xref: /freebsd/sys/kern/vfs_subr.c (revision eacee0ff7ec955b32e09515246bd97b6edcd2b0f)
1 /*
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)vfs_subr.c	8.31 (Berkeley) 5/26/95
39  * $FreeBSD$
40  */
41 
42 /*
43  * External virtual filesystem routines
44  */
45 #include "opt_ddb.h"
46 #include "opt_ffs.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/bio.h>
51 #include <sys/buf.h>
52 #include <sys/conf.h>
53 #include <sys/eventhandler.h>
54 #include <sys/fcntl.h>
55 #include <sys/kernel.h>
56 #include <sys/kthread.h>
57 #include <sys/malloc.h>
58 #include <sys/mount.h>
59 #include <sys/namei.h>
60 #include <sys/stat.h>
61 #include <sys/sysctl.h>
62 #include <sys/syslog.h>
63 #include <sys/vmmeter.h>
64 #include <sys/vnode.h>
65 
66 #include <vm/vm.h>
67 #include <vm/vm_object.h>
68 #include <vm/vm_extern.h>
69 #include <vm/pmap.h>
70 #include <vm/vm_map.h>
71 #include <vm/vm_page.h>
72 #include <vm/vm_zone.h>
73 
74 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
75 
76 static void	addalias __P((struct vnode *vp, dev_t nvp_rdev));
77 static void	insmntque __P((struct vnode *vp, struct mount *mp));
78 static void	vclean __P((struct vnode *vp, int flags, struct thread *td));
79 static void	vlruvp(struct vnode *vp);
80 
81 /*
82  * Number of vnodes in existence.  Increased whenever getnewvnode()
83  * allocates a new vnode, never decreased.
84  */
85 static unsigned long	numvnodes;
86 
87 SYSCTL_LONG(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
88 
89 /*
90  * Conversion tables for conversion from vnode types to inode formats
91  * and back.
92  */
93 enum vtype iftovt_tab[16] = {
94 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
95 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
96 };
97 int vttoif_tab[9] = {
98 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
99 	S_IFSOCK, S_IFIFO, S_IFMT,
100 };
101 
102 /*
103  * List of vnodes that are ready for recycling.
104  */
105 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
106 
107 /*
108  * Minimum number of free vnodes.  If there are fewer than this free vnodes,
109  * getnewvnode() will return a newly allocated vnode.
110  */
111 static u_long wantfreevnodes = 25;
112 SYSCTL_LONG(_debug, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
113 /* Number of vnodes in the free list. */
114 static u_long freevnodes;
115 SYSCTL_LONG(_debug, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
116 
117 #if 0
118 /* Number of vnode allocation. */
119 static u_long vnodeallocs;
120 SYSCTL_LONG(_debug, OID_AUTO, vnodeallocs, CTLFLAG_RD, &vnodeallocs, 0, "");
121 /* Period of vnode recycle from namecache in vnode allocation times. */
122 static u_long vnoderecycleperiod = 1000;
123 SYSCTL_LONG(_debug, OID_AUTO, vnoderecycleperiod, CTLFLAG_RW, &vnoderecycleperiod, 0, "");
124 /* Minimum number of total vnodes required to invoke vnode recycle from namecache. */
125 static u_long vnoderecyclemintotalvn = 2000;
126 SYSCTL_LONG(_debug, OID_AUTO, vnoderecyclemintotalvn, CTLFLAG_RW, &vnoderecyclemintotalvn, 0, "");
127 /* Minimum number of free vnodes required to invoke vnode recycle from namecache. */
128 static u_long vnoderecycleminfreevn = 2000;
129 SYSCTL_LONG(_debug, OID_AUTO, vnoderecycleminfreevn, CTLFLAG_RW, &vnoderecycleminfreevn, 0, "");
130 /* Number of vnodes attempted to recycle at a time. */
131 static u_long vnoderecyclenumber = 3000;
132 SYSCTL_LONG(_debug, OID_AUTO, vnoderecyclenumber, CTLFLAG_RW, &vnoderecyclenumber, 0, "");
133 #endif
134 
135 /*
136  * Various variables used for debugging the new implementation of
137  * reassignbuf().
138  * XXX these are probably of (very) limited utility now.
139  */
140 static int reassignbufcalls;
141 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
142 static int reassignbufloops;
143 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW, &reassignbufloops, 0, "");
144 static int reassignbufsortgood;
145 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW, &reassignbufsortgood, 0, "");
146 static int reassignbufsortbad;
147 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW, &reassignbufsortbad, 0, "");
148 /* Set to 0 for old insertion-sort based reassignbuf, 1 for modern method. */
149 static int reassignbufmethod = 1;
150 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW, &reassignbufmethod, 0, "");
151 static int nameileafonly;
152 SYSCTL_INT(_vfs, OID_AUTO, nameileafonly, CTLFLAG_RW, &nameileafonly, 0, "");
153 
154 #ifdef ENABLE_VFS_IOOPT
155 /* See NOTES for a description of this setting. */
156 int vfs_ioopt;
157 SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, "");
158 #endif
159 
160 /* List of mounted filesystems. */
161 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
162 
163 /* For any iteration/modification of mountlist */
164 struct mtx mountlist_mtx;
165 
166 /* For any iteration/modification of mnt_vnodelist */
167 struct mtx mntvnode_mtx;
168 
169 /*
170  * Cache for the mount type id assigned to NFS.  This is used for
171  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
172  */
173 int	nfs_mount_type = -1;
174 
175 /* To keep more than one thread at a time from running vfs_getnewfsid */
176 static struct mtx mntid_mtx;
177 
178 /* For any iteration/modification of vnode_free_list */
179 static struct mtx vnode_free_list_mtx;
180 
181 /*
182  * For any iteration/modification of dev->si_hlist (linked through
183  * v_specnext)
184  */
185 static struct mtx spechash_mtx;
186 
187 /* Publicly exported FS */
188 struct nfs_public nfs_pub;
189 
190 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
191 static vm_zone_t vnode_zone;
192 static vm_zone_t vnodepoll_zone;
193 
194 /* Set to 1 to print out reclaim of active vnodes */
195 int	prtactive;
196 
197 /*
198  * The workitem queue.
199  *
200  * It is useful to delay writes of file data and filesystem metadata
201  * for tens of seconds so that quickly created and deleted files need
202  * not waste disk bandwidth being created and removed. To realize this,
203  * we append vnodes to a "workitem" queue. When running with a soft
204  * updates implementation, most pending metadata dependencies should
205  * not wait for more than a few seconds. Thus, mounted on block devices
206  * are delayed only about a half the time that file data is delayed.
207  * Similarly, directory updates are more critical, so are only delayed
208  * about a third the time that file data is delayed. Thus, there are
209  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
210  * one each second (driven off the filesystem syncer process). The
211  * syncer_delayno variable indicates the next queue that is to be processed.
212  * Items that need to be processed soon are placed in this queue:
213  *
214  *	syncer_workitem_pending[syncer_delayno]
215  *
216  * A delay of fifteen seconds is done by placing the request fifteen
217  * entries later in the queue:
218  *
219  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
220  *
221  */
222 static int syncer_delayno;
223 static long syncer_mask;
224 LIST_HEAD(synclist, vnode);
225 static struct synclist *syncer_workitem_pending;
226 
227 #define SYNCER_MAXDELAY		32
228 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
229 static int syncdelay = 30;		/* max time to delay syncing data */
230 static int filedelay = 30;		/* time to delay syncing files */
231 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
232 static int dirdelay = 29;		/* time to delay syncing directories */
233 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
234 static int metadelay = 28;		/* time to delay syncing metadata */
235 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
236 static int rushjob;		/* number of slots to run ASAP */
237 static int stat_rush_requests;	/* number of times I/O speeded up */
238 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
239 
240 /*
241  * Number of vnodes we want to exist at any one time.  This is mostly used
242  * to size hash tables in vnode-related code.  It is normally not used in
243  * getnewvnode(), as wantfreevnodes is normally nonzero.)
244  *
245  * XXX desiredvnodes is historical cruft and should not exist.
246  */
247 int desiredvnodes;
248 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
249     &desiredvnodes, 0, "Maximum number of vnodes");
250 static int minvnodes;
251 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
252     &minvnodes, 0, "Minimum number of vnodes");
253 static int vnlru_nowhere;
254 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0,
255     "Number of times the vnlru process ran without success");
256 
257 void
258 v_addpollinfo(struct vnode *vp)
259 {
260 	vp->v_pollinfo = zalloc(vnodepoll_zone);
261 	mtx_init(&vp->v_pollinfo->vpi_lock, "vnode pollinfo", MTX_DEF);
262 }
263 
264 /*
265  * Initialize the vnode management data structures.
266  */
267 static void
268 vntblinit(void *dummy __unused)
269 {
270 
271 	desiredvnodes = maxproc + cnt.v_page_count / 4;
272 	minvnodes = desiredvnodes / 4;
273 	mtx_init(&mountlist_mtx, "mountlist", MTX_DEF);
274 	mtx_init(&mntvnode_mtx, "mntvnode", MTX_DEF);
275 	mtx_init(&mntid_mtx, "mntid", MTX_DEF);
276 	mtx_init(&spechash_mtx, "spechash", MTX_DEF);
277 	TAILQ_INIT(&vnode_free_list);
278 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", MTX_DEF);
279 	vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5);
280 	vnodepoll_zone = zinit("VNODEPOLL", sizeof (struct vpollinfo), 0, 0, 5);
281 	/*
282 	 * Initialize the filesystem syncer.
283 	 */
284 	syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
285 		&syncer_mask);
286 	syncer_maxdelay = syncer_mask + 1;
287 }
288 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL)
289 
290 
291 /*
292  * Mark a mount point as busy. Used to synchronize access and to delay
293  * unmounting. Interlock is not released on failure.
294  */
295 int
296 vfs_busy(mp, flags, interlkp, td)
297 	struct mount *mp;
298 	int flags;
299 	struct mtx *interlkp;
300 	struct thread *td;
301 {
302 	int lkflags;
303 
304 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
305 		if (flags & LK_NOWAIT)
306 			return (ENOENT);
307 		mp->mnt_kern_flag |= MNTK_MWAIT;
308 		/*
309 		 * Since all busy locks are shared except the exclusive
310 		 * lock granted when unmounting, the only place that a
311 		 * wakeup needs to be done is at the release of the
312 		 * exclusive lock at the end of dounmount.
313 		 */
314 		msleep((caddr_t)mp, interlkp, PVFS, "vfs_busy", 0);
315 		return (ENOENT);
316 	}
317 	lkflags = LK_SHARED | LK_NOPAUSE;
318 	if (interlkp)
319 		lkflags |= LK_INTERLOCK;
320 	if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
321 		panic("vfs_busy: unexpected lock failure");
322 	return (0);
323 }
324 
325 /*
326  * Free a busy filesystem.
327  */
328 void
329 vfs_unbusy(mp, td)
330 	struct mount *mp;
331 	struct thread *td;
332 {
333 
334 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
335 }
336 
337 /*
338  * Lookup a filesystem type, and if found allocate and initialize
339  * a mount structure for it.
340  *
341  * Devname is usually updated by mount(8) after booting.
342  */
343 int
344 vfs_rootmountalloc(fstypename, devname, mpp)
345 	char *fstypename;
346 	char *devname;
347 	struct mount **mpp;
348 {
349 	struct thread *td = curthread;	/* XXX */
350 	struct vfsconf *vfsp;
351 	struct mount *mp;
352 
353 	if (fstypename == NULL)
354 		return (ENODEV);
355 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
356 		if (!strcmp(vfsp->vfc_name, fstypename))
357 			break;
358 	if (vfsp == NULL)
359 		return (ENODEV);
360 	mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO);
361 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE);
362 	(void)vfs_busy(mp, LK_NOWAIT, 0, td);
363 	TAILQ_INIT(&mp->mnt_nvnodelist);
364 	TAILQ_INIT(&mp->mnt_reservedvnlist);
365 	mp->mnt_vfc = vfsp;
366 	mp->mnt_op = vfsp->vfc_vfsops;
367 	mp->mnt_flag = MNT_RDONLY;
368 	mp->mnt_vnodecovered = NULLVP;
369 	vfsp->vfc_refcount++;
370 	mp->mnt_iosize_max = DFLTPHYS;
371 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
372 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
373 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
374 	mp->mnt_stat.f_mntonname[0] = '/';
375 	mp->mnt_stat.f_mntonname[1] = 0;
376 	(void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
377 	*mpp = mp;
378 	return (0);
379 }
380 
381 /*
382  * Find an appropriate filesystem to use for the root. If a filesystem
383  * has not been preselected, walk through the list of known filesystems
384  * trying those that have mountroot routines, and try them until one
385  * works or we have tried them all.
386  */
387 #ifdef notdef	/* XXX JH */
388 int
389 lite2_vfs_mountroot()
390 {
391 	struct vfsconf *vfsp;
392 	extern int (*lite2_mountroot) __P((void));
393 	int error;
394 
395 	if (lite2_mountroot != NULL)
396 		return ((*lite2_mountroot)());
397 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
398 		if (vfsp->vfc_mountroot == NULL)
399 			continue;
400 		if ((error = (*vfsp->vfc_mountroot)()) == 0)
401 			return (0);
402 		printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error);
403 	}
404 	return (ENODEV);
405 }
406 #endif
407 
408 /*
409  * Lookup a mount point by filesystem identifier.
410  */
411 struct mount *
412 vfs_getvfs(fsid)
413 	fsid_t *fsid;
414 {
415 	register struct mount *mp;
416 
417 	mtx_lock(&mountlist_mtx);
418 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
419 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
420 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
421 			mtx_unlock(&mountlist_mtx);
422 			return (mp);
423 	    }
424 	}
425 	mtx_unlock(&mountlist_mtx);
426 	return ((struct mount *) 0);
427 }
428 
429 /*
430  * Get a new unique fsid.  Try to make its val[0] unique, since this value
431  * will be used to create fake device numbers for stat().  Also try (but
432  * not so hard) make its val[0] unique mod 2^16, since some emulators only
433  * support 16-bit device numbers.  We end up with unique val[0]'s for the
434  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
435  *
436  * Keep in mind that several mounts may be running in parallel.  Starting
437  * the search one past where the previous search terminated is both a
438  * micro-optimization and a defense against returning the same fsid to
439  * different mounts.
440  */
441 void
442 vfs_getnewfsid(mp)
443 	struct mount *mp;
444 {
445 	static u_int16_t mntid_base;
446 	fsid_t tfsid;
447 	int mtype;
448 
449 	mtx_lock(&mntid_mtx);
450 	mtype = mp->mnt_vfc->vfc_typenum;
451 	tfsid.val[1] = mtype;
452 	mtype = (mtype & 0xFF) << 24;
453 	for (;;) {
454 		tfsid.val[0] = makeudev(255,
455 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
456 		mntid_base++;
457 		if (vfs_getvfs(&tfsid) == NULL)
458 			break;
459 	}
460 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
461 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
462 	mtx_unlock(&mntid_mtx);
463 }
464 
465 /*
466  * Knob to control the precision of file timestamps:
467  *
468  *   0 = seconds only; nanoseconds zeroed.
469  *   1 = seconds and nanoseconds, accurate within 1/HZ.
470  *   2 = seconds and nanoseconds, truncated to microseconds.
471  * >=3 = seconds and nanoseconds, maximum precision.
472  */
473 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
474 
475 static int timestamp_precision = TSP_SEC;
476 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
477     &timestamp_precision, 0, "");
478 
479 /*
480  * Get a current timestamp.
481  */
482 void
483 vfs_timestamp(tsp)
484 	struct timespec *tsp;
485 {
486 	struct timeval tv;
487 
488 	switch (timestamp_precision) {
489 	case TSP_SEC:
490 		tsp->tv_sec = time_second;
491 		tsp->tv_nsec = 0;
492 		break;
493 	case TSP_HZ:
494 		getnanotime(tsp);
495 		break;
496 	case TSP_USEC:
497 		microtime(&tv);
498 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
499 		break;
500 	case TSP_NSEC:
501 	default:
502 		nanotime(tsp);
503 		break;
504 	}
505 }
506 
507 /*
508  * Set vnode attributes to VNOVAL
509  */
510 void
511 vattr_null(vap)
512 	register struct vattr *vap;
513 {
514 
515 	vap->va_type = VNON;
516 	vap->va_size = VNOVAL;
517 	vap->va_bytes = VNOVAL;
518 	vap->va_mode = VNOVAL;
519 	vap->va_nlink = VNOVAL;
520 	vap->va_uid = VNOVAL;
521 	vap->va_gid = VNOVAL;
522 	vap->va_fsid = VNOVAL;
523 	vap->va_fileid = VNOVAL;
524 	vap->va_blocksize = VNOVAL;
525 	vap->va_rdev = VNOVAL;
526 	vap->va_atime.tv_sec = VNOVAL;
527 	vap->va_atime.tv_nsec = VNOVAL;
528 	vap->va_mtime.tv_sec = VNOVAL;
529 	vap->va_mtime.tv_nsec = VNOVAL;
530 	vap->va_ctime.tv_sec = VNOVAL;
531 	vap->va_ctime.tv_nsec = VNOVAL;
532 	vap->va_flags = VNOVAL;
533 	vap->va_gen = VNOVAL;
534 	vap->va_vaflags = 0;
535 }
536 
537 /*
538  * This routine is called when we have too many vnodes.  It attempts
539  * to free <count> vnodes and will potentially free vnodes that still
540  * have VM backing store (VM backing store is typically the cause
541  * of a vnode blowout so we want to do this).  Therefore, this operation
542  * is not considered cheap.
543  *
544  * A number of conditions may prevent a vnode from being reclaimed.
545  * the buffer cache may have references on the vnode, a directory
546  * vnode may still have references due to the namei cache representing
547  * underlying files, or the vnode may be in active use.   It is not
548  * desireable to reuse such vnodes.  These conditions may cause the
549  * number of vnodes to reach some minimum value regardless of what
550  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
551  */
552 static int
553 vlrureclaim(struct mount *mp, int count)
554 {
555 	struct vnode *vp;
556 	int done;
557 	int trigger;
558 	int usevnodes;
559 
560 	/*
561 	 * Calculate the trigger point, don't allow user
562 	 * screwups to blow us up.   This prevents us from
563 	 * recycling vnodes with lots of resident pages.  We
564 	 * aren't trying to free memory, we are trying to
565 	 * free vnodes.
566 	 */
567 	usevnodes = desiredvnodes;
568 	if (usevnodes <= 0)
569 		usevnodes = 1;
570 	trigger = cnt.v_page_count * 2 / usevnodes;
571 
572 	done = 0;
573 	mtx_lock(&mntvnode_mtx);
574 	while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
575 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
576 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
577 
578 		if (vp->v_type != VNON &&
579 		    vp->v_type != VBAD &&
580 		    VMIGHTFREE(vp) &&           /* critical path opt */
581 		    (vp->v_object == NULL || vp->v_object->resident_page_count < trigger) &&
582 		    mtx_trylock(&vp->v_interlock)
583 		) {
584 			mtx_unlock(&mntvnode_mtx);
585 			if (VMIGHTFREE(vp)) {
586 				vgonel(vp, curthread);
587 				done++;
588 			} else {
589 				mtx_unlock(&vp->v_interlock);
590 			}
591 			mtx_lock(&mntvnode_mtx);
592 		}
593 		--count;
594 	}
595 	mtx_unlock(&mntvnode_mtx);
596 	return done;
597 }
598 
599 /*
600  * Attempt to recycle vnodes in a context that is always safe to block.
601  * Calling vlrurecycle() from the bowels of file system code has some
602  * interesting deadlock problems.
603  */
604 static struct proc *vnlruproc;
605 static int vnlruproc_sig;
606 
607 static void
608 vnlru_proc(void)
609 {
610 	struct mount *mp, *nmp;
611 	int s;
612 	int done;
613 	struct proc *p = vnlruproc;
614 	struct thread *td = FIRST_THREAD_IN_PROC(p);	/* XXXKSE */
615 
616 	mtx_lock(&Giant);
617 
618 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
619 	    SHUTDOWN_PRI_FIRST);
620 
621 	s = splbio();
622 	for (;;) {
623 		kthread_suspend_check(p);
624 		if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
625 			vnlruproc_sig = 0;
626 			tsleep(vnlruproc, PVFS, "vlruwt", 0);
627 			continue;
628 		}
629 		done = 0;
630 		mtx_lock(&mountlist_mtx);
631 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
632 			if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
633 				nmp = TAILQ_NEXT(mp, mnt_list);
634 				continue;
635 			}
636 			done += vlrureclaim(mp, 10);
637 			mtx_lock(&mountlist_mtx);
638 			nmp = TAILQ_NEXT(mp, mnt_list);
639 			vfs_unbusy(mp, td);
640 		}
641 		mtx_unlock(&mountlist_mtx);
642 		if (done == 0) {
643 #if 0
644 			/* These messages are temporary debugging aids */
645 			if (vnlru_nowhere < 5)
646 				printf("vnlru process getting nowhere..\n");
647 			else if (vnlru_nowhere == 5)
648 				printf("vnlru process messages stopped.\n");
649 #endif
650 			vnlru_nowhere++;
651 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
652 		}
653 	}
654 	splx(s);
655 }
656 
657 static struct kproc_desc vnlru_kp = {
658 	"vnlru",
659 	vnlru_proc,
660 	&vnlruproc
661 };
662 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
663 
664 
665 /*
666  * Routines having to do with the management of the vnode table.
667  */
668 
669 /*
670  * Return the next vnode from the free list.
671  */
672 int
673 getnewvnode(tag, mp, vops, vpp)
674 	enum vtagtype tag;
675 	struct mount *mp;
676 	vop_t **vops;
677 	struct vnode **vpp;
678 {
679 	int s;
680 	struct thread *td = curthread;	/* XXX */
681 	struct vnode *vp = NULL;
682 	struct mount *vnmp;
683 	vm_object_t object;
684 
685 	s = splbio();
686 	/*
687 	 * Try to reuse vnodes if we hit the max.  This situation only
688 	 * occurs in certain large-memory (2G+) situations.  We cannot
689 	 * attempt to directly reclaim vnodes due to nasty recursion
690 	 * problems.
691 	 */
692 	if (vnlruproc_sig == 0 && numvnodes - freevnodes > desiredvnodes) {
693 		vnlruproc_sig = 1;      /* avoid unnecessary wakeups */
694 		wakeup(vnlruproc);
695 	}
696 
697 	/*
698 	 * Attempt to reuse a vnode already on the free list, allocating
699 	 * a new vnode if we can't find one or if we have not reached a
700 	 * good minimum for good LRU performance.
701 	 */
702 
703 	mtx_lock(&vnode_free_list_mtx);
704 
705 	if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
706 		int count;
707 
708 		for (count = 0; count < freevnodes; count++) {
709 			vp = TAILQ_FIRST(&vnode_free_list);
710 			if (vp == NULL || vp->v_usecount)
711 				panic("getnewvnode: free vnode isn't");
712 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
713 
714 			/*
715 			 * Don't recycle if we still have cached pages or if
716 			 * we cannot get the interlock.
717 			 */
718 			if ((VOP_GETVOBJECT(vp, &object) == 0 &&
719 			     (object->resident_page_count ||
720 			      object->ref_count)) ||
721 			     !mtx_trylock(&vp->v_interlock)) {
722 				TAILQ_INSERT_TAIL(&vnode_free_list, vp,
723 						    v_freelist);
724 				vp = NULL;
725 				continue;
726 			}
727 			if (LIST_FIRST(&vp->v_cache_src)) {
728 				/*
729 				 * note: nameileafonly sysctl is temporary,
730 				 * for debugging only, and will eventually be
731 				 * removed.
732 				 */
733 				if (nameileafonly > 0) {
734 					/*
735 					 * Do not reuse namei-cached directory
736 					 * vnodes that have cached
737 					 * subdirectories.
738 					 */
739 					if (cache_leaf_test(vp) < 0) {
740 						mtx_unlock(&vp->v_interlock);
741 						TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
742 						vp = NULL;
743 						continue;
744 					}
745 				} else if (nameileafonly < 0 ||
746 					    vmiodirenable == 0) {
747 					/*
748 					 * Do not reuse namei-cached directory
749 					 * vnodes if nameileafonly is -1 or
750 					 * if VMIO backing for directories is
751 					 * turned off (otherwise we reuse them
752 					 * too quickly).
753 					 */
754 					mtx_unlock(&vp->v_interlock);
755 					TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
756 					vp = NULL;
757 					continue;
758 				}
759 			}
760 			/*
761 			 * Skip over it if its filesystem is being suspended.
762 			 */
763 			if (vn_start_write(vp, &vnmp, V_NOWAIT) == 0)
764 				break;
765 			mtx_unlock(&vp->v_interlock);
766 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
767 			vp = NULL;
768 		}
769 	}
770 	if (vp) {
771 		vp->v_flag |= VDOOMED;
772 		vp->v_flag &= ~VFREE;
773 		freevnodes--;
774 		mtx_unlock(&vnode_free_list_mtx);
775 		cache_purge(vp);
776 		if (vp->v_type != VBAD) {
777 			vgonel(vp, td);
778 		} else {
779 			mtx_unlock(&vp->v_interlock);
780 		}
781 		vn_finished_write(vnmp);
782 
783 #ifdef INVARIANTS
784 		{
785 			int s;
786 
787 			if (vp->v_data)
788 				panic("cleaned vnode isn't");
789 			s = splbio();
790 			if (vp->v_numoutput)
791 				panic("Clean vnode has pending I/O's");
792 			splx(s);
793 			if (vp->v_writecount != 0)
794 				panic("Non-zero write count");
795 		}
796 #endif
797 		if (vp->v_pollinfo) {
798 			mtx_destroy(&vp->v_pollinfo->vpi_lock);
799 			zfree(vnodepoll_zone, vp->v_pollinfo);
800 		}
801 		vp->v_pollinfo = NULL;
802 		vp->v_flag = 0;
803 		vp->v_lastw = 0;
804 		vp->v_lasta = 0;
805 		vp->v_cstart = 0;
806 		vp->v_clen = 0;
807 		vp->v_socket = 0;
808 	} else {
809 		mtx_unlock(&vnode_free_list_mtx);
810 		vp = (struct vnode *) zalloc(vnode_zone);
811 		bzero((char *) vp, sizeof *vp);
812 		mtx_init(&vp->v_interlock, "vnode interlock", MTX_DEF);
813 		vp->v_dd = vp;
814 		cache_purge(vp);
815 		LIST_INIT(&vp->v_cache_src);
816 		TAILQ_INIT(&vp->v_cache_dst);
817 		numvnodes++;
818 	}
819 
820 	TAILQ_INIT(&vp->v_cleanblkhd);
821 	TAILQ_INIT(&vp->v_dirtyblkhd);
822 	vp->v_type = VNON;
823 	vp->v_tag = tag;
824 	vp->v_op = vops;
825 	lockinit(&vp->v_lock, PVFS, "vnlock", VLKTIMEOUT, LK_NOPAUSE);
826 	insmntque(vp, mp);
827 	*vpp = vp;
828 	vp->v_usecount = 1;
829 	vp->v_data = 0;
830 
831 	splx(s);
832 
833 	vfs_object_create(vp, td, td->td_proc->p_ucred);
834 
835 #if 0
836 	vnodeallocs++;
837 	if (vnodeallocs % vnoderecycleperiod == 0 &&
838 	    freevnodes < vnoderecycleminfreevn &&
839 	    vnoderecyclemintotalvn < numvnodes) {
840 		/* Recycle vnodes. */
841 		cache_purgeleafdirs(vnoderecyclenumber);
842 	}
843 #endif
844 
845 	return (0);
846 }
847 
848 /*
849  * Move a vnode from one mount queue to another.
850  */
851 static void
852 insmntque(vp, mp)
853 	register struct vnode *vp;
854 	register struct mount *mp;
855 {
856 
857 	mtx_lock(&mntvnode_mtx);
858 	/*
859 	 * Delete from old mount point vnode list, if on one.
860 	 */
861 	if (vp->v_mount != NULL)
862 		TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
863 	/*
864 	 * Insert into list of vnodes for the new mount point, if available.
865 	 */
866 	if ((vp->v_mount = mp) == NULL) {
867 		mtx_unlock(&mntvnode_mtx);
868 		return;
869 	}
870 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
871 	mtx_unlock(&mntvnode_mtx);
872 }
873 
874 /*
875  * Update outstanding I/O count and do wakeup if requested.
876  */
877 void
878 vwakeup(bp)
879 	register struct buf *bp;
880 {
881 	register struct vnode *vp;
882 
883 	bp->b_flags &= ~B_WRITEINPROG;
884 	if ((vp = bp->b_vp)) {
885 		vp->v_numoutput--;
886 		if (vp->v_numoutput < 0)
887 			panic("vwakeup: neg numoutput");
888 		if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
889 			vp->v_flag &= ~VBWAIT;
890 			wakeup((caddr_t) &vp->v_numoutput);
891 		}
892 	}
893 }
894 
895 /*
896  * Flush out and invalidate all buffers associated with a vnode.
897  * Called with the underlying object locked.
898  */
899 int
900 vinvalbuf(vp, flags, cred, td, slpflag, slptimeo)
901 	register struct vnode *vp;
902 	int flags;
903 	struct ucred *cred;
904 	struct thread *td;
905 	int slpflag, slptimeo;
906 {
907 	register struct buf *bp;
908 	struct buf *nbp, *blist;
909 	int s, error;
910 	vm_object_t object;
911 
912 	GIANT_REQUIRED;
913 
914 	if (flags & V_SAVE) {
915 		s = splbio();
916 		while (vp->v_numoutput) {
917 			vp->v_flag |= VBWAIT;
918 			error = tsleep((caddr_t)&vp->v_numoutput,
919 			    slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo);
920 			if (error) {
921 				splx(s);
922 				return (error);
923 			}
924 		}
925 		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
926 			splx(s);
927 			if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, td)) != 0)
928 				return (error);
929 			s = splbio();
930 			if (vp->v_numoutput > 0 ||
931 			    !TAILQ_EMPTY(&vp->v_dirtyblkhd))
932 				panic("vinvalbuf: dirty bufs");
933 		}
934 		splx(s);
935   	}
936 	s = splbio();
937 	for (;;) {
938 		blist = TAILQ_FIRST(&vp->v_cleanblkhd);
939 		if (!blist)
940 			blist = TAILQ_FIRST(&vp->v_dirtyblkhd);
941 		if (!blist)
942 			break;
943 
944 		for (bp = blist; bp; bp = nbp) {
945 			nbp = TAILQ_NEXT(bp, b_vnbufs);
946 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
947 				error = BUF_TIMELOCK(bp,
948 				    LK_EXCLUSIVE | LK_SLEEPFAIL,
949 				    "vinvalbuf", slpflag, slptimeo);
950 				if (error == ENOLCK)
951 					break;
952 				splx(s);
953 				return (error);
954 			}
955 			/*
956 			 * XXX Since there are no node locks for NFS, I
957 			 * believe there is a slight chance that a delayed
958 			 * write will occur while sleeping just above, so
959 			 * check for it.  Note that vfs_bio_awrite expects
960 			 * buffers to reside on a queue, while BUF_WRITE and
961 			 * brelse do not.
962 			 */
963 			if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
964 				(flags & V_SAVE)) {
965 
966 				if (bp->b_vp == vp) {
967 					if (bp->b_flags & B_CLUSTEROK) {
968 						BUF_UNLOCK(bp);
969 						vfs_bio_awrite(bp);
970 					} else {
971 						bremfree(bp);
972 						bp->b_flags |= B_ASYNC;
973 						BUF_WRITE(bp);
974 					}
975 				} else {
976 					bremfree(bp);
977 					(void) BUF_WRITE(bp);
978 				}
979 				break;
980 			}
981 			bremfree(bp);
982 			bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
983 			bp->b_flags &= ~B_ASYNC;
984 			brelse(bp);
985 		}
986 	}
987 
988 	/*
989 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
990 	 * have write I/O in-progress but if there is a VM object then the
991 	 * VM object can also have read-I/O in-progress.
992 	 */
993 	do {
994 		while (vp->v_numoutput > 0) {
995 			vp->v_flag |= VBWAIT;
996 			tsleep(&vp->v_numoutput, PVM, "vnvlbv", 0);
997 		}
998 		if (VOP_GETVOBJECT(vp, &object) == 0) {
999 			while (object->paging_in_progress)
1000 			vm_object_pip_sleep(object, "vnvlbx");
1001 		}
1002 	} while (vp->v_numoutput > 0);
1003 
1004 	splx(s);
1005 
1006 	/*
1007 	 * Destroy the copy in the VM cache, too.
1008 	 */
1009 	mtx_lock(&vp->v_interlock);
1010 	if (VOP_GETVOBJECT(vp, &object) == 0) {
1011 		vm_object_page_remove(object, 0, 0,
1012 			(flags & V_SAVE) ? TRUE : FALSE);
1013 	}
1014 	mtx_unlock(&vp->v_interlock);
1015 
1016 	if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd))
1017 		panic("vinvalbuf: flush failed");
1018 	return (0);
1019 }
1020 
1021 /*
1022  * Truncate a file's buffer and pages to a specified length.  This
1023  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1024  * sync activity.
1025  */
1026 int
1027 vtruncbuf(vp, cred, td, length, blksize)
1028 	register struct vnode *vp;
1029 	struct ucred *cred;
1030 	struct thread *td;
1031 	off_t length;
1032 	int blksize;
1033 {
1034 	register struct buf *bp;
1035 	struct buf *nbp;
1036 	int s, anyfreed;
1037 	int trunclbn;
1038 
1039 	/*
1040 	 * Round up to the *next* lbn.
1041 	 */
1042 	trunclbn = (length + blksize - 1) / blksize;
1043 
1044 	s = splbio();
1045 restart:
1046 	anyfreed = 1;
1047 	for (;anyfreed;) {
1048 		anyfreed = 0;
1049 		for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
1050 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1051 			if (bp->b_lblkno >= trunclbn) {
1052 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1053 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1054 					goto restart;
1055 				} else {
1056 					bremfree(bp);
1057 					bp->b_flags |= (B_INVAL | B_RELBUF);
1058 					bp->b_flags &= ~B_ASYNC;
1059 					brelse(bp);
1060 					anyfreed = 1;
1061 				}
1062 				if (nbp &&
1063 				    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1064 				    (nbp->b_vp != vp) ||
1065 				    (nbp->b_flags & B_DELWRI))) {
1066 					goto restart;
1067 				}
1068 			}
1069 		}
1070 
1071 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1072 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1073 			if (bp->b_lblkno >= trunclbn) {
1074 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1075 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1076 					goto restart;
1077 				} else {
1078 					bremfree(bp);
1079 					bp->b_flags |= (B_INVAL | B_RELBUF);
1080 					bp->b_flags &= ~B_ASYNC;
1081 					brelse(bp);
1082 					anyfreed = 1;
1083 				}
1084 				if (nbp &&
1085 				    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1086 				    (nbp->b_vp != vp) ||
1087 				    (nbp->b_flags & B_DELWRI) == 0)) {
1088 					goto restart;
1089 				}
1090 			}
1091 		}
1092 	}
1093 
1094 	if (length > 0) {
1095 restartsync:
1096 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1097 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1098 			if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
1099 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1100 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1101 					goto restart;
1102 				} else {
1103 					bremfree(bp);
1104 					if (bp->b_vp == vp) {
1105 						bp->b_flags |= B_ASYNC;
1106 					} else {
1107 						bp->b_flags &= ~B_ASYNC;
1108 					}
1109 					BUF_WRITE(bp);
1110 				}
1111 				goto restartsync;
1112 			}
1113 
1114 		}
1115 	}
1116 
1117 	while (vp->v_numoutput > 0) {
1118 		vp->v_flag |= VBWAIT;
1119 		tsleep(&vp->v_numoutput, PVM, "vbtrunc", 0);
1120 	}
1121 
1122 	splx(s);
1123 
1124 	vnode_pager_setsize(vp, length);
1125 
1126 	return (0);
1127 }
1128 
1129 /*
1130  * Associate a buffer with a vnode.
1131  */
1132 void
1133 bgetvp(vp, bp)
1134 	register struct vnode *vp;
1135 	register struct buf *bp;
1136 {
1137 	int s;
1138 
1139 	KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
1140 
1141 	vhold(vp);
1142 	bp->b_vp = vp;
1143 	bp->b_dev = vn_todev(vp);
1144 	/*
1145 	 * Insert onto list for new vnode.
1146 	 */
1147 	s = splbio();
1148 	bp->b_xflags |= BX_VNCLEAN;
1149 	bp->b_xflags &= ~BX_VNDIRTY;
1150 	TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
1151 	splx(s);
1152 }
1153 
1154 /*
1155  * Disassociate a buffer from a vnode.
1156  */
1157 void
1158 brelvp(bp)
1159 	register struct buf *bp;
1160 {
1161 	struct vnode *vp;
1162 	struct buflists *listheadp;
1163 	int s;
1164 
1165 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1166 
1167 	/*
1168 	 * Delete from old vnode list, if on one.
1169 	 */
1170 	vp = bp->b_vp;
1171 	s = splbio();
1172 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1173 		if (bp->b_xflags & BX_VNDIRTY)
1174 			listheadp = &vp->v_dirtyblkhd;
1175 		else
1176 			listheadp = &vp->v_cleanblkhd;
1177 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1178 		bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1179 	}
1180 	if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1181 		vp->v_flag &= ~VONWORKLST;
1182 		LIST_REMOVE(vp, v_synclist);
1183 	}
1184 	splx(s);
1185 	bp->b_vp = (struct vnode *) 0;
1186 	vdrop(vp);
1187 }
1188 
1189 /*
1190  * Add an item to the syncer work queue.
1191  */
1192 static void
1193 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
1194 {
1195 	int s, slot;
1196 
1197 	s = splbio();
1198 
1199 	if (vp->v_flag & VONWORKLST) {
1200 		LIST_REMOVE(vp, v_synclist);
1201 	}
1202 
1203 	if (delay > syncer_maxdelay - 2)
1204 		delay = syncer_maxdelay - 2;
1205 	slot = (syncer_delayno + delay) & syncer_mask;
1206 
1207 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
1208 	vp->v_flag |= VONWORKLST;
1209 	splx(s);
1210 }
1211 
1212 struct  proc *updateproc;
1213 static void sched_sync __P((void));
1214 static struct kproc_desc up_kp = {
1215 	"syncer",
1216 	sched_sync,
1217 	&updateproc
1218 };
1219 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1220 
1221 /*
1222  * System filesystem synchronizer daemon.
1223  */
1224 void
1225 sched_sync(void)
1226 {
1227 	struct synclist *slp;
1228 	struct vnode *vp;
1229 	struct mount *mp;
1230 	long starttime;
1231 	int s;
1232 	struct thread *td = FIRST_THREAD_IN_PROC(updateproc);  /* XXXKSE */
1233 
1234 	mtx_lock(&Giant);
1235 
1236 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, td->td_proc,
1237 	    SHUTDOWN_PRI_LAST);
1238 
1239 	for (;;) {
1240 		kthread_suspend_check(td->td_proc);
1241 
1242 		starttime = time_second;
1243 
1244 		/*
1245 		 * Push files whose dirty time has expired.  Be careful
1246 		 * of interrupt race on slp queue.
1247 		 */
1248 		s = splbio();
1249 		slp = &syncer_workitem_pending[syncer_delayno];
1250 		syncer_delayno += 1;
1251 		if (syncer_delayno == syncer_maxdelay)
1252 			syncer_delayno = 0;
1253 		splx(s);
1254 
1255 		while ((vp = LIST_FIRST(slp)) != NULL) {
1256 			if (VOP_ISLOCKED(vp, NULL) == 0 &&
1257 			    vn_start_write(vp, &mp, V_NOWAIT) == 0) {
1258 				vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1259 				(void) VOP_FSYNC(vp, td->td_proc->p_ucred, MNT_LAZY, td);
1260 				VOP_UNLOCK(vp, 0, td);
1261 				vn_finished_write(mp);
1262 			}
1263 			s = splbio();
1264 			if (LIST_FIRST(slp) == vp) {
1265 				/*
1266 				 * Note: v_tag VT_VFS vps can remain on the
1267 				 * worklist too with no dirty blocks, but
1268 				 * since sync_fsync() moves it to a different
1269 				 * slot we are safe.
1270 				 */
1271 				if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1272 				    !vn_isdisk(vp, NULL))
1273 					panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
1274 				/*
1275 				 * Put us back on the worklist.  The worklist
1276 				 * routine will remove us from our current
1277 				 * position and then add us back in at a later
1278 				 * position.
1279 				 */
1280 				vn_syncer_add_to_worklist(vp, syncdelay);
1281 			}
1282 			splx(s);
1283 		}
1284 
1285 		/*
1286 		 * Do soft update processing.
1287 		 */
1288 #ifdef SOFTUPDATES
1289 		softdep_process_worklist(NULL);
1290 #endif
1291 
1292 		/*
1293 		 * The variable rushjob allows the kernel to speed up the
1294 		 * processing of the filesystem syncer process. A rushjob
1295 		 * value of N tells the filesystem syncer to process the next
1296 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1297 		 * is used by the soft update code to speed up the filesystem
1298 		 * syncer process when the incore state is getting so far
1299 		 * ahead of the disk that the kernel memory pool is being
1300 		 * threatened with exhaustion.
1301 		 */
1302 		if (rushjob > 0) {
1303 			rushjob -= 1;
1304 			continue;
1305 		}
1306 		/*
1307 		 * If it has taken us less than a second to process the
1308 		 * current work, then wait. Otherwise start right over
1309 		 * again. We can still lose time if any single round
1310 		 * takes more than two seconds, but it does not really
1311 		 * matter as we are just trying to generally pace the
1312 		 * filesystem activity.
1313 		 */
1314 		if (time_second == starttime)
1315 			tsleep(&lbolt, PPAUSE, "syncer", 0);
1316 	}
1317 }
1318 
1319 /*
1320  * Request the syncer daemon to speed up its work.
1321  * We never push it to speed up more than half of its
1322  * normal turn time, otherwise it could take over the cpu.
1323  * XXXKSE  only one update?
1324  */
1325 int
1326 speedup_syncer()
1327 {
1328 
1329 	mtx_lock_spin(&sched_lock);
1330 	if (FIRST_THREAD_IN_PROC(updateproc)->td_wchan == &lbolt) /* XXXKSE */
1331 		setrunnable(FIRST_THREAD_IN_PROC(updateproc));
1332 	mtx_unlock_spin(&sched_lock);
1333 	if (rushjob < syncdelay / 2) {
1334 		rushjob += 1;
1335 		stat_rush_requests += 1;
1336 		return (1);
1337 	}
1338 	return(0);
1339 }
1340 
1341 /*
1342  * Associate a p-buffer with a vnode.
1343  *
1344  * Also sets B_PAGING flag to indicate that vnode is not fully associated
1345  * with the buffer.  i.e. the bp has not been linked into the vnode or
1346  * ref-counted.
1347  */
1348 void
1349 pbgetvp(vp, bp)
1350 	register struct vnode *vp;
1351 	register struct buf *bp;
1352 {
1353 
1354 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1355 
1356 	bp->b_vp = vp;
1357 	bp->b_flags |= B_PAGING;
1358 	bp->b_dev = vn_todev(vp);
1359 }
1360 
1361 /*
1362  * Disassociate a p-buffer from a vnode.
1363  */
1364 void
1365 pbrelvp(bp)
1366 	register struct buf *bp;
1367 {
1368 
1369 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1370 
1371 	/* XXX REMOVE ME */
1372 	if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1373 		panic(
1374 		    "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1375 		    bp,
1376 		    (int)bp->b_flags
1377 		);
1378 	}
1379 	bp->b_vp = (struct vnode *) 0;
1380 	bp->b_flags &= ~B_PAGING;
1381 }
1382 
1383 /*
1384  * Change the vnode a pager buffer is associated with.
1385  */
1386 void
1387 pbreassignbuf(bp, newvp)
1388 	struct buf *bp;
1389 	struct vnode *newvp;
1390 {
1391 
1392 	KASSERT(bp->b_flags & B_PAGING,
1393 	    ("pbreassignbuf() on non phys bp %p", bp));
1394 	bp->b_vp = newvp;
1395 }
1396 
1397 /*
1398  * Reassign a buffer from one vnode to another.
1399  * Used to assign file specific control information
1400  * (indirect blocks) to the vnode to which they belong.
1401  */
1402 void
1403 reassignbuf(bp, newvp)
1404 	register struct buf *bp;
1405 	register struct vnode *newvp;
1406 {
1407 	struct buflists *listheadp;
1408 	int delay;
1409 	int s;
1410 
1411 	if (newvp == NULL) {
1412 		printf("reassignbuf: NULL");
1413 		return;
1414 	}
1415 	++reassignbufcalls;
1416 
1417 	/*
1418 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1419 	 * is not fully linked in.
1420 	 */
1421 	if (bp->b_flags & B_PAGING)
1422 		panic("cannot reassign paging buffer");
1423 
1424 	s = splbio();
1425 	/*
1426 	 * Delete from old vnode list, if on one.
1427 	 */
1428 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1429 		if (bp->b_xflags & BX_VNDIRTY)
1430 			listheadp = &bp->b_vp->v_dirtyblkhd;
1431 		else
1432 			listheadp = &bp->b_vp->v_cleanblkhd;
1433 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1434 		bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1435 		if (bp->b_vp != newvp) {
1436 			vdrop(bp->b_vp);
1437 			bp->b_vp = NULL;	/* for clarification */
1438 		}
1439 	}
1440 	/*
1441 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1442 	 * of clean buffers.
1443 	 */
1444 	if (bp->b_flags & B_DELWRI) {
1445 		struct buf *tbp;
1446 
1447 		listheadp = &newvp->v_dirtyblkhd;
1448 		if ((newvp->v_flag & VONWORKLST) == 0) {
1449 			switch (newvp->v_type) {
1450 			case VDIR:
1451 				delay = dirdelay;
1452 				break;
1453 			case VCHR:
1454 				if (newvp->v_rdev->si_mountpoint != NULL) {
1455 					delay = metadelay;
1456 					break;
1457 				}
1458 				/* fall through */
1459 			default:
1460 				delay = filedelay;
1461 			}
1462 			vn_syncer_add_to_worklist(newvp, delay);
1463 		}
1464 		bp->b_xflags |= BX_VNDIRTY;
1465 		tbp = TAILQ_FIRST(listheadp);
1466 		if (tbp == NULL ||
1467 		    bp->b_lblkno == 0 ||
1468 		    (bp->b_lblkno > 0 && tbp->b_lblkno < 0) ||
1469 		    (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) {
1470 			TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1471 			++reassignbufsortgood;
1472 		} else if (bp->b_lblkno < 0) {
1473 			TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1474 			++reassignbufsortgood;
1475 		} else if (reassignbufmethod == 1) {
1476 			/*
1477 			 * New sorting algorithm, only handle sequential case,
1478 			 * otherwise append to end (but before metadata)
1479 			 */
1480 			if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL &&
1481 			    (tbp->b_xflags & BX_VNDIRTY)) {
1482 				/*
1483 				 * Found the best place to insert the buffer
1484 				 */
1485 				TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1486 				++reassignbufsortgood;
1487 			} else {
1488 				/*
1489 				 * Missed, append to end, but before meta-data.
1490 				 * We know that the head buffer in the list is
1491 				 * not meta-data due to prior conditionals.
1492 				 *
1493 				 * Indirect effects:  NFS second stage write
1494 				 * tends to wind up here, giving maximum
1495 				 * distance between the unstable write and the
1496 				 * commit rpc.
1497 				 */
1498 				tbp = TAILQ_LAST(listheadp, buflists);
1499 				while (tbp && tbp->b_lblkno < 0)
1500 					tbp = TAILQ_PREV(tbp, buflists, b_vnbufs);
1501 				TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1502 				++reassignbufsortbad;
1503 			}
1504 		} else {
1505 			/*
1506 			 * Old sorting algorithm, scan queue and insert
1507 			 */
1508 			struct buf *ttbp;
1509 			while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1510 			    (ttbp->b_lblkno < bp->b_lblkno)) {
1511 				++reassignbufloops;
1512 				tbp = ttbp;
1513 			}
1514 			TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1515 		}
1516 	} else {
1517 		bp->b_xflags |= BX_VNCLEAN;
1518 		TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1519 		if ((newvp->v_flag & VONWORKLST) &&
1520 		    TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1521 			newvp->v_flag &= ~VONWORKLST;
1522 			LIST_REMOVE(newvp, v_synclist);
1523 		}
1524 	}
1525 	if (bp->b_vp != newvp) {
1526 		bp->b_vp = newvp;
1527 		vhold(bp->b_vp);
1528 	}
1529 	splx(s);
1530 }
1531 
1532 /*
1533  * Create a vnode for a device.
1534  * Used for mounting the root file system.
1535  */
1536 int
1537 bdevvp(dev, vpp)
1538 	dev_t dev;
1539 	struct vnode **vpp;
1540 {
1541 	register struct vnode *vp;
1542 	struct vnode *nvp;
1543 	int error;
1544 
1545 	if (dev == NODEV) {
1546 		*vpp = NULLVP;
1547 		return (ENXIO);
1548 	}
1549 	if (vfinddev(dev, VCHR, vpp))
1550 		return (0);
1551 	error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1552 	if (error) {
1553 		*vpp = NULLVP;
1554 		return (error);
1555 	}
1556 	vp = nvp;
1557 	vp->v_type = VCHR;
1558 	addalias(vp, dev);
1559 	*vpp = vp;
1560 	return (0);
1561 }
1562 
1563 /*
1564  * Add vnode to the alias list hung off the dev_t.
1565  *
1566  * The reason for this gunk is that multiple vnodes can reference
1567  * the same physical device, so checking vp->v_usecount to see
1568  * how many users there are is inadequate; the v_usecount for
1569  * the vnodes need to be accumulated.  vcount() does that.
1570  */
1571 struct vnode *
1572 addaliasu(nvp, nvp_rdev)
1573 	struct vnode *nvp;
1574 	udev_t nvp_rdev;
1575 {
1576 	struct vnode *ovp;
1577 	vop_t **ops;
1578 	dev_t dev;
1579 
1580 	if (nvp->v_type == VBLK)
1581 		return (nvp);
1582 	if (nvp->v_type != VCHR)
1583 		panic("addaliasu on non-special vnode");
1584 	dev = udev2dev(nvp_rdev, 0);
1585 	/*
1586 	 * Check to see if we have a bdevvp vnode with no associated
1587 	 * filesystem. If so, we want to associate the filesystem of
1588 	 * the new newly instigated vnode with the bdevvp vnode and
1589 	 * discard the newly created vnode rather than leaving the
1590 	 * bdevvp vnode lying around with no associated filesystem.
1591 	 */
1592 	if (vfinddev(dev, nvp->v_type, &ovp) == 0 || ovp->v_data != NULL) {
1593 		addalias(nvp, dev);
1594 		return (nvp);
1595 	}
1596 	/*
1597 	 * Discard unneeded vnode, but save its node specific data.
1598 	 * Note that if there is a lock, it is carried over in the
1599 	 * node specific data to the replacement vnode.
1600 	 */
1601 	vref(ovp);
1602 	ovp->v_data = nvp->v_data;
1603 	ovp->v_tag = nvp->v_tag;
1604 	nvp->v_data = NULL;
1605 	lockinit(&ovp->v_lock, PVFS, nvp->v_lock.lk_wmesg,
1606 	    nvp->v_lock.lk_timo, nvp->v_lock.lk_flags & LK_EXTFLG_MASK);
1607 	if (nvp->v_vnlock)
1608 		ovp->v_vnlock = &ovp->v_lock;
1609 	ops = ovp->v_op;
1610 	ovp->v_op = nvp->v_op;
1611 	if (VOP_ISLOCKED(nvp, curthread)) {
1612 		VOP_UNLOCK(nvp, 0, curthread);
1613 		vn_lock(ovp, LK_EXCLUSIVE | LK_RETRY, curthread);
1614 	}
1615 	nvp->v_op = ops;
1616 	insmntque(ovp, nvp->v_mount);
1617 	vrele(nvp);
1618 	vgone(nvp);
1619 	return (ovp);
1620 }
1621 
1622 /* This is a local helper function that do the same as addaliasu, but for a
1623  * dev_t instead of an udev_t. */
1624 static void
1625 addalias(nvp, dev)
1626 	struct vnode *nvp;
1627 	dev_t dev;
1628 {
1629 
1630 	KASSERT(nvp->v_type == VCHR, ("addalias on non-special vnode"));
1631 	nvp->v_rdev = dev;
1632 	mtx_lock(&spechash_mtx);
1633 	SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
1634 	mtx_unlock(&spechash_mtx);
1635 }
1636 
1637 /*
1638  * Grab a particular vnode from the free list, increment its
1639  * reference count and lock it. The vnode lock bit is set if the
1640  * vnode is being eliminated in vgone. The process is awakened
1641  * when the transition is completed, and an error returned to
1642  * indicate that the vnode is no longer usable (possibly having
1643  * been changed to a new file system type).
1644  */
1645 int
1646 vget(vp, flags, td)
1647 	register struct vnode *vp;
1648 	int flags;
1649 	struct thread *td;
1650 {
1651 	int error;
1652 
1653 	/*
1654 	 * If the vnode is in the process of being cleaned out for
1655 	 * another use, we wait for the cleaning to finish and then
1656 	 * return failure. Cleaning is determined by checking that
1657 	 * the VXLOCK flag is set.
1658 	 */
1659 	if ((flags & LK_INTERLOCK) == 0)
1660 		mtx_lock(&vp->v_interlock);
1661 	if (vp->v_flag & VXLOCK) {
1662 		if (vp->v_vxproc == curthread) {
1663 #if 0
1664 			/* this can now occur in normal operation */
1665 			log(LOG_INFO, "VXLOCK interlock avoided\n");
1666 #endif
1667 		} else {
1668 			vp->v_flag |= VXWANT;
1669 			msleep((caddr_t)vp, &vp->v_interlock, PINOD | PDROP,
1670 			    "vget", 0);
1671 			return (ENOENT);
1672 		}
1673 	}
1674 
1675 	vp->v_usecount++;
1676 
1677 	if (VSHOULDBUSY(vp))
1678 		vbusy(vp);
1679 	if (flags & LK_TYPE_MASK) {
1680 		if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
1681 			/*
1682 			 * must expand vrele here because we do not want
1683 			 * to call VOP_INACTIVE if the reference count
1684 			 * drops back to zero since it was never really
1685 			 * active. We must remove it from the free list
1686 			 * before sleeping so that multiple processes do
1687 			 * not try to recycle it.
1688 			 */
1689 			mtx_lock(&vp->v_interlock);
1690 			vp->v_usecount--;
1691 			if (VSHOULDFREE(vp))
1692 				vfree(vp);
1693 			else
1694 				vlruvp(vp);
1695 			mtx_unlock(&vp->v_interlock);
1696 		}
1697 		return (error);
1698 	}
1699 	mtx_unlock(&vp->v_interlock);
1700 	return (0);
1701 }
1702 
1703 /*
1704  * Increase the reference count of a vnode.
1705  */
1706 void
1707 vref(struct vnode *vp)
1708 {
1709 	mtx_lock(&vp->v_interlock);
1710 	vp->v_usecount++;
1711 	mtx_unlock(&vp->v_interlock);
1712 }
1713 
1714 /*
1715  * Vnode put/release.
1716  * If count drops to zero, call inactive routine and return to freelist.
1717  */
1718 void
1719 vrele(vp)
1720 	struct vnode *vp;
1721 {
1722 	struct thread *td = curthread;	/* XXX */
1723 
1724 	KASSERT(vp != NULL, ("vrele: null vp"));
1725 
1726 	mtx_lock(&vp->v_interlock);
1727 
1728 	/* Skip this v_writecount check if we're going to panic below. */
1729 	KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
1730 	    ("vrele: missed vn_close"));
1731 
1732 	if (vp->v_usecount > 1) {
1733 
1734 		vp->v_usecount--;
1735 		mtx_unlock(&vp->v_interlock);
1736 
1737 		return;
1738 	}
1739 
1740 	if (vp->v_usecount == 1) {
1741 		vp->v_usecount--;
1742 		/*
1743 		 * We must call VOP_INACTIVE with the node locked.
1744 		 * If we are doing a vput, the node is already locked,
1745 		 * but, in the case of vrele, we must explicitly lock
1746 		 * the vnode before calling VOP_INACTIVE.
1747 		 */
1748 		if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0)
1749 			VOP_INACTIVE(vp, td);
1750 		if (VSHOULDFREE(vp))
1751 			vfree(vp);
1752 		else
1753 			vlruvp(vp);
1754 
1755 	} else {
1756 #ifdef DIAGNOSTIC
1757 		vprint("vrele: negative ref count", vp);
1758 		mtx_unlock(&vp->v_interlock);
1759 #endif
1760 		panic("vrele: negative ref cnt");
1761 	}
1762 }
1763 
1764 /*
1765  * Release an already locked vnode.  This give the same effects as
1766  * unlock+vrele(), but takes less time and avoids releasing and
1767  * re-aquiring the lock (as vrele() aquires the lock internally.)
1768  */
1769 void
1770 vput(vp)
1771 	struct vnode *vp;
1772 {
1773 	struct thread *td = curthread;	/* XXX */
1774 
1775 	GIANT_REQUIRED;
1776 
1777 	KASSERT(vp != NULL, ("vput: null vp"));
1778 	mtx_lock(&vp->v_interlock);
1779 	/* Skip this v_writecount check if we're going to panic below. */
1780 	KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
1781 	    ("vput: missed vn_close"));
1782 
1783 	if (vp->v_usecount > 1) {
1784 		vp->v_usecount--;
1785 		VOP_UNLOCK(vp, LK_INTERLOCK, td);
1786 		return;
1787 	}
1788 
1789 	if (vp->v_usecount == 1) {
1790 		vp->v_usecount--;
1791 		/*
1792 		 * We must call VOP_INACTIVE with the node locked.
1793 		 * If we are doing a vput, the node is already locked,
1794 		 * so we just need to release the vnode mutex.
1795 		 */
1796 		mtx_unlock(&vp->v_interlock);
1797 		VOP_INACTIVE(vp, td);
1798 		if (VSHOULDFREE(vp))
1799 			vfree(vp);
1800 		else
1801 			vlruvp(vp);
1802 
1803 	} else {
1804 #ifdef DIAGNOSTIC
1805 		vprint("vput: negative ref count", vp);
1806 #endif
1807 		panic("vput: negative ref cnt");
1808 	}
1809 }
1810 
1811 /*
1812  * Somebody doesn't want the vnode recycled.
1813  */
1814 void
1815 vhold(vp)
1816 	register struct vnode *vp;
1817 {
1818 	int s;
1819 
1820   	s = splbio();
1821 	vp->v_holdcnt++;
1822 	if (VSHOULDBUSY(vp))
1823 		vbusy(vp);
1824 	splx(s);
1825 }
1826 
1827 /*
1828  * Note that there is one less who cares about this vnode.  vdrop() is the
1829  * opposite of vhold().
1830  */
1831 void
1832 vdrop(vp)
1833 	register struct vnode *vp;
1834 {
1835 	int s;
1836 
1837 	s = splbio();
1838 	if (vp->v_holdcnt <= 0)
1839 		panic("vdrop: holdcnt");
1840 	vp->v_holdcnt--;
1841 	if (VSHOULDFREE(vp))
1842 		vfree(vp);
1843 	else
1844 		vlruvp(vp);
1845 	splx(s);
1846 }
1847 
1848 /*
1849  * Remove any vnodes in the vnode table belonging to mount point mp.
1850  *
1851  * If FORCECLOSE is not specified, there should not be any active ones,
1852  * return error if any are found (nb: this is a user error, not a
1853  * system error). If FORCECLOSE is specified, detach any active vnodes
1854  * that are found.
1855  *
1856  * If WRITECLOSE is set, only flush out regular file vnodes open for
1857  * writing.
1858  *
1859  * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
1860  *
1861  * `rootrefs' specifies the base reference count for the root vnode
1862  * of this filesystem. The root vnode is considered busy if its
1863  * v_usecount exceeds this value. On a successful return, vflush()
1864  * will call vrele() on the root vnode exactly rootrefs times.
1865  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
1866  * be zero.
1867  */
1868 #ifdef DIAGNOSTIC
1869 static int busyprt = 0;		/* print out busy vnodes */
1870 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1871 #endif
1872 
1873 int
1874 vflush(mp, rootrefs, flags)
1875 	struct mount *mp;
1876 	int rootrefs;
1877 	int flags;
1878 {
1879 	struct thread *td = curthread;	/* XXX */
1880 	struct vnode *vp, *nvp, *rootvp = NULL;
1881 	struct vattr vattr;
1882 	int busy = 0, error;
1883 
1884 	if (rootrefs > 0) {
1885 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
1886 		    ("vflush: bad args"));
1887 		/*
1888 		 * Get the filesystem root vnode. We can vput() it
1889 		 * immediately, since with rootrefs > 0, it won't go away.
1890 		 */
1891 		if ((error = VFS_ROOT(mp, &rootvp)) != 0)
1892 			return (error);
1893 		vput(rootvp);
1894 	}
1895 	mtx_lock(&mntvnode_mtx);
1896 loop:
1897 	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) {
1898 		/*
1899 		 * Make sure this vnode wasn't reclaimed in getnewvnode().
1900 		 * Start over if it has (it won't be on the list anymore).
1901 		 */
1902 		if (vp->v_mount != mp)
1903 			goto loop;
1904 		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
1905 
1906 		mtx_unlock(&mntvnode_mtx);
1907 		mtx_lock(&vp->v_interlock);
1908 		/*
1909 		 * Skip over a vnodes marked VSYSTEM.
1910 		 */
1911 		if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1912 			mtx_unlock(&vp->v_interlock);
1913 			mtx_lock(&mntvnode_mtx);
1914 			continue;
1915 		}
1916 		/*
1917 		 * If WRITECLOSE is set, flush out unlinked but still open
1918 		 * files (even if open only for reading) and regular file
1919 		 * vnodes open for writing.
1920 		 */
1921 		if ((flags & WRITECLOSE) &&
1922 		    (vp->v_type == VNON ||
1923 		    (VOP_GETATTR(vp, &vattr, td->td_proc->p_ucred, td) == 0 &&
1924 		    vattr.va_nlink > 0)) &&
1925 		    (vp->v_writecount == 0 || vp->v_type != VREG)) {
1926 			mtx_unlock(&vp->v_interlock);
1927 			mtx_lock(&mntvnode_mtx);
1928 			continue;
1929 		}
1930 
1931 		/*
1932 		 * With v_usecount == 0, all we need to do is clear out the
1933 		 * vnode data structures and we are done.
1934 		 */
1935 		if (vp->v_usecount == 0) {
1936 			vgonel(vp, td);
1937 			mtx_lock(&mntvnode_mtx);
1938 			continue;
1939 		}
1940 
1941 		/*
1942 		 * If FORCECLOSE is set, forcibly close the vnode. For block
1943 		 * or character devices, revert to an anonymous device. For
1944 		 * all other files, just kill them.
1945 		 */
1946 		if (flags & FORCECLOSE) {
1947 			if (vp->v_type != VCHR) {
1948 				vgonel(vp, td);
1949 			} else {
1950 				vclean(vp, 0, td);
1951 				vp->v_op = spec_vnodeop_p;
1952 				insmntque(vp, (struct mount *) 0);
1953 			}
1954 			mtx_lock(&mntvnode_mtx);
1955 			continue;
1956 		}
1957 #ifdef DIAGNOSTIC
1958 		if (busyprt)
1959 			vprint("vflush: busy vnode", vp);
1960 #endif
1961 		mtx_unlock(&vp->v_interlock);
1962 		mtx_lock(&mntvnode_mtx);
1963 		busy++;
1964 	}
1965 	mtx_unlock(&mntvnode_mtx);
1966 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
1967 		/*
1968 		 * If just the root vnode is busy, and if its refcount
1969 		 * is equal to `rootrefs', then go ahead and kill it.
1970 		 */
1971 		mtx_lock(&rootvp->v_interlock);
1972 		KASSERT(busy > 0, ("vflush: not busy"));
1973 		KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
1974 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
1975 			vgonel(rootvp, td);
1976 			busy = 0;
1977 		} else
1978 			mtx_unlock(&rootvp->v_interlock);
1979 	}
1980 	if (busy)
1981 		return (EBUSY);
1982 	for (; rootrefs > 0; rootrefs--)
1983 		vrele(rootvp);
1984 	return (0);
1985 }
1986 
1987 /*
1988  * This moves a now (likely recyclable) vnode to the end of the
1989  * mountlist.  XXX However, it is temporarily disabled until we
1990  * can clean up ffs_sync() and friends, which have loop restart
1991  * conditions which this code causes to operate O(N^2).
1992  */
1993 static void
1994 vlruvp(struct vnode *vp)
1995 {
1996 #if 0
1997 	struct mount *mp;
1998 
1999 	if ((mp = vp->v_mount) != NULL) {
2000 		mtx_lock(&mntvnode_mtx);
2001 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
2002 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
2003 		mtx_unlock(&mntvnode_mtx);
2004 	}
2005 #endif
2006 }
2007 
2008 /*
2009  * Disassociate the underlying file system from a vnode.
2010  */
2011 static void
2012 vclean(vp, flags, td)
2013 	struct vnode *vp;
2014 	int flags;
2015 	struct thread *td;
2016 {
2017 	int active;
2018 
2019 	/*
2020 	 * Check to see if the vnode is in use. If so we have to reference it
2021 	 * before we clean it out so that its count cannot fall to zero and
2022 	 * generate a race against ourselves to recycle it.
2023 	 */
2024 	if ((active = vp->v_usecount))
2025 		vp->v_usecount++;
2026 
2027 	/*
2028 	 * Prevent the vnode from being recycled or brought into use while we
2029 	 * clean it out.
2030 	 */
2031 	if (vp->v_flag & VXLOCK)
2032 		panic("vclean: deadlock");
2033 	vp->v_flag |= VXLOCK;
2034 	vp->v_vxproc = curthread;
2035 	/*
2036 	 * Even if the count is zero, the VOP_INACTIVE routine may still
2037 	 * have the object locked while it cleans it out. The VOP_LOCK
2038 	 * ensures that the VOP_INACTIVE routine is done with its work.
2039 	 * For active vnodes, it ensures that no other activity can
2040 	 * occur while the underlying object is being cleaned out.
2041 	 */
2042 	VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td);
2043 
2044 	/*
2045 	 * Clean out any buffers associated with the vnode.
2046 	 * If the flush fails, just toss the buffers.
2047 	 */
2048 	if (flags & DOCLOSE) {
2049 		if (TAILQ_FIRST(&vp->v_dirtyblkhd) != NULL)
2050 			(void) vn_write_suspend_wait(vp, NULL, V_WAIT);
2051 		if (vinvalbuf(vp, V_SAVE, NOCRED, td, 0, 0) != 0)
2052 			vinvalbuf(vp, 0, NOCRED, td, 0, 0);
2053 	}
2054 
2055 	VOP_DESTROYVOBJECT(vp);
2056 
2057 	/*
2058 	 * If purging an active vnode, it must be closed and
2059 	 * deactivated before being reclaimed. Note that the
2060 	 * VOP_INACTIVE will unlock the vnode.
2061 	 */
2062 	if (active) {
2063 		if (flags & DOCLOSE)
2064 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2065 		VOP_INACTIVE(vp, td);
2066 	} else {
2067 		/*
2068 		 * Any other processes trying to obtain this lock must first
2069 		 * wait for VXLOCK to clear, then call the new lock operation.
2070 		 */
2071 		VOP_UNLOCK(vp, 0, td);
2072 	}
2073 	/*
2074 	 * Reclaim the vnode.
2075 	 */
2076 	if (VOP_RECLAIM(vp, td))
2077 		panic("vclean: cannot reclaim");
2078 
2079 	if (active) {
2080 		/*
2081 		 * Inline copy of vrele() since VOP_INACTIVE
2082 		 * has already been called.
2083 		 */
2084 		mtx_lock(&vp->v_interlock);
2085 		if (--vp->v_usecount <= 0) {
2086 #ifdef DIAGNOSTIC
2087 			if (vp->v_usecount < 0 || vp->v_writecount != 0) {
2088 				vprint("vclean: bad ref count", vp);
2089 				panic("vclean: ref cnt");
2090 			}
2091 #endif
2092 			vfree(vp);
2093 		}
2094 		mtx_unlock(&vp->v_interlock);
2095 	}
2096 
2097 	cache_purge(vp);
2098 	vp->v_vnlock = NULL;
2099 	lockdestroy(&vp->v_lock);
2100 
2101 	if (VSHOULDFREE(vp))
2102 		vfree(vp);
2103 
2104 	/*
2105 	 * Done with purge, notify sleepers of the grim news.
2106 	 */
2107 	vp->v_op = dead_vnodeop_p;
2108 	if (vp->v_pollinfo != NULL)
2109 		vn_pollgone(vp);
2110 	vp->v_tag = VT_NON;
2111 	vp->v_flag &= ~VXLOCK;
2112 	vp->v_vxproc = NULL;
2113 	if (vp->v_flag & VXWANT) {
2114 		vp->v_flag &= ~VXWANT;
2115 		wakeup((caddr_t) vp);
2116 	}
2117 }
2118 
2119 /*
2120  * Eliminate all activity associated with the requested vnode
2121  * and with all vnodes aliased to the requested vnode.
2122  */
2123 int
2124 vop_revoke(ap)
2125 	struct vop_revoke_args /* {
2126 		struct vnode *a_vp;
2127 		int a_flags;
2128 	} */ *ap;
2129 {
2130 	struct vnode *vp, *vq;
2131 	dev_t dev;
2132 
2133 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
2134 
2135 	vp = ap->a_vp;
2136 	/*
2137 	 * If a vgone (or vclean) is already in progress,
2138 	 * wait until it is done and return.
2139 	 */
2140 	if (vp->v_flag & VXLOCK) {
2141 		vp->v_flag |= VXWANT;
2142 		msleep((caddr_t)vp, &vp->v_interlock, PINOD | PDROP,
2143 		    "vop_revokeall", 0);
2144 		return (0);
2145 	}
2146 	dev = vp->v_rdev;
2147 	for (;;) {
2148 		mtx_lock(&spechash_mtx);
2149 		vq = SLIST_FIRST(&dev->si_hlist);
2150 		mtx_unlock(&spechash_mtx);
2151 		if (!vq)
2152 			break;
2153 		vgone(vq);
2154 	}
2155 	return (0);
2156 }
2157 
2158 /*
2159  * Recycle an unused vnode to the front of the free list.
2160  * Release the passed interlock if the vnode will be recycled.
2161  */
2162 int
2163 vrecycle(vp, inter_lkp, td)
2164 	struct vnode *vp;
2165 	struct mtx *inter_lkp;
2166 	struct thread *td;
2167 {
2168 
2169 	mtx_lock(&vp->v_interlock);
2170 	if (vp->v_usecount == 0) {
2171 		if (inter_lkp) {
2172 			mtx_unlock(inter_lkp);
2173 		}
2174 		vgonel(vp, td);
2175 		return (1);
2176 	}
2177 	mtx_unlock(&vp->v_interlock);
2178 	return (0);
2179 }
2180 
2181 /*
2182  * Eliminate all activity associated with a vnode
2183  * in preparation for reuse.
2184  */
2185 void
2186 vgone(vp)
2187 	register struct vnode *vp;
2188 {
2189 	struct thread *td = curthread;	/* XXX */
2190 
2191 	mtx_lock(&vp->v_interlock);
2192 	vgonel(vp, td);
2193 }
2194 
2195 /*
2196  * vgone, with the vp interlock held.
2197  */
2198 void
2199 vgonel(vp, td)
2200 	struct vnode *vp;
2201 	struct thread *td;
2202 {
2203 	int s;
2204 
2205 	/*
2206 	 * If a vgone (or vclean) is already in progress,
2207 	 * wait until it is done and return.
2208 	 */
2209 	if (vp->v_flag & VXLOCK) {
2210 		vp->v_flag |= VXWANT;
2211 		msleep((caddr_t)vp, &vp->v_interlock, PINOD | PDROP,
2212 		    "vgone", 0);
2213 		return;
2214 	}
2215 
2216 	/*
2217 	 * Clean out the filesystem specific data.
2218 	 */
2219 	vclean(vp, DOCLOSE, td);
2220 	mtx_lock(&vp->v_interlock);
2221 
2222 	/*
2223 	 * Delete from old mount point vnode list, if on one.
2224 	 */
2225 	if (vp->v_mount != NULL)
2226 		insmntque(vp, (struct mount *)0);
2227 	/*
2228 	 * If special device, remove it from special device alias list
2229 	 * if it is on one.
2230 	 */
2231 	if (vp->v_type == VCHR && vp->v_rdev != NULL && vp->v_rdev != NODEV) {
2232 		mtx_lock(&spechash_mtx);
2233 		SLIST_REMOVE(&vp->v_rdev->si_hlist, vp, vnode, v_specnext);
2234 		freedev(vp->v_rdev);
2235 		mtx_unlock(&spechash_mtx);
2236 		vp->v_rdev = NULL;
2237 	}
2238 
2239 	/*
2240 	 * If it is on the freelist and not already at the head,
2241 	 * move it to the head of the list. The test of the
2242 	 * VDOOMED flag and the reference count of zero is because
2243 	 * it will be removed from the free list by getnewvnode,
2244 	 * but will not have its reference count incremented until
2245 	 * after calling vgone. If the reference count were
2246 	 * incremented first, vgone would (incorrectly) try to
2247 	 * close the previous instance of the underlying object.
2248 	 */
2249 	if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
2250 		s = splbio();
2251 		mtx_lock(&vnode_free_list_mtx);
2252 		if (vp->v_flag & VFREE)
2253 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2254 		else
2255 			freevnodes++;
2256 		vp->v_flag |= VFREE;
2257 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2258 		mtx_unlock(&vnode_free_list_mtx);
2259 		splx(s);
2260 	}
2261 
2262 	vp->v_type = VBAD;
2263 	mtx_unlock(&vp->v_interlock);
2264 }
2265 
2266 /*
2267  * Lookup a vnode by device number.
2268  */
2269 int
2270 vfinddev(dev, type, vpp)
2271 	dev_t dev;
2272 	enum vtype type;
2273 	struct vnode **vpp;
2274 {
2275 	struct vnode *vp;
2276 
2277 	mtx_lock(&spechash_mtx);
2278 	SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2279 		if (type == vp->v_type) {
2280 			*vpp = vp;
2281 			mtx_unlock(&spechash_mtx);
2282 			return (1);
2283 		}
2284 	}
2285 	mtx_unlock(&spechash_mtx);
2286 	return (0);
2287 }
2288 
2289 /*
2290  * Calculate the total number of references to a special device.
2291  */
2292 int
2293 vcount(vp)
2294 	struct vnode *vp;
2295 {
2296 	struct vnode *vq;
2297 	int count;
2298 
2299 	count = 0;
2300 	mtx_lock(&spechash_mtx);
2301 	SLIST_FOREACH(vq, &vp->v_rdev->si_hlist, v_specnext)
2302 		count += vq->v_usecount;
2303 	mtx_unlock(&spechash_mtx);
2304 	return (count);
2305 }
2306 
2307 /*
2308  * Same as above, but using the dev_t as argument
2309  */
2310 int
2311 count_dev(dev)
2312 	dev_t dev;
2313 {
2314 	struct vnode *vp;
2315 
2316 	vp = SLIST_FIRST(&dev->si_hlist);
2317 	if (vp == NULL)
2318 		return (0);
2319 	return(vcount(vp));
2320 }
2321 
2322 /*
2323  * Print out a description of a vnode.
2324  */
2325 static char *typename[] =
2326 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2327 
2328 void
2329 vprint(label, vp)
2330 	char *label;
2331 	struct vnode *vp;
2332 {
2333 	char buf[96];
2334 
2335 	if (label != NULL)
2336 		printf("%s: %p: ", label, (void *)vp);
2337 	else
2338 		printf("%p: ", (void *)vp);
2339 	printf("type %s, usecount %d, writecount %d, refcount %d,",
2340 	    typename[vp->v_type], vp->v_usecount, vp->v_writecount,
2341 	    vp->v_holdcnt);
2342 	buf[0] = '\0';
2343 	if (vp->v_flag & VROOT)
2344 		strcat(buf, "|VROOT");
2345 	if (vp->v_flag & VTEXT)
2346 		strcat(buf, "|VTEXT");
2347 	if (vp->v_flag & VSYSTEM)
2348 		strcat(buf, "|VSYSTEM");
2349 	if (vp->v_flag & VXLOCK)
2350 		strcat(buf, "|VXLOCK");
2351 	if (vp->v_flag & VXWANT)
2352 		strcat(buf, "|VXWANT");
2353 	if (vp->v_flag & VBWAIT)
2354 		strcat(buf, "|VBWAIT");
2355 	if (vp->v_flag & VDOOMED)
2356 		strcat(buf, "|VDOOMED");
2357 	if (vp->v_flag & VFREE)
2358 		strcat(buf, "|VFREE");
2359 	if (vp->v_flag & VOBJBUF)
2360 		strcat(buf, "|VOBJBUF");
2361 	if (buf[0] != '\0')
2362 		printf(" flags (%s)", &buf[1]);
2363 	if (vp->v_data == NULL) {
2364 		printf("\n");
2365 	} else {
2366 		printf("\n\t");
2367 		VOP_PRINT(vp);
2368 	}
2369 }
2370 
2371 #ifdef DDB
2372 #include <ddb/ddb.h>
2373 /*
2374  * List all of the locked vnodes in the system.
2375  * Called when debugging the kernel.
2376  */
2377 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
2378 {
2379 	struct thread *td = curthread;	/* XXX */
2380 	struct mount *mp, *nmp;
2381 	struct vnode *vp;
2382 
2383 	printf("Locked vnodes\n");
2384 	mtx_lock(&mountlist_mtx);
2385 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2386 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
2387 			nmp = TAILQ_NEXT(mp, mnt_list);
2388 			continue;
2389 		}
2390 		mtx_lock(&mntvnode_mtx);
2391 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2392 			if (VOP_ISLOCKED(vp, NULL))
2393 				vprint((char *)0, vp);
2394 		}
2395 		mtx_unlock(&mntvnode_mtx);
2396 		mtx_lock(&mountlist_mtx);
2397 		nmp = TAILQ_NEXT(mp, mnt_list);
2398 		vfs_unbusy(mp, td);
2399 	}
2400 	mtx_unlock(&mountlist_mtx);
2401 }
2402 #endif
2403 
2404 /*
2405  * Top level filesystem related information gathering.
2406  */
2407 static int	sysctl_ovfs_conf __P((SYSCTL_HANDLER_ARGS));
2408 
2409 static int
2410 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2411 {
2412 	int *name = (int *)arg1 - 1;	/* XXX */
2413 	u_int namelen = arg2 + 1;	/* XXX */
2414 	struct vfsconf *vfsp;
2415 
2416 #if 1 || defined(COMPAT_PRELITE2)
2417 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2418 	if (namelen == 1)
2419 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2420 #endif
2421 
2422 	/* XXX the below code does not compile; vfs_sysctl does not exist. */
2423 #ifdef notyet
2424 	/* all sysctl names at this level are at least name and field */
2425 	if (namelen < 2)
2426 		return (ENOTDIR);		/* overloaded */
2427 	if (name[0] != VFS_GENERIC) {
2428 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2429 			if (vfsp->vfc_typenum == name[0])
2430 				break;
2431 		if (vfsp == NULL)
2432 			return (EOPNOTSUPP);
2433 		return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2434 		    oldp, oldlenp, newp, newlen, td));
2435 	}
2436 #endif
2437 	switch (name[1]) {
2438 	case VFS_MAXTYPENUM:
2439 		if (namelen != 2)
2440 			return (ENOTDIR);
2441 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2442 	case VFS_CONF:
2443 		if (namelen != 3)
2444 			return (ENOTDIR);	/* overloaded */
2445 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2446 			if (vfsp->vfc_typenum == name[2])
2447 				break;
2448 		if (vfsp == NULL)
2449 			return (EOPNOTSUPP);
2450 		return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2451 	}
2452 	return (EOPNOTSUPP);
2453 }
2454 
2455 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2456 	"Generic filesystem");
2457 
2458 #if 1 || defined(COMPAT_PRELITE2)
2459 
2460 static int
2461 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2462 {
2463 	int error;
2464 	struct vfsconf *vfsp;
2465 	struct ovfsconf ovfs;
2466 
2467 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2468 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
2469 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
2470 		ovfs.vfc_index = vfsp->vfc_typenum;
2471 		ovfs.vfc_refcount = vfsp->vfc_refcount;
2472 		ovfs.vfc_flags = vfsp->vfc_flags;
2473 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2474 		if (error)
2475 			return error;
2476 	}
2477 	return 0;
2478 }
2479 
2480 #endif /* 1 || COMPAT_PRELITE2 */
2481 
2482 #if COMPILING_LINT
2483 #define KINFO_VNODESLOP	10
2484 /*
2485  * Dump vnode list (via sysctl).
2486  * Copyout address of vnode followed by vnode.
2487  */
2488 /* ARGSUSED */
2489 static int
2490 sysctl_vnode(SYSCTL_HANDLER_ARGS)
2491 {
2492 	struct thread *td = curthread;	/* XXX */
2493 	struct mount *mp, *nmp;
2494 	struct vnode *nvp, *vp;
2495 	int error;
2496 
2497 #define VPTRSZ	sizeof (struct vnode *)
2498 #define VNODESZ	sizeof (struct vnode)
2499 
2500 	req->lock = 0;
2501 	if (!req->oldptr) /* Make an estimate */
2502 		return (SYSCTL_OUT(req, 0,
2503 			(numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2504 
2505 	mtx_lock(&mountlist_mtx);
2506 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2507 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
2508 			nmp = TAILQ_NEXT(mp, mnt_list);
2509 			continue;
2510 		}
2511 		mtx_lock(&mntvnode_mtx);
2512 again:
2513 		for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2514 		     vp != NULL;
2515 		     vp = nvp) {
2516 			/*
2517 			 * Check that the vp is still associated with
2518 			 * this filesystem.  RACE: could have been
2519 			 * recycled onto the same filesystem.
2520 			 */
2521 			if (vp->v_mount != mp)
2522 				goto again;
2523 			nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2524 			mtx_unlock(&mntvnode_mtx);
2525 			if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2526 			    (error = SYSCTL_OUT(req, vp, VNODESZ)))
2527 				return (error);
2528 			mtx_lock(&mntvnode_mtx);
2529 		}
2530 		mtx_unlock(&mntvnode_mtx);
2531 		mtx_lock(&mountlist_mtx);
2532 		nmp = TAILQ_NEXT(mp, mnt_list);
2533 		vfs_unbusy(mp, td);
2534 	}
2535 	mtx_unlock(&mountlist_mtx);
2536 
2537 	return (0);
2538 }
2539 
2540 /*
2541  * XXX
2542  * Exporting the vnode list on large systems causes them to crash.
2543  * Exporting the vnode list on medium systems causes sysctl to coredump.
2544  */
2545 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2546 	0, 0, sysctl_vnode, "S,vnode", "");
2547 #endif
2548 
2549 /*
2550  * Check to see if a filesystem is mounted on a block device.
2551  */
2552 int
2553 vfs_mountedon(vp)
2554 	struct vnode *vp;
2555 {
2556 
2557 	if (vp->v_rdev->si_mountpoint != NULL)
2558 		return (EBUSY);
2559 	return (0);
2560 }
2561 
2562 /*
2563  * Unmount all filesystems. The list is traversed in reverse order
2564  * of mounting to avoid dependencies.
2565  */
2566 void
2567 vfs_unmountall()
2568 {
2569 	struct mount *mp;
2570 	struct thread *td;
2571 	int error;
2572 
2573 	if (curthread != NULL)
2574 		td = curthread;
2575 	else
2576 		td = FIRST_THREAD_IN_PROC(initproc); /* XXX XXX proc0? */
2577 	/*
2578 	 * Since this only runs when rebooting, it is not interlocked.
2579 	 */
2580 	while(!TAILQ_EMPTY(&mountlist)) {
2581 		mp = TAILQ_LAST(&mountlist, mntlist);
2582 		error = dounmount(mp, MNT_FORCE, td);
2583 		if (error) {
2584 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
2585 			printf("unmount of %s failed (",
2586 			    mp->mnt_stat.f_mntonname);
2587 			if (error == EBUSY)
2588 				printf("BUSY)\n");
2589 			else
2590 				printf("%d)\n", error);
2591 		} else {
2592 			/* The unmount has removed mp from the mountlist */
2593 		}
2594 	}
2595 }
2596 
2597 /*
2598  * perform msync on all vnodes under a mount point
2599  * the mount point must be locked.
2600  */
2601 void
2602 vfs_msync(struct mount *mp, int flags)
2603 {
2604 	struct vnode *vp, *nvp;
2605 	struct vm_object *obj;
2606 	int tries;
2607 
2608 	GIANT_REQUIRED;
2609 
2610 	tries = 5;
2611 	mtx_lock(&mntvnode_mtx);
2612 loop:
2613 	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) {
2614 		if (vp->v_mount != mp) {
2615 			if (--tries > 0)
2616 				goto loop;
2617 			break;
2618 		}
2619 		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2620 
2621 		if (vp->v_flag & VXLOCK)	/* XXX: what if MNT_WAIT? */
2622 			continue;
2623 
2624 		if (vp->v_flag & VNOSYNC)	/* unlinked, skip it */
2625 			continue;
2626 
2627 		if ((vp->v_flag & VOBJDIRTY) &&
2628 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
2629 			mtx_unlock(&mntvnode_mtx);
2630 			if (!vget(vp,
2631 			    LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, curthread)) {
2632 				if (VOP_GETVOBJECT(vp, &obj) == 0) {
2633 					vm_object_page_clean(obj, 0, 0,
2634 					    flags == MNT_WAIT ?
2635 					    OBJPC_SYNC : OBJPC_NOSYNC);
2636 				}
2637 				vput(vp);
2638 			}
2639 			mtx_lock(&mntvnode_mtx);
2640 			if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) {
2641 				if (--tries > 0)
2642 					goto loop;
2643 				break;
2644 			}
2645 		}
2646 	}
2647 	mtx_unlock(&mntvnode_mtx);
2648 }
2649 
2650 /*
2651  * Create the VM object needed for VMIO and mmap support.  This
2652  * is done for all VREG files in the system.  Some filesystems might
2653  * afford the additional metadata buffering capability of the
2654  * VMIO code by making the device node be VMIO mode also.
2655  *
2656  * vp must be locked when vfs_object_create is called.
2657  */
2658 int
2659 vfs_object_create(vp, td, cred)
2660 	struct vnode *vp;
2661 	struct thread *td;
2662 	struct ucred *cred;
2663 {
2664 	GIANT_REQUIRED;
2665 	return (VOP_CREATEVOBJECT(vp, cred, td));
2666 }
2667 
2668 /*
2669  * Mark a vnode as free, putting it up for recycling.
2670  */
2671 void
2672 vfree(vp)
2673 	struct vnode *vp;
2674 {
2675 	int s;
2676 
2677 	s = splbio();
2678 	mtx_lock(&vnode_free_list_mtx);
2679 	KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free"));
2680 	if (vp->v_flag & VAGE) {
2681 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2682 	} else {
2683 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2684 	}
2685 	freevnodes++;
2686 	mtx_unlock(&vnode_free_list_mtx);
2687 	vp->v_flag &= ~VAGE;
2688 	vp->v_flag |= VFREE;
2689 	splx(s);
2690 }
2691 
2692 /*
2693  * Opposite of vfree() - mark a vnode as in use.
2694  */
2695 void
2696 vbusy(vp)
2697 	struct vnode *vp;
2698 {
2699 	int s;
2700 
2701 	s = splbio();
2702 	mtx_lock(&vnode_free_list_mtx);
2703 	KASSERT((vp->v_flag & VFREE) != 0, ("vnode not free"));
2704 	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2705 	freevnodes--;
2706 	mtx_unlock(&vnode_free_list_mtx);
2707 	vp->v_flag &= ~(VFREE|VAGE);
2708 	splx(s);
2709 }
2710 
2711 /*
2712  * Record a process's interest in events which might happen to
2713  * a vnode.  Because poll uses the historic select-style interface
2714  * internally, this routine serves as both the ``check for any
2715  * pending events'' and the ``record my interest in future events''
2716  * functions.  (These are done together, while the lock is held,
2717  * to avoid race conditions.)
2718  */
2719 int
2720 vn_pollrecord(vp, td, events)
2721 	struct vnode *vp;
2722 	struct thread *td;
2723 	short events;
2724 {
2725 
2726 	if (vp->v_pollinfo == NULL)
2727 		v_addpollinfo(vp);
2728 	mtx_lock(&vp->v_pollinfo->vpi_lock);
2729 	if (vp->v_pollinfo->vpi_revents & events) {
2730 		/*
2731 		 * This leaves events we are not interested
2732 		 * in available for the other process which
2733 		 * which presumably had requested them
2734 		 * (otherwise they would never have been
2735 		 * recorded).
2736 		 */
2737 		events &= vp->v_pollinfo->vpi_revents;
2738 		vp->v_pollinfo->vpi_revents &= ~events;
2739 
2740 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
2741 		return events;
2742 	}
2743 	vp->v_pollinfo->vpi_events |= events;
2744 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
2745 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
2746 	return 0;
2747 }
2748 
2749 /*
2750  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
2751  * it is possible for us to miss an event due to race conditions, but
2752  * that condition is expected to be rare, so for the moment it is the
2753  * preferred interface.
2754  */
2755 void
2756 vn_pollevent(vp, events)
2757 	struct vnode *vp;
2758 	short events;
2759 {
2760 
2761 	if (vp->v_pollinfo == NULL)
2762 		v_addpollinfo(vp);
2763 	mtx_lock(&vp->v_pollinfo->vpi_lock);
2764 	if (vp->v_pollinfo->vpi_events & events) {
2765 		/*
2766 		 * We clear vpi_events so that we don't
2767 		 * call selwakeup() twice if two events are
2768 		 * posted before the polling process(es) is
2769 		 * awakened.  This also ensures that we take at
2770 		 * most one selwakeup() if the polling process
2771 		 * is no longer interested.  However, it does
2772 		 * mean that only one event can be noticed at
2773 		 * a time.  (Perhaps we should only clear those
2774 		 * event bits which we note?) XXX
2775 		 */
2776 		vp->v_pollinfo->vpi_events = 0;	/* &= ~events ??? */
2777 		vp->v_pollinfo->vpi_revents |= events;
2778 		selwakeup(&vp->v_pollinfo->vpi_selinfo);
2779 	}
2780 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
2781 }
2782 
2783 /*
2784  * Wake up anyone polling on vp because it is being revoked.
2785  * This depends on dead_poll() returning POLLHUP for correct
2786  * behavior.
2787  */
2788 void
2789 vn_pollgone(vp)
2790 	struct vnode *vp;
2791 {
2792 
2793 	mtx_lock(&vp->v_pollinfo->vpi_lock);
2794         VN_KNOTE(vp, NOTE_REVOKE);
2795 	if (vp->v_pollinfo->vpi_events) {
2796 		vp->v_pollinfo->vpi_events = 0;
2797 		selwakeup(&vp->v_pollinfo->vpi_selinfo);
2798 	}
2799 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
2800 }
2801 
2802 
2803 
2804 /*
2805  * Routine to create and manage a filesystem syncer vnode.
2806  */
2807 #define sync_close ((int (*) __P((struct  vop_close_args *)))nullop)
2808 static int	sync_fsync __P((struct  vop_fsync_args *));
2809 static int	sync_inactive __P((struct  vop_inactive_args *));
2810 static int	sync_reclaim  __P((struct  vop_reclaim_args *));
2811 #define sync_lock ((int (*) __P((struct  vop_lock_args *)))vop_nolock)
2812 #define sync_unlock ((int (*) __P((struct  vop_unlock_args *)))vop_nounlock)
2813 static int	sync_print __P((struct vop_print_args *));
2814 #define sync_islocked ((int(*) __P((struct vop_islocked_args *)))vop_noislocked)
2815 
2816 static vop_t **sync_vnodeop_p;
2817 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
2818 	{ &vop_default_desc,	(vop_t *) vop_eopnotsupp },
2819 	{ &vop_close_desc,	(vop_t *) sync_close },		/* close */
2820 	{ &vop_fsync_desc,	(vop_t *) sync_fsync },		/* fsync */
2821 	{ &vop_inactive_desc,	(vop_t *) sync_inactive },	/* inactive */
2822 	{ &vop_reclaim_desc,	(vop_t *) sync_reclaim },	/* reclaim */
2823 	{ &vop_lock_desc,	(vop_t *) sync_lock },		/* lock */
2824 	{ &vop_unlock_desc,	(vop_t *) sync_unlock },	/* unlock */
2825 	{ &vop_print_desc,	(vop_t *) sync_print },		/* print */
2826 	{ &vop_islocked_desc,	(vop_t *) sync_islocked },	/* islocked */
2827 	{ NULL, NULL }
2828 };
2829 static struct vnodeopv_desc sync_vnodeop_opv_desc =
2830 	{ &sync_vnodeop_p, sync_vnodeop_entries };
2831 
2832 VNODEOP_SET(sync_vnodeop_opv_desc);
2833 
2834 /*
2835  * Create a new filesystem syncer vnode for the specified mount point.
2836  */
2837 int
2838 vfs_allocate_syncvnode(mp)
2839 	struct mount *mp;
2840 {
2841 	struct vnode *vp;
2842 	static long start, incr, next;
2843 	int error;
2844 
2845 	/* Allocate a new vnode */
2846 	if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
2847 		mp->mnt_syncer = NULL;
2848 		return (error);
2849 	}
2850 	vp->v_type = VNON;
2851 	/*
2852 	 * Place the vnode onto the syncer worklist. We attempt to
2853 	 * scatter them about on the list so that they will go off
2854 	 * at evenly distributed times even if all the filesystems
2855 	 * are mounted at once.
2856 	 */
2857 	next += incr;
2858 	if (next == 0 || next > syncer_maxdelay) {
2859 		start /= 2;
2860 		incr /= 2;
2861 		if (start == 0) {
2862 			start = syncer_maxdelay / 2;
2863 			incr = syncer_maxdelay;
2864 		}
2865 		next = start;
2866 	}
2867 	vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
2868 	mp->mnt_syncer = vp;
2869 	return (0);
2870 }
2871 
2872 /*
2873  * Do a lazy sync of the filesystem.
2874  */
2875 static int
2876 sync_fsync(ap)
2877 	struct vop_fsync_args /* {
2878 		struct vnode *a_vp;
2879 		struct ucred *a_cred;
2880 		int a_waitfor;
2881 		struct thread *a_td;
2882 	} */ *ap;
2883 {
2884 	struct vnode *syncvp = ap->a_vp;
2885 	struct mount *mp = syncvp->v_mount;
2886 	struct thread *td = ap->a_td;
2887 	int asyncflag;
2888 
2889 	/*
2890 	 * We only need to do something if this is a lazy evaluation.
2891 	 */
2892 	if (ap->a_waitfor != MNT_LAZY)
2893 		return (0);
2894 
2895 	/*
2896 	 * Move ourselves to the back of the sync list.
2897 	 */
2898 	vn_syncer_add_to_worklist(syncvp, syncdelay);
2899 
2900 	/*
2901 	 * Walk the list of vnodes pushing all that are dirty and
2902 	 * not already on the sync list.
2903 	 */
2904 	mtx_lock(&mountlist_mtx);
2905 	if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) {
2906 		mtx_unlock(&mountlist_mtx);
2907 		return (0);
2908 	}
2909 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
2910 		vfs_unbusy(mp, td);
2911 		return (0);
2912 	}
2913 	asyncflag = mp->mnt_flag & MNT_ASYNC;
2914 	mp->mnt_flag &= ~MNT_ASYNC;
2915 	vfs_msync(mp, MNT_NOWAIT);
2916 	VFS_SYNC(mp, MNT_LAZY, ap->a_cred, td);
2917 	if (asyncflag)
2918 		mp->mnt_flag |= MNT_ASYNC;
2919 	vn_finished_write(mp);
2920 	vfs_unbusy(mp, td);
2921 	return (0);
2922 }
2923 
2924 /*
2925  * The syncer vnode is no referenced.
2926  */
2927 static int
2928 sync_inactive(ap)
2929 	struct vop_inactive_args /* {
2930 		struct vnode *a_vp;
2931 		struct thread *a_td;
2932 	} */ *ap;
2933 {
2934 
2935 	vgone(ap->a_vp);
2936 	return (0);
2937 }
2938 
2939 /*
2940  * The syncer vnode is no longer needed and is being decommissioned.
2941  *
2942  * Modifications to the worklist must be protected at splbio().
2943  */
2944 static int
2945 sync_reclaim(ap)
2946 	struct vop_reclaim_args /* {
2947 		struct vnode *a_vp;
2948 	} */ *ap;
2949 {
2950 	struct vnode *vp = ap->a_vp;
2951 	int s;
2952 
2953 	s = splbio();
2954 	vp->v_mount->mnt_syncer = NULL;
2955 	if (vp->v_flag & VONWORKLST) {
2956 		LIST_REMOVE(vp, v_synclist);
2957 		vp->v_flag &= ~VONWORKLST;
2958 	}
2959 	splx(s);
2960 
2961 	return (0);
2962 }
2963 
2964 /*
2965  * Print out a syncer vnode.
2966  */
2967 static int
2968 sync_print(ap)
2969 	struct vop_print_args /* {
2970 		struct vnode *a_vp;
2971 	} */ *ap;
2972 {
2973 	struct vnode *vp = ap->a_vp;
2974 
2975 	printf("syncer vnode");
2976 	if (vp->v_vnlock != NULL)
2977 		lockmgr_printinfo(vp->v_vnlock);
2978 	printf("\n");
2979 	return (0);
2980 }
2981 
2982 /*
2983  * extract the dev_t from a VCHR
2984  */
2985 dev_t
2986 vn_todev(vp)
2987 	struct vnode *vp;
2988 {
2989 	if (vp->v_type != VCHR)
2990 		return (NODEV);
2991 	return (vp->v_rdev);
2992 }
2993 
2994 /*
2995  * Check if vnode represents a disk device
2996  */
2997 int
2998 vn_isdisk(vp, errp)
2999 	struct vnode *vp;
3000 	int *errp;
3001 {
3002 	struct cdevsw *cdevsw;
3003 
3004 	if (vp->v_type != VCHR) {
3005 		if (errp != NULL)
3006 			*errp = ENOTBLK;
3007 		return (0);
3008 	}
3009 	if (vp->v_rdev == NULL) {
3010 		if (errp != NULL)
3011 			*errp = ENXIO;
3012 		return (0);
3013 	}
3014 	cdevsw = devsw(vp->v_rdev);
3015 	if (cdevsw == NULL) {
3016 		if (errp != NULL)
3017 			*errp = ENXIO;
3018 		return (0);
3019 	}
3020 	if (!(cdevsw->d_flags & D_DISK)) {
3021 		if (errp != NULL)
3022 			*errp = ENOTBLK;
3023 		return (0);
3024 	}
3025 	if (errp != NULL)
3026 		*errp = 0;
3027 	return (1);
3028 }
3029 
3030 /*
3031  * Free data allocated by namei(); see namei(9) for details.
3032  */
3033 void
3034 NDFREE(ndp, flags)
3035      struct nameidata *ndp;
3036      const uint flags;
3037 {
3038 	if (!(flags & NDF_NO_FREE_PNBUF) &&
3039 	    (ndp->ni_cnd.cn_flags & HASBUF)) {
3040 		zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
3041 		ndp->ni_cnd.cn_flags &= ~HASBUF;
3042 	}
3043 	if (!(flags & NDF_NO_DVP_UNLOCK) &&
3044 	    (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
3045 	    ndp->ni_dvp != ndp->ni_vp)
3046 		VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_thread);
3047 	if (!(flags & NDF_NO_DVP_RELE) &&
3048 	    (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
3049 		vrele(ndp->ni_dvp);
3050 		ndp->ni_dvp = NULL;
3051 	}
3052 	if (!(flags & NDF_NO_VP_UNLOCK) &&
3053 	    (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
3054 		VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_thread);
3055 	if (!(flags & NDF_NO_VP_RELE) &&
3056 	    ndp->ni_vp) {
3057 		vrele(ndp->ni_vp);
3058 		ndp->ni_vp = NULL;
3059 	}
3060 	if (!(flags & NDF_NO_STARTDIR_RELE) &&
3061 	    (ndp->ni_cnd.cn_flags & SAVESTART)) {
3062 		vrele(ndp->ni_startdir);
3063 		ndp->ni_startdir = NULL;
3064 	}
3065 }
3066 
3067 /*
3068  * Common file system object access control check routine.  Accepts a
3069  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3070  * and optional call-by-reference privused argument allowing vaccess()
3071  * to indicate to the caller whether privilege was used to satisfy the
3072  * request.  Returns 0 on success, or an errno on failure.
3073  */
3074 int
3075 vaccess(type, file_mode, file_uid, file_gid, acc_mode, cred, privused)
3076 	enum vtype type;
3077 	mode_t file_mode;
3078 	uid_t file_uid;
3079 	gid_t file_gid;
3080 	mode_t acc_mode;
3081 	struct ucred *cred;
3082 	int *privused;
3083 {
3084 	mode_t dac_granted;
3085 #ifdef CAPABILITIES
3086 	mode_t cap_granted;
3087 #endif
3088 
3089 	/*
3090 	 * Look for a normal, non-privileged way to access the file/directory
3091 	 * as requested.  If it exists, go with that.
3092 	 */
3093 
3094 	if (privused != NULL)
3095 		*privused = 0;
3096 
3097 	dac_granted = 0;
3098 
3099 	/* Check the owner. */
3100 	if (cred->cr_uid == file_uid) {
3101 		dac_granted |= VADMIN;
3102 		if (file_mode & S_IXUSR)
3103 			dac_granted |= VEXEC;
3104 		if (file_mode & S_IRUSR)
3105 			dac_granted |= VREAD;
3106 		if (file_mode & S_IWUSR)
3107 			dac_granted |= VWRITE;
3108 
3109 		if ((acc_mode & dac_granted) == acc_mode)
3110 			return (0);
3111 
3112 		goto privcheck;
3113 	}
3114 
3115 	/* Otherwise, check the groups (first match) */
3116 	if (groupmember(file_gid, cred)) {
3117 		if (file_mode & S_IXGRP)
3118 			dac_granted |= VEXEC;
3119 		if (file_mode & S_IRGRP)
3120 			dac_granted |= VREAD;
3121 		if (file_mode & S_IWGRP)
3122 			dac_granted |= VWRITE;
3123 
3124 		if ((acc_mode & dac_granted) == acc_mode)
3125 			return (0);
3126 
3127 		goto privcheck;
3128 	}
3129 
3130 	/* Otherwise, check everyone else. */
3131 	if (file_mode & S_IXOTH)
3132 		dac_granted |= VEXEC;
3133 	if (file_mode & S_IROTH)
3134 		dac_granted |= VREAD;
3135 	if (file_mode & S_IWOTH)
3136 		dac_granted |= VWRITE;
3137 	if ((acc_mode & dac_granted) == acc_mode)
3138 		return (0);
3139 
3140 privcheck:
3141 	if (!suser_xxx(cred, NULL, PRISON_ROOT)) {
3142 		/* XXX audit: privilege used */
3143 		if (privused != NULL)
3144 			*privused = 1;
3145 		return (0);
3146 	}
3147 
3148 #ifdef CAPABILITIES
3149 	/*
3150 	 * Build a capability mask to determine if the set of capabilities
3151 	 * satisfies the requirements when combined with the granted mask
3152 	 * from above.
3153 	 * For each capability, if the capability is required, bitwise
3154 	 * or the request type onto the cap_granted mask.
3155 	 */
3156 	cap_granted = 0;
3157 
3158 	if (type == VDIR) {
3159 		/*
3160 		 * For directories, use CAP_DAC_READ_SEARCH to satisfy
3161 		 * VEXEC requests, instead of CAP_DAC_EXECUTE.
3162 		 */
3163 		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3164 		    !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3165 			cap_granted |= VEXEC;
3166 	} else {
3167 		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3168 		    !cap_check(cred, NULL, CAP_DAC_EXECUTE, PRISON_ROOT))
3169 			cap_granted |= VEXEC;
3170 	}
3171 
3172 	if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
3173 	    !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3174 		cap_granted |= VREAD;
3175 
3176 	if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3177 	    !cap_check(cred, NULL, CAP_DAC_WRITE, PRISON_ROOT))
3178 		cap_granted |= VWRITE;
3179 
3180 	if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3181 	    !cap_check(cred, NULL, CAP_FOWNER, PRISON_ROOT))
3182 		cap_granted |= VADMIN;
3183 
3184 	if ((acc_mode & (cap_granted | dac_granted)) == acc_mode) {
3185 		/* XXX audit: privilege used */
3186 		if (privused != NULL)
3187 			*privused = 1;
3188 		return (0);
3189 	}
3190 #endif
3191 
3192 	return ((acc_mode & VADMIN) ? EPERM : EACCES);
3193 }
3194 
3195