xref: /freebsd/sys/kern/vfs_subr.c (revision d2387d42b8da231a5b95cbc313825fb2aadf26f6)
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  */
40 
41 /*
42  * External virtual filesystem routines
43  */
44 
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
47 
48 #include "opt_ddb.h"
49 #include "opt_mac.h"
50 
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/bio.h>
54 #include <sys/buf.h>
55 #include <sys/conf.h>
56 #include <sys/eventhandler.h>
57 #include <sys/extattr.h>
58 #include <sys/fcntl.h>
59 #include <sys/kernel.h>
60 #include <sys/kthread.h>
61 #include <sys/mac.h>
62 #include <sys/malloc.h>
63 #include <sys/mount.h>
64 #include <sys/namei.h>
65 #include <sys/sleepqueue.h>
66 #include <sys/stat.h>
67 #include <sys/sysctl.h>
68 #include <sys/syslog.h>
69 #include <sys/vmmeter.h>
70 #include <sys/vnode.h>
71 
72 #include <vm/vm.h>
73 #include <vm/vm_object.h>
74 #include <vm/vm_extern.h>
75 #include <vm/pmap.h>
76 #include <vm/vm_map.h>
77 #include <vm/vm_page.h>
78 #include <vm/vm_kern.h>
79 #include <vm/uma.h>
80 
81 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
82 
83 static void	addalias(struct vnode *vp, dev_t nvp_rdev);
84 static void	insmntque(struct vnode *vp, struct mount *mp);
85 static void	vclean(struct vnode *vp, int flags, struct thread *td);
86 static void	vlruvp(struct vnode *vp);
87 static int	flushbuflist(struct buf *blist, int flags, struct vnode *vp,
88 		    int slpflag, int slptimeo, int *errorp);
89 static int	vtryrecycle(struct vnode *vp);
90 static void	vx_lock(struct vnode *vp);
91 static void	vx_unlock(struct vnode *vp);
92 static void	vgonechrl(struct vnode *vp, struct thread *td);
93 
94 
95 /*
96  * Number of vnodes in existence.  Increased whenever getnewvnode()
97  * allocates a new vnode, never decreased.
98  */
99 static unsigned long	numvnodes;
100 
101 SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
102 
103 /*
104  * Conversion tables for conversion from vnode types to inode formats
105  * and back.
106  */
107 enum vtype iftovt_tab[16] = {
108 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
109 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
110 };
111 int vttoif_tab[9] = {
112 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
113 	S_IFSOCK, S_IFIFO, S_IFMT,
114 };
115 
116 /*
117  * List of vnodes that are ready for recycling.
118  */
119 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
120 
121 /*
122  * Minimum number of free vnodes.  If there are fewer than this free vnodes,
123  * getnewvnode() will return a newly allocated vnode.
124  */
125 static u_long wantfreevnodes = 25;
126 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
127 /* Number of vnodes in the free list. */
128 static u_long freevnodes;
129 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
130 
131 /*
132  * Various variables used for debugging the new implementation of
133  * reassignbuf().
134  * XXX these are probably of (very) limited utility now.
135  */
136 static int reassignbufcalls;
137 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
138 static int nameileafonly;
139 SYSCTL_INT(_vfs, OID_AUTO, nameileafonly, CTLFLAG_RW, &nameileafonly, 0, "");
140 
141 /*
142  * Cache for the mount type id assigned to NFS.  This is used for
143  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
144  */
145 int	nfs_mount_type = -1;
146 
147 /* To keep more than one thread at a time from running vfs_getnewfsid */
148 static struct mtx mntid_mtx;
149 
150 /*
151  * Lock for any access to the following:
152  *	vnode_free_list
153  *	numvnodes
154  *	freevnodes
155  */
156 static struct mtx vnode_free_list_mtx;
157 
158 /*
159  * For any iteration/modification of dev->si_hlist (linked through
160  * v_specnext)
161  */
162 static struct mtx spechash_mtx;
163 
164 /* Publicly exported FS */
165 struct nfs_public nfs_pub;
166 
167 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
168 static uma_zone_t vnode_zone;
169 static uma_zone_t vnodepoll_zone;
170 
171 /* Set to 1 to print out reclaim of active vnodes */
172 int	prtactive;
173 
174 /*
175  * The workitem queue.
176  *
177  * It is useful to delay writes of file data and filesystem metadata
178  * for tens of seconds so that quickly created and deleted files need
179  * not waste disk bandwidth being created and removed. To realize this,
180  * we append vnodes to a "workitem" queue. When running with a soft
181  * updates implementation, most pending metadata dependencies should
182  * not wait for more than a few seconds. Thus, mounted on block devices
183  * are delayed only about a half the time that file data is delayed.
184  * Similarly, directory updates are more critical, so are only delayed
185  * about a third the time that file data is delayed. Thus, there are
186  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
187  * one each second (driven off the filesystem syncer process). The
188  * syncer_delayno variable indicates the next queue that is to be processed.
189  * Items that need to be processed soon are placed in this queue:
190  *
191  *	syncer_workitem_pending[syncer_delayno]
192  *
193  * A delay of fifteen seconds is done by placing the request fifteen
194  * entries later in the queue:
195  *
196  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
197  *
198  */
199 static int syncer_delayno;
200 static long syncer_mask;
201 LIST_HEAD(synclist, vnode);
202 static struct synclist *syncer_workitem_pending;
203 /*
204  * The sync_mtx protects:
205  *	vp->v_synclist
206  *	syncer_delayno
207  *	syncer_workitem_pending
208  *	rushjob
209  */
210 static struct mtx sync_mtx;
211 
212 #define SYNCER_MAXDELAY		32
213 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
214 static int syncdelay = 30;		/* max time to delay syncing data */
215 static int filedelay = 30;		/* time to delay syncing files */
216 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
217 static int dirdelay = 29;		/* time to delay syncing directories */
218 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
219 static int metadelay = 28;		/* time to delay syncing metadata */
220 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
221 static int rushjob;		/* number of slots to run ASAP */
222 static int stat_rush_requests;	/* number of times I/O speeded up */
223 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
224 
225 /*
226  * Number of vnodes we want to exist at any one time.  This is mostly used
227  * to size hash tables in vnode-related code.  It is normally not used in
228  * getnewvnode(), as wantfreevnodes is normally nonzero.)
229  *
230  * XXX desiredvnodes is historical cruft and should not exist.
231  */
232 int desiredvnodes;
233 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
234     &desiredvnodes, 0, "Maximum number of vnodes");
235 static int minvnodes;
236 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
237     &minvnodes, 0, "Minimum number of vnodes");
238 static int vnlru_nowhere;
239 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
240     &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
241 
242 /* Hook for calling soft updates. */
243 int (*softdep_process_worklist_hook)(struct mount *);
244 
245 /*
246  * Initialize the vnode management data structures.
247  */
248 static void
249 vntblinit(void *dummy __unused)
250 {
251 
252 	/*
253 	 * Desiredvnodes is a function of the physical memory size and
254 	 * the kernel's heap size.  Specifically, desiredvnodes scales
255 	 * in proportion to the physical memory size until two fifths
256 	 * of the kernel's heap size is consumed by vnodes and vm
257 	 * objects.
258 	 */
259 	desiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * vm_kmem_size /
260 	    (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
261 	minvnodes = desiredvnodes / 4;
262 	mtx_init(&mountlist_mtx, "mountlist", NULL, MTX_DEF);
263 	mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
264 	mtx_init(&spechash_mtx, "spechash", NULL, MTX_DEF);
265 	TAILQ_INIT(&vnode_free_list);
266 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
267 	vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
268 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
269 	vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
270 	      NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
271 	/*
272 	 * Initialize the filesystem syncer.
273 	 */
274 	syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
275 		&syncer_mask);
276 	syncer_maxdelay = syncer_mask + 1;
277 	mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
278 }
279 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL)
280 
281 
282 /*
283  * Mark a mount point as busy. Used to synchronize access and to delay
284  * unmounting. Interlock is not released on failure.
285  */
286 int
287 vfs_busy(mp, flags, interlkp, td)
288 	struct mount *mp;
289 	int flags;
290 	struct mtx *interlkp;
291 	struct thread *td;
292 {
293 	int lkflags;
294 
295 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
296 		if (flags & LK_NOWAIT)
297 			return (ENOENT);
298 		mp->mnt_kern_flag |= MNTK_MWAIT;
299 		/*
300 		 * Since all busy locks are shared except the exclusive
301 		 * lock granted when unmounting, the only place that a
302 		 * wakeup needs to be done is at the release of the
303 		 * exclusive lock at the end of dounmount.
304 		 */
305 		msleep(mp, interlkp, PVFS, "vfs_busy", 0);
306 		return (ENOENT);
307 	}
308 	lkflags = LK_SHARED | LK_NOPAUSE;
309 	if (interlkp)
310 		lkflags |= LK_INTERLOCK;
311 	if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
312 		panic("vfs_busy: unexpected lock failure");
313 	return (0);
314 }
315 
316 /*
317  * Free a busy filesystem.
318  */
319 void
320 vfs_unbusy(mp, td)
321 	struct mount *mp;
322 	struct thread *td;
323 {
324 
325 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
326 }
327 
328 /*
329  * Lookup a mount point by filesystem identifier.
330  */
331 struct mount *
332 vfs_getvfs(fsid)
333 	fsid_t *fsid;
334 {
335 	register struct mount *mp;
336 
337 	mtx_lock(&mountlist_mtx);
338 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
339 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
340 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
341 			mtx_unlock(&mountlist_mtx);
342 			return (mp);
343 		}
344 	}
345 	mtx_unlock(&mountlist_mtx);
346 	return ((struct mount *) 0);
347 }
348 
349 /*
350  * Get a new unique fsid.  Try to make its val[0] unique, since this value
351  * will be used to create fake device numbers for stat().  Also try (but
352  * not so hard) make its val[0] unique mod 2^16, since some emulators only
353  * support 16-bit device numbers.  We end up with unique val[0]'s for the
354  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
355  *
356  * Keep in mind that several mounts may be running in parallel.  Starting
357  * the search one past where the previous search terminated is both a
358  * micro-optimization and a defense against returning the same fsid to
359  * different mounts.
360  */
361 void
362 vfs_getnewfsid(mp)
363 	struct mount *mp;
364 {
365 	static u_int16_t mntid_base;
366 	fsid_t tfsid;
367 	int mtype;
368 
369 	mtx_lock(&mntid_mtx);
370 	mtype = mp->mnt_vfc->vfc_typenum;
371 	tfsid.val[1] = mtype;
372 	mtype = (mtype & 0xFF) << 24;
373 	for (;;) {
374 		tfsid.val[0] = makeudev(255,
375 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
376 		mntid_base++;
377 		if (vfs_getvfs(&tfsid) == NULL)
378 			break;
379 	}
380 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
381 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
382 	mtx_unlock(&mntid_mtx);
383 }
384 
385 /*
386  * Knob to control the precision of file timestamps:
387  *
388  *   0 = seconds only; nanoseconds zeroed.
389  *   1 = seconds and nanoseconds, accurate within 1/HZ.
390  *   2 = seconds and nanoseconds, truncated to microseconds.
391  * >=3 = seconds and nanoseconds, maximum precision.
392  */
393 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
394 
395 static int timestamp_precision = TSP_SEC;
396 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
397     &timestamp_precision, 0, "");
398 
399 /*
400  * Get a current timestamp.
401  */
402 void
403 vfs_timestamp(tsp)
404 	struct timespec *tsp;
405 {
406 	struct timeval tv;
407 
408 	switch (timestamp_precision) {
409 	case TSP_SEC:
410 		tsp->tv_sec = time_second;
411 		tsp->tv_nsec = 0;
412 		break;
413 	case TSP_HZ:
414 		getnanotime(tsp);
415 		break;
416 	case TSP_USEC:
417 		microtime(&tv);
418 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
419 		break;
420 	case TSP_NSEC:
421 	default:
422 		nanotime(tsp);
423 		break;
424 	}
425 }
426 
427 /*
428  * Set vnode attributes to VNOVAL
429  */
430 void
431 vattr_null(vap)
432 	register struct vattr *vap;
433 {
434 
435 	vap->va_type = VNON;
436 	vap->va_size = VNOVAL;
437 	vap->va_bytes = VNOVAL;
438 	vap->va_mode = VNOVAL;
439 	vap->va_nlink = VNOVAL;
440 	vap->va_uid = VNOVAL;
441 	vap->va_gid = VNOVAL;
442 	vap->va_fsid = VNOVAL;
443 	vap->va_fileid = VNOVAL;
444 	vap->va_blocksize = VNOVAL;
445 	vap->va_rdev = VNOVAL;
446 	vap->va_atime.tv_sec = VNOVAL;
447 	vap->va_atime.tv_nsec = VNOVAL;
448 	vap->va_mtime.tv_sec = VNOVAL;
449 	vap->va_mtime.tv_nsec = VNOVAL;
450 	vap->va_ctime.tv_sec = VNOVAL;
451 	vap->va_ctime.tv_nsec = VNOVAL;
452 	vap->va_birthtime.tv_sec = VNOVAL;
453 	vap->va_birthtime.tv_nsec = VNOVAL;
454 	vap->va_flags = VNOVAL;
455 	vap->va_gen = VNOVAL;
456 	vap->va_vaflags = 0;
457 }
458 
459 /*
460  * This routine is called when we have too many vnodes.  It attempts
461  * to free <count> vnodes and will potentially free vnodes that still
462  * have VM backing store (VM backing store is typically the cause
463  * of a vnode blowout so we want to do this).  Therefore, this operation
464  * is not considered cheap.
465  *
466  * A number of conditions may prevent a vnode from being reclaimed.
467  * the buffer cache may have references on the vnode, a directory
468  * vnode may still have references due to the namei cache representing
469  * underlying files, or the vnode may be in active use.   It is not
470  * desireable to reuse such vnodes.  These conditions may cause the
471  * number of vnodes to reach some minimum value regardless of what
472  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
473  */
474 static int
475 vlrureclaim(struct mount *mp)
476 {
477 	struct vnode *vp;
478 	int done;
479 	int trigger;
480 	int usevnodes;
481 	int count;
482 
483 	/*
484 	 * Calculate the trigger point, don't allow user
485 	 * screwups to blow us up.   This prevents us from
486 	 * recycling vnodes with lots of resident pages.  We
487 	 * aren't trying to free memory, we are trying to
488 	 * free vnodes.
489 	 */
490 	usevnodes = desiredvnodes;
491 	if (usevnodes <= 0)
492 		usevnodes = 1;
493 	trigger = cnt.v_page_count * 2 / usevnodes;
494 
495 	done = 0;
496 	MNT_ILOCK(mp);
497 	count = mp->mnt_nvnodelistsize / 10 + 1;
498 	while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
499 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
500 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
501 
502 		if (vp->v_type != VNON &&
503 		    vp->v_type != VBAD &&
504 		    VI_TRYLOCK(vp)) {
505 			if (VMIGHTFREE(vp) &&           /* critical path opt */
506 			    (vp->v_object == NULL ||
507 			    vp->v_object->resident_page_count < trigger)) {
508 				MNT_IUNLOCK(mp);
509 				vgonel(vp, curthread);
510 				done++;
511 				MNT_ILOCK(mp);
512 			} else
513 				VI_UNLOCK(vp);
514 		}
515 		--count;
516 	}
517 	MNT_IUNLOCK(mp);
518 	return done;
519 }
520 
521 /*
522  * Attempt to recycle vnodes in a context that is always safe to block.
523  * Calling vlrurecycle() from the bowels of filesystem code has some
524  * interesting deadlock problems.
525  */
526 static struct proc *vnlruproc;
527 static int vnlruproc_sig;
528 
529 static void
530 vnlru_proc(void)
531 {
532 	struct mount *mp, *nmp;
533 	int done;
534 	struct proc *p = vnlruproc;
535 	struct thread *td = FIRST_THREAD_IN_PROC(p);
536 
537 	mtx_lock(&Giant);
538 
539 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
540 	    SHUTDOWN_PRI_FIRST);
541 
542 	for (;;) {
543 		kthread_suspend_check(p);
544 		mtx_lock(&vnode_free_list_mtx);
545 		if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
546 			mtx_unlock(&vnode_free_list_mtx);
547 			vnlruproc_sig = 0;
548 			wakeup(&vnlruproc_sig);
549 			tsleep(vnlruproc, PVFS, "vlruwt", hz);
550 			continue;
551 		}
552 		mtx_unlock(&vnode_free_list_mtx);
553 		done = 0;
554 		mtx_lock(&mountlist_mtx);
555 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
556 			if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
557 				nmp = TAILQ_NEXT(mp, mnt_list);
558 				continue;
559 			}
560 			done += vlrureclaim(mp);
561 			mtx_lock(&mountlist_mtx);
562 			nmp = TAILQ_NEXT(mp, mnt_list);
563 			vfs_unbusy(mp, td);
564 		}
565 		mtx_unlock(&mountlist_mtx);
566 		if (done == 0) {
567 #if 0
568 			/* These messages are temporary debugging aids */
569 			if (vnlru_nowhere < 5)
570 				printf("vnlru process getting nowhere..\n");
571 			else if (vnlru_nowhere == 5)
572 				printf("vnlru process messages stopped.\n");
573 #endif
574 			vnlru_nowhere++;
575 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
576 		}
577 	}
578 }
579 
580 static struct kproc_desc vnlru_kp = {
581 	"vnlru",
582 	vnlru_proc,
583 	&vnlruproc
584 };
585 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
586 
587 
588 /*
589  * Routines having to do with the management of the vnode table.
590  */
591 
592 /*
593  * Check to see if a free vnode can be recycled. If it can,
594  * recycle it and return it with the vnode interlock held.
595  */
596 static int
597 vtryrecycle(struct vnode *vp)
598 {
599 	struct thread *td = curthread;
600 	vm_object_t object;
601 	struct mount *vnmp;
602 	int error;
603 
604 	/* Don't recycle if we can't get the interlock */
605 	if (!VI_TRYLOCK(vp))
606 		return (EWOULDBLOCK);
607 	/*
608 	 * This vnode may found and locked via some other list, if so we
609 	 * can't recycle it yet.
610 	 */
611 	if (vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE | LK_NOWAIT, td) != 0)
612 		return (EWOULDBLOCK);
613 	/*
614 	 * Don't recycle if its filesystem is being suspended.
615 	 */
616 	if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
617 		VOP_UNLOCK(vp, 0, td);
618 		return (EBUSY);
619 	}
620 
621 	/*
622 	 * Don't recycle if we still have cached pages.
623 	 */
624 	if (VOP_GETVOBJECT(vp, &object) == 0) {
625 		VM_OBJECT_LOCK(object);
626 		if (object->resident_page_count ||
627 		    object->ref_count) {
628 			VM_OBJECT_UNLOCK(object);
629 			error = EBUSY;
630 			goto done;
631 		}
632 		VM_OBJECT_UNLOCK(object);
633 	}
634 	if (LIST_FIRST(&vp->v_cache_src)) {
635 		/*
636 		 * note: nameileafonly sysctl is temporary,
637 		 * for debugging only, and will eventually be
638 		 * removed.
639 		 */
640 		if (nameileafonly > 0) {
641 			/*
642 			 * Do not reuse namei-cached directory
643 			 * vnodes that have cached
644 			 * subdirectories.
645 			 */
646 			if (cache_leaf_test(vp) < 0) {
647 				error = EISDIR;
648 				goto done;
649 			}
650 		} else if (nameileafonly < 0 ||
651 			    vmiodirenable == 0) {
652 			/*
653 			 * Do not reuse namei-cached directory
654 			 * vnodes if nameileafonly is -1 or
655 			 * if VMIO backing for directories is
656 			 * turned off (otherwise we reuse them
657 			 * too quickly).
658 			 */
659 			error = EBUSY;
660 			goto done;
661 		}
662 	}
663 	/*
664 	 * If we got this far, we need to acquire the interlock and see if
665 	 * anyone picked up this vnode from another list.  If not, we will
666 	 * mark it with XLOCK via vgonel() so that anyone who does find it
667 	 * will skip over it.
668 	 */
669 	VI_LOCK(vp);
670 	if (VSHOULDBUSY(vp) && (vp->v_iflag & VI_XLOCK) == 0) {
671 		VI_UNLOCK(vp);
672 		error = EBUSY;
673 		goto done;
674 	}
675 	mtx_lock(&vnode_free_list_mtx);
676 	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
677 	vp->v_iflag &= ~VI_FREE;
678 	mtx_unlock(&vnode_free_list_mtx);
679 	vp->v_iflag |= VI_DOOMED;
680 	if (vp->v_type != VBAD) {
681 		VOP_UNLOCK(vp, 0, td);
682 		vgonel(vp, td);
683 		VI_LOCK(vp);
684 	} else
685 		VOP_UNLOCK(vp, 0, td);
686 	vn_finished_write(vnmp);
687 	return (0);
688 done:
689 	VOP_UNLOCK(vp, 0, td);
690 	vn_finished_write(vnmp);
691 	return (error);
692 }
693 
694 /*
695  * Return the next vnode from the free list.
696  */
697 int
698 getnewvnode(tag, mp, vops, vpp)
699 	const char *tag;
700 	struct mount *mp;
701 	vop_t **vops;
702 	struct vnode **vpp;
703 {
704 	struct vnode *vp = NULL;
705 	struct vpollinfo *pollinfo = NULL;
706 
707 	mtx_lock(&vnode_free_list_mtx);
708 
709 	/*
710 	 * Try to reuse vnodes if we hit the max.  This situation only
711 	 * occurs in certain large-memory (2G+) situations.  We cannot
712 	 * attempt to directly reclaim vnodes due to nasty recursion
713 	 * problems.
714 	 */
715 	while (numvnodes - freevnodes > desiredvnodes) {
716 		if (vnlruproc_sig == 0) {
717 			vnlruproc_sig = 1;      /* avoid unnecessary wakeups */
718 			wakeup(vnlruproc);
719 		}
720 		mtx_unlock(&vnode_free_list_mtx);
721 		tsleep(&vnlruproc_sig, PVFS, "vlruwk", hz);
722 		mtx_lock(&vnode_free_list_mtx);
723 	}
724 
725 	/*
726 	 * Attempt to reuse a vnode already on the free list, allocating
727 	 * a new vnode if we can't find one or if we have not reached a
728 	 * good minimum for good LRU performance.
729 	 */
730 
731 	if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
732 		int error;
733 		int count;
734 
735 		for (count = 0; count < freevnodes; count++) {
736 			vp = TAILQ_FIRST(&vnode_free_list);
737 
738 			KASSERT(vp->v_usecount == 0 &&
739 			    (vp->v_iflag & VI_DOINGINACT) == 0,
740 			    ("getnewvnode: free vnode isn't"));
741 
742 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
743 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
744 			mtx_unlock(&vnode_free_list_mtx);
745 			error = vtryrecycle(vp);
746 			mtx_lock(&vnode_free_list_mtx);
747 			if (error == 0)
748 				break;
749 			vp = NULL;
750 		}
751 	}
752 	if (vp) {
753 		freevnodes--;
754 		mtx_unlock(&vnode_free_list_mtx);
755 
756 #ifdef INVARIANTS
757 		{
758 			if (vp->v_data)
759 				panic("cleaned vnode isn't");
760 			if (vp->v_numoutput)
761 				panic("Clean vnode has pending I/O's");
762 			if (vp->v_writecount != 0)
763 				panic("Non-zero write count");
764 		}
765 #endif
766 		if ((pollinfo = vp->v_pollinfo) != NULL) {
767 			/*
768 			 * To avoid lock order reversals, the call to
769 			 * uma_zfree() must be delayed until the vnode
770 			 * interlock is released.
771 			 */
772 			vp->v_pollinfo = NULL;
773 		}
774 #ifdef MAC
775 		mac_destroy_vnode(vp);
776 #endif
777 		vp->v_iflag = 0;
778 		vp->v_vflag = 0;
779 		vp->v_lastw = 0;
780 		vp->v_lasta = 0;
781 		vp->v_cstart = 0;
782 		vp->v_clen = 0;
783 		vp->v_socket = 0;
784 		lockdestroy(vp->v_vnlock);
785 		lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOPAUSE);
786 		KASSERT(vp->v_cleanbufcnt == 0, ("cleanbufcnt not 0"));
787 		KASSERT(vp->v_cleanblkroot == NULL, ("cleanblkroot not NULL"));
788 		KASSERT(vp->v_dirtybufcnt == 0, ("dirtybufcnt not 0"));
789 		KASSERT(vp->v_dirtyblkroot == NULL, ("dirtyblkroot not NULL"));
790 	} else {
791 		numvnodes++;
792 		mtx_unlock(&vnode_free_list_mtx);
793 
794 		vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
795 		mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
796 		VI_LOCK(vp);
797 		vp->v_dd = vp;
798 		vp->v_vnlock = &vp->v_lock;
799 		lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOPAUSE);
800 		cache_purge(vp);		/* Sets up v_id. */
801 		LIST_INIT(&vp->v_cache_src);
802 		TAILQ_INIT(&vp->v_cache_dst);
803 	}
804 
805 	TAILQ_INIT(&vp->v_cleanblkhd);
806 	TAILQ_INIT(&vp->v_dirtyblkhd);
807 	vp->v_type = VNON;
808 	vp->v_tag = tag;
809 	vp->v_op = vops;
810 	*vpp = vp;
811 	vp->v_usecount = 1;
812 	vp->v_data = 0;
813 	vp->v_cachedid = -1;
814 	VI_UNLOCK(vp);
815 	if (pollinfo != NULL) {
816 		mtx_destroy(&pollinfo->vpi_lock);
817 		uma_zfree(vnodepoll_zone, pollinfo);
818 	}
819 #ifdef MAC
820 	mac_init_vnode(vp);
821 	if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
822 		mac_associate_vnode_singlelabel(mp, vp);
823 #endif
824 	insmntque(vp, mp);
825 
826 	return (0);
827 }
828 
829 /*
830  * Move a vnode from one mount queue to another.
831  */
832 static void
833 insmntque(vp, mp)
834 	register struct vnode *vp;
835 	register struct mount *mp;
836 {
837 
838 	/*
839 	 * Delete from old mount point vnode list, if on one.
840 	 */
841 	if (vp->v_mount != NULL) {
842 		MNT_ILOCK(vp->v_mount);
843 		KASSERT(vp->v_mount->mnt_nvnodelistsize > 0,
844 			("bad mount point vnode list size"));
845 		TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
846 		vp->v_mount->mnt_nvnodelistsize--;
847 		MNT_IUNLOCK(vp->v_mount);
848 	}
849 	/*
850 	 * Insert into list of vnodes for the new mount point, if available.
851 	 */
852 	if ((vp->v_mount = mp) != NULL) {
853 		MNT_ILOCK(vp->v_mount);
854 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
855 		mp->mnt_nvnodelistsize++;
856 		MNT_IUNLOCK(vp->v_mount);
857 	}
858 }
859 
860 /*
861  * Update outstanding I/O count and do wakeup if requested.
862  */
863 void
864 vwakeup(bp)
865 	register struct buf *bp;
866 {
867 	register struct vnode *vp;
868 
869 	bp->b_flags &= ~B_WRITEINPROG;
870 	if ((vp = bp->b_vp)) {
871 		VI_LOCK(vp);
872 		vp->v_numoutput--;
873 		if (vp->v_numoutput < 0)
874 			panic("vwakeup: neg numoutput");
875 		if ((vp->v_numoutput == 0) && (vp->v_iflag & VI_BWAIT)) {
876 			vp->v_iflag &= ~VI_BWAIT;
877 			wakeup(&vp->v_numoutput);
878 		}
879 		VI_UNLOCK(vp);
880 	}
881 }
882 
883 /*
884  * Flush out and invalidate all buffers associated with a vnode.
885  * Called with the underlying object locked.
886  */
887 int
888 vinvalbuf(vp, flags, cred, td, slpflag, slptimeo)
889 	struct vnode *vp;
890 	int flags;
891 	struct ucred *cred;
892 	struct thread *td;
893 	int slpflag, slptimeo;
894 {
895 	struct buf *blist;
896 	int error;
897 	vm_object_t object;
898 
899 	GIANT_REQUIRED;
900 
901 	ASSERT_VOP_LOCKED(vp, "vinvalbuf");
902 
903 	VI_LOCK(vp);
904 	if (flags & V_SAVE) {
905 		while (vp->v_numoutput) {
906 			vp->v_iflag |= VI_BWAIT;
907 			error = msleep(&vp->v_numoutput, VI_MTX(vp),
908 			    slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo);
909 			if (error) {
910 				VI_UNLOCK(vp);
911 				return (error);
912 			}
913 		}
914 		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
915 			VI_UNLOCK(vp);
916 			if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, td)) != 0)
917 				return (error);
918 			/*
919 			 * XXX We could save a lock/unlock if this was only
920 			 * enabled under INVARIANTS
921 			 */
922 			VI_LOCK(vp);
923 			if (vp->v_numoutput > 0 ||
924 			    !TAILQ_EMPTY(&vp->v_dirtyblkhd))
925 				panic("vinvalbuf: dirty bufs");
926 		}
927 	}
928 	/*
929 	 * If you alter this loop please notice that interlock is dropped and
930 	 * reacquired in flushbuflist.  Special care is needed to ensure that
931 	 * no race conditions occur from this.
932 	 */
933 	for (error = 0;;) {
934 		if ((blist = TAILQ_FIRST(&vp->v_cleanblkhd)) != 0 &&
935 		    flushbuflist(blist, flags, vp, slpflag, slptimeo, &error)) {
936 			if (error)
937 				break;
938 			continue;
939 		}
940 		if ((blist = TAILQ_FIRST(&vp->v_dirtyblkhd)) != 0 &&
941 		    flushbuflist(blist, flags, vp, slpflag, slptimeo, &error)) {
942 			if (error)
943 				break;
944 			continue;
945 		}
946 		break;
947 	}
948 	if (error) {
949 		VI_UNLOCK(vp);
950 		return (error);
951 	}
952 
953 	/*
954 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
955 	 * have write I/O in-progress but if there is a VM object then the
956 	 * VM object can also have read-I/O in-progress.
957 	 */
958 	do {
959 		while (vp->v_numoutput > 0) {
960 			vp->v_iflag |= VI_BWAIT;
961 			msleep(&vp->v_numoutput, VI_MTX(vp), PVM, "vnvlbv", 0);
962 		}
963 		VI_UNLOCK(vp);
964 		if (VOP_GETVOBJECT(vp, &object) == 0) {
965 			VM_OBJECT_LOCK(object);
966 			vm_object_pip_wait(object, "vnvlbx");
967 			VM_OBJECT_UNLOCK(object);
968 		}
969 		VI_LOCK(vp);
970 	} while (vp->v_numoutput > 0);
971 	VI_UNLOCK(vp);
972 
973 	/*
974 	 * Destroy the copy in the VM cache, too.
975 	 */
976 	if (VOP_GETVOBJECT(vp, &object) == 0) {
977 		VM_OBJECT_LOCK(object);
978 		vm_object_page_remove(object, 0, 0,
979 			(flags & V_SAVE) ? TRUE : FALSE);
980 		VM_OBJECT_UNLOCK(object);
981 	}
982 
983 #ifdef INVARIANTS
984 	VI_LOCK(vp);
985 	if ((flags & (V_ALT | V_NORMAL)) == 0 &&
986 	    (!TAILQ_EMPTY(&vp->v_dirtyblkhd) ||
987 	     !TAILQ_EMPTY(&vp->v_cleanblkhd)))
988 		panic("vinvalbuf: flush failed");
989 	VI_UNLOCK(vp);
990 #endif
991 	return (0);
992 }
993 
994 /*
995  * Flush out buffers on the specified list.
996  *
997  */
998 static int
999 flushbuflist(blist, flags, vp, slpflag, slptimeo, errorp)
1000 	struct buf *blist;
1001 	int flags;
1002 	struct vnode *vp;
1003 	int slpflag, slptimeo;
1004 	int *errorp;
1005 {
1006 	struct buf *bp, *nbp;
1007 	int found, error;
1008 
1009 	ASSERT_VI_LOCKED(vp, "flushbuflist");
1010 
1011 	for (found = 0, bp = blist; bp; bp = nbp) {
1012 		nbp = TAILQ_NEXT(bp, b_vnbufs);
1013 		if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1014 		    ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1015 			continue;
1016 		}
1017 		found += 1;
1018 		error = BUF_TIMELOCK(bp,
1019 		    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, VI_MTX(vp),
1020 		    "flushbuf", slpflag, slptimeo);
1021 		if (error) {
1022 			if (error != ENOLCK)
1023 				*errorp = error;
1024 			goto done;
1025 		}
1026 		/*
1027 		 * XXX Since there are no node locks for NFS, I
1028 		 * believe there is a slight chance that a delayed
1029 		 * write will occur while sleeping just above, so
1030 		 * check for it.  Note that vfs_bio_awrite expects
1031 		 * buffers to reside on a queue, while bwrite and
1032 		 * brelse do not.
1033 		 */
1034 		if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1035 			(flags & V_SAVE)) {
1036 
1037 			if (bp->b_vp == vp) {
1038 				if (bp->b_flags & B_CLUSTEROK) {
1039 					vfs_bio_awrite(bp);
1040 				} else {
1041 					bremfree(bp);
1042 					bp->b_flags |= B_ASYNC;
1043 					bwrite(bp);
1044 				}
1045 			} else {
1046 				bremfree(bp);
1047 				(void) bwrite(bp);
1048 			}
1049 			goto done;
1050 		}
1051 		bremfree(bp);
1052 		bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
1053 		bp->b_flags &= ~B_ASYNC;
1054 		brelse(bp);
1055 		VI_LOCK(vp);
1056 	}
1057 	return (found);
1058 done:
1059 	VI_LOCK(vp);
1060 	return (found);
1061 }
1062 
1063 /*
1064  * Truncate a file's buffer and pages to a specified length.  This
1065  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1066  * sync activity.
1067  */
1068 int
1069 vtruncbuf(vp, cred, td, length, blksize)
1070 	register struct vnode *vp;
1071 	struct ucred *cred;
1072 	struct thread *td;
1073 	off_t length;
1074 	int blksize;
1075 {
1076 	register struct buf *bp;
1077 	struct buf *nbp;
1078 	int anyfreed;
1079 	int trunclbn;
1080 
1081 	/*
1082 	 * Round up to the *next* lbn.
1083 	 */
1084 	trunclbn = (length + blksize - 1) / blksize;
1085 
1086 	ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1087 restart:
1088 	VI_LOCK(vp);
1089 	anyfreed = 1;
1090 	for (;anyfreed;) {
1091 		anyfreed = 0;
1092 		for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
1093 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1094 			if (bp->b_lblkno >= trunclbn) {
1095 				if (BUF_LOCK(bp,
1096 				    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1097 				    VI_MTX(vp)) == ENOLCK)
1098 					goto restart;
1099 
1100 				bremfree(bp);
1101 				bp->b_flags |= (B_INVAL | B_RELBUF);
1102 				bp->b_flags &= ~B_ASYNC;
1103 				brelse(bp);
1104 				anyfreed = 1;
1105 
1106 				if (nbp &&
1107 				    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1108 				    (nbp->b_vp != vp) ||
1109 				    (nbp->b_flags & B_DELWRI))) {
1110 					goto restart;
1111 				}
1112 				VI_LOCK(vp);
1113 			}
1114 		}
1115 
1116 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1117 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1118 			if (bp->b_lblkno >= trunclbn) {
1119 				if (BUF_LOCK(bp,
1120 				    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1121 				    VI_MTX(vp)) == ENOLCK)
1122 					goto restart;
1123 				bremfree(bp);
1124 				bp->b_flags |= (B_INVAL | B_RELBUF);
1125 				bp->b_flags &= ~B_ASYNC;
1126 				brelse(bp);
1127 				anyfreed = 1;
1128 				if (nbp &&
1129 				    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1130 				    (nbp->b_vp != vp) ||
1131 				    (nbp->b_flags & B_DELWRI) == 0)) {
1132 					goto restart;
1133 				}
1134 				VI_LOCK(vp);
1135 			}
1136 		}
1137 	}
1138 
1139 	if (length > 0) {
1140 restartsync:
1141 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1142 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1143 			if (bp->b_lblkno > 0)
1144 				continue;
1145 			/*
1146 			 * Since we hold the vnode lock this should only
1147 			 * fail if we're racing with the buf daemon.
1148 			 */
1149 			if (BUF_LOCK(bp,
1150 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1151 			    VI_MTX(vp)) == ENOLCK) {
1152 				goto restart;
1153 			}
1154 			KASSERT((bp->b_flags & B_DELWRI),
1155 			    ("buf(%p) on dirty queue without DELWRI", bp));
1156 
1157 			bremfree(bp);
1158 			bawrite(bp);
1159 			VI_LOCK(vp);
1160 			goto restartsync;
1161 		}
1162 	}
1163 
1164 	while (vp->v_numoutput > 0) {
1165 		vp->v_iflag |= VI_BWAIT;
1166 		msleep(&vp->v_numoutput, VI_MTX(vp), PVM, "vbtrunc", 0);
1167 	}
1168 	VI_UNLOCK(vp);
1169 	vnode_pager_setsize(vp, length);
1170 
1171 	return (0);
1172 }
1173 
1174 /*
1175  * buf_splay() - splay tree core for the clean/dirty list of buffers in
1176  * 		 a vnode.
1177  *
1178  *	NOTE: We have to deal with the special case of a background bitmap
1179  *	buffer, a situation where two buffers will have the same logical
1180  *	block offset.  We want (1) only the foreground buffer to be accessed
1181  *	in a lookup and (2) must differentiate between the foreground and
1182  *	background buffer in the splay tree algorithm because the splay
1183  *	tree cannot normally handle multiple entities with the same 'index'.
1184  *	We accomplish this by adding differentiating flags to the splay tree's
1185  *	numerical domain.
1186  */
1187 static
1188 struct buf *
1189 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1190 {
1191 	struct buf dummy;
1192 	struct buf *lefttreemax, *righttreemin, *y;
1193 
1194 	if (root == NULL)
1195 		return (NULL);
1196 	lefttreemax = righttreemin = &dummy;
1197 	for (;;) {
1198 		if (lblkno < root->b_lblkno ||
1199 		    (lblkno == root->b_lblkno &&
1200 		    (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1201 			if ((y = root->b_left) == NULL)
1202 				break;
1203 			if (lblkno < y->b_lblkno) {
1204 				/* Rotate right. */
1205 				root->b_left = y->b_right;
1206 				y->b_right = root;
1207 				root = y;
1208 				if ((y = root->b_left) == NULL)
1209 					break;
1210 			}
1211 			/* Link into the new root's right tree. */
1212 			righttreemin->b_left = root;
1213 			righttreemin = root;
1214 		} else if (lblkno > root->b_lblkno ||
1215 		    (lblkno == root->b_lblkno &&
1216 		    (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1217 			if ((y = root->b_right) == NULL)
1218 				break;
1219 			if (lblkno > y->b_lblkno) {
1220 				/* Rotate left. */
1221 				root->b_right = y->b_left;
1222 				y->b_left = root;
1223 				root = y;
1224 				if ((y = root->b_right) == NULL)
1225 					break;
1226 			}
1227 			/* Link into the new root's left tree. */
1228 			lefttreemax->b_right = root;
1229 			lefttreemax = root;
1230 		} else {
1231 			break;
1232 		}
1233 		root = y;
1234 	}
1235 	/* Assemble the new root. */
1236 	lefttreemax->b_right = root->b_left;
1237 	righttreemin->b_left = root->b_right;
1238 	root->b_left = dummy.b_right;
1239 	root->b_right = dummy.b_left;
1240 	return (root);
1241 }
1242 
1243 static
1244 void
1245 buf_vlist_remove(struct buf *bp)
1246 {
1247 	struct vnode *vp = bp->b_vp;
1248 	struct buf *root;
1249 
1250 	ASSERT_VI_LOCKED(vp, "buf_vlist_remove");
1251 	if (bp->b_xflags & BX_VNDIRTY) {
1252 		if (bp != vp->v_dirtyblkroot) {
1253 			root = buf_splay(bp->b_lblkno, bp->b_xflags,
1254 			    vp->v_dirtyblkroot);
1255 			KASSERT(root == bp,
1256 			    ("splay lookup failed during dirty remove"));
1257 		}
1258 		if (bp->b_left == NULL) {
1259 			root = bp->b_right;
1260 		} else {
1261 			root = buf_splay(bp->b_lblkno, bp->b_xflags,
1262 			    bp->b_left);
1263 			root->b_right = bp->b_right;
1264 		}
1265 		vp->v_dirtyblkroot = root;
1266 		TAILQ_REMOVE(&vp->v_dirtyblkhd, bp, b_vnbufs);
1267 		vp->v_dirtybufcnt--;
1268 	} else {
1269 		/* KASSERT(bp->b_xflags & BX_VNCLEAN, ("bp wasn't clean")); */
1270 		if (bp != vp->v_cleanblkroot) {
1271 			root = buf_splay(bp->b_lblkno, bp->b_xflags,
1272 			    vp->v_cleanblkroot);
1273 			KASSERT(root == bp,
1274 			    ("splay lookup failed during clean remove"));
1275 		}
1276 		if (bp->b_left == NULL) {
1277 			root = bp->b_right;
1278 		} else {
1279 			root = buf_splay(bp->b_lblkno, bp->b_xflags,
1280 			    bp->b_left);
1281 			root->b_right = bp->b_right;
1282 		}
1283 		vp->v_cleanblkroot = root;
1284 		TAILQ_REMOVE(&vp->v_cleanblkhd, bp, b_vnbufs);
1285 		vp->v_cleanbufcnt--;
1286 	}
1287 	bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1288 }
1289 
1290 /*
1291  * Add the buffer to the sorted clean or dirty block list using a
1292  * splay tree algorithm.
1293  *
1294  * NOTE: xflags is passed as a constant, optimizing this inline function!
1295  */
1296 static
1297 void
1298 buf_vlist_add(struct buf *bp, struct vnode *vp, b_xflags_t xflags)
1299 {
1300 	struct buf *root;
1301 
1302 	ASSERT_VI_LOCKED(vp, "buf_vlist_add");
1303 	bp->b_xflags |= xflags;
1304 	if (xflags & BX_VNDIRTY) {
1305 		root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_dirtyblkroot);
1306 		if (root == NULL) {
1307 			bp->b_left = NULL;
1308 			bp->b_right = NULL;
1309 			TAILQ_INSERT_TAIL(&vp->v_dirtyblkhd, bp, b_vnbufs);
1310 		} else if (bp->b_lblkno < root->b_lblkno ||
1311 		    (bp->b_lblkno == root->b_lblkno &&
1312 		    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1313 			bp->b_left = root->b_left;
1314 			bp->b_right = root;
1315 			root->b_left = NULL;
1316 			TAILQ_INSERT_BEFORE(root, bp, b_vnbufs);
1317 		} else {
1318 			bp->b_right = root->b_right;
1319 			bp->b_left = root;
1320 			root->b_right = NULL;
1321 			TAILQ_INSERT_AFTER(&vp->v_dirtyblkhd,
1322 			    root, bp, b_vnbufs);
1323 		}
1324 		vp->v_dirtybufcnt++;
1325 		vp->v_dirtyblkroot = bp;
1326 	} else {
1327 		/* KASSERT(xflags & BX_VNCLEAN, ("xflags not clean")); */
1328 		root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_cleanblkroot);
1329 		if (root == NULL) {
1330 			bp->b_left = NULL;
1331 			bp->b_right = NULL;
1332 			TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
1333 		} else if (bp->b_lblkno < root->b_lblkno ||
1334 		    (bp->b_lblkno == root->b_lblkno &&
1335 		    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1336 			bp->b_left = root->b_left;
1337 			bp->b_right = root;
1338 			root->b_left = NULL;
1339 			TAILQ_INSERT_BEFORE(root, bp, b_vnbufs);
1340 		} else {
1341 			bp->b_right = root->b_right;
1342 			bp->b_left = root;
1343 			root->b_right = NULL;
1344 			TAILQ_INSERT_AFTER(&vp->v_cleanblkhd,
1345 			    root, bp, b_vnbufs);
1346 		}
1347 		vp->v_cleanbufcnt++;
1348 		vp->v_cleanblkroot = bp;
1349 	}
1350 }
1351 
1352 /*
1353  * Lookup a buffer using the splay tree.  Note that we specifically avoid
1354  * shadow buffers used in background bitmap writes.
1355  *
1356  * This code isn't quite efficient as it could be because we are maintaining
1357  * two sorted lists and do not know which list the block resides in.
1358  *
1359  * During a "make buildworld" the desired buffer is found at one of
1360  * the roots more than 60% of the time.  Thus, checking both roots
1361  * before performing either splay eliminates unnecessary splays on the
1362  * first tree splayed.
1363  */
1364 struct buf *
1365 gbincore(struct vnode *vp, daddr_t lblkno)
1366 {
1367 	struct buf *bp;
1368 
1369 	GIANT_REQUIRED;
1370 
1371 	ASSERT_VI_LOCKED(vp, "gbincore");
1372 	if ((bp = vp->v_cleanblkroot) != NULL &&
1373 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1374 		return (bp);
1375 	if ((bp = vp->v_dirtyblkroot) != NULL &&
1376 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1377 		return (bp);
1378 	if ((bp = vp->v_cleanblkroot) != NULL) {
1379 		vp->v_cleanblkroot = bp = buf_splay(lblkno, 0, bp);
1380 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1381 			return (bp);
1382 	}
1383 	if ((bp = vp->v_dirtyblkroot) != NULL) {
1384 		vp->v_dirtyblkroot = bp = buf_splay(lblkno, 0, bp);
1385 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1386 			return (bp);
1387 	}
1388 	return (NULL);
1389 }
1390 
1391 /*
1392  * Associate a buffer with a vnode.
1393  */
1394 void
1395 bgetvp(vp, bp)
1396 	register struct vnode *vp;
1397 	register struct buf *bp;
1398 {
1399 
1400 	KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
1401 
1402 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1403 	    ("bgetvp: bp already attached! %p", bp));
1404 
1405 	ASSERT_VI_LOCKED(vp, "bgetvp");
1406 	vholdl(vp);
1407 	bp->b_vp = vp;
1408 	bp->b_dev = vn_todev(vp);
1409 	/*
1410 	 * Insert onto list for new vnode.
1411 	 */
1412 	buf_vlist_add(bp, vp, BX_VNCLEAN);
1413 }
1414 
1415 /*
1416  * Disassociate a buffer from a vnode.
1417  */
1418 void
1419 brelvp(bp)
1420 	register struct buf *bp;
1421 {
1422 	struct vnode *vp;
1423 
1424 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1425 
1426 	/*
1427 	 * Delete from old vnode list, if on one.
1428 	 */
1429 	vp = bp->b_vp;
1430 	VI_LOCK(vp);
1431 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1432 		buf_vlist_remove(bp);
1433 	if ((vp->v_iflag & VI_ONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1434 		vp->v_iflag &= ~VI_ONWORKLST;
1435 		mtx_lock(&sync_mtx);
1436 		LIST_REMOVE(vp, v_synclist);
1437 		mtx_unlock(&sync_mtx);
1438 	}
1439 	vdropl(vp);
1440 	bp->b_vp = (struct vnode *) 0;
1441 	if (bp->b_object)
1442 		bp->b_object = NULL;
1443 	VI_UNLOCK(vp);
1444 }
1445 
1446 /*
1447  * Add an item to the syncer work queue.
1448  */
1449 static void
1450 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
1451 {
1452 	int slot;
1453 
1454 	ASSERT_VI_LOCKED(vp, "vn_syncer_add_to_worklist");
1455 
1456 	mtx_lock(&sync_mtx);
1457 	if (vp->v_iflag & VI_ONWORKLST)
1458 		LIST_REMOVE(vp, v_synclist);
1459 	else
1460 		vp->v_iflag |= VI_ONWORKLST;
1461 
1462 	if (delay > syncer_maxdelay - 2)
1463 		delay = syncer_maxdelay - 2;
1464 	slot = (syncer_delayno + delay) & syncer_mask;
1465 
1466 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
1467 	mtx_unlock(&sync_mtx);
1468 }
1469 
1470 struct  proc *updateproc;
1471 static void sched_sync(void);
1472 static struct kproc_desc up_kp = {
1473 	"syncer",
1474 	sched_sync,
1475 	&updateproc
1476 };
1477 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1478 
1479 /*
1480  * System filesystem synchronizer daemon.
1481  */
1482 static void
1483 sched_sync(void)
1484 {
1485 	struct synclist *next;
1486 	struct synclist *slp;
1487 	struct vnode *vp;
1488 	struct mount *mp;
1489 	long starttime;
1490 	struct thread *td = FIRST_THREAD_IN_PROC(updateproc);
1491 
1492 	mtx_lock(&Giant);
1493 
1494 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, td->td_proc,
1495 	    SHUTDOWN_PRI_LAST);
1496 
1497 	for (;;) {
1498 		kthread_suspend_check(td->td_proc);
1499 
1500 		starttime = time_second;
1501 
1502 		/*
1503 		 * Push files whose dirty time has expired.  Be careful
1504 		 * of interrupt race on slp queue.
1505 		 */
1506 		mtx_lock(&sync_mtx);
1507 		slp = &syncer_workitem_pending[syncer_delayno];
1508 		syncer_delayno += 1;
1509 		if (syncer_delayno == syncer_maxdelay)
1510 			syncer_delayno = 0;
1511 		next = &syncer_workitem_pending[syncer_delayno];
1512 
1513 		while ((vp = LIST_FIRST(slp)) != NULL) {
1514 			if (VOP_ISLOCKED(vp, NULL) != 0 ||
1515 			    vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1516 				LIST_REMOVE(vp, v_synclist);
1517 				LIST_INSERT_HEAD(next, vp, v_synclist);
1518 				continue;
1519 			}
1520 			if (VI_TRYLOCK(vp) == 0) {
1521 				LIST_REMOVE(vp, v_synclist);
1522 				LIST_INSERT_HEAD(next, vp, v_synclist);
1523 				vn_finished_write(mp);
1524 				continue;
1525 			}
1526 			/*
1527 			 * We use vhold in case the vnode does not
1528 			 * successfully sync.  vhold prevents the vnode from
1529 			 * going away when we unlock the sync_mtx so that
1530 			 * we can acquire the vnode interlock.
1531 			 */
1532 			vholdl(vp);
1533 			mtx_unlock(&sync_mtx);
1534 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK, td);
1535 			(void) VOP_FSYNC(vp, td->td_ucred, MNT_LAZY, td);
1536 			VOP_UNLOCK(vp, 0, td);
1537 			vn_finished_write(mp);
1538 			VI_LOCK(vp);
1539 			if ((vp->v_iflag & VI_ONWORKLST) != 0) {
1540 				/*
1541 				 * Put us back on the worklist.  The worklist
1542 				 * routine will remove us from our current
1543 				 * position and then add us back in at a later
1544 				 * position.
1545 				 */
1546 				vn_syncer_add_to_worklist(vp, syncdelay);
1547 			}
1548 			vdropl(vp);
1549 			VI_UNLOCK(vp);
1550 			mtx_lock(&sync_mtx);
1551 		}
1552 		mtx_unlock(&sync_mtx);
1553 
1554 		/*
1555 		 * Do soft update processing.
1556 		 */
1557 		if (softdep_process_worklist_hook != NULL)
1558 			(*softdep_process_worklist_hook)(NULL);
1559 
1560 		/*
1561 		 * The variable rushjob allows the kernel to speed up the
1562 		 * processing of the filesystem syncer process. A rushjob
1563 		 * value of N tells the filesystem syncer to process the next
1564 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1565 		 * is used by the soft update code to speed up the filesystem
1566 		 * syncer process when the incore state is getting so far
1567 		 * ahead of the disk that the kernel memory pool is being
1568 		 * threatened with exhaustion.
1569 		 */
1570 		mtx_lock(&sync_mtx);
1571 		if (rushjob > 0) {
1572 			rushjob -= 1;
1573 			mtx_unlock(&sync_mtx);
1574 			continue;
1575 		}
1576 		mtx_unlock(&sync_mtx);
1577 		/*
1578 		 * If it has taken us less than a second to process the
1579 		 * current work, then wait. Otherwise start right over
1580 		 * again. We can still lose time if any single round
1581 		 * takes more than two seconds, but it does not really
1582 		 * matter as we are just trying to generally pace the
1583 		 * filesystem activity.
1584 		 */
1585 		if (time_second == starttime)
1586 			tsleep(&lbolt, PPAUSE, "syncer", 0);
1587 	}
1588 }
1589 
1590 /*
1591  * Request the syncer daemon to speed up its work.
1592  * We never push it to speed up more than half of its
1593  * normal turn time, otherwise it could take over the cpu.
1594  */
1595 int
1596 speedup_syncer()
1597 {
1598 	struct thread *td;
1599 	int ret = 0;
1600 
1601 	td = FIRST_THREAD_IN_PROC(updateproc);
1602 	sleepq_remove(td, &lbolt);
1603 	mtx_lock(&sync_mtx);
1604 	if (rushjob < syncdelay / 2) {
1605 		rushjob += 1;
1606 		stat_rush_requests += 1;
1607 		ret = 1;
1608 	}
1609 	mtx_unlock(&sync_mtx);
1610 	return (ret);
1611 }
1612 
1613 /*
1614  * Associate a p-buffer with a vnode.
1615  *
1616  * Also sets B_PAGING flag to indicate that vnode is not fully associated
1617  * with the buffer.  i.e. the bp has not been linked into the vnode or
1618  * ref-counted.
1619  */
1620 void
1621 pbgetvp(vp, bp)
1622 	register struct vnode *vp;
1623 	register struct buf *bp;
1624 {
1625 
1626 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1627 
1628 	bp->b_vp = vp;
1629 	bp->b_object = vp->v_object;
1630 	bp->b_flags |= B_PAGING;
1631 	bp->b_dev = vn_todev(vp);
1632 }
1633 
1634 /*
1635  * Disassociate a p-buffer from a vnode.
1636  */
1637 void
1638 pbrelvp(bp)
1639 	register struct buf *bp;
1640 {
1641 
1642 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1643 
1644 	/* XXX REMOVE ME */
1645 	VI_LOCK(bp->b_vp);
1646 	if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1647 		panic(
1648 		    "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1649 		    bp,
1650 		    (int)bp->b_flags
1651 		);
1652 	}
1653 	VI_UNLOCK(bp->b_vp);
1654 	bp->b_vp = (struct vnode *) 0;
1655 	bp->b_object = NULL;
1656 	bp->b_flags &= ~B_PAGING;
1657 }
1658 
1659 /*
1660  * Reassign a buffer from one vnode to another.
1661  * Used to assign file specific control information
1662  * (indirect blocks) to the vnode to which they belong.
1663  */
1664 void
1665 reassignbuf(bp, newvp)
1666 	register struct buf *bp;
1667 	register struct vnode *newvp;
1668 {
1669 	struct vnode *vp;
1670 	int delay;
1671 
1672 	if (newvp == NULL) {
1673 		printf("reassignbuf: NULL");
1674 		return;
1675 	}
1676 	vp = bp->b_vp;
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 	/*
1687 	 * Delete from old vnode list, if on one.
1688 	 */
1689 	VI_LOCK(vp);
1690 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1691 		buf_vlist_remove(bp);
1692 		if (vp != newvp) {
1693 			vdropl(bp->b_vp);
1694 			bp->b_vp = NULL;	/* for clarification */
1695 		}
1696 	}
1697 	if (vp != newvp) {
1698 		VI_UNLOCK(vp);
1699 		VI_LOCK(newvp);
1700 	}
1701 	/*
1702 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1703 	 * of clean buffers.
1704 	 */
1705 	if (bp->b_flags & B_DELWRI) {
1706 		if ((newvp->v_iflag & VI_ONWORKLST) == 0) {
1707 			switch (newvp->v_type) {
1708 			case VDIR:
1709 				delay = dirdelay;
1710 				break;
1711 			case VCHR:
1712 				if (newvp->v_rdev->si_mountpoint != NULL) {
1713 					delay = metadelay;
1714 					break;
1715 				}
1716 				/* FALLTHROUGH */
1717 			default:
1718 				delay = filedelay;
1719 			}
1720 			vn_syncer_add_to_worklist(newvp, delay);
1721 		}
1722 		buf_vlist_add(bp, newvp, BX_VNDIRTY);
1723 	} else {
1724 		buf_vlist_add(bp, newvp, BX_VNCLEAN);
1725 
1726 		if ((newvp->v_iflag & VI_ONWORKLST) &&
1727 		    TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1728 			mtx_lock(&sync_mtx);
1729 			LIST_REMOVE(newvp, v_synclist);
1730 			mtx_unlock(&sync_mtx);
1731 			newvp->v_iflag &= ~VI_ONWORKLST;
1732 		}
1733 	}
1734 	if (bp->b_vp != newvp) {
1735 		bp->b_vp = newvp;
1736 		vholdl(bp->b_vp);
1737 	}
1738 	VI_UNLOCK(newvp);
1739 }
1740 
1741 /*
1742  * Create a vnode for a device.
1743  * Used for mounting the root filesystem.
1744  */
1745 int
1746 bdevvp(dev, vpp)
1747 	dev_t dev;
1748 	struct vnode **vpp;
1749 {
1750 	register struct vnode *vp;
1751 	struct vnode *nvp;
1752 	int error;
1753 
1754 	if (dev == NODEV) {
1755 		*vpp = NULLVP;
1756 		return (ENXIO);
1757 	}
1758 	if (vfinddev(dev, vpp))
1759 		return (0);
1760 
1761 	error = getnewvnode("none", (struct mount *)0, spec_vnodeop_p, &nvp);
1762 	if (error) {
1763 		*vpp = NULLVP;
1764 		return (error);
1765 	}
1766 	vp = nvp;
1767 	vp->v_type = VCHR;
1768 	addalias(vp, dev);
1769 	*vpp = vp;
1770 	return (0);
1771 }
1772 
1773 static void
1774 v_incr_usecount(struct vnode *vp, int delta)
1775 {
1776 
1777 	vp->v_usecount += delta;
1778 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1779 		mtx_lock(&spechash_mtx);
1780 		vp->v_rdev->si_usecount += delta;
1781 		mtx_unlock(&spechash_mtx);
1782 	}
1783 }
1784 
1785 /*
1786  * Add vnode to the alias list hung off the dev_t.
1787  *
1788  * The reason for this gunk is that multiple vnodes can reference
1789  * the same physical device, so checking vp->v_usecount to see
1790  * how many users there are is inadequate; the v_usecount for
1791  * the vnodes need to be accumulated.  vcount() does that.
1792  */
1793 struct vnode *
1794 addaliasu(nvp, nvp_rdev)
1795 	struct vnode *nvp;
1796 	udev_t nvp_rdev;
1797 {
1798 	struct vnode *ovp;
1799 	vop_t **ops;
1800 	dev_t dev;
1801 
1802 	if (nvp->v_type == VBLK)
1803 		return (nvp);
1804 	if (nvp->v_type != VCHR)
1805 		panic("addaliasu on non-special vnode");
1806 	dev = udev2dev(nvp_rdev);
1807 	if (dev == NODEV)
1808 		return (nvp);
1809 	/*
1810 	 * Check to see if we have a bdevvp vnode with no associated
1811 	 * filesystem. If so, we want to associate the filesystem of
1812 	 * the new newly instigated vnode with the bdevvp vnode and
1813 	 * discard the newly created vnode rather than leaving the
1814 	 * bdevvp vnode lying around with no associated filesystem.
1815 	 */
1816 	if (vfinddev(dev, &ovp) == 0 || ovp->v_data != NULL) {
1817 		addalias(nvp, dev);
1818 		return (nvp);
1819 	}
1820 	/*
1821 	 * Discard unneeded vnode, but save its node specific data.
1822 	 * Note that if there is a lock, it is carried over in the
1823 	 * node specific data to the replacement vnode.
1824 	 */
1825 	vref(ovp);
1826 	ovp->v_data = nvp->v_data;
1827 	ovp->v_tag = nvp->v_tag;
1828 	nvp->v_data = NULL;
1829 	lockdestroy(ovp->v_vnlock);
1830 	lockinit(ovp->v_vnlock, PVFS, nvp->v_vnlock->lk_wmesg,
1831 	    nvp->v_vnlock->lk_timo, nvp->v_vnlock->lk_flags & LK_EXTFLG_MASK);
1832 	ops = ovp->v_op;
1833 	ovp->v_op = nvp->v_op;
1834 	if (VOP_ISLOCKED(nvp, curthread)) {
1835 		VOP_UNLOCK(nvp, 0, curthread);
1836 		vn_lock(ovp, LK_EXCLUSIVE | LK_RETRY, curthread);
1837 	}
1838 	nvp->v_op = ops;
1839 	insmntque(ovp, nvp->v_mount);
1840 	vrele(nvp);
1841 	vgone(nvp);
1842 	return (ovp);
1843 }
1844 
1845 /* This is a local helper function that do the same as addaliasu, but for a
1846  * dev_t instead of an udev_t. */
1847 static void
1848 addalias(nvp, dev)
1849 	struct vnode *nvp;
1850 	dev_t dev;
1851 {
1852 
1853 	KASSERT(nvp->v_type == VCHR, ("addalias on non-special vnode"));
1854 	dev_ref(dev);
1855 	nvp->v_rdev = dev;
1856 	VI_LOCK(nvp);
1857 	mtx_lock(&spechash_mtx);
1858 	SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
1859 	dev->si_usecount += nvp->v_usecount;
1860 	mtx_unlock(&spechash_mtx);
1861 	VI_UNLOCK(nvp);
1862 }
1863 
1864 /*
1865  * Grab a particular vnode from the free list, increment its
1866  * reference count and lock it. The vnode lock bit is set if the
1867  * vnode is being eliminated in vgone. The process is awakened
1868  * when the transition is completed, and an error returned to
1869  * indicate that the vnode is no longer usable (possibly having
1870  * been changed to a new filesystem type).
1871  */
1872 int
1873 vget(vp, flags, td)
1874 	register struct vnode *vp;
1875 	int flags;
1876 	struct thread *td;
1877 {
1878 	int error;
1879 
1880 	/*
1881 	 * If the vnode is in the process of being cleaned out for
1882 	 * another use, we wait for the cleaning to finish and then
1883 	 * return failure. Cleaning is determined by checking that
1884 	 * the VI_XLOCK flag is set.
1885 	 */
1886 	if ((flags & LK_INTERLOCK) == 0)
1887 		VI_LOCK(vp);
1888 	if (vp->v_iflag & VI_XLOCK && vp->v_vxthread != curthread) {
1889 		if ((flags & LK_NOWAIT) == 0) {
1890 			vp->v_iflag |= VI_XWANT;
1891 			msleep(vp, VI_MTX(vp), PINOD | PDROP, "vget", 0);
1892 			return (ENOENT);
1893 		}
1894 		VI_UNLOCK(vp);
1895 		return (EBUSY);
1896 	}
1897 
1898 	v_incr_usecount(vp, 1);
1899 
1900 	if (VSHOULDBUSY(vp))
1901 		vbusy(vp);
1902 	if (flags & LK_TYPE_MASK) {
1903 		if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
1904 			/*
1905 			 * must expand vrele here because we do not want
1906 			 * to call VOP_INACTIVE if the reference count
1907 			 * drops back to zero since it was never really
1908 			 * active. We must remove it from the free list
1909 			 * before sleeping so that multiple processes do
1910 			 * not try to recycle it.
1911 			 */
1912 			VI_LOCK(vp);
1913 			v_incr_usecount(vp, -1);
1914 			if (VSHOULDFREE(vp))
1915 				vfree(vp);
1916 			else
1917 				vlruvp(vp);
1918 			VI_UNLOCK(vp);
1919 		}
1920 		return (error);
1921 	}
1922 	VI_UNLOCK(vp);
1923 	return (0);
1924 }
1925 
1926 /*
1927  * Increase the reference count of a vnode.
1928  */
1929 void
1930 vref(struct vnode *vp)
1931 {
1932 
1933 	VI_LOCK(vp);
1934 	v_incr_usecount(vp, 1);
1935 	VI_UNLOCK(vp);
1936 }
1937 
1938 /*
1939  * Return reference count of a vnode.
1940  *
1941  * The results of this call are only guaranteed when some mechanism other
1942  * than the VI lock is used to stop other processes from gaining references
1943  * to the vnode.  This may be the case if the caller holds the only reference.
1944  * This is also useful when stale data is acceptable as race conditions may
1945  * be accounted for by some other means.
1946  */
1947 int
1948 vrefcnt(struct vnode *vp)
1949 {
1950 	int usecnt;
1951 
1952 	VI_LOCK(vp);
1953 	usecnt = vp->v_usecount;
1954 	VI_UNLOCK(vp);
1955 
1956 	return (usecnt);
1957 }
1958 
1959 
1960 /*
1961  * Vnode put/release.
1962  * If count drops to zero, call inactive routine and return to freelist.
1963  */
1964 void
1965 vrele(vp)
1966 	struct vnode *vp;
1967 {
1968 	struct thread *td = curthread;	/* XXX */
1969 
1970 	KASSERT(vp != NULL, ("vrele: null vp"));
1971 
1972 	VI_LOCK(vp);
1973 
1974 	/* Skip this v_writecount check if we're going to panic below. */
1975 	KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
1976 	    ("vrele: missed vn_close"));
1977 
1978 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
1979 	    vp->v_usecount == 1)) {
1980 		v_incr_usecount(vp, -1);
1981 		VI_UNLOCK(vp);
1982 
1983 		return;
1984 	}
1985 
1986 	if (vp->v_usecount == 1) {
1987 		v_incr_usecount(vp, -1);
1988 		/*
1989 		 * We must call VOP_INACTIVE with the node locked. Mark
1990 		 * as VI_DOINGINACT to avoid recursion.
1991 		 */
1992 		if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0) {
1993 			VI_LOCK(vp);
1994 			vp->v_iflag |= VI_DOINGINACT;
1995 			VI_UNLOCK(vp);
1996 			VOP_INACTIVE(vp, td);
1997 			VI_LOCK(vp);
1998 			KASSERT(vp->v_iflag & VI_DOINGINACT,
1999 			    ("vrele: lost VI_DOINGINACT"));
2000 			vp->v_iflag &= ~VI_DOINGINACT;
2001 		} else
2002 			VI_LOCK(vp);
2003 		if (VSHOULDFREE(vp))
2004 			vfree(vp);
2005 		else
2006 			vlruvp(vp);
2007 		VI_UNLOCK(vp);
2008 
2009 	} else {
2010 #ifdef DIAGNOSTIC
2011 		vprint("vrele: negative ref count", vp);
2012 #endif
2013 		VI_UNLOCK(vp);
2014 		panic("vrele: negative ref cnt");
2015 	}
2016 }
2017 
2018 /*
2019  * Release an already locked vnode.  This give the same effects as
2020  * unlock+vrele(), but takes less time and avoids releasing and
2021  * re-aquiring the lock (as vrele() aquires the lock internally.)
2022  */
2023 void
2024 vput(vp)
2025 	struct vnode *vp;
2026 {
2027 	struct thread *td = curthread;	/* XXX */
2028 
2029 	GIANT_REQUIRED;
2030 
2031 	KASSERT(vp != NULL, ("vput: null vp"));
2032 	VI_LOCK(vp);
2033 	/* Skip this v_writecount check if we're going to panic below. */
2034 	KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
2035 	    ("vput: missed vn_close"));
2036 
2037 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2038 	    vp->v_usecount == 1)) {
2039 		v_incr_usecount(vp, -1);
2040 		VOP_UNLOCK(vp, LK_INTERLOCK, td);
2041 		return;
2042 	}
2043 
2044 	if (vp->v_usecount == 1) {
2045 		v_incr_usecount(vp, -1);
2046 		/*
2047 		 * We must call VOP_INACTIVE with the node locked, so
2048 		 * we just need to release the vnode mutex. Mark as
2049 		 * as VI_DOINGINACT to avoid recursion.
2050 		 */
2051 		vp->v_iflag |= VI_DOINGINACT;
2052 		VI_UNLOCK(vp);
2053 		VOP_INACTIVE(vp, td);
2054 		VI_LOCK(vp);
2055 		KASSERT(vp->v_iflag & VI_DOINGINACT,
2056 		    ("vput: lost VI_DOINGINACT"));
2057 		vp->v_iflag &= ~VI_DOINGINACT;
2058 		if (VSHOULDFREE(vp))
2059 			vfree(vp);
2060 		else
2061 			vlruvp(vp);
2062 		VI_UNLOCK(vp);
2063 
2064 	} else {
2065 #ifdef DIAGNOSTIC
2066 		vprint("vput: negative ref count", vp);
2067 #endif
2068 		panic("vput: negative ref cnt");
2069 	}
2070 }
2071 
2072 /*
2073  * Somebody doesn't want the vnode recycled.
2074  */
2075 void
2076 vhold(struct vnode *vp)
2077 {
2078 
2079 	VI_LOCK(vp);
2080 	vholdl(vp);
2081 	VI_UNLOCK(vp);
2082 }
2083 
2084 void
2085 vholdl(vp)
2086 	register struct vnode *vp;
2087 {
2088 
2089 	vp->v_holdcnt++;
2090 	if (VSHOULDBUSY(vp))
2091 		vbusy(vp);
2092 }
2093 
2094 /*
2095  * Note that there is one less who cares about this vnode.  vdrop() is the
2096  * opposite of vhold().
2097  */
2098 void
2099 vdrop(struct vnode *vp)
2100 {
2101 
2102 	VI_LOCK(vp);
2103 	vdropl(vp);
2104 	VI_UNLOCK(vp);
2105 }
2106 
2107 void
2108 vdropl(vp)
2109 	register struct vnode *vp;
2110 {
2111 
2112 	if (vp->v_holdcnt <= 0)
2113 		panic("vdrop: holdcnt");
2114 	vp->v_holdcnt--;
2115 	if (VSHOULDFREE(vp))
2116 		vfree(vp);
2117 	else
2118 		vlruvp(vp);
2119 }
2120 
2121 /*
2122  * Remove any vnodes in the vnode table belonging to mount point mp.
2123  *
2124  * If FORCECLOSE is not specified, there should not be any active ones,
2125  * return error if any are found (nb: this is a user error, not a
2126  * system error). If FORCECLOSE is specified, detach any active vnodes
2127  * that are found.
2128  *
2129  * If WRITECLOSE is set, only flush out regular file vnodes open for
2130  * writing.
2131  *
2132  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2133  *
2134  * `rootrefs' specifies the base reference count for the root vnode
2135  * of this filesystem. The root vnode is considered busy if its
2136  * v_usecount exceeds this value. On a successful return, vflush()
2137  * will call vrele() on the root vnode exactly rootrefs times.
2138  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2139  * be zero.
2140  */
2141 #ifdef DIAGNOSTIC
2142 static int busyprt = 0;		/* print out busy vnodes */
2143 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
2144 #endif
2145 
2146 int
2147 vflush(mp, rootrefs, flags)
2148 	struct mount *mp;
2149 	int rootrefs;
2150 	int flags;
2151 {
2152 	struct thread *td = curthread;	/* XXX */
2153 	struct vnode *vp, *nvp, *rootvp = NULL;
2154 	struct vattr vattr;
2155 	int busy = 0, error;
2156 
2157 	if (rootrefs > 0) {
2158 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2159 		    ("vflush: bad args"));
2160 		/*
2161 		 * Get the filesystem root vnode. We can vput() it
2162 		 * immediately, since with rootrefs > 0, it won't go away.
2163 		 */
2164 		if ((error = VFS_ROOT(mp, &rootvp)) != 0)
2165 			return (error);
2166 		vput(rootvp);
2167 
2168 	}
2169 	MNT_ILOCK(mp);
2170 loop:
2171 	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) {
2172 		/*
2173 		 * Make sure this vnode wasn't reclaimed in getnewvnode().
2174 		 * Start over if it has (it won't be on the list anymore).
2175 		 */
2176 		if (vp->v_mount != mp)
2177 			goto loop;
2178 		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2179 
2180 		VI_LOCK(vp);
2181 		MNT_IUNLOCK(mp);
2182 		error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE, td);
2183 		if (error) {
2184 			MNT_ILOCK(mp);
2185 			goto loop;
2186 		}
2187 		/*
2188 		 * Skip over a vnodes marked VV_SYSTEM.
2189 		 */
2190 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2191 			VOP_UNLOCK(vp, 0, td);
2192 			MNT_ILOCK(mp);
2193 			continue;
2194 		}
2195 		/*
2196 		 * If WRITECLOSE is set, flush out unlinked but still open
2197 		 * files (even if open only for reading) and regular file
2198 		 * vnodes open for writing.
2199 		 */
2200 		if (flags & WRITECLOSE) {
2201 			error = VOP_GETATTR(vp, &vattr, td->td_ucred, td);
2202 			VI_LOCK(vp);
2203 
2204 			if ((vp->v_type == VNON ||
2205 			    (error == 0 && vattr.va_nlink > 0)) &&
2206 			    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2207 				VOP_UNLOCK(vp, LK_INTERLOCK, td);
2208 				MNT_ILOCK(mp);
2209 				continue;
2210 			}
2211 		} else
2212 			VI_LOCK(vp);
2213 
2214 		VOP_UNLOCK(vp, 0, td);
2215 
2216 		/*
2217 		 * With v_usecount == 0, all we need to do is clear out the
2218 		 * vnode data structures and we are done.
2219 		 */
2220 		if (vp->v_usecount == 0) {
2221 			vgonel(vp, td);
2222 			MNT_ILOCK(mp);
2223 			continue;
2224 		}
2225 
2226 		/*
2227 		 * If FORCECLOSE is set, forcibly close the vnode. For block
2228 		 * or character devices, revert to an anonymous device. For
2229 		 * all other files, just kill them.
2230 		 */
2231 		if (flags & FORCECLOSE) {
2232 			if (vp->v_type != VCHR)
2233 				vgonel(vp, td);
2234 			else
2235 				vgonechrl(vp, td);
2236 			MNT_ILOCK(mp);
2237 			continue;
2238 		}
2239 #ifdef DIAGNOSTIC
2240 		if (busyprt)
2241 			vprint("vflush: busy vnode", vp);
2242 #endif
2243 		VI_UNLOCK(vp);
2244 		MNT_ILOCK(mp);
2245 		busy++;
2246 	}
2247 	MNT_IUNLOCK(mp);
2248 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2249 		/*
2250 		 * If just the root vnode is busy, and if its refcount
2251 		 * is equal to `rootrefs', then go ahead and kill it.
2252 		 */
2253 		VI_LOCK(rootvp);
2254 		KASSERT(busy > 0, ("vflush: not busy"));
2255 		KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
2256 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2257 			vgonel(rootvp, td);
2258 			busy = 0;
2259 		} else
2260 			VI_UNLOCK(rootvp);
2261 	}
2262 	if (busy)
2263 		return (EBUSY);
2264 	for (; rootrefs > 0; rootrefs--)
2265 		vrele(rootvp);
2266 	return (0);
2267 }
2268 
2269 /*
2270  * This moves a now (likely recyclable) vnode to the end of the
2271  * mountlist.  XXX However, it is temporarily disabled until we
2272  * can clean up ffs_sync() and friends, which have loop restart
2273  * conditions which this code causes to operate O(N^2).
2274  */
2275 static void
2276 vlruvp(struct vnode *vp)
2277 {
2278 #if 0
2279 	struct mount *mp;
2280 
2281 	if ((mp = vp->v_mount) != NULL) {
2282 		MNT_ILOCK(mp);
2283 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
2284 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
2285 		MNT_IUNLOCK(mp);
2286 	}
2287 #endif
2288 }
2289 
2290 static void
2291 vx_lock(struct vnode *vp)
2292 {
2293 
2294 	ASSERT_VI_LOCKED(vp, "vx_lock");
2295 
2296 	/*
2297 	 * Prevent the vnode from being recycled or brought into use while we
2298 	 * clean it out.
2299 	 */
2300 	if (vp->v_iflag & VI_XLOCK)
2301 		panic("vclean: deadlock");
2302 	vp->v_iflag |= VI_XLOCK;
2303 	vp->v_vxthread = curthread;
2304 }
2305 
2306 static void
2307 vx_unlock(struct vnode *vp)
2308 {
2309 	ASSERT_VI_LOCKED(vp, "vx_unlock");
2310 	vp->v_iflag &= ~VI_XLOCK;
2311 	vp->v_vxthread = NULL;
2312 	if (vp->v_iflag & VI_XWANT) {
2313 		vp->v_iflag &= ~VI_XWANT;
2314 		wakeup(vp);
2315 	}
2316 }
2317 
2318 /*
2319  * Disassociate the underlying filesystem from a vnode.
2320  */
2321 static void
2322 vclean(vp, flags, td)
2323 	struct vnode *vp;
2324 	int flags;
2325 	struct thread *td;
2326 {
2327 	int active;
2328 
2329 	ASSERT_VI_LOCKED(vp, "vclean");
2330 	/*
2331 	 * Check to see if the vnode is in use. If so we have to reference it
2332 	 * before we clean it out so that its count cannot fall to zero and
2333 	 * generate a race against ourselves to recycle it.
2334 	 */
2335 	if ((active = vp->v_usecount))
2336 		v_incr_usecount(vp, 1);
2337 
2338 	/*
2339 	 * Even if the count is zero, the VOP_INACTIVE routine may still
2340 	 * have the object locked while it cleans it out. The VOP_LOCK
2341 	 * ensures that the VOP_INACTIVE routine is done with its work.
2342 	 * For active vnodes, it ensures that no other activity can
2343 	 * occur while the underlying object is being cleaned out.
2344 	 */
2345 	VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td);
2346 
2347 	/*
2348 	 * Clean out any buffers associated with the vnode.
2349 	 * If the flush fails, just toss the buffers.
2350 	 */
2351 	if (flags & DOCLOSE) {
2352 		struct buf *bp;
2353 		bp = TAILQ_FIRST(&vp->v_dirtyblkhd);
2354 		if (bp != NULL)
2355 			(void) vn_write_suspend_wait(vp, NULL, V_WAIT);
2356 		if (vinvalbuf(vp, V_SAVE, NOCRED, td, 0, 0) != 0)
2357 			vinvalbuf(vp, 0, NOCRED, td, 0, 0);
2358 	}
2359 
2360 	VOP_DESTROYVOBJECT(vp);
2361 
2362 	/*
2363 	 * Any other processes trying to obtain this lock must first
2364 	 * wait for VXLOCK to clear, then call the new lock operation.
2365 	 */
2366 	VOP_UNLOCK(vp, 0, td);
2367 
2368 	/*
2369 	 * If purging an active vnode, it must be closed and
2370 	 * deactivated before being reclaimed. Note that the
2371 	 * VOP_INACTIVE will unlock the vnode.
2372 	 */
2373 	if (active) {
2374 		if (flags & DOCLOSE)
2375 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2376 		VI_LOCK(vp);
2377 		if ((vp->v_iflag & VI_DOINGINACT) == 0) {
2378 			vp->v_iflag |= VI_DOINGINACT;
2379 			VI_UNLOCK(vp);
2380 			if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT, td) != 0)
2381 				panic("vclean: cannot relock.");
2382 			VOP_INACTIVE(vp, td);
2383 			VI_LOCK(vp);
2384 			KASSERT(vp->v_iflag & VI_DOINGINACT,
2385 			    ("vclean: lost VI_DOINGINACT"));
2386 			vp->v_iflag &= ~VI_DOINGINACT;
2387 		}
2388 		VI_UNLOCK(vp);
2389 	}
2390 	/*
2391 	 * Reclaim the vnode.
2392 	 */
2393 	if (VOP_RECLAIM(vp, td))
2394 		panic("vclean: cannot reclaim");
2395 
2396 	if (active) {
2397 		/*
2398 		 * Inline copy of vrele() since VOP_INACTIVE
2399 		 * has already been called.
2400 		 */
2401 		VI_LOCK(vp);
2402 		v_incr_usecount(vp, -1);
2403 		if (vp->v_usecount <= 0) {
2404 #ifdef INVARIANTS
2405 			if (vp->v_usecount < 0 || vp->v_writecount != 0) {
2406 				vprint("vclean: bad ref count", vp);
2407 				panic("vclean: ref cnt");
2408 			}
2409 #endif
2410 			if (VSHOULDFREE(vp))
2411 				vfree(vp);
2412 		}
2413 		VI_UNLOCK(vp);
2414 	}
2415 	/*
2416 	 * Delete from old mount point vnode list.
2417 	 */
2418 	if (vp->v_mount != NULL)
2419 		insmntque(vp, (struct mount *)0);
2420 	cache_purge(vp);
2421 	VI_LOCK(vp);
2422 	if (VSHOULDFREE(vp))
2423 		vfree(vp);
2424 
2425 	/*
2426 	 * Done with purge, reset to the standard lock and
2427 	 * notify sleepers of the grim news.
2428 	 */
2429 	vp->v_vnlock = &vp->v_lock;
2430 	vp->v_op = dead_vnodeop_p;
2431 	if (vp->v_pollinfo != NULL)
2432 		vn_pollgone(vp);
2433 	vp->v_tag = "none";
2434 }
2435 
2436 /*
2437  * Eliminate all activity associated with the requested vnode
2438  * and with all vnodes aliased to the requested vnode.
2439  */
2440 int
2441 vop_revoke(ap)
2442 	struct vop_revoke_args /* {
2443 		struct vnode *a_vp;
2444 		int a_flags;
2445 	} */ *ap;
2446 {
2447 	struct vnode *vp, *vq;
2448 	dev_t dev;
2449 
2450 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
2451 	vp = ap->a_vp;
2452 	KASSERT((vp->v_type == VCHR), ("vop_revoke: not VCHR"));
2453 
2454 	VI_LOCK(vp);
2455 	/*
2456 	 * If a vgone (or vclean) is already in progress,
2457 	 * wait until it is done and return.
2458 	 */
2459 	if (vp->v_iflag & VI_XLOCK) {
2460 		vp->v_iflag |= VI_XWANT;
2461 		msleep(vp, VI_MTX(vp), PINOD | PDROP,
2462 		    "vop_revokeall", 0);
2463 		return (0);
2464 	}
2465 	VI_UNLOCK(vp);
2466 	dev = vp->v_rdev;
2467 	for (;;) {
2468 		mtx_lock(&spechash_mtx);
2469 		vq = SLIST_FIRST(&dev->si_hlist);
2470 		mtx_unlock(&spechash_mtx);
2471 		if (vq == NULL)
2472 			break;
2473 		vgone(vq);
2474 	}
2475 	return (0);
2476 }
2477 
2478 /*
2479  * Recycle an unused vnode to the front of the free list.
2480  * Release the passed interlock if the vnode will be recycled.
2481  */
2482 int
2483 vrecycle(vp, inter_lkp, td)
2484 	struct vnode *vp;
2485 	struct mtx *inter_lkp;
2486 	struct thread *td;
2487 {
2488 
2489 	VI_LOCK(vp);
2490 	if (vp->v_usecount == 0) {
2491 		if (inter_lkp) {
2492 			mtx_unlock(inter_lkp);
2493 		}
2494 		vgonel(vp, td);
2495 		return (1);
2496 	}
2497 	VI_UNLOCK(vp);
2498 	return (0);
2499 }
2500 
2501 /*
2502  * Eliminate all activity associated with a vnode
2503  * in preparation for reuse.
2504  */
2505 void
2506 vgone(vp)
2507 	register struct vnode *vp;
2508 {
2509 	struct thread *td = curthread;	/* XXX */
2510 
2511 	VI_LOCK(vp);
2512 	vgonel(vp, td);
2513 }
2514 
2515 /*
2516  * Disassociate a character device from the its underlying filesystem and
2517  * attach it to spec.  This is for use when the chr device is still active
2518  * and the filesystem is going away.
2519  */
2520 static void
2521 vgonechrl(struct vnode *vp, struct thread *td)
2522 {
2523 	ASSERT_VI_LOCKED(vp, "vgonechrl");
2524 	vx_lock(vp);
2525 	/*
2526 	 * This is a custom version of vclean() which does not tearm down
2527 	 * the bufs or vm objects held by this vnode.  This allows filesystems
2528 	 * to continue using devices which were discovered via another
2529 	 * filesystem that has been unmounted.
2530 	 */
2531 	if (vp->v_usecount != 0) {
2532 		v_incr_usecount(vp, 1);
2533 		/*
2534 		 * Ensure that no other activity can occur while the
2535 		 * underlying object is being cleaned out.
2536 		 */
2537 		VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td);
2538 		/*
2539 		 * Any other processes trying to obtain this lock must first
2540 		 * wait for VXLOCK to clear, then call the new lock operation.
2541 		 */
2542 		VOP_UNLOCK(vp, 0, td);
2543 		vp->v_vnlock = &vp->v_lock;
2544 		vp->v_tag = "orphanchr";
2545 		vp->v_op = spec_vnodeop_p;
2546 		if (vp->v_mount != NULL)
2547 			insmntque(vp, (struct mount *)0);
2548 		cache_purge(vp);
2549 		vrele(vp);
2550 		VI_LOCK(vp);
2551 	} else
2552 		vclean(vp, 0, td);
2553 	vp->v_op = spec_vnodeop_p;
2554 	vx_unlock(vp);
2555 	VI_UNLOCK(vp);
2556 }
2557 
2558 /*
2559  * vgone, with the vp interlock held.
2560  */
2561 void
2562 vgonel(vp, td)
2563 	struct vnode *vp;
2564 	struct thread *td;
2565 {
2566 	/*
2567 	 * If a vgone (or vclean) is already in progress,
2568 	 * wait until it is done and return.
2569 	 */
2570 	ASSERT_VI_LOCKED(vp, "vgonel");
2571 	if (vp->v_iflag & VI_XLOCK) {
2572 		vp->v_iflag |= VI_XWANT;
2573 		msleep(vp, VI_MTX(vp), PINOD | PDROP, "vgone", 0);
2574 		return;
2575 	}
2576 	vx_lock(vp);
2577 
2578 	/*
2579 	 * Clean out the filesystem specific data.
2580 	 */
2581 	vclean(vp, DOCLOSE, td);
2582 	VI_UNLOCK(vp);
2583 
2584 	/*
2585 	 * If special device, remove it from special device alias list
2586 	 * if it is on one.
2587 	 */
2588 	VI_LOCK(vp);
2589 	if (vp->v_type == VCHR && vp->v_rdev != NODEV) {
2590 		mtx_lock(&spechash_mtx);
2591 		SLIST_REMOVE(&vp->v_rdev->si_hlist, vp, vnode, v_specnext);
2592 		vp->v_rdev->si_usecount -= vp->v_usecount;
2593 		mtx_unlock(&spechash_mtx);
2594 		dev_rel(vp->v_rdev);
2595 		vp->v_rdev = NULL;
2596 	}
2597 
2598 	/*
2599 	 * If it is on the freelist and not already at the head,
2600 	 * move it to the head of the list. The test of the
2601 	 * VDOOMED flag and the reference count of zero is because
2602 	 * it will be removed from the free list by getnewvnode,
2603 	 * but will not have its reference count incremented until
2604 	 * after calling vgone. If the reference count were
2605 	 * incremented first, vgone would (incorrectly) try to
2606 	 * close the previous instance of the underlying object.
2607 	 */
2608 	if (vp->v_usecount == 0 && !(vp->v_iflag & VI_DOOMED)) {
2609 		mtx_lock(&vnode_free_list_mtx);
2610 		if (vp->v_iflag & VI_FREE) {
2611 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2612 		} else {
2613 			vp->v_iflag |= VI_FREE;
2614 			freevnodes++;
2615 		}
2616 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2617 		mtx_unlock(&vnode_free_list_mtx);
2618 	}
2619 
2620 	vp->v_type = VBAD;
2621 	vx_unlock(vp);
2622 	VI_UNLOCK(vp);
2623 }
2624 
2625 /*
2626  * Lookup a vnode by device number.
2627  */
2628 int
2629 vfinddev(dev, vpp)
2630 	dev_t dev;
2631 	struct vnode **vpp;
2632 {
2633 	struct vnode *vp;
2634 
2635 	mtx_lock(&spechash_mtx);
2636 	SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2637 		*vpp = vp;
2638 		mtx_unlock(&spechash_mtx);
2639 		return (1);
2640 	}
2641 	mtx_unlock(&spechash_mtx);
2642 	return (0);
2643 }
2644 
2645 /*
2646  * Calculate the total number of references to a special device.
2647  */
2648 int
2649 vcount(vp)
2650 	struct vnode *vp;
2651 {
2652 	int count;
2653 
2654 	mtx_lock(&spechash_mtx);
2655 	count = vp->v_rdev->si_usecount;
2656 	mtx_unlock(&spechash_mtx);
2657 	return (count);
2658 }
2659 
2660 /*
2661  * Same as above, but using the dev_t as argument
2662  */
2663 int
2664 count_dev(dev)
2665 	dev_t dev;
2666 {
2667 	int count;
2668 
2669 	mtx_lock(&spechash_mtx);
2670 	count = dev->si_usecount;
2671 	mtx_unlock(&spechash_mtx);
2672 	return(count);
2673 }
2674 
2675 /*
2676  * Print out a description of a vnode.
2677  */
2678 static char *typename[] =
2679 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2680 
2681 void
2682 vprint(label, vp)
2683 	char *label;
2684 	struct vnode *vp;
2685 {
2686 	char buf[96];
2687 
2688 	if (label != NULL)
2689 		printf("%s: %p: ", label, (void *)vp);
2690 	else
2691 		printf("%p: ", (void *)vp);
2692 	printf("tag %s, type %s, usecount %d, writecount %d, refcount %d,",
2693 	    vp->v_tag, typename[vp->v_type], vp->v_usecount,
2694 	    vp->v_writecount, vp->v_holdcnt);
2695 	buf[0] = '\0';
2696 	if (vp->v_vflag & VV_ROOT)
2697 		strcat(buf, "|VV_ROOT");
2698 	if (vp->v_vflag & VV_TEXT)
2699 		strcat(buf, "|VV_TEXT");
2700 	if (vp->v_vflag & VV_SYSTEM)
2701 		strcat(buf, "|VV_SYSTEM");
2702 	if (vp->v_iflag & VI_XLOCK)
2703 		strcat(buf, "|VI_XLOCK");
2704 	if (vp->v_iflag & VI_XWANT)
2705 		strcat(buf, "|VI_XWANT");
2706 	if (vp->v_iflag & VI_BWAIT)
2707 		strcat(buf, "|VI_BWAIT");
2708 	if (vp->v_iflag & VI_DOOMED)
2709 		strcat(buf, "|VI_DOOMED");
2710 	if (vp->v_iflag & VI_FREE)
2711 		strcat(buf, "|VI_FREE");
2712 	if (vp->v_vflag & VV_OBJBUF)
2713 		strcat(buf, "|VV_OBJBUF");
2714 	if (buf[0] != '\0')
2715 		printf(" flags (%s),", &buf[1]);
2716 	lockmgr_printinfo(vp->v_vnlock);
2717 	printf("\n");
2718 	if (vp->v_data != NULL)
2719 		VOP_PRINT(vp);
2720 }
2721 
2722 #ifdef DDB
2723 #include <ddb/ddb.h>
2724 /*
2725  * List all of the locked vnodes in the system.
2726  * Called when debugging the kernel.
2727  */
2728 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2729 {
2730 	struct mount *mp, *nmp;
2731 	struct vnode *vp;
2732 
2733 	/*
2734 	 * Note: because this is DDB, we can't obey the locking semantics
2735 	 * for these structures, which means we could catch an inconsistent
2736 	 * state and dereference a nasty pointer.  Not much to be done
2737 	 * about that.
2738 	 */
2739 	printf("Locked vnodes\n");
2740 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2741 		nmp = TAILQ_NEXT(mp, mnt_list);
2742 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2743 			if (VOP_ISLOCKED(vp, NULL))
2744 				vprint(NULL, vp);
2745 		}
2746 		nmp = TAILQ_NEXT(mp, mnt_list);
2747 	}
2748 }
2749 #endif
2750 
2751 /*
2752  * Fill in a struct xvfsconf based on a struct vfsconf.
2753  */
2754 static void
2755 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2756 {
2757 
2758 	strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2759 	xvfsp->vfc_typenum = vfsp->vfc_typenum;
2760 	xvfsp->vfc_refcount = vfsp->vfc_refcount;
2761 	xvfsp->vfc_flags = vfsp->vfc_flags;
2762 	/*
2763 	 * These are unused in userland, we keep them
2764 	 * to not break binary compatibility.
2765 	 */
2766 	xvfsp->vfc_vfsops = NULL;
2767 	xvfsp->vfc_next = NULL;
2768 }
2769 
2770 static int
2771 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2772 {
2773 	struct vfsconf *vfsp;
2774 	struct xvfsconf *xvfsp;
2775 	int cnt, error, i;
2776 
2777 	cnt = 0;
2778 	for (vfsp = vfsconf; vfsp != NULL; vfsp = vfsp->vfc_next)
2779 		cnt++;
2780 	xvfsp = malloc(sizeof(struct xvfsconf) * cnt, M_TEMP, M_WAITOK);
2781 	/*
2782 	 * Handle the race that we will have here when struct vfsconf
2783 	 * will be locked down by using both cnt and checking vfc_next
2784 	 * against NULL to determine the end of the loop.  The race will
2785 	 * happen because we will have to unlock before calling malloc().
2786 	 * We are protected by Giant for now.
2787 	 */
2788 	i = 0;
2789 	for (vfsp = vfsconf; vfsp != NULL && i < cnt; vfsp = vfsp->vfc_next) {
2790 		vfsconf2x(vfsp, xvfsp + i);
2791 		i++;
2792 	}
2793 	error = SYSCTL_OUT(req, xvfsp, sizeof(struct xvfsconf) * i);
2794 	free(xvfsp, M_TEMP);
2795 	return (error);
2796 }
2797 
2798 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
2799     "S,xvfsconf", "List of all configured filesystems");
2800 
2801 /*
2802  * Top level filesystem related information gathering.
2803  */
2804 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2805 
2806 static int
2807 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2808 {
2809 	int *name = (int *)arg1 - 1;	/* XXX */
2810 	u_int namelen = arg2 + 1;	/* XXX */
2811 	struct vfsconf *vfsp;
2812 	struct xvfsconf xvfsp;
2813 
2814 	printf("WARNING: userland calling deprecated sysctl, "
2815 	    "please rebuild world\n");
2816 
2817 #if 1 || defined(COMPAT_PRELITE2)
2818 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2819 	if (namelen == 1)
2820 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2821 #endif
2822 
2823 	switch (name[1]) {
2824 	case VFS_MAXTYPENUM:
2825 		if (namelen != 2)
2826 			return (ENOTDIR);
2827 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2828 	case VFS_CONF:
2829 		if (namelen != 3)
2830 			return (ENOTDIR);	/* overloaded */
2831 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2832 			if (vfsp->vfc_typenum == name[2])
2833 				break;
2834 		if (vfsp == NULL)
2835 			return (EOPNOTSUPP);
2836 		vfsconf2x(vfsp, &xvfsp);
2837 		return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
2838 	}
2839 	return (EOPNOTSUPP);
2840 }
2841 
2842 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP, vfs_sysctl,
2843 	"Generic filesystem");
2844 
2845 #if 1 || defined(COMPAT_PRELITE2)
2846 
2847 static int
2848 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2849 {
2850 	int error;
2851 	struct vfsconf *vfsp;
2852 	struct ovfsconf ovfs;
2853 
2854 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2855 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
2856 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
2857 		ovfs.vfc_index = vfsp->vfc_typenum;
2858 		ovfs.vfc_refcount = vfsp->vfc_refcount;
2859 		ovfs.vfc_flags = vfsp->vfc_flags;
2860 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2861 		if (error)
2862 			return error;
2863 	}
2864 	return 0;
2865 }
2866 
2867 #endif /* 1 || COMPAT_PRELITE2 */
2868 
2869 #define KINFO_VNODESLOP		10
2870 #ifdef notyet
2871 /*
2872  * Dump vnode list (via sysctl).
2873  */
2874 /* ARGSUSED */
2875 static int
2876 sysctl_vnode(SYSCTL_HANDLER_ARGS)
2877 {
2878 	struct xvnode *xvn;
2879 	struct thread *td = req->td;
2880 	struct mount *mp;
2881 	struct vnode *vp;
2882 	int error, len, n;
2883 
2884 	/*
2885 	 * Stale numvnodes access is not fatal here.
2886 	 */
2887 	req->lock = 0;
2888 	len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
2889 	if (!req->oldptr)
2890 		/* Make an estimate */
2891 		return (SYSCTL_OUT(req, 0, len));
2892 
2893 	error = sysctl_wire_old_buffer(req, 0);
2894 	if (error != 0)
2895 		return (error);
2896 	xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
2897 	n = 0;
2898 	mtx_lock(&mountlist_mtx);
2899 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2900 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td))
2901 			continue;
2902 		MNT_ILOCK(mp);
2903 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2904 			if (n == len)
2905 				break;
2906 			vref(vp);
2907 			xvn[n].xv_size = sizeof *xvn;
2908 			xvn[n].xv_vnode = vp;
2909 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
2910 			XV_COPY(usecount);
2911 			XV_COPY(writecount);
2912 			XV_COPY(holdcnt);
2913 			XV_COPY(id);
2914 			XV_COPY(mount);
2915 			XV_COPY(numoutput);
2916 			XV_COPY(type);
2917 #undef XV_COPY
2918 			xvn[n].xv_flag = vp->v_vflag;
2919 
2920 			switch (vp->v_type) {
2921 			case VREG:
2922 			case VDIR:
2923 			case VLNK:
2924 				xvn[n].xv_dev = vp->v_cachedfs;
2925 				xvn[n].xv_ino = vp->v_cachedid;
2926 				break;
2927 			case VBLK:
2928 			case VCHR:
2929 				if (vp->v_rdev == NULL) {
2930 					vrele(vp);
2931 					continue;
2932 				}
2933 				xvn[n].xv_dev = dev2udev(vp->v_rdev);
2934 				break;
2935 			case VSOCK:
2936 				xvn[n].xv_socket = vp->v_socket;
2937 				break;
2938 			case VFIFO:
2939 				xvn[n].xv_fifo = vp->v_fifoinfo;
2940 				break;
2941 			case VNON:
2942 			case VBAD:
2943 			default:
2944 				/* shouldn't happen? */
2945 				vrele(vp);
2946 				continue;
2947 			}
2948 			vrele(vp);
2949 			++n;
2950 		}
2951 		MNT_IUNLOCK(mp);
2952 		mtx_lock(&mountlist_mtx);
2953 		vfs_unbusy(mp, td);
2954 		if (n == len)
2955 			break;
2956 	}
2957 	mtx_unlock(&mountlist_mtx);
2958 
2959 	error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
2960 	free(xvn, M_TEMP);
2961 	return (error);
2962 }
2963 
2964 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2965 	0, 0, sysctl_vnode, "S,xvnode", "");
2966 #endif
2967 
2968 /*
2969  * Check to see if a filesystem is mounted on a block device.
2970  */
2971 int
2972 vfs_mountedon(vp)
2973 	struct vnode *vp;
2974 {
2975 
2976 	if (vp->v_rdev->si_mountpoint != NULL)
2977 		return (EBUSY);
2978 	return (0);
2979 }
2980 
2981 /*
2982  * Unmount all filesystems. The list is traversed in reverse order
2983  * of mounting to avoid dependencies.
2984  */
2985 void
2986 vfs_unmountall()
2987 {
2988 	struct mount *mp;
2989 	struct thread *td;
2990 	int error;
2991 
2992 	if (curthread != NULL)
2993 		td = curthread;
2994 	else
2995 		td = FIRST_THREAD_IN_PROC(initproc); /* XXX XXX proc0? */
2996 	/*
2997 	 * Since this only runs when rebooting, it is not interlocked.
2998 	 */
2999 	while(!TAILQ_EMPTY(&mountlist)) {
3000 		mp = TAILQ_LAST(&mountlist, mntlist);
3001 		error = dounmount(mp, MNT_FORCE, td);
3002 		if (error) {
3003 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
3004 			printf("unmount of %s failed (",
3005 			    mp->mnt_stat.f_mntonname);
3006 			if (error == EBUSY)
3007 				printf("BUSY)\n");
3008 			else
3009 				printf("%d)\n", error);
3010 		} else {
3011 			/* The unmount has removed mp from the mountlist */
3012 		}
3013 	}
3014 }
3015 
3016 /*
3017  * perform msync on all vnodes under a mount point
3018  * the mount point must be locked.
3019  */
3020 void
3021 vfs_msync(struct mount *mp, int flags)
3022 {
3023 	struct vnode *vp, *nvp;
3024 	struct vm_object *obj;
3025 	int tries;
3026 
3027 	GIANT_REQUIRED;
3028 
3029 	tries = 5;
3030 	MNT_ILOCK(mp);
3031 loop:
3032 	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) {
3033 		if (vp->v_mount != mp) {
3034 			if (--tries > 0)
3035 				goto loop;
3036 			break;
3037 		}
3038 		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
3039 
3040 		VI_LOCK(vp);
3041 		if (vp->v_iflag & VI_XLOCK) {
3042 			VI_UNLOCK(vp);
3043 			continue;
3044 		}
3045 
3046 		if ((vp->v_iflag & VI_OBJDIRTY) &&
3047 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
3048 			MNT_IUNLOCK(mp);
3049 			if (!vget(vp,
3050 			    LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3051 			    curthread)) {
3052 				if (vp->v_vflag & VV_NOSYNC) {	/* unlinked */
3053 					vput(vp);
3054 					MNT_ILOCK(mp);
3055 					continue;
3056 				}
3057 
3058 				if (VOP_GETVOBJECT(vp, &obj) == 0) {
3059 					VM_OBJECT_LOCK(obj);
3060 					vm_object_page_clean(obj, 0, 0,
3061 					    flags == MNT_WAIT ?
3062 					    OBJPC_SYNC : OBJPC_NOSYNC);
3063 					VM_OBJECT_UNLOCK(obj);
3064 				}
3065 				vput(vp);
3066 			}
3067 			MNT_ILOCK(mp);
3068 			if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) {
3069 				if (--tries > 0)
3070 					goto loop;
3071 				break;
3072 			}
3073 		} else
3074 			VI_UNLOCK(vp);
3075 	}
3076 	MNT_IUNLOCK(mp);
3077 }
3078 
3079 /*
3080  * Create the VM object needed for VMIO and mmap support.  This
3081  * is done for all VREG files in the system.  Some filesystems might
3082  * afford the additional metadata buffering capability of the
3083  * VMIO code by making the device node be VMIO mode also.
3084  *
3085  * vp must be locked when vfs_object_create is called.
3086  */
3087 int
3088 vfs_object_create(vp, td, cred)
3089 	struct vnode *vp;
3090 	struct thread *td;
3091 	struct ucred *cred;
3092 {
3093 
3094 	GIANT_REQUIRED;
3095 	return (VOP_CREATEVOBJECT(vp, cred, td));
3096 }
3097 
3098 /*
3099  * Mark a vnode as free, putting it up for recycling.
3100  */
3101 void
3102 vfree(vp)
3103 	struct vnode *vp;
3104 {
3105 
3106 	ASSERT_VI_LOCKED(vp, "vfree");
3107 	mtx_lock(&vnode_free_list_mtx);
3108 	KASSERT((vp->v_iflag & VI_FREE) == 0, ("vnode already free"));
3109 	if (vp->v_iflag & VI_AGE) {
3110 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3111 	} else {
3112 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3113 	}
3114 	freevnodes++;
3115 	mtx_unlock(&vnode_free_list_mtx);
3116 	vp->v_iflag &= ~VI_AGE;
3117 	vp->v_iflag |= VI_FREE;
3118 }
3119 
3120 /*
3121  * Opposite of vfree() - mark a vnode as in use.
3122  */
3123 void
3124 vbusy(vp)
3125 	struct vnode *vp;
3126 {
3127 
3128 	ASSERT_VI_LOCKED(vp, "vbusy");
3129 	KASSERT((vp->v_iflag & VI_FREE) != 0, ("vnode not free"));
3130 
3131 	mtx_lock(&vnode_free_list_mtx);
3132 	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3133 	freevnodes--;
3134 	mtx_unlock(&vnode_free_list_mtx);
3135 
3136 	vp->v_iflag &= ~(VI_FREE|VI_AGE);
3137 }
3138 
3139 /*
3140  * Initalize per-vnode helper structure to hold poll-related state.
3141  */
3142 void
3143 v_addpollinfo(struct vnode *vp)
3144 {
3145 
3146 	vp->v_pollinfo = uma_zalloc(vnodepoll_zone, M_WAITOK);
3147 	mtx_init(&vp->v_pollinfo->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3148 }
3149 
3150 /*
3151  * Record a process's interest in events which might happen to
3152  * a vnode.  Because poll uses the historic select-style interface
3153  * internally, this routine serves as both the ``check for any
3154  * pending events'' and the ``record my interest in future events''
3155  * functions.  (These are done together, while the lock is held,
3156  * to avoid race conditions.)
3157  */
3158 int
3159 vn_pollrecord(vp, td, events)
3160 	struct vnode *vp;
3161 	struct thread *td;
3162 	short events;
3163 {
3164 
3165 	if (vp->v_pollinfo == NULL)
3166 		v_addpollinfo(vp);
3167 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3168 	if (vp->v_pollinfo->vpi_revents & events) {
3169 		/*
3170 		 * This leaves events we are not interested
3171 		 * in available for the other process which
3172 		 * which presumably had requested them
3173 		 * (otherwise they would never have been
3174 		 * recorded).
3175 		 */
3176 		events &= vp->v_pollinfo->vpi_revents;
3177 		vp->v_pollinfo->vpi_revents &= ~events;
3178 
3179 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
3180 		return events;
3181 	}
3182 	vp->v_pollinfo->vpi_events |= events;
3183 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3184 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3185 	return 0;
3186 }
3187 
3188 /*
3189  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
3190  * it is possible for us to miss an event due to race conditions, but
3191  * that condition is expected to be rare, so for the moment it is the
3192  * preferred interface.
3193  */
3194 void
3195 vn_pollevent(vp, events)
3196 	struct vnode *vp;
3197 	short events;
3198 {
3199 
3200 	if (vp->v_pollinfo == NULL)
3201 		v_addpollinfo(vp);
3202 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3203 	if (vp->v_pollinfo->vpi_events & events) {
3204 		/*
3205 		 * We clear vpi_events so that we don't
3206 		 * call selwakeup() twice if two events are
3207 		 * posted before the polling process(es) is
3208 		 * awakened.  This also ensures that we take at
3209 		 * most one selwakeup() if the polling process
3210 		 * is no longer interested.  However, it does
3211 		 * mean that only one event can be noticed at
3212 		 * a time.  (Perhaps we should only clear those
3213 		 * event bits which we note?) XXX
3214 		 */
3215 		vp->v_pollinfo->vpi_events = 0;	/* &= ~events ??? */
3216 		vp->v_pollinfo->vpi_revents |= events;
3217 		selwakeuppri(&vp->v_pollinfo->vpi_selinfo, PRIBIO);
3218 	}
3219 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3220 }
3221 
3222 /*
3223  * Wake up anyone polling on vp because it is being revoked.
3224  * This depends on dead_poll() returning POLLHUP for correct
3225  * behavior.
3226  */
3227 void
3228 vn_pollgone(vp)
3229 	struct vnode *vp;
3230 {
3231 
3232 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3233 	VN_KNOTE(vp, NOTE_REVOKE);
3234 	if (vp->v_pollinfo->vpi_events) {
3235 		vp->v_pollinfo->vpi_events = 0;
3236 		selwakeuppri(&vp->v_pollinfo->vpi_selinfo, PRIBIO);
3237 	}
3238 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3239 }
3240 
3241 
3242 
3243 /*
3244  * Routine to create and manage a filesystem syncer vnode.
3245  */
3246 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
3247 static int	sync_fsync(struct  vop_fsync_args *);
3248 static int	sync_inactive(struct  vop_inactive_args *);
3249 static int	sync_reclaim(struct  vop_reclaim_args *);
3250 
3251 static vop_t **sync_vnodeop_p;
3252 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
3253 	{ &vop_default_desc,	(vop_t *) vop_eopnotsupp },
3254 	{ &vop_close_desc,	(vop_t *) sync_close },		/* close */
3255 	{ &vop_fsync_desc,	(vop_t *) sync_fsync },		/* fsync */
3256 	{ &vop_inactive_desc,	(vop_t *) sync_inactive },	/* inactive */
3257 	{ &vop_reclaim_desc,	(vop_t *) sync_reclaim },	/* reclaim */
3258 	{ &vop_lock_desc,	(vop_t *) vop_stdlock },	/* lock */
3259 	{ &vop_unlock_desc,	(vop_t *) vop_stdunlock },	/* unlock */
3260 	{ &vop_islocked_desc,	(vop_t *) vop_stdislocked },	/* islocked */
3261 	{ NULL, NULL }
3262 };
3263 static struct vnodeopv_desc sync_vnodeop_opv_desc =
3264 	{ &sync_vnodeop_p, sync_vnodeop_entries };
3265 
3266 VNODEOP_SET(sync_vnodeop_opv_desc);
3267 
3268 /*
3269  * Create a new filesystem syncer vnode for the specified mount point.
3270  */
3271 int
3272 vfs_allocate_syncvnode(mp)
3273 	struct mount *mp;
3274 {
3275 	struct vnode *vp;
3276 	static long start, incr, next;
3277 	int error;
3278 
3279 	/* Allocate a new vnode */
3280 	if ((error = getnewvnode("syncer", mp, sync_vnodeop_p, &vp)) != 0) {
3281 		mp->mnt_syncer = NULL;
3282 		return (error);
3283 	}
3284 	vp->v_type = VNON;
3285 	/*
3286 	 * Place the vnode onto the syncer worklist. We attempt to
3287 	 * scatter them about on the list so that they will go off
3288 	 * at evenly distributed times even if all the filesystems
3289 	 * are mounted at once.
3290 	 */
3291 	next += incr;
3292 	if (next == 0 || next > syncer_maxdelay) {
3293 		start /= 2;
3294 		incr /= 2;
3295 		if (start == 0) {
3296 			start = syncer_maxdelay / 2;
3297 			incr = syncer_maxdelay;
3298 		}
3299 		next = start;
3300 	}
3301 	VI_LOCK(vp);
3302 	vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
3303 	VI_UNLOCK(vp);
3304 	mp->mnt_syncer = vp;
3305 	return (0);
3306 }
3307 
3308 /*
3309  * Do a lazy sync of the filesystem.
3310  */
3311 static int
3312 sync_fsync(ap)
3313 	struct vop_fsync_args /* {
3314 		struct vnode *a_vp;
3315 		struct ucred *a_cred;
3316 		int a_waitfor;
3317 		struct thread *a_td;
3318 	} */ *ap;
3319 {
3320 	struct vnode *syncvp = ap->a_vp;
3321 	struct mount *mp = syncvp->v_mount;
3322 	struct thread *td = ap->a_td;
3323 	int error, asyncflag;
3324 
3325 	/*
3326 	 * We only need to do something if this is a lazy evaluation.
3327 	 */
3328 	if (ap->a_waitfor != MNT_LAZY)
3329 		return (0);
3330 
3331 	/*
3332 	 * Move ourselves to the back of the sync list.
3333 	 */
3334 	VI_LOCK(syncvp);
3335 	vn_syncer_add_to_worklist(syncvp, syncdelay);
3336 	VI_UNLOCK(syncvp);
3337 
3338 	/*
3339 	 * Walk the list of vnodes pushing all that are dirty and
3340 	 * not already on the sync list.
3341 	 */
3342 	mtx_lock(&mountlist_mtx);
3343 	if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) {
3344 		mtx_unlock(&mountlist_mtx);
3345 		return (0);
3346 	}
3347 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3348 		vfs_unbusy(mp, td);
3349 		return (0);
3350 	}
3351 	asyncflag = mp->mnt_flag & MNT_ASYNC;
3352 	mp->mnt_flag &= ~MNT_ASYNC;
3353 	vfs_msync(mp, MNT_NOWAIT);
3354 	error = VFS_SYNC(mp, MNT_LAZY, ap->a_cred, td);
3355 	if (asyncflag)
3356 		mp->mnt_flag |= MNT_ASYNC;
3357 	vn_finished_write(mp);
3358 	vfs_unbusy(mp, td);
3359 	return (error);
3360 }
3361 
3362 /*
3363  * The syncer vnode is no referenced.
3364  */
3365 static int
3366 sync_inactive(ap)
3367 	struct vop_inactive_args /* {
3368 		struct vnode *a_vp;
3369 		struct thread *a_td;
3370 	} */ *ap;
3371 {
3372 
3373 	VOP_UNLOCK(ap->a_vp, 0, ap->a_td);
3374 	vgone(ap->a_vp);
3375 	return (0);
3376 }
3377 
3378 /*
3379  * The syncer vnode is no longer needed and is being decommissioned.
3380  *
3381  * Modifications to the worklist must be protected by sync_mtx.
3382  */
3383 static int
3384 sync_reclaim(ap)
3385 	struct vop_reclaim_args /* {
3386 		struct vnode *a_vp;
3387 	} */ *ap;
3388 {
3389 	struct vnode *vp = ap->a_vp;
3390 
3391 	VI_LOCK(vp);
3392 	vp->v_mount->mnt_syncer = NULL;
3393 	if (vp->v_iflag & VI_ONWORKLST) {
3394 		mtx_lock(&sync_mtx);
3395 		LIST_REMOVE(vp, v_synclist);
3396 		mtx_unlock(&sync_mtx);
3397 		vp->v_iflag &= ~VI_ONWORKLST;
3398 	}
3399 	VI_UNLOCK(vp);
3400 
3401 	return (0);
3402 }
3403 
3404 /*
3405  * extract the dev_t from a VCHR
3406  */
3407 dev_t
3408 vn_todev(vp)
3409 	struct vnode *vp;
3410 {
3411 
3412 	if (vp->v_type != VCHR)
3413 		return (NODEV);
3414 	return (vp->v_rdev);
3415 }
3416 
3417 /*
3418  * Check if vnode represents a disk device
3419  */
3420 int
3421 vn_isdisk(vp, errp)
3422 	struct vnode *vp;
3423 	int *errp;
3424 {
3425 	int error;
3426 
3427 	error = 0;
3428 	if (vp->v_type != VCHR)
3429 		error = ENOTBLK;
3430 	else if (vp->v_rdev == NULL)
3431 		error = ENXIO;
3432 	else if (!(devsw(vp->v_rdev)->d_flags & D_DISK))
3433 		error = ENOTBLK;
3434 	if (errp != NULL)
3435 		*errp = error;
3436 	return (error == 0);
3437 }
3438 
3439 /*
3440  * Free data allocated by namei(); see namei(9) for details.
3441  */
3442 void
3443 NDFREE(ndp, flags)
3444      struct nameidata *ndp;
3445      const u_int flags;
3446 {
3447 
3448 	if (!(flags & NDF_NO_FREE_PNBUF) &&
3449 	    (ndp->ni_cnd.cn_flags & HASBUF)) {
3450 		uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
3451 		ndp->ni_cnd.cn_flags &= ~HASBUF;
3452 	}
3453 	if (!(flags & NDF_NO_DVP_UNLOCK) &&
3454 	    (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
3455 	    ndp->ni_dvp != ndp->ni_vp)
3456 		VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_thread);
3457 	if (!(flags & NDF_NO_DVP_RELE) &&
3458 	    (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
3459 		vrele(ndp->ni_dvp);
3460 		ndp->ni_dvp = NULL;
3461 	}
3462 	if (!(flags & NDF_NO_VP_UNLOCK) &&
3463 	    (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
3464 		VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_thread);
3465 	if (!(flags & NDF_NO_VP_RELE) &&
3466 	    ndp->ni_vp) {
3467 		vrele(ndp->ni_vp);
3468 		ndp->ni_vp = NULL;
3469 	}
3470 	if (!(flags & NDF_NO_STARTDIR_RELE) &&
3471 	    (ndp->ni_cnd.cn_flags & SAVESTART)) {
3472 		vrele(ndp->ni_startdir);
3473 		ndp->ni_startdir = NULL;
3474 	}
3475 }
3476 
3477 /*
3478  * Common filesystem object access control check routine.  Accepts a
3479  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3480  * and optional call-by-reference privused argument allowing vaccess()
3481  * to indicate to the caller whether privilege was used to satisfy the
3482  * request (obsoleted).  Returns 0 on success, or an errno on failure.
3483  */
3484 int
3485 vaccess(type, file_mode, file_uid, file_gid, acc_mode, cred, privused)
3486 	enum vtype type;
3487 	mode_t file_mode;
3488 	uid_t file_uid;
3489 	gid_t file_gid;
3490 	mode_t acc_mode;
3491 	struct ucred *cred;
3492 	int *privused;
3493 {
3494 	mode_t dac_granted;
3495 #ifdef CAPABILITIES
3496 	mode_t cap_granted;
3497 #endif
3498 
3499 	/*
3500 	 * Look for a normal, non-privileged way to access the file/directory
3501 	 * as requested.  If it exists, go with that.
3502 	 */
3503 
3504 	if (privused != NULL)
3505 		*privused = 0;
3506 
3507 	dac_granted = 0;
3508 
3509 	/* Check the owner. */
3510 	if (cred->cr_uid == file_uid) {
3511 		dac_granted |= VADMIN;
3512 		if (file_mode & S_IXUSR)
3513 			dac_granted |= VEXEC;
3514 		if (file_mode & S_IRUSR)
3515 			dac_granted |= VREAD;
3516 		if (file_mode & S_IWUSR)
3517 			dac_granted |= (VWRITE | VAPPEND);
3518 
3519 		if ((acc_mode & dac_granted) == acc_mode)
3520 			return (0);
3521 
3522 		goto privcheck;
3523 	}
3524 
3525 	/* Otherwise, check the groups (first match) */
3526 	if (groupmember(file_gid, cred)) {
3527 		if (file_mode & S_IXGRP)
3528 			dac_granted |= VEXEC;
3529 		if (file_mode & S_IRGRP)
3530 			dac_granted |= VREAD;
3531 		if (file_mode & S_IWGRP)
3532 			dac_granted |= (VWRITE | VAPPEND);
3533 
3534 		if ((acc_mode & dac_granted) == acc_mode)
3535 			return (0);
3536 
3537 		goto privcheck;
3538 	}
3539 
3540 	/* Otherwise, check everyone else. */
3541 	if (file_mode & S_IXOTH)
3542 		dac_granted |= VEXEC;
3543 	if (file_mode & S_IROTH)
3544 		dac_granted |= VREAD;
3545 	if (file_mode & S_IWOTH)
3546 		dac_granted |= (VWRITE | VAPPEND);
3547 	if ((acc_mode & dac_granted) == acc_mode)
3548 		return (0);
3549 
3550 privcheck:
3551 	if (!suser_cred(cred, PRISON_ROOT)) {
3552 		/* XXX audit: privilege used */
3553 		if (privused != NULL)
3554 			*privused = 1;
3555 		return (0);
3556 	}
3557 
3558 #ifdef CAPABILITIES
3559 	/*
3560 	 * Build a capability mask to determine if the set of capabilities
3561 	 * satisfies the requirements when combined with the granted mask
3562 	 * from above.
3563 	 * For each capability, if the capability is required, bitwise
3564 	 * or the request type onto the cap_granted mask.
3565 	 */
3566 	cap_granted = 0;
3567 
3568 	if (type == VDIR) {
3569 		/*
3570 		 * For directories, use CAP_DAC_READ_SEARCH to satisfy
3571 		 * VEXEC requests, instead of CAP_DAC_EXECUTE.
3572 		 */
3573 		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3574 		    !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3575 			cap_granted |= VEXEC;
3576 	} else {
3577 		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3578 		    !cap_check(cred, NULL, CAP_DAC_EXECUTE, PRISON_ROOT))
3579 			cap_granted |= VEXEC;
3580 	}
3581 
3582 	if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
3583 	    !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3584 		cap_granted |= VREAD;
3585 
3586 	if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3587 	    !cap_check(cred, NULL, CAP_DAC_WRITE, PRISON_ROOT))
3588 		cap_granted |= (VWRITE | VAPPEND);
3589 
3590 	if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3591 	    !cap_check(cred, NULL, CAP_FOWNER, PRISON_ROOT))
3592 		cap_granted |= VADMIN;
3593 
3594 	if ((acc_mode & (cap_granted | dac_granted)) == acc_mode) {
3595 		/* XXX audit: privilege used */
3596 		if (privused != NULL)
3597 			*privused = 1;
3598 		return (0);
3599 	}
3600 #endif
3601 
3602 	return ((acc_mode & VADMIN) ? EPERM : EACCES);
3603 }
3604 
3605 /*
3606  * Credential check based on process requesting service, and per-attribute
3607  * permissions.
3608  */
3609 int
3610 extattr_check_cred(struct vnode *vp, int attrnamespace,
3611     struct ucred *cred, struct thread *td, int access)
3612 {
3613 
3614 	/*
3615 	 * Kernel-invoked always succeeds.
3616 	 */
3617 	if (cred == NOCRED)
3618 		return (0);
3619 
3620 	/*
3621 	 * Do not allow privileged processes in jail to directly
3622 	 * manipulate system attributes.
3623 	 *
3624 	 * XXX What capability should apply here?
3625 	 * Probably CAP_SYS_SETFFLAG.
3626 	 */
3627 	switch (attrnamespace) {
3628 	case EXTATTR_NAMESPACE_SYSTEM:
3629 		/* Potentially should be: return (EPERM); */
3630 		return (suser_cred(cred, 0));
3631 	case EXTATTR_NAMESPACE_USER:
3632 		return (VOP_ACCESS(vp, access, cred, td));
3633 	default:
3634 		return (EPERM);
3635 	}
3636 }
3637 
3638 #ifdef DEBUG_VFS_LOCKS
3639 /*
3640  * This only exists to supress warnings from unlocked specfs accesses.  It is
3641  * no longer ok to have an unlocked VFS.
3642  */
3643 #define	IGNORE_LOCK(vp) ((vp)->v_type == VCHR || (vp)->v_type == VBAD)
3644 
3645 int vfs_badlock_ddb = 1;	/* Drop into debugger on violation. */
3646 int vfs_badlock_mutex = 1;	/* Check for interlock across VOPs. */
3647 int vfs_badlock_print = 1;	/* Print lock violations. */
3648 
3649 static void
3650 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3651 {
3652 
3653 	if (vfs_badlock_print)
3654 		printf("%s: %p %s\n", str, (void *)vp, msg);
3655 	if (vfs_badlock_ddb)
3656 		Debugger("lock violation");
3657 }
3658 
3659 void
3660 assert_vi_locked(struct vnode *vp, const char *str)
3661 {
3662 
3663 	if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3664 		vfs_badlock("interlock is not locked but should be", str, vp);
3665 }
3666 
3667 void
3668 assert_vi_unlocked(struct vnode *vp, const char *str)
3669 {
3670 
3671 	if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3672 		vfs_badlock("interlock is locked but should not be", str, vp);
3673 }
3674 
3675 void
3676 assert_vop_locked(struct vnode *vp, const char *str)
3677 {
3678 
3679 	if (vp && !IGNORE_LOCK(vp) && VOP_ISLOCKED(vp, NULL) == 0)
3680 		vfs_badlock("is not locked but should be", str, vp);
3681 }
3682 
3683 void
3684 assert_vop_unlocked(struct vnode *vp, const char *str)
3685 {
3686 
3687 	if (vp && !IGNORE_LOCK(vp) &&
3688 	    VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE)
3689 		vfs_badlock("is locked but should not be", str, vp);
3690 }
3691 
3692 #if 0
3693 void
3694 assert_vop_elocked(struct vnode *vp, const char *str)
3695 {
3696 
3697 	if (vp && !IGNORE_LOCK(vp) &&
3698 	    VOP_ISLOCKED(vp, curthread) != LK_EXCLUSIVE)
3699 		vfs_badlock("is not exclusive locked but should be", str, vp);
3700 }
3701 
3702 void
3703 assert_vop_elocked_other(struct vnode *vp, const char *str)
3704 {
3705 
3706 	if (vp && !IGNORE_LOCK(vp) &&
3707 	    VOP_ISLOCKED(vp, curthread) != LK_EXCLOTHER)
3708 		vfs_badlock("is not exclusive locked by another thread",
3709 		    str, vp);
3710 }
3711 
3712 void
3713 assert_vop_slocked(struct vnode *vp, const char *str)
3714 {
3715 
3716 	if (vp && !IGNORE_LOCK(vp) &&
3717 	    VOP_ISLOCKED(vp, curthread) != LK_SHARED)
3718 		vfs_badlock("is not locked shared but should be", str, vp);
3719 }
3720 #endif /* 0 */
3721 
3722 void
3723 vop_rename_pre(void *ap)
3724 {
3725 	struct vop_rename_args *a = ap;
3726 
3727 	if (a->a_tvp)
3728 		ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
3729 	ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
3730 	ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
3731 	ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
3732 
3733 	/* Check the source (from). */
3734 	if (a->a_tdvp != a->a_fdvp)
3735 		ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3736 	if (a->a_tvp != a->a_fvp)
3737 		ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: tvp locked");
3738 
3739 	/* Check the target. */
3740 	if (a->a_tvp)
3741 		ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3742 	ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3743 }
3744 
3745 void
3746 vop_strategy_pre(void *ap)
3747 {
3748 	struct vop_strategy_args *a;
3749 	struct buf *bp;
3750 
3751 	a = ap;
3752 	bp = a->a_bp;
3753 
3754 	/*
3755 	 * Cluster ops lock their component buffers but not the IO container.
3756 	 */
3757 	if ((bp->b_flags & B_CLUSTER) != 0)
3758 		return;
3759 
3760 	if (BUF_REFCNT(bp) < 1) {
3761 		if (vfs_badlock_print)
3762 			printf(
3763 			    "VOP_STRATEGY: bp is not locked but should be\n");
3764 		if (vfs_badlock_ddb)
3765 			Debugger("lock violation");
3766 	}
3767 }
3768 
3769 void
3770 vop_lookup_pre(void *ap)
3771 {
3772 	struct vop_lookup_args *a;
3773 	struct vnode *dvp;
3774 
3775 	a = ap;
3776 	dvp = a->a_dvp;
3777 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3778 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3779 }
3780 
3781 void
3782 vop_lookup_post(void *ap, int rc)
3783 {
3784 	struct vop_lookup_args *a;
3785 	struct componentname *cnp;
3786 	struct vnode *dvp;
3787 	struct vnode *vp;
3788 	int flags;
3789 
3790 	a = ap;
3791 	dvp = a->a_dvp;
3792 	cnp = a->a_cnp;
3793 	vp = *(a->a_vpp);
3794 	flags = cnp->cn_flags;
3795 
3796 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3797 
3798 	/*
3799 	 * If this is the last path component for this lookup and LOCKPARENT
3800 	 * is set, OR if there is an error the directory has to be locked.
3801 	 */
3802 	if ((flags & LOCKPARENT) && (flags & ISLASTCN))
3803 		ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP (LOCKPARENT)");
3804 	else if (rc != 0)
3805 		ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP (error)");
3806 	else if (dvp != vp)
3807 		ASSERT_VOP_UNLOCKED(dvp, "VOP_LOOKUP (dvp)");
3808 	if (flags & PDIRUNLOCK)
3809 		ASSERT_VOP_UNLOCKED(dvp, "VOP_LOOKUP (PDIRUNLOCK)");
3810 }
3811 
3812 void
3813 vop_lock_pre(void *ap)
3814 {
3815 	struct vop_lock_args *a = ap;
3816 
3817 	if ((a->a_flags & LK_INTERLOCK) == 0)
3818 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3819 	else
3820 		ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
3821 }
3822 
3823 void
3824 vop_lock_post(void *ap, int rc)
3825 {
3826 	struct vop_lock_args *a = ap;
3827 
3828 	ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3829 	if (rc == 0)
3830 		ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
3831 }
3832 
3833 void
3834 vop_unlock_pre(void *ap)
3835 {
3836 	struct vop_unlock_args *a = ap;
3837 
3838 	if (a->a_flags & LK_INTERLOCK)
3839 		ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
3840 	ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
3841 }
3842 
3843 void
3844 vop_unlock_post(void *ap, int rc)
3845 {
3846 	struct vop_unlock_args *a = ap;
3847 
3848 	if (a->a_flags & LK_INTERLOCK)
3849 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
3850 }
3851 #endif /* DEBUG_VFS_LOCKS */
3852