xref: /freebsd/sys/kern/vfs_subr.c (revision 774f94f14c92bf94afc21d8c8d7a1e8f2fdf5a48)
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 uint16_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 #if 0
804 			/* These messages are temporary debugging aids */
805 			if (vnlru_nowhere < 5)
806 				printf("vnlru process getting nowhere..\n");
807 			else if (vnlru_nowhere == 5)
808 				printf("vnlru process messages stopped.\n");
809 #endif
810 			vnlru_nowhere++;
811 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
812 		} else
813 			uio_yield();
814 	}
815 }
816 
817 static struct kproc_desc vnlru_kp = {
818 	"vnlru",
819 	vnlru_proc,
820 	&vnlruproc
821 };
822 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
823     &vnlru_kp);
824 
825 /*
826  * Routines having to do with the management of the vnode table.
827  */
828 
829 void
830 vdestroy(struct vnode *vp)
831 {
832 	struct bufobj *bo;
833 
834 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
835 	mtx_lock(&vnode_free_list_mtx);
836 	numvnodes--;
837 	mtx_unlock(&vnode_free_list_mtx);
838 	bo = &vp->v_bufobj;
839 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
840 	    ("cleaned vnode still on the free list."));
841 	VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
842 	VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
843 	VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
844 	VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
845 	VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
846 	VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
847 	VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
848 	VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
849 	VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
850 	VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
851 	VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
852 	VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
853 	VI_UNLOCK(vp);
854 #ifdef MAC
855 	mac_vnode_destroy(vp);
856 #endif
857 	if (vp->v_pollinfo != NULL)
858 		destroy_vpollinfo(vp->v_pollinfo);
859 #ifdef INVARIANTS
860 	/* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */
861 	vp->v_op = NULL;
862 #endif
863 	lockdestroy(vp->v_vnlock);
864 	mtx_destroy(&vp->v_interlock);
865 	mtx_destroy(BO_MTX(bo));
866 	uma_zfree(vnode_zone, vp);
867 }
868 
869 /*
870  * Try to recycle a freed vnode.  We abort if anyone picks up a reference
871  * before we actually vgone().  This function must be called with the vnode
872  * held to prevent the vnode from being returned to the free list midway
873  * through vgone().
874  */
875 static int
876 vtryrecycle(struct vnode *vp)
877 {
878 	struct mount *vnmp;
879 
880 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
881 	VNASSERT(vp->v_holdcnt, vp,
882 	    ("vtryrecycle: Recycling vp %p without a reference.", vp));
883 	/*
884 	 * This vnode may found and locked via some other list, if so we
885 	 * can't recycle it yet.
886 	 */
887 	if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
888 		CTR2(KTR_VFS,
889 		    "%s: impossible to recycle, vp %p lock is already held",
890 		    __func__, vp);
891 		return (EWOULDBLOCK);
892 	}
893 	/*
894 	 * Don't recycle if its filesystem is being suspended.
895 	 */
896 	if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
897 		VOP_UNLOCK(vp, 0);
898 		CTR2(KTR_VFS,
899 		    "%s: impossible to recycle, cannot start the write for %p",
900 		    __func__, vp);
901 		return (EBUSY);
902 	}
903 	/*
904 	 * If we got this far, we need to acquire the interlock and see if
905 	 * anyone picked up this vnode from another list.  If not, we will
906 	 * mark it with DOOMED via vgonel() so that anyone who does find it
907 	 * will skip over it.
908 	 */
909 	VI_LOCK(vp);
910 	if (vp->v_usecount) {
911 		VOP_UNLOCK(vp, LK_INTERLOCK);
912 		vn_finished_write(vnmp);
913 		CTR2(KTR_VFS,
914 		    "%s: impossible to recycle, %p is already referenced",
915 		    __func__, vp);
916 		return (EBUSY);
917 	}
918 	if ((vp->v_iflag & VI_DOOMED) == 0)
919 		vgonel(vp);
920 	VOP_UNLOCK(vp, LK_INTERLOCK);
921 	vn_finished_write(vnmp);
922 	return (0);
923 }
924 
925 /*
926  * Return the next vnode from the free list.
927  */
928 int
929 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
930     struct vnode **vpp)
931 {
932 	struct vnode *vp = NULL;
933 	struct bufobj *bo;
934 
935 	CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
936 	mtx_lock(&vnode_free_list_mtx);
937 	/*
938 	 * Lend our context to reclaim vnodes if they've exceeded the max.
939 	 */
940 	if (freevnodes > wantfreevnodes)
941 		vnlru_free(1);
942 	/*
943 	 * Wait for available vnodes.
944 	 */
945 	if (numvnodes > desiredvnodes) {
946 		if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) {
947 			/*
948 			 * File system is beeing suspended, we cannot risk a
949 			 * deadlock here, so allocate new vnode anyway.
950 			 */
951 			if (freevnodes > wantfreevnodes)
952 				vnlru_free(freevnodes - wantfreevnodes);
953 			goto alloc;
954 		}
955 		if (vnlruproc_sig == 0) {
956 			vnlruproc_sig = 1;	/* avoid unnecessary wakeups */
957 			wakeup(vnlruproc);
958 		}
959 		msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
960 		    "vlruwk", hz);
961 #if 0	/* XXX Not all VFS_VGET/ffs_vget callers check returns. */
962 		if (numvnodes > desiredvnodes) {
963 			mtx_unlock(&vnode_free_list_mtx);
964 			return (ENFILE);
965 		}
966 #endif
967 	}
968 alloc:
969 	numvnodes++;
970 	mtx_unlock(&vnode_free_list_mtx);
971 	vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
972 	/*
973 	 * Setup locks.
974 	 */
975 	vp->v_vnlock = &vp->v_lock;
976 	mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
977 	/*
978 	 * By default, don't allow shared locks unless filesystems
979 	 * opt-in.
980 	 */
981 	lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
982 	/*
983 	 * Initialize bufobj.
984 	 */
985 	bo = &vp->v_bufobj;
986 	bo->__bo_vnode = vp;
987 	mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
988 	bo->bo_ops = &buf_ops_bio;
989 	bo->bo_private = vp;
990 	TAILQ_INIT(&bo->bo_clean.bv_hd);
991 	TAILQ_INIT(&bo->bo_dirty.bv_hd);
992 	/*
993 	 * Initialize namecache.
994 	 */
995 	LIST_INIT(&vp->v_cache_src);
996 	TAILQ_INIT(&vp->v_cache_dst);
997 	/*
998 	 * Finalize various vnode identity bits.
999 	 */
1000 	vp->v_type = VNON;
1001 	vp->v_tag = tag;
1002 	vp->v_op = vops;
1003 	v_incr_usecount(vp);
1004 	vp->v_data = 0;
1005 #ifdef MAC
1006 	mac_vnode_init(vp);
1007 	if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1008 		mac_vnode_associate_singlelabel(mp, vp);
1009 	else if (mp == NULL && vops != &dead_vnodeops)
1010 		printf("NULL mp in getnewvnode()\n");
1011 #endif
1012 	if (mp != NULL) {
1013 		bo->bo_bsize = mp->mnt_stat.f_iosize;
1014 		if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1015 			vp->v_vflag |= VV_NOKNOTE;
1016 	}
1017 
1018 	*vpp = vp;
1019 	return (0);
1020 }
1021 
1022 /*
1023  * Delete from old mount point vnode list, if on one.
1024  */
1025 static void
1026 delmntque(struct vnode *vp)
1027 {
1028 	struct mount *mp;
1029 
1030 	mp = vp->v_mount;
1031 	if (mp == NULL)
1032 		return;
1033 	MNT_ILOCK(mp);
1034 	vp->v_mount = NULL;
1035 	VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1036 		("bad mount point vnode list size"));
1037 	TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1038 	mp->mnt_nvnodelistsize--;
1039 	MNT_REL(mp);
1040 	MNT_IUNLOCK(mp);
1041 }
1042 
1043 static void
1044 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1045 {
1046 
1047 	vp->v_data = NULL;
1048 	vp->v_op = &dead_vnodeops;
1049 	/* XXX non mp-safe fs may still call insmntque with vnode
1050 	   unlocked */
1051 	if (!VOP_ISLOCKED(vp))
1052 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1053 	vgone(vp);
1054 	vput(vp);
1055 }
1056 
1057 /*
1058  * Insert into list of vnodes for the new mount point, if available.
1059  */
1060 int
1061 insmntque1(struct vnode *vp, struct mount *mp,
1062 	void (*dtr)(struct vnode *, void *), void *dtr_arg)
1063 {
1064 	int locked;
1065 
1066 	KASSERT(vp->v_mount == NULL,
1067 		("insmntque: vnode already on per mount vnode list"));
1068 	VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1069 #ifdef DEBUG_VFS_LOCKS
1070 	if (!VFS_NEEDSGIANT(mp))
1071 		ASSERT_VOP_ELOCKED(vp,
1072 		    "insmntque: mp-safe fs and non-locked vp");
1073 #endif
1074 	MNT_ILOCK(mp);
1075 	if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1076 	    ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1077 	     mp->mnt_nvnodelistsize == 0)) {
1078 		locked = VOP_ISLOCKED(vp);
1079 		if (!locked || (locked == LK_EXCLUSIVE &&
1080 		     (vp->v_vflag & VV_FORCEINSMQ) == 0)) {
1081 			MNT_IUNLOCK(mp);
1082 			if (dtr != NULL)
1083 				dtr(vp, dtr_arg);
1084 			return (EBUSY);
1085 		}
1086 	}
1087 	vp->v_mount = mp;
1088 	MNT_REF(mp);
1089 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1090 	VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1091 		("neg mount point vnode list size"));
1092 	mp->mnt_nvnodelistsize++;
1093 	MNT_IUNLOCK(mp);
1094 	return (0);
1095 }
1096 
1097 int
1098 insmntque(struct vnode *vp, struct mount *mp)
1099 {
1100 
1101 	return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1102 }
1103 
1104 /*
1105  * Flush out and invalidate all buffers associated with a bufobj
1106  * Called with the underlying object locked.
1107  */
1108 int
1109 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1110 {
1111 	int error;
1112 
1113 	BO_LOCK(bo);
1114 	if (flags & V_SAVE) {
1115 		error = bufobj_wwait(bo, slpflag, slptimeo);
1116 		if (error) {
1117 			BO_UNLOCK(bo);
1118 			return (error);
1119 		}
1120 		if (bo->bo_dirty.bv_cnt > 0) {
1121 			BO_UNLOCK(bo);
1122 			if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1123 				return (error);
1124 			/*
1125 			 * XXX We could save a lock/unlock if this was only
1126 			 * enabled under INVARIANTS
1127 			 */
1128 			BO_LOCK(bo);
1129 			if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1130 				panic("vinvalbuf: dirty bufs");
1131 		}
1132 	}
1133 	/*
1134 	 * If you alter this loop please notice that interlock is dropped and
1135 	 * reacquired in flushbuflist.  Special care is needed to ensure that
1136 	 * no race conditions occur from this.
1137 	 */
1138 	do {
1139 		error = flushbuflist(&bo->bo_clean,
1140 		    flags, bo, slpflag, slptimeo);
1141 		if (error == 0)
1142 			error = flushbuflist(&bo->bo_dirty,
1143 			    flags, bo, slpflag, slptimeo);
1144 		if (error != 0 && error != EAGAIN) {
1145 			BO_UNLOCK(bo);
1146 			return (error);
1147 		}
1148 	} while (error != 0);
1149 
1150 	/*
1151 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1152 	 * have write I/O in-progress but if there is a VM object then the
1153 	 * VM object can also have read-I/O in-progress.
1154 	 */
1155 	do {
1156 		bufobj_wwait(bo, 0, 0);
1157 		BO_UNLOCK(bo);
1158 		if (bo->bo_object != NULL) {
1159 			VM_OBJECT_LOCK(bo->bo_object);
1160 			vm_object_pip_wait(bo->bo_object, "bovlbx");
1161 			VM_OBJECT_UNLOCK(bo->bo_object);
1162 		}
1163 		BO_LOCK(bo);
1164 	} while (bo->bo_numoutput > 0);
1165 	BO_UNLOCK(bo);
1166 
1167 	/*
1168 	 * Destroy the copy in the VM cache, too.
1169 	 */
1170 	if (bo->bo_object != NULL && (flags & (V_ALT | V_NORMAL)) == 0) {
1171 		VM_OBJECT_LOCK(bo->bo_object);
1172 		vm_object_page_remove(bo->bo_object, 0, 0,
1173 			(flags & V_SAVE) ? TRUE : FALSE);
1174 		VM_OBJECT_UNLOCK(bo->bo_object);
1175 	}
1176 
1177 #ifdef INVARIANTS
1178 	BO_LOCK(bo);
1179 	if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1180 	    (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1181 		panic("vinvalbuf: flush failed");
1182 	BO_UNLOCK(bo);
1183 #endif
1184 	return (0);
1185 }
1186 
1187 /*
1188  * Flush out and invalidate all buffers associated with a vnode.
1189  * Called with the underlying object locked.
1190  */
1191 int
1192 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1193 {
1194 
1195 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1196 	ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1197 	return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1198 }
1199 
1200 /*
1201  * Flush out buffers on the specified list.
1202  *
1203  */
1204 static int
1205 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1206     int slptimeo)
1207 {
1208 	struct buf *bp, *nbp;
1209 	int retval, error;
1210 	daddr_t lblkno;
1211 	b_xflags_t xflags;
1212 
1213 	ASSERT_BO_LOCKED(bo);
1214 
1215 	retval = 0;
1216 	TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1217 		if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1218 		    ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1219 			continue;
1220 		}
1221 		lblkno = 0;
1222 		xflags = 0;
1223 		if (nbp != NULL) {
1224 			lblkno = nbp->b_lblkno;
1225 			xflags = nbp->b_xflags &
1226 				(BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1227 		}
1228 		retval = EAGAIN;
1229 		error = BUF_TIMELOCK(bp,
1230 		    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1231 		    "flushbuf", slpflag, slptimeo);
1232 		if (error) {
1233 			BO_LOCK(bo);
1234 			return (error != ENOLCK ? error : EAGAIN);
1235 		}
1236 		KASSERT(bp->b_bufobj == bo,
1237 		    ("bp %p wrong b_bufobj %p should be %p",
1238 		    bp, bp->b_bufobj, bo));
1239 		if (bp->b_bufobj != bo) {	/* XXX: necessary ? */
1240 			BUF_UNLOCK(bp);
1241 			BO_LOCK(bo);
1242 			return (EAGAIN);
1243 		}
1244 		/*
1245 		 * XXX Since there are no node locks for NFS, I
1246 		 * believe there is a slight chance that a delayed
1247 		 * write will occur while sleeping just above, so
1248 		 * check for it.
1249 		 */
1250 		if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1251 		    (flags & V_SAVE)) {
1252 			bremfree(bp);
1253 			bp->b_flags |= B_ASYNC;
1254 			bwrite(bp);
1255 			BO_LOCK(bo);
1256 			return (EAGAIN);	/* XXX: why not loop ? */
1257 		}
1258 		bremfree(bp);
1259 		bp->b_flags |= (B_INVAL | B_RELBUF);
1260 		bp->b_flags &= ~B_ASYNC;
1261 		brelse(bp);
1262 		BO_LOCK(bo);
1263 		if (nbp != NULL &&
1264 		    (nbp->b_bufobj != bo ||
1265 		     nbp->b_lblkno != lblkno ||
1266 		     (nbp->b_xflags &
1267 		      (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1268 			break;			/* nbp invalid */
1269 	}
1270 	return (retval);
1271 }
1272 
1273 /*
1274  * Truncate a file's buffer and pages to a specified length.  This
1275  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1276  * sync activity.
1277  */
1278 int
1279 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
1280     off_t length, int blksize)
1281 {
1282 	struct buf *bp, *nbp;
1283 	int anyfreed;
1284 	int trunclbn;
1285 	struct bufobj *bo;
1286 
1287 	CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1288 	    vp, cred, blksize, (uintmax_t)length);
1289 
1290 	/*
1291 	 * Round up to the *next* lbn.
1292 	 */
1293 	trunclbn = (length + blksize - 1) / blksize;
1294 
1295 	ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1296 restart:
1297 	bo = &vp->v_bufobj;
1298 	BO_LOCK(bo);
1299 	anyfreed = 1;
1300 	for (;anyfreed;) {
1301 		anyfreed = 0;
1302 		TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1303 			if (bp->b_lblkno < trunclbn)
1304 				continue;
1305 			if (BUF_LOCK(bp,
1306 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1307 			    BO_MTX(bo)) == ENOLCK)
1308 				goto restart;
1309 
1310 			bremfree(bp);
1311 			bp->b_flags |= (B_INVAL | B_RELBUF);
1312 			bp->b_flags &= ~B_ASYNC;
1313 			brelse(bp);
1314 			anyfreed = 1;
1315 
1316 			if (nbp != NULL &&
1317 			    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1318 			    (nbp->b_vp != vp) ||
1319 			    (nbp->b_flags & B_DELWRI))) {
1320 				goto restart;
1321 			}
1322 			BO_LOCK(bo);
1323 		}
1324 
1325 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1326 			if (bp->b_lblkno < trunclbn)
1327 				continue;
1328 			if (BUF_LOCK(bp,
1329 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1330 			    BO_MTX(bo)) == ENOLCK)
1331 				goto restart;
1332 			bremfree(bp);
1333 			bp->b_flags |= (B_INVAL | B_RELBUF);
1334 			bp->b_flags &= ~B_ASYNC;
1335 			brelse(bp);
1336 			anyfreed = 1;
1337 			if (nbp != NULL &&
1338 			    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1339 			    (nbp->b_vp != vp) ||
1340 			    (nbp->b_flags & B_DELWRI) == 0)) {
1341 				goto restart;
1342 			}
1343 			BO_LOCK(bo);
1344 		}
1345 	}
1346 
1347 	if (length > 0) {
1348 restartsync:
1349 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1350 			if (bp->b_lblkno > 0)
1351 				continue;
1352 			/*
1353 			 * Since we hold the vnode lock this should only
1354 			 * fail if we're racing with the buf daemon.
1355 			 */
1356 			if (BUF_LOCK(bp,
1357 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1358 			    BO_MTX(bo)) == ENOLCK) {
1359 				goto restart;
1360 			}
1361 			VNASSERT((bp->b_flags & B_DELWRI), vp,
1362 			    ("buf(%p) on dirty queue without DELWRI", bp));
1363 
1364 			bremfree(bp);
1365 			bawrite(bp);
1366 			BO_LOCK(bo);
1367 			goto restartsync;
1368 		}
1369 	}
1370 
1371 	bufobj_wwait(bo, 0, 0);
1372 	BO_UNLOCK(bo);
1373 	vnode_pager_setsize(vp, length);
1374 
1375 	return (0);
1376 }
1377 
1378 /*
1379  * buf_splay() - splay tree core for the clean/dirty list of buffers in
1380  *		 a vnode.
1381  *
1382  *	NOTE: We have to deal with the special case of a background bitmap
1383  *	buffer, a situation where two buffers will have the same logical
1384  *	block offset.  We want (1) only the foreground buffer to be accessed
1385  *	in a lookup and (2) must differentiate between the foreground and
1386  *	background buffer in the splay tree algorithm because the splay
1387  *	tree cannot normally handle multiple entities with the same 'index'.
1388  *	We accomplish this by adding differentiating flags to the splay tree's
1389  *	numerical domain.
1390  */
1391 static
1392 struct buf *
1393 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1394 {
1395 	struct buf dummy;
1396 	struct buf *lefttreemax, *righttreemin, *y;
1397 
1398 	if (root == NULL)
1399 		return (NULL);
1400 	lefttreemax = righttreemin = &dummy;
1401 	for (;;) {
1402 		if (lblkno < root->b_lblkno ||
1403 		    (lblkno == root->b_lblkno &&
1404 		    (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1405 			if ((y = root->b_left) == NULL)
1406 				break;
1407 			if (lblkno < y->b_lblkno) {
1408 				/* Rotate right. */
1409 				root->b_left = y->b_right;
1410 				y->b_right = root;
1411 				root = y;
1412 				if ((y = root->b_left) == NULL)
1413 					break;
1414 			}
1415 			/* Link into the new root's right tree. */
1416 			righttreemin->b_left = root;
1417 			righttreemin = root;
1418 		} else if (lblkno > root->b_lblkno ||
1419 		    (lblkno == root->b_lblkno &&
1420 		    (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1421 			if ((y = root->b_right) == NULL)
1422 				break;
1423 			if (lblkno > y->b_lblkno) {
1424 				/* Rotate left. */
1425 				root->b_right = y->b_left;
1426 				y->b_left = root;
1427 				root = y;
1428 				if ((y = root->b_right) == NULL)
1429 					break;
1430 			}
1431 			/* Link into the new root's left tree. */
1432 			lefttreemax->b_right = root;
1433 			lefttreemax = root;
1434 		} else {
1435 			break;
1436 		}
1437 		root = y;
1438 	}
1439 	/* Assemble the new root. */
1440 	lefttreemax->b_right = root->b_left;
1441 	righttreemin->b_left = root->b_right;
1442 	root->b_left = dummy.b_right;
1443 	root->b_right = dummy.b_left;
1444 	return (root);
1445 }
1446 
1447 static void
1448 buf_vlist_remove(struct buf *bp)
1449 {
1450 	struct buf *root;
1451 	struct bufv *bv;
1452 
1453 	KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1454 	ASSERT_BO_LOCKED(bp->b_bufobj);
1455 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1456 	    (BX_VNDIRTY|BX_VNCLEAN),
1457 	    ("buf_vlist_remove: Buf %p is on two lists", bp));
1458 	if (bp->b_xflags & BX_VNDIRTY)
1459 		bv = &bp->b_bufobj->bo_dirty;
1460 	else
1461 		bv = &bp->b_bufobj->bo_clean;
1462 	if (bp != bv->bv_root) {
1463 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1464 		KASSERT(root == bp, ("splay lookup failed in remove"));
1465 	}
1466 	if (bp->b_left == NULL) {
1467 		root = bp->b_right;
1468 	} else {
1469 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1470 		root->b_right = bp->b_right;
1471 	}
1472 	bv->bv_root = root;
1473 	TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1474 	bv->bv_cnt--;
1475 	bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1476 }
1477 
1478 /*
1479  * Add the buffer to the sorted clean or dirty block list using a
1480  * splay tree algorithm.
1481  *
1482  * NOTE: xflags is passed as a constant, optimizing this inline function!
1483  */
1484 static void
1485 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1486 {
1487 	struct buf *root;
1488 	struct bufv *bv;
1489 
1490 	ASSERT_BO_LOCKED(bo);
1491 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1492 	    ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1493 	bp->b_xflags |= xflags;
1494 	if (xflags & BX_VNDIRTY)
1495 		bv = &bo->bo_dirty;
1496 	else
1497 		bv = &bo->bo_clean;
1498 
1499 	root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1500 	if (root == NULL) {
1501 		bp->b_left = NULL;
1502 		bp->b_right = NULL;
1503 		TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1504 	} else if (bp->b_lblkno < root->b_lblkno ||
1505 	    (bp->b_lblkno == root->b_lblkno &&
1506 	    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1507 		bp->b_left = root->b_left;
1508 		bp->b_right = root;
1509 		root->b_left = NULL;
1510 		TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1511 	} else {
1512 		bp->b_right = root->b_right;
1513 		bp->b_left = root;
1514 		root->b_right = NULL;
1515 		TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1516 	}
1517 	bv->bv_cnt++;
1518 	bv->bv_root = bp;
1519 }
1520 
1521 /*
1522  * Lookup a buffer using the splay tree.  Note that we specifically avoid
1523  * shadow buffers used in background bitmap writes.
1524  *
1525  * This code isn't quite efficient as it could be because we are maintaining
1526  * two sorted lists and do not know which list the block resides in.
1527  *
1528  * During a "make buildworld" the desired buffer is found at one of
1529  * the roots more than 60% of the time.  Thus, checking both roots
1530  * before performing either splay eliminates unnecessary splays on the
1531  * first tree splayed.
1532  */
1533 struct buf *
1534 gbincore(struct bufobj *bo, daddr_t lblkno)
1535 {
1536 	struct buf *bp;
1537 
1538 	ASSERT_BO_LOCKED(bo);
1539 	if ((bp = bo->bo_clean.bv_root) != NULL &&
1540 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1541 		return (bp);
1542 	if ((bp = bo->bo_dirty.bv_root) != NULL &&
1543 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1544 		return (bp);
1545 	if ((bp = bo->bo_clean.bv_root) != NULL) {
1546 		bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1547 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1548 			return (bp);
1549 	}
1550 	if ((bp = bo->bo_dirty.bv_root) != NULL) {
1551 		bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1552 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1553 			return (bp);
1554 	}
1555 	return (NULL);
1556 }
1557 
1558 /*
1559  * Associate a buffer with a vnode.
1560  */
1561 void
1562 bgetvp(struct vnode *vp, struct buf *bp)
1563 {
1564 	struct bufobj *bo;
1565 
1566 	bo = &vp->v_bufobj;
1567 	ASSERT_BO_LOCKED(bo);
1568 	VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1569 
1570 	CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1571 	VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1572 	    ("bgetvp: bp already attached! %p", bp));
1573 
1574 	vhold(vp);
1575 	if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT)
1576 		bp->b_flags |= B_NEEDSGIANT;
1577 	bp->b_vp = vp;
1578 	bp->b_bufobj = bo;
1579 	/*
1580 	 * Insert onto list for new vnode.
1581 	 */
1582 	buf_vlist_add(bp, bo, BX_VNCLEAN);
1583 }
1584 
1585 /*
1586  * Disassociate a buffer from a vnode.
1587  */
1588 void
1589 brelvp(struct buf *bp)
1590 {
1591 	struct bufobj *bo;
1592 	struct vnode *vp;
1593 
1594 	CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1595 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1596 
1597 	/*
1598 	 * Delete from old vnode list, if on one.
1599 	 */
1600 	vp = bp->b_vp;		/* XXX */
1601 	bo = bp->b_bufobj;
1602 	BO_LOCK(bo);
1603 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1604 		buf_vlist_remove(bp);
1605 	else
1606 		panic("brelvp: Buffer %p not on queue.", bp);
1607 	if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1608 		bo->bo_flag &= ~BO_ONWORKLST;
1609 		mtx_lock(&sync_mtx);
1610 		LIST_REMOVE(bo, bo_synclist);
1611 		syncer_worklist_len--;
1612 		mtx_unlock(&sync_mtx);
1613 	}
1614 	bp->b_flags &= ~B_NEEDSGIANT;
1615 	bp->b_vp = NULL;
1616 	bp->b_bufobj = NULL;
1617 	BO_UNLOCK(bo);
1618 	vdrop(vp);
1619 }
1620 
1621 /*
1622  * Add an item to the syncer work queue.
1623  */
1624 static void
1625 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1626 {
1627 	int queue, slot;
1628 
1629 	ASSERT_BO_LOCKED(bo);
1630 
1631 	mtx_lock(&sync_mtx);
1632 	if (bo->bo_flag & BO_ONWORKLST)
1633 		LIST_REMOVE(bo, bo_synclist);
1634 	else {
1635 		bo->bo_flag |= BO_ONWORKLST;
1636 		syncer_worklist_len++;
1637 	}
1638 
1639 	if (delay > syncer_maxdelay - 2)
1640 		delay = syncer_maxdelay - 2;
1641 	slot = (syncer_delayno + delay) & syncer_mask;
1642 
1643 	queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ :
1644 	    WI_MPSAFEQ;
1645 	LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo,
1646 	    bo_synclist);
1647 	mtx_unlock(&sync_mtx);
1648 }
1649 
1650 static int
1651 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1652 {
1653 	int error, len;
1654 
1655 	mtx_lock(&sync_mtx);
1656 	len = syncer_worklist_len - sync_vnode_count;
1657 	mtx_unlock(&sync_mtx);
1658 	error = SYSCTL_OUT(req, &len, sizeof(len));
1659 	return (error);
1660 }
1661 
1662 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1663     sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1664 
1665 static struct proc *updateproc;
1666 static void sched_sync(void);
1667 static struct kproc_desc up_kp = {
1668 	"syncer",
1669 	sched_sync,
1670 	&updateproc
1671 };
1672 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1673 
1674 static int
1675 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1676 {
1677 	struct vnode *vp;
1678 	struct mount *mp;
1679 
1680 	*bo = LIST_FIRST(slp);
1681 	if (*bo == NULL)
1682 		return (0);
1683 	vp = (*bo)->__bo_vnode;	/* XXX */
1684 	if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1685 		return (1);
1686 	/*
1687 	 * We use vhold in case the vnode does not
1688 	 * successfully sync.  vhold prevents the vnode from
1689 	 * going away when we unlock the sync_mtx so that
1690 	 * we can acquire the vnode interlock.
1691 	 */
1692 	vholdl(vp);
1693 	mtx_unlock(&sync_mtx);
1694 	VI_UNLOCK(vp);
1695 	if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1696 		vdrop(vp);
1697 		mtx_lock(&sync_mtx);
1698 		return (*bo == LIST_FIRST(slp));
1699 	}
1700 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1701 	(void) VOP_FSYNC(vp, MNT_LAZY, td);
1702 	VOP_UNLOCK(vp, 0);
1703 	vn_finished_write(mp);
1704 	BO_LOCK(*bo);
1705 	if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1706 		/*
1707 		 * Put us back on the worklist.  The worklist
1708 		 * routine will remove us from our current
1709 		 * position and then add us back in at a later
1710 		 * position.
1711 		 */
1712 		vn_syncer_add_to_worklist(*bo, syncdelay);
1713 	}
1714 	BO_UNLOCK(*bo);
1715 	vdrop(vp);
1716 	mtx_lock(&sync_mtx);
1717 	return (0);
1718 }
1719 
1720 /*
1721  * System filesystem synchronizer daemon.
1722  */
1723 static void
1724 sched_sync(void)
1725 {
1726 	struct synclist *gnext, *next;
1727 	struct synclist *gslp, *slp;
1728 	struct bufobj *bo;
1729 	long starttime;
1730 	struct thread *td = curthread;
1731 	int last_work_seen;
1732 	int net_worklist_len;
1733 	int syncer_final_iter;
1734 	int first_printf;
1735 	int error;
1736 
1737 	last_work_seen = 0;
1738 	syncer_final_iter = 0;
1739 	first_printf = 1;
1740 	syncer_state = SYNCER_RUNNING;
1741 	starttime = time_uptime;
1742 	td->td_pflags |= TDP_NORUNNINGBUF;
1743 
1744 	EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1745 	    SHUTDOWN_PRI_LAST);
1746 
1747 	mtx_lock(&sync_mtx);
1748 	for (;;) {
1749 		if (syncer_state == SYNCER_FINAL_DELAY &&
1750 		    syncer_final_iter == 0) {
1751 			mtx_unlock(&sync_mtx);
1752 			kproc_suspend_check(td->td_proc);
1753 			mtx_lock(&sync_mtx);
1754 		}
1755 		net_worklist_len = syncer_worklist_len - sync_vnode_count;
1756 		if (syncer_state != SYNCER_RUNNING &&
1757 		    starttime != time_uptime) {
1758 			if (first_printf) {
1759 				printf("\nSyncing disks, vnodes remaining...");
1760 				first_printf = 0;
1761 			}
1762 			printf("%d ", net_worklist_len);
1763 		}
1764 		starttime = time_uptime;
1765 
1766 		/*
1767 		 * Push files whose dirty time has expired.  Be careful
1768 		 * of interrupt race on slp queue.
1769 		 *
1770 		 * Skip over empty worklist slots when shutting down.
1771 		 */
1772 		do {
1773 			slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1774 			gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1775 			syncer_delayno += 1;
1776 			if (syncer_delayno == syncer_maxdelay)
1777 				syncer_delayno = 0;
1778 			next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1779 			gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1780 			/*
1781 			 * If the worklist has wrapped since the
1782 			 * it was emptied of all but syncer vnodes,
1783 			 * switch to the FINAL_DELAY state and run
1784 			 * for one more second.
1785 			 */
1786 			if (syncer_state == SYNCER_SHUTTING_DOWN &&
1787 			    net_worklist_len == 0 &&
1788 			    last_work_seen == syncer_delayno) {
1789 				syncer_state = SYNCER_FINAL_DELAY;
1790 				syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1791 			}
1792 		} while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1793 		    LIST_EMPTY(gslp) && syncer_worklist_len > 0);
1794 
1795 		/*
1796 		 * Keep track of the last time there was anything
1797 		 * on the worklist other than syncer vnodes.
1798 		 * Return to the SHUTTING_DOWN state if any
1799 		 * new work appears.
1800 		 */
1801 		if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1802 			last_work_seen = syncer_delayno;
1803 		if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1804 			syncer_state = SYNCER_SHUTTING_DOWN;
1805 		while (!LIST_EMPTY(slp)) {
1806 			error = sync_vnode(slp, &bo, td);
1807 			if (error == 1) {
1808 				LIST_REMOVE(bo, bo_synclist);
1809 				LIST_INSERT_HEAD(next, bo, bo_synclist);
1810 				continue;
1811 			}
1812 		}
1813 		if (!LIST_EMPTY(gslp)) {
1814 			mtx_unlock(&sync_mtx);
1815 			mtx_lock(&Giant);
1816 			mtx_lock(&sync_mtx);
1817 			while (!LIST_EMPTY(gslp)) {
1818 				error = sync_vnode(gslp, &bo, td);
1819 				if (error == 1) {
1820 					LIST_REMOVE(bo, bo_synclist);
1821 					LIST_INSERT_HEAD(gnext, bo,
1822 					    bo_synclist);
1823 					continue;
1824 				}
1825 			}
1826 			mtx_unlock(&Giant);
1827 		}
1828 		if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1829 			syncer_final_iter--;
1830 		/*
1831 		 * The variable rushjob allows the kernel to speed up the
1832 		 * processing of the filesystem syncer process. A rushjob
1833 		 * value of N tells the filesystem syncer to process the next
1834 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1835 		 * is used by the soft update code to speed up the filesystem
1836 		 * syncer process when the incore state is getting so far
1837 		 * ahead of the disk that the kernel memory pool is being
1838 		 * threatened with exhaustion.
1839 		 */
1840 		if (rushjob > 0) {
1841 			rushjob -= 1;
1842 			continue;
1843 		}
1844 		/*
1845 		 * Just sleep for a short period of time between
1846 		 * iterations when shutting down to allow some I/O
1847 		 * to happen.
1848 		 *
1849 		 * If it has taken us less than a second to process the
1850 		 * current work, then wait. Otherwise start right over
1851 		 * again. We can still lose time if any single round
1852 		 * takes more than two seconds, but it does not really
1853 		 * matter as we are just trying to generally pace the
1854 		 * filesystem activity.
1855 		 */
1856 		if (syncer_state != SYNCER_RUNNING)
1857 			cv_timedwait(&sync_wakeup, &sync_mtx,
1858 			    hz / SYNCER_SHUTDOWN_SPEEDUP);
1859 		else if (time_uptime == starttime)
1860 			cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1861 	}
1862 }
1863 
1864 /*
1865  * Request the syncer daemon to speed up its work.
1866  * We never push it to speed up more than half of its
1867  * normal turn time, otherwise it could take over the cpu.
1868  */
1869 int
1870 speedup_syncer(void)
1871 {
1872 	int ret = 0;
1873 
1874 	mtx_lock(&sync_mtx);
1875 	if (rushjob < syncdelay / 2) {
1876 		rushjob += 1;
1877 		stat_rush_requests += 1;
1878 		ret = 1;
1879 	}
1880 	mtx_unlock(&sync_mtx);
1881 	cv_broadcast(&sync_wakeup);
1882 	return (ret);
1883 }
1884 
1885 /*
1886  * Tell the syncer to speed up its work and run though its work
1887  * list several times, then tell it to shut down.
1888  */
1889 static void
1890 syncer_shutdown(void *arg, int howto)
1891 {
1892 
1893 	if (howto & RB_NOSYNC)
1894 		return;
1895 	mtx_lock(&sync_mtx);
1896 	syncer_state = SYNCER_SHUTTING_DOWN;
1897 	rushjob = 0;
1898 	mtx_unlock(&sync_mtx);
1899 	cv_broadcast(&sync_wakeup);
1900 	kproc_shutdown(arg, howto);
1901 }
1902 
1903 /*
1904  * Reassign a buffer from one vnode to another.
1905  * Used to assign file specific control information
1906  * (indirect blocks) to the vnode to which they belong.
1907  */
1908 void
1909 reassignbuf(struct buf *bp)
1910 {
1911 	struct vnode *vp;
1912 	struct bufobj *bo;
1913 	int delay;
1914 #ifdef INVARIANTS
1915 	struct bufv *bv;
1916 #endif
1917 
1918 	vp = bp->b_vp;
1919 	bo = bp->b_bufobj;
1920 	++reassignbufcalls;
1921 
1922 	CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1923 	    bp, bp->b_vp, bp->b_flags);
1924 	/*
1925 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1926 	 * is not fully linked in.
1927 	 */
1928 	if (bp->b_flags & B_PAGING)
1929 		panic("cannot reassign paging buffer");
1930 
1931 	/*
1932 	 * Delete from old vnode list, if on one.
1933 	 */
1934 	BO_LOCK(bo);
1935 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1936 		buf_vlist_remove(bp);
1937 	else
1938 		panic("reassignbuf: Buffer %p not on queue.", bp);
1939 	/*
1940 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1941 	 * of clean buffers.
1942 	 */
1943 	if (bp->b_flags & B_DELWRI) {
1944 		if ((bo->bo_flag & BO_ONWORKLST) == 0) {
1945 			switch (vp->v_type) {
1946 			case VDIR:
1947 				delay = dirdelay;
1948 				break;
1949 			case VCHR:
1950 				delay = metadelay;
1951 				break;
1952 			default:
1953 				delay = filedelay;
1954 			}
1955 			vn_syncer_add_to_worklist(bo, delay);
1956 		}
1957 		buf_vlist_add(bp, bo, BX_VNDIRTY);
1958 	} else {
1959 		buf_vlist_add(bp, bo, BX_VNCLEAN);
1960 
1961 		if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1962 			mtx_lock(&sync_mtx);
1963 			LIST_REMOVE(bo, bo_synclist);
1964 			syncer_worklist_len--;
1965 			mtx_unlock(&sync_mtx);
1966 			bo->bo_flag &= ~BO_ONWORKLST;
1967 		}
1968 	}
1969 #ifdef INVARIANTS
1970 	bv = &bo->bo_clean;
1971 	bp = TAILQ_FIRST(&bv->bv_hd);
1972 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1973 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1974 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
1975 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1976 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1977 	bv = &bo->bo_dirty;
1978 	bp = TAILQ_FIRST(&bv->bv_hd);
1979 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1980 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1981 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
1982 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1983 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1984 #endif
1985 	BO_UNLOCK(bo);
1986 }
1987 
1988 /*
1989  * Increment the use and hold counts on the vnode, taking care to reference
1990  * the driver's usecount if this is a chardev.  The vholdl() will remove
1991  * the vnode from the free list if it is presently free.  Requires the
1992  * vnode interlock and returns with it held.
1993  */
1994 static void
1995 v_incr_usecount(struct vnode *vp)
1996 {
1997 
1998 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
1999 	vp->v_usecount++;
2000 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2001 		dev_lock();
2002 		vp->v_rdev->si_usecount++;
2003 		dev_unlock();
2004 	}
2005 	vholdl(vp);
2006 }
2007 
2008 /*
2009  * Turn a holdcnt into a use+holdcnt such that only one call to
2010  * v_decr_usecount is needed.
2011  */
2012 static void
2013 v_upgrade_usecount(struct vnode *vp)
2014 {
2015 
2016 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2017 	vp->v_usecount++;
2018 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2019 		dev_lock();
2020 		vp->v_rdev->si_usecount++;
2021 		dev_unlock();
2022 	}
2023 }
2024 
2025 /*
2026  * Decrement the vnode use and hold count along with the driver's usecount
2027  * if this is a chardev.  The vdropl() below releases the vnode interlock
2028  * as it may free the vnode.
2029  */
2030 static void
2031 v_decr_usecount(struct vnode *vp)
2032 {
2033 
2034 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2035 	VNASSERT(vp->v_usecount > 0, vp,
2036 	    ("v_decr_usecount: negative usecount"));
2037 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2038 	vp->v_usecount--;
2039 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2040 		dev_lock();
2041 		vp->v_rdev->si_usecount--;
2042 		dev_unlock();
2043 	}
2044 	vdropl(vp);
2045 }
2046 
2047 /*
2048  * Decrement only the use count and driver use count.  This is intended to
2049  * be paired with a follow on vdropl() to release the remaining hold count.
2050  * In this way we may vgone() a vnode with a 0 usecount without risk of
2051  * having it end up on a free list because the hold count is kept above 0.
2052  */
2053 static void
2054 v_decr_useonly(struct vnode *vp)
2055 {
2056 
2057 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2058 	VNASSERT(vp->v_usecount > 0, vp,
2059 	    ("v_decr_useonly: negative usecount"));
2060 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2061 	vp->v_usecount--;
2062 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2063 		dev_lock();
2064 		vp->v_rdev->si_usecount--;
2065 		dev_unlock();
2066 	}
2067 }
2068 
2069 /*
2070  * Grab a particular vnode from the free list, increment its
2071  * reference count and lock it.  VI_DOOMED is set if the vnode
2072  * is being destroyed.  Only callers who specify LK_RETRY will
2073  * see doomed vnodes.  If inactive processing was delayed in
2074  * vput try to do it here.
2075  */
2076 int
2077 vget(struct vnode *vp, int flags, struct thread *td)
2078 {
2079 	int error;
2080 
2081 	error = 0;
2082 	VFS_ASSERT_GIANT(vp->v_mount);
2083 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2084 	    ("vget: invalid lock operation"));
2085 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2086 
2087 	if ((flags & LK_INTERLOCK) == 0)
2088 		VI_LOCK(vp);
2089 	vholdl(vp);
2090 	if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2091 		vdrop(vp);
2092 		CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2093 		    vp);
2094 		return (error);
2095 	}
2096 	if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2097 		panic("vget: vn_lock failed to return ENOENT\n");
2098 	VI_LOCK(vp);
2099 	/* Upgrade our holdcnt to a usecount. */
2100 	v_upgrade_usecount(vp);
2101 	/*
2102 	 * We don't guarantee that any particular close will
2103 	 * trigger inactive processing so just make a best effort
2104 	 * here at preventing a reference to a removed file.  If
2105 	 * we don't succeed no harm is done.
2106 	 */
2107 	if (vp->v_iflag & VI_OWEINACT) {
2108 		if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2109 		    (flags & LK_NOWAIT) == 0)
2110 			vinactive(vp, td);
2111 		vp->v_iflag &= ~VI_OWEINACT;
2112 	}
2113 	VI_UNLOCK(vp);
2114 	return (0);
2115 }
2116 
2117 /*
2118  * Increase the reference count of a vnode.
2119  */
2120 void
2121 vref(struct vnode *vp)
2122 {
2123 
2124 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2125 	VI_LOCK(vp);
2126 	v_incr_usecount(vp);
2127 	VI_UNLOCK(vp);
2128 }
2129 
2130 /*
2131  * Return reference count of a vnode.
2132  *
2133  * The results of this call are only guaranteed when some mechanism other
2134  * than the VI lock is used to stop other processes from gaining references
2135  * to the vnode.  This may be the case if the caller holds the only reference.
2136  * This is also useful when stale data is acceptable as race conditions may
2137  * be accounted for by some other means.
2138  */
2139 int
2140 vrefcnt(struct vnode *vp)
2141 {
2142 	int usecnt;
2143 
2144 	VI_LOCK(vp);
2145 	usecnt = vp->v_usecount;
2146 	VI_UNLOCK(vp);
2147 
2148 	return (usecnt);
2149 }
2150 
2151 #define	VPUTX_VRELE	1
2152 #define	VPUTX_VPUT	2
2153 #define	VPUTX_VUNREF	3
2154 
2155 static void
2156 vputx(struct vnode *vp, int func)
2157 {
2158 	int error;
2159 
2160 	KASSERT(vp != NULL, ("vputx: null vp"));
2161 	if (func == VPUTX_VUNREF)
2162 		ASSERT_VOP_ELOCKED(vp, "vunref");
2163 	else if (func == VPUTX_VPUT)
2164 		ASSERT_VOP_LOCKED(vp, "vput");
2165 	else
2166 		KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2167 	VFS_ASSERT_GIANT(vp->v_mount);
2168 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2169 	VI_LOCK(vp);
2170 
2171 	/* Skip this v_writecount check if we're going to panic below. */
2172 	VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2173 	    ("vputx: missed vn_close"));
2174 	error = 0;
2175 
2176 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2177 	    vp->v_usecount == 1)) {
2178 		if (func == VPUTX_VPUT)
2179 			VOP_UNLOCK(vp, 0);
2180 		v_decr_usecount(vp);
2181 		return;
2182 	}
2183 
2184 	if (vp->v_usecount != 1) {
2185 #ifdef DIAGNOSTIC
2186 		vprint("vputx: negative ref count", vp);
2187 #endif
2188 		panic("vputx: negative ref cnt");
2189 	}
2190 	CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2191 	/*
2192 	 * We want to hold the vnode until the inactive finishes to
2193 	 * prevent vgone() races.  We drop the use count here and the
2194 	 * hold count below when we're done.
2195 	 */
2196 	v_decr_useonly(vp);
2197 	/*
2198 	 * We must call VOP_INACTIVE with the node locked. Mark
2199 	 * as VI_DOINGINACT to avoid recursion.
2200 	 */
2201 	vp->v_iflag |= VI_OWEINACT;
2202 	if (func == VPUTX_VRELE) {
2203 		error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2204 		VI_LOCK(vp);
2205 	} else if (func == VPUTX_VPUT && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2206 		error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK | LK_NOWAIT);
2207 		VI_LOCK(vp);
2208 	}
2209 	if (vp->v_usecount > 0)
2210 		vp->v_iflag &= ~VI_OWEINACT;
2211 	if (error == 0) {
2212 		if (vp->v_iflag & VI_OWEINACT)
2213 			vinactive(vp, curthread);
2214 		if (func != VPUTX_VUNREF)
2215 			VOP_UNLOCK(vp, 0);
2216 	}
2217 	vdropl(vp);
2218 }
2219 
2220 /*
2221  * Vnode put/release.
2222  * If count drops to zero, call inactive routine and return to freelist.
2223  */
2224 void
2225 vrele(struct vnode *vp)
2226 {
2227 
2228 	vputx(vp, VPUTX_VRELE);
2229 }
2230 
2231 /*
2232  * Release an already locked vnode.  This give the same effects as
2233  * unlock+vrele(), but takes less time and avoids releasing and
2234  * re-aquiring the lock (as vrele() acquires the lock internally.)
2235  */
2236 void
2237 vput(struct vnode *vp)
2238 {
2239 
2240 	vputx(vp, VPUTX_VPUT);
2241 }
2242 
2243 /*
2244  * Release an exclusively locked vnode. Do not unlock the vnode lock.
2245  */
2246 void
2247 vunref(struct vnode *vp)
2248 {
2249 
2250 	vputx(vp, VPUTX_VUNREF);
2251 }
2252 
2253 /*
2254  * Somebody doesn't want the vnode recycled.
2255  */
2256 void
2257 vhold(struct vnode *vp)
2258 {
2259 
2260 	VI_LOCK(vp);
2261 	vholdl(vp);
2262 	VI_UNLOCK(vp);
2263 }
2264 
2265 void
2266 vholdl(struct vnode *vp)
2267 {
2268 
2269 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2270 	vp->v_holdcnt++;
2271 	if (VSHOULDBUSY(vp))
2272 		vbusy(vp);
2273 }
2274 
2275 /*
2276  * Note that there is one less who cares about this vnode.  vdrop() is the
2277  * opposite of vhold().
2278  */
2279 void
2280 vdrop(struct vnode *vp)
2281 {
2282 
2283 	VI_LOCK(vp);
2284 	vdropl(vp);
2285 }
2286 
2287 /*
2288  * Drop the hold count of the vnode.  If this is the last reference to
2289  * the vnode we will free it if it has been vgone'd otherwise it is
2290  * placed on the free list.
2291  */
2292 void
2293 vdropl(struct vnode *vp)
2294 {
2295 
2296 	ASSERT_VI_LOCKED(vp, "vdropl");
2297 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2298 	if (vp->v_holdcnt <= 0)
2299 		panic("vdrop: holdcnt %d", vp->v_holdcnt);
2300 	vp->v_holdcnt--;
2301 	if (vp->v_holdcnt == 0) {
2302 		if (vp->v_iflag & VI_DOOMED) {
2303 			CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__,
2304 			    vp);
2305 			vdestroy(vp);
2306 			return;
2307 		} else
2308 			vfree(vp);
2309 	}
2310 	VI_UNLOCK(vp);
2311 }
2312 
2313 /*
2314  * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2315  * flags.  DOINGINACT prevents us from recursing in calls to vinactive.
2316  * OWEINACT tracks whether a vnode missed a call to inactive due to a
2317  * failed lock upgrade.
2318  */
2319 static void
2320 vinactive(struct vnode *vp, struct thread *td)
2321 {
2322 
2323 	ASSERT_VOP_ELOCKED(vp, "vinactive");
2324 	ASSERT_VI_LOCKED(vp, "vinactive");
2325 	VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2326 	    ("vinactive: recursed on VI_DOINGINACT"));
2327 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2328 	vp->v_iflag |= VI_DOINGINACT;
2329 	vp->v_iflag &= ~VI_OWEINACT;
2330 	VI_UNLOCK(vp);
2331 	VOP_INACTIVE(vp, td);
2332 	VI_LOCK(vp);
2333 	VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2334 	    ("vinactive: lost VI_DOINGINACT"));
2335 	vp->v_iflag &= ~VI_DOINGINACT;
2336 }
2337 
2338 /*
2339  * Remove any vnodes in the vnode table belonging to mount point mp.
2340  *
2341  * If FORCECLOSE is not specified, there should not be any active ones,
2342  * return error if any are found (nb: this is a user error, not a
2343  * system error). If FORCECLOSE is specified, detach any active vnodes
2344  * that are found.
2345  *
2346  * If WRITECLOSE is set, only flush out regular file vnodes open for
2347  * writing.
2348  *
2349  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2350  *
2351  * `rootrefs' specifies the base reference count for the root vnode
2352  * of this filesystem. The root vnode is considered busy if its
2353  * v_usecount exceeds this value. On a successful return, vflush(, td)
2354  * will call vrele() on the root vnode exactly rootrefs times.
2355  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2356  * be zero.
2357  */
2358 #ifdef DIAGNOSTIC
2359 static int busyprt = 0;		/* print out busy vnodes */
2360 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
2361 #endif
2362 
2363 int
2364 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
2365 {
2366 	struct vnode *vp, *mvp, *rootvp = NULL;
2367 	struct vattr vattr;
2368 	int busy = 0, error;
2369 
2370 	CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2371 	    rootrefs, flags);
2372 	if (rootrefs > 0) {
2373 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2374 		    ("vflush: bad args"));
2375 		/*
2376 		 * Get the filesystem root vnode. We can vput() it
2377 		 * immediately, since with rootrefs > 0, it won't go away.
2378 		 */
2379 		if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2380 			CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2381 			    __func__, error);
2382 			return (error);
2383 		}
2384 		vput(rootvp);
2385 	}
2386 	MNT_ILOCK(mp);
2387 loop:
2388 	MNT_VNODE_FOREACH(vp, mp, mvp) {
2389 		VI_LOCK(vp);
2390 		vholdl(vp);
2391 		MNT_IUNLOCK(mp);
2392 		error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2393 		if (error) {
2394 			vdrop(vp);
2395 			MNT_ILOCK(mp);
2396 			MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
2397 			goto loop;
2398 		}
2399 		/*
2400 		 * Skip over a vnodes marked VV_SYSTEM.
2401 		 */
2402 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2403 			VOP_UNLOCK(vp, 0);
2404 			vdrop(vp);
2405 			MNT_ILOCK(mp);
2406 			continue;
2407 		}
2408 		/*
2409 		 * If WRITECLOSE is set, flush out unlinked but still open
2410 		 * files (even if open only for reading) and regular file
2411 		 * vnodes open for writing.
2412 		 */
2413 		if (flags & WRITECLOSE) {
2414 			error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2415 			VI_LOCK(vp);
2416 
2417 			if ((vp->v_type == VNON ||
2418 			    (error == 0 && vattr.va_nlink > 0)) &&
2419 			    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2420 				VOP_UNLOCK(vp, 0);
2421 				vdropl(vp);
2422 				MNT_ILOCK(mp);
2423 				continue;
2424 			}
2425 		} else
2426 			VI_LOCK(vp);
2427 		/*
2428 		 * With v_usecount == 0, all we need to do is clear out the
2429 		 * vnode data structures and we are done.
2430 		 *
2431 		 * If FORCECLOSE is set, forcibly close the vnode.
2432 		 */
2433 		if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2434 			VNASSERT(vp->v_usecount == 0 ||
2435 			    (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2436 			    ("device VNODE %p is FORCECLOSED", vp));
2437 			vgonel(vp);
2438 		} else {
2439 			busy++;
2440 #ifdef DIAGNOSTIC
2441 			if (busyprt)
2442 				vprint("vflush: busy vnode", vp);
2443 #endif
2444 		}
2445 		VOP_UNLOCK(vp, 0);
2446 		vdropl(vp);
2447 		MNT_ILOCK(mp);
2448 	}
2449 	MNT_IUNLOCK(mp);
2450 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2451 		/*
2452 		 * If just the root vnode is busy, and if its refcount
2453 		 * is equal to `rootrefs', then go ahead and kill it.
2454 		 */
2455 		VI_LOCK(rootvp);
2456 		KASSERT(busy > 0, ("vflush: not busy"));
2457 		VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2458 		    ("vflush: usecount %d < rootrefs %d",
2459 		     rootvp->v_usecount, rootrefs));
2460 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2461 			VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2462 			vgone(rootvp);
2463 			VOP_UNLOCK(rootvp, 0);
2464 			busy = 0;
2465 		} else
2466 			VI_UNLOCK(rootvp);
2467 	}
2468 	if (busy) {
2469 		CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2470 		    busy);
2471 		return (EBUSY);
2472 	}
2473 	for (; rootrefs > 0; rootrefs--)
2474 		vrele(rootvp);
2475 	return (0);
2476 }
2477 
2478 /*
2479  * Recycle an unused vnode to the front of the free list.
2480  */
2481 int
2482 vrecycle(struct vnode *vp, struct thread *td)
2483 {
2484 	int recycled;
2485 
2486 	ASSERT_VOP_ELOCKED(vp, "vrecycle");
2487 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2488 	recycled = 0;
2489 	VI_LOCK(vp);
2490 	if (vp->v_usecount == 0) {
2491 		recycled = 1;
2492 		vgonel(vp);
2493 	}
2494 	VI_UNLOCK(vp);
2495 	return (recycled);
2496 }
2497 
2498 /*
2499  * Eliminate all activity associated with a vnode
2500  * in preparation for reuse.
2501  */
2502 void
2503 vgone(struct vnode *vp)
2504 {
2505 	VI_LOCK(vp);
2506 	vgonel(vp);
2507 	VI_UNLOCK(vp);
2508 }
2509 
2510 /*
2511  * vgone, with the vp interlock held.
2512  */
2513 void
2514 vgonel(struct vnode *vp)
2515 {
2516 	struct thread *td;
2517 	int oweinact;
2518 	int active;
2519 	struct mount *mp;
2520 
2521 	ASSERT_VOP_ELOCKED(vp, "vgonel");
2522 	ASSERT_VI_LOCKED(vp, "vgonel");
2523 	VNASSERT(vp->v_holdcnt, vp,
2524 	    ("vgonel: vp %p has no reference.", vp));
2525 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2526 	td = curthread;
2527 
2528 	/*
2529 	 * Don't vgonel if we're already doomed.
2530 	 */
2531 	if (vp->v_iflag & VI_DOOMED)
2532 		return;
2533 	vp->v_iflag |= VI_DOOMED;
2534 	/*
2535 	 * Check to see if the vnode is in use.  If so, we have to call
2536 	 * VOP_CLOSE() and VOP_INACTIVE().
2537 	 */
2538 	active = vp->v_usecount;
2539 	oweinact = (vp->v_iflag & VI_OWEINACT);
2540 	VI_UNLOCK(vp);
2541 	/*
2542 	 * Clean out any buffers associated with the vnode.
2543 	 * If the flush fails, just toss the buffers.
2544 	 */
2545 	mp = NULL;
2546 	if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2547 		(void) vn_start_secondary_write(vp, &mp, V_WAIT);
2548 	if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
2549 		vinvalbuf(vp, 0, 0, 0);
2550 
2551 	/*
2552 	 * If purging an active vnode, it must be closed and
2553 	 * deactivated before being reclaimed.
2554 	 */
2555 	if (active)
2556 		VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2557 	if (oweinact || active) {
2558 		VI_LOCK(vp);
2559 		if ((vp->v_iflag & VI_DOINGINACT) == 0)
2560 			vinactive(vp, td);
2561 		VI_UNLOCK(vp);
2562 	}
2563 	/*
2564 	 * Reclaim the vnode.
2565 	 */
2566 	if (VOP_RECLAIM(vp, td))
2567 		panic("vgone: cannot reclaim");
2568 	if (mp != NULL)
2569 		vn_finished_secondary_write(mp);
2570 	VNASSERT(vp->v_object == NULL, vp,
2571 	    ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2572 	/*
2573 	 * Clear the advisory locks and wake up waiting threads.
2574 	 */
2575 	(void)VOP_ADVLOCKPURGE(vp);
2576 	/*
2577 	 * Delete from old mount point vnode list.
2578 	 */
2579 	delmntque(vp);
2580 	cache_purge(vp);
2581 	/*
2582 	 * Done with purge, reset to the standard lock and invalidate
2583 	 * the vnode.
2584 	 */
2585 	VI_LOCK(vp);
2586 	vp->v_vnlock = &vp->v_lock;
2587 	vp->v_op = &dead_vnodeops;
2588 	vp->v_tag = "none";
2589 	vp->v_type = VBAD;
2590 }
2591 
2592 /*
2593  * Calculate the total number of references to a special device.
2594  */
2595 int
2596 vcount(struct vnode *vp)
2597 {
2598 	int count;
2599 
2600 	dev_lock();
2601 	count = vp->v_rdev->si_usecount;
2602 	dev_unlock();
2603 	return (count);
2604 }
2605 
2606 /*
2607  * Same as above, but using the struct cdev *as argument
2608  */
2609 int
2610 count_dev(struct cdev *dev)
2611 {
2612 	int count;
2613 
2614 	dev_lock();
2615 	count = dev->si_usecount;
2616 	dev_unlock();
2617 	return(count);
2618 }
2619 
2620 /*
2621  * Print out a description of a vnode.
2622  */
2623 static char *typename[] =
2624 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2625  "VMARKER"};
2626 
2627 void
2628 vn_printf(struct vnode *vp, const char *fmt, ...)
2629 {
2630 	va_list ap;
2631 	char buf[256], buf2[16];
2632 	u_long flags;
2633 
2634 	va_start(ap, fmt);
2635 	vprintf(fmt, ap);
2636 	va_end(ap);
2637 	printf("%p: ", (void *)vp);
2638 	printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2639 	printf("    usecount %d, writecount %d, refcount %d mountedhere %p\n",
2640 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2641 	buf[0] = '\0';
2642 	buf[1] = '\0';
2643 	if (vp->v_vflag & VV_ROOT)
2644 		strlcat(buf, "|VV_ROOT", sizeof(buf));
2645 	if (vp->v_vflag & VV_ISTTY)
2646 		strlcat(buf, "|VV_ISTTY", sizeof(buf));
2647 	if (vp->v_vflag & VV_NOSYNC)
2648 		strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2649 	if (vp->v_vflag & VV_CACHEDLABEL)
2650 		strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2651 	if (vp->v_vflag & VV_TEXT)
2652 		strlcat(buf, "|VV_TEXT", sizeof(buf));
2653 	if (vp->v_vflag & VV_COPYONWRITE)
2654 		strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2655 	if (vp->v_vflag & VV_SYSTEM)
2656 		strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2657 	if (vp->v_vflag & VV_PROCDEP)
2658 		strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2659 	if (vp->v_vflag & VV_NOKNOTE)
2660 		strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2661 	if (vp->v_vflag & VV_DELETED)
2662 		strlcat(buf, "|VV_DELETED", sizeof(buf));
2663 	if (vp->v_vflag & VV_MD)
2664 		strlcat(buf, "|VV_MD", sizeof(buf));
2665 	flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC |
2666 	    VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2667 	    VV_NOKNOTE | VV_DELETED | VV_MD);
2668 	if (flags != 0) {
2669 		snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2670 		strlcat(buf, buf2, sizeof(buf));
2671 	}
2672 	if (vp->v_iflag & VI_MOUNT)
2673 		strlcat(buf, "|VI_MOUNT", sizeof(buf));
2674 	if (vp->v_iflag & VI_AGE)
2675 		strlcat(buf, "|VI_AGE", sizeof(buf));
2676 	if (vp->v_iflag & VI_DOOMED)
2677 		strlcat(buf, "|VI_DOOMED", sizeof(buf));
2678 	if (vp->v_iflag & VI_FREE)
2679 		strlcat(buf, "|VI_FREE", sizeof(buf));
2680 	if (vp->v_iflag & VI_DOINGINACT)
2681 		strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2682 	if (vp->v_iflag & VI_OWEINACT)
2683 		strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2684 	flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2685 	    VI_DOINGINACT | VI_OWEINACT);
2686 	if (flags != 0) {
2687 		snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2688 		strlcat(buf, buf2, sizeof(buf));
2689 	}
2690 	printf("    flags (%s)\n", buf + 1);
2691 	if (mtx_owned(VI_MTX(vp)))
2692 		printf(" VI_LOCKed");
2693 	if (vp->v_object != NULL)
2694 		printf("    v_object %p ref %d pages %d\n",
2695 		    vp->v_object, vp->v_object->ref_count,
2696 		    vp->v_object->resident_page_count);
2697 	printf("    ");
2698 	lockmgr_printinfo(vp->v_vnlock);
2699 	if (vp->v_data != NULL)
2700 		VOP_PRINT(vp);
2701 }
2702 
2703 #ifdef DDB
2704 /*
2705  * List all of the locked vnodes in the system.
2706  * Called when debugging the kernel.
2707  */
2708 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2709 {
2710 	struct mount *mp, *nmp;
2711 	struct vnode *vp;
2712 
2713 	/*
2714 	 * Note: because this is DDB, we can't obey the locking semantics
2715 	 * for these structures, which means we could catch an inconsistent
2716 	 * state and dereference a nasty pointer.  Not much to be done
2717 	 * about that.
2718 	 */
2719 	db_printf("Locked vnodes\n");
2720 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2721 		nmp = TAILQ_NEXT(mp, mnt_list);
2722 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2723 			if (vp->v_type != VMARKER &&
2724 			    VOP_ISLOCKED(vp))
2725 				vprint("", vp);
2726 		}
2727 		nmp = TAILQ_NEXT(mp, mnt_list);
2728 	}
2729 }
2730 
2731 /*
2732  * Show details about the given vnode.
2733  */
2734 DB_SHOW_COMMAND(vnode, db_show_vnode)
2735 {
2736 	struct vnode *vp;
2737 
2738 	if (!have_addr)
2739 		return;
2740 	vp = (struct vnode *)addr;
2741 	vn_printf(vp, "vnode ");
2742 }
2743 
2744 /*
2745  * Show details about the given mount point.
2746  */
2747 DB_SHOW_COMMAND(mount, db_show_mount)
2748 {
2749 	struct mount *mp;
2750 	struct vfsopt *opt;
2751 	struct statfs *sp;
2752 	struct vnode *vp;
2753 	char buf[512];
2754 	u_int flags;
2755 
2756 	if (!have_addr) {
2757 		/* No address given, print short info about all mount points. */
2758 		TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2759 			db_printf("%p %s on %s (%s)\n", mp,
2760 			    mp->mnt_stat.f_mntfromname,
2761 			    mp->mnt_stat.f_mntonname,
2762 			    mp->mnt_stat.f_fstypename);
2763 			if (db_pager_quit)
2764 				break;
2765 		}
2766 		db_printf("\nMore info: show mount <addr>\n");
2767 		return;
2768 	}
2769 
2770 	mp = (struct mount *)addr;
2771 	db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
2772 	    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
2773 
2774 	buf[0] = '\0';
2775 	flags = mp->mnt_flag;
2776 #define	MNT_FLAG(flag)	do {						\
2777 	if (flags & (flag)) {						\
2778 		if (buf[0] != '\0')					\
2779 			strlcat(buf, ", ", sizeof(buf));		\
2780 		strlcat(buf, (#flag) + 4, sizeof(buf));			\
2781 		flags &= ~(flag);					\
2782 	}								\
2783 } while (0)
2784 	MNT_FLAG(MNT_RDONLY);
2785 	MNT_FLAG(MNT_SYNCHRONOUS);
2786 	MNT_FLAG(MNT_NOEXEC);
2787 	MNT_FLAG(MNT_NOSUID);
2788 	MNT_FLAG(MNT_UNION);
2789 	MNT_FLAG(MNT_ASYNC);
2790 	MNT_FLAG(MNT_SUIDDIR);
2791 	MNT_FLAG(MNT_SOFTDEP);
2792 	MNT_FLAG(MNT_NOSYMFOLLOW);
2793 	MNT_FLAG(MNT_GJOURNAL);
2794 	MNT_FLAG(MNT_MULTILABEL);
2795 	MNT_FLAG(MNT_ACLS);
2796 	MNT_FLAG(MNT_NOATIME);
2797 	MNT_FLAG(MNT_NOCLUSTERR);
2798 	MNT_FLAG(MNT_NOCLUSTERW);
2799 	MNT_FLAG(MNT_NFS4ACLS);
2800 	MNT_FLAG(MNT_EXRDONLY);
2801 	MNT_FLAG(MNT_EXPORTED);
2802 	MNT_FLAG(MNT_DEFEXPORTED);
2803 	MNT_FLAG(MNT_EXPORTANON);
2804 	MNT_FLAG(MNT_EXKERB);
2805 	MNT_FLAG(MNT_EXPUBLIC);
2806 	MNT_FLAG(MNT_LOCAL);
2807 	MNT_FLAG(MNT_QUOTA);
2808 	MNT_FLAG(MNT_ROOTFS);
2809 	MNT_FLAG(MNT_USER);
2810 	MNT_FLAG(MNT_IGNORE);
2811 	MNT_FLAG(MNT_UPDATE);
2812 	MNT_FLAG(MNT_DELEXPORT);
2813 	MNT_FLAG(MNT_RELOAD);
2814 	MNT_FLAG(MNT_FORCE);
2815 	MNT_FLAG(MNT_SNAPSHOT);
2816 	MNT_FLAG(MNT_BYFSID);
2817 	MNT_FLAG(MNT_SOFTDEP);
2818 #undef MNT_FLAG
2819 	if (flags != 0) {
2820 		if (buf[0] != '\0')
2821 			strlcat(buf, ", ", sizeof(buf));
2822 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2823 		    "0x%08x", flags);
2824 	}
2825 	db_printf("    mnt_flag = %s\n", buf);
2826 
2827 	buf[0] = '\0';
2828 	flags = mp->mnt_kern_flag;
2829 #define	MNT_KERN_FLAG(flag)	do {					\
2830 	if (flags & (flag)) {						\
2831 		if (buf[0] != '\0')					\
2832 			strlcat(buf, ", ", sizeof(buf));		\
2833 		strlcat(buf, (#flag) + 5, sizeof(buf));			\
2834 		flags &= ~(flag);					\
2835 	}								\
2836 } while (0)
2837 	MNT_KERN_FLAG(MNTK_UNMOUNTF);
2838 	MNT_KERN_FLAG(MNTK_ASYNC);
2839 	MNT_KERN_FLAG(MNTK_SOFTDEP);
2840 	MNT_KERN_FLAG(MNTK_NOINSMNTQ);
2841 	MNT_KERN_FLAG(MNTK_DRAINING);
2842 	MNT_KERN_FLAG(MNTK_REFEXPIRE);
2843 	MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
2844 	MNT_KERN_FLAG(MNTK_SHARED_WRITES);
2845 	MNT_KERN_FLAG(MNTK_UNMOUNT);
2846 	MNT_KERN_FLAG(MNTK_MWAIT);
2847 	MNT_KERN_FLAG(MNTK_SUSPEND);
2848 	MNT_KERN_FLAG(MNTK_SUSPEND2);
2849 	MNT_KERN_FLAG(MNTK_SUSPENDED);
2850 	MNT_KERN_FLAG(MNTK_MPSAFE);
2851 	MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
2852 	MNT_KERN_FLAG(MNTK_NOKNOTE);
2853 #undef MNT_KERN_FLAG
2854 	if (flags != 0) {
2855 		if (buf[0] != '\0')
2856 			strlcat(buf, ", ", sizeof(buf));
2857 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2858 		    "0x%08x", flags);
2859 	}
2860 	db_printf("    mnt_kern_flag = %s\n", buf);
2861 
2862 	db_printf("    mnt_opt = ");
2863 	opt = TAILQ_FIRST(mp->mnt_opt);
2864 	if (opt != NULL) {
2865 		db_printf("%s", opt->name);
2866 		opt = TAILQ_NEXT(opt, link);
2867 		while (opt != NULL) {
2868 			db_printf(", %s", opt->name);
2869 			opt = TAILQ_NEXT(opt, link);
2870 		}
2871 	}
2872 	db_printf("\n");
2873 
2874 	sp = &mp->mnt_stat;
2875 	db_printf("    mnt_stat = { version=%u type=%u flags=0x%016jx "
2876 	    "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
2877 	    "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
2878 	    "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
2879 	    (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
2880 	    (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
2881 	    (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
2882 	    (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
2883 	    (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
2884 	    (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
2885 	    (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
2886 	    (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
2887 
2888 	db_printf("    mnt_cred = { uid=%u ruid=%u",
2889 	    (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
2890 	if (jailed(mp->mnt_cred))
2891 		db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
2892 	db_printf(" }\n");
2893 	db_printf("    mnt_ref = %d\n", mp->mnt_ref);
2894 	db_printf("    mnt_gen = %d\n", mp->mnt_gen);
2895 	db_printf("    mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
2896 	db_printf("    mnt_writeopcount = %d\n", mp->mnt_writeopcount);
2897 	db_printf("    mnt_noasync = %u\n", mp->mnt_noasync);
2898 	db_printf("    mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
2899 	db_printf("    mnt_iosize_max = %d\n", mp->mnt_iosize_max);
2900 	db_printf("    mnt_hashseed = %u\n", mp->mnt_hashseed);
2901 	db_printf("    mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
2902 	db_printf("    mnt_secondary_accwrites = %d\n",
2903 	    mp->mnt_secondary_accwrites);
2904 	db_printf("    mnt_gjprovider = %s\n",
2905 	    mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
2906 	db_printf("\n");
2907 
2908 	TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2909 		if (vp->v_type != VMARKER) {
2910 			vn_printf(vp, "vnode ");
2911 			if (db_pager_quit)
2912 				break;
2913 		}
2914 	}
2915 }
2916 #endif	/* DDB */
2917 
2918 /*
2919  * Fill in a struct xvfsconf based on a struct vfsconf.
2920  */
2921 static void
2922 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2923 {
2924 
2925 	strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2926 	xvfsp->vfc_typenum = vfsp->vfc_typenum;
2927 	xvfsp->vfc_refcount = vfsp->vfc_refcount;
2928 	xvfsp->vfc_flags = vfsp->vfc_flags;
2929 	/*
2930 	 * These are unused in userland, we keep them
2931 	 * to not break binary compatibility.
2932 	 */
2933 	xvfsp->vfc_vfsops = NULL;
2934 	xvfsp->vfc_next = NULL;
2935 }
2936 
2937 /*
2938  * Top level filesystem related information gathering.
2939  */
2940 static int
2941 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2942 {
2943 	struct vfsconf *vfsp;
2944 	struct xvfsconf xvfsp;
2945 	int error;
2946 
2947 	error = 0;
2948 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2949 		bzero(&xvfsp, sizeof(xvfsp));
2950 		vfsconf2x(vfsp, &xvfsp);
2951 		error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
2952 		if (error)
2953 			break;
2954 	}
2955 	return (error);
2956 }
2957 
2958 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
2959     "S,xvfsconf", "List of all configured filesystems");
2960 
2961 #ifndef BURN_BRIDGES
2962 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2963 
2964 static int
2965 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2966 {
2967 	int *name = (int *)arg1 - 1;	/* XXX */
2968 	u_int namelen = arg2 + 1;	/* XXX */
2969 	struct vfsconf *vfsp;
2970 	struct xvfsconf xvfsp;
2971 
2972 	printf("WARNING: userland calling deprecated sysctl, "
2973 	    "please rebuild world\n");
2974 
2975 #if 1 || defined(COMPAT_PRELITE2)
2976 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2977 	if (namelen == 1)
2978 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2979 #endif
2980 
2981 	switch (name[1]) {
2982 	case VFS_MAXTYPENUM:
2983 		if (namelen != 2)
2984 			return (ENOTDIR);
2985 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2986 	case VFS_CONF:
2987 		if (namelen != 3)
2988 			return (ENOTDIR);	/* overloaded */
2989 		TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
2990 			if (vfsp->vfc_typenum == name[2])
2991 				break;
2992 		if (vfsp == NULL)
2993 			return (EOPNOTSUPP);
2994 		bzero(&xvfsp, sizeof(xvfsp));
2995 		vfsconf2x(vfsp, &xvfsp);
2996 		return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
2997 	}
2998 	return (EOPNOTSUPP);
2999 }
3000 
3001 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
3002 	vfs_sysctl, "Generic filesystem");
3003 
3004 #if 1 || defined(COMPAT_PRELITE2)
3005 
3006 static int
3007 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3008 {
3009 	int error;
3010 	struct vfsconf *vfsp;
3011 	struct ovfsconf ovfs;
3012 
3013 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3014 		bzero(&ovfs, sizeof(ovfs));
3015 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
3016 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
3017 		ovfs.vfc_index = vfsp->vfc_typenum;
3018 		ovfs.vfc_refcount = vfsp->vfc_refcount;
3019 		ovfs.vfc_flags = vfsp->vfc_flags;
3020 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3021 		if (error)
3022 			return error;
3023 	}
3024 	return 0;
3025 }
3026 
3027 #endif /* 1 || COMPAT_PRELITE2 */
3028 #endif /* !BURN_BRIDGES */
3029 
3030 #define KINFO_VNODESLOP		10
3031 #ifdef notyet
3032 /*
3033  * Dump vnode list (via sysctl).
3034  */
3035 /* ARGSUSED */
3036 static int
3037 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3038 {
3039 	struct xvnode *xvn;
3040 	struct mount *mp;
3041 	struct vnode *vp;
3042 	int error, len, n;
3043 
3044 	/*
3045 	 * Stale numvnodes access is not fatal here.
3046 	 */
3047 	req->lock = 0;
3048 	len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3049 	if (!req->oldptr)
3050 		/* Make an estimate */
3051 		return (SYSCTL_OUT(req, 0, len));
3052 
3053 	error = sysctl_wire_old_buffer(req, 0);
3054 	if (error != 0)
3055 		return (error);
3056 	xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3057 	n = 0;
3058 	mtx_lock(&mountlist_mtx);
3059 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3060 		if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3061 			continue;
3062 		MNT_ILOCK(mp);
3063 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3064 			if (n == len)
3065 				break;
3066 			vref(vp);
3067 			xvn[n].xv_size = sizeof *xvn;
3068 			xvn[n].xv_vnode = vp;
3069 			xvn[n].xv_id = 0;	/* XXX compat */
3070 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3071 			XV_COPY(usecount);
3072 			XV_COPY(writecount);
3073 			XV_COPY(holdcnt);
3074 			XV_COPY(mount);
3075 			XV_COPY(numoutput);
3076 			XV_COPY(type);
3077 #undef XV_COPY
3078 			xvn[n].xv_flag = vp->v_vflag;
3079 
3080 			switch (vp->v_type) {
3081 			case VREG:
3082 			case VDIR:
3083 			case VLNK:
3084 				break;
3085 			case VBLK:
3086 			case VCHR:
3087 				if (vp->v_rdev == NULL) {
3088 					vrele(vp);
3089 					continue;
3090 				}
3091 				xvn[n].xv_dev = dev2udev(vp->v_rdev);
3092 				break;
3093 			case VSOCK:
3094 				xvn[n].xv_socket = vp->v_socket;
3095 				break;
3096 			case VFIFO:
3097 				xvn[n].xv_fifo = vp->v_fifoinfo;
3098 				break;
3099 			case VNON:
3100 			case VBAD:
3101 			default:
3102 				/* shouldn't happen? */
3103 				vrele(vp);
3104 				continue;
3105 			}
3106 			vrele(vp);
3107 			++n;
3108 		}
3109 		MNT_IUNLOCK(mp);
3110 		mtx_lock(&mountlist_mtx);
3111 		vfs_unbusy(mp);
3112 		if (n == len)
3113 			break;
3114 	}
3115 	mtx_unlock(&mountlist_mtx);
3116 
3117 	error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3118 	free(xvn, M_TEMP);
3119 	return (error);
3120 }
3121 
3122 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3123 	0, 0, sysctl_vnode, "S,xvnode", "");
3124 #endif
3125 
3126 /*
3127  * Unmount all filesystems. The list is traversed in reverse order
3128  * of mounting to avoid dependencies.
3129  */
3130 void
3131 vfs_unmountall(void)
3132 {
3133 	struct mount *mp;
3134 	struct thread *td;
3135 	int error;
3136 
3137 	KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
3138 	CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3139 	td = curthread;
3140 
3141 	/*
3142 	 * Since this only runs when rebooting, it is not interlocked.
3143 	 */
3144 	while(!TAILQ_EMPTY(&mountlist)) {
3145 		mp = TAILQ_LAST(&mountlist, mntlist);
3146 		error = dounmount(mp, MNT_FORCE, td);
3147 		if (error) {
3148 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
3149 			/*
3150 			 * XXX: Due to the way in which we mount the root
3151 			 * file system off of devfs, devfs will generate a
3152 			 * "busy" warning when we try to unmount it before
3153 			 * the root.  Don't print a warning as a result in
3154 			 * order to avoid false positive errors that may
3155 			 * cause needless upset.
3156 			 */
3157 			if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3158 				printf("unmount of %s failed (",
3159 				    mp->mnt_stat.f_mntonname);
3160 				if (error == EBUSY)
3161 					printf("BUSY)\n");
3162 				else
3163 					printf("%d)\n", error);
3164 			}
3165 		} else {
3166 			/* The unmount has removed mp from the mountlist */
3167 		}
3168 	}
3169 }
3170 
3171 /*
3172  * perform msync on all vnodes under a mount point
3173  * the mount point must be locked.
3174  */
3175 void
3176 vfs_msync(struct mount *mp, int flags)
3177 {
3178 	struct vnode *vp, *mvp;
3179 	struct vm_object *obj;
3180 
3181 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3182 	MNT_ILOCK(mp);
3183 	MNT_VNODE_FOREACH(vp, mp, mvp) {
3184 		VI_LOCK(vp);
3185 		obj = vp->v_object;
3186 		if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3187 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3188 			MNT_IUNLOCK(mp);
3189 			if (!vget(vp,
3190 			    LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3191 			    curthread)) {
3192 				if (vp->v_vflag & VV_NOSYNC) {	/* unlinked */
3193 					vput(vp);
3194 					MNT_ILOCK(mp);
3195 					continue;
3196 				}
3197 
3198 				obj = vp->v_object;
3199 				if (obj != NULL) {
3200 					VM_OBJECT_LOCK(obj);
3201 					vm_object_page_clean(obj, 0, 0,
3202 					    flags == MNT_WAIT ?
3203 					    OBJPC_SYNC : OBJPC_NOSYNC);
3204 					VM_OBJECT_UNLOCK(obj);
3205 				}
3206 				vput(vp);
3207 			}
3208 			MNT_ILOCK(mp);
3209 		} else
3210 			VI_UNLOCK(vp);
3211 	}
3212 	MNT_IUNLOCK(mp);
3213 }
3214 
3215 /*
3216  * Mark a vnode as free, putting it up for recycling.
3217  */
3218 static void
3219 vfree(struct vnode *vp)
3220 {
3221 
3222 	ASSERT_VI_LOCKED(vp, "vfree");
3223 	mtx_lock(&vnode_free_list_mtx);
3224 	VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed."));
3225 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free"));
3226 	VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't"));
3227 	VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp,
3228 	    ("vfree: Freeing doomed vnode"));
3229 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3230 	if (vp->v_iflag & VI_AGE) {
3231 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3232 	} else {
3233 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3234 	}
3235 	freevnodes++;
3236 	vp->v_iflag &= ~VI_AGE;
3237 	vp->v_iflag |= VI_FREE;
3238 	mtx_unlock(&vnode_free_list_mtx);
3239 }
3240 
3241 /*
3242  * Opposite of vfree() - mark a vnode as in use.
3243  */
3244 static void
3245 vbusy(struct vnode *vp)
3246 {
3247 	ASSERT_VI_LOCKED(vp, "vbusy");
3248 	VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
3249 	VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed."));
3250 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3251 
3252 	mtx_lock(&vnode_free_list_mtx);
3253 	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3254 	freevnodes--;
3255 	vp->v_iflag &= ~(VI_FREE|VI_AGE);
3256 	mtx_unlock(&vnode_free_list_mtx);
3257 }
3258 
3259 static void
3260 destroy_vpollinfo(struct vpollinfo *vi)
3261 {
3262 	knlist_destroy(&vi->vpi_selinfo.si_note);
3263 	mtx_destroy(&vi->vpi_lock);
3264 	uma_zfree(vnodepoll_zone, vi);
3265 }
3266 
3267 /*
3268  * Initalize per-vnode helper structure to hold poll-related state.
3269  */
3270 void
3271 v_addpollinfo(struct vnode *vp)
3272 {
3273 	struct vpollinfo *vi;
3274 
3275 	if (vp->v_pollinfo != NULL)
3276 		return;
3277 	vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3278 	mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3279 	knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3280 	    vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3281 	VI_LOCK(vp);
3282 	if (vp->v_pollinfo != NULL) {
3283 		VI_UNLOCK(vp);
3284 		destroy_vpollinfo(vi);
3285 		return;
3286 	}
3287 	vp->v_pollinfo = vi;
3288 	VI_UNLOCK(vp);
3289 }
3290 
3291 /*
3292  * Record a process's interest in events which might happen to
3293  * a vnode.  Because poll uses the historic select-style interface
3294  * internally, this routine serves as both the ``check for any
3295  * pending events'' and the ``record my interest in future events''
3296  * functions.  (These are done together, while the lock is held,
3297  * to avoid race conditions.)
3298  */
3299 int
3300 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3301 {
3302 
3303 	v_addpollinfo(vp);
3304 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3305 	if (vp->v_pollinfo->vpi_revents & events) {
3306 		/*
3307 		 * This leaves events we are not interested
3308 		 * in available for the other process which
3309 		 * which presumably had requested them
3310 		 * (otherwise they would never have been
3311 		 * recorded).
3312 		 */
3313 		events &= vp->v_pollinfo->vpi_revents;
3314 		vp->v_pollinfo->vpi_revents &= ~events;
3315 
3316 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
3317 		return (events);
3318 	}
3319 	vp->v_pollinfo->vpi_events |= events;
3320 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3321 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3322 	return (0);
3323 }
3324 
3325 /*
3326  * Routine to create and manage a filesystem syncer vnode.
3327  */
3328 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
3329 static int	sync_fsync(struct  vop_fsync_args *);
3330 static int	sync_inactive(struct  vop_inactive_args *);
3331 static int	sync_reclaim(struct  vop_reclaim_args *);
3332 
3333 static struct vop_vector sync_vnodeops = {
3334 	.vop_bypass =	VOP_EOPNOTSUPP,
3335 	.vop_close =	sync_close,		/* close */
3336 	.vop_fsync =	sync_fsync,		/* fsync */
3337 	.vop_inactive =	sync_inactive,	/* inactive */
3338 	.vop_reclaim =	sync_reclaim,	/* reclaim */
3339 	.vop_lock1 =	vop_stdlock,	/* lock */
3340 	.vop_unlock =	vop_stdunlock,	/* unlock */
3341 	.vop_islocked =	vop_stdislocked,	/* islocked */
3342 };
3343 
3344 /*
3345  * Create a new filesystem syncer vnode for the specified mount point.
3346  */
3347 int
3348 vfs_allocate_syncvnode(struct mount *mp)
3349 {
3350 	struct vnode *vp;
3351 	struct bufobj *bo;
3352 	static long start, incr, next;
3353 	int error;
3354 
3355 	/* Allocate a new vnode */
3356 	if ((error = getnewvnode("syncer", mp, &sync_vnodeops, &vp)) != 0) {
3357 		mp->mnt_syncer = NULL;
3358 		return (error);
3359 	}
3360 	vp->v_type = VNON;
3361 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3362 	vp->v_vflag |= VV_FORCEINSMQ;
3363 	error = insmntque(vp, mp);
3364 	if (error != 0)
3365 		panic("vfs_allocate_syncvnode: insmntque failed");
3366 	vp->v_vflag &= ~VV_FORCEINSMQ;
3367 	VOP_UNLOCK(vp, 0);
3368 	/*
3369 	 * Place the vnode onto the syncer worklist. We attempt to
3370 	 * scatter them about on the list so that they will go off
3371 	 * at evenly distributed times even if all the filesystems
3372 	 * are mounted at once.
3373 	 */
3374 	next += incr;
3375 	if (next == 0 || next > syncer_maxdelay) {
3376 		start /= 2;
3377 		incr /= 2;
3378 		if (start == 0) {
3379 			start = syncer_maxdelay / 2;
3380 			incr = syncer_maxdelay;
3381 		}
3382 		next = start;
3383 	}
3384 	bo = &vp->v_bufobj;
3385 	BO_LOCK(bo);
3386 	vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3387 	/* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3388 	mtx_lock(&sync_mtx);
3389 	sync_vnode_count++;
3390 	mtx_unlock(&sync_mtx);
3391 	BO_UNLOCK(bo);
3392 	mp->mnt_syncer = vp;
3393 	return (0);
3394 }
3395 
3396 /*
3397  * Do a lazy sync of the filesystem.
3398  */
3399 static int
3400 sync_fsync(struct vop_fsync_args *ap)
3401 {
3402 	struct vnode *syncvp = ap->a_vp;
3403 	struct mount *mp = syncvp->v_mount;
3404 	int error;
3405 	struct bufobj *bo;
3406 
3407 	/*
3408 	 * We only need to do something if this is a lazy evaluation.
3409 	 */
3410 	if (ap->a_waitfor != MNT_LAZY)
3411 		return (0);
3412 
3413 	/*
3414 	 * Move ourselves to the back of the sync list.
3415 	 */
3416 	bo = &syncvp->v_bufobj;
3417 	BO_LOCK(bo);
3418 	vn_syncer_add_to_worklist(bo, syncdelay);
3419 	BO_UNLOCK(bo);
3420 
3421 	/*
3422 	 * Walk the list of vnodes pushing all that are dirty and
3423 	 * not already on the sync list.
3424 	 */
3425 	mtx_lock(&mountlist_mtx);
3426 	if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) {
3427 		mtx_unlock(&mountlist_mtx);
3428 		return (0);
3429 	}
3430 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3431 		vfs_unbusy(mp);
3432 		return (0);
3433 	}
3434 	MNT_ILOCK(mp);
3435 	mp->mnt_noasync++;
3436 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
3437 	MNT_IUNLOCK(mp);
3438 	vfs_msync(mp, MNT_NOWAIT);
3439 	error = VFS_SYNC(mp, MNT_LAZY);
3440 	MNT_ILOCK(mp);
3441 	mp->mnt_noasync--;
3442 	if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
3443 		mp->mnt_kern_flag |= MNTK_ASYNC;
3444 	MNT_IUNLOCK(mp);
3445 	vn_finished_write(mp);
3446 	vfs_unbusy(mp);
3447 	return (error);
3448 }
3449 
3450 /*
3451  * The syncer vnode is no referenced.
3452  */
3453 static int
3454 sync_inactive(struct vop_inactive_args *ap)
3455 {
3456 
3457 	vgone(ap->a_vp);
3458 	return (0);
3459 }
3460 
3461 /*
3462  * The syncer vnode is no longer needed and is being decommissioned.
3463  *
3464  * Modifications to the worklist must be protected by sync_mtx.
3465  */
3466 static int
3467 sync_reclaim(struct vop_reclaim_args *ap)
3468 {
3469 	struct vnode *vp = ap->a_vp;
3470 	struct bufobj *bo;
3471 
3472 	bo = &vp->v_bufobj;
3473 	BO_LOCK(bo);
3474 	vp->v_mount->mnt_syncer = NULL;
3475 	if (bo->bo_flag & BO_ONWORKLST) {
3476 		mtx_lock(&sync_mtx);
3477 		LIST_REMOVE(bo, bo_synclist);
3478 		syncer_worklist_len--;
3479 		sync_vnode_count--;
3480 		mtx_unlock(&sync_mtx);
3481 		bo->bo_flag &= ~BO_ONWORKLST;
3482 	}
3483 	BO_UNLOCK(bo);
3484 
3485 	return (0);
3486 }
3487 
3488 /*
3489  * Check if vnode represents a disk device
3490  */
3491 int
3492 vn_isdisk(struct vnode *vp, int *errp)
3493 {
3494 	int error;
3495 
3496 	error = 0;
3497 	dev_lock();
3498 	if (vp->v_type != VCHR)
3499 		error = ENOTBLK;
3500 	else if (vp->v_rdev == NULL)
3501 		error = ENXIO;
3502 	else if (vp->v_rdev->si_devsw == NULL)
3503 		error = ENXIO;
3504 	else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3505 		error = ENOTBLK;
3506 	dev_unlock();
3507 	if (errp != NULL)
3508 		*errp = error;
3509 	return (error == 0);
3510 }
3511 
3512 /*
3513  * Common filesystem object access control check routine.  Accepts a
3514  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3515  * and optional call-by-reference privused argument allowing vaccess()
3516  * to indicate to the caller whether privilege was used to satisfy the
3517  * request (obsoleted).  Returns 0 on success, or an errno on failure.
3518  *
3519  * The ifdef'd CAPABILITIES version is here for reference, but is not
3520  * actually used.
3521  */
3522 int
3523 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3524     accmode_t accmode, struct ucred *cred, int *privused)
3525 {
3526 	accmode_t dac_granted;
3527 	accmode_t priv_granted;
3528 
3529 	KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3530 	    ("invalid bit in accmode"));
3531 	KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
3532 	    ("VAPPEND without VWRITE"));
3533 
3534 	/*
3535 	 * Look for a normal, non-privileged way to access the file/directory
3536 	 * as requested.  If it exists, go with that.
3537 	 */
3538 
3539 	if (privused != NULL)
3540 		*privused = 0;
3541 
3542 	dac_granted = 0;
3543 
3544 	/* Check the owner. */
3545 	if (cred->cr_uid == file_uid) {
3546 		dac_granted |= VADMIN;
3547 		if (file_mode & S_IXUSR)
3548 			dac_granted |= VEXEC;
3549 		if (file_mode & S_IRUSR)
3550 			dac_granted |= VREAD;
3551 		if (file_mode & S_IWUSR)
3552 			dac_granted |= (VWRITE | VAPPEND);
3553 
3554 		if ((accmode & dac_granted) == accmode)
3555 			return (0);
3556 
3557 		goto privcheck;
3558 	}
3559 
3560 	/* Otherwise, check the groups (first match) */
3561 	if (groupmember(file_gid, cred)) {
3562 		if (file_mode & S_IXGRP)
3563 			dac_granted |= VEXEC;
3564 		if (file_mode & S_IRGRP)
3565 			dac_granted |= VREAD;
3566 		if (file_mode & S_IWGRP)
3567 			dac_granted |= (VWRITE | VAPPEND);
3568 
3569 		if ((accmode & dac_granted) == accmode)
3570 			return (0);
3571 
3572 		goto privcheck;
3573 	}
3574 
3575 	/* Otherwise, check everyone else. */
3576 	if (file_mode & S_IXOTH)
3577 		dac_granted |= VEXEC;
3578 	if (file_mode & S_IROTH)
3579 		dac_granted |= VREAD;
3580 	if (file_mode & S_IWOTH)
3581 		dac_granted |= (VWRITE | VAPPEND);
3582 	if ((accmode & dac_granted) == accmode)
3583 		return (0);
3584 
3585 privcheck:
3586 	/*
3587 	 * Build a privilege mask to determine if the set of privileges
3588 	 * satisfies the requirements when combined with the granted mask
3589 	 * from above.  For each privilege, if the privilege is required,
3590 	 * bitwise or the request type onto the priv_granted mask.
3591 	 */
3592 	priv_granted = 0;
3593 
3594 	if (type == VDIR) {
3595 		/*
3596 		 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3597 		 * requests, instead of PRIV_VFS_EXEC.
3598 		 */
3599 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3600 		    !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3601 			priv_granted |= VEXEC;
3602 	} else {
3603 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3604 		    !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3605 			priv_granted |= VEXEC;
3606 	}
3607 
3608 	if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
3609 	    !priv_check_cred(cred, PRIV_VFS_READ, 0))
3610 		priv_granted |= VREAD;
3611 
3612 	if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3613 	    !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3614 		priv_granted |= (VWRITE | VAPPEND);
3615 
3616 	if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3617 	    !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3618 		priv_granted |= VADMIN;
3619 
3620 	if ((accmode & (priv_granted | dac_granted)) == accmode) {
3621 		/* XXX audit: privilege used */
3622 		if (privused != NULL)
3623 			*privused = 1;
3624 		return (0);
3625 	}
3626 
3627 	return ((accmode & VADMIN) ? EPERM : EACCES);
3628 }
3629 
3630 /*
3631  * Credential check based on process requesting service, and per-attribute
3632  * permissions.
3633  */
3634 int
3635 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3636     struct thread *td, accmode_t accmode)
3637 {
3638 
3639 	/*
3640 	 * Kernel-invoked always succeeds.
3641 	 */
3642 	if (cred == NOCRED)
3643 		return (0);
3644 
3645 	/*
3646 	 * Do not allow privileged processes in jail to directly manipulate
3647 	 * system attributes.
3648 	 */
3649 	switch (attrnamespace) {
3650 	case EXTATTR_NAMESPACE_SYSTEM:
3651 		/* Potentially should be: return (EPERM); */
3652 		return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3653 	case EXTATTR_NAMESPACE_USER:
3654 		return (VOP_ACCESS(vp, accmode, cred, td));
3655 	default:
3656 		return (EPERM);
3657 	}
3658 }
3659 
3660 #ifdef DEBUG_VFS_LOCKS
3661 /*
3662  * This only exists to supress warnings from unlocked specfs accesses.  It is
3663  * no longer ok to have an unlocked VFS.
3664  */
3665 #define	IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL ||		\
3666 	(vp)->v_type == VCHR ||	(vp)->v_type == VBAD)
3667 
3668 int vfs_badlock_ddb = 1;	/* Drop into debugger on violation. */
3669 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, "");
3670 
3671 int vfs_badlock_mutex = 1;	/* Check for interlock across VOPs. */
3672 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, "");
3673 
3674 int vfs_badlock_print = 1;	/* Print lock violations. */
3675 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, "");
3676 
3677 #ifdef KDB
3678 int vfs_badlock_backtrace = 1;	/* Print backtrace at lock violations. */
3679 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, "");
3680 #endif
3681 
3682 static void
3683 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3684 {
3685 
3686 #ifdef KDB
3687 	if (vfs_badlock_backtrace)
3688 		kdb_backtrace();
3689 #endif
3690 	if (vfs_badlock_print)
3691 		printf("%s: %p %s\n", str, (void *)vp, msg);
3692 	if (vfs_badlock_ddb)
3693 		kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3694 }
3695 
3696 void
3697 assert_vi_locked(struct vnode *vp, const char *str)
3698 {
3699 
3700 	if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3701 		vfs_badlock("interlock is not locked but should be", str, vp);
3702 }
3703 
3704 void
3705 assert_vi_unlocked(struct vnode *vp, const char *str)
3706 {
3707 
3708 	if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3709 		vfs_badlock("interlock is locked but should not be", str, vp);
3710 }
3711 
3712 void
3713 assert_vop_locked(struct vnode *vp, const char *str)
3714 {
3715 
3716 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0)
3717 		vfs_badlock("is not locked but should be", str, vp);
3718 }
3719 
3720 void
3721 assert_vop_unlocked(struct vnode *vp, const char *str)
3722 {
3723 
3724 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
3725 		vfs_badlock("is locked but should not be", str, vp);
3726 }
3727 
3728 void
3729 assert_vop_elocked(struct vnode *vp, const char *str)
3730 {
3731 
3732 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
3733 		vfs_badlock("is not exclusive locked but should be", str, vp);
3734 }
3735 
3736 #if 0
3737 void
3738 assert_vop_elocked_other(struct vnode *vp, const char *str)
3739 {
3740 
3741 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
3742 		vfs_badlock("is not exclusive locked by another thread",
3743 		    str, vp);
3744 }
3745 
3746 void
3747 assert_vop_slocked(struct vnode *vp, const char *str)
3748 {
3749 
3750 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
3751 		vfs_badlock("is not locked shared but should be", str, vp);
3752 }
3753 #endif /* 0 */
3754 #endif /* DEBUG_VFS_LOCKS */
3755 
3756 void
3757 vop_rename_fail(struct vop_rename_args *ap)
3758 {
3759 
3760 	if (ap->a_tvp != NULL)
3761 		vput(ap->a_tvp);
3762 	if (ap->a_tdvp == ap->a_tvp)
3763 		vrele(ap->a_tdvp);
3764 	else
3765 		vput(ap->a_tdvp);
3766 	vrele(ap->a_fdvp);
3767 	vrele(ap->a_fvp);
3768 }
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->v_vnlock != a->a_fdvp->v_vnlock &&
3784 	    (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
3785 		ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3786 	if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
3787 		ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
3788 
3789 	/* Check the target. */
3790 	if (a->a_tvp)
3791 		ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3792 	ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3793 #endif
3794 	if (a->a_tdvp != a->a_fdvp)
3795 		vhold(a->a_fdvp);
3796 	if (a->a_tvp != a->a_fvp)
3797 		vhold(a->a_fvp);
3798 	vhold(a->a_tdvp);
3799 	if (a->a_tvp)
3800 		vhold(a->a_tvp);
3801 }
3802 
3803 void
3804 vop_strategy_pre(void *ap)
3805 {
3806 #ifdef DEBUG_VFS_LOCKS
3807 	struct vop_strategy_args *a;
3808 	struct buf *bp;
3809 
3810 	a = ap;
3811 	bp = a->a_bp;
3812 
3813 	/*
3814 	 * Cluster ops lock their component buffers but not the IO container.
3815 	 */
3816 	if ((bp->b_flags & B_CLUSTER) != 0)
3817 		return;
3818 
3819 	if (!BUF_ISLOCKED(bp)) {
3820 		if (vfs_badlock_print)
3821 			printf(
3822 			    "VOP_STRATEGY: bp is not locked but should be\n");
3823 		if (vfs_badlock_ddb)
3824 			kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3825 	}
3826 #endif
3827 }
3828 
3829 void
3830 vop_lookup_pre(void *ap)
3831 {
3832 #ifdef DEBUG_VFS_LOCKS
3833 	struct vop_lookup_args *a;
3834 	struct vnode *dvp;
3835 
3836 	a = ap;
3837 	dvp = a->a_dvp;
3838 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3839 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3840 #endif
3841 }
3842 
3843 void
3844 vop_lookup_post(void *ap, int rc)
3845 {
3846 #ifdef DEBUG_VFS_LOCKS
3847 	struct vop_lookup_args *a;
3848 	struct vnode *dvp;
3849 	struct vnode *vp;
3850 
3851 	a = ap;
3852 	dvp = a->a_dvp;
3853 	vp = *(a->a_vpp);
3854 
3855 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3856 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3857 
3858 	if (!rc)
3859 		ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
3860 #endif
3861 }
3862 
3863 void
3864 vop_lock_pre(void *ap)
3865 {
3866 #ifdef DEBUG_VFS_LOCKS
3867 	struct vop_lock1_args *a = ap;
3868 
3869 	if ((a->a_flags & LK_INTERLOCK) == 0)
3870 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3871 	else
3872 		ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
3873 #endif
3874 }
3875 
3876 void
3877 vop_lock_post(void *ap, int rc)
3878 {
3879 #ifdef DEBUG_VFS_LOCKS
3880 	struct vop_lock1_args *a = ap;
3881 
3882 	ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3883 	if (rc == 0)
3884 		ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
3885 #endif
3886 }
3887 
3888 void
3889 vop_unlock_pre(void *ap)
3890 {
3891 #ifdef DEBUG_VFS_LOCKS
3892 	struct vop_unlock_args *a = ap;
3893 
3894 	if (a->a_flags & LK_INTERLOCK)
3895 		ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
3896 	ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
3897 #endif
3898 }
3899 
3900 void
3901 vop_unlock_post(void *ap, int rc)
3902 {
3903 #ifdef DEBUG_VFS_LOCKS
3904 	struct vop_unlock_args *a = ap;
3905 
3906 	if (a->a_flags & LK_INTERLOCK)
3907 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
3908 #endif
3909 }
3910 
3911 void
3912 vop_create_post(void *ap, int rc)
3913 {
3914 	struct vop_create_args *a = ap;
3915 
3916 	if (!rc)
3917 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3918 }
3919 
3920 void
3921 vop_link_post(void *ap, int rc)
3922 {
3923 	struct vop_link_args *a = ap;
3924 
3925 	if (!rc) {
3926 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
3927 		VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
3928 	}
3929 }
3930 
3931 void
3932 vop_mkdir_post(void *ap, int rc)
3933 {
3934 	struct vop_mkdir_args *a = ap;
3935 
3936 	if (!rc)
3937 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3938 }
3939 
3940 void
3941 vop_mknod_post(void *ap, int rc)
3942 {
3943 	struct vop_mknod_args *a = ap;
3944 
3945 	if (!rc)
3946 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3947 }
3948 
3949 void
3950 vop_remove_post(void *ap, int rc)
3951 {
3952 	struct vop_remove_args *a = ap;
3953 
3954 	if (!rc) {
3955 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3956 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3957 	}
3958 }
3959 
3960 void
3961 vop_rename_post(void *ap, int rc)
3962 {
3963 	struct vop_rename_args *a = ap;
3964 
3965 	if (!rc) {
3966 		VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
3967 		VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
3968 		VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
3969 		if (a->a_tvp)
3970 			VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
3971 	}
3972 	if (a->a_tdvp != a->a_fdvp)
3973 		vdrop(a->a_fdvp);
3974 	if (a->a_tvp != a->a_fvp)
3975 		vdrop(a->a_fvp);
3976 	vdrop(a->a_tdvp);
3977 	if (a->a_tvp)
3978 		vdrop(a->a_tvp);
3979 }
3980 
3981 void
3982 vop_rmdir_post(void *ap, int rc)
3983 {
3984 	struct vop_rmdir_args *a = ap;
3985 
3986 	if (!rc) {
3987 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3988 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3989 	}
3990 }
3991 
3992 void
3993 vop_setattr_post(void *ap, int rc)
3994 {
3995 	struct vop_setattr_args *a = ap;
3996 
3997 	if (!rc)
3998 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
3999 }
4000 
4001 void
4002 vop_symlink_post(void *ap, int rc)
4003 {
4004 	struct vop_symlink_args *a = ap;
4005 
4006 	if (!rc)
4007 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4008 }
4009 
4010 static struct knlist fs_knlist;
4011 
4012 static void
4013 vfs_event_init(void *arg)
4014 {
4015 	knlist_init_mtx(&fs_knlist, NULL);
4016 }
4017 /* XXX - correct order? */
4018 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4019 
4020 void
4021 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
4022 {
4023 
4024 	KNOTE_UNLOCKED(&fs_knlist, event);
4025 }
4026 
4027 static int	filt_fsattach(struct knote *kn);
4028 static void	filt_fsdetach(struct knote *kn);
4029 static int	filt_fsevent(struct knote *kn, long hint);
4030 
4031 struct filterops fs_filtops = {
4032 	.f_isfd = 0,
4033 	.f_attach = filt_fsattach,
4034 	.f_detach = filt_fsdetach,
4035 	.f_event = filt_fsevent
4036 };
4037 
4038 static int
4039 filt_fsattach(struct knote *kn)
4040 {
4041 
4042 	kn->kn_flags |= EV_CLEAR;
4043 	knlist_add(&fs_knlist, kn, 0);
4044 	return (0);
4045 }
4046 
4047 static void
4048 filt_fsdetach(struct knote *kn)
4049 {
4050 
4051 	knlist_remove(&fs_knlist, kn, 0);
4052 }
4053 
4054 static int
4055 filt_fsevent(struct knote *kn, long hint)
4056 {
4057 
4058 	kn->kn_fflags |= hint;
4059 	return (kn->kn_fflags != 0);
4060 }
4061 
4062 static int
4063 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4064 {
4065 	struct vfsidctl vc;
4066 	int error;
4067 	struct mount *mp;
4068 
4069 	error = SYSCTL_IN(req, &vc, sizeof(vc));
4070 	if (error)
4071 		return (error);
4072 	if (vc.vc_vers != VFS_CTL_VERS1)
4073 		return (EINVAL);
4074 	mp = vfs_getvfs(&vc.vc_fsid);
4075 	if (mp == NULL)
4076 		return (ENOENT);
4077 	/* ensure that a specific sysctl goes to the right filesystem. */
4078 	if (strcmp(vc.vc_fstypename, "*") != 0 &&
4079 	    strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4080 		vfs_rel(mp);
4081 		return (EINVAL);
4082 	}
4083 	VCTLTOREQ(&vc, req);
4084 	error = VFS_SYSCTL(mp, vc.vc_op, req);
4085 	vfs_rel(mp);
4086 	return (error);
4087 }
4088 
4089 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "",
4090     "Sysctl by fsid");
4091 
4092 /*
4093  * Function to initialize a va_filerev field sensibly.
4094  * XXX: Wouldn't a random number make a lot more sense ??
4095  */
4096 u_quad_t
4097 init_va_filerev(void)
4098 {
4099 	struct bintime bt;
4100 
4101 	getbinuptime(&bt);
4102 	return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4103 }
4104 
4105 static int	filt_vfsread(struct knote *kn, long hint);
4106 static int	filt_vfswrite(struct knote *kn, long hint);
4107 static int	filt_vfsvnode(struct knote *kn, long hint);
4108 static void	filt_vfsdetach(struct knote *kn);
4109 static struct filterops vfsread_filtops = {
4110 	.f_isfd = 1,
4111 	.f_detach = filt_vfsdetach,
4112 	.f_event = filt_vfsread
4113 };
4114 static struct filterops vfswrite_filtops = {
4115 	.f_isfd = 1,
4116 	.f_detach = filt_vfsdetach,
4117 	.f_event = filt_vfswrite
4118 };
4119 static struct filterops vfsvnode_filtops = {
4120 	.f_isfd = 1,
4121 	.f_detach = filt_vfsdetach,
4122 	.f_event = filt_vfsvnode
4123 };
4124 
4125 static void
4126 vfs_knllock(void *arg)
4127 {
4128 	struct vnode *vp = arg;
4129 
4130 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4131 }
4132 
4133 static void
4134 vfs_knlunlock(void *arg)
4135 {
4136 	struct vnode *vp = arg;
4137 
4138 	VOP_UNLOCK(vp, 0);
4139 }
4140 
4141 static void
4142 vfs_knl_assert_locked(void *arg)
4143 {
4144 #ifdef DEBUG_VFS_LOCKS
4145 	struct vnode *vp = arg;
4146 
4147 	ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4148 #endif
4149 }
4150 
4151 static void
4152 vfs_knl_assert_unlocked(void *arg)
4153 {
4154 #ifdef DEBUG_VFS_LOCKS
4155 	struct vnode *vp = arg;
4156 
4157 	ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4158 #endif
4159 }
4160 
4161 int
4162 vfs_kqfilter(struct vop_kqfilter_args *ap)
4163 {
4164 	struct vnode *vp = ap->a_vp;
4165 	struct knote *kn = ap->a_kn;
4166 	struct knlist *knl;
4167 
4168 	switch (kn->kn_filter) {
4169 	case EVFILT_READ:
4170 		kn->kn_fop = &vfsread_filtops;
4171 		break;
4172 	case EVFILT_WRITE:
4173 		kn->kn_fop = &vfswrite_filtops;
4174 		break;
4175 	case EVFILT_VNODE:
4176 		kn->kn_fop = &vfsvnode_filtops;
4177 		break;
4178 	default:
4179 		return (EINVAL);
4180 	}
4181 
4182 	kn->kn_hook = (caddr_t)vp;
4183 
4184 	v_addpollinfo(vp);
4185 	if (vp->v_pollinfo == NULL)
4186 		return (ENOMEM);
4187 	knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4188 	knlist_add(knl, kn, 0);
4189 
4190 	return (0);
4191 }
4192 
4193 /*
4194  * Detach knote from vnode
4195  */
4196 static void
4197 filt_vfsdetach(struct knote *kn)
4198 {
4199 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4200 
4201 	KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4202 	knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4203 }
4204 
4205 /*ARGSUSED*/
4206 static int
4207 filt_vfsread(struct knote *kn, long hint)
4208 {
4209 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4210 	struct vattr va;
4211 	int res;
4212 
4213 	/*
4214 	 * filesystem is gone, so set the EOF flag and schedule
4215 	 * the knote for deletion.
4216 	 */
4217 	if (hint == NOTE_REVOKE) {
4218 		VI_LOCK(vp);
4219 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4220 		VI_UNLOCK(vp);
4221 		return (1);
4222 	}
4223 
4224 	if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4225 		return (0);
4226 
4227 	VI_LOCK(vp);
4228 	kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4229 	res = (kn->kn_data != 0);
4230 	VI_UNLOCK(vp);
4231 	return (res);
4232 }
4233 
4234 /*ARGSUSED*/
4235 static int
4236 filt_vfswrite(struct knote *kn, long hint)
4237 {
4238 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4239 
4240 	VI_LOCK(vp);
4241 
4242 	/*
4243 	 * filesystem is gone, so set the EOF flag and schedule
4244 	 * the knote for deletion.
4245 	 */
4246 	if (hint == NOTE_REVOKE)
4247 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4248 
4249 	kn->kn_data = 0;
4250 	VI_UNLOCK(vp);
4251 	return (1);
4252 }
4253 
4254 static int
4255 filt_vfsvnode(struct knote *kn, long hint)
4256 {
4257 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4258 	int res;
4259 
4260 	VI_LOCK(vp);
4261 	if (kn->kn_sfflags & hint)
4262 		kn->kn_fflags |= hint;
4263 	if (hint == NOTE_REVOKE) {
4264 		kn->kn_flags |= EV_EOF;
4265 		VI_UNLOCK(vp);
4266 		return (1);
4267 	}
4268 	res = (kn->kn_fflags != 0);
4269 	VI_UNLOCK(vp);
4270 	return (res);
4271 }
4272 
4273 int
4274 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4275 {
4276 	int error;
4277 
4278 	if (dp->d_reclen > ap->a_uio->uio_resid)
4279 		return (ENAMETOOLONG);
4280 	error = uiomove(dp, dp->d_reclen, ap->a_uio);
4281 	if (error) {
4282 		if (ap->a_ncookies != NULL) {
4283 			if (ap->a_cookies != NULL)
4284 				free(ap->a_cookies, M_TEMP);
4285 			ap->a_cookies = NULL;
4286 			*ap->a_ncookies = 0;
4287 		}
4288 		return (error);
4289 	}
4290 	if (ap->a_ncookies == NULL)
4291 		return (0);
4292 
4293 	KASSERT(ap->a_cookies,
4294 	    ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4295 
4296 	*ap->a_cookies = realloc(*ap->a_cookies,
4297 	    (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4298 	(*ap->a_cookies)[*ap->a_ncookies] = off;
4299 	return (0);
4300 }
4301 
4302 /*
4303  * Mark for update the access time of the file if the filesystem
4304  * supports VOP_MARKATIME.  This functionality is used by execve and
4305  * mmap, so we want to avoid the I/O implied by directly setting
4306  * va_atime for the sake of efficiency.
4307  */
4308 void
4309 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4310 {
4311 	struct mount *mp;
4312 
4313 	mp = vp->v_mount;
4314 	VFS_ASSERT_GIANT(mp);
4315 	ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4316 	if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4317 		(void)VOP_MARKATIME(vp);
4318 }
4319 
4320 /*
4321  * The purpose of this routine is to remove granularity from accmode_t,
4322  * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4323  * VADMIN and VAPPEND.
4324  *
4325  * If it returns 0, the caller is supposed to continue with the usual
4326  * access checks using 'accmode' as modified by this routine.  If it
4327  * returns nonzero value, the caller is supposed to return that value
4328  * as errno.
4329  *
4330  * Note that after this routine runs, accmode may be zero.
4331  */
4332 int
4333 vfs_unixify_accmode(accmode_t *accmode)
4334 {
4335 	/*
4336 	 * There is no way to specify explicit "deny" rule using
4337 	 * file mode or POSIX.1e ACLs.
4338 	 */
4339 	if (*accmode & VEXPLICIT_DENY) {
4340 		*accmode = 0;
4341 		return (0);
4342 	}
4343 
4344 	/*
4345 	 * None of these can be translated into usual access bits.
4346 	 * Also, the common case for NFSv4 ACLs is to not contain
4347 	 * either of these bits. Caller should check for VWRITE
4348 	 * on the containing directory instead.
4349 	 */
4350 	if (*accmode & (VDELETE_CHILD | VDELETE))
4351 		return (EPERM);
4352 
4353 	if (*accmode & VADMIN_PERMS) {
4354 		*accmode &= ~VADMIN_PERMS;
4355 		*accmode |= VADMIN;
4356 	}
4357 
4358 	/*
4359 	 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4360 	 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4361 	 */
4362 	*accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
4363 
4364 	return (0);
4365 }
4366