xref: /freebsd/sys/kern/vfs_subr.c (revision 3193579b66fd7067f898dbc54bdea81a0e6f9bd0)
1 /*
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)vfs_subr.c	8.31 (Berkeley) 5/26/95
39  */
40 
41 /*
42  * External virtual filesystem routines
43  */
44 
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
47 
48 #include "opt_ddb.h"
49 #include "opt_mac.h"
50 
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/bio.h>
54 #include <sys/buf.h>
55 #include <sys/conf.h>
56 #include <sys/eventhandler.h>
57 #include <sys/extattr.h>
58 #include <sys/fcntl.h>
59 #include <sys/kernel.h>
60 #include <sys/kthread.h>
61 #include <sys/mac.h>
62 #include <sys/malloc.h>
63 #include <sys/mount.h>
64 #include <sys/namei.h>
65 #include <sys/stat.h>
66 #include <sys/sysctl.h>
67 #include <sys/syslog.h>
68 #include <sys/vmmeter.h>
69 #include <sys/vnode.h>
70 
71 #include <vm/vm.h>
72 #include <vm/vm_object.h>
73 #include <vm/vm_extern.h>
74 #include <vm/pmap.h>
75 #include <vm/vm_map.h>
76 #include <vm/vm_page.h>
77 #include <vm/vm_kern.h>
78 #include <vm/uma.h>
79 
80 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
81 
82 static void	addalias(struct vnode *vp, dev_t nvp_rdev);
83 static void	insmntque(struct vnode *vp, struct mount *mp);
84 static void	vclean(struct vnode *vp, int flags, struct thread *td);
85 static void	vlruvp(struct vnode *vp);
86 static int	flushbuflist(struct buf *blist, int flags, struct vnode *vp,
87 		    int slpflag, int slptimeo, int *errorp);
88 static int	vtryrecycle(struct vnode *vp);
89 static void	vx_lock(struct vnode *vp);
90 static void	vx_unlock(struct vnode *vp);
91 static void	vgonechrl(struct vnode *vp, struct thread *td);
92 
93 
94 /*
95  * Number of vnodes in existence.  Increased whenever getnewvnode()
96  * allocates a new vnode, never decreased.
97  */
98 static unsigned long	numvnodes;
99 
100 SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
101 
102 /*
103  * Conversion tables for conversion from vnode types to inode formats
104  * and back.
105  */
106 enum vtype iftovt_tab[16] = {
107 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
108 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
109 };
110 int vttoif_tab[9] = {
111 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
112 	S_IFSOCK, S_IFIFO, S_IFMT,
113 };
114 
115 /*
116  * List of vnodes that are ready for recycling.
117  */
118 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
119 
120 /*
121  * Minimum number of free vnodes.  If there are fewer than this free vnodes,
122  * getnewvnode() will return a newly allocated vnode.
123  */
124 static u_long wantfreevnodes = 25;
125 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
126 /* Number of vnodes in the free list. */
127 static u_long freevnodes;
128 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
129 
130 /*
131  * Various variables used for debugging the new implementation of
132  * reassignbuf().
133  * XXX these are probably of (very) limited utility now.
134  */
135 static int reassignbufcalls;
136 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
137 static int nameileafonly;
138 SYSCTL_INT(_vfs, OID_AUTO, nameileafonly, CTLFLAG_RW, &nameileafonly, 0, "");
139 
140 /*
141  * Cache for the mount type id assigned to NFS.  This is used for
142  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
143  */
144 int	nfs_mount_type = -1;
145 
146 /* To keep more than one thread at a time from running vfs_getnewfsid */
147 static struct mtx mntid_mtx;
148 
149 /*
150  * Lock for any access to the following:
151  *	vnode_free_list
152  *	numvnodes
153  *	freevnodes
154  */
155 static struct mtx vnode_free_list_mtx;
156 
157 /*
158  * For any iteration/modification of dev->si_hlist (linked through
159  * v_specnext)
160  */
161 static struct mtx spechash_mtx;
162 
163 /* Publicly exported FS */
164 struct nfs_public nfs_pub;
165 
166 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
167 static uma_zone_t vnode_zone;
168 static uma_zone_t vnodepoll_zone;
169 
170 /* Set to 1 to print out reclaim of active vnodes */
171 int	prtactive;
172 
173 /*
174  * The workitem queue.
175  *
176  * It is useful to delay writes of file data and filesystem metadata
177  * for tens of seconds so that quickly created and deleted files need
178  * not waste disk bandwidth being created and removed. To realize this,
179  * we append vnodes to a "workitem" queue. When running with a soft
180  * updates implementation, most pending metadata dependencies should
181  * not wait for more than a few seconds. Thus, mounted on block devices
182  * are delayed only about a half the time that file data is delayed.
183  * Similarly, directory updates are more critical, so are only delayed
184  * about a third the time that file data is delayed. Thus, there are
185  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
186  * one each second (driven off the filesystem syncer process). The
187  * syncer_delayno variable indicates the next queue that is to be processed.
188  * Items that need to be processed soon are placed in this queue:
189  *
190  *	syncer_workitem_pending[syncer_delayno]
191  *
192  * A delay of fifteen seconds is done by placing the request fifteen
193  * entries later in the queue:
194  *
195  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
196  *
197  */
198 static int syncer_delayno;
199 static long syncer_mask;
200 LIST_HEAD(synclist, vnode);
201 static struct synclist *syncer_workitem_pending;
202 /*
203  * The sync_mtx protects:
204  *	vp->v_synclist
205  *	syncer_delayno
206  *	syncer_workitem_pending
207  *	rushjob
208  */
209 static struct mtx sync_mtx;
210 
211 #define SYNCER_MAXDELAY		32
212 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
213 static int syncdelay = 30;		/* max time to delay syncing data */
214 static int filedelay = 30;		/* time to delay syncing files */
215 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
216 static int dirdelay = 29;		/* time to delay syncing directories */
217 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
218 static int metadelay = 28;		/* time to delay syncing metadata */
219 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
220 static int rushjob;		/* number of slots to run ASAP */
221 static int stat_rush_requests;	/* number of times I/O speeded up */
222 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
223 
224 /*
225  * Number of vnodes we want to exist at any one time.  This is mostly used
226  * to size hash tables in vnode-related code.  It is normally not used in
227  * getnewvnode(), as wantfreevnodes is normally nonzero.)
228  *
229  * XXX desiredvnodes is historical cruft and should not exist.
230  */
231 int desiredvnodes;
232 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
233     &desiredvnodes, 0, "Maximum number of vnodes");
234 static int minvnodes;
235 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
236     &minvnodes, 0, "Minimum number of vnodes");
237 static int vnlru_nowhere;
238 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0,
239     "Number of times the vnlru process ran without success");
240 
241 /* Hook for calling soft updates */
242 int (*softdep_process_worklist_hook)(struct mount *);
243 
244 /*
245  * This only exists to supress warnings from unlocked specfs accesses.  It is
246  * no longer ok to have an unlocked VFS.
247  */
248 #define IGNORE_LOCK(vp) ((vp)->v_type == VCHR || (vp)->v_type == VBAD)
249 
250 /* Print lock violations */
251 int vfs_badlock_print = 1;
252 
253 /* Panic on violation */
254 int vfs_badlock_panic = 1;
255 
256 /* Check for interlock across VOPs */
257 int vfs_badlock_mutex = 1;
258 
259 static void
260 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
261 {
262 	if (vfs_badlock_print)
263 		printf("%s: %p %s\n", str, vp, msg);
264 	if (vfs_badlock_panic)
265 		Debugger("Lock violation.\n");
266 }
267 
268 void
269 assert_vi_unlocked(struct vnode *vp, const char *str)
270 {
271 	if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
272 		vfs_badlock("interlock is locked but should not be", str, vp);
273 }
274 
275 void
276 assert_vi_locked(struct vnode *vp, const char *str)
277 {
278 	if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
279 		vfs_badlock("interlock is not locked but should be", str, vp);
280 }
281 
282 void
283 assert_vop_locked(struct vnode *vp, const char *str)
284 {
285 	if (vp && !IGNORE_LOCK(vp) && !VOP_ISLOCKED(vp, NULL))
286 		vfs_badlock("is not locked but should be", str, vp);
287 }
288 
289 void
290 assert_vop_unlocked(struct vnode *vp, const char *str)
291 {
292 	if (vp && !IGNORE_LOCK(vp) &&
293 	    VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE)
294 		vfs_badlock("is locked but should not be", str, vp);
295 }
296 
297 void
298 assert_vop_elocked(struct vnode *vp, const char *str)
299 {
300 	if (vp && !IGNORE_LOCK(vp) &&
301 	    VOP_ISLOCKED(vp, curthread) != LK_EXCLUSIVE)
302 		vfs_badlock("is not exclusive locked but should be", str, vp);
303 }
304 
305 void
306 assert_vop_elocked_other(struct vnode *vp, const char *str)
307 {
308 	if (vp && !IGNORE_LOCK(vp) &&
309 	    VOP_ISLOCKED(vp, curthread) != LK_EXCLOTHER)
310 		vfs_badlock("is not exclusive locked by another thread",
311 		    str, vp);
312 }
313 
314 void
315 assert_vop_slocked(struct vnode *vp, const char *str)
316 {
317 	if (vp && !IGNORE_LOCK(vp) &&
318 	    VOP_ISLOCKED(vp, curthread) != LK_SHARED)
319 		vfs_badlock("is not locked shared but should be", str, vp);
320 }
321 
322 void
323 vop_rename_pre(void *ap)
324 {
325 	struct vop_rename_args *a = ap;
326 
327 	if (a->a_tvp)
328 		ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
329 	ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
330 	ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
331 	ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
332 
333 	/* Check the source (from) */
334 	if (a->a_tdvp != a->a_fdvp)
335 		ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked.\n");
336 	if (a->a_tvp != a->a_fvp)
337 		ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: tvp locked.\n");
338 
339 	/* Check the target */
340 	if (a->a_tvp)
341 		ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked.\n");
342 
343 	ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked.\n");
344 }
345 
346 void
347 vop_strategy_pre(void *ap)
348 {
349 	struct vop_strategy_args *a = ap;
350 	struct buf *bp;
351 
352 	bp = a->a_bp;
353 
354 	/*
355 	 * Cluster ops lock their component buffers but not the IO container.
356 	 */
357 	if ((bp->b_flags & B_CLUSTER) != 0)
358 		return;
359 
360 	if (BUF_REFCNT(bp) < 1) {
361 		if (vfs_badlock_print)
362 			printf("VOP_STRATEGY: bp is not locked but should be.\n");
363 		if (vfs_badlock_panic)
364 			Debugger("Lock violation.\n");
365 	}
366 }
367 
368 void
369 vop_lookup_pre(void *ap)
370 {
371 	struct vop_lookup_args *a = ap;
372 	struct vnode *dvp;
373 
374 	dvp = a->a_dvp;
375 
376 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
377 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
378 }
379 
380 void
381 vop_lookup_post(void *ap, int rc)
382 {
383 	struct vop_lookup_args *a = ap;
384 	struct componentname *cnp;
385 	struct vnode *dvp;
386 	struct vnode *vp;
387 	int flags;
388 
389 	dvp = a->a_dvp;
390 	cnp = a->a_cnp;
391 	vp = *(a->a_vpp);
392 	flags = cnp->cn_flags;
393 
394 
395 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
396 	/*
397 	 * If this is the last path component for this lookup and LOCPARENT
398 	 * is set, OR if there is an error the directory has to be locked.
399 	 */
400 	if ((flags & LOCKPARENT) && (flags & ISLASTCN))
401 		ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP (LOCKPARENT)");
402 	else if (rc != 0)
403 		ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP (error)");
404 	else if (dvp != vp)
405 		ASSERT_VOP_UNLOCKED(dvp, "VOP_LOOKUP (dvp)");
406 
407 	if (flags & PDIRUNLOCK)
408 		ASSERT_VOP_UNLOCKED(dvp, "VOP_LOOKUP (PDIRUNLOCK)");
409 }
410 
411 void
412 vop_unlock_pre(void *ap)
413 {
414 	struct vop_unlock_args *a = ap;
415 
416 	if (a->a_flags & LK_INTERLOCK)
417 		ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
418 
419 	ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
420 }
421 
422 void
423 vop_unlock_post(void *ap, int rc)
424 {
425 	struct vop_unlock_args *a = ap;
426 
427 	if (a->a_flags & LK_INTERLOCK)
428 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
429 }
430 
431 void
432 vop_lock_pre(void *ap)
433 {
434 	struct vop_lock_args *a = ap;
435 
436 	if ((a->a_flags & LK_INTERLOCK) == 0)
437 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
438 	else
439 		ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
440 }
441 
442 void
443 vop_lock_post(void *ap, int rc)
444 {
445 	struct vop_lock_args *a;
446 
447 	a = ap;
448 
449 	ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
450 	if (rc == 0)
451 		ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
452 }
453 
454 void
455 v_addpollinfo(struct vnode *vp)
456 {
457 	vp->v_pollinfo = uma_zalloc(vnodepoll_zone, M_WAITOK);
458 	mtx_init(&vp->v_pollinfo->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
459 }
460 
461 /*
462  * Initialize the vnode management data structures.
463  */
464 static void
465 vntblinit(void *dummy __unused)
466 {
467 
468 	/*
469 	 * Desiredvnodes is a function of the physical memory size and
470 	 * the kernel's heap size.  Specifically, desiredvnodes scales
471 	 * in proportion to the physical memory size until two fifths
472 	 * of the kernel's heap size is consumed by vnodes and vm
473 	 * objects.
474 	 */
475 	desiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * vm_kmem_size /
476 	    (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
477 	minvnodes = desiredvnodes / 4;
478 	mtx_init(&mountlist_mtx, "mountlist", NULL, MTX_DEF);
479 	mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
480 	mtx_init(&spechash_mtx, "spechash", NULL, MTX_DEF);
481 	TAILQ_INIT(&vnode_free_list);
482 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
483 	vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
484 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
485 	vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
486 	      NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
487 	/*
488 	 * Initialize the filesystem syncer.
489 	 */
490 	syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
491 		&syncer_mask);
492 	syncer_maxdelay = syncer_mask + 1;
493 	mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
494 }
495 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL)
496 
497 
498 /*
499  * Mark a mount point as busy. Used to synchronize access and to delay
500  * unmounting. Interlock is not released on failure.
501  */
502 int
503 vfs_busy(mp, flags, interlkp, td)
504 	struct mount *mp;
505 	int flags;
506 	struct mtx *interlkp;
507 	struct thread *td;
508 {
509 	int lkflags;
510 
511 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
512 		if (flags & LK_NOWAIT)
513 			return (ENOENT);
514 		mp->mnt_kern_flag |= MNTK_MWAIT;
515 		/*
516 		 * Since all busy locks are shared except the exclusive
517 		 * lock granted when unmounting, the only place that a
518 		 * wakeup needs to be done is at the release of the
519 		 * exclusive lock at the end of dounmount.
520 		 */
521 		msleep(mp, interlkp, PVFS, "vfs_busy", 0);
522 		return (ENOENT);
523 	}
524 	lkflags = LK_SHARED | LK_NOPAUSE;
525 	if (interlkp)
526 		lkflags |= LK_INTERLOCK;
527 	if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
528 		panic("vfs_busy: unexpected lock failure");
529 	return (0);
530 }
531 
532 /*
533  * Free a busy filesystem.
534  */
535 void
536 vfs_unbusy(mp, td)
537 	struct mount *mp;
538 	struct thread *td;
539 {
540 
541 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
542 }
543 
544 /*
545  * Lookup a mount point by filesystem identifier.
546  */
547 struct mount *
548 vfs_getvfs(fsid)
549 	fsid_t *fsid;
550 {
551 	register struct mount *mp;
552 
553 	mtx_lock(&mountlist_mtx);
554 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
555 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
556 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
557 			mtx_unlock(&mountlist_mtx);
558 			return (mp);
559 		}
560 	}
561 	mtx_unlock(&mountlist_mtx);
562 	return ((struct mount *) 0);
563 }
564 
565 /*
566  * Get a new unique fsid.  Try to make its val[0] unique, since this value
567  * will be used to create fake device numbers for stat().  Also try (but
568  * not so hard) make its val[0] unique mod 2^16, since some emulators only
569  * support 16-bit device numbers.  We end up with unique val[0]'s for the
570  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
571  *
572  * Keep in mind that several mounts may be running in parallel.  Starting
573  * the search one past where the previous search terminated is both a
574  * micro-optimization and a defense against returning the same fsid to
575  * different mounts.
576  */
577 void
578 vfs_getnewfsid(mp)
579 	struct mount *mp;
580 {
581 	static u_int16_t mntid_base;
582 	fsid_t tfsid;
583 	int mtype;
584 
585 	mtx_lock(&mntid_mtx);
586 	mtype = mp->mnt_vfc->vfc_typenum;
587 	tfsid.val[1] = mtype;
588 	mtype = (mtype & 0xFF) << 24;
589 	for (;;) {
590 		tfsid.val[0] = makeudev(255,
591 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
592 		mntid_base++;
593 		if (vfs_getvfs(&tfsid) == NULL)
594 			break;
595 	}
596 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
597 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
598 	mtx_unlock(&mntid_mtx);
599 }
600 
601 /*
602  * Knob to control the precision of file timestamps:
603  *
604  *   0 = seconds only; nanoseconds zeroed.
605  *   1 = seconds and nanoseconds, accurate within 1/HZ.
606  *   2 = seconds and nanoseconds, truncated to microseconds.
607  * >=3 = seconds and nanoseconds, maximum precision.
608  */
609 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
610 
611 static int timestamp_precision = TSP_SEC;
612 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
613     &timestamp_precision, 0, "");
614 
615 /*
616  * Get a current timestamp.
617  */
618 void
619 vfs_timestamp(tsp)
620 	struct timespec *tsp;
621 {
622 	struct timeval tv;
623 
624 	switch (timestamp_precision) {
625 	case TSP_SEC:
626 		tsp->tv_sec = time_second;
627 		tsp->tv_nsec = 0;
628 		break;
629 	case TSP_HZ:
630 		getnanotime(tsp);
631 		break;
632 	case TSP_USEC:
633 		microtime(&tv);
634 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
635 		break;
636 	case TSP_NSEC:
637 	default:
638 		nanotime(tsp);
639 		break;
640 	}
641 }
642 
643 /*
644  * Set vnode attributes to VNOVAL
645  */
646 void
647 vattr_null(vap)
648 	register struct vattr *vap;
649 {
650 
651 	vap->va_type = VNON;
652 	vap->va_size = VNOVAL;
653 	vap->va_bytes = VNOVAL;
654 	vap->va_mode = VNOVAL;
655 	vap->va_nlink = VNOVAL;
656 	vap->va_uid = VNOVAL;
657 	vap->va_gid = VNOVAL;
658 	vap->va_fsid = VNOVAL;
659 	vap->va_fileid = VNOVAL;
660 	vap->va_blocksize = VNOVAL;
661 	vap->va_rdev = VNOVAL;
662 	vap->va_atime.tv_sec = VNOVAL;
663 	vap->va_atime.tv_nsec = VNOVAL;
664 	vap->va_mtime.tv_sec = VNOVAL;
665 	vap->va_mtime.tv_nsec = VNOVAL;
666 	vap->va_ctime.tv_sec = VNOVAL;
667 	vap->va_ctime.tv_nsec = VNOVAL;
668 	vap->va_birthtime.tv_sec = VNOVAL;
669 	vap->va_birthtime.tv_nsec = VNOVAL;
670 	vap->va_flags = VNOVAL;
671 	vap->va_gen = VNOVAL;
672 	vap->va_vaflags = 0;
673 }
674 
675 /*
676  * This routine is called when we have too many vnodes.  It attempts
677  * to free <count> vnodes and will potentially free vnodes that still
678  * have VM backing store (VM backing store is typically the cause
679  * of a vnode blowout so we want to do this).  Therefore, this operation
680  * is not considered cheap.
681  *
682  * A number of conditions may prevent a vnode from being reclaimed.
683  * the buffer cache may have references on the vnode, a directory
684  * vnode may still have references due to the namei cache representing
685  * underlying files, or the vnode may be in active use.   It is not
686  * desireable to reuse such vnodes.  These conditions may cause the
687  * number of vnodes to reach some minimum value regardless of what
688  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
689  */
690 static int
691 vlrureclaim(struct mount *mp)
692 {
693 	struct vnode *vp;
694 	int done;
695 	int trigger;
696 	int usevnodes;
697 	int count;
698 
699 	/*
700 	 * Calculate the trigger point, don't allow user
701 	 * screwups to blow us up.   This prevents us from
702 	 * recycling vnodes with lots of resident pages.  We
703 	 * aren't trying to free memory, we are trying to
704 	 * free vnodes.
705 	 */
706 	usevnodes = desiredvnodes;
707 	if (usevnodes <= 0)
708 		usevnodes = 1;
709 	trigger = cnt.v_page_count * 2 / usevnodes;
710 
711 	done = 0;
712 	MNT_ILOCK(mp);
713 	count = mp->mnt_nvnodelistsize / 10 + 1;
714 	while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
715 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
716 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
717 
718 		if (vp->v_type != VNON &&
719 		    vp->v_type != VBAD &&
720 		    VI_TRYLOCK(vp)) {
721 			if (VMIGHTFREE(vp) &&           /* critical path opt */
722 			    (vp->v_object == NULL ||
723 			    vp->v_object->resident_page_count < trigger)) {
724 				MNT_IUNLOCK(mp);
725 				vgonel(vp, curthread);
726 				done++;
727 				MNT_ILOCK(mp);
728 			} else
729 				VI_UNLOCK(vp);
730 		}
731 		--count;
732 	}
733 	MNT_IUNLOCK(mp);
734 	return done;
735 }
736 
737 /*
738  * Attempt to recycle vnodes in a context that is always safe to block.
739  * Calling vlrurecycle() from the bowels of filesystem code has some
740  * interesting deadlock problems.
741  */
742 static struct proc *vnlruproc;
743 static int vnlruproc_sig;
744 
745 static void
746 vnlru_proc(void)
747 {
748 	struct mount *mp, *nmp;
749 	int done;
750 	struct proc *p = vnlruproc;
751 	struct thread *td = FIRST_THREAD_IN_PROC(p);	/* XXXKSE */
752 
753 	mtx_lock(&Giant);
754 
755 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
756 	    SHUTDOWN_PRI_FIRST);
757 
758 	for (;;) {
759 		kthread_suspend_check(p);
760 		mtx_lock(&vnode_free_list_mtx);
761 		if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
762 			mtx_unlock(&vnode_free_list_mtx);
763 			vnlruproc_sig = 0;
764 			wakeup(&vnlruproc_sig);
765 			tsleep(vnlruproc, PVFS, "vlruwt", hz);
766 			continue;
767 		}
768 		mtx_unlock(&vnode_free_list_mtx);
769 		done = 0;
770 		mtx_lock(&mountlist_mtx);
771 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
772 			if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
773 				nmp = TAILQ_NEXT(mp, mnt_list);
774 				continue;
775 			}
776 			done += vlrureclaim(mp);
777 			mtx_lock(&mountlist_mtx);
778 			nmp = TAILQ_NEXT(mp, mnt_list);
779 			vfs_unbusy(mp, td);
780 		}
781 		mtx_unlock(&mountlist_mtx);
782 		if (done == 0) {
783 #if 0
784 			/* These messages are temporary debugging aids */
785 			if (vnlru_nowhere < 5)
786 				printf("vnlru process getting nowhere..\n");
787 			else if (vnlru_nowhere == 5)
788 				printf("vnlru process messages stopped.\n");
789 #endif
790 			vnlru_nowhere++;
791 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
792 		}
793 	}
794 }
795 
796 static struct kproc_desc vnlru_kp = {
797 	"vnlru",
798 	vnlru_proc,
799 	&vnlruproc
800 };
801 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
802 
803 
804 /*
805  * Routines having to do with the management of the vnode table.
806  */
807 
808 /*
809  * Check to see if a free vnode can be recycled. If it can,
810  * recycle it and return it with the vnode interlock held.
811  */
812 static int
813 vtryrecycle(struct vnode *vp)
814 {
815 	struct thread *td = curthread;
816 	vm_object_t object;
817 	struct mount *vnmp;
818 	int error;
819 
820 	/* Don't recycle if we can't get the interlock */
821 	if (!VI_TRYLOCK(vp))
822 		return (EWOULDBLOCK);
823 	/*
824 	 * This vnode may found and locked via some other list, if so we
825 	 * can't recycle it yet.
826 	 */
827 	if (vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE | LK_NOWAIT, td) != 0)
828 		return (EWOULDBLOCK);
829 	/*
830 	 * Don't recycle if its filesystem is being suspended.
831 	 */
832 	if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
833 		error = EBUSY;
834 		goto done;
835 	}
836 
837 	/*
838 	 * Don't recycle if we still have cached pages.
839 	 */
840 	if (VOP_GETVOBJECT(vp, &object) == 0) {
841 		VM_OBJECT_LOCK(object);
842 		if (object->resident_page_count ||
843 		    object->ref_count) {
844 			VM_OBJECT_UNLOCK(object);
845 			error = EBUSY;
846 			goto done;
847 		}
848 		VM_OBJECT_UNLOCK(object);
849 	}
850 	if (LIST_FIRST(&vp->v_cache_src)) {
851 		/*
852 		 * note: nameileafonly sysctl is temporary,
853 		 * for debugging only, and will eventually be
854 		 * removed.
855 		 */
856 		if (nameileafonly > 0) {
857 			/*
858 			 * Do not reuse namei-cached directory
859 			 * vnodes that have cached
860 			 * subdirectories.
861 			 */
862 			if (cache_leaf_test(vp) < 0) {
863 				error = EISDIR;
864 				goto done;
865 			}
866 		} else if (nameileafonly < 0 ||
867 			    vmiodirenable == 0) {
868 			/*
869 			 * Do not reuse namei-cached directory
870 			 * vnodes if nameileafonly is -1 or
871 			 * if VMIO backing for directories is
872 			 * turned off (otherwise we reuse them
873 			 * too quickly).
874 			 */
875 			error = EBUSY;
876 			goto done;
877 		}
878 	}
879 	/*
880 	 * If we got this far, we need to acquire the interlock and see if
881 	 * anyone picked up this vnode from another list.  If not, we will
882 	 * mark it with XLOCK via vgonel() so that anyone who does find it
883 	 * will skip over it.
884 	 */
885 	VI_LOCK(vp);
886 	if (VSHOULDBUSY(vp) && (vp->v_iflag & VI_XLOCK) == 0) {
887 		VI_UNLOCK(vp);
888 		error = EBUSY;
889 		goto done;
890 	}
891 	mtx_lock(&vnode_free_list_mtx);
892 	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
893 	vp->v_iflag &= ~VI_FREE;
894 	mtx_unlock(&vnode_free_list_mtx);
895 	vp->v_iflag |= VI_DOOMED;
896 	if (vp->v_type != VBAD) {
897 		VOP_UNLOCK(vp, 0, td);
898 		vgonel(vp, td);
899 		VI_LOCK(vp);
900 	} else
901 		VOP_UNLOCK(vp, 0, td);
902 	vn_finished_write(vnmp);
903 	return (0);
904 done:
905 	VOP_UNLOCK(vp, 0, td);
906 	return (error);
907 }
908 
909 /*
910  * Return the next vnode from the free list.
911  */
912 int
913 getnewvnode(tag, mp, vops, vpp)
914 	const char *tag;
915 	struct mount *mp;
916 	vop_t **vops;
917 	struct vnode **vpp;
918 {
919 	struct vnode *vp = NULL;
920 	struct vpollinfo *pollinfo = NULL;
921 
922 	mtx_lock(&vnode_free_list_mtx);
923 
924 	/*
925 	 * Try to reuse vnodes if we hit the max.  This situation only
926 	 * occurs in certain large-memory (2G+) situations.  We cannot
927 	 * attempt to directly reclaim vnodes due to nasty recursion
928 	 * problems.
929 	 */
930 	while (numvnodes - freevnodes > desiredvnodes) {
931 		if (vnlruproc_sig == 0) {
932 			vnlruproc_sig = 1;      /* avoid unnecessary wakeups */
933 			wakeup(vnlruproc);
934 		}
935 		mtx_unlock(&vnode_free_list_mtx);
936 		tsleep(&vnlruproc_sig, PVFS, "vlruwk", hz);
937 		mtx_lock(&vnode_free_list_mtx);
938 	}
939 
940 	/*
941 	 * Attempt to reuse a vnode already on the free list, allocating
942 	 * a new vnode if we can't find one or if we have not reached a
943 	 * good minimum for good LRU performance.
944 	 */
945 
946 	if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
947 		int error;
948 		int count;
949 
950 		for (count = 0; count < freevnodes; count++) {
951 			vp = TAILQ_FIRST(&vnode_free_list);
952 
953 			KASSERT(vp->v_usecount == 0 &&
954 			    (vp->v_iflag & VI_DOINGINACT) == 0,
955 			    ("getnewvnode: free vnode isn't"));
956 
957 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
958 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
959 			mtx_unlock(&vnode_free_list_mtx);
960 			error = vtryrecycle(vp);
961 			mtx_lock(&vnode_free_list_mtx);
962 			if (error == 0)
963 				break;
964 			vp = NULL;
965 		}
966 	}
967 	if (vp) {
968 		freevnodes--;
969 		mtx_unlock(&vnode_free_list_mtx);
970 
971 #ifdef INVARIANTS
972 		{
973 			if (vp->v_data)
974 				panic("cleaned vnode isn't");
975 			if (vp->v_numoutput)
976 				panic("Clean vnode has pending I/O's");
977 			if (vp->v_writecount != 0)
978 				panic("Non-zero write count");
979 		}
980 #endif
981 		if ((pollinfo = vp->v_pollinfo) != NULL) {
982 			/*
983 			 * To avoid lock order reversals, the call to
984 			 * uma_zfree() must be delayed until the vnode
985 			 * interlock is released.
986 			 */
987 			vp->v_pollinfo = NULL;
988 		}
989 #ifdef MAC
990 		mac_destroy_vnode(vp);
991 #endif
992 		vp->v_iflag = 0;
993 		vp->v_vflag = 0;
994 		vp->v_lastw = 0;
995 		vp->v_lasta = 0;
996 		vp->v_cstart = 0;
997 		vp->v_clen = 0;
998 		vp->v_socket = 0;
999 		lockdestroy(vp->v_vnlock);
1000 		lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOPAUSE);
1001 		KASSERT(vp->v_cleanbufcnt == 0, ("cleanbufcnt not 0"));
1002 		KASSERT(vp->v_cleanblkroot == NULL, ("cleanblkroot not NULL"));
1003 		KASSERT(vp->v_dirtybufcnt == 0, ("dirtybufcnt not 0"));
1004 		KASSERT(vp->v_dirtyblkroot == NULL, ("dirtyblkroot not NULL"));
1005 	} else {
1006 		numvnodes++;
1007 		mtx_unlock(&vnode_free_list_mtx);
1008 
1009 		vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
1010 		mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
1011 		VI_LOCK(vp);
1012 		vp->v_dd = vp;
1013 		vp->v_vnlock = &vp->v_lock;
1014 		lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOPAUSE);
1015 		cache_purge(vp);		/* Sets up v_id. */
1016 		LIST_INIT(&vp->v_cache_src);
1017 		TAILQ_INIT(&vp->v_cache_dst);
1018 	}
1019 
1020 	TAILQ_INIT(&vp->v_cleanblkhd);
1021 	TAILQ_INIT(&vp->v_dirtyblkhd);
1022 	vp->v_type = VNON;
1023 	vp->v_tag = tag;
1024 	vp->v_op = vops;
1025 	*vpp = vp;
1026 	vp->v_usecount = 1;
1027 	vp->v_data = 0;
1028 	vp->v_cachedid = -1;
1029 	VI_UNLOCK(vp);
1030 	if (pollinfo != NULL) {
1031 		mtx_destroy(&pollinfo->vpi_lock);
1032 		uma_zfree(vnodepoll_zone, pollinfo);
1033 	}
1034 #ifdef MAC
1035 	mac_init_vnode(vp);
1036 	if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1037 		mac_associate_vnode_singlelabel(mp, vp);
1038 #endif
1039 	insmntque(vp, mp);
1040 
1041 	return (0);
1042 }
1043 
1044 /*
1045  * Move a vnode from one mount queue to another.
1046  */
1047 static void
1048 insmntque(vp, mp)
1049 	register struct vnode *vp;
1050 	register struct mount *mp;
1051 {
1052 
1053 	/*
1054 	 * Delete from old mount point vnode list, if on one.
1055 	 */
1056 	if (vp->v_mount != NULL) {
1057 		MNT_ILOCK(vp->v_mount);
1058 		KASSERT(vp->v_mount->mnt_nvnodelistsize > 0,
1059 			("bad mount point vnode list size"));
1060 		TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
1061 		vp->v_mount->mnt_nvnodelistsize--;
1062 		MNT_IUNLOCK(vp->v_mount);
1063 	}
1064 	/*
1065 	 * Insert into list of vnodes for the new mount point, if available.
1066 	 */
1067 	if ((vp->v_mount = mp) != NULL) {
1068 		MNT_ILOCK(vp->v_mount);
1069 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1070 		mp->mnt_nvnodelistsize++;
1071 		MNT_IUNLOCK(vp->v_mount);
1072 	}
1073 }
1074 
1075 /*
1076  * Update outstanding I/O count and do wakeup if requested.
1077  */
1078 void
1079 vwakeup(bp)
1080 	register struct buf *bp;
1081 {
1082 	register struct vnode *vp;
1083 
1084 	bp->b_flags &= ~B_WRITEINPROG;
1085 	if ((vp = bp->b_vp)) {
1086 		VI_LOCK(vp);
1087 		vp->v_numoutput--;
1088 		if (vp->v_numoutput < 0)
1089 			panic("vwakeup: neg numoutput");
1090 		if ((vp->v_numoutput == 0) && (vp->v_iflag & VI_BWAIT)) {
1091 			vp->v_iflag &= ~VI_BWAIT;
1092 			wakeup(&vp->v_numoutput);
1093 		}
1094 		VI_UNLOCK(vp);
1095 	}
1096 }
1097 
1098 /*
1099  * Flush out and invalidate all buffers associated with a vnode.
1100  * Called with the underlying object locked.
1101  */
1102 int
1103 vinvalbuf(vp, flags, cred, td, slpflag, slptimeo)
1104 	struct vnode *vp;
1105 	int flags;
1106 	struct ucred *cred;
1107 	struct thread *td;
1108 	int slpflag, slptimeo;
1109 {
1110 	struct buf *blist;
1111 	int error;
1112 	vm_object_t object;
1113 
1114 	GIANT_REQUIRED;
1115 
1116 	ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1117 
1118 	VI_LOCK(vp);
1119 	if (flags & V_SAVE) {
1120 		while (vp->v_numoutput) {
1121 			vp->v_iflag |= VI_BWAIT;
1122 			error = msleep(&vp->v_numoutput, VI_MTX(vp),
1123 			    slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo);
1124 			if (error) {
1125 				VI_UNLOCK(vp);
1126 				return (error);
1127 			}
1128 		}
1129 		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1130 			VI_UNLOCK(vp);
1131 			if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, td)) != 0)
1132 				return (error);
1133 			/*
1134 			 * XXX We could save a lock/unlock if this was only
1135 			 * enabled under INVARIANTS
1136 			 */
1137 			VI_LOCK(vp);
1138 			if (vp->v_numoutput > 0 ||
1139 			    !TAILQ_EMPTY(&vp->v_dirtyblkhd))
1140 				panic("vinvalbuf: dirty bufs");
1141 		}
1142 	}
1143 	/*
1144 	 * If you alter this loop please notice that interlock is dropped and
1145 	 * reacquired in flushbuflist.  Special care is needed to ensure that
1146 	 * no race conditions occur from this.
1147 	 */
1148 	for (error = 0;;) {
1149 		if ((blist = TAILQ_FIRST(&vp->v_cleanblkhd)) != 0 &&
1150 		    flushbuflist(blist, flags, vp, slpflag, slptimeo, &error)) {
1151 			if (error)
1152 				break;
1153 			continue;
1154 		}
1155 		if ((blist = TAILQ_FIRST(&vp->v_dirtyblkhd)) != 0 &&
1156 		    flushbuflist(blist, flags, vp, slpflag, slptimeo, &error)) {
1157 			if (error)
1158 				break;
1159 			continue;
1160 		}
1161 		break;
1162 	}
1163 	if (error) {
1164 		VI_UNLOCK(vp);
1165 		return (error);
1166 	}
1167 
1168 	/*
1169 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1170 	 * have write I/O in-progress but if there is a VM object then the
1171 	 * VM object can also have read-I/O in-progress.
1172 	 */
1173 	do {
1174 		while (vp->v_numoutput > 0) {
1175 			vp->v_iflag |= VI_BWAIT;
1176 			msleep(&vp->v_numoutput, VI_MTX(vp), PVM, "vnvlbv", 0);
1177 		}
1178 		VI_UNLOCK(vp);
1179 		if (VOP_GETVOBJECT(vp, &object) == 0) {
1180 			VM_OBJECT_LOCK(object);
1181 			vm_object_pip_wait(object, "vnvlbx");
1182 			VM_OBJECT_UNLOCK(object);
1183 		}
1184 		VI_LOCK(vp);
1185 	} while (vp->v_numoutput > 0);
1186 	VI_UNLOCK(vp);
1187 
1188 	/*
1189 	 * Destroy the copy in the VM cache, too.
1190 	 */
1191 	if (VOP_GETVOBJECT(vp, &object) == 0) {
1192 		VM_OBJECT_LOCK(object);
1193 		vm_object_page_remove(object, 0, 0,
1194 			(flags & V_SAVE) ? TRUE : FALSE);
1195 		VM_OBJECT_UNLOCK(object);
1196 	}
1197 
1198 #ifdef INVARIANTS
1199 	VI_LOCK(vp);
1200 	if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1201 	    (!TAILQ_EMPTY(&vp->v_dirtyblkhd) ||
1202 	     !TAILQ_EMPTY(&vp->v_cleanblkhd)))
1203 		panic("vinvalbuf: flush failed");
1204 	VI_UNLOCK(vp);
1205 #endif
1206 	return (0);
1207 }
1208 
1209 /*
1210  * Flush out buffers on the specified list.
1211  *
1212  */
1213 static int
1214 flushbuflist(blist, flags, vp, slpflag, slptimeo, errorp)
1215 	struct buf *blist;
1216 	int flags;
1217 	struct vnode *vp;
1218 	int slpflag, slptimeo;
1219 	int *errorp;
1220 {
1221 	struct buf *bp, *nbp;
1222 	int found, error;
1223 
1224 	ASSERT_VI_LOCKED(vp, "flushbuflist");
1225 
1226 	for (found = 0, bp = blist; bp; bp = nbp) {
1227 		nbp = TAILQ_NEXT(bp, b_vnbufs);
1228 		if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1229 		    ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1230 			continue;
1231 		}
1232 		found += 1;
1233 		error = BUF_TIMELOCK(bp,
1234 		    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, VI_MTX(vp),
1235 		    "flushbuf", slpflag, slptimeo);
1236 		if (error) {
1237 			if (error != ENOLCK)
1238 				*errorp = error;
1239 			goto done;
1240 		}
1241 		/*
1242 		 * XXX Since there are no node locks for NFS, I
1243 		 * believe there is a slight chance that a delayed
1244 		 * write will occur while sleeping just above, so
1245 		 * check for it.  Note that vfs_bio_awrite expects
1246 		 * buffers to reside on a queue, while BUF_WRITE and
1247 		 * brelse do not.
1248 		 */
1249 		if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1250 			(flags & V_SAVE)) {
1251 
1252 			if (bp->b_vp == vp) {
1253 				if (bp->b_flags & B_CLUSTEROK) {
1254 					vfs_bio_awrite(bp);
1255 				} else {
1256 					bremfree(bp);
1257 					bp->b_flags |= B_ASYNC;
1258 					BUF_WRITE(bp);
1259 				}
1260 			} else {
1261 				bremfree(bp);
1262 				(void) BUF_WRITE(bp);
1263 			}
1264 			goto done;
1265 		}
1266 		bremfree(bp);
1267 		bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
1268 		bp->b_flags &= ~B_ASYNC;
1269 		brelse(bp);
1270 		VI_LOCK(vp);
1271 	}
1272 	return (found);
1273 done:
1274 	VI_LOCK(vp);
1275 	return (found);
1276 }
1277 
1278 /*
1279  * Truncate a file's buffer and pages to a specified length.  This
1280  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1281  * sync activity.
1282  */
1283 int
1284 vtruncbuf(vp, cred, td, length, blksize)
1285 	register struct vnode *vp;
1286 	struct ucred *cred;
1287 	struct thread *td;
1288 	off_t length;
1289 	int blksize;
1290 {
1291 	register struct buf *bp;
1292 	struct buf *nbp;
1293 	int anyfreed;
1294 	int trunclbn;
1295 
1296 	/*
1297 	 * Round up to the *next* lbn.
1298 	 */
1299 	trunclbn = (length + blksize - 1) / blksize;
1300 
1301 	ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1302 restart:
1303 	VI_LOCK(vp);
1304 	anyfreed = 1;
1305 	for (;anyfreed;) {
1306 		anyfreed = 0;
1307 		for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
1308 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1309 			if (bp->b_lblkno >= trunclbn) {
1310 				if (BUF_LOCK(bp,
1311 				    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1312 				    VI_MTX(vp)) == ENOLCK)
1313 					goto restart;
1314 
1315 				bremfree(bp);
1316 				bp->b_flags |= (B_INVAL | B_RELBUF);
1317 				bp->b_flags &= ~B_ASYNC;
1318 				brelse(bp);
1319 				anyfreed = 1;
1320 
1321 				if (nbp &&
1322 				    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1323 				    (nbp->b_vp != vp) ||
1324 				    (nbp->b_flags & B_DELWRI))) {
1325 					goto restart;
1326 				}
1327 				VI_LOCK(vp);
1328 			}
1329 		}
1330 
1331 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1332 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1333 			if (bp->b_lblkno >= trunclbn) {
1334 				if (BUF_LOCK(bp,
1335 				    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1336 				    VI_MTX(vp)) == ENOLCK)
1337 					goto restart;
1338 				bremfree(bp);
1339 				bp->b_flags |= (B_INVAL | B_RELBUF);
1340 				bp->b_flags &= ~B_ASYNC;
1341 				brelse(bp);
1342 				anyfreed = 1;
1343 				if (nbp &&
1344 				    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1345 				    (nbp->b_vp != vp) ||
1346 				    (nbp->b_flags & B_DELWRI) == 0)) {
1347 					goto restart;
1348 				}
1349 				VI_LOCK(vp);
1350 			}
1351 		}
1352 	}
1353 
1354 	if (length > 0) {
1355 restartsync:
1356 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1357 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1358 			if (bp->b_lblkno > 0)
1359 				continue;
1360 			/*
1361 			 * Since we hold the vnode lock this should only
1362 			 * fail if we're racing with the buf daemon.
1363 			 */
1364 			if (BUF_LOCK(bp,
1365 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1366 			    VI_MTX(vp)) == ENOLCK) {
1367 				goto restart;
1368 			}
1369 			KASSERT((bp->b_flags & B_DELWRI),
1370 			    ("buf(%p) on dirty queue without DELWRI.", bp));
1371 
1372 			bremfree(bp);
1373 			bawrite(bp);
1374 			VI_LOCK(vp);
1375 			goto restartsync;
1376 		}
1377 	}
1378 
1379 	while (vp->v_numoutput > 0) {
1380 		vp->v_iflag |= VI_BWAIT;
1381 		msleep(&vp->v_numoutput, VI_MTX(vp), PVM, "vbtrunc", 0);
1382 	}
1383 	VI_UNLOCK(vp);
1384 	vnode_pager_setsize(vp, length);
1385 
1386 	return (0);
1387 }
1388 
1389 /*
1390  * buf_splay() - splay tree core for the clean/dirty list of buffers in
1391  * 		 a vnode.
1392  *
1393  *	NOTE: We have to deal with the special case of a background bitmap
1394  *	buffer, a situation where two buffers will have the same logical
1395  *	block offset.  We want (1) only the foreground buffer to be accessed
1396  *	in a lookup and (2) must differentiate between the foreground and
1397  *	background buffer in the splay tree algorithm because the splay
1398  *	tree cannot normally handle multiple entities with the same 'index'.
1399  *	We accomplish this by adding differentiating flags to the splay tree's
1400  *	numerical domain.
1401  */
1402 static
1403 struct buf *
1404 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1405 {
1406 	struct buf dummy;
1407 	struct buf *lefttreemax, *righttreemin, *y;
1408 
1409 	if (root == NULL)
1410 		return (NULL);
1411 	lefttreemax = righttreemin = &dummy;
1412 	for (;;) {
1413 		if (lblkno < root->b_lblkno ||
1414 		    (lblkno == root->b_lblkno &&
1415 		    (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1416 			if ((y = root->b_left) == NULL)
1417 				break;
1418 			if (lblkno < y->b_lblkno) {
1419 				/* Rotate right. */
1420 				root->b_left = y->b_right;
1421 				y->b_right = root;
1422 				root = y;
1423 				if ((y = root->b_left) == NULL)
1424 					break;
1425 			}
1426 			/* Link into the new root's right tree. */
1427 			righttreemin->b_left = root;
1428 			righttreemin = root;
1429 		} else if (lblkno > root->b_lblkno ||
1430 		    (lblkno == root->b_lblkno &&
1431 		    (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1432 			if ((y = root->b_right) == NULL)
1433 				break;
1434 			if (lblkno > y->b_lblkno) {
1435 				/* Rotate left. */
1436 				root->b_right = y->b_left;
1437 				y->b_left = root;
1438 				root = y;
1439 				if ((y = root->b_right) == NULL)
1440 					break;
1441 			}
1442 			/* Link into the new root's left tree. */
1443 			lefttreemax->b_right = root;
1444 			lefttreemax = root;
1445 		} else {
1446 			break;
1447 		}
1448 		root = y;
1449 	}
1450 	/* Assemble the new root. */
1451 	lefttreemax->b_right = root->b_left;
1452 	righttreemin->b_left = root->b_right;
1453 	root->b_left = dummy.b_right;
1454 	root->b_right = dummy.b_left;
1455 	return (root);
1456 }
1457 
1458 static
1459 void
1460 buf_vlist_remove(struct buf *bp)
1461 {
1462 	struct vnode *vp = bp->b_vp;
1463 	struct buf *root;
1464 
1465 	ASSERT_VI_LOCKED(vp, "buf_vlist_remove");
1466 	if (bp->b_xflags & BX_VNDIRTY) {
1467 		if (bp != vp->v_dirtyblkroot) {
1468 			root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_dirtyblkroot);
1469 			KASSERT(root == bp, ("splay lookup failed during dirty remove"));
1470 		}
1471 		if (bp->b_left == NULL) {
1472 			root = bp->b_right;
1473 		} else {
1474 			root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1475 			root->b_right = bp->b_right;
1476 		}
1477 		vp->v_dirtyblkroot = root;
1478 		TAILQ_REMOVE(&vp->v_dirtyblkhd, bp, b_vnbufs);
1479 		vp->v_dirtybufcnt--;
1480 	} else {
1481 		/* KASSERT(bp->b_xflags & BX_VNCLEAN, ("bp wasn't clean")); */
1482 		if (bp != vp->v_cleanblkroot) {
1483 			root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_cleanblkroot);
1484 			KASSERT(root == bp, ("splay lookup failed during clean remove"));
1485 		}
1486 		if (bp->b_left == NULL) {
1487 			root = bp->b_right;
1488 		} else {
1489 			root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1490 			root->b_right = bp->b_right;
1491 		}
1492 		vp->v_cleanblkroot = root;
1493 		TAILQ_REMOVE(&vp->v_cleanblkhd, bp, b_vnbufs);
1494 		vp->v_cleanbufcnt--;
1495 	}
1496 	bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1497 }
1498 
1499 /*
1500  * Add the buffer to the sorted clean or dirty block list using a
1501  * splay tree algorithm.
1502  *
1503  * NOTE: xflags is passed as a constant, optimizing this inline function!
1504  */
1505 static
1506 void
1507 buf_vlist_add(struct buf *bp, struct vnode *vp, b_xflags_t xflags)
1508 {
1509 	struct buf *root;
1510 
1511 	ASSERT_VI_LOCKED(vp, "buf_vlist_add");
1512 	bp->b_xflags |= xflags;
1513 	if (xflags & BX_VNDIRTY) {
1514 		root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_dirtyblkroot);
1515 		if (root == NULL) {
1516 			bp->b_left = NULL;
1517 			bp->b_right = NULL;
1518 			TAILQ_INSERT_TAIL(&vp->v_dirtyblkhd, bp, b_vnbufs);
1519 		} else if (bp->b_lblkno < root->b_lblkno ||
1520 		    (bp->b_lblkno == root->b_lblkno &&
1521 		    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1522 			bp->b_left = root->b_left;
1523 			bp->b_right = root;
1524 			root->b_left = NULL;
1525 			TAILQ_INSERT_BEFORE(root, bp, b_vnbufs);
1526 		} else {
1527 			bp->b_right = root->b_right;
1528 			bp->b_left = root;
1529 			root->b_right = NULL;
1530 			TAILQ_INSERT_AFTER(&vp->v_dirtyblkhd,
1531 			    root, bp, b_vnbufs);
1532 		}
1533 		vp->v_dirtybufcnt++;
1534 		vp->v_dirtyblkroot = bp;
1535 	} else {
1536 		/* KASSERT(xflags & BX_VNCLEAN, ("xflags not clean")); */
1537 		root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_cleanblkroot);
1538 		if (root == NULL) {
1539 			bp->b_left = NULL;
1540 			bp->b_right = NULL;
1541 			TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
1542 		} else if (bp->b_lblkno < root->b_lblkno ||
1543 		    (bp->b_lblkno == root->b_lblkno &&
1544 		    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1545 			bp->b_left = root->b_left;
1546 			bp->b_right = root;
1547 			root->b_left = NULL;
1548 			TAILQ_INSERT_BEFORE(root, bp, b_vnbufs);
1549 		} else {
1550 			bp->b_right = root->b_right;
1551 			bp->b_left = root;
1552 			root->b_right = NULL;
1553 			TAILQ_INSERT_AFTER(&vp->v_cleanblkhd,
1554 			    root, bp, b_vnbufs);
1555 		}
1556 		vp->v_cleanbufcnt++;
1557 		vp->v_cleanblkroot = bp;
1558 	}
1559 }
1560 
1561 /*
1562  * Lookup a buffer using the splay tree.  Note that we specifically avoid
1563  * shadow buffers used in background bitmap writes.
1564  *
1565  * This code isn't quite efficient as it could be because we are maintaining
1566  * two sorted lists and do not know which list the block resides in.
1567  *
1568  * During a "make buildworld" the desired buffer is found at one of
1569  * the roots more than 60% of the time.  Thus, checking both roots
1570  * before performing either splay eliminates unnecessary splays on the
1571  * first tree splayed.
1572  */
1573 struct buf *
1574 gbincore(struct vnode *vp, daddr_t lblkno)
1575 {
1576 	struct buf *bp;
1577 
1578 	GIANT_REQUIRED;
1579 
1580 	ASSERT_VI_LOCKED(vp, "gbincore");
1581 	if ((bp = vp->v_cleanblkroot) != NULL &&
1582 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1583 		return (bp);
1584 	if ((bp = vp->v_dirtyblkroot) != NULL &&
1585 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1586 		return (bp);
1587 	if ((bp = vp->v_cleanblkroot) != NULL) {
1588 		vp->v_cleanblkroot = bp = buf_splay(lblkno, 0, bp);
1589 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1590 			return (bp);
1591 	}
1592 	if ((bp = vp->v_dirtyblkroot) != NULL) {
1593 		vp->v_dirtyblkroot = bp = buf_splay(lblkno, 0, bp);
1594 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1595 			return (bp);
1596 	}
1597 	return (NULL);
1598 }
1599 
1600 /*
1601  * Associate a buffer with a vnode.
1602  */
1603 void
1604 bgetvp(vp, bp)
1605 	register struct vnode *vp;
1606 	register struct buf *bp;
1607 {
1608 	KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
1609 
1610 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1611 	    ("bgetvp: bp already attached! %p", bp));
1612 
1613 	ASSERT_VI_LOCKED(vp, "bgetvp");
1614 	vholdl(vp);
1615 	bp->b_vp = vp;
1616 	bp->b_dev = vn_todev(vp);
1617 	/*
1618 	 * Insert onto list for new vnode.
1619 	 */
1620 	buf_vlist_add(bp, vp, BX_VNCLEAN);
1621 }
1622 
1623 /*
1624  * Disassociate a buffer from a vnode.
1625  */
1626 void
1627 brelvp(bp)
1628 	register struct buf *bp;
1629 {
1630 	struct vnode *vp;
1631 
1632 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1633 
1634 	/*
1635 	 * Delete from old vnode list, if on one.
1636 	 */
1637 	vp = bp->b_vp;
1638 	VI_LOCK(vp);
1639 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1640 		buf_vlist_remove(bp);
1641 	if ((vp->v_iflag & VI_ONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1642 		vp->v_iflag &= ~VI_ONWORKLST;
1643 		mtx_lock(&sync_mtx);
1644 		LIST_REMOVE(vp, v_synclist);
1645 		mtx_unlock(&sync_mtx);
1646 	}
1647 	vdropl(vp);
1648 	bp->b_vp = (struct vnode *) 0;
1649 	if (bp->b_object)
1650 		bp->b_object = NULL;
1651 	VI_UNLOCK(vp);
1652 }
1653 
1654 /*
1655  * Add an item to the syncer work queue.
1656  */
1657 static void
1658 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
1659 {
1660 	int slot;
1661 
1662 	ASSERT_VI_LOCKED(vp, "vn_syncer_add_to_worklist");
1663 
1664 	mtx_lock(&sync_mtx);
1665 	if (vp->v_iflag & VI_ONWORKLST)
1666 		LIST_REMOVE(vp, v_synclist);
1667 	else
1668 		vp->v_iflag |= VI_ONWORKLST;
1669 
1670 	if (delay > syncer_maxdelay - 2)
1671 		delay = syncer_maxdelay - 2;
1672 	slot = (syncer_delayno + delay) & syncer_mask;
1673 
1674 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
1675 	mtx_unlock(&sync_mtx);
1676 }
1677 
1678 struct  proc *updateproc;
1679 static void sched_sync(void);
1680 static struct kproc_desc up_kp = {
1681 	"syncer",
1682 	sched_sync,
1683 	&updateproc
1684 };
1685 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1686 
1687 /*
1688  * System filesystem synchronizer daemon.
1689  */
1690 static void
1691 sched_sync(void)
1692 {
1693 	struct synclist *next;
1694 	struct synclist *slp;
1695 	struct vnode *vp;
1696 	struct mount *mp;
1697 	long starttime;
1698 	struct thread *td = FIRST_THREAD_IN_PROC(updateproc);  /* XXXKSE */
1699 
1700 	mtx_lock(&Giant);
1701 
1702 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, td->td_proc,
1703 	    SHUTDOWN_PRI_LAST);
1704 
1705 	for (;;) {
1706 		kthread_suspend_check(td->td_proc);
1707 
1708 		starttime = time_second;
1709 
1710 		/*
1711 		 * Push files whose dirty time has expired.  Be careful
1712 		 * of interrupt race on slp queue.
1713 		 */
1714 		mtx_lock(&sync_mtx);
1715 		slp = &syncer_workitem_pending[syncer_delayno];
1716 		syncer_delayno += 1;
1717 		if (syncer_delayno == syncer_maxdelay)
1718 			syncer_delayno = 0;
1719 		next = &syncer_workitem_pending[syncer_delayno];
1720 
1721 		while ((vp = LIST_FIRST(slp)) != NULL) {
1722 			if (VOP_ISLOCKED(vp, NULL) != 0 ||
1723 			    vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1724 				LIST_REMOVE(vp, v_synclist);
1725 				LIST_INSERT_HEAD(next, vp, v_synclist);
1726 				continue;
1727 			}
1728 			if (VI_TRYLOCK(vp) == 0) {
1729 				LIST_REMOVE(vp, v_synclist);
1730 				LIST_INSERT_HEAD(next, vp, v_synclist);
1731 				vn_finished_write(mp);
1732 				continue;
1733 			}
1734 			/*
1735 			 * We use vhold in case the vnode does not
1736 			 * successfully sync.  vhold prevents the vnode from
1737 			 * going away when we unlock the sync_mtx so that
1738 			 * we can acquire the vnode interlock.
1739 			 */
1740 			vholdl(vp);
1741 			mtx_unlock(&sync_mtx);
1742 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK, td);
1743 			(void) VOP_FSYNC(vp, td->td_ucred, MNT_LAZY, td);
1744 			VOP_UNLOCK(vp, 0, td);
1745 			vn_finished_write(mp);
1746 			VI_LOCK(vp);
1747 			if ((vp->v_iflag & VI_ONWORKLST) != 0) {
1748 				/*
1749 				 * Put us back on the worklist.  The worklist
1750 				 * routine will remove us from our current
1751 				 * position and then add us back in at a later
1752 				 * position.
1753 				 */
1754 				vn_syncer_add_to_worklist(vp, syncdelay);
1755 			}
1756 			vdropl(vp);
1757 			VI_UNLOCK(vp);
1758 			mtx_lock(&sync_mtx);
1759 		}
1760 		mtx_unlock(&sync_mtx);
1761 
1762 		/*
1763 		 * Do soft update processing.
1764 		 */
1765 		if (softdep_process_worklist_hook != NULL)
1766 			(*softdep_process_worklist_hook)(NULL);
1767 
1768 		/*
1769 		 * The variable rushjob allows the kernel to speed up the
1770 		 * processing of the filesystem syncer process. A rushjob
1771 		 * value of N tells the filesystem syncer to process the next
1772 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1773 		 * is used by the soft update code to speed up the filesystem
1774 		 * syncer process when the incore state is getting so far
1775 		 * ahead of the disk that the kernel memory pool is being
1776 		 * threatened with exhaustion.
1777 		 */
1778 		mtx_lock(&sync_mtx);
1779 		if (rushjob > 0) {
1780 			rushjob -= 1;
1781 			mtx_unlock(&sync_mtx);
1782 			continue;
1783 		}
1784 		mtx_unlock(&sync_mtx);
1785 		/*
1786 		 * If it has taken us less than a second to process the
1787 		 * current work, then wait. Otherwise start right over
1788 		 * again. We can still lose time if any single round
1789 		 * takes more than two seconds, but it does not really
1790 		 * matter as we are just trying to generally pace the
1791 		 * filesystem activity.
1792 		 */
1793 		if (time_second == starttime)
1794 			tsleep(&lbolt, PPAUSE, "syncer", 0);
1795 	}
1796 }
1797 
1798 /*
1799  * Request the syncer daemon to speed up its work.
1800  * We never push it to speed up more than half of its
1801  * normal turn time, otherwise it could take over the cpu.
1802  * XXXKSE  only one update?
1803  */
1804 int
1805 speedup_syncer()
1806 {
1807 	struct thread *td;
1808 	int ret = 0;
1809 
1810 	td = FIRST_THREAD_IN_PROC(updateproc);
1811 	mtx_lock_spin(&sched_lock);
1812 	if (td->td_wchan == &lbolt) {
1813 		unsleep(td);
1814 		TD_CLR_SLEEPING(td);
1815 		setrunnable(td);
1816 	}
1817 	mtx_unlock_spin(&sched_lock);
1818 	mtx_lock(&sync_mtx);
1819 	if (rushjob < syncdelay / 2) {
1820 		rushjob += 1;
1821 		stat_rush_requests += 1;
1822 		ret = 1;
1823 	}
1824 	mtx_unlock(&sync_mtx);
1825 	return (ret);
1826 }
1827 
1828 /*
1829  * Associate a p-buffer with a vnode.
1830  *
1831  * Also sets B_PAGING flag to indicate that vnode is not fully associated
1832  * with the buffer.  i.e. the bp has not been linked into the vnode or
1833  * ref-counted.
1834  */
1835 void
1836 pbgetvp(vp, bp)
1837 	register struct vnode *vp;
1838 	register struct buf *bp;
1839 {
1840 
1841 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1842 
1843 	bp->b_vp = vp;
1844 	bp->b_object = vp->v_object;
1845 	bp->b_flags |= B_PAGING;
1846 	bp->b_dev = vn_todev(vp);
1847 }
1848 
1849 /*
1850  * Disassociate a p-buffer from a vnode.
1851  */
1852 void
1853 pbrelvp(bp)
1854 	register struct buf *bp;
1855 {
1856 
1857 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1858 
1859 	/* XXX REMOVE ME */
1860 	VI_LOCK(bp->b_vp);
1861 	if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1862 		panic(
1863 		    "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1864 		    bp,
1865 		    (int)bp->b_flags
1866 		);
1867 	}
1868 	VI_UNLOCK(bp->b_vp);
1869 	bp->b_vp = (struct vnode *) 0;
1870 	bp->b_object = NULL;
1871 	bp->b_flags &= ~B_PAGING;
1872 }
1873 
1874 /*
1875  * Reassign a buffer from one vnode to another.
1876  * Used to assign file specific control information
1877  * (indirect blocks) to the vnode to which they belong.
1878  */
1879 void
1880 reassignbuf(bp, newvp)
1881 	register struct buf *bp;
1882 	register struct vnode *newvp;
1883 {
1884 	struct vnode *vp;
1885 	int delay;
1886 
1887 	if (newvp == NULL) {
1888 		printf("reassignbuf: NULL");
1889 		return;
1890 	}
1891 	vp = bp->b_vp;
1892 	++reassignbufcalls;
1893 
1894 	/*
1895 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1896 	 * is not fully linked in.
1897 	 */
1898 	if (bp->b_flags & B_PAGING)
1899 		panic("cannot reassign paging buffer");
1900 
1901 	/*
1902 	 * Delete from old vnode list, if on one.
1903 	 */
1904 	VI_LOCK(vp);
1905 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1906 		buf_vlist_remove(bp);
1907 		if (vp != newvp) {
1908 			vdropl(bp->b_vp);
1909 			bp->b_vp = NULL;	/* for clarification */
1910 		}
1911 	}
1912 	if (vp != newvp) {
1913 		VI_UNLOCK(vp);
1914 		VI_LOCK(newvp);
1915 	}
1916 	/*
1917 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1918 	 * of clean buffers.
1919 	 */
1920 	if (bp->b_flags & B_DELWRI) {
1921 		if ((newvp->v_iflag & VI_ONWORKLST) == 0) {
1922 			switch (newvp->v_type) {
1923 			case VDIR:
1924 				delay = dirdelay;
1925 				break;
1926 			case VCHR:
1927 				if (newvp->v_rdev->si_mountpoint != NULL) {
1928 					delay = metadelay;
1929 					break;
1930 				}
1931 				/* FALLTHROUGH */
1932 			default:
1933 				delay = filedelay;
1934 			}
1935 			vn_syncer_add_to_worklist(newvp, delay);
1936 		}
1937 		buf_vlist_add(bp, newvp, BX_VNDIRTY);
1938 	} else {
1939 		buf_vlist_add(bp, newvp, BX_VNCLEAN);
1940 
1941 		if ((newvp->v_iflag & VI_ONWORKLST) &&
1942 		    TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1943 			mtx_lock(&sync_mtx);
1944 			LIST_REMOVE(newvp, v_synclist);
1945 			mtx_unlock(&sync_mtx);
1946 			newvp->v_iflag &= ~VI_ONWORKLST;
1947 		}
1948 	}
1949 	if (bp->b_vp != newvp) {
1950 		bp->b_vp = newvp;
1951 		vholdl(bp->b_vp);
1952 	}
1953 	VI_UNLOCK(newvp);
1954 }
1955 
1956 /*
1957  * Create a vnode for a device.
1958  * Used for mounting the root filesystem.
1959  */
1960 int
1961 bdevvp(dev, vpp)
1962 	dev_t dev;
1963 	struct vnode **vpp;
1964 {
1965 	register struct vnode *vp;
1966 	struct vnode *nvp;
1967 	int error;
1968 
1969 	if (dev == NODEV) {
1970 		*vpp = NULLVP;
1971 		return (ENXIO);
1972 	}
1973 	if (vfinddev(dev, VCHR, vpp))
1974 		return (0);
1975 	error = getnewvnode("none", (struct mount *)0, spec_vnodeop_p, &nvp);
1976 	if (error) {
1977 		*vpp = NULLVP;
1978 		return (error);
1979 	}
1980 	vp = nvp;
1981 	vp->v_type = VCHR;
1982 	addalias(vp, dev);
1983 	*vpp = vp;
1984 	return (0);
1985 }
1986 
1987 static void
1988 v_incr_usecount(struct vnode *vp, int delta)
1989 {
1990 	vp->v_usecount += delta;
1991 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1992 		mtx_lock(&spechash_mtx);
1993 		vp->v_rdev->si_usecount += delta;
1994 		mtx_unlock(&spechash_mtx);
1995 	}
1996 }
1997 
1998 /*
1999  * Add vnode to the alias list hung off the dev_t.
2000  *
2001  * The reason for this gunk is that multiple vnodes can reference
2002  * the same physical device, so checking vp->v_usecount to see
2003  * how many users there are is inadequate; the v_usecount for
2004  * the vnodes need to be accumulated.  vcount() does that.
2005  */
2006 struct vnode *
2007 addaliasu(nvp, nvp_rdev)
2008 	struct vnode *nvp;
2009 	udev_t nvp_rdev;
2010 {
2011 	struct vnode *ovp;
2012 	vop_t **ops;
2013 	dev_t dev;
2014 
2015 	if (nvp->v_type == VBLK)
2016 		return (nvp);
2017 	if (nvp->v_type != VCHR)
2018 		panic("addaliasu on non-special vnode");
2019 	dev = udev2dev(nvp_rdev, 0);
2020 	/*
2021 	 * Check to see if we have a bdevvp vnode with no associated
2022 	 * filesystem. If so, we want to associate the filesystem of
2023 	 * the new newly instigated vnode with the bdevvp vnode and
2024 	 * discard the newly created vnode rather than leaving the
2025 	 * bdevvp vnode lying around with no associated filesystem.
2026 	 */
2027 	if (vfinddev(dev, nvp->v_type, &ovp) == 0 || ovp->v_data != NULL) {
2028 		addalias(nvp, dev);
2029 		return (nvp);
2030 	}
2031 	/*
2032 	 * Discard unneeded vnode, but save its node specific data.
2033 	 * Note that if there is a lock, it is carried over in the
2034 	 * node specific data to the replacement vnode.
2035 	 */
2036 	vref(ovp);
2037 	ovp->v_data = nvp->v_data;
2038 	ovp->v_tag = nvp->v_tag;
2039 	nvp->v_data = NULL;
2040 	lockdestroy(ovp->v_vnlock);
2041 	lockinit(ovp->v_vnlock, PVFS, nvp->v_vnlock->lk_wmesg,
2042 	    nvp->v_vnlock->lk_timo, nvp->v_vnlock->lk_flags & LK_EXTFLG_MASK);
2043 	ops = ovp->v_op;
2044 	ovp->v_op = nvp->v_op;
2045 	if (VOP_ISLOCKED(nvp, curthread)) {
2046 		VOP_UNLOCK(nvp, 0, curthread);
2047 		vn_lock(ovp, LK_EXCLUSIVE | LK_RETRY, curthread);
2048 	}
2049 	nvp->v_op = ops;
2050 	insmntque(ovp, nvp->v_mount);
2051 	vrele(nvp);
2052 	vgone(nvp);
2053 	return (ovp);
2054 }
2055 
2056 /* This is a local helper function that do the same as addaliasu, but for a
2057  * dev_t instead of an udev_t. */
2058 static void
2059 addalias(nvp, dev)
2060 	struct vnode *nvp;
2061 	dev_t dev;
2062 {
2063 
2064 	KASSERT(nvp->v_type == VCHR, ("addalias on non-special vnode"));
2065 	nvp->v_rdev = dev;
2066 	VI_LOCK(nvp);
2067 	mtx_lock(&spechash_mtx);
2068 	SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
2069 	dev->si_usecount += nvp->v_usecount;
2070 	mtx_unlock(&spechash_mtx);
2071 	VI_UNLOCK(nvp);
2072 }
2073 
2074 /*
2075  * Grab a particular vnode from the free list, increment its
2076  * reference count and lock it. The vnode lock bit is set if the
2077  * vnode is being eliminated in vgone. The process is awakened
2078  * when the transition is completed, and an error returned to
2079  * indicate that the vnode is no longer usable (possibly having
2080  * been changed to a new filesystem type).
2081  */
2082 int
2083 vget(vp, flags, td)
2084 	register struct vnode *vp;
2085 	int flags;
2086 	struct thread *td;
2087 {
2088 	int error;
2089 
2090 	/*
2091 	 * If the vnode is in the process of being cleaned out for
2092 	 * another use, we wait for the cleaning to finish and then
2093 	 * return failure. Cleaning is determined by checking that
2094 	 * the VI_XLOCK flag is set.
2095 	 */
2096 	if ((flags & LK_INTERLOCK) == 0)
2097 		VI_LOCK(vp);
2098 	if (vp->v_iflag & VI_XLOCK && vp->v_vxproc != curthread) {
2099 		if ((flags & LK_NOWAIT) == 0) {
2100 			vp->v_iflag |= VI_XWANT;
2101 			msleep(vp, VI_MTX(vp), PINOD | PDROP, "vget", 0);
2102 		}
2103 		return (ENOENT);
2104 	}
2105 
2106 	v_incr_usecount(vp, 1);
2107 
2108 	if (VSHOULDBUSY(vp))
2109 		vbusy(vp);
2110 	if (flags & LK_TYPE_MASK) {
2111 		if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
2112 			/*
2113 			 * must expand vrele here because we do not want
2114 			 * to call VOP_INACTIVE if the reference count
2115 			 * drops back to zero since it was never really
2116 			 * active. We must remove it from the free list
2117 			 * before sleeping so that multiple processes do
2118 			 * not try to recycle it.
2119 			 */
2120 			VI_LOCK(vp);
2121 			v_incr_usecount(vp, -1);
2122 			if (VSHOULDFREE(vp))
2123 				vfree(vp);
2124 			else
2125 				vlruvp(vp);
2126 			VI_UNLOCK(vp);
2127 		}
2128 		return (error);
2129 	}
2130 	VI_UNLOCK(vp);
2131 	return (0);
2132 }
2133 
2134 /*
2135  * Increase the reference count of a vnode.
2136  */
2137 void
2138 vref(struct vnode *vp)
2139 {
2140 	VI_LOCK(vp);
2141 	v_incr_usecount(vp, 1);
2142 	VI_UNLOCK(vp);
2143 }
2144 
2145 /*
2146  * Return reference count of a vnode.
2147  *
2148  * The results of this call are only guaranteed when some mechanism other
2149  * than the VI lock is used to stop other processes from gaining references
2150  * to the vnode.  This may be the case if the caller holds the only reference.
2151  * This is also useful when stale data is acceptable as race conditions may
2152  * be accounted for by some other means.
2153  */
2154 int
2155 vrefcnt(struct vnode *vp)
2156 {
2157 	int usecnt;
2158 
2159 	VI_LOCK(vp);
2160 	usecnt = vp->v_usecount;
2161 	VI_UNLOCK(vp);
2162 
2163 	return (usecnt);
2164 }
2165 
2166 
2167 /*
2168  * Vnode put/release.
2169  * If count drops to zero, call inactive routine and return to freelist.
2170  */
2171 void
2172 vrele(vp)
2173 	struct vnode *vp;
2174 {
2175 	struct thread *td = curthread;	/* XXX */
2176 
2177 	KASSERT(vp != NULL, ("vrele: null vp"));
2178 
2179 	VI_LOCK(vp);
2180 
2181 	/* Skip this v_writecount check if we're going to panic below. */
2182 	KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
2183 	    ("vrele: missed vn_close"));
2184 
2185 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2186 	    vp->v_usecount == 1)) {
2187 		v_incr_usecount(vp, -1);
2188 		VI_UNLOCK(vp);
2189 
2190 		return;
2191 	}
2192 
2193 	if (vp->v_usecount == 1) {
2194 		v_incr_usecount(vp, -1);
2195 		/*
2196 		 * We must call VOP_INACTIVE with the node locked. Mark
2197 		 * as VI_DOINGINACT to avoid recursion.
2198 		 */
2199 		if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0) {
2200 			VI_LOCK(vp);
2201 			vp->v_iflag |= VI_DOINGINACT;
2202 			VI_UNLOCK(vp);
2203 			VOP_INACTIVE(vp, td);
2204 			VI_LOCK(vp);
2205 			KASSERT(vp->v_iflag & VI_DOINGINACT,
2206 			    ("vrele: lost VI_DOINGINACT"));
2207 			vp->v_iflag &= ~VI_DOINGINACT;
2208 		} else
2209 			VI_LOCK(vp);
2210 		if (VSHOULDFREE(vp))
2211 			vfree(vp);
2212 		else
2213 			vlruvp(vp);
2214 		VI_UNLOCK(vp);
2215 
2216 	} else {
2217 #ifdef DIAGNOSTIC
2218 		vprint("vrele: negative ref count", vp);
2219 #endif
2220 		VI_UNLOCK(vp);
2221 		panic("vrele: negative ref cnt");
2222 	}
2223 }
2224 
2225 /*
2226  * Release an already locked vnode.  This give the same effects as
2227  * unlock+vrele(), but takes less time and avoids releasing and
2228  * re-aquiring the lock (as vrele() aquires the lock internally.)
2229  */
2230 void
2231 vput(vp)
2232 	struct vnode *vp;
2233 {
2234 	struct thread *td = curthread;	/* XXX */
2235 
2236 	GIANT_REQUIRED;
2237 
2238 	KASSERT(vp != NULL, ("vput: null vp"));
2239 	VI_LOCK(vp);
2240 	/* Skip this v_writecount check if we're going to panic below. */
2241 	KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
2242 	    ("vput: missed vn_close"));
2243 
2244 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2245 	    vp->v_usecount == 1)) {
2246 		v_incr_usecount(vp, -1);
2247 		VOP_UNLOCK(vp, LK_INTERLOCK, td);
2248 		return;
2249 	}
2250 
2251 	if (vp->v_usecount == 1) {
2252 		v_incr_usecount(vp, -1);
2253 		/*
2254 		 * We must call VOP_INACTIVE with the node locked, so
2255 		 * we just need to release the vnode mutex. Mark as
2256 		 * as VI_DOINGINACT to avoid recursion.
2257 		 */
2258 		vp->v_iflag |= VI_DOINGINACT;
2259 		VI_UNLOCK(vp);
2260 		VOP_INACTIVE(vp, td);
2261 		VI_LOCK(vp);
2262 		KASSERT(vp->v_iflag & VI_DOINGINACT,
2263 		    ("vput: lost VI_DOINGINACT"));
2264 		vp->v_iflag &= ~VI_DOINGINACT;
2265 		if (VSHOULDFREE(vp))
2266 			vfree(vp);
2267 		else
2268 			vlruvp(vp);
2269 		VI_UNLOCK(vp);
2270 
2271 	} else {
2272 #ifdef DIAGNOSTIC
2273 		vprint("vput: negative ref count", vp);
2274 #endif
2275 		panic("vput: negative ref cnt");
2276 	}
2277 }
2278 
2279 /*
2280  * Somebody doesn't want the vnode recycled.
2281  */
2282 void
2283 vhold(struct vnode *vp)
2284 {
2285 	VI_LOCK(vp);
2286 	vholdl(vp);
2287 	VI_UNLOCK(vp);
2288 }
2289 
2290 void
2291 vholdl(vp)
2292 	register struct vnode *vp;
2293 {
2294 	vp->v_holdcnt++;
2295 	if (VSHOULDBUSY(vp))
2296 		vbusy(vp);
2297 }
2298 
2299 /*
2300  * Note that there is one less who cares about this vnode.  vdrop() is the
2301  * opposite of vhold().
2302  */
2303 void
2304 vdrop(struct vnode *vp)
2305 {
2306 	VI_LOCK(vp);
2307 	vdropl(vp);
2308 	VI_UNLOCK(vp);
2309 }
2310 
2311 void
2312 vdropl(vp)
2313 	register struct vnode *vp;
2314 {
2315 	if (vp->v_holdcnt <= 0)
2316 		panic("vdrop: holdcnt");
2317 	vp->v_holdcnt--;
2318 	if (VSHOULDFREE(vp))
2319 		vfree(vp);
2320 	else
2321 		vlruvp(vp);
2322 }
2323 
2324 /*
2325  * Remove any vnodes in the vnode table belonging to mount point mp.
2326  *
2327  * If FORCECLOSE is not specified, there should not be any active ones,
2328  * return error if any are found (nb: this is a user error, not a
2329  * system error). If FORCECLOSE is specified, detach any active vnodes
2330  * that are found.
2331  *
2332  * If WRITECLOSE is set, only flush out regular file vnodes open for
2333  * writing.
2334  *
2335  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2336  *
2337  * `rootrefs' specifies the base reference count for the root vnode
2338  * of this filesystem. The root vnode is considered busy if its
2339  * v_usecount exceeds this value. On a successful return, vflush()
2340  * will call vrele() on the root vnode exactly rootrefs times.
2341  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2342  * be zero.
2343  */
2344 #ifdef DIAGNOSTIC
2345 static int busyprt = 0;		/* print out busy vnodes */
2346 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
2347 #endif
2348 
2349 int
2350 vflush(mp, rootrefs, flags)
2351 	struct mount *mp;
2352 	int rootrefs;
2353 	int flags;
2354 {
2355 	struct thread *td = curthread;	/* XXX */
2356 	struct vnode *vp, *nvp, *rootvp = NULL;
2357 	struct vattr vattr;
2358 	int busy = 0, error;
2359 
2360 	if (rootrefs > 0) {
2361 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2362 		    ("vflush: bad args"));
2363 		/*
2364 		 * Get the filesystem root vnode. We can vput() it
2365 		 * immediately, since with rootrefs > 0, it won't go away.
2366 		 */
2367 		if ((error = VFS_ROOT(mp, &rootvp)) != 0)
2368 			return (error);
2369 		vput(rootvp);
2370 
2371 	}
2372 	MNT_ILOCK(mp);
2373 loop:
2374 	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) {
2375 		/*
2376 		 * Make sure this vnode wasn't reclaimed in getnewvnode().
2377 		 * Start over if it has (it won't be on the list anymore).
2378 		 */
2379 		if (vp->v_mount != mp)
2380 			goto loop;
2381 		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2382 
2383 		VI_LOCK(vp);
2384 		MNT_IUNLOCK(mp);
2385 		error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE, td);
2386 		if (error) {
2387 			MNT_ILOCK(mp);
2388 			goto loop;
2389 		}
2390 		/*
2391 		 * Skip over a vnodes marked VV_SYSTEM.
2392 		 */
2393 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2394 			VOP_UNLOCK(vp, 0, td);
2395 			MNT_ILOCK(mp);
2396 			continue;
2397 		}
2398 		/*
2399 		 * If WRITECLOSE is set, flush out unlinked but still open
2400 		 * files (even if open only for reading) and regular file
2401 		 * vnodes open for writing.
2402 		 */
2403 		if (flags & WRITECLOSE) {
2404 			error = VOP_GETATTR(vp, &vattr, td->td_ucred, td);
2405 			VI_LOCK(vp);
2406 
2407 			if ((vp->v_type == VNON ||
2408 			    (error == 0 && vattr.va_nlink > 0)) &&
2409 			    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2410 				VOP_UNLOCK(vp, LK_INTERLOCK, td);
2411 				MNT_ILOCK(mp);
2412 				continue;
2413 			}
2414 		} else
2415 			VI_LOCK(vp);
2416 
2417 		VOP_UNLOCK(vp, 0, td);
2418 
2419 		/*
2420 		 * With v_usecount == 0, all we need to do is clear out the
2421 		 * vnode data structures and we are done.
2422 		 */
2423 		if (vp->v_usecount == 0) {
2424 			vgonel(vp, td);
2425 			MNT_ILOCK(mp);
2426 			continue;
2427 		}
2428 
2429 		/*
2430 		 * If FORCECLOSE is set, forcibly close the vnode. For block
2431 		 * or character devices, revert to an anonymous device. For
2432 		 * all other files, just kill them.
2433 		 */
2434 		if (flags & FORCECLOSE) {
2435 			if (vp->v_type != VCHR)
2436 				vgonel(vp, td);
2437 			else
2438 				vgonechrl(vp, td);
2439 			MNT_ILOCK(mp);
2440 			continue;
2441 		}
2442 #ifdef DIAGNOSTIC
2443 		if (busyprt)
2444 			vprint("vflush: busy vnode", vp);
2445 #endif
2446 		VI_UNLOCK(vp);
2447 		MNT_ILOCK(mp);
2448 		busy++;
2449 	}
2450 	MNT_IUNLOCK(mp);
2451 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2452 		/*
2453 		 * If just the root vnode is busy, and if its refcount
2454 		 * is equal to `rootrefs', then go ahead and kill it.
2455 		 */
2456 		VI_LOCK(rootvp);
2457 		KASSERT(busy > 0, ("vflush: not busy"));
2458 		KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
2459 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2460 			vgonel(rootvp, td);
2461 			busy = 0;
2462 		} else
2463 			VI_UNLOCK(rootvp);
2464 	}
2465 	if (busy)
2466 		return (EBUSY);
2467 	for (; rootrefs > 0; rootrefs--)
2468 		vrele(rootvp);
2469 	return (0);
2470 }
2471 
2472 /*
2473  * This moves a now (likely recyclable) vnode to the end of the
2474  * mountlist.  XXX However, it is temporarily disabled until we
2475  * can clean up ffs_sync() and friends, which have loop restart
2476  * conditions which this code causes to operate O(N^2).
2477  */
2478 static void
2479 vlruvp(struct vnode *vp)
2480 {
2481 #if 0
2482 	struct mount *mp;
2483 
2484 	if ((mp = vp->v_mount) != NULL) {
2485 		MNT_ILOCK(mp);
2486 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
2487 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
2488 		MNT_IUNLOCK(mp);
2489 	}
2490 #endif
2491 }
2492 
2493 static void
2494 vx_lock(struct vnode *vp)
2495 {
2496 	ASSERT_VI_LOCKED(vp, "vx_lock");
2497 
2498 	/*
2499 	 * Prevent the vnode from being recycled or brought into use while we
2500 	 * clean it out.
2501 	 */
2502 	if (vp->v_iflag & VI_XLOCK)
2503 		panic("vclean: deadlock");
2504 	vp->v_iflag |= VI_XLOCK;
2505 	vp->v_vxproc = curthread;
2506 }
2507 
2508 static void
2509 vx_unlock(struct vnode *vp)
2510 {
2511 	ASSERT_VI_LOCKED(vp, "vx_unlock");
2512 	vp->v_iflag &= ~VI_XLOCK;
2513 	vp->v_vxproc = NULL;
2514 	if (vp->v_iflag & VI_XWANT) {
2515 		vp->v_iflag &= ~VI_XWANT;
2516 		wakeup(vp);
2517 	}
2518 }
2519 
2520 
2521 /*
2522  * Disassociate the underlying filesystem from a vnode.
2523  */
2524 static void
2525 vclean(vp, flags, td)
2526 	struct vnode *vp;
2527 	int flags;
2528 	struct thread *td;
2529 {
2530 	int active;
2531 
2532 	ASSERT_VI_LOCKED(vp, "vclean");
2533 	/*
2534 	 * Check to see if the vnode is in use. If so we have to reference it
2535 	 * before we clean it out so that its count cannot fall to zero and
2536 	 * generate a race against ourselves to recycle it.
2537 	 */
2538 	if ((active = vp->v_usecount))
2539 		v_incr_usecount(vp, 1);
2540 
2541 	/*
2542 	 * Even if the count is zero, the VOP_INACTIVE routine may still
2543 	 * have the object locked while it cleans it out. The VOP_LOCK
2544 	 * ensures that the VOP_INACTIVE routine is done with its work.
2545 	 * For active vnodes, it ensures that no other activity can
2546 	 * occur while the underlying object is being cleaned out.
2547 	 */
2548 	VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td);
2549 
2550 	/*
2551 	 * Clean out any buffers associated with the vnode.
2552 	 * If the flush fails, just toss the buffers.
2553 	 */
2554 	if (flags & DOCLOSE) {
2555 		struct buf *bp;
2556 		bp = TAILQ_FIRST(&vp->v_dirtyblkhd);
2557 		if (bp != NULL)
2558 			(void) vn_write_suspend_wait(vp, NULL, V_WAIT);
2559 		if (vinvalbuf(vp, V_SAVE, NOCRED, td, 0, 0) != 0)
2560 			vinvalbuf(vp, 0, NOCRED, td, 0, 0);
2561 	}
2562 
2563 	VOP_DESTROYVOBJECT(vp);
2564 
2565 	/*
2566 	 * Any other processes trying to obtain this lock must first
2567 	 * wait for VXLOCK to clear, then call the new lock operation.
2568 	 */
2569 	VOP_UNLOCK(vp, 0, td);
2570 
2571 	/*
2572 	 * If purging an active vnode, it must be closed and
2573 	 * deactivated before being reclaimed. Note that the
2574 	 * VOP_INACTIVE will unlock the vnode.
2575 	 */
2576 	if (active) {
2577 		if (flags & DOCLOSE)
2578 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2579 		VI_LOCK(vp);
2580 		if ((vp->v_iflag & VI_DOINGINACT) == 0) {
2581 			vp->v_iflag |= VI_DOINGINACT;
2582 			VI_UNLOCK(vp);
2583 			if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT, td) != 0)
2584 				panic("vclean: cannot relock.");
2585 			VOP_INACTIVE(vp, td);
2586 			VI_LOCK(vp);
2587 			KASSERT(vp->v_iflag & VI_DOINGINACT,
2588 			    ("vclean: lost VI_DOINGINACT"));
2589 			vp->v_iflag &= ~VI_DOINGINACT;
2590 		}
2591 		VI_UNLOCK(vp);
2592 	}
2593 	/*
2594 	 * Reclaim the vnode.
2595 	 */
2596 	if (VOP_RECLAIM(vp, td))
2597 		panic("vclean: cannot reclaim");
2598 
2599 	if (active) {
2600 		/*
2601 		 * Inline copy of vrele() since VOP_INACTIVE
2602 		 * has already been called.
2603 		 */
2604 		VI_LOCK(vp);
2605 		v_incr_usecount(vp, -1);
2606 		if (vp->v_usecount <= 0) {
2607 #ifdef INVARIANTS
2608 			if (vp->v_usecount < 0 || vp->v_writecount != 0) {
2609 				vprint("vclean: bad ref count", vp);
2610 				panic("vclean: ref cnt");
2611 			}
2612 #endif
2613 			if (VSHOULDFREE(vp))
2614 				vfree(vp);
2615 		}
2616 		VI_UNLOCK(vp);
2617 	}
2618 	/*
2619 	 * Delete from old mount point vnode list.
2620 	 */
2621 	if (vp->v_mount != NULL)
2622 		insmntque(vp, (struct mount *)0);
2623 	cache_purge(vp);
2624 	VI_LOCK(vp);
2625 	if (VSHOULDFREE(vp))
2626 		vfree(vp);
2627 
2628 	/*
2629 	 * Done with purge, reset to the standard lock and
2630 	 * notify sleepers of the grim news.
2631 	 */
2632 	vp->v_vnlock = &vp->v_lock;
2633 	vp->v_op = dead_vnodeop_p;
2634 	if (vp->v_pollinfo != NULL)
2635 		vn_pollgone(vp);
2636 	vp->v_tag = "none";
2637 }
2638 
2639 /*
2640  * Eliminate all activity associated with the requested vnode
2641  * and with all vnodes aliased to the requested vnode.
2642  */
2643 int
2644 vop_revoke(ap)
2645 	struct vop_revoke_args /* {
2646 		struct vnode *a_vp;
2647 		int a_flags;
2648 	} */ *ap;
2649 {
2650 	struct vnode *vp, *vq;
2651 	dev_t dev;
2652 
2653 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
2654 	vp = ap->a_vp;
2655 	KASSERT((vp->v_type == VCHR), ("vop_revoke: not VCHR"));
2656 
2657 	VI_LOCK(vp);
2658 	/*
2659 	 * If a vgone (or vclean) is already in progress,
2660 	 * wait until it is done and return.
2661 	 */
2662 	if (vp->v_iflag & VI_XLOCK) {
2663 		vp->v_iflag |= VI_XWANT;
2664 		msleep(vp, VI_MTX(vp), PINOD | PDROP,
2665 		    "vop_revokeall", 0);
2666 		return (0);
2667 	}
2668 	VI_UNLOCK(vp);
2669 	dev = vp->v_rdev;
2670 	for (;;) {
2671 		mtx_lock(&spechash_mtx);
2672 		vq = SLIST_FIRST(&dev->si_hlist);
2673 		mtx_unlock(&spechash_mtx);
2674 		if (!vq)
2675 			break;
2676 		vgone(vq);
2677 	}
2678 	return (0);
2679 }
2680 
2681 /*
2682  * Recycle an unused vnode to the front of the free list.
2683  * Release the passed interlock if the vnode will be recycled.
2684  */
2685 int
2686 vrecycle(vp, inter_lkp, td)
2687 	struct vnode *vp;
2688 	struct mtx *inter_lkp;
2689 	struct thread *td;
2690 {
2691 
2692 	VI_LOCK(vp);
2693 	if (vp->v_usecount == 0) {
2694 		if (inter_lkp) {
2695 			mtx_unlock(inter_lkp);
2696 		}
2697 		vgonel(vp, td);
2698 		return (1);
2699 	}
2700 	VI_UNLOCK(vp);
2701 	return (0);
2702 }
2703 
2704 /*
2705  * Eliminate all activity associated with a vnode
2706  * in preparation for reuse.
2707  */
2708 void
2709 vgone(vp)
2710 	register struct vnode *vp;
2711 {
2712 	struct thread *td = curthread;	/* XXX */
2713 
2714 	VI_LOCK(vp);
2715 	vgonel(vp, td);
2716 }
2717 
2718 /*
2719  * Disassociate a character device from the its underlying filesystem and
2720  * attach it to spec.  This is for use when the chr device is still active
2721  * and the filesystem is going away.
2722  */
2723 static void
2724 vgonechrl(struct vnode *vp, struct thread *td)
2725 {
2726 	ASSERT_VI_LOCKED(vp, "vgonechrl");
2727 	vx_lock(vp);
2728 	vclean(vp, 0, td);
2729 	vp->v_op = spec_vnodeop_p;
2730 	vx_unlock(vp);
2731 	VI_UNLOCK(vp);
2732 }
2733 /*
2734  * vgone, with the vp interlock held.
2735  */
2736 void
2737 vgonel(vp, td)
2738 	struct vnode *vp;
2739 	struct thread *td;
2740 {
2741 	/*
2742 	 * If a vgone (or vclean) is already in progress,
2743 	 * wait until it is done and return.
2744 	 */
2745 	ASSERT_VI_LOCKED(vp, "vgonel");
2746 	if (vp->v_iflag & VI_XLOCK) {
2747 		vp->v_iflag |= VI_XWANT;
2748 		msleep(vp, VI_MTX(vp), PINOD | PDROP, "vgone", 0);
2749 		return;
2750 	}
2751 	vx_lock(vp);
2752 
2753 	/*
2754 	 * Clean out the filesystem specific data.
2755 	 */
2756 	vclean(vp, DOCLOSE, td);
2757 	VI_UNLOCK(vp);
2758 
2759 	/*
2760 	 * If special device, remove it from special device alias list
2761 	 * if it is on one.
2762 	 */
2763 	VI_LOCK(vp);
2764 	if (vp->v_type == VCHR && vp->v_rdev != NULL && vp->v_rdev != NODEV) {
2765 		mtx_lock(&spechash_mtx);
2766 		SLIST_REMOVE(&vp->v_rdev->si_hlist, vp, vnode, v_specnext);
2767 		vp->v_rdev->si_usecount -= vp->v_usecount;
2768 		mtx_unlock(&spechash_mtx);
2769 		vp->v_rdev = NULL;
2770 	}
2771 
2772 	/*
2773 	 * If it is on the freelist and not already at the head,
2774 	 * move it to the head of the list. The test of the
2775 	 * VDOOMED flag and the reference count of zero is because
2776 	 * it will be removed from the free list by getnewvnode,
2777 	 * but will not have its reference count incremented until
2778 	 * after calling vgone. If the reference count were
2779 	 * incremented first, vgone would (incorrectly) try to
2780 	 * close the previous instance of the underlying object.
2781 	 */
2782 	if (vp->v_usecount == 0 && !(vp->v_iflag & VI_DOOMED)) {
2783 		mtx_lock(&vnode_free_list_mtx);
2784 		if (vp->v_iflag & VI_FREE) {
2785 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2786 		} else {
2787 			vp->v_iflag |= VI_FREE;
2788 			freevnodes++;
2789 		}
2790 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2791 		mtx_unlock(&vnode_free_list_mtx);
2792 	}
2793 
2794 	vp->v_type = VBAD;
2795 	vx_unlock(vp);
2796 	VI_UNLOCK(vp);
2797 }
2798 
2799 /*
2800  * Lookup a vnode by device number.
2801  */
2802 int
2803 vfinddev(dev, type, vpp)
2804 	dev_t dev;
2805 	enum vtype type;
2806 	struct vnode **vpp;
2807 {
2808 	struct vnode *vp;
2809 
2810 	mtx_lock(&spechash_mtx);
2811 	SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2812 		if (type == vp->v_type) {
2813 			*vpp = vp;
2814 			mtx_unlock(&spechash_mtx);
2815 			return (1);
2816 		}
2817 	}
2818 	mtx_unlock(&spechash_mtx);
2819 	return (0);
2820 }
2821 
2822 /*
2823  * Calculate the total number of references to a special device.
2824  */
2825 int
2826 vcount(vp)
2827 	struct vnode *vp;
2828 {
2829 	int count;
2830 
2831 	mtx_lock(&spechash_mtx);
2832 	count = vp->v_rdev->si_usecount;
2833 	mtx_unlock(&spechash_mtx);
2834 	return (count);
2835 }
2836 
2837 /*
2838  * Same as above, but using the dev_t as argument
2839  */
2840 int
2841 count_dev(dev)
2842 	dev_t dev;
2843 {
2844 	int count;
2845 
2846 	mtx_lock(&spechash_mtx);
2847 	count = dev->si_usecount;
2848 	mtx_unlock(&spechash_mtx);
2849 	return(count);
2850 }
2851 
2852 /*
2853  * Print out a description of a vnode.
2854  */
2855 static char *typename[] =
2856 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2857 
2858 void
2859 vprint(label, vp)
2860 	char *label;
2861 	struct vnode *vp;
2862 {
2863 	char buf[96];
2864 
2865 	if (label != NULL)
2866 		printf("%s: %p: ", label, (void *)vp);
2867 	else
2868 		printf("%p: ", (void *)vp);
2869 	printf("tag %s, type %s, usecount %d, writecount %d, refcount %d,",
2870 	    vp->v_tag, typename[vp->v_type], vp->v_usecount,
2871 	    vp->v_writecount, vp->v_holdcnt);
2872 	buf[0] = '\0';
2873 	if (vp->v_vflag & VV_ROOT)
2874 		strcat(buf, "|VV_ROOT");
2875 	if (vp->v_vflag & VV_TEXT)
2876 		strcat(buf, "|VV_TEXT");
2877 	if (vp->v_vflag & VV_SYSTEM)
2878 		strcat(buf, "|VV_SYSTEM");
2879 	if (vp->v_iflag & VI_XLOCK)
2880 		strcat(buf, "|VI_XLOCK");
2881 	if (vp->v_iflag & VI_XWANT)
2882 		strcat(buf, "|VI_XWANT");
2883 	if (vp->v_iflag & VI_BWAIT)
2884 		strcat(buf, "|VI_BWAIT");
2885 	if (vp->v_iflag & VI_DOOMED)
2886 		strcat(buf, "|VI_DOOMED");
2887 	if (vp->v_iflag & VI_FREE)
2888 		strcat(buf, "|VI_FREE");
2889 	if (vp->v_vflag & VV_OBJBUF)
2890 		strcat(buf, "|VV_OBJBUF");
2891 	if (buf[0] != '\0')
2892 		printf(" flags (%s),", &buf[1]);
2893 	lockmgr_printinfo(vp->v_vnlock);
2894 	printf("\n");
2895 	if (vp->v_data != NULL)
2896 		VOP_PRINT(vp);
2897 }
2898 
2899 #ifdef DDB
2900 #include <ddb/ddb.h>
2901 /*
2902  * List all of the locked vnodes in the system.
2903  * Called when debugging the kernel.
2904  */
2905 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2906 {
2907 	struct mount *mp, *nmp;
2908 	struct vnode *vp;
2909 
2910 	/*
2911 	 * Note: because this is DDB, we can't obey the locking semantics
2912 	 * for these structures, which means we could catch an inconsistent
2913 	 * state and dereference a nasty pointer.  Not much to be done
2914 	 * about that.
2915 	 */
2916 	printf("Locked vnodes\n");
2917 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2918 		nmp = TAILQ_NEXT(mp, mnt_list);
2919 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2920 			if (VOP_ISLOCKED(vp, NULL))
2921 				vprint(NULL, vp);
2922 		}
2923 		nmp = TAILQ_NEXT(mp, mnt_list);
2924 	}
2925 }
2926 #endif
2927 
2928 /*
2929  * Fill in a struct xvfsconf based on a struct vfsconf.
2930  */
2931 static void
2932 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2933 {
2934 
2935 	strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2936 	xvfsp->vfc_typenum = vfsp->vfc_typenum;
2937 	xvfsp->vfc_refcount = vfsp->vfc_refcount;
2938 	xvfsp->vfc_flags = vfsp->vfc_flags;
2939 	/*
2940 	 * These are unused in userland, we keep them
2941 	 * to not break binary compatibility.
2942 	 */
2943 	xvfsp->vfc_vfsops = NULL;
2944 	xvfsp->vfc_next = NULL;
2945 }
2946 
2947 static int
2948 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2949 {
2950 	struct vfsconf *vfsp;
2951 	struct xvfsconf *xvfsp;
2952 	int cnt, error, i;
2953 
2954 	cnt = 0;
2955 	for (vfsp = vfsconf; vfsp != NULL; vfsp = vfsp->vfc_next)
2956 		cnt++;
2957 	xvfsp = malloc(sizeof(struct xvfsconf) * cnt, M_TEMP, M_WAITOK);
2958 	/*
2959 	 * Handle the race that we will have here when struct vfsconf
2960 	 * will be locked down by using both cnt and checking vfc_next
2961 	 * against NULL to determine the end of the loop.  The race will
2962 	 * happen because we will have to unlock before calling malloc().
2963 	 * We are protected by Giant for now.
2964 	 */
2965 	i = 0;
2966 	for (vfsp = vfsconf; vfsp != NULL && i < cnt; vfsp = vfsp->vfc_next) {
2967 		vfsconf2x(vfsp, xvfsp + i);
2968 		i++;
2969 	}
2970 	error = SYSCTL_OUT(req, xvfsp, sizeof(struct xvfsconf) * i);
2971 	free(xvfsp, M_TEMP);
2972 	return (error);
2973 }
2974 
2975 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
2976     "S,xvfsconf", "List of all configured filesystems");
2977 
2978 /*
2979  * Top level filesystem related information gathering.
2980  */
2981 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2982 
2983 static int
2984 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2985 {
2986 	int *name = (int *)arg1 - 1;	/* XXX */
2987 	u_int namelen = arg2 + 1;	/* XXX */
2988 	struct vfsconf *vfsp;
2989 	struct xvfsconf xvfsp;
2990 
2991 	printf("WARNING: userland calling deprecated sysctl, "
2992 	    "please rebuild world\n");
2993 
2994 #if 1 || defined(COMPAT_PRELITE2)
2995 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2996 	if (namelen == 1)
2997 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2998 #endif
2999 
3000 	switch (name[1]) {
3001 	case VFS_MAXTYPENUM:
3002 		if (namelen != 2)
3003 			return (ENOTDIR);
3004 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3005 	case VFS_CONF:
3006 		if (namelen != 3)
3007 			return (ENOTDIR);	/* overloaded */
3008 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
3009 			if (vfsp->vfc_typenum == name[2])
3010 				break;
3011 		if (vfsp == NULL)
3012 			return (EOPNOTSUPP);
3013 		vfsconf2x(vfsp, &xvfsp);
3014 		return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3015 	}
3016 	return (EOPNOTSUPP);
3017 }
3018 
3019 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP, vfs_sysctl,
3020 	"Generic filesystem");
3021 
3022 #if 1 || defined(COMPAT_PRELITE2)
3023 
3024 static int
3025 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3026 {
3027 	int error;
3028 	struct vfsconf *vfsp;
3029 	struct ovfsconf ovfs;
3030 
3031 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
3032 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
3033 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
3034 		ovfs.vfc_index = vfsp->vfc_typenum;
3035 		ovfs.vfc_refcount = vfsp->vfc_refcount;
3036 		ovfs.vfc_flags = vfsp->vfc_flags;
3037 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3038 		if (error)
3039 			return error;
3040 	}
3041 	return 0;
3042 }
3043 
3044 #endif /* 1 || COMPAT_PRELITE2 */
3045 
3046 #define KINFO_VNODESLOP		10
3047 #ifdef notyet
3048 /*
3049  * Dump vnode list (via sysctl).
3050  */
3051 /* ARGSUSED */
3052 static int
3053 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3054 {
3055 	struct xvnode *xvn;
3056 	struct thread *td = req->td;
3057 	struct mount *mp;
3058 	struct vnode *vp;
3059 	int error, len, n;
3060 
3061 	/*
3062 	 * Stale numvnodes access is not fatal here.
3063 	 */
3064 	req->lock = 0;
3065 	len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3066 	if (!req->oldptr)
3067 		/* Make an estimate */
3068 		return (SYSCTL_OUT(req, 0, len));
3069 
3070 	sysctl_wire_old_buffer(req, 0);
3071 	xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3072 	n = 0;
3073 	mtx_lock(&mountlist_mtx);
3074 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3075 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td))
3076 			continue;
3077 		MNT_ILOCK(mp);
3078 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3079 			if (n == len)
3080 				break;
3081 			vref(vp);
3082 			xvn[n].xv_size = sizeof *xvn;
3083 			xvn[n].xv_vnode = vp;
3084 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3085 			XV_COPY(usecount);
3086 			XV_COPY(writecount);
3087 			XV_COPY(holdcnt);
3088 			XV_COPY(id);
3089 			XV_COPY(mount);
3090 			XV_COPY(numoutput);
3091 			XV_COPY(type);
3092 #undef XV_COPY
3093 			xvn[n].xv_flag = vp->v_vflag;
3094 
3095 			switch (vp->v_type) {
3096 			case VREG:
3097 			case VDIR:
3098 			case VLNK:
3099 				xvn[n].xv_dev = vp->v_cachedfs;
3100 				xvn[n].xv_ino = vp->v_cachedid;
3101 				break;
3102 			case VBLK:
3103 			case VCHR:
3104 				if (vp->v_rdev == NULL) {
3105 					vrele(vp);
3106 					continue;
3107 				}
3108 				xvn[n].xv_dev = dev2udev(vp->v_rdev);
3109 				break;
3110 			case VSOCK:
3111 				xvn[n].xv_socket = vp->v_socket;
3112 				break;
3113 			case VFIFO:
3114 				xvn[n].xv_fifo = vp->v_fifoinfo;
3115 				break;
3116 			case VNON:
3117 			case VBAD:
3118 			default:
3119 				/* shouldn't happen? */
3120 				vrele(vp);
3121 				continue;
3122 			}
3123 			vrele(vp);
3124 			++n;
3125 		}
3126 		MNT_IUNLOCK(mp);
3127 		mtx_lock(&mountlist_mtx);
3128 		vfs_unbusy(mp, td);
3129 		if (n == len)
3130 			break;
3131 	}
3132 	mtx_unlock(&mountlist_mtx);
3133 
3134 	error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3135 	free(xvn, M_TEMP);
3136 	return (error);
3137 }
3138 
3139 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3140 	0, 0, sysctl_vnode, "S,xvnode", "");
3141 #endif
3142 
3143 /*
3144  * Check to see if a filesystem is mounted on a block device.
3145  */
3146 int
3147 vfs_mountedon(vp)
3148 	struct vnode *vp;
3149 {
3150 
3151 	if (vp->v_rdev->si_mountpoint != NULL)
3152 		return (EBUSY);
3153 	return (0);
3154 }
3155 
3156 /*
3157  * Unmount all filesystems. The list is traversed in reverse order
3158  * of mounting to avoid dependencies.
3159  */
3160 void
3161 vfs_unmountall()
3162 {
3163 	struct mount *mp;
3164 	struct thread *td;
3165 	int error;
3166 
3167 	if (curthread != NULL)
3168 		td = curthread;
3169 	else
3170 		td = FIRST_THREAD_IN_PROC(initproc); /* XXX XXX proc0? */
3171 	/*
3172 	 * Since this only runs when rebooting, it is not interlocked.
3173 	 */
3174 	while(!TAILQ_EMPTY(&mountlist)) {
3175 		mp = TAILQ_LAST(&mountlist, mntlist);
3176 		error = dounmount(mp, MNT_FORCE, td);
3177 		if (error) {
3178 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
3179 			printf("unmount of %s failed (",
3180 			    mp->mnt_stat.f_mntonname);
3181 			if (error == EBUSY)
3182 				printf("BUSY)\n");
3183 			else
3184 				printf("%d)\n", error);
3185 		} else {
3186 			/* The unmount has removed mp from the mountlist */
3187 		}
3188 	}
3189 }
3190 
3191 /*
3192  * perform msync on all vnodes under a mount point
3193  * the mount point must be locked.
3194  */
3195 void
3196 vfs_msync(struct mount *mp, int flags)
3197 {
3198 	struct vnode *vp, *nvp;
3199 	struct vm_object *obj;
3200 	int tries;
3201 
3202 	GIANT_REQUIRED;
3203 
3204 	tries = 5;
3205 	MNT_ILOCK(mp);
3206 loop:
3207 	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) {
3208 		if (vp->v_mount != mp) {
3209 			if (--tries > 0)
3210 				goto loop;
3211 			break;
3212 		}
3213 		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
3214 
3215 		VI_LOCK(vp);
3216 		if (vp->v_iflag & VI_XLOCK) {
3217 			VI_UNLOCK(vp);
3218 			continue;
3219 		}
3220 
3221 		if ((vp->v_iflag & VI_OBJDIRTY) &&
3222 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
3223 			MNT_IUNLOCK(mp);
3224 			if (!vget(vp,
3225 			    LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3226 			    curthread)) {
3227 				if (vp->v_vflag & VV_NOSYNC) {	/* unlinked */
3228 					vput(vp);
3229 					MNT_ILOCK(mp);
3230 					continue;
3231 				}
3232 
3233 				if (VOP_GETVOBJECT(vp, &obj) == 0) {
3234 					VM_OBJECT_LOCK(obj);
3235 					vm_object_page_clean(obj, 0, 0,
3236 					    flags == MNT_WAIT ?
3237 					    OBJPC_SYNC : OBJPC_NOSYNC);
3238 					VM_OBJECT_UNLOCK(obj);
3239 				}
3240 				vput(vp);
3241 			}
3242 			MNT_ILOCK(mp);
3243 			if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) {
3244 				if (--tries > 0)
3245 					goto loop;
3246 				break;
3247 			}
3248 		} else
3249 			VI_UNLOCK(vp);
3250 	}
3251 	MNT_IUNLOCK(mp);
3252 }
3253 
3254 /*
3255  * Create the VM object needed for VMIO and mmap support.  This
3256  * is done for all VREG files in the system.  Some filesystems might
3257  * afford the additional metadata buffering capability of the
3258  * VMIO code by making the device node be VMIO mode also.
3259  *
3260  * vp must be locked when vfs_object_create is called.
3261  */
3262 int
3263 vfs_object_create(vp, td, cred)
3264 	struct vnode *vp;
3265 	struct thread *td;
3266 	struct ucred *cred;
3267 {
3268 	GIANT_REQUIRED;
3269 	return (VOP_CREATEVOBJECT(vp, cred, td));
3270 }
3271 
3272 /*
3273  * Mark a vnode as free, putting it up for recycling.
3274  */
3275 void
3276 vfree(vp)
3277 	struct vnode *vp;
3278 {
3279 	ASSERT_VI_LOCKED(vp, "vfree");
3280 	mtx_lock(&vnode_free_list_mtx);
3281 	KASSERT((vp->v_iflag & VI_FREE) == 0, ("vnode already free"));
3282 	if (vp->v_iflag & VI_AGE) {
3283 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3284 	} else {
3285 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3286 	}
3287 	freevnodes++;
3288 	mtx_unlock(&vnode_free_list_mtx);
3289 	vp->v_iflag &= ~VI_AGE;
3290 	vp->v_iflag |= VI_FREE;
3291 }
3292 
3293 /*
3294  * Opposite of vfree() - mark a vnode as in use.
3295  */
3296 void
3297 vbusy(vp)
3298 	struct vnode *vp;
3299 {
3300 	ASSERT_VI_LOCKED(vp, "vbusy");
3301 	KASSERT((vp->v_iflag & VI_FREE) != 0, ("vnode not free"));
3302 
3303 	mtx_lock(&vnode_free_list_mtx);
3304 	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3305 	freevnodes--;
3306 	mtx_unlock(&vnode_free_list_mtx);
3307 
3308 	vp->v_iflag &= ~(VI_FREE|VI_AGE);
3309 }
3310 
3311 /*
3312  * Record a process's interest in events which might happen to
3313  * a vnode.  Because poll uses the historic select-style interface
3314  * internally, this routine serves as both the ``check for any
3315  * pending events'' and the ``record my interest in future events''
3316  * functions.  (These are done together, while the lock is held,
3317  * to avoid race conditions.)
3318  */
3319 int
3320 vn_pollrecord(vp, td, events)
3321 	struct vnode *vp;
3322 	struct thread *td;
3323 	short events;
3324 {
3325 
3326 	if (vp->v_pollinfo == NULL)
3327 		v_addpollinfo(vp);
3328 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3329 	if (vp->v_pollinfo->vpi_revents & events) {
3330 		/*
3331 		 * This leaves events we are not interested
3332 		 * in available for the other process which
3333 		 * which presumably had requested them
3334 		 * (otherwise they would never have been
3335 		 * recorded).
3336 		 */
3337 		events &= vp->v_pollinfo->vpi_revents;
3338 		vp->v_pollinfo->vpi_revents &= ~events;
3339 
3340 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
3341 		return events;
3342 	}
3343 	vp->v_pollinfo->vpi_events |= events;
3344 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3345 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3346 	return 0;
3347 }
3348 
3349 /*
3350  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
3351  * it is possible for us to miss an event due to race conditions, but
3352  * that condition is expected to be rare, so for the moment it is the
3353  * preferred interface.
3354  */
3355 void
3356 vn_pollevent(vp, events)
3357 	struct vnode *vp;
3358 	short events;
3359 {
3360 
3361 	if (vp->v_pollinfo == NULL)
3362 		v_addpollinfo(vp);
3363 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3364 	if (vp->v_pollinfo->vpi_events & events) {
3365 		/*
3366 		 * We clear vpi_events so that we don't
3367 		 * call selwakeup() twice if two events are
3368 		 * posted before the polling process(es) is
3369 		 * awakened.  This also ensures that we take at
3370 		 * most one selwakeup() if the polling process
3371 		 * is no longer interested.  However, it does
3372 		 * mean that only one event can be noticed at
3373 		 * a time.  (Perhaps we should only clear those
3374 		 * event bits which we note?) XXX
3375 		 */
3376 		vp->v_pollinfo->vpi_events = 0;	/* &= ~events ??? */
3377 		vp->v_pollinfo->vpi_revents |= events;
3378 		selwakeuppri(&vp->v_pollinfo->vpi_selinfo, PRIBIO);
3379 	}
3380 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3381 }
3382 
3383 /*
3384  * Wake up anyone polling on vp because it is being revoked.
3385  * This depends on dead_poll() returning POLLHUP for correct
3386  * behavior.
3387  */
3388 void
3389 vn_pollgone(vp)
3390 	struct vnode *vp;
3391 {
3392 
3393 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3394 	VN_KNOTE(vp, NOTE_REVOKE);
3395 	if (vp->v_pollinfo->vpi_events) {
3396 		vp->v_pollinfo->vpi_events = 0;
3397 		selwakeuppri(&vp->v_pollinfo->vpi_selinfo, PRIBIO);
3398 	}
3399 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3400 }
3401 
3402 
3403 
3404 /*
3405  * Routine to create and manage a filesystem syncer vnode.
3406  */
3407 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
3408 static int	sync_fsync(struct  vop_fsync_args *);
3409 static int	sync_inactive(struct  vop_inactive_args *);
3410 static int	sync_reclaim(struct  vop_reclaim_args *);
3411 
3412 static vop_t **sync_vnodeop_p;
3413 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
3414 	{ &vop_default_desc,	(vop_t *) vop_eopnotsupp },
3415 	{ &vop_close_desc,	(vop_t *) sync_close },		/* close */
3416 	{ &vop_fsync_desc,	(vop_t *) sync_fsync },		/* fsync */
3417 	{ &vop_inactive_desc,	(vop_t *) sync_inactive },	/* inactive */
3418 	{ &vop_reclaim_desc,	(vop_t *) sync_reclaim },	/* reclaim */
3419 	{ &vop_lock_desc,	(vop_t *) vop_stdlock },	/* lock */
3420 	{ &vop_unlock_desc,	(vop_t *) vop_stdunlock },	/* unlock */
3421 	{ &vop_islocked_desc,	(vop_t *) vop_stdislocked },	/* islocked */
3422 	{ NULL, NULL }
3423 };
3424 static struct vnodeopv_desc sync_vnodeop_opv_desc =
3425 	{ &sync_vnodeop_p, sync_vnodeop_entries };
3426 
3427 VNODEOP_SET(sync_vnodeop_opv_desc);
3428 
3429 /*
3430  * Create a new filesystem syncer vnode for the specified mount point.
3431  */
3432 int
3433 vfs_allocate_syncvnode(mp)
3434 	struct mount *mp;
3435 {
3436 	struct vnode *vp;
3437 	static long start, incr, next;
3438 	int error;
3439 
3440 	/* Allocate a new vnode */
3441 	if ((error = getnewvnode("syncer", mp, sync_vnodeop_p, &vp)) != 0) {
3442 		mp->mnt_syncer = NULL;
3443 		return (error);
3444 	}
3445 	vp->v_type = VNON;
3446 	/*
3447 	 * Place the vnode onto the syncer worklist. We attempt to
3448 	 * scatter them about on the list so that they will go off
3449 	 * at evenly distributed times even if all the filesystems
3450 	 * are mounted at once.
3451 	 */
3452 	next += incr;
3453 	if (next == 0 || next > syncer_maxdelay) {
3454 		start /= 2;
3455 		incr /= 2;
3456 		if (start == 0) {
3457 			start = syncer_maxdelay / 2;
3458 			incr = syncer_maxdelay;
3459 		}
3460 		next = start;
3461 	}
3462 	VI_LOCK(vp);
3463 	vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
3464 	VI_UNLOCK(vp);
3465 	mp->mnt_syncer = vp;
3466 	return (0);
3467 }
3468 
3469 /*
3470  * Do a lazy sync of the filesystem.
3471  */
3472 static int
3473 sync_fsync(ap)
3474 	struct vop_fsync_args /* {
3475 		struct vnode *a_vp;
3476 		struct ucred *a_cred;
3477 		int a_waitfor;
3478 		struct thread *a_td;
3479 	} */ *ap;
3480 {
3481 	struct vnode *syncvp = ap->a_vp;
3482 	struct mount *mp = syncvp->v_mount;
3483 	struct thread *td = ap->a_td;
3484 	int error, asyncflag;
3485 
3486 	/*
3487 	 * We only need to do something if this is a lazy evaluation.
3488 	 */
3489 	if (ap->a_waitfor != MNT_LAZY)
3490 		return (0);
3491 
3492 	/*
3493 	 * Move ourselves to the back of the sync list.
3494 	 */
3495 	VI_LOCK(syncvp);
3496 	vn_syncer_add_to_worklist(syncvp, syncdelay);
3497 	VI_UNLOCK(syncvp);
3498 
3499 	/*
3500 	 * Walk the list of vnodes pushing all that are dirty and
3501 	 * not already on the sync list.
3502 	 */
3503 	mtx_lock(&mountlist_mtx);
3504 	if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) {
3505 		mtx_unlock(&mountlist_mtx);
3506 		return (0);
3507 	}
3508 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3509 		vfs_unbusy(mp, td);
3510 		return (0);
3511 	}
3512 	asyncflag = mp->mnt_flag & MNT_ASYNC;
3513 	mp->mnt_flag &= ~MNT_ASYNC;
3514 	vfs_msync(mp, MNT_NOWAIT);
3515 	error = VFS_SYNC(mp, MNT_LAZY, ap->a_cred, td);
3516 	if (asyncflag)
3517 		mp->mnt_flag |= MNT_ASYNC;
3518 	vn_finished_write(mp);
3519 	vfs_unbusy(mp, td);
3520 	return (error);
3521 }
3522 
3523 /*
3524  * The syncer vnode is no referenced.
3525  */
3526 static int
3527 sync_inactive(ap)
3528 	struct vop_inactive_args /* {
3529 		struct vnode *a_vp;
3530 		struct thread *a_td;
3531 	} */ *ap;
3532 {
3533 
3534 	VOP_UNLOCK(ap->a_vp, 0, ap->a_td);
3535 	vgone(ap->a_vp);
3536 	return (0);
3537 }
3538 
3539 /*
3540  * The syncer vnode is no longer needed and is being decommissioned.
3541  *
3542  * Modifications to the worklist must be protected by sync_mtx.
3543  */
3544 static int
3545 sync_reclaim(ap)
3546 	struct vop_reclaim_args /* {
3547 		struct vnode *a_vp;
3548 	} */ *ap;
3549 {
3550 	struct vnode *vp = ap->a_vp;
3551 
3552 	VI_LOCK(vp);
3553 	vp->v_mount->mnt_syncer = NULL;
3554 	if (vp->v_iflag & VI_ONWORKLST) {
3555 		mtx_lock(&sync_mtx);
3556 		LIST_REMOVE(vp, v_synclist);
3557 		mtx_unlock(&sync_mtx);
3558 		vp->v_iflag &= ~VI_ONWORKLST;
3559 	}
3560 	VI_UNLOCK(vp);
3561 
3562 	return (0);
3563 }
3564 
3565 /*
3566  * extract the dev_t from a VCHR
3567  */
3568 dev_t
3569 vn_todev(vp)
3570 	struct vnode *vp;
3571 {
3572 	if (vp->v_type != VCHR)
3573 		return (NODEV);
3574 	return (vp->v_rdev);
3575 }
3576 
3577 /*
3578  * Check if vnode represents a disk device
3579  */
3580 int
3581 vn_isdisk(vp, errp)
3582 	struct vnode *vp;
3583 	int *errp;
3584 {
3585 	int error;
3586 
3587 	error = 0;
3588 	if (vp->v_type != VCHR)
3589 		error = ENOTBLK;
3590 	else if (vp->v_rdev == NULL)
3591 		error = ENXIO;
3592 	else if (!(devsw(vp->v_rdev)->d_flags & D_DISK))
3593 		error = ENOTBLK;
3594 	if (errp != NULL)
3595 		*errp = error;
3596 	return (error == 0);
3597 }
3598 
3599 /*
3600  * Free data allocated by namei(); see namei(9) for details.
3601  */
3602 void
3603 NDFREE(ndp, flags)
3604      struct nameidata *ndp;
3605      const u_int flags;
3606 {
3607 	if (!(flags & NDF_NO_FREE_PNBUF) &&
3608 	    (ndp->ni_cnd.cn_flags & HASBUF)) {
3609 		uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
3610 		ndp->ni_cnd.cn_flags &= ~HASBUF;
3611 	}
3612 	if (!(flags & NDF_NO_DVP_UNLOCK) &&
3613 	    (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
3614 	    ndp->ni_dvp != ndp->ni_vp)
3615 		VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_thread);
3616 	if (!(flags & NDF_NO_DVP_RELE) &&
3617 	    (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
3618 		vrele(ndp->ni_dvp);
3619 		ndp->ni_dvp = NULL;
3620 	}
3621 	if (!(flags & NDF_NO_VP_UNLOCK) &&
3622 	    (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
3623 		VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_thread);
3624 	if (!(flags & NDF_NO_VP_RELE) &&
3625 	    ndp->ni_vp) {
3626 		vrele(ndp->ni_vp);
3627 		ndp->ni_vp = NULL;
3628 	}
3629 	if (!(flags & NDF_NO_STARTDIR_RELE) &&
3630 	    (ndp->ni_cnd.cn_flags & SAVESTART)) {
3631 		vrele(ndp->ni_startdir);
3632 		ndp->ni_startdir = NULL;
3633 	}
3634 }
3635 
3636 /*
3637  * Common filesystem object access control check routine.  Accepts a
3638  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3639  * and optional call-by-reference privused argument allowing vaccess()
3640  * to indicate to the caller whether privilege was used to satisfy the
3641  * request (obsoleted).  Returns 0 on success, or an errno on failure.
3642  */
3643 int
3644 vaccess(type, file_mode, file_uid, file_gid, acc_mode, cred, privused)
3645 	enum vtype type;
3646 	mode_t file_mode;
3647 	uid_t file_uid;
3648 	gid_t file_gid;
3649 	mode_t acc_mode;
3650 	struct ucred *cred;
3651 	int *privused;
3652 {
3653 	mode_t dac_granted;
3654 #ifdef CAPABILITIES
3655 	mode_t cap_granted;
3656 #endif
3657 
3658 	/*
3659 	 * Look for a normal, non-privileged way to access the file/directory
3660 	 * as requested.  If it exists, go with that.
3661 	 */
3662 
3663 	if (privused != NULL)
3664 		*privused = 0;
3665 
3666 	dac_granted = 0;
3667 
3668 	/* Check the owner. */
3669 	if (cred->cr_uid == file_uid) {
3670 		dac_granted |= VADMIN;
3671 		if (file_mode & S_IXUSR)
3672 			dac_granted |= VEXEC;
3673 		if (file_mode & S_IRUSR)
3674 			dac_granted |= VREAD;
3675 		if (file_mode & S_IWUSR)
3676 			dac_granted |= (VWRITE | VAPPEND);
3677 
3678 		if ((acc_mode & dac_granted) == acc_mode)
3679 			return (0);
3680 
3681 		goto privcheck;
3682 	}
3683 
3684 	/* Otherwise, check the groups (first match) */
3685 	if (groupmember(file_gid, cred)) {
3686 		if (file_mode & S_IXGRP)
3687 			dac_granted |= VEXEC;
3688 		if (file_mode & S_IRGRP)
3689 			dac_granted |= VREAD;
3690 		if (file_mode & S_IWGRP)
3691 			dac_granted |= (VWRITE | VAPPEND);
3692 
3693 		if ((acc_mode & dac_granted) == acc_mode)
3694 			return (0);
3695 
3696 		goto privcheck;
3697 	}
3698 
3699 	/* Otherwise, check everyone else. */
3700 	if (file_mode & S_IXOTH)
3701 		dac_granted |= VEXEC;
3702 	if (file_mode & S_IROTH)
3703 		dac_granted |= VREAD;
3704 	if (file_mode & S_IWOTH)
3705 		dac_granted |= (VWRITE | VAPPEND);
3706 	if ((acc_mode & dac_granted) == acc_mode)
3707 		return (0);
3708 
3709 privcheck:
3710 	if (!suser_cred(cred, PRISON_ROOT)) {
3711 		/* XXX audit: privilege used */
3712 		if (privused != NULL)
3713 			*privused = 1;
3714 		return (0);
3715 	}
3716 
3717 #ifdef CAPABILITIES
3718 	/*
3719 	 * Build a capability mask to determine if the set of capabilities
3720 	 * satisfies the requirements when combined with the granted mask
3721 	 * from above.
3722 	 * For each capability, if the capability is required, bitwise
3723 	 * or the request type onto the cap_granted mask.
3724 	 */
3725 	cap_granted = 0;
3726 
3727 	if (type == VDIR) {
3728 		/*
3729 		 * For directories, use CAP_DAC_READ_SEARCH to satisfy
3730 		 * VEXEC requests, instead of CAP_DAC_EXECUTE.
3731 		 */
3732 		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3733 		    !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3734 			cap_granted |= VEXEC;
3735 	} else {
3736 		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3737 		    !cap_check(cred, NULL, CAP_DAC_EXECUTE, PRISON_ROOT))
3738 			cap_granted |= VEXEC;
3739 	}
3740 
3741 	if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
3742 	    !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3743 		cap_granted |= VREAD;
3744 
3745 	if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3746 	    !cap_check(cred, NULL, CAP_DAC_WRITE, PRISON_ROOT))
3747 		cap_granted |= (VWRITE | VAPPEND);
3748 
3749 	if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3750 	    !cap_check(cred, NULL, CAP_FOWNER, PRISON_ROOT))
3751 		cap_granted |= VADMIN;
3752 
3753 	if ((acc_mode & (cap_granted | dac_granted)) == acc_mode) {
3754 		/* XXX audit: privilege used */
3755 		if (privused != NULL)
3756 			*privused = 1;
3757 		return (0);
3758 	}
3759 #endif
3760 
3761 	return ((acc_mode & VADMIN) ? EPERM : EACCES);
3762 }
3763 
3764 /*
3765  * Credential check based on process requesting service, and per-attribute
3766  * permissions.
3767  */
3768 int
3769 extattr_check_cred(struct vnode *vp, int attrnamespace,
3770     struct ucred *cred, struct thread *td, int access)
3771 {
3772 
3773 	/*
3774 	 * Kernel-invoked always succeeds.
3775 	 */
3776 	if (cred == NOCRED)
3777 		return (0);
3778 
3779 	/*
3780 	 * Do not allow privileged processes in jail to directly
3781 	 * manipulate system attributes.
3782 	 *
3783 	 * XXX What capability should apply here?
3784 	 * Probably CAP_SYS_SETFFLAG.
3785 	 */
3786 	switch (attrnamespace) {
3787 	case EXTATTR_NAMESPACE_SYSTEM:
3788 		/* Potentially should be: return (EPERM); */
3789 		return (suser_cred(cred, 0));
3790 	case EXTATTR_NAMESPACE_USER:
3791 		return (VOP_ACCESS(vp, access, cred, td));
3792 	default:
3793 		return (EPERM);
3794 	}
3795 }
3796