xref: /freebsd/sys/kern/vfs_subr.c (revision 1d66272a85cde1c8a69c58f4b5dd649babd6eca6)
1 /*
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)vfs_subr.c	8.31 (Berkeley) 5/26/95
39  * $FreeBSD$
40  */
41 
42 /*
43  * External virtual filesystem routines
44  */
45 #include "opt_ddb.h"
46 #include "opt_ffs.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/bio.h>
51 #include <sys/buf.h>
52 #include <sys/conf.h>
53 #include <sys/dirent.h>
54 #include <sys/domain.h>
55 #include <sys/eventhandler.h>
56 #include <sys/fcntl.h>
57 #include <sys/kernel.h>
58 #include <sys/kthread.h>
59 #include <sys/ktr.h>
60 #include <sys/malloc.h>
61 #include <sys/mount.h>
62 #include <sys/mutex.h>
63 #include <sys/namei.h>
64 #include <sys/proc.h>
65 #include <sys/reboot.h>
66 #include <sys/socket.h>
67 #include <sys/stat.h>
68 #include <sys/sysctl.h>
69 #include <sys/vmmeter.h>
70 #include <sys/vnode.h>
71 
72 #include <machine/limits.h>
73 
74 #include <vm/vm.h>
75 #include <vm/vm_object.h>
76 #include <vm/vm_extern.h>
77 #include <vm/pmap.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_page.h>
80 #include <vm/vm_pager.h>
81 #include <vm/vnode_pager.h>
82 #include <vm/vm_zone.h>
83 
84 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
85 
86 static void	addalias __P((struct vnode *vp, dev_t nvp_rdev));
87 static void	insmntque __P((struct vnode *vp, struct mount *mp));
88 static void	vclean __P((struct vnode *vp, int flags, struct proc *p));
89 
90 /*
91  * Number of vnodes in existence.  Increased whenever getnewvnode()
92  * allocates a new vnode, never decreased.
93  */
94 static unsigned long	numvnodes;
95 SYSCTL_LONG(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
96 
97 /*
98  * Conversion tables for conversion from vnode types to inode formats
99  * and back.
100  */
101 enum vtype iftovt_tab[16] = {
102 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
103 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
104 };
105 int vttoif_tab[9] = {
106 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
107 	S_IFSOCK, S_IFIFO, S_IFMT,
108 };
109 
110 /*
111  * List of vnodes that are ready for recycling.
112  */
113 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
114 
115 /*
116  * Minimum number of free vnodes.  If there are fewer than this free vnodes,
117  * getnewvnode() will return a newly allocated vnode.
118  */
119 static u_long wantfreevnodes = 25;
120 SYSCTL_LONG(_debug, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
121 /* Number of vnodes in the free list. */
122 static u_long freevnodes = 0;
123 SYSCTL_LONG(_debug, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
124 
125 /*
126  * Various variables used for debugging the new implementation of
127  * reassignbuf().
128  * XXX these are probably of (very) limited utility now.
129  */
130 static int reassignbufcalls;
131 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
132 static int reassignbufloops;
133 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW, &reassignbufloops, 0, "");
134 static int reassignbufsortgood;
135 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW, &reassignbufsortgood, 0, "");
136 static int reassignbufsortbad;
137 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW, &reassignbufsortbad, 0, "");
138 /* Set to 0 for old insertion-sort based reassignbuf, 1 for modern method. */
139 static int reassignbufmethod = 1;
140 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW, &reassignbufmethod, 0, "");
141 
142 #ifdef ENABLE_VFS_IOOPT
143 /* See NOTES for a description of this setting. */
144 int vfs_ioopt = 0;
145 SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, "");
146 #endif
147 
148 /* List of mounted filesystems. */
149 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
150 
151 /* For any iteration/modification of mountlist */
152 struct mtx mountlist_mtx;
153 
154 /* For any iteration/modification of mnt_vnodelist */
155 struct simplelock mntvnode_slock;
156 /*
157  * Cache for the mount type id assigned to NFS.  This is used for
158  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
159  */
160 int	nfs_mount_type = -1;
161 
162 #ifndef NULL_SIMPLELOCKS
163 /* To keep more than one thread at a time from running vfs_getnewfsid */
164 static struct simplelock mntid_slock;
165 
166 /* For any iteration/modification of vnode_free_list */
167 static struct simplelock vnode_free_list_slock;
168 
169 /*
170  * For any iteration/modification of dev->si_hlist (linked through
171  * v_specnext)
172  */
173 static struct simplelock spechash_slock;
174 #endif
175 
176 /* Publicly exported FS */
177 struct nfs_public nfs_pub;
178 
179 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
180 static vm_zone_t vnode_zone;
181 
182 /* Set to 1 to print out reclaim of active vnodes */
183 int	prtactive = 0;
184 
185 /*
186  * The workitem queue.
187  *
188  * It is useful to delay writes of file data and filesystem metadata
189  * for tens of seconds so that quickly created and deleted files need
190  * not waste disk bandwidth being created and removed. To realize this,
191  * we append vnodes to a "workitem" queue. When running with a soft
192  * updates implementation, most pending metadata dependencies should
193  * not wait for more than a few seconds. Thus, mounted on block devices
194  * are delayed only about a half the time that file data is delayed.
195  * Similarly, directory updates are more critical, so are only delayed
196  * about a third the time that file data is delayed. Thus, there are
197  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
198  * one each second (driven off the filesystem syncer process). The
199  * syncer_delayno variable indicates the next queue that is to be processed.
200  * Items that need to be processed soon are placed in this queue:
201  *
202  *	syncer_workitem_pending[syncer_delayno]
203  *
204  * A delay of fifteen seconds is done by placing the request fifteen
205  * entries later in the queue:
206  *
207  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
208  *
209  */
210 static int syncer_delayno = 0;
211 static long syncer_mask;
212 LIST_HEAD(synclist, vnode);
213 static struct synclist *syncer_workitem_pending;
214 
215 #define SYNCER_MAXDELAY		32
216 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
217 time_t syncdelay = 30;		/* max time to delay syncing data */
218 time_t filedelay = 30;		/* time to delay syncing files */
219 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
220 time_t dirdelay = 29;		/* time to delay syncing directories */
221 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
222 time_t metadelay = 28;		/* time to delay syncing metadata */
223 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
224 static int rushjob;		/* number of slots to run ASAP */
225 static int stat_rush_requests;	/* number of times I/O speeded up */
226 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
227 
228 /*
229  * Number of vnodes we want to exist at any one time.  This is mostly used
230  * to size hash tables in vnode-related code.  It is normally not used in
231  * getnewvnode(), as wantfreevnodes is normally nonzero.)
232  *
233  * XXX desiredvnodes is historical cruft and should not exist.
234  */
235 int desiredvnodes;
236 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
237     &desiredvnodes, 0, "Maximum number of vnodes");
238 
239 static void	vfs_free_addrlist __P((struct netexport *nep));
240 static int	vfs_free_netcred __P((struct radix_node *rn, void *w));
241 static int	vfs_hang_addrlist __P((struct mount *mp, struct netexport *nep,
242 				       struct export_args *argp));
243 
244 /*
245  * Initialize the vnode management data structures.
246  */
247 static void
248 vntblinit(void *dummy __unused)
249 {
250 
251 	desiredvnodes = maxproc + cnt.v_page_count / 4;
252 	mtx_init(&mountlist_mtx, "mountlist", MTX_DEF);
253 	simple_lock_init(&mntvnode_slock);
254 	simple_lock_init(&mntid_slock);
255 	simple_lock_init(&spechash_slock);
256 	TAILQ_INIT(&vnode_free_list);
257 	simple_lock_init(&vnode_free_list_slock);
258 	vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5);
259 	/*
260 	 * Initialize the filesystem syncer.
261 	 */
262 	syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
263 		&syncer_mask);
264 	syncer_maxdelay = syncer_mask + 1;
265 }
266 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL)
267 
268 
269 /*
270  * Mark a mount point as busy. Used to synchronize access and to delay
271  * unmounting. Interlock is not released on failure.
272  */
273 int
274 vfs_busy(mp, flags, interlkp, p)
275 	struct mount *mp;
276 	int flags;
277 	struct mtx *interlkp;
278 	struct proc *p;
279 {
280 	int lkflags;
281 
282 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
283 		if (flags & LK_NOWAIT)
284 			return (ENOENT);
285 		mp->mnt_kern_flag |= MNTK_MWAIT;
286 		/*
287 		 * Since all busy locks are shared except the exclusive
288 		 * lock granted when unmounting, the only place that a
289 		 * wakeup needs to be done is at the release of the
290 		 * exclusive lock at the end of dounmount.
291 		 */
292 		msleep((caddr_t)mp, interlkp, PVFS, "vfs_busy", 0);
293 		return (ENOENT);
294 	}
295 	lkflags = LK_SHARED | LK_NOPAUSE;
296 	if (interlkp)
297 		lkflags |= LK_INTERLOCK;
298 	if (lockmgr(&mp->mnt_lock, lkflags, interlkp, p))
299 		panic("vfs_busy: unexpected lock failure");
300 	return (0);
301 }
302 
303 /*
304  * Free a busy filesystem.
305  */
306 void
307 vfs_unbusy(mp, p)
308 	struct mount *mp;
309 	struct proc *p;
310 {
311 
312 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, p);
313 }
314 
315 /*
316  * Lookup a filesystem type, and if found allocate and initialize
317  * a mount structure for it.
318  *
319  * Devname is usually updated by mount(8) after booting.
320  */
321 int
322 vfs_rootmountalloc(fstypename, devname, mpp)
323 	char *fstypename;
324 	char *devname;
325 	struct mount **mpp;
326 {
327 	struct proc *p = curproc;	/* XXX */
328 	struct vfsconf *vfsp;
329 	struct mount *mp;
330 
331 	if (fstypename == NULL)
332 		return (ENODEV);
333 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
334 		if (!strcmp(vfsp->vfc_name, fstypename))
335 			break;
336 	if (vfsp == NULL)
337 		return (ENODEV);
338 	mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO);
339 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE);
340 	(void)vfs_busy(mp, LK_NOWAIT, 0, p);
341 	LIST_INIT(&mp->mnt_vnodelist);
342 	mp->mnt_vfc = vfsp;
343 	mp->mnt_op = vfsp->vfc_vfsops;
344 	mp->mnt_flag = MNT_RDONLY;
345 	mp->mnt_vnodecovered = NULLVP;
346 	vfsp->vfc_refcount++;
347 	mp->mnt_iosize_max = DFLTPHYS;
348 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
349 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
350 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
351 	mp->mnt_stat.f_mntonname[0] = '/';
352 	mp->mnt_stat.f_mntonname[1] = 0;
353 	(void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
354 	*mpp = mp;
355 	return (0);
356 }
357 
358 /*
359  * Find an appropriate filesystem to use for the root. If a filesystem
360  * has not been preselected, walk through the list of known filesystems
361  * trying those that have mountroot routines, and try them until one
362  * works or we have tried them all.
363  */
364 #ifdef notdef	/* XXX JH */
365 int
366 lite2_vfs_mountroot()
367 {
368 	struct vfsconf *vfsp;
369 	extern int (*lite2_mountroot) __P((void));
370 	int error;
371 
372 	if (lite2_mountroot != NULL)
373 		return ((*lite2_mountroot)());
374 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
375 		if (vfsp->vfc_mountroot == NULL)
376 			continue;
377 		if ((error = (*vfsp->vfc_mountroot)()) == 0)
378 			return (0);
379 		printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error);
380 	}
381 	return (ENODEV);
382 }
383 #endif
384 
385 /*
386  * Lookup a mount point by filesystem identifier.
387  */
388 struct mount *
389 vfs_getvfs(fsid)
390 	fsid_t *fsid;
391 {
392 	register struct mount *mp;
393 
394 	mtx_enter(&mountlist_mtx, MTX_DEF);
395 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
396 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
397 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
398 			mtx_exit(&mountlist_mtx, MTX_DEF);
399 			return (mp);
400 	    }
401 	}
402 	mtx_exit(&mountlist_mtx, MTX_DEF);
403 	return ((struct mount *) 0);
404 }
405 
406 /*
407  * Get a new unique fsid.  Try to make its val[0] unique, since this value
408  * will be used to create fake device numbers for stat().  Also try (but
409  * not so hard) make its val[0] unique mod 2^16, since some emulators only
410  * support 16-bit device numbers.  We end up with unique val[0]'s for the
411  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
412  *
413  * Keep in mind that several mounts may be running in parallel.  Starting
414  * the search one past where the previous search terminated is both a
415  * micro-optimization and a defense against returning the same fsid to
416  * different mounts.
417  */
418 void
419 vfs_getnewfsid(mp)
420 	struct mount *mp;
421 {
422 	static u_int16_t mntid_base;
423 	fsid_t tfsid;
424 	int mtype;
425 
426 	simple_lock(&mntid_slock);
427 	mtype = mp->mnt_vfc->vfc_typenum;
428 	tfsid.val[1] = mtype;
429 	mtype = (mtype & 0xFF) << 24;
430 	for (;;) {
431 		tfsid.val[0] = makeudev(255,
432 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
433 		mntid_base++;
434 		if (vfs_getvfs(&tfsid) == NULL)
435 			break;
436 	}
437 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
438 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
439 	simple_unlock(&mntid_slock);
440 }
441 
442 /*
443  * Knob to control the precision of file timestamps:
444  *
445  *   0 = seconds only; nanoseconds zeroed.
446  *   1 = seconds and nanoseconds, accurate within 1/HZ.
447  *   2 = seconds and nanoseconds, truncated to microseconds.
448  * >=3 = seconds and nanoseconds, maximum precision.
449  */
450 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
451 
452 static int timestamp_precision = TSP_SEC;
453 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
454     &timestamp_precision, 0, "");
455 
456 /*
457  * Get a current timestamp.
458  */
459 void
460 vfs_timestamp(tsp)
461 	struct timespec *tsp;
462 {
463 	struct timeval tv;
464 
465 	switch (timestamp_precision) {
466 	case TSP_SEC:
467 		tsp->tv_sec = time_second;
468 		tsp->tv_nsec = 0;
469 		break;
470 	case TSP_HZ:
471 		getnanotime(tsp);
472 		break;
473 	case TSP_USEC:
474 		microtime(&tv);
475 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
476 		break;
477 	case TSP_NSEC:
478 	default:
479 		nanotime(tsp);
480 		break;
481 	}
482 }
483 
484 /*
485  * Set vnode attributes to VNOVAL
486  */
487 void
488 vattr_null(vap)
489 	register struct vattr *vap;
490 {
491 
492 	vap->va_type = VNON;
493 	vap->va_size = VNOVAL;
494 	vap->va_bytes = VNOVAL;
495 	vap->va_mode = VNOVAL;
496 	vap->va_nlink = VNOVAL;
497 	vap->va_uid = VNOVAL;
498 	vap->va_gid = VNOVAL;
499 	vap->va_fsid = VNOVAL;
500 	vap->va_fileid = VNOVAL;
501 	vap->va_blocksize = VNOVAL;
502 	vap->va_rdev = VNOVAL;
503 	vap->va_atime.tv_sec = VNOVAL;
504 	vap->va_atime.tv_nsec = VNOVAL;
505 	vap->va_mtime.tv_sec = VNOVAL;
506 	vap->va_mtime.tv_nsec = VNOVAL;
507 	vap->va_ctime.tv_sec = VNOVAL;
508 	vap->va_ctime.tv_nsec = VNOVAL;
509 	vap->va_flags = VNOVAL;
510 	vap->va_gen = VNOVAL;
511 	vap->va_vaflags = 0;
512 }
513 
514 /*
515  * Routines having to do with the management of the vnode table.
516  */
517 
518 /*
519  * Return the next vnode from the free list.
520  */
521 int
522 getnewvnode(tag, mp, vops, vpp)
523 	enum vtagtype tag;
524 	struct mount *mp;
525 	vop_t **vops;
526 	struct vnode **vpp;
527 {
528 	int s, count;
529 	struct proc *p = curproc;	/* XXX */
530 	struct vnode *vp = NULL;
531 	struct mount *vnmp;
532 	vm_object_t object;
533 
534 	/*
535 	 * We take the least recently used vnode from the freelist
536 	 * if we can get it and it has no cached pages, and no
537 	 * namecache entries are relative to it.
538 	 * Otherwise we allocate a new vnode
539 	 */
540 
541 	s = splbio();
542 	simple_lock(&vnode_free_list_slock);
543 
544 	if (wantfreevnodes && freevnodes < wantfreevnodes) {
545 		vp = NULL;
546 	} else if (!wantfreevnodes && freevnodes <= desiredvnodes) {
547 		/*
548 		 * XXX: this is only here to be backwards compatible
549 		 */
550 		vp = NULL;
551 	} else for (count = 0; count < freevnodes; count++) {
552 		vp = TAILQ_FIRST(&vnode_free_list);
553 		if (vp == NULL || vp->v_usecount)
554 			panic("getnewvnode: free vnode isn't");
555 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
556 		/*
557 		 * Don't recycle if active in the namecache or
558 		 * if it still has cached pages or we cannot get
559 		 * its interlock.
560 		 */
561 		if (LIST_FIRST(&vp->v_cache_src) != NULL ||
562 		    (VOP_GETVOBJECT(vp, &object) == 0 &&
563 		     (object->resident_page_count || object->ref_count)) ||
564 		    !mtx_try_enter(&vp->v_interlock, MTX_DEF)) {
565 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
566 			vp = NULL;
567 			continue;
568 		}
569 		/*
570 		 * Skip over it if its filesystem is being suspended.
571 		 */
572 		if (vn_start_write(vp, &vnmp, V_NOWAIT) == 0)
573 			break;
574 		mtx_exit(&vp->v_interlock, MTX_DEF);
575 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
576 		vp = NULL;
577 	}
578 	if (vp) {
579 		vp->v_flag |= VDOOMED;
580 		vp->v_flag &= ~VFREE;
581 		freevnodes--;
582 		simple_unlock(&vnode_free_list_slock);
583 		cache_purge(vp);
584 		vp->v_lease = NULL;
585 		if (vp->v_type != VBAD) {
586 			vgonel(vp, p);
587 		} else {
588 			mtx_exit(&vp->v_interlock, MTX_DEF);
589 		}
590 		vn_finished_write(vnmp);
591 
592 #ifdef INVARIANTS
593 		{
594 			int s;
595 
596 			if (vp->v_data)
597 				panic("cleaned vnode isn't");
598 			s = splbio();
599 			if (vp->v_numoutput)
600 				panic("Clean vnode has pending I/O's");
601 			splx(s);
602 			if (vp->v_writecount != 0)
603 				panic("Non-zero write count");
604 		}
605 #endif
606 		vp->v_flag = 0;
607 		vp->v_lastw = 0;
608 		vp->v_lasta = 0;
609 		vp->v_cstart = 0;
610 		vp->v_clen = 0;
611 		vp->v_socket = 0;
612 	} else {
613 		simple_unlock(&vnode_free_list_slock);
614 		vp = (struct vnode *) zalloc(vnode_zone);
615 		bzero((char *) vp, sizeof *vp);
616 		mtx_init(&vp->v_interlock, "vnode interlock", MTX_DEF);
617 		vp->v_dd = vp;
618 		cache_purge(vp);
619 		LIST_INIT(&vp->v_cache_src);
620 		TAILQ_INIT(&vp->v_cache_dst);
621 		numvnodes++;
622 	}
623 
624 	TAILQ_INIT(&vp->v_cleanblkhd);
625 	TAILQ_INIT(&vp->v_dirtyblkhd);
626 	vp->v_type = VNON;
627 	vp->v_tag = tag;
628 	vp->v_op = vops;
629 	lockinit(&vp->v_lock, PVFS, "vnlock", 0, LK_NOPAUSE);
630 	insmntque(vp, mp);
631 	*vpp = vp;
632 	vp->v_usecount = 1;
633 	vp->v_data = 0;
634 	splx(s);
635 
636 	vfs_object_create(vp, p, p->p_ucred);
637 	return (0);
638 }
639 
640 /*
641  * Move a vnode from one mount queue to another.
642  */
643 static void
644 insmntque(vp, mp)
645 	register struct vnode *vp;
646 	register struct mount *mp;
647 {
648 
649 	simple_lock(&mntvnode_slock);
650 	/*
651 	 * Delete from old mount point vnode list, if on one.
652 	 */
653 	if (vp->v_mount != NULL)
654 		LIST_REMOVE(vp, v_mntvnodes);
655 	/*
656 	 * Insert into list of vnodes for the new mount point, if available.
657 	 */
658 	if ((vp->v_mount = mp) == NULL) {
659 		simple_unlock(&mntvnode_slock);
660 		return;
661 	}
662 	LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes);
663 	simple_unlock(&mntvnode_slock);
664 }
665 
666 /*
667  * Update outstanding I/O count and do wakeup if requested.
668  */
669 void
670 vwakeup(bp)
671 	register struct buf *bp;
672 {
673 	register struct vnode *vp;
674 
675 	bp->b_flags &= ~B_WRITEINPROG;
676 	if ((vp = bp->b_vp)) {
677 		vp->v_numoutput--;
678 		if (vp->v_numoutput < 0)
679 			panic("vwakeup: neg numoutput");
680 		if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
681 			vp->v_flag &= ~VBWAIT;
682 			wakeup((caddr_t) &vp->v_numoutput);
683 		}
684 	}
685 }
686 
687 /*
688  * Flush out and invalidate all buffers associated with a vnode.
689  * Called with the underlying object locked.
690  */
691 int
692 vinvalbuf(vp, flags, cred, p, slpflag, slptimeo)
693 	register struct vnode *vp;
694 	int flags;
695 	struct ucred *cred;
696 	struct proc *p;
697 	int slpflag, slptimeo;
698 {
699 	register struct buf *bp;
700 	struct buf *nbp, *blist;
701 	int s, error;
702 	vm_object_t object;
703 
704 	if (flags & V_SAVE) {
705 		s = splbio();
706 		while (vp->v_numoutput) {
707 			vp->v_flag |= VBWAIT;
708 			error = tsleep((caddr_t)&vp->v_numoutput,
709 			    slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo);
710 			if (error) {
711 				splx(s);
712 				return (error);
713 			}
714 		}
715 		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
716 			splx(s);
717 			if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, p)) != 0)
718 				return (error);
719 			s = splbio();
720 			if (vp->v_numoutput > 0 ||
721 			    !TAILQ_EMPTY(&vp->v_dirtyblkhd))
722 				panic("vinvalbuf: dirty bufs");
723 		}
724 		splx(s);
725   	}
726 	s = splbio();
727 	for (;;) {
728 		blist = TAILQ_FIRST(&vp->v_cleanblkhd);
729 		if (!blist)
730 			blist = TAILQ_FIRST(&vp->v_dirtyblkhd);
731 		if (!blist)
732 			break;
733 
734 		for (bp = blist; bp; bp = nbp) {
735 			nbp = TAILQ_NEXT(bp, b_vnbufs);
736 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
737 				error = BUF_TIMELOCK(bp,
738 				    LK_EXCLUSIVE | LK_SLEEPFAIL,
739 				    "vinvalbuf", slpflag, slptimeo);
740 				if (error == ENOLCK)
741 					break;
742 				splx(s);
743 				return (error);
744 			}
745 			/*
746 			 * XXX Since there are no node locks for NFS, I
747 			 * believe there is a slight chance that a delayed
748 			 * write will occur while sleeping just above, so
749 			 * check for it.  Note that vfs_bio_awrite expects
750 			 * buffers to reside on a queue, while VOP_BWRITE and
751 			 * brelse do not.
752 			 */
753 			if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
754 				(flags & V_SAVE)) {
755 
756 				if (bp->b_vp == vp) {
757 					if (bp->b_flags & B_CLUSTEROK) {
758 						BUF_UNLOCK(bp);
759 						vfs_bio_awrite(bp);
760 					} else {
761 						bremfree(bp);
762 						bp->b_flags |= B_ASYNC;
763 						BUF_WRITE(bp);
764 					}
765 				} else {
766 					bremfree(bp);
767 					(void) BUF_WRITE(bp);
768 				}
769 				break;
770 			}
771 			bremfree(bp);
772 			bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
773 			bp->b_flags &= ~B_ASYNC;
774 			brelse(bp);
775 		}
776 	}
777 
778 	while (vp->v_numoutput > 0) {
779 		vp->v_flag |= VBWAIT;
780 		tsleep(&vp->v_numoutput, PVM, "vnvlbv", 0);
781 	}
782 
783 	splx(s);
784 
785 	/*
786 	 * Destroy the copy in the VM cache, too.
787 	 */
788 	mtx_enter(&vp->v_interlock, MTX_DEF);
789 	if (VOP_GETVOBJECT(vp, &object) == 0) {
790 		vm_object_page_remove(object, 0, 0,
791 			(flags & V_SAVE) ? TRUE : FALSE);
792 	}
793 	mtx_exit(&vp->v_interlock, MTX_DEF);
794 
795 	if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd))
796 		panic("vinvalbuf: flush failed");
797 	return (0);
798 }
799 
800 /*
801  * Truncate a file's buffer and pages to a specified length.  This
802  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
803  * sync activity.
804  */
805 int
806 vtruncbuf(vp, cred, p, length, blksize)
807 	register struct vnode *vp;
808 	struct ucred *cred;
809 	struct proc *p;
810 	off_t length;
811 	int blksize;
812 {
813 	register struct buf *bp;
814 	struct buf *nbp;
815 	int s, anyfreed;
816 	int trunclbn;
817 
818 	/*
819 	 * Round up to the *next* lbn.
820 	 */
821 	trunclbn = (length + blksize - 1) / blksize;
822 
823 	s = splbio();
824 restart:
825 	anyfreed = 1;
826 	for (;anyfreed;) {
827 		anyfreed = 0;
828 		for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
829 			nbp = TAILQ_NEXT(bp, b_vnbufs);
830 			if (bp->b_lblkno >= trunclbn) {
831 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
832 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
833 					goto restart;
834 				} else {
835 					bremfree(bp);
836 					bp->b_flags |= (B_INVAL | B_RELBUF);
837 					bp->b_flags &= ~B_ASYNC;
838 					brelse(bp);
839 					anyfreed = 1;
840 				}
841 				if (nbp &&
842 				    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
843 				    (nbp->b_vp != vp) ||
844 				    (nbp->b_flags & B_DELWRI))) {
845 					goto restart;
846 				}
847 			}
848 		}
849 
850 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
851 			nbp = TAILQ_NEXT(bp, b_vnbufs);
852 			if (bp->b_lblkno >= trunclbn) {
853 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
854 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
855 					goto restart;
856 				} else {
857 					bremfree(bp);
858 					bp->b_flags |= (B_INVAL | B_RELBUF);
859 					bp->b_flags &= ~B_ASYNC;
860 					brelse(bp);
861 					anyfreed = 1;
862 				}
863 				if (nbp &&
864 				    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
865 				    (nbp->b_vp != vp) ||
866 				    (nbp->b_flags & B_DELWRI) == 0)) {
867 					goto restart;
868 				}
869 			}
870 		}
871 	}
872 
873 	if (length > 0) {
874 restartsync:
875 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
876 			nbp = TAILQ_NEXT(bp, b_vnbufs);
877 			if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
878 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
879 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
880 					goto restart;
881 				} else {
882 					bremfree(bp);
883 					if (bp->b_vp == vp) {
884 						bp->b_flags |= B_ASYNC;
885 					} else {
886 						bp->b_flags &= ~B_ASYNC;
887 					}
888 					BUF_WRITE(bp);
889 				}
890 				goto restartsync;
891 			}
892 
893 		}
894 	}
895 
896 	while (vp->v_numoutput > 0) {
897 		vp->v_flag |= VBWAIT;
898 		tsleep(&vp->v_numoutput, PVM, "vbtrunc", 0);
899 	}
900 
901 	splx(s);
902 
903 	vnode_pager_setsize(vp, length);
904 
905 	return (0);
906 }
907 
908 /*
909  * Associate a buffer with a vnode.
910  */
911 void
912 bgetvp(vp, bp)
913 	register struct vnode *vp;
914 	register struct buf *bp;
915 {
916 	int s;
917 
918 	KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
919 
920 	vhold(vp);
921 	bp->b_vp = vp;
922 	bp->b_dev = vn_todev(vp);
923 	/*
924 	 * Insert onto list for new vnode.
925 	 */
926 	s = splbio();
927 	bp->b_xflags |= BX_VNCLEAN;
928 	bp->b_xflags &= ~BX_VNDIRTY;
929 	TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
930 	splx(s);
931 }
932 
933 /*
934  * Disassociate a buffer from a vnode.
935  */
936 void
937 brelvp(bp)
938 	register struct buf *bp;
939 {
940 	struct vnode *vp;
941 	struct buflists *listheadp;
942 	int s;
943 
944 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
945 
946 	/*
947 	 * Delete from old vnode list, if on one.
948 	 */
949 	vp = bp->b_vp;
950 	s = splbio();
951 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
952 		if (bp->b_xflags & BX_VNDIRTY)
953 			listheadp = &vp->v_dirtyblkhd;
954 		else
955 			listheadp = &vp->v_cleanblkhd;
956 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
957 		bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
958 	}
959 	if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
960 		vp->v_flag &= ~VONWORKLST;
961 		LIST_REMOVE(vp, v_synclist);
962 	}
963 	splx(s);
964 	bp->b_vp = (struct vnode *) 0;
965 	vdrop(vp);
966 }
967 
968 /*
969  * Add an item to the syncer work queue.
970  */
971 static void
972 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
973 {
974 	int s, slot;
975 
976 	s = splbio();
977 
978 	if (vp->v_flag & VONWORKLST) {
979 		LIST_REMOVE(vp, v_synclist);
980 	}
981 
982 	if (delay > syncer_maxdelay - 2)
983 		delay = syncer_maxdelay - 2;
984 	slot = (syncer_delayno + delay) & syncer_mask;
985 
986 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
987 	vp->v_flag |= VONWORKLST;
988 	splx(s);
989 }
990 
991 struct  proc *updateproc;
992 static void sched_sync __P((void));
993 static struct kproc_desc up_kp = {
994 	"syncer",
995 	sched_sync,
996 	&updateproc
997 };
998 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
999 
1000 /*
1001  * System filesystem synchronizer daemon.
1002  */
1003 void
1004 sched_sync(void)
1005 {
1006 	struct synclist *slp;
1007 	struct vnode *vp;
1008 	struct mount *mp;
1009 	long starttime;
1010 	int s;
1011 	struct proc *p = updateproc;
1012 
1013 	mtx_enter(&Giant, MTX_DEF);
1014 
1015 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
1016 	    SHUTDOWN_PRI_LAST);
1017 
1018 	for (;;) {
1019 		kthread_suspend_check(p);
1020 
1021 		starttime = time_second;
1022 
1023 		/*
1024 		 * Push files whose dirty time has expired.  Be careful
1025 		 * of interrupt race on slp queue.
1026 		 */
1027 		s = splbio();
1028 		slp = &syncer_workitem_pending[syncer_delayno];
1029 		syncer_delayno += 1;
1030 		if (syncer_delayno == syncer_maxdelay)
1031 			syncer_delayno = 0;
1032 		splx(s);
1033 
1034 		while ((vp = LIST_FIRST(slp)) != NULL) {
1035 			if (VOP_ISLOCKED(vp, NULL) == 0 &&
1036 			    vn_start_write(vp, &mp, V_NOWAIT) == 0) {
1037 				vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
1038 				(void) VOP_FSYNC(vp, p->p_ucred, MNT_LAZY, p);
1039 				VOP_UNLOCK(vp, 0, p);
1040 				vn_finished_write(mp);
1041 			}
1042 			s = splbio();
1043 			if (LIST_FIRST(slp) == vp) {
1044 				/*
1045 				 * Note: v_tag VT_VFS vps can remain on the
1046 				 * worklist too with no dirty blocks, but
1047 				 * since sync_fsync() moves it to a different
1048 				 * slot we are safe.
1049 				 */
1050 				if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1051 				    !vn_isdisk(vp, NULL))
1052 					panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
1053 				/*
1054 				 * Put us back on the worklist.  The worklist
1055 				 * routine will remove us from our current
1056 				 * position and then add us back in at a later
1057 				 * position.
1058 				 */
1059 				vn_syncer_add_to_worklist(vp, syncdelay);
1060 			}
1061 			splx(s);
1062 		}
1063 
1064 		/*
1065 		 * Do soft update processing.
1066 		 */
1067 #ifdef SOFTUPDATES
1068 		softdep_process_worklist(NULL);
1069 #endif
1070 
1071 		/*
1072 		 * The variable rushjob allows the kernel to speed up the
1073 		 * processing of the filesystem syncer process. A rushjob
1074 		 * value of N tells the filesystem syncer to process the next
1075 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1076 		 * is used by the soft update code to speed up the filesystem
1077 		 * syncer process when the incore state is getting so far
1078 		 * ahead of the disk that the kernel memory pool is being
1079 		 * threatened with exhaustion.
1080 		 */
1081 		if (rushjob > 0) {
1082 			rushjob -= 1;
1083 			continue;
1084 		}
1085 		/*
1086 		 * If it has taken us less than a second to process the
1087 		 * current work, then wait. Otherwise start right over
1088 		 * again. We can still lose time if any single round
1089 		 * takes more than two seconds, but it does not really
1090 		 * matter as we are just trying to generally pace the
1091 		 * filesystem activity.
1092 		 */
1093 		if (time_second == starttime)
1094 			tsleep(&lbolt, PPAUSE, "syncer", 0);
1095 	}
1096 }
1097 
1098 /*
1099  * Request the syncer daemon to speed up its work.
1100  * We never push it to speed up more than half of its
1101  * normal turn time, otherwise it could take over the cpu.
1102  */
1103 int
1104 speedup_syncer()
1105 {
1106 
1107 	mtx_enter(&sched_lock, MTX_SPIN);
1108 	if (updateproc->p_wchan == &lbolt)
1109 		setrunnable(updateproc);
1110 	mtx_exit(&sched_lock, MTX_SPIN);
1111 	if (rushjob < syncdelay / 2) {
1112 		rushjob += 1;
1113 		stat_rush_requests += 1;
1114 		return (1);
1115 	}
1116 	return(0);
1117 }
1118 
1119 /*
1120  * Associate a p-buffer with a vnode.
1121  *
1122  * Also sets B_PAGING flag to indicate that vnode is not fully associated
1123  * with the buffer.  i.e. the bp has not been linked into the vnode or
1124  * ref-counted.
1125  */
1126 void
1127 pbgetvp(vp, bp)
1128 	register struct vnode *vp;
1129 	register struct buf *bp;
1130 {
1131 
1132 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1133 
1134 	bp->b_vp = vp;
1135 	bp->b_flags |= B_PAGING;
1136 	bp->b_dev = vn_todev(vp);
1137 }
1138 
1139 /*
1140  * Disassociate a p-buffer from a vnode.
1141  */
1142 void
1143 pbrelvp(bp)
1144 	register struct buf *bp;
1145 {
1146 
1147 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1148 
1149 	/* XXX REMOVE ME */
1150 	if (bp->b_vnbufs.tqe_next != NULL) {
1151 		panic(
1152 		    "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1153 		    bp,
1154 		    (int)bp->b_flags
1155 		);
1156 	}
1157 	bp->b_vp = (struct vnode *) 0;
1158 	bp->b_flags &= ~B_PAGING;
1159 }
1160 
1161 /*
1162  * Change the vnode a pager buffer is associated with.
1163  */
1164 void
1165 pbreassignbuf(bp, newvp)
1166 	struct buf *bp;
1167 	struct vnode *newvp;
1168 {
1169 
1170 	KASSERT(bp->b_flags & B_PAGING,
1171 	    ("pbreassignbuf() on non phys bp %p", bp));
1172 	bp->b_vp = newvp;
1173 }
1174 
1175 /*
1176  * Reassign a buffer from one vnode to another.
1177  * Used to assign file specific control information
1178  * (indirect blocks) to the vnode to which they belong.
1179  */
1180 void
1181 reassignbuf(bp, newvp)
1182 	register struct buf *bp;
1183 	register struct vnode *newvp;
1184 {
1185 	struct buflists *listheadp;
1186 	int delay;
1187 	int s;
1188 
1189 	if (newvp == NULL) {
1190 		printf("reassignbuf: NULL");
1191 		return;
1192 	}
1193 	++reassignbufcalls;
1194 
1195 	/*
1196 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1197 	 * is not fully linked in.
1198 	 */
1199 	if (bp->b_flags & B_PAGING)
1200 		panic("cannot reassign paging buffer");
1201 
1202 	s = splbio();
1203 	/*
1204 	 * Delete from old vnode list, if on one.
1205 	 */
1206 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1207 		if (bp->b_xflags & BX_VNDIRTY)
1208 			listheadp = &bp->b_vp->v_dirtyblkhd;
1209 		else
1210 			listheadp = &bp->b_vp->v_cleanblkhd;
1211 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1212 		bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1213 		if (bp->b_vp != newvp) {
1214 			vdrop(bp->b_vp);
1215 			bp->b_vp = NULL;	/* for clarification */
1216 		}
1217 	}
1218 	/*
1219 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1220 	 * of clean buffers.
1221 	 */
1222 	if (bp->b_flags & B_DELWRI) {
1223 		struct buf *tbp;
1224 
1225 		listheadp = &newvp->v_dirtyblkhd;
1226 		if ((newvp->v_flag & VONWORKLST) == 0) {
1227 			switch (newvp->v_type) {
1228 			case VDIR:
1229 				delay = dirdelay;
1230 				break;
1231 			case VCHR:
1232 				if (newvp->v_rdev->si_mountpoint != NULL) {
1233 					delay = metadelay;
1234 					break;
1235 				}
1236 				/* fall through */
1237 			default:
1238 				delay = filedelay;
1239 			}
1240 			vn_syncer_add_to_worklist(newvp, delay);
1241 		}
1242 		bp->b_xflags |= BX_VNDIRTY;
1243 		tbp = TAILQ_FIRST(listheadp);
1244 		if (tbp == NULL ||
1245 		    bp->b_lblkno == 0 ||
1246 		    (bp->b_lblkno > 0 && tbp->b_lblkno < 0) ||
1247 		    (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) {
1248 			TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1249 			++reassignbufsortgood;
1250 		} else if (bp->b_lblkno < 0) {
1251 			TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1252 			++reassignbufsortgood;
1253 		} else if (reassignbufmethod == 1) {
1254 			/*
1255 			 * New sorting algorithm, only handle sequential case,
1256 			 * otherwise append to end (but before metadata)
1257 			 */
1258 			if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL &&
1259 			    (tbp->b_xflags & BX_VNDIRTY)) {
1260 				/*
1261 				 * Found the best place to insert the buffer
1262 				 */
1263 				TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1264 				++reassignbufsortgood;
1265 			} else {
1266 				/*
1267 				 * Missed, append to end, but before meta-data.
1268 				 * We know that the head buffer in the list is
1269 				 * not meta-data due to prior conditionals.
1270 				 *
1271 				 * Indirect effects:  NFS second stage write
1272 				 * tends to wind up here, giving maximum
1273 				 * distance between the unstable write and the
1274 				 * commit rpc.
1275 				 */
1276 				tbp = TAILQ_LAST(listheadp, buflists);
1277 				while (tbp && tbp->b_lblkno < 0)
1278 					tbp = TAILQ_PREV(tbp, buflists, b_vnbufs);
1279 				TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1280 				++reassignbufsortbad;
1281 			}
1282 		} else {
1283 			/*
1284 			 * Old sorting algorithm, scan queue and insert
1285 			 */
1286 			struct buf *ttbp;
1287 			while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1288 			    (ttbp->b_lblkno < bp->b_lblkno)) {
1289 				++reassignbufloops;
1290 				tbp = ttbp;
1291 			}
1292 			TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1293 		}
1294 	} else {
1295 		bp->b_xflags |= BX_VNCLEAN;
1296 		TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1297 		if ((newvp->v_flag & VONWORKLST) &&
1298 		    TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1299 			newvp->v_flag &= ~VONWORKLST;
1300 			LIST_REMOVE(newvp, v_synclist);
1301 		}
1302 	}
1303 	if (bp->b_vp != newvp) {
1304 		bp->b_vp = newvp;
1305 		vhold(bp->b_vp);
1306 	}
1307 	splx(s);
1308 }
1309 
1310 /*
1311  * Create a vnode for a device.
1312  * Used for mounting the root file system.
1313  */
1314 int
1315 bdevvp(dev, vpp)
1316 	dev_t dev;
1317 	struct vnode **vpp;
1318 {
1319 	register struct vnode *vp;
1320 	struct vnode *nvp;
1321 	int error;
1322 
1323 	if (dev == NODEV) {
1324 		*vpp = NULLVP;
1325 		return (ENXIO);
1326 	}
1327 	if (vfinddev(dev, VCHR, vpp))
1328 		return (0);
1329 	error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1330 	if (error) {
1331 		*vpp = NULLVP;
1332 		return (error);
1333 	}
1334 	vp = nvp;
1335 	vp->v_type = VCHR;
1336 	addalias(vp, dev);
1337 	*vpp = vp;
1338 	return (0);
1339 }
1340 
1341 /*
1342  * Add vnode to the alias list hung off the dev_t.
1343  *
1344  * The reason for this gunk is that multiple vnodes can reference
1345  * the same physical device, so checking vp->v_usecount to see
1346  * how many users there are is inadequate; the v_usecount for
1347  * the vnodes need to be accumulated.  vcount() does that.
1348  */
1349 struct vnode *
1350 addaliasu(nvp, nvp_rdev)
1351 	struct vnode *nvp;
1352 	udev_t nvp_rdev;
1353 {
1354 	struct vnode *ovp;
1355 	vop_t **ops;
1356 	dev_t dev;
1357 
1358 	if (nvp->v_type == VBLK)
1359 		return (nvp);
1360 	if (nvp->v_type != VCHR)
1361 		panic("addaliasu on non-special vnode");
1362 	dev = udev2dev(nvp_rdev, 0);
1363 	/*
1364 	 * Check to see if we have a bdevvp vnode with no associated
1365 	 * filesystem. If so, we want to associate the filesystem of
1366 	 * the new newly instigated vnode with the bdevvp vnode and
1367 	 * discard the newly created vnode rather than leaving the
1368 	 * bdevvp vnode lying around with no associated filesystem.
1369 	 */
1370 	if (vfinddev(dev, nvp->v_type, &ovp) == 0 || ovp->v_data != NULL) {
1371 		addalias(nvp, dev);
1372 		return (nvp);
1373 	}
1374 	/*
1375 	 * Discard unneeded vnode, but save its node specific data.
1376 	 * Note that if there is a lock, it is carried over in the
1377 	 * node specific data to the replacement vnode.
1378 	 */
1379 	vref(ovp);
1380 	ovp->v_data = nvp->v_data;
1381 	ovp->v_tag = nvp->v_tag;
1382 	nvp->v_data = NULL;
1383 	ops = nvp->v_op;
1384 	nvp->v_op = ovp->v_op;
1385 	ovp->v_op = ops;
1386 	lockinit(&ovp->v_lock, PVFS, "vnlock", 0, LK_NOPAUSE);
1387 	if (nvp->v_vnlock)
1388 		ovp->v_vnlock = &ovp->v_lock;
1389 	insmntque(ovp, nvp->v_mount);
1390 	vrele(nvp);
1391 	vgone(nvp);
1392 	return (ovp);
1393 }
1394 
1395 /* This is a local helper function that do the same as addaliasu, but for a
1396  * dev_t instead of an udev_t. */
1397 static void
1398 addalias(nvp, dev)
1399 	struct vnode *nvp;
1400 	dev_t dev;
1401 {
1402 
1403 	KASSERT(nvp->v_type == VCHR, ("addalias on non-special vnode"));
1404 	nvp->v_rdev = dev;
1405 	simple_lock(&spechash_slock);
1406 	SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
1407 	simple_unlock(&spechash_slock);
1408 }
1409 
1410 /*
1411  * Grab a particular vnode from the free list, increment its
1412  * reference count and lock it. The vnode lock bit is set if the
1413  * vnode is being eliminated in vgone. The process is awakened
1414  * when the transition is completed, and an error returned to
1415  * indicate that the vnode is no longer usable (possibly having
1416  * been changed to a new file system type).
1417  */
1418 int
1419 vget(vp, flags, p)
1420 	register struct vnode *vp;
1421 	int flags;
1422 	struct proc *p;
1423 {
1424 	int error;
1425 
1426 	/*
1427 	 * If the vnode is in the process of being cleaned out for
1428 	 * another use, we wait for the cleaning to finish and then
1429 	 * return failure. Cleaning is determined by checking that
1430 	 * the VXLOCK flag is set.
1431 	 */
1432 	if ((flags & LK_INTERLOCK) == 0)
1433 		mtx_enter(&vp->v_interlock, MTX_DEF);
1434 	if (vp->v_flag & VXLOCK) {
1435 		if (vp->v_vxproc == curproc) {
1436 			printf("VXLOCK interlock avoided\n");
1437 		} else {
1438 			vp->v_flag |= VXWANT;
1439 			msleep((caddr_t)vp, &vp->v_interlock, PINOD | PDROP,
1440 			    "vget", 0);
1441 			return (ENOENT);
1442 		}
1443 	}
1444 
1445 	vp->v_usecount++;
1446 
1447 	if (VSHOULDBUSY(vp))
1448 		vbusy(vp);
1449 	if (flags & LK_TYPE_MASK) {
1450 		if ((error = vn_lock(vp, flags | LK_INTERLOCK, p)) != 0) {
1451 			/*
1452 			 * must expand vrele here because we do not want
1453 			 * to call VOP_INACTIVE if the reference count
1454 			 * drops back to zero since it was never really
1455 			 * active. We must remove it from the free list
1456 			 * before sleeping so that multiple processes do
1457 			 * not try to recycle it.
1458 			 */
1459 			mtx_enter(&vp->v_interlock, MTX_DEF);
1460 			vp->v_usecount--;
1461 			if (VSHOULDFREE(vp))
1462 				vfree(vp);
1463 			mtx_exit(&vp->v_interlock, MTX_DEF);
1464 		}
1465 		return (error);
1466 	}
1467 	mtx_exit(&vp->v_interlock, MTX_DEF);
1468 	return (0);
1469 }
1470 
1471 /*
1472  * Increase the reference count of a vnode.
1473  */
1474 void
1475 vref(struct vnode *vp)
1476 {
1477 	mtx_enter(&vp->v_interlock, MTX_DEF);
1478 	vp->v_usecount++;
1479 	mtx_exit(&vp->v_interlock, MTX_DEF);
1480 }
1481 
1482 /*
1483  * Vnode put/release.
1484  * If count drops to zero, call inactive routine and return to freelist.
1485  */
1486 void
1487 vrele(vp)
1488 	struct vnode *vp;
1489 {
1490 	struct proc *p = curproc;	/* XXX */
1491 
1492 	KASSERT(vp != NULL, ("vrele: null vp"));
1493 
1494 	mtx_enter(&vp->v_interlock, MTX_DEF);
1495 
1496 	KASSERT(vp->v_writecount < vp->v_usecount, ("vrele: missed vn_close"));
1497 
1498 	if (vp->v_usecount > 1) {
1499 
1500 		vp->v_usecount--;
1501 		mtx_exit(&vp->v_interlock, MTX_DEF);
1502 
1503 		return;
1504 	}
1505 
1506 	if (vp->v_usecount == 1) {
1507 
1508 		vp->v_usecount--;
1509 		if (VSHOULDFREE(vp))
1510 			vfree(vp);
1511 	/*
1512 	 * If we are doing a vput, the node is already locked, and we must
1513 	 * call VOP_INACTIVE with the node locked.  So, in the case of
1514 	 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE.
1515 	 */
1516 		if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, p) == 0) {
1517 			VOP_INACTIVE(vp, p);
1518 		}
1519 
1520 	} else {
1521 #ifdef DIAGNOSTIC
1522 		vprint("vrele: negative ref count", vp);
1523 		mtx_exit(&vp->v_interlock, MTX_DEF);
1524 #endif
1525 		panic("vrele: negative ref cnt");
1526 	}
1527 }
1528 
1529 /*
1530  * Release an already locked vnode.  This give the same effects as
1531  * unlock+vrele(), but takes less time and avoids releasing and
1532  * re-aquiring the lock (as vrele() aquires the lock internally.)
1533  */
1534 void
1535 vput(vp)
1536 	struct vnode *vp;
1537 {
1538 	struct proc *p = curproc;	/* XXX */
1539 
1540 	KASSERT(vp != NULL, ("vput: null vp"));
1541 	mtx_enter(&vp->v_interlock, MTX_DEF);
1542 	KASSERT(vp->v_writecount < vp->v_usecount, ("vput: missed vn_close"));
1543 
1544 	if (vp->v_usecount > 1) {
1545 
1546 		vp->v_usecount--;
1547 		VOP_UNLOCK(vp, LK_INTERLOCK, p);
1548 		return;
1549 
1550 	}
1551 
1552 	if (vp->v_usecount == 1) {
1553 
1554 		vp->v_usecount--;
1555 		if (VSHOULDFREE(vp))
1556 			vfree(vp);
1557 	/*
1558 	 * If we are doing a vput, the node is already locked, and we must
1559 	 * call VOP_INACTIVE with the node locked.  So, in the case of
1560 	 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE.
1561 	 */
1562 		mtx_exit(&vp->v_interlock, MTX_DEF);
1563 		VOP_INACTIVE(vp, p);
1564 
1565 	} else {
1566 #ifdef DIAGNOSTIC
1567 		vprint("vput: negative ref count", vp);
1568 #endif
1569 		panic("vput: negative ref cnt");
1570 	}
1571 }
1572 
1573 /*
1574  * Somebody doesn't want the vnode recycled.
1575  */
1576 void
1577 vhold(vp)
1578 	register struct vnode *vp;
1579 {
1580 	int s;
1581 
1582   	s = splbio();
1583 	vp->v_holdcnt++;
1584 	if (VSHOULDBUSY(vp))
1585 		vbusy(vp);
1586 	splx(s);
1587 }
1588 
1589 /*
1590  * Note that there is one less who cares about this vnode.  vdrop() is the
1591  * opposite of vhold().
1592  */
1593 void
1594 vdrop(vp)
1595 	register struct vnode *vp;
1596 {
1597 	int s;
1598 
1599 	s = splbio();
1600 	if (vp->v_holdcnt <= 0)
1601 		panic("vdrop: holdcnt");
1602 	vp->v_holdcnt--;
1603 	if (VSHOULDFREE(vp))
1604 		vfree(vp);
1605 	splx(s);
1606 }
1607 
1608 /*
1609  * Remove any vnodes in the vnode table belonging to mount point mp.
1610  *
1611  * If MNT_NOFORCE is specified, there should not be any active ones,
1612  * return error if any are found (nb: this is a user error, not a
1613  * system error). If MNT_FORCE is specified, detach any active vnodes
1614  * that are found.
1615  */
1616 #ifdef DIAGNOSTIC
1617 static int busyprt = 0;		/* print out busy vnodes */
1618 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1619 #endif
1620 
1621 int
1622 vflush(mp, skipvp, flags)
1623 	struct mount *mp;
1624 	struct vnode *skipvp;
1625 	int flags;
1626 {
1627 	struct proc *p = curproc;	/* XXX */
1628 	struct vnode *vp, *nvp;
1629 	int busy = 0;
1630 
1631 	simple_lock(&mntvnode_slock);
1632 loop:
1633 	for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp; vp = nvp) {
1634 		/*
1635 		 * Make sure this vnode wasn't reclaimed in getnewvnode().
1636 		 * Start over if it has (it won't be on the list anymore).
1637 		 */
1638 		if (vp->v_mount != mp)
1639 			goto loop;
1640 		nvp = LIST_NEXT(vp, v_mntvnodes);
1641 		/*
1642 		 * Skip over a selected vnode.
1643 		 */
1644 		if (vp == skipvp)
1645 			continue;
1646 
1647 		mtx_enter(&vp->v_interlock, MTX_DEF);
1648 		/*
1649 		 * Skip over a vnodes marked VSYSTEM.
1650 		 */
1651 		if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1652 			mtx_exit(&vp->v_interlock, MTX_DEF);
1653 			continue;
1654 		}
1655 		/*
1656 		 * If WRITECLOSE is set, only flush out regular file vnodes
1657 		 * open for writing.
1658 		 */
1659 		if ((flags & WRITECLOSE) &&
1660 		    (vp->v_writecount == 0 || vp->v_type != VREG)) {
1661 			mtx_exit(&vp->v_interlock, MTX_DEF);
1662 			continue;
1663 		}
1664 
1665 		/*
1666 		 * With v_usecount == 0, all we need to do is clear out the
1667 		 * vnode data structures and we are done.
1668 		 */
1669 		if (vp->v_usecount == 0) {
1670 			simple_unlock(&mntvnode_slock);
1671 			vgonel(vp, p);
1672 			simple_lock(&mntvnode_slock);
1673 			continue;
1674 		}
1675 
1676 		/*
1677 		 * If FORCECLOSE is set, forcibly close the vnode. For block
1678 		 * or character devices, revert to an anonymous device. For
1679 		 * all other files, just kill them.
1680 		 */
1681 		if (flags & FORCECLOSE) {
1682 			simple_unlock(&mntvnode_slock);
1683 			if (vp->v_type != VCHR) {
1684 				vgonel(vp, p);
1685 			} else {
1686 				vclean(vp, 0, p);
1687 				vp->v_op = spec_vnodeop_p;
1688 				insmntque(vp, (struct mount *) 0);
1689 			}
1690 			simple_lock(&mntvnode_slock);
1691 			continue;
1692 		}
1693 #ifdef DIAGNOSTIC
1694 		if (busyprt)
1695 			vprint("vflush: busy vnode", vp);
1696 #endif
1697 		mtx_exit(&vp->v_interlock, MTX_DEF);
1698 		busy++;
1699 	}
1700 	simple_unlock(&mntvnode_slock);
1701 	if (busy)
1702 		return (EBUSY);
1703 	return (0);
1704 }
1705 
1706 /*
1707  * Disassociate the underlying file system from a vnode.
1708  */
1709 static void
1710 vclean(vp, flags, p)
1711 	struct vnode *vp;
1712 	int flags;
1713 	struct proc *p;
1714 {
1715 	int active;
1716 
1717 	/*
1718 	 * Check to see if the vnode is in use. If so we have to reference it
1719 	 * before we clean it out so that its count cannot fall to zero and
1720 	 * generate a race against ourselves to recycle it.
1721 	 */
1722 	if ((active = vp->v_usecount))
1723 		vp->v_usecount++;
1724 
1725 	/*
1726 	 * Prevent the vnode from being recycled or brought into use while we
1727 	 * clean it out.
1728 	 */
1729 	if (vp->v_flag & VXLOCK)
1730 		panic("vclean: deadlock");
1731 	vp->v_flag |= VXLOCK;
1732 	vp->v_vxproc = curproc;
1733 	/*
1734 	 * Even if the count is zero, the VOP_INACTIVE routine may still
1735 	 * have the object locked while it cleans it out. The VOP_LOCK
1736 	 * ensures that the VOP_INACTIVE routine is done with its work.
1737 	 * For active vnodes, it ensures that no other activity can
1738 	 * occur while the underlying object is being cleaned out.
1739 	 */
1740 	VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, p);
1741 
1742 	/*
1743 	 * Clean out any buffers associated with the vnode.
1744 	 * If the flush fails, just toss the buffers.
1745 	 */
1746 	if (flags & DOCLOSE) {
1747 		if (TAILQ_FIRST(&vp->v_dirtyblkhd) != NULL)
1748 			(void) vn_write_suspend_wait(vp, NULL, V_WAIT);
1749 		if (vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0) != 0)
1750 			vinvalbuf(vp, 0, NOCRED, p, 0, 0);
1751 	}
1752 
1753 	VOP_DESTROYVOBJECT(vp);
1754 
1755 	/*
1756 	 * If purging an active vnode, it must be closed and
1757 	 * deactivated before being reclaimed. Note that the
1758 	 * VOP_INACTIVE will unlock the vnode.
1759 	 */
1760 	if (active) {
1761 		if (flags & DOCLOSE)
1762 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, p);
1763 		VOP_INACTIVE(vp, p);
1764 	} else {
1765 		/*
1766 		 * Any other processes trying to obtain this lock must first
1767 		 * wait for VXLOCK to clear, then call the new lock operation.
1768 		 */
1769 		VOP_UNLOCK(vp, 0, p);
1770 	}
1771 	/*
1772 	 * Reclaim the vnode.
1773 	 */
1774 	if (VOP_RECLAIM(vp, p))
1775 		panic("vclean: cannot reclaim");
1776 
1777 	if (active) {
1778 		/*
1779 		 * Inline copy of vrele() since VOP_INACTIVE
1780 		 * has already been called.
1781 		 */
1782 		mtx_enter(&vp->v_interlock, MTX_DEF);
1783 		if (--vp->v_usecount <= 0) {
1784 #ifdef DIAGNOSTIC
1785 			if (vp->v_usecount < 0 || vp->v_writecount != 0) {
1786 				vprint("vclean: bad ref count", vp);
1787 				panic("vclean: ref cnt");
1788 			}
1789 #endif
1790 			vfree(vp);
1791 		}
1792 		mtx_exit(&vp->v_interlock, MTX_DEF);
1793 	}
1794 
1795 	cache_purge(vp);
1796 	vp->v_vnlock = NULL;
1797 	lockdestroy(&vp->v_lock);
1798 
1799 	if (VSHOULDFREE(vp))
1800 		vfree(vp);
1801 
1802 	/*
1803 	 * Done with purge, notify sleepers of the grim news.
1804 	 */
1805 	vp->v_op = dead_vnodeop_p;
1806 	vn_pollgone(vp);
1807 	vp->v_tag = VT_NON;
1808 	vp->v_flag &= ~VXLOCK;
1809 	vp->v_vxproc = NULL;
1810 	if (vp->v_flag & VXWANT) {
1811 		vp->v_flag &= ~VXWANT;
1812 		wakeup((caddr_t) vp);
1813 	}
1814 }
1815 
1816 /*
1817  * Eliminate all activity associated with the requested vnode
1818  * and with all vnodes aliased to the requested vnode.
1819  */
1820 int
1821 vop_revoke(ap)
1822 	struct vop_revoke_args /* {
1823 		struct vnode *a_vp;
1824 		int a_flags;
1825 	} */ *ap;
1826 {
1827 	struct vnode *vp, *vq;
1828 	dev_t dev;
1829 
1830 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
1831 
1832 	vp = ap->a_vp;
1833 	/*
1834 	 * If a vgone (or vclean) is already in progress,
1835 	 * wait until it is done and return.
1836 	 */
1837 	if (vp->v_flag & VXLOCK) {
1838 		vp->v_flag |= VXWANT;
1839 		msleep((caddr_t)vp, &vp->v_interlock, PINOD | PDROP,
1840 		    "vop_revokeall", 0);
1841 		return (0);
1842 	}
1843 	dev = vp->v_rdev;
1844 	for (;;) {
1845 		simple_lock(&spechash_slock);
1846 		vq = SLIST_FIRST(&dev->si_hlist);
1847 		simple_unlock(&spechash_slock);
1848 		if (!vq)
1849 			break;
1850 		vgone(vq);
1851 	}
1852 	return (0);
1853 }
1854 
1855 /*
1856  * Recycle an unused vnode to the front of the free list.
1857  * Release the passed interlock if the vnode will be recycled.
1858  */
1859 int
1860 vrecycle(vp, inter_lkp, p)
1861 	struct vnode *vp;
1862 	struct simplelock *inter_lkp;
1863 	struct proc *p;
1864 {
1865 
1866 	mtx_enter(&vp->v_interlock, MTX_DEF);
1867 	if (vp->v_usecount == 0) {
1868 		if (inter_lkp) {
1869 			simple_unlock(inter_lkp);
1870 		}
1871 		vgonel(vp, p);
1872 		return (1);
1873 	}
1874 	mtx_exit(&vp->v_interlock, MTX_DEF);
1875 	return (0);
1876 }
1877 
1878 /*
1879  * Eliminate all activity associated with a vnode
1880  * in preparation for reuse.
1881  */
1882 void
1883 vgone(vp)
1884 	register struct vnode *vp;
1885 {
1886 	struct proc *p = curproc;	/* XXX */
1887 
1888 	mtx_enter(&vp->v_interlock, MTX_DEF);
1889 	vgonel(vp, p);
1890 }
1891 
1892 /*
1893  * vgone, with the vp interlock held.
1894  */
1895 void
1896 vgonel(vp, p)
1897 	struct vnode *vp;
1898 	struct proc *p;
1899 {
1900 	int s;
1901 
1902 	/*
1903 	 * If a vgone (or vclean) is already in progress,
1904 	 * wait until it is done and return.
1905 	 */
1906 	if (vp->v_flag & VXLOCK) {
1907 		vp->v_flag |= VXWANT;
1908 		msleep((caddr_t)vp, &vp->v_interlock, PINOD | PDROP,
1909 		    "vgone", 0);
1910 		return;
1911 	}
1912 
1913 	/*
1914 	 * Clean out the filesystem specific data.
1915 	 */
1916 	vclean(vp, DOCLOSE, p);
1917 	mtx_enter(&vp->v_interlock, MTX_DEF);
1918 
1919 	/*
1920 	 * Delete from old mount point vnode list, if on one.
1921 	 */
1922 	if (vp->v_mount != NULL)
1923 		insmntque(vp, (struct mount *)0);
1924 	/*
1925 	 * If special device, remove it from special device alias list
1926 	 * if it is on one.
1927 	 */
1928 	if (vp->v_type == VCHR && vp->v_rdev != NULL && vp->v_rdev != NODEV) {
1929 		simple_lock(&spechash_slock);
1930 		SLIST_REMOVE(&vp->v_rdev->si_hlist, vp, vnode, v_specnext);
1931 		freedev(vp->v_rdev);
1932 		simple_unlock(&spechash_slock);
1933 		vp->v_rdev = NULL;
1934 	}
1935 
1936 	/*
1937 	 * If it is on the freelist and not already at the head,
1938 	 * move it to the head of the list. The test of the
1939 	 * VDOOMED flag and the reference count of zero is because
1940 	 * it will be removed from the free list by getnewvnode,
1941 	 * but will not have its reference count incremented until
1942 	 * after calling vgone. If the reference count were
1943 	 * incremented first, vgone would (incorrectly) try to
1944 	 * close the previous instance of the underlying object.
1945 	 */
1946 	if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
1947 		s = splbio();
1948 		simple_lock(&vnode_free_list_slock);
1949 		if (vp->v_flag & VFREE)
1950 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
1951 		else
1952 			freevnodes++;
1953 		vp->v_flag |= VFREE;
1954 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
1955 		simple_unlock(&vnode_free_list_slock);
1956 		splx(s);
1957 	}
1958 
1959 	vp->v_type = VBAD;
1960 	mtx_exit(&vp->v_interlock, MTX_DEF);
1961 }
1962 
1963 /*
1964  * Lookup a vnode by device number.
1965  */
1966 int
1967 vfinddev(dev, type, vpp)
1968 	dev_t dev;
1969 	enum vtype type;
1970 	struct vnode **vpp;
1971 {
1972 	struct vnode *vp;
1973 
1974 	simple_lock(&spechash_slock);
1975 	SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
1976 		if (type == vp->v_type) {
1977 			*vpp = vp;
1978 			simple_unlock(&spechash_slock);
1979 			return (1);
1980 		}
1981 	}
1982 	simple_unlock(&spechash_slock);
1983 	return (0);
1984 }
1985 
1986 /*
1987  * Calculate the total number of references to a special device.
1988  */
1989 int
1990 vcount(vp)
1991 	struct vnode *vp;
1992 {
1993 	struct vnode *vq;
1994 	int count;
1995 
1996 	count = 0;
1997 	simple_lock(&spechash_slock);
1998 	SLIST_FOREACH(vq, &vp->v_rdev->si_hlist, v_specnext)
1999 		count += vq->v_usecount;
2000 	simple_unlock(&spechash_slock);
2001 	return (count);
2002 }
2003 
2004 /*
2005  * Same as above, but using the dev_t as argument
2006  */
2007 int
2008 count_dev(dev)
2009 	dev_t dev;
2010 {
2011 	struct vnode *vp;
2012 
2013 	vp = SLIST_FIRST(&dev->si_hlist);
2014 	if (vp == NULL)
2015 		return (0);
2016 	return(vcount(vp));
2017 }
2018 
2019 /*
2020  * Print out a description of a vnode.
2021  */
2022 static char *typename[] =
2023 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2024 
2025 void
2026 vprint(label, vp)
2027 	char *label;
2028 	struct vnode *vp;
2029 {
2030 	char buf[96];
2031 
2032 	if (label != NULL)
2033 		printf("%s: %p: ", label, (void *)vp);
2034 	else
2035 		printf("%p: ", (void *)vp);
2036 	printf("type %s, usecount %d, writecount %d, refcount %d,",
2037 	    typename[vp->v_type], vp->v_usecount, vp->v_writecount,
2038 	    vp->v_holdcnt);
2039 	buf[0] = '\0';
2040 	if (vp->v_flag & VROOT)
2041 		strcat(buf, "|VROOT");
2042 	if (vp->v_flag & VTEXT)
2043 		strcat(buf, "|VTEXT");
2044 	if (vp->v_flag & VSYSTEM)
2045 		strcat(buf, "|VSYSTEM");
2046 	if (vp->v_flag & VXLOCK)
2047 		strcat(buf, "|VXLOCK");
2048 	if (vp->v_flag & VXWANT)
2049 		strcat(buf, "|VXWANT");
2050 	if (vp->v_flag & VBWAIT)
2051 		strcat(buf, "|VBWAIT");
2052 	if (vp->v_flag & VDOOMED)
2053 		strcat(buf, "|VDOOMED");
2054 	if (vp->v_flag & VFREE)
2055 		strcat(buf, "|VFREE");
2056 	if (vp->v_flag & VOBJBUF)
2057 		strcat(buf, "|VOBJBUF");
2058 	if (buf[0] != '\0')
2059 		printf(" flags (%s)", &buf[1]);
2060 	if (vp->v_data == NULL) {
2061 		printf("\n");
2062 	} else {
2063 		printf("\n\t");
2064 		VOP_PRINT(vp);
2065 	}
2066 }
2067 
2068 #ifdef DDB
2069 #include <ddb/ddb.h>
2070 /*
2071  * List all of the locked vnodes in the system.
2072  * Called when debugging the kernel.
2073  */
2074 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
2075 {
2076 	struct proc *p = curproc;	/* XXX */
2077 	struct mount *mp, *nmp;
2078 	struct vnode *vp;
2079 
2080 	printf("Locked vnodes\n");
2081 	mtx_enter(&mountlist_mtx, MTX_DEF);
2082 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2083 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, p)) {
2084 			nmp = TAILQ_NEXT(mp, mnt_list);
2085 			continue;
2086 		}
2087 		LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2088 			if (VOP_ISLOCKED(vp, NULL))
2089 				vprint((char *)0, vp);
2090 		}
2091 		mtx_enter(&mountlist_mtx, MTX_DEF);
2092 		nmp = TAILQ_NEXT(mp, mnt_list);
2093 		vfs_unbusy(mp, p);
2094 	}
2095 	mtx_exit(&mountlist_mtx, MTX_DEF);
2096 }
2097 #endif
2098 
2099 /*
2100  * Top level filesystem related information gathering.
2101  */
2102 static int	sysctl_ovfs_conf __P((SYSCTL_HANDLER_ARGS));
2103 
2104 static int
2105 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2106 {
2107 	int *name = (int *)arg1 - 1;	/* XXX */
2108 	u_int namelen = arg2 + 1;	/* XXX */
2109 	struct vfsconf *vfsp;
2110 
2111 #if 1 || defined(COMPAT_PRELITE2)
2112 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2113 	if (namelen == 1)
2114 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2115 #endif
2116 
2117 	/* XXX the below code does not compile; vfs_sysctl does not exist. */
2118 #ifdef notyet
2119 	/* all sysctl names at this level are at least name and field */
2120 	if (namelen < 2)
2121 		return (ENOTDIR);		/* overloaded */
2122 	if (name[0] != VFS_GENERIC) {
2123 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2124 			if (vfsp->vfc_typenum == name[0])
2125 				break;
2126 		if (vfsp == NULL)
2127 			return (EOPNOTSUPP);
2128 		return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2129 		    oldp, oldlenp, newp, newlen, p));
2130 	}
2131 #endif
2132 	switch (name[1]) {
2133 	case VFS_MAXTYPENUM:
2134 		if (namelen != 2)
2135 			return (ENOTDIR);
2136 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2137 	case VFS_CONF:
2138 		if (namelen != 3)
2139 			return (ENOTDIR);	/* overloaded */
2140 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2141 			if (vfsp->vfc_typenum == name[2])
2142 				break;
2143 		if (vfsp == NULL)
2144 			return (EOPNOTSUPP);
2145 		return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2146 	}
2147 	return (EOPNOTSUPP);
2148 }
2149 
2150 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2151 	"Generic filesystem");
2152 
2153 #if 1 || defined(COMPAT_PRELITE2)
2154 
2155 static int
2156 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2157 {
2158 	int error;
2159 	struct vfsconf *vfsp;
2160 	struct ovfsconf ovfs;
2161 
2162 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2163 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
2164 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
2165 		ovfs.vfc_index = vfsp->vfc_typenum;
2166 		ovfs.vfc_refcount = vfsp->vfc_refcount;
2167 		ovfs.vfc_flags = vfsp->vfc_flags;
2168 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2169 		if (error)
2170 			return error;
2171 	}
2172 	return 0;
2173 }
2174 
2175 #endif /* 1 || COMPAT_PRELITE2 */
2176 
2177 #if COMPILING_LINT
2178 #define KINFO_VNODESLOP	10
2179 /*
2180  * Dump vnode list (via sysctl).
2181  * Copyout address of vnode followed by vnode.
2182  */
2183 /* ARGSUSED */
2184 static int
2185 sysctl_vnode(SYSCTL_HANDLER_ARGS)
2186 {
2187 	struct proc *p = curproc;	/* XXX */
2188 	struct mount *mp, *nmp;
2189 	struct vnode *nvp, *vp;
2190 	int error;
2191 
2192 #define VPTRSZ	sizeof (struct vnode *)
2193 #define VNODESZ	sizeof (struct vnode)
2194 
2195 	req->lock = 0;
2196 	if (!req->oldptr) /* Make an estimate */
2197 		return (SYSCTL_OUT(req, 0,
2198 			(numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2199 
2200 	mtx_enter(&mountlist_mtx, MTX_DEF);
2201 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2202 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, p)) {
2203 			nmp = TAILQ_NEXT(mp, mnt_list);
2204 			continue;
2205 		}
2206 again:
2207 		simple_lock(&mntvnode_slock);
2208 		for (vp = LIST_FIRST(&mp->mnt_vnodelist);
2209 		     vp != NULL;
2210 		     vp = nvp) {
2211 			/*
2212 			 * Check that the vp is still associated with
2213 			 * this filesystem.  RACE: could have been
2214 			 * recycled onto the same filesystem.
2215 			 */
2216 			if (vp->v_mount != mp) {
2217 				simple_unlock(&mntvnode_slock);
2218 				goto again;
2219 			}
2220 			nvp = LIST_NEXT(vp, v_mntvnodes);
2221 			simple_unlock(&mntvnode_slock);
2222 			if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2223 			    (error = SYSCTL_OUT(req, vp, VNODESZ)))
2224 				return (error);
2225 			simple_lock(&mntvnode_slock);
2226 		}
2227 		simple_unlock(&mntvnode_slock);
2228 		mtx_enter(&mountlist_mtx, MTX_DEF);
2229 		nmp = TAILQ_NEXT(mp, mnt_list);
2230 		vfs_unbusy(mp, p);
2231 	}
2232 	mtx_exit(&mountlist_mtx, MTX_DEF);
2233 
2234 	return (0);
2235 }
2236 
2237 /*
2238  * XXX
2239  * Exporting the vnode list on large systems causes them to crash.
2240  * Exporting the vnode list on medium systems causes sysctl to coredump.
2241  */
2242 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2243 	0, 0, sysctl_vnode, "S,vnode", "");
2244 #endif
2245 
2246 /*
2247  * Check to see if a filesystem is mounted on a block device.
2248  */
2249 int
2250 vfs_mountedon(vp)
2251 	struct vnode *vp;
2252 {
2253 
2254 	if (vp->v_rdev->si_mountpoint != NULL)
2255 		return (EBUSY);
2256 	return (0);
2257 }
2258 
2259 /*
2260  * Unmount all filesystems. The list is traversed in reverse order
2261  * of mounting to avoid dependencies.
2262  */
2263 void
2264 vfs_unmountall()
2265 {
2266 	struct mount *mp;
2267 	struct proc *p;
2268 	int error;
2269 
2270 	if (curproc != NULL)
2271 		p = curproc;
2272 	else
2273 		p = initproc;	/* XXX XXX should this be proc0? */
2274 	/*
2275 	 * Since this only runs when rebooting, it is not interlocked.
2276 	 */
2277 	while(!TAILQ_EMPTY(&mountlist)) {
2278 		mp = TAILQ_LAST(&mountlist, mntlist);
2279 		error = dounmount(mp, MNT_FORCE, p);
2280 		if (error) {
2281 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
2282 			printf("unmount of %s failed (",
2283 			    mp->mnt_stat.f_mntonname);
2284 			if (error == EBUSY)
2285 				printf("BUSY)\n");
2286 			else
2287 				printf("%d)\n", error);
2288 		} else {
2289 			/* The unmount has removed mp from the mountlist */
2290 		}
2291 	}
2292 }
2293 
2294 /*
2295  * Build hash lists of net addresses and hang them off the mount point.
2296  * Called by ufs_mount() to set up the lists of export addresses.
2297  */
2298 static int
2299 vfs_hang_addrlist(mp, nep, argp)
2300 	struct mount *mp;
2301 	struct netexport *nep;
2302 	struct export_args *argp;
2303 {
2304 	register struct netcred *np;
2305 	register struct radix_node_head *rnh;
2306 	register int i;
2307 	struct radix_node *rn;
2308 	struct sockaddr *saddr, *smask = 0;
2309 	struct domain *dom;
2310 	int error;
2311 
2312 	if (argp->ex_addrlen == 0) {
2313 		if (mp->mnt_flag & MNT_DEFEXPORTED)
2314 			return (EPERM);
2315 		np = &nep->ne_defexported;
2316 		np->netc_exflags = argp->ex_flags;
2317 		np->netc_anon = argp->ex_anon;
2318 		np->netc_anon.cr_ref = 1;
2319 		mp->mnt_flag |= MNT_DEFEXPORTED;
2320 		return (0);
2321 	}
2322 	i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
2323 	np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK | M_ZERO);
2324 	saddr = (struct sockaddr *) (np + 1);
2325 	if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
2326 		goto out;
2327 	if (saddr->sa_len > argp->ex_addrlen)
2328 		saddr->sa_len = argp->ex_addrlen;
2329 	if (argp->ex_masklen) {
2330 		smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen);
2331 		error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen);
2332 		if (error)
2333 			goto out;
2334 		if (smask->sa_len > argp->ex_masklen)
2335 			smask->sa_len = argp->ex_masklen;
2336 	}
2337 	i = saddr->sa_family;
2338 	if ((rnh = nep->ne_rtable[i]) == 0) {
2339 		/*
2340 		 * Seems silly to initialize every AF when most are not used,
2341 		 * do so on demand here
2342 		 */
2343 		for (dom = domains; dom; dom = dom->dom_next)
2344 			if (dom->dom_family == i && dom->dom_rtattach) {
2345 				dom->dom_rtattach((void **) &nep->ne_rtable[i],
2346 				    dom->dom_rtoffset);
2347 				break;
2348 			}
2349 		if ((rnh = nep->ne_rtable[i]) == 0) {
2350 			error = ENOBUFS;
2351 			goto out;
2352 		}
2353 	}
2354 	rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh,
2355 	    np->netc_rnodes);
2356 	if (rn == 0 || np != (struct netcred *) rn) {	/* already exists */
2357 		error = EPERM;
2358 		goto out;
2359 	}
2360 	np->netc_exflags = argp->ex_flags;
2361 	np->netc_anon = argp->ex_anon;
2362 	np->netc_anon.cr_ref = 1;
2363 	return (0);
2364 out:
2365 	free(np, M_NETADDR);
2366 	return (error);
2367 }
2368 
2369 /* Helper for vfs_free_addrlist. */
2370 /* ARGSUSED */
2371 static int
2372 vfs_free_netcred(rn, w)
2373 	struct radix_node *rn;
2374 	void *w;
2375 {
2376 	register struct radix_node_head *rnh = (struct radix_node_head *) w;
2377 
2378 	(*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
2379 	free((caddr_t) rn, M_NETADDR);
2380 	return (0);
2381 }
2382 
2383 /*
2384  * Free the net address hash lists that are hanging off the mount points.
2385  */
2386 static void
2387 vfs_free_addrlist(nep)
2388 	struct netexport *nep;
2389 {
2390 	register int i;
2391 	register struct radix_node_head *rnh;
2392 
2393 	for (i = 0; i <= AF_MAX; i++)
2394 		if ((rnh = nep->ne_rtable[i])) {
2395 			(*rnh->rnh_walktree) (rnh, vfs_free_netcred,
2396 			    (caddr_t) rnh);
2397 			free((caddr_t) rnh, M_RTABLE);
2398 			nep->ne_rtable[i] = 0;
2399 		}
2400 }
2401 
2402 /*
2403  * High level function to manipulate export options on a mount point
2404  * and the passed in netexport.
2405  * Struct export_args *argp is the variable used to twiddle options,
2406  * the structure is described in sys/mount.h
2407  */
2408 int
2409 vfs_export(mp, nep, argp)
2410 	struct mount *mp;
2411 	struct netexport *nep;
2412 	struct export_args *argp;
2413 {
2414 	int error;
2415 
2416 	if (argp->ex_flags & MNT_DELEXPORT) {
2417 		if (mp->mnt_flag & MNT_EXPUBLIC) {
2418 			vfs_setpublicfs(NULL, NULL, NULL);
2419 			mp->mnt_flag &= ~MNT_EXPUBLIC;
2420 		}
2421 		vfs_free_addrlist(nep);
2422 		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
2423 	}
2424 	if (argp->ex_flags & MNT_EXPORTED) {
2425 		if (argp->ex_flags & MNT_EXPUBLIC) {
2426 			if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
2427 				return (error);
2428 			mp->mnt_flag |= MNT_EXPUBLIC;
2429 		}
2430 		if ((error = vfs_hang_addrlist(mp, nep, argp)))
2431 			return (error);
2432 		mp->mnt_flag |= MNT_EXPORTED;
2433 	}
2434 	return (0);
2435 }
2436 
2437 /*
2438  * Set the publicly exported filesystem (WebNFS). Currently, only
2439  * one public filesystem is possible in the spec (RFC 2054 and 2055)
2440  */
2441 int
2442 vfs_setpublicfs(mp, nep, argp)
2443 	struct mount *mp;
2444 	struct netexport *nep;
2445 	struct export_args *argp;
2446 {
2447 	int error;
2448 	struct vnode *rvp;
2449 	char *cp;
2450 
2451 	/*
2452 	 * mp == NULL -> invalidate the current info, the FS is
2453 	 * no longer exported. May be called from either vfs_export
2454 	 * or unmount, so check if it hasn't already been done.
2455 	 */
2456 	if (mp == NULL) {
2457 		if (nfs_pub.np_valid) {
2458 			nfs_pub.np_valid = 0;
2459 			if (nfs_pub.np_index != NULL) {
2460 				FREE(nfs_pub.np_index, M_TEMP);
2461 				nfs_pub.np_index = NULL;
2462 			}
2463 		}
2464 		return (0);
2465 	}
2466 
2467 	/*
2468 	 * Only one allowed at a time.
2469 	 */
2470 	if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
2471 		return (EBUSY);
2472 
2473 	/*
2474 	 * Get real filehandle for root of exported FS.
2475 	 */
2476 	bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
2477 	nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
2478 
2479 	if ((error = VFS_ROOT(mp, &rvp)))
2480 		return (error);
2481 
2482 	if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
2483 		return (error);
2484 
2485 	vput(rvp);
2486 
2487 	/*
2488 	 * If an indexfile was specified, pull it in.
2489 	 */
2490 	if (argp->ex_indexfile != NULL) {
2491 		MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP,
2492 		    M_WAITOK);
2493 		error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
2494 		    MAXNAMLEN, (size_t *)0);
2495 		if (!error) {
2496 			/*
2497 			 * Check for illegal filenames.
2498 			 */
2499 			for (cp = nfs_pub.np_index; *cp; cp++) {
2500 				if (*cp == '/') {
2501 					error = EINVAL;
2502 					break;
2503 				}
2504 			}
2505 		}
2506 		if (error) {
2507 			FREE(nfs_pub.np_index, M_TEMP);
2508 			return (error);
2509 		}
2510 	}
2511 
2512 	nfs_pub.np_mount = mp;
2513 	nfs_pub.np_valid = 1;
2514 	return (0);
2515 }
2516 
2517 /*
2518  * Used by the filesystems to determine if a given network address
2519  * (passed in 'nam') is present in thier exports list, returns a pointer
2520  * to struct netcred so that the filesystem can examine it for
2521  * access rights (read/write/etc).
2522  */
2523 struct netcred *
2524 vfs_export_lookup(mp, nep, nam)
2525 	register struct mount *mp;
2526 	struct netexport *nep;
2527 	struct sockaddr *nam;
2528 {
2529 	register struct netcred *np;
2530 	register struct radix_node_head *rnh;
2531 	struct sockaddr *saddr;
2532 
2533 	np = NULL;
2534 	if (mp->mnt_flag & MNT_EXPORTED) {
2535 		/*
2536 		 * Lookup in the export list first.
2537 		 */
2538 		if (nam != NULL) {
2539 			saddr = nam;
2540 			rnh = nep->ne_rtable[saddr->sa_family];
2541 			if (rnh != NULL) {
2542 				np = (struct netcred *)
2543 					(*rnh->rnh_matchaddr)((caddr_t)saddr,
2544 							      rnh);
2545 				if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
2546 					np = NULL;
2547 			}
2548 		}
2549 		/*
2550 		 * If no address match, use the default if it exists.
2551 		 */
2552 		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
2553 			np = &nep->ne_defexported;
2554 	}
2555 	return (np);
2556 }
2557 
2558 /*
2559  * perform msync on all vnodes under a mount point
2560  * the mount point must be locked.
2561  */
2562 void
2563 vfs_msync(struct mount *mp, int flags) {
2564 	struct vnode *vp, *nvp;
2565 	struct vm_object *obj;
2566 	int anyio, tries;
2567 
2568 	tries = 5;
2569 loop:
2570 	anyio = 0;
2571 	for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp != NULL; vp = nvp) {
2572 
2573 		nvp = LIST_NEXT(vp, v_mntvnodes);
2574 
2575 		if (vp->v_mount != mp) {
2576 			goto loop;
2577 		}
2578 
2579 		if (vp->v_flag & VXLOCK)	/* XXX: what if MNT_WAIT? */
2580 			continue;
2581 
2582 		if (flags != MNT_WAIT) {
2583 			if (VOP_GETVOBJECT(vp, &obj) != 0 ||
2584 			    (obj->flags & OBJ_MIGHTBEDIRTY) == 0)
2585 				continue;
2586 			if (VOP_ISLOCKED(vp, NULL))
2587 				continue;
2588 		}
2589 
2590 		mtx_enter(&vp->v_interlock, MTX_DEF);
2591 		if (VOP_GETVOBJECT(vp, &obj) == 0 &&
2592 		    (obj->flags & OBJ_MIGHTBEDIRTY)) {
2593 			if (!vget(vp,
2594 				LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, curproc)) {
2595 				if (VOP_GETVOBJECT(vp, &obj) == 0) {
2596 					vm_object_page_clean(obj, 0, 0, flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
2597 					anyio = 1;
2598 				}
2599 				vput(vp);
2600 			}
2601 		} else {
2602 			mtx_exit(&vp->v_interlock, MTX_DEF);
2603 		}
2604 	}
2605 	if (anyio && (--tries > 0))
2606 		goto loop;
2607 }
2608 
2609 /*
2610  * Create the VM object needed for VMIO and mmap support.  This
2611  * is done for all VREG files in the system.  Some filesystems might
2612  * afford the additional metadata buffering capability of the
2613  * VMIO code by making the device node be VMIO mode also.
2614  *
2615  * vp must be locked when vfs_object_create is called.
2616  */
2617 int
2618 vfs_object_create(vp, p, cred)
2619 	struct vnode *vp;
2620 	struct proc *p;
2621 	struct ucred *cred;
2622 {
2623 	return (VOP_CREATEVOBJECT(vp, cred, p));
2624 }
2625 
2626 /*
2627  * Mark a vnode as free, putting it up for recycling.
2628  */
2629 void
2630 vfree(vp)
2631 	struct vnode *vp;
2632 {
2633 	int s;
2634 
2635 	s = splbio();
2636 	simple_lock(&vnode_free_list_slock);
2637 	KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free"));
2638 	if (vp->v_flag & VAGE) {
2639 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2640 	} else {
2641 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2642 	}
2643 	freevnodes++;
2644 	simple_unlock(&vnode_free_list_slock);
2645 	vp->v_flag &= ~VAGE;
2646 	vp->v_flag |= VFREE;
2647 	splx(s);
2648 }
2649 
2650 /*
2651  * Opposite of vfree() - mark a vnode as in use.
2652  */
2653 void
2654 vbusy(vp)
2655 	struct vnode *vp;
2656 {
2657 	int s;
2658 
2659 	s = splbio();
2660 	simple_lock(&vnode_free_list_slock);
2661 	KASSERT((vp->v_flag & VFREE) != 0, ("vnode not free"));
2662 	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2663 	freevnodes--;
2664 	simple_unlock(&vnode_free_list_slock);
2665 	vp->v_flag &= ~(VFREE|VAGE);
2666 	splx(s);
2667 }
2668 
2669 /*
2670  * Record a process's interest in events which might happen to
2671  * a vnode.  Because poll uses the historic select-style interface
2672  * internally, this routine serves as both the ``check for any
2673  * pending events'' and the ``record my interest in future events''
2674  * functions.  (These are done together, while the lock is held,
2675  * to avoid race conditions.)
2676  */
2677 int
2678 vn_pollrecord(vp, p, events)
2679 	struct vnode *vp;
2680 	struct proc *p;
2681 	short events;
2682 {
2683 	simple_lock(&vp->v_pollinfo.vpi_lock);
2684 	if (vp->v_pollinfo.vpi_revents & events) {
2685 		/*
2686 		 * This leaves events we are not interested
2687 		 * in available for the other process which
2688 		 * which presumably had requested them
2689 		 * (otherwise they would never have been
2690 		 * recorded).
2691 		 */
2692 		events &= vp->v_pollinfo.vpi_revents;
2693 		vp->v_pollinfo.vpi_revents &= ~events;
2694 
2695 		simple_unlock(&vp->v_pollinfo.vpi_lock);
2696 		return events;
2697 	}
2698 	vp->v_pollinfo.vpi_events |= events;
2699 	selrecord(p, &vp->v_pollinfo.vpi_selinfo);
2700 	simple_unlock(&vp->v_pollinfo.vpi_lock);
2701 	return 0;
2702 }
2703 
2704 /*
2705  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
2706  * it is possible for us to miss an event due to race conditions, but
2707  * that condition is expected to be rare, so for the moment it is the
2708  * preferred interface.
2709  */
2710 void
2711 vn_pollevent(vp, events)
2712 	struct vnode *vp;
2713 	short events;
2714 {
2715 	simple_lock(&vp->v_pollinfo.vpi_lock);
2716 	if (vp->v_pollinfo.vpi_events & events) {
2717 		/*
2718 		 * We clear vpi_events so that we don't
2719 		 * call selwakeup() twice if two events are
2720 		 * posted before the polling process(es) is
2721 		 * awakened.  This also ensures that we take at
2722 		 * most one selwakeup() if the polling process
2723 		 * is no longer interested.  However, it does
2724 		 * mean that only one event can be noticed at
2725 		 * a time.  (Perhaps we should only clear those
2726 		 * event bits which we note?) XXX
2727 		 */
2728 		vp->v_pollinfo.vpi_events = 0;	/* &= ~events ??? */
2729 		vp->v_pollinfo.vpi_revents |= events;
2730 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2731 	}
2732 	simple_unlock(&vp->v_pollinfo.vpi_lock);
2733 }
2734 
2735 /*
2736  * Wake up anyone polling on vp because it is being revoked.
2737  * This depends on dead_poll() returning POLLHUP for correct
2738  * behavior.
2739  */
2740 void
2741 vn_pollgone(vp)
2742 	struct vnode *vp;
2743 {
2744 	simple_lock(&vp->v_pollinfo.vpi_lock);
2745 	if (vp->v_pollinfo.vpi_events) {
2746 		vp->v_pollinfo.vpi_events = 0;
2747 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2748 	}
2749 	simple_unlock(&vp->v_pollinfo.vpi_lock);
2750 }
2751 
2752 
2753 
2754 /*
2755  * Routine to create and manage a filesystem syncer vnode.
2756  */
2757 #define sync_close ((int (*) __P((struct  vop_close_args *)))nullop)
2758 static int	sync_fsync __P((struct  vop_fsync_args *));
2759 static int	sync_inactive __P((struct  vop_inactive_args *));
2760 static int	sync_reclaim  __P((struct  vop_reclaim_args *));
2761 #define sync_lock ((int (*) __P((struct  vop_lock_args *)))vop_nolock)
2762 #define sync_unlock ((int (*) __P((struct  vop_unlock_args *)))vop_nounlock)
2763 static int	sync_print __P((struct vop_print_args *));
2764 #define sync_islocked ((int(*) __P((struct vop_islocked_args *)))vop_noislocked)
2765 
2766 static vop_t **sync_vnodeop_p;
2767 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
2768 	{ &vop_default_desc,	(vop_t *) vop_eopnotsupp },
2769 	{ &vop_close_desc,	(vop_t *) sync_close },		/* close */
2770 	{ &vop_fsync_desc,	(vop_t *) sync_fsync },		/* fsync */
2771 	{ &vop_inactive_desc,	(vop_t *) sync_inactive },	/* inactive */
2772 	{ &vop_reclaim_desc,	(vop_t *) sync_reclaim },	/* reclaim */
2773 	{ &vop_lock_desc,	(vop_t *) sync_lock },		/* lock */
2774 	{ &vop_unlock_desc,	(vop_t *) sync_unlock },	/* unlock */
2775 	{ &vop_print_desc,	(vop_t *) sync_print },		/* print */
2776 	{ &vop_islocked_desc,	(vop_t *) sync_islocked },	/* islocked */
2777 	{ NULL, NULL }
2778 };
2779 static struct vnodeopv_desc sync_vnodeop_opv_desc =
2780 	{ &sync_vnodeop_p, sync_vnodeop_entries };
2781 
2782 VNODEOP_SET(sync_vnodeop_opv_desc);
2783 
2784 /*
2785  * Create a new filesystem syncer vnode for the specified mount point.
2786  */
2787 int
2788 vfs_allocate_syncvnode(mp)
2789 	struct mount *mp;
2790 {
2791 	struct vnode *vp;
2792 	static long start, incr, next;
2793 	int error;
2794 
2795 	/* Allocate a new vnode */
2796 	if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
2797 		mp->mnt_syncer = NULL;
2798 		return (error);
2799 	}
2800 	vp->v_type = VNON;
2801 	/*
2802 	 * Place the vnode onto the syncer worklist. We attempt to
2803 	 * scatter them about on the list so that they will go off
2804 	 * at evenly distributed times even if all the filesystems
2805 	 * are mounted at once.
2806 	 */
2807 	next += incr;
2808 	if (next == 0 || next > syncer_maxdelay) {
2809 		start /= 2;
2810 		incr /= 2;
2811 		if (start == 0) {
2812 			start = syncer_maxdelay / 2;
2813 			incr = syncer_maxdelay;
2814 		}
2815 		next = start;
2816 	}
2817 	vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
2818 	mp->mnt_syncer = vp;
2819 	return (0);
2820 }
2821 
2822 /*
2823  * Do a lazy sync of the filesystem.
2824  */
2825 static int
2826 sync_fsync(ap)
2827 	struct vop_fsync_args /* {
2828 		struct vnode *a_vp;
2829 		struct ucred *a_cred;
2830 		int a_waitfor;
2831 		struct proc *a_p;
2832 	} */ *ap;
2833 {
2834 	struct vnode *syncvp = ap->a_vp;
2835 	struct mount *mp = syncvp->v_mount;
2836 	struct proc *p = ap->a_p;
2837 	int asyncflag;
2838 
2839 	/*
2840 	 * We only need to do something if this is a lazy evaluation.
2841 	 */
2842 	if (ap->a_waitfor != MNT_LAZY)
2843 		return (0);
2844 
2845 	/*
2846 	 * Move ourselves to the back of the sync list.
2847 	 */
2848 	vn_syncer_add_to_worklist(syncvp, syncdelay);
2849 
2850 	/*
2851 	 * Walk the list of vnodes pushing all that are dirty and
2852 	 * not already on the sync list.
2853 	 */
2854 	mtx_enter(&mountlist_mtx, MTX_DEF);
2855 	if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, p) != 0) {
2856 		mtx_exit(&mountlist_mtx, MTX_DEF);
2857 		return (0);
2858 	}
2859 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
2860 		vfs_unbusy(mp, p);
2861 		return (0);
2862 	}
2863 	asyncflag = mp->mnt_flag & MNT_ASYNC;
2864 	mp->mnt_flag &= ~MNT_ASYNC;
2865 	vfs_msync(mp, MNT_NOWAIT);
2866 	VFS_SYNC(mp, MNT_LAZY, ap->a_cred, p);
2867 	if (asyncflag)
2868 		mp->mnt_flag |= MNT_ASYNC;
2869 	vn_finished_write(mp);
2870 	vfs_unbusy(mp, p);
2871 	return (0);
2872 }
2873 
2874 /*
2875  * The syncer vnode is no referenced.
2876  */
2877 static int
2878 sync_inactive(ap)
2879 	struct vop_inactive_args /* {
2880 		struct vnode *a_vp;
2881 		struct proc *a_p;
2882 	} */ *ap;
2883 {
2884 
2885 	vgone(ap->a_vp);
2886 	return (0);
2887 }
2888 
2889 /*
2890  * The syncer vnode is no longer needed and is being decommissioned.
2891  *
2892  * Modifications to the worklist must be protected at splbio().
2893  */
2894 static int
2895 sync_reclaim(ap)
2896 	struct vop_reclaim_args /* {
2897 		struct vnode *a_vp;
2898 	} */ *ap;
2899 {
2900 	struct vnode *vp = ap->a_vp;
2901 	int s;
2902 
2903 	s = splbio();
2904 	vp->v_mount->mnt_syncer = NULL;
2905 	if (vp->v_flag & VONWORKLST) {
2906 		LIST_REMOVE(vp, v_synclist);
2907 		vp->v_flag &= ~VONWORKLST;
2908 	}
2909 	splx(s);
2910 
2911 	return (0);
2912 }
2913 
2914 /*
2915  * Print out a syncer vnode.
2916  */
2917 static int
2918 sync_print(ap)
2919 	struct vop_print_args /* {
2920 		struct vnode *a_vp;
2921 	} */ *ap;
2922 {
2923 	struct vnode *vp = ap->a_vp;
2924 
2925 	printf("syncer vnode");
2926 	if (vp->v_vnlock != NULL)
2927 		lockmgr_printinfo(vp->v_vnlock);
2928 	printf("\n");
2929 	return (0);
2930 }
2931 
2932 /*
2933  * extract the dev_t from a VCHR
2934  */
2935 dev_t
2936 vn_todev(vp)
2937 	struct vnode *vp;
2938 {
2939 	if (vp->v_type != VCHR)
2940 		return (NODEV);
2941 	return (vp->v_rdev);
2942 }
2943 
2944 /*
2945  * Check if vnode represents a disk device
2946  */
2947 int
2948 vn_isdisk(vp, errp)
2949 	struct vnode *vp;
2950 	int *errp;
2951 {
2952 	struct cdevsw *cdevsw;
2953 
2954 	if (vp->v_type != VCHR) {
2955 		if (errp != NULL)
2956 			*errp = ENOTBLK;
2957 		return (0);
2958 	}
2959 	if (vp->v_rdev == NULL) {
2960 		if (errp != NULL)
2961 			*errp = ENXIO;
2962 		return (0);
2963 	}
2964 	cdevsw = devsw(vp->v_rdev);
2965 	if (cdevsw == NULL) {
2966 		if (errp != NULL)
2967 			*errp = ENXIO;
2968 		return (0);
2969 	}
2970 	if (!(cdevsw->d_flags & D_DISK)) {
2971 		if (errp != NULL)
2972 			*errp = ENOTBLK;
2973 		return (0);
2974 	}
2975 	if (errp != NULL)
2976 		*errp = 0;
2977 	return (1);
2978 }
2979 
2980 /*
2981  * Free data allocated by namei(); see namei(9) for details.
2982  */
2983 void
2984 NDFREE(ndp, flags)
2985      struct nameidata *ndp;
2986      const uint flags;
2987 {
2988 	if (!(flags & NDF_NO_FREE_PNBUF) &&
2989 	    (ndp->ni_cnd.cn_flags & HASBUF)) {
2990 		zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
2991 		ndp->ni_cnd.cn_flags &= ~HASBUF;
2992 	}
2993 	if (!(flags & NDF_NO_DVP_UNLOCK) &&
2994 	    (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
2995 	    ndp->ni_dvp != ndp->ni_vp)
2996 		VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_proc);
2997 	if (!(flags & NDF_NO_DVP_RELE) &&
2998 	    (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
2999 		vrele(ndp->ni_dvp);
3000 		ndp->ni_dvp = NULL;
3001 	}
3002 	if (!(flags & NDF_NO_VP_UNLOCK) &&
3003 	    (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
3004 		VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_proc);
3005 	if (!(flags & NDF_NO_VP_RELE) &&
3006 	    ndp->ni_vp) {
3007 		vrele(ndp->ni_vp);
3008 		ndp->ni_vp = NULL;
3009 	}
3010 	if (!(flags & NDF_NO_STARTDIR_RELE) &&
3011 	    (ndp->ni_cnd.cn_flags & SAVESTART)) {
3012 		vrele(ndp->ni_startdir);
3013 		ndp->ni_startdir = NULL;
3014 	}
3015 }
3016 
3017 /*
3018  * Common file system object access control check routine.  Accepts a
3019  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3020  * and optional call-by-reference privused argument allowing vaccess()
3021  * to indicate to the caller whether privilege was used to satisfy the
3022  * request.  Returns 0 on success, or an errno on failure.
3023  */
3024 int
3025 vaccess(type, file_mode, file_uid, file_gid, acc_mode, cred, privused)
3026 	enum vtype type;
3027 	mode_t file_mode;
3028 	uid_t file_uid;
3029 	gid_t file_gid;
3030 	mode_t acc_mode;
3031 	struct ucred *cred;
3032 	int *privused;
3033 {
3034 	mode_t dac_granted;
3035 #ifdef CAPABILITIES
3036 	mode_t cap_granted;
3037 #endif
3038 
3039 	/*
3040 	 * Look for a normal, non-privileged way to access the file/directory
3041 	 * as requested.  If it exists, go with that.
3042 	 */
3043 
3044 	if (privused != NULL)
3045 		*privused = 0;
3046 
3047 	dac_granted = 0;
3048 
3049 	/* Check the owner. */
3050 	if (cred->cr_uid == file_uid) {
3051 		dac_granted |= VADMIN;
3052 		if (file_mode & S_IXUSR)
3053 			dac_granted |= VEXEC;
3054 		if (file_mode & S_IRUSR)
3055 			dac_granted |= VREAD;
3056 		if (file_mode & S_IWUSR)
3057 			dac_granted |= VWRITE;
3058 
3059 		if ((acc_mode & dac_granted) == acc_mode)
3060 			return (0);
3061 
3062 		goto privcheck;
3063 	}
3064 
3065 	/* Otherwise, check the groups (first match) */
3066 	if (groupmember(file_gid, cred)) {
3067 		if (file_mode & S_IXGRP)
3068 			dac_granted |= VEXEC;
3069 		if (file_mode & S_IRGRP)
3070 			dac_granted |= VREAD;
3071 		if (file_mode & S_IWGRP)
3072 			dac_granted |= VWRITE;
3073 
3074 		if ((acc_mode & dac_granted) == acc_mode)
3075 			return (0);
3076 
3077 		goto privcheck;
3078 	}
3079 
3080 	/* Otherwise, check everyone else. */
3081 	if (file_mode & S_IXOTH)
3082 		dac_granted |= VEXEC;
3083 	if (file_mode & S_IROTH)
3084 		dac_granted |= VREAD;
3085 	if (file_mode & S_IWOTH)
3086 		dac_granted |= VWRITE;
3087 	if ((acc_mode & dac_granted) == acc_mode)
3088 		return (0);
3089 
3090 privcheck:
3091 	if (!suser_xxx(cred, NULL, PRISON_ROOT)) {
3092 		/* XXX audit: privilege used */
3093 		if (privused != NULL)
3094 			*privused = 1;
3095 		return (0);
3096 	}
3097 
3098 #ifdef CAPABILITIES
3099 	/*
3100 	 * Build a capability mask to determine if the set of capabilities
3101 	 * satisfies the requirements when combined with the granted mask
3102 	 * from above.
3103 	 * For each capability, if the capability is required, bitwise
3104 	 * or the request type onto the cap_granted mask.
3105 	 */
3106 	cap_granted = 0;
3107 	if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3108 	    !cap_check_xxx(cred, NULL, CAP_DAC_EXECUTE, PRISON_ROOT))
3109 	    cap_granted |= VEXEC;
3110 
3111 	if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
3112 	    !cap_check_xxx(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3113 		cap_granted |= VREAD;
3114 
3115 	if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3116 	    !cap_check_xxx(cred, NULL, CAP_DAC_WRITE, PRISON_ROOT))
3117 		cap_granted |= VWRITE;
3118 
3119 	if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3120 	    !cap_check_xxx(cred, NULL, CAP_FOWNER, PRISON_ROOT))
3121 		cap_granted |= VADMIN;
3122 
3123 	if ((acc_mode & (cap_granted | dac_granted)) == acc_mode) {
3124 		/* XXX audit: privilege used */
3125 		if (privused != NULL)
3126 			*privused = 1;
3127 		return (0);
3128 	}
3129 #endif
3130 
3131 	return (EACCES);
3132 }
3133