xref: /freebsd/sys/kern/vfs_subr.c (revision 09e8dea79366f1e5b3a73e8a271b26e4b6bf2e6a)
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(_vfs, 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(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
113 /* Number of vnodes in the free list. */
114 static u_long freevnodes;
115 SYSCTL_LONG(_vfs, 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  * Build a linked list of mount options from a struct uio.
493  */
494 int
495 vfs_buildopts(struct uio *auio, struct vfsoptlist **options)
496 {
497 	struct vfsoptlist *opts;
498 	struct vfsopt *opt;
499 	unsigned int i, iovcnt;
500 	int error, namelen, optlen;
501 
502 	iovcnt = auio->uio_iovcnt;
503 	opts = malloc(sizeof(struct vfsoptlist), M_MOUNT, M_WAITOK);
504 	TAILQ_INIT(opts);
505 	for (i = 0; i < iovcnt; i += 2) {
506 		opt = malloc(sizeof(struct vfsopt), M_MOUNT, M_WAITOK);
507 		namelen = auio->uio_iov[i].iov_len;
508 		optlen = auio->uio_iov[i + 1].iov_len;
509 		opt->name = malloc(namelen, M_MOUNT, M_WAITOK);
510 		opt->value = malloc(optlen, M_MOUNT, M_WAITOK);
511 		opt->len = optlen;
512 		if (auio->uio_segflg == UIO_SYSSPACE) {
513 			bcopy(auio->uio_iov[i].iov_base, opt->name, namelen);
514 			bcopy(auio->uio_iov[i + 1].iov_base, opt->value,
515 			    optlen);
516 		} else {
517 			error = copyin(auio->uio_iov[i].iov_base, opt->name,
518 			    namelen);
519 			if (!error)
520 				error = copyin(auio->uio_iov[i + 1].iov_base,
521 				    opt->value, optlen);
522 			if (error)
523 				goto bad;
524 		}
525 		TAILQ_INSERT_TAIL(opts, opt, link);
526 	}
527 	*options = opts;
528 	return (0);
529 bad:
530 	vfs_freeopts(opts);
531 	return (error);
532 }
533 
534 /*
535  * Get a mount option by its name.
536  *
537  * Return 0 if the option was found, ENOENT otherwise.
538  * If len is non-NULL it will be filled with the length
539  * of the option. If buf is non-NULL, it will be filled
540  * with the address of the option.
541  */
542 int
543 vfs_getopt(opts, name, buf, len)
544 	struct vfsoptlist *opts;
545 	const char *name;
546 	void **buf;
547 	int *len;
548 {
549 	struct vfsopt *opt;
550 
551 	TAILQ_FOREACH(opt, opts, link) {
552 		if (strcmp(name, opt->name) == 0) {
553 			if (len != NULL)
554 				*len = opt->len;
555 			if (buf != NULL)
556 				*buf = opt->value;
557 			return (0);
558 		}
559 	}
560 	return (ENOENT);
561 }
562 
563 /*
564  * Find and copy a mount option.
565  *
566  * The size of the buffer has to be specified
567  * in len, if it is not the same length as the
568  * mount option, EINVAL is returned.
569  * Returns ENOENT if the option is not found.
570  */
571 int
572 vfs_copyopt(opts, name, dest, len)
573 	struct vfsoptlist *opts;
574 	const char *name;
575 	void *dest;
576 	int len;
577 {
578 	struct vfsopt *opt;
579 
580 	TAILQ_FOREACH(opt, opts, link) {
581 		if (strcmp(name, opt->name) == 0) {
582 			if (len != opt->len)
583 				return (EINVAL);
584 			bcopy(opt->value, dest, opt->len);
585 			return (0);
586 		}
587 	}
588 	return (ENOENT);
589 }
590 
591 /*
592  * Set vnode attributes to VNOVAL
593  */
594 void
595 vattr_null(vap)
596 	register struct vattr *vap;
597 {
598 
599 	vap->va_type = VNON;
600 	vap->va_size = VNOVAL;
601 	vap->va_bytes = VNOVAL;
602 	vap->va_mode = VNOVAL;
603 	vap->va_nlink = VNOVAL;
604 	vap->va_uid = VNOVAL;
605 	vap->va_gid = VNOVAL;
606 	vap->va_fsid = VNOVAL;
607 	vap->va_fileid = VNOVAL;
608 	vap->va_blocksize = VNOVAL;
609 	vap->va_rdev = VNOVAL;
610 	vap->va_atime.tv_sec = VNOVAL;
611 	vap->va_atime.tv_nsec = VNOVAL;
612 	vap->va_mtime.tv_sec = VNOVAL;
613 	vap->va_mtime.tv_nsec = VNOVAL;
614 	vap->va_ctime.tv_sec = VNOVAL;
615 	vap->va_ctime.tv_nsec = VNOVAL;
616 	vap->va_flags = VNOVAL;
617 	vap->va_gen = VNOVAL;
618 	vap->va_vaflags = 0;
619 }
620 
621 /*
622  * This routine is called when we have too many vnodes.  It attempts
623  * to free <count> vnodes and will potentially free vnodes that still
624  * have VM backing store (VM backing store is typically the cause
625  * of a vnode blowout so we want to do this).  Therefore, this operation
626  * is not considered cheap.
627  *
628  * A number of conditions may prevent a vnode from being reclaimed.
629  * the buffer cache may have references on the vnode, a directory
630  * vnode may still have references due to the namei cache representing
631  * underlying files, or the vnode may be in active use.   It is not
632  * desireable to reuse such vnodes.  These conditions may cause the
633  * number of vnodes to reach some minimum value regardless of what
634  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
635  */
636 static int
637 vlrureclaim(struct mount *mp, int count)
638 {
639 	struct vnode *vp;
640 	int done;
641 	int trigger;
642 	int usevnodes;
643 
644 	/*
645 	 * Calculate the trigger point, don't allow user
646 	 * screwups to blow us up.   This prevents us from
647 	 * recycling vnodes with lots of resident pages.  We
648 	 * aren't trying to free memory, we are trying to
649 	 * free vnodes.
650 	 */
651 	usevnodes = desiredvnodes;
652 	if (usevnodes <= 0)
653 		usevnodes = 1;
654 	trigger = cnt.v_page_count * 2 / usevnodes;
655 
656 	done = 0;
657 	mtx_lock(&mntvnode_mtx);
658 	while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
659 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
660 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
661 
662 		if (vp->v_type != VNON &&
663 		    vp->v_type != VBAD &&
664 		    VMIGHTFREE(vp) &&           /* critical path opt */
665 		    (vp->v_object == NULL || vp->v_object->resident_page_count < trigger) &&
666 		    mtx_trylock(&vp->v_interlock)
667 		) {
668 			mtx_unlock(&mntvnode_mtx);
669 			if (VMIGHTFREE(vp)) {
670 				vgonel(vp, curthread);
671 				done++;
672 			} else {
673 				mtx_unlock(&vp->v_interlock);
674 			}
675 			mtx_lock(&mntvnode_mtx);
676 		}
677 		--count;
678 	}
679 	mtx_unlock(&mntvnode_mtx);
680 	return done;
681 }
682 
683 /*
684  * Attempt to recycle vnodes in a context that is always safe to block.
685  * Calling vlrurecycle() from the bowels of filesystem code has some
686  * interesting deadlock problems.
687  */
688 static struct proc *vnlruproc;
689 static int vnlruproc_sig;
690 
691 static void
692 vnlru_proc(void)
693 {
694 	struct mount *mp, *nmp;
695 	int s;
696 	int done;
697 	struct proc *p = vnlruproc;
698 	struct thread *td = FIRST_THREAD_IN_PROC(p);	/* XXXKSE */
699 
700 	mtx_lock(&Giant);
701 
702 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
703 	    SHUTDOWN_PRI_FIRST);
704 
705 	s = splbio();
706 	for (;;) {
707 		kthread_suspend_check(p);
708 		if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
709 			vnlruproc_sig = 0;
710 			tsleep(vnlruproc, PVFS, "vlruwt", 0);
711 			continue;
712 		}
713 		done = 0;
714 		mtx_lock(&mountlist_mtx);
715 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
716 			if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
717 				nmp = TAILQ_NEXT(mp, mnt_list);
718 				continue;
719 			}
720 			done += vlrureclaim(mp, 10);
721 			mtx_lock(&mountlist_mtx);
722 			nmp = TAILQ_NEXT(mp, mnt_list);
723 			vfs_unbusy(mp, td);
724 		}
725 		mtx_unlock(&mountlist_mtx);
726 		if (done == 0) {
727 #if 0
728 			/* These messages are temporary debugging aids */
729 			if (vnlru_nowhere < 5)
730 				printf("vnlru process getting nowhere..\n");
731 			else if (vnlru_nowhere == 5)
732 				printf("vnlru process messages stopped.\n");
733 #endif
734 			vnlru_nowhere++;
735 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
736 		}
737 	}
738 	splx(s);
739 }
740 
741 static struct kproc_desc vnlru_kp = {
742 	"vnlru",
743 	vnlru_proc,
744 	&vnlruproc
745 };
746 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
747 
748 
749 /*
750  * Routines having to do with the management of the vnode table.
751  */
752 
753 /*
754  * Return the next vnode from the free list.
755  */
756 int
757 getnewvnode(tag, mp, vops, vpp)
758 	enum vtagtype tag;
759 	struct mount *mp;
760 	vop_t **vops;
761 	struct vnode **vpp;
762 {
763 	int s;
764 	struct thread *td = curthread;	/* XXX */
765 	struct vnode *vp = NULL;
766 	struct mount *vnmp;
767 	vm_object_t object;
768 
769 	s = splbio();
770 	/*
771 	 * Try to reuse vnodes if we hit the max.  This situation only
772 	 * occurs in certain large-memory (2G+) situations.  We cannot
773 	 * attempt to directly reclaim vnodes due to nasty recursion
774 	 * problems.
775 	 */
776 	if (vnlruproc_sig == 0 && numvnodes - freevnodes > desiredvnodes) {
777 		vnlruproc_sig = 1;      /* avoid unnecessary wakeups */
778 		wakeup(vnlruproc);
779 	}
780 
781 	/*
782 	 * Attempt to reuse a vnode already on the free list, allocating
783 	 * a new vnode if we can't find one or if we have not reached a
784 	 * good minimum for good LRU performance.
785 	 */
786 
787 	mtx_lock(&vnode_free_list_mtx);
788 
789 	if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
790 		int count;
791 
792 		for (count = 0; count < freevnodes; count++) {
793 			vp = TAILQ_FIRST(&vnode_free_list);
794 			if (vp == NULL || vp->v_usecount)
795 				panic("getnewvnode: free vnode isn't");
796 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
797 
798 			/* Don't recycle if we can't get the interlock */
799 			if (!mtx_trylock(&vp->v_interlock)) {
800 				vp = NULL;
801 				continue;
802 			}
803 
804 			/* We should be able to immediately acquire this */
805 			if (vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE, td) != 0)
806 				continue;
807 			/*
808 			 * Don't recycle if we still have cached pages.
809 			 */
810 			if (VOP_GETVOBJECT(vp, &object) == 0 &&
811 			     (object->resident_page_count ||
812 			      object->ref_count)) {
813 				TAILQ_INSERT_TAIL(&vnode_free_list, vp,
814 						    v_freelist);
815 				vp = NULL;
816 				VOP_UNLOCK(vp, 0, td);
817 				continue;
818 			}
819 			if (LIST_FIRST(&vp->v_cache_src)) {
820 				/*
821 				 * note: nameileafonly sysctl is temporary,
822 				 * for debugging only, and will eventually be
823 				 * removed.
824 				 */
825 				if (nameileafonly > 0) {
826 					/*
827 					 * Do not reuse namei-cached directory
828 					 * vnodes that have cached
829 					 * subdirectories.
830 					 */
831 					if (cache_leaf_test(vp) < 0) {
832 						VOP_UNLOCK(vp, 0, td);
833 						TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
834 						vp = NULL;
835 						continue;
836 					}
837 				} else if (nameileafonly < 0 ||
838 					    vmiodirenable == 0) {
839 					/*
840 					 * Do not reuse namei-cached directory
841 					 * vnodes if nameileafonly is -1 or
842 					 * if VMIO backing for directories is
843 					 * turned off (otherwise we reuse them
844 					 * too quickly).
845 					 */
846 					VOP_UNLOCK(vp, 0, td);
847 					TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
848 					vp = NULL;
849 					continue;
850 				}
851 			}
852 			/*
853 			 * Skip over it if its filesystem is being suspended.
854 			 */
855 			if (vn_start_write(vp, &vnmp, V_NOWAIT) == 0)
856 				break;
857 			VOP_UNLOCK(vp, 0, td);
858 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
859 			vp = NULL;
860 		}
861 	}
862 	if (vp) {
863 		vp->v_flag |= VDOOMED;
864 		vp->v_flag &= ~VFREE;
865 		freevnodes--;
866 		mtx_unlock(&vnode_free_list_mtx);
867 		cache_purge(vp);
868 		if (vp->v_type != VBAD) {
869 			VOP_UNLOCK(vp, 0, td);
870 			vgone(vp);
871 		} else {
872 			VOP_UNLOCK(vp, 0, td);
873 		}
874 		vn_finished_write(vnmp);
875 
876 #ifdef INVARIANTS
877 		{
878 			int s;
879 
880 			if (vp->v_data)
881 				panic("cleaned vnode isn't");
882 			s = splbio();
883 			if (vp->v_numoutput)
884 				panic("Clean vnode has pending I/O's");
885 			splx(s);
886 			if (vp->v_writecount != 0)
887 				panic("Non-zero write count");
888 		}
889 #endif
890 		if (vp->v_pollinfo) {
891 			mtx_destroy(&vp->v_pollinfo->vpi_lock);
892 			uma_zfree(vnodepoll_zone, vp->v_pollinfo);
893 		}
894 		vp->v_pollinfo = NULL;
895 		vp->v_flag = 0;
896 		vp->v_lastw = 0;
897 		vp->v_lasta = 0;
898 		vp->v_cstart = 0;
899 		vp->v_clen = 0;
900 		vp->v_socket = 0;
901 	} else {
902 		mtx_unlock(&vnode_free_list_mtx);
903 		vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK);
904 		bzero((char *) vp, sizeof *vp);
905 		mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
906 		vp->v_dd = vp;
907 		cache_purge(vp);
908 		LIST_INIT(&vp->v_cache_src);
909 		TAILQ_INIT(&vp->v_cache_dst);
910 		numvnodes++;
911 	}
912 
913 	TAILQ_INIT(&vp->v_cleanblkhd);
914 	TAILQ_INIT(&vp->v_dirtyblkhd);
915 	vp->v_type = VNON;
916 	vp->v_tag = tag;
917 	vp->v_op = vops;
918 	lockinit(&vp->v_lock, PVFS, "vnlock", VLKTIMEOUT, LK_NOPAUSE);
919 	insmntque(vp, mp);
920 	*vpp = vp;
921 	vp->v_usecount = 1;
922 	vp->v_data = 0;
923 
924 	splx(s);
925 
926 #if 0
927 	vnodeallocs++;
928 	if (vnodeallocs % vnoderecycleperiod == 0 &&
929 	    freevnodes < vnoderecycleminfreevn &&
930 	    vnoderecyclemintotalvn < numvnodes) {
931 		/* Recycle vnodes. */
932 		cache_purgeleafdirs(vnoderecyclenumber);
933 	}
934 #endif
935 
936 	return (0);
937 }
938 
939 /*
940  * Move a vnode from one mount queue to another.
941  */
942 static void
943 insmntque(vp, mp)
944 	register struct vnode *vp;
945 	register struct mount *mp;
946 {
947 
948 	mtx_lock(&mntvnode_mtx);
949 	/*
950 	 * Delete from old mount point vnode list, if on one.
951 	 */
952 	if (vp->v_mount != NULL)
953 		TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
954 	/*
955 	 * Insert into list of vnodes for the new mount point, if available.
956 	 */
957 	if ((vp->v_mount = mp) == NULL) {
958 		mtx_unlock(&mntvnode_mtx);
959 		return;
960 	}
961 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
962 	mtx_unlock(&mntvnode_mtx);
963 }
964 
965 /*
966  * Update outstanding I/O count and do wakeup if requested.
967  */
968 void
969 vwakeup(bp)
970 	register struct buf *bp;
971 {
972 	register struct vnode *vp;
973 
974 	bp->b_flags &= ~B_WRITEINPROG;
975 	if ((vp = bp->b_vp)) {
976 		vp->v_numoutput--;
977 		if (vp->v_numoutput < 0)
978 			panic("vwakeup: neg numoutput");
979 		if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
980 			vp->v_flag &= ~VBWAIT;
981 			wakeup((caddr_t) &vp->v_numoutput);
982 		}
983 	}
984 }
985 
986 /*
987  * Flush out and invalidate all buffers associated with a vnode.
988  * Called with the underlying object locked.
989  */
990 int
991 vinvalbuf(vp, flags, cred, td, slpflag, slptimeo)
992 	register struct vnode *vp;
993 	int flags;
994 	struct ucred *cred;
995 	struct thread *td;
996 	int slpflag, slptimeo;
997 {
998 	register struct buf *bp;
999 	struct buf *nbp, *blist;
1000 	int s, error;
1001 	vm_object_t object;
1002 
1003 	GIANT_REQUIRED;
1004 
1005 	if (flags & V_SAVE) {
1006 		s = splbio();
1007 		while (vp->v_numoutput) {
1008 			vp->v_flag |= VBWAIT;
1009 			error = tsleep((caddr_t)&vp->v_numoutput,
1010 			    slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo);
1011 			if (error) {
1012 				splx(s);
1013 				return (error);
1014 			}
1015 		}
1016 		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1017 			splx(s);
1018 			if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, td)) != 0)
1019 				return (error);
1020 			s = splbio();
1021 			if (vp->v_numoutput > 0 ||
1022 			    !TAILQ_EMPTY(&vp->v_dirtyblkhd))
1023 				panic("vinvalbuf: dirty bufs");
1024 		}
1025 		splx(s);
1026 	}
1027 	s = splbio();
1028 	for (;;) {
1029 		blist = TAILQ_FIRST(&vp->v_cleanblkhd);
1030 		if (!blist)
1031 			blist = TAILQ_FIRST(&vp->v_dirtyblkhd);
1032 		if (!blist)
1033 			break;
1034 
1035 		for (bp = blist; bp; bp = nbp) {
1036 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1037 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1038 				error = BUF_TIMELOCK(bp,
1039 				    LK_EXCLUSIVE | LK_SLEEPFAIL,
1040 				    "vinvalbuf", slpflag, slptimeo);
1041 				if (error == ENOLCK)
1042 					break;
1043 				splx(s);
1044 				return (error);
1045 			}
1046 			/*
1047 			 * XXX Since there are no node locks for NFS, I
1048 			 * believe there is a slight chance that a delayed
1049 			 * write will occur while sleeping just above, so
1050 			 * check for it.  Note that vfs_bio_awrite expects
1051 			 * buffers to reside on a queue, while BUF_WRITE and
1052 			 * brelse do not.
1053 			 */
1054 			if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1055 				(flags & V_SAVE)) {
1056 
1057 				if (bp->b_vp == vp) {
1058 					if (bp->b_flags & B_CLUSTEROK) {
1059 						BUF_UNLOCK(bp);
1060 						vfs_bio_awrite(bp);
1061 					} else {
1062 						bremfree(bp);
1063 						bp->b_flags |= B_ASYNC;
1064 						BUF_WRITE(bp);
1065 					}
1066 				} else {
1067 					bremfree(bp);
1068 					(void) BUF_WRITE(bp);
1069 				}
1070 				break;
1071 			}
1072 			bremfree(bp);
1073 			bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
1074 			bp->b_flags &= ~B_ASYNC;
1075 			brelse(bp);
1076 		}
1077 	}
1078 
1079 	/*
1080 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1081 	 * have write I/O in-progress but if there is a VM object then the
1082 	 * VM object can also have read-I/O in-progress.
1083 	 */
1084 	do {
1085 		while (vp->v_numoutput > 0) {
1086 			vp->v_flag |= VBWAIT;
1087 			tsleep(&vp->v_numoutput, PVM, "vnvlbv", 0);
1088 		}
1089 		if (VOP_GETVOBJECT(vp, &object) == 0) {
1090 			while (object->paging_in_progress)
1091 			vm_object_pip_sleep(object, "vnvlbx");
1092 		}
1093 	} while (vp->v_numoutput > 0);
1094 
1095 	splx(s);
1096 
1097 	/*
1098 	 * Destroy the copy in the VM cache, too.
1099 	 */
1100 	mtx_lock(&vp->v_interlock);
1101 	if (VOP_GETVOBJECT(vp, &object) == 0) {
1102 		vm_object_page_remove(object, 0, 0,
1103 			(flags & V_SAVE) ? TRUE : FALSE);
1104 	}
1105 	mtx_unlock(&vp->v_interlock);
1106 
1107 	if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd))
1108 		panic("vinvalbuf: flush failed");
1109 	return (0);
1110 }
1111 
1112 /*
1113  * Truncate a file's buffer and pages to a specified length.  This
1114  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1115  * sync activity.
1116  */
1117 int
1118 vtruncbuf(vp, cred, td, length, blksize)
1119 	register struct vnode *vp;
1120 	struct ucred *cred;
1121 	struct thread *td;
1122 	off_t length;
1123 	int blksize;
1124 {
1125 	register struct buf *bp;
1126 	struct buf *nbp;
1127 	int s, anyfreed;
1128 	int trunclbn;
1129 
1130 	/*
1131 	 * Round up to the *next* lbn.
1132 	 */
1133 	trunclbn = (length + blksize - 1) / blksize;
1134 
1135 	s = splbio();
1136 restart:
1137 	anyfreed = 1;
1138 	for (;anyfreed;) {
1139 		anyfreed = 0;
1140 		for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
1141 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1142 			if (bp->b_lblkno >= trunclbn) {
1143 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1144 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1145 					goto restart;
1146 				} else {
1147 					bremfree(bp);
1148 					bp->b_flags |= (B_INVAL | B_RELBUF);
1149 					bp->b_flags &= ~B_ASYNC;
1150 					brelse(bp);
1151 					anyfreed = 1;
1152 				}
1153 				if (nbp &&
1154 				    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1155 				    (nbp->b_vp != vp) ||
1156 				    (nbp->b_flags & B_DELWRI))) {
1157 					goto restart;
1158 				}
1159 			}
1160 		}
1161 
1162 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1163 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1164 			if (bp->b_lblkno >= trunclbn) {
1165 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1166 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1167 					goto restart;
1168 				} else {
1169 					bremfree(bp);
1170 					bp->b_flags |= (B_INVAL | B_RELBUF);
1171 					bp->b_flags &= ~B_ASYNC;
1172 					brelse(bp);
1173 					anyfreed = 1;
1174 				}
1175 				if (nbp &&
1176 				    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1177 				    (nbp->b_vp != vp) ||
1178 				    (nbp->b_flags & B_DELWRI) == 0)) {
1179 					goto restart;
1180 				}
1181 			}
1182 		}
1183 	}
1184 
1185 	if (length > 0) {
1186 restartsync:
1187 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1188 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1189 			if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
1190 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1191 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1192 					goto restart;
1193 				} else {
1194 					bremfree(bp);
1195 					if (bp->b_vp == vp) {
1196 						bp->b_flags |= B_ASYNC;
1197 					} else {
1198 						bp->b_flags &= ~B_ASYNC;
1199 					}
1200 					BUF_WRITE(bp);
1201 				}
1202 				goto restartsync;
1203 			}
1204 
1205 		}
1206 	}
1207 
1208 	while (vp->v_numoutput > 0) {
1209 		vp->v_flag |= VBWAIT;
1210 		tsleep(&vp->v_numoutput, PVM, "vbtrunc", 0);
1211 	}
1212 
1213 	splx(s);
1214 
1215 	vnode_pager_setsize(vp, length);
1216 
1217 	return (0);
1218 }
1219 
1220 /*
1221  * Associate a buffer with a vnode.
1222  */
1223 void
1224 bgetvp(vp, bp)
1225 	register struct vnode *vp;
1226 	register struct buf *bp;
1227 {
1228 	int s;
1229 
1230 	KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
1231 
1232 	vhold(vp);
1233 	bp->b_vp = vp;
1234 	bp->b_dev = vn_todev(vp);
1235 	/*
1236 	 * Insert onto list for new vnode.
1237 	 */
1238 	s = splbio();
1239 	bp->b_xflags |= BX_VNCLEAN;
1240 	bp->b_xflags &= ~BX_VNDIRTY;
1241 	TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
1242 	splx(s);
1243 }
1244 
1245 /*
1246  * Disassociate a buffer from a vnode.
1247  */
1248 void
1249 brelvp(bp)
1250 	register struct buf *bp;
1251 {
1252 	struct vnode *vp;
1253 	struct buflists *listheadp;
1254 	int s;
1255 
1256 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1257 
1258 	/*
1259 	 * Delete from old vnode list, if on one.
1260 	 */
1261 	vp = bp->b_vp;
1262 	s = splbio();
1263 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1264 		if (bp->b_xflags & BX_VNDIRTY)
1265 			listheadp = &vp->v_dirtyblkhd;
1266 		else
1267 			listheadp = &vp->v_cleanblkhd;
1268 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1269 		bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1270 	}
1271 	if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1272 		vp->v_flag &= ~VONWORKLST;
1273 		LIST_REMOVE(vp, v_synclist);
1274 	}
1275 	splx(s);
1276 	bp->b_vp = (struct vnode *) 0;
1277 	vdrop(vp);
1278 }
1279 
1280 /*
1281  * Add an item to the syncer work queue.
1282  */
1283 static void
1284 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
1285 {
1286 	int s, slot;
1287 
1288 	s = splbio();
1289 
1290 	if (vp->v_flag & VONWORKLST) {
1291 		LIST_REMOVE(vp, v_synclist);
1292 	}
1293 
1294 	if (delay > syncer_maxdelay - 2)
1295 		delay = syncer_maxdelay - 2;
1296 	slot = (syncer_delayno + delay) & syncer_mask;
1297 
1298 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
1299 	vp->v_flag |= VONWORKLST;
1300 	splx(s);
1301 }
1302 
1303 struct  proc *updateproc;
1304 static void sched_sync(void);
1305 static struct kproc_desc up_kp = {
1306 	"syncer",
1307 	sched_sync,
1308 	&updateproc
1309 };
1310 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1311 
1312 /*
1313  * System filesystem synchronizer daemon.
1314  */
1315 void
1316 sched_sync(void)
1317 {
1318 	struct synclist *slp;
1319 	struct vnode *vp;
1320 	struct mount *mp;
1321 	long starttime;
1322 	int s;
1323 	struct thread *td = FIRST_THREAD_IN_PROC(updateproc);  /* XXXKSE */
1324 
1325 	mtx_lock(&Giant);
1326 
1327 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, td->td_proc,
1328 	    SHUTDOWN_PRI_LAST);
1329 
1330 	for (;;) {
1331 		kthread_suspend_check(td->td_proc);
1332 
1333 		starttime = time_second;
1334 
1335 		/*
1336 		 * Push files whose dirty time has expired.  Be careful
1337 		 * of interrupt race on slp queue.
1338 		 */
1339 		s = splbio();
1340 		slp = &syncer_workitem_pending[syncer_delayno];
1341 		syncer_delayno += 1;
1342 		if (syncer_delayno == syncer_maxdelay)
1343 			syncer_delayno = 0;
1344 		splx(s);
1345 
1346 		while ((vp = LIST_FIRST(slp)) != NULL) {
1347 			if (VOP_ISLOCKED(vp, NULL) == 0 &&
1348 			    vn_start_write(vp, &mp, V_NOWAIT) == 0) {
1349 				vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1350 				(void) VOP_FSYNC(vp, td->td_ucred, MNT_LAZY, td);
1351 				VOP_UNLOCK(vp, 0, td);
1352 				vn_finished_write(mp);
1353 			}
1354 			s = splbio();
1355 			if (LIST_FIRST(slp) == vp) {
1356 				/*
1357 				 * Note: v_tag VT_VFS vps can remain on the
1358 				 * worklist too with no dirty blocks, but
1359 				 * since sync_fsync() moves it to a different
1360 				 * slot we are safe.
1361 				 */
1362 				if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1363 				    !vn_isdisk(vp, NULL))
1364 					panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
1365 				/*
1366 				 * Put us back on the worklist.  The worklist
1367 				 * routine will remove us from our current
1368 				 * position and then add us back in at a later
1369 				 * position.
1370 				 */
1371 				vn_syncer_add_to_worklist(vp, syncdelay);
1372 			}
1373 			splx(s);
1374 		}
1375 
1376 		/*
1377 		 * Do soft update processing.
1378 		 */
1379 #ifdef SOFTUPDATES
1380 		softdep_process_worklist(NULL);
1381 #endif
1382 
1383 		/*
1384 		 * The variable rushjob allows the kernel to speed up the
1385 		 * processing of the filesystem syncer process. A rushjob
1386 		 * value of N tells the filesystem syncer to process the next
1387 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1388 		 * is used by the soft update code to speed up the filesystem
1389 		 * syncer process when the incore state is getting so far
1390 		 * ahead of the disk that the kernel memory pool is being
1391 		 * threatened with exhaustion.
1392 		 */
1393 		if (rushjob > 0) {
1394 			rushjob -= 1;
1395 			continue;
1396 		}
1397 		/*
1398 		 * If it has taken us less than a second to process the
1399 		 * current work, then wait. Otherwise start right over
1400 		 * again. We can still lose time if any single round
1401 		 * takes more than two seconds, but it does not really
1402 		 * matter as we are just trying to generally pace the
1403 		 * filesystem activity.
1404 		 */
1405 		if (time_second == starttime)
1406 			tsleep(&lbolt, PPAUSE, "syncer", 0);
1407 	}
1408 }
1409 
1410 /*
1411  * Request the syncer daemon to speed up its work.
1412  * We never push it to speed up more than half of its
1413  * normal turn time, otherwise it could take over the cpu.
1414  * XXXKSE  only one update?
1415  */
1416 int
1417 speedup_syncer()
1418 {
1419 
1420 	mtx_lock_spin(&sched_lock);
1421 	if (FIRST_THREAD_IN_PROC(updateproc)->td_wchan == &lbolt) /* XXXKSE */
1422 		setrunnable(FIRST_THREAD_IN_PROC(updateproc));
1423 	mtx_unlock_spin(&sched_lock);
1424 	if (rushjob < syncdelay / 2) {
1425 		rushjob += 1;
1426 		stat_rush_requests += 1;
1427 		return (1);
1428 	}
1429 	return(0);
1430 }
1431 
1432 /*
1433  * Associate a p-buffer with a vnode.
1434  *
1435  * Also sets B_PAGING flag to indicate that vnode is not fully associated
1436  * with the buffer.  i.e. the bp has not been linked into the vnode or
1437  * ref-counted.
1438  */
1439 void
1440 pbgetvp(vp, bp)
1441 	register struct vnode *vp;
1442 	register struct buf *bp;
1443 {
1444 
1445 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1446 
1447 	bp->b_vp = vp;
1448 	bp->b_flags |= B_PAGING;
1449 	bp->b_dev = vn_todev(vp);
1450 }
1451 
1452 /*
1453  * Disassociate a p-buffer from a vnode.
1454  */
1455 void
1456 pbrelvp(bp)
1457 	register struct buf *bp;
1458 {
1459 
1460 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1461 
1462 	/* XXX REMOVE ME */
1463 	if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1464 		panic(
1465 		    "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1466 		    bp,
1467 		    (int)bp->b_flags
1468 		);
1469 	}
1470 	bp->b_vp = (struct vnode *) 0;
1471 	bp->b_flags &= ~B_PAGING;
1472 }
1473 
1474 /*
1475  * Reassign a buffer from one vnode to another.
1476  * Used to assign file specific control information
1477  * (indirect blocks) to the vnode to which they belong.
1478  */
1479 void
1480 reassignbuf(bp, newvp)
1481 	register struct buf *bp;
1482 	register struct vnode *newvp;
1483 {
1484 	struct buflists *listheadp;
1485 	int delay;
1486 	int s;
1487 
1488 	if (newvp == NULL) {
1489 		printf("reassignbuf: NULL");
1490 		return;
1491 	}
1492 	++reassignbufcalls;
1493 
1494 	/*
1495 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1496 	 * is not fully linked in.
1497 	 */
1498 	if (bp->b_flags & B_PAGING)
1499 		panic("cannot reassign paging buffer");
1500 
1501 	s = splbio();
1502 	/*
1503 	 * Delete from old vnode list, if on one.
1504 	 */
1505 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1506 		if (bp->b_xflags & BX_VNDIRTY)
1507 			listheadp = &bp->b_vp->v_dirtyblkhd;
1508 		else
1509 			listheadp = &bp->b_vp->v_cleanblkhd;
1510 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1511 		bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1512 		if (bp->b_vp != newvp) {
1513 			vdrop(bp->b_vp);
1514 			bp->b_vp = NULL;	/* for clarification */
1515 		}
1516 	}
1517 	/*
1518 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1519 	 * of clean buffers.
1520 	 */
1521 	if (bp->b_flags & B_DELWRI) {
1522 		struct buf *tbp;
1523 
1524 		listheadp = &newvp->v_dirtyblkhd;
1525 		if ((newvp->v_flag & VONWORKLST) == 0) {
1526 			switch (newvp->v_type) {
1527 			case VDIR:
1528 				delay = dirdelay;
1529 				break;
1530 			case VCHR:
1531 				if (newvp->v_rdev->si_mountpoint != NULL) {
1532 					delay = metadelay;
1533 					break;
1534 				}
1535 				/* fall through */
1536 			default:
1537 				delay = filedelay;
1538 			}
1539 			vn_syncer_add_to_worklist(newvp, delay);
1540 		}
1541 		bp->b_xflags |= BX_VNDIRTY;
1542 		tbp = TAILQ_FIRST(listheadp);
1543 		if (tbp == NULL ||
1544 		    bp->b_lblkno == 0 ||
1545 		    (bp->b_lblkno > 0 && tbp->b_lblkno < 0) ||
1546 		    (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) {
1547 			TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1548 			++reassignbufsortgood;
1549 		} else if (bp->b_lblkno < 0) {
1550 			TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1551 			++reassignbufsortgood;
1552 		} else if (reassignbufmethod == 1) {
1553 			/*
1554 			 * New sorting algorithm, only handle sequential case,
1555 			 * otherwise append to end (but before metadata)
1556 			 */
1557 			if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL &&
1558 			    (tbp->b_xflags & BX_VNDIRTY)) {
1559 				/*
1560 				 * Found the best place to insert the buffer
1561 				 */
1562 				TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1563 				++reassignbufsortgood;
1564 			} else {
1565 				/*
1566 				 * Missed, append to end, but before meta-data.
1567 				 * We know that the head buffer in the list is
1568 				 * not meta-data due to prior conditionals.
1569 				 *
1570 				 * Indirect effects:  NFS second stage write
1571 				 * tends to wind up here, giving maximum
1572 				 * distance between the unstable write and the
1573 				 * commit rpc.
1574 				 */
1575 				tbp = TAILQ_LAST(listheadp, buflists);
1576 				while (tbp && tbp->b_lblkno < 0)
1577 					tbp = TAILQ_PREV(tbp, buflists, b_vnbufs);
1578 				TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1579 				++reassignbufsortbad;
1580 			}
1581 		} else {
1582 			/*
1583 			 * Old sorting algorithm, scan queue and insert
1584 			 */
1585 			struct buf *ttbp;
1586 			while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1587 			    (ttbp->b_lblkno < bp->b_lblkno)) {
1588 				++reassignbufloops;
1589 				tbp = ttbp;
1590 			}
1591 			TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1592 		}
1593 	} else {
1594 		bp->b_xflags |= BX_VNCLEAN;
1595 		TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1596 		if ((newvp->v_flag & VONWORKLST) &&
1597 		    TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1598 			newvp->v_flag &= ~VONWORKLST;
1599 			LIST_REMOVE(newvp, v_synclist);
1600 		}
1601 	}
1602 	if (bp->b_vp != newvp) {
1603 		bp->b_vp = newvp;
1604 		vhold(bp->b_vp);
1605 	}
1606 	splx(s);
1607 }
1608 
1609 /*
1610  * Create a vnode for a device.
1611  * Used for mounting the root filesystem.
1612  */
1613 int
1614 bdevvp(dev, vpp)
1615 	dev_t dev;
1616 	struct vnode **vpp;
1617 {
1618 	register struct vnode *vp;
1619 	struct vnode *nvp;
1620 	int error;
1621 
1622 	if (dev == NODEV) {
1623 		*vpp = NULLVP;
1624 		return (ENXIO);
1625 	}
1626 	if (vfinddev(dev, VCHR, vpp))
1627 		return (0);
1628 	error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1629 	if (error) {
1630 		*vpp = NULLVP;
1631 		return (error);
1632 	}
1633 	vp = nvp;
1634 	vp->v_type = VCHR;
1635 	addalias(vp, dev);
1636 	*vpp = vp;
1637 	return (0);
1638 }
1639 
1640 /*
1641  * Add vnode to the alias list hung off the dev_t.
1642  *
1643  * The reason for this gunk is that multiple vnodes can reference
1644  * the same physical device, so checking vp->v_usecount to see
1645  * how many users there are is inadequate; the v_usecount for
1646  * the vnodes need to be accumulated.  vcount() does that.
1647  */
1648 struct vnode *
1649 addaliasu(nvp, nvp_rdev)
1650 	struct vnode *nvp;
1651 	udev_t nvp_rdev;
1652 {
1653 	struct vnode *ovp;
1654 	vop_t **ops;
1655 	dev_t dev;
1656 
1657 	if (nvp->v_type == VBLK)
1658 		return (nvp);
1659 	if (nvp->v_type != VCHR)
1660 		panic("addaliasu on non-special vnode");
1661 	dev = udev2dev(nvp_rdev, 0);
1662 	/*
1663 	 * Check to see if we have a bdevvp vnode with no associated
1664 	 * filesystem. If so, we want to associate the filesystem of
1665 	 * the new newly instigated vnode with the bdevvp vnode and
1666 	 * discard the newly created vnode rather than leaving the
1667 	 * bdevvp vnode lying around with no associated filesystem.
1668 	 */
1669 	if (vfinddev(dev, nvp->v_type, &ovp) == 0 || ovp->v_data != NULL) {
1670 		addalias(nvp, dev);
1671 		return (nvp);
1672 	}
1673 	/*
1674 	 * Discard unneeded vnode, but save its node specific data.
1675 	 * Note that if there is a lock, it is carried over in the
1676 	 * node specific data to the replacement vnode.
1677 	 */
1678 	vref(ovp);
1679 	ovp->v_data = nvp->v_data;
1680 	ovp->v_tag = nvp->v_tag;
1681 	nvp->v_data = NULL;
1682 	lockinit(&ovp->v_lock, PVFS, nvp->v_lock.lk_wmesg,
1683 	    nvp->v_lock.lk_timo, nvp->v_lock.lk_flags & LK_EXTFLG_MASK);
1684 	if (nvp->v_vnlock)
1685 		ovp->v_vnlock = &ovp->v_lock;
1686 	ops = ovp->v_op;
1687 	ovp->v_op = nvp->v_op;
1688 	if (VOP_ISLOCKED(nvp, curthread)) {
1689 		VOP_UNLOCK(nvp, 0, curthread);
1690 		vn_lock(ovp, LK_EXCLUSIVE | LK_RETRY, curthread);
1691 	}
1692 	nvp->v_op = ops;
1693 	insmntque(ovp, nvp->v_mount);
1694 	vrele(nvp);
1695 	vgone(nvp);
1696 	return (ovp);
1697 }
1698 
1699 /* This is a local helper function that do the same as addaliasu, but for a
1700  * dev_t instead of an udev_t. */
1701 static void
1702 addalias(nvp, dev)
1703 	struct vnode *nvp;
1704 	dev_t dev;
1705 {
1706 
1707 	KASSERT(nvp->v_type == VCHR, ("addalias on non-special vnode"));
1708 	nvp->v_rdev = dev;
1709 	mtx_lock(&spechash_mtx);
1710 	SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
1711 	mtx_unlock(&spechash_mtx);
1712 }
1713 
1714 /*
1715  * Grab a particular vnode from the free list, increment its
1716  * reference count and lock it. The vnode lock bit is set if the
1717  * vnode is being eliminated in vgone. The process is awakened
1718  * when the transition is completed, and an error returned to
1719  * indicate that the vnode is no longer usable (possibly having
1720  * been changed to a new filesystem type).
1721  */
1722 int
1723 vget(vp, flags, td)
1724 	register struct vnode *vp;
1725 	int flags;
1726 	struct thread *td;
1727 {
1728 	int error;
1729 
1730 	/*
1731 	 * If the vnode is in the process of being cleaned out for
1732 	 * another use, we wait for the cleaning to finish and then
1733 	 * return failure. Cleaning is determined by checking that
1734 	 * the VXLOCK flag is set.
1735 	 */
1736 	if ((flags & LK_INTERLOCK) == 0)
1737 		mtx_lock(&vp->v_interlock);
1738 	if (vp->v_flag & VXLOCK) {
1739 		if (vp->v_vxproc == curthread) {
1740 #if 0
1741 			/* this can now occur in normal operation */
1742 			log(LOG_INFO, "VXLOCK interlock avoided\n");
1743 #endif
1744 		} else {
1745 			vp->v_flag |= VXWANT;
1746 			msleep((caddr_t)vp, &vp->v_interlock, PINOD | PDROP,
1747 			    "vget", 0);
1748 			return (ENOENT);
1749 		}
1750 	}
1751 
1752 	vp->v_usecount++;
1753 
1754 	if (VSHOULDBUSY(vp))
1755 		vbusy(vp);
1756 	if (flags & LK_TYPE_MASK) {
1757 		if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
1758 			/*
1759 			 * must expand vrele here because we do not want
1760 			 * to call VOP_INACTIVE if the reference count
1761 			 * drops back to zero since it was never really
1762 			 * active. We must remove it from the free list
1763 			 * before sleeping so that multiple processes do
1764 			 * not try to recycle it.
1765 			 */
1766 			mtx_lock(&vp->v_interlock);
1767 			vp->v_usecount--;
1768 			if (VSHOULDFREE(vp))
1769 				vfree(vp);
1770 			else
1771 				vlruvp(vp);
1772 			mtx_unlock(&vp->v_interlock);
1773 		}
1774 		return (error);
1775 	}
1776 	mtx_unlock(&vp->v_interlock);
1777 	return (0);
1778 }
1779 
1780 /*
1781  * Increase the reference count of a vnode.
1782  */
1783 void
1784 vref(struct vnode *vp)
1785 {
1786 	mtx_lock(&vp->v_interlock);
1787 	vp->v_usecount++;
1788 	mtx_unlock(&vp->v_interlock);
1789 }
1790 
1791 /*
1792  * Vnode put/release.
1793  * If count drops to zero, call inactive routine and return to freelist.
1794  */
1795 void
1796 vrele(vp)
1797 	struct vnode *vp;
1798 {
1799 	struct thread *td = curthread;	/* XXX */
1800 
1801 	KASSERT(vp != NULL, ("vrele: null vp"));
1802 
1803 	mtx_lock(&vp->v_interlock);
1804 
1805 	/* Skip this v_writecount check if we're going to panic below. */
1806 	KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
1807 	    ("vrele: missed vn_close"));
1808 
1809 	if (vp->v_usecount > 1) {
1810 
1811 		vp->v_usecount--;
1812 		mtx_unlock(&vp->v_interlock);
1813 
1814 		return;
1815 	}
1816 
1817 	if (vp->v_usecount == 1) {
1818 		vp->v_usecount--;
1819 		/*
1820 		 * We must call VOP_INACTIVE with the node locked.
1821 		 * If we are doing a vput, the node is already locked,
1822 		 * but, in the case of vrele, we must explicitly lock
1823 		 * the vnode before calling VOP_INACTIVE.
1824 		 */
1825 		if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0)
1826 			VOP_INACTIVE(vp, td);
1827 		if (VSHOULDFREE(vp))
1828 			vfree(vp);
1829 		else
1830 			vlruvp(vp);
1831 
1832 	} else {
1833 #ifdef DIAGNOSTIC
1834 		vprint("vrele: negative ref count", vp);
1835 		mtx_unlock(&vp->v_interlock);
1836 #endif
1837 		panic("vrele: negative ref cnt");
1838 	}
1839 }
1840 
1841 /*
1842  * Release an already locked vnode.  This give the same effects as
1843  * unlock+vrele(), but takes less time and avoids releasing and
1844  * re-aquiring the lock (as vrele() aquires the lock internally.)
1845  */
1846 void
1847 vput(vp)
1848 	struct vnode *vp;
1849 {
1850 	struct thread *td = curthread;	/* XXX */
1851 
1852 	GIANT_REQUIRED;
1853 
1854 	KASSERT(vp != NULL, ("vput: null vp"));
1855 	mtx_lock(&vp->v_interlock);
1856 	/* Skip this v_writecount check if we're going to panic below. */
1857 	KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
1858 	    ("vput: missed vn_close"));
1859 
1860 	if (vp->v_usecount > 1) {
1861 		vp->v_usecount--;
1862 		VOP_UNLOCK(vp, LK_INTERLOCK, td);
1863 		return;
1864 	}
1865 
1866 	if (vp->v_usecount == 1) {
1867 		vp->v_usecount--;
1868 		/*
1869 		 * We must call VOP_INACTIVE with the node locked.
1870 		 * If we are doing a vput, the node is already locked,
1871 		 * so we just need to release the vnode mutex.
1872 		 */
1873 		mtx_unlock(&vp->v_interlock);
1874 		VOP_INACTIVE(vp, td);
1875 		if (VSHOULDFREE(vp))
1876 			vfree(vp);
1877 		else
1878 			vlruvp(vp);
1879 
1880 	} else {
1881 #ifdef DIAGNOSTIC
1882 		vprint("vput: negative ref count", vp);
1883 #endif
1884 		panic("vput: negative ref cnt");
1885 	}
1886 }
1887 
1888 /*
1889  * Somebody doesn't want the vnode recycled.
1890  */
1891 void
1892 vhold(vp)
1893 	register struct vnode *vp;
1894 {
1895 	int s;
1896 
1897 	s = splbio();
1898 	vp->v_holdcnt++;
1899 	if (VSHOULDBUSY(vp))
1900 		vbusy(vp);
1901 	splx(s);
1902 }
1903 
1904 /*
1905  * Note that there is one less who cares about this vnode.  vdrop() is the
1906  * opposite of vhold().
1907  */
1908 void
1909 vdrop(vp)
1910 	register struct vnode *vp;
1911 {
1912 	int s;
1913 
1914 	s = splbio();
1915 	if (vp->v_holdcnt <= 0)
1916 		panic("vdrop: holdcnt");
1917 	vp->v_holdcnt--;
1918 	if (VSHOULDFREE(vp))
1919 		vfree(vp);
1920 	else
1921 		vlruvp(vp);
1922 	splx(s);
1923 }
1924 
1925 /*
1926  * Remove any vnodes in the vnode table belonging to mount point mp.
1927  *
1928  * If FORCECLOSE is not specified, there should not be any active ones,
1929  * return error if any are found (nb: this is a user error, not a
1930  * system error). If FORCECLOSE is specified, detach any active vnodes
1931  * that are found.
1932  *
1933  * If WRITECLOSE is set, only flush out regular file vnodes open for
1934  * writing.
1935  *
1936  * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
1937  *
1938  * `rootrefs' specifies the base reference count for the root vnode
1939  * of this filesystem. The root vnode is considered busy if its
1940  * v_usecount exceeds this value. On a successful return, vflush()
1941  * will call vrele() on the root vnode exactly rootrefs times.
1942  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
1943  * be zero.
1944  */
1945 #ifdef DIAGNOSTIC
1946 static int busyprt = 0;		/* print out busy vnodes */
1947 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1948 #endif
1949 
1950 int
1951 vflush(mp, rootrefs, flags)
1952 	struct mount *mp;
1953 	int rootrefs;
1954 	int flags;
1955 {
1956 	struct thread *td = curthread;	/* XXX */
1957 	struct vnode *vp, *nvp, *rootvp = NULL;
1958 	struct vattr vattr;
1959 	int busy = 0, error;
1960 
1961 	if (rootrefs > 0) {
1962 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
1963 		    ("vflush: bad args"));
1964 		/*
1965 		 * Get the filesystem root vnode. We can vput() it
1966 		 * immediately, since with rootrefs > 0, it won't go away.
1967 		 */
1968 		if ((error = VFS_ROOT(mp, &rootvp)) != 0)
1969 			return (error);
1970 		vput(rootvp);
1971 	}
1972 	mtx_lock(&mntvnode_mtx);
1973 loop:
1974 	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) {
1975 		/*
1976 		 * Make sure this vnode wasn't reclaimed in getnewvnode().
1977 		 * Start over if it has (it won't be on the list anymore).
1978 		 */
1979 		if (vp->v_mount != mp)
1980 			goto loop;
1981 		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
1982 
1983 		mtx_unlock(&mntvnode_mtx);
1984 		mtx_lock(&vp->v_interlock);
1985 		/*
1986 		 * Skip over a vnodes marked VSYSTEM.
1987 		 */
1988 		if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1989 			mtx_unlock(&vp->v_interlock);
1990 			mtx_lock(&mntvnode_mtx);
1991 			continue;
1992 		}
1993 		/*
1994 		 * If WRITECLOSE is set, flush out unlinked but still open
1995 		 * files (even if open only for reading) and regular file
1996 		 * vnodes open for writing.
1997 		 */
1998 		if ((flags & WRITECLOSE) &&
1999 		    (vp->v_type == VNON ||
2000 		    (VOP_GETATTR(vp, &vattr, td->td_ucred, td) == 0 &&
2001 		    vattr.va_nlink > 0)) &&
2002 		    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2003 			mtx_unlock(&vp->v_interlock);
2004 			mtx_lock(&mntvnode_mtx);
2005 			continue;
2006 		}
2007 
2008 		/*
2009 		 * With v_usecount == 0, all we need to do is clear out the
2010 		 * vnode data structures and we are done.
2011 		 */
2012 		if (vp->v_usecount == 0) {
2013 			vgonel(vp, td);
2014 			mtx_lock(&mntvnode_mtx);
2015 			continue;
2016 		}
2017 
2018 		/*
2019 		 * If FORCECLOSE is set, forcibly close the vnode. For block
2020 		 * or character devices, revert to an anonymous device. For
2021 		 * all other files, just kill them.
2022 		 */
2023 		if (flags & FORCECLOSE) {
2024 			if (vp->v_type != VCHR) {
2025 				vgonel(vp, td);
2026 			} else {
2027 				vclean(vp, 0, td);
2028 				vp->v_op = spec_vnodeop_p;
2029 				insmntque(vp, (struct mount *) 0);
2030 			}
2031 			mtx_lock(&mntvnode_mtx);
2032 			continue;
2033 		}
2034 #ifdef DIAGNOSTIC
2035 		if (busyprt)
2036 			vprint("vflush: busy vnode", vp);
2037 #endif
2038 		mtx_unlock(&vp->v_interlock);
2039 		mtx_lock(&mntvnode_mtx);
2040 		busy++;
2041 	}
2042 	mtx_unlock(&mntvnode_mtx);
2043 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2044 		/*
2045 		 * If just the root vnode is busy, and if its refcount
2046 		 * is equal to `rootrefs', then go ahead and kill it.
2047 		 */
2048 		mtx_lock(&rootvp->v_interlock);
2049 		KASSERT(busy > 0, ("vflush: not busy"));
2050 		KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
2051 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2052 			vgonel(rootvp, td);
2053 			busy = 0;
2054 		} else
2055 			mtx_unlock(&rootvp->v_interlock);
2056 	}
2057 	if (busy)
2058 		return (EBUSY);
2059 	for (; rootrefs > 0; rootrefs--)
2060 		vrele(rootvp);
2061 	return (0);
2062 }
2063 
2064 /*
2065  * This moves a now (likely recyclable) vnode to the end of the
2066  * mountlist.  XXX However, it is temporarily disabled until we
2067  * can clean up ffs_sync() and friends, which have loop restart
2068  * conditions which this code causes to operate O(N^2).
2069  */
2070 static void
2071 vlruvp(struct vnode *vp)
2072 {
2073 #if 0
2074 	struct mount *mp;
2075 
2076 	if ((mp = vp->v_mount) != NULL) {
2077 		mtx_lock(&mntvnode_mtx);
2078 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
2079 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
2080 		mtx_unlock(&mntvnode_mtx);
2081 	}
2082 #endif
2083 }
2084 
2085 /*
2086  * Disassociate the underlying filesystem from a vnode.
2087  */
2088 static void
2089 vclean(vp, flags, td)
2090 	struct vnode *vp;
2091 	int flags;
2092 	struct thread *td;
2093 {
2094 	int active;
2095 
2096 	/*
2097 	 * Check to see if the vnode is in use. If so we have to reference it
2098 	 * before we clean it out so that its count cannot fall to zero and
2099 	 * generate a race against ourselves to recycle it.
2100 	 */
2101 	if ((active = vp->v_usecount))
2102 		vp->v_usecount++;
2103 
2104 	/*
2105 	 * Prevent the vnode from being recycled or brought into use while we
2106 	 * clean it out.
2107 	 */
2108 	if (vp->v_flag & VXLOCK)
2109 		panic("vclean: deadlock");
2110 	vp->v_flag |= VXLOCK;
2111 	vp->v_vxproc = curthread;
2112 	/*
2113 	 * Even if the count is zero, the VOP_INACTIVE routine may still
2114 	 * have the object locked while it cleans it out. The VOP_LOCK
2115 	 * ensures that the VOP_INACTIVE routine is done with its work.
2116 	 * For active vnodes, it ensures that no other activity can
2117 	 * occur while the underlying object is being cleaned out.
2118 	 */
2119 	VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td);
2120 
2121 	/*
2122 	 * Clean out any buffers associated with the vnode.
2123 	 * If the flush fails, just toss the buffers.
2124 	 */
2125 	if (flags & DOCLOSE) {
2126 		if (TAILQ_FIRST(&vp->v_dirtyblkhd) != NULL)
2127 			(void) vn_write_suspend_wait(vp, NULL, V_WAIT);
2128 		if (vinvalbuf(vp, V_SAVE, NOCRED, td, 0, 0) != 0)
2129 			vinvalbuf(vp, 0, NOCRED, td, 0, 0);
2130 	}
2131 
2132 	VOP_DESTROYVOBJECT(vp);
2133 
2134 	/*
2135 	 * If purging an active vnode, it must be closed and
2136 	 * deactivated before being reclaimed. Note that the
2137 	 * VOP_INACTIVE will unlock the vnode.
2138 	 */
2139 	if (active) {
2140 		if (flags & DOCLOSE)
2141 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2142 		VOP_INACTIVE(vp, td);
2143 	} else {
2144 		/*
2145 		 * Any other processes trying to obtain this lock must first
2146 		 * wait for VXLOCK to clear, then call the new lock operation.
2147 		 */
2148 		VOP_UNLOCK(vp, 0, td);
2149 	}
2150 	/*
2151 	 * Reclaim the vnode.
2152 	 */
2153 	if (VOP_RECLAIM(vp, td))
2154 		panic("vclean: cannot reclaim");
2155 
2156 	if (active) {
2157 		/*
2158 		 * Inline copy of vrele() since VOP_INACTIVE
2159 		 * has already been called.
2160 		 */
2161 		mtx_lock(&vp->v_interlock);
2162 		if (--vp->v_usecount <= 0) {
2163 #ifdef DIAGNOSTIC
2164 			if (vp->v_usecount < 0 || vp->v_writecount != 0) {
2165 				vprint("vclean: bad ref count", vp);
2166 				panic("vclean: ref cnt");
2167 			}
2168 #endif
2169 			vfree(vp);
2170 		}
2171 		mtx_unlock(&vp->v_interlock);
2172 	}
2173 
2174 	cache_purge(vp);
2175 	vp->v_vnlock = NULL;
2176 	lockdestroy(&vp->v_lock);
2177 
2178 	if (VSHOULDFREE(vp))
2179 		vfree(vp);
2180 
2181 	/*
2182 	 * Done with purge, notify sleepers of the grim news.
2183 	 */
2184 	vp->v_op = dead_vnodeop_p;
2185 	if (vp->v_pollinfo != NULL)
2186 		vn_pollgone(vp);
2187 	vp->v_tag = VT_NON;
2188 	vp->v_flag &= ~VXLOCK;
2189 	vp->v_vxproc = NULL;
2190 	if (vp->v_flag & VXWANT) {
2191 		vp->v_flag &= ~VXWANT;
2192 		wakeup((caddr_t) vp);
2193 	}
2194 }
2195 
2196 /*
2197  * Eliminate all activity associated with the requested vnode
2198  * and with all vnodes aliased to the requested vnode.
2199  */
2200 int
2201 vop_revoke(ap)
2202 	struct vop_revoke_args /* {
2203 		struct vnode *a_vp;
2204 		int a_flags;
2205 	} */ *ap;
2206 {
2207 	struct vnode *vp, *vq;
2208 	dev_t dev;
2209 
2210 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
2211 
2212 	vp = ap->a_vp;
2213 	/*
2214 	 * If a vgone (or vclean) is already in progress,
2215 	 * wait until it is done and return.
2216 	 */
2217 	if (vp->v_flag & VXLOCK) {
2218 		vp->v_flag |= VXWANT;
2219 		msleep((caddr_t)vp, &vp->v_interlock, PINOD | PDROP,
2220 		    "vop_revokeall", 0);
2221 		return (0);
2222 	}
2223 	dev = vp->v_rdev;
2224 	for (;;) {
2225 		mtx_lock(&spechash_mtx);
2226 		vq = SLIST_FIRST(&dev->si_hlist);
2227 		mtx_unlock(&spechash_mtx);
2228 		if (!vq)
2229 			break;
2230 		vgone(vq);
2231 	}
2232 	return (0);
2233 }
2234 
2235 /*
2236  * Recycle an unused vnode to the front of the free list.
2237  * Release the passed interlock if the vnode will be recycled.
2238  */
2239 int
2240 vrecycle(vp, inter_lkp, td)
2241 	struct vnode *vp;
2242 	struct mtx *inter_lkp;
2243 	struct thread *td;
2244 {
2245 
2246 	mtx_lock(&vp->v_interlock);
2247 	if (vp->v_usecount == 0) {
2248 		if (inter_lkp) {
2249 			mtx_unlock(inter_lkp);
2250 		}
2251 		vgonel(vp, td);
2252 		return (1);
2253 	}
2254 	mtx_unlock(&vp->v_interlock);
2255 	return (0);
2256 }
2257 
2258 /*
2259  * Eliminate all activity associated with a vnode
2260  * in preparation for reuse.
2261  */
2262 void
2263 vgone(vp)
2264 	register struct vnode *vp;
2265 {
2266 	struct thread *td = curthread;	/* XXX */
2267 
2268 	mtx_lock(&vp->v_interlock);
2269 	vgonel(vp, td);
2270 }
2271 
2272 /*
2273  * vgone, with the vp interlock held.
2274  */
2275 void
2276 vgonel(vp, td)
2277 	struct vnode *vp;
2278 	struct thread *td;
2279 {
2280 	int s;
2281 
2282 	/*
2283 	 * If a vgone (or vclean) is already in progress,
2284 	 * wait until it is done and return.
2285 	 */
2286 	if (vp->v_flag & VXLOCK) {
2287 		vp->v_flag |= VXWANT;
2288 		msleep((caddr_t)vp, &vp->v_interlock, PINOD | PDROP,
2289 		    "vgone", 0);
2290 		return;
2291 	}
2292 
2293 	/*
2294 	 * Clean out the filesystem specific data.
2295 	 */
2296 	vclean(vp, DOCLOSE, td);
2297 	mtx_lock(&vp->v_interlock);
2298 
2299 	/*
2300 	 * Delete from old mount point vnode list, if on one.
2301 	 */
2302 	if (vp->v_mount != NULL)
2303 		insmntque(vp, (struct mount *)0);
2304 	/*
2305 	 * If special device, remove it from special device alias list
2306 	 * if it is on one.
2307 	 */
2308 	if (vp->v_type == VCHR && vp->v_rdev != NULL && vp->v_rdev != NODEV) {
2309 		mtx_lock(&spechash_mtx);
2310 		SLIST_REMOVE(&vp->v_rdev->si_hlist, vp, vnode, v_specnext);
2311 		freedev(vp->v_rdev);
2312 		mtx_unlock(&spechash_mtx);
2313 		vp->v_rdev = NULL;
2314 	}
2315 
2316 	/*
2317 	 * If it is on the freelist and not already at the head,
2318 	 * move it to the head of the list. The test of the
2319 	 * VDOOMED flag and the reference count of zero is because
2320 	 * it will be removed from the free list by getnewvnode,
2321 	 * but will not have its reference count incremented until
2322 	 * after calling vgone. If the reference count were
2323 	 * incremented first, vgone would (incorrectly) try to
2324 	 * close the previous instance of the underlying object.
2325 	 */
2326 	if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
2327 		s = splbio();
2328 		mtx_lock(&vnode_free_list_mtx);
2329 		if (vp->v_flag & VFREE)
2330 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2331 		else
2332 			freevnodes++;
2333 		vp->v_flag |= VFREE;
2334 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2335 		mtx_unlock(&vnode_free_list_mtx);
2336 		splx(s);
2337 	}
2338 
2339 	vp->v_type = VBAD;
2340 	mtx_unlock(&vp->v_interlock);
2341 }
2342 
2343 /*
2344  * Lookup a vnode by device number.
2345  */
2346 int
2347 vfinddev(dev, type, vpp)
2348 	dev_t dev;
2349 	enum vtype type;
2350 	struct vnode **vpp;
2351 {
2352 	struct vnode *vp;
2353 
2354 	mtx_lock(&spechash_mtx);
2355 	SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2356 		if (type == vp->v_type) {
2357 			*vpp = vp;
2358 			mtx_unlock(&spechash_mtx);
2359 			return (1);
2360 		}
2361 	}
2362 	mtx_unlock(&spechash_mtx);
2363 	return (0);
2364 }
2365 
2366 /*
2367  * Calculate the total number of references to a special device.
2368  */
2369 int
2370 vcount(vp)
2371 	struct vnode *vp;
2372 {
2373 	struct vnode *vq;
2374 	int count;
2375 
2376 	count = 0;
2377 	mtx_lock(&spechash_mtx);
2378 	SLIST_FOREACH(vq, &vp->v_rdev->si_hlist, v_specnext)
2379 		count += vq->v_usecount;
2380 	mtx_unlock(&spechash_mtx);
2381 	return (count);
2382 }
2383 
2384 /*
2385  * Same as above, but using the dev_t as argument
2386  */
2387 int
2388 count_dev(dev)
2389 	dev_t dev;
2390 {
2391 	struct vnode *vp;
2392 
2393 	vp = SLIST_FIRST(&dev->si_hlist);
2394 	if (vp == NULL)
2395 		return (0);
2396 	return(vcount(vp));
2397 }
2398 
2399 /*
2400  * Print out a description of a vnode.
2401  */
2402 static char *typename[] =
2403 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2404 
2405 void
2406 vprint(label, vp)
2407 	char *label;
2408 	struct vnode *vp;
2409 {
2410 	char buf[96];
2411 
2412 	if (label != NULL)
2413 		printf("%s: %p: ", label, (void *)vp);
2414 	else
2415 		printf("%p: ", (void *)vp);
2416 	printf("type %s, usecount %d, writecount %d, refcount %d,",
2417 	    typename[vp->v_type], vp->v_usecount, vp->v_writecount,
2418 	    vp->v_holdcnt);
2419 	buf[0] = '\0';
2420 	if (vp->v_flag & VROOT)
2421 		strcat(buf, "|VROOT");
2422 	if (vp->v_flag & VTEXT)
2423 		strcat(buf, "|VTEXT");
2424 	if (vp->v_flag & VSYSTEM)
2425 		strcat(buf, "|VSYSTEM");
2426 	if (vp->v_flag & VXLOCK)
2427 		strcat(buf, "|VXLOCK");
2428 	if (vp->v_flag & VXWANT)
2429 		strcat(buf, "|VXWANT");
2430 	if (vp->v_flag & VBWAIT)
2431 		strcat(buf, "|VBWAIT");
2432 	if (vp->v_flag & VDOOMED)
2433 		strcat(buf, "|VDOOMED");
2434 	if (vp->v_flag & VFREE)
2435 		strcat(buf, "|VFREE");
2436 	if (vp->v_flag & VOBJBUF)
2437 		strcat(buf, "|VOBJBUF");
2438 	if (buf[0] != '\0')
2439 		printf(" flags (%s)", &buf[1]);
2440 	if (vp->v_data == NULL) {
2441 		printf("\n");
2442 	} else {
2443 		printf("\n\t");
2444 		VOP_PRINT(vp);
2445 	}
2446 }
2447 
2448 #ifdef DDB
2449 #include <ddb/ddb.h>
2450 /*
2451  * List all of the locked vnodes in the system.
2452  * Called when debugging the kernel.
2453  */
2454 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
2455 {
2456 	struct thread *td = curthread;	/* XXX */
2457 	struct mount *mp, *nmp;
2458 	struct vnode *vp;
2459 
2460 	printf("Locked vnodes\n");
2461 	mtx_lock(&mountlist_mtx);
2462 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2463 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
2464 			nmp = TAILQ_NEXT(mp, mnt_list);
2465 			continue;
2466 		}
2467 		mtx_lock(&mntvnode_mtx);
2468 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2469 			if (VOP_ISLOCKED(vp, NULL))
2470 				vprint((char *)0, vp);
2471 		}
2472 		mtx_unlock(&mntvnode_mtx);
2473 		mtx_lock(&mountlist_mtx);
2474 		nmp = TAILQ_NEXT(mp, mnt_list);
2475 		vfs_unbusy(mp, td);
2476 	}
2477 	mtx_unlock(&mountlist_mtx);
2478 }
2479 #endif
2480 
2481 /*
2482  * Top level filesystem related information gathering.
2483  */
2484 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2485 
2486 static int
2487 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2488 {
2489 	int *name = (int *)arg1 - 1;	/* XXX */
2490 	u_int namelen = arg2 + 1;	/* XXX */
2491 	struct vfsconf *vfsp;
2492 
2493 #if 1 || defined(COMPAT_PRELITE2)
2494 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2495 	if (namelen == 1)
2496 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2497 #endif
2498 
2499 	/* XXX the below code does not compile; vfs_sysctl does not exist. */
2500 #ifdef notyet
2501 	/* all sysctl names at this level are at least name and field */
2502 	if (namelen < 2)
2503 		return (ENOTDIR);		/* overloaded */
2504 	if (name[0] != VFS_GENERIC) {
2505 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2506 			if (vfsp->vfc_typenum == name[0])
2507 				break;
2508 		if (vfsp == NULL)
2509 			return (EOPNOTSUPP);
2510 		return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2511 		    oldp, oldlenp, newp, newlen, td));
2512 	}
2513 #endif
2514 	switch (name[1]) {
2515 	case VFS_MAXTYPENUM:
2516 		if (namelen != 2)
2517 			return (ENOTDIR);
2518 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2519 	case VFS_CONF:
2520 		if (namelen != 3)
2521 			return (ENOTDIR);	/* overloaded */
2522 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2523 			if (vfsp->vfc_typenum == name[2])
2524 				break;
2525 		if (vfsp == NULL)
2526 			return (EOPNOTSUPP);
2527 		return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2528 	}
2529 	return (EOPNOTSUPP);
2530 }
2531 
2532 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2533 	"Generic filesystem");
2534 
2535 #if 1 || defined(COMPAT_PRELITE2)
2536 
2537 static int
2538 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2539 {
2540 	int error;
2541 	struct vfsconf *vfsp;
2542 	struct ovfsconf ovfs;
2543 
2544 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2545 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
2546 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
2547 		ovfs.vfc_index = vfsp->vfc_typenum;
2548 		ovfs.vfc_refcount = vfsp->vfc_refcount;
2549 		ovfs.vfc_flags = vfsp->vfc_flags;
2550 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2551 		if (error)
2552 			return error;
2553 	}
2554 	return 0;
2555 }
2556 
2557 #endif /* 1 || COMPAT_PRELITE2 */
2558 
2559 #if COMPILING_LINT
2560 #define KINFO_VNODESLOP	10
2561 /*
2562  * Dump vnode list (via sysctl).
2563  * Copyout address of vnode followed by vnode.
2564  */
2565 /* ARGSUSED */
2566 static int
2567 sysctl_vnode(SYSCTL_HANDLER_ARGS)
2568 {
2569 	struct thread *td = curthread;	/* XXX */
2570 	struct mount *mp, *nmp;
2571 	struct vnode *nvp, *vp;
2572 	int error;
2573 
2574 #define VPTRSZ	sizeof (struct vnode *)
2575 #define VNODESZ	sizeof (struct vnode)
2576 
2577 	req->lock = 0;
2578 	if (!req->oldptr) /* Make an estimate */
2579 		return (SYSCTL_OUT(req, 0,
2580 			(numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2581 
2582 	mtx_lock(&mountlist_mtx);
2583 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2584 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
2585 			nmp = TAILQ_NEXT(mp, mnt_list);
2586 			continue;
2587 		}
2588 		mtx_lock(&mntvnode_mtx);
2589 again:
2590 		for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
2591 		     vp != NULL;
2592 		     vp = nvp) {
2593 			/*
2594 			 * Check that the vp is still associated with
2595 			 * this filesystem.  RACE: could have been
2596 			 * recycled onto the same filesystem.
2597 			 */
2598 			if (vp->v_mount != mp)
2599 				goto again;
2600 			nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2601 			mtx_unlock(&mntvnode_mtx);
2602 			if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2603 			    (error = SYSCTL_OUT(req, vp, VNODESZ)))
2604 				return (error);
2605 			mtx_lock(&mntvnode_mtx);
2606 		}
2607 		mtx_unlock(&mntvnode_mtx);
2608 		mtx_lock(&mountlist_mtx);
2609 		nmp = TAILQ_NEXT(mp, mnt_list);
2610 		vfs_unbusy(mp, td);
2611 	}
2612 	mtx_unlock(&mountlist_mtx);
2613 
2614 	return (0);
2615 }
2616 
2617 /*
2618  * XXX
2619  * Exporting the vnode list on large systems causes them to crash.
2620  * Exporting the vnode list on medium systems causes sysctl to coredump.
2621  */
2622 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2623 	0, 0, sysctl_vnode, "S,vnode", "");
2624 #endif
2625 
2626 /*
2627  * Check to see if a filesystem is mounted on a block device.
2628  */
2629 int
2630 vfs_mountedon(vp)
2631 	struct vnode *vp;
2632 {
2633 
2634 	if (vp->v_rdev->si_mountpoint != NULL)
2635 		return (EBUSY);
2636 	return (0);
2637 }
2638 
2639 /*
2640  * Unmount all filesystems. The list is traversed in reverse order
2641  * of mounting to avoid dependencies.
2642  */
2643 void
2644 vfs_unmountall()
2645 {
2646 	struct mount *mp;
2647 	struct thread *td;
2648 	int error;
2649 
2650 	if (curthread != NULL)
2651 		td = curthread;
2652 	else
2653 		td = FIRST_THREAD_IN_PROC(initproc); /* XXX XXX proc0? */
2654 	/*
2655 	 * Since this only runs when rebooting, it is not interlocked.
2656 	 */
2657 	while(!TAILQ_EMPTY(&mountlist)) {
2658 		mp = TAILQ_LAST(&mountlist, mntlist);
2659 		error = dounmount(mp, MNT_FORCE, td);
2660 		if (error) {
2661 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
2662 			printf("unmount of %s failed (",
2663 			    mp->mnt_stat.f_mntonname);
2664 			if (error == EBUSY)
2665 				printf("BUSY)\n");
2666 			else
2667 				printf("%d)\n", error);
2668 		} else {
2669 			/* The unmount has removed mp from the mountlist */
2670 		}
2671 	}
2672 }
2673 
2674 /*
2675  * perform msync on all vnodes under a mount point
2676  * the mount point must be locked.
2677  */
2678 void
2679 vfs_msync(struct mount *mp, int flags)
2680 {
2681 	struct vnode *vp, *nvp;
2682 	struct vm_object *obj;
2683 	int tries;
2684 
2685 	GIANT_REQUIRED;
2686 
2687 	tries = 5;
2688 	mtx_lock(&mntvnode_mtx);
2689 loop:
2690 	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) {
2691 		if (vp->v_mount != mp) {
2692 			if (--tries > 0)
2693 				goto loop;
2694 			break;
2695 		}
2696 		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2697 
2698 		if (vp->v_flag & VXLOCK)	/* XXX: what if MNT_WAIT? */
2699 			continue;
2700 
2701 		if (vp->v_flag & VNOSYNC)	/* unlinked, skip it */
2702 			continue;
2703 
2704 		if ((vp->v_flag & VOBJDIRTY) &&
2705 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
2706 			mtx_unlock(&mntvnode_mtx);
2707 			if (!vget(vp,
2708 			    LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, curthread)) {
2709 				if (VOP_GETVOBJECT(vp, &obj) == 0) {
2710 					vm_object_page_clean(obj, 0, 0,
2711 					    flags == MNT_WAIT ?
2712 					    OBJPC_SYNC : OBJPC_NOSYNC);
2713 				}
2714 				vput(vp);
2715 			}
2716 			mtx_lock(&mntvnode_mtx);
2717 			if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) {
2718 				if (--tries > 0)
2719 					goto loop;
2720 				break;
2721 			}
2722 		}
2723 	}
2724 	mtx_unlock(&mntvnode_mtx);
2725 }
2726 
2727 /*
2728  * Create the VM object needed for VMIO and mmap support.  This
2729  * is done for all VREG files in the system.  Some filesystems might
2730  * afford the additional metadata buffering capability of the
2731  * VMIO code by making the device node be VMIO mode also.
2732  *
2733  * vp must be locked when vfs_object_create is called.
2734  */
2735 int
2736 vfs_object_create(vp, td, cred)
2737 	struct vnode *vp;
2738 	struct thread *td;
2739 	struct ucred *cred;
2740 {
2741 	GIANT_REQUIRED;
2742 	return (VOP_CREATEVOBJECT(vp, cred, td));
2743 }
2744 
2745 /*
2746  * Mark a vnode as free, putting it up for recycling.
2747  */
2748 void
2749 vfree(vp)
2750 	struct vnode *vp;
2751 {
2752 	int s;
2753 
2754 	s = splbio();
2755 	mtx_lock(&vnode_free_list_mtx);
2756 	KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free"));
2757 	if (vp->v_flag & VAGE) {
2758 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2759 	} else {
2760 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2761 	}
2762 	freevnodes++;
2763 	mtx_unlock(&vnode_free_list_mtx);
2764 	vp->v_flag &= ~VAGE;
2765 	vp->v_flag |= VFREE;
2766 	splx(s);
2767 }
2768 
2769 /*
2770  * Opposite of vfree() - mark a vnode as in use.
2771  */
2772 void
2773 vbusy(vp)
2774 	struct vnode *vp;
2775 {
2776 	int s;
2777 
2778 	s = splbio();
2779 	mtx_lock(&vnode_free_list_mtx);
2780 	KASSERT((vp->v_flag & VFREE) != 0, ("vnode not free"));
2781 	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2782 	freevnodes--;
2783 	mtx_unlock(&vnode_free_list_mtx);
2784 	vp->v_flag &= ~(VFREE|VAGE);
2785 	splx(s);
2786 }
2787 
2788 /*
2789  * Record a process's interest in events which might happen to
2790  * a vnode.  Because poll uses the historic select-style interface
2791  * internally, this routine serves as both the ``check for any
2792  * pending events'' and the ``record my interest in future events''
2793  * functions.  (These are done together, while the lock is held,
2794  * to avoid race conditions.)
2795  */
2796 int
2797 vn_pollrecord(vp, td, events)
2798 	struct vnode *vp;
2799 	struct thread *td;
2800 	short events;
2801 {
2802 
2803 	if (vp->v_pollinfo == NULL)
2804 		v_addpollinfo(vp);
2805 	mtx_lock(&vp->v_pollinfo->vpi_lock);
2806 	if (vp->v_pollinfo->vpi_revents & events) {
2807 		/*
2808 		 * This leaves events we are not interested
2809 		 * in available for the other process which
2810 		 * which presumably had requested them
2811 		 * (otherwise they would never have been
2812 		 * recorded).
2813 		 */
2814 		events &= vp->v_pollinfo->vpi_revents;
2815 		vp->v_pollinfo->vpi_revents &= ~events;
2816 
2817 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
2818 		return events;
2819 	}
2820 	vp->v_pollinfo->vpi_events |= events;
2821 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
2822 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
2823 	return 0;
2824 }
2825 
2826 /*
2827  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
2828  * it is possible for us to miss an event due to race conditions, but
2829  * that condition is expected to be rare, so for the moment it is the
2830  * preferred interface.
2831  */
2832 void
2833 vn_pollevent(vp, events)
2834 	struct vnode *vp;
2835 	short events;
2836 {
2837 
2838 	if (vp->v_pollinfo == NULL)
2839 		v_addpollinfo(vp);
2840 	mtx_lock(&vp->v_pollinfo->vpi_lock);
2841 	if (vp->v_pollinfo->vpi_events & events) {
2842 		/*
2843 		 * We clear vpi_events so that we don't
2844 		 * call selwakeup() twice if two events are
2845 		 * posted before the polling process(es) is
2846 		 * awakened.  This also ensures that we take at
2847 		 * most one selwakeup() if the polling process
2848 		 * is no longer interested.  However, it does
2849 		 * mean that only one event can be noticed at
2850 		 * a time.  (Perhaps we should only clear those
2851 		 * event bits which we note?) XXX
2852 		 */
2853 		vp->v_pollinfo->vpi_events = 0;	/* &= ~events ??? */
2854 		vp->v_pollinfo->vpi_revents |= events;
2855 		selwakeup(&vp->v_pollinfo->vpi_selinfo);
2856 	}
2857 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
2858 }
2859 
2860 /*
2861  * Wake up anyone polling on vp because it is being revoked.
2862  * This depends on dead_poll() returning POLLHUP for correct
2863  * behavior.
2864  */
2865 void
2866 vn_pollgone(vp)
2867 	struct vnode *vp;
2868 {
2869 
2870 	mtx_lock(&vp->v_pollinfo->vpi_lock);
2871 	VN_KNOTE(vp, NOTE_REVOKE);
2872 	if (vp->v_pollinfo->vpi_events) {
2873 		vp->v_pollinfo->vpi_events = 0;
2874 		selwakeup(&vp->v_pollinfo->vpi_selinfo);
2875 	}
2876 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
2877 }
2878 
2879 
2880 
2881 /*
2882  * Routine to create and manage a filesystem syncer vnode.
2883  */
2884 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
2885 static int	sync_fsync(struct  vop_fsync_args *);
2886 static int	sync_inactive(struct  vop_inactive_args *);
2887 static int	sync_reclaim(struct  vop_reclaim_args *);
2888 #define sync_lock ((int (*)(struct  vop_lock_args *))vop_nolock)
2889 #define sync_unlock ((int (*)(struct  vop_unlock_args *))vop_nounlock)
2890 static int	sync_print(struct vop_print_args *);
2891 #define sync_islocked ((int(*)(struct vop_islocked_args *))vop_noislocked)
2892 
2893 static vop_t **sync_vnodeop_p;
2894 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
2895 	{ &vop_default_desc,	(vop_t *) vop_eopnotsupp },
2896 	{ &vop_close_desc,	(vop_t *) sync_close },		/* close */
2897 	{ &vop_fsync_desc,	(vop_t *) sync_fsync },		/* fsync */
2898 	{ &vop_inactive_desc,	(vop_t *) sync_inactive },	/* inactive */
2899 	{ &vop_reclaim_desc,	(vop_t *) sync_reclaim },	/* reclaim */
2900 	{ &vop_lock_desc,	(vop_t *) sync_lock },		/* lock */
2901 	{ &vop_unlock_desc,	(vop_t *) sync_unlock },	/* unlock */
2902 	{ &vop_print_desc,	(vop_t *) sync_print },		/* print */
2903 	{ &vop_islocked_desc,	(vop_t *) sync_islocked },	/* islocked */
2904 	{ NULL, NULL }
2905 };
2906 static struct vnodeopv_desc sync_vnodeop_opv_desc =
2907 	{ &sync_vnodeop_p, sync_vnodeop_entries };
2908 
2909 VNODEOP_SET(sync_vnodeop_opv_desc);
2910 
2911 /*
2912  * Create a new filesystem syncer vnode for the specified mount point.
2913  */
2914 int
2915 vfs_allocate_syncvnode(mp)
2916 	struct mount *mp;
2917 {
2918 	struct vnode *vp;
2919 	static long start, incr, next;
2920 	int error;
2921 
2922 	/* Allocate a new vnode */
2923 	if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
2924 		mp->mnt_syncer = NULL;
2925 		return (error);
2926 	}
2927 	vp->v_type = VNON;
2928 	/*
2929 	 * Place the vnode onto the syncer worklist. We attempt to
2930 	 * scatter them about on the list so that they will go off
2931 	 * at evenly distributed times even if all the filesystems
2932 	 * are mounted at once.
2933 	 */
2934 	next += incr;
2935 	if (next == 0 || next > syncer_maxdelay) {
2936 		start /= 2;
2937 		incr /= 2;
2938 		if (start == 0) {
2939 			start = syncer_maxdelay / 2;
2940 			incr = syncer_maxdelay;
2941 		}
2942 		next = start;
2943 	}
2944 	vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
2945 	mp->mnt_syncer = vp;
2946 	return (0);
2947 }
2948 
2949 /*
2950  * Do a lazy sync of the filesystem.
2951  */
2952 static int
2953 sync_fsync(ap)
2954 	struct vop_fsync_args /* {
2955 		struct vnode *a_vp;
2956 		struct ucred *a_cred;
2957 		int a_waitfor;
2958 		struct thread *a_td;
2959 	} */ *ap;
2960 {
2961 	struct vnode *syncvp = ap->a_vp;
2962 	struct mount *mp = syncvp->v_mount;
2963 	struct thread *td = ap->a_td;
2964 	int asyncflag;
2965 
2966 	/*
2967 	 * We only need to do something if this is a lazy evaluation.
2968 	 */
2969 	if (ap->a_waitfor != MNT_LAZY)
2970 		return (0);
2971 
2972 	/*
2973 	 * Move ourselves to the back of the sync list.
2974 	 */
2975 	vn_syncer_add_to_worklist(syncvp, syncdelay);
2976 
2977 	/*
2978 	 * Walk the list of vnodes pushing all that are dirty and
2979 	 * not already on the sync list.
2980 	 */
2981 	mtx_lock(&mountlist_mtx);
2982 	if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) {
2983 		mtx_unlock(&mountlist_mtx);
2984 		return (0);
2985 	}
2986 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
2987 		vfs_unbusy(mp, td);
2988 		return (0);
2989 	}
2990 	asyncflag = mp->mnt_flag & MNT_ASYNC;
2991 	mp->mnt_flag &= ~MNT_ASYNC;
2992 	vfs_msync(mp, MNT_NOWAIT);
2993 	VFS_SYNC(mp, MNT_LAZY, ap->a_cred, td);
2994 	if (asyncflag)
2995 		mp->mnt_flag |= MNT_ASYNC;
2996 	vn_finished_write(mp);
2997 	vfs_unbusy(mp, td);
2998 	return (0);
2999 }
3000 
3001 /*
3002  * The syncer vnode is no referenced.
3003  */
3004 static int
3005 sync_inactive(ap)
3006 	struct vop_inactive_args /* {
3007 		struct vnode *a_vp;
3008 		struct thread *a_td;
3009 	} */ *ap;
3010 {
3011 
3012 	vgone(ap->a_vp);
3013 	return (0);
3014 }
3015 
3016 /*
3017  * The syncer vnode is no longer needed and is being decommissioned.
3018  *
3019  * Modifications to the worklist must be protected at splbio().
3020  */
3021 static int
3022 sync_reclaim(ap)
3023 	struct vop_reclaim_args /* {
3024 		struct vnode *a_vp;
3025 	} */ *ap;
3026 {
3027 	struct vnode *vp = ap->a_vp;
3028 	int s;
3029 
3030 	s = splbio();
3031 	vp->v_mount->mnt_syncer = NULL;
3032 	if (vp->v_flag & VONWORKLST) {
3033 		LIST_REMOVE(vp, v_synclist);
3034 		vp->v_flag &= ~VONWORKLST;
3035 	}
3036 	splx(s);
3037 
3038 	return (0);
3039 }
3040 
3041 /*
3042  * Print out a syncer vnode.
3043  */
3044 static int
3045 sync_print(ap)
3046 	struct vop_print_args /* {
3047 		struct vnode *a_vp;
3048 	} */ *ap;
3049 {
3050 	struct vnode *vp = ap->a_vp;
3051 
3052 	printf("syncer vnode");
3053 	if (vp->v_vnlock != NULL)
3054 		lockmgr_printinfo(vp->v_vnlock);
3055 	printf("\n");
3056 	return (0);
3057 }
3058 
3059 /*
3060  * extract the dev_t from a VCHR
3061  */
3062 dev_t
3063 vn_todev(vp)
3064 	struct vnode *vp;
3065 {
3066 	if (vp->v_type != VCHR)
3067 		return (NODEV);
3068 	return (vp->v_rdev);
3069 }
3070 
3071 /*
3072  * Check if vnode represents a disk device
3073  */
3074 int
3075 vn_isdisk(vp, errp)
3076 	struct vnode *vp;
3077 	int *errp;
3078 {
3079 	struct cdevsw *cdevsw;
3080 
3081 	if (vp->v_type != VCHR) {
3082 		if (errp != NULL)
3083 			*errp = ENOTBLK;
3084 		return (0);
3085 	}
3086 	if (vp->v_rdev == NULL) {
3087 		if (errp != NULL)
3088 			*errp = ENXIO;
3089 		return (0);
3090 	}
3091 	cdevsw = devsw(vp->v_rdev);
3092 	if (cdevsw == NULL) {
3093 		if (errp != NULL)
3094 			*errp = ENXIO;
3095 		return (0);
3096 	}
3097 	if (!(cdevsw->d_flags & D_DISK)) {
3098 		if (errp != NULL)
3099 			*errp = ENOTBLK;
3100 		return (0);
3101 	}
3102 	if (errp != NULL)
3103 		*errp = 0;
3104 	return (1);
3105 }
3106 
3107 /*
3108  * Free data allocated by namei(); see namei(9) for details.
3109  */
3110 void
3111 NDFREE(ndp, flags)
3112      struct nameidata *ndp;
3113      const uint flags;
3114 {
3115 	if (!(flags & NDF_NO_FREE_PNBUF) &&
3116 	    (ndp->ni_cnd.cn_flags & HASBUF)) {
3117 		uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
3118 		ndp->ni_cnd.cn_flags &= ~HASBUF;
3119 	}
3120 	if (!(flags & NDF_NO_DVP_UNLOCK) &&
3121 	    (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
3122 	    ndp->ni_dvp != ndp->ni_vp)
3123 		VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_thread);
3124 	if (!(flags & NDF_NO_DVP_RELE) &&
3125 	    (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
3126 		vrele(ndp->ni_dvp);
3127 		ndp->ni_dvp = NULL;
3128 	}
3129 	if (!(flags & NDF_NO_VP_UNLOCK) &&
3130 	    (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
3131 		VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_thread);
3132 	if (!(flags & NDF_NO_VP_RELE) &&
3133 	    ndp->ni_vp) {
3134 		vrele(ndp->ni_vp);
3135 		ndp->ni_vp = NULL;
3136 	}
3137 	if (!(flags & NDF_NO_STARTDIR_RELE) &&
3138 	    (ndp->ni_cnd.cn_flags & SAVESTART)) {
3139 		vrele(ndp->ni_startdir);
3140 		ndp->ni_startdir = NULL;
3141 	}
3142 }
3143 
3144 /*
3145  * Common filesystem object access control check routine.  Accepts a
3146  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3147  * and optional call-by-reference privused argument allowing vaccess()
3148  * to indicate to the caller whether privilege was used to satisfy the
3149  * request.  Returns 0 on success, or an errno on failure.
3150  */
3151 int
3152 vaccess(type, file_mode, file_uid, file_gid, acc_mode, cred, privused)
3153 	enum vtype type;
3154 	mode_t file_mode;
3155 	uid_t file_uid;
3156 	gid_t file_gid;
3157 	mode_t acc_mode;
3158 	struct ucred *cred;
3159 	int *privused;
3160 {
3161 	mode_t dac_granted;
3162 #ifdef CAPABILITIES
3163 	mode_t cap_granted;
3164 #endif
3165 
3166 	/*
3167 	 * Look for a normal, non-privileged way to access the file/directory
3168 	 * as requested.  If it exists, go with that.
3169 	 */
3170 
3171 	if (privused != NULL)
3172 		*privused = 0;
3173 
3174 	dac_granted = 0;
3175 
3176 	/* Check the owner. */
3177 	if (cred->cr_uid == file_uid) {
3178 		dac_granted |= VADMIN;
3179 		if (file_mode & S_IXUSR)
3180 			dac_granted |= VEXEC;
3181 		if (file_mode & S_IRUSR)
3182 			dac_granted |= VREAD;
3183 		if (file_mode & S_IWUSR)
3184 			dac_granted |= VWRITE;
3185 
3186 		if ((acc_mode & dac_granted) == acc_mode)
3187 			return (0);
3188 
3189 		goto privcheck;
3190 	}
3191 
3192 	/* Otherwise, check the groups (first match) */
3193 	if (groupmember(file_gid, cred)) {
3194 		if (file_mode & S_IXGRP)
3195 			dac_granted |= VEXEC;
3196 		if (file_mode & S_IRGRP)
3197 			dac_granted |= VREAD;
3198 		if (file_mode & S_IWGRP)
3199 			dac_granted |= VWRITE;
3200 
3201 		if ((acc_mode & dac_granted) == acc_mode)
3202 			return (0);
3203 
3204 		goto privcheck;
3205 	}
3206 
3207 	/* Otherwise, check everyone else. */
3208 	if (file_mode & S_IXOTH)
3209 		dac_granted |= VEXEC;
3210 	if (file_mode & S_IROTH)
3211 		dac_granted |= VREAD;
3212 	if (file_mode & S_IWOTH)
3213 		dac_granted |= VWRITE;
3214 	if ((acc_mode & dac_granted) == acc_mode)
3215 		return (0);
3216 
3217 privcheck:
3218 	if (!suser_cred(cred, PRISON_ROOT)) {
3219 		/* XXX audit: privilege used */
3220 		if (privused != NULL)
3221 			*privused = 1;
3222 		return (0);
3223 	}
3224 
3225 #ifdef CAPABILITIES
3226 	/*
3227 	 * Build a capability mask to determine if the set of capabilities
3228 	 * satisfies the requirements when combined with the granted mask
3229 	 * from above.
3230 	 * For each capability, if the capability is required, bitwise
3231 	 * or the request type onto the cap_granted mask.
3232 	 */
3233 	cap_granted = 0;
3234 
3235 	if (type == VDIR) {
3236 		/*
3237 		 * For directories, use CAP_DAC_READ_SEARCH to satisfy
3238 		 * VEXEC requests, instead of CAP_DAC_EXECUTE.
3239 		 */
3240 		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3241 		    !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3242 			cap_granted |= VEXEC;
3243 	} else {
3244 		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3245 		    !cap_check(cred, NULL, CAP_DAC_EXECUTE, PRISON_ROOT))
3246 			cap_granted |= VEXEC;
3247 	}
3248 
3249 	if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
3250 	    !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3251 		cap_granted |= VREAD;
3252 
3253 	if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3254 	    !cap_check(cred, NULL, CAP_DAC_WRITE, PRISON_ROOT))
3255 		cap_granted |= VWRITE;
3256 
3257 	if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3258 	    !cap_check(cred, NULL, CAP_FOWNER, PRISON_ROOT))
3259 		cap_granted |= VADMIN;
3260 
3261 	if ((acc_mode & (cap_granted | dac_granted)) == acc_mode) {
3262 		/* XXX audit: privilege used */
3263 		if (privused != NULL)
3264 			*privused = 1;
3265 		return (0);
3266 	}
3267 #endif
3268 
3269 	return ((acc_mode & VADMIN) ? EPERM : EACCES);
3270 }
3271