xref: /freebsd/sys/kern/vfs_subr.c (revision 8655c70597b0e0918c82114b1186df5669b83eb6)
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  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)vfs_subr.c	8.31 (Berkeley) 5/26/95
35  */
36 
37 /*
38  * External virtual filesystem routines
39  */
40 
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
43 
44 #include "opt_ddb.h"
45 #include "opt_mac.h"
46 
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/bio.h>
50 #include <sys/buf.h>
51 #include <sys/condvar.h>
52 #include <sys/conf.h>
53 #include <sys/dirent.h>
54 #include <sys/event.h>
55 #include <sys/eventhandler.h>
56 #include <sys/extattr.h>
57 #include <sys/file.h>
58 #include <sys/fcntl.h>
59 #include <sys/jail.h>
60 #include <sys/kdb.h>
61 #include <sys/kernel.h>
62 #include <sys/kthread.h>
63 #include <sys/lockf.h>
64 #include <sys/malloc.h>
65 #include <sys/mount.h>
66 #include <sys/namei.h>
67 #include <sys/priv.h>
68 #include <sys/reboot.h>
69 #include <sys/sleepqueue.h>
70 #include <sys/stat.h>
71 #include <sys/sysctl.h>
72 #include <sys/syslog.h>
73 #include <sys/vmmeter.h>
74 #include <sys/vnode.h>
75 
76 #include <machine/stdarg.h>
77 
78 #include <security/mac/mac_framework.h>
79 
80 #include <vm/vm.h>
81 #include <vm/vm_object.h>
82 #include <vm/vm_extern.h>
83 #include <vm/pmap.h>
84 #include <vm/vm_map.h>
85 #include <vm/vm_page.h>
86 #include <vm/vm_kern.h>
87 #include <vm/uma.h>
88 
89 #ifdef DDB
90 #include <ddb/ddb.h>
91 #endif
92 
93 #define	WI_MPSAFEQ	0
94 #define	WI_GIANTQ	1
95 
96 static MALLOC_DEFINE(M_NETADDR, "subr_export_host", "Export host address structure");
97 
98 static void	delmntque(struct vnode *vp);
99 static int	flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
100 		    int slpflag, int slptimeo);
101 static void	syncer_shutdown(void *arg, int howto);
102 static int	vtryrecycle(struct vnode *vp);
103 static void	vbusy(struct vnode *vp);
104 static void	vinactive(struct vnode *, struct thread *);
105 static void	v_incr_usecount(struct vnode *);
106 static void	v_decr_usecount(struct vnode *);
107 static void	v_decr_useonly(struct vnode *);
108 static void	v_upgrade_usecount(struct vnode *);
109 static void	vfree(struct vnode *);
110 static void	vnlru_free(int);
111 static void	vgonel(struct vnode *);
112 static void	vfs_knllock(void *arg);
113 static void	vfs_knlunlock(void *arg);
114 static int	vfs_knllocked(void *arg);
115 static void	destroy_vpollinfo(struct vpollinfo *vi);
116 
117 /*
118  * Enable Giant pushdown based on whether or not the vm is mpsafe in this
119  * build.  Without mpsafevm the buffer cache can not run Giant free.
120  */
121 int mpsafe_vfs = 1;
122 TUNABLE_INT("debug.mpsafevfs", &mpsafe_vfs);
123 SYSCTL_INT(_debug, OID_AUTO, mpsafevfs, CTLFLAG_RD, &mpsafe_vfs, 0,
124     "MPSAFE VFS");
125 
126 /*
127  * Number of vnodes in existence.  Increased whenever getnewvnode()
128  * allocates a new vnode, decreased on vdestroy() called on VI_DOOMed
129  * vnode.
130  */
131 static unsigned long	numvnodes;
132 
133 SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
134 
135 /*
136  * Conversion tables for conversion from vnode types to inode formats
137  * and back.
138  */
139 enum vtype iftovt_tab[16] = {
140 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
141 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
142 };
143 int vttoif_tab[10] = {
144 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
145 	S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
146 };
147 
148 /*
149  * List of vnodes that are ready for recycling.
150  */
151 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
152 
153 /*
154  * Free vnode target.  Free vnodes may simply be files which have been stat'd
155  * but not read.  This is somewhat common, and a small cache of such files
156  * should be kept to avoid recreation costs.
157  */
158 static u_long wantfreevnodes;
159 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
160 /* Number of vnodes in the free list. */
161 static u_long freevnodes;
162 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
163 
164 /*
165  * Various variables used for debugging the new implementation of
166  * reassignbuf().
167  * XXX these are probably of (very) limited utility now.
168  */
169 static int reassignbufcalls;
170 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
171 
172 /*
173  * Cache for the mount type id assigned to NFS.  This is used for
174  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
175  */
176 int	nfs_mount_type = -1;
177 
178 /* To keep more than one thread at a time from running vfs_getnewfsid */
179 static struct mtx mntid_mtx;
180 
181 /*
182  * Lock for any access to the following:
183  *	vnode_free_list
184  *	numvnodes
185  *	freevnodes
186  */
187 static struct mtx vnode_free_list_mtx;
188 
189 /* Publicly exported FS */
190 struct nfs_public nfs_pub;
191 
192 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
193 static uma_zone_t vnode_zone;
194 static uma_zone_t vnodepoll_zone;
195 
196 /* Set to 1 to print out reclaim of active vnodes */
197 int	prtactive;
198 
199 /*
200  * The workitem queue.
201  *
202  * It is useful to delay writes of file data and filesystem metadata
203  * for tens of seconds so that quickly created and deleted files need
204  * not waste disk bandwidth being created and removed. To realize this,
205  * we append vnodes to a "workitem" queue. When running with a soft
206  * updates implementation, most pending metadata dependencies should
207  * not wait for more than a few seconds. Thus, mounted on block devices
208  * are delayed only about a half the time that file data is delayed.
209  * Similarly, directory updates are more critical, so are only delayed
210  * about a third the time that file data is delayed. Thus, there are
211  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
212  * one each second (driven off the filesystem syncer process). The
213  * syncer_delayno variable indicates the next queue that is to be processed.
214  * Items that need to be processed soon are placed in this queue:
215  *
216  *	syncer_workitem_pending[syncer_delayno]
217  *
218  * A delay of fifteen seconds is done by placing the request fifteen
219  * entries later in the queue:
220  *
221  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
222  *
223  */
224 static int syncer_delayno;
225 static long syncer_mask;
226 LIST_HEAD(synclist, bufobj);
227 static struct synclist *syncer_workitem_pending[2];
228 /*
229  * The sync_mtx protects:
230  *	bo->bo_synclist
231  *	sync_vnode_count
232  *	syncer_delayno
233  *	syncer_state
234  *	syncer_workitem_pending
235  *	syncer_worklist_len
236  *	rushjob
237  */
238 static struct mtx sync_mtx;
239 static struct cv sync_wakeup;
240 
241 #define SYNCER_MAXDELAY		32
242 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
243 static int syncdelay = 30;		/* max time to delay syncing data */
244 static int filedelay = 30;		/* time to delay syncing files */
245 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
246 static int dirdelay = 29;		/* time to delay syncing directories */
247 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
248 static int metadelay = 28;		/* time to delay syncing metadata */
249 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
250 static int rushjob;		/* number of slots to run ASAP */
251 static int stat_rush_requests;	/* number of times I/O speeded up */
252 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
253 
254 /*
255  * When shutting down the syncer, run it at four times normal speed.
256  */
257 #define SYNCER_SHUTDOWN_SPEEDUP		4
258 static int sync_vnode_count;
259 static int syncer_worklist_len;
260 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
261     syncer_state;
262 
263 /*
264  * Number of vnodes we want to exist at any one time.  This is mostly used
265  * to size hash tables in vnode-related code.  It is normally not used in
266  * getnewvnode(), as wantfreevnodes is normally nonzero.)
267  *
268  * XXX desiredvnodes is historical cruft and should not exist.
269  */
270 int desiredvnodes;
271 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
272     &desiredvnodes, 0, "Maximum number of vnodes");
273 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
274     &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
275 static int vnlru_nowhere;
276 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
277     &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
278 
279 /*
280  * Macros to control when a vnode is freed and recycled.  All require
281  * the vnode interlock.
282  */
283 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
284 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
285 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
286 
287 
288 /*
289  * Initialize the vnode management data structures.
290  */
291 #ifndef	MAXVNODES_MAX
292 #define	MAXVNODES_MAX	100000
293 #endif
294 static void
295 vntblinit(void *dummy __unused)
296 {
297 
298 	/*
299 	 * Desiredvnodes is a function of the physical memory size and
300 	 * the kernel's heap size.  Specifically, desiredvnodes scales
301 	 * in proportion to the physical memory size until two fifths
302 	 * of the kernel's heap size is consumed by vnodes and vm
303 	 * objects.
304 	 */
305 	desiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * vm_kmem_size /
306 	    (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
307 	if (desiredvnodes > MAXVNODES_MAX) {
308 		if (bootverbose)
309 			printf("Reducing kern.maxvnodes %d -> %d\n",
310 			    desiredvnodes, MAXVNODES_MAX);
311 		desiredvnodes = MAXVNODES_MAX;
312 	}
313 	wantfreevnodes = desiredvnodes / 4;
314 	mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
315 	TAILQ_INIT(&vnode_free_list);
316 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
317 	vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
318 	    NULL, NULL, UMA_ALIGN_PTR, 0);
319 	vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
320 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
321 	/*
322 	 * Initialize the filesystem syncer.
323 	 */
324 	syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE,
325 	    &syncer_mask);
326 	syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE,
327 	    &syncer_mask);
328 	syncer_maxdelay = syncer_mask + 1;
329 	mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
330 	cv_init(&sync_wakeup, "syncer");
331 }
332 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
333 
334 
335 /*
336  * Mark a mount point as busy. Used to synchronize access and to delay
337  * unmounting. Eventually, mountlist_mtx is not released on failure.
338  */
339 int
340 vfs_busy(struct mount *mp, int flags)
341 {
342 
343 	MPASS((flags & ~MBF_MASK) == 0);
344 	CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
345 
346 	MNT_ILOCK(mp);
347 	MNT_REF(mp);
348 	/*
349 	 * If mount point is currenly being unmounted, sleep until the
350 	 * mount point fate is decided.  If thread doing the unmounting fails,
351 	 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
352 	 * that this mount point has survived the unmount attempt and vfs_busy
353 	 * should retry.  Otherwise the unmounter thread will set MNTK_REFEXPIRE
354 	 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
355 	 * about to be really destroyed.  vfs_busy needs to release its
356 	 * reference on the mount point in this case and return with ENOENT,
357 	 * telling the caller that mount mount it tried to busy is no longer
358 	 * valid.
359 	 */
360 	while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
361 		if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
362 			MNT_REL(mp);
363 			MNT_IUNLOCK(mp);
364 			CTR1(KTR_VFS, "%s: failed busying before sleeping",
365 			    __func__);
366 			return (ENOENT);
367 		}
368 		if (flags & MBF_MNTLSTLOCK)
369 			mtx_unlock(&mountlist_mtx);
370 		mp->mnt_kern_flag |= MNTK_MWAIT;
371 		msleep(mp, MNT_MTX(mp), PVFS, "vfs_busy", 0);
372 		if (flags & MBF_MNTLSTLOCK)
373 			mtx_lock(&mountlist_mtx);
374 	}
375 	if (flags & MBF_MNTLSTLOCK)
376 		mtx_unlock(&mountlist_mtx);
377 	mp->mnt_lockref++;
378 	MNT_IUNLOCK(mp);
379 	return (0);
380 }
381 
382 /*
383  * Free a busy filesystem.
384  */
385 void
386 vfs_unbusy(struct mount *mp)
387 {
388 
389 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
390 	MNT_ILOCK(mp);
391 	MNT_REL(mp);
392 	KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
393 	mp->mnt_lockref--;
394 	if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
395 		MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
396 		CTR1(KTR_VFS, "%s: waking up waiters", __func__);
397 		mp->mnt_kern_flag &= ~MNTK_DRAINING;
398 		wakeup(&mp->mnt_lockref);
399 	}
400 	MNT_IUNLOCK(mp);
401 }
402 
403 /*
404  * Lookup a mount point by filesystem identifier.
405  */
406 struct mount *
407 vfs_getvfs(fsid_t *fsid)
408 {
409 	struct mount *mp;
410 
411 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
412 	mtx_lock(&mountlist_mtx);
413 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
414 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
415 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
416 			vfs_ref(mp);
417 			mtx_unlock(&mountlist_mtx);
418 			return (mp);
419 		}
420 	}
421 	mtx_unlock(&mountlist_mtx);
422 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
423 	return ((struct mount *) 0);
424 }
425 
426 /*
427  * Lookup a mount point by filesystem identifier, busying it before
428  * returning.
429  */
430 struct mount *
431 vfs_busyfs(fsid_t *fsid)
432 {
433 	struct mount *mp;
434 	int error;
435 
436 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
437 	mtx_lock(&mountlist_mtx);
438 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
439 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
440 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
441 			error = vfs_busy(mp, MBF_MNTLSTLOCK);
442 			if (error) {
443 				mtx_unlock(&mountlist_mtx);
444 				return (NULL);
445 			}
446 			return (mp);
447 		}
448 	}
449 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
450 	mtx_unlock(&mountlist_mtx);
451 	return ((struct mount *) 0);
452 }
453 
454 /*
455  * Check if a user can access privileged mount options.
456  */
457 int
458 vfs_suser(struct mount *mp, struct thread *td)
459 {
460 	int error;
461 
462 	/*
463 	 * If the thread is jailed, but this is not a jail-friendly file
464 	 * system, deny immediately.
465 	 */
466 	if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
467 		return (EPERM);
468 
469 	/*
470 	 * If the file system was mounted outside a jail and a jailed thread
471 	 * tries to access it, deny immediately.
472 	 */
473 	if (!jailed(mp->mnt_cred) && jailed(td->td_ucred))
474 		return (EPERM);
475 
476 	/*
477 	 * If the file system was mounted inside different jail that the jail of
478 	 * the calling thread, deny immediately.
479 	 */
480 	if (jailed(mp->mnt_cred) && jailed(td->td_ucred) &&
481 	    mp->mnt_cred->cr_prison != td->td_ucred->cr_prison) {
482 		return (EPERM);
483 	}
484 
485 	/*
486 	 * If file system supports delegated administration, we don't check
487 	 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
488 	 * by the file system itself.
489 	 * If this is not the user that did original mount, we check for
490 	 * the PRIV_VFS_MOUNT_OWNER privilege.
491 	 */
492 	if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
493 	    mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
494 		if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
495 			return (error);
496 	}
497 	return (0);
498 }
499 
500 /*
501  * Get a new unique fsid.  Try to make its val[0] unique, since this value
502  * will be used to create fake device numbers for stat().  Also try (but
503  * not so hard) make its val[0] unique mod 2^16, since some emulators only
504  * support 16-bit device numbers.  We end up with unique val[0]'s for the
505  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
506  *
507  * Keep in mind that several mounts may be running in parallel.  Starting
508  * the search one past where the previous search terminated is both a
509  * micro-optimization and a defense against returning the same fsid to
510  * different mounts.
511  */
512 void
513 vfs_getnewfsid(struct mount *mp)
514 {
515 	static u_int16_t mntid_base;
516 	struct mount *nmp;
517 	fsid_t tfsid;
518 	int mtype;
519 
520 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
521 	mtx_lock(&mntid_mtx);
522 	mtype = mp->mnt_vfc->vfc_typenum;
523 	tfsid.val[1] = mtype;
524 	mtype = (mtype & 0xFF) << 24;
525 	for (;;) {
526 		tfsid.val[0] = makedev(255,
527 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
528 		mntid_base++;
529 		if ((nmp = vfs_getvfs(&tfsid)) == NULL)
530 			break;
531 		vfs_rel(nmp);
532 	}
533 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
534 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
535 	mtx_unlock(&mntid_mtx);
536 }
537 
538 /*
539  * Knob to control the precision of file timestamps:
540  *
541  *   0 = seconds only; nanoseconds zeroed.
542  *   1 = seconds and nanoseconds, accurate within 1/HZ.
543  *   2 = seconds and nanoseconds, truncated to microseconds.
544  * >=3 = seconds and nanoseconds, maximum precision.
545  */
546 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
547 
548 static int timestamp_precision = TSP_SEC;
549 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
550     &timestamp_precision, 0, "");
551 
552 /*
553  * Get a current timestamp.
554  */
555 void
556 vfs_timestamp(struct timespec *tsp)
557 {
558 	struct timeval tv;
559 
560 	switch (timestamp_precision) {
561 	case TSP_SEC:
562 		tsp->tv_sec = time_second;
563 		tsp->tv_nsec = 0;
564 		break;
565 	case TSP_HZ:
566 		getnanotime(tsp);
567 		break;
568 	case TSP_USEC:
569 		microtime(&tv);
570 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
571 		break;
572 	case TSP_NSEC:
573 	default:
574 		nanotime(tsp);
575 		break;
576 	}
577 }
578 
579 /*
580  * Set vnode attributes to VNOVAL
581  */
582 void
583 vattr_null(struct vattr *vap)
584 {
585 
586 	vap->va_type = VNON;
587 	vap->va_size = VNOVAL;
588 	vap->va_bytes = VNOVAL;
589 	vap->va_mode = VNOVAL;
590 	vap->va_nlink = VNOVAL;
591 	vap->va_uid = VNOVAL;
592 	vap->va_gid = VNOVAL;
593 	vap->va_fsid = VNOVAL;
594 	vap->va_fileid = VNOVAL;
595 	vap->va_blocksize = VNOVAL;
596 	vap->va_rdev = VNOVAL;
597 	vap->va_atime.tv_sec = VNOVAL;
598 	vap->va_atime.tv_nsec = VNOVAL;
599 	vap->va_mtime.tv_sec = VNOVAL;
600 	vap->va_mtime.tv_nsec = VNOVAL;
601 	vap->va_ctime.tv_sec = VNOVAL;
602 	vap->va_ctime.tv_nsec = VNOVAL;
603 	vap->va_birthtime.tv_sec = VNOVAL;
604 	vap->va_birthtime.tv_nsec = VNOVAL;
605 	vap->va_flags = VNOVAL;
606 	vap->va_gen = VNOVAL;
607 	vap->va_vaflags = 0;
608 }
609 
610 /*
611  * This routine is called when we have too many vnodes.  It attempts
612  * to free <count> vnodes and will potentially free vnodes that still
613  * have VM backing store (VM backing store is typically the cause
614  * of a vnode blowout so we want to do this).  Therefore, this operation
615  * is not considered cheap.
616  *
617  * A number of conditions may prevent a vnode from being reclaimed.
618  * the buffer cache may have references on the vnode, a directory
619  * vnode may still have references due to the namei cache representing
620  * underlying files, or the vnode may be in active use.   It is not
621  * desireable to reuse such vnodes.  These conditions may cause the
622  * number of vnodes to reach some minimum value regardless of what
623  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
624  */
625 static int
626 vlrureclaim(struct mount *mp)
627 {
628 	struct vnode *vp;
629 	int done;
630 	int trigger;
631 	int usevnodes;
632 	int count;
633 
634 	/*
635 	 * Calculate the trigger point, don't allow user
636 	 * screwups to blow us up.   This prevents us from
637 	 * recycling vnodes with lots of resident pages.  We
638 	 * aren't trying to free memory, we are trying to
639 	 * free vnodes.
640 	 */
641 	usevnodes = desiredvnodes;
642 	if (usevnodes <= 0)
643 		usevnodes = 1;
644 	trigger = cnt.v_page_count * 2 / usevnodes;
645 	done = 0;
646 	vn_start_write(NULL, &mp, V_WAIT);
647 	MNT_ILOCK(mp);
648 	count = mp->mnt_nvnodelistsize / 10 + 1;
649 	while (count != 0) {
650 		vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
651 		while (vp != NULL && vp->v_type == VMARKER)
652 			vp = TAILQ_NEXT(vp, v_nmntvnodes);
653 		if (vp == NULL)
654 			break;
655 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
656 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
657 		--count;
658 		if (!VI_TRYLOCK(vp))
659 			goto next_iter;
660 		/*
661 		 * If it's been deconstructed already, it's still
662 		 * referenced, or it exceeds the trigger, skip it.
663 		 */
664 		if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) ||
665 		    (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
666 		    vp->v_object->resident_page_count > trigger)) {
667 			VI_UNLOCK(vp);
668 			goto next_iter;
669 		}
670 		MNT_IUNLOCK(mp);
671 		vholdl(vp);
672 		if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
673 			vdrop(vp);
674 			goto next_iter_mntunlocked;
675 		}
676 		VI_LOCK(vp);
677 		/*
678 		 * v_usecount may have been bumped after VOP_LOCK() dropped
679 		 * the vnode interlock and before it was locked again.
680 		 *
681 		 * It is not necessary to recheck VI_DOOMED because it can
682 		 * only be set by another thread that holds both the vnode
683 		 * lock and vnode interlock.  If another thread has the
684 		 * vnode lock before we get to VOP_LOCK() and obtains the
685 		 * vnode interlock after VOP_LOCK() drops the vnode
686 		 * interlock, the other thread will be unable to drop the
687 		 * vnode lock before our VOP_LOCK() call fails.
688 		 */
689 		if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) ||
690 		    (vp->v_object != NULL &&
691 		    vp->v_object->resident_page_count > trigger)) {
692 			VOP_UNLOCK(vp, LK_INTERLOCK);
693 			goto next_iter_mntunlocked;
694 		}
695 		KASSERT((vp->v_iflag & VI_DOOMED) == 0,
696 		    ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
697 		vgonel(vp);
698 		VOP_UNLOCK(vp, 0);
699 		vdropl(vp);
700 		done++;
701 next_iter_mntunlocked:
702 		if ((count % 256) != 0)
703 			goto relock_mnt;
704 		goto yield;
705 next_iter:
706 		if ((count % 256) != 0)
707 			continue;
708 		MNT_IUNLOCK(mp);
709 yield:
710 		uio_yield();
711 relock_mnt:
712 		MNT_ILOCK(mp);
713 	}
714 	MNT_IUNLOCK(mp);
715 	vn_finished_write(mp);
716 	return done;
717 }
718 
719 /*
720  * Attempt to keep the free list at wantfreevnodes length.
721  */
722 static void
723 vnlru_free(int count)
724 {
725 	struct vnode *vp;
726 	int vfslocked;
727 
728 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
729 	for (; count > 0; count--) {
730 		vp = TAILQ_FIRST(&vnode_free_list);
731 		/*
732 		 * The list can be modified while the free_list_mtx
733 		 * has been dropped and vp could be NULL here.
734 		 */
735 		if (!vp)
736 			break;
737 		VNASSERT(vp->v_op != NULL, vp,
738 		    ("vnlru_free: vnode already reclaimed."));
739 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
740 		/*
741 		 * Don't recycle if we can't get the interlock.
742 		 */
743 		if (!VI_TRYLOCK(vp)) {
744 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
745 			continue;
746 		}
747 		VNASSERT(VCANRECYCLE(vp), vp,
748 		    ("vp inconsistent on freelist"));
749 		freevnodes--;
750 		vp->v_iflag &= ~VI_FREE;
751 		vholdl(vp);
752 		mtx_unlock(&vnode_free_list_mtx);
753 		VI_UNLOCK(vp);
754 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
755 		vtryrecycle(vp);
756 		VFS_UNLOCK_GIANT(vfslocked);
757 		/*
758 		 * If the recycled succeeded this vdrop will actually free
759 		 * the vnode.  If not it will simply place it back on
760 		 * the free list.
761 		 */
762 		vdrop(vp);
763 		mtx_lock(&vnode_free_list_mtx);
764 	}
765 }
766 /*
767  * Attempt to recycle vnodes in a context that is always safe to block.
768  * Calling vlrurecycle() from the bowels of filesystem code has some
769  * interesting deadlock problems.
770  */
771 static struct proc *vnlruproc;
772 static int vnlruproc_sig;
773 
774 static void
775 vnlru_proc(void)
776 {
777 	struct mount *mp, *nmp;
778 	int done, vfslocked;
779 	struct proc *p = vnlruproc;
780 
781 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
782 	    SHUTDOWN_PRI_FIRST);
783 
784 	for (;;) {
785 		kproc_suspend_check(p);
786 		mtx_lock(&vnode_free_list_mtx);
787 		if (freevnodes > wantfreevnodes)
788 			vnlru_free(freevnodes - wantfreevnodes);
789 		if (numvnodes <= desiredvnodes * 9 / 10) {
790 			vnlruproc_sig = 0;
791 			wakeup(&vnlruproc_sig);
792 			msleep(vnlruproc, &vnode_free_list_mtx,
793 			    PVFS|PDROP, "vlruwt", hz);
794 			continue;
795 		}
796 		mtx_unlock(&vnode_free_list_mtx);
797 		done = 0;
798 		mtx_lock(&mountlist_mtx);
799 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
800 			if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
801 				nmp = TAILQ_NEXT(mp, mnt_list);
802 				continue;
803 			}
804 			vfslocked = VFS_LOCK_GIANT(mp);
805 			done += vlrureclaim(mp);
806 			VFS_UNLOCK_GIANT(vfslocked);
807 			mtx_lock(&mountlist_mtx);
808 			nmp = TAILQ_NEXT(mp, mnt_list);
809 			vfs_unbusy(mp);
810 		}
811 		mtx_unlock(&mountlist_mtx);
812 		if (done == 0) {
813 			EVENTHANDLER_INVOKE(vfs_lowvnodes, desiredvnodes / 10);
814 #if 0
815 			/* These messages are temporary debugging aids */
816 			if (vnlru_nowhere < 5)
817 				printf("vnlru process getting nowhere..\n");
818 			else if (vnlru_nowhere == 5)
819 				printf("vnlru process messages stopped.\n");
820 #endif
821 			vnlru_nowhere++;
822 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
823 		} else
824 			uio_yield();
825 	}
826 }
827 
828 static struct kproc_desc vnlru_kp = {
829 	"vnlru",
830 	vnlru_proc,
831 	&vnlruproc
832 };
833 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
834     &vnlru_kp);
835 
836 /*
837  * Routines having to do with the management of the vnode table.
838  */
839 
840 void
841 vdestroy(struct vnode *vp)
842 {
843 	struct bufobj *bo;
844 
845 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
846 	mtx_lock(&vnode_free_list_mtx);
847 	numvnodes--;
848 	mtx_unlock(&vnode_free_list_mtx);
849 	bo = &vp->v_bufobj;
850 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
851 	    ("cleaned vnode still on the free list."));
852 	VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
853 	VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
854 	VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
855 	VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
856 	VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
857 	VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
858 	VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
859 	VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
860 	VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
861 	VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
862 	VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
863 	VI_UNLOCK(vp);
864 #ifdef MAC
865 	mac_vnode_destroy(vp);
866 #endif
867 	if (vp->v_pollinfo != NULL)
868 		destroy_vpollinfo(vp->v_pollinfo);
869 #ifdef INVARIANTS
870 	/* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */
871 	vp->v_op = NULL;
872 #endif
873 	lockdestroy(vp->v_vnlock);
874 	mtx_destroy(&vp->v_interlock);
875 	mtx_destroy(BO_MTX(bo));
876 	uma_zfree(vnode_zone, vp);
877 }
878 
879 /*
880  * Try to recycle a freed vnode.  We abort if anyone picks up a reference
881  * before we actually vgone().  This function must be called with the vnode
882  * held to prevent the vnode from being returned to the free list midway
883  * through vgone().
884  */
885 static int
886 vtryrecycle(struct vnode *vp)
887 {
888 	struct mount *vnmp;
889 
890 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
891 	VNASSERT(vp->v_holdcnt, vp,
892 	    ("vtryrecycle: Recycling vp %p without a reference.", vp));
893 	/*
894 	 * This vnode may found and locked via some other list, if so we
895 	 * can't recycle it yet.
896 	 */
897 	if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
898 		CTR2(KTR_VFS,
899 		    "%s: impossible to recycle, vp %p lock is already held",
900 		    __func__, vp);
901 		return (EWOULDBLOCK);
902 	}
903 	/*
904 	 * Don't recycle if its filesystem is being suspended.
905 	 */
906 	if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
907 		VOP_UNLOCK(vp, 0);
908 		CTR2(KTR_VFS,
909 		    "%s: impossible to recycle, cannot start the write for %p",
910 		    __func__, vp);
911 		return (EBUSY);
912 	}
913 	/*
914 	 * If we got this far, we need to acquire the interlock and see if
915 	 * anyone picked up this vnode from another list.  If not, we will
916 	 * mark it with DOOMED via vgonel() so that anyone who does find it
917 	 * will skip over it.
918 	 */
919 	VI_LOCK(vp);
920 	if (vp->v_usecount) {
921 		VOP_UNLOCK(vp, LK_INTERLOCK);
922 		vn_finished_write(vnmp);
923 		CTR2(KTR_VFS,
924 		    "%s: impossible to recycle, %p is already referenced",
925 		    __func__, vp);
926 		return (EBUSY);
927 	}
928 	if ((vp->v_iflag & VI_DOOMED) == 0)
929 		vgonel(vp);
930 	VOP_UNLOCK(vp, LK_INTERLOCK);
931 	vn_finished_write(vnmp);
932 	return (0);
933 }
934 
935 /*
936  * Return the next vnode from the free list.
937  */
938 int
939 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
940     struct vnode **vpp)
941 {
942 	struct vnode *vp = NULL;
943 	struct bufobj *bo;
944 
945 	CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
946 	mtx_lock(&vnode_free_list_mtx);
947 	/*
948 	 * Lend our context to reclaim vnodes if they've exceeded the max.
949 	 */
950 	if (freevnodes > wantfreevnodes)
951 		vnlru_free(1);
952 	/*
953 	 * Wait for available vnodes.
954 	 */
955 	if (numvnodes > desiredvnodes) {
956 		if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) {
957 			/*
958 			 * File system is beeing suspended, we cannot risk a
959 			 * deadlock here, so allocate new vnode anyway.
960 			 */
961 			if (freevnodes > wantfreevnodes)
962 				vnlru_free(freevnodes - wantfreevnodes);
963 			goto alloc;
964 		}
965 		if (vnlruproc_sig == 0) {
966 			vnlruproc_sig = 1;	/* avoid unnecessary wakeups */
967 			wakeup(vnlruproc);
968 		}
969 		msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
970 		    "vlruwk", hz);
971 #if 0	/* XXX Not all VFS_VGET/ffs_vget callers check returns. */
972 		if (numvnodes > desiredvnodes) {
973 			mtx_unlock(&vnode_free_list_mtx);
974 			return (ENFILE);
975 		}
976 #endif
977 	}
978 alloc:
979 	numvnodes++;
980 	mtx_unlock(&vnode_free_list_mtx);
981 	vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
982 	/*
983 	 * Setup locks.
984 	 */
985 	vp->v_vnlock = &vp->v_lock;
986 	mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
987 	/*
988 	 * By default, don't allow shared locks unless filesystems
989 	 * opt-in.
990 	 */
991 	lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
992 	/*
993 	 * Initialize bufobj.
994 	 */
995 	bo = &vp->v_bufobj;
996 	bo->__bo_vnode = vp;
997 	mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
998 	bo->bo_ops = &buf_ops_bio;
999 	bo->bo_private = vp;
1000 	TAILQ_INIT(&bo->bo_clean.bv_hd);
1001 	TAILQ_INIT(&bo->bo_dirty.bv_hd);
1002 	/*
1003 	 * Initialize namecache.
1004 	 */
1005 	LIST_INIT(&vp->v_cache_src);
1006 	TAILQ_INIT(&vp->v_cache_dst);
1007 	/*
1008 	 * Finalize various vnode identity bits.
1009 	 */
1010 	vp->v_type = VNON;
1011 	vp->v_tag = tag;
1012 	vp->v_op = vops;
1013 	v_incr_usecount(vp);
1014 	vp->v_data = 0;
1015 #ifdef MAC
1016 	mac_vnode_init(vp);
1017 	if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1018 		mac_vnode_associate_singlelabel(mp, vp);
1019 	else if (mp == NULL && vops != &dead_vnodeops)
1020 		printf("NULL mp in getnewvnode()\n");
1021 #endif
1022 	if (mp != NULL) {
1023 		bo->bo_bsize = mp->mnt_stat.f_iosize;
1024 		if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1025 			vp->v_vflag |= VV_NOKNOTE;
1026 	}
1027 
1028 	*vpp = vp;
1029 	return (0);
1030 }
1031 
1032 /*
1033  * Delete from old mount point vnode list, if on one.
1034  */
1035 static void
1036 delmntque(struct vnode *vp)
1037 {
1038 	struct mount *mp;
1039 
1040 	mp = vp->v_mount;
1041 	if (mp == NULL)
1042 		return;
1043 	MNT_ILOCK(mp);
1044 	vp->v_mount = NULL;
1045 	VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1046 		("bad mount point vnode list size"));
1047 	TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1048 	mp->mnt_nvnodelistsize--;
1049 	MNT_REL(mp);
1050 	MNT_IUNLOCK(mp);
1051 }
1052 
1053 static void
1054 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1055 {
1056 
1057 	vp->v_data = NULL;
1058 	vp->v_op = &dead_vnodeops;
1059 	/* XXX non mp-safe fs may still call insmntque with vnode
1060 	   unlocked */
1061 	if (!VOP_ISLOCKED(vp))
1062 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1063 	vgone(vp);
1064 	vput(vp);
1065 }
1066 
1067 /*
1068  * Insert into list of vnodes for the new mount point, if available.
1069  */
1070 int
1071 insmntque1(struct vnode *vp, struct mount *mp,
1072 	void (*dtr)(struct vnode *, void *), void *dtr_arg)
1073 {
1074 	int locked;
1075 
1076 	KASSERT(vp->v_mount == NULL,
1077 		("insmntque: vnode already on per mount vnode list"));
1078 	VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1079 #ifdef DEBUG_VFS_LOCKS
1080 	if (!VFS_NEEDSGIANT(mp))
1081 		ASSERT_VOP_ELOCKED(vp,
1082 		    "insmntque: mp-safe fs and non-locked vp");
1083 #endif
1084 	MNT_ILOCK(mp);
1085 	if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1086 	    ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1087 	     mp->mnt_nvnodelistsize == 0)) {
1088 		locked = VOP_ISLOCKED(vp);
1089 		if (!locked || (locked == LK_EXCLUSIVE &&
1090 		     (vp->v_vflag & VV_FORCEINSMQ) == 0)) {
1091 			MNT_IUNLOCK(mp);
1092 			if (dtr != NULL)
1093 				dtr(vp, dtr_arg);
1094 			return (EBUSY);
1095 		}
1096 	}
1097 	vp->v_mount = mp;
1098 	MNT_REF(mp);
1099 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1100 	VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1101 		("neg mount point vnode list size"));
1102 	mp->mnt_nvnodelistsize++;
1103 	MNT_IUNLOCK(mp);
1104 	return (0);
1105 }
1106 
1107 int
1108 insmntque(struct vnode *vp, struct mount *mp)
1109 {
1110 
1111 	return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1112 }
1113 
1114 /*
1115  * Flush out and invalidate all buffers associated with a bufobj
1116  * Called with the underlying object locked.
1117  */
1118 int
1119 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1120 {
1121 	int error;
1122 
1123 	BO_LOCK(bo);
1124 	if (flags & V_SAVE) {
1125 		error = bufobj_wwait(bo, slpflag, slptimeo);
1126 		if (error) {
1127 			BO_UNLOCK(bo);
1128 			return (error);
1129 		}
1130 		if (bo->bo_dirty.bv_cnt > 0) {
1131 			BO_UNLOCK(bo);
1132 			if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1133 				return (error);
1134 			/*
1135 			 * XXX We could save a lock/unlock if this was only
1136 			 * enabled under INVARIANTS
1137 			 */
1138 			BO_LOCK(bo);
1139 			if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1140 				panic("vinvalbuf: dirty bufs");
1141 		}
1142 	}
1143 	/*
1144 	 * If you alter this loop please notice that interlock is dropped and
1145 	 * reacquired in flushbuflist.  Special care is needed to ensure that
1146 	 * no race conditions occur from this.
1147 	 */
1148 	do {
1149 		error = flushbuflist(&bo->bo_clean,
1150 		    flags, bo, slpflag, slptimeo);
1151 		if (error == 0)
1152 			error = flushbuflist(&bo->bo_dirty,
1153 			    flags, bo, slpflag, slptimeo);
1154 		if (error != 0 && error != EAGAIN) {
1155 			BO_UNLOCK(bo);
1156 			return (error);
1157 		}
1158 	} while (error != 0);
1159 
1160 	/*
1161 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1162 	 * have write I/O in-progress but if there is a VM object then the
1163 	 * VM object can also have read-I/O in-progress.
1164 	 */
1165 	do {
1166 		bufobj_wwait(bo, 0, 0);
1167 		BO_UNLOCK(bo);
1168 		if (bo->bo_object != NULL) {
1169 			VM_OBJECT_LOCK(bo->bo_object);
1170 			vm_object_pip_wait(bo->bo_object, "bovlbx");
1171 			VM_OBJECT_UNLOCK(bo->bo_object);
1172 		}
1173 		BO_LOCK(bo);
1174 	} while (bo->bo_numoutput > 0);
1175 	BO_UNLOCK(bo);
1176 
1177 	/*
1178 	 * Destroy the copy in the VM cache, too.
1179 	 */
1180 	if (bo->bo_object != NULL && (flags & (V_ALT | V_NORMAL)) == 0) {
1181 		VM_OBJECT_LOCK(bo->bo_object);
1182 		vm_object_page_remove(bo->bo_object, 0, 0,
1183 			(flags & V_SAVE) ? TRUE : FALSE);
1184 		VM_OBJECT_UNLOCK(bo->bo_object);
1185 	}
1186 
1187 #ifdef INVARIANTS
1188 	BO_LOCK(bo);
1189 	if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1190 	    (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1191 		panic("vinvalbuf: flush failed");
1192 	BO_UNLOCK(bo);
1193 #endif
1194 	return (0);
1195 }
1196 
1197 /*
1198  * Flush out and invalidate all buffers associated with a vnode.
1199  * Called with the underlying object locked.
1200  */
1201 int
1202 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1203 {
1204 
1205 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1206 	ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1207 	return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1208 }
1209 
1210 /*
1211  * Flush out buffers on the specified list.
1212  *
1213  */
1214 static int
1215 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1216     int slptimeo)
1217 {
1218 	struct buf *bp, *nbp;
1219 	int retval, error;
1220 	daddr_t lblkno;
1221 	b_xflags_t xflags;
1222 
1223 	ASSERT_BO_LOCKED(bo);
1224 
1225 	retval = 0;
1226 	TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1227 		if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1228 		    ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1229 			continue;
1230 		}
1231 		lblkno = 0;
1232 		xflags = 0;
1233 		if (nbp != NULL) {
1234 			lblkno = nbp->b_lblkno;
1235 			xflags = nbp->b_xflags &
1236 				(BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1237 		}
1238 		retval = EAGAIN;
1239 		error = BUF_TIMELOCK(bp,
1240 		    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1241 		    "flushbuf", slpflag, slptimeo);
1242 		if (error) {
1243 			BO_LOCK(bo);
1244 			return (error != ENOLCK ? error : EAGAIN);
1245 		}
1246 		KASSERT(bp->b_bufobj == bo,
1247 		    ("bp %p wrong b_bufobj %p should be %p",
1248 		    bp, bp->b_bufobj, bo));
1249 		if (bp->b_bufobj != bo) {	/* XXX: necessary ? */
1250 			BUF_UNLOCK(bp);
1251 			BO_LOCK(bo);
1252 			return (EAGAIN);
1253 		}
1254 		/*
1255 		 * XXX Since there are no node locks for NFS, I
1256 		 * believe there is a slight chance that a delayed
1257 		 * write will occur while sleeping just above, so
1258 		 * check for it.
1259 		 */
1260 		if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1261 		    (flags & V_SAVE)) {
1262 			bremfree(bp);
1263 			bp->b_flags |= B_ASYNC;
1264 			bwrite(bp);
1265 			BO_LOCK(bo);
1266 			return (EAGAIN);	/* XXX: why not loop ? */
1267 		}
1268 		bremfree(bp);
1269 		bp->b_flags |= (B_INVAL | B_RELBUF);
1270 		bp->b_flags &= ~B_ASYNC;
1271 		brelse(bp);
1272 		BO_LOCK(bo);
1273 		if (nbp != NULL &&
1274 		    (nbp->b_bufobj != bo ||
1275 		     nbp->b_lblkno != lblkno ||
1276 		     (nbp->b_xflags &
1277 		      (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1278 			break;			/* nbp invalid */
1279 	}
1280 	return (retval);
1281 }
1282 
1283 /*
1284  * Truncate a file's buffer and pages to a specified length.  This
1285  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1286  * sync activity.
1287  */
1288 int
1289 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
1290     off_t length, int blksize)
1291 {
1292 	struct buf *bp, *nbp;
1293 	int anyfreed;
1294 	int trunclbn;
1295 	struct bufobj *bo;
1296 
1297 	CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1298 	    vp, cred, blksize, (uintmax_t)length);
1299 
1300 	/*
1301 	 * Round up to the *next* lbn.
1302 	 */
1303 	trunclbn = (length + blksize - 1) / blksize;
1304 
1305 	ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1306 restart:
1307 	bo = &vp->v_bufobj;
1308 	BO_LOCK(bo);
1309 	anyfreed = 1;
1310 	for (;anyfreed;) {
1311 		anyfreed = 0;
1312 		TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1313 			if (bp->b_lblkno < trunclbn)
1314 				continue;
1315 			if (BUF_LOCK(bp,
1316 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1317 			    BO_MTX(bo)) == ENOLCK)
1318 				goto restart;
1319 
1320 			bremfree(bp);
1321 			bp->b_flags |= (B_INVAL | B_RELBUF);
1322 			bp->b_flags &= ~B_ASYNC;
1323 			brelse(bp);
1324 			anyfreed = 1;
1325 
1326 			if (nbp != NULL &&
1327 			    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1328 			    (nbp->b_vp != vp) ||
1329 			    (nbp->b_flags & B_DELWRI))) {
1330 				goto restart;
1331 			}
1332 			BO_LOCK(bo);
1333 		}
1334 
1335 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1336 			if (bp->b_lblkno < trunclbn)
1337 				continue;
1338 			if (BUF_LOCK(bp,
1339 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1340 			    BO_MTX(bo)) == ENOLCK)
1341 				goto restart;
1342 			bremfree(bp);
1343 			bp->b_flags |= (B_INVAL | B_RELBUF);
1344 			bp->b_flags &= ~B_ASYNC;
1345 			brelse(bp);
1346 			anyfreed = 1;
1347 			if (nbp != NULL &&
1348 			    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1349 			    (nbp->b_vp != vp) ||
1350 			    (nbp->b_flags & B_DELWRI) == 0)) {
1351 				goto restart;
1352 			}
1353 			BO_LOCK(bo);
1354 		}
1355 	}
1356 
1357 	if (length > 0) {
1358 restartsync:
1359 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1360 			if (bp->b_lblkno > 0)
1361 				continue;
1362 			/*
1363 			 * Since we hold the vnode lock this should only
1364 			 * fail if we're racing with the buf daemon.
1365 			 */
1366 			if (BUF_LOCK(bp,
1367 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1368 			    BO_MTX(bo)) == ENOLCK) {
1369 				goto restart;
1370 			}
1371 			VNASSERT((bp->b_flags & B_DELWRI), vp,
1372 			    ("buf(%p) on dirty queue without DELWRI", bp));
1373 
1374 			bremfree(bp);
1375 			bawrite(bp);
1376 			BO_LOCK(bo);
1377 			goto restartsync;
1378 		}
1379 	}
1380 
1381 	bufobj_wwait(bo, 0, 0);
1382 	BO_UNLOCK(bo);
1383 	vnode_pager_setsize(vp, length);
1384 
1385 	return (0);
1386 }
1387 
1388 /*
1389  * buf_splay() - splay tree core for the clean/dirty list of buffers in
1390  * 		 a vnode.
1391  *
1392  *	NOTE: We have to deal with the special case of a background bitmap
1393  *	buffer, a situation where two buffers will have the same logical
1394  *	block offset.  We want (1) only the foreground buffer to be accessed
1395  *	in a lookup and (2) must differentiate between the foreground and
1396  *	background buffer in the splay tree algorithm because the splay
1397  *	tree cannot normally handle multiple entities with the same 'index'.
1398  *	We accomplish this by adding differentiating flags to the splay tree's
1399  *	numerical domain.
1400  */
1401 static
1402 struct buf *
1403 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1404 {
1405 	struct buf dummy;
1406 	struct buf *lefttreemax, *righttreemin, *y;
1407 
1408 	if (root == NULL)
1409 		return (NULL);
1410 	lefttreemax = righttreemin = &dummy;
1411 	for (;;) {
1412 		if (lblkno < root->b_lblkno ||
1413 		    (lblkno == root->b_lblkno &&
1414 		    (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1415 			if ((y = root->b_left) == NULL)
1416 				break;
1417 			if (lblkno < y->b_lblkno) {
1418 				/* Rotate right. */
1419 				root->b_left = y->b_right;
1420 				y->b_right = root;
1421 				root = y;
1422 				if ((y = root->b_left) == NULL)
1423 					break;
1424 			}
1425 			/* Link into the new root's right tree. */
1426 			righttreemin->b_left = root;
1427 			righttreemin = root;
1428 		} else if (lblkno > root->b_lblkno ||
1429 		    (lblkno == root->b_lblkno &&
1430 		    (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1431 			if ((y = root->b_right) == NULL)
1432 				break;
1433 			if (lblkno > y->b_lblkno) {
1434 				/* Rotate left. */
1435 				root->b_right = y->b_left;
1436 				y->b_left = root;
1437 				root = y;
1438 				if ((y = root->b_right) == NULL)
1439 					break;
1440 			}
1441 			/* Link into the new root's left tree. */
1442 			lefttreemax->b_right = root;
1443 			lefttreemax = root;
1444 		} else {
1445 			break;
1446 		}
1447 		root = y;
1448 	}
1449 	/* Assemble the new root. */
1450 	lefttreemax->b_right = root->b_left;
1451 	righttreemin->b_left = root->b_right;
1452 	root->b_left = dummy.b_right;
1453 	root->b_right = dummy.b_left;
1454 	return (root);
1455 }
1456 
1457 static void
1458 buf_vlist_remove(struct buf *bp)
1459 {
1460 	struct buf *root;
1461 	struct bufv *bv;
1462 
1463 	KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1464 	ASSERT_BO_LOCKED(bp->b_bufobj);
1465 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1466 	    (BX_VNDIRTY|BX_VNCLEAN),
1467 	    ("buf_vlist_remove: Buf %p is on two lists", bp));
1468 	if (bp->b_xflags & BX_VNDIRTY)
1469 		bv = &bp->b_bufobj->bo_dirty;
1470 	else
1471 		bv = &bp->b_bufobj->bo_clean;
1472 	if (bp != bv->bv_root) {
1473 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1474 		KASSERT(root == bp, ("splay lookup failed in remove"));
1475 	}
1476 	if (bp->b_left == NULL) {
1477 		root = bp->b_right;
1478 	} else {
1479 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1480 		root->b_right = bp->b_right;
1481 	}
1482 	bv->bv_root = root;
1483 	TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1484 	bv->bv_cnt--;
1485 	bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1486 }
1487 
1488 /*
1489  * Add the buffer to the sorted clean or dirty block list using a
1490  * splay tree algorithm.
1491  *
1492  * NOTE: xflags is passed as a constant, optimizing this inline function!
1493  */
1494 static void
1495 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1496 {
1497 	struct buf *root;
1498 	struct bufv *bv;
1499 
1500 	ASSERT_BO_LOCKED(bo);
1501 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1502 	    ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1503 	bp->b_xflags |= xflags;
1504 	if (xflags & BX_VNDIRTY)
1505 		bv = &bo->bo_dirty;
1506 	else
1507 		bv = &bo->bo_clean;
1508 
1509 	root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1510 	if (root == NULL) {
1511 		bp->b_left = NULL;
1512 		bp->b_right = NULL;
1513 		TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1514 	} else if (bp->b_lblkno < root->b_lblkno ||
1515 	    (bp->b_lblkno == root->b_lblkno &&
1516 	    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1517 		bp->b_left = root->b_left;
1518 		bp->b_right = root;
1519 		root->b_left = NULL;
1520 		TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1521 	} else {
1522 		bp->b_right = root->b_right;
1523 		bp->b_left = root;
1524 		root->b_right = NULL;
1525 		TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1526 	}
1527 	bv->bv_cnt++;
1528 	bv->bv_root = bp;
1529 }
1530 
1531 /*
1532  * Lookup a buffer using the splay tree.  Note that we specifically avoid
1533  * shadow buffers used in background bitmap writes.
1534  *
1535  * This code isn't quite efficient as it could be because we are maintaining
1536  * two sorted lists and do not know which list the block resides in.
1537  *
1538  * During a "make buildworld" the desired buffer is found at one of
1539  * the roots more than 60% of the time.  Thus, checking both roots
1540  * before performing either splay eliminates unnecessary splays on the
1541  * first tree splayed.
1542  */
1543 struct buf *
1544 gbincore(struct bufobj *bo, daddr_t lblkno)
1545 {
1546 	struct buf *bp;
1547 
1548 	ASSERT_BO_LOCKED(bo);
1549 	if ((bp = bo->bo_clean.bv_root) != NULL &&
1550 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1551 		return (bp);
1552 	if ((bp = bo->bo_dirty.bv_root) != NULL &&
1553 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1554 		return (bp);
1555 	if ((bp = bo->bo_clean.bv_root) != NULL) {
1556 		bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1557 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1558 			return (bp);
1559 	}
1560 	if ((bp = bo->bo_dirty.bv_root) != NULL) {
1561 		bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1562 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1563 			return (bp);
1564 	}
1565 	return (NULL);
1566 }
1567 
1568 /*
1569  * Associate a buffer with a vnode.
1570  */
1571 void
1572 bgetvp(struct vnode *vp, struct buf *bp)
1573 {
1574 	struct bufobj *bo;
1575 
1576 	bo = &vp->v_bufobj;
1577 	ASSERT_BO_LOCKED(bo);
1578 	VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1579 
1580 	CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1581 	VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1582 	    ("bgetvp: bp already attached! %p", bp));
1583 
1584 	vhold(vp);
1585 	if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT)
1586 		bp->b_flags |= B_NEEDSGIANT;
1587 	bp->b_vp = vp;
1588 	bp->b_bufobj = bo;
1589 	/*
1590 	 * Insert onto list for new vnode.
1591 	 */
1592 	buf_vlist_add(bp, bo, BX_VNCLEAN);
1593 }
1594 
1595 /*
1596  * Disassociate a buffer from a vnode.
1597  */
1598 void
1599 brelvp(struct buf *bp)
1600 {
1601 	struct bufobj *bo;
1602 	struct vnode *vp;
1603 
1604 	CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1605 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1606 
1607 	/*
1608 	 * Delete from old vnode list, if on one.
1609 	 */
1610 	vp = bp->b_vp;		/* XXX */
1611 	bo = bp->b_bufobj;
1612 	BO_LOCK(bo);
1613 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1614 		buf_vlist_remove(bp);
1615 	else
1616 		panic("brelvp: Buffer %p not on queue.", bp);
1617 	if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1618 		bo->bo_flag &= ~BO_ONWORKLST;
1619 		mtx_lock(&sync_mtx);
1620 		LIST_REMOVE(bo, bo_synclist);
1621 		syncer_worklist_len--;
1622 		mtx_unlock(&sync_mtx);
1623 	}
1624 	bp->b_flags &= ~B_NEEDSGIANT;
1625 	bp->b_vp = NULL;
1626 	bp->b_bufobj = NULL;
1627 	BO_UNLOCK(bo);
1628 	vdrop(vp);
1629 }
1630 
1631 /*
1632  * Add an item to the syncer work queue.
1633  */
1634 static void
1635 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1636 {
1637 	int queue, slot;
1638 
1639 	ASSERT_BO_LOCKED(bo);
1640 
1641 	mtx_lock(&sync_mtx);
1642 	if (bo->bo_flag & BO_ONWORKLST)
1643 		LIST_REMOVE(bo, bo_synclist);
1644 	else {
1645 		bo->bo_flag |= BO_ONWORKLST;
1646 		syncer_worklist_len++;
1647 	}
1648 
1649 	if (delay > syncer_maxdelay - 2)
1650 		delay = syncer_maxdelay - 2;
1651 	slot = (syncer_delayno + delay) & syncer_mask;
1652 
1653 	queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ :
1654 	    WI_MPSAFEQ;
1655 	LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo,
1656 	    bo_synclist);
1657 	mtx_unlock(&sync_mtx);
1658 }
1659 
1660 static int
1661 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1662 {
1663 	int error, len;
1664 
1665 	mtx_lock(&sync_mtx);
1666 	len = syncer_worklist_len - sync_vnode_count;
1667 	mtx_unlock(&sync_mtx);
1668 	error = SYSCTL_OUT(req, &len, sizeof(len));
1669 	return (error);
1670 }
1671 
1672 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1673     sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1674 
1675 static struct proc *updateproc;
1676 static void sched_sync(void);
1677 static struct kproc_desc up_kp = {
1678 	"syncer",
1679 	sched_sync,
1680 	&updateproc
1681 };
1682 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1683 
1684 static int
1685 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1686 {
1687 	struct vnode *vp;
1688 	struct mount *mp;
1689 
1690 	*bo = LIST_FIRST(slp);
1691 	if (*bo == NULL)
1692 		return (0);
1693 	vp = (*bo)->__bo_vnode;	/* XXX */
1694 	if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1695 		return (1);
1696 	/*
1697 	 * We use vhold in case the vnode does not
1698 	 * successfully sync.  vhold prevents the vnode from
1699 	 * going away when we unlock the sync_mtx so that
1700 	 * we can acquire the vnode interlock.
1701 	 */
1702 	vholdl(vp);
1703 	mtx_unlock(&sync_mtx);
1704 	VI_UNLOCK(vp);
1705 	if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1706 		vdrop(vp);
1707 		mtx_lock(&sync_mtx);
1708 		return (*bo == LIST_FIRST(slp));
1709 	}
1710 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1711 	(void) VOP_FSYNC(vp, MNT_LAZY, td);
1712 	VOP_UNLOCK(vp, 0);
1713 	vn_finished_write(mp);
1714 	BO_LOCK(*bo);
1715 	if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1716 		/*
1717 		 * Put us back on the worklist.  The worklist
1718 		 * routine will remove us from our current
1719 		 * position and then add us back in at a later
1720 		 * position.
1721 		 */
1722 		vn_syncer_add_to_worklist(*bo, syncdelay);
1723 	}
1724 	BO_UNLOCK(*bo);
1725 	vdrop(vp);
1726 	mtx_lock(&sync_mtx);
1727 	return (0);
1728 }
1729 
1730 /*
1731  * System filesystem synchronizer daemon.
1732  */
1733 static void
1734 sched_sync(void)
1735 {
1736 	struct synclist *gnext, *next;
1737 	struct synclist *gslp, *slp;
1738 	struct bufobj *bo;
1739 	long starttime;
1740 	struct thread *td = curthread;
1741 	int last_work_seen;
1742 	int net_worklist_len;
1743 	int syncer_final_iter;
1744 	int first_printf;
1745 	int error;
1746 
1747 	last_work_seen = 0;
1748 	syncer_final_iter = 0;
1749 	first_printf = 1;
1750 	syncer_state = SYNCER_RUNNING;
1751 	starttime = time_uptime;
1752 	td->td_pflags |= TDP_NORUNNINGBUF;
1753 
1754 	EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1755 	    SHUTDOWN_PRI_LAST);
1756 
1757 	mtx_lock(&sync_mtx);
1758 	for (;;) {
1759 		if (syncer_state == SYNCER_FINAL_DELAY &&
1760 		    syncer_final_iter == 0) {
1761 			mtx_unlock(&sync_mtx);
1762 			kproc_suspend_check(td->td_proc);
1763 			mtx_lock(&sync_mtx);
1764 		}
1765 		net_worklist_len = syncer_worklist_len - sync_vnode_count;
1766 		if (syncer_state != SYNCER_RUNNING &&
1767 		    starttime != time_uptime) {
1768 			if (first_printf) {
1769 				printf("\nSyncing disks, vnodes remaining...");
1770 				first_printf = 0;
1771 			}
1772 			printf("%d ", net_worklist_len);
1773 		}
1774 		starttime = time_uptime;
1775 
1776 		/*
1777 		 * Push files whose dirty time has expired.  Be careful
1778 		 * of interrupt race on slp queue.
1779 		 *
1780 		 * Skip over empty worklist slots when shutting down.
1781 		 */
1782 		do {
1783 			slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1784 			gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1785 			syncer_delayno += 1;
1786 			if (syncer_delayno == syncer_maxdelay)
1787 				syncer_delayno = 0;
1788 			next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1789 			gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1790 			/*
1791 			 * If the worklist has wrapped since the
1792 			 * it was emptied of all but syncer vnodes,
1793 			 * switch to the FINAL_DELAY state and run
1794 			 * for one more second.
1795 			 */
1796 			if (syncer_state == SYNCER_SHUTTING_DOWN &&
1797 			    net_worklist_len == 0 &&
1798 			    last_work_seen == syncer_delayno) {
1799 				syncer_state = SYNCER_FINAL_DELAY;
1800 				syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1801 			}
1802 		} while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1803 		    LIST_EMPTY(gslp) && syncer_worklist_len > 0);
1804 
1805 		/*
1806 		 * Keep track of the last time there was anything
1807 		 * on the worklist other than syncer vnodes.
1808 		 * Return to the SHUTTING_DOWN state if any
1809 		 * new work appears.
1810 		 */
1811 		if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1812 			last_work_seen = syncer_delayno;
1813 		if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1814 			syncer_state = SYNCER_SHUTTING_DOWN;
1815 		while (!LIST_EMPTY(slp)) {
1816 			error = sync_vnode(slp, &bo, td);
1817 			if (error == 1) {
1818 				LIST_REMOVE(bo, bo_synclist);
1819 				LIST_INSERT_HEAD(next, bo, bo_synclist);
1820 				continue;
1821 			}
1822 		}
1823 		if (!LIST_EMPTY(gslp)) {
1824 			mtx_unlock(&sync_mtx);
1825 			mtx_lock(&Giant);
1826 			mtx_lock(&sync_mtx);
1827 			while (!LIST_EMPTY(gslp)) {
1828 				error = sync_vnode(gslp, &bo, td);
1829 				if (error == 1) {
1830 					LIST_REMOVE(bo, bo_synclist);
1831 					LIST_INSERT_HEAD(gnext, bo,
1832 					    bo_synclist);
1833 					continue;
1834 				}
1835 			}
1836 			mtx_unlock(&Giant);
1837 		}
1838 		if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1839 			syncer_final_iter--;
1840 		/*
1841 		 * The variable rushjob allows the kernel to speed up the
1842 		 * processing of the filesystem syncer process. A rushjob
1843 		 * value of N tells the filesystem syncer to process the next
1844 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1845 		 * is used by the soft update code to speed up the filesystem
1846 		 * syncer process when the incore state is getting so far
1847 		 * ahead of the disk that the kernel memory pool is being
1848 		 * threatened with exhaustion.
1849 		 */
1850 		if (rushjob > 0) {
1851 			rushjob -= 1;
1852 			continue;
1853 		}
1854 		/*
1855 		 * Just sleep for a short period of time between
1856 		 * iterations when shutting down to allow some I/O
1857 		 * to happen.
1858 		 *
1859 		 * If it has taken us less than a second to process the
1860 		 * current work, then wait. Otherwise start right over
1861 		 * again. We can still lose time if any single round
1862 		 * takes more than two seconds, but it does not really
1863 		 * matter as we are just trying to generally pace the
1864 		 * filesystem activity.
1865 		 */
1866 		if (syncer_state != SYNCER_RUNNING)
1867 			cv_timedwait(&sync_wakeup, &sync_mtx,
1868 			    hz / SYNCER_SHUTDOWN_SPEEDUP);
1869 		else if (time_uptime == starttime)
1870 			cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1871 	}
1872 }
1873 
1874 /*
1875  * Request the syncer daemon to speed up its work.
1876  * We never push it to speed up more than half of its
1877  * normal turn time, otherwise it could take over the cpu.
1878  */
1879 int
1880 speedup_syncer(void)
1881 {
1882 	int ret = 0;
1883 
1884 	mtx_lock(&sync_mtx);
1885 	if (rushjob < syncdelay / 2) {
1886 		rushjob += 1;
1887 		stat_rush_requests += 1;
1888 		ret = 1;
1889 	}
1890 	mtx_unlock(&sync_mtx);
1891 	cv_broadcast(&sync_wakeup);
1892 	return (ret);
1893 }
1894 
1895 /*
1896  * Tell the syncer to speed up its work and run though its work
1897  * list several times, then tell it to shut down.
1898  */
1899 static void
1900 syncer_shutdown(void *arg, int howto)
1901 {
1902 
1903 	if (howto & RB_NOSYNC)
1904 		return;
1905 	mtx_lock(&sync_mtx);
1906 	syncer_state = SYNCER_SHUTTING_DOWN;
1907 	rushjob = 0;
1908 	mtx_unlock(&sync_mtx);
1909 	cv_broadcast(&sync_wakeup);
1910 	kproc_shutdown(arg, howto);
1911 }
1912 
1913 /*
1914  * Reassign a buffer from one vnode to another.
1915  * Used to assign file specific control information
1916  * (indirect blocks) to the vnode to which they belong.
1917  */
1918 void
1919 reassignbuf(struct buf *bp)
1920 {
1921 	struct vnode *vp;
1922 	struct bufobj *bo;
1923 	int delay;
1924 #ifdef INVARIANTS
1925 	struct bufv *bv;
1926 #endif
1927 
1928 	vp = bp->b_vp;
1929 	bo = bp->b_bufobj;
1930 	++reassignbufcalls;
1931 
1932 	CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1933 	    bp, bp->b_vp, bp->b_flags);
1934 	/*
1935 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1936 	 * is not fully linked in.
1937 	 */
1938 	if (bp->b_flags & B_PAGING)
1939 		panic("cannot reassign paging buffer");
1940 
1941 	/*
1942 	 * Delete from old vnode list, if on one.
1943 	 */
1944 	BO_LOCK(bo);
1945 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1946 		buf_vlist_remove(bp);
1947 	else
1948 		panic("reassignbuf: Buffer %p not on queue.", bp);
1949 	/*
1950 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1951 	 * of clean buffers.
1952 	 */
1953 	if (bp->b_flags & B_DELWRI) {
1954 		if ((bo->bo_flag & BO_ONWORKLST) == 0) {
1955 			switch (vp->v_type) {
1956 			case VDIR:
1957 				delay = dirdelay;
1958 				break;
1959 			case VCHR:
1960 				delay = metadelay;
1961 				break;
1962 			default:
1963 				delay = filedelay;
1964 			}
1965 			vn_syncer_add_to_worklist(bo, delay);
1966 		}
1967 		buf_vlist_add(bp, bo, BX_VNDIRTY);
1968 	} else {
1969 		buf_vlist_add(bp, bo, BX_VNCLEAN);
1970 
1971 		if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1972 			mtx_lock(&sync_mtx);
1973 			LIST_REMOVE(bo, bo_synclist);
1974 			syncer_worklist_len--;
1975 			mtx_unlock(&sync_mtx);
1976 			bo->bo_flag &= ~BO_ONWORKLST;
1977 		}
1978 	}
1979 #ifdef INVARIANTS
1980 	bv = &bo->bo_clean;
1981 	bp = TAILQ_FIRST(&bv->bv_hd);
1982 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1983 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1984 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
1985 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1986 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1987 	bv = &bo->bo_dirty;
1988 	bp = TAILQ_FIRST(&bv->bv_hd);
1989 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1990 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1991 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
1992 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1993 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1994 #endif
1995 	BO_UNLOCK(bo);
1996 }
1997 
1998 /*
1999  * Increment the use and hold counts on the vnode, taking care to reference
2000  * the driver's usecount if this is a chardev.  The vholdl() will remove
2001  * the vnode from the free list if it is presently free.  Requires the
2002  * vnode interlock and returns with it held.
2003  */
2004 static void
2005 v_incr_usecount(struct vnode *vp)
2006 {
2007 
2008 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2009 	vp->v_usecount++;
2010 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2011 		dev_lock();
2012 		vp->v_rdev->si_usecount++;
2013 		dev_unlock();
2014 	}
2015 	vholdl(vp);
2016 }
2017 
2018 /*
2019  * Turn a holdcnt into a use+holdcnt such that only one call to
2020  * v_decr_usecount is needed.
2021  */
2022 static void
2023 v_upgrade_usecount(struct vnode *vp)
2024 {
2025 
2026 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2027 	vp->v_usecount++;
2028 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2029 		dev_lock();
2030 		vp->v_rdev->si_usecount++;
2031 		dev_unlock();
2032 	}
2033 }
2034 
2035 /*
2036  * Decrement the vnode use and hold count along with the driver's usecount
2037  * if this is a chardev.  The vdropl() below releases the vnode interlock
2038  * as it may free the vnode.
2039  */
2040 static void
2041 v_decr_usecount(struct vnode *vp)
2042 {
2043 
2044 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2045 	VNASSERT(vp->v_usecount > 0, vp,
2046 	    ("v_decr_usecount: negative usecount"));
2047 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2048 	vp->v_usecount--;
2049 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2050 		dev_lock();
2051 		vp->v_rdev->si_usecount--;
2052 		dev_unlock();
2053 	}
2054 	vdropl(vp);
2055 }
2056 
2057 /*
2058  * Decrement only the use count and driver use count.  This is intended to
2059  * be paired with a follow on vdropl() to release the remaining hold count.
2060  * In this way we may vgone() a vnode with a 0 usecount without risk of
2061  * having it end up on a free list because the hold count is kept above 0.
2062  */
2063 static void
2064 v_decr_useonly(struct vnode *vp)
2065 {
2066 
2067 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2068 	VNASSERT(vp->v_usecount > 0, vp,
2069 	    ("v_decr_useonly: negative usecount"));
2070 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2071 	vp->v_usecount--;
2072 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2073 		dev_lock();
2074 		vp->v_rdev->si_usecount--;
2075 		dev_unlock();
2076 	}
2077 }
2078 
2079 /*
2080  * Grab a particular vnode from the free list, increment its
2081  * reference count and lock it.  VI_DOOMED is set if the vnode
2082  * is being destroyed.  Only callers who specify LK_RETRY will
2083  * see doomed vnodes.  If inactive processing was delayed in
2084  * vput try to do it here.
2085  */
2086 int
2087 vget(struct vnode *vp, int flags, struct thread *td)
2088 {
2089 	int error;
2090 
2091 	error = 0;
2092 	VFS_ASSERT_GIANT(vp->v_mount);
2093 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2094 	    ("vget: invalid lock operation"));
2095 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2096 
2097 	if ((flags & LK_INTERLOCK) == 0)
2098 		VI_LOCK(vp);
2099 	vholdl(vp);
2100 	if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2101 		vdrop(vp);
2102 		CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2103 		    vp);
2104 		return (error);
2105 	}
2106 	if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2107 		panic("vget: vn_lock failed to return ENOENT\n");
2108 	VI_LOCK(vp);
2109 	/* Upgrade our holdcnt to a usecount. */
2110 	v_upgrade_usecount(vp);
2111 	/*
2112  	 * We don't guarantee that any particular close will
2113 	 * trigger inactive processing so just make a best effort
2114 	 * here at preventing a reference to a removed file.  If
2115 	 * we don't succeed no harm is done.
2116 	 */
2117 	if (vp->v_iflag & VI_OWEINACT) {
2118 		if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2119 		    (flags & LK_NOWAIT) == 0)
2120 			vinactive(vp, td);
2121 		vp->v_iflag &= ~VI_OWEINACT;
2122 	}
2123 	VI_UNLOCK(vp);
2124 	return (0);
2125 }
2126 
2127 /*
2128  * Increase the reference count of a vnode.
2129  */
2130 void
2131 vref(struct vnode *vp)
2132 {
2133 
2134 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2135 	VI_LOCK(vp);
2136 	v_incr_usecount(vp);
2137 	VI_UNLOCK(vp);
2138 }
2139 
2140 /*
2141  * Return reference count of a vnode.
2142  *
2143  * The results of this call are only guaranteed when some mechanism other
2144  * than the VI lock is used to stop other processes from gaining references
2145  * to the vnode.  This may be the case if the caller holds the only reference.
2146  * This is also useful when stale data is acceptable as race conditions may
2147  * be accounted for by some other means.
2148  */
2149 int
2150 vrefcnt(struct vnode *vp)
2151 {
2152 	int usecnt;
2153 
2154 	VI_LOCK(vp);
2155 	usecnt = vp->v_usecount;
2156 	VI_UNLOCK(vp);
2157 
2158 	return (usecnt);
2159 }
2160 
2161 
2162 /*
2163  * Vnode put/release.
2164  * If count drops to zero, call inactive routine and return to freelist.
2165  */
2166 void
2167 vrele(struct vnode *vp)
2168 {
2169 	struct thread *td = curthread;	/* XXX */
2170 
2171 	KASSERT(vp != NULL, ("vrele: null vp"));
2172 	VFS_ASSERT_GIANT(vp->v_mount);
2173 
2174 	VI_LOCK(vp);
2175 
2176 	/* Skip this v_writecount check if we're going to panic below. */
2177 	VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2178 	    ("vrele: missed vn_close"));
2179 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2180 
2181 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2182 	    vp->v_usecount == 1)) {
2183 		v_decr_usecount(vp);
2184 		return;
2185 	}
2186 	if (vp->v_usecount != 1) {
2187 #ifdef DIAGNOSTIC
2188 		vprint("vrele: negative ref count", vp);
2189 #endif
2190 		VI_UNLOCK(vp);
2191 		panic("vrele: negative ref cnt");
2192 	}
2193 	CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2194 	/*
2195 	 * We want to hold the vnode until the inactive finishes to
2196 	 * prevent vgone() races.  We drop the use count here and the
2197 	 * hold count below when we're done.
2198 	 */
2199 	v_decr_useonly(vp);
2200 	/*
2201 	 * We must call VOP_INACTIVE with the node locked. Mark
2202 	 * as VI_DOINGINACT to avoid recursion.
2203 	 */
2204 	vp->v_iflag |= VI_OWEINACT;
2205 	if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK) == 0) {
2206 		VI_LOCK(vp);
2207 		if (vp->v_usecount > 0)
2208 			vp->v_iflag &= ~VI_OWEINACT;
2209 		if (vp->v_iflag & VI_OWEINACT)
2210 			vinactive(vp, td);
2211 		VOP_UNLOCK(vp, 0);
2212 	} else {
2213 		VI_LOCK(vp);
2214 		if (vp->v_usecount > 0)
2215 			vp->v_iflag &= ~VI_OWEINACT;
2216 	}
2217 	vdropl(vp);
2218 }
2219 
2220 /*
2221  * Release an already locked vnode.  This give the same effects as
2222  * unlock+vrele(), but takes less time and avoids releasing and
2223  * re-aquiring the lock (as vrele() acquires the lock internally.)
2224  */
2225 void
2226 vput(struct vnode *vp)
2227 {
2228 	struct thread *td = curthread;	/* XXX */
2229 	int error;
2230 
2231 	KASSERT(vp != NULL, ("vput: null vp"));
2232 	ASSERT_VOP_LOCKED(vp, "vput");
2233 	VFS_ASSERT_GIANT(vp->v_mount);
2234 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2235 	VI_LOCK(vp);
2236 	/* Skip this v_writecount check if we're going to panic below. */
2237 	VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2238 	    ("vput: missed vn_close"));
2239 	error = 0;
2240 
2241 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2242 	    vp->v_usecount == 1)) {
2243 		VOP_UNLOCK(vp, 0);
2244 		v_decr_usecount(vp);
2245 		return;
2246 	}
2247 
2248 	if (vp->v_usecount != 1) {
2249 #ifdef DIAGNOSTIC
2250 		vprint("vput: negative ref count", vp);
2251 #endif
2252 		panic("vput: negative ref cnt");
2253 	}
2254 	CTR2(KTR_VFS, "%s: return to freelist the vnode %p", __func__, vp);
2255 	/*
2256 	 * We want to hold the vnode until the inactive finishes to
2257 	 * prevent vgone() races.  We drop the use count here and the
2258 	 * hold count below when we're done.
2259 	 */
2260 	v_decr_useonly(vp);
2261 	vp->v_iflag |= VI_OWEINACT;
2262 	if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2263 		error = VOP_LOCK(vp, LK_UPGRADE|LK_INTERLOCK|LK_NOWAIT);
2264 		VI_LOCK(vp);
2265 		if (error) {
2266 			if (vp->v_usecount > 0)
2267 				vp->v_iflag &= ~VI_OWEINACT;
2268 			goto done;
2269 		}
2270 	}
2271 	if (vp->v_usecount > 0)
2272 		vp->v_iflag &= ~VI_OWEINACT;
2273 	if (vp->v_iflag & VI_OWEINACT)
2274 		vinactive(vp, td);
2275 	VOP_UNLOCK(vp, 0);
2276 done:
2277 	vdropl(vp);
2278 }
2279 
2280 /*
2281  * Somebody doesn't want the vnode recycled.
2282  */
2283 void
2284 vhold(struct vnode *vp)
2285 {
2286 
2287 	VI_LOCK(vp);
2288 	vholdl(vp);
2289 	VI_UNLOCK(vp);
2290 }
2291 
2292 void
2293 vholdl(struct vnode *vp)
2294 {
2295 
2296 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2297 	vp->v_holdcnt++;
2298 	if (VSHOULDBUSY(vp))
2299 		vbusy(vp);
2300 }
2301 
2302 /*
2303  * Note that there is one less who cares about this vnode.  vdrop() is the
2304  * opposite of vhold().
2305  */
2306 void
2307 vdrop(struct vnode *vp)
2308 {
2309 
2310 	VI_LOCK(vp);
2311 	vdropl(vp);
2312 }
2313 
2314 /*
2315  * Drop the hold count of the vnode.  If this is the last reference to
2316  * the vnode we will free it if it has been vgone'd otherwise it is
2317  * placed on the free list.
2318  */
2319 void
2320 vdropl(struct vnode *vp)
2321 {
2322 
2323 	ASSERT_VI_LOCKED(vp, "vdropl");
2324 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2325 	if (vp->v_holdcnt <= 0)
2326 		panic("vdrop: holdcnt %d", vp->v_holdcnt);
2327 	vp->v_holdcnt--;
2328 	if (vp->v_holdcnt == 0) {
2329 		if (vp->v_iflag & VI_DOOMED) {
2330 			CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__,
2331 			    vp);
2332 			vdestroy(vp);
2333 			return;
2334 		} else
2335 			vfree(vp);
2336 	}
2337 	VI_UNLOCK(vp);
2338 }
2339 
2340 /*
2341  * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2342  * flags.  DOINGINACT prevents us from recursing in calls to vinactive.
2343  * OWEINACT tracks whether a vnode missed a call to inactive due to a
2344  * failed lock upgrade.
2345  */
2346 static void
2347 vinactive(struct vnode *vp, struct thread *td)
2348 {
2349 
2350 	ASSERT_VOP_ELOCKED(vp, "vinactive");
2351 	ASSERT_VI_LOCKED(vp, "vinactive");
2352 	VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2353 	    ("vinactive: recursed on VI_DOINGINACT"));
2354 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2355 	vp->v_iflag |= VI_DOINGINACT;
2356 	vp->v_iflag &= ~VI_OWEINACT;
2357 	VI_UNLOCK(vp);
2358 	VOP_INACTIVE(vp, td);
2359 	VI_LOCK(vp);
2360 	VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2361 	    ("vinactive: lost VI_DOINGINACT"));
2362 	vp->v_iflag &= ~VI_DOINGINACT;
2363 }
2364 
2365 /*
2366  * Remove any vnodes in the vnode table belonging to mount point mp.
2367  *
2368  * If FORCECLOSE is not specified, there should not be any active ones,
2369  * return error if any are found (nb: this is a user error, not a
2370  * system error). If FORCECLOSE is specified, detach any active vnodes
2371  * that are found.
2372  *
2373  * If WRITECLOSE is set, only flush out regular file vnodes open for
2374  * writing.
2375  *
2376  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2377  *
2378  * `rootrefs' specifies the base reference count for the root vnode
2379  * of this filesystem. The root vnode is considered busy if its
2380  * v_usecount exceeds this value. On a successful return, vflush(, td)
2381  * will call vrele() on the root vnode exactly rootrefs times.
2382  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2383  * be zero.
2384  */
2385 #ifdef DIAGNOSTIC
2386 static int busyprt = 0;		/* print out busy vnodes */
2387 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
2388 #endif
2389 
2390 int
2391 vflush( struct mount *mp, int rootrefs, int flags, struct thread *td)
2392 {
2393 	struct vnode *vp, *mvp, *rootvp = NULL;
2394 	struct vattr vattr;
2395 	int busy = 0, error;
2396 
2397 	CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2398 	    rootrefs, flags);
2399 	if (rootrefs > 0) {
2400 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2401 		    ("vflush: bad args"));
2402 		/*
2403 		 * Get the filesystem root vnode. We can vput() it
2404 		 * immediately, since with rootrefs > 0, it won't go away.
2405 		 */
2406 		if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp, td)) != 0) {
2407 			CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2408 			    __func__, error);
2409 			return (error);
2410 		}
2411 		vput(rootvp);
2412 
2413 	}
2414 	MNT_ILOCK(mp);
2415 loop:
2416 	MNT_VNODE_FOREACH(vp, mp, mvp) {
2417 
2418 		VI_LOCK(vp);
2419 		vholdl(vp);
2420 		MNT_IUNLOCK(mp);
2421 		error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2422 		if (error) {
2423 			vdrop(vp);
2424 			MNT_ILOCK(mp);
2425 			MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
2426 			goto loop;
2427 		}
2428 		/*
2429 		 * Skip over a vnodes marked VV_SYSTEM.
2430 		 */
2431 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2432 			VOP_UNLOCK(vp, 0);
2433 			vdrop(vp);
2434 			MNT_ILOCK(mp);
2435 			continue;
2436 		}
2437 		/*
2438 		 * If WRITECLOSE is set, flush out unlinked but still open
2439 		 * files (even if open only for reading) and regular file
2440 		 * vnodes open for writing.
2441 		 */
2442 		if (flags & WRITECLOSE) {
2443 			error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2444 			VI_LOCK(vp);
2445 
2446 			if ((vp->v_type == VNON ||
2447 			    (error == 0 && vattr.va_nlink > 0)) &&
2448 			    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2449 				VOP_UNLOCK(vp, 0);
2450 				vdropl(vp);
2451 				MNT_ILOCK(mp);
2452 				continue;
2453 			}
2454 		} else
2455 			VI_LOCK(vp);
2456 		/*
2457 		 * With v_usecount == 0, all we need to do is clear out the
2458 		 * vnode data structures and we are done.
2459 		 *
2460 		 * If FORCECLOSE is set, forcibly close the vnode.
2461 		 */
2462 		if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2463 			VNASSERT(vp->v_usecount == 0 ||
2464 			    (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2465 			    ("device VNODE %p is FORCECLOSED", vp));
2466 			vgonel(vp);
2467 		} else {
2468 			busy++;
2469 #ifdef DIAGNOSTIC
2470 			if (busyprt)
2471 				vprint("vflush: busy vnode", vp);
2472 #endif
2473 		}
2474 		VOP_UNLOCK(vp, 0);
2475 		vdropl(vp);
2476 		MNT_ILOCK(mp);
2477 	}
2478 	MNT_IUNLOCK(mp);
2479 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2480 		/*
2481 		 * If just the root vnode is busy, and if its refcount
2482 		 * is equal to `rootrefs', then go ahead and kill it.
2483 		 */
2484 		VI_LOCK(rootvp);
2485 		KASSERT(busy > 0, ("vflush: not busy"));
2486 		VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2487 		    ("vflush: usecount %d < rootrefs %d",
2488 		     rootvp->v_usecount, rootrefs));
2489 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2490 			VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2491 			vgone(rootvp);
2492 			VOP_UNLOCK(rootvp, 0);
2493 			busy = 0;
2494 		} else
2495 			VI_UNLOCK(rootvp);
2496 	}
2497 	if (busy) {
2498 		CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2499 		    busy);
2500 		return (EBUSY);
2501 	}
2502 	for (; rootrefs > 0; rootrefs--)
2503 		vrele(rootvp);
2504 	return (0);
2505 }
2506 
2507 /*
2508  * Recycle an unused vnode to the front of the free list.
2509  */
2510 int
2511 vrecycle(struct vnode *vp, struct thread *td)
2512 {
2513 	int recycled;
2514 
2515 	ASSERT_VOP_ELOCKED(vp, "vrecycle");
2516 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2517 	recycled = 0;
2518 	VI_LOCK(vp);
2519 	if (vp->v_usecount == 0) {
2520 		recycled = 1;
2521 		vgonel(vp);
2522 	}
2523 	VI_UNLOCK(vp);
2524 	return (recycled);
2525 }
2526 
2527 /*
2528  * Eliminate all activity associated with a vnode
2529  * in preparation for reuse.
2530  */
2531 void
2532 vgone(struct vnode *vp)
2533 {
2534 	VI_LOCK(vp);
2535 	vgonel(vp);
2536 	VI_UNLOCK(vp);
2537 }
2538 
2539 /*
2540  * vgone, with the vp interlock held.
2541  */
2542 void
2543 vgonel(struct vnode *vp)
2544 {
2545 	struct thread *td;
2546 	int oweinact;
2547 	int active;
2548 	struct mount *mp;
2549 
2550 	ASSERT_VOP_ELOCKED(vp, "vgonel");
2551 	ASSERT_VI_LOCKED(vp, "vgonel");
2552 	VNASSERT(vp->v_holdcnt, vp,
2553 	    ("vgonel: vp %p has no reference.", vp));
2554 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2555 	td = curthread;
2556 
2557 	/*
2558 	 * Don't vgonel if we're already doomed.
2559 	 */
2560 	if (vp->v_iflag & VI_DOOMED)
2561 		return;
2562 	vp->v_iflag |= VI_DOOMED;
2563 	/*
2564 	 * Check to see if the vnode is in use.  If so, we have to call
2565 	 * VOP_CLOSE() and VOP_INACTIVE().
2566 	 */
2567 	active = vp->v_usecount;
2568 	oweinact = (vp->v_iflag & VI_OWEINACT);
2569 	VI_UNLOCK(vp);
2570 	/*
2571 	 * Clean out any buffers associated with the vnode.
2572 	 * If the flush fails, just toss the buffers.
2573 	 */
2574 	mp = NULL;
2575 	if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2576 		(void) vn_start_secondary_write(vp, &mp, V_WAIT);
2577 	if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
2578 		vinvalbuf(vp, 0, 0, 0);
2579 
2580 	/*
2581 	 * If purging an active vnode, it must be closed and
2582 	 * deactivated before being reclaimed.
2583 	 */
2584 	if (active)
2585 		VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2586 	if (oweinact || active) {
2587 		VI_LOCK(vp);
2588 		if ((vp->v_iflag & VI_DOINGINACT) == 0)
2589 			vinactive(vp, td);
2590 		VI_UNLOCK(vp);
2591 	}
2592 	/*
2593 	 * Reclaim the vnode.
2594 	 */
2595 	if (VOP_RECLAIM(vp, td))
2596 		panic("vgone: cannot reclaim");
2597 	if (mp != NULL)
2598 		vn_finished_secondary_write(mp);
2599 	VNASSERT(vp->v_object == NULL, vp,
2600 	    ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2601 	/*
2602 	 * Clear the advisory locks and wake up waiting threads.
2603 	 */
2604 	lf_purgelocks(vp, &(vp->v_lockf));
2605 	/*
2606 	 * Delete from old mount point vnode list.
2607 	 */
2608 	delmntque(vp);
2609 	cache_purge(vp);
2610 	/*
2611 	 * Done with purge, reset to the standard lock and invalidate
2612 	 * the vnode.
2613 	 */
2614 	VI_LOCK(vp);
2615 	vp->v_vnlock = &vp->v_lock;
2616 	vp->v_op = &dead_vnodeops;
2617 	vp->v_tag = "none";
2618 	vp->v_type = VBAD;
2619 }
2620 
2621 /*
2622  * Calculate the total number of references to a special device.
2623  */
2624 int
2625 vcount(struct vnode *vp)
2626 {
2627 	int count;
2628 
2629 	dev_lock();
2630 	count = vp->v_rdev->si_usecount;
2631 	dev_unlock();
2632 	return (count);
2633 }
2634 
2635 /*
2636  * Same as above, but using the struct cdev *as argument
2637  */
2638 int
2639 count_dev(struct cdev *dev)
2640 {
2641 	int count;
2642 
2643 	dev_lock();
2644 	count = dev->si_usecount;
2645 	dev_unlock();
2646 	return(count);
2647 }
2648 
2649 /*
2650  * Print out a description of a vnode.
2651  */
2652 static char *typename[] =
2653 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2654  "VMARKER"};
2655 
2656 void
2657 vn_printf(struct vnode *vp, const char *fmt, ...)
2658 {
2659 	va_list ap;
2660 	char buf[256], buf2[16];
2661 	u_long flags;
2662 
2663 	va_start(ap, fmt);
2664 	vprintf(fmt, ap);
2665 	va_end(ap);
2666 	printf("%p: ", (void *)vp);
2667 	printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2668 	printf("    usecount %d, writecount %d, refcount %d mountedhere %p\n",
2669 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2670 	buf[0] = '\0';
2671 	buf[1] = '\0';
2672 	if (vp->v_vflag & VV_ROOT)
2673 		strlcat(buf, "|VV_ROOT", sizeof(buf));
2674 	if (vp->v_vflag & VV_ISTTY)
2675 		strlcat(buf, "|VV_ISTTY", sizeof(buf));
2676 	if (vp->v_vflag & VV_NOSYNC)
2677 		strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2678 	if (vp->v_vflag & VV_CACHEDLABEL)
2679 		strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2680 	if (vp->v_vflag & VV_TEXT)
2681 		strlcat(buf, "|VV_TEXT", sizeof(buf));
2682 	if (vp->v_vflag & VV_COPYONWRITE)
2683 		strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2684 	if (vp->v_vflag & VV_SYSTEM)
2685 		strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2686 	if (vp->v_vflag & VV_PROCDEP)
2687 		strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2688 	if (vp->v_vflag & VV_NOKNOTE)
2689 		strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2690 	if (vp->v_vflag & VV_DELETED)
2691 		strlcat(buf, "|VV_DELETED", sizeof(buf));
2692 	if (vp->v_vflag & VV_MD)
2693 		strlcat(buf, "|VV_MD", sizeof(buf));
2694 	flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC |
2695 	    VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2696 	    VV_NOKNOTE | VV_DELETED | VV_MD);
2697 	if (flags != 0) {
2698 		snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2699 		strlcat(buf, buf2, sizeof(buf));
2700 	}
2701 	if (vp->v_iflag & VI_MOUNT)
2702 		strlcat(buf, "|VI_MOUNT", sizeof(buf));
2703 	if (vp->v_iflag & VI_AGE)
2704 		strlcat(buf, "|VI_AGE", sizeof(buf));
2705 	if (vp->v_iflag & VI_DOOMED)
2706 		strlcat(buf, "|VI_DOOMED", sizeof(buf));
2707 	if (vp->v_iflag & VI_FREE)
2708 		strlcat(buf, "|VI_FREE", sizeof(buf));
2709 	if (vp->v_iflag & VI_OBJDIRTY)
2710 		strlcat(buf, "|VI_OBJDIRTY", sizeof(buf));
2711 	if (vp->v_iflag & VI_DOINGINACT)
2712 		strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2713 	if (vp->v_iflag & VI_OWEINACT)
2714 		strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2715 	flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2716 	    VI_OBJDIRTY | VI_DOINGINACT | VI_OWEINACT);
2717 	if (flags != 0) {
2718 		snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2719 		strlcat(buf, buf2, sizeof(buf));
2720 	}
2721 	printf("    flags (%s)\n", buf + 1);
2722 	if (mtx_owned(VI_MTX(vp)))
2723 		printf(" VI_LOCKed");
2724 	if (vp->v_object != NULL)
2725 		printf("    v_object %p ref %d pages %d\n",
2726 		    vp->v_object, vp->v_object->ref_count,
2727 		    vp->v_object->resident_page_count);
2728 	printf("    ");
2729 	lockmgr_printinfo(vp->v_vnlock);
2730 	if (vp->v_data != NULL)
2731 		VOP_PRINT(vp);
2732 }
2733 
2734 #ifdef DDB
2735 /*
2736  * List all of the locked vnodes in the system.
2737  * Called when debugging the kernel.
2738  */
2739 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2740 {
2741 	struct mount *mp, *nmp;
2742 	struct vnode *vp;
2743 
2744 	/*
2745 	 * Note: because this is DDB, we can't obey the locking semantics
2746 	 * for these structures, which means we could catch an inconsistent
2747 	 * state and dereference a nasty pointer.  Not much to be done
2748 	 * about that.
2749 	 */
2750 	db_printf("Locked vnodes\n");
2751 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2752 		nmp = TAILQ_NEXT(mp, mnt_list);
2753 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2754 			if (vp->v_type != VMARKER &&
2755 			    VOP_ISLOCKED(vp))
2756 				vprint("", vp);
2757 		}
2758 		nmp = TAILQ_NEXT(mp, mnt_list);
2759 	}
2760 }
2761 
2762 /*
2763  * Show details about the given vnode.
2764  */
2765 DB_SHOW_COMMAND(vnode, db_show_vnode)
2766 {
2767 	struct vnode *vp;
2768 
2769 	if (!have_addr)
2770 		return;
2771 	vp = (struct vnode *)addr;
2772 	vn_printf(vp, "vnode ");
2773 }
2774 
2775 /*
2776  * Show details about the given mount point.
2777  */
2778 DB_SHOW_COMMAND(mount, db_show_mount)
2779 {
2780 	struct mount *mp;
2781 	struct statfs *sp;
2782 	struct vnode *vp;
2783 	char buf[512];
2784 	u_int flags;
2785 
2786 	if (!have_addr) {
2787 		/* No address given, print short info about all mount points. */
2788 		TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2789 			db_printf("%p %s on %s (%s)\n", mp,
2790 			    mp->mnt_stat.f_mntfromname,
2791 			    mp->mnt_stat.f_mntonname,
2792 			    mp->mnt_stat.f_fstypename);
2793 			if (db_pager_quit)
2794 				break;
2795 		}
2796 		db_printf("\nMore info: show mount <addr>\n");
2797 		return;
2798 	}
2799 
2800 	mp = (struct mount *)addr;
2801 	db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
2802 	    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
2803 
2804 	buf[0] = '\0';
2805 	flags = mp->mnt_flag;
2806 #define	MNT_FLAG(flag)	do {						\
2807 	if (flags & (flag)) {						\
2808 		if (buf[0] != '\0')					\
2809 			strlcat(buf, ", ", sizeof(buf));		\
2810 		strlcat(buf, (#flag) + 4, sizeof(buf));			\
2811 		flags &= ~(flag);					\
2812 	}								\
2813 } while (0)
2814 	MNT_FLAG(MNT_RDONLY);
2815 	MNT_FLAG(MNT_SYNCHRONOUS);
2816 	MNT_FLAG(MNT_NOEXEC);
2817 	MNT_FLAG(MNT_NOSUID);
2818 	MNT_FLAG(MNT_UNION);
2819 	MNT_FLAG(MNT_ASYNC);
2820 	MNT_FLAG(MNT_SUIDDIR);
2821 	MNT_FLAG(MNT_SOFTDEP);
2822 	MNT_FLAG(MNT_NOSYMFOLLOW);
2823 	MNT_FLAG(MNT_GJOURNAL);
2824 	MNT_FLAG(MNT_MULTILABEL);
2825 	MNT_FLAG(MNT_ACLS);
2826 	MNT_FLAG(MNT_NOATIME);
2827 	MNT_FLAG(MNT_NOCLUSTERR);
2828 	MNT_FLAG(MNT_NOCLUSTERW);
2829 	MNT_FLAG(MNT_EXRDONLY);
2830 	MNT_FLAG(MNT_EXPORTED);
2831 	MNT_FLAG(MNT_DEFEXPORTED);
2832 	MNT_FLAG(MNT_EXPORTANON);
2833 	MNT_FLAG(MNT_EXKERB);
2834 	MNT_FLAG(MNT_EXPUBLIC);
2835 	MNT_FLAG(MNT_LOCAL);
2836 	MNT_FLAG(MNT_QUOTA);
2837 	MNT_FLAG(MNT_ROOTFS);
2838 	MNT_FLAG(MNT_USER);
2839 	MNT_FLAG(MNT_IGNORE);
2840 	MNT_FLAG(MNT_UPDATE);
2841 	MNT_FLAG(MNT_DELEXPORT);
2842 	MNT_FLAG(MNT_RELOAD);
2843 	MNT_FLAG(MNT_FORCE);
2844 	MNT_FLAG(MNT_SNAPSHOT);
2845 	MNT_FLAG(MNT_BYFSID);
2846 #undef MNT_FLAG
2847 	if (flags != 0) {
2848 		if (buf[0] != '\0')
2849 			strlcat(buf, ", ", sizeof(buf));
2850 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2851 		    "0x%08x", flags);
2852 	}
2853 	db_printf("    mnt_flag = %s\n", buf);
2854 
2855 	buf[0] = '\0';
2856 	flags = mp->mnt_kern_flag;
2857 #define	MNT_KERN_FLAG(flag)	do {					\
2858 	if (flags & (flag)) {						\
2859 		if (buf[0] != '\0')					\
2860 			strlcat(buf, ", ", sizeof(buf));		\
2861 		strlcat(buf, (#flag) + 5, sizeof(buf));			\
2862 		flags &= ~(flag);					\
2863 	}								\
2864 } while (0)
2865 	MNT_KERN_FLAG(MNTK_UNMOUNTF);
2866 	MNT_KERN_FLAG(MNTK_ASYNC);
2867 	MNT_KERN_FLAG(MNTK_SOFTDEP);
2868 	MNT_KERN_FLAG(MNTK_NOINSMNTQ);
2869 	MNT_KERN_FLAG(MNTK_UNMOUNT);
2870 	MNT_KERN_FLAG(MNTK_MWAIT);
2871 	MNT_KERN_FLAG(MNTK_SUSPEND);
2872 	MNT_KERN_FLAG(MNTK_SUSPEND2);
2873 	MNT_KERN_FLAG(MNTK_SUSPENDED);
2874 	MNT_KERN_FLAG(MNTK_MPSAFE);
2875 	MNT_KERN_FLAG(MNTK_NOKNOTE);
2876 	MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
2877 #undef MNT_KERN_FLAG
2878 	if (flags != 0) {
2879 		if (buf[0] != '\0')
2880 			strlcat(buf, ", ", sizeof(buf));
2881 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2882 		    "0x%08x", flags);
2883 	}
2884 	db_printf("    mnt_kern_flag = %s\n", buf);
2885 
2886 	sp = &mp->mnt_stat;
2887 	db_printf("    mnt_stat = { version=%u type=%u flags=0x%016jx "
2888 	    "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
2889 	    "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
2890 	    "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
2891 	    (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
2892 	    (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
2893 	    (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
2894 	    (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
2895 	    (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
2896 	    (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
2897 	    (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
2898 	    (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
2899 
2900 	db_printf("    mnt_cred = { uid=%u ruid=%u",
2901 	    (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
2902 	if (mp->mnt_cred->cr_prison != NULL)
2903 		db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
2904 	db_printf(" }\n");
2905 	db_printf("    mnt_ref = %d\n", mp->mnt_ref);
2906 	db_printf("    mnt_gen = %d\n", mp->mnt_gen);
2907 	db_printf("    mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
2908 	db_printf("    mnt_writeopcount = %d\n", mp->mnt_writeopcount);
2909 	db_printf("    mnt_noasync = %u\n", mp->mnt_noasync);
2910 	db_printf("    mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
2911 	db_printf("    mnt_iosize_max = %d\n", mp->mnt_iosize_max);
2912 	db_printf("    mnt_hashseed = %u\n", mp->mnt_hashseed);
2913 	db_printf("    mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
2914 	db_printf("    mnt_secondary_accwrites = %d\n",
2915 	    mp->mnt_secondary_accwrites);
2916 	db_printf("    mnt_gjprovider = %s\n",
2917 	    mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
2918 	db_printf("\n");
2919 
2920 	TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2921 		if (vp->v_type != VMARKER) {
2922 			vn_printf(vp, "vnode ");
2923 			if (db_pager_quit)
2924 				break;
2925 		}
2926 	}
2927 }
2928 #endif	/* DDB */
2929 
2930 /*
2931  * Fill in a struct xvfsconf based on a struct vfsconf.
2932  */
2933 static void
2934 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2935 {
2936 
2937 	strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2938 	xvfsp->vfc_typenum = vfsp->vfc_typenum;
2939 	xvfsp->vfc_refcount = vfsp->vfc_refcount;
2940 	xvfsp->vfc_flags = vfsp->vfc_flags;
2941 	/*
2942 	 * These are unused in userland, we keep them
2943 	 * to not break binary compatibility.
2944 	 */
2945 	xvfsp->vfc_vfsops = NULL;
2946 	xvfsp->vfc_next = NULL;
2947 }
2948 
2949 /*
2950  * Top level filesystem related information gathering.
2951  */
2952 static int
2953 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2954 {
2955 	struct vfsconf *vfsp;
2956 	struct xvfsconf xvfsp;
2957 	int error;
2958 
2959 	error = 0;
2960 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2961 		bzero(&xvfsp, sizeof(xvfsp));
2962 		vfsconf2x(vfsp, &xvfsp);
2963 		error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
2964 		if (error)
2965 			break;
2966 	}
2967 	return (error);
2968 }
2969 
2970 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
2971     "S,xvfsconf", "List of all configured filesystems");
2972 
2973 #ifndef BURN_BRIDGES
2974 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2975 
2976 static int
2977 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2978 {
2979 	int *name = (int *)arg1 - 1;	/* XXX */
2980 	u_int namelen = arg2 + 1;	/* XXX */
2981 	struct vfsconf *vfsp;
2982 	struct xvfsconf xvfsp;
2983 
2984 	printf("WARNING: userland calling deprecated sysctl, "
2985 	    "please rebuild world\n");
2986 
2987 #if 1 || defined(COMPAT_PRELITE2)
2988 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2989 	if (namelen == 1)
2990 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2991 #endif
2992 
2993 	switch (name[1]) {
2994 	case VFS_MAXTYPENUM:
2995 		if (namelen != 2)
2996 			return (ENOTDIR);
2997 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2998 	case VFS_CONF:
2999 		if (namelen != 3)
3000 			return (ENOTDIR);	/* overloaded */
3001 		TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
3002 			if (vfsp->vfc_typenum == name[2])
3003 				break;
3004 		if (vfsp == NULL)
3005 			return (EOPNOTSUPP);
3006 		bzero(&xvfsp, sizeof(xvfsp));
3007 		vfsconf2x(vfsp, &xvfsp);
3008 		return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3009 	}
3010 	return (EOPNOTSUPP);
3011 }
3012 
3013 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
3014 	vfs_sysctl, "Generic filesystem");
3015 
3016 #if 1 || defined(COMPAT_PRELITE2)
3017 
3018 static int
3019 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3020 {
3021 	int error;
3022 	struct vfsconf *vfsp;
3023 	struct ovfsconf ovfs;
3024 
3025 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3026 		bzero(&ovfs, sizeof(ovfs));
3027 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
3028 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
3029 		ovfs.vfc_index = vfsp->vfc_typenum;
3030 		ovfs.vfc_refcount = vfsp->vfc_refcount;
3031 		ovfs.vfc_flags = vfsp->vfc_flags;
3032 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3033 		if (error)
3034 			return error;
3035 	}
3036 	return 0;
3037 }
3038 
3039 #endif /* 1 || COMPAT_PRELITE2 */
3040 #endif /* !BURN_BRIDGES */
3041 
3042 #define KINFO_VNODESLOP		10
3043 #ifdef notyet
3044 /*
3045  * Dump vnode list (via sysctl).
3046  */
3047 /* ARGSUSED */
3048 static int
3049 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3050 {
3051 	struct xvnode *xvn;
3052 	struct mount *mp;
3053 	struct vnode *vp;
3054 	int error, len, n;
3055 
3056 	/*
3057 	 * Stale numvnodes access is not fatal here.
3058 	 */
3059 	req->lock = 0;
3060 	len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3061 	if (!req->oldptr)
3062 		/* Make an estimate */
3063 		return (SYSCTL_OUT(req, 0, len));
3064 
3065 	error = sysctl_wire_old_buffer(req, 0);
3066 	if (error != 0)
3067 		return (error);
3068 	xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3069 	n = 0;
3070 	mtx_lock(&mountlist_mtx);
3071 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3072 		if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3073 			continue;
3074 		MNT_ILOCK(mp);
3075 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3076 			if (n == len)
3077 				break;
3078 			vref(vp);
3079 			xvn[n].xv_size = sizeof *xvn;
3080 			xvn[n].xv_vnode = vp;
3081 			xvn[n].xv_id = 0;	/* XXX compat */
3082 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3083 			XV_COPY(usecount);
3084 			XV_COPY(writecount);
3085 			XV_COPY(holdcnt);
3086 			XV_COPY(mount);
3087 			XV_COPY(numoutput);
3088 			XV_COPY(type);
3089 #undef XV_COPY
3090 			xvn[n].xv_flag = vp->v_vflag;
3091 
3092 			switch (vp->v_type) {
3093 			case VREG:
3094 			case VDIR:
3095 			case VLNK:
3096 				break;
3097 			case VBLK:
3098 			case VCHR:
3099 				if (vp->v_rdev == NULL) {
3100 					vrele(vp);
3101 					continue;
3102 				}
3103 				xvn[n].xv_dev = dev2udev(vp->v_rdev);
3104 				break;
3105 			case VSOCK:
3106 				xvn[n].xv_socket = vp->v_socket;
3107 				break;
3108 			case VFIFO:
3109 				xvn[n].xv_fifo = vp->v_fifoinfo;
3110 				break;
3111 			case VNON:
3112 			case VBAD:
3113 			default:
3114 				/* shouldn't happen? */
3115 				vrele(vp);
3116 				continue;
3117 			}
3118 			vrele(vp);
3119 			++n;
3120 		}
3121 		MNT_IUNLOCK(mp);
3122 		mtx_lock(&mountlist_mtx);
3123 		vfs_unbusy(mp);
3124 		if (n == len)
3125 			break;
3126 	}
3127 	mtx_unlock(&mountlist_mtx);
3128 
3129 	error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3130 	free(xvn, M_TEMP);
3131 	return (error);
3132 }
3133 
3134 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3135 	0, 0, sysctl_vnode, "S,xvnode", "");
3136 #endif
3137 
3138 /*
3139  * Unmount all filesystems. The list is traversed in reverse order
3140  * of mounting to avoid dependencies.
3141  */
3142 void
3143 vfs_unmountall(void)
3144 {
3145 	struct mount *mp;
3146 	struct thread *td;
3147 	int error;
3148 
3149 	KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
3150 	CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3151 	td = curthread;
3152 
3153 	/*
3154 	 * Since this only runs when rebooting, it is not interlocked.
3155 	 */
3156 	while(!TAILQ_EMPTY(&mountlist)) {
3157 		mp = TAILQ_LAST(&mountlist, mntlist);
3158 		error = dounmount(mp, MNT_FORCE, td);
3159 		if (error) {
3160 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
3161 			/*
3162 			 * XXX: Due to the way in which we mount the root
3163 			 * file system off of devfs, devfs will generate a
3164 			 * "busy" warning when we try to unmount it before
3165 			 * the root.  Don't print a warning as a result in
3166 			 * order to avoid false positive errors that may
3167 			 * cause needless upset.
3168 			 */
3169 			if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3170 				printf("unmount of %s failed (",
3171 				    mp->mnt_stat.f_mntonname);
3172 				if (error == EBUSY)
3173 					printf("BUSY)\n");
3174 				else
3175 					printf("%d)\n", error);
3176 			}
3177 		} else {
3178 			/* The unmount has removed mp from the mountlist */
3179 		}
3180 	}
3181 }
3182 
3183 /*
3184  * perform msync on all vnodes under a mount point
3185  * the mount point must be locked.
3186  */
3187 void
3188 vfs_msync(struct mount *mp, int flags)
3189 {
3190 	struct vnode *vp, *mvp;
3191 	struct vm_object *obj;
3192 
3193 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3194 	MNT_ILOCK(mp);
3195 	MNT_VNODE_FOREACH(vp, mp, mvp) {
3196 		VI_LOCK(vp);
3197 		if ((vp->v_iflag & VI_OBJDIRTY) &&
3198 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3199 			MNT_IUNLOCK(mp);
3200 			if (!vget(vp,
3201 			    LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3202 			    curthread)) {
3203 				if (vp->v_vflag & VV_NOSYNC) {	/* unlinked */
3204 					vput(vp);
3205 					MNT_ILOCK(mp);
3206 					continue;
3207 				}
3208 
3209 				obj = vp->v_object;
3210 				if (obj != NULL) {
3211 					VM_OBJECT_LOCK(obj);
3212 					vm_object_page_clean(obj, 0, 0,
3213 					    flags == MNT_WAIT ?
3214 					    OBJPC_SYNC : OBJPC_NOSYNC);
3215 					VM_OBJECT_UNLOCK(obj);
3216 				}
3217 				vput(vp);
3218 			}
3219 			MNT_ILOCK(mp);
3220 		} else
3221 			VI_UNLOCK(vp);
3222 	}
3223 	MNT_IUNLOCK(mp);
3224 }
3225 
3226 /*
3227  * Mark a vnode as free, putting it up for recycling.
3228  */
3229 static void
3230 vfree(struct vnode *vp)
3231 {
3232 
3233 	ASSERT_VI_LOCKED(vp, "vfree");
3234 	mtx_lock(&vnode_free_list_mtx);
3235 	VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed."));
3236 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free"));
3237 	VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't"));
3238 	VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp,
3239 	    ("vfree: Freeing doomed vnode"));
3240 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3241 	if (vp->v_iflag & VI_AGE) {
3242 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3243 	} else {
3244 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3245 	}
3246 	freevnodes++;
3247 	vp->v_iflag &= ~VI_AGE;
3248 	vp->v_iflag |= VI_FREE;
3249 	mtx_unlock(&vnode_free_list_mtx);
3250 }
3251 
3252 /*
3253  * Opposite of vfree() - mark a vnode as in use.
3254  */
3255 static void
3256 vbusy(struct vnode *vp)
3257 {
3258 	ASSERT_VI_LOCKED(vp, "vbusy");
3259 	VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
3260 	VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed."));
3261 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3262 
3263 	mtx_lock(&vnode_free_list_mtx);
3264 	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3265 	freevnodes--;
3266 	vp->v_iflag &= ~(VI_FREE|VI_AGE);
3267 	mtx_unlock(&vnode_free_list_mtx);
3268 }
3269 
3270 static void
3271 destroy_vpollinfo(struct vpollinfo *vi)
3272 {
3273 	knlist_destroy(&vi->vpi_selinfo.si_note);
3274 	mtx_destroy(&vi->vpi_lock);
3275 	uma_zfree(vnodepoll_zone, vi);
3276 }
3277 
3278 /*
3279  * Initalize per-vnode helper structure to hold poll-related state.
3280  */
3281 void
3282 v_addpollinfo(struct vnode *vp)
3283 {
3284 	struct vpollinfo *vi;
3285 
3286 	if (vp->v_pollinfo != NULL)
3287 		return;
3288 	vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3289 	mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3290 	knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3291 	    vfs_knlunlock, vfs_knllocked);
3292 	VI_LOCK(vp);
3293 	if (vp->v_pollinfo != NULL) {
3294 		VI_UNLOCK(vp);
3295 		destroy_vpollinfo(vi);
3296 		return;
3297 	}
3298 	vp->v_pollinfo = vi;
3299 	VI_UNLOCK(vp);
3300 }
3301 
3302 /*
3303  * Record a process's interest in events which might happen to
3304  * a vnode.  Because poll uses the historic select-style interface
3305  * internally, this routine serves as both the ``check for any
3306  * pending events'' and the ``record my interest in future events''
3307  * functions.  (These are done together, while the lock is held,
3308  * to avoid race conditions.)
3309  */
3310 int
3311 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3312 {
3313 
3314 	v_addpollinfo(vp);
3315 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3316 	if (vp->v_pollinfo->vpi_revents & events) {
3317 		/*
3318 		 * This leaves events we are not interested
3319 		 * in available for the other process which
3320 		 * which presumably had requested them
3321 		 * (otherwise they would never have been
3322 		 * recorded).
3323 		 */
3324 		events &= vp->v_pollinfo->vpi_revents;
3325 		vp->v_pollinfo->vpi_revents &= ~events;
3326 
3327 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
3328 		return (events);
3329 	}
3330 	vp->v_pollinfo->vpi_events |= events;
3331 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3332 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3333 	return (0);
3334 }
3335 
3336 /*
3337  * Routine to create and manage a filesystem syncer vnode.
3338  */
3339 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
3340 static int	sync_fsync(struct  vop_fsync_args *);
3341 static int	sync_inactive(struct  vop_inactive_args *);
3342 static int	sync_reclaim(struct  vop_reclaim_args *);
3343 
3344 static struct vop_vector sync_vnodeops = {
3345 	.vop_bypass =	VOP_EOPNOTSUPP,
3346 	.vop_close =	sync_close,		/* close */
3347 	.vop_fsync =	sync_fsync,		/* fsync */
3348 	.vop_inactive =	sync_inactive,	/* inactive */
3349 	.vop_reclaim =	sync_reclaim,	/* reclaim */
3350 	.vop_lock1 =	vop_stdlock,	/* lock */
3351 	.vop_unlock =	vop_stdunlock,	/* unlock */
3352 	.vop_islocked =	vop_stdislocked,	/* islocked */
3353 };
3354 
3355 /*
3356  * Create a new filesystem syncer vnode for the specified mount point.
3357  */
3358 int
3359 vfs_allocate_syncvnode(struct mount *mp)
3360 {
3361 	struct vnode *vp;
3362 	struct bufobj *bo;
3363 	static long start, incr, next;
3364 	int error;
3365 
3366 	/* Allocate a new vnode */
3367 	if ((error = getnewvnode("syncer", mp, &sync_vnodeops, &vp)) != 0) {
3368 		mp->mnt_syncer = NULL;
3369 		return (error);
3370 	}
3371 	vp->v_type = VNON;
3372 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3373 	vp->v_vflag |= VV_FORCEINSMQ;
3374 	error = insmntque(vp, mp);
3375 	if (error != 0)
3376 		panic("vfs_allocate_syncvnode: insmntque failed");
3377 	vp->v_vflag &= ~VV_FORCEINSMQ;
3378 	VOP_UNLOCK(vp, 0);
3379 	/*
3380 	 * Place the vnode onto the syncer worklist. We attempt to
3381 	 * scatter them about on the list so that they will go off
3382 	 * at evenly distributed times even if all the filesystems
3383 	 * are mounted at once.
3384 	 */
3385 	next += incr;
3386 	if (next == 0 || next > syncer_maxdelay) {
3387 		start /= 2;
3388 		incr /= 2;
3389 		if (start == 0) {
3390 			start = syncer_maxdelay / 2;
3391 			incr = syncer_maxdelay;
3392 		}
3393 		next = start;
3394 	}
3395 	bo = &vp->v_bufobj;
3396 	BO_LOCK(bo);
3397 	vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3398 	/* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3399 	mtx_lock(&sync_mtx);
3400 	sync_vnode_count++;
3401 	mtx_unlock(&sync_mtx);
3402 	BO_UNLOCK(bo);
3403 	mp->mnt_syncer = vp;
3404 	return (0);
3405 }
3406 
3407 /*
3408  * Do a lazy sync of the filesystem.
3409  */
3410 static int
3411 sync_fsync(struct vop_fsync_args *ap)
3412 {
3413 	struct vnode *syncvp = ap->a_vp;
3414 	struct mount *mp = syncvp->v_mount;
3415 	int error;
3416 	struct bufobj *bo;
3417 
3418 	/*
3419 	 * We only need to do something if this is a lazy evaluation.
3420 	 */
3421 	if (ap->a_waitfor != MNT_LAZY)
3422 		return (0);
3423 
3424 	/*
3425 	 * Move ourselves to the back of the sync list.
3426 	 */
3427 	bo = &syncvp->v_bufobj;
3428 	BO_LOCK(bo);
3429 	vn_syncer_add_to_worklist(bo, syncdelay);
3430 	BO_UNLOCK(bo);
3431 
3432 	/*
3433 	 * Walk the list of vnodes pushing all that are dirty and
3434 	 * not already on the sync list.
3435 	 */
3436 	mtx_lock(&mountlist_mtx);
3437 	if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) {
3438 		mtx_unlock(&mountlist_mtx);
3439 		return (0);
3440 	}
3441 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3442 		vfs_unbusy(mp);
3443 		return (0);
3444 	}
3445 	MNT_ILOCK(mp);
3446 	mp->mnt_noasync++;
3447 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
3448 	MNT_IUNLOCK(mp);
3449 	vfs_msync(mp, MNT_NOWAIT);
3450 	error = VFS_SYNC(mp, MNT_LAZY, ap->a_td);
3451 	MNT_ILOCK(mp);
3452 	mp->mnt_noasync--;
3453 	if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
3454 		mp->mnt_kern_flag |= MNTK_ASYNC;
3455 	MNT_IUNLOCK(mp);
3456 	vn_finished_write(mp);
3457 	vfs_unbusy(mp);
3458 	return (error);
3459 }
3460 
3461 /*
3462  * The syncer vnode is no referenced.
3463  */
3464 static int
3465 sync_inactive(struct vop_inactive_args *ap)
3466 {
3467 
3468 	vgone(ap->a_vp);
3469 	return (0);
3470 }
3471 
3472 /*
3473  * The syncer vnode is no longer needed and is being decommissioned.
3474  *
3475  * Modifications to the worklist must be protected by sync_mtx.
3476  */
3477 static int
3478 sync_reclaim(struct vop_reclaim_args *ap)
3479 {
3480 	struct vnode *vp = ap->a_vp;
3481 	struct bufobj *bo;
3482 
3483 	bo = &vp->v_bufobj;
3484 	BO_LOCK(bo);
3485 	vp->v_mount->mnt_syncer = NULL;
3486 	if (bo->bo_flag & BO_ONWORKLST) {
3487 		mtx_lock(&sync_mtx);
3488 		LIST_REMOVE(bo, bo_synclist);
3489 		syncer_worklist_len--;
3490 		sync_vnode_count--;
3491 		mtx_unlock(&sync_mtx);
3492 		bo->bo_flag &= ~BO_ONWORKLST;
3493 	}
3494 	BO_UNLOCK(bo);
3495 
3496 	return (0);
3497 }
3498 
3499 /*
3500  * Check if vnode represents a disk device
3501  */
3502 int
3503 vn_isdisk(struct vnode *vp, int *errp)
3504 {
3505 	int error;
3506 
3507 	error = 0;
3508 	dev_lock();
3509 	if (vp->v_type != VCHR)
3510 		error = ENOTBLK;
3511 	else if (vp->v_rdev == NULL)
3512 		error = ENXIO;
3513 	else if (vp->v_rdev->si_devsw == NULL)
3514 		error = ENXIO;
3515 	else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3516 		error = ENOTBLK;
3517 	dev_unlock();
3518 	if (errp != NULL)
3519 		*errp = error;
3520 	return (error == 0);
3521 }
3522 
3523 /*
3524  * Common filesystem object access control check routine.  Accepts a
3525  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3526  * and optional call-by-reference privused argument allowing vaccess()
3527  * to indicate to the caller whether privilege was used to satisfy the
3528  * request (obsoleted).  Returns 0 on success, or an errno on failure.
3529  *
3530  * The ifdef'd CAPABILITIES version is here for reference, but is not
3531  * actually used.
3532  */
3533 int
3534 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3535     accmode_t accmode, struct ucred *cred, int *privused)
3536 {
3537 	accmode_t dac_granted;
3538 	accmode_t priv_granted;
3539 
3540 	/*
3541 	 * Look for a normal, non-privileged way to access the file/directory
3542 	 * as requested.  If it exists, go with that.
3543 	 */
3544 
3545 	if (privused != NULL)
3546 		*privused = 0;
3547 
3548 	dac_granted = 0;
3549 
3550 	/* Check the owner. */
3551 	if (cred->cr_uid == file_uid) {
3552 		dac_granted |= VADMIN;
3553 		if (file_mode & S_IXUSR)
3554 			dac_granted |= VEXEC;
3555 		if (file_mode & S_IRUSR)
3556 			dac_granted |= VREAD;
3557 		if (file_mode & S_IWUSR)
3558 			dac_granted |= (VWRITE | VAPPEND);
3559 
3560 		if ((accmode & dac_granted) == accmode)
3561 			return (0);
3562 
3563 		goto privcheck;
3564 	}
3565 
3566 	/* Otherwise, check the groups (first match) */
3567 	if (groupmember(file_gid, cred)) {
3568 		if (file_mode & S_IXGRP)
3569 			dac_granted |= VEXEC;
3570 		if (file_mode & S_IRGRP)
3571 			dac_granted |= VREAD;
3572 		if (file_mode & S_IWGRP)
3573 			dac_granted |= (VWRITE | VAPPEND);
3574 
3575 		if ((accmode & dac_granted) == accmode)
3576 			return (0);
3577 
3578 		goto privcheck;
3579 	}
3580 
3581 	/* Otherwise, check everyone else. */
3582 	if (file_mode & S_IXOTH)
3583 		dac_granted |= VEXEC;
3584 	if (file_mode & S_IROTH)
3585 		dac_granted |= VREAD;
3586 	if (file_mode & S_IWOTH)
3587 		dac_granted |= (VWRITE | VAPPEND);
3588 	if ((accmode & dac_granted) == accmode)
3589 		return (0);
3590 
3591 privcheck:
3592 	/*
3593 	 * Build a privilege mask to determine if the set of privileges
3594 	 * satisfies the requirements when combined with the granted mask
3595 	 * from above.  For each privilege, if the privilege is required,
3596 	 * bitwise or the request type onto the priv_granted mask.
3597 	 */
3598 	priv_granted = 0;
3599 
3600 	if (type == VDIR) {
3601 		/*
3602 		 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3603 		 * requests, instead of PRIV_VFS_EXEC.
3604 		 */
3605 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3606 		    !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3607 			priv_granted |= VEXEC;
3608 	} else {
3609 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3610 		    !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3611 			priv_granted |= VEXEC;
3612 	}
3613 
3614 	if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
3615 	    !priv_check_cred(cred, PRIV_VFS_READ, 0))
3616 		priv_granted |= VREAD;
3617 
3618 	if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3619 	    !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3620 		priv_granted |= (VWRITE | VAPPEND);
3621 
3622 	if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3623 	    !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3624 		priv_granted |= VADMIN;
3625 
3626 	if ((accmode & (priv_granted | dac_granted)) == accmode) {
3627 		/* XXX audit: privilege used */
3628 		if (privused != NULL)
3629 			*privused = 1;
3630 		return (0);
3631 	}
3632 
3633 	return ((accmode & VADMIN) ? EPERM : EACCES);
3634 }
3635 
3636 /*
3637  * Credential check based on process requesting service, and per-attribute
3638  * permissions.
3639  */
3640 int
3641 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3642     struct thread *td, accmode_t accmode)
3643 {
3644 
3645 	/*
3646 	 * Kernel-invoked always succeeds.
3647 	 */
3648 	if (cred == NOCRED)
3649 		return (0);
3650 
3651 	/*
3652 	 * Do not allow privileged processes in jail to directly manipulate
3653 	 * system attributes.
3654 	 */
3655 	switch (attrnamespace) {
3656 	case EXTATTR_NAMESPACE_SYSTEM:
3657 		/* Potentially should be: return (EPERM); */
3658 		return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3659 	case EXTATTR_NAMESPACE_USER:
3660 		return (VOP_ACCESS(vp, accmode, cred, td));
3661 	default:
3662 		return (EPERM);
3663 	}
3664 }
3665 
3666 #ifdef DEBUG_VFS_LOCKS
3667 /*
3668  * This only exists to supress warnings from unlocked specfs accesses.  It is
3669  * no longer ok to have an unlocked VFS.
3670  */
3671 #define	IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL ||		\
3672 	(vp)->v_type == VCHR ||	(vp)->v_type == VBAD)
3673 
3674 int vfs_badlock_ddb = 1;	/* Drop into debugger on violation. */
3675 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, "");
3676 
3677 int vfs_badlock_mutex = 1;	/* Check for interlock across VOPs. */
3678 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, "");
3679 
3680 int vfs_badlock_print = 1;	/* Print lock violations. */
3681 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, "");
3682 
3683 #ifdef KDB
3684 int vfs_badlock_backtrace = 1;	/* Print backtrace at lock violations. */
3685 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, "");
3686 #endif
3687 
3688 static void
3689 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3690 {
3691 
3692 #ifdef KDB
3693 	if (vfs_badlock_backtrace)
3694 		kdb_backtrace();
3695 #endif
3696 	if (vfs_badlock_print)
3697 		printf("%s: %p %s\n", str, (void *)vp, msg);
3698 	if (vfs_badlock_ddb)
3699 		kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3700 }
3701 
3702 void
3703 assert_vi_locked(struct vnode *vp, const char *str)
3704 {
3705 
3706 	if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3707 		vfs_badlock("interlock is not locked but should be", str, vp);
3708 }
3709 
3710 void
3711 assert_vi_unlocked(struct vnode *vp, const char *str)
3712 {
3713 
3714 	if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3715 		vfs_badlock("interlock is locked but should not be", str, vp);
3716 }
3717 
3718 void
3719 assert_vop_locked(struct vnode *vp, const char *str)
3720 {
3721 
3722 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0)
3723 		vfs_badlock("is not locked but should be", str, vp);
3724 }
3725 
3726 void
3727 assert_vop_unlocked(struct vnode *vp, const char *str)
3728 {
3729 
3730 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
3731 		vfs_badlock("is locked but should not be", str, vp);
3732 }
3733 
3734 void
3735 assert_vop_elocked(struct vnode *vp, const char *str)
3736 {
3737 
3738 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
3739 		vfs_badlock("is not exclusive locked but should be", str, vp);
3740 }
3741 
3742 #if 0
3743 void
3744 assert_vop_elocked_other(struct vnode *vp, const char *str)
3745 {
3746 
3747 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
3748 		vfs_badlock("is not exclusive locked by another thread",
3749 		    str, vp);
3750 }
3751 
3752 void
3753 assert_vop_slocked(struct vnode *vp, const char *str)
3754 {
3755 
3756 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
3757 		vfs_badlock("is not locked shared but should be", str, vp);
3758 }
3759 #endif /* 0 */
3760 #endif /* DEBUG_VFS_LOCKS */
3761 
3762 void
3763 vop_rename_pre(void *ap)
3764 {
3765 	struct vop_rename_args *a = ap;
3766 
3767 #ifdef DEBUG_VFS_LOCKS
3768 	if (a->a_tvp)
3769 		ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
3770 	ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
3771 	ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
3772 	ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
3773 
3774 	/* Check the source (from). */
3775 	if (a->a_tdvp != a->a_fdvp && a->a_tvp != a->a_fdvp)
3776 		ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3777 	if (a->a_tvp != a->a_fvp)
3778 		ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
3779 
3780 	/* Check the target. */
3781 	if (a->a_tvp)
3782 		ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3783 	ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3784 #endif
3785 	if (a->a_tdvp != a->a_fdvp)
3786 		vhold(a->a_fdvp);
3787 	if (a->a_tvp != a->a_fvp)
3788 		vhold(a->a_fvp);
3789 	vhold(a->a_tdvp);
3790 	if (a->a_tvp)
3791 		vhold(a->a_tvp);
3792 }
3793 
3794 void
3795 vop_strategy_pre(void *ap)
3796 {
3797 #ifdef DEBUG_VFS_LOCKS
3798 	struct vop_strategy_args *a;
3799 	struct buf *bp;
3800 
3801 	a = ap;
3802 	bp = a->a_bp;
3803 
3804 	/*
3805 	 * Cluster ops lock their component buffers but not the IO container.
3806 	 */
3807 	if ((bp->b_flags & B_CLUSTER) != 0)
3808 		return;
3809 
3810 	if (!BUF_ISLOCKED(bp)) {
3811 		if (vfs_badlock_print)
3812 			printf(
3813 			    "VOP_STRATEGY: bp is not locked but should be\n");
3814 		if (vfs_badlock_ddb)
3815 			kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3816 	}
3817 #endif
3818 }
3819 
3820 void
3821 vop_lookup_pre(void *ap)
3822 {
3823 #ifdef DEBUG_VFS_LOCKS
3824 	struct vop_lookup_args *a;
3825 	struct vnode *dvp;
3826 
3827 	a = ap;
3828 	dvp = a->a_dvp;
3829 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3830 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3831 #endif
3832 }
3833 
3834 void
3835 vop_lookup_post(void *ap, int rc)
3836 {
3837 #ifdef DEBUG_VFS_LOCKS
3838 	struct vop_lookup_args *a;
3839 	struct vnode *dvp;
3840 	struct vnode *vp;
3841 
3842 	a = ap;
3843 	dvp = a->a_dvp;
3844 	vp = *(a->a_vpp);
3845 
3846 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3847 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3848 
3849 	if (!rc)
3850 		ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
3851 #endif
3852 }
3853 
3854 void
3855 vop_lock_pre(void *ap)
3856 {
3857 #ifdef DEBUG_VFS_LOCKS
3858 	struct vop_lock1_args *a = ap;
3859 
3860 	if ((a->a_flags & LK_INTERLOCK) == 0)
3861 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3862 	else
3863 		ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
3864 #endif
3865 }
3866 
3867 void
3868 vop_lock_post(void *ap, int rc)
3869 {
3870 #ifdef DEBUG_VFS_LOCKS
3871 	struct vop_lock1_args *a = ap;
3872 
3873 	ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3874 	if (rc == 0)
3875 		ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
3876 #endif
3877 }
3878 
3879 void
3880 vop_unlock_pre(void *ap)
3881 {
3882 #ifdef DEBUG_VFS_LOCKS
3883 	struct vop_unlock_args *a = ap;
3884 
3885 	if (a->a_flags & LK_INTERLOCK)
3886 		ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
3887 	ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
3888 #endif
3889 }
3890 
3891 void
3892 vop_unlock_post(void *ap, int rc)
3893 {
3894 #ifdef DEBUG_VFS_LOCKS
3895 	struct vop_unlock_args *a = ap;
3896 
3897 	if (a->a_flags & LK_INTERLOCK)
3898 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
3899 #endif
3900 }
3901 
3902 void
3903 vop_create_post(void *ap, int rc)
3904 {
3905 	struct vop_create_args *a = ap;
3906 
3907 	if (!rc)
3908 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3909 }
3910 
3911 void
3912 vop_link_post(void *ap, int rc)
3913 {
3914 	struct vop_link_args *a = ap;
3915 
3916 	if (!rc) {
3917 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
3918 		VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
3919 	}
3920 }
3921 
3922 void
3923 vop_mkdir_post(void *ap, int rc)
3924 {
3925 	struct vop_mkdir_args *a = ap;
3926 
3927 	if (!rc)
3928 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3929 }
3930 
3931 void
3932 vop_mknod_post(void *ap, int rc)
3933 {
3934 	struct vop_mknod_args *a = ap;
3935 
3936 	if (!rc)
3937 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3938 }
3939 
3940 void
3941 vop_remove_post(void *ap, int rc)
3942 {
3943 	struct vop_remove_args *a = ap;
3944 
3945 	if (!rc) {
3946 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3947 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3948 	}
3949 }
3950 
3951 void
3952 vop_rename_post(void *ap, int rc)
3953 {
3954 	struct vop_rename_args *a = ap;
3955 
3956 	if (!rc) {
3957 		VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
3958 		VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
3959 		VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
3960 		if (a->a_tvp)
3961 			VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
3962 	}
3963 	if (a->a_tdvp != a->a_fdvp)
3964 		vdrop(a->a_fdvp);
3965 	if (a->a_tvp != a->a_fvp)
3966 		vdrop(a->a_fvp);
3967 	vdrop(a->a_tdvp);
3968 	if (a->a_tvp)
3969 		vdrop(a->a_tvp);
3970 }
3971 
3972 void
3973 vop_rmdir_post(void *ap, int rc)
3974 {
3975 	struct vop_rmdir_args *a = ap;
3976 
3977 	if (!rc) {
3978 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3979 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3980 	}
3981 }
3982 
3983 void
3984 vop_setattr_post(void *ap, int rc)
3985 {
3986 	struct vop_setattr_args *a = ap;
3987 
3988 	if (!rc)
3989 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
3990 }
3991 
3992 void
3993 vop_symlink_post(void *ap, int rc)
3994 {
3995 	struct vop_symlink_args *a = ap;
3996 
3997 	if (!rc)
3998 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3999 }
4000 
4001 static struct knlist fs_knlist;
4002 
4003 static void
4004 vfs_event_init(void *arg)
4005 {
4006 	knlist_init(&fs_knlist, NULL, NULL, NULL, NULL);
4007 }
4008 /* XXX - correct order? */
4009 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4010 
4011 void
4012 vfs_event_signal(fsid_t *fsid, u_int32_t event, intptr_t data __unused)
4013 {
4014 
4015 	KNOTE_UNLOCKED(&fs_knlist, event);
4016 }
4017 
4018 static int	filt_fsattach(struct knote *kn);
4019 static void	filt_fsdetach(struct knote *kn);
4020 static int	filt_fsevent(struct knote *kn, long hint);
4021 
4022 struct filterops fs_filtops =
4023 	{ 0, filt_fsattach, filt_fsdetach, filt_fsevent };
4024 
4025 static int
4026 filt_fsattach(struct knote *kn)
4027 {
4028 
4029 	kn->kn_flags |= EV_CLEAR;
4030 	knlist_add(&fs_knlist, kn, 0);
4031 	return (0);
4032 }
4033 
4034 static void
4035 filt_fsdetach(struct knote *kn)
4036 {
4037 
4038 	knlist_remove(&fs_knlist, kn, 0);
4039 }
4040 
4041 static int
4042 filt_fsevent(struct knote *kn, long hint)
4043 {
4044 
4045 	kn->kn_fflags |= hint;
4046 	return (kn->kn_fflags != 0);
4047 }
4048 
4049 static int
4050 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4051 {
4052 	struct vfsidctl vc;
4053 	int error;
4054 	struct mount *mp;
4055 
4056 	error = SYSCTL_IN(req, &vc, sizeof(vc));
4057 	if (error)
4058 		return (error);
4059 	if (vc.vc_vers != VFS_CTL_VERS1)
4060 		return (EINVAL);
4061 	mp = vfs_getvfs(&vc.vc_fsid);
4062 	if (mp == NULL)
4063 		return (ENOENT);
4064 	/* ensure that a specific sysctl goes to the right filesystem. */
4065 	if (strcmp(vc.vc_fstypename, "*") != 0 &&
4066 	    strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4067 		vfs_rel(mp);
4068 		return (EINVAL);
4069 	}
4070 	VCTLTOREQ(&vc, req);
4071 	error = VFS_SYSCTL(mp, vc.vc_op, req);
4072 	vfs_rel(mp);
4073 	return (error);
4074 }
4075 
4076 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "",
4077     "Sysctl by fsid");
4078 
4079 /*
4080  * Function to initialize a va_filerev field sensibly.
4081  * XXX: Wouldn't a random number make a lot more sense ??
4082  */
4083 u_quad_t
4084 init_va_filerev(void)
4085 {
4086 	struct bintime bt;
4087 
4088 	getbinuptime(&bt);
4089 	return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4090 }
4091 
4092 static int	filt_vfsread(struct knote *kn, long hint);
4093 static int	filt_vfswrite(struct knote *kn, long hint);
4094 static int	filt_vfsvnode(struct knote *kn, long hint);
4095 static void	filt_vfsdetach(struct knote *kn);
4096 static struct filterops vfsread_filtops =
4097 	{ 1, NULL, filt_vfsdetach, filt_vfsread };
4098 static struct filterops vfswrite_filtops =
4099 	{ 1, NULL, filt_vfsdetach, filt_vfswrite };
4100 static struct filterops vfsvnode_filtops =
4101 	{ 1, NULL, filt_vfsdetach, filt_vfsvnode };
4102 
4103 static void
4104 vfs_knllock(void *arg)
4105 {
4106 	struct vnode *vp = arg;
4107 
4108 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4109 }
4110 
4111 static void
4112 vfs_knlunlock(void *arg)
4113 {
4114 	struct vnode *vp = arg;
4115 
4116 	VOP_UNLOCK(vp, 0);
4117 }
4118 
4119 static int
4120 vfs_knllocked(void *arg)
4121 {
4122 	struct vnode *vp = arg;
4123 
4124 	return (VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
4125 }
4126 
4127 int
4128 vfs_kqfilter(struct vop_kqfilter_args *ap)
4129 {
4130 	struct vnode *vp = ap->a_vp;
4131 	struct knote *kn = ap->a_kn;
4132 	struct knlist *knl;
4133 
4134 	switch (kn->kn_filter) {
4135 	case EVFILT_READ:
4136 		kn->kn_fop = &vfsread_filtops;
4137 		break;
4138 	case EVFILT_WRITE:
4139 		kn->kn_fop = &vfswrite_filtops;
4140 		break;
4141 	case EVFILT_VNODE:
4142 		kn->kn_fop = &vfsvnode_filtops;
4143 		break;
4144 	default:
4145 		return (EINVAL);
4146 	}
4147 
4148 	kn->kn_hook = (caddr_t)vp;
4149 
4150 	v_addpollinfo(vp);
4151 	if (vp->v_pollinfo == NULL)
4152 		return (ENOMEM);
4153 	knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4154 	knlist_add(knl, kn, 0);
4155 
4156 	return (0);
4157 }
4158 
4159 /*
4160  * Detach knote from vnode
4161  */
4162 static void
4163 filt_vfsdetach(struct knote *kn)
4164 {
4165 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4166 
4167 	KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4168 	knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4169 }
4170 
4171 /*ARGSUSED*/
4172 static int
4173 filt_vfsread(struct knote *kn, long hint)
4174 {
4175 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4176 	struct vattr va;
4177 
4178 	/*
4179 	 * filesystem is gone, so set the EOF flag and schedule
4180 	 * the knote for deletion.
4181 	 */
4182 	if (hint == NOTE_REVOKE) {
4183 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4184 		return (1);
4185 	}
4186 
4187 	if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4188 		return (0);
4189 
4190 	kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4191 	return (kn->kn_data != 0);
4192 }
4193 
4194 /*ARGSUSED*/
4195 static int
4196 filt_vfswrite(struct knote *kn, long hint)
4197 {
4198 	/*
4199 	 * filesystem is gone, so set the EOF flag and schedule
4200 	 * the knote for deletion.
4201 	 */
4202 	if (hint == NOTE_REVOKE)
4203 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4204 
4205 	kn->kn_data = 0;
4206 	return (1);
4207 }
4208 
4209 static int
4210 filt_vfsvnode(struct knote *kn, long hint)
4211 {
4212 	if (kn->kn_sfflags & hint)
4213 		kn->kn_fflags |= hint;
4214 	if (hint == NOTE_REVOKE) {
4215 		kn->kn_flags |= EV_EOF;
4216 		return (1);
4217 	}
4218 	return (kn->kn_fflags != 0);
4219 }
4220 
4221 int
4222 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4223 {
4224 	int error;
4225 
4226 	if (dp->d_reclen > ap->a_uio->uio_resid)
4227 		return (ENAMETOOLONG);
4228 	error = uiomove(dp, dp->d_reclen, ap->a_uio);
4229 	if (error) {
4230 		if (ap->a_ncookies != NULL) {
4231 			if (ap->a_cookies != NULL)
4232 				free(ap->a_cookies, M_TEMP);
4233 			ap->a_cookies = NULL;
4234 			*ap->a_ncookies = 0;
4235 		}
4236 		return (error);
4237 	}
4238 	if (ap->a_ncookies == NULL)
4239 		return (0);
4240 
4241 	KASSERT(ap->a_cookies,
4242 	    ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4243 
4244 	*ap->a_cookies = realloc(*ap->a_cookies,
4245 	    (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4246 	(*ap->a_cookies)[*ap->a_ncookies] = off;
4247 	return (0);
4248 }
4249 
4250 /*
4251  * Mark for update the access time of the file if the filesystem
4252  * supports VOP_MARKATIME.  This functionality is used by execve and
4253  * mmap, so we want to avoid the I/O implied by directly setting
4254  * va_atime for the sake of efficiency.
4255  */
4256 void
4257 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4258 {
4259 
4260 	if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4261 		(void)VOP_MARKATIME(vp);
4262 }
4263