xref: /freebsd/sys/kern/vfs_subr.c (revision ee2ea5ceafed78a5bd9810beb9e3ca927180c226)
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 /*
118  * Various variables used for debugging the new implementation of
119  * reassignbuf().
120  * XXX these are probably of (very) limited utility now.
121  */
122 static int reassignbufcalls;
123 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
124 static int reassignbufloops;
125 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW, &reassignbufloops, 0, "");
126 static int reassignbufsortgood;
127 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW, &reassignbufsortgood, 0, "");
128 static int reassignbufsortbad;
129 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW, &reassignbufsortbad, 0, "");
130 /* Set to 0 for old insertion-sort based reassignbuf, 1 for modern method. */
131 static int reassignbufmethod = 1;
132 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW, &reassignbufmethod, 0, "");
133 static int nameileafonly;
134 SYSCTL_INT(_vfs, OID_AUTO, nameileafonly, CTLFLAG_RW, &nameileafonly, 0, "");
135 
136 #ifdef ENABLE_VFS_IOOPT
137 /* See NOTES for a description of this setting. */
138 int vfs_ioopt;
139 SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, "");
140 #endif
141 
142 /* List of mounted filesystems. */
143 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
144 
145 /* For any iteration/modification of mountlist */
146 struct mtx mountlist_mtx;
147 
148 /* For any iteration/modification of mnt_vnodelist */
149 struct mtx mntvnode_mtx;
150 
151 /*
152  * Cache for the mount type id assigned to NFS.  This is used for
153  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
154  */
155 int	nfs_mount_type = -1;
156 
157 /* To keep more than one thread at a time from running vfs_getnewfsid */
158 static struct mtx mntid_mtx;
159 
160 /* For any iteration/modification of vnode_free_list */
161 static struct mtx vnode_free_list_mtx;
162 
163 /*
164  * For any iteration/modification of dev->si_hlist (linked through
165  * v_specnext)
166  */
167 static struct mtx spechash_mtx;
168 
169 /* Publicly exported FS */
170 struct nfs_public nfs_pub;
171 
172 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
173 static uma_zone_t vnode_zone;
174 static uma_zone_t vnodepoll_zone;
175 
176 /* Set to 1 to print out reclaim of active vnodes */
177 int	prtactive;
178 
179 /*
180  * The workitem queue.
181  *
182  * It is useful to delay writes of file data and filesystem metadata
183  * for tens of seconds so that quickly created and deleted files need
184  * not waste disk bandwidth being created and removed. To realize this,
185  * we append vnodes to a "workitem" queue. When running with a soft
186  * updates implementation, most pending metadata dependencies should
187  * not wait for more than a few seconds. Thus, mounted on block devices
188  * are delayed only about a half the time that file data is delayed.
189  * Similarly, directory updates are more critical, so are only delayed
190  * about a third the time that file data is delayed. Thus, there are
191  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
192  * one each second (driven off the filesystem syncer process). The
193  * syncer_delayno variable indicates the next queue that is to be processed.
194  * Items that need to be processed soon are placed in this queue:
195  *
196  *	syncer_workitem_pending[syncer_delayno]
197  *
198  * A delay of fifteen seconds is done by placing the request fifteen
199  * entries later in the queue:
200  *
201  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
202  *
203  */
204 static int syncer_delayno;
205 static long syncer_mask;
206 LIST_HEAD(synclist, vnode);
207 static struct synclist *syncer_workitem_pending;
208 
209 #define SYNCER_MAXDELAY		32
210 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
211 static int syncdelay = 30;		/* max time to delay syncing data */
212 static int filedelay = 30;		/* time to delay syncing files */
213 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
214 static int dirdelay = 29;		/* time to delay syncing directories */
215 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
216 static int metadelay = 28;		/* time to delay syncing metadata */
217 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
218 static int rushjob;		/* number of slots to run ASAP */
219 static int stat_rush_requests;	/* number of times I/O speeded up */
220 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
221 
222 /*
223  * Number of vnodes we want to exist at any one time.  This is mostly used
224  * to size hash tables in vnode-related code.  It is normally not used in
225  * getnewvnode(), as wantfreevnodes is normally nonzero.)
226  *
227  * XXX desiredvnodes is historical cruft and should not exist.
228  */
229 int desiredvnodes;
230 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
231     &desiredvnodes, 0, "Maximum number of vnodes");
232 static int minvnodes;
233 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
234     &minvnodes, 0, "Minimum number of vnodes");
235 static int vnlru_nowhere;
236 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0,
237     "Number of times the vnlru process ran without success");
238 
239 void
240 v_addpollinfo(struct vnode *vp)
241 {
242 	vp->v_pollinfo = uma_zalloc(vnodepoll_zone, M_WAITOK);
243 	mtx_init(&vp->v_pollinfo->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
244 }
245 
246 /*
247  * Initialize the vnode management data structures.
248  */
249 static void
250 vntblinit(void *dummy __unused)
251 {
252 
253 	desiredvnodes = maxproc + cnt.v_page_count / 4;
254 	minvnodes = desiredvnodes / 4;
255 	mtx_init(&mountlist_mtx, "mountlist", NULL, MTX_DEF);
256 	mtx_init(&mntvnode_mtx, "mntvnode", NULL, MTX_DEF);
257 	mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
258 	mtx_init(&spechash_mtx, "spechash", NULL, MTX_DEF);
259 	TAILQ_INIT(&vnode_free_list);
260 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
261 	vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
262 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
263 	vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
264 	      NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
265 	/*
266 	 * Initialize the filesystem syncer.
267 	 */
268 	syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
269 		&syncer_mask);
270 	syncer_maxdelay = syncer_mask + 1;
271 }
272 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL)
273 
274 
275 /*
276  * Mark a mount point as busy. Used to synchronize access and to delay
277  * unmounting. Interlock is not released on failure.
278  */
279 int
280 vfs_busy(mp, flags, interlkp, td)
281 	struct mount *mp;
282 	int flags;
283 	struct mtx *interlkp;
284 	struct thread *td;
285 {
286 	int lkflags;
287 
288 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
289 		if (flags & LK_NOWAIT)
290 			return (ENOENT);
291 		mp->mnt_kern_flag |= MNTK_MWAIT;
292 		/*
293 		 * Since all busy locks are shared except the exclusive
294 		 * lock granted when unmounting, the only place that a
295 		 * wakeup needs to be done is at the release of the
296 		 * exclusive lock at the end of dounmount.
297 		 */
298 		msleep((caddr_t)mp, interlkp, PVFS, "vfs_busy", 0);
299 		return (ENOENT);
300 	}
301 	lkflags = LK_SHARED | LK_NOPAUSE;
302 	if (interlkp)
303 		lkflags |= LK_INTERLOCK;
304 	if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
305 		panic("vfs_busy: unexpected lock failure");
306 	return (0);
307 }
308 
309 /*
310  * Free a busy filesystem.
311  */
312 void
313 vfs_unbusy(mp, td)
314 	struct mount *mp;
315 	struct thread *td;
316 {
317 
318 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
319 }
320 
321 /*
322  * Lookup a filesystem type, and if found allocate and initialize
323  * a mount structure for it.
324  *
325  * Devname is usually updated by mount(8) after booting.
326  */
327 int
328 vfs_rootmountalloc(fstypename, devname, mpp)
329 	char *fstypename;
330 	char *devname;
331 	struct mount **mpp;
332 {
333 	struct thread *td = curthread;	/* XXX */
334 	struct vfsconf *vfsp;
335 	struct mount *mp;
336 
337 	if (fstypename == NULL)
338 		return (ENODEV);
339 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
340 		if (!strcmp(vfsp->vfc_name, fstypename))
341 			break;
342 	if (vfsp == NULL)
343 		return (ENODEV);
344 	mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO);
345 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE);
346 	(void)vfs_busy(mp, LK_NOWAIT, 0, td);
347 	TAILQ_INIT(&mp->mnt_nvnodelist);
348 	TAILQ_INIT(&mp->mnt_reservedvnlist);
349 	mp->mnt_vfc = vfsp;
350 	mp->mnt_op = vfsp->vfc_vfsops;
351 	mp->mnt_flag = MNT_RDONLY;
352 	mp->mnt_vnodecovered = NULLVP;
353 	vfsp->vfc_refcount++;
354 	mp->mnt_iosize_max = DFLTPHYS;
355 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
356 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
357 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
358 	mp->mnt_stat.f_mntonname[0] = '/';
359 	mp->mnt_stat.f_mntonname[1] = 0;
360 	(void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
361 	*mpp = mp;
362 	return (0);
363 }
364 
365 /*
366  * Find an appropriate filesystem to use for the root. If a filesystem
367  * has not been preselected, walk through the list of known filesystems
368  * trying those that have mountroot routines, and try them until one
369  * works or we have tried them all.
370  */
371 #ifdef notdef	/* XXX JH */
372 int
373 lite2_vfs_mountroot()
374 {
375 	struct vfsconf *vfsp;
376 	extern int (*lite2_mountroot)(void);
377 	int error;
378 
379 	if (lite2_mountroot != NULL)
380 		return ((*lite2_mountroot)());
381 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
382 		if (vfsp->vfc_mountroot == NULL)
383 			continue;
384 		if ((error = (*vfsp->vfc_mountroot)()) == 0)
385 			return (0);
386 		printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error);
387 	}
388 	return (ENODEV);
389 }
390 #endif
391 
392 /*
393  * Lookup a mount point by filesystem identifier.
394  */
395 struct mount *
396 vfs_getvfs(fsid)
397 	fsid_t *fsid;
398 {
399 	register struct mount *mp;
400 
401 	mtx_lock(&mountlist_mtx);
402 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
403 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
404 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
405 			mtx_unlock(&mountlist_mtx);
406 			return (mp);
407 	    }
408 	}
409 	mtx_unlock(&mountlist_mtx);
410 	return ((struct mount *) 0);
411 }
412 
413 /*
414  * Get a new unique fsid.  Try to make its val[0] unique, since this value
415  * will be used to create fake device numbers for stat().  Also try (but
416  * not so hard) make its val[0] unique mod 2^16, since some emulators only
417  * support 16-bit device numbers.  We end up with unique val[0]'s for the
418  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
419  *
420  * Keep in mind that several mounts may be running in parallel.  Starting
421  * the search one past where the previous search terminated is both a
422  * micro-optimization and a defense against returning the same fsid to
423  * different mounts.
424  */
425 void
426 vfs_getnewfsid(mp)
427 	struct mount *mp;
428 {
429 	static u_int16_t mntid_base;
430 	fsid_t tfsid;
431 	int mtype;
432 
433 	mtx_lock(&mntid_mtx);
434 	mtype = mp->mnt_vfc->vfc_typenum;
435 	tfsid.val[1] = mtype;
436 	mtype = (mtype & 0xFF) << 24;
437 	for (;;) {
438 		tfsid.val[0] = makeudev(255,
439 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
440 		mntid_base++;
441 		if (vfs_getvfs(&tfsid) == NULL)
442 			break;
443 	}
444 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
445 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
446 	mtx_unlock(&mntid_mtx);
447 }
448 
449 /*
450  * Knob to control the precision of file timestamps:
451  *
452  *   0 = seconds only; nanoseconds zeroed.
453  *   1 = seconds and nanoseconds, accurate within 1/HZ.
454  *   2 = seconds and nanoseconds, truncated to microseconds.
455  * >=3 = seconds and nanoseconds, maximum precision.
456  */
457 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
458 
459 static int timestamp_precision = TSP_SEC;
460 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
461     &timestamp_precision, 0, "");
462 
463 /*
464  * Get a current timestamp.
465  */
466 void
467 vfs_timestamp(tsp)
468 	struct timespec *tsp;
469 {
470 	struct timeval tv;
471 
472 	switch (timestamp_precision) {
473 	case TSP_SEC:
474 		tsp->tv_sec = time_second;
475 		tsp->tv_nsec = 0;
476 		break;
477 	case TSP_HZ:
478 		getnanotime(tsp);
479 		break;
480 	case TSP_USEC:
481 		microtime(&tv);
482 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
483 		break;
484 	case TSP_NSEC:
485 	default:
486 		nanotime(tsp);
487 		break;
488 	}
489 }
490 
491 /*
492  * Get a mount option by its name.
493  *
494  * Return 0 if the option was found.
495  * Return ENOENT if the option wasn't found.
496  * If len is a non-NULL pointer and *len
497  * a integer different from 0, then the size
498  * of the option will be compared with *len and
499  * if they doesn't match, EINVAL is returned.
500  * If len is non-NULL and *len == 0, it will
501  * be filled with the length of the option.
502  * Finally, if buf is non-NULL, it will be
503  * filled with the address of the option.
504  */
505 int
506 vfs_getopt(opts, name, buf, len)
507 	struct vfsoptlist *opts;
508 	const char *name;
509 	void **buf;
510 	int *len;
511 {
512 	struct vfsopt *opt;
513 	int i;
514 
515 	i = 0;
516 	opt = opts->opt;
517 	while (i++ < opts->optcnt) {
518 		if (strcmp(name, opt->name) == 0) {
519 			if (len != NULL) {
520 			       	if ((*len != 0) && (*len != opt->len))
521 					return (EINVAL);
522 				*len = opt->len;
523 			}
524 			if (buf != NULL)
525 				*buf = opt->value;
526 			return (0);
527 		}
528 		opt++;
529 	}
530 	return (ENOENT);
531 }
532 
533 /*
534  * Find and copy a mount option.
535  * The size of the buffer has to be specified
536  * in len, if it is not big enough, EINVAL is
537  * returned.  Returns ENOENT if the option is
538  * not found.  Otherwise, the number of bytes
539  * actually copied are put in done if it's
540  * non-NULL and 0 is returned.
541  */
542 int
543 vfs_copyopt(opts, name, dest, len, done)
544 	struct vfsoptlist *opts;
545 	const char *name;
546 	void *dest;
547 	int len, *done;
548 {
549 	struct vfsopt *opt;
550 	int i;
551 
552 	i = 0;
553 	opt = opts->opt;
554 	while (i++ < opts->optcnt) {
555 		if (strcmp(name, opt->name) == 0) {
556 			if (len < opt->len)
557 				return (EINVAL);
558 			bcopy(dest, opt->value, opt->len);
559 			if (done != NULL)
560 				*done = opt->len;
561 			return (0);
562 		}
563 		opt++;
564 	}
565 	return (ENOENT);
566 }
567 
568 /*
569  * Set vnode attributes to VNOVAL
570  */
571 void
572 vattr_null(vap)
573 	register struct vattr *vap;
574 {
575 
576 	vap->va_type = VNON;
577 	vap->va_size = VNOVAL;
578 	vap->va_bytes = VNOVAL;
579 	vap->va_mode = VNOVAL;
580 	vap->va_nlink = VNOVAL;
581 	vap->va_uid = VNOVAL;
582 	vap->va_gid = VNOVAL;
583 	vap->va_fsid = VNOVAL;
584 	vap->va_fileid = VNOVAL;
585 	vap->va_blocksize = VNOVAL;
586 	vap->va_rdev = VNOVAL;
587 	vap->va_atime.tv_sec = VNOVAL;
588 	vap->va_atime.tv_nsec = VNOVAL;
589 	vap->va_mtime.tv_sec = VNOVAL;
590 	vap->va_mtime.tv_nsec = VNOVAL;
591 	vap->va_ctime.tv_sec = VNOVAL;
592 	vap->va_ctime.tv_nsec = VNOVAL;
593 	vap->va_flags = VNOVAL;
594 	vap->va_gen = VNOVAL;
595 	vap->va_vaflags = 0;
596 }
597 
598 /*
599  * This routine is called when we have too many vnodes.  It attempts
600  * to free <count> vnodes and will potentially free vnodes that still
601  * have VM backing store (VM backing store is typically the cause
602  * of a vnode blowout so we want to do this).  Therefore, this operation
603  * is not considered cheap.
604  *
605  * A number of conditions may prevent a vnode from being reclaimed.
606  * the buffer cache may have references on the vnode, a directory
607  * vnode may still have references due to the namei cache representing
608  * underlying files, or the vnode may be in active use.   It is not
609  * desireable to reuse such vnodes.  These conditions may cause the
610  * number of vnodes to reach some minimum value regardless of what
611  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
612  */
613 static int
614 vlrureclaim(struct mount *mp, int count)
615 {
616 	struct vnode *vp;
617 	int done;
618 	int trigger;
619 	int usevnodes;
620 
621 	/*
622 	 * Calculate the trigger point, don't allow user
623 	 * screwups to blow us up.   This prevents us from
624 	 * recycling vnodes with lots of resident pages.  We
625 	 * aren't trying to free memory, we are trying to
626 	 * free vnodes.
627 	 */
628 	usevnodes = desiredvnodes;
629 	if (usevnodes <= 0)
630 		usevnodes = 1;
631 	trigger = cnt.v_page_count * 2 / usevnodes;
632 
633 	done = 0;
634 	mtx_lock(&mntvnode_mtx);
635 	while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
636 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
637 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
638 
639 		if (vp->v_type != VNON &&
640 		    vp->v_type != VBAD &&
641 		    VMIGHTFREE(vp) &&           /* critical path opt */
642 		    (vp->v_object == NULL || vp->v_object->resident_page_count < trigger) &&
643 		    mtx_trylock(&vp->v_interlock)
644 		) {
645 			mtx_unlock(&mntvnode_mtx);
646 			if (VMIGHTFREE(vp)) {
647 				vgonel(vp, curthread);
648 				done++;
649 			} else {
650 				mtx_unlock(&vp->v_interlock);
651 			}
652 			mtx_lock(&mntvnode_mtx);
653 		}
654 		--count;
655 	}
656 	mtx_unlock(&mntvnode_mtx);
657 	return done;
658 }
659 
660 /*
661  * Attempt to recycle vnodes in a context that is always safe to block.
662  * Calling vlrurecycle() from the bowels of file system code has some
663  * interesting deadlock problems.
664  */
665 static struct proc *vnlruproc;
666 static int vnlruproc_sig;
667 
668 static void
669 vnlru_proc(void)
670 {
671 	struct mount *mp, *nmp;
672 	int s;
673 	int done;
674 	struct proc *p = vnlruproc;
675 	struct thread *td = FIRST_THREAD_IN_PROC(p);	/* XXXKSE */
676 
677 	mtx_lock(&Giant);
678 
679 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
680 	    SHUTDOWN_PRI_FIRST);
681 
682 	s = splbio();
683 	for (;;) {
684 		kthread_suspend_check(p);
685 		if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
686 			vnlruproc_sig = 0;
687 			tsleep(vnlruproc, PVFS, "vlruwt", 0);
688 			continue;
689 		}
690 		done = 0;
691 		mtx_lock(&mountlist_mtx);
692 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
693 			if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
694 				nmp = TAILQ_NEXT(mp, mnt_list);
695 				continue;
696 			}
697 			done += vlrureclaim(mp, 10);
698 			mtx_lock(&mountlist_mtx);
699 			nmp = TAILQ_NEXT(mp, mnt_list);
700 			vfs_unbusy(mp, td);
701 		}
702 		mtx_unlock(&mountlist_mtx);
703 		if (done == 0) {
704 #if 0
705 			/* These messages are temporary debugging aids */
706 			if (vnlru_nowhere < 5)
707 				printf("vnlru process getting nowhere..\n");
708 			else if (vnlru_nowhere == 5)
709 				printf("vnlru process messages stopped.\n");
710 #endif
711 			vnlru_nowhere++;
712 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
713 		}
714 	}
715 	splx(s);
716 }
717 
718 static struct kproc_desc vnlru_kp = {
719 	"vnlru",
720 	vnlru_proc,
721 	&vnlruproc
722 };
723 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
724 
725 
726 /*
727  * Routines having to do with the management of the vnode table.
728  */
729 
730 /*
731  * Return the next vnode from the free list.
732  */
733 int
734 getnewvnode(tag, mp, vops, vpp)
735 	enum vtagtype tag;
736 	struct mount *mp;
737 	vop_t **vops;
738 	struct vnode **vpp;
739 {
740 	int s;
741 	struct thread *td = curthread;	/* XXX */
742 	struct vnode *vp = NULL;
743 	struct mount *vnmp;
744 	vm_object_t object;
745 
746 	s = splbio();
747 	/*
748 	 * Try to reuse vnodes if we hit the max.  This situation only
749 	 * occurs in certain large-memory (2G+) situations.  We cannot
750 	 * attempt to directly reclaim vnodes due to nasty recursion
751 	 * problems.
752 	 */
753 	if (vnlruproc_sig == 0 && numvnodes - freevnodes > desiredvnodes) {
754 		vnlruproc_sig = 1;      /* avoid unnecessary wakeups */
755 		wakeup(vnlruproc);
756 	}
757 
758 	/*
759 	 * Attempt to reuse a vnode already on the free list, allocating
760 	 * a new vnode if we can't find one or if we have not reached a
761 	 * good minimum for good LRU performance.
762 	 */
763 
764 	mtx_lock(&vnode_free_list_mtx);
765 
766 	if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
767 		int count;
768 
769 		for (count = 0; count < freevnodes; count++) {
770 			vp = TAILQ_FIRST(&vnode_free_list);
771 			if (vp == NULL || vp->v_usecount)
772 				panic("getnewvnode: free vnode isn't");
773 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
774 
775 			if (vn_lock(vp, LK_EXCLUSIVE, td) != 0)
776 				continue;
777 			/*
778 			 * Don't recycle if we still have cached pages or if
779 			 * we cannot get the interlock.
780 			 */
781 			if ((VOP_GETVOBJECT(vp, &object) == 0 &&
782 			     (object->resident_page_count ||
783 			      object->ref_count)) ||
784 			     !mtx_trylock(&vp->v_interlock)) {
785 				TAILQ_INSERT_TAIL(&vnode_free_list, vp,
786 						    v_freelist);
787 				vp = NULL;
788 				VOP_UNLOCK(vp, 0, td);
789 				continue;
790 			}
791 			VOP_UNLOCK(vp, 0, td);
792 			if (LIST_FIRST(&vp->v_cache_src)) {
793 				/*
794 				 * note: nameileafonly sysctl is temporary,
795 				 * for debugging only, and will eventually be
796 				 * removed.
797 				 */
798 				if (nameileafonly > 0) {
799 					/*
800 					 * Do not reuse namei-cached directory
801 					 * vnodes that have cached
802 					 * subdirectories.
803 					 */
804 					if (cache_leaf_test(vp) < 0) {
805 						mtx_unlock(&vp->v_interlock);
806 						TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
807 						vp = NULL;
808 						continue;
809 					}
810 				} else if (nameileafonly < 0 ||
811 					    vmiodirenable == 0) {
812 					/*
813 					 * Do not reuse namei-cached directory
814 					 * vnodes if nameileafonly is -1 or
815 					 * if VMIO backing for directories is
816 					 * turned off (otherwise we reuse them
817 					 * too quickly).
818 					 */
819 					mtx_unlock(&vp->v_interlock);
820 					TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
821 					vp = NULL;
822 					continue;
823 				}
824 			}
825 			/*
826 			 * Skip over it if its filesystem is being suspended.
827 			 */
828 			if (vn_start_write(vp, &vnmp, V_NOWAIT) == 0)
829 				break;
830 			mtx_unlock(&vp->v_interlock);
831 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
832 			vp = NULL;
833 		}
834 	}
835 	if (vp) {
836 		vp->v_flag |= VDOOMED;
837 		vp->v_flag &= ~VFREE;
838 		freevnodes--;
839 		mtx_unlock(&vnode_free_list_mtx);
840 		cache_purge(vp);
841 		if (vp->v_type != VBAD) {
842 			vgonel(vp, td);
843 		} else {
844 			mtx_unlock(&vp->v_interlock);
845 		}
846 		vn_finished_write(vnmp);
847 
848 #ifdef INVARIANTS
849 		{
850 			int s;
851 
852 			if (vp->v_data)
853 				panic("cleaned vnode isn't");
854 			s = splbio();
855 			if (vp->v_numoutput)
856 				panic("Clean vnode has pending I/O's");
857 			splx(s);
858 			if (vp->v_writecount != 0)
859 				panic("Non-zero write count");
860 		}
861 #endif
862 		if (vp->v_pollinfo) {
863 			mtx_destroy(&vp->v_pollinfo->vpi_lock);
864 			uma_zfree(vnodepoll_zone, vp->v_pollinfo);
865 		}
866 		vp->v_pollinfo = NULL;
867 		vp->v_flag = 0;
868 		vp->v_lastw = 0;
869 		vp->v_lasta = 0;
870 		vp->v_cstart = 0;
871 		vp->v_clen = 0;
872 		vp->v_socket = 0;
873 	} else {
874 		mtx_unlock(&vnode_free_list_mtx);
875 		vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK);
876 		bzero((char *) vp, sizeof *vp);
877 		mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
878 		vp->v_dd = vp;
879 		cache_purge(vp);
880 		LIST_INIT(&vp->v_cache_src);
881 		TAILQ_INIT(&vp->v_cache_dst);
882 		numvnodes++;
883 	}
884 
885 	TAILQ_INIT(&vp->v_cleanblkhd);
886 	TAILQ_INIT(&vp->v_dirtyblkhd);
887 	vp->v_type = VNON;
888 	vp->v_tag = tag;
889 	vp->v_op = vops;
890 	lockinit(&vp->v_lock, PVFS, "vnlock", VLKTIMEOUT, LK_NOPAUSE);
891 	insmntque(vp, mp);
892 	*vpp = vp;
893 	vp->v_usecount = 1;
894 	vp->v_data = 0;
895 
896 	splx(s);
897 
898 #if 0
899 	vnodeallocs++;
900 	if (vnodeallocs % vnoderecycleperiod == 0 &&
901 	    freevnodes < vnoderecycleminfreevn &&
902 	    vnoderecyclemintotalvn < numvnodes) {
903 		/* Recycle vnodes. */
904 		cache_purgeleafdirs(vnoderecyclenumber);
905 	}
906 #endif
907 
908 	return (0);
909 }
910 
911 /*
912  * Move a vnode from one mount queue to another.
913  */
914 static void
915 insmntque(vp, mp)
916 	register struct vnode *vp;
917 	register struct mount *mp;
918 {
919 
920 	mtx_lock(&mntvnode_mtx);
921 	/*
922 	 * Delete from old mount point vnode list, if on one.
923 	 */
924 	if (vp->v_mount != NULL)
925 		TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
926 	/*
927 	 * Insert into list of vnodes for the new mount point, if available.
928 	 */
929 	if ((vp->v_mount = mp) == NULL) {
930 		mtx_unlock(&mntvnode_mtx);
931 		return;
932 	}
933 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
934 	mtx_unlock(&mntvnode_mtx);
935 }
936 
937 /*
938  * Update outstanding I/O count and do wakeup if requested.
939  */
940 void
941 vwakeup(bp)
942 	register struct buf *bp;
943 {
944 	register struct vnode *vp;
945 
946 	bp->b_flags &= ~B_WRITEINPROG;
947 	if ((vp = bp->b_vp)) {
948 		vp->v_numoutput--;
949 		if (vp->v_numoutput < 0)
950 			panic("vwakeup: neg numoutput");
951 		if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
952 			vp->v_flag &= ~VBWAIT;
953 			wakeup((caddr_t) &vp->v_numoutput);
954 		}
955 	}
956 }
957 
958 /*
959  * Flush out and invalidate all buffers associated with a vnode.
960  * Called with the underlying object locked.
961  */
962 int
963 vinvalbuf(vp, flags, cred, td, slpflag, slptimeo)
964 	register struct vnode *vp;
965 	int flags;
966 	struct ucred *cred;
967 	struct thread *td;
968 	int slpflag, slptimeo;
969 {
970 	register struct buf *bp;
971 	struct buf *nbp, *blist;
972 	int s, error;
973 	vm_object_t object;
974 
975 	GIANT_REQUIRED;
976 
977 	if (flags & V_SAVE) {
978 		s = splbio();
979 		while (vp->v_numoutput) {
980 			vp->v_flag |= VBWAIT;
981 			error = tsleep((caddr_t)&vp->v_numoutput,
982 			    slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo);
983 			if (error) {
984 				splx(s);
985 				return (error);
986 			}
987 		}
988 		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
989 			splx(s);
990 			if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, td)) != 0)
991 				return (error);
992 			s = splbio();
993 			if (vp->v_numoutput > 0 ||
994 			    !TAILQ_EMPTY(&vp->v_dirtyblkhd))
995 				panic("vinvalbuf: dirty bufs");
996 		}
997 		splx(s);
998   	}
999 	s = splbio();
1000 	for (;;) {
1001 		blist = TAILQ_FIRST(&vp->v_cleanblkhd);
1002 		if (!blist)
1003 			blist = TAILQ_FIRST(&vp->v_dirtyblkhd);
1004 		if (!blist)
1005 			break;
1006 
1007 		for (bp = blist; bp; bp = nbp) {
1008 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1009 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1010 				error = BUF_TIMELOCK(bp,
1011 				    LK_EXCLUSIVE | LK_SLEEPFAIL,
1012 				    "vinvalbuf", slpflag, slptimeo);
1013 				if (error == ENOLCK)
1014 					break;
1015 				splx(s);
1016 				return (error);
1017 			}
1018 			/*
1019 			 * XXX Since there are no node locks for NFS, I
1020 			 * believe there is a slight chance that a delayed
1021 			 * write will occur while sleeping just above, so
1022 			 * check for it.  Note that vfs_bio_awrite expects
1023 			 * buffers to reside on a queue, while BUF_WRITE and
1024 			 * brelse do not.
1025 			 */
1026 			if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1027 				(flags & V_SAVE)) {
1028 
1029 				if (bp->b_vp == vp) {
1030 					if (bp->b_flags & B_CLUSTEROK) {
1031 						BUF_UNLOCK(bp);
1032 						vfs_bio_awrite(bp);
1033 					} else {
1034 						bremfree(bp);
1035 						bp->b_flags |= B_ASYNC;
1036 						BUF_WRITE(bp);
1037 					}
1038 				} else {
1039 					bremfree(bp);
1040 					(void) BUF_WRITE(bp);
1041 				}
1042 				break;
1043 			}
1044 			bremfree(bp);
1045 			bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
1046 			bp->b_flags &= ~B_ASYNC;
1047 			brelse(bp);
1048 		}
1049 	}
1050 
1051 	/*
1052 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1053 	 * have write I/O in-progress but if there is a VM object then the
1054 	 * VM object can also have read-I/O in-progress.
1055 	 */
1056 	do {
1057 		while (vp->v_numoutput > 0) {
1058 			vp->v_flag |= VBWAIT;
1059 			tsleep(&vp->v_numoutput, PVM, "vnvlbv", 0);
1060 		}
1061 		if (VOP_GETVOBJECT(vp, &object) == 0) {
1062 			while (object->paging_in_progress)
1063 			vm_object_pip_sleep(object, "vnvlbx");
1064 		}
1065 	} while (vp->v_numoutput > 0);
1066 
1067 	splx(s);
1068 
1069 	/*
1070 	 * Destroy the copy in the VM cache, too.
1071 	 */
1072 	mtx_lock(&vp->v_interlock);
1073 	if (VOP_GETVOBJECT(vp, &object) == 0) {
1074 		vm_object_page_remove(object, 0, 0,
1075 			(flags & V_SAVE) ? TRUE : FALSE);
1076 	}
1077 	mtx_unlock(&vp->v_interlock);
1078 
1079 	if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd))
1080 		panic("vinvalbuf: flush failed");
1081 	return (0);
1082 }
1083 
1084 /*
1085  * Truncate a file's buffer and pages to a specified length.  This
1086  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1087  * sync activity.
1088  */
1089 int
1090 vtruncbuf(vp, cred, td, length, blksize)
1091 	register struct vnode *vp;
1092 	struct ucred *cred;
1093 	struct thread *td;
1094 	off_t length;
1095 	int blksize;
1096 {
1097 	register struct buf *bp;
1098 	struct buf *nbp;
1099 	int s, anyfreed;
1100 	int trunclbn;
1101 
1102 	/*
1103 	 * Round up to the *next* lbn.
1104 	 */
1105 	trunclbn = (length + blksize - 1) / blksize;
1106 
1107 	s = splbio();
1108 restart:
1109 	anyfreed = 1;
1110 	for (;anyfreed;) {
1111 		anyfreed = 0;
1112 		for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
1113 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1114 			if (bp->b_lblkno >= trunclbn) {
1115 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1116 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1117 					goto restart;
1118 				} else {
1119 					bremfree(bp);
1120 					bp->b_flags |= (B_INVAL | B_RELBUF);
1121 					bp->b_flags &= ~B_ASYNC;
1122 					brelse(bp);
1123 					anyfreed = 1;
1124 				}
1125 				if (nbp &&
1126 				    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1127 				    (nbp->b_vp != vp) ||
1128 				    (nbp->b_flags & B_DELWRI))) {
1129 					goto restart;
1130 				}
1131 			}
1132 		}
1133 
1134 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1135 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1136 			if (bp->b_lblkno >= trunclbn) {
1137 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1138 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1139 					goto restart;
1140 				} else {
1141 					bremfree(bp);
1142 					bp->b_flags |= (B_INVAL | B_RELBUF);
1143 					bp->b_flags &= ~B_ASYNC;
1144 					brelse(bp);
1145 					anyfreed = 1;
1146 				}
1147 				if (nbp &&
1148 				    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1149 				    (nbp->b_vp != vp) ||
1150 				    (nbp->b_flags & B_DELWRI) == 0)) {
1151 					goto restart;
1152 				}
1153 			}
1154 		}
1155 	}
1156 
1157 	if (length > 0) {
1158 restartsync:
1159 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1160 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1161 			if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
1162 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1163 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1164 					goto restart;
1165 				} else {
1166 					bremfree(bp);
1167 					if (bp->b_vp == vp) {
1168 						bp->b_flags |= B_ASYNC;
1169 					} else {
1170 						bp->b_flags &= ~B_ASYNC;
1171 					}
1172 					BUF_WRITE(bp);
1173 				}
1174 				goto restartsync;
1175 			}
1176 
1177 		}
1178 	}
1179 
1180 	while (vp->v_numoutput > 0) {
1181 		vp->v_flag |= VBWAIT;
1182 		tsleep(&vp->v_numoutput, PVM, "vbtrunc", 0);
1183 	}
1184 
1185 	splx(s);
1186 
1187 	vnode_pager_setsize(vp, length);
1188 
1189 	return (0);
1190 }
1191 
1192 /*
1193  * Associate a buffer with a vnode.
1194  */
1195 void
1196 bgetvp(vp, bp)
1197 	register struct vnode *vp;
1198 	register struct buf *bp;
1199 {
1200 	int s;
1201 
1202 	KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
1203 
1204 	vhold(vp);
1205 	bp->b_vp = vp;
1206 	bp->b_dev = vn_todev(vp);
1207 	/*
1208 	 * Insert onto list for new vnode.
1209 	 */
1210 	s = splbio();
1211 	bp->b_xflags |= BX_VNCLEAN;
1212 	bp->b_xflags &= ~BX_VNDIRTY;
1213 	TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
1214 	splx(s);
1215 }
1216 
1217 /*
1218  * Disassociate a buffer from a vnode.
1219  */
1220 void
1221 brelvp(bp)
1222 	register struct buf *bp;
1223 {
1224 	struct vnode *vp;
1225 	struct buflists *listheadp;
1226 	int s;
1227 
1228 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1229 
1230 	/*
1231 	 * Delete from old vnode list, if on one.
1232 	 */
1233 	vp = bp->b_vp;
1234 	s = splbio();
1235 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1236 		if (bp->b_xflags & BX_VNDIRTY)
1237 			listheadp = &vp->v_dirtyblkhd;
1238 		else
1239 			listheadp = &vp->v_cleanblkhd;
1240 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1241 		bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1242 	}
1243 	if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1244 		vp->v_flag &= ~VONWORKLST;
1245 		LIST_REMOVE(vp, v_synclist);
1246 	}
1247 	splx(s);
1248 	bp->b_vp = (struct vnode *) 0;
1249 	vdrop(vp);
1250 }
1251 
1252 /*
1253  * Add an item to the syncer work queue.
1254  */
1255 static void
1256 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
1257 {
1258 	int s, slot;
1259 
1260 	s = splbio();
1261 
1262 	if (vp->v_flag & VONWORKLST) {
1263 		LIST_REMOVE(vp, v_synclist);
1264 	}
1265 
1266 	if (delay > syncer_maxdelay - 2)
1267 		delay = syncer_maxdelay - 2;
1268 	slot = (syncer_delayno + delay) & syncer_mask;
1269 
1270 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
1271 	vp->v_flag |= VONWORKLST;
1272 	splx(s);
1273 }
1274 
1275 struct  proc *updateproc;
1276 static void sched_sync(void);
1277 static struct kproc_desc up_kp = {
1278 	"syncer",
1279 	sched_sync,
1280 	&updateproc
1281 };
1282 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1283 
1284 /*
1285  * System filesystem synchronizer daemon.
1286  */
1287 void
1288 sched_sync(void)
1289 {
1290 	struct synclist *slp;
1291 	struct vnode *vp;
1292 	struct mount *mp;
1293 	long starttime;
1294 	int s;
1295 	struct thread *td = FIRST_THREAD_IN_PROC(updateproc);  /* XXXKSE */
1296 
1297 	mtx_lock(&Giant);
1298 
1299 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, td->td_proc,
1300 	    SHUTDOWN_PRI_LAST);
1301 
1302 	for (;;) {
1303 		kthread_suspend_check(td->td_proc);
1304 
1305 		starttime = time_second;
1306 
1307 		/*
1308 		 * Push files whose dirty time has expired.  Be careful
1309 		 * of interrupt race on slp queue.
1310 		 */
1311 		s = splbio();
1312 		slp = &syncer_workitem_pending[syncer_delayno];
1313 		syncer_delayno += 1;
1314 		if (syncer_delayno == syncer_maxdelay)
1315 			syncer_delayno = 0;
1316 		splx(s);
1317 
1318 		while ((vp = LIST_FIRST(slp)) != NULL) {
1319 			if (VOP_ISLOCKED(vp, NULL) == 0 &&
1320 			    vn_start_write(vp, &mp, V_NOWAIT) == 0) {
1321 				vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1322 				(void) VOP_FSYNC(vp, td->td_ucred, MNT_LAZY, td);
1323 				VOP_UNLOCK(vp, 0, td);
1324 				vn_finished_write(mp);
1325 			}
1326 			s = splbio();
1327 			if (LIST_FIRST(slp) == vp) {
1328 				/*
1329 				 * Note: v_tag VT_VFS vps can remain on the
1330 				 * worklist too with no dirty blocks, but
1331 				 * since sync_fsync() moves it to a different
1332 				 * slot we are safe.
1333 				 */
1334 				if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1335 				    !vn_isdisk(vp, NULL))
1336 					panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
1337 				/*
1338 				 * Put us back on the worklist.  The worklist
1339 				 * routine will remove us from our current
1340 				 * position and then add us back in at a later
1341 				 * position.
1342 				 */
1343 				vn_syncer_add_to_worklist(vp, syncdelay);
1344 			}
1345 			splx(s);
1346 		}
1347 
1348 		/*
1349 		 * Do soft update processing.
1350 		 */
1351 #ifdef SOFTUPDATES
1352 		softdep_process_worklist(NULL);
1353 #endif
1354 
1355 		/*
1356 		 * The variable rushjob allows the kernel to speed up the
1357 		 * processing of the filesystem syncer process. A rushjob
1358 		 * value of N tells the filesystem syncer to process the next
1359 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1360 		 * is used by the soft update code to speed up the filesystem
1361 		 * syncer process when the incore state is getting so far
1362 		 * ahead of the disk that the kernel memory pool is being
1363 		 * threatened with exhaustion.
1364 		 */
1365 		if (rushjob > 0) {
1366 			rushjob -= 1;
1367 			continue;
1368 		}
1369 		/*
1370 		 * If it has taken us less than a second to process the
1371 		 * current work, then wait. Otherwise start right over
1372 		 * again. We can still lose time if any single round
1373 		 * takes more than two seconds, but it does not really
1374 		 * matter as we are just trying to generally pace the
1375 		 * filesystem activity.
1376 		 */
1377 		if (time_second == starttime)
1378 			tsleep(&lbolt, PPAUSE, "syncer", 0);
1379 	}
1380 }
1381 
1382 /*
1383  * Request the syncer daemon to speed up its work.
1384  * We never push it to speed up more than half of its
1385  * normal turn time, otherwise it could take over the cpu.
1386  * XXXKSE  only one update?
1387  */
1388 int
1389 speedup_syncer()
1390 {
1391 
1392 	mtx_lock_spin(&sched_lock);
1393 	if (FIRST_THREAD_IN_PROC(updateproc)->td_wchan == &lbolt) /* XXXKSE */
1394 		setrunnable(FIRST_THREAD_IN_PROC(updateproc));
1395 	mtx_unlock_spin(&sched_lock);
1396 	if (rushjob < syncdelay / 2) {
1397 		rushjob += 1;
1398 		stat_rush_requests += 1;
1399 		return (1);
1400 	}
1401 	return(0);
1402 }
1403 
1404 /*
1405  * Associate a p-buffer with a vnode.
1406  *
1407  * Also sets B_PAGING flag to indicate that vnode is not fully associated
1408  * with the buffer.  i.e. the bp has not been linked into the vnode or
1409  * ref-counted.
1410  */
1411 void
1412 pbgetvp(vp, bp)
1413 	register struct vnode *vp;
1414 	register struct buf *bp;
1415 {
1416 
1417 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1418 
1419 	bp->b_vp = vp;
1420 	bp->b_flags |= B_PAGING;
1421 	bp->b_dev = vn_todev(vp);
1422 }
1423 
1424 /*
1425  * Disassociate a p-buffer from a vnode.
1426  */
1427 void
1428 pbrelvp(bp)
1429 	register struct buf *bp;
1430 {
1431 
1432 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1433 
1434 	/* XXX REMOVE ME */
1435 	if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1436 		panic(
1437 		    "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1438 		    bp,
1439 		    (int)bp->b_flags
1440 		);
1441 	}
1442 	bp->b_vp = (struct vnode *) 0;
1443 	bp->b_flags &= ~B_PAGING;
1444 }
1445 
1446 /*
1447  * Reassign a buffer from one vnode to another.
1448  * Used to assign file specific control information
1449  * (indirect blocks) to the vnode to which they belong.
1450  */
1451 void
1452 reassignbuf(bp, newvp)
1453 	register struct buf *bp;
1454 	register struct vnode *newvp;
1455 {
1456 	struct buflists *listheadp;
1457 	int delay;
1458 	int s;
1459 
1460 	if (newvp == NULL) {
1461 		printf("reassignbuf: NULL");
1462 		return;
1463 	}
1464 	++reassignbufcalls;
1465 
1466 	/*
1467 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1468 	 * is not fully linked in.
1469 	 */
1470 	if (bp->b_flags & B_PAGING)
1471 		panic("cannot reassign paging buffer");
1472 
1473 	s = splbio();
1474 	/*
1475 	 * Delete from old vnode list, if on one.
1476 	 */
1477 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1478 		if (bp->b_xflags & BX_VNDIRTY)
1479 			listheadp = &bp->b_vp->v_dirtyblkhd;
1480 		else
1481 			listheadp = &bp->b_vp->v_cleanblkhd;
1482 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1483 		bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1484 		if (bp->b_vp != newvp) {
1485 			vdrop(bp->b_vp);
1486 			bp->b_vp = NULL;	/* for clarification */
1487 		}
1488 	}
1489 	/*
1490 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1491 	 * of clean buffers.
1492 	 */
1493 	if (bp->b_flags & B_DELWRI) {
1494 		struct buf *tbp;
1495 
1496 		listheadp = &newvp->v_dirtyblkhd;
1497 		if ((newvp->v_flag & VONWORKLST) == 0) {
1498 			switch (newvp->v_type) {
1499 			case VDIR:
1500 				delay = dirdelay;
1501 				break;
1502 			case VCHR:
1503 				if (newvp->v_rdev->si_mountpoint != NULL) {
1504 					delay = metadelay;
1505 					break;
1506 				}
1507 				/* fall through */
1508 			default:
1509 				delay = filedelay;
1510 			}
1511 			vn_syncer_add_to_worklist(newvp, delay);
1512 		}
1513 		bp->b_xflags |= BX_VNDIRTY;
1514 		tbp = TAILQ_FIRST(listheadp);
1515 		if (tbp == NULL ||
1516 		    bp->b_lblkno == 0 ||
1517 		    (bp->b_lblkno > 0 && tbp->b_lblkno < 0) ||
1518 		    (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) {
1519 			TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1520 			++reassignbufsortgood;
1521 		} else if (bp->b_lblkno < 0) {
1522 			TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1523 			++reassignbufsortgood;
1524 		} else if (reassignbufmethod == 1) {
1525 			/*
1526 			 * New sorting algorithm, only handle sequential case,
1527 			 * otherwise append to end (but before metadata)
1528 			 */
1529 			if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL &&
1530 			    (tbp->b_xflags & BX_VNDIRTY)) {
1531 				/*
1532 				 * Found the best place to insert the buffer
1533 				 */
1534 				TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1535 				++reassignbufsortgood;
1536 			} else {
1537 				/*
1538 				 * Missed, append to end, but before meta-data.
1539 				 * We know that the head buffer in the list is
1540 				 * not meta-data due to prior conditionals.
1541 				 *
1542 				 * Indirect effects:  NFS second stage write
1543 				 * tends to wind up here, giving maximum
1544 				 * distance between the unstable write and the
1545 				 * commit rpc.
1546 				 */
1547 				tbp = TAILQ_LAST(listheadp, buflists);
1548 				while (tbp && tbp->b_lblkno < 0)
1549 					tbp = TAILQ_PREV(tbp, buflists, b_vnbufs);
1550 				TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1551 				++reassignbufsortbad;
1552 			}
1553 		} else {
1554 			/*
1555 			 * Old sorting algorithm, scan queue and insert
1556 			 */
1557 			struct buf *ttbp;
1558 			while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1559 			    (ttbp->b_lblkno < bp->b_lblkno)) {
1560 				++reassignbufloops;
1561 				tbp = ttbp;
1562 			}
1563 			TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1564 		}
1565 	} else {
1566 		bp->b_xflags |= BX_VNCLEAN;
1567 		TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1568 		if ((newvp->v_flag & VONWORKLST) &&
1569 		    TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1570 			newvp->v_flag &= ~VONWORKLST;
1571 			LIST_REMOVE(newvp, v_synclist);
1572 		}
1573 	}
1574 	if (bp->b_vp != newvp) {
1575 		bp->b_vp = newvp;
1576 		vhold(bp->b_vp);
1577 	}
1578 	splx(s);
1579 }
1580 
1581 /*
1582  * Create a vnode for a device.
1583  * Used for mounting the root file system.
1584  */
1585 int
1586 bdevvp(dev, vpp)
1587 	dev_t dev;
1588 	struct vnode **vpp;
1589 {
1590 	register struct vnode *vp;
1591 	struct vnode *nvp;
1592 	int error;
1593 
1594 	if (dev == NODEV) {
1595 		*vpp = NULLVP;
1596 		return (ENXIO);
1597 	}
1598 	if (vfinddev(dev, VCHR, vpp))
1599 		return (0);
1600 	error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1601 	if (error) {
1602 		*vpp = NULLVP;
1603 		return (error);
1604 	}
1605 	vp = nvp;
1606 	vp->v_type = VCHR;
1607 	addalias(vp, dev);
1608 	*vpp = vp;
1609 	return (0);
1610 }
1611 
1612 /*
1613  * Add vnode to the alias list hung off the dev_t.
1614  *
1615  * The reason for this gunk is that multiple vnodes can reference
1616  * the same physical device, so checking vp->v_usecount to see
1617  * how many users there are is inadequate; the v_usecount for
1618  * the vnodes need to be accumulated.  vcount() does that.
1619  */
1620 struct vnode *
1621 addaliasu(nvp, nvp_rdev)
1622 	struct vnode *nvp;
1623 	udev_t nvp_rdev;
1624 {
1625 	struct vnode *ovp;
1626 	vop_t **ops;
1627 	dev_t dev;
1628 
1629 	if (nvp->v_type == VBLK)
1630 		return (nvp);
1631 	if (nvp->v_type != VCHR)
1632 		panic("addaliasu on non-special vnode");
1633 	dev = udev2dev(nvp_rdev, 0);
1634 	/*
1635 	 * Check to see if we have a bdevvp vnode with no associated
1636 	 * filesystem. If so, we want to associate the filesystem of
1637 	 * the new newly instigated vnode with the bdevvp vnode and
1638 	 * discard the newly created vnode rather than leaving the
1639 	 * bdevvp vnode lying around with no associated filesystem.
1640 	 */
1641 	if (vfinddev(dev, nvp->v_type, &ovp) == 0 || ovp->v_data != NULL) {
1642 		addalias(nvp, dev);
1643 		return (nvp);
1644 	}
1645 	/*
1646 	 * Discard unneeded vnode, but save its node specific data.
1647 	 * Note that if there is a lock, it is carried over in the
1648 	 * node specific data to the replacement vnode.
1649 	 */
1650 	vref(ovp);
1651 	ovp->v_data = nvp->v_data;
1652 	ovp->v_tag = nvp->v_tag;
1653 	nvp->v_data = NULL;
1654 	lockinit(&ovp->v_lock, PVFS, nvp->v_lock.lk_wmesg,
1655 	    nvp->v_lock.lk_timo, nvp->v_lock.lk_flags & LK_EXTFLG_MASK);
1656 	if (nvp->v_vnlock)
1657 		ovp->v_vnlock = &ovp->v_lock;
1658 	ops = ovp->v_op;
1659 	ovp->v_op = nvp->v_op;
1660 	if (VOP_ISLOCKED(nvp, curthread)) {
1661 		VOP_UNLOCK(nvp, 0, curthread);
1662 		vn_lock(ovp, LK_EXCLUSIVE | LK_RETRY, curthread);
1663 	}
1664 	nvp->v_op = ops;
1665 	insmntque(ovp, nvp->v_mount);
1666 	vrele(nvp);
1667 	vgone(nvp);
1668 	return (ovp);
1669 }
1670 
1671 /* This is a local helper function that do the same as addaliasu, but for a
1672  * dev_t instead of an udev_t. */
1673 static void
1674 addalias(nvp, dev)
1675 	struct vnode *nvp;
1676 	dev_t dev;
1677 {
1678 
1679 	KASSERT(nvp->v_type == VCHR, ("addalias on non-special vnode"));
1680 	nvp->v_rdev = dev;
1681 	mtx_lock(&spechash_mtx);
1682 	SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
1683 	mtx_unlock(&spechash_mtx);
1684 }
1685 
1686 /*
1687  * Grab a particular vnode from the free list, increment its
1688  * reference count and lock it. The vnode lock bit is set if the
1689  * vnode is being eliminated in vgone. The process is awakened
1690  * when the transition is completed, and an error returned to
1691  * indicate that the vnode is no longer usable (possibly having
1692  * been changed to a new file system type).
1693  */
1694 int
1695 vget(vp, flags, td)
1696 	register struct vnode *vp;
1697 	int flags;
1698 	struct thread *td;
1699 {
1700 	int error;
1701 
1702 	/*
1703 	 * If the vnode is in the process of being cleaned out for
1704 	 * another use, we wait for the cleaning to finish and then
1705 	 * return failure. Cleaning is determined by checking that
1706 	 * the VXLOCK flag is set.
1707 	 */
1708 	if ((flags & LK_INTERLOCK) == 0)
1709 		mtx_lock(&vp->v_interlock);
1710 	if (vp->v_flag & VXLOCK) {
1711 		if (vp->v_vxproc == curthread) {
1712 #if 0
1713 			/* this can now occur in normal operation */
1714 			log(LOG_INFO, "VXLOCK interlock avoided\n");
1715 #endif
1716 		} else {
1717 			vp->v_flag |= VXWANT;
1718 			msleep((caddr_t)vp, &vp->v_interlock, PINOD | PDROP,
1719 			    "vget", 0);
1720 			return (ENOENT);
1721 		}
1722 	}
1723 
1724 	vp->v_usecount++;
1725 
1726 	if (VSHOULDBUSY(vp))
1727 		vbusy(vp);
1728 	if (flags & LK_TYPE_MASK) {
1729 		if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
1730 			/*
1731 			 * must expand vrele here because we do not want
1732 			 * to call VOP_INACTIVE if the reference count
1733 			 * drops back to zero since it was never really
1734 			 * active. We must remove it from the free list
1735 			 * before sleeping so that multiple processes do
1736 			 * not try to recycle it.
1737 			 */
1738 			mtx_lock(&vp->v_interlock);
1739 			vp->v_usecount--;
1740 			if (VSHOULDFREE(vp))
1741 				vfree(vp);
1742 			else
1743 				vlruvp(vp);
1744 			mtx_unlock(&vp->v_interlock);
1745 		}
1746 		return (error);
1747 	}
1748 	mtx_unlock(&vp->v_interlock);
1749 	return (0);
1750 }
1751 
1752 /*
1753  * Increase the reference count of a vnode.
1754  */
1755 void
1756 vref(struct vnode *vp)
1757 {
1758 	mtx_lock(&vp->v_interlock);
1759 	vp->v_usecount++;
1760 	mtx_unlock(&vp->v_interlock);
1761 }
1762 
1763 /*
1764  * Vnode put/release.
1765  * If count drops to zero, call inactive routine and return to freelist.
1766  */
1767 void
1768 vrele(vp)
1769 	struct vnode *vp;
1770 {
1771 	struct thread *td = curthread;	/* XXX */
1772 
1773 	KASSERT(vp != NULL, ("vrele: null vp"));
1774 
1775 	mtx_lock(&vp->v_interlock);
1776 
1777 	/* Skip this v_writecount check if we're going to panic below. */
1778 	KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
1779 	    ("vrele: missed vn_close"));
1780 
1781 	if (vp->v_usecount > 1) {
1782 
1783 		vp->v_usecount--;
1784 		mtx_unlock(&vp->v_interlock);
1785 
1786 		return;
1787 	}
1788 
1789 	if (vp->v_usecount == 1) {
1790 		vp->v_usecount--;
1791 		/*
1792 		 * We must call VOP_INACTIVE with the node locked.
1793 		 * If we are doing a vput, the node is already locked,
1794 		 * but, in the case of vrele, we must explicitly lock
1795 		 * the vnode before calling VOP_INACTIVE.
1796 		 */
1797 		if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0)
1798 			VOP_INACTIVE(vp, td);
1799 		if (VSHOULDFREE(vp))
1800 			vfree(vp);
1801 		else
1802 			vlruvp(vp);
1803 
1804 	} else {
1805 #ifdef DIAGNOSTIC
1806 		vprint("vrele: negative ref count", vp);
1807 		mtx_unlock(&vp->v_interlock);
1808 #endif
1809 		panic("vrele: negative ref cnt");
1810 	}
1811 }
1812 
1813 /*
1814  * Release an already locked vnode.  This give the same effects as
1815  * unlock+vrele(), but takes less time and avoids releasing and
1816  * re-aquiring the lock (as vrele() aquires the lock internally.)
1817  */
1818 void
1819 vput(vp)
1820 	struct vnode *vp;
1821 {
1822 	struct thread *td = curthread;	/* XXX */
1823 
1824 	GIANT_REQUIRED;
1825 
1826 	KASSERT(vp != NULL, ("vput: null vp"));
1827 	mtx_lock(&vp->v_interlock);
1828 	/* Skip this v_writecount check if we're going to panic below. */
1829 	KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
1830 	    ("vput: missed vn_close"));
1831 
1832 	if (vp->v_usecount > 1) {
1833 		vp->v_usecount--;
1834 		VOP_UNLOCK(vp, LK_INTERLOCK, td);
1835 		return;
1836 	}
1837 
1838 	if (vp->v_usecount == 1) {
1839 		vp->v_usecount--;
1840 		/*
1841 		 * We must call VOP_INACTIVE with the node locked.
1842 		 * If we are doing a vput, the node is already locked,
1843 		 * so we just need to release the vnode mutex.
1844 		 */
1845 		mtx_unlock(&vp->v_interlock);
1846 		VOP_INACTIVE(vp, td);
1847 		if (VSHOULDFREE(vp))
1848 			vfree(vp);
1849 		else
1850 			vlruvp(vp);
1851 
1852 	} else {
1853 #ifdef DIAGNOSTIC
1854 		vprint("vput: negative ref count", vp);
1855 #endif
1856 		panic("vput: negative ref cnt");
1857 	}
1858 }
1859 
1860 /*
1861  * Somebody doesn't want the vnode recycled.
1862  */
1863 void
1864 vhold(vp)
1865 	register struct vnode *vp;
1866 {
1867 	int s;
1868 
1869   	s = splbio();
1870 	vp->v_holdcnt++;
1871 	if (VSHOULDBUSY(vp))
1872 		vbusy(vp);
1873 	splx(s);
1874 }
1875 
1876 /*
1877  * Note that there is one less who cares about this vnode.  vdrop() is the
1878  * opposite of vhold().
1879  */
1880 void
1881 vdrop(vp)
1882 	register struct vnode *vp;
1883 {
1884 	int s;
1885 
1886 	s = splbio();
1887 	if (vp->v_holdcnt <= 0)
1888 		panic("vdrop: holdcnt");
1889 	vp->v_holdcnt--;
1890 	if (VSHOULDFREE(vp))
1891 		vfree(vp);
1892 	else
1893 		vlruvp(vp);
1894 	splx(s);
1895 }
1896 
1897 /*
1898  * Remove any vnodes in the vnode table belonging to mount point mp.
1899  *
1900  * If FORCECLOSE is not specified, there should not be any active ones,
1901  * return error if any are found (nb: this is a user error, not a
1902  * system error). If FORCECLOSE is specified, detach any active vnodes
1903  * that are found.
1904  *
1905  * If WRITECLOSE is set, only flush out regular file vnodes open for
1906  * writing.
1907  *
1908  * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
1909  *
1910  * `rootrefs' specifies the base reference count for the root vnode
1911  * of this filesystem. The root vnode is considered busy if its
1912  * v_usecount exceeds this value. On a successful return, vflush()
1913  * will call vrele() on the root vnode exactly rootrefs times.
1914  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
1915  * be zero.
1916  */
1917 #ifdef DIAGNOSTIC
1918 static int busyprt = 0;		/* print out busy vnodes */
1919 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1920 #endif
1921 
1922 int
1923 vflush(mp, rootrefs, flags)
1924 	struct mount *mp;
1925 	int rootrefs;
1926 	int flags;
1927 {
1928 	struct thread *td = curthread;	/* XXX */
1929 	struct vnode *vp, *nvp, *rootvp = NULL;
1930 	struct vattr vattr;
1931 	int busy = 0, error;
1932 
1933 	if (rootrefs > 0) {
1934 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
1935 		    ("vflush: bad args"));
1936 		/*
1937 		 * Get the filesystem root vnode. We can vput() it
1938 		 * immediately, since with rootrefs > 0, it won't go away.
1939 		 */
1940 		if ((error = VFS_ROOT(mp, &rootvp)) != 0)
1941 			return (error);
1942 		vput(rootvp);
1943 	}
1944 	mtx_lock(&mntvnode_mtx);
1945 loop:
1946 	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) {
1947 		/*
1948 		 * Make sure this vnode wasn't reclaimed in getnewvnode().
1949 		 * Start over if it has (it won't be on the list anymore).
1950 		 */
1951 		if (vp->v_mount != mp)
1952 			goto loop;
1953 		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
1954 
1955 		mtx_unlock(&mntvnode_mtx);
1956 		mtx_lock(&vp->v_interlock);
1957 		/*
1958 		 * Skip over a vnodes marked VSYSTEM.
1959 		 */
1960 		if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1961 			mtx_unlock(&vp->v_interlock);
1962 			mtx_lock(&mntvnode_mtx);
1963 			continue;
1964 		}
1965 		/*
1966 		 * If WRITECLOSE is set, flush out unlinked but still open
1967 		 * files (even if open only for reading) and regular file
1968 		 * vnodes open for writing.
1969 		 */
1970 		if ((flags & WRITECLOSE) &&
1971 		    (vp->v_type == VNON ||
1972 		    (VOP_GETATTR(vp, &vattr, td->td_ucred, td) == 0 &&
1973 		    vattr.va_nlink > 0)) &&
1974 		    (vp->v_writecount == 0 || vp->v_type != VREG)) {
1975 			mtx_unlock(&vp->v_interlock);
1976 			mtx_lock(&mntvnode_mtx);
1977 			continue;
1978 		}
1979 
1980 		/*
1981 		 * With v_usecount == 0, all we need to do is clear out the
1982 		 * vnode data structures and we are done.
1983 		 */
1984 		if (vp->v_usecount == 0) {
1985 			vgonel(vp, td);
1986 			mtx_lock(&mntvnode_mtx);
1987 			continue;
1988 		}
1989 
1990 		/*
1991 		 * If FORCECLOSE is set, forcibly close the vnode. For block
1992 		 * or character devices, revert to an anonymous device. For
1993 		 * all other files, just kill them.
1994 		 */
1995 		if (flags & FORCECLOSE) {
1996 			if (vp->v_type != VCHR) {
1997 				vgonel(vp, td);
1998 			} else {
1999 				vclean(vp, 0, td);
2000 				vp->v_op = spec_vnodeop_p;
2001 				insmntque(vp, (struct mount *) 0);
2002 			}
2003 			mtx_lock(&mntvnode_mtx);
2004 			continue;
2005 		}
2006 #ifdef DIAGNOSTIC
2007 		if (busyprt)
2008 			vprint("vflush: busy vnode", vp);
2009 #endif
2010 		mtx_unlock(&vp->v_interlock);
2011 		mtx_lock(&mntvnode_mtx);
2012 		busy++;
2013 	}
2014 	mtx_unlock(&mntvnode_mtx);
2015 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2016 		/*
2017 		 * If just the root vnode is busy, and if its refcount
2018 		 * is equal to `rootrefs', then go ahead and kill it.
2019 		 */
2020 		mtx_lock(&rootvp->v_interlock);
2021 		KASSERT(busy > 0, ("vflush: not busy"));
2022 		KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
2023 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2024 			vgonel(rootvp, td);
2025 			busy = 0;
2026 		} else
2027 			mtx_unlock(&rootvp->v_interlock);
2028 	}
2029 	if (busy)
2030 		return (EBUSY);
2031 	for (; rootrefs > 0; rootrefs--)
2032 		vrele(rootvp);
2033 	return (0);
2034 }
2035 
2036 /*
2037  * This moves a now (likely recyclable) vnode to the end of the
2038  * mountlist.  XXX However, it is temporarily disabled until we
2039  * can clean up ffs_sync() and friends, which have loop restart
2040  * conditions which this code causes to operate O(N^2).
2041  */
2042 static void
2043 vlruvp(struct vnode *vp)
2044 {
2045 #if 0
2046 	struct mount *mp;
2047 
2048 	if ((mp = vp->v_mount) != NULL) {
2049 		mtx_lock(&mntvnode_mtx);
2050 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
2051 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
2052 		mtx_unlock(&mntvnode_mtx);
2053 	}
2054 #endif
2055 }
2056 
2057 /*
2058  * Disassociate the underlying file system from a vnode.
2059  */
2060 static void
2061 vclean(vp, flags, td)
2062 	struct vnode *vp;
2063 	int flags;
2064 	struct thread *td;
2065 {
2066 	int active;
2067 
2068 	/*
2069 	 * Check to see if the vnode is in use. If so we have to reference it
2070 	 * before we clean it out so that its count cannot fall to zero and
2071 	 * generate a race against ourselves to recycle it.
2072 	 */
2073 	if ((active = vp->v_usecount))
2074 		vp->v_usecount++;
2075 
2076 	/*
2077 	 * Prevent the vnode from being recycled or brought into use while we
2078 	 * clean it out.
2079 	 */
2080 	if (vp->v_flag & VXLOCK)
2081 		panic("vclean: deadlock");
2082 	vp->v_flag |= VXLOCK;
2083 	vp->v_vxproc = curthread;
2084 	/*
2085 	 * Even if the count is zero, the VOP_INACTIVE routine may still
2086 	 * have the object locked while it cleans it out. The VOP_LOCK
2087 	 * ensures that the VOP_INACTIVE routine is done with its work.
2088 	 * For active vnodes, it ensures that no other activity can
2089 	 * occur while the underlying object is being cleaned out.
2090 	 */
2091 	VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td);
2092 
2093 	/*
2094 	 * Clean out any buffers associated with the vnode.
2095 	 * If the flush fails, just toss the buffers.
2096 	 */
2097 	if (flags & DOCLOSE) {
2098 		if (TAILQ_FIRST(&vp->v_dirtyblkhd) != NULL)
2099 			(void) vn_write_suspend_wait(vp, NULL, V_WAIT);
2100 		if (vinvalbuf(vp, V_SAVE, NOCRED, td, 0, 0) != 0)
2101 			vinvalbuf(vp, 0, NOCRED, td, 0, 0);
2102 	}
2103 
2104 	VOP_DESTROYVOBJECT(vp);
2105 
2106 	/*
2107 	 * If purging an active vnode, it must be closed and
2108 	 * deactivated before being reclaimed. Note that the
2109 	 * VOP_INACTIVE will unlock the vnode.
2110 	 */
2111 	if (active) {
2112 		if (flags & DOCLOSE)
2113 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2114 		VOP_INACTIVE(vp, td);
2115 	} else {
2116 		/*
2117 		 * Any other processes trying to obtain this lock must first
2118 		 * wait for VXLOCK to clear, then call the new lock operation.
2119 		 */
2120 		VOP_UNLOCK(vp, 0, td);
2121 	}
2122 	/*
2123 	 * Reclaim the vnode.
2124 	 */
2125 	if (VOP_RECLAIM(vp, td))
2126 		panic("vclean: cannot reclaim");
2127 
2128 	if (active) {
2129 		/*
2130 		 * Inline copy of vrele() since VOP_INACTIVE
2131 		 * has already been called.
2132 		 */
2133 		mtx_lock(&vp->v_interlock);
2134 		if (--vp->v_usecount <= 0) {
2135 #ifdef DIAGNOSTIC
2136 			if (vp->v_usecount < 0 || vp->v_writecount != 0) {
2137 				vprint("vclean: bad ref count", vp);
2138 				panic("vclean: ref cnt");
2139 			}
2140 #endif
2141 			vfree(vp);
2142 		}
2143 		mtx_unlock(&vp->v_interlock);
2144 	}
2145 
2146 	cache_purge(vp);
2147 	vp->v_vnlock = NULL;
2148 	lockdestroy(&vp->v_lock);
2149 
2150 	if (VSHOULDFREE(vp))
2151 		vfree(vp);
2152 
2153 	/*
2154 	 * Done with purge, notify sleepers of the grim news.
2155 	 */
2156 	vp->v_op = dead_vnodeop_p;
2157 	if (vp->v_pollinfo != NULL)
2158 		vn_pollgone(vp);
2159 	vp->v_tag = VT_NON;
2160 	vp->v_flag &= ~VXLOCK;
2161 	vp->v_vxproc = NULL;
2162 	if (vp->v_flag & VXWANT) {
2163 		vp->v_flag &= ~VXWANT;
2164 		wakeup((caddr_t) vp);
2165 	}
2166 }
2167 
2168 /*
2169  * Eliminate all activity associated with the requested vnode
2170  * and with all vnodes aliased to the requested vnode.
2171  */
2172 int
2173 vop_revoke(ap)
2174 	struct vop_revoke_args /* {
2175 		struct vnode *a_vp;
2176 		int a_flags;
2177 	} */ *ap;
2178 {
2179 	struct vnode *vp, *vq;
2180 	dev_t dev;
2181 
2182 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
2183 
2184 	vp = ap->a_vp;
2185 	/*
2186 	 * If a vgone (or vclean) is already in progress,
2187 	 * wait until it is done and return.
2188 	 */
2189 	if (vp->v_flag & VXLOCK) {
2190 		vp->v_flag |= VXWANT;
2191 		msleep((caddr_t)vp, &vp->v_interlock, PINOD | PDROP,
2192 		    "vop_revokeall", 0);
2193 		return (0);
2194 	}
2195 	dev = vp->v_rdev;
2196 	for (;;) {
2197 		mtx_lock(&spechash_mtx);
2198 		vq = SLIST_FIRST(&dev->si_hlist);
2199 		mtx_unlock(&spechash_mtx);
2200 		if (!vq)
2201 			break;
2202 		vgone(vq);
2203 	}
2204 	return (0);
2205 }
2206 
2207 /*
2208  * Recycle an unused vnode to the front of the free list.
2209  * Release the passed interlock if the vnode will be recycled.
2210  */
2211 int
2212 vrecycle(vp, inter_lkp, td)
2213 	struct vnode *vp;
2214 	struct mtx *inter_lkp;
2215 	struct thread *td;
2216 {
2217 
2218 	mtx_lock(&vp->v_interlock);
2219 	if (vp->v_usecount == 0) {
2220 		if (inter_lkp) {
2221 			mtx_unlock(inter_lkp);
2222 		}
2223 		vgonel(vp, td);
2224 		return (1);
2225 	}
2226 	mtx_unlock(&vp->v_interlock);
2227 	return (0);
2228 }
2229 
2230 /*
2231  * Eliminate all activity associated with a vnode
2232  * in preparation for reuse.
2233  */
2234 void
2235 vgone(vp)
2236 	register struct vnode *vp;
2237 {
2238 	struct thread *td = curthread;	/* XXX */
2239 
2240 	mtx_lock(&vp->v_interlock);
2241 	vgonel(vp, td);
2242 }
2243 
2244 /*
2245  * vgone, with the vp interlock held.
2246  */
2247 void
2248 vgonel(vp, td)
2249 	struct vnode *vp;
2250 	struct thread *td;
2251 {
2252 	int s;
2253 
2254 	/*
2255 	 * If a vgone (or vclean) is already in progress,
2256 	 * wait until it is done and return.
2257 	 */
2258 	if (vp->v_flag & VXLOCK) {
2259 		vp->v_flag |= VXWANT;
2260 		msleep((caddr_t)vp, &vp->v_interlock, PINOD | PDROP,
2261 		    "vgone", 0);
2262 		return;
2263 	}
2264 
2265 	/*
2266 	 * Clean out the filesystem specific data.
2267 	 */
2268 	vclean(vp, DOCLOSE, td);
2269 	mtx_lock(&vp->v_interlock);
2270 
2271 	/*
2272 	 * Delete from old mount point vnode list, if on one.
2273 	 */
2274 	if (vp->v_mount != NULL)
2275 		insmntque(vp, (struct mount *)0);
2276 	/*
2277 	 * If special device, remove it from special device alias list
2278 	 * if it is on one.
2279 	 */
2280 	if (vp->v_type == VCHR && vp->v_rdev != NULL && vp->v_rdev != NODEV) {
2281 		mtx_lock(&spechash_mtx);
2282 		SLIST_REMOVE(&vp->v_rdev->si_hlist, vp, vnode, v_specnext);
2283 		freedev(vp->v_rdev);
2284 		mtx_unlock(&spechash_mtx);
2285 		vp->v_rdev = NULL;
2286 	}
2287 
2288 	/*
2289 	 * If it is on the freelist and not already at the head,
2290 	 * move it to the head of the list. The test of the
2291 	 * VDOOMED flag and the reference count of zero is because
2292 	 * it will be removed from the free list by getnewvnode,
2293 	 * but will not have its reference count incremented until
2294 	 * after calling vgone. If the reference count were
2295 	 * incremented first, vgone would (incorrectly) try to
2296 	 * close the previous instance of the underlying object.
2297 	 */
2298 	if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
2299 		s = splbio();
2300 		mtx_lock(&vnode_free_list_mtx);
2301 		if (vp->v_flag & VFREE)
2302 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2303 		else
2304 			freevnodes++;
2305 		vp->v_flag |= VFREE;
2306 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2307 		mtx_unlock(&vnode_free_list_mtx);
2308 		splx(s);
2309 	}
2310 
2311 	vp->v_type = VBAD;
2312 	mtx_unlock(&vp->v_interlock);
2313 }
2314 
2315 /*
2316  * Lookup a vnode by device number.
2317  */
2318 int
2319 vfinddev(dev, type, vpp)
2320 	dev_t dev;
2321 	enum vtype type;
2322 	struct vnode **vpp;
2323 {
2324 	struct vnode *vp;
2325 
2326 	mtx_lock(&spechash_mtx);
2327 	SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2328 		if (type == vp->v_type) {
2329 			*vpp = vp;
2330 			mtx_unlock(&spechash_mtx);
2331 			return (1);
2332 		}
2333 	}
2334 	mtx_unlock(&spechash_mtx);
2335 	return (0);
2336 }
2337 
2338 /*
2339  * Calculate the total number of references to a special device.
2340  */
2341 int
2342 vcount(vp)
2343 	struct vnode *vp;
2344 {
2345 	struct vnode *vq;
2346 	int count;
2347 
2348 	count = 0;
2349 	mtx_lock(&spechash_mtx);
2350 	SLIST_FOREACH(vq, &vp->v_rdev->si_hlist, v_specnext)
2351 		count += vq->v_usecount;
2352 	mtx_unlock(&spechash_mtx);
2353 	return (count);
2354 }
2355 
2356 /*
2357  * Same as above, but using the dev_t as argument
2358  */
2359 int
2360 count_dev(dev)
2361 	dev_t dev;
2362 {
2363 	struct vnode *vp;
2364 
2365 	vp = SLIST_FIRST(&dev->si_hlist);
2366 	if (vp == NULL)
2367 		return (0);
2368 	return(vcount(vp));
2369 }
2370 
2371 /*
2372  * Print out a description of a vnode.
2373  */
2374 static char *typename[] =
2375 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2376 
2377 void
2378 vprint(label, vp)
2379 	char *label;
2380 	struct vnode *vp;
2381 {
2382 	char buf[96];
2383 
2384 	if (label != NULL)
2385 		printf("%s: %p: ", label, (void *)vp);
2386 	else
2387 		printf("%p: ", (void *)vp);
2388 	printf("type %s, usecount %d, writecount %d, refcount %d,",
2389 	    typename[vp->v_type], vp->v_usecount, vp->v_writecount,
2390 	    vp->v_holdcnt);
2391 	buf[0] = '\0';
2392 	if (vp->v_flag & VROOT)
2393 		strcat(buf, "|VROOT");
2394 	if (vp->v_flag & VTEXT)
2395 		strcat(buf, "|VTEXT");
2396 	if (vp->v_flag & VSYSTEM)
2397 		strcat(buf, "|VSYSTEM");
2398 	if (vp->v_flag & VXLOCK)
2399 		strcat(buf, "|VXLOCK");
2400 	if (vp->v_flag & VXWANT)
2401 		strcat(buf, "|VXWANT");
2402 	if (vp->v_flag & VBWAIT)
2403 		strcat(buf, "|VBWAIT");
2404 	if (vp->v_flag & VDOOMED)
2405 		strcat(buf, "|VDOOMED");
2406 	if (vp->v_flag & VFREE)
2407 		strcat(buf, "|VFREE");
2408 	if (vp->v_flag & VOBJBUF)
2409 		strcat(buf, "|VOBJBUF");
2410 	if (buf[0] != '\0')
2411 		printf(" flags (%s)", &buf[1]);
2412 	if (vp->v_data == NULL) {
2413 		printf("\n");
2414 	} else {
2415 		printf("\n\t");
2416 		VOP_PRINT(vp);
2417 	}
2418 }
2419 
2420 #ifdef DDB
2421 #include <ddb/ddb.h>
2422 /*
2423  * List all of the locked vnodes in the system.
2424  * Called when debugging the kernel.
2425  */
2426 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
2427 {
2428 	struct thread *td = curthread;	/* XXX */
2429 	struct mount *mp, *nmp;
2430 	struct vnode *vp;
2431 
2432 	printf("Locked vnodes\n");
2433 	mtx_lock(&mountlist_mtx);
2434 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2435 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
2436 			nmp = TAILQ_NEXT(mp, mnt_list);
2437 			continue;
2438 		}
2439 		mtx_lock(&mntvnode_mtx);
2440 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2441 			if (VOP_ISLOCKED(vp, NULL))
2442 				vprint((char *)0, vp);
2443 		}
2444 		mtx_unlock(&mntvnode_mtx);
2445 		mtx_lock(&mountlist_mtx);
2446 		nmp = TAILQ_NEXT(mp, mnt_list);
2447 		vfs_unbusy(mp, td);
2448 	}
2449 	mtx_unlock(&mountlist_mtx);
2450 }
2451 #endif
2452 
2453 /*
2454  * Top level filesystem related information gathering.
2455  */
2456 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2457 
2458 static int
2459 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2460 {
2461 	int *name = (int *)arg1 - 1;	/* XXX */
2462 	u_int namelen = arg2 + 1;	/* XXX */
2463 	struct vfsconf *vfsp;
2464 
2465 #if 1 || defined(COMPAT_PRELITE2)
2466 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2467 	if (namelen == 1)
2468 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2469 #endif
2470 
2471 	/* XXX the below code does not compile; vfs_sysctl does not exist. */
2472 #ifdef notyet
2473 	/* all sysctl names at this level are at least name and field */
2474 	if (namelen < 2)
2475 		return (ENOTDIR);		/* overloaded */
2476 	if (name[0] != VFS_GENERIC) {
2477 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2478 			if (vfsp->vfc_typenum == name[0])
2479 				break;
2480 		if (vfsp == NULL)
2481 			return (EOPNOTSUPP);
2482 		return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2483 		    oldp, oldlenp, newp, newlen, td));
2484 	}
2485 #endif
2486 	switch (name[1]) {
2487 	case VFS_MAXTYPENUM:
2488 		if (namelen != 2)
2489 			return (ENOTDIR);
2490 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2491 	case VFS_CONF:
2492 		if (namelen != 3)
2493 			return (ENOTDIR);	/* overloaded */
2494 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2495 			if (vfsp->vfc_typenum == name[2])
2496 				break;
2497 		if (vfsp == NULL)
2498 			return (EOPNOTSUPP);
2499 		return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2500 	}
2501 	return (EOPNOTSUPP);
2502 }
2503 
2504 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2505 	"Generic filesystem");
2506 
2507 #if 1 || defined(COMPAT_PRELITE2)
2508 
2509 static int
2510 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2511 {
2512 	int error;
2513 	struct vfsconf *vfsp;
2514 	struct ovfsconf ovfs;
2515 
2516 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2517 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
2518 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
2519 		ovfs.vfc_index = vfsp->vfc_typenum;
2520 		ovfs.vfc_refcount = vfsp->vfc_refcount;
2521 		ovfs.vfc_flags = vfsp->vfc_flags;
2522 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2523 		if (error)
2524 			return error;
2525 	}
2526 	return 0;
2527 }
2528 
2529 #endif /* 1 || COMPAT_PRELITE2 */
2530 
2531 #if COMPILING_LINT
2532 #define KINFO_VNODESLOP	10
2533 /*
2534  * Dump vnode list (via sysctl).
2535  * Copyout address of vnode followed by vnode.
2536  */
2537 /* ARGSUSED */
2538 static int
2539 sysctl_vnode(SYSCTL_HANDLER_ARGS)
2540 {
2541 	struct thread *td = curthread;	/* XXX */
2542 	struct mount *mp, *nmp;
2543 	struct vnode *nvp, *vp;
2544 	int error;
2545 
2546 #define VPTRSZ	sizeof (struct vnode *)
2547 #define VNODESZ	sizeof (struct vnode)
2548 
2549 	req->lock = 0;
2550 	if (!req->oldptr) /* Make an estimate */
2551 		return (SYSCTL_OUT(req, 0,
2552 			(numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2553 
2554 	mtx_lock(&mountlist_mtx);
2555 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2556 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
2557 			nmp = TAILQ_NEXT(mp, mnt_list);
2558 			continue;
2559 		}
2560 		mtx_lock(&mntvnode_mtx);
2561 again:
2562 		for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2563 		     vp != NULL;
2564 		     vp = nvp) {
2565 			/*
2566 			 * Check that the vp is still associated with
2567 			 * this filesystem.  RACE: could have been
2568 			 * recycled onto the same filesystem.
2569 			 */
2570 			if (vp->v_mount != mp)
2571 				goto again;
2572 			nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2573 			mtx_unlock(&mntvnode_mtx);
2574 			if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2575 			    (error = SYSCTL_OUT(req, vp, VNODESZ)))
2576 				return (error);
2577 			mtx_lock(&mntvnode_mtx);
2578 		}
2579 		mtx_unlock(&mntvnode_mtx);
2580 		mtx_lock(&mountlist_mtx);
2581 		nmp = TAILQ_NEXT(mp, mnt_list);
2582 		vfs_unbusy(mp, td);
2583 	}
2584 	mtx_unlock(&mountlist_mtx);
2585 
2586 	return (0);
2587 }
2588 
2589 /*
2590  * XXX
2591  * Exporting the vnode list on large systems causes them to crash.
2592  * Exporting the vnode list on medium systems causes sysctl to coredump.
2593  */
2594 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2595 	0, 0, sysctl_vnode, "S,vnode", "");
2596 #endif
2597 
2598 /*
2599  * Check to see if a filesystem is mounted on a block device.
2600  */
2601 int
2602 vfs_mountedon(vp)
2603 	struct vnode *vp;
2604 {
2605 
2606 	if (vp->v_rdev->si_mountpoint != NULL)
2607 		return (EBUSY);
2608 	return (0);
2609 }
2610 
2611 /*
2612  * Unmount all filesystems. The list is traversed in reverse order
2613  * of mounting to avoid dependencies.
2614  */
2615 void
2616 vfs_unmountall()
2617 {
2618 	struct mount *mp;
2619 	struct thread *td;
2620 	int error;
2621 
2622 	if (curthread != NULL)
2623 		td = curthread;
2624 	else
2625 		td = FIRST_THREAD_IN_PROC(initproc); /* XXX XXX proc0? */
2626 	/*
2627 	 * Since this only runs when rebooting, it is not interlocked.
2628 	 */
2629 	while(!TAILQ_EMPTY(&mountlist)) {
2630 		mp = TAILQ_LAST(&mountlist, mntlist);
2631 		error = dounmount(mp, MNT_FORCE, td);
2632 		if (error) {
2633 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
2634 			printf("unmount of %s failed (",
2635 			    mp->mnt_stat.f_mntonname);
2636 			if (error == EBUSY)
2637 				printf("BUSY)\n");
2638 			else
2639 				printf("%d)\n", error);
2640 		} else {
2641 			/* The unmount has removed mp from the mountlist */
2642 		}
2643 	}
2644 }
2645 
2646 /*
2647  * perform msync on all vnodes under a mount point
2648  * the mount point must be locked.
2649  */
2650 void
2651 vfs_msync(struct mount *mp, int flags)
2652 {
2653 	struct vnode *vp, *nvp;
2654 	struct vm_object *obj;
2655 	int tries;
2656 
2657 	GIANT_REQUIRED;
2658 
2659 	tries = 5;
2660 	mtx_lock(&mntvnode_mtx);
2661 loop:
2662 	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) {
2663 		if (vp->v_mount != mp) {
2664 			if (--tries > 0)
2665 				goto loop;
2666 			break;
2667 		}
2668 		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2669 
2670 		if (vp->v_flag & VXLOCK)	/* XXX: what if MNT_WAIT? */
2671 			continue;
2672 
2673 		if (vp->v_flag & VNOSYNC)	/* unlinked, skip it */
2674 			continue;
2675 
2676 		if ((vp->v_flag & VOBJDIRTY) &&
2677 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
2678 			mtx_unlock(&mntvnode_mtx);
2679 			if (!vget(vp,
2680 			    LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, curthread)) {
2681 				if (VOP_GETVOBJECT(vp, &obj) == 0) {
2682 					vm_object_page_clean(obj, 0, 0,
2683 					    flags == MNT_WAIT ?
2684 					    OBJPC_SYNC : OBJPC_NOSYNC);
2685 				}
2686 				vput(vp);
2687 			}
2688 			mtx_lock(&mntvnode_mtx);
2689 			if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) {
2690 				if (--tries > 0)
2691 					goto loop;
2692 				break;
2693 			}
2694 		}
2695 	}
2696 	mtx_unlock(&mntvnode_mtx);
2697 }
2698 
2699 /*
2700  * Create the VM object needed for VMIO and mmap support.  This
2701  * is done for all VREG files in the system.  Some filesystems might
2702  * afford the additional metadata buffering capability of the
2703  * VMIO code by making the device node be VMIO mode also.
2704  *
2705  * vp must be locked when vfs_object_create is called.
2706  */
2707 int
2708 vfs_object_create(vp, td, cred)
2709 	struct vnode *vp;
2710 	struct thread *td;
2711 	struct ucred *cred;
2712 {
2713 	GIANT_REQUIRED;
2714 	return (VOP_CREATEVOBJECT(vp, cred, td));
2715 }
2716 
2717 /*
2718  * Mark a vnode as free, putting it up for recycling.
2719  */
2720 void
2721 vfree(vp)
2722 	struct vnode *vp;
2723 {
2724 	int s;
2725 
2726 	s = splbio();
2727 	mtx_lock(&vnode_free_list_mtx);
2728 	KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free"));
2729 	if (vp->v_flag & VAGE) {
2730 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2731 	} else {
2732 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2733 	}
2734 	freevnodes++;
2735 	mtx_unlock(&vnode_free_list_mtx);
2736 	vp->v_flag &= ~VAGE;
2737 	vp->v_flag |= VFREE;
2738 	splx(s);
2739 }
2740 
2741 /*
2742  * Opposite of vfree() - mark a vnode as in use.
2743  */
2744 void
2745 vbusy(vp)
2746 	struct vnode *vp;
2747 {
2748 	int s;
2749 
2750 	s = splbio();
2751 	mtx_lock(&vnode_free_list_mtx);
2752 	KASSERT((vp->v_flag & VFREE) != 0, ("vnode not free"));
2753 	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2754 	freevnodes--;
2755 	mtx_unlock(&vnode_free_list_mtx);
2756 	vp->v_flag &= ~(VFREE|VAGE);
2757 	splx(s);
2758 }
2759 
2760 /*
2761  * Record a process's interest in events which might happen to
2762  * a vnode.  Because poll uses the historic select-style interface
2763  * internally, this routine serves as both the ``check for any
2764  * pending events'' and the ``record my interest in future events''
2765  * functions.  (These are done together, while the lock is held,
2766  * to avoid race conditions.)
2767  */
2768 int
2769 vn_pollrecord(vp, td, events)
2770 	struct vnode *vp;
2771 	struct thread *td;
2772 	short events;
2773 {
2774 
2775 	if (vp->v_pollinfo == NULL)
2776 		v_addpollinfo(vp);
2777 	mtx_lock(&vp->v_pollinfo->vpi_lock);
2778 	if (vp->v_pollinfo->vpi_revents & events) {
2779 		/*
2780 		 * This leaves events we are not interested
2781 		 * in available for the other process which
2782 		 * which presumably had requested them
2783 		 * (otherwise they would never have been
2784 		 * recorded).
2785 		 */
2786 		events &= vp->v_pollinfo->vpi_revents;
2787 		vp->v_pollinfo->vpi_revents &= ~events;
2788 
2789 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
2790 		return events;
2791 	}
2792 	vp->v_pollinfo->vpi_events |= events;
2793 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
2794 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
2795 	return 0;
2796 }
2797 
2798 /*
2799  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
2800  * it is possible for us to miss an event due to race conditions, but
2801  * that condition is expected to be rare, so for the moment it is the
2802  * preferred interface.
2803  */
2804 void
2805 vn_pollevent(vp, events)
2806 	struct vnode *vp;
2807 	short events;
2808 {
2809 
2810 	if (vp->v_pollinfo == NULL)
2811 		v_addpollinfo(vp);
2812 	mtx_lock(&vp->v_pollinfo->vpi_lock);
2813 	if (vp->v_pollinfo->vpi_events & events) {
2814 		/*
2815 		 * We clear vpi_events so that we don't
2816 		 * call selwakeup() twice if two events are
2817 		 * posted before the polling process(es) is
2818 		 * awakened.  This also ensures that we take at
2819 		 * most one selwakeup() if the polling process
2820 		 * is no longer interested.  However, it does
2821 		 * mean that only one event can be noticed at
2822 		 * a time.  (Perhaps we should only clear those
2823 		 * event bits which we note?) XXX
2824 		 */
2825 		vp->v_pollinfo->vpi_events = 0;	/* &= ~events ??? */
2826 		vp->v_pollinfo->vpi_revents |= events;
2827 		selwakeup(&vp->v_pollinfo->vpi_selinfo);
2828 	}
2829 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
2830 }
2831 
2832 /*
2833  * Wake up anyone polling on vp because it is being revoked.
2834  * This depends on dead_poll() returning POLLHUP for correct
2835  * behavior.
2836  */
2837 void
2838 vn_pollgone(vp)
2839 	struct vnode *vp;
2840 {
2841 
2842 	mtx_lock(&vp->v_pollinfo->vpi_lock);
2843         VN_KNOTE(vp, NOTE_REVOKE);
2844 	if (vp->v_pollinfo->vpi_events) {
2845 		vp->v_pollinfo->vpi_events = 0;
2846 		selwakeup(&vp->v_pollinfo->vpi_selinfo);
2847 	}
2848 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
2849 }
2850 
2851 
2852 
2853 /*
2854  * Routine to create and manage a filesystem syncer vnode.
2855  */
2856 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
2857 static int	sync_fsync(struct  vop_fsync_args *);
2858 static int	sync_inactive(struct  vop_inactive_args *);
2859 static int	sync_reclaim(struct  vop_reclaim_args *);
2860 #define sync_lock ((int (*)(struct  vop_lock_args *))vop_nolock)
2861 #define sync_unlock ((int (*)(struct  vop_unlock_args *))vop_nounlock)
2862 static int	sync_print(struct vop_print_args *);
2863 #define sync_islocked ((int(*)(struct vop_islocked_args *))vop_noislocked)
2864 
2865 static vop_t **sync_vnodeop_p;
2866 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
2867 	{ &vop_default_desc,	(vop_t *) vop_eopnotsupp },
2868 	{ &vop_close_desc,	(vop_t *) sync_close },		/* close */
2869 	{ &vop_fsync_desc,	(vop_t *) sync_fsync },		/* fsync */
2870 	{ &vop_inactive_desc,	(vop_t *) sync_inactive },	/* inactive */
2871 	{ &vop_reclaim_desc,	(vop_t *) sync_reclaim },	/* reclaim */
2872 	{ &vop_lock_desc,	(vop_t *) sync_lock },		/* lock */
2873 	{ &vop_unlock_desc,	(vop_t *) sync_unlock },	/* unlock */
2874 	{ &vop_print_desc,	(vop_t *) sync_print },		/* print */
2875 	{ &vop_islocked_desc,	(vop_t *) sync_islocked },	/* islocked */
2876 	{ NULL, NULL }
2877 };
2878 static struct vnodeopv_desc sync_vnodeop_opv_desc =
2879 	{ &sync_vnodeop_p, sync_vnodeop_entries };
2880 
2881 VNODEOP_SET(sync_vnodeop_opv_desc);
2882 
2883 /*
2884  * Create a new filesystem syncer vnode for the specified mount point.
2885  */
2886 int
2887 vfs_allocate_syncvnode(mp)
2888 	struct mount *mp;
2889 {
2890 	struct vnode *vp;
2891 	static long start, incr, next;
2892 	int error;
2893 
2894 	/* Allocate a new vnode */
2895 	if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
2896 		mp->mnt_syncer = NULL;
2897 		return (error);
2898 	}
2899 	vp->v_type = VNON;
2900 	/*
2901 	 * Place the vnode onto the syncer worklist. We attempt to
2902 	 * scatter them about on the list so that they will go off
2903 	 * at evenly distributed times even if all the filesystems
2904 	 * are mounted at once.
2905 	 */
2906 	next += incr;
2907 	if (next == 0 || next > syncer_maxdelay) {
2908 		start /= 2;
2909 		incr /= 2;
2910 		if (start == 0) {
2911 			start = syncer_maxdelay / 2;
2912 			incr = syncer_maxdelay;
2913 		}
2914 		next = start;
2915 	}
2916 	vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
2917 	mp->mnt_syncer = vp;
2918 	return (0);
2919 }
2920 
2921 /*
2922  * Do a lazy sync of the filesystem.
2923  */
2924 static int
2925 sync_fsync(ap)
2926 	struct vop_fsync_args /* {
2927 		struct vnode *a_vp;
2928 		struct ucred *a_cred;
2929 		int a_waitfor;
2930 		struct thread *a_td;
2931 	} */ *ap;
2932 {
2933 	struct vnode *syncvp = ap->a_vp;
2934 	struct mount *mp = syncvp->v_mount;
2935 	struct thread *td = ap->a_td;
2936 	int asyncflag;
2937 
2938 	/*
2939 	 * We only need to do something if this is a lazy evaluation.
2940 	 */
2941 	if (ap->a_waitfor != MNT_LAZY)
2942 		return (0);
2943 
2944 	/*
2945 	 * Move ourselves to the back of the sync list.
2946 	 */
2947 	vn_syncer_add_to_worklist(syncvp, syncdelay);
2948 
2949 	/*
2950 	 * Walk the list of vnodes pushing all that are dirty and
2951 	 * not already on the sync list.
2952 	 */
2953 	mtx_lock(&mountlist_mtx);
2954 	if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) {
2955 		mtx_unlock(&mountlist_mtx);
2956 		return (0);
2957 	}
2958 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
2959 		vfs_unbusy(mp, td);
2960 		return (0);
2961 	}
2962 	asyncflag = mp->mnt_flag & MNT_ASYNC;
2963 	mp->mnt_flag &= ~MNT_ASYNC;
2964 	vfs_msync(mp, MNT_NOWAIT);
2965 	VFS_SYNC(mp, MNT_LAZY, ap->a_cred, td);
2966 	if (asyncflag)
2967 		mp->mnt_flag |= MNT_ASYNC;
2968 	vn_finished_write(mp);
2969 	vfs_unbusy(mp, td);
2970 	return (0);
2971 }
2972 
2973 /*
2974  * The syncer vnode is no referenced.
2975  */
2976 static int
2977 sync_inactive(ap)
2978 	struct vop_inactive_args /* {
2979 		struct vnode *a_vp;
2980 		struct thread *a_td;
2981 	} */ *ap;
2982 {
2983 
2984 	vgone(ap->a_vp);
2985 	return (0);
2986 }
2987 
2988 /*
2989  * The syncer vnode is no longer needed and is being decommissioned.
2990  *
2991  * Modifications to the worklist must be protected at splbio().
2992  */
2993 static int
2994 sync_reclaim(ap)
2995 	struct vop_reclaim_args /* {
2996 		struct vnode *a_vp;
2997 	} */ *ap;
2998 {
2999 	struct vnode *vp = ap->a_vp;
3000 	int s;
3001 
3002 	s = splbio();
3003 	vp->v_mount->mnt_syncer = NULL;
3004 	if (vp->v_flag & VONWORKLST) {
3005 		LIST_REMOVE(vp, v_synclist);
3006 		vp->v_flag &= ~VONWORKLST;
3007 	}
3008 	splx(s);
3009 
3010 	return (0);
3011 }
3012 
3013 /*
3014  * Print out a syncer vnode.
3015  */
3016 static int
3017 sync_print(ap)
3018 	struct vop_print_args /* {
3019 		struct vnode *a_vp;
3020 	} */ *ap;
3021 {
3022 	struct vnode *vp = ap->a_vp;
3023 
3024 	printf("syncer vnode");
3025 	if (vp->v_vnlock != NULL)
3026 		lockmgr_printinfo(vp->v_vnlock);
3027 	printf("\n");
3028 	return (0);
3029 }
3030 
3031 /*
3032  * extract the dev_t from a VCHR
3033  */
3034 dev_t
3035 vn_todev(vp)
3036 	struct vnode *vp;
3037 {
3038 	if (vp->v_type != VCHR)
3039 		return (NODEV);
3040 	return (vp->v_rdev);
3041 }
3042 
3043 /*
3044  * Check if vnode represents a disk device
3045  */
3046 int
3047 vn_isdisk(vp, errp)
3048 	struct vnode *vp;
3049 	int *errp;
3050 {
3051 	struct cdevsw *cdevsw;
3052 
3053 	if (vp->v_type != VCHR) {
3054 		if (errp != NULL)
3055 			*errp = ENOTBLK;
3056 		return (0);
3057 	}
3058 	if (vp->v_rdev == NULL) {
3059 		if (errp != NULL)
3060 			*errp = ENXIO;
3061 		return (0);
3062 	}
3063 	cdevsw = devsw(vp->v_rdev);
3064 	if (cdevsw == NULL) {
3065 		if (errp != NULL)
3066 			*errp = ENXIO;
3067 		return (0);
3068 	}
3069 	if (!(cdevsw->d_flags & D_DISK)) {
3070 		if (errp != NULL)
3071 			*errp = ENOTBLK;
3072 		return (0);
3073 	}
3074 	if (errp != NULL)
3075 		*errp = 0;
3076 	return (1);
3077 }
3078 
3079 /*
3080  * Free data allocated by namei(); see namei(9) for details.
3081  */
3082 void
3083 NDFREE(ndp, flags)
3084      struct nameidata *ndp;
3085      const uint flags;
3086 {
3087 	if (!(flags & NDF_NO_FREE_PNBUF) &&
3088 	    (ndp->ni_cnd.cn_flags & HASBUF)) {
3089 		uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
3090 		ndp->ni_cnd.cn_flags &= ~HASBUF;
3091 	}
3092 	if (!(flags & NDF_NO_DVP_UNLOCK) &&
3093 	    (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
3094 	    ndp->ni_dvp != ndp->ni_vp)
3095 		VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_thread);
3096 	if (!(flags & NDF_NO_DVP_RELE) &&
3097 	    (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
3098 		vrele(ndp->ni_dvp);
3099 		ndp->ni_dvp = NULL;
3100 	}
3101 	if (!(flags & NDF_NO_VP_UNLOCK) &&
3102 	    (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
3103 		VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_thread);
3104 	if (!(flags & NDF_NO_VP_RELE) &&
3105 	    ndp->ni_vp) {
3106 		vrele(ndp->ni_vp);
3107 		ndp->ni_vp = NULL;
3108 	}
3109 	if (!(flags & NDF_NO_STARTDIR_RELE) &&
3110 	    (ndp->ni_cnd.cn_flags & SAVESTART)) {
3111 		vrele(ndp->ni_startdir);
3112 		ndp->ni_startdir = NULL;
3113 	}
3114 }
3115 
3116 /*
3117  * Common file system object access control check routine.  Accepts a
3118  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3119  * and optional call-by-reference privused argument allowing vaccess()
3120  * to indicate to the caller whether privilege was used to satisfy the
3121  * request.  Returns 0 on success, or an errno on failure.
3122  */
3123 int
3124 vaccess(type, file_mode, file_uid, file_gid, acc_mode, cred, privused)
3125 	enum vtype type;
3126 	mode_t file_mode;
3127 	uid_t file_uid;
3128 	gid_t file_gid;
3129 	mode_t acc_mode;
3130 	struct ucred *cred;
3131 	int *privused;
3132 {
3133 	mode_t dac_granted;
3134 #ifdef CAPABILITIES
3135 	mode_t cap_granted;
3136 #endif
3137 
3138 	/*
3139 	 * Look for a normal, non-privileged way to access the file/directory
3140 	 * as requested.  If it exists, go with that.
3141 	 */
3142 
3143 	if (privused != NULL)
3144 		*privused = 0;
3145 
3146 	dac_granted = 0;
3147 
3148 	/* Check the owner. */
3149 	if (cred->cr_uid == file_uid) {
3150 		dac_granted |= VADMIN;
3151 		if (file_mode & S_IXUSR)
3152 			dac_granted |= VEXEC;
3153 		if (file_mode & S_IRUSR)
3154 			dac_granted |= VREAD;
3155 		if (file_mode & S_IWUSR)
3156 			dac_granted |= VWRITE;
3157 
3158 		if ((acc_mode & dac_granted) == acc_mode)
3159 			return (0);
3160 
3161 		goto privcheck;
3162 	}
3163 
3164 	/* Otherwise, check the groups (first match) */
3165 	if (groupmember(file_gid, cred)) {
3166 		if (file_mode & S_IXGRP)
3167 			dac_granted |= VEXEC;
3168 		if (file_mode & S_IRGRP)
3169 			dac_granted |= VREAD;
3170 		if (file_mode & S_IWGRP)
3171 			dac_granted |= VWRITE;
3172 
3173 		if ((acc_mode & dac_granted) == acc_mode)
3174 			return (0);
3175 
3176 		goto privcheck;
3177 	}
3178 
3179 	/* Otherwise, check everyone else. */
3180 	if (file_mode & S_IXOTH)
3181 		dac_granted |= VEXEC;
3182 	if (file_mode & S_IROTH)
3183 		dac_granted |= VREAD;
3184 	if (file_mode & S_IWOTH)
3185 		dac_granted |= VWRITE;
3186 	if ((acc_mode & dac_granted) == acc_mode)
3187 		return (0);
3188 
3189 privcheck:
3190 	if (!suser_cred(cred, PRISON_ROOT)) {
3191 		/* XXX audit: privilege used */
3192 		if (privused != NULL)
3193 			*privused = 1;
3194 		return (0);
3195 	}
3196 
3197 #ifdef CAPABILITIES
3198 	/*
3199 	 * Build a capability mask to determine if the set of capabilities
3200 	 * satisfies the requirements when combined with the granted mask
3201 	 * from above.
3202 	 * For each capability, if the capability is required, bitwise
3203 	 * or the request type onto the cap_granted mask.
3204 	 */
3205 	cap_granted = 0;
3206 
3207 	if (type == VDIR) {
3208 		/*
3209 		 * For directories, use CAP_DAC_READ_SEARCH to satisfy
3210 		 * VEXEC requests, instead of CAP_DAC_EXECUTE.
3211 		 */
3212 		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3213 		    !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3214 			cap_granted |= VEXEC;
3215 	} else {
3216 		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3217 		    !cap_check(cred, NULL, CAP_DAC_EXECUTE, PRISON_ROOT))
3218 			cap_granted |= VEXEC;
3219 	}
3220 
3221 	if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
3222 	    !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3223 		cap_granted |= VREAD;
3224 
3225 	if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3226 	    !cap_check(cred, NULL, CAP_DAC_WRITE, PRISON_ROOT))
3227 		cap_granted |= VWRITE;
3228 
3229 	if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3230 	    !cap_check(cred, NULL, CAP_FOWNER, PRISON_ROOT))
3231 		cap_granted |= VADMIN;
3232 
3233 	if ((acc_mode & (cap_granted | dac_granted)) == acc_mode) {
3234 		/* XXX audit: privilege used */
3235 		if (privused != NULL)
3236 			*privused = 1;
3237 		return (0);
3238 	}
3239 #endif
3240 
3241 	return ((acc_mode & VADMIN) ? EPERM : EACCES);
3242 }
3243 
3244