xref: /dflybsd-src/sys/kern/vfs_subr.c (revision 15b85273287111e7f177850e47c2bd7a1f804ca1)
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: src/sys/kern/vfs_subr.c,v 1.249.2.30 2003/04/04 20:35:57 tegge Exp $
40  * $DragonFly: src/sys/kern/vfs_subr.c,v 1.118 2008/09/17 21:44:18 dillon Exp $
41  */
42 
43 /*
44  * External virtual filesystem routines
45  */
46 #include "opt_ddb.h"
47 
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/buf.h>
51 #include <sys/conf.h>
52 #include <sys/dirent.h>
53 #include <sys/domain.h>
54 #include <sys/eventhandler.h>
55 #include <sys/fcntl.h>
56 #include <sys/kernel.h>
57 #include <sys/kthread.h>
58 #include <sys/malloc.h>
59 #include <sys/mbuf.h>
60 #include <sys/mount.h>
61 #include <sys/proc.h>
62 #include <sys/reboot.h>
63 #include <sys/socket.h>
64 #include <sys/stat.h>
65 #include <sys/sysctl.h>
66 #include <sys/syslog.h>
67 #include <sys/unistd.h>
68 #include <sys/vmmeter.h>
69 #include <sys/vnode.h>
70 
71 #include <machine/limits.h>
72 
73 #include <vm/vm.h>
74 #include <vm/vm_object.h>
75 #include <vm/vm_extern.h>
76 #include <vm/vm_kern.h>
77 #include <vm/pmap.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_page.h>
80 #include <vm/vm_pager.h>
81 #include <vm/vnode_pager.h>
82 #include <vm/vm_zone.h>
83 
84 #include <sys/buf2.h>
85 #include <sys/thread2.h>
86 #include <sys/sysref2.h>
87 
88 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
89 
90 int numvnodes;
91 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
92 int vfs_fastdev = 1;
93 SYSCTL_INT(_vfs, OID_AUTO, fastdev, CTLFLAG_RW, &vfs_fastdev, 0, "");
94 
95 enum vtype iftovt_tab[16] = {
96 	VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
97 	VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
98 };
99 int vttoif_tab[9] = {
100 	0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
101 	S_IFSOCK, S_IFIFO, S_IFMT,
102 };
103 
104 static int reassignbufcalls;
105 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW,
106 		&reassignbufcalls, 0, "");
107 static int reassignbufloops;
108 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW,
109 		&reassignbufloops, 0, "");
110 static int reassignbufsortgood;
111 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW,
112 		&reassignbufsortgood, 0, "");
113 static int reassignbufsortbad;
114 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW,
115 		&reassignbufsortbad, 0, "");
116 static int reassignbufmethod = 1;
117 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW,
118 		&reassignbufmethod, 0, "");
119 
120 int	nfs_mount_type = -1;
121 static struct lwkt_token spechash_token;
122 struct nfs_public nfs_pub;	/* publicly exported FS */
123 
124 int desiredvnodes;
125 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
126 		&desiredvnodes, 0, "Maximum number of vnodes");
127 
128 static void	vfs_free_addrlist (struct netexport *nep);
129 static int	vfs_free_netcred (struct radix_node *rn, void *w);
130 static int	vfs_hang_addrlist (struct mount *mp, struct netexport *nep,
131 				       const struct export_args *argp);
132 
133 extern int dev_ref_debug;
134 
135 /*
136  * Red black tree functions
137  */
138 static int rb_buf_compare(struct buf *b1, struct buf *b2);
139 RB_GENERATE2(buf_rb_tree, buf, b_rbnode, rb_buf_compare, off_t, b_loffset);
140 RB_GENERATE2(buf_rb_hash, buf, b_rbhash, rb_buf_compare, off_t, b_loffset);
141 
142 static int
143 rb_buf_compare(struct buf *b1, struct buf *b2)
144 {
145 	if (b1->b_loffset < b2->b_loffset)
146 		return(-1);
147 	if (b1->b_loffset > b2->b_loffset)
148 		return(1);
149 	return(0);
150 }
151 
152 /*
153  * Returns non-zero if the vnode is a candidate for lazy msyncing.
154  */
155 static __inline int
156 vshouldmsync(struct vnode *vp)
157 {
158 	if (vp->v_auxrefs != 0 || vp->v_sysref.refcnt > 0)
159 		return (0);		/* other holders */
160 	if (vp->v_object &&
161 	    (vp->v_object->ref_count || vp->v_object->resident_page_count)) {
162 		return (0);
163 	}
164 	return (1);
165 }
166 
167 /*
168  * Initialize the vnode management data structures.
169  *
170  * Called from vfsinit()
171  */
172 void
173 vfs_subr_init(void)
174 {
175 	/*
176 	 * Desiredvnodes is kern.maxvnodes.  We want to scale it
177 	 * according to available system memory but we may also have
178 	 * to limit it based on available KVM, which is capped on 32 bit
179 	 * systems.
180 	 */
181 	desiredvnodes = min(maxproc + vmstats.v_page_count / 4,
182 			    KvaSize / (20 *
183 			    (sizeof(struct vm_object) + sizeof(struct vnode))));
184 
185 	lwkt_token_init(&spechash_token);
186 }
187 
188 /*
189  * Knob to control the precision of file timestamps:
190  *
191  *   0 = seconds only; nanoseconds zeroed.
192  *   1 = seconds and nanoseconds, accurate within 1/HZ.
193  *   2 = seconds and nanoseconds, truncated to microseconds.
194  * >=3 = seconds and nanoseconds, maximum precision.
195  */
196 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
197 
198 static int timestamp_precision = TSP_SEC;
199 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
200 		&timestamp_precision, 0, "");
201 
202 /*
203  * Get a current timestamp.
204  */
205 void
206 vfs_timestamp(struct timespec *tsp)
207 {
208 	struct timeval tv;
209 
210 	switch (timestamp_precision) {
211 	case TSP_SEC:
212 		tsp->tv_sec = time_second;
213 		tsp->tv_nsec = 0;
214 		break;
215 	case TSP_HZ:
216 		getnanotime(tsp);
217 		break;
218 	case TSP_USEC:
219 		microtime(&tv);
220 		TIMEVAL_TO_TIMESPEC(&tv, tsp);
221 		break;
222 	case TSP_NSEC:
223 	default:
224 		nanotime(tsp);
225 		break;
226 	}
227 }
228 
229 /*
230  * Set vnode attributes to VNOVAL
231  */
232 void
233 vattr_null(struct vattr *vap)
234 {
235 	vap->va_type = VNON;
236 	vap->va_size = VNOVAL;
237 	vap->va_bytes = VNOVAL;
238 	vap->va_mode = VNOVAL;
239 	vap->va_nlink = VNOVAL;
240 	vap->va_uid = VNOVAL;
241 	vap->va_gid = VNOVAL;
242 	vap->va_fsid = VNOVAL;
243 	vap->va_fileid = VNOVAL;
244 	vap->va_blocksize = VNOVAL;
245 	vap->va_rmajor = VNOVAL;
246 	vap->va_rminor = VNOVAL;
247 	vap->va_atime.tv_sec = VNOVAL;
248 	vap->va_atime.tv_nsec = VNOVAL;
249 	vap->va_mtime.tv_sec = VNOVAL;
250 	vap->va_mtime.tv_nsec = VNOVAL;
251 	vap->va_ctime.tv_sec = VNOVAL;
252 	vap->va_ctime.tv_nsec = VNOVAL;
253 	vap->va_flags = VNOVAL;
254 	vap->va_gen = VNOVAL;
255 	vap->va_vaflags = 0;
256 	vap->va_fsmid = VNOVAL;
257 	/* va_*_uuid fields are only valid if related flags are set */
258 }
259 
260 /*
261  * Flush out and invalidate all buffers associated with a vnode.
262  *
263  * vp must be locked.
264  */
265 static int vinvalbuf_bp(struct buf *bp, void *data);
266 
267 struct vinvalbuf_bp_info {
268 	struct vnode *vp;
269 	int slptimeo;
270 	int lkflags;
271 	int flags;
272 };
273 
274 void
275 vupdatefsmid(struct vnode *vp)
276 {
277 	atomic_set_int(&vp->v_flag, VFSMID);
278 }
279 
280 int
281 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
282 {
283 	struct vinvalbuf_bp_info info;
284 	int error;
285 	vm_object_t object;
286 
287 	/*
288 	 * If we are being asked to save, call fsync to ensure that the inode
289 	 * is updated.
290 	 */
291 	if (flags & V_SAVE) {
292 		crit_enter();
293 		while (vp->v_track_write.bk_active) {
294 			vp->v_track_write.bk_waitflag = 1;
295 			error = tsleep(&vp->v_track_write, slpflag,
296 					"vinvlbuf", slptimeo);
297 			if (error) {
298 				crit_exit();
299 				return (error);
300 			}
301 		}
302 		if (!RB_EMPTY(&vp->v_rbdirty_tree)) {
303 			crit_exit();
304 			if ((error = VOP_FSYNC(vp, MNT_WAIT)) != 0)
305 				return (error);
306 			crit_enter();
307 			if (vp->v_track_write.bk_active > 0 ||
308 			    !RB_EMPTY(&vp->v_rbdirty_tree))
309 				panic("vinvalbuf: dirty bufs");
310 		}
311 		crit_exit();
312   	}
313 	crit_enter();
314 	info.slptimeo = slptimeo;
315 	info.lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
316 	if (slpflag & PCATCH)
317 		info.lkflags |= LK_PCATCH;
318 	info.flags = flags;
319 	info.vp = vp;
320 
321 	/*
322 	 * Flush the buffer cache until nothing is left.
323 	 */
324 	while (!RB_EMPTY(&vp->v_rbclean_tree) ||
325 	    !RB_EMPTY(&vp->v_rbdirty_tree)) {
326 		error = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, NULL,
327 				vinvalbuf_bp, &info);
328 		if (error == 0) {
329 			error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
330 					vinvalbuf_bp, &info);
331 		}
332 	}
333 
334 	/*
335 	 * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
336 	 * have write I/O in-progress but if there is a VM object then the
337 	 * VM object can also have read-I/O in-progress.
338 	 */
339 	do {
340 		while (vp->v_track_write.bk_active > 0) {
341 			vp->v_track_write.bk_waitflag = 1;
342 			tsleep(&vp->v_track_write, 0, "vnvlbv", 0);
343 		}
344 		if ((object = vp->v_object) != NULL) {
345 			while (object->paging_in_progress)
346 				vm_object_pip_sleep(object, "vnvlbx");
347 		}
348 	} while (vp->v_track_write.bk_active > 0);
349 
350 	crit_exit();
351 
352 	/*
353 	 * Destroy the copy in the VM cache, too.
354 	 */
355 	if ((object = vp->v_object) != NULL) {
356 		vm_object_page_remove(object, 0, 0,
357 			(flags & V_SAVE) ? TRUE : FALSE);
358 	}
359 
360 	if (!RB_EMPTY(&vp->v_rbdirty_tree) || !RB_EMPTY(&vp->v_rbclean_tree))
361 		panic("vinvalbuf: flush failed");
362 	if (!RB_EMPTY(&vp->v_rbhash_tree))
363 		panic("vinvalbuf: flush failed, buffers still present");
364 	return (0);
365 }
366 
367 static int
368 vinvalbuf_bp(struct buf *bp, void *data)
369 {
370 	struct vinvalbuf_bp_info *info = data;
371 	int error;
372 
373 	if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
374 		error = BUF_TIMELOCK(bp, info->lkflags,
375 				     "vinvalbuf", info->slptimeo);
376 		if (error == 0) {
377 			BUF_UNLOCK(bp);
378 			error = ENOLCK;
379 		}
380 		if (error == ENOLCK)
381 			return(0);
382 		return (-error);
383 	}
384 
385 	KKASSERT(bp->b_vp == info->vp);
386 
387 	/*
388 	 * XXX Since there are no node locks for NFS, I
389 	 * believe there is a slight chance that a delayed
390 	 * write will occur while sleeping just above, so
391 	 * check for it.  Note that vfs_bio_awrite expects
392 	 * buffers to reside on a queue, while bwrite() and
393 	 * brelse() do not.
394 	 *
395 	 * NOTE:  NO B_LOCKED CHECK.  Also no buf_checkwrite()
396 	 * check.  This code will write out the buffer, period.
397 	 */
398 	if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
399 	    (info->flags & V_SAVE)) {
400 		if (bp->b_vp == info->vp) {
401 			if (bp->b_flags & B_CLUSTEROK) {
402 				vfs_bio_awrite(bp);
403 			} else {
404 				bremfree(bp);
405 				bp->b_flags |= B_ASYNC;
406 				bwrite(bp);
407 			}
408 		} else {
409 			bremfree(bp);
410 			bwrite(bp);
411 		}
412 	} else if (info->flags & V_SAVE) {
413 		/*
414 		 * Cannot set B_NOCACHE on a clean buffer as this will
415 		 * destroy the VM backing store which might actually
416 		 * be dirty (and unsynchronized).
417 		 */
418 		bremfree(bp);
419 		bp->b_flags |= (B_INVAL | B_RELBUF);
420 		bp->b_flags &= ~B_ASYNC;
421 		brelse(bp);
422 	} else {
423 		bremfree(bp);
424 		bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
425 		bp->b_flags &= ~B_ASYNC;
426 		brelse(bp);
427 	}
428 	return(0);
429 }
430 
431 /*
432  * Truncate a file's buffer and pages to a specified length.  This
433  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
434  * sync activity.
435  *
436  * The vnode must be locked.
437  */
438 static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data);
439 static int vtruncbuf_bp_trunc(struct buf *bp, void *data);
440 static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data);
441 static int vtruncbuf_bp_metasync(struct buf *bp, void *data);
442 
443 int
444 vtruncbuf(struct vnode *vp, off_t length, int blksize)
445 {
446 	off_t truncloffset;
447 	int count;
448 	const char *filename;
449 
450 	/*
451 	 * Round up to the *next* block, then destroy the buffers in question.
452 	 * Since we are only removing some of the buffers we must rely on the
453 	 * scan count to determine whether a loop is necessary.
454 	 */
455 	if ((count = (int)(length % blksize)) != 0)
456 		truncloffset = length + (blksize - count);
457 	else
458 		truncloffset = length;
459 
460 	crit_enter();
461 	do {
462 		count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
463 				vtruncbuf_bp_trunc_cmp,
464 				vtruncbuf_bp_trunc, &truncloffset);
465 		count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
466 				vtruncbuf_bp_trunc_cmp,
467 				vtruncbuf_bp_trunc, &truncloffset);
468 	} while(count);
469 
470 	/*
471 	 * For safety, fsync any remaining metadata if the file is not being
472 	 * truncated to 0.  Since the metadata does not represent the entire
473 	 * dirty list we have to rely on the hit count to ensure that we get
474 	 * all of it.
475 	 */
476 	if (length > 0) {
477 		do {
478 			count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
479 					vtruncbuf_bp_metasync_cmp,
480 					vtruncbuf_bp_metasync, vp);
481 		} while (count);
482 	}
483 
484 	/*
485 	 * Clean out any left over VM backing store.
486 	 */
487 	crit_exit();
488 
489 	vnode_pager_setsize(vp, length);
490 
491 	crit_enter();
492 
493 	/*
494 	 * It is possible to have in-progress I/O from buffers that were
495 	 * not part of the truncation.  This should not happen if we
496 	 * are truncating to 0-length.
497 	 */
498 	filename = TAILQ_FIRST(&vp->v_namecache) ?
499 		   TAILQ_FIRST(&vp->v_namecache)->nc_name : "?";
500 
501 	while ((count = vp->v_track_write.bk_active) > 0) {
502 		vp->v_track_write.bk_waitflag = 1;
503 		tsleep(&vp->v_track_write, 0, "vbtrunc", 0);
504 		if (length == 0) {
505 			kprintf("Warning: vtruncbuf(): Had to wait for "
506 			       "%d buffer I/Os to finish in %s\n",
507 			       count, filename);
508 		}
509 	}
510 
511 	/*
512 	 * Make sure no buffers were instantiated while we were trying
513 	 * to clean out the remaining VM pages.  This could occur due
514 	 * to busy dirty VM pages being flushed out to disk.
515 	 */
516 	do {
517 		count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
518 				vtruncbuf_bp_trunc_cmp,
519 				vtruncbuf_bp_trunc, &truncloffset);
520 		count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
521 				vtruncbuf_bp_trunc_cmp,
522 				vtruncbuf_bp_trunc, &truncloffset);
523 		if (count) {
524 			kprintf("Warning: vtruncbuf():  Had to re-clean %d "
525 			       "left over buffers in %s\n", count, filename);
526 		}
527 	} while(count);
528 
529 	crit_exit();
530 
531 	return (0);
532 }
533 
534 /*
535  * The callback buffer is beyond the new file EOF and must be destroyed.
536  * Note that the compare function must conform to the RB_SCAN's requirements.
537  */
538 static
539 int
540 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data)
541 {
542 	if (bp->b_loffset >= *(off_t *)data)
543 		return(0);
544 	return(-1);
545 }
546 
547 static
548 int
549 vtruncbuf_bp_trunc(struct buf *bp, void *data)
550 {
551 	/*
552 	 * Do not try to use a buffer we cannot immediately lock, but sleep
553 	 * anyway to prevent a livelock.  The code will loop until all buffers
554 	 * can be acted upon.
555 	 */
556 	if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
557 		if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
558 			BUF_UNLOCK(bp);
559 	} else {
560 		bremfree(bp);
561 		bp->b_flags |= (B_INVAL | B_RELBUF | B_NOCACHE);
562 		bp->b_flags &= ~B_ASYNC;
563 		brelse(bp);
564 	}
565 	return(1);
566 }
567 
568 /*
569  * Fsync all meta-data after truncating a file to be non-zero.  Only metadata
570  * blocks (with a negative loffset) are scanned.
571  * Note that the compare function must conform to the RB_SCAN's requirements.
572  */
573 static int
574 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data)
575 {
576 	if (bp->b_loffset < 0)
577 		return(0);
578 	return(1);
579 }
580 
581 static int
582 vtruncbuf_bp_metasync(struct buf *bp, void *data)
583 {
584 	struct vnode *vp = data;
585 
586 	if (bp->b_flags & B_DELWRI) {
587 		/*
588 		 * Do not try to use a buffer we cannot immediately lock,
589 		 * but sleep anyway to prevent a livelock.  The code will
590 		 * loop until all buffers can be acted upon.
591 		 */
592 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
593 			if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
594 				BUF_UNLOCK(bp);
595 		} else {
596 			bremfree(bp);
597 			if (bp->b_vp == vp) {
598 				bp->b_flags |= B_ASYNC;
599 			} else {
600 				bp->b_flags &= ~B_ASYNC;
601 			}
602 			bwrite(bp);
603 		}
604 		return(1);
605 	} else {
606 		return(0);
607 	}
608 }
609 
610 /*
611  * vfsync - implements a multipass fsync on a file which understands
612  * dependancies and meta-data.  The passed vnode must be locked.  The
613  * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY.
614  *
615  * When fsyncing data asynchronously just do one consolidated pass starting
616  * with the most negative block number.  This may not get all the data due
617  * to dependancies.
618  *
619  * When fsyncing data synchronously do a data pass, then a metadata pass,
620  * then do additional data+metadata passes to try to get all the data out.
621  */
622 static int vfsync_wait_output(struct vnode *vp,
623 			    int (*waitoutput)(struct vnode *, struct thread *));
624 static int vfsync_data_only_cmp(struct buf *bp, void *data);
625 static int vfsync_meta_only_cmp(struct buf *bp, void *data);
626 static int vfsync_lazy_range_cmp(struct buf *bp, void *data);
627 static int vfsync_bp(struct buf *bp, void *data);
628 
629 struct vfsync_info {
630 	struct vnode *vp;
631 	int synchronous;
632 	int syncdeps;
633 	int lazycount;
634 	int lazylimit;
635 	int skippedbufs;
636 	int (*checkdef)(struct buf *);
637 };
638 
639 int
640 vfsync(struct vnode *vp, int waitfor, int passes,
641 	int (*checkdef)(struct buf *),
642 	int (*waitoutput)(struct vnode *, struct thread *))
643 {
644 	struct vfsync_info info;
645 	int error;
646 
647 	bzero(&info, sizeof(info));
648 	info.vp = vp;
649 	if ((info.checkdef = checkdef) == NULL)
650 		info.syncdeps = 1;
651 
652 	crit_enter_id("vfsync");
653 
654 	switch(waitfor) {
655 	case MNT_LAZY:
656 		/*
657 		 * Lazy (filesystem syncer typ) Asynchronous plus limit the
658 		 * number of data (not meta) pages we try to flush to 1MB.
659 		 * A non-zero return means that lazy limit was reached.
660 		 */
661 		info.lazylimit = 1024 * 1024;
662 		info.syncdeps = 1;
663 		error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
664 				vfsync_lazy_range_cmp, vfsync_bp, &info);
665 		RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
666 				vfsync_meta_only_cmp, vfsync_bp, &info);
667 		if (error == 0)
668 			vp->v_lazyw = 0;
669 		else if (!RB_EMPTY(&vp->v_rbdirty_tree))
670 			vn_syncer_add_to_worklist(vp, 1);
671 		error = 0;
672 		break;
673 	case MNT_NOWAIT:
674 		/*
675 		 * Asynchronous.  Do a data-only pass and a meta-only pass.
676 		 */
677 		info.syncdeps = 1;
678 		RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
679 			vfsync_bp, &info);
680 		RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp,
681 			vfsync_bp, &info);
682 		error = 0;
683 		break;
684 	default:
685 		/*
686 		 * Synchronous.  Do a data-only pass, then a meta-data+data
687 		 * pass, then additional integrated passes to try to get
688 		 * all the dependancies flushed.
689 		 */
690 		RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
691 			vfsync_bp, &info);
692 		error = vfsync_wait_output(vp, waitoutput);
693 		if (error == 0) {
694 			info.skippedbufs = 0;
695 			RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
696 				vfsync_bp, &info);
697 			error = vfsync_wait_output(vp, waitoutput);
698 			if (info.skippedbufs)
699 				kprintf("Warning: vfsync skipped %d dirty bufs in pass2!\n", info.skippedbufs);
700 		}
701 		while (error == 0 && passes > 0 &&
702 		    !RB_EMPTY(&vp->v_rbdirty_tree)) {
703 			if (--passes == 0) {
704 				info.synchronous = 1;
705 				info.syncdeps = 1;
706 			}
707 			error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
708 				vfsync_bp, &info);
709 			if (error < 0)
710 				error = -error;
711 			info.syncdeps = 1;
712 			if (error == 0)
713 				error = vfsync_wait_output(vp, waitoutput);
714 		}
715 		break;
716 	}
717 	crit_exit_id("vfsync");
718 	return(error);
719 }
720 
721 static int
722 vfsync_wait_output(struct vnode *vp, int (*waitoutput)(struct vnode *, struct thread *))
723 {
724 	int error = 0;
725 
726 	while (vp->v_track_write.bk_active) {
727 		vp->v_track_write.bk_waitflag = 1;
728 		tsleep(&vp->v_track_write, 0, "fsfsn", 0);
729 	}
730 	if (waitoutput)
731 		error = waitoutput(vp, curthread);
732 	return(error);
733 }
734 
735 static int
736 vfsync_data_only_cmp(struct buf *bp, void *data)
737 {
738 	if (bp->b_loffset < 0)
739 		return(-1);
740 	return(0);
741 }
742 
743 static int
744 vfsync_meta_only_cmp(struct buf *bp, void *data)
745 {
746 	if (bp->b_loffset < 0)
747 		return(0);
748 	return(1);
749 }
750 
751 static int
752 vfsync_lazy_range_cmp(struct buf *bp, void *data)
753 {
754 	struct vfsync_info *info = data;
755 	if (bp->b_loffset < info->vp->v_lazyw)
756 		return(-1);
757 	return(0);
758 }
759 
760 static int
761 vfsync_bp(struct buf *bp, void *data)
762 {
763 	struct vfsync_info *info = data;
764 	struct vnode *vp = info->vp;
765 	int error;
766 
767 	/*
768 	 * if syncdeps is not set we do not try to write buffers which have
769 	 * dependancies.
770 	 */
771 	if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp))
772 		return(0);
773 
774 	/*
775 	 * Ignore buffers that we cannot immediately lock.  XXX
776 	 */
777 	if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
778 		kprintf("Warning: vfsync_bp skipping dirty buffer %p\n", bp);
779 		++info->skippedbufs;
780 		return(0);
781 	}
782 	if ((bp->b_flags & B_DELWRI) == 0)
783 		panic("vfsync_bp: buffer not dirty");
784 	if (vp != bp->b_vp)
785 		panic("vfsync_bp: buffer vp mismatch");
786 
787 	/*
788 	 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer
789 	 * has been written but an additional handshake with the device
790 	 * is required before we can dispose of the buffer.  We have no idea
791 	 * how to do this so we have to skip these buffers.
792 	 */
793 	if (bp->b_flags & B_NEEDCOMMIT) {
794 		BUF_UNLOCK(bp);
795 		return(0);
796 	}
797 
798 	/*
799 	 * Ask bioops if it is ok to sync
800 	 */
801 	if (LIST_FIRST(&bp->b_dep) != NULL && buf_checkwrite(bp)) {
802 		bremfree(bp);
803 		brelse(bp);
804 		return(0);
805 	}
806 
807 	if (info->synchronous) {
808 		/*
809 		 * Synchronous flushing.  An error may be returned.
810 		 */
811 		bremfree(bp);
812 		crit_exit_id("vfsync");
813 		error = bwrite(bp);
814 		crit_enter_id("vfsync");
815 	} else {
816 		/*
817 		 * Asynchronous flushing.  A negative return value simply
818 		 * stops the scan and is not considered an error.  We use
819 		 * this to support limited MNT_LAZY flushes.
820 		 */
821 		vp->v_lazyw = bp->b_loffset;
822 		if ((vp->v_flag & VOBJBUF) && (bp->b_flags & B_CLUSTEROK)) {
823 			info->lazycount += vfs_bio_awrite(bp);
824 		} else {
825 			info->lazycount += bp->b_bufsize;
826 			bremfree(bp);
827 			crit_exit_id("vfsync");
828 			bawrite(bp);
829 			crit_enter_id("vfsync");
830 		}
831 		if (info->lazylimit && info->lazycount >= info->lazylimit)
832 			error = 1;
833 		else
834 			error = 0;
835 	}
836 	return(-error);
837 }
838 
839 /*
840  * Associate a buffer with a vnode.
841  */
842 void
843 bgetvp(struct vnode *vp, struct buf *bp)
844 {
845 	KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
846 	KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0);
847 
848 	vhold(vp);
849 	/*
850 	 * Insert onto list for new vnode.
851 	 */
852 	crit_enter();
853 	bp->b_vp = vp;
854 	bp->b_flags |= B_HASHED;
855 	if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp))
856 		panic("reassignbuf: dup lblk vp %p bp %p", vp, bp);
857 
858 	bp->b_flags |= B_VNCLEAN;
859 	if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp))
860 		panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp);
861 	crit_exit();
862 }
863 
864 /*
865  * Disassociate a buffer from a vnode.
866  */
867 void
868 brelvp(struct buf *bp)
869 {
870 	struct vnode *vp;
871 
872 	KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
873 
874 	/*
875 	 * Delete from old vnode list, if on one.
876 	 */
877 	vp = bp->b_vp;
878 	crit_enter();
879 	if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) {
880 		if (bp->b_flags & B_VNDIRTY)
881 			buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
882 		else
883 			buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
884 		bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN);
885 	}
886 	if (bp->b_flags & B_HASHED) {
887 		buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp);
888 		bp->b_flags &= ~B_HASHED;
889 	}
890 	if ((vp->v_flag & VONWORKLST) && RB_EMPTY(&vp->v_rbdirty_tree)) {
891 		vp->v_flag &= ~VONWORKLST;
892 		LIST_REMOVE(vp, v_synclist);
893 	}
894 	crit_exit();
895 	bp->b_vp = NULL;
896 	vdrop(vp);
897 }
898 
899 /*
900  * Reassign the buffer to the proper clean/dirty list based on B_DELWRI.
901  * This routine is called when the state of the B_DELWRI bit is changed.
902  */
903 void
904 reassignbuf(struct buf *bp)
905 {
906 	struct vnode *vp = bp->b_vp;
907 	int delay;
908 
909 	KKASSERT(vp != NULL);
910 	++reassignbufcalls;
911 
912 	/*
913 	 * B_PAGING flagged buffers cannot be reassigned because their vp
914 	 * is not fully linked in.
915 	 */
916 	if (bp->b_flags & B_PAGING)
917 		panic("cannot reassign paging buffer");
918 
919 	crit_enter();
920 	if (bp->b_flags & B_DELWRI) {
921 		/*
922 		 * Move to the dirty list, add the vnode to the worklist
923 		 */
924 		if (bp->b_flags & B_VNCLEAN) {
925 			buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
926 			bp->b_flags &= ~B_VNCLEAN;
927 		}
928 		if ((bp->b_flags & B_VNDIRTY) == 0) {
929 			if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) {
930 				panic("reassignbuf: dup lblk vp %p bp %p",
931 				      vp, bp);
932 			}
933 			bp->b_flags |= B_VNDIRTY;
934 		}
935 		if ((vp->v_flag & VONWORKLST) == 0) {
936 			switch (vp->v_type) {
937 			case VDIR:
938 				delay = dirdelay;
939 				break;
940 			case VCHR:
941 			case VBLK:
942 				if (vp->v_rdev &&
943 				    vp->v_rdev->si_mountpoint != NULL) {
944 					delay = metadelay;
945 					break;
946 				}
947 				/* fall through */
948 			default:
949 				delay = filedelay;
950 			}
951 			vn_syncer_add_to_worklist(vp, delay);
952 		}
953 	} else {
954 		/*
955 		 * Move to the clean list, remove the vnode from the worklist
956 		 * if no dirty blocks remain.
957 		 */
958 		if (bp->b_flags & B_VNDIRTY) {
959 			buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
960 			bp->b_flags &= ~B_VNDIRTY;
961 		}
962 		if ((bp->b_flags & B_VNCLEAN) == 0) {
963 			if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) {
964 				panic("reassignbuf: dup lblk vp %p bp %p",
965 				      vp, bp);
966 			}
967 			bp->b_flags |= B_VNCLEAN;
968 		}
969 		if ((vp->v_flag & VONWORKLST) &&
970 		    RB_EMPTY(&vp->v_rbdirty_tree)) {
971 			vp->v_flag &= ~VONWORKLST;
972 			LIST_REMOVE(vp, v_synclist);
973 		}
974 	}
975 	crit_exit();
976 }
977 
978 /*
979  * Create a vnode for a block device.
980  * Used for mounting the root file system.
981  */
982 int
983 bdevvp(cdev_t dev, struct vnode **vpp)
984 {
985 	struct vnode *vp;
986 	struct vnode *nvp;
987 	int error;
988 
989 	if (dev == NULL) {
990 		*vpp = NULLVP;
991 		return (ENXIO);
992 	}
993 	error = getspecialvnode(VT_NON, NULL, &spec_vnode_vops_p, &nvp, 0, 0);
994 	if (error) {
995 		*vpp = NULLVP;
996 		return (error);
997 	}
998 	vp = nvp;
999 	vp->v_type = VCHR;
1000 	vp->v_umajor = dev->si_umajor;
1001 	vp->v_uminor = dev->si_uminor;
1002 	vx_unlock(vp);
1003 	*vpp = vp;
1004 	return (0);
1005 }
1006 
1007 int
1008 v_associate_rdev(struct vnode *vp, cdev_t dev)
1009 {
1010 	lwkt_tokref ilock;
1011 
1012 	if (dev == NULL)
1013 		return(ENXIO);
1014 	if (dev_is_good(dev) == 0)
1015 		return(ENXIO);
1016 	KKASSERT(vp->v_rdev == NULL);
1017 	if (dev_ref_debug)
1018 		kprintf("Z1");
1019 	vp->v_rdev = reference_dev(dev);
1020 	lwkt_gettoken(&ilock, &spechash_token);
1021 	SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext);
1022 	lwkt_reltoken(&ilock);
1023 	return(0);
1024 }
1025 
1026 void
1027 v_release_rdev(struct vnode *vp)
1028 {
1029 	lwkt_tokref ilock;
1030 	cdev_t dev;
1031 
1032 	if ((dev = vp->v_rdev) != NULL) {
1033 		lwkt_gettoken(&ilock, &spechash_token);
1034 		SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_cdevnext);
1035 		vp->v_rdev = NULL;
1036 		release_dev(dev);
1037 		lwkt_reltoken(&ilock);
1038 	}
1039 }
1040 
1041 /*
1042  * Add a vnode to the alias list hung off the cdev_t.  We only associate
1043  * the device number with the vnode.  The actual device is not associated
1044  * until the vnode is opened (usually in spec_open()), and will be
1045  * disassociated on last close.
1046  */
1047 void
1048 addaliasu(struct vnode *nvp, int x, int y)
1049 {
1050 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
1051 		panic("addaliasu on non-special vnode");
1052 	nvp->v_umajor = x;
1053 	nvp->v_uminor = y;
1054 }
1055 
1056 /*
1057  * Simple call that a filesystem can make to try to get rid of a
1058  * vnode.  It will fail if anyone is referencing the vnode (including
1059  * the caller).
1060  *
1061  * The filesystem can check whether its in-memory inode structure still
1062  * references the vp on return.
1063  */
1064 void
1065 vclean_unlocked(struct vnode *vp)
1066 {
1067 	vx_get(vp);
1068 	if (sysref_isactive(&vp->v_sysref) == 0)
1069 		vgone_vxlocked(vp);
1070 	vx_put(vp);
1071 }
1072 
1073 /*
1074  * Disassociate a vnode from its underlying filesystem.
1075  *
1076  * The vnode must be VX locked and referenced.  In all normal situations
1077  * there are no active references.  If vclean_vxlocked() is called while
1078  * there are active references, the vnode is being ripped out and we have
1079  * to call VOP_CLOSE() as appropriate before we can reclaim it.
1080  */
1081 void
1082 vclean_vxlocked(struct vnode *vp, int flags)
1083 {
1084 	int active;
1085 	int n;
1086 	vm_object_t object;
1087 
1088 	/*
1089 	 * If the vnode has already been reclaimed we have nothing to do.
1090 	 */
1091 	if (vp->v_flag & VRECLAIMED)
1092 		return;
1093 	vp->v_flag |= VRECLAIMED;
1094 
1095 	/*
1096 	 * Scrap the vfs cache
1097 	 */
1098 	while (cache_inval_vp(vp, 0) != 0) {
1099 		kprintf("Warning: vnode %p clean/cache_resolution race detected\n", vp);
1100 		tsleep(vp, 0, "vclninv", 2);
1101 	}
1102 
1103 	/*
1104 	 * Check to see if the vnode is in use. If so we have to reference it
1105 	 * before we clean it out so that its count cannot fall to zero and
1106 	 * generate a race against ourselves to recycle it.
1107 	 */
1108 	active = sysref_isactive(&vp->v_sysref);
1109 
1110 	/*
1111 	 * Clean out any buffers associated with the vnode and destroy its
1112 	 * object, if it has one.
1113 	 */
1114 	vinvalbuf(vp, V_SAVE, 0, 0);
1115 
1116 	/*
1117 	 * If purging an active vnode (typically during a forced unmount
1118 	 * or reboot), it must be closed and deactivated before being
1119 	 * reclaimed.  This isn't really all that safe, but what can
1120 	 * we do? XXX.
1121 	 *
1122 	 * Note that neither of these routines unlocks the vnode.
1123 	 */
1124 	if (active && (flags & DOCLOSE)) {
1125 		while ((n = vp->v_opencount) != 0) {
1126 			if (vp->v_writecount)
1127 				VOP_CLOSE(vp, FWRITE|FNONBLOCK);
1128 			else
1129 				VOP_CLOSE(vp, FNONBLOCK);
1130 			if (vp->v_opencount == n) {
1131 				kprintf("Warning: unable to force-close"
1132 				       " vnode %p\n", vp);
1133 				break;
1134 			}
1135 		}
1136 	}
1137 
1138 	/*
1139 	 * If the vnode has not been deactivated, deactivated it.  Deactivation
1140 	 * can create new buffers and VM pages so we have to call vinvalbuf()
1141 	 * again to make sure they all get flushed.
1142 	 *
1143 	 * This can occur if a file with a link count of 0 needs to be
1144 	 * truncated.
1145 	 */
1146 	if ((vp->v_flag & VINACTIVE) == 0) {
1147 		vp->v_flag |= VINACTIVE;
1148 		VOP_INACTIVE(vp);
1149 		vinvalbuf(vp, V_SAVE, 0, 0);
1150 	}
1151 
1152 	/*
1153 	 * If the vnode has an object, destroy it.
1154 	 */
1155 	if ((object = vp->v_object) != NULL) {
1156 		if (object->ref_count == 0) {
1157 			if ((object->flags & OBJ_DEAD) == 0)
1158 				vm_object_terminate(object);
1159 		} else {
1160 			vm_pager_deallocate(object);
1161 		}
1162 		vp->v_flag &= ~VOBJBUF;
1163 	}
1164 	KKASSERT((vp->v_flag & VOBJBUF) == 0);
1165 
1166 	/*
1167 	 * Reclaim the vnode.
1168 	 */
1169 	if (VOP_RECLAIM(vp))
1170 		panic("vclean: cannot reclaim");
1171 
1172 	/*
1173 	 * Done with purge, notify sleepers of the grim news.
1174 	 */
1175 	vp->v_ops = &dead_vnode_vops_p;
1176 	vn_pollgone(vp);
1177 	vp->v_tag = VT_NON;
1178 
1179 	/*
1180 	 * If we are destroying an active vnode, reactivate it now that
1181 	 * we have reassociated it with deadfs.  This prevents the system
1182 	 * from crashing on the vnode due to it being unexpectedly marked
1183 	 * as inactive or reclaimed.
1184 	 */
1185 	if (active && (flags & DOCLOSE)) {
1186 		vp->v_flag &= ~(VINACTIVE|VRECLAIMED);
1187 	}
1188 }
1189 
1190 /*
1191  * Eliminate all activity associated with the requested vnode
1192  * and with all vnodes aliased to the requested vnode.
1193  *
1194  * The vnode must be referenced and vx_lock()'d
1195  *
1196  * revoke { struct vnode *a_vp, int a_flags }
1197  */
1198 int
1199 vop_stdrevoke(struct vop_revoke_args *ap)
1200 {
1201 	struct vnode *vp, *vq;
1202 	lwkt_tokref ilock;
1203 	cdev_t dev;
1204 
1205 	KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
1206 
1207 	vp = ap->a_vp;
1208 
1209 	/*
1210 	 * If the vnode is already dead don't try to revoke it
1211 	 */
1212 	if (vp->v_flag & VRECLAIMED)
1213 		return (0);
1214 
1215 	/*
1216 	 * If the vnode has a device association, scrap all vnodes associated
1217 	 * with the device.  Don't let the device disappear on us while we
1218 	 * are scrapping the vnodes.
1219 	 *
1220 	 * The passed vp will probably show up in the list, do not VX lock
1221 	 * it twice!
1222 	 */
1223 	if (vp->v_type != VCHR)
1224 		return(0);
1225 	if ((dev = vp->v_rdev) == NULL) {
1226 		if ((dev = get_dev(vp->v_umajor, vp->v_uminor)) == NULL)
1227 			return(0);
1228 	}
1229 	reference_dev(dev);
1230 	lwkt_gettoken(&ilock, &spechash_token);
1231 	while ((vq = SLIST_FIRST(&dev->si_hlist)) != NULL) {
1232 		if (vp != vq)
1233 			vx_get(vq);
1234 		if (vq == SLIST_FIRST(&dev->si_hlist))
1235 			vgone_vxlocked(vq);
1236 		if (vp != vq)
1237 			vx_put(vq);
1238 	}
1239 	lwkt_reltoken(&ilock);
1240 	release_dev(dev);
1241 	return (0);
1242 }
1243 
1244 /*
1245  * This is called when the object underlying a vnode is being destroyed,
1246  * such as in a remove().  Try to recycle the vnode immediately if the
1247  * only active reference is our reference.
1248  *
1249  * Directory vnodes in the namecache with children cannot be immediately
1250  * recycled because numerous VOP_N*() ops require them to be stable.
1251  */
1252 int
1253 vrecycle(struct vnode *vp)
1254 {
1255 	if (vp->v_sysref.refcnt <= 1) {
1256 		if (cache_inval_vp_nonblock(vp))
1257 			return(0);
1258 		vgone_vxlocked(vp);
1259 		return (1);
1260 	}
1261 	return (0);
1262 }
1263 
1264 /*
1265  * Return the maximum I/O size allowed for strategy calls on VP.
1266  *
1267  * If vp is VCHR or VBLK we dive the device, otherwise we use
1268  * the vp's mount info.
1269  */
1270 int
1271 vmaxiosize(struct vnode *vp)
1272 {
1273 	if (vp->v_type == VBLK || vp->v_type == VCHR) {
1274 		return(vp->v_rdev->si_iosize_max);
1275 	} else {
1276 		return(vp->v_mount->mnt_iosize_max);
1277 	}
1278 }
1279 
1280 /*
1281  * Eliminate all activity associated with a vnode in preparation for reuse.
1282  *
1283  * The vnode must be VX locked and refd and will remain VX locked and refd
1284  * on return.  This routine may be called with the vnode in any state, as
1285  * long as it is VX locked.  The vnode will be cleaned out and marked
1286  * VRECLAIMED but will not actually be reused until all existing refs and
1287  * holds go away.
1288  *
1289  * NOTE: This routine may be called on a vnode which has not yet been
1290  * already been deactivated (VOP_INACTIVE), or on a vnode which has
1291  * already been reclaimed.
1292  *
1293  * This routine is not responsible for placing us back on the freelist.
1294  * Instead, it happens automatically when the caller releases the VX lock
1295  * (assuming there aren't any other references).
1296  */
1297 
1298 void
1299 vgone_vxlocked(struct vnode *vp)
1300 {
1301 	/*
1302 	 * assert that the VX lock is held.  This is an absolute requirement
1303 	 * now for vgone_vxlocked() to be called.
1304 	 */
1305 	KKASSERT(vp->v_lock.lk_exclusivecount == 1);
1306 
1307 	/*
1308 	 * Clean out the filesystem specific data and set the VRECLAIMED
1309 	 * bit.  Also deactivate the vnode if necessary.
1310 	 */
1311 	vclean_vxlocked(vp, DOCLOSE);
1312 
1313 	/*
1314 	 * Delete from old mount point vnode list, if on one.
1315 	 */
1316 	if (vp->v_mount != NULL)
1317 		insmntque(vp, NULL);
1318 
1319 	/*
1320 	 * If special device, remove it from special device alias list
1321 	 * if it is on one.  This should normally only occur if a vnode is
1322 	 * being revoked as the device should otherwise have been released
1323 	 * naturally.
1324 	 */
1325 	if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
1326 		v_release_rdev(vp);
1327 	}
1328 
1329 	/*
1330 	 * Set us to VBAD
1331 	 */
1332 	vp->v_type = VBAD;
1333 }
1334 
1335 /*
1336  * Lookup a vnode by device number.
1337  */
1338 int
1339 vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp)
1340 {
1341 	lwkt_tokref ilock;
1342 	struct vnode *vp;
1343 
1344 	lwkt_gettoken(&ilock, &spechash_token);
1345 	SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
1346 		if (type == vp->v_type) {
1347 			*vpp = vp;
1348 			lwkt_reltoken(&ilock);
1349 			return (1);
1350 		}
1351 	}
1352 	lwkt_reltoken(&ilock);
1353 	return (0);
1354 }
1355 
1356 /*
1357  * Calculate the total number of references to a special device.  This
1358  * routine may only be called for VBLK and VCHR vnodes since v_rdev is
1359  * an overloaded field.  Since udev2dev can now return NULL, we have
1360  * to check for a NULL v_rdev.
1361  */
1362 int
1363 count_dev(cdev_t dev)
1364 {
1365 	lwkt_tokref ilock;
1366 	struct vnode *vp;
1367 	int count = 0;
1368 
1369 	if (SLIST_FIRST(&dev->si_hlist)) {
1370 		lwkt_gettoken(&ilock, &spechash_token);
1371 		SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
1372 			if (vp->v_sysref.refcnt > 0)
1373 				count += vp->v_sysref.refcnt;
1374 		}
1375 		lwkt_reltoken(&ilock);
1376 	}
1377 	return(count);
1378 }
1379 
1380 int
1381 count_udev(int x, int y)
1382 {
1383 	cdev_t dev;
1384 
1385 	if ((dev = get_dev(x, y)) == NULL)
1386 		return(0);
1387 	return(count_dev(dev));
1388 }
1389 
1390 int
1391 vcount(struct vnode *vp)
1392 {
1393 	if (vp->v_rdev == NULL)
1394 		return(0);
1395 	return(count_dev(vp->v_rdev));
1396 }
1397 
1398 /*
1399  * Initialize VMIO for a vnode.  This routine MUST be called before a
1400  * VFS can issue buffer cache ops on a vnode.  It is typically called
1401  * when a vnode is initialized from its inode.
1402  */
1403 int
1404 vinitvmio(struct vnode *vp, off_t filesize)
1405 {
1406 	vm_object_t object;
1407 	int error = 0;
1408 
1409 retry:
1410 	if ((object = vp->v_object) == NULL) {
1411 		object = vnode_pager_alloc(vp, filesize, 0, 0);
1412 		/*
1413 		 * Dereference the reference we just created.  This assumes
1414 		 * that the object is associated with the vp.
1415 		 */
1416 		object->ref_count--;
1417 		vrele(vp);
1418 	} else {
1419 		if (object->flags & OBJ_DEAD) {
1420 			vn_unlock(vp);
1421 			vm_object_dead_sleep(object, "vodead");
1422 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1423 			goto retry;
1424 		}
1425 	}
1426 	KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object"));
1427 	vp->v_flag |= VOBJBUF;
1428 	return (error);
1429 }
1430 
1431 
1432 /*
1433  * Print out a description of a vnode.
1434  */
1435 static char *typename[] =
1436 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1437 
1438 void
1439 vprint(char *label, struct vnode *vp)
1440 {
1441 	char buf[96];
1442 
1443 	if (label != NULL)
1444 		kprintf("%s: %p: ", label, (void *)vp);
1445 	else
1446 		kprintf("%p: ", (void *)vp);
1447 	kprintf("type %s, sysrefs %d, writecount %d, holdcnt %d,",
1448 		typename[vp->v_type],
1449 		vp->v_sysref.refcnt, vp->v_writecount, vp->v_auxrefs);
1450 	buf[0] = '\0';
1451 	if (vp->v_flag & VROOT)
1452 		strcat(buf, "|VROOT");
1453 	if (vp->v_flag & VPFSROOT)
1454 		strcat(buf, "|VPFSROOT");
1455 	if (vp->v_flag & VTEXT)
1456 		strcat(buf, "|VTEXT");
1457 	if (vp->v_flag & VSYSTEM)
1458 		strcat(buf, "|VSYSTEM");
1459 	if (vp->v_flag & VFREE)
1460 		strcat(buf, "|VFREE");
1461 	if (vp->v_flag & VOBJBUF)
1462 		strcat(buf, "|VOBJBUF");
1463 	if (buf[0] != '\0')
1464 		kprintf(" flags (%s)", &buf[1]);
1465 	if (vp->v_data == NULL) {
1466 		kprintf("\n");
1467 	} else {
1468 		kprintf("\n\t");
1469 		VOP_PRINT(vp);
1470 	}
1471 }
1472 
1473 #ifdef DDB
1474 #include <ddb/ddb.h>
1475 
1476 static int db_show_locked_vnodes(struct mount *mp, void *data);
1477 
1478 /*
1479  * List all of the locked vnodes in the system.
1480  * Called when debugging the kernel.
1481  */
1482 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
1483 {
1484 	kprintf("Locked vnodes\n");
1485 	mountlist_scan(db_show_locked_vnodes, NULL,
1486 			MNTSCAN_FORWARD|MNTSCAN_NOBUSY);
1487 }
1488 
1489 static int
1490 db_show_locked_vnodes(struct mount *mp, void *data __unused)
1491 {
1492 	struct vnode *vp;
1493 
1494 	TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
1495 		if (vn_islocked(vp))
1496 			vprint((char *)0, vp);
1497 	}
1498 	return(0);
1499 }
1500 #endif
1501 
1502 /*
1503  * Top level filesystem related information gathering.
1504  */
1505 static int	sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
1506 
1507 static int
1508 vfs_sysctl(SYSCTL_HANDLER_ARGS)
1509 {
1510 	int *name = (int *)arg1 - 1;	/* XXX */
1511 	u_int namelen = arg2 + 1;	/* XXX */
1512 	struct vfsconf *vfsp;
1513 
1514 #if 1 || defined(COMPAT_PRELITE2)
1515 	/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1516 	if (namelen == 1)
1517 		return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
1518 #endif
1519 
1520 #ifdef notyet
1521 	/* all sysctl names at this level are at least name and field */
1522 	if (namelen < 2)
1523 		return (ENOTDIR);		/* overloaded */
1524 	if (name[0] != VFS_GENERIC) {
1525 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
1526 			if (vfsp->vfc_typenum == name[0])
1527 				break;
1528 		if (vfsp == NULL)
1529 			return (EOPNOTSUPP);
1530 		return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
1531 		    oldp, oldlenp, newp, newlen, p));
1532 	}
1533 #endif
1534 	switch (name[1]) {
1535 	case VFS_MAXTYPENUM:
1536 		if (namelen != 2)
1537 			return (ENOTDIR);
1538 		return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
1539 	case VFS_CONF:
1540 		if (namelen != 3)
1541 			return (ENOTDIR);	/* overloaded */
1542 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
1543 			if (vfsp->vfc_typenum == name[2])
1544 				break;
1545 		if (vfsp == NULL)
1546 			return (EOPNOTSUPP);
1547 		return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
1548 	}
1549 	return (EOPNOTSUPP);
1550 }
1551 
1552 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
1553 	"Generic filesystem");
1554 
1555 #if 1 || defined(COMPAT_PRELITE2)
1556 
1557 static int
1558 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
1559 {
1560 	int error;
1561 	struct vfsconf *vfsp;
1562 	struct ovfsconf ovfs;
1563 
1564 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
1565 		bzero(&ovfs, sizeof(ovfs));
1566 		ovfs.vfc_vfsops = vfsp->vfc_vfsops;	/* XXX used as flag */
1567 		strcpy(ovfs.vfc_name, vfsp->vfc_name);
1568 		ovfs.vfc_index = vfsp->vfc_typenum;
1569 		ovfs.vfc_refcount = vfsp->vfc_refcount;
1570 		ovfs.vfc_flags = vfsp->vfc_flags;
1571 		error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
1572 		if (error)
1573 			return error;
1574 	}
1575 	return 0;
1576 }
1577 
1578 #endif /* 1 || COMPAT_PRELITE2 */
1579 
1580 /*
1581  * Check to see if a filesystem is mounted on a block device.
1582  */
1583 int
1584 vfs_mountedon(struct vnode *vp)
1585 {
1586 	cdev_t dev;
1587 
1588 	if ((dev = vp->v_rdev) == NULL) {
1589 		if (vp->v_type != VBLK)
1590 			dev = get_dev(vp->v_uminor, vp->v_umajor);
1591 	}
1592 	if (dev != NULL && dev->si_mountpoint)
1593 		return (EBUSY);
1594 	return (0);
1595 }
1596 
1597 /*
1598  * Unmount all filesystems. The list is traversed in reverse order
1599  * of mounting to avoid dependencies.
1600  */
1601 
1602 static int vfs_umountall_callback(struct mount *mp, void *data);
1603 
1604 void
1605 vfs_unmountall(void)
1606 {
1607 	int count;
1608 
1609 	do {
1610 		count = mountlist_scan(vfs_umountall_callback,
1611 					NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY);
1612 	} while (count);
1613 }
1614 
1615 static
1616 int
1617 vfs_umountall_callback(struct mount *mp, void *data)
1618 {
1619 	int error;
1620 
1621 	error = dounmount(mp, MNT_FORCE);
1622 	if (error) {
1623 		mountlist_remove(mp);
1624 		kprintf("unmount of filesystem mounted from %s failed (",
1625 			mp->mnt_stat.f_mntfromname);
1626 		if (error == EBUSY)
1627 			kprintf("BUSY)\n");
1628 		else
1629 			kprintf("%d)\n", error);
1630 	}
1631 	return(1);
1632 }
1633 
1634 /*
1635  * Build hash lists of net addresses and hang them off the mount point.
1636  * Called by ufs_mount() to set up the lists of export addresses.
1637  */
1638 static int
1639 vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
1640 		const struct export_args *argp)
1641 {
1642 	struct netcred *np;
1643 	struct radix_node_head *rnh;
1644 	int i;
1645 	struct radix_node *rn;
1646 	struct sockaddr *saddr, *smask = 0;
1647 	struct domain *dom;
1648 	int error;
1649 
1650 	if (argp->ex_addrlen == 0) {
1651 		if (mp->mnt_flag & MNT_DEFEXPORTED)
1652 			return (EPERM);
1653 		np = &nep->ne_defexported;
1654 		np->netc_exflags = argp->ex_flags;
1655 		np->netc_anon = argp->ex_anon;
1656 		np->netc_anon.cr_ref = 1;
1657 		mp->mnt_flag |= MNT_DEFEXPORTED;
1658 		return (0);
1659 	}
1660 
1661 	if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
1662 		return (EINVAL);
1663 	if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
1664 		return (EINVAL);
1665 
1666 	i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
1667 	np = (struct netcred *) kmalloc(i, M_NETADDR, M_WAITOK | M_ZERO);
1668 	saddr = (struct sockaddr *) (np + 1);
1669 	if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
1670 		goto out;
1671 	if (saddr->sa_len > argp->ex_addrlen)
1672 		saddr->sa_len = argp->ex_addrlen;
1673 	if (argp->ex_masklen) {
1674 		smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
1675 		error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen);
1676 		if (error)
1677 			goto out;
1678 		if (smask->sa_len > argp->ex_masklen)
1679 			smask->sa_len = argp->ex_masklen;
1680 	}
1681 	i = saddr->sa_family;
1682 	if ((rnh = nep->ne_rtable[i]) == 0) {
1683 		/*
1684 		 * Seems silly to initialize every AF when most are not used,
1685 		 * do so on demand here
1686 		 */
1687 		SLIST_FOREACH(dom, &domains, dom_next)
1688 			if (dom->dom_family == i && dom->dom_rtattach) {
1689 				dom->dom_rtattach((void **) &nep->ne_rtable[i],
1690 				    dom->dom_rtoffset);
1691 				break;
1692 			}
1693 		if ((rnh = nep->ne_rtable[i]) == 0) {
1694 			error = ENOBUFS;
1695 			goto out;
1696 		}
1697 	}
1698 	rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh,
1699 	    np->netc_rnodes);
1700 	if (rn == 0 || np != (struct netcred *) rn) {	/* already exists */
1701 		error = EPERM;
1702 		goto out;
1703 	}
1704 	np->netc_exflags = argp->ex_flags;
1705 	np->netc_anon = argp->ex_anon;
1706 	np->netc_anon.cr_ref = 1;
1707 	return (0);
1708 out:
1709 	kfree(np, M_NETADDR);
1710 	return (error);
1711 }
1712 
1713 /* ARGSUSED */
1714 static int
1715 vfs_free_netcred(struct radix_node *rn, void *w)
1716 {
1717 	struct radix_node_head *rnh = (struct radix_node_head *) w;
1718 
1719 	(*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
1720 	kfree((caddr_t) rn, M_NETADDR);
1721 	return (0);
1722 }
1723 
1724 /*
1725  * Free the net address hash lists that are hanging off the mount points.
1726  */
1727 static void
1728 vfs_free_addrlist(struct netexport *nep)
1729 {
1730 	int i;
1731 	struct radix_node_head *rnh;
1732 
1733 	for (i = 0; i <= AF_MAX; i++)
1734 		if ((rnh = nep->ne_rtable[i])) {
1735 			(*rnh->rnh_walktree) (rnh, vfs_free_netcred,
1736 			    (caddr_t) rnh);
1737 			kfree((caddr_t) rnh, M_RTABLE);
1738 			nep->ne_rtable[i] = 0;
1739 		}
1740 }
1741 
1742 int
1743 vfs_export(struct mount *mp, struct netexport *nep,
1744 	   const struct export_args *argp)
1745 {
1746 	int error;
1747 
1748 	if (argp->ex_flags & MNT_DELEXPORT) {
1749 		if (mp->mnt_flag & MNT_EXPUBLIC) {
1750 			vfs_setpublicfs(NULL, NULL, NULL);
1751 			mp->mnt_flag &= ~MNT_EXPUBLIC;
1752 		}
1753 		vfs_free_addrlist(nep);
1754 		mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
1755 	}
1756 	if (argp->ex_flags & MNT_EXPORTED) {
1757 		if (argp->ex_flags & MNT_EXPUBLIC) {
1758 			if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
1759 				return (error);
1760 			mp->mnt_flag |= MNT_EXPUBLIC;
1761 		}
1762 		if ((error = vfs_hang_addrlist(mp, nep, argp)))
1763 			return (error);
1764 		mp->mnt_flag |= MNT_EXPORTED;
1765 	}
1766 	return (0);
1767 }
1768 
1769 
1770 /*
1771  * Set the publicly exported filesystem (WebNFS). Currently, only
1772  * one public filesystem is possible in the spec (RFC 2054 and 2055)
1773  */
1774 int
1775 vfs_setpublicfs(struct mount *mp, struct netexport *nep,
1776 		const struct export_args *argp)
1777 {
1778 	int error;
1779 	struct vnode *rvp;
1780 	char *cp;
1781 
1782 	/*
1783 	 * mp == NULL -> invalidate the current info, the FS is
1784 	 * no longer exported. May be called from either vfs_export
1785 	 * or unmount, so check if it hasn't already been done.
1786 	 */
1787 	if (mp == NULL) {
1788 		if (nfs_pub.np_valid) {
1789 			nfs_pub.np_valid = 0;
1790 			if (nfs_pub.np_index != NULL) {
1791 				FREE(nfs_pub.np_index, M_TEMP);
1792 				nfs_pub.np_index = NULL;
1793 			}
1794 		}
1795 		return (0);
1796 	}
1797 
1798 	/*
1799 	 * Only one allowed at a time.
1800 	 */
1801 	if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
1802 		return (EBUSY);
1803 
1804 	/*
1805 	 * Get real filehandle for root of exported FS.
1806 	 */
1807 	bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
1808 	nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
1809 
1810 	if ((error = VFS_ROOT(mp, &rvp)))
1811 		return (error);
1812 
1813 	if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
1814 		return (error);
1815 
1816 	vput(rvp);
1817 
1818 	/*
1819 	 * If an indexfile was specified, pull it in.
1820 	 */
1821 	if (argp->ex_indexfile != NULL) {
1822 		int namelen;
1823 
1824 		error = vn_get_namelen(rvp, &namelen);
1825 		if (error)
1826 			return (error);
1827 		MALLOC(nfs_pub.np_index, char *, namelen, M_TEMP,
1828 		    M_WAITOK);
1829 		error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
1830 		    namelen, (size_t *)0);
1831 		if (!error) {
1832 			/*
1833 			 * Check for illegal filenames.
1834 			 */
1835 			for (cp = nfs_pub.np_index; *cp; cp++) {
1836 				if (*cp == '/') {
1837 					error = EINVAL;
1838 					break;
1839 				}
1840 			}
1841 		}
1842 		if (error) {
1843 			FREE(nfs_pub.np_index, M_TEMP);
1844 			return (error);
1845 		}
1846 	}
1847 
1848 	nfs_pub.np_mount = mp;
1849 	nfs_pub.np_valid = 1;
1850 	return (0);
1851 }
1852 
1853 struct netcred *
1854 vfs_export_lookup(struct mount *mp, struct netexport *nep,
1855 		struct sockaddr *nam)
1856 {
1857 	struct netcred *np;
1858 	struct radix_node_head *rnh;
1859 	struct sockaddr *saddr;
1860 
1861 	np = NULL;
1862 	if (mp->mnt_flag & MNT_EXPORTED) {
1863 		/*
1864 		 * Lookup in the export list first.
1865 		 */
1866 		if (nam != NULL) {
1867 			saddr = nam;
1868 			rnh = nep->ne_rtable[saddr->sa_family];
1869 			if (rnh != NULL) {
1870 				np = (struct netcred *)
1871 					(*rnh->rnh_matchaddr)((char *)saddr,
1872 							      rnh);
1873 				if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
1874 					np = NULL;
1875 			}
1876 		}
1877 		/*
1878 		 * If no address match, use the default if it exists.
1879 		 */
1880 		if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
1881 			np = &nep->ne_defexported;
1882 	}
1883 	return (np);
1884 }
1885 
1886 /*
1887  * perform msync on all vnodes under a mount point.  The mount point must
1888  * be locked.  This code is also responsible for lazy-freeing unreferenced
1889  * vnodes whos VM objects no longer contain pages.
1890  *
1891  * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
1892  *
1893  * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked,
1894  * but vnode_pager_putpages() doesn't lock the vnode.  We have to do it
1895  * way up in this high level function.
1896  */
1897 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data);
1898 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data);
1899 
1900 void
1901 vfs_msync(struct mount *mp, int flags)
1902 {
1903 	int vmsc_flags;
1904 
1905 	vmsc_flags = VMSC_GETVP;
1906 	if (flags != MNT_WAIT)
1907 		vmsc_flags |= VMSC_NOWAIT;
1908 	vmntvnodescan(mp, vmsc_flags, vfs_msync_scan1, vfs_msync_scan2,
1909 			(void *)flags);
1910 }
1911 
1912 /*
1913  * scan1 is a fast pre-check.  There could be hundreds of thousands of
1914  * vnodes, we cannot afford to do anything heavy weight until we have a
1915  * fairly good indication that there is work to do.
1916  */
1917 static
1918 int
1919 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data)
1920 {
1921 	int flags = (int)data;
1922 
1923 	if ((vp->v_flag & VRECLAIMED) == 0) {
1924 		if (vshouldmsync(vp))
1925 			return(0);	/* call scan2 */
1926 		if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
1927 		    (vp->v_flag & VOBJDIRTY) &&
1928 		    (flags == MNT_WAIT || vn_islocked(vp) == 0)) {
1929 			return(0);	/* call scan2 */
1930 		}
1931 	}
1932 
1933 	/*
1934 	 * do not call scan2, continue the loop
1935 	 */
1936 	return(-1);
1937 }
1938 
1939 /*
1940  * This callback is handed a locked vnode.
1941  */
1942 static
1943 int
1944 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data)
1945 {
1946 	vm_object_t obj;
1947 	int flags = (int)data;
1948 
1949 	if (vp->v_flag & VRECLAIMED)
1950 		return(0);
1951 
1952 	if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) {
1953 		if ((obj = vp->v_object) != NULL) {
1954 			vm_object_page_clean(obj, 0, 0,
1955 			 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
1956 		}
1957 	}
1958 	return(0);
1959 }
1960 
1961 /*
1962  * Record a process's interest in events which might happen to
1963  * a vnode.  Because poll uses the historic select-style interface
1964  * internally, this routine serves as both the ``check for any
1965  * pending events'' and the ``record my interest in future events''
1966  * functions.  (These are done together, while the lock is held,
1967  * to avoid race conditions.)
1968  */
1969 int
1970 vn_pollrecord(struct vnode *vp, int events)
1971 {
1972 	lwkt_tokref ilock;
1973 
1974 	KKASSERT(curthread->td_proc != NULL);
1975 
1976 	lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
1977 	if (vp->v_pollinfo.vpi_revents & events) {
1978 		/*
1979 		 * This leaves events we are not interested
1980 		 * in available for the other process which
1981 		 * which presumably had requested them
1982 		 * (otherwise they would never have been
1983 		 * recorded).
1984 		 */
1985 		events &= vp->v_pollinfo.vpi_revents;
1986 		vp->v_pollinfo.vpi_revents &= ~events;
1987 
1988 		lwkt_reltoken(&ilock);
1989 		return events;
1990 	}
1991 	vp->v_pollinfo.vpi_events |= events;
1992 	selrecord(curthread, &vp->v_pollinfo.vpi_selinfo);
1993 	lwkt_reltoken(&ilock);
1994 	return 0;
1995 }
1996 
1997 /*
1998  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
1999  * it is possible for us to miss an event due to race conditions, but
2000  * that condition is expected to be rare, so for the moment it is the
2001  * preferred interface.
2002  */
2003 void
2004 vn_pollevent(struct vnode *vp, int events)
2005 {
2006 	lwkt_tokref ilock;
2007 
2008 	lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
2009 	if (vp->v_pollinfo.vpi_events & events) {
2010 		/*
2011 		 * We clear vpi_events so that we don't
2012 		 * call selwakeup() twice if two events are
2013 		 * posted before the polling process(es) is
2014 		 * awakened.  This also ensures that we take at
2015 		 * most one selwakeup() if the polling process
2016 		 * is no longer interested.  However, it does
2017 		 * mean that only one event can be noticed at
2018 		 * a time.  (Perhaps we should only clear those
2019 		 * event bits which we note?) XXX
2020 		 */
2021 		vp->v_pollinfo.vpi_events = 0;	/* &= ~events ??? */
2022 		vp->v_pollinfo.vpi_revents |= events;
2023 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2024 	}
2025 	lwkt_reltoken(&ilock);
2026 }
2027 
2028 /*
2029  * Wake up anyone polling on vp because it is being revoked.
2030  * This depends on dead_poll() returning POLLHUP for correct
2031  * behavior.
2032  */
2033 void
2034 vn_pollgone(struct vnode *vp)
2035 {
2036 	lwkt_tokref ilock;
2037 
2038 	lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token);
2039 	if (vp->v_pollinfo.vpi_events) {
2040 		vp->v_pollinfo.vpi_events = 0;
2041 		selwakeup(&vp->v_pollinfo.vpi_selinfo);
2042 	}
2043 	lwkt_reltoken(&ilock);
2044 }
2045 
2046 /*
2047  * extract the cdev_t from a VBLK or VCHR.  The vnode must have been opened
2048  * (or v_rdev might be NULL).
2049  */
2050 cdev_t
2051 vn_todev(struct vnode *vp)
2052 {
2053 	if (vp->v_type != VBLK && vp->v_type != VCHR)
2054 		return (NULL);
2055 	KKASSERT(vp->v_rdev != NULL);
2056 	return (vp->v_rdev);
2057 }
2058 
2059 /*
2060  * Check if vnode represents a disk device.  The vnode does not need to be
2061  * opened.
2062  */
2063 int
2064 vn_isdisk(struct vnode *vp, int *errp)
2065 {
2066 	cdev_t dev;
2067 
2068 	if (vp->v_type != VCHR) {
2069 		if (errp != NULL)
2070 			*errp = ENOTBLK;
2071 		return (0);
2072 	}
2073 
2074 	if ((dev = vp->v_rdev) == NULL)
2075 		dev = get_dev(vp->v_umajor, vp->v_uminor);
2076 
2077 	if (dev == NULL) {
2078 		if (errp != NULL)
2079 			*errp = ENXIO;
2080 		return (0);
2081 	}
2082 	if (dev_is_good(dev) == 0) {
2083 		if (errp != NULL)
2084 			*errp = ENXIO;
2085 		return (0);
2086 	}
2087 	if ((dev_dflags(dev) & D_DISK) == 0) {
2088 		if (errp != NULL)
2089 			*errp = ENOTBLK;
2090 		return (0);
2091 	}
2092 	if (errp != NULL)
2093 		*errp = 0;
2094 	return (1);
2095 }
2096 
2097 int
2098 vn_get_namelen(struct vnode *vp, int *namelen)
2099 {
2100 	int error, retval[2];
2101 
2102 	error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval);
2103 	if (error)
2104 		return (error);
2105 	*namelen = *retval;
2106 	return (0);
2107 }
2108 
2109 int
2110 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type,
2111 		uint16_t d_namlen, const char *d_name)
2112 {
2113 	struct dirent *dp;
2114 	size_t len;
2115 
2116 	len = _DIRENT_RECLEN(d_namlen);
2117 	if (len > uio->uio_resid)
2118 		return(1);
2119 
2120 	dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO);
2121 
2122 	dp->d_ino = d_ino;
2123 	dp->d_namlen = d_namlen;
2124 	dp->d_type = d_type;
2125 	bcopy(d_name, dp->d_name, d_namlen);
2126 
2127 	*error = uiomove((caddr_t)dp, len, uio);
2128 
2129 	kfree(dp, M_TEMP);
2130 
2131 	return(0);
2132 }
2133 
2134