xref: /freebsd/sys/kern/vfs_subr.c (revision fcb560670601b2a4d87bb31d7531c8dcc37ee71b)
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_compat.h"
45 #include "opt_ddb.h"
46 #include "opt_watchdog.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/bio.h>
51 #include <sys/buf.h>
52 #include <sys/condvar.h>
53 #include <sys/conf.h>
54 #include <sys/dirent.h>
55 #include <sys/event.h>
56 #include <sys/eventhandler.h>
57 #include <sys/extattr.h>
58 #include <sys/file.h>
59 #include <sys/fcntl.h>
60 #include <sys/jail.h>
61 #include <sys/kdb.h>
62 #include <sys/kernel.h>
63 #include <sys/kthread.h>
64 #include <sys/lockf.h>
65 #include <sys/malloc.h>
66 #include <sys/mount.h>
67 #include <sys/namei.h>
68 #include <sys/pctrie.h>
69 #include <sys/priv.h>
70 #include <sys/reboot.h>
71 #include <sys/rwlock.h>
72 #include <sys/sched.h>
73 #include <sys/sleepqueue.h>
74 #include <sys/smp.h>
75 #include <sys/stat.h>
76 #include <sys/sysctl.h>
77 #include <sys/syslog.h>
78 #include <sys/vmmeter.h>
79 #include <sys/vnode.h>
80 #include <sys/watchdog.h>
81 
82 #include <machine/stdarg.h>
83 
84 #include <security/mac/mac_framework.h>
85 
86 #include <vm/vm.h>
87 #include <vm/vm_object.h>
88 #include <vm/vm_extern.h>
89 #include <vm/pmap.h>
90 #include <vm/vm_map.h>
91 #include <vm/vm_page.h>
92 #include <vm/vm_kern.h>
93 #include <vm/uma.h>
94 
95 #ifdef DDB
96 #include <ddb/ddb.h>
97 #endif
98 
99 static void	delmntque(struct vnode *vp);
100 static int	flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
101 		    int slpflag, int slptimeo);
102 static void	syncer_shutdown(void *arg, int howto);
103 static int	vtryrecycle(struct vnode *vp);
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	vnlru_free(int);
109 static void	vgonel(struct vnode *);
110 static void	vfs_knllock(void *arg);
111 static void	vfs_knlunlock(void *arg);
112 static void	vfs_knl_assert_locked(void *arg);
113 static void	vfs_knl_assert_unlocked(void *arg);
114 static void	destroy_vpollinfo(struct vpollinfo *vi);
115 
116 /*
117  * Number of vnodes in existence.  Increased whenever getnewvnode()
118  * allocates a new vnode, decreased in vdropl() for VI_DOOMED vnode.
119  */
120 static unsigned long	numvnodes;
121 
122 SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
123     "Number of vnodes in existence");
124 
125 /*
126  * Conversion tables for conversion from vnode types to inode formats
127  * and back.
128  */
129 enum vtype iftovt_tab[16] = {
130 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
131 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
132 };
133 int vttoif_tab[10] = {
134 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
135 	S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
136 };
137 
138 /*
139  * List of vnodes that are ready for recycling.
140  */
141 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
142 
143 /*
144  * Free vnode target.  Free vnodes may simply be files which have been stat'd
145  * but not read.  This is somewhat common, and a small cache of such files
146  * should be kept to avoid recreation costs.
147  */
148 static u_long wantfreevnodes;
149 SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
150 /* Number of vnodes in the free list. */
151 static u_long freevnodes;
152 SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
153     "Number of vnodes in the free list");
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     "Number of calls to reassignbuf");
167 
168 /*
169  * Cache for the mount type id assigned to NFS.  This is used for
170  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
171  */
172 int	nfs_mount_type = -1;
173 
174 /* To keep more than one thread at a time from running vfs_getnewfsid */
175 static struct mtx mntid_mtx;
176 
177 /*
178  * Lock for any access to the following:
179  *	vnode_free_list
180  *	numvnodes
181  *	freevnodes
182  */
183 static struct mtx vnode_free_list_mtx;
184 
185 /* Publicly exported FS */
186 struct nfs_public nfs_pub;
187 
188 static uma_zone_t buf_trie_zone;
189 
190 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
191 static uma_zone_t vnode_zone;
192 static uma_zone_t vnodepoll_zone;
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;
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     "Time to delay syncing files (in seconds)");
242 static int dirdelay = 29;		/* time to delay syncing directories */
243 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
244     "Time to delay syncing directories (in seconds)");
245 static int metadelay = 28;		/* time to delay syncing metadata */
246 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
247     "Time to delay syncing metadata (in seconds)");
248 static int rushjob;		/* number of slots to run ASAP */
249 static int stat_rush_requests;	/* number of times I/O speeded up */
250 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
251     "Number of times I/O speeded up (rush requests)");
252 
253 /*
254  * When shutting down the syncer, run it at four times normal speed.
255  */
256 #define SYNCER_SHUTDOWN_SPEEDUP		4
257 static int sync_vnode_count;
258 static int syncer_worklist_len;
259 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
260     syncer_state;
261 
262 /*
263  * Number of vnodes we want to exist at any one time.  This is mostly used
264  * to size hash tables in vnode-related code.  It is normally not used in
265  * getnewvnode(), as wantfreevnodes is normally nonzero.)
266  *
267  * XXX desiredvnodes is historical cruft and should not exist.
268  */
269 int desiredvnodes;
270 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
271     &desiredvnodes, 0, "Maximum number of vnodes");
272 SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
273     &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
274 static int vnlru_nowhere;
275 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
276     &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
277 
278 /* Shift count for (uintptr_t)vp to initialize vp->v_hash. */
279 static int vnsz2log;
280 
281 /*
282  * Support for the bufobj clean & dirty pctrie.
283  */
284 static void *
285 buf_trie_alloc(struct pctrie *ptree)
286 {
287 
288 	return uma_zalloc(buf_trie_zone, M_NOWAIT);
289 }
290 
291 static void
292 buf_trie_free(struct pctrie *ptree, void *node)
293 {
294 
295 	uma_zfree(buf_trie_zone, node);
296 }
297 PCTRIE_DEFINE(BUF, buf, b_lblkno, buf_trie_alloc, buf_trie_free);
298 
299 /*
300  * Initialize the vnode management data structures.
301  *
302  * Reevaluate the following cap on the number of vnodes after the physical
303  * memory size exceeds 512GB.  In the limit, as the physical memory size
304  * grows, the ratio of physical pages to vnodes approaches sixteen to one.
305  */
306 #ifndef	MAXVNODES_MAX
307 #define	MAXVNODES_MAX	(512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
308 #endif
309 static void
310 vntblinit(void *dummy __unused)
311 {
312 	u_int i;
313 	int physvnodes, virtvnodes;
314 
315 	/*
316 	 * Desiredvnodes is a function of the physical memory size and the
317 	 * kernel's heap size.  Generally speaking, it scales with the
318 	 * physical memory size.  The ratio of desiredvnodes to physical pages
319 	 * is one to four until desiredvnodes exceeds 98,304.  Thereafter, the
320 	 * marginal ratio of desiredvnodes to physical pages is one to
321 	 * sixteen.  However, desiredvnodes is limited by the kernel's heap
322 	 * size.  The memory required by desiredvnodes vnodes and vm objects
323 	 * may not exceed one seventh of the kernel's heap size.
324 	 */
325 	physvnodes = maxproc + vm_cnt.v_page_count / 16 + 3 * min(98304 * 4,
326 	    vm_cnt.v_page_count) / 16;
327 	virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
328 	    sizeof(struct vnode)));
329 	desiredvnodes = min(physvnodes, virtvnodes);
330 	if (desiredvnodes > MAXVNODES_MAX) {
331 		if (bootverbose)
332 			printf("Reducing kern.maxvnodes %d -> %d\n",
333 			    desiredvnodes, MAXVNODES_MAX);
334 		desiredvnodes = MAXVNODES_MAX;
335 	}
336 	wantfreevnodes = desiredvnodes / 4;
337 	mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
338 	TAILQ_INIT(&vnode_free_list);
339 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
340 	vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
341 	    NULL, NULL, UMA_ALIGN_PTR, 0);
342 	vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
343 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
344 	/*
345 	 * Preallocate enough nodes to support one-per buf so that
346 	 * we can not fail an insert.  reassignbuf() callers can not
347 	 * tolerate the insertion failure.
348 	 */
349 	buf_trie_zone = uma_zcreate("BUF TRIE", pctrie_node_size(),
350 	    NULL, NULL, pctrie_zone_init, NULL, UMA_ALIGN_PTR,
351 	    UMA_ZONE_NOFREE | UMA_ZONE_VM);
352 	uma_prealloc(buf_trie_zone, nbuf);
353 	/*
354 	 * Initialize the filesystem syncer.
355 	 */
356 	syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
357 	    &syncer_mask);
358 	syncer_maxdelay = syncer_mask + 1;
359 	mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
360 	cv_init(&sync_wakeup, "syncer");
361 	for (i = 1; i <= sizeof(struct vnode); i <<= 1)
362 		vnsz2log++;
363 	vnsz2log--;
364 }
365 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
366 
367 
368 /*
369  * Mark a mount point as busy. Used to synchronize access and to delay
370  * unmounting. Eventually, mountlist_mtx is not released on failure.
371  *
372  * vfs_busy() is a custom lock, it can block the caller.
373  * vfs_busy() only sleeps if the unmount is active on the mount point.
374  * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
375  * vnode belonging to mp.
376  *
377  * Lookup uses vfs_busy() to traverse mount points.
378  * root fs			var fs
379  * / vnode lock		A	/ vnode lock (/var)		D
380  * /var vnode lock	B	/log vnode lock(/var/log)	E
381  * vfs_busy lock	C	vfs_busy lock			F
382  *
383  * Within each file system, the lock order is C->A->B and F->D->E.
384  *
385  * When traversing across mounts, the system follows that lock order:
386  *
387  *        C->A->B
388  *              |
389  *              +->F->D->E
390  *
391  * The lookup() process for namei("/var") illustrates the process:
392  *  VOP_LOOKUP() obtains B while A is held
393  *  vfs_busy() obtains a shared lock on F while A and B are held
394  *  vput() releases lock on B
395  *  vput() releases lock on A
396  *  VFS_ROOT() obtains lock on D while shared lock on F is held
397  *  vfs_unbusy() releases shared lock on F
398  *  vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
399  *    Attempt to lock A (instead of vp_crossmp) while D is held would
400  *    violate the global order, causing deadlocks.
401  *
402  * dounmount() locks B while F is drained.
403  */
404 int
405 vfs_busy(struct mount *mp, int flags)
406 {
407 
408 	MPASS((flags & ~MBF_MASK) == 0);
409 	CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
410 
411 	MNT_ILOCK(mp);
412 	MNT_REF(mp);
413 	/*
414 	 * If mount point is currenly being unmounted, sleep until the
415 	 * mount point fate is decided.  If thread doing the unmounting fails,
416 	 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
417 	 * that this mount point has survived the unmount attempt and vfs_busy
418 	 * should retry.  Otherwise the unmounter thread will set MNTK_REFEXPIRE
419 	 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
420 	 * about to be really destroyed.  vfs_busy needs to release its
421 	 * reference on the mount point in this case and return with ENOENT,
422 	 * telling the caller that mount mount it tried to busy is no longer
423 	 * valid.
424 	 */
425 	while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
426 		if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
427 			MNT_REL(mp);
428 			MNT_IUNLOCK(mp);
429 			CTR1(KTR_VFS, "%s: failed busying before sleeping",
430 			    __func__);
431 			return (ENOENT);
432 		}
433 		if (flags & MBF_MNTLSTLOCK)
434 			mtx_unlock(&mountlist_mtx);
435 		mp->mnt_kern_flag |= MNTK_MWAIT;
436 		msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
437 		if (flags & MBF_MNTLSTLOCK)
438 			mtx_lock(&mountlist_mtx);
439 		MNT_ILOCK(mp);
440 	}
441 	if (flags & MBF_MNTLSTLOCK)
442 		mtx_unlock(&mountlist_mtx);
443 	mp->mnt_lockref++;
444 	MNT_IUNLOCK(mp);
445 	return (0);
446 }
447 
448 /*
449  * Free a busy filesystem.
450  */
451 void
452 vfs_unbusy(struct mount *mp)
453 {
454 
455 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
456 	MNT_ILOCK(mp);
457 	MNT_REL(mp);
458 	KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
459 	mp->mnt_lockref--;
460 	if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
461 		MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
462 		CTR1(KTR_VFS, "%s: waking up waiters", __func__);
463 		mp->mnt_kern_flag &= ~MNTK_DRAINING;
464 		wakeup(&mp->mnt_lockref);
465 	}
466 	MNT_IUNLOCK(mp);
467 }
468 
469 /*
470  * Lookup a mount point by filesystem identifier.
471  */
472 struct mount *
473 vfs_getvfs(fsid_t *fsid)
474 {
475 	struct mount *mp;
476 
477 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
478 	mtx_lock(&mountlist_mtx);
479 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
480 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
481 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
482 			vfs_ref(mp);
483 			mtx_unlock(&mountlist_mtx);
484 			return (mp);
485 		}
486 	}
487 	mtx_unlock(&mountlist_mtx);
488 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
489 	return ((struct mount *) 0);
490 }
491 
492 /*
493  * Lookup a mount point by filesystem identifier, busying it before
494  * returning.
495  *
496  * To avoid congestion on mountlist_mtx, implement simple direct-mapped
497  * cache for popular filesystem identifiers.  The cache is lockess, using
498  * the fact that struct mount's are never freed.  In worst case we may
499  * get pointer to unmounted or even different filesystem, so we have to
500  * check what we got, and go slow way if so.
501  */
502 struct mount *
503 vfs_busyfs(fsid_t *fsid)
504 {
505 #define	FSID_CACHE_SIZE	256
506 	typedef struct mount * volatile vmp_t;
507 	static vmp_t cache[FSID_CACHE_SIZE];
508 	struct mount *mp;
509 	int error;
510 	uint32_t hash;
511 
512 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
513 	hash = fsid->val[0] ^ fsid->val[1];
514 	hash = (hash >> 16 ^ hash) & (FSID_CACHE_SIZE - 1);
515 	mp = cache[hash];
516 	if (mp == NULL ||
517 	    mp->mnt_stat.f_fsid.val[0] != fsid->val[0] ||
518 	    mp->mnt_stat.f_fsid.val[1] != fsid->val[1])
519 		goto slow;
520 	if (vfs_busy(mp, 0) != 0) {
521 		cache[hash] = NULL;
522 		goto slow;
523 	}
524 	if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
525 	    mp->mnt_stat.f_fsid.val[1] == fsid->val[1])
526 		return (mp);
527 	else
528 	    vfs_unbusy(mp);
529 
530 slow:
531 	mtx_lock(&mountlist_mtx);
532 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
533 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
534 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
535 			error = vfs_busy(mp, MBF_MNTLSTLOCK);
536 			if (error) {
537 				cache[hash] = NULL;
538 				mtx_unlock(&mountlist_mtx);
539 				return (NULL);
540 			}
541 			cache[hash] = mp;
542 			return (mp);
543 		}
544 	}
545 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
546 	mtx_unlock(&mountlist_mtx);
547 	return ((struct mount *) 0);
548 }
549 
550 /*
551  * Check if a user can access privileged mount options.
552  */
553 int
554 vfs_suser(struct mount *mp, struct thread *td)
555 {
556 	int error;
557 
558 	/*
559 	 * If the thread is jailed, but this is not a jail-friendly file
560 	 * system, deny immediately.
561 	 */
562 	if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
563 		return (EPERM);
564 
565 	/*
566 	 * If the file system was mounted outside the jail of the calling
567 	 * thread, deny immediately.
568 	 */
569 	if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
570 		return (EPERM);
571 
572 	/*
573 	 * If file system supports delegated administration, we don't check
574 	 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
575 	 * by the file system itself.
576 	 * If this is not the user that did original mount, we check for
577 	 * the PRIV_VFS_MOUNT_OWNER privilege.
578 	 */
579 	if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
580 	    mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
581 		if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
582 			return (error);
583 	}
584 	return (0);
585 }
586 
587 /*
588  * Get a new unique fsid.  Try to make its val[0] unique, since this value
589  * will be used to create fake device numbers for stat().  Also try (but
590  * not so hard) make its val[0] unique mod 2^16, since some emulators only
591  * support 16-bit device numbers.  We end up with unique val[0]'s for the
592  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
593  *
594  * Keep in mind that several mounts may be running in parallel.  Starting
595  * the search one past where the previous search terminated is both a
596  * micro-optimization and a defense against returning the same fsid to
597  * different mounts.
598  */
599 void
600 vfs_getnewfsid(struct mount *mp)
601 {
602 	static uint16_t mntid_base;
603 	struct mount *nmp;
604 	fsid_t tfsid;
605 	int mtype;
606 
607 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
608 	mtx_lock(&mntid_mtx);
609 	mtype = mp->mnt_vfc->vfc_typenum;
610 	tfsid.val[1] = mtype;
611 	mtype = (mtype & 0xFF) << 24;
612 	for (;;) {
613 		tfsid.val[0] = makedev(255,
614 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
615 		mntid_base++;
616 		if ((nmp = vfs_getvfs(&tfsid)) == NULL)
617 			break;
618 		vfs_rel(nmp);
619 	}
620 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
621 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
622 	mtx_unlock(&mntid_mtx);
623 }
624 
625 /*
626  * Knob to control the precision of file timestamps:
627  *
628  *   0 = seconds only; nanoseconds zeroed.
629  *   1 = seconds and nanoseconds, accurate within 1/HZ.
630  *   2 = seconds and nanoseconds, truncated to microseconds.
631  * >=3 = seconds and nanoseconds, maximum precision.
632  */
633 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
634 
635 static int timestamp_precision = TSP_USEC;
636 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
637     &timestamp_precision, 0, "File timestamp precision (0: seconds, "
638     "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
639     "3+: sec + ns (max. precision))");
640 
641 /*
642  * Get a current timestamp.
643  */
644 void
645 vfs_timestamp(struct timespec *tsp)
646 {
647 	struct timeval tv;
648 
649 	switch (timestamp_precision) {
650 	case TSP_SEC:
651 		tsp->tv_sec = time_second;
652 		tsp->tv_nsec = 0;
653 		break;
654 	case TSP_HZ:
655 		getnanotime(tsp);
656 		break;
657 	case TSP_USEC:
658 		microtime(&tv);
659 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
660 		break;
661 	case TSP_NSEC:
662 	default:
663 		nanotime(tsp);
664 		break;
665 	}
666 }
667 
668 /*
669  * Set vnode attributes to VNOVAL
670  */
671 void
672 vattr_null(struct vattr *vap)
673 {
674 
675 	vap->va_type = VNON;
676 	vap->va_size = VNOVAL;
677 	vap->va_bytes = VNOVAL;
678 	vap->va_mode = VNOVAL;
679 	vap->va_nlink = VNOVAL;
680 	vap->va_uid = VNOVAL;
681 	vap->va_gid = VNOVAL;
682 	vap->va_fsid = VNOVAL;
683 	vap->va_fileid = VNOVAL;
684 	vap->va_blocksize = VNOVAL;
685 	vap->va_rdev = VNOVAL;
686 	vap->va_atime.tv_sec = VNOVAL;
687 	vap->va_atime.tv_nsec = VNOVAL;
688 	vap->va_mtime.tv_sec = VNOVAL;
689 	vap->va_mtime.tv_nsec = VNOVAL;
690 	vap->va_ctime.tv_sec = VNOVAL;
691 	vap->va_ctime.tv_nsec = VNOVAL;
692 	vap->va_birthtime.tv_sec = VNOVAL;
693 	vap->va_birthtime.tv_nsec = VNOVAL;
694 	vap->va_flags = VNOVAL;
695 	vap->va_gen = VNOVAL;
696 	vap->va_vaflags = 0;
697 }
698 
699 /*
700  * This routine is called when we have too many vnodes.  It attempts
701  * to free <count> vnodes and will potentially free vnodes that still
702  * have VM backing store (VM backing store is typically the cause
703  * of a vnode blowout so we want to do this).  Therefore, this operation
704  * is not considered cheap.
705  *
706  * A number of conditions may prevent a vnode from being reclaimed.
707  * the buffer cache may have references on the vnode, a directory
708  * vnode may still have references due to the namei cache representing
709  * underlying files, or the vnode may be in active use.   It is not
710  * desireable to reuse such vnodes.  These conditions may cause the
711  * number of vnodes to reach some minimum value regardless of what
712  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
713  */
714 static int
715 vlrureclaim(struct mount *mp)
716 {
717 	struct vnode *vp;
718 	int done;
719 	int trigger;
720 	int usevnodes;
721 	int count;
722 
723 	/*
724 	 * Calculate the trigger point, don't allow user
725 	 * screwups to blow us up.   This prevents us from
726 	 * recycling vnodes with lots of resident pages.  We
727 	 * aren't trying to free memory, we are trying to
728 	 * free vnodes.
729 	 */
730 	usevnodes = desiredvnodes;
731 	if (usevnodes <= 0)
732 		usevnodes = 1;
733 	trigger = vm_cnt.v_page_count * 2 / usevnodes;
734 	done = 0;
735 	vn_start_write(NULL, &mp, V_WAIT);
736 	MNT_ILOCK(mp);
737 	count = mp->mnt_nvnodelistsize / 10 + 1;
738 	while (count != 0) {
739 		vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
740 		while (vp != NULL && vp->v_type == VMARKER)
741 			vp = TAILQ_NEXT(vp, v_nmntvnodes);
742 		if (vp == NULL)
743 			break;
744 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
745 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
746 		--count;
747 		if (!VI_TRYLOCK(vp))
748 			goto next_iter;
749 		/*
750 		 * If it's been deconstructed already, it's still
751 		 * referenced, or it exceeds the trigger, skip it.
752 		 */
753 		if (vp->v_usecount ||
754 		    (!vlru_allow_cache_src &&
755 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
756 		    (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
757 		    vp->v_object->resident_page_count > trigger)) {
758 			VI_UNLOCK(vp);
759 			goto next_iter;
760 		}
761 		MNT_IUNLOCK(mp);
762 		vholdl(vp);
763 		if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
764 			vdrop(vp);
765 			goto next_iter_mntunlocked;
766 		}
767 		VI_LOCK(vp);
768 		/*
769 		 * v_usecount may have been bumped after VOP_LOCK() dropped
770 		 * the vnode interlock and before it was locked again.
771 		 *
772 		 * It is not necessary to recheck VI_DOOMED because it can
773 		 * only be set by another thread that holds both the vnode
774 		 * lock and vnode interlock.  If another thread has the
775 		 * vnode lock before we get to VOP_LOCK() and obtains the
776 		 * vnode interlock after VOP_LOCK() drops the vnode
777 		 * interlock, the other thread will be unable to drop the
778 		 * vnode lock before our VOP_LOCK() call fails.
779 		 */
780 		if (vp->v_usecount ||
781 		    (!vlru_allow_cache_src &&
782 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
783 		    (vp->v_object != NULL &&
784 		    vp->v_object->resident_page_count > trigger)) {
785 			VOP_UNLOCK(vp, LK_INTERLOCK);
786 			vdrop(vp);
787 			goto next_iter_mntunlocked;
788 		}
789 		KASSERT((vp->v_iflag & VI_DOOMED) == 0,
790 		    ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
791 		vgonel(vp);
792 		VOP_UNLOCK(vp, 0);
793 		vdropl(vp);
794 		done++;
795 next_iter_mntunlocked:
796 		if (!should_yield())
797 			goto relock_mnt;
798 		goto yield;
799 next_iter:
800 		if (!should_yield())
801 			continue;
802 		MNT_IUNLOCK(mp);
803 yield:
804 		kern_yield(PRI_USER);
805 relock_mnt:
806 		MNT_ILOCK(mp);
807 	}
808 	MNT_IUNLOCK(mp);
809 	vn_finished_write(mp);
810 	return done;
811 }
812 
813 /*
814  * Attempt to keep the free list at wantfreevnodes length.
815  */
816 static void
817 vnlru_free(int count)
818 {
819 	struct vnode *vp;
820 
821 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
822 	for (; count > 0; count--) {
823 		vp = TAILQ_FIRST(&vnode_free_list);
824 		/*
825 		 * The list can be modified while the free_list_mtx
826 		 * has been dropped and vp could be NULL here.
827 		 */
828 		if (!vp)
829 			break;
830 		VNASSERT(vp->v_op != NULL, vp,
831 		    ("vnlru_free: vnode already reclaimed."));
832 		KASSERT((vp->v_iflag & VI_FREE) != 0,
833 		    ("Removing vnode not on freelist"));
834 		KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
835 		    ("Mangling active vnode"));
836 		TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
837 		/*
838 		 * Don't recycle if we can't get the interlock.
839 		 */
840 		if (!VI_TRYLOCK(vp)) {
841 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
842 			continue;
843 		}
844 		VNASSERT((vp->v_iflag & VI_FREE) != 0 && vp->v_holdcnt == 0,
845 		    vp, ("vp inconsistent on freelist"));
846 
847 		/*
848 		 * The clear of VI_FREE prevents activation of the
849 		 * vnode.  There is no sense in putting the vnode on
850 		 * the mount point active list, only to remove it
851 		 * later during recycling.  Inline the relevant part
852 		 * of vholdl(), to avoid triggering assertions or
853 		 * activating.
854 		 */
855 		freevnodes--;
856 		vp->v_iflag &= ~VI_FREE;
857 		vp->v_holdcnt++;
858 
859 		mtx_unlock(&vnode_free_list_mtx);
860 		VI_UNLOCK(vp);
861 		vtryrecycle(vp);
862 		/*
863 		 * If the recycled succeeded this vdrop will actually free
864 		 * the vnode.  If not it will simply place it back on
865 		 * the free list.
866 		 */
867 		vdrop(vp);
868 		mtx_lock(&vnode_free_list_mtx);
869 	}
870 }
871 /*
872  * Attempt to recycle vnodes in a context that is always safe to block.
873  * Calling vlrurecycle() from the bowels of filesystem code has some
874  * interesting deadlock problems.
875  */
876 static struct proc *vnlruproc;
877 static int vnlruproc_sig;
878 
879 static void
880 vnlru_proc(void)
881 {
882 	struct mount *mp, *nmp;
883 	int done;
884 	struct proc *p = vnlruproc;
885 
886 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
887 	    SHUTDOWN_PRI_FIRST);
888 
889 	for (;;) {
890 		kproc_suspend_check(p);
891 		mtx_lock(&vnode_free_list_mtx);
892 		if (freevnodes > wantfreevnodes)
893 			vnlru_free(freevnodes - wantfreevnodes);
894 		if (numvnodes <= desiredvnodes * 9 / 10) {
895 			vnlruproc_sig = 0;
896 			wakeup(&vnlruproc_sig);
897 			msleep(vnlruproc, &vnode_free_list_mtx,
898 			    PVFS|PDROP, "vlruwt", hz);
899 			continue;
900 		}
901 		mtx_unlock(&vnode_free_list_mtx);
902 		done = 0;
903 		mtx_lock(&mountlist_mtx);
904 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
905 			if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
906 				nmp = TAILQ_NEXT(mp, mnt_list);
907 				continue;
908 			}
909 			done += vlrureclaim(mp);
910 			mtx_lock(&mountlist_mtx);
911 			nmp = TAILQ_NEXT(mp, mnt_list);
912 			vfs_unbusy(mp);
913 		}
914 		mtx_unlock(&mountlist_mtx);
915 		if (done == 0) {
916 #if 0
917 			/* These messages are temporary debugging aids */
918 			if (vnlru_nowhere < 5)
919 				printf("vnlru process getting nowhere..\n");
920 			else if (vnlru_nowhere == 5)
921 				printf("vnlru process messages stopped.\n");
922 #endif
923 			vnlru_nowhere++;
924 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
925 		} else
926 			kern_yield(PRI_USER);
927 	}
928 }
929 
930 static struct kproc_desc vnlru_kp = {
931 	"vnlru",
932 	vnlru_proc,
933 	&vnlruproc
934 };
935 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
936     &vnlru_kp);
937 
938 /*
939  * Routines having to do with the management of the vnode table.
940  */
941 
942 /*
943  * Try to recycle a freed vnode.  We abort if anyone picks up a reference
944  * before we actually vgone().  This function must be called with the vnode
945  * held to prevent the vnode from being returned to the free list midway
946  * through vgone().
947  */
948 static int
949 vtryrecycle(struct vnode *vp)
950 {
951 	struct mount *vnmp;
952 
953 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
954 	VNASSERT(vp->v_holdcnt, vp,
955 	    ("vtryrecycle: Recycling vp %p without a reference.", vp));
956 	/*
957 	 * This vnode may found and locked via some other list, if so we
958 	 * can't recycle it yet.
959 	 */
960 	if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
961 		CTR2(KTR_VFS,
962 		    "%s: impossible to recycle, vp %p lock is already held",
963 		    __func__, vp);
964 		return (EWOULDBLOCK);
965 	}
966 	/*
967 	 * Don't recycle if its filesystem is being suspended.
968 	 */
969 	if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
970 		VOP_UNLOCK(vp, 0);
971 		CTR2(KTR_VFS,
972 		    "%s: impossible to recycle, cannot start the write for %p",
973 		    __func__, vp);
974 		return (EBUSY);
975 	}
976 	/*
977 	 * If we got this far, we need to acquire the interlock and see if
978 	 * anyone picked up this vnode from another list.  If not, we will
979 	 * mark it with DOOMED via vgonel() so that anyone who does find it
980 	 * will skip over it.
981 	 */
982 	VI_LOCK(vp);
983 	if (vp->v_usecount) {
984 		VOP_UNLOCK(vp, LK_INTERLOCK);
985 		vn_finished_write(vnmp);
986 		CTR2(KTR_VFS,
987 		    "%s: impossible to recycle, %p is already referenced",
988 		    __func__, vp);
989 		return (EBUSY);
990 	}
991 	if ((vp->v_iflag & VI_DOOMED) == 0)
992 		vgonel(vp);
993 	VOP_UNLOCK(vp, LK_INTERLOCK);
994 	vn_finished_write(vnmp);
995 	return (0);
996 }
997 
998 /*
999  * Wait for available vnodes.
1000  */
1001 static int
1002 getnewvnode_wait(int suspended)
1003 {
1004 
1005 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
1006 	if (numvnodes > desiredvnodes) {
1007 		if (suspended) {
1008 			/*
1009 			 * File system is beeing suspended, we cannot risk a
1010 			 * deadlock here, so allocate new vnode anyway.
1011 			 */
1012 			if (freevnodes > wantfreevnodes)
1013 				vnlru_free(freevnodes - wantfreevnodes);
1014 			return (0);
1015 		}
1016 		if (vnlruproc_sig == 0) {
1017 			vnlruproc_sig = 1;	/* avoid unnecessary wakeups */
1018 			wakeup(vnlruproc);
1019 		}
1020 		msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
1021 		    "vlruwk", hz);
1022 	}
1023 	return (numvnodes > desiredvnodes ? ENFILE : 0);
1024 }
1025 
1026 void
1027 getnewvnode_reserve(u_int count)
1028 {
1029 	struct thread *td;
1030 
1031 	td = curthread;
1032 	/* First try to be quick and racy. */
1033 	if (atomic_fetchadd_long(&numvnodes, count) + count <= desiredvnodes) {
1034 		td->td_vp_reserv += count;
1035 		return;
1036 	} else
1037 		atomic_subtract_long(&numvnodes, count);
1038 
1039 	mtx_lock(&vnode_free_list_mtx);
1040 	while (count > 0) {
1041 		if (getnewvnode_wait(0) == 0) {
1042 			count--;
1043 			td->td_vp_reserv++;
1044 			atomic_add_long(&numvnodes, 1);
1045 		}
1046 	}
1047 	mtx_unlock(&vnode_free_list_mtx);
1048 }
1049 
1050 void
1051 getnewvnode_drop_reserve(void)
1052 {
1053 	struct thread *td;
1054 
1055 	td = curthread;
1056 	atomic_subtract_long(&numvnodes, td->td_vp_reserv);
1057 	td->td_vp_reserv = 0;
1058 }
1059 
1060 /*
1061  * Return the next vnode from the free list.
1062  */
1063 int
1064 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
1065     struct vnode **vpp)
1066 {
1067 	struct vnode *vp;
1068 	struct bufobj *bo;
1069 	struct thread *td;
1070 	int error;
1071 
1072 	CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
1073 	vp = NULL;
1074 	td = curthread;
1075 	if (td->td_vp_reserv > 0) {
1076 		td->td_vp_reserv -= 1;
1077 		goto alloc;
1078 	}
1079 	mtx_lock(&vnode_free_list_mtx);
1080 	/*
1081 	 * Lend our context to reclaim vnodes if they've exceeded the max.
1082 	 */
1083 	if (freevnodes > wantfreevnodes)
1084 		vnlru_free(1);
1085 	error = getnewvnode_wait(mp != NULL && (mp->mnt_kern_flag &
1086 	    MNTK_SUSPEND));
1087 #if 0	/* XXX Not all VFS_VGET/ffs_vget callers check returns. */
1088 	if (error != 0) {
1089 		mtx_unlock(&vnode_free_list_mtx);
1090 		return (error);
1091 	}
1092 #endif
1093 	atomic_add_long(&numvnodes, 1);
1094 	mtx_unlock(&vnode_free_list_mtx);
1095 alloc:
1096 	vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
1097 	/*
1098 	 * Setup locks.
1099 	 */
1100 	vp->v_vnlock = &vp->v_lock;
1101 	mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
1102 	/*
1103 	 * By default, don't allow shared locks unless filesystems
1104 	 * opt-in.
1105 	 */
1106 	lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE | LK_IS_VNODE);
1107 	/*
1108 	 * Initialize bufobj.
1109 	 */
1110 	bo = &vp->v_bufobj;
1111 	bo->__bo_vnode = vp;
1112 	rw_init(BO_LOCKPTR(bo), "bufobj interlock");
1113 	bo->bo_ops = &buf_ops_bio;
1114 	bo->bo_private = vp;
1115 	TAILQ_INIT(&bo->bo_clean.bv_hd);
1116 	TAILQ_INIT(&bo->bo_dirty.bv_hd);
1117 	/*
1118 	 * Initialize namecache.
1119 	 */
1120 	LIST_INIT(&vp->v_cache_src);
1121 	TAILQ_INIT(&vp->v_cache_dst);
1122 	/*
1123 	 * Finalize various vnode identity bits.
1124 	 */
1125 	vp->v_type = VNON;
1126 	vp->v_tag = tag;
1127 	vp->v_op = vops;
1128 	v_incr_usecount(vp);
1129 	vp->v_data = NULL;
1130 #ifdef MAC
1131 	mac_vnode_init(vp);
1132 	if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1133 		mac_vnode_associate_singlelabel(mp, vp);
1134 	else if (mp == NULL && vops != &dead_vnodeops)
1135 		printf("NULL mp in getnewvnode()\n");
1136 #endif
1137 	if (mp != NULL) {
1138 		bo->bo_bsize = mp->mnt_stat.f_iosize;
1139 		if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1140 			vp->v_vflag |= VV_NOKNOTE;
1141 	}
1142 	rangelock_init(&vp->v_rl);
1143 
1144 	/*
1145 	 * For the filesystems which do not use vfs_hash_insert(),
1146 	 * still initialize v_hash to have vfs_hash_index() useful.
1147 	 * E.g., nullfs uses vfs_hash_index() on the lower vnode for
1148 	 * its own hashing.
1149 	 */
1150 	vp->v_hash = (uintptr_t)vp >> vnsz2log;
1151 
1152 	*vpp = vp;
1153 	return (0);
1154 }
1155 
1156 /*
1157  * Delete from old mount point vnode list, if on one.
1158  */
1159 static void
1160 delmntque(struct vnode *vp)
1161 {
1162 	struct mount *mp;
1163 	int active;
1164 
1165 	mp = vp->v_mount;
1166 	if (mp == NULL)
1167 		return;
1168 	MNT_ILOCK(mp);
1169 	VI_LOCK(vp);
1170 	KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize,
1171 	    ("Active vnode list size %d > Vnode list size %d",
1172 	     mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize));
1173 	active = vp->v_iflag & VI_ACTIVE;
1174 	vp->v_iflag &= ~VI_ACTIVE;
1175 	if (active) {
1176 		mtx_lock(&vnode_free_list_mtx);
1177 		TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist);
1178 		mp->mnt_activevnodelistsize--;
1179 		mtx_unlock(&vnode_free_list_mtx);
1180 	}
1181 	vp->v_mount = NULL;
1182 	VI_UNLOCK(vp);
1183 	VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1184 		("bad mount point vnode list size"));
1185 	TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1186 	mp->mnt_nvnodelistsize--;
1187 	MNT_REL(mp);
1188 	MNT_IUNLOCK(mp);
1189 }
1190 
1191 static void
1192 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1193 {
1194 
1195 	vp->v_data = NULL;
1196 	vp->v_op = &dead_vnodeops;
1197 	vgone(vp);
1198 	vput(vp);
1199 }
1200 
1201 /*
1202  * Insert into list of vnodes for the new mount point, if available.
1203  */
1204 int
1205 insmntque1(struct vnode *vp, struct mount *mp,
1206 	void (*dtr)(struct vnode *, void *), void *dtr_arg)
1207 {
1208 
1209 	KASSERT(vp->v_mount == NULL,
1210 		("insmntque: vnode already on per mount vnode list"));
1211 	VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1212 	ASSERT_VOP_ELOCKED(vp, "insmntque: non-locked vp");
1213 
1214 	/*
1215 	 * We acquire the vnode interlock early to ensure that the
1216 	 * vnode cannot be recycled by another process releasing a
1217 	 * holdcnt on it before we get it on both the vnode list
1218 	 * and the active vnode list. The mount mutex protects only
1219 	 * manipulation of the vnode list and the vnode freelist
1220 	 * mutex protects only manipulation of the active vnode list.
1221 	 * Hence the need to hold the vnode interlock throughout.
1222 	 */
1223 	MNT_ILOCK(mp);
1224 	VI_LOCK(vp);
1225 	if (((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1226 	    ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1227 	    mp->mnt_nvnodelistsize == 0)) &&
1228 	    (vp->v_vflag & VV_FORCEINSMQ) == 0) {
1229 		VI_UNLOCK(vp);
1230 		MNT_IUNLOCK(mp);
1231 		if (dtr != NULL)
1232 			dtr(vp, dtr_arg);
1233 		return (EBUSY);
1234 	}
1235 	vp->v_mount = mp;
1236 	MNT_REF(mp);
1237 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1238 	VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1239 		("neg mount point vnode list size"));
1240 	mp->mnt_nvnodelistsize++;
1241 	KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
1242 	    ("Activating already active vnode"));
1243 	vp->v_iflag |= VI_ACTIVE;
1244 	mtx_lock(&vnode_free_list_mtx);
1245 	TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
1246 	mp->mnt_activevnodelistsize++;
1247 	mtx_unlock(&vnode_free_list_mtx);
1248 	VI_UNLOCK(vp);
1249 	MNT_IUNLOCK(mp);
1250 	return (0);
1251 }
1252 
1253 int
1254 insmntque(struct vnode *vp, struct mount *mp)
1255 {
1256 
1257 	return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1258 }
1259 
1260 /*
1261  * Flush out and invalidate all buffers associated with a bufobj
1262  * Called with the underlying object locked.
1263  */
1264 int
1265 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1266 {
1267 	int error;
1268 
1269 	BO_LOCK(bo);
1270 	if (flags & V_SAVE) {
1271 		error = bufobj_wwait(bo, slpflag, slptimeo);
1272 		if (error) {
1273 			BO_UNLOCK(bo);
1274 			return (error);
1275 		}
1276 		if (bo->bo_dirty.bv_cnt > 0) {
1277 			BO_UNLOCK(bo);
1278 			if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1279 				return (error);
1280 			/*
1281 			 * XXX We could save a lock/unlock if this was only
1282 			 * enabled under INVARIANTS
1283 			 */
1284 			BO_LOCK(bo);
1285 			if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1286 				panic("vinvalbuf: dirty bufs");
1287 		}
1288 	}
1289 	/*
1290 	 * If you alter this loop please notice that interlock is dropped and
1291 	 * reacquired in flushbuflist.  Special care is needed to ensure that
1292 	 * no race conditions occur from this.
1293 	 */
1294 	do {
1295 		error = flushbuflist(&bo->bo_clean,
1296 		    flags, bo, slpflag, slptimeo);
1297 		if (error == 0 && !(flags & V_CLEANONLY))
1298 			error = flushbuflist(&bo->bo_dirty,
1299 			    flags, bo, slpflag, slptimeo);
1300 		if (error != 0 && error != EAGAIN) {
1301 			BO_UNLOCK(bo);
1302 			return (error);
1303 		}
1304 	} while (error != 0);
1305 
1306 	/*
1307 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1308 	 * have write I/O in-progress but if there is a VM object then the
1309 	 * VM object can also have read-I/O in-progress.
1310 	 */
1311 	do {
1312 		bufobj_wwait(bo, 0, 0);
1313 		BO_UNLOCK(bo);
1314 		if (bo->bo_object != NULL) {
1315 			VM_OBJECT_WLOCK(bo->bo_object);
1316 			vm_object_pip_wait(bo->bo_object, "bovlbx");
1317 			VM_OBJECT_WUNLOCK(bo->bo_object);
1318 		}
1319 		BO_LOCK(bo);
1320 	} while (bo->bo_numoutput > 0);
1321 	BO_UNLOCK(bo);
1322 
1323 	/*
1324 	 * Destroy the copy in the VM cache, too.
1325 	 */
1326 	if (bo->bo_object != NULL &&
1327 	    (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) {
1328 		VM_OBJECT_WLOCK(bo->bo_object);
1329 		vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
1330 		    OBJPR_CLEANONLY : 0);
1331 		VM_OBJECT_WUNLOCK(bo->bo_object);
1332 	}
1333 
1334 #ifdef INVARIANTS
1335 	BO_LOCK(bo);
1336 	if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 &&
1337 	    (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1338 		panic("vinvalbuf: flush failed");
1339 	BO_UNLOCK(bo);
1340 #endif
1341 	return (0);
1342 }
1343 
1344 /*
1345  * Flush out and invalidate all buffers associated with a vnode.
1346  * Called with the underlying object locked.
1347  */
1348 int
1349 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1350 {
1351 
1352 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1353 	ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1354 	if (vp->v_object != NULL && vp->v_object->handle != vp)
1355 		return (0);
1356 	return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1357 }
1358 
1359 /*
1360  * Flush out buffers on the specified list.
1361  *
1362  */
1363 static int
1364 flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1365     int slptimeo)
1366 {
1367 	struct buf *bp, *nbp;
1368 	int retval, error;
1369 	daddr_t lblkno;
1370 	b_xflags_t xflags;
1371 
1372 	ASSERT_BO_WLOCKED(bo);
1373 
1374 	retval = 0;
1375 	TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1376 		if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1377 		    ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1378 			continue;
1379 		}
1380 		lblkno = 0;
1381 		xflags = 0;
1382 		if (nbp != NULL) {
1383 			lblkno = nbp->b_lblkno;
1384 			xflags = nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN);
1385 		}
1386 		retval = EAGAIN;
1387 		error = BUF_TIMELOCK(bp,
1388 		    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo),
1389 		    "flushbuf", slpflag, slptimeo);
1390 		if (error) {
1391 			BO_LOCK(bo);
1392 			return (error != ENOLCK ? error : EAGAIN);
1393 		}
1394 		KASSERT(bp->b_bufobj == bo,
1395 		    ("bp %p wrong b_bufobj %p should be %p",
1396 		    bp, bp->b_bufobj, bo));
1397 		if (bp->b_bufobj != bo) {	/* XXX: necessary ? */
1398 			BUF_UNLOCK(bp);
1399 			BO_LOCK(bo);
1400 			return (EAGAIN);
1401 		}
1402 		/*
1403 		 * XXX Since there are no node locks for NFS, I
1404 		 * believe there is a slight chance that a delayed
1405 		 * write will occur while sleeping just above, so
1406 		 * check for it.
1407 		 */
1408 		if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1409 		    (flags & V_SAVE)) {
1410 			bremfree(bp);
1411 			bp->b_flags |= B_ASYNC;
1412 			bwrite(bp);
1413 			BO_LOCK(bo);
1414 			return (EAGAIN);	/* XXX: why not loop ? */
1415 		}
1416 		bremfree(bp);
1417 		bp->b_flags |= (B_INVAL | B_RELBUF);
1418 		bp->b_flags &= ~B_ASYNC;
1419 		brelse(bp);
1420 		BO_LOCK(bo);
1421 		if (nbp != NULL &&
1422 		    (nbp->b_bufobj != bo ||
1423 		     nbp->b_lblkno != lblkno ||
1424 		     (nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1425 			break;			/* nbp invalid */
1426 	}
1427 	return (retval);
1428 }
1429 
1430 /*
1431  * Truncate a file's buffer and pages to a specified length.  This
1432  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1433  * sync activity.
1434  */
1435 int
1436 vtruncbuf(struct vnode *vp, struct ucred *cred, off_t length, int blksize)
1437 {
1438 	struct buf *bp, *nbp;
1439 	int anyfreed;
1440 	int trunclbn;
1441 	struct bufobj *bo;
1442 
1443 	CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1444 	    vp, cred, blksize, (uintmax_t)length);
1445 
1446 	/*
1447 	 * Round up to the *next* lbn.
1448 	 */
1449 	trunclbn = (length + blksize - 1) / blksize;
1450 
1451 	ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1452 restart:
1453 	bo = &vp->v_bufobj;
1454 	BO_LOCK(bo);
1455 	anyfreed = 1;
1456 	for (;anyfreed;) {
1457 		anyfreed = 0;
1458 		TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1459 			if (bp->b_lblkno < trunclbn)
1460 				continue;
1461 			if (BUF_LOCK(bp,
1462 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1463 			    BO_LOCKPTR(bo)) == ENOLCK)
1464 				goto restart;
1465 
1466 			bremfree(bp);
1467 			bp->b_flags |= (B_INVAL | B_RELBUF);
1468 			bp->b_flags &= ~B_ASYNC;
1469 			brelse(bp);
1470 			anyfreed = 1;
1471 
1472 			BO_LOCK(bo);
1473 			if (nbp != NULL &&
1474 			    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1475 			    (nbp->b_vp != vp) ||
1476 			    (nbp->b_flags & B_DELWRI))) {
1477 				BO_UNLOCK(bo);
1478 				goto restart;
1479 			}
1480 		}
1481 
1482 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1483 			if (bp->b_lblkno < trunclbn)
1484 				continue;
1485 			if (BUF_LOCK(bp,
1486 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1487 			    BO_LOCKPTR(bo)) == ENOLCK)
1488 				goto restart;
1489 			bremfree(bp);
1490 			bp->b_flags |= (B_INVAL | B_RELBUF);
1491 			bp->b_flags &= ~B_ASYNC;
1492 			brelse(bp);
1493 			anyfreed = 1;
1494 
1495 			BO_LOCK(bo);
1496 			if (nbp != NULL &&
1497 			    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1498 			    (nbp->b_vp != vp) ||
1499 			    (nbp->b_flags & B_DELWRI) == 0)) {
1500 				BO_UNLOCK(bo);
1501 				goto restart;
1502 			}
1503 		}
1504 	}
1505 
1506 	if (length > 0) {
1507 restartsync:
1508 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1509 			if (bp->b_lblkno > 0)
1510 				continue;
1511 			/*
1512 			 * Since we hold the vnode lock this should only
1513 			 * fail if we're racing with the buf daemon.
1514 			 */
1515 			if (BUF_LOCK(bp,
1516 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1517 			    BO_LOCKPTR(bo)) == ENOLCK) {
1518 				goto restart;
1519 			}
1520 			VNASSERT((bp->b_flags & B_DELWRI), vp,
1521 			    ("buf(%p) on dirty queue without DELWRI", bp));
1522 
1523 			bremfree(bp);
1524 			bawrite(bp);
1525 			BO_LOCK(bo);
1526 			goto restartsync;
1527 		}
1528 	}
1529 
1530 	bufobj_wwait(bo, 0, 0);
1531 	BO_UNLOCK(bo);
1532 	vnode_pager_setsize(vp, length);
1533 
1534 	return (0);
1535 }
1536 
1537 static void
1538 buf_vlist_remove(struct buf *bp)
1539 {
1540 	struct bufv *bv;
1541 
1542 	KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1543 	ASSERT_BO_WLOCKED(bp->b_bufobj);
1544 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1545 	    (BX_VNDIRTY|BX_VNCLEAN),
1546 	    ("buf_vlist_remove: Buf %p is on two lists", bp));
1547 	if (bp->b_xflags & BX_VNDIRTY)
1548 		bv = &bp->b_bufobj->bo_dirty;
1549 	else
1550 		bv = &bp->b_bufobj->bo_clean;
1551 	BUF_PCTRIE_REMOVE(&bv->bv_root, bp->b_lblkno);
1552 	TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1553 	bv->bv_cnt--;
1554 	bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1555 }
1556 
1557 /*
1558  * Add the buffer to the sorted clean or dirty block list.
1559  *
1560  * NOTE: xflags is passed as a constant, optimizing this inline function!
1561  */
1562 static void
1563 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1564 {
1565 	struct bufv *bv;
1566 	struct buf *n;
1567 	int error;
1568 
1569 	ASSERT_BO_WLOCKED(bo);
1570 	KASSERT((bo->bo_flag & BO_DEAD) == 0, ("dead bo %p", bo));
1571 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1572 	    ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1573 	bp->b_xflags |= xflags;
1574 	if (xflags & BX_VNDIRTY)
1575 		bv = &bo->bo_dirty;
1576 	else
1577 		bv = &bo->bo_clean;
1578 
1579 	/*
1580 	 * Keep the list ordered.  Optimize empty list insertion.  Assume
1581 	 * we tend to grow at the tail so lookup_le should usually be cheaper
1582 	 * than _ge.
1583 	 */
1584 	if (bv->bv_cnt == 0 ||
1585 	    bp->b_lblkno > TAILQ_LAST(&bv->bv_hd, buflists)->b_lblkno)
1586 		TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1587 	else if ((n = BUF_PCTRIE_LOOKUP_LE(&bv->bv_root, bp->b_lblkno)) == NULL)
1588 		TAILQ_INSERT_HEAD(&bv->bv_hd, bp, b_bobufs);
1589 	else
1590 		TAILQ_INSERT_AFTER(&bv->bv_hd, n, bp, b_bobufs);
1591 	error = BUF_PCTRIE_INSERT(&bv->bv_root, bp);
1592 	if (error)
1593 		panic("buf_vlist_add:  Preallocated nodes insufficient.");
1594 	bv->bv_cnt++;
1595 }
1596 
1597 /*
1598  * Lookup a buffer using the splay tree.  Note that we specifically avoid
1599  * shadow buffers used in background bitmap writes.
1600  *
1601  * This code isn't quite efficient as it could be because we are maintaining
1602  * two sorted lists and do not know which list the block resides in.
1603  *
1604  * During a "make buildworld" the desired buffer is found at one of
1605  * the roots more than 60% of the time.  Thus, checking both roots
1606  * before performing either splay eliminates unnecessary splays on the
1607  * first tree splayed.
1608  */
1609 struct buf *
1610 gbincore(struct bufobj *bo, daddr_t lblkno)
1611 {
1612 	struct buf *bp;
1613 
1614 	ASSERT_BO_LOCKED(bo);
1615 	bp = BUF_PCTRIE_LOOKUP(&bo->bo_clean.bv_root, lblkno);
1616 	if (bp != NULL)
1617 		return (bp);
1618 	return BUF_PCTRIE_LOOKUP(&bo->bo_dirty.bv_root, lblkno);
1619 }
1620 
1621 /*
1622  * Associate a buffer with a vnode.
1623  */
1624 void
1625 bgetvp(struct vnode *vp, struct buf *bp)
1626 {
1627 	struct bufobj *bo;
1628 
1629 	bo = &vp->v_bufobj;
1630 	ASSERT_BO_WLOCKED(bo);
1631 	VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1632 
1633 	CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1634 	VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1635 	    ("bgetvp: bp already attached! %p", bp));
1636 
1637 	vhold(vp);
1638 	bp->b_vp = vp;
1639 	bp->b_bufobj = bo;
1640 	/*
1641 	 * Insert onto list for new vnode.
1642 	 */
1643 	buf_vlist_add(bp, bo, BX_VNCLEAN);
1644 }
1645 
1646 /*
1647  * Disassociate a buffer from a vnode.
1648  */
1649 void
1650 brelvp(struct buf *bp)
1651 {
1652 	struct bufobj *bo;
1653 	struct vnode *vp;
1654 
1655 	CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1656 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1657 
1658 	/*
1659 	 * Delete from old vnode list, if on one.
1660 	 */
1661 	vp = bp->b_vp;		/* XXX */
1662 	bo = bp->b_bufobj;
1663 	BO_LOCK(bo);
1664 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1665 		buf_vlist_remove(bp);
1666 	else
1667 		panic("brelvp: Buffer %p not on queue.", bp);
1668 	if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1669 		bo->bo_flag &= ~BO_ONWORKLST;
1670 		mtx_lock(&sync_mtx);
1671 		LIST_REMOVE(bo, bo_synclist);
1672 		syncer_worklist_len--;
1673 		mtx_unlock(&sync_mtx);
1674 	}
1675 	bp->b_vp = NULL;
1676 	bp->b_bufobj = NULL;
1677 	BO_UNLOCK(bo);
1678 	vdrop(vp);
1679 }
1680 
1681 /*
1682  * Add an item to the syncer work queue.
1683  */
1684 static void
1685 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1686 {
1687 	int slot;
1688 
1689 	ASSERT_BO_WLOCKED(bo);
1690 
1691 	mtx_lock(&sync_mtx);
1692 	if (bo->bo_flag & BO_ONWORKLST)
1693 		LIST_REMOVE(bo, bo_synclist);
1694 	else {
1695 		bo->bo_flag |= BO_ONWORKLST;
1696 		syncer_worklist_len++;
1697 	}
1698 
1699 	if (delay > syncer_maxdelay - 2)
1700 		delay = syncer_maxdelay - 2;
1701 	slot = (syncer_delayno + delay) & syncer_mask;
1702 
1703 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist);
1704 	mtx_unlock(&sync_mtx);
1705 }
1706 
1707 static int
1708 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1709 {
1710 	int error, len;
1711 
1712 	mtx_lock(&sync_mtx);
1713 	len = syncer_worklist_len - sync_vnode_count;
1714 	mtx_unlock(&sync_mtx);
1715 	error = SYSCTL_OUT(req, &len, sizeof(len));
1716 	return (error);
1717 }
1718 
1719 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1720     sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1721 
1722 static struct proc *updateproc;
1723 static void sched_sync(void);
1724 static struct kproc_desc up_kp = {
1725 	"syncer",
1726 	sched_sync,
1727 	&updateproc
1728 };
1729 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1730 
1731 static int
1732 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1733 {
1734 	struct vnode *vp;
1735 	struct mount *mp;
1736 
1737 	*bo = LIST_FIRST(slp);
1738 	if (*bo == NULL)
1739 		return (0);
1740 	vp = (*bo)->__bo_vnode;	/* XXX */
1741 	if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1742 		return (1);
1743 	/*
1744 	 * We use vhold in case the vnode does not
1745 	 * successfully sync.  vhold prevents the vnode from
1746 	 * going away when we unlock the sync_mtx so that
1747 	 * we can acquire the vnode interlock.
1748 	 */
1749 	vholdl(vp);
1750 	mtx_unlock(&sync_mtx);
1751 	VI_UNLOCK(vp);
1752 	if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1753 		vdrop(vp);
1754 		mtx_lock(&sync_mtx);
1755 		return (*bo == LIST_FIRST(slp));
1756 	}
1757 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1758 	(void) VOP_FSYNC(vp, MNT_LAZY, td);
1759 	VOP_UNLOCK(vp, 0);
1760 	vn_finished_write(mp);
1761 	BO_LOCK(*bo);
1762 	if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1763 		/*
1764 		 * Put us back on the worklist.  The worklist
1765 		 * routine will remove us from our current
1766 		 * position and then add us back in at a later
1767 		 * position.
1768 		 */
1769 		vn_syncer_add_to_worklist(*bo, syncdelay);
1770 	}
1771 	BO_UNLOCK(*bo);
1772 	vdrop(vp);
1773 	mtx_lock(&sync_mtx);
1774 	return (0);
1775 }
1776 
1777 static int first_printf = 1;
1778 
1779 /*
1780  * System filesystem synchronizer daemon.
1781  */
1782 static void
1783 sched_sync(void)
1784 {
1785 	struct synclist *next, *slp;
1786 	struct bufobj *bo;
1787 	long starttime;
1788 	struct thread *td = curthread;
1789 	int last_work_seen;
1790 	int net_worklist_len;
1791 	int syncer_final_iter;
1792 	int error;
1793 
1794 	last_work_seen = 0;
1795 	syncer_final_iter = 0;
1796 	syncer_state = SYNCER_RUNNING;
1797 	starttime = time_uptime;
1798 	td->td_pflags |= TDP_NORUNNINGBUF;
1799 
1800 	EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1801 	    SHUTDOWN_PRI_LAST);
1802 
1803 	mtx_lock(&sync_mtx);
1804 	for (;;) {
1805 		if (syncer_state == SYNCER_FINAL_DELAY &&
1806 		    syncer_final_iter == 0) {
1807 			mtx_unlock(&sync_mtx);
1808 			kproc_suspend_check(td->td_proc);
1809 			mtx_lock(&sync_mtx);
1810 		}
1811 		net_worklist_len = syncer_worklist_len - sync_vnode_count;
1812 		if (syncer_state != SYNCER_RUNNING &&
1813 		    starttime != time_uptime) {
1814 			if (first_printf) {
1815 				printf("\nSyncing disks, vnodes remaining...");
1816 				first_printf = 0;
1817 			}
1818 			printf("%d ", net_worklist_len);
1819 		}
1820 		starttime = time_uptime;
1821 
1822 		/*
1823 		 * Push files whose dirty time has expired.  Be careful
1824 		 * of interrupt race on slp queue.
1825 		 *
1826 		 * Skip over empty worklist slots when shutting down.
1827 		 */
1828 		do {
1829 			slp = &syncer_workitem_pending[syncer_delayno];
1830 			syncer_delayno += 1;
1831 			if (syncer_delayno == syncer_maxdelay)
1832 				syncer_delayno = 0;
1833 			next = &syncer_workitem_pending[syncer_delayno];
1834 			/*
1835 			 * If the worklist has wrapped since the
1836 			 * it was emptied of all but syncer vnodes,
1837 			 * switch to the FINAL_DELAY state and run
1838 			 * for one more second.
1839 			 */
1840 			if (syncer_state == SYNCER_SHUTTING_DOWN &&
1841 			    net_worklist_len == 0 &&
1842 			    last_work_seen == syncer_delayno) {
1843 				syncer_state = SYNCER_FINAL_DELAY;
1844 				syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1845 			}
1846 		} while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1847 		    syncer_worklist_len > 0);
1848 
1849 		/*
1850 		 * Keep track of the last time there was anything
1851 		 * on the worklist other than syncer vnodes.
1852 		 * Return to the SHUTTING_DOWN state if any
1853 		 * new work appears.
1854 		 */
1855 		if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1856 			last_work_seen = syncer_delayno;
1857 		if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1858 			syncer_state = SYNCER_SHUTTING_DOWN;
1859 		while (!LIST_EMPTY(slp)) {
1860 			error = sync_vnode(slp, &bo, td);
1861 			if (error == 1) {
1862 				LIST_REMOVE(bo, bo_synclist);
1863 				LIST_INSERT_HEAD(next, bo, bo_synclist);
1864 				continue;
1865 			}
1866 
1867 			if (first_printf == 0) {
1868 				/*
1869 				 * Drop the sync mutex, because some watchdog
1870 				 * drivers need to sleep while patting
1871 				 */
1872 				mtx_unlock(&sync_mtx);
1873 				wdog_kern_pat(WD_LASTVAL);
1874 				mtx_lock(&sync_mtx);
1875 			}
1876 
1877 		}
1878 		if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1879 			syncer_final_iter--;
1880 		/*
1881 		 * The variable rushjob allows the kernel to speed up the
1882 		 * processing of the filesystem syncer process. A rushjob
1883 		 * value of N tells the filesystem syncer to process the next
1884 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1885 		 * is used by the soft update code to speed up the filesystem
1886 		 * syncer process when the incore state is getting so far
1887 		 * ahead of the disk that the kernel memory pool is being
1888 		 * threatened with exhaustion.
1889 		 */
1890 		if (rushjob > 0) {
1891 			rushjob -= 1;
1892 			continue;
1893 		}
1894 		/*
1895 		 * Just sleep for a short period of time between
1896 		 * iterations when shutting down to allow some I/O
1897 		 * to happen.
1898 		 *
1899 		 * If it has taken us less than a second to process the
1900 		 * current work, then wait. Otherwise start right over
1901 		 * again. We can still lose time if any single round
1902 		 * takes more than two seconds, but it does not really
1903 		 * matter as we are just trying to generally pace the
1904 		 * filesystem activity.
1905 		 */
1906 		if (syncer_state != SYNCER_RUNNING ||
1907 		    time_uptime == starttime) {
1908 			thread_lock(td);
1909 			sched_prio(td, PPAUSE);
1910 			thread_unlock(td);
1911 		}
1912 		if (syncer_state != SYNCER_RUNNING)
1913 			cv_timedwait(&sync_wakeup, &sync_mtx,
1914 			    hz / SYNCER_SHUTDOWN_SPEEDUP);
1915 		else if (time_uptime == starttime)
1916 			cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1917 	}
1918 }
1919 
1920 /*
1921  * Request the syncer daemon to speed up its work.
1922  * We never push it to speed up more than half of its
1923  * normal turn time, otherwise it could take over the cpu.
1924  */
1925 int
1926 speedup_syncer(void)
1927 {
1928 	int ret = 0;
1929 
1930 	mtx_lock(&sync_mtx);
1931 	if (rushjob < syncdelay / 2) {
1932 		rushjob += 1;
1933 		stat_rush_requests += 1;
1934 		ret = 1;
1935 	}
1936 	mtx_unlock(&sync_mtx);
1937 	cv_broadcast(&sync_wakeup);
1938 	return (ret);
1939 }
1940 
1941 /*
1942  * Tell the syncer to speed up its work and run though its work
1943  * list several times, then tell it to shut down.
1944  */
1945 static void
1946 syncer_shutdown(void *arg, int howto)
1947 {
1948 
1949 	if (howto & RB_NOSYNC)
1950 		return;
1951 	mtx_lock(&sync_mtx);
1952 	syncer_state = SYNCER_SHUTTING_DOWN;
1953 	rushjob = 0;
1954 	mtx_unlock(&sync_mtx);
1955 	cv_broadcast(&sync_wakeup);
1956 	kproc_shutdown(arg, howto);
1957 }
1958 
1959 void
1960 syncer_suspend(void)
1961 {
1962 
1963 	syncer_shutdown(updateproc, 0);
1964 }
1965 
1966 void
1967 syncer_resume(void)
1968 {
1969 
1970 	mtx_lock(&sync_mtx);
1971 	first_printf = 1;
1972 	syncer_state = SYNCER_RUNNING;
1973 	mtx_unlock(&sync_mtx);
1974 	cv_broadcast(&sync_wakeup);
1975 	kproc_resume(updateproc);
1976 }
1977 
1978 /*
1979  * Reassign a buffer from one vnode to another.
1980  * Used to assign file specific control information
1981  * (indirect blocks) to the vnode to which they belong.
1982  */
1983 void
1984 reassignbuf(struct buf *bp)
1985 {
1986 	struct vnode *vp;
1987 	struct bufobj *bo;
1988 	int delay;
1989 #ifdef INVARIANTS
1990 	struct bufv *bv;
1991 #endif
1992 
1993 	vp = bp->b_vp;
1994 	bo = bp->b_bufobj;
1995 	++reassignbufcalls;
1996 
1997 	CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1998 	    bp, bp->b_vp, bp->b_flags);
1999 	/*
2000 	 * B_PAGING flagged buffers cannot be reassigned because their vp
2001 	 * is not fully linked in.
2002 	 */
2003 	if (bp->b_flags & B_PAGING)
2004 		panic("cannot reassign paging buffer");
2005 
2006 	/*
2007 	 * Delete from old vnode list, if on one.
2008 	 */
2009 	BO_LOCK(bo);
2010 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
2011 		buf_vlist_remove(bp);
2012 	else
2013 		panic("reassignbuf: Buffer %p not on queue.", bp);
2014 	/*
2015 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
2016 	 * of clean buffers.
2017 	 */
2018 	if (bp->b_flags & B_DELWRI) {
2019 		if ((bo->bo_flag & BO_ONWORKLST) == 0) {
2020 			switch (vp->v_type) {
2021 			case VDIR:
2022 				delay = dirdelay;
2023 				break;
2024 			case VCHR:
2025 				delay = metadelay;
2026 				break;
2027 			default:
2028 				delay = filedelay;
2029 			}
2030 			vn_syncer_add_to_worklist(bo, delay);
2031 		}
2032 		buf_vlist_add(bp, bo, BX_VNDIRTY);
2033 	} else {
2034 		buf_vlist_add(bp, bo, BX_VNCLEAN);
2035 
2036 		if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
2037 			mtx_lock(&sync_mtx);
2038 			LIST_REMOVE(bo, bo_synclist);
2039 			syncer_worklist_len--;
2040 			mtx_unlock(&sync_mtx);
2041 			bo->bo_flag &= ~BO_ONWORKLST;
2042 		}
2043 	}
2044 #ifdef INVARIANTS
2045 	bv = &bo->bo_clean;
2046 	bp = TAILQ_FIRST(&bv->bv_hd);
2047 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2048 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2049 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
2050 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2051 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2052 	bv = &bo->bo_dirty;
2053 	bp = TAILQ_FIRST(&bv->bv_hd);
2054 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2055 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2056 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
2057 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2058 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2059 #endif
2060 	BO_UNLOCK(bo);
2061 }
2062 
2063 /*
2064  * Increment the use and hold counts on the vnode, taking care to reference
2065  * the driver's usecount if this is a chardev.  The vholdl() will remove
2066  * the vnode from the free list if it is presently free.  Requires the
2067  * vnode interlock and returns with it held.
2068  */
2069 static void
2070 v_incr_usecount(struct vnode *vp)
2071 {
2072 
2073 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2074 	vholdl(vp);
2075 	vp->v_usecount++;
2076 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2077 		dev_lock();
2078 		vp->v_rdev->si_usecount++;
2079 		dev_unlock();
2080 	}
2081 }
2082 
2083 /*
2084  * Turn a holdcnt into a use+holdcnt such that only one call to
2085  * v_decr_usecount is needed.
2086  */
2087 static void
2088 v_upgrade_usecount(struct vnode *vp)
2089 {
2090 
2091 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2092 	vp->v_usecount++;
2093 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2094 		dev_lock();
2095 		vp->v_rdev->si_usecount++;
2096 		dev_unlock();
2097 	}
2098 }
2099 
2100 /*
2101  * Decrement the vnode use and hold count along with the driver's usecount
2102  * if this is a chardev.  The vdropl() below releases the vnode interlock
2103  * as it may free the vnode.
2104  */
2105 static void
2106 v_decr_usecount(struct vnode *vp)
2107 {
2108 
2109 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2110 	VNASSERT(vp->v_usecount > 0, vp,
2111 	    ("v_decr_usecount: negative usecount"));
2112 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2113 	vp->v_usecount--;
2114 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2115 		dev_lock();
2116 		vp->v_rdev->si_usecount--;
2117 		dev_unlock();
2118 	}
2119 	vdropl(vp);
2120 }
2121 
2122 /*
2123  * Decrement only the use count and driver use count.  This is intended to
2124  * be paired with a follow on vdropl() to release the remaining hold count.
2125  * In this way we may vgone() a vnode with a 0 usecount without risk of
2126  * having it end up on a free list because the hold count is kept above 0.
2127  */
2128 static void
2129 v_decr_useonly(struct vnode *vp)
2130 {
2131 
2132 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2133 	VNASSERT(vp->v_usecount > 0, vp,
2134 	    ("v_decr_useonly: negative usecount"));
2135 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2136 	vp->v_usecount--;
2137 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2138 		dev_lock();
2139 		vp->v_rdev->si_usecount--;
2140 		dev_unlock();
2141 	}
2142 }
2143 
2144 /*
2145  * Grab a particular vnode from the free list, increment its
2146  * reference count and lock it.  VI_DOOMED is set if the vnode
2147  * is being destroyed.  Only callers who specify LK_RETRY will
2148  * see doomed vnodes.  If inactive processing was delayed in
2149  * vput try to do it here.
2150  */
2151 int
2152 vget(struct vnode *vp, int flags, struct thread *td)
2153 {
2154 	int error;
2155 
2156 	error = 0;
2157 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2158 	    ("vget: invalid lock operation"));
2159 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2160 
2161 	if ((flags & LK_INTERLOCK) == 0)
2162 		VI_LOCK(vp);
2163 	vholdl(vp);
2164 	if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2165 		vdrop(vp);
2166 		CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2167 		    vp);
2168 		return (error);
2169 	}
2170 	if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2171 		panic("vget: vn_lock failed to return ENOENT\n");
2172 	VI_LOCK(vp);
2173 	/* Upgrade our holdcnt to a usecount. */
2174 	v_upgrade_usecount(vp);
2175 	/*
2176 	 * We don't guarantee that any particular close will
2177 	 * trigger inactive processing so just make a best effort
2178 	 * here at preventing a reference to a removed file.  If
2179 	 * we don't succeed no harm is done.
2180 	 */
2181 	if (vp->v_iflag & VI_OWEINACT) {
2182 		if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2183 		    (flags & LK_NOWAIT) == 0)
2184 			vinactive(vp, td);
2185 		vp->v_iflag &= ~VI_OWEINACT;
2186 	}
2187 	VI_UNLOCK(vp);
2188 	return (0);
2189 }
2190 
2191 /*
2192  * Increase the reference count of a vnode.
2193  */
2194 void
2195 vref(struct vnode *vp)
2196 {
2197 
2198 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2199 	VI_LOCK(vp);
2200 	v_incr_usecount(vp);
2201 	VI_UNLOCK(vp);
2202 }
2203 
2204 /*
2205  * Return reference count of a vnode.
2206  *
2207  * The results of this call are only guaranteed when some mechanism other
2208  * than the VI lock is used to stop other processes from gaining references
2209  * to the vnode.  This may be the case if the caller holds the only reference.
2210  * This is also useful when stale data is acceptable as race conditions may
2211  * be accounted for by some other means.
2212  */
2213 int
2214 vrefcnt(struct vnode *vp)
2215 {
2216 	int usecnt;
2217 
2218 	VI_LOCK(vp);
2219 	usecnt = vp->v_usecount;
2220 	VI_UNLOCK(vp);
2221 
2222 	return (usecnt);
2223 }
2224 
2225 #define	VPUTX_VRELE	1
2226 #define	VPUTX_VPUT	2
2227 #define	VPUTX_VUNREF	3
2228 
2229 static void
2230 vputx(struct vnode *vp, int func)
2231 {
2232 	int error;
2233 
2234 	KASSERT(vp != NULL, ("vputx: null vp"));
2235 	if (func == VPUTX_VUNREF)
2236 		ASSERT_VOP_LOCKED(vp, "vunref");
2237 	else if (func == VPUTX_VPUT)
2238 		ASSERT_VOP_LOCKED(vp, "vput");
2239 	else
2240 		KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2241 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2242 	VI_LOCK(vp);
2243 
2244 	/* Skip this v_writecount check if we're going to panic below. */
2245 	VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2246 	    ("vputx: missed vn_close"));
2247 	error = 0;
2248 
2249 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2250 	    vp->v_usecount == 1)) {
2251 		if (func == VPUTX_VPUT)
2252 			VOP_UNLOCK(vp, 0);
2253 		v_decr_usecount(vp);
2254 		return;
2255 	}
2256 
2257 	if (vp->v_usecount != 1) {
2258 		vprint("vputx: negative ref count", vp);
2259 		panic("vputx: negative ref cnt");
2260 	}
2261 	CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2262 	/*
2263 	 * We want to hold the vnode until the inactive finishes to
2264 	 * prevent vgone() races.  We drop the use count here and the
2265 	 * hold count below when we're done.
2266 	 */
2267 	v_decr_useonly(vp);
2268 	/*
2269 	 * We must call VOP_INACTIVE with the node locked. Mark
2270 	 * as VI_DOINGINACT to avoid recursion.
2271 	 */
2272 	vp->v_iflag |= VI_OWEINACT;
2273 	switch (func) {
2274 	case VPUTX_VRELE:
2275 		error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2276 		VI_LOCK(vp);
2277 		break;
2278 	case VPUTX_VPUT:
2279 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2280 			error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
2281 			    LK_NOWAIT);
2282 			VI_LOCK(vp);
2283 		}
2284 		break;
2285 	case VPUTX_VUNREF:
2286 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2287 			error = VOP_LOCK(vp, LK_TRYUPGRADE | LK_INTERLOCK);
2288 			VI_LOCK(vp);
2289 		}
2290 		break;
2291 	}
2292 	if (vp->v_usecount > 0)
2293 		vp->v_iflag &= ~VI_OWEINACT;
2294 	if (error == 0) {
2295 		if (vp->v_iflag & VI_OWEINACT)
2296 			vinactive(vp, curthread);
2297 		if (func != VPUTX_VUNREF)
2298 			VOP_UNLOCK(vp, 0);
2299 	}
2300 	vdropl(vp);
2301 }
2302 
2303 /*
2304  * Vnode put/release.
2305  * If count drops to zero, call inactive routine and return to freelist.
2306  */
2307 void
2308 vrele(struct vnode *vp)
2309 {
2310 
2311 	vputx(vp, VPUTX_VRELE);
2312 }
2313 
2314 /*
2315  * Release an already locked vnode.  This give the same effects as
2316  * unlock+vrele(), but takes less time and avoids releasing and
2317  * re-aquiring the lock (as vrele() acquires the lock internally.)
2318  */
2319 void
2320 vput(struct vnode *vp)
2321 {
2322 
2323 	vputx(vp, VPUTX_VPUT);
2324 }
2325 
2326 /*
2327  * Release an exclusively locked vnode. Do not unlock the vnode lock.
2328  */
2329 void
2330 vunref(struct vnode *vp)
2331 {
2332 
2333 	vputx(vp, VPUTX_VUNREF);
2334 }
2335 
2336 /*
2337  * Somebody doesn't want the vnode recycled.
2338  */
2339 void
2340 vhold(struct vnode *vp)
2341 {
2342 
2343 	VI_LOCK(vp);
2344 	vholdl(vp);
2345 	VI_UNLOCK(vp);
2346 }
2347 
2348 /*
2349  * Increase the hold count and activate if this is the first reference.
2350  */
2351 void
2352 vholdl(struct vnode *vp)
2353 {
2354 	struct mount *mp;
2355 
2356 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2357 #ifdef INVARIANTS
2358 	/* getnewvnode() calls v_incr_usecount() without holding interlock. */
2359 	if (vp->v_type != VNON || vp->v_data != NULL) {
2360 		ASSERT_VI_LOCKED(vp, "vholdl");
2361 		VNASSERT(vp->v_holdcnt > 0 || (vp->v_iflag & VI_FREE) != 0,
2362 		    vp, ("vholdl: free vnode is held"));
2363 	}
2364 #endif
2365 	vp->v_holdcnt++;
2366 	if ((vp->v_iflag & VI_FREE) == 0)
2367 		return;
2368 	VNASSERT(vp->v_holdcnt == 1, vp, ("vholdl: wrong hold count"));
2369 	VNASSERT(vp->v_op != NULL, vp, ("vholdl: vnode already reclaimed."));
2370 	/*
2371 	 * Remove a vnode from the free list, mark it as in use,
2372 	 * and put it on the active list.
2373 	 */
2374 	mtx_lock(&vnode_free_list_mtx);
2375 	TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
2376 	freevnodes--;
2377 	vp->v_iflag &= ~(VI_FREE|VI_AGE);
2378 	KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
2379 	    ("Activating already active vnode"));
2380 	vp->v_iflag |= VI_ACTIVE;
2381 	mp = vp->v_mount;
2382 	TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
2383 	mp->mnt_activevnodelistsize++;
2384 	mtx_unlock(&vnode_free_list_mtx);
2385 }
2386 
2387 /*
2388  * Note that there is one less who cares about this vnode.
2389  * vdrop() is the opposite of vhold().
2390  */
2391 void
2392 vdrop(struct vnode *vp)
2393 {
2394 
2395 	VI_LOCK(vp);
2396 	vdropl(vp);
2397 }
2398 
2399 /*
2400  * Drop the hold count of the vnode.  If this is the last reference to
2401  * the vnode we place it on the free list unless it has been vgone'd
2402  * (marked VI_DOOMED) in which case we will free it.
2403  */
2404 void
2405 vdropl(struct vnode *vp)
2406 {
2407 	struct bufobj *bo;
2408 	struct mount *mp;
2409 	int active;
2410 
2411 	ASSERT_VI_LOCKED(vp, "vdropl");
2412 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2413 	if (vp->v_holdcnt <= 0)
2414 		panic("vdrop: holdcnt %d", vp->v_holdcnt);
2415 	vp->v_holdcnt--;
2416 	VNASSERT(vp->v_holdcnt >= vp->v_usecount, vp,
2417 	    ("hold count less than use count"));
2418 	if (vp->v_holdcnt > 0) {
2419 		VI_UNLOCK(vp);
2420 		return;
2421 	}
2422 	if ((vp->v_iflag & VI_DOOMED) == 0) {
2423 		/*
2424 		 * Mark a vnode as free: remove it from its active list
2425 		 * and put it up for recycling on the freelist.
2426 		 */
2427 		VNASSERT(vp->v_op != NULL, vp,
2428 		    ("vdropl: vnode already reclaimed."));
2429 		VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2430 		    ("vnode already free"));
2431 		VNASSERT(vp->v_holdcnt == 0, vp,
2432 		    ("vdropl: freeing when we shouldn't"));
2433 		active = vp->v_iflag & VI_ACTIVE;
2434 		vp->v_iflag &= ~VI_ACTIVE;
2435 		mp = vp->v_mount;
2436 		mtx_lock(&vnode_free_list_mtx);
2437 		if (active) {
2438 			TAILQ_REMOVE(&mp->mnt_activevnodelist, vp,
2439 			    v_actfreelist);
2440 			mp->mnt_activevnodelistsize--;
2441 		}
2442 		if (vp->v_iflag & VI_AGE) {
2443 			TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_actfreelist);
2444 		} else {
2445 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
2446 		}
2447 		freevnodes++;
2448 		vp->v_iflag &= ~VI_AGE;
2449 		vp->v_iflag |= VI_FREE;
2450 		mtx_unlock(&vnode_free_list_mtx);
2451 		VI_UNLOCK(vp);
2452 		return;
2453 	}
2454 	/*
2455 	 * The vnode has been marked for destruction, so free it.
2456 	 */
2457 	CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp);
2458 	atomic_subtract_long(&numvnodes, 1);
2459 	bo = &vp->v_bufobj;
2460 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2461 	    ("cleaned vnode still on the free list."));
2462 	VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
2463 	VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
2464 	VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
2465 	VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
2466 	VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
2467 	VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
2468 	VNASSERT(pctrie_is_empty(&bo->bo_clean.bv_root), vp,
2469 	    ("clean blk trie not empty"));
2470 	VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
2471 	VNASSERT(pctrie_is_empty(&bo->bo_dirty.bv_root), vp,
2472 	    ("dirty blk trie not empty"));
2473 	VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
2474 	VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
2475 	VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
2476 	VI_UNLOCK(vp);
2477 #ifdef MAC
2478 	mac_vnode_destroy(vp);
2479 #endif
2480 	if (vp->v_pollinfo != NULL)
2481 		destroy_vpollinfo(vp->v_pollinfo);
2482 #ifdef INVARIANTS
2483 	/* XXX Elsewhere we detect an already freed vnode via NULL v_op. */
2484 	vp->v_op = NULL;
2485 #endif
2486 	rangelock_destroy(&vp->v_rl);
2487 	lockdestroy(vp->v_vnlock);
2488 	mtx_destroy(&vp->v_interlock);
2489 	rw_destroy(BO_LOCKPTR(bo));
2490 	uma_zfree(vnode_zone, vp);
2491 }
2492 
2493 /*
2494  * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2495  * flags.  DOINGINACT prevents us from recursing in calls to vinactive.
2496  * OWEINACT tracks whether a vnode missed a call to inactive due to a
2497  * failed lock upgrade.
2498  */
2499 void
2500 vinactive(struct vnode *vp, struct thread *td)
2501 {
2502 	struct vm_object *obj;
2503 
2504 	ASSERT_VOP_ELOCKED(vp, "vinactive");
2505 	ASSERT_VI_LOCKED(vp, "vinactive");
2506 	VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2507 	    ("vinactive: recursed on VI_DOINGINACT"));
2508 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2509 	vp->v_iflag |= VI_DOINGINACT;
2510 	vp->v_iflag &= ~VI_OWEINACT;
2511 	VI_UNLOCK(vp);
2512 	/*
2513 	 * Before moving off the active list, we must be sure that any
2514 	 * modified pages are on the vnode's dirty list since these will
2515 	 * no longer be checked once the vnode is on the inactive list.
2516 	 * Because the vnode vm object keeps a hold reference on the vnode
2517 	 * if there is at least one resident non-cached page, the vnode
2518 	 * cannot leave the active list without the page cleanup done.
2519 	 */
2520 	obj = vp->v_object;
2521 	if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
2522 		VM_OBJECT_WLOCK(obj);
2523 		vm_object_page_clean(obj, 0, 0, OBJPC_NOSYNC);
2524 		VM_OBJECT_WUNLOCK(obj);
2525 	}
2526 	VOP_INACTIVE(vp, td);
2527 	VI_LOCK(vp);
2528 	VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2529 	    ("vinactive: lost VI_DOINGINACT"));
2530 	vp->v_iflag &= ~VI_DOINGINACT;
2531 }
2532 
2533 /*
2534  * Remove any vnodes in the vnode table belonging to mount point mp.
2535  *
2536  * If FORCECLOSE is not specified, there should not be any active ones,
2537  * return error if any are found (nb: this is a user error, not a
2538  * system error). If FORCECLOSE is specified, detach any active vnodes
2539  * that are found.
2540  *
2541  * If WRITECLOSE is set, only flush out regular file vnodes open for
2542  * writing.
2543  *
2544  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2545  *
2546  * `rootrefs' specifies the base reference count for the root vnode
2547  * of this filesystem. The root vnode is considered busy if its
2548  * v_usecount exceeds this value. On a successful return, vflush(, td)
2549  * will call vrele() on the root vnode exactly rootrefs times.
2550  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2551  * be zero.
2552  */
2553 #ifdef DIAGNOSTIC
2554 static int busyprt = 0;		/* print out busy vnodes */
2555 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
2556 #endif
2557 
2558 int
2559 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
2560 {
2561 	struct vnode *vp, *mvp, *rootvp = NULL;
2562 	struct vattr vattr;
2563 	int busy = 0, error;
2564 
2565 	CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2566 	    rootrefs, flags);
2567 	if (rootrefs > 0) {
2568 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2569 		    ("vflush: bad args"));
2570 		/*
2571 		 * Get the filesystem root vnode. We can vput() it
2572 		 * immediately, since with rootrefs > 0, it won't go away.
2573 		 */
2574 		if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2575 			CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2576 			    __func__, error);
2577 			return (error);
2578 		}
2579 		vput(rootvp);
2580 	}
2581 loop:
2582 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
2583 		vholdl(vp);
2584 		error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2585 		if (error) {
2586 			vdrop(vp);
2587 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2588 			goto loop;
2589 		}
2590 		/*
2591 		 * Skip over a vnodes marked VV_SYSTEM.
2592 		 */
2593 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2594 			VOP_UNLOCK(vp, 0);
2595 			vdrop(vp);
2596 			continue;
2597 		}
2598 		/*
2599 		 * If WRITECLOSE is set, flush out unlinked but still open
2600 		 * files (even if open only for reading) and regular file
2601 		 * vnodes open for writing.
2602 		 */
2603 		if (flags & WRITECLOSE) {
2604 			if (vp->v_object != NULL) {
2605 				VM_OBJECT_WLOCK(vp->v_object);
2606 				vm_object_page_clean(vp->v_object, 0, 0, 0);
2607 				VM_OBJECT_WUNLOCK(vp->v_object);
2608 			}
2609 			error = VOP_FSYNC(vp, MNT_WAIT, td);
2610 			if (error != 0) {
2611 				VOP_UNLOCK(vp, 0);
2612 				vdrop(vp);
2613 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2614 				return (error);
2615 			}
2616 			error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2617 			VI_LOCK(vp);
2618 
2619 			if ((vp->v_type == VNON ||
2620 			    (error == 0 && vattr.va_nlink > 0)) &&
2621 			    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2622 				VOP_UNLOCK(vp, 0);
2623 				vdropl(vp);
2624 				continue;
2625 			}
2626 		} else
2627 			VI_LOCK(vp);
2628 		/*
2629 		 * With v_usecount == 0, all we need to do is clear out the
2630 		 * vnode data structures and we are done.
2631 		 *
2632 		 * If FORCECLOSE is set, forcibly close the vnode.
2633 		 */
2634 		if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2635 			VNASSERT(vp->v_usecount == 0 ||
2636 			    (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2637 			    ("device VNODE %p is FORCECLOSED", vp));
2638 			vgonel(vp);
2639 		} else {
2640 			busy++;
2641 #ifdef DIAGNOSTIC
2642 			if (busyprt)
2643 				vprint("vflush: busy vnode", vp);
2644 #endif
2645 		}
2646 		VOP_UNLOCK(vp, 0);
2647 		vdropl(vp);
2648 	}
2649 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2650 		/*
2651 		 * If just the root vnode is busy, and if its refcount
2652 		 * is equal to `rootrefs', then go ahead and kill it.
2653 		 */
2654 		VI_LOCK(rootvp);
2655 		KASSERT(busy > 0, ("vflush: not busy"));
2656 		VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2657 		    ("vflush: usecount %d < rootrefs %d",
2658 		     rootvp->v_usecount, rootrefs));
2659 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2660 			VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2661 			vgone(rootvp);
2662 			VOP_UNLOCK(rootvp, 0);
2663 			busy = 0;
2664 		} else
2665 			VI_UNLOCK(rootvp);
2666 	}
2667 	if (busy) {
2668 		CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2669 		    busy);
2670 		return (EBUSY);
2671 	}
2672 	for (; rootrefs > 0; rootrefs--)
2673 		vrele(rootvp);
2674 	return (0);
2675 }
2676 
2677 /*
2678  * Recycle an unused vnode to the front of the free list.
2679  */
2680 int
2681 vrecycle(struct vnode *vp)
2682 {
2683 	int recycled;
2684 
2685 	ASSERT_VOP_ELOCKED(vp, "vrecycle");
2686 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2687 	recycled = 0;
2688 	VI_LOCK(vp);
2689 	if (vp->v_usecount == 0) {
2690 		recycled = 1;
2691 		vgonel(vp);
2692 	}
2693 	VI_UNLOCK(vp);
2694 	return (recycled);
2695 }
2696 
2697 /*
2698  * Eliminate all activity associated with a vnode
2699  * in preparation for reuse.
2700  */
2701 void
2702 vgone(struct vnode *vp)
2703 {
2704 	VI_LOCK(vp);
2705 	vgonel(vp);
2706 	VI_UNLOCK(vp);
2707 }
2708 
2709 static void
2710 notify_lowervp_vfs_dummy(struct mount *mp __unused,
2711     struct vnode *lowervp __unused)
2712 {
2713 }
2714 
2715 /*
2716  * Notify upper mounts about reclaimed or unlinked vnode.
2717  */
2718 void
2719 vfs_notify_upper(struct vnode *vp, int event)
2720 {
2721 	static struct vfsops vgonel_vfsops = {
2722 		.vfs_reclaim_lowervp = notify_lowervp_vfs_dummy,
2723 		.vfs_unlink_lowervp = notify_lowervp_vfs_dummy,
2724 	};
2725 	struct mount *mp, *ump, *mmp;
2726 
2727 	mp = vp->v_mount;
2728 	if (mp == NULL)
2729 		return;
2730 
2731 	MNT_ILOCK(mp);
2732 	if (TAILQ_EMPTY(&mp->mnt_uppers))
2733 		goto unlock;
2734 	MNT_IUNLOCK(mp);
2735 	mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO);
2736 	mmp->mnt_op = &vgonel_vfsops;
2737 	mmp->mnt_kern_flag |= MNTK_MARKER;
2738 	MNT_ILOCK(mp);
2739 	mp->mnt_kern_flag |= MNTK_VGONE_UPPER;
2740 	for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) {
2741 		if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) {
2742 			ump = TAILQ_NEXT(ump, mnt_upper_link);
2743 			continue;
2744 		}
2745 		TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link);
2746 		MNT_IUNLOCK(mp);
2747 		switch (event) {
2748 		case VFS_NOTIFY_UPPER_RECLAIM:
2749 			VFS_RECLAIM_LOWERVP(ump, vp);
2750 			break;
2751 		case VFS_NOTIFY_UPPER_UNLINK:
2752 			VFS_UNLINK_LOWERVP(ump, vp);
2753 			break;
2754 		default:
2755 			KASSERT(0, ("invalid event %d", event));
2756 			break;
2757 		}
2758 		MNT_ILOCK(mp);
2759 		ump = TAILQ_NEXT(mmp, mnt_upper_link);
2760 		TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link);
2761 	}
2762 	free(mmp, M_TEMP);
2763 	mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER;
2764 	if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) {
2765 		mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER;
2766 		wakeup(&mp->mnt_uppers);
2767 	}
2768 unlock:
2769 	MNT_IUNLOCK(mp);
2770 }
2771 
2772 /*
2773  * vgone, with the vp interlock held.
2774  */
2775 void
2776 vgonel(struct vnode *vp)
2777 {
2778 	struct thread *td;
2779 	int oweinact;
2780 	int active;
2781 	struct mount *mp;
2782 
2783 	ASSERT_VOP_ELOCKED(vp, "vgonel");
2784 	ASSERT_VI_LOCKED(vp, "vgonel");
2785 	VNASSERT(vp->v_holdcnt, vp,
2786 	    ("vgonel: vp %p has no reference.", vp));
2787 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2788 	td = curthread;
2789 
2790 	/*
2791 	 * Don't vgonel if we're already doomed.
2792 	 */
2793 	if (vp->v_iflag & VI_DOOMED)
2794 		return;
2795 	vp->v_iflag |= VI_DOOMED;
2796 
2797 	/*
2798 	 * Check to see if the vnode is in use.  If so, we have to call
2799 	 * VOP_CLOSE() and VOP_INACTIVE().
2800 	 */
2801 	active = vp->v_usecount;
2802 	oweinact = (vp->v_iflag & VI_OWEINACT);
2803 	VI_UNLOCK(vp);
2804 	vfs_notify_upper(vp, VFS_NOTIFY_UPPER_RECLAIM);
2805 
2806 	/*
2807 	 * If purging an active vnode, it must be closed and
2808 	 * deactivated before being reclaimed.
2809 	 */
2810 	if (active)
2811 		VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2812 	if (oweinact || active) {
2813 		VI_LOCK(vp);
2814 		if ((vp->v_iflag & VI_DOINGINACT) == 0)
2815 			vinactive(vp, td);
2816 		VI_UNLOCK(vp);
2817 	}
2818 	if (vp->v_type == VSOCK)
2819 		vfs_unp_reclaim(vp);
2820 
2821 	/*
2822 	 * Clean out any buffers associated with the vnode.
2823 	 * If the flush fails, just toss the buffers.
2824 	 */
2825 	mp = NULL;
2826 	if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2827 		(void) vn_start_secondary_write(vp, &mp, V_WAIT);
2828 	if (vinvalbuf(vp, V_SAVE, 0, 0) != 0) {
2829 		while (vinvalbuf(vp, 0, 0, 0) != 0)
2830 			;
2831 	}
2832 #ifdef INVARIANTS
2833 	BO_LOCK(&vp->v_bufobj);
2834 	KASSERT(TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd) &&
2835 	    vp->v_bufobj.bo_dirty.bv_cnt == 0 &&
2836 	    TAILQ_EMPTY(&vp->v_bufobj.bo_clean.bv_hd) &&
2837 	    vp->v_bufobj.bo_clean.bv_cnt == 0,
2838 	    ("vp %p bufobj not invalidated", vp));
2839 	vp->v_bufobj.bo_flag |= BO_DEAD;
2840 	BO_UNLOCK(&vp->v_bufobj);
2841 #endif
2842 
2843 	/*
2844 	 * Reclaim the vnode.
2845 	 */
2846 	if (VOP_RECLAIM(vp, td))
2847 		panic("vgone: cannot reclaim");
2848 	if (mp != NULL)
2849 		vn_finished_secondary_write(mp);
2850 	VNASSERT(vp->v_object == NULL, vp,
2851 	    ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2852 	/*
2853 	 * Clear the advisory locks and wake up waiting threads.
2854 	 */
2855 	(void)VOP_ADVLOCKPURGE(vp);
2856 	/*
2857 	 * Delete from old mount point vnode list.
2858 	 */
2859 	delmntque(vp);
2860 	cache_purge(vp);
2861 	/*
2862 	 * Done with purge, reset to the standard lock and invalidate
2863 	 * the vnode.
2864 	 */
2865 	VI_LOCK(vp);
2866 	vp->v_vnlock = &vp->v_lock;
2867 	vp->v_op = &dead_vnodeops;
2868 	vp->v_tag = "none";
2869 	vp->v_type = VBAD;
2870 }
2871 
2872 /*
2873  * Calculate the total number of references to a special device.
2874  */
2875 int
2876 vcount(struct vnode *vp)
2877 {
2878 	int count;
2879 
2880 	dev_lock();
2881 	count = vp->v_rdev->si_usecount;
2882 	dev_unlock();
2883 	return (count);
2884 }
2885 
2886 /*
2887  * Same as above, but using the struct cdev *as argument
2888  */
2889 int
2890 count_dev(struct cdev *dev)
2891 {
2892 	int count;
2893 
2894 	dev_lock();
2895 	count = dev->si_usecount;
2896 	dev_unlock();
2897 	return(count);
2898 }
2899 
2900 /*
2901  * Print out a description of a vnode.
2902  */
2903 static char *typename[] =
2904 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2905  "VMARKER"};
2906 
2907 void
2908 vn_printf(struct vnode *vp, const char *fmt, ...)
2909 {
2910 	va_list ap;
2911 	char buf[256], buf2[16];
2912 	u_long flags;
2913 
2914 	va_start(ap, fmt);
2915 	vprintf(fmt, ap);
2916 	va_end(ap);
2917 	printf("%p: ", (void *)vp);
2918 	printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2919 	printf("    usecount %d, writecount %d, refcount %d mountedhere %p\n",
2920 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2921 	buf[0] = '\0';
2922 	buf[1] = '\0';
2923 	if (vp->v_vflag & VV_ROOT)
2924 		strlcat(buf, "|VV_ROOT", sizeof(buf));
2925 	if (vp->v_vflag & VV_ISTTY)
2926 		strlcat(buf, "|VV_ISTTY", sizeof(buf));
2927 	if (vp->v_vflag & VV_NOSYNC)
2928 		strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2929 	if (vp->v_vflag & VV_ETERNALDEV)
2930 		strlcat(buf, "|VV_ETERNALDEV", sizeof(buf));
2931 	if (vp->v_vflag & VV_CACHEDLABEL)
2932 		strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2933 	if (vp->v_vflag & VV_TEXT)
2934 		strlcat(buf, "|VV_TEXT", sizeof(buf));
2935 	if (vp->v_vflag & VV_COPYONWRITE)
2936 		strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2937 	if (vp->v_vflag & VV_SYSTEM)
2938 		strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2939 	if (vp->v_vflag & VV_PROCDEP)
2940 		strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2941 	if (vp->v_vflag & VV_NOKNOTE)
2942 		strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2943 	if (vp->v_vflag & VV_DELETED)
2944 		strlcat(buf, "|VV_DELETED", sizeof(buf));
2945 	if (vp->v_vflag & VV_MD)
2946 		strlcat(buf, "|VV_MD", sizeof(buf));
2947 	if (vp->v_vflag & VV_FORCEINSMQ)
2948 		strlcat(buf, "|VV_FORCEINSMQ", sizeof(buf));
2949 	flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | VV_ETERNALDEV |
2950 	    VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2951 	    VV_NOKNOTE | VV_DELETED | VV_MD | VV_FORCEINSMQ);
2952 	if (flags != 0) {
2953 		snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2954 		strlcat(buf, buf2, sizeof(buf));
2955 	}
2956 	if (vp->v_iflag & VI_MOUNT)
2957 		strlcat(buf, "|VI_MOUNT", sizeof(buf));
2958 	if (vp->v_iflag & VI_AGE)
2959 		strlcat(buf, "|VI_AGE", sizeof(buf));
2960 	if (vp->v_iflag & VI_DOOMED)
2961 		strlcat(buf, "|VI_DOOMED", sizeof(buf));
2962 	if (vp->v_iflag & VI_FREE)
2963 		strlcat(buf, "|VI_FREE", sizeof(buf));
2964 	if (vp->v_iflag & VI_ACTIVE)
2965 		strlcat(buf, "|VI_ACTIVE", sizeof(buf));
2966 	if (vp->v_iflag & VI_DOINGINACT)
2967 		strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2968 	if (vp->v_iflag & VI_OWEINACT)
2969 		strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2970 	flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2971 	    VI_ACTIVE | VI_DOINGINACT | VI_OWEINACT);
2972 	if (flags != 0) {
2973 		snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2974 		strlcat(buf, buf2, sizeof(buf));
2975 	}
2976 	printf("    flags (%s)\n", buf + 1);
2977 	if (mtx_owned(VI_MTX(vp)))
2978 		printf(" VI_LOCKed");
2979 	if (vp->v_object != NULL)
2980 		printf("    v_object %p ref %d pages %d "
2981 		    "cleanbuf %d dirtybuf %d\n",
2982 		    vp->v_object, vp->v_object->ref_count,
2983 		    vp->v_object->resident_page_count,
2984 		    vp->v_bufobj.bo_dirty.bv_cnt,
2985 		    vp->v_bufobj.bo_clean.bv_cnt);
2986 	printf("    ");
2987 	lockmgr_printinfo(vp->v_vnlock);
2988 	if (vp->v_data != NULL)
2989 		VOP_PRINT(vp);
2990 }
2991 
2992 #ifdef DDB
2993 /*
2994  * List all of the locked vnodes in the system.
2995  * Called when debugging the kernel.
2996  */
2997 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2998 {
2999 	struct mount *mp;
3000 	struct vnode *vp;
3001 
3002 	/*
3003 	 * Note: because this is DDB, we can't obey the locking semantics
3004 	 * for these structures, which means we could catch an inconsistent
3005 	 * state and dereference a nasty pointer.  Not much to be done
3006 	 * about that.
3007 	 */
3008 	db_printf("Locked vnodes\n");
3009 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3010 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3011 			if (vp->v_type != VMARKER && VOP_ISLOCKED(vp))
3012 				vprint("", vp);
3013 		}
3014 	}
3015 }
3016 
3017 /*
3018  * Show details about the given vnode.
3019  */
3020 DB_SHOW_COMMAND(vnode, db_show_vnode)
3021 {
3022 	struct vnode *vp;
3023 
3024 	if (!have_addr)
3025 		return;
3026 	vp = (struct vnode *)addr;
3027 	vn_printf(vp, "vnode ");
3028 }
3029 
3030 /*
3031  * Show details about the given mount point.
3032  */
3033 DB_SHOW_COMMAND(mount, db_show_mount)
3034 {
3035 	struct mount *mp;
3036 	struct vfsopt *opt;
3037 	struct statfs *sp;
3038 	struct vnode *vp;
3039 	char buf[512];
3040 	uint64_t mflags;
3041 	u_int flags;
3042 
3043 	if (!have_addr) {
3044 		/* No address given, print short info about all mount points. */
3045 		TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3046 			db_printf("%p %s on %s (%s)\n", mp,
3047 			    mp->mnt_stat.f_mntfromname,
3048 			    mp->mnt_stat.f_mntonname,
3049 			    mp->mnt_stat.f_fstypename);
3050 			if (db_pager_quit)
3051 				break;
3052 		}
3053 		db_printf("\nMore info: show mount <addr>\n");
3054 		return;
3055 	}
3056 
3057 	mp = (struct mount *)addr;
3058 	db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
3059 	    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
3060 
3061 	buf[0] = '\0';
3062 	mflags = mp->mnt_flag;
3063 #define	MNT_FLAG(flag)	do {						\
3064 	if (mflags & (flag)) {						\
3065 		if (buf[0] != '\0')					\
3066 			strlcat(buf, ", ", sizeof(buf));		\
3067 		strlcat(buf, (#flag) + 4, sizeof(buf));			\
3068 		mflags &= ~(flag);					\
3069 	}								\
3070 } while (0)
3071 	MNT_FLAG(MNT_RDONLY);
3072 	MNT_FLAG(MNT_SYNCHRONOUS);
3073 	MNT_FLAG(MNT_NOEXEC);
3074 	MNT_FLAG(MNT_NOSUID);
3075 	MNT_FLAG(MNT_NFS4ACLS);
3076 	MNT_FLAG(MNT_UNION);
3077 	MNT_FLAG(MNT_ASYNC);
3078 	MNT_FLAG(MNT_SUIDDIR);
3079 	MNT_FLAG(MNT_SOFTDEP);
3080 	MNT_FLAG(MNT_NOSYMFOLLOW);
3081 	MNT_FLAG(MNT_GJOURNAL);
3082 	MNT_FLAG(MNT_MULTILABEL);
3083 	MNT_FLAG(MNT_ACLS);
3084 	MNT_FLAG(MNT_NOATIME);
3085 	MNT_FLAG(MNT_NOCLUSTERR);
3086 	MNT_FLAG(MNT_NOCLUSTERW);
3087 	MNT_FLAG(MNT_SUJ);
3088 	MNT_FLAG(MNT_EXRDONLY);
3089 	MNT_FLAG(MNT_EXPORTED);
3090 	MNT_FLAG(MNT_DEFEXPORTED);
3091 	MNT_FLAG(MNT_EXPORTANON);
3092 	MNT_FLAG(MNT_EXKERB);
3093 	MNT_FLAG(MNT_EXPUBLIC);
3094 	MNT_FLAG(MNT_LOCAL);
3095 	MNT_FLAG(MNT_QUOTA);
3096 	MNT_FLAG(MNT_ROOTFS);
3097 	MNT_FLAG(MNT_USER);
3098 	MNT_FLAG(MNT_IGNORE);
3099 	MNT_FLAG(MNT_UPDATE);
3100 	MNT_FLAG(MNT_DELEXPORT);
3101 	MNT_FLAG(MNT_RELOAD);
3102 	MNT_FLAG(MNT_FORCE);
3103 	MNT_FLAG(MNT_SNAPSHOT);
3104 	MNT_FLAG(MNT_BYFSID);
3105 #undef MNT_FLAG
3106 	if (mflags != 0) {
3107 		if (buf[0] != '\0')
3108 			strlcat(buf, ", ", sizeof(buf));
3109 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3110 		    "0x%016jx", mflags);
3111 	}
3112 	db_printf("    mnt_flag = %s\n", buf);
3113 
3114 	buf[0] = '\0';
3115 	flags = mp->mnt_kern_flag;
3116 #define	MNT_KERN_FLAG(flag)	do {					\
3117 	if (flags & (flag)) {						\
3118 		if (buf[0] != '\0')					\
3119 			strlcat(buf, ", ", sizeof(buf));		\
3120 		strlcat(buf, (#flag) + 5, sizeof(buf));			\
3121 		flags &= ~(flag);					\
3122 	}								\
3123 } while (0)
3124 	MNT_KERN_FLAG(MNTK_UNMOUNTF);
3125 	MNT_KERN_FLAG(MNTK_ASYNC);
3126 	MNT_KERN_FLAG(MNTK_SOFTDEP);
3127 	MNT_KERN_FLAG(MNTK_NOINSMNTQ);
3128 	MNT_KERN_FLAG(MNTK_DRAINING);
3129 	MNT_KERN_FLAG(MNTK_REFEXPIRE);
3130 	MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
3131 	MNT_KERN_FLAG(MNTK_SHARED_WRITES);
3132 	MNT_KERN_FLAG(MNTK_NO_IOPF);
3133 	MNT_KERN_FLAG(MNTK_VGONE_UPPER);
3134 	MNT_KERN_FLAG(MNTK_VGONE_WAITER);
3135 	MNT_KERN_FLAG(MNTK_LOOKUP_EXCL_DOTDOT);
3136 	MNT_KERN_FLAG(MNTK_MARKER);
3137 	MNT_KERN_FLAG(MNTK_NOASYNC);
3138 	MNT_KERN_FLAG(MNTK_UNMOUNT);
3139 	MNT_KERN_FLAG(MNTK_MWAIT);
3140 	MNT_KERN_FLAG(MNTK_SUSPEND);
3141 	MNT_KERN_FLAG(MNTK_SUSPEND2);
3142 	MNT_KERN_FLAG(MNTK_SUSPENDED);
3143 	MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
3144 	MNT_KERN_FLAG(MNTK_NOKNOTE);
3145 #undef MNT_KERN_FLAG
3146 	if (flags != 0) {
3147 		if (buf[0] != '\0')
3148 			strlcat(buf, ", ", sizeof(buf));
3149 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3150 		    "0x%08x", flags);
3151 	}
3152 	db_printf("    mnt_kern_flag = %s\n", buf);
3153 
3154 	db_printf("    mnt_opt = ");
3155 	opt = TAILQ_FIRST(mp->mnt_opt);
3156 	if (opt != NULL) {
3157 		db_printf("%s", opt->name);
3158 		opt = TAILQ_NEXT(opt, link);
3159 		while (opt != NULL) {
3160 			db_printf(", %s", opt->name);
3161 			opt = TAILQ_NEXT(opt, link);
3162 		}
3163 	}
3164 	db_printf("\n");
3165 
3166 	sp = &mp->mnt_stat;
3167 	db_printf("    mnt_stat = { version=%u type=%u flags=0x%016jx "
3168 	    "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
3169 	    "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
3170 	    "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
3171 	    (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
3172 	    (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
3173 	    (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
3174 	    (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
3175 	    (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
3176 	    (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
3177 	    (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
3178 	    (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
3179 
3180 	db_printf("    mnt_cred = { uid=%u ruid=%u",
3181 	    (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
3182 	if (jailed(mp->mnt_cred))
3183 		db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
3184 	db_printf(" }\n");
3185 	db_printf("    mnt_ref = %d\n", mp->mnt_ref);
3186 	db_printf("    mnt_gen = %d\n", mp->mnt_gen);
3187 	db_printf("    mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
3188 	db_printf("    mnt_activevnodelistsize = %d\n",
3189 	    mp->mnt_activevnodelistsize);
3190 	db_printf("    mnt_writeopcount = %d\n", mp->mnt_writeopcount);
3191 	db_printf("    mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
3192 	db_printf("    mnt_iosize_max = %d\n", mp->mnt_iosize_max);
3193 	db_printf("    mnt_hashseed = %u\n", mp->mnt_hashseed);
3194 	db_printf("    mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
3195 	db_printf("    mnt_secondary_accwrites = %d\n",
3196 	    mp->mnt_secondary_accwrites);
3197 	db_printf("    mnt_gjprovider = %s\n",
3198 	    mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
3199 
3200 	db_printf("\n\nList of active vnodes\n");
3201 	TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) {
3202 		if (vp->v_type != VMARKER) {
3203 			vn_printf(vp, "vnode ");
3204 			if (db_pager_quit)
3205 				break;
3206 		}
3207 	}
3208 	db_printf("\n\nList of inactive vnodes\n");
3209 	TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3210 		if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) {
3211 			vn_printf(vp, "vnode ");
3212 			if (db_pager_quit)
3213 				break;
3214 		}
3215 	}
3216 }
3217 #endif	/* DDB */
3218 
3219 /*
3220  * Fill in a struct xvfsconf based on a struct vfsconf.
3221  */
3222 static int
3223 vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp)
3224 {
3225 	struct xvfsconf xvfsp;
3226 
3227 	bzero(&xvfsp, sizeof(xvfsp));
3228 	strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3229 	xvfsp.vfc_typenum = vfsp->vfc_typenum;
3230 	xvfsp.vfc_refcount = vfsp->vfc_refcount;
3231 	xvfsp.vfc_flags = vfsp->vfc_flags;
3232 	/*
3233 	 * These are unused in userland, we keep them
3234 	 * to not break binary compatibility.
3235 	 */
3236 	xvfsp.vfc_vfsops = NULL;
3237 	xvfsp.vfc_next = NULL;
3238 	return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3239 }
3240 
3241 #ifdef COMPAT_FREEBSD32
3242 struct xvfsconf32 {
3243 	uint32_t	vfc_vfsops;
3244 	char		vfc_name[MFSNAMELEN];
3245 	int32_t		vfc_typenum;
3246 	int32_t		vfc_refcount;
3247 	int32_t		vfc_flags;
3248 	uint32_t	vfc_next;
3249 };
3250 
3251 static int
3252 vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp)
3253 {
3254 	struct xvfsconf32 xvfsp;
3255 
3256 	strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3257 	xvfsp.vfc_typenum = vfsp->vfc_typenum;
3258 	xvfsp.vfc_refcount = vfsp->vfc_refcount;
3259 	xvfsp.vfc_flags = vfsp->vfc_flags;
3260 	xvfsp.vfc_vfsops = 0;
3261 	xvfsp.vfc_next = 0;
3262 	return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3263 }
3264 #endif
3265 
3266 /*
3267  * Top level filesystem related information gathering.
3268  */
3269 static int
3270 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
3271 {
3272 	struct vfsconf *vfsp;
3273 	int error;
3274 
3275 	error = 0;
3276 	vfsconf_slock();
3277 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3278 #ifdef COMPAT_FREEBSD32
3279 		if (req->flags & SCTL_MASK32)
3280 			error = vfsconf2x32(req, vfsp);
3281 		else
3282 #endif
3283 			error = vfsconf2x(req, vfsp);
3284 		if (error)
3285 			break;
3286 	}
3287 	vfsconf_sunlock();
3288 	return (error);
3289 }
3290 
3291 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD |
3292     CTLFLAG_MPSAFE, NULL, 0, sysctl_vfs_conflist,
3293     "S,xvfsconf", "List of all configured filesystems");
3294 
3295 #ifndef BURN_BRIDGES
3296 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
3297 
3298 static int
3299 vfs_sysctl(SYSCTL_HANDLER_ARGS)
3300 {
3301 	int *name = (int *)arg1 - 1;	/* XXX */
3302 	u_int namelen = arg2 + 1;	/* XXX */
3303 	struct vfsconf *vfsp;
3304 
3305 	log(LOG_WARNING, "userland calling deprecated sysctl, "
3306 	    "please rebuild world\n");
3307 
3308 #if 1 || defined(COMPAT_PRELITE2)
3309 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3310 	if (namelen == 1)
3311 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3312 #endif
3313 
3314 	switch (name[1]) {
3315 	case VFS_MAXTYPENUM:
3316 		if (namelen != 2)
3317 			return (ENOTDIR);
3318 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3319 	case VFS_CONF:
3320 		if (namelen != 3)
3321 			return (ENOTDIR);	/* overloaded */
3322 		vfsconf_slock();
3323 		TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3324 			if (vfsp->vfc_typenum == name[2])
3325 				break;
3326 		}
3327 		vfsconf_sunlock();
3328 		if (vfsp == NULL)
3329 			return (EOPNOTSUPP);
3330 #ifdef COMPAT_FREEBSD32
3331 		if (req->flags & SCTL_MASK32)
3332 			return (vfsconf2x32(req, vfsp));
3333 		else
3334 #endif
3335 			return (vfsconf2x(req, vfsp));
3336 	}
3337 	return (EOPNOTSUPP);
3338 }
3339 
3340 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP |
3341     CTLFLAG_MPSAFE, vfs_sysctl,
3342     "Generic filesystem");
3343 
3344 #if 1 || defined(COMPAT_PRELITE2)
3345 
3346 static int
3347 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3348 {
3349 	int error;
3350 	struct vfsconf *vfsp;
3351 	struct ovfsconf ovfs;
3352 
3353 	vfsconf_slock();
3354 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3355 		bzero(&ovfs, sizeof(ovfs));
3356 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
3357 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
3358 		ovfs.vfc_index = vfsp->vfc_typenum;
3359 		ovfs.vfc_refcount = vfsp->vfc_refcount;
3360 		ovfs.vfc_flags = vfsp->vfc_flags;
3361 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3362 		if (error != 0) {
3363 			vfsconf_sunlock();
3364 			return (error);
3365 		}
3366 	}
3367 	vfsconf_sunlock();
3368 	return (0);
3369 }
3370 
3371 #endif /* 1 || COMPAT_PRELITE2 */
3372 #endif /* !BURN_BRIDGES */
3373 
3374 #define KINFO_VNODESLOP		10
3375 #ifdef notyet
3376 /*
3377  * Dump vnode list (via sysctl).
3378  */
3379 /* ARGSUSED */
3380 static int
3381 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3382 {
3383 	struct xvnode *xvn;
3384 	struct mount *mp;
3385 	struct vnode *vp;
3386 	int error, len, n;
3387 
3388 	/*
3389 	 * Stale numvnodes access is not fatal here.
3390 	 */
3391 	req->lock = 0;
3392 	len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3393 	if (!req->oldptr)
3394 		/* Make an estimate */
3395 		return (SYSCTL_OUT(req, 0, len));
3396 
3397 	error = sysctl_wire_old_buffer(req, 0);
3398 	if (error != 0)
3399 		return (error);
3400 	xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3401 	n = 0;
3402 	mtx_lock(&mountlist_mtx);
3403 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3404 		if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3405 			continue;
3406 		MNT_ILOCK(mp);
3407 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3408 			if (n == len)
3409 				break;
3410 			vref(vp);
3411 			xvn[n].xv_size = sizeof *xvn;
3412 			xvn[n].xv_vnode = vp;
3413 			xvn[n].xv_id = 0;	/* XXX compat */
3414 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3415 			XV_COPY(usecount);
3416 			XV_COPY(writecount);
3417 			XV_COPY(holdcnt);
3418 			XV_COPY(mount);
3419 			XV_COPY(numoutput);
3420 			XV_COPY(type);
3421 #undef XV_COPY
3422 			xvn[n].xv_flag = vp->v_vflag;
3423 
3424 			switch (vp->v_type) {
3425 			case VREG:
3426 			case VDIR:
3427 			case VLNK:
3428 				break;
3429 			case VBLK:
3430 			case VCHR:
3431 				if (vp->v_rdev == NULL) {
3432 					vrele(vp);
3433 					continue;
3434 				}
3435 				xvn[n].xv_dev = dev2udev(vp->v_rdev);
3436 				break;
3437 			case VSOCK:
3438 				xvn[n].xv_socket = vp->v_socket;
3439 				break;
3440 			case VFIFO:
3441 				xvn[n].xv_fifo = vp->v_fifoinfo;
3442 				break;
3443 			case VNON:
3444 			case VBAD:
3445 			default:
3446 				/* shouldn't happen? */
3447 				vrele(vp);
3448 				continue;
3449 			}
3450 			vrele(vp);
3451 			++n;
3452 		}
3453 		MNT_IUNLOCK(mp);
3454 		mtx_lock(&mountlist_mtx);
3455 		vfs_unbusy(mp);
3456 		if (n == len)
3457 			break;
3458 	}
3459 	mtx_unlock(&mountlist_mtx);
3460 
3461 	error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3462 	free(xvn, M_TEMP);
3463 	return (error);
3464 }
3465 
3466 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE | CTLFLAG_RD |
3467     CTLFLAG_MPSAFE, 0, 0, sysctl_vnode, "S,xvnode",
3468     "");
3469 #endif
3470 
3471 /*
3472  * Unmount all filesystems. The list is traversed in reverse order
3473  * of mounting to avoid dependencies.
3474  */
3475 void
3476 vfs_unmountall(void)
3477 {
3478 	struct mount *mp;
3479 	struct thread *td;
3480 	int error;
3481 
3482 	CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3483 	td = curthread;
3484 
3485 	/*
3486 	 * Since this only runs when rebooting, it is not interlocked.
3487 	 */
3488 	while(!TAILQ_EMPTY(&mountlist)) {
3489 		mp = TAILQ_LAST(&mountlist, mntlist);
3490 		error = dounmount(mp, MNT_FORCE, td);
3491 		if (error) {
3492 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
3493 			/*
3494 			 * XXX: Due to the way in which we mount the root
3495 			 * file system off of devfs, devfs will generate a
3496 			 * "busy" warning when we try to unmount it before
3497 			 * the root.  Don't print a warning as a result in
3498 			 * order to avoid false positive errors that may
3499 			 * cause needless upset.
3500 			 */
3501 			if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3502 				printf("unmount of %s failed (",
3503 				    mp->mnt_stat.f_mntonname);
3504 				if (error == EBUSY)
3505 					printf("BUSY)\n");
3506 				else
3507 					printf("%d)\n", error);
3508 			}
3509 		} else {
3510 			/* The unmount has removed mp from the mountlist */
3511 		}
3512 	}
3513 }
3514 
3515 /*
3516  * perform msync on all vnodes under a mount point
3517  * the mount point must be locked.
3518  */
3519 void
3520 vfs_msync(struct mount *mp, int flags)
3521 {
3522 	struct vnode *vp, *mvp;
3523 	struct vm_object *obj;
3524 
3525 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3526 	MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
3527 		obj = vp->v_object;
3528 		if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3529 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3530 			if (!vget(vp,
3531 			    LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3532 			    curthread)) {
3533 				if (vp->v_vflag & VV_NOSYNC) {	/* unlinked */
3534 					vput(vp);
3535 					continue;
3536 				}
3537 
3538 				obj = vp->v_object;
3539 				if (obj != NULL) {
3540 					VM_OBJECT_WLOCK(obj);
3541 					vm_object_page_clean(obj, 0, 0,
3542 					    flags == MNT_WAIT ?
3543 					    OBJPC_SYNC : OBJPC_NOSYNC);
3544 					VM_OBJECT_WUNLOCK(obj);
3545 				}
3546 				vput(vp);
3547 			}
3548 		} else
3549 			VI_UNLOCK(vp);
3550 	}
3551 }
3552 
3553 static void
3554 destroy_vpollinfo_free(struct vpollinfo *vi)
3555 {
3556 
3557 	knlist_destroy(&vi->vpi_selinfo.si_note);
3558 	mtx_destroy(&vi->vpi_lock);
3559 	uma_zfree(vnodepoll_zone, vi);
3560 }
3561 
3562 static void
3563 destroy_vpollinfo(struct vpollinfo *vi)
3564 {
3565 
3566 	knlist_clear(&vi->vpi_selinfo.si_note, 1);
3567 	seldrain(&vi->vpi_selinfo);
3568 	destroy_vpollinfo_free(vi);
3569 }
3570 
3571 /*
3572  * Initalize per-vnode helper structure to hold poll-related state.
3573  */
3574 void
3575 v_addpollinfo(struct vnode *vp)
3576 {
3577 	struct vpollinfo *vi;
3578 
3579 	if (vp->v_pollinfo != NULL)
3580 		return;
3581 	vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3582 	mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3583 	knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3584 	    vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3585 	VI_LOCK(vp);
3586 	if (vp->v_pollinfo != NULL) {
3587 		VI_UNLOCK(vp);
3588 		destroy_vpollinfo_free(vi);
3589 		return;
3590 	}
3591 	vp->v_pollinfo = vi;
3592 	VI_UNLOCK(vp);
3593 }
3594 
3595 /*
3596  * Record a process's interest in events which might happen to
3597  * a vnode.  Because poll uses the historic select-style interface
3598  * internally, this routine serves as both the ``check for any
3599  * pending events'' and the ``record my interest in future events''
3600  * functions.  (These are done together, while the lock is held,
3601  * to avoid race conditions.)
3602  */
3603 int
3604 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3605 {
3606 
3607 	v_addpollinfo(vp);
3608 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3609 	if (vp->v_pollinfo->vpi_revents & events) {
3610 		/*
3611 		 * This leaves events we are not interested
3612 		 * in available for the other process which
3613 		 * which presumably had requested them
3614 		 * (otherwise they would never have been
3615 		 * recorded).
3616 		 */
3617 		events &= vp->v_pollinfo->vpi_revents;
3618 		vp->v_pollinfo->vpi_revents &= ~events;
3619 
3620 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
3621 		return (events);
3622 	}
3623 	vp->v_pollinfo->vpi_events |= events;
3624 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3625 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3626 	return (0);
3627 }
3628 
3629 /*
3630  * Routine to create and manage a filesystem syncer vnode.
3631  */
3632 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
3633 static int	sync_fsync(struct  vop_fsync_args *);
3634 static int	sync_inactive(struct  vop_inactive_args *);
3635 static int	sync_reclaim(struct  vop_reclaim_args *);
3636 
3637 static struct vop_vector sync_vnodeops = {
3638 	.vop_bypass =	VOP_EOPNOTSUPP,
3639 	.vop_close =	sync_close,		/* close */
3640 	.vop_fsync =	sync_fsync,		/* fsync */
3641 	.vop_inactive =	sync_inactive,	/* inactive */
3642 	.vop_reclaim =	sync_reclaim,	/* reclaim */
3643 	.vop_lock1 =	vop_stdlock,	/* lock */
3644 	.vop_unlock =	vop_stdunlock,	/* unlock */
3645 	.vop_islocked =	vop_stdislocked,	/* islocked */
3646 };
3647 
3648 /*
3649  * Create a new filesystem syncer vnode for the specified mount point.
3650  */
3651 void
3652 vfs_allocate_syncvnode(struct mount *mp)
3653 {
3654 	struct vnode *vp;
3655 	struct bufobj *bo;
3656 	static long start, incr, next;
3657 	int error;
3658 
3659 	/* Allocate a new vnode */
3660 	error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
3661 	if (error != 0)
3662 		panic("vfs_allocate_syncvnode: getnewvnode() failed");
3663 	vp->v_type = VNON;
3664 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3665 	vp->v_vflag |= VV_FORCEINSMQ;
3666 	error = insmntque(vp, mp);
3667 	if (error != 0)
3668 		panic("vfs_allocate_syncvnode: insmntque() failed");
3669 	vp->v_vflag &= ~VV_FORCEINSMQ;
3670 	VOP_UNLOCK(vp, 0);
3671 	/*
3672 	 * Place the vnode onto the syncer worklist. We attempt to
3673 	 * scatter them about on the list so that they will go off
3674 	 * at evenly distributed times even if all the filesystems
3675 	 * are mounted at once.
3676 	 */
3677 	next += incr;
3678 	if (next == 0 || next > syncer_maxdelay) {
3679 		start /= 2;
3680 		incr /= 2;
3681 		if (start == 0) {
3682 			start = syncer_maxdelay / 2;
3683 			incr = syncer_maxdelay;
3684 		}
3685 		next = start;
3686 	}
3687 	bo = &vp->v_bufobj;
3688 	BO_LOCK(bo);
3689 	vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3690 	/* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3691 	mtx_lock(&sync_mtx);
3692 	sync_vnode_count++;
3693 	if (mp->mnt_syncer == NULL) {
3694 		mp->mnt_syncer = vp;
3695 		vp = NULL;
3696 	}
3697 	mtx_unlock(&sync_mtx);
3698 	BO_UNLOCK(bo);
3699 	if (vp != NULL) {
3700 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3701 		vgone(vp);
3702 		vput(vp);
3703 	}
3704 }
3705 
3706 void
3707 vfs_deallocate_syncvnode(struct mount *mp)
3708 {
3709 	struct vnode *vp;
3710 
3711 	mtx_lock(&sync_mtx);
3712 	vp = mp->mnt_syncer;
3713 	if (vp != NULL)
3714 		mp->mnt_syncer = NULL;
3715 	mtx_unlock(&sync_mtx);
3716 	if (vp != NULL)
3717 		vrele(vp);
3718 }
3719 
3720 /*
3721  * Do a lazy sync of the filesystem.
3722  */
3723 static int
3724 sync_fsync(struct vop_fsync_args *ap)
3725 {
3726 	struct vnode *syncvp = ap->a_vp;
3727 	struct mount *mp = syncvp->v_mount;
3728 	int error, save;
3729 	struct bufobj *bo;
3730 
3731 	/*
3732 	 * We only need to do something if this is a lazy evaluation.
3733 	 */
3734 	if (ap->a_waitfor != MNT_LAZY)
3735 		return (0);
3736 
3737 	/*
3738 	 * Move ourselves to the back of the sync list.
3739 	 */
3740 	bo = &syncvp->v_bufobj;
3741 	BO_LOCK(bo);
3742 	vn_syncer_add_to_worklist(bo, syncdelay);
3743 	BO_UNLOCK(bo);
3744 
3745 	/*
3746 	 * Walk the list of vnodes pushing all that are dirty and
3747 	 * not already on the sync list.
3748 	 */
3749 	if (vfs_busy(mp, MBF_NOWAIT) != 0)
3750 		return (0);
3751 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3752 		vfs_unbusy(mp);
3753 		return (0);
3754 	}
3755 	save = curthread_pflags_set(TDP_SYNCIO);
3756 	vfs_msync(mp, MNT_NOWAIT);
3757 	error = VFS_SYNC(mp, MNT_LAZY);
3758 	curthread_pflags_restore(save);
3759 	vn_finished_write(mp);
3760 	vfs_unbusy(mp);
3761 	return (error);
3762 }
3763 
3764 /*
3765  * The syncer vnode is no referenced.
3766  */
3767 static int
3768 sync_inactive(struct vop_inactive_args *ap)
3769 {
3770 
3771 	vgone(ap->a_vp);
3772 	return (0);
3773 }
3774 
3775 /*
3776  * The syncer vnode is no longer needed and is being decommissioned.
3777  *
3778  * Modifications to the worklist must be protected by sync_mtx.
3779  */
3780 static int
3781 sync_reclaim(struct vop_reclaim_args *ap)
3782 {
3783 	struct vnode *vp = ap->a_vp;
3784 	struct bufobj *bo;
3785 
3786 	bo = &vp->v_bufobj;
3787 	BO_LOCK(bo);
3788 	mtx_lock(&sync_mtx);
3789 	if (vp->v_mount->mnt_syncer == vp)
3790 		vp->v_mount->mnt_syncer = NULL;
3791 	if (bo->bo_flag & BO_ONWORKLST) {
3792 		LIST_REMOVE(bo, bo_synclist);
3793 		syncer_worklist_len--;
3794 		sync_vnode_count--;
3795 		bo->bo_flag &= ~BO_ONWORKLST;
3796 	}
3797 	mtx_unlock(&sync_mtx);
3798 	BO_UNLOCK(bo);
3799 
3800 	return (0);
3801 }
3802 
3803 /*
3804  * Check if vnode represents a disk device
3805  */
3806 int
3807 vn_isdisk(struct vnode *vp, int *errp)
3808 {
3809 	int error;
3810 
3811 	if (vp->v_type != VCHR) {
3812 		error = ENOTBLK;
3813 		goto out;
3814 	}
3815 	error = 0;
3816 	dev_lock();
3817 	if (vp->v_rdev == NULL)
3818 		error = ENXIO;
3819 	else if (vp->v_rdev->si_devsw == NULL)
3820 		error = ENXIO;
3821 	else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3822 		error = ENOTBLK;
3823 	dev_unlock();
3824 out:
3825 	if (errp != NULL)
3826 		*errp = error;
3827 	return (error == 0);
3828 }
3829 
3830 /*
3831  * Common filesystem object access control check routine.  Accepts a
3832  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3833  * and optional call-by-reference privused argument allowing vaccess()
3834  * to indicate to the caller whether privilege was used to satisfy the
3835  * request (obsoleted).  Returns 0 on success, or an errno on failure.
3836  */
3837 int
3838 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3839     accmode_t accmode, struct ucred *cred, int *privused)
3840 {
3841 	accmode_t dac_granted;
3842 	accmode_t priv_granted;
3843 
3844 	KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3845 	    ("invalid bit in accmode"));
3846 	KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
3847 	    ("VAPPEND without VWRITE"));
3848 
3849 	/*
3850 	 * Look for a normal, non-privileged way to access the file/directory
3851 	 * as requested.  If it exists, go with that.
3852 	 */
3853 
3854 	if (privused != NULL)
3855 		*privused = 0;
3856 
3857 	dac_granted = 0;
3858 
3859 	/* Check the owner. */
3860 	if (cred->cr_uid == file_uid) {
3861 		dac_granted |= VADMIN;
3862 		if (file_mode & S_IXUSR)
3863 			dac_granted |= VEXEC;
3864 		if (file_mode & S_IRUSR)
3865 			dac_granted |= VREAD;
3866 		if (file_mode & S_IWUSR)
3867 			dac_granted |= (VWRITE | VAPPEND);
3868 
3869 		if ((accmode & dac_granted) == accmode)
3870 			return (0);
3871 
3872 		goto privcheck;
3873 	}
3874 
3875 	/* Otherwise, check the groups (first match) */
3876 	if (groupmember(file_gid, cred)) {
3877 		if (file_mode & S_IXGRP)
3878 			dac_granted |= VEXEC;
3879 		if (file_mode & S_IRGRP)
3880 			dac_granted |= VREAD;
3881 		if (file_mode & S_IWGRP)
3882 			dac_granted |= (VWRITE | VAPPEND);
3883 
3884 		if ((accmode & dac_granted) == accmode)
3885 			return (0);
3886 
3887 		goto privcheck;
3888 	}
3889 
3890 	/* Otherwise, check everyone else. */
3891 	if (file_mode & S_IXOTH)
3892 		dac_granted |= VEXEC;
3893 	if (file_mode & S_IROTH)
3894 		dac_granted |= VREAD;
3895 	if (file_mode & S_IWOTH)
3896 		dac_granted |= (VWRITE | VAPPEND);
3897 	if ((accmode & dac_granted) == accmode)
3898 		return (0);
3899 
3900 privcheck:
3901 	/*
3902 	 * Build a privilege mask to determine if the set of privileges
3903 	 * satisfies the requirements when combined with the granted mask
3904 	 * from above.  For each privilege, if the privilege is required,
3905 	 * bitwise or the request type onto the priv_granted mask.
3906 	 */
3907 	priv_granted = 0;
3908 
3909 	if (type == VDIR) {
3910 		/*
3911 		 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3912 		 * requests, instead of PRIV_VFS_EXEC.
3913 		 */
3914 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3915 		    !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3916 			priv_granted |= VEXEC;
3917 	} else {
3918 		/*
3919 		 * Ensure that at least one execute bit is on. Otherwise,
3920 		 * a privileged user will always succeed, and we don't want
3921 		 * this to happen unless the file really is executable.
3922 		 */
3923 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3924 		    (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
3925 		    !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3926 			priv_granted |= VEXEC;
3927 	}
3928 
3929 	if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
3930 	    !priv_check_cred(cred, PRIV_VFS_READ, 0))
3931 		priv_granted |= VREAD;
3932 
3933 	if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3934 	    !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3935 		priv_granted |= (VWRITE | VAPPEND);
3936 
3937 	if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3938 	    !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3939 		priv_granted |= VADMIN;
3940 
3941 	if ((accmode & (priv_granted | dac_granted)) == accmode) {
3942 		/* XXX audit: privilege used */
3943 		if (privused != NULL)
3944 			*privused = 1;
3945 		return (0);
3946 	}
3947 
3948 	return ((accmode & VADMIN) ? EPERM : EACCES);
3949 }
3950 
3951 /*
3952  * Credential check based on process requesting service, and per-attribute
3953  * permissions.
3954  */
3955 int
3956 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3957     struct thread *td, accmode_t accmode)
3958 {
3959 
3960 	/*
3961 	 * Kernel-invoked always succeeds.
3962 	 */
3963 	if (cred == NOCRED)
3964 		return (0);
3965 
3966 	/*
3967 	 * Do not allow privileged processes in jail to directly manipulate
3968 	 * system attributes.
3969 	 */
3970 	switch (attrnamespace) {
3971 	case EXTATTR_NAMESPACE_SYSTEM:
3972 		/* Potentially should be: return (EPERM); */
3973 		return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3974 	case EXTATTR_NAMESPACE_USER:
3975 		return (VOP_ACCESS(vp, accmode, cred, td));
3976 	default:
3977 		return (EPERM);
3978 	}
3979 }
3980 
3981 #ifdef DEBUG_VFS_LOCKS
3982 /*
3983  * This only exists to supress warnings from unlocked specfs accesses.  It is
3984  * no longer ok to have an unlocked VFS.
3985  */
3986 #define	IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL ||		\
3987 	(vp)->v_type == VCHR ||	(vp)->v_type == VBAD)
3988 
3989 int vfs_badlock_ddb = 1;	/* Drop into debugger on violation. */
3990 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
3991     "Drop into debugger on lock violation");
3992 
3993 int vfs_badlock_mutex = 1;	/* Check for interlock across VOPs. */
3994 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
3995     0, "Check for interlock across VOPs");
3996 
3997 int vfs_badlock_print = 1;	/* Print lock violations. */
3998 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
3999     0, "Print lock violations");
4000 
4001 #ifdef KDB
4002 int vfs_badlock_backtrace = 1;	/* Print backtrace at lock violations. */
4003 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
4004     &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
4005 #endif
4006 
4007 static void
4008 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
4009 {
4010 
4011 #ifdef KDB
4012 	if (vfs_badlock_backtrace)
4013 		kdb_backtrace();
4014 #endif
4015 	if (vfs_badlock_print)
4016 		printf("%s: %p %s\n", str, (void *)vp, msg);
4017 	if (vfs_badlock_ddb)
4018 		kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
4019 }
4020 
4021 void
4022 assert_vi_locked(struct vnode *vp, const char *str)
4023 {
4024 
4025 	if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
4026 		vfs_badlock("interlock is not locked but should be", str, vp);
4027 }
4028 
4029 void
4030 assert_vi_unlocked(struct vnode *vp, const char *str)
4031 {
4032 
4033 	if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
4034 		vfs_badlock("interlock is locked but should not be", str, vp);
4035 }
4036 
4037 void
4038 assert_vop_locked(struct vnode *vp, const char *str)
4039 {
4040 	int locked;
4041 
4042 	if (!IGNORE_LOCK(vp)) {
4043 		locked = VOP_ISLOCKED(vp);
4044 		if (locked == 0 || locked == LK_EXCLOTHER)
4045 			vfs_badlock("is not locked but should be", str, vp);
4046 	}
4047 }
4048 
4049 void
4050 assert_vop_unlocked(struct vnode *vp, const char *str)
4051 {
4052 
4053 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
4054 		vfs_badlock("is locked but should not be", str, vp);
4055 }
4056 
4057 void
4058 assert_vop_elocked(struct vnode *vp, const char *str)
4059 {
4060 
4061 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
4062 		vfs_badlock("is not exclusive locked but should be", str, vp);
4063 }
4064 
4065 #if 0
4066 void
4067 assert_vop_elocked_other(struct vnode *vp, const char *str)
4068 {
4069 
4070 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
4071 		vfs_badlock("is not exclusive locked by another thread",
4072 		    str, vp);
4073 }
4074 
4075 void
4076 assert_vop_slocked(struct vnode *vp, const char *str)
4077 {
4078 
4079 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
4080 		vfs_badlock("is not locked shared but should be", str, vp);
4081 }
4082 #endif /* 0 */
4083 #endif /* DEBUG_VFS_LOCKS */
4084 
4085 void
4086 vop_rename_fail(struct vop_rename_args *ap)
4087 {
4088 
4089 	if (ap->a_tvp != NULL)
4090 		vput(ap->a_tvp);
4091 	if (ap->a_tdvp == ap->a_tvp)
4092 		vrele(ap->a_tdvp);
4093 	else
4094 		vput(ap->a_tdvp);
4095 	vrele(ap->a_fdvp);
4096 	vrele(ap->a_fvp);
4097 }
4098 
4099 void
4100 vop_rename_pre(void *ap)
4101 {
4102 	struct vop_rename_args *a = ap;
4103 
4104 #ifdef DEBUG_VFS_LOCKS
4105 	if (a->a_tvp)
4106 		ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
4107 	ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
4108 	ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
4109 	ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
4110 
4111 	/* Check the source (from). */
4112 	if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
4113 	    (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
4114 		ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
4115 	if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
4116 		ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
4117 
4118 	/* Check the target. */
4119 	if (a->a_tvp)
4120 		ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
4121 	ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
4122 #endif
4123 	if (a->a_tdvp != a->a_fdvp)
4124 		vhold(a->a_fdvp);
4125 	if (a->a_tvp != a->a_fvp)
4126 		vhold(a->a_fvp);
4127 	vhold(a->a_tdvp);
4128 	if (a->a_tvp)
4129 		vhold(a->a_tvp);
4130 }
4131 
4132 void
4133 vop_strategy_pre(void *ap)
4134 {
4135 #ifdef DEBUG_VFS_LOCKS
4136 	struct vop_strategy_args *a;
4137 	struct buf *bp;
4138 
4139 	a = ap;
4140 	bp = a->a_bp;
4141 
4142 	/*
4143 	 * Cluster ops lock their component buffers but not the IO container.
4144 	 */
4145 	if ((bp->b_flags & B_CLUSTER) != 0)
4146 		return;
4147 
4148 	if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
4149 		if (vfs_badlock_print)
4150 			printf(
4151 			    "VOP_STRATEGY: bp is not locked but should be\n");
4152 		if (vfs_badlock_ddb)
4153 			kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
4154 	}
4155 #endif
4156 }
4157 
4158 void
4159 vop_lock_pre(void *ap)
4160 {
4161 #ifdef DEBUG_VFS_LOCKS
4162 	struct vop_lock1_args *a = ap;
4163 
4164 	if ((a->a_flags & LK_INTERLOCK) == 0)
4165 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4166 	else
4167 		ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
4168 #endif
4169 }
4170 
4171 void
4172 vop_lock_post(void *ap, int rc)
4173 {
4174 #ifdef DEBUG_VFS_LOCKS
4175 	struct vop_lock1_args *a = ap;
4176 
4177 	ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4178 	if (rc == 0 && (a->a_flags & LK_EXCLOTHER) == 0)
4179 		ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
4180 #endif
4181 }
4182 
4183 void
4184 vop_unlock_pre(void *ap)
4185 {
4186 #ifdef DEBUG_VFS_LOCKS
4187 	struct vop_unlock_args *a = ap;
4188 
4189 	if (a->a_flags & LK_INTERLOCK)
4190 		ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
4191 	ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
4192 #endif
4193 }
4194 
4195 void
4196 vop_unlock_post(void *ap, int rc)
4197 {
4198 #ifdef DEBUG_VFS_LOCKS
4199 	struct vop_unlock_args *a = ap;
4200 
4201 	if (a->a_flags & LK_INTERLOCK)
4202 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
4203 #endif
4204 }
4205 
4206 void
4207 vop_create_post(void *ap, int rc)
4208 {
4209 	struct vop_create_args *a = ap;
4210 
4211 	if (!rc)
4212 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4213 }
4214 
4215 void
4216 vop_deleteextattr_post(void *ap, int rc)
4217 {
4218 	struct vop_deleteextattr_args *a = ap;
4219 
4220 	if (!rc)
4221 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4222 }
4223 
4224 void
4225 vop_link_post(void *ap, int rc)
4226 {
4227 	struct vop_link_args *a = ap;
4228 
4229 	if (!rc) {
4230 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
4231 		VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
4232 	}
4233 }
4234 
4235 void
4236 vop_mkdir_post(void *ap, int rc)
4237 {
4238 	struct vop_mkdir_args *a = ap;
4239 
4240 	if (!rc)
4241 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4242 }
4243 
4244 void
4245 vop_mknod_post(void *ap, int rc)
4246 {
4247 	struct vop_mknod_args *a = ap;
4248 
4249 	if (!rc)
4250 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4251 }
4252 
4253 void
4254 vop_remove_post(void *ap, int rc)
4255 {
4256 	struct vop_remove_args *a = ap;
4257 
4258 	if (!rc) {
4259 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4260 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4261 	}
4262 }
4263 
4264 void
4265 vop_rename_post(void *ap, int rc)
4266 {
4267 	struct vop_rename_args *a = ap;
4268 
4269 	if (!rc) {
4270 		VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
4271 		VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
4272 		VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
4273 		if (a->a_tvp)
4274 			VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
4275 	}
4276 	if (a->a_tdvp != a->a_fdvp)
4277 		vdrop(a->a_fdvp);
4278 	if (a->a_tvp != a->a_fvp)
4279 		vdrop(a->a_fvp);
4280 	vdrop(a->a_tdvp);
4281 	if (a->a_tvp)
4282 		vdrop(a->a_tvp);
4283 }
4284 
4285 void
4286 vop_rmdir_post(void *ap, int rc)
4287 {
4288 	struct vop_rmdir_args *a = ap;
4289 
4290 	if (!rc) {
4291 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4292 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4293 	}
4294 }
4295 
4296 void
4297 vop_setattr_post(void *ap, int rc)
4298 {
4299 	struct vop_setattr_args *a = ap;
4300 
4301 	if (!rc)
4302 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4303 }
4304 
4305 void
4306 vop_setextattr_post(void *ap, int rc)
4307 {
4308 	struct vop_setextattr_args *a = ap;
4309 
4310 	if (!rc)
4311 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4312 }
4313 
4314 void
4315 vop_symlink_post(void *ap, int rc)
4316 {
4317 	struct vop_symlink_args *a = ap;
4318 
4319 	if (!rc)
4320 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4321 }
4322 
4323 static struct knlist fs_knlist;
4324 
4325 static void
4326 vfs_event_init(void *arg)
4327 {
4328 	knlist_init_mtx(&fs_knlist, NULL);
4329 }
4330 /* XXX - correct order? */
4331 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4332 
4333 void
4334 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
4335 {
4336 
4337 	KNOTE_UNLOCKED(&fs_knlist, event);
4338 }
4339 
4340 static int	filt_fsattach(struct knote *kn);
4341 static void	filt_fsdetach(struct knote *kn);
4342 static int	filt_fsevent(struct knote *kn, long hint);
4343 
4344 struct filterops fs_filtops = {
4345 	.f_isfd = 0,
4346 	.f_attach = filt_fsattach,
4347 	.f_detach = filt_fsdetach,
4348 	.f_event = filt_fsevent
4349 };
4350 
4351 static int
4352 filt_fsattach(struct knote *kn)
4353 {
4354 
4355 	kn->kn_flags |= EV_CLEAR;
4356 	knlist_add(&fs_knlist, kn, 0);
4357 	return (0);
4358 }
4359 
4360 static void
4361 filt_fsdetach(struct knote *kn)
4362 {
4363 
4364 	knlist_remove(&fs_knlist, kn, 0);
4365 }
4366 
4367 static int
4368 filt_fsevent(struct knote *kn, long hint)
4369 {
4370 
4371 	kn->kn_fflags |= hint;
4372 	return (kn->kn_fflags != 0);
4373 }
4374 
4375 static int
4376 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4377 {
4378 	struct vfsidctl vc;
4379 	int error;
4380 	struct mount *mp;
4381 
4382 	error = SYSCTL_IN(req, &vc, sizeof(vc));
4383 	if (error)
4384 		return (error);
4385 	if (vc.vc_vers != VFS_CTL_VERS1)
4386 		return (EINVAL);
4387 	mp = vfs_getvfs(&vc.vc_fsid);
4388 	if (mp == NULL)
4389 		return (ENOENT);
4390 	/* ensure that a specific sysctl goes to the right filesystem. */
4391 	if (strcmp(vc.vc_fstypename, "*") != 0 &&
4392 	    strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4393 		vfs_rel(mp);
4394 		return (EINVAL);
4395 	}
4396 	VCTLTOREQ(&vc, req);
4397 	error = VFS_SYSCTL(mp, vc.vc_op, req);
4398 	vfs_rel(mp);
4399 	return (error);
4400 }
4401 
4402 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
4403     NULL, 0, sysctl_vfs_ctl, "",
4404     "Sysctl by fsid");
4405 
4406 /*
4407  * Function to initialize a va_filerev field sensibly.
4408  * XXX: Wouldn't a random number make a lot more sense ??
4409  */
4410 u_quad_t
4411 init_va_filerev(void)
4412 {
4413 	struct bintime bt;
4414 
4415 	getbinuptime(&bt);
4416 	return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4417 }
4418 
4419 static int	filt_vfsread(struct knote *kn, long hint);
4420 static int	filt_vfswrite(struct knote *kn, long hint);
4421 static int	filt_vfsvnode(struct knote *kn, long hint);
4422 static void	filt_vfsdetach(struct knote *kn);
4423 static struct filterops vfsread_filtops = {
4424 	.f_isfd = 1,
4425 	.f_detach = filt_vfsdetach,
4426 	.f_event = filt_vfsread
4427 };
4428 static struct filterops vfswrite_filtops = {
4429 	.f_isfd = 1,
4430 	.f_detach = filt_vfsdetach,
4431 	.f_event = filt_vfswrite
4432 };
4433 static struct filterops vfsvnode_filtops = {
4434 	.f_isfd = 1,
4435 	.f_detach = filt_vfsdetach,
4436 	.f_event = filt_vfsvnode
4437 };
4438 
4439 static void
4440 vfs_knllock(void *arg)
4441 {
4442 	struct vnode *vp = arg;
4443 
4444 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4445 }
4446 
4447 static void
4448 vfs_knlunlock(void *arg)
4449 {
4450 	struct vnode *vp = arg;
4451 
4452 	VOP_UNLOCK(vp, 0);
4453 }
4454 
4455 static void
4456 vfs_knl_assert_locked(void *arg)
4457 {
4458 #ifdef DEBUG_VFS_LOCKS
4459 	struct vnode *vp = arg;
4460 
4461 	ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4462 #endif
4463 }
4464 
4465 static void
4466 vfs_knl_assert_unlocked(void *arg)
4467 {
4468 #ifdef DEBUG_VFS_LOCKS
4469 	struct vnode *vp = arg;
4470 
4471 	ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4472 #endif
4473 }
4474 
4475 int
4476 vfs_kqfilter(struct vop_kqfilter_args *ap)
4477 {
4478 	struct vnode *vp = ap->a_vp;
4479 	struct knote *kn = ap->a_kn;
4480 	struct knlist *knl;
4481 
4482 	switch (kn->kn_filter) {
4483 	case EVFILT_READ:
4484 		kn->kn_fop = &vfsread_filtops;
4485 		break;
4486 	case EVFILT_WRITE:
4487 		kn->kn_fop = &vfswrite_filtops;
4488 		break;
4489 	case EVFILT_VNODE:
4490 		kn->kn_fop = &vfsvnode_filtops;
4491 		break;
4492 	default:
4493 		return (EINVAL);
4494 	}
4495 
4496 	kn->kn_hook = (caddr_t)vp;
4497 
4498 	v_addpollinfo(vp);
4499 	if (vp->v_pollinfo == NULL)
4500 		return (ENOMEM);
4501 	knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4502 	vhold(vp);
4503 	knlist_add(knl, kn, 0);
4504 
4505 	return (0);
4506 }
4507 
4508 /*
4509  * Detach knote from vnode
4510  */
4511 static void
4512 filt_vfsdetach(struct knote *kn)
4513 {
4514 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4515 
4516 	KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4517 	knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4518 	vdrop(vp);
4519 }
4520 
4521 /*ARGSUSED*/
4522 static int
4523 filt_vfsread(struct knote *kn, long hint)
4524 {
4525 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4526 	struct vattr va;
4527 	int res;
4528 
4529 	/*
4530 	 * filesystem is gone, so set the EOF flag and schedule
4531 	 * the knote for deletion.
4532 	 */
4533 	if (hint == NOTE_REVOKE) {
4534 		VI_LOCK(vp);
4535 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4536 		VI_UNLOCK(vp);
4537 		return (1);
4538 	}
4539 
4540 	if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4541 		return (0);
4542 
4543 	VI_LOCK(vp);
4544 	kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4545 	res = (kn->kn_data != 0);
4546 	VI_UNLOCK(vp);
4547 	return (res);
4548 }
4549 
4550 /*ARGSUSED*/
4551 static int
4552 filt_vfswrite(struct knote *kn, long hint)
4553 {
4554 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4555 
4556 	VI_LOCK(vp);
4557 
4558 	/*
4559 	 * filesystem is gone, so set the EOF flag and schedule
4560 	 * the knote for deletion.
4561 	 */
4562 	if (hint == NOTE_REVOKE)
4563 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4564 
4565 	kn->kn_data = 0;
4566 	VI_UNLOCK(vp);
4567 	return (1);
4568 }
4569 
4570 static int
4571 filt_vfsvnode(struct knote *kn, long hint)
4572 {
4573 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4574 	int res;
4575 
4576 	VI_LOCK(vp);
4577 	if (kn->kn_sfflags & hint)
4578 		kn->kn_fflags |= hint;
4579 	if (hint == NOTE_REVOKE) {
4580 		kn->kn_flags |= EV_EOF;
4581 		VI_UNLOCK(vp);
4582 		return (1);
4583 	}
4584 	res = (kn->kn_fflags != 0);
4585 	VI_UNLOCK(vp);
4586 	return (res);
4587 }
4588 
4589 int
4590 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4591 {
4592 	int error;
4593 
4594 	if (dp->d_reclen > ap->a_uio->uio_resid)
4595 		return (ENAMETOOLONG);
4596 	error = uiomove(dp, dp->d_reclen, ap->a_uio);
4597 	if (error) {
4598 		if (ap->a_ncookies != NULL) {
4599 			if (ap->a_cookies != NULL)
4600 				free(ap->a_cookies, M_TEMP);
4601 			ap->a_cookies = NULL;
4602 			*ap->a_ncookies = 0;
4603 		}
4604 		return (error);
4605 	}
4606 	if (ap->a_ncookies == NULL)
4607 		return (0);
4608 
4609 	KASSERT(ap->a_cookies,
4610 	    ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4611 
4612 	*ap->a_cookies = realloc(*ap->a_cookies,
4613 	    (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4614 	(*ap->a_cookies)[*ap->a_ncookies] = off;
4615 	return (0);
4616 }
4617 
4618 /*
4619  * Mark for update the access time of the file if the filesystem
4620  * supports VOP_MARKATIME.  This functionality is used by execve and
4621  * mmap, so we want to avoid the I/O implied by directly setting
4622  * va_atime for the sake of efficiency.
4623  */
4624 void
4625 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4626 {
4627 	struct mount *mp;
4628 
4629 	mp = vp->v_mount;
4630 	ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4631 	if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4632 		(void)VOP_MARKATIME(vp);
4633 }
4634 
4635 /*
4636  * The purpose of this routine is to remove granularity from accmode_t,
4637  * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4638  * VADMIN and VAPPEND.
4639  *
4640  * If it returns 0, the caller is supposed to continue with the usual
4641  * access checks using 'accmode' as modified by this routine.  If it
4642  * returns nonzero value, the caller is supposed to return that value
4643  * as errno.
4644  *
4645  * Note that after this routine runs, accmode may be zero.
4646  */
4647 int
4648 vfs_unixify_accmode(accmode_t *accmode)
4649 {
4650 	/*
4651 	 * There is no way to specify explicit "deny" rule using
4652 	 * file mode or POSIX.1e ACLs.
4653 	 */
4654 	if (*accmode & VEXPLICIT_DENY) {
4655 		*accmode = 0;
4656 		return (0);
4657 	}
4658 
4659 	/*
4660 	 * None of these can be translated into usual access bits.
4661 	 * Also, the common case for NFSv4 ACLs is to not contain
4662 	 * either of these bits. Caller should check for VWRITE
4663 	 * on the containing directory instead.
4664 	 */
4665 	if (*accmode & (VDELETE_CHILD | VDELETE))
4666 		return (EPERM);
4667 
4668 	if (*accmode & VADMIN_PERMS) {
4669 		*accmode &= ~VADMIN_PERMS;
4670 		*accmode |= VADMIN;
4671 	}
4672 
4673 	/*
4674 	 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4675 	 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4676 	 */
4677 	*accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
4678 
4679 	return (0);
4680 }
4681 
4682 /*
4683  * These are helper functions for filesystems to traverse all
4684  * their vnodes.  See MNT_VNODE_FOREACH_ALL() in sys/mount.h.
4685  *
4686  * This interface replaces MNT_VNODE_FOREACH.
4687  */
4688 
4689 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
4690 
4691 struct vnode *
4692 __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp)
4693 {
4694 	struct vnode *vp;
4695 
4696 	if (should_yield())
4697 		kern_yield(PRI_USER);
4698 	MNT_ILOCK(mp);
4699 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4700 	vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
4701 	while (vp != NULL && (vp->v_type == VMARKER ||
4702 	    (vp->v_iflag & VI_DOOMED) != 0))
4703 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
4704 
4705 	/* Check if we are done */
4706 	if (vp == NULL) {
4707 		__mnt_vnode_markerfree_all(mvp, mp);
4708 		/* MNT_IUNLOCK(mp); -- done in above function */
4709 		mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
4710 		return (NULL);
4711 	}
4712 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4713 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4714 	VI_LOCK(vp);
4715 	MNT_IUNLOCK(mp);
4716 	return (vp);
4717 }
4718 
4719 struct vnode *
4720 __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp)
4721 {
4722 	struct vnode *vp;
4723 
4724 	*mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4725 	MNT_ILOCK(mp);
4726 	MNT_REF(mp);
4727 	(*mvp)->v_type = VMARKER;
4728 
4729 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
4730 	while (vp != NULL && (vp->v_type == VMARKER ||
4731 	    (vp->v_iflag & VI_DOOMED) != 0))
4732 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
4733 
4734 	/* Check if we are done */
4735 	if (vp == NULL) {
4736 		MNT_REL(mp);
4737 		MNT_IUNLOCK(mp);
4738 		free(*mvp, M_VNODE_MARKER);
4739 		*mvp = NULL;
4740 		return (NULL);
4741 	}
4742 	(*mvp)->v_mount = mp;
4743 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4744 	VI_LOCK(vp);
4745 	MNT_IUNLOCK(mp);
4746 	return (vp);
4747 }
4748 
4749 
4750 void
4751 __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp)
4752 {
4753 
4754 	if (*mvp == NULL) {
4755 		MNT_IUNLOCK(mp);
4756 		return;
4757 	}
4758 
4759 	mtx_assert(MNT_MTX(mp), MA_OWNED);
4760 
4761 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4762 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4763 	MNT_REL(mp);
4764 	MNT_IUNLOCK(mp);
4765 	free(*mvp, M_VNODE_MARKER);
4766 	*mvp = NULL;
4767 }
4768 
4769 /*
4770  * These are helper functions for filesystems to traverse their
4771  * active vnodes.  See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h
4772  */
4773 static void
4774 mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4775 {
4776 
4777 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4778 
4779 	MNT_ILOCK(mp);
4780 	MNT_REL(mp);
4781 	MNT_IUNLOCK(mp);
4782 	free(*mvp, M_VNODE_MARKER);
4783 	*mvp = NULL;
4784 }
4785 
4786 static struct vnode *
4787 mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4788 {
4789 	struct vnode *vp, *nvp;
4790 
4791 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
4792 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4793 restart:
4794 	vp = TAILQ_NEXT(*mvp, v_actfreelist);
4795 	TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4796 	while (vp != NULL) {
4797 		if (vp->v_type == VMARKER) {
4798 			vp = TAILQ_NEXT(vp, v_actfreelist);
4799 			continue;
4800 		}
4801 		if (!VI_TRYLOCK(vp)) {
4802 			if (mp_ncpus == 1 || should_yield()) {
4803 				TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
4804 				mtx_unlock(&vnode_free_list_mtx);
4805 				pause("vnacti", 1);
4806 				mtx_lock(&vnode_free_list_mtx);
4807 				goto restart;
4808 			}
4809 			continue;
4810 		}
4811 		KASSERT(vp->v_type != VMARKER, ("locked marker %p", vp));
4812 		KASSERT(vp->v_mount == mp || vp->v_mount == NULL,
4813 		    ("alien vnode on the active list %p %p", vp, mp));
4814 		if (vp->v_mount == mp && (vp->v_iflag & VI_DOOMED) == 0)
4815 			break;
4816 		nvp = TAILQ_NEXT(vp, v_actfreelist);
4817 		VI_UNLOCK(vp);
4818 		vp = nvp;
4819 	}
4820 
4821 	/* Check if we are done */
4822 	if (vp == NULL) {
4823 		mtx_unlock(&vnode_free_list_mtx);
4824 		mnt_vnode_markerfree_active(mvp, mp);
4825 		return (NULL);
4826 	}
4827 	TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
4828 	mtx_unlock(&vnode_free_list_mtx);
4829 	ASSERT_VI_LOCKED(vp, "active iter");
4830 	KASSERT((vp->v_iflag & VI_ACTIVE) != 0, ("Non-active vp %p", vp));
4831 	return (vp);
4832 }
4833 
4834 struct vnode *
4835 __mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4836 {
4837 
4838 	if (should_yield())
4839 		kern_yield(PRI_USER);
4840 	mtx_lock(&vnode_free_list_mtx);
4841 	return (mnt_vnode_next_active(mvp, mp));
4842 }
4843 
4844 struct vnode *
4845 __mnt_vnode_first_active(struct vnode **mvp, struct mount *mp)
4846 {
4847 	struct vnode *vp;
4848 
4849 	*mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4850 	MNT_ILOCK(mp);
4851 	MNT_REF(mp);
4852 	MNT_IUNLOCK(mp);
4853 	(*mvp)->v_type = VMARKER;
4854 	(*mvp)->v_mount = mp;
4855 
4856 	mtx_lock(&vnode_free_list_mtx);
4857 	vp = TAILQ_FIRST(&mp->mnt_activevnodelist);
4858 	if (vp == NULL) {
4859 		mtx_unlock(&vnode_free_list_mtx);
4860 		mnt_vnode_markerfree_active(mvp, mp);
4861 		return (NULL);
4862 	}
4863 	TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
4864 	return (mnt_vnode_next_active(mvp, mp));
4865 }
4866 
4867 void
4868 __mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4869 {
4870 
4871 	if (*mvp == NULL)
4872 		return;
4873 
4874 	mtx_lock(&vnode_free_list_mtx);
4875 	TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4876 	mtx_unlock(&vnode_free_list_mtx);
4877 	mnt_vnode_markerfree_active(mvp, mp);
4878 }
4879