xref: /freebsd/sys/kern/vfs_subr.c (revision 409a390c3341fb4f162cd7de1fd595a323ebbfd8)
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 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/bio.h>
49 #include <sys/buf.h>
50 #include <sys/condvar.h>
51 #include <sys/conf.h>
52 #include <sys/dirent.h>
53 #include <sys/event.h>
54 #include <sys/eventhandler.h>
55 #include <sys/extattr.h>
56 #include <sys/file.h>
57 #include <sys/fcntl.h>
58 #include <sys/jail.h>
59 #include <sys/kdb.h>
60 #include <sys/kernel.h>
61 #include <sys/kthread.h>
62 #include <sys/lockf.h>
63 #include <sys/malloc.h>
64 #include <sys/mount.h>
65 #include <sys/namei.h>
66 #include <sys/priv.h>
67 #include <sys/reboot.h>
68 #include <sys/sleepqueue.h>
69 #include <sys/stat.h>
70 #include <sys/sysctl.h>
71 #include <sys/syslog.h>
72 #include <sys/vmmeter.h>
73 #include <sys/vnode.h>
74 
75 #include <machine/stdarg.h>
76 
77 #include <security/mac/mac_framework.h>
78 
79 #include <vm/vm.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_extern.h>
82 #include <vm/pmap.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_page.h>
85 #include <vm/vm_kern.h>
86 #include <vm/uma.h>
87 
88 #ifdef DDB
89 #include <ddb/ddb.h>
90 #endif
91 
92 #define	WI_MPSAFEQ	0
93 #define	WI_GIANTQ	1
94 
95 static MALLOC_DEFINE(M_NETADDR, "subr_export_host", "Export host address structure");
96 
97 static void	delmntque(struct vnode *vp);
98 static int	flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
99 		    int slpflag, int slptimeo);
100 static void	syncer_shutdown(void *arg, int howto);
101 static int	vtryrecycle(struct vnode *vp);
102 static void	vbusy(struct vnode *vp);
103 static void	vinactive(struct vnode *, struct thread *);
104 static void	v_incr_usecount(struct vnode *);
105 static void	v_decr_usecount(struct vnode *);
106 static void	v_decr_useonly(struct vnode *);
107 static void	v_upgrade_usecount(struct vnode *);
108 static void	vfree(struct vnode *);
109 static void	vnlru_free(int);
110 static void	vgonel(struct vnode *);
111 static void	vfs_knllock(void *arg);
112 static void	vfs_knlunlock(void *arg);
113 static void	vfs_knl_assert_locked(void *arg);
114 static void	vfs_knl_assert_unlocked(void *arg);
115 static void	destroy_vpollinfo(struct vpollinfo *vi);
116 
117 /*
118  * Number of vnodes in existence.  Increased whenever getnewvnode()
119  * allocates a new vnode, decreased on vdestroy() called on VI_DOOMed
120  * vnode.
121  */
122 static unsigned long	numvnodes;
123 
124 SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
125 
126 /*
127  * Conversion tables for conversion from vnode types to inode formats
128  * and back.
129  */
130 enum vtype iftovt_tab[16] = {
131 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
132 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
133 };
134 int vttoif_tab[10] = {
135 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
136 	S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
137 };
138 
139 /*
140  * List of vnodes that are ready for recycling.
141  */
142 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
143 
144 /*
145  * Free vnode target.  Free vnodes may simply be files which have been stat'd
146  * but not read.  This is somewhat common, and a small cache of such files
147  * should be kept to avoid recreation costs.
148  */
149 static u_long wantfreevnodes;
150 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
151 /* Number of vnodes in the free list. */
152 static u_long freevnodes;
153 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
154 
155 static int vlru_allow_cache_src;
156 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
157     &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
158 
159 /*
160  * Various variables used for debugging the new implementation of
161  * reassignbuf().
162  * XXX these are probably of (very) limited utility now.
163  */
164 static int reassignbufcalls;
165 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
166 
167 /*
168  * Cache for the mount type id assigned to NFS.  This is used for
169  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
170  */
171 int	nfs_mount_type = -1;
172 
173 /* To keep more than one thread at a time from running vfs_getnewfsid */
174 static struct mtx mntid_mtx;
175 
176 /*
177  * Lock for any access to the following:
178  *	vnode_free_list
179  *	numvnodes
180  *	freevnodes
181  */
182 static struct mtx vnode_free_list_mtx;
183 
184 /* Publicly exported FS */
185 struct nfs_public nfs_pub;
186 
187 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
188 static uma_zone_t vnode_zone;
189 static uma_zone_t vnodepoll_zone;
190 
191 /* Set to 1 to print out reclaim of active vnodes */
192 int	prtactive;
193 
194 /*
195  * The workitem queue.
196  *
197  * It is useful to delay writes of file data and filesystem metadata
198  * for tens of seconds so that quickly created and deleted files need
199  * not waste disk bandwidth being created and removed. To realize this,
200  * we append vnodes to a "workitem" queue. When running with a soft
201  * updates implementation, most pending metadata dependencies should
202  * not wait for more than a few seconds. Thus, mounted on block devices
203  * are delayed only about a half the time that file data is delayed.
204  * Similarly, directory updates are more critical, so are only delayed
205  * about a third the time that file data is delayed. Thus, there are
206  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
207  * one each second (driven off the filesystem syncer process). The
208  * syncer_delayno variable indicates the next queue that is to be processed.
209  * Items that need to be processed soon are placed in this queue:
210  *
211  *	syncer_workitem_pending[syncer_delayno]
212  *
213  * A delay of fifteen seconds is done by placing the request fifteen
214  * entries later in the queue:
215  *
216  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
217  *
218  */
219 static int syncer_delayno;
220 static long syncer_mask;
221 LIST_HEAD(synclist, bufobj);
222 static struct synclist *syncer_workitem_pending[2];
223 /*
224  * The sync_mtx protects:
225  *	bo->bo_synclist
226  *	sync_vnode_count
227  *	syncer_delayno
228  *	syncer_state
229  *	syncer_workitem_pending
230  *	syncer_worklist_len
231  *	rushjob
232  */
233 static struct mtx sync_mtx;
234 static struct cv sync_wakeup;
235 
236 #define SYNCER_MAXDELAY		32
237 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
238 static int syncdelay = 30;		/* max time to delay syncing data */
239 static int filedelay = 30;		/* time to delay syncing files */
240 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
241 static int dirdelay = 29;		/* time to delay syncing directories */
242 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
243 static int metadelay = 28;		/* time to delay syncing metadata */
244 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
245 static int rushjob;		/* number of slots to run ASAP */
246 static int stat_rush_requests;	/* number of times I/O speeded up */
247 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
248 
249 /*
250  * When shutting down the syncer, run it at four times normal speed.
251  */
252 #define SYNCER_SHUTDOWN_SPEEDUP		4
253 static int sync_vnode_count;
254 static int syncer_worklist_len;
255 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
256     syncer_state;
257 
258 /*
259  * Number of vnodes we want to exist at any one time.  This is mostly used
260  * to size hash tables in vnode-related code.  It is normally not used in
261  * getnewvnode(), as wantfreevnodes is normally nonzero.)
262  *
263  * XXX desiredvnodes is historical cruft and should not exist.
264  */
265 int desiredvnodes;
266 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
267     &desiredvnodes, 0, "Maximum number of vnodes");
268 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
269     &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
270 static int vnlru_nowhere;
271 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
272     &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
273 
274 /*
275  * Macros to control when a vnode is freed and recycled.  All require
276  * the vnode interlock.
277  */
278 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
279 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
280 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
281 
282 
283 /*
284  * Initialize the vnode management data structures.
285  */
286 #ifndef	MAXVNODES_MAX
287 #define	MAXVNODES_MAX	100000
288 #endif
289 static void
290 vntblinit(void *dummy __unused)
291 {
292 
293 	/*
294 	 * Desiredvnodes is a function of the physical memory size and
295 	 * the kernel's heap size.  Specifically, desiredvnodes scales
296 	 * in proportion to the physical memory size until two fifths
297 	 * of the kernel's heap size is consumed by vnodes and vm
298 	 * objects.
299 	 */
300 	desiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * vm_kmem_size /
301 	    (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
302 	if (desiredvnodes > MAXVNODES_MAX) {
303 		if (bootverbose)
304 			printf("Reducing kern.maxvnodes %d -> %d\n",
305 			    desiredvnodes, MAXVNODES_MAX);
306 		desiredvnodes = MAXVNODES_MAX;
307 	}
308 	wantfreevnodes = desiredvnodes / 4;
309 	mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
310 	TAILQ_INIT(&vnode_free_list);
311 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
312 	vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
313 	    NULL, NULL, UMA_ALIGN_PTR, 0);
314 	vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
315 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
316 	/*
317 	 * Initialize the filesystem syncer.
318 	 */
319 	syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE,
320 	    &syncer_mask);
321 	syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE,
322 	    &syncer_mask);
323 	syncer_maxdelay = syncer_mask + 1;
324 	mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
325 	cv_init(&sync_wakeup, "syncer");
326 }
327 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
328 
329 
330 /*
331  * Mark a mount point as busy. Used to synchronize access and to delay
332  * unmounting. Eventually, mountlist_mtx is not released on failure.
333  */
334 int
335 vfs_busy(struct mount *mp, int flags)
336 {
337 
338 	MPASS((flags & ~MBF_MASK) == 0);
339 	CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
340 
341 	MNT_ILOCK(mp);
342 	MNT_REF(mp);
343 	/*
344 	 * If mount point is currenly being unmounted, sleep until the
345 	 * mount point fate is decided.  If thread doing the unmounting fails,
346 	 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
347 	 * that this mount point has survived the unmount attempt and vfs_busy
348 	 * should retry.  Otherwise the unmounter thread will set MNTK_REFEXPIRE
349 	 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
350 	 * about to be really destroyed.  vfs_busy needs to release its
351 	 * reference on the mount point in this case and return with ENOENT,
352 	 * telling the caller that mount mount it tried to busy is no longer
353 	 * valid.
354 	 */
355 	while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
356 		if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
357 			MNT_REL(mp);
358 			MNT_IUNLOCK(mp);
359 			CTR1(KTR_VFS, "%s: failed busying before sleeping",
360 			    __func__);
361 			return (ENOENT);
362 		}
363 		if (flags & MBF_MNTLSTLOCK)
364 			mtx_unlock(&mountlist_mtx);
365 		mp->mnt_kern_flag |= MNTK_MWAIT;
366 		msleep(mp, MNT_MTX(mp), PVFS, "vfs_busy", 0);
367 		if (flags & MBF_MNTLSTLOCK)
368 			mtx_lock(&mountlist_mtx);
369 	}
370 	if (flags & MBF_MNTLSTLOCK)
371 		mtx_unlock(&mountlist_mtx);
372 	mp->mnt_lockref++;
373 	MNT_IUNLOCK(mp);
374 	return (0);
375 }
376 
377 /*
378  * Free a busy filesystem.
379  */
380 void
381 vfs_unbusy(struct mount *mp)
382 {
383 
384 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
385 	MNT_ILOCK(mp);
386 	MNT_REL(mp);
387 	KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
388 	mp->mnt_lockref--;
389 	if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
390 		MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
391 		CTR1(KTR_VFS, "%s: waking up waiters", __func__);
392 		mp->mnt_kern_flag &= ~MNTK_DRAINING;
393 		wakeup(&mp->mnt_lockref);
394 	}
395 	MNT_IUNLOCK(mp);
396 }
397 
398 /*
399  * Lookup a mount point by filesystem identifier.
400  */
401 struct mount *
402 vfs_getvfs(fsid_t *fsid)
403 {
404 	struct mount *mp;
405 
406 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
407 	mtx_lock(&mountlist_mtx);
408 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
409 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
410 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
411 			vfs_ref(mp);
412 			mtx_unlock(&mountlist_mtx);
413 			return (mp);
414 		}
415 	}
416 	mtx_unlock(&mountlist_mtx);
417 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
418 	return ((struct mount *) 0);
419 }
420 
421 /*
422  * Lookup a mount point by filesystem identifier, busying it before
423  * returning.
424  */
425 struct mount *
426 vfs_busyfs(fsid_t *fsid)
427 {
428 	struct mount *mp;
429 	int error;
430 
431 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
432 	mtx_lock(&mountlist_mtx);
433 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
434 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
435 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
436 			error = vfs_busy(mp, MBF_MNTLSTLOCK);
437 			if (error) {
438 				mtx_unlock(&mountlist_mtx);
439 				return (NULL);
440 			}
441 			return (mp);
442 		}
443 	}
444 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
445 	mtx_unlock(&mountlist_mtx);
446 	return ((struct mount *) 0);
447 }
448 
449 /*
450  * Check if a user can access privileged mount options.
451  */
452 int
453 vfs_suser(struct mount *mp, struct thread *td)
454 {
455 	int error;
456 
457 	/*
458 	 * If the thread is jailed, but this is not a jail-friendly file
459 	 * system, deny immediately.
460 	 */
461 	if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
462 		return (EPERM);
463 
464 	/*
465 	 * If the file system was mounted outside the jail of the calling
466 	 * thread, deny immediately.
467 	 */
468 	if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
469 		return (EPERM);
470 
471 	/*
472 	 * If file system supports delegated administration, we don't check
473 	 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
474 	 * by the file system itself.
475 	 * If this is not the user that did original mount, we check for
476 	 * the PRIV_VFS_MOUNT_OWNER privilege.
477 	 */
478 	if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
479 	    mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
480 		if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
481 			return (error);
482 	}
483 	return (0);
484 }
485 
486 /*
487  * Get a new unique fsid.  Try to make its val[0] unique, since this value
488  * will be used to create fake device numbers for stat().  Also try (but
489  * not so hard) make its val[0] unique mod 2^16, since some emulators only
490  * support 16-bit device numbers.  We end up with unique val[0]'s for the
491  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
492  *
493  * Keep in mind that several mounts may be running in parallel.  Starting
494  * the search one past where the previous search terminated is both a
495  * micro-optimization and a defense against returning the same fsid to
496  * different mounts.
497  */
498 void
499 vfs_getnewfsid(struct mount *mp)
500 {
501 	static u_int16_t mntid_base;
502 	struct mount *nmp;
503 	fsid_t tfsid;
504 	int mtype;
505 
506 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
507 	mtx_lock(&mntid_mtx);
508 	mtype = mp->mnt_vfc->vfc_typenum;
509 	tfsid.val[1] = mtype;
510 	mtype = (mtype & 0xFF) << 24;
511 	for (;;) {
512 		tfsid.val[0] = makedev(255,
513 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
514 		mntid_base++;
515 		if ((nmp = vfs_getvfs(&tfsid)) == NULL)
516 			break;
517 		vfs_rel(nmp);
518 	}
519 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
520 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
521 	mtx_unlock(&mntid_mtx);
522 }
523 
524 /*
525  * Knob to control the precision of file timestamps:
526  *
527  *   0 = seconds only; nanoseconds zeroed.
528  *   1 = seconds and nanoseconds, accurate within 1/HZ.
529  *   2 = seconds and nanoseconds, truncated to microseconds.
530  * >=3 = seconds and nanoseconds, maximum precision.
531  */
532 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
533 
534 static int timestamp_precision = TSP_SEC;
535 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
536     &timestamp_precision, 0, "");
537 
538 /*
539  * Get a current timestamp.
540  */
541 void
542 vfs_timestamp(struct timespec *tsp)
543 {
544 	struct timeval tv;
545 
546 	switch (timestamp_precision) {
547 	case TSP_SEC:
548 		tsp->tv_sec = time_second;
549 		tsp->tv_nsec = 0;
550 		break;
551 	case TSP_HZ:
552 		getnanotime(tsp);
553 		break;
554 	case TSP_USEC:
555 		microtime(&tv);
556 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
557 		break;
558 	case TSP_NSEC:
559 	default:
560 		nanotime(tsp);
561 		break;
562 	}
563 }
564 
565 /*
566  * Set vnode attributes to VNOVAL
567  */
568 void
569 vattr_null(struct vattr *vap)
570 {
571 
572 	vap->va_type = VNON;
573 	vap->va_size = VNOVAL;
574 	vap->va_bytes = VNOVAL;
575 	vap->va_mode = VNOVAL;
576 	vap->va_nlink = VNOVAL;
577 	vap->va_uid = VNOVAL;
578 	vap->va_gid = VNOVAL;
579 	vap->va_fsid = VNOVAL;
580 	vap->va_fileid = VNOVAL;
581 	vap->va_blocksize = VNOVAL;
582 	vap->va_rdev = VNOVAL;
583 	vap->va_atime.tv_sec = VNOVAL;
584 	vap->va_atime.tv_nsec = VNOVAL;
585 	vap->va_mtime.tv_sec = VNOVAL;
586 	vap->va_mtime.tv_nsec = VNOVAL;
587 	vap->va_ctime.tv_sec = VNOVAL;
588 	vap->va_ctime.tv_nsec = VNOVAL;
589 	vap->va_birthtime.tv_sec = VNOVAL;
590 	vap->va_birthtime.tv_nsec = VNOVAL;
591 	vap->va_flags = VNOVAL;
592 	vap->va_gen = VNOVAL;
593 	vap->va_vaflags = 0;
594 }
595 
596 /*
597  * This routine is called when we have too many vnodes.  It attempts
598  * to free <count> vnodes and will potentially free vnodes that still
599  * have VM backing store (VM backing store is typically the cause
600  * of a vnode blowout so we want to do this).  Therefore, this operation
601  * is not considered cheap.
602  *
603  * A number of conditions may prevent a vnode from being reclaimed.
604  * the buffer cache may have references on the vnode, a directory
605  * vnode may still have references due to the namei cache representing
606  * underlying files, or the vnode may be in active use.   It is not
607  * desireable to reuse such vnodes.  These conditions may cause the
608  * number of vnodes to reach some minimum value regardless of what
609  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
610  */
611 static int
612 vlrureclaim(struct mount *mp)
613 {
614 	struct vnode *vp;
615 	int done;
616 	int trigger;
617 	int usevnodes;
618 	int count;
619 
620 	/*
621 	 * Calculate the trigger point, don't allow user
622 	 * screwups to blow us up.   This prevents us from
623 	 * recycling vnodes with lots of resident pages.  We
624 	 * aren't trying to free memory, we are trying to
625 	 * free vnodes.
626 	 */
627 	usevnodes = desiredvnodes;
628 	if (usevnodes <= 0)
629 		usevnodes = 1;
630 	trigger = cnt.v_page_count * 2 / usevnodes;
631 	done = 0;
632 	vn_start_write(NULL, &mp, V_WAIT);
633 	MNT_ILOCK(mp);
634 	count = mp->mnt_nvnodelistsize / 10 + 1;
635 	while (count != 0) {
636 		vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
637 		while (vp != NULL && vp->v_type == VMARKER)
638 			vp = TAILQ_NEXT(vp, v_nmntvnodes);
639 		if (vp == NULL)
640 			break;
641 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
642 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
643 		--count;
644 		if (!VI_TRYLOCK(vp))
645 			goto next_iter;
646 		/*
647 		 * If it's been deconstructed already, it's still
648 		 * referenced, or it exceeds the trigger, skip it.
649 		 */
650 		if (vp->v_usecount ||
651 		    (!vlru_allow_cache_src &&
652 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
653 		    (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
654 		    vp->v_object->resident_page_count > trigger)) {
655 			VI_UNLOCK(vp);
656 			goto next_iter;
657 		}
658 		MNT_IUNLOCK(mp);
659 		vholdl(vp);
660 		if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
661 			vdrop(vp);
662 			goto next_iter_mntunlocked;
663 		}
664 		VI_LOCK(vp);
665 		/*
666 		 * v_usecount may have been bumped after VOP_LOCK() dropped
667 		 * the vnode interlock and before it was locked again.
668 		 *
669 		 * It is not necessary to recheck VI_DOOMED because it can
670 		 * only be set by another thread that holds both the vnode
671 		 * lock and vnode interlock.  If another thread has the
672 		 * vnode lock before we get to VOP_LOCK() and obtains the
673 		 * vnode interlock after VOP_LOCK() drops the vnode
674 		 * interlock, the other thread will be unable to drop the
675 		 * vnode lock before our VOP_LOCK() call fails.
676 		 */
677 		if (vp->v_usecount ||
678 		    (!vlru_allow_cache_src &&
679 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
680 		    (vp->v_object != NULL &&
681 		    vp->v_object->resident_page_count > trigger)) {
682 			VOP_UNLOCK(vp, LK_INTERLOCK);
683 			goto next_iter_mntunlocked;
684 		}
685 		KASSERT((vp->v_iflag & VI_DOOMED) == 0,
686 		    ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
687 		vgonel(vp);
688 		VOP_UNLOCK(vp, 0);
689 		vdropl(vp);
690 		done++;
691 next_iter_mntunlocked:
692 		if ((count % 256) != 0)
693 			goto relock_mnt;
694 		goto yield;
695 next_iter:
696 		if ((count % 256) != 0)
697 			continue;
698 		MNT_IUNLOCK(mp);
699 yield:
700 		uio_yield();
701 relock_mnt:
702 		MNT_ILOCK(mp);
703 	}
704 	MNT_IUNLOCK(mp);
705 	vn_finished_write(mp);
706 	return done;
707 }
708 
709 /*
710  * Attempt to keep the free list at wantfreevnodes length.
711  */
712 static void
713 vnlru_free(int count)
714 {
715 	struct vnode *vp;
716 	int vfslocked;
717 
718 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
719 	for (; count > 0; count--) {
720 		vp = TAILQ_FIRST(&vnode_free_list);
721 		/*
722 		 * The list can be modified while the free_list_mtx
723 		 * has been dropped and vp could be NULL here.
724 		 */
725 		if (!vp)
726 			break;
727 		VNASSERT(vp->v_op != NULL, vp,
728 		    ("vnlru_free: vnode already reclaimed."));
729 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
730 		/*
731 		 * Don't recycle if we can't get the interlock.
732 		 */
733 		if (!VI_TRYLOCK(vp)) {
734 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
735 			continue;
736 		}
737 		VNASSERT(VCANRECYCLE(vp), vp,
738 		    ("vp inconsistent on freelist"));
739 		freevnodes--;
740 		vp->v_iflag &= ~VI_FREE;
741 		vholdl(vp);
742 		mtx_unlock(&vnode_free_list_mtx);
743 		VI_UNLOCK(vp);
744 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
745 		vtryrecycle(vp);
746 		VFS_UNLOCK_GIANT(vfslocked);
747 		/*
748 		 * If the recycled succeeded this vdrop will actually free
749 		 * the vnode.  If not it will simply place it back on
750 		 * the free list.
751 		 */
752 		vdrop(vp);
753 		mtx_lock(&vnode_free_list_mtx);
754 	}
755 }
756 /*
757  * Attempt to recycle vnodes in a context that is always safe to block.
758  * Calling vlrurecycle() from the bowels of filesystem code has some
759  * interesting deadlock problems.
760  */
761 static struct proc *vnlruproc;
762 static int vnlruproc_sig;
763 
764 static void
765 vnlru_proc(void)
766 {
767 	struct mount *mp, *nmp;
768 	int done, vfslocked;
769 	struct proc *p = vnlruproc;
770 
771 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
772 	    SHUTDOWN_PRI_FIRST);
773 
774 	for (;;) {
775 		kproc_suspend_check(p);
776 		mtx_lock(&vnode_free_list_mtx);
777 		if (freevnodes > wantfreevnodes)
778 			vnlru_free(freevnodes - wantfreevnodes);
779 		if (numvnodes <= desiredvnodes * 9 / 10) {
780 			vnlruproc_sig = 0;
781 			wakeup(&vnlruproc_sig);
782 			msleep(vnlruproc, &vnode_free_list_mtx,
783 			    PVFS|PDROP, "vlruwt", hz);
784 			continue;
785 		}
786 		mtx_unlock(&vnode_free_list_mtx);
787 		done = 0;
788 		mtx_lock(&mountlist_mtx);
789 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
790 			if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
791 				nmp = TAILQ_NEXT(mp, mnt_list);
792 				continue;
793 			}
794 			vfslocked = VFS_LOCK_GIANT(mp);
795 			done += vlrureclaim(mp);
796 			VFS_UNLOCK_GIANT(vfslocked);
797 			mtx_lock(&mountlist_mtx);
798 			nmp = TAILQ_NEXT(mp, mnt_list);
799 			vfs_unbusy(mp);
800 		}
801 		mtx_unlock(&mountlist_mtx);
802 		if (done == 0) {
803 			EVENTHANDLER_INVOKE(vfs_lowvnodes, desiredvnodes / 10);
804 #if 0
805 			/* These messages are temporary debugging aids */
806 			if (vnlru_nowhere < 5)
807 				printf("vnlru process getting nowhere..\n");
808 			else if (vnlru_nowhere == 5)
809 				printf("vnlru process messages stopped.\n");
810 #endif
811 			vnlru_nowhere++;
812 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
813 		} else
814 			uio_yield();
815 	}
816 }
817 
818 static struct kproc_desc vnlru_kp = {
819 	"vnlru",
820 	vnlru_proc,
821 	&vnlruproc
822 };
823 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
824     &vnlru_kp);
825 
826 /*
827  * Routines having to do with the management of the vnode table.
828  */
829 
830 void
831 vdestroy(struct vnode *vp)
832 {
833 	struct bufobj *bo;
834 
835 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
836 	mtx_lock(&vnode_free_list_mtx);
837 	numvnodes--;
838 	mtx_unlock(&vnode_free_list_mtx);
839 	bo = &vp->v_bufobj;
840 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
841 	    ("cleaned vnode still on the free list."));
842 	VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
843 	VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
844 	VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
845 	VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
846 	VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
847 	VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
848 	VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
849 	VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
850 	VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
851 	VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
852 	VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
853 	VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
854 	VI_UNLOCK(vp);
855 #ifdef MAC
856 	mac_vnode_destroy(vp);
857 #endif
858 	if (vp->v_pollinfo != NULL)
859 		destroy_vpollinfo(vp->v_pollinfo);
860 #ifdef INVARIANTS
861 	/* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */
862 	vp->v_op = NULL;
863 #endif
864 	lockdestroy(vp->v_vnlock);
865 	mtx_destroy(&vp->v_interlock);
866 	mtx_destroy(BO_MTX(bo));
867 	uma_zfree(vnode_zone, vp);
868 }
869 
870 /*
871  * Try to recycle a freed vnode.  We abort if anyone picks up a reference
872  * before we actually vgone().  This function must be called with the vnode
873  * held to prevent the vnode from being returned to the free list midway
874  * through vgone().
875  */
876 static int
877 vtryrecycle(struct vnode *vp)
878 {
879 	struct mount *vnmp;
880 
881 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
882 	VNASSERT(vp->v_holdcnt, vp,
883 	    ("vtryrecycle: Recycling vp %p without a reference.", vp));
884 	/*
885 	 * This vnode may found and locked via some other list, if so we
886 	 * can't recycle it yet.
887 	 */
888 	if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
889 		CTR2(KTR_VFS,
890 		    "%s: impossible to recycle, vp %p lock is already held",
891 		    __func__, vp);
892 		return (EWOULDBLOCK);
893 	}
894 	/*
895 	 * Don't recycle if its filesystem is being suspended.
896 	 */
897 	if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
898 		VOP_UNLOCK(vp, 0);
899 		CTR2(KTR_VFS,
900 		    "%s: impossible to recycle, cannot start the write for %p",
901 		    __func__, vp);
902 		return (EBUSY);
903 	}
904 	/*
905 	 * If we got this far, we need to acquire the interlock and see if
906 	 * anyone picked up this vnode from another list.  If not, we will
907 	 * mark it with DOOMED via vgonel() so that anyone who does find it
908 	 * will skip over it.
909 	 */
910 	VI_LOCK(vp);
911 	if (vp->v_usecount) {
912 		VOP_UNLOCK(vp, LK_INTERLOCK);
913 		vn_finished_write(vnmp);
914 		CTR2(KTR_VFS,
915 		    "%s: impossible to recycle, %p is already referenced",
916 		    __func__, vp);
917 		return (EBUSY);
918 	}
919 	if ((vp->v_iflag & VI_DOOMED) == 0)
920 		vgonel(vp);
921 	VOP_UNLOCK(vp, LK_INTERLOCK);
922 	vn_finished_write(vnmp);
923 	return (0);
924 }
925 
926 /*
927  * Return the next vnode from the free list.
928  */
929 int
930 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
931     struct vnode **vpp)
932 {
933 	struct vnode *vp = NULL;
934 	struct bufobj *bo;
935 
936 	CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
937 	mtx_lock(&vnode_free_list_mtx);
938 	/*
939 	 * Lend our context to reclaim vnodes if they've exceeded the max.
940 	 */
941 	if (freevnodes > wantfreevnodes)
942 		vnlru_free(1);
943 	/*
944 	 * Wait for available vnodes.
945 	 */
946 	if (numvnodes > desiredvnodes) {
947 		if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) {
948 			/*
949 			 * File system is beeing suspended, we cannot risk a
950 			 * deadlock here, so allocate new vnode anyway.
951 			 */
952 			if (freevnodes > wantfreevnodes)
953 				vnlru_free(freevnodes - wantfreevnodes);
954 			goto alloc;
955 		}
956 		if (vnlruproc_sig == 0) {
957 			vnlruproc_sig = 1;	/* avoid unnecessary wakeups */
958 			wakeup(vnlruproc);
959 		}
960 		msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
961 		    "vlruwk", hz);
962 #if 0	/* XXX Not all VFS_VGET/ffs_vget callers check returns. */
963 		if (numvnodes > desiredvnodes) {
964 			mtx_unlock(&vnode_free_list_mtx);
965 			return (ENFILE);
966 		}
967 #endif
968 	}
969 alloc:
970 	numvnodes++;
971 	mtx_unlock(&vnode_free_list_mtx);
972 	vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
973 	/*
974 	 * Setup locks.
975 	 */
976 	vp->v_vnlock = &vp->v_lock;
977 	mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
978 	/*
979 	 * By default, don't allow shared locks unless filesystems
980 	 * opt-in.
981 	 */
982 	lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
983 	/*
984 	 * Initialize bufobj.
985 	 */
986 	bo = &vp->v_bufobj;
987 	bo->__bo_vnode = vp;
988 	mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
989 	bo->bo_ops = &buf_ops_bio;
990 	bo->bo_private = vp;
991 	TAILQ_INIT(&bo->bo_clean.bv_hd);
992 	TAILQ_INIT(&bo->bo_dirty.bv_hd);
993 	/*
994 	 * Initialize namecache.
995 	 */
996 	LIST_INIT(&vp->v_cache_src);
997 	TAILQ_INIT(&vp->v_cache_dst);
998 	/*
999 	 * Finalize various vnode identity bits.
1000 	 */
1001 	vp->v_type = VNON;
1002 	vp->v_tag = tag;
1003 	vp->v_op = vops;
1004 	v_incr_usecount(vp);
1005 	vp->v_data = 0;
1006 #ifdef MAC
1007 	mac_vnode_init(vp);
1008 	if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1009 		mac_vnode_associate_singlelabel(mp, vp);
1010 	else if (mp == NULL && vops != &dead_vnodeops)
1011 		printf("NULL mp in getnewvnode()\n");
1012 #endif
1013 	if (mp != NULL) {
1014 		bo->bo_bsize = mp->mnt_stat.f_iosize;
1015 		if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1016 			vp->v_vflag |= VV_NOKNOTE;
1017 	}
1018 
1019 	*vpp = vp;
1020 	return (0);
1021 }
1022 
1023 /*
1024  * Delete from old mount point vnode list, if on one.
1025  */
1026 static void
1027 delmntque(struct vnode *vp)
1028 {
1029 	struct mount *mp;
1030 
1031 	mp = vp->v_mount;
1032 	if (mp == NULL)
1033 		return;
1034 	MNT_ILOCK(mp);
1035 	vp->v_mount = NULL;
1036 	VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1037 		("bad mount point vnode list size"));
1038 	TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1039 	mp->mnt_nvnodelistsize--;
1040 	MNT_REL(mp);
1041 	MNT_IUNLOCK(mp);
1042 }
1043 
1044 static void
1045 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1046 {
1047 
1048 	vp->v_data = NULL;
1049 	vp->v_op = &dead_vnodeops;
1050 	/* XXX non mp-safe fs may still call insmntque with vnode
1051 	   unlocked */
1052 	if (!VOP_ISLOCKED(vp))
1053 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1054 	vgone(vp);
1055 	vput(vp);
1056 }
1057 
1058 /*
1059  * Insert into list of vnodes for the new mount point, if available.
1060  */
1061 int
1062 insmntque1(struct vnode *vp, struct mount *mp,
1063 	void (*dtr)(struct vnode *, void *), void *dtr_arg)
1064 {
1065 	int locked;
1066 
1067 	KASSERT(vp->v_mount == NULL,
1068 		("insmntque: vnode already on per mount vnode list"));
1069 	VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1070 #ifdef DEBUG_VFS_LOCKS
1071 	if (!VFS_NEEDSGIANT(mp))
1072 		ASSERT_VOP_ELOCKED(vp,
1073 		    "insmntque: mp-safe fs and non-locked vp");
1074 #endif
1075 	MNT_ILOCK(mp);
1076 	if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1077 	    ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1078 	     mp->mnt_nvnodelistsize == 0)) {
1079 		locked = VOP_ISLOCKED(vp);
1080 		if (!locked || (locked == LK_EXCLUSIVE &&
1081 		     (vp->v_vflag & VV_FORCEINSMQ) == 0)) {
1082 			MNT_IUNLOCK(mp);
1083 			if (dtr != NULL)
1084 				dtr(vp, dtr_arg);
1085 			return (EBUSY);
1086 		}
1087 	}
1088 	vp->v_mount = mp;
1089 	MNT_REF(mp);
1090 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1091 	VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1092 		("neg mount point vnode list size"));
1093 	mp->mnt_nvnodelistsize++;
1094 	MNT_IUNLOCK(mp);
1095 	return (0);
1096 }
1097 
1098 int
1099 insmntque(struct vnode *vp, struct mount *mp)
1100 {
1101 
1102 	return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1103 }
1104 
1105 /*
1106  * Flush out and invalidate all buffers associated with a bufobj
1107  * Called with the underlying object locked.
1108  */
1109 int
1110 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1111 {
1112 	int error;
1113 
1114 	BO_LOCK(bo);
1115 	if (flags & V_SAVE) {
1116 		error = bufobj_wwait(bo, slpflag, slptimeo);
1117 		if (error) {
1118 			BO_UNLOCK(bo);
1119 			return (error);
1120 		}
1121 		if (bo->bo_dirty.bv_cnt > 0) {
1122 			BO_UNLOCK(bo);
1123 			if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1124 				return (error);
1125 			/*
1126 			 * XXX We could save a lock/unlock if this was only
1127 			 * enabled under INVARIANTS
1128 			 */
1129 			BO_LOCK(bo);
1130 			if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1131 				panic("vinvalbuf: dirty bufs");
1132 		}
1133 	}
1134 	/*
1135 	 * If you alter this loop please notice that interlock is dropped and
1136 	 * reacquired in flushbuflist.  Special care is needed to ensure that
1137 	 * no race conditions occur from this.
1138 	 */
1139 	do {
1140 		error = flushbuflist(&bo->bo_clean,
1141 		    flags, bo, slpflag, slptimeo);
1142 		if (error == 0)
1143 			error = flushbuflist(&bo->bo_dirty,
1144 			    flags, bo, slpflag, slptimeo);
1145 		if (error != 0 && error != EAGAIN) {
1146 			BO_UNLOCK(bo);
1147 			return (error);
1148 		}
1149 	} while (error != 0);
1150 
1151 	/*
1152 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1153 	 * have write I/O in-progress but if there is a VM object then the
1154 	 * VM object can also have read-I/O in-progress.
1155 	 */
1156 	do {
1157 		bufobj_wwait(bo, 0, 0);
1158 		BO_UNLOCK(bo);
1159 		if (bo->bo_object != NULL) {
1160 			VM_OBJECT_LOCK(bo->bo_object);
1161 			vm_object_pip_wait(bo->bo_object, "bovlbx");
1162 			VM_OBJECT_UNLOCK(bo->bo_object);
1163 		}
1164 		BO_LOCK(bo);
1165 	} while (bo->bo_numoutput > 0);
1166 	BO_UNLOCK(bo);
1167 
1168 	/*
1169 	 * Destroy the copy in the VM cache, too.
1170 	 */
1171 	if (bo->bo_object != NULL && (flags & (V_ALT | V_NORMAL)) == 0) {
1172 		VM_OBJECT_LOCK(bo->bo_object);
1173 		vm_object_page_remove(bo->bo_object, 0, 0,
1174 			(flags & V_SAVE) ? TRUE : FALSE);
1175 		VM_OBJECT_UNLOCK(bo->bo_object);
1176 	}
1177 
1178 #ifdef INVARIANTS
1179 	BO_LOCK(bo);
1180 	if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1181 	    (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1182 		panic("vinvalbuf: flush failed");
1183 	BO_UNLOCK(bo);
1184 #endif
1185 	return (0);
1186 }
1187 
1188 /*
1189  * Flush out and invalidate all buffers associated with a vnode.
1190  * Called with the underlying object locked.
1191  */
1192 int
1193 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1194 {
1195 
1196 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1197 	ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1198 	return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1199 }
1200 
1201 /*
1202  * Flush out buffers on the specified list.
1203  *
1204  */
1205 static int
1206 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1207     int slptimeo)
1208 {
1209 	struct buf *bp, *nbp;
1210 	int retval, error;
1211 	daddr_t lblkno;
1212 	b_xflags_t xflags;
1213 
1214 	ASSERT_BO_LOCKED(bo);
1215 
1216 	retval = 0;
1217 	TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1218 		if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1219 		    ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1220 			continue;
1221 		}
1222 		lblkno = 0;
1223 		xflags = 0;
1224 		if (nbp != NULL) {
1225 			lblkno = nbp->b_lblkno;
1226 			xflags = nbp->b_xflags &
1227 				(BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1228 		}
1229 		retval = EAGAIN;
1230 		error = BUF_TIMELOCK(bp,
1231 		    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1232 		    "flushbuf", slpflag, slptimeo);
1233 		if (error) {
1234 			BO_LOCK(bo);
1235 			return (error != ENOLCK ? error : EAGAIN);
1236 		}
1237 		KASSERT(bp->b_bufobj == bo,
1238 		    ("bp %p wrong b_bufobj %p should be %p",
1239 		    bp, bp->b_bufobj, bo));
1240 		if (bp->b_bufobj != bo) {	/* XXX: necessary ? */
1241 			BUF_UNLOCK(bp);
1242 			BO_LOCK(bo);
1243 			return (EAGAIN);
1244 		}
1245 		/*
1246 		 * XXX Since there are no node locks for NFS, I
1247 		 * believe there is a slight chance that a delayed
1248 		 * write will occur while sleeping just above, so
1249 		 * check for it.
1250 		 */
1251 		if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1252 		    (flags & V_SAVE)) {
1253 			bremfree(bp);
1254 			bp->b_flags |= B_ASYNC;
1255 			bwrite(bp);
1256 			BO_LOCK(bo);
1257 			return (EAGAIN);	/* XXX: why not loop ? */
1258 		}
1259 		bremfree(bp);
1260 		bp->b_flags |= (B_INVAL | B_RELBUF);
1261 		bp->b_flags &= ~B_ASYNC;
1262 		brelse(bp);
1263 		BO_LOCK(bo);
1264 		if (nbp != NULL &&
1265 		    (nbp->b_bufobj != bo ||
1266 		     nbp->b_lblkno != lblkno ||
1267 		     (nbp->b_xflags &
1268 		      (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1269 			break;			/* nbp invalid */
1270 	}
1271 	return (retval);
1272 }
1273 
1274 /*
1275  * Truncate a file's buffer and pages to a specified length.  This
1276  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1277  * sync activity.
1278  */
1279 int
1280 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
1281     off_t length, int blksize)
1282 {
1283 	struct buf *bp, *nbp;
1284 	int anyfreed;
1285 	int trunclbn;
1286 	struct bufobj *bo;
1287 
1288 	CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1289 	    vp, cred, blksize, (uintmax_t)length);
1290 
1291 	/*
1292 	 * Round up to the *next* lbn.
1293 	 */
1294 	trunclbn = (length + blksize - 1) / blksize;
1295 
1296 	ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1297 restart:
1298 	bo = &vp->v_bufobj;
1299 	BO_LOCK(bo);
1300 	anyfreed = 1;
1301 	for (;anyfreed;) {
1302 		anyfreed = 0;
1303 		TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1304 			if (bp->b_lblkno < trunclbn)
1305 				continue;
1306 			if (BUF_LOCK(bp,
1307 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1308 			    BO_MTX(bo)) == ENOLCK)
1309 				goto restart;
1310 
1311 			bremfree(bp);
1312 			bp->b_flags |= (B_INVAL | B_RELBUF);
1313 			bp->b_flags &= ~B_ASYNC;
1314 			brelse(bp);
1315 			anyfreed = 1;
1316 
1317 			if (nbp != NULL &&
1318 			    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1319 			    (nbp->b_vp != vp) ||
1320 			    (nbp->b_flags & B_DELWRI))) {
1321 				goto restart;
1322 			}
1323 			BO_LOCK(bo);
1324 		}
1325 
1326 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1327 			if (bp->b_lblkno < trunclbn)
1328 				continue;
1329 			if (BUF_LOCK(bp,
1330 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1331 			    BO_MTX(bo)) == ENOLCK)
1332 				goto restart;
1333 			bremfree(bp);
1334 			bp->b_flags |= (B_INVAL | B_RELBUF);
1335 			bp->b_flags &= ~B_ASYNC;
1336 			brelse(bp);
1337 			anyfreed = 1;
1338 			if (nbp != NULL &&
1339 			    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1340 			    (nbp->b_vp != vp) ||
1341 			    (nbp->b_flags & B_DELWRI) == 0)) {
1342 				goto restart;
1343 			}
1344 			BO_LOCK(bo);
1345 		}
1346 	}
1347 
1348 	if (length > 0) {
1349 restartsync:
1350 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1351 			if (bp->b_lblkno > 0)
1352 				continue;
1353 			/*
1354 			 * Since we hold the vnode lock this should only
1355 			 * fail if we're racing with the buf daemon.
1356 			 */
1357 			if (BUF_LOCK(bp,
1358 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1359 			    BO_MTX(bo)) == ENOLCK) {
1360 				goto restart;
1361 			}
1362 			VNASSERT((bp->b_flags & B_DELWRI), vp,
1363 			    ("buf(%p) on dirty queue without DELWRI", bp));
1364 
1365 			bremfree(bp);
1366 			bawrite(bp);
1367 			BO_LOCK(bo);
1368 			goto restartsync;
1369 		}
1370 	}
1371 
1372 	bufobj_wwait(bo, 0, 0);
1373 	BO_UNLOCK(bo);
1374 	vnode_pager_setsize(vp, length);
1375 
1376 	return (0);
1377 }
1378 
1379 /*
1380  * buf_splay() - splay tree core for the clean/dirty list of buffers in
1381  * 		 a vnode.
1382  *
1383  *	NOTE: We have to deal with the special case of a background bitmap
1384  *	buffer, a situation where two buffers will have the same logical
1385  *	block offset.  We want (1) only the foreground buffer to be accessed
1386  *	in a lookup and (2) must differentiate between the foreground and
1387  *	background buffer in the splay tree algorithm because the splay
1388  *	tree cannot normally handle multiple entities with the same 'index'.
1389  *	We accomplish this by adding differentiating flags to the splay tree's
1390  *	numerical domain.
1391  */
1392 static
1393 struct buf *
1394 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1395 {
1396 	struct buf dummy;
1397 	struct buf *lefttreemax, *righttreemin, *y;
1398 
1399 	if (root == NULL)
1400 		return (NULL);
1401 	lefttreemax = righttreemin = &dummy;
1402 	for (;;) {
1403 		if (lblkno < root->b_lblkno ||
1404 		    (lblkno == root->b_lblkno &&
1405 		    (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1406 			if ((y = root->b_left) == NULL)
1407 				break;
1408 			if (lblkno < y->b_lblkno) {
1409 				/* Rotate right. */
1410 				root->b_left = y->b_right;
1411 				y->b_right = root;
1412 				root = y;
1413 				if ((y = root->b_left) == NULL)
1414 					break;
1415 			}
1416 			/* Link into the new root's right tree. */
1417 			righttreemin->b_left = root;
1418 			righttreemin = root;
1419 		} else if (lblkno > root->b_lblkno ||
1420 		    (lblkno == root->b_lblkno &&
1421 		    (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1422 			if ((y = root->b_right) == NULL)
1423 				break;
1424 			if (lblkno > y->b_lblkno) {
1425 				/* Rotate left. */
1426 				root->b_right = y->b_left;
1427 				y->b_left = root;
1428 				root = y;
1429 				if ((y = root->b_right) == NULL)
1430 					break;
1431 			}
1432 			/* Link into the new root's left tree. */
1433 			lefttreemax->b_right = root;
1434 			lefttreemax = root;
1435 		} else {
1436 			break;
1437 		}
1438 		root = y;
1439 	}
1440 	/* Assemble the new root. */
1441 	lefttreemax->b_right = root->b_left;
1442 	righttreemin->b_left = root->b_right;
1443 	root->b_left = dummy.b_right;
1444 	root->b_right = dummy.b_left;
1445 	return (root);
1446 }
1447 
1448 static void
1449 buf_vlist_remove(struct buf *bp)
1450 {
1451 	struct buf *root;
1452 	struct bufv *bv;
1453 
1454 	KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1455 	ASSERT_BO_LOCKED(bp->b_bufobj);
1456 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1457 	    (BX_VNDIRTY|BX_VNCLEAN),
1458 	    ("buf_vlist_remove: Buf %p is on two lists", bp));
1459 	if (bp->b_xflags & BX_VNDIRTY)
1460 		bv = &bp->b_bufobj->bo_dirty;
1461 	else
1462 		bv = &bp->b_bufobj->bo_clean;
1463 	if (bp != bv->bv_root) {
1464 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1465 		KASSERT(root == bp, ("splay lookup failed in remove"));
1466 	}
1467 	if (bp->b_left == NULL) {
1468 		root = bp->b_right;
1469 	} else {
1470 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1471 		root->b_right = bp->b_right;
1472 	}
1473 	bv->bv_root = root;
1474 	TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1475 	bv->bv_cnt--;
1476 	bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1477 }
1478 
1479 /*
1480  * Add the buffer to the sorted clean or dirty block list using a
1481  * splay tree algorithm.
1482  *
1483  * NOTE: xflags is passed as a constant, optimizing this inline function!
1484  */
1485 static void
1486 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1487 {
1488 	struct buf *root;
1489 	struct bufv *bv;
1490 
1491 	ASSERT_BO_LOCKED(bo);
1492 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1493 	    ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1494 	bp->b_xflags |= xflags;
1495 	if (xflags & BX_VNDIRTY)
1496 		bv = &bo->bo_dirty;
1497 	else
1498 		bv = &bo->bo_clean;
1499 
1500 	root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1501 	if (root == NULL) {
1502 		bp->b_left = NULL;
1503 		bp->b_right = NULL;
1504 		TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1505 	} else if (bp->b_lblkno < root->b_lblkno ||
1506 	    (bp->b_lblkno == root->b_lblkno &&
1507 	    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1508 		bp->b_left = root->b_left;
1509 		bp->b_right = root;
1510 		root->b_left = NULL;
1511 		TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1512 	} else {
1513 		bp->b_right = root->b_right;
1514 		bp->b_left = root;
1515 		root->b_right = NULL;
1516 		TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1517 	}
1518 	bv->bv_cnt++;
1519 	bv->bv_root = bp;
1520 }
1521 
1522 /*
1523  * Lookup a buffer using the splay tree.  Note that we specifically avoid
1524  * shadow buffers used in background bitmap writes.
1525  *
1526  * This code isn't quite efficient as it could be because we are maintaining
1527  * two sorted lists and do not know which list the block resides in.
1528  *
1529  * During a "make buildworld" the desired buffer is found at one of
1530  * the roots more than 60% of the time.  Thus, checking both roots
1531  * before performing either splay eliminates unnecessary splays on the
1532  * first tree splayed.
1533  */
1534 struct buf *
1535 gbincore(struct bufobj *bo, daddr_t lblkno)
1536 {
1537 	struct buf *bp;
1538 
1539 	ASSERT_BO_LOCKED(bo);
1540 	if ((bp = bo->bo_clean.bv_root) != NULL &&
1541 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1542 		return (bp);
1543 	if ((bp = bo->bo_dirty.bv_root) != NULL &&
1544 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1545 		return (bp);
1546 	if ((bp = bo->bo_clean.bv_root) != NULL) {
1547 		bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1548 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1549 			return (bp);
1550 	}
1551 	if ((bp = bo->bo_dirty.bv_root) != NULL) {
1552 		bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1553 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1554 			return (bp);
1555 	}
1556 	return (NULL);
1557 }
1558 
1559 /*
1560  * Associate a buffer with a vnode.
1561  */
1562 void
1563 bgetvp(struct vnode *vp, struct buf *bp)
1564 {
1565 	struct bufobj *bo;
1566 
1567 	bo = &vp->v_bufobj;
1568 	ASSERT_BO_LOCKED(bo);
1569 	VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1570 
1571 	CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1572 	VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1573 	    ("bgetvp: bp already attached! %p", bp));
1574 
1575 	vhold(vp);
1576 	if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT)
1577 		bp->b_flags |= B_NEEDSGIANT;
1578 	bp->b_vp = vp;
1579 	bp->b_bufobj = bo;
1580 	/*
1581 	 * Insert onto list for new vnode.
1582 	 */
1583 	buf_vlist_add(bp, bo, BX_VNCLEAN);
1584 }
1585 
1586 /*
1587  * Disassociate a buffer from a vnode.
1588  */
1589 void
1590 brelvp(struct buf *bp)
1591 {
1592 	struct bufobj *bo;
1593 	struct vnode *vp;
1594 
1595 	CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1596 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1597 
1598 	/*
1599 	 * Delete from old vnode list, if on one.
1600 	 */
1601 	vp = bp->b_vp;		/* XXX */
1602 	bo = bp->b_bufobj;
1603 	BO_LOCK(bo);
1604 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1605 		buf_vlist_remove(bp);
1606 	else
1607 		panic("brelvp: Buffer %p not on queue.", bp);
1608 	if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1609 		bo->bo_flag &= ~BO_ONWORKLST;
1610 		mtx_lock(&sync_mtx);
1611 		LIST_REMOVE(bo, bo_synclist);
1612 		syncer_worklist_len--;
1613 		mtx_unlock(&sync_mtx);
1614 	}
1615 	bp->b_flags &= ~B_NEEDSGIANT;
1616 	bp->b_vp = NULL;
1617 	bp->b_bufobj = NULL;
1618 	BO_UNLOCK(bo);
1619 	vdrop(vp);
1620 }
1621 
1622 /*
1623  * Add an item to the syncer work queue.
1624  */
1625 static void
1626 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1627 {
1628 	int queue, slot;
1629 
1630 	ASSERT_BO_LOCKED(bo);
1631 
1632 	mtx_lock(&sync_mtx);
1633 	if (bo->bo_flag & BO_ONWORKLST)
1634 		LIST_REMOVE(bo, bo_synclist);
1635 	else {
1636 		bo->bo_flag |= BO_ONWORKLST;
1637 		syncer_worklist_len++;
1638 	}
1639 
1640 	if (delay > syncer_maxdelay - 2)
1641 		delay = syncer_maxdelay - 2;
1642 	slot = (syncer_delayno + delay) & syncer_mask;
1643 
1644 	queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ :
1645 	    WI_MPSAFEQ;
1646 	LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo,
1647 	    bo_synclist);
1648 	mtx_unlock(&sync_mtx);
1649 }
1650 
1651 static int
1652 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1653 {
1654 	int error, len;
1655 
1656 	mtx_lock(&sync_mtx);
1657 	len = syncer_worklist_len - sync_vnode_count;
1658 	mtx_unlock(&sync_mtx);
1659 	error = SYSCTL_OUT(req, &len, sizeof(len));
1660 	return (error);
1661 }
1662 
1663 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1664     sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1665 
1666 static struct proc *updateproc;
1667 static void sched_sync(void);
1668 static struct kproc_desc up_kp = {
1669 	"syncer",
1670 	sched_sync,
1671 	&updateproc
1672 };
1673 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1674 
1675 static int
1676 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1677 {
1678 	struct vnode *vp;
1679 	struct mount *mp;
1680 
1681 	*bo = LIST_FIRST(slp);
1682 	if (*bo == NULL)
1683 		return (0);
1684 	vp = (*bo)->__bo_vnode;	/* XXX */
1685 	if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1686 		return (1);
1687 	/*
1688 	 * We use vhold in case the vnode does not
1689 	 * successfully sync.  vhold prevents the vnode from
1690 	 * going away when we unlock the sync_mtx so that
1691 	 * we can acquire the vnode interlock.
1692 	 */
1693 	vholdl(vp);
1694 	mtx_unlock(&sync_mtx);
1695 	VI_UNLOCK(vp);
1696 	if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1697 		vdrop(vp);
1698 		mtx_lock(&sync_mtx);
1699 		return (*bo == LIST_FIRST(slp));
1700 	}
1701 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1702 	(void) VOP_FSYNC(vp, MNT_LAZY, td);
1703 	VOP_UNLOCK(vp, 0);
1704 	vn_finished_write(mp);
1705 	BO_LOCK(*bo);
1706 	if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1707 		/*
1708 		 * Put us back on the worklist.  The worklist
1709 		 * routine will remove us from our current
1710 		 * position and then add us back in at a later
1711 		 * position.
1712 		 */
1713 		vn_syncer_add_to_worklist(*bo, syncdelay);
1714 	}
1715 	BO_UNLOCK(*bo);
1716 	vdrop(vp);
1717 	mtx_lock(&sync_mtx);
1718 	return (0);
1719 }
1720 
1721 /*
1722  * System filesystem synchronizer daemon.
1723  */
1724 static void
1725 sched_sync(void)
1726 {
1727 	struct synclist *gnext, *next;
1728 	struct synclist *gslp, *slp;
1729 	struct bufobj *bo;
1730 	long starttime;
1731 	struct thread *td = curthread;
1732 	int last_work_seen;
1733 	int net_worklist_len;
1734 	int syncer_final_iter;
1735 	int first_printf;
1736 	int error;
1737 
1738 	last_work_seen = 0;
1739 	syncer_final_iter = 0;
1740 	first_printf = 1;
1741 	syncer_state = SYNCER_RUNNING;
1742 	starttime = time_uptime;
1743 	td->td_pflags |= TDP_NORUNNINGBUF;
1744 
1745 	EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1746 	    SHUTDOWN_PRI_LAST);
1747 
1748 	mtx_lock(&sync_mtx);
1749 	for (;;) {
1750 		if (syncer_state == SYNCER_FINAL_DELAY &&
1751 		    syncer_final_iter == 0) {
1752 			mtx_unlock(&sync_mtx);
1753 			kproc_suspend_check(td->td_proc);
1754 			mtx_lock(&sync_mtx);
1755 		}
1756 		net_worklist_len = syncer_worklist_len - sync_vnode_count;
1757 		if (syncer_state != SYNCER_RUNNING &&
1758 		    starttime != time_uptime) {
1759 			if (first_printf) {
1760 				printf("\nSyncing disks, vnodes remaining...");
1761 				first_printf = 0;
1762 			}
1763 			printf("%d ", net_worklist_len);
1764 		}
1765 		starttime = time_uptime;
1766 
1767 		/*
1768 		 * Push files whose dirty time has expired.  Be careful
1769 		 * of interrupt race on slp queue.
1770 		 *
1771 		 * Skip over empty worklist slots when shutting down.
1772 		 */
1773 		do {
1774 			slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1775 			gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1776 			syncer_delayno += 1;
1777 			if (syncer_delayno == syncer_maxdelay)
1778 				syncer_delayno = 0;
1779 			next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1780 			gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1781 			/*
1782 			 * If the worklist has wrapped since the
1783 			 * it was emptied of all but syncer vnodes,
1784 			 * switch to the FINAL_DELAY state and run
1785 			 * for one more second.
1786 			 */
1787 			if (syncer_state == SYNCER_SHUTTING_DOWN &&
1788 			    net_worklist_len == 0 &&
1789 			    last_work_seen == syncer_delayno) {
1790 				syncer_state = SYNCER_FINAL_DELAY;
1791 				syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1792 			}
1793 		} while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1794 		    LIST_EMPTY(gslp) && syncer_worklist_len > 0);
1795 
1796 		/*
1797 		 * Keep track of the last time there was anything
1798 		 * on the worklist other than syncer vnodes.
1799 		 * Return to the SHUTTING_DOWN state if any
1800 		 * new work appears.
1801 		 */
1802 		if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1803 			last_work_seen = syncer_delayno;
1804 		if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1805 			syncer_state = SYNCER_SHUTTING_DOWN;
1806 		while (!LIST_EMPTY(slp)) {
1807 			error = sync_vnode(slp, &bo, td);
1808 			if (error == 1) {
1809 				LIST_REMOVE(bo, bo_synclist);
1810 				LIST_INSERT_HEAD(next, bo, bo_synclist);
1811 				continue;
1812 			}
1813 		}
1814 		if (!LIST_EMPTY(gslp)) {
1815 			mtx_unlock(&sync_mtx);
1816 			mtx_lock(&Giant);
1817 			mtx_lock(&sync_mtx);
1818 			while (!LIST_EMPTY(gslp)) {
1819 				error = sync_vnode(gslp, &bo, td);
1820 				if (error == 1) {
1821 					LIST_REMOVE(bo, bo_synclist);
1822 					LIST_INSERT_HEAD(gnext, bo,
1823 					    bo_synclist);
1824 					continue;
1825 				}
1826 			}
1827 			mtx_unlock(&Giant);
1828 		}
1829 		if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1830 			syncer_final_iter--;
1831 		/*
1832 		 * The variable rushjob allows the kernel to speed up the
1833 		 * processing of the filesystem syncer process. A rushjob
1834 		 * value of N tells the filesystem syncer to process the next
1835 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1836 		 * is used by the soft update code to speed up the filesystem
1837 		 * syncer process when the incore state is getting so far
1838 		 * ahead of the disk that the kernel memory pool is being
1839 		 * threatened with exhaustion.
1840 		 */
1841 		if (rushjob > 0) {
1842 			rushjob -= 1;
1843 			continue;
1844 		}
1845 		/*
1846 		 * Just sleep for a short period of time between
1847 		 * iterations when shutting down to allow some I/O
1848 		 * to happen.
1849 		 *
1850 		 * If it has taken us less than a second to process the
1851 		 * current work, then wait. Otherwise start right over
1852 		 * again. We can still lose time if any single round
1853 		 * takes more than two seconds, but it does not really
1854 		 * matter as we are just trying to generally pace the
1855 		 * filesystem activity.
1856 		 */
1857 		if (syncer_state != SYNCER_RUNNING)
1858 			cv_timedwait(&sync_wakeup, &sync_mtx,
1859 			    hz / SYNCER_SHUTDOWN_SPEEDUP);
1860 		else if (time_uptime == starttime)
1861 			cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1862 	}
1863 }
1864 
1865 /*
1866  * Request the syncer daemon to speed up its work.
1867  * We never push it to speed up more than half of its
1868  * normal turn time, otherwise it could take over the cpu.
1869  */
1870 int
1871 speedup_syncer(void)
1872 {
1873 	int ret = 0;
1874 
1875 	mtx_lock(&sync_mtx);
1876 	if (rushjob < syncdelay / 2) {
1877 		rushjob += 1;
1878 		stat_rush_requests += 1;
1879 		ret = 1;
1880 	}
1881 	mtx_unlock(&sync_mtx);
1882 	cv_broadcast(&sync_wakeup);
1883 	return (ret);
1884 }
1885 
1886 /*
1887  * Tell the syncer to speed up its work and run though its work
1888  * list several times, then tell it to shut down.
1889  */
1890 static void
1891 syncer_shutdown(void *arg, int howto)
1892 {
1893 
1894 	if (howto & RB_NOSYNC)
1895 		return;
1896 	mtx_lock(&sync_mtx);
1897 	syncer_state = SYNCER_SHUTTING_DOWN;
1898 	rushjob = 0;
1899 	mtx_unlock(&sync_mtx);
1900 	cv_broadcast(&sync_wakeup);
1901 	kproc_shutdown(arg, howto);
1902 }
1903 
1904 /*
1905  * Reassign a buffer from one vnode to another.
1906  * Used to assign file specific control information
1907  * (indirect blocks) to the vnode to which they belong.
1908  */
1909 void
1910 reassignbuf(struct buf *bp)
1911 {
1912 	struct vnode *vp;
1913 	struct bufobj *bo;
1914 	int delay;
1915 #ifdef INVARIANTS
1916 	struct bufv *bv;
1917 #endif
1918 
1919 	vp = bp->b_vp;
1920 	bo = bp->b_bufobj;
1921 	++reassignbufcalls;
1922 
1923 	CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1924 	    bp, bp->b_vp, bp->b_flags);
1925 	/*
1926 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1927 	 * is not fully linked in.
1928 	 */
1929 	if (bp->b_flags & B_PAGING)
1930 		panic("cannot reassign paging buffer");
1931 
1932 	/*
1933 	 * Delete from old vnode list, if on one.
1934 	 */
1935 	BO_LOCK(bo);
1936 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1937 		buf_vlist_remove(bp);
1938 	else
1939 		panic("reassignbuf: Buffer %p not on queue.", bp);
1940 	/*
1941 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1942 	 * of clean buffers.
1943 	 */
1944 	if (bp->b_flags & B_DELWRI) {
1945 		if ((bo->bo_flag & BO_ONWORKLST) == 0) {
1946 			switch (vp->v_type) {
1947 			case VDIR:
1948 				delay = dirdelay;
1949 				break;
1950 			case VCHR:
1951 				delay = metadelay;
1952 				break;
1953 			default:
1954 				delay = filedelay;
1955 			}
1956 			vn_syncer_add_to_worklist(bo, delay);
1957 		}
1958 		buf_vlist_add(bp, bo, BX_VNDIRTY);
1959 	} else {
1960 		buf_vlist_add(bp, bo, BX_VNCLEAN);
1961 
1962 		if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1963 			mtx_lock(&sync_mtx);
1964 			LIST_REMOVE(bo, bo_synclist);
1965 			syncer_worklist_len--;
1966 			mtx_unlock(&sync_mtx);
1967 			bo->bo_flag &= ~BO_ONWORKLST;
1968 		}
1969 	}
1970 #ifdef INVARIANTS
1971 	bv = &bo->bo_clean;
1972 	bp = TAILQ_FIRST(&bv->bv_hd);
1973 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1974 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1975 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
1976 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1977 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1978 	bv = &bo->bo_dirty;
1979 	bp = TAILQ_FIRST(&bv->bv_hd);
1980 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1981 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1982 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
1983 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1984 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1985 #endif
1986 	BO_UNLOCK(bo);
1987 }
1988 
1989 /*
1990  * Increment the use and hold counts on the vnode, taking care to reference
1991  * the driver's usecount if this is a chardev.  The vholdl() will remove
1992  * the vnode from the free list if it is presently free.  Requires the
1993  * vnode interlock and returns with it held.
1994  */
1995 static void
1996 v_incr_usecount(struct vnode *vp)
1997 {
1998 
1999 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2000 	vp->v_usecount++;
2001 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2002 		dev_lock();
2003 		vp->v_rdev->si_usecount++;
2004 		dev_unlock();
2005 	}
2006 	vholdl(vp);
2007 }
2008 
2009 /*
2010  * Turn a holdcnt into a use+holdcnt such that only one call to
2011  * v_decr_usecount is needed.
2012  */
2013 static void
2014 v_upgrade_usecount(struct vnode *vp)
2015 {
2016 
2017 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2018 	vp->v_usecount++;
2019 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2020 		dev_lock();
2021 		vp->v_rdev->si_usecount++;
2022 		dev_unlock();
2023 	}
2024 }
2025 
2026 /*
2027  * Decrement the vnode use and hold count along with the driver's usecount
2028  * if this is a chardev.  The vdropl() below releases the vnode interlock
2029  * as it may free the vnode.
2030  */
2031 static void
2032 v_decr_usecount(struct vnode *vp)
2033 {
2034 
2035 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2036 	VNASSERT(vp->v_usecount > 0, vp,
2037 	    ("v_decr_usecount: negative usecount"));
2038 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2039 	vp->v_usecount--;
2040 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2041 		dev_lock();
2042 		vp->v_rdev->si_usecount--;
2043 		dev_unlock();
2044 	}
2045 	vdropl(vp);
2046 }
2047 
2048 /*
2049  * Decrement only the use count and driver use count.  This is intended to
2050  * be paired with a follow on vdropl() to release the remaining hold count.
2051  * In this way we may vgone() a vnode with a 0 usecount without risk of
2052  * having it end up on a free list because the hold count is kept above 0.
2053  */
2054 static void
2055 v_decr_useonly(struct vnode *vp)
2056 {
2057 
2058 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2059 	VNASSERT(vp->v_usecount > 0, vp,
2060 	    ("v_decr_useonly: negative usecount"));
2061 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2062 	vp->v_usecount--;
2063 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2064 		dev_lock();
2065 		vp->v_rdev->si_usecount--;
2066 		dev_unlock();
2067 	}
2068 }
2069 
2070 /*
2071  * Grab a particular vnode from the free list, increment its
2072  * reference count and lock it.  VI_DOOMED is set if the vnode
2073  * is being destroyed.  Only callers who specify LK_RETRY will
2074  * see doomed vnodes.  If inactive processing was delayed in
2075  * vput try to do it here.
2076  */
2077 int
2078 vget(struct vnode *vp, int flags, struct thread *td)
2079 {
2080 	int error;
2081 
2082 	error = 0;
2083 	VFS_ASSERT_GIANT(vp->v_mount);
2084 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2085 	    ("vget: invalid lock operation"));
2086 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2087 
2088 	if ((flags & LK_INTERLOCK) == 0)
2089 		VI_LOCK(vp);
2090 	vholdl(vp);
2091 	if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2092 		vdrop(vp);
2093 		CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2094 		    vp);
2095 		return (error);
2096 	}
2097 	if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2098 		panic("vget: vn_lock failed to return ENOENT\n");
2099 	VI_LOCK(vp);
2100 	/* Upgrade our holdcnt to a usecount. */
2101 	v_upgrade_usecount(vp);
2102 	/*
2103  	 * We don't guarantee that any particular close will
2104 	 * trigger inactive processing so just make a best effort
2105 	 * here at preventing a reference to a removed file.  If
2106 	 * we don't succeed no harm is done.
2107 	 */
2108 	if (vp->v_iflag & VI_OWEINACT) {
2109 		if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2110 		    (flags & LK_NOWAIT) == 0)
2111 			vinactive(vp, td);
2112 		vp->v_iflag &= ~VI_OWEINACT;
2113 	}
2114 	VI_UNLOCK(vp);
2115 	return (0);
2116 }
2117 
2118 /*
2119  * Increase the reference count of a vnode.
2120  */
2121 void
2122 vref(struct vnode *vp)
2123 {
2124 
2125 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2126 	VI_LOCK(vp);
2127 	v_incr_usecount(vp);
2128 	VI_UNLOCK(vp);
2129 }
2130 
2131 /*
2132  * Return reference count of a vnode.
2133  *
2134  * The results of this call are only guaranteed when some mechanism other
2135  * than the VI lock is used to stop other processes from gaining references
2136  * to the vnode.  This may be the case if the caller holds the only reference.
2137  * This is also useful when stale data is acceptable as race conditions may
2138  * be accounted for by some other means.
2139  */
2140 int
2141 vrefcnt(struct vnode *vp)
2142 {
2143 	int usecnt;
2144 
2145 	VI_LOCK(vp);
2146 	usecnt = vp->v_usecount;
2147 	VI_UNLOCK(vp);
2148 
2149 	return (usecnt);
2150 }
2151 
2152 
2153 /*
2154  * Vnode put/release.
2155  * If count drops to zero, call inactive routine and return to freelist.
2156  */
2157 void
2158 vrele(struct vnode *vp)
2159 {
2160 	struct thread *td = curthread;	/* XXX */
2161 
2162 	KASSERT(vp != NULL, ("vrele: null vp"));
2163 	VFS_ASSERT_GIANT(vp->v_mount);
2164 
2165 	VI_LOCK(vp);
2166 
2167 	/* Skip this v_writecount check if we're going to panic below. */
2168 	VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2169 	    ("vrele: missed vn_close"));
2170 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2171 
2172 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2173 	    vp->v_usecount == 1)) {
2174 		v_decr_usecount(vp);
2175 		return;
2176 	}
2177 	if (vp->v_usecount != 1) {
2178 #ifdef DIAGNOSTIC
2179 		vprint("vrele: negative ref count", vp);
2180 #endif
2181 		VI_UNLOCK(vp);
2182 		panic("vrele: negative ref cnt");
2183 	}
2184 	CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2185 	/*
2186 	 * We want to hold the vnode until the inactive finishes to
2187 	 * prevent vgone() races.  We drop the use count here and the
2188 	 * hold count below when we're done.
2189 	 */
2190 	v_decr_useonly(vp);
2191 	/*
2192 	 * We must call VOP_INACTIVE with the node locked. Mark
2193 	 * as VI_DOINGINACT to avoid recursion.
2194 	 */
2195 	vp->v_iflag |= VI_OWEINACT;
2196 	if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK) == 0) {
2197 		VI_LOCK(vp);
2198 		if (vp->v_usecount > 0)
2199 			vp->v_iflag &= ~VI_OWEINACT;
2200 		if (vp->v_iflag & VI_OWEINACT)
2201 			vinactive(vp, td);
2202 		VOP_UNLOCK(vp, 0);
2203 	} else {
2204 		VI_LOCK(vp);
2205 		if (vp->v_usecount > 0)
2206 			vp->v_iflag &= ~VI_OWEINACT;
2207 	}
2208 	vdropl(vp);
2209 }
2210 
2211 /*
2212  * Release an already locked vnode.  This give the same effects as
2213  * unlock+vrele(), but takes less time and avoids releasing and
2214  * re-aquiring the lock (as vrele() acquires the lock internally.)
2215  */
2216 void
2217 vput(struct vnode *vp)
2218 {
2219 	struct thread *td = curthread;	/* XXX */
2220 	int error;
2221 
2222 	KASSERT(vp != NULL, ("vput: null vp"));
2223 	ASSERT_VOP_LOCKED(vp, "vput");
2224 	VFS_ASSERT_GIANT(vp->v_mount);
2225 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2226 	VI_LOCK(vp);
2227 	/* Skip this v_writecount check if we're going to panic below. */
2228 	VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2229 	    ("vput: missed vn_close"));
2230 	error = 0;
2231 
2232 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2233 	    vp->v_usecount == 1)) {
2234 		VOP_UNLOCK(vp, 0);
2235 		v_decr_usecount(vp);
2236 		return;
2237 	}
2238 
2239 	if (vp->v_usecount != 1) {
2240 #ifdef DIAGNOSTIC
2241 		vprint("vput: negative ref count", vp);
2242 #endif
2243 		panic("vput: negative ref cnt");
2244 	}
2245 	CTR2(KTR_VFS, "%s: return to freelist the vnode %p", __func__, vp);
2246 	/*
2247 	 * We want to hold the vnode until the inactive finishes to
2248 	 * prevent vgone() races.  We drop the use count here and the
2249 	 * hold count below when we're done.
2250 	 */
2251 	v_decr_useonly(vp);
2252 	vp->v_iflag |= VI_OWEINACT;
2253 	if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2254 		error = VOP_LOCK(vp, LK_UPGRADE|LK_INTERLOCK|LK_NOWAIT);
2255 		VI_LOCK(vp);
2256 		if (error) {
2257 			if (vp->v_usecount > 0)
2258 				vp->v_iflag &= ~VI_OWEINACT;
2259 			goto done;
2260 		}
2261 	}
2262 	if (vp->v_usecount > 0)
2263 		vp->v_iflag &= ~VI_OWEINACT;
2264 	if (vp->v_iflag & VI_OWEINACT)
2265 		vinactive(vp, td);
2266 	VOP_UNLOCK(vp, 0);
2267 done:
2268 	vdropl(vp);
2269 }
2270 
2271 /*
2272  * Somebody doesn't want the vnode recycled.
2273  */
2274 void
2275 vhold(struct vnode *vp)
2276 {
2277 
2278 	VI_LOCK(vp);
2279 	vholdl(vp);
2280 	VI_UNLOCK(vp);
2281 }
2282 
2283 void
2284 vholdl(struct vnode *vp)
2285 {
2286 
2287 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2288 	vp->v_holdcnt++;
2289 	if (VSHOULDBUSY(vp))
2290 		vbusy(vp);
2291 }
2292 
2293 /*
2294  * Note that there is one less who cares about this vnode.  vdrop() is the
2295  * opposite of vhold().
2296  */
2297 void
2298 vdrop(struct vnode *vp)
2299 {
2300 
2301 	VI_LOCK(vp);
2302 	vdropl(vp);
2303 }
2304 
2305 /*
2306  * Drop the hold count of the vnode.  If this is the last reference to
2307  * the vnode we will free it if it has been vgone'd otherwise it is
2308  * placed on the free list.
2309  */
2310 void
2311 vdropl(struct vnode *vp)
2312 {
2313 
2314 	ASSERT_VI_LOCKED(vp, "vdropl");
2315 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2316 	if (vp->v_holdcnt <= 0)
2317 		panic("vdrop: holdcnt %d", vp->v_holdcnt);
2318 	vp->v_holdcnt--;
2319 	if (vp->v_holdcnt == 0) {
2320 		if (vp->v_iflag & VI_DOOMED) {
2321 			CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__,
2322 			    vp);
2323 			vdestroy(vp);
2324 			return;
2325 		} else
2326 			vfree(vp);
2327 	}
2328 	VI_UNLOCK(vp);
2329 }
2330 
2331 /*
2332  * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2333  * flags.  DOINGINACT prevents us from recursing in calls to vinactive.
2334  * OWEINACT tracks whether a vnode missed a call to inactive due to a
2335  * failed lock upgrade.
2336  */
2337 static void
2338 vinactive(struct vnode *vp, struct thread *td)
2339 {
2340 
2341 	ASSERT_VOP_ELOCKED(vp, "vinactive");
2342 	ASSERT_VI_LOCKED(vp, "vinactive");
2343 	VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2344 	    ("vinactive: recursed on VI_DOINGINACT"));
2345 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2346 	vp->v_iflag |= VI_DOINGINACT;
2347 	vp->v_iflag &= ~VI_OWEINACT;
2348 	VI_UNLOCK(vp);
2349 	VOP_INACTIVE(vp, td);
2350 	VI_LOCK(vp);
2351 	VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2352 	    ("vinactive: lost VI_DOINGINACT"));
2353 	vp->v_iflag &= ~VI_DOINGINACT;
2354 }
2355 
2356 /*
2357  * Remove any vnodes in the vnode table belonging to mount point mp.
2358  *
2359  * If FORCECLOSE is not specified, there should not be any active ones,
2360  * return error if any are found (nb: this is a user error, not a
2361  * system error). If FORCECLOSE is specified, detach any active vnodes
2362  * that are found.
2363  *
2364  * If WRITECLOSE is set, only flush out regular file vnodes open for
2365  * writing.
2366  *
2367  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2368  *
2369  * `rootrefs' specifies the base reference count for the root vnode
2370  * of this filesystem. The root vnode is considered busy if its
2371  * v_usecount exceeds this value. On a successful return, vflush(, td)
2372  * will call vrele() on the root vnode exactly rootrefs times.
2373  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2374  * be zero.
2375  */
2376 #ifdef DIAGNOSTIC
2377 static int busyprt = 0;		/* print out busy vnodes */
2378 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
2379 #endif
2380 
2381 int
2382 vflush( struct mount *mp, int rootrefs, int flags, struct thread *td)
2383 {
2384 	struct vnode *vp, *mvp, *rootvp = NULL;
2385 	struct vattr vattr;
2386 	int busy = 0, error;
2387 
2388 	CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2389 	    rootrefs, flags);
2390 	if (rootrefs > 0) {
2391 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2392 		    ("vflush: bad args"));
2393 		/*
2394 		 * Get the filesystem root vnode. We can vput() it
2395 		 * immediately, since with rootrefs > 0, it won't go away.
2396 		 */
2397 		if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2398 			CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2399 			    __func__, error);
2400 			return (error);
2401 		}
2402 		vput(rootvp);
2403 
2404 	}
2405 	MNT_ILOCK(mp);
2406 loop:
2407 	MNT_VNODE_FOREACH(vp, mp, mvp) {
2408 
2409 		VI_LOCK(vp);
2410 		vholdl(vp);
2411 		MNT_IUNLOCK(mp);
2412 		error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2413 		if (error) {
2414 			vdrop(vp);
2415 			MNT_ILOCK(mp);
2416 			MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
2417 			goto loop;
2418 		}
2419 		/*
2420 		 * Skip over a vnodes marked VV_SYSTEM.
2421 		 */
2422 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2423 			VOP_UNLOCK(vp, 0);
2424 			vdrop(vp);
2425 			MNT_ILOCK(mp);
2426 			continue;
2427 		}
2428 		/*
2429 		 * If WRITECLOSE is set, flush out unlinked but still open
2430 		 * files (even if open only for reading) and regular file
2431 		 * vnodes open for writing.
2432 		 */
2433 		if (flags & WRITECLOSE) {
2434 			error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2435 			VI_LOCK(vp);
2436 
2437 			if ((vp->v_type == VNON ||
2438 			    (error == 0 && vattr.va_nlink > 0)) &&
2439 			    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2440 				VOP_UNLOCK(vp, 0);
2441 				vdropl(vp);
2442 				MNT_ILOCK(mp);
2443 				continue;
2444 			}
2445 		} else
2446 			VI_LOCK(vp);
2447 		/*
2448 		 * With v_usecount == 0, all we need to do is clear out the
2449 		 * vnode data structures and we are done.
2450 		 *
2451 		 * If FORCECLOSE is set, forcibly close the vnode.
2452 		 */
2453 		if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2454 			VNASSERT(vp->v_usecount == 0 ||
2455 			    (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2456 			    ("device VNODE %p is FORCECLOSED", vp));
2457 			vgonel(vp);
2458 		} else {
2459 			busy++;
2460 #ifdef DIAGNOSTIC
2461 			if (busyprt)
2462 				vprint("vflush: busy vnode", vp);
2463 #endif
2464 		}
2465 		VOP_UNLOCK(vp, 0);
2466 		vdropl(vp);
2467 		MNT_ILOCK(mp);
2468 	}
2469 	MNT_IUNLOCK(mp);
2470 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2471 		/*
2472 		 * If just the root vnode is busy, and if its refcount
2473 		 * is equal to `rootrefs', then go ahead and kill it.
2474 		 */
2475 		VI_LOCK(rootvp);
2476 		KASSERT(busy > 0, ("vflush: not busy"));
2477 		VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2478 		    ("vflush: usecount %d < rootrefs %d",
2479 		     rootvp->v_usecount, rootrefs));
2480 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2481 			VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2482 			vgone(rootvp);
2483 			VOP_UNLOCK(rootvp, 0);
2484 			busy = 0;
2485 		} else
2486 			VI_UNLOCK(rootvp);
2487 	}
2488 	if (busy) {
2489 		CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2490 		    busy);
2491 		return (EBUSY);
2492 	}
2493 	for (; rootrefs > 0; rootrefs--)
2494 		vrele(rootvp);
2495 	return (0);
2496 }
2497 
2498 /*
2499  * Recycle an unused vnode to the front of the free list.
2500  */
2501 int
2502 vrecycle(struct vnode *vp, struct thread *td)
2503 {
2504 	int recycled;
2505 
2506 	ASSERT_VOP_ELOCKED(vp, "vrecycle");
2507 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2508 	recycled = 0;
2509 	VI_LOCK(vp);
2510 	if (vp->v_usecount == 0) {
2511 		recycled = 1;
2512 		vgonel(vp);
2513 	}
2514 	VI_UNLOCK(vp);
2515 	return (recycled);
2516 }
2517 
2518 /*
2519  * Eliminate all activity associated with a vnode
2520  * in preparation for reuse.
2521  */
2522 void
2523 vgone(struct vnode *vp)
2524 {
2525 	VI_LOCK(vp);
2526 	vgonel(vp);
2527 	VI_UNLOCK(vp);
2528 }
2529 
2530 /*
2531  * vgone, with the vp interlock held.
2532  */
2533 void
2534 vgonel(struct vnode *vp)
2535 {
2536 	struct thread *td;
2537 	int oweinact;
2538 	int active;
2539 	struct mount *mp;
2540 
2541 	ASSERT_VOP_ELOCKED(vp, "vgonel");
2542 	ASSERT_VI_LOCKED(vp, "vgonel");
2543 	VNASSERT(vp->v_holdcnt, vp,
2544 	    ("vgonel: vp %p has no reference.", vp));
2545 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2546 	td = curthread;
2547 
2548 	/*
2549 	 * Don't vgonel if we're already doomed.
2550 	 */
2551 	if (vp->v_iflag & VI_DOOMED)
2552 		return;
2553 	vp->v_iflag |= VI_DOOMED;
2554 	/*
2555 	 * Check to see if the vnode is in use.  If so, we have to call
2556 	 * VOP_CLOSE() and VOP_INACTIVE().
2557 	 */
2558 	active = vp->v_usecount;
2559 	oweinact = (vp->v_iflag & VI_OWEINACT);
2560 	VI_UNLOCK(vp);
2561 	/*
2562 	 * Clean out any buffers associated with the vnode.
2563 	 * If the flush fails, just toss the buffers.
2564 	 */
2565 	mp = NULL;
2566 	if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2567 		(void) vn_start_secondary_write(vp, &mp, V_WAIT);
2568 	if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
2569 		vinvalbuf(vp, 0, 0, 0);
2570 
2571 	/*
2572 	 * If purging an active vnode, it must be closed and
2573 	 * deactivated before being reclaimed.
2574 	 */
2575 	if (active)
2576 		VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2577 	if (oweinact || active) {
2578 		VI_LOCK(vp);
2579 		if ((vp->v_iflag & VI_DOINGINACT) == 0)
2580 			vinactive(vp, td);
2581 		VI_UNLOCK(vp);
2582 	}
2583 	/*
2584 	 * Reclaim the vnode.
2585 	 */
2586 	if (VOP_RECLAIM(vp, td))
2587 		panic("vgone: cannot reclaim");
2588 	if (mp != NULL)
2589 		vn_finished_secondary_write(mp);
2590 	VNASSERT(vp->v_object == NULL, vp,
2591 	    ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2592 	/*
2593 	 * Clear the advisory locks and wake up waiting threads.
2594 	 */
2595 	lf_purgelocks(vp, &(vp->v_lockf));
2596 	/*
2597 	 * Delete from old mount point vnode list.
2598 	 */
2599 	delmntque(vp);
2600 	cache_purge(vp);
2601 	/*
2602 	 * Done with purge, reset to the standard lock and invalidate
2603 	 * the vnode.
2604 	 */
2605 	VI_LOCK(vp);
2606 	vp->v_vnlock = &vp->v_lock;
2607 	vp->v_op = &dead_vnodeops;
2608 	vp->v_tag = "none";
2609 	vp->v_type = VBAD;
2610 }
2611 
2612 /*
2613  * Calculate the total number of references to a special device.
2614  */
2615 int
2616 vcount(struct vnode *vp)
2617 {
2618 	int count;
2619 
2620 	dev_lock();
2621 	count = vp->v_rdev->si_usecount;
2622 	dev_unlock();
2623 	return (count);
2624 }
2625 
2626 /*
2627  * Same as above, but using the struct cdev *as argument
2628  */
2629 int
2630 count_dev(struct cdev *dev)
2631 {
2632 	int count;
2633 
2634 	dev_lock();
2635 	count = dev->si_usecount;
2636 	dev_unlock();
2637 	return(count);
2638 }
2639 
2640 /*
2641  * Print out a description of a vnode.
2642  */
2643 static char *typename[] =
2644 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2645  "VMARKER"};
2646 
2647 void
2648 vn_printf(struct vnode *vp, const char *fmt, ...)
2649 {
2650 	va_list ap;
2651 	char buf[256], buf2[16];
2652 	u_long flags;
2653 
2654 	va_start(ap, fmt);
2655 	vprintf(fmt, ap);
2656 	va_end(ap);
2657 	printf("%p: ", (void *)vp);
2658 	printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2659 	printf("    usecount %d, writecount %d, refcount %d mountedhere %p\n",
2660 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2661 	buf[0] = '\0';
2662 	buf[1] = '\0';
2663 	if (vp->v_vflag & VV_ROOT)
2664 		strlcat(buf, "|VV_ROOT", sizeof(buf));
2665 	if (vp->v_vflag & VV_ISTTY)
2666 		strlcat(buf, "|VV_ISTTY", sizeof(buf));
2667 	if (vp->v_vflag & VV_NOSYNC)
2668 		strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2669 	if (vp->v_vflag & VV_CACHEDLABEL)
2670 		strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2671 	if (vp->v_vflag & VV_TEXT)
2672 		strlcat(buf, "|VV_TEXT", sizeof(buf));
2673 	if (vp->v_vflag & VV_COPYONWRITE)
2674 		strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2675 	if (vp->v_vflag & VV_SYSTEM)
2676 		strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2677 	if (vp->v_vflag & VV_PROCDEP)
2678 		strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2679 	if (vp->v_vflag & VV_NOKNOTE)
2680 		strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2681 	if (vp->v_vflag & VV_DELETED)
2682 		strlcat(buf, "|VV_DELETED", sizeof(buf));
2683 	if (vp->v_vflag & VV_MD)
2684 		strlcat(buf, "|VV_MD", sizeof(buf));
2685 	flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC |
2686 	    VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2687 	    VV_NOKNOTE | VV_DELETED | VV_MD);
2688 	if (flags != 0) {
2689 		snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2690 		strlcat(buf, buf2, sizeof(buf));
2691 	}
2692 	if (vp->v_iflag & VI_MOUNT)
2693 		strlcat(buf, "|VI_MOUNT", sizeof(buf));
2694 	if (vp->v_iflag & VI_AGE)
2695 		strlcat(buf, "|VI_AGE", sizeof(buf));
2696 	if (vp->v_iflag & VI_DOOMED)
2697 		strlcat(buf, "|VI_DOOMED", sizeof(buf));
2698 	if (vp->v_iflag & VI_FREE)
2699 		strlcat(buf, "|VI_FREE", sizeof(buf));
2700 	if (vp->v_iflag & VI_DOINGINACT)
2701 		strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2702 	if (vp->v_iflag & VI_OWEINACT)
2703 		strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2704 	flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2705 	    VI_DOINGINACT | VI_OWEINACT);
2706 	if (flags != 0) {
2707 		snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2708 		strlcat(buf, buf2, sizeof(buf));
2709 	}
2710 	printf("    flags (%s)\n", buf + 1);
2711 	if (mtx_owned(VI_MTX(vp)))
2712 		printf(" VI_LOCKed");
2713 	if (vp->v_object != NULL)
2714 		printf("    v_object %p ref %d pages %d\n",
2715 		    vp->v_object, vp->v_object->ref_count,
2716 		    vp->v_object->resident_page_count);
2717 	printf("    ");
2718 	lockmgr_printinfo(vp->v_vnlock);
2719 	if (vp->v_data != NULL)
2720 		VOP_PRINT(vp);
2721 }
2722 
2723 #ifdef DDB
2724 /*
2725  * List all of the locked vnodes in the system.
2726  * Called when debugging the kernel.
2727  */
2728 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2729 {
2730 	struct mount *mp, *nmp;
2731 	struct vnode *vp;
2732 
2733 	/*
2734 	 * Note: because this is DDB, we can't obey the locking semantics
2735 	 * for these structures, which means we could catch an inconsistent
2736 	 * state and dereference a nasty pointer.  Not much to be done
2737 	 * about that.
2738 	 */
2739 	db_printf("Locked vnodes\n");
2740 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2741 		nmp = TAILQ_NEXT(mp, mnt_list);
2742 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2743 			if (vp->v_type != VMARKER &&
2744 			    VOP_ISLOCKED(vp))
2745 				vprint("", vp);
2746 		}
2747 		nmp = TAILQ_NEXT(mp, mnt_list);
2748 	}
2749 }
2750 
2751 /*
2752  * Show details about the given vnode.
2753  */
2754 DB_SHOW_COMMAND(vnode, db_show_vnode)
2755 {
2756 	struct vnode *vp;
2757 
2758 	if (!have_addr)
2759 		return;
2760 	vp = (struct vnode *)addr;
2761 	vn_printf(vp, "vnode ");
2762 }
2763 
2764 /*
2765  * Show details about the given mount point.
2766  */
2767 DB_SHOW_COMMAND(mount, db_show_mount)
2768 {
2769 	struct mount *mp;
2770 	struct vfsopt *opt;
2771 	struct statfs *sp;
2772 	struct vnode *vp;
2773 	char buf[512];
2774 	u_int flags;
2775 
2776 	if (!have_addr) {
2777 		/* No address given, print short info about all mount points. */
2778 		TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2779 			db_printf("%p %s on %s (%s)\n", mp,
2780 			    mp->mnt_stat.f_mntfromname,
2781 			    mp->mnt_stat.f_mntonname,
2782 			    mp->mnt_stat.f_fstypename);
2783 			if (db_pager_quit)
2784 				break;
2785 		}
2786 		db_printf("\nMore info: show mount <addr>\n");
2787 		return;
2788 	}
2789 
2790 	mp = (struct mount *)addr;
2791 	db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
2792 	    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
2793 
2794 	buf[0] = '\0';
2795 	flags = mp->mnt_flag;
2796 #define	MNT_FLAG(flag)	do {						\
2797 	if (flags & (flag)) {						\
2798 		if (buf[0] != '\0')					\
2799 			strlcat(buf, ", ", sizeof(buf));		\
2800 		strlcat(buf, (#flag) + 4, sizeof(buf));			\
2801 		flags &= ~(flag);					\
2802 	}								\
2803 } while (0)
2804 	MNT_FLAG(MNT_RDONLY);
2805 	MNT_FLAG(MNT_SYNCHRONOUS);
2806 	MNT_FLAG(MNT_NOEXEC);
2807 	MNT_FLAG(MNT_NOSUID);
2808 	MNT_FLAG(MNT_UNION);
2809 	MNT_FLAG(MNT_ASYNC);
2810 	MNT_FLAG(MNT_SUIDDIR);
2811 	MNT_FLAG(MNT_SOFTDEP);
2812 	MNT_FLAG(MNT_NOSYMFOLLOW);
2813 	MNT_FLAG(MNT_GJOURNAL);
2814 	MNT_FLAG(MNT_MULTILABEL);
2815 	MNT_FLAG(MNT_ACLS);
2816 	MNT_FLAG(MNT_NOATIME);
2817 	MNT_FLAG(MNT_NOCLUSTERR);
2818 	MNT_FLAG(MNT_NOCLUSTERW);
2819 	MNT_FLAG(MNT_EXRDONLY);
2820 	MNT_FLAG(MNT_EXPORTED);
2821 	MNT_FLAG(MNT_DEFEXPORTED);
2822 	MNT_FLAG(MNT_EXPORTANON);
2823 	MNT_FLAG(MNT_EXKERB);
2824 	MNT_FLAG(MNT_EXPUBLIC);
2825 	MNT_FLAG(MNT_LOCAL);
2826 	MNT_FLAG(MNT_QUOTA);
2827 	MNT_FLAG(MNT_ROOTFS);
2828 	MNT_FLAG(MNT_USER);
2829 	MNT_FLAG(MNT_IGNORE);
2830 	MNT_FLAG(MNT_UPDATE);
2831 	MNT_FLAG(MNT_DELEXPORT);
2832 	MNT_FLAG(MNT_RELOAD);
2833 	MNT_FLAG(MNT_FORCE);
2834 	MNT_FLAG(MNT_SNAPSHOT);
2835 	MNT_FLAG(MNT_BYFSID);
2836 #undef MNT_FLAG
2837 	if (flags != 0) {
2838 		if (buf[0] != '\0')
2839 			strlcat(buf, ", ", sizeof(buf));
2840 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2841 		    "0x%08x", flags);
2842 	}
2843 	db_printf("    mnt_flag = %s\n", buf);
2844 
2845 	buf[0] = '\0';
2846 	flags = mp->mnt_kern_flag;
2847 #define	MNT_KERN_FLAG(flag)	do {					\
2848 	if (flags & (flag)) {						\
2849 		if (buf[0] != '\0')					\
2850 			strlcat(buf, ", ", sizeof(buf));		\
2851 		strlcat(buf, (#flag) + 5, sizeof(buf));			\
2852 		flags &= ~(flag);					\
2853 	}								\
2854 } while (0)
2855 	MNT_KERN_FLAG(MNTK_UNMOUNTF);
2856 	MNT_KERN_FLAG(MNTK_ASYNC);
2857 	MNT_KERN_FLAG(MNTK_SOFTDEP);
2858 	MNT_KERN_FLAG(MNTK_NOINSMNTQ);
2859 	MNT_KERN_FLAG(MNTK_UNMOUNT);
2860 	MNT_KERN_FLAG(MNTK_MWAIT);
2861 	MNT_KERN_FLAG(MNTK_SUSPEND);
2862 	MNT_KERN_FLAG(MNTK_SUSPEND2);
2863 	MNT_KERN_FLAG(MNTK_SUSPENDED);
2864 	MNT_KERN_FLAG(MNTK_MPSAFE);
2865 	MNT_KERN_FLAG(MNTK_NOKNOTE);
2866 	MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
2867 #undef MNT_KERN_FLAG
2868 	if (flags != 0) {
2869 		if (buf[0] != '\0')
2870 			strlcat(buf, ", ", sizeof(buf));
2871 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2872 		    "0x%08x", flags);
2873 	}
2874 	db_printf("    mnt_kern_flag = %s\n", buf);
2875 
2876 	db_printf("    mnt_opt = ");
2877 	opt = TAILQ_FIRST(mp->mnt_opt);
2878 	if (opt != NULL) {
2879 		db_printf("%s", opt->name);
2880 		opt = TAILQ_NEXT(opt, link);
2881 		while (opt != NULL) {
2882 			db_printf(", %s", opt->name);
2883 			opt = TAILQ_NEXT(opt, link);
2884 		}
2885 	}
2886 	db_printf("\n");
2887 
2888 	sp = &mp->mnt_stat;
2889 	db_printf("    mnt_stat = { version=%u type=%u flags=0x%016jx "
2890 	    "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
2891 	    "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
2892 	    "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
2893 	    (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
2894 	    (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
2895 	    (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
2896 	    (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
2897 	    (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
2898 	    (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
2899 	    (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
2900 	    (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
2901 
2902 	db_printf("    mnt_cred = { uid=%u ruid=%u",
2903 	    (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
2904 	if (jailed(mp->mnt_cred))
2905 		db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
2906 	db_printf(" }\n");
2907 	db_printf("    mnt_ref = %d\n", mp->mnt_ref);
2908 	db_printf("    mnt_gen = %d\n", mp->mnt_gen);
2909 	db_printf("    mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
2910 	db_printf("    mnt_writeopcount = %d\n", mp->mnt_writeopcount);
2911 	db_printf("    mnt_noasync = %u\n", mp->mnt_noasync);
2912 	db_printf("    mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
2913 	db_printf("    mnt_iosize_max = %d\n", mp->mnt_iosize_max);
2914 	db_printf("    mnt_hashseed = %u\n", mp->mnt_hashseed);
2915 	db_printf("    mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
2916 	db_printf("    mnt_secondary_accwrites = %d\n",
2917 	    mp->mnt_secondary_accwrites);
2918 	db_printf("    mnt_gjprovider = %s\n",
2919 	    mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
2920 	db_printf("\n");
2921 
2922 	TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2923 		if (vp->v_type != VMARKER) {
2924 			vn_printf(vp, "vnode ");
2925 			if (db_pager_quit)
2926 				break;
2927 		}
2928 	}
2929 }
2930 #endif	/* DDB */
2931 
2932 /*
2933  * Fill in a struct xvfsconf based on a struct vfsconf.
2934  */
2935 static void
2936 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2937 {
2938 
2939 	strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2940 	xvfsp->vfc_typenum = vfsp->vfc_typenum;
2941 	xvfsp->vfc_refcount = vfsp->vfc_refcount;
2942 	xvfsp->vfc_flags = vfsp->vfc_flags;
2943 	/*
2944 	 * These are unused in userland, we keep them
2945 	 * to not break binary compatibility.
2946 	 */
2947 	xvfsp->vfc_vfsops = NULL;
2948 	xvfsp->vfc_next = NULL;
2949 }
2950 
2951 /*
2952  * Top level filesystem related information gathering.
2953  */
2954 static int
2955 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2956 {
2957 	struct vfsconf *vfsp;
2958 	struct xvfsconf xvfsp;
2959 	int error;
2960 
2961 	error = 0;
2962 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2963 		bzero(&xvfsp, sizeof(xvfsp));
2964 		vfsconf2x(vfsp, &xvfsp);
2965 		error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
2966 		if (error)
2967 			break;
2968 	}
2969 	return (error);
2970 }
2971 
2972 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
2973     "S,xvfsconf", "List of all configured filesystems");
2974 
2975 #ifndef BURN_BRIDGES
2976 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2977 
2978 static int
2979 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2980 {
2981 	int *name = (int *)arg1 - 1;	/* XXX */
2982 	u_int namelen = arg2 + 1;	/* XXX */
2983 	struct vfsconf *vfsp;
2984 	struct xvfsconf xvfsp;
2985 
2986 	printf("WARNING: userland calling deprecated sysctl, "
2987 	    "please rebuild world\n");
2988 
2989 #if 1 || defined(COMPAT_PRELITE2)
2990 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2991 	if (namelen == 1)
2992 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2993 #endif
2994 
2995 	switch (name[1]) {
2996 	case VFS_MAXTYPENUM:
2997 		if (namelen != 2)
2998 			return (ENOTDIR);
2999 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3000 	case VFS_CONF:
3001 		if (namelen != 3)
3002 			return (ENOTDIR);	/* overloaded */
3003 		TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
3004 			if (vfsp->vfc_typenum == name[2])
3005 				break;
3006 		if (vfsp == NULL)
3007 			return (EOPNOTSUPP);
3008 		bzero(&xvfsp, sizeof(xvfsp));
3009 		vfsconf2x(vfsp, &xvfsp);
3010 		return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3011 	}
3012 	return (EOPNOTSUPP);
3013 }
3014 
3015 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
3016 	vfs_sysctl, "Generic filesystem");
3017 
3018 #if 1 || defined(COMPAT_PRELITE2)
3019 
3020 static int
3021 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3022 {
3023 	int error;
3024 	struct vfsconf *vfsp;
3025 	struct ovfsconf ovfs;
3026 
3027 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3028 		bzero(&ovfs, sizeof(ovfs));
3029 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
3030 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
3031 		ovfs.vfc_index = vfsp->vfc_typenum;
3032 		ovfs.vfc_refcount = vfsp->vfc_refcount;
3033 		ovfs.vfc_flags = vfsp->vfc_flags;
3034 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3035 		if (error)
3036 			return error;
3037 	}
3038 	return 0;
3039 }
3040 
3041 #endif /* 1 || COMPAT_PRELITE2 */
3042 #endif /* !BURN_BRIDGES */
3043 
3044 #define KINFO_VNODESLOP		10
3045 #ifdef notyet
3046 /*
3047  * Dump vnode list (via sysctl).
3048  */
3049 /* ARGSUSED */
3050 static int
3051 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3052 {
3053 	struct xvnode *xvn;
3054 	struct mount *mp;
3055 	struct vnode *vp;
3056 	int error, len, n;
3057 
3058 	/*
3059 	 * Stale numvnodes access is not fatal here.
3060 	 */
3061 	req->lock = 0;
3062 	len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3063 	if (!req->oldptr)
3064 		/* Make an estimate */
3065 		return (SYSCTL_OUT(req, 0, len));
3066 
3067 	error = sysctl_wire_old_buffer(req, 0);
3068 	if (error != 0)
3069 		return (error);
3070 	xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3071 	n = 0;
3072 	mtx_lock(&mountlist_mtx);
3073 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3074 		if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3075 			continue;
3076 		MNT_ILOCK(mp);
3077 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3078 			if (n == len)
3079 				break;
3080 			vref(vp);
3081 			xvn[n].xv_size = sizeof *xvn;
3082 			xvn[n].xv_vnode = vp;
3083 			xvn[n].xv_id = 0;	/* XXX compat */
3084 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3085 			XV_COPY(usecount);
3086 			XV_COPY(writecount);
3087 			XV_COPY(holdcnt);
3088 			XV_COPY(mount);
3089 			XV_COPY(numoutput);
3090 			XV_COPY(type);
3091 #undef XV_COPY
3092 			xvn[n].xv_flag = vp->v_vflag;
3093 
3094 			switch (vp->v_type) {
3095 			case VREG:
3096 			case VDIR:
3097 			case VLNK:
3098 				break;
3099 			case VBLK:
3100 			case VCHR:
3101 				if (vp->v_rdev == NULL) {
3102 					vrele(vp);
3103 					continue;
3104 				}
3105 				xvn[n].xv_dev = dev2udev(vp->v_rdev);
3106 				break;
3107 			case VSOCK:
3108 				xvn[n].xv_socket = vp->v_socket;
3109 				break;
3110 			case VFIFO:
3111 				xvn[n].xv_fifo = vp->v_fifoinfo;
3112 				break;
3113 			case VNON:
3114 			case VBAD:
3115 			default:
3116 				/* shouldn't happen? */
3117 				vrele(vp);
3118 				continue;
3119 			}
3120 			vrele(vp);
3121 			++n;
3122 		}
3123 		MNT_IUNLOCK(mp);
3124 		mtx_lock(&mountlist_mtx);
3125 		vfs_unbusy(mp);
3126 		if (n == len)
3127 			break;
3128 	}
3129 	mtx_unlock(&mountlist_mtx);
3130 
3131 	error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3132 	free(xvn, M_TEMP);
3133 	return (error);
3134 }
3135 
3136 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3137 	0, 0, sysctl_vnode, "S,xvnode", "");
3138 #endif
3139 
3140 /*
3141  * Unmount all filesystems. The list is traversed in reverse order
3142  * of mounting to avoid dependencies.
3143  */
3144 void
3145 vfs_unmountall(void)
3146 {
3147 	struct mount *mp;
3148 	struct thread *td;
3149 	int error;
3150 
3151 	KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
3152 	CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3153 	td = curthread;
3154 
3155 	/*
3156 	 * Since this only runs when rebooting, it is not interlocked.
3157 	 */
3158 	while(!TAILQ_EMPTY(&mountlist)) {
3159 		mp = TAILQ_LAST(&mountlist, mntlist);
3160 		error = dounmount(mp, MNT_FORCE, td);
3161 		if (error) {
3162 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
3163 			/*
3164 			 * XXX: Due to the way in which we mount the root
3165 			 * file system off of devfs, devfs will generate a
3166 			 * "busy" warning when we try to unmount it before
3167 			 * the root.  Don't print a warning as a result in
3168 			 * order to avoid false positive errors that may
3169 			 * cause needless upset.
3170 			 */
3171 			if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3172 				printf("unmount of %s failed (",
3173 				    mp->mnt_stat.f_mntonname);
3174 				if (error == EBUSY)
3175 					printf("BUSY)\n");
3176 				else
3177 					printf("%d)\n", error);
3178 			}
3179 		} else {
3180 			/* The unmount has removed mp from the mountlist */
3181 		}
3182 	}
3183 }
3184 
3185 /*
3186  * perform msync on all vnodes under a mount point
3187  * the mount point must be locked.
3188  */
3189 void
3190 vfs_msync(struct mount *mp, int flags)
3191 {
3192 	struct vnode *vp, *mvp;
3193 	struct vm_object *obj;
3194 
3195 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3196 	MNT_ILOCK(mp);
3197 	MNT_VNODE_FOREACH(vp, mp, mvp) {
3198 		VI_LOCK(vp);
3199 		obj = vp->v_object;
3200 		if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3201 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3202 			MNT_IUNLOCK(mp);
3203 			if (!vget(vp,
3204 			    LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3205 			    curthread)) {
3206 				if (vp->v_vflag & VV_NOSYNC) {	/* unlinked */
3207 					vput(vp);
3208 					MNT_ILOCK(mp);
3209 					continue;
3210 				}
3211 
3212 				obj = vp->v_object;
3213 				if (obj != NULL) {
3214 					VM_OBJECT_LOCK(obj);
3215 					vm_object_page_clean(obj, 0, 0,
3216 					    flags == MNT_WAIT ?
3217 					    OBJPC_SYNC : OBJPC_NOSYNC);
3218 					VM_OBJECT_UNLOCK(obj);
3219 				}
3220 				vput(vp);
3221 			}
3222 			MNT_ILOCK(mp);
3223 		} else
3224 			VI_UNLOCK(vp);
3225 	}
3226 	MNT_IUNLOCK(mp);
3227 }
3228 
3229 /*
3230  * Mark a vnode as free, putting it up for recycling.
3231  */
3232 static void
3233 vfree(struct vnode *vp)
3234 {
3235 
3236 	ASSERT_VI_LOCKED(vp, "vfree");
3237 	mtx_lock(&vnode_free_list_mtx);
3238 	VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed."));
3239 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free"));
3240 	VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't"));
3241 	VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp,
3242 	    ("vfree: Freeing doomed vnode"));
3243 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3244 	if (vp->v_iflag & VI_AGE) {
3245 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3246 	} else {
3247 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3248 	}
3249 	freevnodes++;
3250 	vp->v_iflag &= ~VI_AGE;
3251 	vp->v_iflag |= VI_FREE;
3252 	mtx_unlock(&vnode_free_list_mtx);
3253 }
3254 
3255 /*
3256  * Opposite of vfree() - mark a vnode as in use.
3257  */
3258 static void
3259 vbusy(struct vnode *vp)
3260 {
3261 	ASSERT_VI_LOCKED(vp, "vbusy");
3262 	VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
3263 	VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed."));
3264 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3265 
3266 	mtx_lock(&vnode_free_list_mtx);
3267 	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3268 	freevnodes--;
3269 	vp->v_iflag &= ~(VI_FREE|VI_AGE);
3270 	mtx_unlock(&vnode_free_list_mtx);
3271 }
3272 
3273 static void
3274 destroy_vpollinfo(struct vpollinfo *vi)
3275 {
3276 	knlist_destroy(&vi->vpi_selinfo.si_note);
3277 	mtx_destroy(&vi->vpi_lock);
3278 	uma_zfree(vnodepoll_zone, vi);
3279 }
3280 
3281 /*
3282  * Initalize per-vnode helper structure to hold poll-related state.
3283  */
3284 void
3285 v_addpollinfo(struct vnode *vp)
3286 {
3287 	struct vpollinfo *vi;
3288 
3289 	if (vp->v_pollinfo != NULL)
3290 		return;
3291 	vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3292 	mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3293 	knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3294 	    vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3295 	VI_LOCK(vp);
3296 	if (vp->v_pollinfo != NULL) {
3297 		VI_UNLOCK(vp);
3298 		destroy_vpollinfo(vi);
3299 		return;
3300 	}
3301 	vp->v_pollinfo = vi;
3302 	VI_UNLOCK(vp);
3303 }
3304 
3305 /*
3306  * Record a process's interest in events which might happen to
3307  * a vnode.  Because poll uses the historic select-style interface
3308  * internally, this routine serves as both the ``check for any
3309  * pending events'' and the ``record my interest in future events''
3310  * functions.  (These are done together, while the lock is held,
3311  * to avoid race conditions.)
3312  */
3313 int
3314 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3315 {
3316 
3317 	v_addpollinfo(vp);
3318 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3319 	if (vp->v_pollinfo->vpi_revents & events) {
3320 		/*
3321 		 * This leaves events we are not interested
3322 		 * in available for the other process which
3323 		 * which presumably had requested them
3324 		 * (otherwise they would never have been
3325 		 * recorded).
3326 		 */
3327 		events &= vp->v_pollinfo->vpi_revents;
3328 		vp->v_pollinfo->vpi_revents &= ~events;
3329 
3330 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
3331 		return (events);
3332 	}
3333 	vp->v_pollinfo->vpi_events |= events;
3334 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3335 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3336 	return (0);
3337 }
3338 
3339 /*
3340  * Routine to create and manage a filesystem syncer vnode.
3341  */
3342 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
3343 static int	sync_fsync(struct  vop_fsync_args *);
3344 static int	sync_inactive(struct  vop_inactive_args *);
3345 static int	sync_reclaim(struct  vop_reclaim_args *);
3346 
3347 static struct vop_vector sync_vnodeops = {
3348 	.vop_bypass =	VOP_EOPNOTSUPP,
3349 	.vop_close =	sync_close,		/* close */
3350 	.vop_fsync =	sync_fsync,		/* fsync */
3351 	.vop_inactive =	sync_inactive,	/* inactive */
3352 	.vop_reclaim =	sync_reclaim,	/* reclaim */
3353 	.vop_lock1 =	vop_stdlock,	/* lock */
3354 	.vop_unlock =	vop_stdunlock,	/* unlock */
3355 	.vop_islocked =	vop_stdislocked,	/* islocked */
3356 };
3357 
3358 /*
3359  * Create a new filesystem syncer vnode for the specified mount point.
3360  */
3361 int
3362 vfs_allocate_syncvnode(struct mount *mp)
3363 {
3364 	struct vnode *vp;
3365 	struct bufobj *bo;
3366 	static long start, incr, next;
3367 	int error;
3368 
3369 	/* Allocate a new vnode */
3370 	if ((error = getnewvnode("syncer", mp, &sync_vnodeops, &vp)) != 0) {
3371 		mp->mnt_syncer = NULL;
3372 		return (error);
3373 	}
3374 	vp->v_type = VNON;
3375 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3376 	vp->v_vflag |= VV_FORCEINSMQ;
3377 	error = insmntque(vp, mp);
3378 	if (error != 0)
3379 		panic("vfs_allocate_syncvnode: insmntque failed");
3380 	vp->v_vflag &= ~VV_FORCEINSMQ;
3381 	VOP_UNLOCK(vp, 0);
3382 	/*
3383 	 * Place the vnode onto the syncer worklist. We attempt to
3384 	 * scatter them about on the list so that they will go off
3385 	 * at evenly distributed times even if all the filesystems
3386 	 * are mounted at once.
3387 	 */
3388 	next += incr;
3389 	if (next == 0 || next > syncer_maxdelay) {
3390 		start /= 2;
3391 		incr /= 2;
3392 		if (start == 0) {
3393 			start = syncer_maxdelay / 2;
3394 			incr = syncer_maxdelay;
3395 		}
3396 		next = start;
3397 	}
3398 	bo = &vp->v_bufobj;
3399 	BO_LOCK(bo);
3400 	vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3401 	/* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3402 	mtx_lock(&sync_mtx);
3403 	sync_vnode_count++;
3404 	mtx_unlock(&sync_mtx);
3405 	BO_UNLOCK(bo);
3406 	mp->mnt_syncer = vp;
3407 	return (0);
3408 }
3409 
3410 /*
3411  * Do a lazy sync of the filesystem.
3412  */
3413 static int
3414 sync_fsync(struct vop_fsync_args *ap)
3415 {
3416 	struct vnode *syncvp = ap->a_vp;
3417 	struct mount *mp = syncvp->v_mount;
3418 	int error;
3419 	struct bufobj *bo;
3420 
3421 	/*
3422 	 * We only need to do something if this is a lazy evaluation.
3423 	 */
3424 	if (ap->a_waitfor != MNT_LAZY)
3425 		return (0);
3426 
3427 	/*
3428 	 * Move ourselves to the back of the sync list.
3429 	 */
3430 	bo = &syncvp->v_bufobj;
3431 	BO_LOCK(bo);
3432 	vn_syncer_add_to_worklist(bo, syncdelay);
3433 	BO_UNLOCK(bo);
3434 
3435 	/*
3436 	 * Walk the list of vnodes pushing all that are dirty and
3437 	 * not already on the sync list.
3438 	 */
3439 	mtx_lock(&mountlist_mtx);
3440 	if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) {
3441 		mtx_unlock(&mountlist_mtx);
3442 		return (0);
3443 	}
3444 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3445 		vfs_unbusy(mp);
3446 		return (0);
3447 	}
3448 	MNT_ILOCK(mp);
3449 	mp->mnt_noasync++;
3450 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
3451 	MNT_IUNLOCK(mp);
3452 	vfs_msync(mp, MNT_NOWAIT);
3453 	error = VFS_SYNC(mp, MNT_LAZY);
3454 	MNT_ILOCK(mp);
3455 	mp->mnt_noasync--;
3456 	if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
3457 		mp->mnt_kern_flag |= MNTK_ASYNC;
3458 	MNT_IUNLOCK(mp);
3459 	vn_finished_write(mp);
3460 	vfs_unbusy(mp);
3461 	return (error);
3462 }
3463 
3464 /*
3465  * The syncer vnode is no referenced.
3466  */
3467 static int
3468 sync_inactive(struct vop_inactive_args *ap)
3469 {
3470 
3471 	vgone(ap->a_vp);
3472 	return (0);
3473 }
3474 
3475 /*
3476  * The syncer vnode is no longer needed and is being decommissioned.
3477  *
3478  * Modifications to the worklist must be protected by sync_mtx.
3479  */
3480 static int
3481 sync_reclaim(struct vop_reclaim_args *ap)
3482 {
3483 	struct vnode *vp = ap->a_vp;
3484 	struct bufobj *bo;
3485 
3486 	bo = &vp->v_bufobj;
3487 	BO_LOCK(bo);
3488 	vp->v_mount->mnt_syncer = NULL;
3489 	if (bo->bo_flag & BO_ONWORKLST) {
3490 		mtx_lock(&sync_mtx);
3491 		LIST_REMOVE(bo, bo_synclist);
3492 		syncer_worklist_len--;
3493 		sync_vnode_count--;
3494 		mtx_unlock(&sync_mtx);
3495 		bo->bo_flag &= ~BO_ONWORKLST;
3496 	}
3497 	BO_UNLOCK(bo);
3498 
3499 	return (0);
3500 }
3501 
3502 /*
3503  * Check if vnode represents a disk device
3504  */
3505 int
3506 vn_isdisk(struct vnode *vp, int *errp)
3507 {
3508 	int error;
3509 
3510 	error = 0;
3511 	dev_lock();
3512 	if (vp->v_type != VCHR)
3513 		error = ENOTBLK;
3514 	else if (vp->v_rdev == NULL)
3515 		error = ENXIO;
3516 	else if (vp->v_rdev->si_devsw == NULL)
3517 		error = ENXIO;
3518 	else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3519 		error = ENOTBLK;
3520 	dev_unlock();
3521 	if (errp != NULL)
3522 		*errp = error;
3523 	return (error == 0);
3524 }
3525 
3526 /*
3527  * Common filesystem object access control check routine.  Accepts a
3528  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3529  * and optional call-by-reference privused argument allowing vaccess()
3530  * to indicate to the caller whether privilege was used to satisfy the
3531  * request (obsoleted).  Returns 0 on success, or an errno on failure.
3532  *
3533  * The ifdef'd CAPABILITIES version is here for reference, but is not
3534  * actually used.
3535  */
3536 int
3537 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3538     accmode_t accmode, struct ucred *cred, int *privused)
3539 {
3540 	accmode_t dac_granted;
3541 	accmode_t priv_granted;
3542 
3543 	KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3544 	    ("invalid bit in accmode"));
3545 	KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
3546 	    	("VAPPEND without VWRITE"));
3547 
3548 	/*
3549 	 * Look for a normal, non-privileged way to access the file/directory
3550 	 * as requested.  If it exists, go with that.
3551 	 */
3552 
3553 	if (privused != NULL)
3554 		*privused = 0;
3555 
3556 	dac_granted = 0;
3557 
3558 	/* Check the owner. */
3559 	if (cred->cr_uid == file_uid) {
3560 		dac_granted |= VADMIN;
3561 		if (file_mode & S_IXUSR)
3562 			dac_granted |= VEXEC;
3563 		if (file_mode & S_IRUSR)
3564 			dac_granted |= VREAD;
3565 		if (file_mode & S_IWUSR)
3566 			dac_granted |= (VWRITE | VAPPEND);
3567 
3568 		if ((accmode & dac_granted) == accmode)
3569 			return (0);
3570 
3571 		goto privcheck;
3572 	}
3573 
3574 	/* Otherwise, check the groups (first match) */
3575 	if (groupmember(file_gid, cred)) {
3576 		if (file_mode & S_IXGRP)
3577 			dac_granted |= VEXEC;
3578 		if (file_mode & S_IRGRP)
3579 			dac_granted |= VREAD;
3580 		if (file_mode & S_IWGRP)
3581 			dac_granted |= (VWRITE | VAPPEND);
3582 
3583 		if ((accmode & dac_granted) == accmode)
3584 			return (0);
3585 
3586 		goto privcheck;
3587 	}
3588 
3589 	/* Otherwise, check everyone else. */
3590 	if (file_mode & S_IXOTH)
3591 		dac_granted |= VEXEC;
3592 	if (file_mode & S_IROTH)
3593 		dac_granted |= VREAD;
3594 	if (file_mode & S_IWOTH)
3595 		dac_granted |= (VWRITE | VAPPEND);
3596 	if ((accmode & dac_granted) == accmode)
3597 		return (0);
3598 
3599 privcheck:
3600 	/*
3601 	 * Build a privilege mask to determine if the set of privileges
3602 	 * satisfies the requirements when combined with the granted mask
3603 	 * from above.  For each privilege, if the privilege is required,
3604 	 * bitwise or the request type onto the priv_granted mask.
3605 	 */
3606 	priv_granted = 0;
3607 
3608 	if (type == VDIR) {
3609 		/*
3610 		 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3611 		 * requests, instead of PRIV_VFS_EXEC.
3612 		 */
3613 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3614 		    !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3615 			priv_granted |= VEXEC;
3616 	} else {
3617 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3618 		    !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3619 			priv_granted |= VEXEC;
3620 	}
3621 
3622 	if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
3623 	    !priv_check_cred(cred, PRIV_VFS_READ, 0))
3624 		priv_granted |= VREAD;
3625 
3626 	if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3627 	    !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3628 		priv_granted |= (VWRITE | VAPPEND);
3629 
3630 	if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3631 	    !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3632 		priv_granted |= VADMIN;
3633 
3634 	if ((accmode & (priv_granted | dac_granted)) == accmode) {
3635 		/* XXX audit: privilege used */
3636 		if (privused != NULL)
3637 			*privused = 1;
3638 		return (0);
3639 	}
3640 
3641 	return ((accmode & VADMIN) ? EPERM : EACCES);
3642 }
3643 
3644 /*
3645  * Credential check based on process requesting service, and per-attribute
3646  * permissions.
3647  */
3648 int
3649 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3650     struct thread *td, accmode_t accmode)
3651 {
3652 
3653 	/*
3654 	 * Kernel-invoked always succeeds.
3655 	 */
3656 	if (cred == NOCRED)
3657 		return (0);
3658 
3659 	/*
3660 	 * Do not allow privileged processes in jail to directly manipulate
3661 	 * system attributes.
3662 	 */
3663 	switch (attrnamespace) {
3664 	case EXTATTR_NAMESPACE_SYSTEM:
3665 		/* Potentially should be: return (EPERM); */
3666 		return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3667 	case EXTATTR_NAMESPACE_USER:
3668 		return (VOP_ACCESS(vp, accmode, cred, td));
3669 	default:
3670 		return (EPERM);
3671 	}
3672 }
3673 
3674 #ifdef DEBUG_VFS_LOCKS
3675 /*
3676  * This only exists to supress warnings from unlocked specfs accesses.  It is
3677  * no longer ok to have an unlocked VFS.
3678  */
3679 #define	IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL ||		\
3680 	(vp)->v_type == VCHR ||	(vp)->v_type == VBAD)
3681 
3682 int vfs_badlock_ddb = 1;	/* Drop into debugger on violation. */
3683 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, "");
3684 
3685 int vfs_badlock_mutex = 1;	/* Check for interlock across VOPs. */
3686 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, "");
3687 
3688 int vfs_badlock_print = 1;	/* Print lock violations. */
3689 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, "");
3690 
3691 #ifdef KDB
3692 int vfs_badlock_backtrace = 1;	/* Print backtrace at lock violations. */
3693 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, "");
3694 #endif
3695 
3696 static void
3697 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3698 {
3699 
3700 #ifdef KDB
3701 	if (vfs_badlock_backtrace)
3702 		kdb_backtrace();
3703 #endif
3704 	if (vfs_badlock_print)
3705 		printf("%s: %p %s\n", str, (void *)vp, msg);
3706 	if (vfs_badlock_ddb)
3707 		kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3708 }
3709 
3710 void
3711 assert_vi_locked(struct vnode *vp, const char *str)
3712 {
3713 
3714 	if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3715 		vfs_badlock("interlock is not locked but should be", str, vp);
3716 }
3717 
3718 void
3719 assert_vi_unlocked(struct vnode *vp, const char *str)
3720 {
3721 
3722 	if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3723 		vfs_badlock("interlock is locked but should not be", str, vp);
3724 }
3725 
3726 void
3727 assert_vop_locked(struct vnode *vp, const char *str)
3728 {
3729 
3730 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0)
3731 		vfs_badlock("is not locked but should be", str, vp);
3732 }
3733 
3734 void
3735 assert_vop_unlocked(struct vnode *vp, const char *str)
3736 {
3737 
3738 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
3739 		vfs_badlock("is locked but should not be", str, vp);
3740 }
3741 
3742 void
3743 assert_vop_elocked(struct vnode *vp, const char *str)
3744 {
3745 
3746 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
3747 		vfs_badlock("is not exclusive locked but should be", str, vp);
3748 }
3749 
3750 #if 0
3751 void
3752 assert_vop_elocked_other(struct vnode *vp, const char *str)
3753 {
3754 
3755 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
3756 		vfs_badlock("is not exclusive locked by another thread",
3757 		    str, vp);
3758 }
3759 
3760 void
3761 assert_vop_slocked(struct vnode *vp, const char *str)
3762 {
3763 
3764 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
3765 		vfs_badlock("is not locked shared but should be", str, vp);
3766 }
3767 #endif /* 0 */
3768 #endif /* DEBUG_VFS_LOCKS */
3769 
3770 void
3771 vop_rename_pre(void *ap)
3772 {
3773 	struct vop_rename_args *a = ap;
3774 
3775 #ifdef DEBUG_VFS_LOCKS
3776 	if (a->a_tvp)
3777 		ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
3778 	ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
3779 	ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
3780 	ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
3781 
3782 	/* Check the source (from). */
3783 	if (a->a_tdvp != a->a_fdvp && a->a_tvp != a->a_fdvp)
3784 		ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3785 	if (a->a_tvp != a->a_fvp)
3786 		ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
3787 
3788 	/* Check the target. */
3789 	if (a->a_tvp)
3790 		ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3791 	ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3792 #endif
3793 	if (a->a_tdvp != a->a_fdvp)
3794 		vhold(a->a_fdvp);
3795 	if (a->a_tvp != a->a_fvp)
3796 		vhold(a->a_fvp);
3797 	vhold(a->a_tdvp);
3798 	if (a->a_tvp)
3799 		vhold(a->a_tvp);
3800 }
3801 
3802 void
3803 vop_strategy_pre(void *ap)
3804 {
3805 #ifdef DEBUG_VFS_LOCKS
3806 	struct vop_strategy_args *a;
3807 	struct buf *bp;
3808 
3809 	a = ap;
3810 	bp = a->a_bp;
3811 
3812 	/*
3813 	 * Cluster ops lock their component buffers but not the IO container.
3814 	 */
3815 	if ((bp->b_flags & B_CLUSTER) != 0)
3816 		return;
3817 
3818 	if (!BUF_ISLOCKED(bp)) {
3819 		if (vfs_badlock_print)
3820 			printf(
3821 			    "VOP_STRATEGY: bp is not locked but should be\n");
3822 		if (vfs_badlock_ddb)
3823 			kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3824 	}
3825 #endif
3826 }
3827 
3828 void
3829 vop_lookup_pre(void *ap)
3830 {
3831 #ifdef DEBUG_VFS_LOCKS
3832 	struct vop_lookup_args *a;
3833 	struct vnode *dvp;
3834 
3835 	a = ap;
3836 	dvp = a->a_dvp;
3837 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3838 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3839 #endif
3840 }
3841 
3842 void
3843 vop_lookup_post(void *ap, int rc)
3844 {
3845 #ifdef DEBUG_VFS_LOCKS
3846 	struct vop_lookup_args *a;
3847 	struct vnode *dvp;
3848 	struct vnode *vp;
3849 
3850 	a = ap;
3851 	dvp = a->a_dvp;
3852 	vp = *(a->a_vpp);
3853 
3854 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3855 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3856 
3857 	if (!rc)
3858 		ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
3859 #endif
3860 }
3861 
3862 void
3863 vop_lock_pre(void *ap)
3864 {
3865 #ifdef DEBUG_VFS_LOCKS
3866 	struct vop_lock1_args *a = ap;
3867 
3868 	if ((a->a_flags & LK_INTERLOCK) == 0)
3869 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3870 	else
3871 		ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
3872 #endif
3873 }
3874 
3875 void
3876 vop_lock_post(void *ap, int rc)
3877 {
3878 #ifdef DEBUG_VFS_LOCKS
3879 	struct vop_lock1_args *a = ap;
3880 
3881 	ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3882 	if (rc == 0)
3883 		ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
3884 #endif
3885 }
3886 
3887 void
3888 vop_unlock_pre(void *ap)
3889 {
3890 #ifdef DEBUG_VFS_LOCKS
3891 	struct vop_unlock_args *a = ap;
3892 
3893 	if (a->a_flags & LK_INTERLOCK)
3894 		ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
3895 	ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
3896 #endif
3897 }
3898 
3899 void
3900 vop_unlock_post(void *ap, int rc)
3901 {
3902 #ifdef DEBUG_VFS_LOCKS
3903 	struct vop_unlock_args *a = ap;
3904 
3905 	if (a->a_flags & LK_INTERLOCK)
3906 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
3907 #endif
3908 }
3909 
3910 void
3911 vop_create_post(void *ap, int rc)
3912 {
3913 	struct vop_create_args *a = ap;
3914 
3915 	if (!rc)
3916 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3917 }
3918 
3919 void
3920 vop_link_post(void *ap, int rc)
3921 {
3922 	struct vop_link_args *a = ap;
3923 
3924 	if (!rc) {
3925 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
3926 		VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
3927 	}
3928 }
3929 
3930 void
3931 vop_mkdir_post(void *ap, int rc)
3932 {
3933 	struct vop_mkdir_args *a = ap;
3934 
3935 	if (!rc)
3936 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3937 }
3938 
3939 void
3940 vop_mknod_post(void *ap, int rc)
3941 {
3942 	struct vop_mknod_args *a = ap;
3943 
3944 	if (!rc)
3945 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3946 }
3947 
3948 void
3949 vop_remove_post(void *ap, int rc)
3950 {
3951 	struct vop_remove_args *a = ap;
3952 
3953 	if (!rc) {
3954 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3955 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3956 	}
3957 }
3958 
3959 void
3960 vop_rename_post(void *ap, int rc)
3961 {
3962 	struct vop_rename_args *a = ap;
3963 
3964 	if (!rc) {
3965 		VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
3966 		VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
3967 		VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
3968 		if (a->a_tvp)
3969 			VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
3970 	}
3971 	if (a->a_tdvp != a->a_fdvp)
3972 		vdrop(a->a_fdvp);
3973 	if (a->a_tvp != a->a_fvp)
3974 		vdrop(a->a_fvp);
3975 	vdrop(a->a_tdvp);
3976 	if (a->a_tvp)
3977 		vdrop(a->a_tvp);
3978 }
3979 
3980 void
3981 vop_rmdir_post(void *ap, int rc)
3982 {
3983 	struct vop_rmdir_args *a = ap;
3984 
3985 	if (!rc) {
3986 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3987 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3988 	}
3989 }
3990 
3991 void
3992 vop_setattr_post(void *ap, int rc)
3993 {
3994 	struct vop_setattr_args *a = ap;
3995 
3996 	if (!rc)
3997 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
3998 }
3999 
4000 void
4001 vop_symlink_post(void *ap, int rc)
4002 {
4003 	struct vop_symlink_args *a = ap;
4004 
4005 	if (!rc)
4006 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4007 }
4008 
4009 static struct knlist fs_knlist;
4010 
4011 static void
4012 vfs_event_init(void *arg)
4013 {
4014 	knlist_init_mtx(&fs_knlist, NULL);
4015 }
4016 /* XXX - correct order? */
4017 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4018 
4019 void
4020 vfs_event_signal(fsid_t *fsid, u_int32_t event, intptr_t data __unused)
4021 {
4022 
4023 	KNOTE_UNLOCKED(&fs_knlist, event);
4024 }
4025 
4026 static int	filt_fsattach(struct knote *kn);
4027 static void	filt_fsdetach(struct knote *kn);
4028 static int	filt_fsevent(struct knote *kn, long hint);
4029 
4030 struct filterops fs_filtops = {
4031 	.f_isfd = 0,
4032 	.f_attach = filt_fsattach,
4033 	.f_detach = filt_fsdetach,
4034 	.f_event = filt_fsevent
4035 };
4036 
4037 static int
4038 filt_fsattach(struct knote *kn)
4039 {
4040 
4041 	kn->kn_flags |= EV_CLEAR;
4042 	knlist_add(&fs_knlist, kn, 0);
4043 	return (0);
4044 }
4045 
4046 static void
4047 filt_fsdetach(struct knote *kn)
4048 {
4049 
4050 	knlist_remove(&fs_knlist, kn, 0);
4051 }
4052 
4053 static int
4054 filt_fsevent(struct knote *kn, long hint)
4055 {
4056 
4057 	kn->kn_fflags |= hint;
4058 	return (kn->kn_fflags != 0);
4059 }
4060 
4061 static int
4062 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4063 {
4064 	struct vfsidctl vc;
4065 	int error;
4066 	struct mount *mp;
4067 
4068 	error = SYSCTL_IN(req, &vc, sizeof(vc));
4069 	if (error)
4070 		return (error);
4071 	if (vc.vc_vers != VFS_CTL_VERS1)
4072 		return (EINVAL);
4073 	mp = vfs_getvfs(&vc.vc_fsid);
4074 	if (mp == NULL)
4075 		return (ENOENT);
4076 	/* ensure that a specific sysctl goes to the right filesystem. */
4077 	if (strcmp(vc.vc_fstypename, "*") != 0 &&
4078 	    strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4079 		vfs_rel(mp);
4080 		return (EINVAL);
4081 	}
4082 	VCTLTOREQ(&vc, req);
4083 	error = VFS_SYSCTL(mp, vc.vc_op, req);
4084 	vfs_rel(mp);
4085 	return (error);
4086 }
4087 
4088 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "",
4089     "Sysctl by fsid");
4090 
4091 /*
4092  * Function to initialize a va_filerev field sensibly.
4093  * XXX: Wouldn't a random number make a lot more sense ??
4094  */
4095 u_quad_t
4096 init_va_filerev(void)
4097 {
4098 	struct bintime bt;
4099 
4100 	getbinuptime(&bt);
4101 	return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4102 }
4103 
4104 static int	filt_vfsread(struct knote *kn, long hint);
4105 static int	filt_vfswrite(struct knote *kn, long hint);
4106 static int	filt_vfsvnode(struct knote *kn, long hint);
4107 static void	filt_vfsdetach(struct knote *kn);
4108 static struct filterops vfsread_filtops = {
4109 	.f_isfd = 1,
4110 	.f_detach = filt_vfsdetach,
4111 	.f_event = filt_vfsread
4112 };
4113 static struct filterops vfswrite_filtops = {
4114 	.f_isfd = 1,
4115 	.f_detach = filt_vfsdetach,
4116 	.f_event = filt_vfswrite
4117 };
4118 static struct filterops vfsvnode_filtops = {
4119 	.f_isfd = 1,
4120 	.f_detach = filt_vfsdetach,
4121 	.f_event = filt_vfsvnode
4122 };
4123 
4124 static void
4125 vfs_knllock(void *arg)
4126 {
4127 	struct vnode *vp = arg;
4128 
4129 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4130 }
4131 
4132 static void
4133 vfs_knlunlock(void *arg)
4134 {
4135 	struct vnode *vp = arg;
4136 
4137 	VOP_UNLOCK(vp, 0);
4138 }
4139 
4140 static void
4141 vfs_knl_assert_locked(void *arg)
4142 {
4143 #ifdef DEBUG_VFS_LOCKS
4144 	struct vnode *vp = arg;
4145 
4146 	ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4147 #endif
4148 }
4149 
4150 static void
4151 vfs_knl_assert_unlocked(void *arg)
4152 {
4153 #ifdef DEBUG_VFS_LOCKS
4154 	struct vnode *vp = arg;
4155 
4156 	ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4157 #endif
4158 }
4159 
4160 int
4161 vfs_kqfilter(struct vop_kqfilter_args *ap)
4162 {
4163 	struct vnode *vp = ap->a_vp;
4164 	struct knote *kn = ap->a_kn;
4165 	struct knlist *knl;
4166 
4167 	switch (kn->kn_filter) {
4168 	case EVFILT_READ:
4169 		kn->kn_fop = &vfsread_filtops;
4170 		break;
4171 	case EVFILT_WRITE:
4172 		kn->kn_fop = &vfswrite_filtops;
4173 		break;
4174 	case EVFILT_VNODE:
4175 		kn->kn_fop = &vfsvnode_filtops;
4176 		break;
4177 	default:
4178 		return (EINVAL);
4179 	}
4180 
4181 	kn->kn_hook = (caddr_t)vp;
4182 
4183 	v_addpollinfo(vp);
4184 	if (vp->v_pollinfo == NULL)
4185 		return (ENOMEM);
4186 	knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4187 	knlist_add(knl, kn, 0);
4188 
4189 	return (0);
4190 }
4191 
4192 /*
4193  * Detach knote from vnode
4194  */
4195 static void
4196 filt_vfsdetach(struct knote *kn)
4197 {
4198 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4199 
4200 	KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4201 	knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4202 }
4203 
4204 /*ARGSUSED*/
4205 static int
4206 filt_vfsread(struct knote *kn, long hint)
4207 {
4208 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4209 	struct vattr va;
4210 	int res;
4211 
4212 	/*
4213 	 * filesystem is gone, so set the EOF flag and schedule
4214 	 * the knote for deletion.
4215 	 */
4216 	if (hint == NOTE_REVOKE) {
4217 		VI_LOCK(vp);
4218 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4219 		VI_UNLOCK(vp);
4220 		return (1);
4221 	}
4222 
4223 	if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4224 		return (0);
4225 
4226 	VI_LOCK(vp);
4227 	kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4228 	res = (kn->kn_data != 0);
4229 	VI_UNLOCK(vp);
4230 	return (res);
4231 }
4232 
4233 /*ARGSUSED*/
4234 static int
4235 filt_vfswrite(struct knote *kn, long hint)
4236 {
4237 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4238 
4239 	VI_LOCK(vp);
4240 
4241 	/*
4242 	 * filesystem is gone, so set the EOF flag and schedule
4243 	 * the knote for deletion.
4244 	 */
4245 	if (hint == NOTE_REVOKE)
4246 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4247 
4248 	kn->kn_data = 0;
4249 	VI_UNLOCK(vp);
4250 	return (1);
4251 }
4252 
4253 static int
4254 filt_vfsvnode(struct knote *kn, long hint)
4255 {
4256 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4257 	int res;
4258 
4259 	VI_LOCK(vp);
4260 	if (kn->kn_sfflags & hint)
4261 		kn->kn_fflags |= hint;
4262 	if (hint == NOTE_REVOKE) {
4263 		kn->kn_flags |= EV_EOF;
4264 		VI_UNLOCK(vp);
4265 		return (1);
4266 	}
4267 	res = (kn->kn_fflags != 0);
4268 	VI_UNLOCK(vp);
4269 	return (res);
4270 }
4271 
4272 int
4273 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4274 {
4275 	int error;
4276 
4277 	if (dp->d_reclen > ap->a_uio->uio_resid)
4278 		return (ENAMETOOLONG);
4279 	error = uiomove(dp, dp->d_reclen, ap->a_uio);
4280 	if (error) {
4281 		if (ap->a_ncookies != NULL) {
4282 			if (ap->a_cookies != NULL)
4283 				free(ap->a_cookies, M_TEMP);
4284 			ap->a_cookies = NULL;
4285 			*ap->a_ncookies = 0;
4286 		}
4287 		return (error);
4288 	}
4289 	if (ap->a_ncookies == NULL)
4290 		return (0);
4291 
4292 	KASSERT(ap->a_cookies,
4293 	    ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4294 
4295 	*ap->a_cookies = realloc(*ap->a_cookies,
4296 	    (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4297 	(*ap->a_cookies)[*ap->a_ncookies] = off;
4298 	return (0);
4299 }
4300 
4301 /*
4302  * Mark for update the access time of the file if the filesystem
4303  * supports VOP_MARKATIME.  This functionality is used by execve and
4304  * mmap, so we want to avoid the I/O implied by directly setting
4305  * va_atime for the sake of efficiency.
4306  */
4307 void
4308 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4309 {
4310 	struct mount *mp;
4311 
4312 	mp = vp->v_mount;
4313 	VFS_ASSERT_GIANT(mp);
4314 	ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4315 	if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4316 		(void)VOP_MARKATIME(vp);
4317 }
4318 
4319 /*
4320  * The purpose of this routine is to remove granularity from accmode_t,
4321  * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4322  * VADMIN and VAPPEND.
4323  *
4324  * If it returns 0, the caller is supposed to continue with the usual
4325  * access checks using 'accmode' as modified by this routine.  If it
4326  * returns nonzero value, the caller is supposed to return that value
4327  * as errno.
4328  *
4329  * Note that after this routine runs, accmode may be zero.
4330  */
4331 int
4332 vfs_unixify_accmode(accmode_t *accmode)
4333 {
4334 	/*
4335 	 * There is no way to specify explicit "deny" rule using
4336 	 * file mode or POSIX.1e ACLs.
4337 	 */
4338 	if (*accmode & VEXPLICIT_DENY) {
4339 		*accmode = 0;
4340 		return (0);
4341 	}
4342 
4343 	/*
4344 	 * None of these can be translated into usual access bits.
4345 	 * Also, the common case for NFSv4 ACLs is to not contain
4346 	 * either of these bits. Caller should check for VWRITE
4347 	 * on the containing directory instead.
4348 	 */
4349 	if (*accmode & (VDELETE_CHILD | VDELETE))
4350 		return (EPERM);
4351 
4352 	if (*accmode & VADMIN_PERMS) {
4353 		*accmode &= ~VADMIN_PERMS;
4354 		*accmode |= VADMIN;
4355 	}
4356 
4357 	/*
4358 	 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4359 	 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4360 	 */
4361 	*accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
4362 
4363 	return (0);
4364 }
4365