xref: /freebsd/sys/kern/vfs_subr.c (revision 7750ad47a9a7dbc83f87158464170c8640723293)
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 #ifdef SW_WATCHDOG
77 #include <sys/watchdog.h>
78 #endif
79 
80 #include <machine/stdarg.h>
81 
82 #include <security/mac/mac_framework.h>
83 
84 #include <vm/vm.h>
85 #include <vm/vm_object.h>
86 #include <vm/vm_extern.h>
87 #include <vm/pmap.h>
88 #include <vm/vm_map.h>
89 #include <vm/vm_page.h>
90 #include <vm/vm_kern.h>
91 #include <vm/uma.h>
92 
93 #ifdef DDB
94 #include <ddb/ddb.h>
95 #endif
96 
97 #define	WI_MPSAFEQ	0
98 #define	WI_GIANTQ	1
99 
100 static void	delmntque(struct vnode *vp);
101 static int	flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
102 		    int slpflag, int slptimeo);
103 static void	syncer_shutdown(void *arg, int howto);
104 static int	vtryrecycle(struct vnode *vp);
105 static void	v_incr_usecount(struct vnode *);
106 static void	v_decr_usecount(struct vnode *);
107 static void	v_decr_useonly(struct vnode *);
108 static void	v_upgrade_usecount(struct vnode *);
109 static void	vnlru_free(int);
110 static void	vgonel(struct vnode *);
111 static void	vfs_knllock(void *arg);
112 static void	vfs_knlunlock(void *arg);
113 static void	vfs_knl_assert_locked(void *arg);
114 static void	vfs_knl_assert_unlocked(void *arg);
115 static void	destroy_vpollinfo(struct vpollinfo *vi);
116 
117 /*
118  * Number of vnodes in existence.  Increased whenever getnewvnode()
119  * allocates a new vnode, decreased in vdropl() for VI_DOOMED vnode.
120  */
121 static unsigned long	numvnodes;
122 
123 SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
124     "Number of vnodes in existence");
125 
126 /*
127  * Conversion tables for conversion from vnode types to inode formats
128  * and back.
129  */
130 enum vtype iftovt_tab[16] = {
131 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
132 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
133 };
134 int vttoif_tab[10] = {
135 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
136 	S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
137 };
138 
139 /*
140  * List of vnodes that are ready for recycling.
141  */
142 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
143 
144 /*
145  * Free vnode target.  Free vnodes may simply be files which have been stat'd
146  * but not read.  This is somewhat common, and a small cache of such files
147  * should be kept to avoid recreation costs.
148  */
149 static u_long wantfreevnodes;
150 SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
151 /* Number of vnodes in the free list. */
152 static u_long freevnodes;
153 SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
154     "Number of vnodes in the free list");
155 
156 static int vlru_allow_cache_src;
157 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
158     &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
159 
160 /*
161  * Various variables used for debugging the new implementation of
162  * reassignbuf().
163  * XXX these are probably of (very) limited utility now.
164  */
165 static int reassignbufcalls;
166 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
167     "Number of calls to reassignbuf");
168 
169 /*
170  * Cache for the mount type id assigned to NFS.  This is used for
171  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
172  */
173 int	nfs_mount_type = -1;
174 
175 /* To keep more than one thread at a time from running vfs_getnewfsid */
176 static struct mtx mntid_mtx;
177 
178 /*
179  * Lock for any access to the following:
180  *	vnode_free_list
181  *	numvnodes
182  *	freevnodes
183  */
184 static struct mtx vnode_free_list_mtx;
185 
186 /* Publicly exported FS */
187 struct nfs_public nfs_pub;
188 
189 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
190 static uma_zone_t vnode_zone;
191 static uma_zone_t vnodepoll_zone;
192 
193 /*
194  * The workitem queue.
195  *
196  * It is useful to delay writes of file data and filesystem metadata
197  * for tens of seconds so that quickly created and deleted files need
198  * not waste disk bandwidth being created and removed. To realize this,
199  * we append vnodes to a "workitem" queue. When running with a soft
200  * updates implementation, most pending metadata dependencies should
201  * not wait for more than a few seconds. Thus, mounted on block devices
202  * are delayed only about a half the time that file data is delayed.
203  * Similarly, directory updates are more critical, so are only delayed
204  * about a third the time that file data is delayed. Thus, there are
205  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
206  * one each second (driven off the filesystem syncer process). The
207  * syncer_delayno variable indicates the next queue that is to be processed.
208  * Items that need to be processed soon are placed in this queue:
209  *
210  *	syncer_workitem_pending[syncer_delayno]
211  *
212  * A delay of fifteen seconds is done by placing the request fifteen
213  * entries later in the queue:
214  *
215  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
216  *
217  */
218 static int syncer_delayno;
219 static long syncer_mask;
220 LIST_HEAD(synclist, bufobj);
221 static struct synclist *syncer_workitem_pending[2];
222 /*
223  * The sync_mtx protects:
224  *	bo->bo_synclist
225  *	sync_vnode_count
226  *	syncer_delayno
227  *	syncer_state
228  *	syncer_workitem_pending
229  *	syncer_worklist_len
230  *	rushjob
231  */
232 static struct mtx sync_mtx;
233 static struct cv sync_wakeup;
234 
235 #define SYNCER_MAXDELAY		32
236 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
237 static int syncdelay = 30;		/* max time to delay syncing data */
238 static int filedelay = 30;		/* time to delay syncing files */
239 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
240     "Time to delay syncing files (in seconds)");
241 static int dirdelay = 29;		/* time to delay syncing directories */
242 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
243     "Time to delay syncing directories (in seconds)");
244 static int metadelay = 28;		/* time to delay syncing metadata */
245 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
246     "Time to delay syncing metadata (in seconds)");
247 static int rushjob;		/* number of slots to run ASAP */
248 static int stat_rush_requests;	/* number of times I/O speeded up */
249 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
250     "Number of times I/O speeded up (rush requests)");
251 
252 /*
253  * When shutting down the syncer, run it at four times normal speed.
254  */
255 #define SYNCER_SHUTDOWN_SPEEDUP		4
256 static int sync_vnode_count;
257 static int syncer_worklist_len;
258 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
259     syncer_state;
260 
261 /*
262  * Number of vnodes we want to exist at any one time.  This is mostly used
263  * to size hash tables in vnode-related code.  It is normally not used in
264  * getnewvnode(), as wantfreevnodes is normally nonzero.)
265  *
266  * XXX desiredvnodes is historical cruft and should not exist.
267  */
268 int desiredvnodes;
269 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
270     &desiredvnodes, 0, "Maximum number of vnodes");
271 SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
272     &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
273 static int vnlru_nowhere;
274 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
275     &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
276 
277 /*
278  * Macros to control when a vnode is freed and recycled.  All require
279  * the vnode interlock.
280  */
281 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
282 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
283 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
284 
285 
286 /*
287  * Initialize the vnode management data structures.
288  *
289  * Reevaluate the following cap on the number of vnodes after the physical
290  * memory size exceeds 512GB.  In the limit, as the physical memory size
291  * grows, the ratio of physical pages to vnodes approaches sixteen to one.
292  */
293 #ifndef	MAXVNODES_MAX
294 #define	MAXVNODES_MAX	(512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
295 #endif
296 static void
297 vntblinit(void *dummy __unused)
298 {
299 	int physvnodes, virtvnodes;
300 
301 	/*
302 	 * Desiredvnodes is a function of the physical memory size and the
303 	 * kernel's heap size.  Generally speaking, it scales with the
304 	 * physical memory size.  The ratio of desiredvnodes to physical pages
305 	 * is one to four until desiredvnodes exceeds 98,304.  Thereafter, the
306 	 * marginal ratio of desiredvnodes to physical pages is one to
307 	 * sixteen.  However, desiredvnodes is limited by the kernel's heap
308 	 * size.  The memory required by desiredvnodes vnodes and vm objects
309 	 * may not exceed one seventh of the kernel's heap size.
310 	 */
311 	physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
312 	    cnt.v_page_count) / 16;
313 	virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
314 	    sizeof(struct vnode)));
315 	desiredvnodes = min(physvnodes, virtvnodes);
316 	if (desiredvnodes > MAXVNODES_MAX) {
317 		if (bootverbose)
318 			printf("Reducing kern.maxvnodes %d -> %d\n",
319 			    desiredvnodes, MAXVNODES_MAX);
320 		desiredvnodes = MAXVNODES_MAX;
321 	}
322 	wantfreevnodes = desiredvnodes / 4;
323 	mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
324 	TAILQ_INIT(&vnode_free_list);
325 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
326 	vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
327 	    NULL, NULL, UMA_ALIGN_PTR, 0);
328 	vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
329 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
330 	/*
331 	 * Initialize the filesystem syncer.
332 	 */
333 	syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE,
334 	    &syncer_mask);
335 	syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE,
336 	    &syncer_mask);
337 	syncer_maxdelay = syncer_mask + 1;
338 	mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
339 	cv_init(&sync_wakeup, "syncer");
340 }
341 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
342 
343 
344 /*
345  * Mark a mount point as busy. Used to synchronize access and to delay
346  * unmounting. Eventually, mountlist_mtx is not released on failure.
347  *
348  * vfs_busy() is a custom lock, it can block the caller.
349  * vfs_busy() only sleeps if the unmount is active on the mount point.
350  * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
351  * vnode belonging to mp.
352  *
353  * Lookup uses vfs_busy() to traverse mount points.
354  * root fs			var fs
355  * / vnode lock		A	/ vnode lock (/var)		D
356  * /var vnode lock	B	/log vnode lock(/var/log)	E
357  * vfs_busy lock	C	vfs_busy lock			F
358  *
359  * Within each file system, the lock order is C->A->B and F->D->E.
360  *
361  * When traversing across mounts, the system follows that lock order:
362  *
363  *        C->A->B
364  *              |
365  *              +->F->D->E
366  *
367  * The lookup() process for namei("/var") illustrates the process:
368  *  VOP_LOOKUP() obtains B while A is held
369  *  vfs_busy() obtains a shared lock on F while A and B are held
370  *  vput() releases lock on B
371  *  vput() releases lock on A
372  *  VFS_ROOT() obtains lock on D while shared lock on F is held
373  *  vfs_unbusy() releases shared lock on F
374  *  vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
375  *    Attempt to lock A (instead of vp_crossmp) while D is held would
376  *    violate the global order, causing deadlocks.
377  *
378  * dounmount() locks B while F is drained.
379  */
380 int
381 vfs_busy(struct mount *mp, int flags)
382 {
383 
384 	MPASS((flags & ~MBF_MASK) == 0);
385 	CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
386 
387 	MNT_ILOCK(mp);
388 	MNT_REF(mp);
389 	/*
390 	 * If mount point is currenly being unmounted, sleep until the
391 	 * mount point fate is decided.  If thread doing the unmounting fails,
392 	 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
393 	 * that this mount point has survived the unmount attempt and vfs_busy
394 	 * should retry.  Otherwise the unmounter thread will set MNTK_REFEXPIRE
395 	 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
396 	 * about to be really destroyed.  vfs_busy needs to release its
397 	 * reference on the mount point in this case and return with ENOENT,
398 	 * telling the caller that mount mount it tried to busy is no longer
399 	 * valid.
400 	 */
401 	while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
402 		if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
403 			MNT_REL(mp);
404 			MNT_IUNLOCK(mp);
405 			CTR1(KTR_VFS, "%s: failed busying before sleeping",
406 			    __func__);
407 			return (ENOENT);
408 		}
409 		if (flags & MBF_MNTLSTLOCK)
410 			mtx_unlock(&mountlist_mtx);
411 		mp->mnt_kern_flag |= MNTK_MWAIT;
412 		msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
413 		if (flags & MBF_MNTLSTLOCK)
414 			mtx_lock(&mountlist_mtx);
415 		MNT_ILOCK(mp);
416 	}
417 	if (flags & MBF_MNTLSTLOCK)
418 		mtx_unlock(&mountlist_mtx);
419 	mp->mnt_lockref++;
420 	MNT_IUNLOCK(mp);
421 	return (0);
422 }
423 
424 /*
425  * Free a busy filesystem.
426  */
427 void
428 vfs_unbusy(struct mount *mp)
429 {
430 
431 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
432 	MNT_ILOCK(mp);
433 	MNT_REL(mp);
434 	KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
435 	mp->mnt_lockref--;
436 	if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
437 		MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
438 		CTR1(KTR_VFS, "%s: waking up waiters", __func__);
439 		mp->mnt_kern_flag &= ~MNTK_DRAINING;
440 		wakeup(&mp->mnt_lockref);
441 	}
442 	MNT_IUNLOCK(mp);
443 }
444 
445 /*
446  * Lookup a mount point by filesystem identifier.
447  */
448 struct mount *
449 vfs_getvfs(fsid_t *fsid)
450 {
451 	struct mount *mp;
452 
453 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
454 	mtx_lock(&mountlist_mtx);
455 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
456 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
457 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
458 			vfs_ref(mp);
459 			mtx_unlock(&mountlist_mtx);
460 			return (mp);
461 		}
462 	}
463 	mtx_unlock(&mountlist_mtx);
464 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
465 	return ((struct mount *) 0);
466 }
467 
468 /*
469  * Lookup a mount point by filesystem identifier, busying it before
470  * returning.
471  */
472 struct mount *
473 vfs_busyfs(fsid_t *fsid)
474 {
475 	struct mount *mp;
476 	int error;
477 
478 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
479 	mtx_lock(&mountlist_mtx);
480 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
481 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
482 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
483 			error = vfs_busy(mp, MBF_MNTLSTLOCK);
484 			if (error) {
485 				mtx_unlock(&mountlist_mtx);
486 				return (NULL);
487 			}
488 			return (mp);
489 		}
490 	}
491 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
492 	mtx_unlock(&mountlist_mtx);
493 	return ((struct mount *) 0);
494 }
495 
496 /*
497  * Check if a user can access privileged mount options.
498  */
499 int
500 vfs_suser(struct mount *mp, struct thread *td)
501 {
502 	int error;
503 
504 	/*
505 	 * If the thread is jailed, but this is not a jail-friendly file
506 	 * system, deny immediately.
507 	 */
508 	if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
509 		return (EPERM);
510 
511 	/*
512 	 * If the file system was mounted outside the jail of the calling
513 	 * thread, deny immediately.
514 	 */
515 	if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
516 		return (EPERM);
517 
518 	/*
519 	 * If file system supports delegated administration, we don't check
520 	 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
521 	 * by the file system itself.
522 	 * If this is not the user that did original mount, we check for
523 	 * the PRIV_VFS_MOUNT_OWNER privilege.
524 	 */
525 	if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
526 	    mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
527 		if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
528 			return (error);
529 	}
530 	return (0);
531 }
532 
533 /*
534  * Get a new unique fsid.  Try to make its val[0] unique, since this value
535  * will be used to create fake device numbers for stat().  Also try (but
536  * not so hard) make its val[0] unique mod 2^16, since some emulators only
537  * support 16-bit device numbers.  We end up with unique val[0]'s for the
538  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
539  *
540  * Keep in mind that several mounts may be running in parallel.  Starting
541  * the search one past where the previous search terminated is both a
542  * micro-optimization and a defense against returning the same fsid to
543  * different mounts.
544  */
545 void
546 vfs_getnewfsid(struct mount *mp)
547 {
548 	static uint16_t mntid_base;
549 	struct mount *nmp;
550 	fsid_t tfsid;
551 	int mtype;
552 
553 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
554 	mtx_lock(&mntid_mtx);
555 	mtype = mp->mnt_vfc->vfc_typenum;
556 	tfsid.val[1] = mtype;
557 	mtype = (mtype & 0xFF) << 24;
558 	for (;;) {
559 		tfsid.val[0] = makedev(255,
560 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
561 		mntid_base++;
562 		if ((nmp = vfs_getvfs(&tfsid)) == NULL)
563 			break;
564 		vfs_rel(nmp);
565 	}
566 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
567 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
568 	mtx_unlock(&mntid_mtx);
569 }
570 
571 /*
572  * Knob to control the precision of file timestamps:
573  *
574  *   0 = seconds only; nanoseconds zeroed.
575  *   1 = seconds and nanoseconds, accurate within 1/HZ.
576  *   2 = seconds and nanoseconds, truncated to microseconds.
577  * >=3 = seconds and nanoseconds, maximum precision.
578  */
579 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
580 
581 static int timestamp_precision = TSP_SEC;
582 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
583     &timestamp_precision, 0, "File timestamp precision (0: seconds, "
584     "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
585     "3+: sec + ns (max. precision))");
586 
587 /*
588  * Get a current timestamp.
589  */
590 void
591 vfs_timestamp(struct timespec *tsp)
592 {
593 	struct timeval tv;
594 
595 	switch (timestamp_precision) {
596 	case TSP_SEC:
597 		tsp->tv_sec = time_second;
598 		tsp->tv_nsec = 0;
599 		break;
600 	case TSP_HZ:
601 		getnanotime(tsp);
602 		break;
603 	case TSP_USEC:
604 		microtime(&tv);
605 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
606 		break;
607 	case TSP_NSEC:
608 	default:
609 		nanotime(tsp);
610 		break;
611 	}
612 }
613 
614 /*
615  * Set vnode attributes to VNOVAL
616  */
617 void
618 vattr_null(struct vattr *vap)
619 {
620 
621 	vap->va_type = VNON;
622 	vap->va_size = VNOVAL;
623 	vap->va_bytes = VNOVAL;
624 	vap->va_mode = VNOVAL;
625 	vap->va_nlink = VNOVAL;
626 	vap->va_uid = VNOVAL;
627 	vap->va_gid = VNOVAL;
628 	vap->va_fsid = VNOVAL;
629 	vap->va_fileid = VNOVAL;
630 	vap->va_blocksize = VNOVAL;
631 	vap->va_rdev = VNOVAL;
632 	vap->va_atime.tv_sec = VNOVAL;
633 	vap->va_atime.tv_nsec = VNOVAL;
634 	vap->va_mtime.tv_sec = VNOVAL;
635 	vap->va_mtime.tv_nsec = VNOVAL;
636 	vap->va_ctime.tv_sec = VNOVAL;
637 	vap->va_ctime.tv_nsec = VNOVAL;
638 	vap->va_birthtime.tv_sec = VNOVAL;
639 	vap->va_birthtime.tv_nsec = VNOVAL;
640 	vap->va_flags = VNOVAL;
641 	vap->va_gen = VNOVAL;
642 	vap->va_vaflags = 0;
643 }
644 
645 /*
646  * This routine is called when we have too many vnodes.  It attempts
647  * to free <count> vnodes and will potentially free vnodes that still
648  * have VM backing store (VM backing store is typically the cause
649  * of a vnode blowout so we want to do this).  Therefore, this operation
650  * is not considered cheap.
651  *
652  * A number of conditions may prevent a vnode from being reclaimed.
653  * the buffer cache may have references on the vnode, a directory
654  * vnode may still have references due to the namei cache representing
655  * underlying files, or the vnode may be in active use.   It is not
656  * desireable to reuse such vnodes.  These conditions may cause the
657  * number of vnodes to reach some minimum value regardless of what
658  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
659  */
660 static int
661 vlrureclaim(struct mount *mp)
662 {
663 	struct vnode *vp;
664 	int done;
665 	int trigger;
666 	int usevnodes;
667 	int count;
668 
669 	/*
670 	 * Calculate the trigger point, don't allow user
671 	 * screwups to blow us up.   This prevents us from
672 	 * recycling vnodes with lots of resident pages.  We
673 	 * aren't trying to free memory, we are trying to
674 	 * free vnodes.
675 	 */
676 	usevnodes = desiredvnodes;
677 	if (usevnodes <= 0)
678 		usevnodes = 1;
679 	trigger = cnt.v_page_count * 2 / usevnodes;
680 	done = 0;
681 	vn_start_write(NULL, &mp, V_WAIT);
682 	MNT_ILOCK(mp);
683 	count = mp->mnt_nvnodelistsize / 10 + 1;
684 	while (count != 0) {
685 		vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
686 		while (vp != NULL && vp->v_type == VMARKER)
687 			vp = TAILQ_NEXT(vp, v_nmntvnodes);
688 		if (vp == NULL)
689 			break;
690 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
691 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
692 		--count;
693 		if (!VI_TRYLOCK(vp))
694 			goto next_iter;
695 		/*
696 		 * If it's been deconstructed already, it's still
697 		 * referenced, or it exceeds the trigger, skip it.
698 		 */
699 		if (vp->v_usecount ||
700 		    (!vlru_allow_cache_src &&
701 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
702 		    (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
703 		    vp->v_object->resident_page_count > trigger)) {
704 			VI_UNLOCK(vp);
705 			goto next_iter;
706 		}
707 		MNT_IUNLOCK(mp);
708 		vholdl(vp);
709 		if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
710 			vdrop(vp);
711 			goto next_iter_mntunlocked;
712 		}
713 		VI_LOCK(vp);
714 		/*
715 		 * v_usecount may have been bumped after VOP_LOCK() dropped
716 		 * the vnode interlock and before it was locked again.
717 		 *
718 		 * It is not necessary to recheck VI_DOOMED because it can
719 		 * only be set by another thread that holds both the vnode
720 		 * lock and vnode interlock.  If another thread has the
721 		 * vnode lock before we get to VOP_LOCK() and obtains the
722 		 * vnode interlock after VOP_LOCK() drops the vnode
723 		 * interlock, the other thread will be unable to drop the
724 		 * vnode lock before our VOP_LOCK() call fails.
725 		 */
726 		if (vp->v_usecount ||
727 		    (!vlru_allow_cache_src &&
728 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
729 		    (vp->v_object != NULL &&
730 		    vp->v_object->resident_page_count > trigger)) {
731 			VOP_UNLOCK(vp, LK_INTERLOCK);
732 			goto next_iter_mntunlocked;
733 		}
734 		KASSERT((vp->v_iflag & VI_DOOMED) == 0,
735 		    ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
736 		vgonel(vp);
737 		VOP_UNLOCK(vp, 0);
738 		vdropl(vp);
739 		done++;
740 next_iter_mntunlocked:
741 		if (!should_yield())
742 			goto relock_mnt;
743 		goto yield;
744 next_iter:
745 		if (!should_yield())
746 			continue;
747 		MNT_IUNLOCK(mp);
748 yield:
749 		kern_yield(PRI_UNCHANGED);
750 relock_mnt:
751 		MNT_ILOCK(mp);
752 	}
753 	MNT_IUNLOCK(mp);
754 	vn_finished_write(mp);
755 	return done;
756 }
757 
758 /*
759  * Attempt to keep the free list at wantfreevnodes length.
760  */
761 static void
762 vnlru_free(int count)
763 {
764 	struct vnode *vp;
765 	int vfslocked;
766 
767 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
768 	for (; count > 0; count--) {
769 		vp = TAILQ_FIRST(&vnode_free_list);
770 		/*
771 		 * The list can be modified while the free_list_mtx
772 		 * has been dropped and vp could be NULL here.
773 		 */
774 		if (!vp)
775 			break;
776 		VNASSERT(vp->v_op != NULL, vp,
777 		    ("vnlru_free: vnode already reclaimed."));
778 		KASSERT((vp->v_iflag & VI_FREE) != 0,
779 		    ("Removing vnode not on freelist"));
780 		KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
781 		    ("Mangling active vnode"));
782 		TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
783 		/*
784 		 * Don't recycle if we can't get the interlock.
785 		 */
786 		if (!VI_TRYLOCK(vp)) {
787 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
788 			continue;
789 		}
790 		VNASSERT(VCANRECYCLE(vp), vp,
791 		    ("vp inconsistent on freelist"));
792 		freevnodes--;
793 		vp->v_iflag &= ~VI_FREE;
794 		vholdl(vp);
795 		mtx_unlock(&vnode_free_list_mtx);
796 		VI_UNLOCK(vp);
797 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
798 		vtryrecycle(vp);
799 		VFS_UNLOCK_GIANT(vfslocked);
800 		/*
801 		 * If the recycled succeeded this vdrop will actually free
802 		 * the vnode.  If not it will simply place it back on
803 		 * the free list.
804 		 */
805 		vdrop(vp);
806 		mtx_lock(&vnode_free_list_mtx);
807 	}
808 }
809 /*
810  * Attempt to recycle vnodes in a context that is always safe to block.
811  * Calling vlrurecycle() from the bowels of filesystem code has some
812  * interesting deadlock problems.
813  */
814 static struct proc *vnlruproc;
815 static int vnlruproc_sig;
816 
817 static void
818 vnlru_proc(void)
819 {
820 	struct mount *mp, *nmp;
821 	int done, vfslocked;
822 	struct proc *p = vnlruproc;
823 
824 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
825 	    SHUTDOWN_PRI_FIRST);
826 
827 	for (;;) {
828 		kproc_suspend_check(p);
829 		mtx_lock(&vnode_free_list_mtx);
830 		if (freevnodes > wantfreevnodes)
831 			vnlru_free(freevnodes - wantfreevnodes);
832 		if (numvnodes <= desiredvnodes * 9 / 10) {
833 			vnlruproc_sig = 0;
834 			wakeup(&vnlruproc_sig);
835 			msleep(vnlruproc, &vnode_free_list_mtx,
836 			    PVFS|PDROP, "vlruwt", hz);
837 			continue;
838 		}
839 		mtx_unlock(&vnode_free_list_mtx);
840 		done = 0;
841 		mtx_lock(&mountlist_mtx);
842 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
843 			if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
844 				nmp = TAILQ_NEXT(mp, mnt_list);
845 				continue;
846 			}
847 			vfslocked = VFS_LOCK_GIANT(mp);
848 			done += vlrureclaim(mp);
849 			VFS_UNLOCK_GIANT(vfslocked);
850 			mtx_lock(&mountlist_mtx);
851 			nmp = TAILQ_NEXT(mp, mnt_list);
852 			vfs_unbusy(mp);
853 		}
854 		mtx_unlock(&mountlist_mtx);
855 		if (done == 0) {
856 #if 0
857 			/* These messages are temporary debugging aids */
858 			if (vnlru_nowhere < 5)
859 				printf("vnlru process getting nowhere..\n");
860 			else if (vnlru_nowhere == 5)
861 				printf("vnlru process messages stopped.\n");
862 #endif
863 			vnlru_nowhere++;
864 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
865 		} else
866 			kern_yield(PRI_UNCHANGED);
867 	}
868 }
869 
870 static struct kproc_desc vnlru_kp = {
871 	"vnlru",
872 	vnlru_proc,
873 	&vnlruproc
874 };
875 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
876     &vnlru_kp);
877 
878 /*
879  * Routines having to do with the management of the vnode table.
880  */
881 
882 /*
883  * Try to recycle a freed vnode.  We abort if anyone picks up a reference
884  * before we actually vgone().  This function must be called with the vnode
885  * held to prevent the vnode from being returned to the free list midway
886  * through vgone().
887  */
888 static int
889 vtryrecycle(struct vnode *vp)
890 {
891 	struct mount *vnmp;
892 
893 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
894 	VNASSERT(vp->v_holdcnt, vp,
895 	    ("vtryrecycle: Recycling vp %p without a reference.", vp));
896 	/*
897 	 * This vnode may found and locked via some other list, if so we
898 	 * can't recycle it yet.
899 	 */
900 	if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
901 		CTR2(KTR_VFS,
902 		    "%s: impossible to recycle, vp %p lock is already held",
903 		    __func__, vp);
904 		return (EWOULDBLOCK);
905 	}
906 	/*
907 	 * Don't recycle if its filesystem is being suspended.
908 	 */
909 	if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
910 		VOP_UNLOCK(vp, 0);
911 		CTR2(KTR_VFS,
912 		    "%s: impossible to recycle, cannot start the write for %p",
913 		    __func__, vp);
914 		return (EBUSY);
915 	}
916 	/*
917 	 * If we got this far, we need to acquire the interlock and see if
918 	 * anyone picked up this vnode from another list.  If not, we will
919 	 * mark it with DOOMED via vgonel() so that anyone who does find it
920 	 * will skip over it.
921 	 */
922 	VI_LOCK(vp);
923 	if (vp->v_usecount) {
924 		VOP_UNLOCK(vp, LK_INTERLOCK);
925 		vn_finished_write(vnmp);
926 		CTR2(KTR_VFS,
927 		    "%s: impossible to recycle, %p is already referenced",
928 		    __func__, vp);
929 		return (EBUSY);
930 	}
931 	if ((vp->v_iflag & VI_DOOMED) == 0)
932 		vgonel(vp);
933 	VOP_UNLOCK(vp, LK_INTERLOCK);
934 	vn_finished_write(vnmp);
935 	return (0);
936 }
937 
938 /*
939  * Return the next vnode from the free list.
940  */
941 int
942 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
943     struct vnode **vpp)
944 {
945 	struct vnode *vp = NULL;
946 	struct bufobj *bo;
947 
948 	CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
949 	mtx_lock(&vnode_free_list_mtx);
950 	/*
951 	 * Lend our context to reclaim vnodes if they've exceeded the max.
952 	 */
953 	if (freevnodes > wantfreevnodes)
954 		vnlru_free(1);
955 	/*
956 	 * Wait for available vnodes.
957 	 */
958 	if (numvnodes > desiredvnodes) {
959 		if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) {
960 			/*
961 			 * File system is beeing suspended, we cannot risk a
962 			 * deadlock here, so allocate new vnode anyway.
963 			 */
964 			if (freevnodes > wantfreevnodes)
965 				vnlru_free(freevnodes - wantfreevnodes);
966 			goto alloc;
967 		}
968 		if (vnlruproc_sig == 0) {
969 			vnlruproc_sig = 1;	/* avoid unnecessary wakeups */
970 			wakeup(vnlruproc);
971 		}
972 		msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
973 		    "vlruwk", hz);
974 #if 0	/* XXX Not all VFS_VGET/ffs_vget callers check returns. */
975 		if (numvnodes > desiredvnodes) {
976 			mtx_unlock(&vnode_free_list_mtx);
977 			return (ENFILE);
978 		}
979 #endif
980 	}
981 alloc:
982 	numvnodes++;
983 	mtx_unlock(&vnode_free_list_mtx);
984 	vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
985 	/*
986 	 * Setup locks.
987 	 */
988 	vp->v_vnlock = &vp->v_lock;
989 	mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
990 	/*
991 	 * By default, don't allow shared locks unless filesystems
992 	 * opt-in.
993 	 */
994 	lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
995 	/*
996 	 * Initialize bufobj.
997 	 */
998 	bo = &vp->v_bufobj;
999 	bo->__bo_vnode = vp;
1000 	mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
1001 	bo->bo_ops = &buf_ops_bio;
1002 	bo->bo_private = vp;
1003 	TAILQ_INIT(&bo->bo_clean.bv_hd);
1004 	TAILQ_INIT(&bo->bo_dirty.bv_hd);
1005 	/*
1006 	 * Initialize namecache.
1007 	 */
1008 	LIST_INIT(&vp->v_cache_src);
1009 	TAILQ_INIT(&vp->v_cache_dst);
1010 	/*
1011 	 * Finalize various vnode identity bits.
1012 	 */
1013 	vp->v_type = VNON;
1014 	vp->v_tag = tag;
1015 	vp->v_op = vops;
1016 	v_incr_usecount(vp);
1017 	vp->v_data = NULL;
1018 #ifdef MAC
1019 	mac_vnode_init(vp);
1020 	if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1021 		mac_vnode_associate_singlelabel(mp, vp);
1022 	else if (mp == NULL && vops != &dead_vnodeops)
1023 		printf("NULL mp in getnewvnode()\n");
1024 #endif
1025 	if (mp != NULL) {
1026 		bo->bo_bsize = mp->mnt_stat.f_iosize;
1027 		if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1028 			vp->v_vflag |= VV_NOKNOTE;
1029 	}
1030 	rangelock_init(&vp->v_rl);
1031 
1032 	*vpp = vp;
1033 	return (0);
1034 }
1035 
1036 /*
1037  * Delete from old mount point vnode list, if on one.
1038  */
1039 static void
1040 delmntque(struct vnode *vp)
1041 {
1042 	struct mount *mp;
1043 	int active;
1044 
1045 	mp = vp->v_mount;
1046 	if (mp == NULL)
1047 		return;
1048 	MNT_ILOCK(mp);
1049 	VI_LOCK(vp);
1050 	KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize,
1051 	    ("Active vnode list size %d > Vnode list size %d",
1052 	     mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize));
1053 	active = vp->v_iflag & VI_ACTIVE;
1054 	vp->v_iflag &= ~VI_ACTIVE;
1055 	if (active) {
1056 		mtx_lock(&vnode_free_list_mtx);
1057 		TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist);
1058 		mp->mnt_activevnodelistsize--;
1059 		mtx_unlock(&vnode_free_list_mtx);
1060 	}
1061 	vp->v_mount = NULL;
1062 	VI_UNLOCK(vp);
1063 	VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1064 		("bad mount point vnode list size"));
1065 	TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1066 	mp->mnt_nvnodelistsize--;
1067 	MNT_REL(mp);
1068 	MNT_IUNLOCK(mp);
1069 }
1070 
1071 static void
1072 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1073 {
1074 
1075 	vp->v_data = NULL;
1076 	vp->v_op = &dead_vnodeops;
1077 	/* XXX non mp-safe fs may still call insmntque with vnode
1078 	   unlocked */
1079 	if (!VOP_ISLOCKED(vp))
1080 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1081 	vgone(vp);
1082 	vput(vp);
1083 }
1084 
1085 /*
1086  * Insert into list of vnodes for the new mount point, if available.
1087  */
1088 int
1089 insmntque1(struct vnode *vp, struct mount *mp,
1090 	void (*dtr)(struct vnode *, void *), void *dtr_arg)
1091 {
1092 	int locked;
1093 
1094 	KASSERT(vp->v_mount == NULL,
1095 		("insmntque: vnode already on per mount vnode list"));
1096 	VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1097 #ifdef DEBUG_VFS_LOCKS
1098 	if (!VFS_NEEDSGIANT(mp))
1099 		ASSERT_VOP_ELOCKED(vp,
1100 		    "insmntque: mp-safe fs and non-locked vp");
1101 #endif
1102 	/*
1103 	 * We acquire the vnode interlock early to ensure that the
1104 	 * vnode cannot be recycled by another process releasing a
1105 	 * holdcnt on it before we get it on both the vnode list
1106 	 * and the active vnode list. The mount mutex protects only
1107 	 * manipulation of the vnode list and the vnode freelist
1108 	 * mutex protects only manipulation of the active vnode list.
1109 	 * Hence the need to hold the vnode interlock throughout.
1110 	 */
1111 	MNT_ILOCK(mp);
1112 	VI_LOCK(vp);
1113 	if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1114 	    ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1115 	     mp->mnt_nvnodelistsize == 0)) {
1116 		locked = VOP_ISLOCKED(vp);
1117 		if (!locked || (locked == LK_EXCLUSIVE &&
1118 		     (vp->v_vflag & VV_FORCEINSMQ) == 0)) {
1119 			VI_UNLOCK(vp);
1120 			MNT_IUNLOCK(mp);
1121 			if (dtr != NULL)
1122 				dtr(vp, dtr_arg);
1123 			return (EBUSY);
1124 		}
1125 	}
1126 	vp->v_mount = mp;
1127 	MNT_REF(mp);
1128 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1129 	VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1130 		("neg mount point vnode list size"));
1131 	mp->mnt_nvnodelistsize++;
1132 	KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
1133 	    ("Activating already active vnode"));
1134 	vp->v_iflag |= VI_ACTIVE;
1135 	mtx_lock(&vnode_free_list_mtx);
1136 	TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
1137 	mp->mnt_activevnodelistsize++;
1138 	mtx_unlock(&vnode_free_list_mtx);
1139 	VI_UNLOCK(vp);
1140 	MNT_IUNLOCK(mp);
1141 	return (0);
1142 }
1143 
1144 int
1145 insmntque(struct vnode *vp, struct mount *mp)
1146 {
1147 
1148 	return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1149 }
1150 
1151 /*
1152  * Flush out and invalidate all buffers associated with a bufobj
1153  * Called with the underlying object locked.
1154  */
1155 int
1156 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1157 {
1158 	int error;
1159 
1160 	BO_LOCK(bo);
1161 	if (flags & V_SAVE) {
1162 		error = bufobj_wwait(bo, slpflag, slptimeo);
1163 		if (error) {
1164 			BO_UNLOCK(bo);
1165 			return (error);
1166 		}
1167 		if (bo->bo_dirty.bv_cnt > 0) {
1168 			BO_UNLOCK(bo);
1169 			if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1170 				return (error);
1171 			/*
1172 			 * XXX We could save a lock/unlock if this was only
1173 			 * enabled under INVARIANTS
1174 			 */
1175 			BO_LOCK(bo);
1176 			if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1177 				panic("vinvalbuf: dirty bufs");
1178 		}
1179 	}
1180 	/*
1181 	 * If you alter this loop please notice that interlock is dropped and
1182 	 * reacquired in flushbuflist.  Special care is needed to ensure that
1183 	 * no race conditions occur from this.
1184 	 */
1185 	do {
1186 		error = flushbuflist(&bo->bo_clean,
1187 		    flags, bo, slpflag, slptimeo);
1188 		if (error == 0 && !(flags & V_CLEANONLY))
1189 			error = flushbuflist(&bo->bo_dirty,
1190 			    flags, bo, slpflag, slptimeo);
1191 		if (error != 0 && error != EAGAIN) {
1192 			BO_UNLOCK(bo);
1193 			return (error);
1194 		}
1195 	} while (error != 0);
1196 
1197 	/*
1198 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1199 	 * have write I/O in-progress but if there is a VM object then the
1200 	 * VM object can also have read-I/O in-progress.
1201 	 */
1202 	do {
1203 		bufobj_wwait(bo, 0, 0);
1204 		BO_UNLOCK(bo);
1205 		if (bo->bo_object != NULL) {
1206 			VM_OBJECT_LOCK(bo->bo_object);
1207 			vm_object_pip_wait(bo->bo_object, "bovlbx");
1208 			VM_OBJECT_UNLOCK(bo->bo_object);
1209 		}
1210 		BO_LOCK(bo);
1211 	} while (bo->bo_numoutput > 0);
1212 	BO_UNLOCK(bo);
1213 
1214 	/*
1215 	 * Destroy the copy in the VM cache, too.
1216 	 */
1217 	if (bo->bo_object != NULL &&
1218 	    (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) {
1219 		VM_OBJECT_LOCK(bo->bo_object);
1220 		vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
1221 		    OBJPR_CLEANONLY : 0);
1222 		VM_OBJECT_UNLOCK(bo->bo_object);
1223 	}
1224 
1225 #ifdef INVARIANTS
1226 	BO_LOCK(bo);
1227 	if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 &&
1228 	    (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1229 		panic("vinvalbuf: flush failed");
1230 	BO_UNLOCK(bo);
1231 #endif
1232 	return (0);
1233 }
1234 
1235 /*
1236  * Flush out and invalidate all buffers associated with a vnode.
1237  * Called with the underlying object locked.
1238  */
1239 int
1240 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1241 {
1242 
1243 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1244 	ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1245 	return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1246 }
1247 
1248 /*
1249  * Flush out buffers on the specified list.
1250  *
1251  */
1252 static int
1253 flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1254     int slptimeo)
1255 {
1256 	struct buf *bp, *nbp;
1257 	int retval, error;
1258 	daddr_t lblkno;
1259 	b_xflags_t xflags;
1260 
1261 	ASSERT_BO_LOCKED(bo);
1262 
1263 	retval = 0;
1264 	TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1265 		if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1266 		    ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1267 			continue;
1268 		}
1269 		lblkno = 0;
1270 		xflags = 0;
1271 		if (nbp != NULL) {
1272 			lblkno = nbp->b_lblkno;
1273 			xflags = nbp->b_xflags &
1274 				(BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1275 		}
1276 		retval = EAGAIN;
1277 		error = BUF_TIMELOCK(bp,
1278 		    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1279 		    "flushbuf", slpflag, slptimeo);
1280 		if (error) {
1281 			BO_LOCK(bo);
1282 			return (error != ENOLCK ? error : EAGAIN);
1283 		}
1284 		KASSERT(bp->b_bufobj == bo,
1285 		    ("bp %p wrong b_bufobj %p should be %p",
1286 		    bp, bp->b_bufobj, bo));
1287 		if (bp->b_bufobj != bo) {	/* XXX: necessary ? */
1288 			BUF_UNLOCK(bp);
1289 			BO_LOCK(bo);
1290 			return (EAGAIN);
1291 		}
1292 		/*
1293 		 * XXX Since there are no node locks for NFS, I
1294 		 * believe there is a slight chance that a delayed
1295 		 * write will occur while sleeping just above, so
1296 		 * check for it.
1297 		 */
1298 		if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1299 		    (flags & V_SAVE)) {
1300 			BO_LOCK(bo);
1301 			bremfree(bp);
1302 			BO_UNLOCK(bo);
1303 			bp->b_flags |= B_ASYNC;
1304 			bwrite(bp);
1305 			BO_LOCK(bo);
1306 			return (EAGAIN);	/* XXX: why not loop ? */
1307 		}
1308 		BO_LOCK(bo);
1309 		bremfree(bp);
1310 		BO_UNLOCK(bo);
1311 		bp->b_flags |= (B_INVAL | B_RELBUF);
1312 		bp->b_flags &= ~B_ASYNC;
1313 		brelse(bp);
1314 		BO_LOCK(bo);
1315 		if (nbp != NULL &&
1316 		    (nbp->b_bufobj != bo ||
1317 		     nbp->b_lblkno != lblkno ||
1318 		     (nbp->b_xflags &
1319 		      (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1320 			break;			/* nbp invalid */
1321 	}
1322 	return (retval);
1323 }
1324 
1325 /*
1326  * Truncate a file's buffer and pages to a specified length.  This
1327  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1328  * sync activity.
1329  */
1330 int
1331 vtruncbuf(struct vnode *vp, struct ucred *cred, off_t length, int blksize)
1332 {
1333 	struct buf *bp, *nbp;
1334 	int anyfreed;
1335 	int trunclbn;
1336 	struct bufobj *bo;
1337 
1338 	CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1339 	    vp, cred, blksize, (uintmax_t)length);
1340 
1341 	/*
1342 	 * Round up to the *next* lbn.
1343 	 */
1344 	trunclbn = (length + blksize - 1) / blksize;
1345 
1346 	ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1347 restart:
1348 	bo = &vp->v_bufobj;
1349 	BO_LOCK(bo);
1350 	anyfreed = 1;
1351 	for (;anyfreed;) {
1352 		anyfreed = 0;
1353 		TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1354 			if (bp->b_lblkno < trunclbn)
1355 				continue;
1356 			if (BUF_LOCK(bp,
1357 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1358 			    BO_MTX(bo)) == ENOLCK)
1359 				goto restart;
1360 
1361 			BO_LOCK(bo);
1362 			bremfree(bp);
1363 			BO_UNLOCK(bo);
1364 			bp->b_flags |= (B_INVAL | B_RELBUF);
1365 			bp->b_flags &= ~B_ASYNC;
1366 			brelse(bp);
1367 			anyfreed = 1;
1368 
1369 			BO_LOCK(bo);
1370 			if (nbp != NULL &&
1371 			    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1372 			    (nbp->b_vp != vp) ||
1373 			    (nbp->b_flags & B_DELWRI))) {
1374 				BO_UNLOCK(bo);
1375 				goto restart;
1376 			}
1377 		}
1378 
1379 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1380 			if (bp->b_lblkno < trunclbn)
1381 				continue;
1382 			if (BUF_LOCK(bp,
1383 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1384 			    BO_MTX(bo)) == ENOLCK)
1385 				goto restart;
1386 			BO_LOCK(bo);
1387 			bremfree(bp);
1388 			BO_UNLOCK(bo);
1389 			bp->b_flags |= (B_INVAL | B_RELBUF);
1390 			bp->b_flags &= ~B_ASYNC;
1391 			brelse(bp);
1392 			anyfreed = 1;
1393 
1394 			BO_LOCK(bo);
1395 			if (nbp != NULL &&
1396 			    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1397 			    (nbp->b_vp != vp) ||
1398 			    (nbp->b_flags & B_DELWRI) == 0)) {
1399 				BO_UNLOCK(bo);
1400 				goto restart;
1401 			}
1402 		}
1403 	}
1404 
1405 	if (length > 0) {
1406 restartsync:
1407 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1408 			if (bp->b_lblkno > 0)
1409 				continue;
1410 			/*
1411 			 * Since we hold the vnode lock this should only
1412 			 * fail if we're racing with the buf daemon.
1413 			 */
1414 			if (BUF_LOCK(bp,
1415 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1416 			    BO_MTX(bo)) == ENOLCK) {
1417 				goto restart;
1418 			}
1419 			VNASSERT((bp->b_flags & B_DELWRI), vp,
1420 			    ("buf(%p) on dirty queue without DELWRI", bp));
1421 
1422 			BO_LOCK(bo);
1423 			bremfree(bp);
1424 			BO_UNLOCK(bo);
1425 			bawrite(bp);
1426 			BO_LOCK(bo);
1427 			goto restartsync;
1428 		}
1429 	}
1430 
1431 	bufobj_wwait(bo, 0, 0);
1432 	BO_UNLOCK(bo);
1433 	vnode_pager_setsize(vp, length);
1434 
1435 	return (0);
1436 }
1437 
1438 /*
1439  * buf_splay() - splay tree core for the clean/dirty list of buffers in
1440  *		 a vnode.
1441  *
1442  *	NOTE: We have to deal with the special case of a background bitmap
1443  *	buffer, a situation where two buffers will have the same logical
1444  *	block offset.  We want (1) only the foreground buffer to be accessed
1445  *	in a lookup and (2) must differentiate between the foreground and
1446  *	background buffer in the splay tree algorithm because the splay
1447  *	tree cannot normally handle multiple entities with the same 'index'.
1448  *	We accomplish this by adding differentiating flags to the splay tree's
1449  *	numerical domain.
1450  */
1451 static
1452 struct buf *
1453 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1454 {
1455 	struct buf dummy;
1456 	struct buf *lefttreemax, *righttreemin, *y;
1457 
1458 	if (root == NULL)
1459 		return (NULL);
1460 	lefttreemax = righttreemin = &dummy;
1461 	for (;;) {
1462 		if (lblkno < root->b_lblkno ||
1463 		    (lblkno == root->b_lblkno &&
1464 		    (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1465 			if ((y = root->b_left) == NULL)
1466 				break;
1467 			if (lblkno < y->b_lblkno) {
1468 				/* Rotate right. */
1469 				root->b_left = y->b_right;
1470 				y->b_right = root;
1471 				root = y;
1472 				if ((y = root->b_left) == NULL)
1473 					break;
1474 			}
1475 			/* Link into the new root's right tree. */
1476 			righttreemin->b_left = root;
1477 			righttreemin = root;
1478 		} else if (lblkno > root->b_lblkno ||
1479 		    (lblkno == root->b_lblkno &&
1480 		    (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1481 			if ((y = root->b_right) == NULL)
1482 				break;
1483 			if (lblkno > y->b_lblkno) {
1484 				/* Rotate left. */
1485 				root->b_right = y->b_left;
1486 				y->b_left = root;
1487 				root = y;
1488 				if ((y = root->b_right) == NULL)
1489 					break;
1490 			}
1491 			/* Link into the new root's left tree. */
1492 			lefttreemax->b_right = root;
1493 			lefttreemax = root;
1494 		} else {
1495 			break;
1496 		}
1497 		root = y;
1498 	}
1499 	/* Assemble the new root. */
1500 	lefttreemax->b_right = root->b_left;
1501 	righttreemin->b_left = root->b_right;
1502 	root->b_left = dummy.b_right;
1503 	root->b_right = dummy.b_left;
1504 	return (root);
1505 }
1506 
1507 static void
1508 buf_vlist_remove(struct buf *bp)
1509 {
1510 	struct buf *root;
1511 	struct bufv *bv;
1512 
1513 	KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1514 	ASSERT_BO_LOCKED(bp->b_bufobj);
1515 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1516 	    (BX_VNDIRTY|BX_VNCLEAN),
1517 	    ("buf_vlist_remove: Buf %p is on two lists", bp));
1518 	if (bp->b_xflags & BX_VNDIRTY)
1519 		bv = &bp->b_bufobj->bo_dirty;
1520 	else
1521 		bv = &bp->b_bufobj->bo_clean;
1522 	if (bp != bv->bv_root) {
1523 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1524 		KASSERT(root == bp, ("splay lookup failed in remove"));
1525 	}
1526 	if (bp->b_left == NULL) {
1527 		root = bp->b_right;
1528 	} else {
1529 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1530 		root->b_right = bp->b_right;
1531 	}
1532 	bv->bv_root = root;
1533 	TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1534 	bv->bv_cnt--;
1535 	bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1536 }
1537 
1538 /*
1539  * Add the buffer to the sorted clean or dirty block list using a
1540  * splay tree algorithm.
1541  *
1542  * NOTE: xflags is passed as a constant, optimizing this inline function!
1543  */
1544 static void
1545 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1546 {
1547 	struct buf *root;
1548 	struct bufv *bv;
1549 
1550 	ASSERT_BO_LOCKED(bo);
1551 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1552 	    ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1553 	bp->b_xflags |= xflags;
1554 	if (xflags & BX_VNDIRTY)
1555 		bv = &bo->bo_dirty;
1556 	else
1557 		bv = &bo->bo_clean;
1558 
1559 	root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1560 	if (root == NULL) {
1561 		bp->b_left = NULL;
1562 		bp->b_right = NULL;
1563 		TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1564 	} else if (bp->b_lblkno < root->b_lblkno ||
1565 	    (bp->b_lblkno == root->b_lblkno &&
1566 	    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1567 		bp->b_left = root->b_left;
1568 		bp->b_right = root;
1569 		root->b_left = NULL;
1570 		TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1571 	} else {
1572 		bp->b_right = root->b_right;
1573 		bp->b_left = root;
1574 		root->b_right = NULL;
1575 		TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1576 	}
1577 	bv->bv_cnt++;
1578 	bv->bv_root = bp;
1579 }
1580 
1581 /*
1582  * Lookup a buffer using the splay tree.  Note that we specifically avoid
1583  * shadow buffers used in background bitmap writes.
1584  *
1585  * This code isn't quite efficient as it could be because we are maintaining
1586  * two sorted lists and do not know which list the block resides in.
1587  *
1588  * During a "make buildworld" the desired buffer is found at one of
1589  * the roots more than 60% of the time.  Thus, checking both roots
1590  * before performing either splay eliminates unnecessary splays on the
1591  * first tree splayed.
1592  */
1593 struct buf *
1594 gbincore(struct bufobj *bo, daddr_t lblkno)
1595 {
1596 	struct buf *bp;
1597 
1598 	ASSERT_BO_LOCKED(bo);
1599 	if ((bp = bo->bo_clean.bv_root) != NULL &&
1600 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1601 		return (bp);
1602 	if ((bp = bo->bo_dirty.bv_root) != NULL &&
1603 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1604 		return (bp);
1605 	if ((bp = bo->bo_clean.bv_root) != NULL) {
1606 		bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1607 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1608 			return (bp);
1609 	}
1610 	if ((bp = bo->bo_dirty.bv_root) != NULL) {
1611 		bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1612 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1613 			return (bp);
1614 	}
1615 	return (NULL);
1616 }
1617 
1618 /*
1619  * Associate a buffer with a vnode.
1620  */
1621 void
1622 bgetvp(struct vnode *vp, struct buf *bp)
1623 {
1624 	struct bufobj *bo;
1625 
1626 	bo = &vp->v_bufobj;
1627 	ASSERT_BO_LOCKED(bo);
1628 	VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1629 
1630 	CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1631 	VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1632 	    ("bgetvp: bp already attached! %p", bp));
1633 
1634 	vhold(vp);
1635 	if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT)
1636 		bp->b_flags |= B_NEEDSGIANT;
1637 	bp->b_vp = vp;
1638 	bp->b_bufobj = bo;
1639 	/*
1640 	 * Insert onto list for new vnode.
1641 	 */
1642 	buf_vlist_add(bp, bo, BX_VNCLEAN);
1643 }
1644 
1645 /*
1646  * Disassociate a buffer from a vnode.
1647  */
1648 void
1649 brelvp(struct buf *bp)
1650 {
1651 	struct bufobj *bo;
1652 	struct vnode *vp;
1653 
1654 	CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1655 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1656 
1657 	/*
1658 	 * Delete from old vnode list, if on one.
1659 	 */
1660 	vp = bp->b_vp;		/* XXX */
1661 	bo = bp->b_bufobj;
1662 	BO_LOCK(bo);
1663 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1664 		buf_vlist_remove(bp);
1665 	else
1666 		panic("brelvp: Buffer %p not on queue.", bp);
1667 	if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1668 		bo->bo_flag &= ~BO_ONWORKLST;
1669 		mtx_lock(&sync_mtx);
1670 		LIST_REMOVE(bo, bo_synclist);
1671 		syncer_worklist_len--;
1672 		mtx_unlock(&sync_mtx);
1673 	}
1674 	bp->b_flags &= ~B_NEEDSGIANT;
1675 	bp->b_vp = NULL;
1676 	bp->b_bufobj = NULL;
1677 	BO_UNLOCK(bo);
1678 	vdrop(vp);
1679 }
1680 
1681 /*
1682  * Add an item to the syncer work queue.
1683  */
1684 static void
1685 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1686 {
1687 	int queue, slot;
1688 
1689 	ASSERT_BO_LOCKED(bo);
1690 
1691 	mtx_lock(&sync_mtx);
1692 	if (bo->bo_flag & BO_ONWORKLST)
1693 		LIST_REMOVE(bo, bo_synclist);
1694 	else {
1695 		bo->bo_flag |= BO_ONWORKLST;
1696 		syncer_worklist_len++;
1697 	}
1698 
1699 	if (delay > syncer_maxdelay - 2)
1700 		delay = syncer_maxdelay - 2;
1701 	slot = (syncer_delayno + delay) & syncer_mask;
1702 
1703 	queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ :
1704 	    WI_MPSAFEQ;
1705 	LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo,
1706 	    bo_synclist);
1707 	mtx_unlock(&sync_mtx);
1708 }
1709 
1710 static int
1711 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1712 {
1713 	int error, len;
1714 
1715 	mtx_lock(&sync_mtx);
1716 	len = syncer_worklist_len - sync_vnode_count;
1717 	mtx_unlock(&sync_mtx);
1718 	error = SYSCTL_OUT(req, &len, sizeof(len));
1719 	return (error);
1720 }
1721 
1722 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1723     sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1724 
1725 static struct proc *updateproc;
1726 static void sched_sync(void);
1727 static struct kproc_desc up_kp = {
1728 	"syncer",
1729 	sched_sync,
1730 	&updateproc
1731 };
1732 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1733 
1734 static int
1735 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1736 {
1737 	struct vnode *vp;
1738 	struct mount *mp;
1739 
1740 	*bo = LIST_FIRST(slp);
1741 	if (*bo == NULL)
1742 		return (0);
1743 	vp = (*bo)->__bo_vnode;	/* XXX */
1744 	if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1745 		return (1);
1746 	/*
1747 	 * We use vhold in case the vnode does not
1748 	 * successfully sync.  vhold prevents the vnode from
1749 	 * going away when we unlock the sync_mtx so that
1750 	 * we can acquire the vnode interlock.
1751 	 */
1752 	vholdl(vp);
1753 	mtx_unlock(&sync_mtx);
1754 	VI_UNLOCK(vp);
1755 	if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1756 		vdrop(vp);
1757 		mtx_lock(&sync_mtx);
1758 		return (*bo == LIST_FIRST(slp));
1759 	}
1760 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1761 	(void) VOP_FSYNC(vp, MNT_LAZY, td);
1762 	VOP_UNLOCK(vp, 0);
1763 	vn_finished_write(mp);
1764 	BO_LOCK(*bo);
1765 	if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1766 		/*
1767 		 * Put us back on the worklist.  The worklist
1768 		 * routine will remove us from our current
1769 		 * position and then add us back in at a later
1770 		 * position.
1771 		 */
1772 		vn_syncer_add_to_worklist(*bo, syncdelay);
1773 	}
1774 	BO_UNLOCK(*bo);
1775 	vdrop(vp);
1776 	mtx_lock(&sync_mtx);
1777 	return (0);
1778 }
1779 
1780 /*
1781  * System filesystem synchronizer daemon.
1782  */
1783 static void
1784 sched_sync(void)
1785 {
1786 	struct synclist *gnext, *next;
1787 	struct synclist *gslp, *slp;
1788 	struct bufobj *bo;
1789 	long starttime;
1790 	struct thread *td = curthread;
1791 	int last_work_seen;
1792 	int net_worklist_len;
1793 	int syncer_final_iter;
1794 	int first_printf;
1795 	int error;
1796 
1797 	last_work_seen = 0;
1798 	syncer_final_iter = 0;
1799 	first_printf = 1;
1800 	syncer_state = SYNCER_RUNNING;
1801 	starttime = time_uptime;
1802 	td->td_pflags |= TDP_NORUNNINGBUF;
1803 
1804 	EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1805 	    SHUTDOWN_PRI_LAST);
1806 
1807 	mtx_lock(&sync_mtx);
1808 	for (;;) {
1809 		if (syncer_state == SYNCER_FINAL_DELAY &&
1810 		    syncer_final_iter == 0) {
1811 			mtx_unlock(&sync_mtx);
1812 			kproc_suspend_check(td->td_proc);
1813 			mtx_lock(&sync_mtx);
1814 		}
1815 		net_worklist_len = syncer_worklist_len - sync_vnode_count;
1816 		if (syncer_state != SYNCER_RUNNING &&
1817 		    starttime != time_uptime) {
1818 			if (first_printf) {
1819 				printf("\nSyncing disks, vnodes remaining...");
1820 				first_printf = 0;
1821 			}
1822 			printf("%d ", net_worklist_len);
1823 		}
1824 		starttime = time_uptime;
1825 
1826 		/*
1827 		 * Push files whose dirty time has expired.  Be careful
1828 		 * of interrupt race on slp queue.
1829 		 *
1830 		 * Skip over empty worklist slots when shutting down.
1831 		 */
1832 		do {
1833 			slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1834 			gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1835 			syncer_delayno += 1;
1836 			if (syncer_delayno == syncer_maxdelay)
1837 				syncer_delayno = 0;
1838 			next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1839 			gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1840 			/*
1841 			 * If the worklist has wrapped since the
1842 			 * it was emptied of all but syncer vnodes,
1843 			 * switch to the FINAL_DELAY state and run
1844 			 * for one more second.
1845 			 */
1846 			if (syncer_state == SYNCER_SHUTTING_DOWN &&
1847 			    net_worklist_len == 0 &&
1848 			    last_work_seen == syncer_delayno) {
1849 				syncer_state = SYNCER_FINAL_DELAY;
1850 				syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1851 			}
1852 		} while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1853 		    LIST_EMPTY(gslp) && syncer_worklist_len > 0);
1854 
1855 		/*
1856 		 * Keep track of the last time there was anything
1857 		 * on the worklist other than syncer vnodes.
1858 		 * Return to the SHUTTING_DOWN state if any
1859 		 * new work appears.
1860 		 */
1861 		if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1862 			last_work_seen = syncer_delayno;
1863 		if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1864 			syncer_state = SYNCER_SHUTTING_DOWN;
1865 		while (!LIST_EMPTY(slp)) {
1866 			error = sync_vnode(slp, &bo, td);
1867 			if (error == 1) {
1868 				LIST_REMOVE(bo, bo_synclist);
1869 				LIST_INSERT_HEAD(next, bo, bo_synclist);
1870 				continue;
1871 			}
1872 #ifdef SW_WATCHDOG
1873 			if (first_printf == 0)
1874 				wdog_kern_pat(WD_LASTVAL);
1875 #endif
1876 		}
1877 		if (!LIST_EMPTY(gslp)) {
1878 			mtx_unlock(&sync_mtx);
1879 			mtx_lock(&Giant);
1880 			mtx_lock(&sync_mtx);
1881 			while (!LIST_EMPTY(gslp)) {
1882 				error = sync_vnode(gslp, &bo, td);
1883 				if (error == 1) {
1884 					LIST_REMOVE(bo, bo_synclist);
1885 					LIST_INSERT_HEAD(gnext, bo,
1886 					    bo_synclist);
1887 					continue;
1888 				}
1889 			}
1890 			mtx_unlock(&Giant);
1891 		}
1892 		if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1893 			syncer_final_iter--;
1894 		/*
1895 		 * The variable rushjob allows the kernel to speed up the
1896 		 * processing of the filesystem syncer process. A rushjob
1897 		 * value of N tells the filesystem syncer to process the next
1898 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1899 		 * is used by the soft update code to speed up the filesystem
1900 		 * syncer process when the incore state is getting so far
1901 		 * ahead of the disk that the kernel memory pool is being
1902 		 * threatened with exhaustion.
1903 		 */
1904 		if (rushjob > 0) {
1905 			rushjob -= 1;
1906 			continue;
1907 		}
1908 		/*
1909 		 * Just sleep for a short period of time between
1910 		 * iterations when shutting down to allow some I/O
1911 		 * to happen.
1912 		 *
1913 		 * If it has taken us less than a second to process the
1914 		 * current work, then wait. Otherwise start right over
1915 		 * again. We can still lose time if any single round
1916 		 * takes more than two seconds, but it does not really
1917 		 * matter as we are just trying to generally pace the
1918 		 * filesystem activity.
1919 		 */
1920 		if (syncer_state != SYNCER_RUNNING ||
1921 		    time_uptime == starttime) {
1922 			thread_lock(td);
1923 			sched_prio(td, PPAUSE);
1924 			thread_unlock(td);
1925 		}
1926 		if (syncer_state != SYNCER_RUNNING)
1927 			cv_timedwait(&sync_wakeup, &sync_mtx,
1928 			    hz / SYNCER_SHUTDOWN_SPEEDUP);
1929 		else if (time_uptime == starttime)
1930 			cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1931 	}
1932 }
1933 
1934 /*
1935  * Request the syncer daemon to speed up its work.
1936  * We never push it to speed up more than half of its
1937  * normal turn time, otherwise it could take over the cpu.
1938  */
1939 int
1940 speedup_syncer(void)
1941 {
1942 	int ret = 0;
1943 
1944 	mtx_lock(&sync_mtx);
1945 	if (rushjob < syncdelay / 2) {
1946 		rushjob += 1;
1947 		stat_rush_requests += 1;
1948 		ret = 1;
1949 	}
1950 	mtx_unlock(&sync_mtx);
1951 	cv_broadcast(&sync_wakeup);
1952 	return (ret);
1953 }
1954 
1955 /*
1956  * Tell the syncer to speed up its work and run though its work
1957  * list several times, then tell it to shut down.
1958  */
1959 static void
1960 syncer_shutdown(void *arg, int howto)
1961 {
1962 
1963 	if (howto & RB_NOSYNC)
1964 		return;
1965 	mtx_lock(&sync_mtx);
1966 	syncer_state = SYNCER_SHUTTING_DOWN;
1967 	rushjob = 0;
1968 	mtx_unlock(&sync_mtx);
1969 	cv_broadcast(&sync_wakeup);
1970 	kproc_shutdown(arg, howto);
1971 }
1972 
1973 /*
1974  * Reassign a buffer from one vnode to another.
1975  * Used to assign file specific control information
1976  * (indirect blocks) to the vnode to which they belong.
1977  */
1978 void
1979 reassignbuf(struct buf *bp)
1980 {
1981 	struct vnode *vp;
1982 	struct bufobj *bo;
1983 	int delay;
1984 #ifdef INVARIANTS
1985 	struct bufv *bv;
1986 #endif
1987 
1988 	vp = bp->b_vp;
1989 	bo = bp->b_bufobj;
1990 	++reassignbufcalls;
1991 
1992 	CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1993 	    bp, bp->b_vp, bp->b_flags);
1994 	/*
1995 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1996 	 * is not fully linked in.
1997 	 */
1998 	if (bp->b_flags & B_PAGING)
1999 		panic("cannot reassign paging buffer");
2000 
2001 	/*
2002 	 * Delete from old vnode list, if on one.
2003 	 */
2004 	BO_LOCK(bo);
2005 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
2006 		buf_vlist_remove(bp);
2007 	else
2008 		panic("reassignbuf: Buffer %p not on queue.", bp);
2009 	/*
2010 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
2011 	 * of clean buffers.
2012 	 */
2013 	if (bp->b_flags & B_DELWRI) {
2014 		if ((bo->bo_flag & BO_ONWORKLST) == 0) {
2015 			switch (vp->v_type) {
2016 			case VDIR:
2017 				delay = dirdelay;
2018 				break;
2019 			case VCHR:
2020 				delay = metadelay;
2021 				break;
2022 			default:
2023 				delay = filedelay;
2024 			}
2025 			vn_syncer_add_to_worklist(bo, delay);
2026 		}
2027 		buf_vlist_add(bp, bo, BX_VNDIRTY);
2028 	} else {
2029 		buf_vlist_add(bp, bo, BX_VNCLEAN);
2030 
2031 		if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
2032 			mtx_lock(&sync_mtx);
2033 			LIST_REMOVE(bo, bo_synclist);
2034 			syncer_worklist_len--;
2035 			mtx_unlock(&sync_mtx);
2036 			bo->bo_flag &= ~BO_ONWORKLST;
2037 		}
2038 	}
2039 #ifdef INVARIANTS
2040 	bv = &bo->bo_clean;
2041 	bp = TAILQ_FIRST(&bv->bv_hd);
2042 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2043 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2044 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
2045 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2046 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2047 	bv = &bo->bo_dirty;
2048 	bp = TAILQ_FIRST(&bv->bv_hd);
2049 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2050 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2051 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
2052 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2053 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2054 #endif
2055 	BO_UNLOCK(bo);
2056 }
2057 
2058 /*
2059  * Increment the use and hold counts on the vnode, taking care to reference
2060  * the driver's usecount if this is a chardev.  The vholdl() will remove
2061  * the vnode from the free list if it is presently free.  Requires the
2062  * vnode interlock and returns with it held.
2063  */
2064 static void
2065 v_incr_usecount(struct vnode *vp)
2066 {
2067 
2068 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2069 	vp->v_usecount++;
2070 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2071 		dev_lock();
2072 		vp->v_rdev->si_usecount++;
2073 		dev_unlock();
2074 	}
2075 	vholdl(vp);
2076 }
2077 
2078 /*
2079  * Turn a holdcnt into a use+holdcnt such that only one call to
2080  * v_decr_usecount is needed.
2081  */
2082 static void
2083 v_upgrade_usecount(struct vnode *vp)
2084 {
2085 
2086 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2087 	vp->v_usecount++;
2088 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2089 		dev_lock();
2090 		vp->v_rdev->si_usecount++;
2091 		dev_unlock();
2092 	}
2093 }
2094 
2095 /*
2096  * Decrement the vnode use and hold count along with the driver's usecount
2097  * if this is a chardev.  The vdropl() below releases the vnode interlock
2098  * as it may free the vnode.
2099  */
2100 static void
2101 v_decr_usecount(struct vnode *vp)
2102 {
2103 
2104 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2105 	VNASSERT(vp->v_usecount > 0, vp,
2106 	    ("v_decr_usecount: negative usecount"));
2107 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2108 	vp->v_usecount--;
2109 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2110 		dev_lock();
2111 		vp->v_rdev->si_usecount--;
2112 		dev_unlock();
2113 	}
2114 	vdropl(vp);
2115 }
2116 
2117 /*
2118  * Decrement only the use count and driver use count.  This is intended to
2119  * be paired with a follow on vdropl() to release the remaining hold count.
2120  * In this way we may vgone() a vnode with a 0 usecount without risk of
2121  * having it end up on a free list because the hold count is kept above 0.
2122  */
2123 static void
2124 v_decr_useonly(struct vnode *vp)
2125 {
2126 
2127 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2128 	VNASSERT(vp->v_usecount > 0, vp,
2129 	    ("v_decr_useonly: negative usecount"));
2130 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2131 	vp->v_usecount--;
2132 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2133 		dev_lock();
2134 		vp->v_rdev->si_usecount--;
2135 		dev_unlock();
2136 	}
2137 }
2138 
2139 /*
2140  * Grab a particular vnode from the free list, increment its
2141  * reference count and lock it.  VI_DOOMED is set if the vnode
2142  * is being destroyed.  Only callers who specify LK_RETRY will
2143  * see doomed vnodes.  If inactive processing was delayed in
2144  * vput try to do it here.
2145  */
2146 int
2147 vget(struct vnode *vp, int flags, struct thread *td)
2148 {
2149 	int error;
2150 
2151 	error = 0;
2152 	VFS_ASSERT_GIANT(vp->v_mount);
2153 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2154 	    ("vget: invalid lock operation"));
2155 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2156 
2157 	if ((flags & LK_INTERLOCK) == 0)
2158 		VI_LOCK(vp);
2159 	vholdl(vp);
2160 	if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2161 		vdrop(vp);
2162 		CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2163 		    vp);
2164 		return (error);
2165 	}
2166 	if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2167 		panic("vget: vn_lock failed to return ENOENT\n");
2168 	VI_LOCK(vp);
2169 	/* Upgrade our holdcnt to a usecount. */
2170 	v_upgrade_usecount(vp);
2171 	/*
2172 	 * We don't guarantee that any particular close will
2173 	 * trigger inactive processing so just make a best effort
2174 	 * here at preventing a reference to a removed file.  If
2175 	 * we don't succeed no harm is done.
2176 	 */
2177 	if (vp->v_iflag & VI_OWEINACT) {
2178 		if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2179 		    (flags & LK_NOWAIT) == 0)
2180 			vinactive(vp, td);
2181 		vp->v_iflag &= ~VI_OWEINACT;
2182 	}
2183 	VI_UNLOCK(vp);
2184 	return (0);
2185 }
2186 
2187 /*
2188  * Increase the reference count of a vnode.
2189  */
2190 void
2191 vref(struct vnode *vp)
2192 {
2193 
2194 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2195 	VI_LOCK(vp);
2196 	v_incr_usecount(vp);
2197 	VI_UNLOCK(vp);
2198 }
2199 
2200 /*
2201  * Return reference count of a vnode.
2202  *
2203  * The results of this call are only guaranteed when some mechanism other
2204  * than the VI lock is used to stop other processes from gaining references
2205  * to the vnode.  This may be the case if the caller holds the only reference.
2206  * This is also useful when stale data is acceptable as race conditions may
2207  * be accounted for by some other means.
2208  */
2209 int
2210 vrefcnt(struct vnode *vp)
2211 {
2212 	int usecnt;
2213 
2214 	VI_LOCK(vp);
2215 	usecnt = vp->v_usecount;
2216 	VI_UNLOCK(vp);
2217 
2218 	return (usecnt);
2219 }
2220 
2221 #define	VPUTX_VRELE	1
2222 #define	VPUTX_VPUT	2
2223 #define	VPUTX_VUNREF	3
2224 
2225 static void
2226 vputx(struct vnode *vp, int func)
2227 {
2228 	int error;
2229 
2230 	KASSERT(vp != NULL, ("vputx: null vp"));
2231 	if (func == VPUTX_VUNREF)
2232 		ASSERT_VOP_LOCKED(vp, "vunref");
2233 	else if (func == VPUTX_VPUT)
2234 		ASSERT_VOP_LOCKED(vp, "vput");
2235 	else
2236 		KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2237 	VFS_ASSERT_GIANT(vp->v_mount);
2238 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2239 	VI_LOCK(vp);
2240 
2241 	/* Skip this v_writecount check if we're going to panic below. */
2242 	VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2243 	    ("vputx: missed vn_close"));
2244 	error = 0;
2245 
2246 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2247 	    vp->v_usecount == 1)) {
2248 		if (func == VPUTX_VPUT)
2249 			VOP_UNLOCK(vp, 0);
2250 		v_decr_usecount(vp);
2251 		return;
2252 	}
2253 
2254 	if (vp->v_usecount != 1) {
2255 		vprint("vputx: negative ref count", vp);
2256 		panic("vputx: negative ref cnt");
2257 	}
2258 	CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2259 	/*
2260 	 * We want to hold the vnode until the inactive finishes to
2261 	 * prevent vgone() races.  We drop the use count here and the
2262 	 * hold count below when we're done.
2263 	 */
2264 	v_decr_useonly(vp);
2265 	/*
2266 	 * We must call VOP_INACTIVE with the node locked. Mark
2267 	 * as VI_DOINGINACT to avoid recursion.
2268 	 */
2269 	vp->v_iflag |= VI_OWEINACT;
2270 	switch (func) {
2271 	case VPUTX_VRELE:
2272 		error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2273 		VI_LOCK(vp);
2274 		break;
2275 	case VPUTX_VPUT:
2276 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2277 			error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
2278 			    LK_NOWAIT);
2279 			VI_LOCK(vp);
2280 		}
2281 		break;
2282 	case VPUTX_VUNREF:
2283 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
2284 			error = EBUSY;
2285 		break;
2286 	}
2287 	if (vp->v_usecount > 0)
2288 		vp->v_iflag &= ~VI_OWEINACT;
2289 	if (error == 0) {
2290 		if (vp->v_iflag & VI_OWEINACT)
2291 			vinactive(vp, curthread);
2292 		if (func != VPUTX_VUNREF)
2293 			VOP_UNLOCK(vp, 0);
2294 	}
2295 	vdropl(vp);
2296 }
2297 
2298 /*
2299  * Vnode put/release.
2300  * If count drops to zero, call inactive routine and return to freelist.
2301  */
2302 void
2303 vrele(struct vnode *vp)
2304 {
2305 
2306 	vputx(vp, VPUTX_VRELE);
2307 }
2308 
2309 /*
2310  * Release an already locked vnode.  This give the same effects as
2311  * unlock+vrele(), but takes less time and avoids releasing and
2312  * re-aquiring the lock (as vrele() acquires the lock internally.)
2313  */
2314 void
2315 vput(struct vnode *vp)
2316 {
2317 
2318 	vputx(vp, VPUTX_VPUT);
2319 }
2320 
2321 /*
2322  * Release an exclusively locked vnode. Do not unlock the vnode lock.
2323  */
2324 void
2325 vunref(struct vnode *vp)
2326 {
2327 
2328 	vputx(vp, VPUTX_VUNREF);
2329 }
2330 
2331 /*
2332  * Somebody doesn't want the vnode recycled.
2333  */
2334 void
2335 vhold(struct vnode *vp)
2336 {
2337 
2338 	VI_LOCK(vp);
2339 	vholdl(vp);
2340 	VI_UNLOCK(vp);
2341 }
2342 
2343 /*
2344  * Increase the hold count and activate if this is the first reference.
2345  */
2346 void
2347 vholdl(struct vnode *vp)
2348 {
2349 	struct mount *mp;
2350 
2351 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2352 	vp->v_holdcnt++;
2353 	if (!VSHOULDBUSY(vp))
2354 		return;
2355 	ASSERT_VI_LOCKED(vp, "vholdl");
2356 	VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
2357 	VNASSERT(vp->v_op != NULL, vp, ("vholdl: vnode already reclaimed."));
2358 	/*
2359 	 * Remove a vnode from the free list, mark it as in use,
2360 	 * and put it on the active list.
2361 	 */
2362 	mtx_lock(&vnode_free_list_mtx);
2363 	TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
2364 	freevnodes--;
2365 	vp->v_iflag &= ~(VI_FREE|VI_AGE);
2366 	KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
2367 	    ("Activating already active vnode"));
2368 	vp->v_iflag |= VI_ACTIVE;
2369 	mp = vp->v_mount;
2370 	TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
2371 	mp->mnt_activevnodelistsize++;
2372 	mtx_unlock(&vnode_free_list_mtx);
2373 }
2374 
2375 /*
2376  * Note that there is one less who cares about this vnode.
2377  * vdrop() is the opposite of vhold().
2378  */
2379 void
2380 vdrop(struct vnode *vp)
2381 {
2382 
2383 	VI_LOCK(vp);
2384 	vdropl(vp);
2385 }
2386 
2387 /*
2388  * Drop the hold count of the vnode.  If this is the last reference to
2389  * the vnode we place it on the free list unless it has been vgone'd
2390  * (marked VI_DOOMED) in which case we will free it.
2391  */
2392 void
2393 vdropl(struct vnode *vp)
2394 {
2395 	struct bufobj *bo;
2396 	struct mount *mp;
2397 	int active;
2398 
2399 	ASSERT_VI_LOCKED(vp, "vdropl");
2400 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2401 	if (vp->v_holdcnt <= 0)
2402 		panic("vdrop: holdcnt %d", vp->v_holdcnt);
2403 	vp->v_holdcnt--;
2404 	if (vp->v_holdcnt > 0) {
2405 		VI_UNLOCK(vp);
2406 		return;
2407 	}
2408 	if ((vp->v_iflag & VI_DOOMED) == 0) {
2409 		/*
2410 		 * Mark a vnode as free: remove it from its active list
2411 		 * and put it up for recycling on the freelist.
2412 		 */
2413 		VNASSERT(vp->v_op != NULL, vp,
2414 		    ("vdropl: vnode already reclaimed."));
2415 		VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2416 		    ("vnode already free"));
2417 		VNASSERT(VSHOULDFREE(vp), vp,
2418 		    ("vdropl: freeing when we shouldn't"));
2419 		active = vp->v_iflag & VI_ACTIVE;
2420 		vp->v_iflag &= ~VI_ACTIVE;
2421 		mp = vp->v_mount;
2422 		mtx_lock(&vnode_free_list_mtx);
2423 		if (active) {
2424 			TAILQ_REMOVE(&mp->mnt_activevnodelist, vp,
2425 			    v_actfreelist);
2426 			mp->mnt_activevnodelistsize--;
2427 		}
2428 		if (vp->v_iflag & VI_AGE) {
2429 			TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_actfreelist);
2430 		} else {
2431 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
2432 		}
2433 		freevnodes++;
2434 		vp->v_iflag &= ~VI_AGE;
2435 		vp->v_iflag |= VI_FREE;
2436 		mtx_unlock(&vnode_free_list_mtx);
2437 		VI_UNLOCK(vp);
2438 		return;
2439 	}
2440 	/*
2441 	 * The vnode has been marked for destruction, so free it.
2442 	 */
2443 	CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp);
2444 	mtx_lock(&vnode_free_list_mtx);
2445 	numvnodes--;
2446 	mtx_unlock(&vnode_free_list_mtx);
2447 	bo = &vp->v_bufobj;
2448 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2449 	    ("cleaned vnode still on the free list."));
2450 	VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
2451 	VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
2452 	VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
2453 	VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
2454 	VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
2455 	VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
2456 	VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
2457 	VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
2458 	VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
2459 	VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
2460 	VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
2461 	VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
2462 	VI_UNLOCK(vp);
2463 #ifdef MAC
2464 	mac_vnode_destroy(vp);
2465 #endif
2466 	if (vp->v_pollinfo != NULL)
2467 		destroy_vpollinfo(vp->v_pollinfo);
2468 #ifdef INVARIANTS
2469 	/* XXX Elsewhere we detect an already freed vnode via NULL v_op. */
2470 	vp->v_op = NULL;
2471 #endif
2472 	rangelock_destroy(&vp->v_rl);
2473 	lockdestroy(vp->v_vnlock);
2474 	mtx_destroy(&vp->v_interlock);
2475 	mtx_destroy(BO_MTX(bo));
2476 	uma_zfree(vnode_zone, vp);
2477 }
2478 
2479 /*
2480  * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2481  * flags.  DOINGINACT prevents us from recursing in calls to vinactive.
2482  * OWEINACT tracks whether a vnode missed a call to inactive due to a
2483  * failed lock upgrade.
2484  */
2485 void
2486 vinactive(struct vnode *vp, struct thread *td)
2487 {
2488 	struct vm_object *obj;
2489 
2490 	ASSERT_VOP_ELOCKED(vp, "vinactive");
2491 	ASSERT_VI_LOCKED(vp, "vinactive");
2492 	VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2493 	    ("vinactive: recursed on VI_DOINGINACT"));
2494 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2495 	vp->v_iflag |= VI_DOINGINACT;
2496 	vp->v_iflag &= ~VI_OWEINACT;
2497 	VI_UNLOCK(vp);
2498 	/*
2499 	 * Before moving off the active list, we must be sure that any
2500 	 * modified pages are on the vnode's dirty list since these will
2501 	 * no longer be checked once the vnode is on the inactive list.
2502 	 */
2503 	obj = vp->v_object;
2504 	if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
2505 		VM_OBJECT_LOCK(obj);
2506 		vm_object_page_clean(obj, 0, 0, OBJPC_NOSYNC);
2507 		VM_OBJECT_UNLOCK(obj);
2508 	}
2509 	VOP_INACTIVE(vp, td);
2510 	VI_LOCK(vp);
2511 	VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2512 	    ("vinactive: lost VI_DOINGINACT"));
2513 	vp->v_iflag &= ~VI_DOINGINACT;
2514 }
2515 
2516 /*
2517  * Remove any vnodes in the vnode table belonging to mount point mp.
2518  *
2519  * If FORCECLOSE is not specified, there should not be any active ones,
2520  * return error if any are found (nb: this is a user error, not a
2521  * system error). If FORCECLOSE is specified, detach any active vnodes
2522  * that are found.
2523  *
2524  * If WRITECLOSE is set, only flush out regular file vnodes open for
2525  * writing.
2526  *
2527  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2528  *
2529  * `rootrefs' specifies the base reference count for the root vnode
2530  * of this filesystem. The root vnode is considered busy if its
2531  * v_usecount exceeds this value. On a successful return, vflush(, td)
2532  * will call vrele() on the root vnode exactly rootrefs times.
2533  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2534  * be zero.
2535  */
2536 #ifdef DIAGNOSTIC
2537 static int busyprt = 0;		/* print out busy vnodes */
2538 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
2539 #endif
2540 
2541 int
2542 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
2543 {
2544 	struct vnode *vp, *mvp, *rootvp = NULL;
2545 	struct vattr vattr;
2546 	int busy = 0, error;
2547 
2548 	CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2549 	    rootrefs, flags);
2550 	if (rootrefs > 0) {
2551 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2552 		    ("vflush: bad args"));
2553 		/*
2554 		 * Get the filesystem root vnode. We can vput() it
2555 		 * immediately, since with rootrefs > 0, it won't go away.
2556 		 */
2557 		if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2558 			CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2559 			    __func__, error);
2560 			return (error);
2561 		}
2562 		vput(rootvp);
2563 	}
2564 loop:
2565 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
2566 		vholdl(vp);
2567 		error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2568 		if (error) {
2569 			vdrop(vp);
2570 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2571 			goto loop;
2572 		}
2573 		/*
2574 		 * Skip over a vnodes marked VV_SYSTEM.
2575 		 */
2576 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2577 			VOP_UNLOCK(vp, 0);
2578 			vdrop(vp);
2579 			continue;
2580 		}
2581 		/*
2582 		 * If WRITECLOSE is set, flush out unlinked but still open
2583 		 * files (even if open only for reading) and regular file
2584 		 * vnodes open for writing.
2585 		 */
2586 		if (flags & WRITECLOSE) {
2587 			if (vp->v_object != NULL) {
2588 				VM_OBJECT_LOCK(vp->v_object);
2589 				vm_object_page_clean(vp->v_object, 0, 0, 0);
2590 				VM_OBJECT_UNLOCK(vp->v_object);
2591 			}
2592 			error = VOP_FSYNC(vp, MNT_WAIT, td);
2593 			if (error != 0) {
2594 				VOP_UNLOCK(vp, 0);
2595 				vdrop(vp);
2596 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2597 				return (error);
2598 			}
2599 			error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2600 			VI_LOCK(vp);
2601 
2602 			if ((vp->v_type == VNON ||
2603 			    (error == 0 && vattr.va_nlink > 0)) &&
2604 			    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2605 				VOP_UNLOCK(vp, 0);
2606 				vdropl(vp);
2607 				continue;
2608 			}
2609 		} else
2610 			VI_LOCK(vp);
2611 		/*
2612 		 * With v_usecount == 0, all we need to do is clear out the
2613 		 * vnode data structures and we are done.
2614 		 *
2615 		 * If FORCECLOSE is set, forcibly close the vnode.
2616 		 */
2617 		if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2618 			VNASSERT(vp->v_usecount == 0 ||
2619 			    (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2620 			    ("device VNODE %p is FORCECLOSED", vp));
2621 			vgonel(vp);
2622 		} else {
2623 			busy++;
2624 #ifdef DIAGNOSTIC
2625 			if (busyprt)
2626 				vprint("vflush: busy vnode", vp);
2627 #endif
2628 		}
2629 		VOP_UNLOCK(vp, 0);
2630 		vdropl(vp);
2631 	}
2632 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2633 		/*
2634 		 * If just the root vnode is busy, and if its refcount
2635 		 * is equal to `rootrefs', then go ahead and kill it.
2636 		 */
2637 		VI_LOCK(rootvp);
2638 		KASSERT(busy > 0, ("vflush: not busy"));
2639 		VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2640 		    ("vflush: usecount %d < rootrefs %d",
2641 		     rootvp->v_usecount, rootrefs));
2642 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2643 			VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2644 			vgone(rootvp);
2645 			VOP_UNLOCK(rootvp, 0);
2646 			busy = 0;
2647 		} else
2648 			VI_UNLOCK(rootvp);
2649 	}
2650 	if (busy) {
2651 		CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2652 		    busy);
2653 		return (EBUSY);
2654 	}
2655 	for (; rootrefs > 0; rootrefs--)
2656 		vrele(rootvp);
2657 	return (0);
2658 }
2659 
2660 /*
2661  * Recycle an unused vnode to the front of the free list.
2662  */
2663 int
2664 vrecycle(struct vnode *vp)
2665 {
2666 	int recycled;
2667 
2668 	ASSERT_VOP_ELOCKED(vp, "vrecycle");
2669 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2670 	recycled = 0;
2671 	VI_LOCK(vp);
2672 	if (vp->v_usecount == 0) {
2673 		recycled = 1;
2674 		vgonel(vp);
2675 	}
2676 	VI_UNLOCK(vp);
2677 	return (recycled);
2678 }
2679 
2680 /*
2681  * Eliminate all activity associated with a vnode
2682  * in preparation for reuse.
2683  */
2684 void
2685 vgone(struct vnode *vp)
2686 {
2687 	VI_LOCK(vp);
2688 	vgonel(vp);
2689 	VI_UNLOCK(vp);
2690 }
2691 
2692 /*
2693  * vgone, with the vp interlock held.
2694  */
2695 void
2696 vgonel(struct vnode *vp)
2697 {
2698 	struct thread *td;
2699 	int oweinact;
2700 	int active;
2701 	struct mount *mp;
2702 
2703 	ASSERT_VOP_ELOCKED(vp, "vgonel");
2704 	ASSERT_VI_LOCKED(vp, "vgonel");
2705 	VNASSERT(vp->v_holdcnt, vp,
2706 	    ("vgonel: vp %p has no reference.", vp));
2707 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2708 	td = curthread;
2709 
2710 	/*
2711 	 * Don't vgonel if we're already doomed.
2712 	 */
2713 	if (vp->v_iflag & VI_DOOMED)
2714 		return;
2715 	vp->v_iflag |= VI_DOOMED;
2716 	/*
2717 	 * Check to see if the vnode is in use.  If so, we have to call
2718 	 * VOP_CLOSE() and VOP_INACTIVE().
2719 	 */
2720 	active = vp->v_usecount;
2721 	oweinact = (vp->v_iflag & VI_OWEINACT);
2722 	VI_UNLOCK(vp);
2723 	/*
2724 	 * Clean out any buffers associated with the vnode.
2725 	 * If the flush fails, just toss the buffers.
2726 	 */
2727 	mp = NULL;
2728 	if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2729 		(void) vn_start_secondary_write(vp, &mp, V_WAIT);
2730 	if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
2731 		vinvalbuf(vp, 0, 0, 0);
2732 
2733 	/*
2734 	 * If purging an active vnode, it must be closed and
2735 	 * deactivated before being reclaimed.
2736 	 */
2737 	if (active)
2738 		VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2739 	if (oweinact || active) {
2740 		VI_LOCK(vp);
2741 		if ((vp->v_iflag & VI_DOINGINACT) == 0)
2742 			vinactive(vp, td);
2743 		VI_UNLOCK(vp);
2744 	}
2745 	if (vp->v_type == VSOCK)
2746 		vfs_unp_reclaim(vp);
2747 	/*
2748 	 * Reclaim the vnode.
2749 	 */
2750 	if (VOP_RECLAIM(vp, td))
2751 		panic("vgone: cannot reclaim");
2752 	if (mp != NULL)
2753 		vn_finished_secondary_write(mp);
2754 	VNASSERT(vp->v_object == NULL, vp,
2755 	    ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2756 	/*
2757 	 * Clear the advisory locks and wake up waiting threads.
2758 	 */
2759 	(void)VOP_ADVLOCKPURGE(vp);
2760 	/*
2761 	 * Delete from old mount point vnode list.
2762 	 */
2763 	delmntque(vp);
2764 	cache_purge(vp);
2765 	/*
2766 	 * Done with purge, reset to the standard lock and invalidate
2767 	 * the vnode.
2768 	 */
2769 	VI_LOCK(vp);
2770 	vp->v_vnlock = &vp->v_lock;
2771 	vp->v_op = &dead_vnodeops;
2772 	vp->v_tag = "none";
2773 	vp->v_type = VBAD;
2774 }
2775 
2776 /*
2777  * Calculate the total number of references to a special device.
2778  */
2779 int
2780 vcount(struct vnode *vp)
2781 {
2782 	int count;
2783 
2784 	dev_lock();
2785 	count = vp->v_rdev->si_usecount;
2786 	dev_unlock();
2787 	return (count);
2788 }
2789 
2790 /*
2791  * Same as above, but using the struct cdev *as argument
2792  */
2793 int
2794 count_dev(struct cdev *dev)
2795 {
2796 	int count;
2797 
2798 	dev_lock();
2799 	count = dev->si_usecount;
2800 	dev_unlock();
2801 	return(count);
2802 }
2803 
2804 /*
2805  * Print out a description of a vnode.
2806  */
2807 static char *typename[] =
2808 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2809  "VMARKER"};
2810 
2811 void
2812 vn_printf(struct vnode *vp, const char *fmt, ...)
2813 {
2814 	va_list ap;
2815 	char buf[256], buf2[16];
2816 	u_long flags;
2817 
2818 	va_start(ap, fmt);
2819 	vprintf(fmt, ap);
2820 	va_end(ap);
2821 	printf("%p: ", (void *)vp);
2822 	printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2823 	printf("    usecount %d, writecount %d, refcount %d mountedhere %p\n",
2824 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2825 	buf[0] = '\0';
2826 	buf[1] = '\0';
2827 	if (vp->v_vflag & VV_ROOT)
2828 		strlcat(buf, "|VV_ROOT", sizeof(buf));
2829 	if (vp->v_vflag & VV_ISTTY)
2830 		strlcat(buf, "|VV_ISTTY", sizeof(buf));
2831 	if (vp->v_vflag & VV_NOSYNC)
2832 		strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2833 	if (vp->v_vflag & VV_CACHEDLABEL)
2834 		strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2835 	if (vp->v_vflag & VV_TEXT)
2836 		strlcat(buf, "|VV_TEXT", sizeof(buf));
2837 	if (vp->v_vflag & VV_COPYONWRITE)
2838 		strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2839 	if (vp->v_vflag & VV_SYSTEM)
2840 		strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2841 	if (vp->v_vflag & VV_PROCDEP)
2842 		strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2843 	if (vp->v_vflag & VV_NOKNOTE)
2844 		strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2845 	if (vp->v_vflag & VV_DELETED)
2846 		strlcat(buf, "|VV_DELETED", sizeof(buf));
2847 	if (vp->v_vflag & VV_MD)
2848 		strlcat(buf, "|VV_MD", sizeof(buf));
2849 	flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC |
2850 	    VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2851 	    VV_NOKNOTE | VV_DELETED | VV_MD);
2852 	if (flags != 0) {
2853 		snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2854 		strlcat(buf, buf2, sizeof(buf));
2855 	}
2856 	if (vp->v_iflag & VI_MOUNT)
2857 		strlcat(buf, "|VI_MOUNT", sizeof(buf));
2858 	if (vp->v_iflag & VI_AGE)
2859 		strlcat(buf, "|VI_AGE", sizeof(buf));
2860 	if (vp->v_iflag & VI_DOOMED)
2861 		strlcat(buf, "|VI_DOOMED", sizeof(buf));
2862 	if (vp->v_iflag & VI_FREE)
2863 		strlcat(buf, "|VI_FREE", sizeof(buf));
2864 	if (vp->v_iflag & VI_DOINGINACT)
2865 		strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2866 	if (vp->v_iflag & VI_OWEINACT)
2867 		strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2868 	flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2869 	    VI_DOINGINACT | VI_OWEINACT);
2870 	if (flags != 0) {
2871 		snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2872 		strlcat(buf, buf2, sizeof(buf));
2873 	}
2874 	printf("    flags (%s)\n", buf + 1);
2875 	if (mtx_owned(VI_MTX(vp)))
2876 		printf(" VI_LOCKed");
2877 	if (vp->v_object != NULL)
2878 		printf("    v_object %p ref %d pages %d\n",
2879 		    vp->v_object, vp->v_object->ref_count,
2880 		    vp->v_object->resident_page_count);
2881 	printf("    ");
2882 	lockmgr_printinfo(vp->v_vnlock);
2883 	if (vp->v_data != NULL)
2884 		VOP_PRINT(vp);
2885 }
2886 
2887 #ifdef DDB
2888 /*
2889  * List all of the locked vnodes in the system.
2890  * Called when debugging the kernel.
2891  */
2892 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2893 {
2894 	struct mount *mp, *nmp;
2895 	struct vnode *vp;
2896 
2897 	/*
2898 	 * Note: because this is DDB, we can't obey the locking semantics
2899 	 * for these structures, which means we could catch an inconsistent
2900 	 * state and dereference a nasty pointer.  Not much to be done
2901 	 * about that.
2902 	 */
2903 	db_printf("Locked vnodes\n");
2904 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2905 		nmp = TAILQ_NEXT(mp, mnt_list);
2906 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2907 			if (vp->v_type != VMARKER &&
2908 			    VOP_ISLOCKED(vp))
2909 				vprint("", vp);
2910 		}
2911 		nmp = TAILQ_NEXT(mp, mnt_list);
2912 	}
2913 }
2914 
2915 /*
2916  * Show details about the given vnode.
2917  */
2918 DB_SHOW_COMMAND(vnode, db_show_vnode)
2919 {
2920 	struct vnode *vp;
2921 
2922 	if (!have_addr)
2923 		return;
2924 	vp = (struct vnode *)addr;
2925 	vn_printf(vp, "vnode ");
2926 }
2927 
2928 /*
2929  * Show details about the given mount point.
2930  */
2931 DB_SHOW_COMMAND(mount, db_show_mount)
2932 {
2933 	struct mount *mp;
2934 	struct vfsopt *opt;
2935 	struct statfs *sp;
2936 	struct vnode *vp;
2937 	char buf[512];
2938 	uint64_t mflags;
2939 	u_int flags;
2940 
2941 	if (!have_addr) {
2942 		/* No address given, print short info about all mount points. */
2943 		TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2944 			db_printf("%p %s on %s (%s)\n", mp,
2945 			    mp->mnt_stat.f_mntfromname,
2946 			    mp->mnt_stat.f_mntonname,
2947 			    mp->mnt_stat.f_fstypename);
2948 			if (db_pager_quit)
2949 				break;
2950 		}
2951 		db_printf("\nMore info: show mount <addr>\n");
2952 		return;
2953 	}
2954 
2955 	mp = (struct mount *)addr;
2956 	db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
2957 	    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
2958 
2959 	buf[0] = '\0';
2960 	mflags = mp->mnt_flag;
2961 #define	MNT_FLAG(flag)	do {						\
2962 	if (mflags & (flag)) {						\
2963 		if (buf[0] != '\0')					\
2964 			strlcat(buf, ", ", sizeof(buf));		\
2965 		strlcat(buf, (#flag) + 4, sizeof(buf));			\
2966 		mflags &= ~(flag);					\
2967 	}								\
2968 } while (0)
2969 	MNT_FLAG(MNT_RDONLY);
2970 	MNT_FLAG(MNT_SYNCHRONOUS);
2971 	MNT_FLAG(MNT_NOEXEC);
2972 	MNT_FLAG(MNT_NOSUID);
2973 	MNT_FLAG(MNT_UNION);
2974 	MNT_FLAG(MNT_ASYNC);
2975 	MNT_FLAG(MNT_SUIDDIR);
2976 	MNT_FLAG(MNT_SOFTDEP);
2977 	MNT_FLAG(MNT_SUJ);
2978 	MNT_FLAG(MNT_NOSYMFOLLOW);
2979 	MNT_FLAG(MNT_GJOURNAL);
2980 	MNT_FLAG(MNT_MULTILABEL);
2981 	MNT_FLAG(MNT_ACLS);
2982 	MNT_FLAG(MNT_NOATIME);
2983 	MNT_FLAG(MNT_NOCLUSTERR);
2984 	MNT_FLAG(MNT_NOCLUSTERW);
2985 	MNT_FLAG(MNT_NFS4ACLS);
2986 	MNT_FLAG(MNT_EXRDONLY);
2987 	MNT_FLAG(MNT_EXPORTED);
2988 	MNT_FLAG(MNT_DEFEXPORTED);
2989 	MNT_FLAG(MNT_EXPORTANON);
2990 	MNT_FLAG(MNT_EXKERB);
2991 	MNT_FLAG(MNT_EXPUBLIC);
2992 	MNT_FLAG(MNT_LOCAL);
2993 	MNT_FLAG(MNT_QUOTA);
2994 	MNT_FLAG(MNT_ROOTFS);
2995 	MNT_FLAG(MNT_USER);
2996 	MNT_FLAG(MNT_IGNORE);
2997 	MNT_FLAG(MNT_UPDATE);
2998 	MNT_FLAG(MNT_DELEXPORT);
2999 	MNT_FLAG(MNT_RELOAD);
3000 	MNT_FLAG(MNT_FORCE);
3001 	MNT_FLAG(MNT_SNAPSHOT);
3002 	MNT_FLAG(MNT_BYFSID);
3003 #undef MNT_FLAG
3004 	if (mflags != 0) {
3005 		if (buf[0] != '\0')
3006 			strlcat(buf, ", ", sizeof(buf));
3007 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3008 		    "0x%016jx", mflags);
3009 	}
3010 	db_printf("    mnt_flag = %s\n", buf);
3011 
3012 	buf[0] = '\0';
3013 	flags = mp->mnt_kern_flag;
3014 #define	MNT_KERN_FLAG(flag)	do {					\
3015 	if (flags & (flag)) {						\
3016 		if (buf[0] != '\0')					\
3017 			strlcat(buf, ", ", sizeof(buf));		\
3018 		strlcat(buf, (#flag) + 5, sizeof(buf));			\
3019 		flags &= ~(flag);					\
3020 	}								\
3021 } while (0)
3022 	MNT_KERN_FLAG(MNTK_UNMOUNTF);
3023 	MNT_KERN_FLAG(MNTK_ASYNC);
3024 	MNT_KERN_FLAG(MNTK_SOFTDEP);
3025 	MNT_KERN_FLAG(MNTK_NOINSMNTQ);
3026 	MNT_KERN_FLAG(MNTK_DRAINING);
3027 	MNT_KERN_FLAG(MNTK_REFEXPIRE);
3028 	MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
3029 	MNT_KERN_FLAG(MNTK_SHARED_WRITES);
3030 	MNT_KERN_FLAG(MNTK_NOASYNC);
3031 	MNT_KERN_FLAG(MNTK_UNMOUNT);
3032 	MNT_KERN_FLAG(MNTK_MWAIT);
3033 	MNT_KERN_FLAG(MNTK_SUSPEND);
3034 	MNT_KERN_FLAG(MNTK_SUSPEND2);
3035 	MNT_KERN_FLAG(MNTK_SUSPENDED);
3036 	MNT_KERN_FLAG(MNTK_MPSAFE);
3037 	MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
3038 	MNT_KERN_FLAG(MNTK_NOKNOTE);
3039 #undef MNT_KERN_FLAG
3040 	if (flags != 0) {
3041 		if (buf[0] != '\0')
3042 			strlcat(buf, ", ", sizeof(buf));
3043 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3044 		    "0x%08x", flags);
3045 	}
3046 	db_printf("    mnt_kern_flag = %s\n", buf);
3047 
3048 	db_printf("    mnt_opt = ");
3049 	opt = TAILQ_FIRST(mp->mnt_opt);
3050 	if (opt != NULL) {
3051 		db_printf("%s", opt->name);
3052 		opt = TAILQ_NEXT(opt, link);
3053 		while (opt != NULL) {
3054 			db_printf(", %s", opt->name);
3055 			opt = TAILQ_NEXT(opt, link);
3056 		}
3057 	}
3058 	db_printf("\n");
3059 
3060 	sp = &mp->mnt_stat;
3061 	db_printf("    mnt_stat = { version=%u type=%u flags=0x%016jx "
3062 	    "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
3063 	    "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
3064 	    "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
3065 	    (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
3066 	    (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
3067 	    (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
3068 	    (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
3069 	    (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
3070 	    (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
3071 	    (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
3072 	    (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
3073 
3074 	db_printf("    mnt_cred = { uid=%u ruid=%u",
3075 	    (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
3076 	if (jailed(mp->mnt_cred))
3077 		db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
3078 	db_printf(" }\n");
3079 	db_printf("    mnt_ref = %d\n", mp->mnt_ref);
3080 	db_printf("    mnt_gen = %d\n", mp->mnt_gen);
3081 	db_printf("    mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
3082 	db_printf("    mnt_activevnodelistsize = %d\n",
3083 	    mp->mnt_activevnodelistsize);
3084 	db_printf("    mnt_writeopcount = %d\n", mp->mnt_writeopcount);
3085 	db_printf("    mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
3086 	db_printf("    mnt_iosize_max = %d\n", mp->mnt_iosize_max);
3087 	db_printf("    mnt_hashseed = %u\n", mp->mnt_hashseed);
3088 	db_printf("    mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
3089 	db_printf("    mnt_secondary_accwrites = %d\n",
3090 	    mp->mnt_secondary_accwrites);
3091 	db_printf("    mnt_gjprovider = %s\n",
3092 	    mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
3093 
3094 	db_printf("\n\nList of active vnodes\n");
3095 	TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) {
3096 		if (vp->v_type != VMARKER) {
3097 			vn_printf(vp, "vnode ");
3098 			if (db_pager_quit)
3099 				break;
3100 		}
3101 	}
3102 	db_printf("\n\nList of inactive vnodes\n");
3103 	TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3104 		if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) {
3105 			vn_printf(vp, "vnode ");
3106 			if (db_pager_quit)
3107 				break;
3108 		}
3109 	}
3110 }
3111 #endif	/* DDB */
3112 
3113 /*
3114  * Fill in a struct xvfsconf based on a struct vfsconf.
3115  */
3116 static void
3117 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
3118 {
3119 
3120 	strcpy(xvfsp->vfc_name, vfsp->vfc_name);
3121 	xvfsp->vfc_typenum = vfsp->vfc_typenum;
3122 	xvfsp->vfc_refcount = vfsp->vfc_refcount;
3123 	xvfsp->vfc_flags = vfsp->vfc_flags;
3124 	/*
3125 	 * These are unused in userland, we keep them
3126 	 * to not break binary compatibility.
3127 	 */
3128 	xvfsp->vfc_vfsops = NULL;
3129 	xvfsp->vfc_next = NULL;
3130 }
3131 
3132 /*
3133  * Top level filesystem related information gathering.
3134  */
3135 static int
3136 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
3137 {
3138 	struct vfsconf *vfsp;
3139 	struct xvfsconf xvfsp;
3140 	int error;
3141 
3142 	error = 0;
3143 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3144 		bzero(&xvfsp, sizeof(xvfsp));
3145 		vfsconf2x(vfsp, &xvfsp);
3146 		error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
3147 		if (error)
3148 			break;
3149 	}
3150 	return (error);
3151 }
3152 
3153 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD,
3154     NULL, 0, sysctl_vfs_conflist,
3155     "S,xvfsconf", "List of all configured filesystems");
3156 
3157 #ifndef BURN_BRIDGES
3158 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
3159 
3160 static int
3161 vfs_sysctl(SYSCTL_HANDLER_ARGS)
3162 {
3163 	int *name = (int *)arg1 - 1;	/* XXX */
3164 	u_int namelen = arg2 + 1;	/* XXX */
3165 	struct vfsconf *vfsp;
3166 	struct xvfsconf xvfsp;
3167 
3168 	log(LOG_WARNING, "userland calling deprecated sysctl, "
3169 	    "please rebuild world\n");
3170 
3171 #if 1 || defined(COMPAT_PRELITE2)
3172 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3173 	if (namelen == 1)
3174 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3175 #endif
3176 
3177 	switch (name[1]) {
3178 	case VFS_MAXTYPENUM:
3179 		if (namelen != 2)
3180 			return (ENOTDIR);
3181 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3182 	case VFS_CONF:
3183 		if (namelen != 3)
3184 			return (ENOTDIR);	/* overloaded */
3185 		TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
3186 			if (vfsp->vfc_typenum == name[2])
3187 				break;
3188 		if (vfsp == NULL)
3189 			return (EOPNOTSUPP);
3190 		bzero(&xvfsp, sizeof(xvfsp));
3191 		vfsconf2x(vfsp, &xvfsp);
3192 		return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3193 	}
3194 	return (EOPNOTSUPP);
3195 }
3196 
3197 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
3198     vfs_sysctl, "Generic filesystem");
3199 
3200 #if 1 || defined(COMPAT_PRELITE2)
3201 
3202 static int
3203 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3204 {
3205 	int error;
3206 	struct vfsconf *vfsp;
3207 	struct ovfsconf ovfs;
3208 
3209 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3210 		bzero(&ovfs, sizeof(ovfs));
3211 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
3212 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
3213 		ovfs.vfc_index = vfsp->vfc_typenum;
3214 		ovfs.vfc_refcount = vfsp->vfc_refcount;
3215 		ovfs.vfc_flags = vfsp->vfc_flags;
3216 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3217 		if (error)
3218 			return error;
3219 	}
3220 	return 0;
3221 }
3222 
3223 #endif /* 1 || COMPAT_PRELITE2 */
3224 #endif /* !BURN_BRIDGES */
3225 
3226 #define KINFO_VNODESLOP		10
3227 #ifdef notyet
3228 /*
3229  * Dump vnode list (via sysctl).
3230  */
3231 /* ARGSUSED */
3232 static int
3233 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3234 {
3235 	struct xvnode *xvn;
3236 	struct mount *mp;
3237 	struct vnode *vp;
3238 	int error, len, n;
3239 
3240 	/*
3241 	 * Stale numvnodes access is not fatal here.
3242 	 */
3243 	req->lock = 0;
3244 	len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3245 	if (!req->oldptr)
3246 		/* Make an estimate */
3247 		return (SYSCTL_OUT(req, 0, len));
3248 
3249 	error = sysctl_wire_old_buffer(req, 0);
3250 	if (error != 0)
3251 		return (error);
3252 	xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3253 	n = 0;
3254 	mtx_lock(&mountlist_mtx);
3255 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3256 		if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3257 			continue;
3258 		MNT_ILOCK(mp);
3259 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3260 			if (n == len)
3261 				break;
3262 			vref(vp);
3263 			xvn[n].xv_size = sizeof *xvn;
3264 			xvn[n].xv_vnode = vp;
3265 			xvn[n].xv_id = 0;	/* XXX compat */
3266 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3267 			XV_COPY(usecount);
3268 			XV_COPY(writecount);
3269 			XV_COPY(holdcnt);
3270 			XV_COPY(mount);
3271 			XV_COPY(numoutput);
3272 			XV_COPY(type);
3273 #undef XV_COPY
3274 			xvn[n].xv_flag = vp->v_vflag;
3275 
3276 			switch (vp->v_type) {
3277 			case VREG:
3278 			case VDIR:
3279 			case VLNK:
3280 				break;
3281 			case VBLK:
3282 			case VCHR:
3283 				if (vp->v_rdev == NULL) {
3284 					vrele(vp);
3285 					continue;
3286 				}
3287 				xvn[n].xv_dev = dev2udev(vp->v_rdev);
3288 				break;
3289 			case VSOCK:
3290 				xvn[n].xv_socket = vp->v_socket;
3291 				break;
3292 			case VFIFO:
3293 				xvn[n].xv_fifo = vp->v_fifoinfo;
3294 				break;
3295 			case VNON:
3296 			case VBAD:
3297 			default:
3298 				/* shouldn't happen? */
3299 				vrele(vp);
3300 				continue;
3301 			}
3302 			vrele(vp);
3303 			++n;
3304 		}
3305 		MNT_IUNLOCK(mp);
3306 		mtx_lock(&mountlist_mtx);
3307 		vfs_unbusy(mp);
3308 		if (n == len)
3309 			break;
3310 	}
3311 	mtx_unlock(&mountlist_mtx);
3312 
3313 	error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3314 	free(xvn, M_TEMP);
3315 	return (error);
3316 }
3317 
3318 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3319     0, 0, sysctl_vnode, "S,xvnode", "");
3320 #endif
3321 
3322 /*
3323  * Unmount all filesystems. The list is traversed in reverse order
3324  * of mounting to avoid dependencies.
3325  */
3326 void
3327 vfs_unmountall(void)
3328 {
3329 	struct mount *mp;
3330 	struct thread *td;
3331 	int error;
3332 
3333 	KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
3334 	CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3335 	td = curthread;
3336 
3337 	/*
3338 	 * Since this only runs when rebooting, it is not interlocked.
3339 	 */
3340 	while(!TAILQ_EMPTY(&mountlist)) {
3341 		mp = TAILQ_LAST(&mountlist, mntlist);
3342 		error = dounmount(mp, MNT_FORCE, td);
3343 		if (error) {
3344 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
3345 			/*
3346 			 * XXX: Due to the way in which we mount the root
3347 			 * file system off of devfs, devfs will generate a
3348 			 * "busy" warning when we try to unmount it before
3349 			 * the root.  Don't print a warning as a result in
3350 			 * order to avoid false positive errors that may
3351 			 * cause needless upset.
3352 			 */
3353 			if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3354 				printf("unmount of %s failed (",
3355 				    mp->mnt_stat.f_mntonname);
3356 				if (error == EBUSY)
3357 					printf("BUSY)\n");
3358 				else
3359 					printf("%d)\n", error);
3360 			}
3361 		} else {
3362 			/* The unmount has removed mp from the mountlist */
3363 		}
3364 	}
3365 }
3366 
3367 /*
3368  * perform msync on all vnodes under a mount point
3369  * the mount point must be locked.
3370  */
3371 void
3372 vfs_msync(struct mount *mp, int flags)
3373 {
3374 	struct vnode *vp, *mvp;
3375 	struct vm_object *obj;
3376 
3377 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3378 	MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
3379 		obj = vp->v_object;
3380 		if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3381 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3382 			if (!vget(vp,
3383 			    LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3384 			    curthread)) {
3385 				if (vp->v_vflag & VV_NOSYNC) {	/* unlinked */
3386 					vput(vp);
3387 					continue;
3388 				}
3389 
3390 				obj = vp->v_object;
3391 				if (obj != NULL) {
3392 					VM_OBJECT_LOCK(obj);
3393 					vm_object_page_clean(obj, 0, 0,
3394 					    flags == MNT_WAIT ?
3395 					    OBJPC_SYNC : OBJPC_NOSYNC);
3396 					VM_OBJECT_UNLOCK(obj);
3397 				}
3398 				vput(vp);
3399 			}
3400 		} else
3401 			VI_UNLOCK(vp);
3402 	}
3403 }
3404 
3405 static void
3406 destroy_vpollinfo(struct vpollinfo *vi)
3407 {
3408 	seldrain(&vi->vpi_selinfo);
3409 	knlist_destroy(&vi->vpi_selinfo.si_note);
3410 	mtx_destroy(&vi->vpi_lock);
3411 	uma_zfree(vnodepoll_zone, vi);
3412 }
3413 
3414 /*
3415  * Initalize per-vnode helper structure to hold poll-related state.
3416  */
3417 void
3418 v_addpollinfo(struct vnode *vp)
3419 {
3420 	struct vpollinfo *vi;
3421 
3422 	if (vp->v_pollinfo != NULL)
3423 		return;
3424 	vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3425 	mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3426 	knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3427 	    vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3428 	VI_LOCK(vp);
3429 	if (vp->v_pollinfo != NULL) {
3430 		VI_UNLOCK(vp);
3431 		destroy_vpollinfo(vi);
3432 		return;
3433 	}
3434 	vp->v_pollinfo = vi;
3435 	VI_UNLOCK(vp);
3436 }
3437 
3438 /*
3439  * Record a process's interest in events which might happen to
3440  * a vnode.  Because poll uses the historic select-style interface
3441  * internally, this routine serves as both the ``check for any
3442  * pending events'' and the ``record my interest in future events''
3443  * functions.  (These are done together, while the lock is held,
3444  * to avoid race conditions.)
3445  */
3446 int
3447 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3448 {
3449 
3450 	v_addpollinfo(vp);
3451 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3452 	if (vp->v_pollinfo->vpi_revents & events) {
3453 		/*
3454 		 * This leaves events we are not interested
3455 		 * in available for the other process which
3456 		 * which presumably had requested them
3457 		 * (otherwise they would never have been
3458 		 * recorded).
3459 		 */
3460 		events &= vp->v_pollinfo->vpi_revents;
3461 		vp->v_pollinfo->vpi_revents &= ~events;
3462 
3463 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
3464 		return (events);
3465 	}
3466 	vp->v_pollinfo->vpi_events |= events;
3467 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3468 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3469 	return (0);
3470 }
3471 
3472 /*
3473  * Routine to create and manage a filesystem syncer vnode.
3474  */
3475 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
3476 static int	sync_fsync(struct  vop_fsync_args *);
3477 static int	sync_inactive(struct  vop_inactive_args *);
3478 static int	sync_reclaim(struct  vop_reclaim_args *);
3479 
3480 static struct vop_vector sync_vnodeops = {
3481 	.vop_bypass =	VOP_EOPNOTSUPP,
3482 	.vop_close =	sync_close,		/* close */
3483 	.vop_fsync =	sync_fsync,		/* fsync */
3484 	.vop_inactive =	sync_inactive,	/* inactive */
3485 	.vop_reclaim =	sync_reclaim,	/* reclaim */
3486 	.vop_lock1 =	vop_stdlock,	/* lock */
3487 	.vop_unlock =	vop_stdunlock,	/* unlock */
3488 	.vop_islocked =	vop_stdislocked,	/* islocked */
3489 };
3490 
3491 /*
3492  * Create a new filesystem syncer vnode for the specified mount point.
3493  */
3494 void
3495 vfs_allocate_syncvnode(struct mount *mp)
3496 {
3497 	struct vnode *vp;
3498 	struct bufobj *bo;
3499 	static long start, incr, next;
3500 	int error;
3501 
3502 	/* Allocate a new vnode */
3503 	error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
3504 	if (error != 0)
3505 		panic("vfs_allocate_syncvnode: getnewvnode() failed");
3506 	vp->v_type = VNON;
3507 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3508 	vp->v_vflag |= VV_FORCEINSMQ;
3509 	error = insmntque(vp, mp);
3510 	if (error != 0)
3511 		panic("vfs_allocate_syncvnode: insmntque() failed");
3512 	vp->v_vflag &= ~VV_FORCEINSMQ;
3513 	VOP_UNLOCK(vp, 0);
3514 	/*
3515 	 * Place the vnode onto the syncer worklist. We attempt to
3516 	 * scatter them about on the list so that they will go off
3517 	 * at evenly distributed times even if all the filesystems
3518 	 * are mounted at once.
3519 	 */
3520 	next += incr;
3521 	if (next == 0 || next > syncer_maxdelay) {
3522 		start /= 2;
3523 		incr /= 2;
3524 		if (start == 0) {
3525 			start = syncer_maxdelay / 2;
3526 			incr = syncer_maxdelay;
3527 		}
3528 		next = start;
3529 	}
3530 	bo = &vp->v_bufobj;
3531 	BO_LOCK(bo);
3532 	vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3533 	/* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3534 	mtx_lock(&sync_mtx);
3535 	sync_vnode_count++;
3536 	if (mp->mnt_syncer == NULL) {
3537 		mp->mnt_syncer = vp;
3538 		vp = NULL;
3539 	}
3540 	mtx_unlock(&sync_mtx);
3541 	BO_UNLOCK(bo);
3542 	if (vp != NULL) {
3543 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3544 		vgone(vp);
3545 		vput(vp);
3546 	}
3547 }
3548 
3549 void
3550 vfs_deallocate_syncvnode(struct mount *mp)
3551 {
3552 	struct vnode *vp;
3553 
3554 	mtx_lock(&sync_mtx);
3555 	vp = mp->mnt_syncer;
3556 	if (vp != NULL)
3557 		mp->mnt_syncer = NULL;
3558 	mtx_unlock(&sync_mtx);
3559 	if (vp != NULL)
3560 		vrele(vp);
3561 }
3562 
3563 /*
3564  * Do a lazy sync of the filesystem.
3565  */
3566 static int
3567 sync_fsync(struct vop_fsync_args *ap)
3568 {
3569 	struct vnode *syncvp = ap->a_vp;
3570 	struct mount *mp = syncvp->v_mount;
3571 	int error, save;
3572 	struct bufobj *bo;
3573 
3574 	/*
3575 	 * We only need to do something if this is a lazy evaluation.
3576 	 */
3577 	if (ap->a_waitfor != MNT_LAZY)
3578 		return (0);
3579 
3580 	/*
3581 	 * Move ourselves to the back of the sync list.
3582 	 */
3583 	bo = &syncvp->v_bufobj;
3584 	BO_LOCK(bo);
3585 	vn_syncer_add_to_worklist(bo, syncdelay);
3586 	BO_UNLOCK(bo);
3587 
3588 	/*
3589 	 * Walk the list of vnodes pushing all that are dirty and
3590 	 * not already on the sync list.
3591 	 */
3592 	mtx_lock(&mountlist_mtx);
3593 	if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) {
3594 		mtx_unlock(&mountlist_mtx);
3595 		return (0);
3596 	}
3597 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3598 		vfs_unbusy(mp);
3599 		return (0);
3600 	}
3601 	save = curthread_pflags_set(TDP_SYNCIO);
3602 	vfs_msync(mp, MNT_NOWAIT);
3603 	error = VFS_SYNC(mp, MNT_LAZY);
3604 	curthread_pflags_restore(save);
3605 	vn_finished_write(mp);
3606 	vfs_unbusy(mp);
3607 	return (error);
3608 }
3609 
3610 /*
3611  * The syncer vnode is no referenced.
3612  */
3613 static int
3614 sync_inactive(struct vop_inactive_args *ap)
3615 {
3616 
3617 	vgone(ap->a_vp);
3618 	return (0);
3619 }
3620 
3621 /*
3622  * The syncer vnode is no longer needed and is being decommissioned.
3623  *
3624  * Modifications to the worklist must be protected by sync_mtx.
3625  */
3626 static int
3627 sync_reclaim(struct vop_reclaim_args *ap)
3628 {
3629 	struct vnode *vp = ap->a_vp;
3630 	struct bufobj *bo;
3631 
3632 	bo = &vp->v_bufobj;
3633 	BO_LOCK(bo);
3634 	mtx_lock(&sync_mtx);
3635 	if (vp->v_mount->mnt_syncer == vp)
3636 		vp->v_mount->mnt_syncer = NULL;
3637 	if (bo->bo_flag & BO_ONWORKLST) {
3638 		LIST_REMOVE(bo, bo_synclist);
3639 		syncer_worklist_len--;
3640 		sync_vnode_count--;
3641 		bo->bo_flag &= ~BO_ONWORKLST;
3642 	}
3643 	mtx_unlock(&sync_mtx);
3644 	BO_UNLOCK(bo);
3645 
3646 	return (0);
3647 }
3648 
3649 /*
3650  * Check if vnode represents a disk device
3651  */
3652 int
3653 vn_isdisk(struct vnode *vp, int *errp)
3654 {
3655 	int error;
3656 
3657 	error = 0;
3658 	dev_lock();
3659 	if (vp->v_type != VCHR)
3660 		error = ENOTBLK;
3661 	else if (vp->v_rdev == NULL)
3662 		error = ENXIO;
3663 	else if (vp->v_rdev->si_devsw == NULL)
3664 		error = ENXIO;
3665 	else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3666 		error = ENOTBLK;
3667 	dev_unlock();
3668 	if (errp != NULL)
3669 		*errp = error;
3670 	return (error == 0);
3671 }
3672 
3673 /*
3674  * Common filesystem object access control check routine.  Accepts a
3675  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3676  * and optional call-by-reference privused argument allowing vaccess()
3677  * to indicate to the caller whether privilege was used to satisfy the
3678  * request (obsoleted).  Returns 0 on success, or an errno on failure.
3679  */
3680 int
3681 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3682     accmode_t accmode, struct ucred *cred, int *privused)
3683 {
3684 	accmode_t dac_granted;
3685 	accmode_t priv_granted;
3686 
3687 	KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3688 	    ("invalid bit in accmode"));
3689 	KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
3690 	    ("VAPPEND without VWRITE"));
3691 
3692 	/*
3693 	 * Look for a normal, non-privileged way to access the file/directory
3694 	 * as requested.  If it exists, go with that.
3695 	 */
3696 
3697 	if (privused != NULL)
3698 		*privused = 0;
3699 
3700 	dac_granted = 0;
3701 
3702 	/* Check the owner. */
3703 	if (cred->cr_uid == file_uid) {
3704 		dac_granted |= VADMIN;
3705 		if (file_mode & S_IXUSR)
3706 			dac_granted |= VEXEC;
3707 		if (file_mode & S_IRUSR)
3708 			dac_granted |= VREAD;
3709 		if (file_mode & S_IWUSR)
3710 			dac_granted |= (VWRITE | VAPPEND);
3711 
3712 		if ((accmode & dac_granted) == accmode)
3713 			return (0);
3714 
3715 		goto privcheck;
3716 	}
3717 
3718 	/* Otherwise, check the groups (first match) */
3719 	if (groupmember(file_gid, cred)) {
3720 		if (file_mode & S_IXGRP)
3721 			dac_granted |= VEXEC;
3722 		if (file_mode & S_IRGRP)
3723 			dac_granted |= VREAD;
3724 		if (file_mode & S_IWGRP)
3725 			dac_granted |= (VWRITE | VAPPEND);
3726 
3727 		if ((accmode & dac_granted) == accmode)
3728 			return (0);
3729 
3730 		goto privcheck;
3731 	}
3732 
3733 	/* Otherwise, check everyone else. */
3734 	if (file_mode & S_IXOTH)
3735 		dac_granted |= VEXEC;
3736 	if (file_mode & S_IROTH)
3737 		dac_granted |= VREAD;
3738 	if (file_mode & S_IWOTH)
3739 		dac_granted |= (VWRITE | VAPPEND);
3740 	if ((accmode & dac_granted) == accmode)
3741 		return (0);
3742 
3743 privcheck:
3744 	/*
3745 	 * Build a privilege mask to determine if the set of privileges
3746 	 * satisfies the requirements when combined with the granted mask
3747 	 * from above.  For each privilege, if the privilege is required,
3748 	 * bitwise or the request type onto the priv_granted mask.
3749 	 */
3750 	priv_granted = 0;
3751 
3752 	if (type == VDIR) {
3753 		/*
3754 		 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3755 		 * requests, instead of PRIV_VFS_EXEC.
3756 		 */
3757 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3758 		    !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3759 			priv_granted |= VEXEC;
3760 	} else {
3761 		/*
3762 		 * Ensure that at least one execute bit is on. Otherwise,
3763 		 * a privileged user will always succeed, and we don't want
3764 		 * this to happen unless the file really is executable.
3765 		 */
3766 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3767 		    (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
3768 		    !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3769 			priv_granted |= VEXEC;
3770 	}
3771 
3772 	if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
3773 	    !priv_check_cred(cred, PRIV_VFS_READ, 0))
3774 		priv_granted |= VREAD;
3775 
3776 	if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3777 	    !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3778 		priv_granted |= (VWRITE | VAPPEND);
3779 
3780 	if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3781 	    !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3782 		priv_granted |= VADMIN;
3783 
3784 	if ((accmode & (priv_granted | dac_granted)) == accmode) {
3785 		/* XXX audit: privilege used */
3786 		if (privused != NULL)
3787 			*privused = 1;
3788 		return (0);
3789 	}
3790 
3791 	return ((accmode & VADMIN) ? EPERM : EACCES);
3792 }
3793 
3794 /*
3795  * Credential check based on process requesting service, and per-attribute
3796  * permissions.
3797  */
3798 int
3799 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3800     struct thread *td, accmode_t accmode)
3801 {
3802 
3803 	/*
3804 	 * Kernel-invoked always succeeds.
3805 	 */
3806 	if (cred == NOCRED)
3807 		return (0);
3808 
3809 	/*
3810 	 * Do not allow privileged processes in jail to directly manipulate
3811 	 * system attributes.
3812 	 */
3813 	switch (attrnamespace) {
3814 	case EXTATTR_NAMESPACE_SYSTEM:
3815 		/* Potentially should be: return (EPERM); */
3816 		return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3817 	case EXTATTR_NAMESPACE_USER:
3818 		return (VOP_ACCESS(vp, accmode, cred, td));
3819 	default:
3820 		return (EPERM);
3821 	}
3822 }
3823 
3824 #ifdef DEBUG_VFS_LOCKS
3825 /*
3826  * This only exists to supress warnings from unlocked specfs accesses.  It is
3827  * no longer ok to have an unlocked VFS.
3828  */
3829 #define	IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL ||		\
3830 	(vp)->v_type == VCHR ||	(vp)->v_type == VBAD)
3831 
3832 int vfs_badlock_ddb = 1;	/* Drop into debugger on violation. */
3833 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
3834     "Drop into debugger on lock violation");
3835 
3836 int vfs_badlock_mutex = 1;	/* Check for interlock across VOPs. */
3837 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
3838     0, "Check for interlock across VOPs");
3839 
3840 int vfs_badlock_print = 1;	/* Print lock violations. */
3841 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
3842     0, "Print lock violations");
3843 
3844 #ifdef KDB
3845 int vfs_badlock_backtrace = 1;	/* Print backtrace at lock violations. */
3846 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
3847     &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
3848 #endif
3849 
3850 static void
3851 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3852 {
3853 
3854 #ifdef KDB
3855 	if (vfs_badlock_backtrace)
3856 		kdb_backtrace();
3857 #endif
3858 	if (vfs_badlock_print)
3859 		printf("%s: %p %s\n", str, (void *)vp, msg);
3860 	if (vfs_badlock_ddb)
3861 		kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3862 }
3863 
3864 void
3865 assert_vi_locked(struct vnode *vp, const char *str)
3866 {
3867 
3868 	if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3869 		vfs_badlock("interlock is not locked but should be", str, vp);
3870 }
3871 
3872 void
3873 assert_vi_unlocked(struct vnode *vp, const char *str)
3874 {
3875 
3876 	if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3877 		vfs_badlock("interlock is locked but should not be", str, vp);
3878 }
3879 
3880 void
3881 assert_vop_locked(struct vnode *vp, const char *str)
3882 {
3883 
3884 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0)
3885 		vfs_badlock("is not locked but should be", str, vp);
3886 }
3887 
3888 void
3889 assert_vop_unlocked(struct vnode *vp, const char *str)
3890 {
3891 
3892 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
3893 		vfs_badlock("is locked but should not be", str, vp);
3894 }
3895 
3896 void
3897 assert_vop_elocked(struct vnode *vp, const char *str)
3898 {
3899 
3900 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
3901 		vfs_badlock("is not exclusive locked but should be", str, vp);
3902 }
3903 
3904 #if 0
3905 void
3906 assert_vop_elocked_other(struct vnode *vp, const char *str)
3907 {
3908 
3909 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
3910 		vfs_badlock("is not exclusive locked by another thread",
3911 		    str, vp);
3912 }
3913 
3914 void
3915 assert_vop_slocked(struct vnode *vp, const char *str)
3916 {
3917 
3918 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
3919 		vfs_badlock("is not locked shared but should be", str, vp);
3920 }
3921 #endif /* 0 */
3922 #endif /* DEBUG_VFS_LOCKS */
3923 
3924 void
3925 vop_rename_fail(struct vop_rename_args *ap)
3926 {
3927 
3928 	if (ap->a_tvp != NULL)
3929 		vput(ap->a_tvp);
3930 	if (ap->a_tdvp == ap->a_tvp)
3931 		vrele(ap->a_tdvp);
3932 	else
3933 		vput(ap->a_tdvp);
3934 	vrele(ap->a_fdvp);
3935 	vrele(ap->a_fvp);
3936 }
3937 
3938 void
3939 vop_rename_pre(void *ap)
3940 {
3941 	struct vop_rename_args *a = ap;
3942 
3943 #ifdef DEBUG_VFS_LOCKS
3944 	if (a->a_tvp)
3945 		ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
3946 	ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
3947 	ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
3948 	ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
3949 
3950 	/* Check the source (from). */
3951 	if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
3952 	    (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
3953 		ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3954 	if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
3955 		ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
3956 
3957 	/* Check the target. */
3958 	if (a->a_tvp)
3959 		ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3960 	ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3961 #endif
3962 	if (a->a_tdvp != a->a_fdvp)
3963 		vhold(a->a_fdvp);
3964 	if (a->a_tvp != a->a_fvp)
3965 		vhold(a->a_fvp);
3966 	vhold(a->a_tdvp);
3967 	if (a->a_tvp)
3968 		vhold(a->a_tvp);
3969 }
3970 
3971 void
3972 vop_strategy_pre(void *ap)
3973 {
3974 #ifdef DEBUG_VFS_LOCKS
3975 	struct vop_strategy_args *a;
3976 	struct buf *bp;
3977 
3978 	a = ap;
3979 	bp = a->a_bp;
3980 
3981 	/*
3982 	 * Cluster ops lock their component buffers but not the IO container.
3983 	 */
3984 	if ((bp->b_flags & B_CLUSTER) != 0)
3985 		return;
3986 
3987 	if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
3988 		if (vfs_badlock_print)
3989 			printf(
3990 			    "VOP_STRATEGY: bp is not locked but should be\n");
3991 		if (vfs_badlock_ddb)
3992 			kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3993 	}
3994 #endif
3995 }
3996 
3997 void
3998 vop_lookup_pre(void *ap)
3999 {
4000 #ifdef DEBUG_VFS_LOCKS
4001 	struct vop_lookup_args *a;
4002 	struct vnode *dvp;
4003 
4004 	a = ap;
4005 	dvp = a->a_dvp;
4006 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
4007 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
4008 #endif
4009 }
4010 
4011 void
4012 vop_lookup_post(void *ap, int rc)
4013 {
4014 #ifdef DEBUG_VFS_LOCKS
4015 	struct vop_lookup_args *a;
4016 	struct vnode *dvp;
4017 	struct vnode *vp;
4018 
4019 	a = ap;
4020 	dvp = a->a_dvp;
4021 	vp = *(a->a_vpp);
4022 
4023 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
4024 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
4025 
4026 	if (!rc)
4027 		ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
4028 #endif
4029 }
4030 
4031 void
4032 vop_lock_pre(void *ap)
4033 {
4034 #ifdef DEBUG_VFS_LOCKS
4035 	struct vop_lock1_args *a = ap;
4036 
4037 	if ((a->a_flags & LK_INTERLOCK) == 0)
4038 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4039 	else
4040 		ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
4041 #endif
4042 }
4043 
4044 void
4045 vop_lock_post(void *ap, int rc)
4046 {
4047 #ifdef DEBUG_VFS_LOCKS
4048 	struct vop_lock1_args *a = ap;
4049 
4050 	ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4051 	if (rc == 0)
4052 		ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
4053 #endif
4054 }
4055 
4056 void
4057 vop_unlock_pre(void *ap)
4058 {
4059 #ifdef DEBUG_VFS_LOCKS
4060 	struct vop_unlock_args *a = ap;
4061 
4062 	if (a->a_flags & LK_INTERLOCK)
4063 		ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
4064 	ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
4065 #endif
4066 }
4067 
4068 void
4069 vop_unlock_post(void *ap, int rc)
4070 {
4071 #ifdef DEBUG_VFS_LOCKS
4072 	struct vop_unlock_args *a = ap;
4073 
4074 	if (a->a_flags & LK_INTERLOCK)
4075 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
4076 #endif
4077 }
4078 
4079 void
4080 vop_create_post(void *ap, int rc)
4081 {
4082 	struct vop_create_args *a = ap;
4083 
4084 	if (!rc)
4085 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4086 }
4087 
4088 void
4089 vop_deleteextattr_post(void *ap, int rc)
4090 {
4091 	struct vop_deleteextattr_args *a = ap;
4092 
4093 	if (!rc)
4094 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4095 }
4096 
4097 void
4098 vop_link_post(void *ap, int rc)
4099 {
4100 	struct vop_link_args *a = ap;
4101 
4102 	if (!rc) {
4103 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
4104 		VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
4105 	}
4106 }
4107 
4108 void
4109 vop_mkdir_post(void *ap, int rc)
4110 {
4111 	struct vop_mkdir_args *a = ap;
4112 
4113 	if (!rc)
4114 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4115 }
4116 
4117 void
4118 vop_mknod_post(void *ap, int rc)
4119 {
4120 	struct vop_mknod_args *a = ap;
4121 
4122 	if (!rc)
4123 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4124 }
4125 
4126 void
4127 vop_remove_post(void *ap, int rc)
4128 {
4129 	struct vop_remove_args *a = ap;
4130 
4131 	if (!rc) {
4132 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4133 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4134 	}
4135 }
4136 
4137 void
4138 vop_rename_post(void *ap, int rc)
4139 {
4140 	struct vop_rename_args *a = ap;
4141 
4142 	if (!rc) {
4143 		VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
4144 		VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
4145 		VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
4146 		if (a->a_tvp)
4147 			VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
4148 	}
4149 	if (a->a_tdvp != a->a_fdvp)
4150 		vdrop(a->a_fdvp);
4151 	if (a->a_tvp != a->a_fvp)
4152 		vdrop(a->a_fvp);
4153 	vdrop(a->a_tdvp);
4154 	if (a->a_tvp)
4155 		vdrop(a->a_tvp);
4156 }
4157 
4158 void
4159 vop_rmdir_post(void *ap, int rc)
4160 {
4161 	struct vop_rmdir_args *a = ap;
4162 
4163 	if (!rc) {
4164 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4165 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4166 	}
4167 }
4168 
4169 void
4170 vop_setattr_post(void *ap, int rc)
4171 {
4172 	struct vop_setattr_args *a = ap;
4173 
4174 	if (!rc)
4175 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4176 }
4177 
4178 void
4179 vop_setextattr_post(void *ap, int rc)
4180 {
4181 	struct vop_setextattr_args *a = ap;
4182 
4183 	if (!rc)
4184 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4185 }
4186 
4187 void
4188 vop_symlink_post(void *ap, int rc)
4189 {
4190 	struct vop_symlink_args *a = ap;
4191 
4192 	if (!rc)
4193 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4194 }
4195 
4196 static struct knlist fs_knlist;
4197 
4198 static void
4199 vfs_event_init(void *arg)
4200 {
4201 	knlist_init_mtx(&fs_knlist, NULL);
4202 }
4203 /* XXX - correct order? */
4204 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4205 
4206 void
4207 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
4208 {
4209 
4210 	KNOTE_UNLOCKED(&fs_knlist, event);
4211 }
4212 
4213 static int	filt_fsattach(struct knote *kn);
4214 static void	filt_fsdetach(struct knote *kn);
4215 static int	filt_fsevent(struct knote *kn, long hint);
4216 
4217 struct filterops fs_filtops = {
4218 	.f_isfd = 0,
4219 	.f_attach = filt_fsattach,
4220 	.f_detach = filt_fsdetach,
4221 	.f_event = filt_fsevent
4222 };
4223 
4224 static int
4225 filt_fsattach(struct knote *kn)
4226 {
4227 
4228 	kn->kn_flags |= EV_CLEAR;
4229 	knlist_add(&fs_knlist, kn, 0);
4230 	return (0);
4231 }
4232 
4233 static void
4234 filt_fsdetach(struct knote *kn)
4235 {
4236 
4237 	knlist_remove(&fs_knlist, kn, 0);
4238 }
4239 
4240 static int
4241 filt_fsevent(struct knote *kn, long hint)
4242 {
4243 
4244 	kn->kn_fflags |= hint;
4245 	return (kn->kn_fflags != 0);
4246 }
4247 
4248 static int
4249 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4250 {
4251 	struct vfsidctl vc;
4252 	int error;
4253 	struct mount *mp;
4254 
4255 	error = SYSCTL_IN(req, &vc, sizeof(vc));
4256 	if (error)
4257 		return (error);
4258 	if (vc.vc_vers != VFS_CTL_VERS1)
4259 		return (EINVAL);
4260 	mp = vfs_getvfs(&vc.vc_fsid);
4261 	if (mp == NULL)
4262 		return (ENOENT);
4263 	/* ensure that a specific sysctl goes to the right filesystem. */
4264 	if (strcmp(vc.vc_fstypename, "*") != 0 &&
4265 	    strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4266 		vfs_rel(mp);
4267 		return (EINVAL);
4268 	}
4269 	VCTLTOREQ(&vc, req);
4270 	error = VFS_SYSCTL(mp, vc.vc_op, req);
4271 	vfs_rel(mp);
4272 	return (error);
4273 }
4274 
4275 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
4276     NULL, 0, sysctl_vfs_ctl, "",
4277     "Sysctl by fsid");
4278 
4279 /*
4280  * Function to initialize a va_filerev field sensibly.
4281  * XXX: Wouldn't a random number make a lot more sense ??
4282  */
4283 u_quad_t
4284 init_va_filerev(void)
4285 {
4286 	struct bintime bt;
4287 
4288 	getbinuptime(&bt);
4289 	return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4290 }
4291 
4292 static int	filt_vfsread(struct knote *kn, long hint);
4293 static int	filt_vfswrite(struct knote *kn, long hint);
4294 static int	filt_vfsvnode(struct knote *kn, long hint);
4295 static void	filt_vfsdetach(struct knote *kn);
4296 static struct filterops vfsread_filtops = {
4297 	.f_isfd = 1,
4298 	.f_detach = filt_vfsdetach,
4299 	.f_event = filt_vfsread
4300 };
4301 static struct filterops vfswrite_filtops = {
4302 	.f_isfd = 1,
4303 	.f_detach = filt_vfsdetach,
4304 	.f_event = filt_vfswrite
4305 };
4306 static struct filterops vfsvnode_filtops = {
4307 	.f_isfd = 1,
4308 	.f_detach = filt_vfsdetach,
4309 	.f_event = filt_vfsvnode
4310 };
4311 
4312 static void
4313 vfs_knllock(void *arg)
4314 {
4315 	struct vnode *vp = arg;
4316 
4317 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4318 }
4319 
4320 static void
4321 vfs_knlunlock(void *arg)
4322 {
4323 	struct vnode *vp = arg;
4324 
4325 	VOP_UNLOCK(vp, 0);
4326 }
4327 
4328 static void
4329 vfs_knl_assert_locked(void *arg)
4330 {
4331 #ifdef DEBUG_VFS_LOCKS
4332 	struct vnode *vp = arg;
4333 
4334 	ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4335 #endif
4336 }
4337 
4338 static void
4339 vfs_knl_assert_unlocked(void *arg)
4340 {
4341 #ifdef DEBUG_VFS_LOCKS
4342 	struct vnode *vp = arg;
4343 
4344 	ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4345 #endif
4346 }
4347 
4348 int
4349 vfs_kqfilter(struct vop_kqfilter_args *ap)
4350 {
4351 	struct vnode *vp = ap->a_vp;
4352 	struct knote *kn = ap->a_kn;
4353 	struct knlist *knl;
4354 
4355 	switch (kn->kn_filter) {
4356 	case EVFILT_READ:
4357 		kn->kn_fop = &vfsread_filtops;
4358 		break;
4359 	case EVFILT_WRITE:
4360 		kn->kn_fop = &vfswrite_filtops;
4361 		break;
4362 	case EVFILT_VNODE:
4363 		kn->kn_fop = &vfsvnode_filtops;
4364 		break;
4365 	default:
4366 		return (EINVAL);
4367 	}
4368 
4369 	kn->kn_hook = (caddr_t)vp;
4370 
4371 	v_addpollinfo(vp);
4372 	if (vp->v_pollinfo == NULL)
4373 		return (ENOMEM);
4374 	knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4375 	knlist_add(knl, kn, 0);
4376 
4377 	return (0);
4378 }
4379 
4380 /*
4381  * Detach knote from vnode
4382  */
4383 static void
4384 filt_vfsdetach(struct knote *kn)
4385 {
4386 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4387 
4388 	KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4389 	knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4390 }
4391 
4392 /*ARGSUSED*/
4393 static int
4394 filt_vfsread(struct knote *kn, long hint)
4395 {
4396 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4397 	struct vattr va;
4398 	int res;
4399 
4400 	/*
4401 	 * filesystem is gone, so set the EOF flag and schedule
4402 	 * the knote for deletion.
4403 	 */
4404 	if (hint == NOTE_REVOKE) {
4405 		VI_LOCK(vp);
4406 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4407 		VI_UNLOCK(vp);
4408 		return (1);
4409 	}
4410 
4411 	if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4412 		return (0);
4413 
4414 	VI_LOCK(vp);
4415 	kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4416 	res = (kn->kn_data != 0);
4417 	VI_UNLOCK(vp);
4418 	return (res);
4419 }
4420 
4421 /*ARGSUSED*/
4422 static int
4423 filt_vfswrite(struct knote *kn, long hint)
4424 {
4425 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4426 
4427 	VI_LOCK(vp);
4428 
4429 	/*
4430 	 * filesystem is gone, so set the EOF flag and schedule
4431 	 * the knote for deletion.
4432 	 */
4433 	if (hint == NOTE_REVOKE)
4434 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4435 
4436 	kn->kn_data = 0;
4437 	VI_UNLOCK(vp);
4438 	return (1);
4439 }
4440 
4441 static int
4442 filt_vfsvnode(struct knote *kn, long hint)
4443 {
4444 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4445 	int res;
4446 
4447 	VI_LOCK(vp);
4448 	if (kn->kn_sfflags & hint)
4449 		kn->kn_fflags |= hint;
4450 	if (hint == NOTE_REVOKE) {
4451 		kn->kn_flags |= EV_EOF;
4452 		VI_UNLOCK(vp);
4453 		return (1);
4454 	}
4455 	res = (kn->kn_fflags != 0);
4456 	VI_UNLOCK(vp);
4457 	return (res);
4458 }
4459 
4460 int
4461 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4462 {
4463 	int error;
4464 
4465 	if (dp->d_reclen > ap->a_uio->uio_resid)
4466 		return (ENAMETOOLONG);
4467 	error = uiomove(dp, dp->d_reclen, ap->a_uio);
4468 	if (error) {
4469 		if (ap->a_ncookies != NULL) {
4470 			if (ap->a_cookies != NULL)
4471 				free(ap->a_cookies, M_TEMP);
4472 			ap->a_cookies = NULL;
4473 			*ap->a_ncookies = 0;
4474 		}
4475 		return (error);
4476 	}
4477 	if (ap->a_ncookies == NULL)
4478 		return (0);
4479 
4480 	KASSERT(ap->a_cookies,
4481 	    ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4482 
4483 	*ap->a_cookies = realloc(*ap->a_cookies,
4484 	    (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4485 	(*ap->a_cookies)[*ap->a_ncookies] = off;
4486 	return (0);
4487 }
4488 
4489 /*
4490  * Mark for update the access time of the file if the filesystem
4491  * supports VOP_MARKATIME.  This functionality is used by execve and
4492  * mmap, so we want to avoid the I/O implied by directly setting
4493  * va_atime for the sake of efficiency.
4494  */
4495 void
4496 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4497 {
4498 	struct mount *mp;
4499 
4500 	mp = vp->v_mount;
4501 	VFS_ASSERT_GIANT(mp);
4502 	ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4503 	if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4504 		(void)VOP_MARKATIME(vp);
4505 }
4506 
4507 /*
4508  * The purpose of this routine is to remove granularity from accmode_t,
4509  * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4510  * VADMIN and VAPPEND.
4511  *
4512  * If it returns 0, the caller is supposed to continue with the usual
4513  * access checks using 'accmode' as modified by this routine.  If it
4514  * returns nonzero value, the caller is supposed to return that value
4515  * as errno.
4516  *
4517  * Note that after this routine runs, accmode may be zero.
4518  */
4519 int
4520 vfs_unixify_accmode(accmode_t *accmode)
4521 {
4522 	/*
4523 	 * There is no way to specify explicit "deny" rule using
4524 	 * file mode or POSIX.1e ACLs.
4525 	 */
4526 	if (*accmode & VEXPLICIT_DENY) {
4527 		*accmode = 0;
4528 		return (0);
4529 	}
4530 
4531 	/*
4532 	 * None of these can be translated into usual access bits.
4533 	 * Also, the common case for NFSv4 ACLs is to not contain
4534 	 * either of these bits. Caller should check for VWRITE
4535 	 * on the containing directory instead.
4536 	 */
4537 	if (*accmode & (VDELETE_CHILD | VDELETE))
4538 		return (EPERM);
4539 
4540 	if (*accmode & VADMIN_PERMS) {
4541 		*accmode &= ~VADMIN_PERMS;
4542 		*accmode |= VADMIN;
4543 	}
4544 
4545 	/*
4546 	 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4547 	 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4548 	 */
4549 	*accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
4550 
4551 	return (0);
4552 }
4553 
4554 /*
4555  * These are helper functions for filesystems to traverse all
4556  * their vnodes.  See MNT_VNODE_FOREACH_ALL() in sys/mount.h.
4557  *
4558  * This interface replaces MNT_VNODE_FOREACH.
4559  */
4560 
4561 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
4562 
4563 struct vnode *
4564 __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp)
4565 {
4566 	struct vnode *vp;
4567 
4568 	if (should_yield())
4569 		kern_yield(PRI_UNCHANGED);
4570 	MNT_ILOCK(mp);
4571 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4572 	vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
4573 	while (vp != NULL && (vp->v_type == VMARKER ||
4574 	    (vp->v_iflag & VI_DOOMED) != 0))
4575 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
4576 
4577 	/* Check if we are done */
4578 	if (vp == NULL) {
4579 		__mnt_vnode_markerfree_all(mvp, mp);
4580 		/* MNT_IUNLOCK(mp); -- done in above function */
4581 		mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
4582 		return (NULL);
4583 	}
4584 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4585 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4586 	VI_LOCK(vp);
4587 	MNT_IUNLOCK(mp);
4588 	return (vp);
4589 }
4590 
4591 struct vnode *
4592 __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp)
4593 {
4594 	struct vnode *vp;
4595 
4596 	*mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4597 	MNT_ILOCK(mp);
4598 	MNT_REF(mp);
4599 	(*mvp)->v_type = VMARKER;
4600 
4601 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
4602 	while (vp != NULL && (vp->v_type == VMARKER ||
4603 	    (vp->v_iflag & VI_DOOMED) != 0))
4604 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
4605 
4606 	/* Check if we are done */
4607 	if (vp == NULL) {
4608 		MNT_REL(mp);
4609 		MNT_IUNLOCK(mp);
4610 		free(*mvp, M_VNODE_MARKER);
4611 		*mvp = NULL;
4612 		return (NULL);
4613 	}
4614 	(*mvp)->v_mount = mp;
4615 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4616 	VI_LOCK(vp);
4617 	MNT_IUNLOCK(mp);
4618 	return (vp);
4619 }
4620 
4621 
4622 void
4623 __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp)
4624 {
4625 
4626 	if (*mvp == NULL) {
4627 		MNT_IUNLOCK(mp);
4628 		return;
4629 	}
4630 
4631 	mtx_assert(MNT_MTX(mp), MA_OWNED);
4632 
4633 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4634 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4635 	MNT_REL(mp);
4636 	MNT_IUNLOCK(mp);
4637 	free(*mvp, M_VNODE_MARKER);
4638 	*mvp = NULL;
4639 }
4640 
4641 /*
4642  * These are helper functions for filesystems to traverse their
4643  * active vnodes.  See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h
4644  */
4645 struct vnode *
4646 __mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4647 {
4648 	struct vnode *vp, *nvp;
4649 
4650 	if (should_yield())
4651 		kern_yield(PRI_UNCHANGED);
4652 	MNT_ILOCK(mp);
4653 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4654 	vp = TAILQ_NEXT(*mvp, v_actfreelist);
4655 	while (vp != NULL) {
4656 		VI_LOCK(vp);
4657 		if (vp->v_mount == mp && vp->v_type != VMARKER &&
4658 		    (vp->v_iflag & VI_DOOMED) == 0)
4659 			break;
4660 		nvp = TAILQ_NEXT(vp, v_actfreelist);
4661 		VI_UNLOCK(vp);
4662 		vp = nvp;
4663 	}
4664 
4665 	/* Check if we are done */
4666 	if (vp == NULL) {
4667 		__mnt_vnode_markerfree_active(mvp, mp);
4668 		/* MNT_IUNLOCK(mp); -- done in above function */
4669 		mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
4670 		return (NULL);
4671 	}
4672 	mtx_lock(&vnode_free_list_mtx);
4673 	TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4674 	TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
4675 	mtx_unlock(&vnode_free_list_mtx);
4676 	MNT_IUNLOCK(mp);
4677 	return (vp);
4678 }
4679 
4680 struct vnode *
4681 __mnt_vnode_first_active(struct vnode **mvp, struct mount *mp)
4682 {
4683 	struct vnode *vp, *nvp;
4684 
4685 	*mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4686 	MNT_ILOCK(mp);
4687 	MNT_REF(mp);
4688 	(*mvp)->v_type = VMARKER;
4689 
4690 	vp = TAILQ_NEXT(*mvp, v_actfreelist);
4691 	while (vp != NULL) {
4692 		VI_LOCK(vp);
4693 		if (vp->v_mount == mp && vp->v_type != VMARKER &&
4694 		    (vp->v_iflag & VI_DOOMED) == 0)
4695 			break;
4696 		nvp = TAILQ_NEXT(vp, v_actfreelist);
4697 		VI_UNLOCK(vp);
4698 		vp = nvp;
4699 	}
4700 
4701 	/* Check if we are done */
4702 	if (vp == NULL) {
4703 		MNT_REL(mp);
4704 		MNT_IUNLOCK(mp);
4705 		free(*mvp, M_VNODE_MARKER);
4706 		*mvp = NULL;
4707 		return (NULL);
4708 	}
4709 	(*mvp)->v_mount = mp;
4710 	mtx_lock(&vnode_free_list_mtx);
4711 	TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
4712 	mtx_unlock(&vnode_free_list_mtx);
4713 	MNT_IUNLOCK(mp);
4714 	return (vp);
4715 }
4716 
4717 void
4718 __mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4719 {
4720 
4721 	if (*mvp == NULL) {
4722 		MNT_IUNLOCK(mp);
4723 		return;
4724 	}
4725 
4726 	mtx_assert(MNT_MTX(mp), MA_OWNED);
4727 
4728 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4729 	mtx_lock(&vnode_free_list_mtx);
4730 	TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4731 	mtx_unlock(&vnode_free_list_mtx);
4732 	MNT_REL(mp);
4733 	MNT_IUNLOCK(mp);
4734 	free(*mvp, M_VNODE_MARKER);
4735 	*mvp = NULL;
4736 }
4737