xref: /freebsd/sys/kern/vfs_subr.c (revision 0fddbf874719b9bd50cf66ac26d1140bb3f2be69)
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  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)vfs_subr.c	8.31 (Berkeley) 5/26/95
39  * $FreeBSD$
40  */
41 
42 /*
43  * External virtual filesystem routines
44  */
45 #include "opt_ddb.h"
46 #include "opt_ffs.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/bio.h>
51 #include <sys/buf.h>
52 #include <sys/conf.h>
53 #include <sys/eventhandler.h>
54 #include <sys/fcntl.h>
55 #include <sys/kernel.h>
56 #include <sys/kthread.h>
57 #include <sys/malloc.h>
58 #include <sys/mount.h>
59 #include <sys/namei.h>
60 #include <sys/stat.h>
61 #include <sys/sysctl.h>
62 #include <sys/vmmeter.h>
63 #include <sys/vnode.h>
64 
65 #include <vm/vm.h>
66 #include <vm/vm_object.h>
67 #include <vm/vm_extern.h>
68 #include <vm/pmap.h>
69 #include <vm/vm_map.h>
70 #include <vm/vm_page.h>
71 #include <vm/vm_zone.h>
72 
73 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
74 
75 static void	addalias __P((struct vnode *vp, dev_t nvp_rdev));
76 static void	insmntque __P((struct vnode *vp, struct mount *mp));
77 static void	vclean __P((struct vnode *vp, int flags, struct proc *p));
78 
79 /*
80  * Number of vnodes in existence.  Increased whenever getnewvnode()
81  * allocates a new vnode, never decreased.
82  */
83 static unsigned long	numvnodes;
84 SYSCTL_LONG(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
85 
86 /*
87  * Conversion tables for conversion from vnode types to inode formats
88  * and back.
89  */
90 enum vtype iftovt_tab[16] = {
91 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
92 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
93 };
94 int vttoif_tab[9] = {
95 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
96 	S_IFSOCK, S_IFIFO, S_IFMT,
97 };
98 
99 /*
100  * List of vnodes that are ready for recycling.
101  */
102 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
103 
104 /*
105  * Minimum number of free vnodes.  If there are fewer than this free vnodes,
106  * getnewvnode() will return a newly allocated vnode.
107  */
108 static u_long wantfreevnodes = 25;
109 SYSCTL_LONG(_debug, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
110 /* Number of vnodes in the free list. */
111 static u_long freevnodes = 0;
112 SYSCTL_LONG(_debug, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
113 /* Number of vnode allocation. */
114 static u_long vnodeallocs = 0;
115 SYSCTL_LONG(_debug, OID_AUTO, vnodeallocs, CTLFLAG_RD, &vnodeallocs, 0, "");
116 /* Period of vnode recycle from namecache in vnode allocation times. */
117 static u_long vnoderecycleperiod = 1000;
118 SYSCTL_LONG(_debug, OID_AUTO, vnoderecycleperiod, CTLFLAG_RW, &vnoderecycleperiod, 0, "");
119 /* Minimum number of total vnodes required to invoke vnode recycle from namecache. */
120 static u_long vnoderecyclemintotalvn = 2000;
121 SYSCTL_LONG(_debug, OID_AUTO, vnoderecyclemintotalvn, CTLFLAG_RW, &vnoderecyclemintotalvn, 0, "");
122 /* Minimum number of free vnodes required to invoke vnode recycle from namecache. */
123 static u_long vnoderecycleminfreevn = 2000;
124 SYSCTL_LONG(_debug, OID_AUTO, vnoderecycleminfreevn, CTLFLAG_RW, &vnoderecycleminfreevn, 0, "");
125 /* Number of vnodes attempted to recycle at a time. */
126 static u_long vnoderecyclenumber = 3000;
127 SYSCTL_LONG(_debug, OID_AUTO, vnoderecyclenumber, CTLFLAG_RW, &vnoderecyclenumber, 0, "");
128 
129 /*
130  * Various variables used for debugging the new implementation of
131  * reassignbuf().
132  * XXX these are probably of (very) limited utility now.
133  */
134 static int reassignbufcalls;
135 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
136 static int reassignbufloops;
137 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW, &reassignbufloops, 0, "");
138 static int reassignbufsortgood;
139 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW, &reassignbufsortgood, 0, "");
140 static int reassignbufsortbad;
141 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW, &reassignbufsortbad, 0, "");
142 /* Set to 0 for old insertion-sort based reassignbuf, 1 for modern method. */
143 static int reassignbufmethod = 1;
144 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW, &reassignbufmethod, 0, "");
145 
146 #ifdef ENABLE_VFS_IOOPT
147 /* See NOTES for a description of this setting. */
148 int vfs_ioopt = 0;
149 SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, "");
150 #endif
151 
152 /* List of mounted filesystems. */
153 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
154 
155 /* For any iteration/modification of mountlist */
156 struct mtx mountlist_mtx;
157 
158 /* For any iteration/modification of mnt_vnodelist */
159 struct mtx mntvnode_mtx;
160 
161 /*
162  * Cache for the mount type id assigned to NFS.  This is used for
163  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
164  */
165 int	nfs_mount_type = -1;
166 
167 /* To keep more than one thread at a time from running vfs_getnewfsid */
168 static struct mtx mntid_mtx;
169 
170 /* For any iteration/modification of vnode_free_list */
171 static struct mtx vnode_free_list_mtx;
172 
173 /*
174  * For any iteration/modification of dev->si_hlist (linked through
175  * v_specnext)
176  */
177 static struct mtx spechash_mtx;
178 
179 /* Publicly exported FS */
180 struct nfs_public nfs_pub;
181 
182 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
183 static vm_zone_t vnode_zone;
184 
185 /* Set to 1 to print out reclaim of active vnodes */
186 int	prtactive = 0;
187 
188 /*
189  * The workitem queue.
190  *
191  * It is useful to delay writes of file data and filesystem metadata
192  * for tens of seconds so that quickly created and deleted files need
193  * not waste disk bandwidth being created and removed. To realize this,
194  * we append vnodes to a "workitem" queue. When running with a soft
195  * updates implementation, most pending metadata dependencies should
196  * not wait for more than a few seconds. Thus, mounted on block devices
197  * are delayed only about a half the time that file data is delayed.
198  * Similarly, directory updates are more critical, so are only delayed
199  * about a third the time that file data is delayed. Thus, there are
200  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
201  * one each second (driven off the filesystem syncer process). The
202  * syncer_delayno variable indicates the next queue that is to be processed.
203  * Items that need to be processed soon are placed in this queue:
204  *
205  *	syncer_workitem_pending[syncer_delayno]
206  *
207  * A delay of fifteen seconds is done by placing the request fifteen
208  * entries later in the queue:
209  *
210  *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
211  *
212  */
213 static int syncer_delayno = 0;
214 static long syncer_mask;
215 LIST_HEAD(synclist, vnode);
216 static struct synclist *syncer_workitem_pending;
217 
218 #define SYNCER_MAXDELAY		32
219 static int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
220 time_t syncdelay = 30;		/* max time to delay syncing data */
221 time_t filedelay = 30;		/* time to delay syncing files */
222 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
223 time_t dirdelay = 29;		/* time to delay syncing directories */
224 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
225 time_t metadelay = 28;		/* time to delay syncing metadata */
226 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
227 static int rushjob;		/* number of slots to run ASAP */
228 static int stat_rush_requests;	/* number of times I/O speeded up */
229 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
230 
231 /*
232  * Number of vnodes we want to exist at any one time.  This is mostly used
233  * to size hash tables in vnode-related code.  It is normally not used in
234  * getnewvnode(), as wantfreevnodes is normally nonzero.)
235  *
236  * XXX desiredvnodes is historical cruft and should not exist.
237  */
238 int desiredvnodes;
239 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
240     &desiredvnodes, 0, "Maximum number of vnodes");
241 
242 /*
243  * Initialize the vnode management data structures.
244  */
245 static void
246 vntblinit(void *dummy __unused)
247 {
248 
249 	desiredvnodes = maxproc + cnt.v_page_count / 4;
250 	mtx_init(&mountlist_mtx, "mountlist", MTX_DEF);
251 	mtx_init(&mntvnode_mtx, "mntvnode", MTX_DEF);
252 	mtx_init(&mntid_mtx, "mntid", MTX_DEF);
253 	mtx_init(&spechash_mtx, "spechash", MTX_DEF);
254 	TAILQ_INIT(&vnode_free_list);
255 	mtx_init(&vnode_free_list_mtx, "vnode_free_list", MTX_DEF);
256 	vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5);
257 	/*
258 	 * Initialize the filesystem syncer.
259 	 */
260 	syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
261 		&syncer_mask);
262 	syncer_maxdelay = syncer_mask + 1;
263 }
264 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL)
265 
266 
267 /*
268  * Mark a mount point as busy. Used to synchronize access and to delay
269  * unmounting. Interlock is not released on failure.
270  */
271 int
272 vfs_busy(mp, flags, interlkp, p)
273 	struct mount *mp;
274 	int flags;
275 	struct mtx *interlkp;
276 	struct proc *p;
277 {
278 	int lkflags;
279 
280 	if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
281 		if (flags & LK_NOWAIT)
282 			return (ENOENT);
283 		mp->mnt_kern_flag |= MNTK_MWAIT;
284 		/*
285 		 * Since all busy locks are shared except the exclusive
286 		 * lock granted when unmounting, the only place that a
287 		 * wakeup needs to be done is at the release of the
288 		 * exclusive lock at the end of dounmount.
289 		 */
290 		msleep((caddr_t)mp, interlkp, PVFS, "vfs_busy", 0);
291 		return (ENOENT);
292 	}
293 	lkflags = LK_SHARED | LK_NOPAUSE;
294 	if (interlkp)
295 		lkflags |= LK_INTERLOCK;
296 	if (lockmgr(&mp->mnt_lock, lkflags, interlkp, p))
297 		panic("vfs_busy: unexpected lock failure");
298 	return (0);
299 }
300 
301 /*
302  * Free a busy filesystem.
303  */
304 void
305 vfs_unbusy(mp, p)
306 	struct mount *mp;
307 	struct proc *p;
308 {
309 
310 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, p);
311 }
312 
313 /*
314  * Lookup a filesystem type, and if found allocate and initialize
315  * a mount structure for it.
316  *
317  * Devname is usually updated by mount(8) after booting.
318  */
319 int
320 vfs_rootmountalloc(fstypename, devname, mpp)
321 	char *fstypename;
322 	char *devname;
323 	struct mount **mpp;
324 {
325 	struct proc *p = curproc;	/* XXX */
326 	struct vfsconf *vfsp;
327 	struct mount *mp;
328 
329 	if (fstypename == NULL)
330 		return (ENODEV);
331 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
332 		if (!strcmp(vfsp->vfc_name, fstypename))
333 			break;
334 	if (vfsp == NULL)
335 		return (ENODEV);
336 	mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO);
337 	lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE);
338 	(void)vfs_busy(mp, LK_NOWAIT, 0, p);
339 	LIST_INIT(&mp->mnt_vnodelist);
340 	mp->mnt_vfc = vfsp;
341 	mp->mnt_op = vfsp->vfc_vfsops;
342 	mp->mnt_flag = MNT_RDONLY;
343 	mp->mnt_vnodecovered = NULLVP;
344 	vfsp->vfc_refcount++;
345 	mp->mnt_iosize_max = DFLTPHYS;
346 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
347 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
348 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
349 	mp->mnt_stat.f_mntonname[0] = '/';
350 	mp->mnt_stat.f_mntonname[1] = 0;
351 	(void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
352 	*mpp = mp;
353 	return (0);
354 }
355 
356 /*
357  * Find an appropriate filesystem to use for the root. If a filesystem
358  * has not been preselected, walk through the list of known filesystems
359  * trying those that have mountroot routines, and try them until one
360  * works or we have tried them all.
361  */
362 #ifdef notdef	/* XXX JH */
363 int
364 lite2_vfs_mountroot()
365 {
366 	struct vfsconf *vfsp;
367 	extern int (*lite2_mountroot) __P((void));
368 	int error;
369 
370 	if (lite2_mountroot != NULL)
371 		return ((*lite2_mountroot)());
372 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
373 		if (vfsp->vfc_mountroot == NULL)
374 			continue;
375 		if ((error = (*vfsp->vfc_mountroot)()) == 0)
376 			return (0);
377 		printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error);
378 	}
379 	return (ENODEV);
380 }
381 #endif
382 
383 /*
384  * Lookup a mount point by filesystem identifier.
385  */
386 struct mount *
387 vfs_getvfs(fsid)
388 	fsid_t *fsid;
389 {
390 	register 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 			mtx_unlock(&mountlist_mtx);
397 			return (mp);
398 	    }
399 	}
400 	mtx_unlock(&mountlist_mtx);
401 	return ((struct mount *) 0);
402 }
403 
404 /*
405  * Get a new unique fsid.  Try to make its val[0] unique, since this value
406  * will be used to create fake device numbers for stat().  Also try (but
407  * not so hard) make its val[0] unique mod 2^16, since some emulators only
408  * support 16-bit device numbers.  We end up with unique val[0]'s for the
409  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
410  *
411  * Keep in mind that several mounts may be running in parallel.  Starting
412  * the search one past where the previous search terminated is both a
413  * micro-optimization and a defense against returning the same fsid to
414  * different mounts.
415  */
416 void
417 vfs_getnewfsid(mp)
418 	struct mount *mp;
419 {
420 	static u_int16_t mntid_base;
421 	fsid_t tfsid;
422 	int mtype;
423 
424 	mtx_lock(&mntid_mtx);
425 	mtype = mp->mnt_vfc->vfc_typenum;
426 	tfsid.val[1] = mtype;
427 	mtype = (mtype & 0xFF) << 24;
428 	for (;;) {
429 		tfsid.val[0] = makeudev(255,
430 		    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
431 		mntid_base++;
432 		if (vfs_getvfs(&tfsid) == NULL)
433 			break;
434 	}
435 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
436 	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
437 	mtx_unlock(&mntid_mtx);
438 }
439 
440 /*
441  * Knob to control the precision of file timestamps:
442  *
443  *   0 = seconds only; nanoseconds zeroed.
444  *   1 = seconds and nanoseconds, accurate within 1/HZ.
445  *   2 = seconds and nanoseconds, truncated to microseconds.
446  * >=3 = seconds and nanoseconds, maximum precision.
447  */
448 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
449 
450 static int timestamp_precision = TSP_SEC;
451 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
452     &timestamp_precision, 0, "");
453 
454 /*
455  * Get a current timestamp.
456  */
457 void
458 vfs_timestamp(tsp)
459 	struct timespec *tsp;
460 {
461 	struct timeval tv;
462 
463 	switch (timestamp_precision) {
464 	case TSP_SEC:
465 		tsp->tv_sec = time_second;
466 		tsp->tv_nsec = 0;
467 		break;
468 	case TSP_HZ:
469 		getnanotime(tsp);
470 		break;
471 	case TSP_USEC:
472 		microtime(&tv);
473 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
474 		break;
475 	case TSP_NSEC:
476 	default:
477 		nanotime(tsp);
478 		break;
479 	}
480 }
481 
482 /*
483  * Set vnode attributes to VNOVAL
484  */
485 void
486 vattr_null(vap)
487 	register struct vattr *vap;
488 {
489 
490 	vap->va_type = VNON;
491 	vap->va_size = VNOVAL;
492 	vap->va_bytes = VNOVAL;
493 	vap->va_mode = VNOVAL;
494 	vap->va_nlink = VNOVAL;
495 	vap->va_uid = VNOVAL;
496 	vap->va_gid = VNOVAL;
497 	vap->va_fsid = VNOVAL;
498 	vap->va_fileid = VNOVAL;
499 	vap->va_blocksize = VNOVAL;
500 	vap->va_rdev = VNOVAL;
501 	vap->va_atime.tv_sec = VNOVAL;
502 	vap->va_atime.tv_nsec = VNOVAL;
503 	vap->va_mtime.tv_sec = VNOVAL;
504 	vap->va_mtime.tv_nsec = VNOVAL;
505 	vap->va_ctime.tv_sec = VNOVAL;
506 	vap->va_ctime.tv_nsec = VNOVAL;
507 	vap->va_flags = VNOVAL;
508 	vap->va_gen = VNOVAL;
509 	vap->va_vaflags = 0;
510 }
511 
512 /*
513  * Routines having to do with the management of the vnode table.
514  */
515 
516 /*
517  * Return the next vnode from the free list.
518  */
519 int
520 getnewvnode(tag, mp, vops, vpp)
521 	enum vtagtype tag;
522 	struct mount *mp;
523 	vop_t **vops;
524 	struct vnode **vpp;
525 {
526 	int s, count;
527 	struct proc *p = curproc;	/* XXX */
528 	struct vnode *vp = NULL;
529 	struct mount *vnmp;
530 	vm_object_t object;
531 
532 	/*
533 	 * We take the least recently used vnode from the freelist
534 	 * if we can get it and it has no cached pages, and no
535 	 * namecache entries are relative to it.
536 	 * Otherwise we allocate a new vnode
537 	 */
538 
539 	s = splbio();
540 	mtx_lock(&vnode_free_list_mtx);
541 
542 	if (wantfreevnodes && freevnodes < wantfreevnodes) {
543 		vp = NULL;
544 	} else if (!wantfreevnodes && freevnodes <= desiredvnodes) {
545 		/*
546 		 * XXX: this is only here to be backwards compatible
547 		 */
548 		vp = NULL;
549 	} else for (count = 0; count < freevnodes; count++) {
550 		vp = TAILQ_FIRST(&vnode_free_list);
551 		if (vp == NULL || vp->v_usecount)
552 			panic("getnewvnode: free vnode isn't");
553 		TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
554 
555 		/*
556 		 * Don't recycle if active in the namecache or
557 		 * if it still has cached pages or we cannot get
558 		 * its interlock.
559 		 */
560 		if (LIST_FIRST(&vp->v_cache_src) != NULL ||
561 		    (VOP_GETVOBJECT(vp, &object) == 0 &&
562 		     (object->resident_page_count || object->ref_count)) ||
563 		    !mtx_trylock(&vp->v_interlock)) {
564 			TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
565 			vp = NULL;
566 			continue;
567 		}
568 		/*
569 		 * Skip over it if its filesystem is being suspended.
570 		 */
571 		if (vn_start_write(vp, &vnmp, V_NOWAIT) == 0)
572 			break;
573 		mtx_unlock(&vp->v_interlock);
574 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
575 		vp = NULL;
576 	}
577 	if (vp) {
578 		vp->v_flag |= VDOOMED;
579 		vp->v_flag &= ~VFREE;
580 		freevnodes--;
581 		mtx_unlock(&vnode_free_list_mtx);
582 		cache_purge(vp);
583 		vp->v_lease = NULL;
584 		if (vp->v_type != VBAD) {
585 			vgonel(vp, p);
586 		} else {
587 			mtx_unlock(&vp->v_interlock);
588 		}
589 		vn_finished_write(vnmp);
590 
591 #ifdef INVARIANTS
592 		{
593 			int s;
594 
595 			if (vp->v_data)
596 				panic("cleaned vnode isn't");
597 			s = splbio();
598 			if (vp->v_numoutput)
599 				panic("Clean vnode has pending I/O's");
600 			splx(s);
601 			if (vp->v_writecount != 0)
602 				panic("Non-zero write count");
603 		}
604 #endif
605 		vp->v_flag = 0;
606 		vp->v_lastw = 0;
607 		vp->v_lasta = 0;
608 		vp->v_cstart = 0;
609 		vp->v_clen = 0;
610 		vp->v_socket = 0;
611 	} else {
612 		mtx_unlock(&vnode_free_list_mtx);
613 		vp = (struct vnode *) zalloc(vnode_zone);
614 		bzero((char *) vp, sizeof *vp);
615 		mtx_init(&vp->v_interlock, "vnode interlock", MTX_DEF);
616 		vp->v_dd = vp;
617 		mtx_init(&vp->v_pollinfo.vpi_lock, "vnode pollinfo", MTX_DEF);
618 		cache_purge(vp);
619 		LIST_INIT(&vp->v_cache_src);
620 		TAILQ_INIT(&vp->v_cache_dst);
621 		numvnodes++;
622 	}
623 
624 	TAILQ_INIT(&vp->v_cleanblkhd);
625 	TAILQ_INIT(&vp->v_dirtyblkhd);
626 	vp->v_type = VNON;
627 	vp->v_tag = tag;
628 	vp->v_op = vops;
629 	lockinit(&vp->v_lock, PVFS, "vnlock", 0, LK_NOPAUSE);
630 	insmntque(vp, mp);
631 	*vpp = vp;
632 	vp->v_usecount = 1;
633 	vp->v_data = 0;
634 
635 	splx(s);
636 
637 	vfs_object_create(vp, p, p->p_ucred);
638 
639 	vnodeallocs++;
640 	if (vnodeallocs % vnoderecycleperiod == 0 &&
641 	    freevnodes < vnoderecycleminfreevn &&
642 	    vnoderecyclemintotalvn < numvnodes) {
643 		/* Recycle vnodes. */
644 		cache_purgeleafdirs(vnoderecyclenumber);
645 	}
646 
647 	return (0);
648 }
649 
650 /*
651  * Move a vnode from one mount queue to another.
652  */
653 static void
654 insmntque(vp, mp)
655 	register struct vnode *vp;
656 	register struct mount *mp;
657 {
658 
659 	mtx_lock(&mntvnode_mtx);
660 	/*
661 	 * Delete from old mount point vnode list, if on one.
662 	 */
663 	if (vp->v_mount != NULL)
664 		LIST_REMOVE(vp, v_mntvnodes);
665 	/*
666 	 * Insert into list of vnodes for the new mount point, if available.
667 	 */
668 	if ((vp->v_mount = mp) == NULL) {
669 		mtx_unlock(&mntvnode_mtx);
670 		return;
671 	}
672 	LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes);
673 	mtx_unlock(&mntvnode_mtx);
674 }
675 
676 /*
677  * Update outstanding I/O count and do wakeup if requested.
678  */
679 void
680 vwakeup(bp)
681 	register struct buf *bp;
682 {
683 	register struct vnode *vp;
684 
685 	bp->b_flags &= ~B_WRITEINPROG;
686 	if ((vp = bp->b_vp)) {
687 		vp->v_numoutput--;
688 		if (vp->v_numoutput < 0)
689 			panic("vwakeup: neg numoutput");
690 		if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) {
691 			vp->v_flag &= ~VBWAIT;
692 			wakeup((caddr_t) &vp->v_numoutput);
693 		}
694 	}
695 }
696 
697 /*
698  * Flush out and invalidate all buffers associated with a vnode.
699  * Called with the underlying object locked.
700  */
701 int
702 vinvalbuf(vp, flags, cred, p, slpflag, slptimeo)
703 	register struct vnode *vp;
704 	int flags;
705 	struct ucred *cred;
706 	struct proc *p;
707 	int slpflag, slptimeo;
708 {
709 	register struct buf *bp;
710 	struct buf *nbp, *blist;
711 	int s, error;
712 	vm_object_t object;
713 
714 	GIANT_REQUIRED;
715 
716 	if (flags & V_SAVE) {
717 		s = splbio();
718 		while (vp->v_numoutput) {
719 			vp->v_flag |= VBWAIT;
720 			error = tsleep((caddr_t)&vp->v_numoutput,
721 			    slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo);
722 			if (error) {
723 				splx(s);
724 				return (error);
725 			}
726 		}
727 		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
728 			splx(s);
729 			if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, p)) != 0)
730 				return (error);
731 			s = splbio();
732 			if (vp->v_numoutput > 0 ||
733 			    !TAILQ_EMPTY(&vp->v_dirtyblkhd))
734 				panic("vinvalbuf: dirty bufs");
735 		}
736 		splx(s);
737   	}
738 	s = splbio();
739 	for (;;) {
740 		blist = TAILQ_FIRST(&vp->v_cleanblkhd);
741 		if (!blist)
742 			blist = TAILQ_FIRST(&vp->v_dirtyblkhd);
743 		if (!blist)
744 			break;
745 
746 		for (bp = blist; bp; bp = nbp) {
747 			nbp = TAILQ_NEXT(bp, b_vnbufs);
748 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
749 				error = BUF_TIMELOCK(bp,
750 				    LK_EXCLUSIVE | LK_SLEEPFAIL,
751 				    "vinvalbuf", slpflag, slptimeo);
752 				if (error == ENOLCK)
753 					break;
754 				splx(s);
755 				return (error);
756 			}
757 			/*
758 			 * XXX Since there are no node locks for NFS, I
759 			 * believe there is a slight chance that a delayed
760 			 * write will occur while sleeping just above, so
761 			 * check for it.  Note that vfs_bio_awrite expects
762 			 * buffers to reside on a queue, while BUF_WRITE and
763 			 * brelse do not.
764 			 */
765 			if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
766 				(flags & V_SAVE)) {
767 
768 				if (bp->b_vp == vp) {
769 					if (bp->b_flags & B_CLUSTEROK) {
770 						BUF_UNLOCK(bp);
771 						vfs_bio_awrite(bp);
772 					} else {
773 						bremfree(bp);
774 						bp->b_flags |= B_ASYNC;
775 						BUF_WRITE(bp);
776 					}
777 				} else {
778 					bremfree(bp);
779 					(void) BUF_WRITE(bp);
780 				}
781 				break;
782 			}
783 			bremfree(bp);
784 			bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
785 			bp->b_flags &= ~B_ASYNC;
786 			brelse(bp);
787 		}
788 	}
789 
790 	while (vp->v_numoutput > 0) {
791 		vp->v_flag |= VBWAIT;
792 		tsleep(&vp->v_numoutput, PVM, "vnvlbv", 0);
793 	}
794 
795 	splx(s);
796 
797 	/*
798 	 * Destroy the copy in the VM cache, too.
799 	 */
800 	mtx_lock(&vp->v_interlock);
801 	if (VOP_GETVOBJECT(vp, &object) == 0) {
802 		vm_object_page_remove(object, 0, 0,
803 			(flags & V_SAVE) ? TRUE : FALSE);
804 	}
805 	mtx_unlock(&vp->v_interlock);
806 
807 	if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd))
808 		panic("vinvalbuf: flush failed");
809 	return (0);
810 }
811 
812 /*
813  * Truncate a file's buffer and pages to a specified length.  This
814  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
815  * sync activity.
816  */
817 int
818 vtruncbuf(vp, cred, p, length, blksize)
819 	register struct vnode *vp;
820 	struct ucred *cred;
821 	struct proc *p;
822 	off_t length;
823 	int blksize;
824 {
825 	register struct buf *bp;
826 	struct buf *nbp;
827 	int s, anyfreed;
828 	int trunclbn;
829 
830 	/*
831 	 * Round up to the *next* lbn.
832 	 */
833 	trunclbn = (length + blksize - 1) / blksize;
834 
835 	s = splbio();
836 restart:
837 	anyfreed = 1;
838 	for (;anyfreed;) {
839 		anyfreed = 0;
840 		for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
841 			nbp = TAILQ_NEXT(bp, b_vnbufs);
842 			if (bp->b_lblkno >= trunclbn) {
843 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
844 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
845 					goto restart;
846 				} else {
847 					bremfree(bp);
848 					bp->b_flags |= (B_INVAL | B_RELBUF);
849 					bp->b_flags &= ~B_ASYNC;
850 					brelse(bp);
851 					anyfreed = 1;
852 				}
853 				if (nbp &&
854 				    (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
855 				    (nbp->b_vp != vp) ||
856 				    (nbp->b_flags & B_DELWRI))) {
857 					goto restart;
858 				}
859 			}
860 		}
861 
862 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
863 			nbp = TAILQ_NEXT(bp, b_vnbufs);
864 			if (bp->b_lblkno >= trunclbn) {
865 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
866 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
867 					goto restart;
868 				} else {
869 					bremfree(bp);
870 					bp->b_flags |= (B_INVAL | B_RELBUF);
871 					bp->b_flags &= ~B_ASYNC;
872 					brelse(bp);
873 					anyfreed = 1;
874 				}
875 				if (nbp &&
876 				    (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
877 				    (nbp->b_vp != vp) ||
878 				    (nbp->b_flags & B_DELWRI) == 0)) {
879 					goto restart;
880 				}
881 			}
882 		}
883 	}
884 
885 	if (length > 0) {
886 restartsync:
887 		for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
888 			nbp = TAILQ_NEXT(bp, b_vnbufs);
889 			if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) {
890 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
891 					BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL);
892 					goto restart;
893 				} else {
894 					bremfree(bp);
895 					if (bp->b_vp == vp) {
896 						bp->b_flags |= B_ASYNC;
897 					} else {
898 						bp->b_flags &= ~B_ASYNC;
899 					}
900 					BUF_WRITE(bp);
901 				}
902 				goto restartsync;
903 			}
904 
905 		}
906 	}
907 
908 	while (vp->v_numoutput > 0) {
909 		vp->v_flag |= VBWAIT;
910 		tsleep(&vp->v_numoutput, PVM, "vbtrunc", 0);
911 	}
912 
913 	splx(s);
914 
915 	vnode_pager_setsize(vp, length);
916 
917 	return (0);
918 }
919 
920 /*
921  * Associate a buffer with a vnode.
922  */
923 void
924 bgetvp(vp, bp)
925 	register struct vnode *vp;
926 	register struct buf *bp;
927 {
928 	int s;
929 
930 	KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
931 
932 	vhold(vp);
933 	bp->b_vp = vp;
934 	bp->b_dev = vn_todev(vp);
935 	/*
936 	 * Insert onto list for new vnode.
937 	 */
938 	s = splbio();
939 	bp->b_xflags |= BX_VNCLEAN;
940 	bp->b_xflags &= ~BX_VNDIRTY;
941 	TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
942 	splx(s);
943 }
944 
945 /*
946  * Disassociate a buffer from a vnode.
947  */
948 void
949 brelvp(bp)
950 	register struct buf *bp;
951 {
952 	struct vnode *vp;
953 	struct buflists *listheadp;
954 	int s;
955 
956 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
957 
958 	/*
959 	 * Delete from old vnode list, if on one.
960 	 */
961 	vp = bp->b_vp;
962 	s = splbio();
963 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
964 		if (bp->b_xflags & BX_VNDIRTY)
965 			listheadp = &vp->v_dirtyblkhd;
966 		else
967 			listheadp = &vp->v_cleanblkhd;
968 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
969 		bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
970 	}
971 	if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
972 		vp->v_flag &= ~VONWORKLST;
973 		LIST_REMOVE(vp, v_synclist);
974 	}
975 	splx(s);
976 	bp->b_vp = (struct vnode *) 0;
977 	vdrop(vp);
978 }
979 
980 /*
981  * Add an item to the syncer work queue.
982  */
983 static void
984 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
985 {
986 	int s, slot;
987 
988 	s = splbio();
989 
990 	if (vp->v_flag & VONWORKLST) {
991 		LIST_REMOVE(vp, v_synclist);
992 	}
993 
994 	if (delay > syncer_maxdelay - 2)
995 		delay = syncer_maxdelay - 2;
996 	slot = (syncer_delayno + delay) & syncer_mask;
997 
998 	LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
999 	vp->v_flag |= VONWORKLST;
1000 	splx(s);
1001 }
1002 
1003 struct  proc *updateproc;
1004 static void sched_sync __P((void));
1005 static struct kproc_desc up_kp = {
1006 	"syncer",
1007 	sched_sync,
1008 	&updateproc
1009 };
1010 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1011 
1012 /*
1013  * System filesystem synchronizer daemon.
1014  */
1015 void
1016 sched_sync(void)
1017 {
1018 	struct synclist *slp;
1019 	struct vnode *vp;
1020 	struct mount *mp;
1021 	long starttime;
1022 	int s;
1023 	struct proc *p = updateproc;
1024 
1025 	mtx_lock(&Giant);
1026 
1027 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
1028 	    SHUTDOWN_PRI_LAST);
1029 
1030 	for (;;) {
1031 		kthread_suspend_check(p);
1032 
1033 		starttime = time_second;
1034 
1035 		/*
1036 		 * Push files whose dirty time has expired.  Be careful
1037 		 * of interrupt race on slp queue.
1038 		 */
1039 		s = splbio();
1040 		slp = &syncer_workitem_pending[syncer_delayno];
1041 		syncer_delayno += 1;
1042 		if (syncer_delayno == syncer_maxdelay)
1043 			syncer_delayno = 0;
1044 		splx(s);
1045 
1046 		while ((vp = LIST_FIRST(slp)) != NULL) {
1047 			if (VOP_ISLOCKED(vp, NULL) == 0 &&
1048 			    vn_start_write(vp, &mp, V_NOWAIT) == 0) {
1049 				vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
1050 				(void) VOP_FSYNC(vp, p->p_ucred, MNT_LAZY, p);
1051 				VOP_UNLOCK(vp, 0, p);
1052 				vn_finished_write(mp);
1053 			}
1054 			s = splbio();
1055 			if (LIST_FIRST(slp) == vp) {
1056 				/*
1057 				 * Note: v_tag VT_VFS vps can remain on the
1058 				 * worklist too with no dirty blocks, but
1059 				 * since sync_fsync() moves it to a different
1060 				 * slot we are safe.
1061 				 */
1062 				if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
1063 				    !vn_isdisk(vp, NULL))
1064 					panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
1065 				/*
1066 				 * Put us back on the worklist.  The worklist
1067 				 * routine will remove us from our current
1068 				 * position and then add us back in at a later
1069 				 * position.
1070 				 */
1071 				vn_syncer_add_to_worklist(vp, syncdelay);
1072 			}
1073 			splx(s);
1074 		}
1075 
1076 		/*
1077 		 * Do soft update processing.
1078 		 */
1079 #ifdef SOFTUPDATES
1080 		softdep_process_worklist(NULL);
1081 #endif
1082 
1083 		/*
1084 		 * The variable rushjob allows the kernel to speed up the
1085 		 * processing of the filesystem syncer process. A rushjob
1086 		 * value of N tells the filesystem syncer to process the next
1087 		 * N seconds worth of work on its queue ASAP. Currently rushjob
1088 		 * is used by the soft update code to speed up the filesystem
1089 		 * syncer process when the incore state is getting so far
1090 		 * ahead of the disk that the kernel memory pool is being
1091 		 * threatened with exhaustion.
1092 		 */
1093 		if (rushjob > 0) {
1094 			rushjob -= 1;
1095 			continue;
1096 		}
1097 		/*
1098 		 * If it has taken us less than a second to process the
1099 		 * current work, then wait. Otherwise start right over
1100 		 * again. We can still lose time if any single round
1101 		 * takes more than two seconds, but it does not really
1102 		 * matter as we are just trying to generally pace the
1103 		 * filesystem activity.
1104 		 */
1105 		if (time_second == starttime)
1106 			tsleep(&lbolt, PPAUSE, "syncer", 0);
1107 	}
1108 }
1109 
1110 /*
1111  * Request the syncer daemon to speed up its work.
1112  * We never push it to speed up more than half of its
1113  * normal turn time, otherwise it could take over the cpu.
1114  */
1115 int
1116 speedup_syncer()
1117 {
1118 
1119 	mtx_lock_spin(&sched_lock);
1120 	if (updateproc->p_wchan == &lbolt)
1121 		setrunnable(updateproc);
1122 	mtx_unlock_spin(&sched_lock);
1123 	if (rushjob < syncdelay / 2) {
1124 		rushjob += 1;
1125 		stat_rush_requests += 1;
1126 		return (1);
1127 	}
1128 	return(0);
1129 }
1130 
1131 /*
1132  * Associate a p-buffer with a vnode.
1133  *
1134  * Also sets B_PAGING flag to indicate that vnode is not fully associated
1135  * with the buffer.  i.e. the bp has not been linked into the vnode or
1136  * ref-counted.
1137  */
1138 void
1139 pbgetvp(vp, bp)
1140 	register struct vnode *vp;
1141 	register struct buf *bp;
1142 {
1143 
1144 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1145 
1146 	bp->b_vp = vp;
1147 	bp->b_flags |= B_PAGING;
1148 	bp->b_dev = vn_todev(vp);
1149 }
1150 
1151 /*
1152  * Disassociate a p-buffer from a vnode.
1153  */
1154 void
1155 pbrelvp(bp)
1156 	register struct buf *bp;
1157 {
1158 
1159 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1160 
1161 	/* XXX REMOVE ME */
1162 	if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1163 		panic(
1164 		    "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1165 		    bp,
1166 		    (int)bp->b_flags
1167 		);
1168 	}
1169 	bp->b_vp = (struct vnode *) 0;
1170 	bp->b_flags &= ~B_PAGING;
1171 }
1172 
1173 /*
1174  * Change the vnode a pager buffer is associated with.
1175  */
1176 void
1177 pbreassignbuf(bp, newvp)
1178 	struct buf *bp;
1179 	struct vnode *newvp;
1180 {
1181 
1182 	KASSERT(bp->b_flags & B_PAGING,
1183 	    ("pbreassignbuf() on non phys bp %p", bp));
1184 	bp->b_vp = newvp;
1185 }
1186 
1187 /*
1188  * Reassign a buffer from one vnode to another.
1189  * Used to assign file specific control information
1190  * (indirect blocks) to the vnode to which they belong.
1191  */
1192 void
1193 reassignbuf(bp, newvp)
1194 	register struct buf *bp;
1195 	register struct vnode *newvp;
1196 {
1197 	struct buflists *listheadp;
1198 	int delay;
1199 	int s;
1200 
1201 	if (newvp == NULL) {
1202 		printf("reassignbuf: NULL");
1203 		return;
1204 	}
1205 	++reassignbufcalls;
1206 
1207 	/*
1208 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1209 	 * is not fully linked in.
1210 	 */
1211 	if (bp->b_flags & B_PAGING)
1212 		panic("cannot reassign paging buffer");
1213 
1214 	s = splbio();
1215 	/*
1216 	 * Delete from old vnode list, if on one.
1217 	 */
1218 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1219 		if (bp->b_xflags & BX_VNDIRTY)
1220 			listheadp = &bp->b_vp->v_dirtyblkhd;
1221 		else
1222 			listheadp = &bp->b_vp->v_cleanblkhd;
1223 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1224 		bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1225 		if (bp->b_vp != newvp) {
1226 			vdrop(bp->b_vp);
1227 			bp->b_vp = NULL;	/* for clarification */
1228 		}
1229 	}
1230 	/*
1231 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1232 	 * of clean buffers.
1233 	 */
1234 	if (bp->b_flags & B_DELWRI) {
1235 		struct buf *tbp;
1236 
1237 		listheadp = &newvp->v_dirtyblkhd;
1238 		if ((newvp->v_flag & VONWORKLST) == 0) {
1239 			switch (newvp->v_type) {
1240 			case VDIR:
1241 				delay = dirdelay;
1242 				break;
1243 			case VCHR:
1244 				if (newvp->v_rdev->si_mountpoint != NULL) {
1245 					delay = metadelay;
1246 					break;
1247 				}
1248 				/* fall through */
1249 			default:
1250 				delay = filedelay;
1251 			}
1252 			vn_syncer_add_to_worklist(newvp, delay);
1253 		}
1254 		bp->b_xflags |= BX_VNDIRTY;
1255 		tbp = TAILQ_FIRST(listheadp);
1256 		if (tbp == NULL ||
1257 		    bp->b_lblkno == 0 ||
1258 		    (bp->b_lblkno > 0 && tbp->b_lblkno < 0) ||
1259 		    (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) {
1260 			TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1261 			++reassignbufsortgood;
1262 		} else if (bp->b_lblkno < 0) {
1263 			TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1264 			++reassignbufsortgood;
1265 		} else if (reassignbufmethod == 1) {
1266 			/*
1267 			 * New sorting algorithm, only handle sequential case,
1268 			 * otherwise append to end (but before metadata)
1269 			 */
1270 			if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL &&
1271 			    (tbp->b_xflags & BX_VNDIRTY)) {
1272 				/*
1273 				 * Found the best place to insert the buffer
1274 				 */
1275 				TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1276 				++reassignbufsortgood;
1277 			} else {
1278 				/*
1279 				 * Missed, append to end, but before meta-data.
1280 				 * We know that the head buffer in the list is
1281 				 * not meta-data due to prior conditionals.
1282 				 *
1283 				 * Indirect effects:  NFS second stage write
1284 				 * tends to wind up here, giving maximum
1285 				 * distance between the unstable write and the
1286 				 * commit rpc.
1287 				 */
1288 				tbp = TAILQ_LAST(listheadp, buflists);
1289 				while (tbp && tbp->b_lblkno < 0)
1290 					tbp = TAILQ_PREV(tbp, buflists, b_vnbufs);
1291 				TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1292 				++reassignbufsortbad;
1293 			}
1294 		} else {
1295 			/*
1296 			 * Old sorting algorithm, scan queue and insert
1297 			 */
1298 			struct buf *ttbp;
1299 			while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1300 			    (ttbp->b_lblkno < bp->b_lblkno)) {
1301 				++reassignbufloops;
1302 				tbp = ttbp;
1303 			}
1304 			TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1305 		}
1306 	} else {
1307 		bp->b_xflags |= BX_VNCLEAN;
1308 		TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1309 		if ((newvp->v_flag & VONWORKLST) &&
1310 		    TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1311 			newvp->v_flag &= ~VONWORKLST;
1312 			LIST_REMOVE(newvp, v_synclist);
1313 		}
1314 	}
1315 	if (bp->b_vp != newvp) {
1316 		bp->b_vp = newvp;
1317 		vhold(bp->b_vp);
1318 	}
1319 	splx(s);
1320 }
1321 
1322 /*
1323  * Create a vnode for a device.
1324  * Used for mounting the root file system.
1325  */
1326 int
1327 bdevvp(dev, vpp)
1328 	dev_t dev;
1329 	struct vnode **vpp;
1330 {
1331 	register struct vnode *vp;
1332 	struct vnode *nvp;
1333 	int error;
1334 
1335 	if (dev == NODEV) {
1336 		*vpp = NULLVP;
1337 		return (ENXIO);
1338 	}
1339 	if (vfinddev(dev, VCHR, vpp))
1340 		return (0);
1341 	error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1342 	if (error) {
1343 		*vpp = NULLVP;
1344 		return (error);
1345 	}
1346 	vp = nvp;
1347 	vp->v_type = VCHR;
1348 	addalias(vp, dev);
1349 	*vpp = vp;
1350 	return (0);
1351 }
1352 
1353 /*
1354  * Add vnode to the alias list hung off the dev_t.
1355  *
1356  * The reason for this gunk is that multiple vnodes can reference
1357  * the same physical device, so checking vp->v_usecount to see
1358  * how many users there are is inadequate; the v_usecount for
1359  * the vnodes need to be accumulated.  vcount() does that.
1360  */
1361 struct vnode *
1362 addaliasu(nvp, nvp_rdev)
1363 	struct vnode *nvp;
1364 	udev_t nvp_rdev;
1365 {
1366 	struct vnode *ovp;
1367 	vop_t **ops;
1368 	dev_t dev;
1369 
1370 	if (nvp->v_type == VBLK)
1371 		return (nvp);
1372 	if (nvp->v_type != VCHR)
1373 		panic("addaliasu on non-special vnode");
1374 	dev = udev2dev(nvp_rdev, 0);
1375 	/*
1376 	 * Check to see if we have a bdevvp vnode with no associated
1377 	 * filesystem. If so, we want to associate the filesystem of
1378 	 * the new newly instigated vnode with the bdevvp vnode and
1379 	 * discard the newly created vnode rather than leaving the
1380 	 * bdevvp vnode lying around with no associated filesystem.
1381 	 */
1382 	if (vfinddev(dev, nvp->v_type, &ovp) == 0 || ovp->v_data != NULL) {
1383 		addalias(nvp, dev);
1384 		return (nvp);
1385 	}
1386 	/*
1387 	 * Discard unneeded vnode, but save its node specific data.
1388 	 * Note that if there is a lock, it is carried over in the
1389 	 * node specific data to the replacement vnode.
1390 	 */
1391 	vref(ovp);
1392 	ovp->v_data = nvp->v_data;
1393 	ovp->v_tag = nvp->v_tag;
1394 	nvp->v_data = NULL;
1395 	lockinit(&ovp->v_lock, PVFS, nvp->v_lock.lk_wmesg,
1396 	    nvp->v_lock.lk_timo, nvp->v_lock.lk_flags & LK_EXTFLG_MASK);
1397 	if (nvp->v_vnlock)
1398 		ovp->v_vnlock = &ovp->v_lock;
1399 	ops = ovp->v_op;
1400 	ovp->v_op = nvp->v_op;
1401 	if (VOP_ISLOCKED(nvp, curproc)) {
1402 		VOP_UNLOCK(nvp, 0, curproc);
1403 		vn_lock(ovp, LK_EXCLUSIVE | LK_RETRY, curproc);
1404 	}
1405 	nvp->v_op = ops;
1406 	insmntque(ovp, nvp->v_mount);
1407 	vrele(nvp);
1408 	vgone(nvp);
1409 	return (ovp);
1410 }
1411 
1412 /* This is a local helper function that do the same as addaliasu, but for a
1413  * dev_t instead of an udev_t. */
1414 static void
1415 addalias(nvp, dev)
1416 	struct vnode *nvp;
1417 	dev_t dev;
1418 {
1419 
1420 	KASSERT(nvp->v_type == VCHR, ("addalias on non-special vnode"));
1421 	nvp->v_rdev = dev;
1422 	mtx_lock(&spechash_mtx);
1423 	SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
1424 	mtx_unlock(&spechash_mtx);
1425 }
1426 
1427 /*
1428  * Grab a particular vnode from the free list, increment its
1429  * reference count and lock it. The vnode lock bit is set if the
1430  * vnode is being eliminated in vgone. The process is awakened
1431  * when the transition is completed, and an error returned to
1432  * indicate that the vnode is no longer usable (possibly having
1433  * been changed to a new file system type).
1434  */
1435 int
1436 vget(vp, flags, p)
1437 	register struct vnode *vp;
1438 	int flags;
1439 	struct proc *p;
1440 {
1441 	int error;
1442 
1443 	/*
1444 	 * If the vnode is in the process of being cleaned out for
1445 	 * another use, we wait for the cleaning to finish and then
1446 	 * return failure. Cleaning is determined by checking that
1447 	 * the VXLOCK flag is set.
1448 	 */
1449 	if ((flags & LK_INTERLOCK) == 0)
1450 		mtx_lock(&vp->v_interlock);
1451 	if (vp->v_flag & VXLOCK) {
1452 		if (vp->v_vxproc == curproc) {
1453 			printf("VXLOCK interlock avoided\n");
1454 		} else {
1455 			vp->v_flag |= VXWANT;
1456 			msleep((caddr_t)vp, &vp->v_interlock, PINOD | PDROP,
1457 			    "vget", 0);
1458 			return (ENOENT);
1459 		}
1460 	}
1461 
1462 	vp->v_usecount++;
1463 
1464 	if (VSHOULDBUSY(vp))
1465 		vbusy(vp);
1466 	if (flags & LK_TYPE_MASK) {
1467 		if ((error = vn_lock(vp, flags | LK_INTERLOCK, p)) != 0) {
1468 			/*
1469 			 * must expand vrele here because we do not want
1470 			 * to call VOP_INACTIVE if the reference count
1471 			 * drops back to zero since it was never really
1472 			 * active. We must remove it from the free list
1473 			 * before sleeping so that multiple processes do
1474 			 * not try to recycle it.
1475 			 */
1476 			mtx_lock(&vp->v_interlock);
1477 			vp->v_usecount--;
1478 			if (VSHOULDFREE(vp))
1479 				vfree(vp);
1480 			mtx_unlock(&vp->v_interlock);
1481 		}
1482 		return (error);
1483 	}
1484 	mtx_unlock(&vp->v_interlock);
1485 	return (0);
1486 }
1487 
1488 /*
1489  * Increase the reference count of a vnode.
1490  */
1491 void
1492 vref(struct vnode *vp)
1493 {
1494 	mtx_lock(&vp->v_interlock);
1495 	vp->v_usecount++;
1496 	mtx_unlock(&vp->v_interlock);
1497 }
1498 
1499 /*
1500  * Vnode put/release.
1501  * If count drops to zero, call inactive routine and return to freelist.
1502  */
1503 void
1504 vrele(vp)
1505 	struct vnode *vp;
1506 {
1507 	struct proc *p = curproc;	/* XXX */
1508 
1509 	KASSERT(vp != NULL, ("vrele: null vp"));
1510 
1511 	mtx_lock(&vp->v_interlock);
1512 
1513 	/* Skip this v_writecount check if we're going to panic below. */
1514 	KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
1515 	    ("vrele: missed vn_close"));
1516 
1517 	if (vp->v_usecount > 1) {
1518 
1519 		vp->v_usecount--;
1520 		mtx_unlock(&vp->v_interlock);
1521 
1522 		return;
1523 	}
1524 
1525 	if (vp->v_usecount == 1) {
1526 
1527 		vp->v_usecount--;
1528 		if (VSHOULDFREE(vp))
1529 			vfree(vp);
1530 	/*
1531 	 * If we are doing a vput, the node is already locked, and we must
1532 	 * call VOP_INACTIVE with the node locked.  So, in the case of
1533 	 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE.
1534 	 */
1535 		if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, p) == 0) {
1536 			VOP_INACTIVE(vp, p);
1537 		}
1538 
1539 	} else {
1540 #ifdef DIAGNOSTIC
1541 		vprint("vrele: negative ref count", vp);
1542 		mtx_unlock(&vp->v_interlock);
1543 #endif
1544 		panic("vrele: negative ref cnt");
1545 	}
1546 }
1547 
1548 /*
1549  * Release an already locked vnode.  This give the same effects as
1550  * unlock+vrele(), but takes less time and avoids releasing and
1551  * re-aquiring the lock (as vrele() aquires the lock internally.)
1552  */
1553 void
1554 vput(vp)
1555 	struct vnode *vp;
1556 {
1557 	struct proc *p = curproc;	/* XXX */
1558 
1559 	GIANT_REQUIRED;
1560 
1561 	KASSERT(vp != NULL, ("vput: null vp"));
1562 	mtx_lock(&vp->v_interlock);
1563 	/* Skip this v_writecount check if we're going to panic below. */
1564 	KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
1565 	    ("vput: missed vn_close"));
1566 
1567 	if (vp->v_usecount > 1) {
1568 
1569 		vp->v_usecount--;
1570 		VOP_UNLOCK(vp, LK_INTERLOCK, p);
1571 		return;
1572 
1573 	}
1574 
1575 	if (vp->v_usecount == 1) {
1576 
1577 		vp->v_usecount--;
1578 		if (VSHOULDFREE(vp))
1579 			vfree(vp);
1580 	/*
1581 	 * If we are doing a vput, the node is already locked, and we must
1582 	 * call VOP_INACTIVE with the node locked.  So, in the case of
1583 	 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE.
1584 	 */
1585 		mtx_unlock(&vp->v_interlock);
1586 		VOP_INACTIVE(vp, p);
1587 
1588 	} else {
1589 #ifdef DIAGNOSTIC
1590 		vprint("vput: negative ref count", vp);
1591 #endif
1592 		panic("vput: negative ref cnt");
1593 	}
1594 }
1595 
1596 /*
1597  * Somebody doesn't want the vnode recycled.
1598  */
1599 void
1600 vhold(vp)
1601 	register struct vnode *vp;
1602 {
1603 	int s;
1604 
1605   	s = splbio();
1606 	vp->v_holdcnt++;
1607 	if (VSHOULDBUSY(vp))
1608 		vbusy(vp);
1609 	splx(s);
1610 }
1611 
1612 /*
1613  * Note that there is one less who cares about this vnode.  vdrop() is the
1614  * opposite of vhold().
1615  */
1616 void
1617 vdrop(vp)
1618 	register struct vnode *vp;
1619 {
1620 	int s;
1621 
1622 	s = splbio();
1623 	if (vp->v_holdcnt <= 0)
1624 		panic("vdrop: holdcnt");
1625 	vp->v_holdcnt--;
1626 	if (VSHOULDFREE(vp))
1627 		vfree(vp);
1628 	splx(s);
1629 }
1630 
1631 /*
1632  * Remove any vnodes in the vnode table belonging to mount point mp.
1633  *
1634  * If FORCECLOSE is not specified, there should not be any active ones,
1635  * return error if any are found (nb: this is a user error, not a
1636  * system error). If FORCECLOSE is specified, detach any active vnodes
1637  * that are found.
1638  *
1639  * If WRITECLOSE is set, only flush out regular file vnodes open for
1640  * writing.
1641  *
1642  * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
1643  *
1644  * `rootrefs' specifies the base reference count for the root vnode
1645  * of this filesystem. The root vnode is considered busy if its
1646  * v_usecount exceeds this value. On a successful return, vflush()
1647  * will call vrele() on the root vnode exactly rootrefs times.
1648  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
1649  * be zero.
1650  */
1651 #ifdef DIAGNOSTIC
1652 static int busyprt = 0;		/* print out busy vnodes */
1653 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1654 #endif
1655 
1656 int
1657 vflush(mp, rootrefs, flags)
1658 	struct mount *mp;
1659 	int rootrefs;
1660 	int flags;
1661 {
1662 	struct proc *p = curproc;	/* XXX */
1663 	struct vnode *vp, *nvp, *rootvp = NULL;
1664 	int busy = 0, error;
1665 
1666 	if (rootrefs > 0) {
1667 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
1668 		    ("vflush: bad args"));
1669 		/*
1670 		 * Get the filesystem root vnode. We can vput() it
1671 		 * immediately, since with rootrefs > 0, it won't go away.
1672 		 */
1673 		if ((error = VFS_ROOT(mp, &rootvp)) != 0)
1674 			return (error);
1675 		vput(rootvp);
1676 	}
1677 	mtx_lock(&mntvnode_mtx);
1678 loop:
1679 	for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp; vp = nvp) {
1680 		/*
1681 		 * Make sure this vnode wasn't reclaimed in getnewvnode().
1682 		 * Start over if it has (it won't be on the list anymore).
1683 		 */
1684 		if (vp->v_mount != mp)
1685 			goto loop;
1686 		nvp = LIST_NEXT(vp, v_mntvnodes);
1687 
1688 		mtx_unlock(&mntvnode_mtx);
1689 		mtx_lock(&vp->v_interlock);
1690 		/*
1691 		 * Skip over a vnodes marked VSYSTEM.
1692 		 */
1693 		if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1694 			mtx_unlock(&vp->v_interlock);
1695 			mtx_lock(&mntvnode_mtx);
1696 			continue;
1697 		}
1698 		/*
1699 		 * If WRITECLOSE is set, only flush out regular file vnodes
1700 		 * open for writing.
1701 		 */
1702 		if ((flags & WRITECLOSE) &&
1703 		    (vp->v_writecount == 0 || vp->v_type != VREG)) {
1704 			mtx_unlock(&vp->v_interlock);
1705 			mtx_lock(&mntvnode_mtx);
1706 			continue;
1707 		}
1708 
1709 		/*
1710 		 * With v_usecount == 0, all we need to do is clear out the
1711 		 * vnode data structures and we are done.
1712 		 */
1713 		if (vp->v_usecount == 0) {
1714 			vgonel(vp, p);
1715 			mtx_lock(&mntvnode_mtx);
1716 			continue;
1717 		}
1718 
1719 		/*
1720 		 * If FORCECLOSE is set, forcibly close the vnode. For block
1721 		 * or character devices, revert to an anonymous device. For
1722 		 * all other files, just kill them.
1723 		 */
1724 		if (flags & FORCECLOSE) {
1725 			if (vp->v_type != VCHR) {
1726 				vgonel(vp, p);
1727 			} else {
1728 				vclean(vp, 0, p);
1729 				vp->v_op = spec_vnodeop_p;
1730 				insmntque(vp, (struct mount *) 0);
1731 			}
1732 			mtx_lock(&mntvnode_mtx);
1733 			continue;
1734 		}
1735 #ifdef DIAGNOSTIC
1736 		if (busyprt)
1737 			vprint("vflush: busy vnode", vp);
1738 #endif
1739 		mtx_unlock(&vp->v_interlock);
1740 		mtx_lock(&mntvnode_mtx);
1741 		busy++;
1742 	}
1743 	mtx_unlock(&mntvnode_mtx);
1744 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
1745 		/*
1746 		 * If just the root vnode is busy, and if its refcount
1747 		 * is equal to `rootrefs', then go ahead and kill it.
1748 		 */
1749 		mtx_lock(&rootvp->v_interlock);
1750 		KASSERT(busy > 0, ("vflush: not busy"));
1751 		KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
1752 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
1753 			vgonel(rootvp, p);
1754 			busy = 0;
1755 		} else
1756 			mtx_unlock(&rootvp->v_interlock);
1757 	}
1758 	if (busy)
1759 		return (EBUSY);
1760 	for (; rootrefs > 0; rootrefs--)
1761 		vrele(rootvp);
1762 	return (0);
1763 }
1764 
1765 /*
1766  * Disassociate the underlying file system from a vnode.
1767  */
1768 static void
1769 vclean(vp, flags, p)
1770 	struct vnode *vp;
1771 	int flags;
1772 	struct proc *p;
1773 {
1774 	int active;
1775 
1776 	/*
1777 	 * Check to see if the vnode is in use. If so we have to reference it
1778 	 * before we clean it out so that its count cannot fall to zero and
1779 	 * generate a race against ourselves to recycle it.
1780 	 */
1781 	if ((active = vp->v_usecount))
1782 		vp->v_usecount++;
1783 
1784 	/*
1785 	 * Prevent the vnode from being recycled or brought into use while we
1786 	 * clean it out.
1787 	 */
1788 	if (vp->v_flag & VXLOCK)
1789 		panic("vclean: deadlock");
1790 	vp->v_flag |= VXLOCK;
1791 	vp->v_vxproc = curproc;
1792 	/*
1793 	 * Even if the count is zero, the VOP_INACTIVE routine may still
1794 	 * have the object locked while it cleans it out. The VOP_LOCK
1795 	 * ensures that the VOP_INACTIVE routine is done with its work.
1796 	 * For active vnodes, it ensures that no other activity can
1797 	 * occur while the underlying object is being cleaned out.
1798 	 */
1799 	VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, p);
1800 
1801 	/*
1802 	 * Clean out any buffers associated with the vnode.
1803 	 * If the flush fails, just toss the buffers.
1804 	 */
1805 	if (flags & DOCLOSE) {
1806 		if (TAILQ_FIRST(&vp->v_dirtyblkhd) != NULL)
1807 			(void) vn_write_suspend_wait(vp, NULL, V_WAIT);
1808 		if (vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0) != 0)
1809 			vinvalbuf(vp, 0, NOCRED, p, 0, 0);
1810 	}
1811 
1812 	VOP_DESTROYVOBJECT(vp);
1813 
1814 	/*
1815 	 * If purging an active vnode, it must be closed and
1816 	 * deactivated before being reclaimed. Note that the
1817 	 * VOP_INACTIVE will unlock the vnode.
1818 	 */
1819 	if (active) {
1820 		if (flags & DOCLOSE)
1821 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, p);
1822 		VOP_INACTIVE(vp, p);
1823 	} else {
1824 		/*
1825 		 * Any other processes trying to obtain this lock must first
1826 		 * wait for VXLOCK to clear, then call the new lock operation.
1827 		 */
1828 		VOP_UNLOCK(vp, 0, p);
1829 	}
1830 	/*
1831 	 * Reclaim the vnode.
1832 	 */
1833 	if (VOP_RECLAIM(vp, p))
1834 		panic("vclean: cannot reclaim");
1835 
1836 	if (active) {
1837 		/*
1838 		 * Inline copy of vrele() since VOP_INACTIVE
1839 		 * has already been called.
1840 		 */
1841 		mtx_lock(&vp->v_interlock);
1842 		if (--vp->v_usecount <= 0) {
1843 #ifdef DIAGNOSTIC
1844 			if (vp->v_usecount < 0 || vp->v_writecount != 0) {
1845 				vprint("vclean: bad ref count", vp);
1846 				panic("vclean: ref cnt");
1847 			}
1848 #endif
1849 			vfree(vp);
1850 		}
1851 		mtx_unlock(&vp->v_interlock);
1852 	}
1853 
1854 	cache_purge(vp);
1855 	vp->v_vnlock = NULL;
1856 	lockdestroy(&vp->v_lock);
1857 
1858 	if (VSHOULDFREE(vp))
1859 		vfree(vp);
1860 
1861 	/*
1862 	 * Done with purge, notify sleepers of the grim news.
1863 	 */
1864 	vp->v_op = dead_vnodeop_p;
1865 	vn_pollgone(vp);
1866 	vp->v_tag = VT_NON;
1867 	vp->v_flag &= ~VXLOCK;
1868 	vp->v_vxproc = NULL;
1869 	if (vp->v_flag & VXWANT) {
1870 		vp->v_flag &= ~VXWANT;
1871 		wakeup((caddr_t) vp);
1872 	}
1873 }
1874 
1875 /*
1876  * Eliminate all activity associated with the requested vnode
1877  * and with all vnodes aliased to the requested vnode.
1878  */
1879 int
1880 vop_revoke(ap)
1881 	struct vop_revoke_args /* {
1882 		struct vnode *a_vp;
1883 		int a_flags;
1884 	} */ *ap;
1885 {
1886 	struct vnode *vp, *vq;
1887 	dev_t dev;
1888 
1889 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
1890 
1891 	vp = ap->a_vp;
1892 	/*
1893 	 * If a vgone (or vclean) is already in progress,
1894 	 * wait until it is done and return.
1895 	 */
1896 	if (vp->v_flag & VXLOCK) {
1897 		vp->v_flag |= VXWANT;
1898 		msleep((caddr_t)vp, &vp->v_interlock, PINOD | PDROP,
1899 		    "vop_revokeall", 0);
1900 		return (0);
1901 	}
1902 	dev = vp->v_rdev;
1903 	for (;;) {
1904 		mtx_lock(&spechash_mtx);
1905 		vq = SLIST_FIRST(&dev->si_hlist);
1906 		mtx_unlock(&spechash_mtx);
1907 		if (!vq)
1908 			break;
1909 		vgone(vq);
1910 	}
1911 	return (0);
1912 }
1913 
1914 /*
1915  * Recycle an unused vnode to the front of the free list.
1916  * Release the passed interlock if the vnode will be recycled.
1917  */
1918 int
1919 vrecycle(vp, inter_lkp, p)
1920 	struct vnode *vp;
1921 	struct mtx *inter_lkp;
1922 	struct proc *p;
1923 {
1924 
1925 	mtx_lock(&vp->v_interlock);
1926 	if (vp->v_usecount == 0) {
1927 		if (inter_lkp) {
1928 			mtx_unlock(inter_lkp);
1929 		}
1930 		vgonel(vp, p);
1931 		return (1);
1932 	}
1933 	mtx_unlock(&vp->v_interlock);
1934 	return (0);
1935 }
1936 
1937 /*
1938  * Eliminate all activity associated with a vnode
1939  * in preparation for reuse.
1940  */
1941 void
1942 vgone(vp)
1943 	register struct vnode *vp;
1944 {
1945 	struct proc *p = curproc;	/* XXX */
1946 
1947 	mtx_lock(&vp->v_interlock);
1948 	vgonel(vp, p);
1949 }
1950 
1951 /*
1952  * vgone, with the vp interlock held.
1953  */
1954 void
1955 vgonel(vp, p)
1956 	struct vnode *vp;
1957 	struct proc *p;
1958 {
1959 	int s;
1960 
1961 	/*
1962 	 * If a vgone (or vclean) is already in progress,
1963 	 * wait until it is done and return.
1964 	 */
1965 	if (vp->v_flag & VXLOCK) {
1966 		vp->v_flag |= VXWANT;
1967 		msleep((caddr_t)vp, &vp->v_interlock, PINOD | PDROP,
1968 		    "vgone", 0);
1969 		return;
1970 	}
1971 
1972 	/*
1973 	 * Clean out the filesystem specific data.
1974 	 */
1975 	vclean(vp, DOCLOSE, p);
1976 	mtx_lock(&vp->v_interlock);
1977 
1978 	/*
1979 	 * Delete from old mount point vnode list, if on one.
1980 	 */
1981 	if (vp->v_mount != NULL)
1982 		insmntque(vp, (struct mount *)0);
1983 	/*
1984 	 * If special device, remove it from special device alias list
1985 	 * if it is on one.
1986 	 */
1987 	if (vp->v_type == VCHR && vp->v_rdev != NULL && vp->v_rdev != NODEV) {
1988 		mtx_lock(&spechash_mtx);
1989 		SLIST_REMOVE(&vp->v_rdev->si_hlist, vp, vnode, v_specnext);
1990 		freedev(vp->v_rdev);
1991 		mtx_unlock(&spechash_mtx);
1992 		vp->v_rdev = NULL;
1993 	}
1994 
1995 	/*
1996 	 * If it is on the freelist and not already at the head,
1997 	 * move it to the head of the list. The test of the
1998 	 * VDOOMED flag and the reference count of zero is because
1999 	 * it will be removed from the free list by getnewvnode,
2000 	 * but will not have its reference count incremented until
2001 	 * after calling vgone. If the reference count were
2002 	 * incremented first, vgone would (incorrectly) try to
2003 	 * close the previous instance of the underlying object.
2004 	 */
2005 	if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
2006 		s = splbio();
2007 		mtx_lock(&vnode_free_list_mtx);
2008 		if (vp->v_flag & VFREE)
2009 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2010 		else
2011 			freevnodes++;
2012 		vp->v_flag |= VFREE;
2013 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2014 		mtx_unlock(&vnode_free_list_mtx);
2015 		splx(s);
2016 	}
2017 
2018 	vp->v_type = VBAD;
2019 	mtx_unlock(&vp->v_interlock);
2020 }
2021 
2022 /*
2023  * Lookup a vnode by device number.
2024  */
2025 int
2026 vfinddev(dev, type, vpp)
2027 	dev_t dev;
2028 	enum vtype type;
2029 	struct vnode **vpp;
2030 {
2031 	struct vnode *vp;
2032 
2033 	mtx_lock(&spechash_mtx);
2034 	SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2035 		if (type == vp->v_type) {
2036 			*vpp = vp;
2037 			mtx_unlock(&spechash_mtx);
2038 			return (1);
2039 		}
2040 	}
2041 	mtx_unlock(&spechash_mtx);
2042 	return (0);
2043 }
2044 
2045 /*
2046  * Calculate the total number of references to a special device.
2047  */
2048 int
2049 vcount(vp)
2050 	struct vnode *vp;
2051 {
2052 	struct vnode *vq;
2053 	int count;
2054 
2055 	count = 0;
2056 	mtx_lock(&spechash_mtx);
2057 	SLIST_FOREACH(vq, &vp->v_rdev->si_hlist, v_specnext)
2058 		count += vq->v_usecount;
2059 	mtx_unlock(&spechash_mtx);
2060 	return (count);
2061 }
2062 
2063 /*
2064  * Same as above, but using the dev_t as argument
2065  */
2066 int
2067 count_dev(dev)
2068 	dev_t dev;
2069 {
2070 	struct vnode *vp;
2071 
2072 	vp = SLIST_FIRST(&dev->si_hlist);
2073 	if (vp == NULL)
2074 		return (0);
2075 	return(vcount(vp));
2076 }
2077 
2078 /*
2079  * Print out a description of a vnode.
2080  */
2081 static char *typename[] =
2082 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2083 
2084 void
2085 vprint(label, vp)
2086 	char *label;
2087 	struct vnode *vp;
2088 {
2089 	char buf[96];
2090 
2091 	if (label != NULL)
2092 		printf("%s: %p: ", label, (void *)vp);
2093 	else
2094 		printf("%p: ", (void *)vp);
2095 	printf("type %s, usecount %d, writecount %d, refcount %d,",
2096 	    typename[vp->v_type], vp->v_usecount, vp->v_writecount,
2097 	    vp->v_holdcnt);
2098 	buf[0] = '\0';
2099 	if (vp->v_flag & VROOT)
2100 		strcat(buf, "|VROOT");
2101 	if (vp->v_flag & VTEXT)
2102 		strcat(buf, "|VTEXT");
2103 	if (vp->v_flag & VSYSTEM)
2104 		strcat(buf, "|VSYSTEM");
2105 	if (vp->v_flag & VXLOCK)
2106 		strcat(buf, "|VXLOCK");
2107 	if (vp->v_flag & VXWANT)
2108 		strcat(buf, "|VXWANT");
2109 	if (vp->v_flag & VBWAIT)
2110 		strcat(buf, "|VBWAIT");
2111 	if (vp->v_flag & VDOOMED)
2112 		strcat(buf, "|VDOOMED");
2113 	if (vp->v_flag & VFREE)
2114 		strcat(buf, "|VFREE");
2115 	if (vp->v_flag & VOBJBUF)
2116 		strcat(buf, "|VOBJBUF");
2117 	if (buf[0] != '\0')
2118 		printf(" flags (%s)", &buf[1]);
2119 	if (vp->v_data == NULL) {
2120 		printf("\n");
2121 	} else {
2122 		printf("\n\t");
2123 		VOP_PRINT(vp);
2124 	}
2125 }
2126 
2127 #ifdef DDB
2128 #include <ddb/ddb.h>
2129 /*
2130  * List all of the locked vnodes in the system.
2131  * Called when debugging the kernel.
2132  */
2133 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
2134 {
2135 	struct proc *p = curproc;	/* XXX */
2136 	struct mount *mp, *nmp;
2137 	struct vnode *vp;
2138 
2139 	printf("Locked vnodes\n");
2140 	mtx_lock(&mountlist_mtx);
2141 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2142 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, p)) {
2143 			nmp = TAILQ_NEXT(mp, mnt_list);
2144 			continue;
2145 		}
2146 		mtx_lock(&mntvnode_mtx);
2147 		LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2148 			if (VOP_ISLOCKED(vp, NULL))
2149 				vprint((char *)0, vp);
2150 		}
2151 		mtx_unlock(&mntvnode_mtx);
2152 		mtx_lock(&mountlist_mtx);
2153 		nmp = TAILQ_NEXT(mp, mnt_list);
2154 		vfs_unbusy(mp, p);
2155 	}
2156 	mtx_unlock(&mountlist_mtx);
2157 }
2158 #endif
2159 
2160 /*
2161  * Top level filesystem related information gathering.
2162  */
2163 static int	sysctl_ovfs_conf __P((SYSCTL_HANDLER_ARGS));
2164 
2165 static int
2166 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2167 {
2168 	int *name = (int *)arg1 - 1;	/* XXX */
2169 	u_int namelen = arg2 + 1;	/* XXX */
2170 	struct vfsconf *vfsp;
2171 
2172 #if 1 || defined(COMPAT_PRELITE2)
2173 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2174 	if (namelen == 1)
2175 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2176 #endif
2177 
2178 	/* XXX the below code does not compile; vfs_sysctl does not exist. */
2179 #ifdef notyet
2180 	/* all sysctl names at this level are at least name and field */
2181 	if (namelen < 2)
2182 		return (ENOTDIR);		/* overloaded */
2183 	if (name[0] != VFS_GENERIC) {
2184 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2185 			if (vfsp->vfc_typenum == name[0])
2186 				break;
2187 		if (vfsp == NULL)
2188 			return (EOPNOTSUPP);
2189 		return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2190 		    oldp, oldlenp, newp, newlen, p));
2191 	}
2192 #endif
2193 	switch (name[1]) {
2194 	case VFS_MAXTYPENUM:
2195 		if (namelen != 2)
2196 			return (ENOTDIR);
2197 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2198 	case VFS_CONF:
2199 		if (namelen != 3)
2200 			return (ENOTDIR);	/* overloaded */
2201 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2202 			if (vfsp->vfc_typenum == name[2])
2203 				break;
2204 		if (vfsp == NULL)
2205 			return (EOPNOTSUPP);
2206 		return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2207 	}
2208 	return (EOPNOTSUPP);
2209 }
2210 
2211 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2212 	"Generic filesystem");
2213 
2214 #if 1 || defined(COMPAT_PRELITE2)
2215 
2216 static int
2217 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2218 {
2219 	int error;
2220 	struct vfsconf *vfsp;
2221 	struct ovfsconf ovfs;
2222 
2223 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2224 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
2225 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
2226 		ovfs.vfc_index = vfsp->vfc_typenum;
2227 		ovfs.vfc_refcount = vfsp->vfc_refcount;
2228 		ovfs.vfc_flags = vfsp->vfc_flags;
2229 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2230 		if (error)
2231 			return error;
2232 	}
2233 	return 0;
2234 }
2235 
2236 #endif /* 1 || COMPAT_PRELITE2 */
2237 
2238 #if COMPILING_LINT
2239 #define KINFO_VNODESLOP	10
2240 /*
2241  * Dump vnode list (via sysctl).
2242  * Copyout address of vnode followed by vnode.
2243  */
2244 /* ARGSUSED */
2245 static int
2246 sysctl_vnode(SYSCTL_HANDLER_ARGS)
2247 {
2248 	struct proc *p = curproc;	/* XXX */
2249 	struct mount *mp, *nmp;
2250 	struct vnode *nvp, *vp;
2251 	int error;
2252 
2253 #define VPTRSZ	sizeof (struct vnode *)
2254 #define VNODESZ	sizeof (struct vnode)
2255 
2256 	req->lock = 0;
2257 	if (!req->oldptr) /* Make an estimate */
2258 		return (SYSCTL_OUT(req, 0,
2259 			(numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2260 
2261 	mtx_lock(&mountlist_mtx);
2262 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2263 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, p)) {
2264 			nmp = TAILQ_NEXT(mp, mnt_list);
2265 			continue;
2266 		}
2267 		mtx_lock(&mntvnode_mtx);
2268 again:
2269 		for (vp = LIST_FIRST(&mp->mnt_vnodelist);
2270 		     vp != NULL;
2271 		     vp = nvp) {
2272 			/*
2273 			 * Check that the vp is still associated with
2274 			 * this filesystem.  RACE: could have been
2275 			 * recycled onto the same filesystem.
2276 			 */
2277 			if (vp->v_mount != mp)
2278 				goto again;
2279 			nvp = LIST_NEXT(vp, v_mntvnodes);
2280 			mtx_unlock(&mntvnode_mtx);
2281 			if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2282 			    (error = SYSCTL_OUT(req, vp, VNODESZ)))
2283 				return (error);
2284 			mtx_lock(&mntvnode_mtx);
2285 		}
2286 		mtx_unlock(&mntvnode_mtx);
2287 		mtx_lock(&mountlist_mtx);
2288 		nmp = TAILQ_NEXT(mp, mnt_list);
2289 		vfs_unbusy(mp, p);
2290 	}
2291 	mtx_unlock(&mountlist_mtx);
2292 
2293 	return (0);
2294 }
2295 
2296 /*
2297  * XXX
2298  * Exporting the vnode list on large systems causes them to crash.
2299  * Exporting the vnode list on medium systems causes sysctl to coredump.
2300  */
2301 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2302 	0, 0, sysctl_vnode, "S,vnode", "");
2303 #endif
2304 
2305 /*
2306  * Check to see if a filesystem is mounted on a block device.
2307  */
2308 int
2309 vfs_mountedon(vp)
2310 	struct vnode *vp;
2311 {
2312 
2313 	if (vp->v_rdev->si_mountpoint != NULL)
2314 		return (EBUSY);
2315 	return (0);
2316 }
2317 
2318 /*
2319  * Unmount all filesystems. The list is traversed in reverse order
2320  * of mounting to avoid dependencies.
2321  */
2322 void
2323 vfs_unmountall()
2324 {
2325 	struct mount *mp;
2326 	struct proc *p;
2327 	int error;
2328 
2329 	if (curproc != NULL)
2330 		p = curproc;
2331 	else
2332 		p = initproc;	/* XXX XXX should this be proc0? */
2333 	/*
2334 	 * Since this only runs when rebooting, it is not interlocked.
2335 	 */
2336 	while(!TAILQ_EMPTY(&mountlist)) {
2337 		mp = TAILQ_LAST(&mountlist, mntlist);
2338 		error = dounmount(mp, MNT_FORCE, p);
2339 		if (error) {
2340 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
2341 			printf("unmount of %s failed (",
2342 			    mp->mnt_stat.f_mntonname);
2343 			if (error == EBUSY)
2344 				printf("BUSY)\n");
2345 			else
2346 				printf("%d)\n", error);
2347 		} else {
2348 			/* The unmount has removed mp from the mountlist */
2349 		}
2350 	}
2351 }
2352 
2353 /*
2354  * perform msync on all vnodes under a mount point
2355  * the mount point must be locked.
2356  */
2357 void
2358 vfs_msync(struct mount *mp, int flags) {
2359 	struct vnode *vp, *nvp;
2360 	struct vm_object *obj;
2361 	int anyio, tries;
2362 
2363 	GIANT_REQUIRED;
2364 
2365 	tries = 5;
2366 loop:
2367 	anyio = 0;
2368 	mtx_lock(&mntvnode_mtx);
2369 	for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp != NULL; vp = nvp) {
2370 
2371 		nvp = LIST_NEXT(vp, v_mntvnodes);
2372 
2373 		if (vp->v_mount != mp) {
2374 			mtx_unlock(&mntvnode_mtx);
2375 			goto loop;
2376 		}
2377 
2378 		if (vp->v_flag & VXLOCK)	/* XXX: what if MNT_WAIT? */
2379 			continue;
2380 
2381 		if (vp->v_flag & VNOSYNC)	/* unlinked, skip it */
2382 			continue;
2383 
2384 		if (flags != MNT_WAIT) {
2385 			if (VOP_GETVOBJECT(vp, &obj) != 0 ||
2386 			    (obj->flags & OBJ_MIGHTBEDIRTY) == 0)
2387 				continue;
2388 			if (VOP_ISLOCKED(vp, NULL))
2389 				continue;
2390 		}
2391 
2392 		mtx_unlock(&mntvnode_mtx);
2393 		mtx_lock(&vp->v_interlock);
2394 		if (VOP_GETVOBJECT(vp, &obj) == 0 &&
2395 		    (obj->flags & OBJ_MIGHTBEDIRTY)) {
2396 			if (!vget(vp,
2397 				LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, curproc)) {
2398 				if (VOP_GETVOBJECT(vp, &obj) == 0) {
2399 					vm_object_page_clean(obj, 0, 0,
2400 					    flags == MNT_WAIT ?
2401 					    OBJPC_SYNC : OBJPC_NOSYNC);
2402 					anyio = 1;
2403 				}
2404 				vput(vp);
2405 			}
2406 		} else {
2407 			mtx_unlock(&vp->v_interlock);
2408 		}
2409 		mtx_lock(&mntvnode_mtx);
2410 	}
2411 	mtx_unlock(&mntvnode_mtx);
2412 	if (anyio && (--tries > 0))
2413 		goto loop;
2414 }
2415 
2416 /*
2417  * Create the VM object needed for VMIO and mmap support.  This
2418  * is done for all VREG files in the system.  Some filesystems might
2419  * afford the additional metadata buffering capability of the
2420  * VMIO code by making the device node be VMIO mode also.
2421  *
2422  * vp must be locked when vfs_object_create is called.
2423  */
2424 int
2425 vfs_object_create(vp, p, cred)
2426 	struct vnode *vp;
2427 	struct proc *p;
2428 	struct ucred *cred;
2429 {
2430 	GIANT_REQUIRED;
2431 	return (VOP_CREATEVOBJECT(vp, cred, p));
2432 }
2433 
2434 /*
2435  * Mark a vnode as free, putting it up for recycling.
2436  */
2437 void
2438 vfree(vp)
2439 	struct vnode *vp;
2440 {
2441 	int s;
2442 
2443 	s = splbio();
2444 	mtx_lock(&vnode_free_list_mtx);
2445 	KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free"));
2446 	if (vp->v_flag & VAGE) {
2447 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2448 	} else {
2449 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2450 	}
2451 	freevnodes++;
2452 	mtx_unlock(&vnode_free_list_mtx);
2453 	vp->v_flag &= ~VAGE;
2454 	vp->v_flag |= VFREE;
2455 	splx(s);
2456 }
2457 
2458 /*
2459  * Opposite of vfree() - mark a vnode as in use.
2460  */
2461 void
2462 vbusy(vp)
2463 	struct vnode *vp;
2464 {
2465 	int s;
2466 
2467 	s = splbio();
2468 	mtx_lock(&vnode_free_list_mtx);
2469 	KASSERT((vp->v_flag & VFREE) != 0, ("vnode not free"));
2470 	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2471 	freevnodes--;
2472 	mtx_unlock(&vnode_free_list_mtx);
2473 	vp->v_flag &= ~(VFREE|VAGE);
2474 	splx(s);
2475 }
2476 
2477 /*
2478  * Record a process's interest in events which might happen to
2479  * a vnode.  Because poll uses the historic select-style interface
2480  * internally, this routine serves as both the ``check for any
2481  * pending events'' and the ``record my interest in future events''
2482  * functions.  (These are done together, while the lock is held,
2483  * to avoid race conditions.)
2484  */
2485 int
2486 vn_pollrecord(vp, p, events)
2487 	struct vnode *vp;
2488 	struct proc *p;
2489 	short events;
2490 {
2491 	mtx_lock(&vp->v_pollinfo.vpi_lock);
2492 	if (vp->v_pollinfo.vpi_revents & events) {
2493 		/*
2494 		 * This leaves events we are not interested
2495 		 * in available for the other process which
2496 		 * which presumably had requested them
2497 		 * (otherwise they would never have been
2498 		 * recorded).
2499 		 */
2500 		events &= vp->v_pollinfo.vpi_revents;
2501 		vp->v_pollinfo.vpi_revents &= ~events;
2502 
2503 		mtx_unlock(&vp->v_pollinfo.vpi_lock);
2504 		return events;
2505 	}
2506 	vp->v_pollinfo.vpi_events |= events;
2507 	selrecord(p, &vp->v_pollinfo.vpi_selinfo);
2508 	mtx_unlock(&vp->v_pollinfo.vpi_lock);
2509 	return 0;
2510 }
2511 
2512 /*
2513  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
2514  * it is possible for us to miss an event due to race conditions, but
2515  * that condition is expected to be rare, so for the moment it is the
2516  * preferred interface.
2517  */
2518 void
2519 vn_pollevent(vp, events)
2520 	struct vnode *vp;
2521 	short events;
2522 {
2523 	mtx_lock(&vp->v_pollinfo.vpi_lock);
2524 	if (vp->v_pollinfo.vpi_events & events) {
2525 		/*
2526 		 * We clear vpi_events so that we don't
2527 		 * call selwakeup() twice if two events are
2528 		 * posted before the polling process(es) is
2529 		 * awakened.  This also ensures that we take at
2530 		 * most one selwakeup() if the polling process
2531 		 * is no longer interested.  However, it does
2532 		 * mean that only one event can be noticed at
2533 		 * a time.  (Perhaps we should only clear those
2534 		 * event bits which we note?) XXX
2535 		 */
2536 		vp->v_pollinfo.vpi_events = 0;	/* &= ~events ??? */
2537 		vp->v_pollinfo.vpi_revents |= events;
2538 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2539 	}
2540 	mtx_unlock(&vp->v_pollinfo.vpi_lock);
2541 }
2542 
2543 #define VN_KNOTE(vp, b) \
2544 	KNOTE((struct klist *)&vp->v_pollinfo.vpi_selinfo.si_note, (b))
2545 
2546 /*
2547  * Wake up anyone polling on vp because it is being revoked.
2548  * This depends on dead_poll() returning POLLHUP for correct
2549  * behavior.
2550  */
2551 void
2552 vn_pollgone(vp)
2553 	struct vnode *vp;
2554 {
2555 	mtx_lock(&vp->v_pollinfo.vpi_lock);
2556         VN_KNOTE(vp, NOTE_REVOKE);
2557 	if (vp->v_pollinfo.vpi_events) {
2558 		vp->v_pollinfo.vpi_events = 0;
2559 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2560 	}
2561 	mtx_unlock(&vp->v_pollinfo.vpi_lock);
2562 }
2563 
2564 
2565 
2566 /*
2567  * Routine to create and manage a filesystem syncer vnode.
2568  */
2569 #define sync_close ((int (*) __P((struct  vop_close_args *)))nullop)
2570 static int	sync_fsync __P((struct  vop_fsync_args *));
2571 static int	sync_inactive __P((struct  vop_inactive_args *));
2572 static int	sync_reclaim  __P((struct  vop_reclaim_args *));
2573 #define sync_lock ((int (*) __P((struct  vop_lock_args *)))vop_nolock)
2574 #define sync_unlock ((int (*) __P((struct  vop_unlock_args *)))vop_nounlock)
2575 static int	sync_print __P((struct vop_print_args *));
2576 #define sync_islocked ((int(*) __P((struct vop_islocked_args *)))vop_noislocked)
2577 
2578 static vop_t **sync_vnodeop_p;
2579 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
2580 	{ &vop_default_desc,	(vop_t *) vop_eopnotsupp },
2581 	{ &vop_close_desc,	(vop_t *) sync_close },		/* close */
2582 	{ &vop_fsync_desc,	(vop_t *) sync_fsync },		/* fsync */
2583 	{ &vop_inactive_desc,	(vop_t *) sync_inactive },	/* inactive */
2584 	{ &vop_reclaim_desc,	(vop_t *) sync_reclaim },	/* reclaim */
2585 	{ &vop_lock_desc,	(vop_t *) sync_lock },		/* lock */
2586 	{ &vop_unlock_desc,	(vop_t *) sync_unlock },	/* unlock */
2587 	{ &vop_print_desc,	(vop_t *) sync_print },		/* print */
2588 	{ &vop_islocked_desc,	(vop_t *) sync_islocked },	/* islocked */
2589 	{ NULL, NULL }
2590 };
2591 static struct vnodeopv_desc sync_vnodeop_opv_desc =
2592 	{ &sync_vnodeop_p, sync_vnodeop_entries };
2593 
2594 VNODEOP_SET(sync_vnodeop_opv_desc);
2595 
2596 /*
2597  * Create a new filesystem syncer vnode for the specified mount point.
2598  */
2599 int
2600 vfs_allocate_syncvnode(mp)
2601 	struct mount *mp;
2602 {
2603 	struct vnode *vp;
2604 	static long start, incr, next;
2605 	int error;
2606 
2607 	/* Allocate a new vnode */
2608 	if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
2609 		mp->mnt_syncer = NULL;
2610 		return (error);
2611 	}
2612 	vp->v_type = VNON;
2613 	/*
2614 	 * Place the vnode onto the syncer worklist. We attempt to
2615 	 * scatter them about on the list so that they will go off
2616 	 * at evenly distributed times even if all the filesystems
2617 	 * are mounted at once.
2618 	 */
2619 	next += incr;
2620 	if (next == 0 || next > syncer_maxdelay) {
2621 		start /= 2;
2622 		incr /= 2;
2623 		if (start == 0) {
2624 			start = syncer_maxdelay / 2;
2625 			incr = syncer_maxdelay;
2626 		}
2627 		next = start;
2628 	}
2629 	vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
2630 	mp->mnt_syncer = vp;
2631 	return (0);
2632 }
2633 
2634 /*
2635  * Do a lazy sync of the filesystem.
2636  */
2637 static int
2638 sync_fsync(ap)
2639 	struct vop_fsync_args /* {
2640 		struct vnode *a_vp;
2641 		struct ucred *a_cred;
2642 		int a_waitfor;
2643 		struct proc *a_p;
2644 	} */ *ap;
2645 {
2646 	struct vnode *syncvp = ap->a_vp;
2647 	struct mount *mp = syncvp->v_mount;
2648 	struct proc *p = ap->a_p;
2649 	int asyncflag;
2650 
2651 	/*
2652 	 * We only need to do something if this is a lazy evaluation.
2653 	 */
2654 	if (ap->a_waitfor != MNT_LAZY)
2655 		return (0);
2656 
2657 	/*
2658 	 * Move ourselves to the back of the sync list.
2659 	 */
2660 	vn_syncer_add_to_worklist(syncvp, syncdelay);
2661 
2662 	/*
2663 	 * Walk the list of vnodes pushing all that are dirty and
2664 	 * not already on the sync list.
2665 	 */
2666 	mtx_lock(&mountlist_mtx);
2667 	if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, p) != 0) {
2668 		mtx_unlock(&mountlist_mtx);
2669 		return (0);
2670 	}
2671 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
2672 		vfs_unbusy(mp, p);
2673 		return (0);
2674 	}
2675 	asyncflag = mp->mnt_flag & MNT_ASYNC;
2676 	mp->mnt_flag &= ~MNT_ASYNC;
2677 	vfs_msync(mp, MNT_NOWAIT);
2678 	VFS_SYNC(mp, MNT_LAZY, ap->a_cred, p);
2679 	if (asyncflag)
2680 		mp->mnt_flag |= MNT_ASYNC;
2681 	vn_finished_write(mp);
2682 	vfs_unbusy(mp, p);
2683 	return (0);
2684 }
2685 
2686 /*
2687  * The syncer vnode is no referenced.
2688  */
2689 static int
2690 sync_inactive(ap)
2691 	struct vop_inactive_args /* {
2692 		struct vnode *a_vp;
2693 		struct proc *a_p;
2694 	} */ *ap;
2695 {
2696 
2697 	vgone(ap->a_vp);
2698 	return (0);
2699 }
2700 
2701 /*
2702  * The syncer vnode is no longer needed and is being decommissioned.
2703  *
2704  * Modifications to the worklist must be protected at splbio().
2705  */
2706 static int
2707 sync_reclaim(ap)
2708 	struct vop_reclaim_args /* {
2709 		struct vnode *a_vp;
2710 	} */ *ap;
2711 {
2712 	struct vnode *vp = ap->a_vp;
2713 	int s;
2714 
2715 	s = splbio();
2716 	vp->v_mount->mnt_syncer = NULL;
2717 	if (vp->v_flag & VONWORKLST) {
2718 		LIST_REMOVE(vp, v_synclist);
2719 		vp->v_flag &= ~VONWORKLST;
2720 	}
2721 	splx(s);
2722 
2723 	return (0);
2724 }
2725 
2726 /*
2727  * Print out a syncer vnode.
2728  */
2729 static int
2730 sync_print(ap)
2731 	struct vop_print_args /* {
2732 		struct vnode *a_vp;
2733 	} */ *ap;
2734 {
2735 	struct vnode *vp = ap->a_vp;
2736 
2737 	printf("syncer vnode");
2738 	if (vp->v_vnlock != NULL)
2739 		lockmgr_printinfo(vp->v_vnlock);
2740 	printf("\n");
2741 	return (0);
2742 }
2743 
2744 /*
2745  * extract the dev_t from a VCHR
2746  */
2747 dev_t
2748 vn_todev(vp)
2749 	struct vnode *vp;
2750 {
2751 	if (vp->v_type != VCHR)
2752 		return (NODEV);
2753 	return (vp->v_rdev);
2754 }
2755 
2756 /*
2757  * Check if vnode represents a disk device
2758  */
2759 int
2760 vn_isdisk(vp, errp)
2761 	struct vnode *vp;
2762 	int *errp;
2763 {
2764 	struct cdevsw *cdevsw;
2765 
2766 	if (vp->v_type != VCHR) {
2767 		if (errp != NULL)
2768 			*errp = ENOTBLK;
2769 		return (0);
2770 	}
2771 	if (vp->v_rdev == NULL) {
2772 		if (errp != NULL)
2773 			*errp = ENXIO;
2774 		return (0);
2775 	}
2776 	cdevsw = devsw(vp->v_rdev);
2777 	if (cdevsw == NULL) {
2778 		if (errp != NULL)
2779 			*errp = ENXIO;
2780 		return (0);
2781 	}
2782 	if (!(cdevsw->d_flags & D_DISK)) {
2783 		if (errp != NULL)
2784 			*errp = ENOTBLK;
2785 		return (0);
2786 	}
2787 	if (errp != NULL)
2788 		*errp = 0;
2789 	return (1);
2790 }
2791 
2792 /*
2793  * Free data allocated by namei(); see namei(9) for details.
2794  */
2795 void
2796 NDFREE(ndp, flags)
2797      struct nameidata *ndp;
2798      const uint flags;
2799 {
2800 	if (!(flags & NDF_NO_FREE_PNBUF) &&
2801 	    (ndp->ni_cnd.cn_flags & HASBUF)) {
2802 		zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
2803 		ndp->ni_cnd.cn_flags &= ~HASBUF;
2804 	}
2805 	if (!(flags & NDF_NO_DVP_UNLOCK) &&
2806 	    (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
2807 	    ndp->ni_dvp != ndp->ni_vp)
2808 		VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_proc);
2809 	if (!(flags & NDF_NO_DVP_RELE) &&
2810 	    (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
2811 		vrele(ndp->ni_dvp);
2812 		ndp->ni_dvp = NULL;
2813 	}
2814 	if (!(flags & NDF_NO_VP_UNLOCK) &&
2815 	    (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
2816 		VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_proc);
2817 	if (!(flags & NDF_NO_VP_RELE) &&
2818 	    ndp->ni_vp) {
2819 		vrele(ndp->ni_vp);
2820 		ndp->ni_vp = NULL;
2821 	}
2822 	if (!(flags & NDF_NO_STARTDIR_RELE) &&
2823 	    (ndp->ni_cnd.cn_flags & SAVESTART)) {
2824 		vrele(ndp->ni_startdir);
2825 		ndp->ni_startdir = NULL;
2826 	}
2827 }
2828 
2829 /*
2830  * Common file system object access control check routine.  Accepts a
2831  * vnode's type, "mode", uid and gid, requested access mode, credentials,
2832  * and optional call-by-reference privused argument allowing vaccess()
2833  * to indicate to the caller whether privilege was used to satisfy the
2834  * request.  Returns 0 on success, or an errno on failure.
2835  */
2836 int
2837 vaccess(type, file_mode, file_uid, file_gid, acc_mode, cred, privused)
2838 	enum vtype type;
2839 	mode_t file_mode;
2840 	uid_t file_uid;
2841 	gid_t file_gid;
2842 	mode_t acc_mode;
2843 	struct ucred *cred;
2844 	int *privused;
2845 {
2846 	mode_t dac_granted;
2847 #ifdef CAPABILITIES
2848 	mode_t cap_granted;
2849 #endif
2850 
2851 	/*
2852 	 * Look for a normal, non-privileged way to access the file/directory
2853 	 * as requested.  If it exists, go with that.
2854 	 */
2855 
2856 	if (privused != NULL)
2857 		*privused = 0;
2858 
2859 	dac_granted = 0;
2860 
2861 	/* Check the owner. */
2862 	if (cred->cr_uid == file_uid) {
2863 		dac_granted |= VADMIN;
2864 		if (file_mode & S_IXUSR)
2865 			dac_granted |= VEXEC;
2866 		if (file_mode & S_IRUSR)
2867 			dac_granted |= VREAD;
2868 		if (file_mode & S_IWUSR)
2869 			dac_granted |= VWRITE;
2870 
2871 		if ((acc_mode & dac_granted) == acc_mode)
2872 			return (0);
2873 
2874 		goto privcheck;
2875 	}
2876 
2877 	/* Otherwise, check the groups (first match) */
2878 	if (groupmember(file_gid, cred)) {
2879 		if (file_mode & S_IXGRP)
2880 			dac_granted |= VEXEC;
2881 		if (file_mode & S_IRGRP)
2882 			dac_granted |= VREAD;
2883 		if (file_mode & S_IWGRP)
2884 			dac_granted |= VWRITE;
2885 
2886 		if ((acc_mode & dac_granted) == acc_mode)
2887 			return (0);
2888 
2889 		goto privcheck;
2890 	}
2891 
2892 	/* Otherwise, check everyone else. */
2893 	if (file_mode & S_IXOTH)
2894 		dac_granted |= VEXEC;
2895 	if (file_mode & S_IROTH)
2896 		dac_granted |= VREAD;
2897 	if (file_mode & S_IWOTH)
2898 		dac_granted |= VWRITE;
2899 	if ((acc_mode & dac_granted) == acc_mode)
2900 		return (0);
2901 
2902 privcheck:
2903 	if (!suser_xxx(cred, NULL, PRISON_ROOT)) {
2904 		/* XXX audit: privilege used */
2905 		if (privused != NULL)
2906 			*privused = 1;
2907 		return (0);
2908 	}
2909 
2910 #ifdef CAPABILITIES
2911 	/*
2912 	 * Build a capability mask to determine if the set of capabilities
2913 	 * satisfies the requirements when combined with the granted mask
2914 	 * from above.
2915 	 * For each capability, if the capability is required, bitwise
2916 	 * or the request type onto the cap_granted mask.
2917 	 */
2918 	cap_granted = 0;
2919 	if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
2920 	    !cap_check_xxx(cred, NULL, CAP_DAC_EXECUTE, PRISON_ROOT))
2921 	    cap_granted |= VEXEC;
2922 
2923 	if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
2924 	    !cap_check_xxx(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
2925 		cap_granted |= VREAD;
2926 
2927 	if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
2928 	    !cap_check_xxx(cred, NULL, CAP_DAC_WRITE, PRISON_ROOT))
2929 		cap_granted |= VWRITE;
2930 
2931 	if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
2932 	    !cap_check_xxx(cred, NULL, CAP_FOWNER, PRISON_ROOT))
2933 		cap_granted |= VADMIN;
2934 
2935 	if ((acc_mode & (cap_granted | dac_granted)) == acc_mode) {
2936 		/* XXX audit: privilege used */
2937 		if (privused != NULL)
2938 			*privused = 1;
2939 		return (0);
2940 	}
2941 #endif
2942 
2943 	return ((acc_mode & VADMIN) ? EPERM : EACCES);
2944 }
2945 
2946