xref: /freebsd/sys/kern/vfs_subr.c (revision 91c878a6935c5c2e99866eb267e5bc3028bf6d2f)
1 /*-
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)vfs_subr.c	8.31 (Berkeley) 5/26/95
35  */
36 
37 /*
38  * External virtual filesystem routines
39  */
40 
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
43 
44 #include "opt_ddb.h"
45 #include "opt_mac.h"
46 
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/bio.h>
50 #include <sys/buf.h>
51 #include <sys/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/kdb.h>
59 #include <sys/kernel.h>
60 #include <sys/kthread.h>
61 #include <sys/malloc.h>
62 #include <sys/mount.h>
63 #include <sys/namei.h>
64 #include <sys/reboot.h>
65 #include <sys/sleepqueue.h>
66 #include <sys/stat.h>
67 #include <sys/sysctl.h>
68 #include <sys/syslog.h>
69 #include <sys/vmmeter.h>
70 #include <sys/vnode.h>
71 
72 #include <machine/stdarg.h>
73 
74 #include <security/mac/mac_framework.h>
75 
76 #include <vm/vm.h>
77 #include <vm/vm_object.h>
78 #include <vm/vm_extern.h>
79 #include <vm/pmap.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_page.h>
82 #include <vm/vm_kern.h>
83 #include <vm/uma.h>
84 
85 #ifdef DDB
86 #include <ddb/ddb.h>
87 #endif
88 
89 static MALLOC_DEFINE(M_NETADDR, "subr_export_host", "Export host address structure");
90 
91 static void	delmntque(struct vnode *vp);
92 static void	insmntque(struct vnode *vp, struct mount *mp);
93 static int	flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
94 		    int slpflag, int slptimeo);
95 static void	syncer_shutdown(void *arg, int howto);
96 static int	vtryrecycle(struct vnode *vp);
97 static void	vbusy(struct vnode *vp);
98 static void	vdropl(struct vnode *vp);
99 static void	vinactive(struct vnode *, struct thread *);
100 static void	v_incr_usecount(struct vnode *);
101 static void	v_decr_usecount(struct vnode *);
102 static void	v_decr_useonly(struct vnode *);
103 static void	v_upgrade_usecount(struct vnode *);
104 static void	vfree(struct vnode *);
105 static void	vnlru_free(int);
106 static void	vdestroy(struct vnode *);
107 static void	vgonel(struct vnode *);
108 static void	vfs_knllock(void *arg);
109 static void	vfs_knlunlock(void *arg);
110 static int	vfs_knllocked(void *arg);
111 
112 
113 /*
114  * Enable Giant pushdown based on whether or not the vm is mpsafe in this
115  * build.  Without mpsafevm the buffer cache can not run Giant free.
116  */
117 #if !defined(__powerpc__)
118 int mpsafe_vfs = 1;
119 #else
120 int mpsafe_vfs;
121 #endif
122 TUNABLE_INT("debug.mpsafevfs", &mpsafe_vfs);
123 SYSCTL_INT(_debug, OID_AUTO, mpsafevfs, CTLFLAG_RD, &mpsafe_vfs, 0,
124     "MPSAFE VFS");
125 
126 /*
127  * Number of vnodes in existence.  Increased whenever getnewvnode()
128  * allocates a new vnode, decreased on vdestroy() called on VI_DOOMed
129  * vnode.
130  */
131 static unsigned long	numvnodes;
132 
133 SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
134 
135 /*
136  * Conversion tables for conversion from vnode types to inode formats
137  * and back.
138  */
139 enum vtype iftovt_tab[16] = {
140 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
141 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
142 };
143 int vttoif_tab[10] = {
144 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
145 	S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
146 };
147 
148 /*
149  * List of vnodes that are ready for recycling.
150  */
151 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
152 
153 /*
154  * Free vnode target.  Free vnodes may simply be files which have been stat'd
155  * but not read.  This is somewhat common, and a small cache of such files
156  * should be kept to avoid recreation costs.
157  */
158 static u_long wantfreevnodes;
159 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
160 /* Number of vnodes in the free list. */
161 static u_long freevnodes;
162 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
163 
164 /*
165  * Various variables used for debugging the new implementation of
166  * reassignbuf().
167  * XXX these are probably of (very) limited utility now.
168  */
169 static int reassignbufcalls;
170 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
171 
172 /*
173  * Cache for the mount type id assigned to NFS.  This is used for
174  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
175  */
176 int	nfs_mount_type = -1;
177 
178 /* To keep more than one thread at a time from running vfs_getnewfsid */
179 static struct mtx mntid_mtx;
180 
181 /*
182  * Lock for any access to the following:
183  *	vnode_free_list
184  *	numvnodes
185  *	freevnodes
186  */
187 static struct mtx vnode_free_list_mtx;
188 
189 /* Publicly exported FS */
190 struct nfs_public nfs_pub;
191 
192 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
193 static uma_zone_t vnode_zone;
194 static uma_zone_t vnodepoll_zone;
195 
196 /* Set to 1 to print out reclaim of active vnodes */
197 int	prtactive;
198 
199 /*
200  * The workitem queue.
201  *
202  * It is useful to delay writes of file data and filesystem metadata
203  * for tens of seconds so that quickly created and deleted files need
204  * not waste disk bandwidth being created and removed. To realize this,
205  * we append vnodes to a "workitem" queue. When running with a soft
206  * updates implementation, most pending metadata dependencies should
207  * not wait for more than a few seconds. Thus, mounted on block devices
208  * are delayed only about a half the time that file data is delayed.
209  * Similarly, directory updates are more critical, so are only delayed
210  * about a third the time that file data is delayed. Thus, there are
211  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
212  * one each second (driven off the filesystem syncer process). The
213  * syncer_delayno variable indicates the next queue that is to be processed.
214  * Items that need to be processed soon are placed in this queue:
215  *
216  *	syncer_workitem_pending[syncer_delayno]
217  *
218  * A delay of fifteen seconds is done by placing the request fifteen
219  * entries later in the queue:
220  *
221  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
222  *
223  */
224 static int syncer_delayno;
225 static long syncer_mask;
226 LIST_HEAD(synclist, bufobj);
227 static struct synclist *syncer_workitem_pending;
228 /*
229  * The sync_mtx protects:
230  *	bo->bo_synclist
231  *	sync_vnode_count
232  *	syncer_delayno
233  *	syncer_state
234  *	syncer_workitem_pending
235  *	syncer_worklist_len
236  *	rushjob
237  */
238 static struct mtx sync_mtx;
239 
240 #define SYNCER_MAXDELAY		32
241 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
242 static int syncdelay = 30;		/* max time to delay syncing data */
243 static int filedelay = 30;		/* time to delay syncing files */
244 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
245 static int dirdelay = 29;		/* time to delay syncing directories */
246 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
247 static int metadelay = 28;		/* time to delay syncing metadata */
248 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
249 static int rushjob;		/* number of slots to run ASAP */
250 static int stat_rush_requests;	/* number of times I/O speeded up */
251 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
252 
253 /*
254  * When shutting down the syncer, run it at four times normal speed.
255  */
256 #define SYNCER_SHUTDOWN_SPEEDUP		4
257 static int sync_vnode_count;
258 static int syncer_worklist_len;
259 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
260     syncer_state;
261 
262 /*
263  * Number of vnodes we want to exist at any one time.  This is mostly used
264  * to size hash tables in vnode-related code.  It is normally not used in
265  * getnewvnode(), as wantfreevnodes is normally nonzero.)
266  *
267  * XXX desiredvnodes is historical cruft and should not exist.
268  */
269 int desiredvnodes;
270 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
271     &desiredvnodes, 0, "Maximum number of vnodes");
272 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
273     &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
274 static int vnlru_nowhere;
275 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
276     &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
277 
278 /*
279  * Macros to control when a vnode is freed and recycled.  All require
280  * the vnode interlock.
281  */
282 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
283 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
284 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
285 
286 
287 /*
288  * Initialize the vnode management data structures.
289  */
290 #ifndef	MAXVNODES_MAX
291 #define	MAXVNODES_MAX	100000
292 #endif
293 static void
294 vntblinit(void *dummy __unused)
295 {
296 
297 	/*
298 	 * Desiredvnodes is a function of the physical memory size and
299 	 * the kernel's heap size.  Specifically, desiredvnodes scales
300 	 * in proportion to the physical memory size until two fifths
301 	 * of the kernel's heap size is consumed by vnodes and vm
302 	 * objects.
303 	 */
304 	desiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * vm_kmem_size /
305 	    (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
306 	if (desiredvnodes > MAXVNODES_MAX) {
307 		if (bootverbose)
308 			printf("Reducing kern.maxvnodes %d -> %d\n",
309 			    desiredvnodes, MAXVNODES_MAX);
310 		desiredvnodes = MAXVNODES_MAX;
311 	}
312 	wantfreevnodes = desiredvnodes / 4;
313 	mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
314 	TAILQ_INIT(&vnode_free_list);
315 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
316 	vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
317 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
318 	vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
319 	      NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
320 	/*
321 	 * Initialize the filesystem syncer.
322 	 */
323 	syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
324 		&syncer_mask);
325 	syncer_maxdelay = syncer_mask + 1;
326 	mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
327 }
328 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL)
329 
330 
331 /*
332  * Mark a mount point as busy. Used to synchronize access and to delay
333  * unmounting. Interlock is not released on failure.
334  */
335 int
336 vfs_busy(struct mount *mp, int flags, struct mtx *interlkp,
337     struct thread *td)
338 {
339 	int lkflags;
340 
341 	MNT_ILOCK(mp);
342 	MNT_REF(mp);
343 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
344 		if (flags & LK_NOWAIT) {
345 			MNT_REL(mp);
346 			MNT_IUNLOCK(mp);
347 			return (ENOENT);
348 		}
349 		if (interlkp)
350 			mtx_unlock(interlkp);
351 		mp->mnt_kern_flag |= MNTK_MWAIT;
352 		/*
353 		 * Since all busy locks are shared except the exclusive
354 		 * lock granted when unmounting, the only place that a
355 		 * wakeup needs to be done is at the release of the
356 		 * exclusive lock at the end of dounmount.
357 		 */
358 		msleep(mp, MNT_MTX(mp), PVFS, "vfs_busy", 0);
359 		MNT_REL(mp);
360 		MNT_IUNLOCK(mp);
361 		if (interlkp)
362 			mtx_lock(interlkp);
363 		return (ENOENT);
364 	}
365 	if (interlkp)
366 		mtx_unlock(interlkp);
367 	lkflags = LK_SHARED | LK_INTERLOCK;
368 	if (lockmgr(&mp->mnt_lock, lkflags, MNT_MTX(mp), td))
369 		panic("vfs_busy: unexpected lock failure");
370 	return (0);
371 }
372 
373 /*
374  * Free a busy filesystem.
375  */
376 void
377 vfs_unbusy(struct mount *mp, struct thread *td)
378 {
379 
380 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
381 	vfs_rel(mp);
382 }
383 
384 /*
385  * Lookup a mount point by filesystem identifier.
386  */
387 struct mount *
388 vfs_getvfs(fsid_t *fsid)
389 {
390 	struct mount *mp;
391 
392 	mtx_lock(&mountlist_mtx);
393 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
394 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
395 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
396 			vfs_ref(mp);
397 			mtx_unlock(&mountlist_mtx);
398 			return (mp);
399 		}
400 	}
401 	mtx_unlock(&mountlist_mtx);
402 	return ((struct mount *) 0);
403 }
404 
405 /*
406  * Check if a user can access priveledged mount options.
407  */
408 int
409 vfs_suser(struct mount *mp, struct thread *td)
410 {
411 	int error;
412 
413 	if ((mp->mnt_flag & MNT_USER) == 0 ||
414 	    mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
415 		if ((error = suser(td)) != 0)
416 			return (error);
417 	}
418 	return (0);
419 }
420 
421 /*
422  * Get a new unique fsid.  Try to make its val[0] unique, since this value
423  * will be used to create fake device numbers for stat().  Also try (but
424  * not so hard) make its val[0] unique mod 2^16, since some emulators only
425  * support 16-bit device numbers.  We end up with unique val[0]'s for the
426  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
427  *
428  * Keep in mind that several mounts may be running in parallel.  Starting
429  * the search one past where the previous search terminated is both a
430  * micro-optimization and a defense against returning the same fsid to
431  * different mounts.
432  */
433 void
434 vfs_getnewfsid(struct mount *mp)
435 {
436 	static u_int16_t mntid_base;
437 	struct mount *nmp;
438 	fsid_t tfsid;
439 	int mtype;
440 
441 	mtx_lock(&mntid_mtx);
442 	mtype = mp->mnt_vfc->vfc_typenum;
443 	tfsid.val[1] = mtype;
444 	mtype = (mtype & 0xFF) << 24;
445 	for (;;) {
446 		tfsid.val[0] = makedev(255,
447 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
448 		mntid_base++;
449 		if ((nmp = vfs_getvfs(&tfsid)) == NULL)
450 			break;
451 		vfs_rel(nmp);
452 	}
453 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
454 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
455 	mtx_unlock(&mntid_mtx);
456 }
457 
458 /*
459  * Knob to control the precision of file timestamps:
460  *
461  *   0 = seconds only; nanoseconds zeroed.
462  *   1 = seconds and nanoseconds, accurate within 1/HZ.
463  *   2 = seconds and nanoseconds, truncated to microseconds.
464  * >=3 = seconds and nanoseconds, maximum precision.
465  */
466 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
467 
468 static int timestamp_precision = TSP_SEC;
469 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
470     &timestamp_precision, 0, "");
471 
472 /*
473  * Get a current timestamp.
474  */
475 void
476 vfs_timestamp(struct timespec *tsp)
477 {
478 	struct timeval tv;
479 
480 	switch (timestamp_precision) {
481 	case TSP_SEC:
482 		tsp->tv_sec = time_second;
483 		tsp->tv_nsec = 0;
484 		break;
485 	case TSP_HZ:
486 		getnanotime(tsp);
487 		break;
488 	case TSP_USEC:
489 		microtime(&tv);
490 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
491 		break;
492 	case TSP_NSEC:
493 	default:
494 		nanotime(tsp);
495 		break;
496 	}
497 }
498 
499 /*
500  * Set vnode attributes to VNOVAL
501  */
502 void
503 vattr_null(struct vattr *vap)
504 {
505 
506 	vap->va_type = VNON;
507 	vap->va_size = VNOVAL;
508 	vap->va_bytes = VNOVAL;
509 	vap->va_mode = VNOVAL;
510 	vap->va_nlink = VNOVAL;
511 	vap->va_uid = VNOVAL;
512 	vap->va_gid = VNOVAL;
513 	vap->va_fsid = VNOVAL;
514 	vap->va_fileid = VNOVAL;
515 	vap->va_blocksize = VNOVAL;
516 	vap->va_rdev = VNOVAL;
517 	vap->va_atime.tv_sec = VNOVAL;
518 	vap->va_atime.tv_nsec = VNOVAL;
519 	vap->va_mtime.tv_sec = VNOVAL;
520 	vap->va_mtime.tv_nsec = VNOVAL;
521 	vap->va_ctime.tv_sec = VNOVAL;
522 	vap->va_ctime.tv_nsec = VNOVAL;
523 	vap->va_birthtime.tv_sec = VNOVAL;
524 	vap->va_birthtime.tv_nsec = VNOVAL;
525 	vap->va_flags = VNOVAL;
526 	vap->va_gen = VNOVAL;
527 	vap->va_vaflags = 0;
528 }
529 
530 /*
531  * This routine is called when we have too many vnodes.  It attempts
532  * to free <count> vnodes and will potentially free vnodes that still
533  * have VM backing store (VM backing store is typically the cause
534  * of a vnode blowout so we want to do this).  Therefore, this operation
535  * is not considered cheap.
536  *
537  * A number of conditions may prevent a vnode from being reclaimed.
538  * the buffer cache may have references on the vnode, a directory
539  * vnode may still have references due to the namei cache representing
540  * underlying files, or the vnode may be in active use.   It is not
541  * desireable to reuse such vnodes.  These conditions may cause the
542  * number of vnodes to reach some minimum value regardless of what
543  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
544  */
545 static int
546 vlrureclaim(struct mount *mp)
547 {
548 	struct thread *td;
549 	struct vnode *vp;
550 	int done;
551 	int trigger;
552 	int usevnodes;
553 	int count;
554 
555 	/*
556 	 * Calculate the trigger point, don't allow user
557 	 * screwups to blow us up.   This prevents us from
558 	 * recycling vnodes with lots of resident pages.  We
559 	 * aren't trying to free memory, we are trying to
560 	 * free vnodes.
561 	 */
562 	usevnodes = desiredvnodes;
563 	if (usevnodes <= 0)
564 		usevnodes = 1;
565 	trigger = cnt.v_page_count * 2 / usevnodes;
566 	done = 0;
567 	td = curthread;
568 	vn_start_write(NULL, &mp, V_WAIT);
569 	MNT_ILOCK(mp);
570 	count = mp->mnt_nvnodelistsize / 10 + 1;
571 	while (count != 0) {
572 		vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
573 		while (vp != NULL && vp->v_type == VMARKER)
574 			vp = TAILQ_NEXT(vp, v_nmntvnodes);
575 		if (vp == NULL)
576 			break;
577 		TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
578 		TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
579 		--count;
580 		if (!VI_TRYLOCK(vp))
581 			goto next_iter;
582 		/*
583 		 * If it's been deconstructed already, it's still
584 		 * referenced, or it exceeds the trigger, skip it.
585 		 */
586 		if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) ||
587 		    (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
588 		    vp->v_object->resident_page_count > trigger)) {
589 			VI_UNLOCK(vp);
590 			goto next_iter;
591 		}
592 		MNT_IUNLOCK(mp);
593 		vholdl(vp);
594 		if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT, td)) {
595 			vdrop(vp);
596 			goto next_iter_mntunlocked;
597 		}
598 		VI_LOCK(vp);
599 		/*
600 		 * v_usecount may have been bumped after VOP_LOCK() dropped
601 		 * the vnode interlock and before it was locked again.
602 		 *
603 		 * It is not necessary to recheck VI_DOOMED because it can
604 		 * only be set by another thread that holds both the vnode
605 		 * lock and vnode interlock.  If another thread has the
606 		 * vnode lock before we get to VOP_LOCK() and obtains the
607 		 * vnode interlock after VOP_LOCK() drops the vnode
608 		 * interlock, the other thread will be unable to drop the
609 		 * vnode lock before our VOP_LOCK() call fails.
610 		 */
611 		if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) ||
612 		    (vp->v_object != NULL &&
613 		    vp->v_object->resident_page_count > trigger)) {
614 			VOP_UNLOCK(vp, LK_INTERLOCK, td);
615 			goto next_iter_mntunlocked;
616 		}
617 		KASSERT((vp->v_iflag & VI_DOOMED) == 0,
618 		    ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
619 		vgonel(vp);
620 		VOP_UNLOCK(vp, 0, td);
621 		vdropl(vp);
622 		done++;
623 next_iter_mntunlocked:
624 		if ((count % 256) != 0)
625 			goto relock_mnt;
626 		goto yield;
627 next_iter:
628 		if ((count % 256) != 0)
629 			continue;
630 		MNT_IUNLOCK(mp);
631 yield:
632 		uio_yield();
633 relock_mnt:
634 		MNT_ILOCK(mp);
635 	}
636 	MNT_IUNLOCK(mp);
637 	vn_finished_write(mp);
638 	return done;
639 }
640 
641 /*
642  * Attempt to keep the free list at wantfreevnodes length.
643  */
644 static void
645 vnlru_free(int count)
646 {
647 	struct vnode *vp;
648 	int vfslocked;
649 
650 	mtx_assert(&vnode_free_list_mtx, MA_OWNED);
651 	for (; count > 0; count--) {
652 		vp = TAILQ_FIRST(&vnode_free_list);
653 		/*
654 		 * The list can be modified while the free_list_mtx
655 		 * has been dropped and vp could be NULL here.
656 		 */
657 		if (!vp)
658 			break;
659 		VNASSERT(vp->v_op != NULL, vp,
660 		    ("vnlru_free: vnode already reclaimed."));
661 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
662 		/*
663 		 * Don't recycle if we can't get the interlock.
664 		 */
665 		if (!VI_TRYLOCK(vp)) {
666 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
667 			continue;
668 		}
669 		VNASSERT(VCANRECYCLE(vp), vp,
670 		    ("vp inconsistent on freelist"));
671 		freevnodes--;
672 		vp->v_iflag &= ~VI_FREE;
673 		vholdl(vp);
674 		mtx_unlock(&vnode_free_list_mtx);
675 		VI_UNLOCK(vp);
676 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
677 		vtryrecycle(vp);
678 		VFS_UNLOCK_GIANT(vfslocked);
679 		/*
680 		 * If the recycled succeeded this vdrop will actually free
681 		 * the vnode.  If not it will simply place it back on
682 		 * the free list.
683 		 */
684 		vdrop(vp);
685 		mtx_lock(&vnode_free_list_mtx);
686 	}
687 }
688 /*
689  * Attempt to recycle vnodes in a context that is always safe to block.
690  * Calling vlrurecycle() from the bowels of filesystem code has some
691  * interesting deadlock problems.
692  */
693 static struct proc *vnlruproc;
694 static int vnlruproc_sig;
695 
696 static void
697 vnlru_proc(void)
698 {
699 	struct mount *mp, *nmp;
700 	int done;
701 	struct proc *p = vnlruproc;
702 	struct thread *td = FIRST_THREAD_IN_PROC(p);
703 
704 	mtx_lock(&Giant);
705 
706 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
707 	    SHUTDOWN_PRI_FIRST);
708 
709 	for (;;) {
710 		kthread_suspend_check(p);
711 		mtx_lock(&vnode_free_list_mtx);
712 		if (freevnodes > wantfreevnodes)
713 			vnlru_free(freevnodes - wantfreevnodes);
714 		if (numvnodes <= desiredvnodes * 9 / 10) {
715 			vnlruproc_sig = 0;
716 			wakeup(&vnlruproc_sig);
717 			msleep(vnlruproc, &vnode_free_list_mtx,
718 			    PVFS|PDROP, "vlruwt", hz);
719 			continue;
720 		}
721 		mtx_unlock(&vnode_free_list_mtx);
722 		done = 0;
723 		mtx_lock(&mountlist_mtx);
724 		for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
725 			int vfsunlocked;
726 			if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
727 				nmp = TAILQ_NEXT(mp, mnt_list);
728 				continue;
729 			}
730 			if (!VFS_NEEDSGIANT(mp)) {
731 				mtx_unlock(&Giant);
732 				vfsunlocked = 1;
733 			} else
734 				vfsunlocked = 0;
735 			done += vlrureclaim(mp);
736 			if (vfsunlocked)
737 				mtx_lock(&Giant);
738 			mtx_lock(&mountlist_mtx);
739 			nmp = TAILQ_NEXT(mp, mnt_list);
740 			vfs_unbusy(mp, td);
741 		}
742 		mtx_unlock(&mountlist_mtx);
743 		if (done == 0) {
744 #if 0
745 			/* These messages are temporary debugging aids */
746 			if (vnlru_nowhere < 5)
747 				printf("vnlru process getting nowhere..\n");
748 			else if (vnlru_nowhere == 5)
749 				printf("vnlru process messages stopped.\n");
750 #endif
751 			vnlru_nowhere++;
752 			tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
753 		} else
754 			uio_yield();
755 	}
756 }
757 
758 static struct kproc_desc vnlru_kp = {
759 	"vnlru",
760 	vnlru_proc,
761 	&vnlruproc
762 };
763 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
764 
765 /*
766  * Routines having to do with the management of the vnode table.
767  */
768 
769 static void
770 vdestroy(struct vnode *vp)
771 {
772 	struct bufobj *bo;
773 
774 	CTR1(KTR_VFS, "vdestroy vp %p", vp);
775 	mtx_lock(&vnode_free_list_mtx);
776 	numvnodes--;
777 	mtx_unlock(&vnode_free_list_mtx);
778 	bo = &vp->v_bufobj;
779 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
780 	    ("cleaned vnode still on the free list."));
781 	VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
782 	VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
783 	VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
784 	VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
785 	VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
786 	VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
787 	VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
788 	VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
789 	VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
790 	VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
791 	VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
792 	VI_UNLOCK(vp);
793 #ifdef MAC
794 	mac_destroy_vnode(vp);
795 #endif
796 	if (vp->v_pollinfo != NULL) {
797 		knlist_destroy(&vp->v_pollinfo->vpi_selinfo.si_note);
798 		mtx_destroy(&vp->v_pollinfo->vpi_lock);
799 		uma_zfree(vnodepoll_zone, vp->v_pollinfo);
800 	}
801 #ifdef INVARIANTS
802 	/* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */
803 	vp->v_op = NULL;
804 #endif
805 	lockdestroy(vp->v_vnlock);
806 	mtx_destroy(&vp->v_interlock);
807 	uma_zfree(vnode_zone, vp);
808 }
809 
810 /*
811  * Try to recycle a freed vnode.  We abort if anyone picks up a reference
812  * before we actually vgone().  This function must be called with the vnode
813  * held to prevent the vnode from being returned to the free list midway
814  * through vgone().
815  */
816 static int
817 vtryrecycle(struct vnode *vp)
818 {
819 	struct thread *td = curthread;
820 	struct mount *vnmp;
821 
822 	CTR1(KTR_VFS, "vtryrecycle: trying vp %p", vp);
823 	VNASSERT(vp->v_holdcnt, vp,
824 	    ("vtryrecycle: Recycling vp %p without a reference.", vp));
825 	/*
826 	 * This vnode may found and locked via some other list, if so we
827 	 * can't recycle it yet.
828 	 */
829 	if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT, td) != 0)
830 		return (EWOULDBLOCK);
831 	/*
832 	 * Don't recycle if its filesystem is being suspended.
833 	 */
834 	if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
835 		VOP_UNLOCK(vp, 0, td);
836 		return (EBUSY);
837 	}
838 	/*
839 	 * If we got this far, we need to acquire the interlock and see if
840 	 * anyone picked up this vnode from another list.  If not, we will
841 	 * mark it with DOOMED via vgonel() so that anyone who does find it
842 	 * will skip over it.
843 	 */
844 	VI_LOCK(vp);
845 	if (vp->v_usecount) {
846 		VOP_UNLOCK(vp, LK_INTERLOCK, td);
847 		vn_finished_write(vnmp);
848 		return (EBUSY);
849 	}
850 	if ((vp->v_iflag & VI_DOOMED) == 0)
851 		vgonel(vp);
852 	VOP_UNLOCK(vp, LK_INTERLOCK, td);
853 	vn_finished_write(vnmp);
854 	CTR1(KTR_VFS, "vtryrecycle: recycled vp %p", vp);
855 	return (0);
856 }
857 
858 /*
859  * Return the next vnode from the free list.
860  */
861 int
862 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
863     struct vnode **vpp)
864 {
865 	struct vnode *vp = NULL;
866 	struct bufobj *bo;
867 
868 	mtx_lock(&vnode_free_list_mtx);
869 	/*
870 	 * Lend our context to reclaim vnodes if they've exceeded the max.
871 	 */
872 	if (freevnodes > wantfreevnodes)
873 		vnlru_free(1);
874 	/*
875 	 * Wait for available vnodes.
876 	 */
877 	if (numvnodes > desiredvnodes) {
878 		if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) {
879 			/*
880 			 * File system is beeing suspended, we cannot risk a
881 			 * deadlock here, so allocate new vnode anyway.
882 			 */
883 			if (freevnodes > wantfreevnodes)
884 				vnlru_free(freevnodes - wantfreevnodes);
885 			goto alloc;
886 		}
887 		if (vnlruproc_sig == 0) {
888 			vnlruproc_sig = 1;      /* avoid unnecessary wakeups */
889 			wakeup(vnlruproc);
890 		}
891 		msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
892 		    "vlruwk", hz);
893 #if 0	/* XXX Not all VFS_VGET/ffs_vget callers check returns. */
894 		if (numvnodes > desiredvnodes) {
895 			mtx_unlock(&vnode_free_list_mtx);
896 			return (ENFILE);
897 		}
898 #endif
899 	}
900 alloc:
901 	numvnodes++;
902 	mtx_unlock(&vnode_free_list_mtx);
903 	vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
904 	/*
905 	 * Setup locks.
906 	 */
907 	vp->v_vnlock = &vp->v_lock;
908 	mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
909 	/*
910 	 * By default, don't allow shared locks unless filesystems
911 	 * opt-in.
912 	 */
913 	lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
914 	/*
915 	 * Initialize bufobj.
916 	 */
917 	bo = &vp->v_bufobj;
918 	bo->__bo_vnode = vp;
919 	bo->bo_mtx = &vp->v_interlock;
920 	bo->bo_ops = &buf_ops_bio;
921 	bo->bo_private = vp;
922 	TAILQ_INIT(&bo->bo_clean.bv_hd);
923 	TAILQ_INIT(&bo->bo_dirty.bv_hd);
924 	/*
925 	 * Initialize namecache.
926 	 */
927 	LIST_INIT(&vp->v_cache_src);
928 	TAILQ_INIT(&vp->v_cache_dst);
929 	/*
930 	 * Finalize various vnode identity bits.
931 	 */
932 	vp->v_type = VNON;
933 	vp->v_tag = tag;
934 	vp->v_op = vops;
935 	v_incr_usecount(vp);
936 	vp->v_data = 0;
937 #ifdef MAC
938 	mac_init_vnode(vp);
939 	if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
940 		mac_associate_vnode_singlelabel(mp, vp);
941 	else if (mp == NULL)
942 		printf("NULL mp in getnewvnode()\n");
943 #endif
944 	if (mp != NULL) {
945 		insmntque(vp, mp);
946 		bo->bo_bsize = mp->mnt_stat.f_iosize;
947 		if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
948 			vp->v_vflag |= VV_NOKNOTE;
949 	}
950 
951 	CTR2(KTR_VFS, "getnewvnode: mp %p vp %p", mp, vp);
952 	*vpp = vp;
953 	return (0);
954 }
955 
956 /*
957  * Delete from old mount point vnode list, if on one.
958  */
959 static void
960 delmntque(struct vnode *vp)
961 {
962 	struct mount *mp;
963 
964 	mp = vp->v_mount;
965 	if (mp == NULL)
966 		return;
967 	MNT_ILOCK(mp);
968 	vp->v_mount = NULL;
969 	VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
970 		("bad mount point vnode list size"));
971 	TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
972 	mp->mnt_nvnodelistsize--;
973 	MNT_REL(mp);
974 	MNT_IUNLOCK(mp);
975 }
976 
977 /*
978  * Insert into list of vnodes for the new mount point, if available.
979  */
980 static void
981 insmntque(struct vnode *vp, struct mount *mp)
982 {
983 
984 	vp->v_mount = mp;
985 	VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
986 	MNT_ILOCK(mp);
987 	MNT_REF(mp);
988 	TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
989 	VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
990 		("neg mount point vnode list size"));
991 	mp->mnt_nvnodelistsize++;
992 	MNT_IUNLOCK(mp);
993 }
994 
995 /*
996  * Flush out and invalidate all buffers associated with a bufobj
997  * Called with the underlying object locked.
998  */
999 int
1000 bufobj_invalbuf(struct bufobj *bo, int flags, struct thread *td, int slpflag,
1001     int slptimeo)
1002 {
1003 	int error;
1004 
1005 	BO_LOCK(bo);
1006 	if (flags & V_SAVE) {
1007 		error = bufobj_wwait(bo, slpflag, slptimeo);
1008 		if (error) {
1009 			BO_UNLOCK(bo);
1010 			return (error);
1011 		}
1012 		if (bo->bo_dirty.bv_cnt > 0) {
1013 			BO_UNLOCK(bo);
1014 			if ((error = BO_SYNC(bo, MNT_WAIT, td)) != 0)
1015 				return (error);
1016 			/*
1017 			 * XXX We could save a lock/unlock if this was only
1018 			 * enabled under INVARIANTS
1019 			 */
1020 			BO_LOCK(bo);
1021 			if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1022 				panic("vinvalbuf: dirty bufs");
1023 		}
1024 	}
1025 	/*
1026 	 * If you alter this loop please notice that interlock is dropped and
1027 	 * reacquired in flushbuflist.  Special care is needed to ensure that
1028 	 * no race conditions occur from this.
1029 	 */
1030 	do {
1031 		error = flushbuflist(&bo->bo_clean,
1032 		    flags, bo, slpflag, slptimeo);
1033 		if (error == 0)
1034 			error = flushbuflist(&bo->bo_dirty,
1035 			    flags, bo, slpflag, slptimeo);
1036 		if (error != 0 && error != EAGAIN) {
1037 			BO_UNLOCK(bo);
1038 			return (error);
1039 		}
1040 	} while (error != 0);
1041 
1042 	/*
1043 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
1044 	 * have write I/O in-progress but if there is a VM object then the
1045 	 * VM object can also have read-I/O in-progress.
1046 	 */
1047 	do {
1048 		bufobj_wwait(bo, 0, 0);
1049 		BO_UNLOCK(bo);
1050 		if (bo->bo_object != NULL) {
1051 			VM_OBJECT_LOCK(bo->bo_object);
1052 			vm_object_pip_wait(bo->bo_object, "bovlbx");
1053 			VM_OBJECT_UNLOCK(bo->bo_object);
1054 		}
1055 		BO_LOCK(bo);
1056 	} while (bo->bo_numoutput > 0);
1057 	BO_UNLOCK(bo);
1058 
1059 	/*
1060 	 * Destroy the copy in the VM cache, too.
1061 	 */
1062 	if (bo->bo_object != NULL) {
1063 		VM_OBJECT_LOCK(bo->bo_object);
1064 		vm_object_page_remove(bo->bo_object, 0, 0,
1065 			(flags & V_SAVE) ? TRUE : FALSE);
1066 		VM_OBJECT_UNLOCK(bo->bo_object);
1067 	}
1068 
1069 #ifdef INVARIANTS
1070 	BO_LOCK(bo);
1071 	if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1072 	    (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1073 		panic("vinvalbuf: flush failed");
1074 	BO_UNLOCK(bo);
1075 #endif
1076 	return (0);
1077 }
1078 
1079 /*
1080  * Flush out and invalidate all buffers associated with a vnode.
1081  * Called with the underlying object locked.
1082  */
1083 int
1084 vinvalbuf(struct vnode *vp, int flags, struct thread *td, int slpflag,
1085     int slptimeo)
1086 {
1087 
1088 	CTR2(KTR_VFS, "vinvalbuf vp %p flags %d", vp, flags);
1089 	ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1090 	return (bufobj_invalbuf(&vp->v_bufobj, flags, td, slpflag, slptimeo));
1091 }
1092 
1093 /*
1094  * Flush out buffers on the specified list.
1095  *
1096  */
1097 static int
1098 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1099     int slptimeo)
1100 {
1101 	struct buf *bp, *nbp;
1102 	int retval, error;
1103 	daddr_t lblkno;
1104 	b_xflags_t xflags;
1105 
1106 	ASSERT_BO_LOCKED(bo);
1107 
1108 	retval = 0;
1109 	TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1110 		if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1111 		    ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1112 			continue;
1113 		}
1114 		lblkno = 0;
1115 		xflags = 0;
1116 		if (nbp != NULL) {
1117 			lblkno = nbp->b_lblkno;
1118 			xflags = nbp->b_xflags &
1119 				(BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1120 		}
1121 		retval = EAGAIN;
1122 		error = BUF_TIMELOCK(bp,
1123 		    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1124 		    "flushbuf", slpflag, slptimeo);
1125 		if (error) {
1126 			BO_LOCK(bo);
1127 			return (error != ENOLCK ? error : EAGAIN);
1128 		}
1129 		KASSERT(bp->b_bufobj == bo,
1130 	            ("bp %p wrong b_bufobj %p should be %p",
1131 		    bp, bp->b_bufobj, bo));
1132 		if (bp->b_bufobj != bo) {	/* XXX: necessary ? */
1133 			BUF_UNLOCK(bp);
1134 			BO_LOCK(bo);
1135 			return (EAGAIN);
1136 		}
1137 		/*
1138 		 * XXX Since there are no node locks for NFS, I
1139 		 * believe there is a slight chance that a delayed
1140 		 * write will occur while sleeping just above, so
1141 		 * check for it.
1142 		 */
1143 		if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1144 		    (flags & V_SAVE)) {
1145 			bremfree(bp);
1146 			bp->b_flags |= B_ASYNC;
1147 			bwrite(bp);
1148 			BO_LOCK(bo);
1149 			return (EAGAIN);	/* XXX: why not loop ? */
1150 		}
1151 		bremfree(bp);
1152 		bp->b_flags |= (B_INVAL | B_RELBUF);
1153 		bp->b_flags &= ~B_ASYNC;
1154 		brelse(bp);
1155 		BO_LOCK(bo);
1156 		if (nbp != NULL &&
1157 		    (nbp->b_bufobj != bo ||
1158 		     nbp->b_lblkno != lblkno ||
1159 		     (nbp->b_xflags &
1160 		      (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1161 			break;			/* nbp invalid */
1162 	}
1163 	return (retval);
1164 }
1165 
1166 /*
1167  * Truncate a file's buffer and pages to a specified length.  This
1168  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1169  * sync activity.
1170  */
1171 int
1172 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
1173     off_t length, int blksize)
1174 {
1175 	struct buf *bp, *nbp;
1176 	int anyfreed;
1177 	int trunclbn;
1178 	struct bufobj *bo;
1179 
1180 	CTR2(KTR_VFS, "vtruncbuf vp %p length %jd", vp, length);
1181 	/*
1182 	 * Round up to the *next* lbn.
1183 	 */
1184 	trunclbn = (length + blksize - 1) / blksize;
1185 
1186 	ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1187 restart:
1188 	VI_LOCK(vp);
1189 	bo = &vp->v_bufobj;
1190 	anyfreed = 1;
1191 	for (;anyfreed;) {
1192 		anyfreed = 0;
1193 		TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1194 			if (bp->b_lblkno < trunclbn)
1195 				continue;
1196 			if (BUF_LOCK(bp,
1197 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1198 			    VI_MTX(vp)) == ENOLCK)
1199 				goto restart;
1200 
1201 			bremfree(bp);
1202 			bp->b_flags |= (B_INVAL | B_RELBUF);
1203 			bp->b_flags &= ~B_ASYNC;
1204 			brelse(bp);
1205 			anyfreed = 1;
1206 
1207 			if (nbp != NULL &&
1208 			    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1209 			    (nbp->b_vp != vp) ||
1210 			    (nbp->b_flags & B_DELWRI))) {
1211 				goto restart;
1212 			}
1213 			VI_LOCK(vp);
1214 		}
1215 
1216 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1217 			if (bp->b_lblkno < trunclbn)
1218 				continue;
1219 			if (BUF_LOCK(bp,
1220 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1221 			    VI_MTX(vp)) == ENOLCK)
1222 				goto restart;
1223 			bremfree(bp);
1224 			bp->b_flags |= (B_INVAL | B_RELBUF);
1225 			bp->b_flags &= ~B_ASYNC;
1226 			brelse(bp);
1227 			anyfreed = 1;
1228 			if (nbp != NULL &&
1229 			    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1230 			    (nbp->b_vp != vp) ||
1231 			    (nbp->b_flags & B_DELWRI) == 0)) {
1232 				goto restart;
1233 			}
1234 			VI_LOCK(vp);
1235 		}
1236 	}
1237 
1238 	if (length > 0) {
1239 restartsync:
1240 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1241 			if (bp->b_lblkno > 0)
1242 				continue;
1243 			/*
1244 			 * Since we hold the vnode lock this should only
1245 			 * fail if we're racing with the buf daemon.
1246 			 */
1247 			if (BUF_LOCK(bp,
1248 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1249 			    VI_MTX(vp)) == ENOLCK) {
1250 				goto restart;
1251 			}
1252 			VNASSERT((bp->b_flags & B_DELWRI), vp,
1253 			    ("buf(%p) on dirty queue without DELWRI", bp));
1254 
1255 			bremfree(bp);
1256 			bawrite(bp);
1257 			VI_LOCK(vp);
1258 			goto restartsync;
1259 		}
1260 	}
1261 
1262 	bufobj_wwait(bo, 0, 0);
1263 	VI_UNLOCK(vp);
1264 	vnode_pager_setsize(vp, length);
1265 
1266 	return (0);
1267 }
1268 
1269 /*
1270  * buf_splay() - splay tree core for the clean/dirty list of buffers in
1271  * 		 a vnode.
1272  *
1273  *	NOTE: We have to deal with the special case of a background bitmap
1274  *	buffer, a situation where two buffers will have the same logical
1275  *	block offset.  We want (1) only the foreground buffer to be accessed
1276  *	in a lookup and (2) must differentiate between the foreground and
1277  *	background buffer in the splay tree algorithm because the splay
1278  *	tree cannot normally handle multiple entities with the same 'index'.
1279  *	We accomplish this by adding differentiating flags to the splay tree's
1280  *	numerical domain.
1281  */
1282 static
1283 struct buf *
1284 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1285 {
1286 	struct buf dummy;
1287 	struct buf *lefttreemax, *righttreemin, *y;
1288 
1289 	if (root == NULL)
1290 		return (NULL);
1291 	lefttreemax = righttreemin = &dummy;
1292 	for (;;) {
1293 		if (lblkno < root->b_lblkno ||
1294 		    (lblkno == root->b_lblkno &&
1295 		    (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1296 			if ((y = root->b_left) == NULL)
1297 				break;
1298 			if (lblkno < y->b_lblkno) {
1299 				/* Rotate right. */
1300 				root->b_left = y->b_right;
1301 				y->b_right = root;
1302 				root = y;
1303 				if ((y = root->b_left) == NULL)
1304 					break;
1305 			}
1306 			/* Link into the new root's right tree. */
1307 			righttreemin->b_left = root;
1308 			righttreemin = root;
1309 		} else if (lblkno > root->b_lblkno ||
1310 		    (lblkno == root->b_lblkno &&
1311 		    (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1312 			if ((y = root->b_right) == NULL)
1313 				break;
1314 			if (lblkno > y->b_lblkno) {
1315 				/* Rotate left. */
1316 				root->b_right = y->b_left;
1317 				y->b_left = root;
1318 				root = y;
1319 				if ((y = root->b_right) == NULL)
1320 					break;
1321 			}
1322 			/* Link into the new root's left tree. */
1323 			lefttreemax->b_right = root;
1324 			lefttreemax = root;
1325 		} else {
1326 			break;
1327 		}
1328 		root = y;
1329 	}
1330 	/* Assemble the new root. */
1331 	lefttreemax->b_right = root->b_left;
1332 	righttreemin->b_left = root->b_right;
1333 	root->b_left = dummy.b_right;
1334 	root->b_right = dummy.b_left;
1335 	return (root);
1336 }
1337 
1338 static void
1339 buf_vlist_remove(struct buf *bp)
1340 {
1341 	struct buf *root;
1342 	struct bufv *bv;
1343 
1344 	KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1345 	ASSERT_BO_LOCKED(bp->b_bufobj);
1346 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1347 	    (BX_VNDIRTY|BX_VNCLEAN),
1348 	    ("buf_vlist_remove: Buf %p is on two lists", bp));
1349 	if (bp->b_xflags & BX_VNDIRTY)
1350 		bv = &bp->b_bufobj->bo_dirty;
1351 	else
1352 		bv = &bp->b_bufobj->bo_clean;
1353 	if (bp != bv->bv_root) {
1354 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1355 		KASSERT(root == bp, ("splay lookup failed in remove"));
1356 	}
1357 	if (bp->b_left == NULL) {
1358 		root = bp->b_right;
1359 	} else {
1360 		root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1361 		root->b_right = bp->b_right;
1362 	}
1363 	bv->bv_root = root;
1364 	TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1365 	bv->bv_cnt--;
1366 	bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1367 }
1368 
1369 /*
1370  * Add the buffer to the sorted clean or dirty block list using a
1371  * splay tree algorithm.
1372  *
1373  * NOTE: xflags is passed as a constant, optimizing this inline function!
1374  */
1375 static void
1376 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1377 {
1378 	struct buf *root;
1379 	struct bufv *bv;
1380 
1381 	ASSERT_BO_LOCKED(bo);
1382 	KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1383 	    ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1384 	bp->b_xflags |= xflags;
1385 	if (xflags & BX_VNDIRTY)
1386 		bv = &bo->bo_dirty;
1387 	else
1388 		bv = &bo->bo_clean;
1389 
1390 	root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1391 	if (root == NULL) {
1392 		bp->b_left = NULL;
1393 		bp->b_right = NULL;
1394 		TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1395 	} else if (bp->b_lblkno < root->b_lblkno ||
1396 	    (bp->b_lblkno == root->b_lblkno &&
1397 	    (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1398 		bp->b_left = root->b_left;
1399 		bp->b_right = root;
1400 		root->b_left = NULL;
1401 		TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1402 	} else {
1403 		bp->b_right = root->b_right;
1404 		bp->b_left = root;
1405 		root->b_right = NULL;
1406 		TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1407 	}
1408 	bv->bv_cnt++;
1409 	bv->bv_root = bp;
1410 }
1411 
1412 /*
1413  * Lookup a buffer using the splay tree.  Note that we specifically avoid
1414  * shadow buffers used in background bitmap writes.
1415  *
1416  * This code isn't quite efficient as it could be because we are maintaining
1417  * two sorted lists and do not know which list the block resides in.
1418  *
1419  * During a "make buildworld" the desired buffer is found at one of
1420  * the roots more than 60% of the time.  Thus, checking both roots
1421  * before performing either splay eliminates unnecessary splays on the
1422  * first tree splayed.
1423  */
1424 struct buf *
1425 gbincore(struct bufobj *bo, daddr_t lblkno)
1426 {
1427 	struct buf *bp;
1428 
1429 	ASSERT_BO_LOCKED(bo);
1430 	if ((bp = bo->bo_clean.bv_root) != NULL &&
1431 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1432 		return (bp);
1433 	if ((bp = bo->bo_dirty.bv_root) != NULL &&
1434 	    bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1435 		return (bp);
1436 	if ((bp = bo->bo_clean.bv_root) != NULL) {
1437 		bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1438 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1439 			return (bp);
1440 	}
1441 	if ((bp = bo->bo_dirty.bv_root) != NULL) {
1442 		bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1443 		if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1444 			return (bp);
1445 	}
1446 	return (NULL);
1447 }
1448 
1449 /*
1450  * Associate a buffer with a vnode.
1451  */
1452 void
1453 bgetvp(struct vnode *vp, struct buf *bp)
1454 {
1455 
1456 	VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1457 
1458 	CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1459 	VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1460 	    ("bgetvp: bp already attached! %p", bp));
1461 
1462 	ASSERT_VI_LOCKED(vp, "bgetvp");
1463 	vholdl(vp);
1464 	if (VFS_NEEDSGIANT(vp->v_mount) ||
1465 	    vp->v_bufobj.bo_flag & BO_NEEDSGIANT)
1466 		bp->b_flags |= B_NEEDSGIANT;
1467 	bp->b_vp = vp;
1468 	bp->b_bufobj = &vp->v_bufobj;
1469 	/*
1470 	 * Insert onto list for new vnode.
1471 	 */
1472 	buf_vlist_add(bp, &vp->v_bufobj, BX_VNCLEAN);
1473 }
1474 
1475 /*
1476  * Disassociate a buffer from a vnode.
1477  */
1478 void
1479 brelvp(struct buf *bp)
1480 {
1481 	struct bufobj *bo;
1482 	struct vnode *vp;
1483 
1484 	CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1485 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1486 
1487 	/*
1488 	 * Delete from old vnode list, if on one.
1489 	 */
1490 	vp = bp->b_vp;		/* XXX */
1491 	bo = bp->b_bufobj;
1492 	BO_LOCK(bo);
1493 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1494 		buf_vlist_remove(bp);
1495 	else
1496 		panic("brelvp: Buffer %p not on queue.", bp);
1497 	if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1498 		bo->bo_flag &= ~BO_ONWORKLST;
1499 		mtx_lock(&sync_mtx);
1500 		LIST_REMOVE(bo, bo_synclist);
1501  		syncer_worklist_len--;
1502 		mtx_unlock(&sync_mtx);
1503 	}
1504 	bp->b_flags &= ~B_NEEDSGIANT;
1505 	bp->b_vp = NULL;
1506 	bp->b_bufobj = NULL;
1507 	vdropl(vp);
1508 }
1509 
1510 /*
1511  * Add an item to the syncer work queue.
1512  */
1513 static void
1514 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1515 {
1516 	int slot;
1517 
1518 	ASSERT_BO_LOCKED(bo);
1519 
1520 	mtx_lock(&sync_mtx);
1521 	if (bo->bo_flag & BO_ONWORKLST)
1522 		LIST_REMOVE(bo, bo_synclist);
1523 	else {
1524 		bo->bo_flag |= BO_ONWORKLST;
1525  		syncer_worklist_len++;
1526 	}
1527 
1528 	if (delay > syncer_maxdelay - 2)
1529 		delay = syncer_maxdelay - 2;
1530 	slot = (syncer_delayno + delay) & syncer_mask;
1531 
1532 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist);
1533 	mtx_unlock(&sync_mtx);
1534 }
1535 
1536 static int
1537 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1538 {
1539 	int error, len;
1540 
1541 	mtx_lock(&sync_mtx);
1542 	len = syncer_worklist_len - sync_vnode_count;
1543 	mtx_unlock(&sync_mtx);
1544 	error = SYSCTL_OUT(req, &len, sizeof(len));
1545 	return (error);
1546 }
1547 
1548 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1549     sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1550 
1551 static struct proc *updateproc;
1552 static void sched_sync(void);
1553 static struct kproc_desc up_kp = {
1554 	"syncer",
1555 	sched_sync,
1556 	&updateproc
1557 };
1558 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1559 
1560 static int
1561 sync_vnode(struct bufobj *bo, struct thread *td)
1562 {
1563 	struct vnode *vp;
1564 	struct mount *mp;
1565 
1566 	vp = bo->__bo_vnode; 	/* XXX */
1567 	if (VOP_ISLOCKED(vp, NULL) != 0)
1568 		return (1);
1569 	if (VI_TRYLOCK(vp) == 0)
1570 		return (1);
1571 	/*
1572 	 * We use vhold in case the vnode does not
1573 	 * successfully sync.  vhold prevents the vnode from
1574 	 * going away when we unlock the sync_mtx so that
1575 	 * we can acquire the vnode interlock.
1576 	 */
1577 	vholdl(vp);
1578 	mtx_unlock(&sync_mtx);
1579 	VI_UNLOCK(vp);
1580 	if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1581 		vdrop(vp);
1582 		mtx_lock(&sync_mtx);
1583 		return (1);
1584 	}
1585 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1586 	(void) VOP_FSYNC(vp, MNT_LAZY, td);
1587 	VOP_UNLOCK(vp, 0, td);
1588 	vn_finished_write(mp);
1589 	VI_LOCK(vp);
1590 	if ((bo->bo_flag & BO_ONWORKLST) != 0) {
1591 		/*
1592 		 * Put us back on the worklist.  The worklist
1593 		 * routine will remove us from our current
1594 		 * position and then add us back in at a later
1595 		 * position.
1596 		 */
1597 		vn_syncer_add_to_worklist(bo, syncdelay);
1598 	}
1599 	vdropl(vp);
1600 	mtx_lock(&sync_mtx);
1601 	return (0);
1602 }
1603 
1604 /*
1605  * System filesystem synchronizer daemon.
1606  */
1607 static void
1608 sched_sync(void)
1609 {
1610 	struct synclist *next;
1611 	struct synclist *slp;
1612 	struct bufobj *bo;
1613 	long starttime;
1614 	struct thread *td = FIRST_THREAD_IN_PROC(updateproc);
1615 	static int dummychan;
1616 	int last_work_seen;
1617 	int net_worklist_len;
1618 	int syncer_final_iter;
1619 	int first_printf;
1620 	int error;
1621 
1622 	mtx_lock(&Giant);
1623 	last_work_seen = 0;
1624 	syncer_final_iter = 0;
1625 	first_printf = 1;
1626 	syncer_state = SYNCER_RUNNING;
1627 	starttime = time_uptime;
1628 	td->td_pflags |= TDP_NORUNNINGBUF;
1629 
1630 	EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1631 	    SHUTDOWN_PRI_LAST);
1632 
1633 	for (;;) {
1634 		mtx_lock(&sync_mtx);
1635 		if (syncer_state == SYNCER_FINAL_DELAY &&
1636 		    syncer_final_iter == 0) {
1637 			mtx_unlock(&sync_mtx);
1638 			kthread_suspend_check(td->td_proc);
1639 			mtx_lock(&sync_mtx);
1640 		}
1641 		net_worklist_len = syncer_worklist_len - sync_vnode_count;
1642 		if (syncer_state != SYNCER_RUNNING &&
1643 		    starttime != time_uptime) {
1644 			if (first_printf) {
1645 				printf("\nSyncing disks, vnodes remaining...");
1646 				first_printf = 0;
1647 			}
1648 			printf("%d ", net_worklist_len);
1649 		}
1650 		starttime = time_uptime;
1651 
1652 		/*
1653 		 * Push files whose dirty time has expired.  Be careful
1654 		 * of interrupt race on slp queue.
1655 		 *
1656 		 * Skip over empty worklist slots when shutting down.
1657 		 */
1658 		do {
1659 			slp = &syncer_workitem_pending[syncer_delayno];
1660 			syncer_delayno += 1;
1661 			if (syncer_delayno == syncer_maxdelay)
1662 				syncer_delayno = 0;
1663 			next = &syncer_workitem_pending[syncer_delayno];
1664 			/*
1665 			 * If the worklist has wrapped since the
1666 			 * it was emptied of all but syncer vnodes,
1667 			 * switch to the FINAL_DELAY state and run
1668 			 * for one more second.
1669 			 */
1670 			if (syncer_state == SYNCER_SHUTTING_DOWN &&
1671 			    net_worklist_len == 0 &&
1672 			    last_work_seen == syncer_delayno) {
1673 				syncer_state = SYNCER_FINAL_DELAY;
1674 				syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1675 			}
1676 		} while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1677 		    syncer_worklist_len > 0);
1678 
1679 		/*
1680 		 * Keep track of the last time there was anything
1681 		 * on the worklist other than syncer vnodes.
1682 		 * Return to the SHUTTING_DOWN state if any
1683 		 * new work appears.
1684 		 */
1685 		if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1686 			last_work_seen = syncer_delayno;
1687 		if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1688 			syncer_state = SYNCER_SHUTTING_DOWN;
1689 		while ((bo = LIST_FIRST(slp)) != NULL) {
1690 			error = sync_vnode(bo, td);
1691 			if (error == 1) {
1692 				LIST_REMOVE(bo, bo_synclist);
1693 				LIST_INSERT_HEAD(next, bo, bo_synclist);
1694 				continue;
1695 			}
1696 		}
1697 		if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1698 			syncer_final_iter--;
1699 		mtx_unlock(&sync_mtx);
1700 		/*
1701 		 * The variable rushjob allows the kernel to speed up the
1702 		 * processing of the filesystem syncer process. A rushjob
1703 		 * value of N tells the filesystem syncer to process the next
1704 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1705 		 * is used by the soft update code to speed up the filesystem
1706 		 * syncer process when the incore state is getting so far
1707 		 * ahead of the disk that the kernel memory pool is being
1708 		 * threatened with exhaustion.
1709 		 */
1710 		mtx_lock(&sync_mtx);
1711 		if (rushjob > 0) {
1712 			rushjob -= 1;
1713 			mtx_unlock(&sync_mtx);
1714 			continue;
1715 		}
1716 		mtx_unlock(&sync_mtx);
1717 		/*
1718 		 * Just sleep for a short period if time between
1719 		 * iterations when shutting down to allow some I/O
1720 		 * to happen.
1721 		 *
1722 		 * If it has taken us less than a second to process the
1723 		 * current work, then wait. Otherwise start right over
1724 		 * again. We can still lose time if any single round
1725 		 * takes more than two seconds, but it does not really
1726 		 * matter as we are just trying to generally pace the
1727 		 * filesystem activity.
1728 		 */
1729 		if (syncer_state != SYNCER_RUNNING)
1730 			tsleep(&dummychan, PPAUSE, "syncfnl",
1731 			    hz / SYNCER_SHUTDOWN_SPEEDUP);
1732 		else if (time_uptime == starttime)
1733 			tsleep(&lbolt, PPAUSE, "syncer", 0);
1734 	}
1735 }
1736 
1737 /*
1738  * Request the syncer daemon to speed up its work.
1739  * We never push it to speed up more than half of its
1740  * normal turn time, otherwise it could take over the cpu.
1741  */
1742 int
1743 speedup_syncer()
1744 {
1745 	struct thread *td;
1746 	int ret = 0;
1747 
1748 	td = FIRST_THREAD_IN_PROC(updateproc);
1749 	sleepq_remove(td, &lbolt);
1750 	mtx_lock(&sync_mtx);
1751 	if (rushjob < syncdelay / 2) {
1752 		rushjob += 1;
1753 		stat_rush_requests += 1;
1754 		ret = 1;
1755 	}
1756 	mtx_unlock(&sync_mtx);
1757 	return (ret);
1758 }
1759 
1760 /*
1761  * Tell the syncer to speed up its work and run though its work
1762  * list several times, then tell it to shut down.
1763  */
1764 static void
1765 syncer_shutdown(void *arg, int howto)
1766 {
1767 	struct thread *td;
1768 
1769 	if (howto & RB_NOSYNC)
1770 		return;
1771 	td = FIRST_THREAD_IN_PROC(updateproc);
1772 	sleepq_remove(td, &lbolt);
1773 	mtx_lock(&sync_mtx);
1774 	syncer_state = SYNCER_SHUTTING_DOWN;
1775 	rushjob = 0;
1776 	mtx_unlock(&sync_mtx);
1777 	kproc_shutdown(arg, howto);
1778 }
1779 
1780 /*
1781  * Reassign a buffer from one vnode to another.
1782  * Used to assign file specific control information
1783  * (indirect blocks) to the vnode to which they belong.
1784  */
1785 void
1786 reassignbuf(struct buf *bp)
1787 {
1788 	struct vnode *vp;
1789 	struct bufobj *bo;
1790 	int delay;
1791 #ifdef INVARIANTS
1792 	struct bufv *bv;
1793 #endif
1794 
1795 	vp = bp->b_vp;
1796 	bo = bp->b_bufobj;
1797 	++reassignbufcalls;
1798 
1799 	CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1800 	    bp, bp->b_vp, bp->b_flags);
1801 	/*
1802 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1803 	 * is not fully linked in.
1804 	 */
1805 	if (bp->b_flags & B_PAGING)
1806 		panic("cannot reassign paging buffer");
1807 
1808 	/*
1809 	 * Delete from old vnode list, if on one.
1810 	 */
1811 	VI_LOCK(vp);
1812 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1813 		buf_vlist_remove(bp);
1814 	else
1815 		panic("reassignbuf: Buffer %p not on queue.", bp);
1816 	/*
1817 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1818 	 * of clean buffers.
1819 	 */
1820 	if (bp->b_flags & B_DELWRI) {
1821 		if ((bo->bo_flag & BO_ONWORKLST) == 0) {
1822 			switch (vp->v_type) {
1823 			case VDIR:
1824 				delay = dirdelay;
1825 				break;
1826 			case VCHR:
1827 				delay = metadelay;
1828 				break;
1829 			default:
1830 				delay = filedelay;
1831 			}
1832 			vn_syncer_add_to_worklist(bo, delay);
1833 		}
1834 		buf_vlist_add(bp, bo, BX_VNDIRTY);
1835 	} else {
1836 		buf_vlist_add(bp, bo, BX_VNCLEAN);
1837 
1838 		if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1839 			mtx_lock(&sync_mtx);
1840 			LIST_REMOVE(bo, bo_synclist);
1841  			syncer_worklist_len--;
1842 			mtx_unlock(&sync_mtx);
1843 			bo->bo_flag &= ~BO_ONWORKLST;
1844 		}
1845 	}
1846 #ifdef INVARIANTS
1847 	bv = &bo->bo_clean;
1848 	bp = TAILQ_FIRST(&bv->bv_hd);
1849 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1850 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1851 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
1852 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1853 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1854 	bv = &bo->bo_dirty;
1855 	bp = TAILQ_FIRST(&bv->bv_hd);
1856 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1857 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1858 	bp = TAILQ_LAST(&bv->bv_hd, buflists);
1859 	KASSERT(bp == NULL || bp->b_bufobj == bo,
1860 	    ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1861 #endif
1862 	VI_UNLOCK(vp);
1863 }
1864 
1865 /*
1866  * Increment the use and hold counts on the vnode, taking care to reference
1867  * the driver's usecount if this is a chardev.  The vholdl() will remove
1868  * the vnode from the free list if it is presently free.  Requires the
1869  * vnode interlock and returns with it held.
1870  */
1871 static void
1872 v_incr_usecount(struct vnode *vp)
1873 {
1874 
1875 	CTR3(KTR_VFS, "v_incr_usecount: vp %p holdcnt %d usecount %d\n",
1876 	    vp, vp->v_holdcnt, vp->v_usecount);
1877 	vp->v_usecount++;
1878 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1879 		dev_lock();
1880 		vp->v_rdev->si_usecount++;
1881 		dev_unlock();
1882 	}
1883 	vholdl(vp);
1884 }
1885 
1886 /*
1887  * Turn a holdcnt into a use+holdcnt such that only one call to
1888  * v_decr_usecount is needed.
1889  */
1890 static void
1891 v_upgrade_usecount(struct vnode *vp)
1892 {
1893 
1894 	CTR3(KTR_VFS, "v_upgrade_usecount: vp %p holdcnt %d usecount %d\n",
1895 	    vp, vp->v_holdcnt, vp->v_usecount);
1896 	vp->v_usecount++;
1897 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1898 		dev_lock();
1899 		vp->v_rdev->si_usecount++;
1900 		dev_unlock();
1901 	}
1902 }
1903 
1904 /*
1905  * Decrement the vnode use and hold count along with the driver's usecount
1906  * if this is a chardev.  The vdropl() below releases the vnode interlock
1907  * as it may free the vnode.
1908  */
1909 static void
1910 v_decr_usecount(struct vnode *vp)
1911 {
1912 
1913 	CTR3(KTR_VFS, "v_decr_usecount: vp %p holdcnt %d usecount %d\n",
1914 	    vp, vp->v_holdcnt, vp->v_usecount);
1915 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
1916 	VNASSERT(vp->v_usecount > 0, vp,
1917 	    ("v_decr_usecount: negative usecount"));
1918 	vp->v_usecount--;
1919 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1920 		dev_lock();
1921 		vp->v_rdev->si_usecount--;
1922 		dev_unlock();
1923 	}
1924 	vdropl(vp);
1925 }
1926 
1927 /*
1928  * Decrement only the use count and driver use count.  This is intended to
1929  * be paired with a follow on vdropl() to release the remaining hold count.
1930  * In this way we may vgone() a vnode with a 0 usecount without risk of
1931  * having it end up on a free list because the hold count is kept above 0.
1932  */
1933 static void
1934 v_decr_useonly(struct vnode *vp)
1935 {
1936 
1937 	CTR3(KTR_VFS, "v_decr_useonly: vp %p holdcnt %d usecount %d\n",
1938 	    vp, vp->v_holdcnt, vp->v_usecount);
1939 	ASSERT_VI_LOCKED(vp, __FUNCTION__);
1940 	VNASSERT(vp->v_usecount > 0, vp,
1941 	    ("v_decr_useonly: negative usecount"));
1942 	vp->v_usecount--;
1943 	if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1944 		dev_lock();
1945 		vp->v_rdev->si_usecount--;
1946 		dev_unlock();
1947 	}
1948 }
1949 
1950 /*
1951  * Grab a particular vnode from the free list, increment its
1952  * reference count and lock it. The vnode lock bit is set if the
1953  * vnode is being eliminated in vgone. The process is awakened
1954  * when the transition is completed, and an error returned to
1955  * indicate that the vnode is no longer usable (possibly having
1956  * been changed to a new filesystem type).
1957  */
1958 int
1959 vget(struct vnode *vp, int flags, struct thread *td)
1960 {
1961 	int oweinact;
1962 	int oldflags;
1963 	int error;
1964 
1965 	error = 0;
1966 	oldflags = flags;
1967 	oweinact = 0;
1968 	VFS_ASSERT_GIANT(vp->v_mount);
1969 	if ((flags & LK_INTERLOCK) == 0)
1970 		VI_LOCK(vp);
1971 	/*
1972 	 * If the inactive call was deferred because vput() was called
1973 	 * with a shared lock, we have to do it here before another thread
1974 	 * gets a reference to data that should be dead.
1975 	 */
1976 	if (vp->v_iflag & VI_OWEINACT) {
1977 		if (flags & LK_NOWAIT) {
1978 			VI_UNLOCK(vp);
1979 			return (EBUSY);
1980 		}
1981 		flags &= ~LK_TYPE_MASK;
1982 		flags |= LK_EXCLUSIVE;
1983 		oweinact = 1;
1984 	}
1985 	vholdl(vp);
1986 	if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
1987 		vdrop(vp);
1988 		return (error);
1989 	}
1990 	VI_LOCK(vp);
1991 	/* Upgrade our holdcnt to a usecount. */
1992 	v_upgrade_usecount(vp);
1993 	if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
1994 		panic("vget: vn_lock failed to return ENOENT\n");
1995 	if (oweinact) {
1996 		if (vp->v_iflag & VI_OWEINACT)
1997 			vinactive(vp, td);
1998 		VI_UNLOCK(vp);
1999 		if ((oldflags & LK_TYPE_MASK) == 0)
2000 			VOP_UNLOCK(vp, 0, td);
2001 	} else
2002 		VI_UNLOCK(vp);
2003 	return (0);
2004 }
2005 
2006 /*
2007  * Increase the reference count of a vnode.
2008  */
2009 void
2010 vref(struct vnode *vp)
2011 {
2012 
2013 	VI_LOCK(vp);
2014 	v_incr_usecount(vp);
2015 	VI_UNLOCK(vp);
2016 }
2017 
2018 /*
2019  * Return reference count of a vnode.
2020  *
2021  * The results of this call are only guaranteed when some mechanism other
2022  * than the VI lock is used to stop other processes from gaining references
2023  * to the vnode.  This may be the case if the caller holds the only reference.
2024  * This is also useful when stale data is acceptable as race conditions may
2025  * be accounted for by some other means.
2026  */
2027 int
2028 vrefcnt(struct vnode *vp)
2029 {
2030 	int usecnt;
2031 
2032 	VI_LOCK(vp);
2033 	usecnt = vp->v_usecount;
2034 	VI_UNLOCK(vp);
2035 
2036 	return (usecnt);
2037 }
2038 
2039 
2040 /*
2041  * Vnode put/release.
2042  * If count drops to zero, call inactive routine and return to freelist.
2043  */
2044 void
2045 vrele(struct vnode *vp)
2046 {
2047 	struct thread *td = curthread;	/* XXX */
2048 
2049 	KASSERT(vp != NULL, ("vrele: null vp"));
2050 	VFS_ASSERT_GIANT(vp->v_mount);
2051 
2052 	VI_LOCK(vp);
2053 
2054 	/* Skip this v_writecount check if we're going to panic below. */
2055 	VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2056 	    ("vrele: missed vn_close"));
2057 
2058 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2059 	    vp->v_usecount == 1)) {
2060 		v_decr_usecount(vp);
2061 		return;
2062 	}
2063 	if (vp->v_usecount != 1) {
2064 #ifdef DIAGNOSTIC
2065 		vprint("vrele: negative ref count", vp);
2066 #endif
2067 		VI_UNLOCK(vp);
2068 		panic("vrele: negative ref cnt");
2069 	}
2070 	/*
2071 	 * We want to hold the vnode until the inactive finishes to
2072 	 * prevent vgone() races.  We drop the use count here and the
2073 	 * hold count below when we're done.
2074 	 */
2075 	v_decr_useonly(vp);
2076 	/*
2077 	 * We must call VOP_INACTIVE with the node locked. Mark
2078 	 * as VI_DOINGINACT to avoid recursion.
2079 	 */
2080 	vp->v_iflag |= VI_OWEINACT;
2081 	if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0) {
2082 		VI_LOCK(vp);
2083 		if (vp->v_usecount > 0)
2084 			vp->v_iflag &= ~VI_OWEINACT;
2085 		if (vp->v_iflag & VI_OWEINACT)
2086 			vinactive(vp, td);
2087 		VOP_UNLOCK(vp, 0, td);
2088 	} else {
2089 		VI_LOCK(vp);
2090 		if (vp->v_usecount > 0)
2091 			vp->v_iflag &= ~VI_OWEINACT;
2092 	}
2093 	vdropl(vp);
2094 }
2095 
2096 /*
2097  * Release an already locked vnode.  This give the same effects as
2098  * unlock+vrele(), but takes less time and avoids releasing and
2099  * re-aquiring the lock (as vrele() aquires the lock internally.)
2100  */
2101 void
2102 vput(struct vnode *vp)
2103 {
2104 	struct thread *td = curthread;	/* XXX */
2105 	int error;
2106 
2107 	KASSERT(vp != NULL, ("vput: null vp"));
2108 	ASSERT_VOP_LOCKED(vp, "vput");
2109 	VFS_ASSERT_GIANT(vp->v_mount);
2110 	VI_LOCK(vp);
2111 	/* Skip this v_writecount check if we're going to panic below. */
2112 	VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2113 	    ("vput: missed vn_close"));
2114 	error = 0;
2115 
2116 	if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2117 	    vp->v_usecount == 1)) {
2118 		VOP_UNLOCK(vp, 0, td);
2119 		v_decr_usecount(vp);
2120 		return;
2121 	}
2122 
2123 	if (vp->v_usecount != 1) {
2124 #ifdef DIAGNOSTIC
2125 		vprint("vput: negative ref count", vp);
2126 #endif
2127 		panic("vput: negative ref cnt");
2128 	}
2129 	/*
2130 	 * We want to hold the vnode until the inactive finishes to
2131 	 * prevent vgone() races.  We drop the use count here and the
2132 	 * hold count below when we're done.
2133 	 */
2134 	v_decr_useonly(vp);
2135 	vp->v_iflag |= VI_OWEINACT;
2136 	if (VOP_ISLOCKED(vp, NULL) != LK_EXCLUSIVE) {
2137 		error = VOP_LOCK(vp, LK_EXCLUPGRADE|LK_INTERLOCK|LK_NOWAIT, td);
2138 		VI_LOCK(vp);
2139 		if (error) {
2140 			if (vp->v_usecount > 0)
2141 				vp->v_iflag &= ~VI_OWEINACT;
2142 			goto done;
2143 		}
2144 	}
2145 	if (vp->v_usecount > 0)
2146 		vp->v_iflag &= ~VI_OWEINACT;
2147 	if (vp->v_iflag & VI_OWEINACT)
2148 		vinactive(vp, td);
2149 	VOP_UNLOCK(vp, 0, td);
2150 done:
2151 	vdropl(vp);
2152 }
2153 
2154 /*
2155  * Somebody doesn't want the vnode recycled.
2156  */
2157 void
2158 vhold(struct vnode *vp)
2159 {
2160 
2161 	VI_LOCK(vp);
2162 	vholdl(vp);
2163 	VI_UNLOCK(vp);
2164 }
2165 
2166 void
2167 vholdl(struct vnode *vp)
2168 {
2169 
2170 	vp->v_holdcnt++;
2171 	if (VSHOULDBUSY(vp))
2172 		vbusy(vp);
2173 }
2174 
2175 /*
2176  * Note that there is one less who cares about this vnode.  vdrop() is the
2177  * opposite of vhold().
2178  */
2179 void
2180 vdrop(struct vnode *vp)
2181 {
2182 
2183 	VI_LOCK(vp);
2184 	vdropl(vp);
2185 }
2186 
2187 /*
2188  * Drop the hold count of the vnode.  If this is the last reference to
2189  * the vnode we will free it if it has been vgone'd otherwise it is
2190  * placed on the free list.
2191  */
2192 static void
2193 vdropl(struct vnode *vp)
2194 {
2195 
2196 	if (vp->v_holdcnt <= 0)
2197 		panic("vdrop: holdcnt %d", vp->v_holdcnt);
2198 	vp->v_holdcnt--;
2199 	if (vp->v_holdcnt == 0) {
2200 		if (vp->v_iflag & VI_DOOMED) {
2201 			vdestroy(vp);
2202 			return;
2203 		} else
2204 			vfree(vp);
2205 	}
2206 	VI_UNLOCK(vp);
2207 }
2208 
2209 /*
2210  * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2211  * flags.  DOINGINACT prevents us from recursing in calls to vinactive.
2212  * OWEINACT tracks whether a vnode missed a call to inactive due to a
2213  * failed lock upgrade.
2214  */
2215 static void
2216 vinactive(struct vnode *vp, struct thread *td)
2217 {
2218 
2219 	ASSERT_VOP_LOCKED(vp, "vinactive");
2220 	ASSERT_VI_LOCKED(vp, "vinactive");
2221 	VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2222 	    ("vinactive: recursed on VI_DOINGINACT"));
2223 	vp->v_iflag |= VI_DOINGINACT;
2224 	vp->v_iflag &= ~VI_OWEINACT;
2225 	VI_UNLOCK(vp);
2226 	VOP_INACTIVE(vp, td);
2227 	VI_LOCK(vp);
2228 	VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2229 	    ("vinactive: lost VI_DOINGINACT"));
2230 	vp->v_iflag &= ~VI_DOINGINACT;
2231 }
2232 
2233 /*
2234  * Remove any vnodes in the vnode table belonging to mount point mp.
2235  *
2236  * If FORCECLOSE is not specified, there should not be any active ones,
2237  * return error if any are found (nb: this is a user error, not a
2238  * system error). If FORCECLOSE is specified, detach any active vnodes
2239  * that are found.
2240  *
2241  * If WRITECLOSE is set, only flush out regular file vnodes open for
2242  * writing.
2243  *
2244  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2245  *
2246  * `rootrefs' specifies the base reference count for the root vnode
2247  * of this filesystem. The root vnode is considered busy if its
2248  * v_usecount exceeds this value. On a successful return, vflush(, td)
2249  * will call vrele() on the root vnode exactly rootrefs times.
2250  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2251  * be zero.
2252  */
2253 #ifdef DIAGNOSTIC
2254 static int busyprt = 0;		/* print out busy vnodes */
2255 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
2256 #endif
2257 
2258 int
2259 vflush( struct mount *mp, int rootrefs, int flags, struct thread *td)
2260 {
2261 	struct vnode *vp, *mvp, *rootvp = NULL;
2262 	struct vattr vattr;
2263 	int busy = 0, error;
2264 
2265 	CTR1(KTR_VFS, "vflush: mp %p", mp);
2266 	if (rootrefs > 0) {
2267 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2268 		    ("vflush: bad args"));
2269 		/*
2270 		 * Get the filesystem root vnode. We can vput() it
2271 		 * immediately, since with rootrefs > 0, it won't go away.
2272 		 */
2273 		if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp, td)) != 0)
2274 			return (error);
2275 		vput(rootvp);
2276 
2277 	}
2278 	MNT_ILOCK(mp);
2279 loop:
2280 	MNT_VNODE_FOREACH(vp, mp, mvp) {
2281 
2282 		VI_LOCK(vp);
2283 		vholdl(vp);
2284 		MNT_IUNLOCK(mp);
2285 		error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE, td);
2286 		if (error) {
2287 			vdrop(vp);
2288 			MNT_ILOCK(mp);
2289 			MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
2290 			goto loop;
2291 		}
2292 		/*
2293 		 * Skip over a vnodes marked VV_SYSTEM.
2294 		 */
2295 		if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2296 			VOP_UNLOCK(vp, 0, td);
2297 			vdrop(vp);
2298 			MNT_ILOCK(mp);
2299 			continue;
2300 		}
2301 		/*
2302 		 * If WRITECLOSE is set, flush out unlinked but still open
2303 		 * files (even if open only for reading) and regular file
2304 		 * vnodes open for writing.
2305 		 */
2306 		if (flags & WRITECLOSE) {
2307 			error = VOP_GETATTR(vp, &vattr, td->td_ucred, td);
2308 			VI_LOCK(vp);
2309 
2310 			if ((vp->v_type == VNON ||
2311 			    (error == 0 && vattr.va_nlink > 0)) &&
2312 			    (vp->v_writecount == 0 || vp->v_type != VREG)) {
2313 				VOP_UNLOCK(vp, 0, td);
2314 				vdropl(vp);
2315 				MNT_ILOCK(mp);
2316 				continue;
2317 			}
2318 		} else
2319 			VI_LOCK(vp);
2320 		/*
2321 		 * With v_usecount == 0, all we need to do is clear out the
2322 		 * vnode data structures and we are done.
2323 		 *
2324 		 * If FORCECLOSE is set, forcibly close the vnode.
2325 		 */
2326 		if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2327 			VNASSERT(vp->v_usecount == 0 ||
2328 			    (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2329 			    ("device VNODE %p is FORCECLOSED", vp));
2330 			vgonel(vp);
2331 		} else {
2332 			busy++;
2333 #ifdef DIAGNOSTIC
2334 			if (busyprt)
2335 				vprint("vflush: busy vnode", vp);
2336 #endif
2337 		}
2338 		VOP_UNLOCK(vp, 0, td);
2339 		vdropl(vp);
2340 		MNT_ILOCK(mp);
2341 	}
2342 	MNT_IUNLOCK(mp);
2343 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2344 		/*
2345 		 * If just the root vnode is busy, and if its refcount
2346 		 * is equal to `rootrefs', then go ahead and kill it.
2347 		 */
2348 		VI_LOCK(rootvp);
2349 		KASSERT(busy > 0, ("vflush: not busy"));
2350 		VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2351 		    ("vflush: usecount %d < rootrefs %d",
2352 		     rootvp->v_usecount, rootrefs));
2353 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
2354 			VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK, td);
2355 			vgone(rootvp);
2356 			VOP_UNLOCK(rootvp, 0, td);
2357 			busy = 0;
2358 		} else
2359 			VI_UNLOCK(rootvp);
2360 	}
2361 	if (busy)
2362 		return (EBUSY);
2363 	for (; rootrefs > 0; rootrefs--)
2364 		vrele(rootvp);
2365 	return (0);
2366 }
2367 
2368 /*
2369  * Recycle an unused vnode to the front of the free list.
2370  */
2371 int
2372 vrecycle(struct vnode *vp, struct thread *td)
2373 {
2374 	int recycled;
2375 
2376 	ASSERT_VOP_LOCKED(vp, "vrecycle");
2377 	recycled = 0;
2378 	VI_LOCK(vp);
2379 	if (vp->v_usecount == 0) {
2380 		recycled = 1;
2381 		vgonel(vp);
2382 	}
2383 	VI_UNLOCK(vp);
2384 	return (recycled);
2385 }
2386 
2387 /*
2388  * Eliminate all activity associated with a vnode
2389  * in preparation for reuse.
2390  */
2391 void
2392 vgone(struct vnode *vp)
2393 {
2394 	VI_LOCK(vp);
2395 	vgonel(vp);
2396 	VI_UNLOCK(vp);
2397 }
2398 
2399 /*
2400  * vgone, with the vp interlock held.
2401  */
2402 void
2403 vgonel(struct vnode *vp)
2404 {
2405 	struct thread *td;
2406 	int oweinact;
2407 	int active;
2408 	struct mount *mp;
2409 
2410 	CTR1(KTR_VFS, "vgonel: vp %p", vp);
2411 	ASSERT_VOP_LOCKED(vp, "vgonel");
2412 	ASSERT_VI_LOCKED(vp, "vgonel");
2413 	VNASSERT(vp->v_holdcnt, vp,
2414 	    ("vgonel: vp %p has no reference.", vp));
2415 	td = curthread;
2416 
2417 	/*
2418 	 * Don't vgonel if we're already doomed.
2419 	 */
2420 	if (vp->v_iflag & VI_DOOMED)
2421 		return;
2422 	vp->v_iflag |= VI_DOOMED;
2423 	/*
2424 	 * Check to see if the vnode is in use.  If so, we have to call
2425 	 * VOP_CLOSE() and VOP_INACTIVE().
2426 	 */
2427 	active = vp->v_usecount;
2428 	oweinact = (vp->v_iflag & VI_OWEINACT);
2429 	VI_UNLOCK(vp);
2430 	/*
2431 	 * Clean out any buffers associated with the vnode.
2432 	 * If the flush fails, just toss the buffers.
2433 	 */
2434 	mp = NULL;
2435 	if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2436 		(void) vn_start_secondary_write(vp, &mp, V_WAIT);
2437 	if (vinvalbuf(vp, V_SAVE, td, 0, 0) != 0)
2438 		vinvalbuf(vp, 0, td, 0, 0);
2439 
2440 	/*
2441 	 * If purging an active vnode, it must be closed and
2442 	 * deactivated before being reclaimed.
2443 	 */
2444 	if (active)
2445 		VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2446 	if (oweinact || active) {
2447 		VI_LOCK(vp);
2448 		if ((vp->v_iflag & VI_DOINGINACT) == 0)
2449 			vinactive(vp, td);
2450 		VI_UNLOCK(vp);
2451 	}
2452 	/*
2453 	 * Reclaim the vnode.
2454 	 */
2455 	if (VOP_RECLAIM(vp, td))
2456 		panic("vgone: cannot reclaim");
2457 	if (mp != NULL)
2458 		vn_finished_secondary_write(mp);
2459 	VNASSERT(vp->v_object == NULL, vp,
2460 	    ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2461 	/*
2462 	 * Delete from old mount point vnode list.
2463 	 */
2464 	delmntque(vp);
2465 	cache_purge(vp);
2466 	/*
2467 	 * Done with purge, reset to the standard lock and invalidate
2468 	 * the vnode.
2469 	 */
2470 	VI_LOCK(vp);
2471 	vp->v_vnlock = &vp->v_lock;
2472 	vp->v_op = &dead_vnodeops;
2473 	vp->v_tag = "none";
2474 	vp->v_type = VBAD;
2475 }
2476 
2477 /*
2478  * Calculate the total number of references to a special device.
2479  */
2480 int
2481 vcount(struct vnode *vp)
2482 {
2483 	int count;
2484 
2485 	dev_lock();
2486 	count = vp->v_rdev->si_usecount;
2487 	dev_unlock();
2488 	return (count);
2489 }
2490 
2491 /*
2492  * Same as above, but using the struct cdev *as argument
2493  */
2494 int
2495 count_dev(struct cdev *dev)
2496 {
2497 	int count;
2498 
2499 	dev_lock();
2500 	count = dev->si_usecount;
2501 	dev_unlock();
2502 	return(count);
2503 }
2504 
2505 /*
2506  * Print out a description of a vnode.
2507  */
2508 static char *typename[] =
2509 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2510  "VMARKER"};
2511 
2512 void
2513 vn_printf(struct vnode *vp, const char *fmt, ...)
2514 {
2515 	va_list ap;
2516 	char buf[96];
2517 
2518 	va_start(ap, fmt);
2519 	vprintf(fmt, ap);
2520 	va_end(ap);
2521 	printf("%p: ", (void *)vp);
2522 	printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2523 	printf("    usecount %d, writecount %d, refcount %d mountedhere %p\n",
2524 	    vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2525 	buf[0] = '\0';
2526 	buf[1] = '\0';
2527 	if (vp->v_vflag & VV_ROOT)
2528 		strcat(buf, "|VV_ROOT");
2529 	if (vp->v_vflag & VV_TEXT)
2530 		strcat(buf, "|VV_TEXT");
2531 	if (vp->v_vflag & VV_SYSTEM)
2532 		strcat(buf, "|VV_SYSTEM");
2533 	if (vp->v_iflag & VI_DOOMED)
2534 		strcat(buf, "|VI_DOOMED");
2535 	if (vp->v_iflag & VI_FREE)
2536 		strcat(buf, "|VI_FREE");
2537 	printf("    flags (%s)\n", buf + 1);
2538 	if (mtx_owned(VI_MTX(vp)))
2539 		printf(" VI_LOCKed");
2540 	if (vp->v_object != NULL)
2541 		printf("    v_object %p ref %d pages %d\n",
2542 		    vp->v_object, vp->v_object->ref_count,
2543 		    vp->v_object->resident_page_count);
2544 	printf("    ");
2545 	lockmgr_printinfo(vp->v_vnlock);
2546 	printf("\n");
2547 	if (vp->v_data != NULL)
2548 		VOP_PRINT(vp);
2549 }
2550 
2551 #ifdef DDB
2552 /*
2553  * List all of the locked vnodes in the system.
2554  * Called when debugging the kernel.
2555  */
2556 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2557 {
2558 	struct mount *mp, *nmp;
2559 	struct vnode *vp;
2560 
2561 	/*
2562 	 * Note: because this is DDB, we can't obey the locking semantics
2563 	 * for these structures, which means we could catch an inconsistent
2564 	 * state and dereference a nasty pointer.  Not much to be done
2565 	 * about that.
2566 	 */
2567 	printf("Locked vnodes\n");
2568 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2569 		nmp = TAILQ_NEXT(mp, mnt_list);
2570 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2571 			if (vp->v_type != VMARKER && VOP_ISLOCKED(vp, NULL))
2572 				vprint("", vp);
2573 		}
2574 		nmp = TAILQ_NEXT(mp, mnt_list);
2575 	}
2576 }
2577 
2578 /*
2579  * Show details about the given vnode.
2580  */
2581 DB_SHOW_COMMAND(vnode, db_show_vnode)
2582 {
2583 	struct vnode *vp;
2584 
2585 	if (!have_addr)
2586 		return;
2587 	vp = (struct vnode *)addr;
2588 	vn_printf(vp, "vnode ");
2589 }
2590 #endif	/* DDB */
2591 
2592 /*
2593  * Fill in a struct xvfsconf based on a struct vfsconf.
2594  */
2595 static void
2596 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2597 {
2598 
2599 	strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2600 	xvfsp->vfc_typenum = vfsp->vfc_typenum;
2601 	xvfsp->vfc_refcount = vfsp->vfc_refcount;
2602 	xvfsp->vfc_flags = vfsp->vfc_flags;
2603 	/*
2604 	 * These are unused in userland, we keep them
2605 	 * to not break binary compatibility.
2606 	 */
2607 	xvfsp->vfc_vfsops = NULL;
2608 	xvfsp->vfc_next = NULL;
2609 }
2610 
2611 /*
2612  * Top level filesystem related information gathering.
2613  */
2614 static int
2615 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2616 {
2617 	struct vfsconf *vfsp;
2618 	struct xvfsconf xvfsp;
2619 	int error;
2620 
2621 	error = 0;
2622 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2623 		bzero(&xvfsp, sizeof(xvfsp));
2624 		vfsconf2x(vfsp, &xvfsp);
2625 		error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
2626 		if (error)
2627 			break;
2628 	}
2629 	return (error);
2630 }
2631 
2632 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
2633     "S,xvfsconf", "List of all configured filesystems");
2634 
2635 #ifndef BURN_BRIDGES
2636 static int	sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2637 
2638 static int
2639 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2640 {
2641 	int *name = (int *)arg1 - 1;	/* XXX */
2642 	u_int namelen = arg2 + 1;	/* XXX */
2643 	struct vfsconf *vfsp;
2644 	struct xvfsconf xvfsp;
2645 
2646 	printf("WARNING: userland calling deprecated sysctl, "
2647 	    "please rebuild world\n");
2648 
2649 #if 1 || defined(COMPAT_PRELITE2)
2650 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2651 	if (namelen == 1)
2652 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2653 #endif
2654 
2655 	switch (name[1]) {
2656 	case VFS_MAXTYPENUM:
2657 		if (namelen != 2)
2658 			return (ENOTDIR);
2659 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2660 	case VFS_CONF:
2661 		if (namelen != 3)
2662 			return (ENOTDIR);	/* overloaded */
2663 		TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
2664 			if (vfsp->vfc_typenum == name[2])
2665 				break;
2666 		if (vfsp == NULL)
2667 			return (EOPNOTSUPP);
2668 		bzero(&xvfsp, sizeof(xvfsp));
2669 		vfsconf2x(vfsp, &xvfsp);
2670 		return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
2671 	}
2672 	return (EOPNOTSUPP);
2673 }
2674 
2675 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
2676 	vfs_sysctl, "Generic filesystem");
2677 
2678 #if 1 || defined(COMPAT_PRELITE2)
2679 
2680 static int
2681 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2682 {
2683 	int error;
2684 	struct vfsconf *vfsp;
2685 	struct ovfsconf ovfs;
2686 
2687 	TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2688 		bzero(&ovfs, sizeof(ovfs));
2689 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
2690 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
2691 		ovfs.vfc_index = vfsp->vfc_typenum;
2692 		ovfs.vfc_refcount = vfsp->vfc_refcount;
2693 		ovfs.vfc_flags = vfsp->vfc_flags;
2694 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2695 		if (error)
2696 			return error;
2697 	}
2698 	return 0;
2699 }
2700 
2701 #endif /* 1 || COMPAT_PRELITE2 */
2702 #endif /* !BURN_BRIDGES */
2703 
2704 #define KINFO_VNODESLOP		10
2705 #ifdef notyet
2706 /*
2707  * Dump vnode list (via sysctl).
2708  */
2709 /* ARGSUSED */
2710 static int
2711 sysctl_vnode(SYSCTL_HANDLER_ARGS)
2712 {
2713 	struct xvnode *xvn;
2714 	struct thread *td = req->td;
2715 	struct mount *mp;
2716 	struct vnode *vp;
2717 	int error, len, n;
2718 
2719 	/*
2720 	 * Stale numvnodes access is not fatal here.
2721 	 */
2722 	req->lock = 0;
2723 	len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
2724 	if (!req->oldptr)
2725 		/* Make an estimate */
2726 		return (SYSCTL_OUT(req, 0, len));
2727 
2728 	error = sysctl_wire_old_buffer(req, 0);
2729 	if (error != 0)
2730 		return (error);
2731 	xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
2732 	n = 0;
2733 	mtx_lock(&mountlist_mtx);
2734 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2735 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td))
2736 			continue;
2737 		MNT_ILOCK(mp);
2738 		TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2739 			if (n == len)
2740 				break;
2741 			vref(vp);
2742 			xvn[n].xv_size = sizeof *xvn;
2743 			xvn[n].xv_vnode = vp;
2744 			xvn[n].xv_id = 0;	/* XXX compat */
2745 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
2746 			XV_COPY(usecount);
2747 			XV_COPY(writecount);
2748 			XV_COPY(holdcnt);
2749 			XV_COPY(mount);
2750 			XV_COPY(numoutput);
2751 			XV_COPY(type);
2752 #undef XV_COPY
2753 			xvn[n].xv_flag = vp->v_vflag;
2754 
2755 			switch (vp->v_type) {
2756 			case VREG:
2757 			case VDIR:
2758 			case VLNK:
2759 				break;
2760 			case VBLK:
2761 			case VCHR:
2762 				if (vp->v_rdev == NULL) {
2763 					vrele(vp);
2764 					continue;
2765 				}
2766 				xvn[n].xv_dev = dev2udev(vp->v_rdev);
2767 				break;
2768 			case VSOCK:
2769 				xvn[n].xv_socket = vp->v_socket;
2770 				break;
2771 			case VFIFO:
2772 				xvn[n].xv_fifo = vp->v_fifoinfo;
2773 				break;
2774 			case VNON:
2775 			case VBAD:
2776 			default:
2777 				/* shouldn't happen? */
2778 				vrele(vp);
2779 				continue;
2780 			}
2781 			vrele(vp);
2782 			++n;
2783 		}
2784 		MNT_IUNLOCK(mp);
2785 		mtx_lock(&mountlist_mtx);
2786 		vfs_unbusy(mp, td);
2787 		if (n == len)
2788 			break;
2789 	}
2790 	mtx_unlock(&mountlist_mtx);
2791 
2792 	error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
2793 	free(xvn, M_TEMP);
2794 	return (error);
2795 }
2796 
2797 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2798 	0, 0, sysctl_vnode, "S,xvnode", "");
2799 #endif
2800 
2801 /*
2802  * Unmount all filesystems. The list is traversed in reverse order
2803  * of mounting to avoid dependencies.
2804  */
2805 void
2806 vfs_unmountall(void)
2807 {
2808 	struct mount *mp;
2809 	struct thread *td;
2810 	int error;
2811 
2812 	KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
2813 	td = curthread;
2814 	/*
2815 	 * Since this only runs when rebooting, it is not interlocked.
2816 	 */
2817 	while(!TAILQ_EMPTY(&mountlist)) {
2818 		mp = TAILQ_LAST(&mountlist, mntlist);
2819 		error = dounmount(mp, MNT_FORCE, td);
2820 		if (error) {
2821 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
2822 			/*
2823 			 * XXX: Due to the way in which we mount the root
2824 			 * file system off of devfs, devfs will generate a
2825 			 * "busy" warning when we try to unmount it before
2826 			 * the root.  Don't print a warning as a result in
2827 			 * order to avoid false positive errors that may
2828 			 * cause needless upset.
2829 			 */
2830 			if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
2831 				printf("unmount of %s failed (",
2832 				    mp->mnt_stat.f_mntonname);
2833 				if (error == EBUSY)
2834 					printf("BUSY)\n");
2835 				else
2836 					printf("%d)\n", error);
2837 			}
2838 		} else {
2839 			/* The unmount has removed mp from the mountlist */
2840 		}
2841 	}
2842 }
2843 
2844 /*
2845  * perform msync on all vnodes under a mount point
2846  * the mount point must be locked.
2847  */
2848 void
2849 vfs_msync(struct mount *mp, int flags)
2850 {
2851 	struct vnode *vp, *mvp;
2852 	struct vm_object *obj;
2853 
2854 	MNT_ILOCK(mp);
2855 	MNT_VNODE_FOREACH(vp, mp, mvp) {
2856 		VI_LOCK(vp);
2857 		if ((vp->v_iflag & VI_OBJDIRTY) &&
2858 		    (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
2859 			MNT_IUNLOCK(mp);
2860 			if (!vget(vp,
2861 			    LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
2862 			    curthread)) {
2863 				if (vp->v_vflag & VV_NOSYNC) {	/* unlinked */
2864 					vput(vp);
2865 					MNT_ILOCK(mp);
2866 					continue;
2867 				}
2868 
2869 				obj = vp->v_object;
2870 				if (obj != NULL) {
2871 					VM_OBJECT_LOCK(obj);
2872 					vm_object_page_clean(obj, 0, 0,
2873 					    flags == MNT_WAIT ?
2874 					    OBJPC_SYNC : OBJPC_NOSYNC);
2875 					VM_OBJECT_UNLOCK(obj);
2876 				}
2877 				vput(vp);
2878 			}
2879 			MNT_ILOCK(mp);
2880 		} else
2881 			VI_UNLOCK(vp);
2882 	}
2883 	MNT_IUNLOCK(mp);
2884 }
2885 
2886 /*
2887  * Mark a vnode as free, putting it up for recycling.
2888  */
2889 static void
2890 vfree(struct vnode *vp)
2891 {
2892 
2893 	CTR1(KTR_VFS, "vfree vp %p", vp);
2894 	ASSERT_VI_LOCKED(vp, "vfree");
2895 	mtx_lock(&vnode_free_list_mtx);
2896 	VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed."));
2897 	VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free"));
2898 	VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't"));
2899 	VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp,
2900 	    ("vfree: Freeing doomed vnode"));
2901 	if (vp->v_iflag & VI_AGE) {
2902 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2903 	} else {
2904 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2905 	}
2906 	freevnodes++;
2907 	vp->v_iflag &= ~VI_AGE;
2908 	vp->v_iflag |= VI_FREE;
2909 	mtx_unlock(&vnode_free_list_mtx);
2910 }
2911 
2912 /*
2913  * Opposite of vfree() - mark a vnode as in use.
2914  */
2915 static void
2916 vbusy(struct vnode *vp)
2917 {
2918 	CTR1(KTR_VFS, "vbusy vp %p", vp);
2919 	ASSERT_VI_LOCKED(vp, "vbusy");
2920 	VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
2921 	VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed."));
2922 
2923 	mtx_lock(&vnode_free_list_mtx);
2924 	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2925 	freevnodes--;
2926 	vp->v_iflag &= ~(VI_FREE|VI_AGE);
2927 	mtx_unlock(&vnode_free_list_mtx);
2928 }
2929 
2930 /*
2931  * Initalize per-vnode helper structure to hold poll-related state.
2932  */
2933 void
2934 v_addpollinfo(struct vnode *vp)
2935 {
2936 	struct vpollinfo *vi;
2937 
2938 	vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
2939 	if (vp->v_pollinfo != NULL) {
2940 		uma_zfree(vnodepoll_zone, vi);
2941 		return;
2942 	}
2943 	vp->v_pollinfo = vi;
2944 	mtx_init(&vp->v_pollinfo->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
2945 	knlist_init(&vp->v_pollinfo->vpi_selinfo.si_note, vp, vfs_knllock,
2946 	    vfs_knlunlock, vfs_knllocked);
2947 }
2948 
2949 /*
2950  * Record a process's interest in events which might happen to
2951  * a vnode.  Because poll uses the historic select-style interface
2952  * internally, this routine serves as both the ``check for any
2953  * pending events'' and the ``record my interest in future events''
2954  * functions.  (These are done together, while the lock is held,
2955  * to avoid race conditions.)
2956  */
2957 int
2958 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
2959 {
2960 
2961 	if (vp->v_pollinfo == NULL)
2962 		v_addpollinfo(vp);
2963 	mtx_lock(&vp->v_pollinfo->vpi_lock);
2964 	if (vp->v_pollinfo->vpi_revents & events) {
2965 		/*
2966 		 * This leaves events we are not interested
2967 		 * in available for the other process which
2968 		 * which presumably had requested them
2969 		 * (otherwise they would never have been
2970 		 * recorded).
2971 		 */
2972 		events &= vp->v_pollinfo->vpi_revents;
2973 		vp->v_pollinfo->vpi_revents &= ~events;
2974 
2975 		mtx_unlock(&vp->v_pollinfo->vpi_lock);
2976 		return events;
2977 	}
2978 	vp->v_pollinfo->vpi_events |= events;
2979 	selrecord(td, &vp->v_pollinfo->vpi_selinfo);
2980 	mtx_unlock(&vp->v_pollinfo->vpi_lock);
2981 	return 0;
2982 }
2983 
2984 /*
2985  * Routine to create and manage a filesystem syncer vnode.
2986  */
2987 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
2988 static int	sync_fsync(struct  vop_fsync_args *);
2989 static int	sync_inactive(struct  vop_inactive_args *);
2990 static int	sync_reclaim(struct  vop_reclaim_args *);
2991 
2992 static struct vop_vector sync_vnodeops = {
2993 	.vop_bypass =	VOP_EOPNOTSUPP,
2994 	.vop_close =	sync_close,		/* close */
2995 	.vop_fsync =	sync_fsync,		/* fsync */
2996 	.vop_inactive =	sync_inactive,	/* inactive */
2997 	.vop_reclaim =	sync_reclaim,	/* reclaim */
2998 	.vop_lock =	vop_stdlock,	/* lock */
2999 	.vop_unlock =	vop_stdunlock,	/* unlock */
3000 	.vop_islocked =	vop_stdislocked,	/* islocked */
3001 };
3002 
3003 /*
3004  * Create a new filesystem syncer vnode for the specified mount point.
3005  */
3006 int
3007 vfs_allocate_syncvnode(struct mount *mp)
3008 {
3009 	struct vnode *vp;
3010 	static long start, incr, next;
3011 	int error;
3012 
3013 	/* Allocate a new vnode */
3014 	if ((error = getnewvnode("syncer", mp, &sync_vnodeops, &vp)) != 0) {
3015 		mp->mnt_syncer = NULL;
3016 		return (error);
3017 	}
3018 	vp->v_type = VNON;
3019 	/*
3020 	 * Place the vnode onto the syncer worklist. We attempt to
3021 	 * scatter them about on the list so that they will go off
3022 	 * at evenly distributed times even if all the filesystems
3023 	 * are mounted at once.
3024 	 */
3025 	next += incr;
3026 	if (next == 0 || next > syncer_maxdelay) {
3027 		start /= 2;
3028 		incr /= 2;
3029 		if (start == 0) {
3030 			start = syncer_maxdelay / 2;
3031 			incr = syncer_maxdelay;
3032 		}
3033 		next = start;
3034 	}
3035 	VI_LOCK(vp);
3036 	vn_syncer_add_to_worklist(&vp->v_bufobj,
3037 	    syncdelay > 0 ? next % syncdelay : 0);
3038 	/* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3039 	mtx_lock(&sync_mtx);
3040 	sync_vnode_count++;
3041 	mtx_unlock(&sync_mtx);
3042 	VI_UNLOCK(vp);
3043 	mp->mnt_syncer = vp;
3044 	return (0);
3045 }
3046 
3047 /*
3048  * Do a lazy sync of the filesystem.
3049  */
3050 static int
3051 sync_fsync(struct vop_fsync_args *ap)
3052 {
3053 	struct vnode *syncvp = ap->a_vp;
3054 	struct mount *mp = syncvp->v_mount;
3055 	struct thread *td = ap->a_td;
3056 	int error;
3057 	struct bufobj *bo;
3058 
3059 	/*
3060 	 * We only need to do something if this is a lazy evaluation.
3061 	 */
3062 	if (ap->a_waitfor != MNT_LAZY)
3063 		return (0);
3064 
3065 	/*
3066 	 * Move ourselves to the back of the sync list.
3067 	 */
3068 	bo = &syncvp->v_bufobj;
3069 	BO_LOCK(bo);
3070 	vn_syncer_add_to_worklist(bo, syncdelay);
3071 	BO_UNLOCK(bo);
3072 
3073 	/*
3074 	 * Walk the list of vnodes pushing all that are dirty and
3075 	 * not already on the sync list.
3076 	 */
3077 	mtx_lock(&mountlist_mtx);
3078 	if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) {
3079 		mtx_unlock(&mountlist_mtx);
3080 		return (0);
3081 	}
3082 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3083 		vfs_unbusy(mp, td);
3084 		return (0);
3085 	}
3086 	MNT_ILOCK(mp);
3087 	mp->mnt_noasync++;
3088 	mp->mnt_kern_flag &= ~MNTK_ASYNC;
3089 	MNT_IUNLOCK(mp);
3090 	vfs_msync(mp, MNT_NOWAIT);
3091 	error = VFS_SYNC(mp, MNT_LAZY, td);
3092 	MNT_ILOCK(mp);
3093 	mp->mnt_noasync--;
3094 	if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
3095 		mp->mnt_kern_flag |= MNTK_ASYNC;
3096 	MNT_IUNLOCK(mp);
3097 	vn_finished_write(mp);
3098 	vfs_unbusy(mp, td);
3099 	return (error);
3100 }
3101 
3102 /*
3103  * The syncer vnode is no referenced.
3104  */
3105 static int
3106 sync_inactive(struct vop_inactive_args *ap)
3107 {
3108 
3109 	vgone(ap->a_vp);
3110 	return (0);
3111 }
3112 
3113 /*
3114  * The syncer vnode is no longer needed and is being decommissioned.
3115  *
3116  * Modifications to the worklist must be protected by sync_mtx.
3117  */
3118 static int
3119 sync_reclaim(struct vop_reclaim_args *ap)
3120 {
3121 	struct vnode *vp = ap->a_vp;
3122 	struct bufobj *bo;
3123 
3124 	VI_LOCK(vp);
3125 	bo = &vp->v_bufobj;
3126 	vp->v_mount->mnt_syncer = NULL;
3127 	if (bo->bo_flag & BO_ONWORKLST) {
3128 		mtx_lock(&sync_mtx);
3129 		LIST_REMOVE(bo, bo_synclist);
3130  		syncer_worklist_len--;
3131 		sync_vnode_count--;
3132 		mtx_unlock(&sync_mtx);
3133 		bo->bo_flag &= ~BO_ONWORKLST;
3134 	}
3135 	VI_UNLOCK(vp);
3136 
3137 	return (0);
3138 }
3139 
3140 /*
3141  * Check if vnode represents a disk device
3142  */
3143 int
3144 vn_isdisk(struct vnode *vp, int *errp)
3145 {
3146 	int error;
3147 
3148 	error = 0;
3149 	dev_lock();
3150 	if (vp->v_type != VCHR)
3151 		error = ENOTBLK;
3152 	else if (vp->v_rdev == NULL)
3153 		error = ENXIO;
3154 	else if (vp->v_rdev->si_devsw == NULL)
3155 		error = ENXIO;
3156 	else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3157 		error = ENOTBLK;
3158 	dev_unlock();
3159 	if (errp != NULL)
3160 		*errp = error;
3161 	return (error == 0);
3162 }
3163 
3164 /*
3165  * Common filesystem object access control check routine.  Accepts a
3166  * vnode's type, "mode", uid and gid, requested access mode, credentials,
3167  * and optional call-by-reference privused argument allowing vaccess()
3168  * to indicate to the caller whether privilege was used to satisfy the
3169  * request (obsoleted).  Returns 0 on success, or an errno on failure.
3170  *
3171  * The ifdef'd CAPABILITIES version is here for reference, but is not
3172  * actually used.
3173  */
3174 int
3175 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3176     mode_t acc_mode, struct ucred *cred, int *privused)
3177 {
3178 	mode_t dac_granted;
3179 #ifdef CAPABILITIES
3180 	mode_t cap_granted;
3181 #endif
3182 
3183 	/*
3184 	 * Look for a normal, non-privileged way to access the file/directory
3185 	 * as requested.  If it exists, go with that.
3186 	 */
3187 
3188 	if (privused != NULL)
3189 		*privused = 0;
3190 
3191 	dac_granted = 0;
3192 
3193 	/* Check the owner. */
3194 	if (cred->cr_uid == file_uid) {
3195 		dac_granted |= VADMIN;
3196 		if (file_mode & S_IXUSR)
3197 			dac_granted |= VEXEC;
3198 		if (file_mode & S_IRUSR)
3199 			dac_granted |= VREAD;
3200 		if (file_mode & S_IWUSR)
3201 			dac_granted |= (VWRITE | VAPPEND);
3202 
3203 		if ((acc_mode & dac_granted) == acc_mode)
3204 			return (0);
3205 
3206 		goto privcheck;
3207 	}
3208 
3209 	/* Otherwise, check the groups (first match) */
3210 	if (groupmember(file_gid, cred)) {
3211 		if (file_mode & S_IXGRP)
3212 			dac_granted |= VEXEC;
3213 		if (file_mode & S_IRGRP)
3214 			dac_granted |= VREAD;
3215 		if (file_mode & S_IWGRP)
3216 			dac_granted |= (VWRITE | VAPPEND);
3217 
3218 		if ((acc_mode & dac_granted) == acc_mode)
3219 			return (0);
3220 
3221 		goto privcheck;
3222 	}
3223 
3224 	/* Otherwise, check everyone else. */
3225 	if (file_mode & S_IXOTH)
3226 		dac_granted |= VEXEC;
3227 	if (file_mode & S_IROTH)
3228 		dac_granted |= VREAD;
3229 	if (file_mode & S_IWOTH)
3230 		dac_granted |= (VWRITE | VAPPEND);
3231 	if ((acc_mode & dac_granted) == acc_mode)
3232 		return (0);
3233 
3234 privcheck:
3235 	if (!suser_cred(cred, SUSER_ALLOWJAIL)) {
3236 		/* XXX audit: privilege used */
3237 		if (privused != NULL)
3238 			*privused = 1;
3239 		return (0);
3240 	}
3241 
3242 #ifdef CAPABILITIES
3243 	/*
3244 	 * Build a capability mask to determine if the set of capabilities
3245 	 * satisfies the requirements when combined with the granted mask
3246 	 * from above.  For each capability, if the capability is required,
3247 	 * bitwise or the request type onto the cap_granted mask.
3248 	 *
3249 	 * Note: This is never actually used, but is here for reference
3250 	 * purposes.
3251 	 */
3252 	cap_granted = 0;
3253 
3254 	if (type == VDIR) {
3255 		/*
3256 		 * For directories, use CAP_DAC_READ_SEARCH to satisfy
3257 		 * VEXEC requests, instead of CAP_DAC_EXECUTE.
3258 		 */
3259 		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3260 		    !cap_check(cred, NULL, CAP_DAC_READ_SEARCH,
3261 		    SUSER_ALLOWJAIL))
3262 			cap_granted |= VEXEC;
3263 	} else {
3264 		if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3265 		    !cap_check(cred, NULL, CAP_DAC_EXECUTE, SUSER_ALLOWJAIL))
3266 			cap_granted |= VEXEC;
3267 	}
3268 
3269 	if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
3270 	    !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, SUSER_ALLOWJAIL))
3271 		cap_granted |= VREAD;
3272 
3273 	if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3274 	    !cap_check(cred, NULL, CAP_DAC_WRITE, SUSER_ALLOWJAIL))
3275 		cap_granted |= (VWRITE | VAPPEND);
3276 
3277 	if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3278 	    !cap_check(cred, NULL, CAP_FOWNER, SUSER_ALLOWJAIL))
3279 		cap_granted |= VADMIN;
3280 
3281 	if ((acc_mode & (cap_granted | dac_granted)) == acc_mode) {
3282 		/* XXX audit: privilege used */
3283 		if (privused != NULL)
3284 			*privused = 1;
3285 		return (0);
3286 	}
3287 #endif
3288 
3289 	return ((acc_mode & VADMIN) ? EPERM : EACCES);
3290 }
3291 
3292 /*
3293  * Credential check based on process requesting service, and per-attribute
3294  * permissions.
3295  */
3296 int
3297 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3298     struct thread *td, int access)
3299 {
3300 
3301 	/*
3302 	 * Kernel-invoked always succeeds.
3303 	 */
3304 	if (cred == NOCRED)
3305 		return (0);
3306 
3307 	/*
3308 	 * Do not allow privileged processes in jail to directly
3309 	 * manipulate system attributes.
3310 	 *
3311 	 * XXX What capability should apply here?
3312 	 * Probably CAP_SYS_SETFFLAG.
3313 	 */
3314 	switch (attrnamespace) {
3315 	case EXTATTR_NAMESPACE_SYSTEM:
3316 		/* Potentially should be: return (EPERM); */
3317 		return (suser_cred(cred, 0));
3318 	case EXTATTR_NAMESPACE_USER:
3319 		return (VOP_ACCESS(vp, access, cred, td));
3320 	default:
3321 		return (EPERM);
3322 	}
3323 }
3324 
3325 #ifdef DEBUG_VFS_LOCKS
3326 /*
3327  * This only exists to supress warnings from unlocked specfs accesses.  It is
3328  * no longer ok to have an unlocked VFS.
3329  */
3330 #define	IGNORE_LOCK(vp) ((vp)->v_type == VCHR || (vp)->v_type == VBAD)
3331 
3332 int vfs_badlock_ddb = 1;	/* Drop into debugger on violation. */
3333 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, "");
3334 
3335 int vfs_badlock_mutex = 1;	/* Check for interlock across VOPs. */
3336 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, "");
3337 
3338 int vfs_badlock_print = 1;	/* Print lock violations. */
3339 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, "");
3340 
3341 #ifdef KDB
3342 int vfs_badlock_backtrace = 1;	/* Print backtrace at lock violations. */
3343 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, "");
3344 #endif
3345 
3346 static void
3347 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3348 {
3349 
3350 #ifdef KDB
3351 	if (vfs_badlock_backtrace)
3352 		kdb_backtrace();
3353 #endif
3354 	if (vfs_badlock_print)
3355 		printf("%s: %p %s\n", str, (void *)vp, msg);
3356 	if (vfs_badlock_ddb)
3357 		kdb_enter("lock violation");
3358 }
3359 
3360 void
3361 assert_vi_locked(struct vnode *vp, const char *str)
3362 {
3363 
3364 	if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3365 		vfs_badlock("interlock is not locked but should be", str, vp);
3366 }
3367 
3368 void
3369 assert_vi_unlocked(struct vnode *vp, const char *str)
3370 {
3371 
3372 	if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3373 		vfs_badlock("interlock is locked but should not be", str, vp);
3374 }
3375 
3376 void
3377 assert_vop_locked(struct vnode *vp, const char *str)
3378 {
3379 
3380 	if (vp && !IGNORE_LOCK(vp) && VOP_ISLOCKED(vp, NULL) == 0)
3381 		vfs_badlock("is not locked but should be", str, vp);
3382 }
3383 
3384 void
3385 assert_vop_unlocked(struct vnode *vp, const char *str)
3386 {
3387 
3388 	if (vp && !IGNORE_LOCK(vp) &&
3389 	    VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE)
3390 		vfs_badlock("is locked but should not be", str, vp);
3391 }
3392 
3393 void
3394 assert_vop_elocked(struct vnode *vp, const char *str)
3395 {
3396 
3397 	if (vp && !IGNORE_LOCK(vp) &&
3398 	    VOP_ISLOCKED(vp, curthread) != LK_EXCLUSIVE)
3399 		vfs_badlock("is not exclusive locked but should be", str, vp);
3400 }
3401 
3402 #if 0
3403 void
3404 assert_vop_elocked_other(struct vnode *vp, const char *str)
3405 {
3406 
3407 	if (vp && !IGNORE_LOCK(vp) &&
3408 	    VOP_ISLOCKED(vp, curthread) != LK_EXCLOTHER)
3409 		vfs_badlock("is not exclusive locked by another thread",
3410 		    str, vp);
3411 }
3412 
3413 void
3414 assert_vop_slocked(struct vnode *vp, const char *str)
3415 {
3416 
3417 	if (vp && !IGNORE_LOCK(vp) &&
3418 	    VOP_ISLOCKED(vp, curthread) != LK_SHARED)
3419 		vfs_badlock("is not locked shared but should be", str, vp);
3420 }
3421 #endif /* 0 */
3422 #endif /* DEBUG_VFS_LOCKS */
3423 
3424 void
3425 vop_rename_pre(void *ap)
3426 {
3427 	struct vop_rename_args *a = ap;
3428 
3429 #ifdef DEBUG_VFS_LOCKS
3430 	if (a->a_tvp)
3431 		ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
3432 	ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
3433 	ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
3434 	ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
3435 
3436 	/* Check the source (from). */
3437 	if (a->a_tdvp != a->a_fdvp && a->a_tvp != a->a_fdvp)
3438 		ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3439 	if (a->a_tvp != a->a_fvp)
3440 		ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: tvp locked");
3441 
3442 	/* Check the target. */
3443 	if (a->a_tvp)
3444 		ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3445 	ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3446 #endif
3447 	if (a->a_tdvp != a->a_fdvp)
3448 		vhold(a->a_fdvp);
3449 	if (a->a_tvp != a->a_fvp)
3450 		vhold(a->a_fvp);
3451 	vhold(a->a_tdvp);
3452 	if (a->a_tvp)
3453 		vhold(a->a_tvp);
3454 }
3455 
3456 void
3457 vop_strategy_pre(void *ap)
3458 {
3459 #ifdef DEBUG_VFS_LOCKS
3460 	struct vop_strategy_args *a;
3461 	struct buf *bp;
3462 
3463 	a = ap;
3464 	bp = a->a_bp;
3465 
3466 	/*
3467 	 * Cluster ops lock their component buffers but not the IO container.
3468 	 */
3469 	if ((bp->b_flags & B_CLUSTER) != 0)
3470 		return;
3471 
3472 	if (BUF_REFCNT(bp) < 1) {
3473 		if (vfs_badlock_print)
3474 			printf(
3475 			    "VOP_STRATEGY: bp is not locked but should be\n");
3476 		if (vfs_badlock_ddb)
3477 			kdb_enter("lock violation");
3478 	}
3479 #endif
3480 }
3481 
3482 void
3483 vop_lookup_pre(void *ap)
3484 {
3485 #ifdef DEBUG_VFS_LOCKS
3486 	struct vop_lookup_args *a;
3487 	struct vnode *dvp;
3488 
3489 	a = ap;
3490 	dvp = a->a_dvp;
3491 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3492 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3493 #endif
3494 }
3495 
3496 void
3497 vop_lookup_post(void *ap, int rc)
3498 {
3499 #ifdef DEBUG_VFS_LOCKS
3500 	struct vop_lookup_args *a;
3501 	struct vnode *dvp;
3502 	struct vnode *vp;
3503 
3504 	a = ap;
3505 	dvp = a->a_dvp;
3506 	vp = *(a->a_vpp);
3507 
3508 	ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3509 	ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3510 
3511 	if (!rc)
3512 		ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
3513 #endif
3514 }
3515 
3516 void
3517 vop_lock_pre(void *ap)
3518 {
3519 #ifdef DEBUG_VFS_LOCKS
3520 	struct vop_lock_args *a = ap;
3521 
3522 	if ((a->a_flags & LK_INTERLOCK) == 0)
3523 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3524 	else
3525 		ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
3526 #endif
3527 }
3528 
3529 void
3530 vop_lock_post(void *ap, int rc)
3531 {
3532 #ifdef DEBUG_VFS_LOCKS
3533 	struct vop_lock_args *a = ap;
3534 
3535 	ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3536 	if (rc == 0)
3537 		ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
3538 #endif
3539 }
3540 
3541 void
3542 vop_unlock_pre(void *ap)
3543 {
3544 #ifdef DEBUG_VFS_LOCKS
3545 	struct vop_unlock_args *a = ap;
3546 
3547 	if (a->a_flags & LK_INTERLOCK)
3548 		ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
3549 	ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
3550 #endif
3551 }
3552 
3553 void
3554 vop_unlock_post(void *ap, int rc)
3555 {
3556 #ifdef DEBUG_VFS_LOCKS
3557 	struct vop_unlock_args *a = ap;
3558 
3559 	if (a->a_flags & LK_INTERLOCK)
3560 		ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
3561 #endif
3562 }
3563 
3564 void
3565 vop_create_post(void *ap, int rc)
3566 {
3567 	struct vop_create_args *a = ap;
3568 
3569 	if (!rc)
3570 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3571 }
3572 
3573 void
3574 vop_link_post(void *ap, int rc)
3575 {
3576 	struct vop_link_args *a = ap;
3577 
3578 	if (!rc) {
3579 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
3580 		VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
3581 	}
3582 }
3583 
3584 void
3585 vop_mkdir_post(void *ap, int rc)
3586 {
3587 	struct vop_mkdir_args *a = ap;
3588 
3589 	if (!rc)
3590 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3591 }
3592 
3593 void
3594 vop_mknod_post(void *ap, int rc)
3595 {
3596 	struct vop_mknod_args *a = ap;
3597 
3598 	if (!rc)
3599 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3600 }
3601 
3602 void
3603 vop_remove_post(void *ap, int rc)
3604 {
3605 	struct vop_remove_args *a = ap;
3606 
3607 	if (!rc) {
3608 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3609 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3610 	}
3611 }
3612 
3613 void
3614 vop_rename_post(void *ap, int rc)
3615 {
3616 	struct vop_rename_args *a = ap;
3617 
3618 	if (!rc) {
3619 		VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
3620 		VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
3621 		VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
3622 		if (a->a_tvp)
3623 			VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
3624 	}
3625 	if (a->a_tdvp != a->a_fdvp)
3626 		vdrop(a->a_fdvp);
3627 	if (a->a_tvp != a->a_fvp)
3628 		vdrop(a->a_fvp);
3629 	vdrop(a->a_tdvp);
3630 	if (a->a_tvp)
3631 		vdrop(a->a_tvp);
3632 }
3633 
3634 void
3635 vop_rmdir_post(void *ap, int rc)
3636 {
3637 	struct vop_rmdir_args *a = ap;
3638 
3639 	if (!rc) {
3640 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3641 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3642 	}
3643 }
3644 
3645 void
3646 vop_setattr_post(void *ap, int rc)
3647 {
3648 	struct vop_setattr_args *a = ap;
3649 
3650 	if (!rc)
3651 		VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
3652 }
3653 
3654 void
3655 vop_symlink_post(void *ap, int rc)
3656 {
3657 	struct vop_symlink_args *a = ap;
3658 
3659 	if (!rc)
3660 		VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3661 }
3662 
3663 static struct knlist fs_knlist;
3664 
3665 static void
3666 vfs_event_init(void *arg)
3667 {
3668 	knlist_init(&fs_knlist, NULL, NULL, NULL, NULL);
3669 }
3670 /* XXX - correct order? */
3671 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
3672 
3673 void
3674 vfs_event_signal(fsid_t *fsid, u_int32_t event, intptr_t data __unused)
3675 {
3676 
3677 	KNOTE_UNLOCKED(&fs_knlist, event);
3678 }
3679 
3680 static int	filt_fsattach(struct knote *kn);
3681 static void	filt_fsdetach(struct knote *kn);
3682 static int	filt_fsevent(struct knote *kn, long hint);
3683 
3684 struct filterops fs_filtops =
3685 	{ 0, filt_fsattach, filt_fsdetach, filt_fsevent };
3686 
3687 static int
3688 filt_fsattach(struct knote *kn)
3689 {
3690 
3691 	kn->kn_flags |= EV_CLEAR;
3692 	knlist_add(&fs_knlist, kn, 0);
3693 	return (0);
3694 }
3695 
3696 static void
3697 filt_fsdetach(struct knote *kn)
3698 {
3699 
3700 	knlist_remove(&fs_knlist, kn, 0);
3701 }
3702 
3703 static int
3704 filt_fsevent(struct knote *kn, long hint)
3705 {
3706 
3707 	kn->kn_fflags |= hint;
3708 	return (kn->kn_fflags != 0);
3709 }
3710 
3711 static int
3712 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
3713 {
3714 	struct vfsidctl vc;
3715 	int error;
3716 	struct mount *mp;
3717 
3718 	error = SYSCTL_IN(req, &vc, sizeof(vc));
3719 	if (error)
3720 		return (error);
3721 	if (vc.vc_vers != VFS_CTL_VERS1)
3722 		return (EINVAL);
3723 	mp = vfs_getvfs(&vc.vc_fsid);
3724 	if (mp == NULL)
3725 		return (ENOENT);
3726 	/* ensure that a specific sysctl goes to the right filesystem. */
3727 	if (strcmp(vc.vc_fstypename, "*") != 0 &&
3728 	    strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
3729 		vfs_rel(mp);
3730 		return (EINVAL);
3731 	}
3732 	VCTLTOREQ(&vc, req);
3733 	error = VFS_SYSCTL(mp, vc.vc_op, req);
3734 	vfs_rel(mp);
3735 	return (error);
3736 }
3737 
3738 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLFLAG_WR,
3739         NULL, 0, sysctl_vfs_ctl, "", "Sysctl by fsid");
3740 
3741 /*
3742  * Function to initialize a va_filerev field sensibly.
3743  * XXX: Wouldn't a random number make a lot more sense ??
3744  */
3745 u_quad_t
3746 init_va_filerev(void)
3747 {
3748 	struct bintime bt;
3749 
3750 	getbinuptime(&bt);
3751 	return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
3752 }
3753 
3754 static int	filt_vfsread(struct knote *kn, long hint);
3755 static int	filt_vfswrite(struct knote *kn, long hint);
3756 static int	filt_vfsvnode(struct knote *kn, long hint);
3757 static void	filt_vfsdetach(struct knote *kn);
3758 static struct filterops vfsread_filtops =
3759 	{ 1, NULL, filt_vfsdetach, filt_vfsread };
3760 static struct filterops vfswrite_filtops =
3761 	{ 1, NULL, filt_vfsdetach, filt_vfswrite };
3762 static struct filterops vfsvnode_filtops =
3763 	{ 1, NULL, filt_vfsdetach, filt_vfsvnode };
3764 
3765 static void
3766 vfs_knllock(void *arg)
3767 {
3768 	struct vnode *vp = arg;
3769 
3770 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread);
3771 }
3772 
3773 static void
3774 vfs_knlunlock(void *arg)
3775 {
3776 	struct vnode *vp = arg;
3777 
3778 	VOP_UNLOCK(vp, 0, curthread);
3779 }
3780 
3781 static int
3782 vfs_knllocked(void *arg)
3783 {
3784 	struct vnode *vp = arg;
3785 
3786 	return (VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE);
3787 }
3788 
3789 int
3790 vfs_kqfilter(struct vop_kqfilter_args *ap)
3791 {
3792 	struct vnode *vp = ap->a_vp;
3793 	struct knote *kn = ap->a_kn;
3794 	struct knlist *knl;
3795 
3796 	switch (kn->kn_filter) {
3797 	case EVFILT_READ:
3798 		kn->kn_fop = &vfsread_filtops;
3799 		break;
3800 	case EVFILT_WRITE:
3801 		kn->kn_fop = &vfswrite_filtops;
3802 		break;
3803 	case EVFILT_VNODE:
3804 		kn->kn_fop = &vfsvnode_filtops;
3805 		break;
3806 	default:
3807 		return (EINVAL);
3808 	}
3809 
3810 	kn->kn_hook = (caddr_t)vp;
3811 
3812 	if (vp->v_pollinfo == NULL)
3813 		v_addpollinfo(vp);
3814 	if (vp->v_pollinfo == NULL)
3815 		return (ENOMEM);
3816 	knl = &vp->v_pollinfo->vpi_selinfo.si_note;
3817 	knlist_add(knl, kn, 0);
3818 
3819 	return (0);
3820 }
3821 
3822 /*
3823  * Detach knote from vnode
3824  */
3825 static void
3826 filt_vfsdetach(struct knote *kn)
3827 {
3828 	struct vnode *vp = (struct vnode *)kn->kn_hook;
3829 
3830 	KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
3831 	knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
3832 }
3833 
3834 /*ARGSUSED*/
3835 static int
3836 filt_vfsread(struct knote *kn, long hint)
3837 {
3838 	struct vnode *vp = (struct vnode *)kn->kn_hook;
3839 	struct vattr va;
3840 
3841 	/*
3842 	 * filesystem is gone, so set the EOF flag and schedule
3843 	 * the knote for deletion.
3844 	 */
3845 	if (hint == NOTE_REVOKE) {
3846 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
3847 		return (1);
3848 	}
3849 
3850 	if (VOP_GETATTR(vp, &va, curthread->td_ucred, curthread))
3851 		return (0);
3852 
3853 	kn->kn_data = va.va_size - kn->kn_fp->f_offset;
3854 	return (kn->kn_data != 0);
3855 }
3856 
3857 /*ARGSUSED*/
3858 static int
3859 filt_vfswrite(struct knote *kn, long hint)
3860 {
3861 	/*
3862 	 * filesystem is gone, so set the EOF flag and schedule
3863 	 * the knote for deletion.
3864 	 */
3865 	if (hint == NOTE_REVOKE)
3866 		kn->kn_flags |= (EV_EOF | EV_ONESHOT);
3867 
3868 	kn->kn_data = 0;
3869 	return (1);
3870 }
3871 
3872 static int
3873 filt_vfsvnode(struct knote *kn, long hint)
3874 {
3875 	if (kn->kn_sfflags & hint)
3876 		kn->kn_fflags |= hint;
3877 	if (hint == NOTE_REVOKE) {
3878 		kn->kn_flags |= EV_EOF;
3879 		return (1);
3880 	}
3881 	return (kn->kn_fflags != 0);
3882 }
3883 
3884 int
3885 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
3886 {
3887 	int error;
3888 
3889 	if (dp->d_reclen > ap->a_uio->uio_resid)
3890 		return (ENAMETOOLONG);
3891 	error = uiomove(dp, dp->d_reclen, ap->a_uio);
3892 	if (error) {
3893 		if (ap->a_ncookies != NULL) {
3894 			if (ap->a_cookies != NULL)
3895 				free(ap->a_cookies, M_TEMP);
3896 			ap->a_cookies = NULL;
3897 			*ap->a_ncookies = 0;
3898 		}
3899 		return (error);
3900 	}
3901 	if (ap->a_ncookies == NULL)
3902 		return (0);
3903 
3904 	KASSERT(ap->a_cookies,
3905 	    ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
3906 
3907 	*ap->a_cookies = realloc(*ap->a_cookies,
3908 	    (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
3909 	(*ap->a_cookies)[*ap->a_ncookies] = off;
3910 	return (0);
3911 }
3912 
3913 /*
3914  * Mark for update the access time of the file if the filesystem
3915  * supports VA_MARK_ATIME.  This functionality is used by execve
3916  * and mmap, so we want to avoid the synchronous I/O implied by
3917  * directly setting va_atime for the sake of efficiency.
3918  */
3919 void
3920 vfs_mark_atime(struct vnode *vp, struct thread *td)
3921 {
3922 	struct vattr atimeattr;
3923 
3924 	if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) {
3925 		VATTR_NULL(&atimeattr);
3926 		atimeattr.va_vaflags |= VA_MARK_ATIME;
3927 		(void)VOP_SETATTR(vp, &atimeattr, td->td_ucred, td);
3928 	}
3929 }
3930