xref: /freebsd/sys/kern/vfs_subr.c (revision 145992504973bd16cf3518af9ba5ce185fefa82a)
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/priv.h>
69 #include <sys/reboot.h>
70 #include <sys/sched.h>
71 #include <sys/sleepqueue.h>
72 #include <sys/stat.h>
73 #include <sys/sysctl.h>
74 #include <sys/syslog.h>
75 #include <sys/vmmeter.h>
76 #include <sys/vnode.h>
77 #include <sys/watchdog.h>
78 
79 #include <machine/stdarg.h>
80 
81 #include <security/mac/mac_framework.h>
82 
83 #include <vm/vm.h>
84 #include <vm/vm_object.h>
85 #include <vm/vm_extern.h>
86 #include <vm/pmap.h>
87 #include <vm/vm_map.h>
88 #include <vm/vm_page.h>
89 #include <vm/vm_kern.h>
90 #include <vm/uma.h>
91 
92 #ifdef DDB
93 #include <ddb/ddb.h>
94 #endif
95 
96 static void	delmntque(struct vnode *vp);
97 static int	flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
98 		    int slpflag, int slptimeo);
99 static void	syncer_shutdown(void *arg, int howto);
100 static int	vtryrecycle(struct vnode *vp);
101 static void	v_incr_usecount(struct vnode *);
102 static void	v_decr_usecount(struct vnode *);
103 static void	v_decr_useonly(struct vnode *);
104 static void	v_upgrade_usecount(struct vnode *);
105 static void	vnlru_free(int);
106 static void	vgonel(struct vnode *);
107 static void	vfs_knllock(void *arg);
108 static void	vfs_knlunlock(void *arg);
109 static void	vfs_knl_assert_locked(void *arg);
110 static void	vfs_knl_assert_unlocked(void *arg);
111 static void	destroy_vpollinfo(struct vpollinfo *vi);
112 
113 /*
114  * Number of vnodes in existence.  Increased whenever getnewvnode()
115  * allocates a new vnode, decreased in vdropl() for VI_DOOMED vnode.
116  */
117 static unsigned long	numvnodes;
118 
119 SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
120     "Number of vnodes in existence");
121 
122 /*
123  * Conversion tables for conversion from vnode types to inode formats
124  * and back.
125  */
126 enum vtype iftovt_tab[16] = {
127 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
128 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
129 };
130 int vttoif_tab[10] = {
131 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
132 	S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
133 };
134 
135 /*
136  * List of vnodes that are ready for recycling.
137  */
138 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
139 
140 /*
141  * Free vnode target.  Free vnodes may simply be files which have been stat'd
142  * but not read.  This is somewhat common, and a small cache of such files
143  * should be kept to avoid recreation costs.
144  */
145 static u_long wantfreevnodes;
146 SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
147 /* Number of vnodes in the free list. */
148 static u_long freevnodes;
149 SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
150     "Number of vnodes in the free list");
151 
152 static int vlru_allow_cache_src;
153 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
154     &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
155 
156 /*
157  * Various variables used for debugging the new implementation of
158  * reassignbuf().
159  * XXX these are probably of (very) limited utility now.
160  */
161 static int reassignbufcalls;
162 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
163     "Number of calls to reassignbuf");
164 
165 /*
166  * Cache for the mount type id assigned to NFS.  This is used for
167  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
168  */
169 int	nfs_mount_type = -1;
170 
171 /* To keep more than one thread at a time from running vfs_getnewfsid */
172 static struct mtx mntid_mtx;
173 
174 /*
175  * Lock for any access to the following:
176  *	vnode_free_list
177  *	numvnodes
178  *	freevnodes
179  */
180 static struct mtx vnode_free_list_mtx;
181 
182 /* Publicly exported FS */
183 struct nfs_public nfs_pub;
184 
185 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
186 static uma_zone_t vnode_zone;
187 static uma_zone_t vnodepoll_zone;
188 
189 /*
190  * The workitem queue.
191  *
192  * It is useful to delay writes of file data and filesystem metadata
193  * for tens of seconds so that quickly created and deleted files need
194  * not waste disk bandwidth being created and removed. To realize this,
195  * we append vnodes to a "workitem" queue. When running with a soft
196  * updates implementation, most pending metadata dependencies should
197  * not wait for more than a few seconds. Thus, mounted on block devices
198  * are delayed only about a half the time that file data is delayed.
199  * Similarly, directory updates are more critical, so are only delayed
200  * about a third the time that file data is delayed. Thus, there are
201  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
202  * one each second (driven off the filesystem syncer process). The
203  * syncer_delayno variable indicates the next queue that is to be processed.
204  * Items that need to be processed soon are placed in this queue:
205  *
206  *	syncer_workitem_pending[syncer_delayno]
207  *
208  * A delay of fifteen seconds is done by placing the request fifteen
209  * entries later in the queue:
210  *
211  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
212  *
213  */
214 static int syncer_delayno;
215 static long syncer_mask;
216 LIST_HEAD(synclist, bufobj);
217 static struct synclist *syncer_workitem_pending;
218 /*
219  * The sync_mtx protects:
220  *	bo->bo_synclist
221  *	sync_vnode_count
222  *	syncer_delayno
223  *	syncer_state
224  *	syncer_workitem_pending
225  *	syncer_worklist_len
226  *	rushjob
227  */
228 static struct mtx sync_mtx;
229 static struct cv sync_wakeup;
230 
231 #define SYNCER_MAXDELAY		32
232 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
233 static int syncdelay = 30;		/* max time to delay syncing data */
234 static int filedelay = 30;		/* time to delay syncing files */
235 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
236     "Time to delay syncing files (in seconds)");
237 static int dirdelay = 29;		/* time to delay syncing directories */
238 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
239     "Time to delay syncing directories (in seconds)");
240 static int metadelay = 28;		/* time to delay syncing metadata */
241 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
242     "Time to delay syncing metadata (in seconds)");
243 static int rushjob;		/* number of slots to run ASAP */
244 static int stat_rush_requests;	/* number of times I/O speeded up */
245 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
246     "Number of times I/O speeded up (rush requests)");
247 
248 /*
249  * When shutting down the syncer, run it at four times normal speed.
250  */
251 #define SYNCER_SHUTDOWN_SPEEDUP		4
252 static int sync_vnode_count;
253 static int syncer_worklist_len;
254 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
255     syncer_state;
256 
257 /*
258  * Number of vnodes we want to exist at any one time.  This is mostly used
259  * to size hash tables in vnode-related code.  It is normally not used in
260  * getnewvnode(), as wantfreevnodes is normally nonzero.)
261  *
262  * XXX desiredvnodes is historical cruft and should not exist.
263  */
264 int desiredvnodes;
265 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
266     &desiredvnodes, 0, "Maximum number of vnodes");
267 SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
268     &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
269 static int vnlru_nowhere;
270 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
271     &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
272 
273 /*
274  * Macros to control when a vnode is freed and recycled.  All require
275  * the vnode interlock.
276  */
277 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
278 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
279 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
280 
281 
282 /*
283  * Initialize the vnode management data structures.
284  *
285  * Reevaluate the following cap on the number of vnodes after the physical
286  * memory size exceeds 512GB.  In the limit, as the physical memory size
287  * grows, the ratio of physical pages to vnodes approaches sixteen to one.
288  */
289 #ifndef	MAXVNODES_MAX
290 #define	MAXVNODES_MAX	(512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
291 #endif
292 static void
293 vntblinit(void *dummy __unused)
294 {
295 	int physvnodes, virtvnodes;
296 
297 	/*
298 	 * Desiredvnodes is a function of the physical memory size and the
299 	 * kernel's heap size.  Generally speaking, it scales with the
300 	 * physical memory size.  The ratio of desiredvnodes to physical pages
301 	 * is one to four until desiredvnodes exceeds 98,304.  Thereafter, the
302 	 * marginal ratio of desiredvnodes to physical pages is one to
303 	 * sixteen.  However, desiredvnodes is limited by the kernel's heap
304 	 * size.  The memory required by desiredvnodes vnodes and vm objects
305 	 * may not exceed one seventh of the kernel's heap size.
306 	 */
307 	physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
308 	    cnt.v_page_count) / 16;
309 	virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
310 	    sizeof(struct vnode)));
311 	desiredvnodes = min(physvnodes, virtvnodes);
312 	if (desiredvnodes > MAXVNODES_MAX) {
313 		if (bootverbose)
314 			printf("Reducing kern.maxvnodes %d -> %d\n",
315 			    desiredvnodes, MAXVNODES_MAX);
316 		desiredvnodes = MAXVNODES_MAX;
317 	}
318 	wantfreevnodes = desiredvnodes / 4;
319 	mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
320 	TAILQ_INIT(&vnode_free_list);
321 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
322 	vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
323 	    NULL, NULL, UMA_ALIGN_PTR, 0);
324 	vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
325 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
326 	/*
327 	 * Initialize the filesystem syncer.
328 	 */
329 	syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
330 	    &syncer_mask);
331 	syncer_maxdelay = syncer_mask + 1;
332 	mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
333 	cv_init(&sync_wakeup, "syncer");
334 }
335 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
336 
337 
338 /*
339  * Mark a mount point as busy. Used to synchronize access and to delay
340  * unmounting. Eventually, mountlist_mtx is not released on failure.
341  *
342  * vfs_busy() is a custom lock, it can block the caller.
343  * vfs_busy() only sleeps if the unmount is active on the mount point.
344  * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
345  * vnode belonging to mp.
346  *
347  * Lookup uses vfs_busy() to traverse mount points.
348  * root fs			var fs
349  * / vnode lock		A	/ vnode lock (/var)		D
350  * /var vnode lock	B	/log vnode lock(/var/log)	E
351  * vfs_busy lock	C	vfs_busy lock			F
352  *
353  * Within each file system, the lock order is C->A->B and F->D->E.
354  *
355  * When traversing across mounts, the system follows that lock order:
356  *
357  *        C->A->B
358  *              |
359  *              +->F->D->E
360  *
361  * The lookup() process for namei("/var") illustrates the process:
362  *  VOP_LOOKUP() obtains B while A is held
363  *  vfs_busy() obtains a shared lock on F while A and B are held
364  *  vput() releases lock on B
365  *  vput() releases lock on A
366  *  VFS_ROOT() obtains lock on D while shared lock on F is held
367  *  vfs_unbusy() releases shared lock on F
368  *  vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
369  *    Attempt to lock A (instead of vp_crossmp) while D is held would
370  *    violate the global order, causing deadlocks.
371  *
372  * dounmount() locks B while F is drained.
373  */
374 int
375 vfs_busy(struct mount *mp, int flags)
376 {
377 
378 	MPASS((flags & ~MBF_MASK) == 0);
379 	CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
380 
381 	MNT_ILOCK(mp);
382 	MNT_REF(mp);
383 	/*
384 	 * If mount point is currenly being unmounted, sleep until the
385 	 * mount point fate is decided.  If thread doing the unmounting fails,
386 	 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
387 	 * that this mount point has survived the unmount attempt and vfs_busy
388 	 * should retry.  Otherwise the unmounter thread will set MNTK_REFEXPIRE
389 	 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
390 	 * about to be really destroyed.  vfs_busy needs to release its
391 	 * reference on the mount point in this case and return with ENOENT,
392 	 * telling the caller that mount mount it tried to busy is no longer
393 	 * valid.
394 	 */
395 	while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
396 		if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
397 			MNT_REL(mp);
398 			MNT_IUNLOCK(mp);
399 			CTR1(KTR_VFS, "%s: failed busying before sleeping",
400 			    __func__);
401 			return (ENOENT);
402 		}
403 		if (flags & MBF_MNTLSTLOCK)
404 			mtx_unlock(&mountlist_mtx);
405 		mp->mnt_kern_flag |= MNTK_MWAIT;
406 		msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
407 		if (flags & MBF_MNTLSTLOCK)
408 			mtx_lock(&mountlist_mtx);
409 		MNT_ILOCK(mp);
410 	}
411 	if (flags & MBF_MNTLSTLOCK)
412 		mtx_unlock(&mountlist_mtx);
413 	mp->mnt_lockref++;
414 	MNT_IUNLOCK(mp);
415 	return (0);
416 }
417 
418 /*
419  * Free a busy filesystem.
420  */
421 void
422 vfs_unbusy(struct mount *mp)
423 {
424 
425 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
426 	MNT_ILOCK(mp);
427 	MNT_REL(mp);
428 	KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
429 	mp->mnt_lockref--;
430 	if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
431 		MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
432 		CTR1(KTR_VFS, "%s: waking up waiters", __func__);
433 		mp->mnt_kern_flag &= ~MNTK_DRAINING;
434 		wakeup(&mp->mnt_lockref);
435 	}
436 	MNT_IUNLOCK(mp);
437 }
438 
439 /*
440  * Lookup a mount point by filesystem identifier.
441  */
442 struct mount *
443 vfs_getvfs(fsid_t *fsid)
444 {
445 	struct mount *mp;
446 
447 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
448 	mtx_lock(&mountlist_mtx);
449 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
450 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
451 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
452 			vfs_ref(mp);
453 			mtx_unlock(&mountlist_mtx);
454 			return (mp);
455 		}
456 	}
457 	mtx_unlock(&mountlist_mtx);
458 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
459 	return ((struct mount *) 0);
460 }
461 
462 /*
463  * Lookup a mount point by filesystem identifier, busying it before
464  * returning.
465  */
466 struct mount *
467 vfs_busyfs(fsid_t *fsid)
468 {
469 	struct mount *mp;
470 	int error;
471 
472 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
473 	mtx_lock(&mountlist_mtx);
474 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
475 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
476 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
477 			error = vfs_busy(mp, MBF_MNTLSTLOCK);
478 			if (error) {
479 				mtx_unlock(&mountlist_mtx);
480 				return (NULL);
481 			}
482 			return (mp);
483 		}
484 	}
485 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
486 	mtx_unlock(&mountlist_mtx);
487 	return ((struct mount *) 0);
488 }
489 
490 /*
491  * Check if a user can access privileged mount options.
492  */
493 int
494 vfs_suser(struct mount *mp, struct thread *td)
495 {
496 	int error;
497 
498 	/*
499 	 * If the thread is jailed, but this is not a jail-friendly file
500 	 * system, deny immediately.
501 	 */
502 	if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
503 		return (EPERM);
504 
505 	/*
506 	 * If the file system was mounted outside the jail of the calling
507 	 * thread, deny immediately.
508 	 */
509 	if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
510 		return (EPERM);
511 
512 	/*
513 	 * If file system supports delegated administration, we don't check
514 	 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
515 	 * by the file system itself.
516 	 * If this is not the user that did original mount, we check for
517 	 * the PRIV_VFS_MOUNT_OWNER privilege.
518 	 */
519 	if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
520 	    mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
521 		if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
522 			return (error);
523 	}
524 	return (0);
525 }
526 
527 /*
528  * Get a new unique fsid.  Try to make its val[0] unique, since this value
529  * will be used to create fake device numbers for stat().  Also try (but
530  * not so hard) make its val[0] unique mod 2^16, since some emulators only
531  * support 16-bit device numbers.  We end up with unique val[0]'s for the
532  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
533  *
534  * Keep in mind that several mounts may be running in parallel.  Starting
535  * the search one past where the previous search terminated is both a
536  * micro-optimization and a defense against returning the same fsid to
537  * different mounts.
538  */
539 void
540 vfs_getnewfsid(struct mount *mp)
541 {
542 	static uint16_t mntid_base;
543 	struct mount *nmp;
544 	fsid_t tfsid;
545 	int mtype;
546 
547 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
548 	mtx_lock(&mntid_mtx);
549 	mtype = mp->mnt_vfc->vfc_typenum;
550 	tfsid.val[1] = mtype;
551 	mtype = (mtype & 0xFF) << 24;
552 	for (;;) {
553 		tfsid.val[0] = makedev(255,
554 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
555 		mntid_base++;
556 		if ((nmp = vfs_getvfs(&tfsid)) == NULL)
557 			break;
558 		vfs_rel(nmp);
559 	}
560 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
561 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
562 	mtx_unlock(&mntid_mtx);
563 }
564 
565 /*
566  * Knob to control the precision of file timestamps:
567  *
568  *   0 = seconds only; nanoseconds zeroed.
569  *   1 = seconds and nanoseconds, accurate within 1/HZ.
570  *   2 = seconds and nanoseconds, truncated to microseconds.
571  * >=3 = seconds and nanoseconds, maximum precision.
572  */
573 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
574 
575 static int timestamp_precision = TSP_SEC;
576 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
577     &timestamp_precision, 0, "File timestamp precision (0: seconds, "
578     "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
579     "3+: sec + ns (max. precision))");
580 
581 /*
582  * Get a current timestamp.
583  */
584 void
585 vfs_timestamp(struct timespec *tsp)
586 {
587 	struct timeval tv;
588 
589 	switch (timestamp_precision) {
590 	case TSP_SEC:
591 		tsp->tv_sec = time_second;
592 		tsp->tv_nsec = 0;
593 		break;
594 	case TSP_HZ:
595 		getnanotime(tsp);
596 		break;
597 	case TSP_USEC:
598 		microtime(&tv);
599 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
600 		break;
601 	case TSP_NSEC:
602 	default:
603 		nanotime(tsp);
604 		break;
605 	}
606 }
607 
608 /*
609  * Set vnode attributes to VNOVAL
610  */
611 void
612 vattr_null(struct vattr *vap)
613 {
614 
615 	vap->va_type = VNON;
616 	vap->va_size = VNOVAL;
617 	vap->va_bytes = VNOVAL;
618 	vap->va_mode = VNOVAL;
619 	vap->va_nlink = VNOVAL;
620 	vap->va_uid = VNOVAL;
621 	vap->va_gid = VNOVAL;
622 	vap->va_fsid = VNOVAL;
623 	vap->va_fileid = VNOVAL;
624 	vap->va_blocksize = VNOVAL;
625 	vap->va_rdev = VNOVAL;
626 	vap->va_atime.tv_sec = VNOVAL;
627 	vap->va_atime.tv_nsec = VNOVAL;
628 	vap->va_mtime.tv_sec = VNOVAL;
629 	vap->va_mtime.tv_nsec = VNOVAL;
630 	vap->va_ctime.tv_sec = VNOVAL;
631 	vap->va_ctime.tv_nsec = VNOVAL;
632 	vap->va_birthtime.tv_sec = VNOVAL;
633 	vap->va_birthtime.tv_nsec = VNOVAL;
634 	vap->va_flags = VNOVAL;
635 	vap->va_gen = VNOVAL;
636 	vap->va_vaflags = 0;
637 }
638 
639 /*
640  * This routine is called when we have too many vnodes.  It attempts
641  * to free <count> vnodes and will potentially free vnodes that still
642  * have VM backing store (VM backing store is typically the cause
643  * of a vnode blowout so we want to do this).  Therefore, this operation
644  * is not considered cheap.
645  *
646  * A number of conditions may prevent a vnode from being reclaimed.
647  * the buffer cache may have references on the vnode, a directory
648  * vnode may still have references due to the namei cache representing
649  * underlying files, or the vnode may be in active use.   It is not
650  * desireable to reuse such vnodes.  These conditions may cause the
651  * number of vnodes to reach some minimum value regardless of what
652  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
653  */
654 static int
655 vlrureclaim(struct mount *mp)
656 {
657 	struct vnode *vp;
658 	int done;
659 	int trigger;
660 	int usevnodes;
661 	int count;
662 
663 	/*
664 	 * Calculate the trigger point, don't allow user
665 	 * screwups to blow us up.   This prevents us from
666 	 * recycling vnodes with lots of resident pages.  We
667 	 * aren't trying to free memory, we are trying to
668 	 * free vnodes.
669 	 */
670 	usevnodes = desiredvnodes;
671 	if (usevnodes <= 0)
672 		usevnodes = 1;
673 	trigger = cnt.v_page_count * 2 / usevnodes;
674 	done = 0;
675 	vn_start_write(NULL, &mp, V_WAIT);
676 	MNT_ILOCK(mp);
677 	count = mp->mnt_nvnodelistsize / 10 + 1;
678 	while (count != 0) {
679 		vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
680 		while (vp != NULL && vp->v_type == VMARKER)
681 			vp = TAILQ_NEXT(vp, v_nmntvnodes);
682 		if (vp == NULL)
683 			break;
684 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
685 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
686 		--count;
687 		if (!VI_TRYLOCK(vp))
688 			goto next_iter;
689 		/*
690 		 * If it's been deconstructed already, it's still
691 		 * referenced, or it exceeds the trigger, skip it.
692 		 */
693 		if (vp->v_usecount ||
694 		    (!vlru_allow_cache_src &&
695 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
696 		    (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
697 		    vp->v_object->resident_page_count > trigger)) {
698 			VI_UNLOCK(vp);
699 			goto next_iter;
700 		}
701 		MNT_IUNLOCK(mp);
702 		vholdl(vp);
703 		if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
704 			vdrop(vp);
705 			goto next_iter_mntunlocked;
706 		}
707 		VI_LOCK(vp);
708 		/*
709 		 * v_usecount may have been bumped after VOP_LOCK() dropped
710 		 * the vnode interlock and before it was locked again.
711 		 *
712 		 * It is not necessary to recheck VI_DOOMED because it can
713 		 * only be set by another thread that holds both the vnode
714 		 * lock and vnode interlock.  If another thread has the
715 		 * vnode lock before we get to VOP_LOCK() and obtains the
716 		 * vnode interlock after VOP_LOCK() drops the vnode
717 		 * interlock, the other thread will be unable to drop the
718 		 * vnode lock before our VOP_LOCK() call fails.
719 		 */
720 		if (vp->v_usecount ||
721 		    (!vlru_allow_cache_src &&
722 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
723 		    (vp->v_object != NULL &&
724 		    vp->v_object->resident_page_count > trigger)) {
725 			VOP_UNLOCK(vp, LK_INTERLOCK);
726 			goto next_iter_mntunlocked;
727 		}
728 		KASSERT((vp->v_iflag & VI_DOOMED) == 0,
729 		    ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
730 		vgonel(vp);
731 		VOP_UNLOCK(vp, 0);
732 		vdropl(vp);
733 		done++;
734 next_iter_mntunlocked:
735 		if (!should_yield())
736 			goto relock_mnt;
737 		goto yield;
738 next_iter:
739 		if (!should_yield())
740 			continue;
741 		MNT_IUNLOCK(mp);
742 yield:
743 		kern_yield(PRI_UNCHANGED);
744 relock_mnt:
745 		MNT_ILOCK(mp);
746 	}
747 	MNT_IUNLOCK(mp);
748 	vn_finished_write(mp);
749 	return done;
750 }
751 
752 /*
753  * Attempt to keep the free list at wantfreevnodes length.
754  */
755 static void
756 vnlru_free(int count)
757 {
758 	struct vnode *vp;
759 
760 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
761 	for (; count > 0; count--) {
762 		vp = TAILQ_FIRST(&vnode_free_list);
763 		/*
764 		 * The list can be modified while the free_list_mtx
765 		 * has been dropped and vp could be NULL here.
766 		 */
767 		if (!vp)
768 			break;
769 		VNASSERT(vp->v_op != NULL, vp,
770 		    ("vnlru_free: vnode already reclaimed."));
771 		KASSERT((vp->v_iflag & VI_FREE) != 0,
772 		    ("Removing vnode not on freelist"));
773 		KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
774 		    ("Mangling active vnode"));
775 		TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
776 		/*
777 		 * Don't recycle if we can't get the interlock.
778 		 */
779 		if (!VI_TRYLOCK(vp)) {
780 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
781 			continue;
782 		}
783 		VNASSERT(VCANRECYCLE(vp), vp,
784 		    ("vp inconsistent on freelist"));
785 		freevnodes--;
786 		vp->v_iflag &= ~VI_FREE;
787 		vholdl(vp);
788 		mtx_unlock(&vnode_free_list_mtx);
789 		VI_UNLOCK(vp);
790 		vtryrecycle(vp);
791 		/*
792 		 * If the recycled succeeded this vdrop will actually free
793 		 * the vnode.  If not it will simply place it back on
794 		 * the free list.
795 		 */
796 		vdrop(vp);
797 		mtx_lock(&vnode_free_list_mtx);
798 	}
799 }
800 /*
801  * Attempt to recycle vnodes in a context that is always safe to block.
802  * Calling vlrurecycle() from the bowels of filesystem code has some
803  * interesting deadlock problems.
804  */
805 static struct proc *vnlruproc;
806 static int vnlruproc_sig;
807 
808 static void
809 vnlru_proc(void)
810 {
811 	struct mount *mp, *nmp;
812 	int done;
813 	struct proc *p = vnlruproc;
814 
815 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
816 	    SHUTDOWN_PRI_FIRST);
817 
818 	for (;;) {
819 		kproc_suspend_check(p);
820 		mtx_lock(&vnode_free_list_mtx);
821 		if (freevnodes > wantfreevnodes)
822 			vnlru_free(freevnodes - wantfreevnodes);
823 		if (numvnodes <= desiredvnodes * 9 / 10) {
824 			vnlruproc_sig = 0;
825 			wakeup(&vnlruproc_sig);
826 			msleep(vnlruproc, &vnode_free_list_mtx,
827 			    PVFS|PDROP, "vlruwt", hz);
828 			continue;
829 		}
830 		mtx_unlock(&vnode_free_list_mtx);
831 		done = 0;
832 		mtx_lock(&mountlist_mtx);
833 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
834 			if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
835 				nmp = TAILQ_NEXT(mp, mnt_list);
836 				continue;
837 			}
838 			done += vlrureclaim(mp);
839 			mtx_lock(&mountlist_mtx);
840 			nmp = TAILQ_NEXT(mp, mnt_list);
841 			vfs_unbusy(mp);
842 		}
843 		mtx_unlock(&mountlist_mtx);
844 		if (done == 0) {
845 #if 0
846 			/* These messages are temporary debugging aids */
847 			if (vnlru_nowhere < 5)
848 				printf("vnlru process getting nowhere..\n");
849 			else if (vnlru_nowhere == 5)
850 				printf("vnlru process messages stopped.\n");
851 #endif
852 			vnlru_nowhere++;
853 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
854 		} else
855 			kern_yield(PRI_UNCHANGED);
856 	}
857 }
858 
859 static struct kproc_desc vnlru_kp = {
860 	"vnlru",
861 	vnlru_proc,
862 	&vnlruproc
863 };
864 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
865     &vnlru_kp);
866 
867 /*
868  * Routines having to do with the management of the vnode table.
869  */
870 
871 /*
872  * Try to recycle a freed vnode.  We abort if anyone picks up a reference
873  * before we actually vgone().  This function must be called with the vnode
874  * held to prevent the vnode from being returned to the free list midway
875  * through vgone().
876  */
877 static int
878 vtryrecycle(struct vnode *vp)
879 {
880 	struct mount *vnmp;
881 
882 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
883 	VNASSERT(vp->v_holdcnt, vp,
884 	    ("vtryrecycle: Recycling vp %p without a reference.", vp));
885 	/*
886 	 * This vnode may found and locked via some other list, if so we
887 	 * can't recycle it yet.
888 	 */
889 	if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
890 		CTR2(KTR_VFS,
891 		    "%s: impossible to recycle, vp %p lock is already held",
892 		    __func__, vp);
893 		return (EWOULDBLOCK);
894 	}
895 	/*
896 	 * Don't recycle if its filesystem is being suspended.
897 	 */
898 	if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
899 		VOP_UNLOCK(vp, 0);
900 		CTR2(KTR_VFS,
901 		    "%s: impossible to recycle, cannot start the write for %p",
902 		    __func__, vp);
903 		return (EBUSY);
904 	}
905 	/*
906 	 * If we got this far, we need to acquire the interlock and see if
907 	 * anyone picked up this vnode from another list.  If not, we will
908 	 * mark it with DOOMED via vgonel() so that anyone who does find it
909 	 * will skip over it.
910 	 */
911 	VI_LOCK(vp);
912 	if (vp->v_usecount) {
913 		VOP_UNLOCK(vp, LK_INTERLOCK);
914 		vn_finished_write(vnmp);
915 		CTR2(KTR_VFS,
916 		    "%s: impossible to recycle, %p is already referenced",
917 		    __func__, vp);
918 		return (EBUSY);
919 	}
920 	if ((vp->v_iflag & VI_DOOMED) == 0)
921 		vgonel(vp);
922 	VOP_UNLOCK(vp, LK_INTERLOCK);
923 	vn_finished_write(vnmp);
924 	return (0);
925 }
926 
927 /*
928  * Wait for available vnodes.
929  */
930 static int
931 getnewvnode_wait(int suspended)
932 {
933 
934 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
935 	if (numvnodes > desiredvnodes) {
936 		if (suspended) {
937 			/*
938 			 * File system is beeing suspended, we cannot risk a
939 			 * deadlock here, so allocate new vnode anyway.
940 			 */
941 			if (freevnodes > wantfreevnodes)
942 				vnlru_free(freevnodes - wantfreevnodes);
943 			return (0);
944 		}
945 		if (vnlruproc_sig == 0) {
946 			vnlruproc_sig = 1;	/* avoid unnecessary wakeups */
947 			wakeup(vnlruproc);
948 		}
949 		msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
950 		    "vlruwk", hz);
951 	}
952 	return (numvnodes > desiredvnodes ? ENFILE : 0);
953 }
954 
955 void
956 getnewvnode_reserve(u_int count)
957 {
958 	struct thread *td;
959 
960 	td = curthread;
961 	mtx_lock(&vnode_free_list_mtx);
962 	while (count > 0) {
963 		if (getnewvnode_wait(0) == 0) {
964 			count--;
965 			td->td_vp_reserv++;
966 			numvnodes++;
967 		}
968 	}
969 	mtx_unlock(&vnode_free_list_mtx);
970 }
971 
972 void
973 getnewvnode_drop_reserve(void)
974 {
975 	struct thread *td;
976 
977 	td = curthread;
978 	mtx_lock(&vnode_free_list_mtx);
979 	KASSERT(numvnodes >= td->td_vp_reserv, ("reserve too large"));
980 	numvnodes -= td->td_vp_reserv;
981 	mtx_unlock(&vnode_free_list_mtx);
982 	td->td_vp_reserv = 0;
983 }
984 
985 /*
986  * Return the next vnode from the free list.
987  */
988 int
989 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
990     struct vnode **vpp)
991 {
992 	struct vnode *vp;
993 	struct bufobj *bo;
994 	struct thread *td;
995 	int error;
996 
997 	CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
998 	vp = NULL;
999 	td = curthread;
1000 	if (td->td_vp_reserv > 0) {
1001 		td->td_vp_reserv -= 1;
1002 		goto alloc;
1003 	}
1004 	mtx_lock(&vnode_free_list_mtx);
1005 	/*
1006 	 * Lend our context to reclaim vnodes if they've exceeded the max.
1007 	 */
1008 	if (freevnodes > wantfreevnodes)
1009 		vnlru_free(1);
1010 	error = getnewvnode_wait(mp != NULL && (mp->mnt_kern_flag &
1011 	    MNTK_SUSPEND));
1012 #if 0	/* XXX Not all VFS_VGET/ffs_vget callers check returns. */
1013 	if (error != 0) {
1014 		mtx_unlock(&vnode_free_list_mtx);
1015 		return (error);
1016 	}
1017 #endif
1018 	numvnodes++;
1019 	mtx_unlock(&vnode_free_list_mtx);
1020 alloc:
1021 	vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
1022 	/*
1023 	 * Setup locks.
1024 	 */
1025 	vp->v_vnlock = &vp->v_lock;
1026 	mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
1027 	/*
1028 	 * By default, don't allow shared locks unless filesystems
1029 	 * opt-in.
1030 	 */
1031 	lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
1032 	/*
1033 	 * Initialize bufobj.
1034 	 */
1035 	bo = &vp->v_bufobj;
1036 	bo->__bo_vnode = vp;
1037 	mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
1038 	bo->bo_ops = &buf_ops_bio;
1039 	bo->bo_private = vp;
1040 	TAILQ_INIT(&bo->bo_clean.bv_hd);
1041 	TAILQ_INIT(&bo->bo_dirty.bv_hd);
1042 	/*
1043 	 * Initialize namecache.
1044 	 */
1045 	LIST_INIT(&vp->v_cache_src);
1046 	TAILQ_INIT(&vp->v_cache_dst);
1047 	/*
1048 	 * Finalize various vnode identity bits.
1049 	 */
1050 	vp->v_type = VNON;
1051 	vp->v_tag = tag;
1052 	vp->v_op = vops;
1053 	v_incr_usecount(vp);
1054 	vp->v_data = NULL;
1055 #ifdef MAC
1056 	mac_vnode_init(vp);
1057 	if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1058 		mac_vnode_associate_singlelabel(mp, vp);
1059 	else if (mp == NULL && vops != &dead_vnodeops)
1060 		printf("NULL mp in getnewvnode()\n");
1061 #endif
1062 	if (mp != NULL) {
1063 		bo->bo_bsize = mp->mnt_stat.f_iosize;
1064 		if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1065 			vp->v_vflag |= VV_NOKNOTE;
1066 	}
1067 	rangelock_init(&vp->v_rl);
1068 
1069 	*vpp = vp;
1070 	return (0);
1071 }
1072 
1073 /*
1074  * Delete from old mount point vnode list, if on one.
1075  */
1076 static void
1077 delmntque(struct vnode *vp)
1078 {
1079 	struct mount *mp;
1080 	int active;
1081 
1082 	mp = vp->v_mount;
1083 	if (mp == NULL)
1084 		return;
1085 	MNT_ILOCK(mp);
1086 	VI_LOCK(vp);
1087 	KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize,
1088 	    ("Active vnode list size %d > Vnode list size %d",
1089 	     mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize));
1090 	active = vp->v_iflag & VI_ACTIVE;
1091 	vp->v_iflag &= ~VI_ACTIVE;
1092 	if (active) {
1093 		mtx_lock(&vnode_free_list_mtx);
1094 		TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist);
1095 		mp->mnt_activevnodelistsize--;
1096 		mtx_unlock(&vnode_free_list_mtx);
1097 	}
1098 	vp->v_mount = NULL;
1099 	VI_UNLOCK(vp);
1100 	VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1101 		("bad mount point vnode list size"));
1102 	TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1103 	mp->mnt_nvnodelistsize--;
1104 	MNT_REL(mp);
1105 	MNT_IUNLOCK(mp);
1106 }
1107 
1108 static void
1109 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1110 {
1111 
1112 	vp->v_data = NULL;
1113 	vp->v_op = &dead_vnodeops;
1114 	/* XXX non mp-safe fs may still call insmntque with vnode
1115 	   unlocked */
1116 	if (!VOP_ISLOCKED(vp))
1117 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1118 	vgone(vp);
1119 	vput(vp);
1120 }
1121 
1122 /*
1123  * Insert into list of vnodes for the new mount point, if available.
1124  */
1125 int
1126 insmntque1(struct vnode *vp, struct mount *mp,
1127 	void (*dtr)(struct vnode *, void *), void *dtr_arg)
1128 {
1129 	int locked;
1130 
1131 	KASSERT(vp->v_mount == NULL,
1132 		("insmntque: vnode already on per mount vnode list"));
1133 	VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1134 	ASSERT_VOP_ELOCKED(vp, "insmntque: non-locked vp");
1135 
1136 	/*
1137 	 * We acquire the vnode interlock early to ensure that the
1138 	 * vnode cannot be recycled by another process releasing a
1139 	 * holdcnt on it before we get it on both the vnode list
1140 	 * and the active vnode list. The mount mutex protects only
1141 	 * manipulation of the vnode list and the vnode freelist
1142 	 * mutex protects only manipulation of the active vnode list.
1143 	 * Hence the need to hold the vnode interlock throughout.
1144 	 */
1145 	MNT_ILOCK(mp);
1146 	VI_LOCK(vp);
1147 	if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1148 	    ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1149 	     mp->mnt_nvnodelistsize == 0)) {
1150 		locked = VOP_ISLOCKED(vp);
1151 		if (!locked || (locked == LK_EXCLUSIVE &&
1152 		     (vp->v_vflag & VV_FORCEINSMQ) == 0)) {
1153 			VI_UNLOCK(vp);
1154 			MNT_IUNLOCK(mp);
1155 			if (dtr != NULL)
1156 				dtr(vp, dtr_arg);
1157 			return (EBUSY);
1158 		}
1159 	}
1160 	vp->v_mount = mp;
1161 	MNT_REF(mp);
1162 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1163 	VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1164 		("neg mount point vnode list size"));
1165 	mp->mnt_nvnodelistsize++;
1166 	KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
1167 	    ("Activating already active vnode"));
1168 	vp->v_iflag |= VI_ACTIVE;
1169 	mtx_lock(&vnode_free_list_mtx);
1170 	TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
1171 	mp->mnt_activevnodelistsize++;
1172 	mtx_unlock(&vnode_free_list_mtx);
1173 	VI_UNLOCK(vp);
1174 	MNT_IUNLOCK(mp);
1175 	return (0);
1176 }
1177 
1178 int
1179 insmntque(struct vnode *vp, struct mount *mp)
1180 {
1181 
1182 	return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1183 }
1184 
1185 /*
1186  * Flush out and invalidate all buffers associated with a bufobj
1187  * Called with the underlying object locked.
1188  */
1189 int
1190 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1191 {
1192 	int error;
1193 
1194 	BO_LOCK(bo);
1195 	if (flags & V_SAVE) {
1196 		error = bufobj_wwait(bo, slpflag, slptimeo);
1197 		if (error) {
1198 			BO_UNLOCK(bo);
1199 			return (error);
1200 		}
1201 		if (bo->bo_dirty.bv_cnt > 0) {
1202 			BO_UNLOCK(bo);
1203 			if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1204 				return (error);
1205 			/*
1206 			 * XXX We could save a lock/unlock if this was only
1207 			 * enabled under INVARIANTS
1208 			 */
1209 			BO_LOCK(bo);
1210 			if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1211 				panic("vinvalbuf: dirty bufs");
1212 		}
1213 	}
1214 	/*
1215 	 * If you alter this loop please notice that interlock is dropped and
1216 	 * reacquired in flushbuflist.  Special care is needed to ensure that
1217 	 * no race conditions occur from this.
1218 	 */
1219 	do {
1220 		error = flushbuflist(&bo->bo_clean,
1221 		    flags, bo, slpflag, slptimeo);
1222 		if (error == 0 && !(flags & V_CLEANONLY))
1223 			error = flushbuflist(&bo->bo_dirty,
1224 			    flags, bo, slpflag, slptimeo);
1225 		if (error != 0 && error != EAGAIN) {
1226 			BO_UNLOCK(bo);
1227 			return (error);
1228 		}
1229 	} while (error != 0);
1230 
1231 	/*
1232 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1233 	 * have write I/O in-progress but if there is a VM object then the
1234 	 * VM object can also have read-I/O in-progress.
1235 	 */
1236 	do {
1237 		bufobj_wwait(bo, 0, 0);
1238 		BO_UNLOCK(bo);
1239 		if (bo->bo_object != NULL) {
1240 			VM_OBJECT_LOCK(bo->bo_object);
1241 			vm_object_pip_wait(bo->bo_object, "bovlbx");
1242 			VM_OBJECT_UNLOCK(bo->bo_object);
1243 		}
1244 		BO_LOCK(bo);
1245 	} while (bo->bo_numoutput > 0);
1246 	BO_UNLOCK(bo);
1247 
1248 	/*
1249 	 * Destroy the copy in the VM cache, too.
1250 	 */
1251 	if (bo->bo_object != NULL &&
1252 	    (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) {
1253 		VM_OBJECT_LOCK(bo->bo_object);
1254 		vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
1255 		    OBJPR_CLEANONLY : 0);
1256 		VM_OBJECT_UNLOCK(bo->bo_object);
1257 	}
1258 
1259 #ifdef INVARIANTS
1260 	BO_LOCK(bo);
1261 	if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 &&
1262 	    (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1263 		panic("vinvalbuf: flush failed");
1264 	BO_UNLOCK(bo);
1265 #endif
1266 	return (0);
1267 }
1268 
1269 /*
1270  * Flush out and invalidate all buffers associated with a vnode.
1271  * Called with the underlying object locked.
1272  */
1273 int
1274 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1275 {
1276 
1277 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1278 	ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1279 	return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1280 }
1281 
1282 /*
1283  * Flush out buffers on the specified list.
1284  *
1285  */
1286 static int
1287 flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1288     int slptimeo)
1289 {
1290 	struct buf *bp, *nbp;
1291 	int retval, error;
1292 	daddr_t lblkno;
1293 	b_xflags_t xflags;
1294 
1295 	ASSERT_BO_LOCKED(bo);
1296 
1297 	retval = 0;
1298 	TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1299 		if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1300 		    ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1301 			continue;
1302 		}
1303 		lblkno = 0;
1304 		xflags = 0;
1305 		if (nbp != NULL) {
1306 			lblkno = nbp->b_lblkno;
1307 			xflags = nbp->b_xflags &
1308 				(BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1309 		}
1310 		retval = EAGAIN;
1311 		error = BUF_TIMELOCK(bp,
1312 		    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1313 		    "flushbuf", slpflag, slptimeo);
1314 		if (error) {
1315 			BO_LOCK(bo);
1316 			return (error != ENOLCK ? error : EAGAIN);
1317 		}
1318 		KASSERT(bp->b_bufobj == bo,
1319 		    ("bp %p wrong b_bufobj %p should be %p",
1320 		    bp, bp->b_bufobj, bo));
1321 		if (bp->b_bufobj != bo) {	/* XXX: necessary ? */
1322 			BUF_UNLOCK(bp);
1323 			BO_LOCK(bo);
1324 			return (EAGAIN);
1325 		}
1326 		/*
1327 		 * XXX Since there are no node locks for NFS, I
1328 		 * believe there is a slight chance that a delayed
1329 		 * write will occur while sleeping just above, so
1330 		 * check for it.
1331 		 */
1332 		if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1333 		    (flags & V_SAVE)) {
1334 			BO_LOCK(bo);
1335 			bremfree(bp);
1336 			BO_UNLOCK(bo);
1337 			bp->b_flags |= B_ASYNC;
1338 			bwrite(bp);
1339 			BO_LOCK(bo);
1340 			return (EAGAIN);	/* XXX: why not loop ? */
1341 		}
1342 		BO_LOCK(bo);
1343 		bremfree(bp);
1344 		BO_UNLOCK(bo);
1345 		bp->b_flags |= (B_INVAL | B_RELBUF);
1346 		bp->b_flags &= ~B_ASYNC;
1347 		brelse(bp);
1348 		BO_LOCK(bo);
1349 		if (nbp != NULL &&
1350 		    (nbp->b_bufobj != bo ||
1351 		     nbp->b_lblkno != lblkno ||
1352 		     (nbp->b_xflags &
1353 		      (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1354 			break;			/* nbp invalid */
1355 	}
1356 	return (retval);
1357 }
1358 
1359 /*
1360  * Truncate a file's buffer and pages to a specified length.  This
1361  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1362  * sync activity.
1363  */
1364 int
1365 vtruncbuf(struct vnode *vp, struct ucred *cred, off_t length, int blksize)
1366 {
1367 	struct buf *bp, *nbp;
1368 	int anyfreed;
1369 	int trunclbn;
1370 	struct bufobj *bo;
1371 
1372 	CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1373 	    vp, cred, blksize, (uintmax_t)length);
1374 
1375 	/*
1376 	 * Round up to the *next* lbn.
1377 	 */
1378 	trunclbn = (length + blksize - 1) / blksize;
1379 
1380 	ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1381 restart:
1382 	bo = &vp->v_bufobj;
1383 	BO_LOCK(bo);
1384 	anyfreed = 1;
1385 	for (;anyfreed;) {
1386 		anyfreed = 0;
1387 		TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1388 			if (bp->b_lblkno < trunclbn)
1389 				continue;
1390 			if (BUF_LOCK(bp,
1391 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1392 			    BO_MTX(bo)) == ENOLCK)
1393 				goto restart;
1394 
1395 			BO_LOCK(bo);
1396 			bremfree(bp);
1397 			BO_UNLOCK(bo);
1398 			bp->b_flags |= (B_INVAL | B_RELBUF);
1399 			bp->b_flags &= ~B_ASYNC;
1400 			brelse(bp);
1401 			anyfreed = 1;
1402 
1403 			BO_LOCK(bo);
1404 			if (nbp != NULL &&
1405 			    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1406 			    (nbp->b_vp != vp) ||
1407 			    (nbp->b_flags & B_DELWRI))) {
1408 				BO_UNLOCK(bo);
1409 				goto restart;
1410 			}
1411 		}
1412 
1413 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1414 			if (bp->b_lblkno < trunclbn)
1415 				continue;
1416 			if (BUF_LOCK(bp,
1417 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1418 			    BO_MTX(bo)) == ENOLCK)
1419 				goto restart;
1420 			BO_LOCK(bo);
1421 			bremfree(bp);
1422 			BO_UNLOCK(bo);
1423 			bp->b_flags |= (B_INVAL | B_RELBUF);
1424 			bp->b_flags &= ~B_ASYNC;
1425 			brelse(bp);
1426 			anyfreed = 1;
1427 
1428 			BO_LOCK(bo);
1429 			if (nbp != NULL &&
1430 			    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1431 			    (nbp->b_vp != vp) ||
1432 			    (nbp->b_flags & B_DELWRI) == 0)) {
1433 				BO_UNLOCK(bo);
1434 				goto restart;
1435 			}
1436 		}
1437 	}
1438 
1439 	if (length > 0) {
1440 restartsync:
1441 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1442 			if (bp->b_lblkno > 0)
1443 				continue;
1444 			/*
1445 			 * Since we hold the vnode lock this should only
1446 			 * fail if we're racing with the buf daemon.
1447 			 */
1448 			if (BUF_LOCK(bp,
1449 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1450 			    BO_MTX(bo)) == ENOLCK) {
1451 				goto restart;
1452 			}
1453 			VNASSERT((bp->b_flags & B_DELWRI), vp,
1454 			    ("buf(%p) on dirty queue without DELWRI", bp));
1455 
1456 			BO_LOCK(bo);
1457 			bremfree(bp);
1458 			BO_UNLOCK(bo);
1459 			bawrite(bp);
1460 			BO_LOCK(bo);
1461 			goto restartsync;
1462 		}
1463 	}
1464 
1465 	bufobj_wwait(bo, 0, 0);
1466 	BO_UNLOCK(bo);
1467 	vnode_pager_setsize(vp, length);
1468 
1469 	return (0);
1470 }
1471 
1472 /*
1473  * buf_splay() - splay tree core for the clean/dirty list of buffers in
1474  *		 a vnode.
1475  *
1476  *	NOTE: We have to deal with the special case of a background bitmap
1477  *	buffer, a situation where two buffers will have the same logical
1478  *	block offset.  We want (1) only the foreground buffer to be accessed
1479  *	in a lookup and (2) must differentiate between the foreground and
1480  *	background buffer in the splay tree algorithm because the splay
1481  *	tree cannot normally handle multiple entities with the same 'index'.
1482  *	We accomplish this by adding differentiating flags to the splay tree's
1483  *	numerical domain.
1484  */
1485 static
1486 struct buf *
1487 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1488 {
1489 	struct buf dummy;
1490 	struct buf *lefttreemax, *righttreemin, *y;
1491 
1492 	if (root == NULL)
1493 		return (NULL);
1494 	lefttreemax = righttreemin = &dummy;
1495 	for (;;) {
1496 		if (lblkno < root->b_lblkno ||
1497 		    (lblkno == root->b_lblkno &&
1498 		    (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1499 			if ((y = root->b_left) == NULL)
1500 				break;
1501 			if (lblkno < y->b_lblkno) {
1502 				/* Rotate right. */
1503 				root->b_left = y->b_right;
1504 				y->b_right = root;
1505 				root = y;
1506 				if ((y = root->b_left) == NULL)
1507 					break;
1508 			}
1509 			/* Link into the new root's right tree. */
1510 			righttreemin->b_left = root;
1511 			righttreemin = root;
1512 		} else if (lblkno > root->b_lblkno ||
1513 		    (lblkno == root->b_lblkno &&
1514 		    (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1515 			if ((y = root->b_right) == NULL)
1516 				break;
1517 			if (lblkno > y->b_lblkno) {
1518 				/* Rotate left. */
1519 				root->b_right = y->b_left;
1520 				y->b_left = root;
1521 				root = y;
1522 				if ((y = root->b_right) == NULL)
1523 					break;
1524 			}
1525 			/* Link into the new root's left tree. */
1526 			lefttreemax->b_right = root;
1527 			lefttreemax = root;
1528 		} else {
1529 			break;
1530 		}
1531 		root = y;
1532 	}
1533 	/* Assemble the new root. */
1534 	lefttreemax->b_right = root->b_left;
1535 	righttreemin->b_left = root->b_right;
1536 	root->b_left = dummy.b_right;
1537 	root->b_right = dummy.b_left;
1538 	return (root);
1539 }
1540 
1541 static void
1542 buf_vlist_remove(struct buf *bp)
1543 {
1544 	struct buf *root;
1545 	struct bufv *bv;
1546 
1547 	KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1548 	ASSERT_BO_LOCKED(bp->b_bufobj);
1549 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1550 	    (BX_VNDIRTY|BX_VNCLEAN),
1551 	    ("buf_vlist_remove: Buf %p is on two lists", bp));
1552 	if (bp->b_xflags & BX_VNDIRTY)
1553 		bv = &bp->b_bufobj->bo_dirty;
1554 	else
1555 		bv = &bp->b_bufobj->bo_clean;
1556 	if (bp != bv->bv_root) {
1557 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1558 		KASSERT(root == bp, ("splay lookup failed in remove"));
1559 	}
1560 	if (bp->b_left == NULL) {
1561 		root = bp->b_right;
1562 	} else {
1563 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1564 		root->b_right = bp->b_right;
1565 	}
1566 	bv->bv_root = root;
1567 	TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1568 	bv->bv_cnt--;
1569 	bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1570 }
1571 
1572 /*
1573  * Add the buffer to the sorted clean or dirty block list using a
1574  * splay tree algorithm.
1575  *
1576  * NOTE: xflags is passed as a constant, optimizing this inline function!
1577  */
1578 static void
1579 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1580 {
1581 	struct buf *root;
1582 	struct bufv *bv;
1583 
1584 	ASSERT_BO_LOCKED(bo);
1585 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1586 	    ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1587 	bp->b_xflags |= xflags;
1588 	if (xflags & BX_VNDIRTY)
1589 		bv = &bo->bo_dirty;
1590 	else
1591 		bv = &bo->bo_clean;
1592 
1593 	root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1594 	if (root == NULL) {
1595 		bp->b_left = NULL;
1596 		bp->b_right = NULL;
1597 		TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1598 	} else if (bp->b_lblkno < root->b_lblkno ||
1599 	    (bp->b_lblkno == root->b_lblkno &&
1600 	    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1601 		bp->b_left = root->b_left;
1602 		bp->b_right = root;
1603 		root->b_left = NULL;
1604 		TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1605 	} else {
1606 		bp->b_right = root->b_right;
1607 		bp->b_left = root;
1608 		root->b_right = NULL;
1609 		TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1610 	}
1611 	bv->bv_cnt++;
1612 	bv->bv_root = bp;
1613 }
1614 
1615 /*
1616  * Lookup a buffer using the splay tree.  Note that we specifically avoid
1617  * shadow buffers used in background bitmap writes.
1618  *
1619  * This code isn't quite efficient as it could be because we are maintaining
1620  * two sorted lists and do not know which list the block resides in.
1621  *
1622  * During a "make buildworld" the desired buffer is found at one of
1623  * the roots more than 60% of the time.  Thus, checking both roots
1624  * before performing either splay eliminates unnecessary splays on the
1625  * first tree splayed.
1626  */
1627 struct buf *
1628 gbincore(struct bufobj *bo, daddr_t lblkno)
1629 {
1630 	struct buf *bp;
1631 
1632 	ASSERT_BO_LOCKED(bo);
1633 	if ((bp = bo->bo_clean.bv_root) != NULL &&
1634 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1635 		return (bp);
1636 	if ((bp = bo->bo_dirty.bv_root) != NULL &&
1637 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1638 		return (bp);
1639 	if ((bp = bo->bo_clean.bv_root) != NULL) {
1640 		bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1641 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1642 			return (bp);
1643 	}
1644 	if ((bp = bo->bo_dirty.bv_root) != NULL) {
1645 		bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1646 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1647 			return (bp);
1648 	}
1649 	return (NULL);
1650 }
1651 
1652 /*
1653  * Associate a buffer with a vnode.
1654  */
1655 void
1656 bgetvp(struct vnode *vp, struct buf *bp)
1657 {
1658 	struct bufobj *bo;
1659 
1660 	bo = &vp->v_bufobj;
1661 	ASSERT_BO_LOCKED(bo);
1662 	VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1663 
1664 	CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1665 	VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1666 	    ("bgetvp: bp already attached! %p", bp));
1667 
1668 	vhold(vp);
1669 	bp->b_vp = vp;
1670 	bp->b_bufobj = bo;
1671 	/*
1672 	 * Insert onto list for new vnode.
1673 	 */
1674 	buf_vlist_add(bp, bo, BX_VNCLEAN);
1675 }
1676 
1677 /*
1678  * Disassociate a buffer from a vnode.
1679  */
1680 void
1681 brelvp(struct buf *bp)
1682 {
1683 	struct bufobj *bo;
1684 	struct vnode *vp;
1685 
1686 	CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1687 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1688 
1689 	/*
1690 	 * Delete from old vnode list, if on one.
1691 	 */
1692 	vp = bp->b_vp;		/* XXX */
1693 	bo = bp->b_bufobj;
1694 	BO_LOCK(bo);
1695 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1696 		buf_vlist_remove(bp);
1697 	else
1698 		panic("brelvp: Buffer %p not on queue.", bp);
1699 	if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1700 		bo->bo_flag &= ~BO_ONWORKLST;
1701 		mtx_lock(&sync_mtx);
1702 		LIST_REMOVE(bo, bo_synclist);
1703 		syncer_worklist_len--;
1704 		mtx_unlock(&sync_mtx);
1705 	}
1706 	bp->b_vp = NULL;
1707 	bp->b_bufobj = NULL;
1708 	BO_UNLOCK(bo);
1709 	vdrop(vp);
1710 }
1711 
1712 /*
1713  * Add an item to the syncer work queue.
1714  */
1715 static void
1716 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1717 {
1718 	int slot;
1719 
1720 	ASSERT_BO_LOCKED(bo);
1721 
1722 	mtx_lock(&sync_mtx);
1723 	if (bo->bo_flag & BO_ONWORKLST)
1724 		LIST_REMOVE(bo, bo_synclist);
1725 	else {
1726 		bo->bo_flag |= BO_ONWORKLST;
1727 		syncer_worklist_len++;
1728 	}
1729 
1730 	if (delay > syncer_maxdelay - 2)
1731 		delay = syncer_maxdelay - 2;
1732 	slot = (syncer_delayno + delay) & syncer_mask;
1733 
1734 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist);
1735 	mtx_unlock(&sync_mtx);
1736 }
1737 
1738 static int
1739 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1740 {
1741 	int error, len;
1742 
1743 	mtx_lock(&sync_mtx);
1744 	len = syncer_worklist_len - sync_vnode_count;
1745 	mtx_unlock(&sync_mtx);
1746 	error = SYSCTL_OUT(req, &len, sizeof(len));
1747 	return (error);
1748 }
1749 
1750 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1751     sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1752 
1753 static struct proc *updateproc;
1754 static void sched_sync(void);
1755 static struct kproc_desc up_kp = {
1756 	"syncer",
1757 	sched_sync,
1758 	&updateproc
1759 };
1760 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1761 
1762 static int
1763 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1764 {
1765 	struct vnode *vp;
1766 	struct mount *mp;
1767 
1768 	*bo = LIST_FIRST(slp);
1769 	if (*bo == NULL)
1770 		return (0);
1771 	vp = (*bo)->__bo_vnode;	/* XXX */
1772 	if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1773 		return (1);
1774 	/*
1775 	 * We use vhold in case the vnode does not
1776 	 * successfully sync.  vhold prevents the vnode from
1777 	 * going away when we unlock the sync_mtx so that
1778 	 * we can acquire the vnode interlock.
1779 	 */
1780 	vholdl(vp);
1781 	mtx_unlock(&sync_mtx);
1782 	VI_UNLOCK(vp);
1783 	if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1784 		vdrop(vp);
1785 		mtx_lock(&sync_mtx);
1786 		return (*bo == LIST_FIRST(slp));
1787 	}
1788 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1789 	(void) VOP_FSYNC(vp, MNT_LAZY, td);
1790 	VOP_UNLOCK(vp, 0);
1791 	vn_finished_write(mp);
1792 	BO_LOCK(*bo);
1793 	if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1794 		/*
1795 		 * Put us back on the worklist.  The worklist
1796 		 * routine will remove us from our current
1797 		 * position and then add us back in at a later
1798 		 * position.
1799 		 */
1800 		vn_syncer_add_to_worklist(*bo, syncdelay);
1801 	}
1802 	BO_UNLOCK(*bo);
1803 	vdrop(vp);
1804 	mtx_lock(&sync_mtx);
1805 	return (0);
1806 }
1807 
1808 /*
1809  * System filesystem synchronizer daemon.
1810  */
1811 static void
1812 sched_sync(void)
1813 {
1814 	struct synclist *next, *slp;
1815 	struct bufobj *bo;
1816 	long starttime;
1817 	struct thread *td = curthread;
1818 	int last_work_seen;
1819 	int net_worklist_len;
1820 	int syncer_final_iter;
1821 	int first_printf;
1822 	int error;
1823 
1824 	last_work_seen = 0;
1825 	syncer_final_iter = 0;
1826 	first_printf = 1;
1827 	syncer_state = SYNCER_RUNNING;
1828 	starttime = time_uptime;
1829 	td->td_pflags |= TDP_NORUNNINGBUF;
1830 
1831 	EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1832 	    SHUTDOWN_PRI_LAST);
1833 
1834 	mtx_lock(&sync_mtx);
1835 	for (;;) {
1836 		if (syncer_state == SYNCER_FINAL_DELAY &&
1837 		    syncer_final_iter == 0) {
1838 			mtx_unlock(&sync_mtx);
1839 			kproc_suspend_check(td->td_proc);
1840 			mtx_lock(&sync_mtx);
1841 		}
1842 		net_worklist_len = syncer_worklist_len - sync_vnode_count;
1843 		if (syncer_state != SYNCER_RUNNING &&
1844 		    starttime != time_uptime) {
1845 			if (first_printf) {
1846 				printf("\nSyncing disks, vnodes remaining...");
1847 				first_printf = 0;
1848 			}
1849 			printf("%d ", net_worklist_len);
1850 		}
1851 		starttime = time_uptime;
1852 
1853 		/*
1854 		 * Push files whose dirty time has expired.  Be careful
1855 		 * of interrupt race on slp queue.
1856 		 *
1857 		 * Skip over empty worklist slots when shutting down.
1858 		 */
1859 		do {
1860 			slp = &syncer_workitem_pending[syncer_delayno];
1861 			syncer_delayno += 1;
1862 			if (syncer_delayno == syncer_maxdelay)
1863 				syncer_delayno = 0;
1864 			next = &syncer_workitem_pending[syncer_delayno];
1865 			/*
1866 			 * If the worklist has wrapped since the
1867 			 * it was emptied of all but syncer vnodes,
1868 			 * switch to the FINAL_DELAY state and run
1869 			 * for one more second.
1870 			 */
1871 			if (syncer_state == SYNCER_SHUTTING_DOWN &&
1872 			    net_worklist_len == 0 &&
1873 			    last_work_seen == syncer_delayno) {
1874 				syncer_state = SYNCER_FINAL_DELAY;
1875 				syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1876 			}
1877 		} while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1878 		    syncer_worklist_len > 0);
1879 
1880 		/*
1881 		 * Keep track of the last time there was anything
1882 		 * on the worklist other than syncer vnodes.
1883 		 * Return to the SHUTTING_DOWN state if any
1884 		 * new work appears.
1885 		 */
1886 		if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1887 			last_work_seen = syncer_delayno;
1888 		if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1889 			syncer_state = SYNCER_SHUTTING_DOWN;
1890 		while (!LIST_EMPTY(slp)) {
1891 			error = sync_vnode(slp, &bo, td);
1892 			if (error == 1) {
1893 				LIST_REMOVE(bo, bo_synclist);
1894 				LIST_INSERT_HEAD(next, bo, bo_synclist);
1895 				continue;
1896 			}
1897 
1898 			if (first_printf == 0)
1899 				wdog_kern_pat(WD_LASTVAL);
1900 
1901 		}
1902 		if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1903 			syncer_final_iter--;
1904 		/*
1905 		 * The variable rushjob allows the kernel to speed up the
1906 		 * processing of the filesystem syncer process. A rushjob
1907 		 * value of N tells the filesystem syncer to process the next
1908 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1909 		 * is used by the soft update code to speed up the filesystem
1910 		 * syncer process when the incore state is getting so far
1911 		 * ahead of the disk that the kernel memory pool is being
1912 		 * threatened with exhaustion.
1913 		 */
1914 		if (rushjob > 0) {
1915 			rushjob -= 1;
1916 			continue;
1917 		}
1918 		/*
1919 		 * Just sleep for a short period of time between
1920 		 * iterations when shutting down to allow some I/O
1921 		 * to happen.
1922 		 *
1923 		 * If it has taken us less than a second to process the
1924 		 * current work, then wait. Otherwise start right over
1925 		 * again. We can still lose time if any single round
1926 		 * takes more than two seconds, but it does not really
1927 		 * matter as we are just trying to generally pace the
1928 		 * filesystem activity.
1929 		 */
1930 		if (syncer_state != SYNCER_RUNNING ||
1931 		    time_uptime == starttime) {
1932 			thread_lock(td);
1933 			sched_prio(td, PPAUSE);
1934 			thread_unlock(td);
1935 		}
1936 		if (syncer_state != SYNCER_RUNNING)
1937 			cv_timedwait(&sync_wakeup, &sync_mtx,
1938 			    hz / SYNCER_SHUTDOWN_SPEEDUP);
1939 		else if (time_uptime == starttime)
1940 			cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1941 	}
1942 }
1943 
1944 /*
1945  * Request the syncer daemon to speed up its work.
1946  * We never push it to speed up more than half of its
1947  * normal turn time, otherwise it could take over the cpu.
1948  */
1949 int
1950 speedup_syncer(void)
1951 {
1952 	int ret = 0;
1953 
1954 	mtx_lock(&sync_mtx);
1955 	if (rushjob < syncdelay / 2) {
1956 		rushjob += 1;
1957 		stat_rush_requests += 1;
1958 		ret = 1;
1959 	}
1960 	mtx_unlock(&sync_mtx);
1961 	cv_broadcast(&sync_wakeup);
1962 	return (ret);
1963 }
1964 
1965 /*
1966  * Tell the syncer to speed up its work and run though its work
1967  * list several times, then tell it to shut down.
1968  */
1969 static void
1970 syncer_shutdown(void *arg, int howto)
1971 {
1972 
1973 	if (howto & RB_NOSYNC)
1974 		return;
1975 	mtx_lock(&sync_mtx);
1976 	syncer_state = SYNCER_SHUTTING_DOWN;
1977 	rushjob = 0;
1978 	mtx_unlock(&sync_mtx);
1979 	cv_broadcast(&sync_wakeup);
1980 	kproc_shutdown(arg, howto);
1981 }
1982 
1983 /*
1984  * Reassign a buffer from one vnode to another.
1985  * Used to assign file specific control information
1986  * (indirect blocks) to the vnode to which they belong.
1987  */
1988 void
1989 reassignbuf(struct buf *bp)
1990 {
1991 	struct vnode *vp;
1992 	struct bufobj *bo;
1993 	int delay;
1994 #ifdef INVARIANTS
1995 	struct bufv *bv;
1996 #endif
1997 
1998 	vp = bp->b_vp;
1999 	bo = bp->b_bufobj;
2000 	++reassignbufcalls;
2001 
2002 	CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
2003 	    bp, bp->b_vp, bp->b_flags);
2004 	/*
2005 	 * B_PAGING flagged buffers cannot be reassigned because their vp
2006 	 * is not fully linked in.
2007 	 */
2008 	if (bp->b_flags & B_PAGING)
2009 		panic("cannot reassign paging buffer");
2010 
2011 	/*
2012 	 * Delete from old vnode list, if on one.
2013 	 */
2014 	BO_LOCK(bo);
2015 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
2016 		buf_vlist_remove(bp);
2017 	else
2018 		panic("reassignbuf: Buffer %p not on queue.", bp);
2019 	/*
2020 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
2021 	 * of clean buffers.
2022 	 */
2023 	if (bp->b_flags & B_DELWRI) {
2024 		if ((bo->bo_flag & BO_ONWORKLST) == 0) {
2025 			switch (vp->v_type) {
2026 			case VDIR:
2027 				delay = dirdelay;
2028 				break;
2029 			case VCHR:
2030 				delay = metadelay;
2031 				break;
2032 			default:
2033 				delay = filedelay;
2034 			}
2035 			vn_syncer_add_to_worklist(bo, delay);
2036 		}
2037 		buf_vlist_add(bp, bo, BX_VNDIRTY);
2038 	} else {
2039 		buf_vlist_add(bp, bo, BX_VNCLEAN);
2040 
2041 		if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
2042 			mtx_lock(&sync_mtx);
2043 			LIST_REMOVE(bo, bo_synclist);
2044 			syncer_worklist_len--;
2045 			mtx_unlock(&sync_mtx);
2046 			bo->bo_flag &= ~BO_ONWORKLST;
2047 		}
2048 	}
2049 #ifdef INVARIANTS
2050 	bv = &bo->bo_clean;
2051 	bp = TAILQ_FIRST(&bv->bv_hd);
2052 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2053 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2054 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
2055 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2056 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2057 	bv = &bo->bo_dirty;
2058 	bp = TAILQ_FIRST(&bv->bv_hd);
2059 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2060 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2061 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
2062 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2063 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2064 #endif
2065 	BO_UNLOCK(bo);
2066 }
2067 
2068 /*
2069  * Increment the use and hold counts on the vnode, taking care to reference
2070  * the driver's usecount if this is a chardev.  The vholdl() will remove
2071  * the vnode from the free list if it is presently free.  Requires the
2072  * vnode interlock and returns with it held.
2073  */
2074 static void
2075 v_incr_usecount(struct vnode *vp)
2076 {
2077 
2078 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2079 	vp->v_usecount++;
2080 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2081 		dev_lock();
2082 		vp->v_rdev->si_usecount++;
2083 		dev_unlock();
2084 	}
2085 	vholdl(vp);
2086 }
2087 
2088 /*
2089  * Turn a holdcnt into a use+holdcnt such that only one call to
2090  * v_decr_usecount is needed.
2091  */
2092 static void
2093 v_upgrade_usecount(struct vnode *vp)
2094 {
2095 
2096 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2097 	vp->v_usecount++;
2098 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2099 		dev_lock();
2100 		vp->v_rdev->si_usecount++;
2101 		dev_unlock();
2102 	}
2103 }
2104 
2105 /*
2106  * Decrement the vnode use and hold count along with the driver's usecount
2107  * if this is a chardev.  The vdropl() below releases the vnode interlock
2108  * as it may free the vnode.
2109  */
2110 static void
2111 v_decr_usecount(struct vnode *vp)
2112 {
2113 
2114 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2115 	VNASSERT(vp->v_usecount > 0, vp,
2116 	    ("v_decr_usecount: negative usecount"));
2117 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2118 	vp->v_usecount--;
2119 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2120 		dev_lock();
2121 		vp->v_rdev->si_usecount--;
2122 		dev_unlock();
2123 	}
2124 	vdropl(vp);
2125 }
2126 
2127 /*
2128  * Decrement only the use count and driver use count.  This is intended to
2129  * be paired with a follow on vdropl() to release the remaining hold count.
2130  * In this way we may vgone() a vnode with a 0 usecount without risk of
2131  * having it end up on a free list because the hold count is kept above 0.
2132  */
2133 static void
2134 v_decr_useonly(struct vnode *vp)
2135 {
2136 
2137 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2138 	VNASSERT(vp->v_usecount > 0, vp,
2139 	    ("v_decr_useonly: negative usecount"));
2140 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2141 	vp->v_usecount--;
2142 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2143 		dev_lock();
2144 		vp->v_rdev->si_usecount--;
2145 		dev_unlock();
2146 	}
2147 }
2148 
2149 /*
2150  * Grab a particular vnode from the free list, increment its
2151  * reference count and lock it.  VI_DOOMED is set if the vnode
2152  * is being destroyed.  Only callers who specify LK_RETRY will
2153  * see doomed vnodes.  If inactive processing was delayed in
2154  * vput try to do it here.
2155  */
2156 int
2157 vget(struct vnode *vp, int flags, struct thread *td)
2158 {
2159 	int error;
2160 
2161 	error = 0;
2162 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2163 	    ("vget: invalid lock operation"));
2164 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2165 
2166 	if ((flags & LK_INTERLOCK) == 0)
2167 		VI_LOCK(vp);
2168 	vholdl(vp);
2169 	if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2170 		vdrop(vp);
2171 		CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2172 		    vp);
2173 		return (error);
2174 	}
2175 	if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2176 		panic("vget: vn_lock failed to return ENOENT\n");
2177 	VI_LOCK(vp);
2178 	/* Upgrade our holdcnt to a usecount. */
2179 	v_upgrade_usecount(vp);
2180 	/*
2181 	 * We don't guarantee that any particular close will
2182 	 * trigger inactive processing so just make a best effort
2183 	 * here at preventing a reference to a removed file.  If
2184 	 * we don't succeed no harm is done.
2185 	 */
2186 	if (vp->v_iflag & VI_OWEINACT) {
2187 		if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2188 		    (flags & LK_NOWAIT) == 0)
2189 			vinactive(vp, td);
2190 		vp->v_iflag &= ~VI_OWEINACT;
2191 	}
2192 	VI_UNLOCK(vp);
2193 	return (0);
2194 }
2195 
2196 /*
2197  * Increase the reference count of a vnode.
2198  */
2199 void
2200 vref(struct vnode *vp)
2201 {
2202 
2203 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2204 	VI_LOCK(vp);
2205 	v_incr_usecount(vp);
2206 	VI_UNLOCK(vp);
2207 }
2208 
2209 /*
2210  * Return reference count of a vnode.
2211  *
2212  * The results of this call are only guaranteed when some mechanism other
2213  * than the VI lock is used to stop other processes from gaining references
2214  * to the vnode.  This may be the case if the caller holds the only reference.
2215  * This is also useful when stale data is acceptable as race conditions may
2216  * be accounted for by some other means.
2217  */
2218 int
2219 vrefcnt(struct vnode *vp)
2220 {
2221 	int usecnt;
2222 
2223 	VI_LOCK(vp);
2224 	usecnt = vp->v_usecount;
2225 	VI_UNLOCK(vp);
2226 
2227 	return (usecnt);
2228 }
2229 
2230 #define	VPUTX_VRELE	1
2231 #define	VPUTX_VPUT	2
2232 #define	VPUTX_VUNREF	3
2233 
2234 static void
2235 vputx(struct vnode *vp, int func)
2236 {
2237 	int error;
2238 
2239 	KASSERT(vp != NULL, ("vputx: null vp"));
2240 	if (func == VPUTX_VUNREF)
2241 		ASSERT_VOP_LOCKED(vp, "vunref");
2242 	else if (func == VPUTX_VPUT)
2243 		ASSERT_VOP_LOCKED(vp, "vput");
2244 	else
2245 		KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2246 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2247 	VI_LOCK(vp);
2248 
2249 	/* Skip this v_writecount check if we're going to panic below. */
2250 	VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2251 	    ("vputx: missed vn_close"));
2252 	error = 0;
2253 
2254 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2255 	    vp->v_usecount == 1)) {
2256 		if (func == VPUTX_VPUT)
2257 			VOP_UNLOCK(vp, 0);
2258 		v_decr_usecount(vp);
2259 		return;
2260 	}
2261 
2262 	if (vp->v_usecount != 1) {
2263 		vprint("vputx: negative ref count", vp);
2264 		panic("vputx: negative ref cnt");
2265 	}
2266 	CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2267 	/*
2268 	 * We want to hold the vnode until the inactive finishes to
2269 	 * prevent vgone() races.  We drop the use count here and the
2270 	 * hold count below when we're done.
2271 	 */
2272 	v_decr_useonly(vp);
2273 	/*
2274 	 * We must call VOP_INACTIVE with the node locked. Mark
2275 	 * as VI_DOINGINACT to avoid recursion.
2276 	 */
2277 	vp->v_iflag |= VI_OWEINACT;
2278 	switch (func) {
2279 	case VPUTX_VRELE:
2280 		error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2281 		VI_LOCK(vp);
2282 		break;
2283 	case VPUTX_VPUT:
2284 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2285 			error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
2286 			    LK_NOWAIT);
2287 			VI_LOCK(vp);
2288 		}
2289 		break;
2290 	case VPUTX_VUNREF:
2291 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
2292 			error = EBUSY;
2293 		break;
2294 	}
2295 	if (vp->v_usecount > 0)
2296 		vp->v_iflag &= ~VI_OWEINACT;
2297 	if (error == 0) {
2298 		if (vp->v_iflag & VI_OWEINACT)
2299 			vinactive(vp, curthread);
2300 		if (func != VPUTX_VUNREF)
2301 			VOP_UNLOCK(vp, 0);
2302 	}
2303 	vdropl(vp);
2304 }
2305 
2306 /*
2307  * Vnode put/release.
2308  * If count drops to zero, call inactive routine and return to freelist.
2309  */
2310 void
2311 vrele(struct vnode *vp)
2312 {
2313 
2314 	vputx(vp, VPUTX_VRELE);
2315 }
2316 
2317 /*
2318  * Release an already locked vnode.  This give the same effects as
2319  * unlock+vrele(), but takes less time and avoids releasing and
2320  * re-aquiring the lock (as vrele() acquires the lock internally.)
2321  */
2322 void
2323 vput(struct vnode *vp)
2324 {
2325 
2326 	vputx(vp, VPUTX_VPUT);
2327 }
2328 
2329 /*
2330  * Release an exclusively locked vnode. Do not unlock the vnode lock.
2331  */
2332 void
2333 vunref(struct vnode *vp)
2334 {
2335 
2336 	vputx(vp, VPUTX_VUNREF);
2337 }
2338 
2339 /*
2340  * Somebody doesn't want the vnode recycled.
2341  */
2342 void
2343 vhold(struct vnode *vp)
2344 {
2345 
2346 	VI_LOCK(vp);
2347 	vholdl(vp);
2348 	VI_UNLOCK(vp);
2349 }
2350 
2351 /*
2352  * Increase the hold count and activate if this is the first reference.
2353  */
2354 void
2355 vholdl(struct vnode *vp)
2356 {
2357 	struct mount *mp;
2358 
2359 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2360 	vp->v_holdcnt++;
2361 	if (!VSHOULDBUSY(vp))
2362 		return;
2363 	ASSERT_VI_LOCKED(vp, "vholdl");
2364 	VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
2365 	VNASSERT(vp->v_op != NULL, vp, ("vholdl: vnode already reclaimed."));
2366 	/*
2367 	 * Remove a vnode from the free list, mark it as in use,
2368 	 * and put it on the active list.
2369 	 */
2370 	mtx_lock(&vnode_free_list_mtx);
2371 	TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
2372 	freevnodes--;
2373 	vp->v_iflag &= ~(VI_FREE|VI_AGE);
2374 	KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
2375 	    ("Activating already active vnode"));
2376 	vp->v_iflag |= VI_ACTIVE;
2377 	mp = vp->v_mount;
2378 	TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
2379 	mp->mnt_activevnodelistsize++;
2380 	mtx_unlock(&vnode_free_list_mtx);
2381 }
2382 
2383 /*
2384  * Note that there is one less who cares about this vnode.
2385  * vdrop() is the opposite of vhold().
2386  */
2387 void
2388 vdrop(struct vnode *vp)
2389 {
2390 
2391 	VI_LOCK(vp);
2392 	vdropl(vp);
2393 }
2394 
2395 /*
2396  * Drop the hold count of the vnode.  If this is the last reference to
2397  * the vnode we place it on the free list unless it has been vgone'd
2398  * (marked VI_DOOMED) in which case we will free it.
2399  */
2400 void
2401 vdropl(struct vnode *vp)
2402 {
2403 	struct bufobj *bo;
2404 	struct mount *mp;
2405 	int active;
2406 
2407 	ASSERT_VI_LOCKED(vp, "vdropl");
2408 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2409 	if (vp->v_holdcnt <= 0)
2410 		panic("vdrop: holdcnt %d", vp->v_holdcnt);
2411 	vp->v_holdcnt--;
2412 	if (vp->v_holdcnt > 0) {
2413 		VI_UNLOCK(vp);
2414 		return;
2415 	}
2416 	if ((vp->v_iflag & VI_DOOMED) == 0) {
2417 		/*
2418 		 * Mark a vnode as free: remove it from its active list
2419 		 * and put it up for recycling on the freelist.
2420 		 */
2421 		VNASSERT(vp->v_op != NULL, vp,
2422 		    ("vdropl: vnode already reclaimed."));
2423 		VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2424 		    ("vnode already free"));
2425 		VNASSERT(VSHOULDFREE(vp), vp,
2426 		    ("vdropl: freeing when we shouldn't"));
2427 		active = vp->v_iflag & VI_ACTIVE;
2428 		vp->v_iflag &= ~VI_ACTIVE;
2429 		mp = vp->v_mount;
2430 		mtx_lock(&vnode_free_list_mtx);
2431 		if (active) {
2432 			TAILQ_REMOVE(&mp->mnt_activevnodelist, vp,
2433 			    v_actfreelist);
2434 			mp->mnt_activevnodelistsize--;
2435 		}
2436 		if (vp->v_iflag & VI_AGE) {
2437 			TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_actfreelist);
2438 		} else {
2439 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
2440 		}
2441 		freevnodes++;
2442 		vp->v_iflag &= ~VI_AGE;
2443 		vp->v_iflag |= VI_FREE;
2444 		mtx_unlock(&vnode_free_list_mtx);
2445 		VI_UNLOCK(vp);
2446 		return;
2447 	}
2448 	/*
2449 	 * The vnode has been marked for destruction, so free it.
2450 	 */
2451 	CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp);
2452 	mtx_lock(&vnode_free_list_mtx);
2453 	numvnodes--;
2454 	mtx_unlock(&vnode_free_list_mtx);
2455 	bo = &vp->v_bufobj;
2456 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2457 	    ("cleaned vnode still on the free list."));
2458 	VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
2459 	VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
2460 	VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
2461 	VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
2462 	VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
2463 	VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
2464 	VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
2465 	VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
2466 	VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
2467 	VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
2468 	VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
2469 	VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
2470 	VI_UNLOCK(vp);
2471 #ifdef MAC
2472 	mac_vnode_destroy(vp);
2473 #endif
2474 	if (vp->v_pollinfo != NULL)
2475 		destroy_vpollinfo(vp->v_pollinfo);
2476 #ifdef INVARIANTS
2477 	/* XXX Elsewhere we detect an already freed vnode via NULL v_op. */
2478 	vp->v_op = NULL;
2479 #endif
2480 	rangelock_destroy(&vp->v_rl);
2481 	lockdestroy(vp->v_vnlock);
2482 	mtx_destroy(&vp->v_interlock);
2483 	mtx_destroy(BO_MTX(bo));
2484 	uma_zfree(vnode_zone, vp);
2485 }
2486 
2487 /*
2488  * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2489  * flags.  DOINGINACT prevents us from recursing in calls to vinactive.
2490  * OWEINACT tracks whether a vnode missed a call to inactive due to a
2491  * failed lock upgrade.
2492  */
2493 void
2494 vinactive(struct vnode *vp, struct thread *td)
2495 {
2496 	struct vm_object *obj;
2497 
2498 	ASSERT_VOP_ELOCKED(vp, "vinactive");
2499 	ASSERT_VI_LOCKED(vp, "vinactive");
2500 	VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2501 	    ("vinactive: recursed on VI_DOINGINACT"));
2502 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2503 	vp->v_iflag |= VI_DOINGINACT;
2504 	vp->v_iflag &= ~VI_OWEINACT;
2505 	VI_UNLOCK(vp);
2506 	/*
2507 	 * Before moving off the active list, we must be sure that any
2508 	 * modified pages are on the vnode's dirty list since these will
2509 	 * no longer be checked once the vnode is on the inactive list.
2510 	 */
2511 	obj = vp->v_object;
2512 	if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
2513 		VM_OBJECT_LOCK(obj);
2514 		vm_object_page_clean(obj, 0, 0, OBJPC_NOSYNC);
2515 		VM_OBJECT_UNLOCK(obj);
2516 	}
2517 	VOP_INACTIVE(vp, td);
2518 	VI_LOCK(vp);
2519 	VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2520 	    ("vinactive: lost VI_DOINGINACT"));
2521 	vp->v_iflag &= ~VI_DOINGINACT;
2522 }
2523 
2524 /*
2525  * Remove any vnodes in the vnode table belonging to mount point mp.
2526  *
2527  * If FORCECLOSE is not specified, there should not be any active ones,
2528  * return error if any are found (nb: this is a user error, not a
2529  * system error). If FORCECLOSE is specified, detach any active vnodes
2530  * that are found.
2531  *
2532  * If WRITECLOSE is set, only flush out regular file vnodes open for
2533  * writing.
2534  *
2535  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2536  *
2537  * `rootrefs' specifies the base reference count for the root vnode
2538  * of this filesystem. The root vnode is considered busy if its
2539  * v_usecount exceeds this value. On a successful return, vflush(, td)
2540  * will call vrele() on the root vnode exactly rootrefs times.
2541  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2542  * be zero.
2543  */
2544 #ifdef DIAGNOSTIC
2545 static int busyprt = 0;		/* print out busy vnodes */
2546 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
2547 #endif
2548 
2549 int
2550 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
2551 {
2552 	struct vnode *vp, *mvp, *rootvp = NULL;
2553 	struct vattr vattr;
2554 	int busy = 0, error;
2555 
2556 	CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2557 	    rootrefs, flags);
2558 	if (rootrefs > 0) {
2559 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2560 		    ("vflush: bad args"));
2561 		/*
2562 		 * Get the filesystem root vnode. We can vput() it
2563 		 * immediately, since with rootrefs > 0, it won't go away.
2564 		 */
2565 		if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2566 			CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2567 			    __func__, error);
2568 			return (error);
2569 		}
2570 		vput(rootvp);
2571 	}
2572 loop:
2573 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
2574 		vholdl(vp);
2575 		error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2576 		if (error) {
2577 			vdrop(vp);
2578 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2579 			goto loop;
2580 		}
2581 		/*
2582 		 * Skip over a vnodes marked VV_SYSTEM.
2583 		 */
2584 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2585 			VOP_UNLOCK(vp, 0);
2586 			vdrop(vp);
2587 			continue;
2588 		}
2589 		/*
2590 		 * If WRITECLOSE is set, flush out unlinked but still open
2591 		 * files (even if open only for reading) and regular file
2592 		 * vnodes open for writing.
2593 		 */
2594 		if (flags & WRITECLOSE) {
2595 			if (vp->v_object != NULL) {
2596 				VM_OBJECT_LOCK(vp->v_object);
2597 				vm_object_page_clean(vp->v_object, 0, 0, 0);
2598 				VM_OBJECT_UNLOCK(vp->v_object);
2599 			}
2600 			error = VOP_FSYNC(vp, MNT_WAIT, td);
2601 			if (error != 0) {
2602 				VOP_UNLOCK(vp, 0);
2603 				vdrop(vp);
2604 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2605 				return (error);
2606 			}
2607 			error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2608 			VI_LOCK(vp);
2609 
2610 			if ((vp->v_type == VNON ||
2611 			    (error == 0 && vattr.va_nlink > 0)) &&
2612 			    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2613 				VOP_UNLOCK(vp, 0);
2614 				vdropl(vp);
2615 				continue;
2616 			}
2617 		} else
2618 			VI_LOCK(vp);
2619 		/*
2620 		 * With v_usecount == 0, all we need to do is clear out the
2621 		 * vnode data structures and we are done.
2622 		 *
2623 		 * If FORCECLOSE is set, forcibly close the vnode.
2624 		 */
2625 		if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2626 			VNASSERT(vp->v_usecount == 0 ||
2627 			    (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2628 			    ("device VNODE %p is FORCECLOSED", vp));
2629 			vgonel(vp);
2630 		} else {
2631 			busy++;
2632 #ifdef DIAGNOSTIC
2633 			if (busyprt)
2634 				vprint("vflush: busy vnode", vp);
2635 #endif
2636 		}
2637 		VOP_UNLOCK(vp, 0);
2638 		vdropl(vp);
2639 	}
2640 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2641 		/*
2642 		 * If just the root vnode is busy, and if its refcount
2643 		 * is equal to `rootrefs', then go ahead and kill it.
2644 		 */
2645 		VI_LOCK(rootvp);
2646 		KASSERT(busy > 0, ("vflush: not busy"));
2647 		VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2648 		    ("vflush: usecount %d < rootrefs %d",
2649 		     rootvp->v_usecount, rootrefs));
2650 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2651 			VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2652 			vgone(rootvp);
2653 			VOP_UNLOCK(rootvp, 0);
2654 			busy = 0;
2655 		} else
2656 			VI_UNLOCK(rootvp);
2657 	}
2658 	if (busy) {
2659 		CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2660 		    busy);
2661 		return (EBUSY);
2662 	}
2663 	for (; rootrefs > 0; rootrefs--)
2664 		vrele(rootvp);
2665 	return (0);
2666 }
2667 
2668 /*
2669  * Recycle an unused vnode to the front of the free list.
2670  */
2671 int
2672 vrecycle(struct vnode *vp)
2673 {
2674 	int recycled;
2675 
2676 	ASSERT_VOP_ELOCKED(vp, "vrecycle");
2677 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2678 	recycled = 0;
2679 	VI_LOCK(vp);
2680 	if (vp->v_usecount == 0) {
2681 		recycled = 1;
2682 		vgonel(vp);
2683 	}
2684 	VI_UNLOCK(vp);
2685 	return (recycled);
2686 }
2687 
2688 /*
2689  * Eliminate all activity associated with a vnode
2690  * in preparation for reuse.
2691  */
2692 void
2693 vgone(struct vnode *vp)
2694 {
2695 	VI_LOCK(vp);
2696 	vgonel(vp);
2697 	VI_UNLOCK(vp);
2698 }
2699 
2700 static void
2701 vgonel_reclaim_lowervp_vfs(struct mount *mp __unused,
2702     struct vnode *lowervp __unused)
2703 {
2704 }
2705 
2706 /*
2707  * Notify upper mounts about reclaimed vnode.
2708  */
2709 static void
2710 vgonel_reclaim_lowervp(struct vnode *vp)
2711 {
2712 	static struct vfsops vgonel_vfsops = {
2713 		.vfs_reclaim_lowervp = vgonel_reclaim_lowervp_vfs
2714 	};
2715 	struct mount *mp, *ump, *mmp;
2716 
2717 	mp = vp->v_mount;
2718 	if (mp == NULL)
2719 		return;
2720 
2721 	MNT_ILOCK(mp);
2722 	if (TAILQ_EMPTY(&mp->mnt_uppers))
2723 		goto unlock;
2724 	MNT_IUNLOCK(mp);
2725 	mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO);
2726 	mmp->mnt_op = &vgonel_vfsops;
2727 	mmp->mnt_kern_flag |= MNTK_MARKER;
2728 	MNT_ILOCK(mp);
2729 	mp->mnt_kern_flag |= MNTK_VGONE_UPPER;
2730 	for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) {
2731 		if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) {
2732 			ump = TAILQ_NEXT(ump, mnt_upper_link);
2733 			continue;
2734 		}
2735 		TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link);
2736 		MNT_IUNLOCK(mp);
2737 		VFS_RECLAIM_LOWERVP(ump, vp);
2738 		MNT_ILOCK(mp);
2739 		ump = TAILQ_NEXT(mmp, mnt_upper_link);
2740 		TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link);
2741 	}
2742 	free(mmp, M_TEMP);
2743 	mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER;
2744 	if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) {
2745 		mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER;
2746 		wakeup(&mp->mnt_uppers);
2747 	}
2748 unlock:
2749 	MNT_IUNLOCK(mp);
2750 }
2751 
2752 /*
2753  * vgone, with the vp interlock held.
2754  */
2755 void
2756 vgonel(struct vnode *vp)
2757 {
2758 	struct thread *td;
2759 	int oweinact;
2760 	int active;
2761 	struct mount *mp;
2762 
2763 	ASSERT_VOP_ELOCKED(vp, "vgonel");
2764 	ASSERT_VI_LOCKED(vp, "vgonel");
2765 	VNASSERT(vp->v_holdcnt, vp,
2766 	    ("vgonel: vp %p has no reference.", vp));
2767 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2768 	td = curthread;
2769 
2770 	/*
2771 	 * Don't vgonel if we're already doomed.
2772 	 */
2773 	if (vp->v_iflag & VI_DOOMED)
2774 		return;
2775 	vp->v_iflag |= VI_DOOMED;
2776 
2777 	/*
2778 	 * Check to see if the vnode is in use.  If so, we have to call
2779 	 * VOP_CLOSE() and VOP_INACTIVE().
2780 	 */
2781 	active = vp->v_usecount;
2782 	oweinact = (vp->v_iflag & VI_OWEINACT);
2783 	VI_UNLOCK(vp);
2784 	vgonel_reclaim_lowervp(vp);
2785 
2786 	/*
2787 	 * Clean out any buffers associated with the vnode.
2788 	 * If the flush fails, just toss the buffers.
2789 	 */
2790 	mp = NULL;
2791 	if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2792 		(void) vn_start_secondary_write(vp, &mp, V_WAIT);
2793 	if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
2794 		vinvalbuf(vp, 0, 0, 0);
2795 
2796 	/*
2797 	 * If purging an active vnode, it must be closed and
2798 	 * deactivated before being reclaimed.
2799 	 */
2800 	if (active)
2801 		VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2802 	if (oweinact || active) {
2803 		VI_LOCK(vp);
2804 		if ((vp->v_iflag & VI_DOINGINACT) == 0)
2805 			vinactive(vp, td);
2806 		VI_UNLOCK(vp);
2807 	}
2808 	if (vp->v_type == VSOCK)
2809 		vfs_unp_reclaim(vp);
2810 	/*
2811 	 * Reclaim the vnode.
2812 	 */
2813 	if (VOP_RECLAIM(vp, td))
2814 		panic("vgone: cannot reclaim");
2815 	if (mp != NULL)
2816 		vn_finished_secondary_write(mp);
2817 	VNASSERT(vp->v_object == NULL, vp,
2818 	    ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2819 	/*
2820 	 * Clear the advisory locks and wake up waiting threads.
2821 	 */
2822 	(void)VOP_ADVLOCKPURGE(vp);
2823 	/*
2824 	 * Delete from old mount point vnode list.
2825 	 */
2826 	delmntque(vp);
2827 	cache_purge(vp);
2828 	/*
2829 	 * Done with purge, reset to the standard lock and invalidate
2830 	 * the vnode.
2831 	 */
2832 	VI_LOCK(vp);
2833 	vp->v_vnlock = &vp->v_lock;
2834 	vp->v_op = &dead_vnodeops;
2835 	vp->v_tag = "none";
2836 	vp->v_type = VBAD;
2837 }
2838 
2839 /*
2840  * Calculate the total number of references to a special device.
2841  */
2842 int
2843 vcount(struct vnode *vp)
2844 {
2845 	int count;
2846 
2847 	dev_lock();
2848 	count = vp->v_rdev->si_usecount;
2849 	dev_unlock();
2850 	return (count);
2851 }
2852 
2853 /*
2854  * Same as above, but using the struct cdev *as argument
2855  */
2856 int
2857 count_dev(struct cdev *dev)
2858 {
2859 	int count;
2860 
2861 	dev_lock();
2862 	count = dev->si_usecount;
2863 	dev_unlock();
2864 	return(count);
2865 }
2866 
2867 /*
2868  * Print out a description of a vnode.
2869  */
2870 static char *typename[] =
2871 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2872  "VMARKER"};
2873 
2874 void
2875 vn_printf(struct vnode *vp, const char *fmt, ...)
2876 {
2877 	va_list ap;
2878 	char buf[256], buf2[16];
2879 	u_long flags;
2880 
2881 	va_start(ap, fmt);
2882 	vprintf(fmt, ap);
2883 	va_end(ap);
2884 	printf("%p: ", (void *)vp);
2885 	printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2886 	printf("    usecount %d, writecount %d, refcount %d mountedhere %p\n",
2887 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2888 	buf[0] = '\0';
2889 	buf[1] = '\0';
2890 	if (vp->v_vflag & VV_ROOT)
2891 		strlcat(buf, "|VV_ROOT", sizeof(buf));
2892 	if (vp->v_vflag & VV_ISTTY)
2893 		strlcat(buf, "|VV_ISTTY", sizeof(buf));
2894 	if (vp->v_vflag & VV_NOSYNC)
2895 		strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2896 	if (vp->v_vflag & VV_CACHEDLABEL)
2897 		strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2898 	if (vp->v_vflag & VV_TEXT)
2899 		strlcat(buf, "|VV_TEXT", sizeof(buf));
2900 	if (vp->v_vflag & VV_COPYONWRITE)
2901 		strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2902 	if (vp->v_vflag & VV_SYSTEM)
2903 		strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2904 	if (vp->v_vflag & VV_PROCDEP)
2905 		strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2906 	if (vp->v_vflag & VV_NOKNOTE)
2907 		strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2908 	if (vp->v_vflag & VV_DELETED)
2909 		strlcat(buf, "|VV_DELETED", sizeof(buf));
2910 	if (vp->v_vflag & VV_MD)
2911 		strlcat(buf, "|VV_MD", sizeof(buf));
2912 	flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC |
2913 	    VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2914 	    VV_NOKNOTE | VV_DELETED | VV_MD);
2915 	if (flags != 0) {
2916 		snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2917 		strlcat(buf, buf2, sizeof(buf));
2918 	}
2919 	if (vp->v_iflag & VI_MOUNT)
2920 		strlcat(buf, "|VI_MOUNT", sizeof(buf));
2921 	if (vp->v_iflag & VI_AGE)
2922 		strlcat(buf, "|VI_AGE", sizeof(buf));
2923 	if (vp->v_iflag & VI_DOOMED)
2924 		strlcat(buf, "|VI_DOOMED", sizeof(buf));
2925 	if (vp->v_iflag & VI_FREE)
2926 		strlcat(buf, "|VI_FREE", sizeof(buf));
2927 	if (vp->v_iflag & VI_DOINGINACT)
2928 		strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2929 	if (vp->v_iflag & VI_OWEINACT)
2930 		strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2931 	flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2932 	    VI_DOINGINACT | VI_OWEINACT);
2933 	if (flags != 0) {
2934 		snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2935 		strlcat(buf, buf2, sizeof(buf));
2936 	}
2937 	printf("    flags (%s)\n", buf + 1);
2938 	if (mtx_owned(VI_MTX(vp)))
2939 		printf(" VI_LOCKed");
2940 	if (vp->v_object != NULL)
2941 		printf("    v_object %p ref %d pages %d\n",
2942 		    vp->v_object, vp->v_object->ref_count,
2943 		    vp->v_object->resident_page_count);
2944 	printf("    ");
2945 	lockmgr_printinfo(vp->v_vnlock);
2946 	if (vp->v_data != NULL)
2947 		VOP_PRINT(vp);
2948 }
2949 
2950 #ifdef DDB
2951 /*
2952  * List all of the locked vnodes in the system.
2953  * Called when debugging the kernel.
2954  */
2955 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2956 {
2957 	struct mount *mp, *nmp;
2958 	struct vnode *vp;
2959 
2960 	/*
2961 	 * Note: because this is DDB, we can't obey the locking semantics
2962 	 * for these structures, which means we could catch an inconsistent
2963 	 * state and dereference a nasty pointer.  Not much to be done
2964 	 * about that.
2965 	 */
2966 	db_printf("Locked vnodes\n");
2967 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2968 		nmp = TAILQ_NEXT(mp, mnt_list);
2969 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2970 			if (vp->v_type != VMARKER &&
2971 			    VOP_ISLOCKED(vp))
2972 				vprint("", vp);
2973 		}
2974 		nmp = TAILQ_NEXT(mp, mnt_list);
2975 	}
2976 }
2977 
2978 /*
2979  * Show details about the given vnode.
2980  */
2981 DB_SHOW_COMMAND(vnode, db_show_vnode)
2982 {
2983 	struct vnode *vp;
2984 
2985 	if (!have_addr)
2986 		return;
2987 	vp = (struct vnode *)addr;
2988 	vn_printf(vp, "vnode ");
2989 }
2990 
2991 /*
2992  * Show details about the given mount point.
2993  */
2994 DB_SHOW_COMMAND(mount, db_show_mount)
2995 {
2996 	struct mount *mp;
2997 	struct vfsopt *opt;
2998 	struct statfs *sp;
2999 	struct vnode *vp;
3000 	char buf[512];
3001 	uint64_t mflags;
3002 	u_int flags;
3003 
3004 	if (!have_addr) {
3005 		/* No address given, print short info about all mount points. */
3006 		TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3007 			db_printf("%p %s on %s (%s)\n", mp,
3008 			    mp->mnt_stat.f_mntfromname,
3009 			    mp->mnt_stat.f_mntonname,
3010 			    mp->mnt_stat.f_fstypename);
3011 			if (db_pager_quit)
3012 				break;
3013 		}
3014 		db_printf("\nMore info: show mount <addr>\n");
3015 		return;
3016 	}
3017 
3018 	mp = (struct mount *)addr;
3019 	db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
3020 	    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
3021 
3022 	buf[0] = '\0';
3023 	mflags = mp->mnt_flag;
3024 #define	MNT_FLAG(flag)	do {						\
3025 	if (mflags & (flag)) {						\
3026 		if (buf[0] != '\0')					\
3027 			strlcat(buf, ", ", sizeof(buf));		\
3028 		strlcat(buf, (#flag) + 4, sizeof(buf));			\
3029 		mflags &= ~(flag);					\
3030 	}								\
3031 } while (0)
3032 	MNT_FLAG(MNT_RDONLY);
3033 	MNT_FLAG(MNT_SYNCHRONOUS);
3034 	MNT_FLAG(MNT_NOEXEC);
3035 	MNT_FLAG(MNT_NOSUID);
3036 	MNT_FLAG(MNT_UNION);
3037 	MNT_FLAG(MNT_ASYNC);
3038 	MNT_FLAG(MNT_SUIDDIR);
3039 	MNT_FLAG(MNT_SOFTDEP);
3040 	MNT_FLAG(MNT_SUJ);
3041 	MNT_FLAG(MNT_NOSYMFOLLOW);
3042 	MNT_FLAG(MNT_GJOURNAL);
3043 	MNT_FLAG(MNT_MULTILABEL);
3044 	MNT_FLAG(MNT_ACLS);
3045 	MNT_FLAG(MNT_NOATIME);
3046 	MNT_FLAG(MNT_NOCLUSTERR);
3047 	MNT_FLAG(MNT_NOCLUSTERW);
3048 	MNT_FLAG(MNT_NFS4ACLS);
3049 	MNT_FLAG(MNT_EXRDONLY);
3050 	MNT_FLAG(MNT_EXPORTED);
3051 	MNT_FLAG(MNT_DEFEXPORTED);
3052 	MNT_FLAG(MNT_EXPORTANON);
3053 	MNT_FLAG(MNT_EXKERB);
3054 	MNT_FLAG(MNT_EXPUBLIC);
3055 	MNT_FLAG(MNT_LOCAL);
3056 	MNT_FLAG(MNT_QUOTA);
3057 	MNT_FLAG(MNT_ROOTFS);
3058 	MNT_FLAG(MNT_USER);
3059 	MNT_FLAG(MNT_IGNORE);
3060 	MNT_FLAG(MNT_UPDATE);
3061 	MNT_FLAG(MNT_DELEXPORT);
3062 	MNT_FLAG(MNT_RELOAD);
3063 	MNT_FLAG(MNT_FORCE);
3064 	MNT_FLAG(MNT_SNAPSHOT);
3065 	MNT_FLAG(MNT_BYFSID);
3066 #undef MNT_FLAG
3067 	if (mflags != 0) {
3068 		if (buf[0] != '\0')
3069 			strlcat(buf, ", ", sizeof(buf));
3070 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3071 		    "0x%016jx", mflags);
3072 	}
3073 	db_printf("    mnt_flag = %s\n", buf);
3074 
3075 	buf[0] = '\0';
3076 	flags = mp->mnt_kern_flag;
3077 #define	MNT_KERN_FLAG(flag)	do {					\
3078 	if (flags & (flag)) {						\
3079 		if (buf[0] != '\0')					\
3080 			strlcat(buf, ", ", sizeof(buf));		\
3081 		strlcat(buf, (#flag) + 5, sizeof(buf));			\
3082 		flags &= ~(flag);					\
3083 	}								\
3084 } while (0)
3085 	MNT_KERN_FLAG(MNTK_UNMOUNTF);
3086 	MNT_KERN_FLAG(MNTK_ASYNC);
3087 	MNT_KERN_FLAG(MNTK_SOFTDEP);
3088 	MNT_KERN_FLAG(MNTK_NOINSMNTQ);
3089 	MNT_KERN_FLAG(MNTK_DRAINING);
3090 	MNT_KERN_FLAG(MNTK_REFEXPIRE);
3091 	MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
3092 	MNT_KERN_FLAG(MNTK_SHARED_WRITES);
3093 	MNT_KERN_FLAG(MNTK_NOASYNC);
3094 	MNT_KERN_FLAG(MNTK_UNMOUNT);
3095 	MNT_KERN_FLAG(MNTK_MWAIT);
3096 	MNT_KERN_FLAG(MNTK_SUSPEND);
3097 	MNT_KERN_FLAG(MNTK_SUSPEND2);
3098 	MNT_KERN_FLAG(MNTK_SUSPENDED);
3099 	MNT_KERN_FLAG(MNTK_MPSAFE);
3100 	MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
3101 	MNT_KERN_FLAG(MNTK_NOKNOTE);
3102 #undef MNT_KERN_FLAG
3103 	if (flags != 0) {
3104 		if (buf[0] != '\0')
3105 			strlcat(buf, ", ", sizeof(buf));
3106 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3107 		    "0x%08x", flags);
3108 	}
3109 	db_printf("    mnt_kern_flag = %s\n", buf);
3110 
3111 	db_printf("    mnt_opt = ");
3112 	opt = TAILQ_FIRST(mp->mnt_opt);
3113 	if (opt != NULL) {
3114 		db_printf("%s", opt->name);
3115 		opt = TAILQ_NEXT(opt, link);
3116 		while (opt != NULL) {
3117 			db_printf(", %s", opt->name);
3118 			opt = TAILQ_NEXT(opt, link);
3119 		}
3120 	}
3121 	db_printf("\n");
3122 
3123 	sp = &mp->mnt_stat;
3124 	db_printf("    mnt_stat = { version=%u type=%u flags=0x%016jx "
3125 	    "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
3126 	    "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
3127 	    "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
3128 	    (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
3129 	    (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
3130 	    (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
3131 	    (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
3132 	    (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
3133 	    (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
3134 	    (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
3135 	    (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
3136 
3137 	db_printf("    mnt_cred = { uid=%u ruid=%u",
3138 	    (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
3139 	if (jailed(mp->mnt_cred))
3140 		db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
3141 	db_printf(" }\n");
3142 	db_printf("    mnt_ref = %d\n", mp->mnt_ref);
3143 	db_printf("    mnt_gen = %d\n", mp->mnt_gen);
3144 	db_printf("    mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
3145 	db_printf("    mnt_activevnodelistsize = %d\n",
3146 	    mp->mnt_activevnodelistsize);
3147 	db_printf("    mnt_writeopcount = %d\n", mp->mnt_writeopcount);
3148 	db_printf("    mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
3149 	db_printf("    mnt_iosize_max = %d\n", mp->mnt_iosize_max);
3150 	db_printf("    mnt_hashseed = %u\n", mp->mnt_hashseed);
3151 	db_printf("    mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
3152 	db_printf("    mnt_secondary_accwrites = %d\n",
3153 	    mp->mnt_secondary_accwrites);
3154 	db_printf("    mnt_gjprovider = %s\n",
3155 	    mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
3156 
3157 	db_printf("\n\nList of active vnodes\n");
3158 	TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) {
3159 		if (vp->v_type != VMARKER) {
3160 			vn_printf(vp, "vnode ");
3161 			if (db_pager_quit)
3162 				break;
3163 		}
3164 	}
3165 	db_printf("\n\nList of inactive vnodes\n");
3166 	TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3167 		if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) {
3168 			vn_printf(vp, "vnode ");
3169 			if (db_pager_quit)
3170 				break;
3171 		}
3172 	}
3173 }
3174 #endif	/* DDB */
3175 
3176 /*
3177  * Fill in a struct xvfsconf based on a struct vfsconf.
3178  */
3179 static int
3180 vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp)
3181 {
3182 	struct xvfsconf xvfsp;
3183 
3184 	bzero(&xvfsp, sizeof(xvfsp));
3185 	strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3186 	xvfsp.vfc_typenum = vfsp->vfc_typenum;
3187 	xvfsp.vfc_refcount = vfsp->vfc_refcount;
3188 	xvfsp.vfc_flags = vfsp->vfc_flags;
3189 	/*
3190 	 * These are unused in userland, we keep them
3191 	 * to not break binary compatibility.
3192 	 */
3193 	xvfsp.vfc_vfsops = NULL;
3194 	xvfsp.vfc_next = NULL;
3195 	return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3196 }
3197 
3198 #ifdef COMPAT_FREEBSD32
3199 struct xvfsconf32 {
3200 	uint32_t	vfc_vfsops;
3201 	char		vfc_name[MFSNAMELEN];
3202 	int32_t		vfc_typenum;
3203 	int32_t		vfc_refcount;
3204 	int32_t		vfc_flags;
3205 	uint32_t	vfc_next;
3206 };
3207 
3208 static int
3209 vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp)
3210 {
3211 	struct xvfsconf32 xvfsp;
3212 
3213 	strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3214 	xvfsp.vfc_typenum = vfsp->vfc_typenum;
3215 	xvfsp.vfc_refcount = vfsp->vfc_refcount;
3216 	xvfsp.vfc_flags = vfsp->vfc_flags;
3217 	xvfsp.vfc_vfsops = 0;
3218 	xvfsp.vfc_next = 0;
3219 	return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3220 }
3221 #endif
3222 
3223 /*
3224  * Top level filesystem related information gathering.
3225  */
3226 static int
3227 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
3228 {
3229 	struct vfsconf *vfsp;
3230 	int error;
3231 
3232 	error = 0;
3233 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3234 #ifdef COMPAT_FREEBSD32
3235 		if (req->flags & SCTL_MASK32)
3236 			error = vfsconf2x32(req, vfsp);
3237 		else
3238 #endif
3239 			error = vfsconf2x(req, vfsp);
3240 		if (error)
3241 			break;
3242 	}
3243 	return (error);
3244 }
3245 
3246 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD,
3247     NULL, 0, sysctl_vfs_conflist,
3248     "S,xvfsconf", "List of all configured filesystems");
3249 
3250 #ifndef BURN_BRIDGES
3251 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
3252 
3253 static int
3254 vfs_sysctl(SYSCTL_HANDLER_ARGS)
3255 {
3256 	int *name = (int *)arg1 - 1;	/* XXX */
3257 	u_int namelen = arg2 + 1;	/* XXX */
3258 	struct vfsconf *vfsp;
3259 
3260 	log(LOG_WARNING, "userland calling deprecated sysctl, "
3261 	    "please rebuild world\n");
3262 
3263 #if 1 || defined(COMPAT_PRELITE2)
3264 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3265 	if (namelen == 1)
3266 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3267 #endif
3268 
3269 	switch (name[1]) {
3270 	case VFS_MAXTYPENUM:
3271 		if (namelen != 2)
3272 			return (ENOTDIR);
3273 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3274 	case VFS_CONF:
3275 		if (namelen != 3)
3276 			return (ENOTDIR);	/* overloaded */
3277 		TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
3278 			if (vfsp->vfc_typenum == name[2])
3279 				break;
3280 		if (vfsp == NULL)
3281 			return (EOPNOTSUPP);
3282 #ifdef COMPAT_FREEBSD32
3283 		if (req->flags & SCTL_MASK32)
3284 			return (vfsconf2x32(req, vfsp));
3285 		else
3286 #endif
3287 			return (vfsconf2x(req, vfsp));
3288 	}
3289 	return (EOPNOTSUPP);
3290 }
3291 
3292 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
3293     vfs_sysctl, "Generic filesystem");
3294 
3295 #if 1 || defined(COMPAT_PRELITE2)
3296 
3297 static int
3298 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3299 {
3300 	int error;
3301 	struct vfsconf *vfsp;
3302 	struct ovfsconf ovfs;
3303 
3304 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3305 		bzero(&ovfs, sizeof(ovfs));
3306 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
3307 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
3308 		ovfs.vfc_index = vfsp->vfc_typenum;
3309 		ovfs.vfc_refcount = vfsp->vfc_refcount;
3310 		ovfs.vfc_flags = vfsp->vfc_flags;
3311 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3312 		if (error)
3313 			return error;
3314 	}
3315 	return 0;
3316 }
3317 
3318 #endif /* 1 || COMPAT_PRELITE2 */
3319 #endif /* !BURN_BRIDGES */
3320 
3321 #define KINFO_VNODESLOP		10
3322 #ifdef notyet
3323 /*
3324  * Dump vnode list (via sysctl).
3325  */
3326 /* ARGSUSED */
3327 static int
3328 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3329 {
3330 	struct xvnode *xvn;
3331 	struct mount *mp;
3332 	struct vnode *vp;
3333 	int error, len, n;
3334 
3335 	/*
3336 	 * Stale numvnodes access is not fatal here.
3337 	 */
3338 	req->lock = 0;
3339 	len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3340 	if (!req->oldptr)
3341 		/* Make an estimate */
3342 		return (SYSCTL_OUT(req, 0, len));
3343 
3344 	error = sysctl_wire_old_buffer(req, 0);
3345 	if (error != 0)
3346 		return (error);
3347 	xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3348 	n = 0;
3349 	mtx_lock(&mountlist_mtx);
3350 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3351 		if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3352 			continue;
3353 		MNT_ILOCK(mp);
3354 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3355 			if (n == len)
3356 				break;
3357 			vref(vp);
3358 			xvn[n].xv_size = sizeof *xvn;
3359 			xvn[n].xv_vnode = vp;
3360 			xvn[n].xv_id = 0;	/* XXX compat */
3361 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3362 			XV_COPY(usecount);
3363 			XV_COPY(writecount);
3364 			XV_COPY(holdcnt);
3365 			XV_COPY(mount);
3366 			XV_COPY(numoutput);
3367 			XV_COPY(type);
3368 #undef XV_COPY
3369 			xvn[n].xv_flag = vp->v_vflag;
3370 
3371 			switch (vp->v_type) {
3372 			case VREG:
3373 			case VDIR:
3374 			case VLNK:
3375 				break;
3376 			case VBLK:
3377 			case VCHR:
3378 				if (vp->v_rdev == NULL) {
3379 					vrele(vp);
3380 					continue;
3381 				}
3382 				xvn[n].xv_dev = dev2udev(vp->v_rdev);
3383 				break;
3384 			case VSOCK:
3385 				xvn[n].xv_socket = vp->v_socket;
3386 				break;
3387 			case VFIFO:
3388 				xvn[n].xv_fifo = vp->v_fifoinfo;
3389 				break;
3390 			case VNON:
3391 			case VBAD:
3392 			default:
3393 				/* shouldn't happen? */
3394 				vrele(vp);
3395 				continue;
3396 			}
3397 			vrele(vp);
3398 			++n;
3399 		}
3400 		MNT_IUNLOCK(mp);
3401 		mtx_lock(&mountlist_mtx);
3402 		vfs_unbusy(mp);
3403 		if (n == len)
3404 			break;
3405 	}
3406 	mtx_unlock(&mountlist_mtx);
3407 
3408 	error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3409 	free(xvn, M_TEMP);
3410 	return (error);
3411 }
3412 
3413 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3414     0, 0, sysctl_vnode, "S,xvnode", "");
3415 #endif
3416 
3417 /*
3418  * Unmount all filesystems. The list is traversed in reverse order
3419  * of mounting to avoid dependencies.
3420  */
3421 void
3422 vfs_unmountall(void)
3423 {
3424 	struct mount *mp;
3425 	struct thread *td;
3426 	int error;
3427 
3428 	CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3429 	td = curthread;
3430 
3431 	/*
3432 	 * Since this only runs when rebooting, it is not interlocked.
3433 	 */
3434 	while(!TAILQ_EMPTY(&mountlist)) {
3435 		mp = TAILQ_LAST(&mountlist, mntlist);
3436 		error = dounmount(mp, MNT_FORCE, td);
3437 		if (error) {
3438 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
3439 			/*
3440 			 * XXX: Due to the way in which we mount the root
3441 			 * file system off of devfs, devfs will generate a
3442 			 * "busy" warning when we try to unmount it before
3443 			 * the root.  Don't print a warning as a result in
3444 			 * order to avoid false positive errors that may
3445 			 * cause needless upset.
3446 			 */
3447 			if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3448 				printf("unmount of %s failed (",
3449 				    mp->mnt_stat.f_mntonname);
3450 				if (error == EBUSY)
3451 					printf("BUSY)\n");
3452 				else
3453 					printf("%d)\n", error);
3454 			}
3455 		} else {
3456 			/* The unmount has removed mp from the mountlist */
3457 		}
3458 	}
3459 }
3460 
3461 /*
3462  * perform msync on all vnodes under a mount point
3463  * the mount point must be locked.
3464  */
3465 void
3466 vfs_msync(struct mount *mp, int flags)
3467 {
3468 	struct vnode *vp, *mvp;
3469 	struct vm_object *obj;
3470 
3471 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3472 	MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
3473 		obj = vp->v_object;
3474 		if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3475 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3476 			if (!vget(vp,
3477 			    LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3478 			    curthread)) {
3479 				if (vp->v_vflag & VV_NOSYNC) {	/* unlinked */
3480 					vput(vp);
3481 					continue;
3482 				}
3483 
3484 				obj = vp->v_object;
3485 				if (obj != NULL) {
3486 					VM_OBJECT_LOCK(obj);
3487 					vm_object_page_clean(obj, 0, 0,
3488 					    flags == MNT_WAIT ?
3489 					    OBJPC_SYNC : OBJPC_NOSYNC);
3490 					VM_OBJECT_UNLOCK(obj);
3491 				}
3492 				vput(vp);
3493 			}
3494 		} else
3495 			VI_UNLOCK(vp);
3496 	}
3497 }
3498 
3499 static void
3500 destroy_vpollinfo(struct vpollinfo *vi)
3501 {
3502 	seldrain(&vi->vpi_selinfo);
3503 	knlist_destroy(&vi->vpi_selinfo.si_note);
3504 	mtx_destroy(&vi->vpi_lock);
3505 	uma_zfree(vnodepoll_zone, vi);
3506 }
3507 
3508 /*
3509  * Initalize per-vnode helper structure to hold poll-related state.
3510  */
3511 void
3512 v_addpollinfo(struct vnode *vp)
3513 {
3514 	struct vpollinfo *vi;
3515 
3516 	if (vp->v_pollinfo != NULL)
3517 		return;
3518 	vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3519 	mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3520 	knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3521 	    vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3522 	VI_LOCK(vp);
3523 	if (vp->v_pollinfo != NULL) {
3524 		VI_UNLOCK(vp);
3525 		destroy_vpollinfo(vi);
3526 		return;
3527 	}
3528 	vp->v_pollinfo = vi;
3529 	VI_UNLOCK(vp);
3530 }
3531 
3532 /*
3533  * Record a process's interest in events which might happen to
3534  * a vnode.  Because poll uses the historic select-style interface
3535  * internally, this routine serves as both the ``check for any
3536  * pending events'' and the ``record my interest in future events''
3537  * functions.  (These are done together, while the lock is held,
3538  * to avoid race conditions.)
3539  */
3540 int
3541 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3542 {
3543 
3544 	v_addpollinfo(vp);
3545 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3546 	if (vp->v_pollinfo->vpi_revents & events) {
3547 		/*
3548 		 * This leaves events we are not interested
3549 		 * in available for the other process which
3550 		 * which presumably had requested them
3551 		 * (otherwise they would never have been
3552 		 * recorded).
3553 		 */
3554 		events &= vp->v_pollinfo->vpi_revents;
3555 		vp->v_pollinfo->vpi_revents &= ~events;
3556 
3557 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
3558 		return (events);
3559 	}
3560 	vp->v_pollinfo->vpi_events |= events;
3561 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3562 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3563 	return (0);
3564 }
3565 
3566 /*
3567  * Routine to create and manage a filesystem syncer vnode.
3568  */
3569 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
3570 static int	sync_fsync(struct  vop_fsync_args *);
3571 static int	sync_inactive(struct  vop_inactive_args *);
3572 static int	sync_reclaim(struct  vop_reclaim_args *);
3573 
3574 static struct vop_vector sync_vnodeops = {
3575 	.vop_bypass =	VOP_EOPNOTSUPP,
3576 	.vop_close =	sync_close,		/* close */
3577 	.vop_fsync =	sync_fsync,		/* fsync */
3578 	.vop_inactive =	sync_inactive,	/* inactive */
3579 	.vop_reclaim =	sync_reclaim,	/* reclaim */
3580 	.vop_lock1 =	vop_stdlock,	/* lock */
3581 	.vop_unlock =	vop_stdunlock,	/* unlock */
3582 	.vop_islocked =	vop_stdislocked,	/* islocked */
3583 };
3584 
3585 /*
3586  * Create a new filesystem syncer vnode for the specified mount point.
3587  */
3588 void
3589 vfs_allocate_syncvnode(struct mount *mp)
3590 {
3591 	struct vnode *vp;
3592 	struct bufobj *bo;
3593 	static long start, incr, next;
3594 	int error;
3595 
3596 	/* Allocate a new vnode */
3597 	error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
3598 	if (error != 0)
3599 		panic("vfs_allocate_syncvnode: getnewvnode() failed");
3600 	vp->v_type = VNON;
3601 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3602 	vp->v_vflag |= VV_FORCEINSMQ;
3603 	error = insmntque(vp, mp);
3604 	if (error != 0)
3605 		panic("vfs_allocate_syncvnode: insmntque() failed");
3606 	vp->v_vflag &= ~VV_FORCEINSMQ;
3607 	VOP_UNLOCK(vp, 0);
3608 	/*
3609 	 * Place the vnode onto the syncer worklist. We attempt to
3610 	 * scatter them about on the list so that they will go off
3611 	 * at evenly distributed times even if all the filesystems
3612 	 * are mounted at once.
3613 	 */
3614 	next += incr;
3615 	if (next == 0 || next > syncer_maxdelay) {
3616 		start /= 2;
3617 		incr /= 2;
3618 		if (start == 0) {
3619 			start = syncer_maxdelay / 2;
3620 			incr = syncer_maxdelay;
3621 		}
3622 		next = start;
3623 	}
3624 	bo = &vp->v_bufobj;
3625 	BO_LOCK(bo);
3626 	vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3627 	/* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3628 	mtx_lock(&sync_mtx);
3629 	sync_vnode_count++;
3630 	if (mp->mnt_syncer == NULL) {
3631 		mp->mnt_syncer = vp;
3632 		vp = NULL;
3633 	}
3634 	mtx_unlock(&sync_mtx);
3635 	BO_UNLOCK(bo);
3636 	if (vp != NULL) {
3637 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3638 		vgone(vp);
3639 		vput(vp);
3640 	}
3641 }
3642 
3643 void
3644 vfs_deallocate_syncvnode(struct mount *mp)
3645 {
3646 	struct vnode *vp;
3647 
3648 	mtx_lock(&sync_mtx);
3649 	vp = mp->mnt_syncer;
3650 	if (vp != NULL)
3651 		mp->mnt_syncer = NULL;
3652 	mtx_unlock(&sync_mtx);
3653 	if (vp != NULL)
3654 		vrele(vp);
3655 }
3656 
3657 /*
3658  * Do a lazy sync of the filesystem.
3659  */
3660 static int
3661 sync_fsync(struct vop_fsync_args *ap)
3662 {
3663 	struct vnode *syncvp = ap->a_vp;
3664 	struct mount *mp = syncvp->v_mount;
3665 	int error, save;
3666 	struct bufobj *bo;
3667 
3668 	/*
3669 	 * We only need to do something if this is a lazy evaluation.
3670 	 */
3671 	if (ap->a_waitfor != MNT_LAZY)
3672 		return (0);
3673 
3674 	/*
3675 	 * Move ourselves to the back of the sync list.
3676 	 */
3677 	bo = &syncvp->v_bufobj;
3678 	BO_LOCK(bo);
3679 	vn_syncer_add_to_worklist(bo, syncdelay);
3680 	BO_UNLOCK(bo);
3681 
3682 	/*
3683 	 * Walk the list of vnodes pushing all that are dirty and
3684 	 * not already on the sync list.
3685 	 */
3686 	mtx_lock(&mountlist_mtx);
3687 	if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) {
3688 		mtx_unlock(&mountlist_mtx);
3689 		return (0);
3690 	}
3691 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3692 		vfs_unbusy(mp);
3693 		return (0);
3694 	}
3695 	save = curthread_pflags_set(TDP_SYNCIO);
3696 	vfs_msync(mp, MNT_NOWAIT);
3697 	error = VFS_SYNC(mp, MNT_LAZY);
3698 	curthread_pflags_restore(save);
3699 	vn_finished_write(mp);
3700 	vfs_unbusy(mp);
3701 	return (error);
3702 }
3703 
3704 /*
3705  * The syncer vnode is no referenced.
3706  */
3707 static int
3708 sync_inactive(struct vop_inactive_args *ap)
3709 {
3710 
3711 	vgone(ap->a_vp);
3712 	return (0);
3713 }
3714 
3715 /*
3716  * The syncer vnode is no longer needed and is being decommissioned.
3717  *
3718  * Modifications to the worklist must be protected by sync_mtx.
3719  */
3720 static int
3721 sync_reclaim(struct vop_reclaim_args *ap)
3722 {
3723 	struct vnode *vp = ap->a_vp;
3724 	struct bufobj *bo;
3725 
3726 	bo = &vp->v_bufobj;
3727 	BO_LOCK(bo);
3728 	mtx_lock(&sync_mtx);
3729 	if (vp->v_mount->mnt_syncer == vp)
3730 		vp->v_mount->mnt_syncer = NULL;
3731 	if (bo->bo_flag & BO_ONWORKLST) {
3732 		LIST_REMOVE(bo, bo_synclist);
3733 		syncer_worklist_len--;
3734 		sync_vnode_count--;
3735 		bo->bo_flag &= ~BO_ONWORKLST;
3736 	}
3737 	mtx_unlock(&sync_mtx);
3738 	BO_UNLOCK(bo);
3739 
3740 	return (0);
3741 }
3742 
3743 /*
3744  * Check if vnode represents a disk device
3745  */
3746 int
3747 vn_isdisk(struct vnode *vp, int *errp)
3748 {
3749 	int error;
3750 
3751 	error = 0;
3752 	dev_lock();
3753 	if (vp->v_type != VCHR)
3754 		error = ENOTBLK;
3755 	else if (vp->v_rdev == NULL)
3756 		error = ENXIO;
3757 	else if (vp->v_rdev->si_devsw == NULL)
3758 		error = ENXIO;
3759 	else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3760 		error = ENOTBLK;
3761 	dev_unlock();
3762 	if (errp != NULL)
3763 		*errp = error;
3764 	return (error == 0);
3765 }
3766 
3767 /*
3768  * Common filesystem object access control check routine.  Accepts a
3769  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3770  * and optional call-by-reference privused argument allowing vaccess()
3771  * to indicate to the caller whether privilege was used to satisfy the
3772  * request (obsoleted).  Returns 0 on success, or an errno on failure.
3773  */
3774 int
3775 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3776     accmode_t accmode, struct ucred *cred, int *privused)
3777 {
3778 	accmode_t dac_granted;
3779 	accmode_t priv_granted;
3780 
3781 	KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3782 	    ("invalid bit in accmode"));
3783 	KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
3784 	    ("VAPPEND without VWRITE"));
3785 
3786 	/*
3787 	 * Look for a normal, non-privileged way to access the file/directory
3788 	 * as requested.  If it exists, go with that.
3789 	 */
3790 
3791 	if (privused != NULL)
3792 		*privused = 0;
3793 
3794 	dac_granted = 0;
3795 
3796 	/* Check the owner. */
3797 	if (cred->cr_uid == file_uid) {
3798 		dac_granted |= VADMIN;
3799 		if (file_mode & S_IXUSR)
3800 			dac_granted |= VEXEC;
3801 		if (file_mode & S_IRUSR)
3802 			dac_granted |= VREAD;
3803 		if (file_mode & S_IWUSR)
3804 			dac_granted |= (VWRITE | VAPPEND);
3805 
3806 		if ((accmode & dac_granted) == accmode)
3807 			return (0);
3808 
3809 		goto privcheck;
3810 	}
3811 
3812 	/* Otherwise, check the groups (first match) */
3813 	if (groupmember(file_gid, cred)) {
3814 		if (file_mode & S_IXGRP)
3815 			dac_granted |= VEXEC;
3816 		if (file_mode & S_IRGRP)
3817 			dac_granted |= VREAD;
3818 		if (file_mode & S_IWGRP)
3819 			dac_granted |= (VWRITE | VAPPEND);
3820 
3821 		if ((accmode & dac_granted) == accmode)
3822 			return (0);
3823 
3824 		goto privcheck;
3825 	}
3826 
3827 	/* Otherwise, check everyone else. */
3828 	if (file_mode & S_IXOTH)
3829 		dac_granted |= VEXEC;
3830 	if (file_mode & S_IROTH)
3831 		dac_granted |= VREAD;
3832 	if (file_mode & S_IWOTH)
3833 		dac_granted |= (VWRITE | VAPPEND);
3834 	if ((accmode & dac_granted) == accmode)
3835 		return (0);
3836 
3837 privcheck:
3838 	/*
3839 	 * Build a privilege mask to determine if the set of privileges
3840 	 * satisfies the requirements when combined with the granted mask
3841 	 * from above.  For each privilege, if the privilege is required,
3842 	 * bitwise or the request type onto the priv_granted mask.
3843 	 */
3844 	priv_granted = 0;
3845 
3846 	if (type == VDIR) {
3847 		/*
3848 		 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3849 		 * requests, instead of PRIV_VFS_EXEC.
3850 		 */
3851 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3852 		    !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3853 			priv_granted |= VEXEC;
3854 	} else {
3855 		/*
3856 		 * Ensure that at least one execute bit is on. Otherwise,
3857 		 * a privileged user will always succeed, and we don't want
3858 		 * this to happen unless the file really is executable.
3859 		 */
3860 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3861 		    (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
3862 		    !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3863 			priv_granted |= VEXEC;
3864 	}
3865 
3866 	if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
3867 	    !priv_check_cred(cred, PRIV_VFS_READ, 0))
3868 		priv_granted |= VREAD;
3869 
3870 	if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3871 	    !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3872 		priv_granted |= (VWRITE | VAPPEND);
3873 
3874 	if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3875 	    !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3876 		priv_granted |= VADMIN;
3877 
3878 	if ((accmode & (priv_granted | dac_granted)) == accmode) {
3879 		/* XXX audit: privilege used */
3880 		if (privused != NULL)
3881 			*privused = 1;
3882 		return (0);
3883 	}
3884 
3885 	return ((accmode & VADMIN) ? EPERM : EACCES);
3886 }
3887 
3888 /*
3889  * Credential check based on process requesting service, and per-attribute
3890  * permissions.
3891  */
3892 int
3893 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3894     struct thread *td, accmode_t accmode)
3895 {
3896 
3897 	/*
3898 	 * Kernel-invoked always succeeds.
3899 	 */
3900 	if (cred == NOCRED)
3901 		return (0);
3902 
3903 	/*
3904 	 * Do not allow privileged processes in jail to directly manipulate
3905 	 * system attributes.
3906 	 */
3907 	switch (attrnamespace) {
3908 	case EXTATTR_NAMESPACE_SYSTEM:
3909 		/* Potentially should be: return (EPERM); */
3910 		return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3911 	case EXTATTR_NAMESPACE_USER:
3912 		return (VOP_ACCESS(vp, accmode, cred, td));
3913 	default:
3914 		return (EPERM);
3915 	}
3916 }
3917 
3918 #ifdef DEBUG_VFS_LOCKS
3919 /*
3920  * This only exists to supress warnings from unlocked specfs accesses.  It is
3921  * no longer ok to have an unlocked VFS.
3922  */
3923 #define	IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL ||		\
3924 	(vp)->v_type == VCHR ||	(vp)->v_type == VBAD)
3925 
3926 int vfs_badlock_ddb = 1;	/* Drop into debugger on violation. */
3927 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
3928     "Drop into debugger on lock violation");
3929 
3930 int vfs_badlock_mutex = 1;	/* Check for interlock across VOPs. */
3931 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
3932     0, "Check for interlock across VOPs");
3933 
3934 int vfs_badlock_print = 1;	/* Print lock violations. */
3935 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
3936     0, "Print lock violations");
3937 
3938 #ifdef KDB
3939 int vfs_badlock_backtrace = 1;	/* Print backtrace at lock violations. */
3940 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
3941     &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
3942 #endif
3943 
3944 static void
3945 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3946 {
3947 
3948 #ifdef KDB
3949 	if (vfs_badlock_backtrace)
3950 		kdb_backtrace();
3951 #endif
3952 	if (vfs_badlock_print)
3953 		printf("%s: %p %s\n", str, (void *)vp, msg);
3954 	if (vfs_badlock_ddb)
3955 		kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3956 }
3957 
3958 void
3959 assert_vi_locked(struct vnode *vp, const char *str)
3960 {
3961 
3962 	if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3963 		vfs_badlock("interlock is not locked but should be", str, vp);
3964 }
3965 
3966 void
3967 assert_vi_unlocked(struct vnode *vp, const char *str)
3968 {
3969 
3970 	if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3971 		vfs_badlock("interlock is locked but should not be", str, vp);
3972 }
3973 
3974 void
3975 assert_vop_locked(struct vnode *vp, const char *str)
3976 {
3977 
3978 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0)
3979 		vfs_badlock("is not locked but should be", str, vp);
3980 }
3981 
3982 void
3983 assert_vop_unlocked(struct vnode *vp, const char *str)
3984 {
3985 
3986 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
3987 		vfs_badlock("is locked but should not be", str, vp);
3988 }
3989 
3990 void
3991 assert_vop_elocked(struct vnode *vp, const char *str)
3992 {
3993 
3994 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
3995 		vfs_badlock("is not exclusive locked but should be", str, vp);
3996 }
3997 
3998 #if 0
3999 void
4000 assert_vop_elocked_other(struct vnode *vp, const char *str)
4001 {
4002 
4003 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
4004 		vfs_badlock("is not exclusive locked by another thread",
4005 		    str, vp);
4006 }
4007 
4008 void
4009 assert_vop_slocked(struct vnode *vp, const char *str)
4010 {
4011 
4012 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
4013 		vfs_badlock("is not locked shared but should be", str, vp);
4014 }
4015 #endif /* 0 */
4016 #endif /* DEBUG_VFS_LOCKS */
4017 
4018 void
4019 vop_rename_fail(struct vop_rename_args *ap)
4020 {
4021 
4022 	if (ap->a_tvp != NULL)
4023 		vput(ap->a_tvp);
4024 	if (ap->a_tdvp == ap->a_tvp)
4025 		vrele(ap->a_tdvp);
4026 	else
4027 		vput(ap->a_tdvp);
4028 	vrele(ap->a_fdvp);
4029 	vrele(ap->a_fvp);
4030 }
4031 
4032 void
4033 vop_rename_pre(void *ap)
4034 {
4035 	struct vop_rename_args *a = ap;
4036 
4037 #ifdef DEBUG_VFS_LOCKS
4038 	if (a->a_tvp)
4039 		ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
4040 	ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
4041 	ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
4042 	ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
4043 
4044 	/* Check the source (from). */
4045 	if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
4046 	    (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
4047 		ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
4048 	if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
4049 		ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
4050 
4051 	/* Check the target. */
4052 	if (a->a_tvp)
4053 		ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
4054 	ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
4055 #endif
4056 	if (a->a_tdvp != a->a_fdvp)
4057 		vhold(a->a_fdvp);
4058 	if (a->a_tvp != a->a_fvp)
4059 		vhold(a->a_fvp);
4060 	vhold(a->a_tdvp);
4061 	if (a->a_tvp)
4062 		vhold(a->a_tvp);
4063 }
4064 
4065 void
4066 vop_strategy_pre(void *ap)
4067 {
4068 #ifdef DEBUG_VFS_LOCKS
4069 	struct vop_strategy_args *a;
4070 	struct buf *bp;
4071 
4072 	a = ap;
4073 	bp = a->a_bp;
4074 
4075 	/*
4076 	 * Cluster ops lock their component buffers but not the IO container.
4077 	 */
4078 	if ((bp->b_flags & B_CLUSTER) != 0)
4079 		return;
4080 
4081 	if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
4082 		if (vfs_badlock_print)
4083 			printf(
4084 			    "VOP_STRATEGY: bp is not locked but should be\n");
4085 		if (vfs_badlock_ddb)
4086 			kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
4087 	}
4088 #endif
4089 }
4090 
4091 void
4092 vop_lookup_pre(void *ap)
4093 {
4094 #ifdef DEBUG_VFS_LOCKS
4095 	struct vop_lookup_args *a;
4096 	struct vnode *dvp;
4097 
4098 	a = ap;
4099 	dvp = a->a_dvp;
4100 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
4101 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
4102 #endif
4103 }
4104 
4105 void
4106 vop_lookup_post(void *ap, int rc)
4107 {
4108 #ifdef DEBUG_VFS_LOCKS
4109 	struct vop_lookup_args *a;
4110 	struct vnode *dvp;
4111 	struct vnode *vp;
4112 
4113 	a = ap;
4114 	dvp = a->a_dvp;
4115 	vp = *(a->a_vpp);
4116 
4117 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
4118 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
4119 
4120 	if (!rc)
4121 		ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
4122 #endif
4123 }
4124 
4125 void
4126 vop_lock_pre(void *ap)
4127 {
4128 #ifdef DEBUG_VFS_LOCKS
4129 	struct vop_lock1_args *a = ap;
4130 
4131 	if ((a->a_flags & LK_INTERLOCK) == 0)
4132 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4133 	else
4134 		ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
4135 #endif
4136 }
4137 
4138 void
4139 vop_lock_post(void *ap, int rc)
4140 {
4141 #ifdef DEBUG_VFS_LOCKS
4142 	struct vop_lock1_args *a = ap;
4143 
4144 	ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4145 	if (rc == 0)
4146 		ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
4147 #endif
4148 }
4149 
4150 void
4151 vop_unlock_pre(void *ap)
4152 {
4153 #ifdef DEBUG_VFS_LOCKS
4154 	struct vop_unlock_args *a = ap;
4155 
4156 	if (a->a_flags & LK_INTERLOCK)
4157 		ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
4158 	ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
4159 #endif
4160 }
4161 
4162 void
4163 vop_unlock_post(void *ap, int rc)
4164 {
4165 #ifdef DEBUG_VFS_LOCKS
4166 	struct vop_unlock_args *a = ap;
4167 
4168 	if (a->a_flags & LK_INTERLOCK)
4169 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
4170 #endif
4171 }
4172 
4173 void
4174 vop_create_post(void *ap, int rc)
4175 {
4176 	struct vop_create_args *a = ap;
4177 
4178 	if (!rc)
4179 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4180 }
4181 
4182 void
4183 vop_deleteextattr_post(void *ap, int rc)
4184 {
4185 	struct vop_deleteextattr_args *a = ap;
4186 
4187 	if (!rc)
4188 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4189 }
4190 
4191 void
4192 vop_link_post(void *ap, int rc)
4193 {
4194 	struct vop_link_args *a = ap;
4195 
4196 	if (!rc) {
4197 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
4198 		VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
4199 	}
4200 }
4201 
4202 void
4203 vop_mkdir_post(void *ap, int rc)
4204 {
4205 	struct vop_mkdir_args *a = ap;
4206 
4207 	if (!rc)
4208 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4209 }
4210 
4211 void
4212 vop_mknod_post(void *ap, int rc)
4213 {
4214 	struct vop_mknod_args *a = ap;
4215 
4216 	if (!rc)
4217 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4218 }
4219 
4220 void
4221 vop_remove_post(void *ap, int rc)
4222 {
4223 	struct vop_remove_args *a = ap;
4224 
4225 	if (!rc) {
4226 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4227 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4228 	}
4229 }
4230 
4231 void
4232 vop_rename_post(void *ap, int rc)
4233 {
4234 	struct vop_rename_args *a = ap;
4235 
4236 	if (!rc) {
4237 		VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
4238 		VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
4239 		VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
4240 		if (a->a_tvp)
4241 			VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
4242 	}
4243 	if (a->a_tdvp != a->a_fdvp)
4244 		vdrop(a->a_fdvp);
4245 	if (a->a_tvp != a->a_fvp)
4246 		vdrop(a->a_fvp);
4247 	vdrop(a->a_tdvp);
4248 	if (a->a_tvp)
4249 		vdrop(a->a_tvp);
4250 }
4251 
4252 void
4253 vop_rmdir_post(void *ap, int rc)
4254 {
4255 	struct vop_rmdir_args *a = ap;
4256 
4257 	if (!rc) {
4258 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4259 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4260 	}
4261 }
4262 
4263 void
4264 vop_setattr_post(void *ap, int rc)
4265 {
4266 	struct vop_setattr_args *a = ap;
4267 
4268 	if (!rc)
4269 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4270 }
4271 
4272 void
4273 vop_setextattr_post(void *ap, int rc)
4274 {
4275 	struct vop_setextattr_args *a = ap;
4276 
4277 	if (!rc)
4278 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4279 }
4280 
4281 void
4282 vop_symlink_post(void *ap, int rc)
4283 {
4284 	struct vop_symlink_args *a = ap;
4285 
4286 	if (!rc)
4287 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4288 }
4289 
4290 static struct knlist fs_knlist;
4291 
4292 static void
4293 vfs_event_init(void *arg)
4294 {
4295 	knlist_init_mtx(&fs_knlist, NULL);
4296 }
4297 /* XXX - correct order? */
4298 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4299 
4300 void
4301 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
4302 {
4303 
4304 	KNOTE_UNLOCKED(&fs_knlist, event);
4305 }
4306 
4307 static int	filt_fsattach(struct knote *kn);
4308 static void	filt_fsdetach(struct knote *kn);
4309 static int	filt_fsevent(struct knote *kn, long hint);
4310 
4311 struct filterops fs_filtops = {
4312 	.f_isfd = 0,
4313 	.f_attach = filt_fsattach,
4314 	.f_detach = filt_fsdetach,
4315 	.f_event = filt_fsevent
4316 };
4317 
4318 static int
4319 filt_fsattach(struct knote *kn)
4320 {
4321 
4322 	kn->kn_flags |= EV_CLEAR;
4323 	knlist_add(&fs_knlist, kn, 0);
4324 	return (0);
4325 }
4326 
4327 static void
4328 filt_fsdetach(struct knote *kn)
4329 {
4330 
4331 	knlist_remove(&fs_knlist, kn, 0);
4332 }
4333 
4334 static int
4335 filt_fsevent(struct knote *kn, long hint)
4336 {
4337 
4338 	kn->kn_fflags |= hint;
4339 	return (kn->kn_fflags != 0);
4340 }
4341 
4342 static int
4343 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4344 {
4345 	struct vfsidctl vc;
4346 	int error;
4347 	struct mount *mp;
4348 
4349 	error = SYSCTL_IN(req, &vc, sizeof(vc));
4350 	if (error)
4351 		return (error);
4352 	if (vc.vc_vers != VFS_CTL_VERS1)
4353 		return (EINVAL);
4354 	mp = vfs_getvfs(&vc.vc_fsid);
4355 	if (mp == NULL)
4356 		return (ENOENT);
4357 	/* ensure that a specific sysctl goes to the right filesystem. */
4358 	if (strcmp(vc.vc_fstypename, "*") != 0 &&
4359 	    strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4360 		vfs_rel(mp);
4361 		return (EINVAL);
4362 	}
4363 	VCTLTOREQ(&vc, req);
4364 	error = VFS_SYSCTL(mp, vc.vc_op, req);
4365 	vfs_rel(mp);
4366 	return (error);
4367 }
4368 
4369 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
4370     NULL, 0, sysctl_vfs_ctl, "",
4371     "Sysctl by fsid");
4372 
4373 /*
4374  * Function to initialize a va_filerev field sensibly.
4375  * XXX: Wouldn't a random number make a lot more sense ??
4376  */
4377 u_quad_t
4378 init_va_filerev(void)
4379 {
4380 	struct bintime bt;
4381 
4382 	getbinuptime(&bt);
4383 	return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4384 }
4385 
4386 static int	filt_vfsread(struct knote *kn, long hint);
4387 static int	filt_vfswrite(struct knote *kn, long hint);
4388 static int	filt_vfsvnode(struct knote *kn, long hint);
4389 static void	filt_vfsdetach(struct knote *kn);
4390 static struct filterops vfsread_filtops = {
4391 	.f_isfd = 1,
4392 	.f_detach = filt_vfsdetach,
4393 	.f_event = filt_vfsread
4394 };
4395 static struct filterops vfswrite_filtops = {
4396 	.f_isfd = 1,
4397 	.f_detach = filt_vfsdetach,
4398 	.f_event = filt_vfswrite
4399 };
4400 static struct filterops vfsvnode_filtops = {
4401 	.f_isfd = 1,
4402 	.f_detach = filt_vfsdetach,
4403 	.f_event = filt_vfsvnode
4404 };
4405 
4406 static void
4407 vfs_knllock(void *arg)
4408 {
4409 	struct vnode *vp = arg;
4410 
4411 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4412 }
4413 
4414 static void
4415 vfs_knlunlock(void *arg)
4416 {
4417 	struct vnode *vp = arg;
4418 
4419 	VOP_UNLOCK(vp, 0);
4420 }
4421 
4422 static void
4423 vfs_knl_assert_locked(void *arg)
4424 {
4425 #ifdef DEBUG_VFS_LOCKS
4426 	struct vnode *vp = arg;
4427 
4428 	ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4429 #endif
4430 }
4431 
4432 static void
4433 vfs_knl_assert_unlocked(void *arg)
4434 {
4435 #ifdef DEBUG_VFS_LOCKS
4436 	struct vnode *vp = arg;
4437 
4438 	ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4439 #endif
4440 }
4441 
4442 int
4443 vfs_kqfilter(struct vop_kqfilter_args *ap)
4444 {
4445 	struct vnode *vp = ap->a_vp;
4446 	struct knote *kn = ap->a_kn;
4447 	struct knlist *knl;
4448 
4449 	switch (kn->kn_filter) {
4450 	case EVFILT_READ:
4451 		kn->kn_fop = &vfsread_filtops;
4452 		break;
4453 	case EVFILT_WRITE:
4454 		kn->kn_fop = &vfswrite_filtops;
4455 		break;
4456 	case EVFILT_VNODE:
4457 		kn->kn_fop = &vfsvnode_filtops;
4458 		break;
4459 	default:
4460 		return (EINVAL);
4461 	}
4462 
4463 	kn->kn_hook = (caddr_t)vp;
4464 
4465 	v_addpollinfo(vp);
4466 	if (vp->v_pollinfo == NULL)
4467 		return (ENOMEM);
4468 	knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4469 	knlist_add(knl, kn, 0);
4470 
4471 	return (0);
4472 }
4473 
4474 /*
4475  * Detach knote from vnode
4476  */
4477 static void
4478 filt_vfsdetach(struct knote *kn)
4479 {
4480 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4481 
4482 	KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4483 	knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4484 }
4485 
4486 /*ARGSUSED*/
4487 static int
4488 filt_vfsread(struct knote *kn, long hint)
4489 {
4490 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4491 	struct vattr va;
4492 	int res;
4493 
4494 	/*
4495 	 * filesystem is gone, so set the EOF flag and schedule
4496 	 * the knote for deletion.
4497 	 */
4498 	if (hint == NOTE_REVOKE) {
4499 		VI_LOCK(vp);
4500 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4501 		VI_UNLOCK(vp);
4502 		return (1);
4503 	}
4504 
4505 	if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4506 		return (0);
4507 
4508 	VI_LOCK(vp);
4509 	kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4510 	res = (kn->kn_data != 0);
4511 	VI_UNLOCK(vp);
4512 	return (res);
4513 }
4514 
4515 /*ARGSUSED*/
4516 static int
4517 filt_vfswrite(struct knote *kn, long hint)
4518 {
4519 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4520 
4521 	VI_LOCK(vp);
4522 
4523 	/*
4524 	 * filesystem is gone, so set the EOF flag and schedule
4525 	 * the knote for deletion.
4526 	 */
4527 	if (hint == NOTE_REVOKE)
4528 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4529 
4530 	kn->kn_data = 0;
4531 	VI_UNLOCK(vp);
4532 	return (1);
4533 }
4534 
4535 static int
4536 filt_vfsvnode(struct knote *kn, long hint)
4537 {
4538 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4539 	int res;
4540 
4541 	VI_LOCK(vp);
4542 	if (kn->kn_sfflags & hint)
4543 		kn->kn_fflags |= hint;
4544 	if (hint == NOTE_REVOKE) {
4545 		kn->kn_flags |= EV_EOF;
4546 		VI_UNLOCK(vp);
4547 		return (1);
4548 	}
4549 	res = (kn->kn_fflags != 0);
4550 	VI_UNLOCK(vp);
4551 	return (res);
4552 }
4553 
4554 int
4555 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4556 {
4557 	int error;
4558 
4559 	if (dp->d_reclen > ap->a_uio->uio_resid)
4560 		return (ENAMETOOLONG);
4561 	error = uiomove(dp, dp->d_reclen, ap->a_uio);
4562 	if (error) {
4563 		if (ap->a_ncookies != NULL) {
4564 			if (ap->a_cookies != NULL)
4565 				free(ap->a_cookies, M_TEMP);
4566 			ap->a_cookies = NULL;
4567 			*ap->a_ncookies = 0;
4568 		}
4569 		return (error);
4570 	}
4571 	if (ap->a_ncookies == NULL)
4572 		return (0);
4573 
4574 	KASSERT(ap->a_cookies,
4575 	    ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4576 
4577 	*ap->a_cookies = realloc(*ap->a_cookies,
4578 	    (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4579 	(*ap->a_cookies)[*ap->a_ncookies] = off;
4580 	return (0);
4581 }
4582 
4583 /*
4584  * Mark for update the access time of the file if the filesystem
4585  * supports VOP_MARKATIME.  This functionality is used by execve and
4586  * mmap, so we want to avoid the I/O implied by directly setting
4587  * va_atime for the sake of efficiency.
4588  */
4589 void
4590 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4591 {
4592 	struct mount *mp;
4593 
4594 	mp = vp->v_mount;
4595 	ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4596 	if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4597 		(void)VOP_MARKATIME(vp);
4598 }
4599 
4600 /*
4601  * The purpose of this routine is to remove granularity from accmode_t,
4602  * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4603  * VADMIN and VAPPEND.
4604  *
4605  * If it returns 0, the caller is supposed to continue with the usual
4606  * access checks using 'accmode' as modified by this routine.  If it
4607  * returns nonzero value, the caller is supposed to return that value
4608  * as errno.
4609  *
4610  * Note that after this routine runs, accmode may be zero.
4611  */
4612 int
4613 vfs_unixify_accmode(accmode_t *accmode)
4614 {
4615 	/*
4616 	 * There is no way to specify explicit "deny" rule using
4617 	 * file mode or POSIX.1e ACLs.
4618 	 */
4619 	if (*accmode & VEXPLICIT_DENY) {
4620 		*accmode = 0;
4621 		return (0);
4622 	}
4623 
4624 	/*
4625 	 * None of these can be translated into usual access bits.
4626 	 * Also, the common case for NFSv4 ACLs is to not contain
4627 	 * either of these bits. Caller should check for VWRITE
4628 	 * on the containing directory instead.
4629 	 */
4630 	if (*accmode & (VDELETE_CHILD | VDELETE))
4631 		return (EPERM);
4632 
4633 	if (*accmode & VADMIN_PERMS) {
4634 		*accmode &= ~VADMIN_PERMS;
4635 		*accmode |= VADMIN;
4636 	}
4637 
4638 	/*
4639 	 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4640 	 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4641 	 */
4642 	*accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
4643 
4644 	return (0);
4645 }
4646 
4647 /*
4648  * These are helper functions for filesystems to traverse all
4649  * their vnodes.  See MNT_VNODE_FOREACH_ALL() in sys/mount.h.
4650  *
4651  * This interface replaces MNT_VNODE_FOREACH.
4652  */
4653 
4654 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
4655 
4656 struct vnode *
4657 __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp)
4658 {
4659 	struct vnode *vp;
4660 
4661 	if (should_yield())
4662 		kern_yield(PRI_UNCHANGED);
4663 	MNT_ILOCK(mp);
4664 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4665 	vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
4666 	while (vp != NULL && (vp->v_type == VMARKER ||
4667 	    (vp->v_iflag & VI_DOOMED) != 0))
4668 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
4669 
4670 	/* Check if we are done */
4671 	if (vp == NULL) {
4672 		__mnt_vnode_markerfree_all(mvp, mp);
4673 		/* MNT_IUNLOCK(mp); -- done in above function */
4674 		mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
4675 		return (NULL);
4676 	}
4677 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4678 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4679 	VI_LOCK(vp);
4680 	MNT_IUNLOCK(mp);
4681 	return (vp);
4682 }
4683 
4684 struct vnode *
4685 __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp)
4686 {
4687 	struct vnode *vp;
4688 
4689 	*mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4690 	MNT_ILOCK(mp);
4691 	MNT_REF(mp);
4692 	(*mvp)->v_type = VMARKER;
4693 
4694 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
4695 	while (vp != NULL && (vp->v_type == VMARKER ||
4696 	    (vp->v_iflag & VI_DOOMED) != 0))
4697 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
4698 
4699 	/* Check if we are done */
4700 	if (vp == NULL) {
4701 		MNT_REL(mp);
4702 		MNT_IUNLOCK(mp);
4703 		free(*mvp, M_VNODE_MARKER);
4704 		*mvp = NULL;
4705 		return (NULL);
4706 	}
4707 	(*mvp)->v_mount = mp;
4708 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4709 	VI_LOCK(vp);
4710 	MNT_IUNLOCK(mp);
4711 	return (vp);
4712 }
4713 
4714 
4715 void
4716 __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp)
4717 {
4718 
4719 	if (*mvp == NULL) {
4720 		MNT_IUNLOCK(mp);
4721 		return;
4722 	}
4723 
4724 	mtx_assert(MNT_MTX(mp), MA_OWNED);
4725 
4726 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4727 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4728 	MNT_REL(mp);
4729 	MNT_IUNLOCK(mp);
4730 	free(*mvp, M_VNODE_MARKER);
4731 	*mvp = NULL;
4732 }
4733 
4734 /*
4735  * These are helper functions for filesystems to traverse their
4736  * active vnodes.  See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h
4737  */
4738 struct vnode *
4739 __mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4740 {
4741 	struct vnode *vp, *nvp;
4742 
4743 	if (should_yield())
4744 		kern_yield(PRI_UNCHANGED);
4745 	MNT_ILOCK(mp);
4746 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4747 	vp = TAILQ_NEXT(*mvp, v_actfreelist);
4748 	while (vp != NULL) {
4749 		VI_LOCK(vp);
4750 		if (vp->v_mount == mp && vp->v_type != VMARKER &&
4751 		    (vp->v_iflag & VI_DOOMED) == 0)
4752 			break;
4753 		nvp = TAILQ_NEXT(vp, v_actfreelist);
4754 		VI_UNLOCK(vp);
4755 		vp = nvp;
4756 	}
4757 
4758 	/* Check if we are done */
4759 	if (vp == NULL) {
4760 		__mnt_vnode_markerfree_active(mvp, mp);
4761 		/* MNT_IUNLOCK(mp); -- done in above function */
4762 		mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
4763 		return (NULL);
4764 	}
4765 	mtx_lock(&vnode_free_list_mtx);
4766 	TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4767 	TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
4768 	mtx_unlock(&vnode_free_list_mtx);
4769 	MNT_IUNLOCK(mp);
4770 	return (vp);
4771 }
4772 
4773 struct vnode *
4774 __mnt_vnode_first_active(struct vnode **mvp, struct mount *mp)
4775 {
4776 	struct vnode *vp, *nvp;
4777 
4778 	*mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4779 	MNT_ILOCK(mp);
4780 	MNT_REF(mp);
4781 	(*mvp)->v_type = VMARKER;
4782 
4783 	vp = TAILQ_NEXT(*mvp, v_actfreelist);
4784 	while (vp != NULL) {
4785 		VI_LOCK(vp);
4786 		if (vp->v_mount == mp && vp->v_type != VMARKER &&
4787 		    (vp->v_iflag & VI_DOOMED) == 0)
4788 			break;
4789 		nvp = TAILQ_NEXT(vp, v_actfreelist);
4790 		VI_UNLOCK(vp);
4791 		vp = nvp;
4792 	}
4793 
4794 	/* Check if we are done */
4795 	if (vp == NULL) {
4796 		MNT_REL(mp);
4797 		MNT_IUNLOCK(mp);
4798 		free(*mvp, M_VNODE_MARKER);
4799 		*mvp = NULL;
4800 		return (NULL);
4801 	}
4802 	(*mvp)->v_mount = mp;
4803 	mtx_lock(&vnode_free_list_mtx);
4804 	TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
4805 	mtx_unlock(&vnode_free_list_mtx);
4806 	MNT_IUNLOCK(mp);
4807 	return (vp);
4808 }
4809 
4810 void
4811 __mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4812 {
4813 
4814 	if (*mvp == NULL) {
4815 		MNT_IUNLOCK(mp);
4816 		return;
4817 	}
4818 
4819 	mtx_assert(MNT_MTX(mp), MA_OWNED);
4820 
4821 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4822 	mtx_lock(&vnode_free_list_mtx);
4823 	TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4824 	mtx_unlock(&vnode_free_list_mtx);
4825 	MNT_REL(mp);
4826 	MNT_IUNLOCK(mp);
4827 	free(*mvp, M_VNODE_MARKER);
4828 	*mvp = NULL;
4829 }
4830