xref: /netbsd-src/sys/kern/vfs_subr.c (revision 1ffa7b76c40339c17a0fb2a09fac93f287cfc046)
1 /*	$NetBSD: vfs_subr.c,v 1.194 2003/04/22 13:11:23 christos Exp $	*/
2 
3 /*-
4  * Copyright (c) 1997, 1998 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9  * NASA Ames Research Center.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. All advertising materials mentioning features or use of this software
20  *    must display the following acknowledgement:
21  *	This product includes software developed by the NetBSD
22  *	Foundation, Inc. and its contributors.
23  * 4. Neither the name of The NetBSD Foundation nor the names of its
24  *    contributors may be used to endorse or promote products derived
25  *    from this software without specific prior written permission.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
28  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
29  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
30  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
31  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37  * POSSIBILITY OF SUCH DAMAGE.
38  */
39 
40 /*
41  * Copyright (c) 1989, 1993
42  *	The Regents of the University of California.  All rights reserved.
43  * (c) UNIX System Laboratories, Inc.
44  * All or some portions of this file are derived from material licensed
45  * to the University of California by American Telephone and Telegraph
46  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
47  * the permission of UNIX System Laboratories, Inc.
48  *
49  * Redistribution and use in source and binary forms, with or without
50  * modification, are permitted provided that the following conditions
51  * are met:
52  * 1. Redistributions of source code must retain the above copyright
53  *    notice, this list of conditions and the following disclaimer.
54  * 2. Redistributions in binary form must reproduce the above copyright
55  *    notice, this list of conditions and the following disclaimer in the
56  *    documentation and/or other materials provided with the distribution.
57  * 3. All advertising materials mentioning features or use of this software
58  *    must display the following acknowledgement:
59  *	This product includes software developed by the University of
60  *	California, Berkeley and its contributors.
61  * 4. Neither the name of the University nor the names of its contributors
62  *    may be used to endorse or promote products derived from this software
63  *    without specific prior written permission.
64  *
65  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
66  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
67  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
68  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
69  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
70  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
71  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
72  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
73  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
74  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
75  * SUCH DAMAGE.
76  *
77  *	@(#)vfs_subr.c	8.13 (Berkeley) 4/18/94
78  */
79 
80 /*
81  * External virtual filesystem routines
82  */
83 
84 #include <sys/cdefs.h>
85 __KERNEL_RCSID(0, "$NetBSD: vfs_subr.c,v 1.194 2003/04/22 13:11:23 christos Exp $");
86 
87 #include "opt_ddb.h"
88 #include "opt_compat_netbsd.h"
89 #include "opt_compat_43.h"
90 
91 #include <sys/param.h>
92 #include <sys/systm.h>
93 #include <sys/proc.h>
94 #include <sys/kernel.h>
95 #include <sys/mount.h>
96 #include <sys/time.h>
97 #include <sys/event.h>
98 #include <sys/fcntl.h>
99 #include <sys/vnode.h>
100 #include <sys/stat.h>
101 #include <sys/namei.h>
102 #include <sys/ucred.h>
103 #include <sys/buf.h>
104 #include <sys/errno.h>
105 #include <sys/malloc.h>
106 #include <sys/domain.h>
107 #include <sys/mbuf.h>
108 #include <sys/sa.h>
109 #include <sys/syscallargs.h>
110 #include <sys/device.h>
111 #include <sys/dirent.h>
112 #include <sys/filedesc.h>
113 
114 #include <miscfs/specfs/specdev.h>
115 #include <miscfs/genfs/genfs.h>
116 #include <miscfs/syncfs/syncfs.h>
117 
118 #include <uvm/uvm.h>
119 #include <uvm/uvm_ddb.h>
120 
121 #include <sys/sysctl.h>
122 
123 const enum vtype iftovt_tab[16] = {
124 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
125 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
126 };
127 const int	vttoif_tab[9] = {
128 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
129 	S_IFSOCK, S_IFIFO, S_IFMT,
130 };
131 
132 int doforce = 1;		/* 1 => permit forcible unmounting */
133 int prtactive = 0;		/* 1 => print out reclaim of active vnodes */
134 
135 extern int dovfsusermount;	/* 1 => permit any user to mount filesystems */
136 
137 /*
138  * Insq/Remq for the vnode usage lists.
139  */
140 #define	bufinsvn(bp, dp)	LIST_INSERT_HEAD(dp, bp, b_vnbufs)
141 #define	bufremvn(bp) {							\
142 	LIST_REMOVE(bp, b_vnbufs);					\
143 	(bp)->b_vnbufs.le_next = NOLIST;				\
144 }
145 /* TAILQ_HEAD(freelst, vnode) vnode_free_list =	vnode free list (in vnode.h) */
146 struct freelst vnode_free_list = TAILQ_HEAD_INITIALIZER(vnode_free_list);
147 struct freelst vnode_hold_list = TAILQ_HEAD_INITIALIZER(vnode_hold_list);
148 
149 struct mntlist mountlist =			/* mounted filesystem list */
150     CIRCLEQ_HEAD_INITIALIZER(mountlist);
151 struct vfs_list_head vfs_list =			/* vfs list */
152     LIST_HEAD_INITIALIZER(vfs_list);
153 
154 struct nfs_public nfs_pub;			/* publicly exported FS */
155 
156 struct simplelock mountlist_slock = SIMPLELOCK_INITIALIZER;
157 static struct simplelock mntid_slock = SIMPLELOCK_INITIALIZER;
158 struct simplelock mntvnode_slock = SIMPLELOCK_INITIALIZER;
159 struct simplelock vnode_free_list_slock = SIMPLELOCK_INITIALIZER;
160 struct simplelock spechash_slock = SIMPLELOCK_INITIALIZER;
161 
162 /* XXX - gross; single global lock to protect v_numoutput */
163 struct simplelock global_v_numoutput_slock = SIMPLELOCK_INITIALIZER;
164 
165 /*
166  * These define the root filesystem and device.
167  */
168 struct mount *rootfs;
169 struct vnode *rootvnode;
170 struct device *root_device;			/* root device */
171 
172 struct pool vnode_pool;				/* memory pool for vnodes */
173 
174 MALLOC_DEFINE(M_VNODE, "vnodes", "Dynamically allocated vnodes");
175 
176 /*
177  * Local declarations.
178  */
179 void insmntque __P((struct vnode *, struct mount *));
180 int getdevvp __P((dev_t, struct vnode **, enum vtype));
181 void vgoneall __P((struct vnode *));
182 
183 void vclean(struct vnode *, int, struct proc *);
184 
185 static int vfs_hang_addrlist __P((struct mount *, struct netexport *,
186 				  struct export_args *));
187 static int vfs_free_netcred __P((struct radix_node *, void *));
188 static void vfs_free_addrlist __P((struct netexport *));
189 
190 #ifdef DEBUG
191 void printlockedvnodes __P((void));
192 #endif
193 
194 /*
195  * Initialize the vnode management data structures.
196  */
197 void
198 vntblinit()
199 {
200 
201 	pool_init(&vnode_pool, sizeof(struct vnode), 0, 0, 0, "vnodepl",
202 	    &pool_allocator_nointr);
203 
204 	/*
205 	 * Initialize the filesystem syncer.
206 	 */
207 	vn_initialize_syncerd();
208 }
209 
210 /*
211  * Mark a mount point as busy. Used to synchronize access and to delay
212  * unmounting. Interlock is not released on failure.
213  */
214 int
215 vfs_busy(mp, flags, interlkp)
216 	struct mount *mp;
217 	int flags;
218 	struct simplelock *interlkp;
219 {
220 	int lkflags;
221 
222 	while (mp->mnt_flag & MNT_UNMOUNT) {
223 		int gone;
224 
225 		if (flags & LK_NOWAIT)
226 			return (ENOENT);
227 		if ((flags & LK_RECURSEFAIL) && mp->mnt_unmounter != NULL
228 		    && mp->mnt_unmounter == curproc)
229 			return (EDEADLK);
230 		if (interlkp)
231 			simple_unlock(interlkp);
232 		/*
233 		 * Since all busy locks are shared except the exclusive
234 		 * lock granted when unmounting, the only place that a
235 		 * wakeup needs to be done is at the release of the
236 		 * exclusive lock at the end of dounmount.
237 		 *
238 		 * XXX MP: add spinlock protecting mnt_wcnt here once you
239 		 * can atomically unlock-and-sleep.
240 		 */
241 		mp->mnt_wcnt++;
242 		tsleep((caddr_t)mp, PVFS, "vfs_busy", 0);
243 		mp->mnt_wcnt--;
244 		gone = mp->mnt_flag & MNT_GONE;
245 
246 		if (mp->mnt_wcnt == 0)
247 			wakeup(&mp->mnt_wcnt);
248 		if (interlkp)
249 			simple_lock(interlkp);
250 		if (gone)
251 			return (ENOENT);
252 	}
253 	lkflags = LK_SHARED;
254 	if (interlkp)
255 		lkflags |= LK_INTERLOCK;
256 	if (lockmgr(&mp->mnt_lock, lkflags, interlkp))
257 		panic("vfs_busy: unexpected lock failure");
258 	return (0);
259 }
260 
261 /*
262  * Free a busy filesystem.
263  */
264 void
265 vfs_unbusy(mp)
266 	struct mount *mp;
267 {
268 
269 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL);
270 }
271 
272 /*
273  * Lookup a filesystem type, and if found allocate and initialize
274  * a mount structure for it.
275  *
276  * Devname is usually updated by mount(8) after booting.
277  */
278 int
279 vfs_rootmountalloc(fstypename, devname, mpp)
280 	char *fstypename;
281 	char *devname;
282 	struct mount **mpp;
283 {
284 	struct vfsops *vfsp = NULL;
285 	struct mount *mp;
286 
287 	LIST_FOREACH(vfsp, &vfs_list, vfs_list)
288 		if (!strncmp(vfsp->vfs_name, fstypename, MFSNAMELEN))
289 			break;
290 
291 	if (vfsp == NULL)
292 		return (ENODEV);
293 	mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
294 	memset((char *)mp, 0, (u_long)sizeof(struct mount));
295 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0);
296 	(void)vfs_busy(mp, LK_NOWAIT, 0);
297 	LIST_INIT(&mp->mnt_vnodelist);
298 	mp->mnt_op = vfsp;
299 	mp->mnt_flag = MNT_RDONLY;
300 	mp->mnt_vnodecovered = NULLVP;
301 	vfsp->vfs_refcount++;
302 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfs_name, MFSNAMELEN);
303 	mp->mnt_stat.f_mntonname[0] = '/';
304 	(void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
305 	*mpp = mp;
306 	return (0);
307 }
308 
309 /*
310  * Lookup a mount point by filesystem identifier.
311  */
312 struct mount *
313 vfs_getvfs(fsid)
314 	fsid_t *fsid;
315 {
316 	struct mount *mp;
317 
318 	simple_lock(&mountlist_slock);
319 	CIRCLEQ_FOREACH(mp, &mountlist, mnt_list) {
320 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
321 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
322 			simple_unlock(&mountlist_slock);
323 			return (mp);
324 		}
325 	}
326 	simple_unlock(&mountlist_slock);
327 	return ((struct mount *)0);
328 }
329 
330 /*
331  * Get a new unique fsid
332  */
333 void
334 vfs_getnewfsid(mp)
335 	struct mount *mp;
336 {
337 	static u_short xxxfs_mntid;
338 	fsid_t tfsid;
339 	int mtype;
340 
341 	simple_lock(&mntid_slock);
342 	mtype = makefstype(mp->mnt_op->vfs_name);
343 	mp->mnt_stat.f_fsid.val[0] = makedev(mtype, 0);
344 	mp->mnt_stat.f_fsid.val[1] = mtype;
345 	if (xxxfs_mntid == 0)
346 		++xxxfs_mntid;
347 	tfsid.val[0] = makedev(mtype & 0xff, xxxfs_mntid);
348 	tfsid.val[1] = mtype;
349 	if (!CIRCLEQ_EMPTY(&mountlist)) {
350 		while (vfs_getvfs(&tfsid)) {
351 			tfsid.val[0]++;
352 			xxxfs_mntid++;
353 		}
354 	}
355 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
356 	simple_unlock(&mntid_slock);
357 }
358 
359 /*
360  * Make a 'unique' number from a mount type name.
361  */
362 long
363 makefstype(type)
364 	const char *type;
365 {
366 	long rv;
367 
368 	for (rv = 0; *type; type++) {
369 		rv <<= 2;
370 		rv ^= *type;
371 	}
372 	return rv;
373 }
374 
375 
376 /*
377  * Set vnode attributes to VNOVAL
378  */
379 void
380 vattr_null(vap)
381 	struct vattr *vap;
382 {
383 
384 	vap->va_type = VNON;
385 
386 	/*
387 	 * Assign individually so that it is safe even if size and
388 	 * sign of each member are varied.
389 	 */
390 	vap->va_mode = VNOVAL;
391 	vap->va_nlink = VNOVAL;
392 	vap->va_uid = VNOVAL;
393 	vap->va_gid = VNOVAL;
394 	vap->va_fsid = VNOVAL;
395 	vap->va_fileid = VNOVAL;
396 	vap->va_size = VNOVAL;
397 	vap->va_blocksize = VNOVAL;
398 	vap->va_atime.tv_sec =
399 	    vap->va_mtime.tv_sec =
400 	    vap->va_ctime.tv_sec =
401 	    vap->va_birthtime.tv_sec = VNOVAL;
402 	vap->va_atime.tv_nsec =
403 	    vap->va_mtime.tv_nsec =
404 	    vap->va_ctime.tv_nsec =
405 	    vap->va_birthtime.tv_nsec = VNOVAL;
406 	vap->va_gen = VNOVAL;
407 	vap->va_flags = VNOVAL;
408 	vap->va_rdev = VNOVAL;
409 	vap->va_bytes = VNOVAL;
410 	vap->va_vaflags = 0;
411 }
412 
413 /*
414  * Routines having to do with the management of the vnode table.
415  */
416 extern int (**dead_vnodeop_p) __P((void *));
417 long numvnodes;
418 
419 /*
420  * Return the next vnode from the free list.
421  */
422 int
423 getnewvnode(tag, mp, vops, vpp)
424 	enum vtagtype tag;
425 	struct mount *mp;
426 	int (**vops) __P((void *));
427 	struct vnode **vpp;
428 {
429 	extern struct uvm_pagerops uvm_vnodeops;
430 	struct uvm_object *uobj;
431 	struct proc *p = curproc;	/* XXX */
432 	struct freelst *listhd;
433 	static int toggle;
434 	struct vnode *vp;
435 	int error = 0, tryalloc;
436 
437  try_again:
438 	if (mp) {
439 		/*
440 		 * Mark filesystem busy while we're creating a vnode.
441 		 * If unmount is in progress, this will wait; if the
442 		 * unmount succeeds (only if umount -f), this will
443 		 * return an error.  If the unmount fails, we'll keep
444 		 * going afterwards.
445 		 * (This puts the per-mount vnode list logically under
446 		 * the protection of the vfs_busy lock).
447 		 */
448 		error = vfs_busy(mp, LK_RECURSEFAIL, 0);
449 		if (error && error != EDEADLK)
450 			return error;
451 	}
452 
453 	/*
454 	 * We must choose whether to allocate a new vnode or recycle an
455 	 * existing one. The criterion for allocating a new one is that
456 	 * the total number of vnodes is less than the number desired or
457 	 * there are no vnodes on either free list. Generally we only
458 	 * want to recycle vnodes that have no buffers associated with
459 	 * them, so we look first on the vnode_free_list. If it is empty,
460 	 * we next consider vnodes with referencing buffers on the
461 	 * vnode_hold_list. The toggle ensures that half the time we
462 	 * will use a buffer from the vnode_hold_list, and half the time
463 	 * we will allocate a new one unless the list has grown to twice
464 	 * the desired size. We are reticent to recycle vnodes from the
465 	 * vnode_hold_list because we will lose the identity of all its
466 	 * referencing buffers.
467 	 */
468 
469 	vp = NULL;
470 
471 	simple_lock(&vnode_free_list_slock);
472 
473 	toggle ^= 1;
474 	if (numvnodes > 2 * desiredvnodes)
475 		toggle = 0;
476 
477 	tryalloc = numvnodes < desiredvnodes ||
478 	    (TAILQ_FIRST(&vnode_free_list) == NULL &&
479 	     (TAILQ_FIRST(&vnode_hold_list) == NULL || toggle));
480 
481 	if (tryalloc &&
482 	    (vp = pool_get(&vnode_pool, PR_NOWAIT)) != NULL) {
483 		simple_unlock(&vnode_free_list_slock);
484 		memset(vp, 0, sizeof(*vp));
485 		simple_lock_init(&vp->v_interlock);
486 		uobj = &vp->v_uobj;
487 		uobj->pgops = &uvm_vnodeops;
488 		uobj->uo_npages = 0;
489 		TAILQ_INIT(&uobj->memq);
490 		numvnodes++;
491 	} else {
492 		if ((vp = TAILQ_FIRST(listhd = &vnode_free_list)) == NULL)
493 			vp = TAILQ_FIRST(listhd = &vnode_hold_list);
494 		for (; vp != NULL; vp = TAILQ_NEXT(vp, v_freelist)) {
495 			if (simple_lock_try(&vp->v_interlock)) {
496 				if ((vp->v_flag & VLAYER) == 0) {
497 					break;
498 				}
499 				if (VOP_ISLOCKED(vp) == 0)
500 					break;
501 				else
502 					simple_unlock(&vp->v_interlock);
503 			}
504 		}
505 		/*
506 		 * Unless this is a bad time of the month, at most
507 		 * the first NCPUS items on the free list are
508 		 * locked, so this is close enough to being empty.
509 		 */
510 		if (vp == NULLVP) {
511 			simple_unlock(&vnode_free_list_slock);
512 			if (mp && error != EDEADLK)
513 				vfs_unbusy(mp);
514 			if (tryalloc) {
515 				printf("WARNING: unable to allocate new "
516 				    "vnode, retrying...\n");
517 				(void) tsleep(&lbolt, PRIBIO, "newvn", hz);
518 				goto try_again;
519 			}
520 			tablefull("vnode", "increase kern.maxvnodes or NVNODE");
521 			*vpp = 0;
522 			return (ENFILE);
523 		}
524 		if (vp->v_usecount)
525 			panic("free vnode isn't, vp %p", vp);
526 		TAILQ_REMOVE(listhd, vp, v_freelist);
527 		/* see comment on why 0xdeadb is set at end of vgone (below) */
528 		vp->v_freelist.tqe_prev = (struct vnode **)0xdeadb;
529 		simple_unlock(&vnode_free_list_slock);
530 		vp->v_lease = NULL;
531 
532 		if (vp->v_type != VBAD)
533 			vgonel(vp, p);
534 		else
535 			simple_unlock(&vp->v_interlock);
536 #ifdef DIAGNOSTIC
537 		if (vp->v_data || vp->v_uobj.uo_npages ||
538 		    TAILQ_FIRST(&vp->v_uobj.memq))
539 			panic("cleaned vnode isn't, vp %p", vp);
540 		if (vp->v_numoutput)
541 			panic("clean vnode has pending I/O's, vp %p", vp);
542 #endif
543 		KASSERT((vp->v_flag & VONWORKLST) == 0);
544 		vp->v_flag = 0;
545 		vp->v_socket = NULL;
546 #ifdef VERIFIED_EXEC
547 		vp->fp_status = FINGERPRINT_INVALID;
548 #endif
549 	}
550 	vp->v_type = VNON;
551 	vp->v_vnlock = &vp->v_lock;
552 	lockinit(vp->v_vnlock, PVFS, "vnlock", 0, 0);
553 	cache_purge(vp);
554 	vp->v_tag = tag;
555 	vp->v_op = vops;
556 	insmntque(vp, mp);
557 	*vpp = vp;
558 	vp->v_usecount = 1;
559 	vp->v_data = 0;
560 	simple_lock_init(&vp->v_uobj.vmobjlock);
561 
562 	/*
563 	 * initialize uvm_object within vnode.
564 	 */
565 
566 	uobj = &vp->v_uobj;
567 	KASSERT(uobj->pgops == &uvm_vnodeops);
568 	KASSERT(uobj->uo_npages == 0);
569 	KASSERT(TAILQ_FIRST(&uobj->memq) == NULL);
570 	vp->v_size = VSIZENOTSET;
571 
572 	if (mp && error != EDEADLK)
573 		vfs_unbusy(mp);
574 	return (0);
575 }
576 
577 /*
578  * This is really just the reverse of getnewvnode(). Needed for
579  * VFS_VGET functions who may need to push back a vnode in case
580  * of a locking race.
581  */
582 void
583 ungetnewvnode(vp)
584 	struct vnode *vp;
585 {
586 #ifdef DIAGNOSTIC
587 	if (vp->v_usecount != 1)
588 		panic("ungetnewvnode: busy vnode");
589 #endif
590 	vp->v_usecount--;
591 	insmntque(vp, NULL);
592 	vp->v_type = VBAD;
593 
594 	simple_lock(&vp->v_interlock);
595 	/*
596 	 * Insert at head of LRU list
597 	 */
598 	simple_lock(&vnode_free_list_slock);
599 	if (vp->v_holdcnt > 0)
600 		TAILQ_INSERT_HEAD(&vnode_hold_list, vp, v_freelist);
601 	else
602 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
603 	simple_unlock(&vnode_free_list_slock);
604 	simple_unlock(&vp->v_interlock);
605 }
606 
607 /*
608  * Move a vnode from one mount queue to another.
609  */
610 void
611 insmntque(vp, mp)
612 	struct vnode *vp;
613 	struct mount *mp;
614 {
615 
616 #ifdef DIAGNOSTIC
617 	if ((mp != NULL) &&
618 	    (mp->mnt_flag & MNT_UNMOUNT) &&
619 	    !(mp->mnt_flag & MNT_SOFTDEP) &&
620 	    vp->v_tag != VT_VFS) {
621 		panic("insmntque into dying filesystem");
622 	}
623 #endif
624 
625 	simple_lock(&mntvnode_slock);
626 	/*
627 	 * Delete from old mount point vnode list, if on one.
628 	 */
629 	if (vp->v_mount != NULL)
630 		LIST_REMOVE(vp, v_mntvnodes);
631 	/*
632 	 * Insert into list of vnodes for the new mount point, if available.
633 	 */
634 	if ((vp->v_mount = mp) != NULL)
635 		LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes);
636 	simple_unlock(&mntvnode_slock);
637 }
638 
639 /*
640  * Update outstanding I/O count and do wakeup if requested.
641  */
642 void
643 vwakeup(bp)
644 	struct buf *bp;
645 {
646 	struct vnode *vp;
647 
648 	if ((vp = bp->b_vp) != NULL) {
649 		/* XXX global lock hack
650 		 * can't use v_interlock here since this is called
651 		 * in interrupt context from biodone().
652 		 */
653 		simple_lock(&global_v_numoutput_slock);
654 		if (--vp->v_numoutput < 0)
655 			panic("vwakeup: neg numoutput, vp %p", vp);
656 		if ((vp->v_flag & VBWAIT) && vp->v_numoutput <= 0) {
657 			vp->v_flag &= ~VBWAIT;
658 			wakeup((caddr_t)&vp->v_numoutput);
659 		}
660 		simple_unlock(&global_v_numoutput_slock);
661 	}
662 }
663 
664 /*
665  * Flush out and invalidate all buffers associated with a vnode.
666  * Called with the underlying vnode locked, which should prevent new dirty
667  * buffers from being queued.
668  */
669 int
670 vinvalbuf(vp, flags, cred, p, slpflag, slptimeo)
671 	struct vnode *vp;
672 	int flags;
673 	struct ucred *cred;
674 	struct proc *p;
675 	int slpflag, slptimeo;
676 {
677 	struct buf *bp, *nbp;
678 	int s, error;
679 	int flushflags = PGO_ALLPAGES | PGO_FREE | PGO_SYNCIO |
680 		(flags & V_SAVE ? PGO_CLEANIT : 0);
681 
682 	/* XXXUBC this doesn't look at flags or slp* */
683 	simple_lock(&vp->v_interlock);
684 	error = VOP_PUTPAGES(vp, 0, 0, flushflags);
685 	if (error) {
686 		return error;
687 	}
688 
689 	if (flags & V_SAVE) {
690 		error = VOP_FSYNC(vp, cred, FSYNC_WAIT|FSYNC_RECLAIM, 0, 0, p);
691 		if (error)
692 		        return (error);
693 #ifdef DIAGNOSTIC
694 		s = splbio();
695 		if (vp->v_numoutput > 0 || !LIST_EMPTY(&vp->v_dirtyblkhd))
696 		        panic("vinvalbuf: dirty bufs, vp %p", vp);
697 		splx(s);
698 #endif
699 	}
700 
701 	s = splbio();
702 
703 restart:
704 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
705 		nbp = LIST_NEXT(bp, b_vnbufs);
706 		simple_lock(&bp->b_interlock);
707 		if (bp->b_flags & B_BUSY) {
708 			bp->b_flags |= B_WANTED;
709 			error = ltsleep((caddr_t)bp,
710 				    slpflag | (PRIBIO + 1) | PNORELOCK,
711 				    "vinvalbuf", slptimeo, &bp->b_interlock);
712 			if (error) {
713 				splx(s);
714 				return (error);
715 			}
716 			goto restart;
717 		}
718 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
719 		simple_unlock(&bp->b_interlock);
720 		brelse(bp);
721 	}
722 
723 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
724 		nbp = LIST_NEXT(bp, b_vnbufs);
725 		simple_lock(&bp->b_interlock);
726 		if (bp->b_flags & B_BUSY) {
727 			bp->b_flags |= B_WANTED;
728 			error = ltsleep((caddr_t)bp,
729 				    slpflag | (PRIBIO + 1) | PNORELOCK,
730 				    "vinvalbuf", slptimeo, &bp->b_interlock);
731 			if (error) {
732 				splx(s);
733 				return (error);
734 			}
735 			goto restart;
736 		}
737 		/*
738 		 * XXX Since there are no node locks for NFS, I believe
739 		 * there is a slight chance that a delayed write will
740 		 * occur while sleeping just above, so check for it.
741 		 */
742 		if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) {
743 #ifdef DEBUG
744 			printf("buffer still DELWRI\n");
745 #endif
746 			bp->b_flags |= B_BUSY | B_VFLUSH;
747 			simple_unlock(&bp->b_interlock);
748 			VOP_BWRITE(bp);
749 			goto restart;
750 		}
751 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
752 		simple_unlock(&bp->b_interlock);
753 		brelse(bp);
754 	}
755 
756 #ifdef DIAGNOSTIC
757 	if (!LIST_EMPTY(&vp->v_cleanblkhd) || !LIST_EMPTY(&vp->v_dirtyblkhd))
758 		panic("vinvalbuf: flush failed, vp %p", vp);
759 #endif
760 
761 	splx(s);
762 
763 	return (0);
764 }
765 
766 /*
767  * Destroy any in core blocks past the truncation length.
768  * Called with the underlying vnode locked, which should prevent new dirty
769  * buffers from being queued.
770  */
771 int
772 vtruncbuf(vp, lbn, slpflag, slptimeo)
773 	struct vnode *vp;
774 	daddr_t lbn;
775 	int slpflag, slptimeo;
776 {
777 	struct buf *bp, *nbp;
778 	int s, error;
779 	voff_t off;
780 
781 	off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift);
782 	simple_lock(&vp->v_interlock);
783 	error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO);
784 	if (error) {
785 		return error;
786 	}
787 
788 	s = splbio();
789 
790 restart:
791 	for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
792 		nbp = LIST_NEXT(bp, b_vnbufs);
793 		if (bp->b_lblkno < lbn)
794 			continue;
795 		simple_lock(&bp->b_interlock);
796 		if (bp->b_flags & B_BUSY) {
797 			bp->b_flags |= B_WANTED;
798 			error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK,
799 			    "vtruncbuf", slptimeo, &bp->b_interlock);
800 			if (error) {
801 				splx(s);
802 				return (error);
803 			}
804 			goto restart;
805 		}
806 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
807 		simple_unlock(&bp->b_interlock);
808 		brelse(bp);
809 	}
810 
811 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
812 		nbp = LIST_NEXT(bp, b_vnbufs);
813 		if (bp->b_lblkno < lbn)
814 			continue;
815 		simple_lock(&bp->b_interlock);
816 		if (bp->b_flags & B_BUSY) {
817 			bp->b_flags |= B_WANTED;
818 			error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK,
819 			    "vtruncbuf", slptimeo, &bp->b_interlock);
820 			if (error) {
821 				splx(s);
822 				return (error);
823 			}
824 			goto restart;
825 		}
826 		bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH;
827 		simple_unlock(&bp->b_interlock);
828 		brelse(bp);
829 	}
830 
831 	splx(s);
832 
833 	return (0);
834 }
835 
836 void
837 vflushbuf(vp, sync)
838 	struct vnode *vp;
839 	int sync;
840 {
841 	struct buf *bp, *nbp;
842 	int flags = PGO_CLEANIT | PGO_ALLPAGES | (sync ? PGO_SYNCIO : 0);
843 	int s;
844 
845 	simple_lock(&vp->v_interlock);
846 	(void) VOP_PUTPAGES(vp, 0, 0, flags);
847 
848 loop:
849 	s = splbio();
850 	for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
851 		nbp = LIST_NEXT(bp, b_vnbufs);
852 		simple_lock(&bp->b_interlock);
853 		if ((bp->b_flags & B_BUSY)) {
854 			simple_unlock(&bp->b_interlock);
855 			continue;
856 		}
857 		if ((bp->b_flags & B_DELWRI) == 0)
858 			panic("vflushbuf: not dirty, bp %p", bp);
859 		bp->b_flags |= B_BUSY | B_VFLUSH;
860 		simple_unlock(&bp->b_interlock);
861 		splx(s);
862 		/*
863 		 * Wait for I/O associated with indirect blocks to complete,
864 		 * since there is no way to quickly wait for them below.
865 		 */
866 		if (bp->b_vp == vp || sync == 0)
867 			(void) bawrite(bp);
868 		else
869 			(void) bwrite(bp);
870 		goto loop;
871 	}
872 	if (sync == 0) {
873 		splx(s);
874 		return;
875 	}
876 	simple_lock(&global_v_numoutput_slock);
877 	while (vp->v_numoutput) {
878 		vp->v_flag |= VBWAIT;
879 		ltsleep((caddr_t)&vp->v_numoutput, PRIBIO + 1, "vflushbuf", 0,
880 			&global_v_numoutput_slock);
881 	}
882 	simple_unlock(&global_v_numoutput_slock);
883 	splx(s);
884 	if (!LIST_EMPTY(&vp->v_dirtyblkhd)) {
885 		vprint("vflushbuf: dirty", vp);
886 		goto loop;
887 	}
888 }
889 
890 /*
891  * Associate a buffer with a vnode.
892  */
893 void
894 bgetvp(vp, bp)
895 	struct vnode *vp;
896 	struct buf *bp;
897 {
898 	int s;
899 
900 	if (bp->b_vp)
901 		panic("bgetvp: not free, bp %p", bp);
902 	VHOLD(vp);
903 	s = splbio();
904 	bp->b_vp = vp;
905 	if (vp->v_type == VBLK || vp->v_type == VCHR)
906 		bp->b_dev = vp->v_rdev;
907 	else
908 		bp->b_dev = NODEV;
909 	/*
910 	 * Insert onto list for new vnode.
911 	 */
912 	bufinsvn(bp, &vp->v_cleanblkhd);
913 	splx(s);
914 }
915 
916 /*
917  * Disassociate a buffer from a vnode.
918  */
919 void
920 brelvp(bp)
921 	struct buf *bp;
922 {
923 	struct vnode *vp;
924 	int s;
925 
926 	if (bp->b_vp == NULL)
927 		panic("brelvp: vp NULL, bp %p", bp);
928 
929 	s = splbio();
930 	vp = bp->b_vp;
931 	/*
932 	 * Delete from old vnode list, if on one.
933 	 */
934 	if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
935 		bufremvn(bp);
936 
937 	if (TAILQ_EMPTY(&vp->v_uobj.memq) && (vp->v_flag & VONWORKLST) &&
938 	    LIST_FIRST(&vp->v_dirtyblkhd) == NULL) {
939 		vp->v_flag &= ~VONWORKLST;
940 		LIST_REMOVE(vp, v_synclist);
941 	}
942 
943 	bp->b_vp = NULL;
944 	HOLDRELE(vp);
945 	splx(s);
946 }
947 
948 /*
949  * Reassign a buffer from one vnode to another.
950  * Used to assign file specific control information
951  * (indirect blocks) to the vnode to which they belong.
952  *
953  * This function must be called at splbio().
954  */
955 void
956 reassignbuf(bp, newvp)
957 	struct buf *bp;
958 	struct vnode *newvp;
959 {
960 	struct buflists *listheadp;
961 	int delay;
962 
963 	/*
964 	 * Delete from old vnode list, if on one.
965 	 */
966 	if (LIST_NEXT(bp, b_vnbufs) != NOLIST)
967 		bufremvn(bp);
968 	/*
969 	 * If dirty, put on list of dirty buffers;
970 	 * otherwise insert onto list of clean buffers.
971 	 */
972 	if ((bp->b_flags & B_DELWRI) == 0) {
973 		listheadp = &newvp->v_cleanblkhd;
974 		if (TAILQ_EMPTY(&newvp->v_uobj.memq) &&
975 		    (newvp->v_flag & VONWORKLST) &&
976 		    LIST_FIRST(&newvp->v_dirtyblkhd) == NULL) {
977 			newvp->v_flag &= ~VONWORKLST;
978 			LIST_REMOVE(newvp, v_synclist);
979 		}
980 	} else {
981 		listheadp = &newvp->v_dirtyblkhd;
982 		if ((newvp->v_flag & VONWORKLST) == 0) {
983 			switch (newvp->v_type) {
984 			case VDIR:
985 				delay = dirdelay;
986 				break;
987 			case VBLK:
988 				if (newvp->v_specmountpoint != NULL) {
989 					delay = metadelay;
990 					break;
991 				}
992 				/* fall through */
993 			default:
994 				delay = filedelay;
995 				break;
996 			}
997 			if (!newvp->v_mount ||
998 			    (newvp->v_mount->mnt_flag & MNT_ASYNC) == 0)
999 				vn_syncer_add_to_worklist(newvp, delay);
1000 		}
1001 	}
1002 	bufinsvn(bp, listheadp);
1003 }
1004 
1005 /*
1006  * Create a vnode for a block device.
1007  * Used for root filesystem and swap areas.
1008  * Also used for memory file system special devices.
1009  */
1010 int
1011 bdevvp(dev, vpp)
1012 	dev_t dev;
1013 	struct vnode **vpp;
1014 {
1015 
1016 	return (getdevvp(dev, vpp, VBLK));
1017 }
1018 
1019 /*
1020  * Create a vnode for a character device.
1021  * Used for kernfs and some console handling.
1022  */
1023 int
1024 cdevvp(dev, vpp)
1025 	dev_t dev;
1026 	struct vnode **vpp;
1027 {
1028 
1029 	return (getdevvp(dev, vpp, VCHR));
1030 }
1031 
1032 /*
1033  * Create a vnode for a device.
1034  * Used by bdevvp (block device) for root file system etc.,
1035  * and by cdevvp (character device) for console and kernfs.
1036  */
1037 int
1038 getdevvp(dev, vpp, type)
1039 	dev_t dev;
1040 	struct vnode **vpp;
1041 	enum vtype type;
1042 {
1043 	struct vnode *vp;
1044 	struct vnode *nvp;
1045 	int error;
1046 
1047 	if (dev == NODEV) {
1048 		*vpp = NULLVP;
1049 		return (0);
1050 	}
1051 	error = getnewvnode(VT_NON, NULL, spec_vnodeop_p, &nvp);
1052 	if (error) {
1053 		*vpp = NULLVP;
1054 		return (error);
1055 	}
1056 	vp = nvp;
1057 	vp->v_type = type;
1058 	if ((nvp = checkalias(vp, dev, NULL)) != 0) {
1059 		vput(vp);
1060 		vp = nvp;
1061 	}
1062 	*vpp = vp;
1063 	return (0);
1064 }
1065 
1066 /*
1067  * Check to see if the new vnode represents a special device
1068  * for which we already have a vnode (either because of
1069  * bdevvp() or because of a different vnode representing
1070  * the same block device). If such an alias exists, deallocate
1071  * the existing contents and return the aliased vnode. The
1072  * caller is responsible for filling it with its new contents.
1073  */
1074 struct vnode *
1075 checkalias(nvp, nvp_rdev, mp)
1076 	struct vnode *nvp;
1077 	dev_t nvp_rdev;
1078 	struct mount *mp;
1079 {
1080 	struct proc *p = curproc;       /* XXX */
1081 	struct vnode *vp;
1082 	struct vnode **vpp;
1083 
1084 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1085 		return (NULLVP);
1086 
1087 	vpp = &speclisth[SPECHASH(nvp_rdev)];
1088 loop:
1089 	simple_lock(&spechash_slock);
1090 	for (vp = *vpp; vp; vp = vp->v_specnext) {
1091 		if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type)
1092 			continue;
1093 		/*
1094 		 * Alias, but not in use, so flush it out.
1095 		 */
1096 		simple_lock(&vp->v_interlock);
1097 		if (vp->v_usecount == 0) {
1098 			simple_unlock(&spechash_slock);
1099 			vgonel(vp, p);
1100 			goto loop;
1101 		}
1102 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK)) {
1103 			simple_unlock(&spechash_slock);
1104 			goto loop;
1105 		}
1106 		break;
1107 	}
1108 	if (vp == NULL || vp->v_tag != VT_NON || vp->v_type != VBLK) {
1109 		MALLOC(nvp->v_specinfo, struct specinfo *,
1110 			sizeof(struct specinfo), M_VNODE, M_NOWAIT);
1111 		/* XXX Erg. */
1112 		if (nvp->v_specinfo == NULL) {
1113 			simple_unlock(&spechash_slock);
1114 			uvm_wait("checkalias");
1115 			goto loop;
1116 		}
1117 
1118 		nvp->v_rdev = nvp_rdev;
1119 		nvp->v_hashchain = vpp;
1120 		nvp->v_specnext = *vpp;
1121 		nvp->v_specmountpoint = NULL;
1122 		simple_unlock(&spechash_slock);
1123 		nvp->v_speclockf = NULL;
1124 		*vpp = nvp;
1125 		if (vp != NULLVP) {
1126 			nvp->v_flag |= VALIASED;
1127 			vp->v_flag |= VALIASED;
1128 			vput(vp);
1129 		}
1130 		return (NULLVP);
1131 	}
1132 	simple_unlock(&spechash_slock);
1133 	VOP_UNLOCK(vp, 0);
1134 	simple_lock(&vp->v_interlock);
1135 	vclean(vp, 0, p);
1136 	vp->v_op = nvp->v_op;
1137 	vp->v_tag = nvp->v_tag;
1138 	vp->v_vnlock = &vp->v_lock;
1139 	lockinit(vp->v_vnlock, PVFS, "vnlock", 0, 0);
1140 	nvp->v_type = VNON;
1141 	insmntque(vp, mp);
1142 	return (vp);
1143 }
1144 
1145 /*
1146  * Grab a particular vnode from the free list, increment its
1147  * reference count and lock it. If the vnode lock bit is set the
1148  * vnode is being eliminated in vgone. In that case, we can not
1149  * grab the vnode, so the process is awakened when the transition is
1150  * completed, and an error returned to indicate that the vnode is no
1151  * longer usable (possibly having been changed to a new file system type).
1152  */
1153 int
1154 vget(vp, flags)
1155 	struct vnode *vp;
1156 	int flags;
1157 {
1158 	int error;
1159 
1160 	/*
1161 	 * If the vnode is in the process of being cleaned out for
1162 	 * another use, we wait for the cleaning to finish and then
1163 	 * return failure. Cleaning is determined by checking that
1164 	 * the VXLOCK flag is set.
1165 	 */
1166 
1167 	if ((flags & LK_INTERLOCK) == 0)
1168 		simple_lock(&vp->v_interlock);
1169 	if (vp->v_flag & VXLOCK) {
1170 		if (flags & LK_NOWAIT) {
1171 			simple_unlock(&vp->v_interlock);
1172 			return EBUSY;
1173 		}
1174 		vp->v_flag |= VXWANT;
1175 		ltsleep(vp, PINOD|PNORELOCK, "vget", 0, &vp->v_interlock);
1176 		return (ENOENT);
1177 	}
1178 	if (vp->v_usecount == 0) {
1179 		simple_lock(&vnode_free_list_slock);
1180 		if (vp->v_holdcnt > 0)
1181 			TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist);
1182 		else
1183 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1184 		simple_unlock(&vnode_free_list_slock);
1185 	}
1186 	vp->v_usecount++;
1187 #ifdef DIAGNOSTIC
1188 	if (vp->v_usecount == 0) {
1189 		vprint("vget", vp);
1190 		panic("vget: usecount overflow, vp %p", vp);
1191 	}
1192 #endif
1193 	if (flags & LK_TYPE_MASK) {
1194 		if ((error = vn_lock(vp, flags | LK_INTERLOCK))) {
1195 			/*
1196 			 * must expand vrele here because we do not want
1197 			 * to call VOP_INACTIVE if the reference count
1198 			 * drops back to zero since it was never really
1199 			 * active. We must remove it from the free list
1200 			 * before sleeping so that multiple processes do
1201 			 * not try to recycle it.
1202 			 */
1203 			simple_lock(&vp->v_interlock);
1204 			vp->v_usecount--;
1205 			if (vp->v_usecount > 0) {
1206 				simple_unlock(&vp->v_interlock);
1207 				return (error);
1208 			}
1209 			/*
1210 			 * insert at tail of LRU list
1211 			 */
1212 			simple_lock(&vnode_free_list_slock);
1213 			if (vp->v_holdcnt > 0)
1214 				TAILQ_INSERT_TAIL(&vnode_hold_list, vp,
1215 				    v_freelist);
1216 			else
1217 				TAILQ_INSERT_TAIL(&vnode_free_list, vp,
1218 				    v_freelist);
1219 			simple_unlock(&vnode_free_list_slock);
1220 			simple_unlock(&vp->v_interlock);
1221 		}
1222 		return (error);
1223 	}
1224 	simple_unlock(&vp->v_interlock);
1225 	return (0);
1226 }
1227 
1228 /*
1229  * vput(), just unlock and vrele()
1230  */
1231 void
1232 vput(vp)
1233 	struct vnode *vp;
1234 {
1235 	struct proc *p = curproc;	/* XXX */
1236 
1237 #ifdef DIAGNOSTIC
1238 	if (vp == NULL)
1239 		panic("vput: null vp");
1240 #endif
1241 	simple_lock(&vp->v_interlock);
1242 	vp->v_usecount--;
1243 	if (vp->v_usecount > 0) {
1244 		simple_unlock(&vp->v_interlock);
1245 		VOP_UNLOCK(vp, 0);
1246 		return;
1247 	}
1248 #ifdef DIAGNOSTIC
1249 	if (vp->v_usecount < 0 || vp->v_writecount != 0) {
1250 		vprint("vput: bad ref count", vp);
1251 		panic("vput: ref cnt");
1252 	}
1253 #endif
1254 	/*
1255 	 * Insert at tail of LRU list.
1256 	 */
1257 	simple_lock(&vnode_free_list_slock);
1258 	if (vp->v_holdcnt > 0)
1259 		TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist);
1260 	else
1261 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
1262 	simple_unlock(&vnode_free_list_slock);
1263 	if (vp->v_flag & VEXECMAP) {
1264 		uvmexp.execpages -= vp->v_uobj.uo_npages;
1265 		uvmexp.filepages += vp->v_uobj.uo_npages;
1266 	}
1267 	vp->v_flag &= ~(VTEXT|VEXECMAP);
1268 	simple_unlock(&vp->v_interlock);
1269 	VOP_INACTIVE(vp, p);
1270 }
1271 
1272 /*
1273  * Vnode release.
1274  * If count drops to zero, call inactive routine and return to freelist.
1275  */
1276 void
1277 vrele(vp)
1278 	struct vnode *vp;
1279 {
1280 	struct proc *p = curproc;	/* XXX */
1281 
1282 #ifdef DIAGNOSTIC
1283 	if (vp == NULL)
1284 		panic("vrele: null vp");
1285 #endif
1286 	simple_lock(&vp->v_interlock);
1287 	vp->v_usecount--;
1288 	if (vp->v_usecount > 0) {
1289 		simple_unlock(&vp->v_interlock);
1290 		return;
1291 	}
1292 #ifdef DIAGNOSTIC
1293 	if (vp->v_usecount < 0 || vp->v_writecount != 0) {
1294 		vprint("vrele: bad ref count", vp);
1295 		panic("vrele: ref cnt vp %p", vp);
1296 	}
1297 #endif
1298 	/*
1299 	 * Insert at tail of LRU list.
1300 	 */
1301 	simple_lock(&vnode_free_list_slock);
1302 	if (vp->v_holdcnt > 0)
1303 		TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist);
1304 	else
1305 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
1306 	simple_unlock(&vnode_free_list_slock);
1307 	if (vp->v_flag & VEXECMAP) {
1308 		uvmexp.execpages -= vp->v_uobj.uo_npages;
1309 		uvmexp.filepages += vp->v_uobj.uo_npages;
1310 	}
1311 	vp->v_flag &= ~(VTEXT|VEXECMAP);
1312 	if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK) == 0)
1313 		VOP_INACTIVE(vp, p);
1314 }
1315 
1316 #ifdef DIAGNOSTIC
1317 /*
1318  * Page or buffer structure gets a reference.
1319  */
1320 void
1321 vhold(vp)
1322 	struct vnode *vp;
1323 {
1324 
1325 	/*
1326 	 * If it is on the freelist and the hold count is currently
1327 	 * zero, move it to the hold list. The test of the back
1328 	 * pointer and the use reference count of zero is because
1329 	 * it will be removed from a free list by getnewvnode,
1330 	 * but will not have its reference count incremented until
1331 	 * after calling vgone. If the reference count were
1332 	 * incremented first, vgone would (incorrectly) try to
1333 	 * close the previous instance of the underlying object.
1334 	 * So, the back pointer is explicitly set to `0xdeadb' in
1335 	 * getnewvnode after removing it from a freelist to ensure
1336 	 * that we do not try to move it here.
1337 	 */
1338   	simple_lock(&vp->v_interlock);
1339 	if ((vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb) &&
1340 	    vp->v_holdcnt == 0 && vp->v_usecount == 0) {
1341 		simple_lock(&vnode_free_list_slock);
1342 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1343 		TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist);
1344 		simple_unlock(&vnode_free_list_slock);
1345 	}
1346 	vp->v_holdcnt++;
1347 	simple_unlock(&vp->v_interlock);
1348 }
1349 
1350 /*
1351  * Page or buffer structure frees a reference.
1352  */
1353 void
1354 holdrele(vp)
1355 	struct vnode *vp;
1356 {
1357 
1358 	simple_lock(&vp->v_interlock);
1359 	if (vp->v_holdcnt <= 0)
1360 		panic("holdrele: holdcnt vp %p", vp);
1361 	vp->v_holdcnt--;
1362 
1363 	/*
1364 	 * If it is on the holdlist and the hold count drops to
1365 	 * zero, move it to the free list. The test of the back
1366 	 * pointer and the use reference count of zero is because
1367 	 * it will be removed from a free list by getnewvnode,
1368 	 * but will not have its reference count incremented until
1369 	 * after calling vgone. If the reference count were
1370 	 * incremented first, vgone would (incorrectly) try to
1371 	 * close the previous instance of the underlying object.
1372 	 * So, the back pointer is explicitly set to `0xdeadb' in
1373 	 * getnewvnode after removing it from a freelist to ensure
1374 	 * that we do not try to move it here.
1375 	 */
1376 
1377 	if ((vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb) &&
1378 	    vp->v_holdcnt == 0 && vp->v_usecount == 0) {
1379 		simple_lock(&vnode_free_list_slock);
1380 		TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist);
1381 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
1382 		simple_unlock(&vnode_free_list_slock);
1383 	}
1384 	simple_unlock(&vp->v_interlock);
1385 }
1386 
1387 /*
1388  * Vnode reference.
1389  */
1390 void
1391 vref(vp)
1392 	struct vnode *vp;
1393 {
1394 
1395 	simple_lock(&vp->v_interlock);
1396 	if (vp->v_usecount <= 0)
1397 		panic("vref used where vget required, vp %p", vp);
1398 	vp->v_usecount++;
1399 #ifdef DIAGNOSTIC
1400 	if (vp->v_usecount == 0) {
1401 		vprint("vref", vp);
1402 		panic("vref: usecount overflow, vp %p", vp);
1403 	}
1404 #endif
1405 	simple_unlock(&vp->v_interlock);
1406 }
1407 #endif /* DIAGNOSTIC */
1408 
1409 /*
1410  * Remove any vnodes in the vnode table belonging to mount point mp.
1411  *
1412  * If FORCECLOSE is not specified, there should not be any active ones,
1413  * return error if any are found (nb: this is a user error, not a
1414  * system error). If FORCECLOSE is specified, detach any active vnodes
1415  * that are found.
1416  *
1417  * If WRITECLOSE is set, only flush out regular file vnodes open for
1418  * writing.
1419  *
1420  * SKIPSYSTEM causes any vnodes marked V_SYSTEM to be skipped.
1421  */
1422 #ifdef DEBUG
1423 int busyprt = 0;	/* print out busy vnodes */
1424 struct ctldebug debug1 = { "busyprt", &busyprt };
1425 #endif
1426 
1427 int
1428 vflush(mp, skipvp, flags)
1429 	struct mount *mp;
1430 	struct vnode *skipvp;
1431 	int flags;
1432 {
1433 	struct proc *p = curproc;	/* XXX */
1434 	struct vnode *vp, *nvp;
1435 	int busy = 0;
1436 
1437 	simple_lock(&mntvnode_slock);
1438 loop:
1439 	for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp; vp = nvp) {
1440 		if (vp->v_mount != mp)
1441 			goto loop;
1442 		nvp = LIST_NEXT(vp, v_mntvnodes);
1443 		/*
1444 		 * Skip over a selected vnode.
1445 		 */
1446 		if (vp == skipvp)
1447 			continue;
1448 		simple_lock(&vp->v_interlock);
1449 		/*
1450 		 * Skip over a vnodes marked VSYSTEM.
1451 		 */
1452 		if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1453 			simple_unlock(&vp->v_interlock);
1454 			continue;
1455 		}
1456 		/*
1457 		 * If WRITECLOSE is set, only flush out regular file
1458 		 * vnodes open for writing.
1459 		 */
1460 		if ((flags & WRITECLOSE) &&
1461 		    (vp->v_writecount == 0 || vp->v_type != VREG)) {
1462 			simple_unlock(&vp->v_interlock);
1463 			continue;
1464 		}
1465 		/*
1466 		 * With v_usecount == 0, all we need to do is clear
1467 		 * out the vnode data structures and we are done.
1468 		 */
1469 		if (vp->v_usecount == 0) {
1470 			simple_unlock(&mntvnode_slock);
1471 			vgonel(vp, p);
1472 			simple_lock(&mntvnode_slock);
1473 			continue;
1474 		}
1475 		/*
1476 		 * If FORCECLOSE is set, forcibly close the vnode.
1477 		 * For block or character devices, revert to an
1478 		 * anonymous device. For all other files, just kill them.
1479 		 */
1480 		if (flags & FORCECLOSE) {
1481 			simple_unlock(&mntvnode_slock);
1482 			if (vp->v_type != VBLK && vp->v_type != VCHR) {
1483 				vgonel(vp, p);
1484 			} else {
1485 				vclean(vp, 0, p);
1486 				vp->v_op = spec_vnodeop_p;
1487 				insmntque(vp, (struct mount *)0);
1488 			}
1489 			simple_lock(&mntvnode_slock);
1490 			continue;
1491 		}
1492 #ifdef DEBUG
1493 		if (busyprt)
1494 			vprint("vflush: busy vnode", vp);
1495 #endif
1496 		simple_unlock(&vp->v_interlock);
1497 		busy++;
1498 	}
1499 	simple_unlock(&mntvnode_slock);
1500 	if (busy)
1501 		return (EBUSY);
1502 	return (0);
1503 }
1504 
1505 /*
1506  * Disassociate the underlying file system from a vnode.
1507  */
1508 void
1509 vclean(vp, flags, p)
1510 	struct vnode *vp;
1511 	int flags;
1512 	struct proc *p;
1513 {
1514 	int active;
1515 
1516 	LOCK_ASSERT(simple_lock_held(&vp->v_interlock));
1517 
1518 	/*
1519 	 * Check to see if the vnode is in use.
1520 	 * If so we have to reference it before we clean it out
1521 	 * so that its count cannot fall to zero and generate a
1522 	 * race against ourselves to recycle it.
1523 	 */
1524 
1525 	if ((active = vp->v_usecount) != 0) {
1526 		vp->v_usecount++;
1527 #ifdef DIAGNOSTIC
1528 		if (vp->v_usecount == 0) {
1529 			vprint("vclean", vp);
1530 			panic("vclean: usecount overflow");
1531 		}
1532 #endif
1533 	}
1534 
1535 	/*
1536 	 * Prevent the vnode from being recycled or
1537 	 * brought into use while we clean it out.
1538 	 */
1539 	if (vp->v_flag & VXLOCK)
1540 		panic("vclean: deadlock, vp %p", vp);
1541 	vp->v_flag |= VXLOCK;
1542 	if (vp->v_flag & VEXECMAP) {
1543 		uvmexp.execpages -= vp->v_uobj.uo_npages;
1544 		uvmexp.filepages += vp->v_uobj.uo_npages;
1545 	}
1546 	vp->v_flag &= ~(VTEXT|VEXECMAP);
1547 
1548 	/*
1549 	 * Even if the count is zero, the VOP_INACTIVE routine may still
1550 	 * have the object locked while it cleans it out. The VOP_LOCK
1551 	 * ensures that the VOP_INACTIVE routine is done with its work.
1552 	 * For active vnodes, it ensures that no other activity can
1553 	 * occur while the underlying object is being cleaned out.
1554 	 */
1555 	VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK);
1556 
1557 	/*
1558 	 * Clean out any cached data associated with the vnode.
1559 	 */
1560 	if (flags & DOCLOSE) {
1561 		vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0);
1562 		KASSERT((vp->v_flag & VONWORKLST) == 0);
1563 	}
1564 	LOCK_ASSERT(!simple_lock_held(&vp->v_interlock));
1565 
1566 	/*
1567 	 * If purging an active vnode, it must be closed and
1568 	 * deactivated before being reclaimed. Note that the
1569 	 * VOP_INACTIVE will unlock the vnode.
1570 	 */
1571 	if (active) {
1572 		if (flags & DOCLOSE)
1573 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, NULL);
1574 		VOP_INACTIVE(vp, p);
1575 	} else {
1576 		/*
1577 		 * Any other processes trying to obtain this lock must first
1578 		 * wait for VXLOCK to clear, then call the new lock operation.
1579 		 */
1580 		VOP_UNLOCK(vp, 0);
1581 	}
1582 	/*
1583 	 * Reclaim the vnode.
1584 	 */
1585 	if (VOP_RECLAIM(vp, p))
1586 		panic("vclean: cannot reclaim, vp %p", vp);
1587 	if (active) {
1588 		/*
1589 		 * Inline copy of vrele() since VOP_INACTIVE
1590 		 * has already been called.
1591 		 */
1592 		simple_lock(&vp->v_interlock);
1593 		if (--vp->v_usecount <= 0) {
1594 #ifdef DIAGNOSTIC
1595 			if (vp->v_usecount < 0 || vp->v_writecount != 0) {
1596 				vprint("vclean: bad ref count", vp);
1597 				panic("vclean: ref cnt");
1598 			}
1599 #endif
1600 			/*
1601 			 * Insert at tail of LRU list.
1602 			 */
1603 
1604 			simple_unlock(&vp->v_interlock);
1605 			simple_lock(&vnode_free_list_slock);
1606 #ifdef DIAGNOSTIC
1607 			if (vp->v_holdcnt > 0)
1608 				panic("vclean: not clean, vp %p", vp);
1609 #endif
1610 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
1611 			simple_unlock(&vnode_free_list_slock);
1612 		} else
1613 			simple_unlock(&vp->v_interlock);
1614 	}
1615 
1616 	KASSERT(vp->v_uobj.uo_npages == 0);
1617 	cache_purge(vp);
1618 
1619 	/*
1620 	 * Done with purge, notify sleepers of the grim news.
1621 	 */
1622 	vp->v_op = dead_vnodeop_p;
1623 	vp->v_tag = VT_NON;
1624 	simple_lock(&vp->v_interlock);
1625 	VN_KNOTE(vp, NOTE_REVOKE);	/* FreeBSD has this in vn_pollgone() */
1626 	vp->v_flag &= ~VXLOCK;
1627 	if (vp->v_flag & VXWANT) {
1628 		vp->v_flag &= ~VXWANT;
1629 		simple_unlock(&vp->v_interlock);
1630 		wakeup((caddr_t)vp);
1631 	} else
1632 		simple_unlock(&vp->v_interlock);
1633 }
1634 
1635 /*
1636  * Recycle an unused vnode to the front of the free list.
1637  * Release the passed interlock if the vnode will be recycled.
1638  */
1639 int
1640 vrecycle(vp, inter_lkp, p)
1641 	struct vnode *vp;
1642 	struct simplelock *inter_lkp;
1643 	struct proc *p;
1644 {
1645 
1646 	simple_lock(&vp->v_interlock);
1647 	if (vp->v_usecount == 0) {
1648 		if (inter_lkp)
1649 			simple_unlock(inter_lkp);
1650 		vgonel(vp, p);
1651 		return (1);
1652 	}
1653 	simple_unlock(&vp->v_interlock);
1654 	return (0);
1655 }
1656 
1657 /*
1658  * Eliminate all activity associated with a vnode
1659  * in preparation for reuse.
1660  */
1661 void
1662 vgone(vp)
1663 	struct vnode *vp;
1664 {
1665 	struct proc *p = curproc;	/* XXX */
1666 
1667 	simple_lock(&vp->v_interlock);
1668 	vgonel(vp, p);
1669 }
1670 
1671 /*
1672  * vgone, with the vp interlock held.
1673  */
1674 void
1675 vgonel(vp, p)
1676 	struct vnode *vp;
1677 	struct proc *p;
1678 {
1679 	struct vnode *vq;
1680 	struct vnode *vx;
1681 
1682 	LOCK_ASSERT(simple_lock_held(&vp->v_interlock));
1683 
1684 	/*
1685 	 * If a vgone (or vclean) is already in progress,
1686 	 * wait until it is done and return.
1687 	 */
1688 
1689 	if (vp->v_flag & VXLOCK) {
1690 		vp->v_flag |= VXWANT;
1691 		ltsleep(vp, PINOD | PNORELOCK, "vgone", 0, &vp->v_interlock);
1692 		return;
1693 	}
1694 
1695 	/*
1696 	 * Clean out the filesystem specific data.
1697 	 */
1698 
1699 	vclean(vp, DOCLOSE, p);
1700 	KASSERT((vp->v_flag & VONWORKLST) == 0);
1701 
1702 	/*
1703 	 * Delete from old mount point vnode list, if on one.
1704 	 */
1705 
1706 	if (vp->v_mount != NULL)
1707 		insmntque(vp, (struct mount *)0);
1708 
1709 	/*
1710 	 * If special device, remove it from special device alias list.
1711 	 * if it is on one.
1712 	 */
1713 
1714 	if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_specinfo != 0) {
1715 		simple_lock(&spechash_slock);
1716 		if (vp->v_hashchain != NULL) {
1717 			if (*vp->v_hashchain == vp) {
1718 				*vp->v_hashchain = vp->v_specnext;
1719 			} else {
1720 				for (vq = *vp->v_hashchain; vq;
1721 							vq = vq->v_specnext) {
1722 					if (vq->v_specnext != vp)
1723 						continue;
1724 					vq->v_specnext = vp->v_specnext;
1725 					break;
1726 				}
1727 				if (vq == NULL)
1728 					panic("missing bdev");
1729 			}
1730 			if (vp->v_flag & VALIASED) {
1731 				vx = NULL;
1732 				for (vq = *vp->v_hashchain; vq;
1733 							vq = vq->v_specnext) {
1734 					if (vq->v_rdev != vp->v_rdev ||
1735 					    vq->v_type != vp->v_type)
1736 						continue;
1737 					if (vx)
1738 						break;
1739 					vx = vq;
1740 				}
1741 				if (vx == NULL)
1742 					panic("missing alias");
1743 				if (vq == NULL)
1744 					vx->v_flag &= ~VALIASED;
1745 				vp->v_flag &= ~VALIASED;
1746 			}
1747 		}
1748 		simple_unlock(&spechash_slock);
1749 		FREE(vp->v_specinfo, M_VNODE);
1750 		vp->v_specinfo = NULL;
1751 	}
1752 
1753 	/*
1754 	 * If it is on the freelist and not already at the head,
1755 	 * move it to the head of the list. The test of the back
1756 	 * pointer and the reference count of zero is because
1757 	 * it will be removed from the free list by getnewvnode,
1758 	 * but will not have its reference count incremented until
1759 	 * after calling vgone. If the reference count were
1760 	 * incremented first, vgone would (incorrectly) try to
1761 	 * close the previous instance of the underlying object.
1762 	 * So, the back pointer is explicitly set to `0xdeadb' in
1763 	 * getnewvnode after removing it from the freelist to ensure
1764 	 * that we do not try to move it here.
1765 	 */
1766 
1767 	if (vp->v_usecount == 0) {
1768 		simple_lock(&vnode_free_list_slock);
1769 		if (vp->v_holdcnt > 0)
1770 			panic("vgonel: not clean, vp %p", vp);
1771 		if (vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb &&
1772 		    TAILQ_FIRST(&vnode_free_list) != vp) {
1773 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1774 			TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
1775 		}
1776 		simple_unlock(&vnode_free_list_slock);
1777 	}
1778 	vp->v_type = VBAD;
1779 }
1780 
1781 /*
1782  * Lookup a vnode by device number.
1783  */
1784 int
1785 vfinddev(dev, type, vpp)
1786 	dev_t dev;
1787 	enum vtype type;
1788 	struct vnode **vpp;
1789 {
1790 	struct vnode *vp;
1791 	int rc = 0;
1792 
1793 	simple_lock(&spechash_slock);
1794 	for (vp = speclisth[SPECHASH(dev)]; vp; vp = vp->v_specnext) {
1795 		if (dev != vp->v_rdev || type != vp->v_type)
1796 			continue;
1797 		*vpp = vp;
1798 		rc = 1;
1799 		break;
1800 	}
1801 	simple_unlock(&spechash_slock);
1802 	return (rc);
1803 }
1804 
1805 /*
1806  * Revoke all the vnodes corresponding to the specified minor number
1807  * range (endpoints inclusive) of the specified major.
1808  */
1809 void
1810 vdevgone(maj, minl, minh, type)
1811 	int maj, minl, minh;
1812 	enum vtype type;
1813 {
1814 	struct vnode *vp;
1815 	int mn;
1816 
1817 	for (mn = minl; mn <= minh; mn++)
1818 		if (vfinddev(makedev(maj, mn), type, &vp))
1819 			VOP_REVOKE(vp, REVOKEALL);
1820 }
1821 
1822 /*
1823  * Calculate the total number of references to a special device.
1824  */
1825 int
1826 vcount(vp)
1827 	struct vnode *vp;
1828 {
1829 	struct vnode *vq, *vnext;
1830 	int count;
1831 
1832 loop:
1833 	if ((vp->v_flag & VALIASED) == 0)
1834 		return (vp->v_usecount);
1835 	simple_lock(&spechash_slock);
1836 	for (count = 0, vq = *vp->v_hashchain; vq; vq = vnext) {
1837 		vnext = vq->v_specnext;
1838 		if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type)
1839 			continue;
1840 		/*
1841 		 * Alias, but not in use, so flush it out.
1842 		 */
1843 		if (vq->v_usecount == 0 && vq != vp &&
1844 		    (vq->v_flag & VXLOCK) == 0) {
1845 			simple_unlock(&spechash_slock);
1846 			vgone(vq);
1847 			goto loop;
1848 		}
1849 		count += vq->v_usecount;
1850 	}
1851 	simple_unlock(&spechash_slock);
1852 	return (count);
1853 }
1854 
1855 /*
1856  * Print out a description of a vnode.
1857  */
1858 const char * const vnode_types[] = {
1859 	"VNON",
1860 	"VREG",
1861 	"VDIR",
1862 	"VBLK",
1863 	"VCHR",
1864 	"VLNK",
1865 	"VSOCK",
1866 	"VFIFO",
1867 	"VBAD"
1868 };
1869 
1870 void
1871 vprint(label, vp)
1872 	char *label;
1873 	struct vnode *vp;
1874 {
1875 	char buf[96];
1876 
1877 	if (label != NULL)
1878 		printf("%s: ", label);
1879 	printf("tag %d type %s, usecount %d, writecount %ld, refcount %ld,",
1880 	    vp->v_tag, vnode_types[vp->v_type],
1881 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt);
1882 	buf[0] = '\0';
1883 	if (vp->v_flag & VROOT)
1884 		strcat(buf, "|VROOT");
1885 	if (vp->v_flag & VTEXT)
1886 		strcat(buf, "|VTEXT");
1887 	if (vp->v_flag & VEXECMAP)
1888 		strcat(buf, "|VEXECMAP");
1889 	if (vp->v_flag & VSYSTEM)
1890 		strcat(buf, "|VSYSTEM");
1891 	if (vp->v_flag & VXLOCK)
1892 		strcat(buf, "|VXLOCK");
1893 	if (vp->v_flag & VXWANT)
1894 		strcat(buf, "|VXWANT");
1895 	if (vp->v_flag & VBWAIT)
1896 		strcat(buf, "|VBWAIT");
1897 	if (vp->v_flag & VALIASED)
1898 		strcat(buf, "|VALIASED");
1899 	if (buf[0] != '\0')
1900 		printf(" flags (%s)", &buf[1]);
1901 	if (vp->v_data == NULL) {
1902 		printf("\n");
1903 	} else {
1904 		printf("\n\t");
1905 		VOP_PRINT(vp);
1906 	}
1907 }
1908 
1909 #ifdef DEBUG
1910 /*
1911  * List all of the locked vnodes in the system.
1912  * Called when debugging the kernel.
1913  */
1914 void
1915 printlockedvnodes()
1916 {
1917 	struct mount *mp, *nmp;
1918 	struct vnode *vp;
1919 
1920 	printf("Locked vnodes\n");
1921 	simple_lock(&mountlist_slock);
1922 	for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist;
1923 	     mp = nmp) {
1924 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) {
1925 			nmp = CIRCLEQ_NEXT(mp, mnt_list);
1926 			continue;
1927 		}
1928 		LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
1929 			if (VOP_ISLOCKED(vp))
1930 				vprint(NULL, vp);
1931 		}
1932 		simple_lock(&mountlist_slock);
1933 		nmp = CIRCLEQ_NEXT(mp, mnt_list);
1934 		vfs_unbusy(mp);
1935 	}
1936 	simple_unlock(&mountlist_slock);
1937 }
1938 #endif
1939 
1940 /*
1941  * Top level filesystem related information gathering.
1942  */
1943 int
1944 vfs_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p)
1945 	int *name;
1946 	u_int namelen;
1947 	void *oldp;
1948 	size_t *oldlenp;
1949 	void *newp;
1950 	size_t newlen;
1951 	struct proc *p;
1952 {
1953 #if defined(COMPAT_09) || defined(COMPAT_43) || defined(COMPAT_44)
1954 	struct vfsconf vfc;
1955 	extern const char * const mountcompatnames[];
1956 	extern int nmountcompatnames;
1957 #endif
1958 	struct vfsops *vfsp;
1959 
1960 	/* all sysctl names at this level are at least name and field */
1961 	if (namelen < 2)
1962 		return (ENOTDIR);		/* overloaded */
1963 
1964 	/* Not generic: goes to file system. */
1965 	if (name[0] != VFS_GENERIC) {
1966 		static const struct ctlname vfsnames[VFS_MAXID+1]=CTL_VFS_NAMES;
1967 		const char *vfsname;
1968 
1969 		if (name[0] < 0 || name[0] > VFS_MAXID
1970 		    || (vfsname = vfsnames[name[0]].ctl_name) == NULL)
1971 			return (EOPNOTSUPP);
1972 
1973 		vfsp = vfs_getopsbyname(vfsname);
1974 		if (vfsp == NULL || vfsp->vfs_sysctl == NULL)
1975 			return (EOPNOTSUPP);
1976 		return ((*vfsp->vfs_sysctl)(&name[1], namelen - 1,
1977 		    oldp, oldlenp, newp, newlen, p));
1978 	}
1979 
1980 	/* The rest are generic vfs sysctls. */
1981 	switch (name[1]) {
1982 	case VFS_USERMOUNT:
1983 		return sysctl_int(oldp, oldlenp, newp, newlen, &dovfsusermount);
1984 #if defined(COMPAT_09) || defined(COMPAT_43) || defined(COMPAT_44)
1985 	case VFS_MAXTYPENUM:
1986 		/*
1987 		 * Provided for 4.4BSD-Lite2 compatibility.
1988 		 */
1989 		return (sysctl_rdint(oldp, oldlenp, newp, nmountcompatnames));
1990 	case VFS_CONF:
1991 		/*
1992 		 * Special: a node, next is a file system name.
1993 		 * Provided for 4.4BSD-Lite2 compatibility.
1994 		 */
1995 		if (namelen < 3)
1996 			return (ENOTDIR);	/* overloaded */
1997 		if (name[2] >= nmountcompatnames || name[2] < 0 ||
1998 		    mountcompatnames[name[2]] == NULL)
1999 			return (EOPNOTSUPP);
2000 		vfsp = vfs_getopsbyname(mountcompatnames[name[2]]);
2001 		if (vfsp == NULL)
2002 			return (EOPNOTSUPP);
2003 		vfc.vfc_vfsops = vfsp;
2004 		strncpy(vfc.vfc_name, vfsp->vfs_name, MFSNAMELEN);
2005 		vfc.vfc_typenum = name[2];
2006 		vfc.vfc_refcount = vfsp->vfs_refcount;
2007 		vfc.vfc_flags = 0;
2008 		vfc.vfc_mountroot = vfsp->vfs_mountroot;
2009 		vfc.vfc_next = NULL;
2010 		return (sysctl_rdstruct(oldp, oldlenp, newp, &vfc,
2011 		    sizeof(struct vfsconf)));
2012 #endif
2013 	default:
2014 		break;
2015 	}
2016 	return (EOPNOTSUPP);
2017 }
2018 
2019 int kinfo_vdebug = 1;
2020 int kinfo_vgetfailed;
2021 #define KINFO_VNODESLOP	10
2022 /*
2023  * Dump vnode list (via sysctl).
2024  * Copyout address of vnode followed by vnode.
2025  */
2026 /* ARGSUSED */
2027 int
2028 sysctl_vnode(where, sizep, p)
2029 	char *where;
2030 	size_t *sizep;
2031 	struct proc *p;
2032 {
2033 	struct mount *mp, *nmp;
2034 	struct vnode *nvp, *vp;
2035 	char *bp = where, *savebp;
2036 	char *ewhere;
2037 	int error;
2038 
2039 #define VPTRSZ	sizeof(struct vnode *)
2040 #define VNODESZ	sizeof(struct vnode)
2041 	if (where == NULL) {
2042 		*sizep = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ);
2043 		return (0);
2044 	}
2045 	ewhere = where + *sizep;
2046 
2047 	simple_lock(&mountlist_slock);
2048 	for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist;
2049 	     mp = nmp) {
2050 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) {
2051 			nmp = CIRCLEQ_NEXT(mp, mnt_list);
2052 			continue;
2053 		}
2054 		savebp = bp;
2055 again:
2056 		simple_lock(&mntvnode_slock);
2057 		for (vp = LIST_FIRST(&mp->mnt_vnodelist);
2058 		     vp != NULL;
2059 		     vp = nvp) {
2060 			/*
2061 			 * Check that the vp is still associated with
2062 			 * this filesystem.  RACE: could have been
2063 			 * recycled onto the same filesystem.
2064 			 */
2065 			if (vp->v_mount != mp) {
2066 				simple_unlock(&mntvnode_slock);
2067 				if (kinfo_vdebug)
2068 					printf("kinfo: vp changed\n");
2069 				bp = savebp;
2070 				goto again;
2071 			}
2072 			nvp = LIST_NEXT(vp, v_mntvnodes);
2073 			if (bp + VPTRSZ + VNODESZ > ewhere) {
2074 				simple_unlock(&mntvnode_slock);
2075 				*sizep = bp - where;
2076 				return (ENOMEM);
2077 			}
2078 			simple_unlock(&mntvnode_slock);
2079 			if ((error = copyout((caddr_t)&vp, bp, VPTRSZ)) ||
2080 			   (error = copyout((caddr_t)vp, bp + VPTRSZ, VNODESZ)))
2081 				return (error);
2082 			bp += VPTRSZ + VNODESZ;
2083 			simple_lock(&mntvnode_slock);
2084 		}
2085 		simple_unlock(&mntvnode_slock);
2086 		simple_lock(&mountlist_slock);
2087 		nmp = CIRCLEQ_NEXT(mp, mnt_list);
2088 		vfs_unbusy(mp);
2089 	}
2090 	simple_unlock(&mountlist_slock);
2091 
2092 	*sizep = bp - where;
2093 	return (0);
2094 }
2095 
2096 /*
2097  * Check to see if a filesystem is mounted on a block device.
2098  */
2099 int
2100 vfs_mountedon(vp)
2101 	struct vnode *vp;
2102 {
2103 	struct vnode *vq;
2104 	int error = 0;
2105 
2106 	if (vp->v_specmountpoint != NULL)
2107 		return (EBUSY);
2108 	if (vp->v_flag & VALIASED) {
2109 		simple_lock(&spechash_slock);
2110 		for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
2111 			if (vq->v_rdev != vp->v_rdev ||
2112 			    vq->v_type != vp->v_type)
2113 				continue;
2114 			if (vq->v_specmountpoint != NULL) {
2115 				error = EBUSY;
2116 				break;
2117 			}
2118 		}
2119 		simple_unlock(&spechash_slock);
2120 	}
2121 	return (error);
2122 }
2123 
2124 /*
2125  * Build hash lists of net addresses and hang them off the mount point.
2126  * Called by ufs_mount() to set up the lists of export addresses.
2127  */
2128 static int
2129 vfs_hang_addrlist(mp, nep, argp)
2130 	struct mount *mp;
2131 	struct netexport *nep;
2132 	struct export_args *argp;
2133 {
2134 	struct netcred *np, *enp;
2135 	struct radix_node_head *rnh;
2136 	int i;
2137 	struct radix_node *rn;
2138 	struct sockaddr *saddr, *smask = 0;
2139 	struct domain *dom;
2140 	int error;
2141 
2142 	if (argp->ex_addrlen == 0) {
2143 		if (mp->mnt_flag & MNT_DEFEXPORTED)
2144 			return (EPERM);
2145 		np = &nep->ne_defexported;
2146 		np->netc_exflags = argp->ex_flags;
2147 		crcvt(&np->netc_anon, &argp->ex_anon);
2148 		np->netc_anon.cr_ref = 1;
2149 		mp->mnt_flag |= MNT_DEFEXPORTED;
2150 		return (0);
2151 	}
2152 
2153 	if (argp->ex_addrlen > MLEN)
2154 		return (EINVAL);
2155 
2156 	i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2157 	np = (struct netcred *)malloc(i, M_NETADDR, M_WAITOK);
2158 	memset((caddr_t)np, 0, i);
2159 	saddr = (struct sockaddr *)(np + 1);
2160 	error = copyin(argp->ex_addr, (caddr_t)saddr, argp->ex_addrlen);
2161 	if (error)
2162 		goto out;
2163 	if (saddr->sa_len > argp->ex_addrlen)
2164 		saddr->sa_len = argp->ex_addrlen;
2165 	if (argp->ex_masklen) {
2166 		smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
2167 		error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen);
2168 		if (error)
2169 			goto out;
2170 		if (smask->sa_len > argp->ex_masklen)
2171 			smask->sa_len = argp->ex_masklen;
2172 	}
2173 	i = saddr->sa_family;
2174 	if ((rnh = nep->ne_rtable[i]) == 0) {
2175 		/*
2176 		 * Seems silly to initialize every AF when most are not
2177 		 * used, do so on demand here
2178 		 */
2179 		for (dom = domains; dom; dom = dom->dom_next)
2180 			if (dom->dom_family == i && dom->dom_rtattach) {
2181 				dom->dom_rtattach((void **)&nep->ne_rtable[i],
2182 					dom->dom_rtoffset);
2183 				break;
2184 			}
2185 		if ((rnh = nep->ne_rtable[i]) == 0) {
2186 			error = ENOBUFS;
2187 			goto out;
2188 		}
2189 	}
2190 	rn = (*rnh->rnh_addaddr)((caddr_t)saddr, (caddr_t)smask, rnh,
2191 		np->netc_rnodes);
2192 	if (rn == 0 || np != (struct netcred *)rn) { /* already exists */
2193 		if (rn == 0) {
2194 			enp = (struct netcred *)(*rnh->rnh_lookup)(saddr,
2195 				smask, rnh);
2196 			if (enp == 0) {
2197 				error = EPERM;
2198 				goto out;
2199 			}
2200 		} else
2201 			enp = (struct netcred *)rn;
2202 
2203 		if (enp->netc_exflags != argp->ex_flags ||
2204 		    enp->netc_anon.cr_uid != argp->ex_anon.cr_uid ||
2205 		    enp->netc_anon.cr_gid != argp->ex_anon.cr_gid ||
2206 		    enp->netc_anon.cr_ngroups !=
2207 		    			(uint32_t) argp->ex_anon.cr_ngroups ||
2208 		    memcmp(&enp->netc_anon.cr_groups, &argp->ex_anon.cr_groups,
2209 			enp->netc_anon.cr_ngroups))
2210 				error = EPERM;
2211 		else
2212 			error = 0;
2213 		goto out;
2214 	}
2215 	np->netc_exflags = argp->ex_flags;
2216 	crcvt(&np->netc_anon, &argp->ex_anon);
2217 	np->netc_anon.cr_ref = 1;
2218 	return (0);
2219 out:
2220 	free(np, M_NETADDR);
2221 	return (error);
2222 }
2223 
2224 /* ARGSUSED */
2225 static int
2226 vfs_free_netcred(rn, w)
2227 	struct radix_node *rn;
2228 	void *w;
2229 {
2230 	struct radix_node_head *rnh = (struct radix_node_head *)w;
2231 
2232 	(*rnh->rnh_deladdr)(rn->rn_key, rn->rn_mask, rnh);
2233 	free((caddr_t)rn, M_NETADDR);
2234 	return (0);
2235 }
2236 
2237 /*
2238  * Free the net address hash lists that are hanging off the mount points.
2239  */
2240 static void
2241 vfs_free_addrlist(nep)
2242 	struct netexport *nep;
2243 {
2244 	int i;
2245 	struct radix_node_head *rnh;
2246 
2247 	for (i = 0; i <= AF_MAX; i++)
2248 		if ((rnh = nep->ne_rtable[i]) != NULL) {
2249 			(*rnh->rnh_walktree)(rnh, vfs_free_netcred, rnh);
2250 			free((caddr_t)rnh, M_RTABLE);
2251 			nep->ne_rtable[i] = 0;
2252 		}
2253 }
2254 
2255 int
2256 vfs_export(mp, nep, argp)
2257 	struct mount *mp;
2258 	struct netexport *nep;
2259 	struct export_args *argp;
2260 {
2261 	int error;
2262 
2263 	if (argp->ex_flags & MNT_DELEXPORT) {
2264 		if (mp->mnt_flag & MNT_EXPUBLIC) {
2265 			vfs_setpublicfs(NULL, NULL, NULL);
2266 			mp->mnt_flag &= ~MNT_EXPUBLIC;
2267 		}
2268 		vfs_free_addrlist(nep);
2269 		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2270 	}
2271 	if (argp->ex_flags & MNT_EXPORTED) {
2272 		if (argp->ex_flags & MNT_EXPUBLIC) {
2273 			if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2274 				return (error);
2275 			mp->mnt_flag |= MNT_EXPUBLIC;
2276 		}
2277 		if ((error = vfs_hang_addrlist(mp, nep, argp)) != 0)
2278 			return (error);
2279 		mp->mnt_flag |= MNT_EXPORTED;
2280 	}
2281 	return (0);
2282 }
2283 
2284 /*
2285  * Set the publicly exported filesystem (WebNFS). Currently, only
2286  * one public filesystem is possible in the spec (RFC 2054 and 2055)
2287  */
2288 int
2289 vfs_setpublicfs(mp, nep, argp)
2290 	struct mount *mp;
2291 	struct netexport *nep;
2292 	struct export_args *argp;
2293 {
2294 	int error;
2295 	struct vnode *rvp;
2296 	char *cp;
2297 
2298 	/*
2299 	 * mp == NULL -> invalidate the current info, the FS is
2300 	 * no longer exported. May be called from either vfs_export
2301 	 * or unmount, so check if it hasn't already been done.
2302 	 */
2303 	if (mp == NULL) {
2304 		if (nfs_pub.np_valid) {
2305 			nfs_pub.np_valid = 0;
2306 			if (nfs_pub.np_index != NULL) {
2307 				FREE(nfs_pub.np_index, M_TEMP);
2308 				nfs_pub.np_index = NULL;
2309 			}
2310 		}
2311 		return (0);
2312 	}
2313 
2314 	/*
2315 	 * Only one allowed at a time.
2316 	 */
2317 	if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2318 		return (EBUSY);
2319 
2320 	/*
2321 	 * Get real filehandle for root of exported FS.
2322 	 */
2323 	memset((caddr_t)&nfs_pub.np_handle, 0, sizeof(nfs_pub.np_handle));
2324 	nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2325 
2326 	if ((error = VFS_ROOT(mp, &rvp)))
2327 		return (error);
2328 
2329 	if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2330 		return (error);
2331 
2332 	vput(rvp);
2333 
2334 	/*
2335 	 * If an indexfile was specified, pull it in.
2336 	 */
2337 	if (argp->ex_indexfile != NULL) {
2338 		MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
2339 		    M_WAITOK);
2340 		error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2341 		    MAXNAMLEN, (size_t *)0);
2342 		if (!error) {
2343 			/*
2344 			 * Check for illegal filenames.
2345 			 */
2346 			for (cp = nfs_pub.np_index; *cp; cp++) {
2347 				if (*cp == '/') {
2348 					error = EINVAL;
2349 					break;
2350 				}
2351 			}
2352 		}
2353 		if (error) {
2354 			FREE(nfs_pub.np_index, M_TEMP);
2355 			return (error);
2356 		}
2357 	}
2358 
2359 	nfs_pub.np_mount = mp;
2360 	nfs_pub.np_valid = 1;
2361 	return (0);
2362 }
2363 
2364 struct netcred *
2365 vfs_export_lookup(mp, nep, nam)
2366 	struct mount *mp;
2367 	struct netexport *nep;
2368 	struct mbuf *nam;
2369 {
2370 	struct netcred *np;
2371 	struct radix_node_head *rnh;
2372 	struct sockaddr *saddr;
2373 
2374 	np = NULL;
2375 	if (mp->mnt_flag & MNT_EXPORTED) {
2376 		/*
2377 		 * Lookup in the export list first.
2378 		 */
2379 		if (nam != NULL) {
2380 			saddr = mtod(nam, struct sockaddr *);
2381 			rnh = nep->ne_rtable[saddr->sa_family];
2382 			if (rnh != NULL) {
2383 				np = (struct netcred *)
2384 					(*rnh->rnh_matchaddr)((caddr_t)saddr,
2385 							      rnh);
2386 				if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2387 					np = NULL;
2388 			}
2389 		}
2390 		/*
2391 		 * If no address match, use the default if it exists.
2392 		 */
2393 		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2394 			np = &nep->ne_defexported;
2395 	}
2396 	return (np);
2397 }
2398 
2399 /*
2400  * Do the usual access checking.
2401  * file_mode, uid and gid are from the vnode in question,
2402  * while acc_mode and cred are from the VOP_ACCESS parameter list
2403  */
2404 int
2405 vaccess(type, file_mode, uid, gid, acc_mode, cred)
2406 	enum vtype type;
2407 	mode_t file_mode;
2408 	uid_t uid;
2409 	gid_t gid;
2410 	mode_t acc_mode;
2411 	struct ucred *cred;
2412 {
2413 	mode_t mask;
2414 
2415 	/*
2416 	 * Super-user always gets read/write access, but execute access depends
2417 	 * on at least one execute bit being set.
2418 	 */
2419 	if (cred->cr_uid == 0) {
2420 		if ((acc_mode & VEXEC) && type != VDIR &&
2421 		    (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)
2422 			return (EACCES);
2423 		return (0);
2424 	}
2425 
2426 	mask = 0;
2427 
2428 	/* Otherwise, check the owner. */
2429 	if (cred->cr_uid == uid) {
2430 		if (acc_mode & VEXEC)
2431 			mask |= S_IXUSR;
2432 		if (acc_mode & VREAD)
2433 			mask |= S_IRUSR;
2434 		if (acc_mode & VWRITE)
2435 			mask |= S_IWUSR;
2436 		return ((file_mode & mask) == mask ? 0 : EACCES);
2437 	}
2438 
2439 	/* Otherwise, check the groups. */
2440 	if (cred->cr_gid == gid || groupmember(gid, cred)) {
2441 		if (acc_mode & VEXEC)
2442 			mask |= S_IXGRP;
2443 		if (acc_mode & VREAD)
2444 			mask |= S_IRGRP;
2445 		if (acc_mode & VWRITE)
2446 			mask |= S_IWGRP;
2447 		return ((file_mode & mask) == mask ? 0 : EACCES);
2448 	}
2449 
2450 	/* Otherwise, check everyone else. */
2451 	if (acc_mode & VEXEC)
2452 		mask |= S_IXOTH;
2453 	if (acc_mode & VREAD)
2454 		mask |= S_IROTH;
2455 	if (acc_mode & VWRITE)
2456 		mask |= S_IWOTH;
2457 	return ((file_mode & mask) == mask ? 0 : EACCES);
2458 }
2459 
2460 /*
2461  * Unmount all file systems.
2462  * We traverse the list in reverse order under the assumption that doing so
2463  * will avoid needing to worry about dependencies.
2464  */
2465 void
2466 vfs_unmountall(p)
2467 	struct proc *p;
2468 {
2469 	struct mount *mp, *nmp;
2470 	int allerror, error;
2471 
2472 	for (allerror = 0,
2473 	     mp = mountlist.cqh_last; mp != (void *)&mountlist; mp = nmp) {
2474 		nmp = mp->mnt_list.cqe_prev;
2475 #ifdef DEBUG
2476 		printf("unmounting %s (%s)...\n",
2477 		    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_mntfromname);
2478 #endif
2479 		/*
2480 		 * XXX Freeze syncer.  Must do this before locking the
2481 		 * mount point.  See dounmount() for details.
2482 		 */
2483 		lockmgr(&syncer_lock, LK_EXCLUSIVE, NULL);
2484 		if (vfs_busy(mp, 0, 0)) {
2485 			lockmgr(&syncer_lock, LK_RELEASE, NULL);
2486 			continue;
2487 		}
2488 		if ((error = dounmount(mp, MNT_FORCE, p)) != 0) {
2489 			printf("unmount of %s failed with error %d\n",
2490 			    mp->mnt_stat.f_mntonname, error);
2491 			allerror = 1;
2492 		}
2493 	}
2494 	if (allerror)
2495 		printf("WARNING: some file systems would not unmount\n");
2496 }
2497 
2498 /*
2499  * Sync and unmount file systems before shutting down.
2500  */
2501 void
2502 vfs_shutdown()
2503 {
2504 	struct buf *bp;
2505 	int iter, nbusy, nbusy_prev = 0, dcount, s;
2506 	struct lwp *l = curlwp;
2507 	struct proc *p;
2508 
2509 	/* XXX we're certainly not running in proc0's context! */
2510 	if (l == NULL || (p = l->l_proc) == NULL)
2511 		p = &proc0;
2512 
2513 	printf("syncing disks... ");
2514 
2515 	/* remove user process from run queue */
2516 	suspendsched();
2517 	(void) spl0();
2518 
2519 	/* avoid coming back this way again if we panic. */
2520 	doing_shutdown = 1;
2521 
2522 	sys_sync(l, NULL, NULL);
2523 
2524 	/* Wait for sync to finish. */
2525 	dcount = 10000;
2526 	for (iter = 0; iter < 20;) {
2527 		nbusy = 0;
2528 		for (bp = &buf[nbuf]; --bp >= buf; ) {
2529 			if ((bp->b_flags & (B_BUSY|B_INVAL|B_READ)) == B_BUSY)
2530 				nbusy++;
2531 			/*
2532 			 * With soft updates, some buffers that are
2533 			 * written will be remarked as dirty until other
2534 			 * buffers are written.
2535 			 */
2536 			if (bp->b_vp && bp->b_vp->v_mount
2537 			    && (bp->b_vp->v_mount->mnt_flag & MNT_SOFTDEP)
2538 			    && (bp->b_flags & B_DELWRI)) {
2539 				s = splbio();
2540 				bremfree(bp);
2541 				bp->b_flags |= B_BUSY;
2542 				splx(s);
2543 				nbusy++;
2544 				bawrite(bp);
2545 				if (dcount-- <= 0) {
2546 					printf("softdep ");
2547 					goto fail;
2548 				}
2549 			}
2550 		}
2551 		if (nbusy == 0)
2552 			break;
2553 		if (nbusy_prev == 0)
2554 			nbusy_prev = nbusy;
2555 		printf("%d ", nbusy);
2556 		tsleep(&nbusy, PRIBIO, "bflush",
2557 		    (iter == 0) ? 1 : hz / 25 * iter);
2558 		if (nbusy >= nbusy_prev) /* we didn't flush anything */
2559 			iter++;
2560 		else
2561 			nbusy_prev = nbusy;
2562 	}
2563 	if (nbusy) {
2564 fail:
2565 #if defined(DEBUG) || defined(DEBUG_HALT_BUSY)
2566 		printf("giving up\nPrinting vnodes for busy buffers\n");
2567 		for (bp = &buf[nbuf]; --bp >= buf; )
2568 			if ((bp->b_flags & (B_BUSY|B_INVAL|B_READ)) == B_BUSY)
2569 				vprint(NULL, bp->b_vp);
2570 
2571 #if defined(DDB) && defined(DEBUG_HALT_BUSY)
2572 		Debugger();
2573 #endif
2574 
2575 #else  /* defined(DEBUG) || defined(DEBUG_HALT_BUSY) */
2576 		printf("giving up\n");
2577 #endif /* defined(DEBUG) || defined(DEBUG_HALT_BUSY) */
2578 		return;
2579 	} else
2580 		printf("done\n");
2581 
2582 	/*
2583 	 * If we've panic'd, don't make the situation potentially
2584 	 * worse by unmounting the file systems.
2585 	 */
2586 	if (panicstr != NULL)
2587 		return;
2588 
2589 	/* Release inodes held by texts before update. */
2590 #ifdef notdef
2591 	vnshutdown();
2592 #endif
2593 	/* Unmount file systems. */
2594 	vfs_unmountall(p);
2595 }
2596 
2597 /*
2598  * Mount the root file system.  If the operator didn't specify a
2599  * file system to use, try all possible file systems until one
2600  * succeeds.
2601  */
2602 int
2603 vfs_mountroot()
2604 {
2605 	struct vfsops *v;
2606 
2607 	if (root_device == NULL)
2608 		panic("vfs_mountroot: root device unknown");
2609 
2610 	switch (root_device->dv_class) {
2611 	case DV_IFNET:
2612 		if (rootdev != NODEV)
2613 			panic("vfs_mountroot: rootdev set for DV_IFNET "
2614 			    "(0x%08x -> %d,%d)", rootdev,
2615 			    major(rootdev), minor(rootdev));
2616 		break;
2617 
2618 	case DV_DISK:
2619 		if (rootdev == NODEV)
2620 			panic("vfs_mountroot: rootdev not set for DV_DISK");
2621 		break;
2622 
2623 	default:
2624 		printf("%s: inappropriate for root file system\n",
2625 		    root_device->dv_xname);
2626 		return (ENODEV);
2627 	}
2628 
2629 	/*
2630 	 * If user specified a file system, use it.
2631 	 */
2632 	if (mountroot != NULL)
2633 		return ((*mountroot)());
2634 
2635 	/*
2636 	 * Try each file system currently configured into the kernel.
2637 	 */
2638 	for (v = LIST_FIRST(&vfs_list); v != NULL; v = LIST_NEXT(v, vfs_list)) {
2639 		if (v->vfs_mountroot == NULL)
2640 			continue;
2641 #ifdef DEBUG
2642 		printf("mountroot: trying %s...\n", v->vfs_name);
2643 #endif
2644 		if ((*v->vfs_mountroot)() == 0) {
2645 			printf("root file system type: %s\n", v->vfs_name);
2646 			break;
2647 		}
2648 	}
2649 
2650 	if (v == NULL) {
2651 		printf("no file system for %s", root_device->dv_xname);
2652 		if (root_device->dv_class == DV_DISK)
2653 			printf(" (dev 0x%x)", rootdev);
2654 		printf("\n");
2655 		return (EFTYPE);
2656 	}
2657 	return (0);
2658 }
2659 
2660 /*
2661  * Given a file system name, look up the vfsops for that
2662  * file system, or return NULL if file system isn't present
2663  * in the kernel.
2664  */
2665 struct vfsops *
2666 vfs_getopsbyname(name)
2667 	const char *name;
2668 {
2669 	struct vfsops *v;
2670 
2671 	for (v = LIST_FIRST(&vfs_list); v != NULL; v = LIST_NEXT(v, vfs_list)) {
2672 		if (strcmp(v->vfs_name, name) == 0)
2673 			break;
2674 	}
2675 
2676 	return (v);
2677 }
2678 
2679 /*
2680  * Establish a file system and initialize it.
2681  */
2682 int
2683 vfs_attach(vfs)
2684 	struct vfsops *vfs;
2685 {
2686 	struct vfsops *v;
2687 	int error = 0;
2688 
2689 
2690 	/*
2691 	 * Make sure this file system doesn't already exist.
2692 	 */
2693 	LIST_FOREACH(v, &vfs_list, vfs_list) {
2694 		if (strcmp(vfs->vfs_name, v->vfs_name) == 0) {
2695 			error = EEXIST;
2696 			goto out;
2697 		}
2698 	}
2699 
2700 	/*
2701 	 * Initialize the vnode operations for this file system.
2702 	 */
2703 	vfs_opv_init(vfs->vfs_opv_descs);
2704 
2705 	/*
2706 	 * Now initialize the file system itself.
2707 	 */
2708 	(*vfs->vfs_init)();
2709 
2710 	/*
2711 	 * ...and link it into the kernel's list.
2712 	 */
2713 	LIST_INSERT_HEAD(&vfs_list, vfs, vfs_list);
2714 
2715 	/*
2716 	 * Sanity: make sure the reference count is 0.
2717 	 */
2718 	vfs->vfs_refcount = 0;
2719 
2720  out:
2721 	return (error);
2722 }
2723 
2724 /*
2725  * Remove a file system from the kernel.
2726  */
2727 int
2728 vfs_detach(vfs)
2729 	struct vfsops *vfs;
2730 {
2731 	struct vfsops *v;
2732 
2733 	/*
2734 	 * Make sure no one is using the filesystem.
2735 	 */
2736 	if (vfs->vfs_refcount != 0)
2737 		return (EBUSY);
2738 
2739 	/*
2740 	 * ...and remove it from the kernel's list.
2741 	 */
2742 	LIST_FOREACH(v, &vfs_list, vfs_list) {
2743 		if (v == vfs) {
2744 			LIST_REMOVE(v, vfs_list);
2745 			break;
2746 		}
2747 	}
2748 
2749 	if (v == NULL)
2750 		return (ESRCH);
2751 
2752 	/*
2753 	 * Now run the file system-specific cleanups.
2754 	 */
2755 	(*vfs->vfs_done)();
2756 
2757 	/*
2758 	 * Free the vnode operations vector.
2759 	 */
2760 	vfs_opv_free(vfs->vfs_opv_descs);
2761 	return (0);
2762 }
2763 
2764 void
2765 vfs_reinit(void)
2766 {
2767 	struct vfsops *vfs;
2768 
2769 	LIST_FOREACH(vfs, &vfs_list, vfs_list) {
2770 		if (vfs->vfs_reinit) {
2771 			(*vfs->vfs_reinit)();
2772 		}
2773 	}
2774 }
2775 
2776 void
2777 copy_statfs_info(struct statfs *sbp, const struct mount *mp)
2778 {
2779 	const struct statfs *mbp;
2780 
2781 	if (sbp == (mbp = &mp->mnt_stat))
2782 		return;
2783 
2784 	sbp->f_oflags = mbp->f_oflags;
2785 	sbp->f_type = mbp->f_type;
2786 	(void)memcpy(&sbp->f_fsid, &mbp->f_fsid, sizeof(sbp->f_fsid));
2787 	sbp->f_owner = mbp->f_owner;
2788 	sbp->f_flags = mbp->f_flags;
2789 	sbp->f_syncwrites = mbp->f_syncwrites;
2790 	sbp->f_asyncwrites = mbp->f_asyncwrites;
2791 	sbp->f_spare[0] = mbp->f_spare[0];
2792 	(void)memcpy(sbp->f_fstypename, mbp->f_fstypename,
2793 	    sizeof(sbp->f_fstypename));
2794 	(void)memcpy(sbp->f_mntonname, mbp->f_mntonname,
2795 	    sizeof(sbp->f_mntonname));
2796 	(void)memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname,
2797 	    sizeof(sbp->f_mntfromname));
2798 }
2799 
2800 int
2801 set_statfs_info(const char *onp, int ukon, const char *fromp, int ukfrom,
2802     struct mount *mp, struct proc *p)
2803 {
2804 	int error;
2805 	size_t size;
2806 	struct statfs *sfs = &mp->mnt_stat;
2807 	int (*fun)(const void *, void *, size_t, size_t *);
2808 
2809 	(void)strncpy(mp->mnt_stat.f_fstypename, mp->mnt_op->vfs_name,
2810 	    sizeof(mp->mnt_stat.f_fstypename));
2811 
2812 	if (onp) {
2813 		struct cwdinfo *cwdi = p->p_cwdi;
2814 		fun = (ukon == UIO_SYSSPACE) ? copystr : copyinstr;
2815 		if (cwdi->cwdi_rdir != NULL) {
2816 			size_t len;
2817 			char *bp;
2818 			char *path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
2819 
2820 			if (!path)
2821 				return ENOMEM;
2822 
2823 			bp = path + MAXPATHLEN;
2824 			*--bp = '\0';
2825 			error = getcwd_common(cwdi->cwdi_rdir, rootvnode, &bp,
2826 			    path, MAXPATHLEN / 2, 0, p);
2827 			if (error) {
2828 				free(path, M_TEMP);
2829 				return error;
2830 			}
2831 
2832 			len = strlen(bp);
2833 			if (len > sizeof(sfs->f_mntonname) - 1)
2834 				len = sizeof(sfs->f_mntonname) - 1;
2835 			(void)strncpy(sfs->f_mntonname, bp, len);
2836 			free(path, M_TEMP);
2837 
2838 			if (len < sizeof(sfs->f_mntonname) - 1) {
2839 				error = (*fun)(onp, &sfs->f_mntonname[len],
2840 				    sizeof(sfs->f_mntonname) - len - 1, &size);
2841 				if (error)
2842 					return error;
2843 				size += len;
2844 			} else {
2845 				size = len;
2846 			}
2847 		} else {
2848 			error = (*fun)(onp, &sfs->f_mntonname,
2849 			    sizeof(sfs->f_mntonname) - 1, &size);
2850 			if (error)
2851 				return error;
2852 		}
2853 		(void)memset(sfs->f_mntonname + size, 0,
2854 		    sizeof(sfs->f_mntonname) - size);
2855 	}
2856 
2857 	if (fromp) {
2858 		fun = (ukfrom == UIO_SYSSPACE) ? copystr : copyinstr;
2859 		error = (*fun)(fromp, sfs->f_mntfromname,
2860 		    sizeof(sfs->f_mntfromname) - 1, &size);
2861 		if (error)
2862 			return error;
2863 		(void)memset(sfs->f_mntfromname + size, 0,
2864 		    sizeof(sfs->f_mntfromname) - size);
2865 	}
2866 	return 0;
2867 }
2868 
2869 #ifdef DDB
2870 const char buf_flagbits[] =
2871 	"\20\1AGE\2NEEDCOMMIT\3ASYNC\4BAD\5BUSY\6SCANNED\7CALL\10DELWRI"
2872 	"\11DIRTY\12DONE\13EINTR\14ERROR\15GATHERED\16INVAL\17LOCKED\20NOCACHE"
2873 	"\21ORDERED\22CACHE\23PHYS\24RAW\25READ\26TAPE\30WANTED"
2874 	"\32XXX\33VFLUSH";
2875 
2876 void
2877 vfs_buf_print(bp, full, pr)
2878 	struct buf *bp;
2879 	int full;
2880 	void (*pr) __P((const char *, ...));
2881 {
2882 	char buf[1024];
2883 
2884 	(*pr)("  vp %p lblkno 0x%x blkno 0x%x dev 0x%x\n",
2885 		  bp->b_vp, bp->b_lblkno, bp->b_blkno, bp->b_dev);
2886 
2887 	bitmask_snprintf(bp->b_flags, buf_flagbits, buf, sizeof(buf));
2888 	(*pr)("  error %d flags 0x%s\n", bp->b_error, buf);
2889 
2890 	(*pr)("  bufsize 0x%lx bcount 0x%lx resid 0x%lx\n",
2891 		  bp->b_bufsize, bp->b_bcount, bp->b_resid);
2892 	(*pr)("  data %p saveaddr %p dep %p\n",
2893 		  bp->b_data, bp->b_saveaddr, LIST_FIRST(&bp->b_dep));
2894 	(*pr)("  iodone %p\n", bp->b_iodone);
2895 }
2896 
2897 
2898 const char vnode_flagbits[] =
2899 	"\20\1ROOT\2TEXT\3SYSTEM\4ISTTY\5EXECMAP"
2900 	"\11XLOCK\12XWANT\13BWAIT\14ALIASED"
2901 	"\15DIROP\16LAYER\17ONWORKLIST\20DIRTY";
2902 
2903 const char * const vnode_tags[] = {
2904 	"VT_NON",
2905 	"VT_UFS",
2906 	"VT_NFS",
2907 	"VT_MFS",
2908 	"VT_MSDOSFS",
2909 	"VT_LFS",
2910 	"VT_LOFS",
2911 	"VT_FDESC",
2912 	"VT_PORTAL",
2913 	"VT_NULL",
2914 	"VT_UMAP",
2915 	"VT_KERNFS",
2916 	"VT_PROCFS",
2917 	"VT_AFS",
2918 	"VT_ISOFS",
2919 	"VT_UNION",
2920 	"VT_ADOSFS",
2921 	"VT_EXT2FS",
2922 	"VT_CODA",
2923 	"VT_FILECORE",
2924 	"VT_NTFS",
2925 	"VT_VFS",
2926 	"VT_OVERLAY",
2927 	"VT_SMBFS"
2928 };
2929 
2930 void
2931 vfs_vnode_print(vp, full, pr)
2932 	struct vnode *vp;
2933 	int full;
2934 	void (*pr) __P((const char *, ...));
2935 {
2936 	char buf[256];
2937 	const char *vtype, *vtag;
2938 
2939 	uvm_object_printit(&vp->v_uobj, full, pr);
2940 	bitmask_snprintf(vp->v_flag, vnode_flagbits, buf, sizeof(buf));
2941 	(*pr)("\nVNODE flags %s\n", buf);
2942 	(*pr)("mp %p numoutput %d size 0x%llx\n",
2943 	      vp->v_mount, vp->v_numoutput, vp->v_size);
2944 
2945 	(*pr)("data %p usecount %d writecount %ld holdcnt %ld numoutput %d\n",
2946 	      vp->v_data, vp->v_usecount, vp->v_writecount,
2947 	      vp->v_holdcnt, vp->v_numoutput);
2948 
2949 	vtype = (vp->v_type >= 0 &&
2950 		 vp->v_type < sizeof(vnode_types) / sizeof(vnode_types[0])) ?
2951 		vnode_types[vp->v_type] : "UNKNOWN";
2952 	vtag = (vp->v_tag >= 0 &&
2953 		vp->v_tag < sizeof(vnode_tags) / sizeof(vnode_tags[0])) ?
2954 		vnode_tags[vp->v_tag] : "UNKNOWN";
2955 
2956 	(*pr)("type %s(%d) tag %s(%d) id 0x%lx mount %p typedata %p\n",
2957 	      vtype, vp->v_type, vtag, vp->v_tag,
2958 	      vp->v_id, vp->v_mount, vp->v_mountedhere);
2959 
2960 	if (full) {
2961 		struct buf *bp;
2962 
2963 		(*pr)("clean bufs:\n");
2964 		LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) {
2965 			(*pr)(" bp %p\n", bp);
2966 			vfs_buf_print(bp, full, pr);
2967 		}
2968 
2969 		(*pr)("dirty bufs:\n");
2970 		LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) {
2971 			(*pr)(" bp %p\n", bp);
2972 			vfs_buf_print(bp, full, pr);
2973 		}
2974 	}
2975 }
2976 #endif
2977