xref: /freebsd/sys/kern/vfs_subr.c (revision 8be96e101f2691b80ff9562b72f874da82e735aa)
1 /*-
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)vfs_subr.c	8.31 (Berkeley) 5/26/95
35  */
36 
37 /*
38  * External virtual filesystem routines
39  */
40 
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
43 
44 #include "opt_ddb.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/bio.h>
49 #include <sys/buf.h>
50 #include <sys/condvar.h>
51 #include <sys/conf.h>
52 #include <sys/dirent.h>
53 #include <sys/event.h>
54 #include <sys/eventhandler.h>
55 #include <sys/extattr.h>
56 #include <sys/file.h>
57 #include <sys/fcntl.h>
58 #include <sys/jail.h>
59 #include <sys/kdb.h>
60 #include <sys/kernel.h>
61 #include <sys/kthread.h>
62 #include <sys/lockf.h>
63 #include <sys/malloc.h>
64 #include <sys/mount.h>
65 #include <sys/namei.h>
66 #include <sys/priv.h>
67 #include <sys/reboot.h>
68 #include <sys/sleepqueue.h>
69 #include <sys/stat.h>
70 #include <sys/sysctl.h>
71 #include <sys/syslog.h>
72 #include <sys/vmmeter.h>
73 #include <sys/vnode.h>
74 
75 #include <machine/stdarg.h>
76 
77 #include <security/mac/mac_framework.h>
78 
79 #include <vm/vm.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_extern.h>
82 #include <vm/pmap.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_page.h>
85 #include <vm/vm_kern.h>
86 #include <vm/uma.h>
87 
88 #ifdef DDB
89 #include <ddb/ddb.h>
90 #endif
91 
92 #define	WI_MPSAFEQ	0
93 #define	WI_GIANTQ	1
94 
95 static MALLOC_DEFINE(M_NETADDR, "subr_export_host", "Export host address structure");
96 
97 static void	delmntque(struct vnode *vp);
98 static int	flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
99 		    int slpflag, int slptimeo);
100 static void	syncer_shutdown(void *arg, int howto);
101 static int	vtryrecycle(struct vnode *vp);
102 static void	vbusy(struct vnode *vp);
103 static void	vinactive(struct vnode *, struct thread *);
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	vfree(struct vnode *);
109 static void	vnlru_free(int);
110 static void	vgonel(struct vnode *);
111 static void	vfs_knllock(void *arg);
112 static void	vfs_knlunlock(void *arg);
113 static void	vfs_knl_assert_locked(void *arg);
114 static void	vfs_knl_assert_unlocked(void *arg);
115 static void	destroy_vpollinfo(struct vpollinfo *vi);
116 
117 /*
118  * Number of vnodes in existence.  Increased whenever getnewvnode()
119  * allocates a new vnode, decreased on vdestroy() called on VI_DOOMed
120  * vnode.
121  */
122 static unsigned long	numvnodes;
123 
124 SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
125 
126 /*
127  * Conversion tables for conversion from vnode types to inode formats
128  * and back.
129  */
130 enum vtype iftovt_tab[16] = {
131 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
132 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
133 };
134 int vttoif_tab[10] = {
135 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
136 	S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
137 };
138 
139 /*
140  * List of vnodes that are ready for recycling.
141  */
142 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
143 
144 /*
145  * Free vnode target.  Free vnodes may simply be files which have been stat'd
146  * but not read.  This is somewhat common, and a small cache of such files
147  * should be kept to avoid recreation costs.
148  */
149 static u_long wantfreevnodes;
150 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
151 /* Number of vnodes in the free list. */
152 static u_long freevnodes;
153 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
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 
167 /*
168  * Cache for the mount type id assigned to NFS.  This is used for
169  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
170  */
171 int	nfs_mount_type = -1;
172 
173 /* To keep more than one thread at a time from running vfs_getnewfsid */
174 static struct mtx mntid_mtx;
175 
176 /*
177  * Lock for any access to the following:
178  *	vnode_free_list
179  *	numvnodes
180  *	freevnodes
181  */
182 static struct mtx vnode_free_list_mtx;
183 
184 /* Publicly exported FS */
185 struct nfs_public nfs_pub;
186 
187 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
188 static uma_zone_t vnode_zone;
189 static uma_zone_t vnodepoll_zone;
190 
191 /* Set to 1 to print out reclaim of active vnodes */
192 int	prtactive;
193 
194 /*
195  * The workitem queue.
196  *
197  * It is useful to delay writes of file data and filesystem metadata
198  * for tens of seconds so that quickly created and deleted files need
199  * not waste disk bandwidth being created and removed. To realize this,
200  * we append vnodes to a "workitem" queue. When running with a soft
201  * updates implementation, most pending metadata dependencies should
202  * not wait for more than a few seconds. Thus, mounted on block devices
203  * are delayed only about a half the time that file data is delayed.
204  * Similarly, directory updates are more critical, so are only delayed
205  * about a third the time that file data is delayed. Thus, there are
206  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
207  * one each second (driven off the filesystem syncer process). The
208  * syncer_delayno variable indicates the next queue that is to be processed.
209  * Items that need to be processed soon are placed in this queue:
210  *
211  *	syncer_workitem_pending[syncer_delayno]
212  *
213  * A delay of fifteen seconds is done by placing the request fifteen
214  * entries later in the queue:
215  *
216  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
217  *
218  */
219 static int syncer_delayno;
220 static long syncer_mask;
221 LIST_HEAD(synclist, bufobj);
222 static struct synclist *syncer_workitem_pending[2];
223 /*
224  * The sync_mtx protects:
225  *	bo->bo_synclist
226  *	sync_vnode_count
227  *	syncer_delayno
228  *	syncer_state
229  *	syncer_workitem_pending
230  *	syncer_worklist_len
231  *	rushjob
232  */
233 static struct mtx sync_mtx;
234 static struct cv sync_wakeup;
235 
236 #define SYNCER_MAXDELAY		32
237 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
238 static int syncdelay = 30;		/* max time to delay syncing data */
239 static int filedelay = 30;		/* time to delay syncing files */
240 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
241 static int dirdelay = 29;		/* time to delay syncing directories */
242 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
243 static int metadelay = 28;		/* time to delay syncing metadata */
244 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
245 static int rushjob;		/* number of slots to run ASAP */
246 static int stat_rush_requests;	/* number of times I/O speeded up */
247 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
248 
249 /*
250  * When shutting down the syncer, run it at four times normal speed.
251  */
252 #define SYNCER_SHUTDOWN_SPEEDUP		4
253 static int sync_vnode_count;
254 static int syncer_worklist_len;
255 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
256     syncer_state;
257 
258 /*
259  * Number of vnodes we want to exist at any one time.  This is mostly used
260  * to size hash tables in vnode-related code.  It is normally not used in
261  * getnewvnode(), as wantfreevnodes is normally nonzero.)
262  *
263  * XXX desiredvnodes is historical cruft and should not exist.
264  */
265 int desiredvnodes;
266 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
267     &desiredvnodes, 0, "Maximum number of vnodes");
268 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
269     &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
270 static int vnlru_nowhere;
271 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
272     &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
273 
274 /*
275  * Macros to control when a vnode is freed and recycled.  All require
276  * the vnode interlock.
277  */
278 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
279 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
280 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
281 
282 
283 /*
284  * Initialize the vnode management data structures.
285  *
286  * Reevaluate the following cap on the number of vnodes after the physical
287  * memory size exceeds 512GB.  In the limit, as the physical memory size
288  * grows, the ratio of physical pages to vnodes approaches sixteen to one.
289  */
290 #ifndef	MAXVNODES_MAX
291 #define	MAXVNODES_MAX	(512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
292 #endif
293 static void
294 vntblinit(void *dummy __unused)
295 {
296 	int physvnodes, virtvnodes;
297 
298 	/*
299 	 * Desiredvnodes is a function of the physical memory size and the
300 	 * kernel's heap size.  Generally speaking, it scales with the
301 	 * physical memory size.  The ratio of desiredvnodes to physical pages
302 	 * is one to four until desiredvnodes exceeds 98,304.  Thereafter, the
303 	 * marginal ratio of desiredvnodes to physical pages is one to
304 	 * sixteen.  However, desiredvnodes is limited by the kernel's heap
305 	 * size.  The memory required by desiredvnodes vnodes and vm objects
306 	 * may not exceed one seventh of the kernel's heap size.
307 	 */
308 	physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
309 	    cnt.v_page_count) / 16;
310 	virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
311 	    sizeof(struct vnode)));
312 	desiredvnodes = min(physvnodes, virtvnodes);
313 	if (desiredvnodes > MAXVNODES_MAX) {
314 		if (bootverbose)
315 			printf("Reducing kern.maxvnodes %d -> %d\n",
316 			    desiredvnodes, MAXVNODES_MAX);
317 		desiredvnodes = MAXVNODES_MAX;
318 	}
319 	wantfreevnodes = desiredvnodes / 4;
320 	mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
321 	TAILQ_INIT(&vnode_free_list);
322 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
323 	vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
324 	    NULL, NULL, UMA_ALIGN_PTR, 0);
325 	vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
326 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
327 	/*
328 	 * Initialize the filesystem syncer.
329 	 */
330 	syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE,
331 	    &syncer_mask);
332 	syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE,
333 	    &syncer_mask);
334 	syncer_maxdelay = syncer_mask + 1;
335 	mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
336 	cv_init(&sync_wakeup, "syncer");
337 }
338 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
339 
340 
341 /*
342  * Mark a mount point as busy. Used to synchronize access and to delay
343  * unmounting. Eventually, mountlist_mtx is not released on failure.
344  */
345 int
346 vfs_busy(struct mount *mp, int flags)
347 {
348 
349 	MPASS((flags & ~MBF_MASK) == 0);
350 	CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
351 
352 	MNT_ILOCK(mp);
353 	MNT_REF(mp);
354 	/*
355 	 * If mount point is currenly being unmounted, sleep until the
356 	 * mount point fate is decided.  If thread doing the unmounting fails,
357 	 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
358 	 * that this mount point has survived the unmount attempt and vfs_busy
359 	 * should retry.  Otherwise the unmounter thread will set MNTK_REFEXPIRE
360 	 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
361 	 * about to be really destroyed.  vfs_busy needs to release its
362 	 * reference on the mount point in this case and return with ENOENT,
363 	 * telling the caller that mount mount it tried to busy is no longer
364 	 * valid.
365 	 */
366 	while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
367 		if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
368 			MNT_REL(mp);
369 			MNT_IUNLOCK(mp);
370 			CTR1(KTR_VFS, "%s: failed busying before sleeping",
371 			    __func__);
372 			return (ENOENT);
373 		}
374 		if (flags & MBF_MNTLSTLOCK)
375 			mtx_unlock(&mountlist_mtx);
376 		mp->mnt_kern_flag |= MNTK_MWAIT;
377 		msleep(mp, MNT_MTX(mp), PVFS, "vfs_busy", 0);
378 		if (flags & MBF_MNTLSTLOCK)
379 			mtx_lock(&mountlist_mtx);
380 	}
381 	if (flags & MBF_MNTLSTLOCK)
382 		mtx_unlock(&mountlist_mtx);
383 	mp->mnt_lockref++;
384 	MNT_IUNLOCK(mp);
385 	return (0);
386 }
387 
388 /*
389  * Free a busy filesystem.
390  */
391 void
392 vfs_unbusy(struct mount *mp)
393 {
394 
395 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
396 	MNT_ILOCK(mp);
397 	MNT_REL(mp);
398 	KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
399 	mp->mnt_lockref--;
400 	if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
401 		MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
402 		CTR1(KTR_VFS, "%s: waking up waiters", __func__);
403 		mp->mnt_kern_flag &= ~MNTK_DRAINING;
404 		wakeup(&mp->mnt_lockref);
405 	}
406 	MNT_IUNLOCK(mp);
407 }
408 
409 /*
410  * Lookup a mount point by filesystem identifier.
411  */
412 struct mount *
413 vfs_getvfs(fsid_t *fsid)
414 {
415 	struct mount *mp;
416 
417 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
418 	mtx_lock(&mountlist_mtx);
419 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
420 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
421 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
422 			vfs_ref(mp);
423 			mtx_unlock(&mountlist_mtx);
424 			return (mp);
425 		}
426 	}
427 	mtx_unlock(&mountlist_mtx);
428 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
429 	return ((struct mount *) 0);
430 }
431 
432 /*
433  * Lookup a mount point by filesystem identifier, busying it before
434  * returning.
435  */
436 struct mount *
437 vfs_busyfs(fsid_t *fsid)
438 {
439 	struct mount *mp;
440 	int error;
441 
442 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
443 	mtx_lock(&mountlist_mtx);
444 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
445 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
446 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
447 			error = vfs_busy(mp, MBF_MNTLSTLOCK);
448 			if (error) {
449 				mtx_unlock(&mountlist_mtx);
450 				return (NULL);
451 			}
452 			return (mp);
453 		}
454 	}
455 	CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
456 	mtx_unlock(&mountlist_mtx);
457 	return ((struct mount *) 0);
458 }
459 
460 /*
461  * Check if a user can access privileged mount options.
462  */
463 int
464 vfs_suser(struct mount *mp, struct thread *td)
465 {
466 	int error;
467 
468 	/*
469 	 * If the thread is jailed, but this is not a jail-friendly file
470 	 * system, deny immediately.
471 	 */
472 	if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
473 		return (EPERM);
474 
475 	/*
476 	 * If the file system was mounted outside the jail of the calling
477 	 * thread, deny immediately.
478 	 */
479 	if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
480 		return (EPERM);
481 
482 	/*
483 	 * If file system supports delegated administration, we don't check
484 	 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
485 	 * by the file system itself.
486 	 * If this is not the user that did original mount, we check for
487 	 * the PRIV_VFS_MOUNT_OWNER privilege.
488 	 */
489 	if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
490 	    mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
491 		if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
492 			return (error);
493 	}
494 	return (0);
495 }
496 
497 /*
498  * Get a new unique fsid.  Try to make its val[0] unique, since this value
499  * will be used to create fake device numbers for stat().  Also try (but
500  * not so hard) make its val[0] unique mod 2^16, since some emulators only
501  * support 16-bit device numbers.  We end up with unique val[0]'s for the
502  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
503  *
504  * Keep in mind that several mounts may be running in parallel.  Starting
505  * the search one past where the previous search terminated is both a
506  * micro-optimization and a defense against returning the same fsid to
507  * different mounts.
508  */
509 void
510 vfs_getnewfsid(struct mount *mp)
511 {
512 	static uint16_t mntid_base;
513 	struct mount *nmp;
514 	fsid_t tfsid;
515 	int mtype;
516 
517 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
518 	mtx_lock(&mntid_mtx);
519 	mtype = mp->mnt_vfc->vfc_typenum;
520 	tfsid.val[1] = mtype;
521 	mtype = (mtype & 0xFF) << 24;
522 	for (;;) {
523 		tfsid.val[0] = makedev(255,
524 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
525 		mntid_base++;
526 		if ((nmp = vfs_getvfs(&tfsid)) == NULL)
527 			break;
528 		vfs_rel(nmp);
529 	}
530 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
531 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
532 	mtx_unlock(&mntid_mtx);
533 }
534 
535 /*
536  * Knob to control the precision of file timestamps:
537  *
538  *   0 = seconds only; nanoseconds zeroed.
539  *   1 = seconds and nanoseconds, accurate within 1/HZ.
540  *   2 = seconds and nanoseconds, truncated to microseconds.
541  * >=3 = seconds and nanoseconds, maximum precision.
542  */
543 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
544 
545 static int timestamp_precision = TSP_SEC;
546 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
547     &timestamp_precision, 0, "");
548 
549 /*
550  * Get a current timestamp.
551  */
552 void
553 vfs_timestamp(struct timespec *tsp)
554 {
555 	struct timeval tv;
556 
557 	switch (timestamp_precision) {
558 	case TSP_SEC:
559 		tsp->tv_sec = time_second;
560 		tsp->tv_nsec = 0;
561 		break;
562 	case TSP_HZ:
563 		getnanotime(tsp);
564 		break;
565 	case TSP_USEC:
566 		microtime(&tv);
567 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
568 		break;
569 	case TSP_NSEC:
570 	default:
571 		nanotime(tsp);
572 		break;
573 	}
574 }
575 
576 /*
577  * Set vnode attributes to VNOVAL
578  */
579 void
580 vattr_null(struct vattr *vap)
581 {
582 
583 	vap->va_type = VNON;
584 	vap->va_size = VNOVAL;
585 	vap->va_bytes = VNOVAL;
586 	vap->va_mode = VNOVAL;
587 	vap->va_nlink = VNOVAL;
588 	vap->va_uid = VNOVAL;
589 	vap->va_gid = VNOVAL;
590 	vap->va_fsid = VNOVAL;
591 	vap->va_fileid = VNOVAL;
592 	vap->va_blocksize = VNOVAL;
593 	vap->va_rdev = VNOVAL;
594 	vap->va_atime.tv_sec = VNOVAL;
595 	vap->va_atime.tv_nsec = VNOVAL;
596 	vap->va_mtime.tv_sec = VNOVAL;
597 	vap->va_mtime.tv_nsec = VNOVAL;
598 	vap->va_ctime.tv_sec = VNOVAL;
599 	vap->va_ctime.tv_nsec = VNOVAL;
600 	vap->va_birthtime.tv_sec = VNOVAL;
601 	vap->va_birthtime.tv_nsec = VNOVAL;
602 	vap->va_flags = VNOVAL;
603 	vap->va_gen = VNOVAL;
604 	vap->va_vaflags = 0;
605 }
606 
607 /*
608  * This routine is called when we have too many vnodes.  It attempts
609  * to free <count> vnodes and will potentially free vnodes that still
610  * have VM backing store (VM backing store is typically the cause
611  * of a vnode blowout so we want to do this).  Therefore, this operation
612  * is not considered cheap.
613  *
614  * A number of conditions may prevent a vnode from being reclaimed.
615  * the buffer cache may have references on the vnode, a directory
616  * vnode may still have references due to the namei cache representing
617  * underlying files, or the vnode may be in active use.   It is not
618  * desireable to reuse such vnodes.  These conditions may cause the
619  * number of vnodes to reach some minimum value regardless of what
620  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
621  */
622 static int
623 vlrureclaim(struct mount *mp)
624 {
625 	struct vnode *vp;
626 	int done;
627 	int trigger;
628 	int usevnodes;
629 	int count;
630 
631 	/*
632 	 * Calculate the trigger point, don't allow user
633 	 * screwups to blow us up.   This prevents us from
634 	 * recycling vnodes with lots of resident pages.  We
635 	 * aren't trying to free memory, we are trying to
636 	 * free vnodes.
637 	 */
638 	usevnodes = desiredvnodes;
639 	if (usevnodes <= 0)
640 		usevnodes = 1;
641 	trigger = cnt.v_page_count * 2 / usevnodes;
642 	done = 0;
643 	vn_start_write(NULL, &mp, V_WAIT);
644 	MNT_ILOCK(mp);
645 	count = mp->mnt_nvnodelistsize / 10 + 1;
646 	while (count != 0) {
647 		vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
648 		while (vp != NULL && vp->v_type == VMARKER)
649 			vp = TAILQ_NEXT(vp, v_nmntvnodes);
650 		if (vp == NULL)
651 			break;
652 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
653 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
654 		--count;
655 		if (!VI_TRYLOCK(vp))
656 			goto next_iter;
657 		/*
658 		 * If it's been deconstructed already, it's still
659 		 * referenced, or it exceeds the trigger, skip it.
660 		 */
661 		if (vp->v_usecount ||
662 		    (!vlru_allow_cache_src &&
663 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
664 		    (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
665 		    vp->v_object->resident_page_count > trigger)) {
666 			VI_UNLOCK(vp);
667 			goto next_iter;
668 		}
669 		MNT_IUNLOCK(mp);
670 		vholdl(vp);
671 		if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
672 			vdrop(vp);
673 			goto next_iter_mntunlocked;
674 		}
675 		VI_LOCK(vp);
676 		/*
677 		 * v_usecount may have been bumped after VOP_LOCK() dropped
678 		 * the vnode interlock and before it was locked again.
679 		 *
680 		 * It is not necessary to recheck VI_DOOMED because it can
681 		 * only be set by another thread that holds both the vnode
682 		 * lock and vnode interlock.  If another thread has the
683 		 * vnode lock before we get to VOP_LOCK() and obtains the
684 		 * vnode interlock after VOP_LOCK() drops the vnode
685 		 * interlock, the other thread will be unable to drop the
686 		 * vnode lock before our VOP_LOCK() call fails.
687 		 */
688 		if (vp->v_usecount ||
689 		    (!vlru_allow_cache_src &&
690 			!LIST_EMPTY(&(vp)->v_cache_src)) ||
691 		    (vp->v_object != NULL &&
692 		    vp->v_object->resident_page_count > trigger)) {
693 			VOP_UNLOCK(vp, LK_INTERLOCK);
694 			goto next_iter_mntunlocked;
695 		}
696 		KASSERT((vp->v_iflag & VI_DOOMED) == 0,
697 		    ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
698 		vgonel(vp);
699 		VOP_UNLOCK(vp, 0);
700 		vdropl(vp);
701 		done++;
702 next_iter_mntunlocked:
703 		if ((count % 256) != 0)
704 			goto relock_mnt;
705 		goto yield;
706 next_iter:
707 		if ((count % 256) != 0)
708 			continue;
709 		MNT_IUNLOCK(mp);
710 yield:
711 		uio_yield();
712 relock_mnt:
713 		MNT_ILOCK(mp);
714 	}
715 	MNT_IUNLOCK(mp);
716 	vn_finished_write(mp);
717 	return done;
718 }
719 
720 /*
721  * Attempt to keep the free list at wantfreevnodes length.
722  */
723 static void
724 vnlru_free(int count)
725 {
726 	struct vnode *vp;
727 	int vfslocked;
728 
729 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
730 	for (; count > 0; count--) {
731 		vp = TAILQ_FIRST(&vnode_free_list);
732 		/*
733 		 * The list can be modified while the free_list_mtx
734 		 * has been dropped and vp could be NULL here.
735 		 */
736 		if (!vp)
737 			break;
738 		VNASSERT(vp->v_op != NULL, vp,
739 		    ("vnlru_free: vnode already reclaimed."));
740 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
741 		/*
742 		 * Don't recycle if we can't get the interlock.
743 		 */
744 		if (!VI_TRYLOCK(vp)) {
745 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
746 			continue;
747 		}
748 		VNASSERT(VCANRECYCLE(vp), vp,
749 		    ("vp inconsistent on freelist"));
750 		freevnodes--;
751 		vp->v_iflag &= ~VI_FREE;
752 		vholdl(vp);
753 		mtx_unlock(&vnode_free_list_mtx);
754 		VI_UNLOCK(vp);
755 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
756 		vtryrecycle(vp);
757 		VFS_UNLOCK_GIANT(vfslocked);
758 		/*
759 		 * If the recycled succeeded this vdrop will actually free
760 		 * the vnode.  If not it will simply place it back on
761 		 * the free list.
762 		 */
763 		vdrop(vp);
764 		mtx_lock(&vnode_free_list_mtx);
765 	}
766 }
767 /*
768  * Attempt to recycle vnodes in a context that is always safe to block.
769  * Calling vlrurecycle() from the bowels of filesystem code has some
770  * interesting deadlock problems.
771  */
772 static struct proc *vnlruproc;
773 static int vnlruproc_sig;
774 
775 static void
776 vnlru_proc(void)
777 {
778 	struct mount *mp, *nmp;
779 	int done, vfslocked;
780 	struct proc *p = vnlruproc;
781 
782 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
783 	    SHUTDOWN_PRI_FIRST);
784 
785 	for (;;) {
786 		kproc_suspend_check(p);
787 		mtx_lock(&vnode_free_list_mtx);
788 		if (freevnodes > wantfreevnodes)
789 			vnlru_free(freevnodes - wantfreevnodes);
790 		if (numvnodes <= desiredvnodes * 9 / 10) {
791 			vnlruproc_sig = 0;
792 			wakeup(&vnlruproc_sig);
793 			msleep(vnlruproc, &vnode_free_list_mtx,
794 			    PVFS|PDROP, "vlruwt", hz);
795 			continue;
796 		}
797 		mtx_unlock(&vnode_free_list_mtx);
798 		done = 0;
799 		mtx_lock(&mountlist_mtx);
800 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
801 			if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
802 				nmp = TAILQ_NEXT(mp, mnt_list);
803 				continue;
804 			}
805 			vfslocked = VFS_LOCK_GIANT(mp);
806 			done += vlrureclaim(mp);
807 			VFS_UNLOCK_GIANT(vfslocked);
808 			mtx_lock(&mountlist_mtx);
809 			nmp = TAILQ_NEXT(mp, mnt_list);
810 			vfs_unbusy(mp);
811 		}
812 		mtx_unlock(&mountlist_mtx);
813 		if (done == 0) {
814 #if 0
815 			/* These messages are temporary debugging aids */
816 			if (vnlru_nowhere < 5)
817 				printf("vnlru process getting nowhere..\n");
818 			else if (vnlru_nowhere == 5)
819 				printf("vnlru process messages stopped.\n");
820 #endif
821 			vnlru_nowhere++;
822 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
823 		} else
824 			uio_yield();
825 	}
826 }
827 
828 static struct kproc_desc vnlru_kp = {
829 	"vnlru",
830 	vnlru_proc,
831 	&vnlruproc
832 };
833 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
834     &vnlru_kp);
835 
836 /*
837  * Routines having to do with the management of the vnode table.
838  */
839 
840 void
841 vdestroy(struct vnode *vp)
842 {
843 	struct bufobj *bo;
844 
845 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
846 	mtx_lock(&vnode_free_list_mtx);
847 	numvnodes--;
848 	mtx_unlock(&vnode_free_list_mtx);
849 	bo = &vp->v_bufobj;
850 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
851 	    ("cleaned vnode still on the free list."));
852 	VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
853 	VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
854 	VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
855 	VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
856 	VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
857 	VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
858 	VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
859 	VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
860 	VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
861 	VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
862 	VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
863 	VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
864 	VI_UNLOCK(vp);
865 #ifdef MAC
866 	mac_vnode_destroy(vp);
867 #endif
868 	if (vp->v_pollinfo != NULL)
869 		destroy_vpollinfo(vp->v_pollinfo);
870 #ifdef INVARIANTS
871 	/* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */
872 	vp->v_op = NULL;
873 #endif
874 	lockdestroy(vp->v_vnlock);
875 	mtx_destroy(&vp->v_interlock);
876 	mtx_destroy(BO_MTX(bo));
877 	uma_zfree(vnode_zone, vp);
878 }
879 
880 /*
881  * Try to recycle a freed vnode.  We abort if anyone picks up a reference
882  * before we actually vgone().  This function must be called with the vnode
883  * held to prevent the vnode from being returned to the free list midway
884  * through vgone().
885  */
886 static int
887 vtryrecycle(struct vnode *vp)
888 {
889 	struct mount *vnmp;
890 
891 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
892 	VNASSERT(vp->v_holdcnt, vp,
893 	    ("vtryrecycle: Recycling vp %p without a reference.", vp));
894 	/*
895 	 * This vnode may found and locked via some other list, if so we
896 	 * can't recycle it yet.
897 	 */
898 	if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
899 		CTR2(KTR_VFS,
900 		    "%s: impossible to recycle, vp %p lock is already held",
901 		    __func__, vp);
902 		return (EWOULDBLOCK);
903 	}
904 	/*
905 	 * Don't recycle if its filesystem is being suspended.
906 	 */
907 	if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
908 		VOP_UNLOCK(vp, 0);
909 		CTR2(KTR_VFS,
910 		    "%s: impossible to recycle, cannot start the write for %p",
911 		    __func__, vp);
912 		return (EBUSY);
913 	}
914 	/*
915 	 * If we got this far, we need to acquire the interlock and see if
916 	 * anyone picked up this vnode from another list.  If not, we will
917 	 * mark it with DOOMED via vgonel() so that anyone who does find it
918 	 * will skip over it.
919 	 */
920 	VI_LOCK(vp);
921 	if (vp->v_usecount) {
922 		VOP_UNLOCK(vp, LK_INTERLOCK);
923 		vn_finished_write(vnmp);
924 		CTR2(KTR_VFS,
925 		    "%s: impossible to recycle, %p is already referenced",
926 		    __func__, vp);
927 		return (EBUSY);
928 	}
929 	if ((vp->v_iflag & VI_DOOMED) == 0)
930 		vgonel(vp);
931 	VOP_UNLOCK(vp, LK_INTERLOCK);
932 	vn_finished_write(vnmp);
933 	return (0);
934 }
935 
936 /*
937  * Return the next vnode from the free list.
938  */
939 int
940 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
941     struct vnode **vpp)
942 {
943 	struct vnode *vp = NULL;
944 	struct bufobj *bo;
945 
946 	CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
947 	mtx_lock(&vnode_free_list_mtx);
948 	/*
949 	 * Lend our context to reclaim vnodes if they've exceeded the max.
950 	 */
951 	if (freevnodes > wantfreevnodes)
952 		vnlru_free(1);
953 	/*
954 	 * Wait for available vnodes.
955 	 */
956 	if (numvnodes > desiredvnodes) {
957 		if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) {
958 			/*
959 			 * File system is beeing suspended, we cannot risk a
960 			 * deadlock here, so allocate new vnode anyway.
961 			 */
962 			if (freevnodes > wantfreevnodes)
963 				vnlru_free(freevnodes - wantfreevnodes);
964 			goto alloc;
965 		}
966 		if (vnlruproc_sig == 0) {
967 			vnlruproc_sig = 1;	/* avoid unnecessary wakeups */
968 			wakeup(vnlruproc);
969 		}
970 		msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
971 		    "vlruwk", hz);
972 #if 0	/* XXX Not all VFS_VGET/ffs_vget callers check returns. */
973 		if (numvnodes > desiredvnodes) {
974 			mtx_unlock(&vnode_free_list_mtx);
975 			return (ENFILE);
976 		}
977 #endif
978 	}
979 alloc:
980 	numvnodes++;
981 	mtx_unlock(&vnode_free_list_mtx);
982 	vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
983 	/*
984 	 * Setup locks.
985 	 */
986 	vp->v_vnlock = &vp->v_lock;
987 	mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
988 	/*
989 	 * By default, don't allow shared locks unless filesystems
990 	 * opt-in.
991 	 */
992 	lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
993 	/*
994 	 * Initialize bufobj.
995 	 */
996 	bo = &vp->v_bufobj;
997 	bo->__bo_vnode = vp;
998 	mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
999 	bo->bo_ops = &buf_ops_bio;
1000 	bo->bo_private = vp;
1001 	TAILQ_INIT(&bo->bo_clean.bv_hd);
1002 	TAILQ_INIT(&bo->bo_dirty.bv_hd);
1003 	/*
1004 	 * Initialize namecache.
1005 	 */
1006 	LIST_INIT(&vp->v_cache_src);
1007 	TAILQ_INIT(&vp->v_cache_dst);
1008 	/*
1009 	 * Finalize various vnode identity bits.
1010 	 */
1011 	vp->v_type = VNON;
1012 	vp->v_tag = tag;
1013 	vp->v_op = vops;
1014 	v_incr_usecount(vp);
1015 	vp->v_data = 0;
1016 #ifdef MAC
1017 	mac_vnode_init(vp);
1018 	if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1019 		mac_vnode_associate_singlelabel(mp, vp);
1020 	else if (mp == NULL && vops != &dead_vnodeops)
1021 		printf("NULL mp in getnewvnode()\n");
1022 #endif
1023 	if (mp != NULL) {
1024 		bo->bo_bsize = mp->mnt_stat.f_iosize;
1025 		if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1026 			vp->v_vflag |= VV_NOKNOTE;
1027 	}
1028 
1029 	*vpp = vp;
1030 	return (0);
1031 }
1032 
1033 /*
1034  * Delete from old mount point vnode list, if on one.
1035  */
1036 static void
1037 delmntque(struct vnode *vp)
1038 {
1039 	struct mount *mp;
1040 
1041 	mp = vp->v_mount;
1042 	if (mp == NULL)
1043 		return;
1044 	MNT_ILOCK(mp);
1045 	vp->v_mount = NULL;
1046 	VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1047 		("bad mount point vnode list size"));
1048 	TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1049 	mp->mnt_nvnodelistsize--;
1050 	MNT_REL(mp);
1051 	MNT_IUNLOCK(mp);
1052 }
1053 
1054 static void
1055 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1056 {
1057 
1058 	vp->v_data = NULL;
1059 	vp->v_op = &dead_vnodeops;
1060 	/* XXX non mp-safe fs may still call insmntque with vnode
1061 	   unlocked */
1062 	if (!VOP_ISLOCKED(vp))
1063 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1064 	vgone(vp);
1065 	vput(vp);
1066 }
1067 
1068 /*
1069  * Insert into list of vnodes for the new mount point, if available.
1070  */
1071 int
1072 insmntque1(struct vnode *vp, struct mount *mp,
1073 	void (*dtr)(struct vnode *, void *), void *dtr_arg)
1074 {
1075 	int locked;
1076 
1077 	KASSERT(vp->v_mount == NULL,
1078 		("insmntque: vnode already on per mount vnode list"));
1079 	VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1080 #ifdef DEBUG_VFS_LOCKS
1081 	if (!VFS_NEEDSGIANT(mp))
1082 		ASSERT_VOP_ELOCKED(vp,
1083 		    "insmntque: mp-safe fs and non-locked vp");
1084 #endif
1085 	MNT_ILOCK(mp);
1086 	if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1087 	    ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1088 	     mp->mnt_nvnodelistsize == 0)) {
1089 		locked = VOP_ISLOCKED(vp);
1090 		if (!locked || (locked == LK_EXCLUSIVE &&
1091 		     (vp->v_vflag & VV_FORCEINSMQ) == 0)) {
1092 			MNT_IUNLOCK(mp);
1093 			if (dtr != NULL)
1094 				dtr(vp, dtr_arg);
1095 			return (EBUSY);
1096 		}
1097 	}
1098 	vp->v_mount = mp;
1099 	MNT_REF(mp);
1100 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1101 	VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1102 		("neg mount point vnode list size"));
1103 	mp->mnt_nvnodelistsize++;
1104 	MNT_IUNLOCK(mp);
1105 	return (0);
1106 }
1107 
1108 int
1109 insmntque(struct vnode *vp, struct mount *mp)
1110 {
1111 
1112 	return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1113 }
1114 
1115 /*
1116  * Flush out and invalidate all buffers associated with a bufobj
1117  * Called with the underlying object locked.
1118  */
1119 int
1120 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1121 {
1122 	int error;
1123 
1124 	BO_LOCK(bo);
1125 	if (flags & V_SAVE) {
1126 		error = bufobj_wwait(bo, slpflag, slptimeo);
1127 		if (error) {
1128 			BO_UNLOCK(bo);
1129 			return (error);
1130 		}
1131 		if (bo->bo_dirty.bv_cnt > 0) {
1132 			BO_UNLOCK(bo);
1133 			if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1134 				return (error);
1135 			/*
1136 			 * XXX We could save a lock/unlock if this was only
1137 			 * enabled under INVARIANTS
1138 			 */
1139 			BO_LOCK(bo);
1140 			if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1141 				panic("vinvalbuf: dirty bufs");
1142 		}
1143 	}
1144 	/*
1145 	 * If you alter this loop please notice that interlock is dropped and
1146 	 * reacquired in flushbuflist.  Special care is needed to ensure that
1147 	 * no race conditions occur from this.
1148 	 */
1149 	do {
1150 		error = flushbuflist(&bo->bo_clean,
1151 		    flags, bo, slpflag, slptimeo);
1152 		if (error == 0)
1153 			error = flushbuflist(&bo->bo_dirty,
1154 			    flags, bo, slpflag, slptimeo);
1155 		if (error != 0 && error != EAGAIN) {
1156 			BO_UNLOCK(bo);
1157 			return (error);
1158 		}
1159 	} while (error != 0);
1160 
1161 	/*
1162 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1163 	 * have write I/O in-progress but if there is a VM object then the
1164 	 * VM object can also have read-I/O in-progress.
1165 	 */
1166 	do {
1167 		bufobj_wwait(bo, 0, 0);
1168 		BO_UNLOCK(bo);
1169 		if (bo->bo_object != NULL) {
1170 			VM_OBJECT_LOCK(bo->bo_object);
1171 			vm_object_pip_wait(bo->bo_object, "bovlbx");
1172 			VM_OBJECT_UNLOCK(bo->bo_object);
1173 		}
1174 		BO_LOCK(bo);
1175 	} while (bo->bo_numoutput > 0);
1176 	BO_UNLOCK(bo);
1177 
1178 	/*
1179 	 * Destroy the copy in the VM cache, too.
1180 	 */
1181 	if (bo->bo_object != NULL && (flags & (V_ALT | V_NORMAL)) == 0) {
1182 		VM_OBJECT_LOCK(bo->bo_object);
1183 		vm_object_page_remove(bo->bo_object, 0, 0,
1184 			(flags & V_SAVE) ? TRUE : FALSE);
1185 		VM_OBJECT_UNLOCK(bo->bo_object);
1186 	}
1187 
1188 #ifdef INVARIANTS
1189 	BO_LOCK(bo);
1190 	if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1191 	    (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1192 		panic("vinvalbuf: flush failed");
1193 	BO_UNLOCK(bo);
1194 #endif
1195 	return (0);
1196 }
1197 
1198 /*
1199  * Flush out and invalidate all buffers associated with a vnode.
1200  * Called with the underlying object locked.
1201  */
1202 int
1203 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1204 {
1205 
1206 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1207 	ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1208 	return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1209 }
1210 
1211 /*
1212  * Flush out buffers on the specified list.
1213  *
1214  */
1215 static int
1216 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1217     int slptimeo)
1218 {
1219 	struct buf *bp, *nbp;
1220 	int retval, error;
1221 	daddr_t lblkno;
1222 	b_xflags_t xflags;
1223 
1224 	ASSERT_BO_LOCKED(bo);
1225 
1226 	retval = 0;
1227 	TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1228 		if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1229 		    ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1230 			continue;
1231 		}
1232 		lblkno = 0;
1233 		xflags = 0;
1234 		if (nbp != NULL) {
1235 			lblkno = nbp->b_lblkno;
1236 			xflags = nbp->b_xflags &
1237 				(BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1238 		}
1239 		retval = EAGAIN;
1240 		error = BUF_TIMELOCK(bp,
1241 		    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1242 		    "flushbuf", slpflag, slptimeo);
1243 		if (error) {
1244 			BO_LOCK(bo);
1245 			return (error != ENOLCK ? error : EAGAIN);
1246 		}
1247 		KASSERT(bp->b_bufobj == bo,
1248 		    ("bp %p wrong b_bufobj %p should be %p",
1249 		    bp, bp->b_bufobj, bo));
1250 		if (bp->b_bufobj != bo) {	/* XXX: necessary ? */
1251 			BUF_UNLOCK(bp);
1252 			BO_LOCK(bo);
1253 			return (EAGAIN);
1254 		}
1255 		/*
1256 		 * XXX Since there are no node locks for NFS, I
1257 		 * believe there is a slight chance that a delayed
1258 		 * write will occur while sleeping just above, so
1259 		 * check for it.
1260 		 */
1261 		if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1262 		    (flags & V_SAVE)) {
1263 			bremfree(bp);
1264 			bp->b_flags |= B_ASYNC;
1265 			bwrite(bp);
1266 			BO_LOCK(bo);
1267 			return (EAGAIN);	/* XXX: why not loop ? */
1268 		}
1269 		bremfree(bp);
1270 		bp->b_flags |= (B_INVAL | B_RELBUF);
1271 		bp->b_flags &= ~B_ASYNC;
1272 		brelse(bp);
1273 		BO_LOCK(bo);
1274 		if (nbp != NULL &&
1275 		    (nbp->b_bufobj != bo ||
1276 		     nbp->b_lblkno != lblkno ||
1277 		     (nbp->b_xflags &
1278 		      (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1279 			break;			/* nbp invalid */
1280 	}
1281 	return (retval);
1282 }
1283 
1284 /*
1285  * Truncate a file's buffer and pages to a specified length.  This
1286  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1287  * sync activity.
1288  */
1289 int
1290 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
1291     off_t length, int blksize)
1292 {
1293 	struct buf *bp, *nbp;
1294 	int anyfreed;
1295 	int trunclbn;
1296 	struct bufobj *bo;
1297 
1298 	CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1299 	    vp, cred, blksize, (uintmax_t)length);
1300 
1301 	/*
1302 	 * Round up to the *next* lbn.
1303 	 */
1304 	trunclbn = (length + blksize - 1) / blksize;
1305 
1306 	ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1307 restart:
1308 	bo = &vp->v_bufobj;
1309 	BO_LOCK(bo);
1310 	anyfreed = 1;
1311 	for (;anyfreed;) {
1312 		anyfreed = 0;
1313 		TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1314 			if (bp->b_lblkno < trunclbn)
1315 				continue;
1316 			if (BUF_LOCK(bp,
1317 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1318 			    BO_MTX(bo)) == ENOLCK)
1319 				goto restart;
1320 
1321 			bremfree(bp);
1322 			bp->b_flags |= (B_INVAL | B_RELBUF);
1323 			bp->b_flags &= ~B_ASYNC;
1324 			brelse(bp);
1325 			anyfreed = 1;
1326 
1327 			if (nbp != NULL &&
1328 			    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1329 			    (nbp->b_vp != vp) ||
1330 			    (nbp->b_flags & B_DELWRI))) {
1331 				goto restart;
1332 			}
1333 			BO_LOCK(bo);
1334 		}
1335 
1336 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1337 			if (bp->b_lblkno < trunclbn)
1338 				continue;
1339 			if (BUF_LOCK(bp,
1340 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1341 			    BO_MTX(bo)) == ENOLCK)
1342 				goto restart;
1343 			bremfree(bp);
1344 			bp->b_flags |= (B_INVAL | B_RELBUF);
1345 			bp->b_flags &= ~B_ASYNC;
1346 			brelse(bp);
1347 			anyfreed = 1;
1348 			if (nbp != NULL &&
1349 			    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1350 			    (nbp->b_vp != vp) ||
1351 			    (nbp->b_flags & B_DELWRI) == 0)) {
1352 				goto restart;
1353 			}
1354 			BO_LOCK(bo);
1355 		}
1356 	}
1357 
1358 	if (length > 0) {
1359 restartsync:
1360 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1361 			if (bp->b_lblkno > 0)
1362 				continue;
1363 			/*
1364 			 * Since we hold the vnode lock this should only
1365 			 * fail if we're racing with the buf daemon.
1366 			 */
1367 			if (BUF_LOCK(bp,
1368 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1369 			    BO_MTX(bo)) == ENOLCK) {
1370 				goto restart;
1371 			}
1372 			VNASSERT((bp->b_flags & B_DELWRI), vp,
1373 			    ("buf(%p) on dirty queue without DELWRI", bp));
1374 
1375 			bremfree(bp);
1376 			bawrite(bp);
1377 			BO_LOCK(bo);
1378 			goto restartsync;
1379 		}
1380 	}
1381 
1382 	bufobj_wwait(bo, 0, 0);
1383 	BO_UNLOCK(bo);
1384 	vnode_pager_setsize(vp, length);
1385 
1386 	return (0);
1387 }
1388 
1389 /*
1390  * buf_splay() - splay tree core for the clean/dirty list of buffers in
1391  *		 a vnode.
1392  *
1393  *	NOTE: We have to deal with the special case of a background bitmap
1394  *	buffer, a situation where two buffers will have the same logical
1395  *	block offset.  We want (1) only the foreground buffer to be accessed
1396  *	in a lookup and (2) must differentiate between the foreground and
1397  *	background buffer in the splay tree algorithm because the splay
1398  *	tree cannot normally handle multiple entities with the same 'index'.
1399  *	We accomplish this by adding differentiating flags to the splay tree's
1400  *	numerical domain.
1401  */
1402 static
1403 struct buf *
1404 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1405 {
1406 	struct buf dummy;
1407 	struct buf *lefttreemax, *righttreemin, *y;
1408 
1409 	if (root == NULL)
1410 		return (NULL);
1411 	lefttreemax = righttreemin = &dummy;
1412 	for (;;) {
1413 		if (lblkno < root->b_lblkno ||
1414 		    (lblkno == root->b_lblkno &&
1415 		    (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1416 			if ((y = root->b_left) == NULL)
1417 				break;
1418 			if (lblkno < y->b_lblkno) {
1419 				/* Rotate right. */
1420 				root->b_left = y->b_right;
1421 				y->b_right = root;
1422 				root = y;
1423 				if ((y = root->b_left) == NULL)
1424 					break;
1425 			}
1426 			/* Link into the new root's right tree. */
1427 			righttreemin->b_left = root;
1428 			righttreemin = root;
1429 		} else if (lblkno > root->b_lblkno ||
1430 		    (lblkno == root->b_lblkno &&
1431 		    (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1432 			if ((y = root->b_right) == NULL)
1433 				break;
1434 			if (lblkno > y->b_lblkno) {
1435 				/* Rotate left. */
1436 				root->b_right = y->b_left;
1437 				y->b_left = root;
1438 				root = y;
1439 				if ((y = root->b_right) == NULL)
1440 					break;
1441 			}
1442 			/* Link into the new root's left tree. */
1443 			lefttreemax->b_right = root;
1444 			lefttreemax = root;
1445 		} else {
1446 			break;
1447 		}
1448 		root = y;
1449 	}
1450 	/* Assemble the new root. */
1451 	lefttreemax->b_right = root->b_left;
1452 	righttreemin->b_left = root->b_right;
1453 	root->b_left = dummy.b_right;
1454 	root->b_right = dummy.b_left;
1455 	return (root);
1456 }
1457 
1458 static void
1459 buf_vlist_remove(struct buf *bp)
1460 {
1461 	struct buf *root;
1462 	struct bufv *bv;
1463 
1464 	KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1465 	ASSERT_BO_LOCKED(bp->b_bufobj);
1466 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1467 	    (BX_VNDIRTY|BX_VNCLEAN),
1468 	    ("buf_vlist_remove: Buf %p is on two lists", bp));
1469 	if (bp->b_xflags & BX_VNDIRTY)
1470 		bv = &bp->b_bufobj->bo_dirty;
1471 	else
1472 		bv = &bp->b_bufobj->bo_clean;
1473 	if (bp != bv->bv_root) {
1474 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1475 		KASSERT(root == bp, ("splay lookup failed in remove"));
1476 	}
1477 	if (bp->b_left == NULL) {
1478 		root = bp->b_right;
1479 	} else {
1480 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1481 		root->b_right = bp->b_right;
1482 	}
1483 	bv->bv_root = root;
1484 	TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1485 	bv->bv_cnt--;
1486 	bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1487 }
1488 
1489 /*
1490  * Add the buffer to the sorted clean or dirty block list using a
1491  * splay tree algorithm.
1492  *
1493  * NOTE: xflags is passed as a constant, optimizing this inline function!
1494  */
1495 static void
1496 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1497 {
1498 	struct buf *root;
1499 	struct bufv *bv;
1500 
1501 	ASSERT_BO_LOCKED(bo);
1502 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1503 	    ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1504 	bp->b_xflags |= xflags;
1505 	if (xflags & BX_VNDIRTY)
1506 		bv = &bo->bo_dirty;
1507 	else
1508 		bv = &bo->bo_clean;
1509 
1510 	root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1511 	if (root == NULL) {
1512 		bp->b_left = NULL;
1513 		bp->b_right = NULL;
1514 		TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1515 	} else if (bp->b_lblkno < root->b_lblkno ||
1516 	    (bp->b_lblkno == root->b_lblkno &&
1517 	    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1518 		bp->b_left = root->b_left;
1519 		bp->b_right = root;
1520 		root->b_left = NULL;
1521 		TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1522 	} else {
1523 		bp->b_right = root->b_right;
1524 		bp->b_left = root;
1525 		root->b_right = NULL;
1526 		TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1527 	}
1528 	bv->bv_cnt++;
1529 	bv->bv_root = bp;
1530 }
1531 
1532 /*
1533  * Lookup a buffer using the splay tree.  Note that we specifically avoid
1534  * shadow buffers used in background bitmap writes.
1535  *
1536  * This code isn't quite efficient as it could be because we are maintaining
1537  * two sorted lists and do not know which list the block resides in.
1538  *
1539  * During a "make buildworld" the desired buffer is found at one of
1540  * the roots more than 60% of the time.  Thus, checking both roots
1541  * before performing either splay eliminates unnecessary splays on the
1542  * first tree splayed.
1543  */
1544 struct buf *
1545 gbincore(struct bufobj *bo, daddr_t lblkno)
1546 {
1547 	struct buf *bp;
1548 
1549 	ASSERT_BO_LOCKED(bo);
1550 	if ((bp = bo->bo_clean.bv_root) != NULL &&
1551 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1552 		return (bp);
1553 	if ((bp = bo->bo_dirty.bv_root) != NULL &&
1554 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1555 		return (bp);
1556 	if ((bp = bo->bo_clean.bv_root) != NULL) {
1557 		bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1558 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1559 			return (bp);
1560 	}
1561 	if ((bp = bo->bo_dirty.bv_root) != NULL) {
1562 		bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1563 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1564 			return (bp);
1565 	}
1566 	return (NULL);
1567 }
1568 
1569 /*
1570  * Associate a buffer with a vnode.
1571  */
1572 void
1573 bgetvp(struct vnode *vp, struct buf *bp)
1574 {
1575 	struct bufobj *bo;
1576 
1577 	bo = &vp->v_bufobj;
1578 	ASSERT_BO_LOCKED(bo);
1579 	VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1580 
1581 	CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1582 	VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1583 	    ("bgetvp: bp already attached! %p", bp));
1584 
1585 	vhold(vp);
1586 	if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT)
1587 		bp->b_flags |= B_NEEDSGIANT;
1588 	bp->b_vp = vp;
1589 	bp->b_bufobj = bo;
1590 	/*
1591 	 * Insert onto list for new vnode.
1592 	 */
1593 	buf_vlist_add(bp, bo, BX_VNCLEAN);
1594 }
1595 
1596 /*
1597  * Disassociate a buffer from a vnode.
1598  */
1599 void
1600 brelvp(struct buf *bp)
1601 {
1602 	struct bufobj *bo;
1603 	struct vnode *vp;
1604 
1605 	CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1606 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1607 
1608 	/*
1609 	 * Delete from old vnode list, if on one.
1610 	 */
1611 	vp = bp->b_vp;		/* XXX */
1612 	bo = bp->b_bufobj;
1613 	BO_LOCK(bo);
1614 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1615 		buf_vlist_remove(bp);
1616 	else
1617 		panic("brelvp: Buffer %p not on queue.", bp);
1618 	if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1619 		bo->bo_flag &= ~BO_ONWORKLST;
1620 		mtx_lock(&sync_mtx);
1621 		LIST_REMOVE(bo, bo_synclist);
1622 		syncer_worklist_len--;
1623 		mtx_unlock(&sync_mtx);
1624 	}
1625 	bp->b_flags &= ~B_NEEDSGIANT;
1626 	bp->b_vp = NULL;
1627 	bp->b_bufobj = NULL;
1628 	BO_UNLOCK(bo);
1629 	vdrop(vp);
1630 }
1631 
1632 /*
1633  * Add an item to the syncer work queue.
1634  */
1635 static void
1636 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1637 {
1638 	int queue, slot;
1639 
1640 	ASSERT_BO_LOCKED(bo);
1641 
1642 	mtx_lock(&sync_mtx);
1643 	if (bo->bo_flag & BO_ONWORKLST)
1644 		LIST_REMOVE(bo, bo_synclist);
1645 	else {
1646 		bo->bo_flag |= BO_ONWORKLST;
1647 		syncer_worklist_len++;
1648 	}
1649 
1650 	if (delay > syncer_maxdelay - 2)
1651 		delay = syncer_maxdelay - 2;
1652 	slot = (syncer_delayno + delay) & syncer_mask;
1653 
1654 	queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ :
1655 	    WI_MPSAFEQ;
1656 	LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo,
1657 	    bo_synclist);
1658 	mtx_unlock(&sync_mtx);
1659 }
1660 
1661 static int
1662 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1663 {
1664 	int error, len;
1665 
1666 	mtx_lock(&sync_mtx);
1667 	len = syncer_worklist_len - sync_vnode_count;
1668 	mtx_unlock(&sync_mtx);
1669 	error = SYSCTL_OUT(req, &len, sizeof(len));
1670 	return (error);
1671 }
1672 
1673 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1674     sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1675 
1676 static struct proc *updateproc;
1677 static void sched_sync(void);
1678 static struct kproc_desc up_kp = {
1679 	"syncer",
1680 	sched_sync,
1681 	&updateproc
1682 };
1683 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1684 
1685 static int
1686 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1687 {
1688 	struct vnode *vp;
1689 	struct mount *mp;
1690 
1691 	*bo = LIST_FIRST(slp);
1692 	if (*bo == NULL)
1693 		return (0);
1694 	vp = (*bo)->__bo_vnode;	/* XXX */
1695 	if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1696 		return (1);
1697 	/*
1698 	 * We use vhold in case the vnode does not
1699 	 * successfully sync.  vhold prevents the vnode from
1700 	 * going away when we unlock the sync_mtx so that
1701 	 * we can acquire the vnode interlock.
1702 	 */
1703 	vholdl(vp);
1704 	mtx_unlock(&sync_mtx);
1705 	VI_UNLOCK(vp);
1706 	if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1707 		vdrop(vp);
1708 		mtx_lock(&sync_mtx);
1709 		return (*bo == LIST_FIRST(slp));
1710 	}
1711 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1712 	(void) VOP_FSYNC(vp, MNT_LAZY, td);
1713 	VOP_UNLOCK(vp, 0);
1714 	vn_finished_write(mp);
1715 	BO_LOCK(*bo);
1716 	if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1717 		/*
1718 		 * Put us back on the worklist.  The worklist
1719 		 * routine will remove us from our current
1720 		 * position and then add us back in at a later
1721 		 * position.
1722 		 */
1723 		vn_syncer_add_to_worklist(*bo, syncdelay);
1724 	}
1725 	BO_UNLOCK(*bo);
1726 	vdrop(vp);
1727 	mtx_lock(&sync_mtx);
1728 	return (0);
1729 }
1730 
1731 /*
1732  * System filesystem synchronizer daemon.
1733  */
1734 static void
1735 sched_sync(void)
1736 {
1737 	struct synclist *gnext, *next;
1738 	struct synclist *gslp, *slp;
1739 	struct bufobj *bo;
1740 	long starttime;
1741 	struct thread *td = curthread;
1742 	int last_work_seen;
1743 	int net_worklist_len;
1744 	int syncer_final_iter;
1745 	int first_printf;
1746 	int error;
1747 
1748 	last_work_seen = 0;
1749 	syncer_final_iter = 0;
1750 	first_printf = 1;
1751 	syncer_state = SYNCER_RUNNING;
1752 	starttime = time_uptime;
1753 	td->td_pflags |= TDP_NORUNNINGBUF;
1754 
1755 	EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1756 	    SHUTDOWN_PRI_LAST);
1757 
1758 	mtx_lock(&sync_mtx);
1759 	for (;;) {
1760 		if (syncer_state == SYNCER_FINAL_DELAY &&
1761 		    syncer_final_iter == 0) {
1762 			mtx_unlock(&sync_mtx);
1763 			kproc_suspend_check(td->td_proc);
1764 			mtx_lock(&sync_mtx);
1765 		}
1766 		net_worklist_len = syncer_worklist_len - sync_vnode_count;
1767 		if (syncer_state != SYNCER_RUNNING &&
1768 		    starttime != time_uptime) {
1769 			if (first_printf) {
1770 				printf("\nSyncing disks, vnodes remaining...");
1771 				first_printf = 0;
1772 			}
1773 			printf("%d ", net_worklist_len);
1774 		}
1775 		starttime = time_uptime;
1776 
1777 		/*
1778 		 * Push files whose dirty time has expired.  Be careful
1779 		 * of interrupt race on slp queue.
1780 		 *
1781 		 * Skip over empty worklist slots when shutting down.
1782 		 */
1783 		do {
1784 			slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1785 			gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1786 			syncer_delayno += 1;
1787 			if (syncer_delayno == syncer_maxdelay)
1788 				syncer_delayno = 0;
1789 			next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1790 			gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1791 			/*
1792 			 * If the worklist has wrapped since the
1793 			 * it was emptied of all but syncer vnodes,
1794 			 * switch to the FINAL_DELAY state and run
1795 			 * for one more second.
1796 			 */
1797 			if (syncer_state == SYNCER_SHUTTING_DOWN &&
1798 			    net_worklist_len == 0 &&
1799 			    last_work_seen == syncer_delayno) {
1800 				syncer_state = SYNCER_FINAL_DELAY;
1801 				syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1802 			}
1803 		} while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1804 		    LIST_EMPTY(gslp) && syncer_worklist_len > 0);
1805 
1806 		/*
1807 		 * Keep track of the last time there was anything
1808 		 * on the worklist other than syncer vnodes.
1809 		 * Return to the SHUTTING_DOWN state if any
1810 		 * new work appears.
1811 		 */
1812 		if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1813 			last_work_seen = syncer_delayno;
1814 		if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1815 			syncer_state = SYNCER_SHUTTING_DOWN;
1816 		while (!LIST_EMPTY(slp)) {
1817 			error = sync_vnode(slp, &bo, td);
1818 			if (error == 1) {
1819 				LIST_REMOVE(bo, bo_synclist);
1820 				LIST_INSERT_HEAD(next, bo, bo_synclist);
1821 				continue;
1822 			}
1823 		}
1824 		if (!LIST_EMPTY(gslp)) {
1825 			mtx_unlock(&sync_mtx);
1826 			mtx_lock(&Giant);
1827 			mtx_lock(&sync_mtx);
1828 			while (!LIST_EMPTY(gslp)) {
1829 				error = sync_vnode(gslp, &bo, td);
1830 				if (error == 1) {
1831 					LIST_REMOVE(bo, bo_synclist);
1832 					LIST_INSERT_HEAD(gnext, bo,
1833 					    bo_synclist);
1834 					continue;
1835 				}
1836 			}
1837 			mtx_unlock(&Giant);
1838 		}
1839 		if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1840 			syncer_final_iter--;
1841 		/*
1842 		 * The variable rushjob allows the kernel to speed up the
1843 		 * processing of the filesystem syncer process. A rushjob
1844 		 * value of N tells the filesystem syncer to process the next
1845 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1846 		 * is used by the soft update code to speed up the filesystem
1847 		 * syncer process when the incore state is getting so far
1848 		 * ahead of the disk that the kernel memory pool is being
1849 		 * threatened with exhaustion.
1850 		 */
1851 		if (rushjob > 0) {
1852 			rushjob -= 1;
1853 			continue;
1854 		}
1855 		/*
1856 		 * Just sleep for a short period of time between
1857 		 * iterations when shutting down to allow some I/O
1858 		 * to happen.
1859 		 *
1860 		 * If it has taken us less than a second to process the
1861 		 * current work, then wait. Otherwise start right over
1862 		 * again. We can still lose time if any single round
1863 		 * takes more than two seconds, but it does not really
1864 		 * matter as we are just trying to generally pace the
1865 		 * filesystem activity.
1866 		 */
1867 		if (syncer_state != SYNCER_RUNNING)
1868 			cv_timedwait(&sync_wakeup, &sync_mtx,
1869 			    hz / SYNCER_SHUTDOWN_SPEEDUP);
1870 		else if (time_uptime == starttime)
1871 			cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1872 	}
1873 }
1874 
1875 /*
1876  * Request the syncer daemon to speed up its work.
1877  * We never push it to speed up more than half of its
1878  * normal turn time, otherwise it could take over the cpu.
1879  */
1880 int
1881 speedup_syncer(void)
1882 {
1883 	int ret = 0;
1884 
1885 	mtx_lock(&sync_mtx);
1886 	if (rushjob < syncdelay / 2) {
1887 		rushjob += 1;
1888 		stat_rush_requests += 1;
1889 		ret = 1;
1890 	}
1891 	mtx_unlock(&sync_mtx);
1892 	cv_broadcast(&sync_wakeup);
1893 	return (ret);
1894 }
1895 
1896 /*
1897  * Tell the syncer to speed up its work and run though its work
1898  * list several times, then tell it to shut down.
1899  */
1900 static void
1901 syncer_shutdown(void *arg, int howto)
1902 {
1903 
1904 	if (howto & RB_NOSYNC)
1905 		return;
1906 	mtx_lock(&sync_mtx);
1907 	syncer_state = SYNCER_SHUTTING_DOWN;
1908 	rushjob = 0;
1909 	mtx_unlock(&sync_mtx);
1910 	cv_broadcast(&sync_wakeup);
1911 	kproc_shutdown(arg, howto);
1912 }
1913 
1914 /*
1915  * Reassign a buffer from one vnode to another.
1916  * Used to assign file specific control information
1917  * (indirect blocks) to the vnode to which they belong.
1918  */
1919 void
1920 reassignbuf(struct buf *bp)
1921 {
1922 	struct vnode *vp;
1923 	struct bufobj *bo;
1924 	int delay;
1925 #ifdef INVARIANTS
1926 	struct bufv *bv;
1927 #endif
1928 
1929 	vp = bp->b_vp;
1930 	bo = bp->b_bufobj;
1931 	++reassignbufcalls;
1932 
1933 	CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1934 	    bp, bp->b_vp, bp->b_flags);
1935 	/*
1936 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1937 	 * is not fully linked in.
1938 	 */
1939 	if (bp->b_flags & B_PAGING)
1940 		panic("cannot reassign paging buffer");
1941 
1942 	/*
1943 	 * Delete from old vnode list, if on one.
1944 	 */
1945 	BO_LOCK(bo);
1946 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1947 		buf_vlist_remove(bp);
1948 	else
1949 		panic("reassignbuf: Buffer %p not on queue.", bp);
1950 	/*
1951 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1952 	 * of clean buffers.
1953 	 */
1954 	if (bp->b_flags & B_DELWRI) {
1955 		if ((bo->bo_flag & BO_ONWORKLST) == 0) {
1956 			switch (vp->v_type) {
1957 			case VDIR:
1958 				delay = dirdelay;
1959 				break;
1960 			case VCHR:
1961 				delay = metadelay;
1962 				break;
1963 			default:
1964 				delay = filedelay;
1965 			}
1966 			vn_syncer_add_to_worklist(bo, delay);
1967 		}
1968 		buf_vlist_add(bp, bo, BX_VNDIRTY);
1969 	} else {
1970 		buf_vlist_add(bp, bo, BX_VNCLEAN);
1971 
1972 		if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1973 			mtx_lock(&sync_mtx);
1974 			LIST_REMOVE(bo, bo_synclist);
1975 			syncer_worklist_len--;
1976 			mtx_unlock(&sync_mtx);
1977 			bo->bo_flag &= ~BO_ONWORKLST;
1978 		}
1979 	}
1980 #ifdef INVARIANTS
1981 	bv = &bo->bo_clean;
1982 	bp = TAILQ_FIRST(&bv->bv_hd);
1983 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1984 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1985 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
1986 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1987 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1988 	bv = &bo->bo_dirty;
1989 	bp = TAILQ_FIRST(&bv->bv_hd);
1990 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1991 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1992 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
1993 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1994 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1995 #endif
1996 	BO_UNLOCK(bo);
1997 }
1998 
1999 /*
2000  * Increment the use and hold counts on the vnode, taking care to reference
2001  * the driver's usecount if this is a chardev.  The vholdl() will remove
2002  * the vnode from the free list if it is presently free.  Requires the
2003  * vnode interlock and returns with it held.
2004  */
2005 static void
2006 v_incr_usecount(struct vnode *vp)
2007 {
2008 
2009 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2010 	vp->v_usecount++;
2011 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2012 		dev_lock();
2013 		vp->v_rdev->si_usecount++;
2014 		dev_unlock();
2015 	}
2016 	vholdl(vp);
2017 }
2018 
2019 /*
2020  * Turn a holdcnt into a use+holdcnt such that only one call to
2021  * v_decr_usecount is needed.
2022  */
2023 static void
2024 v_upgrade_usecount(struct vnode *vp)
2025 {
2026 
2027 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2028 	vp->v_usecount++;
2029 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2030 		dev_lock();
2031 		vp->v_rdev->si_usecount++;
2032 		dev_unlock();
2033 	}
2034 }
2035 
2036 /*
2037  * Decrement the vnode use and hold count along with the driver's usecount
2038  * if this is a chardev.  The vdropl() below releases the vnode interlock
2039  * as it may free the vnode.
2040  */
2041 static void
2042 v_decr_usecount(struct vnode *vp)
2043 {
2044 
2045 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2046 	VNASSERT(vp->v_usecount > 0, vp,
2047 	    ("v_decr_usecount: negative usecount"));
2048 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2049 	vp->v_usecount--;
2050 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2051 		dev_lock();
2052 		vp->v_rdev->si_usecount--;
2053 		dev_unlock();
2054 	}
2055 	vdropl(vp);
2056 }
2057 
2058 /*
2059  * Decrement only the use count and driver use count.  This is intended to
2060  * be paired with a follow on vdropl() to release the remaining hold count.
2061  * In this way we may vgone() a vnode with a 0 usecount without risk of
2062  * having it end up on a free list because the hold count is kept above 0.
2063  */
2064 static void
2065 v_decr_useonly(struct vnode *vp)
2066 {
2067 
2068 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
2069 	VNASSERT(vp->v_usecount > 0, vp,
2070 	    ("v_decr_useonly: negative usecount"));
2071 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2072 	vp->v_usecount--;
2073 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2074 		dev_lock();
2075 		vp->v_rdev->si_usecount--;
2076 		dev_unlock();
2077 	}
2078 }
2079 
2080 /*
2081  * Grab a particular vnode from the free list, increment its
2082  * reference count and lock it.  VI_DOOMED is set if the vnode
2083  * is being destroyed.  Only callers who specify LK_RETRY will
2084  * see doomed vnodes.  If inactive processing was delayed in
2085  * vput try to do it here.
2086  */
2087 int
2088 vget(struct vnode *vp, int flags, struct thread *td)
2089 {
2090 	int error;
2091 
2092 	error = 0;
2093 	VFS_ASSERT_GIANT(vp->v_mount);
2094 	VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2095 	    ("vget: invalid lock operation"));
2096 	CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2097 
2098 	if ((flags & LK_INTERLOCK) == 0)
2099 		VI_LOCK(vp);
2100 	vholdl(vp);
2101 	if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2102 		vdrop(vp);
2103 		CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2104 		    vp);
2105 		return (error);
2106 	}
2107 	if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2108 		panic("vget: vn_lock failed to return ENOENT\n");
2109 	VI_LOCK(vp);
2110 	/* Upgrade our holdcnt to a usecount. */
2111 	v_upgrade_usecount(vp);
2112 	/*
2113 	 * We don't guarantee that any particular close will
2114 	 * trigger inactive processing so just make a best effort
2115 	 * here at preventing a reference to a removed file.  If
2116 	 * we don't succeed no harm is done.
2117 	 */
2118 	if (vp->v_iflag & VI_OWEINACT) {
2119 		if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2120 		    (flags & LK_NOWAIT) == 0)
2121 			vinactive(vp, td);
2122 		vp->v_iflag &= ~VI_OWEINACT;
2123 	}
2124 	VI_UNLOCK(vp);
2125 	return (0);
2126 }
2127 
2128 /*
2129  * Increase the reference count of a vnode.
2130  */
2131 void
2132 vref(struct vnode *vp)
2133 {
2134 
2135 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2136 	VI_LOCK(vp);
2137 	v_incr_usecount(vp);
2138 	VI_UNLOCK(vp);
2139 }
2140 
2141 /*
2142  * Return reference count of a vnode.
2143  *
2144  * The results of this call are only guaranteed when some mechanism other
2145  * than the VI lock is used to stop other processes from gaining references
2146  * to the vnode.  This may be the case if the caller holds the only reference.
2147  * This is also useful when stale data is acceptable as race conditions may
2148  * be accounted for by some other means.
2149  */
2150 int
2151 vrefcnt(struct vnode *vp)
2152 {
2153 	int usecnt;
2154 
2155 	VI_LOCK(vp);
2156 	usecnt = vp->v_usecount;
2157 	VI_UNLOCK(vp);
2158 
2159 	return (usecnt);
2160 }
2161 
2162 #define	VPUTX_VRELE	1
2163 #define	VPUTX_VPUT	2
2164 #define	VPUTX_VUNREF	3
2165 
2166 static void
2167 vputx(struct vnode *vp, int func)
2168 {
2169 	int error;
2170 
2171 	KASSERT(vp != NULL, ("vputx: null vp"));
2172 	if (func == VPUTX_VUNREF)
2173 		ASSERT_VOP_ELOCKED(vp, "vunref");
2174 	else if (func == VPUTX_VPUT)
2175 		ASSERT_VOP_LOCKED(vp, "vput");
2176 	else
2177 		KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2178 	VFS_ASSERT_GIANT(vp->v_mount);
2179 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2180 	VI_LOCK(vp);
2181 
2182 	/* Skip this v_writecount check if we're going to panic below. */
2183 	VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2184 	    ("vputx: missed vn_close"));
2185 	error = 0;
2186 
2187 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2188 	    vp->v_usecount == 1)) {
2189 		if (func == VPUTX_VPUT)
2190 			VOP_UNLOCK(vp, 0);
2191 		v_decr_usecount(vp);
2192 		return;
2193 	}
2194 
2195 	if (vp->v_usecount != 1) {
2196 #ifdef DIAGNOSTIC
2197 		vprint("vputx: negative ref count", vp);
2198 #endif
2199 		panic("vputx: negative ref cnt");
2200 	}
2201 	CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2202 	/*
2203 	 * We want to hold the vnode until the inactive finishes to
2204 	 * prevent vgone() races.  We drop the use count here and the
2205 	 * hold count below when we're done.
2206 	 */
2207 	v_decr_useonly(vp);
2208 	/*
2209 	 * We must call VOP_INACTIVE with the node locked. Mark
2210 	 * as VI_DOINGINACT to avoid recursion.
2211 	 */
2212 	vp->v_iflag |= VI_OWEINACT;
2213 	if (func == VPUTX_VRELE) {
2214 		error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2215 		VI_LOCK(vp);
2216 	} else if (func == VPUTX_VPUT && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2217 		error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK | LK_NOWAIT);
2218 		VI_LOCK(vp);
2219 	}
2220 	if (vp->v_usecount > 0)
2221 		vp->v_iflag &= ~VI_OWEINACT;
2222 	if (error == 0) {
2223 		if (vp->v_iflag & VI_OWEINACT)
2224 			vinactive(vp, curthread);
2225 		if (func != VPUTX_VUNREF)
2226 			VOP_UNLOCK(vp, 0);
2227 	}
2228 	vdropl(vp);
2229 }
2230 
2231 /*
2232  * Vnode put/release.
2233  * If count drops to zero, call inactive routine and return to freelist.
2234  */
2235 void
2236 vrele(struct vnode *vp)
2237 {
2238 
2239 	vputx(vp, VPUTX_VRELE);
2240 }
2241 
2242 /*
2243  * Release an already locked vnode.  This give the same effects as
2244  * unlock+vrele(), but takes less time and avoids releasing and
2245  * re-aquiring the lock (as vrele() acquires the lock internally.)
2246  */
2247 void
2248 vput(struct vnode *vp)
2249 {
2250 
2251 	vputx(vp, VPUTX_VPUT);
2252 }
2253 
2254 /*
2255  * Release an exclusively locked vnode. Do not unlock the vnode lock.
2256  */
2257 void
2258 vunref(struct vnode *vp)
2259 {
2260 
2261 	vputx(vp, VPUTX_VUNREF);
2262 }
2263 
2264 /*
2265  * Somebody doesn't want the vnode recycled.
2266  */
2267 void
2268 vhold(struct vnode *vp)
2269 {
2270 
2271 	VI_LOCK(vp);
2272 	vholdl(vp);
2273 	VI_UNLOCK(vp);
2274 }
2275 
2276 void
2277 vholdl(struct vnode *vp)
2278 {
2279 
2280 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2281 	vp->v_holdcnt++;
2282 	if (VSHOULDBUSY(vp))
2283 		vbusy(vp);
2284 }
2285 
2286 /*
2287  * Note that there is one less who cares about this vnode.  vdrop() is the
2288  * opposite of vhold().
2289  */
2290 void
2291 vdrop(struct vnode *vp)
2292 {
2293 
2294 	VI_LOCK(vp);
2295 	vdropl(vp);
2296 }
2297 
2298 /*
2299  * Drop the hold count of the vnode.  If this is the last reference to
2300  * the vnode we will free it if it has been vgone'd otherwise it is
2301  * placed on the free list.
2302  */
2303 void
2304 vdropl(struct vnode *vp)
2305 {
2306 
2307 	ASSERT_VI_LOCKED(vp, "vdropl");
2308 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2309 	if (vp->v_holdcnt <= 0)
2310 		panic("vdrop: holdcnt %d", vp->v_holdcnt);
2311 	vp->v_holdcnt--;
2312 	if (vp->v_holdcnt == 0) {
2313 		if (vp->v_iflag & VI_DOOMED) {
2314 			CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__,
2315 			    vp);
2316 			vdestroy(vp);
2317 			return;
2318 		} else
2319 			vfree(vp);
2320 	}
2321 	VI_UNLOCK(vp);
2322 }
2323 
2324 /*
2325  * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2326  * flags.  DOINGINACT prevents us from recursing in calls to vinactive.
2327  * OWEINACT tracks whether a vnode missed a call to inactive due to a
2328  * failed lock upgrade.
2329  */
2330 static void
2331 vinactive(struct vnode *vp, struct thread *td)
2332 {
2333 
2334 	ASSERT_VOP_ELOCKED(vp, "vinactive");
2335 	ASSERT_VI_LOCKED(vp, "vinactive");
2336 	VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2337 	    ("vinactive: recursed on VI_DOINGINACT"));
2338 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2339 	vp->v_iflag |= VI_DOINGINACT;
2340 	vp->v_iflag &= ~VI_OWEINACT;
2341 	VI_UNLOCK(vp);
2342 	VOP_INACTIVE(vp, td);
2343 	VI_LOCK(vp);
2344 	VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2345 	    ("vinactive: lost VI_DOINGINACT"));
2346 	vp->v_iflag &= ~VI_DOINGINACT;
2347 }
2348 
2349 /*
2350  * Remove any vnodes in the vnode table belonging to mount point mp.
2351  *
2352  * If FORCECLOSE is not specified, there should not be any active ones,
2353  * return error if any are found (nb: this is a user error, not a
2354  * system error). If FORCECLOSE is specified, detach any active vnodes
2355  * that are found.
2356  *
2357  * If WRITECLOSE is set, only flush out regular file vnodes open for
2358  * writing.
2359  *
2360  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2361  *
2362  * `rootrefs' specifies the base reference count for the root vnode
2363  * of this filesystem. The root vnode is considered busy if its
2364  * v_usecount exceeds this value. On a successful return, vflush(, td)
2365  * will call vrele() on the root vnode exactly rootrefs times.
2366  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2367  * be zero.
2368  */
2369 #ifdef DIAGNOSTIC
2370 static int busyprt = 0;		/* print out busy vnodes */
2371 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
2372 #endif
2373 
2374 int
2375 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
2376 {
2377 	struct vnode *vp, *mvp, *rootvp = NULL;
2378 	struct vattr vattr;
2379 	int busy = 0, error;
2380 
2381 	CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2382 	    rootrefs, flags);
2383 	if (rootrefs > 0) {
2384 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2385 		    ("vflush: bad args"));
2386 		/*
2387 		 * Get the filesystem root vnode. We can vput() it
2388 		 * immediately, since with rootrefs > 0, it won't go away.
2389 		 */
2390 		if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2391 			CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2392 			    __func__, error);
2393 			return (error);
2394 		}
2395 		vput(rootvp);
2396 	}
2397 	MNT_ILOCK(mp);
2398 loop:
2399 	MNT_VNODE_FOREACH(vp, mp, mvp) {
2400 		VI_LOCK(vp);
2401 		vholdl(vp);
2402 		MNT_IUNLOCK(mp);
2403 		error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2404 		if (error) {
2405 			vdrop(vp);
2406 			MNT_ILOCK(mp);
2407 			MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
2408 			goto loop;
2409 		}
2410 		/*
2411 		 * Skip over a vnodes marked VV_SYSTEM.
2412 		 */
2413 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2414 			VOP_UNLOCK(vp, 0);
2415 			vdrop(vp);
2416 			MNT_ILOCK(mp);
2417 			continue;
2418 		}
2419 		/*
2420 		 * If WRITECLOSE is set, flush out unlinked but still open
2421 		 * files (even if open only for reading) and regular file
2422 		 * vnodes open for writing.
2423 		 */
2424 		if (flags & WRITECLOSE) {
2425 			error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2426 			VI_LOCK(vp);
2427 
2428 			if ((vp->v_type == VNON ||
2429 			    (error == 0 && vattr.va_nlink > 0)) &&
2430 			    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2431 				VOP_UNLOCK(vp, 0);
2432 				vdropl(vp);
2433 				MNT_ILOCK(mp);
2434 				continue;
2435 			}
2436 		} else
2437 			VI_LOCK(vp);
2438 		/*
2439 		 * With v_usecount == 0, all we need to do is clear out the
2440 		 * vnode data structures and we are done.
2441 		 *
2442 		 * If FORCECLOSE is set, forcibly close the vnode.
2443 		 */
2444 		if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2445 			VNASSERT(vp->v_usecount == 0 ||
2446 			    (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2447 			    ("device VNODE %p is FORCECLOSED", vp));
2448 			vgonel(vp);
2449 		} else {
2450 			busy++;
2451 #ifdef DIAGNOSTIC
2452 			if (busyprt)
2453 				vprint("vflush: busy vnode", vp);
2454 #endif
2455 		}
2456 		VOP_UNLOCK(vp, 0);
2457 		vdropl(vp);
2458 		MNT_ILOCK(mp);
2459 	}
2460 	MNT_IUNLOCK(mp);
2461 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2462 		/*
2463 		 * If just the root vnode is busy, and if its refcount
2464 		 * is equal to `rootrefs', then go ahead and kill it.
2465 		 */
2466 		VI_LOCK(rootvp);
2467 		KASSERT(busy > 0, ("vflush: not busy"));
2468 		VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2469 		    ("vflush: usecount %d < rootrefs %d",
2470 		     rootvp->v_usecount, rootrefs));
2471 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2472 			VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2473 			vgone(rootvp);
2474 			VOP_UNLOCK(rootvp, 0);
2475 			busy = 0;
2476 		} else
2477 			VI_UNLOCK(rootvp);
2478 	}
2479 	if (busy) {
2480 		CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2481 		    busy);
2482 		return (EBUSY);
2483 	}
2484 	for (; rootrefs > 0; rootrefs--)
2485 		vrele(rootvp);
2486 	return (0);
2487 }
2488 
2489 /*
2490  * Recycle an unused vnode to the front of the free list.
2491  */
2492 int
2493 vrecycle(struct vnode *vp, struct thread *td)
2494 {
2495 	int recycled;
2496 
2497 	ASSERT_VOP_ELOCKED(vp, "vrecycle");
2498 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2499 	recycled = 0;
2500 	VI_LOCK(vp);
2501 	if (vp->v_usecount == 0) {
2502 		recycled = 1;
2503 		vgonel(vp);
2504 	}
2505 	VI_UNLOCK(vp);
2506 	return (recycled);
2507 }
2508 
2509 /*
2510  * Eliminate all activity associated with a vnode
2511  * in preparation for reuse.
2512  */
2513 void
2514 vgone(struct vnode *vp)
2515 {
2516 	VI_LOCK(vp);
2517 	vgonel(vp);
2518 	VI_UNLOCK(vp);
2519 }
2520 
2521 /*
2522  * vgone, with the vp interlock held.
2523  */
2524 void
2525 vgonel(struct vnode *vp)
2526 {
2527 	struct thread *td;
2528 	int oweinact;
2529 	int active;
2530 	struct mount *mp;
2531 
2532 	ASSERT_VOP_ELOCKED(vp, "vgonel");
2533 	ASSERT_VI_LOCKED(vp, "vgonel");
2534 	VNASSERT(vp->v_holdcnt, vp,
2535 	    ("vgonel: vp %p has no reference.", vp));
2536 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2537 	td = curthread;
2538 
2539 	/*
2540 	 * Don't vgonel if we're already doomed.
2541 	 */
2542 	if (vp->v_iflag & VI_DOOMED)
2543 		return;
2544 	vp->v_iflag |= VI_DOOMED;
2545 	/*
2546 	 * Check to see if the vnode is in use.  If so, we have to call
2547 	 * VOP_CLOSE() and VOP_INACTIVE().
2548 	 */
2549 	active = vp->v_usecount;
2550 	oweinact = (vp->v_iflag & VI_OWEINACT);
2551 	VI_UNLOCK(vp);
2552 	/*
2553 	 * Clean out any buffers associated with the vnode.
2554 	 * If the flush fails, just toss the buffers.
2555 	 */
2556 	mp = NULL;
2557 	if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2558 		(void) vn_start_secondary_write(vp, &mp, V_WAIT);
2559 	if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
2560 		vinvalbuf(vp, 0, 0, 0);
2561 
2562 	/*
2563 	 * If purging an active vnode, it must be closed and
2564 	 * deactivated before being reclaimed.
2565 	 */
2566 	if (active)
2567 		VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2568 	if (oweinact || active) {
2569 		VI_LOCK(vp);
2570 		if ((vp->v_iflag & VI_DOINGINACT) == 0)
2571 			vinactive(vp, td);
2572 		VI_UNLOCK(vp);
2573 	}
2574 	/*
2575 	 * Reclaim the vnode.
2576 	 */
2577 	if (VOP_RECLAIM(vp, td))
2578 		panic("vgone: cannot reclaim");
2579 	if (mp != NULL)
2580 		vn_finished_secondary_write(mp);
2581 	VNASSERT(vp->v_object == NULL, vp,
2582 	    ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2583 	/*
2584 	 * Clear the advisory locks and wake up waiting threads.
2585 	 */
2586 	(void)VOP_ADVLOCKPURGE(vp);
2587 	/*
2588 	 * Delete from old mount point vnode list.
2589 	 */
2590 	delmntque(vp);
2591 	cache_purge(vp);
2592 	/*
2593 	 * Done with purge, reset to the standard lock and invalidate
2594 	 * the vnode.
2595 	 */
2596 	VI_LOCK(vp);
2597 	vp->v_vnlock = &vp->v_lock;
2598 	vp->v_op = &dead_vnodeops;
2599 	vp->v_tag = "none";
2600 	vp->v_type = VBAD;
2601 }
2602 
2603 /*
2604  * Calculate the total number of references to a special device.
2605  */
2606 int
2607 vcount(struct vnode *vp)
2608 {
2609 	int count;
2610 
2611 	dev_lock();
2612 	count = vp->v_rdev->si_usecount;
2613 	dev_unlock();
2614 	return (count);
2615 }
2616 
2617 /*
2618  * Same as above, but using the struct cdev *as argument
2619  */
2620 int
2621 count_dev(struct cdev *dev)
2622 {
2623 	int count;
2624 
2625 	dev_lock();
2626 	count = dev->si_usecount;
2627 	dev_unlock();
2628 	return(count);
2629 }
2630 
2631 /*
2632  * Print out a description of a vnode.
2633  */
2634 static char *typename[] =
2635 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2636  "VMARKER"};
2637 
2638 void
2639 vn_printf(struct vnode *vp, const char *fmt, ...)
2640 {
2641 	va_list ap;
2642 	char buf[256], buf2[16];
2643 	u_long flags;
2644 
2645 	va_start(ap, fmt);
2646 	vprintf(fmt, ap);
2647 	va_end(ap);
2648 	printf("%p: ", (void *)vp);
2649 	printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2650 	printf("    usecount %d, writecount %d, refcount %d mountedhere %p\n",
2651 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2652 	buf[0] = '\0';
2653 	buf[1] = '\0';
2654 	if (vp->v_vflag & VV_ROOT)
2655 		strlcat(buf, "|VV_ROOT", sizeof(buf));
2656 	if (vp->v_vflag & VV_ISTTY)
2657 		strlcat(buf, "|VV_ISTTY", sizeof(buf));
2658 	if (vp->v_vflag & VV_NOSYNC)
2659 		strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2660 	if (vp->v_vflag & VV_CACHEDLABEL)
2661 		strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2662 	if (vp->v_vflag & VV_TEXT)
2663 		strlcat(buf, "|VV_TEXT", sizeof(buf));
2664 	if (vp->v_vflag & VV_COPYONWRITE)
2665 		strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2666 	if (vp->v_vflag & VV_SYSTEM)
2667 		strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2668 	if (vp->v_vflag & VV_PROCDEP)
2669 		strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2670 	if (vp->v_vflag & VV_NOKNOTE)
2671 		strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2672 	if (vp->v_vflag & VV_DELETED)
2673 		strlcat(buf, "|VV_DELETED", sizeof(buf));
2674 	if (vp->v_vflag & VV_MD)
2675 		strlcat(buf, "|VV_MD", sizeof(buf));
2676 	flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC |
2677 	    VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2678 	    VV_NOKNOTE | VV_DELETED | VV_MD);
2679 	if (flags != 0) {
2680 		snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2681 		strlcat(buf, buf2, sizeof(buf));
2682 	}
2683 	if (vp->v_iflag & VI_MOUNT)
2684 		strlcat(buf, "|VI_MOUNT", sizeof(buf));
2685 	if (vp->v_iflag & VI_AGE)
2686 		strlcat(buf, "|VI_AGE", sizeof(buf));
2687 	if (vp->v_iflag & VI_DOOMED)
2688 		strlcat(buf, "|VI_DOOMED", sizeof(buf));
2689 	if (vp->v_iflag & VI_FREE)
2690 		strlcat(buf, "|VI_FREE", sizeof(buf));
2691 	if (vp->v_iflag & VI_DOINGINACT)
2692 		strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2693 	if (vp->v_iflag & VI_OWEINACT)
2694 		strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2695 	flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2696 	    VI_DOINGINACT | VI_OWEINACT);
2697 	if (flags != 0) {
2698 		snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2699 		strlcat(buf, buf2, sizeof(buf));
2700 	}
2701 	printf("    flags (%s)\n", buf + 1);
2702 	if (mtx_owned(VI_MTX(vp)))
2703 		printf(" VI_LOCKed");
2704 	if (vp->v_object != NULL)
2705 		printf("    v_object %p ref %d pages %d\n",
2706 		    vp->v_object, vp->v_object->ref_count,
2707 		    vp->v_object->resident_page_count);
2708 	printf("    ");
2709 	lockmgr_printinfo(vp->v_vnlock);
2710 	if (vp->v_data != NULL)
2711 		VOP_PRINT(vp);
2712 }
2713 
2714 #ifdef DDB
2715 /*
2716  * List all of the locked vnodes in the system.
2717  * Called when debugging the kernel.
2718  */
2719 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2720 {
2721 	struct mount *mp, *nmp;
2722 	struct vnode *vp;
2723 
2724 	/*
2725 	 * Note: because this is DDB, we can't obey the locking semantics
2726 	 * for these structures, which means we could catch an inconsistent
2727 	 * state and dereference a nasty pointer.  Not much to be done
2728 	 * about that.
2729 	 */
2730 	db_printf("Locked vnodes\n");
2731 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2732 		nmp = TAILQ_NEXT(mp, mnt_list);
2733 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2734 			if (vp->v_type != VMARKER &&
2735 			    VOP_ISLOCKED(vp))
2736 				vprint("", vp);
2737 		}
2738 		nmp = TAILQ_NEXT(mp, mnt_list);
2739 	}
2740 }
2741 
2742 /*
2743  * Show details about the given vnode.
2744  */
2745 DB_SHOW_COMMAND(vnode, db_show_vnode)
2746 {
2747 	struct vnode *vp;
2748 
2749 	if (!have_addr)
2750 		return;
2751 	vp = (struct vnode *)addr;
2752 	vn_printf(vp, "vnode ");
2753 }
2754 
2755 /*
2756  * Show details about the given mount point.
2757  */
2758 DB_SHOW_COMMAND(mount, db_show_mount)
2759 {
2760 	struct mount *mp;
2761 	struct vfsopt *opt;
2762 	struct statfs *sp;
2763 	struct vnode *vp;
2764 	char buf[512];
2765 	u_int flags;
2766 
2767 	if (!have_addr) {
2768 		/* No address given, print short info about all mount points. */
2769 		TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2770 			db_printf("%p %s on %s (%s)\n", mp,
2771 			    mp->mnt_stat.f_mntfromname,
2772 			    mp->mnt_stat.f_mntonname,
2773 			    mp->mnt_stat.f_fstypename);
2774 			if (db_pager_quit)
2775 				break;
2776 		}
2777 		db_printf("\nMore info: show mount <addr>\n");
2778 		return;
2779 	}
2780 
2781 	mp = (struct mount *)addr;
2782 	db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
2783 	    mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
2784 
2785 	buf[0] = '\0';
2786 	flags = mp->mnt_flag;
2787 #define	MNT_FLAG(flag)	do {						\
2788 	if (flags & (flag)) {						\
2789 		if (buf[0] != '\0')					\
2790 			strlcat(buf, ", ", sizeof(buf));		\
2791 		strlcat(buf, (#flag) + 4, sizeof(buf));			\
2792 		flags &= ~(flag);					\
2793 	}								\
2794 } while (0)
2795 	MNT_FLAG(MNT_RDONLY);
2796 	MNT_FLAG(MNT_SYNCHRONOUS);
2797 	MNT_FLAG(MNT_NOEXEC);
2798 	MNT_FLAG(MNT_NOSUID);
2799 	MNT_FLAG(MNT_UNION);
2800 	MNT_FLAG(MNT_ASYNC);
2801 	MNT_FLAG(MNT_SUIDDIR);
2802 	MNT_FLAG(MNT_SOFTDEP);
2803 	MNT_FLAG(MNT_NOSYMFOLLOW);
2804 	MNT_FLAG(MNT_GJOURNAL);
2805 	MNT_FLAG(MNT_MULTILABEL);
2806 	MNT_FLAG(MNT_ACLS);
2807 	MNT_FLAG(MNT_NOATIME);
2808 	MNT_FLAG(MNT_NOCLUSTERR);
2809 	MNT_FLAG(MNT_NOCLUSTERW);
2810 	MNT_FLAG(MNT_NFS4ACLS);
2811 	MNT_FLAG(MNT_EXRDONLY);
2812 	MNT_FLAG(MNT_EXPORTED);
2813 	MNT_FLAG(MNT_DEFEXPORTED);
2814 	MNT_FLAG(MNT_EXPORTANON);
2815 	MNT_FLAG(MNT_EXKERB);
2816 	MNT_FLAG(MNT_EXPUBLIC);
2817 	MNT_FLAG(MNT_LOCAL);
2818 	MNT_FLAG(MNT_QUOTA);
2819 	MNT_FLAG(MNT_ROOTFS);
2820 	MNT_FLAG(MNT_USER);
2821 	MNT_FLAG(MNT_IGNORE);
2822 	MNT_FLAG(MNT_UPDATE);
2823 	MNT_FLAG(MNT_DELEXPORT);
2824 	MNT_FLAG(MNT_RELOAD);
2825 	MNT_FLAG(MNT_FORCE);
2826 	MNT_FLAG(MNT_SNAPSHOT);
2827 	MNT_FLAG(MNT_BYFSID);
2828 	MNT_FLAG(MNT_SOFTDEP);
2829 #undef MNT_FLAG
2830 	if (flags != 0) {
2831 		if (buf[0] != '\0')
2832 			strlcat(buf, ", ", sizeof(buf));
2833 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2834 		    "0x%08x", flags);
2835 	}
2836 	db_printf("    mnt_flag = %s\n", buf);
2837 
2838 	buf[0] = '\0';
2839 	flags = mp->mnt_kern_flag;
2840 #define	MNT_KERN_FLAG(flag)	do {					\
2841 	if (flags & (flag)) {						\
2842 		if (buf[0] != '\0')					\
2843 			strlcat(buf, ", ", sizeof(buf));		\
2844 		strlcat(buf, (#flag) + 5, sizeof(buf));			\
2845 		flags &= ~(flag);					\
2846 	}								\
2847 } while (0)
2848 	MNT_KERN_FLAG(MNTK_UNMOUNTF);
2849 	MNT_KERN_FLAG(MNTK_ASYNC);
2850 	MNT_KERN_FLAG(MNTK_SOFTDEP);
2851 	MNT_KERN_FLAG(MNTK_NOINSMNTQ);
2852 	MNT_KERN_FLAG(MNTK_DRAINING);
2853 	MNT_KERN_FLAG(MNTK_REFEXPIRE);
2854 	MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
2855 	MNT_KERN_FLAG(MNTK_SHARED_WRITES);
2856 	MNT_KERN_FLAG(MNTK_UNMOUNT);
2857 	MNT_KERN_FLAG(MNTK_MWAIT);
2858 	MNT_KERN_FLAG(MNTK_SUSPEND);
2859 	MNT_KERN_FLAG(MNTK_SUSPEND2);
2860 	MNT_KERN_FLAG(MNTK_SUSPENDED);
2861 	MNT_KERN_FLAG(MNTK_MPSAFE);
2862 	MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
2863 	MNT_KERN_FLAG(MNTK_NOKNOTE);
2864 #undef MNT_KERN_FLAG
2865 	if (flags != 0) {
2866 		if (buf[0] != '\0')
2867 			strlcat(buf, ", ", sizeof(buf));
2868 		snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2869 		    "0x%08x", flags);
2870 	}
2871 	db_printf("    mnt_kern_flag = %s\n", buf);
2872 
2873 	db_printf("    mnt_opt = ");
2874 	opt = TAILQ_FIRST(mp->mnt_opt);
2875 	if (opt != NULL) {
2876 		db_printf("%s", opt->name);
2877 		opt = TAILQ_NEXT(opt, link);
2878 		while (opt != NULL) {
2879 			db_printf(", %s", opt->name);
2880 			opt = TAILQ_NEXT(opt, link);
2881 		}
2882 	}
2883 	db_printf("\n");
2884 
2885 	sp = &mp->mnt_stat;
2886 	db_printf("    mnt_stat = { version=%u type=%u flags=0x%016jx "
2887 	    "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
2888 	    "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
2889 	    "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
2890 	    (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
2891 	    (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
2892 	    (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
2893 	    (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
2894 	    (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
2895 	    (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
2896 	    (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
2897 	    (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
2898 
2899 	db_printf("    mnt_cred = { uid=%u ruid=%u",
2900 	    (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
2901 	if (jailed(mp->mnt_cred))
2902 		db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
2903 	db_printf(" }\n");
2904 	db_printf("    mnt_ref = %d\n", mp->mnt_ref);
2905 	db_printf("    mnt_gen = %d\n", mp->mnt_gen);
2906 	db_printf("    mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
2907 	db_printf("    mnt_writeopcount = %d\n", mp->mnt_writeopcount);
2908 	db_printf("    mnt_noasync = %u\n", mp->mnt_noasync);
2909 	db_printf("    mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
2910 	db_printf("    mnt_iosize_max = %d\n", mp->mnt_iosize_max);
2911 	db_printf("    mnt_hashseed = %u\n", mp->mnt_hashseed);
2912 	db_printf("    mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
2913 	db_printf("    mnt_secondary_accwrites = %d\n",
2914 	    mp->mnt_secondary_accwrites);
2915 	db_printf("    mnt_gjprovider = %s\n",
2916 	    mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
2917 	db_printf("\n");
2918 
2919 	TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2920 		if (vp->v_type != VMARKER) {
2921 			vn_printf(vp, "vnode ");
2922 			if (db_pager_quit)
2923 				break;
2924 		}
2925 	}
2926 }
2927 #endif	/* DDB */
2928 
2929 /*
2930  * Fill in a struct xvfsconf based on a struct vfsconf.
2931  */
2932 static void
2933 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2934 {
2935 
2936 	strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2937 	xvfsp->vfc_typenum = vfsp->vfc_typenum;
2938 	xvfsp->vfc_refcount = vfsp->vfc_refcount;
2939 	xvfsp->vfc_flags = vfsp->vfc_flags;
2940 	/*
2941 	 * These are unused in userland, we keep them
2942 	 * to not break binary compatibility.
2943 	 */
2944 	xvfsp->vfc_vfsops = NULL;
2945 	xvfsp->vfc_next = NULL;
2946 }
2947 
2948 /*
2949  * Top level filesystem related information gathering.
2950  */
2951 static int
2952 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2953 {
2954 	struct vfsconf *vfsp;
2955 	struct xvfsconf xvfsp;
2956 	int error;
2957 
2958 	error = 0;
2959 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2960 		bzero(&xvfsp, sizeof(xvfsp));
2961 		vfsconf2x(vfsp, &xvfsp);
2962 		error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
2963 		if (error)
2964 			break;
2965 	}
2966 	return (error);
2967 }
2968 
2969 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
2970     "S,xvfsconf", "List of all configured filesystems");
2971 
2972 #ifndef BURN_BRIDGES
2973 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2974 
2975 static int
2976 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2977 {
2978 	int *name = (int *)arg1 - 1;	/* XXX */
2979 	u_int namelen = arg2 + 1;	/* XXX */
2980 	struct vfsconf *vfsp;
2981 	struct xvfsconf xvfsp;
2982 
2983 	printf("WARNING: userland calling deprecated sysctl, "
2984 	    "please rebuild world\n");
2985 
2986 #if 1 || defined(COMPAT_PRELITE2)
2987 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2988 	if (namelen == 1)
2989 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2990 #endif
2991 
2992 	switch (name[1]) {
2993 	case VFS_MAXTYPENUM:
2994 		if (namelen != 2)
2995 			return (ENOTDIR);
2996 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2997 	case VFS_CONF:
2998 		if (namelen != 3)
2999 			return (ENOTDIR);	/* overloaded */
3000 		TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
3001 			if (vfsp->vfc_typenum == name[2])
3002 				break;
3003 		if (vfsp == NULL)
3004 			return (EOPNOTSUPP);
3005 		bzero(&xvfsp, sizeof(xvfsp));
3006 		vfsconf2x(vfsp, &xvfsp);
3007 		return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3008 	}
3009 	return (EOPNOTSUPP);
3010 }
3011 
3012 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
3013 	vfs_sysctl, "Generic filesystem");
3014 
3015 #if 1 || defined(COMPAT_PRELITE2)
3016 
3017 static int
3018 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3019 {
3020 	int error;
3021 	struct vfsconf *vfsp;
3022 	struct ovfsconf ovfs;
3023 
3024 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3025 		bzero(&ovfs, sizeof(ovfs));
3026 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
3027 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
3028 		ovfs.vfc_index = vfsp->vfc_typenum;
3029 		ovfs.vfc_refcount = vfsp->vfc_refcount;
3030 		ovfs.vfc_flags = vfsp->vfc_flags;
3031 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3032 		if (error)
3033 			return error;
3034 	}
3035 	return 0;
3036 }
3037 
3038 #endif /* 1 || COMPAT_PRELITE2 */
3039 #endif /* !BURN_BRIDGES */
3040 
3041 #define KINFO_VNODESLOP		10
3042 #ifdef notyet
3043 /*
3044  * Dump vnode list (via sysctl).
3045  */
3046 /* ARGSUSED */
3047 static int
3048 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3049 {
3050 	struct xvnode *xvn;
3051 	struct mount *mp;
3052 	struct vnode *vp;
3053 	int error, len, n;
3054 
3055 	/*
3056 	 * Stale numvnodes access is not fatal here.
3057 	 */
3058 	req->lock = 0;
3059 	len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3060 	if (!req->oldptr)
3061 		/* Make an estimate */
3062 		return (SYSCTL_OUT(req, 0, len));
3063 
3064 	error = sysctl_wire_old_buffer(req, 0);
3065 	if (error != 0)
3066 		return (error);
3067 	xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3068 	n = 0;
3069 	mtx_lock(&mountlist_mtx);
3070 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3071 		if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3072 			continue;
3073 		MNT_ILOCK(mp);
3074 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3075 			if (n == len)
3076 				break;
3077 			vref(vp);
3078 			xvn[n].xv_size = sizeof *xvn;
3079 			xvn[n].xv_vnode = vp;
3080 			xvn[n].xv_id = 0;	/* XXX compat */
3081 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3082 			XV_COPY(usecount);
3083 			XV_COPY(writecount);
3084 			XV_COPY(holdcnt);
3085 			XV_COPY(mount);
3086 			XV_COPY(numoutput);
3087 			XV_COPY(type);
3088 #undef XV_COPY
3089 			xvn[n].xv_flag = vp->v_vflag;
3090 
3091 			switch (vp->v_type) {
3092 			case VREG:
3093 			case VDIR:
3094 			case VLNK:
3095 				break;
3096 			case VBLK:
3097 			case VCHR:
3098 				if (vp->v_rdev == NULL) {
3099 					vrele(vp);
3100 					continue;
3101 				}
3102 				xvn[n].xv_dev = dev2udev(vp->v_rdev);
3103 				break;
3104 			case VSOCK:
3105 				xvn[n].xv_socket = vp->v_socket;
3106 				break;
3107 			case VFIFO:
3108 				xvn[n].xv_fifo = vp->v_fifoinfo;
3109 				break;
3110 			case VNON:
3111 			case VBAD:
3112 			default:
3113 				/* shouldn't happen? */
3114 				vrele(vp);
3115 				continue;
3116 			}
3117 			vrele(vp);
3118 			++n;
3119 		}
3120 		MNT_IUNLOCK(mp);
3121 		mtx_lock(&mountlist_mtx);
3122 		vfs_unbusy(mp);
3123 		if (n == len)
3124 			break;
3125 	}
3126 	mtx_unlock(&mountlist_mtx);
3127 
3128 	error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3129 	free(xvn, M_TEMP);
3130 	return (error);
3131 }
3132 
3133 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3134 	0, 0, sysctl_vnode, "S,xvnode", "");
3135 #endif
3136 
3137 /*
3138  * Unmount all filesystems. The list is traversed in reverse order
3139  * of mounting to avoid dependencies.
3140  */
3141 void
3142 vfs_unmountall(void)
3143 {
3144 	struct mount *mp;
3145 	struct thread *td;
3146 	int error;
3147 
3148 	KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
3149 	CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3150 	td = curthread;
3151 
3152 	/*
3153 	 * Since this only runs when rebooting, it is not interlocked.
3154 	 */
3155 	while(!TAILQ_EMPTY(&mountlist)) {
3156 		mp = TAILQ_LAST(&mountlist, mntlist);
3157 		error = dounmount(mp, MNT_FORCE, td);
3158 		if (error) {
3159 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
3160 			/*
3161 			 * XXX: Due to the way in which we mount the root
3162 			 * file system off of devfs, devfs will generate a
3163 			 * "busy" warning when we try to unmount it before
3164 			 * the root.  Don't print a warning as a result in
3165 			 * order to avoid false positive errors that may
3166 			 * cause needless upset.
3167 			 */
3168 			if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3169 				printf("unmount of %s failed (",
3170 				    mp->mnt_stat.f_mntonname);
3171 				if (error == EBUSY)
3172 					printf("BUSY)\n");
3173 				else
3174 					printf("%d)\n", error);
3175 			}
3176 		} else {
3177 			/* The unmount has removed mp from the mountlist */
3178 		}
3179 	}
3180 }
3181 
3182 /*
3183  * perform msync on all vnodes under a mount point
3184  * the mount point must be locked.
3185  */
3186 void
3187 vfs_msync(struct mount *mp, int flags)
3188 {
3189 	struct vnode *vp, *mvp;
3190 	struct vm_object *obj;
3191 
3192 	CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3193 	MNT_ILOCK(mp);
3194 	MNT_VNODE_FOREACH(vp, mp, mvp) {
3195 		VI_LOCK(vp);
3196 		obj = vp->v_object;
3197 		if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3198 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3199 			MNT_IUNLOCK(mp);
3200 			if (!vget(vp,
3201 			    LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3202 			    curthread)) {
3203 				if (vp->v_vflag & VV_NOSYNC) {	/* unlinked */
3204 					vput(vp);
3205 					MNT_ILOCK(mp);
3206 					continue;
3207 				}
3208 
3209 				obj = vp->v_object;
3210 				if (obj != NULL) {
3211 					VM_OBJECT_LOCK(obj);
3212 					vm_object_page_clean(obj, 0, 0,
3213 					    flags == MNT_WAIT ?
3214 					    OBJPC_SYNC : OBJPC_NOSYNC);
3215 					VM_OBJECT_UNLOCK(obj);
3216 				}
3217 				vput(vp);
3218 			}
3219 			MNT_ILOCK(mp);
3220 		} else
3221 			VI_UNLOCK(vp);
3222 	}
3223 	MNT_IUNLOCK(mp);
3224 }
3225 
3226 /*
3227  * Mark a vnode as free, putting it up for recycling.
3228  */
3229 static void
3230 vfree(struct vnode *vp)
3231 {
3232 
3233 	ASSERT_VI_LOCKED(vp, "vfree");
3234 	mtx_lock(&vnode_free_list_mtx);
3235 	VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed."));
3236 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free"));
3237 	VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't"));
3238 	VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp,
3239 	    ("vfree: Freeing doomed vnode"));
3240 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3241 	if (vp->v_iflag & VI_AGE) {
3242 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3243 	} else {
3244 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3245 	}
3246 	freevnodes++;
3247 	vp->v_iflag &= ~VI_AGE;
3248 	vp->v_iflag |= VI_FREE;
3249 	mtx_unlock(&vnode_free_list_mtx);
3250 }
3251 
3252 /*
3253  * Opposite of vfree() - mark a vnode as in use.
3254  */
3255 static void
3256 vbusy(struct vnode *vp)
3257 {
3258 	ASSERT_VI_LOCKED(vp, "vbusy");
3259 	VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
3260 	VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed."));
3261 	CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3262 
3263 	mtx_lock(&vnode_free_list_mtx);
3264 	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3265 	freevnodes--;
3266 	vp->v_iflag &= ~(VI_FREE|VI_AGE);
3267 	mtx_unlock(&vnode_free_list_mtx);
3268 }
3269 
3270 static void
3271 destroy_vpollinfo(struct vpollinfo *vi)
3272 {
3273 	knlist_destroy(&vi->vpi_selinfo.si_note);
3274 	mtx_destroy(&vi->vpi_lock);
3275 	uma_zfree(vnodepoll_zone, vi);
3276 }
3277 
3278 /*
3279  * Initalize per-vnode helper structure to hold poll-related state.
3280  */
3281 void
3282 v_addpollinfo(struct vnode *vp)
3283 {
3284 	struct vpollinfo *vi;
3285 
3286 	if (vp->v_pollinfo != NULL)
3287 		return;
3288 	vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3289 	mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3290 	knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3291 	    vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3292 	VI_LOCK(vp);
3293 	if (vp->v_pollinfo != NULL) {
3294 		VI_UNLOCK(vp);
3295 		destroy_vpollinfo(vi);
3296 		return;
3297 	}
3298 	vp->v_pollinfo = vi;
3299 	VI_UNLOCK(vp);
3300 }
3301 
3302 /*
3303  * Record a process's interest in events which might happen to
3304  * a vnode.  Because poll uses the historic select-style interface
3305  * internally, this routine serves as both the ``check for any
3306  * pending events'' and the ``record my interest in future events''
3307  * functions.  (These are done together, while the lock is held,
3308  * to avoid race conditions.)
3309  */
3310 int
3311 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3312 {
3313 
3314 	v_addpollinfo(vp);
3315 	mtx_lock(&vp->v_pollinfo->vpi_lock);
3316 	if (vp->v_pollinfo->vpi_revents & events) {
3317 		/*
3318 		 * This leaves events we are not interested
3319 		 * in available for the other process which
3320 		 * which presumably had requested them
3321 		 * (otherwise they would never have been
3322 		 * recorded).
3323 		 */
3324 		events &= vp->v_pollinfo->vpi_revents;
3325 		vp->v_pollinfo->vpi_revents &= ~events;
3326 
3327 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
3328 		return (events);
3329 	}
3330 	vp->v_pollinfo->vpi_events |= events;
3331 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3332 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
3333 	return (0);
3334 }
3335 
3336 /*
3337  * Routine to create and manage a filesystem syncer vnode.
3338  */
3339 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
3340 static int	sync_fsync(struct  vop_fsync_args *);
3341 static int	sync_inactive(struct  vop_inactive_args *);
3342 static int	sync_reclaim(struct  vop_reclaim_args *);
3343 
3344 static struct vop_vector sync_vnodeops = {
3345 	.vop_bypass =	VOP_EOPNOTSUPP,
3346 	.vop_close =	sync_close,		/* close */
3347 	.vop_fsync =	sync_fsync,		/* fsync */
3348 	.vop_inactive =	sync_inactive,	/* inactive */
3349 	.vop_reclaim =	sync_reclaim,	/* reclaim */
3350 	.vop_lock1 =	vop_stdlock,	/* lock */
3351 	.vop_unlock =	vop_stdunlock,	/* unlock */
3352 	.vop_islocked =	vop_stdislocked,	/* islocked */
3353 };
3354 
3355 /*
3356  * Create a new filesystem syncer vnode for the specified mount point.
3357  */
3358 int
3359 vfs_allocate_syncvnode(struct mount *mp)
3360 {
3361 	struct vnode *vp;
3362 	struct bufobj *bo;
3363 	static long start, incr, next;
3364 	int error;
3365 
3366 	/* Allocate a new vnode */
3367 	if ((error = getnewvnode("syncer", mp, &sync_vnodeops, &vp)) != 0) {
3368 		mp->mnt_syncer = NULL;
3369 		return (error);
3370 	}
3371 	vp->v_type = VNON;
3372 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3373 	vp->v_vflag |= VV_FORCEINSMQ;
3374 	error = insmntque(vp, mp);
3375 	if (error != 0)
3376 		panic("vfs_allocate_syncvnode: insmntque failed");
3377 	vp->v_vflag &= ~VV_FORCEINSMQ;
3378 	VOP_UNLOCK(vp, 0);
3379 	/*
3380 	 * Place the vnode onto the syncer worklist. We attempt to
3381 	 * scatter them about on the list so that they will go off
3382 	 * at evenly distributed times even if all the filesystems
3383 	 * are mounted at once.
3384 	 */
3385 	next += incr;
3386 	if (next == 0 || next > syncer_maxdelay) {
3387 		start /= 2;
3388 		incr /= 2;
3389 		if (start == 0) {
3390 			start = syncer_maxdelay / 2;
3391 			incr = syncer_maxdelay;
3392 		}
3393 		next = start;
3394 	}
3395 	bo = &vp->v_bufobj;
3396 	BO_LOCK(bo);
3397 	vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3398 	/* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3399 	mtx_lock(&sync_mtx);
3400 	sync_vnode_count++;
3401 	mtx_unlock(&sync_mtx);
3402 	BO_UNLOCK(bo);
3403 	mp->mnt_syncer = vp;
3404 	return (0);
3405 }
3406 
3407 /*
3408  * Do a lazy sync of the filesystem.
3409  */
3410 static int
3411 sync_fsync(struct vop_fsync_args *ap)
3412 {
3413 	struct vnode *syncvp = ap->a_vp;
3414 	struct mount *mp = syncvp->v_mount;
3415 	int error;
3416 	struct bufobj *bo;
3417 
3418 	/*
3419 	 * We only need to do something if this is a lazy evaluation.
3420 	 */
3421 	if (ap->a_waitfor != MNT_LAZY)
3422 		return (0);
3423 
3424 	/*
3425 	 * Move ourselves to the back of the sync list.
3426 	 */
3427 	bo = &syncvp->v_bufobj;
3428 	BO_LOCK(bo);
3429 	vn_syncer_add_to_worklist(bo, syncdelay);
3430 	BO_UNLOCK(bo);
3431 
3432 	/*
3433 	 * Walk the list of vnodes pushing all that are dirty and
3434 	 * not already on the sync list.
3435 	 */
3436 	mtx_lock(&mountlist_mtx);
3437 	if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) {
3438 		mtx_unlock(&mountlist_mtx);
3439 		return (0);
3440 	}
3441 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3442 		vfs_unbusy(mp);
3443 		return (0);
3444 	}
3445 	MNT_ILOCK(mp);
3446 	mp->mnt_noasync++;
3447 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
3448 	MNT_IUNLOCK(mp);
3449 	vfs_msync(mp, MNT_NOWAIT);
3450 	error = VFS_SYNC(mp, MNT_LAZY);
3451 	MNT_ILOCK(mp);
3452 	mp->mnt_noasync--;
3453 	if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
3454 		mp->mnt_kern_flag |= MNTK_ASYNC;
3455 	MNT_IUNLOCK(mp);
3456 	vn_finished_write(mp);
3457 	vfs_unbusy(mp);
3458 	return (error);
3459 }
3460 
3461 /*
3462  * The syncer vnode is no referenced.
3463  */
3464 static int
3465 sync_inactive(struct vop_inactive_args *ap)
3466 {
3467 
3468 	vgone(ap->a_vp);
3469 	return (0);
3470 }
3471 
3472 /*
3473  * The syncer vnode is no longer needed and is being decommissioned.
3474  *
3475  * Modifications to the worklist must be protected by sync_mtx.
3476  */
3477 static int
3478 sync_reclaim(struct vop_reclaim_args *ap)
3479 {
3480 	struct vnode *vp = ap->a_vp;
3481 	struct bufobj *bo;
3482 
3483 	bo = &vp->v_bufobj;
3484 	BO_LOCK(bo);
3485 	vp->v_mount->mnt_syncer = NULL;
3486 	if (bo->bo_flag & BO_ONWORKLST) {
3487 		mtx_lock(&sync_mtx);
3488 		LIST_REMOVE(bo, bo_synclist);
3489 		syncer_worklist_len--;
3490 		sync_vnode_count--;
3491 		mtx_unlock(&sync_mtx);
3492 		bo->bo_flag &= ~BO_ONWORKLST;
3493 	}
3494 	BO_UNLOCK(bo);
3495 
3496 	return (0);
3497 }
3498 
3499 /*
3500  * Check if vnode represents a disk device
3501  */
3502 int
3503 vn_isdisk(struct vnode *vp, int *errp)
3504 {
3505 	int error;
3506 
3507 	error = 0;
3508 	dev_lock();
3509 	if (vp->v_type != VCHR)
3510 		error = ENOTBLK;
3511 	else if (vp->v_rdev == NULL)
3512 		error = ENXIO;
3513 	else if (vp->v_rdev->si_devsw == NULL)
3514 		error = ENXIO;
3515 	else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3516 		error = ENOTBLK;
3517 	dev_unlock();
3518 	if (errp != NULL)
3519 		*errp = error;
3520 	return (error == 0);
3521 }
3522 
3523 /*
3524  * Common filesystem object access control check routine.  Accepts a
3525  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3526  * and optional call-by-reference privused argument allowing vaccess()
3527  * to indicate to the caller whether privilege was used to satisfy the
3528  * request (obsoleted).  Returns 0 on success, or an errno on failure.
3529  *
3530  * The ifdef'd CAPABILITIES version is here for reference, but is not
3531  * actually used.
3532  */
3533 int
3534 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3535     accmode_t accmode, struct ucred *cred, int *privused)
3536 {
3537 	accmode_t dac_granted;
3538 	accmode_t priv_granted;
3539 
3540 	KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3541 	    ("invalid bit in accmode"));
3542 	KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
3543 	    ("VAPPEND without VWRITE"));
3544 
3545 	/*
3546 	 * Look for a normal, non-privileged way to access the file/directory
3547 	 * as requested.  If it exists, go with that.
3548 	 */
3549 
3550 	if (privused != NULL)
3551 		*privused = 0;
3552 
3553 	dac_granted = 0;
3554 
3555 	/* Check the owner. */
3556 	if (cred->cr_uid == file_uid) {
3557 		dac_granted |= VADMIN;
3558 		if (file_mode & S_IXUSR)
3559 			dac_granted |= VEXEC;
3560 		if (file_mode & S_IRUSR)
3561 			dac_granted |= VREAD;
3562 		if (file_mode & S_IWUSR)
3563 			dac_granted |= (VWRITE | VAPPEND);
3564 
3565 		if ((accmode & dac_granted) == accmode)
3566 			return (0);
3567 
3568 		goto privcheck;
3569 	}
3570 
3571 	/* Otherwise, check the groups (first match) */
3572 	if (groupmember(file_gid, cred)) {
3573 		if (file_mode & S_IXGRP)
3574 			dac_granted |= VEXEC;
3575 		if (file_mode & S_IRGRP)
3576 			dac_granted |= VREAD;
3577 		if (file_mode & S_IWGRP)
3578 			dac_granted |= (VWRITE | VAPPEND);
3579 
3580 		if ((accmode & dac_granted) == accmode)
3581 			return (0);
3582 
3583 		goto privcheck;
3584 	}
3585 
3586 	/* Otherwise, check everyone else. */
3587 	if (file_mode & S_IXOTH)
3588 		dac_granted |= VEXEC;
3589 	if (file_mode & S_IROTH)
3590 		dac_granted |= VREAD;
3591 	if (file_mode & S_IWOTH)
3592 		dac_granted |= (VWRITE | VAPPEND);
3593 	if ((accmode & dac_granted) == accmode)
3594 		return (0);
3595 
3596 privcheck:
3597 	/*
3598 	 * Build a privilege mask to determine if the set of privileges
3599 	 * satisfies the requirements when combined with the granted mask
3600 	 * from above.  For each privilege, if the privilege is required,
3601 	 * bitwise or the request type onto the priv_granted mask.
3602 	 */
3603 	priv_granted = 0;
3604 
3605 	if (type == VDIR) {
3606 		/*
3607 		 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3608 		 * requests, instead of PRIV_VFS_EXEC.
3609 		 */
3610 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3611 		    !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3612 			priv_granted |= VEXEC;
3613 	} else {
3614 		if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3615 		    !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3616 			priv_granted |= VEXEC;
3617 	}
3618 
3619 	if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
3620 	    !priv_check_cred(cred, PRIV_VFS_READ, 0))
3621 		priv_granted |= VREAD;
3622 
3623 	if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3624 	    !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3625 		priv_granted |= (VWRITE | VAPPEND);
3626 
3627 	if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3628 	    !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3629 		priv_granted |= VADMIN;
3630 
3631 	if ((accmode & (priv_granted | dac_granted)) == accmode) {
3632 		/* XXX audit: privilege used */
3633 		if (privused != NULL)
3634 			*privused = 1;
3635 		return (0);
3636 	}
3637 
3638 	return ((accmode & VADMIN) ? EPERM : EACCES);
3639 }
3640 
3641 /*
3642  * Credential check based on process requesting service, and per-attribute
3643  * permissions.
3644  */
3645 int
3646 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3647     struct thread *td, accmode_t accmode)
3648 {
3649 
3650 	/*
3651 	 * Kernel-invoked always succeeds.
3652 	 */
3653 	if (cred == NOCRED)
3654 		return (0);
3655 
3656 	/*
3657 	 * Do not allow privileged processes in jail to directly manipulate
3658 	 * system attributes.
3659 	 */
3660 	switch (attrnamespace) {
3661 	case EXTATTR_NAMESPACE_SYSTEM:
3662 		/* Potentially should be: return (EPERM); */
3663 		return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3664 	case EXTATTR_NAMESPACE_USER:
3665 		return (VOP_ACCESS(vp, accmode, cred, td));
3666 	default:
3667 		return (EPERM);
3668 	}
3669 }
3670 
3671 #ifdef DEBUG_VFS_LOCKS
3672 /*
3673  * This only exists to supress warnings from unlocked specfs accesses.  It is
3674  * no longer ok to have an unlocked VFS.
3675  */
3676 #define	IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL ||		\
3677 	(vp)->v_type == VCHR ||	(vp)->v_type == VBAD)
3678 
3679 int vfs_badlock_ddb = 1;	/* Drop into debugger on violation. */
3680 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, "");
3681 
3682 int vfs_badlock_mutex = 1;	/* Check for interlock across VOPs. */
3683 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, "");
3684 
3685 int vfs_badlock_print = 1;	/* Print lock violations. */
3686 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, "");
3687 
3688 #ifdef KDB
3689 int vfs_badlock_backtrace = 1;	/* Print backtrace at lock violations. */
3690 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, "");
3691 #endif
3692 
3693 static void
3694 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3695 {
3696 
3697 #ifdef KDB
3698 	if (vfs_badlock_backtrace)
3699 		kdb_backtrace();
3700 #endif
3701 	if (vfs_badlock_print)
3702 		printf("%s: %p %s\n", str, (void *)vp, msg);
3703 	if (vfs_badlock_ddb)
3704 		kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3705 }
3706 
3707 void
3708 assert_vi_locked(struct vnode *vp, const char *str)
3709 {
3710 
3711 	if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3712 		vfs_badlock("interlock is not locked but should be", str, vp);
3713 }
3714 
3715 void
3716 assert_vi_unlocked(struct vnode *vp, const char *str)
3717 {
3718 
3719 	if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3720 		vfs_badlock("interlock is locked but should not be", str, vp);
3721 }
3722 
3723 void
3724 assert_vop_locked(struct vnode *vp, const char *str)
3725 {
3726 
3727 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0)
3728 		vfs_badlock("is not locked but should be", str, vp);
3729 }
3730 
3731 void
3732 assert_vop_unlocked(struct vnode *vp, const char *str)
3733 {
3734 
3735 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
3736 		vfs_badlock("is locked but should not be", str, vp);
3737 }
3738 
3739 void
3740 assert_vop_elocked(struct vnode *vp, const char *str)
3741 {
3742 
3743 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
3744 		vfs_badlock("is not exclusive locked but should be", str, vp);
3745 }
3746 
3747 #if 0
3748 void
3749 assert_vop_elocked_other(struct vnode *vp, const char *str)
3750 {
3751 
3752 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
3753 		vfs_badlock("is not exclusive locked by another thread",
3754 		    str, vp);
3755 }
3756 
3757 void
3758 assert_vop_slocked(struct vnode *vp, const char *str)
3759 {
3760 
3761 	if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
3762 		vfs_badlock("is not locked shared but should be", str, vp);
3763 }
3764 #endif /* 0 */
3765 #endif /* DEBUG_VFS_LOCKS */
3766 
3767 void
3768 vop_rename_fail(struct vop_rename_args *ap)
3769 {
3770 
3771 	if (ap->a_tvp != NULL)
3772 		vput(ap->a_tvp);
3773 	if (ap->a_tdvp == ap->a_tvp)
3774 		vrele(ap->a_tdvp);
3775 	else
3776 		vput(ap->a_tdvp);
3777 	vrele(ap->a_fdvp);
3778 	vrele(ap->a_fvp);
3779 }
3780 
3781 void
3782 vop_rename_pre(void *ap)
3783 {
3784 	struct vop_rename_args *a = ap;
3785 
3786 #ifdef DEBUG_VFS_LOCKS
3787 	if (a->a_tvp)
3788 		ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
3789 	ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
3790 	ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
3791 	ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
3792 
3793 	/* Check the source (from). */
3794 	if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
3795 	    (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
3796 		ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3797 	if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
3798 		ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
3799 
3800 	/* Check the target. */
3801 	if (a->a_tvp)
3802 		ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3803 	ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3804 #endif
3805 	if (a->a_tdvp != a->a_fdvp)
3806 		vhold(a->a_fdvp);
3807 	if (a->a_tvp != a->a_fvp)
3808 		vhold(a->a_fvp);
3809 	vhold(a->a_tdvp);
3810 	if (a->a_tvp)
3811 		vhold(a->a_tvp);
3812 }
3813 
3814 void
3815 vop_strategy_pre(void *ap)
3816 {
3817 #ifdef DEBUG_VFS_LOCKS
3818 	struct vop_strategy_args *a;
3819 	struct buf *bp;
3820 
3821 	a = ap;
3822 	bp = a->a_bp;
3823 
3824 	/*
3825 	 * Cluster ops lock their component buffers but not the IO container.
3826 	 */
3827 	if ((bp->b_flags & B_CLUSTER) != 0)
3828 		return;
3829 
3830 	if (!BUF_ISLOCKED(bp)) {
3831 		if (vfs_badlock_print)
3832 			printf(
3833 			    "VOP_STRATEGY: bp is not locked but should be\n");
3834 		if (vfs_badlock_ddb)
3835 			kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3836 	}
3837 #endif
3838 }
3839 
3840 void
3841 vop_lookup_pre(void *ap)
3842 {
3843 #ifdef DEBUG_VFS_LOCKS
3844 	struct vop_lookup_args *a;
3845 	struct vnode *dvp;
3846 
3847 	a = ap;
3848 	dvp = a->a_dvp;
3849 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3850 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3851 #endif
3852 }
3853 
3854 void
3855 vop_lookup_post(void *ap, int rc)
3856 {
3857 #ifdef DEBUG_VFS_LOCKS
3858 	struct vop_lookup_args *a;
3859 	struct vnode *dvp;
3860 	struct vnode *vp;
3861 
3862 	a = ap;
3863 	dvp = a->a_dvp;
3864 	vp = *(a->a_vpp);
3865 
3866 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3867 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3868 
3869 	if (!rc)
3870 		ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
3871 #endif
3872 }
3873 
3874 void
3875 vop_lock_pre(void *ap)
3876 {
3877 #ifdef DEBUG_VFS_LOCKS
3878 	struct vop_lock1_args *a = ap;
3879 
3880 	if ((a->a_flags & LK_INTERLOCK) == 0)
3881 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3882 	else
3883 		ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
3884 #endif
3885 }
3886 
3887 void
3888 vop_lock_post(void *ap, int rc)
3889 {
3890 #ifdef DEBUG_VFS_LOCKS
3891 	struct vop_lock1_args *a = ap;
3892 
3893 	ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3894 	if (rc == 0)
3895 		ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
3896 #endif
3897 }
3898 
3899 void
3900 vop_unlock_pre(void *ap)
3901 {
3902 #ifdef DEBUG_VFS_LOCKS
3903 	struct vop_unlock_args *a = ap;
3904 
3905 	if (a->a_flags & LK_INTERLOCK)
3906 		ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
3907 	ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
3908 #endif
3909 }
3910 
3911 void
3912 vop_unlock_post(void *ap, int rc)
3913 {
3914 #ifdef DEBUG_VFS_LOCKS
3915 	struct vop_unlock_args *a = ap;
3916 
3917 	if (a->a_flags & LK_INTERLOCK)
3918 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
3919 #endif
3920 }
3921 
3922 void
3923 vop_create_post(void *ap, int rc)
3924 {
3925 	struct vop_create_args *a = ap;
3926 
3927 	if (!rc)
3928 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3929 }
3930 
3931 void
3932 vop_link_post(void *ap, int rc)
3933 {
3934 	struct vop_link_args *a = ap;
3935 
3936 	if (!rc) {
3937 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
3938 		VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
3939 	}
3940 }
3941 
3942 void
3943 vop_mkdir_post(void *ap, int rc)
3944 {
3945 	struct vop_mkdir_args *a = ap;
3946 
3947 	if (!rc)
3948 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3949 }
3950 
3951 void
3952 vop_mknod_post(void *ap, int rc)
3953 {
3954 	struct vop_mknod_args *a = ap;
3955 
3956 	if (!rc)
3957 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3958 }
3959 
3960 void
3961 vop_remove_post(void *ap, int rc)
3962 {
3963 	struct vop_remove_args *a = ap;
3964 
3965 	if (!rc) {
3966 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3967 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3968 	}
3969 }
3970 
3971 void
3972 vop_rename_post(void *ap, int rc)
3973 {
3974 	struct vop_rename_args *a = ap;
3975 
3976 	if (!rc) {
3977 		VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
3978 		VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
3979 		VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
3980 		if (a->a_tvp)
3981 			VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
3982 	}
3983 	if (a->a_tdvp != a->a_fdvp)
3984 		vdrop(a->a_fdvp);
3985 	if (a->a_tvp != a->a_fvp)
3986 		vdrop(a->a_fvp);
3987 	vdrop(a->a_tdvp);
3988 	if (a->a_tvp)
3989 		vdrop(a->a_tvp);
3990 }
3991 
3992 void
3993 vop_rmdir_post(void *ap, int rc)
3994 {
3995 	struct vop_rmdir_args *a = ap;
3996 
3997 	if (!rc) {
3998 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3999 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4000 	}
4001 }
4002 
4003 void
4004 vop_setattr_post(void *ap, int rc)
4005 {
4006 	struct vop_setattr_args *a = ap;
4007 
4008 	if (!rc)
4009 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4010 }
4011 
4012 void
4013 vop_symlink_post(void *ap, int rc)
4014 {
4015 	struct vop_symlink_args *a = ap;
4016 
4017 	if (!rc)
4018 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4019 }
4020 
4021 static struct knlist fs_knlist;
4022 
4023 static void
4024 vfs_event_init(void *arg)
4025 {
4026 	knlist_init_mtx(&fs_knlist, NULL);
4027 }
4028 /* XXX - correct order? */
4029 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4030 
4031 void
4032 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
4033 {
4034 
4035 	KNOTE_UNLOCKED(&fs_knlist, event);
4036 }
4037 
4038 static int	filt_fsattach(struct knote *kn);
4039 static void	filt_fsdetach(struct knote *kn);
4040 static int	filt_fsevent(struct knote *kn, long hint);
4041 
4042 struct filterops fs_filtops = {
4043 	.f_isfd = 0,
4044 	.f_attach = filt_fsattach,
4045 	.f_detach = filt_fsdetach,
4046 	.f_event = filt_fsevent
4047 };
4048 
4049 static int
4050 filt_fsattach(struct knote *kn)
4051 {
4052 
4053 	kn->kn_flags |= EV_CLEAR;
4054 	knlist_add(&fs_knlist, kn, 0);
4055 	return (0);
4056 }
4057 
4058 static void
4059 filt_fsdetach(struct knote *kn)
4060 {
4061 
4062 	knlist_remove(&fs_knlist, kn, 0);
4063 }
4064 
4065 static int
4066 filt_fsevent(struct knote *kn, long hint)
4067 {
4068 
4069 	kn->kn_fflags |= hint;
4070 	return (kn->kn_fflags != 0);
4071 }
4072 
4073 static int
4074 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4075 {
4076 	struct vfsidctl vc;
4077 	int error;
4078 	struct mount *mp;
4079 
4080 	error = SYSCTL_IN(req, &vc, sizeof(vc));
4081 	if (error)
4082 		return (error);
4083 	if (vc.vc_vers != VFS_CTL_VERS1)
4084 		return (EINVAL);
4085 	mp = vfs_getvfs(&vc.vc_fsid);
4086 	if (mp == NULL)
4087 		return (ENOENT);
4088 	/* ensure that a specific sysctl goes to the right filesystem. */
4089 	if (strcmp(vc.vc_fstypename, "*") != 0 &&
4090 	    strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4091 		vfs_rel(mp);
4092 		return (EINVAL);
4093 	}
4094 	VCTLTOREQ(&vc, req);
4095 	error = VFS_SYSCTL(mp, vc.vc_op, req);
4096 	vfs_rel(mp);
4097 	return (error);
4098 }
4099 
4100 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "",
4101     "Sysctl by fsid");
4102 
4103 /*
4104  * Function to initialize a va_filerev field sensibly.
4105  * XXX: Wouldn't a random number make a lot more sense ??
4106  */
4107 u_quad_t
4108 init_va_filerev(void)
4109 {
4110 	struct bintime bt;
4111 
4112 	getbinuptime(&bt);
4113 	return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4114 }
4115 
4116 static int	filt_vfsread(struct knote *kn, long hint);
4117 static int	filt_vfswrite(struct knote *kn, long hint);
4118 static int	filt_vfsvnode(struct knote *kn, long hint);
4119 static void	filt_vfsdetach(struct knote *kn);
4120 static struct filterops vfsread_filtops = {
4121 	.f_isfd = 1,
4122 	.f_detach = filt_vfsdetach,
4123 	.f_event = filt_vfsread
4124 };
4125 static struct filterops vfswrite_filtops = {
4126 	.f_isfd = 1,
4127 	.f_detach = filt_vfsdetach,
4128 	.f_event = filt_vfswrite
4129 };
4130 static struct filterops vfsvnode_filtops = {
4131 	.f_isfd = 1,
4132 	.f_detach = filt_vfsdetach,
4133 	.f_event = filt_vfsvnode
4134 };
4135 
4136 static void
4137 vfs_knllock(void *arg)
4138 {
4139 	struct vnode *vp = arg;
4140 
4141 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4142 }
4143 
4144 static void
4145 vfs_knlunlock(void *arg)
4146 {
4147 	struct vnode *vp = arg;
4148 
4149 	VOP_UNLOCK(vp, 0);
4150 }
4151 
4152 static void
4153 vfs_knl_assert_locked(void *arg)
4154 {
4155 #ifdef DEBUG_VFS_LOCKS
4156 	struct vnode *vp = arg;
4157 
4158 	ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4159 #endif
4160 }
4161 
4162 static void
4163 vfs_knl_assert_unlocked(void *arg)
4164 {
4165 #ifdef DEBUG_VFS_LOCKS
4166 	struct vnode *vp = arg;
4167 
4168 	ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4169 #endif
4170 }
4171 
4172 int
4173 vfs_kqfilter(struct vop_kqfilter_args *ap)
4174 {
4175 	struct vnode *vp = ap->a_vp;
4176 	struct knote *kn = ap->a_kn;
4177 	struct knlist *knl;
4178 
4179 	switch (kn->kn_filter) {
4180 	case EVFILT_READ:
4181 		kn->kn_fop = &vfsread_filtops;
4182 		break;
4183 	case EVFILT_WRITE:
4184 		kn->kn_fop = &vfswrite_filtops;
4185 		break;
4186 	case EVFILT_VNODE:
4187 		kn->kn_fop = &vfsvnode_filtops;
4188 		break;
4189 	default:
4190 		return (EINVAL);
4191 	}
4192 
4193 	kn->kn_hook = (caddr_t)vp;
4194 
4195 	v_addpollinfo(vp);
4196 	if (vp->v_pollinfo == NULL)
4197 		return (ENOMEM);
4198 	knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4199 	knlist_add(knl, kn, 0);
4200 
4201 	return (0);
4202 }
4203 
4204 /*
4205  * Detach knote from vnode
4206  */
4207 static void
4208 filt_vfsdetach(struct knote *kn)
4209 {
4210 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4211 
4212 	KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4213 	knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4214 }
4215 
4216 /*ARGSUSED*/
4217 static int
4218 filt_vfsread(struct knote *kn, long hint)
4219 {
4220 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4221 	struct vattr va;
4222 	int res;
4223 
4224 	/*
4225 	 * filesystem is gone, so set the EOF flag and schedule
4226 	 * the knote for deletion.
4227 	 */
4228 	if (hint == NOTE_REVOKE) {
4229 		VI_LOCK(vp);
4230 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4231 		VI_UNLOCK(vp);
4232 		return (1);
4233 	}
4234 
4235 	if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4236 		return (0);
4237 
4238 	VI_LOCK(vp);
4239 	kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4240 	res = (kn->kn_data != 0);
4241 	VI_UNLOCK(vp);
4242 	return (res);
4243 }
4244 
4245 /*ARGSUSED*/
4246 static int
4247 filt_vfswrite(struct knote *kn, long hint)
4248 {
4249 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4250 
4251 	VI_LOCK(vp);
4252 
4253 	/*
4254 	 * filesystem is gone, so set the EOF flag and schedule
4255 	 * the knote for deletion.
4256 	 */
4257 	if (hint == NOTE_REVOKE)
4258 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4259 
4260 	kn->kn_data = 0;
4261 	VI_UNLOCK(vp);
4262 	return (1);
4263 }
4264 
4265 static int
4266 filt_vfsvnode(struct knote *kn, long hint)
4267 {
4268 	struct vnode *vp = (struct vnode *)kn->kn_hook;
4269 	int res;
4270 
4271 	VI_LOCK(vp);
4272 	if (kn->kn_sfflags & hint)
4273 		kn->kn_fflags |= hint;
4274 	if (hint == NOTE_REVOKE) {
4275 		kn->kn_flags |= EV_EOF;
4276 		VI_UNLOCK(vp);
4277 		return (1);
4278 	}
4279 	res = (kn->kn_fflags != 0);
4280 	VI_UNLOCK(vp);
4281 	return (res);
4282 }
4283 
4284 int
4285 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4286 {
4287 	int error;
4288 
4289 	if (dp->d_reclen > ap->a_uio->uio_resid)
4290 		return (ENAMETOOLONG);
4291 	error = uiomove(dp, dp->d_reclen, ap->a_uio);
4292 	if (error) {
4293 		if (ap->a_ncookies != NULL) {
4294 			if (ap->a_cookies != NULL)
4295 				free(ap->a_cookies, M_TEMP);
4296 			ap->a_cookies = NULL;
4297 			*ap->a_ncookies = 0;
4298 		}
4299 		return (error);
4300 	}
4301 	if (ap->a_ncookies == NULL)
4302 		return (0);
4303 
4304 	KASSERT(ap->a_cookies,
4305 	    ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4306 
4307 	*ap->a_cookies = realloc(*ap->a_cookies,
4308 	    (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4309 	(*ap->a_cookies)[*ap->a_ncookies] = off;
4310 	return (0);
4311 }
4312 
4313 /*
4314  * Mark for update the access time of the file if the filesystem
4315  * supports VOP_MARKATIME.  This functionality is used by execve and
4316  * mmap, so we want to avoid the I/O implied by directly setting
4317  * va_atime for the sake of efficiency.
4318  */
4319 void
4320 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4321 {
4322 	struct mount *mp;
4323 
4324 	mp = vp->v_mount;
4325 	VFS_ASSERT_GIANT(mp);
4326 	ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4327 	if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4328 		(void)VOP_MARKATIME(vp);
4329 }
4330 
4331 /*
4332  * The purpose of this routine is to remove granularity from accmode_t,
4333  * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4334  * VADMIN and VAPPEND.
4335  *
4336  * If it returns 0, the caller is supposed to continue with the usual
4337  * access checks using 'accmode' as modified by this routine.  If it
4338  * returns nonzero value, the caller is supposed to return that value
4339  * as errno.
4340  *
4341  * Note that after this routine runs, accmode may be zero.
4342  */
4343 int
4344 vfs_unixify_accmode(accmode_t *accmode)
4345 {
4346 	/*
4347 	 * There is no way to specify explicit "deny" rule using
4348 	 * file mode or POSIX.1e ACLs.
4349 	 */
4350 	if (*accmode & VEXPLICIT_DENY) {
4351 		*accmode = 0;
4352 		return (0);
4353 	}
4354 
4355 	/*
4356 	 * None of these can be translated into usual access bits.
4357 	 * Also, the common case for NFSv4 ACLs is to not contain
4358 	 * either of these bits. Caller should check for VWRITE
4359 	 * on the containing directory instead.
4360 	 */
4361 	if (*accmode & (VDELETE_CHILD | VDELETE))
4362 		return (EPERM);
4363 
4364 	if (*accmode & VADMIN_PERMS) {
4365 		*accmode &= ~VADMIN_PERMS;
4366 		*accmode |= VADMIN;
4367 	}
4368 
4369 	/*
4370 	 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4371 	 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4372 	 */
4373 	*accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
4374 
4375 	return (0);
4376 }
4377