vfs_subr.c revision 105077 
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: head/sys/kern/vfs_subr.c 105077 2002-10-14 03:20:36Z mckusick $
40 */
41
42/*
43 * External virtual filesystem routines
44 */
45#include "opt_ddb.h"
46#include "opt_mac.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/eventhandler.h>
54#include <sys/extattr.h>
55#include <sys/fcntl.h>
56#include <sys/kernel.h>
57#include <sys/kthread.h>
58#include <sys/mac.h>
59#include <sys/malloc.h>
60#include <sys/mount.h>
61#include <sys/namei.h>
62#include <sys/stat.h>
63#include <sys/sysctl.h>
64#include <sys/syslog.h>
65#include <sys/vmmeter.h>
66#include <sys/vnode.h>
67
68#include <vm/vm.h>
69#include <vm/vm_object.h>
70#include <vm/vm_extern.h>
71#include <vm/pmap.h>
72#include <vm/vm_map.h>
73#include <vm/vm_page.h>
74#include <vm/uma.h>
75
76static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
77
78static void	addalias(struct vnode *vp, dev_t nvp_rdev);
79static void	insmntque(struct vnode *vp, struct mount *mp);
80static void	vclean(struct vnode *vp, int flags, struct thread *td);
81static void	vlruvp(struct vnode *vp);
82static int	flushbuflist(struct buf *blist, int flags, struct vnode *vp,
83		    int slpflag, int slptimeo, int *errorp);
84static int	vcanrecycle(struct vnode *vp, struct mount **vnmpp);
85
86
87/*
88 * Number of vnodes in existence.  Increased whenever getnewvnode()
89 * allocates a new vnode, never decreased.
90 */
91static unsigned long	numvnodes;
92
93SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
94
95/*
96 * Conversion tables for conversion from vnode types to inode formats
97 * and back.
98 */
99enum vtype iftovt_tab[16] = {
100	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
101	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
102};
103int vttoif_tab[9] = {
104	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
105	S_IFSOCK, S_IFIFO, S_IFMT,
106};
107
108/*
109 * List of vnodes that are ready for recycling.
110 */
111static TAILQ_HEAD(freelst, vnode) vnode_free_list;
112
113/*
114 * Minimum number of free vnodes.  If there are fewer than this free vnodes,
115 * getnewvnode() will return a newly allocated vnode.
116 */
117static u_long wantfreevnodes = 25;
118SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
119/* Number of vnodes in the free list. */
120static u_long freevnodes;
121SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
122
123/*
124 * Various variables used for debugging the new implementation of
125 * reassignbuf().
126 * XXX these are probably of (very) limited utility now.
127 */
128static int reassignbufcalls;
129SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
130static int nameileafonly;
131SYSCTL_INT(_vfs, OID_AUTO, nameileafonly, CTLFLAG_RW, &nameileafonly, 0, "");
132
133#ifdef ENABLE_VFS_IOOPT
134/* See NOTES for a description of this setting. */
135int vfs_ioopt;
136SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, "");
137#endif
138
139/*
140 * Cache for the mount type id assigned to NFS.  This is used for
141 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
142 */
143int	nfs_mount_type = -1;
144
145/* To keep more than one thread at a time from running vfs_getnewfsid */
146static struct mtx mntid_mtx;
147
148/*
149 * Lock for any access to the following:
150 *	vnode_free_list
151 *	numvnodes
152 *	freevnodes
153 */
154static struct mtx vnode_free_list_mtx;
155
156/*
157 * For any iteration/modification of dev->si_hlist (linked through
158 * v_specnext)
159 */
160static struct mtx spechash_mtx;
161
162/* Publicly exported FS */
163struct nfs_public nfs_pub;
164
165/* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
166static uma_zone_t vnode_zone;
167static uma_zone_t vnodepoll_zone;
168
169/* Set to 1 to print out reclaim of active vnodes */
170int	prtactive;
171
172/*
173 * The workitem queue.
174 *
175 * It is useful to delay writes of file data and filesystem metadata
176 * for tens of seconds so that quickly created and deleted files need
177 * not waste disk bandwidth being created and removed. To realize this,
178 * we append vnodes to a "workitem" queue. When running with a soft
179 * updates implementation, most pending metadata dependencies should
180 * not wait for more than a few seconds. Thus, mounted on block devices
181 * are delayed only about a half the time that file data is delayed.
182 * Similarly, directory updates are more critical, so are only delayed
183 * about a third the time that file data is delayed. Thus, there are
184 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
185 * one each second (driven off the filesystem syncer process). The
186 * syncer_delayno variable indicates the next queue that is to be processed.
187 * Items that need to be processed soon are placed in this queue:
188 *
189 *	syncer_workitem_pending[syncer_delayno]
190 *
191 * A delay of fifteen seconds is done by placing the request fifteen
192 * entries later in the queue:
193 *
194 *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
195 *
196 */
197static int syncer_delayno;
198static long syncer_mask;
199LIST_HEAD(synclist, vnode);
200static struct synclist *syncer_workitem_pending;
201/*
202 * The sync_mtx protects:
203 *	vp->v_synclist
204 *	syncer_delayno
205 *	syncer_workitem_pending
206 *	rushjob
207 */
208static struct mtx sync_mtx;
209
210#define SYNCER_MAXDELAY		32
211static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
212static int syncdelay = 30;		/* max time to delay syncing data */
213static int filedelay = 30;		/* time to delay syncing files */
214SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
215static int dirdelay = 29;		/* time to delay syncing directories */
216SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
217static int metadelay = 28;		/* time to delay syncing metadata */
218SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
219static int rushjob;		/* number of slots to run ASAP */
220static int stat_rush_requests;	/* number of times I/O speeded up */
221SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
222
223/*
224 * Number of vnodes we want to exist at any one time.  This is mostly used
225 * to size hash tables in vnode-related code.  It is normally not used in
226 * getnewvnode(), as wantfreevnodes is normally nonzero.)
227 *
228 * XXX desiredvnodes is historical cruft and should not exist.
229 */
230int desiredvnodes;
231SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
232    &desiredvnodes, 0, "Maximum number of vnodes");
233static int minvnodes;
234SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
235    &minvnodes, 0, "Minimum number of vnodes");
236static int vnlru_nowhere;
237SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0,
238    "Number of times the vnlru process ran without success");
239
240/* Hook for calling soft updates */
241int (*softdep_process_worklist_hook)(struct mount *);
242
243/*
244 * This only exists to supress warnings from unlocked specfs accesses.  It is
245 * no longer ok to have an unlocked VFS.
246 */
247#define IGNORE_LOCK(vp) ((vp)->v_type == VCHR || (vp)->v_type == VBAD)
248
249/* Print lock violations */
250int vfs_badlock_print = 1;
251
252/* Panic on violation */
253int vfs_badlock_panic = 1;
254
255/* Check for interlock across VOPs */
256int vfs_badlock_mutex = 1;
257
258static void
259vfs_badlock(char *msg, char *str, struct vnode *vp)
260{
261	if (vfs_badlock_print)
262		printf("%s: %p %s\n", str, vp, msg);
263	if (vfs_badlock_panic)
264		Debugger("Lock violation.\n");
265}
266
267void
268assert_vi_unlocked(struct vnode *vp, char *str)
269{
270	if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
271		vfs_badlock("interlock is locked but should not be", str, vp);
272}
273
274void
275assert_vi_locked(struct vnode *vp, char *str)
276{
277	if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
278		vfs_badlock("interlock is not locked but should be", str, vp);
279}
280
281void
282assert_vop_locked(struct vnode *vp, char *str)
283{
284	if (vp && !IGNORE_LOCK(vp) && !VOP_ISLOCKED(vp, NULL))
285		vfs_badlock("is not locked but should be", str, vp);
286}
287
288void
289assert_vop_unlocked(struct vnode *vp, char *str)
290{
291	if (vp && !IGNORE_LOCK(vp) &&
292	    VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE)
293		vfs_badlock("is locked but should not be", str, vp);
294}
295
296void
297assert_vop_elocked(struct vnode *vp, char *str)
298{
299	if (vp && !IGNORE_LOCK(vp) &&
300	    VOP_ISLOCKED(vp, curthread) != LK_EXCLUSIVE)
301		vfs_badlock("is not exclusive locked but should be", str, vp);
302}
303
304void
305assert_vop_elocked_other(struct vnode *vp, char *str)
306{
307	if (vp && !IGNORE_LOCK(vp) &&
308	    VOP_ISLOCKED(vp, curthread) != LK_EXCLOTHER)
309		vfs_badlock("is not exclusive locked by another thread",
310		    str, vp);
311}
312
313void
314assert_vop_slocked(struct vnode *vp, char *str)
315{
316	if (vp && !IGNORE_LOCK(vp) &&
317	    VOP_ISLOCKED(vp, curthread) != LK_SHARED)
318		vfs_badlock("is not locked shared but should be", str, vp);
319}
320
321void
322vop_rename_pre(void *ap)
323{
324	struct vop_rename_args *a = ap;
325
326	if (a->a_tvp)
327		ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
328	ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
329	ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
330	ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
331
332	/* Check the source (from) */
333	if (a->a_tdvp != a->a_fdvp)
334		ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked.\n");
335	if (a->a_tvp != a->a_fvp)
336		ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: tvp locked.\n");
337
338	/* Check the target */
339	if (a->a_tvp)
340		ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked.\n");
341
342	ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked.\n");
343}
344
345void
346vop_strategy_pre(void *ap)
347{
348	struct vop_strategy_args *a = ap;
349	struct buf *bp;
350
351	bp = a->a_bp;
352
353	/*
354	 * Cluster ops lock their component buffers but not the IO container.
355	 */
356	if ((bp->b_flags & B_CLUSTER) != 0)
357		return;
358
359	if (BUF_REFCNT(bp) < 1) {
360		if (vfs_badlock_print)
361			printf("VOP_STRATEGY: bp is not locked but should be.\n");
362		if (vfs_badlock_panic)
363			Debugger("Lock violation.\n");
364	}
365}
366
367void
368vop_lookup_pre(void *ap)
369{
370	struct vop_lookup_args *a = ap;
371	struct vnode *dvp;
372
373	dvp = a->a_dvp;
374
375	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
376	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
377}
378
379void
380vop_lookup_post(void *ap, int rc)
381{
382	struct vop_lookup_args *a = ap;
383	struct componentname *cnp;
384	struct vnode *dvp;
385	struct vnode *vp;
386	int flags;
387
388	dvp = a->a_dvp;
389	cnp = a->a_cnp;
390	vp = *(a->a_vpp);
391	flags = cnp->cn_flags;
392
393
394	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
395	/*
396	 * If this is the last path component for this lookup and LOCPARENT
397	 * is set, OR if there is an error the directory has to be locked.
398	 */
399	if ((flags & LOCKPARENT) && (flags & ISLASTCN))
400		ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP (LOCKPARENT)");
401	else if (rc != 0)
402		ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP (error)");
403	else if (dvp != vp)
404		ASSERT_VOP_UNLOCKED(dvp, "VOP_LOOKUP (dvp)");
405
406	if (flags & PDIRUNLOCK)
407		ASSERT_VOP_UNLOCKED(dvp, "VOP_LOOKUP (PDIRUNLOCK)");
408}
409
410void
411vop_unlock_pre(void *ap)
412{
413	struct vop_unlock_args *a = ap;
414
415	if (a->a_flags & LK_INTERLOCK)
416		ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
417
418	ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
419}
420
421void
422vop_unlock_post(void *ap, int rc)
423{
424	struct vop_unlock_args *a = ap;
425
426	if (a->a_flags & LK_INTERLOCK)
427		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
428}
429
430void
431vop_lock_pre(void *ap)
432{
433	struct vop_lock_args *a = ap;
434
435	if ((a->a_flags & LK_INTERLOCK) == 0)
436		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
437	else
438		ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
439}
440
441void
442vop_lock_post(void *ap, int rc)
443{
444	struct vop_lock_args *a;
445
446	a = ap;
447
448	ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
449	ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
450}
451
452void
453v_addpollinfo(struct vnode *vp)
454{
455	vp->v_pollinfo = uma_zalloc(vnodepoll_zone, M_WAITOK);
456	mtx_init(&vp->v_pollinfo->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
457}
458
459/*
460 * Initialize the vnode management data structures.
461 */
462static void
463vntblinit(void *dummy __unused)
464{
465
466	desiredvnodes = maxproc + cnt.v_page_count / 4;
467	minvnodes = desiredvnodes / 4;
468	mtx_init(&mountlist_mtx, "mountlist", NULL, MTX_DEF);
469	mtx_init(&mntvnode_mtx, "mntvnode", NULL, MTX_DEF);
470	mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
471	mtx_init(&spechash_mtx, "spechash", NULL, MTX_DEF);
472	TAILQ_INIT(&vnode_free_list);
473	mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
474	vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
475	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
476	vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
477	      NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
478	/*
479	 * Initialize the filesystem syncer.
480	 */
481	syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
482		&syncer_mask);
483	syncer_maxdelay = syncer_mask + 1;
484	mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
485}
486SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL)
487
488
489/*
490 * Mark a mount point as busy. Used to synchronize access and to delay
491 * unmounting. Interlock is not released on failure.
492 */
493int
494vfs_busy(mp, flags, interlkp, td)
495	struct mount *mp;
496	int flags;
497	struct mtx *interlkp;
498	struct thread *td;
499{
500	int lkflags;
501
502	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
503		if (flags & LK_NOWAIT)
504			return (ENOENT);
505		mp->mnt_kern_flag |= MNTK_MWAIT;
506		/*
507		 * Since all busy locks are shared except the exclusive
508		 * lock granted when unmounting, the only place that a
509		 * wakeup needs to be done is at the release of the
510		 * exclusive lock at the end of dounmount.
511		 */
512		msleep(mp, interlkp, PVFS, "vfs_busy", 0);
513		return (ENOENT);
514	}
515	lkflags = LK_SHARED | LK_NOPAUSE;
516	if (interlkp)
517		lkflags |= LK_INTERLOCK;
518	if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
519		panic("vfs_busy: unexpected lock failure");
520	return (0);
521}
522
523/*
524 * Free a busy filesystem.
525 */
526void
527vfs_unbusy(mp, td)
528	struct mount *mp;
529	struct thread *td;
530{
531
532	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
533}
534
535/*
536 * Lookup a mount point by filesystem identifier.
537 */
538struct mount *
539vfs_getvfs(fsid)
540	fsid_t *fsid;
541{
542	register struct mount *mp;
543
544	mtx_lock(&mountlist_mtx);
545	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
546		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
547		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
548			mtx_unlock(&mountlist_mtx);
549			return (mp);
550		}
551	}
552	mtx_unlock(&mountlist_mtx);
553	return ((struct mount *) 0);
554}
555
556/*
557 * Get a new unique fsid.  Try to make its val[0] unique, since this value
558 * will be used to create fake device numbers for stat().  Also try (but
559 * not so hard) make its val[0] unique mod 2^16, since some emulators only
560 * support 16-bit device numbers.  We end up with unique val[0]'s for the
561 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
562 *
563 * Keep in mind that several mounts may be running in parallel.  Starting
564 * the search one past where the previous search terminated is both a
565 * micro-optimization and a defense against returning the same fsid to
566 * different mounts.
567 */
568void
569vfs_getnewfsid(mp)
570	struct mount *mp;
571{
572	static u_int16_t mntid_base;
573	fsid_t tfsid;
574	int mtype;
575
576	mtx_lock(&mntid_mtx);
577	mtype = mp->mnt_vfc->vfc_typenum;
578	tfsid.val[1] = mtype;
579	mtype = (mtype & 0xFF) << 24;
580	for (;;) {
581		tfsid.val[0] = makeudev(255,
582		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
583		mntid_base++;
584		if (vfs_getvfs(&tfsid) == NULL)
585			break;
586	}
587	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
588	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
589	mtx_unlock(&mntid_mtx);
590}
591
592/*
593 * Knob to control the precision of file timestamps:
594 *
595 *   0 = seconds only; nanoseconds zeroed.
596 *   1 = seconds and nanoseconds, accurate within 1/HZ.
597 *   2 = seconds and nanoseconds, truncated to microseconds.
598 * >=3 = seconds and nanoseconds, maximum precision.
599 */
600enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
601
602static int timestamp_precision = TSP_SEC;
603SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
604    &timestamp_precision, 0, "");
605
606/*
607 * Get a current timestamp.
608 */
609void
610vfs_timestamp(tsp)
611	struct timespec *tsp;
612{
613	struct timeval tv;
614
615	switch (timestamp_precision) {
616	case TSP_SEC:
617		tsp->tv_sec = time_second;
618		tsp->tv_nsec = 0;
619		break;
620	case TSP_HZ:
621		getnanotime(tsp);
622		break;
623	case TSP_USEC:
624		microtime(&tv);
625		TIMEVAL_TO_TIMESPEC(&tv, tsp);
626		break;
627	case TSP_NSEC:
628	default:
629		nanotime(tsp);
630		break;
631	}
632}
633
634/*
635 * Set vnode attributes to VNOVAL
636 */
637void
638vattr_null(vap)
639	register struct vattr *vap;
640{
641
642	vap->va_type = VNON;
643	vap->va_size = VNOVAL;
644	vap->va_bytes = VNOVAL;
645	vap->va_mode = VNOVAL;
646	vap->va_nlink = VNOVAL;
647	vap->va_uid = VNOVAL;
648	vap->va_gid = VNOVAL;
649	vap->va_fsid = VNOVAL;
650	vap->va_fileid = VNOVAL;
651	vap->va_blocksize = VNOVAL;
652	vap->va_rdev = VNOVAL;
653	vap->va_atime.tv_sec = VNOVAL;
654	vap->va_atime.tv_nsec = VNOVAL;
655	vap->va_mtime.tv_sec = VNOVAL;
656	vap->va_mtime.tv_nsec = VNOVAL;
657	vap->va_ctime.tv_sec = VNOVAL;
658	vap->va_ctime.tv_nsec = VNOVAL;
659	vap->va_birthtime.tv_sec = VNOVAL;
660	vap->va_birthtime.tv_nsec = VNOVAL;
661	vap->va_flags = VNOVAL;
662	vap->va_gen = VNOVAL;
663	vap->va_vaflags = 0;
664}
665
666/*
667 * This routine is called when we have too many vnodes.  It attempts
668 * to free <count> vnodes and will potentially free vnodes that still
669 * have VM backing store (VM backing store is typically the cause
670 * of a vnode blowout so we want to do this).  Therefore, this operation
671 * is not considered cheap.
672 *
673 * A number of conditions may prevent a vnode from being reclaimed.
674 * the buffer cache may have references on the vnode, a directory
675 * vnode may still have references due to the namei cache representing
676 * underlying files, or the vnode may be in active use.   It is not
677 * desireable to reuse such vnodes.  These conditions may cause the
678 * number of vnodes to reach some minimum value regardless of what
679 * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
680 */
681static int
682vlrureclaim(struct mount *mp, int count)
683{
684	struct vnode *vp;
685	int done;
686	int trigger;
687	int usevnodes;
688
689	/*
690	 * Calculate the trigger point, don't allow user
691	 * screwups to blow us up.   This prevents us from
692	 * recycling vnodes with lots of resident pages.  We
693	 * aren't trying to free memory, we are trying to
694	 * free vnodes.
695	 */
696	usevnodes = desiredvnodes;
697	if (usevnodes <= 0)
698		usevnodes = 1;
699	trigger = cnt.v_page_count * 2 / usevnodes;
700
701	done = 0;
702	mtx_lock(&mntvnode_mtx);
703	while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
704		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
705		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
706
707		if (vp->v_type != VNON &&
708		    vp->v_type != VBAD &&
709		    VI_TRYLOCK(vp)) {
710			if (VMIGHTFREE(vp) &&           /* critical path opt */
711			    (vp->v_object == NULL ||
712			    vp->v_object->resident_page_count < trigger)) {
713				mtx_unlock(&mntvnode_mtx);
714				vgonel(vp, curthread);
715				done++;
716				mtx_lock(&mntvnode_mtx);
717			} else
718				VI_UNLOCK(vp);
719		}
720		--count;
721	}
722	mtx_unlock(&mntvnode_mtx);
723	return done;
724}
725
726/*
727 * Attempt to recycle vnodes in a context that is always safe to block.
728 * Calling vlrurecycle() from the bowels of filesystem code has some
729 * interesting deadlock problems.
730 */
731static struct proc *vnlruproc;
732static int vnlruproc_sig;
733
734static void
735vnlru_proc(void)
736{
737	struct mount *mp, *nmp;
738	int s;
739	int done;
740	struct proc *p = vnlruproc;
741	struct thread *td = FIRST_THREAD_IN_PROC(p);	/* XXXKSE */
742
743	mtx_lock(&Giant);
744
745	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
746	    SHUTDOWN_PRI_FIRST);
747
748	s = splbio();
749	for (;;) {
750		kthread_suspend_check(p);
751		mtx_lock(&vnode_free_list_mtx);
752		if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
753			mtx_unlock(&vnode_free_list_mtx);
754			vnlruproc_sig = 0;
755			tsleep(vnlruproc, PVFS, "vlruwt", 0);
756			continue;
757		}
758		mtx_unlock(&vnode_free_list_mtx);
759		done = 0;
760		mtx_lock(&mountlist_mtx);
761		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
762			if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
763				nmp = TAILQ_NEXT(mp, mnt_list);
764				continue;
765			}
766			done += vlrureclaim(mp, 10);
767			mtx_lock(&mountlist_mtx);
768			nmp = TAILQ_NEXT(mp, mnt_list);
769			vfs_unbusy(mp, td);
770		}
771		mtx_unlock(&mountlist_mtx);
772		if (done == 0) {
773#if 0
774			/* These messages are temporary debugging aids */
775			if (vnlru_nowhere < 5)
776				printf("vnlru process getting nowhere..\n");
777			else if (vnlru_nowhere == 5)
778				printf("vnlru process messages stopped.\n");
779#endif
780			vnlru_nowhere++;
781			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
782		}
783	}
784	splx(s);
785}
786
787static struct kproc_desc vnlru_kp = {
788	"vnlru",
789	vnlru_proc,
790	&vnlruproc
791};
792SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
793
794
795/*
796 * Routines having to do with the management of the vnode table.
797 */
798
799/*
800 * Check to see if a free vnode can be recycled. If it can,
801 * return it locked with the vn lock, but not interlock. Also
802 * get the vn_start_write lock. Otherwise indicate the error.
803 */
804static int
805vcanrecycle(struct vnode *vp, struct mount **vnmpp)
806{
807	struct thread *td = curthread;
808	vm_object_t object;
809	int error;
810
811	/* Don't recycle if we can't get the interlock */
812	if (!VI_TRYLOCK(vp))
813		return (EWOULDBLOCK);
814
815	/* We should be able to immediately acquire this */
816	/* XXX This looks like it should panic if it fails */
817	if (vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE, td) != 0) {
818		if (VOP_ISLOCKED(vp, td))
819			panic("vcanrecycle: locked vnode");
820		return (EWOULDBLOCK);
821	}
822
823	/*
824	 * Don't recycle if its filesystem is being suspended.
825	 */
826	if (vn_start_write(vp, vnmpp, V_NOWAIT) != 0) {
827		error = EBUSY;
828		goto done;
829	}
830
831	/*
832	 * Don't recycle if we still have cached pages.
833	 */
834	if (VOP_GETVOBJECT(vp, &object) == 0 &&
835	     (object->resident_page_count ||
836	      object->ref_count)) {
837		error = EBUSY;
838		goto done;
839	}
840	if (LIST_FIRST(&vp->v_cache_src)) {
841		/*
842		 * note: nameileafonly sysctl is temporary,
843		 * for debugging only, and will eventually be
844		 * removed.
845		 */
846		if (nameileafonly > 0) {
847			/*
848			 * Do not reuse namei-cached directory
849			 * vnodes that have cached
850			 * subdirectories.
851			 */
852			if (cache_leaf_test(vp) < 0) {
853				error = EISDIR;
854				goto done;
855			}
856		} else if (nameileafonly < 0 ||
857			    vmiodirenable == 0) {
858			/*
859			 * Do not reuse namei-cached directory
860			 * vnodes if nameileafonly is -1 or
861			 * if VMIO backing for directories is
862			 * turned off (otherwise we reuse them
863			 * too quickly).
864			 */
865			error = EBUSY;
866			goto done;
867		}
868	}
869	return (0);
870done:
871	VOP_UNLOCK(vp, 0, td);
872	return (error);
873}
874
875/*
876 * Return the next vnode from the free list.
877 */
878int
879getnewvnode(tag, mp, vops, vpp)
880	const char *tag;
881	struct mount *mp;
882	vop_t **vops;
883	struct vnode **vpp;
884{
885	int s;
886	struct thread *td = curthread;	/* XXX */
887	struct vnode *vp = NULL;
888	struct mount *vnmp;
889
890	s = splbio();
891	mtx_lock(&vnode_free_list_mtx);
892
893	/*
894	 * Try to reuse vnodes if we hit the max.  This situation only
895	 * occurs in certain large-memory (2G+) situations.  We cannot
896	 * attempt to directly reclaim vnodes due to nasty recursion
897	 * problems.
898	 */
899	if (vnlruproc_sig == 0 && numvnodes - freevnodes > desiredvnodes) {
900		vnlruproc_sig = 1;      /* avoid unnecessary wakeups */
901		wakeup(vnlruproc);
902	}
903
904	/*
905	 * Attempt to reuse a vnode already on the free list, allocating
906	 * a new vnode if we can't find one or if we have not reached a
907	 * good minimum for good LRU performance.
908	 */
909
910	if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
911		int error;
912		int count;
913
914		for (count = 0; count < freevnodes; count++) {
915			vp = TAILQ_FIRST(&vnode_free_list);
916
917			KASSERT(vp->v_usecount == 0,
918			    ("getnewvnode: free vnode isn't"));
919
920			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
921			/*
922			 * We have to drop the free list mtx to avoid lock
923			 * order reversals with interlock.
924			 */
925			mtx_unlock(&vnode_free_list_mtx);
926			error = vcanrecycle(vp, &vnmp);
927			mtx_lock(&vnode_free_list_mtx);
928			if (error != 0)
929				TAILQ_INSERT_TAIL(&vnode_free_list, vp,
930				    v_freelist);
931			else
932				break;
933		}
934	}
935	if (vp) {
936		freevnodes--;
937		mtx_unlock(&vnode_free_list_mtx);
938
939		cache_purge(vp);
940		VI_LOCK(vp);
941		vp->v_iflag |= VI_DOOMED;
942		vp->v_iflag &= ~VI_FREE;
943		if (vp->v_type != VBAD) {
944			VOP_UNLOCK(vp, 0, td);
945			vgonel(vp, td);
946			VI_LOCK(vp);
947		} else {
948			VOP_UNLOCK(vp, 0, td);
949		}
950		vn_finished_write(vnmp);
951
952#ifdef INVARIANTS
953		{
954			if (vp->v_data)
955				panic("cleaned vnode isn't");
956			if (vp->v_numoutput)
957				panic("Clean vnode has pending I/O's");
958			if (vp->v_writecount != 0)
959				panic("Non-zero write count");
960		}
961#endif
962		if (vp->v_pollinfo) {
963			mtx_destroy(&vp->v_pollinfo->vpi_lock);
964			uma_zfree(vnodepoll_zone, vp->v_pollinfo);
965		}
966		vp->v_pollinfo = NULL;
967#ifdef MAC
968		mac_destroy_vnode(vp);
969#endif
970		vp->v_iflag = 0;
971		vp->v_vflag = 0;
972		vp->v_lastw = 0;
973		vp->v_lasta = 0;
974		vp->v_cstart = 0;
975		vp->v_clen = 0;
976		vp->v_socket = 0;
977		lockdestroy(vp->v_vnlock);
978		lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOPAUSE);
979		KASSERT(vp->v_cleanblkroot == NULL, ("cleanblkroot not NULL"));
980		KASSERT(vp->v_dirtyblkroot == NULL, ("dirtyblkroot not NULL"));
981	} else {
982		numvnodes++;
983		mtx_unlock(&vnode_free_list_mtx);
984
985		vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
986		mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
987		VI_LOCK(vp);
988		vp->v_dd = vp;
989		vp->v_vnlock = &vp->v_lock;
990		lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOPAUSE);
991		cache_purge(vp);
992		LIST_INIT(&vp->v_cache_src);
993		TAILQ_INIT(&vp->v_cache_dst);
994	}
995
996	TAILQ_INIT(&vp->v_cleanblkhd);
997	TAILQ_INIT(&vp->v_dirtyblkhd);
998	vp->v_type = VNON;
999	vp->v_tag = tag;
1000	vp->v_op = vops;
1001	*vpp = vp;
1002	vp->v_usecount = 1;
1003	vp->v_data = 0;
1004	vp->v_cachedid = -1;
1005	VI_UNLOCK(vp);
1006#ifdef MAC
1007	mac_init_vnode(vp);
1008#endif
1009	insmntque(vp, mp);
1010
1011	return (0);
1012}
1013
1014/*
1015 * Move a vnode from one mount queue to another.
1016 */
1017static void
1018insmntque(vp, mp)
1019	register struct vnode *vp;
1020	register struct mount *mp;
1021{
1022
1023	mtx_lock(&mntvnode_mtx);
1024	/*
1025	 * Delete from old mount point vnode list, if on one.
1026	 */
1027	if (vp->v_mount != NULL)
1028		TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
1029	/*
1030	 * Insert into list of vnodes for the new mount point, if available.
1031	 */
1032	if ((vp->v_mount = mp) == NULL) {
1033		mtx_unlock(&mntvnode_mtx);
1034		return;
1035	}
1036	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1037	mtx_unlock(&mntvnode_mtx);
1038}
1039
1040/*
1041 * Update outstanding I/O count and do wakeup if requested.
1042 */
1043void
1044vwakeup(bp)
1045	register struct buf *bp;
1046{
1047	register struct vnode *vp;
1048
1049	bp->b_flags &= ~B_WRITEINPROG;
1050	if ((vp = bp->b_vp)) {
1051		VI_LOCK(vp);
1052		vp->v_numoutput--;
1053		if (vp->v_numoutput < 0)
1054			panic("vwakeup: neg numoutput");
1055		if ((vp->v_numoutput == 0) && (vp->v_iflag & VI_BWAIT)) {
1056			vp->v_iflag &= ~VI_BWAIT;
1057			wakeup(&vp->v_numoutput);
1058		}
1059		VI_UNLOCK(vp);
1060	}
1061}
1062
1063/*
1064 * Flush out and invalidate all buffers associated with a vnode.
1065 * Called with the underlying object locked.
1066 */
1067int
1068vinvalbuf(vp, flags, cred, td, slpflag, slptimeo)
1069	struct vnode *vp;
1070	int flags;
1071	struct ucred *cred;
1072	struct thread *td;
1073	int slpflag, slptimeo;
1074{
1075	struct buf *blist;
1076	int s, error;
1077	vm_object_t object;
1078
1079	GIANT_REQUIRED;
1080
1081	ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1082
1083	VI_LOCK(vp);
1084	if (flags & V_SAVE) {
1085		s = splbio();
1086		while (vp->v_numoutput) {
1087			vp->v_iflag |= VI_BWAIT;
1088			error = msleep(&vp->v_numoutput, VI_MTX(vp),
1089			    slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo);
1090			if (error) {
1091				VI_UNLOCK(vp);
1092				splx(s);
1093				return (error);
1094			}
1095		}
1096		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1097			splx(s);
1098			VI_UNLOCK(vp);
1099			if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, td)) != 0)
1100				return (error);
1101			/*
1102			 * XXX We could save a lock/unlock if this was only
1103			 * enabled under INVARIANTS
1104			 */
1105			VI_LOCK(vp);
1106			s = splbio();
1107			if (vp->v_numoutput > 0 ||
1108			    !TAILQ_EMPTY(&vp->v_dirtyblkhd))
1109				panic("vinvalbuf: dirty bufs");
1110		}
1111		splx(s);
1112	}
1113	s = splbio();
1114	/*
1115	 * If you alter this loop please notice that interlock is dropped and
1116	 * reacquired in flushbuflist.  Special care is needed to ensure that
1117	 * no race conditions occur from this.
1118	 */
1119	for (error = 0;;) {
1120		if ((blist = TAILQ_FIRST(&vp->v_cleanblkhd)) != 0 &&
1121		    flushbuflist(blist, flags, vp, slpflag, slptimeo, &error)) {
1122			if (error)
1123				break;
1124			continue;
1125		}
1126		if ((blist = TAILQ_FIRST(&vp->v_dirtyblkhd)) != 0 &&
1127		    flushbuflist(blist, flags, vp, slpflag, slptimeo, &error)) {
1128			if (error)
1129				break;
1130			continue;
1131		}
1132		break;
1133	}
1134	if (error) {
1135		splx(s);
1136		VI_UNLOCK(vp);
1137		return (error);
1138	}
1139
1140	/*
1141	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1142	 * have write I/O in-progress but if there is a VM object then the
1143	 * VM object can also have read-I/O in-progress.
1144	 */
1145	do {
1146		while (vp->v_numoutput > 0) {
1147			vp->v_iflag |= VI_BWAIT;
1148			msleep(&vp->v_numoutput, VI_MTX(vp), PVM, "vnvlbv", 0);
1149		}
1150		VI_UNLOCK(vp);
1151		if (VOP_GETVOBJECT(vp, &object) == 0) {
1152			while (object->paging_in_progress)
1153			vm_object_pip_sleep(object, "vnvlbx");
1154		}
1155		VI_LOCK(vp);
1156	} while (vp->v_numoutput > 0);
1157	VI_UNLOCK(vp);
1158
1159	splx(s);
1160
1161	/*
1162	 * Destroy the copy in the VM cache, too.
1163	 */
1164	if (VOP_GETVOBJECT(vp, &object) == 0) {
1165		vm_object_page_remove(object, 0, 0,
1166			(flags & V_SAVE) ? TRUE : FALSE);
1167	}
1168
1169#ifdef INVARIANTS
1170	VI_LOCK(vp);
1171	if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1172	    (!TAILQ_EMPTY(&vp->v_dirtyblkhd) ||
1173	     !TAILQ_EMPTY(&vp->v_cleanblkhd)))
1174		panic("vinvalbuf: flush failed");
1175	VI_UNLOCK(vp);
1176#endif
1177	return (0);
1178}
1179
1180/*
1181 * Flush out buffers on the specified list.
1182 *
1183 */
1184static int
1185flushbuflist(blist, flags, vp, slpflag, slptimeo, errorp)
1186	struct buf *blist;
1187	int flags;
1188	struct vnode *vp;
1189	int slpflag, slptimeo;
1190	int *errorp;
1191{
1192	struct buf *bp, *nbp;
1193	int found, error;
1194
1195	ASSERT_VI_LOCKED(vp, "flushbuflist");
1196
1197	for (found = 0, bp = blist; bp; bp = nbp) {
1198		nbp = TAILQ_NEXT(bp, b_vnbufs);
1199		VI_UNLOCK(vp);
1200		if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1201		    ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1202			VI_LOCK(vp);
1203			continue;
1204		}
1205		found += 1;
1206		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1207			error = BUF_TIMELOCK(bp,
1208			    LK_EXCLUSIVE | LK_SLEEPFAIL,
1209			    "flushbuf", slpflag, slptimeo);
1210			if (error != ENOLCK)
1211				*errorp = error;
1212			goto done;
1213		}
1214		/*
1215		 * XXX Since there are no node locks for NFS, I
1216		 * believe there is a slight chance that a delayed
1217		 * write will occur while sleeping just above, so
1218		 * check for it.  Note that vfs_bio_awrite expects
1219		 * buffers to reside on a queue, while BUF_WRITE and
1220		 * brelse do not.
1221		 */
1222		if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1223			(flags & V_SAVE)) {
1224
1225			if (bp->b_vp == vp) {
1226				if (bp->b_flags & B_CLUSTEROK) {
1227					BUF_UNLOCK(bp);
1228					vfs_bio_awrite(bp);
1229				} else {
1230					bremfree(bp);
1231					bp->b_flags |= B_ASYNC;
1232					BUF_WRITE(bp);
1233				}
1234			} else {
1235				bremfree(bp);
1236				(void) BUF_WRITE(bp);
1237			}
1238			goto done;
1239		}
1240		bremfree(bp);
1241		bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
1242		bp->b_flags &= ~B_ASYNC;
1243		brelse(bp);
1244		VI_LOCK(vp);
1245	}
1246	return (found);
1247done:
1248	VI_LOCK(vp);
1249	return (found);
1250}
1251
1252/*
1253 * Truncate a file's buffer and pages to a specified length.  This
1254 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1255 * sync activity.
1256 */
1257int
1258vtruncbuf(vp, cred, td, length, blksize)
1259	register struct vnode *vp;
1260	struct ucred *cred;
1261	struct thread *td;
1262	off_t length;
1263	int blksize;
1264{
1265	register struct buf *bp;
1266	struct buf *nbp;
1267	int s, anyfreed;
1268	int trunclbn;
1269
1270	/*
1271	 * Round up to the *next* lbn.
1272	 */
1273	trunclbn = (length + blksize - 1) / blksize;
1274
1275	s = splbio();
1276	ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1277restart:
1278	VI_LOCK(vp);
1279	anyfreed = 1;
1280	for (;anyfreed;) {
1281		anyfreed = 0;
1282		for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
1283			nbp = TAILQ_NEXT(bp, b_vnbufs);
1284			VI_UNLOCK(vp);
1285			if (bp->b_lblkno >= trunclbn) {
1286				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1287					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1288					goto restart;
1289				} else {
1290					bremfree(bp);
1291					bp->b_flags |= (B_INVAL | B_RELBUF);
1292					bp->b_flags &= ~B_ASYNC;
1293					brelse(bp);
1294					anyfreed = 1;
1295				}
1296				if (nbp &&
1297				    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1298				    (nbp->b_vp != vp) ||
1299				    (nbp->b_flags & B_DELWRI))) {
1300					goto restart;
1301				}
1302			}
1303			VI_LOCK(vp);
1304		}
1305
1306		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1307			nbp = TAILQ_NEXT(bp, b_vnbufs);
1308			VI_UNLOCK(vp);
1309			if (bp->b_lblkno >= trunclbn) {
1310				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1311					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1312					goto restart;
1313				} else {
1314					bremfree(bp);
1315					bp->b_flags |= (B_INVAL | B_RELBUF);
1316					bp->b_flags &= ~B_ASYNC;
1317					brelse(bp);
1318					anyfreed = 1;
1319				}
1320				if (nbp &&
1321				    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1322				    (nbp->b_vp != vp) ||
1323				    (nbp->b_flags & B_DELWRI) == 0)) {
1324					goto restart;
1325				}
1326			}
1327			VI_LOCK(vp);
1328		}
1329	}
1330
1331	if (length > 0) {
1332restartsync:
1333		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1334			nbp = TAILQ_NEXT(bp, b_vnbufs);
1335			VI_UNLOCK(vp);
1336			if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
1337				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1338					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1339					goto restart;
1340				} else {
1341					bremfree(bp);
1342					if (bp->b_vp == vp) {
1343						bp->b_flags |= B_ASYNC;
1344					} else {
1345						bp->b_flags &= ~B_ASYNC;
1346					}
1347					BUF_WRITE(bp);
1348				}
1349				VI_LOCK(vp);
1350				goto restartsync;
1351			}
1352			VI_LOCK(vp);
1353		}
1354	}
1355
1356	while (vp->v_numoutput > 0) {
1357		vp->v_iflag |= VI_BWAIT;
1358		msleep(&vp->v_numoutput, VI_MTX(vp), PVM, "vbtrunc", 0);
1359	}
1360	VI_UNLOCK(vp);
1361	splx(s);
1362
1363	vnode_pager_setsize(vp, length);
1364
1365	return (0);
1366}
1367
1368/*
1369 * buf_splay() - splay tree core for the clean/dirty list of buffers in
1370 * 		 a vnode.
1371 *
1372 *	NOTE: We have to deal with the special case of a background bitmap
1373 *	buffer, a situation where two buffers will have the same logical
1374 *	block offset.  We want (1) only the foreground buffer to be accessed
1375 *	in a lookup and (2) must differentiate between the foreground and
1376 *	background buffer in the splay tree algorithm because the splay
1377 *	tree cannot normally handle multiple entities with the same 'index'.
1378 *	We accomplish this by adding differentiating flags to the splay tree's
1379 *	numerical domain.
1380 */
1381static
1382struct buf *
1383buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1384{
1385	struct buf dummy;
1386	struct buf *lefttreemax, *righttreemin, *y;
1387
1388	if (root == NULL)
1389		return (NULL);
1390	lefttreemax = righttreemin = &dummy;
1391	for (;;) {
1392		if (lblkno < root->b_lblkno ||
1393		    (lblkno == root->b_lblkno &&
1394		    (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1395			if ((y = root->b_left) == NULL)
1396				break;
1397			if (lblkno < y->b_lblkno) {
1398				/* Rotate right. */
1399				root->b_left = y->b_right;
1400				y->b_right = root;
1401				root = y;
1402				if ((y = root->b_left) == NULL)
1403					break;
1404			}
1405			/* Link into the new root's right tree. */
1406			righttreemin->b_left = root;
1407			righttreemin = root;
1408		} else if (lblkno > root->b_lblkno ||
1409		    (lblkno == root->b_lblkno &&
1410		    (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1411			if ((y = root->b_right) == NULL)
1412				break;
1413			if (lblkno > y->b_lblkno) {
1414				/* Rotate left. */
1415				root->b_right = y->b_left;
1416				y->b_left = root;
1417				root = y;
1418				if ((y = root->b_right) == NULL)
1419					break;
1420			}
1421			/* Link into the new root's left tree. */
1422			lefttreemax->b_right = root;
1423			lefttreemax = root;
1424		} else {
1425			break;
1426		}
1427		root = y;
1428	}
1429	/* Assemble the new root. */
1430	lefttreemax->b_right = root->b_left;
1431	righttreemin->b_left = root->b_right;
1432	root->b_left = dummy.b_right;
1433	root->b_right = dummy.b_left;
1434	return (root);
1435}
1436
1437static
1438void
1439buf_vlist_remove(struct buf *bp)
1440{
1441	struct vnode *vp = bp->b_vp;
1442	struct buf *root;
1443
1444	ASSERT_VI_LOCKED(vp, "buf_vlist_remove");
1445	if (bp->b_xflags & BX_VNDIRTY) {
1446		if (bp != vp->v_dirtyblkroot) {
1447			root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_dirtyblkroot);
1448			KASSERT(root == bp, ("splay lookup failed during dirty remove"));
1449		}
1450		if (bp->b_left == NULL) {
1451			root = bp->b_right;
1452		} else {
1453			root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1454			root->b_right = bp->b_right;
1455		}
1456		vp->v_dirtyblkroot = root;
1457		TAILQ_REMOVE(&vp->v_dirtyblkhd, bp, b_vnbufs);
1458	} else {
1459		/* KASSERT(bp->b_xflags & BX_VNCLEAN, ("bp wasn't clean")); */
1460		if (bp != vp->v_cleanblkroot) {
1461			root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_cleanblkroot);
1462			KASSERT(root == bp, ("splay lookup failed during clean remove"));
1463		}
1464		if (bp->b_left == NULL) {
1465			root = bp->b_right;
1466		} else {
1467			root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1468			root->b_right = bp->b_right;
1469		}
1470		vp->v_cleanblkroot = root;
1471		TAILQ_REMOVE(&vp->v_cleanblkhd, bp, b_vnbufs);
1472	}
1473	bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1474}
1475
1476/*
1477 * Add the buffer to the sorted clean or dirty block list using a
1478 * splay tree algorithm.
1479 *
1480 * NOTE: xflags is passed as a constant, optimizing this inline function!
1481 */
1482static
1483void
1484buf_vlist_add(struct buf *bp, struct vnode *vp, b_xflags_t xflags)
1485{
1486	struct buf *root;
1487
1488	ASSERT_VI_LOCKED(vp, "buf_vlist_add");
1489	bp->b_xflags |= xflags;
1490	if (xflags & BX_VNDIRTY) {
1491		root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_dirtyblkroot);
1492		if (root == NULL) {
1493			bp->b_left = NULL;
1494			bp->b_right = NULL;
1495			TAILQ_INSERT_TAIL(&vp->v_dirtyblkhd, bp, b_vnbufs);
1496		} else if (bp->b_lblkno < root->b_lblkno ||
1497		    (bp->b_lblkno == root->b_lblkno &&
1498		    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1499			bp->b_left = root->b_left;
1500			bp->b_right = root;
1501			root->b_left = NULL;
1502			TAILQ_INSERT_BEFORE(root, bp, b_vnbufs);
1503		} else {
1504			bp->b_right = root->b_right;
1505			bp->b_left = root;
1506			root->b_right = NULL;
1507			TAILQ_INSERT_AFTER(&vp->v_dirtyblkhd,
1508			    root, bp, b_vnbufs);
1509		}
1510		vp->v_dirtyblkroot = bp;
1511	} else {
1512		/* KASSERT(xflags & BX_VNCLEAN, ("xflags not clean")); */
1513		root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_cleanblkroot);
1514		if (root == NULL) {
1515			bp->b_left = NULL;
1516			bp->b_right = NULL;
1517			TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
1518		} else if (bp->b_lblkno < root->b_lblkno ||
1519		    (bp->b_lblkno == root->b_lblkno &&
1520		    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1521			bp->b_left = root->b_left;
1522			bp->b_right = root;
1523			root->b_left = NULL;
1524			TAILQ_INSERT_BEFORE(root, bp, b_vnbufs);
1525		} else {
1526			bp->b_right = root->b_right;
1527			bp->b_left = root;
1528			root->b_right = NULL;
1529			TAILQ_INSERT_AFTER(&vp->v_cleanblkhd,
1530			    root, bp, b_vnbufs);
1531		}
1532		vp->v_cleanblkroot = bp;
1533	}
1534}
1535
1536#ifndef USE_BUFHASH
1537
1538/*
1539 * Lookup a buffer using the splay tree.  Note that we specifically avoid
1540 * shadow buffers used in background bitmap writes.
1541 *
1542 * This code isn't quite efficient as it could be because we are maintaining
1543 * two sorted lists and do not know which list the block resides in.
1544 */
1545struct buf *
1546gbincore(struct vnode *vp, daddr_t lblkno)
1547{
1548	struct buf *bp;
1549
1550	GIANT_REQUIRED;
1551
1552	ASSERT_VI_LOCKED(vp, "gbincore");
1553	bp = vp->v_cleanblkroot = buf_splay(lblkno, 0, vp->v_cleanblkroot);
1554	if (bp && bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1555		return(bp);
1556	bp = vp->v_dirtyblkroot = buf_splay(lblkno, 0, vp->v_dirtyblkroot);
1557	if (bp && bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1558		return(bp);
1559	return(NULL);
1560}
1561
1562#endif
1563
1564/*
1565 * Associate a buffer with a vnode.
1566 */
1567void
1568bgetvp(vp, bp)
1569	register struct vnode *vp;
1570	register struct buf *bp;
1571{
1572	int s;
1573
1574	KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
1575
1576	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1577	    ("bgetvp: bp already attached! %p", bp));
1578
1579	VI_LOCK(vp);
1580	vholdl(vp);
1581	bp->b_vp = vp;
1582	bp->b_dev = vn_todev(vp);
1583	/*
1584	 * Insert onto list for new vnode.
1585	 */
1586	s = splbio();
1587	buf_vlist_add(bp, vp, BX_VNCLEAN);
1588	splx(s);
1589	VI_UNLOCK(vp);
1590}
1591
1592/*
1593 * Disassociate a buffer from a vnode.
1594 */
1595void
1596brelvp(bp)
1597	register struct buf *bp;
1598{
1599	struct vnode *vp;
1600	int s;
1601
1602	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1603
1604	/*
1605	 * Delete from old vnode list, if on one.
1606	 */
1607	vp = bp->b_vp;
1608	s = splbio();
1609	VI_LOCK(vp);
1610	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1611		buf_vlist_remove(bp);
1612	if ((vp->v_iflag & VI_ONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1613		vp->v_iflag &= ~VI_ONWORKLST;
1614		mtx_lock(&sync_mtx);
1615		LIST_REMOVE(vp, v_synclist);
1616		mtx_unlock(&sync_mtx);
1617	}
1618	vdropl(vp);
1619	VI_UNLOCK(vp);
1620	bp->b_vp = (struct vnode *) 0;
1621	if (bp->b_object)
1622		bp->b_object = NULL;
1623	splx(s);
1624}
1625
1626/*
1627 * Add an item to the syncer work queue.
1628 */
1629static void
1630vn_syncer_add_to_worklist(struct vnode *vp, int delay)
1631{
1632	int s, slot;
1633
1634	s = splbio();
1635	ASSERT_VI_LOCKED(vp, "vn_syncer_add_to_worklist");
1636
1637	mtx_lock(&sync_mtx);
1638	if (vp->v_iflag & VI_ONWORKLST)
1639		LIST_REMOVE(vp, v_synclist);
1640	else
1641		vp->v_iflag |= VI_ONWORKLST;
1642
1643	if (delay > syncer_maxdelay - 2)
1644		delay = syncer_maxdelay - 2;
1645	slot = (syncer_delayno + delay) & syncer_mask;
1646
1647	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
1648	mtx_unlock(&sync_mtx);
1649
1650	splx(s);
1651}
1652
1653struct  proc *updateproc;
1654static void sched_sync(void);
1655static struct kproc_desc up_kp = {
1656	"syncer",
1657	sched_sync,
1658	&updateproc
1659};
1660SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1661
1662/*
1663 * System filesystem synchronizer daemon.
1664 */
1665static void
1666sched_sync(void)
1667{
1668	struct synclist *slp;
1669	struct vnode *vp;
1670	struct mount *mp;
1671	long starttime;
1672	int s;
1673	struct thread *td = FIRST_THREAD_IN_PROC(updateproc);  /* XXXKSE */
1674
1675	mtx_lock(&Giant);
1676
1677	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, td->td_proc,
1678	    SHUTDOWN_PRI_LAST);
1679
1680	for (;;) {
1681		kthread_suspend_check(td->td_proc);
1682
1683		starttime = time_second;
1684
1685		/*
1686		 * Push files whose dirty time has expired.  Be careful
1687		 * of interrupt race on slp queue.
1688		 */
1689		s = splbio();
1690		mtx_lock(&sync_mtx);
1691		slp = &syncer_workitem_pending[syncer_delayno];
1692		syncer_delayno += 1;
1693		if (syncer_delayno == syncer_maxdelay)
1694			syncer_delayno = 0;
1695		splx(s);
1696
1697		while ((vp = LIST_FIRST(slp)) != NULL) {
1698			mtx_unlock(&sync_mtx);
1699			if (VOP_ISLOCKED(vp, NULL) == 0 &&
1700			    vn_start_write(vp, &mp, V_NOWAIT) == 0) {
1701				vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1702				(void) VOP_FSYNC(vp, td->td_ucred, MNT_LAZY, td);
1703				VOP_UNLOCK(vp, 0, td);
1704				vn_finished_write(mp);
1705			}
1706			s = splbio();
1707			mtx_lock(&sync_mtx);
1708			if (LIST_FIRST(slp) == vp) {
1709				mtx_unlock(&sync_mtx);
1710				/*
1711				 * Note: VFS vnodes can remain on the
1712				 * worklist too with no dirty blocks, but
1713				 * since sync_fsync() moves it to a different
1714				 * slot we are safe.
1715				 */
1716				VI_LOCK(vp);
1717				if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1718				    !vn_isdisk(vp, NULL)) {
1719					panic("sched_sync: fsync failed "
1720					      "vp %p tag %s", vp, vp->v_tag);
1721				}
1722				/*
1723				 * Put us back on the worklist.  The worklist
1724				 * routine will remove us from our current
1725				 * position and then add us back in at a later
1726				 * position.
1727				 */
1728				vn_syncer_add_to_worklist(vp, syncdelay);
1729				VI_UNLOCK(vp);
1730				mtx_lock(&sync_mtx);
1731			}
1732			splx(s);
1733		}
1734		mtx_unlock(&sync_mtx);
1735
1736		/*
1737		 * Do soft update processing.
1738		 */
1739		if (softdep_process_worklist_hook != NULL)
1740			(*softdep_process_worklist_hook)(NULL);
1741
1742		/*
1743		 * The variable rushjob allows the kernel to speed up the
1744		 * processing of the filesystem syncer process. A rushjob
1745		 * value of N tells the filesystem syncer to process the next
1746		 * N seconds worth of work on its queue ASAP. Currently rushjob
1747		 * is used by the soft update code to speed up the filesystem
1748		 * syncer process when the incore state is getting so far
1749		 * ahead of the disk that the kernel memory pool is being
1750		 * threatened with exhaustion.
1751		 */
1752		mtx_lock(&sync_mtx);
1753		if (rushjob > 0) {
1754			rushjob -= 1;
1755			mtx_unlock(&sync_mtx);
1756			continue;
1757		}
1758		mtx_unlock(&sync_mtx);
1759		/*
1760		 * If it has taken us less than a second to process the
1761		 * current work, then wait. Otherwise start right over
1762		 * again. We can still lose time if any single round
1763		 * takes more than two seconds, but it does not really
1764		 * matter as we are just trying to generally pace the
1765		 * filesystem activity.
1766		 */
1767		if (time_second == starttime)
1768			tsleep(&lbolt, PPAUSE, "syncer", 0);
1769	}
1770}
1771
1772/*
1773 * Request the syncer daemon to speed up its work.
1774 * We never push it to speed up more than half of its
1775 * normal turn time, otherwise it could take over the cpu.
1776 * XXXKSE  only one update?
1777 */
1778int
1779speedup_syncer()
1780{
1781	struct thread *td;
1782	int ret = 0;
1783
1784	td = FIRST_THREAD_IN_PROC(updateproc);
1785	mtx_lock_spin(&sched_lock);
1786	if (td->td_wchan == &lbolt) {
1787		unsleep(td);
1788		TD_CLR_SLEEPING(td);
1789		setrunnable(td);
1790	}
1791	mtx_unlock_spin(&sched_lock);
1792	mtx_lock(&sync_mtx);
1793	if (rushjob < syncdelay / 2) {
1794		rushjob += 1;
1795		stat_rush_requests += 1;
1796		ret = 1;
1797	}
1798	mtx_unlock(&sync_mtx);
1799	return (ret);
1800}
1801
1802/*
1803 * Associate a p-buffer with a vnode.
1804 *
1805 * Also sets B_PAGING flag to indicate that vnode is not fully associated
1806 * with the buffer.  i.e. the bp has not been linked into the vnode or
1807 * ref-counted.
1808 */
1809void
1810pbgetvp(vp, bp)
1811	register struct vnode *vp;
1812	register struct buf *bp;
1813{
1814
1815	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1816
1817	bp->b_vp = vp;
1818	bp->b_flags |= B_PAGING;
1819	bp->b_dev = vn_todev(vp);
1820}
1821
1822/*
1823 * Disassociate a p-buffer from a vnode.
1824 */
1825void
1826pbrelvp(bp)
1827	register struct buf *bp;
1828{
1829
1830	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1831
1832	/* XXX REMOVE ME */
1833	VI_LOCK(bp->b_vp);
1834	if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1835		panic(
1836		    "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1837		    bp,
1838		    (int)bp->b_flags
1839		);
1840	}
1841	VI_UNLOCK(bp->b_vp);
1842	bp->b_vp = (struct vnode *) 0;
1843	bp->b_flags &= ~B_PAGING;
1844}
1845
1846/*
1847 * Reassign a buffer from one vnode to another.
1848 * Used to assign file specific control information
1849 * (indirect blocks) to the vnode to which they belong.
1850 */
1851void
1852reassignbuf(bp, newvp)
1853	register struct buf *bp;
1854	register struct vnode *newvp;
1855{
1856	int delay;
1857	int s;
1858
1859	if (newvp == NULL) {
1860		printf("reassignbuf: NULL");
1861		return;
1862	}
1863	++reassignbufcalls;
1864
1865	/*
1866	 * B_PAGING flagged buffers cannot be reassigned because their vp
1867	 * is not fully linked in.
1868	 */
1869	if (bp->b_flags & B_PAGING)
1870		panic("cannot reassign paging buffer");
1871
1872	s = splbio();
1873	/*
1874	 * Delete from old vnode list, if on one.
1875	 */
1876	VI_LOCK(bp->b_vp);
1877	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1878		buf_vlist_remove(bp);
1879		if (bp->b_vp != newvp) {
1880			vdropl(bp->b_vp);
1881			bp->b_vp = NULL;	/* for clarification */
1882		}
1883	}
1884	VI_UNLOCK(bp->b_vp);
1885	/*
1886	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1887	 * of clean buffers.
1888	 */
1889	VI_LOCK(newvp);
1890	if (bp->b_flags & B_DELWRI) {
1891		if ((newvp->v_iflag & VI_ONWORKLST) == 0) {
1892			switch (newvp->v_type) {
1893			case VDIR:
1894				delay = dirdelay;
1895				break;
1896			case VCHR:
1897				if (newvp->v_rdev->si_mountpoint != NULL) {
1898					delay = metadelay;
1899					break;
1900				}
1901				/* FALLTHROUGH */
1902			default:
1903				delay = filedelay;
1904			}
1905			vn_syncer_add_to_worklist(newvp, delay);
1906		}
1907		buf_vlist_add(bp, newvp, BX_VNDIRTY);
1908	} else {
1909		buf_vlist_add(bp, newvp, BX_VNCLEAN);
1910
1911		if ((newvp->v_iflag & VI_ONWORKLST) &&
1912		    TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1913			mtx_lock(&sync_mtx);
1914			LIST_REMOVE(newvp, v_synclist);
1915			mtx_unlock(&sync_mtx);
1916			newvp->v_iflag &= ~VI_ONWORKLST;
1917		}
1918	}
1919	if (bp->b_vp != newvp) {
1920		bp->b_vp = newvp;
1921		vholdl(bp->b_vp);
1922	}
1923	VI_UNLOCK(newvp);
1924	splx(s);
1925}
1926
1927/*
1928 * Create a vnode for a device.
1929 * Used for mounting the root filesystem.
1930 */
1931int
1932bdevvp(dev, vpp)
1933	dev_t dev;
1934	struct vnode **vpp;
1935{
1936	register struct vnode *vp;
1937	struct vnode *nvp;
1938	int error;
1939
1940	if (dev == NODEV) {
1941		*vpp = NULLVP;
1942		return (ENXIO);
1943	}
1944	if (vfinddev(dev, VCHR, vpp))
1945		return (0);
1946	error = getnewvnode("none", (struct mount *)0, spec_vnodeop_p, &nvp);
1947	if (error) {
1948		*vpp = NULLVP;
1949		return (error);
1950	}
1951	vp = nvp;
1952	vp->v_type = VCHR;
1953	addalias(vp, dev);
1954	*vpp = vp;
1955	return (0);
1956}
1957
1958/*
1959 * Add vnode to the alias list hung off the dev_t.
1960 *
1961 * The reason for this gunk is that multiple vnodes can reference
1962 * the same physical device, so checking vp->v_usecount to see
1963 * how many users there are is inadequate; the v_usecount for
1964 * the vnodes need to be accumulated.  vcount() does that.
1965 */
1966struct vnode *
1967addaliasu(nvp, nvp_rdev)
1968	struct vnode *nvp;
1969	udev_t nvp_rdev;
1970{
1971	struct vnode *ovp;
1972	vop_t **ops;
1973	dev_t dev;
1974
1975	if (nvp->v_type == VBLK)
1976		return (nvp);
1977	if (nvp->v_type != VCHR)
1978		panic("addaliasu on non-special vnode");
1979	dev = udev2dev(nvp_rdev, 0);
1980	/*
1981	 * Check to see if we have a bdevvp vnode with no associated
1982	 * filesystem. If so, we want to associate the filesystem of
1983	 * the new newly instigated vnode with the bdevvp vnode and
1984	 * discard the newly created vnode rather than leaving the
1985	 * bdevvp vnode lying around with no associated filesystem.
1986	 */
1987	if (vfinddev(dev, nvp->v_type, &ovp) == 0 || ovp->v_data != NULL) {
1988		addalias(nvp, dev);
1989		return (nvp);
1990	}
1991	/*
1992	 * Discard unneeded vnode, but save its node specific data.
1993	 * Note that if there is a lock, it is carried over in the
1994	 * node specific data to the replacement vnode.
1995	 */
1996	vref(ovp);
1997	ovp->v_data = nvp->v_data;
1998	ovp->v_tag = nvp->v_tag;
1999	nvp->v_data = NULL;
2000	lockdestroy(ovp->v_vnlock);
2001	lockinit(ovp->v_vnlock, PVFS, nvp->v_vnlock->lk_wmesg,
2002	    nvp->v_vnlock->lk_timo, nvp->v_vnlock->lk_flags & LK_EXTFLG_MASK);
2003	ops = ovp->v_op;
2004	ovp->v_op = nvp->v_op;
2005	if (VOP_ISLOCKED(nvp, curthread)) {
2006		VOP_UNLOCK(nvp, 0, curthread);
2007		vn_lock(ovp, LK_EXCLUSIVE | LK_RETRY, curthread);
2008	}
2009	nvp->v_op = ops;
2010	insmntque(ovp, nvp->v_mount);
2011	vrele(nvp);
2012	vgone(nvp);
2013	return (ovp);
2014}
2015
2016/* This is a local helper function that do the same as addaliasu, but for a
2017 * dev_t instead of an udev_t. */
2018static void
2019addalias(nvp, dev)
2020	struct vnode *nvp;
2021	dev_t dev;
2022{
2023
2024	KASSERT(nvp->v_type == VCHR, ("addalias on non-special vnode"));
2025	nvp->v_rdev = dev;
2026	mtx_lock(&spechash_mtx);
2027	SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
2028	mtx_unlock(&spechash_mtx);
2029}
2030
2031/*
2032 * Grab a particular vnode from the free list, increment its
2033 * reference count and lock it. The vnode lock bit is set if the
2034 * vnode is being eliminated in vgone. The process is awakened
2035 * when the transition is completed, and an error returned to
2036 * indicate that the vnode is no longer usable (possibly having
2037 * been changed to a new filesystem type).
2038 */
2039int
2040vget(vp, flags, td)
2041	register struct vnode *vp;
2042	int flags;
2043	struct thread *td;
2044{
2045	int error;
2046
2047	/*
2048	 * If the vnode is in the process of being cleaned out for
2049	 * another use, we wait for the cleaning to finish and then
2050	 * return failure. Cleaning is determined by checking that
2051	 * the VI_XLOCK flag is set.
2052	 */
2053	if ((flags & LK_INTERLOCK) == 0)
2054		VI_LOCK(vp);
2055	if (vp->v_iflag & VI_XLOCK && vp->v_vxproc != curthread) {
2056		vp->v_iflag |= VI_XWANT;
2057		msleep(vp, VI_MTX(vp), PINOD | PDROP, "vget", 0);
2058		return (ENOENT);
2059	}
2060
2061	vp->v_usecount++;
2062
2063	if (VSHOULDBUSY(vp))
2064		vbusy(vp);
2065	if (flags & LK_TYPE_MASK) {
2066		if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
2067			/*
2068			 * must expand vrele here because we do not want
2069			 * to call VOP_INACTIVE if the reference count
2070			 * drops back to zero since it was never really
2071			 * active. We must remove it from the free list
2072			 * before sleeping so that multiple processes do
2073			 * not try to recycle it.
2074			 */
2075			VI_LOCK(vp);
2076			vp->v_usecount--;
2077			if (VSHOULDFREE(vp))
2078				vfree(vp);
2079			else
2080				vlruvp(vp);
2081			VI_UNLOCK(vp);
2082		}
2083		return (error);
2084	}
2085	VI_UNLOCK(vp);
2086	return (0);
2087}
2088
2089/*
2090 * Increase the reference count of a vnode.
2091 */
2092void
2093vref(struct vnode *vp)
2094{
2095	VI_LOCK(vp);
2096	vp->v_usecount++;
2097	VI_UNLOCK(vp);
2098}
2099
2100/*
2101 * Return reference count of a vnode.
2102 *
2103 * The results of this call are only guaranteed when some mechanism other
2104 * than the VI lock is used to stop other processes from gaining references
2105 * to the vnode.  This may be the case if the caller holds the only reference.
2106 * This is also useful when stale data is acceptable as race conditions may
2107 * be accounted for by some other means.
2108 */
2109int
2110vrefcnt(struct vnode *vp)
2111{
2112	int usecnt;
2113
2114	VI_LOCK(vp);
2115	usecnt = vp->v_usecount;
2116	VI_UNLOCK(vp);
2117
2118	return (usecnt);
2119}
2120
2121
2122/*
2123 * Vnode put/release.
2124 * If count drops to zero, call inactive routine and return to freelist.
2125 */
2126void
2127vrele(vp)
2128	struct vnode *vp;
2129{
2130	struct thread *td = curthread;	/* XXX */
2131
2132	KASSERT(vp != NULL, ("vrele: null vp"));
2133
2134	VI_LOCK(vp);
2135
2136	/* Skip this v_writecount check if we're going to panic below. */
2137	KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
2138	    ("vrele: missed vn_close"));
2139
2140	if (vp->v_usecount > 1) {
2141
2142		vp->v_usecount--;
2143		VI_UNLOCK(vp);
2144
2145		return;
2146	}
2147
2148	if (vp->v_usecount == 1) {
2149		vp->v_usecount--;
2150		/*
2151		 * We must call VOP_INACTIVE with the node locked.
2152		 * If we are doing a vput, the node is already locked,
2153		 * but, in the case of vrele, we must explicitly lock
2154		 * the vnode before calling VOP_INACTIVE.
2155		 */
2156		if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0)
2157			VOP_INACTIVE(vp, td);
2158		VI_LOCK(vp);
2159		if (VSHOULDFREE(vp))
2160			vfree(vp);
2161		else
2162			vlruvp(vp);
2163		VI_UNLOCK(vp);
2164
2165	} else {
2166#ifdef DIAGNOSTIC
2167		vprint("vrele: negative ref count", vp);
2168#endif
2169		VI_UNLOCK(vp);
2170		panic("vrele: negative ref cnt");
2171	}
2172}
2173
2174/*
2175 * Release an already locked vnode.  This give the same effects as
2176 * unlock+vrele(), but takes less time and avoids releasing and
2177 * re-aquiring the lock (as vrele() aquires the lock internally.)
2178 */
2179void
2180vput(vp)
2181	struct vnode *vp;
2182{
2183	struct thread *td = curthread;	/* XXX */
2184
2185	GIANT_REQUIRED;
2186
2187	KASSERT(vp != NULL, ("vput: null vp"));
2188	VI_LOCK(vp);
2189	/* Skip this v_writecount check if we're going to panic below. */
2190	KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
2191	    ("vput: missed vn_close"));
2192
2193	if (vp->v_usecount > 1) {
2194		vp->v_usecount--;
2195		VOP_UNLOCK(vp, LK_INTERLOCK, td);
2196		return;
2197	}
2198
2199	if (vp->v_usecount == 1) {
2200		vp->v_usecount--;
2201		/*
2202		 * We must call VOP_INACTIVE with the node locked.
2203		 * If we are doing a vput, the node is already locked,
2204		 * so we just need to release the vnode mutex.
2205		 */
2206		VI_UNLOCK(vp);
2207		VOP_INACTIVE(vp, td);
2208		VI_LOCK(vp);
2209		if (VSHOULDFREE(vp))
2210			vfree(vp);
2211		else
2212			vlruvp(vp);
2213		VI_UNLOCK(vp);
2214
2215	} else {
2216#ifdef DIAGNOSTIC
2217		vprint("vput: negative ref count", vp);
2218#endif
2219		panic("vput: negative ref cnt");
2220	}
2221}
2222
2223/*
2224 * Somebody doesn't want the vnode recycled.
2225 */
2226void
2227vhold(struct vnode *vp)
2228{
2229	VI_LOCK(vp);
2230	vholdl(vp);
2231	VI_UNLOCK(vp);
2232}
2233
2234void
2235vholdl(vp)
2236	register struct vnode *vp;
2237{
2238	int s;
2239
2240	s = splbio();
2241	vp->v_holdcnt++;
2242	if (VSHOULDBUSY(vp))
2243		vbusy(vp);
2244	splx(s);
2245}
2246
2247/*
2248 * Note that there is one less who cares about this vnode.  vdrop() is the
2249 * opposite of vhold().
2250 */
2251void
2252vdrop(struct vnode *vp)
2253{
2254	VI_LOCK(vp);
2255	vdropl(vp);
2256	VI_UNLOCK(vp);
2257}
2258
2259void
2260vdropl(vp)
2261	register struct vnode *vp;
2262{
2263	int s;
2264
2265	s = splbio();
2266	if (vp->v_holdcnt <= 0)
2267		panic("vdrop: holdcnt");
2268	vp->v_holdcnt--;
2269	if (VSHOULDFREE(vp))
2270		vfree(vp);
2271	else
2272		vlruvp(vp);
2273	splx(s);
2274}
2275
2276/*
2277 * Remove any vnodes in the vnode table belonging to mount point mp.
2278 *
2279 * If FORCECLOSE is not specified, there should not be any active ones,
2280 * return error if any are found (nb: this is a user error, not a
2281 * system error). If FORCECLOSE is specified, detach any active vnodes
2282 * that are found.
2283 *
2284 * If WRITECLOSE is set, only flush out regular file vnodes open for
2285 * writing.
2286 *
2287 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2288 *
2289 * `rootrefs' specifies the base reference count for the root vnode
2290 * of this filesystem. The root vnode is considered busy if its
2291 * v_usecount exceeds this value. On a successful return, vflush()
2292 * will call vrele() on the root vnode exactly rootrefs times.
2293 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2294 * be zero.
2295 */
2296#ifdef DIAGNOSTIC
2297static int busyprt = 0;		/* print out busy vnodes */
2298SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
2299#endif
2300
2301int
2302vflush(mp, rootrefs, flags)
2303	struct mount *mp;
2304	int rootrefs;
2305	int flags;
2306{
2307	struct thread *td = curthread;	/* XXX */
2308	struct vnode *vp, *nvp, *rootvp = NULL;
2309	struct vattr vattr;
2310	int busy = 0, error;
2311
2312	if (rootrefs > 0) {
2313		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2314		    ("vflush: bad args"));
2315		/*
2316		 * Get the filesystem root vnode. We can vput() it
2317		 * immediately, since with rootrefs > 0, it won't go away.
2318		 */
2319		if ((error = VFS_ROOT(mp, &rootvp)) != 0)
2320			return (error);
2321		vput(rootvp);
2322
2323	}
2324	mtx_lock(&mntvnode_mtx);
2325loop:
2326	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) {
2327		/*
2328		 * Make sure this vnode wasn't reclaimed in getnewvnode().
2329		 * Start over if it has (it won't be on the list anymore).
2330		 */
2331		if (vp->v_mount != mp)
2332			goto loop;
2333		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2334
2335		VI_LOCK(vp);
2336		mtx_unlock(&mntvnode_mtx);
2337		vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY, td);
2338		/*
2339		 * Skip over a vnodes marked VV_SYSTEM.
2340		 */
2341		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2342			VOP_UNLOCK(vp, 0, td);
2343			mtx_lock(&mntvnode_mtx);
2344			continue;
2345		}
2346		/*
2347		 * If WRITECLOSE is set, flush out unlinked but still open
2348		 * files (even if open only for reading) and regular file
2349		 * vnodes open for writing.
2350		 */
2351		if (flags & WRITECLOSE) {
2352			error = VOP_GETATTR(vp, &vattr, td->td_ucred, td);
2353			VI_LOCK(vp);
2354
2355			if ((vp->v_type == VNON ||
2356			    (error == 0 && vattr.va_nlink > 0)) &&
2357			    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2358				VOP_UNLOCK(vp, LK_INTERLOCK, td);
2359				mtx_lock(&mntvnode_mtx);
2360				continue;
2361			}
2362		} else
2363			VI_LOCK(vp);
2364
2365		VOP_UNLOCK(vp, 0, td);
2366
2367		/*
2368		 * With v_usecount == 0, all we need to do is clear out the
2369		 * vnode data structures and we are done.
2370		 */
2371		if (vp->v_usecount == 0) {
2372			vgonel(vp, td);
2373			mtx_lock(&mntvnode_mtx);
2374			continue;
2375		}
2376
2377		/*
2378		 * If FORCECLOSE is set, forcibly close the vnode. For block
2379		 * or character devices, revert to an anonymous device. For
2380		 * all other files, just kill them.
2381		 */
2382		if (flags & FORCECLOSE) {
2383			if (vp->v_type != VCHR) {
2384				vgonel(vp, td);
2385			} else {
2386				vclean(vp, 0, td);
2387				VI_UNLOCK(vp);
2388				vp->v_op = spec_vnodeop_p;
2389				insmntque(vp, (struct mount *) 0);
2390			}
2391			mtx_lock(&mntvnode_mtx);
2392			continue;
2393		}
2394#ifdef DIAGNOSTIC
2395		if (busyprt)
2396			vprint("vflush: busy vnode", vp);
2397#endif
2398		VI_UNLOCK(vp);
2399		mtx_lock(&mntvnode_mtx);
2400		busy++;
2401	}
2402	mtx_unlock(&mntvnode_mtx);
2403	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2404		/*
2405		 * If just the root vnode is busy, and if its refcount
2406		 * is equal to `rootrefs', then go ahead and kill it.
2407		 */
2408		VI_LOCK(rootvp);
2409		KASSERT(busy > 0, ("vflush: not busy"));
2410		KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
2411		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2412			vgonel(rootvp, td);
2413			busy = 0;
2414		} else
2415			VI_UNLOCK(rootvp);
2416	}
2417	if (busy)
2418		return (EBUSY);
2419	for (; rootrefs > 0; rootrefs--)
2420		vrele(rootvp);
2421	return (0);
2422}
2423
2424/*
2425 * This moves a now (likely recyclable) vnode to the end of the
2426 * mountlist.  XXX However, it is temporarily disabled until we
2427 * can clean up ffs_sync() and friends, which have loop restart
2428 * conditions which this code causes to operate O(N^2).
2429 */
2430static void
2431vlruvp(struct vnode *vp)
2432{
2433#if 0
2434	struct mount *mp;
2435
2436	if ((mp = vp->v_mount) != NULL) {
2437		mtx_lock(&mntvnode_mtx);
2438		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
2439		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
2440		mtx_unlock(&mntvnode_mtx);
2441	}
2442#endif
2443}
2444
2445/*
2446 * Disassociate the underlying filesystem from a vnode.
2447 */
2448static void
2449vclean(vp, flags, td)
2450	struct vnode *vp;
2451	int flags;
2452	struct thread *td;
2453{
2454	int active;
2455
2456	ASSERT_VI_LOCKED(vp, "vclean");
2457	/*
2458	 * Check to see if the vnode is in use. If so we have to reference it
2459	 * before we clean it out so that its count cannot fall to zero and
2460	 * generate a race against ourselves to recycle it.
2461	 */
2462	if ((active = vp->v_usecount))
2463		vp->v_usecount++;
2464
2465	/*
2466	 * Prevent the vnode from being recycled or brought into use while we
2467	 * clean it out.
2468	 */
2469	if (vp->v_iflag & VI_XLOCK)
2470		panic("vclean: deadlock");
2471	vp->v_iflag |= VI_XLOCK;
2472	vp->v_vxproc = curthread;
2473	/*
2474	 * Even if the count is zero, the VOP_INACTIVE routine may still
2475	 * have the object locked while it cleans it out. The VOP_LOCK
2476	 * ensures that the VOP_INACTIVE routine is done with its work.
2477	 * For active vnodes, it ensures that no other activity can
2478	 * occur while the underlying object is being cleaned out.
2479	 */
2480	VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td);
2481
2482	/*
2483	 * Clean out any buffers associated with the vnode.
2484	 * If the flush fails, just toss the buffers.
2485	 */
2486	if (flags & DOCLOSE) {
2487		struct buf *bp;
2488		VI_LOCK(vp);
2489		bp = TAILQ_FIRST(&vp->v_dirtyblkhd);
2490		VI_UNLOCK(vp);
2491		if (bp != NULL)
2492			(void) vn_write_suspend_wait(vp, NULL, V_WAIT);
2493		if (vinvalbuf(vp, V_SAVE, NOCRED, td, 0, 0) != 0)
2494			vinvalbuf(vp, 0, NOCRED, td, 0, 0);
2495	}
2496
2497	VOP_DESTROYVOBJECT(vp);
2498
2499	/*
2500	 * Any other processes trying to obtain this lock must first
2501	 * wait for VXLOCK to clear, then call the new lock operation.
2502	 */
2503	VOP_UNLOCK(vp, 0, td);
2504
2505	/*
2506	 * If purging an active vnode, it must be closed and
2507	 * deactivated before being reclaimed. Note that the
2508	 * VOP_INACTIVE will unlock the vnode.
2509	 */
2510	if (active) {
2511		if (flags & DOCLOSE)
2512			VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2513		if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT, td) != 0)
2514			panic("vclean: cannot relock.");
2515		VOP_INACTIVE(vp, td);
2516	}
2517
2518	/*
2519	 * Reclaim the vnode.
2520	 */
2521	if (VOP_RECLAIM(vp, td))
2522		panic("vclean: cannot reclaim");
2523
2524	if (active) {
2525		/*
2526		 * Inline copy of vrele() since VOP_INACTIVE
2527		 * has already been called.
2528		 */
2529		VI_LOCK(vp);
2530		if (--vp->v_usecount <= 0) {
2531#ifdef DIAGNOSTIC
2532			if (vp->v_usecount < 0 || vp->v_writecount != 0) {
2533				vprint("vclean: bad ref count", vp);
2534				panic("vclean: ref cnt");
2535			}
2536#endif
2537			vfree(vp);
2538		}
2539		VI_UNLOCK(vp);
2540	}
2541
2542	cache_purge(vp);
2543	VI_LOCK(vp);
2544	if (VSHOULDFREE(vp))
2545		vfree(vp);
2546
2547	/*
2548	 * Done with purge, notify sleepers of the grim news.
2549	 */
2550	vp->v_op = dead_vnodeop_p;
2551	if (vp->v_pollinfo != NULL)
2552		vn_pollgone(vp);
2553	vp->v_tag = "none";
2554	vp->v_iflag &= ~VI_XLOCK;
2555	vp->v_vxproc = NULL;
2556	if (vp->v_iflag & VI_XWANT) {
2557		vp->v_iflag &= ~VI_XWANT;
2558		wakeup(vp);
2559	}
2560}
2561
2562/*
2563 * Eliminate all activity associated with the requested vnode
2564 * and with all vnodes aliased to the requested vnode.
2565 */
2566int
2567vop_revoke(ap)
2568	struct vop_revoke_args /* {
2569		struct vnode *a_vp;
2570		int a_flags;
2571	} */ *ap;
2572{
2573	struct vnode *vp, *vq;
2574	dev_t dev;
2575
2576	KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
2577
2578	vp = ap->a_vp;
2579	VI_LOCK(vp);
2580	/*
2581	 * If a vgone (or vclean) is already in progress,
2582	 * wait until it is done and return.
2583	 */
2584	if (vp->v_iflag & VI_XLOCK) {
2585		vp->v_iflag |= VI_XWANT;
2586		msleep(vp, VI_MTX(vp), PINOD | PDROP,
2587		    "vop_revokeall", 0);
2588		return (0);
2589	}
2590	VI_UNLOCK(vp);
2591	dev = vp->v_rdev;
2592	for (;;) {
2593		mtx_lock(&spechash_mtx);
2594		vq = SLIST_FIRST(&dev->si_hlist);
2595		mtx_unlock(&spechash_mtx);
2596		if (!vq)
2597			break;
2598		vgone(vq);
2599	}
2600	return (0);
2601}
2602
2603/*
2604 * Recycle an unused vnode to the front of the free list.
2605 * Release the passed interlock if the vnode will be recycled.
2606 */
2607int
2608vrecycle(vp, inter_lkp, td)
2609	struct vnode *vp;
2610	struct mtx *inter_lkp;
2611	struct thread *td;
2612{
2613
2614	VI_LOCK(vp);
2615	if (vp->v_usecount == 0) {
2616		if (inter_lkp) {
2617			mtx_unlock(inter_lkp);
2618		}
2619		vgonel(vp, td);
2620		return (1);
2621	}
2622	VI_UNLOCK(vp);
2623	return (0);
2624}
2625
2626/*
2627 * Eliminate all activity associated with a vnode
2628 * in preparation for reuse.
2629 */
2630void
2631vgone(vp)
2632	register struct vnode *vp;
2633{
2634	struct thread *td = curthread;	/* XXX */
2635
2636	VI_LOCK(vp);
2637	vgonel(vp, td);
2638}
2639
2640/*
2641 * vgone, with the vp interlock held.
2642 */
2643void
2644vgonel(vp, td)
2645	struct vnode *vp;
2646	struct thread *td;
2647{
2648	int s;
2649
2650	/*
2651	 * If a vgone (or vclean) is already in progress,
2652	 * wait until it is done and return.
2653	 */
2654	ASSERT_VI_LOCKED(vp, "vgonel");
2655	if (vp->v_iflag & VI_XLOCK) {
2656		vp->v_iflag |= VI_XWANT;
2657		msleep(vp, VI_MTX(vp), PINOD | PDROP, "vgone", 0);
2658		return;
2659	}
2660
2661	/*
2662	 * Clean out the filesystem specific data.
2663	 */
2664	vclean(vp, DOCLOSE, td);
2665	VI_UNLOCK(vp);
2666
2667	/*
2668	 * Delete from old mount point vnode list, if on one.
2669	 */
2670	if (vp->v_mount != NULL)
2671		insmntque(vp, (struct mount *)0);
2672	/*
2673	 * If special device, remove it from special device alias list
2674	 * if it is on one.
2675	 */
2676	if (vp->v_type == VCHR && vp->v_rdev != NULL && vp->v_rdev != NODEV) {
2677		mtx_lock(&spechash_mtx);
2678		SLIST_REMOVE(&vp->v_rdev->si_hlist, vp, vnode, v_specnext);
2679		freedev(vp->v_rdev);
2680		mtx_unlock(&spechash_mtx);
2681		vp->v_rdev = NULL;
2682	}
2683
2684	/*
2685	 * If it is on the freelist and not already at the head,
2686	 * move it to the head of the list. The test of the
2687	 * VDOOMED flag and the reference count of zero is because
2688	 * it will be removed from the free list by getnewvnode,
2689	 * but will not have its reference count incremented until
2690	 * after calling vgone. If the reference count were
2691	 * incremented first, vgone would (incorrectly) try to
2692	 * close the previous instance of the underlying object.
2693	 */
2694	VI_LOCK(vp);
2695	if (vp->v_usecount == 0 && !(vp->v_iflag & VI_DOOMED)) {
2696		s = splbio();
2697		mtx_lock(&vnode_free_list_mtx);
2698		if (vp->v_iflag & VI_FREE) {
2699			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2700		} else {
2701			vp->v_iflag |= VI_FREE;
2702			freevnodes++;
2703		}
2704		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2705		mtx_unlock(&vnode_free_list_mtx);
2706		splx(s);
2707	}
2708
2709	vp->v_type = VBAD;
2710	VI_UNLOCK(vp);
2711}
2712
2713/*
2714 * Lookup a vnode by device number.
2715 */
2716int
2717vfinddev(dev, type, vpp)
2718	dev_t dev;
2719	enum vtype type;
2720	struct vnode **vpp;
2721{
2722	struct vnode *vp;
2723
2724	mtx_lock(&spechash_mtx);
2725	SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2726		if (type == vp->v_type) {
2727			*vpp = vp;
2728			mtx_unlock(&spechash_mtx);
2729			return (1);
2730		}
2731	}
2732	mtx_unlock(&spechash_mtx);
2733	return (0);
2734}
2735
2736/*
2737 * Calculate the total number of references to a special device.
2738 */
2739int
2740vcount(vp)
2741	struct vnode *vp;
2742{
2743	struct vnode *vq;
2744	int count;
2745
2746	count = 0;
2747	mtx_lock(&spechash_mtx);
2748	SLIST_FOREACH(vq, &vp->v_rdev->si_hlist, v_specnext) {
2749		if (vq != vp)
2750			VI_LOCK(vq);
2751		count += vq->v_usecount;
2752		if (vq != vp)
2753			VI_UNLOCK(vq);
2754	}
2755	mtx_unlock(&spechash_mtx);
2756	return (count);
2757}
2758
2759/*
2760 * Same as above, but using the dev_t as argument
2761 */
2762int
2763count_dev(dev)
2764	dev_t dev;
2765{
2766	struct vnode *vp;
2767
2768	vp = SLIST_FIRST(&dev->si_hlist);
2769	if (vp == NULL)
2770		return (0);
2771	return(vcount(vp));
2772}
2773
2774/*
2775 * Print out a description of a vnode.
2776 */
2777static char *typename[] =
2778{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2779
2780void
2781vprint(label, vp)
2782	char *label;
2783	struct vnode *vp;
2784{
2785	char buf[96];
2786
2787	if (label != NULL)
2788		printf("%s: %p: ", label, (void *)vp);
2789	else
2790		printf("%p: ", (void *)vp);
2791	printf("tag %s, type %s, usecount %d, writecount %d, refcount %d,",
2792	    vp->v_tag, typename[vp->v_type], vp->v_usecount,
2793	    vp->v_writecount, vp->v_holdcnt);
2794	buf[0] = '\0';
2795	if (vp->v_vflag & VV_ROOT)
2796		strcat(buf, "|VV_ROOT");
2797	if (vp->v_vflag & VV_TEXT)
2798		strcat(buf, "|VV_TEXT");
2799	if (vp->v_vflag & VV_SYSTEM)
2800		strcat(buf, "|VV_SYSTEM");
2801	if (vp->v_iflag & VI_XLOCK)
2802		strcat(buf, "|VI_XLOCK");
2803	if (vp->v_iflag & VI_XWANT)
2804		strcat(buf, "|VI_XWANT");
2805	if (vp->v_iflag & VI_BWAIT)
2806		strcat(buf, "|VI_BWAIT");
2807	if (vp->v_iflag & VI_DOOMED)
2808		strcat(buf, "|VI_DOOMED");
2809	if (vp->v_iflag & VI_FREE)
2810		strcat(buf, "|VI_FREE");
2811	if (vp->v_vflag & VV_OBJBUF)
2812		strcat(buf, "|VV_OBJBUF");
2813	if (buf[0] != '\0')
2814		printf(" flags (%s),", &buf[1]);
2815	lockmgr_printinfo(vp->v_vnlock);
2816	printf("\n");
2817	if (vp->v_data != NULL) {
2818		printf("\t");
2819		VOP_PRINT(vp);
2820	}
2821}
2822
2823#ifdef DDB
2824#include <ddb/ddb.h>
2825/*
2826 * List all of the locked vnodes in the system.
2827 * Called when debugging the kernel.
2828 */
2829DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2830{
2831	struct thread *td = curthread;	/* XXX */
2832	struct mount *mp, *nmp;
2833	struct vnode *vp;
2834
2835	printf("Locked vnodes\n");
2836	mtx_lock(&mountlist_mtx);
2837	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2838		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
2839			nmp = TAILQ_NEXT(mp, mnt_list);
2840			continue;
2841		}
2842		mtx_lock(&mntvnode_mtx);
2843		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2844			if (VOP_ISLOCKED(vp, NULL))
2845				vprint((char *)0, vp);
2846		}
2847		mtx_unlock(&mntvnode_mtx);
2848		mtx_lock(&mountlist_mtx);
2849		nmp = TAILQ_NEXT(mp, mnt_list);
2850		vfs_unbusy(mp, td);
2851	}
2852	mtx_unlock(&mountlist_mtx);
2853}
2854#endif
2855
2856/*
2857 * Fill in a struct xvfsconf based on a struct vfsconf.
2858 */
2859static void
2860vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2861{
2862
2863	strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2864	xvfsp->vfc_typenum = vfsp->vfc_typenum;
2865	xvfsp->vfc_refcount = vfsp->vfc_refcount;
2866	xvfsp->vfc_flags = vfsp->vfc_flags;
2867	/*
2868	 * These are unused in userland, we keep them
2869	 * to not break binary compatibility.
2870	 */
2871	xvfsp->vfc_vfsops = NULL;
2872	xvfsp->vfc_next = NULL;
2873}
2874
2875static int
2876sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2877{
2878	struct vfsconf *vfsp;
2879	struct xvfsconf *xvfsp;
2880	int cnt, error, i;
2881
2882	cnt = 0;
2883	for (vfsp = vfsconf; vfsp != NULL; vfsp = vfsp->vfc_next)
2884		cnt++;
2885	xvfsp = malloc(sizeof(struct xvfsconf) * cnt, M_TEMP, M_WAITOK);
2886	/*
2887	 * Handle the race that we will have here when struct vfsconf
2888	 * will be locked down by using both cnt and checking vfc_next
2889	 * against NULL to determine the end of the loop.  The race will
2890	 * happen because we will have to unlock before calling malloc().
2891	 * We are protected by Giant for now.
2892	 */
2893	i = 0;
2894	for (vfsp = vfsconf; vfsp != NULL && i < cnt; vfsp = vfsp->vfc_next) {
2895		vfsconf2x(vfsp, xvfsp + i);
2896		i++;
2897	}
2898	error = SYSCTL_OUT(req, xvfsp, sizeof(struct xvfsconf) * i);
2899	free(xvfsp, M_TEMP);
2900	return (error);
2901}
2902
2903SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
2904    "S,xvfsconf", "List of all configured filesystems");
2905
2906/*
2907 * Top level filesystem related information gathering.
2908 */
2909static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2910
2911static int
2912vfs_sysctl(SYSCTL_HANDLER_ARGS)
2913{
2914	int *name = (int *)arg1 - 1;	/* XXX */
2915	u_int namelen = arg2 + 1;	/* XXX */
2916	struct vfsconf *vfsp;
2917	struct xvfsconf xvfsp;
2918
2919	printf("WARNING: userland calling deprecated sysctl, "
2920	    "please rebuild world\n");
2921
2922#if 1 || defined(COMPAT_PRELITE2)
2923	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2924	if (namelen == 1)
2925		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2926#endif
2927
2928	switch (name[1]) {
2929	case VFS_MAXTYPENUM:
2930		if (namelen != 2)
2931			return (ENOTDIR);
2932		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2933	case VFS_CONF:
2934		if (namelen != 3)
2935			return (ENOTDIR);	/* overloaded */
2936		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2937			if (vfsp->vfc_typenum == name[2])
2938				break;
2939		if (vfsp == NULL)
2940			return (EOPNOTSUPP);
2941		vfsconf2x(vfsp, &xvfsp);
2942		return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
2943	}
2944	return (EOPNOTSUPP);
2945}
2946
2947SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP, vfs_sysctl,
2948	"Generic filesystem");
2949
2950#if 1 || defined(COMPAT_PRELITE2)
2951
2952static int
2953sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2954{
2955	int error;
2956	struct vfsconf *vfsp;
2957	struct ovfsconf ovfs;
2958
2959	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2960		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
2961		strcpy(ovfs.vfc_name, vfsp->vfc_name);
2962		ovfs.vfc_index = vfsp->vfc_typenum;
2963		ovfs.vfc_refcount = vfsp->vfc_refcount;
2964		ovfs.vfc_flags = vfsp->vfc_flags;
2965		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2966		if (error)
2967			return error;
2968	}
2969	return 0;
2970}
2971
2972#endif /* 1 || COMPAT_PRELITE2 */
2973
2974#define KINFO_VNODESLOP		10
2975/*
2976 * Dump vnode list (via sysctl).
2977 */
2978/* ARGSUSED */
2979static int
2980sysctl_vnode(SYSCTL_HANDLER_ARGS)
2981{
2982	struct xvnode *xvn;
2983	struct thread *td = req->td;
2984	struct mount *mp;
2985	struct vnode *vp;
2986	int error, len, n;
2987
2988	/*
2989	 * Stale numvnodes access is not fatal here.
2990	 */
2991	req->lock = 0;
2992	len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
2993	if (!req->oldptr)
2994		/* Make an estimate */
2995		return (SYSCTL_OUT(req, 0, len));
2996
2997	sysctl_wire_old_buffer(req, 0);
2998	xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
2999	n = 0;
3000	mtx_lock(&mountlist_mtx);
3001	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3002		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td))
3003			continue;
3004		mtx_lock(&mntvnode_mtx);
3005		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3006			if (n == len)
3007				break;
3008			vref(vp);
3009			xvn[n].xv_size = sizeof *xvn;
3010			xvn[n].xv_vnode = vp;
3011#define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3012			XV_COPY(usecount);
3013			XV_COPY(writecount);
3014			XV_COPY(holdcnt);
3015			XV_COPY(id);
3016			XV_COPY(mount);
3017			XV_COPY(numoutput);
3018			XV_COPY(type);
3019#undef XV_COPY
3020			xvn[n].xv_flag = vp->v_vflag;
3021
3022			switch (vp->v_type) {
3023			case VREG:
3024			case VDIR:
3025			case VLNK:
3026				xvn[n].xv_dev = vp->v_cachedfs;
3027				xvn[n].xv_ino = vp->v_cachedid;
3028				break;
3029			case VBLK:
3030			case VCHR:
3031				if (vp->v_rdev == NULL) {
3032					vrele(vp);
3033					continue;
3034				}
3035				xvn[n].xv_dev = dev2udev(vp->v_rdev);
3036				break;
3037			case VSOCK:
3038				xvn[n].xv_socket = vp->v_socket;
3039				break;
3040			case VFIFO:
3041				xvn[n].xv_fifo = vp->v_fifoinfo;
3042				break;
3043			case VNON:
3044			case VBAD:
3045			default:
3046				/* shouldn't happen? */
3047				vrele(vp);
3048				continue;
3049			}
3050			vrele(vp);
3051			++n;
3052		}
3053		mtx_unlock(&mntvnode_mtx);
3054		mtx_lock(&mountlist_mtx);
3055		vfs_unbusy(mp, td);
3056		if (n == len)
3057			break;
3058	}
3059	mtx_unlock(&mountlist_mtx);
3060
3061	error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3062	free(xvn, M_TEMP);
3063	return (error);
3064}
3065
3066SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3067	0, 0, sysctl_vnode, "S,xvnode", "");
3068
3069/*
3070 * Check to see if a filesystem is mounted on a block device.
3071 */
3072int
3073vfs_mountedon(vp)
3074	struct vnode *vp;
3075{
3076
3077	if (vp->v_rdev->si_mountpoint != NULL)
3078		return (EBUSY);
3079	return (0);
3080}
3081
3082/*
3083 * Unmount all filesystems. The list is traversed in reverse order
3084 * of mounting to avoid dependencies.
3085 */
3086void
3087vfs_unmountall()
3088{
3089	struct mount *mp;
3090	struct thread *td;
3091	int error;
3092
3093	if (curthread != NULL)
3094		td = curthread;
3095	else
3096		td = FIRST_THREAD_IN_PROC(initproc); /* XXX XXX proc0? */
3097	/*
3098	 * Since this only runs when rebooting, it is not interlocked.
3099	 */
3100	while(!TAILQ_EMPTY(&mountlist)) {
3101		mp = TAILQ_LAST(&mountlist, mntlist);
3102		error = dounmount(mp, MNT_FORCE, td);
3103		if (error) {
3104			TAILQ_REMOVE(&mountlist, mp, mnt_list);
3105			printf("unmount of %s failed (",
3106			    mp->mnt_stat.f_mntonname);
3107			if (error == EBUSY)
3108				printf("BUSY)\n");
3109			else
3110				printf("%d)\n", error);
3111		} else {
3112			/* The unmount has removed mp from the mountlist */
3113		}
3114	}
3115}
3116
3117/*
3118 * perform msync on all vnodes under a mount point
3119 * the mount point must be locked.
3120 */
3121void
3122vfs_msync(struct mount *mp, int flags)
3123{
3124	struct vnode *vp, *nvp;
3125	struct vm_object *obj;
3126	int tries;
3127
3128	GIANT_REQUIRED;
3129
3130	tries = 5;
3131	mtx_lock(&mntvnode_mtx);
3132loop:
3133	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) {
3134		if (vp->v_mount != mp) {
3135			if (--tries > 0)
3136				goto loop;
3137			break;
3138		}
3139		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
3140
3141		VI_LOCK(vp);
3142		if (vp->v_iflag & VI_XLOCK) {	/* XXX: what if MNT_WAIT? */
3143			VI_UNLOCK(vp);
3144			continue;
3145		}
3146
3147		if ((vp->v_iflag & VI_OBJDIRTY) &&
3148		    (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
3149			mtx_unlock(&mntvnode_mtx);
3150			if (!vget(vp,
3151			    LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3152			    curthread)) {
3153				if (vp->v_vflag & VV_NOSYNC) {	/* unlinked */
3154					vput(vp);
3155					mtx_lock(&mntvnode_mtx);
3156					continue;
3157				}
3158
3159				if (VOP_GETVOBJECT(vp, &obj) == 0) {
3160					vm_object_page_clean(obj, 0, 0,
3161					    flags == MNT_WAIT ?
3162					    OBJPC_SYNC : OBJPC_NOSYNC);
3163				}
3164				vput(vp);
3165			}
3166			mtx_lock(&mntvnode_mtx);
3167			if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) {
3168				if (--tries > 0)
3169					goto loop;
3170				break;
3171			}
3172		} else
3173			VI_UNLOCK(vp);
3174	}
3175	mtx_unlock(&mntvnode_mtx);
3176}
3177
3178/*
3179 * Create the VM object needed for VMIO and mmap support.  This
3180 * is done for all VREG files in the system.  Some filesystems might
3181 * afford the additional metadata buffering capability of the
3182 * VMIO code by making the device node be VMIO mode also.
3183 *
3184 * vp must be locked when vfs_object_create is called.
3185 */
3186int
3187vfs_object_create(vp, td, cred)
3188	struct vnode *vp;
3189	struct thread *td;
3190	struct ucred *cred;
3191{
3192	GIANT_REQUIRED;
3193	return (VOP_CREATEVOBJECT(vp, cred, td));
3194}
3195
3196/*
3197 * Mark a vnode as free, putting it up for recycling.
3198 */
3199void
3200vfree(vp)
3201	struct vnode *vp;
3202{
3203	int s;
3204
3205	ASSERT_VI_LOCKED(vp, "vfree");
3206	s = splbio();
3207	mtx_lock(&vnode_free_list_mtx);
3208	KASSERT((vp->v_iflag & VI_FREE) == 0, ("vnode already free"));
3209	if (vp->v_iflag & VI_AGE) {
3210		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3211	} else {
3212		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3213	}
3214	freevnodes++;
3215	mtx_unlock(&vnode_free_list_mtx);
3216	vp->v_iflag &= ~VI_AGE;
3217	vp->v_iflag |= VI_FREE;
3218	splx(s);
3219}
3220
3221/*
3222 * Opposite of vfree() - mark a vnode as in use.
3223 */
3224void
3225vbusy(vp)
3226	struct vnode *vp;
3227{
3228	int s;
3229
3230	s = splbio();
3231	ASSERT_VI_LOCKED(vp, "vbusy");
3232	KASSERT((vp->v_iflag & VI_FREE) != 0, ("vnode not free"));
3233
3234	mtx_lock(&vnode_free_list_mtx);
3235	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3236	freevnodes--;
3237	mtx_unlock(&vnode_free_list_mtx);
3238
3239	vp->v_iflag &= ~(VI_FREE|VI_AGE);
3240	splx(s);
3241}
3242
3243/*
3244 * Record a process's interest in events which might happen to
3245 * a vnode.  Because poll uses the historic select-style interface
3246 * internally, this routine serves as both the ``check for any
3247 * pending events'' and the ``record my interest in future events''
3248 * functions.  (These are done together, while the lock is held,
3249 * to avoid race conditions.)
3250 */
3251int
3252vn_pollrecord(vp, td, events)
3253	struct vnode *vp;
3254	struct thread *td;
3255	short events;
3256{
3257
3258	if (vp->v_pollinfo == NULL)
3259		v_addpollinfo(vp);
3260	mtx_lock(&vp->v_pollinfo->vpi_lock);
3261	if (vp->v_pollinfo->vpi_revents & events) {
3262		/*
3263		 * This leaves events we are not interested
3264		 * in available for the other process which
3265		 * which presumably had requested them
3266		 * (otherwise they would never have been
3267		 * recorded).
3268		 */
3269		events &= vp->v_pollinfo->vpi_revents;
3270		vp->v_pollinfo->vpi_revents &= ~events;
3271
3272		mtx_unlock(&vp->v_pollinfo->vpi_lock);
3273		return events;
3274	}
3275	vp->v_pollinfo->vpi_events |= events;
3276	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3277	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3278	return 0;
3279}
3280
3281/*
3282 * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
3283 * it is possible for us to miss an event due to race conditions, but
3284 * that condition is expected to be rare, so for the moment it is the
3285 * preferred interface.
3286 */
3287void
3288vn_pollevent(vp, events)
3289	struct vnode *vp;
3290	short events;
3291{
3292
3293	if (vp->v_pollinfo == NULL)
3294		v_addpollinfo(vp);
3295	mtx_lock(&vp->v_pollinfo->vpi_lock);
3296	if (vp->v_pollinfo->vpi_events & events) {
3297		/*
3298		 * We clear vpi_events so that we don't
3299		 * call selwakeup() twice if two events are
3300		 * posted before the polling process(es) is
3301		 * awakened.  This also ensures that we take at
3302		 * most one selwakeup() if the polling process
3303		 * is no longer interested.  However, it does
3304		 * mean that only one event can be noticed at
3305		 * a time.  (Perhaps we should only clear those
3306		 * event bits which we note?) XXX
3307		 */
3308		vp->v_pollinfo->vpi_events = 0;	/* &= ~events ??? */
3309		vp->v_pollinfo->vpi_revents |= events;
3310		selwakeup(&vp->v_pollinfo->vpi_selinfo);
3311	}
3312	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3313}
3314
3315/*
3316 * Wake up anyone polling on vp because it is being revoked.
3317 * This depends on dead_poll() returning POLLHUP for correct
3318 * behavior.
3319 */
3320void
3321vn_pollgone(vp)
3322	struct vnode *vp;
3323{
3324
3325	mtx_lock(&vp->v_pollinfo->vpi_lock);
3326	VN_KNOTE(vp, NOTE_REVOKE);
3327	if (vp->v_pollinfo->vpi_events) {
3328		vp->v_pollinfo->vpi_events = 0;
3329		selwakeup(&vp->v_pollinfo->vpi_selinfo);
3330	}
3331	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3332}
3333
3334
3335
3336/*
3337 * Routine to create and manage a filesystem syncer vnode.
3338 */
3339#define sync_close ((int (*)(struct  vop_close_args *))nullop)
3340static int	sync_fsync(struct  vop_fsync_args *);
3341static int	sync_inactive(struct  vop_inactive_args *);
3342static int	sync_reclaim(struct  vop_reclaim_args *);
3343static int	sync_print(struct vop_print_args *);
3344
3345static vop_t **sync_vnodeop_p;
3346static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
3347	{ &vop_default_desc,	(vop_t *) vop_eopnotsupp },
3348	{ &vop_close_desc,	(vop_t *) sync_close },		/* close */
3349	{ &vop_fsync_desc,	(vop_t *) sync_fsync },		/* fsync */
3350	{ &vop_inactive_desc,	(vop_t *) sync_inactive },	/* inactive */
3351	{ &vop_reclaim_desc,	(vop_t *) sync_reclaim },	/* reclaim */
3352	{ &vop_lock_desc,	(vop_t *) vop_stdlock },	/* lock */
3353	{ &vop_unlock_desc,	(vop_t *) vop_stdunlock },	/* unlock */
3354	{ &vop_print_desc,	(vop_t *) sync_print },		/* print */
3355	{ &vop_islocked_desc,	(vop_t *) vop_stdislocked },	/* islocked */
3356	{ NULL, NULL }
3357};
3358static struct vnodeopv_desc sync_vnodeop_opv_desc =
3359	{ &sync_vnodeop_p, sync_vnodeop_entries };
3360
3361VNODEOP_SET(sync_vnodeop_opv_desc);
3362
3363/*
3364 * Create a new filesystem syncer vnode for the specified mount point.
3365 */
3366int
3367vfs_allocate_syncvnode(mp)
3368	struct mount *mp;
3369{
3370	struct vnode *vp;
3371	static long start, incr, next;
3372	int error;
3373
3374	/* Allocate a new vnode */
3375	if ((error = getnewvnode("vfs", mp, sync_vnodeop_p, &vp)) != 0) {
3376		mp->mnt_syncer = NULL;
3377		return (error);
3378	}
3379	vp->v_type = VNON;
3380	/*
3381	 * Place the vnode onto the syncer worklist. We attempt to
3382	 * scatter them about on the list so that they will go off
3383	 * at evenly distributed times even if all the filesystems
3384	 * are mounted at once.
3385	 */
3386	next += incr;
3387	if (next == 0 || next > syncer_maxdelay) {
3388		start /= 2;
3389		incr /= 2;
3390		if (start == 0) {
3391			start = syncer_maxdelay / 2;
3392			incr = syncer_maxdelay;
3393		}
3394		next = start;
3395	}
3396	VI_LOCK(vp);
3397	vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
3398	VI_UNLOCK(vp);
3399	mp->mnt_syncer = vp;
3400	return (0);
3401}
3402
3403/*
3404 * Do a lazy sync of the filesystem.
3405 */
3406static int
3407sync_fsync(ap)
3408	struct vop_fsync_args /* {
3409		struct vnode *a_vp;
3410		struct ucred *a_cred;
3411		int a_waitfor;
3412		struct thread *a_td;
3413	} */ *ap;
3414{
3415	struct vnode *syncvp = ap->a_vp;
3416	struct mount *mp = syncvp->v_mount;
3417	struct thread *td = ap->a_td;
3418	int asyncflag;
3419
3420	/*
3421	 * We only need to do something if this is a lazy evaluation.
3422	 */
3423	if (ap->a_waitfor != MNT_LAZY)
3424		return (0);
3425
3426	/*
3427	 * Move ourselves to the back of the sync list.
3428	 */
3429	VI_LOCK(syncvp);
3430	vn_syncer_add_to_worklist(syncvp, syncdelay);
3431	VI_UNLOCK(syncvp);
3432
3433	/*
3434	 * Walk the list of vnodes pushing all that are dirty and
3435	 * not already on the sync list.
3436	 */
3437	mtx_lock(&mountlist_mtx);
3438	if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) {
3439		mtx_unlock(&mountlist_mtx);
3440		return (0);
3441	}
3442	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3443		vfs_unbusy(mp, td);
3444		return (0);
3445	}
3446	asyncflag = mp->mnt_flag & MNT_ASYNC;
3447	mp->mnt_flag &= ~MNT_ASYNC;
3448	vfs_msync(mp, MNT_NOWAIT);
3449	VFS_SYNC(mp, MNT_LAZY, ap->a_cred, td);
3450	if (asyncflag)
3451		mp->mnt_flag |= MNT_ASYNC;
3452	vn_finished_write(mp);
3453	vfs_unbusy(mp, td);
3454	return (0);
3455}
3456
3457/*
3458 * The syncer vnode is no referenced.
3459 */
3460static int
3461sync_inactive(ap)
3462	struct vop_inactive_args /* {
3463		struct vnode *a_vp;
3464		struct thread *a_td;
3465	} */ *ap;
3466{
3467
3468	VOP_UNLOCK(ap->a_vp, 0, ap->a_td);
3469	vgone(ap->a_vp);
3470	return (0);
3471}
3472
3473/*
3474 * The syncer vnode is no longer needed and is being decommissioned.
3475 *
3476 * Modifications to the worklist must be protected at splbio().
3477 */
3478static int
3479sync_reclaim(ap)
3480	struct vop_reclaim_args /* {
3481		struct vnode *a_vp;
3482	} */ *ap;
3483{
3484	struct vnode *vp = ap->a_vp;
3485	int s;
3486
3487	s = splbio();
3488	vp->v_mount->mnt_syncer = NULL;
3489	VI_LOCK(vp);
3490	if (vp->v_iflag & VI_ONWORKLST) {
3491		mtx_lock(&sync_mtx);
3492		LIST_REMOVE(vp, v_synclist);
3493		mtx_unlock(&sync_mtx);
3494		vp->v_iflag &= ~VI_ONWORKLST;
3495	}
3496	VI_UNLOCK(vp);
3497	splx(s);
3498
3499	return (0);
3500}
3501
3502/*
3503 * Print out a syncer vnode.
3504 */
3505static int
3506sync_print(ap)
3507	struct vop_print_args /* {
3508		struct vnode *a_vp;
3509	} */ *ap;
3510{
3511	struct vnode *vp = ap->a_vp;
3512
3513	printf("syncer vnode");
3514	if (vp->v_vnlock != NULL)
3515		lockmgr_printinfo(vp->v_vnlock);
3516	printf("\n");
3517	return (0);
3518}
3519
3520/*
3521 * extract the dev_t from a VCHR
3522 */
3523dev_t
3524vn_todev(vp)
3525	struct vnode *vp;
3526{
3527	if (vp->v_type != VCHR)
3528		return (NODEV);
3529	return (vp->v_rdev);
3530}
3531
3532/*
3533 * Check if vnode represents a disk device
3534 */
3535int
3536vn_isdisk(vp, errp)
3537	struct vnode *vp;
3538	int *errp;
3539{
3540	struct cdevsw *cdevsw;
3541
3542	if (vp->v_type != VCHR) {
3543		if (errp != NULL)
3544			*errp = ENOTBLK;
3545		return (0);
3546	}
3547	if (vp->v_rdev == NULL) {
3548		if (errp != NULL)
3549			*errp = ENXIO;
3550		return (0);
3551	}
3552	cdevsw = devsw(vp->v_rdev);
3553	if (cdevsw == NULL) {
3554		if (errp != NULL)
3555			*errp = ENXIO;
3556		return (0);
3557	}
3558	if (!(cdevsw->d_flags & D_DISK)) {
3559		if (errp != NULL)
3560			*errp = ENOTBLK;
3561		return (0);
3562	}
3563	if (errp != NULL)
3564		*errp = 0;
3565	return (1);
3566}
3567
3568/*
3569 * Free data allocated by namei(); see namei(9) for details.
3570 */
3571void
3572NDFREE(ndp, flags)
3573     struct nameidata *ndp;
3574     const uint flags;
3575{
3576	if (!(flags & NDF_NO_FREE_PNBUF) &&
3577	    (ndp->ni_cnd.cn_flags & HASBUF)) {
3578		uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
3579		ndp->ni_cnd.cn_flags &= ~HASBUF;
3580	}
3581	if (!(flags & NDF_NO_DVP_UNLOCK) &&
3582	    (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
3583	    ndp->ni_dvp != ndp->ni_vp)
3584		VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_thread);
3585	if (!(flags & NDF_NO_DVP_RELE) &&
3586	    (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
3587		vrele(ndp->ni_dvp);
3588		ndp->ni_dvp = NULL;
3589	}
3590	if (!(flags & NDF_NO_VP_UNLOCK) &&
3591	    (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
3592		VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_thread);
3593	if (!(flags & NDF_NO_VP_RELE) &&
3594	    ndp->ni_vp) {
3595		vrele(ndp->ni_vp);
3596		ndp->ni_vp = NULL;
3597	}
3598	if (!(flags & NDF_NO_STARTDIR_RELE) &&
3599	    (ndp->ni_cnd.cn_flags & SAVESTART)) {
3600		vrele(ndp->ni_startdir);
3601		ndp->ni_startdir = NULL;
3602	}
3603}
3604
3605/*
3606 * Common filesystem object access control check routine.  Accepts a
3607 * vnode's type, "mode", uid and gid, requested access mode, credentials,
3608 * and optional call-by-reference privused argument allowing vaccess()
3609 * to indicate to the caller whether privilege was used to satisfy the
3610 * request (obsoleted).  Returns 0 on success, or an errno on failure.
3611 */
3612int
3613vaccess(type, file_mode, file_uid, file_gid, acc_mode, cred, privused)
3614	enum vtype type;
3615	mode_t file_mode;
3616	uid_t file_uid;
3617	gid_t file_gid;
3618	mode_t acc_mode;
3619	struct ucred *cred;
3620	int *privused;
3621{
3622	mode_t dac_granted;
3623#ifdef CAPABILITIES
3624	mode_t cap_granted;
3625#endif
3626
3627	/*
3628	 * Look for a normal, non-privileged way to access the file/directory
3629	 * as requested.  If it exists, go with that.
3630	 */
3631
3632	if (privused != NULL)
3633		*privused = 0;
3634
3635	dac_granted = 0;
3636
3637	/* Check the owner. */
3638	if (cred->cr_uid == file_uid) {
3639		dac_granted |= VADMIN;
3640		if (file_mode & S_IXUSR)
3641			dac_granted |= VEXEC;
3642		if (file_mode & S_IRUSR)
3643			dac_granted |= VREAD;
3644		if (file_mode & S_IWUSR)
3645			dac_granted |= (VWRITE | VAPPEND);
3646
3647		if ((acc_mode & dac_granted) == acc_mode)
3648			return (0);
3649
3650		goto privcheck;
3651	}
3652
3653	/* Otherwise, check the groups (first match) */
3654	if (groupmember(file_gid, cred)) {
3655		if (file_mode & S_IXGRP)
3656			dac_granted |= VEXEC;
3657		if (file_mode & S_IRGRP)
3658			dac_granted |= VREAD;
3659		if (file_mode & S_IWGRP)
3660			dac_granted |= (VWRITE | VAPPEND);
3661
3662		if ((acc_mode & dac_granted) == acc_mode)
3663			return (0);
3664
3665		goto privcheck;
3666	}
3667
3668	/* Otherwise, check everyone else. */
3669	if (file_mode & S_IXOTH)
3670		dac_granted |= VEXEC;
3671	if (file_mode & S_IROTH)
3672		dac_granted |= VREAD;
3673	if (file_mode & S_IWOTH)
3674		dac_granted |= (VWRITE | VAPPEND);
3675	if ((acc_mode & dac_granted) == acc_mode)
3676		return (0);
3677
3678privcheck:
3679	if (!suser_cred(cred, PRISON_ROOT)) {
3680		/* XXX audit: privilege used */
3681		if (privused != NULL)
3682			*privused = 1;
3683		return (0);
3684	}
3685
3686#ifdef CAPABILITIES
3687	/*
3688	 * Build a capability mask to determine if the set of capabilities
3689	 * satisfies the requirements when combined with the granted mask
3690	 * from above.
3691	 * For each capability, if the capability is required, bitwise
3692	 * or the request type onto the cap_granted mask.
3693	 */
3694	cap_granted = 0;
3695
3696	if (type == VDIR) {
3697		/*
3698		 * For directories, use CAP_DAC_READ_SEARCH to satisfy
3699		 * VEXEC requests, instead of CAP_DAC_EXECUTE.
3700		 */
3701		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3702		    !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3703			cap_granted |= VEXEC;
3704	} else {
3705		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3706		    !cap_check(cred, NULL, CAP_DAC_EXECUTE, PRISON_ROOT))
3707			cap_granted |= VEXEC;
3708	}
3709
3710	if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
3711	    !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3712		cap_granted |= VREAD;
3713
3714	if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3715	    !cap_check(cred, NULL, CAP_DAC_WRITE, PRISON_ROOT))
3716		cap_granted |= (VWRITE | VAPPEND);
3717
3718	if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3719	    !cap_check(cred, NULL, CAP_FOWNER, PRISON_ROOT))
3720		cap_granted |= VADMIN;
3721
3722	if ((acc_mode & (cap_granted | dac_granted)) == acc_mode) {
3723		/* XXX audit: privilege used */
3724		if (privused != NULL)
3725			*privused = 1;
3726		return (0);
3727	}
3728#endif
3729
3730	return ((acc_mode & VADMIN) ? EPERM : EACCES);
3731}
3732
3733/*
3734 * Credential check based on process requesting service, and per-attribute
3735 * permissions.
3736 */
3737int
3738extattr_check_cred(struct vnode *vp, int attrnamespace,
3739    struct ucred *cred, struct thread *td, int access)
3740{
3741
3742	/*
3743	 * Kernel-invoked always succeeds.
3744	 */
3745	if (cred == NOCRED)
3746		return (0);
3747
3748	/*
3749	 * Do not allow privileged processes in jail to directly
3750	 * manipulate system attributes.
3751	 *
3752	 * XXX What capability should apply here?
3753	 * Probably CAP_SYS_SETFFLAG.
3754	 */
3755	switch (attrnamespace) {
3756	case EXTATTR_NAMESPACE_SYSTEM:
3757		/* Potentially should be: return (EPERM); */
3758		return (suser_cred(cred, 0));
3759	case EXTATTR_NAMESPACE_USER:
3760		return (VOP_ACCESS(vp, access, cred, td));
3761	default:
3762		return (EPERM);
3763	}
3764}
3765