xref: /freebsd/sys/kern/vfs_subr.c (revision cacdd70cc751fb68dec4b86c5e5b8c969b6e26ef)
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     struct thread *td)
343 {
344 	int lkflags;
345 
346 	MNT_ILOCK(mp);
347 	MNT_REF(mp);
348 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
349 		if (flags & LK_NOWAIT) {
350 			MNT_REL(mp);
351 			MNT_IUNLOCK(mp);
352 			return (ENOENT);
353 		}
354 		if (interlkp)
355 			mtx_unlock(interlkp);
356 		mp->mnt_kern_flag |= MNTK_MWAIT;
357 		/*
358 		 * Since all busy locks are shared except the exclusive
359 		 * lock granted when unmounting, the only place that a
360 		 * wakeup needs to be done is at the release of the
361 		 * exclusive lock at the end of dounmount.
362 		 */
363 		msleep(mp, MNT_MTX(mp), PVFS, "vfs_busy", 0);
364 		MNT_REL(mp);
365 		MNT_IUNLOCK(mp);
366 		if (interlkp)
367 			mtx_lock(interlkp);
368 		return (ENOENT);
369 	}
370 	if (interlkp)
371 		mtx_unlock(interlkp);
372 	lkflags = LK_SHARED | LK_INTERLOCK;
373 	if (lockmgr(&mp->mnt_lock, lkflags, MNT_MTX(mp)))
374 		panic("vfs_busy: unexpected lock failure");
375 	return (0);
376 }
377 
378 /*
379  * Free a busy filesystem.
380  */
381 void
382 vfs_unbusy(struct mount *mp, struct thread *td)
383 {
384 
385 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL);
386 	vfs_rel(mp);
387 }
388 
389 /*
390  * Lookup a mount point by filesystem identifier.
391  */
392 struct mount *
393 vfs_getvfs(fsid_t *fsid)
394 {
395 	struct mount *mp;
396 
397 	mtx_lock(&mountlist_mtx);
398 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
399 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
400 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
401 			vfs_ref(mp);
402 			mtx_unlock(&mountlist_mtx);
403 			return (mp);
404 		}
405 	}
406 	mtx_unlock(&mountlist_mtx);
407 	return ((struct mount *) 0);
408 }
409 
410 /*
411  * Check if a user can access privileged mount options.
412  */
413 int
414 vfs_suser(struct mount *mp, struct thread *td)
415 {
416 	int error;
417 
418 	/*
419 	 * If the thread is jailed, but this is not a jail-friendly file
420 	 * system, deny immediately.
421 	 */
422 	if (jailed(td->td_ucred) && !(mp->mnt_vfc->vfc_flags & VFCF_JAIL))
423 		return (EPERM);
424 
425 	/*
426 	 * If the file system was mounted outside a jail and a jailed thread
427 	 * tries to access it, deny immediately.
428 	 */
429 	if (!jailed(mp->mnt_cred) && jailed(td->td_ucred))
430 		return (EPERM);
431 
432 	/*
433 	 * If the file system was mounted inside different jail that the jail of
434 	 * the calling thread, deny immediately.
435 	 */
436 	if (jailed(mp->mnt_cred) && jailed(td->td_ucred) &&
437 	    mp->mnt_cred->cr_prison != td->td_ucred->cr_prison) {
438 		return (EPERM);
439 	}
440 
441 	if ((mp->mnt_flag & MNT_USER) == 0 ||
442 	    mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
443 		if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
444 			return (error);
445 	}
446 	return (0);
447 }
448 
449 /*
450  * Get a new unique fsid.  Try to make its val[0] unique, since this value
451  * will be used to create fake device numbers for stat().  Also try (but
452  * not so hard) make its val[0] unique mod 2^16, since some emulators only
453  * support 16-bit device numbers.  We end up with unique val[0]'s for the
454  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
455  *
456  * Keep in mind that several mounts may be running in parallel.  Starting
457  * the search one past where the previous search terminated is both a
458  * micro-optimization and a defense against returning the same fsid to
459  * different mounts.
460  */
461 void
462 vfs_getnewfsid(struct mount *mp)
463 {
464 	static u_int16_t mntid_base;
465 	struct mount *nmp;
466 	fsid_t tfsid;
467 	int mtype;
468 
469 	mtx_lock(&mntid_mtx);
470 	mtype = mp->mnt_vfc->vfc_typenum;
471 	tfsid.val[1] = mtype;
472 	mtype = (mtype & 0xFF) << 24;
473 	for (;;) {
474 		tfsid.val[0] = makedev(255,
475 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
476 		mntid_base++;
477 		if ((nmp = vfs_getvfs(&tfsid)) == NULL)
478 			break;
479 		vfs_rel(nmp);
480 	}
481 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
482 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
483 	mtx_unlock(&mntid_mtx);
484 }
485 
486 /*
487  * Knob to control the precision of file timestamps:
488  *
489  *   0 = seconds only; nanoseconds zeroed.
490  *   1 = seconds and nanoseconds, accurate within 1/HZ.
491  *   2 = seconds and nanoseconds, truncated to microseconds.
492  * >=3 = seconds and nanoseconds, maximum precision.
493  */
494 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
495 
496 static int timestamp_precision = TSP_SEC;
497 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
498     &timestamp_precision, 0, "");
499 
500 /*
501  * Get a current timestamp.
502  */
503 void
504 vfs_timestamp(struct timespec *tsp)
505 {
506 	struct timeval tv;
507 
508 	switch (timestamp_precision) {
509 	case TSP_SEC:
510 		tsp->tv_sec = time_second;
511 		tsp->tv_nsec = 0;
512 		break;
513 	case TSP_HZ:
514 		getnanotime(tsp);
515 		break;
516 	case TSP_USEC:
517 		microtime(&tv);
518 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
519 		break;
520 	case TSP_NSEC:
521 	default:
522 		nanotime(tsp);
523 		break;
524 	}
525 }
526 
527 /*
528  * Set vnode attributes to VNOVAL
529  */
530 void
531 vattr_null(struct vattr *vap)
532 {
533 
534 	vap->va_type = VNON;
535 	vap->va_size = VNOVAL;
536 	vap->va_bytes = VNOVAL;
537 	vap->va_mode = VNOVAL;
538 	vap->va_nlink = VNOVAL;
539 	vap->va_uid = VNOVAL;
540 	vap->va_gid = VNOVAL;
541 	vap->va_fsid = VNOVAL;
542 	vap->va_fileid = VNOVAL;
543 	vap->va_blocksize = VNOVAL;
544 	vap->va_rdev = VNOVAL;
545 	vap->va_atime.tv_sec = VNOVAL;
546 	vap->va_atime.tv_nsec = VNOVAL;
547 	vap->va_mtime.tv_sec = VNOVAL;
548 	vap->va_mtime.tv_nsec = VNOVAL;
549 	vap->va_ctime.tv_sec = VNOVAL;
550 	vap->va_ctime.tv_nsec = VNOVAL;
551 	vap->va_birthtime.tv_sec = VNOVAL;
552 	vap->va_birthtime.tv_nsec = VNOVAL;
553 	vap->va_flags = VNOVAL;
554 	vap->va_gen = VNOVAL;
555 	vap->va_vaflags = 0;
556 }
557 
558 /*
559  * This routine is called when we have too many vnodes.  It attempts
560  * to free <count> vnodes and will potentially free vnodes that still
561  * have VM backing store (VM backing store is typically the cause
562  * of a vnode blowout so we want to do this).  Therefore, this operation
563  * is not considered cheap.
564  *
565  * A number of conditions may prevent a vnode from being reclaimed.
566  * the buffer cache may have references on the vnode, a directory
567  * vnode may still have references due to the namei cache representing
568  * underlying files, or the vnode may be in active use.   It is not
569  * desireable to reuse such vnodes.  These conditions may cause the
570  * number of vnodes to reach some minimum value regardless of what
571  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
572  */
573 static int
574 vlrureclaim(struct mount *mp)
575 {
576 	struct thread *td;
577 	struct vnode *vp;
578 	int done;
579 	int trigger;
580 	int usevnodes;
581 	int count;
582 
583 	/*
584 	 * Calculate the trigger point, don't allow user
585 	 * screwups to blow us up.   This prevents us from
586 	 * recycling vnodes with lots of resident pages.  We
587 	 * aren't trying to free memory, we are trying to
588 	 * free vnodes.
589 	 */
590 	usevnodes = desiredvnodes;
591 	if (usevnodes <= 0)
592 		usevnodes = 1;
593 	trigger = cnt.v_page_count * 2 / usevnodes;
594 	done = 0;
595 	td = curthread;
596 	vn_start_write(NULL, &mp, V_WAIT);
597 	MNT_ILOCK(mp);
598 	count = mp->mnt_nvnodelistsize / 10 + 1;
599 	while (count != 0) {
600 		vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
601 		while (vp != NULL && vp->v_type == VMARKER)
602 			vp = TAILQ_NEXT(vp, v_nmntvnodes);
603 		if (vp == NULL)
604 			break;
605 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
606 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
607 		--count;
608 		if (!VI_TRYLOCK(vp))
609 			goto next_iter;
610 		/*
611 		 * If it's been deconstructed already, it's still
612 		 * referenced, or it exceeds the trigger, skip it.
613 		 */
614 		if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) ||
615 		    (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
616 		    vp->v_object->resident_page_count > trigger)) {
617 			VI_UNLOCK(vp);
618 			goto next_iter;
619 		}
620 		MNT_IUNLOCK(mp);
621 		vholdl(vp);
622 		if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
623 			vdrop(vp);
624 			goto next_iter_mntunlocked;
625 		}
626 		VI_LOCK(vp);
627 		/*
628 		 * v_usecount may have been bumped after VOP_LOCK() dropped
629 		 * the vnode interlock and before it was locked again.
630 		 *
631 		 * It is not necessary to recheck VI_DOOMED because it can
632 		 * only be set by another thread that holds both the vnode
633 		 * lock and vnode interlock.  If another thread has the
634 		 * vnode lock before we get to VOP_LOCK() and obtains the
635 		 * vnode interlock after VOP_LOCK() drops the vnode
636 		 * interlock, the other thread will be unable to drop the
637 		 * vnode lock before our VOP_LOCK() call fails.
638 		 */
639 		if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) ||
640 		    (vp->v_object != NULL &&
641 		    vp->v_object->resident_page_count > trigger)) {
642 			VOP_UNLOCK(vp, LK_INTERLOCK);
643 			goto next_iter_mntunlocked;
644 		}
645 		KASSERT((vp->v_iflag & VI_DOOMED) == 0,
646 		    ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
647 		vgonel(vp);
648 		VOP_UNLOCK(vp, 0);
649 		vdropl(vp);
650 		done++;
651 next_iter_mntunlocked:
652 		if ((count % 256) != 0)
653 			goto relock_mnt;
654 		goto yield;
655 next_iter:
656 		if ((count % 256) != 0)
657 			continue;
658 		MNT_IUNLOCK(mp);
659 yield:
660 		uio_yield();
661 relock_mnt:
662 		MNT_ILOCK(mp);
663 	}
664 	MNT_IUNLOCK(mp);
665 	vn_finished_write(mp);
666 	return done;
667 }
668 
669 /*
670  * Attempt to keep the free list at wantfreevnodes length.
671  */
672 static void
673 vnlru_free(int count)
674 {
675 	struct vnode *vp;
676 	int vfslocked;
677 
678 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
679 	for (; count > 0; count--) {
680 		vp = TAILQ_FIRST(&vnode_free_list);
681 		/*
682 		 * The list can be modified while the free_list_mtx
683 		 * has been dropped and vp could be NULL here.
684 		 */
685 		if (!vp)
686 			break;
687 		VNASSERT(vp->v_op != NULL, vp,
688 		    ("vnlru_free: vnode already reclaimed."));
689 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
690 		/*
691 		 * Don't recycle if we can't get the interlock.
692 		 */
693 		if (!VI_TRYLOCK(vp)) {
694 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
695 			continue;
696 		}
697 		VNASSERT(VCANRECYCLE(vp), vp,
698 		    ("vp inconsistent on freelist"));
699 		freevnodes--;
700 		vp->v_iflag &= ~VI_FREE;
701 		vholdl(vp);
702 		mtx_unlock(&vnode_free_list_mtx);
703 		VI_UNLOCK(vp);
704 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
705 		vtryrecycle(vp);
706 		VFS_UNLOCK_GIANT(vfslocked);
707 		/*
708 		 * If the recycled succeeded this vdrop will actually free
709 		 * the vnode.  If not it will simply place it back on
710 		 * the free list.
711 		 */
712 		vdrop(vp);
713 		mtx_lock(&vnode_free_list_mtx);
714 	}
715 }
716 /*
717  * Attempt to recycle vnodes in a context that is always safe to block.
718  * Calling vlrurecycle() from the bowels of filesystem code has some
719  * interesting deadlock problems.
720  */
721 static struct proc *vnlruproc;
722 static int vnlruproc_sig;
723 
724 static void
725 vnlru_proc(void)
726 {
727 	struct mount *mp, *nmp;
728 	int done;
729 	struct proc *p = vnlruproc;
730 	struct thread *td = curthread;
731 
732 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
733 	    SHUTDOWN_PRI_FIRST);
734 
735 	mtx_lock(&Giant);
736 
737 	for (;;) {
738 		kproc_suspend_check(p);
739 		mtx_lock(&vnode_free_list_mtx);
740 		if (freevnodes > wantfreevnodes)
741 			vnlru_free(freevnodes - wantfreevnodes);
742 		if (numvnodes <= desiredvnodes * 9 / 10) {
743 			vnlruproc_sig = 0;
744 			wakeup(&vnlruproc_sig);
745 			msleep(vnlruproc, &vnode_free_list_mtx,
746 			    PVFS|PDROP, "vlruwt", hz);
747 			continue;
748 		}
749 		mtx_unlock(&vnode_free_list_mtx);
750 		done = 0;
751 		mtx_lock(&mountlist_mtx);
752 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
753 			int vfsunlocked;
754 			if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
755 				nmp = TAILQ_NEXT(mp, mnt_list);
756 				continue;
757 			}
758 			if (!VFS_NEEDSGIANT(mp)) {
759 				mtx_unlock(&Giant);
760 				vfsunlocked = 1;
761 			} else
762 				vfsunlocked = 0;
763 			done += vlrureclaim(mp);
764 			if (vfsunlocked)
765 				mtx_lock(&Giant);
766 			mtx_lock(&mountlist_mtx);
767 			nmp = TAILQ_NEXT(mp, mnt_list);
768 			vfs_unbusy(mp, td);
769 		}
770 		mtx_unlock(&mountlist_mtx);
771 		if (done == 0) {
772 			EVENTHANDLER_INVOKE(vfs_lowvnodes, desiredvnodes / 10);
773 #if 0
774 			/* These messages are temporary debugging aids */
775 			if (vnlru_nowhere < 5)
776 				printf("vnlru process getting nowhere..\n");
777 			else if (vnlru_nowhere == 5)
778 				printf("vnlru process messages stopped.\n");
779 #endif
780 			vnlru_nowhere++;
781 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
782 		} else
783 			uio_yield();
784 	}
785 }
786 
787 static struct kproc_desc vnlru_kp = {
788 	"vnlru",
789 	vnlru_proc,
790 	&vnlruproc
791 };
792 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
793     &vnlru_kp);
794 
795 /*
796  * Routines having to do with the management of the vnode table.
797  */
798 
799 static void
800 vdestroy(struct vnode *vp)
801 {
802 	struct bufobj *bo;
803 
804 	CTR1(KTR_VFS, "vdestroy vp %p", vp);
805 	mtx_lock(&vnode_free_list_mtx);
806 	numvnodes--;
807 	mtx_unlock(&vnode_free_list_mtx);
808 	bo = &vp->v_bufobj;
809 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
810 	    ("cleaned vnode still on the free list."));
811 	VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
812 	VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
813 	VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
814 	VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
815 	VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
816 	VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
817 	VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
818 	VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
819 	VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
820 	VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
821 	VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
822 	VI_UNLOCK(vp);
823 #ifdef MAC
824 	mac_vnode_destroy(vp);
825 #endif
826 	if (vp->v_pollinfo != NULL) {
827 		knlist_destroy(&vp->v_pollinfo->vpi_selinfo.si_note);
828 		mtx_destroy(&vp->v_pollinfo->vpi_lock);
829 		uma_zfree(vnodepoll_zone, vp->v_pollinfo);
830 	}
831 #ifdef INVARIANTS
832 	/* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */
833 	vp->v_op = NULL;
834 #endif
835 	lockdestroy(vp->v_vnlock);
836 	mtx_destroy(&vp->v_interlock);
837 	mtx_destroy(BO_MTX(bo));
838 	uma_zfree(vnode_zone, vp);
839 }
840 
841 /*
842  * Try to recycle a freed vnode.  We abort if anyone picks up a reference
843  * before we actually vgone().  This function must be called with the vnode
844  * held to prevent the vnode from being returned to the free list midway
845  * through vgone().
846  */
847 static int
848 vtryrecycle(struct vnode *vp)
849 {
850 	struct mount *vnmp;
851 
852 	CTR1(KTR_VFS, "vtryrecycle: trying vp %p", vp);
853 	VNASSERT(vp->v_holdcnt, vp,
854 	    ("vtryrecycle: Recycling vp %p without a reference.", vp));
855 	/*
856 	 * This vnode may found and locked via some other list, if so we
857 	 * can't recycle it yet.
858 	 */
859 	if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0)
860 		return (EWOULDBLOCK);
861 	/*
862 	 * Don't recycle if its filesystem is being suspended.
863 	 */
864 	if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
865 		VOP_UNLOCK(vp, 0);
866 		return (EBUSY);
867 	}
868 	/*
869 	 * If we got this far, we need to acquire the interlock and see if
870 	 * anyone picked up this vnode from another list.  If not, we will
871 	 * mark it with DOOMED via vgonel() so that anyone who does find it
872 	 * will skip over it.
873 	 */
874 	VI_LOCK(vp);
875 	if (vp->v_usecount) {
876 		VOP_UNLOCK(vp, LK_INTERLOCK);
877 		vn_finished_write(vnmp);
878 		return (EBUSY);
879 	}
880 	if ((vp->v_iflag & VI_DOOMED) == 0)
881 		vgonel(vp);
882 	VOP_UNLOCK(vp, LK_INTERLOCK);
883 	vn_finished_write(vnmp);
884 	CTR1(KTR_VFS, "vtryrecycle: recycled vp %p", vp);
885 	return (0);
886 }
887 
888 /*
889  * Return the next vnode from the free list.
890  */
891 int
892 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
893     struct vnode **vpp)
894 {
895 	struct vnode *vp = NULL;
896 	struct bufobj *bo;
897 
898 	mtx_lock(&vnode_free_list_mtx);
899 	/*
900 	 * Lend our context to reclaim vnodes if they've exceeded the max.
901 	 */
902 	if (freevnodes > wantfreevnodes)
903 		vnlru_free(1);
904 	/*
905 	 * Wait for available vnodes.
906 	 */
907 	if (numvnodes > desiredvnodes) {
908 		if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) {
909 			/*
910 			 * File system is beeing suspended, we cannot risk a
911 			 * deadlock here, so allocate new vnode anyway.
912 			 */
913 			if (freevnodes > wantfreevnodes)
914 				vnlru_free(freevnodes - wantfreevnodes);
915 			goto alloc;
916 		}
917 		if (vnlruproc_sig == 0) {
918 			vnlruproc_sig = 1;	/* avoid unnecessary wakeups */
919 			wakeup(vnlruproc);
920 		}
921 		msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
922 		    "vlruwk", hz);
923 #if 0	/* XXX Not all VFS_VGET/ffs_vget callers check returns. */
924 		if (numvnodes > desiredvnodes) {
925 			mtx_unlock(&vnode_free_list_mtx);
926 			return (ENFILE);
927 		}
928 #endif
929 	}
930 alloc:
931 	numvnodes++;
932 	mtx_unlock(&vnode_free_list_mtx);
933 	vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
934 	/*
935 	 * Setup locks.
936 	 */
937 	vp->v_vnlock = &vp->v_lock;
938 	mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
939 	/*
940 	 * By default, don't allow shared locks unless filesystems
941 	 * opt-in.
942 	 */
943 	lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
944 	/*
945 	 * Initialize bufobj.
946 	 */
947 	bo = &vp->v_bufobj;
948 	bo->__bo_vnode = vp;
949 	mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
950 	bo->bo_ops = &buf_ops_bio;
951 	bo->bo_private = vp;
952 	TAILQ_INIT(&bo->bo_clean.bv_hd);
953 	TAILQ_INIT(&bo->bo_dirty.bv_hd);
954 	/*
955 	 * Initialize namecache.
956 	 */
957 	LIST_INIT(&vp->v_cache_src);
958 	TAILQ_INIT(&vp->v_cache_dst);
959 	/*
960 	 * Finalize various vnode identity bits.
961 	 */
962 	vp->v_type = VNON;
963 	vp->v_tag = tag;
964 	vp->v_op = vops;
965 	v_incr_usecount(vp);
966 	vp->v_data = 0;
967 #ifdef MAC
968 	mac_vnode_init(vp);
969 	if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
970 		mac_vnode_associate_singlelabel(mp, vp);
971 	else if (mp == NULL)
972 		printf("NULL mp in getnewvnode()\n");
973 #endif
974 	if (mp != NULL) {
975 		bo->bo_bsize = mp->mnt_stat.f_iosize;
976 		if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
977 			vp->v_vflag |= VV_NOKNOTE;
978 	}
979 
980 	CTR2(KTR_VFS, "getnewvnode: mp %p vp %p", mp, vp);
981 	*vpp = vp;
982 	return (0);
983 }
984 
985 /*
986  * Delete from old mount point vnode list, if on one.
987  */
988 static void
989 delmntque(struct vnode *vp)
990 {
991 	struct mount *mp;
992 
993 	mp = vp->v_mount;
994 	if (mp == NULL)
995 		return;
996 	MNT_ILOCK(mp);
997 	vp->v_mount = NULL;
998 	VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
999 		("bad mount point vnode list size"));
1000 	TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1001 	mp->mnt_nvnodelistsize--;
1002 	MNT_REL(mp);
1003 	MNT_IUNLOCK(mp);
1004 }
1005 
1006 static void
1007 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1008 {
1009 
1010 	vp->v_data = NULL;
1011 	vp->v_op = &dead_vnodeops;
1012 	/* XXX non mp-safe fs may still call insmntque with vnode
1013 	   unlocked */
1014 	if (!VOP_ISLOCKED(vp))
1015 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1016 	vgone(vp);
1017 	vput(vp);
1018 }
1019 
1020 /*
1021  * Insert into list of vnodes for the new mount point, if available.
1022  */
1023 int
1024 insmntque1(struct vnode *vp, struct mount *mp,
1025 	void (*dtr)(struct vnode *, void *), void *dtr_arg)
1026 {
1027 
1028 	KASSERT(vp->v_mount == NULL,
1029 		("insmntque: vnode already on per mount vnode list"));
1030 	VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1031 	MNT_ILOCK(mp);
1032 	if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1033 	    mp->mnt_nvnodelistsize == 0) {
1034 		MNT_IUNLOCK(mp);
1035 		if (dtr != NULL)
1036 			dtr(vp, dtr_arg);
1037 		return (EBUSY);
1038 	}
1039 	vp->v_mount = mp;
1040 	MNT_REF(mp);
1041 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1042 	VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1043 		("neg mount point vnode list size"));
1044 	mp->mnt_nvnodelistsize++;
1045 	MNT_IUNLOCK(mp);
1046 	return (0);
1047 }
1048 
1049 int
1050 insmntque(struct vnode *vp, struct mount *mp)
1051 {
1052 
1053 	return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1054 }
1055 
1056 /*
1057  * Flush out and invalidate all buffers associated with a bufobj
1058  * Called with the underlying object locked.
1059  */
1060 int
1061 bufobj_invalbuf(struct bufobj *bo, int flags, struct thread *td, int slpflag,
1062     int slptimeo)
1063 {
1064 	int error;
1065 
1066 	BO_LOCK(bo);
1067 	if (flags & V_SAVE) {
1068 		error = bufobj_wwait(bo, slpflag, slptimeo);
1069 		if (error) {
1070 			BO_UNLOCK(bo);
1071 			return (error);
1072 		}
1073 		if (bo->bo_dirty.bv_cnt > 0) {
1074 			BO_UNLOCK(bo);
1075 			if ((error = BO_SYNC(bo, MNT_WAIT, td)) != 0)
1076 				return (error);
1077 			/*
1078 			 * XXX We could save a lock/unlock if this was only
1079 			 * enabled under INVARIANTS
1080 			 */
1081 			BO_LOCK(bo);
1082 			if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1083 				panic("vinvalbuf: dirty bufs");
1084 		}
1085 	}
1086 	/*
1087 	 * If you alter this loop please notice that interlock is dropped and
1088 	 * reacquired in flushbuflist.  Special care is needed to ensure that
1089 	 * no race conditions occur from this.
1090 	 */
1091 	do {
1092 		error = flushbuflist(&bo->bo_clean,
1093 		    flags, bo, slpflag, slptimeo);
1094 		if (error == 0)
1095 			error = flushbuflist(&bo->bo_dirty,
1096 			    flags, bo, slpflag, slptimeo);
1097 		if (error != 0 && error != EAGAIN) {
1098 			BO_UNLOCK(bo);
1099 			return (error);
1100 		}
1101 	} while (error != 0);
1102 
1103 	/*
1104 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1105 	 * have write I/O in-progress but if there is a VM object then the
1106 	 * VM object can also have read-I/O in-progress.
1107 	 */
1108 	do {
1109 		bufobj_wwait(bo, 0, 0);
1110 		BO_UNLOCK(bo);
1111 		if (bo->bo_object != NULL) {
1112 			VM_OBJECT_LOCK(bo->bo_object);
1113 			vm_object_pip_wait(bo->bo_object, "bovlbx");
1114 			VM_OBJECT_UNLOCK(bo->bo_object);
1115 		}
1116 		BO_LOCK(bo);
1117 	} while (bo->bo_numoutput > 0);
1118 	BO_UNLOCK(bo);
1119 
1120 	/*
1121 	 * Destroy the copy in the VM cache, too.
1122 	 */
1123 	if (bo->bo_object != NULL) {
1124 		VM_OBJECT_LOCK(bo->bo_object);
1125 		vm_object_page_remove(bo->bo_object, 0, 0,
1126 			(flags & V_SAVE) ? TRUE : FALSE);
1127 		VM_OBJECT_UNLOCK(bo->bo_object);
1128 	}
1129 
1130 #ifdef INVARIANTS
1131 	BO_LOCK(bo);
1132 	if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1133 	    (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1134 		panic("vinvalbuf: flush failed");
1135 	BO_UNLOCK(bo);
1136 #endif
1137 	return (0);
1138 }
1139 
1140 /*
1141  * Flush out and invalidate all buffers associated with a vnode.
1142  * Called with the underlying object locked.
1143  */
1144 int
1145 vinvalbuf(struct vnode *vp, int flags, struct thread *td, int slpflag,
1146     int slptimeo)
1147 {
1148 
1149 	CTR2(KTR_VFS, "vinvalbuf vp %p flags %d", vp, flags);
1150 	ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1151 	return (bufobj_invalbuf(&vp->v_bufobj, flags, td, slpflag, slptimeo));
1152 }
1153 
1154 /*
1155  * Flush out buffers on the specified list.
1156  *
1157  */
1158 static int
1159 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1160     int slptimeo)
1161 {
1162 	struct buf *bp, *nbp;
1163 	int retval, error;
1164 	daddr_t lblkno;
1165 	b_xflags_t xflags;
1166 
1167 	ASSERT_BO_LOCKED(bo);
1168 
1169 	retval = 0;
1170 	TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1171 		if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1172 		    ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1173 			continue;
1174 		}
1175 		lblkno = 0;
1176 		xflags = 0;
1177 		if (nbp != NULL) {
1178 			lblkno = nbp->b_lblkno;
1179 			xflags = nbp->b_xflags &
1180 				(BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1181 		}
1182 		retval = EAGAIN;
1183 		error = BUF_TIMELOCK(bp,
1184 		    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1185 		    "flushbuf", slpflag, slptimeo);
1186 		if (error) {
1187 			BO_LOCK(bo);
1188 			return (error != ENOLCK ? error : EAGAIN);
1189 		}
1190 		KASSERT(bp->b_bufobj == bo,
1191 		    ("bp %p wrong b_bufobj %p should be %p",
1192 		    bp, bp->b_bufobj, bo));
1193 		if (bp->b_bufobj != bo) {	/* XXX: necessary ? */
1194 			BUF_UNLOCK(bp);
1195 			BO_LOCK(bo);
1196 			return (EAGAIN);
1197 		}
1198 		/*
1199 		 * XXX Since there are no node locks for NFS, I
1200 		 * believe there is a slight chance that a delayed
1201 		 * write will occur while sleeping just above, so
1202 		 * check for it.
1203 		 */
1204 		if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1205 		    (flags & V_SAVE)) {
1206 			bremfree(bp);
1207 			bp->b_flags |= B_ASYNC;
1208 			bwrite(bp);
1209 			BO_LOCK(bo);
1210 			return (EAGAIN);	/* XXX: why not loop ? */
1211 		}
1212 		bremfree(bp);
1213 		bp->b_flags |= (B_INVAL | B_RELBUF);
1214 		bp->b_flags &= ~B_ASYNC;
1215 		brelse(bp);
1216 		BO_LOCK(bo);
1217 		if (nbp != NULL &&
1218 		    (nbp->b_bufobj != bo ||
1219 		     nbp->b_lblkno != lblkno ||
1220 		     (nbp->b_xflags &
1221 		      (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1222 			break;			/* nbp invalid */
1223 	}
1224 	return (retval);
1225 }
1226 
1227 /*
1228  * Truncate a file's buffer and pages to a specified length.  This
1229  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1230  * sync activity.
1231  */
1232 int
1233 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
1234     off_t length, int blksize)
1235 {
1236 	struct buf *bp, *nbp;
1237 	int anyfreed;
1238 	int trunclbn;
1239 	struct bufobj *bo;
1240 
1241 	CTR2(KTR_VFS, "vtruncbuf vp %p length %jd", vp, length);
1242 	/*
1243 	 * Round up to the *next* lbn.
1244 	 */
1245 	trunclbn = (length + blksize - 1) / blksize;
1246 
1247 	ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1248 restart:
1249 	bo = &vp->v_bufobj;
1250 	BO_LOCK(bo);
1251 	anyfreed = 1;
1252 	for (;anyfreed;) {
1253 		anyfreed = 0;
1254 		TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1255 			if (bp->b_lblkno < trunclbn)
1256 				continue;
1257 			if (BUF_LOCK(bp,
1258 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1259 			    BO_MTX(bo)) == ENOLCK)
1260 				goto restart;
1261 
1262 			bremfree(bp);
1263 			bp->b_flags |= (B_INVAL | B_RELBUF);
1264 			bp->b_flags &= ~B_ASYNC;
1265 			brelse(bp);
1266 			anyfreed = 1;
1267 
1268 			if (nbp != NULL &&
1269 			    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1270 			    (nbp->b_vp != vp) ||
1271 			    (nbp->b_flags & B_DELWRI))) {
1272 				goto restart;
1273 			}
1274 			BO_LOCK(bo);
1275 		}
1276 
1277 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1278 			if (bp->b_lblkno < trunclbn)
1279 				continue;
1280 			if (BUF_LOCK(bp,
1281 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1282 			    BO_MTX(bo)) == ENOLCK)
1283 				goto restart;
1284 			bremfree(bp);
1285 			bp->b_flags |= (B_INVAL | B_RELBUF);
1286 			bp->b_flags &= ~B_ASYNC;
1287 			brelse(bp);
1288 			anyfreed = 1;
1289 			if (nbp != NULL &&
1290 			    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1291 			    (nbp->b_vp != vp) ||
1292 			    (nbp->b_flags & B_DELWRI) == 0)) {
1293 				goto restart;
1294 			}
1295 			BO_LOCK(bo);
1296 		}
1297 	}
1298 
1299 	if (length > 0) {
1300 restartsync:
1301 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1302 			if (bp->b_lblkno > 0)
1303 				continue;
1304 			/*
1305 			 * Since we hold the vnode lock this should only
1306 			 * fail if we're racing with the buf daemon.
1307 			 */
1308 			if (BUF_LOCK(bp,
1309 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1310 			    BO_MTX(bo)) == ENOLCK) {
1311 				goto restart;
1312 			}
1313 			VNASSERT((bp->b_flags & B_DELWRI), vp,
1314 			    ("buf(%p) on dirty queue without DELWRI", bp));
1315 
1316 			bremfree(bp);
1317 			bawrite(bp);
1318 			BO_LOCK(bo);
1319 			goto restartsync;
1320 		}
1321 	}
1322 
1323 	bufobj_wwait(bo, 0, 0);
1324 	BO_UNLOCK(bo);
1325 	vnode_pager_setsize(vp, length);
1326 
1327 	return (0);
1328 }
1329 
1330 /*
1331  * buf_splay() - splay tree core for the clean/dirty list of buffers in
1332  * 		 a vnode.
1333  *
1334  *	NOTE: We have to deal with the special case of a background bitmap
1335  *	buffer, a situation where two buffers will have the same logical
1336  *	block offset.  We want (1) only the foreground buffer to be accessed
1337  *	in a lookup and (2) must differentiate between the foreground and
1338  *	background buffer in the splay tree algorithm because the splay
1339  *	tree cannot normally handle multiple entities with the same 'index'.
1340  *	We accomplish this by adding differentiating flags to the splay tree's
1341  *	numerical domain.
1342  */
1343 static
1344 struct buf *
1345 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1346 {
1347 	struct buf dummy;
1348 	struct buf *lefttreemax, *righttreemin, *y;
1349 
1350 	if (root == NULL)
1351 		return (NULL);
1352 	lefttreemax = righttreemin = &dummy;
1353 	for (;;) {
1354 		if (lblkno < root->b_lblkno ||
1355 		    (lblkno == root->b_lblkno &&
1356 		    (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1357 			if ((y = root->b_left) == NULL)
1358 				break;
1359 			if (lblkno < y->b_lblkno) {
1360 				/* Rotate right. */
1361 				root->b_left = y->b_right;
1362 				y->b_right = root;
1363 				root = y;
1364 				if ((y = root->b_left) == NULL)
1365 					break;
1366 			}
1367 			/* Link into the new root's right tree. */
1368 			righttreemin->b_left = root;
1369 			righttreemin = root;
1370 		} else if (lblkno > root->b_lblkno ||
1371 		    (lblkno == root->b_lblkno &&
1372 		    (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1373 			if ((y = root->b_right) == NULL)
1374 				break;
1375 			if (lblkno > y->b_lblkno) {
1376 				/* Rotate left. */
1377 				root->b_right = y->b_left;
1378 				y->b_left = root;
1379 				root = y;
1380 				if ((y = root->b_right) == NULL)
1381 					break;
1382 			}
1383 			/* Link into the new root's left tree. */
1384 			lefttreemax->b_right = root;
1385 			lefttreemax = root;
1386 		} else {
1387 			break;
1388 		}
1389 		root = y;
1390 	}
1391 	/* Assemble the new root. */
1392 	lefttreemax->b_right = root->b_left;
1393 	righttreemin->b_left = root->b_right;
1394 	root->b_left = dummy.b_right;
1395 	root->b_right = dummy.b_left;
1396 	return (root);
1397 }
1398 
1399 static void
1400 buf_vlist_remove(struct buf *bp)
1401 {
1402 	struct buf *root;
1403 	struct bufv *bv;
1404 
1405 	KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1406 	ASSERT_BO_LOCKED(bp->b_bufobj);
1407 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1408 	    (BX_VNDIRTY|BX_VNCLEAN),
1409 	    ("buf_vlist_remove: Buf %p is on two lists", bp));
1410 	if (bp->b_xflags & BX_VNDIRTY)
1411 		bv = &bp->b_bufobj->bo_dirty;
1412 	else
1413 		bv = &bp->b_bufobj->bo_clean;
1414 	if (bp != bv->bv_root) {
1415 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1416 		KASSERT(root == bp, ("splay lookup failed in remove"));
1417 	}
1418 	if (bp->b_left == NULL) {
1419 		root = bp->b_right;
1420 	} else {
1421 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1422 		root->b_right = bp->b_right;
1423 	}
1424 	bv->bv_root = root;
1425 	TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1426 	bv->bv_cnt--;
1427 	bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1428 }
1429 
1430 /*
1431  * Add the buffer to the sorted clean or dirty block list using a
1432  * splay tree algorithm.
1433  *
1434  * NOTE: xflags is passed as a constant, optimizing this inline function!
1435  */
1436 static void
1437 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1438 {
1439 	struct buf *root;
1440 	struct bufv *bv;
1441 
1442 	ASSERT_BO_LOCKED(bo);
1443 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1444 	    ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1445 	bp->b_xflags |= xflags;
1446 	if (xflags & BX_VNDIRTY)
1447 		bv = &bo->bo_dirty;
1448 	else
1449 		bv = &bo->bo_clean;
1450 
1451 	root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1452 	if (root == NULL) {
1453 		bp->b_left = NULL;
1454 		bp->b_right = NULL;
1455 		TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1456 	} else if (bp->b_lblkno < root->b_lblkno ||
1457 	    (bp->b_lblkno == root->b_lblkno &&
1458 	    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1459 		bp->b_left = root->b_left;
1460 		bp->b_right = root;
1461 		root->b_left = NULL;
1462 		TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1463 	} else {
1464 		bp->b_right = root->b_right;
1465 		bp->b_left = root;
1466 		root->b_right = NULL;
1467 		TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1468 	}
1469 	bv->bv_cnt++;
1470 	bv->bv_root = bp;
1471 }
1472 
1473 /*
1474  * Lookup a buffer using the splay tree.  Note that we specifically avoid
1475  * shadow buffers used in background bitmap writes.
1476  *
1477  * This code isn't quite efficient as it could be because we are maintaining
1478  * two sorted lists and do not know which list the block resides in.
1479  *
1480  * During a "make buildworld" the desired buffer is found at one of
1481  * the roots more than 60% of the time.  Thus, checking both roots
1482  * before performing either splay eliminates unnecessary splays on the
1483  * first tree splayed.
1484  */
1485 struct buf *
1486 gbincore(struct bufobj *bo, daddr_t lblkno)
1487 {
1488 	struct buf *bp;
1489 
1490 	ASSERT_BO_LOCKED(bo);
1491 	if ((bp = bo->bo_clean.bv_root) != NULL &&
1492 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1493 		return (bp);
1494 	if ((bp = bo->bo_dirty.bv_root) != NULL &&
1495 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1496 		return (bp);
1497 	if ((bp = bo->bo_clean.bv_root) != NULL) {
1498 		bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1499 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1500 			return (bp);
1501 	}
1502 	if ((bp = bo->bo_dirty.bv_root) != NULL) {
1503 		bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1504 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1505 			return (bp);
1506 	}
1507 	return (NULL);
1508 }
1509 
1510 /*
1511  * Associate a buffer with a vnode.
1512  */
1513 void
1514 bgetvp(struct vnode *vp, struct buf *bp)
1515 {
1516 	struct bufobj *bo;
1517 
1518 	bo = &vp->v_bufobj;
1519 	ASSERT_BO_LOCKED(bo);
1520 	VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1521 
1522 	CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1523 	VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1524 	    ("bgetvp: bp already attached! %p", bp));
1525 
1526 	vhold(vp);
1527 	if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT)
1528 		bp->b_flags |= B_NEEDSGIANT;
1529 	bp->b_vp = vp;
1530 	bp->b_bufobj = bo;
1531 	/*
1532 	 * Insert onto list for new vnode.
1533 	 */
1534 	buf_vlist_add(bp, bo, BX_VNCLEAN);
1535 }
1536 
1537 /*
1538  * Disassociate a buffer from a vnode.
1539  */
1540 void
1541 brelvp(struct buf *bp)
1542 {
1543 	struct bufobj *bo;
1544 	struct vnode *vp;
1545 
1546 	CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1547 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1548 
1549 	/*
1550 	 * Delete from old vnode list, if on one.
1551 	 */
1552 	vp = bp->b_vp;		/* XXX */
1553 	bo = bp->b_bufobj;
1554 	BO_LOCK(bo);
1555 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1556 		buf_vlist_remove(bp);
1557 	else
1558 		panic("brelvp: Buffer %p not on queue.", bp);
1559 	if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1560 		bo->bo_flag &= ~BO_ONWORKLST;
1561 		mtx_lock(&sync_mtx);
1562 		LIST_REMOVE(bo, bo_synclist);
1563 		syncer_worklist_len--;
1564 		mtx_unlock(&sync_mtx);
1565 	}
1566 	bp->b_flags &= ~B_NEEDSGIANT;
1567 	bp->b_vp = NULL;
1568 	bp->b_bufobj = NULL;
1569 	BO_UNLOCK(bo);
1570 	vdrop(vp);
1571 }
1572 
1573 /*
1574  * Add an item to the syncer work queue.
1575  */
1576 static void
1577 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1578 {
1579 	int queue, slot;
1580 
1581 	ASSERT_BO_LOCKED(bo);
1582 
1583 	mtx_lock(&sync_mtx);
1584 	if (bo->bo_flag & BO_ONWORKLST)
1585 		LIST_REMOVE(bo, bo_synclist);
1586 	else {
1587 		bo->bo_flag |= BO_ONWORKLST;
1588 		syncer_worklist_len++;
1589 	}
1590 
1591 	if (delay > syncer_maxdelay - 2)
1592 		delay = syncer_maxdelay - 2;
1593 	slot = (syncer_delayno + delay) & syncer_mask;
1594 
1595 	queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ :
1596 	    WI_MPSAFEQ;
1597 	LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo,
1598 	    bo_synclist);
1599 	mtx_unlock(&sync_mtx);
1600 }
1601 
1602 static int
1603 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1604 {
1605 	int error, len;
1606 
1607 	mtx_lock(&sync_mtx);
1608 	len = syncer_worklist_len - sync_vnode_count;
1609 	mtx_unlock(&sync_mtx);
1610 	error = SYSCTL_OUT(req, &len, sizeof(len));
1611 	return (error);
1612 }
1613 
1614 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1615     sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1616 
1617 static struct proc *updateproc;
1618 static void sched_sync(void);
1619 static struct kproc_desc up_kp = {
1620 	"syncer",
1621 	sched_sync,
1622 	&updateproc
1623 };
1624 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1625 
1626 static int
1627 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1628 {
1629 	struct vnode *vp;
1630 	struct mount *mp;
1631 
1632 	*bo = LIST_FIRST(slp);
1633 	if (*bo == NULL)
1634 		return (0);
1635 	vp = (*bo)->__bo_vnode;	/* XXX */
1636 	if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1637 		return (1);
1638 	/*
1639 	 * We use vhold in case the vnode does not
1640 	 * successfully sync.  vhold prevents the vnode from
1641 	 * going away when we unlock the sync_mtx so that
1642 	 * we can acquire the vnode interlock.
1643 	 */
1644 	vholdl(vp);
1645 	mtx_unlock(&sync_mtx);
1646 	VI_UNLOCK(vp);
1647 	if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1648 		vdrop(vp);
1649 		mtx_lock(&sync_mtx);
1650 		return (*bo == LIST_FIRST(slp));
1651 	}
1652 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1653 	(void) VOP_FSYNC(vp, MNT_LAZY, td);
1654 	VOP_UNLOCK(vp, 0);
1655 	vn_finished_write(mp);
1656 	BO_LOCK(*bo);
1657 	if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1658 		/*
1659 		 * Put us back on the worklist.  The worklist
1660 		 * routine will remove us from our current
1661 		 * position and then add us back in at a later
1662 		 * position.
1663 		 */
1664 		vn_syncer_add_to_worklist(*bo, syncdelay);
1665 	}
1666 	BO_UNLOCK(*bo);
1667 	vdrop(vp);
1668 	mtx_lock(&sync_mtx);
1669 	return (0);
1670 }
1671 
1672 /*
1673  * System filesystem synchronizer daemon.
1674  */
1675 static void
1676 sched_sync(void)
1677 {
1678 	struct synclist *gnext, *next;
1679 	struct synclist *gslp, *slp;
1680 	struct bufobj *bo;
1681 	long starttime;
1682 	struct thread *td = curthread;
1683 	int last_work_seen;
1684 	int net_worklist_len;
1685 	int syncer_final_iter;
1686 	int first_printf;
1687 	int error;
1688 
1689 	last_work_seen = 0;
1690 	syncer_final_iter = 0;
1691 	first_printf = 1;
1692 	syncer_state = SYNCER_RUNNING;
1693 	starttime = time_uptime;
1694 	td->td_pflags |= TDP_NORUNNINGBUF;
1695 
1696 	EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1697 	    SHUTDOWN_PRI_LAST);
1698 
1699 	mtx_lock(&sync_mtx);
1700 	for (;;) {
1701 		if (syncer_state == SYNCER_FINAL_DELAY &&
1702 		    syncer_final_iter == 0) {
1703 			mtx_unlock(&sync_mtx);
1704 			kproc_suspend_check(td->td_proc);
1705 			mtx_lock(&sync_mtx);
1706 		}
1707 		net_worklist_len = syncer_worklist_len - sync_vnode_count;
1708 		if (syncer_state != SYNCER_RUNNING &&
1709 		    starttime != time_uptime) {
1710 			if (first_printf) {
1711 				printf("\nSyncing disks, vnodes remaining...");
1712 				first_printf = 0;
1713 			}
1714 			printf("%d ", net_worklist_len);
1715 		}
1716 		starttime = time_uptime;
1717 
1718 		/*
1719 		 * Push files whose dirty time has expired.  Be careful
1720 		 * of interrupt race on slp queue.
1721 		 *
1722 		 * Skip over empty worklist slots when shutting down.
1723 		 */
1724 		do {
1725 			slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1726 			gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1727 			syncer_delayno += 1;
1728 			if (syncer_delayno == syncer_maxdelay)
1729 				syncer_delayno = 0;
1730 			next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1731 			gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1732 			/*
1733 			 * If the worklist has wrapped since the
1734 			 * it was emptied of all but syncer vnodes,
1735 			 * switch to the FINAL_DELAY state and run
1736 			 * for one more second.
1737 			 */
1738 			if (syncer_state == SYNCER_SHUTTING_DOWN &&
1739 			    net_worklist_len == 0 &&
1740 			    last_work_seen == syncer_delayno) {
1741 				syncer_state = SYNCER_FINAL_DELAY;
1742 				syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1743 			}
1744 		} while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1745 		    LIST_EMPTY(gslp) && syncer_worklist_len > 0);
1746 
1747 		/*
1748 		 * Keep track of the last time there was anything
1749 		 * on the worklist other than syncer vnodes.
1750 		 * Return to the SHUTTING_DOWN state if any
1751 		 * new work appears.
1752 		 */
1753 		if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1754 			last_work_seen = syncer_delayno;
1755 		if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1756 			syncer_state = SYNCER_SHUTTING_DOWN;
1757 		while (!LIST_EMPTY(slp)) {
1758 			error = sync_vnode(slp, &bo, td);
1759 			if (error == 1) {
1760 				LIST_REMOVE(bo, bo_synclist);
1761 				LIST_INSERT_HEAD(next, bo, bo_synclist);
1762 				continue;
1763 			}
1764 		}
1765 		if (!LIST_EMPTY(gslp)) {
1766 			mtx_unlock(&sync_mtx);
1767 			mtx_lock(&Giant);
1768 			mtx_lock(&sync_mtx);
1769 			while (!LIST_EMPTY(gslp)) {
1770 				error = sync_vnode(gslp, &bo, td);
1771 				if (error == 1) {
1772 					LIST_REMOVE(bo, bo_synclist);
1773 					LIST_INSERT_HEAD(gnext, bo,
1774 					    bo_synclist);
1775 					continue;
1776 				}
1777 			}
1778 			mtx_unlock(&Giant);
1779 		}
1780 		if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1781 			syncer_final_iter--;
1782 		/*
1783 		 * The variable rushjob allows the kernel to speed up the
1784 		 * processing of the filesystem syncer process. A rushjob
1785 		 * value of N tells the filesystem syncer to process the next
1786 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1787 		 * is used by the soft update code to speed up the filesystem
1788 		 * syncer process when the incore state is getting so far
1789 		 * ahead of the disk that the kernel memory pool is being
1790 		 * threatened with exhaustion.
1791 		 */
1792 		if (rushjob > 0) {
1793 			rushjob -= 1;
1794 			continue;
1795 		}
1796 		/*
1797 		 * Just sleep for a short period of time between
1798 		 * iterations when shutting down to allow some I/O
1799 		 * to happen.
1800 		 *
1801 		 * If it has taken us less than a second to process the
1802 		 * current work, then wait. Otherwise start right over
1803 		 * again. We can still lose time if any single round
1804 		 * takes more than two seconds, but it does not really
1805 		 * matter as we are just trying to generally pace the
1806 		 * filesystem activity.
1807 		 */
1808 		if (syncer_state != SYNCER_RUNNING)
1809 			cv_timedwait(&sync_wakeup, &sync_mtx,
1810 			    hz / SYNCER_SHUTDOWN_SPEEDUP);
1811 		else if (time_uptime == starttime)
1812 			cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1813 	}
1814 }
1815 
1816 /*
1817  * Request the syncer daemon to speed up its work.
1818  * We never push it to speed up more than half of its
1819  * normal turn time, otherwise it could take over the cpu.
1820  */
1821 int
1822 speedup_syncer(void)
1823 {
1824 	int ret = 0;
1825 
1826 	mtx_lock(&sync_mtx);
1827 	if (rushjob < syncdelay / 2) {
1828 		rushjob += 1;
1829 		stat_rush_requests += 1;
1830 		ret = 1;
1831 	}
1832 	mtx_unlock(&sync_mtx);
1833 	cv_broadcast(&sync_wakeup);
1834 	return (ret);
1835 }
1836 
1837 /*
1838  * Tell the syncer to speed up its work and run though its work
1839  * list several times, then tell it to shut down.
1840  */
1841 static void
1842 syncer_shutdown(void *arg, int howto)
1843 {
1844 
1845 	if (howto & RB_NOSYNC)
1846 		return;
1847 	mtx_lock(&sync_mtx);
1848 	syncer_state = SYNCER_SHUTTING_DOWN;
1849 	rushjob = 0;
1850 	mtx_unlock(&sync_mtx);
1851 	cv_broadcast(&sync_wakeup);
1852 	kproc_shutdown(arg, howto);
1853 }
1854 
1855 /*
1856  * Reassign a buffer from one vnode to another.
1857  * Used to assign file specific control information
1858  * (indirect blocks) to the vnode to which they belong.
1859  */
1860 void
1861 reassignbuf(struct buf *bp)
1862 {
1863 	struct vnode *vp;
1864 	struct bufobj *bo;
1865 	int delay;
1866 #ifdef INVARIANTS
1867 	struct bufv *bv;
1868 #endif
1869 
1870 	vp = bp->b_vp;
1871 	bo = bp->b_bufobj;
1872 	++reassignbufcalls;
1873 
1874 	CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1875 	    bp, bp->b_vp, bp->b_flags);
1876 	/*
1877 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1878 	 * is not fully linked in.
1879 	 */
1880 	if (bp->b_flags & B_PAGING)
1881 		panic("cannot reassign paging buffer");
1882 
1883 	/*
1884 	 * Delete from old vnode list, if on one.
1885 	 */
1886 	BO_LOCK(bo);
1887 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1888 		buf_vlist_remove(bp);
1889 	else
1890 		panic("reassignbuf: Buffer %p not on queue.", bp);
1891 	/*
1892 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1893 	 * of clean buffers.
1894 	 */
1895 	if (bp->b_flags & B_DELWRI) {
1896 		if ((bo->bo_flag & BO_ONWORKLST) == 0) {
1897 			switch (vp->v_type) {
1898 			case VDIR:
1899 				delay = dirdelay;
1900 				break;
1901 			case VCHR:
1902 				delay = metadelay;
1903 				break;
1904 			default:
1905 				delay = filedelay;
1906 			}
1907 			vn_syncer_add_to_worklist(bo, delay);
1908 		}
1909 		buf_vlist_add(bp, bo, BX_VNDIRTY);
1910 	} else {
1911 		buf_vlist_add(bp, bo, BX_VNCLEAN);
1912 
1913 		if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1914 			mtx_lock(&sync_mtx);
1915 			LIST_REMOVE(bo, bo_synclist);
1916 			syncer_worklist_len--;
1917 			mtx_unlock(&sync_mtx);
1918 			bo->bo_flag &= ~BO_ONWORKLST;
1919 		}
1920 	}
1921 #ifdef INVARIANTS
1922 	bv = &bo->bo_clean;
1923 	bp = TAILQ_FIRST(&bv->bv_hd);
1924 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1925 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1926 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
1927 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1928 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1929 	bv = &bo->bo_dirty;
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 #endif
1937 	BO_UNLOCK(bo);
1938 }
1939 
1940 /*
1941  * Increment the use and hold counts on the vnode, taking care to reference
1942  * the driver's usecount if this is a chardev.  The vholdl() will remove
1943  * the vnode from the free list if it is presently free.  Requires the
1944  * vnode interlock and returns with it held.
1945  */
1946 static void
1947 v_incr_usecount(struct vnode *vp)
1948 {
1949 
1950 	CTR3(KTR_VFS, "v_incr_usecount: vp %p holdcnt %d usecount %d\n",
1951 	    vp, vp->v_holdcnt, vp->v_usecount);
1952 	vp->v_usecount++;
1953 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1954 		dev_lock();
1955 		vp->v_rdev->si_usecount++;
1956 		dev_unlock();
1957 	}
1958 	vholdl(vp);
1959 }
1960 
1961 /*
1962  * Turn a holdcnt into a use+holdcnt such that only one call to
1963  * v_decr_usecount is needed.
1964  */
1965 static void
1966 v_upgrade_usecount(struct vnode *vp)
1967 {
1968 
1969 	CTR3(KTR_VFS, "v_upgrade_usecount: vp %p holdcnt %d usecount %d\n",
1970 	    vp, vp->v_holdcnt, vp->v_usecount);
1971 	vp->v_usecount++;
1972 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1973 		dev_lock();
1974 		vp->v_rdev->si_usecount++;
1975 		dev_unlock();
1976 	}
1977 }
1978 
1979 /*
1980  * Decrement the vnode use and hold count along with the driver's usecount
1981  * if this is a chardev.  The vdropl() below releases the vnode interlock
1982  * as it may free the vnode.
1983  */
1984 static void
1985 v_decr_usecount(struct vnode *vp)
1986 {
1987 
1988 	CTR3(KTR_VFS, "v_decr_usecount: vp %p holdcnt %d usecount %d\n",
1989 	    vp, vp->v_holdcnt, vp->v_usecount);
1990 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
1991 	VNASSERT(vp->v_usecount > 0, vp,
1992 	    ("v_decr_usecount: negative usecount"));
1993 	vp->v_usecount--;
1994 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1995 		dev_lock();
1996 		vp->v_rdev->si_usecount--;
1997 		dev_unlock();
1998 	}
1999 	vdropl(vp);
2000 }
2001 
2002 /*
2003  * Decrement only the use count and driver use count.  This is intended to
2004  * be paired with a follow on vdropl() to release the remaining hold count.
2005  * In this way we may vgone() a vnode with a 0 usecount without risk of
2006  * having it end up on a free list because the hold count is kept above 0.
2007  */
2008 static void
2009 v_decr_useonly(struct vnode *vp)
2010 {
2011 
2012 	CTR3(KTR_VFS, "v_decr_useonly: vp %p holdcnt %d usecount %d\n",
2013 	    vp, vp->v_holdcnt, vp->v_usecount);
2014 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2015 	VNASSERT(vp->v_usecount > 0, vp,
2016 	    ("v_decr_useonly: negative usecount"));
2017 	vp->v_usecount--;
2018 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2019 		dev_lock();
2020 		vp->v_rdev->si_usecount--;
2021 		dev_unlock();
2022 	}
2023 }
2024 
2025 /*
2026  * Grab a particular vnode from the free list, increment its
2027  * reference count and lock it.  VI_DOOMED is set if the vnode
2028  * is being destroyed.  Only callers who specify LK_RETRY will
2029  * see doomed vnodes.  If inactive processing was delayed in
2030  * vput try to do it here.
2031  */
2032 int
2033 vget(struct vnode *vp, int flags, struct thread *td)
2034 {
2035 	int error;
2036 
2037 	error = 0;
2038 	VFS_ASSERT_GIANT(vp->v_mount);
2039 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2040 	    ("vget: invalid lock operation"));
2041 	if ((flags & LK_INTERLOCK) == 0)
2042 		VI_LOCK(vp);
2043 	vholdl(vp);
2044 	if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2045 		vdrop(vp);
2046 		return (error);
2047 	}
2048 	if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2049 		panic("vget: vn_lock failed to return ENOENT\n");
2050 	VI_LOCK(vp);
2051 	/* Upgrade our holdcnt to a usecount. */
2052 	v_upgrade_usecount(vp);
2053 	/*
2054  	 * We don't guarantee that any particular close will
2055 	 * trigger inactive processing so just make a best effort
2056 	 * here at preventing a reference to a removed file.  If
2057 	 * we don't succeed no harm is done.
2058 	 */
2059 	if (vp->v_iflag & VI_OWEINACT) {
2060 		if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2061 		    (flags & LK_NOWAIT) == 0)
2062 			vinactive(vp, td);
2063 		vp->v_iflag &= ~VI_OWEINACT;
2064 	}
2065 	VI_UNLOCK(vp);
2066 	return (0);
2067 }
2068 
2069 /*
2070  * Increase the reference count of a vnode.
2071  */
2072 void
2073 vref(struct vnode *vp)
2074 {
2075 
2076 	VI_LOCK(vp);
2077 	v_incr_usecount(vp);
2078 	VI_UNLOCK(vp);
2079 }
2080 
2081 /*
2082  * Return reference count of a vnode.
2083  *
2084  * The results of this call are only guaranteed when some mechanism other
2085  * than the VI lock is used to stop other processes from gaining references
2086  * to the vnode.  This may be the case if the caller holds the only reference.
2087  * This is also useful when stale data is acceptable as race conditions may
2088  * be accounted for by some other means.
2089  */
2090 int
2091 vrefcnt(struct vnode *vp)
2092 {
2093 	int usecnt;
2094 
2095 	VI_LOCK(vp);
2096 	usecnt = vp->v_usecount;
2097 	VI_UNLOCK(vp);
2098 
2099 	return (usecnt);
2100 }
2101 
2102 
2103 /*
2104  * Vnode put/release.
2105  * If count drops to zero, call inactive routine and return to freelist.
2106  */
2107 void
2108 vrele(struct vnode *vp)
2109 {
2110 	struct thread *td = curthread;	/* XXX */
2111 
2112 	KASSERT(vp != NULL, ("vrele: null vp"));
2113 	VFS_ASSERT_GIANT(vp->v_mount);
2114 
2115 	VI_LOCK(vp);
2116 
2117 	/* Skip this v_writecount check if we're going to panic below. */
2118 	VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2119 	    ("vrele: missed vn_close"));
2120 
2121 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2122 	    vp->v_usecount == 1)) {
2123 		v_decr_usecount(vp);
2124 		return;
2125 	}
2126 	if (vp->v_usecount != 1) {
2127 #ifdef DIAGNOSTIC
2128 		vprint("vrele: negative ref count", vp);
2129 #endif
2130 		VI_UNLOCK(vp);
2131 		panic("vrele: negative ref cnt");
2132 	}
2133 	/*
2134 	 * We want to hold the vnode until the inactive finishes to
2135 	 * prevent vgone() races.  We drop the use count here and the
2136 	 * hold count below when we're done.
2137 	 */
2138 	v_decr_useonly(vp);
2139 	/*
2140 	 * We must call VOP_INACTIVE with the node locked. Mark
2141 	 * as VI_DOINGINACT to avoid recursion.
2142 	 */
2143 	vp->v_iflag |= VI_OWEINACT;
2144 	if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK) == 0) {
2145 		VI_LOCK(vp);
2146 		if (vp->v_usecount > 0)
2147 			vp->v_iflag &= ~VI_OWEINACT;
2148 		if (vp->v_iflag & VI_OWEINACT)
2149 			vinactive(vp, td);
2150 		VOP_UNLOCK(vp, 0);
2151 	} else {
2152 		VI_LOCK(vp);
2153 		if (vp->v_usecount > 0)
2154 			vp->v_iflag &= ~VI_OWEINACT;
2155 	}
2156 	vdropl(vp);
2157 }
2158 
2159 /*
2160  * Release an already locked vnode.  This give the same effects as
2161  * unlock+vrele(), but takes less time and avoids releasing and
2162  * re-aquiring the lock (as vrele() acquires the lock internally.)
2163  */
2164 void
2165 vput(struct vnode *vp)
2166 {
2167 	struct thread *td = curthread;	/* XXX */
2168 	int error;
2169 
2170 	KASSERT(vp != NULL, ("vput: null vp"));
2171 	ASSERT_VOP_LOCKED(vp, "vput");
2172 	VFS_ASSERT_GIANT(vp->v_mount);
2173 	VI_LOCK(vp);
2174 	/* Skip this v_writecount check if we're going to panic below. */
2175 	VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2176 	    ("vput: missed vn_close"));
2177 	error = 0;
2178 
2179 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2180 	    vp->v_usecount == 1)) {
2181 		VOP_UNLOCK(vp, 0);
2182 		v_decr_usecount(vp);
2183 		return;
2184 	}
2185 
2186 	if (vp->v_usecount != 1) {
2187 #ifdef DIAGNOSTIC
2188 		vprint("vput: negative ref count", vp);
2189 #endif
2190 		panic("vput: negative ref cnt");
2191 	}
2192 	/*
2193 	 * We want to hold the vnode until the inactive finishes to
2194 	 * prevent vgone() races.  We drop the use count here and the
2195 	 * hold count below when we're done.
2196 	 */
2197 	v_decr_useonly(vp);
2198 	vp->v_iflag |= VI_OWEINACT;
2199 	if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2200 		error = VOP_LOCK(vp, LK_UPGRADE|LK_INTERLOCK|LK_NOWAIT);
2201 		VI_LOCK(vp);
2202 		if (error) {
2203 			if (vp->v_usecount > 0)
2204 				vp->v_iflag &= ~VI_OWEINACT;
2205 			goto done;
2206 		}
2207 	}
2208 	if (vp->v_usecount > 0)
2209 		vp->v_iflag &= ~VI_OWEINACT;
2210 	if (vp->v_iflag & VI_OWEINACT)
2211 		vinactive(vp, td);
2212 	VOP_UNLOCK(vp, 0);
2213 done:
2214 	vdropl(vp);
2215 }
2216 
2217 /*
2218  * Somebody doesn't want the vnode recycled.
2219  */
2220 void
2221 vhold(struct vnode *vp)
2222 {
2223 
2224 	VI_LOCK(vp);
2225 	vholdl(vp);
2226 	VI_UNLOCK(vp);
2227 }
2228 
2229 void
2230 vholdl(struct vnode *vp)
2231 {
2232 
2233 	vp->v_holdcnt++;
2234 	if (VSHOULDBUSY(vp))
2235 		vbusy(vp);
2236 }
2237 
2238 /*
2239  * Note that there is one less who cares about this vnode.  vdrop() is the
2240  * opposite of vhold().
2241  */
2242 void
2243 vdrop(struct vnode *vp)
2244 {
2245 
2246 	VI_LOCK(vp);
2247 	vdropl(vp);
2248 }
2249 
2250 /*
2251  * Drop the hold count of the vnode.  If this is the last reference to
2252  * the vnode we will free it if it has been vgone'd otherwise it is
2253  * placed on the free list.
2254  */
2255 void
2256 vdropl(struct vnode *vp)
2257 {
2258 
2259 	ASSERT_VI_LOCKED(vp, "vdropl");
2260 	if (vp->v_holdcnt <= 0)
2261 		panic("vdrop: holdcnt %d", vp->v_holdcnt);
2262 	vp->v_holdcnt--;
2263 	if (vp->v_holdcnt == 0) {
2264 		if (vp->v_iflag & VI_DOOMED) {
2265 			vdestroy(vp);
2266 			return;
2267 		} else
2268 			vfree(vp);
2269 	}
2270 	VI_UNLOCK(vp);
2271 }
2272 
2273 /*
2274  * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2275  * flags.  DOINGINACT prevents us from recursing in calls to vinactive.
2276  * OWEINACT tracks whether a vnode missed a call to inactive due to a
2277  * failed lock upgrade.
2278  */
2279 static void
2280 vinactive(struct vnode *vp, struct thread *td)
2281 {
2282 
2283 	ASSERT_VOP_ELOCKED(vp, "vinactive");
2284 	ASSERT_VI_LOCKED(vp, "vinactive");
2285 	VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2286 	    ("vinactive: recursed on VI_DOINGINACT"));
2287 	vp->v_iflag |= VI_DOINGINACT;
2288 	vp->v_iflag &= ~VI_OWEINACT;
2289 	VI_UNLOCK(vp);
2290 	VOP_INACTIVE(vp, td);
2291 	VI_LOCK(vp);
2292 	VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2293 	    ("vinactive: lost VI_DOINGINACT"));
2294 	vp->v_iflag &= ~VI_DOINGINACT;
2295 }
2296 
2297 /*
2298  * Remove any vnodes in the vnode table belonging to mount point mp.
2299  *
2300  * If FORCECLOSE is not specified, there should not be any active ones,
2301  * return error if any are found (nb: this is a user error, not a
2302  * system error). If FORCECLOSE is specified, detach any active vnodes
2303  * that are found.
2304  *
2305  * If WRITECLOSE is set, only flush out regular file vnodes open for
2306  * writing.
2307  *
2308  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2309  *
2310  * `rootrefs' specifies the base reference count for the root vnode
2311  * of this filesystem. The root vnode is considered busy if its
2312  * v_usecount exceeds this value. On a successful return, vflush(, td)
2313  * will call vrele() on the root vnode exactly rootrefs times.
2314  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2315  * be zero.
2316  */
2317 #ifdef DIAGNOSTIC
2318 static int busyprt = 0;		/* print out busy vnodes */
2319 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
2320 #endif
2321 
2322 int
2323 vflush( struct mount *mp, int rootrefs, int flags, struct thread *td)
2324 {
2325 	struct vnode *vp, *mvp, *rootvp = NULL;
2326 	struct vattr vattr;
2327 	int busy = 0, error;
2328 
2329 	CTR1(KTR_VFS, "vflush: mp %p", mp);
2330 	if (rootrefs > 0) {
2331 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2332 		    ("vflush: bad args"));
2333 		/*
2334 		 * Get the filesystem root vnode. We can vput() it
2335 		 * immediately, since with rootrefs > 0, it won't go away.
2336 		 */
2337 		if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp, td)) != 0)
2338 			return (error);
2339 		vput(rootvp);
2340 
2341 	}
2342 	MNT_ILOCK(mp);
2343 loop:
2344 	MNT_VNODE_FOREACH(vp, mp, mvp) {
2345 
2346 		VI_LOCK(vp);
2347 		vholdl(vp);
2348 		MNT_IUNLOCK(mp);
2349 		error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2350 		if (error) {
2351 			vdrop(vp);
2352 			MNT_ILOCK(mp);
2353 			MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
2354 			goto loop;
2355 		}
2356 		/*
2357 		 * Skip over a vnodes marked VV_SYSTEM.
2358 		 */
2359 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2360 			VOP_UNLOCK(vp, 0);
2361 			vdrop(vp);
2362 			MNT_ILOCK(mp);
2363 			continue;
2364 		}
2365 		/*
2366 		 * If WRITECLOSE is set, flush out unlinked but still open
2367 		 * files (even if open only for reading) and regular file
2368 		 * vnodes open for writing.
2369 		 */
2370 		if (flags & WRITECLOSE) {
2371 			error = VOP_GETATTR(vp, &vattr, td->td_ucred, td);
2372 			VI_LOCK(vp);
2373 
2374 			if ((vp->v_type == VNON ||
2375 			    (error == 0 && vattr.va_nlink > 0)) &&
2376 			    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2377 				VOP_UNLOCK(vp, 0);
2378 				vdropl(vp);
2379 				MNT_ILOCK(mp);
2380 				continue;
2381 			}
2382 		} else
2383 			VI_LOCK(vp);
2384 		/*
2385 		 * With v_usecount == 0, all we need to do is clear out the
2386 		 * vnode data structures and we are done.
2387 		 *
2388 		 * If FORCECLOSE is set, forcibly close the vnode.
2389 		 */
2390 		if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2391 			VNASSERT(vp->v_usecount == 0 ||
2392 			    (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2393 			    ("device VNODE %p is FORCECLOSED", vp));
2394 			vgonel(vp);
2395 		} else {
2396 			busy++;
2397 #ifdef DIAGNOSTIC
2398 			if (busyprt)
2399 				vprint("vflush: busy vnode", vp);
2400 #endif
2401 		}
2402 		VOP_UNLOCK(vp, 0);
2403 		vdropl(vp);
2404 		MNT_ILOCK(mp);
2405 	}
2406 	MNT_IUNLOCK(mp);
2407 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2408 		/*
2409 		 * If just the root vnode is busy, and if its refcount
2410 		 * is equal to `rootrefs', then go ahead and kill it.
2411 		 */
2412 		VI_LOCK(rootvp);
2413 		KASSERT(busy > 0, ("vflush: not busy"));
2414 		VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2415 		    ("vflush: usecount %d < rootrefs %d",
2416 		     rootvp->v_usecount, rootrefs));
2417 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2418 			VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2419 			vgone(rootvp);
2420 			VOP_UNLOCK(rootvp, 0);
2421 			busy = 0;
2422 		} else
2423 			VI_UNLOCK(rootvp);
2424 	}
2425 	if (busy)
2426 		return (EBUSY);
2427 	for (; rootrefs > 0; rootrefs--)
2428 		vrele(rootvp);
2429 	return (0);
2430 }
2431 
2432 /*
2433  * Recycle an unused vnode to the front of the free list.
2434  */
2435 int
2436 vrecycle(struct vnode *vp, struct thread *td)
2437 {
2438 	int recycled;
2439 
2440 	ASSERT_VOP_ELOCKED(vp, "vrecycle");
2441 	recycled = 0;
2442 	VI_LOCK(vp);
2443 	if (vp->v_usecount == 0) {
2444 		recycled = 1;
2445 		vgonel(vp);
2446 	}
2447 	VI_UNLOCK(vp);
2448 	return (recycled);
2449 }
2450 
2451 /*
2452  * Eliminate all activity associated with a vnode
2453  * in preparation for reuse.
2454  */
2455 void
2456 vgone(struct vnode *vp)
2457 {
2458 	VI_LOCK(vp);
2459 	vgonel(vp);
2460 	VI_UNLOCK(vp);
2461 }
2462 
2463 /*
2464  * vgone, with the vp interlock held.
2465  */
2466 void
2467 vgonel(struct vnode *vp)
2468 {
2469 	struct thread *td;
2470 	int oweinact;
2471 	int active;
2472 	struct mount *mp;
2473 
2474 	CTR1(KTR_VFS, "vgonel: vp %p", vp);
2475 	ASSERT_VOP_ELOCKED(vp, "vgonel");
2476 	ASSERT_VI_LOCKED(vp, "vgonel");
2477 	VNASSERT(vp->v_holdcnt, vp,
2478 	    ("vgonel: vp %p has no reference.", vp));
2479 	td = curthread;
2480 
2481 	/*
2482 	 * Don't vgonel if we're already doomed.
2483 	 */
2484 	if (vp->v_iflag & VI_DOOMED)
2485 		return;
2486 	vp->v_iflag |= VI_DOOMED;
2487 	/*
2488 	 * Check to see if the vnode is in use.  If so, we have to call
2489 	 * VOP_CLOSE() and VOP_INACTIVE().
2490 	 */
2491 	active = vp->v_usecount;
2492 	oweinact = (vp->v_iflag & VI_OWEINACT);
2493 	VI_UNLOCK(vp);
2494 	/*
2495 	 * Clean out any buffers associated with the vnode.
2496 	 * If the flush fails, just toss the buffers.
2497 	 */
2498 	mp = NULL;
2499 	if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2500 		(void) vn_start_secondary_write(vp, &mp, V_WAIT);
2501 	if (vinvalbuf(vp, V_SAVE, td, 0, 0) != 0)
2502 		vinvalbuf(vp, 0, td, 0, 0);
2503 
2504 	/*
2505 	 * If purging an active vnode, it must be closed and
2506 	 * deactivated before being reclaimed.
2507 	 */
2508 	if (active)
2509 		VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2510 	if (oweinact || active) {
2511 		VI_LOCK(vp);
2512 		if ((vp->v_iflag & VI_DOINGINACT) == 0)
2513 			vinactive(vp, td);
2514 		VI_UNLOCK(vp);
2515 	}
2516 	/*
2517 	 * Reclaim the vnode.
2518 	 */
2519 	if (VOP_RECLAIM(vp, td))
2520 		panic("vgone: cannot reclaim");
2521 	if (mp != NULL)
2522 		vn_finished_secondary_write(mp);
2523 	VNASSERT(vp->v_object == NULL, vp,
2524 	    ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2525 	/*
2526 	 * Clear the advisory locks and wake up waiting threads.
2527 	 */
2528 	lf_purgelocks(vp, &(vp->v_lockf));
2529 	/*
2530 	 * Delete from old mount point vnode list.
2531 	 */
2532 	delmntque(vp);
2533 	cache_purge(vp);
2534 	/*
2535 	 * Done with purge, reset to the standard lock and invalidate
2536 	 * the vnode.
2537 	 */
2538 	VI_LOCK(vp);
2539 	vp->v_vnlock = &vp->v_lock;
2540 	vp->v_op = &dead_vnodeops;
2541 	vp->v_tag = "none";
2542 	vp->v_type = VBAD;
2543 }
2544 
2545 /*
2546  * Calculate the total number of references to a special device.
2547  */
2548 int
2549 vcount(struct vnode *vp)
2550 {
2551 	int count;
2552 
2553 	dev_lock();
2554 	count = vp->v_rdev->si_usecount;
2555 	dev_unlock();
2556 	return (count);
2557 }
2558 
2559 /*
2560  * Same as above, but using the struct cdev *as argument
2561  */
2562 int
2563 count_dev(struct cdev *dev)
2564 {
2565 	int count;
2566 
2567 	dev_lock();
2568 	count = dev->si_usecount;
2569 	dev_unlock();
2570 	return(count);
2571 }
2572 
2573 /*
2574  * Print out a description of a vnode.
2575  */
2576 static char *typename[] =
2577 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2578  "VMARKER"};
2579 
2580 void
2581 vn_printf(struct vnode *vp, const char *fmt, ...)
2582 {
2583 	va_list ap;
2584 	char buf[256], buf2[16];
2585 	u_long flags;
2586 
2587 	va_start(ap, fmt);
2588 	vprintf(fmt, ap);
2589 	va_end(ap);
2590 	printf("%p: ", (void *)vp);
2591 	printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2592 	printf("    usecount %d, writecount %d, refcount %d mountedhere %p\n",
2593 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2594 	buf[0] = '\0';
2595 	buf[1] = '\0';
2596 	if (vp->v_vflag & VV_ROOT)
2597 		strlcat(buf, "|VV_ROOT", sizeof(buf));
2598 	if (vp->v_vflag & VV_ISTTY)
2599 		strlcat(buf, "|VV_ISTTY", sizeof(buf));
2600 	if (vp->v_vflag & VV_NOSYNC)
2601 		strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2602 	if (vp->v_vflag & VV_CACHEDLABEL)
2603 		strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2604 	if (vp->v_vflag & VV_TEXT)
2605 		strlcat(buf, "|VV_TEXT", sizeof(buf));
2606 	if (vp->v_vflag & VV_COPYONWRITE)
2607 		strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2608 	if (vp->v_vflag & VV_SYSTEM)
2609 		strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2610 	if (vp->v_vflag & VV_PROCDEP)
2611 		strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2612 	if (vp->v_vflag & VV_NOKNOTE)
2613 		strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2614 	if (vp->v_vflag & VV_DELETED)
2615 		strlcat(buf, "|VV_DELETED", sizeof(buf));
2616 	if (vp->v_vflag & VV_MD)
2617 		strlcat(buf, "|VV_MD", sizeof(buf));
2618 	flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC |
2619 	    VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2620 	    VV_NOKNOTE | VV_DELETED | VV_MD);
2621 	if (flags != 0) {
2622 		snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2623 		strlcat(buf, buf2, sizeof(buf));
2624 	}
2625 	if (vp->v_iflag & VI_MOUNT)
2626 		strlcat(buf, "|VI_MOUNT", sizeof(buf));
2627 	if (vp->v_iflag & VI_AGE)
2628 		strlcat(buf, "|VI_AGE", sizeof(buf));
2629 	if (vp->v_iflag & VI_DOOMED)
2630 		strlcat(buf, "|VI_DOOMED", sizeof(buf));
2631 	if (vp->v_iflag & VI_FREE)
2632 		strlcat(buf, "|VI_FREE", sizeof(buf));
2633 	if (vp->v_iflag & VI_OBJDIRTY)
2634 		strlcat(buf, "|VI_OBJDIRTY", sizeof(buf));
2635 	if (vp->v_iflag & VI_DOINGINACT)
2636 		strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2637 	if (vp->v_iflag & VI_OWEINACT)
2638 		strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2639 	flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2640 	    VI_OBJDIRTY | VI_DOINGINACT | VI_OWEINACT);
2641 	if (flags != 0) {
2642 		snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2643 		strlcat(buf, buf2, sizeof(buf));
2644 	}
2645 	printf("    flags (%s)\n", buf + 1);
2646 	if (mtx_owned(VI_MTX(vp)))
2647 		printf(" VI_LOCKed");
2648 	if (vp->v_object != NULL)
2649 		printf("    v_object %p ref %d pages %d\n",
2650 		    vp->v_object, vp->v_object->ref_count,
2651 		    vp->v_object->resident_page_count);
2652 	printf("    ");
2653 	lockmgr_printinfo(vp->v_vnlock);
2654 	printf("\n");
2655 	if (vp->v_data != NULL)
2656 		VOP_PRINT(vp);
2657 }
2658 
2659 #ifdef DDB
2660 /*
2661  * List all of the locked vnodes in the system.
2662  * Called when debugging the kernel.
2663  */
2664 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2665 {
2666 	struct mount *mp, *nmp;
2667 	struct vnode *vp;
2668 
2669 	/*
2670 	 * Note: because this is DDB, we can't obey the locking semantics
2671 	 * for these structures, which means we could catch an inconsistent
2672 	 * state and dereference a nasty pointer.  Not much to be done
2673 	 * about that.
2674 	 */
2675 	db_printf("Locked vnodes\n");
2676 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2677 		nmp = TAILQ_NEXT(mp, mnt_list);
2678 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2679 			if (vp->v_type != VMARKER &&
2680 			    VOP_ISLOCKED(vp))
2681 				vprint("", vp);
2682 		}
2683 		nmp = TAILQ_NEXT(mp, mnt_list);
2684 	}
2685 }
2686 
2687 /*
2688  * Show details about the given vnode.
2689  */
2690 DB_SHOW_COMMAND(vnode, db_show_vnode)
2691 {
2692 	struct vnode *vp;
2693 
2694 	if (!have_addr)
2695 		return;
2696 	vp = (struct vnode *)addr;
2697 	vn_printf(vp, "vnode ");
2698 }
2699 
2700 /*
2701  * Show details about the given mount point.
2702  */
2703 DB_SHOW_COMMAND(mount, db_show_mount)
2704 {
2705 	struct mount *mp;
2706 	struct statfs *sp;
2707 	struct vnode *vp;
2708 	char buf[512];
2709 	u_int flags;
2710 
2711 	if (!have_addr) {
2712 		/* No address given, print short info about all mount points. */
2713 		TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2714 			db_printf("%p %s on %s (%s)\n", mp,
2715 			    mp->mnt_stat.f_mntfromname,
2716 			    mp->mnt_stat.f_mntonname,
2717 			    mp->mnt_stat.f_fstypename);
2718 			if (db_pager_quit)
2719 				break;
2720 		}
2721 		db_printf("\nMore info: show mount <addr>\n");
2722 		return;
2723 	}
2724 
2725 	mp = (struct mount *)addr;
2726 	db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
2727 	    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
2728 
2729 	buf[0] = '\0';
2730 	flags = mp->mnt_flag;
2731 #define	MNT_FLAG(flag)	do {						\
2732 	if (flags & (flag)) {						\
2733 		if (buf[0] != '\0')					\
2734 			strlcat(buf, ", ", sizeof(buf));		\
2735 		strlcat(buf, (#flag) + 4, sizeof(buf));			\
2736 		flags &= ~(flag);					\
2737 	}								\
2738 } while (0)
2739 	MNT_FLAG(MNT_RDONLY);
2740 	MNT_FLAG(MNT_SYNCHRONOUS);
2741 	MNT_FLAG(MNT_NOEXEC);
2742 	MNT_FLAG(MNT_NOSUID);
2743 	MNT_FLAG(MNT_UNION);
2744 	MNT_FLAG(MNT_ASYNC);
2745 	MNT_FLAG(MNT_SUIDDIR);
2746 	MNT_FLAG(MNT_SOFTDEP);
2747 	MNT_FLAG(MNT_NOSYMFOLLOW);
2748 	MNT_FLAG(MNT_GJOURNAL);
2749 	MNT_FLAG(MNT_MULTILABEL);
2750 	MNT_FLAG(MNT_ACLS);
2751 	MNT_FLAG(MNT_NOATIME);
2752 	MNT_FLAG(MNT_NOCLUSTERR);
2753 	MNT_FLAG(MNT_NOCLUSTERW);
2754 	MNT_FLAG(MNT_EXRDONLY);
2755 	MNT_FLAG(MNT_EXPORTED);
2756 	MNT_FLAG(MNT_DEFEXPORTED);
2757 	MNT_FLAG(MNT_EXPORTANON);
2758 	MNT_FLAG(MNT_EXKERB);
2759 	MNT_FLAG(MNT_EXPUBLIC);
2760 	MNT_FLAG(MNT_LOCAL);
2761 	MNT_FLAG(MNT_QUOTA);
2762 	MNT_FLAG(MNT_ROOTFS);
2763 	MNT_FLAG(MNT_USER);
2764 	MNT_FLAG(MNT_IGNORE);
2765 	MNT_FLAG(MNT_UPDATE);
2766 	MNT_FLAG(MNT_DELEXPORT);
2767 	MNT_FLAG(MNT_RELOAD);
2768 	MNT_FLAG(MNT_FORCE);
2769 	MNT_FLAG(MNT_SNAPSHOT);
2770 	MNT_FLAG(MNT_BYFSID);
2771 #undef MNT_FLAG
2772 	if (flags != 0) {
2773 		if (buf[0] != '\0')
2774 			strlcat(buf, ", ", sizeof(buf));
2775 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2776 		    "0x%08x", flags);
2777 	}
2778 	db_printf("    mnt_flag = %s\n", buf);
2779 
2780 	buf[0] = '\0';
2781 	flags = mp->mnt_kern_flag;
2782 #define	MNT_KERN_FLAG(flag)	do {					\
2783 	if (flags & (flag)) {						\
2784 		if (buf[0] != '\0')					\
2785 			strlcat(buf, ", ", sizeof(buf));		\
2786 		strlcat(buf, (#flag) + 5, sizeof(buf));			\
2787 		flags &= ~(flag);					\
2788 	}								\
2789 } while (0)
2790 	MNT_KERN_FLAG(MNTK_UNMOUNTF);
2791 	MNT_KERN_FLAG(MNTK_ASYNC);
2792 	MNT_KERN_FLAG(MNTK_SOFTDEP);
2793 	MNT_KERN_FLAG(MNTK_NOINSMNTQ);
2794 	MNT_KERN_FLAG(MNTK_UNMOUNT);
2795 	MNT_KERN_FLAG(MNTK_MWAIT);
2796 	MNT_KERN_FLAG(MNTK_SUSPEND);
2797 	MNT_KERN_FLAG(MNTK_SUSPEND2);
2798 	MNT_KERN_FLAG(MNTK_SUSPENDED);
2799 	MNT_KERN_FLAG(MNTK_MPSAFE);
2800 	MNT_KERN_FLAG(MNTK_NOKNOTE);
2801 	MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
2802 #undef MNT_KERN_FLAG
2803 	if (flags != 0) {
2804 		if (buf[0] != '\0')
2805 			strlcat(buf, ", ", sizeof(buf));
2806 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2807 		    "0x%08x", flags);
2808 	}
2809 	db_printf("    mnt_kern_flag = %s\n", buf);
2810 
2811 	sp = &mp->mnt_stat;
2812 	db_printf("    mnt_stat = { version=%u type=%u flags=0x%016jx "
2813 	    "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
2814 	    "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
2815 	    "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
2816 	    (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
2817 	    (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
2818 	    (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
2819 	    (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
2820 	    (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
2821 	    (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
2822 	    (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
2823 	    (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
2824 
2825 	db_printf("    mnt_cred = { uid=%u ruid=%u",
2826 	    (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
2827 	if (mp->mnt_cred->cr_prison != NULL)
2828 		db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
2829 	db_printf(" }\n");
2830 	db_printf("    mnt_ref = %d\n", mp->mnt_ref);
2831 	db_printf("    mnt_gen = %d\n", mp->mnt_gen);
2832 	db_printf("    mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
2833 	db_printf("    mnt_writeopcount = %d\n", mp->mnt_writeopcount);
2834 	db_printf("    mnt_noasync = %u\n", mp->mnt_noasync);
2835 	db_printf("    mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
2836 	db_printf("    mnt_iosize_max = %d\n", mp->mnt_iosize_max);
2837 	db_printf("    mnt_hashseed = %u\n", mp->mnt_hashseed);
2838 	db_printf("    mnt_markercnt = %d\n", mp->mnt_markercnt);
2839 	db_printf("    mnt_holdcnt = %d\n", mp->mnt_holdcnt);
2840 	db_printf("    mnt_holdcntwaiters = %d\n", mp->mnt_holdcntwaiters);
2841 	db_printf("    mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
2842 	db_printf("    mnt_secondary_accwrites = %d\n",
2843 	    mp->mnt_secondary_accwrites);
2844 	db_printf("    mnt_gjprovider = %s\n",
2845 	    mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
2846 	db_printf("\n");
2847 
2848 	TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2849 		if (vp->v_type != VMARKER) {
2850 			vn_printf(vp, "vnode ");
2851 			if (db_pager_quit)
2852 				break;
2853 		}
2854 	}
2855 }
2856 #endif	/* DDB */
2857 
2858 /*
2859  * Fill in a struct xvfsconf based on a struct vfsconf.
2860  */
2861 static void
2862 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2863 {
2864 
2865 	strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2866 	xvfsp->vfc_typenum = vfsp->vfc_typenum;
2867 	xvfsp->vfc_refcount = vfsp->vfc_refcount;
2868 	xvfsp->vfc_flags = vfsp->vfc_flags;
2869 	/*
2870 	 * These are unused in userland, we keep them
2871 	 * to not break binary compatibility.
2872 	 */
2873 	xvfsp->vfc_vfsops = NULL;
2874 	xvfsp->vfc_next = NULL;
2875 }
2876 
2877 /*
2878  * Top level filesystem related information gathering.
2879  */
2880 static int
2881 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2882 {
2883 	struct vfsconf *vfsp;
2884 	struct xvfsconf xvfsp;
2885 	int error;
2886 
2887 	error = 0;
2888 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2889 		bzero(&xvfsp, sizeof(xvfsp));
2890 		vfsconf2x(vfsp, &xvfsp);
2891 		error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
2892 		if (error)
2893 			break;
2894 	}
2895 	return (error);
2896 }
2897 
2898 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
2899     "S,xvfsconf", "List of all configured filesystems");
2900 
2901 #ifndef BURN_BRIDGES
2902 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2903 
2904 static int
2905 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2906 {
2907 	int *name = (int *)arg1 - 1;	/* XXX */
2908 	u_int namelen = arg2 + 1;	/* XXX */
2909 	struct vfsconf *vfsp;
2910 	struct xvfsconf xvfsp;
2911 
2912 	printf("WARNING: userland calling deprecated sysctl, "
2913 	    "please rebuild world\n");
2914 
2915 #if 1 || defined(COMPAT_PRELITE2)
2916 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2917 	if (namelen == 1)
2918 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2919 #endif
2920 
2921 	switch (name[1]) {
2922 	case VFS_MAXTYPENUM:
2923 		if (namelen != 2)
2924 			return (ENOTDIR);
2925 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2926 	case VFS_CONF:
2927 		if (namelen != 3)
2928 			return (ENOTDIR);	/* overloaded */
2929 		TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
2930 			if (vfsp->vfc_typenum == name[2])
2931 				break;
2932 		if (vfsp == NULL)
2933 			return (EOPNOTSUPP);
2934 		bzero(&xvfsp, sizeof(xvfsp));
2935 		vfsconf2x(vfsp, &xvfsp);
2936 		return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
2937 	}
2938 	return (EOPNOTSUPP);
2939 }
2940 
2941 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
2942 	vfs_sysctl, "Generic filesystem");
2943 
2944 #if 1 || defined(COMPAT_PRELITE2)
2945 
2946 static int
2947 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2948 {
2949 	int error;
2950 	struct vfsconf *vfsp;
2951 	struct ovfsconf ovfs;
2952 
2953 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2954 		bzero(&ovfs, sizeof(ovfs));
2955 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
2956 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
2957 		ovfs.vfc_index = vfsp->vfc_typenum;
2958 		ovfs.vfc_refcount = vfsp->vfc_refcount;
2959 		ovfs.vfc_flags = vfsp->vfc_flags;
2960 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2961 		if (error)
2962 			return error;
2963 	}
2964 	return 0;
2965 }
2966 
2967 #endif /* 1 || COMPAT_PRELITE2 */
2968 #endif /* !BURN_BRIDGES */
2969 
2970 #define KINFO_VNODESLOP		10
2971 #ifdef notyet
2972 /*
2973  * Dump vnode list (via sysctl).
2974  */
2975 /* ARGSUSED */
2976 static int
2977 sysctl_vnode(SYSCTL_HANDLER_ARGS)
2978 {
2979 	struct xvnode *xvn;
2980 	struct thread *td = req->td;
2981 	struct mount *mp;
2982 	struct vnode *vp;
2983 	int error, len, n;
2984 
2985 	/*
2986 	 * Stale numvnodes access is not fatal here.
2987 	 */
2988 	req->lock = 0;
2989 	len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
2990 	if (!req->oldptr)
2991 		/* Make an estimate */
2992 		return (SYSCTL_OUT(req, 0, len));
2993 
2994 	error = sysctl_wire_old_buffer(req, 0);
2995 	if (error != 0)
2996 		return (error);
2997 	xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
2998 	n = 0;
2999 	mtx_lock(&mountlist_mtx);
3000 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3001 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td))
3002 			continue;
3003 		MNT_ILOCK(mp);
3004 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3005 			if (n == len)
3006 				break;
3007 			vref(vp);
3008 			xvn[n].xv_size = sizeof *xvn;
3009 			xvn[n].xv_vnode = vp;
3010 			xvn[n].xv_id = 0;	/* XXX compat */
3011 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3012 			XV_COPY(usecount);
3013 			XV_COPY(writecount);
3014 			XV_COPY(holdcnt);
3015 			XV_COPY(mount);
3016 			XV_COPY(numoutput);
3017 			XV_COPY(type);
3018 #undef XV_COPY
3019 			xvn[n].xv_flag = vp->v_vflag;
3020 
3021 			switch (vp->v_type) {
3022 			case VREG:
3023 			case VDIR:
3024 			case VLNK:
3025 				break;
3026 			case VBLK:
3027 			case VCHR:
3028 				if (vp->v_rdev == NULL) {
3029 					vrele(vp);
3030 					continue;
3031 				}
3032 				xvn[n].xv_dev = dev2udev(vp->v_rdev);
3033 				break;
3034 			case VSOCK:
3035 				xvn[n].xv_socket = vp->v_socket;
3036 				break;
3037 			case VFIFO:
3038 				xvn[n].xv_fifo = vp->v_fifoinfo;
3039 				break;
3040 			case VNON:
3041 			case VBAD:
3042 			default:
3043 				/* shouldn't happen? */
3044 				vrele(vp);
3045 				continue;
3046 			}
3047 			vrele(vp);
3048 			++n;
3049 		}
3050 		MNT_IUNLOCK(mp);
3051 		mtx_lock(&mountlist_mtx);
3052 		vfs_unbusy(mp, td);
3053 		if (n == len)
3054 			break;
3055 	}
3056 	mtx_unlock(&mountlist_mtx);
3057 
3058 	error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3059 	free(xvn, M_TEMP);
3060 	return (error);
3061 }
3062 
3063 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3064 	0, 0, sysctl_vnode, "S,xvnode", "");
3065 #endif
3066 
3067 /*
3068  * Unmount all filesystems. The list is traversed in reverse order
3069  * of mounting to avoid dependencies.
3070  */
3071 void
3072 vfs_unmountall(void)
3073 {
3074 	struct mount *mp;
3075 	struct thread *td;
3076 	int error;
3077 
3078 	KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
3079 	td = curthread;
3080 	/*
3081 	 * Since this only runs when rebooting, it is not interlocked.
3082 	 */
3083 	while(!TAILQ_EMPTY(&mountlist)) {
3084 		mp = TAILQ_LAST(&mountlist, mntlist);
3085 		error = dounmount(mp, MNT_FORCE, td);
3086 		if (error) {
3087 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
3088 			/*
3089 			 * XXX: Due to the way in which we mount the root
3090 			 * file system off of devfs, devfs will generate a
3091 			 * "busy" warning when we try to unmount it before
3092 			 * the root.  Don't print a warning as a result in
3093 			 * order to avoid false positive errors that may
3094 			 * cause needless upset.
3095 			 */
3096 			if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3097 				printf("unmount of %s failed (",
3098 				    mp->mnt_stat.f_mntonname);
3099 				if (error == EBUSY)
3100 					printf("BUSY)\n");
3101 				else
3102 					printf("%d)\n", error);
3103 			}
3104 		} else {
3105 			/* The unmount has removed mp from the mountlist */
3106 		}
3107 	}
3108 }
3109 
3110 /*
3111  * perform msync on all vnodes under a mount point
3112  * the mount point must be locked.
3113  */
3114 void
3115 vfs_msync(struct mount *mp, int flags)
3116 {
3117 	struct vnode *vp, *mvp;
3118 	struct vm_object *obj;
3119 
3120 	MNT_ILOCK(mp);
3121 	MNT_VNODE_FOREACH(vp, mp, mvp) {
3122 		VI_LOCK(vp);
3123 		if ((vp->v_iflag & VI_OBJDIRTY) &&
3124 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3125 			MNT_IUNLOCK(mp);
3126 			if (!vget(vp,
3127 			    LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3128 			    curthread)) {
3129 				if (vp->v_vflag & VV_NOSYNC) {	/* unlinked */
3130 					vput(vp);
3131 					MNT_ILOCK(mp);
3132 					continue;
3133 				}
3134 
3135 				obj = vp->v_object;
3136 				if (obj != NULL) {
3137 					VM_OBJECT_LOCK(obj);
3138 					vm_object_page_clean(obj, 0, 0,
3139 					    flags == MNT_WAIT ?
3140 					    OBJPC_SYNC : OBJPC_NOSYNC);
3141 					VM_OBJECT_UNLOCK(obj);
3142 				}
3143 				vput(vp);
3144 			}
3145 			MNT_ILOCK(mp);
3146 		} else
3147 			VI_UNLOCK(vp);
3148 	}
3149 	MNT_IUNLOCK(mp);
3150 }
3151 
3152 /*
3153  * Mark a vnode as free, putting it up for recycling.
3154  */
3155 static void
3156 vfree(struct vnode *vp)
3157 {
3158 
3159 	CTR1(KTR_VFS, "vfree vp %p", vp);
3160 	ASSERT_VI_LOCKED(vp, "vfree");
3161 	mtx_lock(&vnode_free_list_mtx);
3162 	VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed."));
3163 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free"));
3164 	VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't"));
3165 	VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp,
3166 	    ("vfree: Freeing doomed vnode"));
3167 	if (vp->v_iflag & VI_AGE) {
3168 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3169 	} else {
3170 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3171 	}
3172 	freevnodes++;
3173 	vp->v_iflag &= ~VI_AGE;
3174 	vp->v_iflag |= VI_FREE;
3175 	mtx_unlock(&vnode_free_list_mtx);
3176 }
3177 
3178 /*
3179  * Opposite of vfree() - mark a vnode as in use.
3180  */
3181 static void
3182 vbusy(struct vnode *vp)
3183 {
3184 	CTR1(KTR_VFS, "vbusy vp %p", vp);
3185 	ASSERT_VI_LOCKED(vp, "vbusy");
3186 	VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
3187 	VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed."));
3188 
3189 	mtx_lock(&vnode_free_list_mtx);
3190 	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3191 	freevnodes--;
3192 	vp->v_iflag &= ~(VI_FREE|VI_AGE);
3193 	mtx_unlock(&vnode_free_list_mtx);
3194 }
3195 
3196 /*
3197  * Initalize per-vnode helper structure to hold poll-related state.
3198  */
3199 void
3200 v_addpollinfo(struct vnode *vp)
3201 {
3202 	struct vpollinfo *vi;
3203 
3204 	vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3205 	if (vp->v_pollinfo != NULL) {
3206 		uma_zfree(vnodepoll_zone, vi);
3207 		return;
3208 	}
3209 	vp->v_pollinfo = vi;
3210 	mtx_init(&vp->v_pollinfo->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3211 	knlist_init(&vp->v_pollinfo->vpi_selinfo.si_note, vp, vfs_knllock,
3212 	    vfs_knlunlock, vfs_knllocked);
3213 }
3214 
3215 /*
3216  * Record a process's interest in events which might happen to
3217  * a vnode.  Because poll uses the historic select-style interface
3218  * internally, this routine serves as both the ``check for any
3219  * pending events'' and the ``record my interest in future events''
3220  * functions.  (These are done together, while the lock is held,
3221  * to avoid race conditions.)
3222  */
3223 int
3224 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3225 {
3226 
3227 	if (vp->v_pollinfo == NULL)
3228 		v_addpollinfo(vp);
3229 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3230 	if (vp->v_pollinfo->vpi_revents & events) {
3231 		/*
3232 		 * This leaves events we are not interested
3233 		 * in available for the other process which
3234 		 * which presumably had requested them
3235 		 * (otherwise they would never have been
3236 		 * recorded).
3237 		 */
3238 		events &= vp->v_pollinfo->vpi_revents;
3239 		vp->v_pollinfo->vpi_revents &= ~events;
3240 
3241 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
3242 		return events;
3243 	}
3244 	vp->v_pollinfo->vpi_events |= events;
3245 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3246 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3247 	return 0;
3248 }
3249 
3250 /*
3251  * Routine to create and manage a filesystem syncer vnode.
3252  */
3253 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
3254 static int	sync_fsync(struct  vop_fsync_args *);
3255 static int	sync_inactive(struct  vop_inactive_args *);
3256 static int	sync_reclaim(struct  vop_reclaim_args *);
3257 
3258 static struct vop_vector sync_vnodeops = {
3259 	.vop_bypass =	VOP_EOPNOTSUPP,
3260 	.vop_close =	sync_close,		/* close */
3261 	.vop_fsync =	sync_fsync,		/* fsync */
3262 	.vop_inactive =	sync_inactive,	/* inactive */
3263 	.vop_reclaim =	sync_reclaim,	/* reclaim */
3264 	.vop_lock1 =	vop_stdlock,	/* lock */
3265 	.vop_unlock =	vop_stdunlock,	/* unlock */
3266 	.vop_islocked =	vop_stdislocked,	/* islocked */
3267 };
3268 
3269 /*
3270  * Create a new filesystem syncer vnode for the specified mount point.
3271  */
3272 int
3273 vfs_allocate_syncvnode(struct mount *mp)
3274 {
3275 	struct vnode *vp;
3276 	struct bufobj *bo;
3277 	static long start, incr, next;
3278 	int error;
3279 
3280 	/* Allocate a new vnode */
3281 	if ((error = getnewvnode("syncer", mp, &sync_vnodeops, &vp)) != 0) {
3282 		mp->mnt_syncer = NULL;
3283 		return (error);
3284 	}
3285 	vp->v_type = VNON;
3286 	error = insmntque(vp, mp);
3287 	if (error != 0)
3288 		panic("vfs_allocate_syncvnode: insmntque failed");
3289 	/*
3290 	 * Place the vnode onto the syncer worklist. We attempt to
3291 	 * scatter them about on the list so that they will go off
3292 	 * at evenly distributed times even if all the filesystems
3293 	 * are mounted at once.
3294 	 */
3295 	next += incr;
3296 	if (next == 0 || next > syncer_maxdelay) {
3297 		start /= 2;
3298 		incr /= 2;
3299 		if (start == 0) {
3300 			start = syncer_maxdelay / 2;
3301 			incr = syncer_maxdelay;
3302 		}
3303 		next = start;
3304 	}
3305 	bo = &vp->v_bufobj;
3306 	BO_LOCK(bo);
3307 	vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3308 	/* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3309 	mtx_lock(&sync_mtx);
3310 	sync_vnode_count++;
3311 	mtx_unlock(&sync_mtx);
3312 	BO_UNLOCK(bo);
3313 	mp->mnt_syncer = vp;
3314 	return (0);
3315 }
3316 
3317 /*
3318  * Do a lazy sync of the filesystem.
3319  */
3320 static int
3321 sync_fsync(struct vop_fsync_args *ap)
3322 {
3323 	struct vnode *syncvp = ap->a_vp;
3324 	struct mount *mp = syncvp->v_mount;
3325 	struct thread *td = ap->a_td;
3326 	int error;
3327 	struct bufobj *bo;
3328 
3329 	/*
3330 	 * We only need to do something if this is a lazy evaluation.
3331 	 */
3332 	if (ap->a_waitfor != MNT_LAZY)
3333 		return (0);
3334 
3335 	/*
3336 	 * Move ourselves to the back of the sync list.
3337 	 */
3338 	bo = &syncvp->v_bufobj;
3339 	BO_LOCK(bo);
3340 	vn_syncer_add_to_worklist(bo, syncdelay);
3341 	BO_UNLOCK(bo);
3342 
3343 	/*
3344 	 * Walk the list of vnodes pushing all that are dirty and
3345 	 * not already on the sync list.
3346 	 */
3347 	mtx_lock(&mountlist_mtx);
3348 	if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) {
3349 		mtx_unlock(&mountlist_mtx);
3350 		return (0);
3351 	}
3352 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3353 		vfs_unbusy(mp, td);
3354 		return (0);
3355 	}
3356 	MNT_ILOCK(mp);
3357 	mp->mnt_noasync++;
3358 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
3359 	MNT_IUNLOCK(mp);
3360 	vfs_msync(mp, MNT_NOWAIT);
3361 	error = VFS_SYNC(mp, MNT_LAZY, td);
3362 	MNT_ILOCK(mp);
3363 	mp->mnt_noasync--;
3364 	if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
3365 		mp->mnt_kern_flag |= MNTK_ASYNC;
3366 	MNT_IUNLOCK(mp);
3367 	vn_finished_write(mp);
3368 	vfs_unbusy(mp, td);
3369 	return (error);
3370 }
3371 
3372 /*
3373  * The syncer vnode is no referenced.
3374  */
3375 static int
3376 sync_inactive(struct vop_inactive_args *ap)
3377 {
3378 
3379 	vgone(ap->a_vp);
3380 	return (0);
3381 }
3382 
3383 /*
3384  * The syncer vnode is no longer needed and is being decommissioned.
3385  *
3386  * Modifications to the worklist must be protected by sync_mtx.
3387  */
3388 static int
3389 sync_reclaim(struct vop_reclaim_args *ap)
3390 {
3391 	struct vnode *vp = ap->a_vp;
3392 	struct bufobj *bo;
3393 
3394 	bo = &vp->v_bufobj;
3395 	BO_LOCK(bo);
3396 	vp->v_mount->mnt_syncer = NULL;
3397 	if (bo->bo_flag & BO_ONWORKLST) {
3398 		mtx_lock(&sync_mtx);
3399 		LIST_REMOVE(bo, bo_synclist);
3400 		syncer_worklist_len--;
3401 		sync_vnode_count--;
3402 		mtx_unlock(&sync_mtx);
3403 		bo->bo_flag &= ~BO_ONWORKLST;
3404 	}
3405 	BO_UNLOCK(bo);
3406 
3407 	return (0);
3408 }
3409 
3410 /*
3411  * Check if vnode represents a disk device
3412  */
3413 int
3414 vn_isdisk(struct vnode *vp, int *errp)
3415 {
3416 	int error;
3417 
3418 	error = 0;
3419 	dev_lock();
3420 	if (vp->v_type != VCHR)
3421 		error = ENOTBLK;
3422 	else if (vp->v_rdev == NULL)
3423 		error = ENXIO;
3424 	else if (vp->v_rdev->si_devsw == NULL)
3425 		error = ENXIO;
3426 	else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3427 		error = ENOTBLK;
3428 	dev_unlock();
3429 	if (errp != NULL)
3430 		*errp = error;
3431 	return (error == 0);
3432 }
3433 
3434 /*
3435  * Common filesystem object access control check routine.  Accepts a
3436  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3437  * and optional call-by-reference privused argument allowing vaccess()
3438  * to indicate to the caller whether privilege was used to satisfy the
3439  * request (obsoleted).  Returns 0 on success, or an errno on failure.
3440  *
3441  * The ifdef'd CAPABILITIES version is here for reference, but is not
3442  * actually used.
3443  */
3444 int
3445 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3446     mode_t acc_mode, struct ucred *cred, int *privused)
3447 {
3448 	mode_t dac_granted;
3449 	mode_t priv_granted;
3450 
3451 	/*
3452 	 * Look for a normal, non-privileged way to access the file/directory
3453 	 * as requested.  If it exists, go with that.
3454 	 */
3455 
3456 	if (privused != NULL)
3457 		*privused = 0;
3458 
3459 	dac_granted = 0;
3460 
3461 	/* Check the owner. */
3462 	if (cred->cr_uid == file_uid) {
3463 		dac_granted |= VADMIN;
3464 		if (file_mode & S_IXUSR)
3465 			dac_granted |= VEXEC;
3466 		if (file_mode & S_IRUSR)
3467 			dac_granted |= VREAD;
3468 		if (file_mode & S_IWUSR)
3469 			dac_granted |= (VWRITE | VAPPEND);
3470 
3471 		if ((acc_mode & dac_granted) == acc_mode)
3472 			return (0);
3473 
3474 		goto privcheck;
3475 	}
3476 
3477 	/* Otherwise, check the groups (first match) */
3478 	if (groupmember(file_gid, cred)) {
3479 		if (file_mode & S_IXGRP)
3480 			dac_granted |= VEXEC;
3481 		if (file_mode & S_IRGRP)
3482 			dac_granted |= VREAD;
3483 		if (file_mode & S_IWGRP)
3484 			dac_granted |= (VWRITE | VAPPEND);
3485 
3486 		if ((acc_mode & dac_granted) == acc_mode)
3487 			return (0);
3488 
3489 		goto privcheck;
3490 	}
3491 
3492 	/* Otherwise, check everyone else. */
3493 	if (file_mode & S_IXOTH)
3494 		dac_granted |= VEXEC;
3495 	if (file_mode & S_IROTH)
3496 		dac_granted |= VREAD;
3497 	if (file_mode & S_IWOTH)
3498 		dac_granted |= (VWRITE | VAPPEND);
3499 	if ((acc_mode & dac_granted) == acc_mode)
3500 		return (0);
3501 
3502 privcheck:
3503 	/*
3504 	 * Build a privilege mask to determine if the set of privileges
3505 	 * satisfies the requirements when combined with the granted mask
3506 	 * from above.  For each privilege, if the privilege is required,
3507 	 * bitwise or the request type onto the priv_granted mask.
3508 	 */
3509 	priv_granted = 0;
3510 
3511 	if (type == VDIR) {
3512 		/*
3513 		 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3514 		 * requests, instead of PRIV_VFS_EXEC.
3515 		 */
3516 		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3517 		    !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3518 			priv_granted |= VEXEC;
3519 	} else {
3520 		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3521 		    !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3522 			priv_granted |= VEXEC;
3523 	}
3524 
3525 	if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
3526 	    !priv_check_cred(cred, PRIV_VFS_READ, 0))
3527 		priv_granted |= VREAD;
3528 
3529 	if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3530 	    !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3531 		priv_granted |= (VWRITE | VAPPEND);
3532 
3533 	if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3534 	    !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3535 		priv_granted |= VADMIN;
3536 
3537 	if ((acc_mode & (priv_granted | dac_granted)) == acc_mode) {
3538 		/* XXX audit: privilege used */
3539 		if (privused != NULL)
3540 			*privused = 1;
3541 		return (0);
3542 	}
3543 
3544 	return ((acc_mode & VADMIN) ? EPERM : EACCES);
3545 }
3546 
3547 /*
3548  * Credential check based on process requesting service, and per-attribute
3549  * permissions.
3550  */
3551 int
3552 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3553     struct thread *td, int access)
3554 {
3555 
3556 	/*
3557 	 * Kernel-invoked always succeeds.
3558 	 */
3559 	if (cred == NOCRED)
3560 		return (0);
3561 
3562 	/*
3563 	 * Do not allow privileged processes in jail to directly manipulate
3564 	 * system attributes.
3565 	 */
3566 	switch (attrnamespace) {
3567 	case EXTATTR_NAMESPACE_SYSTEM:
3568 		/* Potentially should be: return (EPERM); */
3569 		return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3570 	case EXTATTR_NAMESPACE_USER:
3571 		return (VOP_ACCESS(vp, access, cred, td));
3572 	default:
3573 		return (EPERM);
3574 	}
3575 }
3576 
3577 #ifdef DEBUG_VFS_LOCKS
3578 /*
3579  * This only exists to supress warnings from unlocked specfs accesses.  It is
3580  * no longer ok to have an unlocked VFS.
3581  */
3582 #define	IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL ||		\
3583 	(vp)->v_type == VCHR ||	(vp)->v_type == VBAD)
3584 
3585 int vfs_badlock_ddb = 1;	/* Drop into debugger on violation. */
3586 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, "");
3587 
3588 int vfs_badlock_mutex = 1;	/* Check for interlock across VOPs. */
3589 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, "");
3590 
3591 int vfs_badlock_print = 1;	/* Print lock violations. */
3592 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, "");
3593 
3594 #ifdef KDB
3595 int vfs_badlock_backtrace = 1;	/* Print backtrace at lock violations. */
3596 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, "");
3597 #endif
3598 
3599 static void
3600 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3601 {
3602 
3603 #ifdef KDB
3604 	if (vfs_badlock_backtrace)
3605 		kdb_backtrace();
3606 #endif
3607 	if (vfs_badlock_print)
3608 		printf("%s: %p %s\n", str, (void *)vp, msg);
3609 	if (vfs_badlock_ddb)
3610 		kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3611 }
3612 
3613 void
3614 assert_vi_locked(struct vnode *vp, const char *str)
3615 {
3616 
3617 	if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3618 		vfs_badlock("interlock is not locked but should be", str, vp);
3619 }
3620 
3621 void
3622 assert_vi_unlocked(struct vnode *vp, const char *str)
3623 {
3624 
3625 	if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3626 		vfs_badlock("interlock is locked but should not be", str, vp);
3627 }
3628 
3629 void
3630 assert_vop_locked(struct vnode *vp, const char *str)
3631 {
3632 
3633 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0)
3634 		vfs_badlock("is not locked but should be", str, vp);
3635 }
3636 
3637 void
3638 assert_vop_unlocked(struct vnode *vp, const char *str)
3639 {
3640 
3641 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
3642 		vfs_badlock("is locked but should not be", str, vp);
3643 }
3644 
3645 void
3646 assert_vop_elocked(struct vnode *vp, const char *str)
3647 {
3648 
3649 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
3650 		vfs_badlock("is not exclusive locked but should be", str, vp);
3651 }
3652 
3653 #if 0
3654 void
3655 assert_vop_elocked_other(struct vnode *vp, const char *str)
3656 {
3657 
3658 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
3659 		vfs_badlock("is not exclusive locked by another thread",
3660 		    str, vp);
3661 }
3662 
3663 void
3664 assert_vop_slocked(struct vnode *vp, const char *str)
3665 {
3666 
3667 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
3668 		vfs_badlock("is not locked shared but should be", str, vp);
3669 }
3670 #endif /* 0 */
3671 #endif /* DEBUG_VFS_LOCKS */
3672 
3673 void
3674 vop_rename_pre(void *ap)
3675 {
3676 	struct vop_rename_args *a = ap;
3677 
3678 #ifdef DEBUG_VFS_LOCKS
3679 	if (a->a_tvp)
3680 		ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
3681 	ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
3682 	ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
3683 	ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
3684 
3685 	/* Check the source (from). */
3686 	if (a->a_tdvp != a->a_fdvp && a->a_tvp != a->a_fdvp)
3687 		ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3688 	if (a->a_tvp != a->a_fvp)
3689 		ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
3690 
3691 	/* Check the target. */
3692 	if (a->a_tvp)
3693 		ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3694 	ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3695 #endif
3696 	if (a->a_tdvp != a->a_fdvp)
3697 		vhold(a->a_fdvp);
3698 	if (a->a_tvp != a->a_fvp)
3699 		vhold(a->a_fvp);
3700 	vhold(a->a_tdvp);
3701 	if (a->a_tvp)
3702 		vhold(a->a_tvp);
3703 }
3704 
3705 void
3706 vop_strategy_pre(void *ap)
3707 {
3708 #ifdef DEBUG_VFS_LOCKS
3709 	struct vop_strategy_args *a;
3710 	struct buf *bp;
3711 
3712 	a = ap;
3713 	bp = a->a_bp;
3714 
3715 	/*
3716 	 * Cluster ops lock their component buffers but not the IO container.
3717 	 */
3718 	if ((bp->b_flags & B_CLUSTER) != 0)
3719 		return;
3720 
3721 	if (!BUF_ISLOCKED(bp)) {
3722 		if (vfs_badlock_print)
3723 			printf(
3724 			    "VOP_STRATEGY: bp is not locked but should be\n");
3725 		if (vfs_badlock_ddb)
3726 			kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3727 	}
3728 #endif
3729 }
3730 
3731 void
3732 vop_lookup_pre(void *ap)
3733 {
3734 #ifdef DEBUG_VFS_LOCKS
3735 	struct vop_lookup_args *a;
3736 	struct vnode *dvp;
3737 
3738 	a = ap;
3739 	dvp = a->a_dvp;
3740 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3741 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3742 #endif
3743 }
3744 
3745 void
3746 vop_lookup_post(void *ap, int rc)
3747 {
3748 #ifdef DEBUG_VFS_LOCKS
3749 	struct vop_lookup_args *a;
3750 	struct vnode *dvp;
3751 	struct vnode *vp;
3752 
3753 	a = ap;
3754 	dvp = a->a_dvp;
3755 	vp = *(a->a_vpp);
3756 
3757 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3758 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3759 
3760 	if (!rc)
3761 		ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
3762 #endif
3763 }
3764 
3765 void
3766 vop_lock_pre(void *ap)
3767 {
3768 #ifdef DEBUG_VFS_LOCKS
3769 	struct vop_lock1_args *a = ap;
3770 
3771 	if ((a->a_flags & LK_INTERLOCK) == 0)
3772 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3773 	else
3774 		ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
3775 #endif
3776 }
3777 
3778 void
3779 vop_lock_post(void *ap, int rc)
3780 {
3781 #ifdef DEBUG_VFS_LOCKS
3782 	struct vop_lock1_args *a = ap;
3783 
3784 	ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3785 	if (rc == 0)
3786 		ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
3787 #endif
3788 }
3789 
3790 void
3791 vop_unlock_pre(void *ap)
3792 {
3793 #ifdef DEBUG_VFS_LOCKS
3794 	struct vop_unlock_args *a = ap;
3795 
3796 	if (a->a_flags & LK_INTERLOCK)
3797 		ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
3798 	ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
3799 #endif
3800 }
3801 
3802 void
3803 vop_unlock_post(void *ap, int rc)
3804 {
3805 #ifdef DEBUG_VFS_LOCKS
3806 	struct vop_unlock_args *a = ap;
3807 
3808 	if (a->a_flags & LK_INTERLOCK)
3809 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
3810 #endif
3811 }
3812 
3813 void
3814 vop_create_post(void *ap, int rc)
3815 {
3816 	struct vop_create_args *a = ap;
3817 
3818 	if (!rc)
3819 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3820 }
3821 
3822 void
3823 vop_link_post(void *ap, int rc)
3824 {
3825 	struct vop_link_args *a = ap;
3826 
3827 	if (!rc) {
3828 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
3829 		VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
3830 	}
3831 }
3832 
3833 void
3834 vop_mkdir_post(void *ap, int rc)
3835 {
3836 	struct vop_mkdir_args *a = ap;
3837 
3838 	if (!rc)
3839 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3840 }
3841 
3842 void
3843 vop_mknod_post(void *ap, int rc)
3844 {
3845 	struct vop_mknod_args *a = ap;
3846 
3847 	if (!rc)
3848 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3849 }
3850 
3851 void
3852 vop_remove_post(void *ap, int rc)
3853 {
3854 	struct vop_remove_args *a = ap;
3855 
3856 	if (!rc) {
3857 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3858 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3859 	}
3860 }
3861 
3862 void
3863 vop_rename_post(void *ap, int rc)
3864 {
3865 	struct vop_rename_args *a = ap;
3866 
3867 	if (!rc) {
3868 		VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
3869 		VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
3870 		VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
3871 		if (a->a_tvp)
3872 			VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
3873 	}
3874 	if (a->a_tdvp != a->a_fdvp)
3875 		vdrop(a->a_fdvp);
3876 	if (a->a_tvp != a->a_fvp)
3877 		vdrop(a->a_fvp);
3878 	vdrop(a->a_tdvp);
3879 	if (a->a_tvp)
3880 		vdrop(a->a_tvp);
3881 }
3882 
3883 void
3884 vop_rmdir_post(void *ap, int rc)
3885 {
3886 	struct vop_rmdir_args *a = ap;
3887 
3888 	if (!rc) {
3889 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3890 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3891 	}
3892 }
3893 
3894 void
3895 vop_setattr_post(void *ap, int rc)
3896 {
3897 	struct vop_setattr_args *a = ap;
3898 
3899 	if (!rc)
3900 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
3901 }
3902 
3903 void
3904 vop_symlink_post(void *ap, int rc)
3905 {
3906 	struct vop_symlink_args *a = ap;
3907 
3908 	if (!rc)
3909 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3910 }
3911 
3912 static struct knlist fs_knlist;
3913 
3914 static void
3915 vfs_event_init(void *arg)
3916 {
3917 	knlist_init(&fs_knlist, NULL, NULL, NULL, NULL);
3918 }
3919 /* XXX - correct order? */
3920 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
3921 
3922 void
3923 vfs_event_signal(fsid_t *fsid, u_int32_t event, intptr_t data __unused)
3924 {
3925 
3926 	KNOTE_UNLOCKED(&fs_knlist, event);
3927 }
3928 
3929 static int	filt_fsattach(struct knote *kn);
3930 static void	filt_fsdetach(struct knote *kn);
3931 static int	filt_fsevent(struct knote *kn, long hint);
3932 
3933 struct filterops fs_filtops =
3934 	{ 0, filt_fsattach, filt_fsdetach, filt_fsevent };
3935 
3936 static int
3937 filt_fsattach(struct knote *kn)
3938 {
3939 
3940 	kn->kn_flags |= EV_CLEAR;
3941 	knlist_add(&fs_knlist, kn, 0);
3942 	return (0);
3943 }
3944 
3945 static void
3946 filt_fsdetach(struct knote *kn)
3947 {
3948 
3949 	knlist_remove(&fs_knlist, kn, 0);
3950 }
3951 
3952 static int
3953 filt_fsevent(struct knote *kn, long hint)
3954 {
3955 
3956 	kn->kn_fflags |= hint;
3957 	return (kn->kn_fflags != 0);
3958 }
3959 
3960 static int
3961 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
3962 {
3963 	struct vfsidctl vc;
3964 	int error;
3965 	struct mount *mp;
3966 
3967 	error = SYSCTL_IN(req, &vc, sizeof(vc));
3968 	if (error)
3969 		return (error);
3970 	if (vc.vc_vers != VFS_CTL_VERS1)
3971 		return (EINVAL);
3972 	mp = vfs_getvfs(&vc.vc_fsid);
3973 	if (mp == NULL)
3974 		return (ENOENT);
3975 	/* ensure that a specific sysctl goes to the right filesystem. */
3976 	if (strcmp(vc.vc_fstypename, "*") != 0 &&
3977 	    strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
3978 		vfs_rel(mp);
3979 		return (EINVAL);
3980 	}
3981 	VCTLTOREQ(&vc, req);
3982 	error = VFS_SYSCTL(mp, vc.vc_op, req);
3983 	vfs_rel(mp);
3984 	return (error);
3985 }
3986 
3987 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "",
3988     "Sysctl by fsid");
3989 
3990 /*
3991  * Function to initialize a va_filerev field sensibly.
3992  * XXX: Wouldn't a random number make a lot more sense ??
3993  */
3994 u_quad_t
3995 init_va_filerev(void)
3996 {
3997 	struct bintime bt;
3998 
3999 	getbinuptime(&bt);
4000 	return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4001 }
4002 
4003 static int	filt_vfsread(struct knote *kn, long hint);
4004 static int	filt_vfswrite(struct knote *kn, long hint);
4005 static int	filt_vfsvnode(struct knote *kn, long hint);
4006 static void	filt_vfsdetach(struct knote *kn);
4007 static struct filterops vfsread_filtops =
4008 	{ 1, NULL, filt_vfsdetach, filt_vfsread };
4009 static struct filterops vfswrite_filtops =
4010 	{ 1, NULL, filt_vfsdetach, filt_vfswrite };
4011 static struct filterops vfsvnode_filtops =
4012 	{ 1, NULL, filt_vfsdetach, filt_vfsvnode };
4013 
4014 static void
4015 vfs_knllock(void *arg)
4016 {
4017 	struct vnode *vp = arg;
4018 
4019 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4020 }
4021 
4022 static void
4023 vfs_knlunlock(void *arg)
4024 {
4025 	struct vnode *vp = arg;
4026 
4027 	VOP_UNLOCK(vp, 0);
4028 }
4029 
4030 static int
4031 vfs_knllocked(void *arg)
4032 {
4033 	struct vnode *vp = arg;
4034 
4035 	return (VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
4036 }
4037 
4038 int
4039 vfs_kqfilter(struct vop_kqfilter_args *ap)
4040 {
4041 	struct vnode *vp = ap->a_vp;
4042 	struct knote *kn = ap->a_kn;
4043 	struct knlist *knl;
4044 
4045 	switch (kn->kn_filter) {
4046 	case EVFILT_READ:
4047 		kn->kn_fop = &vfsread_filtops;
4048 		break;
4049 	case EVFILT_WRITE:
4050 		kn->kn_fop = &vfswrite_filtops;
4051 		break;
4052 	case EVFILT_VNODE:
4053 		kn->kn_fop = &vfsvnode_filtops;
4054 		break;
4055 	default:
4056 		return (EINVAL);
4057 	}
4058 
4059 	kn->kn_hook = (caddr_t)vp;
4060 
4061 	if (vp->v_pollinfo == NULL)
4062 		v_addpollinfo(vp);
4063 	if (vp->v_pollinfo == NULL)
4064 		return (ENOMEM);
4065 	knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4066 	knlist_add(knl, kn, 0);
4067 
4068 	return (0);
4069 }
4070 
4071 /*
4072  * Detach knote from vnode
4073  */
4074 static void
4075 filt_vfsdetach(struct knote *kn)
4076 {
4077 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4078 
4079 	KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4080 	knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4081 }
4082 
4083 /*ARGSUSED*/
4084 static int
4085 filt_vfsread(struct knote *kn, long hint)
4086 {
4087 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4088 	struct vattr va;
4089 
4090 	/*
4091 	 * filesystem is gone, so set the EOF flag and schedule
4092 	 * the knote for deletion.
4093 	 */
4094 	if (hint == NOTE_REVOKE) {
4095 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4096 		return (1);
4097 	}
4098 
4099 	if (VOP_GETATTR(vp, &va, curthread->td_ucred, curthread))
4100 		return (0);
4101 
4102 	kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4103 	return (kn->kn_data != 0);
4104 }
4105 
4106 /*ARGSUSED*/
4107 static int
4108 filt_vfswrite(struct knote *kn, long hint)
4109 {
4110 	/*
4111 	 * filesystem is gone, so set the EOF flag and schedule
4112 	 * the knote for deletion.
4113 	 */
4114 	if (hint == NOTE_REVOKE)
4115 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4116 
4117 	kn->kn_data = 0;
4118 	return (1);
4119 }
4120 
4121 static int
4122 filt_vfsvnode(struct knote *kn, long hint)
4123 {
4124 	if (kn->kn_sfflags & hint)
4125 		kn->kn_fflags |= hint;
4126 	if (hint == NOTE_REVOKE) {
4127 		kn->kn_flags |= EV_EOF;
4128 		return (1);
4129 	}
4130 	return (kn->kn_fflags != 0);
4131 }
4132 
4133 int
4134 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4135 {
4136 	int error;
4137 
4138 	if (dp->d_reclen > ap->a_uio->uio_resid)
4139 		return (ENAMETOOLONG);
4140 	error = uiomove(dp, dp->d_reclen, ap->a_uio);
4141 	if (error) {
4142 		if (ap->a_ncookies != NULL) {
4143 			if (ap->a_cookies != NULL)
4144 				free(ap->a_cookies, M_TEMP);
4145 			ap->a_cookies = NULL;
4146 			*ap->a_ncookies = 0;
4147 		}
4148 		return (error);
4149 	}
4150 	if (ap->a_ncookies == NULL)
4151 		return (0);
4152 
4153 	KASSERT(ap->a_cookies,
4154 	    ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4155 
4156 	*ap->a_cookies = realloc(*ap->a_cookies,
4157 	    (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4158 	(*ap->a_cookies)[*ap->a_ncookies] = off;
4159 	return (0);
4160 }
4161 
4162 /*
4163  * Mark for update the access time of the file if the filesystem
4164  * supports VA_MARK_ATIME.  This functionality is used by execve
4165  * and mmap, so we want to avoid the synchronous I/O implied by
4166  * directly setting va_atime for the sake of efficiency.
4167  */
4168 void
4169 vfs_mark_atime(struct vnode *vp, struct thread *td)
4170 {
4171 	struct vattr atimeattr;
4172 
4173 	if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) {
4174 		VATTR_NULL(&atimeattr);
4175 		atimeattr.va_vaflags |= VA_MARK_ATIME;
4176 		(void)VOP_SETATTR(vp, &atimeattr, td->td_ucred, td);
4177 	}
4178 }
4179