xref: /freebsd/sys/kern/vfs_subr.c (revision c17d43407fe04133a94055b0dbc7ea8965654a9f)
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/uma.h>
73 
74 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
75 
76 static void	addalias(struct vnode *vp, dev_t nvp_rdev);
77 static void	insmntque(struct vnode *vp, struct mount *mp);
78 static void	vclean(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 uma_zone_t vnode_zone;
192 static uma_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 = uma_zalloc(vnodepoll_zone, M_WAITOK);
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 = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
280 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
281 	vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
282 	      NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
283 	/*
284 	 * Initialize the filesystem syncer.
285 	 */
286 	syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
287 		&syncer_mask);
288 	syncer_maxdelay = syncer_mask + 1;
289 }
290 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL)
291 
292 
293 /*
294  * Mark a mount point as busy. Used to synchronize access and to delay
295  * unmounting. Interlock is not released on failure.
296  */
297 int
298 vfs_busy(mp, flags, interlkp, td)
299 	struct mount *mp;
300 	int flags;
301 	struct mtx *interlkp;
302 	struct thread *td;
303 {
304 	int lkflags;
305 
306 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
307 		if (flags & LK_NOWAIT)
308 			return (ENOENT);
309 		mp->mnt_kern_flag |= MNTK_MWAIT;
310 		/*
311 		 * Since all busy locks are shared except the exclusive
312 		 * lock granted when unmounting, the only place that a
313 		 * wakeup needs to be done is at the release of the
314 		 * exclusive lock at the end of dounmount.
315 		 */
316 		msleep((caddr_t)mp, interlkp, PVFS, "vfs_busy", 0);
317 		return (ENOENT);
318 	}
319 	lkflags = LK_SHARED | LK_NOPAUSE;
320 	if (interlkp)
321 		lkflags |= LK_INTERLOCK;
322 	if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
323 		panic("vfs_busy: unexpected lock failure");
324 	return (0);
325 }
326 
327 /*
328  * Free a busy filesystem.
329  */
330 void
331 vfs_unbusy(mp, td)
332 	struct mount *mp;
333 	struct thread *td;
334 {
335 
336 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
337 }
338 
339 /*
340  * Lookup a filesystem type, and if found allocate and initialize
341  * a mount structure for it.
342  *
343  * Devname is usually updated by mount(8) after booting.
344  */
345 int
346 vfs_rootmountalloc(fstypename, devname, mpp)
347 	char *fstypename;
348 	char *devname;
349 	struct mount **mpp;
350 {
351 	struct thread *td = curthread;	/* XXX */
352 	struct vfsconf *vfsp;
353 	struct mount *mp;
354 
355 	if (fstypename == NULL)
356 		return (ENODEV);
357 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
358 		if (!strcmp(vfsp->vfc_name, fstypename))
359 			break;
360 	if (vfsp == NULL)
361 		return (ENODEV);
362 	mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO);
363 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE);
364 	(void)vfs_busy(mp, LK_NOWAIT, 0, td);
365 	TAILQ_INIT(&mp->mnt_nvnodelist);
366 	TAILQ_INIT(&mp->mnt_reservedvnlist);
367 	mp->mnt_vfc = vfsp;
368 	mp->mnt_op = vfsp->vfc_vfsops;
369 	mp->mnt_flag = MNT_RDONLY;
370 	mp->mnt_vnodecovered = NULLVP;
371 	vfsp->vfc_refcount++;
372 	mp->mnt_iosize_max = DFLTPHYS;
373 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
374 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
375 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
376 	mp->mnt_stat.f_mntonname[0] = '/';
377 	mp->mnt_stat.f_mntonname[1] = 0;
378 	(void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
379 	*mpp = mp;
380 	return (0);
381 }
382 
383 /*
384  * Find an appropriate filesystem to use for the root. If a filesystem
385  * has not been preselected, walk through the list of known filesystems
386  * trying those that have mountroot routines, and try them until one
387  * works or we have tried them all.
388  */
389 #ifdef notdef	/* XXX JH */
390 int
391 lite2_vfs_mountroot()
392 {
393 	struct vfsconf *vfsp;
394 	extern int (*lite2_mountroot)(void);
395 	int error;
396 
397 	if (lite2_mountroot != NULL)
398 		return ((*lite2_mountroot)());
399 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
400 		if (vfsp->vfc_mountroot == NULL)
401 			continue;
402 		if ((error = (*vfsp->vfc_mountroot)()) == 0)
403 			return (0);
404 		printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error);
405 	}
406 	return (ENODEV);
407 }
408 #endif
409 
410 /*
411  * Lookup a mount point by filesystem identifier.
412  */
413 struct mount *
414 vfs_getvfs(fsid)
415 	fsid_t *fsid;
416 {
417 	register struct mount *mp;
418 
419 	mtx_lock(&mountlist_mtx);
420 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
421 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
422 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
423 			mtx_unlock(&mountlist_mtx);
424 			return (mp);
425 	    }
426 	}
427 	mtx_unlock(&mountlist_mtx);
428 	return ((struct mount *) 0);
429 }
430 
431 /*
432  * Get a new unique fsid.  Try to make its val[0] unique, since this value
433  * will be used to create fake device numbers for stat().  Also try (but
434  * not so hard) make its val[0] unique mod 2^16, since some emulators only
435  * support 16-bit device numbers.  We end up with unique val[0]'s for the
436  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
437  *
438  * Keep in mind that several mounts may be running in parallel.  Starting
439  * the search one past where the previous search terminated is both a
440  * micro-optimization and a defense against returning the same fsid to
441  * different mounts.
442  */
443 void
444 vfs_getnewfsid(mp)
445 	struct mount *mp;
446 {
447 	static u_int16_t mntid_base;
448 	fsid_t tfsid;
449 	int mtype;
450 
451 	mtx_lock(&mntid_mtx);
452 	mtype = mp->mnt_vfc->vfc_typenum;
453 	tfsid.val[1] = mtype;
454 	mtype = (mtype & 0xFF) << 24;
455 	for (;;) {
456 		tfsid.val[0] = makeudev(255,
457 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
458 		mntid_base++;
459 		if (vfs_getvfs(&tfsid) == NULL)
460 			break;
461 	}
462 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
463 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
464 	mtx_unlock(&mntid_mtx);
465 }
466 
467 /*
468  * Knob to control the precision of file timestamps:
469  *
470  *   0 = seconds only; nanoseconds zeroed.
471  *   1 = seconds and nanoseconds, accurate within 1/HZ.
472  *   2 = seconds and nanoseconds, truncated to microseconds.
473  * >=3 = seconds and nanoseconds, maximum precision.
474  */
475 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
476 
477 static int timestamp_precision = TSP_SEC;
478 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
479     &timestamp_precision, 0, "");
480 
481 /*
482  * Get a current timestamp.
483  */
484 void
485 vfs_timestamp(tsp)
486 	struct timespec *tsp;
487 {
488 	struct timeval tv;
489 
490 	switch (timestamp_precision) {
491 	case TSP_SEC:
492 		tsp->tv_sec = time_second;
493 		tsp->tv_nsec = 0;
494 		break;
495 	case TSP_HZ:
496 		getnanotime(tsp);
497 		break;
498 	case TSP_USEC:
499 		microtime(&tv);
500 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
501 		break;
502 	case TSP_NSEC:
503 	default:
504 		nanotime(tsp);
505 		break;
506 	}
507 }
508 
509 /*
510  * Set vnode attributes to VNOVAL
511  */
512 void
513 vattr_null(vap)
514 	register struct vattr *vap;
515 {
516 
517 	vap->va_type = VNON;
518 	vap->va_size = VNOVAL;
519 	vap->va_bytes = VNOVAL;
520 	vap->va_mode = VNOVAL;
521 	vap->va_nlink = VNOVAL;
522 	vap->va_uid = VNOVAL;
523 	vap->va_gid = VNOVAL;
524 	vap->va_fsid = VNOVAL;
525 	vap->va_fileid = VNOVAL;
526 	vap->va_blocksize = VNOVAL;
527 	vap->va_rdev = VNOVAL;
528 	vap->va_atime.tv_sec = VNOVAL;
529 	vap->va_atime.tv_nsec = VNOVAL;
530 	vap->va_mtime.tv_sec = VNOVAL;
531 	vap->va_mtime.tv_nsec = VNOVAL;
532 	vap->va_ctime.tv_sec = VNOVAL;
533 	vap->va_ctime.tv_nsec = VNOVAL;
534 	vap->va_flags = VNOVAL;
535 	vap->va_gen = VNOVAL;
536 	vap->va_vaflags = 0;
537 }
538 
539 /*
540  * This routine is called when we have too many vnodes.  It attempts
541  * to free <count> vnodes and will potentially free vnodes that still
542  * have VM backing store (VM backing store is typically the cause
543  * of a vnode blowout so we want to do this).  Therefore, this operation
544  * is not considered cheap.
545  *
546  * A number of conditions may prevent a vnode from being reclaimed.
547  * the buffer cache may have references on the vnode, a directory
548  * vnode may still have references due to the namei cache representing
549  * underlying files, or the vnode may be in active use.   It is not
550  * desireable to reuse such vnodes.  These conditions may cause the
551  * number of vnodes to reach some minimum value regardless of what
552  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
553  */
554 static int
555 vlrureclaim(struct mount *mp, int count)
556 {
557 	struct vnode *vp;
558 	int done;
559 	int trigger;
560 	int usevnodes;
561 
562 	/*
563 	 * Calculate the trigger point, don't allow user
564 	 * screwups to blow us up.   This prevents us from
565 	 * recycling vnodes with lots of resident pages.  We
566 	 * aren't trying to free memory, we are trying to
567 	 * free vnodes.
568 	 */
569 	usevnodes = desiredvnodes;
570 	if (usevnodes <= 0)
571 		usevnodes = 1;
572 	trigger = cnt.v_page_count * 2 / usevnodes;
573 
574 	done = 0;
575 	mtx_lock(&mntvnode_mtx);
576 	while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
577 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
578 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
579 
580 		if (vp->v_type != VNON &&
581 		    vp->v_type != VBAD &&
582 		    VMIGHTFREE(vp) &&           /* critical path opt */
583 		    (vp->v_object == NULL || vp->v_object->resident_page_count < trigger) &&
584 		    mtx_trylock(&vp->v_interlock)
585 		) {
586 			mtx_unlock(&mntvnode_mtx);
587 			if (VMIGHTFREE(vp)) {
588 				vgonel(vp, curthread);
589 				done++;
590 			} else {
591 				mtx_unlock(&vp->v_interlock);
592 			}
593 			mtx_lock(&mntvnode_mtx);
594 		}
595 		--count;
596 	}
597 	mtx_unlock(&mntvnode_mtx);
598 	return done;
599 }
600 
601 /*
602  * Attempt to recycle vnodes in a context that is always safe to block.
603  * Calling vlrurecycle() from the bowels of file system code has some
604  * interesting deadlock problems.
605  */
606 static struct proc *vnlruproc;
607 static int vnlruproc_sig;
608 
609 static void
610 vnlru_proc(void)
611 {
612 	struct mount *mp, *nmp;
613 	int s;
614 	int done;
615 	struct proc *p = vnlruproc;
616 	struct thread *td = FIRST_THREAD_IN_PROC(p);	/* XXXKSE */
617 
618 	mtx_lock(&Giant);
619 
620 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
621 	    SHUTDOWN_PRI_FIRST);
622 
623 	s = splbio();
624 	for (;;) {
625 		kthread_suspend_check(p);
626 		if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
627 			vnlruproc_sig = 0;
628 			tsleep(vnlruproc, PVFS, "vlruwt", 0);
629 			continue;
630 		}
631 		done = 0;
632 		mtx_lock(&mountlist_mtx);
633 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
634 			if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
635 				nmp = TAILQ_NEXT(mp, mnt_list);
636 				continue;
637 			}
638 			done += vlrureclaim(mp, 10);
639 			mtx_lock(&mountlist_mtx);
640 			nmp = TAILQ_NEXT(mp, mnt_list);
641 			vfs_unbusy(mp, td);
642 		}
643 		mtx_unlock(&mountlist_mtx);
644 		if (done == 0) {
645 #if 0
646 			/* These messages are temporary debugging aids */
647 			if (vnlru_nowhere < 5)
648 				printf("vnlru process getting nowhere..\n");
649 			else if (vnlru_nowhere == 5)
650 				printf("vnlru process messages stopped.\n");
651 #endif
652 			vnlru_nowhere++;
653 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
654 		}
655 	}
656 	splx(s);
657 }
658 
659 static struct kproc_desc vnlru_kp = {
660 	"vnlru",
661 	vnlru_proc,
662 	&vnlruproc
663 };
664 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
665 
666 
667 /*
668  * Routines having to do with the management of the vnode table.
669  */
670 
671 /*
672  * Return the next vnode from the free list.
673  */
674 int
675 getnewvnode(tag, mp, vops, vpp)
676 	enum vtagtype tag;
677 	struct mount *mp;
678 	vop_t **vops;
679 	struct vnode **vpp;
680 {
681 	int s;
682 	struct thread *td = curthread;	/* XXX */
683 	struct vnode *vp = NULL;
684 	struct mount *vnmp;
685 	vm_object_t object;
686 
687 	s = splbio();
688 	/*
689 	 * Try to reuse vnodes if we hit the max.  This situation only
690 	 * occurs in certain large-memory (2G+) situations.  We cannot
691 	 * attempt to directly reclaim vnodes due to nasty recursion
692 	 * problems.
693 	 */
694 	if (vnlruproc_sig == 0 && numvnodes - freevnodes > desiredvnodes) {
695 		vnlruproc_sig = 1;      /* avoid unnecessary wakeups */
696 		wakeup(vnlruproc);
697 	}
698 
699 	/*
700 	 * Attempt to reuse a vnode already on the free list, allocating
701 	 * a new vnode if we can't find one or if we have not reached a
702 	 * good minimum for good LRU performance.
703 	 */
704 
705 	mtx_lock(&vnode_free_list_mtx);
706 
707 	if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
708 		int count;
709 
710 		for (count = 0; count < freevnodes; count++) {
711 			vp = TAILQ_FIRST(&vnode_free_list);
712 			if (vp == NULL || vp->v_usecount)
713 				panic("getnewvnode: free vnode isn't");
714 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
715 
716 			/*
717 			 * Don't recycle if we still have cached pages or if
718 			 * we cannot get the interlock.
719 			 */
720 			if ((VOP_GETVOBJECT(vp, &object) == 0 &&
721 			     (object->resident_page_count ||
722 			      object->ref_count)) ||
723 			     !mtx_trylock(&vp->v_interlock)) {
724 				TAILQ_INSERT_TAIL(&vnode_free_list, vp,
725 						    v_freelist);
726 				vp = NULL;
727 				continue;
728 			}
729 			if (LIST_FIRST(&vp->v_cache_src)) {
730 				/*
731 				 * note: nameileafonly sysctl is temporary,
732 				 * for debugging only, and will eventually be
733 				 * removed.
734 				 */
735 				if (nameileafonly > 0) {
736 					/*
737 					 * Do not reuse namei-cached directory
738 					 * vnodes that have cached
739 					 * subdirectories.
740 					 */
741 					if (cache_leaf_test(vp) < 0) {
742 						mtx_unlock(&vp->v_interlock);
743 						TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
744 						vp = NULL;
745 						continue;
746 					}
747 				} else if (nameileafonly < 0 ||
748 					    vmiodirenable == 0) {
749 					/*
750 					 * Do not reuse namei-cached directory
751 					 * vnodes if nameileafonly is -1 or
752 					 * if VMIO backing for directories is
753 					 * turned off (otherwise we reuse them
754 					 * too quickly).
755 					 */
756 					mtx_unlock(&vp->v_interlock);
757 					TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
758 					vp = NULL;
759 					continue;
760 				}
761 			}
762 			/*
763 			 * Skip over it if its filesystem is being suspended.
764 			 */
765 			if (vn_start_write(vp, &vnmp, V_NOWAIT) == 0)
766 				break;
767 			mtx_unlock(&vp->v_interlock);
768 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
769 			vp = NULL;
770 		}
771 	}
772 	if (vp) {
773 		vp->v_flag |= VDOOMED;
774 		vp->v_flag &= ~VFREE;
775 		freevnodes--;
776 		mtx_unlock(&vnode_free_list_mtx);
777 		cache_purge(vp);
778 		if (vp->v_type != VBAD) {
779 			vgonel(vp, td);
780 		} else {
781 			mtx_unlock(&vp->v_interlock);
782 		}
783 		vn_finished_write(vnmp);
784 
785 #ifdef INVARIANTS
786 		{
787 			int s;
788 
789 			if (vp->v_data)
790 				panic("cleaned vnode isn't");
791 			s = splbio();
792 			if (vp->v_numoutput)
793 				panic("Clean vnode has pending I/O's");
794 			splx(s);
795 			if (vp->v_writecount != 0)
796 				panic("Non-zero write count");
797 		}
798 #endif
799 		if (vp->v_pollinfo) {
800 			mtx_destroy(&vp->v_pollinfo->vpi_lock);
801 			uma_zfree(vnodepoll_zone, vp->v_pollinfo);
802 		}
803 		vp->v_pollinfo = NULL;
804 		vp->v_flag = 0;
805 		vp->v_lastw = 0;
806 		vp->v_lasta = 0;
807 		vp->v_cstart = 0;
808 		vp->v_clen = 0;
809 		vp->v_socket = 0;
810 	} else {
811 		mtx_unlock(&vnode_free_list_mtx);
812 		vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK);
813 		bzero((char *) vp, sizeof *vp);
814 		mtx_init(&vp->v_interlock, "vnode interlock", MTX_DEF);
815 		vp->v_dd = vp;
816 		cache_purge(vp);
817 		LIST_INIT(&vp->v_cache_src);
818 		TAILQ_INIT(&vp->v_cache_dst);
819 		numvnodes++;
820 	}
821 
822 	TAILQ_INIT(&vp->v_cleanblkhd);
823 	TAILQ_INIT(&vp->v_dirtyblkhd);
824 	vp->v_type = VNON;
825 	vp->v_tag = tag;
826 	vp->v_op = vops;
827 	lockinit(&vp->v_lock, PVFS, "vnlock", VLKTIMEOUT, LK_NOPAUSE);
828 	insmntque(vp, mp);
829 	*vpp = vp;
830 	vp->v_usecount = 1;
831 	vp->v_data = 0;
832 
833 	splx(s);
834 
835 	vfs_object_create(vp, td, td->td_ucred);
836 
837 #if 0
838 	vnodeallocs++;
839 	if (vnodeallocs % vnoderecycleperiod == 0 &&
840 	    freevnodes < vnoderecycleminfreevn &&
841 	    vnoderecyclemintotalvn < numvnodes) {
842 		/* Recycle vnodes. */
843 		cache_purgeleafdirs(vnoderecyclenumber);
844 	}
845 #endif
846 
847 	return (0);
848 }
849 
850 /*
851  * Move a vnode from one mount queue to another.
852  */
853 static void
854 insmntque(vp, mp)
855 	register struct vnode *vp;
856 	register struct mount *mp;
857 {
858 
859 	mtx_lock(&mntvnode_mtx);
860 	/*
861 	 * Delete from old mount point vnode list, if on one.
862 	 */
863 	if (vp->v_mount != NULL)
864 		TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
865 	/*
866 	 * Insert into list of vnodes for the new mount point, if available.
867 	 */
868 	if ((vp->v_mount = mp) == NULL) {
869 		mtx_unlock(&mntvnode_mtx);
870 		return;
871 	}
872 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
873 	mtx_unlock(&mntvnode_mtx);
874 }
875 
876 /*
877  * Update outstanding I/O count and do wakeup if requested.
878  */
879 void
880 vwakeup(bp)
881 	register struct buf *bp;
882 {
883 	register struct vnode *vp;
884 
885 	bp->b_flags &= ~B_WRITEINPROG;
886 	if ((vp = bp->b_vp)) {
887 		vp->v_numoutput--;
888 		if (vp->v_numoutput < 0)
889 			panic("vwakeup: neg numoutput");
890 		if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
891 			vp->v_flag &= ~VBWAIT;
892 			wakeup((caddr_t) &vp->v_numoutput);
893 		}
894 	}
895 }
896 
897 /*
898  * Flush out and invalidate all buffers associated with a vnode.
899  * Called with the underlying object locked.
900  */
901 int
902 vinvalbuf(vp, flags, cred, td, slpflag, slptimeo)
903 	register struct vnode *vp;
904 	int flags;
905 	struct ucred *cred;
906 	struct thread *td;
907 	int slpflag, slptimeo;
908 {
909 	register struct buf *bp;
910 	struct buf *nbp, *blist;
911 	int s, error;
912 	vm_object_t object;
913 
914 	GIANT_REQUIRED;
915 
916 	if (flags & V_SAVE) {
917 		s = splbio();
918 		while (vp->v_numoutput) {
919 			vp->v_flag |= VBWAIT;
920 			error = tsleep((caddr_t)&vp->v_numoutput,
921 			    slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo);
922 			if (error) {
923 				splx(s);
924 				return (error);
925 			}
926 		}
927 		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
928 			splx(s);
929 			if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, td)) != 0)
930 				return (error);
931 			s = splbio();
932 			if (vp->v_numoutput > 0 ||
933 			    !TAILQ_EMPTY(&vp->v_dirtyblkhd))
934 				panic("vinvalbuf: dirty bufs");
935 		}
936 		splx(s);
937   	}
938 	s = splbio();
939 	for (;;) {
940 		blist = TAILQ_FIRST(&vp->v_cleanblkhd);
941 		if (!blist)
942 			blist = TAILQ_FIRST(&vp->v_dirtyblkhd);
943 		if (!blist)
944 			break;
945 
946 		for (bp = blist; bp; bp = nbp) {
947 			nbp = TAILQ_NEXT(bp, b_vnbufs);
948 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
949 				error = BUF_TIMELOCK(bp,
950 				    LK_EXCLUSIVE | LK_SLEEPFAIL,
951 				    "vinvalbuf", slpflag, slptimeo);
952 				if (error == ENOLCK)
953 					break;
954 				splx(s);
955 				return (error);
956 			}
957 			/*
958 			 * XXX Since there are no node locks for NFS, I
959 			 * believe there is a slight chance that a delayed
960 			 * write will occur while sleeping just above, so
961 			 * check for it.  Note that vfs_bio_awrite expects
962 			 * buffers to reside on a queue, while BUF_WRITE and
963 			 * brelse do not.
964 			 */
965 			if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
966 				(flags & V_SAVE)) {
967 
968 				if (bp->b_vp == vp) {
969 					if (bp->b_flags & B_CLUSTEROK) {
970 						BUF_UNLOCK(bp);
971 						vfs_bio_awrite(bp);
972 					} else {
973 						bremfree(bp);
974 						bp->b_flags |= B_ASYNC;
975 						BUF_WRITE(bp);
976 					}
977 				} else {
978 					bremfree(bp);
979 					(void) BUF_WRITE(bp);
980 				}
981 				break;
982 			}
983 			bremfree(bp);
984 			bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
985 			bp->b_flags &= ~B_ASYNC;
986 			brelse(bp);
987 		}
988 	}
989 
990 	/*
991 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
992 	 * have write I/O in-progress but if there is a VM object then the
993 	 * VM object can also have read-I/O in-progress.
994 	 */
995 	do {
996 		while (vp->v_numoutput > 0) {
997 			vp->v_flag |= VBWAIT;
998 			tsleep(&vp->v_numoutput, PVM, "vnvlbv", 0);
999 		}
1000 		if (VOP_GETVOBJECT(vp, &object) == 0) {
1001 			while (object->paging_in_progress)
1002 			vm_object_pip_sleep(object, "vnvlbx");
1003 		}
1004 	} while (vp->v_numoutput > 0);
1005 
1006 	splx(s);
1007 
1008 	/*
1009 	 * Destroy the copy in the VM cache, too.
1010 	 */
1011 	mtx_lock(&vp->v_interlock);
1012 	if (VOP_GETVOBJECT(vp, &object) == 0) {
1013 		vm_object_page_remove(object, 0, 0,
1014 			(flags & V_SAVE) ? TRUE : FALSE);
1015 	}
1016 	mtx_unlock(&vp->v_interlock);
1017 
1018 	if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd))
1019 		panic("vinvalbuf: flush failed");
1020 	return (0);
1021 }
1022 
1023 /*
1024  * Truncate a file's buffer and pages to a specified length.  This
1025  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1026  * sync activity.
1027  */
1028 int
1029 vtruncbuf(vp, cred, td, length, blksize)
1030 	register struct vnode *vp;
1031 	struct ucred *cred;
1032 	struct thread *td;
1033 	off_t length;
1034 	int blksize;
1035 {
1036 	register struct buf *bp;
1037 	struct buf *nbp;
1038 	int s, anyfreed;
1039 	int trunclbn;
1040 
1041 	/*
1042 	 * Round up to the *next* lbn.
1043 	 */
1044 	trunclbn = (length + blksize - 1) / blksize;
1045 
1046 	s = splbio();
1047 restart:
1048 	anyfreed = 1;
1049 	for (;anyfreed;) {
1050 		anyfreed = 0;
1051 		for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
1052 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1053 			if (bp->b_lblkno >= trunclbn) {
1054 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1055 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1056 					goto restart;
1057 				} else {
1058 					bremfree(bp);
1059 					bp->b_flags |= (B_INVAL | B_RELBUF);
1060 					bp->b_flags &= ~B_ASYNC;
1061 					brelse(bp);
1062 					anyfreed = 1;
1063 				}
1064 				if (nbp &&
1065 				    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1066 				    (nbp->b_vp != vp) ||
1067 				    (nbp->b_flags & B_DELWRI))) {
1068 					goto restart;
1069 				}
1070 			}
1071 		}
1072 
1073 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1074 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1075 			if (bp->b_lblkno >= trunclbn) {
1076 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1077 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1078 					goto restart;
1079 				} else {
1080 					bremfree(bp);
1081 					bp->b_flags |= (B_INVAL | B_RELBUF);
1082 					bp->b_flags &= ~B_ASYNC;
1083 					brelse(bp);
1084 					anyfreed = 1;
1085 				}
1086 				if (nbp &&
1087 				    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1088 				    (nbp->b_vp != vp) ||
1089 				    (nbp->b_flags & B_DELWRI) == 0)) {
1090 					goto restart;
1091 				}
1092 			}
1093 		}
1094 	}
1095 
1096 	if (length > 0) {
1097 restartsync:
1098 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1099 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1100 			if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
1101 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1102 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1103 					goto restart;
1104 				} else {
1105 					bremfree(bp);
1106 					if (bp->b_vp == vp) {
1107 						bp->b_flags |= B_ASYNC;
1108 					} else {
1109 						bp->b_flags &= ~B_ASYNC;
1110 					}
1111 					BUF_WRITE(bp);
1112 				}
1113 				goto restartsync;
1114 			}
1115 
1116 		}
1117 	}
1118 
1119 	while (vp->v_numoutput > 0) {
1120 		vp->v_flag |= VBWAIT;
1121 		tsleep(&vp->v_numoutput, PVM, "vbtrunc", 0);
1122 	}
1123 
1124 	splx(s);
1125 
1126 	vnode_pager_setsize(vp, length);
1127 
1128 	return (0);
1129 }
1130 
1131 /*
1132  * Associate a buffer with a vnode.
1133  */
1134 void
1135 bgetvp(vp, bp)
1136 	register struct vnode *vp;
1137 	register struct buf *bp;
1138 {
1139 	int s;
1140 
1141 	KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
1142 
1143 	vhold(vp);
1144 	bp->b_vp = vp;
1145 	bp->b_dev = vn_todev(vp);
1146 	/*
1147 	 * Insert onto list for new vnode.
1148 	 */
1149 	s = splbio();
1150 	bp->b_xflags |= BX_VNCLEAN;
1151 	bp->b_xflags &= ~BX_VNDIRTY;
1152 	TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
1153 	splx(s);
1154 }
1155 
1156 /*
1157  * Disassociate a buffer from a vnode.
1158  */
1159 void
1160 brelvp(bp)
1161 	register struct buf *bp;
1162 {
1163 	struct vnode *vp;
1164 	struct buflists *listheadp;
1165 	int s;
1166 
1167 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1168 
1169 	/*
1170 	 * Delete from old vnode list, if on one.
1171 	 */
1172 	vp = bp->b_vp;
1173 	s = splbio();
1174 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1175 		if (bp->b_xflags & BX_VNDIRTY)
1176 			listheadp = &vp->v_dirtyblkhd;
1177 		else
1178 			listheadp = &vp->v_cleanblkhd;
1179 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1180 		bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1181 	}
1182 	if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1183 		vp->v_flag &= ~VONWORKLST;
1184 		LIST_REMOVE(vp, v_synclist);
1185 	}
1186 	splx(s);
1187 	bp->b_vp = (struct vnode *) 0;
1188 	vdrop(vp);
1189 }
1190 
1191 /*
1192  * Add an item to the syncer work queue.
1193  */
1194 static void
1195 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
1196 {
1197 	int s, slot;
1198 
1199 	s = splbio();
1200 
1201 	if (vp->v_flag & VONWORKLST) {
1202 		LIST_REMOVE(vp, v_synclist);
1203 	}
1204 
1205 	if (delay > syncer_maxdelay - 2)
1206 		delay = syncer_maxdelay - 2;
1207 	slot = (syncer_delayno + delay) & syncer_mask;
1208 
1209 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
1210 	vp->v_flag |= VONWORKLST;
1211 	splx(s);
1212 }
1213 
1214 struct  proc *updateproc;
1215 static void sched_sync(void);
1216 static struct kproc_desc up_kp = {
1217 	"syncer",
1218 	sched_sync,
1219 	&updateproc
1220 };
1221 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1222 
1223 /*
1224  * System filesystem synchronizer daemon.
1225  */
1226 void
1227 sched_sync(void)
1228 {
1229 	struct synclist *slp;
1230 	struct vnode *vp;
1231 	struct mount *mp;
1232 	long starttime;
1233 	int s;
1234 	struct thread *td = FIRST_THREAD_IN_PROC(updateproc);  /* XXXKSE */
1235 
1236 	mtx_lock(&Giant);
1237 
1238 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, td->td_proc,
1239 	    SHUTDOWN_PRI_LAST);
1240 
1241 	for (;;) {
1242 		kthread_suspend_check(td->td_proc);
1243 
1244 		starttime = time_second;
1245 
1246 		/*
1247 		 * Push files whose dirty time has expired.  Be careful
1248 		 * of interrupt race on slp queue.
1249 		 */
1250 		s = splbio();
1251 		slp = &syncer_workitem_pending[syncer_delayno];
1252 		syncer_delayno += 1;
1253 		if (syncer_delayno == syncer_maxdelay)
1254 			syncer_delayno = 0;
1255 		splx(s);
1256 
1257 		while ((vp = LIST_FIRST(slp)) != NULL) {
1258 			if (VOP_ISLOCKED(vp, NULL) == 0 &&
1259 			    vn_start_write(vp, &mp, V_NOWAIT) == 0) {
1260 				vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1261 				(void) VOP_FSYNC(vp, td->td_ucred, MNT_LAZY, td);
1262 				VOP_UNLOCK(vp, 0, td);
1263 				vn_finished_write(mp);
1264 			}
1265 			s = splbio();
1266 			if (LIST_FIRST(slp) == vp) {
1267 				/*
1268 				 * Note: v_tag VT_VFS vps can remain on the
1269 				 * worklist too with no dirty blocks, but
1270 				 * since sync_fsync() moves it to a different
1271 				 * slot we are safe.
1272 				 */
1273 				if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1274 				    !vn_isdisk(vp, NULL))
1275 					panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
1276 				/*
1277 				 * Put us back on the worklist.  The worklist
1278 				 * routine will remove us from our current
1279 				 * position and then add us back in at a later
1280 				 * position.
1281 				 */
1282 				vn_syncer_add_to_worklist(vp, syncdelay);
1283 			}
1284 			splx(s);
1285 		}
1286 
1287 		/*
1288 		 * Do soft update processing.
1289 		 */
1290 #ifdef SOFTUPDATES
1291 		softdep_process_worklist(NULL);
1292 #endif
1293 
1294 		/*
1295 		 * The variable rushjob allows the kernel to speed up the
1296 		 * processing of the filesystem syncer process. A rushjob
1297 		 * value of N tells the filesystem syncer to process the next
1298 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1299 		 * is used by the soft update code to speed up the filesystem
1300 		 * syncer process when the incore state is getting so far
1301 		 * ahead of the disk that the kernel memory pool is being
1302 		 * threatened with exhaustion.
1303 		 */
1304 		if (rushjob > 0) {
1305 			rushjob -= 1;
1306 			continue;
1307 		}
1308 		/*
1309 		 * If it has taken us less than a second to process the
1310 		 * current work, then wait. Otherwise start right over
1311 		 * again. We can still lose time if any single round
1312 		 * takes more than two seconds, but it does not really
1313 		 * matter as we are just trying to generally pace the
1314 		 * filesystem activity.
1315 		 */
1316 		if (time_second == starttime)
1317 			tsleep(&lbolt, PPAUSE, "syncer", 0);
1318 	}
1319 }
1320 
1321 /*
1322  * Request the syncer daemon to speed up its work.
1323  * We never push it to speed up more than half of its
1324  * normal turn time, otherwise it could take over the cpu.
1325  * XXXKSE  only one update?
1326  */
1327 int
1328 speedup_syncer()
1329 {
1330 
1331 	mtx_lock_spin(&sched_lock);
1332 	if (FIRST_THREAD_IN_PROC(updateproc)->td_wchan == &lbolt) /* XXXKSE */
1333 		setrunnable(FIRST_THREAD_IN_PROC(updateproc));
1334 	mtx_unlock_spin(&sched_lock);
1335 	if (rushjob < syncdelay / 2) {
1336 		rushjob += 1;
1337 		stat_rush_requests += 1;
1338 		return (1);
1339 	}
1340 	return(0);
1341 }
1342 
1343 /*
1344  * Associate a p-buffer with a vnode.
1345  *
1346  * Also sets B_PAGING flag to indicate that vnode is not fully associated
1347  * with the buffer.  i.e. the bp has not been linked into the vnode or
1348  * ref-counted.
1349  */
1350 void
1351 pbgetvp(vp, bp)
1352 	register struct vnode *vp;
1353 	register struct buf *bp;
1354 {
1355 
1356 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1357 
1358 	bp->b_vp = vp;
1359 	bp->b_flags |= B_PAGING;
1360 	bp->b_dev = vn_todev(vp);
1361 }
1362 
1363 /*
1364  * Disassociate a p-buffer from a vnode.
1365  */
1366 void
1367 pbrelvp(bp)
1368 	register struct buf *bp;
1369 {
1370 
1371 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1372 
1373 	/* XXX REMOVE ME */
1374 	if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1375 		panic(
1376 		    "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1377 		    bp,
1378 		    (int)bp->b_flags
1379 		);
1380 	}
1381 	bp->b_vp = (struct vnode *) 0;
1382 	bp->b_flags &= ~B_PAGING;
1383 }
1384 
1385 /*
1386  * Change the vnode a pager buffer is associated with.
1387  */
1388 void
1389 pbreassignbuf(bp, newvp)
1390 	struct buf *bp;
1391 	struct vnode *newvp;
1392 {
1393 
1394 	KASSERT(bp->b_flags & B_PAGING,
1395 	    ("pbreassignbuf() on non phys bp %p", bp));
1396 	bp->b_vp = newvp;
1397 }
1398 
1399 /*
1400  * Reassign a buffer from one vnode to another.
1401  * Used to assign file specific control information
1402  * (indirect blocks) to the vnode to which they belong.
1403  */
1404 void
1405 reassignbuf(bp, newvp)
1406 	register struct buf *bp;
1407 	register struct vnode *newvp;
1408 {
1409 	struct buflists *listheadp;
1410 	int delay;
1411 	int s;
1412 
1413 	if (newvp == NULL) {
1414 		printf("reassignbuf: NULL");
1415 		return;
1416 	}
1417 	++reassignbufcalls;
1418 
1419 	/*
1420 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1421 	 * is not fully linked in.
1422 	 */
1423 	if (bp->b_flags & B_PAGING)
1424 		panic("cannot reassign paging buffer");
1425 
1426 	s = splbio();
1427 	/*
1428 	 * Delete from old vnode list, if on one.
1429 	 */
1430 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1431 		if (bp->b_xflags & BX_VNDIRTY)
1432 			listheadp = &bp->b_vp->v_dirtyblkhd;
1433 		else
1434 			listheadp = &bp->b_vp->v_cleanblkhd;
1435 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1436 		bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1437 		if (bp->b_vp != newvp) {
1438 			vdrop(bp->b_vp);
1439 			bp->b_vp = NULL;	/* for clarification */
1440 		}
1441 	}
1442 	/*
1443 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1444 	 * of clean buffers.
1445 	 */
1446 	if (bp->b_flags & B_DELWRI) {
1447 		struct buf *tbp;
1448 
1449 		listheadp = &newvp->v_dirtyblkhd;
1450 		if ((newvp->v_flag & VONWORKLST) == 0) {
1451 			switch (newvp->v_type) {
1452 			case VDIR:
1453 				delay = dirdelay;
1454 				break;
1455 			case VCHR:
1456 				if (newvp->v_rdev->si_mountpoint != NULL) {
1457 					delay = metadelay;
1458 					break;
1459 				}
1460 				/* fall through */
1461 			default:
1462 				delay = filedelay;
1463 			}
1464 			vn_syncer_add_to_worklist(newvp, delay);
1465 		}
1466 		bp->b_xflags |= BX_VNDIRTY;
1467 		tbp = TAILQ_FIRST(listheadp);
1468 		if (tbp == NULL ||
1469 		    bp->b_lblkno == 0 ||
1470 		    (bp->b_lblkno > 0 && tbp->b_lblkno < 0) ||
1471 		    (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) {
1472 			TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1473 			++reassignbufsortgood;
1474 		} else if (bp->b_lblkno < 0) {
1475 			TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1476 			++reassignbufsortgood;
1477 		} else if (reassignbufmethod == 1) {
1478 			/*
1479 			 * New sorting algorithm, only handle sequential case,
1480 			 * otherwise append to end (but before metadata)
1481 			 */
1482 			if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL &&
1483 			    (tbp->b_xflags & BX_VNDIRTY)) {
1484 				/*
1485 				 * Found the best place to insert the buffer
1486 				 */
1487 				TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1488 				++reassignbufsortgood;
1489 			} else {
1490 				/*
1491 				 * Missed, append to end, but before meta-data.
1492 				 * We know that the head buffer in the list is
1493 				 * not meta-data due to prior conditionals.
1494 				 *
1495 				 * Indirect effects:  NFS second stage write
1496 				 * tends to wind up here, giving maximum
1497 				 * distance between the unstable write and the
1498 				 * commit rpc.
1499 				 */
1500 				tbp = TAILQ_LAST(listheadp, buflists);
1501 				while (tbp && tbp->b_lblkno < 0)
1502 					tbp = TAILQ_PREV(tbp, buflists, b_vnbufs);
1503 				TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1504 				++reassignbufsortbad;
1505 			}
1506 		} else {
1507 			/*
1508 			 * Old sorting algorithm, scan queue and insert
1509 			 */
1510 			struct buf *ttbp;
1511 			while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1512 			    (ttbp->b_lblkno < bp->b_lblkno)) {
1513 				++reassignbufloops;
1514 				tbp = ttbp;
1515 			}
1516 			TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1517 		}
1518 	} else {
1519 		bp->b_xflags |= BX_VNCLEAN;
1520 		TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1521 		if ((newvp->v_flag & VONWORKLST) &&
1522 		    TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1523 			newvp->v_flag &= ~VONWORKLST;
1524 			LIST_REMOVE(newvp, v_synclist);
1525 		}
1526 	}
1527 	if (bp->b_vp != newvp) {
1528 		bp->b_vp = newvp;
1529 		vhold(bp->b_vp);
1530 	}
1531 	splx(s);
1532 }
1533 
1534 /*
1535  * Create a vnode for a device.
1536  * Used for mounting the root file system.
1537  */
1538 int
1539 bdevvp(dev, vpp)
1540 	dev_t dev;
1541 	struct vnode **vpp;
1542 {
1543 	register struct vnode *vp;
1544 	struct vnode *nvp;
1545 	int error;
1546 
1547 	if (dev == NODEV) {
1548 		*vpp = NULLVP;
1549 		return (ENXIO);
1550 	}
1551 	if (vfinddev(dev, VCHR, vpp))
1552 		return (0);
1553 	error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1554 	if (error) {
1555 		*vpp = NULLVP;
1556 		return (error);
1557 	}
1558 	vp = nvp;
1559 	vp->v_type = VCHR;
1560 	addalias(vp, dev);
1561 	*vpp = vp;
1562 	return (0);
1563 }
1564 
1565 /*
1566  * Add vnode to the alias list hung off the dev_t.
1567  *
1568  * The reason for this gunk is that multiple vnodes can reference
1569  * the same physical device, so checking vp->v_usecount to see
1570  * how many users there are is inadequate; the v_usecount for
1571  * the vnodes need to be accumulated.  vcount() does that.
1572  */
1573 struct vnode *
1574 addaliasu(nvp, nvp_rdev)
1575 	struct vnode *nvp;
1576 	udev_t nvp_rdev;
1577 {
1578 	struct vnode *ovp;
1579 	vop_t **ops;
1580 	dev_t dev;
1581 
1582 	if (nvp->v_type == VBLK)
1583 		return (nvp);
1584 	if (nvp->v_type != VCHR)
1585 		panic("addaliasu on non-special vnode");
1586 	dev = udev2dev(nvp_rdev, 0);
1587 	/*
1588 	 * Check to see if we have a bdevvp vnode with no associated
1589 	 * filesystem. If so, we want to associate the filesystem of
1590 	 * the new newly instigated vnode with the bdevvp vnode and
1591 	 * discard the newly created vnode rather than leaving the
1592 	 * bdevvp vnode lying around with no associated filesystem.
1593 	 */
1594 	if (vfinddev(dev, nvp->v_type, &ovp) == 0 || ovp->v_data != NULL) {
1595 		addalias(nvp, dev);
1596 		return (nvp);
1597 	}
1598 	/*
1599 	 * Discard unneeded vnode, but save its node specific data.
1600 	 * Note that if there is a lock, it is carried over in the
1601 	 * node specific data to the replacement vnode.
1602 	 */
1603 	vref(ovp);
1604 	ovp->v_data = nvp->v_data;
1605 	ovp->v_tag = nvp->v_tag;
1606 	nvp->v_data = NULL;
1607 	lockinit(&ovp->v_lock, PVFS, nvp->v_lock.lk_wmesg,
1608 	    nvp->v_lock.lk_timo, nvp->v_lock.lk_flags & LK_EXTFLG_MASK);
1609 	if (nvp->v_vnlock)
1610 		ovp->v_vnlock = &ovp->v_lock;
1611 	ops = ovp->v_op;
1612 	ovp->v_op = nvp->v_op;
1613 	if (VOP_ISLOCKED(nvp, curthread)) {
1614 		VOP_UNLOCK(nvp, 0, curthread);
1615 		vn_lock(ovp, LK_EXCLUSIVE | LK_RETRY, curthread);
1616 	}
1617 	nvp->v_op = ops;
1618 	insmntque(ovp, nvp->v_mount);
1619 	vrele(nvp);
1620 	vgone(nvp);
1621 	return (ovp);
1622 }
1623 
1624 /* This is a local helper function that do the same as addaliasu, but for a
1625  * dev_t instead of an udev_t. */
1626 static void
1627 addalias(nvp, dev)
1628 	struct vnode *nvp;
1629 	dev_t dev;
1630 {
1631 
1632 	KASSERT(nvp->v_type == VCHR, ("addalias on non-special vnode"));
1633 	nvp->v_rdev = dev;
1634 	mtx_lock(&spechash_mtx);
1635 	SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
1636 	mtx_unlock(&spechash_mtx);
1637 }
1638 
1639 /*
1640  * Grab a particular vnode from the free list, increment its
1641  * reference count and lock it. The vnode lock bit is set if the
1642  * vnode is being eliminated in vgone. The process is awakened
1643  * when the transition is completed, and an error returned to
1644  * indicate that the vnode is no longer usable (possibly having
1645  * been changed to a new file system type).
1646  */
1647 int
1648 vget(vp, flags, td)
1649 	register struct vnode *vp;
1650 	int flags;
1651 	struct thread *td;
1652 {
1653 	int error;
1654 
1655 	/*
1656 	 * If the vnode is in the process of being cleaned out for
1657 	 * another use, we wait for the cleaning to finish and then
1658 	 * return failure. Cleaning is determined by checking that
1659 	 * the VXLOCK flag is set.
1660 	 */
1661 	if ((flags & LK_INTERLOCK) == 0)
1662 		mtx_lock(&vp->v_interlock);
1663 	if (vp->v_flag & VXLOCK) {
1664 		if (vp->v_vxproc == curthread) {
1665 #if 0
1666 			/* this can now occur in normal operation */
1667 			log(LOG_INFO, "VXLOCK interlock avoided\n");
1668 #endif
1669 		} else {
1670 			vp->v_flag |= VXWANT;
1671 			msleep((caddr_t)vp, &vp->v_interlock, PINOD | PDROP,
1672 			    "vget", 0);
1673 			return (ENOENT);
1674 		}
1675 	}
1676 
1677 	vp->v_usecount++;
1678 
1679 	if (VSHOULDBUSY(vp))
1680 		vbusy(vp);
1681 	if (flags & LK_TYPE_MASK) {
1682 		if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
1683 			/*
1684 			 * must expand vrele here because we do not want
1685 			 * to call VOP_INACTIVE if the reference count
1686 			 * drops back to zero since it was never really
1687 			 * active. We must remove it from the free list
1688 			 * before sleeping so that multiple processes do
1689 			 * not try to recycle it.
1690 			 */
1691 			mtx_lock(&vp->v_interlock);
1692 			vp->v_usecount--;
1693 			if (VSHOULDFREE(vp))
1694 				vfree(vp);
1695 			else
1696 				vlruvp(vp);
1697 			mtx_unlock(&vp->v_interlock);
1698 		}
1699 		return (error);
1700 	}
1701 	mtx_unlock(&vp->v_interlock);
1702 	return (0);
1703 }
1704 
1705 /*
1706  * Increase the reference count of a vnode.
1707  */
1708 void
1709 vref(struct vnode *vp)
1710 {
1711 	mtx_lock(&vp->v_interlock);
1712 	vp->v_usecount++;
1713 	mtx_unlock(&vp->v_interlock);
1714 }
1715 
1716 /*
1717  * Vnode put/release.
1718  * If count drops to zero, call inactive routine and return to freelist.
1719  */
1720 void
1721 vrele(vp)
1722 	struct vnode *vp;
1723 {
1724 	struct thread *td = curthread;	/* XXX */
1725 
1726 	KASSERT(vp != NULL, ("vrele: null vp"));
1727 
1728 	mtx_lock(&vp->v_interlock);
1729 
1730 	/* Skip this v_writecount check if we're going to panic below. */
1731 	KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
1732 	    ("vrele: missed vn_close"));
1733 
1734 	if (vp->v_usecount > 1) {
1735 
1736 		vp->v_usecount--;
1737 		mtx_unlock(&vp->v_interlock);
1738 
1739 		return;
1740 	}
1741 
1742 	if (vp->v_usecount == 1) {
1743 		vp->v_usecount--;
1744 		/*
1745 		 * We must call VOP_INACTIVE with the node locked.
1746 		 * If we are doing a vput, the node is already locked,
1747 		 * but, in the case of vrele, we must explicitly lock
1748 		 * the vnode before calling VOP_INACTIVE.
1749 		 */
1750 		if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0)
1751 			VOP_INACTIVE(vp, td);
1752 		if (VSHOULDFREE(vp))
1753 			vfree(vp);
1754 		else
1755 			vlruvp(vp);
1756 
1757 	} else {
1758 #ifdef DIAGNOSTIC
1759 		vprint("vrele: negative ref count", vp);
1760 		mtx_unlock(&vp->v_interlock);
1761 #endif
1762 		panic("vrele: negative ref cnt");
1763 	}
1764 }
1765 
1766 /*
1767  * Release an already locked vnode.  This give the same effects as
1768  * unlock+vrele(), but takes less time and avoids releasing and
1769  * re-aquiring the lock (as vrele() aquires the lock internally.)
1770  */
1771 void
1772 vput(vp)
1773 	struct vnode *vp;
1774 {
1775 	struct thread *td = curthread;	/* XXX */
1776 
1777 	GIANT_REQUIRED;
1778 
1779 	KASSERT(vp != NULL, ("vput: null vp"));
1780 	mtx_lock(&vp->v_interlock);
1781 	/* Skip this v_writecount check if we're going to panic below. */
1782 	KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
1783 	    ("vput: missed vn_close"));
1784 
1785 	if (vp->v_usecount > 1) {
1786 		vp->v_usecount--;
1787 		VOP_UNLOCK(vp, LK_INTERLOCK, td);
1788 		return;
1789 	}
1790 
1791 	if (vp->v_usecount == 1) {
1792 		vp->v_usecount--;
1793 		/*
1794 		 * We must call VOP_INACTIVE with the node locked.
1795 		 * If we are doing a vput, the node is already locked,
1796 		 * so we just need to release the vnode mutex.
1797 		 */
1798 		mtx_unlock(&vp->v_interlock);
1799 		VOP_INACTIVE(vp, td);
1800 		if (VSHOULDFREE(vp))
1801 			vfree(vp);
1802 		else
1803 			vlruvp(vp);
1804 
1805 	} else {
1806 #ifdef DIAGNOSTIC
1807 		vprint("vput: negative ref count", vp);
1808 #endif
1809 		panic("vput: negative ref cnt");
1810 	}
1811 }
1812 
1813 /*
1814  * Somebody doesn't want the vnode recycled.
1815  */
1816 void
1817 vhold(vp)
1818 	register struct vnode *vp;
1819 {
1820 	int s;
1821 
1822   	s = splbio();
1823 	vp->v_holdcnt++;
1824 	if (VSHOULDBUSY(vp))
1825 		vbusy(vp);
1826 	splx(s);
1827 }
1828 
1829 /*
1830  * Note that there is one less who cares about this vnode.  vdrop() is the
1831  * opposite of vhold().
1832  */
1833 void
1834 vdrop(vp)
1835 	register struct vnode *vp;
1836 {
1837 	int s;
1838 
1839 	s = splbio();
1840 	if (vp->v_holdcnt <= 0)
1841 		panic("vdrop: holdcnt");
1842 	vp->v_holdcnt--;
1843 	if (VSHOULDFREE(vp))
1844 		vfree(vp);
1845 	else
1846 		vlruvp(vp);
1847 	splx(s);
1848 }
1849 
1850 /*
1851  * Remove any vnodes in the vnode table belonging to mount point mp.
1852  *
1853  * If FORCECLOSE is not specified, there should not be any active ones,
1854  * return error if any are found (nb: this is a user error, not a
1855  * system error). If FORCECLOSE is specified, detach any active vnodes
1856  * that are found.
1857  *
1858  * If WRITECLOSE is set, only flush out regular file vnodes open for
1859  * writing.
1860  *
1861  * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
1862  *
1863  * `rootrefs' specifies the base reference count for the root vnode
1864  * of this filesystem. The root vnode is considered busy if its
1865  * v_usecount exceeds this value. On a successful return, vflush()
1866  * will call vrele() on the root vnode exactly rootrefs times.
1867  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
1868  * be zero.
1869  */
1870 #ifdef DIAGNOSTIC
1871 static int busyprt = 0;		/* print out busy vnodes */
1872 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1873 #endif
1874 
1875 int
1876 vflush(mp, rootrefs, flags)
1877 	struct mount *mp;
1878 	int rootrefs;
1879 	int flags;
1880 {
1881 	struct thread *td = curthread;	/* XXX */
1882 	struct vnode *vp, *nvp, *rootvp = NULL;
1883 	struct vattr vattr;
1884 	int busy = 0, error;
1885 
1886 	if (rootrefs > 0) {
1887 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
1888 		    ("vflush: bad args"));
1889 		/*
1890 		 * Get the filesystem root vnode. We can vput() it
1891 		 * immediately, since with rootrefs > 0, it won't go away.
1892 		 */
1893 		if ((error = VFS_ROOT(mp, &rootvp)) != 0)
1894 			return (error);
1895 		vput(rootvp);
1896 	}
1897 	mtx_lock(&mntvnode_mtx);
1898 loop:
1899 	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) {
1900 		/*
1901 		 * Make sure this vnode wasn't reclaimed in getnewvnode().
1902 		 * Start over if it has (it won't be on the list anymore).
1903 		 */
1904 		if (vp->v_mount != mp)
1905 			goto loop;
1906 		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
1907 
1908 		mtx_unlock(&mntvnode_mtx);
1909 		mtx_lock(&vp->v_interlock);
1910 		/*
1911 		 * Skip over a vnodes marked VSYSTEM.
1912 		 */
1913 		if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1914 			mtx_unlock(&vp->v_interlock);
1915 			mtx_lock(&mntvnode_mtx);
1916 			continue;
1917 		}
1918 		/*
1919 		 * If WRITECLOSE is set, flush out unlinked but still open
1920 		 * files (even if open only for reading) and regular file
1921 		 * vnodes open for writing.
1922 		 */
1923 		if ((flags & WRITECLOSE) &&
1924 		    (vp->v_type == VNON ||
1925 		    (VOP_GETATTR(vp, &vattr, td->td_ucred, td) == 0 &&
1926 		    vattr.va_nlink > 0)) &&
1927 		    (vp->v_writecount == 0 || vp->v_type != VREG)) {
1928 			mtx_unlock(&vp->v_interlock);
1929 			mtx_lock(&mntvnode_mtx);
1930 			continue;
1931 		}
1932 
1933 		/*
1934 		 * With v_usecount == 0, all we need to do is clear out the
1935 		 * vnode data structures and we are done.
1936 		 */
1937 		if (vp->v_usecount == 0) {
1938 			vgonel(vp, td);
1939 			mtx_lock(&mntvnode_mtx);
1940 			continue;
1941 		}
1942 
1943 		/*
1944 		 * If FORCECLOSE is set, forcibly close the vnode. For block
1945 		 * or character devices, revert to an anonymous device. For
1946 		 * all other files, just kill them.
1947 		 */
1948 		if (flags & FORCECLOSE) {
1949 			if (vp->v_type != VCHR) {
1950 				vgonel(vp, td);
1951 			} else {
1952 				vclean(vp, 0, td);
1953 				vp->v_op = spec_vnodeop_p;
1954 				insmntque(vp, (struct mount *) 0);
1955 			}
1956 			mtx_lock(&mntvnode_mtx);
1957 			continue;
1958 		}
1959 #ifdef DIAGNOSTIC
1960 		if (busyprt)
1961 			vprint("vflush: busy vnode", vp);
1962 #endif
1963 		mtx_unlock(&vp->v_interlock);
1964 		mtx_lock(&mntvnode_mtx);
1965 		busy++;
1966 	}
1967 	mtx_unlock(&mntvnode_mtx);
1968 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
1969 		/*
1970 		 * If just the root vnode is busy, and if its refcount
1971 		 * is equal to `rootrefs', then go ahead and kill it.
1972 		 */
1973 		mtx_lock(&rootvp->v_interlock);
1974 		KASSERT(busy > 0, ("vflush: not busy"));
1975 		KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
1976 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
1977 			vgonel(rootvp, td);
1978 			busy = 0;
1979 		} else
1980 			mtx_unlock(&rootvp->v_interlock);
1981 	}
1982 	if (busy)
1983 		return (EBUSY);
1984 	for (; rootrefs > 0; rootrefs--)
1985 		vrele(rootvp);
1986 	return (0);
1987 }
1988 
1989 /*
1990  * This moves a now (likely recyclable) vnode to the end of the
1991  * mountlist.  XXX However, it is temporarily disabled until we
1992  * can clean up ffs_sync() and friends, which have loop restart
1993  * conditions which this code causes to operate O(N^2).
1994  */
1995 static void
1996 vlruvp(struct vnode *vp)
1997 {
1998 #if 0
1999 	struct mount *mp;
2000 
2001 	if ((mp = vp->v_mount) != NULL) {
2002 		mtx_lock(&mntvnode_mtx);
2003 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
2004 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
2005 		mtx_unlock(&mntvnode_mtx);
2006 	}
2007 #endif
2008 }
2009 
2010 /*
2011  * Disassociate the underlying file system from a vnode.
2012  */
2013 static void
2014 vclean(vp, flags, td)
2015 	struct vnode *vp;
2016 	int flags;
2017 	struct thread *td;
2018 {
2019 	int active;
2020 
2021 	/*
2022 	 * Check to see if the vnode is in use. If so we have to reference it
2023 	 * before we clean it out so that its count cannot fall to zero and
2024 	 * generate a race against ourselves to recycle it.
2025 	 */
2026 	if ((active = vp->v_usecount))
2027 		vp->v_usecount++;
2028 
2029 	/*
2030 	 * Prevent the vnode from being recycled or brought into use while we
2031 	 * clean it out.
2032 	 */
2033 	if (vp->v_flag & VXLOCK)
2034 		panic("vclean: deadlock");
2035 	vp->v_flag |= VXLOCK;
2036 	vp->v_vxproc = curthread;
2037 	/*
2038 	 * Even if the count is zero, the VOP_INACTIVE routine may still
2039 	 * have the object locked while it cleans it out. The VOP_LOCK
2040 	 * ensures that the VOP_INACTIVE routine is done with its work.
2041 	 * For active vnodes, it ensures that no other activity can
2042 	 * occur while the underlying object is being cleaned out.
2043 	 */
2044 	VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td);
2045 
2046 	/*
2047 	 * Clean out any buffers associated with the vnode.
2048 	 * If the flush fails, just toss the buffers.
2049 	 */
2050 	if (flags & DOCLOSE) {
2051 		if (TAILQ_FIRST(&vp->v_dirtyblkhd) != NULL)
2052 			(void) vn_write_suspend_wait(vp, NULL, V_WAIT);
2053 		if (vinvalbuf(vp, V_SAVE, NOCRED, td, 0, 0) != 0)
2054 			vinvalbuf(vp, 0, NOCRED, td, 0, 0);
2055 	}
2056 
2057 	VOP_DESTROYVOBJECT(vp);
2058 
2059 	/*
2060 	 * If purging an active vnode, it must be closed and
2061 	 * deactivated before being reclaimed. Note that the
2062 	 * VOP_INACTIVE will unlock the vnode.
2063 	 */
2064 	if (active) {
2065 		if (flags & DOCLOSE)
2066 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2067 		VOP_INACTIVE(vp, td);
2068 	} else {
2069 		/*
2070 		 * Any other processes trying to obtain this lock must first
2071 		 * wait for VXLOCK to clear, then call the new lock operation.
2072 		 */
2073 		VOP_UNLOCK(vp, 0, td);
2074 	}
2075 	/*
2076 	 * Reclaim the vnode.
2077 	 */
2078 	if (VOP_RECLAIM(vp, td))
2079 		panic("vclean: cannot reclaim");
2080 
2081 	if (active) {
2082 		/*
2083 		 * Inline copy of vrele() since VOP_INACTIVE
2084 		 * has already been called.
2085 		 */
2086 		mtx_lock(&vp->v_interlock);
2087 		if (--vp->v_usecount <= 0) {
2088 #ifdef DIAGNOSTIC
2089 			if (vp->v_usecount < 0 || vp->v_writecount != 0) {
2090 				vprint("vclean: bad ref count", vp);
2091 				panic("vclean: ref cnt");
2092 			}
2093 #endif
2094 			vfree(vp);
2095 		}
2096 		mtx_unlock(&vp->v_interlock);
2097 	}
2098 
2099 	cache_purge(vp);
2100 	vp->v_vnlock = NULL;
2101 	lockdestroy(&vp->v_lock);
2102 
2103 	if (VSHOULDFREE(vp))
2104 		vfree(vp);
2105 
2106 	/*
2107 	 * Done with purge, notify sleepers of the grim news.
2108 	 */
2109 	vp->v_op = dead_vnodeop_p;
2110 	if (vp->v_pollinfo != NULL)
2111 		vn_pollgone(vp);
2112 	vp->v_tag = VT_NON;
2113 	vp->v_flag &= ~VXLOCK;
2114 	vp->v_vxproc = NULL;
2115 	if (vp->v_flag & VXWANT) {
2116 		vp->v_flag &= ~VXWANT;
2117 		wakeup((caddr_t) vp);
2118 	}
2119 }
2120 
2121 /*
2122  * Eliminate all activity associated with the requested vnode
2123  * and with all vnodes aliased to the requested vnode.
2124  */
2125 int
2126 vop_revoke(ap)
2127 	struct vop_revoke_args /* {
2128 		struct vnode *a_vp;
2129 		int a_flags;
2130 	} */ *ap;
2131 {
2132 	struct vnode *vp, *vq;
2133 	dev_t dev;
2134 
2135 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
2136 
2137 	vp = ap->a_vp;
2138 	/*
2139 	 * If a vgone (or vclean) is already in progress,
2140 	 * wait until it is done and return.
2141 	 */
2142 	if (vp->v_flag & VXLOCK) {
2143 		vp->v_flag |= VXWANT;
2144 		msleep((caddr_t)vp, &vp->v_interlock, PINOD | PDROP,
2145 		    "vop_revokeall", 0);
2146 		return (0);
2147 	}
2148 	dev = vp->v_rdev;
2149 	for (;;) {
2150 		mtx_lock(&spechash_mtx);
2151 		vq = SLIST_FIRST(&dev->si_hlist);
2152 		mtx_unlock(&spechash_mtx);
2153 		if (!vq)
2154 			break;
2155 		vgone(vq);
2156 	}
2157 	return (0);
2158 }
2159 
2160 /*
2161  * Recycle an unused vnode to the front of the free list.
2162  * Release the passed interlock if the vnode will be recycled.
2163  */
2164 int
2165 vrecycle(vp, inter_lkp, td)
2166 	struct vnode *vp;
2167 	struct mtx *inter_lkp;
2168 	struct thread *td;
2169 {
2170 
2171 	mtx_lock(&vp->v_interlock);
2172 	if (vp->v_usecount == 0) {
2173 		if (inter_lkp) {
2174 			mtx_unlock(inter_lkp);
2175 		}
2176 		vgonel(vp, td);
2177 		return (1);
2178 	}
2179 	mtx_unlock(&vp->v_interlock);
2180 	return (0);
2181 }
2182 
2183 /*
2184  * Eliminate all activity associated with a vnode
2185  * in preparation for reuse.
2186  */
2187 void
2188 vgone(vp)
2189 	register struct vnode *vp;
2190 {
2191 	struct thread *td = curthread;	/* XXX */
2192 
2193 	mtx_lock(&vp->v_interlock);
2194 	vgonel(vp, td);
2195 }
2196 
2197 /*
2198  * vgone, with the vp interlock held.
2199  */
2200 void
2201 vgonel(vp, td)
2202 	struct vnode *vp;
2203 	struct thread *td;
2204 {
2205 	int s;
2206 
2207 	/*
2208 	 * If a vgone (or vclean) is already in progress,
2209 	 * wait until it is done and return.
2210 	 */
2211 	if (vp->v_flag & VXLOCK) {
2212 		vp->v_flag |= VXWANT;
2213 		msleep((caddr_t)vp, &vp->v_interlock, PINOD | PDROP,
2214 		    "vgone", 0);
2215 		return;
2216 	}
2217 
2218 	/*
2219 	 * Clean out the filesystem specific data.
2220 	 */
2221 	vclean(vp, DOCLOSE, td);
2222 	mtx_lock(&vp->v_interlock);
2223 
2224 	/*
2225 	 * Delete from old mount point vnode list, if on one.
2226 	 */
2227 	if (vp->v_mount != NULL)
2228 		insmntque(vp, (struct mount *)0);
2229 	/*
2230 	 * If special device, remove it from special device alias list
2231 	 * if it is on one.
2232 	 */
2233 	if (vp->v_type == VCHR && vp->v_rdev != NULL && vp->v_rdev != NODEV) {
2234 		mtx_lock(&spechash_mtx);
2235 		SLIST_REMOVE(&vp->v_rdev->si_hlist, vp, vnode, v_specnext);
2236 		freedev(vp->v_rdev);
2237 		mtx_unlock(&spechash_mtx);
2238 		vp->v_rdev = NULL;
2239 	}
2240 
2241 	/*
2242 	 * If it is on the freelist and not already at the head,
2243 	 * move it to the head of the list. The test of the
2244 	 * VDOOMED flag and the reference count of zero is because
2245 	 * it will be removed from the free list by getnewvnode,
2246 	 * but will not have its reference count incremented until
2247 	 * after calling vgone. If the reference count were
2248 	 * incremented first, vgone would (incorrectly) try to
2249 	 * close the previous instance of the underlying object.
2250 	 */
2251 	if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
2252 		s = splbio();
2253 		mtx_lock(&vnode_free_list_mtx);
2254 		if (vp->v_flag & VFREE)
2255 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2256 		else
2257 			freevnodes++;
2258 		vp->v_flag |= VFREE;
2259 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2260 		mtx_unlock(&vnode_free_list_mtx);
2261 		splx(s);
2262 	}
2263 
2264 	vp->v_type = VBAD;
2265 	mtx_unlock(&vp->v_interlock);
2266 }
2267 
2268 /*
2269  * Lookup a vnode by device number.
2270  */
2271 int
2272 vfinddev(dev, type, vpp)
2273 	dev_t dev;
2274 	enum vtype type;
2275 	struct vnode **vpp;
2276 {
2277 	struct vnode *vp;
2278 
2279 	mtx_lock(&spechash_mtx);
2280 	SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2281 		if (type == vp->v_type) {
2282 			*vpp = vp;
2283 			mtx_unlock(&spechash_mtx);
2284 			return (1);
2285 		}
2286 	}
2287 	mtx_unlock(&spechash_mtx);
2288 	return (0);
2289 }
2290 
2291 /*
2292  * Calculate the total number of references to a special device.
2293  */
2294 int
2295 vcount(vp)
2296 	struct vnode *vp;
2297 {
2298 	struct vnode *vq;
2299 	int count;
2300 
2301 	count = 0;
2302 	mtx_lock(&spechash_mtx);
2303 	SLIST_FOREACH(vq, &vp->v_rdev->si_hlist, v_specnext)
2304 		count += vq->v_usecount;
2305 	mtx_unlock(&spechash_mtx);
2306 	return (count);
2307 }
2308 
2309 /*
2310  * Same as above, but using the dev_t as argument
2311  */
2312 int
2313 count_dev(dev)
2314 	dev_t dev;
2315 {
2316 	struct vnode *vp;
2317 
2318 	vp = SLIST_FIRST(&dev->si_hlist);
2319 	if (vp == NULL)
2320 		return (0);
2321 	return(vcount(vp));
2322 }
2323 
2324 /*
2325  * Print out a description of a vnode.
2326  */
2327 static char *typename[] =
2328 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2329 
2330 void
2331 vprint(label, vp)
2332 	char *label;
2333 	struct vnode *vp;
2334 {
2335 	char buf[96];
2336 
2337 	if (label != NULL)
2338 		printf("%s: %p: ", label, (void *)vp);
2339 	else
2340 		printf("%p: ", (void *)vp);
2341 	printf("type %s, usecount %d, writecount %d, refcount %d,",
2342 	    typename[vp->v_type], vp->v_usecount, vp->v_writecount,
2343 	    vp->v_holdcnt);
2344 	buf[0] = '\0';
2345 	if (vp->v_flag & VROOT)
2346 		strcat(buf, "|VROOT");
2347 	if (vp->v_flag & VTEXT)
2348 		strcat(buf, "|VTEXT");
2349 	if (vp->v_flag & VSYSTEM)
2350 		strcat(buf, "|VSYSTEM");
2351 	if (vp->v_flag & VXLOCK)
2352 		strcat(buf, "|VXLOCK");
2353 	if (vp->v_flag & VXWANT)
2354 		strcat(buf, "|VXWANT");
2355 	if (vp->v_flag & VBWAIT)
2356 		strcat(buf, "|VBWAIT");
2357 	if (vp->v_flag & VDOOMED)
2358 		strcat(buf, "|VDOOMED");
2359 	if (vp->v_flag & VFREE)
2360 		strcat(buf, "|VFREE");
2361 	if (vp->v_flag & VOBJBUF)
2362 		strcat(buf, "|VOBJBUF");
2363 	if (buf[0] != '\0')
2364 		printf(" flags (%s)", &buf[1]);
2365 	if (vp->v_data == NULL) {
2366 		printf("\n");
2367 	} else {
2368 		printf("\n\t");
2369 		VOP_PRINT(vp);
2370 	}
2371 }
2372 
2373 #ifdef DDB
2374 #include <ddb/ddb.h>
2375 /*
2376  * List all of the locked vnodes in the system.
2377  * Called when debugging the kernel.
2378  */
2379 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
2380 {
2381 	struct thread *td = curthread;	/* XXX */
2382 	struct mount *mp, *nmp;
2383 	struct vnode *vp;
2384 
2385 	printf("Locked vnodes\n");
2386 	mtx_lock(&mountlist_mtx);
2387 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2388 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
2389 			nmp = TAILQ_NEXT(mp, mnt_list);
2390 			continue;
2391 		}
2392 		mtx_lock(&mntvnode_mtx);
2393 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2394 			if (VOP_ISLOCKED(vp, NULL))
2395 				vprint((char *)0, vp);
2396 		}
2397 		mtx_unlock(&mntvnode_mtx);
2398 		mtx_lock(&mountlist_mtx);
2399 		nmp = TAILQ_NEXT(mp, mnt_list);
2400 		vfs_unbusy(mp, td);
2401 	}
2402 	mtx_unlock(&mountlist_mtx);
2403 }
2404 #endif
2405 
2406 /*
2407  * Top level filesystem related information gathering.
2408  */
2409 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2410 
2411 static int
2412 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2413 {
2414 	int *name = (int *)arg1 - 1;	/* XXX */
2415 	u_int namelen = arg2 + 1;	/* XXX */
2416 	struct vfsconf *vfsp;
2417 
2418 #if 1 || defined(COMPAT_PRELITE2)
2419 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2420 	if (namelen == 1)
2421 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2422 #endif
2423 
2424 	/* XXX the below code does not compile; vfs_sysctl does not exist. */
2425 #ifdef notyet
2426 	/* all sysctl names at this level are at least name and field */
2427 	if (namelen < 2)
2428 		return (ENOTDIR);		/* overloaded */
2429 	if (name[0] != VFS_GENERIC) {
2430 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2431 			if (vfsp->vfc_typenum == name[0])
2432 				break;
2433 		if (vfsp == NULL)
2434 			return (EOPNOTSUPP);
2435 		return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2436 		    oldp, oldlenp, newp, newlen, td));
2437 	}
2438 #endif
2439 	switch (name[1]) {
2440 	case VFS_MAXTYPENUM:
2441 		if (namelen != 2)
2442 			return (ENOTDIR);
2443 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2444 	case VFS_CONF:
2445 		if (namelen != 3)
2446 			return (ENOTDIR);	/* overloaded */
2447 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2448 			if (vfsp->vfc_typenum == name[2])
2449 				break;
2450 		if (vfsp == NULL)
2451 			return (EOPNOTSUPP);
2452 		return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2453 	}
2454 	return (EOPNOTSUPP);
2455 }
2456 
2457 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2458 	"Generic filesystem");
2459 
2460 #if 1 || defined(COMPAT_PRELITE2)
2461 
2462 static int
2463 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2464 {
2465 	int error;
2466 	struct vfsconf *vfsp;
2467 	struct ovfsconf ovfs;
2468 
2469 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2470 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
2471 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
2472 		ovfs.vfc_index = vfsp->vfc_typenum;
2473 		ovfs.vfc_refcount = vfsp->vfc_refcount;
2474 		ovfs.vfc_flags = vfsp->vfc_flags;
2475 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2476 		if (error)
2477 			return error;
2478 	}
2479 	return 0;
2480 }
2481 
2482 #endif /* 1 || COMPAT_PRELITE2 */
2483 
2484 #if COMPILING_LINT
2485 #define KINFO_VNODESLOP	10
2486 /*
2487  * Dump vnode list (via sysctl).
2488  * Copyout address of vnode followed by vnode.
2489  */
2490 /* ARGSUSED */
2491 static int
2492 sysctl_vnode(SYSCTL_HANDLER_ARGS)
2493 {
2494 	struct thread *td = curthread;	/* XXX */
2495 	struct mount *mp, *nmp;
2496 	struct vnode *nvp, *vp;
2497 	int error;
2498 
2499 #define VPTRSZ	sizeof (struct vnode *)
2500 #define VNODESZ	sizeof (struct vnode)
2501 
2502 	req->lock = 0;
2503 	if (!req->oldptr) /* Make an estimate */
2504 		return (SYSCTL_OUT(req, 0,
2505 			(numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2506 
2507 	mtx_lock(&mountlist_mtx);
2508 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2509 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
2510 			nmp = TAILQ_NEXT(mp, mnt_list);
2511 			continue;
2512 		}
2513 		mtx_lock(&mntvnode_mtx);
2514 again:
2515 		for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2516 		     vp != NULL;
2517 		     vp = nvp) {
2518 			/*
2519 			 * Check that the vp is still associated with
2520 			 * this filesystem.  RACE: could have been
2521 			 * recycled onto the same filesystem.
2522 			 */
2523 			if (vp->v_mount != mp)
2524 				goto again;
2525 			nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2526 			mtx_unlock(&mntvnode_mtx);
2527 			if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2528 			    (error = SYSCTL_OUT(req, vp, VNODESZ)))
2529 				return (error);
2530 			mtx_lock(&mntvnode_mtx);
2531 		}
2532 		mtx_unlock(&mntvnode_mtx);
2533 		mtx_lock(&mountlist_mtx);
2534 		nmp = TAILQ_NEXT(mp, mnt_list);
2535 		vfs_unbusy(mp, td);
2536 	}
2537 	mtx_unlock(&mountlist_mtx);
2538 
2539 	return (0);
2540 }
2541 
2542 /*
2543  * XXX
2544  * Exporting the vnode list on large systems causes them to crash.
2545  * Exporting the vnode list on medium systems causes sysctl to coredump.
2546  */
2547 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2548 	0, 0, sysctl_vnode, "S,vnode", "");
2549 #endif
2550 
2551 /*
2552  * Check to see if a filesystem is mounted on a block device.
2553  */
2554 int
2555 vfs_mountedon(vp)
2556 	struct vnode *vp;
2557 {
2558 
2559 	if (vp->v_rdev->si_mountpoint != NULL)
2560 		return (EBUSY);
2561 	return (0);
2562 }
2563 
2564 /*
2565  * Unmount all filesystems. The list is traversed in reverse order
2566  * of mounting to avoid dependencies.
2567  */
2568 void
2569 vfs_unmountall()
2570 {
2571 	struct mount *mp;
2572 	struct thread *td;
2573 	int error;
2574 
2575 	if (curthread != NULL)
2576 		td = curthread;
2577 	else
2578 		td = FIRST_THREAD_IN_PROC(initproc); /* XXX XXX proc0? */
2579 	/*
2580 	 * Since this only runs when rebooting, it is not interlocked.
2581 	 */
2582 	while(!TAILQ_EMPTY(&mountlist)) {
2583 		mp = TAILQ_LAST(&mountlist, mntlist);
2584 		error = dounmount(mp, MNT_FORCE, td);
2585 		if (error) {
2586 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
2587 			printf("unmount of %s failed (",
2588 			    mp->mnt_stat.f_mntonname);
2589 			if (error == EBUSY)
2590 				printf("BUSY)\n");
2591 			else
2592 				printf("%d)\n", error);
2593 		} else {
2594 			/* The unmount has removed mp from the mountlist */
2595 		}
2596 	}
2597 }
2598 
2599 /*
2600  * perform msync on all vnodes under a mount point
2601  * the mount point must be locked.
2602  */
2603 void
2604 vfs_msync(struct mount *mp, int flags)
2605 {
2606 	struct vnode *vp, *nvp;
2607 	struct vm_object *obj;
2608 	int tries;
2609 
2610 	GIANT_REQUIRED;
2611 
2612 	tries = 5;
2613 	mtx_lock(&mntvnode_mtx);
2614 loop:
2615 	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) {
2616 		if (vp->v_mount != mp) {
2617 			if (--tries > 0)
2618 				goto loop;
2619 			break;
2620 		}
2621 		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2622 
2623 		if (vp->v_flag & VXLOCK)	/* XXX: what if MNT_WAIT? */
2624 			continue;
2625 
2626 		if (vp->v_flag & VNOSYNC)	/* unlinked, skip it */
2627 			continue;
2628 
2629 		if ((vp->v_flag & VOBJDIRTY) &&
2630 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
2631 			mtx_unlock(&mntvnode_mtx);
2632 			if (!vget(vp,
2633 			    LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, curthread)) {
2634 				if (VOP_GETVOBJECT(vp, &obj) == 0) {
2635 					vm_object_page_clean(obj, 0, 0,
2636 					    flags == MNT_WAIT ?
2637 					    OBJPC_SYNC : OBJPC_NOSYNC);
2638 				}
2639 				vput(vp);
2640 			}
2641 			mtx_lock(&mntvnode_mtx);
2642 			if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) {
2643 				if (--tries > 0)
2644 					goto loop;
2645 				break;
2646 			}
2647 		}
2648 	}
2649 	mtx_unlock(&mntvnode_mtx);
2650 }
2651 
2652 /*
2653  * Create the VM object needed for VMIO and mmap support.  This
2654  * is done for all VREG files in the system.  Some filesystems might
2655  * afford the additional metadata buffering capability of the
2656  * VMIO code by making the device node be VMIO mode also.
2657  *
2658  * vp must be locked when vfs_object_create is called.
2659  */
2660 int
2661 vfs_object_create(vp, td, cred)
2662 	struct vnode *vp;
2663 	struct thread *td;
2664 	struct ucred *cred;
2665 {
2666 	GIANT_REQUIRED;
2667 	return (VOP_CREATEVOBJECT(vp, cred, td));
2668 }
2669 
2670 /*
2671  * Mark a vnode as free, putting it up for recycling.
2672  */
2673 void
2674 vfree(vp)
2675 	struct vnode *vp;
2676 {
2677 	int s;
2678 
2679 	s = splbio();
2680 	mtx_lock(&vnode_free_list_mtx);
2681 	KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free"));
2682 	if (vp->v_flag & VAGE) {
2683 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2684 	} else {
2685 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2686 	}
2687 	freevnodes++;
2688 	mtx_unlock(&vnode_free_list_mtx);
2689 	vp->v_flag &= ~VAGE;
2690 	vp->v_flag |= VFREE;
2691 	splx(s);
2692 }
2693 
2694 /*
2695  * Opposite of vfree() - mark a vnode as in use.
2696  */
2697 void
2698 vbusy(vp)
2699 	struct vnode *vp;
2700 {
2701 	int s;
2702 
2703 	s = splbio();
2704 	mtx_lock(&vnode_free_list_mtx);
2705 	KASSERT((vp->v_flag & VFREE) != 0, ("vnode not free"));
2706 	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2707 	freevnodes--;
2708 	mtx_unlock(&vnode_free_list_mtx);
2709 	vp->v_flag &= ~(VFREE|VAGE);
2710 	splx(s);
2711 }
2712 
2713 /*
2714  * Record a process's interest in events which might happen to
2715  * a vnode.  Because poll uses the historic select-style interface
2716  * internally, this routine serves as both the ``check for any
2717  * pending events'' and the ``record my interest in future events''
2718  * functions.  (These are done together, while the lock is held,
2719  * to avoid race conditions.)
2720  */
2721 int
2722 vn_pollrecord(vp, td, events)
2723 	struct vnode *vp;
2724 	struct thread *td;
2725 	short events;
2726 {
2727 
2728 	if (vp->v_pollinfo == NULL)
2729 		v_addpollinfo(vp);
2730 	mtx_lock(&vp->v_pollinfo->vpi_lock);
2731 	if (vp->v_pollinfo->vpi_revents & events) {
2732 		/*
2733 		 * This leaves events we are not interested
2734 		 * in available for the other process which
2735 		 * which presumably had requested them
2736 		 * (otherwise they would never have been
2737 		 * recorded).
2738 		 */
2739 		events &= vp->v_pollinfo->vpi_revents;
2740 		vp->v_pollinfo->vpi_revents &= ~events;
2741 
2742 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
2743 		return events;
2744 	}
2745 	vp->v_pollinfo->vpi_events |= events;
2746 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
2747 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
2748 	return 0;
2749 }
2750 
2751 /*
2752  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
2753  * it is possible for us to miss an event due to race conditions, but
2754  * that condition is expected to be rare, so for the moment it is the
2755  * preferred interface.
2756  */
2757 void
2758 vn_pollevent(vp, events)
2759 	struct vnode *vp;
2760 	short events;
2761 {
2762 
2763 	if (vp->v_pollinfo == NULL)
2764 		v_addpollinfo(vp);
2765 	mtx_lock(&vp->v_pollinfo->vpi_lock);
2766 	if (vp->v_pollinfo->vpi_events & events) {
2767 		/*
2768 		 * We clear vpi_events so that we don't
2769 		 * call selwakeup() twice if two events are
2770 		 * posted before the polling process(es) is
2771 		 * awakened.  This also ensures that we take at
2772 		 * most one selwakeup() if the polling process
2773 		 * is no longer interested.  However, it does
2774 		 * mean that only one event can be noticed at
2775 		 * a time.  (Perhaps we should only clear those
2776 		 * event bits which we note?) XXX
2777 		 */
2778 		vp->v_pollinfo->vpi_events = 0;	/* &= ~events ??? */
2779 		vp->v_pollinfo->vpi_revents |= events;
2780 		selwakeup(&vp->v_pollinfo->vpi_selinfo);
2781 	}
2782 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
2783 }
2784 
2785 /*
2786  * Wake up anyone polling on vp because it is being revoked.
2787  * This depends on dead_poll() returning POLLHUP for correct
2788  * behavior.
2789  */
2790 void
2791 vn_pollgone(vp)
2792 	struct vnode *vp;
2793 {
2794 
2795 	mtx_lock(&vp->v_pollinfo->vpi_lock);
2796         VN_KNOTE(vp, NOTE_REVOKE);
2797 	if (vp->v_pollinfo->vpi_events) {
2798 		vp->v_pollinfo->vpi_events = 0;
2799 		selwakeup(&vp->v_pollinfo->vpi_selinfo);
2800 	}
2801 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
2802 }
2803 
2804 
2805 
2806 /*
2807  * Routine to create and manage a filesystem syncer vnode.
2808  */
2809 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
2810 static int	sync_fsync(struct  vop_fsync_args *);
2811 static int	sync_inactive(struct  vop_inactive_args *);
2812 static int	sync_reclaim(struct  vop_reclaim_args *);
2813 #define sync_lock ((int (*)(struct  vop_lock_args *))vop_nolock)
2814 #define sync_unlock ((int (*)(struct  vop_unlock_args *))vop_nounlock)
2815 static int	sync_print(struct vop_print_args *);
2816 #define sync_islocked ((int(*)(struct vop_islocked_args *))vop_noislocked)
2817 
2818 static vop_t **sync_vnodeop_p;
2819 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
2820 	{ &vop_default_desc,	(vop_t *) vop_eopnotsupp },
2821 	{ &vop_close_desc,	(vop_t *) sync_close },		/* close */
2822 	{ &vop_fsync_desc,	(vop_t *) sync_fsync },		/* fsync */
2823 	{ &vop_inactive_desc,	(vop_t *) sync_inactive },	/* inactive */
2824 	{ &vop_reclaim_desc,	(vop_t *) sync_reclaim },	/* reclaim */
2825 	{ &vop_lock_desc,	(vop_t *) sync_lock },		/* lock */
2826 	{ &vop_unlock_desc,	(vop_t *) sync_unlock },	/* unlock */
2827 	{ &vop_print_desc,	(vop_t *) sync_print },		/* print */
2828 	{ &vop_islocked_desc,	(vop_t *) sync_islocked },	/* islocked */
2829 	{ NULL, NULL }
2830 };
2831 static struct vnodeopv_desc sync_vnodeop_opv_desc =
2832 	{ &sync_vnodeop_p, sync_vnodeop_entries };
2833 
2834 VNODEOP_SET(sync_vnodeop_opv_desc);
2835 
2836 /*
2837  * Create a new filesystem syncer vnode for the specified mount point.
2838  */
2839 int
2840 vfs_allocate_syncvnode(mp)
2841 	struct mount *mp;
2842 {
2843 	struct vnode *vp;
2844 	static long start, incr, next;
2845 	int error;
2846 
2847 	/* Allocate a new vnode */
2848 	if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
2849 		mp->mnt_syncer = NULL;
2850 		return (error);
2851 	}
2852 	vp->v_type = VNON;
2853 	/*
2854 	 * Place the vnode onto the syncer worklist. We attempt to
2855 	 * scatter them about on the list so that they will go off
2856 	 * at evenly distributed times even if all the filesystems
2857 	 * are mounted at once.
2858 	 */
2859 	next += incr;
2860 	if (next == 0 || next > syncer_maxdelay) {
2861 		start /= 2;
2862 		incr /= 2;
2863 		if (start == 0) {
2864 			start = syncer_maxdelay / 2;
2865 			incr = syncer_maxdelay;
2866 		}
2867 		next = start;
2868 	}
2869 	vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
2870 	mp->mnt_syncer = vp;
2871 	return (0);
2872 }
2873 
2874 /*
2875  * Do a lazy sync of the filesystem.
2876  */
2877 static int
2878 sync_fsync(ap)
2879 	struct vop_fsync_args /* {
2880 		struct vnode *a_vp;
2881 		struct ucred *a_cred;
2882 		int a_waitfor;
2883 		struct thread *a_td;
2884 	} */ *ap;
2885 {
2886 	struct vnode *syncvp = ap->a_vp;
2887 	struct mount *mp = syncvp->v_mount;
2888 	struct thread *td = ap->a_td;
2889 	int asyncflag;
2890 
2891 	/*
2892 	 * We only need to do something if this is a lazy evaluation.
2893 	 */
2894 	if (ap->a_waitfor != MNT_LAZY)
2895 		return (0);
2896 
2897 	/*
2898 	 * Move ourselves to the back of the sync list.
2899 	 */
2900 	vn_syncer_add_to_worklist(syncvp, syncdelay);
2901 
2902 	/*
2903 	 * Walk the list of vnodes pushing all that are dirty and
2904 	 * not already on the sync list.
2905 	 */
2906 	mtx_lock(&mountlist_mtx);
2907 	if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) {
2908 		mtx_unlock(&mountlist_mtx);
2909 		return (0);
2910 	}
2911 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
2912 		vfs_unbusy(mp, td);
2913 		return (0);
2914 	}
2915 	asyncflag = mp->mnt_flag & MNT_ASYNC;
2916 	mp->mnt_flag &= ~MNT_ASYNC;
2917 	vfs_msync(mp, MNT_NOWAIT);
2918 	VFS_SYNC(mp, MNT_LAZY, ap->a_cred, td);
2919 	if (asyncflag)
2920 		mp->mnt_flag |= MNT_ASYNC;
2921 	vn_finished_write(mp);
2922 	vfs_unbusy(mp, td);
2923 	return (0);
2924 }
2925 
2926 /*
2927  * The syncer vnode is no referenced.
2928  */
2929 static int
2930 sync_inactive(ap)
2931 	struct vop_inactive_args /* {
2932 		struct vnode *a_vp;
2933 		struct thread *a_td;
2934 	} */ *ap;
2935 {
2936 
2937 	vgone(ap->a_vp);
2938 	return (0);
2939 }
2940 
2941 /*
2942  * The syncer vnode is no longer needed and is being decommissioned.
2943  *
2944  * Modifications to the worklist must be protected at splbio().
2945  */
2946 static int
2947 sync_reclaim(ap)
2948 	struct vop_reclaim_args /* {
2949 		struct vnode *a_vp;
2950 	} */ *ap;
2951 {
2952 	struct vnode *vp = ap->a_vp;
2953 	int s;
2954 
2955 	s = splbio();
2956 	vp->v_mount->mnt_syncer = NULL;
2957 	if (vp->v_flag & VONWORKLST) {
2958 		LIST_REMOVE(vp, v_synclist);
2959 		vp->v_flag &= ~VONWORKLST;
2960 	}
2961 	splx(s);
2962 
2963 	return (0);
2964 }
2965 
2966 /*
2967  * Print out a syncer vnode.
2968  */
2969 static int
2970 sync_print(ap)
2971 	struct vop_print_args /* {
2972 		struct vnode *a_vp;
2973 	} */ *ap;
2974 {
2975 	struct vnode *vp = ap->a_vp;
2976 
2977 	printf("syncer vnode");
2978 	if (vp->v_vnlock != NULL)
2979 		lockmgr_printinfo(vp->v_vnlock);
2980 	printf("\n");
2981 	return (0);
2982 }
2983 
2984 /*
2985  * extract the dev_t from a VCHR
2986  */
2987 dev_t
2988 vn_todev(vp)
2989 	struct vnode *vp;
2990 {
2991 	if (vp->v_type != VCHR)
2992 		return (NODEV);
2993 	return (vp->v_rdev);
2994 }
2995 
2996 /*
2997  * Check if vnode represents a disk device
2998  */
2999 int
3000 vn_isdisk(vp, errp)
3001 	struct vnode *vp;
3002 	int *errp;
3003 {
3004 	struct cdevsw *cdevsw;
3005 
3006 	if (vp->v_type != VCHR) {
3007 		if (errp != NULL)
3008 			*errp = ENOTBLK;
3009 		return (0);
3010 	}
3011 	if (vp->v_rdev == NULL) {
3012 		if (errp != NULL)
3013 			*errp = ENXIO;
3014 		return (0);
3015 	}
3016 	cdevsw = devsw(vp->v_rdev);
3017 	if (cdevsw == NULL) {
3018 		if (errp != NULL)
3019 			*errp = ENXIO;
3020 		return (0);
3021 	}
3022 	if (!(cdevsw->d_flags & D_DISK)) {
3023 		if (errp != NULL)
3024 			*errp = ENOTBLK;
3025 		return (0);
3026 	}
3027 	if (errp != NULL)
3028 		*errp = 0;
3029 	return (1);
3030 }
3031 
3032 /*
3033  * Free data allocated by namei(); see namei(9) for details.
3034  */
3035 void
3036 NDFREE(ndp, flags)
3037      struct nameidata *ndp;
3038      const uint flags;
3039 {
3040 	if (!(flags & NDF_NO_FREE_PNBUF) &&
3041 	    (ndp->ni_cnd.cn_flags & HASBUF)) {
3042 		uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
3043 		ndp->ni_cnd.cn_flags &= ~HASBUF;
3044 	}
3045 	if (!(flags & NDF_NO_DVP_UNLOCK) &&
3046 	    (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
3047 	    ndp->ni_dvp != ndp->ni_vp)
3048 		VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_thread);
3049 	if (!(flags & NDF_NO_DVP_RELE) &&
3050 	    (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
3051 		vrele(ndp->ni_dvp);
3052 		ndp->ni_dvp = NULL;
3053 	}
3054 	if (!(flags & NDF_NO_VP_UNLOCK) &&
3055 	    (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
3056 		VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_thread);
3057 	if (!(flags & NDF_NO_VP_RELE) &&
3058 	    ndp->ni_vp) {
3059 		vrele(ndp->ni_vp);
3060 		ndp->ni_vp = NULL;
3061 	}
3062 	if (!(flags & NDF_NO_STARTDIR_RELE) &&
3063 	    (ndp->ni_cnd.cn_flags & SAVESTART)) {
3064 		vrele(ndp->ni_startdir);
3065 		ndp->ni_startdir = NULL;
3066 	}
3067 }
3068 
3069 /*
3070  * Common file system object access control check routine.  Accepts a
3071  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3072  * and optional call-by-reference privused argument allowing vaccess()
3073  * to indicate to the caller whether privilege was used to satisfy the
3074  * request.  Returns 0 on success, or an errno on failure.
3075  */
3076 int
3077 vaccess(type, file_mode, file_uid, file_gid, acc_mode, cred, privused)
3078 	enum vtype type;
3079 	mode_t file_mode;
3080 	uid_t file_uid;
3081 	gid_t file_gid;
3082 	mode_t acc_mode;
3083 	struct ucred *cred;
3084 	int *privused;
3085 {
3086 	mode_t dac_granted;
3087 #ifdef CAPABILITIES
3088 	mode_t cap_granted;
3089 #endif
3090 
3091 	/*
3092 	 * Look for a normal, non-privileged way to access the file/directory
3093 	 * as requested.  If it exists, go with that.
3094 	 */
3095 
3096 	if (privused != NULL)
3097 		*privused = 0;
3098 
3099 	dac_granted = 0;
3100 
3101 	/* Check the owner. */
3102 	if (cred->cr_uid == file_uid) {
3103 		dac_granted |= VADMIN;
3104 		if (file_mode & S_IXUSR)
3105 			dac_granted |= VEXEC;
3106 		if (file_mode & S_IRUSR)
3107 			dac_granted |= VREAD;
3108 		if (file_mode & S_IWUSR)
3109 			dac_granted |= VWRITE;
3110 
3111 		if ((acc_mode & dac_granted) == acc_mode)
3112 			return (0);
3113 
3114 		goto privcheck;
3115 	}
3116 
3117 	/* Otherwise, check the groups (first match) */
3118 	if (groupmember(file_gid, cred)) {
3119 		if (file_mode & S_IXGRP)
3120 			dac_granted |= VEXEC;
3121 		if (file_mode & S_IRGRP)
3122 			dac_granted |= VREAD;
3123 		if (file_mode & S_IWGRP)
3124 			dac_granted |= VWRITE;
3125 
3126 		if ((acc_mode & dac_granted) == acc_mode)
3127 			return (0);
3128 
3129 		goto privcheck;
3130 	}
3131 
3132 	/* Otherwise, check everyone else. */
3133 	if (file_mode & S_IXOTH)
3134 		dac_granted |= VEXEC;
3135 	if (file_mode & S_IROTH)
3136 		dac_granted |= VREAD;
3137 	if (file_mode & S_IWOTH)
3138 		dac_granted |= VWRITE;
3139 	if ((acc_mode & dac_granted) == acc_mode)
3140 		return (0);
3141 
3142 privcheck:
3143 	if (!suser_xxx(cred, NULL, PRISON_ROOT)) {
3144 		/* XXX audit: privilege used */
3145 		if (privused != NULL)
3146 			*privused = 1;
3147 		return (0);
3148 	}
3149 
3150 #ifdef CAPABILITIES
3151 	/*
3152 	 * Build a capability mask to determine if the set of capabilities
3153 	 * satisfies the requirements when combined with the granted mask
3154 	 * from above.
3155 	 * For each capability, if the capability is required, bitwise
3156 	 * or the request type onto the cap_granted mask.
3157 	 */
3158 	cap_granted = 0;
3159 
3160 	if (type == VDIR) {
3161 		/*
3162 		 * For directories, use CAP_DAC_READ_SEARCH to satisfy
3163 		 * VEXEC requests, instead of CAP_DAC_EXECUTE.
3164 		 */
3165 		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3166 		    !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3167 			cap_granted |= VEXEC;
3168 	} else {
3169 		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3170 		    !cap_check(cred, NULL, CAP_DAC_EXECUTE, PRISON_ROOT))
3171 			cap_granted |= VEXEC;
3172 	}
3173 
3174 	if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
3175 	    !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3176 		cap_granted |= VREAD;
3177 
3178 	if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3179 	    !cap_check(cred, NULL, CAP_DAC_WRITE, PRISON_ROOT))
3180 		cap_granted |= VWRITE;
3181 
3182 	if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3183 	    !cap_check(cred, NULL, CAP_FOWNER, PRISON_ROOT))
3184 		cap_granted |= VADMIN;
3185 
3186 	if ((acc_mode & (cap_granted | dac_granted)) == acc_mode) {
3187 		/* XXX audit: privilege used */
3188 		if (privused != NULL)
3189 			*privused = 1;
3190 		return (0);
3191 	}
3192 #endif
3193 
3194 	return ((acc_mode & VADMIN) ? EPERM : EACCES);
3195 }
3196 
3197