xref: /freebsd/sys/kern/vfs_subr.c (revision 5965373e69ca208cedd8ede9716ef0d58e6c3434)
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_mac.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/bio.h>
51 #include <sys/buf.h>
52 #include <sys/conf.h>
53 #include <sys/eventhandler.h>
54 #include <sys/fcntl.h>
55 #include <sys/kernel.h>
56 #include <sys/kthread.h>
57 #include <sys/mac.h>
58 #include <sys/malloc.h>
59 #include <sys/mount.h>
60 #include <sys/namei.h>
61 #include <sys/stat.h>
62 #include <sys/sysctl.h>
63 #include <sys/syslog.h>
64 #include <sys/vmmeter.h>
65 #include <sys/vnode.h>
66 
67 #include <vm/vm.h>
68 #include <vm/vm_object.h>
69 #include <vm/vm_extern.h>
70 #include <vm/pmap.h>
71 #include <vm/vm_map.h>
72 #include <vm/vm_page.h>
73 #include <vm/uma.h>
74 
75 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
76 
77 static void	addalias(struct vnode *vp, dev_t nvp_rdev);
78 static void	insmntque(struct vnode *vp, struct mount *mp);
79 static void	vclean(struct vnode *vp, int flags, struct thread *td);
80 static void	vlruvp(struct vnode *vp);
81 static int	flushbuflist(struct buf *blist, int flags, struct vnode *vp,
82 		    int slpflag, int slptimeo, int *errorp);
83 static int	vcanrecycle(struct vnode *vp);
84 
85 
86 /*
87  * Number of vnodes in existence.  Increased whenever getnewvnode()
88  * allocates a new vnode, never decreased.
89  */
90 static unsigned long	numvnodes;
91 
92 SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
93 
94 /*
95  * Conversion tables for conversion from vnode types to inode formats
96  * and back.
97  */
98 enum vtype iftovt_tab[16] = {
99 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
100 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
101 };
102 int vttoif_tab[9] = {
103 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
104 	S_IFSOCK, S_IFIFO, S_IFMT,
105 };
106 
107 /*
108  * List of vnodes that are ready for recycling.
109  */
110 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
111 
112 /*
113  * Minimum number of free vnodes.  If there are fewer than this free vnodes,
114  * getnewvnode() will return a newly allocated vnode.
115  */
116 static u_long wantfreevnodes = 25;
117 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
118 /* Number of vnodes in the free list. */
119 static u_long freevnodes;
120 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
121 
122 /*
123  * Various variables used for debugging the new implementation of
124  * reassignbuf().
125  * XXX these are probably of (very) limited utility now.
126  */
127 static int reassignbufcalls;
128 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
129 static int nameileafonly;
130 SYSCTL_INT(_vfs, OID_AUTO, nameileafonly, CTLFLAG_RW, &nameileafonly, 0, "");
131 
132 #ifdef ENABLE_VFS_IOOPT
133 /* See NOTES for a description of this setting. */
134 int vfs_ioopt;
135 SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, "");
136 #endif
137 
138 /*
139  * Cache for the mount type id assigned to NFS.  This is used for
140  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
141  */
142 int	nfs_mount_type = -1;
143 
144 /* To keep more than one thread at a time from running vfs_getnewfsid */
145 static struct mtx mntid_mtx;
146 
147 /* For any iteration/modification of vnode_free_list */
148 static struct mtx vnode_free_list_mtx;
149 
150 /*
151  * For any iteration/modification of dev->si_hlist (linked through
152  * v_specnext)
153  */
154 static struct mtx spechash_mtx;
155 
156 /* Publicly exported FS */
157 struct nfs_public nfs_pub;
158 
159 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
160 static uma_zone_t vnode_zone;
161 static uma_zone_t vnodepoll_zone;
162 
163 /* Set to 1 to print out reclaim of active vnodes */
164 int	prtactive;
165 
166 /*
167  * The workitem queue.
168  *
169  * It is useful to delay writes of file data and filesystem metadata
170  * for tens of seconds so that quickly created and deleted files need
171  * not waste disk bandwidth being created and removed. To realize this,
172  * we append vnodes to a "workitem" queue. When running with a soft
173  * updates implementation, most pending metadata dependencies should
174  * not wait for more than a few seconds. Thus, mounted on block devices
175  * are delayed only about a half the time that file data is delayed.
176  * Similarly, directory updates are more critical, so are only delayed
177  * about a third the time that file data is delayed. Thus, there are
178  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
179  * one each second (driven off the filesystem syncer process). The
180  * syncer_delayno variable indicates the next queue that is to be processed.
181  * Items that need to be processed soon are placed in this queue:
182  *
183  *	syncer_workitem_pending[syncer_delayno]
184  *
185  * A delay of fifteen seconds is done by placing the request fifteen
186  * entries later in the queue:
187  *
188  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
189  *
190  */
191 static int syncer_delayno;
192 static long syncer_mask;
193 LIST_HEAD(synclist, vnode);
194 static struct synclist *syncer_workitem_pending;
195 
196 #define SYNCER_MAXDELAY		32
197 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
198 static int syncdelay = 30;		/* max time to delay syncing data */
199 static int filedelay = 30;		/* time to delay syncing files */
200 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
201 static int dirdelay = 29;		/* time to delay syncing directories */
202 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
203 static int metadelay = 28;		/* time to delay syncing metadata */
204 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
205 static int rushjob;		/* number of slots to run ASAP */
206 static int stat_rush_requests;	/* number of times I/O speeded up */
207 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
208 
209 /*
210  * Number of vnodes we want to exist at any one time.  This is mostly used
211  * to size hash tables in vnode-related code.  It is normally not used in
212  * getnewvnode(), as wantfreevnodes is normally nonzero.)
213  *
214  * XXX desiredvnodes is historical cruft and should not exist.
215  */
216 int desiredvnodes;
217 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
218     &desiredvnodes, 0, "Maximum number of vnodes");
219 static int minvnodes;
220 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
221     &minvnodes, 0, "Minimum number of vnodes");
222 static int vnlru_nowhere;
223 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0,
224     "Number of times the vnlru process ran without success");
225 
226 /* Hook for calling soft updates */
227 int (*softdep_process_worklist_hook)(struct mount *);
228 
229 #ifdef DEBUG_VFS_LOCKS
230 /* Print lock violations */
231 int vfs_badlock_print = 1;
232 /* Panic on violation */
233 int vfs_badlock_panic = 1;
234 
235 void
236 vop_rename_pre(void *ap)
237 {
238 	struct vop_rename_args *a = ap;
239 
240 	/* Check the source (from) */
241 	if (a->a_tdvp != a->a_fdvp)
242 		ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked.\n");
243 	if (a->a_tvp != a->a_fvp)
244 		ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: tvp locked.\n");
245 
246 	/* Check the target */
247 	if (a->a_tvp)
248 		ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked.\n");
249 
250 	ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked.\n");
251 }
252 
253 void
254 vop_strategy_pre(void *ap)
255 {
256 	struct vop_strategy_args *a = ap;
257 	struct buf *bp;
258 
259 	bp = a->a_bp;
260 
261 	/*
262 	 * Cluster ops lock their component buffers but not the IO container.
263 	 */
264 	if ((bp->b_flags & B_CLUSTER) != 0)
265 		return;
266 
267 	if (BUF_REFCNT(bp) < 1) {
268 		if (vfs_badlock_print)
269 			printf("VOP_STRATEGY: bp is not locked but should be.\n");
270 		if (vfs_badlock_panic)
271 			Debugger("Lock violation.\n");
272 	}
273 }
274 
275 void
276 vop_lookup_pre(void *ap)
277 {
278 	struct vop_lookup_args *a = ap;
279 	struct vnode *dvp;
280 
281 	dvp = a->a_dvp;
282 
283 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
284 }
285 
286 void
287 vop_lookup_post(void *ap, int rc)
288 {
289 	struct vop_lookup_args *a = ap;
290 	struct componentname *cnp;
291 	struct vnode *dvp;
292 	struct vnode *vp;
293 	int flags;
294 
295 	dvp = a->a_dvp;
296 	cnp = a->a_cnp;
297 	vp = *(a->a_vpp);
298 	flags = cnp->cn_flags;
299 
300 
301 	/*
302 	 * If this is the last path component for this lookup and LOCPARENT
303 	 * is set, OR if there is an error the directory has to be locked.
304 	 */
305 	if ((flags & LOCKPARENT) && (flags & ISLASTCN))
306 		ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP (LOCKPARENT)");
307 	else if (rc != 0)
308 		ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP (error)");
309 	else if (dvp != vp)
310 		ASSERT_VOP_UNLOCKED(dvp, "VOP_LOOKUP (dvp)");
311 
312 	if (flags & PDIRUNLOCK)
313 		ASSERT_VOP_UNLOCKED(dvp, "VOP_LOOKUP (PDIRUNLOCK)");
314 
315 	if (rc == 0)
316 		ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (vpp)");
317 }
318 
319 #endif	/* DEBUG_VFS_LOCKS */
320 
321 void
322 v_addpollinfo(struct vnode *vp)
323 {
324 	vp->v_pollinfo = uma_zalloc(vnodepoll_zone, M_WAITOK);
325 	mtx_init(&vp->v_pollinfo->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
326 }
327 
328 /*
329  * Initialize the vnode management data structures.
330  */
331 static void
332 vntblinit(void *dummy __unused)
333 {
334 
335 	desiredvnodes = maxproc + cnt.v_page_count / 4;
336 	minvnodes = desiredvnodes / 4;
337 	mtx_init(&mountlist_mtx, "mountlist", NULL, MTX_DEF);
338 	mtx_init(&mntvnode_mtx, "mntvnode", NULL, MTX_DEF);
339 	mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
340 	mtx_init(&spechash_mtx, "spechash", NULL, MTX_DEF);
341 	TAILQ_INIT(&vnode_free_list);
342 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
343 	vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
344 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
345 	vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
346 	      NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
347 	/*
348 	 * Initialize the filesystem syncer.
349 	 */
350 	syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
351 		&syncer_mask);
352 	syncer_maxdelay = syncer_mask + 1;
353 }
354 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL)
355 
356 
357 /*
358  * Mark a mount point as busy. Used to synchronize access and to delay
359  * unmounting. Interlock is not released on failure.
360  */
361 int
362 vfs_busy(mp, flags, interlkp, td)
363 	struct mount *mp;
364 	int flags;
365 	struct mtx *interlkp;
366 	struct thread *td;
367 {
368 	int lkflags;
369 
370 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
371 		if (flags & LK_NOWAIT)
372 			return (ENOENT);
373 		mp->mnt_kern_flag |= MNTK_MWAIT;
374 		/*
375 		 * Since all busy locks are shared except the exclusive
376 		 * lock granted when unmounting, the only place that a
377 		 * wakeup needs to be done is at the release of the
378 		 * exclusive lock at the end of dounmount.
379 		 */
380 		msleep(mp, interlkp, PVFS, "vfs_busy", 0);
381 		return (ENOENT);
382 	}
383 	lkflags = LK_SHARED | LK_NOPAUSE;
384 	if (interlkp)
385 		lkflags |= LK_INTERLOCK;
386 	if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
387 		panic("vfs_busy: unexpected lock failure");
388 	return (0);
389 }
390 
391 /*
392  * Free a busy filesystem.
393  */
394 void
395 vfs_unbusy(mp, td)
396 	struct mount *mp;
397 	struct thread *td;
398 {
399 
400 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
401 }
402 
403 /*
404  * Lookup a mount point by filesystem identifier.
405  */
406 struct mount *
407 vfs_getvfs(fsid)
408 	fsid_t *fsid;
409 {
410 	register struct mount *mp;
411 
412 	mtx_lock(&mountlist_mtx);
413 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
414 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
415 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
416 			mtx_unlock(&mountlist_mtx);
417 			return (mp);
418 	    }
419 	}
420 	mtx_unlock(&mountlist_mtx);
421 	return ((struct mount *) 0);
422 }
423 
424 /*
425  * Get a new unique fsid.  Try to make its val[0] unique, since this value
426  * will be used to create fake device numbers for stat().  Also try (but
427  * not so hard) make its val[0] unique mod 2^16, since some emulators only
428  * support 16-bit device numbers.  We end up with unique val[0]'s for the
429  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
430  *
431  * Keep in mind that several mounts may be running in parallel.  Starting
432  * the search one past where the previous search terminated is both a
433  * micro-optimization and a defense against returning the same fsid to
434  * different mounts.
435  */
436 void
437 vfs_getnewfsid(mp)
438 	struct mount *mp;
439 {
440 	static u_int16_t mntid_base;
441 	fsid_t tfsid;
442 	int mtype;
443 
444 	mtx_lock(&mntid_mtx);
445 	mtype = mp->mnt_vfc->vfc_typenum;
446 	tfsid.val[1] = mtype;
447 	mtype = (mtype & 0xFF) << 24;
448 	for (;;) {
449 		tfsid.val[0] = makeudev(255,
450 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
451 		mntid_base++;
452 		if (vfs_getvfs(&tfsid) == NULL)
453 			break;
454 	}
455 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
456 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
457 	mtx_unlock(&mntid_mtx);
458 }
459 
460 /*
461  * Knob to control the precision of file timestamps:
462  *
463  *   0 = seconds only; nanoseconds zeroed.
464  *   1 = seconds and nanoseconds, accurate within 1/HZ.
465  *   2 = seconds and nanoseconds, truncated to microseconds.
466  * >=3 = seconds and nanoseconds, maximum precision.
467  */
468 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
469 
470 static int timestamp_precision = TSP_SEC;
471 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
472     &timestamp_precision, 0, "");
473 
474 /*
475  * Get a current timestamp.
476  */
477 void
478 vfs_timestamp(tsp)
479 	struct timespec *tsp;
480 {
481 	struct timeval tv;
482 
483 	switch (timestamp_precision) {
484 	case TSP_SEC:
485 		tsp->tv_sec = time_second;
486 		tsp->tv_nsec = 0;
487 		break;
488 	case TSP_HZ:
489 		getnanotime(tsp);
490 		break;
491 	case TSP_USEC:
492 		microtime(&tv);
493 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
494 		break;
495 	case TSP_NSEC:
496 	default:
497 		nanotime(tsp);
498 		break;
499 	}
500 }
501 
502 /*
503  * Set vnode attributes to VNOVAL
504  */
505 void
506 vattr_null(vap)
507 	register struct vattr *vap;
508 {
509 
510 	vap->va_type = VNON;
511 	vap->va_size = VNOVAL;
512 	vap->va_bytes = VNOVAL;
513 	vap->va_mode = VNOVAL;
514 	vap->va_nlink = VNOVAL;
515 	vap->va_uid = VNOVAL;
516 	vap->va_gid = VNOVAL;
517 	vap->va_fsid = VNOVAL;
518 	vap->va_fileid = VNOVAL;
519 	vap->va_blocksize = VNOVAL;
520 	vap->va_rdev = VNOVAL;
521 	vap->va_atime.tv_sec = VNOVAL;
522 	vap->va_atime.tv_nsec = VNOVAL;
523 	vap->va_mtime.tv_sec = VNOVAL;
524 	vap->va_mtime.tv_nsec = VNOVAL;
525 	vap->va_ctime.tv_sec = VNOVAL;
526 	vap->va_ctime.tv_nsec = VNOVAL;
527 	vap->va_birthtime.tv_sec = VNOVAL;
528 	vap->va_birthtime.tv_nsec = VNOVAL;
529 	vap->va_flags = VNOVAL;
530 	vap->va_gen = VNOVAL;
531 	vap->va_vaflags = 0;
532 }
533 
534 /*
535  * This routine is called when we have too many vnodes.  It attempts
536  * to free <count> vnodes and will potentially free vnodes that still
537  * have VM backing store (VM backing store is typically the cause
538  * of a vnode blowout so we want to do this).  Therefore, this operation
539  * is not considered cheap.
540  *
541  * A number of conditions may prevent a vnode from being reclaimed.
542  * the buffer cache may have references on the vnode, a directory
543  * vnode may still have references due to the namei cache representing
544  * underlying files, or the vnode may be in active use.   It is not
545  * desireable to reuse such vnodes.  These conditions may cause the
546  * number of vnodes to reach some minimum value regardless of what
547  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
548  */
549 static int
550 vlrureclaim(struct mount *mp, int count)
551 {
552 	struct vnode *vp;
553 	int done;
554 	int trigger;
555 	int usevnodes;
556 
557 	/*
558 	 * Calculate the trigger point, don't allow user
559 	 * screwups to blow us up.   This prevents us from
560 	 * recycling vnodes with lots of resident pages.  We
561 	 * aren't trying to free memory, we are trying to
562 	 * free vnodes.
563 	 */
564 	usevnodes = desiredvnodes;
565 	if (usevnodes <= 0)
566 		usevnodes = 1;
567 	trigger = cnt.v_page_count * 2 / usevnodes;
568 
569 	done = 0;
570 	mtx_lock(&mntvnode_mtx);
571 	while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
572 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
573 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
574 
575 		if (vp->v_type != VNON &&
576 		    vp->v_type != VBAD &&
577 		    VI_TRYLOCK(vp)) {
578 			if (VMIGHTFREE(vp) &&           /* critical path opt */
579 			    (vp->v_object == NULL ||
580 			    vp->v_object->resident_page_count < trigger)) {
581 				mtx_unlock(&mntvnode_mtx);
582 				vgonel(vp, curthread);
583 				done++;
584 				mtx_lock(&mntvnode_mtx);
585 			} else
586 				VI_UNLOCK(vp);
587 		}
588 		--count;
589 	}
590 	mtx_unlock(&mntvnode_mtx);
591 	return done;
592 }
593 
594 /*
595  * Attempt to recycle vnodes in a context that is always safe to block.
596  * Calling vlrurecycle() from the bowels of filesystem code has some
597  * interesting deadlock problems.
598  */
599 static struct proc *vnlruproc;
600 static int vnlruproc_sig;
601 
602 static void
603 vnlru_proc(void)
604 {
605 	struct mount *mp, *nmp;
606 	int s;
607 	int done;
608 	struct proc *p = vnlruproc;
609 	struct thread *td = FIRST_THREAD_IN_PROC(p);	/* XXXKSE */
610 
611 	mtx_lock(&Giant);
612 
613 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
614 	    SHUTDOWN_PRI_FIRST);
615 
616 	s = splbio();
617 	for (;;) {
618 		kthread_suspend_check(p);
619 		if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
620 			vnlruproc_sig = 0;
621 			tsleep(vnlruproc, PVFS, "vlruwt", 0);
622 			continue;
623 		}
624 		done = 0;
625 		mtx_lock(&mountlist_mtx);
626 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
627 			if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
628 				nmp = TAILQ_NEXT(mp, mnt_list);
629 				continue;
630 			}
631 			done += vlrureclaim(mp, 10);
632 			mtx_lock(&mountlist_mtx);
633 			nmp = TAILQ_NEXT(mp, mnt_list);
634 			vfs_unbusy(mp, td);
635 		}
636 		mtx_unlock(&mountlist_mtx);
637 		if (done == 0) {
638 #if 0
639 			/* These messages are temporary debugging aids */
640 			if (vnlru_nowhere < 5)
641 				printf("vnlru process getting nowhere..\n");
642 			else if (vnlru_nowhere == 5)
643 				printf("vnlru process messages stopped.\n");
644 #endif
645 			vnlru_nowhere++;
646 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
647 		}
648 	}
649 	splx(s);
650 }
651 
652 static struct kproc_desc vnlru_kp = {
653 	"vnlru",
654 	vnlru_proc,
655 	&vnlruproc
656 };
657 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
658 
659 
660 /*
661  * Routines having to do with the management of the vnode table.
662  */
663 
664 /*
665  * Check to see if a free vnode can be recycled.  If it can, return it locked
666  * with the vn lock, but not interlock.  Otherwise indicate the error.
667  */
668 static int
669 vcanrecycle(struct vnode *vp)
670 {
671 	struct thread *td = curthread;
672 	vm_object_t object;
673 	int error;
674 
675 	/* Don't recycle if we can't get the interlock */
676 	if (!mtx_trylock(&vp->v_interlock))
677 		return (EWOULDBLOCK);
678 
679 	/* We should be able to immediately acquire this */
680 	/* XXX This looks like it should panic if it fails */
681 	if (vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE, td) != 0)
682 		return (EWOULDBLOCK);
683 	/*
684 	 * Don't recycle if we still have cached pages.
685 	 */
686 	if (VOP_GETVOBJECT(vp, &object) == 0 &&
687 	     (object->resident_page_count ||
688 	      object->ref_count)) {
689 		VOP_UNLOCK(vp, 0, td);
690 		error = EBUSY;
691 		goto done;
692 	}
693 	if (LIST_FIRST(&vp->v_cache_src)) {
694 		/*
695 		 * note: nameileafonly sysctl is temporary,
696 		 * for debugging only, and will eventually be
697 		 * removed.
698 		 */
699 		if (nameileafonly > 0) {
700 			/*
701 			 * Do not reuse namei-cached directory
702 			 * vnodes that have cached
703 			 * subdirectories.
704 			 */
705 			if (cache_leaf_test(vp) < 0) {
706 				error = EISDIR;
707 				goto done;
708 			}
709 		} else if (nameileafonly < 0 ||
710 			    vmiodirenable == 0) {
711 			/*
712 			 * Do not reuse namei-cached directory
713 			 * vnodes if nameileafonly is -1 or
714 			 * if VMIO backing for directories is
715 			 * turned off (otherwise we reuse them
716 			 * too quickly).
717 			 */
718 			error = EBUSY;
719 			goto done;
720 		}
721 	}
722 	return (0);
723 done:
724 	VOP_UNLOCK(vp, 0, td);
725 	return (error);
726 }
727 
728 /*
729  * Return the next vnode from the free list.
730  */
731 int
732 getnewvnode(tag, mp, vops, vpp)
733 	enum vtagtype tag;
734 	struct mount *mp;
735 	vop_t **vops;
736 	struct vnode **vpp;
737 {
738 	int s;
739 	struct thread *td = curthread;	/* XXX */
740 	struct vnode *vp = NULL;
741 	struct mount *vnmp;
742 
743 	s = splbio();
744 	/*
745 	 * Try to reuse vnodes if we hit the max.  This situation only
746 	 * occurs in certain large-memory (2G+) situations.  We cannot
747 	 * attempt to directly reclaim vnodes due to nasty recursion
748 	 * problems.
749 	 */
750 	if (vnlruproc_sig == 0 && numvnodes - freevnodes > desiredvnodes) {
751 		vnlruproc_sig = 1;      /* avoid unnecessary wakeups */
752 		wakeup(vnlruproc);
753 	}
754 
755 	/*
756 	 * Attempt to reuse a vnode already on the free list, allocating
757 	 * a new vnode if we can't find one or if we have not reached a
758 	 * good minimum for good LRU performance.
759 	 */
760 
761 	mtx_lock(&vnode_free_list_mtx);
762 
763 	if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
764 		int error;
765 		int count;
766 
767 		for (count = 0; count < freevnodes; count++) {
768 			vp = TAILQ_FIRST(&vnode_free_list);
769 
770 			KASSERT(vp->v_usecount == 0,
771 			    ("getnewvnode: free vnode isn't"));
772 
773 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
774 			/*
775 			 * We have to drop the free list mtx to avoid lock
776 			 * order reversals with interlock.
777 			 */
778 			mtx_unlock(&vnode_free_list_mtx);
779 			error = vcanrecycle(vp);
780 			/*
781 			 * Skip over it if its filesystem is being suspended.
782 			 */
783 			if (error == 0 &&
784 			    vn_start_write(vp, &vnmp, V_NOWAIT) != 0)
785 				error = EBUSY;
786 
787 			mtx_lock(&vnode_free_list_mtx);
788 			if (error != 0)
789 				TAILQ_INSERT_TAIL(&vnode_free_list, vp,
790 				    v_freelist);
791 			else
792 				break;
793 		}
794 	}
795 	/*
796 	 * Unlocked access to this vp is ok because we are assured that there
797 	 * are no other references to it.
798 	 */
799 	if (vp) {
800 		freevnodes--;
801 		mtx_unlock(&vnode_free_list_mtx);
802 
803 		vp->v_iflag |= VI_DOOMED;
804 		vp->v_iflag &= ~VI_FREE;
805 		cache_purge(vp);
806 		if (vp->v_type != VBAD) {
807 			VOP_UNLOCK(vp, 0, td);
808 			vgone(vp);
809 		} else {
810 			VOP_UNLOCK(vp, 0, td);
811 		}
812 		vn_finished_write(vnmp);
813 
814 #ifdef INVARIANTS
815 		{
816 			if (vp->v_data)
817 				panic("cleaned vnode isn't");
818 			VI_LOCK(vp);
819 			if (vp->v_numoutput)
820 				panic("Clean vnode has pending I/O's");
821 			if (vp->v_writecount != 0)
822 				panic("Non-zero write count");
823 			VI_UNLOCK(vp);
824 		}
825 #endif
826 		if (vp->v_pollinfo) {
827 			mtx_destroy(&vp->v_pollinfo->vpi_lock);
828 			uma_zfree(vnodepoll_zone, vp->v_pollinfo);
829 		}
830 		vp->v_pollinfo = NULL;
831 #ifdef MAC
832 		mac_destroy_vnode(vp);
833 #endif
834 		vp->v_iflag = 0;
835 		vp->v_vflag = 0;
836 		vp->v_lastw = 0;
837 		vp->v_lasta = 0;
838 		vp->v_cstart = 0;
839 		vp->v_clen = 0;
840 		vp->v_socket = 0;
841 		KASSERT(vp->v_cleanblkroot == NULL, ("cleanblkroot not NULL"));
842 		KASSERT(vp->v_dirtyblkroot == NULL, ("dirtyblkroot not NULL"));
843 	} else {
844 		numvnodes++;
845 		mtx_unlock(&vnode_free_list_mtx);
846 
847 		vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
848 		mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
849 		vp->v_dd = vp;
850 		cache_purge(vp);
851 		LIST_INIT(&vp->v_cache_src);
852 		TAILQ_INIT(&vp->v_cache_dst);
853 	}
854 
855 	TAILQ_INIT(&vp->v_cleanblkhd);
856 	TAILQ_INIT(&vp->v_dirtyblkhd);
857 	vp->v_type = VNON;
858 	vp->v_tag = tag;
859 	vp->v_op = vops;
860 	lockinit(&vp->v_lock, PVFS, "vnlock", VLKTIMEOUT, LK_NOPAUSE);
861 #ifdef MAC
862 	mac_init_vnode(vp);
863 #endif
864 	insmntque(vp, mp);
865 	*vpp = vp;
866 	vp->v_usecount = 1;
867 	vp->v_data = 0;
868 	vp->v_cachedid = -1;
869 
870 	splx(s);
871 
872 #if 0
873 	vnodeallocs++;
874 	if (vnodeallocs % vnoderecycleperiod == 0 &&
875 	    freevnodes < vnoderecycleminfreevn &&
876 	    vnoderecyclemintotalvn < numvnodes) {
877 		/* Recycle vnodes. */
878 		cache_purgeleafdirs(vnoderecyclenumber);
879 	}
880 #endif
881 
882 	return (0);
883 }
884 
885 /*
886  * Move a vnode from one mount queue to another.
887  */
888 static void
889 insmntque(vp, mp)
890 	register struct vnode *vp;
891 	register struct mount *mp;
892 {
893 
894 	mtx_lock(&mntvnode_mtx);
895 	/*
896 	 * Delete from old mount point vnode list, if on one.
897 	 */
898 	if (vp->v_mount != NULL)
899 		TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
900 	/*
901 	 * Insert into list of vnodes for the new mount point, if available.
902 	 */
903 	if ((vp->v_mount = mp) == NULL) {
904 		mtx_unlock(&mntvnode_mtx);
905 		return;
906 	}
907 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
908 	mtx_unlock(&mntvnode_mtx);
909 }
910 
911 /*
912  * Update outstanding I/O count and do wakeup if requested.
913  */
914 void
915 vwakeup(bp)
916 	register struct buf *bp;
917 {
918 	register struct vnode *vp;
919 
920 	bp->b_flags &= ~B_WRITEINPROG;
921 	if ((vp = bp->b_vp)) {
922 		VI_LOCK(vp);
923 		vp->v_numoutput--;
924 		if (vp->v_numoutput < 0)
925 			panic("vwakeup: neg numoutput");
926 		if ((vp->v_numoutput == 0) && (vp->v_iflag & VI_BWAIT)) {
927 			vp->v_iflag &= ~VI_BWAIT;
928 			wakeup(&vp->v_numoutput);
929 		}
930 		VI_UNLOCK(vp);
931 	}
932 }
933 
934 /*
935  * Flush out and invalidate all buffers associated with a vnode.
936  * Called with the underlying object locked.
937  */
938 int
939 vinvalbuf(vp, flags, cred, td, slpflag, slptimeo)
940 	struct vnode *vp;
941 	int flags;
942 	struct ucred *cred;
943 	struct thread *td;
944 	int slpflag, slptimeo;
945 {
946 	struct buf *blist;
947 	int s, error;
948 	vm_object_t object;
949 
950 	GIANT_REQUIRED;
951 
952 	if (flags & V_SAVE) {
953 		s = splbio();
954 		VI_LOCK(vp);
955 		while (vp->v_numoutput) {
956 			vp->v_iflag |= VI_BWAIT;
957 			error = msleep(&vp->v_numoutput, VI_MTX(vp),
958 			    slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo);
959 			if (error) {
960 				VI_UNLOCK(vp);
961 				splx(s);
962 				return (error);
963 			}
964 		}
965 		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
966 			splx(s);
967 			VI_UNLOCK(vp);
968 			if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, td)) != 0)
969 				return (error);
970 			/*
971 			 * XXX We could save a lock/unlock if this was only
972 			 * enabled under INVARIANTS
973 			 */
974 			VI_LOCK(vp);
975 			s = splbio();
976 			if (vp->v_numoutput > 0 ||
977 			    !TAILQ_EMPTY(&vp->v_dirtyblkhd))
978 				panic("vinvalbuf: dirty bufs");
979 		}
980 		VI_UNLOCK(vp);
981 		splx(s);
982 	}
983 	s = splbio();
984 	for (error = 0;;) {
985 		if ((blist = TAILQ_FIRST(&vp->v_cleanblkhd)) != 0 &&
986 		    flushbuflist(blist, flags, vp, slpflag, slptimeo, &error)) {
987 			if (error)
988 				break;
989 			continue;
990 		}
991 		if ((blist = TAILQ_FIRST(&vp->v_dirtyblkhd)) != 0 &&
992 		    flushbuflist(blist, flags, vp, slpflag, slptimeo, &error)) {
993 			if (error)
994 				break;
995 			continue;
996 		}
997 		break;
998 	}
999 	if (error) {
1000 		splx(s);
1001 		return (error);
1002 	}
1003 
1004 	/*
1005 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1006 	 * have write I/O in-progress but if there is a VM object then the
1007 	 * VM object can also have read-I/O in-progress.
1008 	 */
1009 	VI_LOCK(vp);
1010 	do {
1011 		while (vp->v_numoutput > 0) {
1012 			vp->v_iflag |= VI_BWAIT;
1013 			msleep(&vp->v_numoutput, VI_MTX(vp), PVM, "vnvlbv", 0);
1014 		}
1015 		VI_UNLOCK(vp);
1016 		if (VOP_GETVOBJECT(vp, &object) == 0) {
1017 			while (object->paging_in_progress)
1018 			vm_object_pip_sleep(object, "vnvlbx");
1019 		}
1020 		VI_LOCK(vp);
1021 	} while (vp->v_numoutput > 0);
1022 	VI_UNLOCK(vp);
1023 
1024 	splx(s);
1025 
1026 	/*
1027 	 * Destroy the copy in the VM cache, too.
1028 	 */
1029 	if (VOP_GETVOBJECT(vp, &object) == 0) {
1030 		vm_object_page_remove(object, 0, 0,
1031 			(flags & V_SAVE) ? TRUE : FALSE);
1032 	}
1033 
1034 	if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1035 	    (!TAILQ_EMPTY(&vp->v_dirtyblkhd) ||
1036 	     !TAILQ_EMPTY(&vp->v_cleanblkhd)))
1037 		panic("vinvalbuf: flush failed");
1038 	return (0);
1039 }
1040 
1041 /*
1042  * Flush out buffers on the specified list.
1043  */
1044 static int
1045 flushbuflist(blist, flags, vp, slpflag, slptimeo, errorp)
1046 	struct buf *blist;
1047 	int flags;
1048 	struct vnode *vp;
1049 	int slpflag, slptimeo;
1050 	int *errorp;
1051 {
1052 	struct buf *bp, *nbp;
1053 	int found, error;
1054 
1055 	for (found = 0, bp = blist; bp; bp = nbp) {
1056 		nbp = TAILQ_NEXT(bp, b_vnbufs);
1057 		if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1058 		    ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0))
1059 			continue;
1060 		found += 1;
1061 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1062 			error = BUF_TIMELOCK(bp,
1063 			    LK_EXCLUSIVE | LK_SLEEPFAIL,
1064 			    "flushbuf", slpflag, slptimeo);
1065 			if (error != ENOLCK)
1066 				*errorp = error;
1067 			return (found);
1068 		}
1069 		/*
1070 		 * XXX Since there are no node locks for NFS, I
1071 		 * believe there is a slight chance that a delayed
1072 		 * write will occur while sleeping just above, so
1073 		 * check for it.  Note that vfs_bio_awrite expects
1074 		 * buffers to reside on a queue, while BUF_WRITE and
1075 		 * brelse do not.
1076 		 */
1077 		if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1078 			(flags & V_SAVE)) {
1079 
1080 			if (bp->b_vp == vp) {
1081 				if (bp->b_flags & B_CLUSTEROK) {
1082 					BUF_UNLOCK(bp);
1083 					vfs_bio_awrite(bp);
1084 				} else {
1085 					bremfree(bp);
1086 					bp->b_flags |= B_ASYNC;
1087 					BUF_WRITE(bp);
1088 				}
1089 			} else {
1090 				bremfree(bp);
1091 				(void) BUF_WRITE(bp);
1092 			}
1093 			return (found);
1094 		}
1095 		bremfree(bp);
1096 		bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
1097 		bp->b_flags &= ~B_ASYNC;
1098 		brelse(bp);
1099 	}
1100 	return (found);
1101 }
1102 
1103 /*
1104  * Truncate a file's buffer and pages to a specified length.  This
1105  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1106  * sync activity.
1107  */
1108 int
1109 vtruncbuf(vp, cred, td, length, blksize)
1110 	register struct vnode *vp;
1111 	struct ucred *cred;
1112 	struct thread *td;
1113 	off_t length;
1114 	int blksize;
1115 {
1116 	register struct buf *bp;
1117 	struct buf *nbp;
1118 	int s, anyfreed;
1119 	int trunclbn;
1120 
1121 	/*
1122 	 * Round up to the *next* lbn.
1123 	 */
1124 	trunclbn = (length + blksize - 1) / blksize;
1125 
1126 	s = splbio();
1127 restart:
1128 	anyfreed = 1;
1129 	for (;anyfreed;) {
1130 		anyfreed = 0;
1131 		for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
1132 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1133 			if (bp->b_lblkno >= trunclbn) {
1134 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1135 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1136 					goto restart;
1137 				} else {
1138 					bremfree(bp);
1139 					bp->b_flags |= (B_INVAL | B_RELBUF);
1140 					bp->b_flags &= ~B_ASYNC;
1141 					brelse(bp);
1142 					anyfreed = 1;
1143 				}
1144 				if (nbp &&
1145 				    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1146 				    (nbp->b_vp != vp) ||
1147 				    (nbp->b_flags & B_DELWRI))) {
1148 					goto restart;
1149 				}
1150 			}
1151 		}
1152 
1153 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1154 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1155 			if (bp->b_lblkno >= trunclbn) {
1156 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1157 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1158 					goto restart;
1159 				} else {
1160 					bremfree(bp);
1161 					bp->b_flags |= (B_INVAL | B_RELBUF);
1162 					bp->b_flags &= ~B_ASYNC;
1163 					brelse(bp);
1164 					anyfreed = 1;
1165 				}
1166 				if (nbp &&
1167 				    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1168 				    (nbp->b_vp != vp) ||
1169 				    (nbp->b_flags & B_DELWRI) == 0)) {
1170 					goto restart;
1171 				}
1172 			}
1173 		}
1174 	}
1175 
1176 	if (length > 0) {
1177 restartsync:
1178 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1179 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1180 			if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
1181 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
1182 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
1183 					goto restart;
1184 				} else {
1185 					bremfree(bp);
1186 					if (bp->b_vp == vp) {
1187 						bp->b_flags |= B_ASYNC;
1188 					} else {
1189 						bp->b_flags &= ~B_ASYNC;
1190 					}
1191 					BUF_WRITE(bp);
1192 				}
1193 				goto restartsync;
1194 			}
1195 
1196 		}
1197 	}
1198 
1199 	VI_LOCK(vp);
1200 	while (vp->v_numoutput > 0) {
1201 		vp->v_iflag |= VI_BWAIT;
1202 		msleep(&vp->v_numoutput, VI_MTX(vp), PVM, "vbtrunc", 0);
1203 	}
1204 	VI_UNLOCK(vp);
1205 	splx(s);
1206 
1207 	vnode_pager_setsize(vp, length);
1208 
1209 	return (0);
1210 }
1211 
1212 /*
1213  * buf_splay() - splay tree core for the clean/dirty list of buffers in
1214  * 		 a vnode.
1215  *
1216  *	NOTE: We have to deal with the special case of a background bitmap
1217  *	buffer, a situation where two buffers will have the same logical
1218  *	block offset.  We want (1) only the foreground buffer to be accessed
1219  *	in a lookup and (2) must differentiate between the foreground and
1220  *	background buffer in the splay tree algorithm because the splay
1221  *	tree cannot normally handle multiple entities with the same 'index'.
1222  *	We accomplish this by adding differentiating flags to the splay tree's
1223  *	numerical domain.
1224  */
1225 static
1226 struct buf *
1227 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1228 {
1229 	struct buf dummy;
1230 	struct buf *lefttreemax, *righttreemin, *y;
1231 
1232 	if (root == NULL)
1233 		return (NULL);
1234 	lefttreemax = righttreemin = &dummy;
1235 	for (;;) {
1236 		if (lblkno < root->b_lblkno ||
1237 		    (lblkno == root->b_lblkno &&
1238 		    (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1239 			if ((y = root->b_left) == NULL)
1240 				break;
1241 			if (lblkno < y->b_lblkno) {
1242 				/* Rotate right. */
1243 				root->b_left = y->b_right;
1244 				y->b_right = root;
1245 				root = y;
1246 				if ((y = root->b_left) == NULL)
1247 					break;
1248 			}
1249 			/* Link into the new root's right tree. */
1250 			righttreemin->b_left = root;
1251 			righttreemin = root;
1252 		} else if (lblkno > root->b_lblkno ||
1253 		    (lblkno == root->b_lblkno &&
1254 		    (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1255 			if ((y = root->b_right) == NULL)
1256 				break;
1257 			if (lblkno > y->b_lblkno) {
1258 				/* Rotate left. */
1259 				root->b_right = y->b_left;
1260 				y->b_left = root;
1261 				root = y;
1262 				if ((y = root->b_right) == NULL)
1263 					break;
1264 			}
1265 			/* Link into the new root's left tree. */
1266 			lefttreemax->b_right = root;
1267 			lefttreemax = root;
1268 		} else {
1269 			break;
1270 		}
1271 		root = y;
1272 	}
1273 	/* Assemble the new root. */
1274 	lefttreemax->b_right = root->b_left;
1275 	righttreemin->b_left = root->b_right;
1276 	root->b_left = dummy.b_right;
1277 	root->b_right = dummy.b_left;
1278 	return (root);
1279 }
1280 
1281 static
1282 void
1283 buf_vlist_remove(struct buf *bp)
1284 {
1285 	struct vnode *vp = bp->b_vp;
1286 	struct buf *root;
1287 
1288 	if (bp->b_xflags & BX_VNDIRTY) {
1289 		if (bp != vp->v_dirtyblkroot) {
1290 			root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_dirtyblkroot);
1291 			KASSERT(root == bp, ("splay lookup failed during dirty remove"));
1292 		}
1293 		if (bp->b_left == NULL) {
1294 			root = bp->b_right;
1295 		} else {
1296 			root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1297 			root->b_right = bp->b_right;
1298 		}
1299 		vp->v_dirtyblkroot = root;
1300 		TAILQ_REMOVE(&vp->v_dirtyblkhd, bp, b_vnbufs);
1301 	} else {
1302 		/* KASSERT(bp->b_xflags & BX_VNCLEAN, ("bp wasn't clean")); */
1303 		if (bp != vp->v_cleanblkroot) {
1304 			root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_cleanblkroot);
1305 			KASSERT(root == bp, ("splay lookup failed during clean remove"));
1306 		}
1307 		if (bp->b_left == NULL) {
1308 			root = bp->b_right;
1309 		} else {
1310 			root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1311 			root->b_right = bp->b_right;
1312 		}
1313 		vp->v_cleanblkroot = root;
1314 		TAILQ_REMOVE(&vp->v_cleanblkhd, bp, b_vnbufs);
1315 	}
1316 	bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1317 }
1318 
1319 /*
1320  * Add the buffer to the sorted clean or dirty block list using a
1321  * splay tree algorithm.
1322  *
1323  * NOTE: xflags is passed as a constant, optimizing this inline function!
1324  */
1325 static
1326 void
1327 buf_vlist_add(struct buf *bp, struct vnode *vp, b_xflags_t xflags)
1328 {
1329 	struct buf *root;
1330 
1331 	bp->b_xflags |= xflags;
1332 	if (xflags & BX_VNDIRTY) {
1333 		root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_dirtyblkroot);
1334 		if (root == NULL) {
1335 			bp->b_left = NULL;
1336 			bp->b_right = NULL;
1337 			TAILQ_INSERT_TAIL(&vp->v_dirtyblkhd, bp, b_vnbufs);
1338 		} else if (bp->b_lblkno < root->b_lblkno ||
1339 		    (bp->b_lblkno == root->b_lblkno &&
1340 		    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1341 			bp->b_left = root->b_left;
1342 			bp->b_right = root;
1343 			root->b_left = NULL;
1344 			TAILQ_INSERT_BEFORE(root, bp, b_vnbufs);
1345 		} else {
1346 			bp->b_right = root->b_right;
1347 			bp->b_left = root;
1348 			root->b_right = NULL;
1349 			TAILQ_INSERT_AFTER(&vp->v_dirtyblkhd,
1350 			    root, bp, b_vnbufs);
1351 		}
1352 		vp->v_dirtyblkroot = bp;
1353 	} else {
1354 		/* KASSERT(xflags & BX_VNCLEAN, ("xflags not clean")); */
1355 		root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_cleanblkroot);
1356 		if (root == NULL) {
1357 			bp->b_left = NULL;
1358 			bp->b_right = NULL;
1359 			TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
1360 		} else if (bp->b_lblkno < root->b_lblkno ||
1361 		    (bp->b_lblkno == root->b_lblkno &&
1362 		    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1363 			bp->b_left = root->b_left;
1364 			bp->b_right = root;
1365 			root->b_left = NULL;
1366 			TAILQ_INSERT_BEFORE(root, bp, b_vnbufs);
1367 		} else {
1368 			bp->b_right = root->b_right;
1369 			bp->b_left = root;
1370 			root->b_right = NULL;
1371 			TAILQ_INSERT_AFTER(&vp->v_cleanblkhd,
1372 			    root, bp, b_vnbufs);
1373 		}
1374 		vp->v_cleanblkroot = bp;
1375 	}
1376 }
1377 
1378 #ifndef USE_BUFHASH
1379 
1380 /*
1381  * Lookup a buffer using the splay tree.  Note that we specifically avoid
1382  * shadow buffers used in background bitmap writes.
1383  *
1384  * This code isn't quite efficient as it could be because we are maintaining
1385  * two sorted lists and do not know which list the block resides in.
1386  */
1387 struct buf *
1388 gbincore(struct vnode *vp, daddr_t lblkno)
1389 {
1390 	struct buf *bp;
1391 
1392 	GIANT_REQUIRED;
1393 
1394 	bp = vp->v_cleanblkroot = buf_splay(lblkno, 0, vp->v_cleanblkroot);
1395 	if (bp && bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1396 		return(bp);
1397 	bp = vp->v_dirtyblkroot = buf_splay(lblkno, 0, vp->v_dirtyblkroot);
1398 	if (bp && bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1399 		return(bp);
1400 	return(NULL);
1401 }
1402 
1403 #endif
1404 
1405 /*
1406  * Associate a buffer with a vnode.
1407  */
1408 void
1409 bgetvp(vp, bp)
1410 	register struct vnode *vp;
1411 	register struct buf *bp;
1412 {
1413 	int s;
1414 
1415 	KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
1416 
1417 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1418 	    ("bgetvp: bp already attached! %p", bp));
1419 
1420 	vhold(vp);
1421 	bp->b_vp = vp;
1422 	bp->b_dev = vn_todev(vp);
1423 	/*
1424 	 * Insert onto list for new vnode.
1425 	 */
1426 	s = splbio();
1427 	buf_vlist_add(bp, vp, BX_VNCLEAN);
1428 	splx(s);
1429 }
1430 
1431 /*
1432  * Disassociate a buffer from a vnode.
1433  */
1434 void
1435 brelvp(bp)
1436 	register struct buf *bp;
1437 {
1438 	struct vnode *vp;
1439 	int s;
1440 
1441 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1442 
1443 	/*
1444 	 * Delete from old vnode list, if on one.
1445 	 */
1446 	vp = bp->b_vp;
1447 	s = splbio();
1448 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1449 		buf_vlist_remove(bp);
1450 	VI_LOCK(vp);
1451 	if ((vp->v_iflag & VI_ONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1452 		vp->v_iflag &= ~VI_ONWORKLST;
1453 		LIST_REMOVE(vp, v_synclist);
1454 	}
1455 	VI_UNLOCK(vp);
1456 	splx(s);
1457 	bp->b_vp = (struct vnode *) 0;
1458 	vdrop(vp);
1459 	if (bp->b_object)
1460 		bp->b_object = NULL;
1461 }
1462 
1463 /*
1464  * Add an item to the syncer work queue.
1465  */
1466 static void
1467 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
1468 {
1469 	int s, slot;
1470 
1471 	s = splbio();
1472 	mtx_assert(VI_MTX(vp), MA_OWNED);
1473 
1474 	if (vp->v_iflag & VI_ONWORKLST)
1475 		LIST_REMOVE(vp, v_synclist);
1476 	else
1477 		vp->v_iflag |= VI_ONWORKLST;
1478 
1479 	if (delay > syncer_maxdelay - 2)
1480 		delay = syncer_maxdelay - 2;
1481 	slot = (syncer_delayno + delay) & syncer_mask;
1482 
1483 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
1484 
1485 	splx(s);
1486 }
1487 
1488 struct  proc *updateproc;
1489 static void sched_sync(void);
1490 static struct kproc_desc up_kp = {
1491 	"syncer",
1492 	sched_sync,
1493 	&updateproc
1494 };
1495 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1496 
1497 /*
1498  * System filesystem synchronizer daemon.
1499  */
1500 void
1501 sched_sync(void)
1502 {
1503 	struct synclist *slp;
1504 	struct vnode *vp;
1505 	struct mount *mp;
1506 	long starttime;
1507 	int s;
1508 	struct thread *td = FIRST_THREAD_IN_PROC(updateproc);  /* XXXKSE */
1509 
1510 	mtx_lock(&Giant);
1511 
1512 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, td->td_proc,
1513 	    SHUTDOWN_PRI_LAST);
1514 
1515 	for (;;) {
1516 		kthread_suspend_check(td->td_proc);
1517 
1518 		starttime = time_second;
1519 
1520 		/*
1521 		 * Push files whose dirty time has expired.  Be careful
1522 		 * of interrupt race on slp queue.
1523 		 */
1524 		s = splbio();
1525 		slp = &syncer_workitem_pending[syncer_delayno];
1526 		syncer_delayno += 1;
1527 		if (syncer_delayno == syncer_maxdelay)
1528 			syncer_delayno = 0;
1529 		splx(s);
1530 
1531 		while ((vp = LIST_FIRST(slp)) != NULL) {
1532 			if (VOP_ISLOCKED(vp, NULL) == 0 &&
1533 			    vn_start_write(vp, &mp, V_NOWAIT) == 0) {
1534 				vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1535 				(void) VOP_FSYNC(vp, td->td_ucred, MNT_LAZY, td);
1536 				VOP_UNLOCK(vp, 0, td);
1537 				vn_finished_write(mp);
1538 			}
1539 			s = splbio();
1540 			if (LIST_FIRST(slp) == vp) {
1541 				/*
1542 				 * Note: v_tag VT_VFS vps can remain on the
1543 				 * worklist too with no dirty blocks, but
1544 				 * since sync_fsync() moves it to a different
1545 				 * slot we are safe.
1546 				 */
1547 				if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1548 				    !vn_isdisk(vp, NULL))
1549 					panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
1550 				/*
1551 				 * Put us back on the worklist.  The worklist
1552 				 * routine will remove us from our current
1553 				 * position and then add us back in at a later
1554 				 * position.
1555 				 */
1556 				VI_LOCK(vp);
1557 				vn_syncer_add_to_worklist(vp, syncdelay);
1558 				VI_UNLOCK(vp);
1559 			}
1560 			splx(s);
1561 		}
1562 
1563 		/*
1564 		 * Do soft update processing.
1565 		 */
1566 		if (softdep_process_worklist_hook != NULL)
1567 			(*softdep_process_worklist_hook)(NULL);
1568 
1569 		/*
1570 		 * The variable rushjob allows the kernel to speed up the
1571 		 * processing of the filesystem syncer process. A rushjob
1572 		 * value of N tells the filesystem syncer to process the next
1573 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1574 		 * is used by the soft update code to speed up the filesystem
1575 		 * syncer process when the incore state is getting so far
1576 		 * ahead of the disk that the kernel memory pool is being
1577 		 * threatened with exhaustion.
1578 		 */
1579 		if (rushjob > 0) {
1580 			rushjob -= 1;
1581 			continue;
1582 		}
1583 		/*
1584 		 * If it has taken us less than a second to process the
1585 		 * current work, then wait. Otherwise start right over
1586 		 * again. We can still lose time if any single round
1587 		 * takes more than two seconds, but it does not really
1588 		 * matter as we are just trying to generally pace the
1589 		 * filesystem activity.
1590 		 */
1591 		if (time_second == starttime)
1592 			tsleep(&lbolt, PPAUSE, "syncer", 0);
1593 	}
1594 }
1595 
1596 /*
1597  * Request the syncer daemon to speed up its work.
1598  * We never push it to speed up more than half of its
1599  * normal turn time, otherwise it could take over the cpu.
1600  * XXXKSE  only one update?
1601  */
1602 int
1603 speedup_syncer()
1604 {
1605 
1606 	mtx_lock_spin(&sched_lock);
1607 	if (FIRST_THREAD_IN_PROC(updateproc)->td_wchan == &lbolt) /* XXXKSE */
1608 		setrunnable(FIRST_THREAD_IN_PROC(updateproc));
1609 	mtx_unlock_spin(&sched_lock);
1610 	if (rushjob < syncdelay / 2) {
1611 		rushjob += 1;
1612 		stat_rush_requests += 1;
1613 		return (1);
1614 	}
1615 	return(0);
1616 }
1617 
1618 /*
1619  * Associate a p-buffer with a vnode.
1620  *
1621  * Also sets B_PAGING flag to indicate that vnode is not fully associated
1622  * with the buffer.  i.e. the bp has not been linked into the vnode or
1623  * ref-counted.
1624  */
1625 void
1626 pbgetvp(vp, bp)
1627 	register struct vnode *vp;
1628 	register struct buf *bp;
1629 {
1630 
1631 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1632 
1633 	bp->b_vp = vp;
1634 	bp->b_flags |= B_PAGING;
1635 	bp->b_dev = vn_todev(vp);
1636 }
1637 
1638 /*
1639  * Disassociate a p-buffer from a vnode.
1640  */
1641 void
1642 pbrelvp(bp)
1643 	register struct buf *bp;
1644 {
1645 
1646 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1647 
1648 	/* XXX REMOVE ME */
1649 	if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1650 		panic(
1651 		    "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1652 		    bp,
1653 		    (int)bp->b_flags
1654 		);
1655 	}
1656 	bp->b_vp = (struct vnode *) 0;
1657 	bp->b_flags &= ~B_PAGING;
1658 }
1659 
1660 /*
1661  * Reassign a buffer from one vnode to another.
1662  * Used to assign file specific control information
1663  * (indirect blocks) to the vnode to which they belong.
1664  */
1665 void
1666 reassignbuf(bp, newvp)
1667 	register struct buf *bp;
1668 	register struct vnode *newvp;
1669 {
1670 	int delay;
1671 	int s;
1672 
1673 	if (newvp == NULL) {
1674 		printf("reassignbuf: NULL");
1675 		return;
1676 	}
1677 	++reassignbufcalls;
1678 
1679 	/*
1680 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1681 	 * is not fully linked in.
1682 	 */
1683 	if (bp->b_flags & B_PAGING)
1684 		panic("cannot reassign paging buffer");
1685 
1686 	s = splbio();
1687 	/*
1688 	 * Delete from old vnode list, if on one.
1689 	 */
1690 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1691 		buf_vlist_remove(bp);
1692 		if (bp->b_vp != newvp) {
1693 			vdrop(bp->b_vp);
1694 			bp->b_vp = NULL;	/* for clarification */
1695 		}
1696 	}
1697 	/*
1698 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1699 	 * of clean buffers.
1700 	 */
1701 	if (bp->b_flags & B_DELWRI) {
1702 		VI_LOCK(newvp);
1703 		if ((newvp->v_iflag & VI_ONWORKLST) == 0) {
1704 			switch (newvp->v_type) {
1705 			case VDIR:
1706 				delay = dirdelay;
1707 				break;
1708 			case VCHR:
1709 				if (newvp->v_rdev->si_mountpoint != NULL) {
1710 					delay = metadelay;
1711 					break;
1712 				}
1713 				/* fall through */
1714 			default:
1715 				delay = filedelay;
1716 			}
1717 			vn_syncer_add_to_worklist(newvp, delay);
1718 		}
1719 		VI_UNLOCK(newvp);
1720 		buf_vlist_add(bp, newvp, BX_VNDIRTY);
1721 	} else {
1722 		buf_vlist_add(bp, newvp, BX_VNCLEAN);
1723 
1724 		VI_LOCK(newvp);
1725 		if ((newvp->v_iflag & VI_ONWORKLST) &&
1726 		    TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1727 			newvp->v_iflag &= ~VI_ONWORKLST;
1728 			LIST_REMOVE(newvp, v_synclist);
1729 		}
1730 		VI_UNLOCK(newvp);
1731 	}
1732 	if (bp->b_vp != newvp) {
1733 		bp->b_vp = newvp;
1734 		vhold(bp->b_vp);
1735 	}
1736 	splx(s);
1737 }
1738 
1739 /*
1740  * Create a vnode for a device.
1741  * Used for mounting the root filesystem.
1742  */
1743 int
1744 bdevvp(dev, vpp)
1745 	dev_t dev;
1746 	struct vnode **vpp;
1747 {
1748 	register struct vnode *vp;
1749 	struct vnode *nvp;
1750 	int error;
1751 
1752 	if (dev == NODEV) {
1753 		*vpp = NULLVP;
1754 		return (ENXIO);
1755 	}
1756 	if (vfinddev(dev, VCHR, vpp))
1757 		return (0);
1758 	error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1759 	if (error) {
1760 		*vpp = NULLVP;
1761 		return (error);
1762 	}
1763 	vp = nvp;
1764 	vp->v_type = VCHR;
1765 	addalias(vp, dev);
1766 	*vpp = vp;
1767 	return (0);
1768 }
1769 
1770 /*
1771  * Add vnode to the alias list hung off the dev_t.
1772  *
1773  * The reason for this gunk is that multiple vnodes can reference
1774  * the same physical device, so checking vp->v_usecount to see
1775  * how many users there are is inadequate; the v_usecount for
1776  * the vnodes need to be accumulated.  vcount() does that.
1777  */
1778 struct vnode *
1779 addaliasu(nvp, nvp_rdev)
1780 	struct vnode *nvp;
1781 	udev_t nvp_rdev;
1782 {
1783 	struct vnode *ovp;
1784 	vop_t **ops;
1785 	dev_t dev;
1786 
1787 	if (nvp->v_type == VBLK)
1788 		return (nvp);
1789 	if (nvp->v_type != VCHR)
1790 		panic("addaliasu on non-special vnode");
1791 	dev = udev2dev(nvp_rdev, 0);
1792 	/*
1793 	 * Check to see if we have a bdevvp vnode with no associated
1794 	 * filesystem. If so, we want to associate the filesystem of
1795 	 * the new newly instigated vnode with the bdevvp vnode and
1796 	 * discard the newly created vnode rather than leaving the
1797 	 * bdevvp vnode lying around with no associated filesystem.
1798 	 */
1799 	if (vfinddev(dev, nvp->v_type, &ovp) == 0 || ovp->v_data != NULL) {
1800 		addalias(nvp, dev);
1801 		return (nvp);
1802 	}
1803 	/*
1804 	 * Discard unneeded vnode, but save its node specific data.
1805 	 * Note that if there is a lock, it is carried over in the
1806 	 * node specific data to the replacement vnode.
1807 	 */
1808 	vref(ovp);
1809 	ovp->v_data = nvp->v_data;
1810 	ovp->v_tag = nvp->v_tag;
1811 	nvp->v_data = NULL;
1812 	lockinit(&ovp->v_lock, PVFS, nvp->v_lock.lk_wmesg,
1813 	    nvp->v_lock.lk_timo, nvp->v_lock.lk_flags & LK_EXTFLG_MASK);
1814 	if (nvp->v_vnlock)
1815 		ovp->v_vnlock = &ovp->v_lock;
1816 	ops = ovp->v_op;
1817 	ovp->v_op = nvp->v_op;
1818 	if (VOP_ISLOCKED(nvp, curthread)) {
1819 		VOP_UNLOCK(nvp, 0, curthread);
1820 		vn_lock(ovp, LK_EXCLUSIVE | LK_RETRY, curthread);
1821 	}
1822 	nvp->v_op = ops;
1823 	insmntque(ovp, nvp->v_mount);
1824 	vrele(nvp);
1825 	vgone(nvp);
1826 	return (ovp);
1827 }
1828 
1829 /* This is a local helper function that do the same as addaliasu, but for a
1830  * dev_t instead of an udev_t. */
1831 static void
1832 addalias(nvp, dev)
1833 	struct vnode *nvp;
1834 	dev_t dev;
1835 {
1836 
1837 	KASSERT(nvp->v_type == VCHR, ("addalias on non-special vnode"));
1838 	nvp->v_rdev = dev;
1839 	mtx_lock(&spechash_mtx);
1840 	SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
1841 	mtx_unlock(&spechash_mtx);
1842 }
1843 
1844 /*
1845  * Grab a particular vnode from the free list, increment its
1846  * reference count and lock it. The vnode lock bit is set if the
1847  * vnode is being eliminated in vgone. The process is awakened
1848  * when the transition is completed, and an error returned to
1849  * indicate that the vnode is no longer usable (possibly having
1850  * been changed to a new filesystem type).
1851  */
1852 int
1853 vget(vp, flags, td)
1854 	register struct vnode *vp;
1855 	int flags;
1856 	struct thread *td;
1857 {
1858 	int error;
1859 
1860 	/*
1861 	 * If the vnode is in the process of being cleaned out for
1862 	 * another use, we wait for the cleaning to finish and then
1863 	 * return failure. Cleaning is determined by checking that
1864 	 * the VI_XLOCK flag is set.
1865 	 */
1866 	if ((flags & LK_INTERLOCK) == 0)
1867 		VI_LOCK(vp);
1868 	if (vp->v_iflag & VI_XLOCK) {
1869 		if (vp->v_vxproc == curthread) {
1870 #if 0
1871 			/* this can now occur in normal operation */
1872 			log(LOG_INFO, "VXLOCK interlock avoided\n");
1873 #endif
1874 		} else {
1875 			vp->v_iflag |= VI_XWANT;
1876 			msleep(vp, VI_MTX(vp), PINOD | PDROP, "vget", 0);
1877 			return (ENOENT);
1878 		}
1879 	}
1880 
1881 	vp->v_usecount++;
1882 
1883 	if (VSHOULDBUSY(vp))
1884 		vbusy(vp);
1885 	if (flags & LK_TYPE_MASK) {
1886 		if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
1887 			/*
1888 			 * must expand vrele here because we do not want
1889 			 * to call VOP_INACTIVE if the reference count
1890 			 * drops back to zero since it was never really
1891 			 * active. We must remove it from the free list
1892 			 * before sleeping so that multiple processes do
1893 			 * not try to recycle it.
1894 			 */
1895 			VI_LOCK(vp);
1896 			vp->v_usecount--;
1897 			if (VSHOULDFREE(vp))
1898 				vfree(vp);
1899 			else
1900 				vlruvp(vp);
1901 			VI_UNLOCK(vp);
1902 		}
1903 		return (error);
1904 	}
1905 	VI_UNLOCK(vp);
1906 	return (0);
1907 }
1908 
1909 /*
1910  * Increase the reference count of a vnode.
1911  */
1912 void
1913 vref(struct vnode *vp)
1914 {
1915 	mtx_lock(&vp->v_interlock);
1916 	vp->v_usecount++;
1917 	mtx_unlock(&vp->v_interlock);
1918 }
1919 
1920 /*
1921  * Vnode put/release.
1922  * If count drops to zero, call inactive routine and return to freelist.
1923  */
1924 void
1925 vrele(vp)
1926 	struct vnode *vp;
1927 {
1928 	struct thread *td = curthread;	/* XXX */
1929 
1930 	KASSERT(vp != NULL, ("vrele: null vp"));
1931 
1932 	VI_LOCK(vp);
1933 
1934 	/* Skip this v_writecount check if we're going to panic below. */
1935 	KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
1936 	    ("vrele: missed vn_close"));
1937 
1938 	if (vp->v_usecount > 1) {
1939 
1940 		vp->v_usecount--;
1941 		VI_UNLOCK(vp);
1942 
1943 		return;
1944 	}
1945 
1946 	if (vp->v_usecount == 1) {
1947 		vp->v_usecount--;
1948 		/*
1949 		 * We must call VOP_INACTIVE with the node locked.
1950 		 * If we are doing a vput, the node is already locked,
1951 		 * but, in the case of vrele, we must explicitly lock
1952 		 * the vnode before calling VOP_INACTIVE.
1953 		 */
1954 		if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0)
1955 			VOP_INACTIVE(vp, td);
1956 		VI_LOCK(vp);
1957 		if (VSHOULDFREE(vp))
1958 			vfree(vp);
1959 		else
1960 			vlruvp(vp);
1961 		VI_UNLOCK(vp);
1962 
1963 	} else {
1964 #ifdef DIAGNOSTIC
1965 		vprint("vrele: negative ref count", vp);
1966 		VI_UNLOCK(vp);
1967 #endif
1968 		panic("vrele: negative ref cnt");
1969 	}
1970 }
1971 
1972 /*
1973  * Release an already locked vnode.  This give the same effects as
1974  * unlock+vrele(), but takes less time and avoids releasing and
1975  * re-aquiring the lock (as vrele() aquires the lock internally.)
1976  */
1977 void
1978 vput(vp)
1979 	struct vnode *vp;
1980 {
1981 	struct thread *td = curthread;	/* XXX */
1982 
1983 	GIANT_REQUIRED;
1984 
1985 	KASSERT(vp != NULL, ("vput: null vp"));
1986 	mtx_lock(&vp->v_interlock);
1987 	/* Skip this v_writecount check if we're going to panic below. */
1988 	KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
1989 	    ("vput: missed vn_close"));
1990 
1991 	if (vp->v_usecount > 1) {
1992 		vp->v_usecount--;
1993 		VOP_UNLOCK(vp, LK_INTERLOCK, td);
1994 		return;
1995 	}
1996 
1997 	if (vp->v_usecount == 1) {
1998 		vp->v_usecount--;
1999 		/*
2000 		 * We must call VOP_INACTIVE with the node locked.
2001 		 * If we are doing a vput, the node is already locked,
2002 		 * so we just need to release the vnode mutex.
2003 		 */
2004 		VI_UNLOCK(vp);
2005 		VOP_INACTIVE(vp, td);
2006 		VI_LOCK(vp);
2007 		if (VSHOULDFREE(vp))
2008 			vfree(vp);
2009 		else
2010 			vlruvp(vp);
2011 		VI_UNLOCK(vp);
2012 
2013 	} else {
2014 #ifdef DIAGNOSTIC
2015 		vprint("vput: negative ref count", vp);
2016 #endif
2017 		panic("vput: negative ref cnt");
2018 	}
2019 }
2020 
2021 /*
2022  * Somebody doesn't want the vnode recycled.
2023  */
2024 void
2025 vhold(vp)
2026 	register struct vnode *vp;
2027 {
2028 	int s;
2029 
2030 	s = splbio();
2031 	vp->v_holdcnt++;
2032 	VI_LOCK(vp);
2033 	if (VSHOULDBUSY(vp))
2034 		vbusy(vp);
2035 	VI_UNLOCK(vp);
2036 	splx(s);
2037 }
2038 
2039 /*
2040  * Note that there is one less who cares about this vnode.  vdrop() is the
2041  * opposite of vhold().
2042  */
2043 void
2044 vdrop(vp)
2045 	register struct vnode *vp;
2046 {
2047 	int s;
2048 
2049 	s = splbio();
2050 	if (vp->v_holdcnt <= 0)
2051 		panic("vdrop: holdcnt");
2052 	vp->v_holdcnt--;
2053 	VI_LOCK(vp);
2054 	if (VSHOULDFREE(vp))
2055 		vfree(vp);
2056 	else
2057 		vlruvp(vp);
2058 	VI_UNLOCK(vp);
2059 	splx(s);
2060 }
2061 
2062 /*
2063  * Remove any vnodes in the vnode table belonging to mount point mp.
2064  *
2065  * If FORCECLOSE is not specified, there should not be any active ones,
2066  * return error if any are found (nb: this is a user error, not a
2067  * system error). If FORCECLOSE is specified, detach any active vnodes
2068  * that are found.
2069  *
2070  * If WRITECLOSE is set, only flush out regular file vnodes open for
2071  * writing.
2072  *
2073  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2074  *
2075  * `rootrefs' specifies the base reference count for the root vnode
2076  * of this filesystem. The root vnode is considered busy if its
2077  * v_usecount exceeds this value. On a successful return, vflush()
2078  * will call vrele() on the root vnode exactly rootrefs times.
2079  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2080  * be zero.
2081  */
2082 #ifdef DIAGNOSTIC
2083 static int busyprt = 0;		/* print out busy vnodes */
2084 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
2085 #endif
2086 
2087 int
2088 vflush(mp, rootrefs, flags)
2089 	struct mount *mp;
2090 	int rootrefs;
2091 	int flags;
2092 {
2093 	struct thread *td = curthread;	/* XXX */
2094 	struct vnode *vp, *nvp, *rootvp = NULL;
2095 	struct vattr vattr;
2096 	int busy = 0, error;
2097 
2098 	if (rootrefs > 0) {
2099 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2100 		    ("vflush: bad args"));
2101 		/*
2102 		 * Get the filesystem root vnode. We can vput() it
2103 		 * immediately, since with rootrefs > 0, it won't go away.
2104 		 */
2105 		if ((error = VFS_ROOT(mp, &rootvp)) != 0)
2106 			return (error);
2107 		vput(rootvp);
2108 
2109 	}
2110 	mtx_lock(&mntvnode_mtx);
2111 loop:
2112 	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) {
2113 		/*
2114 		 * Make sure this vnode wasn't reclaimed in getnewvnode().
2115 		 * Start over if it has (it won't be on the list anymore).
2116 		 */
2117 		if (vp->v_mount != mp)
2118 			goto loop;
2119 		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2120 
2121 		mtx_unlock(&mntvnode_mtx);
2122 		VI_LOCK(vp);
2123 		/*
2124 		 * Skip over a vnodes marked VV_SYSTEM.
2125 		 */
2126 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2127 			VI_UNLOCK(vp);
2128 			mtx_lock(&mntvnode_mtx);
2129 			continue;
2130 		}
2131 		/*
2132 		 * If WRITECLOSE is set, flush out unlinked but still open
2133 		 * files (even if open only for reading) and regular file
2134 		 * vnodes open for writing.
2135 		 */
2136 		mp_fixme("Getattr called with interlock held!");
2137 		if ((flags & WRITECLOSE) &&
2138 		    (vp->v_type == VNON ||
2139 		    (VOP_GETATTR(vp, &vattr, td->td_ucred, td) == 0 &&
2140 		    vattr.va_nlink > 0)) &&
2141 		    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2142 			mtx_unlock(&vp->v_interlock);
2143 			mtx_lock(&mntvnode_mtx);
2144 			continue;
2145 		}
2146 
2147 		/*
2148 		 * With v_usecount == 0, all we need to do is clear out the
2149 		 * vnode data structures and we are done.
2150 		 */
2151 		if (vp->v_usecount == 0) {
2152 			vgonel(vp, td);
2153 			mtx_lock(&mntvnode_mtx);
2154 			continue;
2155 		}
2156 
2157 		/*
2158 		 * If FORCECLOSE is set, forcibly close the vnode. For block
2159 		 * or character devices, revert to an anonymous device. For
2160 		 * all other files, just kill them.
2161 		 */
2162 		if (flags & FORCECLOSE) {
2163 			if (vp->v_type != VCHR) {
2164 				vgonel(vp, td);
2165 			} else {
2166 				vclean(vp, 0, td);
2167 				VI_UNLOCK(vp);
2168 				vp->v_op = spec_vnodeop_p;
2169 				insmntque(vp, (struct mount *) 0);
2170 			}
2171 			mtx_lock(&mntvnode_mtx);
2172 			continue;
2173 		}
2174 #ifdef DIAGNOSTIC
2175 		if (busyprt)
2176 			vprint("vflush: busy vnode", vp);
2177 #endif
2178 		VI_UNLOCK(vp);
2179 		mtx_lock(&mntvnode_mtx);
2180 		busy++;
2181 	}
2182 	mtx_unlock(&mntvnode_mtx);
2183 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2184 		/*
2185 		 * If just the root vnode is busy, and if its refcount
2186 		 * is equal to `rootrefs', then go ahead and kill it.
2187 		 */
2188 		mtx_lock(&rootvp->v_interlock);
2189 		KASSERT(busy > 0, ("vflush: not busy"));
2190 		KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
2191 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2192 			vgonel(rootvp, td);
2193 			busy = 0;
2194 		} else
2195 			mtx_unlock(&rootvp->v_interlock);
2196 	}
2197 	if (busy)
2198 		return (EBUSY);
2199 	for (; rootrefs > 0; rootrefs--)
2200 		vrele(rootvp);
2201 	return (0);
2202 }
2203 
2204 /*
2205  * This moves a now (likely recyclable) vnode to the end of the
2206  * mountlist.  XXX However, it is temporarily disabled until we
2207  * can clean up ffs_sync() and friends, which have loop restart
2208  * conditions which this code causes to operate O(N^2).
2209  */
2210 static void
2211 vlruvp(struct vnode *vp)
2212 {
2213 #if 0
2214 	struct mount *mp;
2215 
2216 	if ((mp = vp->v_mount) != NULL) {
2217 		mtx_lock(&mntvnode_mtx);
2218 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
2219 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
2220 		mtx_unlock(&mntvnode_mtx);
2221 	}
2222 #endif
2223 }
2224 
2225 /*
2226  * Disassociate the underlying filesystem from a vnode.
2227  */
2228 static void
2229 vclean(vp, flags, td)
2230 	struct vnode *vp;
2231 	int flags;
2232 	struct thread *td;
2233 {
2234 	int active;
2235 
2236 	mtx_assert(VI_MTX(vp), MA_OWNED);
2237 	/*
2238 	 * Check to see if the vnode is in use. If so we have to reference it
2239 	 * before we clean it out so that its count cannot fall to zero and
2240 	 * generate a race against ourselves to recycle it.
2241 	 */
2242 	if ((active = vp->v_usecount))
2243 		vp->v_usecount++;
2244 
2245 	/*
2246 	 * Prevent the vnode from being recycled or brought into use while we
2247 	 * clean it out.
2248 	 */
2249 	if (vp->v_iflag & VI_XLOCK)
2250 		panic("vclean: deadlock");
2251 	vp->v_iflag |= VI_XLOCK;
2252 	vp->v_vxproc = curthread;
2253 	/*
2254 	 * Even if the count is zero, the VOP_INACTIVE routine may still
2255 	 * have the object locked while it cleans it out. The VOP_LOCK
2256 	 * ensures that the VOP_INACTIVE routine is done with its work.
2257 	 * For active vnodes, it ensures that no other activity can
2258 	 * occur while the underlying object is being cleaned out.
2259 	 */
2260 	VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td);
2261 
2262 	/*
2263 	 * Clean out any buffers associated with the vnode.
2264 	 * If the flush fails, just toss the buffers.
2265 	 */
2266 	if (flags & DOCLOSE) {
2267 		if (TAILQ_FIRST(&vp->v_dirtyblkhd) != NULL)
2268 			(void) vn_write_suspend_wait(vp, NULL, V_WAIT);
2269 		if (vinvalbuf(vp, V_SAVE, NOCRED, td, 0, 0) != 0)
2270 			vinvalbuf(vp, 0, NOCRED, td, 0, 0);
2271 	}
2272 
2273 	VOP_DESTROYVOBJECT(vp);
2274 
2275 	/*
2276 	 * Any other processes trying to obtain this lock must first
2277 	 * wait for VXLOCK to clear, then call the new lock operation.
2278 	 */
2279 	VOP_UNLOCK(vp, 0, td);
2280 
2281 	/*
2282 	 * If purging an active vnode, it must be closed and
2283 	 * deactivated before being reclaimed. Note that the
2284 	 * VOP_INACTIVE will unlock the vnode.
2285 	 */
2286 	if (active) {
2287 		if (flags & DOCLOSE)
2288 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2289 		if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT, td) != 0)
2290 			panic("vclean: cannot relock.");
2291 		VOP_INACTIVE(vp, td);
2292 	}
2293 
2294 	/*
2295 	 * Reclaim the vnode.
2296 	 */
2297 	if (VOP_RECLAIM(vp, td))
2298 		panic("vclean: cannot reclaim");
2299 
2300 	if (active) {
2301 		/*
2302 		 * Inline copy of vrele() since VOP_INACTIVE
2303 		 * has already been called.
2304 		 */
2305 		VI_LOCK(vp);
2306 		if (--vp->v_usecount <= 0) {
2307 #ifdef DIAGNOSTIC
2308 			if (vp->v_usecount < 0 || vp->v_writecount != 0) {
2309 				vprint("vclean: bad ref count", vp);
2310 				panic("vclean: ref cnt");
2311 			}
2312 #endif
2313 			vfree(vp);
2314 		}
2315 		VI_UNLOCK(vp);
2316 	}
2317 
2318 	cache_purge(vp);
2319 	vp->v_vnlock = NULL;
2320 	lockdestroy(&vp->v_lock);
2321 
2322 	VI_LOCK(vp);
2323 	if (VSHOULDFREE(vp))
2324 		vfree(vp);
2325 
2326 	/*
2327 	 * Done with purge, notify sleepers of the grim news.
2328 	 */
2329 	vp->v_op = dead_vnodeop_p;
2330 	if (vp->v_pollinfo != NULL)
2331 		vn_pollgone(vp);
2332 	vp->v_tag = VT_NON;
2333 	vp->v_iflag &= ~VI_XLOCK;
2334 	vp->v_vxproc = NULL;
2335 	if (vp->v_iflag & VI_XWANT) {
2336 		vp->v_iflag &= ~VI_XWANT;
2337 		wakeup(vp);
2338 	}
2339 }
2340 
2341 /*
2342  * Eliminate all activity associated with the requested vnode
2343  * and with all vnodes aliased to the requested vnode.
2344  */
2345 int
2346 vop_revoke(ap)
2347 	struct vop_revoke_args /* {
2348 		struct vnode *a_vp;
2349 		int a_flags;
2350 	} */ *ap;
2351 {
2352 	struct vnode *vp, *vq;
2353 	dev_t dev;
2354 
2355 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
2356 
2357 	vp = ap->a_vp;
2358 	VI_LOCK(vp);
2359 	/*
2360 	 * If a vgone (or vclean) is already in progress,
2361 	 * wait until it is done and return.
2362 	 */
2363 	if (vp->v_iflag & VI_XLOCK) {
2364 		vp->v_iflag |= VI_XWANT;
2365 		msleep(vp, VI_MTX(vp), PINOD | PDROP,
2366 		    "vop_revokeall", 0);
2367 		VI_UNLOCK(vp);
2368 		return (0);
2369 	}
2370 	VI_UNLOCK(vp);
2371 	dev = vp->v_rdev;
2372 	for (;;) {
2373 		mtx_lock(&spechash_mtx);
2374 		vq = SLIST_FIRST(&dev->si_hlist);
2375 		mtx_unlock(&spechash_mtx);
2376 		if (!vq)
2377 			break;
2378 		vgone(vq);
2379 	}
2380 	return (0);
2381 }
2382 
2383 /*
2384  * Recycle an unused vnode to the front of the free list.
2385  * Release the passed interlock if the vnode will be recycled.
2386  */
2387 int
2388 vrecycle(vp, inter_lkp, td)
2389 	struct vnode *vp;
2390 	struct mtx *inter_lkp;
2391 	struct thread *td;
2392 {
2393 
2394 	mtx_lock(&vp->v_interlock);
2395 	if (vp->v_usecount == 0) {
2396 		if (inter_lkp) {
2397 			mtx_unlock(inter_lkp);
2398 		}
2399 		vgonel(vp, td);
2400 		return (1);
2401 	}
2402 	mtx_unlock(&vp->v_interlock);
2403 	return (0);
2404 }
2405 
2406 /*
2407  * Eliminate all activity associated with a vnode
2408  * in preparation for reuse.
2409  */
2410 void
2411 vgone(vp)
2412 	register struct vnode *vp;
2413 {
2414 	struct thread *td = curthread;	/* XXX */
2415 
2416 	VI_LOCK(vp);
2417 	vgonel(vp, td);
2418 }
2419 
2420 /*
2421  * vgone, with the vp interlock held.
2422  */
2423 void
2424 vgonel(vp, td)
2425 	struct vnode *vp;
2426 	struct thread *td;
2427 {
2428 	int s;
2429 
2430 	/*
2431 	 * If a vgone (or vclean) is already in progress,
2432 	 * wait until it is done and return.
2433 	 */
2434 	mtx_assert(VI_MTX(vp), MA_OWNED);
2435 	if (vp->v_iflag & VI_XLOCK) {
2436 		vp->v_iflag |= VI_XWANT;
2437 		VI_UNLOCK(vp);
2438 		tsleep(vp, PINOD | PDROP, "vgone", 0);
2439 		return;
2440 	}
2441 
2442 	/*
2443 	 * Clean out the filesystem specific data.
2444 	 */
2445 	vclean(vp, DOCLOSE, td);
2446 	VI_UNLOCK(vp);
2447 
2448 	/*
2449 	 * Delete from old mount point vnode list, if on one.
2450 	 */
2451 	if (vp->v_mount != NULL)
2452 		insmntque(vp, (struct mount *)0);
2453 	/*
2454 	 * If special device, remove it from special device alias list
2455 	 * if it is on one.
2456 	 */
2457 	if (vp->v_type == VCHR && vp->v_rdev != NULL && vp->v_rdev != NODEV) {
2458 		mtx_lock(&spechash_mtx);
2459 		SLIST_REMOVE(&vp->v_rdev->si_hlist, vp, vnode, v_specnext);
2460 		freedev(vp->v_rdev);
2461 		mtx_unlock(&spechash_mtx);
2462 		vp->v_rdev = NULL;
2463 	}
2464 
2465 	/*
2466 	 * If it is on the freelist and not already at the head,
2467 	 * move it to the head of the list. The test of the
2468 	 * VDOOMED flag and the reference count of zero is because
2469 	 * it will be removed from the free list by getnewvnode,
2470 	 * but will not have its reference count incremented until
2471 	 * after calling vgone. If the reference count were
2472 	 * incremented first, vgone would (incorrectly) try to
2473 	 * close the previous instance of the underlying object.
2474 	 */
2475 	VI_LOCK(vp);
2476 	if (vp->v_usecount == 0 && !(vp->v_iflag & VI_DOOMED)) {
2477 		s = splbio();
2478 		mtx_lock(&vnode_free_list_mtx);
2479 		if (vp->v_iflag & VI_FREE) {
2480 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2481 		} else {
2482 			vp->v_iflag |= VI_FREE;
2483 			freevnodes++;
2484 		}
2485 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2486 		mtx_unlock(&vnode_free_list_mtx);
2487 		splx(s);
2488 	}
2489 
2490 	vp->v_type = VBAD;
2491 	VI_UNLOCK(vp);
2492 }
2493 
2494 /*
2495  * Lookup a vnode by device number.
2496  */
2497 int
2498 vfinddev(dev, type, vpp)
2499 	dev_t dev;
2500 	enum vtype type;
2501 	struct vnode **vpp;
2502 {
2503 	struct vnode *vp;
2504 
2505 	mtx_lock(&spechash_mtx);
2506 	SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2507 		if (type == vp->v_type) {
2508 			*vpp = vp;
2509 			mtx_unlock(&spechash_mtx);
2510 			return (1);
2511 		}
2512 	}
2513 	mtx_unlock(&spechash_mtx);
2514 	return (0);
2515 }
2516 
2517 /*
2518  * Calculate the total number of references to a special device.
2519  */
2520 int
2521 vcount(vp)
2522 	struct vnode *vp;
2523 {
2524 	struct vnode *vq;
2525 	int count;
2526 
2527 	count = 0;
2528 	mtx_lock(&spechash_mtx);
2529 	SLIST_FOREACH(vq, &vp->v_rdev->si_hlist, v_specnext)
2530 		count += vq->v_usecount;
2531 	mtx_unlock(&spechash_mtx);
2532 	return (count);
2533 }
2534 
2535 /*
2536  * Same as above, but using the dev_t as argument
2537  */
2538 int
2539 count_dev(dev)
2540 	dev_t dev;
2541 {
2542 	struct vnode *vp;
2543 
2544 	vp = SLIST_FIRST(&dev->si_hlist);
2545 	if (vp == NULL)
2546 		return (0);
2547 	return(vcount(vp));
2548 }
2549 
2550 /*
2551  * Print out a description of a vnode.
2552  */
2553 static char *typename[] =
2554 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2555 
2556 void
2557 vprint(label, vp)
2558 	char *label;
2559 	struct vnode *vp;
2560 {
2561 	char buf[96];
2562 
2563 	if (label != NULL)
2564 		printf("%s: %p: ", label, (void *)vp);
2565 	else
2566 		printf("%p: ", (void *)vp);
2567 	printf("type %s, usecount %d, writecount %d, refcount %d,",
2568 	    typename[vp->v_type], vp->v_usecount, vp->v_writecount,
2569 	    vp->v_holdcnt);
2570 	buf[0] = '\0';
2571 	if (vp->v_vflag & VV_ROOT)
2572 		strcat(buf, "|VV_ROOT");
2573 	if (vp->v_vflag & VV_TEXT)
2574 		strcat(buf, "|VV_TEXT");
2575 	if (vp->v_vflag & VV_SYSTEM)
2576 		strcat(buf, "|VV_SYSTEM");
2577 	if (vp->v_iflag & VI_XLOCK)
2578 		strcat(buf, "|VI_XLOCK");
2579 	if (vp->v_iflag & VI_XWANT)
2580 		strcat(buf, "|VI_XWANT");
2581 	if (vp->v_iflag & VI_BWAIT)
2582 		strcat(buf, "|VI_BWAIT");
2583 	if (vp->v_iflag & VI_DOOMED)
2584 		strcat(buf, "|VI_DOOMED");
2585 	if (vp->v_iflag & VI_FREE)
2586 		strcat(buf, "|VI_FREE");
2587 	if (vp->v_vflag & VV_OBJBUF)
2588 		strcat(buf, "|VV_OBJBUF");
2589 	if (buf[0] != '\0')
2590 		printf(" flags (%s)", &buf[1]);
2591 	if (vp->v_data == NULL) {
2592 		printf("\n");
2593 	} else {
2594 		printf("\n\t");
2595 		VOP_PRINT(vp);
2596 	}
2597 }
2598 
2599 #ifdef DDB
2600 #include <ddb/ddb.h>
2601 /*
2602  * List all of the locked vnodes in the system.
2603  * Called when debugging the kernel.
2604  */
2605 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2606 {
2607 	struct thread *td = curthread;	/* XXX */
2608 	struct mount *mp, *nmp;
2609 	struct vnode *vp;
2610 
2611 	printf("Locked vnodes\n");
2612 	mtx_lock(&mountlist_mtx);
2613 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2614 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
2615 			nmp = TAILQ_NEXT(mp, mnt_list);
2616 			continue;
2617 		}
2618 		mtx_lock(&mntvnode_mtx);
2619 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2620 			if (VOP_ISLOCKED(vp, NULL))
2621 				vprint((char *)0, vp);
2622 		}
2623 		mtx_unlock(&mntvnode_mtx);
2624 		mtx_lock(&mountlist_mtx);
2625 		nmp = TAILQ_NEXT(mp, mnt_list);
2626 		vfs_unbusy(mp, td);
2627 	}
2628 	mtx_unlock(&mountlist_mtx);
2629 }
2630 #endif
2631 
2632 /*
2633  * Fill in a struct xvfsconf based on a struct vfsconf.
2634  */
2635 static void
2636 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2637 {
2638 
2639 	strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2640 	xvfsp->vfc_typenum = vfsp->vfc_typenum;
2641 	xvfsp->vfc_refcount = vfsp->vfc_refcount;
2642 	xvfsp->vfc_flags = vfsp->vfc_flags;
2643 	/*
2644 	 * These are unused in userland, we keep them
2645 	 * to not break binary compatibility.
2646 	 */
2647 	xvfsp->vfc_vfsops = NULL;
2648 	xvfsp->vfc_next = NULL;
2649 }
2650 
2651 static int
2652 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2653 {
2654 	struct vfsconf *vfsp;
2655 	struct xvfsconf *xvfsp;
2656 	int cnt, error, i;
2657 
2658 	cnt = 0;
2659 	for (vfsp = vfsconf; vfsp != NULL; vfsp = vfsp->vfc_next)
2660 		cnt++;
2661 	xvfsp = malloc(sizeof(struct xvfsconf) * cnt, M_TEMP, M_WAITOK);
2662 	/*
2663 	 * Handle the race that we will have here when struct vfsconf
2664 	 * will be locked down by using both cnt and checking vfc_next
2665 	 * against NULL to determine the end of the loop.  The race will
2666 	 * happen because we will have to unlock before calling malloc().
2667 	 * We are protected by Giant for now.
2668 	 */
2669 	i = 0;
2670 	for (vfsp = vfsconf; vfsp != NULL && i < cnt; vfsp = vfsp->vfc_next) {
2671 		vfsconf2x(vfsp, xvfsp + i);
2672 		i++;
2673 	}
2674 	error = SYSCTL_OUT(req, xvfsp, sizeof(struct xvfsconf) * i);
2675 	free(xvfsp, M_TEMP);
2676 	return (error);
2677 }
2678 
2679 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
2680     "S,xvfsconf", "List of all configured filesystems");
2681 
2682 /*
2683  * Top level filesystem related information gathering.
2684  */
2685 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2686 
2687 static int
2688 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2689 {
2690 	int *name = (int *)arg1 - 1;	/* XXX */
2691 	u_int namelen = arg2 + 1;	/* XXX */
2692 	struct vfsconf *vfsp;
2693 	struct xvfsconf xvfsp;
2694 
2695 	printf("WARNING: userland calling deprecated sysctl, "
2696 	    "please rebuild world\n");
2697 
2698 #if 1 || defined(COMPAT_PRELITE2)
2699 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2700 	if (namelen == 1)
2701 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2702 #endif
2703 
2704 	switch (name[1]) {
2705 	case VFS_MAXTYPENUM:
2706 		if (namelen != 2)
2707 			return (ENOTDIR);
2708 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2709 	case VFS_CONF:
2710 		if (namelen != 3)
2711 			return (ENOTDIR);	/* overloaded */
2712 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2713 			if (vfsp->vfc_typenum == name[2])
2714 				break;
2715 		if (vfsp == NULL)
2716 			return (EOPNOTSUPP);
2717 		vfsconf2x(vfsp, &xvfsp);
2718 		return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
2719 	}
2720 	return (EOPNOTSUPP);
2721 }
2722 
2723 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP, vfs_sysctl,
2724 	"Generic filesystem");
2725 
2726 #if 1 || defined(COMPAT_PRELITE2)
2727 
2728 static int
2729 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2730 {
2731 	int error;
2732 	struct vfsconf *vfsp;
2733 	struct ovfsconf ovfs;
2734 
2735 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2736 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
2737 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
2738 		ovfs.vfc_index = vfsp->vfc_typenum;
2739 		ovfs.vfc_refcount = vfsp->vfc_refcount;
2740 		ovfs.vfc_flags = vfsp->vfc_flags;
2741 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2742 		if (error)
2743 			return error;
2744 	}
2745 	return 0;
2746 }
2747 
2748 #endif /* 1 || COMPAT_PRELITE2 */
2749 
2750 #define KINFO_VNODESLOP		10
2751 /*
2752  * Dump vnode list (via sysctl).
2753  */
2754 /* ARGSUSED */
2755 static int
2756 sysctl_vnode(SYSCTL_HANDLER_ARGS)
2757 {
2758 	struct xvnode *xvn;
2759 	struct thread *td = req->td;
2760 	struct mount *mp;
2761 	struct vnode *vp;
2762 	int error, len, n;
2763 
2764 	req->lock = 0;
2765 	len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
2766 	if (!req->oldptr)
2767 		/* Make an estimate */
2768 		return (SYSCTL_OUT(req, 0, len));
2769 
2770 	sysctl_wire_old_buffer(req, 0);
2771 	xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
2772 	n = 0;
2773 	mtx_lock(&mountlist_mtx);
2774 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2775 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td))
2776 			continue;
2777 		mtx_lock(&mntvnode_mtx);
2778 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2779 			if (n == len)
2780 				break;
2781 			vref(vp);
2782 			xvn[n].xv_size = sizeof *xvn;
2783 			xvn[n].xv_vnode = vp;
2784 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
2785 			XV_COPY(usecount);
2786 			XV_COPY(writecount);
2787 			XV_COPY(holdcnt);
2788 			XV_COPY(id);
2789 			XV_COPY(mount);
2790 			XV_COPY(numoutput);
2791 			XV_COPY(type);
2792 #undef XV_COPY
2793 			xvn[n].xv_flag = vp->v_vflag;
2794 
2795 			switch (vp->v_type) {
2796 			case VREG:
2797 			case VDIR:
2798 			case VLNK:
2799 				xvn[n].xv_dev = vp->v_cachedfs;
2800 				xvn[n].xv_ino = vp->v_cachedid;
2801 				break;
2802 			case VBLK:
2803 			case VCHR:
2804 				if (vp->v_rdev == NULL) {
2805 					vrele(vp);
2806 					continue;
2807 				}
2808 				xvn[n].xv_dev = dev2udev(vp->v_rdev);
2809 				break;
2810 			case VSOCK:
2811 				xvn[n].xv_socket = vp->v_socket;
2812 				break;
2813 			case VFIFO:
2814 				xvn[n].xv_fifo = vp->v_fifoinfo;
2815 				break;
2816 			case VNON:
2817 			case VBAD:
2818 			default:
2819 				/* shouldn't happen? */
2820 				vrele(vp);
2821 				continue;
2822 			}
2823 			vrele(vp);
2824 			++n;
2825 		}
2826 		mtx_unlock(&mntvnode_mtx);
2827 		mtx_lock(&mountlist_mtx);
2828 		vfs_unbusy(mp, td);
2829 		if (n == len)
2830 			break;
2831 	}
2832 	mtx_unlock(&mountlist_mtx);
2833 
2834 	error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
2835 	free(xvn, M_TEMP);
2836 	return (error);
2837 }
2838 
2839 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2840 	0, 0, sysctl_vnode, "S,xvnode", "");
2841 
2842 /*
2843  * Check to see if a filesystem is mounted on a block device.
2844  */
2845 int
2846 vfs_mountedon(vp)
2847 	struct vnode *vp;
2848 {
2849 
2850 	if (vp->v_rdev->si_mountpoint != NULL)
2851 		return (EBUSY);
2852 	return (0);
2853 }
2854 
2855 /*
2856  * Unmount all filesystems. The list is traversed in reverse order
2857  * of mounting to avoid dependencies.
2858  */
2859 void
2860 vfs_unmountall()
2861 {
2862 	struct mount *mp;
2863 	struct thread *td;
2864 	int error;
2865 
2866 	if (curthread != NULL)
2867 		td = curthread;
2868 	else
2869 		td = FIRST_THREAD_IN_PROC(initproc); /* XXX XXX proc0? */
2870 	/*
2871 	 * Since this only runs when rebooting, it is not interlocked.
2872 	 */
2873 	while(!TAILQ_EMPTY(&mountlist)) {
2874 		mp = TAILQ_LAST(&mountlist, mntlist);
2875 		error = dounmount(mp, MNT_FORCE, td);
2876 		if (error) {
2877 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
2878 			printf("unmount of %s failed (",
2879 			    mp->mnt_stat.f_mntonname);
2880 			if (error == EBUSY)
2881 				printf("BUSY)\n");
2882 			else
2883 				printf("%d)\n", error);
2884 		} else {
2885 			/* The unmount has removed mp from the mountlist */
2886 		}
2887 	}
2888 }
2889 
2890 /*
2891  * perform msync on all vnodes under a mount point
2892  * the mount point must be locked.
2893  */
2894 void
2895 vfs_msync(struct mount *mp, int flags)
2896 {
2897 	struct vnode *vp, *nvp;
2898 	struct vm_object *obj;
2899 	int tries;
2900 
2901 	GIANT_REQUIRED;
2902 
2903 	tries = 5;
2904 	mtx_lock(&mntvnode_mtx);
2905 loop:
2906 	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) {
2907 		if (vp->v_mount != mp) {
2908 			if (--tries > 0)
2909 				goto loop;
2910 			break;
2911 		}
2912 		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2913 
2914 		mp_fixme("What locks do we need here?");
2915 		if (vp->v_iflag & VI_XLOCK)	/* XXX: what if MNT_WAIT? */
2916 			continue;
2917 
2918 		if (vp->v_vflag & VV_NOSYNC)	/* unlinked, skip it */
2919 			continue;
2920 
2921 		if ((vp->v_iflag & VI_OBJDIRTY) &&
2922 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
2923 			mtx_unlock(&mntvnode_mtx);
2924 			if (!vget(vp,
2925 			    LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, curthread)) {
2926 				if (VOP_GETVOBJECT(vp, &obj) == 0) {
2927 					vm_object_page_clean(obj, 0, 0,
2928 					    flags == MNT_WAIT ?
2929 					    OBJPC_SYNC : OBJPC_NOSYNC);
2930 				}
2931 				vput(vp);
2932 			}
2933 			mtx_lock(&mntvnode_mtx);
2934 			if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) {
2935 				if (--tries > 0)
2936 					goto loop;
2937 				break;
2938 			}
2939 		}
2940 	}
2941 	mtx_unlock(&mntvnode_mtx);
2942 }
2943 
2944 /*
2945  * Create the VM object needed for VMIO and mmap support.  This
2946  * is done for all VREG files in the system.  Some filesystems might
2947  * afford the additional metadata buffering capability of the
2948  * VMIO code by making the device node be VMIO mode also.
2949  *
2950  * vp must be locked when vfs_object_create is called.
2951  */
2952 int
2953 vfs_object_create(vp, td, cred)
2954 	struct vnode *vp;
2955 	struct thread *td;
2956 	struct ucred *cred;
2957 {
2958 	GIANT_REQUIRED;
2959 	return (VOP_CREATEVOBJECT(vp, cred, td));
2960 }
2961 
2962 /*
2963  * Mark a vnode as free, putting it up for recycling.
2964  */
2965 void
2966 vfree(vp)
2967 	struct vnode *vp;
2968 {
2969 	int s;
2970 
2971 	mtx_assert(VI_MTX(vp), MA_OWNED);
2972 	s = splbio();
2973 	mtx_lock(&vnode_free_list_mtx);
2974 	KASSERT((vp->v_iflag & VI_FREE) == 0, ("vnode already free"));
2975 	if (vp->v_iflag & VI_AGE) {
2976 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2977 	} else {
2978 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2979 	}
2980 	freevnodes++;
2981 	mtx_unlock(&vnode_free_list_mtx);
2982 	vp->v_iflag &= ~VI_AGE;
2983 	vp->v_iflag |= VI_FREE;
2984 	splx(s);
2985 }
2986 
2987 /*
2988  * Opposite of vfree() - mark a vnode as in use.
2989  */
2990 void
2991 vbusy(vp)
2992 	struct vnode *vp;
2993 {
2994 	int s;
2995 
2996 	s = splbio();
2997 	mtx_assert(VI_MTX(vp), MA_OWNED);
2998 	mtx_lock(&vnode_free_list_mtx);
2999 	KASSERT((vp->v_iflag & VI_FREE) != 0, ("vnode not free"));
3000 	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3001 	freevnodes--;
3002 	mtx_unlock(&vnode_free_list_mtx);
3003 	vp->v_iflag &= ~(VI_FREE|VI_AGE);
3004 	splx(s);
3005 }
3006 
3007 /*
3008  * Record a process's interest in events which might happen to
3009  * a vnode.  Because poll uses the historic select-style interface
3010  * internally, this routine serves as both the ``check for any
3011  * pending events'' and the ``record my interest in future events''
3012  * functions.  (These are done together, while the lock is held,
3013  * to avoid race conditions.)
3014  */
3015 int
3016 vn_pollrecord(vp, td, events)
3017 	struct vnode *vp;
3018 	struct thread *td;
3019 	short events;
3020 {
3021 
3022 	if (vp->v_pollinfo == NULL)
3023 		v_addpollinfo(vp);
3024 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3025 	if (vp->v_pollinfo->vpi_revents & events) {
3026 		/*
3027 		 * This leaves events we are not interested
3028 		 * in available for the other process which
3029 		 * which presumably had requested them
3030 		 * (otherwise they would never have been
3031 		 * recorded).
3032 		 */
3033 		events &= vp->v_pollinfo->vpi_revents;
3034 		vp->v_pollinfo->vpi_revents &= ~events;
3035 
3036 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
3037 		return events;
3038 	}
3039 	vp->v_pollinfo->vpi_events |= events;
3040 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3041 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3042 	return 0;
3043 }
3044 
3045 /*
3046  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
3047  * it is possible for us to miss an event due to race conditions, but
3048  * that condition is expected to be rare, so for the moment it is the
3049  * preferred interface.
3050  */
3051 void
3052 vn_pollevent(vp, events)
3053 	struct vnode *vp;
3054 	short events;
3055 {
3056 
3057 	if (vp->v_pollinfo == NULL)
3058 		v_addpollinfo(vp);
3059 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3060 	if (vp->v_pollinfo->vpi_events & events) {
3061 		/*
3062 		 * We clear vpi_events so that we don't
3063 		 * call selwakeup() twice if two events are
3064 		 * posted before the polling process(es) is
3065 		 * awakened.  This also ensures that we take at
3066 		 * most one selwakeup() if the polling process
3067 		 * is no longer interested.  However, it does
3068 		 * mean that only one event can be noticed at
3069 		 * a time.  (Perhaps we should only clear those
3070 		 * event bits which we note?) XXX
3071 		 */
3072 		vp->v_pollinfo->vpi_events = 0;	/* &= ~events ??? */
3073 		vp->v_pollinfo->vpi_revents |= events;
3074 		selwakeup(&vp->v_pollinfo->vpi_selinfo);
3075 	}
3076 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3077 }
3078 
3079 /*
3080  * Wake up anyone polling on vp because it is being revoked.
3081  * This depends on dead_poll() returning POLLHUP for correct
3082  * behavior.
3083  */
3084 void
3085 vn_pollgone(vp)
3086 	struct vnode *vp;
3087 {
3088 
3089 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3090 	VN_KNOTE(vp, NOTE_REVOKE);
3091 	if (vp->v_pollinfo->vpi_events) {
3092 		vp->v_pollinfo->vpi_events = 0;
3093 		selwakeup(&vp->v_pollinfo->vpi_selinfo);
3094 	}
3095 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3096 }
3097 
3098 
3099 
3100 /*
3101  * Routine to create and manage a filesystem syncer vnode.
3102  */
3103 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
3104 static int	sync_fsync(struct  vop_fsync_args *);
3105 static int	sync_inactive(struct  vop_inactive_args *);
3106 static int	sync_reclaim(struct  vop_reclaim_args *);
3107 static int	sync_print(struct vop_print_args *);
3108 
3109 static vop_t **sync_vnodeop_p;
3110 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
3111 	{ &vop_default_desc,	(vop_t *) vop_eopnotsupp },
3112 	{ &vop_close_desc,	(vop_t *) sync_close },		/* close */
3113 	{ &vop_fsync_desc,	(vop_t *) sync_fsync },		/* fsync */
3114 	{ &vop_inactive_desc,	(vop_t *) sync_inactive },	/* inactive */
3115 	{ &vop_reclaim_desc,	(vop_t *) sync_reclaim },	/* reclaim */
3116 	{ &vop_lock_desc,	(vop_t *) vop_stdlock },	/* lock */
3117 	{ &vop_unlock_desc,	(vop_t *) vop_stdunlock },	/* unlock */
3118 	{ &vop_print_desc,	(vop_t *) sync_print },		/* print */
3119 	{ &vop_islocked_desc,	(vop_t *) vop_stdislocked },	/* islocked */
3120 	{ NULL, NULL }
3121 };
3122 static struct vnodeopv_desc sync_vnodeop_opv_desc =
3123 	{ &sync_vnodeop_p, sync_vnodeop_entries };
3124 
3125 VNODEOP_SET(sync_vnodeop_opv_desc);
3126 
3127 /*
3128  * Create a new filesystem syncer vnode for the specified mount point.
3129  */
3130 int
3131 vfs_allocate_syncvnode(mp)
3132 	struct mount *mp;
3133 {
3134 	struct vnode *vp;
3135 	static long start, incr, next;
3136 	int error;
3137 
3138 	/* Allocate a new vnode */
3139 	if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
3140 		mp->mnt_syncer = NULL;
3141 		return (error);
3142 	}
3143 	vp->v_type = VNON;
3144 	/*
3145 	 * Place the vnode onto the syncer worklist. We attempt to
3146 	 * scatter them about on the list so that they will go off
3147 	 * at evenly distributed times even if all the filesystems
3148 	 * are mounted at once.
3149 	 */
3150 	next += incr;
3151 	if (next == 0 || next > syncer_maxdelay) {
3152 		start /= 2;
3153 		incr /= 2;
3154 		if (start == 0) {
3155 			start = syncer_maxdelay / 2;
3156 			incr = syncer_maxdelay;
3157 		}
3158 		next = start;
3159 	}
3160 	VI_LOCK(vp);
3161 	vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
3162 	VI_UNLOCK(vp);
3163 	mp->mnt_syncer = vp;
3164 	return (0);
3165 }
3166 
3167 /*
3168  * Do a lazy sync of the filesystem.
3169  */
3170 static int
3171 sync_fsync(ap)
3172 	struct vop_fsync_args /* {
3173 		struct vnode *a_vp;
3174 		struct ucred *a_cred;
3175 		int a_waitfor;
3176 		struct thread *a_td;
3177 	} */ *ap;
3178 {
3179 	struct vnode *syncvp = ap->a_vp;
3180 	struct mount *mp = syncvp->v_mount;
3181 	struct thread *td = ap->a_td;
3182 	int asyncflag;
3183 
3184 	/*
3185 	 * We only need to do something if this is a lazy evaluation.
3186 	 */
3187 	if (ap->a_waitfor != MNT_LAZY)
3188 		return (0);
3189 
3190 	/*
3191 	 * Move ourselves to the back of the sync list.
3192 	 */
3193 	VI_LOCK(syncvp);
3194 	vn_syncer_add_to_worklist(syncvp, syncdelay);
3195 	VI_UNLOCK(syncvp);
3196 
3197 	/*
3198 	 * Walk the list of vnodes pushing all that are dirty and
3199 	 * not already on the sync list.
3200 	 */
3201 	mtx_lock(&mountlist_mtx);
3202 	if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) {
3203 		mtx_unlock(&mountlist_mtx);
3204 		return (0);
3205 	}
3206 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3207 		vfs_unbusy(mp, td);
3208 		return (0);
3209 	}
3210 	asyncflag = mp->mnt_flag & MNT_ASYNC;
3211 	mp->mnt_flag &= ~MNT_ASYNC;
3212 	vfs_msync(mp, MNT_NOWAIT);
3213 	VFS_SYNC(mp, MNT_LAZY, ap->a_cred, td);
3214 	if (asyncflag)
3215 		mp->mnt_flag |= MNT_ASYNC;
3216 	vn_finished_write(mp);
3217 	vfs_unbusy(mp, td);
3218 	return (0);
3219 }
3220 
3221 /*
3222  * The syncer vnode is no referenced.
3223  */
3224 static int
3225 sync_inactive(ap)
3226 	struct vop_inactive_args /* {
3227 		struct vnode *a_vp;
3228 		struct thread *a_td;
3229 	} */ *ap;
3230 {
3231 
3232 	VOP_UNLOCK(ap->a_vp, 0, ap->a_td);
3233 	vgone(ap->a_vp);
3234 	return (0);
3235 }
3236 
3237 /*
3238  * The syncer vnode is no longer needed and is being decommissioned.
3239  *
3240  * Modifications to the worklist must be protected at splbio().
3241  */
3242 static int
3243 sync_reclaim(ap)
3244 	struct vop_reclaim_args /* {
3245 		struct vnode *a_vp;
3246 	} */ *ap;
3247 {
3248 	struct vnode *vp = ap->a_vp;
3249 	int s;
3250 
3251 	s = splbio();
3252 	vp->v_mount->mnt_syncer = NULL;
3253 	VI_LOCK(vp);
3254 	if (vp->v_iflag & VI_ONWORKLST) {
3255 		LIST_REMOVE(vp, v_synclist);
3256 		vp->v_iflag &= ~VI_ONWORKLST;
3257 	}
3258 	VI_UNLOCK(vp);
3259 	splx(s);
3260 
3261 	return (0);
3262 }
3263 
3264 /*
3265  * Print out a syncer vnode.
3266  */
3267 static int
3268 sync_print(ap)
3269 	struct vop_print_args /* {
3270 		struct vnode *a_vp;
3271 	} */ *ap;
3272 {
3273 	struct vnode *vp = ap->a_vp;
3274 
3275 	printf("syncer vnode");
3276 	if (vp->v_vnlock != NULL)
3277 		lockmgr_printinfo(vp->v_vnlock);
3278 	printf("\n");
3279 	return (0);
3280 }
3281 
3282 /*
3283  * extract the dev_t from a VCHR
3284  */
3285 dev_t
3286 vn_todev(vp)
3287 	struct vnode *vp;
3288 {
3289 	if (vp->v_type != VCHR)
3290 		return (NODEV);
3291 	return (vp->v_rdev);
3292 }
3293 
3294 /*
3295  * Check if vnode represents a disk device
3296  */
3297 int
3298 vn_isdisk(vp, errp)
3299 	struct vnode *vp;
3300 	int *errp;
3301 {
3302 	struct cdevsw *cdevsw;
3303 
3304 	if (vp->v_type != VCHR) {
3305 		if (errp != NULL)
3306 			*errp = ENOTBLK;
3307 		return (0);
3308 	}
3309 	if (vp->v_rdev == NULL) {
3310 		if (errp != NULL)
3311 			*errp = ENXIO;
3312 		return (0);
3313 	}
3314 	cdevsw = devsw(vp->v_rdev);
3315 	if (cdevsw == NULL) {
3316 		if (errp != NULL)
3317 			*errp = ENXIO;
3318 		return (0);
3319 	}
3320 	if (!(cdevsw->d_flags & D_DISK)) {
3321 		if (errp != NULL)
3322 			*errp = ENOTBLK;
3323 		return (0);
3324 	}
3325 	if (errp != NULL)
3326 		*errp = 0;
3327 	return (1);
3328 }
3329 
3330 /*
3331  * Free data allocated by namei(); see namei(9) for details.
3332  */
3333 void
3334 NDFREE(ndp, flags)
3335      struct nameidata *ndp;
3336      const uint flags;
3337 {
3338 	if (!(flags & NDF_NO_FREE_PNBUF) &&
3339 	    (ndp->ni_cnd.cn_flags & HASBUF)) {
3340 		uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
3341 		ndp->ni_cnd.cn_flags &= ~HASBUF;
3342 	}
3343 	if (!(flags & NDF_NO_DVP_UNLOCK) &&
3344 	    (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
3345 	    ndp->ni_dvp != ndp->ni_vp)
3346 		VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_thread);
3347 	if (!(flags & NDF_NO_DVP_RELE) &&
3348 	    (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
3349 		vrele(ndp->ni_dvp);
3350 		ndp->ni_dvp = NULL;
3351 	}
3352 	if (!(flags & NDF_NO_VP_UNLOCK) &&
3353 	    (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
3354 		VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_thread);
3355 	if (!(flags & NDF_NO_VP_RELE) &&
3356 	    ndp->ni_vp) {
3357 		vrele(ndp->ni_vp);
3358 		ndp->ni_vp = NULL;
3359 	}
3360 	if (!(flags & NDF_NO_STARTDIR_RELE) &&
3361 	    (ndp->ni_cnd.cn_flags & SAVESTART)) {
3362 		vrele(ndp->ni_startdir);
3363 		ndp->ni_startdir = NULL;
3364 	}
3365 }
3366 
3367 /*
3368  * Common filesystem object access control check routine.  Accepts a
3369  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3370  * and optional call-by-reference privused argument allowing vaccess()
3371  * to indicate to the caller whether privilege was used to satisfy the
3372  * request (obsoleted).  Returns 0 on success, or an errno on failure.
3373  */
3374 int
3375 vaccess(type, file_mode, file_uid, file_gid, acc_mode, cred, privused)
3376 	enum vtype type;
3377 	mode_t file_mode;
3378 	uid_t file_uid;
3379 	gid_t file_gid;
3380 	mode_t acc_mode;
3381 	struct ucred *cred;
3382 	int *privused;
3383 {
3384 	mode_t dac_granted;
3385 #ifdef CAPABILITIES
3386 	mode_t cap_granted;
3387 #endif
3388 
3389 	/*
3390 	 * Look for a normal, non-privileged way to access the file/directory
3391 	 * as requested.  If it exists, go with that.
3392 	 */
3393 
3394 	if (privused != NULL)
3395 		*privused = 0;
3396 
3397 	dac_granted = 0;
3398 
3399 	/* Check the owner. */
3400 	if (cred->cr_uid == file_uid) {
3401 		dac_granted |= VADMIN;
3402 		if (file_mode & S_IXUSR)
3403 			dac_granted |= VEXEC;
3404 		if (file_mode & S_IRUSR)
3405 			dac_granted |= VREAD;
3406 		if (file_mode & S_IWUSR)
3407 			dac_granted |= (VWRITE | VAPPEND);
3408 
3409 		if ((acc_mode & dac_granted) == acc_mode)
3410 			return (0);
3411 
3412 		goto privcheck;
3413 	}
3414 
3415 	/* Otherwise, check the groups (first match) */
3416 	if (groupmember(file_gid, cred)) {
3417 		if (file_mode & S_IXGRP)
3418 			dac_granted |= VEXEC;
3419 		if (file_mode & S_IRGRP)
3420 			dac_granted |= VREAD;
3421 		if (file_mode & S_IWGRP)
3422 			dac_granted |= (VWRITE | VAPPEND);
3423 
3424 		if ((acc_mode & dac_granted) == acc_mode)
3425 			return (0);
3426 
3427 		goto privcheck;
3428 	}
3429 
3430 	/* Otherwise, check everyone else. */
3431 	if (file_mode & S_IXOTH)
3432 		dac_granted |= VEXEC;
3433 	if (file_mode & S_IROTH)
3434 		dac_granted |= VREAD;
3435 	if (file_mode & S_IWOTH)
3436 		dac_granted |= (VWRITE | VAPPEND);
3437 	if ((acc_mode & dac_granted) == acc_mode)
3438 		return (0);
3439 
3440 privcheck:
3441 	if (!suser_cred(cred, PRISON_ROOT)) {
3442 		/* XXX audit: privilege used */
3443 		if (privused != NULL)
3444 			*privused = 1;
3445 		return (0);
3446 	}
3447 
3448 #ifdef CAPABILITIES
3449 	/*
3450 	 * Build a capability mask to determine if the set of capabilities
3451 	 * satisfies the requirements when combined with the granted mask
3452 	 * from above.
3453 	 * For each capability, if the capability is required, bitwise
3454 	 * or the request type onto the cap_granted mask.
3455 	 */
3456 	cap_granted = 0;
3457 
3458 	if (type == VDIR) {
3459 		/*
3460 		 * For directories, use CAP_DAC_READ_SEARCH to satisfy
3461 		 * VEXEC requests, instead of CAP_DAC_EXECUTE.
3462 		 */
3463 		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3464 		    !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3465 			cap_granted |= VEXEC;
3466 	} else {
3467 		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3468 		    !cap_check(cred, NULL, CAP_DAC_EXECUTE, PRISON_ROOT))
3469 			cap_granted |= VEXEC;
3470 	}
3471 
3472 	if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
3473 	    !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3474 		cap_granted |= VREAD;
3475 
3476 	if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3477 	    !cap_check(cred, NULL, CAP_DAC_WRITE, PRISON_ROOT))
3478 		cap_granted |= (VWRITE | VAPPEND);
3479 
3480 	if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3481 	    !cap_check(cred, NULL, CAP_FOWNER, PRISON_ROOT))
3482 		cap_granted |= VADMIN;
3483 
3484 	if ((acc_mode & (cap_granted | dac_granted)) == acc_mode) {
3485 		/* XXX audit: privilege used */
3486 		if (privused != NULL)
3487 			*privused = 1;
3488 		return (0);
3489 	}
3490 #endif
3491 
3492 	return ((acc_mode & VADMIN) ? EPERM : EACCES);
3493 }
3494