xref: /freebsd/sys/kern/vfs_subr.c (revision 6990ffd8a95caaba6858ad44ff1b3157d1efba8f)
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 thread *td));
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, td)
273 	struct mount *mp;
274 	int flags;
275 	struct mtx *interlkp;
276 	struct thread *td;
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, td))
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, td)
306 	struct mount *mp;
307 	struct thread *td;
308 {
309 
310 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
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 thread *td = curthread;	/* 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, td);
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 thread *td = curthread;	/* 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, td);
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, td, td->td_proc->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, td, slpflag, slptimeo)
703 	register struct vnode *vp;
704 	int flags;
705 	struct ucred *cred;
706 	struct thread *td;
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, td)) != 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, td, length, blksize)
819 	register struct vnode *vp;
820 	struct ucred *cred;
821 	struct thread *td;
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 thread *td = &updateproc->p_thread;  /* XXXKSE */
1024 
1025 	mtx_lock(&Giant);
1026 
1027 	EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, td->td_proc,
1028 	    SHUTDOWN_PRI_LAST);
1029 
1030 	for (;;) {
1031 		kthread_suspend_check(td->td_proc);
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, td);
1050 				(void) VOP_FSYNC(vp, td->td_proc->p_ucred, MNT_LAZY, td);
1051 				VOP_UNLOCK(vp, 0, td);
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  * XXXKSE  only one update?
1115  */
1116 int
1117 speedup_syncer()
1118 {
1119 
1120 	mtx_lock_spin(&sched_lock);
1121 	if (updateproc->p_thread.td_wchan == &lbolt) /* XXXKSE */
1122 		setrunnable(&updateproc->p_thread);
1123 	mtx_unlock_spin(&sched_lock);
1124 	if (rushjob < syncdelay / 2) {
1125 		rushjob += 1;
1126 		stat_rush_requests += 1;
1127 		return (1);
1128 	}
1129 	return(0);
1130 }
1131 
1132 /*
1133  * Associate a p-buffer with a vnode.
1134  *
1135  * Also sets B_PAGING flag to indicate that vnode is not fully associated
1136  * with the buffer.  i.e. the bp has not been linked into the vnode or
1137  * ref-counted.
1138  */
1139 void
1140 pbgetvp(vp, bp)
1141 	register struct vnode *vp;
1142 	register struct buf *bp;
1143 {
1144 
1145 	KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
1146 
1147 	bp->b_vp = vp;
1148 	bp->b_flags |= B_PAGING;
1149 	bp->b_dev = vn_todev(vp);
1150 }
1151 
1152 /*
1153  * Disassociate a p-buffer from a vnode.
1154  */
1155 void
1156 pbrelvp(bp)
1157 	register struct buf *bp;
1158 {
1159 
1160 	KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
1161 
1162 	/* XXX REMOVE ME */
1163 	if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
1164 		panic(
1165 		    "relpbuf(): b_vp was probably reassignbuf()d %p %x",
1166 		    bp,
1167 		    (int)bp->b_flags
1168 		);
1169 	}
1170 	bp->b_vp = (struct vnode *) 0;
1171 	bp->b_flags &= ~B_PAGING;
1172 }
1173 
1174 /*
1175  * Change the vnode a pager buffer is associated with.
1176  */
1177 void
1178 pbreassignbuf(bp, newvp)
1179 	struct buf *bp;
1180 	struct vnode *newvp;
1181 {
1182 
1183 	KASSERT(bp->b_flags & B_PAGING,
1184 	    ("pbreassignbuf() on non phys bp %p", bp));
1185 	bp->b_vp = newvp;
1186 }
1187 
1188 /*
1189  * Reassign a buffer from one vnode to another.
1190  * Used to assign file specific control information
1191  * (indirect blocks) to the vnode to which they belong.
1192  */
1193 void
1194 reassignbuf(bp, newvp)
1195 	register struct buf *bp;
1196 	register struct vnode *newvp;
1197 {
1198 	struct buflists *listheadp;
1199 	int delay;
1200 	int s;
1201 
1202 	if (newvp == NULL) {
1203 		printf("reassignbuf: NULL");
1204 		return;
1205 	}
1206 	++reassignbufcalls;
1207 
1208 	/*
1209 	 * B_PAGING flagged buffers cannot be reassigned because their vp
1210 	 * is not fully linked in.
1211 	 */
1212 	if (bp->b_flags & B_PAGING)
1213 		panic("cannot reassign paging buffer");
1214 
1215 	s = splbio();
1216 	/*
1217 	 * Delete from old vnode list, if on one.
1218 	 */
1219 	if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
1220 		if (bp->b_xflags & BX_VNDIRTY)
1221 			listheadp = &bp->b_vp->v_dirtyblkhd;
1222 		else
1223 			listheadp = &bp->b_vp->v_cleanblkhd;
1224 		TAILQ_REMOVE(listheadp, bp, b_vnbufs);
1225 		bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1226 		if (bp->b_vp != newvp) {
1227 			vdrop(bp->b_vp);
1228 			bp->b_vp = NULL;	/* for clarification */
1229 		}
1230 	}
1231 	/*
1232 	 * If dirty, put on list of dirty buffers; otherwise insert onto list
1233 	 * of clean buffers.
1234 	 */
1235 	if (bp->b_flags & B_DELWRI) {
1236 		struct buf *tbp;
1237 
1238 		listheadp = &newvp->v_dirtyblkhd;
1239 		if ((newvp->v_flag & VONWORKLST) == 0) {
1240 			switch (newvp->v_type) {
1241 			case VDIR:
1242 				delay = dirdelay;
1243 				break;
1244 			case VCHR:
1245 				if (newvp->v_rdev->si_mountpoint != NULL) {
1246 					delay = metadelay;
1247 					break;
1248 				}
1249 				/* fall through */
1250 			default:
1251 				delay = filedelay;
1252 			}
1253 			vn_syncer_add_to_worklist(newvp, delay);
1254 		}
1255 		bp->b_xflags |= BX_VNDIRTY;
1256 		tbp = TAILQ_FIRST(listheadp);
1257 		if (tbp == NULL ||
1258 		    bp->b_lblkno == 0 ||
1259 		    (bp->b_lblkno > 0 && tbp->b_lblkno < 0) ||
1260 		    (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) {
1261 			TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs);
1262 			++reassignbufsortgood;
1263 		} else if (bp->b_lblkno < 0) {
1264 			TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs);
1265 			++reassignbufsortgood;
1266 		} else if (reassignbufmethod == 1) {
1267 			/*
1268 			 * New sorting algorithm, only handle sequential case,
1269 			 * otherwise append to end (but before metadata)
1270 			 */
1271 			if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL &&
1272 			    (tbp->b_xflags & BX_VNDIRTY)) {
1273 				/*
1274 				 * Found the best place to insert the buffer
1275 				 */
1276 				TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1277 				++reassignbufsortgood;
1278 			} else {
1279 				/*
1280 				 * Missed, append to end, but before meta-data.
1281 				 * We know that the head buffer in the list is
1282 				 * not meta-data due to prior conditionals.
1283 				 *
1284 				 * Indirect effects:  NFS second stage write
1285 				 * tends to wind up here, giving maximum
1286 				 * distance between the unstable write and the
1287 				 * commit rpc.
1288 				 */
1289 				tbp = TAILQ_LAST(listheadp, buflists);
1290 				while (tbp && tbp->b_lblkno < 0)
1291 					tbp = TAILQ_PREV(tbp, buflists, b_vnbufs);
1292 				TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1293 				++reassignbufsortbad;
1294 			}
1295 		} else {
1296 			/*
1297 			 * Old sorting algorithm, scan queue and insert
1298 			 */
1299 			struct buf *ttbp;
1300 			while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) &&
1301 			    (ttbp->b_lblkno < bp->b_lblkno)) {
1302 				++reassignbufloops;
1303 				tbp = ttbp;
1304 			}
1305 			TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs);
1306 		}
1307 	} else {
1308 		bp->b_xflags |= BX_VNCLEAN;
1309 		TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs);
1310 		if ((newvp->v_flag & VONWORKLST) &&
1311 		    TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
1312 			newvp->v_flag &= ~VONWORKLST;
1313 			LIST_REMOVE(newvp, v_synclist);
1314 		}
1315 	}
1316 	if (bp->b_vp != newvp) {
1317 		bp->b_vp = newvp;
1318 		vhold(bp->b_vp);
1319 	}
1320 	splx(s);
1321 }
1322 
1323 /*
1324  * Create a vnode for a device.
1325  * Used for mounting the root file system.
1326  */
1327 int
1328 bdevvp(dev, vpp)
1329 	dev_t dev;
1330 	struct vnode **vpp;
1331 {
1332 	register struct vnode *vp;
1333 	struct vnode *nvp;
1334 	int error;
1335 
1336 	if (dev == NODEV) {
1337 		*vpp = NULLVP;
1338 		return (ENXIO);
1339 	}
1340 	if (vfinddev(dev, VCHR, vpp))
1341 		return (0);
1342 	error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
1343 	if (error) {
1344 		*vpp = NULLVP;
1345 		return (error);
1346 	}
1347 	vp = nvp;
1348 	vp->v_type = VCHR;
1349 	addalias(vp, dev);
1350 	*vpp = vp;
1351 	return (0);
1352 }
1353 
1354 /*
1355  * Add vnode to the alias list hung off the dev_t.
1356  *
1357  * The reason for this gunk is that multiple vnodes can reference
1358  * the same physical device, so checking vp->v_usecount to see
1359  * how many users there are is inadequate; the v_usecount for
1360  * the vnodes need to be accumulated.  vcount() does that.
1361  */
1362 struct vnode *
1363 addaliasu(nvp, nvp_rdev)
1364 	struct vnode *nvp;
1365 	udev_t nvp_rdev;
1366 {
1367 	struct vnode *ovp;
1368 	vop_t **ops;
1369 	dev_t dev;
1370 
1371 	if (nvp->v_type == VBLK)
1372 		return (nvp);
1373 	if (nvp->v_type != VCHR)
1374 		panic("addaliasu on non-special vnode");
1375 	dev = udev2dev(nvp_rdev, 0);
1376 	/*
1377 	 * Check to see if we have a bdevvp vnode with no associated
1378 	 * filesystem. If so, we want to associate the filesystem of
1379 	 * the new newly instigated vnode with the bdevvp vnode and
1380 	 * discard the newly created vnode rather than leaving the
1381 	 * bdevvp vnode lying around with no associated filesystem.
1382 	 */
1383 	if (vfinddev(dev, nvp->v_type, &ovp) == 0 || ovp->v_data != NULL) {
1384 		addalias(nvp, dev);
1385 		return (nvp);
1386 	}
1387 	/*
1388 	 * Discard unneeded vnode, but save its node specific data.
1389 	 * Note that if there is a lock, it is carried over in the
1390 	 * node specific data to the replacement vnode.
1391 	 */
1392 	vref(ovp);
1393 	ovp->v_data = nvp->v_data;
1394 	ovp->v_tag = nvp->v_tag;
1395 	nvp->v_data = NULL;
1396 	lockinit(&ovp->v_lock, PVFS, nvp->v_lock.lk_wmesg,
1397 	    nvp->v_lock.lk_timo, nvp->v_lock.lk_flags & LK_EXTFLG_MASK);
1398 	if (nvp->v_vnlock)
1399 		ovp->v_vnlock = &ovp->v_lock;
1400 	ops = ovp->v_op;
1401 	ovp->v_op = nvp->v_op;
1402 	if (VOP_ISLOCKED(nvp, curthread)) {
1403 		VOP_UNLOCK(nvp, 0, curthread);
1404 		vn_lock(ovp, LK_EXCLUSIVE | LK_RETRY, curthread);
1405 	}
1406 	nvp->v_op = ops;
1407 	insmntque(ovp, nvp->v_mount);
1408 	vrele(nvp);
1409 	vgone(nvp);
1410 	return (ovp);
1411 }
1412 
1413 /* This is a local helper function that do the same as addaliasu, but for a
1414  * dev_t instead of an udev_t. */
1415 static void
1416 addalias(nvp, dev)
1417 	struct vnode *nvp;
1418 	dev_t dev;
1419 {
1420 
1421 	KASSERT(nvp->v_type == VCHR, ("addalias on non-special vnode"));
1422 	nvp->v_rdev = dev;
1423 	mtx_lock(&spechash_mtx);
1424 	SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
1425 	mtx_unlock(&spechash_mtx);
1426 }
1427 
1428 /*
1429  * Grab a particular vnode from the free list, increment its
1430  * reference count and lock it. The vnode lock bit is set if the
1431  * vnode is being eliminated in vgone. The process is awakened
1432  * when the transition is completed, and an error returned to
1433  * indicate that the vnode is no longer usable (possibly having
1434  * been changed to a new file system type).
1435  */
1436 int
1437 vget(vp, flags, td)
1438 	register struct vnode *vp;
1439 	int flags;
1440 	struct thread *td;
1441 {
1442 	int error;
1443 
1444 	/*
1445 	 * If the vnode is in the process of being cleaned out for
1446 	 * another use, we wait for the cleaning to finish and then
1447 	 * return failure. Cleaning is determined by checking that
1448 	 * the VXLOCK flag is set.
1449 	 */
1450 	if ((flags & LK_INTERLOCK) == 0)
1451 		mtx_lock(&vp->v_interlock);
1452 	if (vp->v_flag & VXLOCK) {
1453 		if (vp->v_vxproc == curthread) {
1454 			printf("VXLOCK interlock avoided\n");
1455 		} else {
1456 			vp->v_flag |= VXWANT;
1457 			msleep((caddr_t)vp, &vp->v_interlock, PINOD | PDROP,
1458 			    "vget", 0);
1459 			return (ENOENT);
1460 		}
1461 	}
1462 
1463 	vp->v_usecount++;
1464 
1465 	if (VSHOULDBUSY(vp))
1466 		vbusy(vp);
1467 	if (flags & LK_TYPE_MASK) {
1468 		if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
1469 			/*
1470 			 * must expand vrele here because we do not want
1471 			 * to call VOP_INACTIVE if the reference count
1472 			 * drops back to zero since it was never really
1473 			 * active. We must remove it from the free list
1474 			 * before sleeping so that multiple processes do
1475 			 * not try to recycle it.
1476 			 */
1477 			mtx_lock(&vp->v_interlock);
1478 			vp->v_usecount--;
1479 			if (VSHOULDFREE(vp))
1480 				vfree(vp);
1481 			mtx_unlock(&vp->v_interlock);
1482 		}
1483 		return (error);
1484 	}
1485 	mtx_unlock(&vp->v_interlock);
1486 	return (0);
1487 }
1488 
1489 /*
1490  * Increase the reference count of a vnode.
1491  */
1492 void
1493 vref(struct vnode *vp)
1494 {
1495 	mtx_lock(&vp->v_interlock);
1496 	vp->v_usecount++;
1497 	mtx_unlock(&vp->v_interlock);
1498 }
1499 
1500 /*
1501  * Vnode put/release.
1502  * If count drops to zero, call inactive routine and return to freelist.
1503  */
1504 void
1505 vrele(vp)
1506 	struct vnode *vp;
1507 {
1508 	struct thread *td = curthread;	/* XXX */
1509 
1510 	KASSERT(vp != NULL, ("vrele: null vp"));
1511 
1512 	mtx_lock(&vp->v_interlock);
1513 
1514 	/* Skip this v_writecount check if we're going to panic below. */
1515 	KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
1516 	    ("vrele: missed vn_close"));
1517 
1518 	if (vp->v_usecount > 1) {
1519 
1520 		vp->v_usecount--;
1521 		mtx_unlock(&vp->v_interlock);
1522 
1523 		return;
1524 	}
1525 
1526 	if (vp->v_usecount == 1) {
1527 
1528 		vp->v_usecount--;
1529 		if (VSHOULDFREE(vp))
1530 			vfree(vp);
1531 	/*
1532 	 * If we are doing a vput, the node is already locked, and we must
1533 	 * call VOP_INACTIVE with the node locked.  So, in the case of
1534 	 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE.
1535 	 */
1536 		if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0) {
1537 			VOP_INACTIVE(vp, td);
1538 		}
1539 
1540 	} else {
1541 #ifdef DIAGNOSTIC
1542 		vprint("vrele: negative ref count", vp);
1543 		mtx_unlock(&vp->v_interlock);
1544 #endif
1545 		panic("vrele: negative ref cnt");
1546 	}
1547 }
1548 
1549 /*
1550  * Release an already locked vnode.  This give the same effects as
1551  * unlock+vrele(), but takes less time and avoids releasing and
1552  * re-aquiring the lock (as vrele() aquires the lock internally.)
1553  */
1554 void
1555 vput(vp)
1556 	struct vnode *vp;
1557 {
1558 	struct thread *td = curthread;	/* XXX */
1559 
1560 	GIANT_REQUIRED;
1561 
1562 	KASSERT(vp != NULL, ("vput: null vp"));
1563 	mtx_lock(&vp->v_interlock);
1564 	/* Skip this v_writecount check if we're going to panic below. */
1565 	KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
1566 	    ("vput: missed vn_close"));
1567 
1568 	if (vp->v_usecount > 1) {
1569 
1570 		vp->v_usecount--;
1571 		VOP_UNLOCK(vp, LK_INTERLOCK, td);
1572 		return;
1573 
1574 	}
1575 
1576 	if (vp->v_usecount == 1) {
1577 
1578 		vp->v_usecount--;
1579 		if (VSHOULDFREE(vp))
1580 			vfree(vp);
1581 	/*
1582 	 * If we are doing a vput, the node is already locked, and we must
1583 	 * call VOP_INACTIVE with the node locked.  So, in the case of
1584 	 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE.
1585 	 */
1586 		mtx_unlock(&vp->v_interlock);
1587 		VOP_INACTIVE(vp, td);
1588 
1589 	} else {
1590 #ifdef DIAGNOSTIC
1591 		vprint("vput: negative ref count", vp);
1592 #endif
1593 		panic("vput: negative ref cnt");
1594 	}
1595 }
1596 
1597 /*
1598  * Somebody doesn't want the vnode recycled.
1599  */
1600 void
1601 vhold(vp)
1602 	register struct vnode *vp;
1603 {
1604 	int s;
1605 
1606   	s = splbio();
1607 	vp->v_holdcnt++;
1608 	if (VSHOULDBUSY(vp))
1609 		vbusy(vp);
1610 	splx(s);
1611 }
1612 
1613 /*
1614  * Note that there is one less who cares about this vnode.  vdrop() is the
1615  * opposite of vhold().
1616  */
1617 void
1618 vdrop(vp)
1619 	register struct vnode *vp;
1620 {
1621 	int s;
1622 
1623 	s = splbio();
1624 	if (vp->v_holdcnt <= 0)
1625 		panic("vdrop: holdcnt");
1626 	vp->v_holdcnt--;
1627 	if (VSHOULDFREE(vp))
1628 		vfree(vp);
1629 	splx(s);
1630 }
1631 
1632 /*
1633  * Remove any vnodes in the vnode table belonging to mount point mp.
1634  *
1635  * If FORCECLOSE is not specified, there should not be any active ones,
1636  * return error if any are found (nb: this is a user error, not a
1637  * system error). If FORCECLOSE is specified, detach any active vnodes
1638  * that are found.
1639  *
1640  * If WRITECLOSE is set, only flush out regular file vnodes open for
1641  * writing.
1642  *
1643  * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
1644  *
1645  * `rootrefs' specifies the base reference count for the root vnode
1646  * of this filesystem. The root vnode is considered busy if its
1647  * v_usecount exceeds this value. On a successful return, vflush()
1648  * will call vrele() on the root vnode exactly rootrefs times.
1649  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
1650  * be zero.
1651  */
1652 #ifdef DIAGNOSTIC
1653 static int busyprt = 0;		/* print out busy vnodes */
1654 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1655 #endif
1656 
1657 int
1658 vflush(mp, rootrefs, flags)
1659 	struct mount *mp;
1660 	int rootrefs;
1661 	int flags;
1662 {
1663 	struct thread *td = curthread;	/* XXX */
1664 	struct vnode *vp, *nvp, *rootvp = NULL;
1665 	int busy = 0, error;
1666 
1667 	if (rootrefs > 0) {
1668 		KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
1669 		    ("vflush: bad args"));
1670 		/*
1671 		 * Get the filesystem root vnode. We can vput() it
1672 		 * immediately, since with rootrefs > 0, it won't go away.
1673 		 */
1674 		if ((error = VFS_ROOT(mp, &rootvp)) != 0)
1675 			return (error);
1676 		vput(rootvp);
1677 	}
1678 	mtx_lock(&mntvnode_mtx);
1679 loop:
1680 	for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp; vp = nvp) {
1681 		/*
1682 		 * Make sure this vnode wasn't reclaimed in getnewvnode().
1683 		 * Start over if it has (it won't be on the list anymore).
1684 		 */
1685 		if (vp->v_mount != mp)
1686 			goto loop;
1687 		nvp = LIST_NEXT(vp, v_mntvnodes);
1688 
1689 		mtx_unlock(&mntvnode_mtx);
1690 		mtx_lock(&vp->v_interlock);
1691 		/*
1692 		 * Skip over a vnodes marked VSYSTEM.
1693 		 */
1694 		if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1695 			mtx_unlock(&vp->v_interlock);
1696 			mtx_lock(&mntvnode_mtx);
1697 			continue;
1698 		}
1699 		/*
1700 		 * If WRITECLOSE is set, only flush out regular file vnodes
1701 		 * open for writing.
1702 		 */
1703 		if ((flags & WRITECLOSE) &&
1704 		    (vp->v_writecount == 0 || vp->v_type != VREG)) {
1705 			mtx_unlock(&vp->v_interlock);
1706 			mtx_lock(&mntvnode_mtx);
1707 			continue;
1708 		}
1709 
1710 		/*
1711 		 * With v_usecount == 0, all we need to do is clear out the
1712 		 * vnode data structures and we are done.
1713 		 */
1714 		if (vp->v_usecount == 0) {
1715 			vgonel(vp, td);
1716 			mtx_lock(&mntvnode_mtx);
1717 			continue;
1718 		}
1719 
1720 		/*
1721 		 * If FORCECLOSE is set, forcibly close the vnode. For block
1722 		 * or character devices, revert to an anonymous device. For
1723 		 * all other files, just kill them.
1724 		 */
1725 		if (flags & FORCECLOSE) {
1726 			if (vp->v_type != VCHR) {
1727 				vgonel(vp, td);
1728 			} else {
1729 				vclean(vp, 0, td);
1730 				vp->v_op = spec_vnodeop_p;
1731 				insmntque(vp, (struct mount *) 0);
1732 			}
1733 			mtx_lock(&mntvnode_mtx);
1734 			continue;
1735 		}
1736 #ifdef DIAGNOSTIC
1737 		if (busyprt)
1738 			vprint("vflush: busy vnode", vp);
1739 #endif
1740 		mtx_unlock(&vp->v_interlock);
1741 		mtx_lock(&mntvnode_mtx);
1742 		busy++;
1743 	}
1744 	mtx_unlock(&mntvnode_mtx);
1745 	if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
1746 		/*
1747 		 * If just the root vnode is busy, and if its refcount
1748 		 * is equal to `rootrefs', then go ahead and kill it.
1749 		 */
1750 		mtx_lock(&rootvp->v_interlock);
1751 		KASSERT(busy > 0, ("vflush: not busy"));
1752 		KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
1753 		if (busy == 1 && rootvp->v_usecount == rootrefs) {
1754 			vgonel(rootvp, td);
1755 			busy = 0;
1756 		} else
1757 			mtx_unlock(&rootvp->v_interlock);
1758 	}
1759 	if (busy)
1760 		return (EBUSY);
1761 	for (; rootrefs > 0; rootrefs--)
1762 		vrele(rootvp);
1763 	return (0);
1764 }
1765 
1766 /*
1767  * Disassociate the underlying file system from a vnode.
1768  */
1769 static void
1770 vclean(vp, flags, td)
1771 	struct vnode *vp;
1772 	int flags;
1773 	struct thread *td;
1774 {
1775 	int active;
1776 
1777 	/*
1778 	 * Check to see if the vnode is in use. If so we have to reference it
1779 	 * before we clean it out so that its count cannot fall to zero and
1780 	 * generate a race against ourselves to recycle it.
1781 	 */
1782 	if ((active = vp->v_usecount))
1783 		vp->v_usecount++;
1784 
1785 	/*
1786 	 * Prevent the vnode from being recycled or brought into use while we
1787 	 * clean it out.
1788 	 */
1789 	if (vp->v_flag & VXLOCK)
1790 		panic("vclean: deadlock");
1791 	vp->v_flag |= VXLOCK;
1792 	vp->v_vxproc = curthread;
1793 	/*
1794 	 * Even if the count is zero, the VOP_INACTIVE routine may still
1795 	 * have the object locked while it cleans it out. The VOP_LOCK
1796 	 * ensures that the VOP_INACTIVE routine is done with its work.
1797 	 * For active vnodes, it ensures that no other activity can
1798 	 * occur while the underlying object is being cleaned out.
1799 	 */
1800 	VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td);
1801 
1802 	/*
1803 	 * Clean out any buffers associated with the vnode.
1804 	 * If the flush fails, just toss the buffers.
1805 	 */
1806 	if (flags & DOCLOSE) {
1807 		if (TAILQ_FIRST(&vp->v_dirtyblkhd) != NULL)
1808 			(void) vn_write_suspend_wait(vp, NULL, V_WAIT);
1809 		if (vinvalbuf(vp, V_SAVE, NOCRED, td, 0, 0) != 0)
1810 			vinvalbuf(vp, 0, NOCRED, td, 0, 0);
1811 	}
1812 
1813 	VOP_DESTROYVOBJECT(vp);
1814 
1815 	/*
1816 	 * If purging an active vnode, it must be closed and
1817 	 * deactivated before being reclaimed. Note that the
1818 	 * VOP_INACTIVE will unlock the vnode.
1819 	 */
1820 	if (active) {
1821 		if (flags & DOCLOSE)
1822 			VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
1823 		VOP_INACTIVE(vp, td);
1824 	} else {
1825 		/*
1826 		 * Any other processes trying to obtain this lock must first
1827 		 * wait for VXLOCK to clear, then call the new lock operation.
1828 		 */
1829 		VOP_UNLOCK(vp, 0, td);
1830 	}
1831 	/*
1832 	 * Reclaim the vnode.
1833 	 */
1834 	if (VOP_RECLAIM(vp, td))
1835 		panic("vclean: cannot reclaim");
1836 
1837 	if (active) {
1838 		/*
1839 		 * Inline copy of vrele() since VOP_INACTIVE
1840 		 * has already been called.
1841 		 */
1842 		mtx_lock(&vp->v_interlock);
1843 		if (--vp->v_usecount <= 0) {
1844 #ifdef DIAGNOSTIC
1845 			if (vp->v_usecount < 0 || vp->v_writecount != 0) {
1846 				vprint("vclean: bad ref count", vp);
1847 				panic("vclean: ref cnt");
1848 			}
1849 #endif
1850 			vfree(vp);
1851 		}
1852 		mtx_unlock(&vp->v_interlock);
1853 	}
1854 
1855 	cache_purge(vp);
1856 	vp->v_vnlock = NULL;
1857 	lockdestroy(&vp->v_lock);
1858 
1859 	if (VSHOULDFREE(vp))
1860 		vfree(vp);
1861 
1862 	/*
1863 	 * Done with purge, notify sleepers of the grim news.
1864 	 */
1865 	vp->v_op = dead_vnodeop_p;
1866 	vn_pollgone(vp);
1867 	vp->v_tag = VT_NON;
1868 	vp->v_flag &= ~VXLOCK;
1869 	vp->v_vxproc = NULL;
1870 	if (vp->v_flag & VXWANT) {
1871 		vp->v_flag &= ~VXWANT;
1872 		wakeup((caddr_t) vp);
1873 	}
1874 }
1875 
1876 /*
1877  * Eliminate all activity associated with the requested vnode
1878  * and with all vnodes aliased to the requested vnode.
1879  */
1880 int
1881 vop_revoke(ap)
1882 	struct vop_revoke_args /* {
1883 		struct vnode *a_vp;
1884 		int a_flags;
1885 	} */ *ap;
1886 {
1887 	struct vnode *vp, *vq;
1888 	dev_t dev;
1889 
1890 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
1891 
1892 	vp = ap->a_vp;
1893 	/*
1894 	 * If a vgone (or vclean) is already in progress,
1895 	 * wait until it is done and return.
1896 	 */
1897 	if (vp->v_flag & VXLOCK) {
1898 		vp->v_flag |= VXWANT;
1899 		msleep((caddr_t)vp, &vp->v_interlock, PINOD | PDROP,
1900 		    "vop_revokeall", 0);
1901 		return (0);
1902 	}
1903 	dev = vp->v_rdev;
1904 	for (;;) {
1905 		mtx_lock(&spechash_mtx);
1906 		vq = SLIST_FIRST(&dev->si_hlist);
1907 		mtx_unlock(&spechash_mtx);
1908 		if (!vq)
1909 			break;
1910 		vgone(vq);
1911 	}
1912 	return (0);
1913 }
1914 
1915 /*
1916  * Recycle an unused vnode to the front of the free list.
1917  * Release the passed interlock if the vnode will be recycled.
1918  */
1919 int
1920 vrecycle(vp, inter_lkp, td)
1921 	struct vnode *vp;
1922 	struct mtx *inter_lkp;
1923 	struct thread *td;
1924 {
1925 
1926 	mtx_lock(&vp->v_interlock);
1927 	if (vp->v_usecount == 0) {
1928 		if (inter_lkp) {
1929 			mtx_unlock(inter_lkp);
1930 		}
1931 		vgonel(vp, td);
1932 		return (1);
1933 	}
1934 	mtx_unlock(&vp->v_interlock);
1935 	return (0);
1936 }
1937 
1938 /*
1939  * Eliminate all activity associated with a vnode
1940  * in preparation for reuse.
1941  */
1942 void
1943 vgone(vp)
1944 	register struct vnode *vp;
1945 {
1946 	struct thread *td = curthread;	/* XXX */
1947 
1948 	mtx_lock(&vp->v_interlock);
1949 	vgonel(vp, td);
1950 }
1951 
1952 /*
1953  * vgone, with the vp interlock held.
1954  */
1955 void
1956 vgonel(vp, td)
1957 	struct vnode *vp;
1958 	struct thread *td;
1959 {
1960 	int s;
1961 
1962 	/*
1963 	 * If a vgone (or vclean) is already in progress,
1964 	 * wait until it is done and return.
1965 	 */
1966 	if (vp->v_flag & VXLOCK) {
1967 		vp->v_flag |= VXWANT;
1968 		msleep((caddr_t)vp, &vp->v_interlock, PINOD | PDROP,
1969 		    "vgone", 0);
1970 		return;
1971 	}
1972 
1973 	/*
1974 	 * Clean out the filesystem specific data.
1975 	 */
1976 	vclean(vp, DOCLOSE, td);
1977 	mtx_lock(&vp->v_interlock);
1978 
1979 	/*
1980 	 * Delete from old mount point vnode list, if on one.
1981 	 */
1982 	if (vp->v_mount != NULL)
1983 		insmntque(vp, (struct mount *)0);
1984 	/*
1985 	 * If special device, remove it from special device alias list
1986 	 * if it is on one.
1987 	 */
1988 	if (vp->v_type == VCHR && vp->v_rdev != NULL && vp->v_rdev != NODEV) {
1989 		mtx_lock(&spechash_mtx);
1990 		SLIST_REMOVE(&vp->v_rdev->si_hlist, vp, vnode, v_specnext);
1991 		freedev(vp->v_rdev);
1992 		mtx_unlock(&spechash_mtx);
1993 		vp->v_rdev = NULL;
1994 	}
1995 
1996 	/*
1997 	 * If it is on the freelist and not already at the head,
1998 	 * move it to the head of the list. The test of the
1999 	 * VDOOMED flag and the reference count of zero is because
2000 	 * it will be removed from the free list by getnewvnode,
2001 	 * but will not have its reference count incremented until
2002 	 * after calling vgone. If the reference count were
2003 	 * incremented first, vgone would (incorrectly) try to
2004 	 * close the previous instance of the underlying object.
2005 	 */
2006 	if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) {
2007 		s = splbio();
2008 		mtx_lock(&vnode_free_list_mtx);
2009 		if (vp->v_flag & VFREE)
2010 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2011 		else
2012 			freevnodes++;
2013 		vp->v_flag |= VFREE;
2014 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2015 		mtx_unlock(&vnode_free_list_mtx);
2016 		splx(s);
2017 	}
2018 
2019 	vp->v_type = VBAD;
2020 	mtx_unlock(&vp->v_interlock);
2021 }
2022 
2023 /*
2024  * Lookup a vnode by device number.
2025  */
2026 int
2027 vfinddev(dev, type, vpp)
2028 	dev_t dev;
2029 	enum vtype type;
2030 	struct vnode **vpp;
2031 {
2032 	struct vnode *vp;
2033 
2034 	mtx_lock(&spechash_mtx);
2035 	SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
2036 		if (type == vp->v_type) {
2037 			*vpp = vp;
2038 			mtx_unlock(&spechash_mtx);
2039 			return (1);
2040 		}
2041 	}
2042 	mtx_unlock(&spechash_mtx);
2043 	return (0);
2044 }
2045 
2046 /*
2047  * Calculate the total number of references to a special device.
2048  */
2049 int
2050 vcount(vp)
2051 	struct vnode *vp;
2052 {
2053 	struct vnode *vq;
2054 	int count;
2055 
2056 	count = 0;
2057 	mtx_lock(&spechash_mtx);
2058 	SLIST_FOREACH(vq, &vp->v_rdev->si_hlist, v_specnext)
2059 		count += vq->v_usecount;
2060 	mtx_unlock(&spechash_mtx);
2061 	return (count);
2062 }
2063 
2064 /*
2065  * Same as above, but using the dev_t as argument
2066  */
2067 int
2068 count_dev(dev)
2069 	dev_t dev;
2070 {
2071 	struct vnode *vp;
2072 
2073 	vp = SLIST_FIRST(&dev->si_hlist);
2074 	if (vp == NULL)
2075 		return (0);
2076 	return(vcount(vp));
2077 }
2078 
2079 /*
2080  * Print out a description of a vnode.
2081  */
2082 static char *typename[] =
2083 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
2084 
2085 void
2086 vprint(label, vp)
2087 	char *label;
2088 	struct vnode *vp;
2089 {
2090 	char buf[96];
2091 
2092 	if (label != NULL)
2093 		printf("%s: %p: ", label, (void *)vp);
2094 	else
2095 		printf("%p: ", (void *)vp);
2096 	printf("type %s, usecount %d, writecount %d, refcount %d,",
2097 	    typename[vp->v_type], vp->v_usecount, vp->v_writecount,
2098 	    vp->v_holdcnt);
2099 	buf[0] = '\0';
2100 	if (vp->v_flag & VROOT)
2101 		strcat(buf, "|VROOT");
2102 	if (vp->v_flag & VTEXT)
2103 		strcat(buf, "|VTEXT");
2104 	if (vp->v_flag & VSYSTEM)
2105 		strcat(buf, "|VSYSTEM");
2106 	if (vp->v_flag & VXLOCK)
2107 		strcat(buf, "|VXLOCK");
2108 	if (vp->v_flag & VXWANT)
2109 		strcat(buf, "|VXWANT");
2110 	if (vp->v_flag & VBWAIT)
2111 		strcat(buf, "|VBWAIT");
2112 	if (vp->v_flag & VDOOMED)
2113 		strcat(buf, "|VDOOMED");
2114 	if (vp->v_flag & VFREE)
2115 		strcat(buf, "|VFREE");
2116 	if (vp->v_flag & VOBJBUF)
2117 		strcat(buf, "|VOBJBUF");
2118 	if (buf[0] != '\0')
2119 		printf(" flags (%s)", &buf[1]);
2120 	if (vp->v_data == NULL) {
2121 		printf("\n");
2122 	} else {
2123 		printf("\n\t");
2124 		VOP_PRINT(vp);
2125 	}
2126 }
2127 
2128 #ifdef DDB
2129 #include <ddb/ddb.h>
2130 /*
2131  * List all of the locked vnodes in the system.
2132  * Called when debugging the kernel.
2133  */
2134 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
2135 {
2136 	struct thread *td = curthread;	/* XXX */
2137 	struct mount *mp, *nmp;
2138 	struct vnode *vp;
2139 
2140 	printf("Locked vnodes\n");
2141 	mtx_lock(&mountlist_mtx);
2142 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2143 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
2144 			nmp = TAILQ_NEXT(mp, mnt_list);
2145 			continue;
2146 		}
2147 		mtx_lock(&mntvnode_mtx);
2148 		LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) {
2149 			if (VOP_ISLOCKED(vp, NULL))
2150 				vprint((char *)0, vp);
2151 		}
2152 		mtx_unlock(&mntvnode_mtx);
2153 		mtx_lock(&mountlist_mtx);
2154 		nmp = TAILQ_NEXT(mp, mnt_list);
2155 		vfs_unbusy(mp, td);
2156 	}
2157 	mtx_unlock(&mountlist_mtx);
2158 }
2159 #endif
2160 
2161 /*
2162  * Top level filesystem related information gathering.
2163  */
2164 static int	sysctl_ovfs_conf __P((SYSCTL_HANDLER_ARGS));
2165 
2166 static int
2167 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2168 {
2169 	int *name = (int *)arg1 - 1;	/* XXX */
2170 	u_int namelen = arg2 + 1;	/* XXX */
2171 	struct vfsconf *vfsp;
2172 
2173 #if 1 || defined(COMPAT_PRELITE2)
2174 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2175 	if (namelen == 1)
2176 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2177 #endif
2178 
2179 	/* XXX the below code does not compile; vfs_sysctl does not exist. */
2180 #ifdef notyet
2181 	/* all sysctl names at this level are at least name and field */
2182 	if (namelen < 2)
2183 		return (ENOTDIR);		/* overloaded */
2184 	if (name[0] != VFS_GENERIC) {
2185 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2186 			if (vfsp->vfc_typenum == name[0])
2187 				break;
2188 		if (vfsp == NULL)
2189 			return (EOPNOTSUPP);
2190 		return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
2191 		    oldp, oldlenp, newp, newlen, td));
2192 	}
2193 #endif
2194 	switch (name[1]) {
2195 	case VFS_MAXTYPENUM:
2196 		if (namelen != 2)
2197 			return (ENOTDIR);
2198 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2199 	case VFS_CONF:
2200 		if (namelen != 3)
2201 			return (ENOTDIR);	/* overloaded */
2202 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
2203 			if (vfsp->vfc_typenum == name[2])
2204 				break;
2205 		if (vfsp == NULL)
2206 			return (EOPNOTSUPP);
2207 		return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
2208 	}
2209 	return (EOPNOTSUPP);
2210 }
2211 
2212 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
2213 	"Generic filesystem");
2214 
2215 #if 1 || defined(COMPAT_PRELITE2)
2216 
2217 static int
2218 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2219 {
2220 	int error;
2221 	struct vfsconf *vfsp;
2222 	struct ovfsconf ovfs;
2223 
2224 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
2225 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
2226 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
2227 		ovfs.vfc_index = vfsp->vfc_typenum;
2228 		ovfs.vfc_refcount = vfsp->vfc_refcount;
2229 		ovfs.vfc_flags = vfsp->vfc_flags;
2230 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2231 		if (error)
2232 			return error;
2233 	}
2234 	return 0;
2235 }
2236 
2237 #endif /* 1 || COMPAT_PRELITE2 */
2238 
2239 #if COMPILING_LINT
2240 #define KINFO_VNODESLOP	10
2241 /*
2242  * Dump vnode list (via sysctl).
2243  * Copyout address of vnode followed by vnode.
2244  */
2245 /* ARGSUSED */
2246 static int
2247 sysctl_vnode(SYSCTL_HANDLER_ARGS)
2248 {
2249 	struct thread *td = curthread;	/* XXX */
2250 	struct mount *mp, *nmp;
2251 	struct vnode *nvp, *vp;
2252 	int error;
2253 
2254 #define VPTRSZ	sizeof (struct vnode *)
2255 #define VNODESZ	sizeof (struct vnode)
2256 
2257 	req->lock = 0;
2258 	if (!req->oldptr) /* Make an estimate */
2259 		return (SYSCTL_OUT(req, 0,
2260 			(numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ)));
2261 
2262 	mtx_lock(&mountlist_mtx);
2263 	for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2264 		if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
2265 			nmp = TAILQ_NEXT(mp, mnt_list);
2266 			continue;
2267 		}
2268 		mtx_lock(&mntvnode_mtx);
2269 again:
2270 		for (vp = LIST_FIRST(&mp->mnt_vnodelist);
2271 		     vp != NULL;
2272 		     vp = nvp) {
2273 			/*
2274 			 * Check that the vp is still associated with
2275 			 * this filesystem.  RACE: could have been
2276 			 * recycled onto the same filesystem.
2277 			 */
2278 			if (vp->v_mount != mp)
2279 				goto again;
2280 			nvp = LIST_NEXT(vp, v_mntvnodes);
2281 			mtx_unlock(&mntvnode_mtx);
2282 			if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) ||
2283 			    (error = SYSCTL_OUT(req, vp, VNODESZ)))
2284 				return (error);
2285 			mtx_lock(&mntvnode_mtx);
2286 		}
2287 		mtx_unlock(&mntvnode_mtx);
2288 		mtx_lock(&mountlist_mtx);
2289 		nmp = TAILQ_NEXT(mp, mnt_list);
2290 		vfs_unbusy(mp, td);
2291 	}
2292 	mtx_unlock(&mountlist_mtx);
2293 
2294 	return (0);
2295 }
2296 
2297 /*
2298  * XXX
2299  * Exporting the vnode list on large systems causes them to crash.
2300  * Exporting the vnode list on medium systems causes sysctl to coredump.
2301  */
2302 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2303 	0, 0, sysctl_vnode, "S,vnode", "");
2304 #endif
2305 
2306 /*
2307  * Check to see if a filesystem is mounted on a block device.
2308  */
2309 int
2310 vfs_mountedon(vp)
2311 	struct vnode *vp;
2312 {
2313 
2314 	if (vp->v_rdev->si_mountpoint != NULL)
2315 		return (EBUSY);
2316 	return (0);
2317 }
2318 
2319 /*
2320  * Unmount all filesystems. The list is traversed in reverse order
2321  * of mounting to avoid dependencies.
2322  */
2323 void
2324 vfs_unmountall()
2325 {
2326 	struct mount *mp;
2327 	struct thread *td;
2328 	int error;
2329 
2330 	if (curthread != NULL)
2331 		td = curthread;
2332 	else
2333 		td = &initproc->p_thread;	/* XXX XXX should this be proc0? */
2334 	/*
2335 	 * Since this only runs when rebooting, it is not interlocked.
2336 	 */
2337 	while(!TAILQ_EMPTY(&mountlist)) {
2338 		mp = TAILQ_LAST(&mountlist, mntlist);
2339 		error = dounmount(mp, MNT_FORCE, td);
2340 		if (error) {
2341 			TAILQ_REMOVE(&mountlist, mp, mnt_list);
2342 			printf("unmount of %s failed (",
2343 			    mp->mnt_stat.f_mntonname);
2344 			if (error == EBUSY)
2345 				printf("BUSY)\n");
2346 			else
2347 				printf("%d)\n", error);
2348 		} else {
2349 			/* The unmount has removed mp from the mountlist */
2350 		}
2351 	}
2352 }
2353 
2354 /*
2355  * perform msync on all vnodes under a mount point
2356  * the mount point must be locked.
2357  */
2358 void
2359 vfs_msync(struct mount *mp, int flags) {
2360 	struct vnode *vp, *nvp;
2361 	struct vm_object *obj;
2362 	int anyio, tries;
2363 
2364 	GIANT_REQUIRED;
2365 
2366 	tries = 5;
2367 loop:
2368 	anyio = 0;
2369 	mtx_lock(&mntvnode_mtx);
2370 	for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp != NULL; vp = nvp) {
2371 
2372 		nvp = LIST_NEXT(vp, v_mntvnodes);
2373 
2374 		if (vp->v_mount != mp) {
2375 			mtx_unlock(&mntvnode_mtx);
2376 			goto loop;
2377 		}
2378 
2379 		if (vp->v_flag & VXLOCK)	/* XXX: what if MNT_WAIT? */
2380 			continue;
2381 
2382 		if (vp->v_flag & VNOSYNC)	/* unlinked, skip it */
2383 			continue;
2384 
2385 		if (flags != MNT_WAIT) {
2386 			if (VOP_GETVOBJECT(vp, &obj) != 0 ||
2387 			    (obj->flags & OBJ_MIGHTBEDIRTY) == 0)
2388 				continue;
2389 			if (VOP_ISLOCKED(vp, NULL))
2390 				continue;
2391 		}
2392 
2393 		mtx_unlock(&mntvnode_mtx);
2394 		mtx_lock(&vp->v_interlock);
2395 		if (VOP_GETVOBJECT(vp, &obj) == 0 &&
2396 		    (obj->flags & OBJ_MIGHTBEDIRTY)) {
2397 			if (!vget(vp,
2398 				LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, curthread)) {
2399 				if (VOP_GETVOBJECT(vp, &obj) == 0) {
2400 					vm_object_page_clean(obj, 0, 0,
2401 					    flags == MNT_WAIT ?
2402 					    OBJPC_SYNC : OBJPC_NOSYNC);
2403 					anyio = 1;
2404 				}
2405 				vput(vp);
2406 			}
2407 		} else {
2408 			mtx_unlock(&vp->v_interlock);
2409 		}
2410 		mtx_lock(&mntvnode_mtx);
2411 	}
2412 	mtx_unlock(&mntvnode_mtx);
2413 	if (anyio && (--tries > 0))
2414 		goto loop;
2415 }
2416 
2417 /*
2418  * Create the VM object needed for VMIO and mmap support.  This
2419  * is done for all VREG files in the system.  Some filesystems might
2420  * afford the additional metadata buffering capability of the
2421  * VMIO code by making the device node be VMIO mode also.
2422  *
2423  * vp must be locked when vfs_object_create is called.
2424  */
2425 int
2426 vfs_object_create(vp, td, cred)
2427 	struct vnode *vp;
2428 	struct thread *td;
2429 	struct ucred *cred;
2430 {
2431 	GIANT_REQUIRED;
2432 	return (VOP_CREATEVOBJECT(vp, cred, td));
2433 }
2434 
2435 /*
2436  * Mark a vnode as free, putting it up for recycling.
2437  */
2438 void
2439 vfree(vp)
2440 	struct vnode *vp;
2441 {
2442 	int s;
2443 
2444 	s = splbio();
2445 	mtx_lock(&vnode_free_list_mtx);
2446 	KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free"));
2447 	if (vp->v_flag & VAGE) {
2448 		TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2449 	} else {
2450 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2451 	}
2452 	freevnodes++;
2453 	mtx_unlock(&vnode_free_list_mtx);
2454 	vp->v_flag &= ~VAGE;
2455 	vp->v_flag |= VFREE;
2456 	splx(s);
2457 }
2458 
2459 /*
2460  * Opposite of vfree() - mark a vnode as in use.
2461  */
2462 void
2463 vbusy(vp)
2464 	struct vnode *vp;
2465 {
2466 	int s;
2467 
2468 	s = splbio();
2469 	mtx_lock(&vnode_free_list_mtx);
2470 	KASSERT((vp->v_flag & VFREE) != 0, ("vnode not free"));
2471 	TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
2472 	freevnodes--;
2473 	mtx_unlock(&vnode_free_list_mtx);
2474 	vp->v_flag &= ~(VFREE|VAGE);
2475 	splx(s);
2476 }
2477 
2478 /*
2479  * Record a process's interest in events which might happen to
2480  * a vnode.  Because poll uses the historic select-style interface
2481  * internally, this routine serves as both the ``check for any
2482  * pending events'' and the ``record my interest in future events''
2483  * functions.  (These are done together, while the lock is held,
2484  * to avoid race conditions.)
2485  */
2486 int
2487 vn_pollrecord(vp, td, events)
2488 	struct vnode *vp;
2489 	struct thread *td;
2490 	short events;
2491 {
2492 	mtx_lock(&vp->v_pollinfo.vpi_lock);
2493 	if (vp->v_pollinfo.vpi_revents & events) {
2494 		/*
2495 		 * This leaves events we are not interested
2496 		 * in available for the other process which
2497 		 * which presumably had requested them
2498 		 * (otherwise they would never have been
2499 		 * recorded).
2500 		 */
2501 		events &= vp->v_pollinfo.vpi_revents;
2502 		vp->v_pollinfo.vpi_revents &= ~events;
2503 
2504 		mtx_unlock(&vp->v_pollinfo.vpi_lock);
2505 		return events;
2506 	}
2507 	vp->v_pollinfo.vpi_events |= events;
2508 	selrecord(td, &vp->v_pollinfo.vpi_selinfo);
2509 	mtx_unlock(&vp->v_pollinfo.vpi_lock);
2510 	return 0;
2511 }
2512 
2513 /*
2514  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
2515  * it is possible for us to miss an event due to race conditions, but
2516  * that condition is expected to be rare, so for the moment it is the
2517  * preferred interface.
2518  */
2519 void
2520 vn_pollevent(vp, events)
2521 	struct vnode *vp;
2522 	short events;
2523 {
2524 	mtx_lock(&vp->v_pollinfo.vpi_lock);
2525 	if (vp->v_pollinfo.vpi_events & events) {
2526 		/*
2527 		 * We clear vpi_events so that we don't
2528 		 * call selwakeup() twice if two events are
2529 		 * posted before the polling process(es) is
2530 		 * awakened.  This also ensures that we take at
2531 		 * most one selwakeup() if the polling process
2532 		 * is no longer interested.  However, it does
2533 		 * mean that only one event can be noticed at
2534 		 * a time.  (Perhaps we should only clear those
2535 		 * event bits which we note?) XXX
2536 		 */
2537 		vp->v_pollinfo.vpi_events = 0;	/* &= ~events ??? */
2538 		vp->v_pollinfo.vpi_revents |= events;
2539 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2540 	}
2541 	mtx_unlock(&vp->v_pollinfo.vpi_lock);
2542 }
2543 
2544 #define VN_KNOTE(vp, b) \
2545 	KNOTE((struct klist *)&vp->v_pollinfo.vpi_selinfo.si_note, (b))
2546 
2547 /*
2548  * Wake up anyone polling on vp because it is being revoked.
2549  * This depends on dead_poll() returning POLLHUP for correct
2550  * behavior.
2551  */
2552 void
2553 vn_pollgone(vp)
2554 	struct vnode *vp;
2555 {
2556 	mtx_lock(&vp->v_pollinfo.vpi_lock);
2557         VN_KNOTE(vp, NOTE_REVOKE);
2558 	if (vp->v_pollinfo.vpi_events) {
2559 		vp->v_pollinfo.vpi_events = 0;
2560 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2561 	}
2562 	mtx_unlock(&vp->v_pollinfo.vpi_lock);
2563 }
2564 
2565 
2566 
2567 /*
2568  * Routine to create and manage a filesystem syncer vnode.
2569  */
2570 #define sync_close ((int (*) __P((struct  vop_close_args *)))nullop)
2571 static int	sync_fsync __P((struct  vop_fsync_args *));
2572 static int	sync_inactive __P((struct  vop_inactive_args *));
2573 static int	sync_reclaim  __P((struct  vop_reclaim_args *));
2574 #define sync_lock ((int (*) __P((struct  vop_lock_args *)))vop_nolock)
2575 #define sync_unlock ((int (*) __P((struct  vop_unlock_args *)))vop_nounlock)
2576 static int	sync_print __P((struct vop_print_args *));
2577 #define sync_islocked ((int(*) __P((struct vop_islocked_args *)))vop_noislocked)
2578 
2579 static vop_t **sync_vnodeop_p;
2580 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
2581 	{ &vop_default_desc,	(vop_t *) vop_eopnotsupp },
2582 	{ &vop_close_desc,	(vop_t *) sync_close },		/* close */
2583 	{ &vop_fsync_desc,	(vop_t *) sync_fsync },		/* fsync */
2584 	{ &vop_inactive_desc,	(vop_t *) sync_inactive },	/* inactive */
2585 	{ &vop_reclaim_desc,	(vop_t *) sync_reclaim },	/* reclaim */
2586 	{ &vop_lock_desc,	(vop_t *) sync_lock },		/* lock */
2587 	{ &vop_unlock_desc,	(vop_t *) sync_unlock },	/* unlock */
2588 	{ &vop_print_desc,	(vop_t *) sync_print },		/* print */
2589 	{ &vop_islocked_desc,	(vop_t *) sync_islocked },	/* islocked */
2590 	{ NULL, NULL }
2591 };
2592 static struct vnodeopv_desc sync_vnodeop_opv_desc =
2593 	{ &sync_vnodeop_p, sync_vnodeop_entries };
2594 
2595 VNODEOP_SET(sync_vnodeop_opv_desc);
2596 
2597 /*
2598  * Create a new filesystem syncer vnode for the specified mount point.
2599  */
2600 int
2601 vfs_allocate_syncvnode(mp)
2602 	struct mount *mp;
2603 {
2604 	struct vnode *vp;
2605 	static long start, incr, next;
2606 	int error;
2607 
2608 	/* Allocate a new vnode */
2609 	if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) {
2610 		mp->mnt_syncer = NULL;
2611 		return (error);
2612 	}
2613 	vp->v_type = VNON;
2614 	/*
2615 	 * Place the vnode onto the syncer worklist. We attempt to
2616 	 * scatter them about on the list so that they will go off
2617 	 * at evenly distributed times even if all the filesystems
2618 	 * are mounted at once.
2619 	 */
2620 	next += incr;
2621 	if (next == 0 || next > syncer_maxdelay) {
2622 		start /= 2;
2623 		incr /= 2;
2624 		if (start == 0) {
2625 			start = syncer_maxdelay / 2;
2626 			incr = syncer_maxdelay;
2627 		}
2628 		next = start;
2629 	}
2630 	vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
2631 	mp->mnt_syncer = vp;
2632 	return (0);
2633 }
2634 
2635 /*
2636  * Do a lazy sync of the filesystem.
2637  */
2638 static int
2639 sync_fsync(ap)
2640 	struct vop_fsync_args /* {
2641 		struct vnode *a_vp;
2642 		struct ucred *a_cred;
2643 		int a_waitfor;
2644 		struct thread *a_td;
2645 	} */ *ap;
2646 {
2647 	struct vnode *syncvp = ap->a_vp;
2648 	struct mount *mp = syncvp->v_mount;
2649 	struct thread *td = ap->a_td;
2650 	int asyncflag;
2651 
2652 	/*
2653 	 * We only need to do something if this is a lazy evaluation.
2654 	 */
2655 	if (ap->a_waitfor != MNT_LAZY)
2656 		return (0);
2657 
2658 	/*
2659 	 * Move ourselves to the back of the sync list.
2660 	 */
2661 	vn_syncer_add_to_worklist(syncvp, syncdelay);
2662 
2663 	/*
2664 	 * Walk the list of vnodes pushing all that are dirty and
2665 	 * not already on the sync list.
2666 	 */
2667 	mtx_lock(&mountlist_mtx);
2668 	if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) {
2669 		mtx_unlock(&mountlist_mtx);
2670 		return (0);
2671 	}
2672 	if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
2673 		vfs_unbusy(mp, td);
2674 		return (0);
2675 	}
2676 	asyncflag = mp->mnt_flag & MNT_ASYNC;
2677 	mp->mnt_flag &= ~MNT_ASYNC;
2678 	vfs_msync(mp, MNT_NOWAIT);
2679 	VFS_SYNC(mp, MNT_LAZY, ap->a_cred, td);
2680 	if (asyncflag)
2681 		mp->mnt_flag |= MNT_ASYNC;
2682 	vn_finished_write(mp);
2683 	vfs_unbusy(mp, td);
2684 	return (0);
2685 }
2686 
2687 /*
2688  * The syncer vnode is no referenced.
2689  */
2690 static int
2691 sync_inactive(ap)
2692 	struct vop_inactive_args /* {
2693 		struct vnode *a_vp;
2694 		struct thread *a_td;
2695 	} */ *ap;
2696 {
2697 
2698 	vgone(ap->a_vp);
2699 	return (0);
2700 }
2701 
2702 /*
2703  * The syncer vnode is no longer needed and is being decommissioned.
2704  *
2705  * Modifications to the worklist must be protected at splbio().
2706  */
2707 static int
2708 sync_reclaim(ap)
2709 	struct vop_reclaim_args /* {
2710 		struct vnode *a_vp;
2711 	} */ *ap;
2712 {
2713 	struct vnode *vp = ap->a_vp;
2714 	int s;
2715 
2716 	s = splbio();
2717 	vp->v_mount->mnt_syncer = NULL;
2718 	if (vp->v_flag & VONWORKLST) {
2719 		LIST_REMOVE(vp, v_synclist);
2720 		vp->v_flag &= ~VONWORKLST;
2721 	}
2722 	splx(s);
2723 
2724 	return (0);
2725 }
2726 
2727 /*
2728  * Print out a syncer vnode.
2729  */
2730 static int
2731 sync_print(ap)
2732 	struct vop_print_args /* {
2733 		struct vnode *a_vp;
2734 	} */ *ap;
2735 {
2736 	struct vnode *vp = ap->a_vp;
2737 
2738 	printf("syncer vnode");
2739 	if (vp->v_vnlock != NULL)
2740 		lockmgr_printinfo(vp->v_vnlock);
2741 	printf("\n");
2742 	return (0);
2743 }
2744 
2745 /*
2746  * extract the dev_t from a VCHR
2747  */
2748 dev_t
2749 vn_todev(vp)
2750 	struct vnode *vp;
2751 {
2752 	if (vp->v_type != VCHR)
2753 		return (NODEV);
2754 	return (vp->v_rdev);
2755 }
2756 
2757 /*
2758  * Check if vnode represents a disk device
2759  */
2760 int
2761 vn_isdisk(vp, errp)
2762 	struct vnode *vp;
2763 	int *errp;
2764 {
2765 	struct cdevsw *cdevsw;
2766 
2767 	if (vp->v_type != VCHR) {
2768 		if (errp != NULL)
2769 			*errp = ENOTBLK;
2770 		return (0);
2771 	}
2772 	if (vp->v_rdev == NULL) {
2773 		if (errp != NULL)
2774 			*errp = ENXIO;
2775 		return (0);
2776 	}
2777 	cdevsw = devsw(vp->v_rdev);
2778 	if (cdevsw == NULL) {
2779 		if (errp != NULL)
2780 			*errp = ENXIO;
2781 		return (0);
2782 	}
2783 	if (!(cdevsw->d_flags & D_DISK)) {
2784 		if (errp != NULL)
2785 			*errp = ENOTBLK;
2786 		return (0);
2787 	}
2788 	if (errp != NULL)
2789 		*errp = 0;
2790 	return (1);
2791 }
2792 
2793 /*
2794  * Free data allocated by namei(); see namei(9) for details.
2795  */
2796 void
2797 NDFREE(ndp, flags)
2798      struct nameidata *ndp;
2799      const uint flags;
2800 {
2801 	if (!(flags & NDF_NO_FREE_PNBUF) &&
2802 	    (ndp->ni_cnd.cn_flags & HASBUF)) {
2803 		zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
2804 		ndp->ni_cnd.cn_flags &= ~HASBUF;
2805 	}
2806 	if (!(flags & NDF_NO_DVP_UNLOCK) &&
2807 	    (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
2808 	    ndp->ni_dvp != ndp->ni_vp)
2809 		VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_thread);
2810 	if (!(flags & NDF_NO_DVP_RELE) &&
2811 	    (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
2812 		vrele(ndp->ni_dvp);
2813 		ndp->ni_dvp = NULL;
2814 	}
2815 	if (!(flags & NDF_NO_VP_UNLOCK) &&
2816 	    (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
2817 		VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_thread);
2818 	if (!(flags & NDF_NO_VP_RELE) &&
2819 	    ndp->ni_vp) {
2820 		vrele(ndp->ni_vp);
2821 		ndp->ni_vp = NULL;
2822 	}
2823 	if (!(flags & NDF_NO_STARTDIR_RELE) &&
2824 	    (ndp->ni_cnd.cn_flags & SAVESTART)) {
2825 		vrele(ndp->ni_startdir);
2826 		ndp->ni_startdir = NULL;
2827 	}
2828 }
2829 
2830 /*
2831  * Common file system object access control check routine.  Accepts a
2832  * vnode's type, "mode", uid and gid, requested access mode, credentials,
2833  * and optional call-by-reference privused argument allowing vaccess()
2834  * to indicate to the caller whether privilege was used to satisfy the
2835  * request.  Returns 0 on success, or an errno on failure.
2836  */
2837 int
2838 vaccess(type, file_mode, file_uid, file_gid, acc_mode, cred, privused)
2839 	enum vtype type;
2840 	mode_t file_mode;
2841 	uid_t file_uid;
2842 	gid_t file_gid;
2843 	mode_t acc_mode;
2844 	struct ucred *cred;
2845 	int *privused;
2846 {
2847 	mode_t dac_granted;
2848 #ifdef CAPABILITIES
2849 	mode_t cap_granted;
2850 #endif
2851 
2852 	/*
2853 	 * Look for a normal, non-privileged way to access the file/directory
2854 	 * as requested.  If it exists, go with that.
2855 	 */
2856 
2857 	if (privused != NULL)
2858 		*privused = 0;
2859 
2860 	dac_granted = 0;
2861 
2862 	/* Check the owner. */
2863 	if (cred->cr_uid == file_uid) {
2864 		dac_granted |= VADMIN;
2865 		if (file_mode & S_IXUSR)
2866 			dac_granted |= VEXEC;
2867 		if (file_mode & S_IRUSR)
2868 			dac_granted |= VREAD;
2869 		if (file_mode & S_IWUSR)
2870 			dac_granted |= VWRITE;
2871 
2872 		if ((acc_mode & dac_granted) == acc_mode)
2873 			return (0);
2874 
2875 		goto privcheck;
2876 	}
2877 
2878 	/* Otherwise, check the groups (first match) */
2879 	if (groupmember(file_gid, cred)) {
2880 		if (file_mode & S_IXGRP)
2881 			dac_granted |= VEXEC;
2882 		if (file_mode & S_IRGRP)
2883 			dac_granted |= VREAD;
2884 		if (file_mode & S_IWGRP)
2885 			dac_granted |= VWRITE;
2886 
2887 		if ((acc_mode & dac_granted) == acc_mode)
2888 			return (0);
2889 
2890 		goto privcheck;
2891 	}
2892 
2893 	/* Otherwise, check everyone else. */
2894 	if (file_mode & S_IXOTH)
2895 		dac_granted |= VEXEC;
2896 	if (file_mode & S_IROTH)
2897 		dac_granted |= VREAD;
2898 	if (file_mode & S_IWOTH)
2899 		dac_granted |= VWRITE;
2900 	if ((acc_mode & dac_granted) == acc_mode)
2901 		return (0);
2902 
2903 privcheck:
2904 	if (!suser_xxx(cred, NULL, PRISON_ROOT)) {
2905 		/* XXX audit: privilege used */
2906 		if (privused != NULL)
2907 			*privused = 1;
2908 		return (0);
2909 	}
2910 
2911 #ifdef CAPABILITIES
2912 	/*
2913 	 * Build a capability mask to determine if the set of capabilities
2914 	 * satisfies the requirements when combined with the granted mask
2915 	 * from above.
2916 	 * For each capability, if the capability is required, bitwise
2917 	 * or the request type onto the cap_granted mask.
2918 	 */
2919 	cap_granted = 0;
2920 	if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
2921 	    !cap_check_xxx(cred, NULL, CAP_DAC_EXECUTE, PRISON_ROOT))
2922 	    cap_granted |= VEXEC;
2923 
2924 	if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
2925 	    !cap_check_xxx(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
2926 		cap_granted |= VREAD;
2927 
2928 	if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
2929 	    !cap_check_xxx(cred, NULL, CAP_DAC_WRITE, PRISON_ROOT))
2930 		cap_granted |= VWRITE;
2931 
2932 	if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
2933 	    !cap_check_xxx(cred, NULL, CAP_FOWNER, PRISON_ROOT))
2934 		cap_granted |= VADMIN;
2935 
2936 	if ((acc_mode & (cap_granted | dac_granted)) == acc_mode) {
2937 		/* XXX audit: privilege used */
2938 		if (privused != NULL)
2939 			*privused = 1;
2940 		return (0);
2941 	}
2942 #endif
2943 
2944 	return ((acc_mode & VADMIN) ? EPERM : EACCES);
2945 }
2946 
2947