vput(vp);
537 		vp = nvp;
538 	}
539 	if (vp->v_type == VCHR && cdevsw[major(vp->v_rdev)].d_type == D_TTY)
540 		vp->v_flag |= VISTTY;
541 	*vpp = vp;
542 	return (0);
543 }
544 
545 /*
546  * Check to see if the new vnode represents a special device
547  * for which we already have a vnode (either because of
548  * bdevvp() or because of a different vnode representing
549  * the same block device). If such an alias exists, deallocate
550  * the existing contents and return the aliased vnode. The
551  * caller is responsible for filling it with its new contents.
552  */
553 struct vnode *
554 checkalias(struct vnode *nvp, dev_t nvp_rdev, struct mount *mp)
555 {
556 	struct proc *p = curproc;
557 	struct vnode *vp;
558 	struct vnodechain *vchain;
559 
560 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
561 		return (NULLVP);
562 
563 	vchain = &speclisth[SPECHASH(nvp_rdev)];
564 loop:
565 	SLIST_FOREACH(vp, vchain, v_specnext) {
566 		if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type) {
567 			continue;
568 		}
569 		/*
570 		 * Alias, but not in use, so flush it out.
571 		 */
572 		if (vp->v_usecount == 0) {
573 			vgonel(vp, p);
574 			goto loop;
575 		}
576 		if (vget(vp, LK_EXCLUSIVE)) {
577 			goto loop;
578 		}
579 		break;
580 	}
581 
582 	/*
583 	 * Common case is actually in the if statement
584 	 */
585 	if (vp == NULL || !(vp->v_tag == VT_NON && vp->v_type == VBLK)) {
586 		nvp->v_specinfo = malloc(sizeof(struct specinfo), M_VNODE,
587 			M_WAITOK);
588 		nvp->v_rdev = nvp_rdev;
589 		nvp->v_hashchain = vchain;
590 		nvp->v_specmountpoint = NULL;
591 		nvp->v_speclockf = NULL;
592 		nvp->v_specbitmap = NULL;
593 		if (nvp->v_type == VCHR &&
594 		    (cdevsw[major(nvp_rdev)].d_flags & D_CLONE) &&
595 		    (minor(nvp_rdev) >> CLONE_SHIFT == 0)) {
596 			if (vp != NULLVP)
597 				nvp->v_specbitmap = vp->v_specbitmap;
598 			else
599 				nvp->v_specbitmap = malloc(CLONE_MAPSZ,
600 				    M_VNODE, M_WAITOK | M_ZERO);
601 		}
602 		SLIST_INSERT_HEAD(vchain, nvp, v_specnext);
603 		if (vp != NULLVP) {
604 			nvp->v_flag |= VALIASED;
605 			vp->v_flag |= VALIASED;
606 			vput(vp);
607 		}
608 		return (NULLVP);
609 	}
610 
611 	/*
612 	 * This code is the uncommon case. It is called in case
613 	 * we found an alias that was VT_NON && vtype of VBLK
614 	 * This means we found a block device that was created
615 	 * using bdevvp.
616 	 * An example of such a vnode is the root partition device vnode
617 	 * created in ffs_mountroot.
618 	 *
619 	 * The vnodes created by bdevvp should not be aliased (why?).
620 	 */
621 
622 	VOP_UNLOCK(vp);
623 	vclean(vp, 0, p);
624 	vp->v_op = nvp->v_op;
625 	vp->v_tag = nvp->v_tag;
626 	nvp->v_type = VNON;
627 	insmntque(vp, mp);
628 	return (vp);
629 }
630 
631 /*
632  * Grab a particular vnode from the free list, increment its
633  * reference count and lock it. If the vnode lock bit is set,
634  * the vnode is being eliminated in vgone. In that case, we
635  * cannot grab it, so the process is awakened when the
636  * transition is completed, and an error code is returned to
637  * indicate that the vnode is no longer usable, possibly
638  * having been changed to a new file system type.
639  */
640 int
641 vget(struct vnode *vp, int flags)
642 {
643 	int error, s, onfreelist;
644 
645 	/*
646 	 * If the vnode is in the process of being cleaned out for
647 	 * another use, we wait for the cleaning to finish and then
648 	 * return failure. Cleaning is determined by checking that
649 	 * the VXLOCK flag is set.
650 	 */
651 	mtx_enter(&vnode_mtx);
652 	if (vp->v_lflag & VXLOCK) {
653 		if (flags & LK_NOWAIT) {
654 			mtx_leave(&vnode_mtx);
655 			return (EBUSY);
656 		}
657 
658 		vp->v_lflag |= VXWANT;
659 		msleep_nsec(vp, &vnode_mtx, PINOD, "vget", INFSLP);
660 		mtx_leave(&vnode_mtx);
661 		return (ENOENT);
662 	}
663 	mtx_leave(&vnode_mtx);
664 
665 	onfreelist = vp->v_bioflag & VBIOONFREELIST;
666 	if (vp->v_usecount == 0 && onfreelist) {
667 		s = splbio();
668 		if (vp->v_holdcnt > 0)
669 			TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist);
670 		else
671 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
672 		vp->v_bioflag &= ~VBIOONFREELIST;
673 		splx(s);
674 	}
675 
676 	vp->v_usecount++;
677 	if (flags & LK_TYPE_MASK) {
678 		if ((error = vn_lock(vp, flags)) != 0) {
679 			vp->v_usecount--;
680 			if (vp->v_usecount == 0 && onfreelist)
681 				vputonfreelist(vp);
682 		}
683 		return (error);
684 	}
685 
686 	return (0);
687 }
688 
689 
690 /* Vnode reference. */
691 void
692 vref(struct vnode *vp)
693 {
694 	KERNEL_ASSERT_LOCKED();
695 
696 #ifdef DIAGNOSTIC
697 	if (vp->v_usecount == 0)
698 		panic("vref used where vget required");
699 	if (vp->v_type == VNON)
700 		panic("vref on a VNON vnode");
701 #endif
702 	vp->v_usecount++;
703 }
704 
705 void
706 vputonfreelist(struct vnode *vp)
707 {
708 	int s;
709 	struct freelst *lst;
710 
711 	s = splbio();
712 #ifdef DIAGNOSTIC
713 	if (vp->v_usecount != 0)
714 		panic("Use count is not zero!");
715 
716 	/*
717 	 * If the hold count is still positive, one or many threads could still
718 	 * be waiting on the vnode lock inside uvn_io().
719 	 */
720 	if (vp->v_holdcnt == 0 && vp->v_lockcount != 0)
721 		panic("%s: lock count is not zero", __func__);
722 
723 	if (vp->v_bioflag & VBIOONFREELIST) {
724 		vprint("vnode already on free list: ", vp);
725 		panic("vnode already on free list");
726 	}
727 #endif
728 
729 	vp->v_bioflag |= VBIOONFREELIST;
730 	vp->v_bioflag &= ~VBIOERROR;
731 
732 	if (vp->v_holdcnt > 0)
733 		lst = &vnode_hold_list;
734 	else
735 		lst = &vnode_free_list;
736 
737 	if (vp->v_type == VBAD)
738 		TAILQ_INSERT_HEAD(lst, vp, v_freelist);
739 	else
740 		TAILQ_INSERT_TAIL(lst, vp, v_freelist);
741 
742 	splx(s);
743 }
744 
745 /*
746  * vput(), just unlock and vrele()
747  */
748 void
749 vput(struct vnode *vp)
750 {
751 	struct proc *p = curproc;
752 
753 #ifdef DIAGNOSTIC
754 	if (vp == NULL)
755 		panic("vput: null vp");
756 #endif
757 
758 #ifdef DIAGNOSTIC
759 	if (vp->v_usecount == 0) {
760 		vprint("vput: bad ref count", vp);
761 		panic("vput: ref cnt");
762 	}
763 #endif
764 	vp->v_usecount--;
765 	KASSERT(vp->v_usecount > 0 || vp->v_uvcount == 0);
766 	if (vp->v_usecount > 0) {
767 		VOP_UNLOCK(vp);
768 		return;
769 	}
770 
771 #ifdef DIAGNOSTIC
772 	if (vp->v_writecount != 0) {
773 		vprint("vput: bad writecount", vp);
774 		panic("vput: v_writecount != 0");
775 	}
776 #endif
777 
778 	VOP_INACTIVE(vp, p);
779 
780 	if (vp->v_usecount == 0 && !(vp->v_bioflag & VBIOONFREELIST))
781 		vputonfreelist(vp);
782 }
783 
784 /*
785  * Vnode release - use for active VNODES.
786  * If count drops to zero, call inactive routine and return to freelist.
787  * Returns 0 if it did not sleep.
788  */
789 int
790 vrele(struct vnode *vp)
791 {
792 	struct proc *p = curproc;
793 
794 #ifdef DIAGNOSTIC
795 	if (vp == NULL)
796 		panic("vrele: null vp");
797 #endif
798 #ifdef DIAGNOSTIC
799 	if (vp->v_usecount == 0) {
800 		vprint("vrele: bad ref count", vp);
801 		panic("vrele: ref cnt");
802 	}
803 #endif
804 	vp->v_usecount--;
805 	if (vp->v_usecount > 0) {
806 		return (0);
807 	}
808 
809 #ifdef DIAGNOSTIC
810 	if (vp->v_writecount != 0) {
811 		vprint("vrele: bad writecount", vp);
812 		panic("vrele: v_writecount != 0");
813 	}
814 #endif
815 
816 	if (vn_lock(vp, LK_EXCLUSIVE)) {
817 #ifdef DIAGNOSTIC
818 		vprint("vrele: cannot lock", vp);
819 #endif
820 		return (1);
821 	}
822 
823 	VOP_INACTIVE(vp, p);
824 
825 	if (vp->v_usecount == 0 && !(vp->v_bioflag & VBIOONFREELIST))
826 		vputonfreelist(vp);
827 	return (1);
828 }
829 
830 /* Page or buffer structure gets a reference. */
831 void
832 vhold(struct vnode *vp)
833 {
834 	/*
835 	 * If it is on the freelist and the hold count is currently
836 	 * zero, move it to the hold list.
837 	 */
838 	if ((vp->v_bioflag & VBIOONFREELIST) &&
839 	    vp->v_holdcnt == 0 && vp->v_usecount == 0) {
840 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
841 		TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist);
842 	}
843 	vp->v_holdcnt++;
844 }
845 
846 /* Lose interest in a vnode. */
847 void
848 vdrop(struct vnode *vp)
849 {
850 #ifdef DIAGNOSTIC
851 	if (vp->v_holdcnt == 0)
852 		panic("vdrop: zero holdcnt");
853 #endif
854 
855 	vp->v_holdcnt--;
856 
857 	/*
858 	 * If it is on the holdlist and the hold count drops to
859 	 * zero, move it to the free list.
860 	 */
861 	if ((vp->v_bioflag & VBIOONFREELIST) &&
862 	    vp->v_holdcnt == 0 && vp->v_usecount == 0) {
863 		TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist);
864 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
865 	}
866 }
867 
868 /*
869  * Remove any vnodes in the vnode table belonging to mount point mp.
870  *
871  * If MNT_NOFORCE is specified, there should not be any active ones,
872  * return error if any are found (nb: this is a user error, not a
873  * system error). If MNT_FORCE is specified, detach any active vnodes
874  * that are found.
875  */
876 #ifdef DEBUG_SYSCTL
877 int busyprt = 0;	/* print out busy vnodes */
878 struct ctldebug debug_vfs_busyprt = { "vfs_busyprt", &busyprt };
879 #endif
880 
881 int
882 vfs_mount_foreach_vnode(struct mount *mp,
883     int (*func)(struct vnode *, void *), void *arg) {
884 	struct vnode *vp, *nvp;
885 	int error = 0;
886 
887 loop:
888 	TAILQ_FOREACH_SAFE(vp , &mp->mnt_vnodelist, v_mntvnodes, nvp) {
889 		if (vp->v_mount != mp)
890 			goto loop;
891 
892 		error = func(vp, arg);
893 
894 		if (error != 0)
895 			break;
896 	}
897 
898 	return (error);
899 }
900 
901 struct vflush_args {
902 	struct vnode *skipvp;
903 	int busy;
904 	int flags;
905 };
906 
907 int
908 vflush_vnode(struct vnode *vp, void *arg)
909 {
910 	struct vflush_args *va = arg;
911 	struct proc *p = curproc;
912 
913 	if (vp == va->skipvp) {
914 		return (0);
915 	}
916 
917 	if ((va->flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
918 		return (0);
919 	}
920 
921 	/*
922 	 * If WRITECLOSE is set, only flush out regular file
923 	 * vnodes open for writing.
924 	 */
925 	if ((va->flags & WRITECLOSE) &&
926 	    (vp->v_writecount == 0 || vp->v_type != VREG)) {
927 		return (0);
928 	}
929 
930 	/*
931 	 * With v_usecount == 0, all we need to do is clear
932 	 * out the vnode data structures and we are done.
933 	 */
934 	if (vp->v_usecount == 0) {
935 		vgonel(vp, p);
936 		return (0);
937 	}
938 
939 	/*
940 	 * If FORCECLOSE is set, forcibly close the vnode.
941 	 * For block or character devices, revert to an
942 	 * anonymous device. For all other files, just kill them.
943 	 */
944 	if (va->flags & FORCECLOSE) {
945 		if (vp->v_type != VBLK && vp->v_type != VCHR) {
946 			vgonel(vp, p);
947 		} else {
948 			vclean(vp, 0, p);
949 			vp->v_op = &spec_vops;
950 			insmntque(vp, NULL);
951 		}
952 		return (0);
953 	}
954 
955 	/*
956 	 * If set, this is allowed to ignore vnodes which don't
957 	 * have changes pending to disk.
958 	 * XXX Might be nice to check per-fs "inode" flags, but
959 	 * generally the filesystem is sync'd already, right?
960 	 */
961 	if ((va->flags & IGNORECLEAN) &&
962 	    LIST_EMPTY(&vp->v_dirtyblkhd))
963 		return (0);
964 
965 #ifdef DEBUG_SYSCTL
966 	if (busyprt)
967 		vprint("vflush: busy vnode", vp);
968 #endif
969 	va->busy++;
970 	return (0);
971 }
972 
973 int
974 vflush(struct mount *mp, struct vnode *skipvp, int flags)
975 {
976 	struct vflush_args va;
977 	va.skipvp = skipvp;
978 	va.busy = 0;
979 	va.flags = flags;
980 
981 	vfs_mount_foreach_vnode(mp, vflush_vnode, &va);
982 
983 	if (va.busy)
984 		return (EBUSY);
985 	return (0);
986 }
987 
988 /*
989  * Disassociate the underlying file system from a vnode.
990  */
991 void
992 vclean(struct vnode *vp, int flags, struct proc *p)
993 {
994 	int active, do_wakeup = 0;
995 
996 	/*
997 	 * Check to see if the vnode is in use.
998 	 * If so we have to reference it before we clean it out
999 	 * so that its count cannot fall to zero and generate a
1000 	 * race against ourselves to recycle it.
1001 	 */
1002 	if ((active = vp->v_usecount) != 0)
1003 		vp->v_usecount++;
1004 
1005 	/*
1006 	 * Prevent the vnode from being recycled or
1007 	 * brought into use while we clean it out.
1008 	 */
1009 	mtx_enter(&vnode_mtx);
1010 	if (vp->v_lflag & VXLOCK)
1011 		panic("vclean: deadlock");
1012 	vp->v_lflag |= VXLOCK;
1013 
1014 	if (vp->v_lockcount > 0) {
1015 		/*
1016 		 * Ensure that any thread currently waiting on the same lock has
1017 		 * observed that the vnode is about to be exclusively locked
1018 		 * before continuing.
1019 		 */
1020 		msleep_nsec(&vp->v_lockcount, &vnode_mtx, PINOD, "vop_lock",
1021 		    INFSLP);
1022 		KASSERT(vp->v_lockcount == 0);
1023 	}
1024 	mtx_leave(&vnode_mtx);
1025 
1026 	/*
1027 	 * Even if the count is zero, the VOP_INACTIVE routine may still
1028 	 * have the object locked while it cleans it out. The VOP_LOCK
1029 	 * ensures that the VOP_INACTIVE routine is done with its work.
1030 	 * For active vnodes, it ensures that no other activity can
1031 	 * occur while the underlying object is being cleaned out.
1032 	 */
1033 	VOP_LOCK(vp, LK_EXCLUSIVE | LK_DRAIN);
1034 
1035 	/*
1036 	 * Clean out any VM data associated with the vnode.
1037 	 */
1038 	uvm_vnp_terminate(vp);
1039 	/*
1040 	 * Clean out any buffers associated with the vnode.
1041 	 */
1042 	if (flags & DOCLOSE)
1043 		vinvalbuf(vp, V_SAVE, NOCRED, p, 0, INFSLP);
1044 	/*
1045 	 * If purging an active vnode, it must be closed and
1046 	 * deactivated before being reclaimed. Note that the
1047 	 * VOP_INACTIVE will unlock the vnode
1048 	 */
1049 	if (active) {
1050 		if (flags & DOCLOSE)
1051 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, p);
1052 		VOP_INACTIVE(vp, p);
1053 	} else {
1054 		/*
1055 		 * Any other processes trying to obtain this lock must first
1056 		 * wait for VXLOCK to clear, then call the new lock operation.
1057 		 */
1058 		VOP_UNLOCK(vp);
1059 	}
1060 
1061 	/*
1062 	 * Reclaim the vnode.
1063 	 */
1064 	if (VOP_RECLAIM(vp, p))
1065 		panic("vclean: cannot reclaim");
1066 	if (active) {
1067 		vp->v_usecount--;
1068 		if (vp->v_usecount == 0) {
1069 			if (vp->v_holdcnt > 0)
1070 				panic("vclean: not clean");
1071 			vputonfreelist(vp);
1072 		}
1073 	}
1074 	cache_purge(vp);
1075 
1076 	/*
1077 	 * Done with purge, notify sleepers of the grim news.
1078 	 */
1079 	vp->v_op = &dead_vops;
1080 	VN_KNOTE(vp, NOTE_REVOKE);
1081 	vp->v_tag = VT_NON;
1082 #ifdef VFSLCKDEBUG
1083 	vp->v_flag &= ~VLOCKSWORK;
1084 #endif
1085 	mtx_enter(&vnode_mtx);
1086 	vp->v_lflag &= ~VXLOCK;
1087 	if (vp->v_lflag & VXWANT) {
1088 		vp->v_lflag &= ~VXWANT;
1089 		do_wakeup = 1;
1090 	}
1091 	mtx_leave(&vnode_mtx);
1092 	if (do_wakeup)
1093 		wakeup(vp);
1094 }
1095 
1096 /*
1097  * Recycle an unused vnode to the front of the free list.
1098  */
1099 int
1100 vrecycle(struct vnode *vp, struct proc *p)
1101 {
1102 	if (vp->v_usecount == 0) {
1103 		vgonel(vp, p);
1104 		return (1);
1105 	}
1106 	return (0);
1107 }
1108 
1109 /*
1110  * Eliminate all activity associated with a vnode
1111  * in preparation for reuse.
1112  */
1113 void
1114 vgone(struct vnode *vp)
1115 {
1116 	struct proc *p = curproc;
1117 	vgonel(vp, p);
1118 }
1119 
1120 /*
1121  * vgone, with struct proc.
1122  */
1123 void
1124 vgonel(struct vnode *vp, struct proc *p)
1125 {
1126 	struct vnode *vq;
1127 	struct vnode *vx;
1128 
1129 	KASSERT(vp->v_uvcount == 0);
1130 
1131 	/*
1132 	 * If a vgone (or vclean) is already in progress,
1133 	 * wait until it is done and return.
1134 	 */
1135 	mtx_enter(&vnode_mtx);
1136 	if (vp->v_lflag & VXLOCK) {
1137 		vp->v_lflag |= VXWANT;
1138 		msleep_nsec(vp, &vnode_mtx, PINOD, "vgone", INFSLP);
1139 		mtx_leave(&vnode_mtx);
1140 		return;
1141 	}
1142 	mtx_leave(&vnode_mtx);
1143 
1144 	/*
1145 	 * Clean out the filesystem specific data.
1146 	 */
1147 	vclean(vp, DOCLOSE, p);
1148 	/*
1149 	 * Delete from old mount point vnode list, if on one.
1150 	 */
1151 	if (vp->v_mount != NULL)
1152 		insmntque(vp, NULL);
1153 	/*
1154 	 * If special device, remove it from special device alias list
1155 	 * if it is on one.
1156 	 */
1157 	if ((vp->v_type == VBLK || vp->v_type == VCHR) &&
1158 	    vp->v_specinfo != NULL) {
1159 		if ((vp->v_flag & VALIASED) == 0 && vp->v_type == VCHR &&
1160 		    (cdevsw[major(vp->v_rdev)].d_flags & D_CLONE) &&
1161 		    (minor(vp->v_rdev) >> CLONE_SHIFT == 0)) {
1162 			free(vp->v_specbitmap, M_VNODE, CLONE_MAPSZ);
1163 		}
1164 		SLIST_REMOVE(vp->v_hashchain, vp, vnode, v_specnext);
1165 		if (vp->v_flag & VALIASED) {
1166 			vx = NULL;
1167 			SLIST_FOREACH(vq, vp->v_hashchain, v_specnext) {
1168 				if (vq->v_rdev != vp->v_rdev ||
1169 				    vq->v_type != vp->v_type)
1170 					continue;
1171 				if (vx)
1172 					break;
1173 				vx = vq;
1174 			}
1175 			if (vx == NULL)
1176 				panic("missing alias");
1177 			if (vq == NULL)
1178 				vx->v_flag &= ~VALIASED;
1179 			vp->v_flag &= ~VALIASED;
1180 		}
1181 		lf_purgelocks(&vp->v_speclockf);
1182 		free(vp->v_specinfo, M_VNODE, sizeof(struct specinfo));
1183 		vp->v_specinfo = NULL;
1184 	}
1185 	/*
1186 	 * If it is on the freelist and not already at the head,
1187 	 * move it to the head of the list.
1188 	 */
1189 	vp->v_type = VBAD;
1190 
1191 	/*
1192 	 * Move onto the free list, unless we were called from
1193 	 * getnewvnode and we're not on any free list
1194 	 */
1195 	if (vp->v_usecount == 0 &&
1196 	    (vp->v_bioflag & VBIOONFREELIST)) {
1197 		int s;
1198 
1199 		s = splbio();
1200 
1201 		if (vp->v_holdcnt > 0)
1202 			panic("vgonel: not clean");
1203 
1204 		if (TAILQ_FIRST(&vnode_free_list) != vp) {
1205 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1206 			TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
1207 		}
1208 		splx(s);
1209 	}
1210 }
1211 
1212 /*
1213  * Lookup a vnode by device number.
1214  */
1215 int
1216 vfinddev(dev_t dev, enum vtype type, struct vnode **vpp)
1217 {
1218 	struct vnode *vp;
1219 	int rc =0;
1220 
1221 	SLIST_FOREACH(vp, &speclisth[SPECHASH(dev)], v_specnext) {
1222 		if (dev != vp->v_rdev || type != vp->v_type)
1223 			continue;
1224 		*vpp = vp;
1225 		rc = 1;
1226 		break;
1227 	}
1228 	return (rc);
1229 }
1230 
1231 /*
1232  * Revoke all the vnodes corresponding to the specified minor number
1233  * range (endpoints inclusive) of the specified major.
1234  */
1235 void
1236 vdevgone(int maj, int minl, int minh, enum vtype type)
1237 {
1238 	struct vnode *vp;
1239 	int mn;
1240 
1241 	for (mn = minl; mn <= minh; mn++)
1242 		if (vfinddev(makedev(maj, mn), type, &vp))
1243 			VOP_REVOKE(vp, REVOKEALL);
1244 }
1245 
1246 /*
1247  * Calculate the total number of references to a special device.
1248  */
1249 int
1250 vcount(struct vnode *vp)
1251 {
1252 	struct vnode *vq;
1253 	int count;
1254 
1255 loop:
1256 	if ((vp->v_flag & VALIASED) == 0)
1257 		return (vp->v_usecount);
1258 	count = 0;
1259 	SLIST_FOREACH(vq, vp->v_hashchain, v_specnext) {
1260 		if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type)
1261 			continue;
1262 		/*
1263 		 * Alias, but not in use, so flush it out.
1264 		 */
1265 		if (vq->v_usecount == 0 && vq != vp) {
1266 			vgone(vq);
1267 			goto loop;
1268 		}
1269 		count += vq->v_usecount;
1270 	}
1271 	return (count);
1272 }
1273 
1274 #if defined(DEBUG) || defined(DIAGNOSTIC)
1275 /*
1276  * Print out a description of a vnode.
1277  */
1278 static char *typename[] =
1279    { "VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD" };
1280 
1281 void
1282 vprint(char *label, struct vnode *vp)
1283 {
1284 	char buf[64];
1285 
1286 	if (label != NULL)
1287 		printf("%s: ", label);
1288 	printf("%p, type %s, use %u, write %u, hold %u,",
1289 		vp, typename[vp->v_type], vp->v_usecount, vp->v_writecount,
1290 		vp->v_holdcnt);
1291 	buf[0] = '\0';
1292 	if (vp->v_flag & VROOT)
1293 		strlcat(buf, "|VROOT", sizeof buf);
1294 	if (vp->v_flag & VTEXT)
1295 		strlcat(buf, "|VTEXT", sizeof buf);
1296 	if (vp->v_flag & VSYSTEM)
1297 		strlcat(buf, "|VSYSTEM", sizeof buf);
1298 	if (vp->v_lflag & VXLOCK)
1299 		strlcat(buf, "|VXLOCK", sizeof buf);
1300 	if (vp->v_lflag & VXWANT)
1301 		strlcat(buf, "|VXWANT", sizeof buf);
1302 	if (vp->v_bioflag & VBIOWAIT)
1303 		strlcat(buf, "|VBIOWAIT", sizeof buf);
1304 	if (vp->v_bioflag & VBIOONFREELIST)
1305 		strlcat(buf, "|VBIOONFREELIST", sizeof buf);
1306 	if (vp->v_bioflag & VBIOONSYNCLIST)
1307 		strlcat(buf, "|VBIOONSYNCLIST", sizeof buf);
1308 	if (vp->v_flag & VALIASED)
1309 		strlcat(buf, "|VALIASED", sizeof buf);
1310 	if (buf[0] != '\0')
1311 		printf(" flags (%s)", &buf[1]);
1312 	if (vp->v_data == NULL) {
1313 		printf("\n");
1314 	} else {
1315 		printf("\n\t");
1316 		VOP_PRINT(vp);
1317 	}
1318 }
1319 #endif /* DEBUG || DIAGNOSTIC */
1320 
1321 #ifdef DEBUG
1322 /*
1323  * List all of the locked vnodes in the system.
1324  * Called when debugging the kernel.
1325  */
1326 void
1327 printlockedvnodes(void)
1328 {
1329 	struct mount *mp;
1330 	struct vnode *vp;
1331 
1332 	printf("Locked vnodes\n");
1333 
1334 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
1335 		if (vfs_busy(mp, VB_READ|VB_NOWAIT))
1336 			continue;
1337 		TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
1338 			if (VOP_ISLOCKED(vp))
1339 				vprint(NULL, vp);
1340 		}
1341 		vfs_unbusy(mp);
1342 	}
1343 
1344 }
1345 #endif
1346 
1347 /*
1348  * Top level filesystem related information gathering.
1349  */
1350 int
1351 vfs_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp,
1352     size_t newlen, struct proc *p)
1353 {
1354 	struct vfsconf *vfsp, *tmpvfsp;
1355 	int ret;
1356 
1357 	/* all sysctl names at this level are at least name and field */
1358 	if (namelen < 2)
1359 		return (ENOTDIR);		/* overloaded */
1360 
1361 	if (name[0] != VFS_GENERIC) {
1362 		vfsp = vfs_bytypenum(name[0]);
1363 		if (vfsp == NULL || vfsp->vfc_vfsops->vfs_sysctl == NULL)
1364 			return (EOPNOTSUPP);
1365 
1366 		return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
1367 		    oldp, oldlenp, newp, newlen, p));
1368 	}
1369 
1370 	switch (name[1]) {
1371 	case VFS_MAXTYPENUM:
1372 		return (sysctl_rdint(oldp, oldlenp, newp, maxvfsconf));
1373 
1374 	case VFS_CONF:
1375 		if (namelen < 3)
1376 			return (ENOTDIR);	/* overloaded */
1377 
1378 		vfsp = vfs_bytypenum(name[2]);
1379 		if (vfsp == NULL)
1380 			return (EOPNOTSUPP);
1381 
1382 		/* Make a copy, clear out kernel pointers */
1383 		tmpvfsp = malloc(sizeof(*tmpvfsp), M_TEMP, M_WAITOK|M_ZERO);
1384 		memcpy(tmpvfsp, vfsp, sizeof(*tmpvfsp));
1385 		tmpvfsp->vfc_vfsops = NULL;
1386 
1387 		ret = sysctl_rdstruct(oldp, oldlenp, newp, tmpvfsp,
1388 		    sizeof(struct vfsconf));
1389 
1390 		free(tmpvfsp, M_TEMP, sizeof(*tmpvfsp));
1391 		return (ret);
1392 	case VFS_BCACHESTAT:	/* buffer cache statistics */
1393 		ret = sysctl_rdstruct(oldp, oldlenp, newp, &bcstats,
1394 		    sizeof(struct bcachestats));
1395 		return(ret);
1396 	}
1397 	return (EOPNOTSUPP);
1398 }
1399 
1400 /*
1401  * Check to see if a filesystem is mounted on a block device.
1402  */
1403 int
1404 vfs_mountedon(struct vnode *vp)
1405 {
1406 	struct vnode *vq;
1407 	int error = 0;
1408 
1409 	if (vp->v_specmountpoint != NULL)
1410 		return (EBUSY);
1411 	if (vp->v_flag & VALIASED) {
1412 		SLIST_FOREACH(vq, vp->v_hashchain, v_specnext) {
1413 			if (vq->v_rdev != vp->v_rdev ||
1414 			    vq->v_type != vp->v_type)
1415 				continue;
1416 			if (vq->v_specmountpoint != NULL) {
1417 				error = EBUSY;
1418 				break;
1419 			}
1420 		}
1421 	}
1422 	return (error);
1423 }
1424 
1425 #ifdef NFSSERVER
1426 /*
1427  * Build hash lists of net addresses and hang them off the mount point.
1428  * Called by vfs_export() to set up the lists of export addresses.
1429  */
1430 int
1431 vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
1432     struct export_args *argp)
1433 {
1434 	struct netcred *np;
1435 	struct radix_node_head *rnh;
1436 	int nplen, i;
1437 	struct radix_node *rn;
1438 	struct sockaddr *saddr, *smask = NULL;
1439 	int error;
1440 
1441 	if (argp->ex_addrlen == 0) {
1442 		if (mp->mnt_flag & MNT_DEFEXPORTED)
1443 			return (EPERM);
1444 		np = &nep->ne_defexported;
1445 		/* fill in the kernel's ucred from userspace's xucred */
1446 		if ((error = crfromxucred(&np->netc_anon, &argp->ex_anon)))
1447 			return (error);
1448 		mp->mnt_flag |= MNT_DEFEXPORTED;
1449 		goto finish;
1450 	}
1451 	if (argp->ex_addrlen > MLEN || argp->ex_masklen > MLEN ||
1452 	    argp->ex_addrlen < 0 || argp->ex_masklen < 0)
1453 		return (EINVAL);
1454 	nplen = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
1455 	np = (struct netcred *)malloc(nplen, M_NETADDR, M_WAITOK|M_ZERO);
1456 	np->netc_len = nplen;
1457 	saddr = (struct sockaddr *)(np + 1);
1458 	error = copyin(argp->ex_addr, saddr, argp->ex_addrlen);
1459 	if (error)
1460 		goto out;
1461 	if (saddr->sa_len > argp->ex_addrlen)
1462 		saddr->sa_len = argp->ex_addrlen;
1463 	if (argp->ex_masklen) {
1464 		smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
1465 		error = copyin(argp->ex_mask, smask, argp->ex_masklen);
1466 		if (error)
1467 			goto out;
1468 		if (smask->sa_len > argp->ex_masklen)
1469 			smask->sa_len = argp->ex_masklen;
1470 	}
1471 	/* fill in the kernel's ucred from userspace's xucred */
1472 	if ((error = crfromxucred(&np->netc_anon, &argp->ex_anon)))
1473 		goto out;
1474 	i = saddr->sa_family;
1475 	switch (i) {
1476 	case AF_INET:
1477 		if ((rnh = nep->ne_rtable_inet) == NULL) {
1478 			if (!rn_inithead((void **)&nep->ne_rtable_inet,
1479 			    offsetof(struct sockaddr_in, sin_addr))) {
1480 				error = ENOBUFS;
1481 				goto out;
1482 			}
1483 			rnh = nep->ne_rtable_inet;
1484 		}
1485 		break;
1486 	default:
1487 		error = EINVAL;
1488 		goto out;
1489 	}
1490 	rn = rn_addroute(saddr, smask, rnh, np->netc_rnodes, 0);
1491 	if (rn == NULL || np != (struct netcred *)rn) { /* already exists */
1492 		error = EPERM;
1493 		goto out;
1494 	}
1495 finish:
1496 	np->netc_exflags = argp->ex_flags;
1497 	return (0);
1498 out:
1499 	free(np, M_NETADDR, np->netc_len);
1500 	return (error);
1501 }
1502 
1503 int
1504 vfs_free_netcred(struct radix_node *rn, void *w, u_int id)
1505 {
1506 	struct radix_node_head *rnh = (struct radix_node_head *)w;
1507 	struct netcred * np = (struct netcred *)rn;
1508 
1509 	rn_delete(rn->rn_key, rn->rn_mask, rnh, NULL);
1510 	free(np, M_NETADDR, np->netc_len);
1511 	return (0);
1512 }
1513 
1514 /*
1515  * Free the net address hash lists that are hanging off the mount points.
1516  */
1517 void
1518 vfs_free_addrlist(struct netexport *nep)
1519 {
1520 	struct radix_node_head *rnh;
1521 
1522 	if ((rnh = nep->ne_rtable_inet) != NULL) {
1523 		rn_walktree(rnh, vfs_free_netcred, rnh);
1524 		free(rnh, M_RTABLE, sizeof(*rnh));
1525 		nep->ne_rtable_inet = NULL;
1526 	}
1527 }
1528 #endif /* NFSSERVER */
1529 
1530 int
1531 vfs_export(struct mount *mp, struct netexport *nep, struct export_args *argp)
1532 {
1533 #ifdef NFSSERVER
1534 	int error;
1535 
1536 	if (argp->ex_flags & MNT_DELEXPORT) {
1537 		vfs_free_addrlist(nep);
1538 		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
1539 	}
1540 	if (argp->ex_flags & MNT_EXPORTED) {
1541 		if ((error = vfs_hang_addrlist(mp, nep, argp)) != 0)
1542 			return (error);
1543 		mp->mnt_flag |= MNT_EXPORTED;
1544 	}
1545 	return (0);
1546 #else
1547 	return (ENOTSUP);
1548 #endif /* NFSSERVER */
1549 }
1550 
1551 struct netcred *
1552 vfs_export_lookup(struct mount *mp, struct netexport *nep, struct mbuf *nam)
1553 {
1554 #ifdef NFSSERVER
1555 	struct netcred *np;
1556 	struct radix_node_head *rnh;
1557 	struct sockaddr *saddr;
1558 
1559 	np = NULL;
1560 	if (mp->mnt_flag & MNT_EXPORTED) {
1561 		/*
1562 		 * Lookup in the export list first.
1563 		 */
1564 		if (nam != NULL) {
1565 			saddr = mtod(nam, struct sockaddr *);
1566 			switch(saddr->sa_family) {
1567 			case AF_INET:
1568 				rnh = nep->ne_rtable_inet;
1569 				break;
1570 			default:
1571 				rnh = NULL;
1572 				break;
1573 			}
1574 			if (rnh != NULL)
1575 				np = (struct netcred *)rn_match(saddr, rnh);
1576 		}
1577 		/*
1578 		 * If no address match, use the default if it exists.
1579 		 */
1580 		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
1581 			np = &nep->ne_defexported;
1582 	}
1583 	return (np);
1584 #else
1585 	return (NULL);
1586 #endif /* NFSSERVER */
1587 }
1588 
1589 /*
1590  * Do the usual access checking.
1591  * file_mode, uid and gid are from the vnode in question,
1592  * while acc_mode and cred are from the VOP_ACCESS parameter list
1593  */
1594 int
1595 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid,
1596     mode_t acc_mode, struct ucred *cred)
1597 {
1598 	mode_t mask;
1599 
1600 	/* User id 0 always gets read/write access. */
1601 	if (cred->cr_uid == 0) {
1602 		/* For VEXEC, at least one of the execute bits must be set. */
1603 		if ((acc_mode & VEXEC) && type != VDIR &&
1604 		    (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)
1605 			return EACCES;
1606 		return 0;
1607 	}
1608 
1609 	mask = 0;
1610 
1611 	/* Otherwise, check the owner. */
1612 	if (cred->cr_uid == uid) {
1613 		if (acc_mode & VEXEC)
1614 			mask |= S_IXUSR;
1615 		if (acc_mode & VREAD)
1616 			mask |= S_IRUSR;
1617 		if (acc_mode & VWRITE)
1618 			mask |= S_IWUSR;
1619 		return (file_mode & mask) == mask ? 0 : EACCES;
1620 	}
1621 
1622 	/* Otherwise, check the groups. */
1623 	if (groupmember(gid, cred)) {
1624 		if (acc_mode & VEXEC)
1625 			mask |= S_IXGRP;
1626 		if (acc_mode & VREAD)
1627 			mask |= S_IRGRP;
1628 		if (acc_mode & VWRITE)
1629 			mask |= S_IWGRP;
1630 		return (file_mode & mask) == mask ? 0 : EACCES;
1631 	}
1632 
1633 	/* Otherwise, check everyone else. */
1634 	if (acc_mode & VEXEC)
1635 		mask |= S_IXOTH;
1636 	if (acc_mode & VREAD)
1637 		mask |= S_IROTH;
1638 	if (acc_mode & VWRITE)
1639 		mask |= S_IWOTH;
1640 	return (file_mode & mask) == mask ? 0 : EACCES;
1641 }
1642 
1643 int
1644 vnoperm(struct vnode *vp)
1645 {
1646 	if (vp->v_flag & VROOT || vp->v_mount == NULL)
1647 		return 0;
1648 
1649 	return (vp->v_mount->mnt_flag & MNT_NOPERM);
1650 }
1651 
1652 struct rwlock vfs_stall_lock = RWLOCK_INITIALIZER("vfs_stall");
1653 unsigned int vfs_stalling = 0;
1654 
1655 int
1656 vfs_stall(struct proc *p, int stall)
1657 {
1658 	struct mount *mp;
1659 	int allerror = 0, error;
1660 
1661 	if (stall) {
1662 		atomic_inc_int(&vfs_stalling);
1663 		rw_enter_write(&vfs_stall_lock);
1664 	}
1665 
1666 	/*
1667 	 * The loop variable mp is protected by vfs_busy() so that it cannot
1668 	 * be unmounted while VFS_SYNC() sleeps.  Traverse forward to keep the
1669 	 * lock order consistent with dounmount().
1670 	 */
1671 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
1672 		if (stall) {
1673 			error = vfs_busy(mp, VB_WRITE|VB_WAIT|VB_DUPOK);
1674 			if (error) {
1675 				printf("%s: busy\n", mp->mnt_stat.f_mntonname);
1676 				allerror = error;
1677 				continue;
1678 			}
1679 			uvm_vnp_sync(mp);
1680 			error = VFS_SYNC(mp, MNT_WAIT, stall, p->p_ucred, p);
1681 			if (error) {
1682 				printf("%s: failed to sync\n",
1683 				    mp->mnt_stat.f_mntonname);
1684 				vfs_unbusy(mp);
1685 				allerror = error;
1686 				continue;
1687 			}
1688 			mp->mnt_flag |= MNT_STALLED;
1689 		} else {
1690 			if (mp->mnt_flag & MNT_STALLED) {
1691 				vfs_unbusy(mp);
1692 				mp->mnt_flag &= ~MNT_STALLED;
1693 			}
1694 		}
1695 	}
1696 
1697 	if (!stall) {
1698 		rw_exit_write(&vfs_stall_lock);
1699 		atomic_dec_int(&vfs_stalling);
1700 	}
1701 
1702 	return (allerror);
1703 }
1704 
1705 void
1706 vfs_stall_barrier(void)
1707 {
1708 	if (__predict_false(vfs_stalling)) {
1709 		rw_enter_read(&vfs_stall_lock);
1710 		rw_exit_read(&vfs_stall_lock);
1711 	}
1712 }
1713 
1714 /*
1715  * Unmount all file systems.
1716  * We traverse the list in reverse order under the assumption that doing so
1717  * will avoid needing to worry about dependencies.
1718  */
1719 void
1720 vfs_unmountall(void)
1721 {
1722 	struct mount *mp, *nmp;
1723 	int allerror, error, again = 1;
1724 
1725  retry:
1726 	allerror = 0;
1727 	TAILQ_FOREACH_REVERSE_SAFE(mp, &mountlist, mntlist, mnt_list, nmp) {
1728 		if (vfs_busy(mp, VB_WRITE|VB_NOWAIT))
1729 			continue;
1730 		/* XXX Here is a race, the next pointer is not locked. */
1731 		if ((error = dounmount(mp, MNT_FORCE, curproc)) != 0) {
1732 			printf("unmount of %s failed with error %d\n",
1733 			    mp->mnt_stat.f_mntonname, error);
1734 			allerror = 1;
1735 		}
1736 	}
1737 
1738 	if (allerror) {
1739 		printf("WARNING: some file systems would not unmount\n");
1740 		if (again) {
1741 			printf("retrying\n");
1742 			again = 0;
1743 			goto retry;
1744 		}
1745 	}
1746 }
1747 
1748 /*
1749  * Sync and unmount file systems before shutting down.
1750  */
1751 void
1752 vfs_shutdown(struct proc *p)
1753 {
1754 #ifdef ACCOUNTING
1755 	acct_shutdown();
1756 #endif
1757 
1758 	printf("syncing disks...");
1759 
1760 	if (panicstr == NULL) {
1761 		/* Sync before unmount, in case we hang on something. */
1762 		sys_sync(p, NULL, NULL);
1763 		vfs_unmountall();
1764 	}
1765 
1766 #if NSOFTRAID > 0
1767 	sr_quiesce();
1768 #endif
1769 
1770 	if (vfs_syncwait(p, 1))
1771 		printf(" giving up\n");
1772 	else
1773 		printf(" done\n");
1774 }
1775 
1776 /*
1777  * perform sync() operation and wait for buffers to flush.
1778  */
1779 int
1780 vfs_syncwait(struct proc *p, int verbose)
1781 {
1782 	struct buf *bp;
1783 	int iter, nbusy, dcount, s;
1784 #ifdef MULTIPROCESSOR
1785 	int hold_count;
1786 #endif
1787 
1788 	sys_sync(p, NULL, NULL);
1789 
1790 	/* Wait for sync to finish. */
1791 	dcount = 10000;
1792 	for (iter = 0; iter < 20; iter++) {
1793 		nbusy = 0;
1794 		LIST_FOREACH(bp, &bufhead, b_list) {
1795 			if ((bp->b_flags & (B_BUSY|B_INVAL|B_READ)) == B_BUSY)
1796 				nbusy++;
1797 			/*
1798 			 * With soft updates, some buffers that are
1799 			 * written will be remarked as dirty until other
1800 			 * buffers are written.
1801 			 */
1802 			if (bp->b_flags & B_DELWRI) {
1803 				s = splbio();
1804 				bremfree(bp);
1805 				buf_acquire(bp);
1806 				splx(s);
1807 				nbusy++;
1808 				bawrite(bp);
1809 				if (dcount-- <= 0) {
1810 					if (verbose)
1811 						printf("softdep ");
1812 					return 1;
1813 				}
1814 			}
1815 		}
1816 		if (nbusy == 0)
1817 			break;
1818 		if (verbose)
1819 			printf("%d ", nbusy);
1820 #ifdef MULTIPROCESSOR
1821 		if (_kernel_lock_held())
1822 			hold_count = __mp_release_all(&kernel_lock);
1823 		else
1824 			hold_count = 0;
1825 #endif
1826 		DELAY(40000 * iter);
1827 #ifdef MULTIPROCESSOR
1828 		if (hold_count)
1829 			__mp_acquire_count(&kernel_lock, hold_count);
1830 #endif
1831 	}
1832 
1833 	return nbusy;
1834 }
1835 
1836 /*
1837  * posix file system related system variables.
1838  */
1839 int
1840 fs_posix_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp,
1841     void *newp, size_t newlen, struct proc *p)
1842 {
1843 	/* all sysctl names at this level are terminal */
1844 	if (namelen != 1)
1845 		return (ENOTDIR);
1846 
1847 	switch (name[0]) {
1848 	case FS_POSIX_SETUID:
1849 		return (sysctl_securelevel_int(oldp, oldlenp, newp, newlen,
1850 		    &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