xref: /freebsd/sys/kern/vfs_subr.c (revision 74036c4de983981ee2b23f91e285363293aa7c3c)
1 /*-
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)vfs_subr.c	8.31 (Berkeley) 5/26/95
35  */
36 
37 /*
38  * External virtual filesystem routines
39  */
40 
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
43 
44 #include "opt_compat.h"
45 #include "opt_ddb.h"
46 #include "opt_watchdog.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/bio.h>
51 #include <sys/buf.h>
52 #include <sys/condvar.h>
53 #include <sys/conf.h>
54 #include <sys/dirent.h>
55 #include <sys/event.h>
56 #include <sys/eventhandler.h>
57 #include <sys/extattr.h>
58 #include <sys/file.h>
59 #include <sys/fcntl.h>
60 #include <sys/jail.h>
61 #include <sys/kdb.h>
62 #include <sys/kernel.h>
63 #include <sys/kthread.h>
64 #include <sys/lockf.h>
65 #include <sys/malloc.h>
66 #include <sys/mount.h>
67 #include <sys/namei.h>
68 #include <sys/pctrie.h>
69 #include <sys/priv.h>
70 #include <sys/reboot.h>
71 #include <sys/rwlock.h>
72 #include <sys/sched.h>
73 #include <sys/sleepqueue.h>
74 #include <sys/smp.h>
75 #include <sys/stat.h>
76 #include <sys/sysctl.h>
77 #include <sys/syslog.h>
78 #include <sys/vmmeter.h>
79 #include <sys/vnode.h>
80 #include <sys/watchdog.h>
81 
82 #include <machine/stdarg.h>
83 
84 #include <security/mac/mac_framework.h>
85 
86 #include <vm/vm.h>
87 #include <vm/vm_object.h>
88 #include <vm/vm_extern.h>
89 #include <vm/pmap.h>
90 #include <vm/vm_map.h>
91 #include <vm/vm_page.h>
92 #include <vm/vm_kern.h>
93 #include <vm/uma.h>
94 
95 #ifdef DDB
96 #include <ddb/ddb.h>
97 #endif
98 
99 static void	delmntque(struct vnode *vp);
100 static int	flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
101 		    int slpflag, int slptimeo);
102 static void	syncer_shutdown(void *arg, int howto);
103 static int	vtryrecycle(struct vnode *vp);
104 static void	v_incr_usecount(struct vnode *);
105 static void	v_decr_usecount(struct vnode *);
106 static void	v_decr_useonly(struct vnode *);
107 static void	v_upgrade_usecount(struct vnode *);
108 static void	vnlru_free(int);
109 static void	vgonel(struct vnode *);
110 static void	vfs_knllock(void *arg);
111 static void	vfs_knlunlock(void *arg);
112 static void	vfs_knl_assert_locked(void *arg);
113 static void	vfs_knl_assert_unlocked(void *arg);
114 static void	destroy_vpollinfo(struct vpollinfo *vi);
115 
116 /*
117  * Number of vnodes in existence.  Increased whenever getnewvnode()
118  * allocates a new vnode, decreased in vdropl() for VI_DOOMED vnode.
119  */
120 static unsigned long	numvnodes;
121 
122 SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
123     "Number of vnodes in existence");
124 
125 /*
126  * Conversion tables for conversion from vnode types to inode formats
127  * and back.
128  */
129 enum vtype iftovt_tab[16] = {
130 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
131 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
132 };
133 int vttoif_tab[10] = {
134 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
135 	S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
136 };
137 
138 /*
139  * List of vnodes that are ready for recycling.
140  */
141 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
142 
143 /*
144  * Free vnode target.  Free vnodes may simply be files which have been stat'd
145  * but not read.  This is somewhat common, and a small cache of such files
146  * should be kept to avoid recreation costs.
147  */
148 static u_long wantfreevnodes;
149 SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
150 /* Number of vnodes in the free list. */
151 static u_long freevnodes;
152 SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
153     "Number of vnodes in the free list");
154 
155 static int vlru_allow_cache_src;
156 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
157     &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
158 
159 /*
160  * Various variables used for debugging the new implementation of
161  * reassignbuf().
162  * XXX these are probably of (very) limited utility now.
163  */
164 static int reassignbufcalls;
165 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
166     "Number of calls to reassignbuf");
167 
168 /*
169  * Cache for the mount type id assigned to NFS.  This is used for
170  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
171  */
172 int	nfs_mount_type = -1;
173 
174 /* To keep more than one thread at a time from running vfs_getnewfsid */
175 static struct mtx mntid_mtx;
176 
177 /*
178  * Lock for any access to the following:
179  *	vnode_free_list
180  *	numvnodes
181  *	freevnodes
182  */
183 static struct mtx vnode_free_list_mtx;
184 
185 /* Publicly exported FS */
186 struct nfs_public nfs_pub;
187 
188 static uma_zone_t buf_trie_zone;
189 
190 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
191 static uma_zone_t vnode_zone;
192 static uma_zone_t vnodepoll_zone;
193 
194 /*
195  * The workitem queue.
196  *
197  * It is useful to delay writes of file data and filesystem metadata
198  * for tens of seconds so that quickly created and deleted files need
199  * not waste disk bandwidth being created and removed. To realize this,
200  * we append vnodes to a "workitem" queue. When running with a soft
201  * updates implementation, most pending metadata dependencies should
202  * not wait for more than a few seconds. Thus, mounted on block devices
203  * are delayed only about a half the time that file data is delayed.
204  * Similarly, directory updates are more critical, so are only delayed
205  * about a third the time that file data is delayed. Thus, there are
206  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
207  * one each second (driven off the filesystem syncer process). The
208  * syncer_delayno variable indicates the next queue that is to be processed.
209  * Items that need to be processed soon are placed in this queue:
210  *
211  *	syncer_workitem_pending[syncer_delayno]
212  *
213  * A delay of fifteen seconds is done by placing the request fifteen
214  * entries later in the queue:
215  *
216  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
217  *
218  */
219 static int syncer_delayno;
220 static long syncer_mask;
221 LIST_HEAD(synclist, bufobj);
222 static struct synclist *syncer_workitem_pending;
223 /*
224  * The sync_mtx protects:
225  *	bo->bo_synclist
226  *	sync_vnode_count
227  *	syncer_delayno
228  *	syncer_state
229  *	syncer_workitem_pending
230  *	syncer_worklist_len
231  *	rushjob
232  */
233 static struct mtx sync_mtx;
234 static struct cv sync_wakeup;
235 
236 #define SYNCER_MAXDELAY		32
237 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
238 static int syncdelay = 30;		/* max time to delay syncing data */
239 static int filedelay = 30;		/* time to delay syncing files */
240 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
241     "Time to delay syncing files (in seconds)");
242 static int dirdelay = 29;		/* time to delay syncing directories */
243 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
244     "Time to delay syncing directories (in seconds)");
245 static int metadelay = 28;		/* time to delay syncing metadata */
246 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
247     "Time to delay syncing metadata (in seconds)");
248 static int rushjob;		/* number of slots to run ASAP */
249 static int stat_rush_requests;	/* number of times I/O speeded up */
250 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
251     "Number of times I/O speeded up (rush requests)");
252 
253 /*
254  * When shutting down the syncer, run it at four times normal speed.
255  */
256 #define SYNCER_SHUTDOWN_SPEEDUP		4
257 static int sync_vnode_count;
258 static int syncer_worklist_len;
259 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
260     syncer_state;
261 
262 /*
263  * Number of vnodes we want to exist at any one time.  This is mostly used
264  * to size hash tables in vnode-related code.  It is normally not used in
265  * getnewvnode(), as wantfreevnodes is normally nonzero.)
266  *
267  * XXX desiredvnodes is historical cruft and should not exist.
268  */
269 int desiredvnodes;
270 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
271     &desiredvnodes, 0, "Maximum number of vnodes");
272 SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
273     &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
274 static int vnlru_nowhere;
275 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
276     &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
277 
278 /*
279  * Macros to control when a vnode is freed and recycled.  All require
280  * the vnode interlock.
281  */
282 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
283 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
284 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
285 
286 /* Shift count for (uintptr_t)vp to initialize vp->v_hash. */
287 static int vnsz2log;
288 
289 /*
290  * Support for the bufobj clean & dirty pctrie.
291  */
292 static void *
293 buf_trie_alloc(struct pctrie *ptree)
294 {
295 
296 	return uma_zalloc(buf_trie_zone, M_NOWAIT);
297 }
298 
299 static void
300 buf_trie_free(struct pctrie *ptree, void *node)
301 {
302 
303 	uma_zfree(buf_trie_zone, node);
304 }
305 PCTRIE_DEFINE(BUF, buf, b_lblkno, buf_trie_alloc, buf_trie_free);
306 
307 /*
308  * Initialize the vnode management data structures.
309  *
310  * Reevaluate the following cap on the number of vnodes after the physical
311  * memory size exceeds 512GB.  In the limit, as the physical memory size
312  * grows, the ratio of physical pages to vnodes approaches sixteen to one.
313  */
314 #ifndef	MAXVNODES_MAX
315 #define	MAXVNODES_MAX	(512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
316 #endif
317 static void
318 vntblinit(void *dummy __unused)
319 {
320 	u_int i;
321 	int physvnodes, virtvnodes;
322 
323 	/*
324 	 * Desiredvnodes is a function of the physical memory size and the
325 	 * kernel's heap size.  Generally speaking, it scales with the
326 	 * physical memory size.  The ratio of desiredvnodes to physical pages
327 	 * is one to four until desiredvnodes exceeds 98,304.  Thereafter, the
328 	 * marginal ratio of desiredvnodes to physical pages is one to
329 	 * sixteen.  However, desiredvnodes is limited by the kernel's heap
330 	 * size.  The memory required by desiredvnodes vnodes and vm objects
331 	 * may not exceed one seventh of the kernel's heap size.
332 	 */
333 	physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
334 	    cnt.v_page_count) / 16;
335 	virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
336 	    sizeof(struct vnode)));
337 	desiredvnodes = min(physvnodes, virtvnodes);
338 	if (desiredvnodes > MAXVNODES_MAX) {
339 		if (bootverbose)
340 			printf("Reducing kern.maxvnodes %d -> %d\n",
341 			    desiredvnodes, MAXVNODES_MAX);
342 		desiredvnodes = MAXVNODES_MAX;
343 	}
344 	wantfreevnodes = desiredvnodes / 4;
345 	mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
346 	TAILQ_INIT(&vnode_free_list);
347 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
348 	vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
349 	    NULL, NULL, UMA_ALIGN_PTR, 0);
350 	vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
351 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
352 	/*
353 	 * Preallocate enough nodes to support one-per buf so that
354 	 * we can not fail an insert.  reassignbuf() callers can not
355 	 * tolerate the insertion failure.
356 	 */
357 	buf_trie_zone = uma_zcreate("BUF TRIE", pctrie_node_size(),
358 	    NULL, NULL, pctrie_zone_init, NULL, UMA_ALIGN_PTR,
359 	    UMA_ZONE_NOFREE | UMA_ZONE_VM);
360 	uma_prealloc(buf_trie_zone, nbuf);
361 	/*
362 	 * Initialize the filesystem syncer.
363 	 */
364 	syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
365 	    &syncer_mask);
366 	syncer_maxdelay = syncer_mask + 1;
367 	mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
368 	cv_init(&sync_wakeup, "syncer");
369 	for (i = 1; i <= sizeof(struct vnode); i <<= 1)
370 		vnsz2log++;
371 	vnsz2log--;
372 }
373 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
374 
375 
376 /*
377  * Mark a mount point as busy. Used to synchronize access and to delay
378  * unmounting. Eventually, mountlist_mtx is not released on failure.
379  *
380  * vfs_busy() is a custom lock, it can block the caller.
381  * vfs_busy() only sleeps if the unmount is active on the mount point.
382  * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
383  * vnode belonging to mp.
384  *
385  * Lookup uses vfs_busy() to traverse mount points.
386  * root fs			var fs
387  * / vnode lock		A	/ vnode lock (/var)		D
388  * /var vnode lock	B	/log vnode lock(/var/log)	E
389  * vfs_busy lock	C	vfs_busy lock			F
390  *
391  * Within each file system, the lock order is C->A->B and F->D->E.
392  *
393  * When traversing across mounts, the system follows that lock order:
394  *
395  *        C->A->B
396  *              |
397  *              +->F->D->E
398  *
399  * The lookup() process for namei("/var") illustrates the process:
400  *  VOP_LOOKUP() obtains B while A is held
401  *  vfs_busy() obtains a shared lock on F while A and B are held
402  *  vput() releases lock on B
403  *  vput() releases lock on A
404  *  VFS_ROOT() obtains lock on D while shared lock on F is held
405  *  vfs_unbusy() releases shared lock on F
406  *  vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
407  *    Attempt to lock A (instead of vp_crossmp) while D is held would
408  *    violate the global order, causing deadlocks.
409  *
410  * dounmount() locks B while F is drained.
411  */
412 int
413 vfs_busy(struct mount *mp, int flags)
414 {
415 
416 	MPASS((flags & ~MBF_MASK) == 0);
417 	CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
418 
419 	MNT_ILOCK(mp);
420 	MNT_REF(mp);
421 	/*
422 	 * If mount point is currenly being unmounted, sleep until the
423 	 * mount point fate is decided.  If thread doing the unmounting fails,
424 	 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
425 	 * that this mount point has survived the unmount attempt and vfs_busy
426 	 * should retry.  Otherwise the unmounter thread will set MNTK_REFEXPIRE
427 	 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
428 	 * about to be really destroyed.  vfs_busy needs to release its
429 	 * reference on the mount point in this case and return with ENOENT,
430 	 * telling the caller that mount mount it tried to busy is no longer
431 	 * valid.
432 	 */
433 	while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
434 		if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
435 			MNT_REL(mp);
436 			MNT_IUNLOCK(mp);
437 			CTR1(KTR_VFS, "%s: failed busying before sleeping",
438 			    __func__);
439 			return (ENOENT);
440 		}
441 		if (flags & MBF_MNTLSTLOCK)
442 			mtx_unlock(&mountlist_mtx);
443 		mp->mnt_kern_flag |= MNTK_MWAIT;
444 		msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
445 		if (flags & MBF_MNTLSTLOCK)
446 			mtx_lock(&mountlist_mtx);
447 		MNT_ILOCK(mp);
448 	}
449 	if (flags & MBF_MNTLSTLOCK)
450 		mtx_unlock(&mountlist_mtx);
451 	mp->mnt_lockref++;
452 	MNT_IUNLOCK(mp);
453 	return (0);
454 }
455 
456 /*
457  * Free a busy filesystem.
458  */
459 void
460 vfs_unbusy(struct mount *mp)
461 {
462 
463 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
464 	MNT_ILOCK(mp);
465 	MNT_REL(mp);
466 	KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
467 	mp->mnt_lockref--;
468 	if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
469 		MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
470 		CTR1(KTR_VFS, "%s: waking up waiters", __func__);
471 		mp->mnt_kern_flag &= ~MNTK_DRAINING;
472 		wakeup(&mp->mnt_lockref);
473 	}
474 	MNT_IUNLOCK(mp);
475 }
476 
477 /*
478  * Lookup a mount point by filesystem identifier.
479  */
480 struct mount *
481 vfs_getvfs(fsid_t *fsid)
482 {
483 	struct mount *mp;
484 
485 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
486 	mtx_lock(&mountlist_mtx);
487 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
488 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
489 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
490 			vfs_ref(mp);
491 			mtx_unlock(&mountlist_mtx);
492 			return (mp);
493 		}
494 	}
495 	mtx_unlock(&mountlist_mtx);
496 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
497 	return ((struct mount *) 0);
498 }
499 
500 /*
501  * Lookup a mount point by filesystem identifier, busying it before
502  * returning.
503  */
504 struct mount *
505 vfs_busyfs(fsid_t *fsid)
506 {
507 	struct mount *mp;
508 	int error;
509 
510 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
511 	mtx_lock(&mountlist_mtx);
512 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
513 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
514 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
515 			error = vfs_busy(mp, MBF_MNTLSTLOCK);
516 			if (error) {
517 				mtx_unlock(&mountlist_mtx);
518 				return (NULL);
519 			}
520 			return (mp);
521 		}
522 	}
523 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
524 	mtx_unlock(&mountlist_mtx);
525 	return ((struct mount *) 0);
526 }
527 
528 /*
529  * Check if a user can access privileged mount options.
530  */
531 int
532 vfs_suser(struct mount *mp, struct thread *td)
533 {
534 	int error;
535 
536 	/*
537 	 * If the thread is jailed, but this is not a jail-friendly file
538 	 * system, deny immediately.
539 	 */
540 	if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
541 		return (EPERM);
542 
543 	/*
544 	 * If the file system was mounted outside the jail of the calling
545 	 * thread, deny immediately.
546 	 */
547 	if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
548 		return (EPERM);
549 
550 	/*
551 	 * If file system supports delegated administration, we don't check
552 	 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
553 	 * by the file system itself.
554 	 * If this is not the user that did original mount, we check for
555 	 * the PRIV_VFS_MOUNT_OWNER privilege.
556 	 */
557 	if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
558 	    mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
559 		if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
560 			return (error);
561 	}
562 	return (0);
563 }
564 
565 /*
566  * Get a new unique fsid.  Try to make its val[0] unique, since this value
567  * will be used to create fake device numbers for stat().  Also try (but
568  * not so hard) make its val[0] unique mod 2^16, since some emulators only
569  * support 16-bit device numbers.  We end up with unique val[0]'s for the
570  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
571  *
572  * Keep in mind that several mounts may be running in parallel.  Starting
573  * the search one past where the previous search terminated is both a
574  * micro-optimization and a defense against returning the same fsid to
575  * different mounts.
576  */
577 void
578 vfs_getnewfsid(struct mount *mp)
579 {
580 	static uint16_t mntid_base;
581 	struct mount *nmp;
582 	fsid_t tfsid;
583 	int mtype;
584 
585 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
586 	mtx_lock(&mntid_mtx);
587 	mtype = mp->mnt_vfc->vfc_typenum;
588 	tfsid.val[1] = mtype;
589 	mtype = (mtype & 0xFF) << 24;
590 	for (;;) {
591 		tfsid.val[0] = makedev(255,
592 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
593 		mntid_base++;
594 		if ((nmp = vfs_getvfs(&tfsid)) == NULL)
595 			break;
596 		vfs_rel(nmp);
597 	}
598 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
599 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
600 	mtx_unlock(&mntid_mtx);
601 }
602 
603 /*
604  * Knob to control the precision of file timestamps:
605  *
606  *   0 = seconds only; nanoseconds zeroed.
607  *   1 = seconds and nanoseconds, accurate within 1/HZ.
608  *   2 = seconds and nanoseconds, truncated to microseconds.
609  * >=3 = seconds and nanoseconds, maximum precision.
610  */
611 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
612 
613 static int timestamp_precision = TSP_SEC;
614 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
615     &timestamp_precision, 0, "File timestamp precision (0: seconds, "
616     "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
617     "3+: sec + ns (max. precision))");
618 
619 /*
620  * Get a current timestamp.
621  */
622 void
623 vfs_timestamp(struct timespec *tsp)
624 {
625 	struct timeval tv;
626 
627 	switch (timestamp_precision) {
628 	case TSP_SEC:
629 		tsp->tv_sec = time_second;
630 		tsp->tv_nsec = 0;
631 		break;
632 	case TSP_HZ:
633 		getnanotime(tsp);
634 		break;
635 	case TSP_USEC:
636 		microtime(&tv);
637 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
638 		break;
639 	case TSP_NSEC:
640 	default:
641 		nanotime(tsp);
642 		break;
643 	}
644 }
645 
646 /*
647  * Set vnode attributes to VNOVAL
648  */
649 void
650 vattr_null(struct vattr *vap)
651 {
652 
653 	vap->va_type = VNON;
654 	vap->va_size = VNOVAL;
655 	vap->va_bytes = VNOVAL;
656 	vap->va_mode = VNOVAL;
657 	vap->va_nlink = VNOVAL;
658 	vap->va_uid = VNOVAL;
659 	vap->va_gid = VNOVAL;
660 	vap->va_fsid = VNOVAL;
661 	vap->va_fileid = VNOVAL;
662 	vap->va_blocksize = VNOVAL;
663 	vap->va_rdev = VNOVAL;
664 	vap->va_atime.tv_sec = VNOVAL;
665 	vap->va_atime.tv_nsec = VNOVAL;
666 	vap->va_mtime.tv_sec = VNOVAL;
667 	vap->va_mtime.tv_nsec = VNOVAL;
668 	vap->va_ctime.tv_sec = VNOVAL;
669 	vap->va_ctime.tv_nsec = VNOVAL;
670 	vap->va_birthtime.tv_sec = VNOVAL;
671 	vap->va_birthtime.tv_nsec = VNOVAL;
672 	vap->va_flags = VNOVAL;
673 	vap->va_gen = VNOVAL;
674 	vap->va_vaflags = 0;
675 }
676 
677 /*
678  * This routine is called when we have too many vnodes.  It attempts
679  * to free <count> vnodes and will potentially free vnodes that still
680  * have VM backing store (VM backing store is typically the cause
681  * of a vnode blowout so we want to do this).  Therefore, this operation
682  * is not considered cheap.
683  *
684  * A number of conditions may prevent a vnode from being reclaimed.
685  * the buffer cache may have references on the vnode, a directory
686  * vnode may still have references due to the namei cache representing
687  * underlying files, or the vnode may be in active use.   It is not
688  * desireable to reuse such vnodes.  These conditions may cause the
689  * number of vnodes to reach some minimum value regardless of what
690  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
691  */
692 static int
693 vlrureclaim(struct mount *mp)
694 {
695 	struct vnode *vp;
696 	int done;
697 	int trigger;
698 	int usevnodes;
699 	int count;
700 
701 	/*
702 	 * Calculate the trigger point, don't allow user
703 	 * screwups to blow us up.   This prevents us from
704 	 * recycling vnodes with lots of resident pages.  We
705 	 * aren't trying to free memory, we are trying to
706 	 * free vnodes.
707 	 */
708 	usevnodes = desiredvnodes;
709 	if (usevnodes <= 0)
710 		usevnodes = 1;
711 	trigger = cnt.v_page_count * 2 / usevnodes;
712 	done = 0;
713 	vn_start_write(NULL, &mp, V_WAIT);
714 	MNT_ILOCK(mp);
715 	count = mp->mnt_nvnodelistsize / 10 + 1;
716 	while (count != 0) {
717 		vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
718 		while (vp != NULL && vp->v_type == VMARKER)
719 			vp = TAILQ_NEXT(vp, v_nmntvnodes);
720 		if (vp == NULL)
721 			break;
722 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
723 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
724 		--count;
725 		if (!VI_TRYLOCK(vp))
726 			goto next_iter;
727 		/*
728 		 * If it's been deconstructed already, it's still
729 		 * referenced, or it exceeds the trigger, skip it.
730 		 */
731 		if (vp->v_usecount ||
732 		    (!vlru_allow_cache_src &&
733 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
734 		    (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
735 		    vp->v_object->resident_page_count > trigger)) {
736 			VI_UNLOCK(vp);
737 			goto next_iter;
738 		}
739 		MNT_IUNLOCK(mp);
740 		vholdl(vp);
741 		if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
742 			vdrop(vp);
743 			goto next_iter_mntunlocked;
744 		}
745 		VI_LOCK(vp);
746 		/*
747 		 * v_usecount may have been bumped after VOP_LOCK() dropped
748 		 * the vnode interlock and before it was locked again.
749 		 *
750 		 * It is not necessary to recheck VI_DOOMED because it can
751 		 * only be set by another thread that holds both the vnode
752 		 * lock and vnode interlock.  If another thread has the
753 		 * vnode lock before we get to VOP_LOCK() and obtains the
754 		 * vnode interlock after VOP_LOCK() drops the vnode
755 		 * interlock, the other thread will be unable to drop the
756 		 * vnode lock before our VOP_LOCK() call fails.
757 		 */
758 		if (vp->v_usecount ||
759 		    (!vlru_allow_cache_src &&
760 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
761 		    (vp->v_object != NULL &&
762 		    vp->v_object->resident_page_count > trigger)) {
763 			VOP_UNLOCK(vp, LK_INTERLOCK);
764 			vdrop(vp);
765 			goto next_iter_mntunlocked;
766 		}
767 		KASSERT((vp->v_iflag & VI_DOOMED) == 0,
768 		    ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
769 		vgonel(vp);
770 		VOP_UNLOCK(vp, 0);
771 		vdropl(vp);
772 		done++;
773 next_iter_mntunlocked:
774 		if (!should_yield())
775 			goto relock_mnt;
776 		goto yield;
777 next_iter:
778 		if (!should_yield())
779 			continue;
780 		MNT_IUNLOCK(mp);
781 yield:
782 		kern_yield(PRI_USER);
783 relock_mnt:
784 		MNT_ILOCK(mp);
785 	}
786 	MNT_IUNLOCK(mp);
787 	vn_finished_write(mp);
788 	return done;
789 }
790 
791 /*
792  * Attempt to keep the free list at wantfreevnodes length.
793  */
794 static void
795 vnlru_free(int count)
796 {
797 	struct vnode *vp;
798 
799 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
800 	for (; count > 0; count--) {
801 		vp = TAILQ_FIRST(&vnode_free_list);
802 		/*
803 		 * The list can be modified while the free_list_mtx
804 		 * has been dropped and vp could be NULL here.
805 		 */
806 		if (!vp)
807 			break;
808 		VNASSERT(vp->v_op != NULL, vp,
809 		    ("vnlru_free: vnode already reclaimed."));
810 		KASSERT((vp->v_iflag & VI_FREE) != 0,
811 		    ("Removing vnode not on freelist"));
812 		KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
813 		    ("Mangling active vnode"));
814 		TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
815 		/*
816 		 * Don't recycle if we can't get the interlock.
817 		 */
818 		if (!VI_TRYLOCK(vp)) {
819 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
820 			continue;
821 		}
822 		VNASSERT(VCANRECYCLE(vp), vp,
823 		    ("vp inconsistent on freelist"));
824 		freevnodes--;
825 		vp->v_iflag &= ~VI_FREE;
826 		vholdl(vp);
827 		mtx_unlock(&vnode_free_list_mtx);
828 		VI_UNLOCK(vp);
829 		vtryrecycle(vp);
830 		/*
831 		 * If the recycled succeeded this vdrop will actually free
832 		 * the vnode.  If not it will simply place it back on
833 		 * the free list.
834 		 */
835 		vdrop(vp);
836 		mtx_lock(&vnode_free_list_mtx);
837 	}
838 }
839 /*
840  * Attempt to recycle vnodes in a context that is always safe to block.
841  * Calling vlrurecycle() from the bowels of filesystem code has some
842  * interesting deadlock problems.
843  */
844 static struct proc *vnlruproc;
845 static int vnlruproc_sig;
846 
847 static void
848 vnlru_proc(void)
849 {
850 	struct mount *mp, *nmp;
851 	int done;
852 	struct proc *p = vnlruproc;
853 
854 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
855 	    SHUTDOWN_PRI_FIRST);
856 
857 	for (;;) {
858 		kproc_suspend_check(p);
859 		mtx_lock(&vnode_free_list_mtx);
860 		if (freevnodes > wantfreevnodes)
861 			vnlru_free(freevnodes - wantfreevnodes);
862 		if (numvnodes <= desiredvnodes * 9 / 10) {
863 			vnlruproc_sig = 0;
864 			wakeup(&vnlruproc_sig);
865 			msleep(vnlruproc, &vnode_free_list_mtx,
866 			    PVFS|PDROP, "vlruwt", hz);
867 			continue;
868 		}
869 		mtx_unlock(&vnode_free_list_mtx);
870 		done = 0;
871 		mtx_lock(&mountlist_mtx);
872 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
873 			if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
874 				nmp = TAILQ_NEXT(mp, mnt_list);
875 				continue;
876 			}
877 			done += vlrureclaim(mp);
878 			mtx_lock(&mountlist_mtx);
879 			nmp = TAILQ_NEXT(mp, mnt_list);
880 			vfs_unbusy(mp);
881 		}
882 		mtx_unlock(&mountlist_mtx);
883 		if (done == 0) {
884 #if 0
885 			/* These messages are temporary debugging aids */
886 			if (vnlru_nowhere < 5)
887 				printf("vnlru process getting nowhere..\n");
888 			else if (vnlru_nowhere == 5)
889 				printf("vnlru process messages stopped.\n");
890 #endif
891 			vnlru_nowhere++;
892 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
893 		} else
894 			kern_yield(PRI_USER);
895 	}
896 }
897 
898 static struct kproc_desc vnlru_kp = {
899 	"vnlru",
900 	vnlru_proc,
901 	&vnlruproc
902 };
903 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
904     &vnlru_kp);
905 
906 /*
907  * Routines having to do with the management of the vnode table.
908  */
909 
910 /*
911  * Try to recycle a freed vnode.  We abort if anyone picks up a reference
912  * before we actually vgone().  This function must be called with the vnode
913  * held to prevent the vnode from being returned to the free list midway
914  * through vgone().
915  */
916 static int
917 vtryrecycle(struct vnode *vp)
918 {
919 	struct mount *vnmp;
920 
921 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
922 	VNASSERT(vp->v_holdcnt, vp,
923 	    ("vtryrecycle: Recycling vp %p without a reference.", vp));
924 	/*
925 	 * This vnode may found and locked via some other list, if so we
926 	 * can't recycle it yet.
927 	 */
928 	if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
929 		CTR2(KTR_VFS,
930 		    "%s: impossible to recycle, vp %p lock is already held",
931 		    __func__, vp);
932 		return (EWOULDBLOCK);
933 	}
934 	/*
935 	 * Don't recycle if its filesystem is being suspended.
936 	 */
937 	if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
938 		VOP_UNLOCK(vp, 0);
939 		CTR2(KTR_VFS,
940 		    "%s: impossible to recycle, cannot start the write for %p",
941 		    __func__, vp);
942 		return (EBUSY);
943 	}
944 	/*
945 	 * If we got this far, we need to acquire the interlock and see if
946 	 * anyone picked up this vnode from another list.  If not, we will
947 	 * mark it with DOOMED via vgonel() so that anyone who does find it
948 	 * will skip over it.
949 	 */
950 	VI_LOCK(vp);
951 	if (vp->v_usecount) {
952 		VOP_UNLOCK(vp, LK_INTERLOCK);
953 		vn_finished_write(vnmp);
954 		CTR2(KTR_VFS,
955 		    "%s: impossible to recycle, %p is already referenced",
956 		    __func__, vp);
957 		return (EBUSY);
958 	}
959 	if ((vp->v_iflag & VI_DOOMED) == 0)
960 		vgonel(vp);
961 	VOP_UNLOCK(vp, LK_INTERLOCK);
962 	vn_finished_write(vnmp);
963 	return (0);
964 }
965 
966 /*
967  * Wait for available vnodes.
968  */
969 static int
970 getnewvnode_wait(int suspended)
971 {
972 
973 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
974 	if (numvnodes > desiredvnodes) {
975 		if (suspended) {
976 			/*
977 			 * File system is beeing suspended, we cannot risk a
978 			 * deadlock here, so allocate new vnode anyway.
979 			 */
980 			if (freevnodes > wantfreevnodes)
981 				vnlru_free(freevnodes - wantfreevnodes);
982 			return (0);
983 		}
984 		if (vnlruproc_sig == 0) {
985 			vnlruproc_sig = 1;	/* avoid unnecessary wakeups */
986 			wakeup(vnlruproc);
987 		}
988 		msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
989 		    "vlruwk", hz);
990 	}
991 	return (numvnodes > desiredvnodes ? ENFILE : 0);
992 }
993 
994 void
995 getnewvnode_reserve(u_int count)
996 {
997 	struct thread *td;
998 
999 	td = curthread;
1000 	mtx_lock(&vnode_free_list_mtx);
1001 	while (count > 0) {
1002 		if (getnewvnode_wait(0) == 0) {
1003 			count--;
1004 			td->td_vp_reserv++;
1005 			numvnodes++;
1006 		}
1007 	}
1008 	mtx_unlock(&vnode_free_list_mtx);
1009 }
1010 
1011 void
1012 getnewvnode_drop_reserve(void)
1013 {
1014 	struct thread *td;
1015 
1016 	td = curthread;
1017 	mtx_lock(&vnode_free_list_mtx);
1018 	KASSERT(numvnodes >= td->td_vp_reserv, ("reserve too large"));
1019 	numvnodes -= td->td_vp_reserv;
1020 	mtx_unlock(&vnode_free_list_mtx);
1021 	td->td_vp_reserv = 0;
1022 }
1023 
1024 /*
1025  * Return the next vnode from the free list.
1026  */
1027 int
1028 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
1029     struct vnode **vpp)
1030 {
1031 	struct vnode *vp;
1032 	struct bufobj *bo;
1033 	struct thread *td;
1034 	int error;
1035 
1036 	CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
1037 	vp = NULL;
1038 	td = curthread;
1039 	if (td->td_vp_reserv > 0) {
1040 		td->td_vp_reserv -= 1;
1041 		goto alloc;
1042 	}
1043 	mtx_lock(&vnode_free_list_mtx);
1044 	/*
1045 	 * Lend our context to reclaim vnodes if they've exceeded the max.
1046 	 */
1047 	if (freevnodes > wantfreevnodes)
1048 		vnlru_free(1);
1049 	error = getnewvnode_wait(mp != NULL && (mp->mnt_kern_flag &
1050 	    MNTK_SUSPEND));
1051 #if 0	/* XXX Not all VFS_VGET/ffs_vget callers check returns. */
1052 	if (error != 0) {
1053 		mtx_unlock(&vnode_free_list_mtx);
1054 		return (error);
1055 	}
1056 #endif
1057 	numvnodes++;
1058 	mtx_unlock(&vnode_free_list_mtx);
1059 alloc:
1060 	vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
1061 	/*
1062 	 * Setup locks.
1063 	 */
1064 	vp->v_vnlock = &vp->v_lock;
1065 	mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
1066 	/*
1067 	 * By default, don't allow shared locks unless filesystems
1068 	 * opt-in.
1069 	 */
1070 	lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE | LK_IS_VNODE);
1071 	/*
1072 	 * Initialize bufobj.
1073 	 */
1074 	bo = &vp->v_bufobj;
1075 	bo->__bo_vnode = vp;
1076 	rw_init(BO_LOCKPTR(bo), "bufobj interlock");
1077 	bo->bo_ops = &buf_ops_bio;
1078 	bo->bo_private = vp;
1079 	TAILQ_INIT(&bo->bo_clean.bv_hd);
1080 	TAILQ_INIT(&bo->bo_dirty.bv_hd);
1081 	/*
1082 	 * Initialize namecache.
1083 	 */
1084 	LIST_INIT(&vp->v_cache_src);
1085 	TAILQ_INIT(&vp->v_cache_dst);
1086 	/*
1087 	 * Finalize various vnode identity bits.
1088 	 */
1089 	vp->v_type = VNON;
1090 	vp->v_tag = tag;
1091 	vp->v_op = vops;
1092 	v_incr_usecount(vp);
1093 	vp->v_data = NULL;
1094 #ifdef MAC
1095 	mac_vnode_init(vp);
1096 	if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1097 		mac_vnode_associate_singlelabel(mp, vp);
1098 	else if (mp == NULL && vops != &dead_vnodeops)
1099 		printf("NULL mp in getnewvnode()\n");
1100 #endif
1101 	if (mp != NULL) {
1102 		bo->bo_bsize = mp->mnt_stat.f_iosize;
1103 		if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1104 			vp->v_vflag |= VV_NOKNOTE;
1105 	}
1106 	rangelock_init(&vp->v_rl);
1107 
1108 	/*
1109 	 * For the filesystems which do not use vfs_hash_insert(),
1110 	 * still initialize v_hash to have vfs_hash_index() useful.
1111 	 * E.g., nullfs uses vfs_hash_index() on the lower vnode for
1112 	 * its own hashing.
1113 	 */
1114 	vp->v_hash = (uintptr_t)vp >> vnsz2log;
1115 
1116 	*vpp = vp;
1117 	return (0);
1118 }
1119 
1120 /*
1121  * Delete from old mount point vnode list, if on one.
1122  */
1123 static void
1124 delmntque(struct vnode *vp)
1125 {
1126 	struct mount *mp;
1127 	int active;
1128 
1129 	mp = vp->v_mount;
1130 	if (mp == NULL)
1131 		return;
1132 	MNT_ILOCK(mp);
1133 	VI_LOCK(vp);
1134 	KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize,
1135 	    ("Active vnode list size %d > Vnode list size %d",
1136 	     mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize));
1137 	active = vp->v_iflag & VI_ACTIVE;
1138 	vp->v_iflag &= ~VI_ACTIVE;
1139 	if (active) {
1140 		mtx_lock(&vnode_free_list_mtx);
1141 		TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist);
1142 		mp->mnt_activevnodelistsize--;
1143 		mtx_unlock(&vnode_free_list_mtx);
1144 	}
1145 	vp->v_mount = NULL;
1146 	VI_UNLOCK(vp);
1147 	VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1148 		("bad mount point vnode list size"));
1149 	TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1150 	mp->mnt_nvnodelistsize--;
1151 	MNT_REL(mp);
1152 	MNT_IUNLOCK(mp);
1153 }
1154 
1155 static void
1156 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1157 {
1158 
1159 	vp->v_data = NULL;
1160 	vp->v_op = &dead_vnodeops;
1161 	vgone(vp);
1162 	vput(vp);
1163 }
1164 
1165 /*
1166  * Insert into list of vnodes for the new mount point, if available.
1167  */
1168 int
1169 insmntque1(struct vnode *vp, struct mount *mp,
1170 	void (*dtr)(struct vnode *, void *), void *dtr_arg)
1171 {
1172 
1173 	KASSERT(vp->v_mount == NULL,
1174 		("insmntque: vnode already on per mount vnode list"));
1175 	VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1176 	ASSERT_VOP_ELOCKED(vp, "insmntque: non-locked vp");
1177 
1178 	/*
1179 	 * We acquire the vnode interlock early to ensure that the
1180 	 * vnode cannot be recycled by another process releasing a
1181 	 * holdcnt on it before we get it on both the vnode list
1182 	 * and the active vnode list. The mount mutex protects only
1183 	 * manipulation of the vnode list and the vnode freelist
1184 	 * mutex protects only manipulation of the active vnode list.
1185 	 * Hence the need to hold the vnode interlock throughout.
1186 	 */
1187 	MNT_ILOCK(mp);
1188 	VI_LOCK(vp);
1189 	if (((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1190 	    ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1191 	    mp->mnt_nvnodelistsize == 0)) &&
1192 	    (vp->v_vflag & VV_FORCEINSMQ) == 0) {
1193 		VI_UNLOCK(vp);
1194 		MNT_IUNLOCK(mp);
1195 		if (dtr != NULL)
1196 			dtr(vp, dtr_arg);
1197 		return (EBUSY);
1198 	}
1199 	vp->v_mount = mp;
1200 	MNT_REF(mp);
1201 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1202 	VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1203 		("neg mount point vnode list size"));
1204 	mp->mnt_nvnodelistsize++;
1205 	KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
1206 	    ("Activating already active vnode"));
1207 	vp->v_iflag |= VI_ACTIVE;
1208 	mtx_lock(&vnode_free_list_mtx);
1209 	TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
1210 	mp->mnt_activevnodelistsize++;
1211 	mtx_unlock(&vnode_free_list_mtx);
1212 	VI_UNLOCK(vp);
1213 	MNT_IUNLOCK(mp);
1214 	return (0);
1215 }
1216 
1217 int
1218 insmntque(struct vnode *vp, struct mount *mp)
1219 {
1220 
1221 	return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1222 }
1223 
1224 /*
1225  * Flush out and invalidate all buffers associated with a bufobj
1226  * Called with the underlying object locked.
1227  */
1228 int
1229 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1230 {
1231 	int error;
1232 
1233 	BO_LOCK(bo);
1234 	if (flags & V_SAVE) {
1235 		error = bufobj_wwait(bo, slpflag, slptimeo);
1236 		if (error) {
1237 			BO_UNLOCK(bo);
1238 			return (error);
1239 		}
1240 		if (bo->bo_dirty.bv_cnt > 0) {
1241 			BO_UNLOCK(bo);
1242 			if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1243 				return (error);
1244 			/*
1245 			 * XXX We could save a lock/unlock if this was only
1246 			 * enabled under INVARIANTS
1247 			 */
1248 			BO_LOCK(bo);
1249 			if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1250 				panic("vinvalbuf: dirty bufs");
1251 		}
1252 	}
1253 	/*
1254 	 * If you alter this loop please notice that interlock is dropped and
1255 	 * reacquired in flushbuflist.  Special care is needed to ensure that
1256 	 * no race conditions occur from this.
1257 	 */
1258 	do {
1259 		error = flushbuflist(&bo->bo_clean,
1260 		    flags, bo, slpflag, slptimeo);
1261 		if (error == 0 && !(flags & V_CLEANONLY))
1262 			error = flushbuflist(&bo->bo_dirty,
1263 			    flags, bo, slpflag, slptimeo);
1264 		if (error != 0 && error != EAGAIN) {
1265 			BO_UNLOCK(bo);
1266 			return (error);
1267 		}
1268 	} while (error != 0);
1269 
1270 	/*
1271 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1272 	 * have write I/O in-progress but if there is a VM object then the
1273 	 * VM object can also have read-I/O in-progress.
1274 	 */
1275 	do {
1276 		bufobj_wwait(bo, 0, 0);
1277 		BO_UNLOCK(bo);
1278 		if (bo->bo_object != NULL) {
1279 			VM_OBJECT_WLOCK(bo->bo_object);
1280 			vm_object_pip_wait(bo->bo_object, "bovlbx");
1281 			VM_OBJECT_WUNLOCK(bo->bo_object);
1282 		}
1283 		BO_LOCK(bo);
1284 	} while (bo->bo_numoutput > 0);
1285 	BO_UNLOCK(bo);
1286 
1287 	/*
1288 	 * Destroy the copy in the VM cache, too.
1289 	 */
1290 	if (bo->bo_object != NULL &&
1291 	    (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) {
1292 		VM_OBJECT_WLOCK(bo->bo_object);
1293 		vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
1294 		    OBJPR_CLEANONLY : 0);
1295 		VM_OBJECT_WUNLOCK(bo->bo_object);
1296 	}
1297 
1298 #ifdef INVARIANTS
1299 	BO_LOCK(bo);
1300 	if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 &&
1301 	    (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1302 		panic("vinvalbuf: flush failed");
1303 	BO_UNLOCK(bo);
1304 #endif
1305 	return (0);
1306 }
1307 
1308 /*
1309  * Flush out and invalidate all buffers associated with a vnode.
1310  * Called with the underlying object locked.
1311  */
1312 int
1313 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1314 {
1315 
1316 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1317 	ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1318 	return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1319 }
1320 
1321 /*
1322  * Flush out buffers on the specified list.
1323  *
1324  */
1325 static int
1326 flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1327     int slptimeo)
1328 {
1329 	struct buf *bp, *nbp;
1330 	int retval, error;
1331 	daddr_t lblkno;
1332 	b_xflags_t xflags;
1333 
1334 	ASSERT_BO_WLOCKED(bo);
1335 
1336 	retval = 0;
1337 	TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1338 		if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1339 		    ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1340 			continue;
1341 		}
1342 		lblkno = 0;
1343 		xflags = 0;
1344 		if (nbp != NULL) {
1345 			lblkno = nbp->b_lblkno;
1346 			xflags = nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN);
1347 		}
1348 		retval = EAGAIN;
1349 		error = BUF_TIMELOCK(bp,
1350 		    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo),
1351 		    "flushbuf", slpflag, slptimeo);
1352 		if (error) {
1353 			BO_LOCK(bo);
1354 			return (error != ENOLCK ? error : EAGAIN);
1355 		}
1356 		KASSERT(bp->b_bufobj == bo,
1357 		    ("bp %p wrong b_bufobj %p should be %p",
1358 		    bp, bp->b_bufobj, bo));
1359 		if (bp->b_bufobj != bo) {	/* XXX: necessary ? */
1360 			BUF_UNLOCK(bp);
1361 			BO_LOCK(bo);
1362 			return (EAGAIN);
1363 		}
1364 		/*
1365 		 * XXX Since there are no node locks for NFS, I
1366 		 * believe there is a slight chance that a delayed
1367 		 * write will occur while sleeping just above, so
1368 		 * check for it.
1369 		 */
1370 		if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1371 		    (flags & V_SAVE)) {
1372 			bremfree(bp);
1373 			bp->b_flags |= B_ASYNC;
1374 			bwrite(bp);
1375 			BO_LOCK(bo);
1376 			return (EAGAIN);	/* XXX: why not loop ? */
1377 		}
1378 		bremfree(bp);
1379 		bp->b_flags |= (B_INVAL | B_RELBUF);
1380 		bp->b_flags &= ~B_ASYNC;
1381 		brelse(bp);
1382 		BO_LOCK(bo);
1383 		if (nbp != NULL &&
1384 		    (nbp->b_bufobj != bo ||
1385 		     nbp->b_lblkno != lblkno ||
1386 		     (nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1387 			break;			/* nbp invalid */
1388 	}
1389 	return (retval);
1390 }
1391 
1392 /*
1393  * Truncate a file's buffer and pages to a specified length.  This
1394  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1395  * sync activity.
1396  */
1397 int
1398 vtruncbuf(struct vnode *vp, struct ucred *cred, off_t length, int blksize)
1399 {
1400 	struct buf *bp, *nbp;
1401 	int anyfreed;
1402 	int trunclbn;
1403 	struct bufobj *bo;
1404 
1405 	CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1406 	    vp, cred, blksize, (uintmax_t)length);
1407 
1408 	/*
1409 	 * Round up to the *next* lbn.
1410 	 */
1411 	trunclbn = (length + blksize - 1) / blksize;
1412 
1413 	ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1414 restart:
1415 	bo = &vp->v_bufobj;
1416 	BO_LOCK(bo);
1417 	anyfreed = 1;
1418 	for (;anyfreed;) {
1419 		anyfreed = 0;
1420 		TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1421 			if (bp->b_lblkno < trunclbn)
1422 				continue;
1423 			if (BUF_LOCK(bp,
1424 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1425 			    BO_LOCKPTR(bo)) == ENOLCK)
1426 				goto restart;
1427 
1428 			bremfree(bp);
1429 			bp->b_flags |= (B_INVAL | B_RELBUF);
1430 			bp->b_flags &= ~B_ASYNC;
1431 			brelse(bp);
1432 			anyfreed = 1;
1433 
1434 			BO_LOCK(bo);
1435 			if (nbp != NULL &&
1436 			    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1437 			    (nbp->b_vp != vp) ||
1438 			    (nbp->b_flags & B_DELWRI))) {
1439 				BO_UNLOCK(bo);
1440 				goto restart;
1441 			}
1442 		}
1443 
1444 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1445 			if (bp->b_lblkno < trunclbn)
1446 				continue;
1447 			if (BUF_LOCK(bp,
1448 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1449 			    BO_LOCKPTR(bo)) == ENOLCK)
1450 				goto restart;
1451 			bremfree(bp);
1452 			bp->b_flags |= (B_INVAL | B_RELBUF);
1453 			bp->b_flags &= ~B_ASYNC;
1454 			brelse(bp);
1455 			anyfreed = 1;
1456 
1457 			BO_LOCK(bo);
1458 			if (nbp != NULL &&
1459 			    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1460 			    (nbp->b_vp != vp) ||
1461 			    (nbp->b_flags & B_DELWRI) == 0)) {
1462 				BO_UNLOCK(bo);
1463 				goto restart;
1464 			}
1465 		}
1466 	}
1467 
1468 	if (length > 0) {
1469 restartsync:
1470 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1471 			if (bp->b_lblkno > 0)
1472 				continue;
1473 			/*
1474 			 * Since we hold the vnode lock this should only
1475 			 * fail if we're racing with the buf daemon.
1476 			 */
1477 			if (BUF_LOCK(bp,
1478 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1479 			    BO_LOCKPTR(bo)) == ENOLCK) {
1480 				goto restart;
1481 			}
1482 			VNASSERT((bp->b_flags & B_DELWRI), vp,
1483 			    ("buf(%p) on dirty queue without DELWRI", bp));
1484 
1485 			bremfree(bp);
1486 			bawrite(bp);
1487 			BO_LOCK(bo);
1488 			goto restartsync;
1489 		}
1490 	}
1491 
1492 	bufobj_wwait(bo, 0, 0);
1493 	BO_UNLOCK(bo);
1494 	vnode_pager_setsize(vp, length);
1495 
1496 	return (0);
1497 }
1498 
1499 static void
1500 buf_vlist_remove(struct buf *bp)
1501 {
1502 	struct bufv *bv;
1503 
1504 	KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1505 	ASSERT_BO_WLOCKED(bp->b_bufobj);
1506 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1507 	    (BX_VNDIRTY|BX_VNCLEAN),
1508 	    ("buf_vlist_remove: Buf %p is on two lists", bp));
1509 	if (bp->b_xflags & BX_VNDIRTY)
1510 		bv = &bp->b_bufobj->bo_dirty;
1511 	else
1512 		bv = &bp->b_bufobj->bo_clean;
1513 	BUF_PCTRIE_REMOVE(&bv->bv_root, bp->b_lblkno);
1514 	TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1515 	bv->bv_cnt--;
1516 	bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1517 }
1518 
1519 /*
1520  * Add the buffer to the sorted clean or dirty block list.
1521  *
1522  * NOTE: xflags is passed as a constant, optimizing this inline function!
1523  */
1524 static void
1525 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1526 {
1527 	struct bufv *bv;
1528 	struct buf *n;
1529 	int error;
1530 
1531 	ASSERT_BO_WLOCKED(bo);
1532 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1533 	    ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1534 	bp->b_xflags |= xflags;
1535 	if (xflags & BX_VNDIRTY)
1536 		bv = &bo->bo_dirty;
1537 	else
1538 		bv = &bo->bo_clean;
1539 
1540 	/*
1541 	 * Keep the list ordered.  Optimize empty list insertion.  Assume
1542 	 * we tend to grow at the tail so lookup_le should usually be cheaper
1543 	 * than _ge.
1544 	 */
1545 	if (bv->bv_cnt == 0 ||
1546 	    bp->b_lblkno > TAILQ_LAST(&bv->bv_hd, buflists)->b_lblkno)
1547 		TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1548 	else if ((n = BUF_PCTRIE_LOOKUP_LE(&bv->bv_root, bp->b_lblkno)) == NULL)
1549 		TAILQ_INSERT_HEAD(&bv->bv_hd, bp, b_bobufs);
1550 	else
1551 		TAILQ_INSERT_AFTER(&bv->bv_hd, n, bp, b_bobufs);
1552 	error = BUF_PCTRIE_INSERT(&bv->bv_root, bp);
1553 	if (error)
1554 		panic("buf_vlist_add:  Preallocated nodes insufficient.");
1555 	bv->bv_cnt++;
1556 }
1557 
1558 /*
1559  * Lookup a buffer using the splay tree.  Note that we specifically avoid
1560  * shadow buffers used in background bitmap writes.
1561  *
1562  * This code isn't quite efficient as it could be because we are maintaining
1563  * two sorted lists and do not know which list the block resides in.
1564  *
1565  * During a "make buildworld" the desired buffer is found at one of
1566  * the roots more than 60% of the time.  Thus, checking both roots
1567  * before performing either splay eliminates unnecessary splays on the
1568  * first tree splayed.
1569  */
1570 struct buf *
1571 gbincore(struct bufobj *bo, daddr_t lblkno)
1572 {
1573 	struct buf *bp;
1574 
1575 	ASSERT_BO_LOCKED(bo);
1576 	bp = BUF_PCTRIE_LOOKUP(&bo->bo_clean.bv_root, lblkno);
1577 	if (bp != NULL)
1578 		return (bp);
1579 	return BUF_PCTRIE_LOOKUP(&bo->bo_dirty.bv_root, lblkno);
1580 }
1581 
1582 /*
1583  * Associate a buffer with a vnode.
1584  */
1585 void
1586 bgetvp(struct vnode *vp, struct buf *bp)
1587 {
1588 	struct bufobj *bo;
1589 
1590 	bo = &vp->v_bufobj;
1591 	ASSERT_BO_WLOCKED(bo);
1592 	VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1593 
1594 	CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1595 	VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1596 	    ("bgetvp: bp already attached! %p", bp));
1597 
1598 	vhold(vp);
1599 	bp->b_vp = vp;
1600 	bp->b_bufobj = bo;
1601 	/*
1602 	 * Insert onto list for new vnode.
1603 	 */
1604 	buf_vlist_add(bp, bo, BX_VNCLEAN);
1605 }
1606 
1607 /*
1608  * Disassociate a buffer from a vnode.
1609  */
1610 void
1611 brelvp(struct buf *bp)
1612 {
1613 	struct bufobj *bo;
1614 	struct vnode *vp;
1615 
1616 	CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1617 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1618 
1619 	/*
1620 	 * Delete from old vnode list, if on one.
1621 	 */
1622 	vp = bp->b_vp;		/* XXX */
1623 	bo = bp->b_bufobj;
1624 	BO_LOCK(bo);
1625 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1626 		buf_vlist_remove(bp);
1627 	else
1628 		panic("brelvp: Buffer %p not on queue.", bp);
1629 	if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1630 		bo->bo_flag &= ~BO_ONWORKLST;
1631 		mtx_lock(&sync_mtx);
1632 		LIST_REMOVE(bo, bo_synclist);
1633 		syncer_worklist_len--;
1634 		mtx_unlock(&sync_mtx);
1635 	}
1636 	bp->b_vp = NULL;
1637 	bp->b_bufobj = NULL;
1638 	BO_UNLOCK(bo);
1639 	vdrop(vp);
1640 }
1641 
1642 /*
1643  * Add an item to the syncer work queue.
1644  */
1645 static void
1646 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1647 {
1648 	int slot;
1649 
1650 	ASSERT_BO_WLOCKED(bo);
1651 
1652 	mtx_lock(&sync_mtx);
1653 	if (bo->bo_flag & BO_ONWORKLST)
1654 		LIST_REMOVE(bo, bo_synclist);
1655 	else {
1656 		bo->bo_flag |= BO_ONWORKLST;
1657 		syncer_worklist_len++;
1658 	}
1659 
1660 	if (delay > syncer_maxdelay - 2)
1661 		delay = syncer_maxdelay - 2;
1662 	slot = (syncer_delayno + delay) & syncer_mask;
1663 
1664 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist);
1665 	mtx_unlock(&sync_mtx);
1666 }
1667 
1668 static int
1669 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1670 {
1671 	int error, len;
1672 
1673 	mtx_lock(&sync_mtx);
1674 	len = syncer_worklist_len - sync_vnode_count;
1675 	mtx_unlock(&sync_mtx);
1676 	error = SYSCTL_OUT(req, &len, sizeof(len));
1677 	return (error);
1678 }
1679 
1680 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1681     sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1682 
1683 static struct proc *updateproc;
1684 static void sched_sync(void);
1685 static struct kproc_desc up_kp = {
1686 	"syncer",
1687 	sched_sync,
1688 	&updateproc
1689 };
1690 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1691 
1692 static int
1693 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1694 {
1695 	struct vnode *vp;
1696 	struct mount *mp;
1697 
1698 	*bo = LIST_FIRST(slp);
1699 	if (*bo == NULL)
1700 		return (0);
1701 	vp = (*bo)->__bo_vnode;	/* XXX */
1702 	if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1703 		return (1);
1704 	/*
1705 	 * We use vhold in case the vnode does not
1706 	 * successfully sync.  vhold prevents the vnode from
1707 	 * going away when we unlock the sync_mtx so that
1708 	 * we can acquire the vnode interlock.
1709 	 */
1710 	vholdl(vp);
1711 	mtx_unlock(&sync_mtx);
1712 	VI_UNLOCK(vp);
1713 	if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1714 		vdrop(vp);
1715 		mtx_lock(&sync_mtx);
1716 		return (*bo == LIST_FIRST(slp));
1717 	}
1718 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1719 	(void) VOP_FSYNC(vp, MNT_LAZY, td);
1720 	VOP_UNLOCK(vp, 0);
1721 	vn_finished_write(mp);
1722 	BO_LOCK(*bo);
1723 	if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1724 		/*
1725 		 * Put us back on the worklist.  The worklist
1726 		 * routine will remove us from our current
1727 		 * position and then add us back in at a later
1728 		 * position.
1729 		 */
1730 		vn_syncer_add_to_worklist(*bo, syncdelay);
1731 	}
1732 	BO_UNLOCK(*bo);
1733 	vdrop(vp);
1734 	mtx_lock(&sync_mtx);
1735 	return (0);
1736 }
1737 
1738 /*
1739  * System filesystem synchronizer daemon.
1740  */
1741 static void
1742 sched_sync(void)
1743 {
1744 	struct synclist *next, *slp;
1745 	struct bufobj *bo;
1746 	long starttime;
1747 	struct thread *td = curthread;
1748 	int last_work_seen;
1749 	int net_worklist_len;
1750 	int syncer_final_iter;
1751 	int first_printf;
1752 	int error;
1753 
1754 	last_work_seen = 0;
1755 	syncer_final_iter = 0;
1756 	first_printf = 1;
1757 	syncer_state = SYNCER_RUNNING;
1758 	starttime = time_uptime;
1759 	td->td_pflags |= TDP_NORUNNINGBUF;
1760 
1761 	EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1762 	    SHUTDOWN_PRI_LAST);
1763 
1764 	mtx_lock(&sync_mtx);
1765 	for (;;) {
1766 		if (syncer_state == SYNCER_FINAL_DELAY &&
1767 		    syncer_final_iter == 0) {
1768 			mtx_unlock(&sync_mtx);
1769 			kproc_suspend_check(td->td_proc);
1770 			mtx_lock(&sync_mtx);
1771 		}
1772 		net_worklist_len = syncer_worklist_len - sync_vnode_count;
1773 		if (syncer_state != SYNCER_RUNNING &&
1774 		    starttime != time_uptime) {
1775 			if (first_printf) {
1776 				printf("\nSyncing disks, vnodes remaining...");
1777 				first_printf = 0;
1778 			}
1779 			printf("%d ", net_worklist_len);
1780 		}
1781 		starttime = time_uptime;
1782 
1783 		/*
1784 		 * Push files whose dirty time has expired.  Be careful
1785 		 * of interrupt race on slp queue.
1786 		 *
1787 		 * Skip over empty worklist slots when shutting down.
1788 		 */
1789 		do {
1790 			slp = &syncer_workitem_pending[syncer_delayno];
1791 			syncer_delayno += 1;
1792 			if (syncer_delayno == syncer_maxdelay)
1793 				syncer_delayno = 0;
1794 			next = &syncer_workitem_pending[syncer_delayno];
1795 			/*
1796 			 * If the worklist has wrapped since the
1797 			 * it was emptied of all but syncer vnodes,
1798 			 * switch to the FINAL_DELAY state and run
1799 			 * for one more second.
1800 			 */
1801 			if (syncer_state == SYNCER_SHUTTING_DOWN &&
1802 			    net_worklist_len == 0 &&
1803 			    last_work_seen == syncer_delayno) {
1804 				syncer_state = SYNCER_FINAL_DELAY;
1805 				syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1806 			}
1807 		} while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1808 		    syncer_worklist_len > 0);
1809 
1810 		/*
1811 		 * Keep track of the last time there was anything
1812 		 * on the worklist other than syncer vnodes.
1813 		 * Return to the SHUTTING_DOWN state if any
1814 		 * new work appears.
1815 		 */
1816 		if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1817 			last_work_seen = syncer_delayno;
1818 		if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1819 			syncer_state = SYNCER_SHUTTING_DOWN;
1820 		while (!LIST_EMPTY(slp)) {
1821 			error = sync_vnode(slp, &bo, td);
1822 			if (error == 1) {
1823 				LIST_REMOVE(bo, bo_synclist);
1824 				LIST_INSERT_HEAD(next, bo, bo_synclist);
1825 				continue;
1826 			}
1827 
1828 			if (first_printf == 0)
1829 				wdog_kern_pat(WD_LASTVAL);
1830 
1831 		}
1832 		if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1833 			syncer_final_iter--;
1834 		/*
1835 		 * The variable rushjob allows the kernel to speed up the
1836 		 * processing of the filesystem syncer process. A rushjob
1837 		 * value of N tells the filesystem syncer to process the next
1838 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1839 		 * is used by the soft update code to speed up the filesystem
1840 		 * syncer process when the incore state is getting so far
1841 		 * ahead of the disk that the kernel memory pool is being
1842 		 * threatened with exhaustion.
1843 		 */
1844 		if (rushjob > 0) {
1845 			rushjob -= 1;
1846 			continue;
1847 		}
1848 		/*
1849 		 * Just sleep for a short period of time between
1850 		 * iterations when shutting down to allow some I/O
1851 		 * to happen.
1852 		 *
1853 		 * If it has taken us less than a second to process the
1854 		 * current work, then wait. Otherwise start right over
1855 		 * again. We can still lose time if any single round
1856 		 * takes more than two seconds, but it does not really
1857 		 * matter as we are just trying to generally pace the
1858 		 * filesystem activity.
1859 		 */
1860 		if (syncer_state != SYNCER_RUNNING ||
1861 		    time_uptime == starttime) {
1862 			thread_lock(td);
1863 			sched_prio(td, PPAUSE);
1864 			thread_unlock(td);
1865 		}
1866 		if (syncer_state != SYNCER_RUNNING)
1867 			cv_timedwait(&sync_wakeup, &sync_mtx,
1868 			    hz / SYNCER_SHUTDOWN_SPEEDUP);
1869 		else if (time_uptime == starttime)
1870 			cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1871 	}
1872 }
1873 
1874 /*
1875  * Request the syncer daemon to speed up its work.
1876  * We never push it to speed up more than half of its
1877  * normal turn time, otherwise it could take over the cpu.
1878  */
1879 int
1880 speedup_syncer(void)
1881 {
1882 	int ret = 0;
1883 
1884 	mtx_lock(&sync_mtx);
1885 	if (rushjob < syncdelay / 2) {
1886 		rushjob += 1;
1887 		stat_rush_requests += 1;
1888 		ret = 1;
1889 	}
1890 	mtx_unlock(&sync_mtx);
1891 	cv_broadcast(&sync_wakeup);
1892 	return (ret);
1893 }
1894 
1895 /*
1896  * Tell the syncer to speed up its work and run though its work
1897  * list several times, then tell it to shut down.
1898  */
1899 static void
1900 syncer_shutdown(void *arg, int howto)
1901 {
1902 
1903 	if (howto & RB_NOSYNC)
1904 		return;
1905 	mtx_lock(&sync_mtx);
1906 	syncer_state = SYNCER_SHUTTING_DOWN;
1907 	rushjob = 0;
1908 	mtx_unlock(&sync_mtx);
1909 	cv_broadcast(&sync_wakeup);
1910 	kproc_shutdown(arg, howto);
1911 }
1912 
1913 /*
1914  * Reassign a buffer from one vnode to another.
1915  * Used to assign file specific control information
1916  * (indirect blocks) to the vnode to which they belong.
1917  */
1918 void
1919 reassignbuf(struct buf *bp)
1920 {
1921 	struct vnode *vp;
1922 	struct bufobj *bo;
1923 	int delay;
1924 #ifdef INVARIANTS
1925 	struct bufv *bv;
1926 #endif
1927 
1928 	vp = bp->b_vp;
1929 	bo = bp->b_bufobj;
1930 	++reassignbufcalls;
1931 
1932 	CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1933 	    bp, bp->b_vp, bp->b_flags);
1934 	/*
1935 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1936 	 * is not fully linked in.
1937 	 */
1938 	if (bp->b_flags & B_PAGING)
1939 		panic("cannot reassign paging buffer");
1940 
1941 	/*
1942 	 * Delete from old vnode list, if on one.
1943 	 */
1944 	BO_LOCK(bo);
1945 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1946 		buf_vlist_remove(bp);
1947 	else
1948 		panic("reassignbuf: Buffer %p not on queue.", bp);
1949 	/*
1950 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1951 	 * of clean buffers.
1952 	 */
1953 	if (bp->b_flags & B_DELWRI) {
1954 		if ((bo->bo_flag & BO_ONWORKLST) == 0) {
1955 			switch (vp->v_type) {
1956 			case VDIR:
1957 				delay = dirdelay;
1958 				break;
1959 			case VCHR:
1960 				delay = metadelay;
1961 				break;
1962 			default:
1963 				delay = filedelay;
1964 			}
1965 			vn_syncer_add_to_worklist(bo, delay);
1966 		}
1967 		buf_vlist_add(bp, bo, BX_VNDIRTY);
1968 	} else {
1969 		buf_vlist_add(bp, bo, BX_VNCLEAN);
1970 
1971 		if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1972 			mtx_lock(&sync_mtx);
1973 			LIST_REMOVE(bo, bo_synclist);
1974 			syncer_worklist_len--;
1975 			mtx_unlock(&sync_mtx);
1976 			bo->bo_flag &= ~BO_ONWORKLST;
1977 		}
1978 	}
1979 #ifdef INVARIANTS
1980 	bv = &bo->bo_clean;
1981 	bp = TAILQ_FIRST(&bv->bv_hd);
1982 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1983 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1984 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
1985 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1986 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1987 	bv = &bo->bo_dirty;
1988 	bp = TAILQ_FIRST(&bv->bv_hd);
1989 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1990 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1991 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
1992 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1993 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1994 #endif
1995 	BO_UNLOCK(bo);
1996 }
1997 
1998 /*
1999  * Increment the use and hold counts on the vnode, taking care to reference
2000  * the driver's usecount if this is a chardev.  The vholdl() will remove
2001  * the vnode from the free list if it is presently free.  Requires the
2002  * vnode interlock and returns with it held.
2003  */
2004 static void
2005 v_incr_usecount(struct vnode *vp)
2006 {
2007 
2008 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2009 	vp->v_usecount++;
2010 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2011 		dev_lock();
2012 		vp->v_rdev->si_usecount++;
2013 		dev_unlock();
2014 	}
2015 	vholdl(vp);
2016 }
2017 
2018 /*
2019  * Turn a holdcnt into a use+holdcnt such that only one call to
2020  * v_decr_usecount is needed.
2021  */
2022 static void
2023 v_upgrade_usecount(struct vnode *vp)
2024 {
2025 
2026 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2027 	vp->v_usecount++;
2028 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2029 		dev_lock();
2030 		vp->v_rdev->si_usecount++;
2031 		dev_unlock();
2032 	}
2033 }
2034 
2035 /*
2036  * Decrement the vnode use and hold count along with the driver's usecount
2037  * if this is a chardev.  The vdropl() below releases the vnode interlock
2038  * as it may free the vnode.
2039  */
2040 static void
2041 v_decr_usecount(struct vnode *vp)
2042 {
2043 
2044 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2045 	VNASSERT(vp->v_usecount > 0, vp,
2046 	    ("v_decr_usecount: negative usecount"));
2047 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2048 	vp->v_usecount--;
2049 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2050 		dev_lock();
2051 		vp->v_rdev->si_usecount--;
2052 		dev_unlock();
2053 	}
2054 	vdropl(vp);
2055 }
2056 
2057 /*
2058  * Decrement only the use count and driver use count.  This is intended to
2059  * be paired with a follow on vdropl() to release the remaining hold count.
2060  * In this way we may vgone() a vnode with a 0 usecount without risk of
2061  * having it end up on a free list because the hold count is kept above 0.
2062  */
2063 static void
2064 v_decr_useonly(struct vnode *vp)
2065 {
2066 
2067 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2068 	VNASSERT(vp->v_usecount > 0, vp,
2069 	    ("v_decr_useonly: negative usecount"));
2070 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2071 	vp->v_usecount--;
2072 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2073 		dev_lock();
2074 		vp->v_rdev->si_usecount--;
2075 		dev_unlock();
2076 	}
2077 }
2078 
2079 /*
2080  * Grab a particular vnode from the free list, increment its
2081  * reference count and lock it.  VI_DOOMED is set if the vnode
2082  * is being destroyed.  Only callers who specify LK_RETRY will
2083  * see doomed vnodes.  If inactive processing was delayed in
2084  * vput try to do it here.
2085  */
2086 int
2087 vget(struct vnode *vp, int flags, struct thread *td)
2088 {
2089 	int error;
2090 
2091 	error = 0;
2092 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2093 	    ("vget: invalid lock operation"));
2094 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2095 
2096 	if ((flags & LK_INTERLOCK) == 0)
2097 		VI_LOCK(vp);
2098 	vholdl(vp);
2099 	if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2100 		vdrop(vp);
2101 		CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2102 		    vp);
2103 		return (error);
2104 	}
2105 	if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2106 		panic("vget: vn_lock failed to return ENOENT\n");
2107 	VI_LOCK(vp);
2108 	/* Upgrade our holdcnt to a usecount. */
2109 	v_upgrade_usecount(vp);
2110 	/*
2111 	 * We don't guarantee that any particular close will
2112 	 * trigger inactive processing so just make a best effort
2113 	 * here at preventing a reference to a removed file.  If
2114 	 * we don't succeed no harm is done.
2115 	 */
2116 	if (vp->v_iflag & VI_OWEINACT) {
2117 		if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2118 		    (flags & LK_NOWAIT) == 0)
2119 			vinactive(vp, td);
2120 		vp->v_iflag &= ~VI_OWEINACT;
2121 	}
2122 	VI_UNLOCK(vp);
2123 	return (0);
2124 }
2125 
2126 /*
2127  * Increase the reference count of a vnode.
2128  */
2129 void
2130 vref(struct vnode *vp)
2131 {
2132 
2133 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2134 	VI_LOCK(vp);
2135 	v_incr_usecount(vp);
2136 	VI_UNLOCK(vp);
2137 }
2138 
2139 /*
2140  * Return reference count of a vnode.
2141  *
2142  * The results of this call are only guaranteed when some mechanism other
2143  * than the VI lock is used to stop other processes from gaining references
2144  * to the vnode.  This may be the case if the caller holds the only reference.
2145  * This is also useful when stale data is acceptable as race conditions may
2146  * be accounted for by some other means.
2147  */
2148 int
2149 vrefcnt(struct vnode *vp)
2150 {
2151 	int usecnt;
2152 
2153 	VI_LOCK(vp);
2154 	usecnt = vp->v_usecount;
2155 	VI_UNLOCK(vp);
2156 
2157 	return (usecnt);
2158 }
2159 
2160 #define	VPUTX_VRELE	1
2161 #define	VPUTX_VPUT	2
2162 #define	VPUTX_VUNREF	3
2163 
2164 static void
2165 vputx(struct vnode *vp, int func)
2166 {
2167 	int error;
2168 
2169 	KASSERT(vp != NULL, ("vputx: null vp"));
2170 	if (func == VPUTX_VUNREF)
2171 		ASSERT_VOP_LOCKED(vp, "vunref");
2172 	else if (func == VPUTX_VPUT)
2173 		ASSERT_VOP_LOCKED(vp, "vput");
2174 	else
2175 		KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2176 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2177 	VI_LOCK(vp);
2178 
2179 	/* Skip this v_writecount check if we're going to panic below. */
2180 	VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2181 	    ("vputx: missed vn_close"));
2182 	error = 0;
2183 
2184 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2185 	    vp->v_usecount == 1)) {
2186 		if (func == VPUTX_VPUT)
2187 			VOP_UNLOCK(vp, 0);
2188 		v_decr_usecount(vp);
2189 		return;
2190 	}
2191 
2192 	if (vp->v_usecount != 1) {
2193 		vprint("vputx: negative ref count", vp);
2194 		panic("vputx: negative ref cnt");
2195 	}
2196 	CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2197 	/*
2198 	 * We want to hold the vnode until the inactive finishes to
2199 	 * prevent vgone() races.  We drop the use count here and the
2200 	 * hold count below when we're done.
2201 	 */
2202 	v_decr_useonly(vp);
2203 	/*
2204 	 * We must call VOP_INACTIVE with the node locked. Mark
2205 	 * as VI_DOINGINACT to avoid recursion.
2206 	 */
2207 	vp->v_iflag |= VI_OWEINACT;
2208 	switch (func) {
2209 	case VPUTX_VRELE:
2210 		error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2211 		VI_LOCK(vp);
2212 		break;
2213 	case VPUTX_VPUT:
2214 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2215 			error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
2216 			    LK_NOWAIT);
2217 			VI_LOCK(vp);
2218 		}
2219 		break;
2220 	case VPUTX_VUNREF:
2221 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
2222 			error = EBUSY;
2223 		break;
2224 	}
2225 	if (vp->v_usecount > 0)
2226 		vp->v_iflag &= ~VI_OWEINACT;
2227 	if (error == 0) {
2228 		if (vp->v_iflag & VI_OWEINACT)
2229 			vinactive(vp, curthread);
2230 		if (func != VPUTX_VUNREF)
2231 			VOP_UNLOCK(vp, 0);
2232 	}
2233 	vdropl(vp);
2234 }
2235 
2236 /*
2237  * Vnode put/release.
2238  * If count drops to zero, call inactive routine and return to freelist.
2239  */
2240 void
2241 vrele(struct vnode *vp)
2242 {
2243 
2244 	vputx(vp, VPUTX_VRELE);
2245 }
2246 
2247 /*
2248  * Release an already locked vnode.  This give the same effects as
2249  * unlock+vrele(), but takes less time and avoids releasing and
2250  * re-aquiring the lock (as vrele() acquires the lock internally.)
2251  */
2252 void
2253 vput(struct vnode *vp)
2254 {
2255 
2256 	vputx(vp, VPUTX_VPUT);
2257 }
2258 
2259 /*
2260  * Release an exclusively locked vnode. Do not unlock the vnode lock.
2261  */
2262 void
2263 vunref(struct vnode *vp)
2264 {
2265 
2266 	vputx(vp, VPUTX_VUNREF);
2267 }
2268 
2269 /*
2270  * Somebody doesn't want the vnode recycled.
2271  */
2272 void
2273 vhold(struct vnode *vp)
2274 {
2275 
2276 	VI_LOCK(vp);
2277 	vholdl(vp);
2278 	VI_UNLOCK(vp);
2279 }
2280 
2281 /*
2282  * Increase the hold count and activate if this is the first reference.
2283  */
2284 void
2285 vholdl(struct vnode *vp)
2286 {
2287 	struct mount *mp;
2288 
2289 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2290 	vp->v_holdcnt++;
2291 	if (!VSHOULDBUSY(vp))
2292 		return;
2293 	ASSERT_VI_LOCKED(vp, "vholdl");
2294 	VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
2295 	VNASSERT(vp->v_op != NULL, vp, ("vholdl: vnode already reclaimed."));
2296 	/*
2297 	 * Remove a vnode from the free list, mark it as in use,
2298 	 * and put it on the active list.
2299 	 */
2300 	mtx_lock(&vnode_free_list_mtx);
2301 	TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
2302 	freevnodes--;
2303 	vp->v_iflag &= ~(VI_FREE|VI_AGE);
2304 	KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
2305 	    ("Activating already active vnode"));
2306 	vp->v_iflag |= VI_ACTIVE;
2307 	mp = vp->v_mount;
2308 	TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
2309 	mp->mnt_activevnodelistsize++;
2310 	mtx_unlock(&vnode_free_list_mtx);
2311 }
2312 
2313 /*
2314  * Note that there is one less who cares about this vnode.
2315  * vdrop() is the opposite of vhold().
2316  */
2317 void
2318 vdrop(struct vnode *vp)
2319 {
2320 
2321 	VI_LOCK(vp);
2322 	vdropl(vp);
2323 }
2324 
2325 /*
2326  * Drop the hold count of the vnode.  If this is the last reference to
2327  * the vnode we place it on the free list unless it has been vgone'd
2328  * (marked VI_DOOMED) in which case we will free it.
2329  */
2330 void
2331 vdropl(struct vnode *vp)
2332 {
2333 	struct bufobj *bo;
2334 	struct mount *mp;
2335 	int active;
2336 
2337 	ASSERT_VI_LOCKED(vp, "vdropl");
2338 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2339 	if (vp->v_holdcnt <= 0)
2340 		panic("vdrop: holdcnt %d", vp->v_holdcnt);
2341 	vp->v_holdcnt--;
2342 	if (vp->v_holdcnt > 0) {
2343 		VI_UNLOCK(vp);
2344 		return;
2345 	}
2346 	if ((vp->v_iflag & VI_DOOMED) == 0) {
2347 		/*
2348 		 * Mark a vnode as free: remove it from its active list
2349 		 * and put it up for recycling on the freelist.
2350 		 */
2351 		VNASSERT(vp->v_op != NULL, vp,
2352 		    ("vdropl: vnode already reclaimed."));
2353 		VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2354 		    ("vnode already free"));
2355 		VNASSERT(VSHOULDFREE(vp), vp,
2356 		    ("vdropl: freeing when we shouldn't"));
2357 		active = vp->v_iflag & VI_ACTIVE;
2358 		vp->v_iflag &= ~VI_ACTIVE;
2359 		mp = vp->v_mount;
2360 		mtx_lock(&vnode_free_list_mtx);
2361 		if (active) {
2362 			TAILQ_REMOVE(&mp->mnt_activevnodelist, vp,
2363 			    v_actfreelist);
2364 			mp->mnt_activevnodelistsize--;
2365 		}
2366 		if (vp->v_iflag & VI_AGE) {
2367 			TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_actfreelist);
2368 		} else {
2369 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
2370 		}
2371 		freevnodes++;
2372 		vp->v_iflag &= ~VI_AGE;
2373 		vp->v_iflag |= VI_FREE;
2374 		mtx_unlock(&vnode_free_list_mtx);
2375 		VI_UNLOCK(vp);
2376 		return;
2377 	}
2378 	/*
2379 	 * The vnode has been marked for destruction, so free it.
2380 	 */
2381 	CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp);
2382 	mtx_lock(&vnode_free_list_mtx);
2383 	numvnodes--;
2384 	mtx_unlock(&vnode_free_list_mtx);
2385 	bo = &vp->v_bufobj;
2386 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2387 	    ("cleaned vnode still on the free list."));
2388 	VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
2389 	VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
2390 	VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
2391 	VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
2392 	VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
2393 	VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
2394 	VNASSERT(pctrie_is_empty(&bo->bo_clean.bv_root), vp,
2395 	    ("clean blk trie not empty"));
2396 	VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
2397 	VNASSERT(pctrie_is_empty(&bo->bo_dirty.bv_root), vp,
2398 	    ("dirty blk trie not empty"));
2399 	VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
2400 	VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
2401 	VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
2402 	VI_UNLOCK(vp);
2403 #ifdef MAC
2404 	mac_vnode_destroy(vp);
2405 #endif
2406 	if (vp->v_pollinfo != NULL)
2407 		destroy_vpollinfo(vp->v_pollinfo);
2408 #ifdef INVARIANTS
2409 	/* XXX Elsewhere we detect an already freed vnode via NULL v_op. */
2410 	vp->v_op = NULL;
2411 #endif
2412 	rangelock_destroy(&vp->v_rl);
2413 	lockdestroy(vp->v_vnlock);
2414 	mtx_destroy(&vp->v_interlock);
2415 	rw_destroy(BO_LOCKPTR(bo));
2416 	uma_zfree(vnode_zone, vp);
2417 }
2418 
2419 /*
2420  * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2421  * flags.  DOINGINACT prevents us from recursing in calls to vinactive.
2422  * OWEINACT tracks whether a vnode missed a call to inactive due to a
2423  * failed lock upgrade.
2424  */
2425 void
2426 vinactive(struct vnode *vp, struct thread *td)
2427 {
2428 	struct vm_object *obj;
2429 
2430 	ASSERT_VOP_ELOCKED(vp, "vinactive");
2431 	ASSERT_VI_LOCKED(vp, "vinactive");
2432 	VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2433 	    ("vinactive: recursed on VI_DOINGINACT"));
2434 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2435 	vp->v_iflag |= VI_DOINGINACT;
2436 	vp->v_iflag &= ~VI_OWEINACT;
2437 	VI_UNLOCK(vp);
2438 	/*
2439 	 * Before moving off the active list, we must be sure that any
2440 	 * modified pages are on the vnode's dirty list since these will
2441 	 * no longer be checked once the vnode is on the inactive list.
2442 	 * Because the vnode vm object keeps a hold reference on the vnode
2443 	 * if there is at least one resident non-cached page, the vnode
2444 	 * cannot leave the active list without the page cleanup done.
2445 	 */
2446 	obj = vp->v_object;
2447 	if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
2448 		VM_OBJECT_WLOCK(obj);
2449 		vm_object_page_clean(obj, 0, 0, OBJPC_NOSYNC);
2450 		VM_OBJECT_WUNLOCK(obj);
2451 	}
2452 	VOP_INACTIVE(vp, td);
2453 	VI_LOCK(vp);
2454 	VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2455 	    ("vinactive: lost VI_DOINGINACT"));
2456 	vp->v_iflag &= ~VI_DOINGINACT;
2457 }
2458 
2459 /*
2460  * Remove any vnodes in the vnode table belonging to mount point mp.
2461  *
2462  * If FORCECLOSE is not specified, there should not be any active ones,
2463  * return error if any are found (nb: this is a user error, not a
2464  * system error). If FORCECLOSE is specified, detach any active vnodes
2465  * that are found.
2466  *
2467  * If WRITECLOSE is set, only flush out regular file vnodes open for
2468  * writing.
2469  *
2470  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2471  *
2472  * `rootrefs' specifies the base reference count for the root vnode
2473  * of this filesystem. The root vnode is considered busy if its
2474  * v_usecount exceeds this value. On a successful return, vflush(, td)
2475  * will call vrele() on the root vnode exactly rootrefs times.
2476  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2477  * be zero.
2478  */
2479 #ifdef DIAGNOSTIC
2480 static int busyprt = 0;		/* print out busy vnodes */
2481 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
2482 #endif
2483 
2484 int
2485 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
2486 {
2487 	struct vnode *vp, *mvp, *rootvp = NULL;
2488 	struct vattr vattr;
2489 	int busy = 0, error;
2490 
2491 	CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2492 	    rootrefs, flags);
2493 	if (rootrefs > 0) {
2494 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2495 		    ("vflush: bad args"));
2496 		/*
2497 		 * Get the filesystem root vnode. We can vput() it
2498 		 * immediately, since with rootrefs > 0, it won't go away.
2499 		 */
2500 		if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2501 			CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2502 			    __func__, error);
2503 			return (error);
2504 		}
2505 		vput(rootvp);
2506 	}
2507 loop:
2508 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
2509 		vholdl(vp);
2510 		error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2511 		if (error) {
2512 			vdrop(vp);
2513 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2514 			goto loop;
2515 		}
2516 		/*
2517 		 * Skip over a vnodes marked VV_SYSTEM.
2518 		 */
2519 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2520 			VOP_UNLOCK(vp, 0);
2521 			vdrop(vp);
2522 			continue;
2523 		}
2524 		/*
2525 		 * If WRITECLOSE is set, flush out unlinked but still open
2526 		 * files (even if open only for reading) and regular file
2527 		 * vnodes open for writing.
2528 		 */
2529 		if (flags & WRITECLOSE) {
2530 			if (vp->v_object != NULL) {
2531 				VM_OBJECT_WLOCK(vp->v_object);
2532 				vm_object_page_clean(vp->v_object, 0, 0, 0);
2533 				VM_OBJECT_WUNLOCK(vp->v_object);
2534 			}
2535 			error = VOP_FSYNC(vp, MNT_WAIT, td);
2536 			if (error != 0) {
2537 				VOP_UNLOCK(vp, 0);
2538 				vdrop(vp);
2539 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2540 				return (error);
2541 			}
2542 			error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2543 			VI_LOCK(vp);
2544 
2545 			if ((vp->v_type == VNON ||
2546 			    (error == 0 && vattr.va_nlink > 0)) &&
2547 			    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2548 				VOP_UNLOCK(vp, 0);
2549 				vdropl(vp);
2550 				continue;
2551 			}
2552 		} else
2553 			VI_LOCK(vp);
2554 		/*
2555 		 * With v_usecount == 0, all we need to do is clear out the
2556 		 * vnode data structures and we are done.
2557 		 *
2558 		 * If FORCECLOSE is set, forcibly close the vnode.
2559 		 */
2560 		if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2561 			VNASSERT(vp->v_usecount == 0 ||
2562 			    (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2563 			    ("device VNODE %p is FORCECLOSED", vp));
2564 			vgonel(vp);
2565 		} else {
2566 			busy++;
2567 #ifdef DIAGNOSTIC
2568 			if (busyprt)
2569 				vprint("vflush: busy vnode", vp);
2570 #endif
2571 		}
2572 		VOP_UNLOCK(vp, 0);
2573 		vdropl(vp);
2574 	}
2575 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2576 		/*
2577 		 * If just the root vnode is busy, and if its refcount
2578 		 * is equal to `rootrefs', then go ahead and kill it.
2579 		 */
2580 		VI_LOCK(rootvp);
2581 		KASSERT(busy > 0, ("vflush: not busy"));
2582 		VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2583 		    ("vflush: usecount %d < rootrefs %d",
2584 		     rootvp->v_usecount, rootrefs));
2585 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2586 			VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2587 			vgone(rootvp);
2588 			VOP_UNLOCK(rootvp, 0);
2589 			busy = 0;
2590 		} else
2591 			VI_UNLOCK(rootvp);
2592 	}
2593 	if (busy) {
2594 		CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2595 		    busy);
2596 		return (EBUSY);
2597 	}
2598 	for (; rootrefs > 0; rootrefs--)
2599 		vrele(rootvp);
2600 	return (0);
2601 }
2602 
2603 /*
2604  * Recycle an unused vnode to the front of the free list.
2605  */
2606 int
2607 vrecycle(struct vnode *vp)
2608 {
2609 	int recycled;
2610 
2611 	ASSERT_VOP_ELOCKED(vp, "vrecycle");
2612 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2613 	recycled = 0;
2614 	VI_LOCK(vp);
2615 	if (vp->v_usecount == 0) {
2616 		recycled = 1;
2617 		vgonel(vp);
2618 	}
2619 	VI_UNLOCK(vp);
2620 	return (recycled);
2621 }
2622 
2623 /*
2624  * Eliminate all activity associated with a vnode
2625  * in preparation for reuse.
2626  */
2627 void
2628 vgone(struct vnode *vp)
2629 {
2630 	VI_LOCK(vp);
2631 	vgonel(vp);
2632 	VI_UNLOCK(vp);
2633 }
2634 
2635 static void
2636 notify_lowervp_vfs_dummy(struct mount *mp __unused,
2637     struct vnode *lowervp __unused)
2638 {
2639 }
2640 
2641 /*
2642  * Notify upper mounts about reclaimed or unlinked vnode.
2643  */
2644 void
2645 vfs_notify_upper(struct vnode *vp, int event)
2646 {
2647 	static struct vfsops vgonel_vfsops = {
2648 		.vfs_reclaim_lowervp = notify_lowervp_vfs_dummy,
2649 		.vfs_unlink_lowervp = notify_lowervp_vfs_dummy,
2650 	};
2651 	struct mount *mp, *ump, *mmp;
2652 
2653 	mp = vp->v_mount;
2654 	if (mp == NULL)
2655 		return;
2656 
2657 	MNT_ILOCK(mp);
2658 	if (TAILQ_EMPTY(&mp->mnt_uppers))
2659 		goto unlock;
2660 	MNT_IUNLOCK(mp);
2661 	mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO);
2662 	mmp->mnt_op = &vgonel_vfsops;
2663 	mmp->mnt_kern_flag |= MNTK_MARKER;
2664 	MNT_ILOCK(mp);
2665 	mp->mnt_kern_flag |= MNTK_VGONE_UPPER;
2666 	for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) {
2667 		if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) {
2668 			ump = TAILQ_NEXT(ump, mnt_upper_link);
2669 			continue;
2670 		}
2671 		TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link);
2672 		MNT_IUNLOCK(mp);
2673 		switch (event) {
2674 		case VFS_NOTIFY_UPPER_RECLAIM:
2675 			VFS_RECLAIM_LOWERVP(ump, vp);
2676 			break;
2677 		case VFS_NOTIFY_UPPER_UNLINK:
2678 			VFS_UNLINK_LOWERVP(ump, vp);
2679 			break;
2680 		default:
2681 			KASSERT(0, ("invalid event %d", event));
2682 			break;
2683 		}
2684 		MNT_ILOCK(mp);
2685 		ump = TAILQ_NEXT(mmp, mnt_upper_link);
2686 		TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link);
2687 	}
2688 	free(mmp, M_TEMP);
2689 	mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER;
2690 	if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) {
2691 		mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER;
2692 		wakeup(&mp->mnt_uppers);
2693 	}
2694 unlock:
2695 	MNT_IUNLOCK(mp);
2696 }
2697 
2698 /*
2699  * vgone, with the vp interlock held.
2700  */
2701 void
2702 vgonel(struct vnode *vp)
2703 {
2704 	struct thread *td;
2705 	int oweinact;
2706 	int active;
2707 	struct mount *mp;
2708 
2709 	ASSERT_VOP_ELOCKED(vp, "vgonel");
2710 	ASSERT_VI_LOCKED(vp, "vgonel");
2711 	VNASSERT(vp->v_holdcnt, vp,
2712 	    ("vgonel: vp %p has no reference.", vp));
2713 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2714 	td = curthread;
2715 
2716 	/*
2717 	 * Don't vgonel if we're already doomed.
2718 	 */
2719 	if (vp->v_iflag & VI_DOOMED)
2720 		return;
2721 	vp->v_iflag |= VI_DOOMED;
2722 
2723 	/*
2724 	 * Check to see if the vnode is in use.  If so, we have to call
2725 	 * VOP_CLOSE() and VOP_INACTIVE().
2726 	 */
2727 	active = vp->v_usecount;
2728 	oweinact = (vp->v_iflag & VI_OWEINACT);
2729 	VI_UNLOCK(vp);
2730 	vfs_notify_upper(vp, VFS_NOTIFY_UPPER_RECLAIM);
2731 
2732 	/*
2733 	 * Clean out any buffers associated with the vnode.
2734 	 * If the flush fails, just toss the buffers.
2735 	 */
2736 	mp = NULL;
2737 	if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2738 		(void) vn_start_secondary_write(vp, &mp, V_WAIT);
2739 	if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
2740 		vinvalbuf(vp, 0, 0, 0);
2741 
2742 	/*
2743 	 * If purging an active vnode, it must be closed and
2744 	 * deactivated before being reclaimed.
2745 	 */
2746 	if (active)
2747 		VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2748 	if (oweinact || active) {
2749 		VI_LOCK(vp);
2750 		if ((vp->v_iflag & VI_DOINGINACT) == 0)
2751 			vinactive(vp, td);
2752 		VI_UNLOCK(vp);
2753 	}
2754 	if (vp->v_type == VSOCK)
2755 		vfs_unp_reclaim(vp);
2756 	/*
2757 	 * Reclaim the vnode.
2758 	 */
2759 	if (VOP_RECLAIM(vp, td))
2760 		panic("vgone: cannot reclaim");
2761 	if (mp != NULL)
2762 		vn_finished_secondary_write(mp);
2763 	VNASSERT(vp->v_object == NULL, vp,
2764 	    ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2765 	/*
2766 	 * Clear the advisory locks and wake up waiting threads.
2767 	 */
2768 	(void)VOP_ADVLOCKPURGE(vp);
2769 	/*
2770 	 * Delete from old mount point vnode list.
2771 	 */
2772 	delmntque(vp);
2773 	cache_purge(vp);
2774 	/*
2775 	 * Done with purge, reset to the standard lock and invalidate
2776 	 * the vnode.
2777 	 */
2778 	VI_LOCK(vp);
2779 	vp->v_vnlock = &vp->v_lock;
2780 	vp->v_op = &dead_vnodeops;
2781 	vp->v_tag = "none";
2782 	vp->v_type = VBAD;
2783 }
2784 
2785 /*
2786  * Calculate the total number of references to a special device.
2787  */
2788 int
2789 vcount(struct vnode *vp)
2790 {
2791 	int count;
2792 
2793 	dev_lock();
2794 	count = vp->v_rdev->si_usecount;
2795 	dev_unlock();
2796 	return (count);
2797 }
2798 
2799 /*
2800  * Same as above, but using the struct cdev *as argument
2801  */
2802 int
2803 count_dev(struct cdev *dev)
2804 {
2805 	int count;
2806 
2807 	dev_lock();
2808 	count = dev->si_usecount;
2809 	dev_unlock();
2810 	return(count);
2811 }
2812 
2813 /*
2814  * Print out a description of a vnode.
2815  */
2816 static char *typename[] =
2817 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2818  "VMARKER"};
2819 
2820 void
2821 vn_printf(struct vnode *vp, const char *fmt, ...)
2822 {
2823 	va_list ap;
2824 	char buf[256], buf2[16];
2825 	u_long flags;
2826 
2827 	va_start(ap, fmt);
2828 	vprintf(fmt, ap);
2829 	va_end(ap);
2830 	printf("%p: ", (void *)vp);
2831 	printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2832 	printf("    usecount %d, writecount %d, refcount %d mountedhere %p\n",
2833 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2834 	buf[0] = '\0';
2835 	buf[1] = '\0';
2836 	if (vp->v_vflag & VV_ROOT)
2837 		strlcat(buf, "|VV_ROOT", sizeof(buf));
2838 	if (vp->v_vflag & VV_ISTTY)
2839 		strlcat(buf, "|VV_ISTTY", sizeof(buf));
2840 	if (vp->v_vflag & VV_NOSYNC)
2841 		strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2842 	if (vp->v_vflag & VV_ETERNALDEV)
2843 		strlcat(buf, "|VV_ETERNALDEV", sizeof(buf));
2844 	if (vp->v_vflag & VV_CACHEDLABEL)
2845 		strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2846 	if (vp->v_vflag & VV_TEXT)
2847 		strlcat(buf, "|VV_TEXT", sizeof(buf));
2848 	if (vp->v_vflag & VV_COPYONWRITE)
2849 		strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2850 	if (vp->v_vflag & VV_SYSTEM)
2851 		strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2852 	if (vp->v_vflag & VV_PROCDEP)
2853 		strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2854 	if (vp->v_vflag & VV_NOKNOTE)
2855 		strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2856 	if (vp->v_vflag & VV_DELETED)
2857 		strlcat(buf, "|VV_DELETED", sizeof(buf));
2858 	if (vp->v_vflag & VV_MD)
2859 		strlcat(buf, "|VV_MD", sizeof(buf));
2860 	if (vp->v_vflag & VV_FORCEINSMQ)
2861 		strlcat(buf, "|VV_FORCEINSMQ", sizeof(buf));
2862 	flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | VV_ETERNALDEV |
2863 	    VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2864 	    VV_NOKNOTE | VV_DELETED | VV_MD | VV_FORCEINSMQ);
2865 	if (flags != 0) {
2866 		snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2867 		strlcat(buf, buf2, sizeof(buf));
2868 	}
2869 	if (vp->v_iflag & VI_MOUNT)
2870 		strlcat(buf, "|VI_MOUNT", sizeof(buf));
2871 	if (vp->v_iflag & VI_AGE)
2872 		strlcat(buf, "|VI_AGE", sizeof(buf));
2873 	if (vp->v_iflag & VI_DOOMED)
2874 		strlcat(buf, "|VI_DOOMED", sizeof(buf));
2875 	if (vp->v_iflag & VI_FREE)
2876 		strlcat(buf, "|VI_FREE", sizeof(buf));
2877 	if (vp->v_iflag & VI_ACTIVE)
2878 		strlcat(buf, "|VI_ACTIVE", sizeof(buf));
2879 	if (vp->v_iflag & VI_DOINGINACT)
2880 		strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2881 	if (vp->v_iflag & VI_OWEINACT)
2882 		strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2883 	flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2884 	    VI_ACTIVE | VI_DOINGINACT | VI_OWEINACT);
2885 	if (flags != 0) {
2886 		snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2887 		strlcat(buf, buf2, sizeof(buf));
2888 	}
2889 	printf("    flags (%s)\n", buf + 1);
2890 	if (mtx_owned(VI_MTX(vp)))
2891 		printf(" VI_LOCKed");
2892 	if (vp->v_object != NULL)
2893 		printf("    v_object %p ref %d pages %d\n",
2894 		    vp->v_object, vp->v_object->ref_count,
2895 		    vp->v_object->resident_page_count);
2896 	printf("    ");
2897 	lockmgr_printinfo(vp->v_vnlock);
2898 	if (vp->v_data != NULL)
2899 		VOP_PRINT(vp);
2900 }
2901 
2902 #ifdef DDB
2903 /*
2904  * List all of the locked vnodes in the system.
2905  * Called when debugging the kernel.
2906  */
2907 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2908 {
2909 	struct mount *mp, *nmp;
2910 	struct vnode *vp;
2911 
2912 	/*
2913 	 * Note: because this is DDB, we can't obey the locking semantics
2914 	 * for these structures, which means we could catch an inconsistent
2915 	 * state and dereference a nasty pointer.  Not much to be done
2916 	 * about that.
2917 	 */
2918 	db_printf("Locked vnodes\n");
2919 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2920 		nmp = TAILQ_NEXT(mp, mnt_list);
2921 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2922 			if (vp->v_type != VMARKER &&
2923 			    VOP_ISLOCKED(vp))
2924 				vprint("", vp);
2925 		}
2926 		nmp = TAILQ_NEXT(mp, mnt_list);
2927 	}
2928 }
2929 
2930 /*
2931  * Show details about the given vnode.
2932  */
2933 DB_SHOW_COMMAND(vnode, db_show_vnode)
2934 {
2935 	struct vnode *vp;
2936 
2937 	if (!have_addr)
2938 		return;
2939 	vp = (struct vnode *)addr;
2940 	vn_printf(vp, "vnode ");
2941 }
2942 
2943 /*
2944  * Show details about the given mount point.
2945  */
2946 DB_SHOW_COMMAND(mount, db_show_mount)
2947 {
2948 	struct mount *mp;
2949 	struct vfsopt *opt;
2950 	struct statfs *sp;
2951 	struct vnode *vp;
2952 	char buf[512];
2953 	uint64_t mflags;
2954 	u_int flags;
2955 
2956 	if (!have_addr) {
2957 		/* No address given, print short info about all mount points. */
2958 		TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2959 			db_printf("%p %s on %s (%s)\n", mp,
2960 			    mp->mnt_stat.f_mntfromname,
2961 			    mp->mnt_stat.f_mntonname,
2962 			    mp->mnt_stat.f_fstypename);
2963 			if (db_pager_quit)
2964 				break;
2965 		}
2966 		db_printf("\nMore info: show mount <addr>\n");
2967 		return;
2968 	}
2969 
2970 	mp = (struct mount *)addr;
2971 	db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
2972 	    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
2973 
2974 	buf[0] = '\0';
2975 	mflags = mp->mnt_flag;
2976 #define	MNT_FLAG(flag)	do {						\
2977 	if (mflags & (flag)) {						\
2978 		if (buf[0] != '\0')					\
2979 			strlcat(buf, ", ", sizeof(buf));		\
2980 		strlcat(buf, (#flag) + 4, sizeof(buf));			\
2981 		mflags &= ~(flag);					\
2982 	}								\
2983 } while (0)
2984 	MNT_FLAG(MNT_RDONLY);
2985 	MNT_FLAG(MNT_SYNCHRONOUS);
2986 	MNT_FLAG(MNT_NOEXEC);
2987 	MNT_FLAG(MNT_NOSUID);
2988 	MNT_FLAG(MNT_NFS4ACLS);
2989 	MNT_FLAG(MNT_UNION);
2990 	MNT_FLAG(MNT_ASYNC);
2991 	MNT_FLAG(MNT_SUIDDIR);
2992 	MNT_FLAG(MNT_SOFTDEP);
2993 	MNT_FLAG(MNT_NOSYMFOLLOW);
2994 	MNT_FLAG(MNT_GJOURNAL);
2995 	MNT_FLAG(MNT_MULTILABEL);
2996 	MNT_FLAG(MNT_ACLS);
2997 	MNT_FLAG(MNT_NOATIME);
2998 	MNT_FLAG(MNT_NOCLUSTERR);
2999 	MNT_FLAG(MNT_NOCLUSTERW);
3000 	MNT_FLAG(MNT_SUJ);
3001 	MNT_FLAG(MNT_EXRDONLY);
3002 	MNT_FLAG(MNT_EXPORTED);
3003 	MNT_FLAG(MNT_DEFEXPORTED);
3004 	MNT_FLAG(MNT_EXPORTANON);
3005 	MNT_FLAG(MNT_EXKERB);
3006 	MNT_FLAG(MNT_EXPUBLIC);
3007 	MNT_FLAG(MNT_LOCAL);
3008 	MNT_FLAG(MNT_QUOTA);
3009 	MNT_FLAG(MNT_ROOTFS);
3010 	MNT_FLAG(MNT_USER);
3011 	MNT_FLAG(MNT_IGNORE);
3012 	MNT_FLAG(MNT_UPDATE);
3013 	MNT_FLAG(MNT_DELEXPORT);
3014 	MNT_FLAG(MNT_RELOAD);
3015 	MNT_FLAG(MNT_FORCE);
3016 	MNT_FLAG(MNT_SNAPSHOT);
3017 	MNT_FLAG(MNT_BYFSID);
3018 #undef MNT_FLAG
3019 	if (mflags != 0) {
3020 		if (buf[0] != '\0')
3021 			strlcat(buf, ", ", sizeof(buf));
3022 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3023 		    "0x%016jx", mflags);
3024 	}
3025 	db_printf("    mnt_flag = %s\n", buf);
3026 
3027 	buf[0] = '\0';
3028 	flags = mp->mnt_kern_flag;
3029 #define	MNT_KERN_FLAG(flag)	do {					\
3030 	if (flags & (flag)) {						\
3031 		if (buf[0] != '\0')					\
3032 			strlcat(buf, ", ", sizeof(buf));		\
3033 		strlcat(buf, (#flag) + 5, sizeof(buf));			\
3034 		flags &= ~(flag);					\
3035 	}								\
3036 } while (0)
3037 	MNT_KERN_FLAG(MNTK_UNMOUNTF);
3038 	MNT_KERN_FLAG(MNTK_ASYNC);
3039 	MNT_KERN_FLAG(MNTK_SOFTDEP);
3040 	MNT_KERN_FLAG(MNTK_NOINSMNTQ);
3041 	MNT_KERN_FLAG(MNTK_DRAINING);
3042 	MNT_KERN_FLAG(MNTK_REFEXPIRE);
3043 	MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
3044 	MNT_KERN_FLAG(MNTK_SHARED_WRITES);
3045 	MNT_KERN_FLAG(MNTK_NO_IOPF);
3046 	MNT_KERN_FLAG(MNTK_VGONE_UPPER);
3047 	MNT_KERN_FLAG(MNTK_VGONE_WAITER);
3048 	MNT_KERN_FLAG(MNTK_LOOKUP_EXCL_DOTDOT);
3049 	MNT_KERN_FLAG(MNTK_MARKER);
3050 	MNT_KERN_FLAG(MNTK_NOASYNC);
3051 	MNT_KERN_FLAG(MNTK_UNMOUNT);
3052 	MNT_KERN_FLAG(MNTK_MWAIT);
3053 	MNT_KERN_FLAG(MNTK_SUSPEND);
3054 	MNT_KERN_FLAG(MNTK_SUSPEND2);
3055 	MNT_KERN_FLAG(MNTK_SUSPENDED);
3056 	MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
3057 	MNT_KERN_FLAG(MNTK_NOKNOTE);
3058 #undef MNT_KERN_FLAG
3059 	if (flags != 0) {
3060 		if (buf[0] != '\0')
3061 			strlcat(buf, ", ", sizeof(buf));
3062 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3063 		    "0x%08x", flags);
3064 	}
3065 	db_printf("    mnt_kern_flag = %s\n", buf);
3066 
3067 	db_printf("    mnt_opt = ");
3068 	opt = TAILQ_FIRST(mp->mnt_opt);
3069 	if (opt != NULL) {
3070 		db_printf("%s", opt->name);
3071 		opt = TAILQ_NEXT(opt, link);
3072 		while (opt != NULL) {
3073 			db_printf(", %s", opt->name);
3074 			opt = TAILQ_NEXT(opt, link);
3075 		}
3076 	}
3077 	db_printf("\n");
3078 
3079 	sp = &mp->mnt_stat;
3080 	db_printf("    mnt_stat = { version=%u type=%u flags=0x%016jx "
3081 	    "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
3082 	    "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
3083 	    "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
3084 	    (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
3085 	    (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
3086 	    (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
3087 	    (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
3088 	    (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
3089 	    (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
3090 	    (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
3091 	    (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
3092 
3093 	db_printf("    mnt_cred = { uid=%u ruid=%u",
3094 	    (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
3095 	if (jailed(mp->mnt_cred))
3096 		db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
3097 	db_printf(" }\n");
3098 	db_printf("    mnt_ref = %d\n", mp->mnt_ref);
3099 	db_printf("    mnt_gen = %d\n", mp->mnt_gen);
3100 	db_printf("    mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
3101 	db_printf("    mnt_activevnodelistsize = %d\n",
3102 	    mp->mnt_activevnodelistsize);
3103 	db_printf("    mnt_writeopcount = %d\n", mp->mnt_writeopcount);
3104 	db_printf("    mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
3105 	db_printf("    mnt_iosize_max = %d\n", mp->mnt_iosize_max);
3106 	db_printf("    mnt_hashseed = %u\n", mp->mnt_hashseed);
3107 	db_printf("    mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
3108 	db_printf("    mnt_secondary_accwrites = %d\n",
3109 	    mp->mnt_secondary_accwrites);
3110 	db_printf("    mnt_gjprovider = %s\n",
3111 	    mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
3112 
3113 	db_printf("\n\nList of active vnodes\n");
3114 	TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) {
3115 		if (vp->v_type != VMARKER) {
3116 			vn_printf(vp, "vnode ");
3117 			if (db_pager_quit)
3118 				break;
3119 		}
3120 	}
3121 	db_printf("\n\nList of inactive vnodes\n");
3122 	TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3123 		if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) {
3124 			vn_printf(vp, "vnode ");
3125 			if (db_pager_quit)
3126 				break;
3127 		}
3128 	}
3129 }
3130 #endif	/* DDB */
3131 
3132 /*
3133  * Fill in a struct xvfsconf based on a struct vfsconf.
3134  */
3135 static int
3136 vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp)
3137 {
3138 	struct xvfsconf xvfsp;
3139 
3140 	bzero(&xvfsp, sizeof(xvfsp));
3141 	strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3142 	xvfsp.vfc_typenum = vfsp->vfc_typenum;
3143 	xvfsp.vfc_refcount = vfsp->vfc_refcount;
3144 	xvfsp.vfc_flags = vfsp->vfc_flags;
3145 	/*
3146 	 * These are unused in userland, we keep them
3147 	 * to not break binary compatibility.
3148 	 */
3149 	xvfsp.vfc_vfsops = NULL;
3150 	xvfsp.vfc_next = NULL;
3151 	return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3152 }
3153 
3154 #ifdef COMPAT_FREEBSD32
3155 struct xvfsconf32 {
3156 	uint32_t	vfc_vfsops;
3157 	char		vfc_name[MFSNAMELEN];
3158 	int32_t		vfc_typenum;
3159 	int32_t		vfc_refcount;
3160 	int32_t		vfc_flags;
3161 	uint32_t	vfc_next;
3162 };
3163 
3164 static int
3165 vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp)
3166 {
3167 	struct xvfsconf32 xvfsp;
3168 
3169 	strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3170 	xvfsp.vfc_typenum = vfsp->vfc_typenum;
3171 	xvfsp.vfc_refcount = vfsp->vfc_refcount;
3172 	xvfsp.vfc_flags = vfsp->vfc_flags;
3173 	xvfsp.vfc_vfsops = 0;
3174 	xvfsp.vfc_next = 0;
3175 	return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3176 }
3177 #endif
3178 
3179 /*
3180  * Top level filesystem related information gathering.
3181  */
3182 static int
3183 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
3184 {
3185 	struct vfsconf *vfsp;
3186 	int error;
3187 
3188 	error = 0;
3189 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3190 #ifdef COMPAT_FREEBSD32
3191 		if (req->flags & SCTL_MASK32)
3192 			error = vfsconf2x32(req, vfsp);
3193 		else
3194 #endif
3195 			error = vfsconf2x(req, vfsp);
3196 		if (error)
3197 			break;
3198 	}
3199 	return (error);
3200 }
3201 
3202 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD,
3203     NULL, 0, sysctl_vfs_conflist,
3204     "S,xvfsconf", "List of all configured filesystems");
3205 
3206 #ifndef BURN_BRIDGES
3207 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
3208 
3209 static int
3210 vfs_sysctl(SYSCTL_HANDLER_ARGS)
3211 {
3212 	int *name = (int *)arg1 - 1;	/* XXX */
3213 	u_int namelen = arg2 + 1;	/* XXX */
3214 	struct vfsconf *vfsp;
3215 
3216 	log(LOG_WARNING, "userland calling deprecated sysctl, "
3217 	    "please rebuild world\n");
3218 
3219 #if 1 || defined(COMPAT_PRELITE2)
3220 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3221 	if (namelen == 1)
3222 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3223 #endif
3224 
3225 	switch (name[1]) {
3226 	case VFS_MAXTYPENUM:
3227 		if (namelen != 2)
3228 			return (ENOTDIR);
3229 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3230 	case VFS_CONF:
3231 		if (namelen != 3)
3232 			return (ENOTDIR);	/* overloaded */
3233 		TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
3234 			if (vfsp->vfc_typenum == name[2])
3235 				break;
3236 		if (vfsp == NULL)
3237 			return (EOPNOTSUPP);
3238 #ifdef COMPAT_FREEBSD32
3239 		if (req->flags & SCTL_MASK32)
3240 			return (vfsconf2x32(req, vfsp));
3241 		else
3242 #endif
3243 			return (vfsconf2x(req, vfsp));
3244 	}
3245 	return (EOPNOTSUPP);
3246 }
3247 
3248 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
3249     vfs_sysctl, "Generic filesystem");
3250 
3251 #if 1 || defined(COMPAT_PRELITE2)
3252 
3253 static int
3254 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3255 {
3256 	int error;
3257 	struct vfsconf *vfsp;
3258 	struct ovfsconf ovfs;
3259 
3260 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3261 		bzero(&ovfs, sizeof(ovfs));
3262 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
3263 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
3264 		ovfs.vfc_index = vfsp->vfc_typenum;
3265 		ovfs.vfc_refcount = vfsp->vfc_refcount;
3266 		ovfs.vfc_flags = vfsp->vfc_flags;
3267 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3268 		if (error)
3269 			return error;
3270 	}
3271 	return 0;
3272 }
3273 
3274 #endif /* 1 || COMPAT_PRELITE2 */
3275 #endif /* !BURN_BRIDGES */
3276 
3277 #define KINFO_VNODESLOP		10
3278 #ifdef notyet
3279 /*
3280  * Dump vnode list (via sysctl).
3281  */
3282 /* ARGSUSED */
3283 static int
3284 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3285 {
3286 	struct xvnode *xvn;
3287 	struct mount *mp;
3288 	struct vnode *vp;
3289 	int error, len, n;
3290 
3291 	/*
3292 	 * Stale numvnodes access is not fatal here.
3293 	 */
3294 	req->lock = 0;
3295 	len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3296 	if (!req->oldptr)
3297 		/* Make an estimate */
3298 		return (SYSCTL_OUT(req, 0, len));
3299 
3300 	error = sysctl_wire_old_buffer(req, 0);
3301 	if (error != 0)
3302 		return (error);
3303 	xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3304 	n = 0;
3305 	mtx_lock(&mountlist_mtx);
3306 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3307 		if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3308 			continue;
3309 		MNT_ILOCK(mp);
3310 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3311 			if (n == len)
3312 				break;
3313 			vref(vp);
3314 			xvn[n].xv_size = sizeof *xvn;
3315 			xvn[n].xv_vnode = vp;
3316 			xvn[n].xv_id = 0;	/* XXX compat */
3317 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3318 			XV_COPY(usecount);
3319 			XV_COPY(writecount);
3320 			XV_COPY(holdcnt);
3321 			XV_COPY(mount);
3322 			XV_COPY(numoutput);
3323 			XV_COPY(type);
3324 #undef XV_COPY
3325 			xvn[n].xv_flag = vp->v_vflag;
3326 
3327 			switch (vp->v_type) {
3328 			case VREG:
3329 			case VDIR:
3330 			case VLNK:
3331 				break;
3332 			case VBLK:
3333 			case VCHR:
3334 				if (vp->v_rdev == NULL) {
3335 					vrele(vp);
3336 					continue;
3337 				}
3338 				xvn[n].xv_dev = dev2udev(vp->v_rdev);
3339 				break;
3340 			case VSOCK:
3341 				xvn[n].xv_socket = vp->v_socket;
3342 				break;
3343 			case VFIFO:
3344 				xvn[n].xv_fifo = vp->v_fifoinfo;
3345 				break;
3346 			case VNON:
3347 			case VBAD:
3348 			default:
3349 				/* shouldn't happen? */
3350 				vrele(vp);
3351 				continue;
3352 			}
3353 			vrele(vp);
3354 			++n;
3355 		}
3356 		MNT_IUNLOCK(mp);
3357 		mtx_lock(&mountlist_mtx);
3358 		vfs_unbusy(mp);
3359 		if (n == len)
3360 			break;
3361 	}
3362 	mtx_unlock(&mountlist_mtx);
3363 
3364 	error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3365 	free(xvn, M_TEMP);
3366 	return (error);
3367 }
3368 
3369 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3370     0, 0, sysctl_vnode, "S,xvnode", "");
3371 #endif
3372 
3373 /*
3374  * Unmount all filesystems. The list is traversed in reverse order
3375  * of mounting to avoid dependencies.
3376  */
3377 void
3378 vfs_unmountall(void)
3379 {
3380 	struct mount *mp;
3381 	struct thread *td;
3382 	int error;
3383 
3384 	CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3385 	td = curthread;
3386 
3387 	/*
3388 	 * Since this only runs when rebooting, it is not interlocked.
3389 	 */
3390 	while(!TAILQ_EMPTY(&mountlist)) {
3391 		mp = TAILQ_LAST(&mountlist, mntlist);
3392 		error = dounmount(mp, MNT_FORCE, td);
3393 		if (error) {
3394 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
3395 			/*
3396 			 * XXX: Due to the way in which we mount the root
3397 			 * file system off of devfs, devfs will generate a
3398 			 * "busy" warning when we try to unmount it before
3399 			 * the root.  Don't print a warning as a result in
3400 			 * order to avoid false positive errors that may
3401 			 * cause needless upset.
3402 			 */
3403 			if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3404 				printf("unmount of %s failed (",
3405 				    mp->mnt_stat.f_mntonname);
3406 				if (error == EBUSY)
3407 					printf("BUSY)\n");
3408 				else
3409 					printf("%d)\n", error);
3410 			}
3411 		} else {
3412 			/* The unmount has removed mp from the mountlist */
3413 		}
3414 	}
3415 }
3416 
3417 /*
3418  * perform msync on all vnodes under a mount point
3419  * the mount point must be locked.
3420  */
3421 void
3422 vfs_msync(struct mount *mp, int flags)
3423 {
3424 	struct vnode *vp, *mvp;
3425 	struct vm_object *obj;
3426 
3427 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3428 	MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
3429 		obj = vp->v_object;
3430 		if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3431 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3432 			if (!vget(vp,
3433 			    LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3434 			    curthread)) {
3435 				if (vp->v_vflag & VV_NOSYNC) {	/* unlinked */
3436 					vput(vp);
3437 					continue;
3438 				}
3439 
3440 				obj = vp->v_object;
3441 				if (obj != NULL) {
3442 					VM_OBJECT_WLOCK(obj);
3443 					vm_object_page_clean(obj, 0, 0,
3444 					    flags == MNT_WAIT ?
3445 					    OBJPC_SYNC : OBJPC_NOSYNC);
3446 					VM_OBJECT_WUNLOCK(obj);
3447 				}
3448 				vput(vp);
3449 			}
3450 		} else
3451 			VI_UNLOCK(vp);
3452 	}
3453 }
3454 
3455 static void
3456 destroy_vpollinfo(struct vpollinfo *vi)
3457 {
3458 	seldrain(&vi->vpi_selinfo);
3459 	knlist_destroy(&vi->vpi_selinfo.si_note);
3460 	mtx_destroy(&vi->vpi_lock);
3461 	uma_zfree(vnodepoll_zone, vi);
3462 }
3463 
3464 /*
3465  * Initalize per-vnode helper structure to hold poll-related state.
3466  */
3467 void
3468 v_addpollinfo(struct vnode *vp)
3469 {
3470 	struct vpollinfo *vi;
3471 
3472 	if (vp->v_pollinfo != NULL)
3473 		return;
3474 	vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3475 	mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3476 	knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3477 	    vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3478 	VI_LOCK(vp);
3479 	if (vp->v_pollinfo != NULL) {
3480 		VI_UNLOCK(vp);
3481 		destroy_vpollinfo(vi);
3482 		return;
3483 	}
3484 	vp->v_pollinfo = vi;
3485 	VI_UNLOCK(vp);
3486 }
3487 
3488 /*
3489  * Record a process's interest in events which might happen to
3490  * a vnode.  Because poll uses the historic select-style interface
3491  * internally, this routine serves as both the ``check for any
3492  * pending events'' and the ``record my interest in future events''
3493  * functions.  (These are done together, while the lock is held,
3494  * to avoid race conditions.)
3495  */
3496 int
3497 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3498 {
3499 
3500 	v_addpollinfo(vp);
3501 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3502 	if (vp->v_pollinfo->vpi_revents & events) {
3503 		/*
3504 		 * This leaves events we are not interested
3505 		 * in available for the other process which
3506 		 * which presumably had requested them
3507 		 * (otherwise they would never have been
3508 		 * recorded).
3509 		 */
3510 		events &= vp->v_pollinfo->vpi_revents;
3511 		vp->v_pollinfo->vpi_revents &= ~events;
3512 
3513 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
3514 		return (events);
3515 	}
3516 	vp->v_pollinfo->vpi_events |= events;
3517 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3518 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3519 	return (0);
3520 }
3521 
3522 /*
3523  * Routine to create and manage a filesystem syncer vnode.
3524  */
3525 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
3526 static int	sync_fsync(struct  vop_fsync_args *);
3527 static int	sync_inactive(struct  vop_inactive_args *);
3528 static int	sync_reclaim(struct  vop_reclaim_args *);
3529 
3530 static struct vop_vector sync_vnodeops = {
3531 	.vop_bypass =	VOP_EOPNOTSUPP,
3532 	.vop_close =	sync_close,		/* close */
3533 	.vop_fsync =	sync_fsync,		/* fsync */
3534 	.vop_inactive =	sync_inactive,	/* inactive */
3535 	.vop_reclaim =	sync_reclaim,	/* reclaim */
3536 	.vop_lock1 =	vop_stdlock,	/* lock */
3537 	.vop_unlock =	vop_stdunlock,	/* unlock */
3538 	.vop_islocked =	vop_stdislocked,	/* islocked */
3539 };
3540 
3541 /*
3542  * Create a new filesystem syncer vnode for the specified mount point.
3543  */
3544 void
3545 vfs_allocate_syncvnode(struct mount *mp)
3546 {
3547 	struct vnode *vp;
3548 	struct bufobj *bo;
3549 	static long start, incr, next;
3550 	int error;
3551 
3552 	/* Allocate a new vnode */
3553 	error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
3554 	if (error != 0)
3555 		panic("vfs_allocate_syncvnode: getnewvnode() failed");
3556 	vp->v_type = VNON;
3557 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3558 	vp->v_vflag |= VV_FORCEINSMQ;
3559 	error = insmntque(vp, mp);
3560 	if (error != 0)
3561 		panic("vfs_allocate_syncvnode: insmntque() failed");
3562 	vp->v_vflag &= ~VV_FORCEINSMQ;
3563 	VOP_UNLOCK(vp, 0);
3564 	/*
3565 	 * Place the vnode onto the syncer worklist. We attempt to
3566 	 * scatter them about on the list so that they will go off
3567 	 * at evenly distributed times even if all the filesystems
3568 	 * are mounted at once.
3569 	 */
3570 	next += incr;
3571 	if (next == 0 || next > syncer_maxdelay) {
3572 		start /= 2;
3573 		incr /= 2;
3574 		if (start == 0) {
3575 			start = syncer_maxdelay / 2;
3576 			incr = syncer_maxdelay;
3577 		}
3578 		next = start;
3579 	}
3580 	bo = &vp->v_bufobj;
3581 	BO_LOCK(bo);
3582 	vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3583 	/* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3584 	mtx_lock(&sync_mtx);
3585 	sync_vnode_count++;
3586 	if (mp->mnt_syncer == NULL) {
3587 		mp->mnt_syncer = vp;
3588 		vp = NULL;
3589 	}
3590 	mtx_unlock(&sync_mtx);
3591 	BO_UNLOCK(bo);
3592 	if (vp != NULL) {
3593 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3594 		vgone(vp);
3595 		vput(vp);
3596 	}
3597 }
3598 
3599 void
3600 vfs_deallocate_syncvnode(struct mount *mp)
3601 {
3602 	struct vnode *vp;
3603 
3604 	mtx_lock(&sync_mtx);
3605 	vp = mp->mnt_syncer;
3606 	if (vp != NULL)
3607 		mp->mnt_syncer = NULL;
3608 	mtx_unlock(&sync_mtx);
3609 	if (vp != NULL)
3610 		vrele(vp);
3611 }
3612 
3613 /*
3614  * Do a lazy sync of the filesystem.
3615  */
3616 static int
3617 sync_fsync(struct vop_fsync_args *ap)
3618 {
3619 	struct vnode *syncvp = ap->a_vp;
3620 	struct mount *mp = syncvp->v_mount;
3621 	int error, save;
3622 	struct bufobj *bo;
3623 
3624 	/*
3625 	 * We only need to do something if this is a lazy evaluation.
3626 	 */
3627 	if (ap->a_waitfor != MNT_LAZY)
3628 		return (0);
3629 
3630 	/*
3631 	 * Move ourselves to the back of the sync list.
3632 	 */
3633 	bo = &syncvp->v_bufobj;
3634 	BO_LOCK(bo);
3635 	vn_syncer_add_to_worklist(bo, syncdelay);
3636 	BO_UNLOCK(bo);
3637 
3638 	/*
3639 	 * Walk the list of vnodes pushing all that are dirty and
3640 	 * not already on the sync list.
3641 	 */
3642 	mtx_lock(&mountlist_mtx);
3643 	if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) {
3644 		mtx_unlock(&mountlist_mtx);
3645 		return (0);
3646 	}
3647 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3648 		vfs_unbusy(mp);
3649 		return (0);
3650 	}
3651 	save = curthread_pflags_set(TDP_SYNCIO);
3652 	vfs_msync(mp, MNT_NOWAIT);
3653 	error = VFS_SYNC(mp, MNT_LAZY);
3654 	curthread_pflags_restore(save);
3655 	vn_finished_write(mp);
3656 	vfs_unbusy(mp);
3657 	return (error);
3658 }
3659 
3660 /*
3661  * The syncer vnode is no referenced.
3662  */
3663 static int
3664 sync_inactive(struct vop_inactive_args *ap)
3665 {
3666 
3667 	vgone(ap->a_vp);
3668 	return (0);
3669 }
3670 
3671 /*
3672  * The syncer vnode is no longer needed and is being decommissioned.
3673  *
3674  * Modifications to the worklist must be protected by sync_mtx.
3675  */
3676 static int
3677 sync_reclaim(struct vop_reclaim_args *ap)
3678 {
3679 	struct vnode *vp = ap->a_vp;
3680 	struct bufobj *bo;
3681 
3682 	bo = &vp->v_bufobj;
3683 	BO_LOCK(bo);
3684 	mtx_lock(&sync_mtx);
3685 	if (vp->v_mount->mnt_syncer == vp)
3686 		vp->v_mount->mnt_syncer = NULL;
3687 	if (bo->bo_flag & BO_ONWORKLST) {
3688 		LIST_REMOVE(bo, bo_synclist);
3689 		syncer_worklist_len--;
3690 		sync_vnode_count--;
3691 		bo->bo_flag &= ~BO_ONWORKLST;
3692 	}
3693 	mtx_unlock(&sync_mtx);
3694 	BO_UNLOCK(bo);
3695 
3696 	return (0);
3697 }
3698 
3699 /*
3700  * Check if vnode represents a disk device
3701  */
3702 int
3703 vn_isdisk(struct vnode *vp, int *errp)
3704 {
3705 	int error;
3706 
3707 	error = 0;
3708 	dev_lock();
3709 	if (vp->v_type != VCHR)
3710 		error = ENOTBLK;
3711 	else if (vp->v_rdev == NULL)
3712 		error = ENXIO;
3713 	else if (vp->v_rdev->si_devsw == NULL)
3714 		error = ENXIO;
3715 	else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3716 		error = ENOTBLK;
3717 	dev_unlock();
3718 	if (errp != NULL)
3719 		*errp = error;
3720 	return (error == 0);
3721 }
3722 
3723 /*
3724  * Common filesystem object access control check routine.  Accepts a
3725  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3726  * and optional call-by-reference privused argument allowing vaccess()
3727  * to indicate to the caller whether privilege was used to satisfy the
3728  * request (obsoleted).  Returns 0 on success, or an errno on failure.
3729  */
3730 int
3731 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3732     accmode_t accmode, struct ucred *cred, int *privused)
3733 {
3734 	accmode_t dac_granted;
3735 	accmode_t priv_granted;
3736 
3737 	KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3738 	    ("invalid bit in accmode"));
3739 	KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
3740 	    ("VAPPEND without VWRITE"));
3741 
3742 	/*
3743 	 * Look for a normal, non-privileged way to access the file/directory
3744 	 * as requested.  If it exists, go with that.
3745 	 */
3746 
3747 	if (privused != NULL)
3748 		*privused = 0;
3749 
3750 	dac_granted = 0;
3751 
3752 	/* Check the owner. */
3753 	if (cred->cr_uid == file_uid) {
3754 		dac_granted |= VADMIN;
3755 		if (file_mode & S_IXUSR)
3756 			dac_granted |= VEXEC;
3757 		if (file_mode & S_IRUSR)
3758 			dac_granted |= VREAD;
3759 		if (file_mode & S_IWUSR)
3760 			dac_granted |= (VWRITE | VAPPEND);
3761 
3762 		if ((accmode & dac_granted) == accmode)
3763 			return (0);
3764 
3765 		goto privcheck;
3766 	}
3767 
3768 	/* Otherwise, check the groups (first match) */
3769 	if (groupmember(file_gid, cred)) {
3770 		if (file_mode & S_IXGRP)
3771 			dac_granted |= VEXEC;
3772 		if (file_mode & S_IRGRP)
3773 			dac_granted |= VREAD;
3774 		if (file_mode & S_IWGRP)
3775 			dac_granted |= (VWRITE | VAPPEND);
3776 
3777 		if ((accmode & dac_granted) == accmode)
3778 			return (0);
3779 
3780 		goto privcheck;
3781 	}
3782 
3783 	/* Otherwise, check everyone else. */
3784 	if (file_mode & S_IXOTH)
3785 		dac_granted |= VEXEC;
3786 	if (file_mode & S_IROTH)
3787 		dac_granted |= VREAD;
3788 	if (file_mode & S_IWOTH)
3789 		dac_granted |= (VWRITE | VAPPEND);
3790 	if ((accmode & dac_granted) == accmode)
3791 		return (0);
3792 
3793 privcheck:
3794 	/*
3795 	 * Build a privilege mask to determine if the set of privileges
3796 	 * satisfies the requirements when combined with the granted mask
3797 	 * from above.  For each privilege, if the privilege is required,
3798 	 * bitwise or the request type onto the priv_granted mask.
3799 	 */
3800 	priv_granted = 0;
3801 
3802 	if (type == VDIR) {
3803 		/*
3804 		 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3805 		 * requests, instead of PRIV_VFS_EXEC.
3806 		 */
3807 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3808 		    !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3809 			priv_granted |= VEXEC;
3810 	} else {
3811 		/*
3812 		 * Ensure that at least one execute bit is on. Otherwise,
3813 		 * a privileged user will always succeed, and we don't want
3814 		 * this to happen unless the file really is executable.
3815 		 */
3816 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3817 		    (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
3818 		    !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3819 			priv_granted |= VEXEC;
3820 	}
3821 
3822 	if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
3823 	    !priv_check_cred(cred, PRIV_VFS_READ, 0))
3824 		priv_granted |= VREAD;
3825 
3826 	if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3827 	    !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3828 		priv_granted |= (VWRITE | VAPPEND);
3829 
3830 	if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3831 	    !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3832 		priv_granted |= VADMIN;
3833 
3834 	if ((accmode & (priv_granted | dac_granted)) == accmode) {
3835 		/* XXX audit: privilege used */
3836 		if (privused != NULL)
3837 			*privused = 1;
3838 		return (0);
3839 	}
3840 
3841 	return ((accmode & VADMIN) ? EPERM : EACCES);
3842 }
3843 
3844 /*
3845  * Credential check based on process requesting service, and per-attribute
3846  * permissions.
3847  */
3848 int
3849 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3850     struct thread *td, accmode_t accmode)
3851 {
3852 
3853 	/*
3854 	 * Kernel-invoked always succeeds.
3855 	 */
3856 	if (cred == NOCRED)
3857 		return (0);
3858 
3859 	/*
3860 	 * Do not allow privileged processes in jail to directly manipulate
3861 	 * system attributes.
3862 	 */
3863 	switch (attrnamespace) {
3864 	case EXTATTR_NAMESPACE_SYSTEM:
3865 		/* Potentially should be: return (EPERM); */
3866 		return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3867 	case EXTATTR_NAMESPACE_USER:
3868 		return (VOP_ACCESS(vp, accmode, cred, td));
3869 	default:
3870 		return (EPERM);
3871 	}
3872 }
3873 
3874 #ifdef DEBUG_VFS_LOCKS
3875 /*
3876  * This only exists to supress warnings from unlocked specfs accesses.  It is
3877  * no longer ok to have an unlocked VFS.
3878  */
3879 #define	IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL ||		\
3880 	(vp)->v_type == VCHR ||	(vp)->v_type == VBAD)
3881 
3882 int vfs_badlock_ddb = 1;	/* Drop into debugger on violation. */
3883 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
3884     "Drop into debugger on lock violation");
3885 
3886 int vfs_badlock_mutex = 1;	/* Check for interlock across VOPs. */
3887 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
3888     0, "Check for interlock across VOPs");
3889 
3890 int vfs_badlock_print = 1;	/* Print lock violations. */
3891 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
3892     0, "Print lock violations");
3893 
3894 #ifdef KDB
3895 int vfs_badlock_backtrace = 1;	/* Print backtrace at lock violations. */
3896 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
3897     &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
3898 #endif
3899 
3900 static void
3901 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3902 {
3903 
3904 #ifdef KDB
3905 	if (vfs_badlock_backtrace)
3906 		kdb_backtrace();
3907 #endif
3908 	if (vfs_badlock_print)
3909 		printf("%s: %p %s\n", str, (void *)vp, msg);
3910 	if (vfs_badlock_ddb)
3911 		kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3912 }
3913 
3914 void
3915 assert_vi_locked(struct vnode *vp, const char *str)
3916 {
3917 
3918 	if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3919 		vfs_badlock("interlock is not locked but should be", str, vp);
3920 }
3921 
3922 void
3923 assert_vi_unlocked(struct vnode *vp, const char *str)
3924 {
3925 
3926 	if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3927 		vfs_badlock("interlock is locked but should not be", str, vp);
3928 }
3929 
3930 void
3931 assert_vop_locked(struct vnode *vp, const char *str)
3932 {
3933 	int locked;
3934 
3935 	if (!IGNORE_LOCK(vp)) {
3936 		locked = VOP_ISLOCKED(vp);
3937 		if (locked == 0 || locked == LK_EXCLOTHER)
3938 			vfs_badlock("is not locked but should be", str, vp);
3939 	}
3940 }
3941 
3942 void
3943 assert_vop_unlocked(struct vnode *vp, const char *str)
3944 {
3945 
3946 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
3947 		vfs_badlock("is locked but should not be", str, vp);
3948 }
3949 
3950 void
3951 assert_vop_elocked(struct vnode *vp, const char *str)
3952 {
3953 
3954 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
3955 		vfs_badlock("is not exclusive locked but should be", str, vp);
3956 }
3957 
3958 #if 0
3959 void
3960 assert_vop_elocked_other(struct vnode *vp, const char *str)
3961 {
3962 
3963 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
3964 		vfs_badlock("is not exclusive locked by another thread",
3965 		    str, vp);
3966 }
3967 
3968 void
3969 assert_vop_slocked(struct vnode *vp, const char *str)
3970 {
3971 
3972 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
3973 		vfs_badlock("is not locked shared but should be", str, vp);
3974 }
3975 #endif /* 0 */
3976 #endif /* DEBUG_VFS_LOCKS */
3977 
3978 void
3979 vop_rename_fail(struct vop_rename_args *ap)
3980 {
3981 
3982 	if (ap->a_tvp != NULL)
3983 		vput(ap->a_tvp);
3984 	if (ap->a_tdvp == ap->a_tvp)
3985 		vrele(ap->a_tdvp);
3986 	else
3987 		vput(ap->a_tdvp);
3988 	vrele(ap->a_fdvp);
3989 	vrele(ap->a_fvp);
3990 }
3991 
3992 void
3993 vop_rename_pre(void *ap)
3994 {
3995 	struct vop_rename_args *a = ap;
3996 
3997 #ifdef DEBUG_VFS_LOCKS
3998 	if (a->a_tvp)
3999 		ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
4000 	ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
4001 	ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
4002 	ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
4003 
4004 	/* Check the source (from). */
4005 	if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
4006 	    (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
4007 		ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
4008 	if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
4009 		ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
4010 
4011 	/* Check the target. */
4012 	if (a->a_tvp)
4013 		ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
4014 	ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
4015 #endif
4016 	if (a->a_tdvp != a->a_fdvp)
4017 		vhold(a->a_fdvp);
4018 	if (a->a_tvp != a->a_fvp)
4019 		vhold(a->a_fvp);
4020 	vhold(a->a_tdvp);
4021 	if (a->a_tvp)
4022 		vhold(a->a_tvp);
4023 }
4024 
4025 void
4026 vop_strategy_pre(void *ap)
4027 {
4028 #ifdef DEBUG_VFS_LOCKS
4029 	struct vop_strategy_args *a;
4030 	struct buf *bp;
4031 
4032 	a = ap;
4033 	bp = a->a_bp;
4034 
4035 	/*
4036 	 * Cluster ops lock their component buffers but not the IO container.
4037 	 */
4038 	if ((bp->b_flags & B_CLUSTER) != 0)
4039 		return;
4040 
4041 	if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
4042 		if (vfs_badlock_print)
4043 			printf(
4044 			    "VOP_STRATEGY: bp is not locked but should be\n");
4045 		if (vfs_badlock_ddb)
4046 			kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
4047 	}
4048 #endif
4049 }
4050 
4051 void
4052 vop_lock_pre(void *ap)
4053 {
4054 #ifdef DEBUG_VFS_LOCKS
4055 	struct vop_lock1_args *a = ap;
4056 
4057 	if ((a->a_flags & LK_INTERLOCK) == 0)
4058 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4059 	else
4060 		ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
4061 #endif
4062 }
4063 
4064 void
4065 vop_lock_post(void *ap, int rc)
4066 {
4067 #ifdef DEBUG_VFS_LOCKS
4068 	struct vop_lock1_args *a = ap;
4069 
4070 	ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4071 	if (rc == 0 && (a->a_flags & LK_EXCLOTHER) == 0)
4072 		ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
4073 #endif
4074 }
4075 
4076 void
4077 vop_unlock_pre(void *ap)
4078 {
4079 #ifdef DEBUG_VFS_LOCKS
4080 	struct vop_unlock_args *a = ap;
4081 
4082 	if (a->a_flags & LK_INTERLOCK)
4083 		ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
4084 	ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
4085 #endif
4086 }
4087 
4088 void
4089 vop_unlock_post(void *ap, int rc)
4090 {
4091 #ifdef DEBUG_VFS_LOCKS
4092 	struct vop_unlock_args *a = ap;
4093 
4094 	if (a->a_flags & LK_INTERLOCK)
4095 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
4096 #endif
4097 }
4098 
4099 void
4100 vop_create_post(void *ap, int rc)
4101 {
4102 	struct vop_create_args *a = ap;
4103 
4104 	if (!rc)
4105 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4106 }
4107 
4108 void
4109 vop_deleteextattr_post(void *ap, int rc)
4110 {
4111 	struct vop_deleteextattr_args *a = ap;
4112 
4113 	if (!rc)
4114 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4115 }
4116 
4117 void
4118 vop_link_post(void *ap, int rc)
4119 {
4120 	struct vop_link_args *a = ap;
4121 
4122 	if (!rc) {
4123 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
4124 		VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
4125 	}
4126 }
4127 
4128 void
4129 vop_mkdir_post(void *ap, int rc)
4130 {
4131 	struct vop_mkdir_args *a = ap;
4132 
4133 	if (!rc)
4134 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4135 }
4136 
4137 void
4138 vop_mknod_post(void *ap, int rc)
4139 {
4140 	struct vop_mknod_args *a = ap;
4141 
4142 	if (!rc)
4143 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4144 }
4145 
4146 void
4147 vop_remove_post(void *ap, int rc)
4148 {
4149 	struct vop_remove_args *a = ap;
4150 
4151 	if (!rc) {
4152 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4153 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4154 	}
4155 }
4156 
4157 void
4158 vop_rename_post(void *ap, int rc)
4159 {
4160 	struct vop_rename_args *a = ap;
4161 
4162 	if (!rc) {
4163 		VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
4164 		VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
4165 		VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
4166 		if (a->a_tvp)
4167 			VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
4168 	}
4169 	if (a->a_tdvp != a->a_fdvp)
4170 		vdrop(a->a_fdvp);
4171 	if (a->a_tvp != a->a_fvp)
4172 		vdrop(a->a_fvp);
4173 	vdrop(a->a_tdvp);
4174 	if (a->a_tvp)
4175 		vdrop(a->a_tvp);
4176 }
4177 
4178 void
4179 vop_rmdir_post(void *ap, int rc)
4180 {
4181 	struct vop_rmdir_args *a = ap;
4182 
4183 	if (!rc) {
4184 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4185 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4186 	}
4187 }
4188 
4189 void
4190 vop_setattr_post(void *ap, int rc)
4191 {
4192 	struct vop_setattr_args *a = ap;
4193 
4194 	if (!rc)
4195 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4196 }
4197 
4198 void
4199 vop_setextattr_post(void *ap, int rc)
4200 {
4201 	struct vop_setextattr_args *a = ap;
4202 
4203 	if (!rc)
4204 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4205 }
4206 
4207 void
4208 vop_symlink_post(void *ap, int rc)
4209 {
4210 	struct vop_symlink_args *a = ap;
4211 
4212 	if (!rc)
4213 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4214 }
4215 
4216 static struct knlist fs_knlist;
4217 
4218 static void
4219 vfs_event_init(void *arg)
4220 {
4221 	knlist_init_mtx(&fs_knlist, NULL);
4222 }
4223 /* XXX - correct order? */
4224 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4225 
4226 void
4227 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
4228 {
4229 
4230 	KNOTE_UNLOCKED(&fs_knlist, event);
4231 }
4232 
4233 static int	filt_fsattach(struct knote *kn);
4234 static void	filt_fsdetach(struct knote *kn);
4235 static int	filt_fsevent(struct knote *kn, long hint);
4236 
4237 struct filterops fs_filtops = {
4238 	.f_isfd = 0,
4239 	.f_attach = filt_fsattach,
4240 	.f_detach = filt_fsdetach,
4241 	.f_event = filt_fsevent
4242 };
4243 
4244 static int
4245 filt_fsattach(struct knote *kn)
4246 {
4247 
4248 	kn->kn_flags |= EV_CLEAR;
4249 	knlist_add(&fs_knlist, kn, 0);
4250 	return (0);
4251 }
4252 
4253 static void
4254 filt_fsdetach(struct knote *kn)
4255 {
4256 
4257 	knlist_remove(&fs_knlist, kn, 0);
4258 }
4259 
4260 static int
4261 filt_fsevent(struct knote *kn, long hint)
4262 {
4263 
4264 	kn->kn_fflags |= hint;
4265 	return (kn->kn_fflags != 0);
4266 }
4267 
4268 static int
4269 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4270 {
4271 	struct vfsidctl vc;
4272 	int error;
4273 	struct mount *mp;
4274 
4275 	error = SYSCTL_IN(req, &vc, sizeof(vc));
4276 	if (error)
4277 		return (error);
4278 	if (vc.vc_vers != VFS_CTL_VERS1)
4279 		return (EINVAL);
4280 	mp = vfs_getvfs(&vc.vc_fsid);
4281 	if (mp == NULL)
4282 		return (ENOENT);
4283 	/* ensure that a specific sysctl goes to the right filesystem. */
4284 	if (strcmp(vc.vc_fstypename, "*") != 0 &&
4285 	    strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4286 		vfs_rel(mp);
4287 		return (EINVAL);
4288 	}
4289 	VCTLTOREQ(&vc, req);
4290 	error = VFS_SYSCTL(mp, vc.vc_op, req);
4291 	vfs_rel(mp);
4292 	return (error);
4293 }
4294 
4295 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
4296     NULL, 0, sysctl_vfs_ctl, "",
4297     "Sysctl by fsid");
4298 
4299 /*
4300  * Function to initialize a va_filerev field sensibly.
4301  * XXX: Wouldn't a random number make a lot more sense ??
4302  */
4303 u_quad_t
4304 init_va_filerev(void)
4305 {
4306 	struct bintime bt;
4307 
4308 	getbinuptime(&bt);
4309 	return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4310 }
4311 
4312 static int	filt_vfsread(struct knote *kn, long hint);
4313 static int	filt_vfswrite(struct knote *kn, long hint);
4314 static int	filt_vfsvnode(struct knote *kn, long hint);
4315 static void	filt_vfsdetach(struct knote *kn);
4316 static struct filterops vfsread_filtops = {
4317 	.f_isfd = 1,
4318 	.f_detach = filt_vfsdetach,
4319 	.f_event = filt_vfsread
4320 };
4321 static struct filterops vfswrite_filtops = {
4322 	.f_isfd = 1,
4323 	.f_detach = filt_vfsdetach,
4324 	.f_event = filt_vfswrite
4325 };
4326 static struct filterops vfsvnode_filtops = {
4327 	.f_isfd = 1,
4328 	.f_detach = filt_vfsdetach,
4329 	.f_event = filt_vfsvnode
4330 };
4331 
4332 static void
4333 vfs_knllock(void *arg)
4334 {
4335 	struct vnode *vp = arg;
4336 
4337 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4338 }
4339 
4340 static void
4341 vfs_knlunlock(void *arg)
4342 {
4343 	struct vnode *vp = arg;
4344 
4345 	VOP_UNLOCK(vp, 0);
4346 }
4347 
4348 static void
4349 vfs_knl_assert_locked(void *arg)
4350 {
4351 #ifdef DEBUG_VFS_LOCKS
4352 	struct vnode *vp = arg;
4353 
4354 	ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4355 #endif
4356 }
4357 
4358 static void
4359 vfs_knl_assert_unlocked(void *arg)
4360 {
4361 #ifdef DEBUG_VFS_LOCKS
4362 	struct vnode *vp = arg;
4363 
4364 	ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4365 #endif
4366 }
4367 
4368 int
4369 vfs_kqfilter(struct vop_kqfilter_args *ap)
4370 {
4371 	struct vnode *vp = ap->a_vp;
4372 	struct knote *kn = ap->a_kn;
4373 	struct knlist *knl;
4374 
4375 	switch (kn->kn_filter) {
4376 	case EVFILT_READ:
4377 		kn->kn_fop = &vfsread_filtops;
4378 		break;
4379 	case EVFILT_WRITE:
4380 		kn->kn_fop = &vfswrite_filtops;
4381 		break;
4382 	case EVFILT_VNODE:
4383 		kn->kn_fop = &vfsvnode_filtops;
4384 		break;
4385 	default:
4386 		return (EINVAL);
4387 	}
4388 
4389 	kn->kn_hook = (caddr_t)vp;
4390 
4391 	v_addpollinfo(vp);
4392 	if (vp->v_pollinfo == NULL)
4393 		return (ENOMEM);
4394 	knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4395 	knlist_add(knl, kn, 0);
4396 
4397 	return (0);
4398 }
4399 
4400 /*
4401  * Detach knote from vnode
4402  */
4403 static void
4404 filt_vfsdetach(struct knote *kn)
4405 {
4406 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4407 
4408 	KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4409 	knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4410 }
4411 
4412 /*ARGSUSED*/
4413 static int
4414 filt_vfsread(struct knote *kn, long hint)
4415 {
4416 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4417 	struct vattr va;
4418 	int res;
4419 
4420 	/*
4421 	 * filesystem is gone, so set the EOF flag and schedule
4422 	 * the knote for deletion.
4423 	 */
4424 	if (hint == NOTE_REVOKE) {
4425 		VI_LOCK(vp);
4426 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4427 		VI_UNLOCK(vp);
4428 		return (1);
4429 	}
4430 
4431 	if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4432 		return (0);
4433 
4434 	VI_LOCK(vp);
4435 	kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4436 	res = (kn->kn_data != 0);
4437 	VI_UNLOCK(vp);
4438 	return (res);
4439 }
4440 
4441 /*ARGSUSED*/
4442 static int
4443 filt_vfswrite(struct knote *kn, long hint)
4444 {
4445 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4446 
4447 	VI_LOCK(vp);
4448 
4449 	/*
4450 	 * filesystem is gone, so set the EOF flag and schedule
4451 	 * the knote for deletion.
4452 	 */
4453 	if (hint == NOTE_REVOKE)
4454 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4455 
4456 	kn->kn_data = 0;
4457 	VI_UNLOCK(vp);
4458 	return (1);
4459 }
4460 
4461 static int
4462 filt_vfsvnode(struct knote *kn, long hint)
4463 {
4464 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4465 	int res;
4466 
4467 	VI_LOCK(vp);
4468 	if (kn->kn_sfflags & hint)
4469 		kn->kn_fflags |= hint;
4470 	if (hint == NOTE_REVOKE) {
4471 		kn->kn_flags |= EV_EOF;
4472 		VI_UNLOCK(vp);
4473 		return (1);
4474 	}
4475 	res = (kn->kn_fflags != 0);
4476 	VI_UNLOCK(vp);
4477 	return (res);
4478 }
4479 
4480 int
4481 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4482 {
4483 	int error;
4484 
4485 	if (dp->d_reclen > ap->a_uio->uio_resid)
4486 		return (ENAMETOOLONG);
4487 	error = uiomove(dp, dp->d_reclen, ap->a_uio);
4488 	if (error) {
4489 		if (ap->a_ncookies != NULL) {
4490 			if (ap->a_cookies != NULL)
4491 				free(ap->a_cookies, M_TEMP);
4492 			ap->a_cookies = NULL;
4493 			*ap->a_ncookies = 0;
4494 		}
4495 		return (error);
4496 	}
4497 	if (ap->a_ncookies == NULL)
4498 		return (0);
4499 
4500 	KASSERT(ap->a_cookies,
4501 	    ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4502 
4503 	*ap->a_cookies = realloc(*ap->a_cookies,
4504 	    (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4505 	(*ap->a_cookies)[*ap->a_ncookies] = off;
4506 	return (0);
4507 }
4508 
4509 /*
4510  * Mark for update the access time of the file if the filesystem
4511  * supports VOP_MARKATIME.  This functionality is used by execve and
4512  * mmap, so we want to avoid the I/O implied by directly setting
4513  * va_atime for the sake of efficiency.
4514  */
4515 void
4516 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4517 {
4518 	struct mount *mp;
4519 
4520 	mp = vp->v_mount;
4521 	ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4522 	if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4523 		(void)VOP_MARKATIME(vp);
4524 }
4525 
4526 /*
4527  * The purpose of this routine is to remove granularity from accmode_t,
4528  * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4529  * VADMIN and VAPPEND.
4530  *
4531  * If it returns 0, the caller is supposed to continue with the usual
4532  * access checks using 'accmode' as modified by this routine.  If it
4533  * returns nonzero value, the caller is supposed to return that value
4534  * as errno.
4535  *
4536  * Note that after this routine runs, accmode may be zero.
4537  */
4538 int
4539 vfs_unixify_accmode(accmode_t *accmode)
4540 {
4541 	/*
4542 	 * There is no way to specify explicit "deny" rule using
4543 	 * file mode or POSIX.1e ACLs.
4544 	 */
4545 	if (*accmode & VEXPLICIT_DENY) {
4546 		*accmode = 0;
4547 		return (0);
4548 	}
4549 
4550 	/*
4551 	 * None of these can be translated into usual access bits.
4552 	 * Also, the common case for NFSv4 ACLs is to not contain
4553 	 * either of these bits. Caller should check for VWRITE
4554 	 * on the containing directory instead.
4555 	 */
4556 	if (*accmode & (VDELETE_CHILD | VDELETE))
4557 		return (EPERM);
4558 
4559 	if (*accmode & VADMIN_PERMS) {
4560 		*accmode &= ~VADMIN_PERMS;
4561 		*accmode |= VADMIN;
4562 	}
4563 
4564 	/*
4565 	 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4566 	 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4567 	 */
4568 	*accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
4569 
4570 	return (0);
4571 }
4572 
4573 /*
4574  * These are helper functions for filesystems to traverse all
4575  * their vnodes.  See MNT_VNODE_FOREACH_ALL() in sys/mount.h.
4576  *
4577  * This interface replaces MNT_VNODE_FOREACH.
4578  */
4579 
4580 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
4581 
4582 struct vnode *
4583 __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp)
4584 {
4585 	struct vnode *vp;
4586 
4587 	if (should_yield())
4588 		kern_yield(PRI_USER);
4589 	MNT_ILOCK(mp);
4590 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4591 	vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
4592 	while (vp != NULL && (vp->v_type == VMARKER ||
4593 	    (vp->v_iflag & VI_DOOMED) != 0))
4594 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
4595 
4596 	/* Check if we are done */
4597 	if (vp == NULL) {
4598 		__mnt_vnode_markerfree_all(mvp, mp);
4599 		/* MNT_IUNLOCK(mp); -- done in above function */
4600 		mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
4601 		return (NULL);
4602 	}
4603 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4604 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4605 	VI_LOCK(vp);
4606 	MNT_IUNLOCK(mp);
4607 	return (vp);
4608 }
4609 
4610 struct vnode *
4611 __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp)
4612 {
4613 	struct vnode *vp;
4614 
4615 	*mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4616 	MNT_ILOCK(mp);
4617 	MNT_REF(mp);
4618 	(*mvp)->v_type = VMARKER;
4619 
4620 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
4621 	while (vp != NULL && (vp->v_type == VMARKER ||
4622 	    (vp->v_iflag & VI_DOOMED) != 0))
4623 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
4624 
4625 	/* Check if we are done */
4626 	if (vp == NULL) {
4627 		MNT_REL(mp);
4628 		MNT_IUNLOCK(mp);
4629 		free(*mvp, M_VNODE_MARKER);
4630 		*mvp = NULL;
4631 		return (NULL);
4632 	}
4633 	(*mvp)->v_mount = mp;
4634 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4635 	VI_LOCK(vp);
4636 	MNT_IUNLOCK(mp);
4637 	return (vp);
4638 }
4639 
4640 
4641 void
4642 __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp)
4643 {
4644 
4645 	if (*mvp == NULL) {
4646 		MNT_IUNLOCK(mp);
4647 		return;
4648 	}
4649 
4650 	mtx_assert(MNT_MTX(mp), MA_OWNED);
4651 
4652 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4653 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4654 	MNT_REL(mp);
4655 	MNT_IUNLOCK(mp);
4656 	free(*mvp, M_VNODE_MARKER);
4657 	*mvp = NULL;
4658 }
4659 
4660 /*
4661  * These are helper functions for filesystems to traverse their
4662  * active vnodes.  See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h
4663  */
4664 static void
4665 mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4666 {
4667 
4668 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4669 
4670 	MNT_ILOCK(mp);
4671 	MNT_REL(mp);
4672 	MNT_IUNLOCK(mp);
4673 	free(*mvp, M_VNODE_MARKER);
4674 	*mvp = NULL;
4675 }
4676 
4677 static struct vnode *
4678 mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4679 {
4680 	struct vnode *vp, *nvp;
4681 
4682 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
4683 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4684 restart:
4685 	vp = TAILQ_NEXT(*mvp, v_actfreelist);
4686 	TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4687 	while (vp != NULL) {
4688 		if (vp->v_type == VMARKER) {
4689 			vp = TAILQ_NEXT(vp, v_actfreelist);
4690 			continue;
4691 		}
4692 		if (!VI_TRYLOCK(vp)) {
4693 			if (mp_ncpus == 1 || should_yield()) {
4694 				TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
4695 				mtx_unlock(&vnode_free_list_mtx);
4696 				kern_yield(PRI_USER);
4697 				mtx_lock(&vnode_free_list_mtx);
4698 				goto restart;
4699 			}
4700 			continue;
4701 		}
4702 		KASSERT(vp->v_type != VMARKER, ("locked marker %p", vp));
4703 		KASSERT(vp->v_mount == mp || vp->v_mount == NULL,
4704 		    ("alien vnode on the active list %p %p", vp, mp));
4705 		if (vp->v_mount == mp && (vp->v_iflag & VI_DOOMED) == 0)
4706 			break;
4707 		nvp = TAILQ_NEXT(vp, v_actfreelist);
4708 		VI_UNLOCK(vp);
4709 		vp = nvp;
4710 	}
4711 
4712 	/* Check if we are done */
4713 	if (vp == NULL) {
4714 		mtx_unlock(&vnode_free_list_mtx);
4715 		mnt_vnode_markerfree_active(mvp, mp);
4716 		return (NULL);
4717 	}
4718 	TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
4719 	mtx_unlock(&vnode_free_list_mtx);
4720 	ASSERT_VI_LOCKED(vp, "active iter");
4721 	KASSERT((vp->v_iflag & VI_ACTIVE) != 0, ("Non-active vp %p", vp));
4722 	return (vp);
4723 }
4724 
4725 struct vnode *
4726 __mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4727 {
4728 
4729 	if (should_yield())
4730 		kern_yield(PRI_USER);
4731 	mtx_lock(&vnode_free_list_mtx);
4732 	return (mnt_vnode_next_active(mvp, mp));
4733 }
4734 
4735 struct vnode *
4736 __mnt_vnode_first_active(struct vnode **mvp, struct mount *mp)
4737 {
4738 	struct vnode *vp;
4739 
4740 	*mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4741 	MNT_ILOCK(mp);
4742 	MNT_REF(mp);
4743 	MNT_IUNLOCK(mp);
4744 	(*mvp)->v_type = VMARKER;
4745 	(*mvp)->v_mount = mp;
4746 
4747 	mtx_lock(&vnode_free_list_mtx);
4748 	vp = TAILQ_FIRST(&mp->mnt_activevnodelist);
4749 	if (vp == NULL) {
4750 		mtx_unlock(&vnode_free_list_mtx);
4751 		mnt_vnode_markerfree_active(mvp, mp);
4752 		return (NULL);
4753 	}
4754 	TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
4755 	return (mnt_vnode_next_active(mvp, mp));
4756 }
4757 
4758 void
4759 __mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4760 {
4761 
4762 	if (*mvp == NULL)
4763 		return;
4764 
4765 	mtx_lock(&vnode_free_list_mtx);
4766 	TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4767 	mtx_unlock(&vnode_free_list_mtx);
4768 	mnt_vnode_markerfree_active(mvp, mp);
4769 }
4770