xref: /freebsd/sys/kern/vfs_subr.c (revision 729362425c09cf6b362366aabc6fb547eee8035a)
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)
678 {
679 	struct vnode *vp;
680 	int done;
681 	int trigger;
682 	int usevnodes;
683 	int count;
684 
685 	/*
686 	 * Calculate the trigger point, don't allow user
687 	 * screwups to blow us up.   This prevents us from
688 	 * recycling vnodes with lots of resident pages.  We
689 	 * aren't trying to free memory, we are trying to
690 	 * free vnodes.
691 	 */
692 	usevnodes = desiredvnodes;
693 	if (usevnodes <= 0)
694 		usevnodes = 1;
695 	trigger = cnt.v_page_count * 2 / usevnodes;
696 
697 	done = 0;
698 	mtx_lock(&mntvnode_mtx);
699 	count = mp->mnt_nvnodelistsize / 10 + 1;
700 	while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
701 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
702 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
703 
704 		if (vp->v_type != VNON &&
705 		    vp->v_type != VBAD &&
706 		    VI_TRYLOCK(vp)) {
707 			if (VMIGHTFREE(vp) &&           /* critical path opt */
708 			    (vp->v_object == NULL ||
709 			    vp->v_object->resident_page_count < trigger)) {
710 				mtx_unlock(&mntvnode_mtx);
711 				vgonel(vp, curthread);
712 				done++;
713 				mtx_lock(&mntvnode_mtx);
714 			} else
715 				VI_UNLOCK(vp);
716 		}
717 		--count;
718 	}
719 	mtx_unlock(&mntvnode_mtx);
720 	return done;
721 }
722 
723 /*
724  * Attempt to recycle vnodes in a context that is always safe to block.
725  * Calling vlrurecycle() from the bowels of filesystem code has some
726  * interesting deadlock problems.
727  */
728 static struct proc *vnlruproc;
729 static int vnlruproc_sig;
730 
731 static void
732 vnlru_proc(void)
733 {
734 	struct mount *mp, *nmp;
735 	int s;
736 	int done;
737 	struct proc *p = vnlruproc;
738 	struct thread *td = FIRST_THREAD_IN_PROC(p);	/* XXXKSE */
739 
740 	mtx_lock(&Giant);
741 
742 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
743 	    SHUTDOWN_PRI_FIRST);
744 
745 	s = splbio();
746 	for (;;) {
747 		kthread_suspend_check(p);
748 		mtx_lock(&vnode_free_list_mtx);
749 		if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
750 			mtx_unlock(&vnode_free_list_mtx);
751 			vnlruproc_sig = 0;
752 			wakeup(&vnlruproc_sig);
753 			tsleep(vnlruproc, PVFS, "vlruwt", hz);
754 			continue;
755 		}
756 		mtx_unlock(&vnode_free_list_mtx);
757 		done = 0;
758 		mtx_lock(&mountlist_mtx);
759 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
760 			if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
761 				nmp = TAILQ_NEXT(mp, mnt_list);
762 				continue;
763 			}
764 			done += vlrureclaim(mp);
765 			mtx_lock(&mountlist_mtx);
766 			nmp = TAILQ_NEXT(mp, mnt_list);
767 			vfs_unbusy(mp, td);
768 		}
769 		mtx_unlock(&mountlist_mtx);
770 		if (done == 0) {
771 #if 0
772 			/* These messages are temporary debugging aids */
773 			if (vnlru_nowhere < 5)
774 				printf("vnlru process getting nowhere..\n");
775 			else if (vnlru_nowhere == 5)
776 				printf("vnlru process messages stopped.\n");
777 #endif
778 			vnlru_nowhere++;
779 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
780 		}
781 	}
782 	splx(s);
783 }
784 
785 static struct kproc_desc vnlru_kp = {
786 	"vnlru",
787 	vnlru_proc,
788 	&vnlruproc
789 };
790 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
791 
792 
793 /*
794  * Routines having to do with the management of the vnode table.
795  */
796 
797 /*
798  * Check to see if a free vnode can be recycled. If it can,
799  * return it locked with the vn lock, but not interlock. Also
800  * get the vn_start_write lock. Otherwise indicate the error.
801  */
802 static int
803 vcanrecycle(struct vnode *vp, struct mount **vnmpp)
804 {
805 	struct thread *td = curthread;
806 	vm_object_t object;
807 	int error;
808 
809 	/* Don't recycle if we can't get the interlock */
810 	if (!VI_TRYLOCK(vp))
811 		return (EWOULDBLOCK);
812 
813 	/* We should be able to immediately acquire this */
814 	/* XXX This looks like it should panic if it fails */
815 	if (vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE, td) != 0) {
816 		if (VOP_ISLOCKED(vp, td))
817 			panic("vcanrecycle: locked vnode");
818 		return (EWOULDBLOCK);
819 	}
820 
821 	/*
822 	 * Don't recycle if its filesystem is being suspended.
823 	 */
824 	if (vn_start_write(vp, vnmpp, V_NOWAIT) != 0) {
825 		error = EBUSY;
826 		goto done;
827 	}
828 
829 	/*
830 	 * Don't recycle if we still have cached pages.
831 	 */
832 	if (VOP_GETVOBJECT(vp, &object) == 0 &&
833 	     (object->resident_page_count ||
834 	      object->ref_count)) {
835 		error = EBUSY;
836 		goto done;
837 	}
838 	if (LIST_FIRST(&vp->v_cache_src)) {
839 		/*
840 		 * note: nameileafonly sysctl is temporary,
841 		 * for debugging only, and will eventually be
842 		 * removed.
843 		 */
844 		if (nameileafonly > 0) {
845 			/*
846 			 * Do not reuse namei-cached directory
847 			 * vnodes that have cached
848 			 * subdirectories.
849 			 */
850 			if (cache_leaf_test(vp) < 0) {
851 				error = EISDIR;
852 				goto done;
853 			}
854 		} else if (nameileafonly < 0 ||
855 			    vmiodirenable == 0) {
856 			/*
857 			 * Do not reuse namei-cached directory
858 			 * vnodes if nameileafonly is -1 or
859 			 * if VMIO backing for directories is
860 			 * turned off (otherwise we reuse them
861 			 * too quickly).
862 			 */
863 			error = EBUSY;
864 			goto done;
865 		}
866 	}
867 	return (0);
868 done:
869 	VOP_UNLOCK(vp, 0, td);
870 	return (error);
871 }
872 
873 /*
874  * Return the next vnode from the free list.
875  */
876 int
877 getnewvnode(tag, mp, vops, vpp)
878 	const char *tag;
879 	struct mount *mp;
880 	vop_t **vops;
881 	struct vnode **vpp;
882 {
883 	int s;
884 	struct thread *td = curthread;	/* XXX */
885 	struct vnode *vp = NULL;
886 	struct vpollinfo *pollinfo = NULL;
887 	struct mount *vnmp;
888 
889 	s = splbio();
890 	mtx_lock(&vnode_free_list_mtx);
891 
892 	/*
893 	 * Try to reuse vnodes if we hit the max.  This situation only
894 	 * occurs in certain large-memory (2G+) situations.  We cannot
895 	 * attempt to directly reclaim vnodes due to nasty recursion
896 	 * problems.
897 	 */
898 	while (numvnodes - freevnodes > desiredvnodes) {
899 		if (vnlruproc_sig == 0) {
900 			vnlruproc_sig = 1;      /* avoid unnecessary wakeups */
901 			wakeup(vnlruproc);
902 		}
903 		mtx_unlock(&vnode_free_list_mtx);
904 		tsleep(&vnlruproc_sig, PVFS, "vlruwk", hz);
905 		mtx_lock(&vnode_free_list_mtx);
906 	}
907 
908 	/*
909 	 * Attempt to reuse a vnode already on the free list, allocating
910 	 * a new vnode if we can't find one or if we have not reached a
911 	 * good minimum for good LRU performance.
912 	 */
913 
914 	if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
915 		int error;
916 		int count;
917 
918 		for (count = 0; count < freevnodes; count++) {
919 			vp = TAILQ_FIRST(&vnode_free_list);
920 
921 			KASSERT(vp->v_usecount == 0 &&
922 			    (vp->v_iflag & VI_DOINGINACT) == 0,
923 			    ("getnewvnode: free vnode isn't"));
924 
925 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
926 			/*
927 			 * We have to drop the free list mtx to avoid lock
928 			 * order reversals with interlock.
929 			 */
930 			mtx_unlock(&vnode_free_list_mtx);
931 			error = vcanrecycle(vp, &vnmp);
932 			mtx_lock(&vnode_free_list_mtx);
933 			if (error == 0)
934 				break;
935 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
936 			vp = NULL;
937 		}
938 	}
939 	if (vp) {
940 		freevnodes--;
941 		mtx_unlock(&vnode_free_list_mtx);
942 
943 		cache_purge(vp);
944 		VI_LOCK(vp);
945 		vp->v_iflag |= VI_DOOMED;
946 		vp->v_iflag &= ~VI_FREE;
947 		if (vp->v_type != VBAD) {
948 			VOP_UNLOCK(vp, 0, td);
949 			vgonel(vp, td);
950 			VI_LOCK(vp);
951 		} else {
952 			VOP_UNLOCK(vp, 0, td);
953 		}
954 		vn_finished_write(vnmp);
955 
956 #ifdef INVARIANTS
957 		{
958 			if (vp->v_data)
959 				panic("cleaned vnode isn't");
960 			if (vp->v_numoutput)
961 				panic("Clean vnode has pending I/O's");
962 			if (vp->v_writecount != 0)
963 				panic("Non-zero write count");
964 		}
965 #endif
966 		if ((pollinfo = vp->v_pollinfo) != NULL) {
967 			/*
968 			 * To avoid lock order reversals, the call to
969 			 * uma_zfree() must be delayed until the vnode
970 			 * interlock is released.
971 			 */
972 			vp->v_pollinfo = NULL;
973 		}
974 #ifdef MAC
975 		mac_destroy_vnode(vp);
976 #endif
977 		vp->v_iflag = 0;
978 		vp->v_vflag = 0;
979 		vp->v_lastw = 0;
980 		vp->v_lasta = 0;
981 		vp->v_cstart = 0;
982 		vp->v_clen = 0;
983 		vp->v_socket = 0;
984 		lockdestroy(vp->v_vnlock);
985 		lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOPAUSE);
986 		KASSERT(vp->v_cleanbufcnt == 0, ("cleanbufcnt not 0"));
987 		KASSERT(vp->v_cleanblkroot == NULL, ("cleanblkroot not NULL"));
988 		KASSERT(vp->v_dirtybufcnt == 0, ("dirtybufcnt not 0"));
989 		KASSERT(vp->v_dirtyblkroot == NULL, ("dirtyblkroot not NULL"));
990 	} else {
991 		numvnodes++;
992 		mtx_unlock(&vnode_free_list_mtx);
993 
994 		vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
995 		mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
996 		VI_LOCK(vp);
997 		vp->v_dd = vp;
998 		vp->v_vnlock = &vp->v_lock;
999 		lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOPAUSE);
1000 		cache_purge(vp);
1001 		LIST_INIT(&vp->v_cache_src);
1002 		TAILQ_INIT(&vp->v_cache_dst);
1003 	}
1004 
1005 	TAILQ_INIT(&vp->v_cleanblkhd);
1006 	TAILQ_INIT(&vp->v_dirtyblkhd);
1007 	vp->v_type = VNON;
1008 	vp->v_tag = tag;
1009 	vp->v_op = vops;
1010 	*vpp = vp;
1011 	vp->v_usecount = 1;
1012 	vp->v_data = 0;
1013 	vp->v_cachedid = -1;
1014 	VI_UNLOCK(vp);
1015 	if (pollinfo != NULL) {
1016 		mtx_destroy(&pollinfo->vpi_lock);
1017 		uma_zfree(vnodepoll_zone, pollinfo);
1018 	}
1019 #ifdef MAC
1020 	mac_init_vnode(vp);
1021 	if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1022 		mac_associate_vnode_singlelabel(mp, vp);
1023 #endif
1024 	insmntque(vp, mp);
1025 
1026 	return (0);
1027 }
1028 
1029 /*
1030  * Move a vnode from one mount queue to another.
1031  */
1032 static void
1033 insmntque(vp, mp)
1034 	register struct vnode *vp;
1035 	register struct mount *mp;
1036 {
1037 
1038 	mtx_lock(&mntvnode_mtx);
1039 	/*
1040 	 * Delete from old mount point vnode list, if on one.
1041 	 */
1042 	if (vp->v_mount != NULL) {
1043 		KASSERT(vp->v_mount->mnt_nvnodelistsize > 0,
1044 			("bad mount point vnode list size"));
1045 		TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
1046 		vp->v_mount->mnt_nvnodelistsize--;
1047 	}
1048 	/*
1049 	 * Insert into list of vnodes for the new mount point, if available.
1050 	 */
1051 	if ((vp->v_mount = mp) == NULL) {
1052 		mtx_unlock(&mntvnode_mtx);
1053 		return;
1054 	}
1055 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1056 	mp->mnt_nvnodelistsize++;
1057 	mtx_unlock(&mntvnode_mtx);
1058 }
1059 
1060 /*
1061  * Update outstanding I/O count and do wakeup if requested.
1062  */
1063 void
1064 vwakeup(bp)
1065 	register struct buf *bp;
1066 {
1067 	register struct vnode *vp;
1068 
1069 	bp->b_flags &= ~B_WRITEINPROG;
1070 	if ((vp = bp->b_vp)) {
1071 		VI_LOCK(vp);
1072 		vp->v_numoutput--;
1073 		if (vp->v_numoutput < 0)
1074 			panic("vwakeup: neg numoutput");
1075 		if ((vp->v_numoutput == 0) && (vp->v_iflag & VI_BWAIT)) {
1076 			vp->v_iflag &= ~VI_BWAIT;
1077 			wakeup(&vp->v_numoutput);
1078 		}
1079 		VI_UNLOCK(vp);
1080 	}
1081 }
1082 
1083 /*
1084  * Flush out and invalidate all buffers associated with a vnode.
1085  * Called with the underlying object locked.
1086  */
1087 int
1088 vinvalbuf(vp, flags, cred, td, slpflag, slptimeo)
1089 	struct vnode *vp;
1090 	int flags;
1091 	struct ucred *cred;
1092 	struct thread *td;
1093 	int slpflag, slptimeo;
1094 {
1095 	struct buf *blist;
1096 	int s, error;
1097 	vm_object_t object;
1098 
1099 	GIANT_REQUIRED;
1100 
1101 	ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1102 
1103 	VI_LOCK(vp);
1104 	if (flags & V_SAVE) {
1105 		s = splbio();
1106 		while (vp->v_numoutput) {
1107 			vp->v_iflag |= VI_BWAIT;
1108 			error = msleep(&vp->v_numoutput, VI_MTX(vp),
1109 			    slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo);
1110 			if (error) {
1111 				VI_UNLOCK(vp);
1112 				splx(s);
1113 				return (error);
1114 			}
1115 		}
1116 		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1117 			splx(s);
1118 			VI_UNLOCK(vp);
1119 			if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, td)) != 0)
1120 				return (error);
1121 			/*
1122 			 * XXX We could save a lock/unlock if this was only
1123 			 * enabled under INVARIANTS
1124 			 */
1125 			VI_LOCK(vp);
1126 			s = splbio();
1127 			if (vp->v_numoutput > 0 ||
1128 			    !TAILQ_EMPTY(&vp->v_dirtyblkhd))
1129 				panic("vinvalbuf: dirty bufs");
1130 		}
1131 		splx(s);
1132 	}
1133 	s = splbio();
1134 	/*
1135 	 * If you alter this loop please notice that interlock is dropped and
1136 	 * reacquired in flushbuflist.  Special care is needed to ensure that
1137 	 * no race conditions occur from this.
1138 	 */
1139 	for (error = 0;;) {
1140 		if ((blist = TAILQ_FIRST(&vp->v_cleanblkhd)) != 0 &&
1141 		    flushbuflist(blist, flags, vp, slpflag, slptimeo, &error)) {
1142 			if (error)
1143 				break;
1144 			continue;
1145 		}
1146 		if ((blist = TAILQ_FIRST(&vp->v_dirtyblkhd)) != 0 &&
1147 		    flushbuflist(blist, flags, vp, slpflag, slptimeo, &error)) {
1148 			if (error)
1149 				break;
1150 			continue;
1151 		}
1152 		break;
1153 	}
1154 	if (error) {
1155 		splx(s);
1156 		VI_UNLOCK(vp);
1157 		return (error);
1158 	}
1159 
1160 	/*
1161 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1162 	 * have write I/O in-progress but if there is a VM object then the
1163 	 * VM object can also have read-I/O in-progress.
1164 	 */
1165 	do {
1166 		while (vp->v_numoutput > 0) {
1167 			vp->v_iflag |= VI_BWAIT;
1168 			msleep(&vp->v_numoutput, VI_MTX(vp), PVM, "vnvlbv", 0);
1169 		}
1170 		VI_UNLOCK(vp);
1171 		if (VOP_GETVOBJECT(vp, &object) == 0) {
1172 			while (object->paging_in_progress)
1173 			vm_object_pip_sleep(object, "vnvlbx");
1174 		}
1175 		VI_LOCK(vp);
1176 	} while (vp->v_numoutput > 0);
1177 	VI_UNLOCK(vp);
1178 
1179 	splx(s);
1180 
1181 	/*
1182 	 * Destroy the copy in the VM cache, too.
1183 	 */
1184 	if (VOP_GETVOBJECT(vp, &object) == 0) {
1185 		vm_object_lock(object);
1186 		vm_object_page_remove(object, 0, 0,
1187 			(flags & V_SAVE) ? TRUE : FALSE);
1188 		vm_object_unlock(object);
1189 	}
1190 
1191 #ifdef INVARIANTS
1192 	VI_LOCK(vp);
1193 	if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1194 	    (!TAILQ_EMPTY(&vp->v_dirtyblkhd) ||
1195 	     !TAILQ_EMPTY(&vp->v_cleanblkhd)))
1196 		panic("vinvalbuf: flush failed");
1197 	VI_UNLOCK(vp);
1198 #endif
1199 	return (0);
1200 }
1201 
1202 /*
1203  * Flush out buffers on the specified list.
1204  *
1205  */
1206 static int
1207 flushbuflist(blist, flags, vp, slpflag, slptimeo, errorp)
1208 	struct buf *blist;
1209 	int flags;
1210 	struct vnode *vp;
1211 	int slpflag, slptimeo;
1212 	int *errorp;
1213 {
1214 	struct buf *bp, *nbp;
1215 	int found, error;
1216 
1217 	ASSERT_VI_LOCKED(vp, "flushbuflist");
1218 
1219 	for (found = 0, bp = blist; bp; bp = nbp) {
1220 		nbp = TAILQ_NEXT(bp, b_vnbufs);
1221 		if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1222 		    ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1223 			continue;
1224 		}
1225 		found += 1;
1226 		error = BUF_TIMELOCK(bp,
1227 		    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, VI_MTX(vp),
1228 		    "flushbuf", slpflag, slptimeo);
1229 		if (error) {
1230 			if (error != ENOLCK)
1231 				*errorp = error;
1232 			goto done;
1233 		}
1234 		/*
1235 		 * XXX Since there are no node locks for NFS, I
1236 		 * believe there is a slight chance that a delayed
1237 		 * write will occur while sleeping just above, so
1238 		 * check for it.  Note that vfs_bio_awrite expects
1239 		 * buffers to reside on a queue, while BUF_WRITE and
1240 		 * brelse do not.
1241 		 */
1242 		if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1243 			(flags & V_SAVE)) {
1244 
1245 			if (bp->b_vp == vp) {
1246 				if (bp->b_flags & B_CLUSTEROK) {
1247 					vfs_bio_awrite(bp);
1248 				} else {
1249 					bremfree(bp);
1250 					bp->b_flags |= B_ASYNC;
1251 					BUF_WRITE(bp);
1252 				}
1253 			} else {
1254 				bremfree(bp);
1255 				(void) BUF_WRITE(bp);
1256 			}
1257 			goto done;
1258 		}
1259 		bremfree(bp);
1260 		bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
1261 		bp->b_flags &= ~B_ASYNC;
1262 		brelse(bp);
1263 		VI_LOCK(vp);
1264 	}
1265 	return (found);
1266 done:
1267 	VI_LOCK(vp);
1268 	return (found);
1269 }
1270 
1271 /*
1272  * Truncate a file's buffer and pages to a specified length.  This
1273  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1274  * sync activity.
1275  */
1276 int
1277 vtruncbuf(vp, cred, td, length, blksize)
1278 	register struct vnode *vp;
1279 	struct ucred *cred;
1280 	struct thread *td;
1281 	off_t length;
1282 	int blksize;
1283 {
1284 	register struct buf *bp;
1285 	struct buf *nbp;
1286 	int s, anyfreed;
1287 	int trunclbn;
1288 
1289 	/*
1290 	 * Round up to the *next* lbn.
1291 	 */
1292 	trunclbn = (length + blksize - 1) / blksize;
1293 
1294 	s = splbio();
1295 	ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1296 restart:
1297 	VI_LOCK(vp);
1298 	anyfreed = 1;
1299 	for (;anyfreed;) {
1300 		anyfreed = 0;
1301 		for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
1302 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1303 			if (bp->b_lblkno >= trunclbn) {
1304 				if (BUF_LOCK(bp,
1305 				    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1306 				    VI_MTX(vp)) == ENOLCK)
1307 					goto restart;
1308 
1309 				bremfree(bp);
1310 				bp->b_flags |= (B_INVAL | B_RELBUF);
1311 				bp->b_flags &= ~B_ASYNC;
1312 				brelse(bp);
1313 				anyfreed = 1;
1314 
1315 				if (nbp &&
1316 				    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1317 				    (nbp->b_vp != vp) ||
1318 				    (nbp->b_flags & B_DELWRI))) {
1319 					goto restart;
1320 				}
1321 				VI_LOCK(vp);
1322 			}
1323 		}
1324 
1325 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1326 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1327 			if (bp->b_lblkno >= trunclbn) {
1328 				if (BUF_LOCK(bp,
1329 				    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1330 				    VI_MTX(vp)) == ENOLCK)
1331 					goto restart;
1332 				bremfree(bp);
1333 				bp->b_flags |= (B_INVAL | B_RELBUF);
1334 				bp->b_flags &= ~B_ASYNC;
1335 				brelse(bp);
1336 				anyfreed = 1;
1337 				if (nbp &&
1338 				    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1339 				    (nbp->b_vp != vp) ||
1340 				    (nbp->b_flags & B_DELWRI) == 0)) {
1341 					goto restart;
1342 				}
1343 				VI_LOCK(vp);
1344 			}
1345 		}
1346 	}
1347 
1348 	if (length > 0) {
1349 restartsync:
1350 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1351 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1352 			if (bp->b_lblkno > 0)
1353 				continue;
1354 			/*
1355 			 * Since we hold the vnode lock this should only
1356 			 * fail if we're racing with the buf daemon.
1357 			 */
1358 			if (BUF_LOCK(bp,
1359 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1360 			    VI_MTX(vp)) == ENOLCK) {
1361 				goto restart;
1362 			}
1363 			KASSERT((bp->b_flags & B_DELWRI),
1364 			    ("buf(%p) on dirty queue without DELWRI.", bp));
1365 
1366 			bremfree(bp);
1367 			bawrite(bp);
1368 			VI_LOCK(vp);
1369 			goto restartsync;
1370 		}
1371 	}
1372 
1373 	while (vp->v_numoutput > 0) {
1374 		vp->v_iflag |= VI_BWAIT;
1375 		msleep(&vp->v_numoutput, VI_MTX(vp), PVM, "vbtrunc", 0);
1376 	}
1377 	VI_UNLOCK(vp);
1378 	splx(s);
1379 
1380 	vnode_pager_setsize(vp, length);
1381 
1382 	return (0);
1383 }
1384 
1385 /*
1386  * buf_splay() - splay tree core for the clean/dirty list of buffers in
1387  * 		 a vnode.
1388  *
1389  *	NOTE: We have to deal with the special case of a background bitmap
1390  *	buffer, a situation where two buffers will have the same logical
1391  *	block offset.  We want (1) only the foreground buffer to be accessed
1392  *	in a lookup and (2) must differentiate between the foreground and
1393  *	background buffer in the splay tree algorithm because the splay
1394  *	tree cannot normally handle multiple entities with the same 'index'.
1395  *	We accomplish this by adding differentiating flags to the splay tree's
1396  *	numerical domain.
1397  */
1398 static
1399 struct buf *
1400 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1401 {
1402 	struct buf dummy;
1403 	struct buf *lefttreemax, *righttreemin, *y;
1404 
1405 	if (root == NULL)
1406 		return (NULL);
1407 	lefttreemax = righttreemin = &dummy;
1408 	for (;;) {
1409 		if (lblkno < root->b_lblkno ||
1410 		    (lblkno == root->b_lblkno &&
1411 		    (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1412 			if ((y = root->b_left) == NULL)
1413 				break;
1414 			if (lblkno < y->b_lblkno) {
1415 				/* Rotate right. */
1416 				root->b_left = y->b_right;
1417 				y->b_right = root;
1418 				root = y;
1419 				if ((y = root->b_left) == NULL)
1420 					break;
1421 			}
1422 			/* Link into the new root's right tree. */
1423 			righttreemin->b_left = root;
1424 			righttreemin = root;
1425 		} else if (lblkno > root->b_lblkno ||
1426 		    (lblkno == root->b_lblkno &&
1427 		    (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1428 			if ((y = root->b_right) == NULL)
1429 				break;
1430 			if (lblkno > y->b_lblkno) {
1431 				/* Rotate left. */
1432 				root->b_right = y->b_left;
1433 				y->b_left = root;
1434 				root = y;
1435 				if ((y = root->b_right) == NULL)
1436 					break;
1437 			}
1438 			/* Link into the new root's left tree. */
1439 			lefttreemax->b_right = root;
1440 			lefttreemax = root;
1441 		} else {
1442 			break;
1443 		}
1444 		root = y;
1445 	}
1446 	/* Assemble the new root. */
1447 	lefttreemax->b_right = root->b_left;
1448 	righttreemin->b_left = root->b_right;
1449 	root->b_left = dummy.b_right;
1450 	root->b_right = dummy.b_left;
1451 	return (root);
1452 }
1453 
1454 static
1455 void
1456 buf_vlist_remove(struct buf *bp)
1457 {
1458 	struct vnode *vp = bp->b_vp;
1459 	struct buf *root;
1460 
1461 	ASSERT_VI_LOCKED(vp, "buf_vlist_remove");
1462 	if (bp->b_xflags & BX_VNDIRTY) {
1463 		if (bp != vp->v_dirtyblkroot) {
1464 			root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_dirtyblkroot);
1465 			KASSERT(root == bp, ("splay lookup failed during dirty remove"));
1466 		}
1467 		if (bp->b_left == NULL) {
1468 			root = bp->b_right;
1469 		} else {
1470 			root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1471 			root->b_right = bp->b_right;
1472 		}
1473 		vp->v_dirtyblkroot = root;
1474 		TAILQ_REMOVE(&vp->v_dirtyblkhd, bp, b_vnbufs);
1475 		vp->v_dirtybufcnt--;
1476 	} else {
1477 		/* KASSERT(bp->b_xflags & BX_VNCLEAN, ("bp wasn't clean")); */
1478 		if (bp != vp->v_cleanblkroot) {
1479 			root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_cleanblkroot);
1480 			KASSERT(root == bp, ("splay lookup failed during clean remove"));
1481 		}
1482 		if (bp->b_left == NULL) {
1483 			root = bp->b_right;
1484 		} else {
1485 			root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1486 			root->b_right = bp->b_right;
1487 		}
1488 		vp->v_cleanblkroot = root;
1489 		TAILQ_REMOVE(&vp->v_cleanblkhd, bp, b_vnbufs);
1490 		vp->v_cleanbufcnt--;
1491 	}
1492 	bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1493 }
1494 
1495 /*
1496  * Add the buffer to the sorted clean or dirty block list using a
1497  * splay tree algorithm.
1498  *
1499  * NOTE: xflags is passed as a constant, optimizing this inline function!
1500  */
1501 static
1502 void
1503 buf_vlist_add(struct buf *bp, struct vnode *vp, b_xflags_t xflags)
1504 {
1505 	struct buf *root;
1506 
1507 	ASSERT_VI_LOCKED(vp, "buf_vlist_add");
1508 	bp->b_xflags |= xflags;
1509 	if (xflags & BX_VNDIRTY) {
1510 		root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_dirtyblkroot);
1511 		if (root == NULL) {
1512 			bp->b_left = NULL;
1513 			bp->b_right = NULL;
1514 			TAILQ_INSERT_TAIL(&vp->v_dirtyblkhd, bp, b_vnbufs);
1515 		} else if (bp->b_lblkno < root->b_lblkno ||
1516 		    (bp->b_lblkno == root->b_lblkno &&
1517 		    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1518 			bp->b_left = root->b_left;
1519 			bp->b_right = root;
1520 			root->b_left = NULL;
1521 			TAILQ_INSERT_BEFORE(root, bp, b_vnbufs);
1522 		} else {
1523 			bp->b_right = root->b_right;
1524 			bp->b_left = root;
1525 			root->b_right = NULL;
1526 			TAILQ_INSERT_AFTER(&vp->v_dirtyblkhd,
1527 			    root, bp, b_vnbufs);
1528 		}
1529 		vp->v_dirtybufcnt++;
1530 		vp->v_dirtyblkroot = bp;
1531 	} else {
1532 		/* KASSERT(xflags & BX_VNCLEAN, ("xflags not clean")); */
1533 		root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_cleanblkroot);
1534 		if (root == NULL) {
1535 			bp->b_left = NULL;
1536 			bp->b_right = NULL;
1537 			TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
1538 		} else if (bp->b_lblkno < root->b_lblkno ||
1539 		    (bp->b_lblkno == root->b_lblkno &&
1540 		    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1541 			bp->b_left = root->b_left;
1542 			bp->b_right = root;
1543 			root->b_left = NULL;
1544 			TAILQ_INSERT_BEFORE(root, bp, b_vnbufs);
1545 		} else {
1546 			bp->b_right = root->b_right;
1547 			bp->b_left = root;
1548 			root->b_right = NULL;
1549 			TAILQ_INSERT_AFTER(&vp->v_cleanblkhd,
1550 			    root, bp, b_vnbufs);
1551 		}
1552 		vp->v_cleanbufcnt++;
1553 		vp->v_cleanblkroot = bp;
1554 	}
1555 }
1556 
1557 /*
1558  * Lookup a buffer using the splay tree.  Note that we specifically avoid
1559  * shadow buffers used in background bitmap writes.
1560  *
1561  * This code isn't quite efficient as it could be because we are maintaining
1562  * two sorted lists and do not know which list the block resides in.
1563  */
1564 struct buf *
1565 gbincore(struct vnode *vp, daddr_t lblkno)
1566 {
1567 	struct buf *bp;
1568 
1569 	GIANT_REQUIRED;
1570 
1571 	ASSERT_VI_LOCKED(vp, "gbincore");
1572 	bp = vp->v_cleanblkroot = buf_splay(lblkno, 0, vp->v_cleanblkroot);
1573 	if (bp && bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1574 		return(bp);
1575 	bp = vp->v_dirtyblkroot = buf_splay(lblkno, 0, vp->v_dirtyblkroot);
1576 	if (bp && bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1577 		return(bp);
1578 	return(NULL);
1579 }
1580 
1581 /*
1582  * Associate a buffer with a vnode.
1583  */
1584 void
1585 bgetvp(vp, bp)
1586 	register struct vnode *vp;
1587 	register struct buf *bp;
1588 {
1589 	int s;
1590 
1591 	KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
1592 
1593 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1594 	    ("bgetvp: bp already attached! %p", bp));
1595 
1596 	ASSERT_VI_LOCKED(vp, "bgetvp");
1597 	vholdl(vp);
1598 	bp->b_vp = vp;
1599 	bp->b_dev = vn_todev(vp);
1600 	/*
1601 	 * Insert onto list for new vnode.
1602 	 */
1603 	s = splbio();
1604 	buf_vlist_add(bp, vp, BX_VNCLEAN);
1605 	splx(s);
1606 }
1607 
1608 /*
1609  * Disassociate a buffer from a vnode.
1610  */
1611 void
1612 brelvp(bp)
1613 	register struct buf *bp;
1614 {
1615 	struct vnode *vp;
1616 	int s;
1617 
1618 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1619 
1620 	/*
1621 	 * Delete from old vnode list, if on one.
1622 	 */
1623 	vp = bp->b_vp;
1624 	s = splbio();
1625 	VI_LOCK(vp);
1626 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1627 		buf_vlist_remove(bp);
1628 	if ((vp->v_iflag & VI_ONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1629 		vp->v_iflag &= ~VI_ONWORKLST;
1630 		mtx_lock(&sync_mtx);
1631 		LIST_REMOVE(vp, v_synclist);
1632 		mtx_unlock(&sync_mtx);
1633 	}
1634 	vdropl(vp);
1635 	VI_UNLOCK(vp);
1636 	bp->b_vp = (struct vnode *) 0;
1637 	if (bp->b_object)
1638 		bp->b_object = NULL;
1639 	splx(s);
1640 }
1641 
1642 /*
1643  * Add an item to the syncer work queue.
1644  */
1645 static void
1646 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
1647 {
1648 	int s, slot;
1649 
1650 	s = splbio();
1651 	ASSERT_VI_LOCKED(vp, "vn_syncer_add_to_worklist");
1652 
1653 	mtx_lock(&sync_mtx);
1654 	if (vp->v_iflag & VI_ONWORKLST)
1655 		LIST_REMOVE(vp, v_synclist);
1656 	else
1657 		vp->v_iflag |= VI_ONWORKLST;
1658 
1659 	if (delay > syncer_maxdelay - 2)
1660 		delay = syncer_maxdelay - 2;
1661 	slot = (syncer_delayno + delay) & syncer_mask;
1662 
1663 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
1664 	mtx_unlock(&sync_mtx);
1665 
1666 	splx(s);
1667 }
1668 
1669 struct  proc *updateproc;
1670 static void sched_sync(void);
1671 static struct kproc_desc up_kp = {
1672 	"syncer",
1673 	sched_sync,
1674 	&updateproc
1675 };
1676 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1677 
1678 /*
1679  * System filesystem synchronizer daemon.
1680  */
1681 static void
1682 sched_sync(void)
1683 {
1684 	struct synclist *slp;
1685 	struct vnode *vp;
1686 	struct mount *mp;
1687 	long starttime;
1688 	int s;
1689 	struct thread *td = FIRST_THREAD_IN_PROC(updateproc);  /* XXXKSE */
1690 
1691 	mtx_lock(&Giant);
1692 
1693 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, td->td_proc,
1694 	    SHUTDOWN_PRI_LAST);
1695 
1696 	for (;;) {
1697 		kthread_suspend_check(td->td_proc);
1698 
1699 		starttime = time_second;
1700 
1701 		/*
1702 		 * Push files whose dirty time has expired.  Be careful
1703 		 * of interrupt race on slp queue.
1704 		 */
1705 		s = splbio();
1706 		mtx_lock(&sync_mtx);
1707 		slp = &syncer_workitem_pending[syncer_delayno];
1708 		syncer_delayno += 1;
1709 		if (syncer_delayno == syncer_maxdelay)
1710 			syncer_delayno = 0;
1711 		splx(s);
1712 
1713 		while ((vp = LIST_FIRST(slp)) != NULL) {
1714 			mtx_unlock(&sync_mtx);
1715 			if (VOP_ISLOCKED(vp, NULL) == 0 &&
1716 			    vn_start_write(vp, &mp, V_NOWAIT) == 0) {
1717 				vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1718 				(void) VOP_FSYNC(vp, td->td_ucred, MNT_LAZY, td);
1719 				VOP_UNLOCK(vp, 0, td);
1720 				vn_finished_write(mp);
1721 			}
1722 			s = splbio();
1723 			mtx_lock(&sync_mtx);
1724 			if (LIST_FIRST(slp) == vp) {
1725 				mtx_unlock(&sync_mtx);
1726 				/*
1727 				 * Note: VFS vnodes can remain on the
1728 				 * worklist too with no dirty blocks, but
1729 				 * since sync_fsync() moves it to a different
1730 				 * slot we are safe.
1731 				 */
1732 				VI_LOCK(vp);
1733 				if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1734 				    !vn_isdisk(vp, NULL)) {
1735 					panic("sched_sync: fsync failed "
1736 					      "vp %p tag %s", vp, vp->v_tag);
1737 				}
1738 				/*
1739 				 * Put us back on the worklist.  The worklist
1740 				 * routine will remove us from our current
1741 				 * position and then add us back in at a later
1742 				 * position.
1743 				 */
1744 				vn_syncer_add_to_worklist(vp, syncdelay);
1745 				VI_UNLOCK(vp);
1746 				mtx_lock(&sync_mtx);
1747 			}
1748 			splx(s);
1749 		}
1750 		mtx_unlock(&sync_mtx);
1751 
1752 		/*
1753 		 * Do soft update processing.
1754 		 */
1755 		if (softdep_process_worklist_hook != NULL)
1756 			(*softdep_process_worklist_hook)(NULL);
1757 
1758 		/*
1759 		 * The variable rushjob allows the kernel to speed up the
1760 		 * processing of the filesystem syncer process. A rushjob
1761 		 * value of N tells the filesystem syncer to process the next
1762 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1763 		 * is used by the soft update code to speed up the filesystem
1764 		 * syncer process when the incore state is getting so far
1765 		 * ahead of the disk that the kernel memory pool is being
1766 		 * threatened with exhaustion.
1767 		 */
1768 		mtx_lock(&sync_mtx);
1769 		if (rushjob > 0) {
1770 			rushjob -= 1;
1771 			mtx_unlock(&sync_mtx);
1772 			continue;
1773 		}
1774 		mtx_unlock(&sync_mtx);
1775 		/*
1776 		 * If it has taken us less than a second to process the
1777 		 * current work, then wait. Otherwise start right over
1778 		 * again. We can still lose time if any single round
1779 		 * takes more than two seconds, but it does not really
1780 		 * matter as we are just trying to generally pace the
1781 		 * filesystem activity.
1782 		 */
1783 		if (time_second == starttime)
1784 			tsleep(&lbolt, PPAUSE, "syncer", 0);
1785 	}
1786 }
1787 
1788 /*
1789  * Request the syncer daemon to speed up its work.
1790  * We never push it to speed up more than half of its
1791  * normal turn time, otherwise it could take over the cpu.
1792  * XXXKSE  only one update?
1793  */
1794 int
1795 speedup_syncer()
1796 {
1797 	struct thread *td;
1798 	int ret = 0;
1799 
1800 	td = FIRST_THREAD_IN_PROC(updateproc);
1801 	mtx_lock_spin(&sched_lock);
1802 	if (td->td_wchan == &lbolt) {
1803 		unsleep(td);
1804 		TD_CLR_SLEEPING(td);
1805 		setrunnable(td);
1806 	}
1807 	mtx_unlock_spin(&sched_lock);
1808 	mtx_lock(&sync_mtx);
1809 	if (rushjob < syncdelay / 2) {
1810 		rushjob += 1;
1811 		stat_rush_requests += 1;
1812 		ret = 1;
1813 	}
1814 	mtx_unlock(&sync_mtx);
1815 	return (ret);
1816 }
1817 
1818 /*
1819  * Associate a p-buffer with a vnode.
1820  *
1821  * Also sets B_PAGING flag to indicate that vnode is not fully associated
1822  * with the buffer.  i.e. the bp has not been linked into the vnode or
1823  * ref-counted.
1824  */
1825 void
1826 pbgetvp(vp, bp)
1827 	register struct vnode *vp;
1828 	register struct buf *bp;
1829 {
1830 
1831 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1832 
1833 	bp->b_vp = vp;
1834 	bp->b_flags |= B_PAGING;
1835 	bp->b_dev = vn_todev(vp);
1836 }
1837 
1838 /*
1839  * Disassociate a p-buffer from a vnode.
1840  */
1841 void
1842 pbrelvp(bp)
1843 	register struct buf *bp;
1844 {
1845 
1846 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1847 
1848 	/* XXX REMOVE ME */
1849 	VI_LOCK(bp->b_vp);
1850 	if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1851 		panic(
1852 		    "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1853 		    bp,
1854 		    (int)bp->b_flags
1855 		);
1856 	}
1857 	VI_UNLOCK(bp->b_vp);
1858 	bp->b_vp = (struct vnode *) 0;
1859 	bp->b_flags &= ~B_PAGING;
1860 }
1861 
1862 /*
1863  * Reassign a buffer from one vnode to another.
1864  * Used to assign file specific control information
1865  * (indirect blocks) to the vnode to which they belong.
1866  */
1867 void
1868 reassignbuf(bp, newvp)
1869 	register struct buf *bp;
1870 	register struct vnode *newvp;
1871 {
1872 	int delay;
1873 	int s;
1874 
1875 	if (newvp == NULL) {
1876 		printf("reassignbuf: NULL");
1877 		return;
1878 	}
1879 	++reassignbufcalls;
1880 
1881 	/*
1882 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1883 	 * is not fully linked in.
1884 	 */
1885 	if (bp->b_flags & B_PAGING)
1886 		panic("cannot reassign paging buffer");
1887 
1888 	s = splbio();
1889 	/*
1890 	 * Delete from old vnode list, if on one.
1891 	 */
1892 	VI_LOCK(bp->b_vp);
1893 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1894 		buf_vlist_remove(bp);
1895 		if (bp->b_vp != newvp) {
1896 			vdropl(bp->b_vp);
1897 			bp->b_vp = NULL;	/* for clarification */
1898 		}
1899 	}
1900 	VI_UNLOCK(bp->b_vp);
1901 	/*
1902 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1903 	 * of clean buffers.
1904 	 */
1905 	VI_LOCK(newvp);
1906 	if (bp->b_flags & B_DELWRI) {
1907 		if ((newvp->v_iflag & VI_ONWORKLST) == 0) {
1908 			switch (newvp->v_type) {
1909 			case VDIR:
1910 				delay = dirdelay;
1911 				break;
1912 			case VCHR:
1913 				if (newvp->v_rdev->si_mountpoint != NULL) {
1914 					delay = metadelay;
1915 					break;
1916 				}
1917 				/* FALLTHROUGH */
1918 			default:
1919 				delay = filedelay;
1920 			}
1921 			vn_syncer_add_to_worklist(newvp, delay);
1922 		}
1923 		buf_vlist_add(bp, newvp, BX_VNDIRTY);
1924 	} else {
1925 		buf_vlist_add(bp, newvp, BX_VNCLEAN);
1926 
1927 		if ((newvp->v_iflag & VI_ONWORKLST) &&
1928 		    TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1929 			mtx_lock(&sync_mtx);
1930 			LIST_REMOVE(newvp, v_synclist);
1931 			mtx_unlock(&sync_mtx);
1932 			newvp->v_iflag &= ~VI_ONWORKLST;
1933 		}
1934 	}
1935 	if (bp->b_vp != newvp) {
1936 		bp->b_vp = newvp;
1937 		vholdl(bp->b_vp);
1938 	}
1939 	VI_UNLOCK(newvp);
1940 	splx(s);
1941 }
1942 
1943 /*
1944  * Create a vnode for a device.
1945  * Used for mounting the root filesystem.
1946  */
1947 int
1948 bdevvp(dev, vpp)
1949 	dev_t dev;
1950 	struct vnode **vpp;
1951 {
1952 	register struct vnode *vp;
1953 	struct vnode *nvp;
1954 	int error;
1955 
1956 	if (dev == NODEV) {
1957 		*vpp = NULLVP;
1958 		return (ENXIO);
1959 	}
1960 	if (vfinddev(dev, VCHR, vpp))
1961 		return (0);
1962 	error = getnewvnode("none", (struct mount *)0, spec_vnodeop_p, &nvp);
1963 	if (error) {
1964 		*vpp = NULLVP;
1965 		return (error);
1966 	}
1967 	vp = nvp;
1968 	vp->v_type = VCHR;
1969 	addalias(vp, dev);
1970 	*vpp = vp;
1971 	return (0);
1972 }
1973 
1974 static void
1975 v_incr_usecount(struct vnode *vp, int delta)
1976 {
1977 	vp->v_usecount += delta;
1978 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1979 		mtx_lock(&spechash_mtx);
1980 		vp->v_rdev->si_usecount += delta;
1981 		mtx_unlock(&spechash_mtx);
1982 	}
1983 }
1984 
1985 /*
1986  * Add vnode to the alias list hung off the dev_t.
1987  *
1988  * The reason for this gunk is that multiple vnodes can reference
1989  * the same physical device, so checking vp->v_usecount to see
1990  * how many users there are is inadequate; the v_usecount for
1991  * the vnodes need to be accumulated.  vcount() does that.
1992  */
1993 struct vnode *
1994 addaliasu(nvp, nvp_rdev)
1995 	struct vnode *nvp;
1996 	udev_t nvp_rdev;
1997 {
1998 	struct vnode *ovp;
1999 	vop_t **ops;
2000 	dev_t dev;
2001 
2002 	if (nvp->v_type == VBLK)
2003 		return (nvp);
2004 	if (nvp->v_type != VCHR)
2005 		panic("addaliasu on non-special vnode");
2006 	dev = udev2dev(nvp_rdev, 0);
2007 	/*
2008 	 * Check to see if we have a bdevvp vnode with no associated
2009 	 * filesystem. If so, we want to associate the filesystem of
2010 	 * the new newly instigated vnode with the bdevvp vnode and
2011 	 * discard the newly created vnode rather than leaving the
2012 	 * bdevvp vnode lying around with no associated filesystem.
2013 	 */
2014 	if (vfinddev(dev, nvp->v_type, &ovp) == 0 || ovp->v_data != NULL) {
2015 		addalias(nvp, dev);
2016 		return (nvp);
2017 	}
2018 	/*
2019 	 * Discard unneeded vnode, but save its node specific data.
2020 	 * Note that if there is a lock, it is carried over in the
2021 	 * node specific data to the replacement vnode.
2022 	 */
2023 	vref(ovp);
2024 	ovp->v_data = nvp->v_data;
2025 	ovp->v_tag = nvp->v_tag;
2026 	nvp->v_data = NULL;
2027 	lockdestroy(ovp->v_vnlock);
2028 	lockinit(ovp->v_vnlock, PVFS, nvp->v_vnlock->lk_wmesg,
2029 	    nvp->v_vnlock->lk_timo, nvp->v_vnlock->lk_flags & LK_EXTFLG_MASK);
2030 	ops = ovp->v_op;
2031 	ovp->v_op = nvp->v_op;
2032 	if (VOP_ISLOCKED(nvp, curthread)) {
2033 		VOP_UNLOCK(nvp, 0, curthread);
2034 		vn_lock(ovp, LK_EXCLUSIVE | LK_RETRY, curthread);
2035 	}
2036 	nvp->v_op = ops;
2037 	insmntque(ovp, nvp->v_mount);
2038 	vrele(nvp);
2039 	vgone(nvp);
2040 	return (ovp);
2041 }
2042 
2043 /* This is a local helper function that do the same as addaliasu, but for a
2044  * dev_t instead of an udev_t. */
2045 static void
2046 addalias(nvp, dev)
2047 	struct vnode *nvp;
2048 	dev_t dev;
2049 {
2050 
2051 	KASSERT(nvp->v_type == VCHR, ("addalias on non-special vnode"));
2052 	nvp->v_rdev = dev;
2053 	VI_LOCK(nvp);
2054 	mtx_lock(&spechash_mtx);
2055 	SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
2056 	dev->si_usecount += nvp->v_usecount;
2057 	mtx_unlock(&spechash_mtx);
2058 	VI_UNLOCK(nvp);
2059 }
2060 
2061 /*
2062  * Grab a particular vnode from the free list, increment its
2063  * reference count and lock it. The vnode lock bit is set if the
2064  * vnode is being eliminated in vgone. The process is awakened
2065  * when the transition is completed, and an error returned to
2066  * indicate that the vnode is no longer usable (possibly having
2067  * been changed to a new filesystem type).
2068  */
2069 int
2070 vget(vp, flags, td)
2071 	register struct vnode *vp;
2072 	int flags;
2073 	struct thread *td;
2074 {
2075 	int error;
2076 
2077 	/*
2078 	 * If the vnode is in the process of being cleaned out for
2079 	 * another use, we wait for the cleaning to finish and then
2080 	 * return failure. Cleaning is determined by checking that
2081 	 * the VI_XLOCK flag is set.
2082 	 */
2083 	if ((flags & LK_INTERLOCK) == 0)
2084 		VI_LOCK(vp);
2085 	if (vp->v_iflag & VI_XLOCK && vp->v_vxproc != curthread) {
2086 		vp->v_iflag |= VI_XWANT;
2087 		msleep(vp, VI_MTX(vp), PINOD | PDROP, "vget", 0);
2088 		return (ENOENT);
2089 	}
2090 
2091 	v_incr_usecount(vp, 1);
2092 
2093 	if (VSHOULDBUSY(vp))
2094 		vbusy(vp);
2095 	if (flags & LK_TYPE_MASK) {
2096 		if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
2097 			/*
2098 			 * must expand vrele here because we do not want
2099 			 * to call VOP_INACTIVE if the reference count
2100 			 * drops back to zero since it was never really
2101 			 * active. We must remove it from the free list
2102 			 * before sleeping so that multiple processes do
2103 			 * not try to recycle it.
2104 			 */
2105 			VI_LOCK(vp);
2106 			v_incr_usecount(vp, -1);
2107 			if (VSHOULDFREE(vp))
2108 				vfree(vp);
2109 			else
2110 				vlruvp(vp);
2111 			VI_UNLOCK(vp);
2112 		}
2113 		return (error);
2114 	}
2115 	VI_UNLOCK(vp);
2116 	return (0);
2117 }
2118 
2119 /*
2120  * Increase the reference count of a vnode.
2121  */
2122 void
2123 vref(struct vnode *vp)
2124 {
2125 	VI_LOCK(vp);
2126 	v_incr_usecount(vp, 1);
2127 	VI_UNLOCK(vp);
2128 }
2129 
2130 /*
2131  * Return reference count of a vnode.
2132  *
2133  * The results of this call are only guaranteed when some mechanism other
2134  * than the VI lock is used to stop other processes from gaining references
2135  * to the vnode.  This may be the case if the caller holds the only reference.
2136  * This is also useful when stale data is acceptable as race conditions may
2137  * be accounted for by some other means.
2138  */
2139 int
2140 vrefcnt(struct vnode *vp)
2141 {
2142 	int usecnt;
2143 
2144 	VI_LOCK(vp);
2145 	usecnt = vp->v_usecount;
2146 	VI_UNLOCK(vp);
2147 
2148 	return (usecnt);
2149 }
2150 
2151 
2152 /*
2153  * Vnode put/release.
2154  * If count drops to zero, call inactive routine and return to freelist.
2155  */
2156 void
2157 vrele(vp)
2158 	struct vnode *vp;
2159 {
2160 	struct thread *td = curthread;	/* XXX */
2161 
2162 	KASSERT(vp != NULL, ("vrele: null vp"));
2163 
2164 	VI_LOCK(vp);
2165 
2166 	/* Skip this v_writecount check if we're going to panic below. */
2167 	KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
2168 	    ("vrele: missed vn_close"));
2169 
2170 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2171 	    vp->v_usecount == 1)) {
2172 		v_incr_usecount(vp, -1);
2173 		VI_UNLOCK(vp);
2174 
2175 		return;
2176 	}
2177 
2178 	if (vp->v_usecount == 1) {
2179 		v_incr_usecount(vp, -1);
2180 		/*
2181 		 * We must call VOP_INACTIVE with the node locked. Mark
2182 		 * as VI_DOINGINACT to avoid recursion.
2183 		 */
2184 		if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0) {
2185 			VI_LOCK(vp);
2186 			vp->v_iflag |= VI_DOINGINACT;
2187 			VI_UNLOCK(vp);
2188 			VOP_INACTIVE(vp, td);
2189 			VI_LOCK(vp);
2190 			KASSERT(vp->v_iflag & VI_DOINGINACT,
2191 			    ("vrele: lost VI_DOINGINACT"));
2192 			vp->v_iflag &= ~VI_DOINGINACT;
2193 			VI_UNLOCK(vp);
2194 		}
2195 		VI_LOCK(vp);
2196 		if (VSHOULDFREE(vp))
2197 			vfree(vp);
2198 		else
2199 			vlruvp(vp);
2200 		VI_UNLOCK(vp);
2201 
2202 	} else {
2203 #ifdef DIAGNOSTIC
2204 		vprint("vrele: negative ref count", vp);
2205 #endif
2206 		VI_UNLOCK(vp);
2207 		panic("vrele: negative ref cnt");
2208 	}
2209 }
2210 
2211 /*
2212  * Release an already locked vnode.  This give the same effects as
2213  * unlock+vrele(), but takes less time and avoids releasing and
2214  * re-aquiring the lock (as vrele() aquires the lock internally.)
2215  */
2216 void
2217 vput(vp)
2218 	struct vnode *vp;
2219 {
2220 	struct thread *td = curthread;	/* XXX */
2221 
2222 	GIANT_REQUIRED;
2223 
2224 	KASSERT(vp != NULL, ("vput: null vp"));
2225 	VI_LOCK(vp);
2226 	/* Skip this v_writecount check if we're going to panic below. */
2227 	KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
2228 	    ("vput: missed vn_close"));
2229 
2230 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2231 	    vp->v_usecount == 1)) {
2232 		v_incr_usecount(vp, -1);
2233 		VOP_UNLOCK(vp, LK_INTERLOCK, td);
2234 		return;
2235 	}
2236 
2237 	if (vp->v_usecount == 1) {
2238 		v_incr_usecount(vp, -1);
2239 		/*
2240 		 * We must call VOP_INACTIVE with the node locked, so
2241 		 * we just need to release the vnode mutex. Mark as
2242 		 * as VI_DOINGINACT to avoid recursion.
2243 		 */
2244 		vp->v_iflag |= VI_DOINGINACT;
2245 		VI_UNLOCK(vp);
2246 		VOP_INACTIVE(vp, td);
2247 		VI_LOCK(vp);
2248 		KASSERT(vp->v_iflag & VI_DOINGINACT,
2249 		    ("vput: lost VI_DOINGINACT"));
2250 		vp->v_iflag &= ~VI_DOINGINACT;
2251 		if (VSHOULDFREE(vp))
2252 			vfree(vp);
2253 		else
2254 			vlruvp(vp);
2255 		VI_UNLOCK(vp);
2256 
2257 	} else {
2258 #ifdef DIAGNOSTIC
2259 		vprint("vput: negative ref count", vp);
2260 #endif
2261 		panic("vput: negative ref cnt");
2262 	}
2263 }
2264 
2265 /*
2266  * Somebody doesn't want the vnode recycled.
2267  */
2268 void
2269 vhold(struct vnode *vp)
2270 {
2271 	VI_LOCK(vp);
2272 	vholdl(vp);
2273 	VI_UNLOCK(vp);
2274 }
2275 
2276 void
2277 vholdl(vp)
2278 	register struct vnode *vp;
2279 {
2280 	int s;
2281 
2282 	s = splbio();
2283 	vp->v_holdcnt++;
2284 	if (VSHOULDBUSY(vp))
2285 		vbusy(vp);
2286 	splx(s);
2287 }
2288 
2289 /*
2290  * Note that there is one less who cares about this vnode.  vdrop() is the
2291  * opposite of vhold().
2292  */
2293 void
2294 vdrop(struct vnode *vp)
2295 {
2296 	VI_LOCK(vp);
2297 	vdropl(vp);
2298 	VI_UNLOCK(vp);
2299 }
2300 
2301 void
2302 vdropl(vp)
2303 	register struct vnode *vp;
2304 {
2305 	int s;
2306 
2307 	s = splbio();
2308 	if (vp->v_holdcnt <= 0)
2309 		panic("vdrop: holdcnt");
2310 	vp->v_holdcnt--;
2311 	if (VSHOULDFREE(vp))
2312 		vfree(vp);
2313 	else
2314 		vlruvp(vp);
2315 	splx(s);
2316 }
2317 
2318 /*
2319  * Remove any vnodes in the vnode table belonging to mount point mp.
2320  *
2321  * If FORCECLOSE is not specified, there should not be any active ones,
2322  * return error if any are found (nb: this is a user error, not a
2323  * system error). If FORCECLOSE is specified, detach any active vnodes
2324  * that are found.
2325  *
2326  * If WRITECLOSE is set, only flush out regular file vnodes open for
2327  * writing.
2328  *
2329  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2330  *
2331  * `rootrefs' specifies the base reference count for the root vnode
2332  * of this filesystem. The root vnode is considered busy if its
2333  * v_usecount exceeds this value. On a successful return, vflush()
2334  * will call vrele() on the root vnode exactly rootrefs times.
2335  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2336  * be zero.
2337  */
2338 #ifdef DIAGNOSTIC
2339 static int busyprt = 0;		/* print out busy vnodes */
2340 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
2341 #endif
2342 
2343 int
2344 vflush(mp, rootrefs, flags)
2345 	struct mount *mp;
2346 	int rootrefs;
2347 	int flags;
2348 {
2349 	struct thread *td = curthread;	/* XXX */
2350 	struct vnode *vp, *nvp, *rootvp = NULL;
2351 	struct vattr vattr;
2352 	int busy = 0, error;
2353 
2354 	if (rootrefs > 0) {
2355 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2356 		    ("vflush: bad args"));
2357 		/*
2358 		 * Get the filesystem root vnode. We can vput() it
2359 		 * immediately, since with rootrefs > 0, it won't go away.
2360 		 */
2361 		if ((error = VFS_ROOT(mp, &rootvp)) != 0)
2362 			return (error);
2363 		vput(rootvp);
2364 
2365 	}
2366 	mtx_lock(&mntvnode_mtx);
2367 loop:
2368 	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) {
2369 		/*
2370 		 * Make sure this vnode wasn't reclaimed in getnewvnode().
2371 		 * Start over if it has (it won't be on the list anymore).
2372 		 */
2373 		if (vp->v_mount != mp)
2374 			goto loop;
2375 		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2376 
2377 		VI_LOCK(vp);
2378 		mtx_unlock(&mntvnode_mtx);
2379 		vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY, td);
2380 		/*
2381 		 * Skip over a vnodes marked VV_SYSTEM.
2382 		 */
2383 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2384 			VOP_UNLOCK(vp, 0, td);
2385 			mtx_lock(&mntvnode_mtx);
2386 			continue;
2387 		}
2388 		/*
2389 		 * If WRITECLOSE is set, flush out unlinked but still open
2390 		 * files (even if open only for reading) and regular file
2391 		 * vnodes open for writing.
2392 		 */
2393 		if (flags & WRITECLOSE) {
2394 			error = VOP_GETATTR(vp, &vattr, td->td_ucred, td);
2395 			VI_LOCK(vp);
2396 
2397 			if ((vp->v_type == VNON ||
2398 			    (error == 0 && vattr.va_nlink > 0)) &&
2399 			    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2400 				VOP_UNLOCK(vp, LK_INTERLOCK, td);
2401 				mtx_lock(&mntvnode_mtx);
2402 				continue;
2403 			}
2404 		} else
2405 			VI_LOCK(vp);
2406 
2407 		VOP_UNLOCK(vp, 0, td);
2408 
2409 		/*
2410 		 * With v_usecount == 0, all we need to do is clear out the
2411 		 * vnode data structures and we are done.
2412 		 */
2413 		if (vp->v_usecount == 0) {
2414 			vgonel(vp, td);
2415 			mtx_lock(&mntvnode_mtx);
2416 			continue;
2417 		}
2418 
2419 		/*
2420 		 * If FORCECLOSE is set, forcibly close the vnode. For block
2421 		 * or character devices, revert to an anonymous device. For
2422 		 * all other files, just kill them.
2423 		 */
2424 		if (flags & FORCECLOSE) {
2425 			if (vp->v_type != VCHR) {
2426 				vgonel(vp, td);
2427 			} else {
2428 				vclean(vp, 0, td);
2429 				VI_UNLOCK(vp);
2430 				vp->v_op = spec_vnodeop_p;
2431 				insmntque(vp, (struct mount *) 0);
2432 			}
2433 			mtx_lock(&mntvnode_mtx);
2434 			continue;
2435 		}
2436 #ifdef DIAGNOSTIC
2437 		if (busyprt)
2438 			vprint("vflush: busy vnode", vp);
2439 #endif
2440 		VI_UNLOCK(vp);
2441 		mtx_lock(&mntvnode_mtx);
2442 		busy++;
2443 	}
2444 	mtx_unlock(&mntvnode_mtx);
2445 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2446 		/*
2447 		 * If just the root vnode is busy, and if its refcount
2448 		 * is equal to `rootrefs', then go ahead and kill it.
2449 		 */
2450 		VI_LOCK(rootvp);
2451 		KASSERT(busy > 0, ("vflush: not busy"));
2452 		KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
2453 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2454 			vgonel(rootvp, td);
2455 			busy = 0;
2456 		} else
2457 			VI_UNLOCK(rootvp);
2458 	}
2459 	if (busy)
2460 		return (EBUSY);
2461 	for (; rootrefs > 0; rootrefs--)
2462 		vrele(rootvp);
2463 	return (0);
2464 }
2465 
2466 /*
2467  * This moves a now (likely recyclable) vnode to the end of the
2468  * mountlist.  XXX However, it is temporarily disabled until we
2469  * can clean up ffs_sync() and friends, which have loop restart
2470  * conditions which this code causes to operate O(N^2).
2471  */
2472 static void
2473 vlruvp(struct vnode *vp)
2474 {
2475 #if 0
2476 	struct mount *mp;
2477 
2478 	if ((mp = vp->v_mount) != NULL) {
2479 		mtx_lock(&mntvnode_mtx);
2480 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
2481 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
2482 		mtx_unlock(&mntvnode_mtx);
2483 	}
2484 #endif
2485 }
2486 
2487 /*
2488  * Disassociate the underlying filesystem from a vnode.
2489  */
2490 static void
2491 vclean(vp, flags, td)
2492 	struct vnode *vp;
2493 	int flags;
2494 	struct thread *td;
2495 {
2496 	int active;
2497 
2498 	ASSERT_VI_LOCKED(vp, "vclean");
2499 	/*
2500 	 * Check to see if the vnode is in use. If so we have to reference it
2501 	 * before we clean it out so that its count cannot fall to zero and
2502 	 * generate a race against ourselves to recycle it.
2503 	 */
2504 	if ((active = vp->v_usecount))
2505 		v_incr_usecount(vp, 1);
2506 
2507 	/*
2508 	 * Prevent the vnode from being recycled or brought into use while we
2509 	 * clean it out.
2510 	 */
2511 	if (vp->v_iflag & VI_XLOCK)
2512 		panic("vclean: deadlock");
2513 	vp->v_iflag |= VI_XLOCK;
2514 	vp->v_vxproc = curthread;
2515 	/*
2516 	 * Even if the count is zero, the VOP_INACTIVE routine may still
2517 	 * have the object locked while it cleans it out. The VOP_LOCK
2518 	 * ensures that the VOP_INACTIVE routine is done with its work.
2519 	 * For active vnodes, it ensures that no other activity can
2520 	 * occur while the underlying object is being cleaned out.
2521 	 */
2522 	VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td);
2523 
2524 	/*
2525 	 * Clean out any buffers associated with the vnode.
2526 	 * If the flush fails, just toss the buffers.
2527 	 */
2528 	if (flags & DOCLOSE) {
2529 		struct buf *bp;
2530 		VI_LOCK(vp);
2531 		bp = TAILQ_FIRST(&vp->v_dirtyblkhd);
2532 		VI_UNLOCK(vp);
2533 		if (bp != NULL)
2534 			(void) vn_write_suspend_wait(vp, NULL, V_WAIT);
2535 		if (vinvalbuf(vp, V_SAVE, NOCRED, td, 0, 0) != 0)
2536 			vinvalbuf(vp, 0, NOCRED, td, 0, 0);
2537 	}
2538 
2539 	VOP_DESTROYVOBJECT(vp);
2540 
2541 	/*
2542 	 * Any other processes trying to obtain this lock must first
2543 	 * wait for VXLOCK to clear, then call the new lock operation.
2544 	 */
2545 	VOP_UNLOCK(vp, 0, td);
2546 
2547 	/*
2548 	 * If purging an active vnode, it must be closed and
2549 	 * deactivated before being reclaimed. Note that the
2550 	 * VOP_INACTIVE will unlock the vnode.
2551 	 */
2552 	if (active) {
2553 		if (flags & DOCLOSE)
2554 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2555 		VI_LOCK(vp);
2556 		if ((vp->v_iflag & VI_DOINGINACT) == 0) {
2557 			vp->v_iflag |= VI_DOINGINACT;
2558 			VI_UNLOCK(vp);
2559 			if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT, td) != 0)
2560 				panic("vclean: cannot relock.");
2561 			VOP_INACTIVE(vp, td);
2562 			VI_LOCK(vp);
2563 			KASSERT(vp->v_iflag & VI_DOINGINACT,
2564 			    ("vclean: lost VI_DOINGINACT"));
2565 			vp->v_iflag &= ~VI_DOINGINACT;
2566 		}
2567 		VI_UNLOCK(vp);
2568 	}
2569 
2570 	/*
2571 	 * Reclaim the vnode.
2572 	 */
2573 	if (VOP_RECLAIM(vp, td))
2574 		panic("vclean: cannot reclaim");
2575 
2576 	if (active) {
2577 		/*
2578 		 * Inline copy of vrele() since VOP_INACTIVE
2579 		 * has already been called.
2580 		 */
2581 		VI_LOCK(vp);
2582 		v_incr_usecount(vp, -1);
2583 		if (vp->v_usecount <= 0) {
2584 #ifdef DIAGNOSTIC
2585 			if (vp->v_usecount < 0 || vp->v_writecount != 0) {
2586 				vprint("vclean: bad ref count", vp);
2587 				panic("vclean: ref cnt");
2588 			}
2589 #endif
2590 			vfree(vp);
2591 		}
2592 		VI_UNLOCK(vp);
2593 	}
2594 
2595 	cache_purge(vp);
2596 	VI_LOCK(vp);
2597 	if (VSHOULDFREE(vp))
2598 		vfree(vp);
2599 
2600 	/*
2601 	 * Done with purge, reset to the standard lock and
2602 	 * notify sleepers of the grim news.
2603 	 */
2604 	vp->v_vnlock = &vp->v_lock;
2605 	vp->v_op = dead_vnodeop_p;
2606 	if (vp->v_pollinfo != NULL)
2607 		vn_pollgone(vp);
2608 	vp->v_tag = "none";
2609 	vp->v_iflag &= ~VI_XLOCK;
2610 	vp->v_vxproc = NULL;
2611 	if (vp->v_iflag & VI_XWANT) {
2612 		vp->v_iflag &= ~VI_XWANT;
2613 		wakeup(vp);
2614 	}
2615 }
2616 
2617 /*
2618  * Eliminate all activity associated with the requested vnode
2619  * and with all vnodes aliased to the requested vnode.
2620  */
2621 int
2622 vop_revoke(ap)
2623 	struct vop_revoke_args /* {
2624 		struct vnode *a_vp;
2625 		int a_flags;
2626 	} */ *ap;
2627 {
2628 	struct vnode *vp, *vq;
2629 	dev_t dev;
2630 
2631 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
2632 	vp = ap->a_vp;
2633 	KASSERT((vp->v_type == VCHR), ("vop_revoke: not VCHR"));
2634 
2635 	VI_LOCK(vp);
2636 	/*
2637 	 * If a vgone (or vclean) is already in progress,
2638 	 * wait until it is done and return.
2639 	 */
2640 	if (vp->v_iflag & VI_XLOCK) {
2641 		vp->v_iflag |= VI_XWANT;
2642 		msleep(vp, VI_MTX(vp), PINOD | PDROP,
2643 		    "vop_revokeall", 0);
2644 		return (0);
2645 	}
2646 	VI_UNLOCK(vp);
2647 	dev = vp->v_rdev;
2648 	for (;;) {
2649 		mtx_lock(&spechash_mtx);
2650 		vq = SLIST_FIRST(&dev->si_hlist);
2651 		mtx_unlock(&spechash_mtx);
2652 		if (!vq)
2653 			break;
2654 		vgone(vq);
2655 	}
2656 	return (0);
2657 }
2658 
2659 /*
2660  * Recycle an unused vnode to the front of the free list.
2661  * Release the passed interlock if the vnode will be recycled.
2662  */
2663 int
2664 vrecycle(vp, inter_lkp, td)
2665 	struct vnode *vp;
2666 	struct mtx *inter_lkp;
2667 	struct thread *td;
2668 {
2669 
2670 	VI_LOCK(vp);
2671 	if (vp->v_usecount == 0) {
2672 		if (inter_lkp) {
2673 			mtx_unlock(inter_lkp);
2674 		}
2675 		vgonel(vp, td);
2676 		return (1);
2677 	}
2678 	VI_UNLOCK(vp);
2679 	return (0);
2680 }
2681 
2682 /*
2683  * Eliminate all activity associated with a vnode
2684  * in preparation for reuse.
2685  */
2686 void
2687 vgone(vp)
2688 	register struct vnode *vp;
2689 {
2690 	struct thread *td = curthread;	/* XXX */
2691 
2692 	VI_LOCK(vp);
2693 	vgonel(vp, td);
2694 }
2695 
2696 /*
2697  * vgone, with the vp interlock held.
2698  */
2699 void
2700 vgonel(vp, td)
2701 	struct vnode *vp;
2702 	struct thread *td;
2703 {
2704 	int s;
2705 
2706 	/*
2707 	 * If a vgone (or vclean) is already in progress,
2708 	 * wait until it is done and return.
2709 	 */
2710 	ASSERT_VI_LOCKED(vp, "vgonel");
2711 	if (vp->v_iflag & VI_XLOCK) {
2712 		vp->v_iflag |= VI_XWANT;
2713 		msleep(vp, VI_MTX(vp), PINOD | PDROP, "vgone", 0);
2714 		return;
2715 	}
2716 
2717 	/*
2718 	 * Clean out the filesystem specific data.
2719 	 */
2720 	vclean(vp, DOCLOSE, td);
2721 	VI_UNLOCK(vp);
2722 
2723 	/*
2724 	 * Delete from old mount point vnode list, if on one.
2725 	 */
2726 	if (vp->v_mount != NULL)
2727 		insmntque(vp, (struct mount *)0);
2728 	/*
2729 	 * If special device, remove it from special device alias list
2730 	 * if it is on one.
2731 	 */
2732 	if (vp->v_type == VCHR && vp->v_rdev != NULL && vp->v_rdev != NODEV) {
2733 		VI_LOCK(vp);
2734 		mtx_lock(&spechash_mtx);
2735 		SLIST_REMOVE(&vp->v_rdev->si_hlist, vp, vnode, v_specnext);
2736 		vp->v_rdev->si_usecount -= vp->v_usecount;
2737 		mtx_unlock(&spechash_mtx);
2738 		VI_UNLOCK(vp);
2739 		vp->v_rdev = NULL;
2740 	}
2741 
2742 	/*
2743 	 * If it is on the freelist and not already at the head,
2744 	 * move it to the head of the list. The test of the
2745 	 * VDOOMED flag and the reference count of zero is because
2746 	 * it will be removed from the free list by getnewvnode,
2747 	 * but will not have its reference count incremented until
2748 	 * after calling vgone. If the reference count were
2749 	 * incremented first, vgone would (incorrectly) try to
2750 	 * close the previous instance of the underlying object.
2751 	 */
2752 	VI_LOCK(vp);
2753 	if (vp->v_usecount == 0 && !(vp->v_iflag & VI_DOOMED)) {
2754 		s = splbio();
2755 		mtx_lock(&vnode_free_list_mtx);
2756 		if (vp->v_iflag & VI_FREE) {
2757 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2758 		} else {
2759 			vp->v_iflag |= VI_FREE;
2760 			freevnodes++;
2761 		}
2762 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2763 		mtx_unlock(&vnode_free_list_mtx);
2764 		splx(s);
2765 	}
2766 
2767 	vp->v_type = VBAD;
2768 	VI_UNLOCK(vp);
2769 }
2770 
2771 /*
2772  * Lookup a vnode by device number.
2773  */
2774 int
2775 vfinddev(dev, type, vpp)
2776 	dev_t dev;
2777 	enum vtype type;
2778 	struct vnode **vpp;
2779 {
2780 	struct vnode *vp;
2781 
2782 	mtx_lock(&spechash_mtx);
2783 	SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2784 		if (type == vp->v_type) {
2785 			*vpp = vp;
2786 			mtx_unlock(&spechash_mtx);
2787 			return (1);
2788 		}
2789 	}
2790 	mtx_unlock(&spechash_mtx);
2791 	return (0);
2792 }
2793 
2794 /*
2795  * Calculate the total number of references to a special device.
2796  */
2797 int
2798 vcount(vp)
2799 	struct vnode *vp;
2800 {
2801 	int count;
2802 
2803 	mtx_lock(&spechash_mtx);
2804 	count = vp->v_rdev->si_usecount;
2805 	mtx_unlock(&spechash_mtx);
2806 	return (count);
2807 }
2808 
2809 /*
2810  * Same as above, but using the dev_t as argument
2811  */
2812 int
2813 count_dev(dev)
2814 	dev_t dev;
2815 {
2816 	struct vnode *vp;
2817 
2818 	vp = SLIST_FIRST(&dev->si_hlist);
2819 	if (vp == NULL)
2820 		return (0);
2821 	return(vcount(vp));
2822 }
2823 
2824 /*
2825  * Print out a description of a vnode.
2826  */
2827 static char *typename[] =
2828 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2829 
2830 void
2831 vprint(label, vp)
2832 	char *label;
2833 	struct vnode *vp;
2834 {
2835 	char buf[96];
2836 
2837 	if (label != NULL)
2838 		printf("%s: %p: ", label, (void *)vp);
2839 	else
2840 		printf("%p: ", (void *)vp);
2841 	printf("tag %s, type %s, usecount %d, writecount %d, refcount %d,",
2842 	    vp->v_tag, typename[vp->v_type], vp->v_usecount,
2843 	    vp->v_writecount, vp->v_holdcnt);
2844 	buf[0] = '\0';
2845 	if (vp->v_vflag & VV_ROOT)
2846 		strcat(buf, "|VV_ROOT");
2847 	if (vp->v_vflag & VV_TEXT)
2848 		strcat(buf, "|VV_TEXT");
2849 	if (vp->v_vflag & VV_SYSTEM)
2850 		strcat(buf, "|VV_SYSTEM");
2851 	if (vp->v_iflag & VI_XLOCK)
2852 		strcat(buf, "|VI_XLOCK");
2853 	if (vp->v_iflag & VI_XWANT)
2854 		strcat(buf, "|VI_XWANT");
2855 	if (vp->v_iflag & VI_BWAIT)
2856 		strcat(buf, "|VI_BWAIT");
2857 	if (vp->v_iflag & VI_DOOMED)
2858 		strcat(buf, "|VI_DOOMED");
2859 	if (vp->v_iflag & VI_FREE)
2860 		strcat(buf, "|VI_FREE");
2861 	if (vp->v_vflag & VV_OBJBUF)
2862 		strcat(buf, "|VV_OBJBUF");
2863 	if (buf[0] != '\0')
2864 		printf(" flags (%s),", &buf[1]);
2865 	lockmgr_printinfo(vp->v_vnlock);
2866 	printf("\n");
2867 	if (vp->v_data != NULL)
2868 		VOP_PRINT(vp);
2869 }
2870 
2871 #ifdef DDB
2872 #include <ddb/ddb.h>
2873 /*
2874  * List all of the locked vnodes in the system.
2875  * Called when debugging the kernel.
2876  */
2877 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2878 {
2879 	struct thread *td = curthread;	/* XXX */
2880 	struct mount *mp, *nmp;
2881 	struct vnode *vp;
2882 
2883 	printf("Locked vnodes\n");
2884 	mtx_lock(&mountlist_mtx);
2885 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2886 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
2887 			nmp = TAILQ_NEXT(mp, mnt_list);
2888 			continue;
2889 		}
2890 		mtx_lock(&mntvnode_mtx);
2891 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2892 			if (VOP_ISLOCKED(vp, NULL))
2893 				vprint(NULL, vp);
2894 		}
2895 		mtx_unlock(&mntvnode_mtx);
2896 		mtx_lock(&mountlist_mtx);
2897 		nmp = TAILQ_NEXT(mp, mnt_list);
2898 		vfs_unbusy(mp, td);
2899 	}
2900 	mtx_unlock(&mountlist_mtx);
2901 }
2902 #endif
2903 
2904 /*
2905  * Fill in a struct xvfsconf based on a struct vfsconf.
2906  */
2907 static void
2908 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2909 {
2910 
2911 	strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2912 	xvfsp->vfc_typenum = vfsp->vfc_typenum;
2913 	xvfsp->vfc_refcount = vfsp->vfc_refcount;
2914 	xvfsp->vfc_flags = vfsp->vfc_flags;
2915 	/*
2916 	 * These are unused in userland, we keep them
2917 	 * to not break binary compatibility.
2918 	 */
2919 	xvfsp->vfc_vfsops = NULL;
2920 	xvfsp->vfc_next = NULL;
2921 }
2922 
2923 static int
2924 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2925 {
2926 	struct vfsconf *vfsp;
2927 	struct xvfsconf *xvfsp;
2928 	int cnt, error, i;
2929 
2930 	cnt = 0;
2931 	for (vfsp = vfsconf; vfsp != NULL; vfsp = vfsp->vfc_next)
2932 		cnt++;
2933 	xvfsp = malloc(sizeof(struct xvfsconf) * cnt, M_TEMP, M_WAITOK);
2934 	/*
2935 	 * Handle the race that we will have here when struct vfsconf
2936 	 * will be locked down by using both cnt and checking vfc_next
2937 	 * against NULL to determine the end of the loop.  The race will
2938 	 * happen because we will have to unlock before calling malloc().
2939 	 * We are protected by Giant for now.
2940 	 */
2941 	i = 0;
2942 	for (vfsp = vfsconf; vfsp != NULL && i < cnt; vfsp = vfsp->vfc_next) {
2943 		vfsconf2x(vfsp, xvfsp + i);
2944 		i++;
2945 	}
2946 	error = SYSCTL_OUT(req, xvfsp, sizeof(struct xvfsconf) * i);
2947 	free(xvfsp, M_TEMP);
2948 	return (error);
2949 }
2950 
2951 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
2952     "S,xvfsconf", "List of all configured filesystems");
2953 
2954 /*
2955  * Top level filesystem related information gathering.
2956  */
2957 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2958 
2959 static int
2960 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2961 {
2962 	int *name = (int *)arg1 - 1;	/* XXX */
2963 	u_int namelen = arg2 + 1;	/* XXX */
2964 	struct vfsconf *vfsp;
2965 	struct xvfsconf xvfsp;
2966 
2967 	printf("WARNING: userland calling deprecated sysctl, "
2968 	    "please rebuild world\n");
2969 
2970 #if 1 || defined(COMPAT_PRELITE2)
2971 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2972 	if (namelen == 1)
2973 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2974 #endif
2975 
2976 	switch (name[1]) {
2977 	case VFS_MAXTYPENUM:
2978 		if (namelen != 2)
2979 			return (ENOTDIR);
2980 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2981 	case VFS_CONF:
2982 		if (namelen != 3)
2983 			return (ENOTDIR);	/* overloaded */
2984 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2985 			if (vfsp->vfc_typenum == name[2])
2986 				break;
2987 		if (vfsp == NULL)
2988 			return (EOPNOTSUPP);
2989 		vfsconf2x(vfsp, &xvfsp);
2990 		return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
2991 	}
2992 	return (EOPNOTSUPP);
2993 }
2994 
2995 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP, vfs_sysctl,
2996 	"Generic filesystem");
2997 
2998 #if 1 || defined(COMPAT_PRELITE2)
2999 
3000 static int
3001 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3002 {
3003 	int error;
3004 	struct vfsconf *vfsp;
3005 	struct ovfsconf ovfs;
3006 
3007 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
3008 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
3009 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
3010 		ovfs.vfc_index = vfsp->vfc_typenum;
3011 		ovfs.vfc_refcount = vfsp->vfc_refcount;
3012 		ovfs.vfc_flags = vfsp->vfc_flags;
3013 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3014 		if (error)
3015 			return error;
3016 	}
3017 	return 0;
3018 }
3019 
3020 #endif /* 1 || COMPAT_PRELITE2 */
3021 
3022 #define KINFO_VNODESLOP		10
3023 #ifdef notyet
3024 /*
3025  * Dump vnode list (via sysctl).
3026  */
3027 /* ARGSUSED */
3028 static int
3029 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3030 {
3031 	struct xvnode *xvn;
3032 	struct thread *td = req->td;
3033 	struct mount *mp;
3034 	struct vnode *vp;
3035 	int error, len, n;
3036 
3037 	/*
3038 	 * Stale numvnodes access is not fatal here.
3039 	 */
3040 	req->lock = 0;
3041 	len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3042 	if (!req->oldptr)
3043 		/* Make an estimate */
3044 		return (SYSCTL_OUT(req, 0, len));
3045 
3046 	sysctl_wire_old_buffer(req, 0);
3047 	xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3048 	n = 0;
3049 	mtx_lock(&mountlist_mtx);
3050 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3051 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td))
3052 			continue;
3053 		mtx_lock(&mntvnode_mtx);
3054 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3055 			if (n == len)
3056 				break;
3057 			vref(vp);
3058 			xvn[n].xv_size = sizeof *xvn;
3059 			xvn[n].xv_vnode = vp;
3060 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3061 			XV_COPY(usecount);
3062 			XV_COPY(writecount);
3063 			XV_COPY(holdcnt);
3064 			XV_COPY(id);
3065 			XV_COPY(mount);
3066 			XV_COPY(numoutput);
3067 			XV_COPY(type);
3068 #undef XV_COPY
3069 			xvn[n].xv_flag = vp->v_vflag;
3070 
3071 			switch (vp->v_type) {
3072 			case VREG:
3073 			case VDIR:
3074 			case VLNK:
3075 				xvn[n].xv_dev = vp->v_cachedfs;
3076 				xvn[n].xv_ino = vp->v_cachedid;
3077 				break;
3078 			case VBLK:
3079 			case VCHR:
3080 				if (vp->v_rdev == NULL) {
3081 					vrele(vp);
3082 					continue;
3083 				}
3084 				xvn[n].xv_dev = dev2udev(vp->v_rdev);
3085 				break;
3086 			case VSOCK:
3087 				xvn[n].xv_socket = vp->v_socket;
3088 				break;
3089 			case VFIFO:
3090 				xvn[n].xv_fifo = vp->v_fifoinfo;
3091 				break;
3092 			case VNON:
3093 			case VBAD:
3094 			default:
3095 				/* shouldn't happen? */
3096 				vrele(vp);
3097 				continue;
3098 			}
3099 			vrele(vp);
3100 			++n;
3101 		}
3102 		mtx_unlock(&mntvnode_mtx);
3103 		mtx_lock(&mountlist_mtx);
3104 		vfs_unbusy(mp, td);
3105 		if (n == len)
3106 			break;
3107 	}
3108 	mtx_unlock(&mountlist_mtx);
3109 
3110 	error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3111 	free(xvn, M_TEMP);
3112 	return (error);
3113 }
3114 
3115 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3116 	0, 0, sysctl_vnode, "S,xvnode", "");
3117 #endif
3118 
3119 /*
3120  * Check to see if a filesystem is mounted on a block device.
3121  */
3122 int
3123 vfs_mountedon(vp)
3124 	struct vnode *vp;
3125 {
3126 
3127 	if (vp->v_rdev->si_mountpoint != NULL)
3128 		return (EBUSY);
3129 	return (0);
3130 }
3131 
3132 /*
3133  * Unmount all filesystems. The list is traversed in reverse order
3134  * of mounting to avoid dependencies.
3135  */
3136 void
3137 vfs_unmountall()
3138 {
3139 	struct mount *mp;
3140 	struct thread *td;
3141 	int error;
3142 
3143 	if (curthread != NULL)
3144 		td = curthread;
3145 	else
3146 		td = FIRST_THREAD_IN_PROC(initproc); /* XXX XXX proc0? */
3147 	/*
3148 	 * Since this only runs when rebooting, it is not interlocked.
3149 	 */
3150 	while(!TAILQ_EMPTY(&mountlist)) {
3151 		mp = TAILQ_LAST(&mountlist, mntlist);
3152 		error = dounmount(mp, MNT_FORCE, td);
3153 		if (error) {
3154 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
3155 			printf("unmount of %s failed (",
3156 			    mp->mnt_stat.f_mntonname);
3157 			if (error == EBUSY)
3158 				printf("BUSY)\n");
3159 			else
3160 				printf("%d)\n", error);
3161 		} else {
3162 			/* The unmount has removed mp from the mountlist */
3163 		}
3164 	}
3165 }
3166 
3167 /*
3168  * perform msync on all vnodes under a mount point
3169  * the mount point must be locked.
3170  */
3171 void
3172 vfs_msync(struct mount *mp, int flags)
3173 {
3174 	struct vnode *vp, *nvp;
3175 	struct vm_object *obj;
3176 	int tries;
3177 
3178 	GIANT_REQUIRED;
3179 
3180 	tries = 5;
3181 	mtx_lock(&mntvnode_mtx);
3182 loop:
3183 	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) {
3184 		if (vp->v_mount != mp) {
3185 			if (--tries > 0)
3186 				goto loop;
3187 			break;
3188 		}
3189 		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
3190 
3191 		VI_LOCK(vp);
3192 		if (vp->v_iflag & VI_XLOCK) {	/* XXX: what if MNT_WAIT? */
3193 			VI_UNLOCK(vp);
3194 			continue;
3195 		}
3196 
3197 		if ((vp->v_iflag & VI_OBJDIRTY) &&
3198 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
3199 			mtx_unlock(&mntvnode_mtx);
3200 			if (!vget(vp,
3201 			    LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3202 			    curthread)) {
3203 				if (vp->v_vflag & VV_NOSYNC) {	/* unlinked */
3204 					vput(vp);
3205 					mtx_lock(&mntvnode_mtx);
3206 					continue;
3207 				}
3208 
3209 				if (VOP_GETVOBJECT(vp, &obj) == 0) {
3210 					vm_object_page_clean(obj, 0, 0,
3211 					    flags == MNT_WAIT ?
3212 					    OBJPC_SYNC : OBJPC_NOSYNC);
3213 				}
3214 				vput(vp);
3215 			}
3216 			mtx_lock(&mntvnode_mtx);
3217 			if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) {
3218 				if (--tries > 0)
3219 					goto loop;
3220 				break;
3221 			}
3222 		} else
3223 			VI_UNLOCK(vp);
3224 	}
3225 	mtx_unlock(&mntvnode_mtx);
3226 }
3227 
3228 /*
3229  * Create the VM object needed for VMIO and mmap support.  This
3230  * is done for all VREG files in the system.  Some filesystems might
3231  * afford the additional metadata buffering capability of the
3232  * VMIO code by making the device node be VMIO mode also.
3233  *
3234  * vp must be locked when vfs_object_create is called.
3235  */
3236 int
3237 vfs_object_create(vp, td, cred)
3238 	struct vnode *vp;
3239 	struct thread *td;
3240 	struct ucred *cred;
3241 {
3242 	GIANT_REQUIRED;
3243 	return (VOP_CREATEVOBJECT(vp, cred, td));
3244 }
3245 
3246 /*
3247  * Mark a vnode as free, putting it up for recycling.
3248  */
3249 void
3250 vfree(vp)
3251 	struct vnode *vp;
3252 {
3253 	int s;
3254 
3255 	ASSERT_VI_LOCKED(vp, "vfree");
3256 	s = splbio();
3257 	mtx_lock(&vnode_free_list_mtx);
3258 	KASSERT((vp->v_iflag & VI_FREE) == 0, ("vnode already free"));
3259 	if (vp->v_iflag & VI_AGE) {
3260 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3261 	} else {
3262 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3263 	}
3264 	freevnodes++;
3265 	mtx_unlock(&vnode_free_list_mtx);
3266 	vp->v_iflag &= ~VI_AGE;
3267 	vp->v_iflag |= VI_FREE;
3268 	splx(s);
3269 }
3270 
3271 /*
3272  * Opposite of vfree() - mark a vnode as in use.
3273  */
3274 void
3275 vbusy(vp)
3276 	struct vnode *vp;
3277 {
3278 	int s;
3279 
3280 	s = splbio();
3281 	ASSERT_VI_LOCKED(vp, "vbusy");
3282 	KASSERT((vp->v_iflag & VI_FREE) != 0, ("vnode not free"));
3283 
3284 	mtx_lock(&vnode_free_list_mtx);
3285 	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3286 	freevnodes--;
3287 	mtx_unlock(&vnode_free_list_mtx);
3288 
3289 	vp->v_iflag &= ~(VI_FREE|VI_AGE);
3290 	splx(s);
3291 }
3292 
3293 /*
3294  * Record a process's interest in events which might happen to
3295  * a vnode.  Because poll uses the historic select-style interface
3296  * internally, this routine serves as both the ``check for any
3297  * pending events'' and the ``record my interest in future events''
3298  * functions.  (These are done together, while the lock is held,
3299  * to avoid race conditions.)
3300  */
3301 int
3302 vn_pollrecord(vp, td, events)
3303 	struct vnode *vp;
3304 	struct thread *td;
3305 	short events;
3306 {
3307 
3308 	if (vp->v_pollinfo == NULL)
3309 		v_addpollinfo(vp);
3310 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3311 	if (vp->v_pollinfo->vpi_revents & events) {
3312 		/*
3313 		 * This leaves events we are not interested
3314 		 * in available for the other process which
3315 		 * which presumably had requested them
3316 		 * (otherwise they would never have been
3317 		 * recorded).
3318 		 */
3319 		events &= vp->v_pollinfo->vpi_revents;
3320 		vp->v_pollinfo->vpi_revents &= ~events;
3321 
3322 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
3323 		return events;
3324 	}
3325 	vp->v_pollinfo->vpi_events |= events;
3326 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3327 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3328 	return 0;
3329 }
3330 
3331 /*
3332  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
3333  * it is possible for us to miss an event due to race conditions, but
3334  * that condition is expected to be rare, so for the moment it is the
3335  * preferred interface.
3336  */
3337 void
3338 vn_pollevent(vp, events)
3339 	struct vnode *vp;
3340 	short events;
3341 {
3342 
3343 	if (vp->v_pollinfo == NULL)
3344 		v_addpollinfo(vp);
3345 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3346 	if (vp->v_pollinfo->vpi_events & events) {
3347 		/*
3348 		 * We clear vpi_events so that we don't
3349 		 * call selwakeup() twice if two events are
3350 		 * posted before the polling process(es) is
3351 		 * awakened.  This also ensures that we take at
3352 		 * most one selwakeup() if the polling process
3353 		 * is no longer interested.  However, it does
3354 		 * mean that only one event can be noticed at
3355 		 * a time.  (Perhaps we should only clear those
3356 		 * event bits which we note?) XXX
3357 		 */
3358 		vp->v_pollinfo->vpi_events = 0;	/* &= ~events ??? */
3359 		vp->v_pollinfo->vpi_revents |= events;
3360 		selwakeup(&vp->v_pollinfo->vpi_selinfo);
3361 	}
3362 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3363 }
3364 
3365 /*
3366  * Wake up anyone polling on vp because it is being revoked.
3367  * This depends on dead_poll() returning POLLHUP for correct
3368  * behavior.
3369  */
3370 void
3371 vn_pollgone(vp)
3372 	struct vnode *vp;
3373 {
3374 
3375 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3376 	VN_KNOTE(vp, NOTE_REVOKE);
3377 	if (vp->v_pollinfo->vpi_events) {
3378 		vp->v_pollinfo->vpi_events = 0;
3379 		selwakeup(&vp->v_pollinfo->vpi_selinfo);
3380 	}
3381 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3382 }
3383 
3384 
3385 
3386 /*
3387  * Routine to create and manage a filesystem syncer vnode.
3388  */
3389 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
3390 static int	sync_fsync(struct  vop_fsync_args *);
3391 static int	sync_inactive(struct  vop_inactive_args *);
3392 static int	sync_reclaim(struct  vop_reclaim_args *);
3393 
3394 static vop_t **sync_vnodeop_p;
3395 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
3396 	{ &vop_default_desc,	(vop_t *) vop_eopnotsupp },
3397 	{ &vop_close_desc,	(vop_t *) sync_close },		/* close */
3398 	{ &vop_fsync_desc,	(vop_t *) sync_fsync },		/* fsync */
3399 	{ &vop_inactive_desc,	(vop_t *) sync_inactive },	/* inactive */
3400 	{ &vop_reclaim_desc,	(vop_t *) sync_reclaim },	/* reclaim */
3401 	{ &vop_lock_desc,	(vop_t *) vop_stdlock },	/* lock */
3402 	{ &vop_unlock_desc,	(vop_t *) vop_stdunlock },	/* unlock */
3403 	{ &vop_islocked_desc,	(vop_t *) vop_stdislocked },	/* islocked */
3404 	{ NULL, NULL }
3405 };
3406 static struct vnodeopv_desc sync_vnodeop_opv_desc =
3407 	{ &sync_vnodeop_p, sync_vnodeop_entries };
3408 
3409 VNODEOP_SET(sync_vnodeop_opv_desc);
3410 
3411 /*
3412  * Create a new filesystem syncer vnode for the specified mount point.
3413  */
3414 int
3415 vfs_allocate_syncvnode(mp)
3416 	struct mount *mp;
3417 {
3418 	struct vnode *vp;
3419 	static long start, incr, next;
3420 	int error;
3421 
3422 	/* Allocate a new vnode */
3423 	if ((error = getnewvnode("syncer", mp, sync_vnodeop_p, &vp)) != 0) {
3424 		mp->mnt_syncer = NULL;
3425 		return (error);
3426 	}
3427 	vp->v_type = VNON;
3428 	/*
3429 	 * Place the vnode onto the syncer worklist. We attempt to
3430 	 * scatter them about on the list so that they will go off
3431 	 * at evenly distributed times even if all the filesystems
3432 	 * are mounted at once.
3433 	 */
3434 	next += incr;
3435 	if (next == 0 || next > syncer_maxdelay) {
3436 		start /= 2;
3437 		incr /= 2;
3438 		if (start == 0) {
3439 			start = syncer_maxdelay / 2;
3440 			incr = syncer_maxdelay;
3441 		}
3442 		next = start;
3443 	}
3444 	VI_LOCK(vp);
3445 	vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
3446 	VI_UNLOCK(vp);
3447 	mp->mnt_syncer = vp;
3448 	return (0);
3449 }
3450 
3451 /*
3452  * Do a lazy sync of the filesystem.
3453  */
3454 static int
3455 sync_fsync(ap)
3456 	struct vop_fsync_args /* {
3457 		struct vnode *a_vp;
3458 		struct ucred *a_cred;
3459 		int a_waitfor;
3460 		struct thread *a_td;
3461 	} */ *ap;
3462 {
3463 	struct vnode *syncvp = ap->a_vp;
3464 	struct mount *mp = syncvp->v_mount;
3465 	struct thread *td = ap->a_td;
3466 	int error, asyncflag;
3467 
3468 	/*
3469 	 * We only need to do something if this is a lazy evaluation.
3470 	 */
3471 	if (ap->a_waitfor != MNT_LAZY)
3472 		return (0);
3473 
3474 	/*
3475 	 * Move ourselves to the back of the sync list.
3476 	 */
3477 	VI_LOCK(syncvp);
3478 	vn_syncer_add_to_worklist(syncvp, syncdelay);
3479 	VI_UNLOCK(syncvp);
3480 
3481 	/*
3482 	 * Walk the list of vnodes pushing all that are dirty and
3483 	 * not already on the sync list.
3484 	 */
3485 	mtx_lock(&mountlist_mtx);
3486 	if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) {
3487 		mtx_unlock(&mountlist_mtx);
3488 		return (0);
3489 	}
3490 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3491 		vfs_unbusy(mp, td);
3492 		return (0);
3493 	}
3494 	asyncflag = mp->mnt_flag & MNT_ASYNC;
3495 	mp->mnt_flag &= ~MNT_ASYNC;
3496 	vfs_msync(mp, MNT_NOWAIT);
3497 	error = VFS_SYNC(mp, MNT_LAZY, ap->a_cred, td);
3498 	if (asyncflag)
3499 		mp->mnt_flag |= MNT_ASYNC;
3500 	vn_finished_write(mp);
3501 	vfs_unbusy(mp, td);
3502 	return (error);
3503 }
3504 
3505 /*
3506  * The syncer vnode is no referenced.
3507  */
3508 static int
3509 sync_inactive(ap)
3510 	struct vop_inactive_args /* {
3511 		struct vnode *a_vp;
3512 		struct thread *a_td;
3513 	} */ *ap;
3514 {
3515 
3516 	VOP_UNLOCK(ap->a_vp, 0, ap->a_td);
3517 	vgone(ap->a_vp);
3518 	return (0);
3519 }
3520 
3521 /*
3522  * The syncer vnode is no longer needed and is being decommissioned.
3523  *
3524  * Modifications to the worklist must be protected at splbio().
3525  */
3526 static int
3527 sync_reclaim(ap)
3528 	struct vop_reclaim_args /* {
3529 		struct vnode *a_vp;
3530 	} */ *ap;
3531 {
3532 	struct vnode *vp = ap->a_vp;
3533 	int s;
3534 
3535 	s = splbio();
3536 	vp->v_mount->mnt_syncer = NULL;
3537 	VI_LOCK(vp);
3538 	if (vp->v_iflag & VI_ONWORKLST) {
3539 		mtx_lock(&sync_mtx);
3540 		LIST_REMOVE(vp, v_synclist);
3541 		mtx_unlock(&sync_mtx);
3542 		vp->v_iflag &= ~VI_ONWORKLST;
3543 	}
3544 	VI_UNLOCK(vp);
3545 	splx(s);
3546 
3547 	return (0);
3548 }
3549 
3550 /*
3551  * extract the dev_t from a VCHR
3552  */
3553 dev_t
3554 vn_todev(vp)
3555 	struct vnode *vp;
3556 {
3557 	if (vp->v_type != VCHR)
3558 		return (NODEV);
3559 	return (vp->v_rdev);
3560 }
3561 
3562 /*
3563  * Check if vnode represents a disk device
3564  */
3565 int
3566 vn_isdisk(vp, errp)
3567 	struct vnode *vp;
3568 	int *errp;
3569 {
3570 	struct cdevsw *cdevsw;
3571 
3572 	if (vp->v_type != VCHR) {
3573 		if (errp != NULL)
3574 			*errp = ENOTBLK;
3575 		return (0);
3576 	}
3577 	if (vp->v_rdev == NULL) {
3578 		if (errp != NULL)
3579 			*errp = ENXIO;
3580 		return (0);
3581 	}
3582 	cdevsw = devsw(vp->v_rdev);
3583 	if (cdevsw == NULL) {
3584 		if (errp != NULL)
3585 			*errp = ENXIO;
3586 		return (0);
3587 	}
3588 	if (!(cdevsw->d_flags & D_DISK)) {
3589 		if (errp != NULL)
3590 			*errp = ENOTBLK;
3591 		return (0);
3592 	}
3593 	if (errp != NULL)
3594 		*errp = 0;
3595 	return (1);
3596 }
3597 
3598 /*
3599  * Free data allocated by namei(); see namei(9) for details.
3600  */
3601 void
3602 NDFREE(ndp, flags)
3603      struct nameidata *ndp;
3604      const uint flags;
3605 {
3606 	if (!(flags & NDF_NO_FREE_PNBUF) &&
3607 	    (ndp->ni_cnd.cn_flags & HASBUF)) {
3608 		uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
3609 		ndp->ni_cnd.cn_flags &= ~HASBUF;
3610 	}
3611 	if (!(flags & NDF_NO_DVP_UNLOCK) &&
3612 	    (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
3613 	    ndp->ni_dvp != ndp->ni_vp)
3614 		VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_thread);
3615 	if (!(flags & NDF_NO_DVP_RELE) &&
3616 	    (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
3617 		vrele(ndp->ni_dvp);
3618 		ndp->ni_dvp = NULL;
3619 	}
3620 	if (!(flags & NDF_NO_VP_UNLOCK) &&
3621 	    (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
3622 		VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_thread);
3623 	if (!(flags & NDF_NO_VP_RELE) &&
3624 	    ndp->ni_vp) {
3625 		vrele(ndp->ni_vp);
3626 		ndp->ni_vp = NULL;
3627 	}
3628 	if (!(flags & NDF_NO_STARTDIR_RELE) &&
3629 	    (ndp->ni_cnd.cn_flags & SAVESTART)) {
3630 		vrele(ndp->ni_startdir);
3631 		ndp->ni_startdir = NULL;
3632 	}
3633 }
3634 
3635 /*
3636  * Common filesystem object access control check routine.  Accepts a
3637  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3638  * and optional call-by-reference privused argument allowing vaccess()
3639  * to indicate to the caller whether privilege was used to satisfy the
3640  * request (obsoleted).  Returns 0 on success, or an errno on failure.
3641  */
3642 int
3643 vaccess(type, file_mode, file_uid, file_gid, acc_mode, cred, privused)
3644 	enum vtype type;
3645 	mode_t file_mode;
3646 	uid_t file_uid;
3647 	gid_t file_gid;
3648 	mode_t acc_mode;
3649 	struct ucred *cred;
3650 	int *privused;
3651 {
3652 	mode_t dac_granted;
3653 #ifdef CAPABILITIES
3654 	mode_t cap_granted;
3655 #endif
3656 
3657 	/*
3658 	 * Look for a normal, non-privileged way to access the file/directory
3659 	 * as requested.  If it exists, go with that.
3660 	 */
3661 
3662 	if (privused != NULL)
3663 		*privused = 0;
3664 
3665 	dac_granted = 0;
3666 
3667 	/* Check the owner. */
3668 	if (cred->cr_uid == file_uid) {
3669 		dac_granted |= VADMIN;
3670 		if (file_mode & S_IXUSR)
3671 			dac_granted |= VEXEC;
3672 		if (file_mode & S_IRUSR)
3673 			dac_granted |= VREAD;
3674 		if (file_mode & S_IWUSR)
3675 			dac_granted |= (VWRITE | VAPPEND);
3676 
3677 		if ((acc_mode & dac_granted) == acc_mode)
3678 			return (0);
3679 
3680 		goto privcheck;
3681 	}
3682 
3683 	/* Otherwise, check the groups (first match) */
3684 	if (groupmember(file_gid, cred)) {
3685 		if (file_mode & S_IXGRP)
3686 			dac_granted |= VEXEC;
3687 		if (file_mode & S_IRGRP)
3688 			dac_granted |= VREAD;
3689 		if (file_mode & S_IWGRP)
3690 			dac_granted |= (VWRITE | VAPPEND);
3691 
3692 		if ((acc_mode & dac_granted) == acc_mode)
3693 			return (0);
3694 
3695 		goto privcheck;
3696 	}
3697 
3698 	/* Otherwise, check everyone else. */
3699 	if (file_mode & S_IXOTH)
3700 		dac_granted |= VEXEC;
3701 	if (file_mode & S_IROTH)
3702 		dac_granted |= VREAD;
3703 	if (file_mode & S_IWOTH)
3704 		dac_granted |= (VWRITE | VAPPEND);
3705 	if ((acc_mode & dac_granted) == acc_mode)
3706 		return (0);
3707 
3708 privcheck:
3709 	if (!suser_cred(cred, PRISON_ROOT)) {
3710 		/* XXX audit: privilege used */
3711 		if (privused != NULL)
3712 			*privused = 1;
3713 		return (0);
3714 	}
3715 
3716 #ifdef CAPABILITIES
3717 	/*
3718 	 * Build a capability mask to determine if the set of capabilities
3719 	 * satisfies the requirements when combined with the granted mask
3720 	 * from above.
3721 	 * For each capability, if the capability is required, bitwise
3722 	 * or the request type onto the cap_granted mask.
3723 	 */
3724 	cap_granted = 0;
3725 
3726 	if (type == VDIR) {
3727 		/*
3728 		 * For directories, use CAP_DAC_READ_SEARCH to satisfy
3729 		 * VEXEC requests, instead of CAP_DAC_EXECUTE.
3730 		 */
3731 		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3732 		    !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3733 			cap_granted |= VEXEC;
3734 	} else {
3735 		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3736 		    !cap_check(cred, NULL, CAP_DAC_EXECUTE, PRISON_ROOT))
3737 			cap_granted |= VEXEC;
3738 	}
3739 
3740 	if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
3741 	    !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3742 		cap_granted |= VREAD;
3743 
3744 	if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3745 	    !cap_check(cred, NULL, CAP_DAC_WRITE, PRISON_ROOT))
3746 		cap_granted |= (VWRITE | VAPPEND);
3747 
3748 	if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3749 	    !cap_check(cred, NULL, CAP_FOWNER, PRISON_ROOT))
3750 		cap_granted |= VADMIN;
3751 
3752 	if ((acc_mode & (cap_granted | dac_granted)) == acc_mode) {
3753 		/* XXX audit: privilege used */
3754 		if (privused != NULL)
3755 			*privused = 1;
3756 		return (0);
3757 	}
3758 #endif
3759 
3760 	return ((acc_mode & VADMIN) ? EPERM : EACCES);
3761 }
3762 
3763 /*
3764  * Credential check based on process requesting service, and per-attribute
3765  * permissions.
3766  */
3767 int
3768 extattr_check_cred(struct vnode *vp, int attrnamespace,
3769     struct ucred *cred, struct thread *td, int access)
3770 {
3771 
3772 	/*
3773 	 * Kernel-invoked always succeeds.
3774 	 */
3775 	if (cred == NOCRED)
3776 		return (0);
3777 
3778 	/*
3779 	 * Do not allow privileged processes in jail to directly
3780 	 * manipulate system attributes.
3781 	 *
3782 	 * XXX What capability should apply here?
3783 	 * Probably CAP_SYS_SETFFLAG.
3784 	 */
3785 	switch (attrnamespace) {
3786 	case EXTATTR_NAMESPACE_SYSTEM:
3787 		/* Potentially should be: return (EPERM); */
3788 		return (suser_cred(cred, 0));
3789 	case EXTATTR_NAMESPACE_USER:
3790 		return (VOP_ACCESS(vp, access, cred, td));
3791 	default:
3792 		return (EPERM);
3793 	}
3794 }
3795