xref: /freebsd/sys/kern/vfs_subr.c (revision 3e0efd2ec4fcb4cd68fb8ccf8aea6fc6151c454b)
1 /*-
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)vfs_subr.c	8.31 (Berkeley) 5/26/95
35  */
36 
37 /*
38  * External virtual filesystem routines
39  */
40 
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
43 
44 #include "opt_ddb.h"
45 #include "opt_watchdog.h"
46 
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/bio.h>
50 #include <sys/buf.h>
51 #include <sys/condvar.h>
52 #include <sys/conf.h>
53 #include <sys/dirent.h>
54 #include <sys/event.h>
55 #include <sys/eventhandler.h>
56 #include <sys/extattr.h>
57 #include <sys/file.h>
58 #include <sys/fcntl.h>
59 #include <sys/jail.h>
60 #include <sys/kdb.h>
61 #include <sys/kernel.h>
62 #include <sys/kthread.h>
63 #include <sys/lockf.h>
64 #include <sys/malloc.h>
65 #include <sys/mount.h>
66 #include <sys/namei.h>
67 #include <sys/priv.h>
68 #include <sys/reboot.h>
69 #include <sys/sched.h>
70 #include <sys/sleepqueue.h>
71 #include <sys/stat.h>
72 #include <sys/sysctl.h>
73 #include <sys/syslog.h>
74 #include <sys/vmmeter.h>
75 #include <sys/vnode.h>
76 #include <sys/watchdog.h>
77 
78 #include <machine/stdarg.h>
79 
80 #include <security/mac/mac_framework.h>
81 
82 #include <vm/vm.h>
83 #include <vm/vm_object.h>
84 #include <vm/vm_extern.h>
85 #include <vm/pmap.h>
86 #include <vm/vm_map.h>
87 #include <vm/vm_page.h>
88 #include <vm/vm_kern.h>
89 #include <vm/uma.h>
90 
91 #ifdef DDB
92 #include <ddb/ddb.h>
93 #endif
94 
95 #define	WI_MPSAFEQ	0
96 #define	WI_GIANTQ	1
97 
98 static void	delmntque(struct vnode *vp);
99 static int	flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
100 		    int slpflag, int slptimeo);
101 static void	syncer_shutdown(void *arg, int howto);
102 static int	vtryrecycle(struct vnode *vp);
103 static void	v_incr_usecount(struct vnode *);
104 static void	v_decr_usecount(struct vnode *);
105 static void	v_decr_useonly(struct vnode *);
106 static void	v_upgrade_usecount(struct vnode *);
107 static void	vnlru_free(int);
108 static void	vgonel(struct vnode *);
109 static void	vfs_knllock(void *arg);
110 static void	vfs_knlunlock(void *arg);
111 static void	vfs_knl_assert_locked(void *arg);
112 static void	vfs_knl_assert_unlocked(void *arg);
113 static void	destroy_vpollinfo(struct vpollinfo *vi);
114 
115 /*
116  * Number of vnodes in existence.  Increased whenever getnewvnode()
117  * allocates a new vnode, decreased in vdropl() for VI_DOOMED vnode.
118  */
119 static unsigned long	numvnodes;
120 
121 SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
122     "Number of vnodes in existence");
123 
124 /*
125  * Conversion tables for conversion from vnode types to inode formats
126  * and back.
127  */
128 enum vtype iftovt_tab[16] = {
129 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
130 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
131 };
132 int vttoif_tab[10] = {
133 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
134 	S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
135 };
136 
137 /*
138  * List of vnodes that are ready for recycling.
139  */
140 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
141 
142 /*
143  * Free vnode target.  Free vnodes may simply be files which have been stat'd
144  * but not read.  This is somewhat common, and a small cache of such files
145  * should be kept to avoid recreation costs.
146  */
147 static u_long wantfreevnodes;
148 SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
149 /* Number of vnodes in the free list. */
150 static u_long freevnodes;
151 SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
152     "Number of vnodes in the free list");
153 
154 static int vlru_allow_cache_src;
155 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
156     &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
157 
158 /*
159  * Various variables used for debugging the new implementation of
160  * reassignbuf().
161  * XXX these are probably of (very) limited utility now.
162  */
163 static int reassignbufcalls;
164 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
165     "Number of calls to reassignbuf");
166 
167 /*
168  * Cache for the mount type id assigned to NFS.  This is used for
169  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
170  */
171 int	nfs_mount_type = -1;
172 
173 /* To keep more than one thread at a time from running vfs_getnewfsid */
174 static struct mtx mntid_mtx;
175 
176 /*
177  * Lock for any access to the following:
178  *	vnode_free_list
179  *	numvnodes
180  *	freevnodes
181  */
182 static struct mtx vnode_free_list_mtx;
183 
184 /* Publicly exported FS */
185 struct nfs_public nfs_pub;
186 
187 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
188 static uma_zone_t vnode_zone;
189 static uma_zone_t vnodepoll_zone;
190 
191 /*
192  * The workitem queue.
193  *
194  * It is useful to delay writes of file data and filesystem metadata
195  * for tens of seconds so that quickly created and deleted files need
196  * not waste disk bandwidth being created and removed. To realize this,
197  * we append vnodes to a "workitem" queue. When running with a soft
198  * updates implementation, most pending metadata dependencies should
199  * not wait for more than a few seconds. Thus, mounted on block devices
200  * are delayed only about a half the time that file data is delayed.
201  * Similarly, directory updates are more critical, so are only delayed
202  * about a third the time that file data is delayed. Thus, there are
203  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
204  * one each second (driven off the filesystem syncer process). The
205  * syncer_delayno variable indicates the next queue that is to be processed.
206  * Items that need to be processed soon are placed in this queue:
207  *
208  *	syncer_workitem_pending[syncer_delayno]
209  *
210  * A delay of fifteen seconds is done by placing the request fifteen
211  * entries later in the queue:
212  *
213  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
214  *
215  */
216 static int syncer_delayno;
217 static long syncer_mask;
218 LIST_HEAD(synclist, bufobj);
219 static struct synclist *syncer_workitem_pending[2];
220 /*
221  * The sync_mtx protects:
222  *	bo->bo_synclist
223  *	sync_vnode_count
224  *	syncer_delayno
225  *	syncer_state
226  *	syncer_workitem_pending
227  *	syncer_worklist_len
228  *	rushjob
229  */
230 static struct mtx sync_mtx;
231 static struct cv sync_wakeup;
232 
233 #define SYNCER_MAXDELAY		32
234 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
235 static int syncdelay = 30;		/* max time to delay syncing data */
236 static int filedelay = 30;		/* time to delay syncing files */
237 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
238     "Time to delay syncing files (in seconds)");
239 static int dirdelay = 29;		/* time to delay syncing directories */
240 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
241     "Time to delay syncing directories (in seconds)");
242 static int metadelay = 28;		/* time to delay syncing metadata */
243 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
244     "Time to delay syncing metadata (in seconds)");
245 static int rushjob;		/* number of slots to run ASAP */
246 static int stat_rush_requests;	/* number of times I/O speeded up */
247 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
248     "Number of times I/O speeded up (rush requests)");
249 
250 /*
251  * When shutting down the syncer, run it at four times normal speed.
252  */
253 #define SYNCER_SHUTDOWN_SPEEDUP		4
254 static int sync_vnode_count;
255 static int syncer_worklist_len;
256 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
257     syncer_state;
258 
259 /*
260  * Number of vnodes we want to exist at any one time.  This is mostly used
261  * to size hash tables in vnode-related code.  It is normally not used in
262  * getnewvnode(), as wantfreevnodes is normally nonzero.)
263  *
264  * XXX desiredvnodes is historical cruft and should not exist.
265  */
266 int desiredvnodes;
267 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
268     &desiredvnodes, 0, "Maximum number of vnodes");
269 SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
270     &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
271 static int vnlru_nowhere;
272 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
273     &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
274 
275 /*
276  * Macros to control when a vnode is freed and recycled.  All require
277  * the vnode interlock.
278  */
279 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
280 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
281 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
282 
283 
284 /*
285  * Initialize the vnode management data structures.
286  *
287  * Reevaluate the following cap on the number of vnodes after the physical
288  * memory size exceeds 512GB.  In the limit, as the physical memory size
289  * grows, the ratio of physical pages to vnodes approaches sixteen to one.
290  */
291 #ifndef	MAXVNODES_MAX
292 #define	MAXVNODES_MAX	(512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
293 #endif
294 static void
295 vntblinit(void *dummy __unused)
296 {
297 	int physvnodes, virtvnodes;
298 
299 	/*
300 	 * Desiredvnodes is a function of the physical memory size and the
301 	 * kernel's heap size.  Generally speaking, it scales with the
302 	 * physical memory size.  The ratio of desiredvnodes to physical pages
303 	 * is one to four until desiredvnodes exceeds 98,304.  Thereafter, the
304 	 * marginal ratio of desiredvnodes to physical pages is one to
305 	 * sixteen.  However, desiredvnodes is limited by the kernel's heap
306 	 * size.  The memory required by desiredvnodes vnodes and vm objects
307 	 * may not exceed one seventh of the kernel's heap size.
308 	 */
309 	physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
310 	    cnt.v_page_count) / 16;
311 	virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
312 	    sizeof(struct vnode)));
313 	desiredvnodes = min(physvnodes, virtvnodes);
314 	if (desiredvnodes > MAXVNODES_MAX) {
315 		if (bootverbose)
316 			printf("Reducing kern.maxvnodes %d -> %d\n",
317 			    desiredvnodes, MAXVNODES_MAX);
318 		desiredvnodes = MAXVNODES_MAX;
319 	}
320 	wantfreevnodes = desiredvnodes / 4;
321 	mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
322 	TAILQ_INIT(&vnode_free_list);
323 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
324 	vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
325 	    NULL, NULL, UMA_ALIGN_PTR, 0);
326 	vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
327 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
328 	/*
329 	 * Initialize the filesystem syncer.
330 	 */
331 	syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE,
332 	    &syncer_mask);
333 	syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE,
334 	    &syncer_mask);
335 	syncer_maxdelay = syncer_mask + 1;
336 	mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
337 	cv_init(&sync_wakeup, "syncer");
338 }
339 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
340 
341 
342 /*
343  * Mark a mount point as busy. Used to synchronize access and to delay
344  * unmounting. Eventually, mountlist_mtx is not released on failure.
345  *
346  * vfs_busy() is a custom lock, it can block the caller.
347  * vfs_busy() only sleeps if the unmount is active on the mount point.
348  * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
349  * vnode belonging to mp.
350  *
351  * Lookup uses vfs_busy() to traverse mount points.
352  * root fs			var fs
353  * / vnode lock		A	/ vnode lock (/var)		D
354  * /var vnode lock	B	/log vnode lock(/var/log)	E
355  * vfs_busy lock	C	vfs_busy lock			F
356  *
357  * Within each file system, the lock order is C->A->B and F->D->E.
358  *
359  * When traversing across mounts, the system follows that lock order:
360  *
361  *        C->A->B
362  *              |
363  *              +->F->D->E
364  *
365  * The lookup() process for namei("/var") illustrates the process:
366  *  VOP_LOOKUP() obtains B while A is held
367  *  vfs_busy() obtains a shared lock on F while A and B are held
368  *  vput() releases lock on B
369  *  vput() releases lock on A
370  *  VFS_ROOT() obtains lock on D while shared lock on F is held
371  *  vfs_unbusy() releases shared lock on F
372  *  vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
373  *    Attempt to lock A (instead of vp_crossmp) while D is held would
374  *    violate the global order, causing deadlocks.
375  *
376  * dounmount() locks B while F is drained.
377  */
378 int
379 vfs_busy(struct mount *mp, int flags)
380 {
381 
382 	MPASS((flags & ~MBF_MASK) == 0);
383 	CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
384 
385 	MNT_ILOCK(mp);
386 	MNT_REF(mp);
387 	/*
388 	 * If mount point is currenly being unmounted, sleep until the
389 	 * mount point fate is decided.  If thread doing the unmounting fails,
390 	 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
391 	 * that this mount point has survived the unmount attempt and vfs_busy
392 	 * should retry.  Otherwise the unmounter thread will set MNTK_REFEXPIRE
393 	 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
394 	 * about to be really destroyed.  vfs_busy needs to release its
395 	 * reference on the mount point in this case and return with ENOENT,
396 	 * telling the caller that mount mount it tried to busy is no longer
397 	 * valid.
398 	 */
399 	while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
400 		if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
401 			MNT_REL(mp);
402 			MNT_IUNLOCK(mp);
403 			CTR1(KTR_VFS, "%s: failed busying before sleeping",
404 			    __func__);
405 			return (ENOENT);
406 		}
407 		if (flags & MBF_MNTLSTLOCK)
408 			mtx_unlock(&mountlist_mtx);
409 		mp->mnt_kern_flag |= MNTK_MWAIT;
410 		msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
411 		if (flags & MBF_MNTLSTLOCK)
412 			mtx_lock(&mountlist_mtx);
413 		MNT_ILOCK(mp);
414 	}
415 	if (flags & MBF_MNTLSTLOCK)
416 		mtx_unlock(&mountlist_mtx);
417 	mp->mnt_lockref++;
418 	MNT_IUNLOCK(mp);
419 	return (0);
420 }
421 
422 /*
423  * Free a busy filesystem.
424  */
425 void
426 vfs_unbusy(struct mount *mp)
427 {
428 
429 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
430 	MNT_ILOCK(mp);
431 	MNT_REL(mp);
432 	KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
433 	mp->mnt_lockref--;
434 	if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
435 		MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
436 		CTR1(KTR_VFS, "%s: waking up waiters", __func__);
437 		mp->mnt_kern_flag &= ~MNTK_DRAINING;
438 		wakeup(&mp->mnt_lockref);
439 	}
440 	MNT_IUNLOCK(mp);
441 }
442 
443 /*
444  * Lookup a mount point by filesystem identifier.
445  */
446 struct mount *
447 vfs_getvfs(fsid_t *fsid)
448 {
449 	struct mount *mp;
450 
451 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
452 	mtx_lock(&mountlist_mtx);
453 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
454 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
455 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
456 			vfs_ref(mp);
457 			mtx_unlock(&mountlist_mtx);
458 			return (mp);
459 		}
460 	}
461 	mtx_unlock(&mountlist_mtx);
462 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
463 	return ((struct mount *) 0);
464 }
465 
466 /*
467  * Lookup a mount point by filesystem identifier, busying it before
468  * returning.
469  */
470 struct mount *
471 vfs_busyfs(fsid_t *fsid)
472 {
473 	struct mount *mp;
474 	int error;
475 
476 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
477 	mtx_lock(&mountlist_mtx);
478 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
479 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
480 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
481 			error = vfs_busy(mp, MBF_MNTLSTLOCK);
482 			if (error) {
483 				mtx_unlock(&mountlist_mtx);
484 				return (NULL);
485 			}
486 			return (mp);
487 		}
488 	}
489 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
490 	mtx_unlock(&mountlist_mtx);
491 	return ((struct mount *) 0);
492 }
493 
494 /*
495  * Check if a user can access privileged mount options.
496  */
497 int
498 vfs_suser(struct mount *mp, struct thread *td)
499 {
500 	int error;
501 
502 	/*
503 	 * If the thread is jailed, but this is not a jail-friendly file
504 	 * system, deny immediately.
505 	 */
506 	if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
507 		return (EPERM);
508 
509 	/*
510 	 * If the file system was mounted outside the jail of the calling
511 	 * thread, deny immediately.
512 	 */
513 	if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
514 		return (EPERM);
515 
516 	/*
517 	 * If file system supports delegated administration, we don't check
518 	 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
519 	 * by the file system itself.
520 	 * If this is not the user that did original mount, we check for
521 	 * the PRIV_VFS_MOUNT_OWNER privilege.
522 	 */
523 	if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
524 	    mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
525 		if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
526 			return (error);
527 	}
528 	return (0);
529 }
530 
531 /*
532  * Get a new unique fsid.  Try to make its val[0] unique, since this value
533  * will be used to create fake device numbers for stat().  Also try (but
534  * not so hard) make its val[0] unique mod 2^16, since some emulators only
535  * support 16-bit device numbers.  We end up with unique val[0]'s for the
536  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
537  *
538  * Keep in mind that several mounts may be running in parallel.  Starting
539  * the search one past where the previous search terminated is both a
540  * micro-optimization and a defense against returning the same fsid to
541  * different mounts.
542  */
543 void
544 vfs_getnewfsid(struct mount *mp)
545 {
546 	static uint16_t mntid_base;
547 	struct mount *nmp;
548 	fsid_t tfsid;
549 	int mtype;
550 
551 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
552 	mtx_lock(&mntid_mtx);
553 	mtype = mp->mnt_vfc->vfc_typenum;
554 	tfsid.val[1] = mtype;
555 	mtype = (mtype & 0xFF) << 24;
556 	for (;;) {
557 		tfsid.val[0] = makedev(255,
558 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
559 		mntid_base++;
560 		if ((nmp = vfs_getvfs(&tfsid)) == NULL)
561 			break;
562 		vfs_rel(nmp);
563 	}
564 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
565 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
566 	mtx_unlock(&mntid_mtx);
567 }
568 
569 /*
570  * Knob to control the precision of file timestamps:
571  *
572  *   0 = seconds only; nanoseconds zeroed.
573  *   1 = seconds and nanoseconds, accurate within 1/HZ.
574  *   2 = seconds and nanoseconds, truncated to microseconds.
575  * >=3 = seconds and nanoseconds, maximum precision.
576  */
577 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
578 
579 static int timestamp_precision = TSP_SEC;
580 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
581     &timestamp_precision, 0, "File timestamp precision (0: seconds, "
582     "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
583     "3+: sec + ns (max. precision))");
584 
585 /*
586  * Get a current timestamp.
587  */
588 void
589 vfs_timestamp(struct timespec *tsp)
590 {
591 	struct timeval tv;
592 
593 	switch (timestamp_precision) {
594 	case TSP_SEC:
595 		tsp->tv_sec = time_second;
596 		tsp->tv_nsec = 0;
597 		break;
598 	case TSP_HZ:
599 		getnanotime(tsp);
600 		break;
601 	case TSP_USEC:
602 		microtime(&tv);
603 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
604 		break;
605 	case TSP_NSEC:
606 	default:
607 		nanotime(tsp);
608 		break;
609 	}
610 }
611 
612 /*
613  * Set vnode attributes to VNOVAL
614  */
615 void
616 vattr_null(struct vattr *vap)
617 {
618 
619 	vap->va_type = VNON;
620 	vap->va_size = VNOVAL;
621 	vap->va_bytes = VNOVAL;
622 	vap->va_mode = VNOVAL;
623 	vap->va_nlink = VNOVAL;
624 	vap->va_uid = VNOVAL;
625 	vap->va_gid = VNOVAL;
626 	vap->va_fsid = VNOVAL;
627 	vap->va_fileid = VNOVAL;
628 	vap->va_blocksize = VNOVAL;
629 	vap->va_rdev = VNOVAL;
630 	vap->va_atime.tv_sec = VNOVAL;
631 	vap->va_atime.tv_nsec = VNOVAL;
632 	vap->va_mtime.tv_sec = VNOVAL;
633 	vap->va_mtime.tv_nsec = VNOVAL;
634 	vap->va_ctime.tv_sec = VNOVAL;
635 	vap->va_ctime.tv_nsec = VNOVAL;
636 	vap->va_birthtime.tv_sec = VNOVAL;
637 	vap->va_birthtime.tv_nsec = VNOVAL;
638 	vap->va_flags = VNOVAL;
639 	vap->va_gen = VNOVAL;
640 	vap->va_vaflags = 0;
641 }
642 
643 /*
644  * This routine is called when we have too many vnodes.  It attempts
645  * to free <count> vnodes and will potentially free vnodes that still
646  * have VM backing store (VM backing store is typically the cause
647  * of a vnode blowout so we want to do this).  Therefore, this operation
648  * is not considered cheap.
649  *
650  * A number of conditions may prevent a vnode from being reclaimed.
651  * the buffer cache may have references on the vnode, a directory
652  * vnode may still have references due to the namei cache representing
653  * underlying files, or the vnode may be in active use.   It is not
654  * desireable to reuse such vnodes.  These conditions may cause the
655  * number of vnodes to reach some minimum value regardless of what
656  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
657  */
658 static int
659 vlrureclaim(struct mount *mp)
660 {
661 	struct vnode *vp;
662 	int done;
663 	int trigger;
664 	int usevnodes;
665 	int count;
666 
667 	/*
668 	 * Calculate the trigger point, don't allow user
669 	 * screwups to blow us up.   This prevents us from
670 	 * recycling vnodes with lots of resident pages.  We
671 	 * aren't trying to free memory, we are trying to
672 	 * free vnodes.
673 	 */
674 	usevnodes = desiredvnodes;
675 	if (usevnodes <= 0)
676 		usevnodes = 1;
677 	trigger = cnt.v_page_count * 2 / usevnodes;
678 	done = 0;
679 	vn_start_write(NULL, &mp, V_WAIT);
680 	MNT_ILOCK(mp);
681 	count = mp->mnt_nvnodelistsize / 10 + 1;
682 	while (count != 0) {
683 		vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
684 		while (vp != NULL && vp->v_type == VMARKER)
685 			vp = TAILQ_NEXT(vp, v_nmntvnodes);
686 		if (vp == NULL)
687 			break;
688 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
689 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
690 		--count;
691 		if (!VI_TRYLOCK(vp))
692 			goto next_iter;
693 		/*
694 		 * If it's been deconstructed already, it's still
695 		 * referenced, or it exceeds the trigger, skip it.
696 		 */
697 		if (vp->v_usecount ||
698 		    (!vlru_allow_cache_src &&
699 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
700 		    (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
701 		    vp->v_object->resident_page_count > trigger)) {
702 			VI_UNLOCK(vp);
703 			goto next_iter;
704 		}
705 		MNT_IUNLOCK(mp);
706 		vholdl(vp);
707 		if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
708 			vdrop(vp);
709 			goto next_iter_mntunlocked;
710 		}
711 		VI_LOCK(vp);
712 		/*
713 		 * v_usecount may have been bumped after VOP_LOCK() dropped
714 		 * the vnode interlock and before it was locked again.
715 		 *
716 		 * It is not necessary to recheck VI_DOOMED because it can
717 		 * only be set by another thread that holds both the vnode
718 		 * lock and vnode interlock.  If another thread has the
719 		 * vnode lock before we get to VOP_LOCK() and obtains the
720 		 * vnode interlock after VOP_LOCK() drops the vnode
721 		 * interlock, the other thread will be unable to drop the
722 		 * vnode lock before our VOP_LOCK() call fails.
723 		 */
724 		if (vp->v_usecount ||
725 		    (!vlru_allow_cache_src &&
726 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
727 		    (vp->v_object != NULL &&
728 		    vp->v_object->resident_page_count > trigger)) {
729 			VOP_UNLOCK(vp, LK_INTERLOCK);
730 			goto next_iter_mntunlocked;
731 		}
732 		KASSERT((vp->v_iflag & VI_DOOMED) == 0,
733 		    ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
734 		vgonel(vp);
735 		VOP_UNLOCK(vp, 0);
736 		vdropl(vp);
737 		done++;
738 next_iter_mntunlocked:
739 		if (!should_yield())
740 			goto relock_mnt;
741 		goto yield;
742 next_iter:
743 		if (!should_yield())
744 			continue;
745 		MNT_IUNLOCK(mp);
746 yield:
747 		kern_yield(PRI_UNCHANGED);
748 relock_mnt:
749 		MNT_ILOCK(mp);
750 	}
751 	MNT_IUNLOCK(mp);
752 	vn_finished_write(mp);
753 	return done;
754 }
755 
756 /*
757  * Attempt to keep the free list at wantfreevnodes length.
758  */
759 static void
760 vnlru_free(int count)
761 {
762 	struct vnode *vp;
763 	int vfslocked;
764 
765 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
766 	for (; count > 0; count--) {
767 		vp = TAILQ_FIRST(&vnode_free_list);
768 		/*
769 		 * The list can be modified while the free_list_mtx
770 		 * has been dropped and vp could be NULL here.
771 		 */
772 		if (!vp)
773 			break;
774 		VNASSERT(vp->v_op != NULL, vp,
775 		    ("vnlru_free: vnode already reclaimed."));
776 		KASSERT((vp->v_iflag & VI_FREE) != 0,
777 		    ("Removing vnode not on freelist"));
778 		KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
779 		    ("Mangling active vnode"));
780 		TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
781 		/*
782 		 * Don't recycle if we can't get the interlock.
783 		 */
784 		if (!VI_TRYLOCK(vp)) {
785 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
786 			continue;
787 		}
788 		VNASSERT(VCANRECYCLE(vp), vp,
789 		    ("vp inconsistent on freelist"));
790 		freevnodes--;
791 		vp->v_iflag &= ~VI_FREE;
792 		vholdl(vp);
793 		mtx_unlock(&vnode_free_list_mtx);
794 		VI_UNLOCK(vp);
795 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
796 		vtryrecycle(vp);
797 		VFS_UNLOCK_GIANT(vfslocked);
798 		/*
799 		 * If the recycled succeeded this vdrop will actually free
800 		 * the vnode.  If not it will simply place it back on
801 		 * the free list.
802 		 */
803 		vdrop(vp);
804 		mtx_lock(&vnode_free_list_mtx);
805 	}
806 }
807 /*
808  * Attempt to recycle vnodes in a context that is always safe to block.
809  * Calling vlrurecycle() from the bowels of filesystem code has some
810  * interesting deadlock problems.
811  */
812 static struct proc *vnlruproc;
813 static int vnlruproc_sig;
814 
815 static void
816 vnlru_proc(void)
817 {
818 	struct mount *mp, *nmp;
819 	int done, vfslocked;
820 	struct proc *p = vnlruproc;
821 
822 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
823 	    SHUTDOWN_PRI_FIRST);
824 
825 	for (;;) {
826 		kproc_suspend_check(p);
827 		mtx_lock(&vnode_free_list_mtx);
828 		if (freevnodes > wantfreevnodes)
829 			vnlru_free(freevnodes - wantfreevnodes);
830 		if (numvnodes <= desiredvnodes * 9 / 10) {
831 			vnlruproc_sig = 0;
832 			wakeup(&vnlruproc_sig);
833 			msleep(vnlruproc, &vnode_free_list_mtx,
834 			    PVFS|PDROP, "vlruwt", hz);
835 			continue;
836 		}
837 		mtx_unlock(&vnode_free_list_mtx);
838 		done = 0;
839 		mtx_lock(&mountlist_mtx);
840 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
841 			if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
842 				nmp = TAILQ_NEXT(mp, mnt_list);
843 				continue;
844 			}
845 			vfslocked = VFS_LOCK_GIANT(mp);
846 			done += vlrureclaim(mp);
847 			VFS_UNLOCK_GIANT(vfslocked);
848 			mtx_lock(&mountlist_mtx);
849 			nmp = TAILQ_NEXT(mp, mnt_list);
850 			vfs_unbusy(mp);
851 		}
852 		mtx_unlock(&mountlist_mtx);
853 		if (done == 0) {
854 #if 0
855 			/* These messages are temporary debugging aids */
856 			if (vnlru_nowhere < 5)
857 				printf("vnlru process getting nowhere..\n");
858 			else if (vnlru_nowhere == 5)
859 				printf("vnlru process messages stopped.\n");
860 #endif
861 			vnlru_nowhere++;
862 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
863 		} else
864 			kern_yield(PRI_UNCHANGED);
865 	}
866 }
867 
868 static struct kproc_desc vnlru_kp = {
869 	"vnlru",
870 	vnlru_proc,
871 	&vnlruproc
872 };
873 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
874     &vnlru_kp);
875 
876 /*
877  * Routines having to do with the management of the vnode table.
878  */
879 
880 /*
881  * Try to recycle a freed vnode.  We abort if anyone picks up a reference
882  * before we actually vgone().  This function must be called with the vnode
883  * held to prevent the vnode from being returned to the free list midway
884  * through vgone().
885  */
886 static int
887 vtryrecycle(struct vnode *vp)
888 {
889 	struct mount *vnmp;
890 
891 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
892 	VNASSERT(vp->v_holdcnt, vp,
893 	    ("vtryrecycle: Recycling vp %p without a reference.", vp));
894 	/*
895 	 * This vnode may found and locked via some other list, if so we
896 	 * can't recycle it yet.
897 	 */
898 	if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
899 		CTR2(KTR_VFS,
900 		    "%s: impossible to recycle, vp %p lock is already held",
901 		    __func__, vp);
902 		return (EWOULDBLOCK);
903 	}
904 	/*
905 	 * Don't recycle if its filesystem is being suspended.
906 	 */
907 	if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
908 		VOP_UNLOCK(vp, 0);
909 		CTR2(KTR_VFS,
910 		    "%s: impossible to recycle, cannot start the write for %p",
911 		    __func__, vp);
912 		return (EBUSY);
913 	}
914 	/*
915 	 * If we got this far, we need to acquire the interlock and see if
916 	 * anyone picked up this vnode from another list.  If not, we will
917 	 * mark it with DOOMED via vgonel() so that anyone who does find it
918 	 * will skip over it.
919 	 */
920 	VI_LOCK(vp);
921 	if (vp->v_usecount) {
922 		VOP_UNLOCK(vp, LK_INTERLOCK);
923 		vn_finished_write(vnmp);
924 		CTR2(KTR_VFS,
925 		    "%s: impossible to recycle, %p is already referenced",
926 		    __func__, vp);
927 		return (EBUSY);
928 	}
929 	if ((vp->v_iflag & VI_DOOMED) == 0)
930 		vgonel(vp);
931 	VOP_UNLOCK(vp, LK_INTERLOCK);
932 	vn_finished_write(vnmp);
933 	return (0);
934 }
935 
936 /*
937  * Return the next vnode from the free list.
938  */
939 int
940 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
941     struct vnode **vpp)
942 {
943 	struct vnode *vp = NULL;
944 	struct bufobj *bo;
945 
946 	CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
947 	mtx_lock(&vnode_free_list_mtx);
948 	/*
949 	 * Lend our context to reclaim vnodes if they've exceeded the max.
950 	 */
951 	if (freevnodes > wantfreevnodes)
952 		vnlru_free(1);
953 	/*
954 	 * Wait for available vnodes.
955 	 */
956 	if (numvnodes > desiredvnodes) {
957 		if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) {
958 			/*
959 			 * File system is beeing suspended, we cannot risk a
960 			 * deadlock here, so allocate new vnode anyway.
961 			 */
962 			if (freevnodes > wantfreevnodes)
963 				vnlru_free(freevnodes - wantfreevnodes);
964 			goto alloc;
965 		}
966 		if (vnlruproc_sig == 0) {
967 			vnlruproc_sig = 1;	/* avoid unnecessary wakeups */
968 			wakeup(vnlruproc);
969 		}
970 		msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
971 		    "vlruwk", hz);
972 #if 0	/* XXX Not all VFS_VGET/ffs_vget callers check returns. */
973 		if (numvnodes > desiredvnodes) {
974 			mtx_unlock(&vnode_free_list_mtx);
975 			return (ENFILE);
976 		}
977 #endif
978 	}
979 alloc:
980 	numvnodes++;
981 	mtx_unlock(&vnode_free_list_mtx);
982 	vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
983 	/*
984 	 * Setup locks.
985 	 */
986 	vp->v_vnlock = &vp->v_lock;
987 	mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
988 	/*
989 	 * By default, don't allow shared locks unless filesystems
990 	 * opt-in.
991 	 */
992 	lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
993 	/*
994 	 * Initialize bufobj.
995 	 */
996 	bo = &vp->v_bufobj;
997 	bo->__bo_vnode = vp;
998 	mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
999 	bo->bo_ops = &buf_ops_bio;
1000 	bo->bo_private = vp;
1001 	TAILQ_INIT(&bo->bo_clean.bv_hd);
1002 	TAILQ_INIT(&bo->bo_dirty.bv_hd);
1003 	/*
1004 	 * Initialize namecache.
1005 	 */
1006 	LIST_INIT(&vp->v_cache_src);
1007 	TAILQ_INIT(&vp->v_cache_dst);
1008 	/*
1009 	 * Finalize various vnode identity bits.
1010 	 */
1011 	vp->v_type = VNON;
1012 	vp->v_tag = tag;
1013 	vp->v_op = vops;
1014 	v_incr_usecount(vp);
1015 	vp->v_data = NULL;
1016 #ifdef MAC
1017 	mac_vnode_init(vp);
1018 	if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1019 		mac_vnode_associate_singlelabel(mp, vp);
1020 	else if (mp == NULL && vops != &dead_vnodeops)
1021 		printf("NULL mp in getnewvnode()\n");
1022 #endif
1023 	if (mp != NULL) {
1024 		bo->bo_bsize = mp->mnt_stat.f_iosize;
1025 		if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1026 			vp->v_vflag |= VV_NOKNOTE;
1027 	}
1028 	rangelock_init(&vp->v_rl);
1029 
1030 	*vpp = vp;
1031 	return (0);
1032 }
1033 
1034 /*
1035  * Delete from old mount point vnode list, if on one.
1036  */
1037 static void
1038 delmntque(struct vnode *vp)
1039 {
1040 	struct mount *mp;
1041 	int active;
1042 
1043 	mp = vp->v_mount;
1044 	if (mp == NULL)
1045 		return;
1046 	MNT_ILOCK(mp);
1047 	VI_LOCK(vp);
1048 	KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize,
1049 	    ("Active vnode list size %d > Vnode list size %d",
1050 	     mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize));
1051 	active = vp->v_iflag & VI_ACTIVE;
1052 	vp->v_iflag &= ~VI_ACTIVE;
1053 	if (active) {
1054 		mtx_lock(&vnode_free_list_mtx);
1055 		TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist);
1056 		mp->mnt_activevnodelistsize--;
1057 		mtx_unlock(&vnode_free_list_mtx);
1058 	}
1059 	vp->v_mount = NULL;
1060 	VI_UNLOCK(vp);
1061 	VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1062 		("bad mount point vnode list size"));
1063 	TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1064 	mp->mnt_nvnodelistsize--;
1065 	MNT_REL(mp);
1066 	MNT_IUNLOCK(mp);
1067 }
1068 
1069 static void
1070 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1071 {
1072 
1073 	vp->v_data = NULL;
1074 	vp->v_op = &dead_vnodeops;
1075 	/* XXX non mp-safe fs may still call insmntque with vnode
1076 	   unlocked */
1077 	if (!VOP_ISLOCKED(vp))
1078 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1079 	vgone(vp);
1080 	vput(vp);
1081 }
1082 
1083 /*
1084  * Insert into list of vnodes for the new mount point, if available.
1085  */
1086 int
1087 insmntque1(struct vnode *vp, struct mount *mp,
1088 	void (*dtr)(struct vnode *, void *), void *dtr_arg)
1089 {
1090 	int locked;
1091 
1092 	KASSERT(vp->v_mount == NULL,
1093 		("insmntque: vnode already on per mount vnode list"));
1094 	VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1095 #ifdef DEBUG_VFS_LOCKS
1096 	if (!VFS_NEEDSGIANT(mp))
1097 		ASSERT_VOP_ELOCKED(vp,
1098 		    "insmntque: mp-safe fs and non-locked vp");
1099 #endif
1100 	/*
1101 	 * We acquire the vnode interlock early to ensure that the
1102 	 * vnode cannot be recycled by another process releasing a
1103 	 * holdcnt on it before we get it on both the vnode list
1104 	 * and the active vnode list. The mount mutex protects only
1105 	 * manipulation of the vnode list and the vnode freelist
1106 	 * mutex protects only manipulation of the active vnode list.
1107 	 * Hence the need to hold the vnode interlock throughout.
1108 	 */
1109 	MNT_ILOCK(mp);
1110 	VI_LOCK(vp);
1111 	if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1112 	    ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1113 	     mp->mnt_nvnodelistsize == 0)) {
1114 		locked = VOP_ISLOCKED(vp);
1115 		if (!locked || (locked == LK_EXCLUSIVE &&
1116 		     (vp->v_vflag & VV_FORCEINSMQ) == 0)) {
1117 			VI_UNLOCK(vp);
1118 			MNT_IUNLOCK(mp);
1119 			if (dtr != NULL)
1120 				dtr(vp, dtr_arg);
1121 			return (EBUSY);
1122 		}
1123 	}
1124 	vp->v_mount = mp;
1125 	MNT_REF(mp);
1126 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1127 	VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1128 		("neg mount point vnode list size"));
1129 	mp->mnt_nvnodelistsize++;
1130 	KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
1131 	    ("Activating already active vnode"));
1132 	vp->v_iflag |= VI_ACTIVE;
1133 	mtx_lock(&vnode_free_list_mtx);
1134 	TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
1135 	mp->mnt_activevnodelistsize++;
1136 	mtx_unlock(&vnode_free_list_mtx);
1137 	VI_UNLOCK(vp);
1138 	MNT_IUNLOCK(mp);
1139 	return (0);
1140 }
1141 
1142 int
1143 insmntque(struct vnode *vp, struct mount *mp)
1144 {
1145 
1146 	return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1147 }
1148 
1149 /*
1150  * Flush out and invalidate all buffers associated with a bufobj
1151  * Called with the underlying object locked.
1152  */
1153 int
1154 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1155 {
1156 	int error;
1157 
1158 	BO_LOCK(bo);
1159 	if (flags & V_SAVE) {
1160 		error = bufobj_wwait(bo, slpflag, slptimeo);
1161 		if (error) {
1162 			BO_UNLOCK(bo);
1163 			return (error);
1164 		}
1165 		if (bo->bo_dirty.bv_cnt > 0) {
1166 			BO_UNLOCK(bo);
1167 			if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1168 				return (error);
1169 			/*
1170 			 * XXX We could save a lock/unlock if this was only
1171 			 * enabled under INVARIANTS
1172 			 */
1173 			BO_LOCK(bo);
1174 			if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1175 				panic("vinvalbuf: dirty bufs");
1176 		}
1177 	}
1178 	/*
1179 	 * If you alter this loop please notice that interlock is dropped and
1180 	 * reacquired in flushbuflist.  Special care is needed to ensure that
1181 	 * no race conditions occur from this.
1182 	 */
1183 	do {
1184 		error = flushbuflist(&bo->bo_clean,
1185 		    flags, bo, slpflag, slptimeo);
1186 		if (error == 0 && !(flags & V_CLEANONLY))
1187 			error = flushbuflist(&bo->bo_dirty,
1188 			    flags, bo, slpflag, slptimeo);
1189 		if (error != 0 && error != EAGAIN) {
1190 			BO_UNLOCK(bo);
1191 			return (error);
1192 		}
1193 	} while (error != 0);
1194 
1195 	/*
1196 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1197 	 * have write I/O in-progress but if there is a VM object then the
1198 	 * VM object can also have read-I/O in-progress.
1199 	 */
1200 	do {
1201 		bufobj_wwait(bo, 0, 0);
1202 		BO_UNLOCK(bo);
1203 		if (bo->bo_object != NULL) {
1204 			VM_OBJECT_LOCK(bo->bo_object);
1205 			vm_object_pip_wait(bo->bo_object, "bovlbx");
1206 			VM_OBJECT_UNLOCK(bo->bo_object);
1207 		}
1208 		BO_LOCK(bo);
1209 	} while (bo->bo_numoutput > 0);
1210 	BO_UNLOCK(bo);
1211 
1212 	/*
1213 	 * Destroy the copy in the VM cache, too.
1214 	 */
1215 	if (bo->bo_object != NULL &&
1216 	    (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) {
1217 		VM_OBJECT_LOCK(bo->bo_object);
1218 		vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
1219 		    OBJPR_CLEANONLY : 0);
1220 		VM_OBJECT_UNLOCK(bo->bo_object);
1221 	}
1222 
1223 #ifdef INVARIANTS
1224 	BO_LOCK(bo);
1225 	if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 &&
1226 	    (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1227 		panic("vinvalbuf: flush failed");
1228 	BO_UNLOCK(bo);
1229 #endif
1230 	return (0);
1231 }
1232 
1233 /*
1234  * Flush out and invalidate all buffers associated with a vnode.
1235  * Called with the underlying object locked.
1236  */
1237 int
1238 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1239 {
1240 
1241 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1242 	ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1243 	return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1244 }
1245 
1246 /*
1247  * Flush out buffers on the specified list.
1248  *
1249  */
1250 static int
1251 flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1252     int slptimeo)
1253 {
1254 	struct buf *bp, *nbp;
1255 	int retval, error;
1256 	daddr_t lblkno;
1257 	b_xflags_t xflags;
1258 
1259 	ASSERT_BO_LOCKED(bo);
1260 
1261 	retval = 0;
1262 	TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1263 		if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1264 		    ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1265 			continue;
1266 		}
1267 		lblkno = 0;
1268 		xflags = 0;
1269 		if (nbp != NULL) {
1270 			lblkno = nbp->b_lblkno;
1271 			xflags = nbp->b_xflags &
1272 				(BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1273 		}
1274 		retval = EAGAIN;
1275 		error = BUF_TIMELOCK(bp,
1276 		    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1277 		    "flushbuf", slpflag, slptimeo);
1278 		if (error) {
1279 			BO_LOCK(bo);
1280 			return (error != ENOLCK ? error : EAGAIN);
1281 		}
1282 		KASSERT(bp->b_bufobj == bo,
1283 		    ("bp %p wrong b_bufobj %p should be %p",
1284 		    bp, bp->b_bufobj, bo));
1285 		if (bp->b_bufobj != bo) {	/* XXX: necessary ? */
1286 			BUF_UNLOCK(bp);
1287 			BO_LOCK(bo);
1288 			return (EAGAIN);
1289 		}
1290 		/*
1291 		 * XXX Since there are no node locks for NFS, I
1292 		 * believe there is a slight chance that a delayed
1293 		 * write will occur while sleeping just above, so
1294 		 * check for it.
1295 		 */
1296 		if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1297 		    (flags & V_SAVE)) {
1298 			BO_LOCK(bo);
1299 			bremfree(bp);
1300 			BO_UNLOCK(bo);
1301 			bp->b_flags |= B_ASYNC;
1302 			bwrite(bp);
1303 			BO_LOCK(bo);
1304 			return (EAGAIN);	/* XXX: why not loop ? */
1305 		}
1306 		BO_LOCK(bo);
1307 		bremfree(bp);
1308 		BO_UNLOCK(bo);
1309 		bp->b_flags |= (B_INVAL | B_RELBUF);
1310 		bp->b_flags &= ~B_ASYNC;
1311 		brelse(bp);
1312 		BO_LOCK(bo);
1313 		if (nbp != NULL &&
1314 		    (nbp->b_bufobj != bo ||
1315 		     nbp->b_lblkno != lblkno ||
1316 		     (nbp->b_xflags &
1317 		      (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1318 			break;			/* nbp invalid */
1319 	}
1320 	return (retval);
1321 }
1322 
1323 /*
1324  * Truncate a file's buffer and pages to a specified length.  This
1325  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1326  * sync activity.
1327  */
1328 int
1329 vtruncbuf(struct vnode *vp, struct ucred *cred, off_t length, int blksize)
1330 {
1331 	struct buf *bp, *nbp;
1332 	int anyfreed;
1333 	int trunclbn;
1334 	struct bufobj *bo;
1335 
1336 	CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1337 	    vp, cred, blksize, (uintmax_t)length);
1338 
1339 	/*
1340 	 * Round up to the *next* lbn.
1341 	 */
1342 	trunclbn = (length + blksize - 1) / blksize;
1343 
1344 	ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1345 restart:
1346 	bo = &vp->v_bufobj;
1347 	BO_LOCK(bo);
1348 	anyfreed = 1;
1349 	for (;anyfreed;) {
1350 		anyfreed = 0;
1351 		TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1352 			if (bp->b_lblkno < trunclbn)
1353 				continue;
1354 			if (BUF_LOCK(bp,
1355 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1356 			    BO_MTX(bo)) == ENOLCK)
1357 				goto restart;
1358 
1359 			BO_LOCK(bo);
1360 			bremfree(bp);
1361 			BO_UNLOCK(bo);
1362 			bp->b_flags |= (B_INVAL | B_RELBUF);
1363 			bp->b_flags &= ~B_ASYNC;
1364 			brelse(bp);
1365 			anyfreed = 1;
1366 
1367 			BO_LOCK(bo);
1368 			if (nbp != NULL &&
1369 			    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1370 			    (nbp->b_vp != vp) ||
1371 			    (nbp->b_flags & B_DELWRI))) {
1372 				BO_UNLOCK(bo);
1373 				goto restart;
1374 			}
1375 		}
1376 
1377 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1378 			if (bp->b_lblkno < trunclbn)
1379 				continue;
1380 			if (BUF_LOCK(bp,
1381 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1382 			    BO_MTX(bo)) == ENOLCK)
1383 				goto restart;
1384 			BO_LOCK(bo);
1385 			bremfree(bp);
1386 			BO_UNLOCK(bo);
1387 			bp->b_flags |= (B_INVAL | B_RELBUF);
1388 			bp->b_flags &= ~B_ASYNC;
1389 			brelse(bp);
1390 			anyfreed = 1;
1391 
1392 			BO_LOCK(bo);
1393 			if (nbp != NULL &&
1394 			    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1395 			    (nbp->b_vp != vp) ||
1396 			    (nbp->b_flags & B_DELWRI) == 0)) {
1397 				BO_UNLOCK(bo);
1398 				goto restart;
1399 			}
1400 		}
1401 	}
1402 
1403 	if (length > 0) {
1404 restartsync:
1405 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1406 			if (bp->b_lblkno > 0)
1407 				continue;
1408 			/*
1409 			 * Since we hold the vnode lock this should only
1410 			 * fail if we're racing with the buf daemon.
1411 			 */
1412 			if (BUF_LOCK(bp,
1413 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1414 			    BO_MTX(bo)) == ENOLCK) {
1415 				goto restart;
1416 			}
1417 			VNASSERT((bp->b_flags & B_DELWRI), vp,
1418 			    ("buf(%p) on dirty queue without DELWRI", bp));
1419 
1420 			BO_LOCK(bo);
1421 			bremfree(bp);
1422 			BO_UNLOCK(bo);
1423 			bawrite(bp);
1424 			BO_LOCK(bo);
1425 			goto restartsync;
1426 		}
1427 	}
1428 
1429 	bufobj_wwait(bo, 0, 0);
1430 	BO_UNLOCK(bo);
1431 	vnode_pager_setsize(vp, length);
1432 
1433 	return (0);
1434 }
1435 
1436 /*
1437  * buf_splay() - splay tree core for the clean/dirty list of buffers in
1438  *		 a vnode.
1439  *
1440  *	NOTE: We have to deal with the special case of a background bitmap
1441  *	buffer, a situation where two buffers will have the same logical
1442  *	block offset.  We want (1) only the foreground buffer to be accessed
1443  *	in a lookup and (2) must differentiate between the foreground and
1444  *	background buffer in the splay tree algorithm because the splay
1445  *	tree cannot normally handle multiple entities with the same 'index'.
1446  *	We accomplish this by adding differentiating flags to the splay tree's
1447  *	numerical domain.
1448  */
1449 static
1450 struct buf *
1451 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1452 {
1453 	struct buf dummy;
1454 	struct buf *lefttreemax, *righttreemin, *y;
1455 
1456 	if (root == NULL)
1457 		return (NULL);
1458 	lefttreemax = righttreemin = &dummy;
1459 	for (;;) {
1460 		if (lblkno < root->b_lblkno ||
1461 		    (lblkno == root->b_lblkno &&
1462 		    (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1463 			if ((y = root->b_left) == NULL)
1464 				break;
1465 			if (lblkno < y->b_lblkno) {
1466 				/* Rotate right. */
1467 				root->b_left = y->b_right;
1468 				y->b_right = root;
1469 				root = y;
1470 				if ((y = root->b_left) == NULL)
1471 					break;
1472 			}
1473 			/* Link into the new root's right tree. */
1474 			righttreemin->b_left = root;
1475 			righttreemin = root;
1476 		} else if (lblkno > root->b_lblkno ||
1477 		    (lblkno == root->b_lblkno &&
1478 		    (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1479 			if ((y = root->b_right) == NULL)
1480 				break;
1481 			if (lblkno > y->b_lblkno) {
1482 				/* Rotate left. */
1483 				root->b_right = y->b_left;
1484 				y->b_left = root;
1485 				root = y;
1486 				if ((y = root->b_right) == NULL)
1487 					break;
1488 			}
1489 			/* Link into the new root's left tree. */
1490 			lefttreemax->b_right = root;
1491 			lefttreemax = root;
1492 		} else {
1493 			break;
1494 		}
1495 		root = y;
1496 	}
1497 	/* Assemble the new root. */
1498 	lefttreemax->b_right = root->b_left;
1499 	righttreemin->b_left = root->b_right;
1500 	root->b_left = dummy.b_right;
1501 	root->b_right = dummy.b_left;
1502 	return (root);
1503 }
1504 
1505 static void
1506 buf_vlist_remove(struct buf *bp)
1507 {
1508 	struct buf *root;
1509 	struct bufv *bv;
1510 
1511 	KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1512 	ASSERT_BO_LOCKED(bp->b_bufobj);
1513 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1514 	    (BX_VNDIRTY|BX_VNCLEAN),
1515 	    ("buf_vlist_remove: Buf %p is on two lists", bp));
1516 	if (bp->b_xflags & BX_VNDIRTY)
1517 		bv = &bp->b_bufobj->bo_dirty;
1518 	else
1519 		bv = &bp->b_bufobj->bo_clean;
1520 	if (bp != bv->bv_root) {
1521 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1522 		KASSERT(root == bp, ("splay lookup failed in remove"));
1523 	}
1524 	if (bp->b_left == NULL) {
1525 		root = bp->b_right;
1526 	} else {
1527 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1528 		root->b_right = bp->b_right;
1529 	}
1530 	bv->bv_root = root;
1531 	TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1532 	bv->bv_cnt--;
1533 	bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1534 }
1535 
1536 /*
1537  * Add the buffer to the sorted clean or dirty block list using a
1538  * splay tree algorithm.
1539  *
1540  * NOTE: xflags is passed as a constant, optimizing this inline function!
1541  */
1542 static void
1543 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1544 {
1545 	struct buf *root;
1546 	struct bufv *bv;
1547 
1548 	ASSERT_BO_LOCKED(bo);
1549 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1550 	    ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1551 	bp->b_xflags |= xflags;
1552 	if (xflags & BX_VNDIRTY)
1553 		bv = &bo->bo_dirty;
1554 	else
1555 		bv = &bo->bo_clean;
1556 
1557 	root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1558 	if (root == NULL) {
1559 		bp->b_left = NULL;
1560 		bp->b_right = NULL;
1561 		TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1562 	} else if (bp->b_lblkno < root->b_lblkno ||
1563 	    (bp->b_lblkno == root->b_lblkno &&
1564 	    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1565 		bp->b_left = root->b_left;
1566 		bp->b_right = root;
1567 		root->b_left = NULL;
1568 		TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1569 	} else {
1570 		bp->b_right = root->b_right;
1571 		bp->b_left = root;
1572 		root->b_right = NULL;
1573 		TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1574 	}
1575 	bv->bv_cnt++;
1576 	bv->bv_root = bp;
1577 }
1578 
1579 /*
1580  * Lookup a buffer using the splay tree.  Note that we specifically avoid
1581  * shadow buffers used in background bitmap writes.
1582  *
1583  * This code isn't quite efficient as it could be because we are maintaining
1584  * two sorted lists and do not know which list the block resides in.
1585  *
1586  * During a "make buildworld" the desired buffer is found at one of
1587  * the roots more than 60% of the time.  Thus, checking both roots
1588  * before performing either splay eliminates unnecessary splays on the
1589  * first tree splayed.
1590  */
1591 struct buf *
1592 gbincore(struct bufobj *bo, daddr_t lblkno)
1593 {
1594 	struct buf *bp;
1595 
1596 	ASSERT_BO_LOCKED(bo);
1597 	if ((bp = bo->bo_clean.bv_root) != NULL &&
1598 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1599 		return (bp);
1600 	if ((bp = bo->bo_dirty.bv_root) != NULL &&
1601 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1602 		return (bp);
1603 	if ((bp = bo->bo_clean.bv_root) != NULL) {
1604 		bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1605 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1606 			return (bp);
1607 	}
1608 	if ((bp = bo->bo_dirty.bv_root) != NULL) {
1609 		bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1610 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1611 			return (bp);
1612 	}
1613 	return (NULL);
1614 }
1615 
1616 /*
1617  * Associate a buffer with a vnode.
1618  */
1619 void
1620 bgetvp(struct vnode *vp, struct buf *bp)
1621 {
1622 	struct bufobj *bo;
1623 
1624 	bo = &vp->v_bufobj;
1625 	ASSERT_BO_LOCKED(bo);
1626 	VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1627 
1628 	CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1629 	VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1630 	    ("bgetvp: bp already attached! %p", bp));
1631 
1632 	vhold(vp);
1633 	if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT)
1634 		bp->b_flags |= B_NEEDSGIANT;
1635 	bp->b_vp = vp;
1636 	bp->b_bufobj = bo;
1637 	/*
1638 	 * Insert onto list for new vnode.
1639 	 */
1640 	buf_vlist_add(bp, bo, BX_VNCLEAN);
1641 }
1642 
1643 /*
1644  * Disassociate a buffer from a vnode.
1645  */
1646 void
1647 brelvp(struct buf *bp)
1648 {
1649 	struct bufobj *bo;
1650 	struct vnode *vp;
1651 
1652 	CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1653 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1654 
1655 	/*
1656 	 * Delete from old vnode list, if on one.
1657 	 */
1658 	vp = bp->b_vp;		/* XXX */
1659 	bo = bp->b_bufobj;
1660 	BO_LOCK(bo);
1661 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1662 		buf_vlist_remove(bp);
1663 	else
1664 		panic("brelvp: Buffer %p not on queue.", bp);
1665 	if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1666 		bo->bo_flag &= ~BO_ONWORKLST;
1667 		mtx_lock(&sync_mtx);
1668 		LIST_REMOVE(bo, bo_synclist);
1669 		syncer_worklist_len--;
1670 		mtx_unlock(&sync_mtx);
1671 	}
1672 	bp->b_flags &= ~B_NEEDSGIANT;
1673 	bp->b_vp = NULL;
1674 	bp->b_bufobj = NULL;
1675 	BO_UNLOCK(bo);
1676 	vdrop(vp);
1677 }
1678 
1679 /*
1680  * Add an item to the syncer work queue.
1681  */
1682 static void
1683 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1684 {
1685 	int queue, slot;
1686 
1687 	ASSERT_BO_LOCKED(bo);
1688 
1689 	mtx_lock(&sync_mtx);
1690 	if (bo->bo_flag & BO_ONWORKLST)
1691 		LIST_REMOVE(bo, bo_synclist);
1692 	else {
1693 		bo->bo_flag |= BO_ONWORKLST;
1694 		syncer_worklist_len++;
1695 	}
1696 
1697 	if (delay > syncer_maxdelay - 2)
1698 		delay = syncer_maxdelay - 2;
1699 	slot = (syncer_delayno + delay) & syncer_mask;
1700 
1701 	queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ :
1702 	    WI_MPSAFEQ;
1703 	LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo,
1704 	    bo_synclist);
1705 	mtx_unlock(&sync_mtx);
1706 }
1707 
1708 static int
1709 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1710 {
1711 	int error, len;
1712 
1713 	mtx_lock(&sync_mtx);
1714 	len = syncer_worklist_len - sync_vnode_count;
1715 	mtx_unlock(&sync_mtx);
1716 	error = SYSCTL_OUT(req, &len, sizeof(len));
1717 	return (error);
1718 }
1719 
1720 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1721     sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1722 
1723 static struct proc *updateproc;
1724 static void sched_sync(void);
1725 static struct kproc_desc up_kp = {
1726 	"syncer",
1727 	sched_sync,
1728 	&updateproc
1729 };
1730 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1731 
1732 static int
1733 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1734 {
1735 	struct vnode *vp;
1736 	struct mount *mp;
1737 
1738 	*bo = LIST_FIRST(slp);
1739 	if (*bo == NULL)
1740 		return (0);
1741 	vp = (*bo)->__bo_vnode;	/* XXX */
1742 	if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1743 		return (1);
1744 	/*
1745 	 * We use vhold in case the vnode does not
1746 	 * successfully sync.  vhold prevents the vnode from
1747 	 * going away when we unlock the sync_mtx so that
1748 	 * we can acquire the vnode interlock.
1749 	 */
1750 	vholdl(vp);
1751 	mtx_unlock(&sync_mtx);
1752 	VI_UNLOCK(vp);
1753 	if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1754 		vdrop(vp);
1755 		mtx_lock(&sync_mtx);
1756 		return (*bo == LIST_FIRST(slp));
1757 	}
1758 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1759 	(void) VOP_FSYNC(vp, MNT_LAZY, td);
1760 	VOP_UNLOCK(vp, 0);
1761 	vn_finished_write(mp);
1762 	BO_LOCK(*bo);
1763 	if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1764 		/*
1765 		 * Put us back on the worklist.  The worklist
1766 		 * routine will remove us from our current
1767 		 * position and then add us back in at a later
1768 		 * position.
1769 		 */
1770 		vn_syncer_add_to_worklist(*bo, syncdelay);
1771 	}
1772 	BO_UNLOCK(*bo);
1773 	vdrop(vp);
1774 	mtx_lock(&sync_mtx);
1775 	return (0);
1776 }
1777 
1778 /*
1779  * System filesystem synchronizer daemon.
1780  */
1781 static void
1782 sched_sync(void)
1783 {
1784 	struct synclist *gnext, *next;
1785 	struct synclist *gslp, *slp;
1786 	struct bufobj *bo;
1787 	long starttime;
1788 	struct thread *td = curthread;
1789 	int last_work_seen;
1790 	int net_worklist_len;
1791 	int syncer_final_iter;
1792 	int first_printf;
1793 	int error;
1794 
1795 	last_work_seen = 0;
1796 	syncer_final_iter = 0;
1797 	first_printf = 1;
1798 	syncer_state = SYNCER_RUNNING;
1799 	starttime = time_uptime;
1800 	td->td_pflags |= TDP_NORUNNINGBUF;
1801 
1802 	EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1803 	    SHUTDOWN_PRI_LAST);
1804 
1805 	mtx_lock(&sync_mtx);
1806 	for (;;) {
1807 		if (syncer_state == SYNCER_FINAL_DELAY &&
1808 		    syncer_final_iter == 0) {
1809 			mtx_unlock(&sync_mtx);
1810 			kproc_suspend_check(td->td_proc);
1811 			mtx_lock(&sync_mtx);
1812 		}
1813 		net_worklist_len = syncer_worklist_len - sync_vnode_count;
1814 		if (syncer_state != SYNCER_RUNNING &&
1815 		    starttime != time_uptime) {
1816 			if (first_printf) {
1817 				printf("\nSyncing disks, vnodes remaining...");
1818 				first_printf = 0;
1819 			}
1820 			printf("%d ", net_worklist_len);
1821 		}
1822 		starttime = time_uptime;
1823 
1824 		/*
1825 		 * Push files whose dirty time has expired.  Be careful
1826 		 * of interrupt race on slp queue.
1827 		 *
1828 		 * Skip over empty worklist slots when shutting down.
1829 		 */
1830 		do {
1831 			slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1832 			gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1833 			syncer_delayno += 1;
1834 			if (syncer_delayno == syncer_maxdelay)
1835 				syncer_delayno = 0;
1836 			next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1837 			gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1838 			/*
1839 			 * If the worklist has wrapped since the
1840 			 * it was emptied of all but syncer vnodes,
1841 			 * switch to the FINAL_DELAY state and run
1842 			 * for one more second.
1843 			 */
1844 			if (syncer_state == SYNCER_SHUTTING_DOWN &&
1845 			    net_worklist_len == 0 &&
1846 			    last_work_seen == syncer_delayno) {
1847 				syncer_state = SYNCER_FINAL_DELAY;
1848 				syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1849 			}
1850 		} while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1851 		    LIST_EMPTY(gslp) && syncer_worklist_len > 0);
1852 
1853 		/*
1854 		 * Keep track of the last time there was anything
1855 		 * on the worklist other than syncer vnodes.
1856 		 * Return to the SHUTTING_DOWN state if any
1857 		 * new work appears.
1858 		 */
1859 		if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1860 			last_work_seen = syncer_delayno;
1861 		if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1862 			syncer_state = SYNCER_SHUTTING_DOWN;
1863 		while (!LIST_EMPTY(slp)) {
1864 			error = sync_vnode(slp, &bo, td);
1865 			if (error == 1) {
1866 				LIST_REMOVE(bo, bo_synclist);
1867 				LIST_INSERT_HEAD(next, bo, bo_synclist);
1868 				continue;
1869 			}
1870 
1871 			if (first_printf == 0)
1872 				wdog_kern_pat(WD_LASTVAL);
1873 
1874 		}
1875 		if (!LIST_EMPTY(gslp)) {
1876 			mtx_unlock(&sync_mtx);
1877 			mtx_lock(&Giant);
1878 			mtx_lock(&sync_mtx);
1879 			while (!LIST_EMPTY(gslp)) {
1880 				error = sync_vnode(gslp, &bo, td);
1881 				if (error == 1) {
1882 					LIST_REMOVE(bo, bo_synclist);
1883 					LIST_INSERT_HEAD(gnext, bo,
1884 					    bo_synclist);
1885 					continue;
1886 				}
1887 			}
1888 			mtx_unlock(&Giant);
1889 		}
1890 		if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1891 			syncer_final_iter--;
1892 		/*
1893 		 * The variable rushjob allows the kernel to speed up the
1894 		 * processing of the filesystem syncer process. A rushjob
1895 		 * value of N tells the filesystem syncer to process the next
1896 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1897 		 * is used by the soft update code to speed up the filesystem
1898 		 * syncer process when the incore state is getting so far
1899 		 * ahead of the disk that the kernel memory pool is being
1900 		 * threatened with exhaustion.
1901 		 */
1902 		if (rushjob > 0) {
1903 			rushjob -= 1;
1904 			continue;
1905 		}
1906 		/*
1907 		 * Just sleep for a short period of time between
1908 		 * iterations when shutting down to allow some I/O
1909 		 * to happen.
1910 		 *
1911 		 * If it has taken us less than a second to process the
1912 		 * current work, then wait. Otherwise start right over
1913 		 * again. We can still lose time if any single round
1914 		 * takes more than two seconds, but it does not really
1915 		 * matter as we are just trying to generally pace the
1916 		 * filesystem activity.
1917 		 */
1918 		if (syncer_state != SYNCER_RUNNING ||
1919 		    time_uptime == starttime) {
1920 			thread_lock(td);
1921 			sched_prio(td, PPAUSE);
1922 			thread_unlock(td);
1923 		}
1924 		if (syncer_state != SYNCER_RUNNING)
1925 			cv_timedwait(&sync_wakeup, &sync_mtx,
1926 			    hz / SYNCER_SHUTDOWN_SPEEDUP);
1927 		else if (time_uptime == starttime)
1928 			cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1929 	}
1930 }
1931 
1932 /*
1933  * Request the syncer daemon to speed up its work.
1934  * We never push it to speed up more than half of its
1935  * normal turn time, otherwise it could take over the cpu.
1936  */
1937 int
1938 speedup_syncer(void)
1939 {
1940 	int ret = 0;
1941 
1942 	mtx_lock(&sync_mtx);
1943 	if (rushjob < syncdelay / 2) {
1944 		rushjob += 1;
1945 		stat_rush_requests += 1;
1946 		ret = 1;
1947 	}
1948 	mtx_unlock(&sync_mtx);
1949 	cv_broadcast(&sync_wakeup);
1950 	return (ret);
1951 }
1952 
1953 /*
1954  * Tell the syncer to speed up its work and run though its work
1955  * list several times, then tell it to shut down.
1956  */
1957 static void
1958 syncer_shutdown(void *arg, int howto)
1959 {
1960 
1961 	if (howto & RB_NOSYNC)
1962 		return;
1963 	mtx_lock(&sync_mtx);
1964 	syncer_state = SYNCER_SHUTTING_DOWN;
1965 	rushjob = 0;
1966 	mtx_unlock(&sync_mtx);
1967 	cv_broadcast(&sync_wakeup);
1968 	kproc_shutdown(arg, howto);
1969 }
1970 
1971 /*
1972  * Reassign a buffer from one vnode to another.
1973  * Used to assign file specific control information
1974  * (indirect blocks) to the vnode to which they belong.
1975  */
1976 void
1977 reassignbuf(struct buf *bp)
1978 {
1979 	struct vnode *vp;
1980 	struct bufobj *bo;
1981 	int delay;
1982 #ifdef INVARIANTS
1983 	struct bufv *bv;
1984 #endif
1985 
1986 	vp = bp->b_vp;
1987 	bo = bp->b_bufobj;
1988 	++reassignbufcalls;
1989 
1990 	CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1991 	    bp, bp->b_vp, bp->b_flags);
1992 	/*
1993 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1994 	 * is not fully linked in.
1995 	 */
1996 	if (bp->b_flags & B_PAGING)
1997 		panic("cannot reassign paging buffer");
1998 
1999 	/*
2000 	 * Delete from old vnode list, if on one.
2001 	 */
2002 	BO_LOCK(bo);
2003 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
2004 		buf_vlist_remove(bp);
2005 	else
2006 		panic("reassignbuf: Buffer %p not on queue.", bp);
2007 	/*
2008 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
2009 	 * of clean buffers.
2010 	 */
2011 	if (bp->b_flags & B_DELWRI) {
2012 		if ((bo->bo_flag & BO_ONWORKLST) == 0) {
2013 			switch (vp->v_type) {
2014 			case VDIR:
2015 				delay = dirdelay;
2016 				break;
2017 			case VCHR:
2018 				delay = metadelay;
2019 				break;
2020 			default:
2021 				delay = filedelay;
2022 			}
2023 			vn_syncer_add_to_worklist(bo, delay);
2024 		}
2025 		buf_vlist_add(bp, bo, BX_VNDIRTY);
2026 	} else {
2027 		buf_vlist_add(bp, bo, BX_VNCLEAN);
2028 
2029 		if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
2030 			mtx_lock(&sync_mtx);
2031 			LIST_REMOVE(bo, bo_synclist);
2032 			syncer_worklist_len--;
2033 			mtx_unlock(&sync_mtx);
2034 			bo->bo_flag &= ~BO_ONWORKLST;
2035 		}
2036 	}
2037 #ifdef INVARIANTS
2038 	bv = &bo->bo_clean;
2039 	bp = TAILQ_FIRST(&bv->bv_hd);
2040 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2041 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2042 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
2043 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2044 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2045 	bv = &bo->bo_dirty;
2046 	bp = TAILQ_FIRST(&bv->bv_hd);
2047 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2048 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2049 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
2050 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2051 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2052 #endif
2053 	BO_UNLOCK(bo);
2054 }
2055 
2056 /*
2057  * Increment the use and hold counts on the vnode, taking care to reference
2058  * the driver's usecount if this is a chardev.  The vholdl() will remove
2059  * the vnode from the free list if it is presently free.  Requires the
2060  * vnode interlock and returns with it held.
2061  */
2062 static void
2063 v_incr_usecount(struct vnode *vp)
2064 {
2065 
2066 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2067 	vp->v_usecount++;
2068 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2069 		dev_lock();
2070 		vp->v_rdev->si_usecount++;
2071 		dev_unlock();
2072 	}
2073 	vholdl(vp);
2074 }
2075 
2076 /*
2077  * Turn a holdcnt into a use+holdcnt such that only one call to
2078  * v_decr_usecount is needed.
2079  */
2080 static void
2081 v_upgrade_usecount(struct vnode *vp)
2082 {
2083 
2084 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2085 	vp->v_usecount++;
2086 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2087 		dev_lock();
2088 		vp->v_rdev->si_usecount++;
2089 		dev_unlock();
2090 	}
2091 }
2092 
2093 /*
2094  * Decrement the vnode use and hold count along with the driver's usecount
2095  * if this is a chardev.  The vdropl() below releases the vnode interlock
2096  * as it may free the vnode.
2097  */
2098 static void
2099 v_decr_usecount(struct vnode *vp)
2100 {
2101 
2102 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2103 	VNASSERT(vp->v_usecount > 0, vp,
2104 	    ("v_decr_usecount: negative usecount"));
2105 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2106 	vp->v_usecount--;
2107 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2108 		dev_lock();
2109 		vp->v_rdev->si_usecount--;
2110 		dev_unlock();
2111 	}
2112 	vdropl(vp);
2113 }
2114 
2115 /*
2116  * Decrement only the use count and driver use count.  This is intended to
2117  * be paired with a follow on vdropl() to release the remaining hold count.
2118  * In this way we may vgone() a vnode with a 0 usecount without risk of
2119  * having it end up on a free list because the hold count is kept above 0.
2120  */
2121 static void
2122 v_decr_useonly(struct vnode *vp)
2123 {
2124 
2125 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2126 	VNASSERT(vp->v_usecount > 0, vp,
2127 	    ("v_decr_useonly: negative usecount"));
2128 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2129 	vp->v_usecount--;
2130 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2131 		dev_lock();
2132 		vp->v_rdev->si_usecount--;
2133 		dev_unlock();
2134 	}
2135 }
2136 
2137 /*
2138  * Grab a particular vnode from the free list, increment its
2139  * reference count and lock it.  VI_DOOMED is set if the vnode
2140  * is being destroyed.  Only callers who specify LK_RETRY will
2141  * see doomed vnodes.  If inactive processing was delayed in
2142  * vput try to do it here.
2143  */
2144 int
2145 vget(struct vnode *vp, int flags, struct thread *td)
2146 {
2147 	int error;
2148 
2149 	error = 0;
2150 	VFS_ASSERT_GIANT(vp->v_mount);
2151 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2152 	    ("vget: invalid lock operation"));
2153 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2154 
2155 	if ((flags & LK_INTERLOCK) == 0)
2156 		VI_LOCK(vp);
2157 	vholdl(vp);
2158 	if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2159 		vdrop(vp);
2160 		CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2161 		    vp);
2162 		return (error);
2163 	}
2164 	if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2165 		panic("vget: vn_lock failed to return ENOENT\n");
2166 	VI_LOCK(vp);
2167 	/* Upgrade our holdcnt to a usecount. */
2168 	v_upgrade_usecount(vp);
2169 	/*
2170 	 * We don't guarantee that any particular close will
2171 	 * trigger inactive processing so just make a best effort
2172 	 * here at preventing a reference to a removed file.  If
2173 	 * we don't succeed no harm is done.
2174 	 */
2175 	if (vp->v_iflag & VI_OWEINACT) {
2176 		if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2177 		    (flags & LK_NOWAIT) == 0)
2178 			vinactive(vp, td);
2179 		vp->v_iflag &= ~VI_OWEINACT;
2180 	}
2181 	VI_UNLOCK(vp);
2182 	return (0);
2183 }
2184 
2185 /*
2186  * Increase the reference count of a vnode.
2187  */
2188 void
2189 vref(struct vnode *vp)
2190 {
2191 
2192 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2193 	VI_LOCK(vp);
2194 	v_incr_usecount(vp);
2195 	VI_UNLOCK(vp);
2196 }
2197 
2198 /*
2199  * Return reference count of a vnode.
2200  *
2201  * The results of this call are only guaranteed when some mechanism other
2202  * than the VI lock is used to stop other processes from gaining references
2203  * to the vnode.  This may be the case if the caller holds the only reference.
2204  * This is also useful when stale data is acceptable as race conditions may
2205  * be accounted for by some other means.
2206  */
2207 int
2208 vrefcnt(struct vnode *vp)
2209 {
2210 	int usecnt;
2211 
2212 	VI_LOCK(vp);
2213 	usecnt = vp->v_usecount;
2214 	VI_UNLOCK(vp);
2215 
2216 	return (usecnt);
2217 }
2218 
2219 #define	VPUTX_VRELE	1
2220 #define	VPUTX_VPUT	2
2221 #define	VPUTX_VUNREF	3
2222 
2223 static void
2224 vputx(struct vnode *vp, int func)
2225 {
2226 	int error;
2227 
2228 	KASSERT(vp != NULL, ("vputx: null vp"));
2229 	if (func == VPUTX_VUNREF)
2230 		ASSERT_VOP_LOCKED(vp, "vunref");
2231 	else if (func == VPUTX_VPUT)
2232 		ASSERT_VOP_LOCKED(vp, "vput");
2233 	else
2234 		KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2235 	VFS_ASSERT_GIANT(vp->v_mount);
2236 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2237 	VI_LOCK(vp);
2238 
2239 	/* Skip this v_writecount check if we're going to panic below. */
2240 	VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2241 	    ("vputx: missed vn_close"));
2242 	error = 0;
2243 
2244 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2245 	    vp->v_usecount == 1)) {
2246 		if (func == VPUTX_VPUT)
2247 			VOP_UNLOCK(vp, 0);
2248 		v_decr_usecount(vp);
2249 		return;
2250 	}
2251 
2252 	if (vp->v_usecount != 1) {
2253 		vprint("vputx: negative ref count", vp);
2254 		panic("vputx: negative ref cnt");
2255 	}
2256 	CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2257 	/*
2258 	 * We want to hold the vnode until the inactive finishes to
2259 	 * prevent vgone() races.  We drop the use count here and the
2260 	 * hold count below when we're done.
2261 	 */
2262 	v_decr_useonly(vp);
2263 	/*
2264 	 * We must call VOP_INACTIVE with the node locked. Mark
2265 	 * as VI_DOINGINACT to avoid recursion.
2266 	 */
2267 	vp->v_iflag |= VI_OWEINACT;
2268 	switch (func) {
2269 	case VPUTX_VRELE:
2270 		error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2271 		VI_LOCK(vp);
2272 		break;
2273 	case VPUTX_VPUT:
2274 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2275 			error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
2276 			    LK_NOWAIT);
2277 			VI_LOCK(vp);
2278 		}
2279 		break;
2280 	case VPUTX_VUNREF:
2281 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
2282 			error = EBUSY;
2283 		break;
2284 	}
2285 	if (vp->v_usecount > 0)
2286 		vp->v_iflag &= ~VI_OWEINACT;
2287 	if (error == 0) {
2288 		if (vp->v_iflag & VI_OWEINACT)
2289 			vinactive(vp, curthread);
2290 		if (func != VPUTX_VUNREF)
2291 			VOP_UNLOCK(vp, 0);
2292 	}
2293 	vdropl(vp);
2294 }
2295 
2296 /*
2297  * Vnode put/release.
2298  * If count drops to zero, call inactive routine and return to freelist.
2299  */
2300 void
2301 vrele(struct vnode *vp)
2302 {
2303 
2304 	vputx(vp, VPUTX_VRELE);
2305 }
2306 
2307 /*
2308  * Release an already locked vnode.  This give the same effects as
2309  * unlock+vrele(), but takes less time and avoids releasing and
2310  * re-aquiring the lock (as vrele() acquires the lock internally.)
2311  */
2312 void
2313 vput(struct vnode *vp)
2314 {
2315 
2316 	vputx(vp, VPUTX_VPUT);
2317 }
2318 
2319 /*
2320  * Release an exclusively locked vnode. Do not unlock the vnode lock.
2321  */
2322 void
2323 vunref(struct vnode *vp)
2324 {
2325 
2326 	vputx(vp, VPUTX_VUNREF);
2327 }
2328 
2329 /*
2330  * Somebody doesn't want the vnode recycled.
2331  */
2332 void
2333 vhold(struct vnode *vp)
2334 {
2335 
2336 	VI_LOCK(vp);
2337 	vholdl(vp);
2338 	VI_UNLOCK(vp);
2339 }
2340 
2341 /*
2342  * Increase the hold count and activate if this is the first reference.
2343  */
2344 void
2345 vholdl(struct vnode *vp)
2346 {
2347 	struct mount *mp;
2348 
2349 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2350 	vp->v_holdcnt++;
2351 	if (!VSHOULDBUSY(vp))
2352 		return;
2353 	ASSERT_VI_LOCKED(vp, "vholdl");
2354 	VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
2355 	VNASSERT(vp->v_op != NULL, vp, ("vholdl: vnode already reclaimed."));
2356 	/*
2357 	 * Remove a vnode from the free list, mark it as in use,
2358 	 * and put it on the active list.
2359 	 */
2360 	mtx_lock(&vnode_free_list_mtx);
2361 	TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
2362 	freevnodes--;
2363 	vp->v_iflag &= ~(VI_FREE|VI_AGE);
2364 	KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
2365 	    ("Activating already active vnode"));
2366 	vp->v_iflag |= VI_ACTIVE;
2367 	mp = vp->v_mount;
2368 	TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
2369 	mp->mnt_activevnodelistsize++;
2370 	mtx_unlock(&vnode_free_list_mtx);
2371 }
2372 
2373 /*
2374  * Note that there is one less who cares about this vnode.
2375  * vdrop() is the opposite of vhold().
2376  */
2377 void
2378 vdrop(struct vnode *vp)
2379 {
2380 
2381 	VI_LOCK(vp);
2382 	vdropl(vp);
2383 }
2384 
2385 /*
2386  * Drop the hold count of the vnode.  If this is the last reference to
2387  * the vnode we place it on the free list unless it has been vgone'd
2388  * (marked VI_DOOMED) in which case we will free it.
2389  */
2390 void
2391 vdropl(struct vnode *vp)
2392 {
2393 	struct bufobj *bo;
2394 	struct mount *mp;
2395 	int active;
2396 
2397 	ASSERT_VI_LOCKED(vp, "vdropl");
2398 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2399 	if (vp->v_holdcnt <= 0)
2400 		panic("vdrop: holdcnt %d", vp->v_holdcnt);
2401 	vp->v_holdcnt--;
2402 	if (vp->v_holdcnt > 0) {
2403 		VI_UNLOCK(vp);
2404 		return;
2405 	}
2406 	if ((vp->v_iflag & VI_DOOMED) == 0) {
2407 		/*
2408 		 * Mark a vnode as free: remove it from its active list
2409 		 * and put it up for recycling on the freelist.
2410 		 */
2411 		VNASSERT(vp->v_op != NULL, vp,
2412 		    ("vdropl: vnode already reclaimed."));
2413 		VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2414 		    ("vnode already free"));
2415 		VNASSERT(VSHOULDFREE(vp), vp,
2416 		    ("vdropl: freeing when we shouldn't"));
2417 		active = vp->v_iflag & VI_ACTIVE;
2418 		vp->v_iflag &= ~VI_ACTIVE;
2419 		mp = vp->v_mount;
2420 		mtx_lock(&vnode_free_list_mtx);
2421 		if (active) {
2422 			TAILQ_REMOVE(&mp->mnt_activevnodelist, vp,
2423 			    v_actfreelist);
2424 			mp->mnt_activevnodelistsize--;
2425 		}
2426 		if (vp->v_iflag & VI_AGE) {
2427 			TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_actfreelist);
2428 		} else {
2429 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
2430 		}
2431 		freevnodes++;
2432 		vp->v_iflag &= ~VI_AGE;
2433 		vp->v_iflag |= VI_FREE;
2434 		mtx_unlock(&vnode_free_list_mtx);
2435 		VI_UNLOCK(vp);
2436 		return;
2437 	}
2438 	/*
2439 	 * The vnode has been marked for destruction, so free it.
2440 	 */
2441 	CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp);
2442 	mtx_lock(&vnode_free_list_mtx);
2443 	numvnodes--;
2444 	mtx_unlock(&vnode_free_list_mtx);
2445 	bo = &vp->v_bufobj;
2446 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2447 	    ("cleaned vnode still on the free list."));
2448 	VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
2449 	VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
2450 	VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
2451 	VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
2452 	VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
2453 	VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
2454 	VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
2455 	VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
2456 	VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
2457 	VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
2458 	VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
2459 	VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
2460 	VI_UNLOCK(vp);
2461 #ifdef MAC
2462 	mac_vnode_destroy(vp);
2463 #endif
2464 	if (vp->v_pollinfo != NULL)
2465 		destroy_vpollinfo(vp->v_pollinfo);
2466 #ifdef INVARIANTS
2467 	/* XXX Elsewhere we detect an already freed vnode via NULL v_op. */
2468 	vp->v_op = NULL;
2469 #endif
2470 	rangelock_destroy(&vp->v_rl);
2471 	lockdestroy(vp->v_vnlock);
2472 	mtx_destroy(&vp->v_interlock);
2473 	mtx_destroy(BO_MTX(bo));
2474 	uma_zfree(vnode_zone, vp);
2475 }
2476 
2477 /*
2478  * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2479  * flags.  DOINGINACT prevents us from recursing in calls to vinactive.
2480  * OWEINACT tracks whether a vnode missed a call to inactive due to a
2481  * failed lock upgrade.
2482  */
2483 void
2484 vinactive(struct vnode *vp, struct thread *td)
2485 {
2486 	struct vm_object *obj;
2487 
2488 	ASSERT_VOP_ELOCKED(vp, "vinactive");
2489 	ASSERT_VI_LOCKED(vp, "vinactive");
2490 	VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2491 	    ("vinactive: recursed on VI_DOINGINACT"));
2492 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2493 	vp->v_iflag |= VI_DOINGINACT;
2494 	vp->v_iflag &= ~VI_OWEINACT;
2495 	VI_UNLOCK(vp);
2496 	/*
2497 	 * Before moving off the active list, we must be sure that any
2498 	 * modified pages are on the vnode's dirty list since these will
2499 	 * no longer be checked once the vnode is on the inactive list.
2500 	 */
2501 	obj = vp->v_object;
2502 	if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
2503 		VM_OBJECT_LOCK(obj);
2504 		vm_object_page_clean(obj, 0, 0, OBJPC_NOSYNC);
2505 		VM_OBJECT_UNLOCK(obj);
2506 	}
2507 	VOP_INACTIVE(vp, td);
2508 	VI_LOCK(vp);
2509 	VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2510 	    ("vinactive: lost VI_DOINGINACT"));
2511 	vp->v_iflag &= ~VI_DOINGINACT;
2512 }
2513 
2514 /*
2515  * Remove any vnodes in the vnode table belonging to mount point mp.
2516  *
2517  * If FORCECLOSE is not specified, there should not be any active ones,
2518  * return error if any are found (nb: this is a user error, not a
2519  * system error). If FORCECLOSE is specified, detach any active vnodes
2520  * that are found.
2521  *
2522  * If WRITECLOSE is set, only flush out regular file vnodes open for
2523  * writing.
2524  *
2525  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2526  *
2527  * `rootrefs' specifies the base reference count for the root vnode
2528  * of this filesystem. The root vnode is considered busy if its
2529  * v_usecount exceeds this value. On a successful return, vflush(, td)
2530  * will call vrele() on the root vnode exactly rootrefs times.
2531  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2532  * be zero.
2533  */
2534 #ifdef DIAGNOSTIC
2535 static int busyprt = 0;		/* print out busy vnodes */
2536 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
2537 #endif
2538 
2539 int
2540 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
2541 {
2542 	struct vnode *vp, *mvp, *rootvp = NULL;
2543 	struct vattr vattr;
2544 	int busy = 0, error;
2545 
2546 	CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2547 	    rootrefs, flags);
2548 	if (rootrefs > 0) {
2549 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2550 		    ("vflush: bad args"));
2551 		/*
2552 		 * Get the filesystem root vnode. We can vput() it
2553 		 * immediately, since with rootrefs > 0, it won't go away.
2554 		 */
2555 		if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2556 			CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2557 			    __func__, error);
2558 			return (error);
2559 		}
2560 		vput(rootvp);
2561 	}
2562 loop:
2563 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
2564 		vholdl(vp);
2565 		error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2566 		if (error) {
2567 			vdrop(vp);
2568 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2569 			goto loop;
2570 		}
2571 		/*
2572 		 * Skip over a vnodes marked VV_SYSTEM.
2573 		 */
2574 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2575 			VOP_UNLOCK(vp, 0);
2576 			vdrop(vp);
2577 			continue;
2578 		}
2579 		/*
2580 		 * If WRITECLOSE is set, flush out unlinked but still open
2581 		 * files (even if open only for reading) and regular file
2582 		 * vnodes open for writing.
2583 		 */
2584 		if (flags & WRITECLOSE) {
2585 			if (vp->v_object != NULL) {
2586 				VM_OBJECT_LOCK(vp->v_object);
2587 				vm_object_page_clean(vp->v_object, 0, 0, 0);
2588 				VM_OBJECT_UNLOCK(vp->v_object);
2589 			}
2590 			error = VOP_FSYNC(vp, MNT_WAIT, td);
2591 			if (error != 0) {
2592 				VOP_UNLOCK(vp, 0);
2593 				vdrop(vp);
2594 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2595 				return (error);
2596 			}
2597 			error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2598 			VI_LOCK(vp);
2599 
2600 			if ((vp->v_type == VNON ||
2601 			    (error == 0 && vattr.va_nlink > 0)) &&
2602 			    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2603 				VOP_UNLOCK(vp, 0);
2604 				vdropl(vp);
2605 				continue;
2606 			}
2607 		} else
2608 			VI_LOCK(vp);
2609 		/*
2610 		 * With v_usecount == 0, all we need to do is clear out the
2611 		 * vnode data structures and we are done.
2612 		 *
2613 		 * If FORCECLOSE is set, forcibly close the vnode.
2614 		 */
2615 		if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2616 			VNASSERT(vp->v_usecount == 0 ||
2617 			    (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2618 			    ("device VNODE %p is FORCECLOSED", vp));
2619 			vgonel(vp);
2620 		} else {
2621 			busy++;
2622 #ifdef DIAGNOSTIC
2623 			if (busyprt)
2624 				vprint("vflush: busy vnode", vp);
2625 #endif
2626 		}
2627 		VOP_UNLOCK(vp, 0);
2628 		vdropl(vp);
2629 	}
2630 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2631 		/*
2632 		 * If just the root vnode is busy, and if its refcount
2633 		 * is equal to `rootrefs', then go ahead and kill it.
2634 		 */
2635 		VI_LOCK(rootvp);
2636 		KASSERT(busy > 0, ("vflush: not busy"));
2637 		VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2638 		    ("vflush: usecount %d < rootrefs %d",
2639 		     rootvp->v_usecount, rootrefs));
2640 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2641 			VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2642 			vgone(rootvp);
2643 			VOP_UNLOCK(rootvp, 0);
2644 			busy = 0;
2645 		} else
2646 			VI_UNLOCK(rootvp);
2647 	}
2648 	if (busy) {
2649 		CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2650 		    busy);
2651 		return (EBUSY);
2652 	}
2653 	for (; rootrefs > 0; rootrefs--)
2654 		vrele(rootvp);
2655 	return (0);
2656 }
2657 
2658 /*
2659  * Recycle an unused vnode to the front of the free list.
2660  */
2661 int
2662 vrecycle(struct vnode *vp)
2663 {
2664 	int recycled;
2665 
2666 	ASSERT_VOP_ELOCKED(vp, "vrecycle");
2667 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2668 	recycled = 0;
2669 	VI_LOCK(vp);
2670 	if (vp->v_usecount == 0) {
2671 		recycled = 1;
2672 		vgonel(vp);
2673 	}
2674 	VI_UNLOCK(vp);
2675 	return (recycled);
2676 }
2677 
2678 /*
2679  * Eliminate all activity associated with a vnode
2680  * in preparation for reuse.
2681  */
2682 void
2683 vgone(struct vnode *vp)
2684 {
2685 	VI_LOCK(vp);
2686 	vgonel(vp);
2687 	VI_UNLOCK(vp);
2688 }
2689 
2690 /*
2691  * vgone, with the vp interlock held.
2692  */
2693 void
2694 vgonel(struct vnode *vp)
2695 {
2696 	struct thread *td;
2697 	int oweinact;
2698 	int active;
2699 	struct mount *mp;
2700 
2701 	ASSERT_VOP_ELOCKED(vp, "vgonel");
2702 	ASSERT_VI_LOCKED(vp, "vgonel");
2703 	VNASSERT(vp->v_holdcnt, vp,
2704 	    ("vgonel: vp %p has no reference.", vp));
2705 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2706 	td = curthread;
2707 
2708 	/*
2709 	 * Don't vgonel if we're already doomed.
2710 	 */
2711 	if (vp->v_iflag & VI_DOOMED)
2712 		return;
2713 	vp->v_iflag |= VI_DOOMED;
2714 	/*
2715 	 * Check to see if the vnode is in use.  If so, we have to call
2716 	 * VOP_CLOSE() and VOP_INACTIVE().
2717 	 */
2718 	active = vp->v_usecount;
2719 	oweinact = (vp->v_iflag & VI_OWEINACT);
2720 	VI_UNLOCK(vp);
2721 	/*
2722 	 * Clean out any buffers associated with the vnode.
2723 	 * If the flush fails, just toss the buffers.
2724 	 */
2725 	mp = NULL;
2726 	if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2727 		(void) vn_start_secondary_write(vp, &mp, V_WAIT);
2728 	if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
2729 		vinvalbuf(vp, 0, 0, 0);
2730 
2731 	/*
2732 	 * If purging an active vnode, it must be closed and
2733 	 * deactivated before being reclaimed.
2734 	 */
2735 	if (active)
2736 		VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2737 	if (oweinact || active) {
2738 		VI_LOCK(vp);
2739 		if ((vp->v_iflag & VI_DOINGINACT) == 0)
2740 			vinactive(vp, td);
2741 		VI_UNLOCK(vp);
2742 	}
2743 	if (vp->v_type == VSOCK)
2744 		vfs_unp_reclaim(vp);
2745 	/*
2746 	 * Reclaim the vnode.
2747 	 */
2748 	if (VOP_RECLAIM(vp, td))
2749 		panic("vgone: cannot reclaim");
2750 	if (mp != NULL)
2751 		vn_finished_secondary_write(mp);
2752 	VNASSERT(vp->v_object == NULL, vp,
2753 	    ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2754 	/*
2755 	 * Clear the advisory locks and wake up waiting threads.
2756 	 */
2757 	(void)VOP_ADVLOCKPURGE(vp);
2758 	/*
2759 	 * Delete from old mount point vnode list.
2760 	 */
2761 	delmntque(vp);
2762 	cache_purge(vp);
2763 	/*
2764 	 * Done with purge, reset to the standard lock and invalidate
2765 	 * the vnode.
2766 	 */
2767 	VI_LOCK(vp);
2768 	vp->v_vnlock = &vp->v_lock;
2769 	vp->v_op = &dead_vnodeops;
2770 	vp->v_tag = "none";
2771 	vp->v_type = VBAD;
2772 }
2773 
2774 /*
2775  * Calculate the total number of references to a special device.
2776  */
2777 int
2778 vcount(struct vnode *vp)
2779 {
2780 	int count;
2781 
2782 	dev_lock();
2783 	count = vp->v_rdev->si_usecount;
2784 	dev_unlock();
2785 	return (count);
2786 }
2787 
2788 /*
2789  * Same as above, but using the struct cdev *as argument
2790  */
2791 int
2792 count_dev(struct cdev *dev)
2793 {
2794 	int count;
2795 
2796 	dev_lock();
2797 	count = dev->si_usecount;
2798 	dev_unlock();
2799 	return(count);
2800 }
2801 
2802 /*
2803  * Print out a description of a vnode.
2804  */
2805 static char *typename[] =
2806 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2807  "VMARKER"};
2808 
2809 void
2810 vn_printf(struct vnode *vp, const char *fmt, ...)
2811 {
2812 	va_list ap;
2813 	char buf[256], buf2[16];
2814 	u_long flags;
2815 
2816 	va_start(ap, fmt);
2817 	vprintf(fmt, ap);
2818 	va_end(ap);
2819 	printf("%p: ", (void *)vp);
2820 	printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2821 	printf("    usecount %d, writecount %d, refcount %d mountedhere %p\n",
2822 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2823 	buf[0] = '\0';
2824 	buf[1] = '\0';
2825 	if (vp->v_vflag & VV_ROOT)
2826 		strlcat(buf, "|VV_ROOT", sizeof(buf));
2827 	if (vp->v_vflag & VV_ISTTY)
2828 		strlcat(buf, "|VV_ISTTY", sizeof(buf));
2829 	if (vp->v_vflag & VV_NOSYNC)
2830 		strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2831 	if (vp->v_vflag & VV_CACHEDLABEL)
2832 		strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2833 	if (vp->v_vflag & VV_TEXT)
2834 		strlcat(buf, "|VV_TEXT", sizeof(buf));
2835 	if (vp->v_vflag & VV_COPYONWRITE)
2836 		strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2837 	if (vp->v_vflag & VV_SYSTEM)
2838 		strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2839 	if (vp->v_vflag & VV_PROCDEP)
2840 		strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2841 	if (vp->v_vflag & VV_NOKNOTE)
2842 		strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2843 	if (vp->v_vflag & VV_DELETED)
2844 		strlcat(buf, "|VV_DELETED", sizeof(buf));
2845 	if (vp->v_vflag & VV_MD)
2846 		strlcat(buf, "|VV_MD", sizeof(buf));
2847 	flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC |
2848 	    VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2849 	    VV_NOKNOTE | VV_DELETED | VV_MD);
2850 	if (flags != 0) {
2851 		snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2852 		strlcat(buf, buf2, sizeof(buf));
2853 	}
2854 	if (vp->v_iflag & VI_MOUNT)
2855 		strlcat(buf, "|VI_MOUNT", sizeof(buf));
2856 	if (vp->v_iflag & VI_AGE)
2857 		strlcat(buf, "|VI_AGE", sizeof(buf));
2858 	if (vp->v_iflag & VI_DOOMED)
2859 		strlcat(buf, "|VI_DOOMED", sizeof(buf));
2860 	if (vp->v_iflag & VI_FREE)
2861 		strlcat(buf, "|VI_FREE", sizeof(buf));
2862 	if (vp->v_iflag & VI_DOINGINACT)
2863 		strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2864 	if (vp->v_iflag & VI_OWEINACT)
2865 		strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2866 	flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2867 	    VI_DOINGINACT | VI_OWEINACT);
2868 	if (flags != 0) {
2869 		snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2870 		strlcat(buf, buf2, sizeof(buf));
2871 	}
2872 	printf("    flags (%s)\n", buf + 1);
2873 	if (mtx_owned(VI_MTX(vp)))
2874 		printf(" VI_LOCKed");
2875 	if (vp->v_object != NULL)
2876 		printf("    v_object %p ref %d pages %d\n",
2877 		    vp->v_object, vp->v_object->ref_count,
2878 		    vp->v_object->resident_page_count);
2879 	printf("    ");
2880 	lockmgr_printinfo(vp->v_vnlock);
2881 	if (vp->v_data != NULL)
2882 		VOP_PRINT(vp);
2883 }
2884 
2885 #ifdef DDB
2886 /*
2887  * List all of the locked vnodes in the system.
2888  * Called when debugging the kernel.
2889  */
2890 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2891 {
2892 	struct mount *mp, *nmp;
2893 	struct vnode *vp;
2894 
2895 	/*
2896 	 * Note: because this is DDB, we can't obey the locking semantics
2897 	 * for these structures, which means we could catch an inconsistent
2898 	 * state and dereference a nasty pointer.  Not much to be done
2899 	 * about that.
2900 	 */
2901 	db_printf("Locked vnodes\n");
2902 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2903 		nmp = TAILQ_NEXT(mp, mnt_list);
2904 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2905 			if (vp->v_type != VMARKER &&
2906 			    VOP_ISLOCKED(vp))
2907 				vprint("", vp);
2908 		}
2909 		nmp = TAILQ_NEXT(mp, mnt_list);
2910 	}
2911 }
2912 
2913 /*
2914  * Show details about the given vnode.
2915  */
2916 DB_SHOW_COMMAND(vnode, db_show_vnode)
2917 {
2918 	struct vnode *vp;
2919 
2920 	if (!have_addr)
2921 		return;
2922 	vp = (struct vnode *)addr;
2923 	vn_printf(vp, "vnode ");
2924 }
2925 
2926 /*
2927  * Show details about the given mount point.
2928  */
2929 DB_SHOW_COMMAND(mount, db_show_mount)
2930 {
2931 	struct mount *mp;
2932 	struct vfsopt *opt;
2933 	struct statfs *sp;
2934 	struct vnode *vp;
2935 	char buf[512];
2936 	uint64_t mflags;
2937 	u_int flags;
2938 
2939 	if (!have_addr) {
2940 		/* No address given, print short info about all mount points. */
2941 		TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2942 			db_printf("%p %s on %s (%s)\n", mp,
2943 			    mp->mnt_stat.f_mntfromname,
2944 			    mp->mnt_stat.f_mntonname,
2945 			    mp->mnt_stat.f_fstypename);
2946 			if (db_pager_quit)
2947 				break;
2948 		}
2949 		db_printf("\nMore info: show mount <addr>\n");
2950 		return;
2951 	}
2952 
2953 	mp = (struct mount *)addr;
2954 	db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
2955 	    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
2956 
2957 	buf[0] = '\0';
2958 	mflags = mp->mnt_flag;
2959 #define	MNT_FLAG(flag)	do {						\
2960 	if (mflags & (flag)) {						\
2961 		if (buf[0] != '\0')					\
2962 			strlcat(buf, ", ", sizeof(buf));		\
2963 		strlcat(buf, (#flag) + 4, sizeof(buf));			\
2964 		mflags &= ~(flag);					\
2965 	}								\
2966 } while (0)
2967 	MNT_FLAG(MNT_RDONLY);
2968 	MNT_FLAG(MNT_SYNCHRONOUS);
2969 	MNT_FLAG(MNT_NOEXEC);
2970 	MNT_FLAG(MNT_NOSUID);
2971 	MNT_FLAG(MNT_UNION);
2972 	MNT_FLAG(MNT_ASYNC);
2973 	MNT_FLAG(MNT_SUIDDIR);
2974 	MNT_FLAG(MNT_SOFTDEP);
2975 	MNT_FLAG(MNT_SUJ);
2976 	MNT_FLAG(MNT_NOSYMFOLLOW);
2977 	MNT_FLAG(MNT_GJOURNAL);
2978 	MNT_FLAG(MNT_MULTILABEL);
2979 	MNT_FLAG(MNT_ACLS);
2980 	MNT_FLAG(MNT_NOATIME);
2981 	MNT_FLAG(MNT_NOCLUSTERR);
2982 	MNT_FLAG(MNT_NOCLUSTERW);
2983 	MNT_FLAG(MNT_NFS4ACLS);
2984 	MNT_FLAG(MNT_EXRDONLY);
2985 	MNT_FLAG(MNT_EXPORTED);
2986 	MNT_FLAG(MNT_DEFEXPORTED);
2987 	MNT_FLAG(MNT_EXPORTANON);
2988 	MNT_FLAG(MNT_EXKERB);
2989 	MNT_FLAG(MNT_EXPUBLIC);
2990 	MNT_FLAG(MNT_LOCAL);
2991 	MNT_FLAG(MNT_QUOTA);
2992 	MNT_FLAG(MNT_ROOTFS);
2993 	MNT_FLAG(MNT_USER);
2994 	MNT_FLAG(MNT_IGNORE);
2995 	MNT_FLAG(MNT_UPDATE);
2996 	MNT_FLAG(MNT_DELEXPORT);
2997 	MNT_FLAG(MNT_RELOAD);
2998 	MNT_FLAG(MNT_FORCE);
2999 	MNT_FLAG(MNT_SNAPSHOT);
3000 	MNT_FLAG(MNT_BYFSID);
3001 #undef MNT_FLAG
3002 	if (mflags != 0) {
3003 		if (buf[0] != '\0')
3004 			strlcat(buf, ", ", sizeof(buf));
3005 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3006 		    "0x%016jx", mflags);
3007 	}
3008 	db_printf("    mnt_flag = %s\n", buf);
3009 
3010 	buf[0] = '\0';
3011 	flags = mp->mnt_kern_flag;
3012 #define	MNT_KERN_FLAG(flag)	do {					\
3013 	if (flags & (flag)) {						\
3014 		if (buf[0] != '\0')					\
3015 			strlcat(buf, ", ", sizeof(buf));		\
3016 		strlcat(buf, (#flag) + 5, sizeof(buf));			\
3017 		flags &= ~(flag);					\
3018 	}								\
3019 } while (0)
3020 	MNT_KERN_FLAG(MNTK_UNMOUNTF);
3021 	MNT_KERN_FLAG(MNTK_ASYNC);
3022 	MNT_KERN_FLAG(MNTK_SOFTDEP);
3023 	MNT_KERN_FLAG(MNTK_NOINSMNTQ);
3024 	MNT_KERN_FLAG(MNTK_DRAINING);
3025 	MNT_KERN_FLAG(MNTK_REFEXPIRE);
3026 	MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
3027 	MNT_KERN_FLAG(MNTK_SHARED_WRITES);
3028 	MNT_KERN_FLAG(MNTK_NOASYNC);
3029 	MNT_KERN_FLAG(MNTK_UNMOUNT);
3030 	MNT_KERN_FLAG(MNTK_MWAIT);
3031 	MNT_KERN_FLAG(MNTK_SUSPEND);
3032 	MNT_KERN_FLAG(MNTK_SUSPEND2);
3033 	MNT_KERN_FLAG(MNTK_SUSPENDED);
3034 	MNT_KERN_FLAG(MNTK_MPSAFE);
3035 	MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
3036 	MNT_KERN_FLAG(MNTK_NOKNOTE);
3037 #undef MNT_KERN_FLAG
3038 	if (flags != 0) {
3039 		if (buf[0] != '\0')
3040 			strlcat(buf, ", ", sizeof(buf));
3041 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3042 		    "0x%08x", flags);
3043 	}
3044 	db_printf("    mnt_kern_flag = %s\n", buf);
3045 
3046 	db_printf("    mnt_opt = ");
3047 	opt = TAILQ_FIRST(mp->mnt_opt);
3048 	if (opt != NULL) {
3049 		db_printf("%s", opt->name);
3050 		opt = TAILQ_NEXT(opt, link);
3051 		while (opt != NULL) {
3052 			db_printf(", %s", opt->name);
3053 			opt = TAILQ_NEXT(opt, link);
3054 		}
3055 	}
3056 	db_printf("\n");
3057 
3058 	sp = &mp->mnt_stat;
3059 	db_printf("    mnt_stat = { version=%u type=%u flags=0x%016jx "
3060 	    "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
3061 	    "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
3062 	    "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
3063 	    (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
3064 	    (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
3065 	    (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
3066 	    (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
3067 	    (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
3068 	    (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
3069 	    (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
3070 	    (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
3071 
3072 	db_printf("    mnt_cred = { uid=%u ruid=%u",
3073 	    (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
3074 	if (jailed(mp->mnt_cred))
3075 		db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
3076 	db_printf(" }\n");
3077 	db_printf("    mnt_ref = %d\n", mp->mnt_ref);
3078 	db_printf("    mnt_gen = %d\n", mp->mnt_gen);
3079 	db_printf("    mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
3080 	db_printf("    mnt_activevnodelistsize = %d\n",
3081 	    mp->mnt_activevnodelistsize);
3082 	db_printf("    mnt_writeopcount = %d\n", mp->mnt_writeopcount);
3083 	db_printf("    mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
3084 	db_printf("    mnt_iosize_max = %d\n", mp->mnt_iosize_max);
3085 	db_printf("    mnt_hashseed = %u\n", mp->mnt_hashseed);
3086 	db_printf("    mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
3087 	db_printf("    mnt_secondary_accwrites = %d\n",
3088 	    mp->mnt_secondary_accwrites);
3089 	db_printf("    mnt_gjprovider = %s\n",
3090 	    mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
3091 
3092 	db_printf("\n\nList of active vnodes\n");
3093 	TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) {
3094 		if (vp->v_type != VMARKER) {
3095 			vn_printf(vp, "vnode ");
3096 			if (db_pager_quit)
3097 				break;
3098 		}
3099 	}
3100 	db_printf("\n\nList of inactive vnodes\n");
3101 	TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3102 		if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) {
3103 			vn_printf(vp, "vnode ");
3104 			if (db_pager_quit)
3105 				break;
3106 		}
3107 	}
3108 }
3109 #endif	/* DDB */
3110 
3111 /*
3112  * Fill in a struct xvfsconf based on a struct vfsconf.
3113  */
3114 static void
3115 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
3116 {
3117 
3118 	strcpy(xvfsp->vfc_name, vfsp->vfc_name);
3119 	xvfsp->vfc_typenum = vfsp->vfc_typenum;
3120 	xvfsp->vfc_refcount = vfsp->vfc_refcount;
3121 	xvfsp->vfc_flags = vfsp->vfc_flags;
3122 	/*
3123 	 * These are unused in userland, we keep them
3124 	 * to not break binary compatibility.
3125 	 */
3126 	xvfsp->vfc_vfsops = NULL;
3127 	xvfsp->vfc_next = NULL;
3128 }
3129 
3130 /*
3131  * Top level filesystem related information gathering.
3132  */
3133 static int
3134 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
3135 {
3136 	struct vfsconf *vfsp;
3137 	struct xvfsconf xvfsp;
3138 	int error;
3139 
3140 	error = 0;
3141 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3142 		bzero(&xvfsp, sizeof(xvfsp));
3143 		vfsconf2x(vfsp, &xvfsp);
3144 		error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
3145 		if (error)
3146 			break;
3147 	}
3148 	return (error);
3149 }
3150 
3151 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD,
3152     NULL, 0, sysctl_vfs_conflist,
3153     "S,xvfsconf", "List of all configured filesystems");
3154 
3155 #ifndef BURN_BRIDGES
3156 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
3157 
3158 static int
3159 vfs_sysctl(SYSCTL_HANDLER_ARGS)
3160 {
3161 	int *name = (int *)arg1 - 1;	/* XXX */
3162 	u_int namelen = arg2 + 1;	/* XXX */
3163 	struct vfsconf *vfsp;
3164 	struct xvfsconf xvfsp;
3165 
3166 	log(LOG_WARNING, "userland calling deprecated sysctl, "
3167 	    "please rebuild world\n");
3168 
3169 #if 1 || defined(COMPAT_PRELITE2)
3170 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3171 	if (namelen == 1)
3172 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3173 #endif
3174 
3175 	switch (name[1]) {
3176 	case VFS_MAXTYPENUM:
3177 		if (namelen != 2)
3178 			return (ENOTDIR);
3179 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3180 	case VFS_CONF:
3181 		if (namelen != 3)
3182 			return (ENOTDIR);	/* overloaded */
3183 		TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
3184 			if (vfsp->vfc_typenum == name[2])
3185 				break;
3186 		if (vfsp == NULL)
3187 			return (EOPNOTSUPP);
3188 		bzero(&xvfsp, sizeof(xvfsp));
3189 		vfsconf2x(vfsp, &xvfsp);
3190 		return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3191 	}
3192 	return (EOPNOTSUPP);
3193 }
3194 
3195 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
3196     vfs_sysctl, "Generic filesystem");
3197 
3198 #if 1 || defined(COMPAT_PRELITE2)
3199 
3200 static int
3201 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3202 {
3203 	int error;
3204 	struct vfsconf *vfsp;
3205 	struct ovfsconf ovfs;
3206 
3207 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3208 		bzero(&ovfs, sizeof(ovfs));
3209 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
3210 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
3211 		ovfs.vfc_index = vfsp->vfc_typenum;
3212 		ovfs.vfc_refcount = vfsp->vfc_refcount;
3213 		ovfs.vfc_flags = vfsp->vfc_flags;
3214 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3215 		if (error)
3216 			return error;
3217 	}
3218 	return 0;
3219 }
3220 
3221 #endif /* 1 || COMPAT_PRELITE2 */
3222 #endif /* !BURN_BRIDGES */
3223 
3224 #define KINFO_VNODESLOP		10
3225 #ifdef notyet
3226 /*
3227  * Dump vnode list (via sysctl).
3228  */
3229 /* ARGSUSED */
3230 static int
3231 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3232 {
3233 	struct xvnode *xvn;
3234 	struct mount *mp;
3235 	struct vnode *vp;
3236 	int error, len, n;
3237 
3238 	/*
3239 	 * Stale numvnodes access is not fatal here.
3240 	 */
3241 	req->lock = 0;
3242 	len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3243 	if (!req->oldptr)
3244 		/* Make an estimate */
3245 		return (SYSCTL_OUT(req, 0, len));
3246 
3247 	error = sysctl_wire_old_buffer(req, 0);
3248 	if (error != 0)
3249 		return (error);
3250 	xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3251 	n = 0;
3252 	mtx_lock(&mountlist_mtx);
3253 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3254 		if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3255 			continue;
3256 		MNT_ILOCK(mp);
3257 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3258 			if (n == len)
3259 				break;
3260 			vref(vp);
3261 			xvn[n].xv_size = sizeof *xvn;
3262 			xvn[n].xv_vnode = vp;
3263 			xvn[n].xv_id = 0;	/* XXX compat */
3264 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3265 			XV_COPY(usecount);
3266 			XV_COPY(writecount);
3267 			XV_COPY(holdcnt);
3268 			XV_COPY(mount);
3269 			XV_COPY(numoutput);
3270 			XV_COPY(type);
3271 #undef XV_COPY
3272 			xvn[n].xv_flag = vp->v_vflag;
3273 
3274 			switch (vp->v_type) {
3275 			case VREG:
3276 			case VDIR:
3277 			case VLNK:
3278 				break;
3279 			case VBLK:
3280 			case VCHR:
3281 				if (vp->v_rdev == NULL) {
3282 					vrele(vp);
3283 					continue;
3284 				}
3285 				xvn[n].xv_dev = dev2udev(vp->v_rdev);
3286 				break;
3287 			case VSOCK:
3288 				xvn[n].xv_socket = vp->v_socket;
3289 				break;
3290 			case VFIFO:
3291 				xvn[n].xv_fifo = vp->v_fifoinfo;
3292 				break;
3293 			case VNON:
3294 			case VBAD:
3295 			default:
3296 				/* shouldn't happen? */
3297 				vrele(vp);
3298 				continue;
3299 			}
3300 			vrele(vp);
3301 			++n;
3302 		}
3303 		MNT_IUNLOCK(mp);
3304 		mtx_lock(&mountlist_mtx);
3305 		vfs_unbusy(mp);
3306 		if (n == len)
3307 			break;
3308 	}
3309 	mtx_unlock(&mountlist_mtx);
3310 
3311 	error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3312 	free(xvn, M_TEMP);
3313 	return (error);
3314 }
3315 
3316 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3317     0, 0, sysctl_vnode, "S,xvnode", "");
3318 #endif
3319 
3320 /*
3321  * Unmount all filesystems. The list is traversed in reverse order
3322  * of mounting to avoid dependencies.
3323  */
3324 void
3325 vfs_unmountall(void)
3326 {
3327 	struct mount *mp;
3328 	struct thread *td;
3329 	int error;
3330 
3331 	KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
3332 	CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3333 	td = curthread;
3334 
3335 	/*
3336 	 * Since this only runs when rebooting, it is not interlocked.
3337 	 */
3338 	while(!TAILQ_EMPTY(&mountlist)) {
3339 		mp = TAILQ_LAST(&mountlist, mntlist);
3340 		error = dounmount(mp, MNT_FORCE, td);
3341 		if (error) {
3342 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
3343 			/*
3344 			 * XXX: Due to the way in which we mount the root
3345 			 * file system off of devfs, devfs will generate a
3346 			 * "busy" warning when we try to unmount it before
3347 			 * the root.  Don't print a warning as a result in
3348 			 * order to avoid false positive errors that may
3349 			 * cause needless upset.
3350 			 */
3351 			if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3352 				printf("unmount of %s failed (",
3353 				    mp->mnt_stat.f_mntonname);
3354 				if (error == EBUSY)
3355 					printf("BUSY)\n");
3356 				else
3357 					printf("%d)\n", error);
3358 			}
3359 		} else {
3360 			/* The unmount has removed mp from the mountlist */
3361 		}
3362 	}
3363 }
3364 
3365 /*
3366  * perform msync on all vnodes under a mount point
3367  * the mount point must be locked.
3368  */
3369 void
3370 vfs_msync(struct mount *mp, int flags)
3371 {
3372 	struct vnode *vp, *mvp;
3373 	struct vm_object *obj;
3374 
3375 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3376 	MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
3377 		obj = vp->v_object;
3378 		if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3379 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3380 			if (!vget(vp,
3381 			    LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3382 			    curthread)) {
3383 				if (vp->v_vflag & VV_NOSYNC) {	/* unlinked */
3384 					vput(vp);
3385 					continue;
3386 				}
3387 
3388 				obj = vp->v_object;
3389 				if (obj != NULL) {
3390 					VM_OBJECT_LOCK(obj);
3391 					vm_object_page_clean(obj, 0, 0,
3392 					    flags == MNT_WAIT ?
3393 					    OBJPC_SYNC : OBJPC_NOSYNC);
3394 					VM_OBJECT_UNLOCK(obj);
3395 				}
3396 				vput(vp);
3397 			}
3398 		} else
3399 			VI_UNLOCK(vp);
3400 	}
3401 }
3402 
3403 static void
3404 destroy_vpollinfo(struct vpollinfo *vi)
3405 {
3406 	seldrain(&vi->vpi_selinfo);
3407 	knlist_destroy(&vi->vpi_selinfo.si_note);
3408 	mtx_destroy(&vi->vpi_lock);
3409 	uma_zfree(vnodepoll_zone, vi);
3410 }
3411 
3412 /*
3413  * Initalize per-vnode helper structure to hold poll-related state.
3414  */
3415 void
3416 v_addpollinfo(struct vnode *vp)
3417 {
3418 	struct vpollinfo *vi;
3419 
3420 	if (vp->v_pollinfo != NULL)
3421 		return;
3422 	vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3423 	mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3424 	knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3425 	    vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3426 	VI_LOCK(vp);
3427 	if (vp->v_pollinfo != NULL) {
3428 		VI_UNLOCK(vp);
3429 		destroy_vpollinfo(vi);
3430 		return;
3431 	}
3432 	vp->v_pollinfo = vi;
3433 	VI_UNLOCK(vp);
3434 }
3435 
3436 /*
3437  * Record a process's interest in events which might happen to
3438  * a vnode.  Because poll uses the historic select-style interface
3439  * internally, this routine serves as both the ``check for any
3440  * pending events'' and the ``record my interest in future events''
3441  * functions.  (These are done together, while the lock is held,
3442  * to avoid race conditions.)
3443  */
3444 int
3445 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3446 {
3447 
3448 	v_addpollinfo(vp);
3449 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3450 	if (vp->v_pollinfo->vpi_revents & events) {
3451 		/*
3452 		 * This leaves events we are not interested
3453 		 * in available for the other process which
3454 		 * which presumably had requested them
3455 		 * (otherwise they would never have been
3456 		 * recorded).
3457 		 */
3458 		events &= vp->v_pollinfo->vpi_revents;
3459 		vp->v_pollinfo->vpi_revents &= ~events;
3460 
3461 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
3462 		return (events);
3463 	}
3464 	vp->v_pollinfo->vpi_events |= events;
3465 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3466 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3467 	return (0);
3468 }
3469 
3470 /*
3471  * Routine to create and manage a filesystem syncer vnode.
3472  */
3473 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
3474 static int	sync_fsync(struct  vop_fsync_args *);
3475 static int	sync_inactive(struct  vop_inactive_args *);
3476 static int	sync_reclaim(struct  vop_reclaim_args *);
3477 
3478 static struct vop_vector sync_vnodeops = {
3479 	.vop_bypass =	VOP_EOPNOTSUPP,
3480 	.vop_close =	sync_close,		/* close */
3481 	.vop_fsync =	sync_fsync,		/* fsync */
3482 	.vop_inactive =	sync_inactive,	/* inactive */
3483 	.vop_reclaim =	sync_reclaim,	/* reclaim */
3484 	.vop_lock1 =	vop_stdlock,	/* lock */
3485 	.vop_unlock =	vop_stdunlock,	/* unlock */
3486 	.vop_islocked =	vop_stdislocked,	/* islocked */
3487 };
3488 
3489 /*
3490  * Create a new filesystem syncer vnode for the specified mount point.
3491  */
3492 void
3493 vfs_allocate_syncvnode(struct mount *mp)
3494 {
3495 	struct vnode *vp;
3496 	struct bufobj *bo;
3497 	static long start, incr, next;
3498 	int error;
3499 
3500 	/* Allocate a new vnode */
3501 	error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
3502 	if (error != 0)
3503 		panic("vfs_allocate_syncvnode: getnewvnode() failed");
3504 	vp->v_type = VNON;
3505 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3506 	vp->v_vflag |= VV_FORCEINSMQ;
3507 	error = insmntque(vp, mp);
3508 	if (error != 0)
3509 		panic("vfs_allocate_syncvnode: insmntque() failed");
3510 	vp->v_vflag &= ~VV_FORCEINSMQ;
3511 	VOP_UNLOCK(vp, 0);
3512 	/*
3513 	 * Place the vnode onto the syncer worklist. We attempt to
3514 	 * scatter them about on the list so that they will go off
3515 	 * at evenly distributed times even if all the filesystems
3516 	 * are mounted at once.
3517 	 */
3518 	next += incr;
3519 	if (next == 0 || next > syncer_maxdelay) {
3520 		start /= 2;
3521 		incr /= 2;
3522 		if (start == 0) {
3523 			start = syncer_maxdelay / 2;
3524 			incr = syncer_maxdelay;
3525 		}
3526 		next = start;
3527 	}
3528 	bo = &vp->v_bufobj;
3529 	BO_LOCK(bo);
3530 	vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3531 	/* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3532 	mtx_lock(&sync_mtx);
3533 	sync_vnode_count++;
3534 	if (mp->mnt_syncer == NULL) {
3535 		mp->mnt_syncer = vp;
3536 		vp = NULL;
3537 	}
3538 	mtx_unlock(&sync_mtx);
3539 	BO_UNLOCK(bo);
3540 	if (vp != NULL) {
3541 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3542 		vgone(vp);
3543 		vput(vp);
3544 	}
3545 }
3546 
3547 void
3548 vfs_deallocate_syncvnode(struct mount *mp)
3549 {
3550 	struct vnode *vp;
3551 
3552 	mtx_lock(&sync_mtx);
3553 	vp = mp->mnt_syncer;
3554 	if (vp != NULL)
3555 		mp->mnt_syncer = NULL;
3556 	mtx_unlock(&sync_mtx);
3557 	if (vp != NULL)
3558 		vrele(vp);
3559 }
3560 
3561 /*
3562  * Do a lazy sync of the filesystem.
3563  */
3564 static int
3565 sync_fsync(struct vop_fsync_args *ap)
3566 {
3567 	struct vnode *syncvp = ap->a_vp;
3568 	struct mount *mp = syncvp->v_mount;
3569 	int error, save;
3570 	struct bufobj *bo;
3571 
3572 	/*
3573 	 * We only need to do something if this is a lazy evaluation.
3574 	 */
3575 	if (ap->a_waitfor != MNT_LAZY)
3576 		return (0);
3577 
3578 	/*
3579 	 * Move ourselves to the back of the sync list.
3580 	 */
3581 	bo = &syncvp->v_bufobj;
3582 	BO_LOCK(bo);
3583 	vn_syncer_add_to_worklist(bo, syncdelay);
3584 	BO_UNLOCK(bo);
3585 
3586 	/*
3587 	 * Walk the list of vnodes pushing all that are dirty and
3588 	 * not already on the sync list.
3589 	 */
3590 	mtx_lock(&mountlist_mtx);
3591 	if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) {
3592 		mtx_unlock(&mountlist_mtx);
3593 		return (0);
3594 	}
3595 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3596 		vfs_unbusy(mp);
3597 		return (0);
3598 	}
3599 	save = curthread_pflags_set(TDP_SYNCIO);
3600 	vfs_msync(mp, MNT_NOWAIT);
3601 	error = VFS_SYNC(mp, MNT_LAZY);
3602 	curthread_pflags_restore(save);
3603 	vn_finished_write(mp);
3604 	vfs_unbusy(mp);
3605 	return (error);
3606 }
3607 
3608 /*
3609  * The syncer vnode is no referenced.
3610  */
3611 static int
3612 sync_inactive(struct vop_inactive_args *ap)
3613 {
3614 
3615 	vgone(ap->a_vp);
3616 	return (0);
3617 }
3618 
3619 /*
3620  * The syncer vnode is no longer needed and is being decommissioned.
3621  *
3622  * Modifications to the worklist must be protected by sync_mtx.
3623  */
3624 static int
3625 sync_reclaim(struct vop_reclaim_args *ap)
3626 {
3627 	struct vnode *vp = ap->a_vp;
3628 	struct bufobj *bo;
3629 
3630 	bo = &vp->v_bufobj;
3631 	BO_LOCK(bo);
3632 	mtx_lock(&sync_mtx);
3633 	if (vp->v_mount->mnt_syncer == vp)
3634 		vp->v_mount->mnt_syncer = NULL;
3635 	if (bo->bo_flag & BO_ONWORKLST) {
3636 		LIST_REMOVE(bo, bo_synclist);
3637 		syncer_worklist_len--;
3638 		sync_vnode_count--;
3639 		bo->bo_flag &= ~BO_ONWORKLST;
3640 	}
3641 	mtx_unlock(&sync_mtx);
3642 	BO_UNLOCK(bo);
3643 
3644 	return (0);
3645 }
3646 
3647 /*
3648  * Check if vnode represents a disk device
3649  */
3650 int
3651 vn_isdisk(struct vnode *vp, int *errp)
3652 {
3653 	int error;
3654 
3655 	error = 0;
3656 	dev_lock();
3657 	if (vp->v_type != VCHR)
3658 		error = ENOTBLK;
3659 	else if (vp->v_rdev == NULL)
3660 		error = ENXIO;
3661 	else if (vp->v_rdev->si_devsw == NULL)
3662 		error = ENXIO;
3663 	else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3664 		error = ENOTBLK;
3665 	dev_unlock();
3666 	if (errp != NULL)
3667 		*errp = error;
3668 	return (error == 0);
3669 }
3670 
3671 /*
3672  * Common filesystem object access control check routine.  Accepts a
3673  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3674  * and optional call-by-reference privused argument allowing vaccess()
3675  * to indicate to the caller whether privilege was used to satisfy the
3676  * request (obsoleted).  Returns 0 on success, or an errno on failure.
3677  */
3678 int
3679 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3680     accmode_t accmode, struct ucred *cred, int *privused)
3681 {
3682 	accmode_t dac_granted;
3683 	accmode_t priv_granted;
3684 
3685 	KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3686 	    ("invalid bit in accmode"));
3687 	KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
3688 	    ("VAPPEND without VWRITE"));
3689 
3690 	/*
3691 	 * Look for a normal, non-privileged way to access the file/directory
3692 	 * as requested.  If it exists, go with that.
3693 	 */
3694 
3695 	if (privused != NULL)
3696 		*privused = 0;
3697 
3698 	dac_granted = 0;
3699 
3700 	/* Check the owner. */
3701 	if (cred->cr_uid == file_uid) {
3702 		dac_granted |= VADMIN;
3703 		if (file_mode & S_IXUSR)
3704 			dac_granted |= VEXEC;
3705 		if (file_mode & S_IRUSR)
3706 			dac_granted |= VREAD;
3707 		if (file_mode & S_IWUSR)
3708 			dac_granted |= (VWRITE | VAPPEND);
3709 
3710 		if ((accmode & dac_granted) == accmode)
3711 			return (0);
3712 
3713 		goto privcheck;
3714 	}
3715 
3716 	/* Otherwise, check the groups (first match) */
3717 	if (groupmember(file_gid, cred)) {
3718 		if (file_mode & S_IXGRP)
3719 			dac_granted |= VEXEC;
3720 		if (file_mode & S_IRGRP)
3721 			dac_granted |= VREAD;
3722 		if (file_mode & S_IWGRP)
3723 			dac_granted |= (VWRITE | VAPPEND);
3724 
3725 		if ((accmode & dac_granted) == accmode)
3726 			return (0);
3727 
3728 		goto privcheck;
3729 	}
3730 
3731 	/* Otherwise, check everyone else. */
3732 	if (file_mode & S_IXOTH)
3733 		dac_granted |= VEXEC;
3734 	if (file_mode & S_IROTH)
3735 		dac_granted |= VREAD;
3736 	if (file_mode & S_IWOTH)
3737 		dac_granted |= (VWRITE | VAPPEND);
3738 	if ((accmode & dac_granted) == accmode)
3739 		return (0);
3740 
3741 privcheck:
3742 	/*
3743 	 * Build a privilege mask to determine if the set of privileges
3744 	 * satisfies the requirements when combined with the granted mask
3745 	 * from above.  For each privilege, if the privilege is required,
3746 	 * bitwise or the request type onto the priv_granted mask.
3747 	 */
3748 	priv_granted = 0;
3749 
3750 	if (type == VDIR) {
3751 		/*
3752 		 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3753 		 * requests, instead of PRIV_VFS_EXEC.
3754 		 */
3755 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3756 		    !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3757 			priv_granted |= VEXEC;
3758 	} else {
3759 		/*
3760 		 * Ensure that at least one execute bit is on. Otherwise,
3761 		 * a privileged user will always succeed, and we don't want
3762 		 * this to happen unless the file really is executable.
3763 		 */
3764 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3765 		    (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
3766 		    !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3767 			priv_granted |= VEXEC;
3768 	}
3769 
3770 	if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
3771 	    !priv_check_cred(cred, PRIV_VFS_READ, 0))
3772 		priv_granted |= VREAD;
3773 
3774 	if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3775 	    !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3776 		priv_granted |= (VWRITE | VAPPEND);
3777 
3778 	if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3779 	    !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3780 		priv_granted |= VADMIN;
3781 
3782 	if ((accmode & (priv_granted | dac_granted)) == accmode) {
3783 		/* XXX audit: privilege used */
3784 		if (privused != NULL)
3785 			*privused = 1;
3786 		return (0);
3787 	}
3788 
3789 	return ((accmode & VADMIN) ? EPERM : EACCES);
3790 }
3791 
3792 /*
3793  * Credential check based on process requesting service, and per-attribute
3794  * permissions.
3795  */
3796 int
3797 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3798     struct thread *td, accmode_t accmode)
3799 {
3800 
3801 	/*
3802 	 * Kernel-invoked always succeeds.
3803 	 */
3804 	if (cred == NOCRED)
3805 		return (0);
3806 
3807 	/*
3808 	 * Do not allow privileged processes in jail to directly manipulate
3809 	 * system attributes.
3810 	 */
3811 	switch (attrnamespace) {
3812 	case EXTATTR_NAMESPACE_SYSTEM:
3813 		/* Potentially should be: return (EPERM); */
3814 		return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3815 	case EXTATTR_NAMESPACE_USER:
3816 		return (VOP_ACCESS(vp, accmode, cred, td));
3817 	default:
3818 		return (EPERM);
3819 	}
3820 }
3821 
3822 #ifdef DEBUG_VFS_LOCKS
3823 /*
3824  * This only exists to supress warnings from unlocked specfs accesses.  It is
3825  * no longer ok to have an unlocked VFS.
3826  */
3827 #define	IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL ||		\
3828 	(vp)->v_type == VCHR ||	(vp)->v_type == VBAD)
3829 
3830 int vfs_badlock_ddb = 1;	/* Drop into debugger on violation. */
3831 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
3832     "Drop into debugger on lock violation");
3833 
3834 int vfs_badlock_mutex = 1;	/* Check for interlock across VOPs. */
3835 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
3836     0, "Check for interlock across VOPs");
3837 
3838 int vfs_badlock_print = 1;	/* Print lock violations. */
3839 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
3840     0, "Print lock violations");
3841 
3842 #ifdef KDB
3843 int vfs_badlock_backtrace = 1;	/* Print backtrace at lock violations. */
3844 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
3845     &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
3846 #endif
3847 
3848 static void
3849 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3850 {
3851 
3852 #ifdef KDB
3853 	if (vfs_badlock_backtrace)
3854 		kdb_backtrace();
3855 #endif
3856 	if (vfs_badlock_print)
3857 		printf("%s: %p %s\n", str, (void *)vp, msg);
3858 	if (vfs_badlock_ddb)
3859 		kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3860 }
3861 
3862 void
3863 assert_vi_locked(struct vnode *vp, const char *str)
3864 {
3865 
3866 	if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3867 		vfs_badlock("interlock is not locked but should be", str, vp);
3868 }
3869 
3870 void
3871 assert_vi_unlocked(struct vnode *vp, const char *str)
3872 {
3873 
3874 	if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3875 		vfs_badlock("interlock is locked but should not be", str, vp);
3876 }
3877 
3878 void
3879 assert_vop_locked(struct vnode *vp, const char *str)
3880 {
3881 
3882 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0)
3883 		vfs_badlock("is not locked but should be", str, vp);
3884 }
3885 
3886 void
3887 assert_vop_unlocked(struct vnode *vp, const char *str)
3888 {
3889 
3890 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
3891 		vfs_badlock("is locked but should not be", str, vp);
3892 }
3893 
3894 void
3895 assert_vop_elocked(struct vnode *vp, const char *str)
3896 {
3897 
3898 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
3899 		vfs_badlock("is not exclusive locked but should be", str, vp);
3900 }
3901 
3902 #if 0
3903 void
3904 assert_vop_elocked_other(struct vnode *vp, const char *str)
3905 {
3906 
3907 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
3908 		vfs_badlock("is not exclusive locked by another thread",
3909 		    str, vp);
3910 }
3911 
3912 void
3913 assert_vop_slocked(struct vnode *vp, const char *str)
3914 {
3915 
3916 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
3917 		vfs_badlock("is not locked shared but should be", str, vp);
3918 }
3919 #endif /* 0 */
3920 #endif /* DEBUG_VFS_LOCKS */
3921 
3922 void
3923 vop_rename_fail(struct vop_rename_args *ap)
3924 {
3925 
3926 	if (ap->a_tvp != NULL)
3927 		vput(ap->a_tvp);
3928 	if (ap->a_tdvp == ap->a_tvp)
3929 		vrele(ap->a_tdvp);
3930 	else
3931 		vput(ap->a_tdvp);
3932 	vrele(ap->a_fdvp);
3933 	vrele(ap->a_fvp);
3934 }
3935 
3936 void
3937 vop_rename_pre(void *ap)
3938 {
3939 	struct vop_rename_args *a = ap;
3940 
3941 #ifdef DEBUG_VFS_LOCKS
3942 	if (a->a_tvp)
3943 		ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
3944 	ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
3945 	ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
3946 	ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
3947 
3948 	/* Check the source (from). */
3949 	if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
3950 	    (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
3951 		ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3952 	if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
3953 		ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
3954 
3955 	/* Check the target. */
3956 	if (a->a_tvp)
3957 		ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3958 	ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3959 #endif
3960 	if (a->a_tdvp != a->a_fdvp)
3961 		vhold(a->a_fdvp);
3962 	if (a->a_tvp != a->a_fvp)
3963 		vhold(a->a_fvp);
3964 	vhold(a->a_tdvp);
3965 	if (a->a_tvp)
3966 		vhold(a->a_tvp);
3967 }
3968 
3969 void
3970 vop_strategy_pre(void *ap)
3971 {
3972 #ifdef DEBUG_VFS_LOCKS
3973 	struct vop_strategy_args *a;
3974 	struct buf *bp;
3975 
3976 	a = ap;
3977 	bp = a->a_bp;
3978 
3979 	/*
3980 	 * Cluster ops lock their component buffers but not the IO container.
3981 	 */
3982 	if ((bp->b_flags & B_CLUSTER) != 0)
3983 		return;
3984 
3985 	if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
3986 		if (vfs_badlock_print)
3987 			printf(
3988 			    "VOP_STRATEGY: bp is not locked but should be\n");
3989 		if (vfs_badlock_ddb)
3990 			kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3991 	}
3992 #endif
3993 }
3994 
3995 void
3996 vop_lookup_pre(void *ap)
3997 {
3998 #ifdef DEBUG_VFS_LOCKS
3999 	struct vop_lookup_args *a;
4000 	struct vnode *dvp;
4001 
4002 	a = ap;
4003 	dvp = a->a_dvp;
4004 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
4005 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
4006 #endif
4007 }
4008 
4009 void
4010 vop_lookup_post(void *ap, int rc)
4011 {
4012 #ifdef DEBUG_VFS_LOCKS
4013 	struct vop_lookup_args *a;
4014 	struct vnode *dvp;
4015 	struct vnode *vp;
4016 
4017 	a = ap;
4018 	dvp = a->a_dvp;
4019 	vp = *(a->a_vpp);
4020 
4021 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
4022 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
4023 
4024 	if (!rc)
4025 		ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
4026 #endif
4027 }
4028 
4029 void
4030 vop_lock_pre(void *ap)
4031 {
4032 #ifdef DEBUG_VFS_LOCKS
4033 	struct vop_lock1_args *a = ap;
4034 
4035 	if ((a->a_flags & LK_INTERLOCK) == 0)
4036 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4037 	else
4038 		ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
4039 #endif
4040 }
4041 
4042 void
4043 vop_lock_post(void *ap, int rc)
4044 {
4045 #ifdef DEBUG_VFS_LOCKS
4046 	struct vop_lock1_args *a = ap;
4047 
4048 	ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4049 	if (rc == 0)
4050 		ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
4051 #endif
4052 }
4053 
4054 void
4055 vop_unlock_pre(void *ap)
4056 {
4057 #ifdef DEBUG_VFS_LOCKS
4058 	struct vop_unlock_args *a = ap;
4059 
4060 	if (a->a_flags & LK_INTERLOCK)
4061 		ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
4062 	ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
4063 #endif
4064 }
4065 
4066 void
4067 vop_unlock_post(void *ap, int rc)
4068 {
4069 #ifdef DEBUG_VFS_LOCKS
4070 	struct vop_unlock_args *a = ap;
4071 
4072 	if (a->a_flags & LK_INTERLOCK)
4073 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
4074 #endif
4075 }
4076 
4077 void
4078 vop_create_post(void *ap, int rc)
4079 {
4080 	struct vop_create_args *a = ap;
4081 
4082 	if (!rc)
4083 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4084 }
4085 
4086 void
4087 vop_deleteextattr_post(void *ap, int rc)
4088 {
4089 	struct vop_deleteextattr_args *a = ap;
4090 
4091 	if (!rc)
4092 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4093 }
4094 
4095 void
4096 vop_link_post(void *ap, int rc)
4097 {
4098 	struct vop_link_args *a = ap;
4099 
4100 	if (!rc) {
4101 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
4102 		VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
4103 	}
4104 }
4105 
4106 void
4107 vop_mkdir_post(void *ap, int rc)
4108 {
4109 	struct vop_mkdir_args *a = ap;
4110 
4111 	if (!rc)
4112 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4113 }
4114 
4115 void
4116 vop_mknod_post(void *ap, int rc)
4117 {
4118 	struct vop_mknod_args *a = ap;
4119 
4120 	if (!rc)
4121 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4122 }
4123 
4124 void
4125 vop_remove_post(void *ap, int rc)
4126 {
4127 	struct vop_remove_args *a = ap;
4128 
4129 	if (!rc) {
4130 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4131 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4132 	}
4133 }
4134 
4135 void
4136 vop_rename_post(void *ap, int rc)
4137 {
4138 	struct vop_rename_args *a = ap;
4139 
4140 	if (!rc) {
4141 		VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
4142 		VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
4143 		VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
4144 		if (a->a_tvp)
4145 			VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
4146 	}
4147 	if (a->a_tdvp != a->a_fdvp)
4148 		vdrop(a->a_fdvp);
4149 	if (a->a_tvp != a->a_fvp)
4150 		vdrop(a->a_fvp);
4151 	vdrop(a->a_tdvp);
4152 	if (a->a_tvp)
4153 		vdrop(a->a_tvp);
4154 }
4155 
4156 void
4157 vop_rmdir_post(void *ap, int rc)
4158 {
4159 	struct vop_rmdir_args *a = ap;
4160 
4161 	if (!rc) {
4162 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4163 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4164 	}
4165 }
4166 
4167 void
4168 vop_setattr_post(void *ap, int rc)
4169 {
4170 	struct vop_setattr_args *a = ap;
4171 
4172 	if (!rc)
4173 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4174 }
4175 
4176 void
4177 vop_setextattr_post(void *ap, int rc)
4178 {
4179 	struct vop_setextattr_args *a = ap;
4180 
4181 	if (!rc)
4182 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4183 }
4184 
4185 void
4186 vop_symlink_post(void *ap, int rc)
4187 {
4188 	struct vop_symlink_args *a = ap;
4189 
4190 	if (!rc)
4191 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4192 }
4193 
4194 static struct knlist fs_knlist;
4195 
4196 static void
4197 vfs_event_init(void *arg)
4198 {
4199 	knlist_init_mtx(&fs_knlist, NULL);
4200 }
4201 /* XXX - correct order? */
4202 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4203 
4204 void
4205 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
4206 {
4207 
4208 	KNOTE_UNLOCKED(&fs_knlist, event);
4209 }
4210 
4211 static int	filt_fsattach(struct knote *kn);
4212 static void	filt_fsdetach(struct knote *kn);
4213 static int	filt_fsevent(struct knote *kn, long hint);
4214 
4215 struct filterops fs_filtops = {
4216 	.f_isfd = 0,
4217 	.f_attach = filt_fsattach,
4218 	.f_detach = filt_fsdetach,
4219 	.f_event = filt_fsevent
4220 };
4221 
4222 static int
4223 filt_fsattach(struct knote *kn)
4224 {
4225 
4226 	kn->kn_flags |= EV_CLEAR;
4227 	knlist_add(&fs_knlist, kn, 0);
4228 	return (0);
4229 }
4230 
4231 static void
4232 filt_fsdetach(struct knote *kn)
4233 {
4234 
4235 	knlist_remove(&fs_knlist, kn, 0);
4236 }
4237 
4238 static int
4239 filt_fsevent(struct knote *kn, long hint)
4240 {
4241 
4242 	kn->kn_fflags |= hint;
4243 	return (kn->kn_fflags != 0);
4244 }
4245 
4246 static int
4247 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4248 {
4249 	struct vfsidctl vc;
4250 	int error;
4251 	struct mount *mp;
4252 
4253 	error = SYSCTL_IN(req, &vc, sizeof(vc));
4254 	if (error)
4255 		return (error);
4256 	if (vc.vc_vers != VFS_CTL_VERS1)
4257 		return (EINVAL);
4258 	mp = vfs_getvfs(&vc.vc_fsid);
4259 	if (mp == NULL)
4260 		return (ENOENT);
4261 	/* ensure that a specific sysctl goes to the right filesystem. */
4262 	if (strcmp(vc.vc_fstypename, "*") != 0 &&
4263 	    strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4264 		vfs_rel(mp);
4265 		return (EINVAL);
4266 	}
4267 	VCTLTOREQ(&vc, req);
4268 	error = VFS_SYSCTL(mp, vc.vc_op, req);
4269 	vfs_rel(mp);
4270 	return (error);
4271 }
4272 
4273 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
4274     NULL, 0, sysctl_vfs_ctl, "",
4275     "Sysctl by fsid");
4276 
4277 /*
4278  * Function to initialize a va_filerev field sensibly.
4279  * XXX: Wouldn't a random number make a lot more sense ??
4280  */
4281 u_quad_t
4282 init_va_filerev(void)
4283 {
4284 	struct bintime bt;
4285 
4286 	getbinuptime(&bt);
4287 	return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4288 }
4289 
4290 static int	filt_vfsread(struct knote *kn, long hint);
4291 static int	filt_vfswrite(struct knote *kn, long hint);
4292 static int	filt_vfsvnode(struct knote *kn, long hint);
4293 static void	filt_vfsdetach(struct knote *kn);
4294 static struct filterops vfsread_filtops = {
4295 	.f_isfd = 1,
4296 	.f_detach = filt_vfsdetach,
4297 	.f_event = filt_vfsread
4298 };
4299 static struct filterops vfswrite_filtops = {
4300 	.f_isfd = 1,
4301 	.f_detach = filt_vfsdetach,
4302 	.f_event = filt_vfswrite
4303 };
4304 static struct filterops vfsvnode_filtops = {
4305 	.f_isfd = 1,
4306 	.f_detach = filt_vfsdetach,
4307 	.f_event = filt_vfsvnode
4308 };
4309 
4310 static void
4311 vfs_knllock(void *arg)
4312 {
4313 	struct vnode *vp = arg;
4314 
4315 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4316 }
4317 
4318 static void
4319 vfs_knlunlock(void *arg)
4320 {
4321 	struct vnode *vp = arg;
4322 
4323 	VOP_UNLOCK(vp, 0);
4324 }
4325 
4326 static void
4327 vfs_knl_assert_locked(void *arg)
4328 {
4329 #ifdef DEBUG_VFS_LOCKS
4330 	struct vnode *vp = arg;
4331 
4332 	ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4333 #endif
4334 }
4335 
4336 static void
4337 vfs_knl_assert_unlocked(void *arg)
4338 {
4339 #ifdef DEBUG_VFS_LOCKS
4340 	struct vnode *vp = arg;
4341 
4342 	ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4343 #endif
4344 }
4345 
4346 int
4347 vfs_kqfilter(struct vop_kqfilter_args *ap)
4348 {
4349 	struct vnode *vp = ap->a_vp;
4350 	struct knote *kn = ap->a_kn;
4351 	struct knlist *knl;
4352 
4353 	switch (kn->kn_filter) {
4354 	case EVFILT_READ:
4355 		kn->kn_fop = &vfsread_filtops;
4356 		break;
4357 	case EVFILT_WRITE:
4358 		kn->kn_fop = &vfswrite_filtops;
4359 		break;
4360 	case EVFILT_VNODE:
4361 		kn->kn_fop = &vfsvnode_filtops;
4362 		break;
4363 	default:
4364 		return (EINVAL);
4365 	}
4366 
4367 	kn->kn_hook = (caddr_t)vp;
4368 
4369 	v_addpollinfo(vp);
4370 	if (vp->v_pollinfo == NULL)
4371 		return (ENOMEM);
4372 	knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4373 	knlist_add(knl, kn, 0);
4374 
4375 	return (0);
4376 }
4377 
4378 /*
4379  * Detach knote from vnode
4380  */
4381 static void
4382 filt_vfsdetach(struct knote *kn)
4383 {
4384 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4385 
4386 	KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4387 	knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4388 }
4389 
4390 /*ARGSUSED*/
4391 static int
4392 filt_vfsread(struct knote *kn, long hint)
4393 {
4394 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4395 	struct vattr va;
4396 	int res;
4397 
4398 	/*
4399 	 * filesystem is gone, so set the EOF flag and schedule
4400 	 * the knote for deletion.
4401 	 */
4402 	if (hint == NOTE_REVOKE) {
4403 		VI_LOCK(vp);
4404 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4405 		VI_UNLOCK(vp);
4406 		return (1);
4407 	}
4408 
4409 	if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4410 		return (0);
4411 
4412 	VI_LOCK(vp);
4413 	kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4414 	res = (kn->kn_data != 0);
4415 	VI_UNLOCK(vp);
4416 	return (res);
4417 }
4418 
4419 /*ARGSUSED*/
4420 static int
4421 filt_vfswrite(struct knote *kn, long hint)
4422 {
4423 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4424 
4425 	VI_LOCK(vp);
4426 
4427 	/*
4428 	 * filesystem is gone, so set the EOF flag and schedule
4429 	 * the knote for deletion.
4430 	 */
4431 	if (hint == NOTE_REVOKE)
4432 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4433 
4434 	kn->kn_data = 0;
4435 	VI_UNLOCK(vp);
4436 	return (1);
4437 }
4438 
4439 static int
4440 filt_vfsvnode(struct knote *kn, long hint)
4441 {
4442 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4443 	int res;
4444 
4445 	VI_LOCK(vp);
4446 	if (kn->kn_sfflags & hint)
4447 		kn->kn_fflags |= hint;
4448 	if (hint == NOTE_REVOKE) {
4449 		kn->kn_flags |= EV_EOF;
4450 		VI_UNLOCK(vp);
4451 		return (1);
4452 	}
4453 	res = (kn->kn_fflags != 0);
4454 	VI_UNLOCK(vp);
4455 	return (res);
4456 }
4457 
4458 int
4459 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4460 {
4461 	int error;
4462 
4463 	if (dp->d_reclen > ap->a_uio->uio_resid)
4464 		return (ENAMETOOLONG);
4465 	error = uiomove(dp, dp->d_reclen, ap->a_uio);
4466 	if (error) {
4467 		if (ap->a_ncookies != NULL) {
4468 			if (ap->a_cookies != NULL)
4469 				free(ap->a_cookies, M_TEMP);
4470 			ap->a_cookies = NULL;
4471 			*ap->a_ncookies = 0;
4472 		}
4473 		return (error);
4474 	}
4475 	if (ap->a_ncookies == NULL)
4476 		return (0);
4477 
4478 	KASSERT(ap->a_cookies,
4479 	    ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4480 
4481 	*ap->a_cookies = realloc(*ap->a_cookies,
4482 	    (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4483 	(*ap->a_cookies)[*ap->a_ncookies] = off;
4484 	return (0);
4485 }
4486 
4487 /*
4488  * Mark for update the access time of the file if the filesystem
4489  * supports VOP_MARKATIME.  This functionality is used by execve and
4490  * mmap, so we want to avoid the I/O implied by directly setting
4491  * va_atime for the sake of efficiency.
4492  */
4493 void
4494 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4495 {
4496 	struct mount *mp;
4497 
4498 	mp = vp->v_mount;
4499 	VFS_ASSERT_GIANT(mp);
4500 	ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4501 	if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4502 		(void)VOP_MARKATIME(vp);
4503 }
4504 
4505 /*
4506  * The purpose of this routine is to remove granularity from accmode_t,
4507  * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4508  * VADMIN and VAPPEND.
4509  *
4510  * If it returns 0, the caller is supposed to continue with the usual
4511  * access checks using 'accmode' as modified by this routine.  If it
4512  * returns nonzero value, the caller is supposed to return that value
4513  * as errno.
4514  *
4515  * Note that after this routine runs, accmode may be zero.
4516  */
4517 int
4518 vfs_unixify_accmode(accmode_t *accmode)
4519 {
4520 	/*
4521 	 * There is no way to specify explicit "deny" rule using
4522 	 * file mode or POSIX.1e ACLs.
4523 	 */
4524 	if (*accmode & VEXPLICIT_DENY) {
4525 		*accmode = 0;
4526 		return (0);
4527 	}
4528 
4529 	/*
4530 	 * None of these can be translated into usual access bits.
4531 	 * Also, the common case for NFSv4 ACLs is to not contain
4532 	 * either of these bits. Caller should check for VWRITE
4533 	 * on the containing directory instead.
4534 	 */
4535 	if (*accmode & (VDELETE_CHILD | VDELETE))
4536 		return (EPERM);
4537 
4538 	if (*accmode & VADMIN_PERMS) {
4539 		*accmode &= ~VADMIN_PERMS;
4540 		*accmode |= VADMIN;
4541 	}
4542 
4543 	/*
4544 	 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4545 	 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4546 	 */
4547 	*accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
4548 
4549 	return (0);
4550 }
4551 
4552 /*
4553  * These are helper functions for filesystems to traverse all
4554  * their vnodes.  See MNT_VNODE_FOREACH_ALL() in sys/mount.h.
4555  *
4556  * This interface replaces MNT_VNODE_FOREACH.
4557  */
4558 
4559 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
4560 
4561 struct vnode *
4562 __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp)
4563 {
4564 	struct vnode *vp;
4565 
4566 	if (should_yield())
4567 		kern_yield(PRI_UNCHANGED);
4568 	MNT_ILOCK(mp);
4569 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4570 	vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
4571 	while (vp != NULL && (vp->v_type == VMARKER ||
4572 	    (vp->v_iflag & VI_DOOMED) != 0))
4573 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
4574 
4575 	/* Check if we are done */
4576 	if (vp == NULL) {
4577 		__mnt_vnode_markerfree_all(mvp, mp);
4578 		/* MNT_IUNLOCK(mp); -- done in above function */
4579 		mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
4580 		return (NULL);
4581 	}
4582 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4583 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4584 	VI_LOCK(vp);
4585 	MNT_IUNLOCK(mp);
4586 	return (vp);
4587 }
4588 
4589 struct vnode *
4590 __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp)
4591 {
4592 	struct vnode *vp;
4593 
4594 	*mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4595 	MNT_ILOCK(mp);
4596 	MNT_REF(mp);
4597 	(*mvp)->v_type = VMARKER;
4598 
4599 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
4600 	while (vp != NULL && (vp->v_type == VMARKER ||
4601 	    (vp->v_iflag & VI_DOOMED) != 0))
4602 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
4603 
4604 	/* Check if we are done */
4605 	if (vp == NULL) {
4606 		MNT_REL(mp);
4607 		MNT_IUNLOCK(mp);
4608 		free(*mvp, M_VNODE_MARKER);
4609 		*mvp = NULL;
4610 		return (NULL);
4611 	}
4612 	(*mvp)->v_mount = mp;
4613 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4614 	VI_LOCK(vp);
4615 	MNT_IUNLOCK(mp);
4616 	return (vp);
4617 }
4618 
4619 
4620 void
4621 __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp)
4622 {
4623 
4624 	if (*mvp == NULL) {
4625 		MNT_IUNLOCK(mp);
4626 		return;
4627 	}
4628 
4629 	mtx_assert(MNT_MTX(mp), MA_OWNED);
4630 
4631 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4632 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4633 	MNT_REL(mp);
4634 	MNT_IUNLOCK(mp);
4635 	free(*mvp, M_VNODE_MARKER);
4636 	*mvp = NULL;
4637 }
4638 
4639 /*
4640  * These are helper functions for filesystems to traverse their
4641  * active vnodes.  See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h
4642  */
4643 struct vnode *
4644 __mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4645 {
4646 	struct vnode *vp, *nvp;
4647 
4648 	if (should_yield())
4649 		kern_yield(PRI_UNCHANGED);
4650 	MNT_ILOCK(mp);
4651 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4652 	vp = TAILQ_NEXT(*mvp, v_actfreelist);
4653 	while (vp != NULL) {
4654 		VI_LOCK(vp);
4655 		if (vp->v_mount == mp && vp->v_type != VMARKER &&
4656 		    (vp->v_iflag & VI_DOOMED) == 0)
4657 			break;
4658 		nvp = TAILQ_NEXT(vp, v_actfreelist);
4659 		VI_UNLOCK(vp);
4660 		vp = nvp;
4661 	}
4662 
4663 	/* Check if we are done */
4664 	if (vp == NULL) {
4665 		__mnt_vnode_markerfree_active(mvp, mp);
4666 		/* MNT_IUNLOCK(mp); -- done in above function */
4667 		mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
4668 		return (NULL);
4669 	}
4670 	mtx_lock(&vnode_free_list_mtx);
4671 	TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4672 	TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
4673 	mtx_unlock(&vnode_free_list_mtx);
4674 	MNT_IUNLOCK(mp);
4675 	return (vp);
4676 }
4677 
4678 struct vnode *
4679 __mnt_vnode_first_active(struct vnode **mvp, struct mount *mp)
4680 {
4681 	struct vnode *vp, *nvp;
4682 
4683 	*mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4684 	MNT_ILOCK(mp);
4685 	MNT_REF(mp);
4686 	(*mvp)->v_type = VMARKER;
4687 
4688 	vp = TAILQ_NEXT(*mvp, v_actfreelist);
4689 	while (vp != NULL) {
4690 		VI_LOCK(vp);
4691 		if (vp->v_mount == mp && vp->v_type != VMARKER &&
4692 		    (vp->v_iflag & VI_DOOMED) == 0)
4693 			break;
4694 		nvp = TAILQ_NEXT(vp, v_actfreelist);
4695 		VI_UNLOCK(vp);
4696 		vp = nvp;
4697 	}
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 	mtx_lock(&vnode_free_list_mtx);
4709 	TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
4710 	mtx_unlock(&vnode_free_list_mtx);
4711 	MNT_IUNLOCK(mp);
4712 	return (vp);
4713 }
4714 
4715 void
4716 __mnt_vnode_markerfree_active(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 	mtx_lock(&vnode_free_list_mtx);
4728 	TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4729 	mtx_unlock(&vnode_free_list_mtx);
4730 	MNT_REL(mp);
4731 	MNT_IUNLOCK(mp);
4732 	free(*mvp, M_VNODE_MARKER);
4733 	*mvp = NULL;
4734 }
4735