xref: /freebsd/sys/kern/vfs_subr.c (revision 298cf604ccf133b101c6fad42d1a078a1fac58ca)
1 /*-
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)vfs_subr.c	8.31 (Berkeley) 5/26/95
35  */
36 
37 /*
38  * External virtual filesystem routines
39  */
40 
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
43 
44 #include "opt_compat.h"
45 #include "opt_ddb.h"
46 #include "opt_watchdog.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/bio.h>
51 #include <sys/buf.h>
52 #include <sys/condvar.h>
53 #include <sys/conf.h>
54 #include <sys/dirent.h>
55 #include <sys/event.h>
56 #include <sys/eventhandler.h>
57 #include <sys/extattr.h>
58 #include <sys/file.h>
59 #include <sys/fcntl.h>
60 #include <sys/jail.h>
61 #include <sys/kdb.h>
62 #include <sys/kernel.h>
63 #include <sys/kthread.h>
64 #include <sys/lockf.h>
65 #include <sys/malloc.h>
66 #include <sys/mount.h>
67 #include <sys/namei.h>
68 #include <sys/priv.h>
69 #include <sys/reboot.h>
70 #include <sys/sched.h>
71 #include <sys/sleepqueue.h>
72 #include <sys/smp.h>
73 #include <sys/stat.h>
74 #include <sys/sysctl.h>
75 #include <sys/syslog.h>
76 #include <sys/vmmeter.h>
77 #include <sys/vnode.h>
78 #include <sys/watchdog.h>
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 static void	delmntque(struct vnode *vp);
98 static int	flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
99 		    int slpflag, int slptimeo);
100 static void	syncer_shutdown(void *arg, int howto);
101 static int	vtryrecycle(struct vnode *vp);
102 static void	v_incr_usecount(struct vnode *);
103 static void	v_decr_usecount(struct vnode *);
104 static void	v_decr_useonly(struct vnode *);
105 static void	v_upgrade_usecount(struct vnode *);
106 static void	vnlru_free(int);
107 static void	vgonel(struct vnode *);
108 static void	vfs_knllock(void *arg);
109 static void	vfs_knlunlock(void *arg);
110 static void	vfs_knl_assert_locked(void *arg);
111 static void	vfs_knl_assert_unlocked(void *arg);
112 static void	destroy_vpollinfo(struct vpollinfo *vi);
113 
114 /*
115  * Number of vnodes in existence.  Increased whenever getnewvnode()
116  * allocates a new vnode, decreased in vdropl() for VI_DOOMED vnode.
117  */
118 static unsigned long	numvnodes;
119 
120 SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
121     "Number of vnodes in existence");
122 
123 /*
124  * Conversion tables for conversion from vnode types to inode formats
125  * and back.
126  */
127 enum vtype iftovt_tab[16] = {
128 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
129 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
130 };
131 int vttoif_tab[10] = {
132 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
133 	S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
134 };
135 
136 /*
137  * List of vnodes that are ready for recycling.
138  */
139 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
140 
141 /*
142  * Free vnode target.  Free vnodes may simply be files which have been stat'd
143  * but not read.  This is somewhat common, and a small cache of such files
144  * should be kept to avoid recreation costs.
145  */
146 static u_long wantfreevnodes;
147 SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
148 /* Number of vnodes in the free list. */
149 static u_long freevnodes;
150 SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
151     "Number of vnodes in the free list");
152 
153 static int vlru_allow_cache_src;
154 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
155     &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
156 
157 /*
158  * Various variables used for debugging the new implementation of
159  * reassignbuf().
160  * XXX these are probably of (very) limited utility now.
161  */
162 static int reassignbufcalls;
163 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
164     "Number of calls to reassignbuf");
165 
166 /*
167  * Cache for the mount type id assigned to NFS.  This is used for
168  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
169  */
170 int	nfs_mount_type = -1;
171 
172 /* To keep more than one thread at a time from running vfs_getnewfsid */
173 static struct mtx mntid_mtx;
174 
175 /*
176  * Lock for any access to the following:
177  *	vnode_free_list
178  *	numvnodes
179  *	freevnodes
180  */
181 static struct mtx vnode_free_list_mtx;
182 
183 /* Publicly exported FS */
184 struct nfs_public nfs_pub;
185 
186 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
187 static uma_zone_t vnode_zone;
188 static uma_zone_t vnodepoll_zone;
189 
190 /*
191  * The workitem queue.
192  *
193  * It is useful to delay writes of file data and filesystem metadata
194  * for tens of seconds so that quickly created and deleted files need
195  * not waste disk bandwidth being created and removed. To realize this,
196  * we append vnodes to a "workitem" queue. When running with a soft
197  * updates implementation, most pending metadata dependencies should
198  * not wait for more than a few seconds. Thus, mounted on block devices
199  * are delayed only about a half the time that file data is delayed.
200  * Similarly, directory updates are more critical, so are only delayed
201  * about a third the time that file data is delayed. Thus, there are
202  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
203  * one each second (driven off the filesystem syncer process). The
204  * syncer_delayno variable indicates the next queue that is to be processed.
205  * Items that need to be processed soon are placed in this queue:
206  *
207  *	syncer_workitem_pending[syncer_delayno]
208  *
209  * A delay of fifteen seconds is done by placing the request fifteen
210  * entries later in the queue:
211  *
212  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
213  *
214  */
215 static int syncer_delayno;
216 static long syncer_mask;
217 LIST_HEAD(synclist, bufobj);
218 static struct synclist *syncer_workitem_pending;
219 /*
220  * The sync_mtx protects:
221  *	bo->bo_synclist
222  *	sync_vnode_count
223  *	syncer_delayno
224  *	syncer_state
225  *	syncer_workitem_pending
226  *	syncer_worklist_len
227  *	rushjob
228  */
229 static struct mtx sync_mtx;
230 static struct cv sync_wakeup;
231 
232 #define SYNCER_MAXDELAY		32
233 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
234 static int syncdelay = 30;		/* max time to delay syncing data */
235 static int filedelay = 30;		/* time to delay syncing files */
236 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
237     "Time to delay syncing files (in seconds)");
238 static int dirdelay = 29;		/* time to delay syncing directories */
239 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
240     "Time to delay syncing directories (in seconds)");
241 static int metadelay = 28;		/* time to delay syncing metadata */
242 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
243     "Time to delay syncing metadata (in seconds)");
244 static int rushjob;		/* number of slots to run ASAP */
245 static int stat_rush_requests;	/* number of times I/O speeded up */
246 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
247     "Number of times I/O speeded up (rush requests)");
248 
249 /*
250  * When shutting down the syncer, run it at four times normal speed.
251  */
252 #define SYNCER_SHUTDOWN_SPEEDUP		4
253 static int sync_vnode_count;
254 static int syncer_worklist_len;
255 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
256     syncer_state;
257 
258 /*
259  * Number of vnodes we want to exist at any one time.  This is mostly used
260  * to size hash tables in vnode-related code.  It is normally not used in
261  * getnewvnode(), as wantfreevnodes is normally nonzero.)
262  *
263  * XXX desiredvnodes is historical cruft and should not exist.
264  */
265 int desiredvnodes;
266 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
267     &desiredvnodes, 0, "Maximum number of vnodes");
268 SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
269     &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
270 static int vnlru_nowhere;
271 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
272     &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
273 
274 /*
275  * Macros to control when a vnode is freed and recycled.  All require
276  * the vnode interlock.
277  */
278 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
279 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
280 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
281 
282 
283 /*
284  * Initialize the vnode management data structures.
285  *
286  * Reevaluate the following cap on the number of vnodes after the physical
287  * memory size exceeds 512GB.  In the limit, as the physical memory size
288  * grows, the ratio of physical pages to vnodes approaches sixteen to one.
289  */
290 #ifndef	MAXVNODES_MAX
291 #define	MAXVNODES_MAX	(512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
292 #endif
293 static void
294 vntblinit(void *dummy __unused)
295 {
296 	int physvnodes, virtvnodes;
297 
298 	/*
299 	 * Desiredvnodes is a function of the physical memory size and the
300 	 * kernel's heap size.  Generally speaking, it scales with the
301 	 * physical memory size.  The ratio of desiredvnodes to physical pages
302 	 * is one to four until desiredvnodes exceeds 98,304.  Thereafter, the
303 	 * marginal ratio of desiredvnodes to physical pages is one to
304 	 * sixteen.  However, desiredvnodes is limited by the kernel's heap
305 	 * size.  The memory required by desiredvnodes vnodes and vm objects
306 	 * may not exceed one seventh of the kernel's heap size.
307 	 */
308 	physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
309 	    cnt.v_page_count) / 16;
310 	virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
311 	    sizeof(struct vnode)));
312 	desiredvnodes = min(physvnodes, virtvnodes);
313 	if (desiredvnodes > MAXVNODES_MAX) {
314 		if (bootverbose)
315 			printf("Reducing kern.maxvnodes %d -> %d\n",
316 			    desiredvnodes, MAXVNODES_MAX);
317 		desiredvnodes = MAXVNODES_MAX;
318 	}
319 	wantfreevnodes = desiredvnodes / 4;
320 	mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
321 	TAILQ_INIT(&vnode_free_list);
322 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
323 	vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
324 	    NULL, NULL, UMA_ALIGN_PTR, 0);
325 	vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
326 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
327 	/*
328 	 * Initialize the filesystem syncer.
329 	 */
330 	syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
331 	    &syncer_mask);
332 	syncer_maxdelay = syncer_mask + 1;
333 	mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
334 	cv_init(&sync_wakeup, "syncer");
335 }
336 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
337 
338 
339 /*
340  * Mark a mount point as busy. Used to synchronize access and to delay
341  * unmounting. Eventually, mountlist_mtx is not released on failure.
342  *
343  * vfs_busy() is a custom lock, it can block the caller.
344  * vfs_busy() only sleeps if the unmount is active on the mount point.
345  * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
346  * vnode belonging to mp.
347  *
348  * Lookup uses vfs_busy() to traverse mount points.
349  * root fs			var fs
350  * / vnode lock		A	/ vnode lock (/var)		D
351  * /var vnode lock	B	/log vnode lock(/var/log)	E
352  * vfs_busy lock	C	vfs_busy lock			F
353  *
354  * Within each file system, the lock order is C->A->B and F->D->E.
355  *
356  * When traversing across mounts, the system follows that lock order:
357  *
358  *        C->A->B
359  *              |
360  *              +->F->D->E
361  *
362  * The lookup() process for namei("/var") illustrates the process:
363  *  VOP_LOOKUP() obtains B while A is held
364  *  vfs_busy() obtains a shared lock on F while A and B are held
365  *  vput() releases lock on B
366  *  vput() releases lock on A
367  *  VFS_ROOT() obtains lock on D while shared lock on F is held
368  *  vfs_unbusy() releases shared lock on F
369  *  vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
370  *    Attempt to lock A (instead of vp_crossmp) while D is held would
371  *    violate the global order, causing deadlocks.
372  *
373  * dounmount() locks B while F is drained.
374  */
375 int
376 vfs_busy(struct mount *mp, int flags)
377 {
378 
379 	MPASS((flags & ~MBF_MASK) == 0);
380 	CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
381 
382 	MNT_ILOCK(mp);
383 	MNT_REF(mp);
384 	/*
385 	 * If mount point is currenly being unmounted, sleep until the
386 	 * mount point fate is decided.  If thread doing the unmounting fails,
387 	 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
388 	 * that this mount point has survived the unmount attempt and vfs_busy
389 	 * should retry.  Otherwise the unmounter thread will set MNTK_REFEXPIRE
390 	 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
391 	 * about to be really destroyed.  vfs_busy needs to release its
392 	 * reference on the mount point in this case and return with ENOENT,
393 	 * telling the caller that mount mount it tried to busy is no longer
394 	 * valid.
395 	 */
396 	while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
397 		if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
398 			MNT_REL(mp);
399 			MNT_IUNLOCK(mp);
400 			CTR1(KTR_VFS, "%s: failed busying before sleeping",
401 			    __func__);
402 			return (ENOENT);
403 		}
404 		if (flags & MBF_MNTLSTLOCK)
405 			mtx_unlock(&mountlist_mtx);
406 		mp->mnt_kern_flag |= MNTK_MWAIT;
407 		msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
408 		if (flags & MBF_MNTLSTLOCK)
409 			mtx_lock(&mountlist_mtx);
410 		MNT_ILOCK(mp);
411 	}
412 	if (flags & MBF_MNTLSTLOCK)
413 		mtx_unlock(&mountlist_mtx);
414 	mp->mnt_lockref++;
415 	MNT_IUNLOCK(mp);
416 	return (0);
417 }
418 
419 /*
420  * Free a busy filesystem.
421  */
422 void
423 vfs_unbusy(struct mount *mp)
424 {
425 
426 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
427 	MNT_ILOCK(mp);
428 	MNT_REL(mp);
429 	KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
430 	mp->mnt_lockref--;
431 	if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
432 		MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
433 		CTR1(KTR_VFS, "%s: waking up waiters", __func__);
434 		mp->mnt_kern_flag &= ~MNTK_DRAINING;
435 		wakeup(&mp->mnt_lockref);
436 	}
437 	MNT_IUNLOCK(mp);
438 }
439 
440 /*
441  * Lookup a mount point by filesystem identifier.
442  */
443 struct mount *
444 vfs_getvfs(fsid_t *fsid)
445 {
446 	struct mount *mp;
447 
448 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
449 	mtx_lock(&mountlist_mtx);
450 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
451 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
452 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
453 			vfs_ref(mp);
454 			mtx_unlock(&mountlist_mtx);
455 			return (mp);
456 		}
457 	}
458 	mtx_unlock(&mountlist_mtx);
459 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
460 	return ((struct mount *) 0);
461 }
462 
463 /*
464  * Lookup a mount point by filesystem identifier, busying it before
465  * returning.
466  */
467 struct mount *
468 vfs_busyfs(fsid_t *fsid)
469 {
470 	struct mount *mp;
471 	int error;
472 
473 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
474 	mtx_lock(&mountlist_mtx);
475 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
476 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
477 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
478 			error = vfs_busy(mp, MBF_MNTLSTLOCK);
479 			if (error) {
480 				mtx_unlock(&mountlist_mtx);
481 				return (NULL);
482 			}
483 			return (mp);
484 		}
485 	}
486 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
487 	mtx_unlock(&mountlist_mtx);
488 	return ((struct mount *) 0);
489 }
490 
491 /*
492  * Check if a user can access privileged mount options.
493  */
494 int
495 vfs_suser(struct mount *mp, struct thread *td)
496 {
497 	int error;
498 
499 	/*
500 	 * If the thread is jailed, but this is not a jail-friendly file
501 	 * system, deny immediately.
502 	 */
503 	if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
504 		return (EPERM);
505 
506 	/*
507 	 * If the file system was mounted outside the jail of the calling
508 	 * thread, deny immediately.
509 	 */
510 	if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
511 		return (EPERM);
512 
513 	/*
514 	 * If file system supports delegated administration, we don't check
515 	 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
516 	 * by the file system itself.
517 	 * If this is not the user that did original mount, we check for
518 	 * the PRIV_VFS_MOUNT_OWNER privilege.
519 	 */
520 	if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
521 	    mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
522 		if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
523 			return (error);
524 	}
525 	return (0);
526 }
527 
528 /*
529  * Get a new unique fsid.  Try to make its val[0] unique, since this value
530  * will be used to create fake device numbers for stat().  Also try (but
531  * not so hard) make its val[0] unique mod 2^16, since some emulators only
532  * support 16-bit device numbers.  We end up with unique val[0]'s for the
533  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
534  *
535  * Keep in mind that several mounts may be running in parallel.  Starting
536  * the search one past where the previous search terminated is both a
537  * micro-optimization and a defense against returning the same fsid to
538  * different mounts.
539  */
540 void
541 vfs_getnewfsid(struct mount *mp)
542 {
543 	static uint16_t mntid_base;
544 	struct mount *nmp;
545 	fsid_t tfsid;
546 	int mtype;
547 
548 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
549 	mtx_lock(&mntid_mtx);
550 	mtype = mp->mnt_vfc->vfc_typenum;
551 	tfsid.val[1] = mtype;
552 	mtype = (mtype & 0xFF) << 24;
553 	for (;;) {
554 		tfsid.val[0] = makedev(255,
555 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
556 		mntid_base++;
557 		if ((nmp = vfs_getvfs(&tfsid)) == NULL)
558 			break;
559 		vfs_rel(nmp);
560 	}
561 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
562 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
563 	mtx_unlock(&mntid_mtx);
564 }
565 
566 /*
567  * Knob to control the precision of file timestamps:
568  *
569  *   0 = seconds only; nanoseconds zeroed.
570  *   1 = seconds and nanoseconds, accurate within 1/HZ.
571  *   2 = seconds and nanoseconds, truncated to microseconds.
572  * >=3 = seconds and nanoseconds, maximum precision.
573  */
574 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
575 
576 static int timestamp_precision = TSP_SEC;
577 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
578     &timestamp_precision, 0, "File timestamp precision (0: seconds, "
579     "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
580     "3+: sec + ns (max. precision))");
581 
582 /*
583  * Get a current timestamp.
584  */
585 void
586 vfs_timestamp(struct timespec *tsp)
587 {
588 	struct timeval tv;
589 
590 	switch (timestamp_precision) {
591 	case TSP_SEC:
592 		tsp->tv_sec = time_second;
593 		tsp->tv_nsec = 0;
594 		break;
595 	case TSP_HZ:
596 		getnanotime(tsp);
597 		break;
598 	case TSP_USEC:
599 		microtime(&tv);
600 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
601 		break;
602 	case TSP_NSEC:
603 	default:
604 		nanotime(tsp);
605 		break;
606 	}
607 }
608 
609 /*
610  * Set vnode attributes to VNOVAL
611  */
612 void
613 vattr_null(struct vattr *vap)
614 {
615 
616 	vap->va_type = VNON;
617 	vap->va_size = VNOVAL;
618 	vap->va_bytes = VNOVAL;
619 	vap->va_mode = VNOVAL;
620 	vap->va_nlink = VNOVAL;
621 	vap->va_uid = VNOVAL;
622 	vap->va_gid = VNOVAL;
623 	vap->va_fsid = VNOVAL;
624 	vap->va_fileid = VNOVAL;
625 	vap->va_blocksize = VNOVAL;
626 	vap->va_rdev = VNOVAL;
627 	vap->va_atime.tv_sec = VNOVAL;
628 	vap->va_atime.tv_nsec = VNOVAL;
629 	vap->va_mtime.tv_sec = VNOVAL;
630 	vap->va_mtime.tv_nsec = VNOVAL;
631 	vap->va_ctime.tv_sec = VNOVAL;
632 	vap->va_ctime.tv_nsec = VNOVAL;
633 	vap->va_birthtime.tv_sec = VNOVAL;
634 	vap->va_birthtime.tv_nsec = VNOVAL;
635 	vap->va_flags = VNOVAL;
636 	vap->va_gen = VNOVAL;
637 	vap->va_vaflags = 0;
638 }
639 
640 /*
641  * This routine is called when we have too many vnodes.  It attempts
642  * to free <count> vnodes and will potentially free vnodes that still
643  * have VM backing store (VM backing store is typically the cause
644  * of a vnode blowout so we want to do this).  Therefore, this operation
645  * is not considered cheap.
646  *
647  * A number of conditions may prevent a vnode from being reclaimed.
648  * the buffer cache may have references on the vnode, a directory
649  * vnode may still have references due to the namei cache representing
650  * underlying files, or the vnode may be in active use.   It is not
651  * desireable to reuse such vnodes.  These conditions may cause the
652  * number of vnodes to reach some minimum value regardless of what
653  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
654  */
655 static int
656 vlrureclaim(struct mount *mp)
657 {
658 	struct vnode *vp;
659 	int done;
660 	int trigger;
661 	int usevnodes;
662 	int count;
663 
664 	/*
665 	 * Calculate the trigger point, don't allow user
666 	 * screwups to blow us up.   This prevents us from
667 	 * recycling vnodes with lots of resident pages.  We
668 	 * aren't trying to free memory, we are trying to
669 	 * free vnodes.
670 	 */
671 	usevnodes = desiredvnodes;
672 	if (usevnodes <= 0)
673 		usevnodes = 1;
674 	trigger = cnt.v_page_count * 2 / usevnodes;
675 	done = 0;
676 	vn_start_write(NULL, &mp, V_WAIT);
677 	MNT_ILOCK(mp);
678 	count = mp->mnt_nvnodelistsize / 10 + 1;
679 	while (count != 0) {
680 		vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
681 		while (vp != NULL && vp->v_type == VMARKER)
682 			vp = TAILQ_NEXT(vp, v_nmntvnodes);
683 		if (vp == NULL)
684 			break;
685 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
686 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
687 		--count;
688 		if (!VI_TRYLOCK(vp))
689 			goto next_iter;
690 		/*
691 		 * If it's been deconstructed already, it's still
692 		 * referenced, or it exceeds the trigger, skip it.
693 		 */
694 		if (vp->v_usecount ||
695 		    (!vlru_allow_cache_src &&
696 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
697 		    (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
698 		    vp->v_object->resident_page_count > trigger)) {
699 			VI_UNLOCK(vp);
700 			goto next_iter;
701 		}
702 		MNT_IUNLOCK(mp);
703 		vholdl(vp);
704 		if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
705 			vdrop(vp);
706 			goto next_iter_mntunlocked;
707 		}
708 		VI_LOCK(vp);
709 		/*
710 		 * v_usecount may have been bumped after VOP_LOCK() dropped
711 		 * the vnode interlock and before it was locked again.
712 		 *
713 		 * It is not necessary to recheck VI_DOOMED because it can
714 		 * only be set by another thread that holds both the vnode
715 		 * lock and vnode interlock.  If another thread has the
716 		 * vnode lock before we get to VOP_LOCK() and obtains the
717 		 * vnode interlock after VOP_LOCK() drops the vnode
718 		 * interlock, the other thread will be unable to drop the
719 		 * vnode lock before our VOP_LOCK() call fails.
720 		 */
721 		if (vp->v_usecount ||
722 		    (!vlru_allow_cache_src &&
723 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
724 		    (vp->v_object != NULL &&
725 		    vp->v_object->resident_page_count > trigger)) {
726 			VOP_UNLOCK(vp, LK_INTERLOCK);
727 			goto next_iter_mntunlocked;
728 		}
729 		KASSERT((vp->v_iflag & VI_DOOMED) == 0,
730 		    ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
731 		vgonel(vp);
732 		VOP_UNLOCK(vp, 0);
733 		vdropl(vp);
734 		done++;
735 next_iter_mntunlocked:
736 		if (!should_yield())
737 			goto relock_mnt;
738 		goto yield;
739 next_iter:
740 		if (!should_yield())
741 			continue;
742 		MNT_IUNLOCK(mp);
743 yield:
744 		kern_yield(PRI_USER);
745 relock_mnt:
746 		MNT_ILOCK(mp);
747 	}
748 	MNT_IUNLOCK(mp);
749 	vn_finished_write(mp);
750 	return done;
751 }
752 
753 /*
754  * Attempt to keep the free list at wantfreevnodes length.
755  */
756 static void
757 vnlru_free(int count)
758 {
759 	struct vnode *vp;
760 
761 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
762 	for (; count > 0; count--) {
763 		vp = TAILQ_FIRST(&vnode_free_list);
764 		/*
765 		 * The list can be modified while the free_list_mtx
766 		 * has been dropped and vp could be NULL here.
767 		 */
768 		if (!vp)
769 			break;
770 		VNASSERT(vp->v_op != NULL, vp,
771 		    ("vnlru_free: vnode already reclaimed."));
772 		KASSERT((vp->v_iflag & VI_FREE) != 0,
773 		    ("Removing vnode not on freelist"));
774 		KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
775 		    ("Mangling active vnode"));
776 		TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
777 		/*
778 		 * Don't recycle if we can't get the interlock.
779 		 */
780 		if (!VI_TRYLOCK(vp)) {
781 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
782 			continue;
783 		}
784 		VNASSERT(VCANRECYCLE(vp), vp,
785 		    ("vp inconsistent on freelist"));
786 		freevnodes--;
787 		vp->v_iflag &= ~VI_FREE;
788 		vholdl(vp);
789 		mtx_unlock(&vnode_free_list_mtx);
790 		VI_UNLOCK(vp);
791 		vtryrecycle(vp);
792 		/*
793 		 * If the recycled succeeded this vdrop will actually free
794 		 * the vnode.  If not it will simply place it back on
795 		 * the free list.
796 		 */
797 		vdrop(vp);
798 		mtx_lock(&vnode_free_list_mtx);
799 	}
800 }
801 /*
802  * Attempt to recycle vnodes in a context that is always safe to block.
803  * Calling vlrurecycle() from the bowels of filesystem code has some
804  * interesting deadlock problems.
805  */
806 static struct proc *vnlruproc;
807 static int vnlruproc_sig;
808 
809 static void
810 vnlru_proc(void)
811 {
812 	struct mount *mp, *nmp;
813 	int done;
814 	struct proc *p = vnlruproc;
815 
816 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
817 	    SHUTDOWN_PRI_FIRST);
818 
819 	for (;;) {
820 		kproc_suspend_check(p);
821 		mtx_lock(&vnode_free_list_mtx);
822 		if (freevnodes > wantfreevnodes)
823 			vnlru_free(freevnodes - wantfreevnodes);
824 		if (numvnodes <= desiredvnodes * 9 / 10) {
825 			vnlruproc_sig = 0;
826 			wakeup(&vnlruproc_sig);
827 			msleep(vnlruproc, &vnode_free_list_mtx,
828 			    PVFS|PDROP, "vlruwt", hz);
829 			continue;
830 		}
831 		mtx_unlock(&vnode_free_list_mtx);
832 		done = 0;
833 		mtx_lock(&mountlist_mtx);
834 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
835 			if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
836 				nmp = TAILQ_NEXT(mp, mnt_list);
837 				continue;
838 			}
839 			done += vlrureclaim(mp);
840 			mtx_lock(&mountlist_mtx);
841 			nmp = TAILQ_NEXT(mp, mnt_list);
842 			vfs_unbusy(mp);
843 		}
844 		mtx_unlock(&mountlist_mtx);
845 		if (done == 0) {
846 #if 0
847 			/* These messages are temporary debugging aids */
848 			if (vnlru_nowhere < 5)
849 				printf("vnlru process getting nowhere..\n");
850 			else if (vnlru_nowhere == 5)
851 				printf("vnlru process messages stopped.\n");
852 #endif
853 			vnlru_nowhere++;
854 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
855 		} else
856 			kern_yield(PRI_USER);
857 	}
858 }
859 
860 static struct kproc_desc vnlru_kp = {
861 	"vnlru",
862 	vnlru_proc,
863 	&vnlruproc
864 };
865 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
866     &vnlru_kp);
867 
868 /*
869  * Routines having to do with the management of the vnode table.
870  */
871 
872 /*
873  * Try to recycle a freed vnode.  We abort if anyone picks up a reference
874  * before we actually vgone().  This function must be called with the vnode
875  * held to prevent the vnode from being returned to the free list midway
876  * through vgone().
877  */
878 static int
879 vtryrecycle(struct vnode *vp)
880 {
881 	struct mount *vnmp;
882 
883 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
884 	VNASSERT(vp->v_holdcnt, vp,
885 	    ("vtryrecycle: Recycling vp %p without a reference.", vp));
886 	/*
887 	 * This vnode may found and locked via some other list, if so we
888 	 * can't recycle it yet.
889 	 */
890 	if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
891 		CTR2(KTR_VFS,
892 		    "%s: impossible to recycle, vp %p lock is already held",
893 		    __func__, vp);
894 		return (EWOULDBLOCK);
895 	}
896 	/*
897 	 * Don't recycle if its filesystem is being suspended.
898 	 */
899 	if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
900 		VOP_UNLOCK(vp, 0);
901 		CTR2(KTR_VFS,
902 		    "%s: impossible to recycle, cannot start the write for %p",
903 		    __func__, vp);
904 		return (EBUSY);
905 	}
906 	/*
907 	 * If we got this far, we need to acquire the interlock and see if
908 	 * anyone picked up this vnode from another list.  If not, we will
909 	 * mark it with DOOMED via vgonel() so that anyone who does find it
910 	 * will skip over it.
911 	 */
912 	VI_LOCK(vp);
913 	if (vp->v_usecount) {
914 		VOP_UNLOCK(vp, LK_INTERLOCK);
915 		vn_finished_write(vnmp);
916 		CTR2(KTR_VFS,
917 		    "%s: impossible to recycle, %p is already referenced",
918 		    __func__, vp);
919 		return (EBUSY);
920 	}
921 	if ((vp->v_iflag & VI_DOOMED) == 0)
922 		vgonel(vp);
923 	VOP_UNLOCK(vp, LK_INTERLOCK);
924 	vn_finished_write(vnmp);
925 	return (0);
926 }
927 
928 /*
929  * Wait for available vnodes.
930  */
931 static int
932 getnewvnode_wait(int suspended)
933 {
934 
935 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
936 	if (numvnodes > desiredvnodes) {
937 		if (suspended) {
938 			/*
939 			 * File system is beeing suspended, we cannot risk a
940 			 * deadlock here, so allocate new vnode anyway.
941 			 */
942 			if (freevnodes > wantfreevnodes)
943 				vnlru_free(freevnodes - wantfreevnodes);
944 			return (0);
945 		}
946 		if (vnlruproc_sig == 0) {
947 			vnlruproc_sig = 1;	/* avoid unnecessary wakeups */
948 			wakeup(vnlruproc);
949 		}
950 		msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
951 		    "vlruwk", hz);
952 	}
953 	return (numvnodes > desiredvnodes ? ENFILE : 0);
954 }
955 
956 void
957 getnewvnode_reserve(u_int count)
958 {
959 	struct thread *td;
960 
961 	td = curthread;
962 	mtx_lock(&vnode_free_list_mtx);
963 	while (count > 0) {
964 		if (getnewvnode_wait(0) == 0) {
965 			count--;
966 			td->td_vp_reserv++;
967 			numvnodes++;
968 		}
969 	}
970 	mtx_unlock(&vnode_free_list_mtx);
971 }
972 
973 void
974 getnewvnode_drop_reserve(void)
975 {
976 	struct thread *td;
977 
978 	td = curthread;
979 	mtx_lock(&vnode_free_list_mtx);
980 	KASSERT(numvnodes >= td->td_vp_reserv, ("reserve too large"));
981 	numvnodes -= td->td_vp_reserv;
982 	mtx_unlock(&vnode_free_list_mtx);
983 	td->td_vp_reserv = 0;
984 }
985 
986 /*
987  * Return the next vnode from the free list.
988  */
989 int
990 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
991     struct vnode **vpp)
992 {
993 	struct vnode *vp;
994 	struct bufobj *bo;
995 	struct thread *td;
996 	int error;
997 
998 	CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
999 	vp = NULL;
1000 	td = curthread;
1001 	if (td->td_vp_reserv > 0) {
1002 		td->td_vp_reserv -= 1;
1003 		goto alloc;
1004 	}
1005 	mtx_lock(&vnode_free_list_mtx);
1006 	/*
1007 	 * Lend our context to reclaim vnodes if they've exceeded the max.
1008 	 */
1009 	if (freevnodes > wantfreevnodes)
1010 		vnlru_free(1);
1011 	error = getnewvnode_wait(mp != NULL && (mp->mnt_kern_flag &
1012 	    MNTK_SUSPEND));
1013 #if 0	/* XXX Not all VFS_VGET/ffs_vget callers check returns. */
1014 	if (error != 0) {
1015 		mtx_unlock(&vnode_free_list_mtx);
1016 		return (error);
1017 	}
1018 #endif
1019 	numvnodes++;
1020 	mtx_unlock(&vnode_free_list_mtx);
1021 alloc:
1022 	vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
1023 	/*
1024 	 * Setup locks.
1025 	 */
1026 	vp->v_vnlock = &vp->v_lock;
1027 	mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
1028 	/*
1029 	 * By default, don't allow shared locks unless filesystems
1030 	 * opt-in.
1031 	 */
1032 	lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
1033 	/*
1034 	 * Initialize bufobj.
1035 	 */
1036 	bo = &vp->v_bufobj;
1037 	bo->__bo_vnode = vp;
1038 	mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
1039 	bo->bo_ops = &buf_ops_bio;
1040 	bo->bo_private = vp;
1041 	TAILQ_INIT(&bo->bo_clean.bv_hd);
1042 	TAILQ_INIT(&bo->bo_dirty.bv_hd);
1043 	/*
1044 	 * Initialize namecache.
1045 	 */
1046 	LIST_INIT(&vp->v_cache_src);
1047 	TAILQ_INIT(&vp->v_cache_dst);
1048 	/*
1049 	 * Finalize various vnode identity bits.
1050 	 */
1051 	vp->v_type = VNON;
1052 	vp->v_tag = tag;
1053 	vp->v_op = vops;
1054 	v_incr_usecount(vp);
1055 	vp->v_data = NULL;
1056 #ifdef MAC
1057 	mac_vnode_init(vp);
1058 	if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1059 		mac_vnode_associate_singlelabel(mp, vp);
1060 	else if (mp == NULL && vops != &dead_vnodeops)
1061 		printf("NULL mp in getnewvnode()\n");
1062 #endif
1063 	if (mp != NULL) {
1064 		bo->bo_bsize = mp->mnt_stat.f_iosize;
1065 		if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1066 			vp->v_vflag |= VV_NOKNOTE;
1067 	}
1068 	rangelock_init(&vp->v_rl);
1069 
1070 	*vpp = vp;
1071 	return (0);
1072 }
1073 
1074 /*
1075  * Delete from old mount point vnode list, if on one.
1076  */
1077 static void
1078 delmntque(struct vnode *vp)
1079 {
1080 	struct mount *mp;
1081 	int active;
1082 
1083 	mp = vp->v_mount;
1084 	if (mp == NULL)
1085 		return;
1086 	MNT_ILOCK(mp);
1087 	VI_LOCK(vp);
1088 	KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize,
1089 	    ("Active vnode list size %d > Vnode list size %d",
1090 	     mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize));
1091 	active = vp->v_iflag & VI_ACTIVE;
1092 	vp->v_iflag &= ~VI_ACTIVE;
1093 	if (active) {
1094 		mtx_lock(&vnode_free_list_mtx);
1095 		TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist);
1096 		mp->mnt_activevnodelistsize--;
1097 		mtx_unlock(&vnode_free_list_mtx);
1098 	}
1099 	vp->v_mount = NULL;
1100 	VI_UNLOCK(vp);
1101 	VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1102 		("bad mount point vnode list size"));
1103 	TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1104 	mp->mnt_nvnodelistsize--;
1105 	MNT_REL(mp);
1106 	MNT_IUNLOCK(mp);
1107 }
1108 
1109 static void
1110 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1111 {
1112 
1113 	vp->v_data = NULL;
1114 	vp->v_op = &dead_vnodeops;
1115 	vgone(vp);
1116 	vput(vp);
1117 }
1118 
1119 /*
1120  * Insert into list of vnodes for the new mount point, if available.
1121  */
1122 int
1123 insmntque1(struct vnode *vp, struct mount *mp,
1124 	void (*dtr)(struct vnode *, void *), void *dtr_arg)
1125 {
1126 
1127 	KASSERT(vp->v_mount == NULL,
1128 		("insmntque: vnode already on per mount vnode list"));
1129 	VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1130 	ASSERT_VOP_ELOCKED(vp, "insmntque: non-locked vp");
1131 
1132 	/*
1133 	 * We acquire the vnode interlock early to ensure that the
1134 	 * vnode cannot be recycled by another process releasing a
1135 	 * holdcnt on it before we get it on both the vnode list
1136 	 * and the active vnode list. The mount mutex protects only
1137 	 * manipulation of the vnode list and the vnode freelist
1138 	 * mutex protects only manipulation of the active vnode list.
1139 	 * Hence the need to hold the vnode interlock throughout.
1140 	 */
1141 	MNT_ILOCK(mp);
1142 	VI_LOCK(vp);
1143 	if (((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1144 	    ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1145 	    mp->mnt_nvnodelistsize == 0)) &&
1146 	    (vp->v_vflag & VV_FORCEINSMQ) == 0) {
1147 		VI_UNLOCK(vp);
1148 		MNT_IUNLOCK(mp);
1149 		if (dtr != NULL)
1150 			dtr(vp, dtr_arg);
1151 		return (EBUSY);
1152 	}
1153 	vp->v_mount = mp;
1154 	MNT_REF(mp);
1155 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1156 	VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1157 		("neg mount point vnode list size"));
1158 	mp->mnt_nvnodelistsize++;
1159 	KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
1160 	    ("Activating already active vnode"));
1161 	vp->v_iflag |= VI_ACTIVE;
1162 	mtx_lock(&vnode_free_list_mtx);
1163 	TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
1164 	mp->mnt_activevnodelistsize++;
1165 	mtx_unlock(&vnode_free_list_mtx);
1166 	VI_UNLOCK(vp);
1167 	MNT_IUNLOCK(mp);
1168 	return (0);
1169 }
1170 
1171 int
1172 insmntque(struct vnode *vp, struct mount *mp)
1173 {
1174 
1175 	return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1176 }
1177 
1178 /*
1179  * Flush out and invalidate all buffers associated with a bufobj
1180  * Called with the underlying object locked.
1181  */
1182 int
1183 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1184 {
1185 	int error;
1186 
1187 	BO_LOCK(bo);
1188 	if (flags & V_SAVE) {
1189 		error = bufobj_wwait(bo, slpflag, slptimeo);
1190 		if (error) {
1191 			BO_UNLOCK(bo);
1192 			return (error);
1193 		}
1194 		if (bo->bo_dirty.bv_cnt > 0) {
1195 			BO_UNLOCK(bo);
1196 			if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1197 				return (error);
1198 			/*
1199 			 * XXX We could save a lock/unlock if this was only
1200 			 * enabled under INVARIANTS
1201 			 */
1202 			BO_LOCK(bo);
1203 			if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1204 				panic("vinvalbuf: dirty bufs");
1205 		}
1206 	}
1207 	/*
1208 	 * If you alter this loop please notice that interlock is dropped and
1209 	 * reacquired in flushbuflist.  Special care is needed to ensure that
1210 	 * no race conditions occur from this.
1211 	 */
1212 	do {
1213 		error = flushbuflist(&bo->bo_clean,
1214 		    flags, bo, slpflag, slptimeo);
1215 		if (error == 0 && !(flags & V_CLEANONLY))
1216 			error = flushbuflist(&bo->bo_dirty,
1217 			    flags, bo, slpflag, slptimeo);
1218 		if (error != 0 && error != EAGAIN) {
1219 			BO_UNLOCK(bo);
1220 			return (error);
1221 		}
1222 	} while (error != 0);
1223 
1224 	/*
1225 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1226 	 * have write I/O in-progress but if there is a VM object then the
1227 	 * VM object can also have read-I/O in-progress.
1228 	 */
1229 	do {
1230 		bufobj_wwait(bo, 0, 0);
1231 		BO_UNLOCK(bo);
1232 		if (bo->bo_object != NULL) {
1233 			VM_OBJECT_LOCK(bo->bo_object);
1234 			vm_object_pip_wait(bo->bo_object, "bovlbx");
1235 			VM_OBJECT_UNLOCK(bo->bo_object);
1236 		}
1237 		BO_LOCK(bo);
1238 	} while (bo->bo_numoutput > 0);
1239 	BO_UNLOCK(bo);
1240 
1241 	/*
1242 	 * Destroy the copy in the VM cache, too.
1243 	 */
1244 	if (bo->bo_object != NULL &&
1245 	    (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) {
1246 		VM_OBJECT_LOCK(bo->bo_object);
1247 		vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
1248 		    OBJPR_CLEANONLY : 0);
1249 		VM_OBJECT_UNLOCK(bo->bo_object);
1250 	}
1251 
1252 #ifdef INVARIANTS
1253 	BO_LOCK(bo);
1254 	if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 &&
1255 	    (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1256 		panic("vinvalbuf: flush failed");
1257 	BO_UNLOCK(bo);
1258 #endif
1259 	return (0);
1260 }
1261 
1262 /*
1263  * Flush out and invalidate all buffers associated with a vnode.
1264  * Called with the underlying object locked.
1265  */
1266 int
1267 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1268 {
1269 
1270 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1271 	ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1272 	return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1273 }
1274 
1275 /*
1276  * Flush out buffers on the specified list.
1277  *
1278  */
1279 static int
1280 flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1281     int slptimeo)
1282 {
1283 	struct buf *bp, *nbp;
1284 	int retval, error;
1285 	daddr_t lblkno;
1286 	b_xflags_t xflags;
1287 
1288 	ASSERT_BO_LOCKED(bo);
1289 
1290 	retval = 0;
1291 	TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1292 		if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1293 		    ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1294 			continue;
1295 		}
1296 		lblkno = 0;
1297 		xflags = 0;
1298 		if (nbp != NULL) {
1299 			lblkno = nbp->b_lblkno;
1300 			xflags = nbp->b_xflags &
1301 				(BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1302 		}
1303 		retval = EAGAIN;
1304 		error = BUF_TIMELOCK(bp,
1305 		    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1306 		    "flushbuf", slpflag, slptimeo);
1307 		if (error) {
1308 			BO_LOCK(bo);
1309 			return (error != ENOLCK ? error : EAGAIN);
1310 		}
1311 		KASSERT(bp->b_bufobj == bo,
1312 		    ("bp %p wrong b_bufobj %p should be %p",
1313 		    bp, bp->b_bufobj, bo));
1314 		if (bp->b_bufobj != bo) {	/* XXX: necessary ? */
1315 			BUF_UNLOCK(bp);
1316 			BO_LOCK(bo);
1317 			return (EAGAIN);
1318 		}
1319 		/*
1320 		 * XXX Since there are no node locks for NFS, I
1321 		 * believe there is a slight chance that a delayed
1322 		 * write will occur while sleeping just above, so
1323 		 * check for it.
1324 		 */
1325 		if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1326 		    (flags & V_SAVE)) {
1327 			BO_LOCK(bo);
1328 			bremfree(bp);
1329 			BO_UNLOCK(bo);
1330 			bp->b_flags |= B_ASYNC;
1331 			bwrite(bp);
1332 			BO_LOCK(bo);
1333 			return (EAGAIN);	/* XXX: why not loop ? */
1334 		}
1335 		BO_LOCK(bo);
1336 		bremfree(bp);
1337 		BO_UNLOCK(bo);
1338 		bp->b_flags |= (B_INVAL | B_RELBUF);
1339 		bp->b_flags &= ~B_ASYNC;
1340 		brelse(bp);
1341 		BO_LOCK(bo);
1342 		if (nbp != NULL &&
1343 		    (nbp->b_bufobj != bo ||
1344 		     nbp->b_lblkno != lblkno ||
1345 		     (nbp->b_xflags &
1346 		      (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1347 			break;			/* nbp invalid */
1348 	}
1349 	return (retval);
1350 }
1351 
1352 /*
1353  * Truncate a file's buffer and pages to a specified length.  This
1354  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1355  * sync activity.
1356  */
1357 int
1358 vtruncbuf(struct vnode *vp, struct ucred *cred, off_t length, int blksize)
1359 {
1360 	struct buf *bp, *nbp;
1361 	int anyfreed;
1362 	int trunclbn;
1363 	struct bufobj *bo;
1364 
1365 	CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1366 	    vp, cred, blksize, (uintmax_t)length);
1367 
1368 	/*
1369 	 * Round up to the *next* lbn.
1370 	 */
1371 	trunclbn = (length + blksize - 1) / blksize;
1372 
1373 	ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1374 restart:
1375 	bo = &vp->v_bufobj;
1376 	BO_LOCK(bo);
1377 	anyfreed = 1;
1378 	for (;anyfreed;) {
1379 		anyfreed = 0;
1380 		TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1381 			if (bp->b_lblkno < trunclbn)
1382 				continue;
1383 			if (BUF_LOCK(bp,
1384 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1385 			    BO_MTX(bo)) == ENOLCK)
1386 				goto restart;
1387 
1388 			BO_LOCK(bo);
1389 			bremfree(bp);
1390 			BO_UNLOCK(bo);
1391 			bp->b_flags |= (B_INVAL | B_RELBUF);
1392 			bp->b_flags &= ~B_ASYNC;
1393 			brelse(bp);
1394 			anyfreed = 1;
1395 
1396 			BO_LOCK(bo);
1397 			if (nbp != NULL &&
1398 			    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1399 			    (nbp->b_vp != vp) ||
1400 			    (nbp->b_flags & B_DELWRI))) {
1401 				BO_UNLOCK(bo);
1402 				goto restart;
1403 			}
1404 		}
1405 
1406 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1407 			if (bp->b_lblkno < trunclbn)
1408 				continue;
1409 			if (BUF_LOCK(bp,
1410 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1411 			    BO_MTX(bo)) == ENOLCK)
1412 				goto restart;
1413 			BO_LOCK(bo);
1414 			bremfree(bp);
1415 			BO_UNLOCK(bo);
1416 			bp->b_flags |= (B_INVAL | B_RELBUF);
1417 			bp->b_flags &= ~B_ASYNC;
1418 			brelse(bp);
1419 			anyfreed = 1;
1420 
1421 			BO_LOCK(bo);
1422 			if (nbp != NULL &&
1423 			    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1424 			    (nbp->b_vp != vp) ||
1425 			    (nbp->b_flags & B_DELWRI) == 0)) {
1426 				BO_UNLOCK(bo);
1427 				goto restart;
1428 			}
1429 		}
1430 	}
1431 
1432 	if (length > 0) {
1433 restartsync:
1434 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1435 			if (bp->b_lblkno > 0)
1436 				continue;
1437 			/*
1438 			 * Since we hold the vnode lock this should only
1439 			 * fail if we're racing with the buf daemon.
1440 			 */
1441 			if (BUF_LOCK(bp,
1442 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1443 			    BO_MTX(bo)) == ENOLCK) {
1444 				goto restart;
1445 			}
1446 			VNASSERT((bp->b_flags & B_DELWRI), vp,
1447 			    ("buf(%p) on dirty queue without DELWRI", bp));
1448 
1449 			BO_LOCK(bo);
1450 			bremfree(bp);
1451 			BO_UNLOCK(bo);
1452 			bawrite(bp);
1453 			BO_LOCK(bo);
1454 			goto restartsync;
1455 		}
1456 	}
1457 
1458 	bufobj_wwait(bo, 0, 0);
1459 	BO_UNLOCK(bo);
1460 	vnode_pager_setsize(vp, length);
1461 
1462 	return (0);
1463 }
1464 
1465 /*
1466  * buf_splay() - splay tree core for the clean/dirty list of buffers in
1467  *		 a vnode.
1468  *
1469  *	NOTE: We have to deal with the special case of a background bitmap
1470  *	buffer, a situation where two buffers will have the same logical
1471  *	block offset.  We want (1) only the foreground buffer to be accessed
1472  *	in a lookup and (2) must differentiate between the foreground and
1473  *	background buffer in the splay tree algorithm because the splay
1474  *	tree cannot normally handle multiple entities with the same 'index'.
1475  *	We accomplish this by adding differentiating flags to the splay tree's
1476  *	numerical domain.
1477  */
1478 static
1479 struct buf *
1480 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1481 {
1482 	struct buf dummy;
1483 	struct buf *lefttreemax, *righttreemin, *y;
1484 
1485 	if (root == NULL)
1486 		return (NULL);
1487 	lefttreemax = righttreemin = &dummy;
1488 	for (;;) {
1489 		if (lblkno < root->b_lblkno ||
1490 		    (lblkno == root->b_lblkno &&
1491 		    (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1492 			if ((y = root->b_left) == NULL)
1493 				break;
1494 			if (lblkno < y->b_lblkno) {
1495 				/* Rotate right. */
1496 				root->b_left = y->b_right;
1497 				y->b_right = root;
1498 				root = y;
1499 				if ((y = root->b_left) == NULL)
1500 					break;
1501 			}
1502 			/* Link into the new root's right tree. */
1503 			righttreemin->b_left = root;
1504 			righttreemin = root;
1505 		} else if (lblkno > root->b_lblkno ||
1506 		    (lblkno == root->b_lblkno &&
1507 		    (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1508 			if ((y = root->b_right) == NULL)
1509 				break;
1510 			if (lblkno > y->b_lblkno) {
1511 				/* Rotate left. */
1512 				root->b_right = y->b_left;
1513 				y->b_left = root;
1514 				root = y;
1515 				if ((y = root->b_right) == NULL)
1516 					break;
1517 			}
1518 			/* Link into the new root's left tree. */
1519 			lefttreemax->b_right = root;
1520 			lefttreemax = root;
1521 		} else {
1522 			break;
1523 		}
1524 		root = y;
1525 	}
1526 	/* Assemble the new root. */
1527 	lefttreemax->b_right = root->b_left;
1528 	righttreemin->b_left = root->b_right;
1529 	root->b_left = dummy.b_right;
1530 	root->b_right = dummy.b_left;
1531 	return (root);
1532 }
1533 
1534 static void
1535 buf_vlist_remove(struct buf *bp)
1536 {
1537 	struct buf *root;
1538 	struct bufv *bv;
1539 
1540 	KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1541 	ASSERT_BO_LOCKED(bp->b_bufobj);
1542 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1543 	    (BX_VNDIRTY|BX_VNCLEAN),
1544 	    ("buf_vlist_remove: Buf %p is on two lists", bp));
1545 	if (bp->b_xflags & BX_VNDIRTY)
1546 		bv = &bp->b_bufobj->bo_dirty;
1547 	else
1548 		bv = &bp->b_bufobj->bo_clean;
1549 	if (bp != bv->bv_root) {
1550 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1551 		KASSERT(root == bp, ("splay lookup failed in remove"));
1552 	}
1553 	if (bp->b_left == NULL) {
1554 		root = bp->b_right;
1555 	} else {
1556 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1557 		root->b_right = bp->b_right;
1558 	}
1559 	bv->bv_root = root;
1560 	TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1561 	bv->bv_cnt--;
1562 	bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1563 }
1564 
1565 /*
1566  * Add the buffer to the sorted clean or dirty block list using a
1567  * splay tree algorithm.
1568  *
1569  * NOTE: xflags is passed as a constant, optimizing this inline function!
1570  */
1571 static void
1572 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1573 {
1574 	struct buf *root;
1575 	struct bufv *bv;
1576 
1577 	ASSERT_BO_LOCKED(bo);
1578 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1579 	    ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1580 	bp->b_xflags |= xflags;
1581 	if (xflags & BX_VNDIRTY)
1582 		bv = &bo->bo_dirty;
1583 	else
1584 		bv = &bo->bo_clean;
1585 
1586 	root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1587 	if (root == NULL) {
1588 		bp->b_left = NULL;
1589 		bp->b_right = NULL;
1590 		TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1591 	} else if (bp->b_lblkno < root->b_lblkno ||
1592 	    (bp->b_lblkno == root->b_lblkno &&
1593 	    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1594 		bp->b_left = root->b_left;
1595 		bp->b_right = root;
1596 		root->b_left = NULL;
1597 		TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1598 	} else {
1599 		bp->b_right = root->b_right;
1600 		bp->b_left = root;
1601 		root->b_right = NULL;
1602 		TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1603 	}
1604 	bv->bv_cnt++;
1605 	bv->bv_root = bp;
1606 }
1607 
1608 /*
1609  * Lookup a buffer using the splay tree.  Note that we specifically avoid
1610  * shadow buffers used in background bitmap writes.
1611  *
1612  * This code isn't quite efficient as it could be because we are maintaining
1613  * two sorted lists and do not know which list the block resides in.
1614  *
1615  * During a "make buildworld" the desired buffer is found at one of
1616  * the roots more than 60% of the time.  Thus, checking both roots
1617  * before performing either splay eliminates unnecessary splays on the
1618  * first tree splayed.
1619  */
1620 struct buf *
1621 gbincore(struct bufobj *bo, daddr_t lblkno)
1622 {
1623 	struct buf *bp;
1624 
1625 	ASSERT_BO_LOCKED(bo);
1626 	if ((bp = bo->bo_clean.bv_root) != NULL &&
1627 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1628 		return (bp);
1629 	if ((bp = bo->bo_dirty.bv_root) != NULL &&
1630 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1631 		return (bp);
1632 	if ((bp = bo->bo_clean.bv_root) != NULL) {
1633 		bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1634 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1635 			return (bp);
1636 	}
1637 	if ((bp = bo->bo_dirty.bv_root) != NULL) {
1638 		bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1639 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1640 			return (bp);
1641 	}
1642 	return (NULL);
1643 }
1644 
1645 /*
1646  * Associate a buffer with a vnode.
1647  */
1648 void
1649 bgetvp(struct vnode *vp, struct buf *bp)
1650 {
1651 	struct bufobj *bo;
1652 
1653 	bo = &vp->v_bufobj;
1654 	ASSERT_BO_LOCKED(bo);
1655 	VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1656 
1657 	CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1658 	VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1659 	    ("bgetvp: bp already attached! %p", bp));
1660 
1661 	vhold(vp);
1662 	bp->b_vp = vp;
1663 	bp->b_bufobj = bo;
1664 	/*
1665 	 * Insert onto list for new vnode.
1666 	 */
1667 	buf_vlist_add(bp, bo, BX_VNCLEAN);
1668 }
1669 
1670 /*
1671  * Disassociate a buffer from a vnode.
1672  */
1673 void
1674 brelvp(struct buf *bp)
1675 {
1676 	struct bufobj *bo;
1677 	struct vnode *vp;
1678 
1679 	CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1680 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1681 
1682 	/*
1683 	 * Delete from old vnode list, if on one.
1684 	 */
1685 	vp = bp->b_vp;		/* XXX */
1686 	bo = bp->b_bufobj;
1687 	BO_LOCK(bo);
1688 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1689 		buf_vlist_remove(bp);
1690 	else
1691 		panic("brelvp: Buffer %p not on queue.", bp);
1692 	if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1693 		bo->bo_flag &= ~BO_ONWORKLST;
1694 		mtx_lock(&sync_mtx);
1695 		LIST_REMOVE(bo, bo_synclist);
1696 		syncer_worklist_len--;
1697 		mtx_unlock(&sync_mtx);
1698 	}
1699 	bp->b_vp = NULL;
1700 	bp->b_bufobj = NULL;
1701 	BO_UNLOCK(bo);
1702 	vdrop(vp);
1703 }
1704 
1705 /*
1706  * Add an item to the syncer work queue.
1707  */
1708 static void
1709 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1710 {
1711 	int slot;
1712 
1713 	ASSERT_BO_LOCKED(bo);
1714 
1715 	mtx_lock(&sync_mtx);
1716 	if (bo->bo_flag & BO_ONWORKLST)
1717 		LIST_REMOVE(bo, bo_synclist);
1718 	else {
1719 		bo->bo_flag |= BO_ONWORKLST;
1720 		syncer_worklist_len++;
1721 	}
1722 
1723 	if (delay > syncer_maxdelay - 2)
1724 		delay = syncer_maxdelay - 2;
1725 	slot = (syncer_delayno + delay) & syncer_mask;
1726 
1727 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist);
1728 	mtx_unlock(&sync_mtx);
1729 }
1730 
1731 static int
1732 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1733 {
1734 	int error, len;
1735 
1736 	mtx_lock(&sync_mtx);
1737 	len = syncer_worklist_len - sync_vnode_count;
1738 	mtx_unlock(&sync_mtx);
1739 	error = SYSCTL_OUT(req, &len, sizeof(len));
1740 	return (error);
1741 }
1742 
1743 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1744     sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1745 
1746 static struct proc *updateproc;
1747 static void sched_sync(void);
1748 static struct kproc_desc up_kp = {
1749 	"syncer",
1750 	sched_sync,
1751 	&updateproc
1752 };
1753 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1754 
1755 static int
1756 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1757 {
1758 	struct vnode *vp;
1759 	struct mount *mp;
1760 
1761 	*bo = LIST_FIRST(slp);
1762 	if (*bo == NULL)
1763 		return (0);
1764 	vp = (*bo)->__bo_vnode;	/* XXX */
1765 	if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1766 		return (1);
1767 	/*
1768 	 * We use vhold in case the vnode does not
1769 	 * successfully sync.  vhold prevents the vnode from
1770 	 * going away when we unlock the sync_mtx so that
1771 	 * we can acquire the vnode interlock.
1772 	 */
1773 	vholdl(vp);
1774 	mtx_unlock(&sync_mtx);
1775 	VI_UNLOCK(vp);
1776 	if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1777 		vdrop(vp);
1778 		mtx_lock(&sync_mtx);
1779 		return (*bo == LIST_FIRST(slp));
1780 	}
1781 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1782 	(void) VOP_FSYNC(vp, MNT_LAZY, td);
1783 	VOP_UNLOCK(vp, 0);
1784 	vn_finished_write(mp);
1785 	BO_LOCK(*bo);
1786 	if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1787 		/*
1788 		 * Put us back on the worklist.  The worklist
1789 		 * routine will remove us from our current
1790 		 * position and then add us back in at a later
1791 		 * position.
1792 		 */
1793 		vn_syncer_add_to_worklist(*bo, syncdelay);
1794 	}
1795 	BO_UNLOCK(*bo);
1796 	vdrop(vp);
1797 	mtx_lock(&sync_mtx);
1798 	return (0);
1799 }
1800 
1801 /*
1802  * System filesystem synchronizer daemon.
1803  */
1804 static void
1805 sched_sync(void)
1806 {
1807 	struct synclist *next, *slp;
1808 	struct bufobj *bo;
1809 	long starttime;
1810 	struct thread *td = curthread;
1811 	int last_work_seen;
1812 	int net_worklist_len;
1813 	int syncer_final_iter;
1814 	int first_printf;
1815 	int error;
1816 
1817 	last_work_seen = 0;
1818 	syncer_final_iter = 0;
1819 	first_printf = 1;
1820 	syncer_state = SYNCER_RUNNING;
1821 	starttime = time_uptime;
1822 	td->td_pflags |= TDP_NORUNNINGBUF;
1823 
1824 	EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1825 	    SHUTDOWN_PRI_LAST);
1826 
1827 	mtx_lock(&sync_mtx);
1828 	for (;;) {
1829 		if (syncer_state == SYNCER_FINAL_DELAY &&
1830 		    syncer_final_iter == 0) {
1831 			mtx_unlock(&sync_mtx);
1832 			kproc_suspend_check(td->td_proc);
1833 			mtx_lock(&sync_mtx);
1834 		}
1835 		net_worklist_len = syncer_worklist_len - sync_vnode_count;
1836 		if (syncer_state != SYNCER_RUNNING &&
1837 		    starttime != time_uptime) {
1838 			if (first_printf) {
1839 				printf("\nSyncing disks, vnodes remaining...");
1840 				first_printf = 0;
1841 			}
1842 			printf("%d ", net_worklist_len);
1843 		}
1844 		starttime = time_uptime;
1845 
1846 		/*
1847 		 * Push files whose dirty time has expired.  Be careful
1848 		 * of interrupt race on slp queue.
1849 		 *
1850 		 * Skip over empty worklist slots when shutting down.
1851 		 */
1852 		do {
1853 			slp = &syncer_workitem_pending[syncer_delayno];
1854 			syncer_delayno += 1;
1855 			if (syncer_delayno == syncer_maxdelay)
1856 				syncer_delayno = 0;
1857 			next = &syncer_workitem_pending[syncer_delayno];
1858 			/*
1859 			 * If the worklist has wrapped since the
1860 			 * it was emptied of all but syncer vnodes,
1861 			 * switch to the FINAL_DELAY state and run
1862 			 * for one more second.
1863 			 */
1864 			if (syncer_state == SYNCER_SHUTTING_DOWN &&
1865 			    net_worklist_len == 0 &&
1866 			    last_work_seen == syncer_delayno) {
1867 				syncer_state = SYNCER_FINAL_DELAY;
1868 				syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1869 			}
1870 		} while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1871 		    syncer_worklist_len > 0);
1872 
1873 		/*
1874 		 * Keep track of the last time there was anything
1875 		 * on the worklist other than syncer vnodes.
1876 		 * Return to the SHUTTING_DOWN state if any
1877 		 * new work appears.
1878 		 */
1879 		if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1880 			last_work_seen = syncer_delayno;
1881 		if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1882 			syncer_state = SYNCER_SHUTTING_DOWN;
1883 		while (!LIST_EMPTY(slp)) {
1884 			error = sync_vnode(slp, &bo, td);
1885 			if (error == 1) {
1886 				LIST_REMOVE(bo, bo_synclist);
1887 				LIST_INSERT_HEAD(next, bo, bo_synclist);
1888 				continue;
1889 			}
1890 
1891 			if (first_printf == 0)
1892 				wdog_kern_pat(WD_LASTVAL);
1893 
1894 		}
1895 		if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1896 			syncer_final_iter--;
1897 		/*
1898 		 * The variable rushjob allows the kernel to speed up the
1899 		 * processing of the filesystem syncer process. A rushjob
1900 		 * value of N tells the filesystem syncer to process the next
1901 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1902 		 * is used by the soft update code to speed up the filesystem
1903 		 * syncer process when the incore state is getting so far
1904 		 * ahead of the disk that the kernel memory pool is being
1905 		 * threatened with exhaustion.
1906 		 */
1907 		if (rushjob > 0) {
1908 			rushjob -= 1;
1909 			continue;
1910 		}
1911 		/*
1912 		 * Just sleep for a short period of time between
1913 		 * iterations when shutting down to allow some I/O
1914 		 * to happen.
1915 		 *
1916 		 * If it has taken us less than a second to process the
1917 		 * current work, then wait. Otherwise start right over
1918 		 * again. We can still lose time if any single round
1919 		 * takes more than two seconds, but it does not really
1920 		 * matter as we are just trying to generally pace the
1921 		 * filesystem activity.
1922 		 */
1923 		if (syncer_state != SYNCER_RUNNING ||
1924 		    time_uptime == starttime) {
1925 			thread_lock(td);
1926 			sched_prio(td, PPAUSE);
1927 			thread_unlock(td);
1928 		}
1929 		if (syncer_state != SYNCER_RUNNING)
1930 			cv_timedwait(&sync_wakeup, &sync_mtx,
1931 			    hz / SYNCER_SHUTDOWN_SPEEDUP);
1932 		else if (time_uptime == starttime)
1933 			cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1934 	}
1935 }
1936 
1937 /*
1938  * Request the syncer daemon to speed up its work.
1939  * We never push it to speed up more than half of its
1940  * normal turn time, otherwise it could take over the cpu.
1941  */
1942 int
1943 speedup_syncer(void)
1944 {
1945 	int ret = 0;
1946 
1947 	mtx_lock(&sync_mtx);
1948 	if (rushjob < syncdelay / 2) {
1949 		rushjob += 1;
1950 		stat_rush_requests += 1;
1951 		ret = 1;
1952 	}
1953 	mtx_unlock(&sync_mtx);
1954 	cv_broadcast(&sync_wakeup);
1955 	return (ret);
1956 }
1957 
1958 /*
1959  * Tell the syncer to speed up its work and run though its work
1960  * list several times, then tell it to shut down.
1961  */
1962 static void
1963 syncer_shutdown(void *arg, int howto)
1964 {
1965 
1966 	if (howto & RB_NOSYNC)
1967 		return;
1968 	mtx_lock(&sync_mtx);
1969 	syncer_state = SYNCER_SHUTTING_DOWN;
1970 	rushjob = 0;
1971 	mtx_unlock(&sync_mtx);
1972 	cv_broadcast(&sync_wakeup);
1973 	kproc_shutdown(arg, howto);
1974 }
1975 
1976 /*
1977  * Reassign a buffer from one vnode to another.
1978  * Used to assign file specific control information
1979  * (indirect blocks) to the vnode to which they belong.
1980  */
1981 void
1982 reassignbuf(struct buf *bp)
1983 {
1984 	struct vnode *vp;
1985 	struct bufobj *bo;
1986 	int delay;
1987 #ifdef INVARIANTS
1988 	struct bufv *bv;
1989 #endif
1990 
1991 	vp = bp->b_vp;
1992 	bo = bp->b_bufobj;
1993 	++reassignbufcalls;
1994 
1995 	CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1996 	    bp, bp->b_vp, bp->b_flags);
1997 	/*
1998 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1999 	 * is not fully linked in.
2000 	 */
2001 	if (bp->b_flags & B_PAGING)
2002 		panic("cannot reassign paging buffer");
2003 
2004 	/*
2005 	 * Delete from old vnode list, if on one.
2006 	 */
2007 	BO_LOCK(bo);
2008 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
2009 		buf_vlist_remove(bp);
2010 	else
2011 		panic("reassignbuf: Buffer %p not on queue.", bp);
2012 	/*
2013 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
2014 	 * of clean buffers.
2015 	 */
2016 	if (bp->b_flags & B_DELWRI) {
2017 		if ((bo->bo_flag & BO_ONWORKLST) == 0) {
2018 			switch (vp->v_type) {
2019 			case VDIR:
2020 				delay = dirdelay;
2021 				break;
2022 			case VCHR:
2023 				delay = metadelay;
2024 				break;
2025 			default:
2026 				delay = filedelay;
2027 			}
2028 			vn_syncer_add_to_worklist(bo, delay);
2029 		}
2030 		buf_vlist_add(bp, bo, BX_VNDIRTY);
2031 	} else {
2032 		buf_vlist_add(bp, bo, BX_VNCLEAN);
2033 
2034 		if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
2035 			mtx_lock(&sync_mtx);
2036 			LIST_REMOVE(bo, bo_synclist);
2037 			syncer_worklist_len--;
2038 			mtx_unlock(&sync_mtx);
2039 			bo->bo_flag &= ~BO_ONWORKLST;
2040 		}
2041 	}
2042 #ifdef INVARIANTS
2043 	bv = &bo->bo_clean;
2044 	bp = TAILQ_FIRST(&bv->bv_hd);
2045 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2046 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2047 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
2048 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2049 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2050 	bv = &bo->bo_dirty;
2051 	bp = TAILQ_FIRST(&bv->bv_hd);
2052 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2053 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2054 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
2055 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2056 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2057 #endif
2058 	BO_UNLOCK(bo);
2059 }
2060 
2061 /*
2062  * Increment the use and hold counts on the vnode, taking care to reference
2063  * the driver's usecount if this is a chardev.  The vholdl() will remove
2064  * the vnode from the free list if it is presently free.  Requires the
2065  * vnode interlock and returns with it held.
2066  */
2067 static void
2068 v_incr_usecount(struct vnode *vp)
2069 {
2070 
2071 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2072 	vp->v_usecount++;
2073 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2074 		dev_lock();
2075 		vp->v_rdev->si_usecount++;
2076 		dev_unlock();
2077 	}
2078 	vholdl(vp);
2079 }
2080 
2081 /*
2082  * Turn a holdcnt into a use+holdcnt such that only one call to
2083  * v_decr_usecount is needed.
2084  */
2085 static void
2086 v_upgrade_usecount(struct vnode *vp)
2087 {
2088 
2089 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2090 	vp->v_usecount++;
2091 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2092 		dev_lock();
2093 		vp->v_rdev->si_usecount++;
2094 		dev_unlock();
2095 	}
2096 }
2097 
2098 /*
2099  * Decrement the vnode use and hold count along with the driver's usecount
2100  * if this is a chardev.  The vdropl() below releases the vnode interlock
2101  * as it may free the vnode.
2102  */
2103 static void
2104 v_decr_usecount(struct vnode *vp)
2105 {
2106 
2107 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2108 	VNASSERT(vp->v_usecount > 0, vp,
2109 	    ("v_decr_usecount: negative usecount"));
2110 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2111 	vp->v_usecount--;
2112 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2113 		dev_lock();
2114 		vp->v_rdev->si_usecount--;
2115 		dev_unlock();
2116 	}
2117 	vdropl(vp);
2118 }
2119 
2120 /*
2121  * Decrement only the use count and driver use count.  This is intended to
2122  * be paired with a follow on vdropl() to release the remaining hold count.
2123  * In this way we may vgone() a vnode with a 0 usecount without risk of
2124  * having it end up on a free list because the hold count is kept above 0.
2125  */
2126 static void
2127 v_decr_useonly(struct vnode *vp)
2128 {
2129 
2130 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2131 	VNASSERT(vp->v_usecount > 0, vp,
2132 	    ("v_decr_useonly: negative usecount"));
2133 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2134 	vp->v_usecount--;
2135 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2136 		dev_lock();
2137 		vp->v_rdev->si_usecount--;
2138 		dev_unlock();
2139 	}
2140 }
2141 
2142 /*
2143  * Grab a particular vnode from the free list, increment its
2144  * reference count and lock it.  VI_DOOMED is set if the vnode
2145  * is being destroyed.  Only callers who specify LK_RETRY will
2146  * see doomed vnodes.  If inactive processing was delayed in
2147  * vput try to do it here.
2148  */
2149 int
2150 vget(struct vnode *vp, int flags, struct thread *td)
2151 {
2152 	int error;
2153 
2154 	error = 0;
2155 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2156 	    ("vget: invalid lock operation"));
2157 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2158 
2159 	if ((flags & LK_INTERLOCK) == 0)
2160 		VI_LOCK(vp);
2161 	vholdl(vp);
2162 	if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2163 		vdrop(vp);
2164 		CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2165 		    vp);
2166 		return (error);
2167 	}
2168 	if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2169 		panic("vget: vn_lock failed to return ENOENT\n");
2170 	VI_LOCK(vp);
2171 	/* Upgrade our holdcnt to a usecount. */
2172 	v_upgrade_usecount(vp);
2173 	/*
2174 	 * We don't guarantee that any particular close will
2175 	 * trigger inactive processing so just make a best effort
2176 	 * here at preventing a reference to a removed file.  If
2177 	 * we don't succeed no harm is done.
2178 	 */
2179 	if (vp->v_iflag & VI_OWEINACT) {
2180 		if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2181 		    (flags & LK_NOWAIT) == 0)
2182 			vinactive(vp, td);
2183 		vp->v_iflag &= ~VI_OWEINACT;
2184 	}
2185 	VI_UNLOCK(vp);
2186 	return (0);
2187 }
2188 
2189 /*
2190  * Increase the reference count of a vnode.
2191  */
2192 void
2193 vref(struct vnode *vp)
2194 {
2195 
2196 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2197 	VI_LOCK(vp);
2198 	v_incr_usecount(vp);
2199 	VI_UNLOCK(vp);
2200 }
2201 
2202 /*
2203  * Return reference count of a vnode.
2204  *
2205  * The results of this call are only guaranteed when some mechanism other
2206  * than the VI lock is used to stop other processes from gaining references
2207  * to the vnode.  This may be the case if the caller holds the only reference.
2208  * This is also useful when stale data is acceptable as race conditions may
2209  * be accounted for by some other means.
2210  */
2211 int
2212 vrefcnt(struct vnode *vp)
2213 {
2214 	int usecnt;
2215 
2216 	VI_LOCK(vp);
2217 	usecnt = vp->v_usecount;
2218 	VI_UNLOCK(vp);
2219 
2220 	return (usecnt);
2221 }
2222 
2223 #define	VPUTX_VRELE	1
2224 #define	VPUTX_VPUT	2
2225 #define	VPUTX_VUNREF	3
2226 
2227 static void
2228 vputx(struct vnode *vp, int func)
2229 {
2230 	int error;
2231 
2232 	KASSERT(vp != NULL, ("vputx: null vp"));
2233 	if (func == VPUTX_VUNREF)
2234 		ASSERT_VOP_LOCKED(vp, "vunref");
2235 	else if (func == VPUTX_VPUT)
2236 		ASSERT_VOP_LOCKED(vp, "vput");
2237 	else
2238 		KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2239 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2240 	VI_LOCK(vp);
2241 
2242 	/* Skip this v_writecount check if we're going to panic below. */
2243 	VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2244 	    ("vputx: missed vn_close"));
2245 	error = 0;
2246 
2247 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2248 	    vp->v_usecount == 1)) {
2249 		if (func == VPUTX_VPUT)
2250 			VOP_UNLOCK(vp, 0);
2251 		v_decr_usecount(vp);
2252 		return;
2253 	}
2254 
2255 	if (vp->v_usecount != 1) {
2256 		vprint("vputx: negative ref count", vp);
2257 		panic("vputx: negative ref cnt");
2258 	}
2259 	CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2260 	/*
2261 	 * We want to hold the vnode until the inactive finishes to
2262 	 * prevent vgone() races.  We drop the use count here and the
2263 	 * hold count below when we're done.
2264 	 */
2265 	v_decr_useonly(vp);
2266 	/*
2267 	 * We must call VOP_INACTIVE with the node locked. Mark
2268 	 * as VI_DOINGINACT to avoid recursion.
2269 	 */
2270 	vp->v_iflag |= VI_OWEINACT;
2271 	switch (func) {
2272 	case VPUTX_VRELE:
2273 		error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2274 		VI_LOCK(vp);
2275 		break;
2276 	case VPUTX_VPUT:
2277 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2278 			error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
2279 			    LK_NOWAIT);
2280 			VI_LOCK(vp);
2281 		}
2282 		break;
2283 	case VPUTX_VUNREF:
2284 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
2285 			error = EBUSY;
2286 		break;
2287 	}
2288 	if (vp->v_usecount > 0)
2289 		vp->v_iflag &= ~VI_OWEINACT;
2290 	if (error == 0) {
2291 		if (vp->v_iflag & VI_OWEINACT)
2292 			vinactive(vp, curthread);
2293 		if (func != VPUTX_VUNREF)
2294 			VOP_UNLOCK(vp, 0);
2295 	}
2296 	vdropl(vp);
2297 }
2298 
2299 /*
2300  * Vnode put/release.
2301  * If count drops to zero, call inactive routine and return to freelist.
2302  */
2303 void
2304 vrele(struct vnode *vp)
2305 {
2306 
2307 	vputx(vp, VPUTX_VRELE);
2308 }
2309 
2310 /*
2311  * Release an already locked vnode.  This give the same effects as
2312  * unlock+vrele(), but takes less time and avoids releasing and
2313  * re-aquiring the lock (as vrele() acquires the lock internally.)
2314  */
2315 void
2316 vput(struct vnode *vp)
2317 {
2318 
2319 	vputx(vp, VPUTX_VPUT);
2320 }
2321 
2322 /*
2323  * Release an exclusively locked vnode. Do not unlock the vnode lock.
2324  */
2325 void
2326 vunref(struct vnode *vp)
2327 {
2328 
2329 	vputx(vp, VPUTX_VUNREF);
2330 }
2331 
2332 /*
2333  * Somebody doesn't want the vnode recycled.
2334  */
2335 void
2336 vhold(struct vnode *vp)
2337 {
2338 
2339 	VI_LOCK(vp);
2340 	vholdl(vp);
2341 	VI_UNLOCK(vp);
2342 }
2343 
2344 /*
2345  * Increase the hold count and activate if this is the first reference.
2346  */
2347 void
2348 vholdl(struct vnode *vp)
2349 {
2350 	struct mount *mp;
2351 
2352 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2353 	vp->v_holdcnt++;
2354 	if (!VSHOULDBUSY(vp))
2355 		return;
2356 	ASSERT_VI_LOCKED(vp, "vholdl");
2357 	VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
2358 	VNASSERT(vp->v_op != NULL, vp, ("vholdl: vnode already reclaimed."));
2359 	/*
2360 	 * Remove a vnode from the free list, mark it as in use,
2361 	 * and put it on the active list.
2362 	 */
2363 	mtx_lock(&vnode_free_list_mtx);
2364 	TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
2365 	freevnodes--;
2366 	vp->v_iflag &= ~(VI_FREE|VI_AGE);
2367 	KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
2368 	    ("Activating already active vnode"));
2369 	vp->v_iflag |= VI_ACTIVE;
2370 	mp = vp->v_mount;
2371 	TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
2372 	mp->mnt_activevnodelistsize++;
2373 	mtx_unlock(&vnode_free_list_mtx);
2374 }
2375 
2376 /*
2377  * Note that there is one less who cares about this vnode.
2378  * vdrop() is the opposite of vhold().
2379  */
2380 void
2381 vdrop(struct vnode *vp)
2382 {
2383 
2384 	VI_LOCK(vp);
2385 	vdropl(vp);
2386 }
2387 
2388 /*
2389  * Drop the hold count of the vnode.  If this is the last reference to
2390  * the vnode we place it on the free list unless it has been vgone'd
2391  * (marked VI_DOOMED) in which case we will free it.
2392  */
2393 void
2394 vdropl(struct vnode *vp)
2395 {
2396 	struct bufobj *bo;
2397 	struct mount *mp;
2398 	int active;
2399 
2400 	ASSERT_VI_LOCKED(vp, "vdropl");
2401 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2402 	if (vp->v_holdcnt <= 0)
2403 		panic("vdrop: holdcnt %d", vp->v_holdcnt);
2404 	vp->v_holdcnt--;
2405 	if (vp->v_holdcnt > 0) {
2406 		VI_UNLOCK(vp);
2407 		return;
2408 	}
2409 	if ((vp->v_iflag & VI_DOOMED) == 0) {
2410 		/*
2411 		 * Mark a vnode as free: remove it from its active list
2412 		 * and put it up for recycling on the freelist.
2413 		 */
2414 		VNASSERT(vp->v_op != NULL, vp,
2415 		    ("vdropl: vnode already reclaimed."));
2416 		VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2417 		    ("vnode already free"));
2418 		VNASSERT(VSHOULDFREE(vp), vp,
2419 		    ("vdropl: freeing when we shouldn't"));
2420 		active = vp->v_iflag & VI_ACTIVE;
2421 		vp->v_iflag &= ~VI_ACTIVE;
2422 		mp = vp->v_mount;
2423 		mtx_lock(&vnode_free_list_mtx);
2424 		if (active) {
2425 			TAILQ_REMOVE(&mp->mnt_activevnodelist, vp,
2426 			    v_actfreelist);
2427 			mp->mnt_activevnodelistsize--;
2428 		}
2429 		if (vp->v_iflag & VI_AGE) {
2430 			TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_actfreelist);
2431 		} else {
2432 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
2433 		}
2434 		freevnodes++;
2435 		vp->v_iflag &= ~VI_AGE;
2436 		vp->v_iflag |= VI_FREE;
2437 		mtx_unlock(&vnode_free_list_mtx);
2438 		VI_UNLOCK(vp);
2439 		return;
2440 	}
2441 	/*
2442 	 * The vnode has been marked for destruction, so free it.
2443 	 */
2444 	CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp);
2445 	mtx_lock(&vnode_free_list_mtx);
2446 	numvnodes--;
2447 	mtx_unlock(&vnode_free_list_mtx);
2448 	bo = &vp->v_bufobj;
2449 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2450 	    ("cleaned vnode still on the free list."));
2451 	VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
2452 	VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
2453 	VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
2454 	VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
2455 	VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
2456 	VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
2457 	VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
2458 	VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
2459 	VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
2460 	VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
2461 	VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
2462 	VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
2463 	VI_UNLOCK(vp);
2464 #ifdef MAC
2465 	mac_vnode_destroy(vp);
2466 #endif
2467 	if (vp->v_pollinfo != NULL)
2468 		destroy_vpollinfo(vp->v_pollinfo);
2469 #ifdef INVARIANTS
2470 	/* XXX Elsewhere we detect an already freed vnode via NULL v_op. */
2471 	vp->v_op = NULL;
2472 #endif
2473 	rangelock_destroy(&vp->v_rl);
2474 	lockdestroy(vp->v_vnlock);
2475 	mtx_destroy(&vp->v_interlock);
2476 	mtx_destroy(BO_MTX(bo));
2477 	uma_zfree(vnode_zone, vp);
2478 }
2479 
2480 /*
2481  * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2482  * flags.  DOINGINACT prevents us from recursing in calls to vinactive.
2483  * OWEINACT tracks whether a vnode missed a call to inactive due to a
2484  * failed lock upgrade.
2485  */
2486 void
2487 vinactive(struct vnode *vp, struct thread *td)
2488 {
2489 	struct vm_object *obj;
2490 
2491 	ASSERT_VOP_ELOCKED(vp, "vinactive");
2492 	ASSERT_VI_LOCKED(vp, "vinactive");
2493 	VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2494 	    ("vinactive: recursed on VI_DOINGINACT"));
2495 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2496 	vp->v_iflag |= VI_DOINGINACT;
2497 	vp->v_iflag &= ~VI_OWEINACT;
2498 	VI_UNLOCK(vp);
2499 	/*
2500 	 * Before moving off the active list, we must be sure that any
2501 	 * modified pages are on the vnode's dirty list since these will
2502 	 * no longer be checked once the vnode is on the inactive list.
2503 	 * Because the vnode vm object keeps a hold reference on the vnode
2504 	 * if there is at least one resident non-cached page, the vnode
2505 	 * cannot leave the active list without the page cleanup done.
2506 	 */
2507 	obj = vp->v_object;
2508 	if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
2509 		VM_OBJECT_LOCK(obj);
2510 		vm_object_page_clean(obj, 0, 0, OBJPC_NOSYNC);
2511 		VM_OBJECT_UNLOCK(obj);
2512 	}
2513 	VOP_INACTIVE(vp, td);
2514 	VI_LOCK(vp);
2515 	VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2516 	    ("vinactive: lost VI_DOINGINACT"));
2517 	vp->v_iflag &= ~VI_DOINGINACT;
2518 }
2519 
2520 /*
2521  * Remove any vnodes in the vnode table belonging to mount point mp.
2522  *
2523  * If FORCECLOSE is not specified, there should not be any active ones,
2524  * return error if any are found (nb: this is a user error, not a
2525  * system error). If FORCECLOSE is specified, detach any active vnodes
2526  * that are found.
2527  *
2528  * If WRITECLOSE is set, only flush out regular file vnodes open for
2529  * writing.
2530  *
2531  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2532  *
2533  * `rootrefs' specifies the base reference count for the root vnode
2534  * of this filesystem. The root vnode is considered busy if its
2535  * v_usecount exceeds this value. On a successful return, vflush(, td)
2536  * will call vrele() on the root vnode exactly rootrefs times.
2537  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2538  * be zero.
2539  */
2540 #ifdef DIAGNOSTIC
2541 static int busyprt = 0;		/* print out busy vnodes */
2542 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
2543 #endif
2544 
2545 int
2546 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
2547 {
2548 	struct vnode *vp, *mvp, *rootvp = NULL;
2549 	struct vattr vattr;
2550 	int busy = 0, error;
2551 
2552 	CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2553 	    rootrefs, flags);
2554 	if (rootrefs > 0) {
2555 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2556 		    ("vflush: bad args"));
2557 		/*
2558 		 * Get the filesystem root vnode. We can vput() it
2559 		 * immediately, since with rootrefs > 0, it won't go away.
2560 		 */
2561 		if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2562 			CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2563 			    __func__, error);
2564 			return (error);
2565 		}
2566 		vput(rootvp);
2567 	}
2568 loop:
2569 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
2570 		vholdl(vp);
2571 		error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2572 		if (error) {
2573 			vdrop(vp);
2574 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2575 			goto loop;
2576 		}
2577 		/*
2578 		 * Skip over a vnodes marked VV_SYSTEM.
2579 		 */
2580 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2581 			VOP_UNLOCK(vp, 0);
2582 			vdrop(vp);
2583 			continue;
2584 		}
2585 		/*
2586 		 * If WRITECLOSE is set, flush out unlinked but still open
2587 		 * files (even if open only for reading) and regular file
2588 		 * vnodes open for writing.
2589 		 */
2590 		if (flags & WRITECLOSE) {
2591 			if (vp->v_object != NULL) {
2592 				VM_OBJECT_LOCK(vp->v_object);
2593 				vm_object_page_clean(vp->v_object, 0, 0, 0);
2594 				VM_OBJECT_UNLOCK(vp->v_object);
2595 			}
2596 			error = VOP_FSYNC(vp, MNT_WAIT, td);
2597 			if (error != 0) {
2598 				VOP_UNLOCK(vp, 0);
2599 				vdrop(vp);
2600 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2601 				return (error);
2602 			}
2603 			error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2604 			VI_LOCK(vp);
2605 
2606 			if ((vp->v_type == VNON ||
2607 			    (error == 0 && vattr.va_nlink > 0)) &&
2608 			    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2609 				VOP_UNLOCK(vp, 0);
2610 				vdropl(vp);
2611 				continue;
2612 			}
2613 		} else
2614 			VI_LOCK(vp);
2615 		/*
2616 		 * With v_usecount == 0, all we need to do is clear out the
2617 		 * vnode data structures and we are done.
2618 		 *
2619 		 * If FORCECLOSE is set, forcibly close the vnode.
2620 		 */
2621 		if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2622 			VNASSERT(vp->v_usecount == 0 ||
2623 			    (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2624 			    ("device VNODE %p is FORCECLOSED", vp));
2625 			vgonel(vp);
2626 		} else {
2627 			busy++;
2628 #ifdef DIAGNOSTIC
2629 			if (busyprt)
2630 				vprint("vflush: busy vnode", vp);
2631 #endif
2632 		}
2633 		VOP_UNLOCK(vp, 0);
2634 		vdropl(vp);
2635 	}
2636 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2637 		/*
2638 		 * If just the root vnode is busy, and if its refcount
2639 		 * is equal to `rootrefs', then go ahead and kill it.
2640 		 */
2641 		VI_LOCK(rootvp);
2642 		KASSERT(busy > 0, ("vflush: not busy"));
2643 		VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2644 		    ("vflush: usecount %d < rootrefs %d",
2645 		     rootvp->v_usecount, rootrefs));
2646 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2647 			VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2648 			vgone(rootvp);
2649 			VOP_UNLOCK(rootvp, 0);
2650 			busy = 0;
2651 		} else
2652 			VI_UNLOCK(rootvp);
2653 	}
2654 	if (busy) {
2655 		CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2656 		    busy);
2657 		return (EBUSY);
2658 	}
2659 	for (; rootrefs > 0; rootrefs--)
2660 		vrele(rootvp);
2661 	return (0);
2662 }
2663 
2664 /*
2665  * Recycle an unused vnode to the front of the free list.
2666  */
2667 int
2668 vrecycle(struct vnode *vp)
2669 {
2670 	int recycled;
2671 
2672 	ASSERT_VOP_ELOCKED(vp, "vrecycle");
2673 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2674 	recycled = 0;
2675 	VI_LOCK(vp);
2676 	if (vp->v_usecount == 0) {
2677 		recycled = 1;
2678 		vgonel(vp);
2679 	}
2680 	VI_UNLOCK(vp);
2681 	return (recycled);
2682 }
2683 
2684 /*
2685  * Eliminate all activity associated with a vnode
2686  * in preparation for reuse.
2687  */
2688 void
2689 vgone(struct vnode *vp)
2690 {
2691 	VI_LOCK(vp);
2692 	vgonel(vp);
2693 	VI_UNLOCK(vp);
2694 }
2695 
2696 static void
2697 vgonel_reclaim_lowervp_vfs(struct mount *mp __unused,
2698     struct vnode *lowervp __unused)
2699 {
2700 }
2701 
2702 /*
2703  * Notify upper mounts about reclaimed vnode.
2704  */
2705 static void
2706 vgonel_reclaim_lowervp(struct vnode *vp)
2707 {
2708 	static struct vfsops vgonel_vfsops = {
2709 		.vfs_reclaim_lowervp = vgonel_reclaim_lowervp_vfs
2710 	};
2711 	struct mount *mp, *ump, *mmp;
2712 
2713 	mp = vp->v_mount;
2714 	if (mp == NULL)
2715 		return;
2716 
2717 	MNT_ILOCK(mp);
2718 	if (TAILQ_EMPTY(&mp->mnt_uppers))
2719 		goto unlock;
2720 	MNT_IUNLOCK(mp);
2721 	mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO);
2722 	mmp->mnt_op = &vgonel_vfsops;
2723 	mmp->mnt_kern_flag |= MNTK_MARKER;
2724 	MNT_ILOCK(mp);
2725 	mp->mnt_kern_flag |= MNTK_VGONE_UPPER;
2726 	for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) {
2727 		if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) {
2728 			ump = TAILQ_NEXT(ump, mnt_upper_link);
2729 			continue;
2730 		}
2731 		TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link);
2732 		MNT_IUNLOCK(mp);
2733 		VFS_RECLAIM_LOWERVP(ump, vp);
2734 		MNT_ILOCK(mp);
2735 		ump = TAILQ_NEXT(mmp, mnt_upper_link);
2736 		TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link);
2737 	}
2738 	free(mmp, M_TEMP);
2739 	mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER;
2740 	if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) {
2741 		mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER;
2742 		wakeup(&mp->mnt_uppers);
2743 	}
2744 unlock:
2745 	MNT_IUNLOCK(mp);
2746 }
2747 
2748 /*
2749  * vgone, with the vp interlock held.
2750  */
2751 void
2752 vgonel(struct vnode *vp)
2753 {
2754 	struct thread *td;
2755 	int oweinact;
2756 	int active;
2757 	struct mount *mp;
2758 
2759 	ASSERT_VOP_ELOCKED(vp, "vgonel");
2760 	ASSERT_VI_LOCKED(vp, "vgonel");
2761 	VNASSERT(vp->v_holdcnt, vp,
2762 	    ("vgonel: vp %p has no reference.", vp));
2763 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2764 	td = curthread;
2765 
2766 	/*
2767 	 * Don't vgonel if we're already doomed.
2768 	 */
2769 	if (vp->v_iflag & VI_DOOMED)
2770 		return;
2771 	vp->v_iflag |= VI_DOOMED;
2772 
2773 	/*
2774 	 * Check to see if the vnode is in use.  If so, we have to call
2775 	 * VOP_CLOSE() and VOP_INACTIVE().
2776 	 */
2777 	active = vp->v_usecount;
2778 	oweinact = (vp->v_iflag & VI_OWEINACT);
2779 	VI_UNLOCK(vp);
2780 	vgonel_reclaim_lowervp(vp);
2781 
2782 	/*
2783 	 * Clean out any buffers associated with the vnode.
2784 	 * If the flush fails, just toss the buffers.
2785 	 */
2786 	mp = NULL;
2787 	if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2788 		(void) vn_start_secondary_write(vp, &mp, V_WAIT);
2789 	if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
2790 		vinvalbuf(vp, 0, 0, 0);
2791 
2792 	/*
2793 	 * If purging an active vnode, it must be closed and
2794 	 * deactivated before being reclaimed.
2795 	 */
2796 	if (active)
2797 		VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2798 	if (oweinact || active) {
2799 		VI_LOCK(vp);
2800 		if ((vp->v_iflag & VI_DOINGINACT) == 0)
2801 			vinactive(vp, td);
2802 		VI_UNLOCK(vp);
2803 	}
2804 	if (vp->v_type == VSOCK)
2805 		vfs_unp_reclaim(vp);
2806 	/*
2807 	 * Reclaim the vnode.
2808 	 */
2809 	if (VOP_RECLAIM(vp, td))
2810 		panic("vgone: cannot reclaim");
2811 	if (mp != NULL)
2812 		vn_finished_secondary_write(mp);
2813 	VNASSERT(vp->v_object == NULL, vp,
2814 	    ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2815 	/*
2816 	 * Clear the advisory locks and wake up waiting threads.
2817 	 */
2818 	(void)VOP_ADVLOCKPURGE(vp);
2819 	/*
2820 	 * Delete from old mount point vnode list.
2821 	 */
2822 	delmntque(vp);
2823 	cache_purge(vp);
2824 	/*
2825 	 * Done with purge, reset to the standard lock and invalidate
2826 	 * the vnode.
2827 	 */
2828 	VI_LOCK(vp);
2829 	vp->v_vnlock = &vp->v_lock;
2830 	vp->v_op = &dead_vnodeops;
2831 	vp->v_tag = "none";
2832 	vp->v_type = VBAD;
2833 }
2834 
2835 /*
2836  * Calculate the total number of references to a special device.
2837  */
2838 int
2839 vcount(struct vnode *vp)
2840 {
2841 	int count;
2842 
2843 	dev_lock();
2844 	count = vp->v_rdev->si_usecount;
2845 	dev_unlock();
2846 	return (count);
2847 }
2848 
2849 /*
2850  * Same as above, but using the struct cdev *as argument
2851  */
2852 int
2853 count_dev(struct cdev *dev)
2854 {
2855 	int count;
2856 
2857 	dev_lock();
2858 	count = dev->si_usecount;
2859 	dev_unlock();
2860 	return(count);
2861 }
2862 
2863 /*
2864  * Print out a description of a vnode.
2865  */
2866 static char *typename[] =
2867 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2868  "VMARKER"};
2869 
2870 void
2871 vn_printf(struct vnode *vp, const char *fmt, ...)
2872 {
2873 	va_list ap;
2874 	char buf[256], buf2[16];
2875 	u_long flags;
2876 
2877 	va_start(ap, fmt);
2878 	vprintf(fmt, ap);
2879 	va_end(ap);
2880 	printf("%p: ", (void *)vp);
2881 	printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2882 	printf("    usecount %d, writecount %d, refcount %d mountedhere %p\n",
2883 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2884 	buf[0] = '\0';
2885 	buf[1] = '\0';
2886 	if (vp->v_vflag & VV_ROOT)
2887 		strlcat(buf, "|VV_ROOT", sizeof(buf));
2888 	if (vp->v_vflag & VV_ISTTY)
2889 		strlcat(buf, "|VV_ISTTY", sizeof(buf));
2890 	if (vp->v_vflag & VV_NOSYNC)
2891 		strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2892 	if (vp->v_vflag & VV_ETERNALDEV)
2893 		strlcat(buf, "|VV_ETERNALDEV", sizeof(buf));
2894 	if (vp->v_vflag & VV_CACHEDLABEL)
2895 		strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2896 	if (vp->v_vflag & VV_TEXT)
2897 		strlcat(buf, "|VV_TEXT", sizeof(buf));
2898 	if (vp->v_vflag & VV_COPYONWRITE)
2899 		strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2900 	if (vp->v_vflag & VV_SYSTEM)
2901 		strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2902 	if (vp->v_vflag & VV_PROCDEP)
2903 		strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2904 	if (vp->v_vflag & VV_NOKNOTE)
2905 		strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2906 	if (vp->v_vflag & VV_DELETED)
2907 		strlcat(buf, "|VV_DELETED", sizeof(buf));
2908 	if (vp->v_vflag & VV_MD)
2909 		strlcat(buf, "|VV_MD", sizeof(buf));
2910 	if (vp->v_vflag & VV_FORCEINSMQ)
2911 		strlcat(buf, "|VV_FORCEINSMQ", sizeof(buf));
2912 	flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | VV_ETERNALDEV |
2913 	    VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2914 	    VV_NOKNOTE | VV_DELETED | VV_MD | VV_FORCEINSMQ);
2915 	if (flags != 0) {
2916 		snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2917 		strlcat(buf, buf2, sizeof(buf));
2918 	}
2919 	if (vp->v_iflag & VI_MOUNT)
2920 		strlcat(buf, "|VI_MOUNT", sizeof(buf));
2921 	if (vp->v_iflag & VI_AGE)
2922 		strlcat(buf, "|VI_AGE", sizeof(buf));
2923 	if (vp->v_iflag & VI_DOOMED)
2924 		strlcat(buf, "|VI_DOOMED", sizeof(buf));
2925 	if (vp->v_iflag & VI_FREE)
2926 		strlcat(buf, "|VI_FREE", sizeof(buf));
2927 	if (vp->v_iflag & VI_ACTIVE)
2928 		strlcat(buf, "|VI_ACTIVE", sizeof(buf));
2929 	if (vp->v_iflag & VI_DOINGINACT)
2930 		strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2931 	if (vp->v_iflag & VI_OWEINACT)
2932 		strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2933 	flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2934 	    VI_ACTIVE | VI_DOINGINACT | VI_OWEINACT);
2935 	if (flags != 0) {
2936 		snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2937 		strlcat(buf, buf2, sizeof(buf));
2938 	}
2939 	printf("    flags (%s)\n", buf + 1);
2940 	if (mtx_owned(VI_MTX(vp)))
2941 		printf(" VI_LOCKed");
2942 	if (vp->v_object != NULL)
2943 		printf("    v_object %p ref %d pages %d\n",
2944 		    vp->v_object, vp->v_object->ref_count,
2945 		    vp->v_object->resident_page_count);
2946 	printf("    ");
2947 	lockmgr_printinfo(vp->v_vnlock);
2948 	if (vp->v_data != NULL)
2949 		VOP_PRINT(vp);
2950 }
2951 
2952 #ifdef DDB
2953 /*
2954  * List all of the locked vnodes in the system.
2955  * Called when debugging the kernel.
2956  */
2957 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2958 {
2959 	struct mount *mp, *nmp;
2960 	struct vnode *vp;
2961 
2962 	/*
2963 	 * Note: because this is DDB, we can't obey the locking semantics
2964 	 * for these structures, which means we could catch an inconsistent
2965 	 * state and dereference a nasty pointer.  Not much to be done
2966 	 * about that.
2967 	 */
2968 	db_printf("Locked vnodes\n");
2969 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2970 		nmp = TAILQ_NEXT(mp, mnt_list);
2971 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2972 			if (vp->v_type != VMARKER &&
2973 			    VOP_ISLOCKED(vp))
2974 				vprint("", vp);
2975 		}
2976 		nmp = TAILQ_NEXT(mp, mnt_list);
2977 	}
2978 }
2979 
2980 /*
2981  * Show details about the given vnode.
2982  */
2983 DB_SHOW_COMMAND(vnode, db_show_vnode)
2984 {
2985 	struct vnode *vp;
2986 
2987 	if (!have_addr)
2988 		return;
2989 	vp = (struct vnode *)addr;
2990 	vn_printf(vp, "vnode ");
2991 }
2992 
2993 /*
2994  * Show details about the given mount point.
2995  */
2996 DB_SHOW_COMMAND(mount, db_show_mount)
2997 {
2998 	struct mount *mp;
2999 	struct vfsopt *opt;
3000 	struct statfs *sp;
3001 	struct vnode *vp;
3002 	char buf[512];
3003 	uint64_t mflags;
3004 	u_int flags;
3005 
3006 	if (!have_addr) {
3007 		/* No address given, print short info about all mount points. */
3008 		TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3009 			db_printf("%p %s on %s (%s)\n", mp,
3010 			    mp->mnt_stat.f_mntfromname,
3011 			    mp->mnt_stat.f_mntonname,
3012 			    mp->mnt_stat.f_fstypename);
3013 			if (db_pager_quit)
3014 				break;
3015 		}
3016 		db_printf("\nMore info: show mount <addr>\n");
3017 		return;
3018 	}
3019 
3020 	mp = (struct mount *)addr;
3021 	db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
3022 	    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
3023 
3024 	buf[0] = '\0';
3025 	mflags = mp->mnt_flag;
3026 #define	MNT_FLAG(flag)	do {						\
3027 	if (mflags & (flag)) {						\
3028 		if (buf[0] != '\0')					\
3029 			strlcat(buf, ", ", sizeof(buf));		\
3030 		strlcat(buf, (#flag) + 4, sizeof(buf));			\
3031 		mflags &= ~(flag);					\
3032 	}								\
3033 } while (0)
3034 	MNT_FLAG(MNT_RDONLY);
3035 	MNT_FLAG(MNT_SYNCHRONOUS);
3036 	MNT_FLAG(MNT_NOEXEC);
3037 	MNT_FLAG(MNT_NOSUID);
3038 	MNT_FLAG(MNT_NFS4ACLS);
3039 	MNT_FLAG(MNT_UNION);
3040 	MNT_FLAG(MNT_ASYNC);
3041 	MNT_FLAG(MNT_SUIDDIR);
3042 	MNT_FLAG(MNT_SOFTDEP);
3043 	MNT_FLAG(MNT_NOSYMFOLLOW);
3044 	MNT_FLAG(MNT_GJOURNAL);
3045 	MNT_FLAG(MNT_MULTILABEL);
3046 	MNT_FLAG(MNT_ACLS);
3047 	MNT_FLAG(MNT_NOATIME);
3048 	MNT_FLAG(MNT_NOCLUSTERR);
3049 	MNT_FLAG(MNT_NOCLUSTERW);
3050 	MNT_FLAG(MNT_SUJ);
3051 	MNT_FLAG(MNT_EXRDONLY);
3052 	MNT_FLAG(MNT_EXPORTED);
3053 	MNT_FLAG(MNT_DEFEXPORTED);
3054 	MNT_FLAG(MNT_EXPORTANON);
3055 	MNT_FLAG(MNT_EXKERB);
3056 	MNT_FLAG(MNT_EXPUBLIC);
3057 	MNT_FLAG(MNT_LOCAL);
3058 	MNT_FLAG(MNT_QUOTA);
3059 	MNT_FLAG(MNT_ROOTFS);
3060 	MNT_FLAG(MNT_USER);
3061 	MNT_FLAG(MNT_IGNORE);
3062 	MNT_FLAG(MNT_UPDATE);
3063 	MNT_FLAG(MNT_DELEXPORT);
3064 	MNT_FLAG(MNT_RELOAD);
3065 	MNT_FLAG(MNT_FORCE);
3066 	MNT_FLAG(MNT_SNAPSHOT);
3067 	MNT_FLAG(MNT_BYFSID);
3068 #undef MNT_FLAG
3069 	if (mflags != 0) {
3070 		if (buf[0] != '\0')
3071 			strlcat(buf, ", ", sizeof(buf));
3072 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3073 		    "0x%016jx", mflags);
3074 	}
3075 	db_printf("    mnt_flag = %s\n", buf);
3076 
3077 	buf[0] = '\0';
3078 	flags = mp->mnt_kern_flag;
3079 #define	MNT_KERN_FLAG(flag)	do {					\
3080 	if (flags & (flag)) {						\
3081 		if (buf[0] != '\0')					\
3082 			strlcat(buf, ", ", sizeof(buf));		\
3083 		strlcat(buf, (#flag) + 5, sizeof(buf));			\
3084 		flags &= ~(flag);					\
3085 	}								\
3086 } while (0)
3087 	MNT_KERN_FLAG(MNTK_UNMOUNTF);
3088 	MNT_KERN_FLAG(MNTK_ASYNC);
3089 	MNT_KERN_FLAG(MNTK_SOFTDEP);
3090 	MNT_KERN_FLAG(MNTK_NOINSMNTQ);
3091 	MNT_KERN_FLAG(MNTK_DRAINING);
3092 	MNT_KERN_FLAG(MNTK_REFEXPIRE);
3093 	MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
3094 	MNT_KERN_FLAG(MNTK_SHARED_WRITES);
3095 	MNT_KERN_FLAG(MNTK_NO_IOPF);
3096 	MNT_KERN_FLAG(MNTK_VGONE_UPPER);
3097 	MNT_KERN_FLAG(MNTK_VGONE_WAITER);
3098 	MNT_KERN_FLAG(MNTK_LOOKUP_EXCL_DOTDOT);
3099 	MNT_KERN_FLAG(MNTK_MARKER);
3100 	MNT_KERN_FLAG(MNTK_NOASYNC);
3101 	MNT_KERN_FLAG(MNTK_UNMOUNT);
3102 	MNT_KERN_FLAG(MNTK_MWAIT);
3103 	MNT_KERN_FLAG(MNTK_SUSPEND);
3104 	MNT_KERN_FLAG(MNTK_SUSPEND2);
3105 	MNT_KERN_FLAG(MNTK_SUSPENDED);
3106 	MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
3107 	MNT_KERN_FLAG(MNTK_NOKNOTE);
3108 #undef MNT_KERN_FLAG
3109 	if (flags != 0) {
3110 		if (buf[0] != '\0')
3111 			strlcat(buf, ", ", sizeof(buf));
3112 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3113 		    "0x%08x", flags);
3114 	}
3115 	db_printf("    mnt_kern_flag = %s\n", buf);
3116 
3117 	db_printf("    mnt_opt = ");
3118 	opt = TAILQ_FIRST(mp->mnt_opt);
3119 	if (opt != NULL) {
3120 		db_printf("%s", opt->name);
3121 		opt = TAILQ_NEXT(opt, link);
3122 		while (opt != NULL) {
3123 			db_printf(", %s", opt->name);
3124 			opt = TAILQ_NEXT(opt, link);
3125 		}
3126 	}
3127 	db_printf("\n");
3128 
3129 	sp = &mp->mnt_stat;
3130 	db_printf("    mnt_stat = { version=%u type=%u flags=0x%016jx "
3131 	    "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
3132 	    "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
3133 	    "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
3134 	    (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
3135 	    (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
3136 	    (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
3137 	    (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
3138 	    (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
3139 	    (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
3140 	    (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
3141 	    (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
3142 
3143 	db_printf("    mnt_cred = { uid=%u ruid=%u",
3144 	    (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
3145 	if (jailed(mp->mnt_cred))
3146 		db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
3147 	db_printf(" }\n");
3148 	db_printf("    mnt_ref = %d\n", mp->mnt_ref);
3149 	db_printf("    mnt_gen = %d\n", mp->mnt_gen);
3150 	db_printf("    mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
3151 	db_printf("    mnt_activevnodelistsize = %d\n",
3152 	    mp->mnt_activevnodelistsize);
3153 	db_printf("    mnt_writeopcount = %d\n", mp->mnt_writeopcount);
3154 	db_printf("    mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
3155 	db_printf("    mnt_iosize_max = %d\n", mp->mnt_iosize_max);
3156 	db_printf("    mnt_hashseed = %u\n", mp->mnt_hashseed);
3157 	db_printf("    mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
3158 	db_printf("    mnt_secondary_accwrites = %d\n",
3159 	    mp->mnt_secondary_accwrites);
3160 	db_printf("    mnt_gjprovider = %s\n",
3161 	    mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
3162 
3163 	db_printf("\n\nList of active vnodes\n");
3164 	TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) {
3165 		if (vp->v_type != VMARKER) {
3166 			vn_printf(vp, "vnode ");
3167 			if (db_pager_quit)
3168 				break;
3169 		}
3170 	}
3171 	db_printf("\n\nList of inactive vnodes\n");
3172 	TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3173 		if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) {
3174 			vn_printf(vp, "vnode ");
3175 			if (db_pager_quit)
3176 				break;
3177 		}
3178 	}
3179 }
3180 #endif	/* DDB */
3181 
3182 /*
3183  * Fill in a struct xvfsconf based on a struct vfsconf.
3184  */
3185 static int
3186 vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp)
3187 {
3188 	struct xvfsconf xvfsp;
3189 
3190 	bzero(&xvfsp, sizeof(xvfsp));
3191 	strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3192 	xvfsp.vfc_typenum = vfsp->vfc_typenum;
3193 	xvfsp.vfc_refcount = vfsp->vfc_refcount;
3194 	xvfsp.vfc_flags = vfsp->vfc_flags;
3195 	/*
3196 	 * These are unused in userland, we keep them
3197 	 * to not break binary compatibility.
3198 	 */
3199 	xvfsp.vfc_vfsops = NULL;
3200 	xvfsp.vfc_next = NULL;
3201 	return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3202 }
3203 
3204 #ifdef COMPAT_FREEBSD32
3205 struct xvfsconf32 {
3206 	uint32_t	vfc_vfsops;
3207 	char		vfc_name[MFSNAMELEN];
3208 	int32_t		vfc_typenum;
3209 	int32_t		vfc_refcount;
3210 	int32_t		vfc_flags;
3211 	uint32_t	vfc_next;
3212 };
3213 
3214 static int
3215 vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp)
3216 {
3217 	struct xvfsconf32 xvfsp;
3218 
3219 	strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3220 	xvfsp.vfc_typenum = vfsp->vfc_typenum;
3221 	xvfsp.vfc_refcount = vfsp->vfc_refcount;
3222 	xvfsp.vfc_flags = vfsp->vfc_flags;
3223 	xvfsp.vfc_vfsops = 0;
3224 	xvfsp.vfc_next = 0;
3225 	return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3226 }
3227 #endif
3228 
3229 /*
3230  * Top level filesystem related information gathering.
3231  */
3232 static int
3233 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
3234 {
3235 	struct vfsconf *vfsp;
3236 	int error;
3237 
3238 	error = 0;
3239 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3240 #ifdef COMPAT_FREEBSD32
3241 		if (req->flags & SCTL_MASK32)
3242 			error = vfsconf2x32(req, vfsp);
3243 		else
3244 #endif
3245 			error = vfsconf2x(req, vfsp);
3246 		if (error)
3247 			break;
3248 	}
3249 	return (error);
3250 }
3251 
3252 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD,
3253     NULL, 0, sysctl_vfs_conflist,
3254     "S,xvfsconf", "List of all configured filesystems");
3255 
3256 #ifndef BURN_BRIDGES
3257 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
3258 
3259 static int
3260 vfs_sysctl(SYSCTL_HANDLER_ARGS)
3261 {
3262 	int *name = (int *)arg1 - 1;	/* XXX */
3263 	u_int namelen = arg2 + 1;	/* XXX */
3264 	struct vfsconf *vfsp;
3265 
3266 	log(LOG_WARNING, "userland calling deprecated sysctl, "
3267 	    "please rebuild world\n");
3268 
3269 #if 1 || defined(COMPAT_PRELITE2)
3270 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3271 	if (namelen == 1)
3272 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3273 #endif
3274 
3275 	switch (name[1]) {
3276 	case VFS_MAXTYPENUM:
3277 		if (namelen != 2)
3278 			return (ENOTDIR);
3279 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3280 	case VFS_CONF:
3281 		if (namelen != 3)
3282 			return (ENOTDIR);	/* overloaded */
3283 		TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
3284 			if (vfsp->vfc_typenum == name[2])
3285 				break;
3286 		if (vfsp == NULL)
3287 			return (EOPNOTSUPP);
3288 #ifdef COMPAT_FREEBSD32
3289 		if (req->flags & SCTL_MASK32)
3290 			return (vfsconf2x32(req, vfsp));
3291 		else
3292 #endif
3293 			return (vfsconf2x(req, vfsp));
3294 	}
3295 	return (EOPNOTSUPP);
3296 }
3297 
3298 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
3299     vfs_sysctl, "Generic filesystem");
3300 
3301 #if 1 || defined(COMPAT_PRELITE2)
3302 
3303 static int
3304 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3305 {
3306 	int error;
3307 	struct vfsconf *vfsp;
3308 	struct ovfsconf ovfs;
3309 
3310 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3311 		bzero(&ovfs, sizeof(ovfs));
3312 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
3313 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
3314 		ovfs.vfc_index = vfsp->vfc_typenum;
3315 		ovfs.vfc_refcount = vfsp->vfc_refcount;
3316 		ovfs.vfc_flags = vfsp->vfc_flags;
3317 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3318 		if (error)
3319 			return error;
3320 	}
3321 	return 0;
3322 }
3323 
3324 #endif /* 1 || COMPAT_PRELITE2 */
3325 #endif /* !BURN_BRIDGES */
3326 
3327 #define KINFO_VNODESLOP		10
3328 #ifdef notyet
3329 /*
3330  * Dump vnode list (via sysctl).
3331  */
3332 /* ARGSUSED */
3333 static int
3334 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3335 {
3336 	struct xvnode *xvn;
3337 	struct mount *mp;
3338 	struct vnode *vp;
3339 	int error, len, n;
3340 
3341 	/*
3342 	 * Stale numvnodes access is not fatal here.
3343 	 */
3344 	req->lock = 0;
3345 	len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3346 	if (!req->oldptr)
3347 		/* Make an estimate */
3348 		return (SYSCTL_OUT(req, 0, len));
3349 
3350 	error = sysctl_wire_old_buffer(req, 0);
3351 	if (error != 0)
3352 		return (error);
3353 	xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3354 	n = 0;
3355 	mtx_lock(&mountlist_mtx);
3356 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3357 		if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3358 			continue;
3359 		MNT_ILOCK(mp);
3360 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3361 			if (n == len)
3362 				break;
3363 			vref(vp);
3364 			xvn[n].xv_size = sizeof *xvn;
3365 			xvn[n].xv_vnode = vp;
3366 			xvn[n].xv_id = 0;	/* XXX compat */
3367 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3368 			XV_COPY(usecount);
3369 			XV_COPY(writecount);
3370 			XV_COPY(holdcnt);
3371 			XV_COPY(mount);
3372 			XV_COPY(numoutput);
3373 			XV_COPY(type);
3374 #undef XV_COPY
3375 			xvn[n].xv_flag = vp->v_vflag;
3376 
3377 			switch (vp->v_type) {
3378 			case VREG:
3379 			case VDIR:
3380 			case VLNK:
3381 				break;
3382 			case VBLK:
3383 			case VCHR:
3384 				if (vp->v_rdev == NULL) {
3385 					vrele(vp);
3386 					continue;
3387 				}
3388 				xvn[n].xv_dev = dev2udev(vp->v_rdev);
3389 				break;
3390 			case VSOCK:
3391 				xvn[n].xv_socket = vp->v_socket;
3392 				break;
3393 			case VFIFO:
3394 				xvn[n].xv_fifo = vp->v_fifoinfo;
3395 				break;
3396 			case VNON:
3397 			case VBAD:
3398 			default:
3399 				/* shouldn't happen? */
3400 				vrele(vp);
3401 				continue;
3402 			}
3403 			vrele(vp);
3404 			++n;
3405 		}
3406 		MNT_IUNLOCK(mp);
3407 		mtx_lock(&mountlist_mtx);
3408 		vfs_unbusy(mp);
3409 		if (n == len)
3410 			break;
3411 	}
3412 	mtx_unlock(&mountlist_mtx);
3413 
3414 	error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3415 	free(xvn, M_TEMP);
3416 	return (error);
3417 }
3418 
3419 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3420     0, 0, sysctl_vnode, "S,xvnode", "");
3421 #endif
3422 
3423 /*
3424  * Unmount all filesystems. The list is traversed in reverse order
3425  * of mounting to avoid dependencies.
3426  */
3427 void
3428 vfs_unmountall(void)
3429 {
3430 	struct mount *mp;
3431 	struct thread *td;
3432 	int error;
3433 
3434 	CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3435 	td = curthread;
3436 
3437 	/*
3438 	 * Since this only runs when rebooting, it is not interlocked.
3439 	 */
3440 	while(!TAILQ_EMPTY(&mountlist)) {
3441 		mp = TAILQ_LAST(&mountlist, mntlist);
3442 		error = dounmount(mp, MNT_FORCE, td);
3443 		if (error) {
3444 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
3445 			/*
3446 			 * XXX: Due to the way in which we mount the root
3447 			 * file system off of devfs, devfs will generate a
3448 			 * "busy" warning when we try to unmount it before
3449 			 * the root.  Don't print a warning as a result in
3450 			 * order to avoid false positive errors that may
3451 			 * cause needless upset.
3452 			 */
3453 			if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3454 				printf("unmount of %s failed (",
3455 				    mp->mnt_stat.f_mntonname);
3456 				if (error == EBUSY)
3457 					printf("BUSY)\n");
3458 				else
3459 					printf("%d)\n", error);
3460 			}
3461 		} else {
3462 			/* The unmount has removed mp from the mountlist */
3463 		}
3464 	}
3465 }
3466 
3467 /*
3468  * perform msync on all vnodes under a mount point
3469  * the mount point must be locked.
3470  */
3471 void
3472 vfs_msync(struct mount *mp, int flags)
3473 {
3474 	struct vnode *vp, *mvp;
3475 	struct vm_object *obj;
3476 
3477 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3478 	MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
3479 		obj = vp->v_object;
3480 		if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3481 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3482 			if (!vget(vp,
3483 			    LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3484 			    curthread)) {
3485 				if (vp->v_vflag & VV_NOSYNC) {	/* unlinked */
3486 					vput(vp);
3487 					continue;
3488 				}
3489 
3490 				obj = vp->v_object;
3491 				if (obj != NULL) {
3492 					VM_OBJECT_LOCK(obj);
3493 					vm_object_page_clean(obj, 0, 0,
3494 					    flags == MNT_WAIT ?
3495 					    OBJPC_SYNC : OBJPC_NOSYNC);
3496 					VM_OBJECT_UNLOCK(obj);
3497 				}
3498 				vput(vp);
3499 			}
3500 		} else
3501 			VI_UNLOCK(vp);
3502 	}
3503 }
3504 
3505 static void
3506 destroy_vpollinfo(struct vpollinfo *vi)
3507 {
3508 	seldrain(&vi->vpi_selinfo);
3509 	knlist_destroy(&vi->vpi_selinfo.si_note);
3510 	mtx_destroy(&vi->vpi_lock);
3511 	uma_zfree(vnodepoll_zone, vi);
3512 }
3513 
3514 /*
3515  * Initalize per-vnode helper structure to hold poll-related state.
3516  */
3517 void
3518 v_addpollinfo(struct vnode *vp)
3519 {
3520 	struct vpollinfo *vi;
3521 
3522 	if (vp->v_pollinfo != NULL)
3523 		return;
3524 	vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3525 	mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3526 	knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3527 	    vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3528 	VI_LOCK(vp);
3529 	if (vp->v_pollinfo != NULL) {
3530 		VI_UNLOCK(vp);
3531 		destroy_vpollinfo(vi);
3532 		return;
3533 	}
3534 	vp->v_pollinfo = vi;
3535 	VI_UNLOCK(vp);
3536 }
3537 
3538 /*
3539  * Record a process's interest in events which might happen to
3540  * a vnode.  Because poll uses the historic select-style interface
3541  * internally, this routine serves as both the ``check for any
3542  * pending events'' and the ``record my interest in future events''
3543  * functions.  (These are done together, while the lock is held,
3544  * to avoid race conditions.)
3545  */
3546 int
3547 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3548 {
3549 
3550 	v_addpollinfo(vp);
3551 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3552 	if (vp->v_pollinfo->vpi_revents & events) {
3553 		/*
3554 		 * This leaves events we are not interested
3555 		 * in available for the other process which
3556 		 * which presumably had requested them
3557 		 * (otherwise they would never have been
3558 		 * recorded).
3559 		 */
3560 		events &= vp->v_pollinfo->vpi_revents;
3561 		vp->v_pollinfo->vpi_revents &= ~events;
3562 
3563 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
3564 		return (events);
3565 	}
3566 	vp->v_pollinfo->vpi_events |= events;
3567 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3568 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3569 	return (0);
3570 }
3571 
3572 /*
3573  * Routine to create and manage a filesystem syncer vnode.
3574  */
3575 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
3576 static int	sync_fsync(struct  vop_fsync_args *);
3577 static int	sync_inactive(struct  vop_inactive_args *);
3578 static int	sync_reclaim(struct  vop_reclaim_args *);
3579 
3580 static struct vop_vector sync_vnodeops = {
3581 	.vop_bypass =	VOP_EOPNOTSUPP,
3582 	.vop_close =	sync_close,		/* close */
3583 	.vop_fsync =	sync_fsync,		/* fsync */
3584 	.vop_inactive =	sync_inactive,	/* inactive */
3585 	.vop_reclaim =	sync_reclaim,	/* reclaim */
3586 	.vop_lock1 =	vop_stdlock,	/* lock */
3587 	.vop_unlock =	vop_stdunlock,	/* unlock */
3588 	.vop_islocked =	vop_stdislocked,	/* islocked */
3589 };
3590 
3591 /*
3592  * Create a new filesystem syncer vnode for the specified mount point.
3593  */
3594 void
3595 vfs_allocate_syncvnode(struct mount *mp)
3596 {
3597 	struct vnode *vp;
3598 	struct bufobj *bo;
3599 	static long start, incr, next;
3600 	int error;
3601 
3602 	/* Allocate a new vnode */
3603 	error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
3604 	if (error != 0)
3605 		panic("vfs_allocate_syncvnode: getnewvnode() failed");
3606 	vp->v_type = VNON;
3607 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3608 	vp->v_vflag |= VV_FORCEINSMQ;
3609 	error = insmntque(vp, mp);
3610 	if (error != 0)
3611 		panic("vfs_allocate_syncvnode: insmntque() failed");
3612 	vp->v_vflag &= ~VV_FORCEINSMQ;
3613 	VOP_UNLOCK(vp, 0);
3614 	/*
3615 	 * Place the vnode onto the syncer worklist. We attempt to
3616 	 * scatter them about on the list so that they will go off
3617 	 * at evenly distributed times even if all the filesystems
3618 	 * are mounted at once.
3619 	 */
3620 	next += incr;
3621 	if (next == 0 || next > syncer_maxdelay) {
3622 		start /= 2;
3623 		incr /= 2;
3624 		if (start == 0) {
3625 			start = syncer_maxdelay / 2;
3626 			incr = syncer_maxdelay;
3627 		}
3628 		next = start;
3629 	}
3630 	bo = &vp->v_bufobj;
3631 	BO_LOCK(bo);
3632 	vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3633 	/* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3634 	mtx_lock(&sync_mtx);
3635 	sync_vnode_count++;
3636 	if (mp->mnt_syncer == NULL) {
3637 		mp->mnt_syncer = vp;
3638 		vp = NULL;
3639 	}
3640 	mtx_unlock(&sync_mtx);
3641 	BO_UNLOCK(bo);
3642 	if (vp != NULL) {
3643 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3644 		vgone(vp);
3645 		vput(vp);
3646 	}
3647 }
3648 
3649 void
3650 vfs_deallocate_syncvnode(struct mount *mp)
3651 {
3652 	struct vnode *vp;
3653 
3654 	mtx_lock(&sync_mtx);
3655 	vp = mp->mnt_syncer;
3656 	if (vp != NULL)
3657 		mp->mnt_syncer = NULL;
3658 	mtx_unlock(&sync_mtx);
3659 	if (vp != NULL)
3660 		vrele(vp);
3661 }
3662 
3663 /*
3664  * Do a lazy sync of the filesystem.
3665  */
3666 static int
3667 sync_fsync(struct vop_fsync_args *ap)
3668 {
3669 	struct vnode *syncvp = ap->a_vp;
3670 	struct mount *mp = syncvp->v_mount;
3671 	int error, save;
3672 	struct bufobj *bo;
3673 
3674 	/*
3675 	 * We only need to do something if this is a lazy evaluation.
3676 	 */
3677 	if (ap->a_waitfor != MNT_LAZY)
3678 		return (0);
3679 
3680 	/*
3681 	 * Move ourselves to the back of the sync list.
3682 	 */
3683 	bo = &syncvp->v_bufobj;
3684 	BO_LOCK(bo);
3685 	vn_syncer_add_to_worklist(bo, syncdelay);
3686 	BO_UNLOCK(bo);
3687 
3688 	/*
3689 	 * Walk the list of vnodes pushing all that are dirty and
3690 	 * not already on the sync list.
3691 	 */
3692 	mtx_lock(&mountlist_mtx);
3693 	if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) {
3694 		mtx_unlock(&mountlist_mtx);
3695 		return (0);
3696 	}
3697 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3698 		vfs_unbusy(mp);
3699 		return (0);
3700 	}
3701 	save = curthread_pflags_set(TDP_SYNCIO);
3702 	vfs_msync(mp, MNT_NOWAIT);
3703 	error = VFS_SYNC(mp, MNT_LAZY);
3704 	curthread_pflags_restore(save);
3705 	vn_finished_write(mp);
3706 	vfs_unbusy(mp);
3707 	return (error);
3708 }
3709 
3710 /*
3711  * The syncer vnode is no referenced.
3712  */
3713 static int
3714 sync_inactive(struct vop_inactive_args *ap)
3715 {
3716 
3717 	vgone(ap->a_vp);
3718 	return (0);
3719 }
3720 
3721 /*
3722  * The syncer vnode is no longer needed and is being decommissioned.
3723  *
3724  * Modifications to the worklist must be protected by sync_mtx.
3725  */
3726 static int
3727 sync_reclaim(struct vop_reclaim_args *ap)
3728 {
3729 	struct vnode *vp = ap->a_vp;
3730 	struct bufobj *bo;
3731 
3732 	bo = &vp->v_bufobj;
3733 	BO_LOCK(bo);
3734 	mtx_lock(&sync_mtx);
3735 	if (vp->v_mount->mnt_syncer == vp)
3736 		vp->v_mount->mnt_syncer = NULL;
3737 	if (bo->bo_flag & BO_ONWORKLST) {
3738 		LIST_REMOVE(bo, bo_synclist);
3739 		syncer_worklist_len--;
3740 		sync_vnode_count--;
3741 		bo->bo_flag &= ~BO_ONWORKLST;
3742 	}
3743 	mtx_unlock(&sync_mtx);
3744 	BO_UNLOCK(bo);
3745 
3746 	return (0);
3747 }
3748 
3749 /*
3750  * Check if vnode represents a disk device
3751  */
3752 int
3753 vn_isdisk(struct vnode *vp, int *errp)
3754 {
3755 	int error;
3756 
3757 	error = 0;
3758 	dev_lock();
3759 	if (vp->v_type != VCHR)
3760 		error = ENOTBLK;
3761 	else if (vp->v_rdev == NULL)
3762 		error = ENXIO;
3763 	else if (vp->v_rdev->si_devsw == NULL)
3764 		error = ENXIO;
3765 	else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3766 		error = ENOTBLK;
3767 	dev_unlock();
3768 	if (errp != NULL)
3769 		*errp = error;
3770 	return (error == 0);
3771 }
3772 
3773 /*
3774  * Common filesystem object access control check routine.  Accepts a
3775  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3776  * and optional call-by-reference privused argument allowing vaccess()
3777  * to indicate to the caller whether privilege was used to satisfy the
3778  * request (obsoleted).  Returns 0 on success, or an errno on failure.
3779  */
3780 int
3781 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3782     accmode_t accmode, struct ucred *cred, int *privused)
3783 {
3784 	accmode_t dac_granted;
3785 	accmode_t priv_granted;
3786 
3787 	KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3788 	    ("invalid bit in accmode"));
3789 	KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
3790 	    ("VAPPEND without VWRITE"));
3791 
3792 	/*
3793 	 * Look for a normal, non-privileged way to access the file/directory
3794 	 * as requested.  If it exists, go with that.
3795 	 */
3796 
3797 	if (privused != NULL)
3798 		*privused = 0;
3799 
3800 	dac_granted = 0;
3801 
3802 	/* Check the owner. */
3803 	if (cred->cr_uid == file_uid) {
3804 		dac_granted |= VADMIN;
3805 		if (file_mode & S_IXUSR)
3806 			dac_granted |= VEXEC;
3807 		if (file_mode & S_IRUSR)
3808 			dac_granted |= VREAD;
3809 		if (file_mode & S_IWUSR)
3810 			dac_granted |= (VWRITE | VAPPEND);
3811 
3812 		if ((accmode & dac_granted) == accmode)
3813 			return (0);
3814 
3815 		goto privcheck;
3816 	}
3817 
3818 	/* Otherwise, check the groups (first match) */
3819 	if (groupmember(file_gid, cred)) {
3820 		if (file_mode & S_IXGRP)
3821 			dac_granted |= VEXEC;
3822 		if (file_mode & S_IRGRP)
3823 			dac_granted |= VREAD;
3824 		if (file_mode & S_IWGRP)
3825 			dac_granted |= (VWRITE | VAPPEND);
3826 
3827 		if ((accmode & dac_granted) == accmode)
3828 			return (0);
3829 
3830 		goto privcheck;
3831 	}
3832 
3833 	/* Otherwise, check everyone else. */
3834 	if (file_mode & S_IXOTH)
3835 		dac_granted |= VEXEC;
3836 	if (file_mode & S_IROTH)
3837 		dac_granted |= VREAD;
3838 	if (file_mode & S_IWOTH)
3839 		dac_granted |= (VWRITE | VAPPEND);
3840 	if ((accmode & dac_granted) == accmode)
3841 		return (0);
3842 
3843 privcheck:
3844 	/*
3845 	 * Build a privilege mask to determine if the set of privileges
3846 	 * satisfies the requirements when combined with the granted mask
3847 	 * from above.  For each privilege, if the privilege is required,
3848 	 * bitwise or the request type onto the priv_granted mask.
3849 	 */
3850 	priv_granted = 0;
3851 
3852 	if (type == VDIR) {
3853 		/*
3854 		 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3855 		 * requests, instead of PRIV_VFS_EXEC.
3856 		 */
3857 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3858 		    !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3859 			priv_granted |= VEXEC;
3860 	} else {
3861 		/*
3862 		 * Ensure that at least one execute bit is on. Otherwise,
3863 		 * a privileged user will always succeed, and we don't want
3864 		 * this to happen unless the file really is executable.
3865 		 */
3866 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3867 		    (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
3868 		    !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3869 			priv_granted |= VEXEC;
3870 	}
3871 
3872 	if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
3873 	    !priv_check_cred(cred, PRIV_VFS_READ, 0))
3874 		priv_granted |= VREAD;
3875 
3876 	if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3877 	    !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3878 		priv_granted |= (VWRITE | VAPPEND);
3879 
3880 	if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3881 	    !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3882 		priv_granted |= VADMIN;
3883 
3884 	if ((accmode & (priv_granted | dac_granted)) == accmode) {
3885 		/* XXX audit: privilege used */
3886 		if (privused != NULL)
3887 			*privused = 1;
3888 		return (0);
3889 	}
3890 
3891 	return ((accmode & VADMIN) ? EPERM : EACCES);
3892 }
3893 
3894 /*
3895  * Credential check based on process requesting service, and per-attribute
3896  * permissions.
3897  */
3898 int
3899 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3900     struct thread *td, accmode_t accmode)
3901 {
3902 
3903 	/*
3904 	 * Kernel-invoked always succeeds.
3905 	 */
3906 	if (cred == NOCRED)
3907 		return (0);
3908 
3909 	/*
3910 	 * Do not allow privileged processes in jail to directly manipulate
3911 	 * system attributes.
3912 	 */
3913 	switch (attrnamespace) {
3914 	case EXTATTR_NAMESPACE_SYSTEM:
3915 		/* Potentially should be: return (EPERM); */
3916 		return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3917 	case EXTATTR_NAMESPACE_USER:
3918 		return (VOP_ACCESS(vp, accmode, cred, td));
3919 	default:
3920 		return (EPERM);
3921 	}
3922 }
3923 
3924 #ifdef DEBUG_VFS_LOCKS
3925 /*
3926  * This only exists to supress warnings from unlocked specfs accesses.  It is
3927  * no longer ok to have an unlocked VFS.
3928  */
3929 #define	IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL ||		\
3930 	(vp)->v_type == VCHR ||	(vp)->v_type == VBAD)
3931 
3932 int vfs_badlock_ddb = 1;	/* Drop into debugger on violation. */
3933 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
3934     "Drop into debugger on lock violation");
3935 
3936 int vfs_badlock_mutex = 1;	/* Check for interlock across VOPs. */
3937 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
3938     0, "Check for interlock across VOPs");
3939 
3940 int vfs_badlock_print = 1;	/* Print lock violations. */
3941 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
3942     0, "Print lock violations");
3943 
3944 #ifdef KDB
3945 int vfs_badlock_backtrace = 1;	/* Print backtrace at lock violations. */
3946 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
3947     &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
3948 #endif
3949 
3950 static void
3951 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3952 {
3953 
3954 #ifdef KDB
3955 	if (vfs_badlock_backtrace)
3956 		kdb_backtrace();
3957 #endif
3958 	if (vfs_badlock_print)
3959 		printf("%s: %p %s\n", str, (void *)vp, msg);
3960 	if (vfs_badlock_ddb)
3961 		kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3962 }
3963 
3964 void
3965 assert_vi_locked(struct vnode *vp, const char *str)
3966 {
3967 
3968 	if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3969 		vfs_badlock("interlock is not locked but should be", str, vp);
3970 }
3971 
3972 void
3973 assert_vi_unlocked(struct vnode *vp, const char *str)
3974 {
3975 
3976 	if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3977 		vfs_badlock("interlock is locked but should not be", str, vp);
3978 }
3979 
3980 void
3981 assert_vop_locked(struct vnode *vp, const char *str)
3982 {
3983 	int locked;
3984 
3985 	if (!IGNORE_LOCK(vp)) {
3986 		locked = VOP_ISLOCKED(vp);
3987 		if (locked == 0 || locked == LK_EXCLOTHER)
3988 			vfs_badlock("is not locked but should be", str, vp);
3989 	}
3990 }
3991 
3992 void
3993 assert_vop_unlocked(struct vnode *vp, const char *str)
3994 {
3995 
3996 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
3997 		vfs_badlock("is locked but should not be", str, vp);
3998 }
3999 
4000 void
4001 assert_vop_elocked(struct vnode *vp, const char *str)
4002 {
4003 
4004 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
4005 		vfs_badlock("is not exclusive locked but should be", str, vp);
4006 }
4007 
4008 #if 0
4009 void
4010 assert_vop_elocked_other(struct vnode *vp, const char *str)
4011 {
4012 
4013 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
4014 		vfs_badlock("is not exclusive locked by another thread",
4015 		    str, vp);
4016 }
4017 
4018 void
4019 assert_vop_slocked(struct vnode *vp, const char *str)
4020 {
4021 
4022 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
4023 		vfs_badlock("is not locked shared but should be", str, vp);
4024 }
4025 #endif /* 0 */
4026 #endif /* DEBUG_VFS_LOCKS */
4027 
4028 void
4029 vop_rename_fail(struct vop_rename_args *ap)
4030 {
4031 
4032 	if (ap->a_tvp != NULL)
4033 		vput(ap->a_tvp);
4034 	if (ap->a_tdvp == ap->a_tvp)
4035 		vrele(ap->a_tdvp);
4036 	else
4037 		vput(ap->a_tdvp);
4038 	vrele(ap->a_fdvp);
4039 	vrele(ap->a_fvp);
4040 }
4041 
4042 void
4043 vop_rename_pre(void *ap)
4044 {
4045 	struct vop_rename_args *a = ap;
4046 
4047 #ifdef DEBUG_VFS_LOCKS
4048 	if (a->a_tvp)
4049 		ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
4050 	ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
4051 	ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
4052 	ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
4053 
4054 	/* Check the source (from). */
4055 	if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
4056 	    (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
4057 		ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
4058 	if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
4059 		ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
4060 
4061 	/* Check the target. */
4062 	if (a->a_tvp)
4063 		ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
4064 	ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
4065 #endif
4066 	if (a->a_tdvp != a->a_fdvp)
4067 		vhold(a->a_fdvp);
4068 	if (a->a_tvp != a->a_fvp)
4069 		vhold(a->a_fvp);
4070 	vhold(a->a_tdvp);
4071 	if (a->a_tvp)
4072 		vhold(a->a_tvp);
4073 }
4074 
4075 void
4076 vop_strategy_pre(void *ap)
4077 {
4078 #ifdef DEBUG_VFS_LOCKS
4079 	struct vop_strategy_args *a;
4080 	struct buf *bp;
4081 
4082 	a = ap;
4083 	bp = a->a_bp;
4084 
4085 	/*
4086 	 * Cluster ops lock their component buffers but not the IO container.
4087 	 */
4088 	if ((bp->b_flags & B_CLUSTER) != 0)
4089 		return;
4090 
4091 	if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
4092 		if (vfs_badlock_print)
4093 			printf(
4094 			    "VOP_STRATEGY: bp is not locked but should be\n");
4095 		if (vfs_badlock_ddb)
4096 			kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
4097 	}
4098 #endif
4099 }
4100 
4101 void
4102 vop_lock_pre(void *ap)
4103 {
4104 #ifdef DEBUG_VFS_LOCKS
4105 	struct vop_lock1_args *a = ap;
4106 
4107 	if ((a->a_flags & LK_INTERLOCK) == 0)
4108 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4109 	else
4110 		ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
4111 #endif
4112 }
4113 
4114 void
4115 vop_lock_post(void *ap, int rc)
4116 {
4117 #ifdef DEBUG_VFS_LOCKS
4118 	struct vop_lock1_args *a = ap;
4119 
4120 	ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4121 	if (rc == 0)
4122 		ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
4123 #endif
4124 }
4125 
4126 void
4127 vop_unlock_pre(void *ap)
4128 {
4129 #ifdef DEBUG_VFS_LOCKS
4130 	struct vop_unlock_args *a = ap;
4131 
4132 	if (a->a_flags & LK_INTERLOCK)
4133 		ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
4134 	ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
4135 #endif
4136 }
4137 
4138 void
4139 vop_unlock_post(void *ap, int rc)
4140 {
4141 #ifdef DEBUG_VFS_LOCKS
4142 	struct vop_unlock_args *a = ap;
4143 
4144 	if (a->a_flags & LK_INTERLOCK)
4145 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
4146 #endif
4147 }
4148 
4149 void
4150 vop_create_post(void *ap, int rc)
4151 {
4152 	struct vop_create_args *a = ap;
4153 
4154 	if (!rc)
4155 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4156 }
4157 
4158 void
4159 vop_deleteextattr_post(void *ap, int rc)
4160 {
4161 	struct vop_deleteextattr_args *a = ap;
4162 
4163 	if (!rc)
4164 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4165 }
4166 
4167 void
4168 vop_link_post(void *ap, int rc)
4169 {
4170 	struct vop_link_args *a = ap;
4171 
4172 	if (!rc) {
4173 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
4174 		VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
4175 	}
4176 }
4177 
4178 void
4179 vop_mkdir_post(void *ap, int rc)
4180 {
4181 	struct vop_mkdir_args *a = ap;
4182 
4183 	if (!rc)
4184 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4185 }
4186 
4187 void
4188 vop_mknod_post(void *ap, int rc)
4189 {
4190 	struct vop_mknod_args *a = ap;
4191 
4192 	if (!rc)
4193 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4194 }
4195 
4196 void
4197 vop_remove_post(void *ap, int rc)
4198 {
4199 	struct vop_remove_args *a = ap;
4200 
4201 	if (!rc) {
4202 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4203 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4204 	}
4205 }
4206 
4207 void
4208 vop_rename_post(void *ap, int rc)
4209 {
4210 	struct vop_rename_args *a = ap;
4211 
4212 	if (!rc) {
4213 		VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
4214 		VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
4215 		VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
4216 		if (a->a_tvp)
4217 			VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
4218 	}
4219 	if (a->a_tdvp != a->a_fdvp)
4220 		vdrop(a->a_fdvp);
4221 	if (a->a_tvp != a->a_fvp)
4222 		vdrop(a->a_fvp);
4223 	vdrop(a->a_tdvp);
4224 	if (a->a_tvp)
4225 		vdrop(a->a_tvp);
4226 }
4227 
4228 void
4229 vop_rmdir_post(void *ap, int rc)
4230 {
4231 	struct vop_rmdir_args *a = ap;
4232 
4233 	if (!rc) {
4234 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4235 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4236 	}
4237 }
4238 
4239 void
4240 vop_setattr_post(void *ap, int rc)
4241 {
4242 	struct vop_setattr_args *a = ap;
4243 
4244 	if (!rc)
4245 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4246 }
4247 
4248 void
4249 vop_setextattr_post(void *ap, int rc)
4250 {
4251 	struct vop_setextattr_args *a = ap;
4252 
4253 	if (!rc)
4254 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4255 }
4256 
4257 void
4258 vop_symlink_post(void *ap, int rc)
4259 {
4260 	struct vop_symlink_args *a = ap;
4261 
4262 	if (!rc)
4263 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4264 }
4265 
4266 static struct knlist fs_knlist;
4267 
4268 static void
4269 vfs_event_init(void *arg)
4270 {
4271 	knlist_init_mtx(&fs_knlist, NULL);
4272 }
4273 /* XXX - correct order? */
4274 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4275 
4276 void
4277 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
4278 {
4279 
4280 	KNOTE_UNLOCKED(&fs_knlist, event);
4281 }
4282 
4283 static int	filt_fsattach(struct knote *kn);
4284 static void	filt_fsdetach(struct knote *kn);
4285 static int	filt_fsevent(struct knote *kn, long hint);
4286 
4287 struct filterops fs_filtops = {
4288 	.f_isfd = 0,
4289 	.f_attach = filt_fsattach,
4290 	.f_detach = filt_fsdetach,
4291 	.f_event = filt_fsevent
4292 };
4293 
4294 static int
4295 filt_fsattach(struct knote *kn)
4296 {
4297 
4298 	kn->kn_flags |= EV_CLEAR;
4299 	knlist_add(&fs_knlist, kn, 0);
4300 	return (0);
4301 }
4302 
4303 static void
4304 filt_fsdetach(struct knote *kn)
4305 {
4306 
4307 	knlist_remove(&fs_knlist, kn, 0);
4308 }
4309 
4310 static int
4311 filt_fsevent(struct knote *kn, long hint)
4312 {
4313 
4314 	kn->kn_fflags |= hint;
4315 	return (kn->kn_fflags != 0);
4316 }
4317 
4318 static int
4319 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4320 {
4321 	struct vfsidctl vc;
4322 	int error;
4323 	struct mount *mp;
4324 
4325 	error = SYSCTL_IN(req, &vc, sizeof(vc));
4326 	if (error)
4327 		return (error);
4328 	if (vc.vc_vers != VFS_CTL_VERS1)
4329 		return (EINVAL);
4330 	mp = vfs_getvfs(&vc.vc_fsid);
4331 	if (mp == NULL)
4332 		return (ENOENT);
4333 	/* ensure that a specific sysctl goes to the right filesystem. */
4334 	if (strcmp(vc.vc_fstypename, "*") != 0 &&
4335 	    strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4336 		vfs_rel(mp);
4337 		return (EINVAL);
4338 	}
4339 	VCTLTOREQ(&vc, req);
4340 	error = VFS_SYSCTL(mp, vc.vc_op, req);
4341 	vfs_rel(mp);
4342 	return (error);
4343 }
4344 
4345 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
4346     NULL, 0, sysctl_vfs_ctl, "",
4347     "Sysctl by fsid");
4348 
4349 /*
4350  * Function to initialize a va_filerev field sensibly.
4351  * XXX: Wouldn't a random number make a lot more sense ??
4352  */
4353 u_quad_t
4354 init_va_filerev(void)
4355 {
4356 	struct bintime bt;
4357 
4358 	getbinuptime(&bt);
4359 	return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4360 }
4361 
4362 static int	filt_vfsread(struct knote *kn, long hint);
4363 static int	filt_vfswrite(struct knote *kn, long hint);
4364 static int	filt_vfsvnode(struct knote *kn, long hint);
4365 static void	filt_vfsdetach(struct knote *kn);
4366 static struct filterops vfsread_filtops = {
4367 	.f_isfd = 1,
4368 	.f_detach = filt_vfsdetach,
4369 	.f_event = filt_vfsread
4370 };
4371 static struct filterops vfswrite_filtops = {
4372 	.f_isfd = 1,
4373 	.f_detach = filt_vfsdetach,
4374 	.f_event = filt_vfswrite
4375 };
4376 static struct filterops vfsvnode_filtops = {
4377 	.f_isfd = 1,
4378 	.f_detach = filt_vfsdetach,
4379 	.f_event = filt_vfsvnode
4380 };
4381 
4382 static void
4383 vfs_knllock(void *arg)
4384 {
4385 	struct vnode *vp = arg;
4386 
4387 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4388 }
4389 
4390 static void
4391 vfs_knlunlock(void *arg)
4392 {
4393 	struct vnode *vp = arg;
4394 
4395 	VOP_UNLOCK(vp, 0);
4396 }
4397 
4398 static void
4399 vfs_knl_assert_locked(void *arg)
4400 {
4401 #ifdef DEBUG_VFS_LOCKS
4402 	struct vnode *vp = arg;
4403 
4404 	ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4405 #endif
4406 }
4407 
4408 static void
4409 vfs_knl_assert_unlocked(void *arg)
4410 {
4411 #ifdef DEBUG_VFS_LOCKS
4412 	struct vnode *vp = arg;
4413 
4414 	ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4415 #endif
4416 }
4417 
4418 int
4419 vfs_kqfilter(struct vop_kqfilter_args *ap)
4420 {
4421 	struct vnode *vp = ap->a_vp;
4422 	struct knote *kn = ap->a_kn;
4423 	struct knlist *knl;
4424 
4425 	switch (kn->kn_filter) {
4426 	case EVFILT_READ:
4427 		kn->kn_fop = &vfsread_filtops;
4428 		break;
4429 	case EVFILT_WRITE:
4430 		kn->kn_fop = &vfswrite_filtops;
4431 		break;
4432 	case EVFILT_VNODE:
4433 		kn->kn_fop = &vfsvnode_filtops;
4434 		break;
4435 	default:
4436 		return (EINVAL);
4437 	}
4438 
4439 	kn->kn_hook = (caddr_t)vp;
4440 
4441 	v_addpollinfo(vp);
4442 	if (vp->v_pollinfo == NULL)
4443 		return (ENOMEM);
4444 	knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4445 	knlist_add(knl, kn, 0);
4446 
4447 	return (0);
4448 }
4449 
4450 /*
4451  * Detach knote from vnode
4452  */
4453 static void
4454 filt_vfsdetach(struct knote *kn)
4455 {
4456 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4457 
4458 	KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4459 	knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4460 }
4461 
4462 /*ARGSUSED*/
4463 static int
4464 filt_vfsread(struct knote *kn, long hint)
4465 {
4466 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4467 	struct vattr va;
4468 	int res;
4469 
4470 	/*
4471 	 * filesystem is gone, so set the EOF flag and schedule
4472 	 * the knote for deletion.
4473 	 */
4474 	if (hint == NOTE_REVOKE) {
4475 		VI_LOCK(vp);
4476 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4477 		VI_UNLOCK(vp);
4478 		return (1);
4479 	}
4480 
4481 	if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4482 		return (0);
4483 
4484 	VI_LOCK(vp);
4485 	kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4486 	res = (kn->kn_data != 0);
4487 	VI_UNLOCK(vp);
4488 	return (res);
4489 }
4490 
4491 /*ARGSUSED*/
4492 static int
4493 filt_vfswrite(struct knote *kn, long hint)
4494 {
4495 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4496 
4497 	VI_LOCK(vp);
4498 
4499 	/*
4500 	 * filesystem is gone, so set the EOF flag and schedule
4501 	 * the knote for deletion.
4502 	 */
4503 	if (hint == NOTE_REVOKE)
4504 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4505 
4506 	kn->kn_data = 0;
4507 	VI_UNLOCK(vp);
4508 	return (1);
4509 }
4510 
4511 static int
4512 filt_vfsvnode(struct knote *kn, long hint)
4513 {
4514 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4515 	int res;
4516 
4517 	VI_LOCK(vp);
4518 	if (kn->kn_sfflags & hint)
4519 		kn->kn_fflags |= hint;
4520 	if (hint == NOTE_REVOKE) {
4521 		kn->kn_flags |= EV_EOF;
4522 		VI_UNLOCK(vp);
4523 		return (1);
4524 	}
4525 	res = (kn->kn_fflags != 0);
4526 	VI_UNLOCK(vp);
4527 	return (res);
4528 }
4529 
4530 int
4531 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4532 {
4533 	int error;
4534 
4535 	if (dp->d_reclen > ap->a_uio->uio_resid)
4536 		return (ENAMETOOLONG);
4537 	error = uiomove(dp, dp->d_reclen, ap->a_uio);
4538 	if (error) {
4539 		if (ap->a_ncookies != NULL) {
4540 			if (ap->a_cookies != NULL)
4541 				free(ap->a_cookies, M_TEMP);
4542 			ap->a_cookies = NULL;
4543 			*ap->a_ncookies = 0;
4544 		}
4545 		return (error);
4546 	}
4547 	if (ap->a_ncookies == NULL)
4548 		return (0);
4549 
4550 	KASSERT(ap->a_cookies,
4551 	    ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4552 
4553 	*ap->a_cookies = realloc(*ap->a_cookies,
4554 	    (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4555 	(*ap->a_cookies)[*ap->a_ncookies] = off;
4556 	return (0);
4557 }
4558 
4559 /*
4560  * Mark for update the access time of the file if the filesystem
4561  * supports VOP_MARKATIME.  This functionality is used by execve and
4562  * mmap, so we want to avoid the I/O implied by directly setting
4563  * va_atime for the sake of efficiency.
4564  */
4565 void
4566 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4567 {
4568 	struct mount *mp;
4569 
4570 	mp = vp->v_mount;
4571 	ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4572 	if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4573 		(void)VOP_MARKATIME(vp);
4574 }
4575 
4576 /*
4577  * The purpose of this routine is to remove granularity from accmode_t,
4578  * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4579  * VADMIN and VAPPEND.
4580  *
4581  * If it returns 0, the caller is supposed to continue with the usual
4582  * access checks using 'accmode' as modified by this routine.  If it
4583  * returns nonzero value, the caller is supposed to return that value
4584  * as errno.
4585  *
4586  * Note that after this routine runs, accmode may be zero.
4587  */
4588 int
4589 vfs_unixify_accmode(accmode_t *accmode)
4590 {
4591 	/*
4592 	 * There is no way to specify explicit "deny" rule using
4593 	 * file mode or POSIX.1e ACLs.
4594 	 */
4595 	if (*accmode & VEXPLICIT_DENY) {
4596 		*accmode = 0;
4597 		return (0);
4598 	}
4599 
4600 	/*
4601 	 * None of these can be translated into usual access bits.
4602 	 * Also, the common case for NFSv4 ACLs is to not contain
4603 	 * either of these bits. Caller should check for VWRITE
4604 	 * on the containing directory instead.
4605 	 */
4606 	if (*accmode & (VDELETE_CHILD | VDELETE))
4607 		return (EPERM);
4608 
4609 	if (*accmode & VADMIN_PERMS) {
4610 		*accmode &= ~VADMIN_PERMS;
4611 		*accmode |= VADMIN;
4612 	}
4613 
4614 	/*
4615 	 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4616 	 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4617 	 */
4618 	*accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
4619 
4620 	return (0);
4621 }
4622 
4623 /*
4624  * These are helper functions for filesystems to traverse all
4625  * their vnodes.  See MNT_VNODE_FOREACH_ALL() in sys/mount.h.
4626  *
4627  * This interface replaces MNT_VNODE_FOREACH.
4628  */
4629 
4630 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
4631 
4632 struct vnode *
4633 __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp)
4634 {
4635 	struct vnode *vp;
4636 
4637 	if (should_yield())
4638 		kern_yield(PRI_USER);
4639 	MNT_ILOCK(mp);
4640 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4641 	vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
4642 	while (vp != NULL && (vp->v_type == VMARKER ||
4643 	    (vp->v_iflag & VI_DOOMED) != 0))
4644 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
4645 
4646 	/* Check if we are done */
4647 	if (vp == NULL) {
4648 		__mnt_vnode_markerfree_all(mvp, mp);
4649 		/* MNT_IUNLOCK(mp); -- done in above function */
4650 		mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
4651 		return (NULL);
4652 	}
4653 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4654 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4655 	VI_LOCK(vp);
4656 	MNT_IUNLOCK(mp);
4657 	return (vp);
4658 }
4659 
4660 struct vnode *
4661 __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp)
4662 {
4663 	struct vnode *vp;
4664 
4665 	*mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4666 	MNT_ILOCK(mp);
4667 	MNT_REF(mp);
4668 	(*mvp)->v_type = VMARKER;
4669 
4670 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
4671 	while (vp != NULL && (vp->v_type == VMARKER ||
4672 	    (vp->v_iflag & VI_DOOMED) != 0))
4673 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
4674 
4675 	/* Check if we are done */
4676 	if (vp == NULL) {
4677 		MNT_REL(mp);
4678 		MNT_IUNLOCK(mp);
4679 		free(*mvp, M_VNODE_MARKER);
4680 		*mvp = NULL;
4681 		return (NULL);
4682 	}
4683 	(*mvp)->v_mount = mp;
4684 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4685 	VI_LOCK(vp);
4686 	MNT_IUNLOCK(mp);
4687 	return (vp);
4688 }
4689 
4690 
4691 void
4692 __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp)
4693 {
4694 
4695 	if (*mvp == NULL) {
4696 		MNT_IUNLOCK(mp);
4697 		return;
4698 	}
4699 
4700 	mtx_assert(MNT_MTX(mp), MA_OWNED);
4701 
4702 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4703 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4704 	MNT_REL(mp);
4705 	MNT_IUNLOCK(mp);
4706 	free(*mvp, M_VNODE_MARKER);
4707 	*mvp = NULL;
4708 }
4709 
4710 /*
4711  * These are helper functions for filesystems to traverse their
4712  * active vnodes.  See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h
4713  */
4714 static void
4715 mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4716 {
4717 
4718 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4719 
4720 	MNT_ILOCK(mp);
4721 	MNT_REL(mp);
4722 	MNT_IUNLOCK(mp);
4723 	free(*mvp, M_VNODE_MARKER);
4724 	*mvp = NULL;
4725 }
4726 
4727 #ifdef SMP
4728 #define	ALWAYS_YIELD	(mp_ncpus == 1)
4729 #else
4730 #define	ALWAYS_YIELD	1
4731 #endif
4732 
4733 static struct vnode *
4734 mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4735 {
4736 	struct vnode *vp, *nvp;
4737 
4738 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
4739 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4740 restart:
4741 	vp = TAILQ_NEXT(*mvp, v_actfreelist);
4742 	TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4743 	while (vp != NULL) {
4744 		if (vp->v_type == VMARKER) {
4745 			vp = TAILQ_NEXT(vp, v_actfreelist);
4746 			continue;
4747 		}
4748 		if (!VI_TRYLOCK(vp)) {
4749 			if (ALWAYS_YIELD || should_yield()) {
4750 				TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
4751 				mtx_unlock(&vnode_free_list_mtx);
4752 				kern_yield(PRI_USER);
4753 				mtx_lock(&vnode_free_list_mtx);
4754 				goto restart;
4755 			}
4756 			continue;
4757 		}
4758 		KASSERT(vp->v_type != VMARKER, ("locked marker %p", vp));
4759 		KASSERT(vp->v_mount == mp || vp->v_mount == NULL,
4760 		    ("alien vnode on the active list %p %p", vp, mp));
4761 		if (vp->v_mount == mp && (vp->v_iflag & VI_DOOMED) == 0)
4762 			break;
4763 		nvp = TAILQ_NEXT(vp, v_actfreelist);
4764 		VI_UNLOCK(vp);
4765 		vp = nvp;
4766 	}
4767 
4768 	/* Check if we are done */
4769 	if (vp == NULL) {
4770 		mtx_unlock(&vnode_free_list_mtx);
4771 		mnt_vnode_markerfree_active(mvp, mp);
4772 		return (NULL);
4773 	}
4774 	TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
4775 	mtx_unlock(&vnode_free_list_mtx);
4776 	ASSERT_VI_LOCKED(vp, "active iter");
4777 	KASSERT((vp->v_iflag & VI_ACTIVE) != 0, ("Non-active vp %p", vp));
4778 	return (vp);
4779 }
4780 #undef ALWAYS_YIELD
4781 
4782 struct vnode *
4783 __mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4784 {
4785 
4786 	if (should_yield())
4787 		kern_yield(PRI_USER);
4788 	mtx_lock(&vnode_free_list_mtx);
4789 	return (mnt_vnode_next_active(mvp, mp));
4790 }
4791 
4792 struct vnode *
4793 __mnt_vnode_first_active(struct vnode **mvp, struct mount *mp)
4794 {
4795 	struct vnode *vp;
4796 
4797 	*mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4798 	MNT_ILOCK(mp);
4799 	MNT_REF(mp);
4800 	MNT_IUNLOCK(mp);
4801 	(*mvp)->v_type = VMARKER;
4802 	(*mvp)->v_mount = mp;
4803 
4804 	mtx_lock(&vnode_free_list_mtx);
4805 	vp = TAILQ_FIRST(&mp->mnt_activevnodelist);
4806 	if (vp == NULL) {
4807 		mtx_unlock(&vnode_free_list_mtx);
4808 		mnt_vnode_markerfree_active(mvp, mp);
4809 		return (NULL);
4810 	}
4811 	TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
4812 	return (mnt_vnode_next_active(mvp, mp));
4813 }
4814 
4815 void
4816 __mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4817 {
4818 
4819 	if (*mvp == NULL)
4820 		return;
4821 
4822 	mtx_lock(&vnode_free_list_mtx);
4823 	TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4824 	mtx_unlock(&vnode_free_list_mtx);
4825 	mnt_vnode_markerfree_active(mvp, mp);
4826 }
4827