xref: /freebsd/sys/kern/vfs_subr.c (revision 5ca34122ecdd5abc62bdae39663fec9ac8523d87)
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/refcount.h>
72 #include <sys/rwlock.h>
73 #include <sys/sched.h>
74 #include <sys/sleepqueue.h>
75 #include <sys/smp.h>
76 #include <sys/stat.h>
77 #include <sys/sysctl.h>
78 #include <sys/syslog.h>
79 #include <sys/vmmeter.h>
80 #include <sys/vnode.h>
81 #include <sys/watchdog.h>
82 
83 #include <machine/stdarg.h>
84 
85 #include <security/mac/mac_framework.h>
86 
87 #include <vm/vm.h>
88 #include <vm/vm_object.h>
89 #include <vm/vm_extern.h>
90 #include <vm/pmap.h>
91 #include <vm/vm_map.h>
92 #include <vm/vm_page.h>
93 #include <vm/vm_kern.h>
94 #include <vm/uma.h>
95 
96 #ifdef DDB
97 #include <ddb/ddb.h>
98 #endif
99 
100 static void	delmntque(struct vnode *vp);
101 static int	flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
102 		    int slpflag, int slptimeo);
103 static void	syncer_shutdown(void *arg, int howto);
104 static int	vtryrecycle(struct vnode *vp);
105 static void	v_init_counters(struct vnode *);
106 static void	v_incr_usecount(struct vnode *);
107 static void	v_incr_devcount(struct vnode *);
108 static void	v_decr_devcount(struct vnode *);
109 static void	vnlru_free(int);
110 static void	vgonel(struct vnode *);
111 static void	vfs_knllock(void *arg);
112 static void	vfs_knlunlock(void *arg);
113 static void	vfs_knl_assert_locked(void *arg);
114 static void	vfs_knl_assert_unlocked(void *arg);
115 static void	destroy_vpollinfo(struct vpollinfo *vi);
116 
117 /*
118  * Number of vnodes in existence.  Increased whenever getnewvnode()
119  * allocates a new vnode, decreased in vdropl() for VI_DOOMED vnode.
120  */
121 static unsigned long	numvnodes;
122 
123 SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
124     "Number of vnodes in existence");
125 
126 static u_long vnodes_created;
127 SYSCTL_ULONG(_vfs, OID_AUTO, vnodes_created, CTLFLAG_RD, &vnodes_created,
128     0, "Number of vnodes created by getnewvnode");
129 
130 /*
131  * Conversion tables for conversion from vnode types to inode formats
132  * and back.
133  */
134 enum vtype iftovt_tab[16] = {
135 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
136 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
137 };
138 int vttoif_tab[10] = {
139 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
140 	S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
141 };
142 
143 /*
144  * List of vnodes that are ready for recycling.
145  */
146 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
147 
148 /*
149  * Free vnode target.  Free vnodes may simply be files which have been stat'd
150  * but not read.  This is somewhat common, and a small cache of such files
151  * should be kept to avoid recreation costs.
152  */
153 static u_long wantfreevnodes;
154 SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
155 /* Number of vnodes in the free list. */
156 static u_long freevnodes;
157 SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
158     "Number of vnodes in the free list");
159 
160 static int vlru_allow_cache_src;
161 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
162     &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
163 
164 static u_long recycles_count;
165 SYSCTL_ULONG(_vfs, OID_AUTO, recycles, CTLFLAG_RD, &recycles_count, 0,
166     "Number of vnodes recycled to avoid exceding kern.maxvnodes");
167 
168 /*
169  * Various variables used for debugging the new implementation of
170  * reassignbuf().
171  * XXX these are probably of (very) limited utility now.
172  */
173 static int reassignbufcalls;
174 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
175     "Number of calls to reassignbuf");
176 
177 static u_long free_owe_inact;
178 SYSCTL_ULONG(_vfs, OID_AUTO, free_owe_inact, CTLFLAG_RD, &free_owe_inact, 0,
179     "Number of times free vnodes kept on active list due to VFS "
180     "owing inactivation");
181 
182 /*
183  * Cache for the mount type id assigned to NFS.  This is used for
184  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
185  */
186 int	nfs_mount_type = -1;
187 
188 /* To keep more than one thread at a time from running vfs_getnewfsid */
189 static struct mtx mntid_mtx;
190 
191 /*
192  * Lock for any access to the following:
193  *	vnode_free_list
194  *	numvnodes
195  *	freevnodes
196  */
197 static struct mtx vnode_free_list_mtx;
198 
199 /* Publicly exported FS */
200 struct nfs_public nfs_pub;
201 
202 static uma_zone_t buf_trie_zone;
203 
204 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
205 static uma_zone_t vnode_zone;
206 static uma_zone_t vnodepoll_zone;
207 
208 /*
209  * The workitem queue.
210  *
211  * It is useful to delay writes of file data and filesystem metadata
212  * for tens of seconds so that quickly created and deleted files need
213  * not waste disk bandwidth being created and removed. To realize this,
214  * we append vnodes to a "workitem" queue. When running with a soft
215  * updates implementation, most pending metadata dependencies should
216  * not wait for more than a few seconds. Thus, mounted on block devices
217  * are delayed only about a half the time that file data is delayed.
218  * Similarly, directory updates are more critical, so are only delayed
219  * about a third the time that file data is delayed. Thus, there are
220  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
221  * one each second (driven off the filesystem syncer process). The
222  * syncer_delayno variable indicates the next queue that is to be processed.
223  * Items that need to be processed soon are placed in this queue:
224  *
225  *	syncer_workitem_pending[syncer_delayno]
226  *
227  * A delay of fifteen seconds is done by placing the request fifteen
228  * entries later in the queue:
229  *
230  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
231  *
232  */
233 static int syncer_delayno;
234 static long syncer_mask;
235 LIST_HEAD(synclist, bufobj);
236 static struct synclist *syncer_workitem_pending;
237 /*
238  * The sync_mtx protects:
239  *	bo->bo_synclist
240  *	sync_vnode_count
241  *	syncer_delayno
242  *	syncer_state
243  *	syncer_workitem_pending
244  *	syncer_worklist_len
245  *	rushjob
246  */
247 static struct mtx sync_mtx;
248 static struct cv sync_wakeup;
249 
250 #define SYNCER_MAXDELAY		32
251 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
252 static int syncdelay = 30;		/* max time to delay syncing data */
253 static int filedelay = 30;		/* time to delay syncing files */
254 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
255     "Time to delay syncing files (in seconds)");
256 static int dirdelay = 29;		/* time to delay syncing directories */
257 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
258     "Time to delay syncing directories (in seconds)");
259 static int metadelay = 28;		/* time to delay syncing metadata */
260 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
261     "Time to delay syncing metadata (in seconds)");
262 static int rushjob;		/* number of slots to run ASAP */
263 static int stat_rush_requests;	/* number of times I/O speeded up */
264 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
265     "Number of times I/O speeded up (rush requests)");
266 
267 /*
268  * When shutting down the syncer, run it at four times normal speed.
269  */
270 #define SYNCER_SHUTDOWN_SPEEDUP		4
271 static int sync_vnode_count;
272 static int syncer_worklist_len;
273 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
274     syncer_state;
275 
276 /*
277  * Number of vnodes we want to exist at any one time.  This is mostly used
278  * to size hash tables in vnode-related code.  It is normally not used in
279  * getnewvnode(), as wantfreevnodes is normally nonzero.)
280  *
281  * XXX desiredvnodes is historical cruft and should not exist.
282  */
283 int desiredvnodes;
284 
285 static int
286 sysctl_update_desiredvnodes(SYSCTL_HANDLER_ARGS)
287 {
288 	int error, old_desiredvnodes;
289 
290 	old_desiredvnodes = desiredvnodes;
291 	if ((error = sysctl_handle_int(oidp, arg1, arg2, req)) != 0)
292 		return (error);
293 	if (old_desiredvnodes != desiredvnodes) {
294 		vfs_hash_changesize(desiredvnodes);
295 		cache_changesize(desiredvnodes);
296 	}
297 	return (0);
298 }
299 
300 SYSCTL_PROC(_kern, KERN_MAXVNODES, maxvnodes,
301     CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, &desiredvnodes, 0,
302     sysctl_update_desiredvnodes, "I", "Maximum number of vnodes");
303 SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
304     &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
305 static int vnlru_nowhere;
306 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
307     &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
308 
309 /* Shift count for (uintptr_t)vp to initialize vp->v_hash. */
310 static int vnsz2log;
311 
312 /*
313  * Support for the bufobj clean & dirty pctrie.
314  */
315 static void *
316 buf_trie_alloc(struct pctrie *ptree)
317 {
318 
319 	return uma_zalloc(buf_trie_zone, M_NOWAIT);
320 }
321 
322 static void
323 buf_trie_free(struct pctrie *ptree, void *node)
324 {
325 
326 	uma_zfree(buf_trie_zone, node);
327 }
328 PCTRIE_DEFINE(BUF, buf, b_lblkno, buf_trie_alloc, buf_trie_free);
329 
330 /*
331  * Initialize the vnode management data structures.
332  *
333  * Reevaluate the following cap on the number of vnodes after the physical
334  * memory size exceeds 512GB.  In the limit, as the physical memory size
335  * grows, the ratio of physical pages to vnodes approaches sixteen to one.
336  */
337 #ifndef	MAXVNODES_MAX
338 #define	MAXVNODES_MAX	(512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
339 #endif
340 static void
341 vntblinit(void *dummy __unused)
342 {
343 	u_int i;
344 	int physvnodes, virtvnodes;
345 
346 	/*
347 	 * Desiredvnodes is a function of the physical memory size and the
348 	 * kernel's heap size.  Generally speaking, it scales with the
349 	 * physical memory size.  The ratio of desiredvnodes to physical pages
350 	 * is one to four until desiredvnodes exceeds 98,304.  Thereafter, the
351 	 * marginal ratio of desiredvnodes to physical pages is one to
352 	 * sixteen.  However, desiredvnodes is limited by the kernel's heap
353 	 * size.  The memory required by desiredvnodes vnodes and vm objects
354 	 * may not exceed one seventh of the kernel's heap size.
355 	 */
356 	physvnodes = maxproc + vm_cnt.v_page_count / 16 + 3 * min(98304 * 4,
357 	    vm_cnt.v_page_count) / 16;
358 	virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
359 	    sizeof(struct vnode)));
360 	desiredvnodes = min(physvnodes, virtvnodes);
361 	if (desiredvnodes > MAXVNODES_MAX) {
362 		if (bootverbose)
363 			printf("Reducing kern.maxvnodes %d -> %d\n",
364 			    desiredvnodes, MAXVNODES_MAX);
365 		desiredvnodes = MAXVNODES_MAX;
366 	}
367 	wantfreevnodes = desiredvnodes / 4;
368 	mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
369 	TAILQ_INIT(&vnode_free_list);
370 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
371 	vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
372 	    NULL, NULL, UMA_ALIGN_PTR, 0);
373 	vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
374 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
375 	/*
376 	 * Preallocate enough nodes to support one-per buf so that
377 	 * we can not fail an insert.  reassignbuf() callers can not
378 	 * tolerate the insertion failure.
379 	 */
380 	buf_trie_zone = uma_zcreate("BUF TRIE", pctrie_node_size(),
381 	    NULL, NULL, pctrie_zone_init, NULL, UMA_ALIGN_PTR,
382 	    UMA_ZONE_NOFREE | UMA_ZONE_VM);
383 	uma_prealloc(buf_trie_zone, nbuf);
384 	/*
385 	 * Initialize the filesystem syncer.
386 	 */
387 	syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
388 	    &syncer_mask);
389 	syncer_maxdelay = syncer_mask + 1;
390 	mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
391 	cv_init(&sync_wakeup, "syncer");
392 	for (i = 1; i <= sizeof(struct vnode); i <<= 1)
393 		vnsz2log++;
394 	vnsz2log--;
395 }
396 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
397 
398 
399 /*
400  * Mark a mount point as busy. Used to synchronize access and to delay
401  * unmounting. Eventually, mountlist_mtx is not released on failure.
402  *
403  * vfs_busy() is a custom lock, it can block the caller.
404  * vfs_busy() only sleeps if the unmount is active on the mount point.
405  * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
406  * vnode belonging to mp.
407  *
408  * Lookup uses vfs_busy() to traverse mount points.
409  * root fs			var fs
410  * / vnode lock		A	/ vnode lock (/var)		D
411  * /var vnode lock	B	/log vnode lock(/var/log)	E
412  * vfs_busy lock	C	vfs_busy lock			F
413  *
414  * Within each file system, the lock order is C->A->B and F->D->E.
415  *
416  * When traversing across mounts, the system follows that lock order:
417  *
418  *        C->A->B
419  *              |
420  *              +->F->D->E
421  *
422  * The lookup() process for namei("/var") illustrates the process:
423  *  VOP_LOOKUP() obtains B while A is held
424  *  vfs_busy() obtains a shared lock on F while A and B are held
425  *  vput() releases lock on B
426  *  vput() releases lock on A
427  *  VFS_ROOT() obtains lock on D while shared lock on F is held
428  *  vfs_unbusy() releases shared lock on F
429  *  vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
430  *    Attempt to lock A (instead of vp_crossmp) while D is held would
431  *    violate the global order, causing deadlocks.
432  *
433  * dounmount() locks B while F is drained.
434  */
435 int
436 vfs_busy(struct mount *mp, int flags)
437 {
438 
439 	MPASS((flags & ~MBF_MASK) == 0);
440 	CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
441 
442 	MNT_ILOCK(mp);
443 	MNT_REF(mp);
444 	/*
445 	 * If mount point is currenly being unmounted, sleep until the
446 	 * mount point fate is decided.  If thread doing the unmounting fails,
447 	 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
448 	 * that this mount point has survived the unmount attempt and vfs_busy
449 	 * should retry.  Otherwise the unmounter thread will set MNTK_REFEXPIRE
450 	 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
451 	 * about to be really destroyed.  vfs_busy needs to release its
452 	 * reference on the mount point in this case and return with ENOENT,
453 	 * telling the caller that mount mount it tried to busy is no longer
454 	 * valid.
455 	 */
456 	while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
457 		if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
458 			MNT_REL(mp);
459 			MNT_IUNLOCK(mp);
460 			CTR1(KTR_VFS, "%s: failed busying before sleeping",
461 			    __func__);
462 			return (ENOENT);
463 		}
464 		if (flags & MBF_MNTLSTLOCK)
465 			mtx_unlock(&mountlist_mtx);
466 		mp->mnt_kern_flag |= MNTK_MWAIT;
467 		msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
468 		if (flags & MBF_MNTLSTLOCK)
469 			mtx_lock(&mountlist_mtx);
470 		MNT_ILOCK(mp);
471 	}
472 	if (flags & MBF_MNTLSTLOCK)
473 		mtx_unlock(&mountlist_mtx);
474 	mp->mnt_lockref++;
475 	MNT_IUNLOCK(mp);
476 	return (0);
477 }
478 
479 /*
480  * Free a busy filesystem.
481  */
482 void
483 vfs_unbusy(struct mount *mp)
484 {
485 
486 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
487 	MNT_ILOCK(mp);
488 	MNT_REL(mp);
489 	KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
490 	mp->mnt_lockref--;
491 	if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
492 		MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
493 		CTR1(KTR_VFS, "%s: waking up waiters", __func__);
494 		mp->mnt_kern_flag &= ~MNTK_DRAINING;
495 		wakeup(&mp->mnt_lockref);
496 	}
497 	MNT_IUNLOCK(mp);
498 }
499 
500 /*
501  * Lookup a mount point by filesystem identifier.
502  */
503 struct mount *
504 vfs_getvfs(fsid_t *fsid)
505 {
506 	struct mount *mp;
507 
508 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
509 	mtx_lock(&mountlist_mtx);
510 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
511 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
512 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
513 			vfs_ref(mp);
514 			mtx_unlock(&mountlist_mtx);
515 			return (mp);
516 		}
517 	}
518 	mtx_unlock(&mountlist_mtx);
519 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
520 	return ((struct mount *) 0);
521 }
522 
523 /*
524  * Lookup a mount point by filesystem identifier, busying it before
525  * returning.
526  *
527  * To avoid congestion on mountlist_mtx, implement simple direct-mapped
528  * cache for popular filesystem identifiers.  The cache is lockess, using
529  * the fact that struct mount's are never freed.  In worst case we may
530  * get pointer to unmounted or even different filesystem, so we have to
531  * check what we got, and go slow way if so.
532  */
533 struct mount *
534 vfs_busyfs(fsid_t *fsid)
535 {
536 #define	FSID_CACHE_SIZE	256
537 	typedef struct mount * volatile vmp_t;
538 	static vmp_t cache[FSID_CACHE_SIZE];
539 	struct mount *mp;
540 	int error;
541 	uint32_t hash;
542 
543 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
544 	hash = fsid->val[0] ^ fsid->val[1];
545 	hash = (hash >> 16 ^ hash) & (FSID_CACHE_SIZE - 1);
546 	mp = cache[hash];
547 	if (mp == NULL ||
548 	    mp->mnt_stat.f_fsid.val[0] != fsid->val[0] ||
549 	    mp->mnt_stat.f_fsid.val[1] != fsid->val[1])
550 		goto slow;
551 	if (vfs_busy(mp, 0) != 0) {
552 		cache[hash] = NULL;
553 		goto slow;
554 	}
555 	if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
556 	    mp->mnt_stat.f_fsid.val[1] == fsid->val[1])
557 		return (mp);
558 	else
559 	    vfs_unbusy(mp);
560 
561 slow:
562 	mtx_lock(&mountlist_mtx);
563 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
564 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
565 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
566 			error = vfs_busy(mp, MBF_MNTLSTLOCK);
567 			if (error) {
568 				cache[hash] = NULL;
569 				mtx_unlock(&mountlist_mtx);
570 				return (NULL);
571 			}
572 			cache[hash] = mp;
573 			return (mp);
574 		}
575 	}
576 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
577 	mtx_unlock(&mountlist_mtx);
578 	return ((struct mount *) 0);
579 }
580 
581 /*
582  * Check if a user can access privileged mount options.
583  */
584 int
585 vfs_suser(struct mount *mp, struct thread *td)
586 {
587 	int error;
588 
589 	/*
590 	 * If the thread is jailed, but this is not a jail-friendly file
591 	 * system, deny immediately.
592 	 */
593 	if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
594 		return (EPERM);
595 
596 	/*
597 	 * If the file system was mounted outside the jail of the calling
598 	 * thread, deny immediately.
599 	 */
600 	if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
601 		return (EPERM);
602 
603 	/*
604 	 * If file system supports delegated administration, we don't check
605 	 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
606 	 * by the file system itself.
607 	 * If this is not the user that did original mount, we check for
608 	 * the PRIV_VFS_MOUNT_OWNER privilege.
609 	 */
610 	if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
611 	    mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
612 		if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
613 			return (error);
614 	}
615 	return (0);
616 }
617 
618 /*
619  * Get a new unique fsid.  Try to make its val[0] unique, since this value
620  * will be used to create fake device numbers for stat().  Also try (but
621  * not so hard) make its val[0] unique mod 2^16, since some emulators only
622  * support 16-bit device numbers.  We end up with unique val[0]'s for the
623  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
624  *
625  * Keep in mind that several mounts may be running in parallel.  Starting
626  * the search one past where the previous search terminated is both a
627  * micro-optimization and a defense against returning the same fsid to
628  * different mounts.
629  */
630 void
631 vfs_getnewfsid(struct mount *mp)
632 {
633 	static uint16_t mntid_base;
634 	struct mount *nmp;
635 	fsid_t tfsid;
636 	int mtype;
637 
638 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
639 	mtx_lock(&mntid_mtx);
640 	mtype = mp->mnt_vfc->vfc_typenum;
641 	tfsid.val[1] = mtype;
642 	mtype = (mtype & 0xFF) << 24;
643 	for (;;) {
644 		tfsid.val[0] = makedev(255,
645 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
646 		mntid_base++;
647 		if ((nmp = vfs_getvfs(&tfsid)) == NULL)
648 			break;
649 		vfs_rel(nmp);
650 	}
651 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
652 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
653 	mtx_unlock(&mntid_mtx);
654 }
655 
656 /*
657  * Knob to control the precision of file timestamps:
658  *
659  *   0 = seconds only; nanoseconds zeroed.
660  *   1 = seconds and nanoseconds, accurate within 1/HZ.
661  *   2 = seconds and nanoseconds, truncated to microseconds.
662  * >=3 = seconds and nanoseconds, maximum precision.
663  */
664 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
665 
666 static int timestamp_precision = TSP_USEC;
667 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
668     &timestamp_precision, 0, "File timestamp precision (0: seconds, "
669     "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
670     "3+: sec + ns (max. precision))");
671 
672 /*
673  * Get a current timestamp.
674  */
675 void
676 vfs_timestamp(struct timespec *tsp)
677 {
678 	struct timeval tv;
679 
680 	switch (timestamp_precision) {
681 	case TSP_SEC:
682 		tsp->tv_sec = time_second;
683 		tsp->tv_nsec = 0;
684 		break;
685 	case TSP_HZ:
686 		getnanotime(tsp);
687 		break;
688 	case TSP_USEC:
689 		microtime(&tv);
690 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
691 		break;
692 	case TSP_NSEC:
693 	default:
694 		nanotime(tsp);
695 		break;
696 	}
697 }
698 
699 /*
700  * Set vnode attributes to VNOVAL
701  */
702 void
703 vattr_null(struct vattr *vap)
704 {
705 
706 	vap->va_type = VNON;
707 	vap->va_size = VNOVAL;
708 	vap->va_bytes = VNOVAL;
709 	vap->va_mode = VNOVAL;
710 	vap->va_nlink = VNOVAL;
711 	vap->va_uid = VNOVAL;
712 	vap->va_gid = VNOVAL;
713 	vap->va_fsid = VNOVAL;
714 	vap->va_fileid = VNOVAL;
715 	vap->va_blocksize = VNOVAL;
716 	vap->va_rdev = VNOVAL;
717 	vap->va_atime.tv_sec = VNOVAL;
718 	vap->va_atime.tv_nsec = VNOVAL;
719 	vap->va_mtime.tv_sec = VNOVAL;
720 	vap->va_mtime.tv_nsec = VNOVAL;
721 	vap->va_ctime.tv_sec = VNOVAL;
722 	vap->va_ctime.tv_nsec = VNOVAL;
723 	vap->va_birthtime.tv_sec = VNOVAL;
724 	vap->va_birthtime.tv_nsec = VNOVAL;
725 	vap->va_flags = VNOVAL;
726 	vap->va_gen = VNOVAL;
727 	vap->va_vaflags = 0;
728 }
729 
730 /*
731  * This routine is called when we have too many vnodes.  It attempts
732  * to free <count> vnodes and will potentially free vnodes that still
733  * have VM backing store (VM backing store is typically the cause
734  * of a vnode blowout so we want to do this).  Therefore, this operation
735  * is not considered cheap.
736  *
737  * A number of conditions may prevent a vnode from being reclaimed.
738  * the buffer cache may have references on the vnode, a directory
739  * vnode may still have references due to the namei cache representing
740  * underlying files, or the vnode may be in active use.   It is not
741  * desireable to reuse such vnodes.  These conditions may cause the
742  * number of vnodes to reach some minimum value regardless of what
743  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
744  */
745 static int
746 vlrureclaim(struct mount *mp)
747 {
748 	struct vnode *vp;
749 	int done;
750 	int trigger;
751 	int usevnodes;
752 	int count;
753 
754 	/*
755 	 * Calculate the trigger point, don't allow user
756 	 * screwups to blow us up.   This prevents us from
757 	 * recycling vnodes with lots of resident pages.  We
758 	 * aren't trying to free memory, we are trying to
759 	 * free vnodes.
760 	 */
761 	usevnodes = desiredvnodes;
762 	if (usevnodes <= 0)
763 		usevnodes = 1;
764 	trigger = vm_cnt.v_page_count * 2 / usevnodes;
765 	done = 0;
766 	vn_start_write(NULL, &mp, V_WAIT);
767 	MNT_ILOCK(mp);
768 	count = mp->mnt_nvnodelistsize / 10 + 1;
769 	while (count != 0) {
770 		vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
771 		while (vp != NULL && vp->v_type == VMARKER)
772 			vp = TAILQ_NEXT(vp, v_nmntvnodes);
773 		if (vp == NULL)
774 			break;
775 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
776 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
777 		--count;
778 		if (!VI_TRYLOCK(vp))
779 			goto next_iter;
780 		/*
781 		 * If it's been deconstructed already, it's still
782 		 * referenced, or it exceeds the trigger, skip it.
783 		 */
784 		if (vp->v_usecount ||
785 		    (!vlru_allow_cache_src &&
786 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
787 		    (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
788 		    vp->v_object->resident_page_count > trigger)) {
789 			VI_UNLOCK(vp);
790 			goto next_iter;
791 		}
792 		MNT_IUNLOCK(mp);
793 		vholdl(vp);
794 		if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
795 			vdrop(vp);
796 			goto next_iter_mntunlocked;
797 		}
798 		VI_LOCK(vp);
799 		/*
800 		 * v_usecount may have been bumped after VOP_LOCK() dropped
801 		 * the vnode interlock and before it was locked again.
802 		 *
803 		 * It is not necessary to recheck VI_DOOMED because it can
804 		 * only be set by another thread that holds both the vnode
805 		 * lock and vnode interlock.  If another thread has the
806 		 * vnode lock before we get to VOP_LOCK() and obtains the
807 		 * vnode interlock after VOP_LOCK() drops the vnode
808 		 * interlock, the other thread will be unable to drop the
809 		 * vnode lock before our VOP_LOCK() call fails.
810 		 */
811 		if (vp->v_usecount ||
812 		    (!vlru_allow_cache_src &&
813 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
814 		    (vp->v_object != NULL &&
815 		    vp->v_object->resident_page_count > trigger)) {
816 			VOP_UNLOCK(vp, LK_INTERLOCK);
817 			vdrop(vp);
818 			goto next_iter_mntunlocked;
819 		}
820 		KASSERT((vp->v_iflag & VI_DOOMED) == 0,
821 		    ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
822 		atomic_add_long(&recycles_count, 1);
823 		vgonel(vp);
824 		VOP_UNLOCK(vp, 0);
825 		vdropl(vp);
826 		done++;
827 next_iter_mntunlocked:
828 		if (!should_yield())
829 			goto relock_mnt;
830 		goto yield;
831 next_iter:
832 		if (!should_yield())
833 			continue;
834 		MNT_IUNLOCK(mp);
835 yield:
836 		kern_yield(PRI_USER);
837 relock_mnt:
838 		MNT_ILOCK(mp);
839 	}
840 	MNT_IUNLOCK(mp);
841 	vn_finished_write(mp);
842 	return done;
843 }
844 
845 /*
846  * Attempt to keep the free list at wantfreevnodes length.
847  */
848 static void
849 vnlru_free(int count)
850 {
851 	struct vnode *vp;
852 
853 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
854 	for (; count > 0; count--) {
855 		vp = TAILQ_FIRST(&vnode_free_list);
856 		/*
857 		 * The list can be modified while the free_list_mtx
858 		 * has been dropped and vp could be NULL here.
859 		 */
860 		if (!vp)
861 			break;
862 		VNASSERT(vp->v_op != NULL, vp,
863 		    ("vnlru_free: vnode already reclaimed."));
864 		KASSERT((vp->v_iflag & VI_FREE) != 0,
865 		    ("Removing vnode not on freelist"));
866 		KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
867 		    ("Mangling active vnode"));
868 		TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
869 		/*
870 		 * Don't recycle if we can't get the interlock.
871 		 */
872 		if (!VI_TRYLOCK(vp)) {
873 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
874 			continue;
875 		}
876 		VNASSERT((vp->v_iflag & VI_FREE) != 0 && vp->v_holdcnt == 0,
877 		    vp, ("vp inconsistent on freelist"));
878 
879 		/*
880 		 * The clear of VI_FREE prevents activation of the
881 		 * vnode.  There is no sense in putting the vnode on
882 		 * the mount point active list, only to remove it
883 		 * later during recycling.  Inline the relevant part
884 		 * of vholdl(), to avoid triggering assertions or
885 		 * activating.
886 		 */
887 		freevnodes--;
888 		vp->v_iflag &= ~VI_FREE;
889 		refcount_acquire(&vp->v_holdcnt);
890 
891 		mtx_unlock(&vnode_free_list_mtx);
892 		VI_UNLOCK(vp);
893 		vtryrecycle(vp);
894 		/*
895 		 * If the recycled succeeded this vdrop will actually free
896 		 * the vnode.  If not it will simply place it back on
897 		 * the free list.
898 		 */
899 		vdrop(vp);
900 		mtx_lock(&vnode_free_list_mtx);
901 	}
902 }
903 /*
904  * Attempt to recycle vnodes in a context that is always safe to block.
905  * Calling vlrurecycle() from the bowels of filesystem code has some
906  * interesting deadlock problems.
907  */
908 static struct proc *vnlruproc;
909 static int vnlruproc_sig;
910 
911 static void
912 vnlru_proc(void)
913 {
914 	struct mount *mp, *nmp;
915 	int done;
916 	struct proc *p = vnlruproc;
917 
918 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
919 	    SHUTDOWN_PRI_FIRST);
920 
921 	for (;;) {
922 		kproc_suspend_check(p);
923 		mtx_lock(&vnode_free_list_mtx);
924 		if (freevnodes > wantfreevnodes)
925 			vnlru_free(freevnodes - wantfreevnodes);
926 		if (numvnodes <= desiredvnodes * 9 / 10) {
927 			vnlruproc_sig = 0;
928 			wakeup(&vnlruproc_sig);
929 			msleep(vnlruproc, &vnode_free_list_mtx,
930 			    PVFS|PDROP, "vlruwt", hz);
931 			continue;
932 		}
933 		mtx_unlock(&vnode_free_list_mtx);
934 		done = 0;
935 		mtx_lock(&mountlist_mtx);
936 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
937 			if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
938 				nmp = TAILQ_NEXT(mp, mnt_list);
939 				continue;
940 			}
941 			done += vlrureclaim(mp);
942 			mtx_lock(&mountlist_mtx);
943 			nmp = TAILQ_NEXT(mp, mnt_list);
944 			vfs_unbusy(mp);
945 		}
946 		mtx_unlock(&mountlist_mtx);
947 		if (done == 0) {
948 #if 0
949 			/* These messages are temporary debugging aids */
950 			if (vnlru_nowhere < 5)
951 				printf("vnlru process getting nowhere..\n");
952 			else if (vnlru_nowhere == 5)
953 				printf("vnlru process messages stopped.\n");
954 #endif
955 			vnlru_nowhere++;
956 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
957 		} else
958 			kern_yield(PRI_USER);
959 	}
960 }
961 
962 static struct kproc_desc vnlru_kp = {
963 	"vnlru",
964 	vnlru_proc,
965 	&vnlruproc
966 };
967 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
968     &vnlru_kp);
969 
970 /*
971  * Routines having to do with the management of the vnode table.
972  */
973 
974 /*
975  * Try to recycle a freed vnode.  We abort if anyone picks up a reference
976  * before we actually vgone().  This function must be called with the vnode
977  * held to prevent the vnode from being returned to the free list midway
978  * through vgone().
979  */
980 static int
981 vtryrecycle(struct vnode *vp)
982 {
983 	struct mount *vnmp;
984 
985 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
986 	VNASSERT(vp->v_holdcnt, vp,
987 	    ("vtryrecycle: Recycling vp %p without a reference.", vp));
988 	/*
989 	 * This vnode may found and locked via some other list, if so we
990 	 * can't recycle it yet.
991 	 */
992 	if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
993 		CTR2(KTR_VFS,
994 		    "%s: impossible to recycle, vp %p lock is already held",
995 		    __func__, vp);
996 		return (EWOULDBLOCK);
997 	}
998 	/*
999 	 * Don't recycle if its filesystem is being suspended.
1000 	 */
1001 	if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
1002 		VOP_UNLOCK(vp, 0);
1003 		CTR2(KTR_VFS,
1004 		    "%s: impossible to recycle, cannot start the write for %p",
1005 		    __func__, vp);
1006 		return (EBUSY);
1007 	}
1008 	/*
1009 	 * If we got this far, we need to acquire the interlock and see if
1010 	 * anyone picked up this vnode from another list.  If not, we will
1011 	 * mark it with DOOMED via vgonel() so that anyone who does find it
1012 	 * will skip over it.
1013 	 */
1014 	VI_LOCK(vp);
1015 	if (vp->v_usecount) {
1016 		VOP_UNLOCK(vp, LK_INTERLOCK);
1017 		vn_finished_write(vnmp);
1018 		CTR2(KTR_VFS,
1019 		    "%s: impossible to recycle, %p is already referenced",
1020 		    __func__, vp);
1021 		return (EBUSY);
1022 	}
1023 	if ((vp->v_iflag & VI_DOOMED) == 0) {
1024 		atomic_add_long(&recycles_count, 1);
1025 		vgonel(vp);
1026 	}
1027 	VOP_UNLOCK(vp, LK_INTERLOCK);
1028 	vn_finished_write(vnmp);
1029 	return (0);
1030 }
1031 
1032 /*
1033  * Wait for available vnodes.
1034  */
1035 static int
1036 getnewvnode_wait(int suspended)
1037 {
1038 
1039 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
1040 	if (numvnodes > desiredvnodes) {
1041 		if (suspended) {
1042 			/*
1043 			 * File system is beeing suspended, we cannot risk a
1044 			 * deadlock here, so allocate new vnode anyway.
1045 			 */
1046 			if (freevnodes > wantfreevnodes)
1047 				vnlru_free(freevnodes - wantfreevnodes);
1048 			return (0);
1049 		}
1050 		if (vnlruproc_sig == 0) {
1051 			vnlruproc_sig = 1;	/* avoid unnecessary wakeups */
1052 			wakeup(vnlruproc);
1053 		}
1054 		msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
1055 		    "vlruwk", hz);
1056 	}
1057 	return (numvnodes > desiredvnodes ? ENFILE : 0);
1058 }
1059 
1060 void
1061 getnewvnode_reserve(u_int count)
1062 {
1063 	struct thread *td;
1064 
1065 	td = curthread;
1066 	/* First try to be quick and racy. */
1067 	if (atomic_fetchadd_long(&numvnodes, count) + count <= desiredvnodes) {
1068 		td->td_vp_reserv += count;
1069 		return;
1070 	} else
1071 		atomic_subtract_long(&numvnodes, count);
1072 
1073 	mtx_lock(&vnode_free_list_mtx);
1074 	while (count > 0) {
1075 		if (getnewvnode_wait(0) == 0) {
1076 			count--;
1077 			td->td_vp_reserv++;
1078 			atomic_add_long(&numvnodes, 1);
1079 		}
1080 	}
1081 	mtx_unlock(&vnode_free_list_mtx);
1082 }
1083 
1084 void
1085 getnewvnode_drop_reserve(void)
1086 {
1087 	struct thread *td;
1088 
1089 	td = curthread;
1090 	atomic_subtract_long(&numvnodes, td->td_vp_reserv);
1091 	td->td_vp_reserv = 0;
1092 }
1093 
1094 /*
1095  * Return the next vnode from the free list.
1096  */
1097 int
1098 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
1099     struct vnode **vpp)
1100 {
1101 	struct vnode *vp;
1102 	struct bufobj *bo;
1103 	struct thread *td;
1104 	int error;
1105 
1106 	CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
1107 	vp = NULL;
1108 	td = curthread;
1109 	if (td->td_vp_reserv > 0) {
1110 		td->td_vp_reserv -= 1;
1111 		goto alloc;
1112 	}
1113 	mtx_lock(&vnode_free_list_mtx);
1114 	/*
1115 	 * Lend our context to reclaim vnodes if they've exceeded the max.
1116 	 */
1117 	if (freevnodes > wantfreevnodes)
1118 		vnlru_free(1);
1119 	error = getnewvnode_wait(mp != NULL && (mp->mnt_kern_flag &
1120 	    MNTK_SUSPEND));
1121 #if 0	/* XXX Not all VFS_VGET/ffs_vget callers check returns. */
1122 	if (error != 0) {
1123 		mtx_unlock(&vnode_free_list_mtx);
1124 		return (error);
1125 	}
1126 #endif
1127 	atomic_add_long(&numvnodes, 1);
1128 	mtx_unlock(&vnode_free_list_mtx);
1129 alloc:
1130 	atomic_add_long(&vnodes_created, 1);
1131 	vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
1132 	/*
1133 	 * Setup locks.
1134 	 */
1135 	vp->v_vnlock = &vp->v_lock;
1136 	mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
1137 	/*
1138 	 * By default, don't allow shared locks unless filesystems
1139 	 * opt-in.
1140 	 */
1141 	lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE | LK_IS_VNODE);
1142 	/*
1143 	 * Initialize bufobj.
1144 	 */
1145 	bo = &vp->v_bufobj;
1146 	bo->__bo_vnode = vp;
1147 	rw_init(BO_LOCKPTR(bo), "bufobj interlock");
1148 	bo->bo_ops = &buf_ops_bio;
1149 	bo->bo_private = vp;
1150 	TAILQ_INIT(&bo->bo_clean.bv_hd);
1151 	TAILQ_INIT(&bo->bo_dirty.bv_hd);
1152 	/*
1153 	 * Initialize namecache.
1154 	 */
1155 	LIST_INIT(&vp->v_cache_src);
1156 	TAILQ_INIT(&vp->v_cache_dst);
1157 	/*
1158 	 * Finalize various vnode identity bits.
1159 	 */
1160 	vp->v_type = VNON;
1161 	vp->v_tag = tag;
1162 	vp->v_op = vops;
1163 	v_init_counters(vp);
1164 	vp->v_data = NULL;
1165 #ifdef MAC
1166 	mac_vnode_init(vp);
1167 	if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1168 		mac_vnode_associate_singlelabel(mp, vp);
1169 	else if (mp == NULL && vops != &dead_vnodeops)
1170 		printf("NULL mp in getnewvnode()\n");
1171 #endif
1172 	if (mp != NULL) {
1173 		bo->bo_bsize = mp->mnt_stat.f_iosize;
1174 		if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1175 			vp->v_vflag |= VV_NOKNOTE;
1176 	}
1177 	rangelock_init(&vp->v_rl);
1178 
1179 	/*
1180 	 * For the filesystems which do not use vfs_hash_insert(),
1181 	 * still initialize v_hash to have vfs_hash_index() useful.
1182 	 * E.g., nullfs uses vfs_hash_index() on the lower vnode for
1183 	 * its own hashing.
1184 	 */
1185 	vp->v_hash = (uintptr_t)vp >> vnsz2log;
1186 
1187 	*vpp = vp;
1188 	return (0);
1189 }
1190 
1191 /*
1192  * Delete from old mount point vnode list, if on one.
1193  */
1194 static void
1195 delmntque(struct vnode *vp)
1196 {
1197 	struct mount *mp;
1198 	int active;
1199 
1200 	mp = vp->v_mount;
1201 	if (mp == NULL)
1202 		return;
1203 	MNT_ILOCK(mp);
1204 	VI_LOCK(vp);
1205 	KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize,
1206 	    ("Active vnode list size %d > Vnode list size %d",
1207 	     mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize));
1208 	active = vp->v_iflag & VI_ACTIVE;
1209 	vp->v_iflag &= ~VI_ACTIVE;
1210 	if (active) {
1211 		mtx_lock(&vnode_free_list_mtx);
1212 		TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist);
1213 		mp->mnt_activevnodelistsize--;
1214 		mtx_unlock(&vnode_free_list_mtx);
1215 	}
1216 	vp->v_mount = NULL;
1217 	VI_UNLOCK(vp);
1218 	VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1219 		("bad mount point vnode list size"));
1220 	TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1221 	mp->mnt_nvnodelistsize--;
1222 	MNT_REL(mp);
1223 	MNT_IUNLOCK(mp);
1224 }
1225 
1226 static void
1227 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1228 {
1229 
1230 	vp->v_data = NULL;
1231 	vp->v_op = &dead_vnodeops;
1232 	vgone(vp);
1233 	vput(vp);
1234 }
1235 
1236 /*
1237  * Insert into list of vnodes for the new mount point, if available.
1238  */
1239 int
1240 insmntque1(struct vnode *vp, struct mount *mp,
1241 	void (*dtr)(struct vnode *, void *), void *dtr_arg)
1242 {
1243 
1244 	KASSERT(vp->v_mount == NULL,
1245 		("insmntque: vnode already on per mount vnode list"));
1246 	VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1247 	ASSERT_VOP_ELOCKED(vp, "insmntque: non-locked vp");
1248 
1249 	/*
1250 	 * We acquire the vnode interlock early to ensure that the
1251 	 * vnode cannot be recycled by another process releasing a
1252 	 * holdcnt on it before we get it on both the vnode list
1253 	 * and the active vnode list. The mount mutex protects only
1254 	 * manipulation of the vnode list and the vnode freelist
1255 	 * mutex protects only manipulation of the active vnode list.
1256 	 * Hence the need to hold the vnode interlock throughout.
1257 	 */
1258 	MNT_ILOCK(mp);
1259 	VI_LOCK(vp);
1260 	if (((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1261 	    ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1262 	    mp->mnt_nvnodelistsize == 0)) &&
1263 	    (vp->v_vflag & VV_FORCEINSMQ) == 0) {
1264 		VI_UNLOCK(vp);
1265 		MNT_IUNLOCK(mp);
1266 		if (dtr != NULL)
1267 			dtr(vp, dtr_arg);
1268 		return (EBUSY);
1269 	}
1270 	vp->v_mount = mp;
1271 	MNT_REF(mp);
1272 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1273 	VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1274 		("neg mount point vnode list size"));
1275 	mp->mnt_nvnodelistsize++;
1276 	KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
1277 	    ("Activating already active vnode"));
1278 	vp->v_iflag |= VI_ACTIVE;
1279 	mtx_lock(&vnode_free_list_mtx);
1280 	TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
1281 	mp->mnt_activevnodelistsize++;
1282 	mtx_unlock(&vnode_free_list_mtx);
1283 	VI_UNLOCK(vp);
1284 	MNT_IUNLOCK(mp);
1285 	return (0);
1286 }
1287 
1288 int
1289 insmntque(struct vnode *vp, struct mount *mp)
1290 {
1291 
1292 	return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1293 }
1294 
1295 /*
1296  * Flush out and invalidate all buffers associated with a bufobj
1297  * Called with the underlying object locked.
1298  */
1299 int
1300 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1301 {
1302 	int error;
1303 
1304 	BO_LOCK(bo);
1305 	if (flags & V_SAVE) {
1306 		error = bufobj_wwait(bo, slpflag, slptimeo);
1307 		if (error) {
1308 			BO_UNLOCK(bo);
1309 			return (error);
1310 		}
1311 		if (bo->bo_dirty.bv_cnt > 0) {
1312 			BO_UNLOCK(bo);
1313 			if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1314 				return (error);
1315 			/*
1316 			 * XXX We could save a lock/unlock if this was only
1317 			 * enabled under INVARIANTS
1318 			 */
1319 			BO_LOCK(bo);
1320 			if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1321 				panic("vinvalbuf: dirty bufs");
1322 		}
1323 	}
1324 	/*
1325 	 * If you alter this loop please notice that interlock is dropped and
1326 	 * reacquired in flushbuflist.  Special care is needed to ensure that
1327 	 * no race conditions occur from this.
1328 	 */
1329 	do {
1330 		error = flushbuflist(&bo->bo_clean,
1331 		    flags, bo, slpflag, slptimeo);
1332 		if (error == 0 && !(flags & V_CLEANONLY))
1333 			error = flushbuflist(&bo->bo_dirty,
1334 			    flags, bo, slpflag, slptimeo);
1335 		if (error != 0 && error != EAGAIN) {
1336 			BO_UNLOCK(bo);
1337 			return (error);
1338 		}
1339 	} while (error != 0);
1340 
1341 	/*
1342 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1343 	 * have write I/O in-progress but if there is a VM object then the
1344 	 * VM object can also have read-I/O in-progress.
1345 	 */
1346 	do {
1347 		bufobj_wwait(bo, 0, 0);
1348 		BO_UNLOCK(bo);
1349 		if (bo->bo_object != NULL) {
1350 			VM_OBJECT_WLOCK(bo->bo_object);
1351 			vm_object_pip_wait(bo->bo_object, "bovlbx");
1352 			VM_OBJECT_WUNLOCK(bo->bo_object);
1353 		}
1354 		BO_LOCK(bo);
1355 	} while (bo->bo_numoutput > 0);
1356 	BO_UNLOCK(bo);
1357 
1358 	/*
1359 	 * Destroy the copy in the VM cache, too.
1360 	 */
1361 	if (bo->bo_object != NULL &&
1362 	    (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) {
1363 		VM_OBJECT_WLOCK(bo->bo_object);
1364 		vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
1365 		    OBJPR_CLEANONLY : 0);
1366 		VM_OBJECT_WUNLOCK(bo->bo_object);
1367 	}
1368 
1369 #ifdef INVARIANTS
1370 	BO_LOCK(bo);
1371 	if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 &&
1372 	    (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1373 		panic("vinvalbuf: flush failed");
1374 	BO_UNLOCK(bo);
1375 #endif
1376 	return (0);
1377 }
1378 
1379 /*
1380  * Flush out and invalidate all buffers associated with a vnode.
1381  * Called with the underlying object locked.
1382  */
1383 int
1384 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1385 {
1386 
1387 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1388 	ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1389 	if (vp->v_object != NULL && vp->v_object->handle != vp)
1390 		return (0);
1391 	return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1392 }
1393 
1394 /*
1395  * Flush out buffers on the specified list.
1396  *
1397  */
1398 static int
1399 flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1400     int slptimeo)
1401 {
1402 	struct buf *bp, *nbp;
1403 	int retval, error;
1404 	daddr_t lblkno;
1405 	b_xflags_t xflags;
1406 
1407 	ASSERT_BO_WLOCKED(bo);
1408 
1409 	retval = 0;
1410 	TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1411 		if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1412 		    ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1413 			continue;
1414 		}
1415 		lblkno = 0;
1416 		xflags = 0;
1417 		if (nbp != NULL) {
1418 			lblkno = nbp->b_lblkno;
1419 			xflags = nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN);
1420 		}
1421 		retval = EAGAIN;
1422 		error = BUF_TIMELOCK(bp,
1423 		    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo),
1424 		    "flushbuf", slpflag, slptimeo);
1425 		if (error) {
1426 			BO_LOCK(bo);
1427 			return (error != ENOLCK ? error : EAGAIN);
1428 		}
1429 		KASSERT(bp->b_bufobj == bo,
1430 		    ("bp %p wrong b_bufobj %p should be %p",
1431 		    bp, bp->b_bufobj, bo));
1432 		if (bp->b_bufobj != bo) {	/* XXX: necessary ? */
1433 			BUF_UNLOCK(bp);
1434 			BO_LOCK(bo);
1435 			return (EAGAIN);
1436 		}
1437 		/*
1438 		 * XXX Since there are no node locks for NFS, I
1439 		 * believe there is a slight chance that a delayed
1440 		 * write will occur while sleeping just above, so
1441 		 * check for it.
1442 		 */
1443 		if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1444 		    (flags & V_SAVE)) {
1445 			bremfree(bp);
1446 			bp->b_flags |= B_ASYNC;
1447 			bwrite(bp);
1448 			BO_LOCK(bo);
1449 			return (EAGAIN);	/* XXX: why not loop ? */
1450 		}
1451 		bremfree(bp);
1452 		bp->b_flags |= (B_INVAL | B_RELBUF);
1453 		bp->b_flags &= ~B_ASYNC;
1454 		brelse(bp);
1455 		BO_LOCK(bo);
1456 		if (nbp != NULL &&
1457 		    (nbp->b_bufobj != bo ||
1458 		     nbp->b_lblkno != lblkno ||
1459 		     (nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1460 			break;			/* nbp invalid */
1461 	}
1462 	return (retval);
1463 }
1464 
1465 /*
1466  * Truncate a file's buffer and pages to a specified length.  This
1467  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1468  * sync activity.
1469  */
1470 int
1471 vtruncbuf(struct vnode *vp, struct ucred *cred, off_t length, int blksize)
1472 {
1473 	struct buf *bp, *nbp;
1474 	int anyfreed;
1475 	int trunclbn;
1476 	struct bufobj *bo;
1477 
1478 	CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1479 	    vp, cred, blksize, (uintmax_t)length);
1480 
1481 	/*
1482 	 * Round up to the *next* lbn.
1483 	 */
1484 	trunclbn = (length + blksize - 1) / blksize;
1485 
1486 	ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1487 restart:
1488 	bo = &vp->v_bufobj;
1489 	BO_LOCK(bo);
1490 	anyfreed = 1;
1491 	for (;anyfreed;) {
1492 		anyfreed = 0;
1493 		TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1494 			if (bp->b_lblkno < trunclbn)
1495 				continue;
1496 			if (BUF_LOCK(bp,
1497 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1498 			    BO_LOCKPTR(bo)) == ENOLCK)
1499 				goto restart;
1500 
1501 			bremfree(bp);
1502 			bp->b_flags |= (B_INVAL | B_RELBUF);
1503 			bp->b_flags &= ~B_ASYNC;
1504 			brelse(bp);
1505 			anyfreed = 1;
1506 
1507 			BO_LOCK(bo);
1508 			if (nbp != NULL &&
1509 			    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1510 			    (nbp->b_vp != vp) ||
1511 			    (nbp->b_flags & B_DELWRI))) {
1512 				BO_UNLOCK(bo);
1513 				goto restart;
1514 			}
1515 		}
1516 
1517 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1518 			if (bp->b_lblkno < trunclbn)
1519 				continue;
1520 			if (BUF_LOCK(bp,
1521 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1522 			    BO_LOCKPTR(bo)) == ENOLCK)
1523 				goto restart;
1524 			bremfree(bp);
1525 			bp->b_flags |= (B_INVAL | B_RELBUF);
1526 			bp->b_flags &= ~B_ASYNC;
1527 			brelse(bp);
1528 			anyfreed = 1;
1529 
1530 			BO_LOCK(bo);
1531 			if (nbp != NULL &&
1532 			    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1533 			    (nbp->b_vp != vp) ||
1534 			    (nbp->b_flags & B_DELWRI) == 0)) {
1535 				BO_UNLOCK(bo);
1536 				goto restart;
1537 			}
1538 		}
1539 	}
1540 
1541 	if (length > 0) {
1542 restartsync:
1543 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1544 			if (bp->b_lblkno > 0)
1545 				continue;
1546 			/*
1547 			 * Since we hold the vnode lock this should only
1548 			 * fail if we're racing with the buf daemon.
1549 			 */
1550 			if (BUF_LOCK(bp,
1551 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1552 			    BO_LOCKPTR(bo)) == ENOLCK) {
1553 				goto restart;
1554 			}
1555 			VNASSERT((bp->b_flags & B_DELWRI), vp,
1556 			    ("buf(%p) on dirty queue without DELWRI", bp));
1557 
1558 			bremfree(bp);
1559 			bawrite(bp);
1560 			BO_LOCK(bo);
1561 			goto restartsync;
1562 		}
1563 	}
1564 
1565 	bufobj_wwait(bo, 0, 0);
1566 	BO_UNLOCK(bo);
1567 	vnode_pager_setsize(vp, length);
1568 
1569 	return (0);
1570 }
1571 
1572 static void
1573 buf_vlist_remove(struct buf *bp)
1574 {
1575 	struct bufv *bv;
1576 
1577 	KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1578 	ASSERT_BO_WLOCKED(bp->b_bufobj);
1579 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1580 	    (BX_VNDIRTY|BX_VNCLEAN),
1581 	    ("buf_vlist_remove: Buf %p is on two lists", bp));
1582 	if (bp->b_xflags & BX_VNDIRTY)
1583 		bv = &bp->b_bufobj->bo_dirty;
1584 	else
1585 		bv = &bp->b_bufobj->bo_clean;
1586 	BUF_PCTRIE_REMOVE(&bv->bv_root, bp->b_lblkno);
1587 	TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1588 	bv->bv_cnt--;
1589 	bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1590 }
1591 
1592 /*
1593  * Add the buffer to the sorted clean or dirty block list.
1594  *
1595  * NOTE: xflags is passed as a constant, optimizing this inline function!
1596  */
1597 static void
1598 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1599 {
1600 	struct bufv *bv;
1601 	struct buf *n;
1602 	int error;
1603 
1604 	ASSERT_BO_WLOCKED(bo);
1605 	KASSERT((xflags & BX_VNDIRTY) == 0 || (bo->bo_flag & BO_DEAD) == 0,
1606 	    ("dead bo %p", bo));
1607 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1608 	    ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1609 	bp->b_xflags |= xflags;
1610 	if (xflags & BX_VNDIRTY)
1611 		bv = &bo->bo_dirty;
1612 	else
1613 		bv = &bo->bo_clean;
1614 
1615 	/*
1616 	 * Keep the list ordered.  Optimize empty list insertion.  Assume
1617 	 * we tend to grow at the tail so lookup_le should usually be cheaper
1618 	 * than _ge.
1619 	 */
1620 	if (bv->bv_cnt == 0 ||
1621 	    bp->b_lblkno > TAILQ_LAST(&bv->bv_hd, buflists)->b_lblkno)
1622 		TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1623 	else if ((n = BUF_PCTRIE_LOOKUP_LE(&bv->bv_root, bp->b_lblkno)) == NULL)
1624 		TAILQ_INSERT_HEAD(&bv->bv_hd, bp, b_bobufs);
1625 	else
1626 		TAILQ_INSERT_AFTER(&bv->bv_hd, n, bp, b_bobufs);
1627 	error = BUF_PCTRIE_INSERT(&bv->bv_root, bp);
1628 	if (error)
1629 		panic("buf_vlist_add:  Preallocated nodes insufficient.");
1630 	bv->bv_cnt++;
1631 }
1632 
1633 /*
1634  * Look up a buffer using the buffer tries.
1635  */
1636 struct buf *
1637 gbincore(struct bufobj *bo, daddr_t lblkno)
1638 {
1639 	struct buf *bp;
1640 
1641 	ASSERT_BO_LOCKED(bo);
1642 	bp = BUF_PCTRIE_LOOKUP(&bo->bo_clean.bv_root, lblkno);
1643 	if (bp != NULL)
1644 		return (bp);
1645 	return BUF_PCTRIE_LOOKUP(&bo->bo_dirty.bv_root, lblkno);
1646 }
1647 
1648 /*
1649  * Associate a buffer with a vnode.
1650  */
1651 void
1652 bgetvp(struct vnode *vp, struct buf *bp)
1653 {
1654 	struct bufobj *bo;
1655 
1656 	bo = &vp->v_bufobj;
1657 	ASSERT_BO_WLOCKED(bo);
1658 	VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1659 
1660 	CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1661 	VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1662 	    ("bgetvp: bp already attached! %p", bp));
1663 
1664 	vhold(vp);
1665 	bp->b_vp = vp;
1666 	bp->b_bufobj = bo;
1667 	/*
1668 	 * Insert onto list for new vnode.
1669 	 */
1670 	buf_vlist_add(bp, bo, BX_VNCLEAN);
1671 }
1672 
1673 /*
1674  * Disassociate a buffer from a vnode.
1675  */
1676 void
1677 brelvp(struct buf *bp)
1678 {
1679 	struct bufobj *bo;
1680 	struct vnode *vp;
1681 
1682 	CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1683 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1684 
1685 	/*
1686 	 * Delete from old vnode list, if on one.
1687 	 */
1688 	vp = bp->b_vp;		/* XXX */
1689 	bo = bp->b_bufobj;
1690 	BO_LOCK(bo);
1691 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1692 		buf_vlist_remove(bp);
1693 	else
1694 		panic("brelvp: Buffer %p not on queue.", bp);
1695 	if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1696 		bo->bo_flag &= ~BO_ONWORKLST;
1697 		mtx_lock(&sync_mtx);
1698 		LIST_REMOVE(bo, bo_synclist);
1699 		syncer_worklist_len--;
1700 		mtx_unlock(&sync_mtx);
1701 	}
1702 	bp->b_vp = NULL;
1703 	bp->b_bufobj = NULL;
1704 	BO_UNLOCK(bo);
1705 	vdrop(vp);
1706 }
1707 
1708 /*
1709  * Add an item to the syncer work queue.
1710  */
1711 static void
1712 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1713 {
1714 	int slot;
1715 
1716 	ASSERT_BO_WLOCKED(bo);
1717 
1718 	mtx_lock(&sync_mtx);
1719 	if (bo->bo_flag & BO_ONWORKLST)
1720 		LIST_REMOVE(bo, bo_synclist);
1721 	else {
1722 		bo->bo_flag |= BO_ONWORKLST;
1723 		syncer_worklist_len++;
1724 	}
1725 
1726 	if (delay > syncer_maxdelay - 2)
1727 		delay = syncer_maxdelay - 2;
1728 	slot = (syncer_delayno + delay) & syncer_mask;
1729 
1730 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist);
1731 	mtx_unlock(&sync_mtx);
1732 }
1733 
1734 static int
1735 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1736 {
1737 	int error, len;
1738 
1739 	mtx_lock(&sync_mtx);
1740 	len = syncer_worklist_len - sync_vnode_count;
1741 	mtx_unlock(&sync_mtx);
1742 	error = SYSCTL_OUT(req, &len, sizeof(len));
1743 	return (error);
1744 }
1745 
1746 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1747     sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1748 
1749 static struct proc *updateproc;
1750 static void sched_sync(void);
1751 static struct kproc_desc up_kp = {
1752 	"syncer",
1753 	sched_sync,
1754 	&updateproc
1755 };
1756 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1757 
1758 static int
1759 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1760 {
1761 	struct vnode *vp;
1762 	struct mount *mp;
1763 
1764 	*bo = LIST_FIRST(slp);
1765 	if (*bo == NULL)
1766 		return (0);
1767 	vp = (*bo)->__bo_vnode;	/* XXX */
1768 	if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1769 		return (1);
1770 	/*
1771 	 * We use vhold in case the vnode does not
1772 	 * successfully sync.  vhold prevents the vnode from
1773 	 * going away when we unlock the sync_mtx so that
1774 	 * we can acquire the vnode interlock.
1775 	 */
1776 	vholdl(vp);
1777 	mtx_unlock(&sync_mtx);
1778 	VI_UNLOCK(vp);
1779 	if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1780 		vdrop(vp);
1781 		mtx_lock(&sync_mtx);
1782 		return (*bo == LIST_FIRST(slp));
1783 	}
1784 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1785 	(void) VOP_FSYNC(vp, MNT_LAZY, td);
1786 	VOP_UNLOCK(vp, 0);
1787 	vn_finished_write(mp);
1788 	BO_LOCK(*bo);
1789 	if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1790 		/*
1791 		 * Put us back on the worklist.  The worklist
1792 		 * routine will remove us from our current
1793 		 * position and then add us back in at a later
1794 		 * position.
1795 		 */
1796 		vn_syncer_add_to_worklist(*bo, syncdelay);
1797 	}
1798 	BO_UNLOCK(*bo);
1799 	vdrop(vp);
1800 	mtx_lock(&sync_mtx);
1801 	return (0);
1802 }
1803 
1804 static int first_printf = 1;
1805 
1806 /*
1807  * System filesystem synchronizer daemon.
1808  */
1809 static void
1810 sched_sync(void)
1811 {
1812 	struct synclist *next, *slp;
1813 	struct bufobj *bo;
1814 	long starttime;
1815 	struct thread *td = curthread;
1816 	int last_work_seen;
1817 	int net_worklist_len;
1818 	int syncer_final_iter;
1819 	int error;
1820 
1821 	last_work_seen = 0;
1822 	syncer_final_iter = 0;
1823 	syncer_state = SYNCER_RUNNING;
1824 	starttime = time_uptime;
1825 	td->td_pflags |= TDP_NORUNNINGBUF;
1826 
1827 	EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1828 	    SHUTDOWN_PRI_LAST);
1829 
1830 	mtx_lock(&sync_mtx);
1831 	for (;;) {
1832 		if (syncer_state == SYNCER_FINAL_DELAY &&
1833 		    syncer_final_iter == 0) {
1834 			mtx_unlock(&sync_mtx);
1835 			kproc_suspend_check(td->td_proc);
1836 			mtx_lock(&sync_mtx);
1837 		}
1838 		net_worklist_len = syncer_worklist_len - sync_vnode_count;
1839 		if (syncer_state != SYNCER_RUNNING &&
1840 		    starttime != time_uptime) {
1841 			if (first_printf) {
1842 				printf("\nSyncing disks, vnodes remaining...");
1843 				first_printf = 0;
1844 			}
1845 			printf("%d ", net_worklist_len);
1846 		}
1847 		starttime = time_uptime;
1848 
1849 		/*
1850 		 * Push files whose dirty time has expired.  Be careful
1851 		 * of interrupt race on slp queue.
1852 		 *
1853 		 * Skip over empty worklist slots when shutting down.
1854 		 */
1855 		do {
1856 			slp = &syncer_workitem_pending[syncer_delayno];
1857 			syncer_delayno += 1;
1858 			if (syncer_delayno == syncer_maxdelay)
1859 				syncer_delayno = 0;
1860 			next = &syncer_workitem_pending[syncer_delayno];
1861 			/*
1862 			 * If the worklist has wrapped since the
1863 			 * it was emptied of all but syncer vnodes,
1864 			 * switch to the FINAL_DELAY state and run
1865 			 * for one more second.
1866 			 */
1867 			if (syncer_state == SYNCER_SHUTTING_DOWN &&
1868 			    net_worklist_len == 0 &&
1869 			    last_work_seen == syncer_delayno) {
1870 				syncer_state = SYNCER_FINAL_DELAY;
1871 				syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1872 			}
1873 		} while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1874 		    syncer_worklist_len > 0);
1875 
1876 		/*
1877 		 * Keep track of the last time there was anything
1878 		 * on the worklist other than syncer vnodes.
1879 		 * Return to the SHUTTING_DOWN state if any
1880 		 * new work appears.
1881 		 */
1882 		if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1883 			last_work_seen = syncer_delayno;
1884 		if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1885 			syncer_state = SYNCER_SHUTTING_DOWN;
1886 		while (!LIST_EMPTY(slp)) {
1887 			error = sync_vnode(slp, &bo, td);
1888 			if (error == 1) {
1889 				LIST_REMOVE(bo, bo_synclist);
1890 				LIST_INSERT_HEAD(next, bo, bo_synclist);
1891 				continue;
1892 			}
1893 
1894 			if (first_printf == 0) {
1895 				/*
1896 				 * Drop the sync mutex, because some watchdog
1897 				 * drivers need to sleep while patting
1898 				 */
1899 				mtx_unlock(&sync_mtx);
1900 				wdog_kern_pat(WD_LASTVAL);
1901 				mtx_lock(&sync_mtx);
1902 			}
1903 
1904 		}
1905 		if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1906 			syncer_final_iter--;
1907 		/*
1908 		 * The variable rushjob allows the kernel to speed up the
1909 		 * processing of the filesystem syncer process. A rushjob
1910 		 * value of N tells the filesystem syncer to process the next
1911 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1912 		 * is used by the soft update code to speed up the filesystem
1913 		 * syncer process when the incore state is getting so far
1914 		 * ahead of the disk that the kernel memory pool is being
1915 		 * threatened with exhaustion.
1916 		 */
1917 		if (rushjob > 0) {
1918 			rushjob -= 1;
1919 			continue;
1920 		}
1921 		/*
1922 		 * Just sleep for a short period of time between
1923 		 * iterations when shutting down to allow some I/O
1924 		 * to happen.
1925 		 *
1926 		 * If it has taken us less than a second to process the
1927 		 * current work, then wait. Otherwise start right over
1928 		 * again. We can still lose time if any single round
1929 		 * takes more than two seconds, but it does not really
1930 		 * matter as we are just trying to generally pace the
1931 		 * filesystem activity.
1932 		 */
1933 		if (syncer_state != SYNCER_RUNNING ||
1934 		    time_uptime == starttime) {
1935 			thread_lock(td);
1936 			sched_prio(td, PPAUSE);
1937 			thread_unlock(td);
1938 		}
1939 		if (syncer_state != SYNCER_RUNNING)
1940 			cv_timedwait(&sync_wakeup, &sync_mtx,
1941 			    hz / SYNCER_SHUTDOWN_SPEEDUP);
1942 		else if (time_uptime == starttime)
1943 			cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1944 	}
1945 }
1946 
1947 /*
1948  * Request the syncer daemon to speed up its work.
1949  * We never push it to speed up more than half of its
1950  * normal turn time, otherwise it could take over the cpu.
1951  */
1952 int
1953 speedup_syncer(void)
1954 {
1955 	int ret = 0;
1956 
1957 	mtx_lock(&sync_mtx);
1958 	if (rushjob < syncdelay / 2) {
1959 		rushjob += 1;
1960 		stat_rush_requests += 1;
1961 		ret = 1;
1962 	}
1963 	mtx_unlock(&sync_mtx);
1964 	cv_broadcast(&sync_wakeup);
1965 	return (ret);
1966 }
1967 
1968 /*
1969  * Tell the syncer to speed up its work and run though its work
1970  * list several times, then tell it to shut down.
1971  */
1972 static void
1973 syncer_shutdown(void *arg, int howto)
1974 {
1975 
1976 	if (howto & RB_NOSYNC)
1977 		return;
1978 	mtx_lock(&sync_mtx);
1979 	syncer_state = SYNCER_SHUTTING_DOWN;
1980 	rushjob = 0;
1981 	mtx_unlock(&sync_mtx);
1982 	cv_broadcast(&sync_wakeup);
1983 	kproc_shutdown(arg, howto);
1984 }
1985 
1986 void
1987 syncer_suspend(void)
1988 {
1989 
1990 	syncer_shutdown(updateproc, 0);
1991 }
1992 
1993 void
1994 syncer_resume(void)
1995 {
1996 
1997 	mtx_lock(&sync_mtx);
1998 	first_printf = 1;
1999 	syncer_state = SYNCER_RUNNING;
2000 	mtx_unlock(&sync_mtx);
2001 	cv_broadcast(&sync_wakeup);
2002 	kproc_resume(updateproc);
2003 }
2004 
2005 /*
2006  * Reassign a buffer from one vnode to another.
2007  * Used to assign file specific control information
2008  * (indirect blocks) to the vnode to which they belong.
2009  */
2010 void
2011 reassignbuf(struct buf *bp)
2012 {
2013 	struct vnode *vp;
2014 	struct bufobj *bo;
2015 	int delay;
2016 #ifdef INVARIANTS
2017 	struct bufv *bv;
2018 #endif
2019 
2020 	vp = bp->b_vp;
2021 	bo = bp->b_bufobj;
2022 	++reassignbufcalls;
2023 
2024 	CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
2025 	    bp, bp->b_vp, bp->b_flags);
2026 	/*
2027 	 * B_PAGING flagged buffers cannot be reassigned because their vp
2028 	 * is not fully linked in.
2029 	 */
2030 	if (bp->b_flags & B_PAGING)
2031 		panic("cannot reassign paging buffer");
2032 
2033 	/*
2034 	 * Delete from old vnode list, if on one.
2035 	 */
2036 	BO_LOCK(bo);
2037 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
2038 		buf_vlist_remove(bp);
2039 	else
2040 		panic("reassignbuf: Buffer %p not on queue.", bp);
2041 	/*
2042 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
2043 	 * of clean buffers.
2044 	 */
2045 	if (bp->b_flags & B_DELWRI) {
2046 		if ((bo->bo_flag & BO_ONWORKLST) == 0) {
2047 			switch (vp->v_type) {
2048 			case VDIR:
2049 				delay = dirdelay;
2050 				break;
2051 			case VCHR:
2052 				delay = metadelay;
2053 				break;
2054 			default:
2055 				delay = filedelay;
2056 			}
2057 			vn_syncer_add_to_worklist(bo, delay);
2058 		}
2059 		buf_vlist_add(bp, bo, BX_VNDIRTY);
2060 	} else {
2061 		buf_vlist_add(bp, bo, BX_VNCLEAN);
2062 
2063 		if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
2064 			mtx_lock(&sync_mtx);
2065 			LIST_REMOVE(bo, bo_synclist);
2066 			syncer_worklist_len--;
2067 			mtx_unlock(&sync_mtx);
2068 			bo->bo_flag &= ~BO_ONWORKLST;
2069 		}
2070 	}
2071 #ifdef INVARIANTS
2072 	bv = &bo->bo_clean;
2073 	bp = TAILQ_FIRST(&bv->bv_hd);
2074 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2075 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2076 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
2077 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2078 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2079 	bv = &bo->bo_dirty;
2080 	bp = TAILQ_FIRST(&bv->bv_hd);
2081 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2082 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2083 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
2084 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2085 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2086 #endif
2087 	BO_UNLOCK(bo);
2088 }
2089 
2090 /*
2091  * A temporary hack until refcount_* APIs are sorted out.
2092  */
2093 static __inline int
2094 vfs_refcount_acquire_if_not_zero(volatile u_int *count)
2095 {
2096 	u_int old;
2097 
2098 	for (;;) {
2099 		old = *count;
2100 		if (old == 0)
2101 			return (0);
2102 		if (atomic_cmpset_int(count, old, old + 1))
2103 			return (1);
2104 	}
2105 }
2106 
2107 static __inline int
2108 vfs_refcount_release_if_not_last(volatile u_int *count)
2109 {
2110 	u_int old;
2111 
2112 	for (;;) {
2113 		old = *count;
2114 		if (old == 1)
2115 			return (0);
2116 		if (atomic_cmpset_int(count, old, old - 1))
2117 			return (1);
2118 	}
2119 }
2120 
2121 static void
2122 v_init_counters(struct vnode *vp)
2123 {
2124 
2125 	VNASSERT(vp->v_type == VNON && vp->v_data == NULL && vp->v_iflag == 0,
2126 	    vp, ("%s called for an initialized vnode", __FUNCTION__));
2127 	ASSERT_VI_UNLOCKED(vp, __FUNCTION__);
2128 
2129 	refcount_init(&vp->v_holdcnt, 1);
2130 	refcount_init(&vp->v_usecount, 1);
2131 }
2132 
2133 /*
2134  * Increment the use and hold counts on the vnode, taking care to reference
2135  * the driver's usecount if this is a chardev.  The _vhold() will remove
2136  * the vnode from the free list if it is presently free.
2137  */
2138 static void
2139 v_incr_usecount(struct vnode *vp)
2140 {
2141 
2142 	ASSERT_VI_UNLOCKED(vp, __func__);
2143 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2144 
2145 	if (vp->v_type == VCHR) {
2146 		VI_LOCK(vp);
2147 		_vhold(vp, true);
2148 		if (vp->v_iflag & VI_OWEINACT) {
2149 			VNASSERT(vp->v_usecount == 0, vp,
2150 			    ("vnode with usecount and VI_OWEINACT set"));
2151 			vp->v_iflag &= ~VI_OWEINACT;
2152 		}
2153 		refcount_acquire(&vp->v_usecount);
2154 		v_incr_devcount(vp);
2155 		VI_UNLOCK(vp);
2156 		return;
2157 	}
2158 
2159 	_vhold(vp, false);
2160 	if (vfs_refcount_acquire_if_not_zero(&vp->v_usecount)) {
2161 		VNASSERT((vp->v_iflag & VI_OWEINACT) == 0, vp,
2162 		    ("vnode with usecount and VI_OWEINACT set"));
2163 	} else {
2164 		VI_LOCK(vp);
2165 		if (vp->v_iflag & VI_OWEINACT)
2166 			vp->v_iflag &= ~VI_OWEINACT;
2167 		refcount_acquire(&vp->v_usecount);
2168 		VI_UNLOCK(vp);
2169 	}
2170 }
2171 
2172 /*
2173  * Increment si_usecount of the associated device, if any.
2174  */
2175 static void
2176 v_incr_devcount(struct vnode *vp)
2177 {
2178 
2179 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2180 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2181 		dev_lock();
2182 		vp->v_rdev->si_usecount++;
2183 		dev_unlock();
2184 	}
2185 }
2186 
2187 /*
2188  * Decrement si_usecount of the associated device, if any.
2189  */
2190 static void
2191 v_decr_devcount(struct vnode *vp)
2192 {
2193 
2194 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2195 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2196 		dev_lock();
2197 		vp->v_rdev->si_usecount--;
2198 		dev_unlock();
2199 	}
2200 }
2201 
2202 /*
2203  * Grab a particular vnode from the free list, increment its
2204  * reference count and lock it.  VI_DOOMED is set if the vnode
2205  * is being destroyed.  Only callers who specify LK_RETRY will
2206  * see doomed vnodes.  If inactive processing was delayed in
2207  * vput try to do it here.
2208  *
2209  * Notes on lockless counter manipulation:
2210  * _vhold, vputx and other routines make various decisions based
2211  * on either holdcnt or usecount being 0. As long as either contuner
2212  * is not transitioning 0->1 nor 1->0, the manipulation can be done
2213  * with atomic operations. Otherwise the interlock is taken.
2214  */
2215 int
2216 vget(struct vnode *vp, int flags, struct thread *td)
2217 {
2218 	int error, oweinact;
2219 
2220 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2221 	    ("vget: invalid lock operation"));
2222 
2223 	if ((flags & LK_INTERLOCK) != 0)
2224 		ASSERT_VI_LOCKED(vp, __func__);
2225 	else
2226 		ASSERT_VI_UNLOCKED(vp, __func__);
2227 	if ((flags & LK_VNHELD) != 0)
2228 		VNASSERT((vp->v_holdcnt > 0), vp,
2229 		    ("vget: LK_VNHELD passed but vnode not held"));
2230 
2231 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2232 
2233 	if ((flags & LK_VNHELD) == 0)
2234 		_vhold(vp, (flags & LK_INTERLOCK) != 0);
2235 
2236 	if ((error = vn_lock(vp, flags)) != 0) {
2237 		vdrop(vp);
2238 		CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2239 		    vp);
2240 		return (error);
2241 	}
2242 	if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2243 		panic("vget: vn_lock failed to return ENOENT\n");
2244 	/*
2245 	 * We don't guarantee that any particular close will
2246 	 * trigger inactive processing so just make a best effort
2247 	 * here at preventing a reference to a removed file.  If
2248 	 * we don't succeed no harm is done.
2249 	 *
2250 	 * Upgrade our holdcnt to a usecount.
2251 	 */
2252 	if (vp->v_type != VCHR &&
2253 	    vfs_refcount_acquire_if_not_zero(&vp->v_usecount)) {
2254 		VNASSERT((vp->v_iflag & VI_OWEINACT) == 0, vp,
2255 		    ("vnode with usecount and VI_OWEINACT set"));
2256 	} else {
2257 		VI_LOCK(vp);
2258 		if ((vp->v_iflag & VI_OWEINACT) == 0) {
2259 			oweinact = 0;
2260 		} else {
2261 			oweinact = 1;
2262 			vp->v_iflag &= ~VI_OWEINACT;
2263 		}
2264 		refcount_acquire(&vp->v_usecount);
2265 		v_incr_devcount(vp);
2266 		if (oweinact && VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2267 		    (flags & LK_NOWAIT) == 0)
2268 			vinactive(vp, td);
2269 		VI_UNLOCK(vp);
2270 	}
2271 	return (0);
2272 }
2273 
2274 /*
2275  * Increase the reference count of a vnode.
2276  */
2277 void
2278 vref(struct vnode *vp)
2279 {
2280 
2281 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2282 	v_incr_usecount(vp);
2283 }
2284 
2285 /*
2286  * Return reference count of a vnode.
2287  *
2288  * The results of this call are only guaranteed when some mechanism is used to
2289  * stop other processes from gaining references to the vnode.  This may be the
2290  * case if the caller holds the only reference.  This is also useful when stale
2291  * data is acceptable as race conditions may be accounted for by some other
2292  * means.
2293  */
2294 int
2295 vrefcnt(struct vnode *vp)
2296 {
2297 
2298 	return (vp->v_usecount);
2299 }
2300 
2301 #define	VPUTX_VRELE	1
2302 #define	VPUTX_VPUT	2
2303 #define	VPUTX_VUNREF	3
2304 
2305 /*
2306  * Decrement the use and hold counts for a vnode.
2307  *
2308  * See an explanation near vget() as to why atomic operation is safe.
2309  */
2310 static void
2311 vputx(struct vnode *vp, int func)
2312 {
2313 	int error;
2314 
2315 	KASSERT(vp != NULL, ("vputx: null vp"));
2316 	if (func == VPUTX_VUNREF)
2317 		ASSERT_VOP_LOCKED(vp, "vunref");
2318 	else if (func == VPUTX_VPUT)
2319 		ASSERT_VOP_LOCKED(vp, "vput");
2320 	else
2321 		KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2322 	ASSERT_VI_UNLOCKED(vp, __func__);
2323 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2324 
2325 	if (vp->v_type != VCHR &&
2326 	    vfs_refcount_release_if_not_last(&vp->v_usecount)) {
2327 		if (func == VPUTX_VPUT)
2328 			VOP_UNLOCK(vp, 0);
2329 		vdrop(vp);
2330 		return;
2331 	}
2332 
2333 	VI_LOCK(vp);
2334 
2335 	/*
2336 	 * We want to hold the vnode until the inactive finishes to
2337 	 * prevent vgone() races.  We drop the use count here and the
2338 	 * hold count below when we're done.
2339 	 */
2340 	if (!refcount_release(&vp->v_usecount) ||
2341 	    (vp->v_iflag & VI_DOINGINACT)) {
2342 		if (func == VPUTX_VPUT)
2343 			VOP_UNLOCK(vp, 0);
2344 		v_decr_devcount(vp);
2345 		vdropl(vp);
2346 		return;
2347 	}
2348 
2349 	v_decr_devcount(vp);
2350 
2351 	error = 0;
2352 
2353 	if (vp->v_usecount != 0) {
2354 		vprint("vputx: usecount not zero", vp);
2355 		panic("vputx: usecount not zero");
2356 	}
2357 
2358 	CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2359 
2360 	/*
2361 	 * We must call VOP_INACTIVE with the node locked. Mark
2362 	 * as VI_DOINGINACT to avoid recursion.
2363 	 */
2364 	vp->v_iflag |= VI_OWEINACT;
2365 	switch (func) {
2366 	case VPUTX_VRELE:
2367 		error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2368 		VI_LOCK(vp);
2369 		break;
2370 	case VPUTX_VPUT:
2371 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2372 			error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
2373 			    LK_NOWAIT);
2374 			VI_LOCK(vp);
2375 		}
2376 		break;
2377 	case VPUTX_VUNREF:
2378 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2379 			error = VOP_LOCK(vp, LK_TRYUPGRADE | LK_INTERLOCK);
2380 			VI_LOCK(vp);
2381 		}
2382 		break;
2383 	}
2384 	VNASSERT(vp->v_usecount == 0 || (vp->v_iflag & VI_OWEINACT) == 0, vp,
2385 	    ("vnode with usecount and VI_OWEINACT set"));
2386 	if (error == 0) {
2387 		if (vp->v_iflag & VI_OWEINACT)
2388 			vinactive(vp, curthread);
2389 		if (func != VPUTX_VUNREF)
2390 			VOP_UNLOCK(vp, 0);
2391 	}
2392 	vdropl(vp);
2393 }
2394 
2395 /*
2396  * Vnode put/release.
2397  * If count drops to zero, call inactive routine and return to freelist.
2398  */
2399 void
2400 vrele(struct vnode *vp)
2401 {
2402 
2403 	vputx(vp, VPUTX_VRELE);
2404 }
2405 
2406 /*
2407  * Release an already locked vnode.  This give the same effects as
2408  * unlock+vrele(), but takes less time and avoids releasing and
2409  * re-aquiring the lock (as vrele() acquires the lock internally.)
2410  */
2411 void
2412 vput(struct vnode *vp)
2413 {
2414 
2415 	vputx(vp, VPUTX_VPUT);
2416 }
2417 
2418 /*
2419  * Release an exclusively locked vnode. Do not unlock the vnode lock.
2420  */
2421 void
2422 vunref(struct vnode *vp)
2423 {
2424 
2425 	vputx(vp, VPUTX_VUNREF);
2426 }
2427 
2428 /*
2429  * Increase the hold count and activate if this is the first reference.
2430  */
2431 void
2432 _vhold(struct vnode *vp, bool locked)
2433 {
2434 	struct mount *mp;
2435 
2436 	if (locked)
2437 		ASSERT_VI_LOCKED(vp, __func__);
2438 	else
2439 		ASSERT_VI_UNLOCKED(vp, __func__);
2440 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2441 	if (!locked && vfs_refcount_acquire_if_not_zero(&vp->v_holdcnt)) {
2442 		VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2443 		    ("_vhold: vnode with holdcnt is free"));
2444 		return;
2445 	}
2446 
2447 	if (!locked)
2448 		VI_LOCK(vp);
2449 	if ((vp->v_iflag & VI_FREE) == 0) {
2450 		refcount_acquire(&vp->v_holdcnt);
2451 		if (!locked)
2452 			VI_UNLOCK(vp);
2453 		return;
2454 	}
2455 	VNASSERT(vp->v_holdcnt == 0, vp,
2456 	    ("%s: wrong hold count", __func__));
2457 	VNASSERT(vp->v_op != NULL, vp,
2458 	    ("%s: vnode already reclaimed.", __func__));
2459 	/*
2460 	 * Remove a vnode from the free list, mark it as in use,
2461 	 * and put it on the active list.
2462 	 */
2463 	mtx_lock(&vnode_free_list_mtx);
2464 	TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
2465 	freevnodes--;
2466 	vp->v_iflag &= ~(VI_FREE|VI_AGE);
2467 	KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
2468 	    ("Activating already active vnode"));
2469 	vp->v_iflag |= VI_ACTIVE;
2470 	mp = vp->v_mount;
2471 	TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
2472 	mp->mnt_activevnodelistsize++;
2473 	mtx_unlock(&vnode_free_list_mtx);
2474 	refcount_acquire(&vp->v_holdcnt);
2475 	if (!locked)
2476 		VI_UNLOCK(vp);
2477 }
2478 
2479 /*
2480  * Drop the hold count of the vnode.  If this is the last reference to
2481  * the vnode we place it on the free list unless it has been vgone'd
2482  * (marked VI_DOOMED) in which case we will free it.
2483  */
2484 void
2485 _vdrop(struct vnode *vp, bool locked)
2486 {
2487 	struct bufobj *bo;
2488 	struct mount *mp;
2489 	int active;
2490 
2491 	if (locked)
2492 		ASSERT_VI_LOCKED(vp, __func__);
2493 	else
2494 		ASSERT_VI_UNLOCKED(vp, __func__);
2495 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2496 	if ((int)vp->v_holdcnt <= 0)
2497 		panic("vdrop: holdcnt %d", vp->v_holdcnt);
2498 	if (vfs_refcount_release_if_not_last(&vp->v_holdcnt)) {
2499 		if (locked)
2500 			VI_UNLOCK(vp);
2501 		return;
2502 	}
2503 
2504 	if (!locked)
2505 		VI_LOCK(vp);
2506 	if (refcount_release(&vp->v_holdcnt) == 0) {
2507 		VI_UNLOCK(vp);
2508 		return;
2509 	}
2510 	if ((vp->v_iflag & VI_DOOMED) == 0) {
2511 		/*
2512 		 * Mark a vnode as free: remove it from its active list
2513 		 * and put it up for recycling on the freelist.
2514 		 */
2515 		VNASSERT(vp->v_op != NULL, vp,
2516 		    ("vdropl: vnode already reclaimed."));
2517 		VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2518 		    ("vnode already free"));
2519 		VNASSERT(vp->v_holdcnt == 0, vp,
2520 		    ("vdropl: freeing when we shouldn't"));
2521 		active = vp->v_iflag & VI_ACTIVE;
2522 		if ((vp->v_iflag & VI_OWEINACT) == 0) {
2523 			vp->v_iflag &= ~VI_ACTIVE;
2524 			mp = vp->v_mount;
2525 			mtx_lock(&vnode_free_list_mtx);
2526 			if (active) {
2527 				TAILQ_REMOVE(&mp->mnt_activevnodelist, vp,
2528 				    v_actfreelist);
2529 				mp->mnt_activevnodelistsize--;
2530 			}
2531 			if (vp->v_iflag & VI_AGE) {
2532 				TAILQ_INSERT_HEAD(&vnode_free_list, vp,
2533 				    v_actfreelist);
2534 			} else {
2535 				TAILQ_INSERT_TAIL(&vnode_free_list, vp,
2536 				    v_actfreelist);
2537 			}
2538 			freevnodes++;
2539 			vp->v_iflag &= ~VI_AGE;
2540 			vp->v_iflag |= VI_FREE;
2541 			mtx_unlock(&vnode_free_list_mtx);
2542 		} else {
2543 			atomic_add_long(&free_owe_inact, 1);
2544 		}
2545 		VI_UNLOCK(vp);
2546 		return;
2547 	}
2548 	/*
2549 	 * The vnode has been marked for destruction, so free it.
2550 	 */
2551 	CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp);
2552 	atomic_subtract_long(&numvnodes, 1);
2553 	bo = &vp->v_bufobj;
2554 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2555 	    ("cleaned vnode still on the free list."));
2556 	VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
2557 	VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
2558 	VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
2559 	VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
2560 	VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
2561 	VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
2562 	VNASSERT(pctrie_is_empty(&bo->bo_clean.bv_root), vp,
2563 	    ("clean blk trie not empty"));
2564 	VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
2565 	VNASSERT(pctrie_is_empty(&bo->bo_dirty.bv_root), vp,
2566 	    ("dirty blk trie not empty"));
2567 	VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
2568 	VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
2569 	VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
2570 	VI_UNLOCK(vp);
2571 #ifdef MAC
2572 	mac_vnode_destroy(vp);
2573 #endif
2574 	if (vp->v_pollinfo != NULL)
2575 		destroy_vpollinfo(vp->v_pollinfo);
2576 #ifdef INVARIANTS
2577 	/* XXX Elsewhere we detect an already freed vnode via NULL v_op. */
2578 	vp->v_op = NULL;
2579 #endif
2580 	rangelock_destroy(&vp->v_rl);
2581 	lockdestroy(vp->v_vnlock);
2582 	mtx_destroy(&vp->v_interlock);
2583 	rw_destroy(BO_LOCKPTR(bo));
2584 	uma_zfree(vnode_zone, vp);
2585 }
2586 
2587 /*
2588  * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2589  * flags.  DOINGINACT prevents us from recursing in calls to vinactive.
2590  * OWEINACT tracks whether a vnode missed a call to inactive due to a
2591  * failed lock upgrade.
2592  */
2593 void
2594 vinactive(struct vnode *vp, struct thread *td)
2595 {
2596 	struct vm_object *obj;
2597 
2598 	ASSERT_VOP_ELOCKED(vp, "vinactive");
2599 	ASSERT_VI_LOCKED(vp, "vinactive");
2600 	VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2601 	    ("vinactive: recursed on VI_DOINGINACT"));
2602 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2603 	vp->v_iflag |= VI_DOINGINACT;
2604 	vp->v_iflag &= ~VI_OWEINACT;
2605 	VI_UNLOCK(vp);
2606 	/*
2607 	 * Before moving off the active list, we must be sure that any
2608 	 * modified pages are on the vnode's dirty list since these will
2609 	 * no longer be checked once the vnode is on the inactive list.
2610 	 * Because the vnode vm object keeps a hold reference on the vnode
2611 	 * if there is at least one resident non-cached page, the vnode
2612 	 * cannot leave the active list without the page cleanup done.
2613 	 */
2614 	obj = vp->v_object;
2615 	if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
2616 		VM_OBJECT_WLOCK(obj);
2617 		vm_object_page_clean(obj, 0, 0, OBJPC_NOSYNC);
2618 		VM_OBJECT_WUNLOCK(obj);
2619 	}
2620 	VOP_INACTIVE(vp, td);
2621 	VI_LOCK(vp);
2622 	VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2623 	    ("vinactive: lost VI_DOINGINACT"));
2624 	vp->v_iflag &= ~VI_DOINGINACT;
2625 }
2626 
2627 /*
2628  * Remove any vnodes in the vnode table belonging to mount point mp.
2629  *
2630  * If FORCECLOSE is not specified, there should not be any active ones,
2631  * return error if any are found (nb: this is a user error, not a
2632  * system error). If FORCECLOSE is specified, detach any active vnodes
2633  * that are found.
2634  *
2635  * If WRITECLOSE is set, only flush out regular file vnodes open for
2636  * writing.
2637  *
2638  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2639  *
2640  * `rootrefs' specifies the base reference count for the root vnode
2641  * of this filesystem. The root vnode is considered busy if its
2642  * v_usecount exceeds this value. On a successful return, vflush(, td)
2643  * will call vrele() on the root vnode exactly rootrefs times.
2644  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2645  * be zero.
2646  */
2647 #ifdef DIAGNOSTIC
2648 static int busyprt = 0;		/* print out busy vnodes */
2649 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
2650 #endif
2651 
2652 int
2653 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
2654 {
2655 	struct vnode *vp, *mvp, *rootvp = NULL;
2656 	struct vattr vattr;
2657 	int busy = 0, error;
2658 
2659 	CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2660 	    rootrefs, flags);
2661 	if (rootrefs > 0) {
2662 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2663 		    ("vflush: bad args"));
2664 		/*
2665 		 * Get the filesystem root vnode. We can vput() it
2666 		 * immediately, since with rootrefs > 0, it won't go away.
2667 		 */
2668 		if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2669 			CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2670 			    __func__, error);
2671 			return (error);
2672 		}
2673 		vput(rootvp);
2674 	}
2675 loop:
2676 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
2677 		vholdl(vp);
2678 		error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2679 		if (error) {
2680 			vdrop(vp);
2681 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2682 			goto loop;
2683 		}
2684 		/*
2685 		 * Skip over a vnodes marked VV_SYSTEM.
2686 		 */
2687 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2688 			VOP_UNLOCK(vp, 0);
2689 			vdrop(vp);
2690 			continue;
2691 		}
2692 		/*
2693 		 * If WRITECLOSE is set, flush out unlinked but still open
2694 		 * files (even if open only for reading) and regular file
2695 		 * vnodes open for writing.
2696 		 */
2697 		if (flags & WRITECLOSE) {
2698 			if (vp->v_object != NULL) {
2699 				VM_OBJECT_WLOCK(vp->v_object);
2700 				vm_object_page_clean(vp->v_object, 0, 0, 0);
2701 				VM_OBJECT_WUNLOCK(vp->v_object);
2702 			}
2703 			error = VOP_FSYNC(vp, MNT_WAIT, td);
2704 			if (error != 0) {
2705 				VOP_UNLOCK(vp, 0);
2706 				vdrop(vp);
2707 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2708 				return (error);
2709 			}
2710 			error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2711 			VI_LOCK(vp);
2712 
2713 			if ((vp->v_type == VNON ||
2714 			    (error == 0 && vattr.va_nlink > 0)) &&
2715 			    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2716 				VOP_UNLOCK(vp, 0);
2717 				vdropl(vp);
2718 				continue;
2719 			}
2720 		} else
2721 			VI_LOCK(vp);
2722 		/*
2723 		 * With v_usecount == 0, all we need to do is clear out the
2724 		 * vnode data structures and we are done.
2725 		 *
2726 		 * If FORCECLOSE is set, forcibly close the vnode.
2727 		 */
2728 		if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2729 			vgonel(vp);
2730 		} else {
2731 			busy++;
2732 #ifdef DIAGNOSTIC
2733 			if (busyprt)
2734 				vprint("vflush: busy vnode", vp);
2735 #endif
2736 		}
2737 		VOP_UNLOCK(vp, 0);
2738 		vdropl(vp);
2739 	}
2740 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2741 		/*
2742 		 * If just the root vnode is busy, and if its refcount
2743 		 * is equal to `rootrefs', then go ahead and kill it.
2744 		 */
2745 		VI_LOCK(rootvp);
2746 		KASSERT(busy > 0, ("vflush: not busy"));
2747 		VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2748 		    ("vflush: usecount %d < rootrefs %d",
2749 		     rootvp->v_usecount, rootrefs));
2750 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2751 			VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2752 			vgone(rootvp);
2753 			VOP_UNLOCK(rootvp, 0);
2754 			busy = 0;
2755 		} else
2756 			VI_UNLOCK(rootvp);
2757 	}
2758 	if (busy) {
2759 		CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2760 		    busy);
2761 		return (EBUSY);
2762 	}
2763 	for (; rootrefs > 0; rootrefs--)
2764 		vrele(rootvp);
2765 	return (0);
2766 }
2767 
2768 /*
2769  * Recycle an unused vnode to the front of the free list.
2770  */
2771 int
2772 vrecycle(struct vnode *vp)
2773 {
2774 	int recycled;
2775 
2776 	ASSERT_VOP_ELOCKED(vp, "vrecycle");
2777 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2778 	recycled = 0;
2779 	VI_LOCK(vp);
2780 	if (vp->v_usecount == 0) {
2781 		recycled = 1;
2782 		vgonel(vp);
2783 	}
2784 	VI_UNLOCK(vp);
2785 	return (recycled);
2786 }
2787 
2788 /*
2789  * Eliminate all activity associated with a vnode
2790  * in preparation for reuse.
2791  */
2792 void
2793 vgone(struct vnode *vp)
2794 {
2795 	VI_LOCK(vp);
2796 	vgonel(vp);
2797 	VI_UNLOCK(vp);
2798 }
2799 
2800 static void
2801 notify_lowervp_vfs_dummy(struct mount *mp __unused,
2802     struct vnode *lowervp __unused)
2803 {
2804 }
2805 
2806 /*
2807  * Notify upper mounts about reclaimed or unlinked vnode.
2808  */
2809 void
2810 vfs_notify_upper(struct vnode *vp, int event)
2811 {
2812 	static struct vfsops vgonel_vfsops = {
2813 		.vfs_reclaim_lowervp = notify_lowervp_vfs_dummy,
2814 		.vfs_unlink_lowervp = notify_lowervp_vfs_dummy,
2815 	};
2816 	struct mount *mp, *ump, *mmp;
2817 
2818 	mp = vp->v_mount;
2819 	if (mp == NULL)
2820 		return;
2821 
2822 	MNT_ILOCK(mp);
2823 	if (TAILQ_EMPTY(&mp->mnt_uppers))
2824 		goto unlock;
2825 	MNT_IUNLOCK(mp);
2826 	mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO);
2827 	mmp->mnt_op = &vgonel_vfsops;
2828 	mmp->mnt_kern_flag |= MNTK_MARKER;
2829 	MNT_ILOCK(mp);
2830 	mp->mnt_kern_flag |= MNTK_VGONE_UPPER;
2831 	for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) {
2832 		if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) {
2833 			ump = TAILQ_NEXT(ump, mnt_upper_link);
2834 			continue;
2835 		}
2836 		TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link);
2837 		MNT_IUNLOCK(mp);
2838 		switch (event) {
2839 		case VFS_NOTIFY_UPPER_RECLAIM:
2840 			VFS_RECLAIM_LOWERVP(ump, vp);
2841 			break;
2842 		case VFS_NOTIFY_UPPER_UNLINK:
2843 			VFS_UNLINK_LOWERVP(ump, vp);
2844 			break;
2845 		default:
2846 			KASSERT(0, ("invalid event %d", event));
2847 			break;
2848 		}
2849 		MNT_ILOCK(mp);
2850 		ump = TAILQ_NEXT(mmp, mnt_upper_link);
2851 		TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link);
2852 	}
2853 	free(mmp, M_TEMP);
2854 	mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER;
2855 	if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) {
2856 		mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER;
2857 		wakeup(&mp->mnt_uppers);
2858 	}
2859 unlock:
2860 	MNT_IUNLOCK(mp);
2861 }
2862 
2863 /*
2864  * vgone, with the vp interlock held.
2865  */
2866 static void
2867 vgonel(struct vnode *vp)
2868 {
2869 	struct thread *td;
2870 	int oweinact;
2871 	int active;
2872 	struct mount *mp;
2873 
2874 	ASSERT_VOP_ELOCKED(vp, "vgonel");
2875 	ASSERT_VI_LOCKED(vp, "vgonel");
2876 	VNASSERT(vp->v_holdcnt, vp,
2877 	    ("vgonel: vp %p has no reference.", vp));
2878 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2879 	td = curthread;
2880 
2881 	/*
2882 	 * Don't vgonel if we're already doomed.
2883 	 */
2884 	if (vp->v_iflag & VI_DOOMED)
2885 		return;
2886 	vp->v_iflag |= VI_DOOMED;
2887 
2888 	/*
2889 	 * Check to see if the vnode is in use.  If so, we have to call
2890 	 * VOP_CLOSE() and VOP_INACTIVE().
2891 	 */
2892 	active = vp->v_usecount;
2893 	oweinact = (vp->v_iflag & VI_OWEINACT);
2894 	VI_UNLOCK(vp);
2895 	vfs_notify_upper(vp, VFS_NOTIFY_UPPER_RECLAIM);
2896 
2897 	/*
2898 	 * If purging an active vnode, it must be closed and
2899 	 * deactivated before being reclaimed.
2900 	 */
2901 	if (active)
2902 		VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2903 	if (oweinact || active) {
2904 		VI_LOCK(vp);
2905 		if ((vp->v_iflag & VI_DOINGINACT) == 0)
2906 			vinactive(vp, td);
2907 		VI_UNLOCK(vp);
2908 	}
2909 	if (vp->v_type == VSOCK)
2910 		vfs_unp_reclaim(vp);
2911 
2912 	/*
2913 	 * Clean out any buffers associated with the vnode.
2914 	 * If the flush fails, just toss the buffers.
2915 	 */
2916 	mp = NULL;
2917 	if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2918 		(void) vn_start_secondary_write(vp, &mp, V_WAIT);
2919 	if (vinvalbuf(vp, V_SAVE, 0, 0) != 0) {
2920 		while (vinvalbuf(vp, 0, 0, 0) != 0)
2921 			;
2922 	}
2923 
2924 	BO_LOCK(&vp->v_bufobj);
2925 	KASSERT(TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd) &&
2926 	    vp->v_bufobj.bo_dirty.bv_cnt == 0 &&
2927 	    TAILQ_EMPTY(&vp->v_bufobj.bo_clean.bv_hd) &&
2928 	    vp->v_bufobj.bo_clean.bv_cnt == 0,
2929 	    ("vp %p bufobj not invalidated", vp));
2930 	vp->v_bufobj.bo_flag |= BO_DEAD;
2931 	BO_UNLOCK(&vp->v_bufobj);
2932 
2933 	/*
2934 	 * Reclaim the vnode.
2935 	 */
2936 	if (VOP_RECLAIM(vp, td))
2937 		panic("vgone: cannot reclaim");
2938 	if (mp != NULL)
2939 		vn_finished_secondary_write(mp);
2940 	VNASSERT(vp->v_object == NULL, vp,
2941 	    ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2942 	/*
2943 	 * Clear the advisory locks and wake up waiting threads.
2944 	 */
2945 	(void)VOP_ADVLOCKPURGE(vp);
2946 	/*
2947 	 * Delete from old mount point vnode list.
2948 	 */
2949 	delmntque(vp);
2950 	cache_purge(vp);
2951 	/*
2952 	 * Done with purge, reset to the standard lock and invalidate
2953 	 * the vnode.
2954 	 */
2955 	VI_LOCK(vp);
2956 	vp->v_vnlock = &vp->v_lock;
2957 	vp->v_op = &dead_vnodeops;
2958 	vp->v_tag = "none";
2959 	vp->v_type = VBAD;
2960 }
2961 
2962 /*
2963  * Calculate the total number of references to a special device.
2964  */
2965 int
2966 vcount(struct vnode *vp)
2967 {
2968 	int count;
2969 
2970 	dev_lock();
2971 	count = vp->v_rdev->si_usecount;
2972 	dev_unlock();
2973 	return (count);
2974 }
2975 
2976 /*
2977  * Same as above, but using the struct cdev *as argument
2978  */
2979 int
2980 count_dev(struct cdev *dev)
2981 {
2982 	int count;
2983 
2984 	dev_lock();
2985 	count = dev->si_usecount;
2986 	dev_unlock();
2987 	return(count);
2988 }
2989 
2990 /*
2991  * Print out a description of a vnode.
2992  */
2993 static char *typename[] =
2994 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2995  "VMARKER"};
2996 
2997 void
2998 vn_printf(struct vnode *vp, const char *fmt, ...)
2999 {
3000 	va_list ap;
3001 	char buf[256], buf2[16];
3002 	u_long flags;
3003 
3004 	va_start(ap, fmt);
3005 	vprintf(fmt, ap);
3006 	va_end(ap);
3007 	printf("%p: ", (void *)vp);
3008 	printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
3009 	printf("    usecount %d, writecount %d, refcount %d mountedhere %p\n",
3010 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
3011 	buf[0] = '\0';
3012 	buf[1] = '\0';
3013 	if (vp->v_vflag & VV_ROOT)
3014 		strlcat(buf, "|VV_ROOT", sizeof(buf));
3015 	if (vp->v_vflag & VV_ISTTY)
3016 		strlcat(buf, "|VV_ISTTY", sizeof(buf));
3017 	if (vp->v_vflag & VV_NOSYNC)
3018 		strlcat(buf, "|VV_NOSYNC", sizeof(buf));
3019 	if (vp->v_vflag & VV_ETERNALDEV)
3020 		strlcat(buf, "|VV_ETERNALDEV", sizeof(buf));
3021 	if (vp->v_vflag & VV_CACHEDLABEL)
3022 		strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
3023 	if (vp->v_vflag & VV_TEXT)
3024 		strlcat(buf, "|VV_TEXT", sizeof(buf));
3025 	if (vp->v_vflag & VV_COPYONWRITE)
3026 		strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
3027 	if (vp->v_vflag & VV_SYSTEM)
3028 		strlcat(buf, "|VV_SYSTEM", sizeof(buf));
3029 	if (vp->v_vflag & VV_PROCDEP)
3030 		strlcat(buf, "|VV_PROCDEP", sizeof(buf));
3031 	if (vp->v_vflag & VV_NOKNOTE)
3032 		strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
3033 	if (vp->v_vflag & VV_DELETED)
3034 		strlcat(buf, "|VV_DELETED", sizeof(buf));
3035 	if (vp->v_vflag & VV_MD)
3036 		strlcat(buf, "|VV_MD", sizeof(buf));
3037 	if (vp->v_vflag & VV_FORCEINSMQ)
3038 		strlcat(buf, "|VV_FORCEINSMQ", sizeof(buf));
3039 	flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | VV_ETERNALDEV |
3040 	    VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
3041 	    VV_NOKNOTE | VV_DELETED | VV_MD | VV_FORCEINSMQ);
3042 	if (flags != 0) {
3043 		snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
3044 		strlcat(buf, buf2, sizeof(buf));
3045 	}
3046 	if (vp->v_iflag & VI_MOUNT)
3047 		strlcat(buf, "|VI_MOUNT", sizeof(buf));
3048 	if (vp->v_iflag & VI_AGE)
3049 		strlcat(buf, "|VI_AGE", sizeof(buf));
3050 	if (vp->v_iflag & VI_DOOMED)
3051 		strlcat(buf, "|VI_DOOMED", sizeof(buf));
3052 	if (vp->v_iflag & VI_FREE)
3053 		strlcat(buf, "|VI_FREE", sizeof(buf));
3054 	if (vp->v_iflag & VI_ACTIVE)
3055 		strlcat(buf, "|VI_ACTIVE", sizeof(buf));
3056 	if (vp->v_iflag & VI_DOINGINACT)
3057 		strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
3058 	if (vp->v_iflag & VI_OWEINACT)
3059 		strlcat(buf, "|VI_OWEINACT", sizeof(buf));
3060 	flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
3061 	    VI_ACTIVE | VI_DOINGINACT | VI_OWEINACT);
3062 	if (flags != 0) {
3063 		snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
3064 		strlcat(buf, buf2, sizeof(buf));
3065 	}
3066 	printf("    flags (%s)\n", buf + 1);
3067 	if (mtx_owned(VI_MTX(vp)))
3068 		printf(" VI_LOCKed");
3069 	if (vp->v_object != NULL)
3070 		printf("    v_object %p ref %d pages %d "
3071 		    "cleanbuf %d dirtybuf %d\n",
3072 		    vp->v_object, vp->v_object->ref_count,
3073 		    vp->v_object->resident_page_count,
3074 		    vp->v_bufobj.bo_dirty.bv_cnt,
3075 		    vp->v_bufobj.bo_clean.bv_cnt);
3076 	printf("    ");
3077 	lockmgr_printinfo(vp->v_vnlock);
3078 	if (vp->v_data != NULL)
3079 		VOP_PRINT(vp);
3080 }
3081 
3082 #ifdef DDB
3083 /*
3084  * List all of the locked vnodes in the system.
3085  * Called when debugging the kernel.
3086  */
3087 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
3088 {
3089 	struct mount *mp;
3090 	struct vnode *vp;
3091 
3092 	/*
3093 	 * Note: because this is DDB, we can't obey the locking semantics
3094 	 * for these structures, which means we could catch an inconsistent
3095 	 * state and dereference a nasty pointer.  Not much to be done
3096 	 * about that.
3097 	 */
3098 	db_printf("Locked vnodes\n");
3099 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3100 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3101 			if (vp->v_type != VMARKER && VOP_ISLOCKED(vp))
3102 				vprint("", vp);
3103 		}
3104 	}
3105 }
3106 
3107 /*
3108  * Show details about the given vnode.
3109  */
3110 DB_SHOW_COMMAND(vnode, db_show_vnode)
3111 {
3112 	struct vnode *vp;
3113 
3114 	if (!have_addr)
3115 		return;
3116 	vp = (struct vnode *)addr;
3117 	vn_printf(vp, "vnode ");
3118 }
3119 
3120 /*
3121  * Show details about the given mount point.
3122  */
3123 DB_SHOW_COMMAND(mount, db_show_mount)
3124 {
3125 	struct mount *mp;
3126 	struct vfsopt *opt;
3127 	struct statfs *sp;
3128 	struct vnode *vp;
3129 	char buf[512];
3130 	uint64_t mflags;
3131 	u_int flags;
3132 
3133 	if (!have_addr) {
3134 		/* No address given, print short info about all mount points. */
3135 		TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3136 			db_printf("%p %s on %s (%s)\n", mp,
3137 			    mp->mnt_stat.f_mntfromname,
3138 			    mp->mnt_stat.f_mntonname,
3139 			    mp->mnt_stat.f_fstypename);
3140 			if (db_pager_quit)
3141 				break;
3142 		}
3143 		db_printf("\nMore info: show mount <addr>\n");
3144 		return;
3145 	}
3146 
3147 	mp = (struct mount *)addr;
3148 	db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
3149 	    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
3150 
3151 	buf[0] = '\0';
3152 	mflags = mp->mnt_flag;
3153 #define	MNT_FLAG(flag)	do {						\
3154 	if (mflags & (flag)) {						\
3155 		if (buf[0] != '\0')					\
3156 			strlcat(buf, ", ", sizeof(buf));		\
3157 		strlcat(buf, (#flag) + 4, sizeof(buf));			\
3158 		mflags &= ~(flag);					\
3159 	}								\
3160 } while (0)
3161 	MNT_FLAG(MNT_RDONLY);
3162 	MNT_FLAG(MNT_SYNCHRONOUS);
3163 	MNT_FLAG(MNT_NOEXEC);
3164 	MNT_FLAG(MNT_NOSUID);
3165 	MNT_FLAG(MNT_NFS4ACLS);
3166 	MNT_FLAG(MNT_UNION);
3167 	MNT_FLAG(MNT_ASYNC);
3168 	MNT_FLAG(MNT_SUIDDIR);
3169 	MNT_FLAG(MNT_SOFTDEP);
3170 	MNT_FLAG(MNT_NOSYMFOLLOW);
3171 	MNT_FLAG(MNT_GJOURNAL);
3172 	MNT_FLAG(MNT_MULTILABEL);
3173 	MNT_FLAG(MNT_ACLS);
3174 	MNT_FLAG(MNT_NOATIME);
3175 	MNT_FLAG(MNT_NOCLUSTERR);
3176 	MNT_FLAG(MNT_NOCLUSTERW);
3177 	MNT_FLAG(MNT_SUJ);
3178 	MNT_FLAG(MNT_EXRDONLY);
3179 	MNT_FLAG(MNT_EXPORTED);
3180 	MNT_FLAG(MNT_DEFEXPORTED);
3181 	MNT_FLAG(MNT_EXPORTANON);
3182 	MNT_FLAG(MNT_EXKERB);
3183 	MNT_FLAG(MNT_EXPUBLIC);
3184 	MNT_FLAG(MNT_LOCAL);
3185 	MNT_FLAG(MNT_QUOTA);
3186 	MNT_FLAG(MNT_ROOTFS);
3187 	MNT_FLAG(MNT_USER);
3188 	MNT_FLAG(MNT_IGNORE);
3189 	MNT_FLAG(MNT_UPDATE);
3190 	MNT_FLAG(MNT_DELEXPORT);
3191 	MNT_FLAG(MNT_RELOAD);
3192 	MNT_FLAG(MNT_FORCE);
3193 	MNT_FLAG(MNT_SNAPSHOT);
3194 	MNT_FLAG(MNT_BYFSID);
3195 #undef MNT_FLAG
3196 	if (mflags != 0) {
3197 		if (buf[0] != '\0')
3198 			strlcat(buf, ", ", sizeof(buf));
3199 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3200 		    "0x%016jx", mflags);
3201 	}
3202 	db_printf("    mnt_flag = %s\n", buf);
3203 
3204 	buf[0] = '\0';
3205 	flags = mp->mnt_kern_flag;
3206 #define	MNT_KERN_FLAG(flag)	do {					\
3207 	if (flags & (flag)) {						\
3208 		if (buf[0] != '\0')					\
3209 			strlcat(buf, ", ", sizeof(buf));		\
3210 		strlcat(buf, (#flag) + 5, sizeof(buf));			\
3211 		flags &= ~(flag);					\
3212 	}								\
3213 } while (0)
3214 	MNT_KERN_FLAG(MNTK_UNMOUNTF);
3215 	MNT_KERN_FLAG(MNTK_ASYNC);
3216 	MNT_KERN_FLAG(MNTK_SOFTDEP);
3217 	MNT_KERN_FLAG(MNTK_NOINSMNTQ);
3218 	MNT_KERN_FLAG(MNTK_DRAINING);
3219 	MNT_KERN_FLAG(MNTK_REFEXPIRE);
3220 	MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
3221 	MNT_KERN_FLAG(MNTK_SHARED_WRITES);
3222 	MNT_KERN_FLAG(MNTK_NO_IOPF);
3223 	MNT_KERN_FLAG(MNTK_VGONE_UPPER);
3224 	MNT_KERN_FLAG(MNTK_VGONE_WAITER);
3225 	MNT_KERN_FLAG(MNTK_LOOKUP_EXCL_DOTDOT);
3226 	MNT_KERN_FLAG(MNTK_MARKER);
3227 	MNT_KERN_FLAG(MNTK_USES_BCACHE);
3228 	MNT_KERN_FLAG(MNTK_NOASYNC);
3229 	MNT_KERN_FLAG(MNTK_UNMOUNT);
3230 	MNT_KERN_FLAG(MNTK_MWAIT);
3231 	MNT_KERN_FLAG(MNTK_SUSPEND);
3232 	MNT_KERN_FLAG(MNTK_SUSPEND2);
3233 	MNT_KERN_FLAG(MNTK_SUSPENDED);
3234 	MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
3235 	MNT_KERN_FLAG(MNTK_NOKNOTE);
3236 #undef MNT_KERN_FLAG
3237 	if (flags != 0) {
3238 		if (buf[0] != '\0')
3239 			strlcat(buf, ", ", sizeof(buf));
3240 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3241 		    "0x%08x", flags);
3242 	}
3243 	db_printf("    mnt_kern_flag = %s\n", buf);
3244 
3245 	db_printf("    mnt_opt = ");
3246 	opt = TAILQ_FIRST(mp->mnt_opt);
3247 	if (opt != NULL) {
3248 		db_printf("%s", opt->name);
3249 		opt = TAILQ_NEXT(opt, link);
3250 		while (opt != NULL) {
3251 			db_printf(", %s", opt->name);
3252 			opt = TAILQ_NEXT(opt, link);
3253 		}
3254 	}
3255 	db_printf("\n");
3256 
3257 	sp = &mp->mnt_stat;
3258 	db_printf("    mnt_stat = { version=%u type=%u flags=0x%016jx "
3259 	    "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
3260 	    "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
3261 	    "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
3262 	    (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
3263 	    (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
3264 	    (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
3265 	    (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
3266 	    (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
3267 	    (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
3268 	    (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
3269 	    (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
3270 
3271 	db_printf("    mnt_cred = { uid=%u ruid=%u",
3272 	    (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
3273 	if (jailed(mp->mnt_cred))
3274 		db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
3275 	db_printf(" }\n");
3276 	db_printf("    mnt_ref = %d\n", mp->mnt_ref);
3277 	db_printf("    mnt_gen = %d\n", mp->mnt_gen);
3278 	db_printf("    mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
3279 	db_printf("    mnt_activevnodelistsize = %d\n",
3280 	    mp->mnt_activevnodelistsize);
3281 	db_printf("    mnt_writeopcount = %d\n", mp->mnt_writeopcount);
3282 	db_printf("    mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
3283 	db_printf("    mnt_iosize_max = %d\n", mp->mnt_iosize_max);
3284 	db_printf("    mnt_hashseed = %u\n", mp->mnt_hashseed);
3285 	db_printf("    mnt_lockref = %d\n", mp->mnt_lockref);
3286 	db_printf("    mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
3287 	db_printf("    mnt_secondary_accwrites = %d\n",
3288 	    mp->mnt_secondary_accwrites);
3289 	db_printf("    mnt_gjprovider = %s\n",
3290 	    mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
3291 
3292 	db_printf("\n\nList of active vnodes\n");
3293 	TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) {
3294 		if (vp->v_type != VMARKER) {
3295 			vn_printf(vp, "vnode ");
3296 			if (db_pager_quit)
3297 				break;
3298 		}
3299 	}
3300 	db_printf("\n\nList of inactive vnodes\n");
3301 	TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3302 		if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) {
3303 			vn_printf(vp, "vnode ");
3304 			if (db_pager_quit)
3305 				break;
3306 		}
3307 	}
3308 }
3309 #endif	/* DDB */
3310 
3311 /*
3312  * Fill in a struct xvfsconf based on a struct vfsconf.
3313  */
3314 static int
3315 vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp)
3316 {
3317 	struct xvfsconf xvfsp;
3318 
3319 	bzero(&xvfsp, sizeof(xvfsp));
3320 	strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3321 	xvfsp.vfc_typenum = vfsp->vfc_typenum;
3322 	xvfsp.vfc_refcount = vfsp->vfc_refcount;
3323 	xvfsp.vfc_flags = vfsp->vfc_flags;
3324 	/*
3325 	 * These are unused in userland, we keep them
3326 	 * to not break binary compatibility.
3327 	 */
3328 	xvfsp.vfc_vfsops = NULL;
3329 	xvfsp.vfc_next = NULL;
3330 	return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3331 }
3332 
3333 #ifdef COMPAT_FREEBSD32
3334 struct xvfsconf32 {
3335 	uint32_t	vfc_vfsops;
3336 	char		vfc_name[MFSNAMELEN];
3337 	int32_t		vfc_typenum;
3338 	int32_t		vfc_refcount;
3339 	int32_t		vfc_flags;
3340 	uint32_t	vfc_next;
3341 };
3342 
3343 static int
3344 vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp)
3345 {
3346 	struct xvfsconf32 xvfsp;
3347 
3348 	strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3349 	xvfsp.vfc_typenum = vfsp->vfc_typenum;
3350 	xvfsp.vfc_refcount = vfsp->vfc_refcount;
3351 	xvfsp.vfc_flags = vfsp->vfc_flags;
3352 	xvfsp.vfc_vfsops = 0;
3353 	xvfsp.vfc_next = 0;
3354 	return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3355 }
3356 #endif
3357 
3358 /*
3359  * Top level filesystem related information gathering.
3360  */
3361 static int
3362 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
3363 {
3364 	struct vfsconf *vfsp;
3365 	int error;
3366 
3367 	error = 0;
3368 	vfsconf_slock();
3369 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3370 #ifdef COMPAT_FREEBSD32
3371 		if (req->flags & SCTL_MASK32)
3372 			error = vfsconf2x32(req, vfsp);
3373 		else
3374 #endif
3375 			error = vfsconf2x(req, vfsp);
3376 		if (error)
3377 			break;
3378 	}
3379 	vfsconf_sunlock();
3380 	return (error);
3381 }
3382 
3383 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD |
3384     CTLFLAG_MPSAFE, NULL, 0, sysctl_vfs_conflist,
3385     "S,xvfsconf", "List of all configured filesystems");
3386 
3387 #ifndef BURN_BRIDGES
3388 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
3389 
3390 static int
3391 vfs_sysctl(SYSCTL_HANDLER_ARGS)
3392 {
3393 	int *name = (int *)arg1 - 1;	/* XXX */
3394 	u_int namelen = arg2 + 1;	/* XXX */
3395 	struct vfsconf *vfsp;
3396 
3397 	log(LOG_WARNING, "userland calling deprecated sysctl, "
3398 	    "please rebuild world\n");
3399 
3400 #if 1 || defined(COMPAT_PRELITE2)
3401 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3402 	if (namelen == 1)
3403 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3404 #endif
3405 
3406 	switch (name[1]) {
3407 	case VFS_MAXTYPENUM:
3408 		if (namelen != 2)
3409 			return (ENOTDIR);
3410 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3411 	case VFS_CONF:
3412 		if (namelen != 3)
3413 			return (ENOTDIR);	/* overloaded */
3414 		vfsconf_slock();
3415 		TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3416 			if (vfsp->vfc_typenum == name[2])
3417 				break;
3418 		}
3419 		vfsconf_sunlock();
3420 		if (vfsp == NULL)
3421 			return (EOPNOTSUPP);
3422 #ifdef COMPAT_FREEBSD32
3423 		if (req->flags & SCTL_MASK32)
3424 			return (vfsconf2x32(req, vfsp));
3425 		else
3426 #endif
3427 			return (vfsconf2x(req, vfsp));
3428 	}
3429 	return (EOPNOTSUPP);
3430 }
3431 
3432 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP |
3433     CTLFLAG_MPSAFE, vfs_sysctl,
3434     "Generic filesystem");
3435 
3436 #if 1 || defined(COMPAT_PRELITE2)
3437 
3438 static int
3439 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3440 {
3441 	int error;
3442 	struct vfsconf *vfsp;
3443 	struct ovfsconf ovfs;
3444 
3445 	vfsconf_slock();
3446 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3447 		bzero(&ovfs, sizeof(ovfs));
3448 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
3449 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
3450 		ovfs.vfc_index = vfsp->vfc_typenum;
3451 		ovfs.vfc_refcount = vfsp->vfc_refcount;
3452 		ovfs.vfc_flags = vfsp->vfc_flags;
3453 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3454 		if (error != 0) {
3455 			vfsconf_sunlock();
3456 			return (error);
3457 		}
3458 	}
3459 	vfsconf_sunlock();
3460 	return (0);
3461 }
3462 
3463 #endif /* 1 || COMPAT_PRELITE2 */
3464 #endif /* !BURN_BRIDGES */
3465 
3466 #define KINFO_VNODESLOP		10
3467 #ifdef notyet
3468 /*
3469  * Dump vnode list (via sysctl).
3470  */
3471 /* ARGSUSED */
3472 static int
3473 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3474 {
3475 	struct xvnode *xvn;
3476 	struct mount *mp;
3477 	struct vnode *vp;
3478 	int error, len, n;
3479 
3480 	/*
3481 	 * Stale numvnodes access is not fatal here.
3482 	 */
3483 	req->lock = 0;
3484 	len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3485 	if (!req->oldptr)
3486 		/* Make an estimate */
3487 		return (SYSCTL_OUT(req, 0, len));
3488 
3489 	error = sysctl_wire_old_buffer(req, 0);
3490 	if (error != 0)
3491 		return (error);
3492 	xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3493 	n = 0;
3494 	mtx_lock(&mountlist_mtx);
3495 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3496 		if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3497 			continue;
3498 		MNT_ILOCK(mp);
3499 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3500 			if (n == len)
3501 				break;
3502 			vref(vp);
3503 			xvn[n].xv_size = sizeof *xvn;
3504 			xvn[n].xv_vnode = vp;
3505 			xvn[n].xv_id = 0;	/* XXX compat */
3506 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3507 			XV_COPY(usecount);
3508 			XV_COPY(writecount);
3509 			XV_COPY(holdcnt);
3510 			XV_COPY(mount);
3511 			XV_COPY(numoutput);
3512 			XV_COPY(type);
3513 #undef XV_COPY
3514 			xvn[n].xv_flag = vp->v_vflag;
3515 
3516 			switch (vp->v_type) {
3517 			case VREG:
3518 			case VDIR:
3519 			case VLNK:
3520 				break;
3521 			case VBLK:
3522 			case VCHR:
3523 				if (vp->v_rdev == NULL) {
3524 					vrele(vp);
3525 					continue;
3526 				}
3527 				xvn[n].xv_dev = dev2udev(vp->v_rdev);
3528 				break;
3529 			case VSOCK:
3530 				xvn[n].xv_socket = vp->v_socket;
3531 				break;
3532 			case VFIFO:
3533 				xvn[n].xv_fifo = vp->v_fifoinfo;
3534 				break;
3535 			case VNON:
3536 			case VBAD:
3537 			default:
3538 				/* shouldn't happen? */
3539 				vrele(vp);
3540 				continue;
3541 			}
3542 			vrele(vp);
3543 			++n;
3544 		}
3545 		MNT_IUNLOCK(mp);
3546 		mtx_lock(&mountlist_mtx);
3547 		vfs_unbusy(mp);
3548 		if (n == len)
3549 			break;
3550 	}
3551 	mtx_unlock(&mountlist_mtx);
3552 
3553 	error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3554 	free(xvn, M_TEMP);
3555 	return (error);
3556 }
3557 
3558 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE | CTLFLAG_RD |
3559     CTLFLAG_MPSAFE, 0, 0, sysctl_vnode, "S,xvnode",
3560     "");
3561 #endif
3562 
3563 static void
3564 unmount_or_warn(struct mount *mp)
3565 {
3566 	int error;
3567 
3568 	error = dounmount(mp, MNT_FORCE, curthread);
3569 	if (error != 0) {
3570 		printf("unmount of %s failed (", mp->mnt_stat.f_mntonname);
3571 		if (error == EBUSY)
3572 			printf("BUSY)\n");
3573 		else
3574 			printf("%d)\n", error);
3575 	}
3576 }
3577 
3578 /*
3579  * Unmount all filesystems. The list is traversed in reverse order
3580  * of mounting to avoid dependencies.
3581  */
3582 void
3583 vfs_unmountall(void)
3584 {
3585 	struct mount *mp, *tmp;
3586 
3587 	CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3588 
3589 	/*
3590 	 * Since this only runs when rebooting, it is not interlocked.
3591 	 */
3592 	TAILQ_FOREACH_REVERSE_SAFE(mp, &mountlist, mntlist, mnt_list, tmp) {
3593 		vfs_ref(mp);
3594 
3595 		/*
3596 		 * Forcibly unmounting "/dev" before "/" would prevent clean
3597 		 * unmount of the latter.
3598 		 */
3599 		if (mp == rootdevmp)
3600 			continue;
3601 
3602 		unmount_or_warn(mp);
3603 	}
3604 
3605 	if (rootdevmp != NULL)
3606 		unmount_or_warn(rootdevmp);
3607 }
3608 
3609 /*
3610  * perform msync on all vnodes under a mount point
3611  * the mount point must be locked.
3612  */
3613 void
3614 vfs_msync(struct mount *mp, int flags)
3615 {
3616 	struct vnode *vp, *mvp;
3617 	struct vm_object *obj;
3618 
3619 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3620 	MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
3621 		obj = vp->v_object;
3622 		if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3623 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3624 			if (!vget(vp,
3625 			    LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3626 			    curthread)) {
3627 				if (vp->v_vflag & VV_NOSYNC) {	/* unlinked */
3628 					vput(vp);
3629 					continue;
3630 				}
3631 
3632 				obj = vp->v_object;
3633 				if (obj != NULL) {
3634 					VM_OBJECT_WLOCK(obj);
3635 					vm_object_page_clean(obj, 0, 0,
3636 					    flags == MNT_WAIT ?
3637 					    OBJPC_SYNC : OBJPC_NOSYNC);
3638 					VM_OBJECT_WUNLOCK(obj);
3639 				}
3640 				vput(vp);
3641 			}
3642 		} else
3643 			VI_UNLOCK(vp);
3644 	}
3645 }
3646 
3647 static void
3648 destroy_vpollinfo_free(struct vpollinfo *vi)
3649 {
3650 
3651 	knlist_destroy(&vi->vpi_selinfo.si_note);
3652 	mtx_destroy(&vi->vpi_lock);
3653 	uma_zfree(vnodepoll_zone, vi);
3654 }
3655 
3656 static void
3657 destroy_vpollinfo(struct vpollinfo *vi)
3658 {
3659 
3660 	knlist_clear(&vi->vpi_selinfo.si_note, 1);
3661 	seldrain(&vi->vpi_selinfo);
3662 	destroy_vpollinfo_free(vi);
3663 }
3664 
3665 /*
3666  * Initalize per-vnode helper structure to hold poll-related state.
3667  */
3668 void
3669 v_addpollinfo(struct vnode *vp)
3670 {
3671 	struct vpollinfo *vi;
3672 
3673 	if (vp->v_pollinfo != NULL)
3674 		return;
3675 	vi = uma_zalloc(vnodepoll_zone, M_WAITOK | M_ZERO);
3676 	mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3677 	knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3678 	    vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3679 	VI_LOCK(vp);
3680 	if (vp->v_pollinfo != NULL) {
3681 		VI_UNLOCK(vp);
3682 		destroy_vpollinfo_free(vi);
3683 		return;
3684 	}
3685 	vp->v_pollinfo = vi;
3686 	VI_UNLOCK(vp);
3687 }
3688 
3689 /*
3690  * Record a process's interest in events which might happen to
3691  * a vnode.  Because poll uses the historic select-style interface
3692  * internally, this routine serves as both the ``check for any
3693  * pending events'' and the ``record my interest in future events''
3694  * functions.  (These are done together, while the lock is held,
3695  * to avoid race conditions.)
3696  */
3697 int
3698 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3699 {
3700 
3701 	v_addpollinfo(vp);
3702 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3703 	if (vp->v_pollinfo->vpi_revents & events) {
3704 		/*
3705 		 * This leaves events we are not interested
3706 		 * in available for the other process which
3707 		 * which presumably had requested them
3708 		 * (otherwise they would never have been
3709 		 * recorded).
3710 		 */
3711 		events &= vp->v_pollinfo->vpi_revents;
3712 		vp->v_pollinfo->vpi_revents &= ~events;
3713 
3714 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
3715 		return (events);
3716 	}
3717 	vp->v_pollinfo->vpi_events |= events;
3718 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3719 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3720 	return (0);
3721 }
3722 
3723 /*
3724  * Routine to create and manage a filesystem syncer vnode.
3725  */
3726 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
3727 static int	sync_fsync(struct  vop_fsync_args *);
3728 static int	sync_inactive(struct  vop_inactive_args *);
3729 static int	sync_reclaim(struct  vop_reclaim_args *);
3730 
3731 static struct vop_vector sync_vnodeops = {
3732 	.vop_bypass =	VOP_EOPNOTSUPP,
3733 	.vop_close =	sync_close,		/* close */
3734 	.vop_fsync =	sync_fsync,		/* fsync */
3735 	.vop_inactive =	sync_inactive,	/* inactive */
3736 	.vop_reclaim =	sync_reclaim,	/* reclaim */
3737 	.vop_lock1 =	vop_stdlock,	/* lock */
3738 	.vop_unlock =	vop_stdunlock,	/* unlock */
3739 	.vop_islocked =	vop_stdislocked,	/* islocked */
3740 };
3741 
3742 /*
3743  * Create a new filesystem syncer vnode for the specified mount point.
3744  */
3745 void
3746 vfs_allocate_syncvnode(struct mount *mp)
3747 {
3748 	struct vnode *vp;
3749 	struct bufobj *bo;
3750 	static long start, incr, next;
3751 	int error;
3752 
3753 	/* Allocate a new vnode */
3754 	error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
3755 	if (error != 0)
3756 		panic("vfs_allocate_syncvnode: getnewvnode() failed");
3757 	vp->v_type = VNON;
3758 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3759 	vp->v_vflag |= VV_FORCEINSMQ;
3760 	error = insmntque(vp, mp);
3761 	if (error != 0)
3762 		panic("vfs_allocate_syncvnode: insmntque() failed");
3763 	vp->v_vflag &= ~VV_FORCEINSMQ;
3764 	VOP_UNLOCK(vp, 0);
3765 	/*
3766 	 * Place the vnode onto the syncer worklist. We attempt to
3767 	 * scatter them about on the list so that they will go off
3768 	 * at evenly distributed times even if all the filesystems
3769 	 * are mounted at once.
3770 	 */
3771 	next += incr;
3772 	if (next == 0 || next > syncer_maxdelay) {
3773 		start /= 2;
3774 		incr /= 2;
3775 		if (start == 0) {
3776 			start = syncer_maxdelay / 2;
3777 			incr = syncer_maxdelay;
3778 		}
3779 		next = start;
3780 	}
3781 	bo = &vp->v_bufobj;
3782 	BO_LOCK(bo);
3783 	vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3784 	/* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3785 	mtx_lock(&sync_mtx);
3786 	sync_vnode_count++;
3787 	if (mp->mnt_syncer == NULL) {
3788 		mp->mnt_syncer = vp;
3789 		vp = NULL;
3790 	}
3791 	mtx_unlock(&sync_mtx);
3792 	BO_UNLOCK(bo);
3793 	if (vp != NULL) {
3794 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3795 		vgone(vp);
3796 		vput(vp);
3797 	}
3798 }
3799 
3800 void
3801 vfs_deallocate_syncvnode(struct mount *mp)
3802 {
3803 	struct vnode *vp;
3804 
3805 	mtx_lock(&sync_mtx);
3806 	vp = mp->mnt_syncer;
3807 	if (vp != NULL)
3808 		mp->mnt_syncer = NULL;
3809 	mtx_unlock(&sync_mtx);
3810 	if (vp != NULL)
3811 		vrele(vp);
3812 }
3813 
3814 /*
3815  * Do a lazy sync of the filesystem.
3816  */
3817 static int
3818 sync_fsync(struct vop_fsync_args *ap)
3819 {
3820 	struct vnode *syncvp = ap->a_vp;
3821 	struct mount *mp = syncvp->v_mount;
3822 	int error, save;
3823 	struct bufobj *bo;
3824 
3825 	/*
3826 	 * We only need to do something if this is a lazy evaluation.
3827 	 */
3828 	if (ap->a_waitfor != MNT_LAZY)
3829 		return (0);
3830 
3831 	/*
3832 	 * Move ourselves to the back of the sync list.
3833 	 */
3834 	bo = &syncvp->v_bufobj;
3835 	BO_LOCK(bo);
3836 	vn_syncer_add_to_worklist(bo, syncdelay);
3837 	BO_UNLOCK(bo);
3838 
3839 	/*
3840 	 * Walk the list of vnodes pushing all that are dirty and
3841 	 * not already on the sync list.
3842 	 */
3843 	if (vfs_busy(mp, MBF_NOWAIT) != 0)
3844 		return (0);
3845 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3846 		vfs_unbusy(mp);
3847 		return (0);
3848 	}
3849 	save = curthread_pflags_set(TDP_SYNCIO);
3850 	vfs_msync(mp, MNT_NOWAIT);
3851 	error = VFS_SYNC(mp, MNT_LAZY);
3852 	curthread_pflags_restore(save);
3853 	vn_finished_write(mp);
3854 	vfs_unbusy(mp);
3855 	return (error);
3856 }
3857 
3858 /*
3859  * The syncer vnode is no referenced.
3860  */
3861 static int
3862 sync_inactive(struct vop_inactive_args *ap)
3863 {
3864 
3865 	vgone(ap->a_vp);
3866 	return (0);
3867 }
3868 
3869 /*
3870  * The syncer vnode is no longer needed and is being decommissioned.
3871  *
3872  * Modifications to the worklist must be protected by sync_mtx.
3873  */
3874 static int
3875 sync_reclaim(struct vop_reclaim_args *ap)
3876 {
3877 	struct vnode *vp = ap->a_vp;
3878 	struct bufobj *bo;
3879 
3880 	bo = &vp->v_bufobj;
3881 	BO_LOCK(bo);
3882 	mtx_lock(&sync_mtx);
3883 	if (vp->v_mount->mnt_syncer == vp)
3884 		vp->v_mount->mnt_syncer = NULL;
3885 	if (bo->bo_flag & BO_ONWORKLST) {
3886 		LIST_REMOVE(bo, bo_synclist);
3887 		syncer_worklist_len--;
3888 		sync_vnode_count--;
3889 		bo->bo_flag &= ~BO_ONWORKLST;
3890 	}
3891 	mtx_unlock(&sync_mtx);
3892 	BO_UNLOCK(bo);
3893 
3894 	return (0);
3895 }
3896 
3897 /*
3898  * Check if vnode represents a disk device
3899  */
3900 int
3901 vn_isdisk(struct vnode *vp, int *errp)
3902 {
3903 	int error;
3904 
3905 	if (vp->v_type != VCHR) {
3906 		error = ENOTBLK;
3907 		goto out;
3908 	}
3909 	error = 0;
3910 	dev_lock();
3911 	if (vp->v_rdev == NULL)
3912 		error = ENXIO;
3913 	else if (vp->v_rdev->si_devsw == NULL)
3914 		error = ENXIO;
3915 	else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3916 		error = ENOTBLK;
3917 	dev_unlock();
3918 out:
3919 	if (errp != NULL)
3920 		*errp = error;
3921 	return (error == 0);
3922 }
3923 
3924 /*
3925  * Common filesystem object access control check routine.  Accepts a
3926  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3927  * and optional call-by-reference privused argument allowing vaccess()
3928  * to indicate to the caller whether privilege was used to satisfy the
3929  * request (obsoleted).  Returns 0 on success, or an errno on failure.
3930  */
3931 int
3932 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3933     accmode_t accmode, struct ucred *cred, int *privused)
3934 {
3935 	accmode_t dac_granted;
3936 	accmode_t priv_granted;
3937 
3938 	KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3939 	    ("invalid bit in accmode"));
3940 	KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
3941 	    ("VAPPEND without VWRITE"));
3942 
3943 	/*
3944 	 * Look for a normal, non-privileged way to access the file/directory
3945 	 * as requested.  If it exists, go with that.
3946 	 */
3947 
3948 	if (privused != NULL)
3949 		*privused = 0;
3950 
3951 	dac_granted = 0;
3952 
3953 	/* Check the owner. */
3954 	if (cred->cr_uid == file_uid) {
3955 		dac_granted |= VADMIN;
3956 		if (file_mode & S_IXUSR)
3957 			dac_granted |= VEXEC;
3958 		if (file_mode & S_IRUSR)
3959 			dac_granted |= VREAD;
3960 		if (file_mode & S_IWUSR)
3961 			dac_granted |= (VWRITE | VAPPEND);
3962 
3963 		if ((accmode & dac_granted) == accmode)
3964 			return (0);
3965 
3966 		goto privcheck;
3967 	}
3968 
3969 	/* Otherwise, check the groups (first match) */
3970 	if (groupmember(file_gid, cred)) {
3971 		if (file_mode & S_IXGRP)
3972 			dac_granted |= VEXEC;
3973 		if (file_mode & S_IRGRP)
3974 			dac_granted |= VREAD;
3975 		if (file_mode & S_IWGRP)
3976 			dac_granted |= (VWRITE | VAPPEND);
3977 
3978 		if ((accmode & dac_granted) == accmode)
3979 			return (0);
3980 
3981 		goto privcheck;
3982 	}
3983 
3984 	/* Otherwise, check everyone else. */
3985 	if (file_mode & S_IXOTH)
3986 		dac_granted |= VEXEC;
3987 	if (file_mode & S_IROTH)
3988 		dac_granted |= VREAD;
3989 	if (file_mode & S_IWOTH)
3990 		dac_granted |= (VWRITE | VAPPEND);
3991 	if ((accmode & dac_granted) == accmode)
3992 		return (0);
3993 
3994 privcheck:
3995 	/*
3996 	 * Build a privilege mask to determine if the set of privileges
3997 	 * satisfies the requirements when combined with the granted mask
3998 	 * from above.  For each privilege, if the privilege is required,
3999 	 * bitwise or the request type onto the priv_granted mask.
4000 	 */
4001 	priv_granted = 0;
4002 
4003 	if (type == VDIR) {
4004 		/*
4005 		 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
4006 		 * requests, instead of PRIV_VFS_EXEC.
4007 		 */
4008 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
4009 		    !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
4010 			priv_granted |= VEXEC;
4011 	} else {
4012 		/*
4013 		 * Ensure that at least one execute bit is on. Otherwise,
4014 		 * a privileged user will always succeed, and we don't want
4015 		 * this to happen unless the file really is executable.
4016 		 */
4017 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
4018 		    (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
4019 		    !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
4020 			priv_granted |= VEXEC;
4021 	}
4022 
4023 	if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
4024 	    !priv_check_cred(cred, PRIV_VFS_READ, 0))
4025 		priv_granted |= VREAD;
4026 
4027 	if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
4028 	    !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
4029 		priv_granted |= (VWRITE | VAPPEND);
4030 
4031 	if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
4032 	    !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
4033 		priv_granted |= VADMIN;
4034 
4035 	if ((accmode & (priv_granted | dac_granted)) == accmode) {
4036 		/* XXX audit: privilege used */
4037 		if (privused != NULL)
4038 			*privused = 1;
4039 		return (0);
4040 	}
4041 
4042 	return ((accmode & VADMIN) ? EPERM : EACCES);
4043 }
4044 
4045 /*
4046  * Credential check based on process requesting service, and per-attribute
4047  * permissions.
4048  */
4049 int
4050 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
4051     struct thread *td, accmode_t accmode)
4052 {
4053 
4054 	/*
4055 	 * Kernel-invoked always succeeds.
4056 	 */
4057 	if (cred == NOCRED)
4058 		return (0);
4059 
4060 	/*
4061 	 * Do not allow privileged processes in jail to directly manipulate
4062 	 * system attributes.
4063 	 */
4064 	switch (attrnamespace) {
4065 	case EXTATTR_NAMESPACE_SYSTEM:
4066 		/* Potentially should be: return (EPERM); */
4067 		return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
4068 	case EXTATTR_NAMESPACE_USER:
4069 		return (VOP_ACCESS(vp, accmode, cred, td));
4070 	default:
4071 		return (EPERM);
4072 	}
4073 }
4074 
4075 #ifdef DEBUG_VFS_LOCKS
4076 /*
4077  * This only exists to supress warnings from unlocked specfs accesses.  It is
4078  * no longer ok to have an unlocked VFS.
4079  */
4080 #define	IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL ||		\
4081 	(vp)->v_type == VCHR ||	(vp)->v_type == VBAD)
4082 
4083 int vfs_badlock_ddb = 1;	/* Drop into debugger on violation. */
4084 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
4085     "Drop into debugger on lock violation");
4086 
4087 int vfs_badlock_mutex = 1;	/* Check for interlock across VOPs. */
4088 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
4089     0, "Check for interlock across VOPs");
4090 
4091 int vfs_badlock_print = 1;	/* Print lock violations. */
4092 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
4093     0, "Print lock violations");
4094 
4095 #ifdef KDB
4096 int vfs_badlock_backtrace = 1;	/* Print backtrace at lock violations. */
4097 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
4098     &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
4099 #endif
4100 
4101 static void
4102 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
4103 {
4104 
4105 #ifdef KDB
4106 	if (vfs_badlock_backtrace)
4107 		kdb_backtrace();
4108 #endif
4109 	if (vfs_badlock_print)
4110 		printf("%s: %p %s\n", str, (void *)vp, msg);
4111 	if (vfs_badlock_ddb)
4112 		kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
4113 }
4114 
4115 void
4116 assert_vi_locked(struct vnode *vp, const char *str)
4117 {
4118 
4119 	if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
4120 		vfs_badlock("interlock is not locked but should be", str, vp);
4121 }
4122 
4123 void
4124 assert_vi_unlocked(struct vnode *vp, const char *str)
4125 {
4126 
4127 	if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
4128 		vfs_badlock("interlock is locked but should not be", str, vp);
4129 }
4130 
4131 void
4132 assert_vop_locked(struct vnode *vp, const char *str)
4133 {
4134 	int locked;
4135 
4136 	if (!IGNORE_LOCK(vp)) {
4137 		locked = VOP_ISLOCKED(vp);
4138 		if (locked == 0 || locked == LK_EXCLOTHER)
4139 			vfs_badlock("is not locked but should be", str, vp);
4140 	}
4141 }
4142 
4143 void
4144 assert_vop_unlocked(struct vnode *vp, const char *str)
4145 {
4146 
4147 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
4148 		vfs_badlock("is locked but should not be", str, vp);
4149 }
4150 
4151 void
4152 assert_vop_elocked(struct vnode *vp, const char *str)
4153 {
4154 
4155 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
4156 		vfs_badlock("is not exclusive locked but should be", str, vp);
4157 }
4158 
4159 #if 0
4160 void
4161 assert_vop_elocked_other(struct vnode *vp, const char *str)
4162 {
4163 
4164 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
4165 		vfs_badlock("is not exclusive locked by another thread",
4166 		    str, vp);
4167 }
4168 
4169 void
4170 assert_vop_slocked(struct vnode *vp, const char *str)
4171 {
4172 
4173 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
4174 		vfs_badlock("is not locked shared but should be", str, vp);
4175 }
4176 #endif /* 0 */
4177 #endif /* DEBUG_VFS_LOCKS */
4178 
4179 void
4180 vop_rename_fail(struct vop_rename_args *ap)
4181 {
4182 
4183 	if (ap->a_tvp != NULL)
4184 		vput(ap->a_tvp);
4185 	if (ap->a_tdvp == ap->a_tvp)
4186 		vrele(ap->a_tdvp);
4187 	else
4188 		vput(ap->a_tdvp);
4189 	vrele(ap->a_fdvp);
4190 	vrele(ap->a_fvp);
4191 }
4192 
4193 void
4194 vop_rename_pre(void *ap)
4195 {
4196 	struct vop_rename_args *a = ap;
4197 
4198 #ifdef DEBUG_VFS_LOCKS
4199 	if (a->a_tvp)
4200 		ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
4201 	ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
4202 	ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
4203 	ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
4204 
4205 	/* Check the source (from). */
4206 	if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
4207 	    (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
4208 		ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
4209 	if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
4210 		ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
4211 
4212 	/* Check the target. */
4213 	if (a->a_tvp)
4214 		ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
4215 	ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
4216 #endif
4217 	if (a->a_tdvp != a->a_fdvp)
4218 		vhold(a->a_fdvp);
4219 	if (a->a_tvp != a->a_fvp)
4220 		vhold(a->a_fvp);
4221 	vhold(a->a_tdvp);
4222 	if (a->a_tvp)
4223 		vhold(a->a_tvp);
4224 }
4225 
4226 void
4227 vop_strategy_pre(void *ap)
4228 {
4229 #ifdef DEBUG_VFS_LOCKS
4230 	struct vop_strategy_args *a;
4231 	struct buf *bp;
4232 
4233 	a = ap;
4234 	bp = a->a_bp;
4235 
4236 	/*
4237 	 * Cluster ops lock their component buffers but not the IO container.
4238 	 */
4239 	if ((bp->b_flags & B_CLUSTER) != 0)
4240 		return;
4241 
4242 	if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
4243 		if (vfs_badlock_print)
4244 			printf(
4245 			    "VOP_STRATEGY: bp is not locked but should be\n");
4246 		if (vfs_badlock_ddb)
4247 			kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
4248 	}
4249 #endif
4250 }
4251 
4252 void
4253 vop_lock_pre(void *ap)
4254 {
4255 #ifdef DEBUG_VFS_LOCKS
4256 	struct vop_lock1_args *a = ap;
4257 
4258 	if ((a->a_flags & LK_INTERLOCK) == 0)
4259 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4260 	else
4261 		ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
4262 #endif
4263 }
4264 
4265 void
4266 vop_lock_post(void *ap, int rc)
4267 {
4268 #ifdef DEBUG_VFS_LOCKS
4269 	struct vop_lock1_args *a = ap;
4270 
4271 	ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4272 	if (rc == 0 && (a->a_flags & LK_EXCLOTHER) == 0)
4273 		ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
4274 #endif
4275 }
4276 
4277 void
4278 vop_unlock_pre(void *ap)
4279 {
4280 #ifdef DEBUG_VFS_LOCKS
4281 	struct vop_unlock_args *a = ap;
4282 
4283 	if (a->a_flags & LK_INTERLOCK)
4284 		ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
4285 	ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
4286 #endif
4287 }
4288 
4289 void
4290 vop_unlock_post(void *ap, int rc)
4291 {
4292 #ifdef DEBUG_VFS_LOCKS
4293 	struct vop_unlock_args *a = ap;
4294 
4295 	if (a->a_flags & LK_INTERLOCK)
4296 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
4297 #endif
4298 }
4299 
4300 void
4301 vop_create_post(void *ap, int rc)
4302 {
4303 	struct vop_create_args *a = ap;
4304 
4305 	if (!rc)
4306 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4307 }
4308 
4309 void
4310 vop_deleteextattr_post(void *ap, int rc)
4311 {
4312 	struct vop_deleteextattr_args *a = ap;
4313 
4314 	if (!rc)
4315 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4316 }
4317 
4318 void
4319 vop_link_post(void *ap, int rc)
4320 {
4321 	struct vop_link_args *a = ap;
4322 
4323 	if (!rc) {
4324 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
4325 		VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
4326 	}
4327 }
4328 
4329 void
4330 vop_mkdir_post(void *ap, int rc)
4331 {
4332 	struct vop_mkdir_args *a = ap;
4333 
4334 	if (!rc)
4335 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4336 }
4337 
4338 void
4339 vop_mknod_post(void *ap, int rc)
4340 {
4341 	struct vop_mknod_args *a = ap;
4342 
4343 	if (!rc)
4344 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4345 }
4346 
4347 void
4348 vop_reclaim_post(void *ap, int rc)
4349 {
4350 	struct vop_reclaim_args *a = ap;
4351 
4352 	if (!rc)
4353 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_REVOKE);
4354 }
4355 
4356 void
4357 vop_remove_post(void *ap, int rc)
4358 {
4359 	struct vop_remove_args *a = ap;
4360 
4361 	if (!rc) {
4362 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4363 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4364 	}
4365 }
4366 
4367 void
4368 vop_rename_post(void *ap, int rc)
4369 {
4370 	struct vop_rename_args *a = ap;
4371 
4372 	if (!rc) {
4373 		VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
4374 		VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
4375 		VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
4376 		if (a->a_tvp)
4377 			VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
4378 	}
4379 	if (a->a_tdvp != a->a_fdvp)
4380 		vdrop(a->a_fdvp);
4381 	if (a->a_tvp != a->a_fvp)
4382 		vdrop(a->a_fvp);
4383 	vdrop(a->a_tdvp);
4384 	if (a->a_tvp)
4385 		vdrop(a->a_tvp);
4386 }
4387 
4388 void
4389 vop_rmdir_post(void *ap, int rc)
4390 {
4391 	struct vop_rmdir_args *a = ap;
4392 
4393 	if (!rc) {
4394 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4395 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4396 	}
4397 }
4398 
4399 void
4400 vop_setattr_post(void *ap, int rc)
4401 {
4402 	struct vop_setattr_args *a = ap;
4403 
4404 	if (!rc)
4405 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4406 }
4407 
4408 void
4409 vop_setextattr_post(void *ap, int rc)
4410 {
4411 	struct vop_setextattr_args *a = ap;
4412 
4413 	if (!rc)
4414 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4415 }
4416 
4417 void
4418 vop_symlink_post(void *ap, int rc)
4419 {
4420 	struct vop_symlink_args *a = ap;
4421 
4422 	if (!rc)
4423 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4424 }
4425 
4426 static struct knlist fs_knlist;
4427 
4428 static void
4429 vfs_event_init(void *arg)
4430 {
4431 	knlist_init_mtx(&fs_knlist, NULL);
4432 }
4433 /* XXX - correct order? */
4434 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4435 
4436 void
4437 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
4438 {
4439 
4440 	KNOTE_UNLOCKED(&fs_knlist, event);
4441 }
4442 
4443 static int	filt_fsattach(struct knote *kn);
4444 static void	filt_fsdetach(struct knote *kn);
4445 static int	filt_fsevent(struct knote *kn, long hint);
4446 
4447 struct filterops fs_filtops = {
4448 	.f_isfd = 0,
4449 	.f_attach = filt_fsattach,
4450 	.f_detach = filt_fsdetach,
4451 	.f_event = filt_fsevent
4452 };
4453 
4454 static int
4455 filt_fsattach(struct knote *kn)
4456 {
4457 
4458 	kn->kn_flags |= EV_CLEAR;
4459 	knlist_add(&fs_knlist, kn, 0);
4460 	return (0);
4461 }
4462 
4463 static void
4464 filt_fsdetach(struct knote *kn)
4465 {
4466 
4467 	knlist_remove(&fs_knlist, kn, 0);
4468 }
4469 
4470 static int
4471 filt_fsevent(struct knote *kn, long hint)
4472 {
4473 
4474 	kn->kn_fflags |= hint;
4475 	return (kn->kn_fflags != 0);
4476 }
4477 
4478 static int
4479 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4480 {
4481 	struct vfsidctl vc;
4482 	int error;
4483 	struct mount *mp;
4484 
4485 	error = SYSCTL_IN(req, &vc, sizeof(vc));
4486 	if (error)
4487 		return (error);
4488 	if (vc.vc_vers != VFS_CTL_VERS1)
4489 		return (EINVAL);
4490 	mp = vfs_getvfs(&vc.vc_fsid);
4491 	if (mp == NULL)
4492 		return (ENOENT);
4493 	/* ensure that a specific sysctl goes to the right filesystem. */
4494 	if (strcmp(vc.vc_fstypename, "*") != 0 &&
4495 	    strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4496 		vfs_rel(mp);
4497 		return (EINVAL);
4498 	}
4499 	VCTLTOREQ(&vc, req);
4500 	error = VFS_SYSCTL(mp, vc.vc_op, req);
4501 	vfs_rel(mp);
4502 	return (error);
4503 }
4504 
4505 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
4506     NULL, 0, sysctl_vfs_ctl, "",
4507     "Sysctl by fsid");
4508 
4509 /*
4510  * Function to initialize a va_filerev field sensibly.
4511  * XXX: Wouldn't a random number make a lot more sense ??
4512  */
4513 u_quad_t
4514 init_va_filerev(void)
4515 {
4516 	struct bintime bt;
4517 
4518 	getbinuptime(&bt);
4519 	return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4520 }
4521 
4522 static int	filt_vfsread(struct knote *kn, long hint);
4523 static int	filt_vfswrite(struct knote *kn, long hint);
4524 static int	filt_vfsvnode(struct knote *kn, long hint);
4525 static void	filt_vfsdetach(struct knote *kn);
4526 static struct filterops vfsread_filtops = {
4527 	.f_isfd = 1,
4528 	.f_detach = filt_vfsdetach,
4529 	.f_event = filt_vfsread
4530 };
4531 static struct filterops vfswrite_filtops = {
4532 	.f_isfd = 1,
4533 	.f_detach = filt_vfsdetach,
4534 	.f_event = filt_vfswrite
4535 };
4536 static struct filterops vfsvnode_filtops = {
4537 	.f_isfd = 1,
4538 	.f_detach = filt_vfsdetach,
4539 	.f_event = filt_vfsvnode
4540 };
4541 
4542 static void
4543 vfs_knllock(void *arg)
4544 {
4545 	struct vnode *vp = arg;
4546 
4547 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4548 }
4549 
4550 static void
4551 vfs_knlunlock(void *arg)
4552 {
4553 	struct vnode *vp = arg;
4554 
4555 	VOP_UNLOCK(vp, 0);
4556 }
4557 
4558 static void
4559 vfs_knl_assert_locked(void *arg)
4560 {
4561 #ifdef DEBUG_VFS_LOCKS
4562 	struct vnode *vp = arg;
4563 
4564 	ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4565 #endif
4566 }
4567 
4568 static void
4569 vfs_knl_assert_unlocked(void *arg)
4570 {
4571 #ifdef DEBUG_VFS_LOCKS
4572 	struct vnode *vp = arg;
4573 
4574 	ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4575 #endif
4576 }
4577 
4578 int
4579 vfs_kqfilter(struct vop_kqfilter_args *ap)
4580 {
4581 	struct vnode *vp = ap->a_vp;
4582 	struct knote *kn = ap->a_kn;
4583 	struct knlist *knl;
4584 
4585 	switch (kn->kn_filter) {
4586 	case EVFILT_READ:
4587 		kn->kn_fop = &vfsread_filtops;
4588 		break;
4589 	case EVFILT_WRITE:
4590 		kn->kn_fop = &vfswrite_filtops;
4591 		break;
4592 	case EVFILT_VNODE:
4593 		kn->kn_fop = &vfsvnode_filtops;
4594 		break;
4595 	default:
4596 		return (EINVAL);
4597 	}
4598 
4599 	kn->kn_hook = (caddr_t)vp;
4600 
4601 	v_addpollinfo(vp);
4602 	if (vp->v_pollinfo == NULL)
4603 		return (ENOMEM);
4604 	knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4605 	vhold(vp);
4606 	knlist_add(knl, kn, 0);
4607 
4608 	return (0);
4609 }
4610 
4611 /*
4612  * Detach knote from vnode
4613  */
4614 static void
4615 filt_vfsdetach(struct knote *kn)
4616 {
4617 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4618 
4619 	KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4620 	knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4621 	vdrop(vp);
4622 }
4623 
4624 /*ARGSUSED*/
4625 static int
4626 filt_vfsread(struct knote *kn, long hint)
4627 {
4628 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4629 	struct vattr va;
4630 	int res;
4631 
4632 	/*
4633 	 * filesystem is gone, so set the EOF flag and schedule
4634 	 * the knote for deletion.
4635 	 */
4636 	if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) {
4637 		VI_LOCK(vp);
4638 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4639 		VI_UNLOCK(vp);
4640 		return (1);
4641 	}
4642 
4643 	if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4644 		return (0);
4645 
4646 	VI_LOCK(vp);
4647 	kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4648 	res = (kn->kn_sfflags & NOTE_FILE_POLL) != 0 || kn->kn_data != 0;
4649 	VI_UNLOCK(vp);
4650 	return (res);
4651 }
4652 
4653 /*ARGSUSED*/
4654 static int
4655 filt_vfswrite(struct knote *kn, long hint)
4656 {
4657 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4658 
4659 	VI_LOCK(vp);
4660 
4661 	/*
4662 	 * filesystem is gone, so set the EOF flag and schedule
4663 	 * the knote for deletion.
4664 	 */
4665 	if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD))
4666 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4667 
4668 	kn->kn_data = 0;
4669 	VI_UNLOCK(vp);
4670 	return (1);
4671 }
4672 
4673 static int
4674 filt_vfsvnode(struct knote *kn, long hint)
4675 {
4676 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4677 	int res;
4678 
4679 	VI_LOCK(vp);
4680 	if (kn->kn_sfflags & hint)
4681 		kn->kn_fflags |= hint;
4682 	if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) {
4683 		kn->kn_flags |= EV_EOF;
4684 		VI_UNLOCK(vp);
4685 		return (1);
4686 	}
4687 	res = (kn->kn_fflags != 0);
4688 	VI_UNLOCK(vp);
4689 	return (res);
4690 }
4691 
4692 int
4693 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4694 {
4695 	int error;
4696 
4697 	if (dp->d_reclen > ap->a_uio->uio_resid)
4698 		return (ENAMETOOLONG);
4699 	error = uiomove(dp, dp->d_reclen, ap->a_uio);
4700 	if (error) {
4701 		if (ap->a_ncookies != NULL) {
4702 			if (ap->a_cookies != NULL)
4703 				free(ap->a_cookies, M_TEMP);
4704 			ap->a_cookies = NULL;
4705 			*ap->a_ncookies = 0;
4706 		}
4707 		return (error);
4708 	}
4709 	if (ap->a_ncookies == NULL)
4710 		return (0);
4711 
4712 	KASSERT(ap->a_cookies,
4713 	    ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4714 
4715 	*ap->a_cookies = realloc(*ap->a_cookies,
4716 	    (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4717 	(*ap->a_cookies)[*ap->a_ncookies] = off;
4718 	return (0);
4719 }
4720 
4721 /*
4722  * Mark for update the access time of the file if the filesystem
4723  * supports VOP_MARKATIME.  This functionality is used by execve and
4724  * mmap, so we want to avoid the I/O implied by directly setting
4725  * va_atime for the sake of efficiency.
4726  */
4727 void
4728 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4729 {
4730 	struct mount *mp;
4731 
4732 	mp = vp->v_mount;
4733 	ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4734 	if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4735 		(void)VOP_MARKATIME(vp);
4736 }
4737 
4738 /*
4739  * The purpose of this routine is to remove granularity from accmode_t,
4740  * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4741  * VADMIN and VAPPEND.
4742  *
4743  * If it returns 0, the caller is supposed to continue with the usual
4744  * access checks using 'accmode' as modified by this routine.  If it
4745  * returns nonzero value, the caller is supposed to return that value
4746  * as errno.
4747  *
4748  * Note that after this routine runs, accmode may be zero.
4749  */
4750 int
4751 vfs_unixify_accmode(accmode_t *accmode)
4752 {
4753 	/*
4754 	 * There is no way to specify explicit "deny" rule using
4755 	 * file mode or POSIX.1e ACLs.
4756 	 */
4757 	if (*accmode & VEXPLICIT_DENY) {
4758 		*accmode = 0;
4759 		return (0);
4760 	}
4761 
4762 	/*
4763 	 * None of these can be translated into usual access bits.
4764 	 * Also, the common case for NFSv4 ACLs is to not contain
4765 	 * either of these bits. Caller should check for VWRITE
4766 	 * on the containing directory instead.
4767 	 */
4768 	if (*accmode & (VDELETE_CHILD | VDELETE))
4769 		return (EPERM);
4770 
4771 	if (*accmode & VADMIN_PERMS) {
4772 		*accmode &= ~VADMIN_PERMS;
4773 		*accmode |= VADMIN;
4774 	}
4775 
4776 	/*
4777 	 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4778 	 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4779 	 */
4780 	*accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
4781 
4782 	return (0);
4783 }
4784 
4785 /*
4786  * These are helper functions for filesystems to traverse all
4787  * their vnodes.  See MNT_VNODE_FOREACH_ALL() in sys/mount.h.
4788  *
4789  * This interface replaces MNT_VNODE_FOREACH.
4790  */
4791 
4792 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
4793 
4794 struct vnode *
4795 __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp)
4796 {
4797 	struct vnode *vp;
4798 
4799 	if (should_yield())
4800 		kern_yield(PRI_USER);
4801 	MNT_ILOCK(mp);
4802 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4803 	vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
4804 	while (vp != NULL && (vp->v_type == VMARKER ||
4805 	    (vp->v_iflag & VI_DOOMED) != 0))
4806 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
4807 
4808 	/* Check if we are done */
4809 	if (vp == NULL) {
4810 		__mnt_vnode_markerfree_all(mvp, mp);
4811 		/* MNT_IUNLOCK(mp); -- done in above function */
4812 		mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
4813 		return (NULL);
4814 	}
4815 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4816 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4817 	VI_LOCK(vp);
4818 	MNT_IUNLOCK(mp);
4819 	return (vp);
4820 }
4821 
4822 struct vnode *
4823 __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp)
4824 {
4825 	struct vnode *vp;
4826 
4827 	*mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4828 	MNT_ILOCK(mp);
4829 	MNT_REF(mp);
4830 	(*mvp)->v_type = VMARKER;
4831 
4832 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
4833 	while (vp != NULL && (vp->v_type == VMARKER ||
4834 	    (vp->v_iflag & VI_DOOMED) != 0))
4835 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
4836 
4837 	/* Check if we are done */
4838 	if (vp == NULL) {
4839 		MNT_REL(mp);
4840 		MNT_IUNLOCK(mp);
4841 		free(*mvp, M_VNODE_MARKER);
4842 		*mvp = NULL;
4843 		return (NULL);
4844 	}
4845 	(*mvp)->v_mount = mp;
4846 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4847 	VI_LOCK(vp);
4848 	MNT_IUNLOCK(mp);
4849 	return (vp);
4850 }
4851 
4852 
4853 void
4854 __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp)
4855 {
4856 
4857 	if (*mvp == NULL) {
4858 		MNT_IUNLOCK(mp);
4859 		return;
4860 	}
4861 
4862 	mtx_assert(MNT_MTX(mp), MA_OWNED);
4863 
4864 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4865 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4866 	MNT_REL(mp);
4867 	MNT_IUNLOCK(mp);
4868 	free(*mvp, M_VNODE_MARKER);
4869 	*mvp = NULL;
4870 }
4871 
4872 /*
4873  * These are helper functions for filesystems to traverse their
4874  * active vnodes.  See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h
4875  */
4876 static void
4877 mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4878 {
4879 
4880 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4881 
4882 	MNT_ILOCK(mp);
4883 	MNT_REL(mp);
4884 	MNT_IUNLOCK(mp);
4885 	free(*mvp, M_VNODE_MARKER);
4886 	*mvp = NULL;
4887 }
4888 
4889 static struct vnode *
4890 mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4891 {
4892 	struct vnode *vp, *nvp;
4893 
4894 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
4895 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4896 restart:
4897 	vp = TAILQ_NEXT(*mvp, v_actfreelist);
4898 	TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4899 	while (vp != NULL) {
4900 		if (vp->v_type == VMARKER) {
4901 			vp = TAILQ_NEXT(vp, v_actfreelist);
4902 			continue;
4903 		}
4904 		if (!VI_TRYLOCK(vp)) {
4905 			if (mp_ncpus == 1 || should_yield()) {
4906 				TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
4907 				mtx_unlock(&vnode_free_list_mtx);
4908 				pause("vnacti", 1);
4909 				mtx_lock(&vnode_free_list_mtx);
4910 				goto restart;
4911 			}
4912 			continue;
4913 		}
4914 		KASSERT(vp->v_type != VMARKER, ("locked marker %p", vp));
4915 		KASSERT(vp->v_mount == mp || vp->v_mount == NULL,
4916 		    ("alien vnode on the active list %p %p", vp, mp));
4917 		if (vp->v_mount == mp && (vp->v_iflag & VI_DOOMED) == 0)
4918 			break;
4919 		nvp = TAILQ_NEXT(vp, v_actfreelist);
4920 		VI_UNLOCK(vp);
4921 		vp = nvp;
4922 	}
4923 
4924 	/* Check if we are done */
4925 	if (vp == NULL) {
4926 		mtx_unlock(&vnode_free_list_mtx);
4927 		mnt_vnode_markerfree_active(mvp, mp);
4928 		return (NULL);
4929 	}
4930 	TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
4931 	mtx_unlock(&vnode_free_list_mtx);
4932 	ASSERT_VI_LOCKED(vp, "active iter");
4933 	KASSERT((vp->v_iflag & VI_ACTIVE) != 0, ("Non-active vp %p", vp));
4934 	return (vp);
4935 }
4936 
4937 struct vnode *
4938 __mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4939 {
4940 
4941 	if (should_yield())
4942 		kern_yield(PRI_USER);
4943 	mtx_lock(&vnode_free_list_mtx);
4944 	return (mnt_vnode_next_active(mvp, mp));
4945 }
4946 
4947 struct vnode *
4948 __mnt_vnode_first_active(struct vnode **mvp, struct mount *mp)
4949 {
4950 	struct vnode *vp;
4951 
4952 	*mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4953 	MNT_ILOCK(mp);
4954 	MNT_REF(mp);
4955 	MNT_IUNLOCK(mp);
4956 	(*mvp)->v_type = VMARKER;
4957 	(*mvp)->v_mount = mp;
4958 
4959 	mtx_lock(&vnode_free_list_mtx);
4960 	vp = TAILQ_FIRST(&mp->mnt_activevnodelist);
4961 	if (vp == NULL) {
4962 		mtx_unlock(&vnode_free_list_mtx);
4963 		mnt_vnode_markerfree_active(mvp, mp);
4964 		return (NULL);
4965 	}
4966 	TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
4967 	return (mnt_vnode_next_active(mvp, mp));
4968 }
4969 
4970 void
4971 __mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4972 {
4973 
4974 	if (*mvp == NULL)
4975 		return;
4976 
4977 	mtx_lock(&vnode_free_list_mtx);
4978 	TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4979 	mtx_unlock(&vnode_free_list_mtx);
4980 	mnt_vnode_markerfree_active(mvp, mp);
4981 }
4982