xref: /freebsd/sys/kern/vfs_subr.c (revision c4f02a891fe62fe1277c89859922804ea2c27bcd)
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	vcanrecycle(struct vnode *vp, struct mount **vnmpp);
89 
90 
91 /*
92  * Number of vnodes in existence.  Increased whenever getnewvnode()
93  * allocates a new vnode, never decreased.
94  */
95 static unsigned long	numvnodes;
96 
97 SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
98 
99 /*
100  * Conversion tables for conversion from vnode types to inode formats
101  * and back.
102  */
103 enum vtype iftovt_tab[16] = {
104 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
105 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
106 };
107 int vttoif_tab[9] = {
108 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
109 	S_IFSOCK, S_IFIFO, S_IFMT,
110 };
111 
112 /*
113  * List of vnodes that are ready for recycling.
114  */
115 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
116 
117 /*
118  * Minimum number of free vnodes.  If there are fewer than this free vnodes,
119  * getnewvnode() will return a newly allocated vnode.
120  */
121 static u_long wantfreevnodes = 25;
122 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
123 /* Number of vnodes in the free list. */
124 static u_long freevnodes;
125 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
126 
127 /*
128  * Various variables used for debugging the new implementation of
129  * reassignbuf().
130  * XXX these are probably of (very) limited utility now.
131  */
132 static int reassignbufcalls;
133 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
134 static int nameileafonly;
135 SYSCTL_INT(_vfs, OID_AUTO, nameileafonly, CTLFLAG_RW, &nameileafonly, 0, "");
136 
137 /*
138  * Cache for the mount type id assigned to NFS.  This is used for
139  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
140  */
141 int	nfs_mount_type = -1;
142 
143 /* To keep more than one thread at a time from running vfs_getnewfsid */
144 static struct mtx mntid_mtx;
145 
146 /*
147  * Lock for any access to the following:
148  *	vnode_free_list
149  *	numvnodes
150  *	freevnodes
151  */
152 static struct mtx vnode_free_list_mtx;
153 
154 /*
155  * For any iteration/modification of dev->si_hlist (linked through
156  * v_specnext)
157  */
158 static struct mtx spechash_mtx;
159 
160 /* Publicly exported FS */
161 struct nfs_public nfs_pub;
162 
163 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
164 static uma_zone_t vnode_zone;
165 static uma_zone_t vnodepoll_zone;
166 
167 /* Set to 1 to print out reclaim of active vnodes */
168 int	prtactive;
169 
170 /*
171  * The workitem queue.
172  *
173  * It is useful to delay writes of file data and filesystem metadata
174  * for tens of seconds so that quickly created and deleted files need
175  * not waste disk bandwidth being created and removed. To realize this,
176  * we append vnodes to a "workitem" queue. When running with a soft
177  * updates implementation, most pending metadata dependencies should
178  * not wait for more than a few seconds. Thus, mounted on block devices
179  * are delayed only about a half the time that file data is delayed.
180  * Similarly, directory updates are more critical, so are only delayed
181  * about a third the time that file data is delayed. Thus, there are
182  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
183  * one each second (driven off the filesystem syncer process). The
184  * syncer_delayno variable indicates the next queue that is to be processed.
185  * Items that need to be processed soon are placed in this queue:
186  *
187  *	syncer_workitem_pending[syncer_delayno]
188  *
189  * A delay of fifteen seconds is done by placing the request fifteen
190  * entries later in the queue:
191  *
192  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
193  *
194  */
195 static int syncer_delayno;
196 static long syncer_mask;
197 LIST_HEAD(synclist, vnode);
198 static struct synclist *syncer_workitem_pending;
199 /*
200  * The sync_mtx protects:
201  *	vp->v_synclist
202  *	syncer_delayno
203  *	syncer_workitem_pending
204  *	rushjob
205  */
206 static struct mtx sync_mtx;
207 
208 #define SYNCER_MAXDELAY		32
209 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
210 static int syncdelay = 30;		/* max time to delay syncing data */
211 static int filedelay = 30;		/* time to delay syncing files */
212 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
213 static int dirdelay = 29;		/* time to delay syncing directories */
214 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
215 static int metadelay = 28;		/* time to delay syncing metadata */
216 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
217 static int rushjob;		/* number of slots to run ASAP */
218 static int stat_rush_requests;	/* number of times I/O speeded up */
219 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
220 
221 /*
222  * Number of vnodes we want to exist at any one time.  This is mostly used
223  * to size hash tables in vnode-related code.  It is normally not used in
224  * getnewvnode(), as wantfreevnodes is normally nonzero.)
225  *
226  * XXX desiredvnodes is historical cruft and should not exist.
227  */
228 int desiredvnodes;
229 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
230     &desiredvnodes, 0, "Maximum number of vnodes");
231 static int minvnodes;
232 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
233     &minvnodes, 0, "Minimum number of vnodes");
234 static int vnlru_nowhere;
235 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0,
236     "Number of times the vnlru process ran without success");
237 
238 /* Hook for calling soft updates */
239 int (*softdep_process_worklist_hook)(struct mount *);
240 
241 /*
242  * This only exists to supress warnings from unlocked specfs accesses.  It is
243  * no longer ok to have an unlocked VFS.
244  */
245 #define IGNORE_LOCK(vp) ((vp)->v_type == VCHR || (vp)->v_type == VBAD)
246 
247 /* Print lock violations */
248 int vfs_badlock_print = 1;
249 
250 /* Panic on violation */
251 int vfs_badlock_panic = 1;
252 
253 /* Check for interlock across VOPs */
254 int vfs_badlock_mutex = 1;
255 
256 static void
257 vfs_badlock(char *msg, char *str, struct vnode *vp)
258 {
259 	if (vfs_badlock_print)
260 		printf("%s: %p %s\n", str, vp, msg);
261 	if (vfs_badlock_panic)
262 		Debugger("Lock violation.\n");
263 }
264 
265 void
266 assert_vi_unlocked(struct vnode *vp, char *str)
267 {
268 	if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
269 		vfs_badlock("interlock is locked but should not be", str, vp);
270 }
271 
272 void
273 assert_vi_locked(struct vnode *vp, char *str)
274 {
275 	if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
276 		vfs_badlock("interlock is not locked but should be", str, vp);
277 }
278 
279 void
280 assert_vop_locked(struct vnode *vp, char *str)
281 {
282 	if (vp && !IGNORE_LOCK(vp) && !VOP_ISLOCKED(vp, NULL))
283 		vfs_badlock("is not locked but should be", str, vp);
284 }
285 
286 void
287 assert_vop_unlocked(struct vnode *vp, char *str)
288 {
289 	if (vp && !IGNORE_LOCK(vp) &&
290 	    VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE)
291 		vfs_badlock("is locked but should not be", str, vp);
292 }
293 
294 void
295 assert_vop_elocked(struct vnode *vp, char *str)
296 {
297 	if (vp && !IGNORE_LOCK(vp) &&
298 	    VOP_ISLOCKED(vp, curthread) != LK_EXCLUSIVE)
299 		vfs_badlock("is not exclusive locked but should be", str, vp);
300 }
301 
302 void
303 assert_vop_elocked_other(struct vnode *vp, char *str)
304 {
305 	if (vp && !IGNORE_LOCK(vp) &&
306 	    VOP_ISLOCKED(vp, curthread) != LK_EXCLOTHER)
307 		vfs_badlock("is not exclusive locked by another thread",
308 		    str, vp);
309 }
310 
311 void
312 assert_vop_slocked(struct vnode *vp, char *str)
313 {
314 	if (vp && !IGNORE_LOCK(vp) &&
315 	    VOP_ISLOCKED(vp, curthread) != LK_SHARED)
316 		vfs_badlock("is not locked shared but should be", str, vp);
317 }
318 
319 void
320 vop_rename_pre(void *ap)
321 {
322 	struct vop_rename_args *a = ap;
323 
324 	if (a->a_tvp)
325 		ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
326 	ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
327 	ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
328 	ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
329 
330 	/* Check the source (from) */
331 	if (a->a_tdvp != a->a_fdvp)
332 		ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked.\n");
333 	if (a->a_tvp != a->a_fvp)
334 		ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: tvp locked.\n");
335 
336 	/* Check the target */
337 	if (a->a_tvp)
338 		ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked.\n");
339 
340 	ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked.\n");
341 }
342 
343 void
344 vop_strategy_pre(void *ap)
345 {
346 	struct vop_strategy_args *a = ap;
347 	struct buf *bp;
348 
349 	bp = a->a_bp;
350 
351 	/*
352 	 * Cluster ops lock their component buffers but not the IO container.
353 	 */
354 	if ((bp->b_flags & B_CLUSTER) != 0)
355 		return;
356 
357 	if (BUF_REFCNT(bp) < 1) {
358 		if (vfs_badlock_print)
359 			printf("VOP_STRATEGY: bp is not locked but should be.\n");
360 		if (vfs_badlock_panic)
361 			Debugger("Lock violation.\n");
362 	}
363 }
364 
365 void
366 vop_lookup_pre(void *ap)
367 {
368 	struct vop_lookup_args *a = ap;
369 	struct vnode *dvp;
370 
371 	dvp = a->a_dvp;
372 
373 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
374 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
375 }
376 
377 void
378 vop_lookup_post(void *ap, int rc)
379 {
380 	struct vop_lookup_args *a = ap;
381 	struct componentname *cnp;
382 	struct vnode *dvp;
383 	struct vnode *vp;
384 	int flags;
385 
386 	dvp = a->a_dvp;
387 	cnp = a->a_cnp;
388 	vp = *(a->a_vpp);
389 	flags = cnp->cn_flags;
390 
391 
392 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
393 	/*
394 	 * If this is the last path component for this lookup and LOCPARENT
395 	 * is set, OR if there is an error the directory has to be locked.
396 	 */
397 	if ((flags & LOCKPARENT) && (flags & ISLASTCN))
398 		ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP (LOCKPARENT)");
399 	else if (rc != 0)
400 		ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP (error)");
401 	else if (dvp != vp)
402 		ASSERT_VOP_UNLOCKED(dvp, "VOP_LOOKUP (dvp)");
403 
404 	if (flags & PDIRUNLOCK)
405 		ASSERT_VOP_UNLOCKED(dvp, "VOP_LOOKUP (PDIRUNLOCK)");
406 }
407 
408 void
409 vop_unlock_pre(void *ap)
410 {
411 	struct vop_unlock_args *a = ap;
412 
413 	if (a->a_flags & LK_INTERLOCK)
414 		ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
415 
416 	ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
417 }
418 
419 void
420 vop_unlock_post(void *ap, int rc)
421 {
422 	struct vop_unlock_args *a = ap;
423 
424 	if (a->a_flags & LK_INTERLOCK)
425 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
426 }
427 
428 void
429 vop_lock_pre(void *ap)
430 {
431 	struct vop_lock_args *a = ap;
432 
433 	if ((a->a_flags & LK_INTERLOCK) == 0)
434 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
435 	else
436 		ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
437 }
438 
439 void
440 vop_lock_post(void *ap, int rc)
441 {
442 	struct vop_lock_args *a;
443 
444 	a = ap;
445 
446 	ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
447 	if (rc == 0)
448 		ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
449 }
450 
451 void
452 v_addpollinfo(struct vnode *vp)
453 {
454 	vp->v_pollinfo = uma_zalloc(vnodepoll_zone, M_WAITOK);
455 	mtx_init(&vp->v_pollinfo->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
456 }
457 
458 /*
459  * Initialize the vnode management data structures.
460  */
461 static void
462 vntblinit(void *dummy __unused)
463 {
464 
465 	/*
466 	 * Desiredvnodes is a function of the physical memory size and
467 	 * the kernel's heap size.  Specifically, desiredvnodes scales
468 	 * in proportion to the physical memory size until two fifths
469 	 * of the kernel's heap size is consumed by vnodes and vm
470 	 * objects.
471 	 */
472 	desiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * vm_kmem_size /
473 	    (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
474 	minvnodes = desiredvnodes / 4;
475 	mtx_init(&mountlist_mtx, "mountlist", NULL, MTX_DEF);
476 	mtx_init(&mntvnode_mtx, "mntvnode", NULL, MTX_DEF);
477 	mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
478 	mtx_init(&spechash_mtx, "spechash", NULL, MTX_DEF);
479 	TAILQ_INIT(&vnode_free_list);
480 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
481 	vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
482 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
483 	vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
484 	      NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
485 	/*
486 	 * Initialize the filesystem syncer.
487 	 */
488 	syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
489 		&syncer_mask);
490 	syncer_maxdelay = syncer_mask + 1;
491 	mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
492 }
493 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL)
494 
495 
496 /*
497  * Mark a mount point as busy. Used to synchronize access and to delay
498  * unmounting. Interlock is not released on failure.
499  */
500 int
501 vfs_busy(mp, flags, interlkp, td)
502 	struct mount *mp;
503 	int flags;
504 	struct mtx *interlkp;
505 	struct thread *td;
506 {
507 	int lkflags;
508 
509 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
510 		if (flags & LK_NOWAIT)
511 			return (ENOENT);
512 		mp->mnt_kern_flag |= MNTK_MWAIT;
513 		/*
514 		 * Since all busy locks are shared except the exclusive
515 		 * lock granted when unmounting, the only place that a
516 		 * wakeup needs to be done is at the release of the
517 		 * exclusive lock at the end of dounmount.
518 		 */
519 		msleep(mp, interlkp, PVFS, "vfs_busy", 0);
520 		return (ENOENT);
521 	}
522 	lkflags = LK_SHARED | LK_NOPAUSE;
523 	if (interlkp)
524 		lkflags |= LK_INTERLOCK;
525 	if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
526 		panic("vfs_busy: unexpected lock failure");
527 	return (0);
528 }
529 
530 /*
531  * Free a busy filesystem.
532  */
533 void
534 vfs_unbusy(mp, td)
535 	struct mount *mp;
536 	struct thread *td;
537 {
538 
539 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
540 }
541 
542 /*
543  * Lookup a mount point by filesystem identifier.
544  */
545 struct mount *
546 vfs_getvfs(fsid)
547 	fsid_t *fsid;
548 {
549 	register struct mount *mp;
550 
551 	mtx_lock(&mountlist_mtx);
552 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
553 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
554 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
555 			mtx_unlock(&mountlist_mtx);
556 			return (mp);
557 		}
558 	}
559 	mtx_unlock(&mountlist_mtx);
560 	return ((struct mount *) 0);
561 }
562 
563 /*
564  * Get a new unique fsid.  Try to make its val[0] unique, since this value
565  * will be used to create fake device numbers for stat().  Also try (but
566  * not so hard) make its val[0] unique mod 2^16, since some emulators only
567  * support 16-bit device numbers.  We end up with unique val[0]'s for the
568  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
569  *
570  * Keep in mind that several mounts may be running in parallel.  Starting
571  * the search one past where the previous search terminated is both a
572  * micro-optimization and a defense against returning the same fsid to
573  * different mounts.
574  */
575 void
576 vfs_getnewfsid(mp)
577 	struct mount *mp;
578 {
579 	static u_int16_t mntid_base;
580 	fsid_t tfsid;
581 	int mtype;
582 
583 	mtx_lock(&mntid_mtx);
584 	mtype = mp->mnt_vfc->vfc_typenum;
585 	tfsid.val[1] = mtype;
586 	mtype = (mtype & 0xFF) << 24;
587 	for (;;) {
588 		tfsid.val[0] = makeudev(255,
589 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
590 		mntid_base++;
591 		if (vfs_getvfs(&tfsid) == NULL)
592 			break;
593 	}
594 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
595 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
596 	mtx_unlock(&mntid_mtx);
597 }
598 
599 /*
600  * Knob to control the precision of file timestamps:
601  *
602  *   0 = seconds only; nanoseconds zeroed.
603  *   1 = seconds and nanoseconds, accurate within 1/HZ.
604  *   2 = seconds and nanoseconds, truncated to microseconds.
605  * >=3 = seconds and nanoseconds, maximum precision.
606  */
607 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
608 
609 static int timestamp_precision = TSP_SEC;
610 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
611     &timestamp_precision, 0, "");
612 
613 /*
614  * Get a current timestamp.
615  */
616 void
617 vfs_timestamp(tsp)
618 	struct timespec *tsp;
619 {
620 	struct timeval tv;
621 
622 	switch (timestamp_precision) {
623 	case TSP_SEC:
624 		tsp->tv_sec = time_second;
625 		tsp->tv_nsec = 0;
626 		break;
627 	case TSP_HZ:
628 		getnanotime(tsp);
629 		break;
630 	case TSP_USEC:
631 		microtime(&tv);
632 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
633 		break;
634 	case TSP_NSEC:
635 	default:
636 		nanotime(tsp);
637 		break;
638 	}
639 }
640 
641 /*
642  * Set vnode attributes to VNOVAL
643  */
644 void
645 vattr_null(vap)
646 	register struct vattr *vap;
647 {
648 
649 	vap->va_type = VNON;
650 	vap->va_size = VNOVAL;
651 	vap->va_bytes = VNOVAL;
652 	vap->va_mode = VNOVAL;
653 	vap->va_nlink = VNOVAL;
654 	vap->va_uid = VNOVAL;
655 	vap->va_gid = VNOVAL;
656 	vap->va_fsid = VNOVAL;
657 	vap->va_fileid = VNOVAL;
658 	vap->va_blocksize = VNOVAL;
659 	vap->va_rdev = VNOVAL;
660 	vap->va_atime.tv_sec = VNOVAL;
661 	vap->va_atime.tv_nsec = VNOVAL;
662 	vap->va_mtime.tv_sec = VNOVAL;
663 	vap->va_mtime.tv_nsec = VNOVAL;
664 	vap->va_ctime.tv_sec = VNOVAL;
665 	vap->va_ctime.tv_nsec = VNOVAL;
666 	vap->va_birthtime.tv_sec = VNOVAL;
667 	vap->va_birthtime.tv_nsec = VNOVAL;
668 	vap->va_flags = VNOVAL;
669 	vap->va_gen = VNOVAL;
670 	vap->va_vaflags = 0;
671 }
672 
673 /*
674  * This routine is called when we have too many vnodes.  It attempts
675  * to free <count> vnodes and will potentially free vnodes that still
676  * have VM backing store (VM backing store is typically the cause
677  * of a vnode blowout so we want to do this).  Therefore, this operation
678  * is not considered cheap.
679  *
680  * A number of conditions may prevent a vnode from being reclaimed.
681  * the buffer cache may have references on the vnode, a directory
682  * vnode may still have references due to the namei cache representing
683  * underlying files, or the vnode may be in active use.   It is not
684  * desireable to reuse such vnodes.  These conditions may cause the
685  * number of vnodes to reach some minimum value regardless of what
686  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
687  */
688 static int
689 vlrureclaim(struct mount *mp)
690 {
691 	struct vnode *vp;
692 	int done;
693 	int trigger;
694 	int usevnodes;
695 	int count;
696 
697 	/*
698 	 * Calculate the trigger point, don't allow user
699 	 * screwups to blow us up.   This prevents us from
700 	 * recycling vnodes with lots of resident pages.  We
701 	 * aren't trying to free memory, we are trying to
702 	 * free vnodes.
703 	 */
704 	usevnodes = desiredvnodes;
705 	if (usevnodes <= 0)
706 		usevnodes = 1;
707 	trigger = cnt.v_page_count * 2 / usevnodes;
708 
709 	done = 0;
710 	mtx_lock(&mntvnode_mtx);
711 	count = mp->mnt_nvnodelistsize / 10 + 1;
712 	while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
713 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
714 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
715 
716 		if (vp->v_type != VNON &&
717 		    vp->v_type != VBAD &&
718 		    VI_TRYLOCK(vp)) {
719 			if (VMIGHTFREE(vp) &&           /* critical path opt */
720 			    (vp->v_object == NULL ||
721 			    vp->v_object->resident_page_count < trigger)) {
722 				mtx_unlock(&mntvnode_mtx);
723 				vgonel(vp, curthread);
724 				done++;
725 				mtx_lock(&mntvnode_mtx);
726 			} else
727 				VI_UNLOCK(vp);
728 		}
729 		--count;
730 	}
731 	mtx_unlock(&mntvnode_mtx);
732 	return done;
733 }
734 
735 /*
736  * Attempt to recycle vnodes in a context that is always safe to block.
737  * Calling vlrurecycle() from the bowels of filesystem code has some
738  * interesting deadlock problems.
739  */
740 static struct proc *vnlruproc;
741 static int vnlruproc_sig;
742 
743 static void
744 vnlru_proc(void)
745 {
746 	struct mount *mp, *nmp;
747 	int done;
748 	struct proc *p = vnlruproc;
749 	struct thread *td = FIRST_THREAD_IN_PROC(p);	/* XXXKSE */
750 
751 	mtx_lock(&Giant);
752 
753 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
754 	    SHUTDOWN_PRI_FIRST);
755 
756 	for (;;) {
757 		kthread_suspend_check(p);
758 		mtx_lock(&vnode_free_list_mtx);
759 		if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
760 			mtx_unlock(&vnode_free_list_mtx);
761 			vnlruproc_sig = 0;
762 			wakeup(&vnlruproc_sig);
763 			tsleep(vnlruproc, PVFS, "vlruwt", hz);
764 			continue;
765 		}
766 		mtx_unlock(&vnode_free_list_mtx);
767 		done = 0;
768 		mtx_lock(&mountlist_mtx);
769 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
770 			if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
771 				nmp = TAILQ_NEXT(mp, mnt_list);
772 				continue;
773 			}
774 			done += vlrureclaim(mp);
775 			mtx_lock(&mountlist_mtx);
776 			nmp = TAILQ_NEXT(mp, mnt_list);
777 			vfs_unbusy(mp, td);
778 		}
779 		mtx_unlock(&mountlist_mtx);
780 		if (done == 0) {
781 #if 0
782 			/* These messages are temporary debugging aids */
783 			if (vnlru_nowhere < 5)
784 				printf("vnlru process getting nowhere..\n");
785 			else if (vnlru_nowhere == 5)
786 				printf("vnlru process messages stopped.\n");
787 #endif
788 			vnlru_nowhere++;
789 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
790 		}
791 	}
792 }
793 
794 static struct kproc_desc vnlru_kp = {
795 	"vnlru",
796 	vnlru_proc,
797 	&vnlruproc
798 };
799 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
800 
801 
802 /*
803  * Routines having to do with the management of the vnode table.
804  */
805 
806 /*
807  * Check to see if a free vnode can be recycled. If it can,
808  * return it locked with the vn lock, but not interlock. Also
809  * get the vn_start_write lock. Otherwise indicate the error.
810  */
811 static int
812 vcanrecycle(struct vnode *vp, struct mount **vnmpp)
813 {
814 	struct thread *td = curthread;
815 	vm_object_t object;
816 	int error;
817 
818 	/* Don't recycle if we can't get the interlock */
819 	if (!VI_TRYLOCK(vp))
820 		return (EWOULDBLOCK);
821 
822 	/* We should be able to immediately acquire this */
823 	/* XXX This looks like it should panic if it fails */
824 	if (vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE, td) != 0) {
825 		if (VOP_ISLOCKED(vp, td))
826 			panic("vcanrecycle: locked vnode");
827 		return (EWOULDBLOCK);
828 	}
829 
830 	/*
831 	 * Don't recycle if its filesystem is being suspended.
832 	 */
833 	if (vn_start_write(vp, vnmpp, V_NOWAIT) != 0) {
834 		error = EBUSY;
835 		goto done;
836 	}
837 
838 	/*
839 	 * Don't recycle if we still have cached pages.
840 	 */
841 	if (VOP_GETVOBJECT(vp, &object) == 0) {
842 		VM_OBJECT_LOCK(object);
843 		if (object->resident_page_count ||
844 		    object->ref_count) {
845 			VM_OBJECT_UNLOCK(object);
846 			error = EBUSY;
847 			goto done;
848 		}
849 		VM_OBJECT_UNLOCK(object);
850 	}
851 	if (LIST_FIRST(&vp->v_cache_src)) {
852 		/*
853 		 * note: nameileafonly sysctl is temporary,
854 		 * for debugging only, and will eventually be
855 		 * removed.
856 		 */
857 		if (nameileafonly > 0) {
858 			/*
859 			 * Do not reuse namei-cached directory
860 			 * vnodes that have cached
861 			 * subdirectories.
862 			 */
863 			if (cache_leaf_test(vp) < 0) {
864 				error = EISDIR;
865 				goto done;
866 			}
867 		} else if (nameileafonly < 0 ||
868 			    vmiodirenable == 0) {
869 			/*
870 			 * Do not reuse namei-cached directory
871 			 * vnodes if nameileafonly is -1 or
872 			 * if VMIO backing for directories is
873 			 * turned off (otherwise we reuse them
874 			 * too quickly).
875 			 */
876 			error = EBUSY;
877 			goto done;
878 		}
879 	}
880 	return (0);
881 done:
882 	VOP_UNLOCK(vp, 0, td);
883 	return (error);
884 }
885 
886 /*
887  * Return the next vnode from the free list.
888  */
889 int
890 getnewvnode(tag, mp, vops, vpp)
891 	const char *tag;
892 	struct mount *mp;
893 	vop_t **vops;
894 	struct vnode **vpp;
895 {
896 	struct thread *td = curthread;	/* XXX */
897 	struct vnode *vp = NULL;
898 	struct vpollinfo *pollinfo = NULL;
899 	struct mount *vnmp;
900 
901 	mtx_lock(&vnode_free_list_mtx);
902 
903 	/*
904 	 * Try to reuse vnodes if we hit the max.  This situation only
905 	 * occurs in certain large-memory (2G+) situations.  We cannot
906 	 * attempt to directly reclaim vnodes due to nasty recursion
907 	 * problems.
908 	 */
909 	while (numvnodes - freevnodes > desiredvnodes) {
910 		if (vnlruproc_sig == 0) {
911 			vnlruproc_sig = 1;      /* avoid unnecessary wakeups */
912 			wakeup(vnlruproc);
913 		}
914 		mtx_unlock(&vnode_free_list_mtx);
915 		tsleep(&vnlruproc_sig, PVFS, "vlruwk", hz);
916 		mtx_lock(&vnode_free_list_mtx);
917 	}
918 
919 	/*
920 	 * Attempt to reuse a vnode already on the free list, allocating
921 	 * a new vnode if we can't find one or if we have not reached a
922 	 * good minimum for good LRU performance.
923 	 */
924 
925 	if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
926 		int error;
927 		int count;
928 
929 		for (count = 0; count < freevnodes; count++) {
930 			vp = TAILQ_FIRST(&vnode_free_list);
931 
932 			KASSERT(vp->v_usecount == 0 &&
933 			    (vp->v_iflag & VI_DOINGINACT) == 0,
934 			    ("getnewvnode: free vnode isn't"));
935 
936 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
937 			/*
938 			 * We have to drop the free list mtx to avoid lock
939 			 * order reversals with interlock.
940 			 */
941 			mtx_unlock(&vnode_free_list_mtx);
942 			error = vcanrecycle(vp, &vnmp);
943 			mtx_lock(&vnode_free_list_mtx);
944 			if (error == 0)
945 				break;
946 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
947 			vp = NULL;
948 		}
949 	}
950 	if (vp) {
951 		freevnodes--;
952 		mtx_unlock(&vnode_free_list_mtx);
953 
954 		cache_purge(vp);
955 		VI_LOCK(vp);
956 		vp->v_iflag |= VI_DOOMED;
957 		vp->v_iflag &= ~VI_FREE;
958 		if (vp->v_type != VBAD) {
959 			VOP_UNLOCK(vp, 0, td);
960 			vgonel(vp, td);
961 			VI_LOCK(vp);
962 		} else {
963 			VOP_UNLOCK(vp, 0, td);
964 		}
965 		vn_finished_write(vnmp);
966 
967 #ifdef INVARIANTS
968 		{
969 			if (vp->v_data)
970 				panic("cleaned vnode isn't");
971 			if (vp->v_numoutput)
972 				panic("Clean vnode has pending I/O's");
973 			if (vp->v_writecount != 0)
974 				panic("Non-zero write count");
975 		}
976 #endif
977 		if ((pollinfo = vp->v_pollinfo) != NULL) {
978 			/*
979 			 * To avoid lock order reversals, the call to
980 			 * uma_zfree() must be delayed until the vnode
981 			 * interlock is released.
982 			 */
983 			vp->v_pollinfo = NULL;
984 		}
985 #ifdef MAC
986 		mac_destroy_vnode(vp);
987 #endif
988 		vp->v_iflag = 0;
989 		vp->v_vflag = 0;
990 		vp->v_lastw = 0;
991 		vp->v_lasta = 0;
992 		vp->v_cstart = 0;
993 		vp->v_clen = 0;
994 		vp->v_socket = 0;
995 		lockdestroy(vp->v_vnlock);
996 		lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOPAUSE);
997 		KASSERT(vp->v_cleanbufcnt == 0, ("cleanbufcnt not 0"));
998 		KASSERT(vp->v_cleanblkroot == NULL, ("cleanblkroot not NULL"));
999 		KASSERT(vp->v_dirtybufcnt == 0, ("dirtybufcnt not 0"));
1000 		KASSERT(vp->v_dirtyblkroot == NULL, ("dirtyblkroot not NULL"));
1001 	} else {
1002 		numvnodes++;
1003 		mtx_unlock(&vnode_free_list_mtx);
1004 
1005 		vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
1006 		mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
1007 		VI_LOCK(vp);
1008 		vp->v_dd = vp;
1009 		vp->v_vnlock = &vp->v_lock;
1010 		lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOPAUSE);
1011 		cache_purge(vp);
1012 		LIST_INIT(&vp->v_cache_src);
1013 		TAILQ_INIT(&vp->v_cache_dst);
1014 	}
1015 
1016 	TAILQ_INIT(&vp->v_cleanblkhd);
1017 	TAILQ_INIT(&vp->v_dirtyblkhd);
1018 	vp->v_type = VNON;
1019 	vp->v_tag = tag;
1020 	vp->v_op = vops;
1021 	*vpp = vp;
1022 	vp->v_usecount = 1;
1023 	vp->v_data = 0;
1024 	vp->v_cachedid = -1;
1025 	VI_UNLOCK(vp);
1026 	if (pollinfo != NULL) {
1027 		mtx_destroy(&pollinfo->vpi_lock);
1028 		uma_zfree(vnodepoll_zone, pollinfo);
1029 	}
1030 #ifdef MAC
1031 	mac_init_vnode(vp);
1032 	if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1033 		mac_associate_vnode_singlelabel(mp, vp);
1034 #endif
1035 	insmntque(vp, mp);
1036 
1037 	return (0);
1038 }
1039 
1040 /*
1041  * Move a vnode from one mount queue to another.
1042  */
1043 static void
1044 insmntque(vp, mp)
1045 	register struct vnode *vp;
1046 	register struct mount *mp;
1047 {
1048 
1049 	mtx_lock(&mntvnode_mtx);
1050 	/*
1051 	 * Delete from old mount point vnode list, if on one.
1052 	 */
1053 	if (vp->v_mount != NULL) {
1054 		KASSERT(vp->v_mount->mnt_nvnodelistsize > 0,
1055 			("bad mount point vnode list size"));
1056 		TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
1057 		vp->v_mount->mnt_nvnodelistsize--;
1058 	}
1059 	/*
1060 	 * Insert into list of vnodes for the new mount point, if available.
1061 	 */
1062 	if ((vp->v_mount = mp) == NULL) {
1063 		mtx_unlock(&mntvnode_mtx);
1064 		return;
1065 	}
1066 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1067 	mp->mnt_nvnodelistsize++;
1068 	mtx_unlock(&mntvnode_mtx);
1069 }
1070 
1071 /*
1072  * Update outstanding I/O count and do wakeup if requested.
1073  */
1074 void
1075 vwakeup(bp)
1076 	register struct buf *bp;
1077 {
1078 	register struct vnode *vp;
1079 
1080 	bp->b_flags &= ~B_WRITEINPROG;
1081 	if ((vp = bp->b_vp)) {
1082 		VI_LOCK(vp);
1083 		vp->v_numoutput--;
1084 		if (vp->v_numoutput < 0)
1085 			panic("vwakeup: neg numoutput");
1086 		if ((vp->v_numoutput == 0) && (vp->v_iflag & VI_BWAIT)) {
1087 			vp->v_iflag &= ~VI_BWAIT;
1088 			wakeup(&vp->v_numoutput);
1089 		}
1090 		VI_UNLOCK(vp);
1091 	}
1092 }
1093 
1094 /*
1095  * Flush out and invalidate all buffers associated with a vnode.
1096  * Called with the underlying object locked.
1097  */
1098 int
1099 vinvalbuf(vp, flags, cred, td, slpflag, slptimeo)
1100 	struct vnode *vp;
1101 	int flags;
1102 	struct ucred *cred;
1103 	struct thread *td;
1104 	int slpflag, slptimeo;
1105 {
1106 	struct buf *blist;
1107 	int error;
1108 	vm_object_t object;
1109 
1110 	GIANT_REQUIRED;
1111 
1112 	ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1113 
1114 	VI_LOCK(vp);
1115 	if (flags & V_SAVE) {
1116 		while (vp->v_numoutput) {
1117 			vp->v_iflag |= VI_BWAIT;
1118 			error = msleep(&vp->v_numoutput, VI_MTX(vp),
1119 			    slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo);
1120 			if (error) {
1121 				VI_UNLOCK(vp);
1122 				return (error);
1123 			}
1124 		}
1125 		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1126 			VI_UNLOCK(vp);
1127 			if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, td)) != 0)
1128 				return (error);
1129 			/*
1130 			 * XXX We could save a lock/unlock if this was only
1131 			 * enabled under INVARIANTS
1132 			 */
1133 			VI_LOCK(vp);
1134 			if (vp->v_numoutput > 0 ||
1135 			    !TAILQ_EMPTY(&vp->v_dirtyblkhd))
1136 				panic("vinvalbuf: dirty bufs");
1137 		}
1138 	}
1139 	/*
1140 	 * If you alter this loop please notice that interlock is dropped and
1141 	 * reacquired in flushbuflist.  Special care is needed to ensure that
1142 	 * no race conditions occur from this.
1143 	 */
1144 	for (error = 0;;) {
1145 		if ((blist = TAILQ_FIRST(&vp->v_cleanblkhd)) != 0 &&
1146 		    flushbuflist(blist, flags, vp, slpflag, slptimeo, &error)) {
1147 			if (error)
1148 				break;
1149 			continue;
1150 		}
1151 		if ((blist = TAILQ_FIRST(&vp->v_dirtyblkhd)) != 0 &&
1152 		    flushbuflist(blist, flags, vp, slpflag, slptimeo, &error)) {
1153 			if (error)
1154 				break;
1155 			continue;
1156 		}
1157 		break;
1158 	}
1159 	if (error) {
1160 		VI_UNLOCK(vp);
1161 		return (error);
1162 	}
1163 
1164 	/*
1165 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1166 	 * have write I/O in-progress but if there is a VM object then the
1167 	 * VM object can also have read-I/O in-progress.
1168 	 */
1169 	do {
1170 		while (vp->v_numoutput > 0) {
1171 			vp->v_iflag |= VI_BWAIT;
1172 			msleep(&vp->v_numoutput, VI_MTX(vp), PVM, "vnvlbv", 0);
1173 		}
1174 		VI_UNLOCK(vp);
1175 		if (VOP_GETVOBJECT(vp, &object) == 0) {
1176 			VM_OBJECT_LOCK(object);
1177 			vm_object_pip_wait(object, "vnvlbx");
1178 			VM_OBJECT_UNLOCK(object);
1179 		}
1180 		VI_LOCK(vp);
1181 	} while (vp->v_numoutput > 0);
1182 	VI_UNLOCK(vp);
1183 
1184 	/*
1185 	 * Destroy the copy in the VM cache, too.
1186 	 */
1187 	if (VOP_GETVOBJECT(vp, &object) == 0) {
1188 		VM_OBJECT_LOCK(object);
1189 		vm_object_page_remove(object, 0, 0,
1190 			(flags & V_SAVE) ? TRUE : FALSE);
1191 		VM_OBJECT_UNLOCK(object);
1192 	}
1193 
1194 #ifdef INVARIANTS
1195 	VI_LOCK(vp);
1196 	if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1197 	    (!TAILQ_EMPTY(&vp->v_dirtyblkhd) ||
1198 	     !TAILQ_EMPTY(&vp->v_cleanblkhd)))
1199 		panic("vinvalbuf: flush failed");
1200 	VI_UNLOCK(vp);
1201 #endif
1202 	return (0);
1203 }
1204 
1205 /*
1206  * Flush out buffers on the specified list.
1207  *
1208  */
1209 static int
1210 flushbuflist(blist, flags, vp, slpflag, slptimeo, errorp)
1211 	struct buf *blist;
1212 	int flags;
1213 	struct vnode *vp;
1214 	int slpflag, slptimeo;
1215 	int *errorp;
1216 {
1217 	struct buf *bp, *nbp;
1218 	int found, error;
1219 
1220 	ASSERT_VI_LOCKED(vp, "flushbuflist");
1221 
1222 	for (found = 0, bp = blist; bp; bp = nbp) {
1223 		nbp = TAILQ_NEXT(bp, b_vnbufs);
1224 		if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1225 		    ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1226 			continue;
1227 		}
1228 		found += 1;
1229 		error = BUF_TIMELOCK(bp,
1230 		    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, VI_MTX(vp),
1231 		    "flushbuf", slpflag, slptimeo);
1232 		if (error) {
1233 			if (error != ENOLCK)
1234 				*errorp = error;
1235 			goto done;
1236 		}
1237 		/*
1238 		 * XXX Since there are no node locks for NFS, I
1239 		 * believe there is a slight chance that a delayed
1240 		 * write will occur while sleeping just above, so
1241 		 * check for it.  Note that vfs_bio_awrite expects
1242 		 * buffers to reside on a queue, while BUF_WRITE and
1243 		 * brelse do not.
1244 		 */
1245 		if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1246 			(flags & V_SAVE)) {
1247 
1248 			if (bp->b_vp == vp) {
1249 				if (bp->b_flags & B_CLUSTEROK) {
1250 					vfs_bio_awrite(bp);
1251 				} else {
1252 					bremfree(bp);
1253 					bp->b_flags |= B_ASYNC;
1254 					BUF_WRITE(bp);
1255 				}
1256 			} else {
1257 				bremfree(bp);
1258 				(void) BUF_WRITE(bp);
1259 			}
1260 			goto done;
1261 		}
1262 		bremfree(bp);
1263 		bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
1264 		bp->b_flags &= ~B_ASYNC;
1265 		brelse(bp);
1266 		VI_LOCK(vp);
1267 	}
1268 	return (found);
1269 done:
1270 	VI_LOCK(vp);
1271 	return (found);
1272 }
1273 
1274 /*
1275  * Truncate a file's buffer and pages to a specified length.  This
1276  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1277  * sync activity.
1278  */
1279 int
1280 vtruncbuf(vp, cred, td, length, blksize)
1281 	register struct vnode *vp;
1282 	struct ucred *cred;
1283 	struct thread *td;
1284 	off_t length;
1285 	int blksize;
1286 {
1287 	register struct buf *bp;
1288 	struct buf *nbp;
1289 	int anyfreed;
1290 	int trunclbn;
1291 
1292 	/*
1293 	 * Round up to the *next* lbn.
1294 	 */
1295 	trunclbn = (length + blksize - 1) / blksize;
1296 
1297 	ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1298 restart:
1299 	VI_LOCK(vp);
1300 	anyfreed = 1;
1301 	for (;anyfreed;) {
1302 		anyfreed = 0;
1303 		for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
1304 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1305 			if (bp->b_lblkno >= trunclbn) {
1306 				if (BUF_LOCK(bp,
1307 				    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1308 				    VI_MTX(vp)) == ENOLCK)
1309 					goto restart;
1310 
1311 				bremfree(bp);
1312 				bp->b_flags |= (B_INVAL | B_RELBUF);
1313 				bp->b_flags &= ~B_ASYNC;
1314 				brelse(bp);
1315 				anyfreed = 1;
1316 
1317 				if (nbp &&
1318 				    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1319 				    (nbp->b_vp != vp) ||
1320 				    (nbp->b_flags & B_DELWRI))) {
1321 					goto restart;
1322 				}
1323 				VI_LOCK(vp);
1324 			}
1325 		}
1326 
1327 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1328 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1329 			if (bp->b_lblkno >= trunclbn) {
1330 				if (BUF_LOCK(bp,
1331 				    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1332 				    VI_MTX(vp)) == ENOLCK)
1333 					goto restart;
1334 				bremfree(bp);
1335 				bp->b_flags |= (B_INVAL | B_RELBUF);
1336 				bp->b_flags &= ~B_ASYNC;
1337 				brelse(bp);
1338 				anyfreed = 1;
1339 				if (nbp &&
1340 				    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1341 				    (nbp->b_vp != vp) ||
1342 				    (nbp->b_flags & B_DELWRI) == 0)) {
1343 					goto restart;
1344 				}
1345 				VI_LOCK(vp);
1346 			}
1347 		}
1348 	}
1349 
1350 	if (length > 0) {
1351 restartsync:
1352 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
1353 			nbp = TAILQ_NEXT(bp, b_vnbufs);
1354 			if (bp->b_lblkno > 0)
1355 				continue;
1356 			/*
1357 			 * Since we hold the vnode lock this should only
1358 			 * fail if we're racing with the buf daemon.
1359 			 */
1360 			if (BUF_LOCK(bp,
1361 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1362 			    VI_MTX(vp)) == ENOLCK) {
1363 				goto restart;
1364 			}
1365 			KASSERT((bp->b_flags & B_DELWRI),
1366 			    ("buf(%p) on dirty queue without DELWRI.", bp));
1367 
1368 			bremfree(bp);
1369 			bawrite(bp);
1370 			VI_LOCK(vp);
1371 			goto restartsync;
1372 		}
1373 	}
1374 
1375 	while (vp->v_numoutput > 0) {
1376 		vp->v_iflag |= VI_BWAIT;
1377 		msleep(&vp->v_numoutput, VI_MTX(vp), PVM, "vbtrunc", 0);
1378 	}
1379 	VI_UNLOCK(vp);
1380 	vnode_pager_setsize(vp, length);
1381 
1382 	return (0);
1383 }
1384 
1385 /*
1386  * buf_splay() - splay tree core for the clean/dirty list of buffers in
1387  * 		 a vnode.
1388  *
1389  *	NOTE: We have to deal with the special case of a background bitmap
1390  *	buffer, a situation where two buffers will have the same logical
1391  *	block offset.  We want (1) only the foreground buffer to be accessed
1392  *	in a lookup and (2) must differentiate between the foreground and
1393  *	background buffer in the splay tree algorithm because the splay
1394  *	tree cannot normally handle multiple entities with the same 'index'.
1395  *	We accomplish this by adding differentiating flags to the splay tree's
1396  *	numerical domain.
1397  */
1398 static
1399 struct buf *
1400 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1401 {
1402 	struct buf dummy;
1403 	struct buf *lefttreemax, *righttreemin, *y;
1404 
1405 	if (root == NULL)
1406 		return (NULL);
1407 	lefttreemax = righttreemin = &dummy;
1408 	for (;;) {
1409 		if (lblkno < root->b_lblkno ||
1410 		    (lblkno == root->b_lblkno &&
1411 		    (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1412 			if ((y = root->b_left) == NULL)
1413 				break;
1414 			if (lblkno < y->b_lblkno) {
1415 				/* Rotate right. */
1416 				root->b_left = y->b_right;
1417 				y->b_right = root;
1418 				root = y;
1419 				if ((y = root->b_left) == NULL)
1420 					break;
1421 			}
1422 			/* Link into the new root's right tree. */
1423 			righttreemin->b_left = root;
1424 			righttreemin = root;
1425 		} else if (lblkno > root->b_lblkno ||
1426 		    (lblkno == root->b_lblkno &&
1427 		    (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1428 			if ((y = root->b_right) == NULL)
1429 				break;
1430 			if (lblkno > y->b_lblkno) {
1431 				/* Rotate left. */
1432 				root->b_right = y->b_left;
1433 				y->b_left = root;
1434 				root = y;
1435 				if ((y = root->b_right) == NULL)
1436 					break;
1437 			}
1438 			/* Link into the new root's left tree. */
1439 			lefttreemax->b_right = root;
1440 			lefttreemax = root;
1441 		} else {
1442 			break;
1443 		}
1444 		root = y;
1445 	}
1446 	/* Assemble the new root. */
1447 	lefttreemax->b_right = root->b_left;
1448 	righttreemin->b_left = root->b_right;
1449 	root->b_left = dummy.b_right;
1450 	root->b_right = dummy.b_left;
1451 	return (root);
1452 }
1453 
1454 static
1455 void
1456 buf_vlist_remove(struct buf *bp)
1457 {
1458 	struct vnode *vp = bp->b_vp;
1459 	struct buf *root;
1460 
1461 	ASSERT_VI_LOCKED(vp, "buf_vlist_remove");
1462 	if (bp->b_xflags & BX_VNDIRTY) {
1463 		if (bp != vp->v_dirtyblkroot) {
1464 			root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_dirtyblkroot);
1465 			KASSERT(root == bp, ("splay lookup failed during dirty remove"));
1466 		}
1467 		if (bp->b_left == NULL) {
1468 			root = bp->b_right;
1469 		} else {
1470 			root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1471 			root->b_right = bp->b_right;
1472 		}
1473 		vp->v_dirtyblkroot = root;
1474 		TAILQ_REMOVE(&vp->v_dirtyblkhd, bp, b_vnbufs);
1475 		vp->v_dirtybufcnt--;
1476 	} else {
1477 		/* KASSERT(bp->b_xflags & BX_VNCLEAN, ("bp wasn't clean")); */
1478 		if (bp != vp->v_cleanblkroot) {
1479 			root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_cleanblkroot);
1480 			KASSERT(root == bp, ("splay lookup failed during clean remove"));
1481 		}
1482 		if (bp->b_left == NULL) {
1483 			root = bp->b_right;
1484 		} else {
1485 			root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1486 			root->b_right = bp->b_right;
1487 		}
1488 		vp->v_cleanblkroot = root;
1489 		TAILQ_REMOVE(&vp->v_cleanblkhd, bp, b_vnbufs);
1490 		vp->v_cleanbufcnt--;
1491 	}
1492 	bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1493 }
1494 
1495 /*
1496  * Add the buffer to the sorted clean or dirty block list using a
1497  * splay tree algorithm.
1498  *
1499  * NOTE: xflags is passed as a constant, optimizing this inline function!
1500  */
1501 static
1502 void
1503 buf_vlist_add(struct buf *bp, struct vnode *vp, b_xflags_t xflags)
1504 {
1505 	struct buf *root;
1506 
1507 	ASSERT_VI_LOCKED(vp, "buf_vlist_add");
1508 	bp->b_xflags |= xflags;
1509 	if (xflags & BX_VNDIRTY) {
1510 		root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_dirtyblkroot);
1511 		if (root == NULL) {
1512 			bp->b_left = NULL;
1513 			bp->b_right = NULL;
1514 			TAILQ_INSERT_TAIL(&vp->v_dirtyblkhd, bp, b_vnbufs);
1515 		} else if (bp->b_lblkno < root->b_lblkno ||
1516 		    (bp->b_lblkno == root->b_lblkno &&
1517 		    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1518 			bp->b_left = root->b_left;
1519 			bp->b_right = root;
1520 			root->b_left = NULL;
1521 			TAILQ_INSERT_BEFORE(root, bp, b_vnbufs);
1522 		} else {
1523 			bp->b_right = root->b_right;
1524 			bp->b_left = root;
1525 			root->b_right = NULL;
1526 			TAILQ_INSERT_AFTER(&vp->v_dirtyblkhd,
1527 			    root, bp, b_vnbufs);
1528 		}
1529 		vp->v_dirtybufcnt++;
1530 		vp->v_dirtyblkroot = bp;
1531 	} else {
1532 		/* KASSERT(xflags & BX_VNCLEAN, ("xflags not clean")); */
1533 		root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_cleanblkroot);
1534 		if (root == NULL) {
1535 			bp->b_left = NULL;
1536 			bp->b_right = NULL;
1537 			TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
1538 		} else if (bp->b_lblkno < root->b_lblkno ||
1539 		    (bp->b_lblkno == root->b_lblkno &&
1540 		    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1541 			bp->b_left = root->b_left;
1542 			bp->b_right = root;
1543 			root->b_left = NULL;
1544 			TAILQ_INSERT_BEFORE(root, bp, b_vnbufs);
1545 		} else {
1546 			bp->b_right = root->b_right;
1547 			bp->b_left = root;
1548 			root->b_right = NULL;
1549 			TAILQ_INSERT_AFTER(&vp->v_cleanblkhd,
1550 			    root, bp, b_vnbufs);
1551 		}
1552 		vp->v_cleanbufcnt++;
1553 		vp->v_cleanblkroot = bp;
1554 	}
1555 }
1556 
1557 /*
1558  * Lookup a buffer using the splay tree.  Note that we specifically avoid
1559  * shadow buffers used in background bitmap writes.
1560  *
1561  * This code isn't quite efficient as it could be because we are maintaining
1562  * two sorted lists and do not know which list the block resides in.
1563  *
1564  * During a "make buildworld" the desired buffer is found at one of
1565  * the roots more than 60% of the time.  Thus, checking both roots
1566  * before performing either splay eliminates unnecessary splays on the
1567  * first tree splayed.
1568  */
1569 struct buf *
1570 gbincore(struct vnode *vp, daddr_t lblkno)
1571 {
1572 	struct buf *bp;
1573 
1574 	GIANT_REQUIRED;
1575 
1576 	ASSERT_VI_LOCKED(vp, "gbincore");
1577 	if ((bp = vp->v_cleanblkroot) != NULL &&
1578 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1579 		return (bp);
1580 	if ((bp = vp->v_dirtyblkroot) != NULL &&
1581 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1582 		return (bp);
1583 	if ((bp = vp->v_cleanblkroot) != NULL) {
1584 		vp->v_cleanblkroot = bp = buf_splay(lblkno, 0, bp);
1585 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1586 			return (bp);
1587 	}
1588 	if ((bp = vp->v_dirtyblkroot) != NULL) {
1589 		vp->v_dirtyblkroot = bp = buf_splay(lblkno, 0, bp);
1590 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1591 			return (bp);
1592 	}
1593 	return (NULL);
1594 }
1595 
1596 /*
1597  * Associate a buffer with a vnode.
1598  */
1599 void
1600 bgetvp(vp, bp)
1601 	register struct vnode *vp;
1602 	register struct buf *bp;
1603 {
1604 	KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
1605 
1606 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1607 	    ("bgetvp: bp already attached! %p", bp));
1608 
1609 	ASSERT_VI_LOCKED(vp, "bgetvp");
1610 	vholdl(vp);
1611 	bp->b_vp = vp;
1612 	bp->b_dev = vn_todev(vp);
1613 	/*
1614 	 * Insert onto list for new vnode.
1615 	 */
1616 	buf_vlist_add(bp, vp, BX_VNCLEAN);
1617 }
1618 
1619 /*
1620  * Disassociate a buffer from a vnode.
1621  */
1622 void
1623 brelvp(bp)
1624 	register struct buf *bp;
1625 {
1626 	struct vnode *vp;
1627 
1628 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1629 
1630 	/*
1631 	 * Delete from old vnode list, if on one.
1632 	 */
1633 	vp = bp->b_vp;
1634 	VI_LOCK(vp);
1635 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1636 		buf_vlist_remove(bp);
1637 	if ((vp->v_iflag & VI_ONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
1638 		vp->v_iflag &= ~VI_ONWORKLST;
1639 		mtx_lock(&sync_mtx);
1640 		LIST_REMOVE(vp, v_synclist);
1641 		mtx_unlock(&sync_mtx);
1642 	}
1643 	vdropl(vp);
1644 	bp->b_vp = (struct vnode *) 0;
1645 	if (bp->b_object)
1646 		bp->b_object = NULL;
1647 	VI_UNLOCK(vp);
1648 }
1649 
1650 /*
1651  * Add an item to the syncer work queue.
1652  */
1653 static void
1654 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
1655 {
1656 	int slot;
1657 
1658 	ASSERT_VI_LOCKED(vp, "vn_syncer_add_to_worklist");
1659 
1660 	mtx_lock(&sync_mtx);
1661 	if (vp->v_iflag & VI_ONWORKLST)
1662 		LIST_REMOVE(vp, v_synclist);
1663 	else
1664 		vp->v_iflag |= VI_ONWORKLST;
1665 
1666 	if (delay > syncer_maxdelay - 2)
1667 		delay = syncer_maxdelay - 2;
1668 	slot = (syncer_delayno + delay) & syncer_mask;
1669 
1670 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
1671 	mtx_unlock(&sync_mtx);
1672 }
1673 
1674 struct  proc *updateproc;
1675 static void sched_sync(void);
1676 static struct kproc_desc up_kp = {
1677 	"syncer",
1678 	sched_sync,
1679 	&updateproc
1680 };
1681 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1682 
1683 /*
1684  * System filesystem synchronizer daemon.
1685  */
1686 static void
1687 sched_sync(void)
1688 {
1689 	struct synclist *slp;
1690 	struct vnode *vp;
1691 	struct mount *mp;
1692 	long starttime;
1693 	struct thread *td = FIRST_THREAD_IN_PROC(updateproc);  /* XXXKSE */
1694 
1695 	mtx_lock(&Giant);
1696 
1697 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, td->td_proc,
1698 	    SHUTDOWN_PRI_LAST);
1699 
1700 	for (;;) {
1701 		kthread_suspend_check(td->td_proc);
1702 
1703 		starttime = time_second;
1704 
1705 		/*
1706 		 * Push files whose dirty time has expired.  Be careful
1707 		 * of interrupt race on slp queue.
1708 		 */
1709 		mtx_lock(&sync_mtx);
1710 		slp = &syncer_workitem_pending[syncer_delayno];
1711 		syncer_delayno += 1;
1712 		if (syncer_delayno == syncer_maxdelay)
1713 			syncer_delayno = 0;
1714 
1715 		while ((vp = LIST_FIRST(slp)) != NULL) {
1716 			mtx_unlock(&sync_mtx);
1717 			if (VOP_ISLOCKED(vp, NULL) == 0 &&
1718 			    vn_start_write(vp, &mp, V_NOWAIT) == 0) {
1719 				vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1720 				(void) VOP_FSYNC(vp, td->td_ucred, MNT_LAZY, td);
1721 				VOP_UNLOCK(vp, 0, td);
1722 				vn_finished_write(mp);
1723 			}
1724 			mtx_lock(&sync_mtx);
1725 			if (LIST_FIRST(slp) == vp) {
1726 				mtx_unlock(&sync_mtx);
1727 				/*
1728 				 * Note: VFS vnodes can remain on the
1729 				 * worklist too with no dirty blocks, but
1730 				 * since sync_fsync() moves it to a different
1731 				 * slot we are safe.
1732 				 */
1733 				VI_LOCK(vp);
1734 				if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1735 				    !vn_isdisk(vp, NULL)) {
1736 					panic("sched_sync: fsync failed "
1737 					      "vp %p tag %s", vp, vp->v_tag);
1738 				}
1739 				/*
1740 				 * Put us back on the worklist.  The worklist
1741 				 * routine will remove us from our current
1742 				 * position and then add us back in at a later
1743 				 * position.
1744 				 */
1745 				vn_syncer_add_to_worklist(vp, syncdelay);
1746 				VI_UNLOCK(vp);
1747 				mtx_lock(&sync_mtx);
1748 			}
1749 		}
1750 		mtx_unlock(&sync_mtx);
1751 
1752 		/*
1753 		 * Do soft update processing.
1754 		 */
1755 		if (softdep_process_worklist_hook != NULL)
1756 			(*softdep_process_worklist_hook)(NULL);
1757 
1758 		/*
1759 		 * The variable rushjob allows the kernel to speed up the
1760 		 * processing of the filesystem syncer process. A rushjob
1761 		 * value of N tells the filesystem syncer to process the next
1762 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1763 		 * is used by the soft update code to speed up the filesystem
1764 		 * syncer process when the incore state is getting so far
1765 		 * ahead of the disk that the kernel memory pool is being
1766 		 * threatened with exhaustion.
1767 		 */
1768 		mtx_lock(&sync_mtx);
1769 		if (rushjob > 0) {
1770 			rushjob -= 1;
1771 			mtx_unlock(&sync_mtx);
1772 			continue;
1773 		}
1774 		mtx_unlock(&sync_mtx);
1775 		/*
1776 		 * If it has taken us less than a second to process the
1777 		 * current work, then wait. Otherwise start right over
1778 		 * again. We can still lose time if any single round
1779 		 * takes more than two seconds, but it does not really
1780 		 * matter as we are just trying to generally pace the
1781 		 * filesystem activity.
1782 		 */
1783 		if (time_second == starttime)
1784 			tsleep(&lbolt, PPAUSE, "syncer", 0);
1785 	}
1786 }
1787 
1788 /*
1789  * Request the syncer daemon to speed up its work.
1790  * We never push it to speed up more than half of its
1791  * normal turn time, otherwise it could take over the cpu.
1792  * XXXKSE  only one update?
1793  */
1794 int
1795 speedup_syncer()
1796 {
1797 	struct thread *td;
1798 	int ret = 0;
1799 
1800 	td = FIRST_THREAD_IN_PROC(updateproc);
1801 	mtx_lock_spin(&sched_lock);
1802 	if (td->td_wchan == &lbolt) {
1803 		unsleep(td);
1804 		TD_CLR_SLEEPING(td);
1805 		setrunnable(td);
1806 	}
1807 	mtx_unlock_spin(&sched_lock);
1808 	mtx_lock(&sync_mtx);
1809 	if (rushjob < syncdelay / 2) {
1810 		rushjob += 1;
1811 		stat_rush_requests += 1;
1812 		ret = 1;
1813 	}
1814 	mtx_unlock(&sync_mtx);
1815 	return (ret);
1816 }
1817 
1818 /*
1819  * Associate a p-buffer with a vnode.
1820  *
1821  * Also sets B_PAGING flag to indicate that vnode is not fully associated
1822  * with the buffer.  i.e. the bp has not been linked into the vnode or
1823  * ref-counted.
1824  */
1825 void
1826 pbgetvp(vp, bp)
1827 	register struct vnode *vp;
1828 	register struct buf *bp;
1829 {
1830 
1831 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1832 
1833 	bp->b_vp = vp;
1834 	bp->b_flags |= B_PAGING;
1835 	bp->b_dev = vn_todev(vp);
1836 }
1837 
1838 /*
1839  * Disassociate a p-buffer from a vnode.
1840  */
1841 void
1842 pbrelvp(bp)
1843 	register struct buf *bp;
1844 {
1845 
1846 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1847 
1848 	/* XXX REMOVE ME */
1849 	VI_LOCK(bp->b_vp);
1850 	if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1851 		panic(
1852 		    "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1853 		    bp,
1854 		    (int)bp->b_flags
1855 		);
1856 	}
1857 	VI_UNLOCK(bp->b_vp);
1858 	bp->b_vp = (struct vnode *) 0;
1859 	bp->b_flags &= ~B_PAGING;
1860 }
1861 
1862 /*
1863  * Reassign a buffer from one vnode to another.
1864  * Used to assign file specific control information
1865  * (indirect blocks) to the vnode to which they belong.
1866  */
1867 void
1868 reassignbuf(bp, newvp)
1869 	register struct buf *bp;
1870 	register struct vnode *newvp;
1871 {
1872 	struct vnode *vp;
1873 	int delay;
1874 
1875 	if (newvp == NULL) {
1876 		printf("reassignbuf: NULL");
1877 		return;
1878 	}
1879 	vp = bp->b_vp;
1880 	++reassignbufcalls;
1881 
1882 	/*
1883 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1884 	 * is not fully linked in.
1885 	 */
1886 	if (bp->b_flags & B_PAGING)
1887 		panic("cannot reassign paging buffer");
1888 
1889 	/*
1890 	 * Delete from old vnode list, if on one.
1891 	 */
1892 	VI_LOCK(vp);
1893 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1894 		buf_vlist_remove(bp);
1895 		if (vp != newvp) {
1896 			vdropl(bp->b_vp);
1897 			bp->b_vp = NULL;	/* for clarification */
1898 		}
1899 	}
1900 	if (vp != newvp) {
1901 		VI_UNLOCK(vp);
1902 		VI_LOCK(newvp);
1903 	}
1904 	/*
1905 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1906 	 * of clean buffers.
1907 	 */
1908 	if (bp->b_flags & B_DELWRI) {
1909 		if ((newvp->v_iflag & VI_ONWORKLST) == 0) {
1910 			switch (newvp->v_type) {
1911 			case VDIR:
1912 				delay = dirdelay;
1913 				break;
1914 			case VCHR:
1915 				if (newvp->v_rdev->si_mountpoint != NULL) {
1916 					delay = metadelay;
1917 					break;
1918 				}
1919 				/* FALLTHROUGH */
1920 			default:
1921 				delay = filedelay;
1922 			}
1923 			vn_syncer_add_to_worklist(newvp, delay);
1924 		}
1925 		buf_vlist_add(bp, newvp, BX_VNDIRTY);
1926 	} else {
1927 		buf_vlist_add(bp, newvp, BX_VNCLEAN);
1928 
1929 		if ((newvp->v_iflag & VI_ONWORKLST) &&
1930 		    TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1931 			mtx_lock(&sync_mtx);
1932 			LIST_REMOVE(newvp, v_synclist);
1933 			mtx_unlock(&sync_mtx);
1934 			newvp->v_iflag &= ~VI_ONWORKLST;
1935 		}
1936 	}
1937 	if (bp->b_vp != newvp) {
1938 		bp->b_vp = newvp;
1939 		vholdl(bp->b_vp);
1940 	}
1941 	VI_UNLOCK(newvp);
1942 }
1943 
1944 /*
1945  * Create a vnode for a device.
1946  * Used for mounting the root filesystem.
1947  */
1948 int
1949 bdevvp(dev, vpp)
1950 	dev_t dev;
1951 	struct vnode **vpp;
1952 {
1953 	register struct vnode *vp;
1954 	struct vnode *nvp;
1955 	int error;
1956 
1957 	if (dev == NODEV) {
1958 		*vpp = NULLVP;
1959 		return (ENXIO);
1960 	}
1961 	if (vfinddev(dev, VCHR, vpp))
1962 		return (0);
1963 	error = getnewvnode("none", (struct mount *)0, spec_vnodeop_p, &nvp);
1964 	if (error) {
1965 		*vpp = NULLVP;
1966 		return (error);
1967 	}
1968 	vp = nvp;
1969 	vp->v_type = VCHR;
1970 	addalias(vp, dev);
1971 	*vpp = vp;
1972 	return (0);
1973 }
1974 
1975 static void
1976 v_incr_usecount(struct vnode *vp, int delta)
1977 {
1978 	vp->v_usecount += delta;
1979 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1980 		mtx_lock(&spechash_mtx);
1981 		vp->v_rdev->si_usecount += delta;
1982 		mtx_unlock(&spechash_mtx);
1983 	}
1984 }
1985 
1986 /*
1987  * Add vnode to the alias list hung off the dev_t.
1988  *
1989  * The reason for this gunk is that multiple vnodes can reference
1990  * the same physical device, so checking vp->v_usecount to see
1991  * how many users there are is inadequate; the v_usecount for
1992  * the vnodes need to be accumulated.  vcount() does that.
1993  */
1994 struct vnode *
1995 addaliasu(nvp, nvp_rdev)
1996 	struct vnode *nvp;
1997 	udev_t nvp_rdev;
1998 {
1999 	struct vnode *ovp;
2000 	vop_t **ops;
2001 	dev_t dev;
2002 
2003 	if (nvp->v_type == VBLK)
2004 		return (nvp);
2005 	if (nvp->v_type != VCHR)
2006 		panic("addaliasu on non-special vnode");
2007 	dev = udev2dev(nvp_rdev, 0);
2008 	/*
2009 	 * Check to see if we have a bdevvp vnode with no associated
2010 	 * filesystem. If so, we want to associate the filesystem of
2011 	 * the new newly instigated vnode with the bdevvp vnode and
2012 	 * discard the newly created vnode rather than leaving the
2013 	 * bdevvp vnode lying around with no associated filesystem.
2014 	 */
2015 	if (vfinddev(dev, nvp->v_type, &ovp) == 0 || ovp->v_data != NULL) {
2016 		addalias(nvp, dev);
2017 		return (nvp);
2018 	}
2019 	/*
2020 	 * Discard unneeded vnode, but save its node specific data.
2021 	 * Note that if there is a lock, it is carried over in the
2022 	 * node specific data to the replacement vnode.
2023 	 */
2024 	vref(ovp);
2025 	ovp->v_data = nvp->v_data;
2026 	ovp->v_tag = nvp->v_tag;
2027 	nvp->v_data = NULL;
2028 	lockdestroy(ovp->v_vnlock);
2029 	lockinit(ovp->v_vnlock, PVFS, nvp->v_vnlock->lk_wmesg,
2030 	    nvp->v_vnlock->lk_timo, nvp->v_vnlock->lk_flags & LK_EXTFLG_MASK);
2031 	ops = ovp->v_op;
2032 	ovp->v_op = nvp->v_op;
2033 	if (VOP_ISLOCKED(nvp, curthread)) {
2034 		VOP_UNLOCK(nvp, 0, curthread);
2035 		vn_lock(ovp, LK_EXCLUSIVE | LK_RETRY, curthread);
2036 	}
2037 	nvp->v_op = ops;
2038 	insmntque(ovp, nvp->v_mount);
2039 	vrele(nvp);
2040 	vgone(nvp);
2041 	return (ovp);
2042 }
2043 
2044 /* This is a local helper function that do the same as addaliasu, but for a
2045  * dev_t instead of an udev_t. */
2046 static void
2047 addalias(nvp, dev)
2048 	struct vnode *nvp;
2049 	dev_t dev;
2050 {
2051 
2052 	KASSERT(nvp->v_type == VCHR, ("addalias on non-special vnode"));
2053 	nvp->v_rdev = dev;
2054 	VI_LOCK(nvp);
2055 	mtx_lock(&spechash_mtx);
2056 	SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
2057 	dev->si_usecount += nvp->v_usecount;
2058 	mtx_unlock(&spechash_mtx);
2059 	VI_UNLOCK(nvp);
2060 }
2061 
2062 /*
2063  * Grab a particular vnode from the free list, increment its
2064  * reference count and lock it. The vnode lock bit is set if the
2065  * vnode is being eliminated in vgone. The process is awakened
2066  * when the transition is completed, and an error returned to
2067  * indicate that the vnode is no longer usable (possibly having
2068  * been changed to a new filesystem type).
2069  */
2070 int
2071 vget(vp, flags, td)
2072 	register struct vnode *vp;
2073 	int flags;
2074 	struct thread *td;
2075 {
2076 	int error;
2077 
2078 	/*
2079 	 * If the vnode is in the process of being cleaned out for
2080 	 * another use, we wait for the cleaning to finish and then
2081 	 * return failure. Cleaning is determined by checking that
2082 	 * the VI_XLOCK flag is set.
2083 	 */
2084 	if ((flags & LK_INTERLOCK) == 0)
2085 		VI_LOCK(vp);
2086 	if (vp->v_iflag & VI_XLOCK && vp->v_vxproc != curthread) {
2087 		vp->v_iflag |= VI_XWANT;
2088 		msleep(vp, VI_MTX(vp), PINOD | PDROP, "vget", 0);
2089 		return (ENOENT);
2090 	}
2091 
2092 	v_incr_usecount(vp, 1);
2093 
2094 	if (VSHOULDBUSY(vp))
2095 		vbusy(vp);
2096 	if (flags & LK_TYPE_MASK) {
2097 		if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
2098 			/*
2099 			 * must expand vrele here because we do not want
2100 			 * to call VOP_INACTIVE if the reference count
2101 			 * drops back to zero since it was never really
2102 			 * active. We must remove it from the free list
2103 			 * before sleeping so that multiple processes do
2104 			 * not try to recycle it.
2105 			 */
2106 			VI_LOCK(vp);
2107 			v_incr_usecount(vp, -1);
2108 			if (VSHOULDFREE(vp))
2109 				vfree(vp);
2110 			else
2111 				vlruvp(vp);
2112 			VI_UNLOCK(vp);
2113 		}
2114 		return (error);
2115 	}
2116 	VI_UNLOCK(vp);
2117 	return (0);
2118 }
2119 
2120 /*
2121  * Increase the reference count of a vnode.
2122  */
2123 void
2124 vref(struct vnode *vp)
2125 {
2126 	VI_LOCK(vp);
2127 	v_incr_usecount(vp, 1);
2128 	VI_UNLOCK(vp);
2129 }
2130 
2131 /*
2132  * Return reference count of a vnode.
2133  *
2134  * The results of this call are only guaranteed when some mechanism other
2135  * than the VI lock is used to stop other processes from gaining references
2136  * to the vnode.  This may be the case if the caller holds the only reference.
2137  * This is also useful when stale data is acceptable as race conditions may
2138  * be accounted for by some other means.
2139  */
2140 int
2141 vrefcnt(struct vnode *vp)
2142 {
2143 	int usecnt;
2144 
2145 	VI_LOCK(vp);
2146 	usecnt = vp->v_usecount;
2147 	VI_UNLOCK(vp);
2148 
2149 	return (usecnt);
2150 }
2151 
2152 
2153 /*
2154  * Vnode put/release.
2155  * If count drops to zero, call inactive routine and return to freelist.
2156  */
2157 void
2158 vrele(vp)
2159 	struct vnode *vp;
2160 {
2161 	struct thread *td = curthread;	/* XXX */
2162 
2163 	KASSERT(vp != NULL, ("vrele: null vp"));
2164 
2165 	VI_LOCK(vp);
2166 
2167 	/* Skip this v_writecount check if we're going to panic below. */
2168 	KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
2169 	    ("vrele: missed vn_close"));
2170 
2171 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2172 	    vp->v_usecount == 1)) {
2173 		v_incr_usecount(vp, -1);
2174 		VI_UNLOCK(vp);
2175 
2176 		return;
2177 	}
2178 
2179 	if (vp->v_usecount == 1) {
2180 		v_incr_usecount(vp, -1);
2181 		/*
2182 		 * We must call VOP_INACTIVE with the node locked. Mark
2183 		 * as VI_DOINGINACT to avoid recursion.
2184 		 */
2185 		if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0) {
2186 			VI_LOCK(vp);
2187 			vp->v_iflag |= VI_DOINGINACT;
2188 			VI_UNLOCK(vp);
2189 			VOP_INACTIVE(vp, td);
2190 			VI_LOCK(vp);
2191 			KASSERT(vp->v_iflag & VI_DOINGINACT,
2192 			    ("vrele: lost VI_DOINGINACT"));
2193 			vp->v_iflag &= ~VI_DOINGINACT;
2194 		} else
2195 			VI_LOCK(vp);
2196 		if (VSHOULDFREE(vp))
2197 			vfree(vp);
2198 		else
2199 			vlruvp(vp);
2200 		VI_UNLOCK(vp);
2201 
2202 	} else {
2203 #ifdef DIAGNOSTIC
2204 		vprint("vrele: negative ref count", vp);
2205 #endif
2206 		VI_UNLOCK(vp);
2207 		panic("vrele: negative ref cnt");
2208 	}
2209 }
2210 
2211 /*
2212  * Release an already locked vnode.  This give the same effects as
2213  * unlock+vrele(), but takes less time and avoids releasing and
2214  * re-aquiring the lock (as vrele() aquires the lock internally.)
2215  */
2216 void
2217 vput(vp)
2218 	struct vnode *vp;
2219 {
2220 	struct thread *td = curthread;	/* XXX */
2221 
2222 	GIANT_REQUIRED;
2223 
2224 	KASSERT(vp != NULL, ("vput: null vp"));
2225 	VI_LOCK(vp);
2226 	/* Skip this v_writecount check if we're going to panic below. */
2227 	KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
2228 	    ("vput: missed vn_close"));
2229 
2230 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2231 	    vp->v_usecount == 1)) {
2232 		v_incr_usecount(vp, -1);
2233 		VOP_UNLOCK(vp, LK_INTERLOCK, td);
2234 		return;
2235 	}
2236 
2237 	if (vp->v_usecount == 1) {
2238 		v_incr_usecount(vp, -1);
2239 		/*
2240 		 * We must call VOP_INACTIVE with the node locked, so
2241 		 * we just need to release the vnode mutex. Mark as
2242 		 * as VI_DOINGINACT to avoid recursion.
2243 		 */
2244 		vp->v_iflag |= VI_DOINGINACT;
2245 		VI_UNLOCK(vp);
2246 		VOP_INACTIVE(vp, td);
2247 		VI_LOCK(vp);
2248 		KASSERT(vp->v_iflag & VI_DOINGINACT,
2249 		    ("vput: lost VI_DOINGINACT"));
2250 		vp->v_iflag &= ~VI_DOINGINACT;
2251 		if (VSHOULDFREE(vp))
2252 			vfree(vp);
2253 		else
2254 			vlruvp(vp);
2255 		VI_UNLOCK(vp);
2256 
2257 	} else {
2258 #ifdef DIAGNOSTIC
2259 		vprint("vput: negative ref count", vp);
2260 #endif
2261 		panic("vput: negative ref cnt");
2262 	}
2263 }
2264 
2265 /*
2266  * Somebody doesn't want the vnode recycled.
2267  */
2268 void
2269 vhold(struct vnode *vp)
2270 {
2271 	VI_LOCK(vp);
2272 	vholdl(vp);
2273 	VI_UNLOCK(vp);
2274 }
2275 
2276 void
2277 vholdl(vp)
2278 	register struct vnode *vp;
2279 {
2280 	vp->v_holdcnt++;
2281 	if (VSHOULDBUSY(vp))
2282 		vbusy(vp);
2283 }
2284 
2285 /*
2286  * Note that there is one less who cares about this vnode.  vdrop() is the
2287  * opposite of vhold().
2288  */
2289 void
2290 vdrop(struct vnode *vp)
2291 {
2292 	VI_LOCK(vp);
2293 	vdropl(vp);
2294 	VI_UNLOCK(vp);
2295 }
2296 
2297 void
2298 vdropl(vp)
2299 	register struct vnode *vp;
2300 {
2301 	if (vp->v_holdcnt <= 0)
2302 		panic("vdrop: holdcnt");
2303 	vp->v_holdcnt--;
2304 	if (VSHOULDFREE(vp))
2305 		vfree(vp);
2306 	else
2307 		vlruvp(vp);
2308 }
2309 
2310 /*
2311  * Remove any vnodes in the vnode table belonging to mount point mp.
2312  *
2313  * If FORCECLOSE is not specified, there should not be any active ones,
2314  * return error if any are found (nb: this is a user error, not a
2315  * system error). If FORCECLOSE is specified, detach any active vnodes
2316  * that are found.
2317  *
2318  * If WRITECLOSE is set, only flush out regular file vnodes open for
2319  * writing.
2320  *
2321  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2322  *
2323  * `rootrefs' specifies the base reference count for the root vnode
2324  * of this filesystem. The root vnode is considered busy if its
2325  * v_usecount exceeds this value. On a successful return, vflush()
2326  * will call vrele() on the root vnode exactly rootrefs times.
2327  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2328  * be zero.
2329  */
2330 #ifdef DIAGNOSTIC
2331 static int busyprt = 0;		/* print out busy vnodes */
2332 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
2333 #endif
2334 
2335 int
2336 vflush(mp, rootrefs, flags)
2337 	struct mount *mp;
2338 	int rootrefs;
2339 	int flags;
2340 {
2341 	struct thread *td = curthread;	/* XXX */
2342 	struct vnode *vp, *nvp, *rootvp = NULL;
2343 	struct vattr vattr;
2344 	int busy = 0, error;
2345 
2346 	if (rootrefs > 0) {
2347 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2348 		    ("vflush: bad args"));
2349 		/*
2350 		 * Get the filesystem root vnode. We can vput() it
2351 		 * immediately, since with rootrefs > 0, it won't go away.
2352 		 */
2353 		if ((error = VFS_ROOT(mp, &rootvp)) != 0)
2354 			return (error);
2355 		vput(rootvp);
2356 
2357 	}
2358 	mtx_lock(&mntvnode_mtx);
2359 loop:
2360 	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) {
2361 		/*
2362 		 * Make sure this vnode wasn't reclaimed in getnewvnode().
2363 		 * Start over if it has (it won't be on the list anymore).
2364 		 */
2365 		if (vp->v_mount != mp)
2366 			goto loop;
2367 		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
2368 
2369 		VI_LOCK(vp);
2370 		mtx_unlock(&mntvnode_mtx);
2371 		vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY, td);
2372 		/*
2373 		 * This vnode could have been reclaimed while we were
2374 		 * waiting for the lock since we are not holding a
2375 		 * reference.
2376 		 * Start over if the vnode was reclaimed.
2377 		 */
2378 		if (vp->v_mount != mp) {
2379 			VOP_UNLOCK(vp, 0, td);
2380 			mtx_lock(&mntvnode_mtx);
2381 			goto loop;
2382 		}
2383 		/*
2384 		 * Skip over a vnodes marked VV_SYSTEM.
2385 		 */
2386 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2387 			VOP_UNLOCK(vp, 0, td);
2388 			mtx_lock(&mntvnode_mtx);
2389 			continue;
2390 		}
2391 		/*
2392 		 * If WRITECLOSE is set, flush out unlinked but still open
2393 		 * files (even if open only for reading) and regular file
2394 		 * vnodes open for writing.
2395 		 */
2396 		if (flags & WRITECLOSE) {
2397 			error = VOP_GETATTR(vp, &vattr, td->td_ucred, td);
2398 			VI_LOCK(vp);
2399 
2400 			if ((vp->v_type == VNON ||
2401 			    (error == 0 && vattr.va_nlink > 0)) &&
2402 			    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2403 				VOP_UNLOCK(vp, LK_INTERLOCK, td);
2404 				mtx_lock(&mntvnode_mtx);
2405 				continue;
2406 			}
2407 		} else
2408 			VI_LOCK(vp);
2409 
2410 		VOP_UNLOCK(vp, 0, td);
2411 
2412 		/*
2413 		 * With v_usecount == 0, all we need to do is clear out the
2414 		 * vnode data structures and we are done.
2415 		 */
2416 		if (vp->v_usecount == 0) {
2417 			vgonel(vp, td);
2418 			mtx_lock(&mntvnode_mtx);
2419 			continue;
2420 		}
2421 
2422 		/*
2423 		 * If FORCECLOSE is set, forcibly close the vnode. For block
2424 		 * or character devices, revert to an anonymous device. For
2425 		 * all other files, just kill them.
2426 		 */
2427 		if (flags & FORCECLOSE) {
2428 			if (vp->v_type != VCHR) {
2429 				vgonel(vp, td);
2430 			} else {
2431 				vclean(vp, 0, td);
2432 				VI_UNLOCK(vp);
2433 				vp->v_op = spec_vnodeop_p;
2434 				insmntque(vp, (struct mount *) 0);
2435 			}
2436 			mtx_lock(&mntvnode_mtx);
2437 			continue;
2438 		}
2439 #ifdef DIAGNOSTIC
2440 		if (busyprt)
2441 			vprint("vflush: busy vnode", vp);
2442 #endif
2443 		VI_UNLOCK(vp);
2444 		mtx_lock(&mntvnode_mtx);
2445 		busy++;
2446 	}
2447 	mtx_unlock(&mntvnode_mtx);
2448 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2449 		/*
2450 		 * If just the root vnode is busy, and if its refcount
2451 		 * is equal to `rootrefs', then go ahead and kill it.
2452 		 */
2453 		VI_LOCK(rootvp);
2454 		KASSERT(busy > 0, ("vflush: not busy"));
2455 		KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
2456 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2457 			vgonel(rootvp, td);
2458 			busy = 0;
2459 		} else
2460 			VI_UNLOCK(rootvp);
2461 	}
2462 	if (busy)
2463 		return (EBUSY);
2464 	for (; rootrefs > 0; rootrefs--)
2465 		vrele(rootvp);
2466 	return (0);
2467 }
2468 
2469 /*
2470  * This moves a now (likely recyclable) vnode to the end of the
2471  * mountlist.  XXX However, it is temporarily disabled until we
2472  * can clean up ffs_sync() and friends, which have loop restart
2473  * conditions which this code causes to operate O(N^2).
2474  */
2475 static void
2476 vlruvp(struct vnode *vp)
2477 {
2478 #if 0
2479 	struct mount *mp;
2480 
2481 	if ((mp = vp->v_mount) != NULL) {
2482 		mtx_lock(&mntvnode_mtx);
2483 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
2484 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
2485 		mtx_unlock(&mntvnode_mtx);
2486 	}
2487 #endif
2488 }
2489 
2490 /*
2491  * Disassociate the underlying filesystem from a vnode.
2492  */
2493 static void
2494 vclean(vp, flags, td)
2495 	struct vnode *vp;
2496 	int flags;
2497 	struct thread *td;
2498 {
2499 	int active;
2500 
2501 	ASSERT_VI_LOCKED(vp, "vclean");
2502 	/*
2503 	 * Check to see if the vnode is in use. If so we have to reference it
2504 	 * before we clean it out so that its count cannot fall to zero and
2505 	 * generate a race against ourselves to recycle it.
2506 	 */
2507 	if ((active = vp->v_usecount))
2508 		v_incr_usecount(vp, 1);
2509 
2510 	/*
2511 	 * Prevent the vnode from being recycled or brought into use while we
2512 	 * clean it out.
2513 	 */
2514 	if (vp->v_iflag & VI_XLOCK)
2515 		panic("vclean: deadlock");
2516 	vp->v_iflag |= VI_XLOCK;
2517 	vp->v_vxproc = curthread;
2518 	/*
2519 	 * Even if the count is zero, the VOP_INACTIVE routine may still
2520 	 * have the object locked while it cleans it out. The VOP_LOCK
2521 	 * ensures that the VOP_INACTIVE routine is done with its work.
2522 	 * For active vnodes, it ensures that no other activity can
2523 	 * occur while the underlying object is being cleaned out.
2524 	 */
2525 	VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td);
2526 
2527 	/*
2528 	 * Clean out any buffers associated with the vnode.
2529 	 * If the flush fails, just toss the buffers.
2530 	 */
2531 	if (flags & DOCLOSE) {
2532 		struct buf *bp;
2533 		VI_LOCK(vp);
2534 		bp = TAILQ_FIRST(&vp->v_dirtyblkhd);
2535 		VI_UNLOCK(vp);
2536 		if (bp != NULL)
2537 			(void) vn_write_suspend_wait(vp, NULL, V_WAIT);
2538 		if (vinvalbuf(vp, V_SAVE, NOCRED, td, 0, 0) != 0)
2539 			vinvalbuf(vp, 0, NOCRED, td, 0, 0);
2540 	}
2541 
2542 	VOP_DESTROYVOBJECT(vp);
2543 
2544 	/*
2545 	 * Any other processes trying to obtain this lock must first
2546 	 * wait for VXLOCK to clear, then call the new lock operation.
2547 	 */
2548 	VOP_UNLOCK(vp, 0, td);
2549 
2550 	/*
2551 	 * If purging an active vnode, it must be closed and
2552 	 * deactivated before being reclaimed. Note that the
2553 	 * VOP_INACTIVE will unlock the vnode.
2554 	 */
2555 	if (active) {
2556 		if (flags & DOCLOSE)
2557 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2558 		VI_LOCK(vp);
2559 		if ((vp->v_iflag & VI_DOINGINACT) == 0) {
2560 			vp->v_iflag |= VI_DOINGINACT;
2561 			VI_UNLOCK(vp);
2562 			if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT, td) != 0)
2563 				panic("vclean: cannot relock.");
2564 			VOP_INACTIVE(vp, td);
2565 			VI_LOCK(vp);
2566 			KASSERT(vp->v_iflag & VI_DOINGINACT,
2567 			    ("vclean: lost VI_DOINGINACT"));
2568 			vp->v_iflag &= ~VI_DOINGINACT;
2569 		}
2570 		VI_UNLOCK(vp);
2571 	}
2572 
2573 	/*
2574 	 * Reclaim the vnode.
2575 	 */
2576 	if (VOP_RECLAIM(vp, td))
2577 		panic("vclean: cannot reclaim");
2578 
2579 	if (active) {
2580 		/*
2581 		 * Inline copy of vrele() since VOP_INACTIVE
2582 		 * has already been called.
2583 		 */
2584 		VI_LOCK(vp);
2585 		v_incr_usecount(vp, -1);
2586 		if (vp->v_usecount <= 0) {
2587 #ifdef INVARIANTS
2588 			if (vp->v_usecount < 0 || vp->v_writecount != 0) {
2589 				vprint("vclean: bad ref count", vp);
2590 				panic("vclean: ref cnt");
2591 			}
2592 #endif
2593 			if (VSHOULDFREE(vp))
2594 				vfree(vp);
2595 		}
2596 		VI_UNLOCK(vp);
2597 	}
2598 
2599 	cache_purge(vp);
2600 	VI_LOCK(vp);
2601 	if (VSHOULDFREE(vp))
2602 		vfree(vp);
2603 
2604 	/*
2605 	 * Done with purge, reset to the standard lock and
2606 	 * notify sleepers of the grim news.
2607 	 */
2608 	vp->v_vnlock = &vp->v_lock;
2609 	vp->v_op = dead_vnodeop_p;
2610 	if (vp->v_pollinfo != NULL)
2611 		vn_pollgone(vp);
2612 	vp->v_tag = "none";
2613 	vp->v_iflag &= ~VI_XLOCK;
2614 	vp->v_vxproc = NULL;
2615 	if (vp->v_iflag & VI_XWANT) {
2616 		vp->v_iflag &= ~VI_XWANT;
2617 		wakeup(vp);
2618 	}
2619 }
2620 
2621 /*
2622  * Eliminate all activity associated with the requested vnode
2623  * and with all vnodes aliased to the requested vnode.
2624  */
2625 int
2626 vop_revoke(ap)
2627 	struct vop_revoke_args /* {
2628 		struct vnode *a_vp;
2629 		int a_flags;
2630 	} */ *ap;
2631 {
2632 	struct vnode *vp, *vq;
2633 	dev_t dev;
2634 
2635 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
2636 	vp = ap->a_vp;
2637 	KASSERT((vp->v_type == VCHR), ("vop_revoke: not VCHR"));
2638 
2639 	VI_LOCK(vp);
2640 	/*
2641 	 * If a vgone (or vclean) is already in progress,
2642 	 * wait until it is done and return.
2643 	 */
2644 	if (vp->v_iflag & VI_XLOCK) {
2645 		vp->v_iflag |= VI_XWANT;
2646 		msleep(vp, VI_MTX(vp), PINOD | PDROP,
2647 		    "vop_revokeall", 0);
2648 		return (0);
2649 	}
2650 	VI_UNLOCK(vp);
2651 	dev = vp->v_rdev;
2652 	for (;;) {
2653 		mtx_lock(&spechash_mtx);
2654 		vq = SLIST_FIRST(&dev->si_hlist);
2655 		mtx_unlock(&spechash_mtx);
2656 		if (!vq)
2657 			break;
2658 		vgone(vq);
2659 	}
2660 	return (0);
2661 }
2662 
2663 /*
2664  * Recycle an unused vnode to the front of the free list.
2665  * Release the passed interlock if the vnode will be recycled.
2666  */
2667 int
2668 vrecycle(vp, inter_lkp, td)
2669 	struct vnode *vp;
2670 	struct mtx *inter_lkp;
2671 	struct thread *td;
2672 {
2673 
2674 	VI_LOCK(vp);
2675 	if (vp->v_usecount == 0) {
2676 		if (inter_lkp) {
2677 			mtx_unlock(inter_lkp);
2678 		}
2679 		vgonel(vp, td);
2680 		return (1);
2681 	}
2682 	VI_UNLOCK(vp);
2683 	return (0);
2684 }
2685 
2686 /*
2687  * Eliminate all activity associated with a vnode
2688  * in preparation for reuse.
2689  */
2690 void
2691 vgone(vp)
2692 	register struct vnode *vp;
2693 {
2694 	struct thread *td = curthread;	/* XXX */
2695 
2696 	VI_LOCK(vp);
2697 	vgonel(vp, td);
2698 }
2699 
2700 /*
2701  * vgone, with the vp interlock held.
2702  */
2703 void
2704 vgonel(vp, td)
2705 	struct vnode *vp;
2706 	struct thread *td;
2707 {
2708 	/*
2709 	 * If a vgone (or vclean) is already in progress,
2710 	 * wait until it is done and return.
2711 	 */
2712 	ASSERT_VI_LOCKED(vp, "vgonel");
2713 	if (vp->v_iflag & VI_XLOCK) {
2714 		vp->v_iflag |= VI_XWANT;
2715 		msleep(vp, VI_MTX(vp), PINOD | PDROP, "vgone", 0);
2716 		return;
2717 	}
2718 
2719 	/*
2720 	 * Clean out the filesystem specific data.
2721 	 */
2722 	vclean(vp, DOCLOSE, td);
2723 	VI_UNLOCK(vp);
2724 
2725 	/*
2726 	 * Delete from old mount point vnode list, if on one.
2727 	 */
2728 	if (vp->v_mount != NULL)
2729 		insmntque(vp, (struct mount *)0);
2730 	/*
2731 	 * If special device, remove it from special device alias list
2732 	 * if it is on one.
2733 	 */
2734 	if (vp->v_type == VCHR && vp->v_rdev != NULL && vp->v_rdev != NODEV) {
2735 		VI_LOCK(vp);
2736 		mtx_lock(&spechash_mtx);
2737 		SLIST_REMOVE(&vp->v_rdev->si_hlist, vp, vnode, v_specnext);
2738 		vp->v_rdev->si_usecount -= vp->v_usecount;
2739 		mtx_unlock(&spechash_mtx);
2740 		VI_UNLOCK(vp);
2741 		vp->v_rdev = NULL;
2742 	}
2743 
2744 	/*
2745 	 * If it is on the freelist and not already at the head,
2746 	 * move it to the head of the list. The test of the
2747 	 * VDOOMED flag and the reference count of zero is because
2748 	 * it will be removed from the free list by getnewvnode,
2749 	 * but will not have its reference count incremented until
2750 	 * after calling vgone. If the reference count were
2751 	 * incremented first, vgone would (incorrectly) try to
2752 	 * close the previous instance of the underlying object.
2753 	 */
2754 	VI_LOCK(vp);
2755 	if (vp->v_usecount == 0 && !(vp->v_iflag & VI_DOOMED)) {
2756 		mtx_lock(&vnode_free_list_mtx);
2757 		if (vp->v_iflag & VI_FREE) {
2758 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2759 		} else {
2760 			vp->v_iflag |= VI_FREE;
2761 			freevnodes++;
2762 		}
2763 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2764 		mtx_unlock(&vnode_free_list_mtx);
2765 	}
2766 
2767 	vp->v_type = VBAD;
2768 	VI_UNLOCK(vp);
2769 }
2770 
2771 /*
2772  * Lookup a vnode by device number.
2773  */
2774 int
2775 vfinddev(dev, type, vpp)
2776 	dev_t dev;
2777 	enum vtype type;
2778 	struct vnode **vpp;
2779 {
2780 	struct vnode *vp;
2781 
2782 	mtx_lock(&spechash_mtx);
2783 	SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2784 		if (type == vp->v_type) {
2785 			*vpp = vp;
2786 			mtx_unlock(&spechash_mtx);
2787 			return (1);
2788 		}
2789 	}
2790 	mtx_unlock(&spechash_mtx);
2791 	return (0);
2792 }
2793 
2794 /*
2795  * Calculate the total number of references to a special device.
2796  */
2797 int
2798 vcount(vp)
2799 	struct vnode *vp;
2800 {
2801 	int count;
2802 
2803 	mtx_lock(&spechash_mtx);
2804 	count = vp->v_rdev->si_usecount;
2805 	mtx_unlock(&spechash_mtx);
2806 	return (count);
2807 }
2808 
2809 /*
2810  * Same as above, but using the dev_t as argument
2811  */
2812 int
2813 count_dev(dev)
2814 	dev_t dev;
2815 {
2816 	struct vnode *vp;
2817 
2818 	vp = SLIST_FIRST(&dev->si_hlist);
2819 	if (vp == NULL)
2820 		return (0);
2821 	return(vcount(vp));
2822 }
2823 
2824 /*
2825  * Print out a description of a vnode.
2826  */
2827 static char *typename[] =
2828 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2829 
2830 void
2831 vprint(label, vp)
2832 	char *label;
2833 	struct vnode *vp;
2834 {
2835 	char buf[96];
2836 
2837 	if (label != NULL)
2838 		printf("%s: %p: ", label, (void *)vp);
2839 	else
2840 		printf("%p: ", (void *)vp);
2841 	printf("tag %s, type %s, usecount %d, writecount %d, refcount %d,",
2842 	    vp->v_tag, typename[vp->v_type], vp->v_usecount,
2843 	    vp->v_writecount, vp->v_holdcnt);
2844 	buf[0] = '\0';
2845 	if (vp->v_vflag & VV_ROOT)
2846 		strcat(buf, "|VV_ROOT");
2847 	if (vp->v_vflag & VV_TEXT)
2848 		strcat(buf, "|VV_TEXT");
2849 	if (vp->v_vflag & VV_SYSTEM)
2850 		strcat(buf, "|VV_SYSTEM");
2851 	if (vp->v_iflag & VI_XLOCK)
2852 		strcat(buf, "|VI_XLOCK");
2853 	if (vp->v_iflag & VI_XWANT)
2854 		strcat(buf, "|VI_XWANT");
2855 	if (vp->v_iflag & VI_BWAIT)
2856 		strcat(buf, "|VI_BWAIT");
2857 	if (vp->v_iflag & VI_DOOMED)
2858 		strcat(buf, "|VI_DOOMED");
2859 	if (vp->v_iflag & VI_FREE)
2860 		strcat(buf, "|VI_FREE");
2861 	if (vp->v_vflag & VV_OBJBUF)
2862 		strcat(buf, "|VV_OBJBUF");
2863 	if (buf[0] != '\0')
2864 		printf(" flags (%s),", &buf[1]);
2865 	lockmgr_printinfo(vp->v_vnlock);
2866 	printf("\n");
2867 	if (vp->v_data != NULL)
2868 		VOP_PRINT(vp);
2869 }
2870 
2871 #ifdef DDB
2872 #include <ddb/ddb.h>
2873 /*
2874  * List all of the locked vnodes in the system.
2875  * Called when debugging the kernel.
2876  */
2877 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2878 {
2879 	struct mount *mp, *nmp;
2880 	struct vnode *vp;
2881 
2882 	/*
2883 	 * Note: because this is DDB, we can't obey the locking semantics
2884 	 * for these structures, which means we could catch an inconsistent
2885 	 * state and dereference a nasty pointer.  Not much to be done
2886 	 * about that.
2887 	 */
2888 	printf("Locked vnodes\n");
2889 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2890 		nmp = TAILQ_NEXT(mp, mnt_list);
2891 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2892 			if (VOP_ISLOCKED(vp, NULL))
2893 				vprint(NULL, vp);
2894 		}
2895 		nmp = TAILQ_NEXT(mp, mnt_list);
2896 	}
2897 }
2898 #endif
2899 
2900 /*
2901  * Fill in a struct xvfsconf based on a struct vfsconf.
2902  */
2903 static void
2904 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2905 {
2906 
2907 	strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2908 	xvfsp->vfc_typenum = vfsp->vfc_typenum;
2909 	xvfsp->vfc_refcount = vfsp->vfc_refcount;
2910 	xvfsp->vfc_flags = vfsp->vfc_flags;
2911 	/*
2912 	 * These are unused in userland, we keep them
2913 	 * to not break binary compatibility.
2914 	 */
2915 	xvfsp->vfc_vfsops = NULL;
2916 	xvfsp->vfc_next = NULL;
2917 }
2918 
2919 static int
2920 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2921 {
2922 	struct vfsconf *vfsp;
2923 	struct xvfsconf *xvfsp;
2924 	int cnt, error, i;
2925 
2926 	cnt = 0;
2927 	for (vfsp = vfsconf; vfsp != NULL; vfsp = vfsp->vfc_next)
2928 		cnt++;
2929 	xvfsp = malloc(sizeof(struct xvfsconf) * cnt, M_TEMP, M_WAITOK);
2930 	/*
2931 	 * Handle the race that we will have here when struct vfsconf
2932 	 * will be locked down by using both cnt and checking vfc_next
2933 	 * against NULL to determine the end of the loop.  The race will
2934 	 * happen because we will have to unlock before calling malloc().
2935 	 * We are protected by Giant for now.
2936 	 */
2937 	i = 0;
2938 	for (vfsp = vfsconf; vfsp != NULL && i < cnt; vfsp = vfsp->vfc_next) {
2939 		vfsconf2x(vfsp, xvfsp + i);
2940 		i++;
2941 	}
2942 	error = SYSCTL_OUT(req, xvfsp, sizeof(struct xvfsconf) * i);
2943 	free(xvfsp, M_TEMP);
2944 	return (error);
2945 }
2946 
2947 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
2948     "S,xvfsconf", "List of all configured filesystems");
2949 
2950 /*
2951  * Top level filesystem related information gathering.
2952  */
2953 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2954 
2955 static int
2956 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2957 {
2958 	int *name = (int *)arg1 - 1;	/* XXX */
2959 	u_int namelen = arg2 + 1;	/* XXX */
2960 	struct vfsconf *vfsp;
2961 	struct xvfsconf xvfsp;
2962 
2963 	printf("WARNING: userland calling deprecated sysctl, "
2964 	    "please rebuild world\n");
2965 
2966 #if 1 || defined(COMPAT_PRELITE2)
2967 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2968 	if (namelen == 1)
2969 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2970 #endif
2971 
2972 	switch (name[1]) {
2973 	case VFS_MAXTYPENUM:
2974 		if (namelen != 2)
2975 			return (ENOTDIR);
2976 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2977 	case VFS_CONF:
2978 		if (namelen != 3)
2979 			return (ENOTDIR);	/* overloaded */
2980 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2981 			if (vfsp->vfc_typenum == name[2])
2982 				break;
2983 		if (vfsp == NULL)
2984 			return (EOPNOTSUPP);
2985 		vfsconf2x(vfsp, &xvfsp);
2986 		return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
2987 	}
2988 	return (EOPNOTSUPP);
2989 }
2990 
2991 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP, vfs_sysctl,
2992 	"Generic filesystem");
2993 
2994 #if 1 || defined(COMPAT_PRELITE2)
2995 
2996 static int
2997 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2998 {
2999 	int error;
3000 	struct vfsconf *vfsp;
3001 	struct ovfsconf ovfs;
3002 
3003 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
3004 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
3005 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
3006 		ovfs.vfc_index = vfsp->vfc_typenum;
3007 		ovfs.vfc_refcount = vfsp->vfc_refcount;
3008 		ovfs.vfc_flags = vfsp->vfc_flags;
3009 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3010 		if (error)
3011 			return error;
3012 	}
3013 	return 0;
3014 }
3015 
3016 #endif /* 1 || COMPAT_PRELITE2 */
3017 
3018 #define KINFO_VNODESLOP		10
3019 #ifdef notyet
3020 /*
3021  * Dump vnode list (via sysctl).
3022  */
3023 /* ARGSUSED */
3024 static int
3025 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3026 {
3027 	struct xvnode *xvn;
3028 	struct thread *td = req->td;
3029 	struct mount *mp;
3030 	struct vnode *vp;
3031 	int error, len, n;
3032 
3033 	/*
3034 	 * Stale numvnodes access is not fatal here.
3035 	 */
3036 	req->lock = 0;
3037 	len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3038 	if (!req->oldptr)
3039 		/* Make an estimate */
3040 		return (SYSCTL_OUT(req, 0, len));
3041 
3042 	sysctl_wire_old_buffer(req, 0);
3043 	xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3044 	n = 0;
3045 	mtx_lock(&mountlist_mtx);
3046 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3047 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td))
3048 			continue;
3049 		mtx_lock(&mntvnode_mtx);
3050 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3051 			if (n == len)
3052 				break;
3053 			vref(vp);
3054 			xvn[n].xv_size = sizeof *xvn;
3055 			xvn[n].xv_vnode = vp;
3056 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3057 			XV_COPY(usecount);
3058 			XV_COPY(writecount);
3059 			XV_COPY(holdcnt);
3060 			XV_COPY(id);
3061 			XV_COPY(mount);
3062 			XV_COPY(numoutput);
3063 			XV_COPY(type);
3064 #undef XV_COPY
3065 			xvn[n].xv_flag = vp->v_vflag;
3066 
3067 			switch (vp->v_type) {
3068 			case VREG:
3069 			case VDIR:
3070 			case VLNK:
3071 				xvn[n].xv_dev = vp->v_cachedfs;
3072 				xvn[n].xv_ino = vp->v_cachedid;
3073 				break;
3074 			case VBLK:
3075 			case VCHR:
3076 				if (vp->v_rdev == NULL) {
3077 					vrele(vp);
3078 					continue;
3079 				}
3080 				xvn[n].xv_dev = dev2udev(vp->v_rdev);
3081 				break;
3082 			case VSOCK:
3083 				xvn[n].xv_socket = vp->v_socket;
3084 				break;
3085 			case VFIFO:
3086 				xvn[n].xv_fifo = vp->v_fifoinfo;
3087 				break;
3088 			case VNON:
3089 			case VBAD:
3090 			default:
3091 				/* shouldn't happen? */
3092 				vrele(vp);
3093 				continue;
3094 			}
3095 			vrele(vp);
3096 			++n;
3097 		}
3098 		mtx_unlock(&mntvnode_mtx);
3099 		mtx_lock(&mountlist_mtx);
3100 		vfs_unbusy(mp, td);
3101 		if (n == len)
3102 			break;
3103 	}
3104 	mtx_unlock(&mountlist_mtx);
3105 
3106 	error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3107 	free(xvn, M_TEMP);
3108 	return (error);
3109 }
3110 
3111 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3112 	0, 0, sysctl_vnode, "S,xvnode", "");
3113 #endif
3114 
3115 /*
3116  * Check to see if a filesystem is mounted on a block device.
3117  */
3118 int
3119 vfs_mountedon(vp)
3120 	struct vnode *vp;
3121 {
3122 
3123 	if (vp->v_rdev->si_mountpoint != NULL)
3124 		return (EBUSY);
3125 	return (0);
3126 }
3127 
3128 /*
3129  * Unmount all filesystems. The list is traversed in reverse order
3130  * of mounting to avoid dependencies.
3131  */
3132 void
3133 vfs_unmountall()
3134 {
3135 	struct mount *mp;
3136 	struct thread *td;
3137 	int error;
3138 
3139 	if (curthread != NULL)
3140 		td = curthread;
3141 	else
3142 		td = FIRST_THREAD_IN_PROC(initproc); /* XXX XXX proc0? */
3143 	/*
3144 	 * Since this only runs when rebooting, it is not interlocked.
3145 	 */
3146 	while(!TAILQ_EMPTY(&mountlist)) {
3147 		mp = TAILQ_LAST(&mountlist, mntlist);
3148 		error = dounmount(mp, MNT_FORCE, td);
3149 		if (error) {
3150 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
3151 			printf("unmount of %s failed (",
3152 			    mp->mnt_stat.f_mntonname);
3153 			if (error == EBUSY)
3154 				printf("BUSY)\n");
3155 			else
3156 				printf("%d)\n", error);
3157 		} else {
3158 			/* The unmount has removed mp from the mountlist */
3159 		}
3160 	}
3161 }
3162 
3163 /*
3164  * perform msync on all vnodes under a mount point
3165  * the mount point must be locked.
3166  */
3167 void
3168 vfs_msync(struct mount *mp, int flags)
3169 {
3170 	struct vnode *vp, *nvp;
3171 	struct vm_object *obj;
3172 	int tries;
3173 
3174 	GIANT_REQUIRED;
3175 
3176 	tries = 5;
3177 	mtx_lock(&mntvnode_mtx);
3178 loop:
3179 	for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) {
3180 		if (vp->v_mount != mp) {
3181 			if (--tries > 0)
3182 				goto loop;
3183 			break;
3184 		}
3185 		nvp = TAILQ_NEXT(vp, v_nmntvnodes);
3186 
3187 		VI_LOCK(vp);
3188 		if (vp->v_iflag & VI_XLOCK) {	/* XXX: what if MNT_WAIT? */
3189 			VI_UNLOCK(vp);
3190 			continue;
3191 		}
3192 
3193 		if ((vp->v_iflag & VI_OBJDIRTY) &&
3194 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
3195 			mtx_unlock(&mntvnode_mtx);
3196 			if (!vget(vp,
3197 			    LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3198 			    curthread)) {
3199 				if (vp->v_vflag & VV_NOSYNC) {	/* unlinked */
3200 					vput(vp);
3201 					mtx_lock(&mntvnode_mtx);
3202 					continue;
3203 				}
3204 
3205 				if (VOP_GETVOBJECT(vp, &obj) == 0) {
3206 					VM_OBJECT_LOCK(obj);
3207 					vm_object_page_clean(obj, 0, 0,
3208 					    flags == MNT_WAIT ?
3209 					    OBJPC_SYNC : OBJPC_NOSYNC);
3210 					VM_OBJECT_UNLOCK(obj);
3211 				}
3212 				vput(vp);
3213 			}
3214 			mtx_lock(&mntvnode_mtx);
3215 			if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) {
3216 				if (--tries > 0)
3217 					goto loop;
3218 				break;
3219 			}
3220 		} else
3221 			VI_UNLOCK(vp);
3222 	}
3223 	mtx_unlock(&mntvnode_mtx);
3224 }
3225 
3226 /*
3227  * Create the VM object needed for VMIO and mmap support.  This
3228  * is done for all VREG files in the system.  Some filesystems might
3229  * afford the additional metadata buffering capability of the
3230  * VMIO code by making the device node be VMIO mode also.
3231  *
3232  * vp must be locked when vfs_object_create is called.
3233  */
3234 int
3235 vfs_object_create(vp, td, cred)
3236 	struct vnode *vp;
3237 	struct thread *td;
3238 	struct ucred *cred;
3239 {
3240 	GIANT_REQUIRED;
3241 	return (VOP_CREATEVOBJECT(vp, cred, td));
3242 }
3243 
3244 /*
3245  * Mark a vnode as free, putting it up for recycling.
3246  */
3247 void
3248 vfree(vp)
3249 	struct vnode *vp;
3250 {
3251 	ASSERT_VI_LOCKED(vp, "vfree");
3252 	mtx_lock(&vnode_free_list_mtx);
3253 	KASSERT((vp->v_iflag & VI_FREE) == 0, ("vnode already free"));
3254 	if (vp->v_iflag & VI_AGE) {
3255 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3256 	} else {
3257 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3258 	}
3259 	freevnodes++;
3260 	mtx_unlock(&vnode_free_list_mtx);
3261 	vp->v_iflag &= ~VI_AGE;
3262 	vp->v_iflag |= VI_FREE;
3263 }
3264 
3265 /*
3266  * Opposite of vfree() - mark a vnode as in use.
3267  */
3268 void
3269 vbusy(vp)
3270 	struct vnode *vp;
3271 {
3272 	ASSERT_VI_LOCKED(vp, "vbusy");
3273 	KASSERT((vp->v_iflag & VI_FREE) != 0, ("vnode not free"));
3274 
3275 	mtx_lock(&vnode_free_list_mtx);
3276 	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3277 	freevnodes--;
3278 	mtx_unlock(&vnode_free_list_mtx);
3279 
3280 	vp->v_iflag &= ~(VI_FREE|VI_AGE);
3281 }
3282 
3283 /*
3284  * Record a process's interest in events which might happen to
3285  * a vnode.  Because poll uses the historic select-style interface
3286  * internally, this routine serves as both the ``check for any
3287  * pending events'' and the ``record my interest in future events''
3288  * functions.  (These are done together, while the lock is held,
3289  * to avoid race conditions.)
3290  */
3291 int
3292 vn_pollrecord(vp, td, events)
3293 	struct vnode *vp;
3294 	struct thread *td;
3295 	short events;
3296 {
3297 
3298 	if (vp->v_pollinfo == NULL)
3299 		v_addpollinfo(vp);
3300 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3301 	if (vp->v_pollinfo->vpi_revents & events) {
3302 		/*
3303 		 * This leaves events we are not interested
3304 		 * in available for the other process which
3305 		 * which presumably had requested them
3306 		 * (otherwise they would never have been
3307 		 * recorded).
3308 		 */
3309 		events &= vp->v_pollinfo->vpi_revents;
3310 		vp->v_pollinfo->vpi_revents &= ~events;
3311 
3312 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
3313 		return events;
3314 	}
3315 	vp->v_pollinfo->vpi_events |= events;
3316 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3317 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3318 	return 0;
3319 }
3320 
3321 /*
3322  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
3323  * it is possible for us to miss an event due to race conditions, but
3324  * that condition is expected to be rare, so for the moment it is the
3325  * preferred interface.
3326  */
3327 void
3328 vn_pollevent(vp, events)
3329 	struct vnode *vp;
3330 	short events;
3331 {
3332 
3333 	if (vp->v_pollinfo == NULL)
3334 		v_addpollinfo(vp);
3335 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3336 	if (vp->v_pollinfo->vpi_events & events) {
3337 		/*
3338 		 * We clear vpi_events so that we don't
3339 		 * call selwakeup() twice if two events are
3340 		 * posted before the polling process(es) is
3341 		 * awakened.  This also ensures that we take at
3342 		 * most one selwakeup() if the polling process
3343 		 * is no longer interested.  However, it does
3344 		 * mean that only one event can be noticed at
3345 		 * a time.  (Perhaps we should only clear those
3346 		 * event bits which we note?) XXX
3347 		 */
3348 		vp->v_pollinfo->vpi_events = 0;	/* &= ~events ??? */
3349 		vp->v_pollinfo->vpi_revents |= events;
3350 		selwakeup(&vp->v_pollinfo->vpi_selinfo);
3351 	}
3352 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3353 }
3354 
3355 /*
3356  * Wake up anyone polling on vp because it is being revoked.
3357  * This depends on dead_poll() returning POLLHUP for correct
3358  * behavior.
3359  */
3360 void
3361 vn_pollgone(vp)
3362 	struct vnode *vp;
3363 {
3364 
3365 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3366 	VN_KNOTE(vp, NOTE_REVOKE);
3367 	if (vp->v_pollinfo->vpi_events) {
3368 		vp->v_pollinfo->vpi_events = 0;
3369 		selwakeup(&vp->v_pollinfo->vpi_selinfo);
3370 	}
3371 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3372 }
3373 
3374 
3375 
3376 /*
3377  * Routine to create and manage a filesystem syncer vnode.
3378  */
3379 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
3380 static int	sync_fsync(struct  vop_fsync_args *);
3381 static int	sync_inactive(struct  vop_inactive_args *);
3382 static int	sync_reclaim(struct  vop_reclaim_args *);
3383 
3384 static vop_t **sync_vnodeop_p;
3385 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
3386 	{ &vop_default_desc,	(vop_t *) vop_eopnotsupp },
3387 	{ &vop_close_desc,	(vop_t *) sync_close },		/* close */
3388 	{ &vop_fsync_desc,	(vop_t *) sync_fsync },		/* fsync */
3389 	{ &vop_inactive_desc,	(vop_t *) sync_inactive },	/* inactive */
3390 	{ &vop_reclaim_desc,	(vop_t *) sync_reclaim },	/* reclaim */
3391 	{ &vop_lock_desc,	(vop_t *) vop_stdlock },	/* lock */
3392 	{ &vop_unlock_desc,	(vop_t *) vop_stdunlock },	/* unlock */
3393 	{ &vop_islocked_desc,	(vop_t *) vop_stdislocked },	/* islocked */
3394 	{ NULL, NULL }
3395 };
3396 static struct vnodeopv_desc sync_vnodeop_opv_desc =
3397 	{ &sync_vnodeop_p, sync_vnodeop_entries };
3398 
3399 VNODEOP_SET(sync_vnodeop_opv_desc);
3400 
3401 /*
3402  * Create a new filesystem syncer vnode for the specified mount point.
3403  */
3404 int
3405 vfs_allocate_syncvnode(mp)
3406 	struct mount *mp;
3407 {
3408 	struct vnode *vp;
3409 	static long start, incr, next;
3410 	int error;
3411 
3412 	/* Allocate a new vnode */
3413 	if ((error = getnewvnode("syncer", mp, sync_vnodeop_p, &vp)) != 0) {
3414 		mp->mnt_syncer = NULL;
3415 		return (error);
3416 	}
3417 	vp->v_type = VNON;
3418 	/*
3419 	 * Place the vnode onto the syncer worklist. We attempt to
3420 	 * scatter them about on the list so that they will go off
3421 	 * at evenly distributed times even if all the filesystems
3422 	 * are mounted at once.
3423 	 */
3424 	next += incr;
3425 	if (next == 0 || next > syncer_maxdelay) {
3426 		start /= 2;
3427 		incr /= 2;
3428 		if (start == 0) {
3429 			start = syncer_maxdelay / 2;
3430 			incr = syncer_maxdelay;
3431 		}
3432 		next = start;
3433 	}
3434 	VI_LOCK(vp);
3435 	vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
3436 	VI_UNLOCK(vp);
3437 	mp->mnt_syncer = vp;
3438 	return (0);
3439 }
3440 
3441 /*
3442  * Do a lazy sync of the filesystem.
3443  */
3444 static int
3445 sync_fsync(ap)
3446 	struct vop_fsync_args /* {
3447 		struct vnode *a_vp;
3448 		struct ucred *a_cred;
3449 		int a_waitfor;
3450 		struct thread *a_td;
3451 	} */ *ap;
3452 {
3453 	struct vnode *syncvp = ap->a_vp;
3454 	struct mount *mp = syncvp->v_mount;
3455 	struct thread *td = ap->a_td;
3456 	int error, asyncflag;
3457 
3458 	/*
3459 	 * We only need to do something if this is a lazy evaluation.
3460 	 */
3461 	if (ap->a_waitfor != MNT_LAZY)
3462 		return (0);
3463 
3464 	/*
3465 	 * Move ourselves to the back of the sync list.
3466 	 */
3467 	VI_LOCK(syncvp);
3468 	vn_syncer_add_to_worklist(syncvp, syncdelay);
3469 	VI_UNLOCK(syncvp);
3470 
3471 	/*
3472 	 * Walk the list of vnodes pushing all that are dirty and
3473 	 * not already on the sync list.
3474 	 */
3475 	mtx_lock(&mountlist_mtx);
3476 	if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) {
3477 		mtx_unlock(&mountlist_mtx);
3478 		return (0);
3479 	}
3480 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3481 		vfs_unbusy(mp, td);
3482 		return (0);
3483 	}
3484 	asyncflag = mp->mnt_flag & MNT_ASYNC;
3485 	mp->mnt_flag &= ~MNT_ASYNC;
3486 	vfs_msync(mp, MNT_NOWAIT);
3487 	error = VFS_SYNC(mp, MNT_LAZY, ap->a_cred, td);
3488 	if (asyncflag)
3489 		mp->mnt_flag |= MNT_ASYNC;
3490 	vn_finished_write(mp);
3491 	vfs_unbusy(mp, td);
3492 	return (error);
3493 }
3494 
3495 /*
3496  * The syncer vnode is no referenced.
3497  */
3498 static int
3499 sync_inactive(ap)
3500 	struct vop_inactive_args /* {
3501 		struct vnode *a_vp;
3502 		struct thread *a_td;
3503 	} */ *ap;
3504 {
3505 
3506 	VOP_UNLOCK(ap->a_vp, 0, ap->a_td);
3507 	vgone(ap->a_vp);
3508 	return (0);
3509 }
3510 
3511 /*
3512  * The syncer vnode is no longer needed and is being decommissioned.
3513  *
3514  * Modifications to the worklist must be protected by sync_mtx.
3515  */
3516 static int
3517 sync_reclaim(ap)
3518 	struct vop_reclaim_args /* {
3519 		struct vnode *a_vp;
3520 	} */ *ap;
3521 {
3522 	struct vnode *vp = ap->a_vp;
3523 
3524 	VI_LOCK(vp);
3525 	vp->v_mount->mnt_syncer = NULL;
3526 	if (vp->v_iflag & VI_ONWORKLST) {
3527 		mtx_lock(&sync_mtx);
3528 		LIST_REMOVE(vp, v_synclist);
3529 		mtx_unlock(&sync_mtx);
3530 		vp->v_iflag &= ~VI_ONWORKLST;
3531 	}
3532 	VI_UNLOCK(vp);
3533 
3534 	return (0);
3535 }
3536 
3537 /*
3538  * extract the dev_t from a VCHR
3539  */
3540 dev_t
3541 vn_todev(vp)
3542 	struct vnode *vp;
3543 {
3544 	if (vp->v_type != VCHR)
3545 		return (NODEV);
3546 	return (vp->v_rdev);
3547 }
3548 
3549 /*
3550  * Check if vnode represents a disk device
3551  */
3552 int
3553 vn_isdisk(vp, errp)
3554 	struct vnode *vp;
3555 	int *errp;
3556 {
3557 	struct cdevsw *cdevsw;
3558 
3559 	if (vp->v_type != VCHR) {
3560 		if (errp != NULL)
3561 			*errp = ENOTBLK;
3562 		return (0);
3563 	}
3564 	if (vp->v_rdev == NULL) {
3565 		if (errp != NULL)
3566 			*errp = ENXIO;
3567 		return (0);
3568 	}
3569 	cdevsw = devsw(vp->v_rdev);
3570 	if (cdevsw == NULL) {
3571 		if (errp != NULL)
3572 			*errp = ENXIO;
3573 		return (0);
3574 	}
3575 	if (!(cdevsw->d_flags & D_DISK)) {
3576 		if (errp != NULL)
3577 			*errp = ENOTBLK;
3578 		return (0);
3579 	}
3580 	if (errp != NULL)
3581 		*errp = 0;
3582 	return (1);
3583 }
3584 
3585 /*
3586  * Free data allocated by namei(); see namei(9) for details.
3587  */
3588 void
3589 NDFREE(ndp, flags)
3590      struct nameidata *ndp;
3591      const u_int flags;
3592 {
3593 	if (!(flags & NDF_NO_FREE_PNBUF) &&
3594 	    (ndp->ni_cnd.cn_flags & HASBUF)) {
3595 		uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
3596 		ndp->ni_cnd.cn_flags &= ~HASBUF;
3597 	}
3598 	if (!(flags & NDF_NO_DVP_UNLOCK) &&
3599 	    (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
3600 	    ndp->ni_dvp != ndp->ni_vp)
3601 		VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_thread);
3602 	if (!(flags & NDF_NO_DVP_RELE) &&
3603 	    (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
3604 		vrele(ndp->ni_dvp);
3605 		ndp->ni_dvp = NULL;
3606 	}
3607 	if (!(flags & NDF_NO_VP_UNLOCK) &&
3608 	    (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
3609 		VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_thread);
3610 	if (!(flags & NDF_NO_VP_RELE) &&
3611 	    ndp->ni_vp) {
3612 		vrele(ndp->ni_vp);
3613 		ndp->ni_vp = NULL;
3614 	}
3615 	if (!(flags & NDF_NO_STARTDIR_RELE) &&
3616 	    (ndp->ni_cnd.cn_flags & SAVESTART)) {
3617 		vrele(ndp->ni_startdir);
3618 		ndp->ni_startdir = NULL;
3619 	}
3620 }
3621 
3622 /*
3623  * Common filesystem object access control check routine.  Accepts a
3624  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3625  * and optional call-by-reference privused argument allowing vaccess()
3626  * to indicate to the caller whether privilege was used to satisfy the
3627  * request (obsoleted).  Returns 0 on success, or an errno on failure.
3628  */
3629 int
3630 vaccess(type, file_mode, file_uid, file_gid, acc_mode, cred, privused)
3631 	enum vtype type;
3632 	mode_t file_mode;
3633 	uid_t file_uid;
3634 	gid_t file_gid;
3635 	mode_t acc_mode;
3636 	struct ucred *cred;
3637 	int *privused;
3638 {
3639 	mode_t dac_granted;
3640 #ifdef CAPABILITIES
3641 	mode_t cap_granted;
3642 #endif
3643 
3644 	/*
3645 	 * Look for a normal, non-privileged way to access the file/directory
3646 	 * as requested.  If it exists, go with that.
3647 	 */
3648 
3649 	if (privused != NULL)
3650 		*privused = 0;
3651 
3652 	dac_granted = 0;
3653 
3654 	/* Check the owner. */
3655 	if (cred->cr_uid == file_uid) {
3656 		dac_granted |= VADMIN;
3657 		if (file_mode & S_IXUSR)
3658 			dac_granted |= VEXEC;
3659 		if (file_mode & S_IRUSR)
3660 			dac_granted |= VREAD;
3661 		if (file_mode & S_IWUSR)
3662 			dac_granted |= (VWRITE | VAPPEND);
3663 
3664 		if ((acc_mode & dac_granted) == acc_mode)
3665 			return (0);
3666 
3667 		goto privcheck;
3668 	}
3669 
3670 	/* Otherwise, check the groups (first match) */
3671 	if (groupmember(file_gid, cred)) {
3672 		if (file_mode & S_IXGRP)
3673 			dac_granted |= VEXEC;
3674 		if (file_mode & S_IRGRP)
3675 			dac_granted |= VREAD;
3676 		if (file_mode & S_IWGRP)
3677 			dac_granted |= (VWRITE | VAPPEND);
3678 
3679 		if ((acc_mode & dac_granted) == acc_mode)
3680 			return (0);
3681 
3682 		goto privcheck;
3683 	}
3684 
3685 	/* Otherwise, check everyone else. */
3686 	if (file_mode & S_IXOTH)
3687 		dac_granted |= VEXEC;
3688 	if (file_mode & S_IROTH)
3689 		dac_granted |= VREAD;
3690 	if (file_mode & S_IWOTH)
3691 		dac_granted |= (VWRITE | VAPPEND);
3692 	if ((acc_mode & dac_granted) == acc_mode)
3693 		return (0);
3694 
3695 privcheck:
3696 	if (!suser_cred(cred, PRISON_ROOT)) {
3697 		/* XXX audit: privilege used */
3698 		if (privused != NULL)
3699 			*privused = 1;
3700 		return (0);
3701 	}
3702 
3703 #ifdef CAPABILITIES
3704 	/*
3705 	 * Build a capability mask to determine if the set of capabilities
3706 	 * satisfies the requirements when combined with the granted mask
3707 	 * from above.
3708 	 * For each capability, if the capability is required, bitwise
3709 	 * or the request type onto the cap_granted mask.
3710 	 */
3711 	cap_granted = 0;
3712 
3713 	if (type == VDIR) {
3714 		/*
3715 		 * For directories, use CAP_DAC_READ_SEARCH to satisfy
3716 		 * VEXEC requests, instead of CAP_DAC_EXECUTE.
3717 		 */
3718 		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3719 		    !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3720 			cap_granted |= VEXEC;
3721 	} else {
3722 		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3723 		    !cap_check(cred, NULL, CAP_DAC_EXECUTE, PRISON_ROOT))
3724 			cap_granted |= VEXEC;
3725 	}
3726 
3727 	if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
3728 	    !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
3729 		cap_granted |= VREAD;
3730 
3731 	if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3732 	    !cap_check(cred, NULL, CAP_DAC_WRITE, PRISON_ROOT))
3733 		cap_granted |= (VWRITE | VAPPEND);
3734 
3735 	if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3736 	    !cap_check(cred, NULL, CAP_FOWNER, PRISON_ROOT))
3737 		cap_granted |= VADMIN;
3738 
3739 	if ((acc_mode & (cap_granted | dac_granted)) == acc_mode) {
3740 		/* XXX audit: privilege used */
3741 		if (privused != NULL)
3742 			*privused = 1;
3743 		return (0);
3744 	}
3745 #endif
3746 
3747 	return ((acc_mode & VADMIN) ? EPERM : EACCES);
3748 }
3749 
3750 /*
3751  * Credential check based on process requesting service, and per-attribute
3752  * permissions.
3753  */
3754 int
3755 extattr_check_cred(struct vnode *vp, int attrnamespace,
3756     struct ucred *cred, struct thread *td, int access)
3757 {
3758 
3759 	/*
3760 	 * Kernel-invoked always succeeds.
3761 	 */
3762 	if (cred == NOCRED)
3763 		return (0);
3764 
3765 	/*
3766 	 * Do not allow privileged processes in jail to directly
3767 	 * manipulate system attributes.
3768 	 *
3769 	 * XXX What capability should apply here?
3770 	 * Probably CAP_SYS_SETFFLAG.
3771 	 */
3772 	switch (attrnamespace) {
3773 	case EXTATTR_NAMESPACE_SYSTEM:
3774 		/* Potentially should be: return (EPERM); */
3775 		return (suser_cred(cred, 0));
3776 	case EXTATTR_NAMESPACE_USER:
3777 		return (VOP_ACCESS(vp, access, cred, td));
3778 	default:
3779 		return (EPERM);
3780 	}
3781 }
3782