xref: /freebsd/sys/kern/vfs_subr.c (revision 9a41df2a0e6408e9b329bbd8b9e37c2b44461a1b)
1 /*-
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)vfs_subr.c	8.31 (Berkeley) 5/26/95
35  */
36 
37 /*
38  * External virtual filesystem routines
39  */
40 
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
43 
44 #include "opt_compat.h"
45 #include "opt_ddb.h"
46 #include "opt_watchdog.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/bio.h>
51 #include <sys/buf.h>
52 #include <sys/condvar.h>
53 #include <sys/conf.h>
54 #include <sys/dirent.h>
55 #include <sys/event.h>
56 #include <sys/eventhandler.h>
57 #include <sys/extattr.h>
58 #include <sys/file.h>
59 #include <sys/fcntl.h>
60 #include <sys/jail.h>
61 #include <sys/kdb.h>
62 #include <sys/kernel.h>
63 #include <sys/kthread.h>
64 #include <sys/lockf.h>
65 #include <sys/malloc.h>
66 #include <sys/mount.h>
67 #include <sys/namei.h>
68 #include <sys/priv.h>
69 #include <sys/reboot.h>
70 #include <sys/sched.h>
71 #include <sys/sleepqueue.h>
72 #include <sys/stat.h>
73 #include <sys/sysctl.h>
74 #include <sys/syslog.h>
75 #include <sys/vmmeter.h>
76 #include <sys/vnode.h>
77 #include <sys/watchdog.h>
78 
79 #include <machine/stdarg.h>
80 
81 #include <security/mac/mac_framework.h>
82 
83 #include <vm/vm.h>
84 #include <vm/vm_object.h>
85 #include <vm/vm_extern.h>
86 #include <vm/pmap.h>
87 #include <vm/vm_map.h>
88 #include <vm/vm_page.h>
89 #include <vm/vm_kern.h>
90 #include <vm/uma.h>
91 
92 #ifdef DDB
93 #include <ddb/ddb.h>
94 #endif
95 
96 #define	WI_MPSAFEQ	0
97 #define	WI_GIANTQ	1
98 
99 static void	delmntque(struct vnode *vp);
100 static int	flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
101 		    int slpflag, int slptimeo);
102 static void	syncer_shutdown(void *arg, int howto);
103 static int	vtryrecycle(struct vnode *vp);
104 static void	v_incr_usecount(struct vnode *);
105 static void	v_decr_usecount(struct vnode *);
106 static void	v_decr_useonly(struct vnode *);
107 static void	v_upgrade_usecount(struct vnode *);
108 static void	vnlru_free(int);
109 static void	vgonel(struct vnode *);
110 static void	vfs_knllock(void *arg);
111 static void	vfs_knlunlock(void *arg);
112 static void	vfs_knl_assert_locked(void *arg);
113 static void	vfs_knl_assert_unlocked(void *arg);
114 static void	destroy_vpollinfo(struct vpollinfo *vi);
115 
116 /*
117  * Number of vnodes in existence.  Increased whenever getnewvnode()
118  * allocates a new vnode, decreased in vdropl() for VI_DOOMED vnode.
119  */
120 static unsigned long	numvnodes;
121 
122 SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
123     "Number of vnodes in existence");
124 
125 /*
126  * Conversion tables for conversion from vnode types to inode formats
127  * and back.
128  */
129 enum vtype iftovt_tab[16] = {
130 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
131 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
132 };
133 int vttoif_tab[10] = {
134 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
135 	S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
136 };
137 
138 /*
139  * List of vnodes that are ready for recycling.
140  */
141 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
142 
143 /*
144  * Free vnode target.  Free vnodes may simply be files which have been stat'd
145  * but not read.  This is somewhat common, and a small cache of such files
146  * should be kept to avoid recreation costs.
147  */
148 static u_long wantfreevnodes;
149 SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
150 /* Number of vnodes in the free list. */
151 static u_long freevnodes;
152 SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
153     "Number of vnodes in the free list");
154 
155 static int vlru_allow_cache_src;
156 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
157     &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
158 
159 /*
160  * Various variables used for debugging the new implementation of
161  * reassignbuf().
162  * XXX these are probably of (very) limited utility now.
163  */
164 static int reassignbufcalls;
165 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
166     "Number of calls to reassignbuf");
167 
168 /*
169  * Cache for the mount type id assigned to NFS.  This is used for
170  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
171  */
172 int	nfs_mount_type = -1;
173 
174 /* To keep more than one thread at a time from running vfs_getnewfsid */
175 static struct mtx mntid_mtx;
176 
177 /*
178  * Lock for any access to the following:
179  *	vnode_free_list
180  *	numvnodes
181  *	freevnodes
182  */
183 static struct mtx vnode_free_list_mtx;
184 
185 /* Publicly exported FS */
186 struct nfs_public nfs_pub;
187 
188 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
189 static uma_zone_t vnode_zone;
190 static uma_zone_t vnodepoll_zone;
191 
192 /*
193  * The workitem queue.
194  *
195  * It is useful to delay writes of file data and filesystem metadata
196  * for tens of seconds so that quickly created and deleted files need
197  * not waste disk bandwidth being created and removed. To realize this,
198  * we append vnodes to a "workitem" queue. When running with a soft
199  * updates implementation, most pending metadata dependencies should
200  * not wait for more than a few seconds. Thus, mounted on block devices
201  * are delayed only about a half the time that file data is delayed.
202  * Similarly, directory updates are more critical, so are only delayed
203  * about a third the time that file data is delayed. Thus, there are
204  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
205  * one each second (driven off the filesystem syncer process). The
206  * syncer_delayno variable indicates the next queue that is to be processed.
207  * Items that need to be processed soon are placed in this queue:
208  *
209  *	syncer_workitem_pending[syncer_delayno]
210  *
211  * A delay of fifteen seconds is done by placing the request fifteen
212  * entries later in the queue:
213  *
214  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
215  *
216  */
217 static int syncer_delayno;
218 static long syncer_mask;
219 LIST_HEAD(synclist, bufobj);
220 static struct synclist *syncer_workitem_pending[2];
221 /*
222  * The sync_mtx protects:
223  *	bo->bo_synclist
224  *	sync_vnode_count
225  *	syncer_delayno
226  *	syncer_state
227  *	syncer_workitem_pending
228  *	syncer_worklist_len
229  *	rushjob
230  */
231 static struct mtx sync_mtx;
232 static struct cv sync_wakeup;
233 
234 #define SYNCER_MAXDELAY		32
235 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
236 static int syncdelay = 30;		/* max time to delay syncing data */
237 static int filedelay = 30;		/* time to delay syncing files */
238 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
239     "Time to delay syncing files (in seconds)");
240 static int dirdelay = 29;		/* time to delay syncing directories */
241 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
242     "Time to delay syncing directories (in seconds)");
243 static int metadelay = 28;		/* time to delay syncing metadata */
244 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
245     "Time to delay syncing metadata (in seconds)");
246 static int rushjob;		/* number of slots to run ASAP */
247 static int stat_rush_requests;	/* number of times I/O speeded up */
248 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
249     "Number of times I/O speeded up (rush requests)");
250 
251 /*
252  * When shutting down the syncer, run it at four times normal speed.
253  */
254 #define SYNCER_SHUTDOWN_SPEEDUP		4
255 static int sync_vnode_count;
256 static int syncer_worklist_len;
257 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
258     syncer_state;
259 
260 /*
261  * Number of vnodes we want to exist at any one time.  This is mostly used
262  * to size hash tables in vnode-related code.  It is normally not used in
263  * getnewvnode(), as wantfreevnodes is normally nonzero.)
264  *
265  * XXX desiredvnodes is historical cruft and should not exist.
266  */
267 int desiredvnodes;
268 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
269     &desiredvnodes, 0, "Maximum number of vnodes");
270 SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
271     &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
272 static int vnlru_nowhere;
273 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
274     &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
275 
276 /*
277  * Macros to control when a vnode is freed and recycled.  All require
278  * the vnode interlock.
279  */
280 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
281 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
282 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
283 
284 
285 /*
286  * Initialize the vnode management data structures.
287  *
288  * Reevaluate the following cap on the number of vnodes after the physical
289  * memory size exceeds 512GB.  In the limit, as the physical memory size
290  * grows, the ratio of physical pages to vnodes approaches sixteen to one.
291  */
292 #ifndef	MAXVNODES_MAX
293 #define	MAXVNODES_MAX	(512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
294 #endif
295 static void
296 vntblinit(void *dummy __unused)
297 {
298 	int physvnodes, virtvnodes;
299 
300 	/*
301 	 * Desiredvnodes is a function of the physical memory size and the
302 	 * kernel's heap size.  Generally speaking, it scales with the
303 	 * physical memory size.  The ratio of desiredvnodes to physical pages
304 	 * is one to four until desiredvnodes exceeds 98,304.  Thereafter, the
305 	 * marginal ratio of desiredvnodes to physical pages is one to
306 	 * sixteen.  However, desiredvnodes is limited by the kernel's heap
307 	 * size.  The memory required by desiredvnodes vnodes and vm objects
308 	 * may not exceed one seventh of the kernel's heap size.
309 	 */
310 	physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
311 	    cnt.v_page_count) / 16;
312 	virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
313 	    sizeof(struct vnode)));
314 	desiredvnodes = min(physvnodes, virtvnodes);
315 	if (desiredvnodes > MAXVNODES_MAX) {
316 		if (bootverbose)
317 			printf("Reducing kern.maxvnodes %d -> %d\n",
318 			    desiredvnodes, MAXVNODES_MAX);
319 		desiredvnodes = MAXVNODES_MAX;
320 	}
321 	wantfreevnodes = desiredvnodes / 4;
322 	mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
323 	TAILQ_INIT(&vnode_free_list);
324 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
325 	vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
326 	    NULL, NULL, UMA_ALIGN_PTR, 0);
327 	vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
328 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
329 	/*
330 	 * Initialize the filesystem syncer.
331 	 */
332 	syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE,
333 	    &syncer_mask);
334 	syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE,
335 	    &syncer_mask);
336 	syncer_maxdelay = syncer_mask + 1;
337 	mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
338 	cv_init(&sync_wakeup, "syncer");
339 }
340 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
341 
342 
343 /*
344  * Mark a mount point as busy. Used to synchronize access and to delay
345  * unmounting. Eventually, mountlist_mtx is not released on failure.
346  *
347  * vfs_busy() is a custom lock, it can block the caller.
348  * vfs_busy() only sleeps if the unmount is active on the mount point.
349  * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
350  * vnode belonging to mp.
351  *
352  * Lookup uses vfs_busy() to traverse mount points.
353  * root fs			var fs
354  * / vnode lock		A	/ vnode lock (/var)		D
355  * /var vnode lock	B	/log vnode lock(/var/log)	E
356  * vfs_busy lock	C	vfs_busy lock			F
357  *
358  * Within each file system, the lock order is C->A->B and F->D->E.
359  *
360  * When traversing across mounts, the system follows that lock order:
361  *
362  *        C->A->B
363  *              |
364  *              +->F->D->E
365  *
366  * The lookup() process for namei("/var") illustrates the process:
367  *  VOP_LOOKUP() obtains B while A is held
368  *  vfs_busy() obtains a shared lock on F while A and B are held
369  *  vput() releases lock on B
370  *  vput() releases lock on A
371  *  VFS_ROOT() obtains lock on D while shared lock on F is held
372  *  vfs_unbusy() releases shared lock on F
373  *  vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
374  *    Attempt to lock A (instead of vp_crossmp) while D is held would
375  *    violate the global order, causing deadlocks.
376  *
377  * dounmount() locks B while F is drained.
378  */
379 int
380 vfs_busy(struct mount *mp, int flags)
381 {
382 
383 	MPASS((flags & ~MBF_MASK) == 0);
384 	CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
385 
386 	MNT_ILOCK(mp);
387 	MNT_REF(mp);
388 	/*
389 	 * If mount point is currenly being unmounted, sleep until the
390 	 * mount point fate is decided.  If thread doing the unmounting fails,
391 	 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
392 	 * that this mount point has survived the unmount attempt and vfs_busy
393 	 * should retry.  Otherwise the unmounter thread will set MNTK_REFEXPIRE
394 	 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
395 	 * about to be really destroyed.  vfs_busy needs to release its
396 	 * reference on the mount point in this case and return with ENOENT,
397 	 * telling the caller that mount mount it tried to busy is no longer
398 	 * valid.
399 	 */
400 	while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
401 		if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
402 			MNT_REL(mp);
403 			MNT_IUNLOCK(mp);
404 			CTR1(KTR_VFS, "%s: failed busying before sleeping",
405 			    __func__);
406 			return (ENOENT);
407 		}
408 		if (flags & MBF_MNTLSTLOCK)
409 			mtx_unlock(&mountlist_mtx);
410 		mp->mnt_kern_flag |= MNTK_MWAIT;
411 		msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
412 		if (flags & MBF_MNTLSTLOCK)
413 			mtx_lock(&mountlist_mtx);
414 		MNT_ILOCK(mp);
415 	}
416 	if (flags & MBF_MNTLSTLOCK)
417 		mtx_unlock(&mountlist_mtx);
418 	mp->mnt_lockref++;
419 	MNT_IUNLOCK(mp);
420 	return (0);
421 }
422 
423 /*
424  * Free a busy filesystem.
425  */
426 void
427 vfs_unbusy(struct mount *mp)
428 {
429 
430 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
431 	MNT_ILOCK(mp);
432 	MNT_REL(mp);
433 	KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
434 	mp->mnt_lockref--;
435 	if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
436 		MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
437 		CTR1(KTR_VFS, "%s: waking up waiters", __func__);
438 		mp->mnt_kern_flag &= ~MNTK_DRAINING;
439 		wakeup(&mp->mnt_lockref);
440 	}
441 	MNT_IUNLOCK(mp);
442 }
443 
444 /*
445  * Lookup a mount point by filesystem identifier.
446  */
447 struct mount *
448 vfs_getvfs(fsid_t *fsid)
449 {
450 	struct mount *mp;
451 
452 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
453 	mtx_lock(&mountlist_mtx);
454 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
455 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
456 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
457 			vfs_ref(mp);
458 			mtx_unlock(&mountlist_mtx);
459 			return (mp);
460 		}
461 	}
462 	mtx_unlock(&mountlist_mtx);
463 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
464 	return ((struct mount *) 0);
465 }
466 
467 /*
468  * Lookup a mount point by filesystem identifier, busying it before
469  * returning.
470  */
471 struct mount *
472 vfs_busyfs(fsid_t *fsid)
473 {
474 	struct mount *mp;
475 	int error;
476 
477 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
478 	mtx_lock(&mountlist_mtx);
479 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
480 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
481 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
482 			error = vfs_busy(mp, MBF_MNTLSTLOCK);
483 			if (error) {
484 				mtx_unlock(&mountlist_mtx);
485 				return (NULL);
486 			}
487 			return (mp);
488 		}
489 	}
490 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
491 	mtx_unlock(&mountlist_mtx);
492 	return ((struct mount *) 0);
493 }
494 
495 /*
496  * Check if a user can access privileged mount options.
497  */
498 int
499 vfs_suser(struct mount *mp, struct thread *td)
500 {
501 	int error;
502 
503 	/*
504 	 * If the thread is jailed, but this is not a jail-friendly file
505 	 * system, deny immediately.
506 	 */
507 	if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
508 		return (EPERM);
509 
510 	/*
511 	 * If the file system was mounted outside the jail of the calling
512 	 * thread, deny immediately.
513 	 */
514 	if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
515 		return (EPERM);
516 
517 	/*
518 	 * If file system supports delegated administration, we don't check
519 	 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
520 	 * by the file system itself.
521 	 * If this is not the user that did original mount, we check for
522 	 * the PRIV_VFS_MOUNT_OWNER privilege.
523 	 */
524 	if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
525 	    mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
526 		if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
527 			return (error);
528 	}
529 	return (0);
530 }
531 
532 /*
533  * Get a new unique fsid.  Try to make its val[0] unique, since this value
534  * will be used to create fake device numbers for stat().  Also try (but
535  * not so hard) make its val[0] unique mod 2^16, since some emulators only
536  * support 16-bit device numbers.  We end up with unique val[0]'s for the
537  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
538  *
539  * Keep in mind that several mounts may be running in parallel.  Starting
540  * the search one past where the previous search terminated is both a
541  * micro-optimization and a defense against returning the same fsid to
542  * different mounts.
543  */
544 void
545 vfs_getnewfsid(struct mount *mp)
546 {
547 	static uint16_t mntid_base;
548 	struct mount *nmp;
549 	fsid_t tfsid;
550 	int mtype;
551 
552 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
553 	mtx_lock(&mntid_mtx);
554 	mtype = mp->mnt_vfc->vfc_typenum;
555 	tfsid.val[1] = mtype;
556 	mtype = (mtype & 0xFF) << 24;
557 	for (;;) {
558 		tfsid.val[0] = makedev(255,
559 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
560 		mntid_base++;
561 		if ((nmp = vfs_getvfs(&tfsid)) == NULL)
562 			break;
563 		vfs_rel(nmp);
564 	}
565 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
566 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
567 	mtx_unlock(&mntid_mtx);
568 }
569 
570 /*
571  * Knob to control the precision of file timestamps:
572  *
573  *   0 = seconds only; nanoseconds zeroed.
574  *   1 = seconds and nanoseconds, accurate within 1/HZ.
575  *   2 = seconds and nanoseconds, truncated to microseconds.
576  * >=3 = seconds and nanoseconds, maximum precision.
577  */
578 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
579 
580 static int timestamp_precision = TSP_SEC;
581 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
582     &timestamp_precision, 0, "File timestamp precision (0: seconds, "
583     "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
584     "3+: sec + ns (max. precision))");
585 
586 /*
587  * Get a current timestamp.
588  */
589 void
590 vfs_timestamp(struct timespec *tsp)
591 {
592 	struct timeval tv;
593 
594 	switch (timestamp_precision) {
595 	case TSP_SEC:
596 		tsp->tv_sec = time_second;
597 		tsp->tv_nsec = 0;
598 		break;
599 	case TSP_HZ:
600 		getnanotime(tsp);
601 		break;
602 	case TSP_USEC:
603 		microtime(&tv);
604 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
605 		break;
606 	case TSP_NSEC:
607 	default:
608 		nanotime(tsp);
609 		break;
610 	}
611 }
612 
613 /*
614  * Set vnode attributes to VNOVAL
615  */
616 void
617 vattr_null(struct vattr *vap)
618 {
619 
620 	vap->va_type = VNON;
621 	vap->va_size = VNOVAL;
622 	vap->va_bytes = VNOVAL;
623 	vap->va_mode = VNOVAL;
624 	vap->va_nlink = VNOVAL;
625 	vap->va_uid = VNOVAL;
626 	vap->va_gid = VNOVAL;
627 	vap->va_fsid = VNOVAL;
628 	vap->va_fileid = VNOVAL;
629 	vap->va_blocksize = VNOVAL;
630 	vap->va_rdev = VNOVAL;
631 	vap->va_atime.tv_sec = VNOVAL;
632 	vap->va_atime.tv_nsec = VNOVAL;
633 	vap->va_mtime.tv_sec = VNOVAL;
634 	vap->va_mtime.tv_nsec = VNOVAL;
635 	vap->va_ctime.tv_sec = VNOVAL;
636 	vap->va_ctime.tv_nsec = VNOVAL;
637 	vap->va_birthtime.tv_sec = VNOVAL;
638 	vap->va_birthtime.tv_nsec = VNOVAL;
639 	vap->va_flags = VNOVAL;
640 	vap->va_gen = VNOVAL;
641 	vap->va_vaflags = 0;
642 }
643 
644 /*
645  * This routine is called when we have too many vnodes.  It attempts
646  * to free <count> vnodes and will potentially free vnodes that still
647  * have VM backing store (VM backing store is typically the cause
648  * of a vnode blowout so we want to do this).  Therefore, this operation
649  * is not considered cheap.
650  *
651  * A number of conditions may prevent a vnode from being reclaimed.
652  * the buffer cache may have references on the vnode, a directory
653  * vnode may still have references due to the namei cache representing
654  * underlying files, or the vnode may be in active use.   It is not
655  * desireable to reuse such vnodes.  These conditions may cause the
656  * number of vnodes to reach some minimum value regardless of what
657  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
658  */
659 static int
660 vlrureclaim(struct mount *mp)
661 {
662 	struct vnode *vp;
663 	int done;
664 	int trigger;
665 	int usevnodes;
666 	int count;
667 
668 	/*
669 	 * Calculate the trigger point, don't allow user
670 	 * screwups to blow us up.   This prevents us from
671 	 * recycling vnodes with lots of resident pages.  We
672 	 * aren't trying to free memory, we are trying to
673 	 * free vnodes.
674 	 */
675 	usevnodes = desiredvnodes;
676 	if (usevnodes <= 0)
677 		usevnodes = 1;
678 	trigger = cnt.v_page_count * 2 / usevnodes;
679 	done = 0;
680 	vn_start_write(NULL, &mp, V_WAIT);
681 	MNT_ILOCK(mp);
682 	count = mp->mnt_nvnodelistsize / 10 + 1;
683 	while (count != 0) {
684 		vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
685 		while (vp != NULL && vp->v_type == VMARKER)
686 			vp = TAILQ_NEXT(vp, v_nmntvnodes);
687 		if (vp == NULL)
688 			break;
689 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
690 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
691 		--count;
692 		if (!VI_TRYLOCK(vp))
693 			goto next_iter;
694 		/*
695 		 * If it's been deconstructed already, it's still
696 		 * referenced, or it exceeds the trigger, skip it.
697 		 */
698 		if (vp->v_usecount ||
699 		    (!vlru_allow_cache_src &&
700 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
701 		    (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
702 		    vp->v_object->resident_page_count > trigger)) {
703 			VI_UNLOCK(vp);
704 			goto next_iter;
705 		}
706 		MNT_IUNLOCK(mp);
707 		vholdl(vp);
708 		if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
709 			vdrop(vp);
710 			goto next_iter_mntunlocked;
711 		}
712 		VI_LOCK(vp);
713 		/*
714 		 * v_usecount may have been bumped after VOP_LOCK() dropped
715 		 * the vnode interlock and before it was locked again.
716 		 *
717 		 * It is not necessary to recheck VI_DOOMED because it can
718 		 * only be set by another thread that holds both the vnode
719 		 * lock and vnode interlock.  If another thread has the
720 		 * vnode lock before we get to VOP_LOCK() and obtains the
721 		 * vnode interlock after VOP_LOCK() drops the vnode
722 		 * interlock, the other thread will be unable to drop the
723 		 * vnode lock before our VOP_LOCK() call fails.
724 		 */
725 		if (vp->v_usecount ||
726 		    (!vlru_allow_cache_src &&
727 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
728 		    (vp->v_object != NULL &&
729 		    vp->v_object->resident_page_count > trigger)) {
730 			VOP_UNLOCK(vp, LK_INTERLOCK);
731 			goto next_iter_mntunlocked;
732 		}
733 		KASSERT((vp->v_iflag & VI_DOOMED) == 0,
734 		    ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
735 		vgonel(vp);
736 		VOP_UNLOCK(vp, 0);
737 		vdropl(vp);
738 		done++;
739 next_iter_mntunlocked:
740 		if (!should_yield())
741 			goto relock_mnt;
742 		goto yield;
743 next_iter:
744 		if (!should_yield())
745 			continue;
746 		MNT_IUNLOCK(mp);
747 yield:
748 		kern_yield(PRI_UNCHANGED);
749 relock_mnt:
750 		MNT_ILOCK(mp);
751 	}
752 	MNT_IUNLOCK(mp);
753 	vn_finished_write(mp);
754 	return done;
755 }
756 
757 /*
758  * Attempt to keep the free list at wantfreevnodes length.
759  */
760 static void
761 vnlru_free(int count)
762 {
763 	struct vnode *vp;
764 	int vfslocked;
765 
766 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
767 	for (; count > 0; count--) {
768 		vp = TAILQ_FIRST(&vnode_free_list);
769 		/*
770 		 * The list can be modified while the free_list_mtx
771 		 * has been dropped and vp could be NULL here.
772 		 */
773 		if (!vp)
774 			break;
775 		VNASSERT(vp->v_op != NULL, vp,
776 		    ("vnlru_free: vnode already reclaimed."));
777 		KASSERT((vp->v_iflag & VI_FREE) != 0,
778 		    ("Removing vnode not on freelist"));
779 		KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
780 		    ("Mangling active vnode"));
781 		TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
782 		/*
783 		 * Don't recycle if we can't get the interlock.
784 		 */
785 		if (!VI_TRYLOCK(vp)) {
786 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
787 			continue;
788 		}
789 		VNASSERT(VCANRECYCLE(vp), vp,
790 		    ("vp inconsistent on freelist"));
791 		freevnodes--;
792 		vp->v_iflag &= ~VI_FREE;
793 		vholdl(vp);
794 		mtx_unlock(&vnode_free_list_mtx);
795 		VI_UNLOCK(vp);
796 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
797 		vtryrecycle(vp);
798 		VFS_UNLOCK_GIANT(vfslocked);
799 		/*
800 		 * If the recycled succeeded this vdrop will actually free
801 		 * the vnode.  If not it will simply place it back on
802 		 * the free list.
803 		 */
804 		vdrop(vp);
805 		mtx_lock(&vnode_free_list_mtx);
806 	}
807 }
808 /*
809  * Attempt to recycle vnodes in a context that is always safe to block.
810  * Calling vlrurecycle() from the bowels of filesystem code has some
811  * interesting deadlock problems.
812  */
813 static struct proc *vnlruproc;
814 static int vnlruproc_sig;
815 
816 static void
817 vnlru_proc(void)
818 {
819 	struct mount *mp, *nmp;
820 	int done, vfslocked;
821 	struct proc *p = vnlruproc;
822 
823 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
824 	    SHUTDOWN_PRI_FIRST);
825 
826 	for (;;) {
827 		kproc_suspend_check(p);
828 		mtx_lock(&vnode_free_list_mtx);
829 		if (freevnodes > wantfreevnodes)
830 			vnlru_free(freevnodes - wantfreevnodes);
831 		if (numvnodes <= desiredvnodes * 9 / 10) {
832 			vnlruproc_sig = 0;
833 			wakeup(&vnlruproc_sig);
834 			msleep(vnlruproc, &vnode_free_list_mtx,
835 			    PVFS|PDROP, "vlruwt", hz);
836 			continue;
837 		}
838 		mtx_unlock(&vnode_free_list_mtx);
839 		done = 0;
840 		mtx_lock(&mountlist_mtx);
841 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
842 			if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
843 				nmp = TAILQ_NEXT(mp, mnt_list);
844 				continue;
845 			}
846 			vfslocked = VFS_LOCK_GIANT(mp);
847 			done += vlrureclaim(mp);
848 			VFS_UNLOCK_GIANT(vfslocked);
849 			mtx_lock(&mountlist_mtx);
850 			nmp = TAILQ_NEXT(mp, mnt_list);
851 			vfs_unbusy(mp);
852 		}
853 		mtx_unlock(&mountlist_mtx);
854 		if (done == 0) {
855 #if 0
856 			/* These messages are temporary debugging aids */
857 			if (vnlru_nowhere < 5)
858 				printf("vnlru process getting nowhere..\n");
859 			else if (vnlru_nowhere == 5)
860 				printf("vnlru process messages stopped.\n");
861 #endif
862 			vnlru_nowhere++;
863 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
864 		} else
865 			kern_yield(PRI_UNCHANGED);
866 	}
867 }
868 
869 static struct kproc_desc vnlru_kp = {
870 	"vnlru",
871 	vnlru_proc,
872 	&vnlruproc
873 };
874 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
875     &vnlru_kp);
876 
877 /*
878  * Routines having to do with the management of the vnode table.
879  */
880 
881 /*
882  * Try to recycle a freed vnode.  We abort if anyone picks up a reference
883  * before we actually vgone().  This function must be called with the vnode
884  * held to prevent the vnode from being returned to the free list midway
885  * through vgone().
886  */
887 static int
888 vtryrecycle(struct vnode *vp)
889 {
890 	struct mount *vnmp;
891 
892 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
893 	VNASSERT(vp->v_holdcnt, vp,
894 	    ("vtryrecycle: Recycling vp %p without a reference.", vp));
895 	/*
896 	 * This vnode may found and locked via some other list, if so we
897 	 * can't recycle it yet.
898 	 */
899 	if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
900 		CTR2(KTR_VFS,
901 		    "%s: impossible to recycle, vp %p lock is already held",
902 		    __func__, vp);
903 		return (EWOULDBLOCK);
904 	}
905 	/*
906 	 * Don't recycle if its filesystem is being suspended.
907 	 */
908 	if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
909 		VOP_UNLOCK(vp, 0);
910 		CTR2(KTR_VFS,
911 		    "%s: impossible to recycle, cannot start the write for %p",
912 		    __func__, vp);
913 		return (EBUSY);
914 	}
915 	/*
916 	 * If we got this far, we need to acquire the interlock and see if
917 	 * anyone picked up this vnode from another list.  If not, we will
918 	 * mark it with DOOMED via vgonel() so that anyone who does find it
919 	 * will skip over it.
920 	 */
921 	VI_LOCK(vp);
922 	if (vp->v_usecount) {
923 		VOP_UNLOCK(vp, LK_INTERLOCK);
924 		vn_finished_write(vnmp);
925 		CTR2(KTR_VFS,
926 		    "%s: impossible to recycle, %p is already referenced",
927 		    __func__, vp);
928 		return (EBUSY);
929 	}
930 	if ((vp->v_iflag & VI_DOOMED) == 0)
931 		vgonel(vp);
932 	VOP_UNLOCK(vp, LK_INTERLOCK);
933 	vn_finished_write(vnmp);
934 	return (0);
935 }
936 
937 /*
938  * Wait for available vnodes.
939  */
940 static int
941 getnewvnode_wait(int suspended)
942 {
943 
944 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
945 	if (numvnodes > desiredvnodes) {
946 		if (suspended) {
947 			/*
948 			 * File system is beeing suspended, we cannot risk a
949 			 * deadlock here, so allocate new vnode anyway.
950 			 */
951 			if (freevnodes > wantfreevnodes)
952 				vnlru_free(freevnodes - wantfreevnodes);
953 			return (0);
954 		}
955 		if (vnlruproc_sig == 0) {
956 			vnlruproc_sig = 1;	/* avoid unnecessary wakeups */
957 			wakeup(vnlruproc);
958 		}
959 		msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
960 		    "vlruwk", hz);
961 	}
962 	return (numvnodes > desiredvnodes ? ENFILE : 0);
963 }
964 
965 void
966 getnewvnode_reserve(u_int count)
967 {
968 	struct thread *td;
969 
970 	td = curthread;
971 	mtx_lock(&vnode_free_list_mtx);
972 	while (count > 0) {
973 		if (getnewvnode_wait(0) == 0) {
974 			count--;
975 			td->td_vp_reserv++;
976 			numvnodes++;
977 		}
978 	}
979 	mtx_unlock(&vnode_free_list_mtx);
980 }
981 
982 void
983 getnewvnode_drop_reserve(void)
984 {
985 	struct thread *td;
986 
987 	td = curthread;
988 	mtx_lock(&vnode_free_list_mtx);
989 	KASSERT(numvnodes >= td->td_vp_reserv, ("reserve too large"));
990 	numvnodes -= td->td_vp_reserv;
991 	mtx_unlock(&vnode_free_list_mtx);
992 	td->td_vp_reserv = 0;
993 }
994 
995 /*
996  * Return the next vnode from the free list.
997  */
998 int
999 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
1000     struct vnode **vpp)
1001 {
1002 	struct vnode *vp;
1003 	struct bufobj *bo;
1004 	struct thread *td;
1005 	int error;
1006 
1007 	CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
1008 	vp = NULL;
1009 	td = curthread;
1010 	if (td->td_vp_reserv > 0) {
1011 		td->td_vp_reserv -= 1;
1012 		goto alloc;
1013 	}
1014 	mtx_lock(&vnode_free_list_mtx);
1015 	/*
1016 	 * Lend our context to reclaim vnodes if they've exceeded the max.
1017 	 */
1018 	if (freevnodes > wantfreevnodes)
1019 		vnlru_free(1);
1020 	error = getnewvnode_wait(mp != NULL && (mp->mnt_kern_flag &
1021 	    MNTK_SUSPEND));
1022 #if 0	/* XXX Not all VFS_VGET/ffs_vget callers check returns. */
1023 	if (error != 0) {
1024 		mtx_unlock(&vnode_free_list_mtx);
1025 		return (error);
1026 	}
1027 #endif
1028 	numvnodes++;
1029 	mtx_unlock(&vnode_free_list_mtx);
1030 alloc:
1031 	vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
1032 	/*
1033 	 * Setup locks.
1034 	 */
1035 	vp->v_vnlock = &vp->v_lock;
1036 	mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
1037 	/*
1038 	 * By default, don't allow shared locks unless filesystems
1039 	 * opt-in.
1040 	 */
1041 	lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
1042 	/*
1043 	 * Initialize bufobj.
1044 	 */
1045 	bo = &vp->v_bufobj;
1046 	bo->__bo_vnode = vp;
1047 	mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
1048 	bo->bo_ops = &buf_ops_bio;
1049 	bo->bo_private = vp;
1050 	TAILQ_INIT(&bo->bo_clean.bv_hd);
1051 	TAILQ_INIT(&bo->bo_dirty.bv_hd);
1052 	/*
1053 	 * Initialize namecache.
1054 	 */
1055 	LIST_INIT(&vp->v_cache_src);
1056 	TAILQ_INIT(&vp->v_cache_dst);
1057 	/*
1058 	 * Finalize various vnode identity bits.
1059 	 */
1060 	vp->v_type = VNON;
1061 	vp->v_tag = tag;
1062 	vp->v_op = vops;
1063 	v_incr_usecount(vp);
1064 	vp->v_data = NULL;
1065 #ifdef MAC
1066 	mac_vnode_init(vp);
1067 	if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1068 		mac_vnode_associate_singlelabel(mp, vp);
1069 	else if (mp == NULL && vops != &dead_vnodeops)
1070 		printf("NULL mp in getnewvnode()\n");
1071 #endif
1072 	if (mp != NULL) {
1073 		bo->bo_bsize = mp->mnt_stat.f_iosize;
1074 		if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1075 			vp->v_vflag |= VV_NOKNOTE;
1076 	}
1077 	rangelock_init(&vp->v_rl);
1078 
1079 	*vpp = vp;
1080 	return (0);
1081 }
1082 
1083 /*
1084  * Delete from old mount point vnode list, if on one.
1085  */
1086 static void
1087 delmntque(struct vnode *vp)
1088 {
1089 	struct mount *mp;
1090 	int active;
1091 
1092 	mp = vp->v_mount;
1093 	if (mp == NULL)
1094 		return;
1095 	MNT_ILOCK(mp);
1096 	VI_LOCK(vp);
1097 	KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize,
1098 	    ("Active vnode list size %d > Vnode list size %d",
1099 	     mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize));
1100 	active = vp->v_iflag & VI_ACTIVE;
1101 	vp->v_iflag &= ~VI_ACTIVE;
1102 	if (active) {
1103 		mtx_lock(&vnode_free_list_mtx);
1104 		TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist);
1105 		mp->mnt_activevnodelistsize--;
1106 		mtx_unlock(&vnode_free_list_mtx);
1107 	}
1108 	vp->v_mount = NULL;
1109 	VI_UNLOCK(vp);
1110 	VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1111 		("bad mount point vnode list size"));
1112 	TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1113 	mp->mnt_nvnodelistsize--;
1114 	MNT_REL(mp);
1115 	MNT_IUNLOCK(mp);
1116 }
1117 
1118 static void
1119 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1120 {
1121 
1122 	vp->v_data = NULL;
1123 	vp->v_op = &dead_vnodeops;
1124 	/* XXX non mp-safe fs may still call insmntque with vnode
1125 	   unlocked */
1126 	if (!VOP_ISLOCKED(vp))
1127 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1128 	vgone(vp);
1129 	vput(vp);
1130 }
1131 
1132 /*
1133  * Insert into list of vnodes for the new mount point, if available.
1134  */
1135 int
1136 insmntque1(struct vnode *vp, struct mount *mp,
1137 	void (*dtr)(struct vnode *, void *), void *dtr_arg)
1138 {
1139 	int locked;
1140 
1141 	KASSERT(vp->v_mount == NULL,
1142 		("insmntque: vnode already on per mount vnode list"));
1143 	VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1144 #ifdef DEBUG_VFS_LOCKS
1145 	if (!VFS_NEEDSGIANT(mp))
1146 		ASSERT_VOP_ELOCKED(vp,
1147 		    "insmntque: mp-safe fs and non-locked vp");
1148 #endif
1149 	/*
1150 	 * We acquire the vnode interlock early to ensure that the
1151 	 * vnode cannot be recycled by another process releasing a
1152 	 * holdcnt on it before we get it on both the vnode list
1153 	 * and the active vnode list. The mount mutex protects only
1154 	 * manipulation of the vnode list and the vnode freelist
1155 	 * mutex protects only manipulation of the active vnode list.
1156 	 * Hence the need to hold the vnode interlock throughout.
1157 	 */
1158 	MNT_ILOCK(mp);
1159 	VI_LOCK(vp);
1160 	if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1161 	    ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1162 	     mp->mnt_nvnodelistsize == 0)) {
1163 		locked = VOP_ISLOCKED(vp);
1164 		if (!locked || (locked == LK_EXCLUSIVE &&
1165 		     (vp->v_vflag & VV_FORCEINSMQ) == 0)) {
1166 			VI_UNLOCK(vp);
1167 			MNT_IUNLOCK(mp);
1168 			if (dtr != NULL)
1169 				dtr(vp, dtr_arg);
1170 			return (EBUSY);
1171 		}
1172 	}
1173 	vp->v_mount = mp;
1174 	MNT_REF(mp);
1175 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1176 	VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1177 		("neg mount point vnode list size"));
1178 	mp->mnt_nvnodelistsize++;
1179 	KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
1180 	    ("Activating already active vnode"));
1181 	vp->v_iflag |= VI_ACTIVE;
1182 	mtx_lock(&vnode_free_list_mtx);
1183 	TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
1184 	mp->mnt_activevnodelistsize++;
1185 	mtx_unlock(&vnode_free_list_mtx);
1186 	VI_UNLOCK(vp);
1187 	MNT_IUNLOCK(mp);
1188 	return (0);
1189 }
1190 
1191 int
1192 insmntque(struct vnode *vp, struct mount *mp)
1193 {
1194 
1195 	return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1196 }
1197 
1198 /*
1199  * Flush out and invalidate all buffers associated with a bufobj
1200  * Called with the underlying object locked.
1201  */
1202 int
1203 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1204 {
1205 	int error;
1206 
1207 	BO_LOCK(bo);
1208 	if (flags & V_SAVE) {
1209 		error = bufobj_wwait(bo, slpflag, slptimeo);
1210 		if (error) {
1211 			BO_UNLOCK(bo);
1212 			return (error);
1213 		}
1214 		if (bo->bo_dirty.bv_cnt > 0) {
1215 			BO_UNLOCK(bo);
1216 			if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1217 				return (error);
1218 			/*
1219 			 * XXX We could save a lock/unlock if this was only
1220 			 * enabled under INVARIANTS
1221 			 */
1222 			BO_LOCK(bo);
1223 			if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1224 				panic("vinvalbuf: dirty bufs");
1225 		}
1226 	}
1227 	/*
1228 	 * If you alter this loop please notice that interlock is dropped and
1229 	 * reacquired in flushbuflist.  Special care is needed to ensure that
1230 	 * no race conditions occur from this.
1231 	 */
1232 	do {
1233 		error = flushbuflist(&bo->bo_clean,
1234 		    flags, bo, slpflag, slptimeo);
1235 		if (error == 0 && !(flags & V_CLEANONLY))
1236 			error = flushbuflist(&bo->bo_dirty,
1237 			    flags, bo, slpflag, slptimeo);
1238 		if (error != 0 && error != EAGAIN) {
1239 			BO_UNLOCK(bo);
1240 			return (error);
1241 		}
1242 	} while (error != 0);
1243 
1244 	/*
1245 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1246 	 * have write I/O in-progress but if there is a VM object then the
1247 	 * VM object can also have read-I/O in-progress.
1248 	 */
1249 	do {
1250 		bufobj_wwait(bo, 0, 0);
1251 		BO_UNLOCK(bo);
1252 		if (bo->bo_object != NULL) {
1253 			VM_OBJECT_LOCK(bo->bo_object);
1254 			vm_object_pip_wait(bo->bo_object, "bovlbx");
1255 			VM_OBJECT_UNLOCK(bo->bo_object);
1256 		}
1257 		BO_LOCK(bo);
1258 	} while (bo->bo_numoutput > 0);
1259 	BO_UNLOCK(bo);
1260 
1261 	/*
1262 	 * Destroy the copy in the VM cache, too.
1263 	 */
1264 	if (bo->bo_object != NULL &&
1265 	    (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) {
1266 		VM_OBJECT_LOCK(bo->bo_object);
1267 		vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
1268 		    OBJPR_CLEANONLY : 0);
1269 		VM_OBJECT_UNLOCK(bo->bo_object);
1270 	}
1271 
1272 #ifdef INVARIANTS
1273 	BO_LOCK(bo);
1274 	if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 &&
1275 	    (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1276 		panic("vinvalbuf: flush failed");
1277 	BO_UNLOCK(bo);
1278 #endif
1279 	return (0);
1280 }
1281 
1282 /*
1283  * Flush out and invalidate all buffers associated with a vnode.
1284  * Called with the underlying object locked.
1285  */
1286 int
1287 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1288 {
1289 
1290 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1291 	ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1292 	return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1293 }
1294 
1295 /*
1296  * Flush out buffers on the specified list.
1297  *
1298  */
1299 static int
1300 flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1301     int slptimeo)
1302 {
1303 	struct buf *bp, *nbp;
1304 	int retval, error;
1305 	daddr_t lblkno;
1306 	b_xflags_t xflags;
1307 
1308 	ASSERT_BO_LOCKED(bo);
1309 
1310 	retval = 0;
1311 	TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1312 		if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1313 		    ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1314 			continue;
1315 		}
1316 		lblkno = 0;
1317 		xflags = 0;
1318 		if (nbp != NULL) {
1319 			lblkno = nbp->b_lblkno;
1320 			xflags = nbp->b_xflags &
1321 				(BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1322 		}
1323 		retval = EAGAIN;
1324 		error = BUF_TIMELOCK(bp,
1325 		    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1326 		    "flushbuf", slpflag, slptimeo);
1327 		if (error) {
1328 			BO_LOCK(bo);
1329 			return (error != ENOLCK ? error : EAGAIN);
1330 		}
1331 		KASSERT(bp->b_bufobj == bo,
1332 		    ("bp %p wrong b_bufobj %p should be %p",
1333 		    bp, bp->b_bufobj, bo));
1334 		if (bp->b_bufobj != bo) {	/* XXX: necessary ? */
1335 			BUF_UNLOCK(bp);
1336 			BO_LOCK(bo);
1337 			return (EAGAIN);
1338 		}
1339 		/*
1340 		 * XXX Since there are no node locks for NFS, I
1341 		 * believe there is a slight chance that a delayed
1342 		 * write will occur while sleeping just above, so
1343 		 * check for it.
1344 		 */
1345 		if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1346 		    (flags & V_SAVE)) {
1347 			BO_LOCK(bo);
1348 			bremfree(bp);
1349 			BO_UNLOCK(bo);
1350 			bp->b_flags |= B_ASYNC;
1351 			bwrite(bp);
1352 			BO_LOCK(bo);
1353 			return (EAGAIN);	/* XXX: why not loop ? */
1354 		}
1355 		BO_LOCK(bo);
1356 		bremfree(bp);
1357 		BO_UNLOCK(bo);
1358 		bp->b_flags |= (B_INVAL | B_RELBUF);
1359 		bp->b_flags &= ~B_ASYNC;
1360 		brelse(bp);
1361 		BO_LOCK(bo);
1362 		if (nbp != NULL &&
1363 		    (nbp->b_bufobj != bo ||
1364 		     nbp->b_lblkno != lblkno ||
1365 		     (nbp->b_xflags &
1366 		      (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1367 			break;			/* nbp invalid */
1368 	}
1369 	return (retval);
1370 }
1371 
1372 /*
1373  * Truncate a file's buffer and pages to a specified length.  This
1374  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1375  * sync activity.
1376  */
1377 int
1378 vtruncbuf(struct vnode *vp, struct ucred *cred, off_t length, int blksize)
1379 {
1380 	struct buf *bp, *nbp;
1381 	int anyfreed;
1382 	int trunclbn;
1383 	struct bufobj *bo;
1384 
1385 	CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1386 	    vp, cred, blksize, (uintmax_t)length);
1387 
1388 	/*
1389 	 * Round up to the *next* lbn.
1390 	 */
1391 	trunclbn = (length + blksize - 1) / blksize;
1392 
1393 	ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1394 restart:
1395 	bo = &vp->v_bufobj;
1396 	BO_LOCK(bo);
1397 	anyfreed = 1;
1398 	for (;anyfreed;) {
1399 		anyfreed = 0;
1400 		TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1401 			if (bp->b_lblkno < trunclbn)
1402 				continue;
1403 			if (BUF_LOCK(bp,
1404 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1405 			    BO_MTX(bo)) == ENOLCK)
1406 				goto restart;
1407 
1408 			BO_LOCK(bo);
1409 			bremfree(bp);
1410 			BO_UNLOCK(bo);
1411 			bp->b_flags |= (B_INVAL | B_RELBUF);
1412 			bp->b_flags &= ~B_ASYNC;
1413 			brelse(bp);
1414 			anyfreed = 1;
1415 
1416 			BO_LOCK(bo);
1417 			if (nbp != NULL &&
1418 			    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1419 			    (nbp->b_vp != vp) ||
1420 			    (nbp->b_flags & B_DELWRI))) {
1421 				BO_UNLOCK(bo);
1422 				goto restart;
1423 			}
1424 		}
1425 
1426 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1427 			if (bp->b_lblkno < trunclbn)
1428 				continue;
1429 			if (BUF_LOCK(bp,
1430 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1431 			    BO_MTX(bo)) == ENOLCK)
1432 				goto restart;
1433 			BO_LOCK(bo);
1434 			bremfree(bp);
1435 			BO_UNLOCK(bo);
1436 			bp->b_flags |= (B_INVAL | B_RELBUF);
1437 			bp->b_flags &= ~B_ASYNC;
1438 			brelse(bp);
1439 			anyfreed = 1;
1440 
1441 			BO_LOCK(bo);
1442 			if (nbp != NULL &&
1443 			    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1444 			    (nbp->b_vp != vp) ||
1445 			    (nbp->b_flags & B_DELWRI) == 0)) {
1446 				BO_UNLOCK(bo);
1447 				goto restart;
1448 			}
1449 		}
1450 	}
1451 
1452 	if (length > 0) {
1453 restartsync:
1454 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1455 			if (bp->b_lblkno > 0)
1456 				continue;
1457 			/*
1458 			 * Since we hold the vnode lock this should only
1459 			 * fail if we're racing with the buf daemon.
1460 			 */
1461 			if (BUF_LOCK(bp,
1462 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1463 			    BO_MTX(bo)) == ENOLCK) {
1464 				goto restart;
1465 			}
1466 			VNASSERT((bp->b_flags & B_DELWRI), vp,
1467 			    ("buf(%p) on dirty queue without DELWRI", bp));
1468 
1469 			BO_LOCK(bo);
1470 			bremfree(bp);
1471 			BO_UNLOCK(bo);
1472 			bawrite(bp);
1473 			BO_LOCK(bo);
1474 			goto restartsync;
1475 		}
1476 	}
1477 
1478 	bufobj_wwait(bo, 0, 0);
1479 	BO_UNLOCK(bo);
1480 	vnode_pager_setsize(vp, length);
1481 
1482 	return (0);
1483 }
1484 
1485 /*
1486  * buf_splay() - splay tree core for the clean/dirty list of buffers in
1487  *		 a vnode.
1488  *
1489  *	NOTE: We have to deal with the special case of a background bitmap
1490  *	buffer, a situation where two buffers will have the same logical
1491  *	block offset.  We want (1) only the foreground buffer to be accessed
1492  *	in a lookup and (2) must differentiate between the foreground and
1493  *	background buffer in the splay tree algorithm because the splay
1494  *	tree cannot normally handle multiple entities with the same 'index'.
1495  *	We accomplish this by adding differentiating flags to the splay tree's
1496  *	numerical domain.
1497  */
1498 static
1499 struct buf *
1500 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1501 {
1502 	struct buf dummy;
1503 	struct buf *lefttreemax, *righttreemin, *y;
1504 
1505 	if (root == NULL)
1506 		return (NULL);
1507 	lefttreemax = righttreemin = &dummy;
1508 	for (;;) {
1509 		if (lblkno < root->b_lblkno ||
1510 		    (lblkno == root->b_lblkno &&
1511 		    (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1512 			if ((y = root->b_left) == NULL)
1513 				break;
1514 			if (lblkno < y->b_lblkno) {
1515 				/* Rotate right. */
1516 				root->b_left = y->b_right;
1517 				y->b_right = root;
1518 				root = y;
1519 				if ((y = root->b_left) == NULL)
1520 					break;
1521 			}
1522 			/* Link into the new root's right tree. */
1523 			righttreemin->b_left = root;
1524 			righttreemin = root;
1525 		} else if (lblkno > root->b_lblkno ||
1526 		    (lblkno == root->b_lblkno &&
1527 		    (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1528 			if ((y = root->b_right) == NULL)
1529 				break;
1530 			if (lblkno > y->b_lblkno) {
1531 				/* Rotate left. */
1532 				root->b_right = y->b_left;
1533 				y->b_left = root;
1534 				root = y;
1535 				if ((y = root->b_right) == NULL)
1536 					break;
1537 			}
1538 			/* Link into the new root's left tree. */
1539 			lefttreemax->b_right = root;
1540 			lefttreemax = root;
1541 		} else {
1542 			break;
1543 		}
1544 		root = y;
1545 	}
1546 	/* Assemble the new root. */
1547 	lefttreemax->b_right = root->b_left;
1548 	righttreemin->b_left = root->b_right;
1549 	root->b_left = dummy.b_right;
1550 	root->b_right = dummy.b_left;
1551 	return (root);
1552 }
1553 
1554 static void
1555 buf_vlist_remove(struct buf *bp)
1556 {
1557 	struct buf *root;
1558 	struct bufv *bv;
1559 
1560 	KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1561 	ASSERT_BO_LOCKED(bp->b_bufobj);
1562 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1563 	    (BX_VNDIRTY|BX_VNCLEAN),
1564 	    ("buf_vlist_remove: Buf %p is on two lists", bp));
1565 	if (bp->b_xflags & BX_VNDIRTY)
1566 		bv = &bp->b_bufobj->bo_dirty;
1567 	else
1568 		bv = &bp->b_bufobj->bo_clean;
1569 	if (bp != bv->bv_root) {
1570 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1571 		KASSERT(root == bp, ("splay lookup failed in remove"));
1572 	}
1573 	if (bp->b_left == NULL) {
1574 		root = bp->b_right;
1575 	} else {
1576 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1577 		root->b_right = bp->b_right;
1578 	}
1579 	bv->bv_root = root;
1580 	TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1581 	bv->bv_cnt--;
1582 	bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1583 }
1584 
1585 /*
1586  * Add the buffer to the sorted clean or dirty block list using a
1587  * splay tree algorithm.
1588  *
1589  * NOTE: xflags is passed as a constant, optimizing this inline function!
1590  */
1591 static void
1592 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1593 {
1594 	struct buf *root;
1595 	struct bufv *bv;
1596 
1597 	ASSERT_BO_LOCKED(bo);
1598 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1599 	    ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1600 	bp->b_xflags |= xflags;
1601 	if (xflags & BX_VNDIRTY)
1602 		bv = &bo->bo_dirty;
1603 	else
1604 		bv = &bo->bo_clean;
1605 
1606 	root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1607 	if (root == NULL) {
1608 		bp->b_left = NULL;
1609 		bp->b_right = NULL;
1610 		TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1611 	} else if (bp->b_lblkno < root->b_lblkno ||
1612 	    (bp->b_lblkno == root->b_lblkno &&
1613 	    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1614 		bp->b_left = root->b_left;
1615 		bp->b_right = root;
1616 		root->b_left = NULL;
1617 		TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1618 	} else {
1619 		bp->b_right = root->b_right;
1620 		bp->b_left = root;
1621 		root->b_right = NULL;
1622 		TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1623 	}
1624 	bv->bv_cnt++;
1625 	bv->bv_root = bp;
1626 }
1627 
1628 /*
1629  * Lookup a buffer using the splay tree.  Note that we specifically avoid
1630  * shadow buffers used in background bitmap writes.
1631  *
1632  * This code isn't quite efficient as it could be because we are maintaining
1633  * two sorted lists and do not know which list the block resides in.
1634  *
1635  * During a "make buildworld" the desired buffer is found at one of
1636  * the roots more than 60% of the time.  Thus, checking both roots
1637  * before performing either splay eliminates unnecessary splays on the
1638  * first tree splayed.
1639  */
1640 struct buf *
1641 gbincore(struct bufobj *bo, daddr_t lblkno)
1642 {
1643 	struct buf *bp;
1644 
1645 	ASSERT_BO_LOCKED(bo);
1646 	if ((bp = bo->bo_clean.bv_root) != NULL &&
1647 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1648 		return (bp);
1649 	if ((bp = bo->bo_dirty.bv_root) != NULL &&
1650 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1651 		return (bp);
1652 	if ((bp = bo->bo_clean.bv_root) != NULL) {
1653 		bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1654 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1655 			return (bp);
1656 	}
1657 	if ((bp = bo->bo_dirty.bv_root) != NULL) {
1658 		bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1659 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1660 			return (bp);
1661 	}
1662 	return (NULL);
1663 }
1664 
1665 /*
1666  * Associate a buffer with a vnode.
1667  */
1668 void
1669 bgetvp(struct vnode *vp, struct buf *bp)
1670 {
1671 	struct bufobj *bo;
1672 
1673 	bo = &vp->v_bufobj;
1674 	ASSERT_BO_LOCKED(bo);
1675 	VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1676 
1677 	CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1678 	VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1679 	    ("bgetvp: bp already attached! %p", bp));
1680 
1681 	vhold(vp);
1682 	if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT)
1683 		bp->b_flags |= B_NEEDSGIANT;
1684 	bp->b_vp = vp;
1685 	bp->b_bufobj = bo;
1686 	/*
1687 	 * Insert onto list for new vnode.
1688 	 */
1689 	buf_vlist_add(bp, bo, BX_VNCLEAN);
1690 }
1691 
1692 /*
1693  * Disassociate a buffer from a vnode.
1694  */
1695 void
1696 brelvp(struct buf *bp)
1697 {
1698 	struct bufobj *bo;
1699 	struct vnode *vp;
1700 
1701 	CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1702 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1703 
1704 	/*
1705 	 * Delete from old vnode list, if on one.
1706 	 */
1707 	vp = bp->b_vp;		/* XXX */
1708 	bo = bp->b_bufobj;
1709 	BO_LOCK(bo);
1710 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1711 		buf_vlist_remove(bp);
1712 	else
1713 		panic("brelvp: Buffer %p not on queue.", bp);
1714 	if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1715 		bo->bo_flag &= ~BO_ONWORKLST;
1716 		mtx_lock(&sync_mtx);
1717 		LIST_REMOVE(bo, bo_synclist);
1718 		syncer_worklist_len--;
1719 		mtx_unlock(&sync_mtx);
1720 	}
1721 	bp->b_flags &= ~B_NEEDSGIANT;
1722 	bp->b_vp = NULL;
1723 	bp->b_bufobj = NULL;
1724 	BO_UNLOCK(bo);
1725 	vdrop(vp);
1726 }
1727 
1728 /*
1729  * Add an item to the syncer work queue.
1730  */
1731 static void
1732 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1733 {
1734 	int queue, slot;
1735 
1736 	ASSERT_BO_LOCKED(bo);
1737 
1738 	mtx_lock(&sync_mtx);
1739 	if (bo->bo_flag & BO_ONWORKLST)
1740 		LIST_REMOVE(bo, bo_synclist);
1741 	else {
1742 		bo->bo_flag |= BO_ONWORKLST;
1743 		syncer_worklist_len++;
1744 	}
1745 
1746 	if (delay > syncer_maxdelay - 2)
1747 		delay = syncer_maxdelay - 2;
1748 	slot = (syncer_delayno + delay) & syncer_mask;
1749 
1750 	queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ :
1751 	    WI_MPSAFEQ;
1752 	LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo,
1753 	    bo_synclist);
1754 	mtx_unlock(&sync_mtx);
1755 }
1756 
1757 static int
1758 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1759 {
1760 	int error, len;
1761 
1762 	mtx_lock(&sync_mtx);
1763 	len = syncer_worklist_len - sync_vnode_count;
1764 	mtx_unlock(&sync_mtx);
1765 	error = SYSCTL_OUT(req, &len, sizeof(len));
1766 	return (error);
1767 }
1768 
1769 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1770     sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1771 
1772 static struct proc *updateproc;
1773 static void sched_sync(void);
1774 static struct kproc_desc up_kp = {
1775 	"syncer",
1776 	sched_sync,
1777 	&updateproc
1778 };
1779 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1780 
1781 static int
1782 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1783 {
1784 	struct vnode *vp;
1785 	struct mount *mp;
1786 
1787 	*bo = LIST_FIRST(slp);
1788 	if (*bo == NULL)
1789 		return (0);
1790 	vp = (*bo)->__bo_vnode;	/* XXX */
1791 	if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1792 		return (1);
1793 	/*
1794 	 * We use vhold in case the vnode does not
1795 	 * successfully sync.  vhold prevents the vnode from
1796 	 * going away when we unlock the sync_mtx so that
1797 	 * we can acquire the vnode interlock.
1798 	 */
1799 	vholdl(vp);
1800 	mtx_unlock(&sync_mtx);
1801 	VI_UNLOCK(vp);
1802 	if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1803 		vdrop(vp);
1804 		mtx_lock(&sync_mtx);
1805 		return (*bo == LIST_FIRST(slp));
1806 	}
1807 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1808 	(void) VOP_FSYNC(vp, MNT_LAZY, td);
1809 	VOP_UNLOCK(vp, 0);
1810 	vn_finished_write(mp);
1811 	BO_LOCK(*bo);
1812 	if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1813 		/*
1814 		 * Put us back on the worklist.  The worklist
1815 		 * routine will remove us from our current
1816 		 * position and then add us back in at a later
1817 		 * position.
1818 		 */
1819 		vn_syncer_add_to_worklist(*bo, syncdelay);
1820 	}
1821 	BO_UNLOCK(*bo);
1822 	vdrop(vp);
1823 	mtx_lock(&sync_mtx);
1824 	return (0);
1825 }
1826 
1827 /*
1828  * System filesystem synchronizer daemon.
1829  */
1830 static void
1831 sched_sync(void)
1832 {
1833 	struct synclist *gnext, *next;
1834 	struct synclist *gslp, *slp;
1835 	struct bufobj *bo;
1836 	long starttime;
1837 	struct thread *td = curthread;
1838 	int last_work_seen;
1839 	int net_worklist_len;
1840 	int syncer_final_iter;
1841 	int first_printf;
1842 	int error;
1843 
1844 	last_work_seen = 0;
1845 	syncer_final_iter = 0;
1846 	first_printf = 1;
1847 	syncer_state = SYNCER_RUNNING;
1848 	starttime = time_uptime;
1849 	td->td_pflags |= TDP_NORUNNINGBUF;
1850 
1851 	EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1852 	    SHUTDOWN_PRI_LAST);
1853 
1854 	mtx_lock(&sync_mtx);
1855 	for (;;) {
1856 		if (syncer_state == SYNCER_FINAL_DELAY &&
1857 		    syncer_final_iter == 0) {
1858 			mtx_unlock(&sync_mtx);
1859 			kproc_suspend_check(td->td_proc);
1860 			mtx_lock(&sync_mtx);
1861 		}
1862 		net_worklist_len = syncer_worklist_len - sync_vnode_count;
1863 		if (syncer_state != SYNCER_RUNNING &&
1864 		    starttime != time_uptime) {
1865 			if (first_printf) {
1866 				printf("\nSyncing disks, vnodes remaining...");
1867 				first_printf = 0;
1868 			}
1869 			printf("%d ", net_worklist_len);
1870 		}
1871 		starttime = time_uptime;
1872 
1873 		/*
1874 		 * Push files whose dirty time has expired.  Be careful
1875 		 * of interrupt race on slp queue.
1876 		 *
1877 		 * Skip over empty worklist slots when shutting down.
1878 		 */
1879 		do {
1880 			slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1881 			gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1882 			syncer_delayno += 1;
1883 			if (syncer_delayno == syncer_maxdelay)
1884 				syncer_delayno = 0;
1885 			next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1886 			gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1887 			/*
1888 			 * If the worklist has wrapped since the
1889 			 * it was emptied of all but syncer vnodes,
1890 			 * switch to the FINAL_DELAY state and run
1891 			 * for one more second.
1892 			 */
1893 			if (syncer_state == SYNCER_SHUTTING_DOWN &&
1894 			    net_worklist_len == 0 &&
1895 			    last_work_seen == syncer_delayno) {
1896 				syncer_state = SYNCER_FINAL_DELAY;
1897 				syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1898 			}
1899 		} while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1900 		    LIST_EMPTY(gslp) && syncer_worklist_len > 0);
1901 
1902 		/*
1903 		 * Keep track of the last time there was anything
1904 		 * on the worklist other than syncer vnodes.
1905 		 * Return to the SHUTTING_DOWN state if any
1906 		 * new work appears.
1907 		 */
1908 		if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1909 			last_work_seen = syncer_delayno;
1910 		if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1911 			syncer_state = SYNCER_SHUTTING_DOWN;
1912 		while (!LIST_EMPTY(slp)) {
1913 			error = sync_vnode(slp, &bo, td);
1914 			if (error == 1) {
1915 				LIST_REMOVE(bo, bo_synclist);
1916 				LIST_INSERT_HEAD(next, bo, bo_synclist);
1917 				continue;
1918 			}
1919 
1920 			if (first_printf == 0)
1921 				wdog_kern_pat(WD_LASTVAL);
1922 
1923 		}
1924 		if (!LIST_EMPTY(gslp)) {
1925 			mtx_unlock(&sync_mtx);
1926 			mtx_lock(&Giant);
1927 			mtx_lock(&sync_mtx);
1928 			while (!LIST_EMPTY(gslp)) {
1929 				error = sync_vnode(gslp, &bo, td);
1930 				if (error == 1) {
1931 					LIST_REMOVE(bo, bo_synclist);
1932 					LIST_INSERT_HEAD(gnext, bo,
1933 					    bo_synclist);
1934 					continue;
1935 				}
1936 			}
1937 			mtx_unlock(&Giant);
1938 		}
1939 		if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1940 			syncer_final_iter--;
1941 		/*
1942 		 * The variable rushjob allows the kernel to speed up the
1943 		 * processing of the filesystem syncer process. A rushjob
1944 		 * value of N tells the filesystem syncer to process the next
1945 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1946 		 * is used by the soft update code to speed up the filesystem
1947 		 * syncer process when the incore state is getting so far
1948 		 * ahead of the disk that the kernel memory pool is being
1949 		 * threatened with exhaustion.
1950 		 */
1951 		if (rushjob > 0) {
1952 			rushjob -= 1;
1953 			continue;
1954 		}
1955 		/*
1956 		 * Just sleep for a short period of time between
1957 		 * iterations when shutting down to allow some I/O
1958 		 * to happen.
1959 		 *
1960 		 * If it has taken us less than a second to process the
1961 		 * current work, then wait. Otherwise start right over
1962 		 * again. We can still lose time if any single round
1963 		 * takes more than two seconds, but it does not really
1964 		 * matter as we are just trying to generally pace the
1965 		 * filesystem activity.
1966 		 */
1967 		if (syncer_state != SYNCER_RUNNING ||
1968 		    time_uptime == starttime) {
1969 			thread_lock(td);
1970 			sched_prio(td, PPAUSE);
1971 			thread_unlock(td);
1972 		}
1973 		if (syncer_state != SYNCER_RUNNING)
1974 			cv_timedwait(&sync_wakeup, &sync_mtx,
1975 			    hz / SYNCER_SHUTDOWN_SPEEDUP);
1976 		else if (time_uptime == starttime)
1977 			cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1978 	}
1979 }
1980 
1981 /*
1982  * Request the syncer daemon to speed up its work.
1983  * We never push it to speed up more than half of its
1984  * normal turn time, otherwise it could take over the cpu.
1985  */
1986 int
1987 speedup_syncer(void)
1988 {
1989 	int ret = 0;
1990 
1991 	mtx_lock(&sync_mtx);
1992 	if (rushjob < syncdelay / 2) {
1993 		rushjob += 1;
1994 		stat_rush_requests += 1;
1995 		ret = 1;
1996 	}
1997 	mtx_unlock(&sync_mtx);
1998 	cv_broadcast(&sync_wakeup);
1999 	return (ret);
2000 }
2001 
2002 /*
2003  * Tell the syncer to speed up its work and run though its work
2004  * list several times, then tell it to shut down.
2005  */
2006 static void
2007 syncer_shutdown(void *arg, int howto)
2008 {
2009 
2010 	if (howto & RB_NOSYNC)
2011 		return;
2012 	mtx_lock(&sync_mtx);
2013 	syncer_state = SYNCER_SHUTTING_DOWN;
2014 	rushjob = 0;
2015 	mtx_unlock(&sync_mtx);
2016 	cv_broadcast(&sync_wakeup);
2017 	kproc_shutdown(arg, howto);
2018 }
2019 
2020 /*
2021  * Reassign a buffer from one vnode to another.
2022  * Used to assign file specific control information
2023  * (indirect blocks) to the vnode to which they belong.
2024  */
2025 void
2026 reassignbuf(struct buf *bp)
2027 {
2028 	struct vnode *vp;
2029 	struct bufobj *bo;
2030 	int delay;
2031 #ifdef INVARIANTS
2032 	struct bufv *bv;
2033 #endif
2034 
2035 	vp = bp->b_vp;
2036 	bo = bp->b_bufobj;
2037 	++reassignbufcalls;
2038 
2039 	CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
2040 	    bp, bp->b_vp, bp->b_flags);
2041 	/*
2042 	 * B_PAGING flagged buffers cannot be reassigned because their vp
2043 	 * is not fully linked in.
2044 	 */
2045 	if (bp->b_flags & B_PAGING)
2046 		panic("cannot reassign paging buffer");
2047 
2048 	/*
2049 	 * Delete from old vnode list, if on one.
2050 	 */
2051 	BO_LOCK(bo);
2052 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
2053 		buf_vlist_remove(bp);
2054 	else
2055 		panic("reassignbuf: Buffer %p not on queue.", bp);
2056 	/*
2057 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
2058 	 * of clean buffers.
2059 	 */
2060 	if (bp->b_flags & B_DELWRI) {
2061 		if ((bo->bo_flag & BO_ONWORKLST) == 0) {
2062 			switch (vp->v_type) {
2063 			case VDIR:
2064 				delay = dirdelay;
2065 				break;
2066 			case VCHR:
2067 				delay = metadelay;
2068 				break;
2069 			default:
2070 				delay = filedelay;
2071 			}
2072 			vn_syncer_add_to_worklist(bo, delay);
2073 		}
2074 		buf_vlist_add(bp, bo, BX_VNDIRTY);
2075 	} else {
2076 		buf_vlist_add(bp, bo, BX_VNCLEAN);
2077 
2078 		if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
2079 			mtx_lock(&sync_mtx);
2080 			LIST_REMOVE(bo, bo_synclist);
2081 			syncer_worklist_len--;
2082 			mtx_unlock(&sync_mtx);
2083 			bo->bo_flag &= ~BO_ONWORKLST;
2084 		}
2085 	}
2086 #ifdef INVARIANTS
2087 	bv = &bo->bo_clean;
2088 	bp = TAILQ_FIRST(&bv->bv_hd);
2089 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2090 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2091 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
2092 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2093 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2094 	bv = &bo->bo_dirty;
2095 	bp = TAILQ_FIRST(&bv->bv_hd);
2096 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2097 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2098 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
2099 	KASSERT(bp == NULL || bp->b_bufobj == bo,
2100 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2101 #endif
2102 	BO_UNLOCK(bo);
2103 }
2104 
2105 /*
2106  * Increment the use and hold counts on the vnode, taking care to reference
2107  * the driver's usecount if this is a chardev.  The vholdl() will remove
2108  * the vnode from the free list if it is presently free.  Requires the
2109  * vnode interlock and returns with it held.
2110  */
2111 static void
2112 v_incr_usecount(struct vnode *vp)
2113 {
2114 
2115 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2116 	vp->v_usecount++;
2117 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2118 		dev_lock();
2119 		vp->v_rdev->si_usecount++;
2120 		dev_unlock();
2121 	}
2122 	vholdl(vp);
2123 }
2124 
2125 /*
2126  * Turn a holdcnt into a use+holdcnt such that only one call to
2127  * v_decr_usecount is needed.
2128  */
2129 static void
2130 v_upgrade_usecount(struct vnode *vp)
2131 {
2132 
2133 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2134 	vp->v_usecount++;
2135 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2136 		dev_lock();
2137 		vp->v_rdev->si_usecount++;
2138 		dev_unlock();
2139 	}
2140 }
2141 
2142 /*
2143  * Decrement the vnode use and hold count along with the driver's usecount
2144  * if this is a chardev.  The vdropl() below releases the vnode interlock
2145  * as it may free the vnode.
2146  */
2147 static void
2148 v_decr_usecount(struct vnode *vp)
2149 {
2150 
2151 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2152 	VNASSERT(vp->v_usecount > 0, vp,
2153 	    ("v_decr_usecount: negative usecount"));
2154 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2155 	vp->v_usecount--;
2156 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2157 		dev_lock();
2158 		vp->v_rdev->si_usecount--;
2159 		dev_unlock();
2160 	}
2161 	vdropl(vp);
2162 }
2163 
2164 /*
2165  * Decrement only the use count and driver use count.  This is intended to
2166  * be paired with a follow on vdropl() to release the remaining hold count.
2167  * In this way we may vgone() a vnode with a 0 usecount without risk of
2168  * having it end up on a free list because the hold count is kept above 0.
2169  */
2170 static void
2171 v_decr_useonly(struct vnode *vp)
2172 {
2173 
2174 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2175 	VNASSERT(vp->v_usecount > 0, vp,
2176 	    ("v_decr_useonly: negative usecount"));
2177 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2178 	vp->v_usecount--;
2179 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2180 		dev_lock();
2181 		vp->v_rdev->si_usecount--;
2182 		dev_unlock();
2183 	}
2184 }
2185 
2186 /*
2187  * Grab a particular vnode from the free list, increment its
2188  * reference count and lock it.  VI_DOOMED is set if the vnode
2189  * is being destroyed.  Only callers who specify LK_RETRY will
2190  * see doomed vnodes.  If inactive processing was delayed in
2191  * vput try to do it here.
2192  */
2193 int
2194 vget(struct vnode *vp, int flags, struct thread *td)
2195 {
2196 	int error;
2197 
2198 	error = 0;
2199 	VFS_ASSERT_GIANT(vp->v_mount);
2200 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2201 	    ("vget: invalid lock operation"));
2202 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2203 
2204 	if ((flags & LK_INTERLOCK) == 0)
2205 		VI_LOCK(vp);
2206 	vholdl(vp);
2207 	if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2208 		vdrop(vp);
2209 		CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2210 		    vp);
2211 		return (error);
2212 	}
2213 	if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2214 		panic("vget: vn_lock failed to return ENOENT\n");
2215 	VI_LOCK(vp);
2216 	/* Upgrade our holdcnt to a usecount. */
2217 	v_upgrade_usecount(vp);
2218 	/*
2219 	 * We don't guarantee that any particular close will
2220 	 * trigger inactive processing so just make a best effort
2221 	 * here at preventing a reference to a removed file.  If
2222 	 * we don't succeed no harm is done.
2223 	 */
2224 	if (vp->v_iflag & VI_OWEINACT) {
2225 		if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2226 		    (flags & LK_NOWAIT) == 0)
2227 			vinactive(vp, td);
2228 		vp->v_iflag &= ~VI_OWEINACT;
2229 	}
2230 	VI_UNLOCK(vp);
2231 	return (0);
2232 }
2233 
2234 /*
2235  * Increase the reference count of a vnode.
2236  */
2237 void
2238 vref(struct vnode *vp)
2239 {
2240 
2241 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2242 	VI_LOCK(vp);
2243 	v_incr_usecount(vp);
2244 	VI_UNLOCK(vp);
2245 }
2246 
2247 /*
2248  * Return reference count of a vnode.
2249  *
2250  * The results of this call are only guaranteed when some mechanism other
2251  * than the VI lock is used to stop other processes from gaining references
2252  * to the vnode.  This may be the case if the caller holds the only reference.
2253  * This is also useful when stale data is acceptable as race conditions may
2254  * be accounted for by some other means.
2255  */
2256 int
2257 vrefcnt(struct vnode *vp)
2258 {
2259 	int usecnt;
2260 
2261 	VI_LOCK(vp);
2262 	usecnt = vp->v_usecount;
2263 	VI_UNLOCK(vp);
2264 
2265 	return (usecnt);
2266 }
2267 
2268 #define	VPUTX_VRELE	1
2269 #define	VPUTX_VPUT	2
2270 #define	VPUTX_VUNREF	3
2271 
2272 static void
2273 vputx(struct vnode *vp, int func)
2274 {
2275 	int error;
2276 
2277 	KASSERT(vp != NULL, ("vputx: null vp"));
2278 	if (func == VPUTX_VUNREF)
2279 		ASSERT_VOP_LOCKED(vp, "vunref");
2280 	else if (func == VPUTX_VPUT)
2281 		ASSERT_VOP_LOCKED(vp, "vput");
2282 	else
2283 		KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2284 	VFS_ASSERT_GIANT(vp->v_mount);
2285 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2286 	VI_LOCK(vp);
2287 
2288 	/* Skip this v_writecount check if we're going to panic below. */
2289 	VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2290 	    ("vputx: missed vn_close"));
2291 	error = 0;
2292 
2293 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2294 	    vp->v_usecount == 1)) {
2295 		if (func == VPUTX_VPUT)
2296 			VOP_UNLOCK(vp, 0);
2297 		v_decr_usecount(vp);
2298 		return;
2299 	}
2300 
2301 	if (vp->v_usecount != 1) {
2302 		vprint("vputx: negative ref count", vp);
2303 		panic("vputx: negative ref cnt");
2304 	}
2305 	CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2306 	/*
2307 	 * We want to hold the vnode until the inactive finishes to
2308 	 * prevent vgone() races.  We drop the use count here and the
2309 	 * hold count below when we're done.
2310 	 */
2311 	v_decr_useonly(vp);
2312 	/*
2313 	 * We must call VOP_INACTIVE with the node locked. Mark
2314 	 * as VI_DOINGINACT to avoid recursion.
2315 	 */
2316 	vp->v_iflag |= VI_OWEINACT;
2317 	switch (func) {
2318 	case VPUTX_VRELE:
2319 		error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2320 		VI_LOCK(vp);
2321 		break;
2322 	case VPUTX_VPUT:
2323 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2324 			error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
2325 			    LK_NOWAIT);
2326 			VI_LOCK(vp);
2327 		}
2328 		break;
2329 	case VPUTX_VUNREF:
2330 		if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
2331 			error = EBUSY;
2332 		break;
2333 	}
2334 	if (vp->v_usecount > 0)
2335 		vp->v_iflag &= ~VI_OWEINACT;
2336 	if (error == 0) {
2337 		if (vp->v_iflag & VI_OWEINACT)
2338 			vinactive(vp, curthread);
2339 		if (func != VPUTX_VUNREF)
2340 			VOP_UNLOCK(vp, 0);
2341 	}
2342 	vdropl(vp);
2343 }
2344 
2345 /*
2346  * Vnode put/release.
2347  * If count drops to zero, call inactive routine and return to freelist.
2348  */
2349 void
2350 vrele(struct vnode *vp)
2351 {
2352 
2353 	vputx(vp, VPUTX_VRELE);
2354 }
2355 
2356 /*
2357  * Release an already locked vnode.  This give the same effects as
2358  * unlock+vrele(), but takes less time and avoids releasing and
2359  * re-aquiring the lock (as vrele() acquires the lock internally.)
2360  */
2361 void
2362 vput(struct vnode *vp)
2363 {
2364 
2365 	vputx(vp, VPUTX_VPUT);
2366 }
2367 
2368 /*
2369  * Release an exclusively locked vnode. Do not unlock the vnode lock.
2370  */
2371 void
2372 vunref(struct vnode *vp)
2373 {
2374 
2375 	vputx(vp, VPUTX_VUNREF);
2376 }
2377 
2378 /*
2379  * Somebody doesn't want the vnode recycled.
2380  */
2381 void
2382 vhold(struct vnode *vp)
2383 {
2384 
2385 	VI_LOCK(vp);
2386 	vholdl(vp);
2387 	VI_UNLOCK(vp);
2388 }
2389 
2390 /*
2391  * Increase the hold count and activate if this is the first reference.
2392  */
2393 void
2394 vholdl(struct vnode *vp)
2395 {
2396 	struct mount *mp;
2397 
2398 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2399 	vp->v_holdcnt++;
2400 	if (!VSHOULDBUSY(vp))
2401 		return;
2402 	ASSERT_VI_LOCKED(vp, "vholdl");
2403 	VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
2404 	VNASSERT(vp->v_op != NULL, vp, ("vholdl: vnode already reclaimed."));
2405 	/*
2406 	 * Remove a vnode from the free list, mark it as in use,
2407 	 * and put it on the active list.
2408 	 */
2409 	mtx_lock(&vnode_free_list_mtx);
2410 	TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
2411 	freevnodes--;
2412 	vp->v_iflag &= ~(VI_FREE|VI_AGE);
2413 	KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
2414 	    ("Activating already active vnode"));
2415 	vp->v_iflag |= VI_ACTIVE;
2416 	mp = vp->v_mount;
2417 	TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
2418 	mp->mnt_activevnodelistsize++;
2419 	mtx_unlock(&vnode_free_list_mtx);
2420 }
2421 
2422 /*
2423  * Note that there is one less who cares about this vnode.
2424  * vdrop() is the opposite of vhold().
2425  */
2426 void
2427 vdrop(struct vnode *vp)
2428 {
2429 
2430 	VI_LOCK(vp);
2431 	vdropl(vp);
2432 }
2433 
2434 /*
2435  * Drop the hold count of the vnode.  If this is the last reference to
2436  * the vnode we place it on the free list unless it has been vgone'd
2437  * (marked VI_DOOMED) in which case we will free it.
2438  */
2439 void
2440 vdropl(struct vnode *vp)
2441 {
2442 	struct bufobj *bo;
2443 	struct mount *mp;
2444 	int active;
2445 
2446 	ASSERT_VI_LOCKED(vp, "vdropl");
2447 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2448 	if (vp->v_holdcnt <= 0)
2449 		panic("vdrop: holdcnt %d", vp->v_holdcnt);
2450 	vp->v_holdcnt--;
2451 	if (vp->v_holdcnt > 0) {
2452 		VI_UNLOCK(vp);
2453 		return;
2454 	}
2455 	if ((vp->v_iflag & VI_DOOMED) == 0) {
2456 		/*
2457 		 * Mark a vnode as free: remove it from its active list
2458 		 * and put it up for recycling on the freelist.
2459 		 */
2460 		VNASSERT(vp->v_op != NULL, vp,
2461 		    ("vdropl: vnode already reclaimed."));
2462 		VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2463 		    ("vnode already free"));
2464 		VNASSERT(VSHOULDFREE(vp), vp,
2465 		    ("vdropl: freeing when we shouldn't"));
2466 		active = vp->v_iflag & VI_ACTIVE;
2467 		vp->v_iflag &= ~VI_ACTIVE;
2468 		mp = vp->v_mount;
2469 		mtx_lock(&vnode_free_list_mtx);
2470 		if (active) {
2471 			TAILQ_REMOVE(&mp->mnt_activevnodelist, vp,
2472 			    v_actfreelist);
2473 			mp->mnt_activevnodelistsize--;
2474 		}
2475 		if (vp->v_iflag & VI_AGE) {
2476 			TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_actfreelist);
2477 		} else {
2478 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
2479 		}
2480 		freevnodes++;
2481 		vp->v_iflag &= ~VI_AGE;
2482 		vp->v_iflag |= VI_FREE;
2483 		mtx_unlock(&vnode_free_list_mtx);
2484 		VI_UNLOCK(vp);
2485 		return;
2486 	}
2487 	/*
2488 	 * The vnode has been marked for destruction, so free it.
2489 	 */
2490 	CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp);
2491 	mtx_lock(&vnode_free_list_mtx);
2492 	numvnodes--;
2493 	mtx_unlock(&vnode_free_list_mtx);
2494 	bo = &vp->v_bufobj;
2495 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2496 	    ("cleaned vnode still on the free list."));
2497 	VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
2498 	VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
2499 	VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
2500 	VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
2501 	VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
2502 	VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
2503 	VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
2504 	VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
2505 	VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
2506 	VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
2507 	VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
2508 	VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
2509 	VI_UNLOCK(vp);
2510 #ifdef MAC
2511 	mac_vnode_destroy(vp);
2512 #endif
2513 	if (vp->v_pollinfo != NULL)
2514 		destroy_vpollinfo(vp->v_pollinfo);
2515 #ifdef INVARIANTS
2516 	/* XXX Elsewhere we detect an already freed vnode via NULL v_op. */
2517 	vp->v_op = NULL;
2518 #endif
2519 	rangelock_destroy(&vp->v_rl);
2520 	lockdestroy(vp->v_vnlock);
2521 	mtx_destroy(&vp->v_interlock);
2522 	mtx_destroy(BO_MTX(bo));
2523 	uma_zfree(vnode_zone, vp);
2524 }
2525 
2526 /*
2527  * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2528  * flags.  DOINGINACT prevents us from recursing in calls to vinactive.
2529  * OWEINACT tracks whether a vnode missed a call to inactive due to a
2530  * failed lock upgrade.
2531  */
2532 void
2533 vinactive(struct vnode *vp, struct thread *td)
2534 {
2535 	struct vm_object *obj;
2536 
2537 	ASSERT_VOP_ELOCKED(vp, "vinactive");
2538 	ASSERT_VI_LOCKED(vp, "vinactive");
2539 	VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2540 	    ("vinactive: recursed on VI_DOINGINACT"));
2541 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2542 	vp->v_iflag |= VI_DOINGINACT;
2543 	vp->v_iflag &= ~VI_OWEINACT;
2544 	VI_UNLOCK(vp);
2545 	/*
2546 	 * Before moving off the active list, we must be sure that any
2547 	 * modified pages are on the vnode's dirty list since these will
2548 	 * no longer be checked once the vnode is on the inactive list.
2549 	 */
2550 	obj = vp->v_object;
2551 	if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
2552 		VM_OBJECT_LOCK(obj);
2553 		vm_object_page_clean(obj, 0, 0, OBJPC_NOSYNC);
2554 		VM_OBJECT_UNLOCK(obj);
2555 	}
2556 	VOP_INACTIVE(vp, td);
2557 	VI_LOCK(vp);
2558 	VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2559 	    ("vinactive: lost VI_DOINGINACT"));
2560 	vp->v_iflag &= ~VI_DOINGINACT;
2561 }
2562 
2563 /*
2564  * Remove any vnodes in the vnode table belonging to mount point mp.
2565  *
2566  * If FORCECLOSE is not specified, there should not be any active ones,
2567  * return error if any are found (nb: this is a user error, not a
2568  * system error). If FORCECLOSE is specified, detach any active vnodes
2569  * that are found.
2570  *
2571  * If WRITECLOSE is set, only flush out regular file vnodes open for
2572  * writing.
2573  *
2574  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2575  *
2576  * `rootrefs' specifies the base reference count for the root vnode
2577  * of this filesystem. The root vnode is considered busy if its
2578  * v_usecount exceeds this value. On a successful return, vflush(, td)
2579  * will call vrele() on the root vnode exactly rootrefs times.
2580  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2581  * be zero.
2582  */
2583 #ifdef DIAGNOSTIC
2584 static int busyprt = 0;		/* print out busy vnodes */
2585 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
2586 #endif
2587 
2588 int
2589 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
2590 {
2591 	struct vnode *vp, *mvp, *rootvp = NULL;
2592 	struct vattr vattr;
2593 	int busy = 0, error;
2594 
2595 	CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2596 	    rootrefs, flags);
2597 	if (rootrefs > 0) {
2598 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2599 		    ("vflush: bad args"));
2600 		/*
2601 		 * Get the filesystem root vnode. We can vput() it
2602 		 * immediately, since with rootrefs > 0, it won't go away.
2603 		 */
2604 		if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2605 			CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2606 			    __func__, error);
2607 			return (error);
2608 		}
2609 		vput(rootvp);
2610 	}
2611 loop:
2612 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
2613 		vholdl(vp);
2614 		error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2615 		if (error) {
2616 			vdrop(vp);
2617 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2618 			goto loop;
2619 		}
2620 		/*
2621 		 * Skip over a vnodes marked VV_SYSTEM.
2622 		 */
2623 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2624 			VOP_UNLOCK(vp, 0);
2625 			vdrop(vp);
2626 			continue;
2627 		}
2628 		/*
2629 		 * If WRITECLOSE is set, flush out unlinked but still open
2630 		 * files (even if open only for reading) and regular file
2631 		 * vnodes open for writing.
2632 		 */
2633 		if (flags & WRITECLOSE) {
2634 			if (vp->v_object != NULL) {
2635 				VM_OBJECT_LOCK(vp->v_object);
2636 				vm_object_page_clean(vp->v_object, 0, 0, 0);
2637 				VM_OBJECT_UNLOCK(vp->v_object);
2638 			}
2639 			error = VOP_FSYNC(vp, MNT_WAIT, td);
2640 			if (error != 0) {
2641 				VOP_UNLOCK(vp, 0);
2642 				vdrop(vp);
2643 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2644 				return (error);
2645 			}
2646 			error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2647 			VI_LOCK(vp);
2648 
2649 			if ((vp->v_type == VNON ||
2650 			    (error == 0 && vattr.va_nlink > 0)) &&
2651 			    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2652 				VOP_UNLOCK(vp, 0);
2653 				vdropl(vp);
2654 				continue;
2655 			}
2656 		} else
2657 			VI_LOCK(vp);
2658 		/*
2659 		 * With v_usecount == 0, all we need to do is clear out the
2660 		 * vnode data structures and we are done.
2661 		 *
2662 		 * If FORCECLOSE is set, forcibly close the vnode.
2663 		 */
2664 		if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2665 			VNASSERT(vp->v_usecount == 0 ||
2666 			    (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2667 			    ("device VNODE %p is FORCECLOSED", vp));
2668 			vgonel(vp);
2669 		} else {
2670 			busy++;
2671 #ifdef DIAGNOSTIC
2672 			if (busyprt)
2673 				vprint("vflush: busy vnode", vp);
2674 #endif
2675 		}
2676 		VOP_UNLOCK(vp, 0);
2677 		vdropl(vp);
2678 	}
2679 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2680 		/*
2681 		 * If just the root vnode is busy, and if its refcount
2682 		 * is equal to `rootrefs', then go ahead and kill it.
2683 		 */
2684 		VI_LOCK(rootvp);
2685 		KASSERT(busy > 0, ("vflush: not busy"));
2686 		VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2687 		    ("vflush: usecount %d < rootrefs %d",
2688 		     rootvp->v_usecount, rootrefs));
2689 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2690 			VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2691 			vgone(rootvp);
2692 			VOP_UNLOCK(rootvp, 0);
2693 			busy = 0;
2694 		} else
2695 			VI_UNLOCK(rootvp);
2696 	}
2697 	if (busy) {
2698 		CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2699 		    busy);
2700 		return (EBUSY);
2701 	}
2702 	for (; rootrefs > 0; rootrefs--)
2703 		vrele(rootvp);
2704 	return (0);
2705 }
2706 
2707 /*
2708  * Recycle an unused vnode to the front of the free list.
2709  */
2710 int
2711 vrecycle(struct vnode *vp)
2712 {
2713 	int recycled;
2714 
2715 	ASSERT_VOP_ELOCKED(vp, "vrecycle");
2716 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2717 	recycled = 0;
2718 	VI_LOCK(vp);
2719 	if (vp->v_usecount == 0) {
2720 		recycled = 1;
2721 		vgonel(vp);
2722 	}
2723 	VI_UNLOCK(vp);
2724 	return (recycled);
2725 }
2726 
2727 /*
2728  * Eliminate all activity associated with a vnode
2729  * in preparation for reuse.
2730  */
2731 void
2732 vgone(struct vnode *vp)
2733 {
2734 	VI_LOCK(vp);
2735 	vgonel(vp);
2736 	VI_UNLOCK(vp);
2737 }
2738 
2739 static void
2740 vgonel_reclaim_lowervp_vfs(struct mount *mp __unused,
2741     struct vnode *lowervp __unused)
2742 {
2743 }
2744 
2745 /*
2746  * Notify upper mounts about reclaimed vnode.
2747  */
2748 static void
2749 vgonel_reclaim_lowervp(struct vnode *vp)
2750 {
2751 	static struct vfsops vgonel_vfsops = {
2752 		.vfs_reclaim_lowervp = vgonel_reclaim_lowervp_vfs
2753 	};
2754 	struct mount *mp, *ump, *mmp;
2755 
2756 	mp = vp->v_mount;
2757 	if (mp == NULL)
2758 		return;
2759 
2760 	MNT_ILOCK(mp);
2761 	if (TAILQ_EMPTY(&mp->mnt_uppers))
2762 		goto unlock;
2763 	MNT_IUNLOCK(mp);
2764 	mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO);
2765 	mmp->mnt_op = &vgonel_vfsops;
2766 	mmp->mnt_kern_flag |= MNTK_MARKER;
2767 	MNT_ILOCK(mp);
2768 	mp->mnt_kern_flag |= MNTK_VGONE_UPPER;
2769 	for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) {
2770 		if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) {
2771 			ump = TAILQ_NEXT(ump, mnt_upper_link);
2772 			continue;
2773 		}
2774 		TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link);
2775 		MNT_IUNLOCK(mp);
2776 		VFS_RECLAIM_LOWERVP(ump, vp);
2777 		MNT_ILOCK(mp);
2778 		ump = TAILQ_NEXT(mmp, mnt_upper_link);
2779 		TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link);
2780 	}
2781 	free(mmp, M_TEMP);
2782 	mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER;
2783 	if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) {
2784 		mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER;
2785 		wakeup(&mp->mnt_uppers);
2786 	}
2787 unlock:
2788 	MNT_IUNLOCK(mp);
2789 }
2790 
2791 /*
2792  * vgone, with the vp interlock held.
2793  */
2794 void
2795 vgonel(struct vnode *vp)
2796 {
2797 	struct thread *td;
2798 	int oweinact;
2799 	int active;
2800 	struct mount *mp;
2801 
2802 	ASSERT_VOP_ELOCKED(vp, "vgonel");
2803 	ASSERT_VI_LOCKED(vp, "vgonel");
2804 	VNASSERT(vp->v_holdcnt, vp,
2805 	    ("vgonel: vp %p has no reference.", vp));
2806 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2807 	td = curthread;
2808 
2809 	/*
2810 	 * Don't vgonel if we're already doomed.
2811 	 */
2812 	if (vp->v_iflag & VI_DOOMED)
2813 		return;
2814 	vp->v_iflag |= VI_DOOMED;
2815 
2816 	/*
2817 	 * Check to see if the vnode is in use.  If so, we have to call
2818 	 * VOP_CLOSE() and VOP_INACTIVE().
2819 	 */
2820 	active = vp->v_usecount;
2821 	oweinact = (vp->v_iflag & VI_OWEINACT);
2822 	VI_UNLOCK(vp);
2823 	vgonel_reclaim_lowervp(vp);
2824 
2825 	/*
2826 	 * Clean out any buffers associated with the vnode.
2827 	 * If the flush fails, just toss the buffers.
2828 	 */
2829 	mp = NULL;
2830 	if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2831 		(void) vn_start_secondary_write(vp, &mp, V_WAIT);
2832 	if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
2833 		vinvalbuf(vp, 0, 0, 0);
2834 
2835 	/*
2836 	 * If purging an active vnode, it must be closed and
2837 	 * deactivated before being reclaimed.
2838 	 */
2839 	if (active)
2840 		VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2841 	if (oweinact || active) {
2842 		VI_LOCK(vp);
2843 		if ((vp->v_iflag & VI_DOINGINACT) == 0)
2844 			vinactive(vp, td);
2845 		VI_UNLOCK(vp);
2846 	}
2847 	if (vp->v_type == VSOCK)
2848 		vfs_unp_reclaim(vp);
2849 	/*
2850 	 * Reclaim the vnode.
2851 	 */
2852 	if (VOP_RECLAIM(vp, td))
2853 		panic("vgone: cannot reclaim");
2854 	if (mp != NULL)
2855 		vn_finished_secondary_write(mp);
2856 	VNASSERT(vp->v_object == NULL, vp,
2857 	    ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2858 	/*
2859 	 * Clear the advisory locks and wake up waiting threads.
2860 	 */
2861 	(void)VOP_ADVLOCKPURGE(vp);
2862 	/*
2863 	 * Delete from old mount point vnode list.
2864 	 */
2865 	delmntque(vp);
2866 	cache_purge(vp);
2867 	/*
2868 	 * Done with purge, reset to the standard lock and invalidate
2869 	 * the vnode.
2870 	 */
2871 	VI_LOCK(vp);
2872 	vp->v_vnlock = &vp->v_lock;
2873 	vp->v_op = &dead_vnodeops;
2874 	vp->v_tag = "none";
2875 	vp->v_type = VBAD;
2876 }
2877 
2878 /*
2879  * Calculate the total number of references to a special device.
2880  */
2881 int
2882 vcount(struct vnode *vp)
2883 {
2884 	int count;
2885 
2886 	dev_lock();
2887 	count = vp->v_rdev->si_usecount;
2888 	dev_unlock();
2889 	return (count);
2890 }
2891 
2892 /*
2893  * Same as above, but using the struct cdev *as argument
2894  */
2895 int
2896 count_dev(struct cdev *dev)
2897 {
2898 	int count;
2899 
2900 	dev_lock();
2901 	count = dev->si_usecount;
2902 	dev_unlock();
2903 	return(count);
2904 }
2905 
2906 /*
2907  * Print out a description of a vnode.
2908  */
2909 static char *typename[] =
2910 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2911  "VMARKER"};
2912 
2913 void
2914 vn_printf(struct vnode *vp, const char *fmt, ...)
2915 {
2916 	va_list ap;
2917 	char buf[256], buf2[16];
2918 	u_long flags;
2919 
2920 	va_start(ap, fmt);
2921 	vprintf(fmt, ap);
2922 	va_end(ap);
2923 	printf("%p: ", (void *)vp);
2924 	printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2925 	printf("    usecount %d, writecount %d, refcount %d mountedhere %p\n",
2926 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2927 	buf[0] = '\0';
2928 	buf[1] = '\0';
2929 	if (vp->v_vflag & VV_ROOT)
2930 		strlcat(buf, "|VV_ROOT", sizeof(buf));
2931 	if (vp->v_vflag & VV_ISTTY)
2932 		strlcat(buf, "|VV_ISTTY", sizeof(buf));
2933 	if (vp->v_vflag & VV_NOSYNC)
2934 		strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2935 	if (vp->v_vflag & VV_CACHEDLABEL)
2936 		strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2937 	if (vp->v_vflag & VV_TEXT)
2938 		strlcat(buf, "|VV_TEXT", sizeof(buf));
2939 	if (vp->v_vflag & VV_COPYONWRITE)
2940 		strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2941 	if (vp->v_vflag & VV_SYSTEM)
2942 		strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2943 	if (vp->v_vflag & VV_PROCDEP)
2944 		strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2945 	if (vp->v_vflag & VV_NOKNOTE)
2946 		strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2947 	if (vp->v_vflag & VV_DELETED)
2948 		strlcat(buf, "|VV_DELETED", sizeof(buf));
2949 	if (vp->v_vflag & VV_MD)
2950 		strlcat(buf, "|VV_MD", sizeof(buf));
2951 	flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC |
2952 	    VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2953 	    VV_NOKNOTE | VV_DELETED | VV_MD);
2954 	if (flags != 0) {
2955 		snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2956 		strlcat(buf, buf2, sizeof(buf));
2957 	}
2958 	if (vp->v_iflag & VI_MOUNT)
2959 		strlcat(buf, "|VI_MOUNT", sizeof(buf));
2960 	if (vp->v_iflag & VI_AGE)
2961 		strlcat(buf, "|VI_AGE", sizeof(buf));
2962 	if (vp->v_iflag & VI_DOOMED)
2963 		strlcat(buf, "|VI_DOOMED", sizeof(buf));
2964 	if (vp->v_iflag & VI_FREE)
2965 		strlcat(buf, "|VI_FREE", sizeof(buf));
2966 	if (vp->v_iflag & VI_DOINGINACT)
2967 		strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2968 	if (vp->v_iflag & VI_OWEINACT)
2969 		strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2970 	flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2971 	    VI_DOINGINACT | VI_OWEINACT);
2972 	if (flags != 0) {
2973 		snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2974 		strlcat(buf, buf2, sizeof(buf));
2975 	}
2976 	printf("    flags (%s)\n", buf + 1);
2977 	if (mtx_owned(VI_MTX(vp)))
2978 		printf(" VI_LOCKed");
2979 	if (vp->v_object != NULL)
2980 		printf("    v_object %p ref %d pages %d\n",
2981 		    vp->v_object, vp->v_object->ref_count,
2982 		    vp->v_object->resident_page_count);
2983 	printf("    ");
2984 	lockmgr_printinfo(vp->v_vnlock);
2985 	if (vp->v_data != NULL)
2986 		VOP_PRINT(vp);
2987 }
2988 
2989 #ifdef DDB
2990 /*
2991  * List all of the locked vnodes in the system.
2992  * Called when debugging the kernel.
2993  */
2994 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2995 {
2996 	struct mount *mp, *nmp;
2997 	struct vnode *vp;
2998 
2999 	/*
3000 	 * Note: because this is DDB, we can't obey the locking semantics
3001 	 * for these structures, which means we could catch an inconsistent
3002 	 * state and dereference a nasty pointer.  Not much to be done
3003 	 * about that.
3004 	 */
3005 	db_printf("Locked vnodes\n");
3006 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
3007 		nmp = TAILQ_NEXT(mp, mnt_list);
3008 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3009 			if (vp->v_type != VMARKER &&
3010 			    VOP_ISLOCKED(vp))
3011 				vprint("", vp);
3012 		}
3013 		nmp = TAILQ_NEXT(mp, mnt_list);
3014 	}
3015 }
3016 
3017 /*
3018  * Show details about the given vnode.
3019  */
3020 DB_SHOW_COMMAND(vnode, db_show_vnode)
3021 {
3022 	struct vnode *vp;
3023 
3024 	if (!have_addr)
3025 		return;
3026 	vp = (struct vnode *)addr;
3027 	vn_printf(vp, "vnode ");
3028 }
3029 
3030 /*
3031  * Show details about the given mount point.
3032  */
3033 DB_SHOW_COMMAND(mount, db_show_mount)
3034 {
3035 	struct mount *mp;
3036 	struct vfsopt *opt;
3037 	struct statfs *sp;
3038 	struct vnode *vp;
3039 	char buf[512];
3040 	uint64_t mflags;
3041 	u_int flags;
3042 
3043 	if (!have_addr) {
3044 		/* No address given, print short info about all mount points. */
3045 		TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3046 			db_printf("%p %s on %s (%s)\n", mp,
3047 			    mp->mnt_stat.f_mntfromname,
3048 			    mp->mnt_stat.f_mntonname,
3049 			    mp->mnt_stat.f_fstypename);
3050 			if (db_pager_quit)
3051 				break;
3052 		}
3053 		db_printf("\nMore info: show mount <addr>\n");
3054 		return;
3055 	}
3056 
3057 	mp = (struct mount *)addr;
3058 	db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
3059 	    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
3060 
3061 	buf[0] = '\0';
3062 	mflags = mp->mnt_flag;
3063 #define	MNT_FLAG(flag)	do {						\
3064 	if (mflags & (flag)) {						\
3065 		if (buf[0] != '\0')					\
3066 			strlcat(buf, ", ", sizeof(buf));		\
3067 		strlcat(buf, (#flag) + 4, sizeof(buf));			\
3068 		mflags &= ~(flag);					\
3069 	}								\
3070 } while (0)
3071 	MNT_FLAG(MNT_RDONLY);
3072 	MNT_FLAG(MNT_SYNCHRONOUS);
3073 	MNT_FLAG(MNT_NOEXEC);
3074 	MNT_FLAG(MNT_NOSUID);
3075 	MNT_FLAG(MNT_UNION);
3076 	MNT_FLAG(MNT_ASYNC);
3077 	MNT_FLAG(MNT_SUIDDIR);
3078 	MNT_FLAG(MNT_SOFTDEP);
3079 	MNT_FLAG(MNT_SUJ);
3080 	MNT_FLAG(MNT_NOSYMFOLLOW);
3081 	MNT_FLAG(MNT_GJOURNAL);
3082 	MNT_FLAG(MNT_MULTILABEL);
3083 	MNT_FLAG(MNT_ACLS);
3084 	MNT_FLAG(MNT_NOATIME);
3085 	MNT_FLAG(MNT_NOCLUSTERR);
3086 	MNT_FLAG(MNT_NOCLUSTERW);
3087 	MNT_FLAG(MNT_NFS4ACLS);
3088 	MNT_FLAG(MNT_EXRDONLY);
3089 	MNT_FLAG(MNT_EXPORTED);
3090 	MNT_FLAG(MNT_DEFEXPORTED);
3091 	MNT_FLAG(MNT_EXPORTANON);
3092 	MNT_FLAG(MNT_EXKERB);
3093 	MNT_FLAG(MNT_EXPUBLIC);
3094 	MNT_FLAG(MNT_LOCAL);
3095 	MNT_FLAG(MNT_QUOTA);
3096 	MNT_FLAG(MNT_ROOTFS);
3097 	MNT_FLAG(MNT_USER);
3098 	MNT_FLAG(MNT_IGNORE);
3099 	MNT_FLAG(MNT_UPDATE);
3100 	MNT_FLAG(MNT_DELEXPORT);
3101 	MNT_FLAG(MNT_RELOAD);
3102 	MNT_FLAG(MNT_FORCE);
3103 	MNT_FLAG(MNT_SNAPSHOT);
3104 	MNT_FLAG(MNT_BYFSID);
3105 #undef MNT_FLAG
3106 	if (mflags != 0) {
3107 		if (buf[0] != '\0')
3108 			strlcat(buf, ", ", sizeof(buf));
3109 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3110 		    "0x%016jx", mflags);
3111 	}
3112 	db_printf("    mnt_flag = %s\n", buf);
3113 
3114 	buf[0] = '\0';
3115 	flags = mp->mnt_kern_flag;
3116 #define	MNT_KERN_FLAG(flag)	do {					\
3117 	if (flags & (flag)) {						\
3118 		if (buf[0] != '\0')					\
3119 			strlcat(buf, ", ", sizeof(buf));		\
3120 		strlcat(buf, (#flag) + 5, sizeof(buf));			\
3121 		flags &= ~(flag);					\
3122 	}								\
3123 } while (0)
3124 	MNT_KERN_FLAG(MNTK_UNMOUNTF);
3125 	MNT_KERN_FLAG(MNTK_ASYNC);
3126 	MNT_KERN_FLAG(MNTK_SOFTDEP);
3127 	MNT_KERN_FLAG(MNTK_NOINSMNTQ);
3128 	MNT_KERN_FLAG(MNTK_DRAINING);
3129 	MNT_KERN_FLAG(MNTK_REFEXPIRE);
3130 	MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
3131 	MNT_KERN_FLAG(MNTK_SHARED_WRITES);
3132 	MNT_KERN_FLAG(MNTK_NOASYNC);
3133 	MNT_KERN_FLAG(MNTK_UNMOUNT);
3134 	MNT_KERN_FLAG(MNTK_MWAIT);
3135 	MNT_KERN_FLAG(MNTK_SUSPEND);
3136 	MNT_KERN_FLAG(MNTK_SUSPEND2);
3137 	MNT_KERN_FLAG(MNTK_SUSPENDED);
3138 	MNT_KERN_FLAG(MNTK_MPSAFE);
3139 	MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
3140 	MNT_KERN_FLAG(MNTK_NOKNOTE);
3141 #undef MNT_KERN_FLAG
3142 	if (flags != 0) {
3143 		if (buf[0] != '\0')
3144 			strlcat(buf, ", ", sizeof(buf));
3145 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3146 		    "0x%08x", flags);
3147 	}
3148 	db_printf("    mnt_kern_flag = %s\n", buf);
3149 
3150 	db_printf("    mnt_opt = ");
3151 	opt = TAILQ_FIRST(mp->mnt_opt);
3152 	if (opt != NULL) {
3153 		db_printf("%s", opt->name);
3154 		opt = TAILQ_NEXT(opt, link);
3155 		while (opt != NULL) {
3156 			db_printf(", %s", opt->name);
3157 			opt = TAILQ_NEXT(opt, link);
3158 		}
3159 	}
3160 	db_printf("\n");
3161 
3162 	sp = &mp->mnt_stat;
3163 	db_printf("    mnt_stat = { version=%u type=%u flags=0x%016jx "
3164 	    "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
3165 	    "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
3166 	    "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
3167 	    (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
3168 	    (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
3169 	    (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
3170 	    (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
3171 	    (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
3172 	    (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
3173 	    (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
3174 	    (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
3175 
3176 	db_printf("    mnt_cred = { uid=%u ruid=%u",
3177 	    (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
3178 	if (jailed(mp->mnt_cred))
3179 		db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
3180 	db_printf(" }\n");
3181 	db_printf("    mnt_ref = %d\n", mp->mnt_ref);
3182 	db_printf("    mnt_gen = %d\n", mp->mnt_gen);
3183 	db_printf("    mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
3184 	db_printf("    mnt_activevnodelistsize = %d\n",
3185 	    mp->mnt_activevnodelistsize);
3186 	db_printf("    mnt_writeopcount = %d\n", mp->mnt_writeopcount);
3187 	db_printf("    mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
3188 	db_printf("    mnt_iosize_max = %d\n", mp->mnt_iosize_max);
3189 	db_printf("    mnt_hashseed = %u\n", mp->mnt_hashseed);
3190 	db_printf("    mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
3191 	db_printf("    mnt_secondary_accwrites = %d\n",
3192 	    mp->mnt_secondary_accwrites);
3193 	db_printf("    mnt_gjprovider = %s\n",
3194 	    mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
3195 
3196 	db_printf("\n\nList of active vnodes\n");
3197 	TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) {
3198 		if (vp->v_type != VMARKER) {
3199 			vn_printf(vp, "vnode ");
3200 			if (db_pager_quit)
3201 				break;
3202 		}
3203 	}
3204 	db_printf("\n\nList of inactive vnodes\n");
3205 	TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3206 		if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) {
3207 			vn_printf(vp, "vnode ");
3208 			if (db_pager_quit)
3209 				break;
3210 		}
3211 	}
3212 }
3213 #endif	/* DDB */
3214 
3215 /*
3216  * Fill in a struct xvfsconf based on a struct vfsconf.
3217  */
3218 static int
3219 vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp)
3220 {
3221 	struct xvfsconf xvfsp;
3222 
3223 	bzero(&xvfsp, sizeof(xvfsp));
3224 	strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3225 	xvfsp.vfc_typenum = vfsp->vfc_typenum;
3226 	xvfsp.vfc_refcount = vfsp->vfc_refcount;
3227 	xvfsp.vfc_flags = vfsp->vfc_flags;
3228 	/*
3229 	 * These are unused in userland, we keep them
3230 	 * to not break binary compatibility.
3231 	 */
3232 	xvfsp.vfc_vfsops = NULL;
3233 	xvfsp.vfc_next = NULL;
3234 	return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3235 }
3236 
3237 #ifdef COMPAT_FREEBSD32
3238 struct xvfsconf32 {
3239 	uint32_t	vfc_vfsops;
3240 	char		vfc_name[MFSNAMELEN];
3241 	int32_t		vfc_typenum;
3242 	int32_t		vfc_refcount;
3243 	int32_t		vfc_flags;
3244 	uint32_t	vfc_next;
3245 };
3246 
3247 static int
3248 vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp)
3249 {
3250 	struct xvfsconf32 xvfsp;
3251 
3252 	strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3253 	xvfsp.vfc_typenum = vfsp->vfc_typenum;
3254 	xvfsp.vfc_refcount = vfsp->vfc_refcount;
3255 	xvfsp.vfc_flags = vfsp->vfc_flags;
3256 	xvfsp.vfc_vfsops = 0;
3257 	xvfsp.vfc_next = 0;
3258 	return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3259 }
3260 #endif
3261 
3262 /*
3263  * Top level filesystem related information gathering.
3264  */
3265 static int
3266 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
3267 {
3268 	struct vfsconf *vfsp;
3269 	int error;
3270 
3271 	error = 0;
3272 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3273 #ifdef COMPAT_FREEBSD32
3274 		if (req->flags & SCTL_MASK32)
3275 			error = vfsconf2x32(req, vfsp);
3276 		else
3277 #endif
3278 			error = vfsconf2x(req, vfsp);
3279 		if (error)
3280 			break;
3281 	}
3282 	return (error);
3283 }
3284 
3285 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD,
3286     NULL, 0, sysctl_vfs_conflist,
3287     "S,xvfsconf", "List of all configured filesystems");
3288 
3289 #ifndef BURN_BRIDGES
3290 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
3291 
3292 static int
3293 vfs_sysctl(SYSCTL_HANDLER_ARGS)
3294 {
3295 	int *name = (int *)arg1 - 1;	/* XXX */
3296 	u_int namelen = arg2 + 1;	/* XXX */
3297 	struct vfsconf *vfsp;
3298 
3299 	log(LOG_WARNING, "userland calling deprecated sysctl, "
3300 	    "please rebuild world\n");
3301 
3302 #if 1 || defined(COMPAT_PRELITE2)
3303 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3304 	if (namelen == 1)
3305 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3306 #endif
3307 
3308 	switch (name[1]) {
3309 	case VFS_MAXTYPENUM:
3310 		if (namelen != 2)
3311 			return (ENOTDIR);
3312 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3313 	case VFS_CONF:
3314 		if (namelen != 3)
3315 			return (ENOTDIR);	/* overloaded */
3316 		TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
3317 			if (vfsp->vfc_typenum == name[2])
3318 				break;
3319 		if (vfsp == NULL)
3320 			return (EOPNOTSUPP);
3321 #ifdef COMPAT_FREEBSD32
3322 		if (req->flags & SCTL_MASK32)
3323 			return (vfsconf2x32(req, vfsp));
3324 		else
3325 #endif
3326 			return (vfsconf2x(req, vfsp));
3327 	}
3328 	return (EOPNOTSUPP);
3329 }
3330 
3331 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
3332     vfs_sysctl, "Generic filesystem");
3333 
3334 #if 1 || defined(COMPAT_PRELITE2)
3335 
3336 static int
3337 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3338 {
3339 	int error;
3340 	struct vfsconf *vfsp;
3341 	struct ovfsconf ovfs;
3342 
3343 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3344 		bzero(&ovfs, sizeof(ovfs));
3345 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
3346 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
3347 		ovfs.vfc_index = vfsp->vfc_typenum;
3348 		ovfs.vfc_refcount = vfsp->vfc_refcount;
3349 		ovfs.vfc_flags = vfsp->vfc_flags;
3350 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3351 		if (error)
3352 			return error;
3353 	}
3354 	return 0;
3355 }
3356 
3357 #endif /* 1 || COMPAT_PRELITE2 */
3358 #endif /* !BURN_BRIDGES */
3359 
3360 #define KINFO_VNODESLOP		10
3361 #ifdef notyet
3362 /*
3363  * Dump vnode list (via sysctl).
3364  */
3365 /* ARGSUSED */
3366 static int
3367 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3368 {
3369 	struct xvnode *xvn;
3370 	struct mount *mp;
3371 	struct vnode *vp;
3372 	int error, len, n;
3373 
3374 	/*
3375 	 * Stale numvnodes access is not fatal here.
3376 	 */
3377 	req->lock = 0;
3378 	len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3379 	if (!req->oldptr)
3380 		/* Make an estimate */
3381 		return (SYSCTL_OUT(req, 0, len));
3382 
3383 	error = sysctl_wire_old_buffer(req, 0);
3384 	if (error != 0)
3385 		return (error);
3386 	xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3387 	n = 0;
3388 	mtx_lock(&mountlist_mtx);
3389 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3390 		if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3391 			continue;
3392 		MNT_ILOCK(mp);
3393 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3394 			if (n == len)
3395 				break;
3396 			vref(vp);
3397 			xvn[n].xv_size = sizeof *xvn;
3398 			xvn[n].xv_vnode = vp;
3399 			xvn[n].xv_id = 0;	/* XXX compat */
3400 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3401 			XV_COPY(usecount);
3402 			XV_COPY(writecount);
3403 			XV_COPY(holdcnt);
3404 			XV_COPY(mount);
3405 			XV_COPY(numoutput);
3406 			XV_COPY(type);
3407 #undef XV_COPY
3408 			xvn[n].xv_flag = vp->v_vflag;
3409 
3410 			switch (vp->v_type) {
3411 			case VREG:
3412 			case VDIR:
3413 			case VLNK:
3414 				break;
3415 			case VBLK:
3416 			case VCHR:
3417 				if (vp->v_rdev == NULL) {
3418 					vrele(vp);
3419 					continue;
3420 				}
3421 				xvn[n].xv_dev = dev2udev(vp->v_rdev);
3422 				break;
3423 			case VSOCK:
3424 				xvn[n].xv_socket = vp->v_socket;
3425 				break;
3426 			case VFIFO:
3427 				xvn[n].xv_fifo = vp->v_fifoinfo;
3428 				break;
3429 			case VNON:
3430 			case VBAD:
3431 			default:
3432 				/* shouldn't happen? */
3433 				vrele(vp);
3434 				continue;
3435 			}
3436 			vrele(vp);
3437 			++n;
3438 		}
3439 		MNT_IUNLOCK(mp);
3440 		mtx_lock(&mountlist_mtx);
3441 		vfs_unbusy(mp);
3442 		if (n == len)
3443 			break;
3444 	}
3445 	mtx_unlock(&mountlist_mtx);
3446 
3447 	error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3448 	free(xvn, M_TEMP);
3449 	return (error);
3450 }
3451 
3452 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3453     0, 0, sysctl_vnode, "S,xvnode", "");
3454 #endif
3455 
3456 /*
3457  * Unmount all filesystems. The list is traversed in reverse order
3458  * of mounting to avoid dependencies.
3459  */
3460 void
3461 vfs_unmountall(void)
3462 {
3463 	struct mount *mp;
3464 	struct thread *td;
3465 	int error;
3466 
3467 	CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3468 	td = curthread;
3469 
3470 	/*
3471 	 * Since this only runs when rebooting, it is not interlocked.
3472 	 */
3473 	while(!TAILQ_EMPTY(&mountlist)) {
3474 		mp = TAILQ_LAST(&mountlist, mntlist);
3475 		error = dounmount(mp, MNT_FORCE, td);
3476 		if (error) {
3477 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
3478 			/*
3479 			 * XXX: Due to the way in which we mount the root
3480 			 * file system off of devfs, devfs will generate a
3481 			 * "busy" warning when we try to unmount it before
3482 			 * the root.  Don't print a warning as a result in
3483 			 * order to avoid false positive errors that may
3484 			 * cause needless upset.
3485 			 */
3486 			if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3487 				printf("unmount of %s failed (",
3488 				    mp->mnt_stat.f_mntonname);
3489 				if (error == EBUSY)
3490 					printf("BUSY)\n");
3491 				else
3492 					printf("%d)\n", error);
3493 			}
3494 		} else {
3495 			/* The unmount has removed mp from the mountlist */
3496 		}
3497 	}
3498 }
3499 
3500 /*
3501  * perform msync on all vnodes under a mount point
3502  * the mount point must be locked.
3503  */
3504 void
3505 vfs_msync(struct mount *mp, int flags)
3506 {
3507 	struct vnode *vp, *mvp;
3508 	struct vm_object *obj;
3509 
3510 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3511 	MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
3512 		obj = vp->v_object;
3513 		if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3514 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3515 			if (!vget(vp,
3516 			    LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3517 			    curthread)) {
3518 				if (vp->v_vflag & VV_NOSYNC) {	/* unlinked */
3519 					vput(vp);
3520 					continue;
3521 				}
3522 
3523 				obj = vp->v_object;
3524 				if (obj != NULL) {
3525 					VM_OBJECT_LOCK(obj);
3526 					vm_object_page_clean(obj, 0, 0,
3527 					    flags == MNT_WAIT ?
3528 					    OBJPC_SYNC : OBJPC_NOSYNC);
3529 					VM_OBJECT_UNLOCK(obj);
3530 				}
3531 				vput(vp);
3532 			}
3533 		} else
3534 			VI_UNLOCK(vp);
3535 	}
3536 }
3537 
3538 static void
3539 destroy_vpollinfo(struct vpollinfo *vi)
3540 {
3541 	seldrain(&vi->vpi_selinfo);
3542 	knlist_destroy(&vi->vpi_selinfo.si_note);
3543 	mtx_destroy(&vi->vpi_lock);
3544 	uma_zfree(vnodepoll_zone, vi);
3545 }
3546 
3547 /*
3548  * Initalize per-vnode helper structure to hold poll-related state.
3549  */
3550 void
3551 v_addpollinfo(struct vnode *vp)
3552 {
3553 	struct vpollinfo *vi;
3554 
3555 	if (vp->v_pollinfo != NULL)
3556 		return;
3557 	vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3558 	mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3559 	knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3560 	    vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3561 	VI_LOCK(vp);
3562 	if (vp->v_pollinfo != NULL) {
3563 		VI_UNLOCK(vp);
3564 		destroy_vpollinfo(vi);
3565 		return;
3566 	}
3567 	vp->v_pollinfo = vi;
3568 	VI_UNLOCK(vp);
3569 }
3570 
3571 /*
3572  * Record a process's interest in events which might happen to
3573  * a vnode.  Because poll uses the historic select-style interface
3574  * internally, this routine serves as both the ``check for any
3575  * pending events'' and the ``record my interest in future events''
3576  * functions.  (These are done together, while the lock is held,
3577  * to avoid race conditions.)
3578  */
3579 int
3580 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3581 {
3582 
3583 	v_addpollinfo(vp);
3584 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3585 	if (vp->v_pollinfo->vpi_revents & events) {
3586 		/*
3587 		 * This leaves events we are not interested
3588 		 * in available for the other process which
3589 		 * which presumably had requested them
3590 		 * (otherwise they would never have been
3591 		 * recorded).
3592 		 */
3593 		events &= vp->v_pollinfo->vpi_revents;
3594 		vp->v_pollinfo->vpi_revents &= ~events;
3595 
3596 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
3597 		return (events);
3598 	}
3599 	vp->v_pollinfo->vpi_events |= events;
3600 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3601 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3602 	return (0);
3603 }
3604 
3605 /*
3606  * Routine to create and manage a filesystem syncer vnode.
3607  */
3608 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
3609 static int	sync_fsync(struct  vop_fsync_args *);
3610 static int	sync_inactive(struct  vop_inactive_args *);
3611 static int	sync_reclaim(struct  vop_reclaim_args *);
3612 
3613 static struct vop_vector sync_vnodeops = {
3614 	.vop_bypass =	VOP_EOPNOTSUPP,
3615 	.vop_close =	sync_close,		/* close */
3616 	.vop_fsync =	sync_fsync,		/* fsync */
3617 	.vop_inactive =	sync_inactive,	/* inactive */
3618 	.vop_reclaim =	sync_reclaim,	/* reclaim */
3619 	.vop_lock1 =	vop_stdlock,	/* lock */
3620 	.vop_unlock =	vop_stdunlock,	/* unlock */
3621 	.vop_islocked =	vop_stdislocked,	/* islocked */
3622 };
3623 
3624 /*
3625  * Create a new filesystem syncer vnode for the specified mount point.
3626  */
3627 void
3628 vfs_allocate_syncvnode(struct mount *mp)
3629 {
3630 	struct vnode *vp;
3631 	struct bufobj *bo;
3632 	static long start, incr, next;
3633 	int error;
3634 
3635 	/* Allocate a new vnode */
3636 	error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
3637 	if (error != 0)
3638 		panic("vfs_allocate_syncvnode: getnewvnode() failed");
3639 	vp->v_type = VNON;
3640 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3641 	vp->v_vflag |= VV_FORCEINSMQ;
3642 	error = insmntque(vp, mp);
3643 	if (error != 0)
3644 		panic("vfs_allocate_syncvnode: insmntque() failed");
3645 	vp->v_vflag &= ~VV_FORCEINSMQ;
3646 	VOP_UNLOCK(vp, 0);
3647 	/*
3648 	 * Place the vnode onto the syncer worklist. We attempt to
3649 	 * scatter them about on the list so that they will go off
3650 	 * at evenly distributed times even if all the filesystems
3651 	 * are mounted at once.
3652 	 */
3653 	next += incr;
3654 	if (next == 0 || next > syncer_maxdelay) {
3655 		start /= 2;
3656 		incr /= 2;
3657 		if (start == 0) {
3658 			start = syncer_maxdelay / 2;
3659 			incr = syncer_maxdelay;
3660 		}
3661 		next = start;
3662 	}
3663 	bo = &vp->v_bufobj;
3664 	BO_LOCK(bo);
3665 	vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3666 	/* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3667 	mtx_lock(&sync_mtx);
3668 	sync_vnode_count++;
3669 	if (mp->mnt_syncer == NULL) {
3670 		mp->mnt_syncer = vp;
3671 		vp = NULL;
3672 	}
3673 	mtx_unlock(&sync_mtx);
3674 	BO_UNLOCK(bo);
3675 	if (vp != NULL) {
3676 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3677 		vgone(vp);
3678 		vput(vp);
3679 	}
3680 }
3681 
3682 void
3683 vfs_deallocate_syncvnode(struct mount *mp)
3684 {
3685 	struct vnode *vp;
3686 
3687 	mtx_lock(&sync_mtx);
3688 	vp = mp->mnt_syncer;
3689 	if (vp != NULL)
3690 		mp->mnt_syncer = NULL;
3691 	mtx_unlock(&sync_mtx);
3692 	if (vp != NULL)
3693 		vrele(vp);
3694 }
3695 
3696 /*
3697  * Do a lazy sync of the filesystem.
3698  */
3699 static int
3700 sync_fsync(struct vop_fsync_args *ap)
3701 {
3702 	struct vnode *syncvp = ap->a_vp;
3703 	struct mount *mp = syncvp->v_mount;
3704 	int error, save;
3705 	struct bufobj *bo;
3706 
3707 	/*
3708 	 * We only need to do something if this is a lazy evaluation.
3709 	 */
3710 	if (ap->a_waitfor != MNT_LAZY)
3711 		return (0);
3712 
3713 	/*
3714 	 * Move ourselves to the back of the sync list.
3715 	 */
3716 	bo = &syncvp->v_bufobj;
3717 	BO_LOCK(bo);
3718 	vn_syncer_add_to_worklist(bo, syncdelay);
3719 	BO_UNLOCK(bo);
3720 
3721 	/*
3722 	 * Walk the list of vnodes pushing all that are dirty and
3723 	 * not already on the sync list.
3724 	 */
3725 	mtx_lock(&mountlist_mtx);
3726 	if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) {
3727 		mtx_unlock(&mountlist_mtx);
3728 		return (0);
3729 	}
3730 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3731 		vfs_unbusy(mp);
3732 		return (0);
3733 	}
3734 	save = curthread_pflags_set(TDP_SYNCIO);
3735 	vfs_msync(mp, MNT_NOWAIT);
3736 	error = VFS_SYNC(mp, MNT_LAZY);
3737 	curthread_pflags_restore(save);
3738 	vn_finished_write(mp);
3739 	vfs_unbusy(mp);
3740 	return (error);
3741 }
3742 
3743 /*
3744  * The syncer vnode is no referenced.
3745  */
3746 static int
3747 sync_inactive(struct vop_inactive_args *ap)
3748 {
3749 
3750 	vgone(ap->a_vp);
3751 	return (0);
3752 }
3753 
3754 /*
3755  * The syncer vnode is no longer needed and is being decommissioned.
3756  *
3757  * Modifications to the worklist must be protected by sync_mtx.
3758  */
3759 static int
3760 sync_reclaim(struct vop_reclaim_args *ap)
3761 {
3762 	struct vnode *vp = ap->a_vp;
3763 	struct bufobj *bo;
3764 
3765 	bo = &vp->v_bufobj;
3766 	BO_LOCK(bo);
3767 	mtx_lock(&sync_mtx);
3768 	if (vp->v_mount->mnt_syncer == vp)
3769 		vp->v_mount->mnt_syncer = NULL;
3770 	if (bo->bo_flag & BO_ONWORKLST) {
3771 		LIST_REMOVE(bo, bo_synclist);
3772 		syncer_worklist_len--;
3773 		sync_vnode_count--;
3774 		bo->bo_flag &= ~BO_ONWORKLST;
3775 	}
3776 	mtx_unlock(&sync_mtx);
3777 	BO_UNLOCK(bo);
3778 
3779 	return (0);
3780 }
3781 
3782 /*
3783  * Check if vnode represents a disk device
3784  */
3785 int
3786 vn_isdisk(struct vnode *vp, int *errp)
3787 {
3788 	int error;
3789 
3790 	error = 0;
3791 	dev_lock();
3792 	if (vp->v_type != VCHR)
3793 		error = ENOTBLK;
3794 	else if (vp->v_rdev == NULL)
3795 		error = ENXIO;
3796 	else if (vp->v_rdev->si_devsw == NULL)
3797 		error = ENXIO;
3798 	else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3799 		error = ENOTBLK;
3800 	dev_unlock();
3801 	if (errp != NULL)
3802 		*errp = error;
3803 	return (error == 0);
3804 }
3805 
3806 /*
3807  * Common filesystem object access control check routine.  Accepts a
3808  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3809  * and optional call-by-reference privused argument allowing vaccess()
3810  * to indicate to the caller whether privilege was used to satisfy the
3811  * request (obsoleted).  Returns 0 on success, or an errno on failure.
3812  */
3813 int
3814 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3815     accmode_t accmode, struct ucred *cred, int *privused)
3816 {
3817 	accmode_t dac_granted;
3818 	accmode_t priv_granted;
3819 
3820 	KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3821 	    ("invalid bit in accmode"));
3822 	KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
3823 	    ("VAPPEND without VWRITE"));
3824 
3825 	/*
3826 	 * Look for a normal, non-privileged way to access the file/directory
3827 	 * as requested.  If it exists, go with that.
3828 	 */
3829 
3830 	if (privused != NULL)
3831 		*privused = 0;
3832 
3833 	dac_granted = 0;
3834 
3835 	/* Check the owner. */
3836 	if (cred->cr_uid == file_uid) {
3837 		dac_granted |= VADMIN;
3838 		if (file_mode & S_IXUSR)
3839 			dac_granted |= VEXEC;
3840 		if (file_mode & S_IRUSR)
3841 			dac_granted |= VREAD;
3842 		if (file_mode & S_IWUSR)
3843 			dac_granted |= (VWRITE | VAPPEND);
3844 
3845 		if ((accmode & dac_granted) == accmode)
3846 			return (0);
3847 
3848 		goto privcheck;
3849 	}
3850 
3851 	/* Otherwise, check the groups (first match) */
3852 	if (groupmember(file_gid, cred)) {
3853 		if (file_mode & S_IXGRP)
3854 			dac_granted |= VEXEC;
3855 		if (file_mode & S_IRGRP)
3856 			dac_granted |= VREAD;
3857 		if (file_mode & S_IWGRP)
3858 			dac_granted |= (VWRITE | VAPPEND);
3859 
3860 		if ((accmode & dac_granted) == accmode)
3861 			return (0);
3862 
3863 		goto privcheck;
3864 	}
3865 
3866 	/* Otherwise, check everyone else. */
3867 	if (file_mode & S_IXOTH)
3868 		dac_granted |= VEXEC;
3869 	if (file_mode & S_IROTH)
3870 		dac_granted |= VREAD;
3871 	if (file_mode & S_IWOTH)
3872 		dac_granted |= (VWRITE | VAPPEND);
3873 	if ((accmode & dac_granted) == accmode)
3874 		return (0);
3875 
3876 privcheck:
3877 	/*
3878 	 * Build a privilege mask to determine if the set of privileges
3879 	 * satisfies the requirements when combined with the granted mask
3880 	 * from above.  For each privilege, if the privilege is required,
3881 	 * bitwise or the request type onto the priv_granted mask.
3882 	 */
3883 	priv_granted = 0;
3884 
3885 	if (type == VDIR) {
3886 		/*
3887 		 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3888 		 * requests, instead of PRIV_VFS_EXEC.
3889 		 */
3890 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3891 		    !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3892 			priv_granted |= VEXEC;
3893 	} else {
3894 		/*
3895 		 * Ensure that at least one execute bit is on. Otherwise,
3896 		 * a privileged user will always succeed, and we don't want
3897 		 * this to happen unless the file really is executable.
3898 		 */
3899 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3900 		    (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
3901 		    !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3902 			priv_granted |= VEXEC;
3903 	}
3904 
3905 	if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
3906 	    !priv_check_cred(cred, PRIV_VFS_READ, 0))
3907 		priv_granted |= VREAD;
3908 
3909 	if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3910 	    !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3911 		priv_granted |= (VWRITE | VAPPEND);
3912 
3913 	if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3914 	    !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3915 		priv_granted |= VADMIN;
3916 
3917 	if ((accmode & (priv_granted | dac_granted)) == accmode) {
3918 		/* XXX audit: privilege used */
3919 		if (privused != NULL)
3920 			*privused = 1;
3921 		return (0);
3922 	}
3923 
3924 	return ((accmode & VADMIN) ? EPERM : EACCES);
3925 }
3926 
3927 /*
3928  * Credential check based on process requesting service, and per-attribute
3929  * permissions.
3930  */
3931 int
3932 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3933     struct thread *td, accmode_t accmode)
3934 {
3935 
3936 	/*
3937 	 * Kernel-invoked always succeeds.
3938 	 */
3939 	if (cred == NOCRED)
3940 		return (0);
3941 
3942 	/*
3943 	 * Do not allow privileged processes in jail to directly manipulate
3944 	 * system attributes.
3945 	 */
3946 	switch (attrnamespace) {
3947 	case EXTATTR_NAMESPACE_SYSTEM:
3948 		/* Potentially should be: return (EPERM); */
3949 		return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3950 	case EXTATTR_NAMESPACE_USER:
3951 		return (VOP_ACCESS(vp, accmode, cred, td));
3952 	default:
3953 		return (EPERM);
3954 	}
3955 }
3956 
3957 #ifdef DEBUG_VFS_LOCKS
3958 /*
3959  * This only exists to supress warnings from unlocked specfs accesses.  It is
3960  * no longer ok to have an unlocked VFS.
3961  */
3962 #define	IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL ||		\
3963 	(vp)->v_type == VCHR ||	(vp)->v_type == VBAD)
3964 
3965 int vfs_badlock_ddb = 1;	/* Drop into debugger on violation. */
3966 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
3967     "Drop into debugger on lock violation");
3968 
3969 int vfs_badlock_mutex = 1;	/* Check for interlock across VOPs. */
3970 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
3971     0, "Check for interlock across VOPs");
3972 
3973 int vfs_badlock_print = 1;	/* Print lock violations. */
3974 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
3975     0, "Print lock violations");
3976 
3977 #ifdef KDB
3978 int vfs_badlock_backtrace = 1;	/* Print backtrace at lock violations. */
3979 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
3980     &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
3981 #endif
3982 
3983 static void
3984 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3985 {
3986 
3987 #ifdef KDB
3988 	if (vfs_badlock_backtrace)
3989 		kdb_backtrace();
3990 #endif
3991 	if (vfs_badlock_print)
3992 		printf("%s: %p %s\n", str, (void *)vp, msg);
3993 	if (vfs_badlock_ddb)
3994 		kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3995 }
3996 
3997 void
3998 assert_vi_locked(struct vnode *vp, const char *str)
3999 {
4000 
4001 	if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
4002 		vfs_badlock("interlock is not locked but should be", str, vp);
4003 }
4004 
4005 void
4006 assert_vi_unlocked(struct vnode *vp, const char *str)
4007 {
4008 
4009 	if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
4010 		vfs_badlock("interlock is locked but should not be", str, vp);
4011 }
4012 
4013 void
4014 assert_vop_locked(struct vnode *vp, const char *str)
4015 {
4016 
4017 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0)
4018 		vfs_badlock("is not locked but should be", str, vp);
4019 }
4020 
4021 void
4022 assert_vop_unlocked(struct vnode *vp, const char *str)
4023 {
4024 
4025 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
4026 		vfs_badlock("is locked but should not be", str, vp);
4027 }
4028 
4029 void
4030 assert_vop_elocked(struct vnode *vp, const char *str)
4031 {
4032 
4033 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
4034 		vfs_badlock("is not exclusive locked but should be", str, vp);
4035 }
4036 
4037 #if 0
4038 void
4039 assert_vop_elocked_other(struct vnode *vp, const char *str)
4040 {
4041 
4042 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
4043 		vfs_badlock("is not exclusive locked by another thread",
4044 		    str, vp);
4045 }
4046 
4047 void
4048 assert_vop_slocked(struct vnode *vp, const char *str)
4049 {
4050 
4051 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
4052 		vfs_badlock("is not locked shared but should be", str, vp);
4053 }
4054 #endif /* 0 */
4055 #endif /* DEBUG_VFS_LOCKS */
4056 
4057 void
4058 vop_rename_fail(struct vop_rename_args *ap)
4059 {
4060 
4061 	if (ap->a_tvp != NULL)
4062 		vput(ap->a_tvp);
4063 	if (ap->a_tdvp == ap->a_tvp)
4064 		vrele(ap->a_tdvp);
4065 	else
4066 		vput(ap->a_tdvp);
4067 	vrele(ap->a_fdvp);
4068 	vrele(ap->a_fvp);
4069 }
4070 
4071 void
4072 vop_rename_pre(void *ap)
4073 {
4074 	struct vop_rename_args *a = ap;
4075 
4076 #ifdef DEBUG_VFS_LOCKS
4077 	if (a->a_tvp)
4078 		ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
4079 	ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
4080 	ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
4081 	ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
4082 
4083 	/* Check the source (from). */
4084 	if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
4085 	    (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
4086 		ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
4087 	if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
4088 		ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
4089 
4090 	/* Check the target. */
4091 	if (a->a_tvp)
4092 		ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
4093 	ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
4094 #endif
4095 	if (a->a_tdvp != a->a_fdvp)
4096 		vhold(a->a_fdvp);
4097 	if (a->a_tvp != a->a_fvp)
4098 		vhold(a->a_fvp);
4099 	vhold(a->a_tdvp);
4100 	if (a->a_tvp)
4101 		vhold(a->a_tvp);
4102 }
4103 
4104 void
4105 vop_strategy_pre(void *ap)
4106 {
4107 #ifdef DEBUG_VFS_LOCKS
4108 	struct vop_strategy_args *a;
4109 	struct buf *bp;
4110 
4111 	a = ap;
4112 	bp = a->a_bp;
4113 
4114 	/*
4115 	 * Cluster ops lock their component buffers but not the IO container.
4116 	 */
4117 	if ((bp->b_flags & B_CLUSTER) != 0)
4118 		return;
4119 
4120 	if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
4121 		if (vfs_badlock_print)
4122 			printf(
4123 			    "VOP_STRATEGY: bp is not locked but should be\n");
4124 		if (vfs_badlock_ddb)
4125 			kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
4126 	}
4127 #endif
4128 }
4129 
4130 void
4131 vop_lookup_pre(void *ap)
4132 {
4133 #ifdef DEBUG_VFS_LOCKS
4134 	struct vop_lookup_args *a;
4135 	struct vnode *dvp;
4136 
4137 	a = ap;
4138 	dvp = a->a_dvp;
4139 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
4140 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
4141 #endif
4142 }
4143 
4144 void
4145 vop_lookup_post(void *ap, int rc)
4146 {
4147 #ifdef DEBUG_VFS_LOCKS
4148 	struct vop_lookup_args *a;
4149 	struct vnode *dvp;
4150 	struct vnode *vp;
4151 
4152 	a = ap;
4153 	dvp = a->a_dvp;
4154 	vp = *(a->a_vpp);
4155 
4156 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
4157 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
4158 
4159 	if (!rc)
4160 		ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
4161 #endif
4162 }
4163 
4164 void
4165 vop_lock_pre(void *ap)
4166 {
4167 #ifdef DEBUG_VFS_LOCKS
4168 	struct vop_lock1_args *a = ap;
4169 
4170 	if ((a->a_flags & LK_INTERLOCK) == 0)
4171 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4172 	else
4173 		ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
4174 #endif
4175 }
4176 
4177 void
4178 vop_lock_post(void *ap, int rc)
4179 {
4180 #ifdef DEBUG_VFS_LOCKS
4181 	struct vop_lock1_args *a = ap;
4182 
4183 	ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4184 	if (rc == 0)
4185 		ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
4186 #endif
4187 }
4188 
4189 void
4190 vop_unlock_pre(void *ap)
4191 {
4192 #ifdef DEBUG_VFS_LOCKS
4193 	struct vop_unlock_args *a = ap;
4194 
4195 	if (a->a_flags & LK_INTERLOCK)
4196 		ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
4197 	ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
4198 #endif
4199 }
4200 
4201 void
4202 vop_unlock_post(void *ap, int rc)
4203 {
4204 #ifdef DEBUG_VFS_LOCKS
4205 	struct vop_unlock_args *a = ap;
4206 
4207 	if (a->a_flags & LK_INTERLOCK)
4208 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
4209 #endif
4210 }
4211 
4212 void
4213 vop_create_post(void *ap, int rc)
4214 {
4215 	struct vop_create_args *a = ap;
4216 
4217 	if (!rc)
4218 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4219 }
4220 
4221 void
4222 vop_deleteextattr_post(void *ap, int rc)
4223 {
4224 	struct vop_deleteextattr_args *a = ap;
4225 
4226 	if (!rc)
4227 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4228 }
4229 
4230 void
4231 vop_link_post(void *ap, int rc)
4232 {
4233 	struct vop_link_args *a = ap;
4234 
4235 	if (!rc) {
4236 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
4237 		VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
4238 	}
4239 }
4240 
4241 void
4242 vop_mkdir_post(void *ap, int rc)
4243 {
4244 	struct vop_mkdir_args *a = ap;
4245 
4246 	if (!rc)
4247 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4248 }
4249 
4250 void
4251 vop_mknod_post(void *ap, int rc)
4252 {
4253 	struct vop_mknod_args *a = ap;
4254 
4255 	if (!rc)
4256 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4257 }
4258 
4259 void
4260 vop_remove_post(void *ap, int rc)
4261 {
4262 	struct vop_remove_args *a = ap;
4263 
4264 	if (!rc) {
4265 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4266 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4267 	}
4268 }
4269 
4270 void
4271 vop_rename_post(void *ap, int rc)
4272 {
4273 	struct vop_rename_args *a = ap;
4274 
4275 	if (!rc) {
4276 		VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
4277 		VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
4278 		VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
4279 		if (a->a_tvp)
4280 			VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
4281 	}
4282 	if (a->a_tdvp != a->a_fdvp)
4283 		vdrop(a->a_fdvp);
4284 	if (a->a_tvp != a->a_fvp)
4285 		vdrop(a->a_fvp);
4286 	vdrop(a->a_tdvp);
4287 	if (a->a_tvp)
4288 		vdrop(a->a_tvp);
4289 }
4290 
4291 void
4292 vop_rmdir_post(void *ap, int rc)
4293 {
4294 	struct vop_rmdir_args *a = ap;
4295 
4296 	if (!rc) {
4297 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4298 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4299 	}
4300 }
4301 
4302 void
4303 vop_setattr_post(void *ap, int rc)
4304 {
4305 	struct vop_setattr_args *a = ap;
4306 
4307 	if (!rc)
4308 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4309 }
4310 
4311 void
4312 vop_setextattr_post(void *ap, int rc)
4313 {
4314 	struct vop_setextattr_args *a = ap;
4315 
4316 	if (!rc)
4317 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4318 }
4319 
4320 void
4321 vop_symlink_post(void *ap, int rc)
4322 {
4323 	struct vop_symlink_args *a = ap;
4324 
4325 	if (!rc)
4326 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4327 }
4328 
4329 static struct knlist fs_knlist;
4330 
4331 static void
4332 vfs_event_init(void *arg)
4333 {
4334 	knlist_init_mtx(&fs_knlist, NULL);
4335 }
4336 /* XXX - correct order? */
4337 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4338 
4339 void
4340 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
4341 {
4342 
4343 	KNOTE_UNLOCKED(&fs_knlist, event);
4344 }
4345 
4346 static int	filt_fsattach(struct knote *kn);
4347 static void	filt_fsdetach(struct knote *kn);
4348 static int	filt_fsevent(struct knote *kn, long hint);
4349 
4350 struct filterops fs_filtops = {
4351 	.f_isfd = 0,
4352 	.f_attach = filt_fsattach,
4353 	.f_detach = filt_fsdetach,
4354 	.f_event = filt_fsevent
4355 };
4356 
4357 static int
4358 filt_fsattach(struct knote *kn)
4359 {
4360 
4361 	kn->kn_flags |= EV_CLEAR;
4362 	knlist_add(&fs_knlist, kn, 0);
4363 	return (0);
4364 }
4365 
4366 static void
4367 filt_fsdetach(struct knote *kn)
4368 {
4369 
4370 	knlist_remove(&fs_knlist, kn, 0);
4371 }
4372 
4373 static int
4374 filt_fsevent(struct knote *kn, long hint)
4375 {
4376 
4377 	kn->kn_fflags |= hint;
4378 	return (kn->kn_fflags != 0);
4379 }
4380 
4381 static int
4382 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4383 {
4384 	struct vfsidctl vc;
4385 	int error;
4386 	struct mount *mp;
4387 
4388 	error = SYSCTL_IN(req, &vc, sizeof(vc));
4389 	if (error)
4390 		return (error);
4391 	if (vc.vc_vers != VFS_CTL_VERS1)
4392 		return (EINVAL);
4393 	mp = vfs_getvfs(&vc.vc_fsid);
4394 	if (mp == NULL)
4395 		return (ENOENT);
4396 	/* ensure that a specific sysctl goes to the right filesystem. */
4397 	if (strcmp(vc.vc_fstypename, "*") != 0 &&
4398 	    strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4399 		vfs_rel(mp);
4400 		return (EINVAL);
4401 	}
4402 	VCTLTOREQ(&vc, req);
4403 	error = VFS_SYSCTL(mp, vc.vc_op, req);
4404 	vfs_rel(mp);
4405 	return (error);
4406 }
4407 
4408 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
4409     NULL, 0, sysctl_vfs_ctl, "",
4410     "Sysctl by fsid");
4411 
4412 /*
4413  * Function to initialize a va_filerev field sensibly.
4414  * XXX: Wouldn't a random number make a lot more sense ??
4415  */
4416 u_quad_t
4417 init_va_filerev(void)
4418 {
4419 	struct bintime bt;
4420 
4421 	getbinuptime(&bt);
4422 	return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4423 }
4424 
4425 static int	filt_vfsread(struct knote *kn, long hint);
4426 static int	filt_vfswrite(struct knote *kn, long hint);
4427 static int	filt_vfsvnode(struct knote *kn, long hint);
4428 static void	filt_vfsdetach(struct knote *kn);
4429 static struct filterops vfsread_filtops = {
4430 	.f_isfd = 1,
4431 	.f_detach = filt_vfsdetach,
4432 	.f_event = filt_vfsread
4433 };
4434 static struct filterops vfswrite_filtops = {
4435 	.f_isfd = 1,
4436 	.f_detach = filt_vfsdetach,
4437 	.f_event = filt_vfswrite
4438 };
4439 static struct filterops vfsvnode_filtops = {
4440 	.f_isfd = 1,
4441 	.f_detach = filt_vfsdetach,
4442 	.f_event = filt_vfsvnode
4443 };
4444 
4445 static void
4446 vfs_knllock(void *arg)
4447 {
4448 	struct vnode *vp = arg;
4449 
4450 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4451 }
4452 
4453 static void
4454 vfs_knlunlock(void *arg)
4455 {
4456 	struct vnode *vp = arg;
4457 
4458 	VOP_UNLOCK(vp, 0);
4459 }
4460 
4461 static void
4462 vfs_knl_assert_locked(void *arg)
4463 {
4464 #ifdef DEBUG_VFS_LOCKS
4465 	struct vnode *vp = arg;
4466 
4467 	ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4468 #endif
4469 }
4470 
4471 static void
4472 vfs_knl_assert_unlocked(void *arg)
4473 {
4474 #ifdef DEBUG_VFS_LOCKS
4475 	struct vnode *vp = arg;
4476 
4477 	ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4478 #endif
4479 }
4480 
4481 int
4482 vfs_kqfilter(struct vop_kqfilter_args *ap)
4483 {
4484 	struct vnode *vp = ap->a_vp;
4485 	struct knote *kn = ap->a_kn;
4486 	struct knlist *knl;
4487 
4488 	switch (kn->kn_filter) {
4489 	case EVFILT_READ:
4490 		kn->kn_fop = &vfsread_filtops;
4491 		break;
4492 	case EVFILT_WRITE:
4493 		kn->kn_fop = &vfswrite_filtops;
4494 		break;
4495 	case EVFILT_VNODE:
4496 		kn->kn_fop = &vfsvnode_filtops;
4497 		break;
4498 	default:
4499 		return (EINVAL);
4500 	}
4501 
4502 	kn->kn_hook = (caddr_t)vp;
4503 
4504 	v_addpollinfo(vp);
4505 	if (vp->v_pollinfo == NULL)
4506 		return (ENOMEM);
4507 	knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4508 	knlist_add(knl, kn, 0);
4509 
4510 	return (0);
4511 }
4512 
4513 /*
4514  * Detach knote from vnode
4515  */
4516 static void
4517 filt_vfsdetach(struct knote *kn)
4518 {
4519 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4520 
4521 	KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4522 	knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4523 }
4524 
4525 /*ARGSUSED*/
4526 static int
4527 filt_vfsread(struct knote *kn, long hint)
4528 {
4529 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4530 	struct vattr va;
4531 	int res;
4532 
4533 	/*
4534 	 * filesystem is gone, so set the EOF flag and schedule
4535 	 * the knote for deletion.
4536 	 */
4537 	if (hint == NOTE_REVOKE) {
4538 		VI_LOCK(vp);
4539 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4540 		VI_UNLOCK(vp);
4541 		return (1);
4542 	}
4543 
4544 	if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4545 		return (0);
4546 
4547 	VI_LOCK(vp);
4548 	kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4549 	res = (kn->kn_data != 0);
4550 	VI_UNLOCK(vp);
4551 	return (res);
4552 }
4553 
4554 /*ARGSUSED*/
4555 static int
4556 filt_vfswrite(struct knote *kn, long hint)
4557 {
4558 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4559 
4560 	VI_LOCK(vp);
4561 
4562 	/*
4563 	 * filesystem is gone, so set the EOF flag and schedule
4564 	 * the knote for deletion.
4565 	 */
4566 	if (hint == NOTE_REVOKE)
4567 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4568 
4569 	kn->kn_data = 0;
4570 	VI_UNLOCK(vp);
4571 	return (1);
4572 }
4573 
4574 static int
4575 filt_vfsvnode(struct knote *kn, long hint)
4576 {
4577 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4578 	int res;
4579 
4580 	VI_LOCK(vp);
4581 	if (kn->kn_sfflags & hint)
4582 		kn->kn_fflags |= hint;
4583 	if (hint == NOTE_REVOKE) {
4584 		kn->kn_flags |= EV_EOF;
4585 		VI_UNLOCK(vp);
4586 		return (1);
4587 	}
4588 	res = (kn->kn_fflags != 0);
4589 	VI_UNLOCK(vp);
4590 	return (res);
4591 }
4592 
4593 int
4594 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4595 {
4596 	int error;
4597 
4598 	if (dp->d_reclen > ap->a_uio->uio_resid)
4599 		return (ENAMETOOLONG);
4600 	error = uiomove(dp, dp->d_reclen, ap->a_uio);
4601 	if (error) {
4602 		if (ap->a_ncookies != NULL) {
4603 			if (ap->a_cookies != NULL)
4604 				free(ap->a_cookies, M_TEMP);
4605 			ap->a_cookies = NULL;
4606 			*ap->a_ncookies = 0;
4607 		}
4608 		return (error);
4609 	}
4610 	if (ap->a_ncookies == NULL)
4611 		return (0);
4612 
4613 	KASSERT(ap->a_cookies,
4614 	    ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4615 
4616 	*ap->a_cookies = realloc(*ap->a_cookies,
4617 	    (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4618 	(*ap->a_cookies)[*ap->a_ncookies] = off;
4619 	return (0);
4620 }
4621 
4622 /*
4623  * Mark for update the access time of the file if the filesystem
4624  * supports VOP_MARKATIME.  This functionality is used by execve and
4625  * mmap, so we want to avoid the I/O implied by directly setting
4626  * va_atime for the sake of efficiency.
4627  */
4628 void
4629 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4630 {
4631 	struct mount *mp;
4632 
4633 	mp = vp->v_mount;
4634 	VFS_ASSERT_GIANT(mp);
4635 	ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4636 	if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4637 		(void)VOP_MARKATIME(vp);
4638 }
4639 
4640 /*
4641  * The purpose of this routine is to remove granularity from accmode_t,
4642  * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4643  * VADMIN and VAPPEND.
4644  *
4645  * If it returns 0, the caller is supposed to continue with the usual
4646  * access checks using 'accmode' as modified by this routine.  If it
4647  * returns nonzero value, the caller is supposed to return that value
4648  * as errno.
4649  *
4650  * Note that after this routine runs, accmode may be zero.
4651  */
4652 int
4653 vfs_unixify_accmode(accmode_t *accmode)
4654 {
4655 	/*
4656 	 * There is no way to specify explicit "deny" rule using
4657 	 * file mode or POSIX.1e ACLs.
4658 	 */
4659 	if (*accmode & VEXPLICIT_DENY) {
4660 		*accmode = 0;
4661 		return (0);
4662 	}
4663 
4664 	/*
4665 	 * None of these can be translated into usual access bits.
4666 	 * Also, the common case for NFSv4 ACLs is to not contain
4667 	 * either of these bits. Caller should check for VWRITE
4668 	 * on the containing directory instead.
4669 	 */
4670 	if (*accmode & (VDELETE_CHILD | VDELETE))
4671 		return (EPERM);
4672 
4673 	if (*accmode & VADMIN_PERMS) {
4674 		*accmode &= ~VADMIN_PERMS;
4675 		*accmode |= VADMIN;
4676 	}
4677 
4678 	/*
4679 	 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4680 	 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4681 	 */
4682 	*accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
4683 
4684 	return (0);
4685 }
4686 
4687 /*
4688  * These are helper functions for filesystems to traverse all
4689  * their vnodes.  See MNT_VNODE_FOREACH_ALL() in sys/mount.h.
4690  *
4691  * This interface replaces MNT_VNODE_FOREACH.
4692  */
4693 
4694 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
4695 
4696 struct vnode *
4697 __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp)
4698 {
4699 	struct vnode *vp;
4700 
4701 	if (should_yield())
4702 		kern_yield(PRI_UNCHANGED);
4703 	MNT_ILOCK(mp);
4704 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4705 	vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
4706 	while (vp != NULL && (vp->v_type == VMARKER ||
4707 	    (vp->v_iflag & VI_DOOMED) != 0))
4708 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
4709 
4710 	/* Check if we are done */
4711 	if (vp == NULL) {
4712 		__mnt_vnode_markerfree_all(mvp, mp);
4713 		/* MNT_IUNLOCK(mp); -- done in above function */
4714 		mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
4715 		return (NULL);
4716 	}
4717 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4718 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4719 	VI_LOCK(vp);
4720 	MNT_IUNLOCK(mp);
4721 	return (vp);
4722 }
4723 
4724 struct vnode *
4725 __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp)
4726 {
4727 	struct vnode *vp;
4728 
4729 	*mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4730 	MNT_ILOCK(mp);
4731 	MNT_REF(mp);
4732 	(*mvp)->v_type = VMARKER;
4733 
4734 	vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
4735 	while (vp != NULL && (vp->v_type == VMARKER ||
4736 	    (vp->v_iflag & VI_DOOMED) != 0))
4737 		vp = TAILQ_NEXT(vp, v_nmntvnodes);
4738 
4739 	/* Check if we are done */
4740 	if (vp == NULL) {
4741 		MNT_REL(mp);
4742 		MNT_IUNLOCK(mp);
4743 		free(*mvp, M_VNODE_MARKER);
4744 		*mvp = NULL;
4745 		return (NULL);
4746 	}
4747 	(*mvp)->v_mount = mp;
4748 	TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4749 	VI_LOCK(vp);
4750 	MNT_IUNLOCK(mp);
4751 	return (vp);
4752 }
4753 
4754 
4755 void
4756 __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp)
4757 {
4758 
4759 	if (*mvp == NULL) {
4760 		MNT_IUNLOCK(mp);
4761 		return;
4762 	}
4763 
4764 	mtx_assert(MNT_MTX(mp), MA_OWNED);
4765 
4766 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4767 	TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4768 	MNT_REL(mp);
4769 	MNT_IUNLOCK(mp);
4770 	free(*mvp, M_VNODE_MARKER);
4771 	*mvp = NULL;
4772 }
4773 
4774 /*
4775  * These are helper functions for filesystems to traverse their
4776  * active vnodes.  See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h
4777  */
4778 struct vnode *
4779 __mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4780 {
4781 	struct vnode *vp, *nvp;
4782 
4783 	if (should_yield())
4784 		kern_yield(PRI_UNCHANGED);
4785 	MNT_ILOCK(mp);
4786 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4787 	vp = TAILQ_NEXT(*mvp, v_actfreelist);
4788 	while (vp != NULL) {
4789 		VI_LOCK(vp);
4790 		if (vp->v_mount == mp && vp->v_type != VMARKER &&
4791 		    (vp->v_iflag & VI_DOOMED) == 0)
4792 			break;
4793 		nvp = TAILQ_NEXT(vp, v_actfreelist);
4794 		VI_UNLOCK(vp);
4795 		vp = nvp;
4796 	}
4797 
4798 	/* Check if we are done */
4799 	if (vp == NULL) {
4800 		__mnt_vnode_markerfree_active(mvp, mp);
4801 		/* MNT_IUNLOCK(mp); -- done in above function */
4802 		mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
4803 		return (NULL);
4804 	}
4805 	mtx_lock(&vnode_free_list_mtx);
4806 	TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4807 	TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
4808 	mtx_unlock(&vnode_free_list_mtx);
4809 	MNT_IUNLOCK(mp);
4810 	return (vp);
4811 }
4812 
4813 struct vnode *
4814 __mnt_vnode_first_active(struct vnode **mvp, struct mount *mp)
4815 {
4816 	struct vnode *vp, *nvp;
4817 
4818 	*mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4819 	MNT_ILOCK(mp);
4820 	MNT_REF(mp);
4821 	(*mvp)->v_type = VMARKER;
4822 
4823 	vp = TAILQ_NEXT(*mvp, v_actfreelist);
4824 	while (vp != NULL) {
4825 		VI_LOCK(vp);
4826 		if (vp->v_mount == mp && vp->v_type != VMARKER &&
4827 		    (vp->v_iflag & VI_DOOMED) == 0)
4828 			break;
4829 		nvp = TAILQ_NEXT(vp, v_actfreelist);
4830 		VI_UNLOCK(vp);
4831 		vp = nvp;
4832 	}
4833 
4834 	/* Check if we are done */
4835 	if (vp == NULL) {
4836 		MNT_REL(mp);
4837 		MNT_IUNLOCK(mp);
4838 		free(*mvp, M_VNODE_MARKER);
4839 		*mvp = NULL;
4840 		return (NULL);
4841 	}
4842 	(*mvp)->v_mount = mp;
4843 	mtx_lock(&vnode_free_list_mtx);
4844 	TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
4845 	mtx_unlock(&vnode_free_list_mtx);
4846 	MNT_IUNLOCK(mp);
4847 	return (vp);
4848 }
4849 
4850 void
4851 __mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4852 {
4853 
4854 	if (*mvp == NULL) {
4855 		MNT_IUNLOCK(mp);
4856 		return;
4857 	}
4858 
4859 	mtx_assert(MNT_MTX(mp), MA_OWNED);
4860 
4861 	KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4862 	mtx_lock(&vnode_free_list_mtx);
4863 	TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4864 	mtx_unlock(&vnode_free_list_mtx);
4865 	MNT_REL(mp);
4866 	MNT_IUNLOCK(mp);
4867 	free(*mvp, M_VNODE_MARKER);
4868 	*mvp = NULL;
4869 }
4870