xref: /freebsd/sys/kern/vfs_subr.c (revision 7aa65846327fe5bc7e5961c2f7fd0c61f2ec0b01)
1 /*-
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)vfs_subr.c	8.31 (Berkeley) 5/26/95
35  */
36 
37 /*
38  * External virtual filesystem routines
39  */
40 
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
43 
44 #include "opt_ddb.h"
45 #include "opt_watchdog.h"
46 
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/bio.h>
50 #include <sys/buf.h>
51 #include <sys/condvar.h>
52 #include <sys/conf.h>
53 #include <sys/dirent.h>
54 #include <sys/event.h>
55 #include <sys/eventhandler.h>
56 #include <sys/extattr.h>
57 #include <sys/file.h>
58 #include <sys/fcntl.h>
59 #include <sys/jail.h>
60 #include <sys/kdb.h>
61 #include <sys/kernel.h>
62 #include <sys/kthread.h>
63 #include <sys/lockf.h>
64 #include <sys/malloc.h>
65 #include <sys/mount.h>
66 #include <sys/namei.h>
67 #include <sys/priv.h>
68 #include <sys/reboot.h>
69 #include <sys/sched.h>
70 #include <sys/sleepqueue.h>
71 #include <sys/stat.h>
72 #include <sys/sysctl.h>
73 #include <sys/syslog.h>
74 #include <sys/vmmeter.h>
75 #include <sys/vnode.h>
76 #ifdef SW_WATCHDOG
77 #include <sys/watchdog.h>
78 #endif
79 
80 #include <machine/stdarg.h>
81 
82 #include <security/mac/mac_framework.h>
83 
84 #include <vm/vm.h>
85 #include <vm/vm_object.h>
86 #include <vm/vm_extern.h>
87 #include <vm/pmap.h>
88 #include <vm/vm_map.h>
89 #include <vm/vm_page.h>
90 #include <vm/vm_kern.h>
91 #include <vm/uma.h>
92 
93 #ifdef DDB
94 #include <ddb/ddb.h>
95 #endif
96 
97 #define	WI_MPSAFEQ	0
98 #define	WI_GIANTQ	1
99 
100 static void	delmntque(struct vnode *vp);
101 static int	flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
102 		    int slpflag, int slptimeo);
103 static void	syncer_shutdown(void *arg, int howto);
104 static int	vtryrecycle(struct vnode *vp);
105 static void	v_incr_usecount(struct vnode *);
106 static void	v_decr_usecount(struct vnode *);
107 static void	v_decr_useonly(struct vnode *);
108 static void	v_upgrade_usecount(struct vnode *);
109 static void	vnlru_free(int);
110 static void	vgonel(struct vnode *);
111 static void	vfs_knllock(void *arg);
112 static void	vfs_knlunlock(void *arg);
113 static void	vfs_knl_assert_locked(void *arg);
114 static void	vfs_knl_assert_unlocked(void *arg);
115 static void	destroy_vpollinfo(struct vpollinfo *vi);
116 
117 /*
118  * Number of vnodes in existence.  Increased whenever getnewvnode()
119  * allocates a new vnode, decreased in vdropl() for VI_DOOMED vnode.
120  */
121 static unsigned long	numvnodes;
122 
123 SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
124     "Number of vnodes in existence");
125 
126 /*
127  * Conversion tables for conversion from vnode types to inode formats
128  * and back.
129  */
130 enum vtype iftovt_tab[16] = {
131 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
132 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
133 };
134 int vttoif_tab[10] = {
135 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
136 	S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
137 };
138 
139 /*
140  * List of vnodes that are ready for recycling.
141  */
142 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
143 
144 /*
145  * Free vnode target.  Free vnodes may simply be files which have been stat'd
146  * but not read.  This is somewhat common, and a small cache of such files
147  * should be kept to avoid recreation costs.
148  */
149 static u_long wantfreevnodes;
150 SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
151 /* Number of vnodes in the free list. */
152 static u_long freevnodes;
153 SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
154     "Number of vnodes in the free list");
155 
156 static int vlru_allow_cache_src;
157 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
158     &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
159 
160 /*
161  * Various variables used for debugging the new implementation of
162  * reassignbuf().
163  * XXX these are probably of (very) limited utility now.
164  */
165 static int reassignbufcalls;
166 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
167     "Number of calls to reassignbuf");
168 
169 /*
170  * Cache for the mount type id assigned to NFS.  This is used for
171  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
172  */
173 int	nfs_mount_type = -1;
174 
175 /* To keep more than one thread at a time from running vfs_getnewfsid */
176 static struct mtx mntid_mtx;
177 
178 /*
179  * Lock for any access to the following:
180  *	vnode_free_list
181  *	numvnodes
182  *	freevnodes
183  */
184 static struct mtx vnode_free_list_mtx;
185 
186 /* Publicly exported FS */
187 struct nfs_public nfs_pub;
188 
189 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
190 static uma_zone_t vnode_zone;
191 static uma_zone_t vnodepoll_zone;
192 
193 /*
194  * The workitem queue.
195  *
196  * It is useful to delay writes of file data and filesystem metadata
197  * for tens of seconds so that quickly created and deleted files need
198  * not waste disk bandwidth being created and removed. To realize this,
199  * we append vnodes to a "workitem" queue. When running with a soft
200  * updates implementation, most pending metadata dependencies should
201  * not wait for more than a few seconds. Thus, mounted on block devices
202  * are delayed only about a half the time that file data is delayed.
203  * Similarly, directory updates are more critical, so are only delayed
204  * about a third the time that file data is delayed. Thus, there are
205  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
206  * one each second (driven off the filesystem syncer process). The
207  * syncer_delayno variable indicates the next queue that is to be processed.
208  * Items that need to be processed soon are placed in this queue:
209  *
210  *	syncer_workitem_pending[syncer_delayno]
211  *
212  * A delay of fifteen seconds is done by placing the request fifteen
213  * entries later in the queue:
214  *
215  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
216  *
217  */
218 static int syncer_delayno;
219 static long syncer_mask;
220 LIST_HEAD(synclist, bufobj);
221 static struct synclist *syncer_workitem_pending[2];
222 /*
223  * The sync_mtx protects:
224  *	bo->bo_synclist
225  *	sync_vnode_count
226  *	syncer_delayno
227  *	syncer_state
228  *	syncer_workitem_pending
229  *	syncer_worklist_len
230  *	rushjob
231  */
232 static struct mtx sync_mtx;
233 static struct cv sync_wakeup;
234 
235 #define SYNCER_MAXDELAY		32
236 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
237 static int syncdelay = 30;		/* max time to delay syncing data */
238 static int filedelay = 30;		/* time to delay syncing files */
239 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
240     "Time to delay syncing files (in seconds)");
241 static int dirdelay = 29;		/* time to delay syncing directories */
242 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
243     "Time to delay syncing directories (in seconds)");
244 static int metadelay = 28;		/* time to delay syncing metadata */
245 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
246     "Time to delay syncing metadata (in seconds)");
247 static int rushjob;		/* number of slots to run ASAP */
248 static int stat_rush_requests;	/* number of times I/O speeded up */
249 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
250     "Number of times I/O speeded up (rush requests)");
251 
252 /*
253  * When shutting down the syncer, run it at four times normal speed.
254  */
255 #define SYNCER_SHUTDOWN_SPEEDUP		4
256 static int sync_vnode_count;
257 static int syncer_worklist_len;
258 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
259     syncer_state;
260 
261 /*
262  * Number of vnodes we want to exist at any one time.  This is mostly used
263  * to size hash tables in vnode-related code.  It is normally not used in
264  * getnewvnode(), as wantfreevnodes is normally nonzero.)
265  *
266  * XXX desiredvnodes is historical cruft and should not exist.
267  */
268 int desiredvnodes;
269 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
270     &desiredvnodes, 0, "Maximum number of vnodes");
271 SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
272     &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
273 static int vnlru_nowhere;
274 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
275     &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
276 
277 /*
278  * Macros to control when a vnode is freed and recycled.  All require
279  * the vnode interlock.
280  */
281 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
282 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
283 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
284 
285 
286 /*
287  * Initialize the vnode management data structures.
288  *
289  * Reevaluate the following cap on the number of vnodes after the physical
290  * memory size exceeds 512GB.  In the limit, as the physical memory size
291  * grows, the ratio of physical pages to vnodes approaches sixteen to one.
292  */
293 #ifndef	MAXVNODES_MAX
294 #define	MAXVNODES_MAX	(512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
295 #endif
296 static void
297 vntblinit(void *dummy __unused)
298 {
299 	int physvnodes, virtvnodes;
300 
301 	/*
302 	 * Desiredvnodes is a function of the physical memory size and the
303 	 * kernel's heap size.  Generally speaking, it scales with the
304 	 * physical memory size.  The ratio of desiredvnodes to physical pages
305 	 * is one to four until desiredvnodes exceeds 98,304.  Thereafter, the
306 	 * marginal ratio of desiredvnodes to physical pages is one to
307 	 * sixteen.  However, desiredvnodes is limited by the kernel's heap
308 	 * size.  The memory required by desiredvnodes vnodes and vm objects
309 	 * may not exceed one seventh of the kernel's heap size.
310 	 */
311 	physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
312 	    cnt.v_page_count) / 16;
313 	virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
314 	    sizeof(struct vnode)));
315 	desiredvnodes = min(physvnodes, virtvnodes);
316 	if (desiredvnodes > MAXVNODES_MAX) {
317 		if (bootverbose)
318 			printf("Reducing kern.maxvnodes %d -> %d\n",
319 			    desiredvnodes, MAXVNODES_MAX);
320 		desiredvnodes = MAXVNODES_MAX;
321 	}
322 	wantfreevnodes = desiredvnodes / 4;
323 	mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
324 	TAILQ_INIT(&vnode_free_list);
325 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
326 	vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
327 	    NULL, NULL, UMA_ALIGN_PTR, 0);
328 	vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
329 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
330 	/*
331 	 * Initialize the filesystem syncer.
332 	 */
333 	syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE,
334 	    &syncer_mask);
335 	syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE,
336 	    &syncer_mask);
337 	syncer_maxdelay = syncer_mask + 1;
338 	mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
339 	cv_init(&sync_wakeup, "syncer");
340 }
341 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
342 
343 
344 /*
345  * Mark a mount point as busy. Used to synchronize access and to delay
346  * unmounting. Eventually, mountlist_mtx is not released on failure.
347  *
348  * vfs_busy() is a custom lock, it can block the caller.
349  * vfs_busy() only sleeps if the unmount is active on the mount point.
350  * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
351  * vnode belonging to mp.
352  *
353  * Lookup uses vfs_busy() to traverse mount points.
354  * root fs			var fs
355  * / vnode lock		A	/ vnode lock (/var)		D
356  * /var vnode lock	B	/log vnode lock(/var/log)	E
357  * vfs_busy lock	C	vfs_busy lock			F
358  *
359  * Within each file system, the lock order is C->A->B and F->D->E.
360  *
361  * When traversing across mounts, the system follows that lock order:
362  *
363  *        C->A->B
364  *              |
365  *              +->F->D->E
366  *
367  * The lookup() process for namei("/var") illustrates the process:
368  *  VOP_LOOKUP() obtains B while A is held
369  *  vfs_busy() obtains a shared lock on F while A and B are held
370  *  vput() releases lock on B
371  *  vput() releases lock on A
372  *  VFS_ROOT() obtains lock on D while shared lock on F is held
373  *  vfs_unbusy() releases shared lock on F
374  *  vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
375  *    Attempt to lock A (instead of vp_crossmp) while D is held would
376  *    violate the global order, causing deadlocks.
377  *
378  * dounmount() locks B while F is drained.
379  */
380 int
381 vfs_busy(struct mount *mp, int flags)
382 {
383 
384 	MPASS((flags & ~MBF_MASK) == 0);
385 	CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
386 
387 	MNT_ILOCK(mp);
388 	MNT_REF(mp);
389 	/*
390 	 * If mount point is currenly being unmounted, sleep until the
391 	 * mount point fate is decided.  If thread doing the unmounting fails,
392 	 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
393 	 * that this mount point has survived the unmount attempt and vfs_busy
394 	 * should retry.  Otherwise the unmounter thread will set MNTK_REFEXPIRE
395 	 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
396 	 * about to be really destroyed.  vfs_busy needs to release its
397 	 * reference on the mount point in this case and return with ENOENT,
398 	 * telling the caller that mount mount it tried to busy is no longer
399 	 * valid.
400 	 */
401 	while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
402 		if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
403 			MNT_REL(mp);
404 			MNT_IUNLOCK(mp);
405 			CTR1(KTR_VFS, "%s: failed busying before sleeping",
406 			    __func__);
407 			return (ENOENT);
408 		}
409 		if (flags & MBF_MNTLSTLOCK)
410 			mtx_unlock(&mountlist_mtx);
411 		mp->mnt_kern_flag |= MNTK_MWAIT;
412 		msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
413 		if (flags & MBF_MNTLSTLOCK)
414 			mtx_lock(&mountlist_mtx);
415 		MNT_ILOCK(mp);
416 	}
417 	if (flags & MBF_MNTLSTLOCK)
418 		mtx_unlock(&mountlist_mtx);
419 	mp->mnt_lockref++;
420 	MNT_IUNLOCK(mp);
421 	return (0);
422 }
423 
424 /*
425  * Free a busy filesystem.
426  */
427 void
428 vfs_unbusy(struct mount *mp)
429 {
430 
431 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
432 	MNT_ILOCK(mp);
433 	MNT_REL(mp);
434 	KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
435 	mp->mnt_lockref--;
436 	if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
437 		MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
438 		CTR1(KTR_VFS, "%s: waking up waiters", __func__);
439 		mp->mnt_kern_flag &= ~MNTK_DRAINING;
440 		wakeup(&mp->mnt_lockref);
441 	}
442 	MNT_IUNLOCK(mp);
443 }
444 
445 /*
446  * Lookup a mount point by filesystem identifier.
447  */
448 struct mount *
449 vfs_getvfs(fsid_t *fsid)
450 {
451 	struct mount *mp;
452 
453 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
454 	mtx_lock(&mountlist_mtx);
455 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
456 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
457 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
458 			vfs_ref(mp);
459 			mtx_unlock(&mountlist_mtx);
460 			return (mp);
461 		}
462 	}
463 	mtx_unlock(&mountlist_mtx);
464 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
465 	return ((struct mount *) 0);
466 }
467 
468 /*
469  * Lookup a mount point by filesystem identifier, busying it before
470  * returning.
471  */
472 struct mount *
473 vfs_busyfs(fsid_t *fsid)
474 {
475 	struct mount *mp;
476 	int error;
477 
478 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
479 	mtx_lock(&mountlist_mtx);
480 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
481 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
482 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
483 			error = vfs_busy(mp, MBF_MNTLSTLOCK);
484 			if (error) {
485 				mtx_unlock(&mountlist_mtx);
486 				return (NULL);
487 			}
488 			return (mp);
489 		}
490 	}
491 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
492 	mtx_unlock(&mountlist_mtx);
493 	return ((struct mount *) 0);
494 }
495 
496 /*
497  * Check if a user can access privileged mount options.
498  */
499 int
500 vfs_suser(struct mount *mp, struct thread *td)
501 {
502 	int error;
503 
504 	/*
505 	 * If the thread is jailed, but this is not a jail-friendly file
506 	 * system, deny immediately.
507 	 */
508 	if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
509 		return (EPERM);
510 
511 	/*
512 	 * If the file system was mounted outside the jail of the calling
513 	 * thread, deny immediately.
514 	 */
515 	if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
516 		return (EPERM);
517 
518 	/*
519 	 * If file system supports delegated administration, we don't check
520 	 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
521 	 * by the file system itself.
522 	 * If this is not the user that did original mount, we check for
523 	 * the PRIV_VFS_MOUNT_OWNER privilege.
524 	 */
525 	if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
526 	    mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
527 		if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
528 			return (error);
529 	}
530 	return (0);
531 }
532 
533 /*
534  * Get a new unique fsid.  Try to make its val[0] unique, since this value
535  * will be used to create fake device numbers for stat().  Also try (but
536  * not so hard) make its val[0] unique mod 2^16, since some emulators only
537  * support 16-bit device numbers.  We end up with unique val[0]'s for the
538  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
539  *
540  * Keep in mind that several mounts may be running in parallel.  Starting
541  * the search one past where the previous search terminated is both a
542  * micro-optimization and a defense against returning the same fsid to
543  * different mounts.
544  */
545 void
546 vfs_getnewfsid(struct mount *mp)
547 {
548 	static uint16_t mntid_base;
549 	struct mount *nmp;
550 	fsid_t tfsid;
551 	int mtype;
552 
553 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
554 	mtx_lock(&mntid_mtx);
555 	mtype = mp->mnt_vfc->vfc_typenum;
556 	tfsid.val[1] = mtype;
557 	mtype = (mtype & 0xFF) << 24;
558 	for (;;) {
559 		tfsid.val[0] = makedev(255,
560 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
561 		mntid_base++;
562 		if ((nmp = vfs_getvfs(&tfsid)) == NULL)
563 			break;
564 		vfs_rel(nmp);
565 	}
566 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
567 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
568 	mtx_unlock(&mntid_mtx);
569 }
570 
571 /*
572  * Knob to control the precision of file timestamps:
573  *
574  *   0 = seconds only; nanoseconds zeroed.
575  *   1 = seconds and nanoseconds, accurate within 1/HZ.
576  *   2 = seconds and nanoseconds, truncated to microseconds.
577  * >=3 = seconds and nanoseconds, maximum precision.
578  */
579 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
580 
581 static int timestamp_precision = TSP_SEC;
582 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
583     &timestamp_precision, 0, "File timestamp precision (0: seconds, "
584     "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
585     "3+: sec + ns (max. precision))");
586 
587 /*
588  * Get a current timestamp.
589  */
590 void
591 vfs_timestamp(struct timespec *tsp)
592 {
593 	struct timeval tv;
594 
595 	switch (timestamp_precision) {
596 	case TSP_SEC:
597 		tsp->tv_sec = time_second;
598 		tsp->tv_nsec = 0;
599 		break;
600 	case TSP_HZ:
601 		getnanotime(tsp);
602 		break;
603 	case TSP_USEC:
604 		microtime(&tv);
605 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
606 		break;
607 	case TSP_NSEC:
608 	default:
609 		nanotime(tsp);
610 		break;
611 	}
612 }
613 
614 /*
615  * Set vnode attributes to VNOVAL
616  */
617 void
618 vattr_null(struct vattr *vap)
619 {
620 
621 	vap->va_type = VNON;
622 	vap->va_size = VNOVAL;
623 	vap->va_bytes = VNOVAL;
624 	vap->va_mode = VNOVAL;
625 	vap->va_nlink = VNOVAL;
626 	vap->va_uid = VNOVAL;
627 	vap->va_gid = VNOVAL;
628 	vap->va_fsid = VNOVAL;
629 	vap->va_fileid = VNOVAL;
630 	vap->va_blocksize = VNOVAL;
631 	vap->va_rdev = VNOVAL;
632 	vap->va_atime.tv_sec = VNOVAL;
633 	vap->va_atime.tv_nsec = VNOVAL;
634 	vap->va_mtime.tv_sec = VNOVAL;
635 	vap->va_mtime.tv_nsec = VNOVAL;
636 	vap->va_ctime.tv_sec = VNOVAL;
637 	vap->va_ctime.tv_nsec = VNOVAL;
638 	vap->va_birthtime.tv_sec = VNOVAL;
639 	vap->va_birthtime.tv_nsec = VNOVAL;
640 	vap->va_flags = VNOVAL;
641 	vap->va_gen = VNOVAL;
642 	vap->va_vaflags = 0;
643 }
644 
645 /*
646  * This routine is called when we have too many vnodes.  It attempts
647  * to free <count> vnodes and will potentially free vnodes that still
648  * have VM backing store (VM backing store is typically the cause
649  * of a vnode blowout so we want to do this).  Therefore, this operation
650  * is not considered cheap.
651  *
652  * A number of conditions may prevent a vnode from being reclaimed.
653  * the buffer cache may have references on the vnode, a directory
654  * vnode may still have references due to the namei cache representing
655  * underlying files, or the vnode may be in active use.   It is not
656  * desireable to reuse such vnodes.  These conditions may cause the
657  * number of vnodes to reach some minimum value regardless of what
658  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
659  */
660 static int
661 vlrureclaim(struct mount *mp)
662 {
663 	struct vnode *vp;
664 	int done;
665 	int trigger;
666 	int usevnodes;
667 	int count;
668 
669 	/*
670 	 * Calculate the trigger point, don't allow user
671 	 * screwups to blow us up.   This prevents us from
672 	 * recycling vnodes with lots of resident pages.  We
673 	 * aren't trying to free memory, we are trying to
674 	 * free vnodes.
675 	 */
676 	usevnodes = desiredvnodes;
677 	if (usevnodes <= 0)
678 		usevnodes = 1;
679 	trigger = cnt.v_page_count * 2 / usevnodes;
680 	done = 0;
681 	vn_start_write(NULL, &mp, V_WAIT);
682 	MNT_ILOCK(mp);
683 	count = mp->mnt_nvnodelistsize / 10 + 1;
684 	while (count != 0) {
685 		vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
686 		while (vp != NULL && vp->v_type == VMARKER)
687 			vp = TAILQ_NEXT(vp, v_nmntvnodes);
688 		if (vp == NULL)
689 			break;
690 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
691 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
692 		--count;
693 		if (!VI_TRYLOCK(vp))
694 			goto next_iter;
695 		/*
696 		 * If it's been deconstructed already, it's still
697 		 * referenced, or it exceeds the trigger, skip it.
698 		 */
699 		if (vp->v_usecount ||
700 		    (!vlru_allow_cache_src &&
701 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
702 		    (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
703 		    vp->v_object->resident_page_count > trigger)) {
704 			VI_UNLOCK(vp);
705 			goto next_iter;
706 		}
707 		MNT_IUNLOCK(mp);
708 		vholdl(vp);
709 		if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
710 			vdrop(vp);
711 			goto next_iter_mntunlocked;
712 		}
713 		VI_LOCK(vp);
714 		/*
715 		 * v_usecount may have been bumped after VOP_LOCK() dropped
716 		 * the vnode interlock and before it was locked again.
717 		 *
718 		 * It is not necessary to recheck VI_DOOMED because it can
719 		 * only be set by another thread that holds both the vnode
720 		 * lock and vnode interlock.  If another thread has the
721 		 * vnode lock before we get to VOP_LOCK() and obtains the
722 		 * vnode interlock after VOP_LOCK() drops the vnode
723 		 * interlock, the other thread will be unable to drop the
724 		 * vnode lock before our VOP_LOCK() call fails.
725 		 */
726 		if (vp->v_usecount ||
727 		    (!vlru_allow_cache_src &&
728 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
729 		    (vp->v_object != NULL &&
730 		    vp->v_object->resident_page_count > trigger)) {
731 			VOP_UNLOCK(vp, LK_INTERLOCK);
732 			goto next_iter_mntunlocked;
733 		}
734 		KASSERT((vp->v_iflag & VI_DOOMED) == 0,
735 		    ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
736 		vgonel(vp);
737 		VOP_UNLOCK(vp, 0);
738 		vdropl(vp);
739 		done++;
740 next_iter_mntunlocked:
741 		if (!should_yield())
742 			goto relock_mnt;
743 		goto yield;
744 next_iter:
745 		if (!should_yield())
746 			continue;
747 		MNT_IUNLOCK(mp);
748 yield:
749 		kern_yield(PRI_UNCHANGED);
750 relock_mnt:
751 		MNT_ILOCK(mp);
752 	}
753 	MNT_IUNLOCK(mp);
754 	vn_finished_write(mp);
755 	return done;
756 }
757 
758 /*
759  * Attempt to keep the free list at wantfreevnodes length.
760  */
761 static void
762 vnlru_free(int count)
763 {
764 	struct vnode *vp;
765 	int vfslocked;
766 
767 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
768 	for (; count > 0; count--) {
769 		vp = TAILQ_FIRST(&vnode_free_list);
770 		/*
771 		 * The list can be modified while the free_list_mtx
772 		 * has been dropped and vp could be NULL here.
773 		 */
774 		if (!vp)
775 			break;
776 		VNASSERT(vp->v_op != NULL, vp,
777 		    ("vnlru_free: vnode already reclaimed."));
778 		KASSERT((vp->v_iflag & VI_FREE) != 0,
779 		    ("Removing vnode not on freelist"));
780 		KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
781 		    ("Mangling active vnode"));
782 		TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
783 		/*
784 		 * Don't recycle if we can't get the interlock.
785 		 */
786 		if (!VI_TRYLOCK(vp)) {
787 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
788 			continue;
789 		}
790 		VNASSERT(VCANRECYCLE(vp), vp,
791 		    ("vp inconsistent on freelist"));
792 		freevnodes--;
793 		vp->v_iflag &= ~VI_FREE;
794 		vholdl(vp);
795 		mtx_unlock(&vnode_free_list_mtx);
796 		VI_UNLOCK(vp);
797 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
798 		vtryrecycle(vp);
799 		VFS_UNLOCK_GIANT(vfslocked);
800 		/*
801 		 * If the recycled succeeded this vdrop will actually free
802 		 * the vnode.  If not it will simply place it back on
803 		 * the free list.
804 		 */
805 		vdrop(vp);
806 		mtx_lock(&vnode_free_list_mtx);
807 	}
808 }
809 /*
810  * Attempt to recycle vnodes in a context that is always safe to block.
811  * Calling vlrurecycle() from the bowels of filesystem code has some
812  * interesting deadlock problems.
813  */
814 static struct proc *vnlruproc;
815 static int vnlruproc_sig;
816 
817 static void
818 vnlru_proc(void)
819 {
820 	struct mount *mp, *nmp;
821 	int done, vfslocked;
822 	struct proc *p = vnlruproc;
823 
824 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
825 	    SHUTDOWN_PRI_FIRST);
826 
827 	for (;;) {
828 		kproc_suspend_check(p);
829 		mtx_lock(&vnode_free_list_mtx);
830 		if (freevnodes > wantfreevnodes)
831 			vnlru_free(freevnodes - wantfreevnodes);
832 		if (numvnodes <= desiredvnodes * 9 / 10) {
833 			vnlruproc_sig = 0;
834 			wakeup(&vnlruproc_sig);
835 			msleep(vnlruproc, &vnode_free_list_mtx,
836 			    PVFS|PDROP, "vlruwt", hz);
837 			continue;
838 		}
839 		mtx_unlock(&vnode_free_list_mtx);
840 		done = 0;
841 		mtx_lock(&mountlist_mtx);
842 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
843 			if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
844 				nmp = TAILQ_NEXT(mp, mnt_list);
845 				continue;
846 			}
847 			vfslocked = VFS_LOCK_GIANT(mp);
848 			done += vlrureclaim(mp);
849 			VFS_UNLOCK_GIANT(vfslocked);
850 			mtx_lock(&mountlist_mtx);
851 			nmp = TAILQ_NEXT(mp, mnt_list);
852 			vfs_unbusy(mp);
853 		}
854 		mtx_unlock(&mountlist_mtx);
855 		if (done == 0) {
856 #if 0
857 			/* These messages are temporary debugging aids */
858 			if (vnlru_nowhere < 5)
859 				printf("vnlru process getting nowhere..\n");
860 			else if (vnlru_nowhere == 5)
861 				printf("vnlru process messages stopped.\n");
862 #endif
863 			vnlru_nowhere++;
864 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
865 		} else
866 			kern_yield(PRI_UNCHANGED);
867 	}
868 }
869 
870 static struct kproc_desc vnlru_kp = {
871 	"vnlru",
872 	vnlru_proc,
873 	&vnlruproc
874 };
875 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
876     &vnlru_kp);
877 
878 /*
879  * Routines having to do with the management of the vnode table.
880  */
881 
882 /*
883  * Try to recycle a freed vnode.  We abort if anyone picks up a reference
884  * before we actually vgone().  This function must be called with the vnode
885  * held to prevent the vnode from being returned to the free list midway
886  * through vgone().
887  */
888 static int
889 vtryrecycle(struct vnode *vp)
890 {
891 	struct mount *vnmp;
892 
893 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
894 	VNASSERT(vp->v_holdcnt, vp,
895 	    ("vtryrecycle: Recycling vp %p without a reference.", vp));
896 	/*
897 	 * This vnode may found and locked via some other list, if so we
898 	 * can't recycle it yet.
899 	 */
900 	if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
901 		CTR2(KTR_VFS,
902 		    "%s: impossible to recycle, vp %p lock is already held",
903 		    __func__, vp);
904 		return (EWOULDBLOCK);
905 	}
906 	/*
907 	 * Don't recycle if its filesystem is being suspended.
908 	 */
909 	if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
910 		VOP_UNLOCK(vp, 0);
911 		CTR2(KTR_VFS,
912 		    "%s: impossible to recycle, cannot start the write for %p",
913 		    __func__, vp);
914 		return (EBUSY);
915 	}
916 	/*
917 	 * If we got this far, we need to acquire the interlock and see if
918 	 * anyone picked up this vnode from another list.  If not, we will
919 	 * mark it with DOOMED via vgonel() so that anyone who does find it
920 	 * will skip over it.
921 	 */
922 	VI_LOCK(vp);
923 	if (vp->v_usecount) {
924 		VOP_UNLOCK(vp, LK_INTERLOCK);
925 		vn_finished_write(vnmp);
926 		CTR2(KTR_VFS,
927 		    "%s: impossible to recycle, %p is already referenced",
928 		    __func__, vp);
929 		return (EBUSY);
930 	}
931 	if ((vp->v_iflag & VI_DOOMED) == 0)
932 		vgonel(vp);
933 	VOP_UNLOCK(vp, LK_INTERLOCK);
934 	vn_finished_write(vnmp);
935 	return (0);
936 }
937 
938 /*
939  * Return the next vnode from the free list.
940  */
941 int
942 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
943     struct vnode **vpp)
944 {
945 	struct vnode *vp = NULL;
946 	struct bufobj *bo;
947 
948 	CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
949 	mtx_lock(&vnode_free_list_mtx);
950 	/*
951 	 * Lend our context to reclaim vnodes if they've exceeded the max.
952 	 */
953 	if (freevnodes > wantfreevnodes)
954 		vnlru_free(1);
955 	/*
956 	 * Wait for available vnodes.
957 	 */
958 	if (numvnodes > desiredvnodes) {
959 		if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) {
960 			/*
961 			 * File system is beeing suspended, we cannot risk a
962 			 * deadlock here, so allocate new vnode anyway.
963 			 */
964 			if (freevnodes > wantfreevnodes)
965 				vnlru_free(freevnodes - wantfreevnodes);
966 			goto alloc;
967 		}
968 		if (vnlruproc_sig == 0) {
969 			vnlruproc_sig = 1;	/* avoid unnecessary wakeups */
970 			wakeup(vnlruproc);
971 		}
972 		msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
973 		    "vlruwk", hz);
974 #if 0	/* XXX Not all VFS_VGET/ffs_vget callers check returns. */
975 		if (numvnodes > desiredvnodes) {
976 			mtx_unlock(&vnode_free_list_mtx);
977 			return (ENFILE);
978 		}
979 #endif
980 	}
981 alloc:
982 	numvnodes++;
983 	mtx_unlock(&vnode_free_list_mtx);
984 	vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
985 	/*
986 	 * Setup locks.
987 	 */
988 	vp->v_vnlock = &vp->v_lock;
989 	mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
990 	/*
991 	 * By default, don't allow shared locks unless filesystems
992 	 * opt-in.
993 	 */
994 	lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
995 	/*
996 	 * Initialize bufobj.
997 	 */
998 	bo = &vp->v_bufobj;
999 	bo->__bo_vnode = vp;
1000 	mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
1001 	bo->bo_ops = &buf_ops_bio;
1002 	bo->bo_private = vp;
1003 	TAILQ_INIT(&bo->bo_clean.bv_hd);
1004 	TAILQ_INIT(&bo->bo_dirty.bv_hd);
1005 	/*
1006 	 * Initialize namecache.
1007 	 */
1008 	LIST_INIT(&vp->v_cache_src);
1009 	TAILQ_INIT(&vp->v_cache_dst);
1010 	/*
1011 	 * Finalize various vnode identity bits.
1012 	 */
1013 	vp->v_type = VNON;
1014 	vp->v_tag = tag;
1015 	vp->v_op = vops;
1016 	v_incr_usecount(vp);
1017 	vp->v_data = NULL;
1018 #ifdef MAC
1019 	mac_vnode_init(vp);
1020 	if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1021 		mac_vnode_associate_singlelabel(mp, vp);
1022 	else if (mp == NULL && vops != &dead_vnodeops)
1023 		printf("NULL mp in getnewvnode()\n");
1024 #endif
1025 	if (mp != NULL) {
1026 		bo->bo_bsize = mp->mnt_stat.f_iosize;
1027 		if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1028 			vp->v_vflag |= VV_NOKNOTE;
1029 	}
1030 
1031 	*vpp = vp;
1032 	return (0);
1033 }
1034 
1035 /*
1036  * Delete from old mount point vnode list, if on one.
1037  */
1038 static void
1039 delmntque(struct vnode *vp)
1040 {
1041 	struct mount *mp;
1042 	int active;
1043 
1044 	mp = vp->v_mount;
1045 	if (mp == NULL)
1046 		return;
1047 	MNT_ILOCK(mp);
1048 	VI_LOCK(vp);
1049 	KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize,
1050 	    ("Active vnode list size %d > Vnode list size %d",
1051 	     mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize));
1052 	active = vp->v_iflag & VI_ACTIVE;
1053 	vp->v_iflag &= ~VI_ACTIVE;
1054 	if (active) {
1055 		mtx_lock(&vnode_free_list_mtx);
1056 		TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist);
1057 		mp->mnt_activevnodelistsize--;
1058 		mtx_unlock(&vnode_free_list_mtx);
1059 	}
1060 	vp->v_mount = NULL;
1061 	VI_UNLOCK(vp);
1062 	VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1063 		("bad mount point vnode list size"));
1064 	TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1065 	mp->mnt_nvnodelistsize--;
1066 	MNT_REL(mp);
1067 	MNT_IUNLOCK(mp);
1068 }
1069 
1070 static void
1071 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1072 {
1073 
1074 	vp->v_data = NULL;
1075 	vp->v_op = &dead_vnodeops;
1076 	/* XXX non mp-safe fs may still call insmntque with vnode
1077 	   unlocked */
1078 	if (!VOP_ISLOCKED(vp))
1079 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1080 	vgone(vp);
1081 	vput(vp);
1082 }
1083 
1084 /*
1085  * Insert into list of vnodes for the new mount point, if available.
1086  */
1087 int
1088 insmntque1(struct vnode *vp, struct mount *mp,
1089 	void (*dtr)(struct vnode *, void *), void *dtr_arg)
1090 {
1091 	int locked;
1092 
1093 	KASSERT(vp->v_mount == NULL,
1094 		("insmntque: vnode already on per mount vnode list"));
1095 	VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1096 #ifdef DEBUG_VFS_LOCKS
1097 	if (!VFS_NEEDSGIANT(mp))
1098 		ASSERT_VOP_ELOCKED(vp,
1099 		    "insmntque: mp-safe fs and non-locked vp");
1100 #endif
1101 	/*
1102 	 * We acquire the vnode interlock early to ensure that the
1103 	 * vnode cannot be recycled by another process releasing a
1104 	 * holdcnt on it before we get it on both the vnode list
1105 	 * and the active vnode list. The mount mutex protects only
1106 	 * manipulation of the vnode list and the vnode freelist
1107 	 * mutex protects only manipulation of the active vnode list.
1108 	 * Hence the need to hold the vnode interlock throughout.
1109 	 */
1110 	MNT_ILOCK(mp);
1111 	VI_LOCK(vp);
1112 	if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1113 	    ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1114 	     mp->mnt_nvnodelistsize == 0)) {
1115 		locked = VOP_ISLOCKED(vp);
1116 		if (!locked || (locked == LK_EXCLUSIVE &&
1117 		     (vp->v_vflag & VV_FORCEINSMQ) == 0)) {
1118 			VI_UNLOCK(vp);
1119 			MNT_IUNLOCK(mp);
1120 			if (dtr != NULL)
1121 				dtr(vp, dtr_arg);
1122 			return (EBUSY);
1123 		}
1124 	}
1125 	vp->v_mount = mp;
1126 	MNT_REF(mp);
1127 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1128 	VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1129 		("neg mount point vnode list size"));
1130 	mp->mnt_nvnodelistsize++;
1131 	KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
1132 	    ("Activating already active vnode"));
1133 	vp->v_iflag |= VI_ACTIVE;
1134 	mtx_lock(&vnode_free_list_mtx);
1135 	TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
1136 	mp->mnt_activevnodelistsize++;
1137 	mtx_unlock(&vnode_free_list_mtx);
1138 	VI_UNLOCK(vp);
1139 	MNT_IUNLOCK(mp);
1140 	return (0);
1141 }
1142 
1143 int
1144 insmntque(struct vnode *vp, struct mount *mp)
1145 {
1146 
1147 	return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1148 }
1149 
1150 /*
1151  * Flush out and invalidate all buffers associated with a bufobj
1152  * Called with the underlying object locked.
1153  */
1154 int
1155 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1156 {
1157 	int error;
1158 
1159 	BO_LOCK(bo);
1160 	if (flags & V_SAVE) {
1161 		error = bufobj_wwait(bo, slpflag, slptimeo);
1162 		if (error) {
1163 			BO_UNLOCK(bo);
1164 			return (error);
1165 		}
1166 		if (bo->bo_dirty.bv_cnt > 0) {
1167 			BO_UNLOCK(bo);
1168 			if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1169 				return (error);
1170 			/*
1171 			 * XXX We could save a lock/unlock if this was only
1172 			 * enabled under INVARIANTS
1173 			 */
1174 			BO_LOCK(bo);
1175 			if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1176 				panic("vinvalbuf: dirty bufs");
1177 		}
1178 	}
1179 	/*
1180 	 * If you alter this loop please notice that interlock is dropped and
1181 	 * reacquired in flushbuflist.  Special care is needed to ensure that
1182 	 * no race conditions occur from this.
1183 	 */
1184 	do {
1185 		error = flushbuflist(&bo->bo_clean,
1186 		    flags, bo, slpflag, slptimeo);
1187 		if (error == 0 && !(flags & V_CLEANONLY))
1188 			error = flushbuflist(&bo->bo_dirty,
1189 			    flags, bo, slpflag, slptimeo);
1190 		if (error != 0 && error != EAGAIN) {
1191 			BO_UNLOCK(bo);
1192 			return (error);
1193 		}
1194 	} while (error != 0);
1195 
1196 	/*
1197 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1198 	 * have write I/O in-progress but if there is a VM object then the
1199 	 * VM object can also have read-I/O in-progress.
1200 	 */
1201 	do {
1202 		bufobj_wwait(bo, 0, 0);
1203 		BO_UNLOCK(bo);
1204 		if (bo->bo_object != NULL) {
1205 			VM_OBJECT_LOCK(bo->bo_object);
1206 			vm_object_pip_wait(bo->bo_object, "bovlbx");
1207 			VM_OBJECT_UNLOCK(bo->bo_object);
1208 		}
1209 		BO_LOCK(bo);
1210 	} while (bo->bo_numoutput > 0);
1211 	BO_UNLOCK(bo);
1212 
1213 	/*
1214 	 * Destroy the copy in the VM cache, too.
1215 	 */
1216 	if (bo->bo_object != NULL &&
1217 	    (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) {
1218 		VM_OBJECT_LOCK(bo->bo_object);
1219 		vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
1220 		    OBJPR_CLEANONLY : 0);
1221 		VM_OBJECT_UNLOCK(bo->bo_object);
1222 	}
1223 
1224 #ifdef INVARIANTS
1225 	BO_LOCK(bo);
1226 	if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 &&
1227 	    (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1228 		panic("vinvalbuf: flush failed");
1229 	BO_UNLOCK(bo);
1230 #endif
1231 	return (0);
1232 }
1233 
1234 /*
1235  * Flush out and invalidate all buffers associated with a vnode.
1236  * Called with the underlying object locked.
1237  */
1238 int
1239 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1240 {
1241 
1242 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1243 	ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1244 	return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1245 }
1246 
1247 /*
1248  * Flush out buffers on the specified list.
1249  *
1250  */
1251 static int
1252 flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1253     int slptimeo)
1254 {
1255 	struct buf *bp, *nbp;
1256 	int retval, error;
1257 	daddr_t lblkno;
1258 	b_xflags_t xflags;
1259 
1260 	ASSERT_BO_LOCKED(bo);
1261 
1262 	retval = 0;
1263 	TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1264 		if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1265 		    ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1266 			continue;
1267 		}
1268 		lblkno = 0;
1269 		xflags = 0;
1270 		if (nbp != NULL) {
1271 			lblkno = nbp->b_lblkno;
1272 			xflags = nbp->b_xflags &
1273 				(BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1274 		}
1275 		retval = EAGAIN;
1276 		error = BUF_TIMELOCK(bp,
1277 		    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1278 		    "flushbuf", slpflag, slptimeo);
1279 		if (error) {
1280 			BO_LOCK(bo);
1281 			return (error != ENOLCK ? error : EAGAIN);
1282 		}
1283 		KASSERT(bp->b_bufobj == bo,
1284 		    ("bp %p wrong b_bufobj %p should be %p",
1285 		    bp, bp->b_bufobj, bo));
1286 		if (bp->b_bufobj != bo) {	/* XXX: necessary ? */
1287 			BUF_UNLOCK(bp);
1288 			BO_LOCK(bo);
1289 			return (EAGAIN);
1290 		}
1291 		/*
1292 		 * XXX Since there are no node locks for NFS, I
1293 		 * believe there is a slight chance that a delayed
1294 		 * write will occur while sleeping just above, so
1295 		 * check for it.
1296 		 */
1297 		if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1298 		    (flags & V_SAVE)) {
1299 			BO_LOCK(bo);
1300 			bremfree(bp);
1301 			BO_UNLOCK(bo);
1302 			bp->b_flags |= B_ASYNC;
1303 			bwrite(bp);
1304 			BO_LOCK(bo);
1305 			return (EAGAIN);	/* XXX: why not loop ? */
1306 		}
1307 		BO_LOCK(bo);
1308 		bremfree(bp);
1309 		BO_UNLOCK(bo);
1310 		bp->b_flags |= (B_INVAL | B_RELBUF);
1311 		bp->b_flags &= ~B_ASYNC;
1312 		brelse(bp);
1313 		BO_LOCK(bo);
1314 		if (nbp != NULL &&
1315 		    (nbp->b_bufobj != bo ||
1316 		     nbp->b_lblkno != lblkno ||
1317 		     (nbp->b_xflags &
1318 		      (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1319 			break;			/* nbp invalid */
1320 	}
1321 	return (retval);
1322 }
1323 
1324 /*
1325  * Truncate a file's buffer and pages to a specified length.  This
1326  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1327  * sync activity.
1328  */
1329 int
1330 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
1331     off_t length, int blksize)
1332 {
1333 	struct buf *bp, *nbp;
1334 	int anyfreed;
1335 	int trunclbn;
1336 	struct bufobj *bo;
1337 
1338 	CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1339 	    vp, cred, blksize, (uintmax_t)length);
1340 
1341 	/*
1342 	 * Round up to the *next* lbn.
1343 	 */
1344 	trunclbn = (length + blksize - 1) / blksize;
1345 
1346 	ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1347 restart:
1348 	bo = &vp->v_bufobj;
1349 	BO_LOCK(bo);
1350 	anyfreed = 1;
1351 	for (;anyfreed;) {
1352 		anyfreed = 0;
1353 		TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1354 			if (bp->b_lblkno < trunclbn)
1355 				continue;
1356 			if (BUF_LOCK(bp,
1357 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1358 			    BO_MTX(bo)) == ENOLCK)
1359 				goto restart;
1360 
1361 			BO_LOCK(bo);
1362 			bremfree(bp);
1363 			BO_UNLOCK(bo);
1364 			bp->b_flags |= (B_INVAL | B_RELBUF);
1365 			bp->b_flags &= ~B_ASYNC;
1366 			brelse(bp);
1367 			anyfreed = 1;
1368 
1369 			BO_LOCK(bo);
1370 			if (nbp != NULL &&
1371 			    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1372 			    (nbp->b_vp != vp) ||
1373 			    (nbp->b_flags & B_DELWRI))) {
1374 				BO_UNLOCK(bo);
1375 				goto restart;
1376 			}
1377 		}
1378 
1379 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1380 			if (bp->b_lblkno < trunclbn)
1381 				continue;
1382 			if (BUF_LOCK(bp,
1383 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1384 			    BO_MTX(bo)) == ENOLCK)
1385 				goto restart;
1386 			BO_LOCK(bo);
1387 			bremfree(bp);
1388 			BO_UNLOCK(bo);
1389 			bp->b_flags |= (B_INVAL | B_RELBUF);
1390 			bp->b_flags &= ~B_ASYNC;
1391 			brelse(bp);
1392 			anyfreed = 1;
1393 
1394 			BO_LOCK(bo);
1395 			if (nbp != NULL &&
1396 			    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1397 			    (nbp->b_vp != vp) ||
1398 			    (nbp->b_flags & B_DELWRI) == 0)) {
1399 				BO_UNLOCK(bo);
1400 				goto restart;
1401 			}
1402 		}
1403 	}
1404 
1405 	if (length > 0) {
1406 restartsync:
1407 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1408 			if (bp->b_lblkno > 0)
1409 				continue;
1410 			/*
1411 			 * Since we hold the vnode lock this should only
1412 			 * fail if we're racing with the buf daemon.
1413 			 */
1414 			if (BUF_LOCK(bp,
1415 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1416 			    BO_MTX(bo)) == ENOLCK) {
1417 				goto restart;
1418 			}
1419 			VNASSERT((bp->b_flags & B_DELWRI), vp,
1420 			    ("buf(%p) on dirty queue without DELWRI", bp));
1421 
1422 			BO_LOCK(bo);
1423 			bremfree(bp);
1424 			BO_UNLOCK(bo);
1425 			bawrite(bp);
1426 			BO_LOCK(bo);
1427 			goto restartsync;
1428 		}
1429 	}
1430 
1431 	bufobj_wwait(bo, 0, 0);
1432 	BO_UNLOCK(bo);
1433 	vnode_pager_setsize(vp, length);
1434 
1435 	return (0);
1436 }
1437 
1438 /*
1439  * buf_splay() - splay tree core for the clean/dirty list of buffers in
1440  *		 a vnode.
1441  *
1442  *	NOTE: We have to deal with the special case of a background bitmap
1443  *	buffer, a situation where two buffers will have the same logical
1444  *	block offset.  We want (1) only the foreground buffer to be accessed
1445  *	in a lookup and (2) must differentiate between the foreground and
1446  *	background buffer in the splay tree algorithm because the splay
1447  *	tree cannot normally handle multiple entities with the same 'index'.
1448  *	We accomplish this by adding differentiating flags to the splay tree's
1449  *	numerical domain.
1450  */
1451 static
1452 struct buf *
1453 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1454 {
1455 	struct buf dummy;
1456 	struct buf *lefttreemax, *righttreemin, *y;
1457 
1458 	if (root == NULL)
1459 		return (NULL);
1460 	lefttreemax = righttreemin = &dummy;
1461 	for (;;) {
1462 		if (lblkno < root->b_lblkno ||
1463 		    (lblkno == root->b_lblkno &&
1464 		    (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1465 			if ((y = root->b_left) == NULL)
1466 				break;
1467 			if (lblkno < y->b_lblkno) {
1468 				/* Rotate right. */
1469 				root->b_left = y->b_right;
1470 				y->b_right = root;
1471 				root = y;
1472 				if ((y = root->b_left) == NULL)
1473 					break;
1474 			}
1475 			/* Link into the new root's right tree. */
1476 			righttreemin->b_left = root;
1477 			righttreemin = root;
1478 		} else if (lblkno > root->b_lblkno ||
1479 		    (lblkno == root->b_lblkno &&
1480 		    (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1481 			if ((y = root->b_right) == NULL)
1482 				break;
1483 			if (lblkno > y->b_lblkno) {
1484 				/* Rotate left. */
1485 				root->b_right = y->b_left;
1486 				y->b_left = root;
1487 				root = y;
1488 				if ((y = root->b_right) == NULL)
1489 					break;
1490 			}
1491 			/* Link into the new root's left tree. */
1492 			lefttreemax->b_right = root;
1493 			lefttreemax = root;
1494 		} else {
1495 			break;
1496 		}
1497 		root = y;
1498 	}
1499 	/* Assemble the new root. */
1500 	lefttreemax->b_right = root->b_left;
1501 	righttreemin->b_left = root->b_right;
1502 	root->b_left = dummy.b_right;
1503 	root->b_right = dummy.b_left;
1504 	return (root);
1505 }
1506 
1507 static void
1508 buf_vlist_remove(struct buf *bp)
1509 {
1510 	struct buf *root;
1511 	struct bufv *bv;
1512 
1513 	KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1514 	ASSERT_BO_LOCKED(bp->b_bufobj);
1515 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1516 	    (BX_VNDIRTY|BX_VNCLEAN),
1517 	    ("buf_vlist_remove: Buf %p is on two lists", bp));
1518 	if (bp->b_xflags & BX_VNDIRTY)
1519 		bv = &bp->b_bufobj->bo_dirty;
1520 	else
1521 		bv = &bp->b_bufobj->bo_clean;
1522 	if (bp != bv->bv_root) {
1523 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1524 		KASSERT(root == bp, ("splay lookup failed in remove"));
1525 	}
1526 	if (bp->b_left == NULL) {
1527 		root = bp->b_right;
1528 	} else {
1529 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1530 		root->b_right = bp->b_right;
1531 	}
1532 	bv->bv_root = root;
1533 	TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1534 	bv->bv_cnt--;
1535 	bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1536 }
1537 
1538 /*
1539  * Add the buffer to the sorted clean or dirty block list using a
1540  * splay tree algorithm.
1541  *
1542  * NOTE: xflags is passed as a constant, optimizing this inline function!
1543  */
1544 static void
1545 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1546 {
1547 	struct buf *root;
1548 	struct bufv *bv;
1549 
1550 	ASSERT_BO_LOCKED(bo);
1551 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1552 	    ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1553 	bp->b_xflags |= xflags;
1554 	if (xflags & BX_VNDIRTY)
1555 		bv = &bo->bo_dirty;
1556 	else
1557 		bv = &bo->bo_clean;
1558 
1559 	root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1560 	if (root == NULL) {
1561 		bp->b_left = NULL;
1562 		bp->b_right = NULL;
1563 		TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1564 	} else if (bp->b_lblkno < root->b_lblkno ||
1565 	    (bp->b_lblkno == root->b_lblkno &&
1566 	    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1567 		bp->b_left = root->b_left;
1568 		bp->b_right = root;
1569 		root->b_left = NULL;
1570 		TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1571 	} else {
1572 		bp->b_right = root->b_right;
1573 		bp->b_left = root;
1574 		root->b_right = NULL;
1575 		TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1576 	}
1577 	bv->bv_cnt++;
1578 	bv->bv_root = bp;
1579 }
1580 
1581 /*
1582  * Lookup a buffer using the splay tree.  Note that we specifically avoid
1583  * shadow buffers used in background bitmap writes.
1584  *
1585  * This code isn't quite efficient as it could be because we are maintaining
1586  * two sorted lists and do not know which list the block resides in.
1587  *
1588  * During a "make buildworld" the desired buffer is found at one of
1589  * the roots more than 60% of the time.  Thus, checking both roots
1590  * before performing either splay eliminates unnecessary splays on the
1591  * first tree splayed.
1592  */
1593 struct buf *
1594 gbincore(struct bufobj *bo, daddr_t lblkno)
1595 {
1596 	struct buf *bp;
1597 
1598 	ASSERT_BO_LOCKED(bo);
1599 	if ((bp = bo->bo_clean.bv_root) != NULL &&
1600 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1601 		return (bp);
1602 	if ((bp = bo->bo_dirty.bv_root) != NULL &&
1603 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1604 		return (bp);
1605 	if ((bp = bo->bo_clean.bv_root) != NULL) {
1606 		bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1607 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1608 			return (bp);
1609 	}
1610 	if ((bp = bo->bo_dirty.bv_root) != NULL) {
1611 		bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1612 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1613 			return (bp);
1614 	}
1615 	return (NULL);
1616 }
1617 
1618 /*
1619  * Associate a buffer with a vnode.
1620  */
1621 void
1622 bgetvp(struct vnode *vp, struct buf *bp)
1623 {
1624 	struct bufobj *bo;
1625 
1626 	bo = &vp->v_bufobj;
1627 	ASSERT_BO_LOCKED(bo);
1628 	VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1629 
1630 	CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1631 	VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1632 	    ("bgetvp: bp already attached! %p", bp));
1633 
1634 	vhold(vp);
1635 	if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT)
1636 		bp->b_flags |= B_NEEDSGIANT;
1637 	bp->b_vp = vp;
1638 	bp->b_bufobj = bo;
1639 	/*
1640 	 * Insert onto list for new vnode.
1641 	 */
1642 	buf_vlist_add(bp, bo, BX_VNCLEAN);
1643 }
1644 
1645 /*
1646  * Disassociate a buffer from a vnode.
1647  */
1648 void
1649 brelvp(struct buf *bp)
1650 {
1651 	struct bufobj *bo;
1652 	struct vnode *vp;
1653 
1654 	CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1655 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1656 
1657 	/*
1658 	 * Delete from old vnode list, if on one.
1659 	 */
1660 	vp = bp->b_vp;		/* XXX */
1661 	bo = bp->b_bufobj;
1662 	BO_LOCK(bo);
1663 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1664 		buf_vlist_remove(bp);
1665 	else
1666 		panic("brelvp: Buffer %p not on queue.", bp);
1667 	if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1668 		bo->bo_flag &= ~BO_ONWORKLST;
1669 		mtx_lock(&sync_mtx);
1670 		LIST_REMOVE(bo, bo_synclist);
1671 		syncer_worklist_len--;
1672 		mtx_unlock(&sync_mtx);
1673 	}
1674 	bp->b_flags &= ~B_NEEDSGIANT;
1675 	bp->b_vp = NULL;
1676 	bp->b_bufobj = NULL;
1677 	BO_UNLOCK(bo);
1678 	vdrop(vp);
1679 }
1680 
1681 /*
1682  * Add an item to the syncer work queue.
1683  */
1684 static void
1685 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1686 {
1687 	int queue, slot;
1688 
1689 	ASSERT_BO_LOCKED(bo);
1690 
1691 	mtx_lock(&sync_mtx);
1692 	if (bo->bo_flag & BO_ONWORKLST)
1693 		LIST_REMOVE(bo, bo_synclist);
1694 	else {
1695 		bo->bo_flag |= BO_ONWORKLST;
1696 		syncer_worklist_len++;
1697 	}
1698 
1699 	if (delay > syncer_maxdelay - 2)
1700 		delay = syncer_maxdelay - 2;
1701 	slot = (syncer_delayno + delay) & syncer_mask;
1702 
1703 	queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ :
1704 	    WI_MPSAFEQ;
1705 	LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo,
1706 	    bo_synclist);
1707 	mtx_unlock(&sync_mtx);
1708 }
1709 
1710 static int
1711 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1712 {
1713 	int error, len;
1714 
1715 	mtx_lock(&sync_mtx);
1716 	len = syncer_worklist_len - sync_vnode_count;
1717 	mtx_unlock(&sync_mtx);
1718 	error = SYSCTL_OUT(req, &len, sizeof(len));
1719 	return (error);
1720 }
1721 
1722 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1723     sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1724 
1725 static struct proc *updateproc;
1726 static void sched_sync(void);
1727 static struct kproc_desc up_kp = {
1728 	"syncer",
1729 	sched_sync,
1730 	&updateproc
1731 };
1732 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1733 
1734 static int
1735 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1736 {
1737 	struct vnode *vp;
1738 	struct mount *mp;
1739 
1740 	*bo = LIST_FIRST(slp);
1741 	if (*bo == NULL)
1742 		return (0);
1743 	vp = (*bo)->__bo_vnode;	/* XXX */
1744 	if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1745 		return (1);
1746 	/*
1747 	 * We use vhold in case the vnode does not
1748 	 * successfully sync.  vhold prevents the vnode from
1749 	 * going away when we unlock the sync_mtx so that
1750 	 * we can acquire the vnode interlock.
1751 	 */
1752 	vholdl(vp);
1753 	mtx_unlock(&sync_mtx);
1754 	VI_UNLOCK(vp);
1755 	if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1756 		vdrop(vp);
1757 		mtx_lock(&sync_mtx);
1758 		return (*bo == LIST_FIRST(slp));
1759 	}
1760 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1761 	(void) VOP_FSYNC(vp, MNT_LAZY, td);
1762 	VOP_UNLOCK(vp, 0);
1763 	vn_finished_write(mp);
1764 	BO_LOCK(*bo);
1765 	if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1766 		/*
1767 		 * Put us back on the worklist.  The worklist
1768 		 * routine will remove us from our current
1769 		 * position and then add us back in at a later
1770 		 * position.
1771 		 */
1772 		vn_syncer_add_to_worklist(*bo, syncdelay);
1773 	}
1774 	BO_UNLOCK(*bo);
1775 	vdrop(vp);
1776 	mtx_lock(&sync_mtx);
1777 	return (0);
1778 }
1779 
1780 /*
1781  * System filesystem synchronizer daemon.
1782  */
1783 static void
1784 sched_sync(void)
1785 {
1786 	struct synclist *gnext, *next;
1787 	struct synclist *gslp, *slp;
1788 	struct bufobj *bo;
1789 	long starttime;
1790 	struct thread *td = curthread;
1791 	int last_work_seen;
1792 	int net_worklist_len;
1793 	int syncer_final_iter;
1794 	int first_printf;
1795 	int error;
1796 
1797 	last_work_seen = 0;
1798 	syncer_final_iter = 0;
1799 	first_printf = 1;
1800 	syncer_state = SYNCER_RUNNING;
1801 	starttime = time_uptime;
1802 	td->td_pflags |= TDP_NORUNNINGBUF;
1803 
1804 	EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1805 	    SHUTDOWN_PRI_LAST);
1806 
1807 	mtx_lock(&sync_mtx);
1808 	for (;;) {
1809 		if (syncer_state == SYNCER_FINAL_DELAY &&
1810 		    syncer_final_iter == 0) {
1811 			mtx_unlock(&sync_mtx);
1812 			kproc_suspend_check(td->td_proc);
1813 			mtx_lock(&sync_mtx);
1814 		}
1815 		net_worklist_len = syncer_worklist_len - sync_vnode_count;
1816 		if (syncer_state != SYNCER_RUNNING &&
1817 		    starttime != time_uptime) {
1818 			if (first_printf) {
1819 				printf("\nSyncing disks, vnodes remaining...");
1820 				first_printf = 0;
1821 			}
1822 			printf("%d ", net_worklist_len);
1823 		}
1824 		starttime = time_uptime;
1825 
1826 		/*
1827 		 * Push files whose dirty time has expired.  Be careful
1828 		 * of interrupt race on slp queue.
1829 		 *
1830 		 * Skip over empty worklist slots when shutting down.
1831 		 */
1832 		do {
1833 			slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1834 			gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1835 			syncer_delayno += 1;
1836 			if (syncer_delayno == syncer_maxdelay)
1837 				syncer_delayno = 0;
1838 			next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1839 			gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1840 			/*
1841 			 * If the worklist has wrapped since the
1842 			 * it was emptied of all but syncer vnodes,
1843 			 * switch to the FINAL_DELAY state and run
1844 			 * for one more second.
1845 			 */
1846 			if (syncer_state == SYNCER_SHUTTING_DOWN &&
1847 			    net_worklist_len == 0 &&
1848 			    last_work_seen == syncer_delayno) {
1849 				syncer_state = SYNCER_FINAL_DELAY;
1850 				syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1851 			}
1852 		} while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1853 		    LIST_EMPTY(gslp) && syncer_worklist_len > 0);
1854 
1855 		/*
1856 		 * Keep track of the last time there was anything
1857 		 * on the worklist other than syncer vnodes.
1858 		 * Return to the SHUTTING_DOWN state if any
1859 		 * new work appears.
1860 		 */
1861 		if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1862 			last_work_seen = syncer_delayno;
1863 		if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1864 			syncer_state = SYNCER_SHUTTING_DOWN;
1865 		while (!LIST_EMPTY(slp)) {
1866 			error = sync_vnode(slp, &bo, td);
1867 			if (error == 1) {
1868 				LIST_REMOVE(bo, bo_synclist);
1869 				LIST_INSERT_HEAD(next, bo, bo_synclist);
1870 				continue;
1871 			}
1872 #ifdef SW_WATCHDOG
1873 			if (first_printf == 0)
1874 				wdog_kern_pat(WD_LASTVAL);
1875 #endif
1876 		}
1877 		if (!LIST_EMPTY(gslp)) {
1878 			mtx_unlock(&sync_mtx);
1879 			mtx_lock(&Giant);
1880 			mtx_lock(&sync_mtx);
1881 			while (!LIST_EMPTY(gslp)) {
1882 				error = sync_vnode(gslp, &bo, td);
1883 				if (error == 1) {
1884 					LIST_REMOVE(bo, bo_synclist);
1885 					LIST_INSERT_HEAD(gnext, bo,
1886 					    bo_synclist);
1887 					continue;
1888 				}
1889 			}
1890 			mtx_unlock(&Giant);
1891 		}
1892 		if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1893 			syncer_final_iter--;
1894 		/*
1895 		 * The variable rushjob allows the kernel to speed up the
1896 		 * processing of the filesystem syncer process. A rushjob
1897 		 * value of N tells the filesystem syncer to process the next
1898 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1899 		 * is used by the soft update code to speed up the filesystem
1900 		 * syncer process when the incore state is getting so far
1901 		 * ahead of the disk that the kernel memory pool is being
1902 		 * threatened with exhaustion.
1903 		 */
1904 		if (rushjob > 0) {
1905 			rushjob -= 1;
1906 			continue;
1907 		}
1908 		/*
1909 		 * Just sleep for a short period of time between
1910 		 * iterations when shutting down to allow some I/O
1911 		 * to happen.
1912 		 *
1913 		 * If it has taken us less than a second to process the
1914 		 * current work, then wait. Otherwise start right over
1915 		 * again. We can still lose time if any single round
1916 		 * takes more than two seconds, but it does not really
1917 		 * matter as we are just trying to generally pace the
1918 		 * filesystem activity.
1919 		 */
1920 		if (syncer_state != SYNCER_RUNNING ||
1921 		    time_uptime == starttime) {
1922 			thread_lock(td);
1923 			sched_prio(td, PPAUSE);
1924 			thread_unlock(td);
1925 		}
1926 		if (syncer_state != SYNCER_RUNNING)
1927 			cv_timedwait(&sync_wakeup, &sync_mtx,
1928 			    hz / SYNCER_SHUTDOWN_SPEEDUP);
1929 		else if (time_uptime == starttime)
1930 			cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1931 	}
1932 }
1933 
1934 /*
1935  * Request the syncer daemon to speed up its work.
1936  * We never push it to speed up more than half of its
1937  * normal turn time, otherwise it could take over the cpu.
1938  */
1939 int
1940 speedup_syncer(void)
1941 {
1942 	int ret = 0;
1943 
1944 	mtx_lock(&sync_mtx);
1945 	if (rushjob < syncdelay / 2) {
1946 		rushjob += 1;
1947 		stat_rush_requests += 1;
1948 		ret = 1;
1949 	}
1950 	mtx_unlock(&sync_mtx);
1951 	cv_broadcast(&sync_wakeup);
1952 	return (ret);
1953 }
1954 
1955 /*
1956  * Tell the syncer to speed up its work and run though its work
1957  * list several times, then tell it to shut down.
1958  */
1959 static void
1960 syncer_shutdown(void *arg, int howto)
1961 {
1962 
1963 	if (howto & RB_NOSYNC)
1964 		return;
1965 	mtx_lock(&sync_mtx);
1966 	syncer_state = SYNCER_SHUTTING_DOWN;
1967 	rushjob = 0;
1968 	mtx_unlock(&sync_mtx);
1969 	cv_broadcast(&sync_wakeup);
1970 	kproc_shutdown(arg, howto);
1971 }
1972 
1973 /*
1974  * Reassign a buffer from one vnode to another.
1975  * Used to assign file specific control information
1976  * (indirect blocks) to the vnode to which they belong.
1977  */
1978 void
1979 reassignbuf(struct buf *bp)
1980 {
1981 	struct vnode *vp;
1982 	struct bufobj *bo;
1983 	int delay;
1984 #ifdef INVARIANTS
1985 	struct bufv *bv;
1986 #endif
1987 
1988 	vp = bp->b_vp;
1989 	bo = bp->b_bufobj;
1990 	++reassignbufcalls;
1991 
1992 	CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1993 	    bp, bp->b_vp, bp->b_flags);
1994 	/*
1995 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1996 	 * is not fully linked in.
1997 	 */
1998 	if (bp->b_flags & B_PAGING)
1999 		panic("cannot reassign paging buffer");
2000 
2001 	/*
2002 	 * Delete from old vnode list, if on one.
2003 	 */
2004 	BO_LOCK(bo);
2005 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
2006 		buf_vlist_remove(bp);
2007 	else
2008 		panic("reassignbuf: Buffer %p not on queue.", bp);
2009 	/*
2010 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
2011 	 * of clean buffers.
2012 	 */
2013 	if (bp->b_flags & B_DELWRI) {
2014 		if ((bo->bo_flag & BO_ONWORKLST) == 0) {
2015 			switch (vp->v_type) {
2016 			case VDIR:
2017 				delay = dirdelay;
2018 				break;
2019 			case VCHR:
2020 				delay = metadelay;
2021 				break;
2022 			default:
2023 				delay = filedelay;
2024 			}
2025 			vn_syncer_add_to_worklist(bo, delay);
2026 		}
2027 		buf_vlist_add(bp, bo, BX_VNDIRTY);
2028 	} else {
2029 		buf_vlist_add(bp, bo, BX_VNCLEAN);
2030 
2031 		if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
2032 			mtx_lock(&sync_mtx);
2033 			LIST_REMOVE(bo, bo_synclist);
2034 			syncer_worklist_len--;
2035 			mtx_unlock(&sync_mtx);
2036 			bo->bo_flag &= ~BO_ONWORKLST;
2037 		}
2038 	}
2039 #ifdef INVARIANTS
2040 	bv = &bo->bo_clean;
2041 	bp = TAILQ_FIRST(&bv->bv_hd);
2042 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2043 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2044 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
2045 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2046 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2047 	bv = &bo->bo_dirty;
2048 	bp = TAILQ_FIRST(&bv->bv_hd);
2049 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2050 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2051 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
2052 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2053 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2054 #endif
2055 	BO_UNLOCK(bo);
2056 }
2057 
2058 /*
2059  * Increment the use and hold counts on the vnode, taking care to reference
2060  * the driver's usecount if this is a chardev.  The vholdl() will remove
2061  * the vnode from the free list if it is presently free.  Requires the
2062  * vnode interlock and returns with it held.
2063  */
2064 static void
2065 v_incr_usecount(struct vnode *vp)
2066 {
2067 
2068 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2069 	vp->v_usecount++;
2070 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2071 		dev_lock();
2072 		vp->v_rdev->si_usecount++;
2073 		dev_unlock();
2074 	}
2075 	vholdl(vp);
2076 }
2077 
2078 /*
2079  * Turn a holdcnt into a use+holdcnt such that only one call to
2080  * v_decr_usecount is needed.
2081  */
2082 static void
2083 v_upgrade_usecount(struct vnode *vp)
2084 {
2085 
2086 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2087 	vp->v_usecount++;
2088 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2089 		dev_lock();
2090 		vp->v_rdev->si_usecount++;
2091 		dev_unlock();
2092 	}
2093 }
2094 
2095 /*
2096  * Decrement the vnode use and hold count along with the driver's usecount
2097  * if this is a chardev.  The vdropl() below releases the vnode interlock
2098  * as it may free the vnode.
2099  */
2100 static void
2101 v_decr_usecount(struct vnode *vp)
2102 {
2103 
2104 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2105 	VNASSERT(vp->v_usecount > 0, vp,
2106 	    ("v_decr_usecount: negative usecount"));
2107 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2108 	vp->v_usecount--;
2109 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2110 		dev_lock();
2111 		vp->v_rdev->si_usecount--;
2112 		dev_unlock();
2113 	}
2114 	vdropl(vp);
2115 }
2116 
2117 /*
2118  * Decrement only the use count and driver use count.  This is intended to
2119  * be paired with a follow on vdropl() to release the remaining hold count.
2120  * In this way we may vgone() a vnode with a 0 usecount without risk of
2121  * having it end up on a free list because the hold count is kept above 0.
2122  */
2123 static void
2124 v_decr_useonly(struct vnode *vp)
2125 {
2126 
2127 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2128 	VNASSERT(vp->v_usecount > 0, vp,
2129 	    ("v_decr_useonly: negative usecount"));
2130 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2131 	vp->v_usecount--;
2132 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2133 		dev_lock();
2134 		vp->v_rdev->si_usecount--;
2135 		dev_unlock();
2136 	}
2137 }
2138 
2139 /*
2140  * Grab a particular vnode from the free list, increment its
2141  * reference count and lock it.  VI_DOOMED is set if the vnode
2142  * is being destroyed.  Only callers who specify LK_RETRY will
2143  * see doomed vnodes.  If inactive processing was delayed in
2144  * vput try to do it here.
2145  */
2146 int
2147 vget(struct vnode *vp, int flags, struct thread *td)
2148 {
2149 	int error;
2150 
2151 	error = 0;
2152 	VFS_ASSERT_GIANT(vp->v_mount);
2153 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2154 	    ("vget: invalid lock operation"));
2155 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2156 
2157 	if ((flags & LK_INTERLOCK) == 0)
2158 		VI_LOCK(vp);
2159 	vholdl(vp);
2160 	if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2161 		vdrop(vp);
2162 		CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2163 		    vp);
2164 		return (error);
2165 	}
2166 	if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2167 		panic("vget: vn_lock failed to return ENOENT\n");
2168 	VI_LOCK(vp);
2169 	/* Upgrade our holdcnt to a usecount. */
2170 	v_upgrade_usecount(vp);
2171 	/*
2172 	 * We don't guarantee that any particular close will
2173 	 * trigger inactive processing so just make a best effort
2174 	 * here at preventing a reference to a removed file.  If
2175 	 * we don't succeed no harm is done.
2176 	 */
2177 	if (vp->v_iflag & VI_OWEINACT) {
2178 		if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2179 		    (flags & LK_NOWAIT) == 0)
2180 			vinactive(vp, td);
2181 		vp->v_iflag &= ~VI_OWEINACT;
2182 	}
2183 	VI_UNLOCK(vp);
2184 	return (0);
2185 }
2186 
2187 /*
2188  * Increase the reference count of a vnode.
2189  */
2190 void
2191 vref(struct vnode *vp)
2192 {
2193 
2194 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2195 	VI_LOCK(vp);
2196 	v_incr_usecount(vp);
2197 	VI_UNLOCK(vp);
2198 }
2199 
2200 /*
2201  * Return reference count of a vnode.
2202  *
2203  * The results of this call are only guaranteed when some mechanism other
2204  * than the VI lock is used to stop other processes from gaining references
2205  * to the vnode.  This may be the case if the caller holds the only reference.
2206  * This is also useful when stale data is acceptable as race conditions may
2207  * be accounted for by some other means.
2208  */
2209 int
2210 vrefcnt(struct vnode *vp)
2211 {
2212 	int usecnt;
2213 
2214 	VI_LOCK(vp);
2215 	usecnt = vp->v_usecount;
2216 	VI_UNLOCK(vp);
2217 
2218 	return (usecnt);
2219 }
2220 
2221 #define	VPUTX_VRELE	1
2222 #define	VPUTX_VPUT	2
2223 #define	VPUTX_VUNREF	3
2224 
2225 static void
2226 vputx(struct vnode *vp, int func)
2227 {
2228 	int error;
2229 
2230 	KASSERT(vp != NULL, ("vputx: null vp"));
2231 	if (func == VPUTX_VUNREF)
2232 		ASSERT_VOP_LOCKED(vp, "vunref");
2233 	else if (func == VPUTX_VPUT)
2234 		ASSERT_VOP_LOCKED(vp, "vput");
2235 	else
2236 		KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2237 	VFS_ASSERT_GIANT(vp->v_mount);
2238 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2239 	VI_LOCK(vp);
2240 
2241 	/* Skip this v_writecount check if we're going to panic below. */
2242 	VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2243 	    ("vputx: missed vn_close"));
2244 	error = 0;
2245 
2246 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2247 	    vp->v_usecount == 1)) {
2248 		if (func == VPUTX_VPUT)
2249 			VOP_UNLOCK(vp, 0);
2250 		v_decr_usecount(vp);
2251 		return;
2252 	}
2253 
2254 	if (vp->v_usecount != 1) {
2255 		vprint("vputx: negative ref count", vp);
2256 		panic("vputx: negative ref cnt");
2257 	}
2258 	CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2259 	/*
2260 	 * We want to hold the vnode until the inactive finishes to
2261 	 * prevent vgone() races.  We drop the use count here and the
2262 	 * hold count below when we're done.
2263 	 */
2264 	v_decr_useonly(vp);
2265 	/*
2266 	 * We must call VOP_INACTIVE with the node locked. Mark
2267 	 * as VI_DOINGINACT to avoid recursion.
2268 	 */
2269 	vp->v_iflag |= VI_OWEINACT;
2270 	switch (func) {
2271 	case VPUTX_VRELE:
2272 		error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2273 		VI_LOCK(vp);
2274 		break;
2275 	case VPUTX_VPUT:
2276 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2277 			error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
2278 			    LK_NOWAIT);
2279 			VI_LOCK(vp);
2280 		}
2281 		break;
2282 	case VPUTX_VUNREF:
2283 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
2284 			error = EBUSY;
2285 		break;
2286 	}
2287 	if (vp->v_usecount > 0)
2288 		vp->v_iflag &= ~VI_OWEINACT;
2289 	if (error == 0) {
2290 		if (vp->v_iflag & VI_OWEINACT)
2291 			vinactive(vp, curthread);
2292 		if (func != VPUTX_VUNREF)
2293 			VOP_UNLOCK(vp, 0);
2294 	}
2295 	vdropl(vp);
2296 }
2297 
2298 /*
2299  * Vnode put/release.
2300  * If count drops to zero, call inactive routine and return to freelist.
2301  */
2302 void
2303 vrele(struct vnode *vp)
2304 {
2305 
2306 	vputx(vp, VPUTX_VRELE);
2307 }
2308 
2309 /*
2310  * Release an already locked vnode.  This give the same effects as
2311  * unlock+vrele(), but takes less time and avoids releasing and
2312  * re-aquiring the lock (as vrele() acquires the lock internally.)
2313  */
2314 void
2315 vput(struct vnode *vp)
2316 {
2317 
2318 	vputx(vp, VPUTX_VPUT);
2319 }
2320 
2321 /*
2322  * Release an exclusively locked vnode. Do not unlock the vnode lock.
2323  */
2324 void
2325 vunref(struct vnode *vp)
2326 {
2327 
2328 	vputx(vp, VPUTX_VUNREF);
2329 }
2330 
2331 /*
2332  * Somebody doesn't want the vnode recycled.
2333  */
2334 void
2335 vhold(struct vnode *vp)
2336 {
2337 
2338 	VI_LOCK(vp);
2339 	vholdl(vp);
2340 	VI_UNLOCK(vp);
2341 }
2342 
2343 /*
2344  * Increase the hold count and activate if this is the first reference.
2345  */
2346 void
2347 vholdl(struct vnode *vp)
2348 {
2349 	struct mount *mp;
2350 
2351 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2352 	vp->v_holdcnt++;
2353 	if (!VSHOULDBUSY(vp))
2354 		return;
2355 	ASSERT_VI_LOCKED(vp, "vholdl");
2356 	VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
2357 	VNASSERT(vp->v_op != NULL, vp, ("vholdl: vnode already reclaimed."));
2358 	/*
2359 	 * Remove a vnode from the free list, mark it as in use,
2360 	 * and put it on the active list.
2361 	 */
2362 	mtx_lock(&vnode_free_list_mtx);
2363 	TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
2364 	freevnodes--;
2365 	vp->v_iflag &= ~(VI_FREE|VI_AGE);
2366 	KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
2367 	    ("Activating already active vnode"));
2368 	vp->v_iflag |= VI_ACTIVE;
2369 	mp = vp->v_mount;
2370 	TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
2371 	mp->mnt_activevnodelistsize++;
2372 	mtx_unlock(&vnode_free_list_mtx);
2373 }
2374 
2375 /*
2376  * Note that there is one less who cares about this vnode.
2377  * vdrop() is the opposite of vhold().
2378  */
2379 void
2380 vdrop(struct vnode *vp)
2381 {
2382 
2383 	VI_LOCK(vp);
2384 	vdropl(vp);
2385 }
2386 
2387 /*
2388  * Drop the hold count of the vnode.  If this is the last reference to
2389  * the vnode we place it on the free list unless it has been vgone'd
2390  * (marked VI_DOOMED) in which case we will free it.
2391  */
2392 void
2393 vdropl(struct vnode *vp)
2394 {
2395 	struct bufobj *bo;
2396 	struct mount *mp;
2397 	int active;
2398 
2399 	ASSERT_VI_LOCKED(vp, "vdropl");
2400 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2401 	if (vp->v_holdcnt <= 0)
2402 		panic("vdrop: holdcnt %d", vp->v_holdcnt);
2403 	vp->v_holdcnt--;
2404 	if (vp->v_holdcnt > 0) {
2405 		VI_UNLOCK(vp);
2406 		return;
2407 	}
2408 	if ((vp->v_iflag & VI_DOOMED) == 0) {
2409 		/*
2410 		 * Mark a vnode as free: remove it from its active list
2411 		 * and put it up for recycling on the freelist.
2412 		 */
2413 		VNASSERT(vp->v_op != NULL, vp,
2414 		    ("vdropl: vnode already reclaimed."));
2415 		VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2416 		    ("vnode already free"));
2417 		VNASSERT(VSHOULDFREE(vp), vp,
2418 		    ("vdropl: freeing when we shouldn't"));
2419 		active = vp->v_iflag & VI_ACTIVE;
2420 		vp->v_iflag &= ~VI_ACTIVE;
2421 		mp = vp->v_mount;
2422 		mtx_lock(&vnode_free_list_mtx);
2423 		if (active) {
2424 			TAILQ_REMOVE(&mp->mnt_activevnodelist, vp,
2425 			    v_actfreelist);
2426 			mp->mnt_activevnodelistsize--;
2427 		}
2428 		if (vp->v_iflag & VI_AGE) {
2429 			TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_actfreelist);
2430 		} else {
2431 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
2432 		}
2433 		freevnodes++;
2434 		vp->v_iflag &= ~VI_AGE;
2435 		vp->v_iflag |= VI_FREE;
2436 		mtx_unlock(&vnode_free_list_mtx);
2437 		VI_UNLOCK(vp);
2438 		return;
2439 	}
2440 	/*
2441 	 * The vnode has been marked for destruction, so free it.
2442 	 */
2443 	CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp);
2444 	mtx_lock(&vnode_free_list_mtx);
2445 	numvnodes--;
2446 	mtx_unlock(&vnode_free_list_mtx);
2447 	bo = &vp->v_bufobj;
2448 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2449 	    ("cleaned vnode still on the free list."));
2450 	VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
2451 	VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
2452 	VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
2453 	VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
2454 	VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
2455 	VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
2456 	VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
2457 	VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
2458 	VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
2459 	VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
2460 	VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
2461 	VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
2462 	VI_UNLOCK(vp);
2463 #ifdef MAC
2464 	mac_vnode_destroy(vp);
2465 #endif
2466 	if (vp->v_pollinfo != NULL)
2467 		destroy_vpollinfo(vp->v_pollinfo);
2468 #ifdef INVARIANTS
2469 	/* XXX Elsewhere we detect an already freed vnode via NULL v_op. */
2470 	vp->v_op = NULL;
2471 #endif
2472 	lockdestroy(vp->v_vnlock);
2473 	mtx_destroy(&vp->v_interlock);
2474 	mtx_destroy(BO_MTX(bo));
2475 	uma_zfree(vnode_zone, vp);
2476 }
2477 
2478 /*
2479  * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2480  * flags.  DOINGINACT prevents us from recursing in calls to vinactive.
2481  * OWEINACT tracks whether a vnode missed a call to inactive due to a
2482  * failed lock upgrade.
2483  */
2484 void
2485 vinactive(struct vnode *vp, struct thread *td)
2486 {
2487 	struct vm_object *obj;
2488 
2489 	ASSERT_VOP_ELOCKED(vp, "vinactive");
2490 	ASSERT_VI_LOCKED(vp, "vinactive");
2491 	VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2492 	    ("vinactive: recursed on VI_DOINGINACT"));
2493 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2494 	vp->v_iflag |= VI_DOINGINACT;
2495 	vp->v_iflag &= ~VI_OWEINACT;
2496 	VI_UNLOCK(vp);
2497 	/*
2498 	 * Before moving off the active list, we must be sure that any
2499 	 * modified pages are on the vnode's dirty list since these will
2500 	 * no longer be checked once the vnode is on the inactive list.
2501 	 */
2502 	obj = vp->v_object;
2503 	if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
2504 		VM_OBJECT_LOCK(obj);
2505 		vm_object_page_clean(obj, 0, 0, OBJPC_NOSYNC);
2506 		VM_OBJECT_UNLOCK(obj);
2507 	}
2508 	VOP_INACTIVE(vp, td);
2509 	VI_LOCK(vp);
2510 	VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2511 	    ("vinactive: lost VI_DOINGINACT"));
2512 	vp->v_iflag &= ~VI_DOINGINACT;
2513 }
2514 
2515 /*
2516  * Remove any vnodes in the vnode table belonging to mount point mp.
2517  *
2518  * If FORCECLOSE is not specified, there should not be any active ones,
2519  * return error if any are found (nb: this is a user error, not a
2520  * system error). If FORCECLOSE is specified, detach any active vnodes
2521  * that are found.
2522  *
2523  * If WRITECLOSE is set, only flush out regular file vnodes open for
2524  * writing.
2525  *
2526  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2527  *
2528  * `rootrefs' specifies the base reference count for the root vnode
2529  * of this filesystem. The root vnode is considered busy if its
2530  * v_usecount exceeds this value. On a successful return, vflush(, td)
2531  * will call vrele() on the root vnode exactly rootrefs times.
2532  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2533  * be zero.
2534  */
2535 #ifdef DIAGNOSTIC
2536 static int busyprt = 0;		/* print out busy vnodes */
2537 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
2538 #endif
2539 
2540 int
2541 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
2542 {
2543 	struct vnode *vp, *mvp, *rootvp = NULL;
2544 	struct vattr vattr;
2545 	int busy = 0, error;
2546 
2547 	CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2548 	    rootrefs, flags);
2549 	if (rootrefs > 0) {
2550 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2551 		    ("vflush: bad args"));
2552 		/*
2553 		 * Get the filesystem root vnode. We can vput() it
2554 		 * immediately, since with rootrefs > 0, it won't go away.
2555 		 */
2556 		if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2557 			CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2558 			    __func__, error);
2559 			return (error);
2560 		}
2561 		vput(rootvp);
2562 	}
2563 loop:
2564 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
2565 		vholdl(vp);
2566 		error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2567 		if (error) {
2568 			vdrop(vp);
2569 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2570 			goto loop;
2571 		}
2572 		/*
2573 		 * Skip over a vnodes marked VV_SYSTEM.
2574 		 */
2575 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2576 			VOP_UNLOCK(vp, 0);
2577 			vdrop(vp);
2578 			continue;
2579 		}
2580 		/*
2581 		 * If WRITECLOSE is set, flush out unlinked but still open
2582 		 * files (even if open only for reading) and regular file
2583 		 * vnodes open for writing.
2584 		 */
2585 		if (flags & WRITECLOSE) {
2586 			if (vp->v_object != NULL) {
2587 				VM_OBJECT_LOCK(vp->v_object);
2588 				vm_object_page_clean(vp->v_object, 0, 0, 0);
2589 				VM_OBJECT_UNLOCK(vp->v_object);
2590 			}
2591 			error = VOP_FSYNC(vp, MNT_WAIT, td);
2592 			if (error != 0) {
2593 				VOP_UNLOCK(vp, 0);
2594 				vdrop(vp);
2595 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2596 				return (error);
2597 			}
2598 			error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2599 			VI_LOCK(vp);
2600 
2601 			if ((vp->v_type == VNON ||
2602 			    (error == 0 && vattr.va_nlink > 0)) &&
2603 			    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2604 				VOP_UNLOCK(vp, 0);
2605 				vdropl(vp);
2606 				continue;
2607 			}
2608 		} else
2609 			VI_LOCK(vp);
2610 		/*
2611 		 * With v_usecount == 0, all we need to do is clear out the
2612 		 * vnode data structures and we are done.
2613 		 *
2614 		 * If FORCECLOSE is set, forcibly close the vnode.
2615 		 */
2616 		if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2617 			VNASSERT(vp->v_usecount == 0 ||
2618 			    (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2619 			    ("device VNODE %p is FORCECLOSED", vp));
2620 			vgonel(vp);
2621 		} else {
2622 			busy++;
2623 #ifdef DIAGNOSTIC
2624 			if (busyprt)
2625 				vprint("vflush: busy vnode", vp);
2626 #endif
2627 		}
2628 		VOP_UNLOCK(vp, 0);
2629 		vdropl(vp);
2630 	}
2631 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2632 		/*
2633 		 * If just the root vnode is busy, and if its refcount
2634 		 * is equal to `rootrefs', then go ahead and kill it.
2635 		 */
2636 		VI_LOCK(rootvp);
2637 		KASSERT(busy > 0, ("vflush: not busy"));
2638 		VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2639 		    ("vflush: usecount %d < rootrefs %d",
2640 		     rootvp->v_usecount, rootrefs));
2641 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2642 			VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2643 			vgone(rootvp);
2644 			VOP_UNLOCK(rootvp, 0);
2645 			busy = 0;
2646 		} else
2647 			VI_UNLOCK(rootvp);
2648 	}
2649 	if (busy) {
2650 		CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2651 		    busy);
2652 		return (EBUSY);
2653 	}
2654 	for (; rootrefs > 0; rootrefs--)
2655 		vrele(rootvp);
2656 	return (0);
2657 }
2658 
2659 /*
2660  * Recycle an unused vnode to the front of the free list.
2661  */
2662 int
2663 vrecycle(struct vnode *vp, struct thread *td)
2664 {
2665 	int recycled;
2666 
2667 	ASSERT_VOP_ELOCKED(vp, "vrecycle");
2668 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2669 	recycled = 0;
2670 	VI_LOCK(vp);
2671 	if (vp->v_usecount == 0) {
2672 		recycled = 1;
2673 		vgonel(vp);
2674 	}
2675 	VI_UNLOCK(vp);
2676 	return (recycled);
2677 }
2678 
2679 /*
2680  * Eliminate all activity associated with a vnode
2681  * in preparation for reuse.
2682  */
2683 void
2684 vgone(struct vnode *vp)
2685 {
2686 	VI_LOCK(vp);
2687 	vgonel(vp);
2688 	VI_UNLOCK(vp);
2689 }
2690 
2691 /*
2692  * vgone, with the vp interlock held.
2693  */
2694 void
2695 vgonel(struct vnode *vp)
2696 {
2697 	struct thread *td;
2698 	int oweinact;
2699 	int active;
2700 	struct mount *mp;
2701 
2702 	ASSERT_VOP_ELOCKED(vp, "vgonel");
2703 	ASSERT_VI_LOCKED(vp, "vgonel");
2704 	VNASSERT(vp->v_holdcnt, vp,
2705 	    ("vgonel: vp %p has no reference.", vp));
2706 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2707 	td = curthread;
2708 
2709 	/*
2710 	 * Don't vgonel if we're already doomed.
2711 	 */
2712 	if (vp->v_iflag & VI_DOOMED)
2713 		return;
2714 	vp->v_iflag |= VI_DOOMED;
2715 	/*
2716 	 * Check to see if the vnode is in use.  If so, we have to call
2717 	 * VOP_CLOSE() and VOP_INACTIVE().
2718 	 */
2719 	active = vp->v_usecount;
2720 	oweinact = (vp->v_iflag & VI_OWEINACT);
2721 	VI_UNLOCK(vp);
2722 	/*
2723 	 * Clean out any buffers associated with the vnode.
2724 	 * If the flush fails, just toss the buffers.
2725 	 */
2726 	mp = NULL;
2727 	if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2728 		(void) vn_start_secondary_write(vp, &mp, V_WAIT);
2729 	if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
2730 		vinvalbuf(vp, 0, 0, 0);
2731 
2732 	/*
2733 	 * If purging an active vnode, it must be closed and
2734 	 * deactivated before being reclaimed.
2735 	 */
2736 	if (active)
2737 		VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2738 	if (oweinact || active) {
2739 		VI_LOCK(vp);
2740 		if ((vp->v_iflag & VI_DOINGINACT) == 0)
2741 			vinactive(vp, td);
2742 		VI_UNLOCK(vp);
2743 	}
2744 	if (vp->v_type == VSOCK)
2745 		vfs_unp_reclaim(vp);
2746 	/*
2747 	 * Reclaim the vnode.
2748 	 */
2749 	if (VOP_RECLAIM(vp, td))
2750 		panic("vgone: cannot reclaim");
2751 	if (mp != NULL)
2752 		vn_finished_secondary_write(mp);
2753 	VNASSERT(vp->v_object == NULL, vp,
2754 	    ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2755 	/*
2756 	 * Clear the advisory locks and wake up waiting threads.
2757 	 */
2758 	(void)VOP_ADVLOCKPURGE(vp);
2759 	/*
2760 	 * Delete from old mount point vnode list.
2761 	 */
2762 	delmntque(vp);
2763 	cache_purge(vp);
2764 	/*
2765 	 * Done with purge, reset to the standard lock and invalidate
2766 	 * the vnode.
2767 	 */
2768 	VI_LOCK(vp);
2769 	vp->v_vnlock = &vp->v_lock;
2770 	vp->v_op = &dead_vnodeops;
2771 	vp->v_tag = "none";
2772 	vp->v_type = VBAD;
2773 }
2774 
2775 /*
2776  * Calculate the total number of references to a special device.
2777  */
2778 int
2779 vcount(struct vnode *vp)
2780 {
2781 	int count;
2782 
2783 	dev_lock();
2784 	count = vp->v_rdev->si_usecount;
2785 	dev_unlock();
2786 	return (count);
2787 }
2788 
2789 /*
2790  * Same as above, but using the struct cdev *as argument
2791  */
2792 int
2793 count_dev(struct cdev *dev)
2794 {
2795 	int count;
2796 
2797 	dev_lock();
2798 	count = dev->si_usecount;
2799 	dev_unlock();
2800 	return(count);
2801 }
2802 
2803 /*
2804  * Print out a description of a vnode.
2805  */
2806 static char *typename[] =
2807 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2808  "VMARKER"};
2809 
2810 void
2811 vn_printf(struct vnode *vp, const char *fmt, ...)
2812 {
2813 	va_list ap;
2814 	char buf[256], buf2[16];
2815 	u_long flags;
2816 
2817 	va_start(ap, fmt);
2818 	vprintf(fmt, ap);
2819 	va_end(ap);
2820 	printf("%p: ", (void *)vp);
2821 	printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2822 	printf("    usecount %d, writecount %d, refcount %d mountedhere %p\n",
2823 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2824 	buf[0] = '\0';
2825 	buf[1] = '\0';
2826 	if (vp->v_vflag & VV_ROOT)
2827 		strlcat(buf, "|VV_ROOT", sizeof(buf));
2828 	if (vp->v_vflag & VV_ISTTY)
2829 		strlcat(buf, "|VV_ISTTY", sizeof(buf));
2830 	if (vp->v_vflag & VV_NOSYNC)
2831 		strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2832 	if (vp->v_vflag & VV_CACHEDLABEL)
2833 		strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2834 	if (vp->v_vflag & VV_TEXT)
2835 		strlcat(buf, "|VV_TEXT", sizeof(buf));
2836 	if (vp->v_vflag & VV_COPYONWRITE)
2837 		strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2838 	if (vp->v_vflag & VV_SYSTEM)
2839 		strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2840 	if (vp->v_vflag & VV_PROCDEP)
2841 		strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2842 	if (vp->v_vflag & VV_NOKNOTE)
2843 		strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2844 	if (vp->v_vflag & VV_DELETED)
2845 		strlcat(buf, "|VV_DELETED", sizeof(buf));
2846 	if (vp->v_vflag & VV_MD)
2847 		strlcat(buf, "|VV_MD", sizeof(buf));
2848 	flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC |
2849 	    VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2850 	    VV_NOKNOTE | VV_DELETED | VV_MD);
2851 	if (flags != 0) {
2852 		snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2853 		strlcat(buf, buf2, sizeof(buf));
2854 	}
2855 	if (vp->v_iflag & VI_MOUNT)
2856 		strlcat(buf, "|VI_MOUNT", sizeof(buf));
2857 	if (vp->v_iflag & VI_AGE)
2858 		strlcat(buf, "|VI_AGE", sizeof(buf));
2859 	if (vp->v_iflag & VI_DOOMED)
2860 		strlcat(buf, "|VI_DOOMED", sizeof(buf));
2861 	if (vp->v_iflag & VI_FREE)
2862 		strlcat(buf, "|VI_FREE", sizeof(buf));
2863 	if (vp->v_iflag & VI_DOINGINACT)
2864 		strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2865 	if (vp->v_iflag & VI_OWEINACT)
2866 		strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2867 	flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2868 	    VI_DOINGINACT | VI_OWEINACT);
2869 	if (flags != 0) {
2870 		snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2871 		strlcat(buf, buf2, sizeof(buf));
2872 	}
2873 	printf("    flags (%s)\n", buf + 1);
2874 	if (mtx_owned(VI_MTX(vp)))
2875 		printf(" VI_LOCKed");
2876 	if (vp->v_object != NULL)
2877 		printf("    v_object %p ref %d pages %d\n",
2878 		    vp->v_object, vp->v_object->ref_count,
2879 		    vp->v_object->resident_page_count);
2880 	printf("    ");
2881 	lockmgr_printinfo(vp->v_vnlock);
2882 	if (vp->v_data != NULL)
2883 		VOP_PRINT(vp);
2884 }
2885 
2886 #ifdef DDB
2887 /*
2888  * List all of the locked vnodes in the system.
2889  * Called when debugging the kernel.
2890  */
2891 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2892 {
2893 	struct mount *mp, *nmp;
2894 	struct vnode *vp;
2895 
2896 	/*
2897 	 * Note: because this is DDB, we can't obey the locking semantics
2898 	 * for these structures, which means we could catch an inconsistent
2899 	 * state and dereference a nasty pointer.  Not much to be done
2900 	 * about that.
2901 	 */
2902 	db_printf("Locked vnodes\n");
2903 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2904 		nmp = TAILQ_NEXT(mp, mnt_list);
2905 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2906 			if (vp->v_type != VMARKER &&
2907 			    VOP_ISLOCKED(vp))
2908 				vprint("", vp);
2909 		}
2910 		nmp = TAILQ_NEXT(mp, mnt_list);
2911 	}
2912 }
2913 
2914 /*
2915  * Show details about the given vnode.
2916  */
2917 DB_SHOW_COMMAND(vnode, db_show_vnode)
2918 {
2919 	struct vnode *vp;
2920 
2921 	if (!have_addr)
2922 		return;
2923 	vp = (struct vnode *)addr;
2924 	vn_printf(vp, "vnode ");
2925 }
2926 
2927 /*
2928  * Show details about the given mount point.
2929  */
2930 DB_SHOW_COMMAND(mount, db_show_mount)
2931 {
2932 	struct mount *mp;
2933 	struct vfsopt *opt;
2934 	struct statfs *sp;
2935 	struct vnode *vp;
2936 	char buf[512];
2937 	uint64_t mflags;
2938 	u_int flags;
2939 
2940 	if (!have_addr) {
2941 		/* No address given, print short info about all mount points. */
2942 		TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2943 			db_printf("%p %s on %s (%s)\n", mp,
2944 			    mp->mnt_stat.f_mntfromname,
2945 			    mp->mnt_stat.f_mntonname,
2946 			    mp->mnt_stat.f_fstypename);
2947 			if (db_pager_quit)
2948 				break;
2949 		}
2950 		db_printf("\nMore info: show mount <addr>\n");
2951 		return;
2952 	}
2953 
2954 	mp = (struct mount *)addr;
2955 	db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
2956 	    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
2957 
2958 	buf[0] = '\0';
2959 	mflags = mp->mnt_flag;
2960 #define	MNT_FLAG(flag)	do {						\
2961 	if (mflags & (flag)) {						\
2962 		if (buf[0] != '\0')					\
2963 			strlcat(buf, ", ", sizeof(buf));		\
2964 		strlcat(buf, (#flag) + 4, sizeof(buf));			\
2965 		mflags &= ~(flag);					\
2966 	}								\
2967 } while (0)
2968 	MNT_FLAG(MNT_RDONLY);
2969 	MNT_FLAG(MNT_SYNCHRONOUS);
2970 	MNT_FLAG(MNT_NOEXEC);
2971 	MNT_FLAG(MNT_NOSUID);
2972 	MNT_FLAG(MNT_UNION);
2973 	MNT_FLAG(MNT_ASYNC);
2974 	MNT_FLAG(MNT_SUIDDIR);
2975 	MNT_FLAG(MNT_SOFTDEP);
2976 	MNT_FLAG(MNT_SUJ);
2977 	MNT_FLAG(MNT_NOSYMFOLLOW);
2978 	MNT_FLAG(MNT_GJOURNAL);
2979 	MNT_FLAG(MNT_MULTILABEL);
2980 	MNT_FLAG(MNT_ACLS);
2981 	MNT_FLAG(MNT_NOATIME);
2982 	MNT_FLAG(MNT_NOCLUSTERR);
2983 	MNT_FLAG(MNT_NOCLUSTERW);
2984 	MNT_FLAG(MNT_NFS4ACLS);
2985 	MNT_FLAG(MNT_EXRDONLY);
2986 	MNT_FLAG(MNT_EXPORTED);
2987 	MNT_FLAG(MNT_DEFEXPORTED);
2988 	MNT_FLAG(MNT_EXPORTANON);
2989 	MNT_FLAG(MNT_EXKERB);
2990 	MNT_FLAG(MNT_EXPUBLIC);
2991 	MNT_FLAG(MNT_LOCAL);
2992 	MNT_FLAG(MNT_QUOTA);
2993 	MNT_FLAG(MNT_ROOTFS);
2994 	MNT_FLAG(MNT_USER);
2995 	MNT_FLAG(MNT_IGNORE);
2996 	MNT_FLAG(MNT_UPDATE);
2997 	MNT_FLAG(MNT_DELEXPORT);
2998 	MNT_FLAG(MNT_RELOAD);
2999 	MNT_FLAG(MNT_FORCE);
3000 	MNT_FLAG(MNT_SNAPSHOT);
3001 	MNT_FLAG(MNT_BYFSID);
3002 #undef MNT_FLAG
3003 	if (mflags != 0) {
3004 		if (buf[0] != '\0')
3005 			strlcat(buf, ", ", sizeof(buf));
3006 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3007 		    "0x%016jx", mflags);
3008 	}
3009 	db_printf("    mnt_flag = %s\n", buf);
3010 
3011 	buf[0] = '\0';
3012 	flags = mp->mnt_kern_flag;
3013 #define	MNT_KERN_FLAG(flag)	do {					\
3014 	if (flags & (flag)) {						\
3015 		if (buf[0] != '\0')					\
3016 			strlcat(buf, ", ", sizeof(buf));		\
3017 		strlcat(buf, (#flag) + 5, sizeof(buf));			\
3018 		flags &= ~(flag);					\
3019 	}								\
3020 } while (0)
3021 	MNT_KERN_FLAG(MNTK_UNMOUNTF);
3022 	MNT_KERN_FLAG(MNTK_ASYNC);
3023 	MNT_KERN_FLAG(MNTK_SOFTDEP);
3024 	MNT_KERN_FLAG(MNTK_NOINSMNTQ);
3025 	MNT_KERN_FLAG(MNTK_DRAINING);
3026 	MNT_KERN_FLAG(MNTK_REFEXPIRE);
3027 	MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
3028 	MNT_KERN_FLAG(MNTK_SHARED_WRITES);
3029 	MNT_KERN_FLAG(MNTK_NOASYNC);
3030 	MNT_KERN_FLAG(MNTK_UNMOUNT);
3031 	MNT_KERN_FLAG(MNTK_MWAIT);
3032 	MNT_KERN_FLAG(MNTK_SUSPEND);
3033 	MNT_KERN_FLAG(MNTK_SUSPEND2);
3034 	MNT_KERN_FLAG(MNTK_SUSPENDED);
3035 	MNT_KERN_FLAG(MNTK_MPSAFE);
3036 	MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
3037 	MNT_KERN_FLAG(MNTK_NOKNOTE);
3038 #undef MNT_KERN_FLAG
3039 	if (flags != 0) {
3040 		if (buf[0] != '\0')
3041 			strlcat(buf, ", ", sizeof(buf));
3042 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3043 		    "0x%08x", flags);
3044 	}
3045 	db_printf("    mnt_kern_flag = %s\n", buf);
3046 
3047 	db_printf("    mnt_opt = ");
3048 	opt = TAILQ_FIRST(mp->mnt_opt);
3049 	if (opt != NULL) {
3050 		db_printf("%s", opt->name);
3051 		opt = TAILQ_NEXT(opt, link);
3052 		while (opt != NULL) {
3053 			db_printf(", %s", opt->name);
3054 			opt = TAILQ_NEXT(opt, link);
3055 		}
3056 	}
3057 	db_printf("\n");
3058 
3059 	sp = &mp->mnt_stat;
3060 	db_printf("    mnt_stat = { version=%u type=%u flags=0x%016jx "
3061 	    "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
3062 	    "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
3063 	    "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
3064 	    (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
3065 	    (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
3066 	    (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
3067 	    (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
3068 	    (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
3069 	    (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
3070 	    (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
3071 	    (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
3072 
3073 	db_printf("    mnt_cred = { uid=%u ruid=%u",
3074 	    (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
3075 	if (jailed(mp->mnt_cred))
3076 		db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
3077 	db_printf(" }\n");
3078 	db_printf("    mnt_ref = %d\n", mp->mnt_ref);
3079 	db_printf("    mnt_gen = %d\n", mp->mnt_gen);
3080 	db_printf("    mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
3081 	db_printf("    mnt_activevnodelistsize = %d\n",
3082 	    mp->mnt_activevnodelistsize);
3083 	db_printf("    mnt_writeopcount = %d\n", mp->mnt_writeopcount);
3084 	db_printf("    mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
3085 	db_printf("    mnt_iosize_max = %d\n", mp->mnt_iosize_max);
3086 	db_printf("    mnt_hashseed = %u\n", mp->mnt_hashseed);
3087 	db_printf("    mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
3088 	db_printf("    mnt_secondary_accwrites = %d\n",
3089 	    mp->mnt_secondary_accwrites);
3090 	db_printf("    mnt_gjprovider = %s\n",
3091 	    mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
3092 
3093 	db_printf("\n\nList of active vnodes\n");
3094 	TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) {
3095 		if (vp->v_type != VMARKER) {
3096 			vn_printf(vp, "vnode ");
3097 			if (db_pager_quit)
3098 				break;
3099 		}
3100 	}
3101 	db_printf("\n\nList of inactive vnodes\n");
3102 	TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3103 		if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) {
3104 			vn_printf(vp, "vnode ");
3105 			if (db_pager_quit)
3106 				break;
3107 		}
3108 	}
3109 }
3110 #endif	/* DDB */
3111 
3112 /*
3113  * Fill in a struct xvfsconf based on a struct vfsconf.
3114  */
3115 static void
3116 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
3117 {
3118 
3119 	strcpy(xvfsp->vfc_name, vfsp->vfc_name);
3120 	xvfsp->vfc_typenum = vfsp->vfc_typenum;
3121 	xvfsp->vfc_refcount = vfsp->vfc_refcount;
3122 	xvfsp->vfc_flags = vfsp->vfc_flags;
3123 	/*
3124 	 * These are unused in userland, we keep them
3125 	 * to not break binary compatibility.
3126 	 */
3127 	xvfsp->vfc_vfsops = NULL;
3128 	xvfsp->vfc_next = NULL;
3129 }
3130 
3131 /*
3132  * Top level filesystem related information gathering.
3133  */
3134 static int
3135 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
3136 {
3137 	struct vfsconf *vfsp;
3138 	struct xvfsconf xvfsp;
3139 	int error;
3140 
3141 	error = 0;
3142 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3143 		bzero(&xvfsp, sizeof(xvfsp));
3144 		vfsconf2x(vfsp, &xvfsp);
3145 		error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
3146 		if (error)
3147 			break;
3148 	}
3149 	return (error);
3150 }
3151 
3152 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD,
3153     NULL, 0, sysctl_vfs_conflist,
3154     "S,xvfsconf", "List of all configured filesystems");
3155 
3156 #ifndef BURN_BRIDGES
3157 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
3158 
3159 static int
3160 vfs_sysctl(SYSCTL_HANDLER_ARGS)
3161 {
3162 	int *name = (int *)arg1 - 1;	/* XXX */
3163 	u_int namelen = arg2 + 1;	/* XXX */
3164 	struct vfsconf *vfsp;
3165 	struct xvfsconf xvfsp;
3166 
3167 	log(LOG_WARNING, "userland calling deprecated sysctl, "
3168 	    "please rebuild world\n");
3169 
3170 #if 1 || defined(COMPAT_PRELITE2)
3171 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3172 	if (namelen == 1)
3173 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3174 #endif
3175 
3176 	switch (name[1]) {
3177 	case VFS_MAXTYPENUM:
3178 		if (namelen != 2)
3179 			return (ENOTDIR);
3180 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3181 	case VFS_CONF:
3182 		if (namelen != 3)
3183 			return (ENOTDIR);	/* overloaded */
3184 		TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
3185 			if (vfsp->vfc_typenum == name[2])
3186 				break;
3187 		if (vfsp == NULL)
3188 			return (EOPNOTSUPP);
3189 		bzero(&xvfsp, sizeof(xvfsp));
3190 		vfsconf2x(vfsp, &xvfsp);
3191 		return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3192 	}
3193 	return (EOPNOTSUPP);
3194 }
3195 
3196 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
3197     vfs_sysctl, "Generic filesystem");
3198 
3199 #if 1 || defined(COMPAT_PRELITE2)
3200 
3201 static int
3202 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3203 {
3204 	int error;
3205 	struct vfsconf *vfsp;
3206 	struct ovfsconf ovfs;
3207 
3208 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3209 		bzero(&ovfs, sizeof(ovfs));
3210 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
3211 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
3212 		ovfs.vfc_index = vfsp->vfc_typenum;
3213 		ovfs.vfc_refcount = vfsp->vfc_refcount;
3214 		ovfs.vfc_flags = vfsp->vfc_flags;
3215 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3216 		if (error)
3217 			return error;
3218 	}
3219 	return 0;
3220 }
3221 
3222 #endif /* 1 || COMPAT_PRELITE2 */
3223 #endif /* !BURN_BRIDGES */
3224 
3225 #define KINFO_VNODESLOP		10
3226 #ifdef notyet
3227 /*
3228  * Dump vnode list (via sysctl).
3229  */
3230 /* ARGSUSED */
3231 static int
3232 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3233 {
3234 	struct xvnode *xvn;
3235 	struct mount *mp;
3236 	struct vnode *vp;
3237 	int error, len, n;
3238 
3239 	/*
3240 	 * Stale numvnodes access is not fatal here.
3241 	 */
3242 	req->lock = 0;
3243 	len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3244 	if (!req->oldptr)
3245 		/* Make an estimate */
3246 		return (SYSCTL_OUT(req, 0, len));
3247 
3248 	error = sysctl_wire_old_buffer(req, 0);
3249 	if (error != 0)
3250 		return (error);
3251 	xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3252 	n = 0;
3253 	mtx_lock(&mountlist_mtx);
3254 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3255 		if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3256 			continue;
3257 		MNT_ILOCK(mp);
3258 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3259 			if (n == len)
3260 				break;
3261 			vref(vp);
3262 			xvn[n].xv_size = sizeof *xvn;
3263 			xvn[n].xv_vnode = vp;
3264 			xvn[n].xv_id = 0;	/* XXX compat */
3265 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3266 			XV_COPY(usecount);
3267 			XV_COPY(writecount);
3268 			XV_COPY(holdcnt);
3269 			XV_COPY(mount);
3270 			XV_COPY(numoutput);
3271 			XV_COPY(type);
3272 #undef XV_COPY
3273 			xvn[n].xv_flag = vp->v_vflag;
3274 
3275 			switch (vp->v_type) {
3276 			case VREG:
3277 			case VDIR:
3278 			case VLNK:
3279 				break;
3280 			case VBLK:
3281 			case VCHR:
3282 				if (vp->v_rdev == NULL) {
3283 					vrele(vp);
3284 					continue;
3285 				}
3286 				xvn[n].xv_dev = dev2udev(vp->v_rdev);
3287 				break;
3288 			case VSOCK:
3289 				xvn[n].xv_socket = vp->v_socket;
3290 				break;
3291 			case VFIFO:
3292 				xvn[n].xv_fifo = vp->v_fifoinfo;
3293 				break;
3294 			case VNON:
3295 			case VBAD:
3296 			default:
3297 				/* shouldn't happen? */
3298 				vrele(vp);
3299 				continue;
3300 			}
3301 			vrele(vp);
3302 			++n;
3303 		}
3304 		MNT_IUNLOCK(mp);
3305 		mtx_lock(&mountlist_mtx);
3306 		vfs_unbusy(mp);
3307 		if (n == len)
3308 			break;
3309 	}
3310 	mtx_unlock(&mountlist_mtx);
3311 
3312 	error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3313 	free(xvn, M_TEMP);
3314 	return (error);
3315 }
3316 
3317 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3318     0, 0, sysctl_vnode, "S,xvnode", "");
3319 #endif
3320 
3321 /*
3322  * Unmount all filesystems. The list is traversed in reverse order
3323  * of mounting to avoid dependencies.
3324  */
3325 void
3326 vfs_unmountall(void)
3327 {
3328 	struct mount *mp;
3329 	struct thread *td;
3330 	int error;
3331 
3332 	KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
3333 	CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3334 	td = curthread;
3335 
3336 	/*
3337 	 * Since this only runs when rebooting, it is not interlocked.
3338 	 */
3339 	while(!TAILQ_EMPTY(&mountlist)) {
3340 		mp = TAILQ_LAST(&mountlist, mntlist);
3341 		error = dounmount(mp, MNT_FORCE, td);
3342 		if (error) {
3343 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
3344 			/*
3345 			 * XXX: Due to the way in which we mount the root
3346 			 * file system off of devfs, devfs will generate a
3347 			 * "busy" warning when we try to unmount it before
3348 			 * the root.  Don't print a warning as a result in
3349 			 * order to avoid false positive errors that may
3350 			 * cause needless upset.
3351 			 */
3352 			if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3353 				printf("unmount of %s failed (",
3354 				    mp->mnt_stat.f_mntonname);
3355 				if (error == EBUSY)
3356 					printf("BUSY)\n");
3357 				else
3358 					printf("%d)\n", error);
3359 			}
3360 		} else {
3361 			/* The unmount has removed mp from the mountlist */
3362 		}
3363 	}
3364 }
3365 
3366 /*
3367  * perform msync on all vnodes under a mount point
3368  * the mount point must be locked.
3369  */
3370 void
3371 vfs_msync(struct mount *mp, int flags)
3372 {
3373 	struct vnode *vp, *mvp;
3374 	struct vm_object *obj;
3375 
3376 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3377 	MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
3378 		obj = vp->v_object;
3379 		if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3380 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3381 			if (!vget(vp,
3382 			    LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3383 			    curthread)) {
3384 				if (vp->v_vflag & VV_NOSYNC) {	/* unlinked */
3385 					vput(vp);
3386 					continue;
3387 				}
3388 
3389 				obj = vp->v_object;
3390 				if (obj != NULL) {
3391 					VM_OBJECT_LOCK(obj);
3392 					vm_object_page_clean(obj, 0, 0,
3393 					    flags == MNT_WAIT ?
3394 					    OBJPC_SYNC : OBJPC_NOSYNC);
3395 					VM_OBJECT_UNLOCK(obj);
3396 				}
3397 				vput(vp);
3398 			}
3399 		} else
3400 			VI_UNLOCK(vp);
3401 	}
3402 }
3403 
3404 static void
3405 destroy_vpollinfo(struct vpollinfo *vi)
3406 {
3407 	seldrain(&vi->vpi_selinfo);
3408 	knlist_destroy(&vi->vpi_selinfo.si_note);
3409 	mtx_destroy(&vi->vpi_lock);
3410 	uma_zfree(vnodepoll_zone, vi);
3411 }
3412 
3413 /*
3414  * Initalize per-vnode helper structure to hold poll-related state.
3415  */
3416 void
3417 v_addpollinfo(struct vnode *vp)
3418 {
3419 	struct vpollinfo *vi;
3420 
3421 	if (vp->v_pollinfo != NULL)
3422 		return;
3423 	vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3424 	mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3425 	knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3426 	    vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3427 	VI_LOCK(vp);
3428 	if (vp->v_pollinfo != NULL) {
3429 		VI_UNLOCK(vp);
3430 		destroy_vpollinfo(vi);
3431 		return;
3432 	}
3433 	vp->v_pollinfo = vi;
3434 	VI_UNLOCK(vp);
3435 }
3436 
3437 /*
3438  * Record a process's interest in events which might happen to
3439  * a vnode.  Because poll uses the historic select-style interface
3440  * internally, this routine serves as both the ``check for any
3441  * pending events'' and the ``record my interest in future events''
3442  * functions.  (These are done together, while the lock is held,
3443  * to avoid race conditions.)
3444  */
3445 int
3446 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3447 {
3448 
3449 	v_addpollinfo(vp);
3450 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3451 	if (vp->v_pollinfo->vpi_revents & events) {
3452 		/*
3453 		 * This leaves events we are not interested
3454 		 * in available for the other process which
3455 		 * which presumably had requested them
3456 		 * (otherwise they would never have been
3457 		 * recorded).
3458 		 */
3459 		events &= vp->v_pollinfo->vpi_revents;
3460 		vp->v_pollinfo->vpi_revents &= ~events;
3461 
3462 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
3463 		return (events);
3464 	}
3465 	vp->v_pollinfo->vpi_events |= events;
3466 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3467 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3468 	return (0);
3469 }
3470 
3471 /*
3472  * Routine to create and manage a filesystem syncer vnode.
3473  */
3474 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
3475 static int	sync_fsync(struct  vop_fsync_args *);
3476 static int	sync_inactive(struct  vop_inactive_args *);
3477 static int	sync_reclaim(struct  vop_reclaim_args *);
3478 
3479 static struct vop_vector sync_vnodeops = {
3480 	.vop_bypass =	VOP_EOPNOTSUPP,
3481 	.vop_close =	sync_close,		/* close */
3482 	.vop_fsync =	sync_fsync,		/* fsync */
3483 	.vop_inactive =	sync_inactive,	/* inactive */
3484 	.vop_reclaim =	sync_reclaim,	/* reclaim */
3485 	.vop_lock1 =	vop_stdlock,	/* lock */
3486 	.vop_unlock =	vop_stdunlock,	/* unlock */
3487 	.vop_islocked =	vop_stdislocked,	/* islocked */
3488 };
3489 
3490 /*
3491  * Create a new filesystem syncer vnode for the specified mount point.
3492  */
3493 void
3494 vfs_allocate_syncvnode(struct mount *mp)
3495 {
3496 	struct vnode *vp;
3497 	struct bufobj *bo;
3498 	static long start, incr, next;
3499 	int error;
3500 
3501 	/* Allocate a new vnode */
3502 	error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
3503 	if (error != 0)
3504 		panic("vfs_allocate_syncvnode: getnewvnode() failed");
3505 	vp->v_type = VNON;
3506 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3507 	vp->v_vflag |= VV_FORCEINSMQ;
3508 	error = insmntque(vp, mp);
3509 	if (error != 0)
3510 		panic("vfs_allocate_syncvnode: insmntque() failed");
3511 	vp->v_vflag &= ~VV_FORCEINSMQ;
3512 	VOP_UNLOCK(vp, 0);
3513 	/*
3514 	 * Place the vnode onto the syncer worklist. We attempt to
3515 	 * scatter them about on the list so that they will go off
3516 	 * at evenly distributed times even if all the filesystems
3517 	 * are mounted at once.
3518 	 */
3519 	next += incr;
3520 	if (next == 0 || next > syncer_maxdelay) {
3521 		start /= 2;
3522 		incr /= 2;
3523 		if (start == 0) {
3524 			start = syncer_maxdelay / 2;
3525 			incr = syncer_maxdelay;
3526 		}
3527 		next = start;
3528 	}
3529 	bo = &vp->v_bufobj;
3530 	BO_LOCK(bo);
3531 	vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3532 	/* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3533 	mtx_lock(&sync_mtx);
3534 	sync_vnode_count++;
3535 	if (mp->mnt_syncer == NULL) {
3536 		mp->mnt_syncer = vp;
3537 		vp = NULL;
3538 	}
3539 	mtx_unlock(&sync_mtx);
3540 	BO_UNLOCK(bo);
3541 	if (vp != NULL) {
3542 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3543 		vgone(vp);
3544 		vput(vp);
3545 	}
3546 }
3547 
3548 void
3549 vfs_deallocate_syncvnode(struct mount *mp)
3550 {
3551 	struct vnode *vp;
3552 
3553 	mtx_lock(&sync_mtx);
3554 	vp = mp->mnt_syncer;
3555 	if (vp != NULL)
3556 		mp->mnt_syncer = NULL;
3557 	mtx_unlock(&sync_mtx);
3558 	if (vp != NULL)
3559 		vrele(vp);
3560 }
3561 
3562 /*
3563  * Do a lazy sync of the filesystem.
3564  */
3565 static int
3566 sync_fsync(struct vop_fsync_args *ap)
3567 {
3568 	struct vnode *syncvp = ap->a_vp;
3569 	struct mount *mp = syncvp->v_mount;
3570 	int error, save;
3571 	struct bufobj *bo;
3572 
3573 	/*
3574 	 * We only need to do something if this is a lazy evaluation.
3575 	 */
3576 	if (ap->a_waitfor != MNT_LAZY)
3577 		return (0);
3578 
3579 	/*
3580 	 * Move ourselves to the back of the sync list.
3581 	 */
3582 	bo = &syncvp->v_bufobj;
3583 	BO_LOCK(bo);
3584 	vn_syncer_add_to_worklist(bo, syncdelay);
3585 	BO_UNLOCK(bo);
3586 
3587 	/*
3588 	 * Walk the list of vnodes pushing all that are dirty and
3589 	 * not already on the sync list.
3590 	 */
3591 	mtx_lock(&mountlist_mtx);
3592 	if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) {
3593 		mtx_unlock(&mountlist_mtx);
3594 		return (0);
3595 	}
3596 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3597 		vfs_unbusy(mp);
3598 		return (0);
3599 	}
3600 	save = curthread_pflags_set(TDP_SYNCIO);
3601 	vfs_msync(mp, MNT_NOWAIT);
3602 	error = VFS_SYNC(mp, MNT_LAZY);
3603 	curthread_pflags_restore(save);
3604 	vn_finished_write(mp);
3605 	vfs_unbusy(mp);
3606 	return (error);
3607 }
3608 
3609 /*
3610  * The syncer vnode is no referenced.
3611  */
3612 static int
3613 sync_inactive(struct vop_inactive_args *ap)
3614 {
3615 
3616 	vgone(ap->a_vp);
3617 	return (0);
3618 }
3619 
3620 /*
3621  * The syncer vnode is no longer needed and is being decommissioned.
3622  *
3623  * Modifications to the worklist must be protected by sync_mtx.
3624  */
3625 static int
3626 sync_reclaim(struct vop_reclaim_args *ap)
3627 {
3628 	struct vnode *vp = ap->a_vp;
3629 	struct bufobj *bo;
3630 
3631 	bo = &vp->v_bufobj;
3632 	BO_LOCK(bo);
3633 	mtx_lock(&sync_mtx);
3634 	if (vp->v_mount->mnt_syncer == vp)
3635 		vp->v_mount->mnt_syncer = NULL;
3636 	if (bo->bo_flag & BO_ONWORKLST) {
3637 		LIST_REMOVE(bo, bo_synclist);
3638 		syncer_worklist_len--;
3639 		sync_vnode_count--;
3640 		bo->bo_flag &= ~BO_ONWORKLST;
3641 	}
3642 	mtx_unlock(&sync_mtx);
3643 	BO_UNLOCK(bo);
3644 
3645 	return (0);
3646 }
3647 
3648 /*
3649  * Check if vnode represents a disk device
3650  */
3651 int
3652 vn_isdisk(struct vnode *vp, int *errp)
3653 {
3654 	int error;
3655 
3656 	error = 0;
3657 	dev_lock();
3658 	if (vp->v_type != VCHR)
3659 		error = ENOTBLK;
3660 	else if (vp->v_rdev == NULL)
3661 		error = ENXIO;
3662 	else if (vp->v_rdev->si_devsw == NULL)
3663 		error = ENXIO;
3664 	else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3665 		error = ENOTBLK;
3666 	dev_unlock();
3667 	if (errp != NULL)
3668 		*errp = error;
3669 	return (error == 0);
3670 }
3671 
3672 /*
3673  * Common filesystem object access control check routine.  Accepts a
3674  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3675  * and optional call-by-reference privused argument allowing vaccess()
3676  * to indicate to the caller whether privilege was used to satisfy the
3677  * request (obsoleted).  Returns 0 on success, or an errno on failure.
3678  */
3679 int
3680 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3681     accmode_t accmode, struct ucred *cred, int *privused)
3682 {
3683 	accmode_t dac_granted;
3684 	accmode_t priv_granted;
3685 
3686 	KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3687 	    ("invalid bit in accmode"));
3688 	KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
3689 	    ("VAPPEND without VWRITE"));
3690 
3691 	/*
3692 	 * Look for a normal, non-privileged way to access the file/directory
3693 	 * as requested.  If it exists, go with that.
3694 	 */
3695 
3696 	if (privused != NULL)
3697 		*privused = 0;
3698 
3699 	dac_granted = 0;
3700 
3701 	/* Check the owner. */
3702 	if (cred->cr_uid == file_uid) {
3703 		dac_granted |= VADMIN;
3704 		if (file_mode & S_IXUSR)
3705 			dac_granted |= VEXEC;
3706 		if (file_mode & S_IRUSR)
3707 			dac_granted |= VREAD;
3708 		if (file_mode & S_IWUSR)
3709 			dac_granted |= (VWRITE | VAPPEND);
3710 
3711 		if ((accmode & dac_granted) == accmode)
3712 			return (0);
3713 
3714 		goto privcheck;
3715 	}
3716 
3717 	/* Otherwise, check the groups (first match) */
3718 	if (groupmember(file_gid, cred)) {
3719 		if (file_mode & S_IXGRP)
3720 			dac_granted |= VEXEC;
3721 		if (file_mode & S_IRGRP)
3722 			dac_granted |= VREAD;
3723 		if (file_mode & S_IWGRP)
3724 			dac_granted |= (VWRITE | VAPPEND);
3725 
3726 		if ((accmode & dac_granted) == accmode)
3727 			return (0);
3728 
3729 		goto privcheck;
3730 	}
3731 
3732 	/* Otherwise, check everyone else. */
3733 	if (file_mode & S_IXOTH)
3734 		dac_granted |= VEXEC;
3735 	if (file_mode & S_IROTH)
3736 		dac_granted |= VREAD;
3737 	if (file_mode & S_IWOTH)
3738 		dac_granted |= (VWRITE | VAPPEND);
3739 	if ((accmode & dac_granted) == accmode)
3740 		return (0);
3741 
3742 privcheck:
3743 	/*
3744 	 * Build a privilege mask to determine if the set of privileges
3745 	 * satisfies the requirements when combined with the granted mask
3746 	 * from above.  For each privilege, if the privilege is required,
3747 	 * bitwise or the request type onto the priv_granted mask.
3748 	 */
3749 	priv_granted = 0;
3750 
3751 	if (type == VDIR) {
3752 		/*
3753 		 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3754 		 * requests, instead of PRIV_VFS_EXEC.
3755 		 */
3756 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3757 		    !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3758 			priv_granted |= VEXEC;
3759 	} else {
3760 		/*
3761 		 * Ensure that at least one execute bit is on. Otherwise,
3762 		 * a privileged user will always succeed, and we don't want
3763 		 * this to happen unless the file really is executable.
3764 		 */
3765 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3766 		    (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
3767 		    !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3768 			priv_granted |= VEXEC;
3769 	}
3770 
3771 	if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
3772 	    !priv_check_cred(cred, PRIV_VFS_READ, 0))
3773 		priv_granted |= VREAD;
3774 
3775 	if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3776 	    !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3777 		priv_granted |= (VWRITE | VAPPEND);
3778 
3779 	if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3780 	    !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3781 		priv_granted |= VADMIN;
3782 
3783 	if ((accmode & (priv_granted | dac_granted)) == accmode) {
3784 		/* XXX audit: privilege used */
3785 		if (privused != NULL)
3786 			*privused = 1;
3787 		return (0);
3788 	}
3789 
3790 	return ((accmode & VADMIN) ? EPERM : EACCES);
3791 }
3792 
3793 /*
3794  * Credential check based on process requesting service, and per-attribute
3795  * permissions.
3796  */
3797 int
3798 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3799     struct thread *td, accmode_t accmode)
3800 {
3801 
3802 	/*
3803 	 * Kernel-invoked always succeeds.
3804 	 */
3805 	if (cred == NOCRED)
3806 		return (0);
3807 
3808 	/*
3809 	 * Do not allow privileged processes in jail to directly manipulate
3810 	 * system attributes.
3811 	 */
3812 	switch (attrnamespace) {
3813 	case EXTATTR_NAMESPACE_SYSTEM:
3814 		/* Potentially should be: return (EPERM); */
3815 		return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3816 	case EXTATTR_NAMESPACE_USER:
3817 		return (VOP_ACCESS(vp, accmode, cred, td));
3818 	default:
3819 		return (EPERM);
3820 	}
3821 }
3822 
3823 #ifdef DEBUG_VFS_LOCKS
3824 /*
3825  * This only exists to supress warnings from unlocked specfs accesses.  It is
3826  * no longer ok to have an unlocked VFS.
3827  */
3828 #define	IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL ||		\
3829 	(vp)->v_type == VCHR ||	(vp)->v_type == VBAD)
3830 
3831 int vfs_badlock_ddb = 1;	/* Drop into debugger on violation. */
3832 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
3833     "Drop into debugger on lock violation");
3834 
3835 int vfs_badlock_mutex = 1;	/* Check for interlock across VOPs. */
3836 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
3837     0, "Check for interlock across VOPs");
3838 
3839 int vfs_badlock_print = 1;	/* Print lock violations. */
3840 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
3841     0, "Print lock violations");
3842 
3843 #ifdef KDB
3844 int vfs_badlock_backtrace = 1;	/* Print backtrace at lock violations. */
3845 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
3846     &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
3847 #endif
3848 
3849 static void
3850 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3851 {
3852 
3853 #ifdef KDB
3854 	if (vfs_badlock_backtrace)
3855 		kdb_backtrace();
3856 #endif
3857 	if (vfs_badlock_print)
3858 		printf("%s: %p %s\n", str, (void *)vp, msg);
3859 	if (vfs_badlock_ddb)
3860 		kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3861 }
3862 
3863 void
3864 assert_vi_locked(struct vnode *vp, const char *str)
3865 {
3866 
3867 	if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3868 		vfs_badlock("interlock is not locked but should be", str, vp);
3869 }
3870 
3871 void
3872 assert_vi_unlocked(struct vnode *vp, const char *str)
3873 {
3874 
3875 	if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3876 		vfs_badlock("interlock is locked but should not be", str, vp);
3877 }
3878 
3879 void
3880 assert_vop_locked(struct vnode *vp, const char *str)
3881 {
3882 
3883 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0)
3884 		vfs_badlock("is not locked but should be", str, vp);
3885 }
3886 
3887 void
3888 assert_vop_unlocked(struct vnode *vp, const char *str)
3889 {
3890 
3891 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
3892 		vfs_badlock("is locked but should not be", str, vp);
3893 }
3894 
3895 void
3896 assert_vop_elocked(struct vnode *vp, const char *str)
3897 {
3898 
3899 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
3900 		vfs_badlock("is not exclusive locked but should be", str, vp);
3901 }
3902 
3903 #if 0
3904 void
3905 assert_vop_elocked_other(struct vnode *vp, const char *str)
3906 {
3907 
3908 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
3909 		vfs_badlock("is not exclusive locked by another thread",
3910 		    str, vp);
3911 }
3912 
3913 void
3914 assert_vop_slocked(struct vnode *vp, const char *str)
3915 {
3916 
3917 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
3918 		vfs_badlock("is not locked shared but should be", str, vp);
3919 }
3920 #endif /* 0 */
3921 #endif /* DEBUG_VFS_LOCKS */
3922 
3923 void
3924 vop_rename_fail(struct vop_rename_args *ap)
3925 {
3926 
3927 	if (ap->a_tvp != NULL)
3928 		vput(ap->a_tvp);
3929 	if (ap->a_tdvp == ap->a_tvp)
3930 		vrele(ap->a_tdvp);
3931 	else
3932 		vput(ap->a_tdvp);
3933 	vrele(ap->a_fdvp);
3934 	vrele(ap->a_fvp);
3935 }
3936 
3937 void
3938 vop_rename_pre(void *ap)
3939 {
3940 	struct vop_rename_args *a = ap;
3941 
3942 #ifdef DEBUG_VFS_LOCKS
3943 	if (a->a_tvp)
3944 		ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
3945 	ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
3946 	ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
3947 	ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
3948 
3949 	/* Check the source (from). */
3950 	if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
3951 	    (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
3952 		ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3953 	if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
3954 		ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
3955 
3956 	/* Check the target. */
3957 	if (a->a_tvp)
3958 		ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3959 	ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3960 #endif
3961 	if (a->a_tdvp != a->a_fdvp)
3962 		vhold(a->a_fdvp);
3963 	if (a->a_tvp != a->a_fvp)
3964 		vhold(a->a_fvp);
3965 	vhold(a->a_tdvp);
3966 	if (a->a_tvp)
3967 		vhold(a->a_tvp);
3968 }
3969 
3970 void
3971 vop_strategy_pre(void *ap)
3972 {
3973 #ifdef DEBUG_VFS_LOCKS
3974 	struct vop_strategy_args *a;
3975 	struct buf *bp;
3976 
3977 	a = ap;
3978 	bp = a->a_bp;
3979 
3980 	/*
3981 	 * Cluster ops lock their component buffers but not the IO container.
3982 	 */
3983 	if ((bp->b_flags & B_CLUSTER) != 0)
3984 		return;
3985 
3986 	if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
3987 		if (vfs_badlock_print)
3988 			printf(
3989 			    "VOP_STRATEGY: bp is not locked but should be\n");
3990 		if (vfs_badlock_ddb)
3991 			kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3992 	}
3993 #endif
3994 }
3995 
3996 void
3997 vop_lookup_pre(void *ap)
3998 {
3999 #ifdef DEBUG_VFS_LOCKS
4000 	struct vop_lookup_args *a;
4001 	struct vnode *dvp;
4002 
4003 	a = ap;
4004 	dvp = a->a_dvp;
4005 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
4006 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
4007 #endif
4008 }
4009 
4010 void
4011 vop_lookup_post(void *ap, int rc)
4012 {
4013 #ifdef DEBUG_VFS_LOCKS
4014 	struct vop_lookup_args *a;
4015 	struct vnode *dvp;
4016 	struct vnode *vp;
4017 
4018 	a = ap;
4019 	dvp = a->a_dvp;
4020 	vp = *(a->a_vpp);
4021 
4022 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
4023 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
4024 
4025 	if (!rc)
4026 		ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
4027 #endif
4028 }
4029 
4030 void
4031 vop_lock_pre(void *ap)
4032 {
4033 #ifdef DEBUG_VFS_LOCKS
4034 	struct vop_lock1_args *a = ap;
4035 
4036 	if ((a->a_flags & LK_INTERLOCK) == 0)
4037 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4038 	else
4039 		ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
4040 #endif
4041 }
4042 
4043 void
4044 vop_lock_post(void *ap, int rc)
4045 {
4046 #ifdef DEBUG_VFS_LOCKS
4047 	struct vop_lock1_args *a = ap;
4048 
4049 	ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4050 	if (rc == 0)
4051 		ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
4052 #endif
4053 }
4054 
4055 void
4056 vop_unlock_pre(void *ap)
4057 {
4058 #ifdef DEBUG_VFS_LOCKS
4059 	struct vop_unlock_args *a = ap;
4060 
4061 	if (a->a_flags & LK_INTERLOCK)
4062 		ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
4063 	ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
4064 #endif
4065 }
4066 
4067 void
4068 vop_unlock_post(void *ap, int rc)
4069 {
4070 #ifdef DEBUG_VFS_LOCKS
4071 	struct vop_unlock_args *a = ap;
4072 
4073 	if (a->a_flags & LK_INTERLOCK)
4074 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
4075 #endif
4076 }
4077 
4078 void
4079 vop_create_post(void *ap, int rc)
4080 {
4081 	struct vop_create_args *a = ap;
4082 
4083 	if (!rc)
4084 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4085 }
4086 
4087 void
4088 vop_deleteextattr_post(void *ap, int rc)
4089 {
4090 	struct vop_deleteextattr_args *a = ap;
4091 
4092 	if (!rc)
4093 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4094 }
4095 
4096 void
4097 vop_link_post(void *ap, int rc)
4098 {
4099 	struct vop_link_args *a = ap;
4100 
4101 	if (!rc) {
4102 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
4103 		VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
4104 	}
4105 }
4106 
4107 void
4108 vop_mkdir_post(void *ap, int rc)
4109 {
4110 	struct vop_mkdir_args *a = ap;
4111 
4112 	if (!rc)
4113 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4114 }
4115 
4116 void
4117 vop_mknod_post(void *ap, int rc)
4118 {
4119 	struct vop_mknod_args *a = ap;
4120 
4121 	if (!rc)
4122 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4123 }
4124 
4125 void
4126 vop_remove_post(void *ap, int rc)
4127 {
4128 	struct vop_remove_args *a = ap;
4129 
4130 	if (!rc) {
4131 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4132 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4133 	}
4134 }
4135 
4136 void
4137 vop_rename_post(void *ap, int rc)
4138 {
4139 	struct vop_rename_args *a = ap;
4140 
4141 	if (!rc) {
4142 		VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
4143 		VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
4144 		VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
4145 		if (a->a_tvp)
4146 			VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
4147 	}
4148 	if (a->a_tdvp != a->a_fdvp)
4149 		vdrop(a->a_fdvp);
4150 	if (a->a_tvp != a->a_fvp)
4151 		vdrop(a->a_fvp);
4152 	vdrop(a->a_tdvp);
4153 	if (a->a_tvp)
4154 		vdrop(a->a_tvp);
4155 }
4156 
4157 void
4158 vop_rmdir_post(void *ap, int rc)
4159 {
4160 	struct vop_rmdir_args *a = ap;
4161 
4162 	if (!rc) {
4163 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4164 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4165 	}
4166 }
4167 
4168 void
4169 vop_setattr_post(void *ap, int rc)
4170 {
4171 	struct vop_setattr_args *a = ap;
4172 
4173 	if (!rc)
4174 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4175 }
4176 
4177 void
4178 vop_setextattr_post(void *ap, int rc)
4179 {
4180 	struct vop_setextattr_args *a = ap;
4181 
4182 	if (!rc)
4183 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4184 }
4185 
4186 void
4187 vop_symlink_post(void *ap, int rc)
4188 {
4189 	struct vop_symlink_args *a = ap;
4190 
4191 	if (!rc)
4192 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4193 }
4194 
4195 static struct knlist fs_knlist;
4196 
4197 static void
4198 vfs_event_init(void *arg)
4199 {
4200 	knlist_init_mtx(&fs_knlist, NULL);
4201 }
4202 /* XXX - correct order? */
4203 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4204 
4205 void
4206 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
4207 {
4208 
4209 	KNOTE_UNLOCKED(&fs_knlist, event);
4210 }
4211 
4212 static int	filt_fsattach(struct knote *kn);
4213 static void	filt_fsdetach(struct knote *kn);
4214 static int	filt_fsevent(struct knote *kn, long hint);
4215 
4216 struct filterops fs_filtops = {
4217 	.f_isfd = 0,
4218 	.f_attach = filt_fsattach,
4219 	.f_detach = filt_fsdetach,
4220 	.f_event = filt_fsevent
4221 };
4222 
4223 static int
4224 filt_fsattach(struct knote *kn)
4225 {
4226 
4227 	kn->kn_flags |= EV_CLEAR;
4228 	knlist_add(&fs_knlist, kn, 0);
4229 	return (0);
4230 }
4231 
4232 static void
4233 filt_fsdetach(struct knote *kn)
4234 {
4235 
4236 	knlist_remove(&fs_knlist, kn, 0);
4237 }
4238 
4239 static int
4240 filt_fsevent(struct knote *kn, long hint)
4241 {
4242 
4243 	kn->kn_fflags |= hint;
4244 	return (kn->kn_fflags != 0);
4245 }
4246 
4247 static int
4248 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4249 {
4250 	struct vfsidctl vc;
4251 	int error;
4252 	struct mount *mp;
4253 
4254 	error = SYSCTL_IN(req, &vc, sizeof(vc));
4255 	if (error)
4256 		return (error);
4257 	if (vc.vc_vers != VFS_CTL_VERS1)
4258 		return (EINVAL);
4259 	mp = vfs_getvfs(&vc.vc_fsid);
4260 	if (mp == NULL)
4261 		return (ENOENT);
4262 	/* ensure that a specific sysctl goes to the right filesystem. */
4263 	if (strcmp(vc.vc_fstypename, "*") != 0 &&
4264 	    strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4265 		vfs_rel(mp);
4266 		return (EINVAL);
4267 	}
4268 	VCTLTOREQ(&vc, req);
4269 	error = VFS_SYSCTL(mp, vc.vc_op, req);
4270 	vfs_rel(mp);
4271 	return (error);
4272 }
4273 
4274 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
4275     NULL, 0, sysctl_vfs_ctl, "",
4276     "Sysctl by fsid");
4277 
4278 /*
4279  * Function to initialize a va_filerev field sensibly.
4280  * XXX: Wouldn't a random number make a lot more sense ??
4281  */
4282 u_quad_t
4283 init_va_filerev(void)
4284 {
4285 	struct bintime bt;
4286 
4287 	getbinuptime(&bt);
4288 	return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4289 }
4290 
4291 static int	filt_vfsread(struct knote *kn, long hint);
4292 static int	filt_vfswrite(struct knote *kn, long hint);
4293 static int	filt_vfsvnode(struct knote *kn, long hint);
4294 static void	filt_vfsdetach(struct knote *kn);
4295 static struct filterops vfsread_filtops = {
4296 	.f_isfd = 1,
4297 	.f_detach = filt_vfsdetach,
4298 	.f_event = filt_vfsread
4299 };
4300 static struct filterops vfswrite_filtops = {
4301 	.f_isfd = 1,
4302 	.f_detach = filt_vfsdetach,
4303 	.f_event = filt_vfswrite
4304 };
4305 static struct filterops vfsvnode_filtops = {
4306 	.f_isfd = 1,
4307 	.f_detach = filt_vfsdetach,
4308 	.f_event = filt_vfsvnode
4309 };
4310 
4311 static void
4312 vfs_knllock(void *arg)
4313 {
4314 	struct vnode *vp = arg;
4315 
4316 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4317 }
4318 
4319 static void
4320 vfs_knlunlock(void *arg)
4321 {
4322 	struct vnode *vp = arg;
4323 
4324 	VOP_UNLOCK(vp, 0);
4325 }
4326 
4327 static void
4328 vfs_knl_assert_locked(void *arg)
4329 {
4330 #ifdef DEBUG_VFS_LOCKS
4331 	struct vnode *vp = arg;
4332 
4333 	ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4334 #endif
4335 }
4336 
4337 static void
4338 vfs_knl_assert_unlocked(void *arg)
4339 {
4340 #ifdef DEBUG_VFS_LOCKS
4341 	struct vnode *vp = arg;
4342 
4343 	ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4344 #endif
4345 }
4346 
4347 int
4348 vfs_kqfilter(struct vop_kqfilter_args *ap)
4349 {
4350 	struct vnode *vp = ap->a_vp;
4351 	struct knote *kn = ap->a_kn;
4352 	struct knlist *knl;
4353 
4354 	switch (kn->kn_filter) {
4355 	case EVFILT_READ:
4356 		kn->kn_fop = &vfsread_filtops;
4357 		break;
4358 	case EVFILT_WRITE:
4359 		kn->kn_fop = &vfswrite_filtops;
4360 		break;
4361 	case EVFILT_VNODE:
4362 		kn->kn_fop = &vfsvnode_filtops;
4363 		break;
4364 	default:
4365 		return (EINVAL);
4366 	}
4367 
4368 	kn->kn_hook = (caddr_t)vp;
4369 
4370 	v_addpollinfo(vp);
4371 	if (vp->v_pollinfo == NULL)
4372 		return (ENOMEM);
4373 	knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4374 	knlist_add(knl, kn, 0);
4375 
4376 	return (0);
4377 }
4378 
4379 /*
4380  * Detach knote from vnode
4381  */
4382 static void
4383 filt_vfsdetach(struct knote *kn)
4384 {
4385 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4386 
4387 	KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4388 	knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4389 }
4390 
4391 /*ARGSUSED*/
4392 static int
4393 filt_vfsread(struct knote *kn, long hint)
4394 {
4395 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4396 	struct vattr va;
4397 	int res;
4398 
4399 	/*
4400 	 * filesystem is gone, so set the EOF flag and schedule
4401 	 * the knote for deletion.
4402 	 */
4403 	if (hint == NOTE_REVOKE) {
4404 		VI_LOCK(vp);
4405 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4406 		VI_UNLOCK(vp);
4407 		return (1);
4408 	}
4409 
4410 	if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4411 		return (0);
4412 
4413 	VI_LOCK(vp);
4414 	kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4415 	res = (kn->kn_data != 0);
4416 	VI_UNLOCK(vp);
4417 	return (res);
4418 }
4419 
4420 /*ARGSUSED*/
4421 static int
4422 filt_vfswrite(struct knote *kn, long hint)
4423 {
4424 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4425 
4426 	VI_LOCK(vp);
4427 
4428 	/*
4429 	 * filesystem is gone, so set the EOF flag and schedule
4430 	 * the knote for deletion.
4431 	 */
4432 	if (hint == NOTE_REVOKE)
4433 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4434 
4435 	kn->kn_data = 0;
4436 	VI_UNLOCK(vp);
4437 	return (1);
4438 }
4439 
4440 static int
4441 filt_vfsvnode(struct knote *kn, long hint)
4442 {
4443 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4444 	int res;
4445 
4446 	VI_LOCK(vp);
4447 	if (kn->kn_sfflags & hint)
4448 		kn->kn_fflags |= hint;
4449 	if (hint == NOTE_REVOKE) {
4450 		kn->kn_flags |= EV_EOF;
4451 		VI_UNLOCK(vp);
4452 		return (1);
4453 	}
4454 	res = (kn->kn_fflags != 0);
4455 	VI_UNLOCK(vp);
4456 	return (res);
4457 }
4458 
4459 int
4460 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4461 {
4462 	int error;
4463 
4464 	if (dp->d_reclen > ap->a_uio->uio_resid)
4465 		return (ENAMETOOLONG);
4466 	error = uiomove(dp, dp->d_reclen, ap->a_uio);
4467 	if (error) {
4468 		if (ap->a_ncookies != NULL) {
4469 			if (ap->a_cookies != NULL)
4470 				free(ap->a_cookies, M_TEMP);
4471 			ap->a_cookies = NULL;
4472 			*ap->a_ncookies = 0;
4473 		}
4474 		return (error);
4475 	}
4476 	if (ap->a_ncookies == NULL)
4477 		return (0);
4478 
4479 	KASSERT(ap->a_cookies,
4480 	    ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4481 
4482 	*ap->a_cookies = realloc(*ap->a_cookies,
4483 	    (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4484 	(*ap->a_cookies)[*ap->a_ncookies] = off;
4485 	return (0);
4486 }
4487 
4488 /*
4489  * Mark for update the access time of the file if the filesystem
4490  * supports VOP_MARKATIME.  This functionality is used by execve and
4491  * mmap, so we want to avoid the I/O implied by directly setting
4492  * va_atime for the sake of efficiency.
4493  */
4494 void
4495 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4496 {
4497 	struct mount *mp;
4498 
4499 	mp = vp->v_mount;
4500 	VFS_ASSERT_GIANT(mp);
4501 	ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4502 	if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4503 		(void)VOP_MARKATIME(vp);
4504 }
4505 
4506 /*
4507  * The purpose of this routine is to remove granularity from accmode_t,
4508  * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4509  * VADMIN and VAPPEND.
4510  *
4511  * If it returns 0, the caller is supposed to continue with the usual
4512  * access checks using 'accmode' as modified by this routine.  If it
4513  * returns nonzero value, the caller is supposed to return that value
4514  * as errno.
4515  *
4516  * Note that after this routine runs, accmode may be zero.
4517  */
4518 int
4519 vfs_unixify_accmode(accmode_t *accmode)
4520 {
4521 	/*
4522 	 * There is no way to specify explicit "deny" rule using
4523 	 * file mode or POSIX.1e ACLs.
4524 	 */
4525 	if (*accmode & VEXPLICIT_DENY) {
4526 		*accmode = 0;
4527 		return (0);
4528 	}
4529 
4530 	/*
4531 	 * None of these can be translated into usual access bits.
4532 	 * Also, the common case for NFSv4 ACLs is to not contain
4533 	 * either of these bits. Caller should check for VWRITE
4534 	 * on the containing directory instead.
4535 	 */
4536 	if (*accmode & (VDELETE_CHILD | VDELETE))
4537 		return (EPERM);
4538 
4539 	if (*accmode & VADMIN_PERMS) {
4540 		*accmode &= ~VADMIN_PERMS;
4541 		*accmode |= VADMIN;
4542 	}
4543 
4544 	/*
4545 	 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4546 	 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4547 	 */
4548 	*accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
4549 
4550 	return (0);
4551 }
4552 
4553 /*
4554  * These are helper functions for filesystems to traverse all
4555  * their vnodes.  See MNT_VNODE_FOREACH_ALL() in sys/mount.h.
4556  *
4557  * This interface replaces MNT_VNODE_FOREACH.
4558  */
4559 
4560 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
4561 
4562 struct vnode *
4563 __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp)
4564 {
4565 	struct vnode *vp;
4566 
4567 	if (should_yield())
4568 		kern_yield(PRI_UNCHANGED);
4569 	MNT_ILOCK(mp);
4570 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4571 	vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
4572 	while (vp != NULL && (vp->v_type == VMARKER ||
4573 	    (vp->v_iflag & VI_DOOMED) != 0))
4574 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
4575 
4576 	/* Check if we are done */
4577 	if (vp == NULL) {
4578 		__mnt_vnode_markerfree_all(mvp, mp);
4579 		/* MNT_IUNLOCK(mp); -- done in above function */
4580 		mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
4581 		return (NULL);
4582 	}
4583 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4584 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4585 	VI_LOCK(vp);
4586 	MNT_IUNLOCK(mp);
4587 	return (vp);
4588 }
4589 
4590 struct vnode *
4591 __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp)
4592 {
4593 	struct vnode *vp;
4594 
4595 	*mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4596 	MNT_ILOCK(mp);
4597 	MNT_REF(mp);
4598 	(*mvp)->v_type = VMARKER;
4599 
4600 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
4601 	while (vp != NULL && (vp->v_type == VMARKER ||
4602 	    (vp->v_iflag & VI_DOOMED) != 0))
4603 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
4604 
4605 	/* Check if we are done */
4606 	if (vp == NULL) {
4607 		MNT_REL(mp);
4608 		MNT_IUNLOCK(mp);
4609 		free(*mvp, M_VNODE_MARKER);
4610 		*mvp = NULL;
4611 		return (NULL);
4612 	}
4613 	(*mvp)->v_mount = mp;
4614 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4615 	VI_LOCK(vp);
4616 	MNT_IUNLOCK(mp);
4617 	return (vp);
4618 }
4619 
4620 
4621 void
4622 __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp)
4623 {
4624 
4625 	if (*mvp == NULL) {
4626 		MNT_IUNLOCK(mp);
4627 		return;
4628 	}
4629 
4630 	mtx_assert(MNT_MTX(mp), MA_OWNED);
4631 
4632 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4633 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4634 	MNT_REL(mp);
4635 	MNT_IUNLOCK(mp);
4636 	free(*mvp, M_VNODE_MARKER);
4637 	*mvp = NULL;
4638 }
4639 
4640 /*
4641  * These are helper functions for filesystems to traverse their
4642  * active vnodes.  See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h
4643  */
4644 struct vnode *
4645 __mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4646 {
4647 	struct vnode *vp, *nvp;
4648 
4649 	if (should_yield())
4650 		kern_yield(PRI_UNCHANGED);
4651 	MNT_ILOCK(mp);
4652 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4653 	vp = TAILQ_NEXT(*mvp, v_actfreelist);
4654 	while (vp != NULL) {
4655 		VI_LOCK(vp);
4656 		if (vp->v_mount == mp && vp->v_type != VMARKER &&
4657 		    (vp->v_iflag & VI_DOOMED) == 0)
4658 			break;
4659 		nvp = TAILQ_NEXT(vp, v_actfreelist);
4660 		VI_UNLOCK(vp);
4661 		vp = nvp;
4662 	}
4663 
4664 	/* Check if we are done */
4665 	if (vp == NULL) {
4666 		__mnt_vnode_markerfree_active(mvp, mp);
4667 		/* MNT_IUNLOCK(mp); -- done in above function */
4668 		mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
4669 		return (NULL);
4670 	}
4671 	mtx_lock(&vnode_free_list_mtx);
4672 	TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4673 	TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
4674 	mtx_unlock(&vnode_free_list_mtx);
4675 	MNT_IUNLOCK(mp);
4676 	return (vp);
4677 }
4678 
4679 struct vnode *
4680 __mnt_vnode_first_active(struct vnode **mvp, struct mount *mp)
4681 {
4682 	struct vnode *vp, *nvp;
4683 
4684 	*mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4685 	MNT_ILOCK(mp);
4686 	MNT_REF(mp);
4687 	(*mvp)->v_type = VMARKER;
4688 
4689 	vp = TAILQ_NEXT(*mvp, v_actfreelist);
4690 	while (vp != NULL) {
4691 		VI_LOCK(vp);
4692 		if (vp->v_mount == mp && vp->v_type != VMARKER &&
4693 		    (vp->v_iflag & VI_DOOMED) == 0)
4694 			break;
4695 		nvp = TAILQ_NEXT(vp, v_actfreelist);
4696 		VI_UNLOCK(vp);
4697 		vp = nvp;
4698 	}
4699 
4700 	/* Check if we are done */
4701 	if (vp == NULL) {
4702 		MNT_REL(mp);
4703 		MNT_IUNLOCK(mp);
4704 		free(*mvp, M_VNODE_MARKER);
4705 		*mvp = NULL;
4706 		return (NULL);
4707 	}
4708 	(*mvp)->v_mount = mp;
4709 	mtx_lock(&vnode_free_list_mtx);
4710 	TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
4711 	mtx_unlock(&vnode_free_list_mtx);
4712 	MNT_IUNLOCK(mp);
4713 	return (vp);
4714 }
4715 
4716 void
4717 __mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4718 {
4719 
4720 	if (*mvp == NULL) {
4721 		MNT_IUNLOCK(mp);
4722 		return;
4723 	}
4724 
4725 	mtx_assert(MNT_MTX(mp), MA_OWNED);
4726 
4727 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4728 	mtx_lock(&vnode_free_list_mtx);
4729 	TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4730 	mtx_unlock(&vnode_free_list_mtx);
4731 	MNT_REL(mp);
4732 	MNT_IUNLOCK(mp);
4733 	free(*mvp, M_VNODE_MARKER);
4734 	*mvp = NULL;
4735 }
4736