xref: /freebsd/sys/kern/vfs_subr.c (revision ec3a1a66aad80b656971dcff4277d8ea4967fbb5)
1 /*
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)vfs_subr.c	8.31 (Berkeley) 5/26/95
39  * $FreeBSD$
40  */
41 
42 /*
43  * External virtual filesystem routines
44  */
45 #include "opt_ddb.h"
46 #include "opt_mac.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/bio.h>
51 #include <sys/buf.h>
52 #include <sys/conf.h>
53 #include <sys/eventhandler.h>
54 #include <sys/extattr.h>
55 #include <sys/fcntl.h>
56 #include <sys/kernel.h>
57 #include <sys/kthread.h>
58 #include <sys/mac.h>
59 #include <sys/malloc.h>
60 #include <sys/mount.h>
61 #include <sys/namei.h>
62 #include <sys/stat.h>
63 #include <sys/sysctl.h>
64 #include <sys/syslog.h>
65 #include <sys/vmmeter.h>
66 #include <sys/vnode.h>
67 
68 #include <vm/vm.h>
69 #include <vm/vm_object.h>
70 #include <vm/vm_extern.h>
71 #include <vm/pmap.h>
72 #include <vm/vm_map.h>
73 #include <vm/vm_page.h>
74 #include <vm/uma.h>
75 
76 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
77 
78 static void	addalias(struct vnode *vp, dev_t nvp_rdev);
79 static void	insmntque(struct vnode *vp, struct mount *mp);
80 static void	vclean(struct vnode *vp, int flags, struct thread *td);
81 static void	vlruvp(struct vnode *vp);
82 static int	flushbuflist(struct buf *blist, int flags, struct vnode *vp,
83 		    int slpflag, int slptimeo, int *errorp);
84 static int	vcanrecycle(struct vnode *vp, struct mount **vnmpp);
85 
86 
87 /*
88  * Number of vnodes in existence.  Increased whenever getnewvnode()
89  * allocates a new vnode, never decreased.
90  */
91 static unsigned long	numvnodes;
92 
93 SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
94 
95 /*
96  * Conversion tables for conversion from vnode types to inode formats
97  * and back.
98  */
99 enum vtype iftovt_tab[16] = {
100 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
101 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
102 };
103 int vttoif_tab[9] = {
104 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
105 	S_IFSOCK, S_IFIFO, S_IFMT,
106 };
107 
108 /*
109  * List of vnodes that are ready for recycling.
110  */
111 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
112 
113 /*
114  * Minimum number of free vnodes.  If there are fewer than this free vnodes,
115  * getnewvnode() will return a newly allocated vnode.
116  */
117 static u_long wantfreevnodes = 25;
118 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
119 /* Number of vnodes in the free list. */
120 static u_long freevnodes;
121 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
122 
123 /*
124  * Various variables used for debugging the new implementation of
125  * reassignbuf().
126  * XXX these are probably of (very) limited utility now.
127  */
128 static int reassignbufcalls;
129 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
130 static int nameileafonly;
131 SYSCTL_INT(_vfs, OID_AUTO, nameileafonly, CTLFLAG_RW, &nameileafonly, 0, "");
132 
133 /*
134  * Cache for the mount type id assigned to NFS.  This is used for
135  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
136  */
137 int	nfs_mount_type = -1;
138 
139 /* To keep more than one thread at a time from running vfs_getnewfsid */
140 static struct mtx mntid_mtx;
141 
142 /*
143  * Lock for any access to the following:
144  *	vnode_free_list
145  *	numvnodes
146  *	freevnodes
147  */
148 static struct mtx vnode_free_list_mtx;
149 
150 /*
151  * For any iteration/modification of dev->si_hlist (linked through
152  * v_specnext)
153  */
154 static struct mtx spechash_mtx;
155 
156 /* Publicly exported FS */
157 struct nfs_public nfs_pub;
158 
159 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
160 static uma_zone_t vnode_zone;
161 static uma_zone_t vnodepoll_zone;
162 
163 /* Set to 1 to print out reclaim of active vnodes */
164 int	prtactive;
165 
166 /*
167  * The workitem queue.
168  *
169  * It is useful to delay writes of file data and filesystem metadata
170  * for tens of seconds so that quickly created and deleted files need
171  * not waste disk bandwidth being created and removed. To realize this,
172  * we append vnodes to a "workitem" queue. When running with a soft
173  * updates implementation, most pending metadata dependencies should
174  * not wait for more than a few seconds. Thus, mounted on block devices
175  * are delayed only about a half the time that file data is delayed.
176  * Similarly, directory updates are more critical, so are only delayed
177  * about a third the time that file data is delayed. Thus, there are
178  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
179  * one each second (driven off the filesystem syncer process). The
180  * syncer_delayno variable indicates the next queue that is to be processed.
181  * Items that need to be processed soon are placed in this queue:
182  *
183  *	syncer_workitem_pending[syncer_delayno]
184  *
185  * A delay of fifteen seconds is done by placing the request fifteen
186  * entries later in the queue:
187  *
188  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
189  *
190  */
191 static int syncer_delayno;
192 static long syncer_mask;
193 LIST_HEAD(synclist, vnode);
194 static struct synclist *syncer_workitem_pending;
195 /*
196  * The sync_mtx protects:
197  *	vp->v_synclist
198  *	syncer_delayno
199  *	syncer_workitem_pending
200  *	rushjob
201  */
202 static struct mtx sync_mtx;
203 
204 #define SYNCER_MAXDELAY		32
205 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
206 static int syncdelay = 30;		/* max time to delay syncing data */
207 static int filedelay = 30;		/* time to delay syncing files */
208 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
209 static int dirdelay = 29;		/* time to delay syncing directories */
210 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
211 static int metadelay = 28;		/* time to delay syncing metadata */
212 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
213 static int rushjob;		/* number of slots to run ASAP */
214 static int stat_rush_requests;	/* number of times I/O speeded up */
215 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
216 
217 /*
218  * Number of vnodes we want to exist at any one time.  This is mostly used
219  * to size hash tables in vnode-related code.  It is normally not used in
220  * getnewvnode(), as wantfreevnodes is normally nonzero.)
221  *
222  * XXX desiredvnodes is historical cruft and should not exist.
223  */
224 int desiredvnodes;
225 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
226     &desiredvnodes, 0, "Maximum number of vnodes");
227 static int minvnodes;
228 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
229     &minvnodes, 0, "Minimum number of vnodes");
230 static int vnlru_nowhere;
231 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0,
232     "Number of times the vnlru process ran without success");
233 
234 /* Hook for calling soft updates */
235 int (*softdep_process_worklist_hook)(struct mount *);
236 
237 /*
238  * This only exists to supress warnings from unlocked specfs accesses.  It is
239  * no longer ok to have an unlocked VFS.
240  */
241 #define IGNORE_LOCK(vp) ((vp)->v_type == VCHR || (vp)->v_type == VBAD)
242 
243 /* Print lock violations */
244 int vfs_badlock_print = 1;
245 
246 /* Panic on violation */
247 int vfs_badlock_panic = 1;
248 
249 /* Check for interlock across VOPs */
250 int vfs_badlock_mutex = 1;
251 
252 static void
253 vfs_badlock(char *msg, char *str, struct vnode *vp)
254 {
255 	if (vfs_badlock_print)
256 		printf("%s: %p %s\n", str, vp, msg);
257 	if (vfs_badlock_panic)
258 		Debugger("Lock violation.\n");
259 }
260 
261 void
262 assert_vi_unlocked(struct vnode *vp, char *str)
263 {
264 	if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
265 		vfs_badlock("interlock is locked but should not be", str, vp);
266 }
267 
268 void
269 assert_vi_locked(struct vnode *vp, char *str)
270 {
271 	if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
272 		vfs_badlock("interlock is not locked but should be", str, vp);
273 }
274 
275 void
276 assert_vop_locked(struct vnode *vp, char *str)
277 {
278 	if (vp && !IGNORE_LOCK(vp) && !VOP_ISLOCKED(vp, NULL))
279 		vfs_badlock("is not locked but should be", str, vp);
280 }
281 
282 void
283 assert_vop_unlocked(struct vnode *vp, char *str)
284 {
285 	if (vp && !IGNORE_LOCK(vp) &&
286 	    VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE)
287 		vfs_badlock("is locked but should not be", str, vp);
288 }
289 
290 void
291 assert_vop_elocked(struct vnode *vp, char *str)
292 {
293 	if (vp && !IGNORE_LOCK(vp) &&
294 	    VOP_ISLOCKED(vp, curthread) != LK_EXCLUSIVE)
295 		vfs_badlock("is not exclusive locked but should be", str, vp);
296 }
297 
298 void
299 assert_vop_elocked_other(struct vnode *vp, char *str)
300 {
301 	if (vp && !IGNORE_LOCK(vp) &&
302 	    VOP_ISLOCKED(vp, curthread) != LK_EXCLOTHER)
303 		vfs_badlock("is not exclusive locked by another thread",
304 		    str, vp);
305 }
306 
307 void
308 assert_vop_slocked(struct vnode *vp, char *str)
309 {
310 	if (vp && !IGNORE_LOCK(vp) &&
311 	    VOP_ISLOCKED(vp, curthread) != LK_SHARED)
312 		vfs_badlock("is not locked shared but should be", str, vp);
313 }
314 
315 void
316 vop_rename_pre(void *ap)
317 {
318 	struct vop_rename_args *a = ap;
319 
320 	if (a->a_tvp)
321 		ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
322 	ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
323 	ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
324 	ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
325 
326 	/* Check the source (from) */
327 	if (a->a_tdvp != a->a_fdvp)
328 		ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked.\n");
329 	if (a->a_tvp != a->a_fvp)
330 		ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: tvp locked.\n");
331 
332 	/* Check the target */
333 	if (a->a_tvp)
334 		ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked.\n");
335 
336 	ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked.\n");
337 }
338 
339 void
340 vop_strategy_pre(void *ap)
341 {
342 	struct vop_strategy_args *a = ap;
343 	struct buf *bp;
344 
345 	bp = a->a_bp;
346 
347 	/*
348 	 * Cluster ops lock their component buffers but not the IO container.
349 	 */
350 	if ((bp->b_flags & B_CLUSTER) != 0)
351 		return;
352 
353 	if (BUF_REFCNT(bp) < 1) {
354 		if (vfs_badlock_print)
355 			printf("VOP_STRATEGY: bp is not locked but should be.\n");
356 		if (vfs_badlock_panic)
357 			Debugger("Lock violation.\n");
358 	}
359 }
360 
361 void
362 vop_lookup_pre(void *ap)
363 {
364 	struct vop_lookup_args *a = ap;
365 	struct vnode *dvp;
366 
367 	dvp = a->a_dvp;
368 
369 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
370 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
371 }
372 
373 void
374 vop_lookup_post(void *ap, int rc)
375 {
376 	struct vop_lookup_args *a = ap;
377 	struct componentname *cnp;
378 	struct vnode *dvp;
379 	struct vnode *vp;
380 	int flags;
381 
382 	dvp = a->a_dvp;
383 	cnp = a->a_cnp;
384 	vp = *(a->a_vpp);
385 	flags = cnp->cn_flags;
386 
387 
388 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
389 	/*
390 	 * If this is the last path component for this lookup and LOCPARENT
391 	 * is set, OR if there is an error the directory has to be locked.
392 	 */
393 	if ((flags & LOCKPARENT) && (flags & ISLASTCN))
394 		ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP (LOCKPARENT)");
395 	else if (rc != 0)
396 		ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP (error)");
397 	else if (dvp != vp)
398 		ASSERT_VOP_UNLOCKED(dvp, "VOP_LOOKUP (dvp)");
399 
400 	if (flags & PDIRUNLOCK)
401 		ASSERT_VOP_UNLOCKED(dvp, "VOP_LOOKUP (PDIRUNLOCK)");
402 }
403 
404 void
405 vop_unlock_pre(void *ap)
406 {
407 	struct vop_unlock_args *a = ap;
408 
409 	if (a->a_flags & LK_INTERLOCK)
410 		ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
411 
412 	ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
413 }
414 
415 void
416 vop_unlock_post(void *ap, int rc)
417 {
418 	struct vop_unlock_args *a = ap;
419 
420 	if (a->a_flags & LK_INTERLOCK)
421 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
422 }
423 
424 void
425 vop_lock_pre(void *ap)
426 {
427 	struct vop_lock_args *a = ap;
428 
429 	if ((a->a_flags & LK_INTERLOCK) == 0)
430 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
431 	else
432 		ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
433 }
434 
435 void
436 vop_lock_post(void *ap, int rc)
437 {
438 	struct vop_lock_args *a;
439 
440 	a = ap;
441 
442 	ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
443 	if (rc == 0)
444 		ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
445 }
446 
447 void
448 v_addpollinfo(struct vnode *vp)
449 {
450 	vp->v_pollinfo = uma_zalloc(vnodepoll_zone, M_WAITOK);
451 	mtx_init(&vp->v_pollinfo->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
452 }
453 
454 /*
455  * Initialize the vnode management data structures.
456  */
457 static void
458 vntblinit(void *dummy __unused)
459 {
460 
461 	desiredvnodes = maxproc + cnt.v_page_count / 4;
462 	minvnodes = desiredvnodes / 4;
463 	mtx_init(&mountlist_mtx, "mountlist", NULL, MTX_DEF);
464 	mtx_init(&mntvnode_mtx, "mntvnode", NULL, MTX_DEF);
465 	mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
466 	mtx_init(&spechash_mtx, "spechash", NULL, MTX_DEF);
467 	TAILQ_INIT(&vnode_free_list);
468 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
469 	vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
470 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
471 	vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
472 	      NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
473 	/*
474 	 * Initialize the filesystem syncer.
475 	 */
476 	syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
477 		&syncer_mask);
478 	syncer_maxdelay = syncer_mask + 1;
479 	mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
480 }
481 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL)
482 
483 
484 /*
485  * Mark a mount point as busy. Used to synchronize access and to delay
486  * unmounting. Interlock is not released on failure.
487  */
488 int
489 vfs_busy(mp, flags, interlkp, td)
490 	struct mount *mp;
491 	int flags;
492 	struct mtx *interlkp;
493 	struct thread *td;
494 {
495 	int lkflags;
496 
497 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
498 		if (flags & LK_NOWAIT)
499 			return (ENOENT);
500 		mp->mnt_kern_flag |= MNTK_MWAIT;
501 		/*
502 		 * Since all busy locks are shared except the exclusive
503 		 * lock granted when unmounting, the only place that a
504 		 * wakeup needs to be done is at the release of the
505 		 * exclusive lock at the end of dounmount.
506 		 */
507 		msleep(mp, interlkp, PVFS, "vfs_busy", 0);
508 		return (ENOENT);
509 	}
510 	lkflags = LK_SHARED | LK_NOPAUSE;
511 	if (interlkp)
512 		lkflags |= LK_INTERLOCK;
513 	if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
514 		panic("vfs_busy: unexpected lock failure");
515 	return (0);
516 }
517 
518 /*
519  * Free a busy filesystem.
520  */
521 void
522 vfs_unbusy(mp, td)
523 	struct mount *mp;
524 	struct thread *td;
525 {
526 
527 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
528 }
529 
530 /*
531  * Lookup a mount point by filesystem identifier.
532  */
533 struct mount *
534 vfs_getvfs(fsid)
535 	fsid_t *fsid;
536 {
537 	register struct mount *mp;
538 
539 	mtx_lock(&mountlist_mtx);
540 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
541 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
542 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
543 			mtx_unlock(&mountlist_mtx);
544 			return (mp);
545 		}
546 	}
547 	mtx_unlock(&mountlist_mtx);
548 	return ((struct mount *) 0);
549 }
550 
551 /*
552  * Get a new unique fsid.  Try to make its val[0] unique, since this value
553  * will be used to create fake device numbers for stat().  Also try (but
554  * not so hard) make its val[0] unique mod 2^16, since some emulators only
555  * support 16-bit device numbers.  We end up with unique val[0]'s for the
556  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
557  *
558  * Keep in mind that several mounts may be running in parallel.  Starting
559  * the search one past where the previous search terminated is both a
560  * micro-optimization and a defense against returning the same fsid to
561  * different mounts.
562  */
563 void
564 vfs_getnewfsid(mp)
565 	struct mount *mp;
566 {
567 	static u_int16_t mntid_base;
568 	fsid_t tfsid;
569 	int mtype;
570 
571 	mtx_lock(&mntid_mtx);
572 	mtype = mp->mnt_vfc->vfc_typenum;
573 	tfsid.val[1] = mtype;
574 	mtype = (mtype & 0xFF) << 24;
575 	for (;;) {
576 		tfsid.val[0] = makeudev(255,
577 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
578 		mntid_base++;
579 		if (vfs_getvfs(&tfsid) == NULL)
580 			break;
581 	}
582 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
583 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
584 	mtx_unlock(&mntid_mtx);
585 }
586 
587 /*
588  * Knob to control the precision of file timestamps:
589  *
590  *   0 = seconds only; nanoseconds zeroed.
591  *   1 = seconds and nanoseconds, accurate within 1/HZ.
592  *   2 = seconds and nanoseconds, truncated to microseconds.
593  * >=3 = seconds and nanoseconds, maximum precision.
594  */
595 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
596 
597 static int timestamp_precision = TSP_SEC;
598 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
599     &timestamp_precision, 0, "");
600 
601 /*
602  * Get a current timestamp.
603  */
604 void
605 vfs_timestamp(tsp)
606 	struct timespec *tsp;
607 {
608 	struct timeval tv;
609 
610 	switch (timestamp_precision) {
611 	case TSP_SEC:
612 		tsp->tv_sec = time_second;
613 		tsp->tv_nsec = 0;
614 		break;
615 	case TSP_HZ:
616 		getnanotime(tsp);
617 		break;
618 	case TSP_USEC:
619 		microtime(&tv);
620 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
621 		break;
622 	case TSP_NSEC:
623 	default:
624 		nanotime(tsp);
625 		break;
626 	}
627 }
628 
629 /*
630  * Set vnode attributes to VNOVAL
631  */
632 void
633 vattr_null(vap)
634 	register struct vattr *vap;
635 {
636 
637 	vap->va_type = VNON;
638 	vap->va_size = VNOVAL;
639 	vap->va_bytes = VNOVAL;
640 	vap->va_mode = VNOVAL;
641 	vap->va_nlink = VNOVAL;
642 	vap->va_uid = VNOVAL;
643 	vap->va_gid = VNOVAL;
644 	vap->va_fsid = VNOVAL;
645 	vap->va_fileid = VNOVAL;
646 	vap->va_blocksize = VNOVAL;
647 	vap->va_rdev = VNOVAL;
648 	vap->va_atime.tv_sec = VNOVAL;
649 	vap->va_atime.tv_nsec = VNOVAL;
650 	vap->va_mtime.tv_sec = VNOVAL;
651 	vap->va_mtime.tv_nsec = VNOVAL;
652 	vap->va_ctime.tv_sec = VNOVAL;
653 	vap->va_ctime.tv_nsec = VNOVAL;
654 	vap->va_birthtime.tv_sec = VNOVAL;
655 	vap->va_birthtime.tv_nsec = VNOVAL;
656 	vap->va_flags = VNOVAL;
657 	vap->va_gen = VNOVAL;
658 	vap->va_vaflags = 0;
659 }
660 
661 /*
662  * This routine is called when we have too many vnodes.  It attempts
663  * to free <count> vnodes and will potentially free vnodes that still
664  * have VM backing store (VM backing store is typically the cause
665  * of a vnode blowout so we want to do this).  Therefore, this operation
666  * is not considered cheap.
667  *
668  * A number of conditions may prevent a vnode from being reclaimed.
669  * the buffer cache may have references on the vnode, a directory
670  * vnode may still have references due to the namei cache representing
671  * underlying files, or the vnode may be in active use.   It is not
672  * desireable to reuse such vnodes.  These conditions may cause the
673  * number of vnodes to reach some minimum value regardless of what
674  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
675  */
676 static int
677 vlrureclaim(struct mount *mp, int count)
678 {
679 	struct vnode *vp;
680 	int done;
681 	int trigger;
682 	int usevnodes;
683 
684 	/*
685 	 * Calculate the trigger point, don't allow user
686 	 * screwups to blow us up.   This prevents us from
687 	 * recycling vnodes with lots of resident pages.  We
688 	 * aren't trying to free memory, we are trying to
689 	 * free vnodes.
690 	 */
691 	usevnodes = desiredvnodes;
692 	if (usevnodes <= 0)
693 		usevnodes = 1;
694 	trigger = cnt.v_page_count * 2 / usevnodes;
695 
696 	done = 0;
697 	mtx_lock(&mntvnode_mtx);
698 	while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
699 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
700 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
701 
702 		if (vp->v_type != VNON &&
703 		    vp->v_type != VBAD &&
704 		    VI_TRYLOCK(vp)) {
705 			if (VMIGHTFREE(vp) &&           /* critical path opt */
706 			    (vp->v_object == NULL ||
707 			    vp->v_object->resident_page_count < trigger)) {
708 				mtx_unlock(&mntvnode_mtx);
709 				vgonel(vp, curthread);
710 				done++;
711 				mtx_lock(&mntvnode_mtx);
712 			} else
713 				VI_UNLOCK(vp);
714 		}
715 		--count;
716 	}
717 	mtx_unlock(&mntvnode_mtx);
718 	return done;
719 }
720 
721 /*
722  * Attempt to recycle vnodes in a context that is always safe to block.
723  * Calling vlrurecycle() from the bowels of filesystem code has some
724  * interesting deadlock problems.
725  */
726 static struct proc *vnlruproc;
727 static int vnlruproc_sig;
728 
729 static void
730 vnlru_proc(void)
731 {
732 	struct mount *mp, *nmp;
733 	int s;
734 	int done, take;
735 	struct proc *p = vnlruproc;
736 	struct thread *td = FIRST_THREAD_IN_PROC(p);	/* XXXKSE */
737 
738 	mtx_lock(&Giant);
739 
740 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
741 	    SHUTDOWN_PRI_FIRST);
742 
743 	s = splbio();
744 	for (;;) {
745 		kthread_suspend_check(p);
746 		mtx_lock(&vnode_free_list_mtx);
747 		if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
748 			mtx_unlock(&vnode_free_list_mtx);
749 			vnlruproc_sig = 0;
750 			wakeup(&vnlruproc_sig);
751 			tsleep(vnlruproc, PVFS, "vlruwt", hz);
752 			continue;
753 		}
754 		mtx_unlock(&vnode_free_list_mtx);
755 		done = 0;
756 		mtx_lock(&mountlist_mtx);
757 		take = 0;
758 		TAILQ_FOREACH(mp, &mountlist, mnt_list)
759 			take++;
760 		take = desiredvnodes / (take * 10);
761 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
762 			if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
763 				nmp = TAILQ_NEXT(mp, mnt_list);
764 				continue;
765 			}
766 			done += vlrureclaim(mp, take);
767 			mtx_lock(&mountlist_mtx);
768 			nmp = TAILQ_NEXT(mp, mnt_list);
769 			vfs_unbusy(mp, td);
770 		}
771 		mtx_unlock(&mountlist_mtx);
772 		if (done == 0) {
773 #if 0
774 			/* These messages are temporary debugging aids */
775 			if (vnlru_nowhere < 5)
776 				printf("vnlru process getting nowhere..\n");
777 			else if (vnlru_nowhere == 5)
778 				printf("vnlru process messages stopped.\n");
779 #endif
780 			vnlru_nowhere++;
781 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
782 		}
783 	}
784 	splx(s);
785 }
786 
787 static struct kproc_desc vnlru_kp = {
788 	"vnlru",
789 	vnlru_proc,
790 	&vnlruproc
791 };
792 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
793 
794 
795 /*
796  * Routines having to do with the management of the vnode table.
797  */
798 
799 /*
800  * Check to see if a free vnode can be recycled. If it can,
801  * return it locked with the vn lock, but not interlock. Also
802  * get the vn_start_write lock. Otherwise indicate the error.
803  */
804 static int
805 vcanrecycle(struct vnode *vp, struct mount **vnmpp)
806 {
807 	struct thread *td = curthread;
808 	vm_object_t object;
809 	int error;
810 
811 	/* Don't recycle if we can't get the interlock */
812 	if (!VI_TRYLOCK(vp))
813 		return (EWOULDBLOCK);
814 
815 	/* We should be able to immediately acquire this */
816 	/* XXX This looks like it should panic if it fails */
817 	if (vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE, td) != 0) {
818 		if (VOP_ISLOCKED(vp, td))
819 			panic("vcanrecycle: locked vnode");
820 		return (EWOULDBLOCK);
821 	}
822 
823 	/*
824 	 * Don't recycle if its filesystem is being suspended.
825 	 */
826 	if (vn_start_write(vp, vnmpp, V_NOWAIT) != 0) {
827 		error = EBUSY;
828 		goto done;
829 	}
830 
831 	/*
832 	 * Don't recycle if we still have cached pages.
833 	 */
834 	if (VOP_GETVOBJECT(vp, &object) == 0 &&
835 	     (object->resident_page_count ||
836 	      object->ref_count)) {
837 		error = EBUSY;
838 		goto done;
839 	}
840 	if (LIST_FIRST(&vp->v_cache_src)) {
841 		/*
842 		 * note: nameileafonly sysctl is temporary,
843 		 * for debugging only, and will eventually be
844 		 * removed.
845 		 */
846 		if (nameileafonly > 0) {
847 			/*
848 			 * Do not reuse namei-cached directory
849 			 * vnodes that have cached
850 			 * subdirectories.
851 			 */
852 			if (cache_leaf_test(vp) < 0) {
853 				error = EISDIR;
854 				goto done;
855 			}
856 		} else if (nameileafonly < 0 ||
857 			    vmiodirenable == 0) {
858 			/*
859 			 * Do not reuse namei-cached directory
860 			 * vnodes if nameileafonly is -1 or
861 			 * if VMIO backing for directories is
862 			 * turned off (otherwise we reuse them
863 			 * too quickly).
864 			 */
865 			error = EBUSY;
866 			goto done;
867 		}
868 	}
869 	return (0);
870 done:
871 	VOP_UNLOCK(vp, 0, td);
872 	return (error);
873 }
874 
875 /*
876  * Return the next vnode from the free list.
877  */
878 int
879 getnewvnode(tag, mp, vops, vpp)
880 	const char *tag;
881 	struct mount *mp;
882 	vop_t **vops;
883 	struct vnode **vpp;
884 {
885 	int s;
886 	struct thread *td = curthread;	/* XXX */
887 	struct vnode *vp = NULL;
888 	struct vpollinfo *pollinfo = NULL;
889 	struct mount *vnmp;
890 
891 	s = splbio();
892 	mtx_lock(&vnode_free_list_mtx);
893 
894 	/*
895 	 * Try to reuse vnodes if we hit the max.  This situation only
896 	 * occurs in certain large-memory (2G+) situations.  We cannot
897 	 * attempt to directly reclaim vnodes due to nasty recursion
898 	 * problems.
899 	 */
900 	while (numvnodes - freevnodes > desiredvnodes) {
901 		if (vnlruproc_sig == 0) {
902 			vnlruproc_sig = 1;      /* avoid unnecessary wakeups */
903 			wakeup(vnlruproc);
904 		}
905 		mtx_unlock(&vnode_free_list_mtx);
906 		tsleep(&vnlruproc_sig, PVFS, "vlruwk", hz);
907 		mtx_lock(&vnode_free_list_mtx);
908 	}
909 
910 	/*
911 	 * Attempt to reuse a vnode already on the free list, allocating
912 	 * a new vnode if we can't find one or if we have not reached a
913 	 * good minimum for good LRU performance.
914 	 */
915 
916 	if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
917 		int error;
918 		int count;
919 
920 		for (count = 0; count < freevnodes; count++) {
921 			vp = TAILQ_FIRST(&vnode_free_list);
922 
923 			KASSERT(vp->v_usecount == 0 &&
924 			    (vp->v_iflag & VI_DOINGINACT) == 0,
925 			    ("getnewvnode: free vnode isn't"));
926 
927 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
928 			/*
929 			 * We have to drop the free list mtx to avoid lock
930 			 * order reversals with interlock.
931 			 */
932 			mtx_unlock(&vnode_free_list_mtx);
933 			error = vcanrecycle(vp, &vnmp);
934 			mtx_lock(&vnode_free_list_mtx);
935 			if (error == 0)
936 				break;
937 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
938 			vp = NULL;
939 		}
940 	}
941 	if (vp) {
942 		freevnodes--;
943 		mtx_unlock(&vnode_free_list_mtx);
944 
945 		cache_purge(vp);
946 		VI_LOCK(vp);
947 		vp->v_iflag |= VI_DOOMED;
948 		vp->v_iflag &= ~VI_FREE;
949 		if (vp->v_type != VBAD) {
950 			VOP_UNLOCK(vp, 0, td);
951 			vgonel(vp, td);
952 			VI_LOCK(vp);
953 		} else {
954 			VOP_UNLOCK(vp, 0, td);
955 		}
956 		vn_finished_write(vnmp);
957 
958 #ifdef INVARIANTS
959 		{
960 			if (vp->v_data)
961 				panic("cleaned vnode isn't");
962 			if (vp->v_numoutput)
963 				panic("Clean vnode has pending I/O's");
964 			if (vp->v_writecount != 0)
965 				panic("Non-zero write count");
966 		}
967 #endif
968 		if ((pollinfo = vp->v_pollinfo) != NULL) {
969 			/*
970 			 * To avoid lock order reversals, the call to
971 			 * uma_zfree() must be delayed until the vnode
972 			 * interlock is released.
973 			 */
974 			vp->v_pollinfo = NULL;
975 		}
976 #ifdef MAC
977 		mac_destroy_vnode(vp);
978 #endif
979 		vp->v_iflag = 0;
980 		vp->v_vflag = 0;
981 		vp->v_lastw = 0;
982 		vp->v_lasta = 0;
983 		vp->v_cstart = 0;
984 		vp->v_clen = 0;
985 		vp->v_socket = 0;
986 		lockdestroy(vp->v_vnlock);
987 		lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOPAUSE);
988 		KASSERT(vp->v_cleanbufcnt == 0, ("cleanbufcnt not 0"));
989 		KASSERT(vp->v_cleanblkroot == NULL, ("cleanblkroot not NULL"));
990 		KASSERT(vp->v_dirtybufcnt == 0, ("dirtybufcnt not 0"));
991 		KASSERT(vp->v_dirtyblkroot == NULL, ("dirtyblkroot not NULL"));
992 	} else {
993 		numvnodes++;
994 		mtx_unlock(&vnode_free_list_mtx);
995 
996 		vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
997 		mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
998 		VI_LOCK(vp);
999 		vp->v_dd = vp;
1000 		vp->v_vnlock = &vp->v_lock;
1001 		lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOPAUSE);
1002 		cache_purge(vp);
1003 		LIST_INIT(&vp->v_cache_src);
1004 		TAILQ_INIT(&vp->v_cache_dst);
1005 	}
1006 
1007 	TAILQ_INIT(&vp->v_cleanblkhd);
1008 	TAILQ_INIT(&vp->v_dirtyblkhd);
1009 	vp->v_type = VNON;
1010 	vp->v_tag = tag;
1011 	vp->v_op = vops;
1012 	*vpp = vp;
1013 	vp->v_usecount = 1;
1014 	vp->v_data = 0;
1015 	vp->v_cachedid = -1;
1016 	VI_UNLOCK(vp);
1017 	if (pollinfo != NULL) {
1018 		mtx_destroy(&pollinfo->vpi_lock);
1019 		uma_zfree(vnodepoll_zone, pollinfo);
1020 	}
1021 #ifdef MAC
1022 	mac_init_vnode(vp);
1023 	if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1024 		mac_associate_vnode_singlelabel(mp, vp);
1025 #endif
1026 	insmntque(vp, mp);
1027 
1028 	return (0);
1029 }
1030 
1031 /*
1032  * Move a vnode from one mount queue to another.
1033  */
1034 static void
1035 insmntque(vp, mp)
1036 	register struct vnode *vp;
1037 	register struct mount *mp;
1038 {
1039 
1040 	mtx_lock(&mntvnode_mtx);
1041 	/*
1042 	 * Delete from old mount point vnode list, if on one.
1043 	 */
1044 	if (vp->v_mount != NULL)
1045 		TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
1046 	/*
1047 	 * Insert into list of vnodes for the new mount point, if available.
1048 	 */
1049 	if ((vp->v_mount = mp) == NULL) {
1050 		mtx_unlock(&mntvnode_mtx);
1051 		return;
1052 	}
1053 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1054 	mtx_unlock(&mntvnode_mtx);
1055 }
1056 
1057 /*
1058  * Update outstanding I/O count and do wakeup if requested.
1059  */
1060 void
1061 vwakeup(bp)
1062 	register struct buf *bp;
1063 {
1064 	register struct vnode *vp;
1065 
1066 	bp->b_flags &= ~B_WRITEINPROG;
1067 	if ((vp = bp->b_vp)) {
1068 		VI_LOCK(vp);
1069 		vp->v_numoutput--;
1070 		if (vp->v_numoutput < 0)
1071 			panic("vwakeup: neg numoutput");
1072 		if ((vp->v_numoutput == 0) && (vp->v_iflag & VI_BWAIT)) {
1073 			vp->v_iflag &= ~VI_BWAIT;
1074 			wakeup(&vp->v_numoutput);
1075 		}
1076 		VI_UNLOCK(vp);
1077 	}
1078 }
1079 
1080 /*
1081  * Flush out and invalidate all buffers associated with a vnode.
1082  * Called with the underlying object locked.
1083  */
1084 int
1085 vinvalbuf(vp, flags, cred, td, slpflag, slptimeo)
1086 	struct vnode *vp;
1087 	int flags;
1088 	struct ucred *cred;
1089 	struct thread *td;
1090 	int slpflag, slptimeo;
1091 {
1092 	struct buf *blist;
1093 	int s, error;
1094 	vm_object_t object;
1095 
1096 	GIANT_REQUIRED;
1097 
1098 	ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1099 
1100 	VI_LOCK(vp);
1101 	if (flags & V_SAVE) {
1102 		s = splbio();
1103 		while (vp->v_numoutput) {
1104 			vp->v_iflag |= VI_BWAIT;
1105 			error = msleep(&vp->v_numoutput, VI_MTX(vp),
1106 			    slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo);
1107 			if (error) {
1108 				VI_UNLOCK(vp);
1109 				splx(s);
1110 				return (error);
1111 			}
1112 		}
1113 		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1114 			splx(s);
1115 			VI_UNLOCK(vp);
1116 			if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, td)) != 0)
1117 				return (error);
1118 			/*
1119 			 * XXX We could save a lock/unlock if this was only
1120 			 * enabled under INVARIANTS
1121 			 */
1122 			VI_LOCK(vp);
1123 			s = splbio();
1124 			if (vp->v_numoutput > 0 ||
1125 			    !TAILQ_EMPTY(&vp->v_dirtyblkhd))
1126 				panic("vinvalbuf: dirty bufs");
1127 		}
1128 		splx(s);
1129 	}
1130 	s = splbio();
1131 	/*
1132 	 * If you alter this loop please notice that interlock is dropped and
1133 	 * reacquired in flushbuflist.  Special care is needed to ensure that
1134 	 * no race conditions occur from this.
1135 	 */
1136 	for (error = 0;;) {
1137 		if ((blist = TAILQ_FIRST(&vp->v_cleanblkhd)) != 0 &&
1138 		    flushbuflist(blist, flags, vp, slpflag, slptimeo, &error)) {
1139 			if (error)
1140 				break;
1141 			continue;
1142 		}
1143 		if ((blist = TAILQ_FIRST(&vp->v_dirtyblkhd)) != 0 &&
1144 		    flushbuflist(blist, flags, vp, slpflag, slptimeo, &error)) {
1145 			if (error)
1146 				break;
1147 			continue;
1148 		}
1149 		break;
1150 	}
1151 	if (error) {
1152 		splx(s);
1153 		VI_UNLOCK(vp);
1154 		return (error);
1155 	}
1156 
1157 	/*
1158 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1159 	 * have write I/O in-progress but if there is a VM object then the
1160 	 * VM object can also have read-I/O in-progress.
1161 	 */
1162 	do {
1163 		while (vp->v_numoutput > 0) {
1164 			vp->v_iflag |= VI_BWAIT;
1165 			msleep(&vp->v_numoutput, VI_MTX(vp), PVM, "vnvlbv", 0);
1166 		}
1167 		VI_UNLOCK(vp);
1168 		if (VOP_GETVOBJECT(vp, &object) == 0) {
1169 			while (object->paging_in_progress)
1170 			vm_object_pip_sleep(object, "vnvlbx");
1171 		}
1172 		VI_LOCK(vp);
1173 	} while (vp->v_numoutput > 0);
1174 	VI_UNLOCK(vp);
1175 
1176 	splx(s);
1177 
1178 	/*
1179 	 * Destroy the copy in the VM cache, too.
1180 	 */
1181 	if (VOP_GETVOBJECT(vp, &object) == 0) {
1182 		vm_object_lock(object);
1183 		vm_object_page_remove(object, 0, 0,
1184 			(flags & V_SAVE) ? TRUE : FALSE);
1185 		vm_object_unlock(object);
1186 	}
1187 
1188 #ifdef INVARIANTS
1189 	VI_LOCK(vp);
1190 	if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1191 	    (!TAILQ_EMPTY(&vp->v_dirtyblkhd) ||
1192 	     !TAILQ_EMPTY(&vp->v_cleanblkhd)))
1193 		panic("vinvalbuf: flush failed");
1194 	VI_UNLOCK(vp);
1195 #endif
1196 	return (0);
1197 }
1198 
1199 /*
1200  * Flush out buffers on the specified list.
1201  *
1202  */
1203 static int
1204 flushbuflist(blist, flags, vp, slpflag, slptimeo, errorp)
1205 	struct buf *blist;
1206 	int flags;
1207 	struct vnode *vp;
1208 	int slpflag, slptimeo;
1209 	int *errorp;
1210 {
1211 	struct buf *bp, *nbp;
1212 	int found, error;
1213 
1214 	ASSERT_VI_LOCKED(vp, "flushbuflist");
1215 
1216 	for (found = 0, bp = blist; bp; bp = nbp) {
1217 		nbp = TAILQ_NEXT(bp, b_vnbufs);
1218 		if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1219 		    ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1220 			continue;
1221 		}
1222 		found += 1;
1223 		error = BUF_TIMELOCK(bp,
1224 		    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, VI_MTX(vp),
1225 		    "flushbuf", slpflag, slptimeo);
1226 		if (error) {
1227 			if (error != ENOLCK)
1228 				*errorp = error;
1229 			goto done;
1230 		}
1231 		/*
1232 		 * XXX Since there are no node locks for NFS, I
1233 		 * believe there is a slight chance that a delayed
1234 		 * write will occur while sleeping just above, so
1235 		 * check for it.  Note that vfs_bio_awrite expects
1236 		 * buffers to reside on a queue, while BUF_WRITE and
1237 		 * brelse do not.
1238 		 */
1239 		if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1240 			(flags & V_SAVE)) {
1241 
1242 			if (bp->b_vp == vp) {
1243 				if (bp->b_flags & B_CLUSTEROK) {
1244 					vfs_bio_awrite(bp);
1245 				} else {
1246 					bremfree(bp);
1247 					bp->b_flags |= B_ASYNC;
1248 					BUF_WRITE(bp);
1249 				}
1250 			} else {
1251 				bremfree(bp);
1252 				(void) BUF_WRITE(bp);
1253 			}
1254 			goto done;
1255 		}
1256 		bremfree(bp);
1257 		bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
1258 		bp->b_flags &= ~B_ASYNC;
1259 		brelse(bp);
1260 		VI_LOCK(vp);
1261 	}
1262 	return (found);
1263 done:
1264 	VI_LOCK(vp);
1265 	return (found);
1266 }
1267 
1268 /*
1269  * Truncate a file's buffer and pages to a specified length.  This
1270  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1271  * sync activity.
1272  */
1273 int
1274 vtruncbuf(vp, cred, td, length, blksize)
1275 	register struct vnode *vp;
1276 	struct ucred *cred;
1277 	struct thread *td;
1278 	off_t length;
1279 	int blksize;
1280 {
1281 	register struct buf *bp;
1282 	struct buf *nbp;
1283 	int s, anyfreed;
1284 	int trunclbn;
1285 
1286 	/*
1287 	 * Round up to the *next* lbn.
1288 	 */
1289 	trunclbn = (length + blksize - 1) / blksize;
1290 
1291 	s = splbio();
1292 	ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1293 restart:
1294 	VI_LOCK(vp);
1295 	anyfreed = 1;
1296 	for (;anyfreed;) {
1297 		anyfreed = 0;
1298 		for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
1299 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1300 			if (bp->b_lblkno >= trunclbn) {
1301 				if (BUF_LOCK(bp,
1302 				    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1303 				    VI_MTX(vp)) == ENOLCK)
1304 					goto restart;
1305 
1306 				bremfree(bp);
1307 				bp->b_flags |= (B_INVAL | B_RELBUF);
1308 				bp->b_flags &= ~B_ASYNC;
1309 				brelse(bp);
1310 				anyfreed = 1;
1311 
1312 				if (nbp &&
1313 				    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1314 				    (nbp->b_vp != vp) ||
1315 				    (nbp->b_flags & B_DELWRI))) {
1316 					goto restart;
1317 				}
1318 				VI_LOCK(vp);
1319 			}
1320 		}
1321 
1322 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1323 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1324 			if (bp->b_lblkno >= trunclbn) {
1325 				if (BUF_LOCK(bp,
1326 				    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1327 				    VI_MTX(vp)) == ENOLCK)
1328 					goto restart;
1329 				bremfree(bp);
1330 				bp->b_flags |= (B_INVAL | B_RELBUF);
1331 				bp->b_flags &= ~B_ASYNC;
1332 				brelse(bp);
1333 				anyfreed = 1;
1334 				if (nbp &&
1335 				    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1336 				    (nbp->b_vp != vp) ||
1337 				    (nbp->b_flags & B_DELWRI) == 0)) {
1338 					goto restart;
1339 				}
1340 				VI_LOCK(vp);
1341 			}
1342 		}
1343 	}
1344 
1345 	if (length > 0) {
1346 restartsync:
1347 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1348 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1349 			if (bp->b_lblkno > 0)
1350 				continue;
1351 			/*
1352 			 * Since we hold the vnode lock this should only
1353 			 * fail if we're racing with the buf daemon.
1354 			 */
1355 			if (BUF_LOCK(bp,
1356 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1357 			    VI_MTX(vp)) == ENOLCK) {
1358 				goto restart;
1359 			}
1360 			KASSERT((bp->b_flags & B_DELWRI),
1361 			    ("buf(%p) on dirty queue without DELWRI.", bp));
1362 
1363 			bremfree(bp);
1364 			bawrite(bp);
1365 			VI_LOCK(vp);
1366 			goto restartsync;
1367 		}
1368 	}
1369 
1370 	while (vp->v_numoutput > 0) {
1371 		vp->v_iflag |= VI_BWAIT;
1372 		msleep(&vp->v_numoutput, VI_MTX(vp), PVM, "vbtrunc", 0);
1373 	}
1374 	VI_UNLOCK(vp);
1375 	splx(s);
1376 
1377 	vnode_pager_setsize(vp, length);
1378 
1379 	return (0);
1380 }
1381 
1382 /*
1383  * buf_splay() - splay tree core for the clean/dirty list of buffers in
1384  * 		 a vnode.
1385  *
1386  *	NOTE: We have to deal with the special case of a background bitmap
1387  *	buffer, a situation where two buffers will have the same logical
1388  *	block offset.  We want (1) only the foreground buffer to be accessed
1389  *	in a lookup and (2) must differentiate between the foreground and
1390  *	background buffer in the splay tree algorithm because the splay
1391  *	tree cannot normally handle multiple entities with the same 'index'.
1392  *	We accomplish this by adding differentiating flags to the splay tree's
1393  *	numerical domain.
1394  */
1395 static
1396 struct buf *
1397 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1398 {
1399 	struct buf dummy;
1400 	struct buf *lefttreemax, *righttreemin, *y;
1401 
1402 	if (root == NULL)
1403 		return (NULL);
1404 	lefttreemax = righttreemin = &dummy;
1405 	for (;;) {
1406 		if (lblkno < root->b_lblkno ||
1407 		    (lblkno == root->b_lblkno &&
1408 		    (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1409 			if ((y = root->b_left) == NULL)
1410 				break;
1411 			if (lblkno < y->b_lblkno) {
1412 				/* Rotate right. */
1413 				root->b_left = y->b_right;
1414 				y->b_right = root;
1415 				root = y;
1416 				if ((y = root->b_left) == NULL)
1417 					break;
1418 			}
1419 			/* Link into the new root's right tree. */
1420 			righttreemin->b_left = root;
1421 			righttreemin = root;
1422 		} else if (lblkno > root->b_lblkno ||
1423 		    (lblkno == root->b_lblkno &&
1424 		    (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1425 			if ((y = root->b_right) == NULL)
1426 				break;
1427 			if (lblkno > y->b_lblkno) {
1428 				/* Rotate left. */
1429 				root->b_right = y->b_left;
1430 				y->b_left = root;
1431 				root = y;
1432 				if ((y = root->b_right) == NULL)
1433 					break;
1434 			}
1435 			/* Link into the new root's left tree. */
1436 			lefttreemax->b_right = root;
1437 			lefttreemax = root;
1438 		} else {
1439 			break;
1440 		}
1441 		root = y;
1442 	}
1443 	/* Assemble the new root. */
1444 	lefttreemax->b_right = root->b_left;
1445 	righttreemin->b_left = root->b_right;
1446 	root->b_left = dummy.b_right;
1447 	root->b_right = dummy.b_left;
1448 	return (root);
1449 }
1450 
1451 static
1452 void
1453 buf_vlist_remove(struct buf *bp)
1454 {
1455 	struct vnode *vp = bp->b_vp;
1456 	struct buf *root;
1457 
1458 	ASSERT_VI_LOCKED(vp, "buf_vlist_remove");
1459 	if (bp->b_xflags & BX_VNDIRTY) {
1460 		if (bp != vp->v_dirtyblkroot) {
1461 			root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_dirtyblkroot);
1462 			KASSERT(root == bp, ("splay lookup failed during dirty remove"));
1463 		}
1464 		if (bp->b_left == NULL) {
1465 			root = bp->b_right;
1466 		} else {
1467 			root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1468 			root->b_right = bp->b_right;
1469 		}
1470 		vp->v_dirtyblkroot = root;
1471 		TAILQ_REMOVE(&vp->v_dirtyblkhd, bp, b_vnbufs);
1472 		vp->v_dirtybufcnt--;
1473 	} else {
1474 		/* KASSERT(bp->b_xflags & BX_VNCLEAN, ("bp wasn't clean")); */
1475 		if (bp != vp->v_cleanblkroot) {
1476 			root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_cleanblkroot);
1477 			KASSERT(root == bp, ("splay lookup failed during clean remove"));
1478 		}
1479 		if (bp->b_left == NULL) {
1480 			root = bp->b_right;
1481 		} else {
1482 			root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1483 			root->b_right = bp->b_right;
1484 		}
1485 		vp->v_cleanblkroot = root;
1486 		TAILQ_REMOVE(&vp->v_cleanblkhd, bp, b_vnbufs);
1487 		vp->v_cleanbufcnt--;
1488 	}
1489 	bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1490 }
1491 
1492 /*
1493  * Add the buffer to the sorted clean or dirty block list using a
1494  * splay tree algorithm.
1495  *
1496  * NOTE: xflags is passed as a constant, optimizing this inline function!
1497  */
1498 static
1499 void
1500 buf_vlist_add(struct buf *bp, struct vnode *vp, b_xflags_t xflags)
1501 {
1502 	struct buf *root;
1503 
1504 	ASSERT_VI_LOCKED(vp, "buf_vlist_add");
1505 	bp->b_xflags |= xflags;
1506 	if (xflags & BX_VNDIRTY) {
1507 		root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_dirtyblkroot);
1508 		if (root == NULL) {
1509 			bp->b_left = NULL;
1510 			bp->b_right = NULL;
1511 			TAILQ_INSERT_TAIL(&vp->v_dirtyblkhd, bp, b_vnbufs);
1512 		} else if (bp->b_lblkno < root->b_lblkno ||
1513 		    (bp->b_lblkno == root->b_lblkno &&
1514 		    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1515 			bp->b_left = root->b_left;
1516 			bp->b_right = root;
1517 			root->b_left = NULL;
1518 			TAILQ_INSERT_BEFORE(root, bp, b_vnbufs);
1519 		} else {
1520 			bp->b_right = root->b_right;
1521 			bp->b_left = root;
1522 			root->b_right = NULL;
1523 			TAILQ_INSERT_AFTER(&vp->v_dirtyblkhd,
1524 			    root, bp, b_vnbufs);
1525 		}
1526 		vp->v_dirtybufcnt++;
1527 		vp->v_dirtyblkroot = bp;
1528 	} else {
1529 		/* KASSERT(xflags & BX_VNCLEAN, ("xflags not clean")); */
1530 		root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_cleanblkroot);
1531 		if (root == NULL) {
1532 			bp->b_left = NULL;
1533 			bp->b_right = NULL;
1534 			TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
1535 		} else if (bp->b_lblkno < root->b_lblkno ||
1536 		    (bp->b_lblkno == root->b_lblkno &&
1537 		    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1538 			bp->b_left = root->b_left;
1539 			bp->b_right = root;
1540 			root->b_left = NULL;
1541 			TAILQ_INSERT_BEFORE(root, bp, b_vnbufs);
1542 		} else {
1543 			bp->b_right = root->b_right;
1544 			bp->b_left = root;
1545 			root->b_right = NULL;
1546 			TAILQ_INSERT_AFTER(&vp->v_cleanblkhd,
1547 			    root, bp, b_vnbufs);
1548 		}
1549 		vp->v_cleanbufcnt++;
1550 		vp->v_cleanblkroot = bp;
1551 	}
1552 }
1553 
1554 /*
1555  * Lookup a buffer using the splay tree.  Note that we specifically avoid
1556  * shadow buffers used in background bitmap writes.
1557  *
1558  * This code isn't quite efficient as it could be because we are maintaining
1559  * two sorted lists and do not know which list the block resides in.
1560  */
1561 struct buf *
1562 gbincore(struct vnode *vp, daddr_t lblkno)
1563 {
1564 	struct buf *bp;
1565 
1566 	GIANT_REQUIRED;
1567 
1568 	ASSERT_VI_LOCKED(vp, "gbincore");
1569 	bp = vp->v_cleanblkroot = buf_splay(lblkno, 0, vp->v_cleanblkroot);
1570 	if (bp && bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1571 		return(bp);
1572 	bp = vp->v_dirtyblkroot = buf_splay(lblkno, 0, vp->v_dirtyblkroot);
1573 	if (bp && bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1574 		return(bp);
1575 	return(NULL);
1576 }
1577 
1578 /*
1579  * Associate a buffer with a vnode.
1580  */
1581 void
1582 bgetvp(vp, bp)
1583 	register struct vnode *vp;
1584 	register struct buf *bp;
1585 {
1586 	int s;
1587 
1588 	KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
1589 
1590 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1591 	    ("bgetvp: bp already attached! %p", bp));
1592 
1593 	ASSERT_VI_LOCKED(vp, "bgetvp");
1594 	vholdl(vp);
1595 	bp->b_vp = vp;
1596 	bp->b_dev = vn_todev(vp);
1597 	/*
1598 	 * Insert onto list for new vnode.
1599 	 */
1600 	s = splbio();
1601 	buf_vlist_add(bp, vp, BX_VNCLEAN);
1602 	splx(s);
1603 }
1604 
1605 /*
1606  * Disassociate a buffer from a vnode.
1607  */
1608 void
1609 brelvp(bp)
1610 	register struct buf *bp;
1611 {
1612 	struct vnode *vp;
1613 	int s;
1614 
1615 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1616 
1617 	/*
1618 	 * Delete from old vnode list, if on one.
1619 	 */
1620 	vp = bp->b_vp;
1621 	s = splbio();
1622 	VI_LOCK(vp);
1623 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1624 		buf_vlist_remove(bp);
1625 	if ((vp->v_iflag & VI_ONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1626 		vp->v_iflag &= ~VI_ONWORKLST;
1627 		mtx_lock(&sync_mtx);
1628 		LIST_REMOVE(vp, v_synclist);
1629 		mtx_unlock(&sync_mtx);
1630 	}
1631 	vdropl(vp);
1632 	VI_UNLOCK(vp);
1633 	bp->b_vp = (struct vnode *) 0;
1634 	if (bp->b_object)
1635 		bp->b_object = NULL;
1636 	splx(s);
1637 }
1638 
1639 /*
1640  * Add an item to the syncer work queue.
1641  */
1642 static void
1643 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
1644 {
1645 	int s, slot;
1646 
1647 	s = splbio();
1648 	ASSERT_VI_LOCKED(vp, "vn_syncer_add_to_worklist");
1649 
1650 	mtx_lock(&sync_mtx);
1651 	if (vp->v_iflag & VI_ONWORKLST)
1652 		LIST_REMOVE(vp, v_synclist);
1653 	else
1654 		vp->v_iflag |= VI_ONWORKLST;
1655 
1656 	if (delay > syncer_maxdelay - 2)
1657 		delay = syncer_maxdelay - 2;
1658 	slot = (syncer_delayno + delay) & syncer_mask;
1659 
1660 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
1661 	mtx_unlock(&sync_mtx);
1662 
1663 	splx(s);
1664 }
1665 
1666 struct  proc *updateproc;
1667 static void sched_sync(void);
1668 static struct kproc_desc up_kp = {
1669 	"syncer",
1670 	sched_sync,
1671 	&updateproc
1672 };
1673 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1674 
1675 /*
1676  * System filesystem synchronizer daemon.
1677  */
1678 static void
1679 sched_sync(void)
1680 {
1681 	struct synclist *slp;
1682 	struct vnode *vp;
1683 	struct mount *mp;
1684 	long starttime;
1685 	int s;
1686 	struct thread *td = FIRST_THREAD_IN_PROC(updateproc);  /* XXXKSE */
1687 
1688 	mtx_lock(&Giant);
1689 
1690 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, td->td_proc,
1691 	    SHUTDOWN_PRI_LAST);
1692 
1693 	for (;;) {
1694 		kthread_suspend_check(td->td_proc);
1695 
1696 		starttime = time_second;
1697 
1698 		/*
1699 		 * Push files whose dirty time has expired.  Be careful
1700 		 * of interrupt race on slp queue.
1701 		 */
1702 		s = splbio();
1703 		mtx_lock(&sync_mtx);
1704 		slp = &syncer_workitem_pending[syncer_delayno];
1705 		syncer_delayno += 1;
1706 		if (syncer_delayno == syncer_maxdelay)
1707 			syncer_delayno = 0;
1708 		splx(s);
1709 
1710 		while ((vp = LIST_FIRST(slp)) != NULL) {
1711 			mtx_unlock(&sync_mtx);
1712 			if (VOP_ISLOCKED(vp, NULL) == 0 &&
1713 			    vn_start_write(vp, &mp, V_NOWAIT) == 0) {
1714 				vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1715 				(void) VOP_FSYNC(vp, td->td_ucred, MNT_LAZY, td);
1716 				VOP_UNLOCK(vp, 0, td);
1717 				vn_finished_write(mp);
1718 			}
1719 			s = splbio();
1720 			mtx_lock(&sync_mtx);
1721 			if (LIST_FIRST(slp) == vp) {
1722 				mtx_unlock(&sync_mtx);
1723 				/*
1724 				 * Note: VFS vnodes can remain on the
1725 				 * worklist too with no dirty blocks, but
1726 				 * since sync_fsync() moves it to a different
1727 				 * slot we are safe.
1728 				 */
1729 				VI_LOCK(vp);
1730 				if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1731 				    !vn_isdisk(vp, NULL)) {
1732 					panic("sched_sync: fsync failed "
1733 					      "vp %p tag %s", vp, vp->v_tag);
1734 				}
1735 				/*
1736 				 * Put us back on the worklist.  The worklist
1737 				 * routine will remove us from our current
1738 				 * position and then add us back in at a later
1739 				 * position.
1740 				 */
1741 				vn_syncer_add_to_worklist(vp, syncdelay);
1742 				VI_UNLOCK(vp);
1743 				mtx_lock(&sync_mtx);
1744 			}
1745 			splx(s);
1746 		}
1747 		mtx_unlock(&sync_mtx);
1748 
1749 		/*
1750 		 * Do soft update processing.
1751 		 */
1752 		if (softdep_process_worklist_hook != NULL)
1753 			(*softdep_process_worklist_hook)(NULL);
1754 
1755 		/*
1756 		 * The variable rushjob allows the kernel to speed up the
1757 		 * processing of the filesystem syncer process. A rushjob
1758 		 * value of N tells the filesystem syncer to process the next
1759 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1760 		 * is used by the soft update code to speed up the filesystem
1761 		 * syncer process when the incore state is getting so far
1762 		 * ahead of the disk that the kernel memory pool is being
1763 		 * threatened with exhaustion.
1764 		 */
1765 		mtx_lock(&sync_mtx);
1766 		if (rushjob > 0) {
1767 			rushjob -= 1;
1768 			mtx_unlock(&sync_mtx);
1769 			continue;
1770 		}
1771 		mtx_unlock(&sync_mtx);
1772 		/*
1773 		 * If it has taken us less than a second to process the
1774 		 * current work, then wait. Otherwise start right over
1775 		 * again. We can still lose time if any single round
1776 		 * takes more than two seconds, but it does not really
1777 		 * matter as we are just trying to generally pace the
1778 		 * filesystem activity.
1779 		 */
1780 		if (time_second == starttime)
1781 			tsleep(&lbolt, PPAUSE, "syncer", 0);
1782 	}
1783 }
1784 
1785 /*
1786  * Request the syncer daemon to speed up its work.
1787  * We never push it to speed up more than half of its
1788  * normal turn time, otherwise it could take over the cpu.
1789  * XXXKSE  only one update?
1790  */
1791 int
1792 speedup_syncer()
1793 {
1794 	struct thread *td;
1795 	int ret = 0;
1796 
1797 	td = FIRST_THREAD_IN_PROC(updateproc);
1798 	mtx_lock_spin(&sched_lock);
1799 	if (td->td_wchan == &lbolt) {
1800 		unsleep(td);
1801 		TD_CLR_SLEEPING(td);
1802 		setrunnable(td);
1803 	}
1804 	mtx_unlock_spin(&sched_lock);
1805 	mtx_lock(&sync_mtx);
1806 	if (rushjob < syncdelay / 2) {
1807 		rushjob += 1;
1808 		stat_rush_requests += 1;
1809 		ret = 1;
1810 	}
1811 	mtx_unlock(&sync_mtx);
1812 	return (ret);
1813 }
1814 
1815 /*
1816  * Associate a p-buffer with a vnode.
1817  *
1818  * Also sets B_PAGING flag to indicate that vnode is not fully associated
1819  * with the buffer.  i.e. the bp has not been linked into the vnode or
1820  * ref-counted.
1821  */
1822 void
1823 pbgetvp(vp, bp)
1824 	register struct vnode *vp;
1825 	register struct buf *bp;
1826 {
1827 
1828 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1829 
1830 	bp->b_vp = vp;
1831 	bp->b_flags |= B_PAGING;
1832 	bp->b_dev = vn_todev(vp);
1833 }
1834 
1835 /*
1836  * Disassociate a p-buffer from a vnode.
1837  */
1838 void
1839 pbrelvp(bp)
1840 	register struct buf *bp;
1841 {
1842 
1843 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1844 
1845 	/* XXX REMOVE ME */
1846 	VI_LOCK(bp->b_vp);
1847 	if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1848 		panic(
1849 		    "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1850 		    bp,
1851 		    (int)bp->b_flags
1852 		);
1853 	}
1854 	VI_UNLOCK(bp->b_vp);
1855 	bp->b_vp = (struct vnode *) 0;
1856 	bp->b_flags &= ~B_PAGING;
1857 }
1858 
1859 /*
1860  * Reassign a buffer from one vnode to another.
1861  * Used to assign file specific control information
1862  * (indirect blocks) to the vnode to which they belong.
1863  */
1864 void
1865 reassignbuf(bp, newvp)
1866 	register struct buf *bp;
1867 	register struct vnode *newvp;
1868 {
1869 	int delay;
1870 	int s;
1871 
1872 	if (newvp == NULL) {
1873 		printf("reassignbuf: NULL");
1874 		return;
1875 	}
1876 	++reassignbufcalls;
1877 
1878 	/*
1879 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1880 	 * is not fully linked in.
1881 	 */
1882 	if (bp->b_flags & B_PAGING)
1883 		panic("cannot reassign paging buffer");
1884 
1885 	s = splbio();
1886 	/*
1887 	 * Delete from old vnode list, if on one.
1888 	 */
1889 	VI_LOCK(bp->b_vp);
1890 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1891 		buf_vlist_remove(bp);
1892 		if (bp->b_vp != newvp) {
1893 			vdropl(bp->b_vp);
1894 			bp->b_vp = NULL;	/* for clarification */
1895 		}
1896 	}
1897 	VI_UNLOCK(bp->b_vp);
1898 	/*
1899 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1900 	 * of clean buffers.
1901 	 */
1902 	VI_LOCK(newvp);
1903 	if (bp->b_flags & B_DELWRI) {
1904 		if ((newvp->v_iflag & VI_ONWORKLST) == 0) {
1905 			switch (newvp->v_type) {
1906 			case VDIR:
1907 				delay = dirdelay;
1908 				break;
1909 			case VCHR:
1910 				if (newvp->v_rdev->si_mountpoint != NULL) {
1911 					delay = metadelay;
1912 					break;
1913 				}
1914 				/* FALLTHROUGH */
1915 			default:
1916 				delay = filedelay;
1917 			}
1918 			vn_syncer_add_to_worklist(newvp, delay);
1919 		}
1920 		buf_vlist_add(bp, newvp, BX_VNDIRTY);
1921 	} else {
1922 		buf_vlist_add(bp, newvp, BX_VNCLEAN);
1923 
1924 		if ((newvp->v_iflag & VI_ONWORKLST) &&
1925 		    TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1926 			mtx_lock(&sync_mtx);
1927 			LIST_REMOVE(newvp, v_synclist);
1928 			mtx_unlock(&sync_mtx);
1929 			newvp->v_iflag &= ~VI_ONWORKLST;
1930 		}
1931 	}
1932 	if (bp->b_vp != newvp) {
1933 		bp->b_vp = newvp;
1934 		vholdl(bp->b_vp);
1935 	}
1936 	VI_UNLOCK(newvp);
1937 	splx(s);
1938 }
1939 
1940 /*
1941  * Create a vnode for a device.
1942  * Used for mounting the root filesystem.
1943  */
1944 int
1945 bdevvp(dev, vpp)
1946 	dev_t dev;
1947 	struct vnode **vpp;
1948 {
1949 	register struct vnode *vp;
1950 	struct vnode *nvp;
1951 	int error;
1952 
1953 	if (dev == NODEV) {
1954 		*vpp = NULLVP;
1955 		return (ENXIO);
1956 	}
1957 	if (vfinddev(dev, VCHR, vpp))
1958 		return (0);
1959 	error = getnewvnode("none", (struct mount *)0, spec_vnodeop_p, &nvp);
1960 	if (error) {
1961 		*vpp = NULLVP;
1962 		return (error);
1963 	}
1964 	vp = nvp;
1965 	vp->v_type = VCHR;
1966 	addalias(vp, dev);
1967 	*vpp = vp;
1968 	return (0);
1969 }
1970 
1971 static void
1972 v_incr_usecount(struct vnode *vp, int delta)
1973 {
1974 	vp->v_usecount += delta;
1975 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1976 		mtx_lock(&spechash_mtx);
1977 		vp->v_rdev->si_usecount += delta;
1978 		mtx_unlock(&spechash_mtx);
1979 	}
1980 }
1981 
1982 /*
1983  * Add vnode to the alias list hung off the dev_t.
1984  *
1985  * The reason for this gunk is that multiple vnodes can reference
1986  * the same physical device, so checking vp->v_usecount to see
1987  * how many users there are is inadequate; the v_usecount for
1988  * the vnodes need to be accumulated.  vcount() does that.
1989  */
1990 struct vnode *
1991 addaliasu(nvp, nvp_rdev)
1992 	struct vnode *nvp;
1993 	udev_t nvp_rdev;
1994 {
1995 	struct vnode *ovp;
1996 	vop_t **ops;
1997 	dev_t dev;
1998 
1999 	if (nvp->v_type == VBLK)
2000 		return (nvp);
2001 	if (nvp->v_type != VCHR)
2002 		panic("addaliasu on non-special vnode");
2003 	dev = udev2dev(nvp_rdev, 0);
2004 	/*
2005 	 * Check to see if we have a bdevvp vnode with no associated
2006 	 * filesystem. If so, we want to associate the filesystem of
2007 	 * the new newly instigated vnode with the bdevvp vnode and
2008 	 * discard the newly created vnode rather than leaving the
2009 	 * bdevvp vnode lying around with no associated filesystem.
2010 	 */
2011 	if (vfinddev(dev, nvp->v_type, &ovp) == 0 || ovp->v_data != NULL) {
2012 		addalias(nvp, dev);
2013 		return (nvp);
2014 	}
2015 	/*
2016 	 * Discard unneeded vnode, but save its node specific data.
2017 	 * Note that if there is a lock, it is carried over in the
2018 	 * node specific data to the replacement vnode.
2019 	 */
2020 	vref(ovp);
2021 	ovp->v_data = nvp->v_data;
2022 	ovp->v_tag = nvp->v_tag;
2023 	nvp->v_data = NULL;
2024 	lockdestroy(ovp->v_vnlock);
2025 	lockinit(ovp->v_vnlock, PVFS, nvp->v_vnlock->lk_wmesg,
2026 	    nvp->v_vnlock->lk_timo, nvp->v_vnlock->lk_flags & LK_EXTFLG_MASK);
2027 	ops = ovp->v_op;
2028 	ovp->v_op = nvp->v_op;
2029 	if (VOP_ISLOCKED(nvp, curthread)) {
2030 		VOP_UNLOCK(nvp, 0, curthread);
2031 		vn_lock(ovp, LK_EXCLUSIVE | LK_RETRY, curthread);
2032 	}
2033 	nvp->v_op = ops;
2034 	insmntque(ovp, nvp->v_mount);
2035 	vrele(nvp);
2036 	vgone(nvp);
2037 	return (ovp);
2038 }
2039 
2040 /* This is a local helper function that do the same as addaliasu, but for a
2041  * dev_t instead of an udev_t. */
2042 static void
2043 addalias(nvp, dev)
2044 	struct vnode *nvp;
2045 	dev_t dev;
2046 {
2047 
2048 	KASSERT(nvp->v_type == VCHR, ("addalias on non-special vnode"));
2049 	nvp->v_rdev = dev;
2050 	VI_LOCK(nvp);
2051 	mtx_lock(&spechash_mtx);
2052 	SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
2053 	dev->si_usecount += nvp->v_usecount;
2054 	mtx_unlock(&spechash_mtx);
2055 	VI_UNLOCK(nvp);
2056 }
2057 
2058 /*
2059  * Grab a particular vnode from the free list, increment its
2060  * reference count and lock it. The vnode lock bit is set if the
2061  * vnode is being eliminated in vgone. The process is awakened
2062  * when the transition is completed, and an error returned to
2063  * indicate that the vnode is no longer usable (possibly having
2064  * been changed to a new filesystem type).
2065  */
2066 int
2067 vget(vp, flags, td)
2068 	register struct vnode *vp;
2069 	int flags;
2070 	struct thread *td;
2071 {
2072 	int error;
2073 
2074 	/*
2075 	 * If the vnode is in the process of being cleaned out for
2076 	 * another use, we wait for the cleaning to finish and then
2077 	 * return failure. Cleaning is determined by checking that
2078 	 * the VI_XLOCK flag is set.
2079 	 */
2080 	if ((flags & LK_INTERLOCK) == 0)
2081 		VI_LOCK(vp);
2082 	if (vp->v_iflag & VI_XLOCK && vp->v_vxproc != curthread) {
2083 		vp->v_iflag |= VI_XWANT;
2084 		msleep(vp, VI_MTX(vp), PINOD | PDROP, "vget", 0);
2085 		return (ENOENT);
2086 	}
2087 
2088 	v_incr_usecount(vp, 1);
2089 
2090 	if (VSHOULDBUSY(vp))
2091 		vbusy(vp);
2092 	if (flags & LK_TYPE_MASK) {
2093 		if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
2094 			/*
2095 			 * must expand vrele here because we do not want
2096 			 * to call VOP_INACTIVE if the reference count
2097 			 * drops back to zero since it was never really
2098 			 * active. We must remove it from the free list
2099 			 * before sleeping so that multiple processes do
2100 			 * not try to recycle it.
2101 			 */
2102 			VI_LOCK(vp);
2103 			v_incr_usecount(vp, -1);
2104 			if (VSHOULDFREE(vp))
2105 				vfree(vp);
2106 			else
2107 				vlruvp(vp);
2108 			VI_UNLOCK(vp);
2109 		}
2110 		return (error);
2111 	}
2112 	VI_UNLOCK(vp);
2113 	return (0);
2114 }
2115 
2116 /*
2117  * Increase the reference count of a vnode.
2118  */
2119 void
2120 vref(struct vnode *vp)
2121 {
2122 	VI_LOCK(vp);
2123 	v_incr_usecount(vp, 1);
2124 	VI_UNLOCK(vp);
2125 }
2126 
2127 /*
2128  * Return reference count of a vnode.
2129  *
2130  * The results of this call are only guaranteed when some mechanism other
2131  * than the VI lock is used to stop other processes from gaining references
2132  * to the vnode.  This may be the case if the caller holds the only reference.
2133  * This is also useful when stale data is acceptable as race conditions may
2134  * be accounted for by some other means.
2135  */
2136 int
2137 vrefcnt(struct vnode *vp)
2138 {
2139 	int usecnt;
2140 
2141 	VI_LOCK(vp);
2142 	usecnt = vp->v_usecount;
2143 	VI_UNLOCK(vp);
2144 
2145 	return (usecnt);
2146 }
2147 
2148 
2149 /*
2150  * Vnode put/release.
2151  * If count drops to zero, call inactive routine and return to freelist.
2152  */
2153 void
2154 vrele(vp)
2155 	struct vnode *vp;
2156 {
2157 	struct thread *td = curthread;	/* XXX */
2158 
2159 	KASSERT(vp != NULL, ("vrele: null vp"));
2160 
2161 	VI_LOCK(vp);
2162 
2163 	/* Skip this v_writecount check if we're going to panic below. */
2164 	KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
2165 	    ("vrele: missed vn_close"));
2166 
2167 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2168 	    vp->v_usecount == 1)) {
2169 		v_incr_usecount(vp, -1);
2170 		VI_UNLOCK(vp);
2171 
2172 		return;
2173 	}
2174 
2175 	if (vp->v_usecount == 1) {
2176 		v_incr_usecount(vp, -1);
2177 		/*
2178 		 * We must call VOP_INACTIVE with the node locked. Mark
2179 		 * as VI_DOINGINACT to avoid recursion.
2180 		 */
2181 		if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0) {
2182 			VI_LOCK(vp);
2183 			vp->v_iflag |= VI_DOINGINACT;
2184 			VI_UNLOCK(vp);
2185 			VOP_INACTIVE(vp, td);
2186 			VI_LOCK(vp);
2187 			KASSERT(vp->v_iflag & VI_DOINGINACT,
2188 			    ("vrele: lost VI_DOINGINACT"));
2189 			vp->v_iflag &= ~VI_DOINGINACT;
2190 			VI_UNLOCK(vp);
2191 		}
2192 		VI_LOCK(vp);
2193 		if (VSHOULDFREE(vp))
2194 			vfree(vp);
2195 		else
2196 			vlruvp(vp);
2197 		VI_UNLOCK(vp);
2198 
2199 	} else {
2200 #ifdef DIAGNOSTIC
2201 		vprint("vrele: negative ref count", vp);
2202 #endif
2203 		VI_UNLOCK(vp);
2204 		panic("vrele: negative ref cnt");
2205 	}
2206 }
2207 
2208 /*
2209  * Release an already locked vnode.  This give the same effects as
2210  * unlock+vrele(), but takes less time and avoids releasing and
2211  * re-aquiring the lock (as vrele() aquires the lock internally.)
2212  */
2213 void
2214 vput(vp)
2215 	struct vnode *vp;
2216 {
2217 	struct thread *td = curthread;	/* XXX */
2218 
2219 	GIANT_REQUIRED;
2220 
2221 	KASSERT(vp != NULL, ("vput: null vp"));
2222 	VI_LOCK(vp);
2223 	/* Skip this v_writecount check if we're going to panic below. */
2224 	KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
2225 	    ("vput: missed vn_close"));
2226 
2227 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2228 	    vp->v_usecount == 1)) {
2229 		v_incr_usecount(vp, -1);
2230 		VOP_UNLOCK(vp, LK_INTERLOCK, td);
2231 		return;
2232 	}
2233 
2234 	if (vp->v_usecount == 1) {
2235 		v_incr_usecount(vp, -1);
2236 		/*
2237 		 * We must call VOP_INACTIVE with the node locked, so
2238 		 * we just need to release the vnode mutex. Mark as
2239 		 * as VI_DOINGINACT to avoid recursion.
2240 		 */
2241 		vp->v_iflag |= VI_DOINGINACT;
2242 		VI_UNLOCK(vp);
2243 		VOP_INACTIVE(vp, td);
2244 		VI_LOCK(vp);
2245 		KASSERT(vp->v_iflag & VI_DOINGINACT,
2246 		    ("vput: lost VI_DOINGINACT"));
2247 		vp->v_iflag &= ~VI_DOINGINACT;
2248 		if (VSHOULDFREE(vp))
2249 			vfree(vp);
2250 		else
2251 			vlruvp(vp);
2252 		VI_UNLOCK(vp);
2253 
2254 	} else {
2255 #ifdef DIAGNOSTIC
2256 		vprint("vput: negative ref count", vp);
2257 #endif
2258 		panic("vput: negative ref cnt");
2259 	}
2260 }
2261 
2262 /*
2263  * Somebody doesn't want the vnode recycled.
2264  */
2265 void
2266 vhold(struct vnode *vp)
2267 {
2268 	VI_LOCK(vp);
2269 	vholdl(vp);
2270 	VI_UNLOCK(vp);
2271 }
2272 
2273 void
2274 vholdl(vp)
2275 	register struct vnode *vp;
2276 {
2277 	int s;
2278 
2279 	s = splbio();
2280 	vp->v_holdcnt++;
2281 	if (VSHOULDBUSY(vp))
2282 		vbusy(vp);
2283 	splx(s);
2284 }
2285 
2286 /*
2287  * Note that there is one less who cares about this vnode.  vdrop() is the
2288  * opposite of vhold().
2289  */
2290 void
2291 vdrop(struct vnode *vp)
2292 {
2293 	VI_LOCK(vp);
2294 	vdropl(vp);
2295 	VI_UNLOCK(vp);
2296 }
2297 
2298 void
2299 vdropl(vp)
2300 	register struct vnode *vp;
2301 {
2302 	int s;
2303 
2304 	s = splbio();
2305 	if (vp->v_holdcnt <= 0)
2306 		panic("vdrop: holdcnt");
2307 	vp->v_holdcnt--;
2308 	if (VSHOULDFREE(vp))
2309 		vfree(vp);
2310 	else
2311 		vlruvp(vp);
2312 	splx(s);
2313 }
2314 
2315 /*
2316  * Remove any vnodes in the vnode table belonging to mount point mp.
2317  *
2318  * If FORCECLOSE is not specified, there should not be any active ones,
2319  * return error if any are found (nb: this is a user error, not a
2320  * system error). If FORCECLOSE is specified, detach any active vnodes
2321  * that are found.
2322  *
2323  * If WRITECLOSE is set, only flush out regular file vnodes open for
2324  * writing.
2325  *
2326  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2327  *
2328  * `rootrefs' specifies the base reference count for the root vnode
2329  * of this filesystem. The root vnode is considered busy if its
2330  * v_usecount exceeds this value. On a successful return, vflush()
2331  * will call vrele() on the root vnode exactly rootrefs times.
2332  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2333  * be zero.
2334  */
2335 #ifdef DIAGNOSTIC
2336 static int busyprt = 0;		/* print out busy vnodes */
2337 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
2338 #endif
2339 
2340 int
2341 vflush(mp, rootrefs, flags)
2342 	struct mount *mp;
2343 	int rootrefs;
2344 	int flags;
2345 {
2346 	struct thread *td = curthread;	/* XXX */
2347 	struct vnode *vp, *nvp, *rootvp = NULL;
2348 	struct vattr vattr;
2349 	int busy = 0, error;
2350 
2351 	if (rootrefs > 0) {
2352 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2353 		    ("vflush: bad args"));
2354 		/*
2355 		 * Get the filesystem root vnode. We can vput() it
2356 		 * immediately, since with rootrefs > 0, it won't go away.
2357 		 */
2358 		if ((error = VFS_ROOT(mp, &rootvp)) != 0)
2359 			return (error);
2360 		vput(rootvp);
2361 
2362 	}
2363 	mtx_lock(&mntvnode_mtx);
2364 loop:
2365 	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) {
2366 		/*
2367 		 * Make sure this vnode wasn't reclaimed in getnewvnode().
2368 		 * Start over if it has (it won't be on the list anymore).
2369 		 */
2370 		if (vp->v_mount != mp)
2371 			goto loop;
2372 		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2373 
2374 		VI_LOCK(vp);
2375 		mtx_unlock(&mntvnode_mtx);
2376 		vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY, td);
2377 		/*
2378 		 * Skip over a vnodes marked VV_SYSTEM.
2379 		 */
2380 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2381 			VOP_UNLOCK(vp, 0, td);
2382 			mtx_lock(&mntvnode_mtx);
2383 			continue;
2384 		}
2385 		/*
2386 		 * If WRITECLOSE is set, flush out unlinked but still open
2387 		 * files (even if open only for reading) and regular file
2388 		 * vnodes open for writing.
2389 		 */
2390 		if (flags & WRITECLOSE) {
2391 			error = VOP_GETATTR(vp, &vattr, td->td_ucred, td);
2392 			VI_LOCK(vp);
2393 
2394 			if ((vp->v_type == VNON ||
2395 			    (error == 0 && vattr.va_nlink > 0)) &&
2396 			    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2397 				VOP_UNLOCK(vp, LK_INTERLOCK, td);
2398 				mtx_lock(&mntvnode_mtx);
2399 				continue;
2400 			}
2401 		} else
2402 			VI_LOCK(vp);
2403 
2404 		VOP_UNLOCK(vp, 0, td);
2405 
2406 		/*
2407 		 * With v_usecount == 0, all we need to do is clear out the
2408 		 * vnode data structures and we are done.
2409 		 */
2410 		if (vp->v_usecount == 0) {
2411 			vgonel(vp, td);
2412 			mtx_lock(&mntvnode_mtx);
2413 			continue;
2414 		}
2415 
2416 		/*
2417 		 * If FORCECLOSE is set, forcibly close the vnode. For block
2418 		 * or character devices, revert to an anonymous device. For
2419 		 * all other files, just kill them.
2420 		 */
2421 		if (flags & FORCECLOSE) {
2422 			if (vp->v_type != VCHR) {
2423 				vgonel(vp, td);
2424 			} else {
2425 				vclean(vp, 0, td);
2426 				VI_UNLOCK(vp);
2427 				vp->v_op = spec_vnodeop_p;
2428 				insmntque(vp, (struct mount *) 0);
2429 			}
2430 			mtx_lock(&mntvnode_mtx);
2431 			continue;
2432 		}
2433 #ifdef DIAGNOSTIC
2434 		if (busyprt)
2435 			vprint("vflush: busy vnode", vp);
2436 #endif
2437 		VI_UNLOCK(vp);
2438 		mtx_lock(&mntvnode_mtx);
2439 		busy++;
2440 	}
2441 	mtx_unlock(&mntvnode_mtx);
2442 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2443 		/*
2444 		 * If just the root vnode is busy, and if its refcount
2445 		 * is equal to `rootrefs', then go ahead and kill it.
2446 		 */
2447 		VI_LOCK(rootvp);
2448 		KASSERT(busy > 0, ("vflush: not busy"));
2449 		KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
2450 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2451 			vgonel(rootvp, td);
2452 			busy = 0;
2453 		} else
2454 			VI_UNLOCK(rootvp);
2455 	}
2456 	if (busy)
2457 		return (EBUSY);
2458 	for (; rootrefs > 0; rootrefs--)
2459 		vrele(rootvp);
2460 	return (0);
2461 }
2462 
2463 /*
2464  * This moves a now (likely recyclable) vnode to the end of the
2465  * mountlist.  XXX However, it is temporarily disabled until we
2466  * can clean up ffs_sync() and friends, which have loop restart
2467  * conditions which this code causes to operate O(N^2).
2468  */
2469 static void
2470 vlruvp(struct vnode *vp)
2471 {
2472 #if 0
2473 	struct mount *mp;
2474 
2475 	if ((mp = vp->v_mount) != NULL) {
2476 		mtx_lock(&mntvnode_mtx);
2477 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
2478 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
2479 		mtx_unlock(&mntvnode_mtx);
2480 	}
2481 #endif
2482 }
2483 
2484 /*
2485  * Disassociate the underlying filesystem from a vnode.
2486  */
2487 static void
2488 vclean(vp, flags, td)
2489 	struct vnode *vp;
2490 	int flags;
2491 	struct thread *td;
2492 {
2493 	int active;
2494 
2495 	ASSERT_VI_LOCKED(vp, "vclean");
2496 	/*
2497 	 * Check to see if the vnode is in use. If so we have to reference it
2498 	 * before we clean it out so that its count cannot fall to zero and
2499 	 * generate a race against ourselves to recycle it.
2500 	 */
2501 	if ((active = vp->v_usecount))
2502 		v_incr_usecount(vp, 1);
2503 
2504 	/*
2505 	 * Prevent the vnode from being recycled or brought into use while we
2506 	 * clean it out.
2507 	 */
2508 	if (vp->v_iflag & VI_XLOCK)
2509 		panic("vclean: deadlock");
2510 	vp->v_iflag |= VI_XLOCK;
2511 	vp->v_vxproc = curthread;
2512 	/*
2513 	 * Even if the count is zero, the VOP_INACTIVE routine may still
2514 	 * have the object locked while it cleans it out. The VOP_LOCK
2515 	 * ensures that the VOP_INACTIVE routine is done with its work.
2516 	 * For active vnodes, it ensures that no other activity can
2517 	 * occur while the underlying object is being cleaned out.
2518 	 */
2519 	VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td);
2520 
2521 	/*
2522 	 * Clean out any buffers associated with the vnode.
2523 	 * If the flush fails, just toss the buffers.
2524 	 */
2525 	if (flags & DOCLOSE) {
2526 		struct buf *bp;
2527 		VI_LOCK(vp);
2528 		bp = TAILQ_FIRST(&vp->v_dirtyblkhd);
2529 		VI_UNLOCK(vp);
2530 		if (bp != NULL)
2531 			(void) vn_write_suspend_wait(vp, NULL, V_WAIT);
2532 		if (vinvalbuf(vp, V_SAVE, NOCRED, td, 0, 0) != 0)
2533 			vinvalbuf(vp, 0, NOCRED, td, 0, 0);
2534 	}
2535 
2536 	VOP_DESTROYVOBJECT(vp);
2537 
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 
2544 	/*
2545 	 * If purging an active vnode, it must be closed and
2546 	 * deactivated before being reclaimed. Note that the
2547 	 * VOP_INACTIVE will unlock the vnode.
2548 	 */
2549 	if (active) {
2550 		if (flags & DOCLOSE)
2551 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2552 		VI_LOCK(vp);
2553 		if ((vp->v_iflag & VI_DOINGINACT) == 0) {
2554 			vp->v_iflag |= VI_DOINGINACT;
2555 			VI_UNLOCK(vp);
2556 			if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT, td) != 0)
2557 				panic("vclean: cannot relock.");
2558 			VOP_INACTIVE(vp, td);
2559 			VI_LOCK(vp);
2560 			KASSERT(vp->v_iflag & VI_DOINGINACT,
2561 			    ("vclean: lost VI_DOINGINACT"));
2562 			vp->v_iflag &= ~VI_DOINGINACT;
2563 		}
2564 		VI_UNLOCK(vp);
2565 	}
2566 
2567 	/*
2568 	 * Reclaim the vnode.
2569 	 */
2570 	if (VOP_RECLAIM(vp, td))
2571 		panic("vclean: cannot reclaim");
2572 
2573 	if (active) {
2574 		/*
2575 		 * Inline copy of vrele() since VOP_INACTIVE
2576 		 * has already been called.
2577 		 */
2578 		VI_LOCK(vp);
2579 		v_incr_usecount(vp, -1);
2580 		if (vp->v_usecount <= 0) {
2581 #ifdef DIAGNOSTIC
2582 			if (vp->v_usecount < 0 || vp->v_writecount != 0) {
2583 				vprint("vclean: bad ref count", vp);
2584 				panic("vclean: ref cnt");
2585 			}
2586 #endif
2587 			vfree(vp);
2588 		}
2589 		VI_UNLOCK(vp);
2590 	}
2591 
2592 	cache_purge(vp);
2593 	VI_LOCK(vp);
2594 	if (VSHOULDFREE(vp))
2595 		vfree(vp);
2596 
2597 	/*
2598 	 * Done with purge, reset to the standard lock and
2599 	 * notify sleepers of the grim news.
2600 	 */
2601 	vp->v_vnlock = &vp->v_lock;
2602 	vp->v_op = dead_vnodeop_p;
2603 	if (vp->v_pollinfo != NULL)
2604 		vn_pollgone(vp);
2605 	vp->v_tag = "none";
2606 	vp->v_iflag &= ~VI_XLOCK;
2607 	vp->v_vxproc = NULL;
2608 	if (vp->v_iflag & VI_XWANT) {
2609 		vp->v_iflag &= ~VI_XWANT;
2610 		wakeup(vp);
2611 	}
2612 }
2613 
2614 /*
2615  * Eliminate all activity associated with the requested vnode
2616  * and with all vnodes aliased to the requested vnode.
2617  */
2618 int
2619 vop_revoke(ap)
2620 	struct vop_revoke_args /* {
2621 		struct vnode *a_vp;
2622 		int a_flags;
2623 	} */ *ap;
2624 {
2625 	struct vnode *vp, *vq;
2626 	dev_t dev;
2627 
2628 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
2629 	vp = ap->a_vp;
2630 	KASSERT((vp->v_type == VCHR), ("vop_revoke: not VCHR"));
2631 
2632 	VI_LOCK(vp);
2633 	/*
2634 	 * If a vgone (or vclean) is already in progress,
2635 	 * wait until it is done and return.
2636 	 */
2637 	if (vp->v_iflag & VI_XLOCK) {
2638 		vp->v_iflag |= VI_XWANT;
2639 		msleep(vp, VI_MTX(vp), PINOD | PDROP,
2640 		    "vop_revokeall", 0);
2641 		return (0);
2642 	}
2643 	VI_UNLOCK(vp);
2644 	dev = vp->v_rdev;
2645 	for (;;) {
2646 		mtx_lock(&spechash_mtx);
2647 		vq = SLIST_FIRST(&dev->si_hlist);
2648 		mtx_unlock(&spechash_mtx);
2649 		if (!vq)
2650 			break;
2651 		vgone(vq);
2652 	}
2653 	return (0);
2654 }
2655 
2656 /*
2657  * Recycle an unused vnode to the front of the free list.
2658  * Release the passed interlock if the vnode will be recycled.
2659  */
2660 int
2661 vrecycle(vp, inter_lkp, td)
2662 	struct vnode *vp;
2663 	struct mtx *inter_lkp;
2664 	struct thread *td;
2665 {
2666 
2667 	VI_LOCK(vp);
2668 	if (vp->v_usecount == 0) {
2669 		if (inter_lkp) {
2670 			mtx_unlock(inter_lkp);
2671 		}
2672 		vgonel(vp, td);
2673 		return (1);
2674 	}
2675 	VI_UNLOCK(vp);
2676 	return (0);
2677 }
2678 
2679 /*
2680  * Eliminate all activity associated with a vnode
2681  * in preparation for reuse.
2682  */
2683 void
2684 vgone(vp)
2685 	register struct vnode *vp;
2686 {
2687 	struct thread *td = curthread;	/* XXX */
2688 
2689 	VI_LOCK(vp);
2690 	vgonel(vp, td);
2691 }
2692 
2693 /*
2694  * vgone, with the vp interlock held.
2695  */
2696 void
2697 vgonel(vp, td)
2698 	struct vnode *vp;
2699 	struct thread *td;
2700 {
2701 	int s;
2702 
2703 	/*
2704 	 * If a vgone (or vclean) is already in progress,
2705 	 * wait until it is done and return.
2706 	 */
2707 	ASSERT_VI_LOCKED(vp, "vgonel");
2708 	if (vp->v_iflag & VI_XLOCK) {
2709 		vp->v_iflag |= VI_XWANT;
2710 		msleep(vp, VI_MTX(vp), PINOD | PDROP, "vgone", 0);
2711 		return;
2712 	}
2713 
2714 	/*
2715 	 * Clean out the filesystem specific data.
2716 	 */
2717 	vclean(vp, DOCLOSE, td);
2718 	VI_UNLOCK(vp);
2719 
2720 	/*
2721 	 * Delete from old mount point vnode list, if on one.
2722 	 */
2723 	if (vp->v_mount != NULL)
2724 		insmntque(vp, (struct mount *)0);
2725 	/*
2726 	 * If special device, remove it from special device alias list
2727 	 * if it is on one.
2728 	 */
2729 	if (vp->v_type == VCHR && vp->v_rdev != NULL && vp->v_rdev != NODEV) {
2730 		VI_LOCK(vp);
2731 		mtx_lock(&spechash_mtx);
2732 		SLIST_REMOVE(&vp->v_rdev->si_hlist, vp, vnode, v_specnext);
2733 		vp->v_rdev->si_usecount -= vp->v_usecount;
2734 		mtx_unlock(&spechash_mtx);
2735 		VI_UNLOCK(vp);
2736 		vp->v_rdev = NULL;
2737 	}
2738 
2739 	/*
2740 	 * If it is on the freelist and not already at the head,
2741 	 * move it to the head of the list. The test of the
2742 	 * VDOOMED flag and the reference count of zero is because
2743 	 * it will be removed from the free list by getnewvnode,
2744 	 * but will not have its reference count incremented until
2745 	 * after calling vgone. If the reference count were
2746 	 * incremented first, vgone would (incorrectly) try to
2747 	 * close the previous instance of the underlying object.
2748 	 */
2749 	VI_LOCK(vp);
2750 	if (vp->v_usecount == 0 && !(vp->v_iflag & VI_DOOMED)) {
2751 		s = splbio();
2752 		mtx_lock(&vnode_free_list_mtx);
2753 		if (vp->v_iflag & VI_FREE) {
2754 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2755 		} else {
2756 			vp->v_iflag |= VI_FREE;
2757 			freevnodes++;
2758 		}
2759 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2760 		mtx_unlock(&vnode_free_list_mtx);
2761 		splx(s);
2762 	}
2763 
2764 	vp->v_type = VBAD;
2765 	VI_UNLOCK(vp);
2766 }
2767 
2768 /*
2769  * Lookup a vnode by device number.
2770  */
2771 int
2772 vfinddev(dev, type, vpp)
2773 	dev_t dev;
2774 	enum vtype type;
2775 	struct vnode **vpp;
2776 {
2777 	struct vnode *vp;
2778 
2779 	mtx_lock(&spechash_mtx);
2780 	SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2781 		if (type == vp->v_type) {
2782 			*vpp = vp;
2783 			mtx_unlock(&spechash_mtx);
2784 			return (1);
2785 		}
2786 	}
2787 	mtx_unlock(&spechash_mtx);
2788 	return (0);
2789 }
2790 
2791 /*
2792  * Calculate the total number of references to a special device.
2793  */
2794 int
2795 vcount(vp)
2796 	struct vnode *vp;
2797 {
2798 	int count;
2799 
2800 	mtx_lock(&spechash_mtx);
2801 	count = vp->v_rdev->si_usecount;
2802 	mtx_unlock(&spechash_mtx);
2803 	return (count);
2804 }
2805 
2806 /*
2807  * Same as above, but using the dev_t as argument
2808  */
2809 int
2810 count_dev(dev)
2811 	dev_t dev;
2812 {
2813 	struct vnode *vp;
2814 
2815 	vp = SLIST_FIRST(&dev->si_hlist);
2816 	if (vp == NULL)
2817 		return (0);
2818 	return(vcount(vp));
2819 }
2820 
2821 /*
2822  * Print out a description of a vnode.
2823  */
2824 static char *typename[] =
2825 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2826 
2827 void
2828 vprint(label, vp)
2829 	char *label;
2830 	struct vnode *vp;
2831 {
2832 	char buf[96];
2833 
2834 	if (label != NULL)
2835 		printf("%s: %p: ", label, (void *)vp);
2836 	else
2837 		printf("%p: ", (void *)vp);
2838 	printf("tag %s, type %s, usecount %d, writecount %d, refcount %d,",
2839 	    vp->v_tag, typename[vp->v_type], vp->v_usecount,
2840 	    vp->v_writecount, vp->v_holdcnt);
2841 	buf[0] = '\0';
2842 	if (vp->v_vflag & VV_ROOT)
2843 		strcat(buf, "|VV_ROOT");
2844 	if (vp->v_vflag & VV_TEXT)
2845 		strcat(buf, "|VV_TEXT");
2846 	if (vp->v_vflag & VV_SYSTEM)
2847 		strcat(buf, "|VV_SYSTEM");
2848 	if (vp->v_iflag & VI_XLOCK)
2849 		strcat(buf, "|VI_XLOCK");
2850 	if (vp->v_iflag & VI_XWANT)
2851 		strcat(buf, "|VI_XWANT");
2852 	if (vp->v_iflag & VI_BWAIT)
2853 		strcat(buf, "|VI_BWAIT");
2854 	if (vp->v_iflag & VI_DOOMED)
2855 		strcat(buf, "|VI_DOOMED");
2856 	if (vp->v_iflag & VI_FREE)
2857 		strcat(buf, "|VI_FREE");
2858 	if (vp->v_vflag & VV_OBJBUF)
2859 		strcat(buf, "|VV_OBJBUF");
2860 	if (buf[0] != '\0')
2861 		printf(" flags (%s),", &buf[1]);
2862 	lockmgr_printinfo(vp->v_vnlock);
2863 	printf("\n");
2864 	if (vp->v_data != NULL)
2865 		VOP_PRINT(vp);
2866 }
2867 
2868 #ifdef DDB
2869 #include <ddb/ddb.h>
2870 /*
2871  * List all of the locked vnodes in the system.
2872  * Called when debugging the kernel.
2873  */
2874 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2875 {
2876 	struct thread *td = curthread;	/* XXX */
2877 	struct mount *mp, *nmp;
2878 	struct vnode *vp;
2879 
2880 	printf("Locked vnodes\n");
2881 	mtx_lock(&mountlist_mtx);
2882 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2883 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
2884 			nmp = TAILQ_NEXT(mp, mnt_list);
2885 			continue;
2886 		}
2887 		mtx_lock(&mntvnode_mtx);
2888 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2889 			if (VOP_ISLOCKED(vp, NULL))
2890 				vprint(NULL, vp);
2891 		}
2892 		mtx_unlock(&mntvnode_mtx);
2893 		mtx_lock(&mountlist_mtx);
2894 		nmp = TAILQ_NEXT(mp, mnt_list);
2895 		vfs_unbusy(mp, td);
2896 	}
2897 	mtx_unlock(&mountlist_mtx);
2898 }
2899 #endif
2900 
2901 /*
2902  * Fill in a struct xvfsconf based on a struct vfsconf.
2903  */
2904 static void
2905 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2906 {
2907 
2908 	strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2909 	xvfsp->vfc_typenum = vfsp->vfc_typenum;
2910 	xvfsp->vfc_refcount = vfsp->vfc_refcount;
2911 	xvfsp->vfc_flags = vfsp->vfc_flags;
2912 	/*
2913 	 * These are unused in userland, we keep them
2914 	 * to not break binary compatibility.
2915 	 */
2916 	xvfsp->vfc_vfsops = NULL;
2917 	xvfsp->vfc_next = NULL;
2918 }
2919 
2920 static int
2921 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2922 {
2923 	struct vfsconf *vfsp;
2924 	struct xvfsconf *xvfsp;
2925 	int cnt, error, i;
2926 
2927 	cnt = 0;
2928 	for (vfsp = vfsconf; vfsp != NULL; vfsp = vfsp->vfc_next)
2929 		cnt++;
2930 	xvfsp = malloc(sizeof(struct xvfsconf) * cnt, M_TEMP, M_WAITOK);
2931 	/*
2932 	 * Handle the race that we will have here when struct vfsconf
2933 	 * will be locked down by using both cnt and checking vfc_next
2934 	 * against NULL to determine the end of the loop.  The race will
2935 	 * happen because we will have to unlock before calling malloc().
2936 	 * We are protected by Giant for now.
2937 	 */
2938 	i = 0;
2939 	for (vfsp = vfsconf; vfsp != NULL && i < cnt; vfsp = vfsp->vfc_next) {
2940 		vfsconf2x(vfsp, xvfsp + i);
2941 		i++;
2942 	}
2943 	error = SYSCTL_OUT(req, xvfsp, sizeof(struct xvfsconf) * i);
2944 	free(xvfsp, M_TEMP);
2945 	return (error);
2946 }
2947 
2948 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
2949     "S,xvfsconf", "List of all configured filesystems");
2950 
2951 /*
2952  * Top level filesystem related information gathering.
2953  */
2954 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2955 
2956 static int
2957 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2958 {
2959 	int *name = (int *)arg1 - 1;	/* XXX */
2960 	u_int namelen = arg2 + 1;	/* XXX */
2961 	struct vfsconf *vfsp;
2962 	struct xvfsconf xvfsp;
2963 
2964 	printf("WARNING: userland calling deprecated sysctl, "
2965 	    "please rebuild world\n");
2966 
2967 #if 1 || defined(COMPAT_PRELITE2)
2968 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2969 	if (namelen == 1)
2970 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2971 #endif
2972 
2973 	switch (name[1]) {
2974 	case VFS_MAXTYPENUM:
2975 		if (namelen != 2)
2976 			return (ENOTDIR);
2977 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2978 	case VFS_CONF:
2979 		if (namelen != 3)
2980 			return (ENOTDIR);	/* overloaded */
2981 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2982 			if (vfsp->vfc_typenum == name[2])
2983 				break;
2984 		if (vfsp == NULL)
2985 			return (EOPNOTSUPP);
2986 		vfsconf2x(vfsp, &xvfsp);
2987 		return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
2988 	}
2989 	return (EOPNOTSUPP);
2990 }
2991 
2992 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP, vfs_sysctl,
2993 	"Generic filesystem");
2994 
2995 #if 1 || defined(COMPAT_PRELITE2)
2996 
2997 static int
2998 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2999 {
3000 	int error;
3001 	struct vfsconf *vfsp;
3002 	struct ovfsconf ovfs;
3003 
3004 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
3005 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
3006 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
3007 		ovfs.vfc_index = vfsp->vfc_typenum;
3008 		ovfs.vfc_refcount = vfsp->vfc_refcount;
3009 		ovfs.vfc_flags = vfsp->vfc_flags;
3010 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3011 		if (error)
3012 			return error;
3013 	}
3014 	return 0;
3015 }
3016 
3017 #endif /* 1 || COMPAT_PRELITE2 */
3018 
3019 #define KINFO_VNODESLOP		10
3020 #ifdef notyet
3021 /*
3022  * Dump vnode list (via sysctl).
3023  */
3024 /* ARGSUSED */
3025 static int
3026 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3027 {
3028 	struct xvnode *xvn;
3029 	struct thread *td = req->td;
3030 	struct mount *mp;
3031 	struct vnode *vp;
3032 	int error, len, n;
3033 
3034 	/*
3035 	 * Stale numvnodes access is not fatal here.
3036 	 */
3037 	req->lock = 0;
3038 	len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3039 	if (!req->oldptr)
3040 		/* Make an estimate */
3041 		return (SYSCTL_OUT(req, 0, len));
3042 
3043 	sysctl_wire_old_buffer(req, 0);
3044 	xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3045 	n = 0;
3046 	mtx_lock(&mountlist_mtx);
3047 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3048 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td))
3049 			continue;
3050 		mtx_lock(&mntvnode_mtx);
3051 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3052 			if (n == len)
3053 				break;
3054 			vref(vp);
3055 			xvn[n].xv_size = sizeof *xvn;
3056 			xvn[n].xv_vnode = vp;
3057 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3058 			XV_COPY(usecount);
3059 			XV_COPY(writecount);
3060 			XV_COPY(holdcnt);
3061 			XV_COPY(id);
3062 			XV_COPY(mount);
3063 			XV_COPY(numoutput);
3064 			XV_COPY(type);
3065 #undef XV_COPY
3066 			xvn[n].xv_flag = vp->v_vflag;
3067 
3068 			switch (vp->v_type) {
3069 			case VREG:
3070 			case VDIR:
3071 			case VLNK:
3072 				xvn[n].xv_dev = vp->v_cachedfs;
3073 				xvn[n].xv_ino = vp->v_cachedid;
3074 				break;
3075 			case VBLK:
3076 			case VCHR:
3077 				if (vp->v_rdev == NULL) {
3078 					vrele(vp);
3079 					continue;
3080 				}
3081 				xvn[n].xv_dev = dev2udev(vp->v_rdev);
3082 				break;
3083 			case VSOCK:
3084 				xvn[n].xv_socket = vp->v_socket;
3085 				break;
3086 			case VFIFO:
3087 				xvn[n].xv_fifo = vp->v_fifoinfo;
3088 				break;
3089 			case VNON:
3090 			case VBAD:
3091 			default:
3092 				/* shouldn't happen? */
3093 				vrele(vp);
3094 				continue;
3095 			}
3096 			vrele(vp);
3097 			++n;
3098 		}
3099 		mtx_unlock(&mntvnode_mtx);
3100 		mtx_lock(&mountlist_mtx);
3101 		vfs_unbusy(mp, td);
3102 		if (n == len)
3103 			break;
3104 	}
3105 	mtx_unlock(&mountlist_mtx);
3106 
3107 	error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3108 	free(xvn, M_TEMP);
3109 	return (error);
3110 }
3111 
3112 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3113 	0, 0, sysctl_vnode, "S,xvnode", "");
3114 #endif
3115 
3116 /*
3117  * Check to see if a filesystem is mounted on a block device.
3118  */
3119 int
3120 vfs_mountedon(vp)
3121 	struct vnode *vp;
3122 {
3123 
3124 	if (vp->v_rdev->si_mountpoint != NULL)
3125 		return (EBUSY);
3126 	return (0);
3127 }
3128 
3129 /*
3130  * Unmount all filesystems. The list is traversed in reverse order
3131  * of mounting to avoid dependencies.
3132  */
3133 void
3134 vfs_unmountall()
3135 {
3136 	struct mount *mp;
3137 	struct thread *td;
3138 	int error;
3139 
3140 	if (curthread != NULL)
3141 		td = curthread;
3142 	else
3143 		td = FIRST_THREAD_IN_PROC(initproc); /* XXX XXX proc0? */
3144 	/*
3145 	 * Since this only runs when rebooting, it is not interlocked.
3146 	 */
3147 	while(!TAILQ_EMPTY(&mountlist)) {
3148 		mp = TAILQ_LAST(&mountlist, mntlist);
3149 		error = dounmount(mp, MNT_FORCE, td);
3150 		if (error) {
3151 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
3152 			printf("unmount of %s failed (",
3153 			    mp->mnt_stat.f_mntonname);
3154 			if (error == EBUSY)
3155 				printf("BUSY)\n");
3156 			else
3157 				printf("%d)\n", error);
3158 		} else {
3159 			/* The unmount has removed mp from the mountlist */
3160 		}
3161 	}
3162 }
3163 
3164 /*
3165  * perform msync on all vnodes under a mount point
3166  * the mount point must be locked.
3167  */
3168 void
3169 vfs_msync(struct mount *mp, int flags)
3170 {
3171 	struct vnode *vp, *nvp;
3172 	struct vm_object *obj;
3173 	int tries;
3174 
3175 	GIANT_REQUIRED;
3176 
3177 	tries = 5;
3178 	mtx_lock(&mntvnode_mtx);
3179 loop:
3180 	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) {
3181 		if (vp->v_mount != mp) {
3182 			if (--tries > 0)
3183 				goto loop;
3184 			break;
3185 		}
3186 		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
3187 
3188 		VI_LOCK(vp);
3189 		if (vp->v_iflag & VI_XLOCK) {	/* XXX: what if MNT_WAIT? */
3190 			VI_UNLOCK(vp);
3191 			continue;
3192 		}
3193 
3194 		if ((vp->v_iflag & VI_OBJDIRTY) &&
3195 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
3196 			mtx_unlock(&mntvnode_mtx);
3197 			if (!vget(vp,
3198 			    LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3199 			    curthread)) {
3200 				if (vp->v_vflag & VV_NOSYNC) {	/* unlinked */
3201 					vput(vp);
3202 					mtx_lock(&mntvnode_mtx);
3203 					continue;
3204 				}
3205 
3206 				if (VOP_GETVOBJECT(vp, &obj) == 0) {
3207 					vm_object_page_clean(obj, 0, 0,
3208 					    flags == MNT_WAIT ?
3209 					    OBJPC_SYNC : OBJPC_NOSYNC);
3210 				}
3211 				vput(vp);
3212 			}
3213 			mtx_lock(&mntvnode_mtx);
3214 			if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) {
3215 				if (--tries > 0)
3216 					goto loop;
3217 				break;
3218 			}
3219 		} else
3220 			VI_UNLOCK(vp);
3221 	}
3222 	mtx_unlock(&mntvnode_mtx);
3223 }
3224 
3225 /*
3226  * Create the VM object needed for VMIO and mmap support.  This
3227  * is done for all VREG files in the system.  Some filesystems might
3228  * afford the additional metadata buffering capability of the
3229  * VMIO code by making the device node be VMIO mode also.
3230  *
3231  * vp must be locked when vfs_object_create is called.
3232  */
3233 int
3234 vfs_object_create(vp, td, cred)
3235 	struct vnode *vp;
3236 	struct thread *td;
3237 	struct ucred *cred;
3238 {
3239 	GIANT_REQUIRED;
3240 	return (VOP_CREATEVOBJECT(vp, cred, td));
3241 }
3242 
3243 /*
3244  * Mark a vnode as free, putting it up for recycling.
3245  */
3246 void
3247 vfree(vp)
3248 	struct vnode *vp;
3249 {
3250 	int s;
3251 
3252 	ASSERT_VI_LOCKED(vp, "vfree");
3253 	s = splbio();
3254 	mtx_lock(&vnode_free_list_mtx);
3255 	KASSERT((vp->v_iflag & VI_FREE) == 0, ("vnode already free"));
3256 	if (vp->v_iflag & VI_AGE) {
3257 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3258 	} else {
3259 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3260 	}
3261 	freevnodes++;
3262 	mtx_unlock(&vnode_free_list_mtx);
3263 	vp->v_iflag &= ~VI_AGE;
3264 	vp->v_iflag |= VI_FREE;
3265 	splx(s);
3266 }
3267 
3268 /*
3269  * Opposite of vfree() - mark a vnode as in use.
3270  */
3271 void
3272 vbusy(vp)
3273 	struct vnode *vp;
3274 {
3275 	int s;
3276 
3277 	s = splbio();
3278 	ASSERT_VI_LOCKED(vp, "vbusy");
3279 	KASSERT((vp->v_iflag & VI_FREE) != 0, ("vnode not free"));
3280 
3281 	mtx_lock(&vnode_free_list_mtx);
3282 	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3283 	freevnodes--;
3284 	mtx_unlock(&vnode_free_list_mtx);
3285 
3286 	vp->v_iflag &= ~(VI_FREE|VI_AGE);
3287 	splx(s);
3288 }
3289 
3290 /*
3291  * Record a process's interest in events which might happen to
3292  * a vnode.  Because poll uses the historic select-style interface
3293  * internally, this routine serves as both the ``check for any
3294  * pending events'' and the ``record my interest in future events''
3295  * functions.  (These are done together, while the lock is held,
3296  * to avoid race conditions.)
3297  */
3298 int
3299 vn_pollrecord(vp, td, events)
3300 	struct vnode *vp;
3301 	struct thread *td;
3302 	short events;
3303 {
3304 
3305 	if (vp->v_pollinfo == NULL)
3306 		v_addpollinfo(vp);
3307 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3308 	if (vp->v_pollinfo->vpi_revents & events) {
3309 		/*
3310 		 * This leaves events we are not interested
3311 		 * in available for the other process which
3312 		 * which presumably had requested them
3313 		 * (otherwise they would never have been
3314 		 * recorded).
3315 		 */
3316 		events &= vp->v_pollinfo->vpi_revents;
3317 		vp->v_pollinfo->vpi_revents &= ~events;
3318 
3319 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
3320 		return events;
3321 	}
3322 	vp->v_pollinfo->vpi_events |= events;
3323 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3324 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3325 	return 0;
3326 }
3327 
3328 /*
3329  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
3330  * it is possible for us to miss an event due to race conditions, but
3331  * that condition is expected to be rare, so for the moment it is the
3332  * preferred interface.
3333  */
3334 void
3335 vn_pollevent(vp, events)
3336 	struct vnode *vp;
3337 	short events;
3338 {
3339 
3340 	if (vp->v_pollinfo == NULL)
3341 		v_addpollinfo(vp);
3342 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3343 	if (vp->v_pollinfo->vpi_events & events) {
3344 		/*
3345 		 * We clear vpi_events so that we don't
3346 		 * call selwakeup() twice if two events are
3347 		 * posted before the polling process(es) is
3348 		 * awakened.  This also ensures that we take at
3349 		 * most one selwakeup() if the polling process
3350 		 * is no longer interested.  However, it does
3351 		 * mean that only one event can be noticed at
3352 		 * a time.  (Perhaps we should only clear those
3353 		 * event bits which we note?) XXX
3354 		 */
3355 		vp->v_pollinfo->vpi_events = 0;	/* &= ~events ??? */
3356 		vp->v_pollinfo->vpi_revents |= events;
3357 		selwakeup(&vp->v_pollinfo->vpi_selinfo);
3358 	}
3359 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3360 }
3361 
3362 /*
3363  * Wake up anyone polling on vp because it is being revoked.
3364  * This depends on dead_poll() returning POLLHUP for correct
3365  * behavior.
3366  */
3367 void
3368 vn_pollgone(vp)
3369 	struct vnode *vp;
3370 {
3371 
3372 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3373 	VN_KNOTE(vp, NOTE_REVOKE);
3374 	if (vp->v_pollinfo->vpi_events) {
3375 		vp->v_pollinfo->vpi_events = 0;
3376 		selwakeup(&vp->v_pollinfo->vpi_selinfo);
3377 	}
3378 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3379 }
3380 
3381 
3382 
3383 /*
3384  * Routine to create and manage a filesystem syncer vnode.
3385  */
3386 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
3387 static int	sync_fsync(struct  vop_fsync_args *);
3388 static int	sync_inactive(struct  vop_inactive_args *);
3389 static int	sync_reclaim(struct  vop_reclaim_args *);
3390 
3391 static vop_t **sync_vnodeop_p;
3392 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
3393 	{ &vop_default_desc,	(vop_t *) vop_eopnotsupp },
3394 	{ &vop_close_desc,	(vop_t *) sync_close },		/* close */
3395 	{ &vop_fsync_desc,	(vop_t *) sync_fsync },		/* fsync */
3396 	{ &vop_inactive_desc,	(vop_t *) sync_inactive },	/* inactive */
3397 	{ &vop_reclaim_desc,	(vop_t *) sync_reclaim },	/* reclaim */
3398 	{ &vop_lock_desc,	(vop_t *) vop_stdlock },	/* lock */
3399 	{ &vop_unlock_desc,	(vop_t *) vop_stdunlock },	/* unlock */
3400 	{ &vop_islocked_desc,	(vop_t *) vop_stdislocked },	/* islocked */
3401 	{ NULL, NULL }
3402 };
3403 static struct vnodeopv_desc sync_vnodeop_opv_desc =
3404 	{ &sync_vnodeop_p, sync_vnodeop_entries };
3405 
3406 VNODEOP_SET(sync_vnodeop_opv_desc);
3407 
3408 /*
3409  * Create a new filesystem syncer vnode for the specified mount point.
3410  */
3411 int
3412 vfs_allocate_syncvnode(mp)
3413 	struct mount *mp;
3414 {
3415 	struct vnode *vp;
3416 	static long start, incr, next;
3417 	int error;
3418 
3419 	/* Allocate a new vnode */
3420 	if ((error = getnewvnode("syncer", mp, sync_vnodeop_p, &vp)) != 0) {
3421 		mp->mnt_syncer = NULL;
3422 		return (error);
3423 	}
3424 	vp->v_type = VNON;
3425 	/*
3426 	 * Place the vnode onto the syncer worklist. We attempt to
3427 	 * scatter them about on the list so that they will go off
3428 	 * at evenly distributed times even if all the filesystems
3429 	 * are mounted at once.
3430 	 */
3431 	next += incr;
3432 	if (next == 0 || next > syncer_maxdelay) {
3433 		start /= 2;
3434 		incr /= 2;
3435 		if (start == 0) {
3436 			start = syncer_maxdelay / 2;
3437 			incr = syncer_maxdelay;
3438 		}
3439 		next = start;
3440 	}
3441 	VI_LOCK(vp);
3442 	vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
3443 	VI_UNLOCK(vp);
3444 	mp->mnt_syncer = vp;
3445 	return (0);
3446 }
3447 
3448 /*
3449  * Do a lazy sync of the filesystem.
3450  */
3451 static int
3452 sync_fsync(ap)
3453 	struct vop_fsync_args /* {
3454 		struct vnode *a_vp;
3455 		struct ucred *a_cred;
3456 		int a_waitfor;
3457 		struct thread *a_td;
3458 	} */ *ap;
3459 {
3460 	struct vnode *syncvp = ap->a_vp;
3461 	struct mount *mp = syncvp->v_mount;
3462 	struct thread *td = ap->a_td;
3463 	int error, asyncflag;
3464 
3465 	/*
3466 	 * We only need to do something if this is a lazy evaluation.
3467 	 */
3468 	if (ap->a_waitfor != MNT_LAZY)
3469 		return (0);
3470 
3471 	/*
3472 	 * Move ourselves to the back of the sync list.
3473 	 */
3474 	VI_LOCK(syncvp);
3475 	vn_syncer_add_to_worklist(syncvp, syncdelay);
3476 	VI_UNLOCK(syncvp);
3477 
3478 	/*
3479 	 * Walk the list of vnodes pushing all that are dirty and
3480 	 * not already on the sync list.
3481 	 */
3482 	mtx_lock(&mountlist_mtx);
3483 	if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) {
3484 		mtx_unlock(&mountlist_mtx);
3485 		return (0);
3486 	}
3487 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3488 		vfs_unbusy(mp, td);
3489 		return (0);
3490 	}
3491 	asyncflag = mp->mnt_flag & MNT_ASYNC;
3492 	mp->mnt_flag &= ~MNT_ASYNC;
3493 	vfs_msync(mp, MNT_NOWAIT);
3494 	error = VFS_SYNC(mp, MNT_LAZY, ap->a_cred, td);
3495 	if (asyncflag)
3496 		mp->mnt_flag |= MNT_ASYNC;
3497 	vn_finished_write(mp);
3498 	vfs_unbusy(mp, td);
3499 	return (error);
3500 }
3501 
3502 /*
3503  * The syncer vnode is no referenced.
3504  */
3505 static int
3506 sync_inactive(ap)
3507 	struct vop_inactive_args /* {
3508 		struct vnode *a_vp;
3509 		struct thread *a_td;
3510 	} */ *ap;
3511 {
3512 
3513 	VOP_UNLOCK(ap->a_vp, 0, ap->a_td);
3514 	vgone(ap->a_vp);
3515 	return (0);
3516 }
3517 
3518 /*
3519  * The syncer vnode is no longer needed and is being decommissioned.
3520  *
3521  * Modifications to the worklist must be protected at splbio().
3522  */
3523 static int
3524 sync_reclaim(ap)
3525 	struct vop_reclaim_args /* {
3526 		struct vnode *a_vp;
3527 	} */ *ap;
3528 {
3529 	struct vnode *vp = ap->a_vp;
3530 	int s;
3531 
3532 	s = splbio();
3533 	vp->v_mount->mnt_syncer = NULL;
3534 	VI_LOCK(vp);
3535 	if (vp->v_iflag & VI_ONWORKLST) {
3536 		mtx_lock(&sync_mtx);
3537 		LIST_REMOVE(vp, v_synclist);
3538 		mtx_unlock(&sync_mtx);
3539 		vp->v_iflag &= ~VI_ONWORKLST;
3540 	}
3541 	VI_UNLOCK(vp);
3542 	splx(s);
3543 
3544 	return (0);
3545 }
3546 
3547 /*
3548  * extract the dev_t from a VCHR
3549  */
3550 dev_t
3551 vn_todev(vp)
3552 	struct vnode *vp;
3553 {
3554 	if (vp->v_type != VCHR)
3555 		return (NODEV);
3556 	return (vp->v_rdev);
3557 }
3558 
3559 /*
3560  * Check if vnode represents a disk device
3561  */
3562 int
3563 vn_isdisk(vp, errp)
3564 	struct vnode *vp;
3565 	int *errp;
3566 {
3567 	struct cdevsw *cdevsw;
3568 
3569 	if (vp->v_type != VCHR) {
3570 		if (errp != NULL)
3571 			*errp = ENOTBLK;
3572 		return (0);
3573 	}
3574 	if (vp->v_rdev == NULL) {
3575 		if (errp != NULL)
3576 			*errp = ENXIO;
3577 		return (0);
3578 	}
3579 	cdevsw = devsw(vp->v_rdev);
3580 	if (cdevsw == NULL) {
3581 		if (errp != NULL)
3582 			*errp = ENXIO;
3583 		return (0);
3584 	}
3585 	if (!(cdevsw->d_flags & D_DISK)) {
3586 		if (errp != NULL)
3587 			*errp = ENOTBLK;
3588 		return (0);
3589 	}
3590 	if (errp != NULL)
3591 		*errp = 0;
3592 	return (1);
3593 }
3594 
3595 /*
3596  * Free data allocated by namei(); see namei(9) for details.
3597  */
3598 void
3599 NDFREE(ndp, flags)
3600      struct nameidata *ndp;
3601      const uint flags;
3602 {
3603 	if (!(flags & NDF_NO_FREE_PNBUF) &&
3604 	    (ndp->ni_cnd.cn_flags & HASBUF)) {
3605 		uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
3606 		ndp->ni_cnd.cn_flags &= ~HASBUF;
3607 	}
3608 	if (!(flags & NDF_NO_DVP_UNLOCK) &&
3609 	    (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
3610 	    ndp->ni_dvp != ndp->ni_vp)
3611 		VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_thread);
3612 	if (!(flags & NDF_NO_DVP_RELE) &&
3613 	    (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
3614 		vrele(ndp->ni_dvp);
3615 		ndp->ni_dvp = NULL;
3616 	}
3617 	if (!(flags & NDF_NO_VP_UNLOCK) &&
3618 	    (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
3619 		VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_thread);
3620 	if (!(flags & NDF_NO_VP_RELE) &&
3621 	    ndp->ni_vp) {
3622 		vrele(ndp->ni_vp);
3623 		ndp->ni_vp = NULL;
3624 	}
3625 	if (!(flags & NDF_NO_STARTDIR_RELE) &&
3626 	    (ndp->ni_cnd.cn_flags & SAVESTART)) {
3627 		vrele(ndp->ni_startdir);
3628 		ndp->ni_startdir = NULL;
3629 	}
3630 }
3631 
3632 /*
3633  * Common filesystem object access control check routine.  Accepts a
3634  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3635  * and optional call-by-reference privused argument allowing vaccess()
3636  * to indicate to the caller whether privilege was used to satisfy the
3637  * request (obsoleted).  Returns 0 on success, or an errno on failure.
3638  */
3639 int
3640 vaccess(type, file_mode, file_uid, file_gid, acc_mode, cred, privused)
3641 	enum vtype type;
3642 	mode_t file_mode;
3643 	uid_t file_uid;
3644 	gid_t file_gid;
3645 	mode_t acc_mode;
3646 	struct ucred *cred;
3647 	int *privused;
3648 {
3649 	mode_t dac_granted;
3650 #ifdef CAPABILITIES
3651 	mode_t cap_granted;
3652 #endif
3653 
3654 	/*
3655 	 * Look for a normal, non-privileged way to access the file/directory
3656 	 * as requested.  If it exists, go with that.
3657 	 */
3658 
3659 	if (privused != NULL)
3660 		*privused = 0;
3661 
3662 	dac_granted = 0;
3663 
3664 	/* Check the owner. */
3665 	if (cred->cr_uid == file_uid) {
3666 		dac_granted |= VADMIN;
3667 		if (file_mode & S_IXUSR)
3668 			dac_granted |= VEXEC;
3669 		if (file_mode & S_IRUSR)
3670 			dac_granted |= VREAD;
3671 		if (file_mode & S_IWUSR)
3672 			dac_granted |= (VWRITE | VAPPEND);
3673 
3674 		if ((acc_mode & dac_granted) == acc_mode)
3675 			return (0);
3676 
3677 		goto privcheck;
3678 	}
3679 
3680 	/* Otherwise, check the groups (first match) */
3681 	if (groupmember(file_gid, cred)) {
3682 		if (file_mode & S_IXGRP)
3683 			dac_granted |= VEXEC;
3684 		if (file_mode & S_IRGRP)
3685 			dac_granted |= VREAD;
3686 		if (file_mode & S_IWGRP)
3687 			dac_granted |= (VWRITE | VAPPEND);
3688 
3689 		if ((acc_mode & dac_granted) == acc_mode)
3690 			return (0);
3691 
3692 		goto privcheck;
3693 	}
3694 
3695 	/* Otherwise, check everyone else. */
3696 	if (file_mode & S_IXOTH)
3697 		dac_granted |= VEXEC;
3698 	if (file_mode & S_IROTH)
3699 		dac_granted |= VREAD;
3700 	if (file_mode & S_IWOTH)
3701 		dac_granted |= (VWRITE | VAPPEND);
3702 	if ((acc_mode & dac_granted) == acc_mode)
3703 		return (0);
3704 
3705 privcheck:
3706 	if (!suser_cred(cred, PRISON_ROOT)) {
3707 		/* XXX audit: privilege used */
3708 		if (privused != NULL)
3709 			*privused = 1;
3710 		return (0);
3711 	}
3712 
3713 #ifdef CAPABILITIES
3714 	/*
3715 	 * Build a capability mask to determine if the set of capabilities
3716 	 * satisfies the requirements when combined with the granted mask
3717 	 * from above.
3718 	 * For each capability, if the capability is required, bitwise
3719 	 * or the request type onto the cap_granted mask.
3720 	 */
3721 	cap_granted = 0;
3722 
3723 	if (type == VDIR) {
3724 		/*
3725 		 * For directories, use CAP_DAC_READ_SEARCH to satisfy
3726 		 * VEXEC requests, instead of CAP_DAC_EXECUTE.
3727 		 */
3728 		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3729 		    !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3730 			cap_granted |= VEXEC;
3731 	} else {
3732 		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3733 		    !cap_check(cred, NULL, CAP_DAC_EXECUTE, PRISON_ROOT))
3734 			cap_granted |= VEXEC;
3735 	}
3736 
3737 	if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
3738 	    !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3739 		cap_granted |= VREAD;
3740 
3741 	if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3742 	    !cap_check(cred, NULL, CAP_DAC_WRITE, PRISON_ROOT))
3743 		cap_granted |= (VWRITE | VAPPEND);
3744 
3745 	if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3746 	    !cap_check(cred, NULL, CAP_FOWNER, PRISON_ROOT))
3747 		cap_granted |= VADMIN;
3748 
3749 	if ((acc_mode & (cap_granted | dac_granted)) == acc_mode) {
3750 		/* XXX audit: privilege used */
3751 		if (privused != NULL)
3752 			*privused = 1;
3753 		return (0);
3754 	}
3755 #endif
3756 
3757 	return ((acc_mode & VADMIN) ? EPERM : EACCES);
3758 }
3759 
3760 /*
3761  * Credential check based on process requesting service, and per-attribute
3762  * permissions.
3763  */
3764 int
3765 extattr_check_cred(struct vnode *vp, int attrnamespace,
3766     struct ucred *cred, struct thread *td, int access)
3767 {
3768 
3769 	/*
3770 	 * Kernel-invoked always succeeds.
3771 	 */
3772 	if (cred == NOCRED)
3773 		return (0);
3774 
3775 	/*
3776 	 * Do not allow privileged processes in jail to directly
3777 	 * manipulate system attributes.
3778 	 *
3779 	 * XXX What capability should apply here?
3780 	 * Probably CAP_SYS_SETFFLAG.
3781 	 */
3782 	switch (attrnamespace) {
3783 	case EXTATTR_NAMESPACE_SYSTEM:
3784 		/* Potentially should be: return (EPERM); */
3785 		return (suser_cred(cred, 0));
3786 	case EXTATTR_NAMESPACE_USER:
3787 		return (VOP_ACCESS(vp, access, cred, td));
3788 	default:
3789 		return (EPERM);
3790 	}
3791 }
3792