xref: /freebsd/sys/kern/vfs_subr.c (revision ae2cbf4c649fecd3302a3bea16672345582d2562)
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	vdestroy(struct vnode *);
112 static void	vgonel(struct vnode *);
113 static void	vfs_knllock(void *arg);
114 static void	vfs_knlunlock(void *arg);
115 static int	vfs_knllocked(void *arg);
116 
117 
118 /*
119  * Enable Giant pushdown based on whether or not the vm is mpsafe in this
120  * build.  Without mpsafevm the buffer cache can not run Giant free.
121  */
122 int mpsafe_vfs = 1;
123 TUNABLE_INT("debug.mpsafevfs", &mpsafe_vfs);
124 SYSCTL_INT(_debug, OID_AUTO, mpsafevfs, CTLFLAG_RD, &mpsafe_vfs, 0,
125     "MPSAFE VFS");
126 
127 /*
128  * Number of vnodes in existence.  Increased whenever getnewvnode()
129  * allocates a new vnode, decreased on vdestroy() called on VI_DOOMed
130  * vnode.
131  */
132 static unsigned long	numvnodes;
133 
134 SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
135 
136 /*
137  * Conversion tables for conversion from vnode types to inode formats
138  * and back.
139  */
140 enum vtype iftovt_tab[16] = {
141 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
142 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
143 };
144 int vttoif_tab[10] = {
145 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
146 	S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
147 };
148 
149 /*
150  * List of vnodes that are ready for recycling.
151  */
152 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
153 
154 /*
155  * Free vnode target.  Free vnodes may simply be files which have been stat'd
156  * but not read.  This is somewhat common, and a small cache of such files
157  * should be kept to avoid recreation costs.
158  */
159 static u_long wantfreevnodes;
160 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
161 /* Number of vnodes in the free list. */
162 static u_long freevnodes;
163 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
164 
165 /*
166  * Various variables used for debugging the new implementation of
167  * reassignbuf().
168  * XXX these are probably of (very) limited utility now.
169  */
170 static int reassignbufcalls;
171 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
172 
173 /*
174  * Cache for the mount type id assigned to NFS.  This is used for
175  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
176  */
177 int	nfs_mount_type = -1;
178 
179 /* To keep more than one thread at a time from running vfs_getnewfsid */
180 static struct mtx mntid_mtx;
181 
182 /*
183  * Lock for any access to the following:
184  *	vnode_free_list
185  *	numvnodes
186  *	freevnodes
187  */
188 static struct mtx vnode_free_list_mtx;
189 
190 /* Publicly exported FS */
191 struct nfs_public nfs_pub;
192 
193 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
194 static uma_zone_t vnode_zone;
195 static uma_zone_t vnodepoll_zone;
196 
197 /* Set to 1 to print out reclaim of active vnodes */
198 int	prtactive;
199 
200 /*
201  * The workitem queue.
202  *
203  * It is useful to delay writes of file data and filesystem metadata
204  * for tens of seconds so that quickly created and deleted files need
205  * not waste disk bandwidth being created and removed. To realize this,
206  * we append vnodes to a "workitem" queue. When running with a soft
207  * updates implementation, most pending metadata dependencies should
208  * not wait for more than a few seconds. Thus, mounted on block devices
209  * are delayed only about a half the time that file data is delayed.
210  * Similarly, directory updates are more critical, so are only delayed
211  * about a third the time that file data is delayed. Thus, there are
212  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
213  * one each second (driven off the filesystem syncer process). The
214  * syncer_delayno variable indicates the next queue that is to be processed.
215  * Items that need to be processed soon are placed in this queue:
216  *
217  *	syncer_workitem_pending[syncer_delayno]
218  *
219  * A delay of fifteen seconds is done by placing the request fifteen
220  * entries later in the queue:
221  *
222  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
223  *
224  */
225 static int syncer_delayno;
226 static long syncer_mask;
227 LIST_HEAD(synclist, bufobj);
228 static struct synclist *syncer_workitem_pending[2];
229 /*
230  * The sync_mtx protects:
231  *	bo->bo_synclist
232  *	sync_vnode_count
233  *	syncer_delayno
234  *	syncer_state
235  *	syncer_workitem_pending
236  *	syncer_worklist_len
237  *	rushjob
238  */
239 static struct mtx sync_mtx;
240 static struct cv sync_wakeup;
241 
242 #define SYNCER_MAXDELAY		32
243 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
244 static int syncdelay = 30;		/* max time to delay syncing data */
245 static int filedelay = 30;		/* time to delay syncing files */
246 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
247 static int dirdelay = 29;		/* time to delay syncing directories */
248 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
249 static int metadelay = 28;		/* time to delay syncing metadata */
250 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
251 static int rushjob;		/* number of slots to run ASAP */
252 static int stat_rush_requests;	/* number of times I/O speeded up */
253 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
254 
255 /*
256  * When shutting down the syncer, run it at four times normal speed.
257  */
258 #define SYNCER_SHUTDOWN_SPEEDUP		4
259 static int sync_vnode_count;
260 static int syncer_worklist_len;
261 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
262     syncer_state;
263 
264 /*
265  * Number of vnodes we want to exist at any one time.  This is mostly used
266  * to size hash tables in vnode-related code.  It is normally not used in
267  * getnewvnode(), as wantfreevnodes is normally nonzero.)
268  *
269  * XXX desiredvnodes is historical cruft and should not exist.
270  */
271 int desiredvnodes;
272 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
273     &desiredvnodes, 0, "Maximum number of vnodes");
274 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
275     &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
276 static int vnlru_nowhere;
277 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
278     &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
279 
280 /*
281  * Macros to control when a vnode is freed and recycled.  All require
282  * the vnode interlock.
283  */
284 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
285 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
286 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
287 
288 
289 /*
290  * Initialize the vnode management data structures.
291  */
292 #ifndef	MAXVNODES_MAX
293 #define	MAXVNODES_MAX	100000
294 #endif
295 static void
296 vntblinit(void *dummy __unused)
297 {
298 
299 	/*
300 	 * Desiredvnodes is a function of the physical memory size and
301 	 * the kernel's heap size.  Specifically, desiredvnodes scales
302 	 * in proportion to the physical memory size until two fifths
303 	 * of the kernel's heap size is consumed by vnodes and vm
304 	 * objects.
305 	 */
306 	desiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * vm_kmem_size /
307 	    (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
308 	if (desiredvnodes > MAXVNODES_MAX) {
309 		if (bootverbose)
310 			printf("Reducing kern.maxvnodes %d -> %d\n",
311 			    desiredvnodes, MAXVNODES_MAX);
312 		desiredvnodes = MAXVNODES_MAX;
313 	}
314 	wantfreevnodes = desiredvnodes / 4;
315 	mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
316 	TAILQ_INIT(&vnode_free_list);
317 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
318 	vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
319 	    NULL, NULL, UMA_ALIGN_PTR, 0);
320 	vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
321 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
322 	/*
323 	 * Initialize the filesystem syncer.
324 	 */
325 	syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE,
326 	    &syncer_mask);
327 	syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE,
328 	    &syncer_mask);
329 	syncer_maxdelay = syncer_mask + 1;
330 	mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
331 	cv_init(&sync_wakeup, "syncer");
332 }
333 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
334 
335 
336 /*
337  * Mark a mount point as busy. Used to synchronize access and to delay
338  * unmounting. Interlock is not released on failure.
339  */
340 int
341 vfs_busy(struct mount *mp, int flags, struct mtx *interlkp)
342 {
343 	int lkflags;
344 
345 	MNT_ILOCK(mp);
346 	MNT_REF(mp);
347 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
348 		if (flags & LK_NOWAIT) {
349 			MNT_REL(mp);
350 			MNT_IUNLOCK(mp);
351 			return (ENOENT);
352 		}
353 		if (interlkp)
354 			mtx_unlock(interlkp);
355 		mp->mnt_kern_flag |= MNTK_MWAIT;
356 		/*
357 		 * Since all busy locks are shared except the exclusive
358 		 * lock granted when unmounting, the only place that a
359 		 * wakeup needs to be done is at the release of the
360 		 * exclusive lock at the end of dounmount.
361 		 */
362 		msleep(mp, MNT_MTX(mp), PVFS, "vfs_busy", 0);
363 		MNT_REL(mp);
364 		MNT_IUNLOCK(mp);
365 		if (interlkp)
366 			mtx_lock(interlkp);
367 		return (ENOENT);
368 	}
369 	if (interlkp)
370 		mtx_unlock(interlkp);
371 	lkflags = LK_SHARED | LK_INTERLOCK;
372 	if (lockmgr(&mp->mnt_lock, lkflags, MNT_MTX(mp)))
373 		panic("vfs_busy: unexpected lock failure");
374 	return (0);
375 }
376 
377 /*
378  * Free a busy filesystem.
379  */
380 void
381 vfs_unbusy(struct mount *mp)
382 {
383 
384 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL);
385 	vfs_rel(mp);
386 }
387 
388 /*
389  * Lookup a mount point by filesystem identifier.
390  */
391 struct mount *
392 vfs_getvfs(fsid_t *fsid)
393 {
394 	struct mount *mp;
395 
396 	mtx_lock(&mountlist_mtx);
397 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
398 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
399 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
400 			vfs_ref(mp);
401 			mtx_unlock(&mountlist_mtx);
402 			return (mp);
403 		}
404 	}
405 	mtx_unlock(&mountlist_mtx);
406 	return ((struct mount *) 0);
407 }
408 
409 /*
410  * Check if a user can access privileged mount options.
411  */
412 int
413 vfs_suser(struct mount *mp, struct thread *td)
414 {
415 	int error;
416 
417 	/*
418 	 * If the thread is jailed, but this is not a jail-friendly file
419 	 * system, deny immediately.
420 	 */
421 	if (jailed(td->td_ucred) && !(mp->mnt_vfc->vfc_flags & VFCF_JAIL))
422 		return (EPERM);
423 
424 	/*
425 	 * If the file system was mounted outside a jail and a jailed thread
426 	 * tries to access it, deny immediately.
427 	 */
428 	if (!jailed(mp->mnt_cred) && jailed(td->td_ucred))
429 		return (EPERM);
430 
431 	/*
432 	 * If the file system was mounted inside different jail that the jail of
433 	 * the calling thread, deny immediately.
434 	 */
435 	if (jailed(mp->mnt_cred) && jailed(td->td_ucred) &&
436 	    mp->mnt_cred->cr_prison != td->td_ucred->cr_prison) {
437 		return (EPERM);
438 	}
439 
440 	if ((mp->mnt_flag & MNT_USER) == 0 ||
441 	    mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
442 		if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
443 			return (error);
444 	}
445 	return (0);
446 }
447 
448 /*
449  * Get a new unique fsid.  Try to make its val[0] unique, since this value
450  * will be used to create fake device numbers for stat().  Also try (but
451  * not so hard) make its val[0] unique mod 2^16, since some emulators only
452  * support 16-bit device numbers.  We end up with unique val[0]'s for the
453  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
454  *
455  * Keep in mind that several mounts may be running in parallel.  Starting
456  * the search one past where the previous search terminated is both a
457  * micro-optimization and a defense against returning the same fsid to
458  * different mounts.
459  */
460 void
461 vfs_getnewfsid(struct mount *mp)
462 {
463 	static u_int16_t mntid_base;
464 	struct mount *nmp;
465 	fsid_t tfsid;
466 	int mtype;
467 
468 	mtx_lock(&mntid_mtx);
469 	mtype = mp->mnt_vfc->vfc_typenum;
470 	tfsid.val[1] = mtype;
471 	mtype = (mtype & 0xFF) << 24;
472 	for (;;) {
473 		tfsid.val[0] = makedev(255,
474 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
475 		mntid_base++;
476 		if ((nmp = vfs_getvfs(&tfsid)) == NULL)
477 			break;
478 		vfs_rel(nmp);
479 	}
480 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
481 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
482 	mtx_unlock(&mntid_mtx);
483 }
484 
485 /*
486  * Knob to control the precision of file timestamps:
487  *
488  *   0 = seconds only; nanoseconds zeroed.
489  *   1 = seconds and nanoseconds, accurate within 1/HZ.
490  *   2 = seconds and nanoseconds, truncated to microseconds.
491  * >=3 = seconds and nanoseconds, maximum precision.
492  */
493 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
494 
495 static int timestamp_precision = TSP_SEC;
496 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
497     &timestamp_precision, 0, "");
498 
499 /*
500  * Get a current timestamp.
501  */
502 void
503 vfs_timestamp(struct timespec *tsp)
504 {
505 	struct timeval tv;
506 
507 	switch (timestamp_precision) {
508 	case TSP_SEC:
509 		tsp->tv_sec = time_second;
510 		tsp->tv_nsec = 0;
511 		break;
512 	case TSP_HZ:
513 		getnanotime(tsp);
514 		break;
515 	case TSP_USEC:
516 		microtime(&tv);
517 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
518 		break;
519 	case TSP_NSEC:
520 	default:
521 		nanotime(tsp);
522 		break;
523 	}
524 }
525 
526 /*
527  * Set vnode attributes to VNOVAL
528  */
529 void
530 vattr_null(struct vattr *vap)
531 {
532 
533 	vap->va_type = VNON;
534 	vap->va_size = VNOVAL;
535 	vap->va_bytes = VNOVAL;
536 	vap->va_mode = VNOVAL;
537 	vap->va_nlink = VNOVAL;
538 	vap->va_uid = VNOVAL;
539 	vap->va_gid = VNOVAL;
540 	vap->va_fsid = VNOVAL;
541 	vap->va_fileid = VNOVAL;
542 	vap->va_blocksize = VNOVAL;
543 	vap->va_rdev = VNOVAL;
544 	vap->va_atime.tv_sec = VNOVAL;
545 	vap->va_atime.tv_nsec = VNOVAL;
546 	vap->va_mtime.tv_sec = VNOVAL;
547 	vap->va_mtime.tv_nsec = VNOVAL;
548 	vap->va_ctime.tv_sec = VNOVAL;
549 	vap->va_ctime.tv_nsec = VNOVAL;
550 	vap->va_birthtime.tv_sec = VNOVAL;
551 	vap->va_birthtime.tv_nsec = VNOVAL;
552 	vap->va_flags = VNOVAL;
553 	vap->va_gen = VNOVAL;
554 	vap->va_vaflags = 0;
555 }
556 
557 /*
558  * This routine is called when we have too many vnodes.  It attempts
559  * to free <count> vnodes and will potentially free vnodes that still
560  * have VM backing store (VM backing store is typically the cause
561  * of a vnode blowout so we want to do this).  Therefore, this operation
562  * is not considered cheap.
563  *
564  * A number of conditions may prevent a vnode from being reclaimed.
565  * the buffer cache may have references on the vnode, a directory
566  * vnode may still have references due to the namei cache representing
567  * underlying files, or the vnode may be in active use.   It is not
568  * desireable to reuse such vnodes.  These conditions may cause the
569  * number of vnodes to reach some minimum value regardless of what
570  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
571  */
572 static int
573 vlrureclaim(struct mount *mp)
574 {
575 	struct vnode *vp;
576 	int done;
577 	int trigger;
578 	int usevnodes;
579 	int count;
580 
581 	/*
582 	 * Calculate the trigger point, don't allow user
583 	 * screwups to blow us up.   This prevents us from
584 	 * recycling vnodes with lots of resident pages.  We
585 	 * aren't trying to free memory, we are trying to
586 	 * free vnodes.
587 	 */
588 	usevnodes = desiredvnodes;
589 	if (usevnodes <= 0)
590 		usevnodes = 1;
591 	trigger = cnt.v_page_count * 2 / usevnodes;
592 	done = 0;
593 	vn_start_write(NULL, &mp, V_WAIT);
594 	MNT_ILOCK(mp);
595 	count = mp->mnt_nvnodelistsize / 10 + 1;
596 	while (count != 0) {
597 		vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
598 		while (vp != NULL && vp->v_type == VMARKER)
599 			vp = TAILQ_NEXT(vp, v_nmntvnodes);
600 		if (vp == NULL)
601 			break;
602 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
603 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
604 		--count;
605 		if (!VI_TRYLOCK(vp))
606 			goto next_iter;
607 		/*
608 		 * If it's been deconstructed already, it's still
609 		 * referenced, or it exceeds the trigger, skip it.
610 		 */
611 		if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) ||
612 		    (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
613 		    vp->v_object->resident_page_count > trigger)) {
614 			VI_UNLOCK(vp);
615 			goto next_iter;
616 		}
617 		MNT_IUNLOCK(mp);
618 		vholdl(vp);
619 		if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
620 			vdrop(vp);
621 			goto next_iter_mntunlocked;
622 		}
623 		VI_LOCK(vp);
624 		/*
625 		 * v_usecount may have been bumped after VOP_LOCK() dropped
626 		 * the vnode interlock and before it was locked again.
627 		 *
628 		 * It is not necessary to recheck VI_DOOMED because it can
629 		 * only be set by another thread that holds both the vnode
630 		 * lock and vnode interlock.  If another thread has the
631 		 * vnode lock before we get to VOP_LOCK() and obtains the
632 		 * vnode interlock after VOP_LOCK() drops the vnode
633 		 * interlock, the other thread will be unable to drop the
634 		 * vnode lock before our VOP_LOCK() call fails.
635 		 */
636 		if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) ||
637 		    (vp->v_object != NULL &&
638 		    vp->v_object->resident_page_count > trigger)) {
639 			VOP_UNLOCK(vp, LK_INTERLOCK);
640 			goto next_iter_mntunlocked;
641 		}
642 		KASSERT((vp->v_iflag & VI_DOOMED) == 0,
643 		    ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
644 		vgonel(vp);
645 		VOP_UNLOCK(vp, 0);
646 		vdropl(vp);
647 		done++;
648 next_iter_mntunlocked:
649 		if ((count % 256) != 0)
650 			goto relock_mnt;
651 		goto yield;
652 next_iter:
653 		if ((count % 256) != 0)
654 			continue;
655 		MNT_IUNLOCK(mp);
656 yield:
657 		uio_yield();
658 relock_mnt:
659 		MNT_ILOCK(mp);
660 	}
661 	MNT_IUNLOCK(mp);
662 	vn_finished_write(mp);
663 	return done;
664 }
665 
666 /*
667  * Attempt to keep the free list at wantfreevnodes length.
668  */
669 static void
670 vnlru_free(int count)
671 {
672 	struct vnode *vp;
673 	int vfslocked;
674 
675 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
676 	for (; count > 0; count--) {
677 		vp = TAILQ_FIRST(&vnode_free_list);
678 		/*
679 		 * The list can be modified while the free_list_mtx
680 		 * has been dropped and vp could be NULL here.
681 		 */
682 		if (!vp)
683 			break;
684 		VNASSERT(vp->v_op != NULL, vp,
685 		    ("vnlru_free: vnode already reclaimed."));
686 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
687 		/*
688 		 * Don't recycle if we can't get the interlock.
689 		 */
690 		if (!VI_TRYLOCK(vp)) {
691 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
692 			continue;
693 		}
694 		VNASSERT(VCANRECYCLE(vp), vp,
695 		    ("vp inconsistent on freelist"));
696 		freevnodes--;
697 		vp->v_iflag &= ~VI_FREE;
698 		vholdl(vp);
699 		mtx_unlock(&vnode_free_list_mtx);
700 		VI_UNLOCK(vp);
701 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
702 		vtryrecycle(vp);
703 		VFS_UNLOCK_GIANT(vfslocked);
704 		/*
705 		 * If the recycled succeeded this vdrop will actually free
706 		 * the vnode.  If not it will simply place it back on
707 		 * the free list.
708 		 */
709 		vdrop(vp);
710 		mtx_lock(&vnode_free_list_mtx);
711 	}
712 }
713 /*
714  * Attempt to recycle vnodes in a context that is always safe to block.
715  * Calling vlrurecycle() from the bowels of filesystem code has some
716  * interesting deadlock problems.
717  */
718 static struct proc *vnlruproc;
719 static int vnlruproc_sig;
720 
721 static void
722 vnlru_proc(void)
723 {
724 	struct mount *mp, *nmp;
725 	int done;
726 	struct proc *p = vnlruproc;
727 
728 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
729 	    SHUTDOWN_PRI_FIRST);
730 
731 	mtx_lock(&Giant);
732 
733 	for (;;) {
734 		kproc_suspend_check(p);
735 		mtx_lock(&vnode_free_list_mtx);
736 		if (freevnodes > wantfreevnodes)
737 			vnlru_free(freevnodes - wantfreevnodes);
738 		if (numvnodes <= desiredvnodes * 9 / 10) {
739 			vnlruproc_sig = 0;
740 			wakeup(&vnlruproc_sig);
741 			msleep(vnlruproc, &vnode_free_list_mtx,
742 			    PVFS|PDROP, "vlruwt", hz);
743 			continue;
744 		}
745 		mtx_unlock(&vnode_free_list_mtx);
746 		done = 0;
747 		mtx_lock(&mountlist_mtx);
748 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
749 			int vfsunlocked;
750 			if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx)) {
751 				nmp = TAILQ_NEXT(mp, mnt_list);
752 				continue;
753 			}
754 			if (!VFS_NEEDSGIANT(mp)) {
755 				mtx_unlock(&Giant);
756 				vfsunlocked = 1;
757 			} else
758 				vfsunlocked = 0;
759 			done += vlrureclaim(mp);
760 			if (vfsunlocked)
761 				mtx_lock(&Giant);
762 			mtx_lock(&mountlist_mtx);
763 			nmp = TAILQ_NEXT(mp, mnt_list);
764 			vfs_unbusy(mp);
765 		}
766 		mtx_unlock(&mountlist_mtx);
767 		if (done == 0) {
768 			EVENTHANDLER_INVOKE(vfs_lowvnodes, desiredvnodes / 10);
769 #if 0
770 			/* These messages are temporary debugging aids */
771 			if (vnlru_nowhere < 5)
772 				printf("vnlru process getting nowhere..\n");
773 			else if (vnlru_nowhere == 5)
774 				printf("vnlru process messages stopped.\n");
775 #endif
776 			vnlru_nowhere++;
777 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
778 		} else
779 			uio_yield();
780 	}
781 }
782 
783 static struct kproc_desc vnlru_kp = {
784 	"vnlru",
785 	vnlru_proc,
786 	&vnlruproc
787 };
788 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
789     &vnlru_kp);
790 
791 /*
792  * Routines having to do with the management of the vnode table.
793  */
794 
795 static void
796 vdestroy(struct vnode *vp)
797 {
798 	struct bufobj *bo;
799 
800 	CTR1(KTR_VFS, "vdestroy vp %p", vp);
801 	mtx_lock(&vnode_free_list_mtx);
802 	numvnodes--;
803 	mtx_unlock(&vnode_free_list_mtx);
804 	bo = &vp->v_bufobj;
805 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
806 	    ("cleaned vnode still on the free list."));
807 	VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
808 	VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
809 	VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
810 	VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
811 	VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
812 	VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
813 	VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
814 	VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
815 	VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
816 	VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
817 	VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
818 	VI_UNLOCK(vp);
819 #ifdef MAC
820 	mac_vnode_destroy(vp);
821 #endif
822 	if (vp->v_pollinfo != NULL) {
823 		knlist_destroy(&vp->v_pollinfo->vpi_selinfo.si_note);
824 		mtx_destroy(&vp->v_pollinfo->vpi_lock);
825 		uma_zfree(vnodepoll_zone, vp->v_pollinfo);
826 	}
827 #ifdef INVARIANTS
828 	/* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */
829 	vp->v_op = NULL;
830 #endif
831 	lockdestroy(vp->v_vnlock);
832 	mtx_destroy(&vp->v_interlock);
833 	mtx_destroy(BO_MTX(bo));
834 	uma_zfree(vnode_zone, vp);
835 }
836 
837 /*
838  * Try to recycle a freed vnode.  We abort if anyone picks up a reference
839  * before we actually vgone().  This function must be called with the vnode
840  * held to prevent the vnode from being returned to the free list midway
841  * through vgone().
842  */
843 static int
844 vtryrecycle(struct vnode *vp)
845 {
846 	struct mount *vnmp;
847 
848 	CTR1(KTR_VFS, "vtryrecycle: trying vp %p", vp);
849 	VNASSERT(vp->v_holdcnt, vp,
850 	    ("vtryrecycle: Recycling vp %p without a reference.", vp));
851 	/*
852 	 * This vnode may found and locked via some other list, if so we
853 	 * can't recycle it yet.
854 	 */
855 	if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0)
856 		return (EWOULDBLOCK);
857 	/*
858 	 * Don't recycle if its filesystem is being suspended.
859 	 */
860 	if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
861 		VOP_UNLOCK(vp, 0);
862 		return (EBUSY);
863 	}
864 	/*
865 	 * If we got this far, we need to acquire the interlock and see if
866 	 * anyone picked up this vnode from another list.  If not, we will
867 	 * mark it with DOOMED via vgonel() so that anyone who does find it
868 	 * will skip over it.
869 	 */
870 	VI_LOCK(vp);
871 	if (vp->v_usecount) {
872 		VOP_UNLOCK(vp, LK_INTERLOCK);
873 		vn_finished_write(vnmp);
874 		return (EBUSY);
875 	}
876 	if ((vp->v_iflag & VI_DOOMED) == 0)
877 		vgonel(vp);
878 	VOP_UNLOCK(vp, LK_INTERLOCK);
879 	vn_finished_write(vnmp);
880 	CTR1(KTR_VFS, "vtryrecycle: recycled vp %p", vp);
881 	return (0);
882 }
883 
884 /*
885  * Return the next vnode from the free list.
886  */
887 int
888 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
889     struct vnode **vpp)
890 {
891 	struct vnode *vp = NULL;
892 	struct bufobj *bo;
893 
894 	mtx_lock(&vnode_free_list_mtx);
895 	/*
896 	 * Lend our context to reclaim vnodes if they've exceeded the max.
897 	 */
898 	if (freevnodes > wantfreevnodes)
899 		vnlru_free(1);
900 	/*
901 	 * Wait for available vnodes.
902 	 */
903 	if (numvnodes > desiredvnodes) {
904 		if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) {
905 			/*
906 			 * File system is beeing suspended, we cannot risk a
907 			 * deadlock here, so allocate new vnode anyway.
908 			 */
909 			if (freevnodes > wantfreevnodes)
910 				vnlru_free(freevnodes - wantfreevnodes);
911 			goto alloc;
912 		}
913 		if (vnlruproc_sig == 0) {
914 			vnlruproc_sig = 1;	/* avoid unnecessary wakeups */
915 			wakeup(vnlruproc);
916 		}
917 		msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
918 		    "vlruwk", hz);
919 #if 0	/* XXX Not all VFS_VGET/ffs_vget callers check returns. */
920 		if (numvnodes > desiredvnodes) {
921 			mtx_unlock(&vnode_free_list_mtx);
922 			return (ENFILE);
923 		}
924 #endif
925 	}
926 alloc:
927 	numvnodes++;
928 	mtx_unlock(&vnode_free_list_mtx);
929 	vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
930 	/*
931 	 * Setup locks.
932 	 */
933 	vp->v_vnlock = &vp->v_lock;
934 	mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
935 	/*
936 	 * By default, don't allow shared locks unless filesystems
937 	 * opt-in.
938 	 */
939 	lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
940 	/*
941 	 * Initialize bufobj.
942 	 */
943 	bo = &vp->v_bufobj;
944 	bo->__bo_vnode = vp;
945 	mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
946 	bo->bo_ops = &buf_ops_bio;
947 	bo->bo_private = vp;
948 	TAILQ_INIT(&bo->bo_clean.bv_hd);
949 	TAILQ_INIT(&bo->bo_dirty.bv_hd);
950 	/*
951 	 * Initialize namecache.
952 	 */
953 	LIST_INIT(&vp->v_cache_src);
954 	TAILQ_INIT(&vp->v_cache_dst);
955 	/*
956 	 * Finalize various vnode identity bits.
957 	 */
958 	vp->v_type = VNON;
959 	vp->v_tag = tag;
960 	vp->v_op = vops;
961 	v_incr_usecount(vp);
962 	vp->v_data = 0;
963 #ifdef MAC
964 	mac_vnode_init(vp);
965 	if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
966 		mac_vnode_associate_singlelabel(mp, vp);
967 	else if (mp == NULL && vops != &dead_vnodeops)
968 		printf("NULL mp in getnewvnode()\n");
969 #endif
970 	if (mp != NULL) {
971 		bo->bo_bsize = mp->mnt_stat.f_iosize;
972 		if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
973 			vp->v_vflag |= VV_NOKNOTE;
974 	}
975 
976 	CTR2(KTR_VFS, "getnewvnode: mp %p vp %p", mp, vp);
977 	*vpp = vp;
978 	return (0);
979 }
980 
981 /*
982  * Delete from old mount point vnode list, if on one.
983  */
984 static void
985 delmntque(struct vnode *vp)
986 {
987 	struct mount *mp;
988 
989 	mp = vp->v_mount;
990 	if (mp == NULL)
991 		return;
992 	MNT_ILOCK(mp);
993 	vp->v_mount = NULL;
994 	VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
995 		("bad mount point vnode list size"));
996 	TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
997 	mp->mnt_nvnodelistsize--;
998 	MNT_REL(mp);
999 	MNT_IUNLOCK(mp);
1000 }
1001 
1002 static void
1003 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1004 {
1005 
1006 	vp->v_data = NULL;
1007 	vp->v_op = &dead_vnodeops;
1008 	/* XXX non mp-safe fs may still call insmntque with vnode
1009 	   unlocked */
1010 	if (!VOP_ISLOCKED(vp))
1011 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1012 	vgone(vp);
1013 	vput(vp);
1014 }
1015 
1016 /*
1017  * Insert into list of vnodes for the new mount point, if available.
1018  */
1019 int
1020 insmntque1(struct vnode *vp, struct mount *mp,
1021 	void (*dtr)(struct vnode *, void *), void *dtr_arg)
1022 {
1023 	int locked;
1024 
1025 	KASSERT(vp->v_mount == NULL,
1026 		("insmntque: vnode already on per mount vnode list"));
1027 	VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1028 #ifdef DEBUG_VFS_LOCKS
1029 	if (!VFS_NEEDSGIANT(mp))
1030 		ASSERT_VOP_ELOCKED(vp,
1031 		    "insmntque: mp-safe fs and non-locked vp");
1032 #endif
1033 	MNT_ILOCK(mp);
1034 	if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1035 	    ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1036 	     mp->mnt_nvnodelistsize == 0)) {
1037 		locked = VOP_ISLOCKED(vp);
1038 		if (!locked || (locked == LK_EXCLUSIVE &&
1039 		     (vp->v_vflag & VV_FORCEINSMQ) == 0)) {
1040 			MNT_IUNLOCK(mp);
1041 			if (dtr != NULL)
1042 				dtr(vp, dtr_arg);
1043 			return (EBUSY);
1044 		}
1045 	}
1046 	vp->v_mount = mp;
1047 	MNT_REF(mp);
1048 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1049 	VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1050 		("neg mount point vnode list size"));
1051 	mp->mnt_nvnodelistsize++;
1052 	MNT_IUNLOCK(mp);
1053 	return (0);
1054 }
1055 
1056 int
1057 insmntque(struct vnode *vp, struct mount *mp)
1058 {
1059 
1060 	return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1061 }
1062 
1063 /*
1064  * Flush out and invalidate all buffers associated with a bufobj
1065  * Called with the underlying object locked.
1066  */
1067 int
1068 bufobj_invalbuf(struct bufobj *bo, int flags, struct thread *td, int slpflag,
1069     int slptimeo)
1070 {
1071 	int error;
1072 
1073 	BO_LOCK(bo);
1074 	if (flags & V_SAVE) {
1075 		error = bufobj_wwait(bo, slpflag, slptimeo);
1076 		if (error) {
1077 			BO_UNLOCK(bo);
1078 			return (error);
1079 		}
1080 		if (bo->bo_dirty.bv_cnt > 0) {
1081 			BO_UNLOCK(bo);
1082 			if ((error = BO_SYNC(bo, MNT_WAIT, td)) != 0)
1083 				return (error);
1084 			/*
1085 			 * XXX We could save a lock/unlock if this was only
1086 			 * enabled under INVARIANTS
1087 			 */
1088 			BO_LOCK(bo);
1089 			if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1090 				panic("vinvalbuf: dirty bufs");
1091 		}
1092 	}
1093 	/*
1094 	 * If you alter this loop please notice that interlock is dropped and
1095 	 * reacquired in flushbuflist.  Special care is needed to ensure that
1096 	 * no race conditions occur from this.
1097 	 */
1098 	do {
1099 		error = flushbuflist(&bo->bo_clean,
1100 		    flags, bo, slpflag, slptimeo);
1101 		if (error == 0)
1102 			error = flushbuflist(&bo->bo_dirty,
1103 			    flags, bo, slpflag, slptimeo);
1104 		if (error != 0 && error != EAGAIN) {
1105 			BO_UNLOCK(bo);
1106 			return (error);
1107 		}
1108 	} while (error != 0);
1109 
1110 	/*
1111 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1112 	 * have write I/O in-progress but if there is a VM object then the
1113 	 * VM object can also have read-I/O in-progress.
1114 	 */
1115 	do {
1116 		bufobj_wwait(bo, 0, 0);
1117 		BO_UNLOCK(bo);
1118 		if (bo->bo_object != NULL) {
1119 			VM_OBJECT_LOCK(bo->bo_object);
1120 			vm_object_pip_wait(bo->bo_object, "bovlbx");
1121 			VM_OBJECT_UNLOCK(bo->bo_object);
1122 		}
1123 		BO_LOCK(bo);
1124 	} while (bo->bo_numoutput > 0);
1125 	BO_UNLOCK(bo);
1126 
1127 	/*
1128 	 * Destroy the copy in the VM cache, too.
1129 	 */
1130 	if (bo->bo_object != NULL) {
1131 		VM_OBJECT_LOCK(bo->bo_object);
1132 		vm_object_page_remove(bo->bo_object, 0, 0,
1133 			(flags & V_SAVE) ? TRUE : FALSE);
1134 		VM_OBJECT_UNLOCK(bo->bo_object);
1135 	}
1136 
1137 #ifdef INVARIANTS
1138 	BO_LOCK(bo);
1139 	if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1140 	    (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1141 		panic("vinvalbuf: flush failed");
1142 	BO_UNLOCK(bo);
1143 #endif
1144 	return (0);
1145 }
1146 
1147 /*
1148  * Flush out and invalidate all buffers associated with a vnode.
1149  * Called with the underlying object locked.
1150  */
1151 int
1152 vinvalbuf(struct vnode *vp, int flags, struct thread *td, int slpflag,
1153     int slptimeo)
1154 {
1155 
1156 	CTR2(KTR_VFS, "vinvalbuf vp %p flags %d", vp, flags);
1157 	ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1158 	return (bufobj_invalbuf(&vp->v_bufobj, flags, td, slpflag, slptimeo));
1159 }
1160 
1161 /*
1162  * Flush out buffers on the specified list.
1163  *
1164  */
1165 static int
1166 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1167     int slptimeo)
1168 {
1169 	struct buf *bp, *nbp;
1170 	int retval, error;
1171 	daddr_t lblkno;
1172 	b_xflags_t xflags;
1173 
1174 	ASSERT_BO_LOCKED(bo);
1175 
1176 	retval = 0;
1177 	TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1178 		if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1179 		    ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1180 			continue;
1181 		}
1182 		lblkno = 0;
1183 		xflags = 0;
1184 		if (nbp != NULL) {
1185 			lblkno = nbp->b_lblkno;
1186 			xflags = nbp->b_xflags &
1187 				(BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1188 		}
1189 		retval = EAGAIN;
1190 		error = BUF_TIMELOCK(bp,
1191 		    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1192 		    "flushbuf", slpflag, slptimeo);
1193 		if (error) {
1194 			BO_LOCK(bo);
1195 			return (error != ENOLCK ? error : EAGAIN);
1196 		}
1197 		KASSERT(bp->b_bufobj == bo,
1198 		    ("bp %p wrong b_bufobj %p should be %p",
1199 		    bp, bp->b_bufobj, bo));
1200 		if (bp->b_bufobj != bo) {	/* XXX: necessary ? */
1201 			BUF_UNLOCK(bp);
1202 			BO_LOCK(bo);
1203 			return (EAGAIN);
1204 		}
1205 		/*
1206 		 * XXX Since there are no node locks for NFS, I
1207 		 * believe there is a slight chance that a delayed
1208 		 * write will occur while sleeping just above, so
1209 		 * check for it.
1210 		 */
1211 		if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1212 		    (flags & V_SAVE)) {
1213 			bremfree(bp);
1214 			bp->b_flags |= B_ASYNC;
1215 			bwrite(bp);
1216 			BO_LOCK(bo);
1217 			return (EAGAIN);	/* XXX: why not loop ? */
1218 		}
1219 		bremfree(bp);
1220 		bp->b_flags |= (B_INVAL | B_RELBUF);
1221 		bp->b_flags &= ~B_ASYNC;
1222 		brelse(bp);
1223 		BO_LOCK(bo);
1224 		if (nbp != NULL &&
1225 		    (nbp->b_bufobj != bo ||
1226 		     nbp->b_lblkno != lblkno ||
1227 		     (nbp->b_xflags &
1228 		      (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1229 			break;			/* nbp invalid */
1230 	}
1231 	return (retval);
1232 }
1233 
1234 /*
1235  * Truncate a file's buffer and pages to a specified length.  This
1236  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1237  * sync activity.
1238  */
1239 int
1240 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
1241     off_t length, int blksize)
1242 {
1243 	struct buf *bp, *nbp;
1244 	int anyfreed;
1245 	int trunclbn;
1246 	struct bufobj *bo;
1247 
1248 	CTR2(KTR_VFS, "vtruncbuf vp %p length %jd", vp, length);
1249 	/*
1250 	 * Round up to the *next* lbn.
1251 	 */
1252 	trunclbn = (length + blksize - 1) / blksize;
1253 
1254 	ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1255 restart:
1256 	bo = &vp->v_bufobj;
1257 	BO_LOCK(bo);
1258 	anyfreed = 1;
1259 	for (;anyfreed;) {
1260 		anyfreed = 0;
1261 		TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1262 			if (bp->b_lblkno < trunclbn)
1263 				continue;
1264 			if (BUF_LOCK(bp,
1265 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1266 			    BO_MTX(bo)) == ENOLCK)
1267 				goto restart;
1268 
1269 			bremfree(bp);
1270 			bp->b_flags |= (B_INVAL | B_RELBUF);
1271 			bp->b_flags &= ~B_ASYNC;
1272 			brelse(bp);
1273 			anyfreed = 1;
1274 
1275 			if (nbp != NULL &&
1276 			    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1277 			    (nbp->b_vp != vp) ||
1278 			    (nbp->b_flags & B_DELWRI))) {
1279 				goto restart;
1280 			}
1281 			BO_LOCK(bo);
1282 		}
1283 
1284 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1285 			if (bp->b_lblkno < trunclbn)
1286 				continue;
1287 			if (BUF_LOCK(bp,
1288 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1289 			    BO_MTX(bo)) == ENOLCK)
1290 				goto restart;
1291 			bremfree(bp);
1292 			bp->b_flags |= (B_INVAL | B_RELBUF);
1293 			bp->b_flags &= ~B_ASYNC;
1294 			brelse(bp);
1295 			anyfreed = 1;
1296 			if (nbp != NULL &&
1297 			    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1298 			    (nbp->b_vp != vp) ||
1299 			    (nbp->b_flags & B_DELWRI) == 0)) {
1300 				goto restart;
1301 			}
1302 			BO_LOCK(bo);
1303 		}
1304 	}
1305 
1306 	if (length > 0) {
1307 restartsync:
1308 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1309 			if (bp->b_lblkno > 0)
1310 				continue;
1311 			/*
1312 			 * Since we hold the vnode lock this should only
1313 			 * fail if we're racing with the buf daemon.
1314 			 */
1315 			if (BUF_LOCK(bp,
1316 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1317 			    BO_MTX(bo)) == ENOLCK) {
1318 				goto restart;
1319 			}
1320 			VNASSERT((bp->b_flags & B_DELWRI), vp,
1321 			    ("buf(%p) on dirty queue without DELWRI", bp));
1322 
1323 			bremfree(bp);
1324 			bawrite(bp);
1325 			BO_LOCK(bo);
1326 			goto restartsync;
1327 		}
1328 	}
1329 
1330 	bufobj_wwait(bo, 0, 0);
1331 	BO_UNLOCK(bo);
1332 	vnode_pager_setsize(vp, length);
1333 
1334 	return (0);
1335 }
1336 
1337 /*
1338  * buf_splay() - splay tree core for the clean/dirty list of buffers in
1339  * 		 a vnode.
1340  *
1341  *	NOTE: We have to deal with the special case of a background bitmap
1342  *	buffer, a situation where two buffers will have the same logical
1343  *	block offset.  We want (1) only the foreground buffer to be accessed
1344  *	in a lookup and (2) must differentiate between the foreground and
1345  *	background buffer in the splay tree algorithm because the splay
1346  *	tree cannot normally handle multiple entities with the same 'index'.
1347  *	We accomplish this by adding differentiating flags to the splay tree's
1348  *	numerical domain.
1349  */
1350 static
1351 struct buf *
1352 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1353 {
1354 	struct buf dummy;
1355 	struct buf *lefttreemax, *righttreemin, *y;
1356 
1357 	if (root == NULL)
1358 		return (NULL);
1359 	lefttreemax = righttreemin = &dummy;
1360 	for (;;) {
1361 		if (lblkno < root->b_lblkno ||
1362 		    (lblkno == root->b_lblkno &&
1363 		    (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1364 			if ((y = root->b_left) == NULL)
1365 				break;
1366 			if (lblkno < y->b_lblkno) {
1367 				/* Rotate right. */
1368 				root->b_left = y->b_right;
1369 				y->b_right = root;
1370 				root = y;
1371 				if ((y = root->b_left) == NULL)
1372 					break;
1373 			}
1374 			/* Link into the new root's right tree. */
1375 			righttreemin->b_left = root;
1376 			righttreemin = root;
1377 		} else if (lblkno > root->b_lblkno ||
1378 		    (lblkno == root->b_lblkno &&
1379 		    (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1380 			if ((y = root->b_right) == NULL)
1381 				break;
1382 			if (lblkno > y->b_lblkno) {
1383 				/* Rotate left. */
1384 				root->b_right = y->b_left;
1385 				y->b_left = root;
1386 				root = y;
1387 				if ((y = root->b_right) == NULL)
1388 					break;
1389 			}
1390 			/* Link into the new root's left tree. */
1391 			lefttreemax->b_right = root;
1392 			lefttreemax = root;
1393 		} else {
1394 			break;
1395 		}
1396 		root = y;
1397 	}
1398 	/* Assemble the new root. */
1399 	lefttreemax->b_right = root->b_left;
1400 	righttreemin->b_left = root->b_right;
1401 	root->b_left = dummy.b_right;
1402 	root->b_right = dummy.b_left;
1403 	return (root);
1404 }
1405 
1406 static void
1407 buf_vlist_remove(struct buf *bp)
1408 {
1409 	struct buf *root;
1410 	struct bufv *bv;
1411 
1412 	KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1413 	ASSERT_BO_LOCKED(bp->b_bufobj);
1414 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1415 	    (BX_VNDIRTY|BX_VNCLEAN),
1416 	    ("buf_vlist_remove: Buf %p is on two lists", bp));
1417 	if (bp->b_xflags & BX_VNDIRTY)
1418 		bv = &bp->b_bufobj->bo_dirty;
1419 	else
1420 		bv = &bp->b_bufobj->bo_clean;
1421 	if (bp != bv->bv_root) {
1422 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1423 		KASSERT(root == bp, ("splay lookup failed in remove"));
1424 	}
1425 	if (bp->b_left == NULL) {
1426 		root = bp->b_right;
1427 	} else {
1428 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1429 		root->b_right = bp->b_right;
1430 	}
1431 	bv->bv_root = root;
1432 	TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1433 	bv->bv_cnt--;
1434 	bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1435 }
1436 
1437 /*
1438  * Add the buffer to the sorted clean or dirty block list using a
1439  * splay tree algorithm.
1440  *
1441  * NOTE: xflags is passed as a constant, optimizing this inline function!
1442  */
1443 static void
1444 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1445 {
1446 	struct buf *root;
1447 	struct bufv *bv;
1448 
1449 	ASSERT_BO_LOCKED(bo);
1450 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1451 	    ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1452 	bp->b_xflags |= xflags;
1453 	if (xflags & BX_VNDIRTY)
1454 		bv = &bo->bo_dirty;
1455 	else
1456 		bv = &bo->bo_clean;
1457 
1458 	root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1459 	if (root == NULL) {
1460 		bp->b_left = NULL;
1461 		bp->b_right = NULL;
1462 		TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1463 	} else if (bp->b_lblkno < root->b_lblkno ||
1464 	    (bp->b_lblkno == root->b_lblkno &&
1465 	    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1466 		bp->b_left = root->b_left;
1467 		bp->b_right = root;
1468 		root->b_left = NULL;
1469 		TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1470 	} else {
1471 		bp->b_right = root->b_right;
1472 		bp->b_left = root;
1473 		root->b_right = NULL;
1474 		TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1475 	}
1476 	bv->bv_cnt++;
1477 	bv->bv_root = bp;
1478 }
1479 
1480 /*
1481  * Lookup a buffer using the splay tree.  Note that we specifically avoid
1482  * shadow buffers used in background bitmap writes.
1483  *
1484  * This code isn't quite efficient as it could be because we are maintaining
1485  * two sorted lists and do not know which list the block resides in.
1486  *
1487  * During a "make buildworld" the desired buffer is found at one of
1488  * the roots more than 60% of the time.  Thus, checking both roots
1489  * before performing either splay eliminates unnecessary splays on the
1490  * first tree splayed.
1491  */
1492 struct buf *
1493 gbincore(struct bufobj *bo, daddr_t lblkno)
1494 {
1495 	struct buf *bp;
1496 
1497 	ASSERT_BO_LOCKED(bo);
1498 	if ((bp = bo->bo_clean.bv_root) != NULL &&
1499 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1500 		return (bp);
1501 	if ((bp = bo->bo_dirty.bv_root) != NULL &&
1502 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1503 		return (bp);
1504 	if ((bp = bo->bo_clean.bv_root) != NULL) {
1505 		bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1506 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1507 			return (bp);
1508 	}
1509 	if ((bp = bo->bo_dirty.bv_root) != NULL) {
1510 		bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1511 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1512 			return (bp);
1513 	}
1514 	return (NULL);
1515 }
1516 
1517 /*
1518  * Associate a buffer with a vnode.
1519  */
1520 void
1521 bgetvp(struct vnode *vp, struct buf *bp)
1522 {
1523 	struct bufobj *bo;
1524 
1525 	bo = &vp->v_bufobj;
1526 	ASSERT_BO_LOCKED(bo);
1527 	VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1528 
1529 	CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1530 	VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1531 	    ("bgetvp: bp already attached! %p", bp));
1532 
1533 	vhold(vp);
1534 	if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT)
1535 		bp->b_flags |= B_NEEDSGIANT;
1536 	bp->b_vp = vp;
1537 	bp->b_bufobj = bo;
1538 	/*
1539 	 * Insert onto list for new vnode.
1540 	 */
1541 	buf_vlist_add(bp, bo, BX_VNCLEAN);
1542 }
1543 
1544 /*
1545  * Disassociate a buffer from a vnode.
1546  */
1547 void
1548 brelvp(struct buf *bp)
1549 {
1550 	struct bufobj *bo;
1551 	struct vnode *vp;
1552 
1553 	CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1554 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1555 
1556 	/*
1557 	 * Delete from old vnode list, if on one.
1558 	 */
1559 	vp = bp->b_vp;		/* XXX */
1560 	bo = bp->b_bufobj;
1561 	BO_LOCK(bo);
1562 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1563 		buf_vlist_remove(bp);
1564 	else
1565 		panic("brelvp: Buffer %p not on queue.", bp);
1566 	if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1567 		bo->bo_flag &= ~BO_ONWORKLST;
1568 		mtx_lock(&sync_mtx);
1569 		LIST_REMOVE(bo, bo_synclist);
1570 		syncer_worklist_len--;
1571 		mtx_unlock(&sync_mtx);
1572 	}
1573 	bp->b_flags &= ~B_NEEDSGIANT;
1574 	bp->b_vp = NULL;
1575 	bp->b_bufobj = NULL;
1576 	BO_UNLOCK(bo);
1577 	vdrop(vp);
1578 }
1579 
1580 /*
1581  * Add an item to the syncer work queue.
1582  */
1583 static void
1584 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1585 {
1586 	int queue, slot;
1587 
1588 	ASSERT_BO_LOCKED(bo);
1589 
1590 	mtx_lock(&sync_mtx);
1591 	if (bo->bo_flag & BO_ONWORKLST)
1592 		LIST_REMOVE(bo, bo_synclist);
1593 	else {
1594 		bo->bo_flag |= BO_ONWORKLST;
1595 		syncer_worklist_len++;
1596 	}
1597 
1598 	if (delay > syncer_maxdelay - 2)
1599 		delay = syncer_maxdelay - 2;
1600 	slot = (syncer_delayno + delay) & syncer_mask;
1601 
1602 	queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ :
1603 	    WI_MPSAFEQ;
1604 	LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo,
1605 	    bo_synclist);
1606 	mtx_unlock(&sync_mtx);
1607 }
1608 
1609 static int
1610 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1611 {
1612 	int error, len;
1613 
1614 	mtx_lock(&sync_mtx);
1615 	len = syncer_worklist_len - sync_vnode_count;
1616 	mtx_unlock(&sync_mtx);
1617 	error = SYSCTL_OUT(req, &len, sizeof(len));
1618 	return (error);
1619 }
1620 
1621 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1622     sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1623 
1624 static struct proc *updateproc;
1625 static void sched_sync(void);
1626 static struct kproc_desc up_kp = {
1627 	"syncer",
1628 	sched_sync,
1629 	&updateproc
1630 };
1631 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1632 
1633 static int
1634 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1635 {
1636 	struct vnode *vp;
1637 	struct mount *mp;
1638 
1639 	*bo = LIST_FIRST(slp);
1640 	if (*bo == NULL)
1641 		return (0);
1642 	vp = (*bo)->__bo_vnode;	/* XXX */
1643 	if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1644 		return (1);
1645 	/*
1646 	 * We use vhold in case the vnode does not
1647 	 * successfully sync.  vhold prevents the vnode from
1648 	 * going away when we unlock the sync_mtx so that
1649 	 * we can acquire the vnode interlock.
1650 	 */
1651 	vholdl(vp);
1652 	mtx_unlock(&sync_mtx);
1653 	VI_UNLOCK(vp);
1654 	if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1655 		vdrop(vp);
1656 		mtx_lock(&sync_mtx);
1657 		return (*bo == LIST_FIRST(slp));
1658 	}
1659 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1660 	(void) VOP_FSYNC(vp, MNT_LAZY, td);
1661 	VOP_UNLOCK(vp, 0);
1662 	vn_finished_write(mp);
1663 	BO_LOCK(*bo);
1664 	if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1665 		/*
1666 		 * Put us back on the worklist.  The worklist
1667 		 * routine will remove us from our current
1668 		 * position and then add us back in at a later
1669 		 * position.
1670 		 */
1671 		vn_syncer_add_to_worklist(*bo, syncdelay);
1672 	}
1673 	BO_UNLOCK(*bo);
1674 	vdrop(vp);
1675 	mtx_lock(&sync_mtx);
1676 	return (0);
1677 }
1678 
1679 /*
1680  * System filesystem synchronizer daemon.
1681  */
1682 static void
1683 sched_sync(void)
1684 {
1685 	struct synclist *gnext, *next;
1686 	struct synclist *gslp, *slp;
1687 	struct bufobj *bo;
1688 	long starttime;
1689 	struct thread *td = curthread;
1690 	int last_work_seen;
1691 	int net_worklist_len;
1692 	int syncer_final_iter;
1693 	int first_printf;
1694 	int error;
1695 
1696 	last_work_seen = 0;
1697 	syncer_final_iter = 0;
1698 	first_printf = 1;
1699 	syncer_state = SYNCER_RUNNING;
1700 	starttime = time_uptime;
1701 	td->td_pflags |= TDP_NORUNNINGBUF;
1702 
1703 	EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1704 	    SHUTDOWN_PRI_LAST);
1705 
1706 	mtx_lock(&sync_mtx);
1707 	for (;;) {
1708 		if (syncer_state == SYNCER_FINAL_DELAY &&
1709 		    syncer_final_iter == 0) {
1710 			mtx_unlock(&sync_mtx);
1711 			kproc_suspend_check(td->td_proc);
1712 			mtx_lock(&sync_mtx);
1713 		}
1714 		net_worklist_len = syncer_worklist_len - sync_vnode_count;
1715 		if (syncer_state != SYNCER_RUNNING &&
1716 		    starttime != time_uptime) {
1717 			if (first_printf) {
1718 				printf("\nSyncing disks, vnodes remaining...");
1719 				first_printf = 0;
1720 			}
1721 			printf("%d ", net_worklist_len);
1722 		}
1723 		starttime = time_uptime;
1724 
1725 		/*
1726 		 * Push files whose dirty time has expired.  Be careful
1727 		 * of interrupt race on slp queue.
1728 		 *
1729 		 * Skip over empty worklist slots when shutting down.
1730 		 */
1731 		do {
1732 			slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1733 			gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1734 			syncer_delayno += 1;
1735 			if (syncer_delayno == syncer_maxdelay)
1736 				syncer_delayno = 0;
1737 			next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1738 			gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1739 			/*
1740 			 * If the worklist has wrapped since the
1741 			 * it was emptied of all but syncer vnodes,
1742 			 * switch to the FINAL_DELAY state and run
1743 			 * for one more second.
1744 			 */
1745 			if (syncer_state == SYNCER_SHUTTING_DOWN &&
1746 			    net_worklist_len == 0 &&
1747 			    last_work_seen == syncer_delayno) {
1748 				syncer_state = SYNCER_FINAL_DELAY;
1749 				syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1750 			}
1751 		} while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1752 		    LIST_EMPTY(gslp) && syncer_worklist_len > 0);
1753 
1754 		/*
1755 		 * Keep track of the last time there was anything
1756 		 * on the worklist other than syncer vnodes.
1757 		 * Return to the SHUTTING_DOWN state if any
1758 		 * new work appears.
1759 		 */
1760 		if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1761 			last_work_seen = syncer_delayno;
1762 		if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1763 			syncer_state = SYNCER_SHUTTING_DOWN;
1764 		while (!LIST_EMPTY(slp)) {
1765 			error = sync_vnode(slp, &bo, td);
1766 			if (error == 1) {
1767 				LIST_REMOVE(bo, bo_synclist);
1768 				LIST_INSERT_HEAD(next, bo, bo_synclist);
1769 				continue;
1770 			}
1771 		}
1772 		if (!LIST_EMPTY(gslp)) {
1773 			mtx_unlock(&sync_mtx);
1774 			mtx_lock(&Giant);
1775 			mtx_lock(&sync_mtx);
1776 			while (!LIST_EMPTY(gslp)) {
1777 				error = sync_vnode(gslp, &bo, td);
1778 				if (error == 1) {
1779 					LIST_REMOVE(bo, bo_synclist);
1780 					LIST_INSERT_HEAD(gnext, bo,
1781 					    bo_synclist);
1782 					continue;
1783 				}
1784 			}
1785 			mtx_unlock(&Giant);
1786 		}
1787 		if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1788 			syncer_final_iter--;
1789 		/*
1790 		 * The variable rushjob allows the kernel to speed up the
1791 		 * processing of the filesystem syncer process. A rushjob
1792 		 * value of N tells the filesystem syncer to process the next
1793 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1794 		 * is used by the soft update code to speed up the filesystem
1795 		 * syncer process when the incore state is getting so far
1796 		 * ahead of the disk that the kernel memory pool is being
1797 		 * threatened with exhaustion.
1798 		 */
1799 		if (rushjob > 0) {
1800 			rushjob -= 1;
1801 			continue;
1802 		}
1803 		/*
1804 		 * Just sleep for a short period of time between
1805 		 * iterations when shutting down to allow some I/O
1806 		 * to happen.
1807 		 *
1808 		 * If it has taken us less than a second to process the
1809 		 * current work, then wait. Otherwise start right over
1810 		 * again. We can still lose time if any single round
1811 		 * takes more than two seconds, but it does not really
1812 		 * matter as we are just trying to generally pace the
1813 		 * filesystem activity.
1814 		 */
1815 		if (syncer_state != SYNCER_RUNNING)
1816 			cv_timedwait(&sync_wakeup, &sync_mtx,
1817 			    hz / SYNCER_SHUTDOWN_SPEEDUP);
1818 		else if (time_uptime == starttime)
1819 			cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1820 	}
1821 }
1822 
1823 /*
1824  * Request the syncer daemon to speed up its work.
1825  * We never push it to speed up more than half of its
1826  * normal turn time, otherwise it could take over the cpu.
1827  */
1828 int
1829 speedup_syncer(void)
1830 {
1831 	int ret = 0;
1832 
1833 	mtx_lock(&sync_mtx);
1834 	if (rushjob < syncdelay / 2) {
1835 		rushjob += 1;
1836 		stat_rush_requests += 1;
1837 		ret = 1;
1838 	}
1839 	mtx_unlock(&sync_mtx);
1840 	cv_broadcast(&sync_wakeup);
1841 	return (ret);
1842 }
1843 
1844 /*
1845  * Tell the syncer to speed up its work and run though its work
1846  * list several times, then tell it to shut down.
1847  */
1848 static void
1849 syncer_shutdown(void *arg, int howto)
1850 {
1851 
1852 	if (howto & RB_NOSYNC)
1853 		return;
1854 	mtx_lock(&sync_mtx);
1855 	syncer_state = SYNCER_SHUTTING_DOWN;
1856 	rushjob = 0;
1857 	mtx_unlock(&sync_mtx);
1858 	cv_broadcast(&sync_wakeup);
1859 	kproc_shutdown(arg, howto);
1860 }
1861 
1862 /*
1863  * Reassign a buffer from one vnode to another.
1864  * Used to assign file specific control information
1865  * (indirect blocks) to the vnode to which they belong.
1866  */
1867 void
1868 reassignbuf(struct buf *bp)
1869 {
1870 	struct vnode *vp;
1871 	struct bufobj *bo;
1872 	int delay;
1873 #ifdef INVARIANTS
1874 	struct bufv *bv;
1875 #endif
1876 
1877 	vp = bp->b_vp;
1878 	bo = bp->b_bufobj;
1879 	++reassignbufcalls;
1880 
1881 	CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1882 	    bp, bp->b_vp, bp->b_flags);
1883 	/*
1884 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1885 	 * is not fully linked in.
1886 	 */
1887 	if (bp->b_flags & B_PAGING)
1888 		panic("cannot reassign paging buffer");
1889 
1890 	/*
1891 	 * Delete from old vnode list, if on one.
1892 	 */
1893 	BO_LOCK(bo);
1894 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1895 		buf_vlist_remove(bp);
1896 	else
1897 		panic("reassignbuf: Buffer %p not on queue.", bp);
1898 	/*
1899 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1900 	 * of clean buffers.
1901 	 */
1902 	if (bp->b_flags & B_DELWRI) {
1903 		if ((bo->bo_flag & BO_ONWORKLST) == 0) {
1904 			switch (vp->v_type) {
1905 			case VDIR:
1906 				delay = dirdelay;
1907 				break;
1908 			case VCHR:
1909 				delay = metadelay;
1910 				break;
1911 			default:
1912 				delay = filedelay;
1913 			}
1914 			vn_syncer_add_to_worklist(bo, delay);
1915 		}
1916 		buf_vlist_add(bp, bo, BX_VNDIRTY);
1917 	} else {
1918 		buf_vlist_add(bp, bo, BX_VNCLEAN);
1919 
1920 		if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1921 			mtx_lock(&sync_mtx);
1922 			LIST_REMOVE(bo, bo_synclist);
1923 			syncer_worklist_len--;
1924 			mtx_unlock(&sync_mtx);
1925 			bo->bo_flag &= ~BO_ONWORKLST;
1926 		}
1927 	}
1928 #ifdef INVARIANTS
1929 	bv = &bo->bo_clean;
1930 	bp = TAILQ_FIRST(&bv->bv_hd);
1931 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1932 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1933 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
1934 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1935 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1936 	bv = &bo->bo_dirty;
1937 	bp = TAILQ_FIRST(&bv->bv_hd);
1938 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1939 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1940 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
1941 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1942 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1943 #endif
1944 	BO_UNLOCK(bo);
1945 }
1946 
1947 /*
1948  * Increment the use and hold counts on the vnode, taking care to reference
1949  * the driver's usecount if this is a chardev.  The vholdl() will remove
1950  * the vnode from the free list if it is presently free.  Requires the
1951  * vnode interlock and returns with it held.
1952  */
1953 static void
1954 v_incr_usecount(struct vnode *vp)
1955 {
1956 
1957 	CTR3(KTR_VFS, "v_incr_usecount: vp %p holdcnt %d usecount %d\n",
1958 	    vp, vp->v_holdcnt, vp->v_usecount);
1959 	vp->v_usecount++;
1960 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1961 		dev_lock();
1962 		vp->v_rdev->si_usecount++;
1963 		dev_unlock();
1964 	}
1965 	vholdl(vp);
1966 }
1967 
1968 /*
1969  * Turn a holdcnt into a use+holdcnt such that only one call to
1970  * v_decr_usecount is needed.
1971  */
1972 static void
1973 v_upgrade_usecount(struct vnode *vp)
1974 {
1975 
1976 	CTR3(KTR_VFS, "v_upgrade_usecount: vp %p holdcnt %d usecount %d\n",
1977 	    vp, vp->v_holdcnt, vp->v_usecount);
1978 	vp->v_usecount++;
1979 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1980 		dev_lock();
1981 		vp->v_rdev->si_usecount++;
1982 		dev_unlock();
1983 	}
1984 }
1985 
1986 /*
1987  * Decrement the vnode use and hold count along with the driver's usecount
1988  * if this is a chardev.  The vdropl() below releases the vnode interlock
1989  * as it may free the vnode.
1990  */
1991 static void
1992 v_decr_usecount(struct vnode *vp)
1993 {
1994 
1995 	CTR3(KTR_VFS, "v_decr_usecount: vp %p holdcnt %d usecount %d\n",
1996 	    vp, vp->v_holdcnt, vp->v_usecount);
1997 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
1998 	VNASSERT(vp->v_usecount > 0, vp,
1999 	    ("v_decr_usecount: negative usecount"));
2000 	vp->v_usecount--;
2001 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2002 		dev_lock();
2003 		vp->v_rdev->si_usecount--;
2004 		dev_unlock();
2005 	}
2006 	vdropl(vp);
2007 }
2008 
2009 /*
2010  * Decrement only the use count and driver use count.  This is intended to
2011  * be paired with a follow on vdropl() to release the remaining hold count.
2012  * In this way we may vgone() a vnode with a 0 usecount without risk of
2013  * having it end up on a free list because the hold count is kept above 0.
2014  */
2015 static void
2016 v_decr_useonly(struct vnode *vp)
2017 {
2018 
2019 	CTR3(KTR_VFS, "v_decr_useonly: vp %p holdcnt %d usecount %d\n",
2020 	    vp, vp->v_holdcnt, vp->v_usecount);
2021 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2022 	VNASSERT(vp->v_usecount > 0, vp,
2023 	    ("v_decr_useonly: negative usecount"));
2024 	vp->v_usecount--;
2025 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2026 		dev_lock();
2027 		vp->v_rdev->si_usecount--;
2028 		dev_unlock();
2029 	}
2030 }
2031 
2032 /*
2033  * Grab a particular vnode from the free list, increment its
2034  * reference count and lock it.  VI_DOOMED is set if the vnode
2035  * is being destroyed.  Only callers who specify LK_RETRY will
2036  * see doomed vnodes.  If inactive processing was delayed in
2037  * vput try to do it here.
2038  */
2039 int
2040 vget(struct vnode *vp, int flags, struct thread *td)
2041 {
2042 	int error;
2043 
2044 	error = 0;
2045 	VFS_ASSERT_GIANT(vp->v_mount);
2046 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2047 	    ("vget: invalid lock operation"));
2048 	if ((flags & LK_INTERLOCK) == 0)
2049 		VI_LOCK(vp);
2050 	vholdl(vp);
2051 	if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2052 		vdrop(vp);
2053 		return (error);
2054 	}
2055 	if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2056 		panic("vget: vn_lock failed to return ENOENT\n");
2057 	VI_LOCK(vp);
2058 	/* Upgrade our holdcnt to a usecount. */
2059 	v_upgrade_usecount(vp);
2060 	/*
2061  	 * We don't guarantee that any particular close will
2062 	 * trigger inactive processing so just make a best effort
2063 	 * here at preventing a reference to a removed file.  If
2064 	 * we don't succeed no harm is done.
2065 	 */
2066 	if (vp->v_iflag & VI_OWEINACT) {
2067 		if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2068 		    (flags & LK_NOWAIT) == 0)
2069 			vinactive(vp, td);
2070 		vp->v_iflag &= ~VI_OWEINACT;
2071 	}
2072 	VI_UNLOCK(vp);
2073 	return (0);
2074 }
2075 
2076 /*
2077  * Increase the reference count of a vnode.
2078  */
2079 void
2080 vref(struct vnode *vp)
2081 {
2082 
2083 	VI_LOCK(vp);
2084 	v_incr_usecount(vp);
2085 	VI_UNLOCK(vp);
2086 }
2087 
2088 /*
2089  * Return reference count of a vnode.
2090  *
2091  * The results of this call are only guaranteed when some mechanism other
2092  * than the VI lock is used to stop other processes from gaining references
2093  * to the vnode.  This may be the case if the caller holds the only reference.
2094  * This is also useful when stale data is acceptable as race conditions may
2095  * be accounted for by some other means.
2096  */
2097 int
2098 vrefcnt(struct vnode *vp)
2099 {
2100 	int usecnt;
2101 
2102 	VI_LOCK(vp);
2103 	usecnt = vp->v_usecount;
2104 	VI_UNLOCK(vp);
2105 
2106 	return (usecnt);
2107 }
2108 
2109 
2110 /*
2111  * Vnode put/release.
2112  * If count drops to zero, call inactive routine and return to freelist.
2113  */
2114 void
2115 vrele(struct vnode *vp)
2116 {
2117 	struct thread *td = curthread;	/* XXX */
2118 
2119 	KASSERT(vp != NULL, ("vrele: null vp"));
2120 	VFS_ASSERT_GIANT(vp->v_mount);
2121 
2122 	VI_LOCK(vp);
2123 
2124 	/* Skip this v_writecount check if we're going to panic below. */
2125 	VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2126 	    ("vrele: missed vn_close"));
2127 
2128 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2129 	    vp->v_usecount == 1)) {
2130 		v_decr_usecount(vp);
2131 		return;
2132 	}
2133 	if (vp->v_usecount != 1) {
2134 #ifdef DIAGNOSTIC
2135 		vprint("vrele: negative ref count", vp);
2136 #endif
2137 		VI_UNLOCK(vp);
2138 		panic("vrele: negative ref cnt");
2139 	}
2140 	/*
2141 	 * We want to hold the vnode until the inactive finishes to
2142 	 * prevent vgone() races.  We drop the use count here and the
2143 	 * hold count below when we're done.
2144 	 */
2145 	v_decr_useonly(vp);
2146 	/*
2147 	 * We must call VOP_INACTIVE with the node locked. Mark
2148 	 * as VI_DOINGINACT to avoid recursion.
2149 	 */
2150 	vp->v_iflag |= VI_OWEINACT;
2151 	if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK) == 0) {
2152 		VI_LOCK(vp);
2153 		if (vp->v_usecount > 0)
2154 			vp->v_iflag &= ~VI_OWEINACT;
2155 		if (vp->v_iflag & VI_OWEINACT)
2156 			vinactive(vp, td);
2157 		VOP_UNLOCK(vp, 0);
2158 	} else {
2159 		VI_LOCK(vp);
2160 		if (vp->v_usecount > 0)
2161 			vp->v_iflag &= ~VI_OWEINACT;
2162 	}
2163 	vdropl(vp);
2164 }
2165 
2166 /*
2167  * Release an already locked vnode.  This give the same effects as
2168  * unlock+vrele(), but takes less time and avoids releasing and
2169  * re-aquiring the lock (as vrele() acquires the lock internally.)
2170  */
2171 void
2172 vput(struct vnode *vp)
2173 {
2174 	struct thread *td = curthread;	/* XXX */
2175 	int error;
2176 
2177 	KASSERT(vp != NULL, ("vput: null vp"));
2178 	ASSERT_VOP_LOCKED(vp, "vput");
2179 	VFS_ASSERT_GIANT(vp->v_mount);
2180 	VI_LOCK(vp);
2181 	/* Skip this v_writecount check if we're going to panic below. */
2182 	VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2183 	    ("vput: missed vn_close"));
2184 	error = 0;
2185 
2186 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2187 	    vp->v_usecount == 1)) {
2188 		VOP_UNLOCK(vp, 0);
2189 		v_decr_usecount(vp);
2190 		return;
2191 	}
2192 
2193 	if (vp->v_usecount != 1) {
2194 #ifdef DIAGNOSTIC
2195 		vprint("vput: negative ref count", vp);
2196 #endif
2197 		panic("vput: negative ref cnt");
2198 	}
2199 	/*
2200 	 * We want to hold the vnode until the inactive finishes to
2201 	 * prevent vgone() races.  We drop the use count here and the
2202 	 * hold count below when we're done.
2203 	 */
2204 	v_decr_useonly(vp);
2205 	vp->v_iflag |= VI_OWEINACT;
2206 	if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2207 		error = VOP_LOCK(vp, LK_UPGRADE|LK_INTERLOCK|LK_NOWAIT);
2208 		VI_LOCK(vp);
2209 		if (error) {
2210 			if (vp->v_usecount > 0)
2211 				vp->v_iflag &= ~VI_OWEINACT;
2212 			goto done;
2213 		}
2214 	}
2215 	if (vp->v_usecount > 0)
2216 		vp->v_iflag &= ~VI_OWEINACT;
2217 	if (vp->v_iflag & VI_OWEINACT)
2218 		vinactive(vp, td);
2219 	VOP_UNLOCK(vp, 0);
2220 done:
2221 	vdropl(vp);
2222 }
2223 
2224 /*
2225  * Somebody doesn't want the vnode recycled.
2226  */
2227 void
2228 vhold(struct vnode *vp)
2229 {
2230 
2231 	VI_LOCK(vp);
2232 	vholdl(vp);
2233 	VI_UNLOCK(vp);
2234 }
2235 
2236 void
2237 vholdl(struct vnode *vp)
2238 {
2239 
2240 	vp->v_holdcnt++;
2241 	if (VSHOULDBUSY(vp))
2242 		vbusy(vp);
2243 }
2244 
2245 /*
2246  * Note that there is one less who cares about this vnode.  vdrop() is the
2247  * opposite of vhold().
2248  */
2249 void
2250 vdrop(struct vnode *vp)
2251 {
2252 
2253 	VI_LOCK(vp);
2254 	vdropl(vp);
2255 }
2256 
2257 /*
2258  * Drop the hold count of the vnode.  If this is the last reference to
2259  * the vnode we will free it if it has been vgone'd otherwise it is
2260  * placed on the free list.
2261  */
2262 void
2263 vdropl(struct vnode *vp)
2264 {
2265 
2266 	ASSERT_VI_LOCKED(vp, "vdropl");
2267 	if (vp->v_holdcnt <= 0)
2268 		panic("vdrop: holdcnt %d", vp->v_holdcnt);
2269 	vp->v_holdcnt--;
2270 	if (vp->v_holdcnt == 0) {
2271 		if (vp->v_iflag & VI_DOOMED) {
2272 			vdestroy(vp);
2273 			return;
2274 		} else
2275 			vfree(vp);
2276 	}
2277 	VI_UNLOCK(vp);
2278 }
2279 
2280 /*
2281  * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2282  * flags.  DOINGINACT prevents us from recursing in calls to vinactive.
2283  * OWEINACT tracks whether a vnode missed a call to inactive due to a
2284  * failed lock upgrade.
2285  */
2286 static void
2287 vinactive(struct vnode *vp, struct thread *td)
2288 {
2289 
2290 	ASSERT_VOP_ELOCKED(vp, "vinactive");
2291 	ASSERT_VI_LOCKED(vp, "vinactive");
2292 	VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2293 	    ("vinactive: recursed on VI_DOINGINACT"));
2294 	vp->v_iflag |= VI_DOINGINACT;
2295 	vp->v_iflag &= ~VI_OWEINACT;
2296 	VI_UNLOCK(vp);
2297 	VOP_INACTIVE(vp, td);
2298 	VI_LOCK(vp);
2299 	VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2300 	    ("vinactive: lost VI_DOINGINACT"));
2301 	vp->v_iflag &= ~VI_DOINGINACT;
2302 }
2303 
2304 /*
2305  * Remove any vnodes in the vnode table belonging to mount point mp.
2306  *
2307  * If FORCECLOSE is not specified, there should not be any active ones,
2308  * return error if any are found (nb: this is a user error, not a
2309  * system error). If FORCECLOSE is specified, detach any active vnodes
2310  * that are found.
2311  *
2312  * If WRITECLOSE is set, only flush out regular file vnodes open for
2313  * writing.
2314  *
2315  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2316  *
2317  * `rootrefs' specifies the base reference count for the root vnode
2318  * of this filesystem. The root vnode is considered busy if its
2319  * v_usecount exceeds this value. On a successful return, vflush(, td)
2320  * will call vrele() on the root vnode exactly rootrefs times.
2321  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2322  * be zero.
2323  */
2324 #ifdef DIAGNOSTIC
2325 static int busyprt = 0;		/* print out busy vnodes */
2326 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
2327 #endif
2328 
2329 int
2330 vflush( struct mount *mp, int rootrefs, int flags, struct thread *td)
2331 {
2332 	struct vnode *vp, *mvp, *rootvp = NULL;
2333 	struct vattr vattr;
2334 	int busy = 0, error;
2335 
2336 	CTR1(KTR_VFS, "vflush: mp %p", mp);
2337 	if (rootrefs > 0) {
2338 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2339 		    ("vflush: bad args"));
2340 		/*
2341 		 * Get the filesystem root vnode. We can vput() it
2342 		 * immediately, since with rootrefs > 0, it won't go away.
2343 		 */
2344 		if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp, td)) != 0)
2345 			return (error);
2346 		vput(rootvp);
2347 
2348 	}
2349 	MNT_ILOCK(mp);
2350 loop:
2351 	MNT_VNODE_FOREACH(vp, mp, mvp) {
2352 
2353 		VI_LOCK(vp);
2354 		vholdl(vp);
2355 		MNT_IUNLOCK(mp);
2356 		error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2357 		if (error) {
2358 			vdrop(vp);
2359 			MNT_ILOCK(mp);
2360 			MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
2361 			goto loop;
2362 		}
2363 		/*
2364 		 * Skip over a vnodes marked VV_SYSTEM.
2365 		 */
2366 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2367 			VOP_UNLOCK(vp, 0);
2368 			vdrop(vp);
2369 			MNT_ILOCK(mp);
2370 			continue;
2371 		}
2372 		/*
2373 		 * If WRITECLOSE is set, flush out unlinked but still open
2374 		 * files (even if open only for reading) and regular file
2375 		 * vnodes open for writing.
2376 		 */
2377 		if (flags & WRITECLOSE) {
2378 			error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2379 			VI_LOCK(vp);
2380 
2381 			if ((vp->v_type == VNON ||
2382 			    (error == 0 && vattr.va_nlink > 0)) &&
2383 			    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2384 				VOP_UNLOCK(vp, 0);
2385 				vdropl(vp);
2386 				MNT_ILOCK(mp);
2387 				continue;
2388 			}
2389 		} else
2390 			VI_LOCK(vp);
2391 		/*
2392 		 * With v_usecount == 0, all we need to do is clear out the
2393 		 * vnode data structures and we are done.
2394 		 *
2395 		 * If FORCECLOSE is set, forcibly close the vnode.
2396 		 */
2397 		if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2398 			VNASSERT(vp->v_usecount == 0 ||
2399 			    (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2400 			    ("device VNODE %p is FORCECLOSED", vp));
2401 			vgonel(vp);
2402 		} else {
2403 			busy++;
2404 #ifdef DIAGNOSTIC
2405 			if (busyprt)
2406 				vprint("vflush: busy vnode", vp);
2407 #endif
2408 		}
2409 		VOP_UNLOCK(vp, 0);
2410 		vdropl(vp);
2411 		MNT_ILOCK(mp);
2412 	}
2413 	MNT_IUNLOCK(mp);
2414 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2415 		/*
2416 		 * If just the root vnode is busy, and if its refcount
2417 		 * is equal to `rootrefs', then go ahead and kill it.
2418 		 */
2419 		VI_LOCK(rootvp);
2420 		KASSERT(busy > 0, ("vflush: not busy"));
2421 		VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2422 		    ("vflush: usecount %d < rootrefs %d",
2423 		     rootvp->v_usecount, rootrefs));
2424 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2425 			VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2426 			vgone(rootvp);
2427 			VOP_UNLOCK(rootvp, 0);
2428 			busy = 0;
2429 		} else
2430 			VI_UNLOCK(rootvp);
2431 	}
2432 	if (busy)
2433 		return (EBUSY);
2434 	for (; rootrefs > 0; rootrefs--)
2435 		vrele(rootvp);
2436 	return (0);
2437 }
2438 
2439 /*
2440  * Recycle an unused vnode to the front of the free list.
2441  */
2442 int
2443 vrecycle(struct vnode *vp, struct thread *td)
2444 {
2445 	int recycled;
2446 
2447 	ASSERT_VOP_ELOCKED(vp, "vrecycle");
2448 	recycled = 0;
2449 	VI_LOCK(vp);
2450 	if (vp->v_usecount == 0) {
2451 		recycled = 1;
2452 		vgonel(vp);
2453 	}
2454 	VI_UNLOCK(vp);
2455 	return (recycled);
2456 }
2457 
2458 /*
2459  * Eliminate all activity associated with a vnode
2460  * in preparation for reuse.
2461  */
2462 void
2463 vgone(struct vnode *vp)
2464 {
2465 	VI_LOCK(vp);
2466 	vgonel(vp);
2467 	VI_UNLOCK(vp);
2468 }
2469 
2470 /*
2471  * vgone, with the vp interlock held.
2472  */
2473 void
2474 vgonel(struct vnode *vp)
2475 {
2476 	struct thread *td;
2477 	int oweinact;
2478 	int active;
2479 	struct mount *mp;
2480 
2481 	CTR1(KTR_VFS, "vgonel: vp %p", vp);
2482 	ASSERT_VOP_ELOCKED(vp, "vgonel");
2483 	ASSERT_VI_LOCKED(vp, "vgonel");
2484 	VNASSERT(vp->v_holdcnt, vp,
2485 	    ("vgonel: vp %p has no reference.", vp));
2486 	td = curthread;
2487 
2488 	/*
2489 	 * Don't vgonel if we're already doomed.
2490 	 */
2491 	if (vp->v_iflag & VI_DOOMED)
2492 		return;
2493 	vp->v_iflag |= VI_DOOMED;
2494 	/*
2495 	 * Check to see if the vnode is in use.  If so, we have to call
2496 	 * VOP_CLOSE() and VOP_INACTIVE().
2497 	 */
2498 	active = vp->v_usecount;
2499 	oweinact = (vp->v_iflag & VI_OWEINACT);
2500 	VI_UNLOCK(vp);
2501 	/*
2502 	 * Clean out any buffers associated with the vnode.
2503 	 * If the flush fails, just toss the buffers.
2504 	 */
2505 	mp = NULL;
2506 	if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2507 		(void) vn_start_secondary_write(vp, &mp, V_WAIT);
2508 	if (vinvalbuf(vp, V_SAVE, td, 0, 0) != 0)
2509 		vinvalbuf(vp, 0, td, 0, 0);
2510 
2511 	/*
2512 	 * If purging an active vnode, it must be closed and
2513 	 * deactivated before being reclaimed.
2514 	 */
2515 	if (active)
2516 		VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2517 	if (oweinact || active) {
2518 		VI_LOCK(vp);
2519 		if ((vp->v_iflag & VI_DOINGINACT) == 0)
2520 			vinactive(vp, td);
2521 		VI_UNLOCK(vp);
2522 	}
2523 	/*
2524 	 * Reclaim the vnode.
2525 	 */
2526 	if (VOP_RECLAIM(vp, td))
2527 		panic("vgone: cannot reclaim");
2528 	if (mp != NULL)
2529 		vn_finished_secondary_write(mp);
2530 	VNASSERT(vp->v_object == NULL, vp,
2531 	    ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2532 	/*
2533 	 * Clear the advisory locks and wake up waiting threads.
2534 	 */
2535 	lf_purgelocks(vp, &(vp->v_lockf));
2536 	/*
2537 	 * Delete from old mount point vnode list.
2538 	 */
2539 	delmntque(vp);
2540 	cache_purge(vp);
2541 	/*
2542 	 * Done with purge, reset to the standard lock and invalidate
2543 	 * the vnode.
2544 	 */
2545 	VI_LOCK(vp);
2546 	vp->v_vnlock = &vp->v_lock;
2547 	vp->v_op = &dead_vnodeops;
2548 	vp->v_tag = "none";
2549 	vp->v_type = VBAD;
2550 }
2551 
2552 /*
2553  * Calculate the total number of references to a special device.
2554  */
2555 int
2556 vcount(struct vnode *vp)
2557 {
2558 	int count;
2559 
2560 	dev_lock();
2561 	count = vp->v_rdev->si_usecount;
2562 	dev_unlock();
2563 	return (count);
2564 }
2565 
2566 /*
2567  * Same as above, but using the struct cdev *as argument
2568  */
2569 int
2570 count_dev(struct cdev *dev)
2571 {
2572 	int count;
2573 
2574 	dev_lock();
2575 	count = dev->si_usecount;
2576 	dev_unlock();
2577 	return(count);
2578 }
2579 
2580 /*
2581  * Print out a description of a vnode.
2582  */
2583 static char *typename[] =
2584 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2585  "VMARKER"};
2586 
2587 void
2588 vn_printf(struct vnode *vp, const char *fmt, ...)
2589 {
2590 	va_list ap;
2591 	char buf[256], buf2[16];
2592 	u_long flags;
2593 
2594 	va_start(ap, fmt);
2595 	vprintf(fmt, ap);
2596 	va_end(ap);
2597 	printf("%p: ", (void *)vp);
2598 	printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2599 	printf("    usecount %d, writecount %d, refcount %d mountedhere %p\n",
2600 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2601 	buf[0] = '\0';
2602 	buf[1] = '\0';
2603 	if (vp->v_vflag & VV_ROOT)
2604 		strlcat(buf, "|VV_ROOT", sizeof(buf));
2605 	if (vp->v_vflag & VV_ISTTY)
2606 		strlcat(buf, "|VV_ISTTY", sizeof(buf));
2607 	if (vp->v_vflag & VV_NOSYNC)
2608 		strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2609 	if (vp->v_vflag & VV_CACHEDLABEL)
2610 		strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2611 	if (vp->v_vflag & VV_TEXT)
2612 		strlcat(buf, "|VV_TEXT", sizeof(buf));
2613 	if (vp->v_vflag & VV_COPYONWRITE)
2614 		strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2615 	if (vp->v_vflag & VV_SYSTEM)
2616 		strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2617 	if (vp->v_vflag & VV_PROCDEP)
2618 		strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2619 	if (vp->v_vflag & VV_NOKNOTE)
2620 		strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2621 	if (vp->v_vflag & VV_DELETED)
2622 		strlcat(buf, "|VV_DELETED", sizeof(buf));
2623 	if (vp->v_vflag & VV_MD)
2624 		strlcat(buf, "|VV_MD", sizeof(buf));
2625 	flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC |
2626 	    VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2627 	    VV_NOKNOTE | VV_DELETED | VV_MD);
2628 	if (flags != 0) {
2629 		snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2630 		strlcat(buf, buf2, sizeof(buf));
2631 	}
2632 	if (vp->v_iflag & VI_MOUNT)
2633 		strlcat(buf, "|VI_MOUNT", sizeof(buf));
2634 	if (vp->v_iflag & VI_AGE)
2635 		strlcat(buf, "|VI_AGE", sizeof(buf));
2636 	if (vp->v_iflag & VI_DOOMED)
2637 		strlcat(buf, "|VI_DOOMED", sizeof(buf));
2638 	if (vp->v_iflag & VI_FREE)
2639 		strlcat(buf, "|VI_FREE", sizeof(buf));
2640 	if (vp->v_iflag & VI_OBJDIRTY)
2641 		strlcat(buf, "|VI_OBJDIRTY", sizeof(buf));
2642 	if (vp->v_iflag & VI_DOINGINACT)
2643 		strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2644 	if (vp->v_iflag & VI_OWEINACT)
2645 		strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2646 	flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2647 	    VI_OBJDIRTY | VI_DOINGINACT | VI_OWEINACT);
2648 	if (flags != 0) {
2649 		snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2650 		strlcat(buf, buf2, sizeof(buf));
2651 	}
2652 	printf("    flags (%s)\n", buf + 1);
2653 	if (mtx_owned(VI_MTX(vp)))
2654 		printf(" VI_LOCKed");
2655 	if (vp->v_object != NULL)
2656 		printf("    v_object %p ref %d pages %d\n",
2657 		    vp->v_object, vp->v_object->ref_count,
2658 		    vp->v_object->resident_page_count);
2659 	printf("    ");
2660 	lockmgr_printinfo(vp->v_vnlock);
2661 	printf("\n");
2662 	if (vp->v_data != NULL)
2663 		VOP_PRINT(vp);
2664 }
2665 
2666 #ifdef DDB
2667 /*
2668  * List all of the locked vnodes in the system.
2669  * Called when debugging the kernel.
2670  */
2671 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2672 {
2673 	struct mount *mp, *nmp;
2674 	struct vnode *vp;
2675 
2676 	/*
2677 	 * Note: because this is DDB, we can't obey the locking semantics
2678 	 * for these structures, which means we could catch an inconsistent
2679 	 * state and dereference a nasty pointer.  Not much to be done
2680 	 * about that.
2681 	 */
2682 	db_printf("Locked vnodes\n");
2683 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2684 		nmp = TAILQ_NEXT(mp, mnt_list);
2685 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2686 			if (vp->v_type != VMARKER &&
2687 			    VOP_ISLOCKED(vp))
2688 				vprint("", vp);
2689 		}
2690 		nmp = TAILQ_NEXT(mp, mnt_list);
2691 	}
2692 }
2693 
2694 /*
2695  * Show details about the given vnode.
2696  */
2697 DB_SHOW_COMMAND(vnode, db_show_vnode)
2698 {
2699 	struct vnode *vp;
2700 
2701 	if (!have_addr)
2702 		return;
2703 	vp = (struct vnode *)addr;
2704 	vn_printf(vp, "vnode ");
2705 }
2706 
2707 /*
2708  * Show details about the given mount point.
2709  */
2710 DB_SHOW_COMMAND(mount, db_show_mount)
2711 {
2712 	struct mount *mp;
2713 	struct statfs *sp;
2714 	struct vnode *vp;
2715 	char buf[512];
2716 	u_int flags;
2717 
2718 	if (!have_addr) {
2719 		/* No address given, print short info about all mount points. */
2720 		TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2721 			db_printf("%p %s on %s (%s)\n", mp,
2722 			    mp->mnt_stat.f_mntfromname,
2723 			    mp->mnt_stat.f_mntonname,
2724 			    mp->mnt_stat.f_fstypename);
2725 			if (db_pager_quit)
2726 				break;
2727 		}
2728 		db_printf("\nMore info: show mount <addr>\n");
2729 		return;
2730 	}
2731 
2732 	mp = (struct mount *)addr;
2733 	db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
2734 	    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
2735 
2736 	buf[0] = '\0';
2737 	flags = mp->mnt_flag;
2738 #define	MNT_FLAG(flag)	do {						\
2739 	if (flags & (flag)) {						\
2740 		if (buf[0] != '\0')					\
2741 			strlcat(buf, ", ", sizeof(buf));		\
2742 		strlcat(buf, (#flag) + 4, sizeof(buf));			\
2743 		flags &= ~(flag);					\
2744 	}								\
2745 } while (0)
2746 	MNT_FLAG(MNT_RDONLY);
2747 	MNT_FLAG(MNT_SYNCHRONOUS);
2748 	MNT_FLAG(MNT_NOEXEC);
2749 	MNT_FLAG(MNT_NOSUID);
2750 	MNT_FLAG(MNT_UNION);
2751 	MNT_FLAG(MNT_ASYNC);
2752 	MNT_FLAG(MNT_SUIDDIR);
2753 	MNT_FLAG(MNT_SOFTDEP);
2754 	MNT_FLAG(MNT_NOSYMFOLLOW);
2755 	MNT_FLAG(MNT_GJOURNAL);
2756 	MNT_FLAG(MNT_MULTILABEL);
2757 	MNT_FLAG(MNT_ACLS);
2758 	MNT_FLAG(MNT_NOATIME);
2759 	MNT_FLAG(MNT_NOCLUSTERR);
2760 	MNT_FLAG(MNT_NOCLUSTERW);
2761 	MNT_FLAG(MNT_EXRDONLY);
2762 	MNT_FLAG(MNT_EXPORTED);
2763 	MNT_FLAG(MNT_DEFEXPORTED);
2764 	MNT_FLAG(MNT_EXPORTANON);
2765 	MNT_FLAG(MNT_EXKERB);
2766 	MNT_FLAG(MNT_EXPUBLIC);
2767 	MNT_FLAG(MNT_LOCAL);
2768 	MNT_FLAG(MNT_QUOTA);
2769 	MNT_FLAG(MNT_ROOTFS);
2770 	MNT_FLAG(MNT_USER);
2771 	MNT_FLAG(MNT_IGNORE);
2772 	MNT_FLAG(MNT_UPDATE);
2773 	MNT_FLAG(MNT_DELEXPORT);
2774 	MNT_FLAG(MNT_RELOAD);
2775 	MNT_FLAG(MNT_FORCE);
2776 	MNT_FLAG(MNT_SNAPSHOT);
2777 	MNT_FLAG(MNT_BYFSID);
2778 #undef MNT_FLAG
2779 	if (flags != 0) {
2780 		if (buf[0] != '\0')
2781 			strlcat(buf, ", ", sizeof(buf));
2782 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2783 		    "0x%08x", flags);
2784 	}
2785 	db_printf("    mnt_flag = %s\n", buf);
2786 
2787 	buf[0] = '\0';
2788 	flags = mp->mnt_kern_flag;
2789 #define	MNT_KERN_FLAG(flag)	do {					\
2790 	if (flags & (flag)) {						\
2791 		if (buf[0] != '\0')					\
2792 			strlcat(buf, ", ", sizeof(buf));		\
2793 		strlcat(buf, (#flag) + 5, sizeof(buf));			\
2794 		flags &= ~(flag);					\
2795 	}								\
2796 } while (0)
2797 	MNT_KERN_FLAG(MNTK_UNMOUNTF);
2798 	MNT_KERN_FLAG(MNTK_ASYNC);
2799 	MNT_KERN_FLAG(MNTK_SOFTDEP);
2800 	MNT_KERN_FLAG(MNTK_NOINSMNTQ);
2801 	MNT_KERN_FLAG(MNTK_UNMOUNT);
2802 	MNT_KERN_FLAG(MNTK_MWAIT);
2803 	MNT_KERN_FLAG(MNTK_SUSPEND);
2804 	MNT_KERN_FLAG(MNTK_SUSPEND2);
2805 	MNT_KERN_FLAG(MNTK_SUSPENDED);
2806 	MNT_KERN_FLAG(MNTK_MPSAFE);
2807 	MNT_KERN_FLAG(MNTK_NOKNOTE);
2808 	MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
2809 #undef MNT_KERN_FLAG
2810 	if (flags != 0) {
2811 		if (buf[0] != '\0')
2812 			strlcat(buf, ", ", sizeof(buf));
2813 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2814 		    "0x%08x", flags);
2815 	}
2816 	db_printf("    mnt_kern_flag = %s\n", buf);
2817 
2818 	sp = &mp->mnt_stat;
2819 	db_printf("    mnt_stat = { version=%u type=%u flags=0x%016jx "
2820 	    "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
2821 	    "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
2822 	    "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
2823 	    (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
2824 	    (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
2825 	    (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
2826 	    (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
2827 	    (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
2828 	    (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
2829 	    (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
2830 	    (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
2831 
2832 	db_printf("    mnt_cred = { uid=%u ruid=%u",
2833 	    (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
2834 	if (mp->mnt_cred->cr_prison != NULL)
2835 		db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
2836 	db_printf(" }\n");
2837 	db_printf("    mnt_ref = %d\n", mp->mnt_ref);
2838 	db_printf("    mnt_gen = %d\n", mp->mnt_gen);
2839 	db_printf("    mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
2840 	db_printf("    mnt_writeopcount = %d\n", mp->mnt_writeopcount);
2841 	db_printf("    mnt_noasync = %u\n", mp->mnt_noasync);
2842 	db_printf("    mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
2843 	db_printf("    mnt_iosize_max = %d\n", mp->mnt_iosize_max);
2844 	db_printf("    mnt_hashseed = %u\n", mp->mnt_hashseed);
2845 	db_printf("    mnt_markercnt = %d\n", mp->mnt_markercnt);
2846 	db_printf("    mnt_holdcnt = %d\n", mp->mnt_holdcnt);
2847 	db_printf("    mnt_holdcntwaiters = %d\n", mp->mnt_holdcntwaiters);
2848 	db_printf("    mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
2849 	db_printf("    mnt_secondary_accwrites = %d\n",
2850 	    mp->mnt_secondary_accwrites);
2851 	db_printf("    mnt_gjprovider = %s\n",
2852 	    mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
2853 	db_printf("\n");
2854 
2855 	TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2856 		if (vp->v_type != VMARKER) {
2857 			vn_printf(vp, "vnode ");
2858 			if (db_pager_quit)
2859 				break;
2860 		}
2861 	}
2862 }
2863 #endif	/* DDB */
2864 
2865 /*
2866  * Fill in a struct xvfsconf based on a struct vfsconf.
2867  */
2868 static void
2869 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2870 {
2871 
2872 	strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2873 	xvfsp->vfc_typenum = vfsp->vfc_typenum;
2874 	xvfsp->vfc_refcount = vfsp->vfc_refcount;
2875 	xvfsp->vfc_flags = vfsp->vfc_flags;
2876 	/*
2877 	 * These are unused in userland, we keep them
2878 	 * to not break binary compatibility.
2879 	 */
2880 	xvfsp->vfc_vfsops = NULL;
2881 	xvfsp->vfc_next = NULL;
2882 }
2883 
2884 /*
2885  * Top level filesystem related information gathering.
2886  */
2887 static int
2888 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2889 {
2890 	struct vfsconf *vfsp;
2891 	struct xvfsconf xvfsp;
2892 	int error;
2893 
2894 	error = 0;
2895 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2896 		bzero(&xvfsp, sizeof(xvfsp));
2897 		vfsconf2x(vfsp, &xvfsp);
2898 		error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
2899 		if (error)
2900 			break;
2901 	}
2902 	return (error);
2903 }
2904 
2905 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
2906     "S,xvfsconf", "List of all configured filesystems");
2907 
2908 #ifndef BURN_BRIDGES
2909 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2910 
2911 static int
2912 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2913 {
2914 	int *name = (int *)arg1 - 1;	/* XXX */
2915 	u_int namelen = arg2 + 1;	/* XXX */
2916 	struct vfsconf *vfsp;
2917 	struct xvfsconf xvfsp;
2918 
2919 	printf("WARNING: userland calling deprecated sysctl, "
2920 	    "please rebuild world\n");
2921 
2922 #if 1 || defined(COMPAT_PRELITE2)
2923 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2924 	if (namelen == 1)
2925 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2926 #endif
2927 
2928 	switch (name[1]) {
2929 	case VFS_MAXTYPENUM:
2930 		if (namelen != 2)
2931 			return (ENOTDIR);
2932 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2933 	case VFS_CONF:
2934 		if (namelen != 3)
2935 			return (ENOTDIR);	/* overloaded */
2936 		TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
2937 			if (vfsp->vfc_typenum == name[2])
2938 				break;
2939 		if (vfsp == NULL)
2940 			return (EOPNOTSUPP);
2941 		bzero(&xvfsp, sizeof(xvfsp));
2942 		vfsconf2x(vfsp, &xvfsp);
2943 		return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
2944 	}
2945 	return (EOPNOTSUPP);
2946 }
2947 
2948 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
2949 	vfs_sysctl, "Generic filesystem");
2950 
2951 #if 1 || defined(COMPAT_PRELITE2)
2952 
2953 static int
2954 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2955 {
2956 	int error;
2957 	struct vfsconf *vfsp;
2958 	struct ovfsconf ovfs;
2959 
2960 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2961 		bzero(&ovfs, sizeof(ovfs));
2962 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
2963 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
2964 		ovfs.vfc_index = vfsp->vfc_typenum;
2965 		ovfs.vfc_refcount = vfsp->vfc_refcount;
2966 		ovfs.vfc_flags = vfsp->vfc_flags;
2967 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2968 		if (error)
2969 			return error;
2970 	}
2971 	return 0;
2972 }
2973 
2974 #endif /* 1 || COMPAT_PRELITE2 */
2975 #endif /* !BURN_BRIDGES */
2976 
2977 #define KINFO_VNODESLOP		10
2978 #ifdef notyet
2979 /*
2980  * Dump vnode list (via sysctl).
2981  */
2982 /* ARGSUSED */
2983 static int
2984 sysctl_vnode(SYSCTL_HANDLER_ARGS)
2985 {
2986 	struct xvnode *xvn;
2987 	struct mount *mp;
2988 	struct vnode *vp;
2989 	int error, len, n;
2990 
2991 	/*
2992 	 * Stale numvnodes access is not fatal here.
2993 	 */
2994 	req->lock = 0;
2995 	len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
2996 	if (!req->oldptr)
2997 		/* Make an estimate */
2998 		return (SYSCTL_OUT(req, 0, len));
2999 
3000 	error = sysctl_wire_old_buffer(req, 0);
3001 	if (error != 0)
3002 		return (error);
3003 	xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3004 	n = 0;
3005 	mtx_lock(&mountlist_mtx);
3006 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3007 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx))
3008 			continue;
3009 		MNT_ILOCK(mp);
3010 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3011 			if (n == len)
3012 				break;
3013 			vref(vp);
3014 			xvn[n].xv_size = sizeof *xvn;
3015 			xvn[n].xv_vnode = vp;
3016 			xvn[n].xv_id = 0;	/* XXX compat */
3017 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3018 			XV_COPY(usecount);
3019 			XV_COPY(writecount);
3020 			XV_COPY(holdcnt);
3021 			XV_COPY(mount);
3022 			XV_COPY(numoutput);
3023 			XV_COPY(type);
3024 #undef XV_COPY
3025 			xvn[n].xv_flag = vp->v_vflag;
3026 
3027 			switch (vp->v_type) {
3028 			case VREG:
3029 			case VDIR:
3030 			case VLNK:
3031 				break;
3032 			case VBLK:
3033 			case VCHR:
3034 				if (vp->v_rdev == NULL) {
3035 					vrele(vp);
3036 					continue;
3037 				}
3038 				xvn[n].xv_dev = dev2udev(vp->v_rdev);
3039 				break;
3040 			case VSOCK:
3041 				xvn[n].xv_socket = vp->v_socket;
3042 				break;
3043 			case VFIFO:
3044 				xvn[n].xv_fifo = vp->v_fifoinfo;
3045 				break;
3046 			case VNON:
3047 			case VBAD:
3048 			default:
3049 				/* shouldn't happen? */
3050 				vrele(vp);
3051 				continue;
3052 			}
3053 			vrele(vp);
3054 			++n;
3055 		}
3056 		MNT_IUNLOCK(mp);
3057 		mtx_lock(&mountlist_mtx);
3058 		vfs_unbusy(mp);
3059 		if (n == len)
3060 			break;
3061 	}
3062 	mtx_unlock(&mountlist_mtx);
3063 
3064 	error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3065 	free(xvn, M_TEMP);
3066 	return (error);
3067 }
3068 
3069 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3070 	0, 0, sysctl_vnode, "S,xvnode", "");
3071 #endif
3072 
3073 /*
3074  * Unmount all filesystems. The list is traversed in reverse order
3075  * of mounting to avoid dependencies.
3076  */
3077 void
3078 vfs_unmountall(void)
3079 {
3080 	struct mount *mp;
3081 	struct thread *td;
3082 	int error;
3083 
3084 	KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
3085 	td = curthread;
3086 	/*
3087 	 * Since this only runs when rebooting, it is not interlocked.
3088 	 */
3089 	while(!TAILQ_EMPTY(&mountlist)) {
3090 		mp = TAILQ_LAST(&mountlist, mntlist);
3091 		error = dounmount(mp, MNT_FORCE, td);
3092 		if (error) {
3093 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
3094 			/*
3095 			 * XXX: Due to the way in which we mount the root
3096 			 * file system off of devfs, devfs will generate a
3097 			 * "busy" warning when we try to unmount it before
3098 			 * the root.  Don't print a warning as a result in
3099 			 * order to avoid false positive errors that may
3100 			 * cause needless upset.
3101 			 */
3102 			if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3103 				printf("unmount of %s failed (",
3104 				    mp->mnt_stat.f_mntonname);
3105 				if (error == EBUSY)
3106 					printf("BUSY)\n");
3107 				else
3108 					printf("%d)\n", error);
3109 			}
3110 		} else {
3111 			/* The unmount has removed mp from the mountlist */
3112 		}
3113 	}
3114 }
3115 
3116 /*
3117  * perform msync on all vnodes under a mount point
3118  * the mount point must be locked.
3119  */
3120 void
3121 vfs_msync(struct mount *mp, int flags)
3122 {
3123 	struct vnode *vp, *mvp;
3124 	struct vm_object *obj;
3125 
3126 	MNT_ILOCK(mp);
3127 	MNT_VNODE_FOREACH(vp, mp, mvp) {
3128 		VI_LOCK(vp);
3129 		if ((vp->v_iflag & VI_OBJDIRTY) &&
3130 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3131 			MNT_IUNLOCK(mp);
3132 			if (!vget(vp,
3133 			    LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3134 			    curthread)) {
3135 				if (vp->v_vflag & VV_NOSYNC) {	/* unlinked */
3136 					vput(vp);
3137 					MNT_ILOCK(mp);
3138 					continue;
3139 				}
3140 
3141 				obj = vp->v_object;
3142 				if (obj != NULL) {
3143 					VM_OBJECT_LOCK(obj);
3144 					vm_object_page_clean(obj, 0, 0,
3145 					    flags == MNT_WAIT ?
3146 					    OBJPC_SYNC : OBJPC_NOSYNC);
3147 					VM_OBJECT_UNLOCK(obj);
3148 				}
3149 				vput(vp);
3150 			}
3151 			MNT_ILOCK(mp);
3152 		} else
3153 			VI_UNLOCK(vp);
3154 	}
3155 	MNT_IUNLOCK(mp);
3156 }
3157 
3158 /*
3159  * Mark a vnode as free, putting it up for recycling.
3160  */
3161 static void
3162 vfree(struct vnode *vp)
3163 {
3164 
3165 	CTR1(KTR_VFS, "vfree vp %p", vp);
3166 	ASSERT_VI_LOCKED(vp, "vfree");
3167 	mtx_lock(&vnode_free_list_mtx);
3168 	VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed."));
3169 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free"));
3170 	VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't"));
3171 	VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp,
3172 	    ("vfree: Freeing doomed vnode"));
3173 	if (vp->v_iflag & VI_AGE) {
3174 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3175 	} else {
3176 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3177 	}
3178 	freevnodes++;
3179 	vp->v_iflag &= ~VI_AGE;
3180 	vp->v_iflag |= VI_FREE;
3181 	mtx_unlock(&vnode_free_list_mtx);
3182 }
3183 
3184 /*
3185  * Opposite of vfree() - mark a vnode as in use.
3186  */
3187 static void
3188 vbusy(struct vnode *vp)
3189 {
3190 	CTR1(KTR_VFS, "vbusy vp %p", vp);
3191 	ASSERT_VI_LOCKED(vp, "vbusy");
3192 	VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
3193 	VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed."));
3194 
3195 	mtx_lock(&vnode_free_list_mtx);
3196 	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3197 	freevnodes--;
3198 	vp->v_iflag &= ~(VI_FREE|VI_AGE);
3199 	mtx_unlock(&vnode_free_list_mtx);
3200 }
3201 
3202 /*
3203  * Initalize per-vnode helper structure to hold poll-related state.
3204  */
3205 void
3206 v_addpollinfo(struct vnode *vp)
3207 {
3208 	struct vpollinfo *vi;
3209 
3210 	vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3211 	if (vp->v_pollinfo != NULL) {
3212 		uma_zfree(vnodepoll_zone, vi);
3213 		return;
3214 	}
3215 	vp->v_pollinfo = vi;
3216 	mtx_init(&vp->v_pollinfo->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3217 	knlist_init(&vp->v_pollinfo->vpi_selinfo.si_note, vp, vfs_knllock,
3218 	    vfs_knlunlock, vfs_knllocked);
3219 }
3220 
3221 /*
3222  * Record a process's interest in events which might happen to
3223  * a vnode.  Because poll uses the historic select-style interface
3224  * internally, this routine serves as both the ``check for any
3225  * pending events'' and the ``record my interest in future events''
3226  * functions.  (These are done together, while the lock is held,
3227  * to avoid race conditions.)
3228  */
3229 int
3230 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3231 {
3232 
3233 	if (vp->v_pollinfo == NULL)
3234 		v_addpollinfo(vp);
3235 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3236 	if (vp->v_pollinfo->vpi_revents & events) {
3237 		/*
3238 		 * This leaves events we are not interested
3239 		 * in available for the other process which
3240 		 * which presumably had requested them
3241 		 * (otherwise they would never have been
3242 		 * recorded).
3243 		 */
3244 		events &= vp->v_pollinfo->vpi_revents;
3245 		vp->v_pollinfo->vpi_revents &= ~events;
3246 
3247 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
3248 		return events;
3249 	}
3250 	vp->v_pollinfo->vpi_events |= events;
3251 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3252 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3253 	return 0;
3254 }
3255 
3256 /*
3257  * Routine to create and manage a filesystem syncer vnode.
3258  */
3259 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
3260 static int	sync_fsync(struct  vop_fsync_args *);
3261 static int	sync_inactive(struct  vop_inactive_args *);
3262 static int	sync_reclaim(struct  vop_reclaim_args *);
3263 
3264 static struct vop_vector sync_vnodeops = {
3265 	.vop_bypass =	VOP_EOPNOTSUPP,
3266 	.vop_close =	sync_close,		/* close */
3267 	.vop_fsync =	sync_fsync,		/* fsync */
3268 	.vop_inactive =	sync_inactive,	/* inactive */
3269 	.vop_reclaim =	sync_reclaim,	/* reclaim */
3270 	.vop_lock1 =	vop_stdlock,	/* lock */
3271 	.vop_unlock =	vop_stdunlock,	/* unlock */
3272 	.vop_islocked =	vop_stdislocked,	/* islocked */
3273 };
3274 
3275 /*
3276  * Create a new filesystem syncer vnode for the specified mount point.
3277  */
3278 int
3279 vfs_allocate_syncvnode(struct mount *mp)
3280 {
3281 	struct vnode *vp;
3282 	struct bufobj *bo;
3283 	static long start, incr, next;
3284 	int error;
3285 
3286 	/* Allocate a new vnode */
3287 	if ((error = getnewvnode("syncer", mp, &sync_vnodeops, &vp)) != 0) {
3288 		mp->mnt_syncer = NULL;
3289 		return (error);
3290 	}
3291 	vp->v_type = VNON;
3292 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3293 	vp->v_vflag |= VV_FORCEINSMQ;
3294 	error = insmntque(vp, mp);
3295 	if (error != 0)
3296 		panic("vfs_allocate_syncvnode: insmntque failed");
3297 	vp->v_vflag &= ~VV_FORCEINSMQ;
3298 	VOP_UNLOCK(vp, 0);
3299 	/*
3300 	 * Place the vnode onto the syncer worklist. We attempt to
3301 	 * scatter them about on the list so that they will go off
3302 	 * at evenly distributed times even if all the filesystems
3303 	 * are mounted at once.
3304 	 */
3305 	next += incr;
3306 	if (next == 0 || next > syncer_maxdelay) {
3307 		start /= 2;
3308 		incr /= 2;
3309 		if (start == 0) {
3310 			start = syncer_maxdelay / 2;
3311 			incr = syncer_maxdelay;
3312 		}
3313 		next = start;
3314 	}
3315 	bo = &vp->v_bufobj;
3316 	BO_LOCK(bo);
3317 	vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3318 	/* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3319 	mtx_lock(&sync_mtx);
3320 	sync_vnode_count++;
3321 	mtx_unlock(&sync_mtx);
3322 	BO_UNLOCK(bo);
3323 	mp->mnt_syncer = vp;
3324 	return (0);
3325 }
3326 
3327 /*
3328  * Do a lazy sync of the filesystem.
3329  */
3330 static int
3331 sync_fsync(struct vop_fsync_args *ap)
3332 {
3333 	struct vnode *syncvp = ap->a_vp;
3334 	struct mount *mp = syncvp->v_mount;
3335 	int error;
3336 	struct bufobj *bo;
3337 
3338 	/*
3339 	 * We only need to do something if this is a lazy evaluation.
3340 	 */
3341 	if (ap->a_waitfor != MNT_LAZY)
3342 		return (0);
3343 
3344 	/*
3345 	 * Move ourselves to the back of the sync list.
3346 	 */
3347 	bo = &syncvp->v_bufobj;
3348 	BO_LOCK(bo);
3349 	vn_syncer_add_to_worklist(bo, syncdelay);
3350 	BO_UNLOCK(bo);
3351 
3352 	/*
3353 	 * Walk the list of vnodes pushing all that are dirty and
3354 	 * not already on the sync list.
3355 	 */
3356 	mtx_lock(&mountlist_mtx);
3357 	if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx) != 0) {
3358 		mtx_unlock(&mountlist_mtx);
3359 		return (0);
3360 	}
3361 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3362 		vfs_unbusy(mp);
3363 		return (0);
3364 	}
3365 	MNT_ILOCK(mp);
3366 	mp->mnt_noasync++;
3367 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
3368 	MNT_IUNLOCK(mp);
3369 	vfs_msync(mp, MNT_NOWAIT);
3370 	error = VFS_SYNC(mp, MNT_LAZY, ap->a_td);
3371 	MNT_ILOCK(mp);
3372 	mp->mnt_noasync--;
3373 	if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
3374 		mp->mnt_kern_flag |= MNTK_ASYNC;
3375 	MNT_IUNLOCK(mp);
3376 	vn_finished_write(mp);
3377 	vfs_unbusy(mp);
3378 	return (error);
3379 }
3380 
3381 /*
3382  * The syncer vnode is no referenced.
3383  */
3384 static int
3385 sync_inactive(struct vop_inactive_args *ap)
3386 {
3387 
3388 	vgone(ap->a_vp);
3389 	return (0);
3390 }
3391 
3392 /*
3393  * The syncer vnode is no longer needed and is being decommissioned.
3394  *
3395  * Modifications to the worklist must be protected by sync_mtx.
3396  */
3397 static int
3398 sync_reclaim(struct vop_reclaim_args *ap)
3399 {
3400 	struct vnode *vp = ap->a_vp;
3401 	struct bufobj *bo;
3402 
3403 	bo = &vp->v_bufobj;
3404 	BO_LOCK(bo);
3405 	vp->v_mount->mnt_syncer = NULL;
3406 	if (bo->bo_flag & BO_ONWORKLST) {
3407 		mtx_lock(&sync_mtx);
3408 		LIST_REMOVE(bo, bo_synclist);
3409 		syncer_worklist_len--;
3410 		sync_vnode_count--;
3411 		mtx_unlock(&sync_mtx);
3412 		bo->bo_flag &= ~BO_ONWORKLST;
3413 	}
3414 	BO_UNLOCK(bo);
3415 
3416 	return (0);
3417 }
3418 
3419 /*
3420  * Check if vnode represents a disk device
3421  */
3422 int
3423 vn_isdisk(struct vnode *vp, int *errp)
3424 {
3425 	int error;
3426 
3427 	error = 0;
3428 	dev_lock();
3429 	if (vp->v_type != VCHR)
3430 		error = ENOTBLK;
3431 	else if (vp->v_rdev == NULL)
3432 		error = ENXIO;
3433 	else if (vp->v_rdev->si_devsw == NULL)
3434 		error = ENXIO;
3435 	else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3436 		error = ENOTBLK;
3437 	dev_unlock();
3438 	if (errp != NULL)
3439 		*errp = error;
3440 	return (error == 0);
3441 }
3442 
3443 /*
3444  * Common filesystem object access control check routine.  Accepts a
3445  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3446  * and optional call-by-reference privused argument allowing vaccess()
3447  * to indicate to the caller whether privilege was used to satisfy the
3448  * request (obsoleted).  Returns 0 on success, or an errno on failure.
3449  *
3450  * The ifdef'd CAPABILITIES version is here for reference, but is not
3451  * actually used.
3452  */
3453 int
3454 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3455     mode_t acc_mode, struct ucred *cred, int *privused)
3456 {
3457 	mode_t dac_granted;
3458 	mode_t priv_granted;
3459 
3460 	/*
3461 	 * Look for a normal, non-privileged way to access the file/directory
3462 	 * as requested.  If it exists, go with that.
3463 	 */
3464 
3465 	if (privused != NULL)
3466 		*privused = 0;
3467 
3468 	dac_granted = 0;
3469 
3470 	/* Check the owner. */
3471 	if (cred->cr_uid == file_uid) {
3472 		dac_granted |= VADMIN;
3473 		if (file_mode & S_IXUSR)
3474 			dac_granted |= VEXEC;
3475 		if (file_mode & S_IRUSR)
3476 			dac_granted |= VREAD;
3477 		if (file_mode & S_IWUSR)
3478 			dac_granted |= (VWRITE | VAPPEND);
3479 
3480 		if ((acc_mode & dac_granted) == acc_mode)
3481 			return (0);
3482 
3483 		goto privcheck;
3484 	}
3485 
3486 	/* Otherwise, check the groups (first match) */
3487 	if (groupmember(file_gid, cred)) {
3488 		if (file_mode & S_IXGRP)
3489 			dac_granted |= VEXEC;
3490 		if (file_mode & S_IRGRP)
3491 			dac_granted |= VREAD;
3492 		if (file_mode & S_IWGRP)
3493 			dac_granted |= (VWRITE | VAPPEND);
3494 
3495 		if ((acc_mode & dac_granted) == acc_mode)
3496 			return (0);
3497 
3498 		goto privcheck;
3499 	}
3500 
3501 	/* Otherwise, check everyone else. */
3502 	if (file_mode & S_IXOTH)
3503 		dac_granted |= VEXEC;
3504 	if (file_mode & S_IROTH)
3505 		dac_granted |= VREAD;
3506 	if (file_mode & S_IWOTH)
3507 		dac_granted |= (VWRITE | VAPPEND);
3508 	if ((acc_mode & dac_granted) == acc_mode)
3509 		return (0);
3510 
3511 privcheck:
3512 	/*
3513 	 * Build a privilege mask to determine if the set of privileges
3514 	 * satisfies the requirements when combined with the granted mask
3515 	 * from above.  For each privilege, if the privilege is required,
3516 	 * bitwise or the request type onto the priv_granted mask.
3517 	 */
3518 	priv_granted = 0;
3519 
3520 	if (type == VDIR) {
3521 		/*
3522 		 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3523 		 * requests, instead of PRIV_VFS_EXEC.
3524 		 */
3525 		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3526 		    !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3527 			priv_granted |= VEXEC;
3528 	} else {
3529 		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3530 		    !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3531 			priv_granted |= VEXEC;
3532 	}
3533 
3534 	if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
3535 	    !priv_check_cred(cred, PRIV_VFS_READ, 0))
3536 		priv_granted |= VREAD;
3537 
3538 	if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3539 	    !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3540 		priv_granted |= (VWRITE | VAPPEND);
3541 
3542 	if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3543 	    !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3544 		priv_granted |= VADMIN;
3545 
3546 	if ((acc_mode & (priv_granted | dac_granted)) == acc_mode) {
3547 		/* XXX audit: privilege used */
3548 		if (privused != NULL)
3549 			*privused = 1;
3550 		return (0);
3551 	}
3552 
3553 	return ((acc_mode & VADMIN) ? EPERM : EACCES);
3554 }
3555 
3556 /*
3557  * Credential check based on process requesting service, and per-attribute
3558  * permissions.
3559  */
3560 int
3561 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3562     struct thread *td, int access)
3563 {
3564 
3565 	/*
3566 	 * Kernel-invoked always succeeds.
3567 	 */
3568 	if (cred == NOCRED)
3569 		return (0);
3570 
3571 	/*
3572 	 * Do not allow privileged processes in jail to directly manipulate
3573 	 * system attributes.
3574 	 */
3575 	switch (attrnamespace) {
3576 	case EXTATTR_NAMESPACE_SYSTEM:
3577 		/* Potentially should be: return (EPERM); */
3578 		return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3579 	case EXTATTR_NAMESPACE_USER:
3580 		return (VOP_ACCESS(vp, access, cred, td));
3581 	default:
3582 		return (EPERM);
3583 	}
3584 }
3585 
3586 #ifdef DEBUG_VFS_LOCKS
3587 /*
3588  * This only exists to supress warnings from unlocked specfs accesses.  It is
3589  * no longer ok to have an unlocked VFS.
3590  */
3591 #define	IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL ||		\
3592 	(vp)->v_type == VCHR ||	(vp)->v_type == VBAD)
3593 
3594 int vfs_badlock_ddb = 1;	/* Drop into debugger on violation. */
3595 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, "");
3596 
3597 int vfs_badlock_mutex = 1;	/* Check for interlock across VOPs. */
3598 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, "");
3599 
3600 int vfs_badlock_print = 1;	/* Print lock violations. */
3601 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, "");
3602 
3603 #ifdef KDB
3604 int vfs_badlock_backtrace = 1;	/* Print backtrace at lock violations. */
3605 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, "");
3606 #endif
3607 
3608 static void
3609 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3610 {
3611 
3612 #ifdef KDB
3613 	if (vfs_badlock_backtrace)
3614 		kdb_backtrace();
3615 #endif
3616 	if (vfs_badlock_print)
3617 		printf("%s: %p %s\n", str, (void *)vp, msg);
3618 	if (vfs_badlock_ddb)
3619 		kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3620 }
3621 
3622 void
3623 assert_vi_locked(struct vnode *vp, const char *str)
3624 {
3625 
3626 	if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3627 		vfs_badlock("interlock is not locked but should be", str, vp);
3628 }
3629 
3630 void
3631 assert_vi_unlocked(struct vnode *vp, const char *str)
3632 {
3633 
3634 	if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3635 		vfs_badlock("interlock is locked but should not be", str, vp);
3636 }
3637 
3638 void
3639 assert_vop_locked(struct vnode *vp, const char *str)
3640 {
3641 
3642 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0)
3643 		vfs_badlock("is not locked but should be", str, vp);
3644 }
3645 
3646 void
3647 assert_vop_unlocked(struct vnode *vp, const char *str)
3648 {
3649 
3650 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
3651 		vfs_badlock("is locked but should not be", str, vp);
3652 }
3653 
3654 void
3655 assert_vop_elocked(struct vnode *vp, const char *str)
3656 {
3657 
3658 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
3659 		vfs_badlock("is not exclusive locked but should be", str, vp);
3660 }
3661 
3662 #if 0
3663 void
3664 assert_vop_elocked_other(struct vnode *vp, const char *str)
3665 {
3666 
3667 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
3668 		vfs_badlock("is not exclusive locked by another thread",
3669 		    str, vp);
3670 }
3671 
3672 void
3673 assert_vop_slocked(struct vnode *vp, const char *str)
3674 {
3675 
3676 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
3677 		vfs_badlock("is not locked shared but should be", str, vp);
3678 }
3679 #endif /* 0 */
3680 #endif /* DEBUG_VFS_LOCKS */
3681 
3682 void
3683 vop_rename_pre(void *ap)
3684 {
3685 	struct vop_rename_args *a = ap;
3686 
3687 #ifdef DEBUG_VFS_LOCKS
3688 	if (a->a_tvp)
3689 		ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
3690 	ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
3691 	ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
3692 	ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
3693 
3694 	/* Check the source (from). */
3695 	if (a->a_tdvp != a->a_fdvp && a->a_tvp != a->a_fdvp)
3696 		ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3697 	if (a->a_tvp != a->a_fvp)
3698 		ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
3699 
3700 	/* Check the target. */
3701 	if (a->a_tvp)
3702 		ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3703 	ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3704 #endif
3705 	if (a->a_tdvp != a->a_fdvp)
3706 		vhold(a->a_fdvp);
3707 	if (a->a_tvp != a->a_fvp)
3708 		vhold(a->a_fvp);
3709 	vhold(a->a_tdvp);
3710 	if (a->a_tvp)
3711 		vhold(a->a_tvp);
3712 }
3713 
3714 void
3715 vop_strategy_pre(void *ap)
3716 {
3717 #ifdef DEBUG_VFS_LOCKS
3718 	struct vop_strategy_args *a;
3719 	struct buf *bp;
3720 
3721 	a = ap;
3722 	bp = a->a_bp;
3723 
3724 	/*
3725 	 * Cluster ops lock their component buffers but not the IO container.
3726 	 */
3727 	if ((bp->b_flags & B_CLUSTER) != 0)
3728 		return;
3729 
3730 	if (!BUF_ISLOCKED(bp)) {
3731 		if (vfs_badlock_print)
3732 			printf(
3733 			    "VOP_STRATEGY: bp is not locked but should be\n");
3734 		if (vfs_badlock_ddb)
3735 			kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3736 	}
3737 #endif
3738 }
3739 
3740 void
3741 vop_lookup_pre(void *ap)
3742 {
3743 #ifdef DEBUG_VFS_LOCKS
3744 	struct vop_lookup_args *a;
3745 	struct vnode *dvp;
3746 
3747 	a = ap;
3748 	dvp = a->a_dvp;
3749 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3750 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3751 #endif
3752 }
3753 
3754 void
3755 vop_lookup_post(void *ap, int rc)
3756 {
3757 #ifdef DEBUG_VFS_LOCKS
3758 	struct vop_lookup_args *a;
3759 	struct vnode *dvp;
3760 	struct vnode *vp;
3761 
3762 	a = ap;
3763 	dvp = a->a_dvp;
3764 	vp = *(a->a_vpp);
3765 
3766 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3767 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3768 
3769 	if (!rc)
3770 		ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
3771 #endif
3772 }
3773 
3774 void
3775 vop_lock_pre(void *ap)
3776 {
3777 #ifdef DEBUG_VFS_LOCKS
3778 	struct vop_lock1_args *a = ap;
3779 
3780 	if ((a->a_flags & LK_INTERLOCK) == 0)
3781 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3782 	else
3783 		ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
3784 #endif
3785 }
3786 
3787 void
3788 vop_lock_post(void *ap, int rc)
3789 {
3790 #ifdef DEBUG_VFS_LOCKS
3791 	struct vop_lock1_args *a = ap;
3792 
3793 	ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3794 	if (rc == 0)
3795 		ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
3796 #endif
3797 }
3798 
3799 void
3800 vop_unlock_pre(void *ap)
3801 {
3802 #ifdef DEBUG_VFS_LOCKS
3803 	struct vop_unlock_args *a = ap;
3804 
3805 	if (a->a_flags & LK_INTERLOCK)
3806 		ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
3807 	ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
3808 #endif
3809 }
3810 
3811 void
3812 vop_unlock_post(void *ap, int rc)
3813 {
3814 #ifdef DEBUG_VFS_LOCKS
3815 	struct vop_unlock_args *a = ap;
3816 
3817 	if (a->a_flags & LK_INTERLOCK)
3818 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
3819 #endif
3820 }
3821 
3822 void
3823 vop_create_post(void *ap, int rc)
3824 {
3825 	struct vop_create_args *a = ap;
3826 
3827 	if (!rc)
3828 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3829 }
3830 
3831 void
3832 vop_link_post(void *ap, int rc)
3833 {
3834 	struct vop_link_args *a = ap;
3835 
3836 	if (!rc) {
3837 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
3838 		VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
3839 	}
3840 }
3841 
3842 void
3843 vop_mkdir_post(void *ap, int rc)
3844 {
3845 	struct vop_mkdir_args *a = ap;
3846 
3847 	if (!rc)
3848 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3849 }
3850 
3851 void
3852 vop_mknod_post(void *ap, int rc)
3853 {
3854 	struct vop_mknod_args *a = ap;
3855 
3856 	if (!rc)
3857 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3858 }
3859 
3860 void
3861 vop_remove_post(void *ap, int rc)
3862 {
3863 	struct vop_remove_args *a = ap;
3864 
3865 	if (!rc) {
3866 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3867 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3868 	}
3869 }
3870 
3871 void
3872 vop_rename_post(void *ap, int rc)
3873 {
3874 	struct vop_rename_args *a = ap;
3875 
3876 	if (!rc) {
3877 		VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
3878 		VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
3879 		VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
3880 		if (a->a_tvp)
3881 			VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
3882 	}
3883 	if (a->a_tdvp != a->a_fdvp)
3884 		vdrop(a->a_fdvp);
3885 	if (a->a_tvp != a->a_fvp)
3886 		vdrop(a->a_fvp);
3887 	vdrop(a->a_tdvp);
3888 	if (a->a_tvp)
3889 		vdrop(a->a_tvp);
3890 }
3891 
3892 void
3893 vop_rmdir_post(void *ap, int rc)
3894 {
3895 	struct vop_rmdir_args *a = ap;
3896 
3897 	if (!rc) {
3898 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3899 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3900 	}
3901 }
3902 
3903 void
3904 vop_setattr_post(void *ap, int rc)
3905 {
3906 	struct vop_setattr_args *a = ap;
3907 
3908 	if (!rc)
3909 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
3910 }
3911 
3912 void
3913 vop_symlink_post(void *ap, int rc)
3914 {
3915 	struct vop_symlink_args *a = ap;
3916 
3917 	if (!rc)
3918 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3919 }
3920 
3921 static struct knlist fs_knlist;
3922 
3923 static void
3924 vfs_event_init(void *arg)
3925 {
3926 	knlist_init(&fs_knlist, NULL, NULL, NULL, NULL);
3927 }
3928 /* XXX - correct order? */
3929 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
3930 
3931 void
3932 vfs_event_signal(fsid_t *fsid, u_int32_t event, intptr_t data __unused)
3933 {
3934 
3935 	KNOTE_UNLOCKED(&fs_knlist, event);
3936 }
3937 
3938 static int	filt_fsattach(struct knote *kn);
3939 static void	filt_fsdetach(struct knote *kn);
3940 static int	filt_fsevent(struct knote *kn, long hint);
3941 
3942 struct filterops fs_filtops =
3943 	{ 0, filt_fsattach, filt_fsdetach, filt_fsevent };
3944 
3945 static int
3946 filt_fsattach(struct knote *kn)
3947 {
3948 
3949 	kn->kn_flags |= EV_CLEAR;
3950 	knlist_add(&fs_knlist, kn, 0);
3951 	return (0);
3952 }
3953 
3954 static void
3955 filt_fsdetach(struct knote *kn)
3956 {
3957 
3958 	knlist_remove(&fs_knlist, kn, 0);
3959 }
3960 
3961 static int
3962 filt_fsevent(struct knote *kn, long hint)
3963 {
3964 
3965 	kn->kn_fflags |= hint;
3966 	return (kn->kn_fflags != 0);
3967 }
3968 
3969 static int
3970 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
3971 {
3972 	struct vfsidctl vc;
3973 	int error;
3974 	struct mount *mp;
3975 
3976 	error = SYSCTL_IN(req, &vc, sizeof(vc));
3977 	if (error)
3978 		return (error);
3979 	if (vc.vc_vers != VFS_CTL_VERS1)
3980 		return (EINVAL);
3981 	mp = vfs_getvfs(&vc.vc_fsid);
3982 	if (mp == NULL)
3983 		return (ENOENT);
3984 	/* ensure that a specific sysctl goes to the right filesystem. */
3985 	if (strcmp(vc.vc_fstypename, "*") != 0 &&
3986 	    strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
3987 		vfs_rel(mp);
3988 		return (EINVAL);
3989 	}
3990 	VCTLTOREQ(&vc, req);
3991 	error = VFS_SYSCTL(mp, vc.vc_op, req);
3992 	vfs_rel(mp);
3993 	return (error);
3994 }
3995 
3996 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "",
3997     "Sysctl by fsid");
3998 
3999 /*
4000  * Function to initialize a va_filerev field sensibly.
4001  * XXX: Wouldn't a random number make a lot more sense ??
4002  */
4003 u_quad_t
4004 init_va_filerev(void)
4005 {
4006 	struct bintime bt;
4007 
4008 	getbinuptime(&bt);
4009 	return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4010 }
4011 
4012 static int	filt_vfsread(struct knote *kn, long hint);
4013 static int	filt_vfswrite(struct knote *kn, long hint);
4014 static int	filt_vfsvnode(struct knote *kn, long hint);
4015 static void	filt_vfsdetach(struct knote *kn);
4016 static struct filterops vfsread_filtops =
4017 	{ 1, NULL, filt_vfsdetach, filt_vfsread };
4018 static struct filterops vfswrite_filtops =
4019 	{ 1, NULL, filt_vfsdetach, filt_vfswrite };
4020 static struct filterops vfsvnode_filtops =
4021 	{ 1, NULL, filt_vfsdetach, filt_vfsvnode };
4022 
4023 static void
4024 vfs_knllock(void *arg)
4025 {
4026 	struct vnode *vp = arg;
4027 
4028 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4029 }
4030 
4031 static void
4032 vfs_knlunlock(void *arg)
4033 {
4034 	struct vnode *vp = arg;
4035 
4036 	VOP_UNLOCK(vp, 0);
4037 }
4038 
4039 static int
4040 vfs_knllocked(void *arg)
4041 {
4042 	struct vnode *vp = arg;
4043 
4044 	return (VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
4045 }
4046 
4047 int
4048 vfs_kqfilter(struct vop_kqfilter_args *ap)
4049 {
4050 	struct vnode *vp = ap->a_vp;
4051 	struct knote *kn = ap->a_kn;
4052 	struct knlist *knl;
4053 
4054 	switch (kn->kn_filter) {
4055 	case EVFILT_READ:
4056 		kn->kn_fop = &vfsread_filtops;
4057 		break;
4058 	case EVFILT_WRITE:
4059 		kn->kn_fop = &vfswrite_filtops;
4060 		break;
4061 	case EVFILT_VNODE:
4062 		kn->kn_fop = &vfsvnode_filtops;
4063 		break;
4064 	default:
4065 		return (EINVAL);
4066 	}
4067 
4068 	kn->kn_hook = (caddr_t)vp;
4069 
4070 	if (vp->v_pollinfo == NULL)
4071 		v_addpollinfo(vp);
4072 	if (vp->v_pollinfo == NULL)
4073 		return (ENOMEM);
4074 	knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4075 	knlist_add(knl, kn, 0);
4076 
4077 	return (0);
4078 }
4079 
4080 /*
4081  * Detach knote from vnode
4082  */
4083 static void
4084 filt_vfsdetach(struct knote *kn)
4085 {
4086 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4087 
4088 	KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4089 	knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4090 }
4091 
4092 /*ARGSUSED*/
4093 static int
4094 filt_vfsread(struct knote *kn, long hint)
4095 {
4096 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4097 	struct vattr va;
4098 
4099 	/*
4100 	 * filesystem is gone, so set the EOF flag and schedule
4101 	 * the knote for deletion.
4102 	 */
4103 	if (hint == NOTE_REVOKE) {
4104 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4105 		return (1);
4106 	}
4107 
4108 	if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4109 		return (0);
4110 
4111 	kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4112 	return (kn->kn_data != 0);
4113 }
4114 
4115 /*ARGSUSED*/
4116 static int
4117 filt_vfswrite(struct knote *kn, long hint)
4118 {
4119 	/*
4120 	 * filesystem is gone, so set the EOF flag and schedule
4121 	 * the knote for deletion.
4122 	 */
4123 	if (hint == NOTE_REVOKE)
4124 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4125 
4126 	kn->kn_data = 0;
4127 	return (1);
4128 }
4129 
4130 static int
4131 filt_vfsvnode(struct knote *kn, long hint)
4132 {
4133 	if (kn->kn_sfflags & hint)
4134 		kn->kn_fflags |= hint;
4135 	if (hint == NOTE_REVOKE) {
4136 		kn->kn_flags |= EV_EOF;
4137 		return (1);
4138 	}
4139 	return (kn->kn_fflags != 0);
4140 }
4141 
4142 int
4143 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4144 {
4145 	int error;
4146 
4147 	if (dp->d_reclen > ap->a_uio->uio_resid)
4148 		return (ENAMETOOLONG);
4149 	error = uiomove(dp, dp->d_reclen, ap->a_uio);
4150 	if (error) {
4151 		if (ap->a_ncookies != NULL) {
4152 			if (ap->a_cookies != NULL)
4153 				free(ap->a_cookies, M_TEMP);
4154 			ap->a_cookies = NULL;
4155 			*ap->a_ncookies = 0;
4156 		}
4157 		return (error);
4158 	}
4159 	if (ap->a_ncookies == NULL)
4160 		return (0);
4161 
4162 	KASSERT(ap->a_cookies,
4163 	    ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4164 
4165 	*ap->a_cookies = realloc(*ap->a_cookies,
4166 	    (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4167 	(*ap->a_cookies)[*ap->a_ncookies] = off;
4168 	return (0);
4169 }
4170 
4171 /*
4172  * Mark for update the access time of the file if the filesystem
4173  * supports VA_MARK_ATIME.  This functionality is used by execve
4174  * and mmap, so we want to avoid the synchronous I/O implied by
4175  * directly setting va_atime for the sake of efficiency.
4176  */
4177 void
4178 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4179 {
4180 	struct vattr atimeattr;
4181 
4182 	if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) {
4183 		VATTR_NULL(&atimeattr);
4184 		atimeattr.va_vaflags |= VA_MARK_ATIME;
4185 		(void)VOP_SETATTR(vp, &atimeattr, cred);
4186 	}
4187 }
4188